text
stringlengths 4
2.78M
| meta
dict |
|---|---|
---
address:
- 'Einstein Institute of Mathematics, Hebrew University of Jerusalem, Jerusalem, Israel'
- 'Institute for Advanced Study, Fuld Hall, 1 Einstein Drive, Princeton, New Jersey, USA.'
- 'Department of Combinatorics & Optimization, University of Waterloo, 200 University Avenue West, Waterloo, Ontario, Canada.'
author:
- Karim Adiprasito
- June Huh
- Eric Katz
title: Hodge theory for combinatorial geometries
---
Introduction
============
The combinatorial theory of matroids starts with Whitney [@Whitney], who introduced matroids as models for independence in vector spaces and graphs. By definition, a *matroid* $\mathrm{M}$ is given by a closure operator defined on all subsets of a finite set $E$ satisfying the Steinitz-MacLane exchange property: $$\begin{gathered}
\text{For every subset $I$ of $E$ and every element $a$ not in the closure of $I$,}\\
\text{if $a$ is in the closure of ${I \cup\{ b\}}$, then $b$ is in the closure of $I \cup \{a\}$.}\end{gathered}$$ The matroid is called *loopless* if the empty subset of $E$ is closed, and is called a *combinatorial geometry* if in addition all single element subsets of $E$ are closed. A closed subset of $E$ is called a *flat* of $\mathrm{M}$, and every subset of $E$ has a well-defined rank and corank in the poset of all flats of $\mathrm{M}$. The notion of matroid played a fundamental role in graph theory, coding theory, combinatorial optimization, and mathematical logic; we refer to [@Welsh] and [@Oxley] for general introduction.
As a generalization of the chromatic polynomial of a graph [@Birkhoff; @WhitneyLE], Rota defined for an arbitrary matroid $\mathrm{M}$ the *characteristic polynomial* $$\chi_\mathrm{M}( \lambda)=\sum_{I \subseteq E} (-1)^{|I|}\ \lambda^{\text{crk}(I)},
$$ where the sum is over all subsets $I \subseteq E$ and $\text{crk}(I)$ is the corank of $I$ in $\mathrm{M}$ [@Foundations]. Equivalently, the characteristic polynomial of $\mathrm{M}$ is $$\chi_{\mathrm{M}}(\lambda)\ =\ \sum_{F} \mu(\varnothing,F)
\hspace{0.5mm} \lambda^{\text{crk}(F)},$$ where the sum is over all flats $F$ of $\mathrm{M}$ and $\mu$ is the Möbius function of the poset of flats of $\mathrm{M}$, see Chapters 7 and 8 of [@white]. Among the problems that withstood many advances in matroid theory are the following log-concavity conjectures formulated in the 1970s. Write $r+1$ for the *rank* of $\mathrm{M}$, that is, the rank of $E$ in the poset of flats of $\mathrm{M}$.
\[ConjectureCharacteristic\] Let $w_k(\mathrm{M})$ be the absolute value of the coefficient of $\lambda^{r-k+1}$ in the characteristic polynomial of $\mathrm{M}$. Then the sequence $w_k(\mathrm{M})$ is log-concave: $$w_{k-1}(\mathrm{M}) w_{k+1}(\mathrm{M}) \le w_k(\mathrm{M})^2 \ \ \text{for all $1 \le k\le r$.}$$ In particular, the sequence $w_k(\mathrm{M})$ is unimodal: $$w_0(\mathrm{M}) \le w_1(\mathrm{M}) \le \cdots \le w_l(\mathrm{M}) \ge \cdots \ge w_r(\mathrm{M}) \ge w_{r+1}(\mathrm{M}) \ \ \text{for some index $l$.}$$
We remark that the positivity of the numbers $w_k(\mathrm{M})$ is used to deduce the unimodality from the log-concavity [@WelshBook Chapter 15].
For chromatic polynomials, the unimodality was conjectured by Read, and the log-concavity was conjectured by Hoggar [@Read; @Hoggar]. The prediction of Read was then extended to arbitrary matroids by Rota and Heron, and the conjecture in its full generality was given by Welsh [@Rota; @Heron; @WelshBook]. We refer to [@white Chapter 8] and [@Oxley Chapter 15] for overviews and historical accounts.
A subset $I \subseteq E$ is said to be *independent* in $\mathrm{M}$ if no element $i$ in $I$ is in the closure of $I \setminus \{i\}$. A related conjecture of Welsh and Mason concerns the number of independent subsets of $E$ of given cardinality [@Welsh; @Mason].
\[ConjectureIndependence\] Let $f_k(\mathrm{M})$ be the number of independent subsets of $E$ with cardinality $k$. Then the sequence $f_k(\mathrm{M})$ is log-concave: $$f_{k-1}(\mathrm{M}) f_{k+1}(\mathrm{M}) \le f_k(\mathrm{M})^2 \ \ \text{for all $1 \le k \le r$.}$$ In particular, the sequence $f_k(\mathrm{M})$ is unimodal: $$f_0(\mathrm{M}) \le f_1(\mathrm{M}) \le \cdots \le f_l(\mathrm{M}) \ge \cdots \ge f_r(\mathrm{M}) \ge f_{r+1}(\mathrm{M}) \ \ \text{for some index $l$.}$$
We prove Conjecture \[ConjectureCharacteristic\] and Conjecture \[ConjectureIndependence\] by constructing a “cohomology ring” of $\mathrm{M}$ that satisfies the hard Lefschetz theorem and the Hodge-Riemann relations, see Theorem \[MainTheoremIntroduction\].
Matroid theory has experienced a remarkable development in the past century, and has been connected to diverse areas such as topology [@GM], geometric model theory [@Pillay], and noncommutative geometry [@Neumann]. The study of complex hyperplane arrangements provided a particularly strong connection, see for example [@Arrangements]. Most important for our purposes is the work of de Concini and Procesi on certain “wonderful” compactifications of hyperplane arrangement complements [@DeConcini-Procesi]. The original work focused only on realizable matroids, but Feichtner and Yuzvinsky [@Feichtner-Yuzvinsky] defined a commutative ring associated to an arbitrary matroid that specializes to the cohomology ring of a wonderful compactification in the realizable case.
Let $S_\mathrm{M}$ be the polynomial ring $$S_\mathrm{M}:=\mathbb{R}\big[x_F| \text{$F$ is a nonempty proper flat of $\mathrm{M}$}\big].$$ The *Chow ring* of $\mathrm{M}$ is defined to be the quotient $$A^*(\mathrm{M})_\mathbb{R}:=S_\mathrm{M}/(I_\mathrm{M}+J_\mathrm{M}),$$ where $I_\mathrm{M}$ is the ideal generated by the quadratic monomials $$x_{F_1}x_{F_2}, \ \ \text{$F_1$ and $F_2$ are two incomparable nonempty proper flats of $\mathrm{M}$,}$$ and $J_\mathrm{M}$ is the ideal generated by the linear forms $$\sum_{i_1 \in F} x_F - \sum_{i_2 \in F} x_F, \ \ \text{$i_1$ and $i_2$ are distinct elements of the ground set $E$.}$$
Conjecture \[ConjectureCharacteristic\] was proved for matroids realizable over $\mathbb{C}$ in [@Huh] by relating $w_k(\mathrm{M})$ to the Milnor numbers of a hyperplane arrangement realizing $\mathrm{M}$ over $\mathbb{C}$. Subsequently in [@Huh-Katz], using the intersection theory of wonderful compactifications and the Khovanskii-Teissier inequality [@Lazarsfeld Section 1.6], the conjecture was verified for matroids that are realizable over some field. Lenz used this result to deduce Conjecture \[ConjectureIndependence\] for matroids realizable over some field [@Lenz].
After the completion of [@Huh-Katz], it was gradually realized that the validity of the Hodge-Riemann relations for the Chow ring of $\mathrm{M}$ is a vital ingredient for the proof of the log-concavity conjectures, see Theorem \[MainTheoremIntroduction\] below. While the Chow ring of $\mathrm{M}$ could be defined for arbitrary $\mathrm{M}$, it was unclear how to formulate and prove the Hodge-Riemann relations. From the point of view of [@Feichtner-Yuzvinsky], the ring $A^*(\mathrm{M})_\mathbb{R}$ is the Chow ring of a smooth, but noncompact toric variety $X(\Sigma_{\mathrm{M}})$, and there is no obvious way to reduce to the classical case of projective varieties. In fact, we will see that $X(\Sigma_{\mathrm{M}})$ is “Chow equivalent” to a smooth or mildly singular projective variety over $\mathbb{K}$ if and only if the matroid $\mathrm{M}$ is realizable over $\mathbb{K}$, see Theorem \[Chow-equivalence\].
We are nearing a difficult chasm, as there is no reason to expect a working Hodge theory beyond the case of realizable matroids. Nevertheless, there was some evidence on the existence of such a theory for arbitrary matroids. For example, it was proved in [@Adiprasito], using the method of concentration of measure, that the log-concavity conjectures hold for $c$-arrangements in the sense of Goresky and MacPherson [@GM].
We now state the main theorem of this paper. A function $c$ on the set of nonempty proper subsets of $E$ is said to be *strictly submodular* if $$c_{I_1}+c_{I_2} > c_{I_1 \cap I_2} +c_{I_1 \cup I_2} \ \ \text{for any two incomparable subsets $I_1,I_2 \subseteq E$,}$$ where we replace $c_\varnothing$ and $c_E$ by zero whenever they appear in the above inequality. A strictly submodular function $c$ defines an element $$\ell(c):= \sum_F c_F x_F\in A^1(\mathrm{M})_\mathbb{R},$$ where the sum is over all nonempty proper flats of $\mathrm{M}$. Note that the rank function of *any* matroid on $E$, and more generally any submodular function of the free matroid on $E$, can, when restricted to the set of nonempty proper subsets of $E$, be obtained as a *limit* of strictly submodular functions. We write “$\text{deg}$” for the isomorphism $A^r(\mathrm{M})_\mathbb{R} \simeq \mathbb{R}$ determined by the property that $$\text{deg}(x_{F_1}x_{F_2}\cdots x_{F_r})=1 \ \ \text{for any flag of nonempty proper flats} \ \ F_1 \subsetneq F_2 \subsetneq \cdots \subsetneq F_r.$$
\[MainTheoremIntroduction\] Let $\ell$ be an element of $A^1(\mathrm{M})_\mathbb{R}$ associated to a strictly submodular function.
(Hard Lefschetz theorem) For every nonnegative integer $q \le \frac{r}{2}$, the multiplication by $\ell$ defines an isomorphism $$L_\ell^{q}:\ A^q(\mathrm{M})_{\mathbb{R}} \longrightarrow A^{r-q}(\mathrm{M})_{\mathbb{R}}, \qquad a \longmapsto \ell^{r-2q} \cdot a.$$
(Hodge–Riemann relations) For every nonnegative integer $q \le \frac{r}{2}$, the multiplication by $\ell$ defines a symmetric bilinear form $$Q_\ell^{q}:\ A^q(\mathrm{M})_{\mathbb{R}} \times A^q(\mathrm{M})_{\mathbb{R}} \longrightarrow \mathbb{R}, \qquad (a_1,a_2) \longmapsto (-1)^q \ \text{deg}(a_1 \cdot L^q_\ell \ a_2)$$ that is positive definite on the kernel of $\ell \cdot L^q_\ell$.
In fact, we will prove that the Chow ring of $\mathrm{M}$ satisfies the hard Lefschetz theorem and the Hodge-Riemann relations with respect to any strictly convex piecewise linear function on the tropical linear space $\Sigma_\mathrm{M}$ associated to $\mathrm{M}$, see Theorem \[MainTheoremBody\]. This implies Theorem \[MainTheoremIntroduction\]. Our proof of the hard Lefschetz theorem and the Hodge-Riemann relations for general matroids is inspired by an ingenious inductive proof of the analogous facts for simple polytopes given by McMullen [@McMullen] (compare also [@CM] for related ideas in a different context). To show that this program, with a considerable amount of work, extends beyond polytopes, is our main purpose here.
In Section \[SectionLCConjectures\], we show that the Hodge-Riemann relations, which are in fact stronger than the hard Lefschetz theorem, imply Conjecture \[ConjectureCharacteristic\] and Conjecture \[ConjectureIndependence\]. We remark that, in the context of projective toric varieties, a similar reasoning leads to the Alexandrov-Fenchel inequality on mixed volumes of convex bodies. In this respect, broadly speaking the approach of the present paper can be viewed as following Rota’s idea that log-concavity conjectures should follow from their relation with the theory of mixed volumes of convex bodies, see [@Kung].
There are other combinatorial approaches to intersection theory for matroids. Mikhalkin et. al. introduced an integral Hodge structure for arbitrary matroids modeled on the cohomology of hyperplane arrangement complements [@IKMZ]. Adiprasito and Björner showed that an analogue of the Lefschetz hyperplane section theorem holds for all smooth (i.e. locally matroidal) projective tropical varieties [@ABj-pr]. We will discuss the relations with the above perspectives on Hodge theory for matroids in the upcoming paper [@AHK2].
Theorem \[MainTheoremIntroduction\] should be compared with the counterexample to a version of Hodge conjecture for positive currents in [@BH]: The example used in [@BH] gives a tropical variety that satisfies the Poincaré duality, the hard Lefschetz theorem, but not the Hodge-Riemann relations. We will also discuss this in detail in [@AHK2].
Finally, we remark that Zilber and Hrushovski have worked on subjects related to intersection theory for finitary combinatorial geometries, see [@Hrushovski]. At present the relationship between their approach and ours is unclear.
Acknowledgements {#acknowledgements .unnumbered}
----------------
The authors thank Patrick Brosnan, Eduardo Cattani, Ben Elias, Ehud Hrushovski, Gil Kalai, and Sam Payne for valuable conversations. Karim Adiprasito was supported by a Minerva Fellowship from the Max Planck Society and NSF Grant DMS-1128155. June Huh was supported by a Clay Research Fellowship and NSF Grant DMS-1128155. Eric Katz was supported by an NSERC Discovery grant.
Finite sets and their subsets
=============================
Let $E$ be a nonempty finite set of cardinality $n+1$, say $\{0,1,\ldots,n\}$. We write $\mathbb{Z}^E$ for the free abelian group generated by the standard basis vectors $\mathbf{e}_i$ corresponding to the elements $i \in E$. For an arbitrary subset $I \subseteq E$, we set $$\mathbf{e}_I:=\sum_{i \in I} \mathbf{e}_i.$$ We associate to the set $E$ a dual pair of rank $n$ free abelian groups $$N_E:=\mathbb{Z}^E/ \langle\mathbf{e}_E\rangle, \qquad M_E:=\mathbf{e}_E^\perp \subset \mathbb{Z}^E, \qquad \langle -,- \rangle:N_E \times M_E \longrightarrow \mathbb{Z}.$$ The corresponding real vector spaces will be denoted $$N_{E,\mathbb{R}}:=N_E \otimes_\mathbb{Z} \mathbb{R}, \qquad M_{E,\mathbb{R}}:=M_E \otimes_\mathbb{Z} \mathbb{R}.$$ We use the same symbols $\mathbf{e}_i$ and $\mathbf{e}_I$ to denote their images in $N_E$ and $N_{E,\mathbb{R}}$.
The groups $N$ and $M$ associated to nonempty finite sets are related to each other in a natural way. For example, if $F$ is a nonempty subset of $E$, then there is a surjective homomorphism $$N_E \longrightarrow N_F, \qquad \mathbf{e}_I \longmapsto \mathbf{e}_{I \cap F},$$ and an injective homomorphism $$M_F \longrightarrow M_E, \qquad \mathbf{e}_i-\mathbf{e}_j \longmapsto \mathbf{e}_i-\mathbf{e}_j.$$ If $F$ is a nonempty proper subset of $E$, we have a decomposition $$(\mathbf{e}_F^\perp \subset M_E)=(\mathbf{e}_{E\setminus F}^\perp \subset M_E)=M_F \oplus M_{E \setminus F}.$$ Dually, we have an isomorphism from the quotient space $$N_E/\langle \mathbf{e}_F\rangle=N_E/\langle \mathbf{e}_{E \setminus F} \rangle \longrightarrow N_F \oplus N_{E \setminus F}, \qquad \mathbf{e}_I \longmapsto \mathbf{e}_{I \cap F} \oplus \mathbf{e}_{I \setminus F}.$$ This isomorphism will be used later to analyze local structure of Bergman fans.
More generally, for any map between nonempty finite sets $\pi:E_1 \to E_2$, there is an associated homomorphism $$\pi_N:N_{E_2} \longrightarrow N_{E_1}, \quad \mathbf{e}_I \longmapsto \mathbf{e}_{\pi^{-1}(I)},$$ and the dual homomorphism $$\pi_M:M_{E_1} \longrightarrow M_{E_2}, \qquad \mathbf{e}_i-\mathbf{e}_j \longmapsto \mathbf{e}_{\pi(i)}-\mathbf{e}_{\pi(j)}.$$ When $\pi$ is surjective, $\pi_N$ is injective and $\pi_M$ is surjective.
Let $\mathscr{P}(E)$ be the poset of nonempty proper subsets of $E$. Throughout this section the symbol $\mathscr{F}$ will stand for a totally ordered subset of $\mathscr{P}(E)$, that is, a flag of nonempty proper subsets of $E$: $$\mathscr{F}=\Big\{F_1 \subsetneq F_2 \subsetneq \cdots \subsetneq F_l\Big\} \subseteq \mathscr{P}(E).$$ We write $\text{min}\ \mathscr{F}$ for the intersection of all members of $\mathscr{F}$. By definition, $\text{min}\ \varnothing=E$.
\[BooleanCompatiblePair\] When $I$ is a proper subset of $\text{min}\ \mathscr{F}$, we say that $I$ is *compatible* with $\mathscr{F}$ in $E$, and write $I<\mathscr{F}$.
The set of all compatible pairs in $E$ form a poset under the relation $$(I_1<\mathscr{F}_1) \preceq (I_2<\mathscr{F}_2) \Longleftrightarrow \text{$I_1 \subseteq I_2$ and $\mathscr{F}_1 \subseteq \mathscr{F}_2$}.$$ We note that any maximal compatible pair $I<\mathscr{F}$ gives a basis of the group $N_E$: $$\Big\{\text{$\mathbf{e}_i$ and $\mathbf{e}_F$ for $i \in I$ and $F \in \mathscr{F}$}\Big\} \subseteq N_E.$$ If $0$ is the unique element of $E$ not in $I$ and not in any member of $\mathscr{F}$, then the above basis of $N_E$ is related to the basis $\{\mathbf{e}_1,\mathbf{e}_2,\ldots,\mathbf{e}_n\}$ by an invertible upper triangular matrix.
For each compatible pair $I<\mathscr{F}$ in $E$, we define two polyhedra $$\begin{aligned}
\vartriangle_{I < \mathscr{F}}&:=\text{conv}\Big\{\text{$\mathbf{e}_i$ and $\mathbf{e}_F$ for $i \in I$ and $F \in \mathscr{F}$}\Big\}\subseteq N_{E,\mathbb{R}}, \\[2pt]
\sigma_{I <\mathscr{F}}&:=\text{cone}\Big\{\text{$\mathbf{e}_i$ and $\mathbf{e}_F$ for $i \in I$ and $F \in \mathscr{F}$}\Big\} \subseteq N_{E,\mathbb{R}}.\end{aligned}$$
Since maximal compatible pairs give bases of $N_E$, the polytope $\vartriangle_{I < \mathscr{F}}$ is a simplex, and the cone $\sigma_{I<\mathscr{F}}$ is unimodular. Any proper subset of $E$ is compatible with the empty flag in $\mathscr{P}(E)$, and the empty subset of $E$ is compatible with any flag in $\mathscr{P}(E)$. Therefore we may write $$\vartriangle_{I<\mathscr{F}}\ =\ \vartriangle_{I<\varnothing}*\vartriangle_{\varnothing<\mathscr{F}}\ \ \text{and} \ \ \sigma_{I<\mathscr{F}}=\sigma_{I<\varnothing}+\sigma_{\varnothing<\mathscr{F}}.$$ The set of all simplices of the form $\vartriangle_{I<\mathscr{F}}$ is in fact a simplicial complex, that is, $$\vartriangle_{I_1 < \mathscr{F}_1} \cap \vartriangle_{I_2 < \mathscr{F}_2}=\vartriangle_{I_1 \cap I_2 < \mathscr{F}_1 \cap \mathscr{F}_2}.$$ This gives a geometric realization of the poset of compatible pairs in $E$.
An *order filter* $\mathscr{P}$ of $\mathscr{P}(E)$ is a collection of nonempty proper subsets of $E$ with the following property: $$\text{If $F_1 \subseteq F_2$ are nonempty proper subsets of $E$, then $F_1 \in \mathscr{P}$ implies $F_2 \in \mathscr{P}$.}$$ Any such order filter cuts out a simplicial sphere in the simplicial complex of compatible pairs.
The *Bergman complex* of an order filter $\mathscr{P} \subseteq \mathscr{P}(E)$ is the collection of simplices $$\Delta_\mathscr{P}:=\Big\{\text{$\vartriangle_{I < \mathscr{F}}\ $ for $I \notin \mathscr{P}$ and $\mathscr{F} \subseteq\mathscr{P}$}\Big\}.$$ The *Bergman fan* of an order filter $\mathscr{P} \subseteq \mathscr{P}(E)$ is the collection of simplicial cones $$\Sigma_\mathscr{P}:=\Big\{ \text{$\sigma_{I < \mathscr{F}}\ $ for $I \notin \mathscr{P}$ and $\mathscr{F} \subseteq\mathscr{P}$}\Big\}.$$
The Bergman complex $\Delta_\mathscr{P}$ is a simplicial complex because $\mathscr{P}$ is an order filter. We will see below that the Bergman fan $\Sigma_\mathscr{P}$ indeed is a fan, that is, $$\sigma_{I_1 < \mathscr{F}_1} \cap \sigma_{I_2 < \mathscr{F}_2}=\sigma_{I_1 \cap I_2 < \mathscr{F}_1 \cap \mathscr{F}_2} \ \ \text{for} \ \ \sigma_{I_1 < \mathscr{F}_1}, \sigma_{I_2 < \mathscr{F}_2} \in \Sigma_\mathscr{P}.$$
The extreme cases $\mathscr{P}=\varnothing$ and $\mathscr{P}=\mathscr{P}(E)$ correspond to familiar geometric objects. When $\mathscr{P}$ is empty, the collection $\Sigma_\mathscr{P}$ is the normal fan of the standard $n$-dimensional simplex $$\Delta_n:=\text{conv}\big\{\mathbf{e}_0,\mathbf{e}_1,\ldots,\mathbf{e}_n\big\} \subseteq \mathbb{R}^E.$$ When $\mathscr{P}$ contains all nonempty proper subsets of $E$, the collection $\Sigma_\mathscr{P}$ is the normal fan of the $n$-dimensional permutohedron $$\Pi_n:=\text{conv}\Big\{(x_0,x_1,\ldots,x_n) \mid \text{$x_0,x_1,\ldots,x_n$ is a permutation of $0,1,\ldots,n$}\Big\} \subseteq \mathbb{R}^E.$$ Proposition \[NormalFan\] below shows that, in general, the Bergman complex $\Delta_\mathscr{P}$ is a simplicial sphere and $\Sigma_\mathscr{P}$ is a complete unimodular fan.
\[NormalFan\] For any order filter $\mathscr{P} \subseteq \mathscr{P}(E)$, the collection $\Sigma_\mathscr{P}$ is the normal fan of a polytope.
We show that $\Sigma_\mathscr{P}$ can be obtained from $\Sigma_\varnothing$ by performing a sequence of stellar subdivisions. In other words, a polytope corresponding to $\Sigma_\mathscr{P}$ can be obtained by repeatedly truncating faces of the standard simplex $\Delta_n$.
For this we choose a sequence of order filters obtained by adding a single subset in $\mathscr{P}$ at a time: $$\varnothing,\ldots, \mathscr{P}_{-},\mathscr{P}_+,\ldots,\mathscr{P} \quad \text{with} \quad \mathscr{P}_+=\mathscr{P}_- \cup \{Z\}.$$ The corresponding sequence of $\Sigma$ interpolates between the collections $\Sigma_\varnothing$ and $\Sigma_\mathscr{P}$: $$\Sigma_\varnothing \rightsquigarrow \ldots \rightsquigarrow \Sigma_{\mathscr{P}_{-}}\rightsquigarrow \Sigma_{\mathscr{P}_{+}} \rightsquigarrow \ldots \rightsquigarrow \Sigma_{\mathscr{P}}.$$ The modification in the middle replaces the cones of the form $\sigma_{Z<\mathscr{F}}$ with the sums of the form $$\sigma_{\varnothing<\{Z\}}+\sigma_{I<\mathscr{F}},$$ where $I$ is any proper subset of $Z$. In other words, the modification is the stellar subdivision of $\Sigma_{\mathscr{P}_-}$ relative to the cone $\sigma_{Z<\varnothing}$. Since a stellar subdivision of the normal fan of a polytope is the normal fan of a polytope, by induction we know that the collection $\Sigma_\mathscr{P}$ is the normal fan of a polytope.
Matroids and their flats
========================
Let $\mathrm{M}$ be a loopless matroid of rank $r+1$ on the ground set $E$. We denote $\text{rk}_\mathrm{M}$, $\text{crk}_\mathrm{M}$, and $\text{cl}_\mathrm{M}$ for the rank function, the corank function, and the closure operator of $\mathrm{M}$ respectively. We omit the subscripts when $\mathrm{M}$ is understood from the context. If $F$ is a nonempty proper flat of $\mathrm{M}$, we write $$\begin{aligned}
\mathrm{M}^F&:=\text{the restriction of $\mathrm{M}$ to $F$, a loopless matroid on $F$ of rank $=\text{rk}_\mathrm{M}(F)$},\\
\mathrm{M}_F&:=\text{the contraction of $\mathrm{M}$ by $F$, a loopless matroid on $E \setminus F$ of rank $=\text{crk}_\mathrm{M}(F)$}.\end{aligned}$$ We refer to [@Oxley] and [@WelshBook] for basic notions of matroid theory.
Let $\mathscr{P}(\mathrm{M})$ be the poset of nonempty proper flats of $\mathrm{M}$. There is an injective map from the poset of the restriction $$\iota^F: \mathscr{P}(\mathrm{M}^F) \longrightarrow \mathscr{P}(\mathrm{M}), \qquad G \longmapsto G,$$ and an injective map from the poset of the contraction $$\iota_F:\mathscr{P}(\mathrm{M}_F) \longrightarrow \mathscr{P}(\mathrm{M}), \qquad G \longmapsto G \cup F.$$ We identify the flats of $\mathrm{M}_F$ with the flats of $\mathrm{M}$ containing $F$ using $\iota_F$. If $\mathscr{P}$ is a subset of $\mathscr{P}(\mathrm{M})$, we set $$\mathscr{P}^F:=(\iota^F)^{-1} \mathscr{P} \ \ \text{and} \ \ \mathscr{P}_F:=(\iota_F)^{-1} \mathscr{P}.$$
Throughout this section the symbol $\mathscr{F}$ will stand for a totally ordered subset of $\mathscr{P}(\mathrm{M})$, that is, a flag of nonempty proper flats of $\mathrm{M}$: $$\mathscr{F}=\Big\{F_1 \subsetneq F_2 \subsetneq \cdots \subsetneq F_l\Big\} \subseteq \mathscr{P}(\mathrm{M}).$$ As before, we write $\text{min}\ \mathscr{F}$ for the intersection of all members of $\mathscr{F}$ inside $E$. We extend the notion of compatibility in Definition \[BooleanCompatiblePair\] to the case when the matroid $\mathrm{M}$ is not Boolean.
When $I$ is a subset of $\text{min}\ \mathscr{F}$ of cardinality less than $\text{rk}_\mathrm{M}(\text{min}\ \mathscr{F})$, we say that $I$ is *compatible* with $\mathscr{F}$ in $\mathrm{M}$, and write $I<_\mathrm{M}\mathscr{F}$.
Since any flag of nonempty proper flats of $\mathrm{M}$ has length at most $r$, any cone $$\sigma_{I<_\mathrm{M}\mathscr{F}}=\text{cone}\Big\{\text{$\mathbf{e}_i$ and $\mathbf{e}_F$ for $i \in I$ and $F \in \mathscr{F}$}\Big\}$$ associated to a compatible pair in $\mathrm{M}$ has dimension at most $r$. Conversely, any such cone is contained in an $r$-dimensional cone of the same type: For this one may take $$\begin{aligned}
I' &= \text{a subset that is maximal among those containing $I$ and compatible with $\mathscr{F}$ in $\mathrm{M}$,}\\
\mathscr{F}' &= \text{a flag of flats maximal among those containing $\mathscr{F}$ and compatible with $I'$ in $\mathrm{M}$},\end{aligned}$$ or alternatively take $$\begin{aligned}
\mathscr{F}' &= \text{a flag of flats maximal among those containing $\mathscr{F}$ and compatible with $I$ in $\mathrm{M}$,}\\
I' &= \text{a subset that is maximal among those containing $I$ and compatible with $\mathscr{F}'$ in $\mathrm{M}$.}\end{aligned}$$
We note that any subset of $E$ with cardinality at most $r$ is compatible in $\mathrm{M}$ with the empty flag of flats, and the empty subset of $E$ is compatible in $\mathrm{M}$ with any flag of nonempty proper flats of $\mathrm{M}$. Therefore we may write $$\vartriangle_{I<_\mathrm{M}\mathscr{F}}\ =\ \vartriangle_{I<_\mathrm{M}\varnothing}*\vartriangle_{\varnothing<_\mathrm{M}\mathscr{F}}\ \ \text{and} \ \ \sigma_{I<_\mathrm{M}\mathscr{F}}=\sigma_{I<_\mathrm{M}\varnothing}+\sigma_{\varnothing<_\mathrm{M}\mathscr{F}}.$$ The set of all simplices associated to compatible pairs in $\mathrm{M}$ form a simplicial complex, that is, $$\vartriangle_{I_1 <_\mathrm{M} \mathscr{F}_1} \cap \vartriangle_{I_2 <_\mathrm{M} \mathscr{F}_2}=\vartriangle_{I_1 \cap I_2 <_\mathrm{M} \mathscr{F}_1 \cap \mathscr{F}_2}.$$
An *order filter* $\mathscr{P}$ of $\mathscr{P}(\mathrm{M})$ is a collection of nonempty proper flats of $\mathrm{M}$ with the following property: $$\text{If $F_1 \subseteq F_2$ are nonempty proper flats of $\mathrm{M}$, then $F_1 \in \mathscr{P}$ implies $F_2 \in \mathscr{P}$.}$$ We write $\widehat{\mathscr{P}}:=\mathscr{P} \cup \{E\}$ for the order filter of the lattice of flats of $\mathrm{M}$ generated by $\mathscr{P}$.
\[BergmanFan\] The *Bergman fan* of an order filter $\mathscr{P} \subseteq \mathscr{P}(\mathrm{M})$ is the set of simplicial cones $$\Sigma_{\mathrm{M},\mathscr{P}}:=\Big\{ \text{$\sigma_{I < \mathscr{F}}\ $ for $\text{cl}_\mathrm{M}(I) \notin \widehat{\mathscr{P}}$ and $\mathscr{F} \subseteq\mathscr{P}$}\Big\}.$$ The *reduced Bergman fan* of $\mathscr{P}$ is the subset of the Bergman fan $$\widetilde{\Sigma}_{\mathrm{M},\mathscr{P}}:=\Big\{ \text{$\sigma_{I <_\mathrm{M} \mathscr{F}}\ $ for $\text{cl}_\mathrm{M}(I) \notin \widehat{\mathscr{P}}$ and $\mathscr{F} \subseteq\mathscr{P}$}\Big\}.$$ When $\mathscr{P}=\mathscr{P}(\mathrm{M})$, we omit $\mathscr{P}$ from the notation and write the Bergman fan by $\Sigma_\mathrm{M}$.
We note that the *Bergman complex* and the *reduced Bergman complex* $\widetilde{\Delta}_{\mathrm{M},\mathscr{P}} \subseteq \Delta_{\mathrm{M},\mathscr{P}}$, defined in analogous ways using the simplices $\vartriangle_{I<\mathscr{F}}$ and $\vartriangle_{I<_\mathrm{M}\mathscr{F}}$, share the same set of vertices.
Two extreme cases give familiar geometric objects. When $\mathscr{P}$ is the set of all nonempty proper flats of $\mathrm{M}$, we have $$\Sigma_{\mathrm{M}}=\Sigma_{\mathrm{M},\mathscr{P}}=\widetilde{\Sigma}_{\mathrm{M},\mathscr{P}}=\text{the fine subdivision of the tropical linear space of $\mathrm{M}$ \cite{Ardila-Klivans}}.$$ When $\mathscr{P}$ is empty, we have $$\widetilde{\Sigma}_{\mathrm{M},\varnothing}=\text{the $r$-dimensional skeleton of the normal fan of the simplex $\Delta_n$},$$ and $\Sigma_{\mathrm{M},\varnothing}$ is the fan whose maximal cones are $\sigma_{F<\varnothing}$ for rank $r$ flats $F$ of $\mathrm{M}$. We remark that $$\Delta_{\mathrm{M},\varnothing}=\text{the Alexander dual of the matroid complex $\text{IN}(\textrm{M}^*)$ of the dual matroid $\mathrm{M}^*$}.$$ See [@Bjorner] for basic facts on the matroid complexes and [@Miller-Sturmfels Chapter 5] for the Alexander dual of a simplicial complex.
We show that, in general, the Bergman fan and the reduced Bergman fan are indeed fans, and the reduced Bergman fan is pure of dimension $r$.
\[Subfan\] The collection $\Sigma_{\mathrm{M},\mathscr{P}}$ is a subfan of the normal fan of a polytope.
Since $\mathscr{P}$ is an order filter, any face of a cone in $\Sigma_{\mathrm{M},\mathscr{P}}$ is in $\Sigma_{\mathrm{M},\mathscr{P}}$. Therefore it is enough to show that there is a normal fan of a polytope that contains $\Sigma_{\mathrm{M},\mathscr{P}}$ as a subset.
For this we consider the order filter of $\mathscr{P}(E)$ generated by $\mathscr{P}$, that is, the collection of sets $$\widetilde{\mathscr{P}}:=\big\{\text{nonempty proper subset of $E$ containing a flat in $\mathscr{P}$}\big\} \subseteq \mathscr{P}(E).$$ If the closure of $I \subseteq E$ in $\mathrm{M}$ is not in $\widehat{\mathscr{P}}$, then $I$ does not contain any flat in $\mathscr{P}$, and hence $$\Sigma_{\mathrm{M},\mathscr{P}} \subseteq \Sigma_{\widetilde{\mathscr{P}}}.$$ The latter collection is the normal fan of a polytope by Proposition \[NormalFan\].
Since $\mathscr{P}$ is an order filter, any face of a cone in $\widetilde{\Sigma}_{\mathrm{M},\mathscr{P}}$ is in $\widetilde{\Sigma}_{\mathrm{M},\mathscr{P}}$, and hence $\widetilde{\Sigma}_{\mathrm{M},\mathscr{P}}$ is a subfan of $\Sigma_{\mathrm{M},\mathscr{P}}$. It follows that the reduced Bergman fan also is a subfan of the normal fan of a polytope.
\[Purity\] The reduced Bergman fan $\widetilde{\Sigma}_{\mathrm{M},\mathscr{P}}$ is pure of dimension $r$.
Let $I$ be a subset of $E$ whose closure is not in $\mathscr{P}$, and let $\mathscr{F}$ be a flag of flats in $\mathscr{P}$ compatible with $I$ in $\mathrm{M}$. We show that there are $I'$ containing $I$ and $\mathscr{F}'$ containing $\mathscr{F}$ such that $$I'<_\mathrm{M} \mathscr{F}', \quad \text{cl}_\mathrm{M}(I') \notin \widehat{\mathscr{P}}, \quad \mathscr{F}' \subseteq \mathscr{P}, \quad \text{and} \quad |I'|+|\mathscr{F}'|=r.$$
First choose any flag of flats $\mathscr{F}'$ that is maximal among those containing $\mathscr{F}$, contained in $\mathscr{P}$, and compatible with $I$ in $\mathrm{M}$. Next choose any flat $F$ of $\mathrm{M}$ that is maximal among those containing $I$ and strictly contained in $\text{min}\ \mathscr{F}'$.
We note that, by the maximality of $F$ and the maximality of $\mathscr{F}'$ respectively, $$\text{rk}_\mathrm{M}(F)=\text{rk}_\mathrm{M}(\text{min}\ \mathscr{F}') -1=r-|\mathscr{F}'|.$$ Since the rank of a set is at most its cardinality, the above implies $$|I| \le r-|\mathscr{F}'| \le |F|.$$ This shows that there is $I'$ containing $I$, contained in $F$, and with cardinality exactly $r-|\mathscr{F}'|$. Any such $I'$ is automatically compatible with $\mathscr{F}'$ in $\mathrm{M}$.
We show that the closure of $I'$ is not in $\mathscr{P}$ by showing that the flat $F$ is not in $\mathscr{P}$. If otherwise, by the maximality of $\mathscr{F}'$, the set $I$ cannot be compatible in $\mathrm{M}$ with the flag $\{F\}$, meaning $$|I| \ge \text{rk}_\mathrm{M}(F).$$ The above implies that the closure of $I$ in $\mathrm{M}$, which is not in $\mathscr{P}$, is equal to $F$. This gives the desired contradiction.
Our inductive approach to the hard Lefschetz theorem and the Hodge-Riemann relations for matroids is modeled on the observation that any facet of a permutohedron is the product of two smaller permutohedrons. We note below that the Bergman fan $\Sigma_{\mathrm{M},\mathscr{P}}$ has an analogous local structure when $\mathrm{M}$ has no parallel elements.
Recall that the *star* of a cone $\sigma$ in a fan $\Sigma$ in a latticed vector space $N_\mathbb{R}$ is the fan $$\text{star}(\sigma,\Sigma):=\big\{\overline{\sigma'} \mid \text{$\overline{\sigma'} $ is the image in $N_\mathbb{R}/\langle \sigma\rangle$ of a cone $\sigma'$ in $\Sigma$ containing $\sigma$}\big\}.$$ When $\sigma$ is a ray generated by its primitive generator $\mathbf{e}$, we write $\text{star}(\mathbf{e},\Sigma)$ for the star of $\sigma$ in $\Sigma$.
\[Star\] Let $\mathrm{M}$ be a loopless matroid on $E$, and let $\mathscr{P}$ be an order filter of $\mathscr{P}(\mathrm{M})$.
(1) If $F$ is a flat in $\mathscr{P}$, then the isomorphism $
N_{E}/\langle \mathbf{e}_F\rangle \rightarrow N_{F} \oplus N_{E \setminus F}
$ induces a bijection $$\text{star}(\mathbf{e}_F,\Sigma_{\mathrm{M},\mathscr{P}}) \longrightarrow \Sigma_{\mathrm{M}^F,\mathscr{P}^F} \times \Sigma_{\mathrm{M}_F}.$$
(2) If $\{i\}$ is a proper flat of $\mathrm{M}$, then the isomorphism $
N_E/\langle \mathbf{e}_i \rangle \rightarrow N_{E \setminus \{i\}}
$ induces a bijection $$\text{star}(\mathbf{e}_i,\Sigma_{\mathrm{M},\mathscr{P}}) \longrightarrow \Sigma_{\mathrm{M}_{\{i\}},\mathscr{P}_{\{i\}}}.$$
Under the same assumptions, the stars of $\mathbf{e}_F$ and $\mathbf{e}_i$ in the reduced Bergman fan $\widetilde{\Sigma}_{\mathrm{M},\mathscr{P}}$ admit analogous descriptions.
When $M$ is not a combinatorial geometry, the star of $\mathbf{e}_i$ in $\Sigma_{\mathrm{M},\mathscr{P}}$ is not necessarily a product of smaller Bergman fans. However, when $\mathrm{M}$ is a combinatorial geometry, Proposition \[Star\] shows that the star of every ray in $\Sigma_{\mathrm{M},\mathscr{P}}$ is a product of at most two Bergman fans.
Piecewise linear functions and their convexity {#SectionPLFunctions}
==============================================
{#SectionGroupA1}
Piecewise linear functions on possibly incomplete fans will play an important role throughout the paper. In this section, we prove several general properties concerning convexity of such functions, working with a dual pair free abelian groups $$\langle -,-\rangle: N \times M \longrightarrow \mathbb{Z}, \quad N_\mathbb{R}:=N \otimes_\mathbb{Z} \mathbb{R}, \quad M_\mathbb{R}:=M \otimes_\mathbb{Z} \mathbb{R},$$ and a unimodular fan $\Sigma$ in the latticed vector space $N_\mathbb{R}$. The set of primitive ray generators of $\Sigma$ will be denoted $V_\Sigma$.
We say that a function $\ell:|\Sigma| \rightarrow \mathbb{R}$ is *piecewise linear* if it is continuous and the restriction of $\ell$ to any cone in $\Sigma$ is the restriction of a linear function on $N_{\mathbb{R}}$. The function $\ell$ is said to be *integral* if $$\ell\big( |\Sigma| \cap N\big) \subseteq \mathbb{Z},$$ and the function $\ell$ is said to be *positive* if $$\ell\big( |\Sigma| \setminus \{0\}\big) \subseteq \mathbb{R}_{>0}.$$
An important example of a piecewise linear function on $\Sigma$ is the *Courant function* $x_\mathbf{e}$ associated to a primitive ray generator $\mathbf{e}$ of $\Sigma$, whose values at $V_\Sigma$ are given by the Kronecker delta function. Since $\Sigma$ is unimodular, the Courant functions are integral, and they form a basis of the group of integral piecewise linear functions on $\Sigma$: $$\text{PL}(\Sigma)=\Bigg\{ \sum_{\mathbf{e} \in V_\Sigma} c_\mathbf{e} \hspace{0.5mm}x_\mathbf{e}
\mid c_\mathbf{e} \in \mathbb{Z} \Bigg\} \simeq \mathbb{Z}^{V_{\Sigma}}.$$ An integral linear function on $N_\mathbb{R}$ restricts to an integral piecewise linear function on $\Sigma$, giving a homomorphism $$\text{res}_\Sigma: M \longrightarrow \text{PL}(\Sigma), \qquad m \longmapsto \sum_{\mathbf{e} \in V_\Sigma} \langle \mathbf{e},m \rangle \hspace{0.5mm} x_\mathbf{e}.$$ We denote the cokernel of the restriction map by $$A^1(\Sigma):=\text{PL}(\Sigma)/M.$$ In general, this group may have torsion, even under our assumption that $\Sigma$ is unimodular. When integral piecewise linear functions $\ell$ and $\ell'$ on $\Sigma$ differ by the restriction of an integral linear function on $N_\mathbb{R}$, we say that $\ell$ and $\ell'$ are *equivalent* over $\mathbb{Z}$.
Note that the group of piecewise linear functions modulo linear functions on $\Sigma$ can be identified with the tensor product $$A^1(\Sigma)_\mathbb{R}:=A^1(\Sigma) \otimes_\mathbb{Z} \mathbb{R}.$$ When piecewise linear functions $\ell$ and $\ell'$ on $\Sigma$ differ by the restriction of a linear function on $N_\mathbb{R}$, we say that $\ell$ and $\ell'$ are *equivalent*.
We describe three basic pullback homomorphisms between the groups $A^1$. Let $\Sigma'$ be a subfan of $\Sigma$, and let $\sigma$ be a cone in $\Sigma$.
(1) The restriction of functions from $\Sigma$ to $\Sigma'$ defines a surjective homomorphism $$\text{PL}(\Sigma) \longrightarrow \text{PL}(\Sigma'),$$ and this descends to a surjective homomorphism $$\textrm{p}_{\Sigma' \subseteq \Sigma}: A^1(\Sigma) \longrightarrow A^1(\Sigma').$$ In terms of Courant functions, $\textrm{p}_{\Sigma' \subseteq \Sigma}$ is uniquely determined by its values $$x_\mathbf{e} \longmapsto \begin{cases} x_{\mathbf{e}} & \text{if $\mathbf{e}$ is in $V_{\Sigma'}$,} \\ 0& \text{if otherwise}. \end{cases}$$
(2) Any integral piecewise linear function $\ell$ on $\Sigma$ is equivalent over $\mathbb{Z}$ to an integral $\ell'$ that is zero on $\sigma$, and the choice of such $\ell'$ is unique up to an integral linear function on $N_\mathbb{R}/\langle \sigma\rangle$. Therefore we have a surjective homomorphism $$\textrm{p}_{\sigma \in \Sigma}: A^1(\Sigma) \longrightarrow A^1(\text{star}(\sigma,\Sigma)),$$ uniquely determined by its values on $x_{\mathbf{e}}$ for primitive ray generators $\mathbf{e}$ not contained in $\sigma$: $$x_\mathbf{e} \longmapsto \begin{cases} x_{\overline{\mathbf{e}}} & \text{if there is a cone in $\Sigma$ containing $\mathbf{e}$ and $\sigma$,} \\ 0& \text{if otherwise}. \end{cases}$$ Here $\overline{\mathbf{e}}$ is the image of $\mathbf{e}$ in the quotient space $N_\mathbb{R}/\langle \sigma\rangle$.
(3) A piecewise linear function on the product of two fans $\Sigma_1 \times \Sigma_2$ is the sum of its restrictions to the subfans $$\Sigma_1 \times \{0\} \subseteq \Sigma_1 \times \Sigma_2 \ \ \text{and} \ \ \{0\} \times \Sigma_2 \subseteq \Sigma_1 \times \Sigma_2.$$ Therefore we have an isomorphism $$\text{PL}(\Sigma_1 \times \Sigma_2) \simeq \text{PL}(\Sigma_1) \oplus \text{PL}(\Sigma_2),$$ and this descends to an isomorphism $$\textrm{p}_{\Sigma_1,\Sigma_2}: A^1(\Sigma_1 \times \Sigma_2) \simeq A^1(\Sigma_1) \oplus A^1(\Sigma_2).$$
{#section-9}
We define the *link* of a cone $\sigma$ in $\Sigma$ to be the collection $$\text{link}(\sigma,\Sigma):=\big\{\sigma' \in \Sigma \mid \text{$\sigma'$ is contained in a cone in $\Sigma$ containing $\sigma$, and $\sigma \cap \sigma'=\{0\}$} \big\}.$$ Note that the link of $\sigma$ in $\Sigma$ is a subfan of $\Sigma$.
\[DefinitionConvexity\] Let $\ell$ be a piecewise linear function on $\Sigma$, and let $\sigma$ be a cone in $\Sigma$.
(1) The function $\ell$ is *convex* around $\sigma$ if it is equivalent to a piecewise linear function that is zero on $\sigma$ and nonnegative on the link of $\sigma$.
(2) The function $\ell$ is *strictly convex* around $\sigma$ if it is equivalent to a piecewise linear function that is zero on $\sigma$ and positive on the link of $\sigma$.
The function $\ell$ is *convex* if it is convex around every cone in $\Sigma$, and *strictly convex* if it is strictly convex around every cone in $\Sigma$.
When $\Sigma$ is complete, the function $\ell$ is convex in the sense of Definition \[DefinitionConvexity\] if and only if it is convex in the usual sense: $$\ell(\mathbf{u}_1+\mathbf{u}_2) \le \ell(\mathbf{u}_1)+\ell(\mathbf{u}_2) \ \ \text{for} \ \ \mathbf{u}_1, \mathbf{u}_2 \in N_\mathbb{R}.$$ In general, writing $\iota$ for the inclusion of the torus orbit closure corresponding to $\sigma$ in the toric variety of $\Sigma$, we have $$\text{$\ell$ is convex around $\sigma$} \Longleftrightarrow \text{$\iota^*$ of the class of the divisor associated to $\ell$ is effective}.$$ For a detailed discussion and related notions of convexity from the point of view of toric geometry, see [@Gibney-Maclagan].
The *ample cone* of $\Sigma$ is the open convex cone $$\mathscr{K}_\Sigma:=\big\{\text{classes of strictly convex piecewise linear functions on $\Sigma$}\big\} \subseteq A^1(\Sigma)_\mathbb{R}.$$ The *nef cone* of $\Sigma$ is the closed convex cone $$\mathscr{N}_\Sigma:=\big\{\text{classes of convex piecewise linear functions on $\Sigma$}\big\} \subseteq A^1(\Sigma)_\mathbb{R}.$$
Note that the closure of the ample cone $\mathscr{K}_\Sigma$ is contained in the nef cone $\mathscr{N}_\Sigma$. In many interesting cases, the reverse inclusion also holds.
\[Kleiman\] If $\mathscr{K}_\Sigma$ is nonempty, then $\mathscr{N}_\Sigma$ is the closure of $\mathscr{K}_\Sigma$.
If $\ell_1$ is a convex piecewise linear function and $\ell_2$ is strictly convex piecewise linear function on $\Sigma$, then the sum $\ell_1+\epsilon \hspace{0.5mm} \ell_2$ is strictly convex for every positive number $\epsilon$. This shows that the nef cone of $\Sigma$ is in the closure of the ample cone of $\Sigma$.
We record here that the various pullbacks of an ample class are ample. The proof is straightforward from Definition \[DefinitionConvexity\].
\[PropositionAmplePullback\] Let $\Sigma'$ be a subfan of $\Sigma$, $\sigma$ be a cone in $\Sigma$, and let $\Sigma_1 \times \Sigma_2$ be a product fan.
(1) The pullback homomorphism $\textrm{p}_{\Sigma' \subseteq\Sigma}$ induces a map between the ample cones $$\mathscr{K}_{\Sigma} \longrightarrow \mathscr{K}_{\Sigma'}.$$
(2) The pullback homomorphism $\textrm{p}_{\sigma \in \Sigma}$ induces a map between the ample cones $$\mathscr{K}_\Sigma \longrightarrow \mathscr{K}_{\text{star}(\sigma,\Sigma)}.$$
(3) The isomorphism $\textrm{p}_{\Sigma_1,\Sigma_2}$ induces a bijective map between the ample cones $$\mathscr{K}_{\Sigma_1 \times \Sigma_2} \longrightarrow \mathscr{K}_{\Sigma_1} \times \mathscr{K}_{\Sigma_2}.$$
It follows from the first item that any subfan of the normal fan of a polytope has a nonempty ample cone. In particular, by Proposition \[Subfan\], the Bergman fan $\Sigma_{\mathrm{M},\mathscr{P}}$ has a nonempty ample cone.
Strictly convex piecewise linear functions on the normal fan of the permutohedron can be described in a particularly nice way: A piecewise linear function on $\Sigma_{\mathscr{P}(E)}$ is strictly convex if and only if it is of the form $$\sum_{F \in \mathscr{P}(E)} c_F x_F, \ \ \text{$c_{F_1}+c_{F_2} > c_{F_1 \cap F_2} +c_{F_1 \cup F_2}$ for any incomparable $F_1,F_2$, with $c_\varnothing=c_E=0$}.$$ For this and related results, see [@Batyrev-Blume]. The restriction of any such *strictly submodular function* gives a strictly convex function on the Bergman fan $\Sigma_\mathrm{M}$, and defines an ample class on $\Sigma_\mathrm{M}$.
{#section-10}
We specialize to the case of matroids and prove basic properties of convex piecewise linear functions on the Bergman fan $\Sigma_{\mathrm{M},\mathscr{P}}$. We write $\mathscr{K}_{\mathrm{M},\mathscr{P}}$ for the ample cone of $\Sigma_{\mathrm{M},\mathscr{P}}$, and $\mathscr{N}_{\mathrm{M},\mathscr{P}}$ for the nef cone of $\Sigma_{\mathrm{M},\mathscr{P}}$.
\[MatroidKleiman\] Let $\mathrm{M}$ be a loopless matroid on $E$, and let $\mathscr{P}$ be an order filter of $\mathscr{P}(\mathrm{M})$.
(1) The nef cone of $\Sigma_{\mathrm{M},\mathscr{P}}$ is equal to the closure of the ample cone of $\Sigma_{\mathrm{M},\mathscr{P}}$: $$\overline{\mathscr{K}_{\mathrm{M},\mathscr{P}}}=\mathscr{N}_{\mathrm{M},\mathscr{P}}.$$
(2) The ample cone of $\Sigma_{\mathrm{M},\mathscr{P}}$ is equal to the interior of the nef cone of $\Sigma_{\mathrm{M},\mathscr{P}}$: $$\mathscr{K}_{\mathrm{M},\mathscr{P}}=\mathscr{N}^\circ_{\mathrm{M},\mathscr{P}}.$$
Propositions \[Subfan\] shows that the ample cone $\mathscr{K}_{\mathrm{M},\mathscr{P}}$ is nonempty. Therefore, by Proposition \[Kleiman\], the nef cone $\mathscr{N}_{\mathrm{M},\mathscr{P}}$ is equal to the closure of $\mathscr{K}_{\mathrm{M},\mathscr{P}}$.
The second assertion can be deduced from the first using the following general property of convex sets: An open convex set is equal to the interior of its closure.
The main result here is that the ample cone and its ambient vector space $$\mathscr{K}_{\mathrm{M},\mathscr{P}} \subseteq A^1(\Sigma_{\mathrm{M},\mathscr{P}})_\mathbb{R}$$ depend only on $\mathscr{P}$ and the combinatorial geometry of $\mathrm{M}$, see Proposition \[CombinatorialGeometry\] below. We set $$\overline{E}:=\big\{ A \mid \text{$A$ is a rank $1$ flat of $\mathrm{M}$}\big\}.$$
The *combinatorial geometry* of $\mathrm{M}$ is the simple matroid $\overline{\mathrm{M}}$ on $\overline{E}$ determined by its poset of nonempty proper flats $\mathscr{P}(\overline{\mathrm{M}}) = \mathscr{P}(\mathrm{M})$.
The set of primitive ray generators of $\Sigma_{\mathrm{M},\mathscr{P}}$ is the disjoint union $$\big\{\mathbf{e}_i \mid \text{the closure of $i$ in $\mathrm{M}$ is not in $\mathscr{P}$}\big\}
\cup \big\{ \mathbf{e}_F \mid \text{$F$ is a flat in $\mathscr{P}$}\big\} \subseteq N_{E,\mathbb{R}},$$ and the set of primitive ray generators of $\Sigma_{\overline{\mathrm{M}},\mathscr{P}}$ is the disjoint union $$\big\{\mathbf{e}_A \mid \text{$A$ is a rank $1$ flat of $\mathrm{M}$ not in $\mathscr{P}$}\big\}
\cup \big\{ \mathbf{e}_{F} \mid \text{$F$ is a flat in $\mathscr{P}$}\big\}\subseteq N_{\overline{E},\mathbb{R}}.$$ The corresponding Courant functions on the Bergman fans will be denoted $x_i$, $x_F$, and $x_A$, $x_F$ respectively.
Let $\pi$ be the surjective map between the ground sets of $\mathrm{M}$ and $\overline{\mathrm{M}}$ given by the closure operator of $\mathrm{M}$. We fix an arbitrary section $\iota$ of $\pi$ by choosing an element from each rank $1$ flat: $$\pi: E \longrightarrow \overline{E}, \qquad \iota: \overline{E} \longrightarrow E, \qquad \pi\circ\iota=\text{id}.$$ The maps $\pi$ and $\iota$ induce the horizontal homomorphisms in the diagram $$\xymatrixcolsep{5pc}
\xymatrixrowsep{3pc}
\xymatrix{
\text{PL}(\Sigma_{\mathrm{M},\mathscr{P}}) \ar@<0.5ex>[r]^{\pi_{\text{PL}}} & \text{PL}(\Sigma_{\overline{\mathrm{M}},\mathscr{P}}) \ar@<0.5ex>[l]^{\iota_{\text{PL}}}\\
M_E \ar@<0.5ex>[r]^{\pi_M} \ar[u]^{\text{res}} & M_{\overline{E}}, \ar[u]_{\text{res}} \ar@<0.5ex>[l]^{\iota_M}
}$$ where the homomorphism $\pi_{\text{PL}}$ is obtained by setting $$x_i \longmapsto x_{\pi(i)}, \quad x_F \longmapsto x_F, \quad \text{for elements $i$ whose closure is not in $\mathscr{P}$, and for flats $F$ in $\mathscr{P}$,
}$$ and the homomorphism $\iota_{\text{PL}}$ is obtained by setting $$x_A \longmapsto x_{\iota(A)}, \quad x_F \longmapsto x_F, \quad \text{for rank $1$ flats $A$ not in $\mathscr{P}$, and for flats $F$ in $\mathscr{P}$.}$$ In the diagram above, we have $$\pi_{\text{PL}} \circ \text{res}= \text{res} \circ \pi_M, \quad
\iota_{\text{PL}} \circ \text{res}= \text{res} \circ \iota_M, \quad
\pi_{\text{PL}} \circ \iota_{\text{PL}}=\text{id}, \quad
\pi_{M} \circ \iota_{M}=\text{id}.$$
\[PropositionSimplification\] The homomorphism $\pi_{\text{PL}}$ induces an isomorphism $$\underline{\pi}_{\text{PL}}: A^1(\Sigma_{\mathrm{M},\mathscr{P}}) \longrightarrow A^1(\Sigma_{\overline{\mathrm{M}},\mathscr{P}}).$$ The homomorphism $\iota_{\text{PL}}$ induces the inverse isomorphism $$\underline{\iota}_{\text{PL}}: A^1(\Sigma_{\overline{\mathrm{M}},\mathscr{P}}) \longrightarrow A^1(\Sigma_{\mathrm{M},\mathscr{P}}).$$
We use the same symbols to denote the isomorphisms $
A^1(\Sigma_{\mathrm{M},\mathscr{P}})_\mathbb{R} \leftrightarrows A^1(\Sigma_{\overline{\mathrm{M}},\mathscr{P}})_\mathbb{R}.
$
It is enough to check that the composition $\underline{\iota}_{\text{PL}} \circ \underline{\pi}_{\text{PL}}$ is the identity. Let $i$ and $j$ be elements whose closures are not in $\mathscr{P}$. Consider the linear function on $N_{E,\mathbb{R}}$ given by the integral vector $$\mathbf{e}_i-\mathbf{e}_j \in M_E.$$ The restriction of this linear function to $\Sigma_{M,\mathscr{P}}$ is the linear combination $$\text{res}(\mathbf{e}_i-\mathbf{e}_j)=\Big(x_i+\sum_{i \in F \in \mathscr{P}} x_F\Big)-\Big(x_j+\sum_{j \in F \in \mathscr{P}} x_F\Big).$$ If $i$ and $j$ have the same closure, then a flat contains $i$ if and only if it contains $j$, and hence the linear function witnesses that the piecewise linear functions $x_i$ and $x_j$ are equivalent over $\mathbb{Z}$. It follows that $\underline{\iota}_{\text{PL}} \circ \underline{\pi}_{\text{PL}}=\text{id}$.
The maps $\pi$ and $\iota$ induce simplicial maps between the Bergman complexes $$\xymatrix{
\Delta_{\mathrm{M},\mathscr{P}} \ar@<0.5ex>[rr]^{\pi_\Delta} && \Delta_{\overline{\mathrm{M}},\mathscr{P}} \ar@<0.5ex>[ll]^{\iota_\Delta},
} \qquad
\vartriangle_{I< \mathscr{F}}\ \longmapsto\ \vartriangle_{\pi(I)< \mathscr{F}},
\quad
\vartriangle_{\mathscr{I}< \mathscr{F}}\ \longmapsto\ \vartriangle_{\iota(\mathscr{I})< \mathscr{F}}.$$ The simplicial map $\pi_\Delta$ collapses those simplices containing vectors of parallel elements, and $$\pi_\Delta \circ \iota_\Delta=\text{id}.$$ The other composition $\iota_\Delta \circ \pi_\Delta$ is a deformation retraction. For this note that $$\vartriangle_{I<\mathscr{F}} \ \in \Delta_{\mathrm{M},\mathscr{P}} \Longrightarrow \iota_\Delta \circ \pi_\Delta (\vartriangle_{I<\mathscr{F}}) \ \cup \vartriangle_{I<\mathscr{F}} \ \subseteq \ \vartriangle_{\pi^{-1} \pi I<\mathscr{F}.}$$ The simplex $\vartriangle_{\pi^{-1} \pi I<\mathscr{F}}$ is in $\Delta_{\mathrm{M},\mathscr{P}}$, and hence we can find a homotopy $\iota_\Delta \circ \pi_\Delta \simeq \text{id}$.
\[CombinatorialGeometry\] The isomorphism $\underline{\pi}_{\text{PL}}$ restricts to a bijective map between the ample cones $$\mathscr{K}_{{\mathrm{M},\mathscr{P}}} \longrightarrow \mathscr{K}_{{\overline{\mathrm{M}},\mathscr{P}}}.$$
By Proposition \[MatroidKleiman\], it is enough to show that $\underline{\pi}_{\text{PL}}$ restricts to a bijective map $$\mathscr{N}_{{\mathrm{M},\mathscr{P}}} \longrightarrow \mathscr{N}_{{\overline{\mathrm{M}},\mathscr{P}}}.$$ We use the following maps corresponding to $\pi_\Delta$ and $\iota_\Delta$: $$\xymatrix{
\Sigma_{\mathrm{M},\mathscr{P}} \ar@<0.5ex>[rr]^{\pi_\Sigma} && \Sigma_{\overline{\mathrm{M}},\mathscr{P}} \ar@<0.5ex>[ll]^{\iota_\Sigma},
} \qquad
\sigma_{I< \mathscr{F}}\ \longmapsto\ \sigma_{\pi(I)< \mathscr{F}},
\quad
\sigma_{\mathscr{I}< \mathscr{F}}\ \longmapsto\ \sigma_{\iota(\mathscr{I})< \mathscr{F}}.$$
One direction is more direct: The homomorphism $\iota_{\text{PL}}$ maps a convex piecewise linear function $\overline{\ell}$ to a convex piecewise linear function $\iota_{\text{PL}}(\overline{\ell})$. Indeed, for any cone $\sigma_{I<\mathscr{F}}$ in $\Sigma_{\mathrm{M},\mathscr{P}}$, $$\begin{gathered}
\Big(\text{$\overline{\ell}$ is zero on $\sigma_{\pi(I)<\mathscr{F}}$ and nonnegative on the link of $\sigma_{\pi(I)<\mathscr{F}}$ in $\Sigma_{\overline{\mathrm{M}},\mathscr{P}}$}\Big) \Longrightarrow\\
\Big(\text{$\iota_{\text{PL}}(\overline{\ell})$ is zero on $\sigma_{\pi^{-1}\pi(I)<\mathscr{F}}$ and nonnegative on the link of $\sigma_{\pi^{-1}\pi(I)<\mathscr{F}}$ in $\Sigma_{\mathrm{M},\mathscr{P}}$}\Big)\\
\Longrightarrow
\Big(\text{$\iota_{\text{PL}}(\overline{\ell})$ is zero on $\sigma_{I<\mathscr{F}}$ and nonnegative on the link of $\sigma_{I<\mathscr{F}}$ in $\Sigma_{\mathrm{M},\mathscr{P}}$}\Big).\end{gathered}$$
Next we show the other direction: The homomorphism $\pi_{\text{PL}}$ maps a convex piecewise linear function $\ell$ to a convex piecewise linear function $\pi_{\text{PL}}(\ell)$. The main claim is that, for any cone $\sigma_{\mathscr{I}<\mathscr{F}}$ in $\Sigma_{\overline{\mathrm{M}},\mathscr{P}}$, $$\text{$\ell$ is convex around $\sigma_{\pi^{-1}(\mathscr{I})<\mathscr{F}}$}\Longrightarrow \text{ $\pi_{\text{PL}}(\ell)$ is convex around $\sigma_{\mathscr{I}<\mathscr{F}}$.}$$ This can be deduced from the following identities between the subfans of $\Sigma_{\mathrm{M},\mathscr{P}}$: $$\begin{aligned}
\pi_\Sigma^{-1}\Big( \text{the set of all faces of $\sigma_{\mathscr{I}<\mathscr{F}}$}\Big)&=
\Big( \text{the set of all faces of $\sigma_{\pi^{-1}(\mathscr{I})<\mathscr{F}}$}\Big), \\
\pi_\Sigma^{-1} \Big( \text{the link of $\sigma_{\mathscr{I}<\mathscr{F}}$ in $\Sigma_{\overline{\mathrm{M}},\mathscr{P}}$}\Big)&=
\Big( \text{the link of $\sigma_{\pi^{-1}(\mathscr{I})<\mathscr{F}}$ in $\Sigma_{\mathrm{M},\mathscr{P}}$}\Big).\end{aligned}$$ It is straightforward to check the two equalities from the definitions of $\Sigma_{\mathrm{M},\mathscr{P}}$ and $\Sigma_{\overline{\mathrm{M}},\mathscr{P}}$.
Note that a Bergman fan and the corresponding reduced Bergman fan share the same set of primitive ray generators. Therefore we have isomorphisms
$$\xymatrix{
A^1(\Sigma_{\mathrm{M},\mathscr{P}}) \ar@<0.5ex>[r] \ar@<0.5ex>[d] & A^1(\Sigma_{\overline{\mathrm{M}},\mathscr{P}}) \ar@<0.5ex>[d] \ar@<0.5ex>[l]\\
A^1(\widetilde{\Sigma}_{\mathrm{M},\mathscr{P}}) \ar@<0.5ex>[r]\ar@<0.5ex>[u]& A^1(\widetilde{\Sigma}_{\overline{\mathrm{M}},\mathscr{P}}). \ar@<0.5ex>[l] \ar@<0.5ex>[u]
}$$ We remark that there are inclusion maps between the corresponding ample cones $$\xymatrix{
\mathscr{K}_{\mathrm{M},\mathscr{P}} \ar@{=}[r] \ar@{->}[d]& \mathscr{K}_{\overline{\mathrm{M}},\mathscr{P}} \ar@{->}[d] \\
\widetilde{\mathscr{K}}_{\mathrm{M},\mathscr{P}} & \ar@{->}[l]\widetilde{\mathscr{K}}_{\overline{\mathrm{M}},\mathscr{P}}.
}$$ In general, all three inclusions shown above may be strict.
Homology and cohomology
=======================
{#section-11}
Let $\Sigma$ be a unimodular fan in an $n$-dimensional latticed vector space $N_\mathbb{R}$, and let $\Sigma_k$ be the set of $k$-dimensional cones in $\Sigma$. If $\tau$ is a codimension $1$ face of a unimodular cone $\sigma$, we write $$\mathbf{e}_{\sigma/\tau}:=\text{the primitive generator of the unique $1$-dimensional face of $\sigma$ not in $\tau$}.$$
A *$k$-dimensional Minkowski weight* on $\Sigma$ is a function $$\omega: \Sigma_k \longrightarrow \mathbb{Z}$$ which satisfies the *balancing condition*: For every $(k-1)$-dimensional cone $\tau$ in $\Sigma$, $$\sum_{\tau \subset \sigma} \omega(\sigma) \hspace{0.5mm} \mathbf{e}_{\sigma/\tau} \ \text{is contained in the subspace generated by $\tau$}.$$ The *group of Minkowski weights* on $\Sigma$ is the group $$\text{MW}_*(\Sigma):=\bigoplus_{k \in \mathbb{Z}} \text{MW}_k(\Sigma),$$ where $\text{MW}_k(\Sigma):=\big\{\text{$k$-dimensional Minkowski weights on $\Sigma$}\big\} \subseteq \mathbb{Z}^{\Sigma_k}$.
The group of Minkowski weights was studied by Fulton and Sturmfels in the context of toric geometry [@Fulton-Sturmfels]. An equivalent notion of stress space was independently pursued by Lee in [@Lee]. We record here some immediate properties of the group of Minkowski weights on $\Sigma$.
(1) The group $\text{MW}_0(\Sigma)$ is canonically isomorphic to the group of integers: $$\text{MW}_0(\Sigma) =\mathbb{Z}^{\Sigma_0} \simeq \mathbb{Z}.$$
(2) The group $\text{MW}_1(\Sigma)$ is perpendicular to the image of the restriction map from $M$: $$\text{MW}_1(\Sigma)=\text{im}(\text{res}_\Sigma)^\perp \subseteq \mathbb{Z}^{\Sigma_1}.$$
(3) The group $\text{MW}_k(\Sigma)$ is trivial for $k$ negative or $k$ larger than the dimension of $\Sigma$.
If $\Sigma$ is in addition complete, then an $n$-dimensional weight on $\Sigma$ satisfies the balancing condition if and only if it is constant. Therefore, in this case, there is a canonical isomorphism $$\text{MW}_n(\Sigma) \simeq \mathbb{Z}.$$ We show that the Bergman fan $\Sigma_{\mathrm{M}}$ has the same property with respect to its dimension $r$.
\[PropositionMatroidBalancing\] An $r$-dimensional weight on $\Sigma_\mathrm{M}$ satisfies the balancing condition if and only if it is constant.
It follows that there is a canonical isomorphism $
\text{MW}_r(\Sigma_\mathrm{M}) \simeq \mathbb{Z}.
$
\[LemmaConnectedness\] The Bergman fan $\Sigma_\mathrm{M}$ is connected in codimension $1$.
We remark that Lemma \[LemmaConnectedness\] is a direct consequence of the shellability of $\Delta_\mathrm{M}$, see [@Bjorner].
The claim is that, for any two $r$-dimensional cones $\sigma_\mathscr{F}, \sigma_\mathscr{G}$ in $\Sigma_\mathrm{M}$, there is a sequence $$\sigma_\mathscr{F}= \sigma_0 \supset \tau_1 \subset \sigma_1 \supset \cdots \subset \sigma_{l-1} \supset \tau_l \subset \sigma_l = \sigma_\mathscr{G},$$ where $\tau_i$ is a common facet of $\sigma_{i-1}$ and $\sigma_i$ in $\Sigma_\mathrm{M}$. We express this by writing $\sigma_\mathscr{F} \sim \sigma_\mathscr{G}$.
We prove by induction on the rank of $\mathrm{M}$. If $\text{min}\ \mathscr{F}=\text{min}\ \mathscr{G}$, then the induction hypothesis applied to $\mathrm{M}_{\text{min}\hspace{0.5mm} \mathscr{F}}$ shows that $$\sigma_\mathscr{F} \sim \sigma_{\mathscr{G}}.$$ If otherwise, we choose a flag of nonempty proper flats $\mathscr{H}$ maximal among those satisfying $\text{min}\ \mathscr{F} \cup \text{min}\ \mathscr{G}<\mathscr{H}$. By the induction hypothesis applied to $\mathrm{M}_{\text{min}\hspace{0.5mm} \mathscr{F}}$, we have $$\sigma_\mathscr{F} \sim \sigma_{\{\text{min}\hspace{0.5mm} \mathscr{F}\} \cup \mathscr{H}}.$$ Similarly, by the induction hypothesis applied to $\mathrm{M}_{\text{min}\hspace{0.5mm} \mathscr{G}}$, we have $$\sigma_\mathscr{G} \sim \sigma_{\{\text{min}\hspace{0.5mm} \mathscr{G}\} \cup \mathscr{H}}.$$ Since any $1$-dimensional fan is connected in codimension $1$, this complete the induction.
The proof is based on the *flat partition property* for matroids $\mathrm{M}$ on $E$: $$\text{If $F$ is a flat of $\mathrm{M}$, then the flats of $\mathrm{M}$ that cover $F$ partition $E \setminus F$.}$$ Let $\tau_\mathscr{G}$ be a codimension $1$ cone in the Bergman fan $\Sigma_\mathrm{M}$, and set $$V_{\text{star}(\mathscr{G})}:= \text{the set of primitive ray generators of the star of
$\tau_\mathscr{G}$ in $\Sigma_\mathrm{M}$} \subseteq N_{E,\mathbb{R}}/\langle \tau_\mathscr{F}\rangle.$$ The flat partition property applied to the restrictions of $\mathrm{M}$ shows that, first, the sum of all the vectors in $V_{\text{star}(\mathscr{G})}$ is zero and, second, any proper subset of $V_{\text{star}(\mathscr{G})}$ is linearly independent. Therefore, for an $r$-dimensional weight $\omega$ on $\Sigma_\mathrm{M}$, $$\text{$\omega$ satisfies the balancing condition at $\tau_\mathscr{G}$} \Longleftrightarrow
\text{$\omega$ is constant on cones containing $\tau_{\mathscr{G}}$.}$$ By the connectedness of Lemma \[LemmaConnectedness\], the latter condition for every $\tau_\mathscr{G}$ implies that $\omega$ is constant.
{#section-12}
We continue to work with a unimodular fan $\Sigma$ in $N_\mathbb{R}$. As before, we write $V_\Sigma$ for the set of primitive ray generators of $\Sigma$. Let $S_\Sigma$ be the polynomial ring over $\mathbb{Z}$ with variables indexed by $V_\Sigma$: $$S_\Sigma:=\mathbb{Z}[x_\mathbf{e} ]_{\mathbf{e} \in V_\Sigma}.$$ For each $k$-dimensional cone $\sigma$ in $\Sigma$, we associate a degree $k$ square-free monomial $$x_\sigma:=\prod_{\mathbf{e} \in \sigma} x_\mathbf{e}.$$ The subgroup of $S_\Sigma$ generated by all such monomials $x_\sigma$ will be denoted $$Z^k(\Sigma):=\bigoplus_{\sigma \in \Sigma_k} \mathbb{Z}\hspace{0.5mm} x_\sigma.$$ Let $Z^*(\Sigma)$ be the sum of $Z^k(\Sigma)$ over all nonnegative integers $k$.
The *Chow ring* of $\Sigma$ is the commutative graded algebra $$A^*(\Sigma):=S_\Sigma/(I_\Sigma+J_\Sigma),$$ where $I_\Sigma$ and $J_\Sigma$ are the ideals of $S_\Sigma$ defined by $$\begin{aligned}
I_\Sigma&:=\text{the ideal generated by the square-free monomials not in $Z^*(\Sigma)$,}\\
J_\Sigma&:=\text{the ideal generated by the linear forms $\sum_{\mathbf{e} \in V_\Sigma} \langle \mathbf{e}, m \rangle \hspace{0.5mm} x_\mathbf{e}$ for $m \in M$.}\end{aligned}$$ We write $A^k(\Sigma)$ for the degree $k$ component of $A^*(\Sigma)$, and set $$A^*(\Sigma)_\mathbb{R}:=A^*(\Sigma) \otimes_\mathbb{Z} \mathbb{R} \quad \text{and} \quad A^k(\Sigma)_\mathbb{R}:=A^k(\Sigma)\otimes_\mathbb{Z} \mathbb{R}.$$
If we identify the variables of $S_\Sigma$ with the Courant functions on $\Sigma$, then the degree $1$ component of $A^*(\Sigma)$ agrees with the group introduced in Section \[SectionPLFunctions\]: $$A^1(\Sigma)=\text{PL}(\Sigma)/M.$$ Note that the pullback homomorphisms between $A^1$ introduced in that section uniquely extend to graded ring homomorphisms between $A^*$:
(1) The homomorphism $\textrm{p}_{\Sigma' \subseteq \Sigma}$ uniquely extends to a surjective graded ring homomorphism $$\textrm{p}_{\Sigma' \subseteq \Sigma}: A^*(\Sigma) \longrightarrow A^*(\Sigma').$$
(2) The homomorphism $\textrm{p}_{\sigma \in \Sigma}$ uniquely extends to a surjective graded ring homomorphism $$\textrm{p}_{\sigma \in \Sigma}: A^*(\Sigma) \longrightarrow A^*(\text{star}(\sigma,\Sigma)).$$
(3) The isomorphism $\textrm{p}_{\Sigma_1,\Sigma_2}$ uniquely extends to a graded ring isomorphism $$\textrm{p}_{\Sigma_1,\Sigma_2}: A^*(\Sigma_1 \times \Sigma_2) \longrightarrow A^*(\Sigma_1) \otimes_\mathbb{Z} A^*(\Sigma_2).$$
We remark that the Chow ring $A^*(\Sigma)_\mathbb{R}$ can be identified with the ring of piecewise polynomial functions on $\Sigma$ modulo linear functions on $N_\mathbb{R}$, see [@Billera].
\[PropositionGeneration\] The group $A^k(\Sigma)$ is generated by $Z^k(\Sigma)$ for each nonnegative integer $k$.
In particular, if $k$ larger than the dimension of $\Sigma$, then $A^k(\Sigma)=0$.
Let $\sigma$ be a cone in $\Sigma$, let $\mathbf{e}_1,\mathbf{e}_2,\ldots,\mathbf{e}_l$ be its primitive ray generators. and consider a degree $k$ monomial of the form $$x_{\mathbf{e}_1}^{k_1}x_{\mathbf{e}_2}^{k_2} \cdots x_\mathbf{e_l}^{k_l}, \qquad k_1 \ge k_2 \ge \cdots \ge k_l \ge 1.$$ We show that the image of this monomial in $A^k(\Sigma)$ is in the span of $Z^k(\Sigma)$.
We do this by descending induction on the dimension of $\sigma$. If $\text{dim}\ \sigma=k$, there is nothing to prove. If otherwise, we use the unimodularity of $\sigma$ to choose $m \in M$ such that $$\langle \mathbf{e}_1,m\rangle=-1 \ \ \text{and} \ \ \langle \mathbf{e}_2,m\rangle=\cdots=\langle \mathbf{e}_{l},m\rangle=0.$$ This shows that, modulo the relations given by $I_\Sigma$ and $J_\Sigma$, we have $$x_{\mathbf{e}_1}^{k_1}x_{\mathbf{e}_2}^{k_2} \cdots x_\mathbf{e_k}^{k_l}=x_{\mathbf{e}_1}^{k_1-1} x_{\mathbf{e}_2}^{k_2}\cdots x_\mathbf{e_l}^{k_l}\ \sum_{\mathbf{e} \in \text{link}(\sigma)} \langle \mathbf{e},m\rangle \hspace{0.5mm}x_\mathbf{e},$$ where the sum is over the set of primitive ray generators of the link of $\sigma$ in $\Sigma$. The induction hypothesis applies to each of the terms in the expansion of the right-hand side.
The group of $k$-dimensional weights on $\Sigma$ can be identified with the dual of $Z^k(\Sigma)$ under the tautological isomorphism $$\mathrm{t}_\Sigma: \mathbb{Z}^{\Sigma_k} \longrightarrow \text{Hom}_\mathbb{Z}(Z^k(\Sigma),\mathbb{Z}), \qquad
\omega \longmapsto \Big( x_\sigma \longmapsto \omega(\sigma)\Big).$$ By Proposition \[PropositionGeneration\], the target of $\mathrm{t}_\Sigma$ contains $\text{Hom}_\mathbb{Z}(A^k(\Sigma),\mathbb{Z})$ as a subgroup.
\[PropositionBasicDuality\] The isomorphism $\mathrm{t}_\Sigma$ restricts to the bijection between the subgroups $$\text{MW}_k(\Sigma) \longrightarrow \text{Hom}_\mathbb{Z}(A^k(\Sigma),\mathbb{Z}).$$
The bijection in Proposition \[PropositionBasicDuality\] is an analogue of the Kronecker duality homomorphism in algebraic topology. We use it to define the *cap product* $$A^l(\Sigma) \times \text{MW}_k(\Sigma) \longrightarrow \text{MW}_{k-l}(\Sigma), \qquad \xi \cap \omega \hspace{0.5mm}(\sigma) := \mathrm{t}_\Sigma\hspace{0.5mm} \omega \hspace{0.5mm}(\xi \cdot x_\sigma).$$ This makes the group $\text{MW}_*(\Sigma)$ a graded module over the Chow ring $A^*(\Sigma)$.
The homomorphisms from $A^k(\Sigma)$ to $\mathbb{Z}$ bijectively correspond to the homomorphisms from $Z^k(\Sigma)$ to $\mathbb{Z}$ which vanish on the subgroup $$Z^k(\Sigma) \cap (I_\Sigma+J_\Sigma) \subseteq Z^k(\Sigma).$$ The main point is that this subgroup is generated by polynomials of the form $$\Bigg( \sum_{\mathbf{e} \in \text{link}(\tau)} \langle \mathbf{e}, m \rangle \hspace{0.5mm} x_\mathbf{e}\Bigg) x_\tau,$$ where $\tau$ is a $(k-1)$-dimensional cone of $\Sigma$ and $m$ is an element perpendicular to $\langle \tau \rangle$. It follows that a $k$-dimensional weight $\omega$ corresponds to a homomorphism $A^k(\Sigma) \to \mathbb{Z}$ if and only if $$\sum_{\tau \subset \sigma} \omega(\sigma)\hspace{0.5mm} \langle \mathbf{e}_{\sigma/\tau}, m \rangle =0 \ \text{for all $m \in \langle \tau \rangle^\perp$},$$ where the sum is over all $k$-dimensional cones $\sigma$ in $\Sigma$ containing $\tau$. Since $\langle\tau \rangle^{\perp\perp}=\langle \tau \rangle$, the latter condition is equivalent to the balancing condition on $\omega$ at $\tau$.
{#section-13}
The ideals $I_\Sigma$ and $J_\Sigma$ have a particularly simple description when $\Sigma=\Sigma_\mathrm{M}$. In this case, we label the variables of $S_\Sigma$ by the nonempty proper flats of $\mathrm{M}$, and write $$S_{\Sigma}=\mathbb{Z}[x_F]_{F \in \mathscr{P}(\mathrm{M})}.$$ For a flag of nonempty proper flats $\mathscr{F}$, we set $
x_\mathscr{F}= \prod_{F \in \mathscr{F}} x_F.
$
(1) The ideal $I_\Sigma$ is generated by the quadratic monomials $$x_{F_1}x_{F_2},$$ where $F_1$ and $F_2$ are two incomparable nonempty proper flats of $\mathrm{M}$.
(2) The ideal $J_\Sigma$ is generated by the linear forms $$\sum_{i_1 \in F} x_F - \sum_{i_2 \in F} x_F,$$ where $i_1$ and $i_2$ are distinct elements of the ground set $E$.
The quotient ring $A^*(\Sigma_\mathrm{M})$ and its generalizations were studied by Feichtner and Yuzvinsky in [@Feichtner-Yuzvinsky].
\[DefinitionAlphaBeta\] To an element $i$ in $E$, we associate linear forms $$\alpha_{\mathrm{M},i}:=\sum_{i \in F} x_F, \quad \beta_{\mathrm{M},i}:=\sum_{i \notin F} x_F.$$ Their classes in $A^*(\Sigma_\mathrm{M})$, which are independent of $i$, will be written $\alpha_\mathrm{M}$ and $\beta_\mathrm{M}$ respectively.
We show that $A^r(\Sigma_\mathrm{M})$ is generated by the element $\alpha_\mathrm{M}^r$, where $r$ is the dimension of $\Sigma_\mathrm{M}$.
\[PropositionFundamentalClass\] Let $F_1 \subsetneq F_2 \subsetneq \cdots \subsetneq F_k$ be any flag of nonempty proper flats of $\mathrm{M}$.
(1) If the rank of $F_m$ is not $m$ for some $m \le k$, then $$x_{F_1}x_{F_2} \cdots x_{F_{k}} \hspace{0.5mm}\alpha_\mathrm{M}^{r-k}=\hspace{0.5mm} 0 \hspace{0.5mm}\in A^r(\Sigma_\mathrm{M}).$$
(2) If the rank of $F_m$ is $m$ for all $m \le k$, then $$x_{F_1}x_{F_2} \cdots x_{F_{k}} \hspace{0.5mm}\alpha_\mathrm{M}^{r-k}=\alpha_\mathrm{M}^r \in A^r(\Sigma_\mathrm{M}).$$
In particular, for any two maximal flags of nonempty proper flats $\mathscr{F}_1$ and $\mathscr{F}_2$ of $\mathrm{M}$, $$x_{\mathscr{F}_1}=x_{\mathscr{F}_2} \in A^r(\Sigma_\mathrm{M}).$$ Since $\text{MW}_r(\Sigma_\mathrm{M})$ is isomorphic to $\mathbb{Z}$, this implies that $A^r(\Sigma_\mathrm{M})$ is isomorphic to $\mathbb{Z}$, see Proposition \[PropositionDegreeMap\].
As a general observation, we note that for any element $i$ not in a nonempty proper flat $F$, $$x_F \ \alpha_\mathrm{M}= x_F \ \Big( \sum_G x_G\Big) \in A^*(\Sigma_\mathrm{M}),$$ where the sum is over all proper flats containing $F$ and $\{i\}$. In particular, if the rank of $F$ is $r$, then the product is zero.
We prove the first assertion by descending induction on $k$, which is necessarily less than $r$. If $k=r-1$, then the rank of $F_k$ should be $r$, and hence the product is zero. For general $k$, we choose an element $i$ not in $F_k$. By the observation made above, we have $$x_{F_1}x_{F_2} \cdots x_{F_{k}} \hspace{0.5mm}\alpha_\mathrm{M}^{r-k}=
x_{F_1}x_{F_2} \cdots x_{F_{k}} \hspace{0.5mm} \Big(\sum_{G} x_G\Big) \hspace{0.5mm}\alpha_\mathrm{M}^{r-k-1},$$ where the sum is over all proper flats containing $F$ and $\{i\}$. The right-hand side is zero by the induction hypothesis for $k+1$ applied to each of the terms in the expansion.
We prove the second assertion by ascending induction on $k$. When $k=1$, we choose an element $i$ in $F_k$. By the first part of the proposition for $k=1$, we have $$x_{F_1} \hspace{0.5mm}\alpha_\mathrm{M}^{r-1}=\alpha_\mathrm{M}^{r}.$$ For general $k$, we choose an element $i$ in $F_k \setminus F_{k-1}$. By the first part of the proposition for $k$, we have $$x_{F_1}x_{F_2} \cdots x_{F_{k-1}} x_{F_{k}} \hspace{0.5mm}\alpha_\mathrm{M}^{r-k}=
x_{F_1}x_{F_2} \cdots x_{F_{k-1}} \hspace{0.5mm} \Big(\sum_{G} x_G\Big) \hspace{0.5mm}\alpha_\mathrm{M}^{r-k},$$ where the sum is over all proper flats containing $F_{k-1}$ and $\{i\}$. The right-hand side is $\alpha_\mathrm{M}^r$ by the induction hypothesis for $k-1$.
When $\Sigma$ is complete, Fulton and Sturmfels showed in [@Fulton-Sturmfels] that there is an isomorphism $$A^{k}(\Sigma) \longrightarrow \text{MW}_{n-k}(\Sigma), \qquad \xi \longmapsto \big( \sigma \longmapsto \text{deg} \ \xi \cdot x_\sigma \big),$$ where $n$ is the dimension of $\Sigma$ and $``\text{deg}"$ is the degree map of the complete toric variety of $\Sigma$. In Theorem \[PoincareDuality\], we show that there is an isomorphism for the Bergman fan $$A^{k}(\Sigma_{\mathrm{M}}) \longrightarrow \text{MW}_{r-k}(\Sigma_{\mathrm{M}}), \qquad \xi \longmapsto \big( \sigma_\mathscr{F} \longmapsto \text{deg} \ \xi \cdot x_\mathscr{F} \big),$$ where $r$ is the dimension of $\Sigma_\mathrm{M}$ and $``\text{deg}"$ is a homomorphism constructed from $\mathrm{M}$. These isomorphisms are analogues of the Poincaré duality homomorphism in algebraic topology.
\[MatroidDegreeMap\] The *degree map* of $\mathrm{M}$ is the homomorphism obtained by taking the cap product $$\text{deg}: A^r(\Sigma_\mathrm{M}) \longrightarrow \mathbb{Z}, \qquad \xi \longmapsto \xi \cap 1_\mathrm{M},$$ where $1_\mathrm{M}=1$ is the constant $r$-dimensional Minkowski weight on $\Sigma_\mathrm{M}$.
By Proposition \[PropositionGeneration\], the homomorphism $\text{deg}$ is uniquely determined by its property $$\text{deg}(x_\mathscr{F})=1 \ \ \text{for all monomials $x_\mathscr{F}$ corresponding to an $r$-dimensional cone in $\Sigma_\mathrm{M}$.}$$
\[PropositionDegreeMap\] The degree map of $\mathrm{M}$ is an isomorphism.
The second part of Proposition \[PropositionFundamentalClass\] shows that $A^r(\Sigma_\mathrm{M})$ is generated by the element $\alpha_\mathrm{M}^r$, and that $\text{deg} (\alpha_\mathrm{M}^r)=\text{deg}(x_\mathscr{F})=1$.
{#section-14}
We remark on algebraic geometric properties of Bergman fans, working over a fixed field $\mathbb{K}$. For basics on toric varieties, we refer to [@Fulton]. The results of this subsection will be independent from the remainder of the paper.
The main object is the smooth toric variety $X(\Sigma)$ over $\mathbb{K}$ associated to a unimodular fan $\Sigma$ in $N_\mathbb{R}$: $$X(\Sigma):=\bigcup_{\sigma \in \Sigma} \text{Spec} \ \mathbb{K}[\sigma^\vee \cap M].$$ It is known that the Chow ring of $\Sigma$ is naturally isomorphic to the Chow ring of $X(\Sigma)$: $$A^*(\Sigma) \longrightarrow A^*(X({\Sigma})), \qquad x_\sigma \longmapsto [X({\text{star}(\sigma)})].$$ See [@Danilov Section 10] for the proof when $\Sigma$ is complete, and see [@Bifet-DeConcini-Procesi] and [@Brion] for the general case.
A morphism between smooth algebraic varieties $X_1 \to X_2$ is a *Chow equivalence* if the induced homomorphism between the Chow rings $A^*(X_2) \to A^*(X_1)$ is an isomorphism.
In fact, the results of this subsection will be valid for any variety that is locally a quotient of a manifold by a finite group so that $A^*(X) \otimes_\mathbb{Z}\mathbb{Q}$ has the structure of a graded algebra over $\mathbb{Q}$. Matroids provide nontrivial examples of Chow equivalences. For example, consider the subfan ${\widetilde{\Sigma}_{\mathrm{M},\mathscr{P}}}\subseteq {\Sigma_{\mathrm{M},\mathscr{P}}}$ and the corresponding open subset $$X({\widetilde{\Sigma}_{\mathrm{M},\mathscr{P}}})\subseteq X({\Sigma_{\mathrm{M},\mathscr{P}}}).$$ In Proposition \[PropositionOpenInclusion\], we show that the above inclusion is a Chow equivalence for any $\mathrm{M}$ and $\mathscr{P}$.
We remark that, when $\mathbb{K}=\mathbb{C}$, a Chow equivalence need not induce an isomorphism between singular cohomology rings. For example, consider any line in a projective plane minus two points $$\mathbb{C}\mathbb{P}^1 \subseteq \mathbb{C}\mathbb{P}^2 \setminus \{p_1,p_2\}.$$ The inclusion is a Chow equivalence for any two distinct points $p_1,p_2$ outside $\mathbb{C}\mathbb{P}^1$, but the two spaces have different singular cohomology rings.
We show that the notion of Chow equivalence can be used to characterize the realizability of matroids.
\[Chow-equivalence\] There is a Chow equivalence from a smooth projective variety over $\mathbb{K}$ to $X(\Sigma_{\mathrm{M}})$ if and only if the matroid $\mathrm{M}$ is realizable over $\mathbb{K}$.
This is a classical variant of the tropical characterization of the realizability of matroids in [@Katz-Payne]. We write $r$ for the dimension of $\Sigma_\mathrm{M}$, and $n$ for the dimension of $X({\Sigma_\mathrm{M}})$. As before, the ground set of $\mathrm{M}$ will be $E=\{0,1,\ldots,n\}$.
The “if" direction follows from the construction of De Concini-Procesi wonderful models [@DeConcini-Procesi]. Suppose that the loopless matroid $\mathrm{M}$ is realized by a spanning set of nonzero vectors $$\mathscr{R}=\{f_0,f_1,\ldots,f_n\} \subseteq V/\mathbb{K}.$$ The realization $\mathscr{R}$ gives an injective linear map between two projective spaces $$L_\mathscr{R}: \mathbb{P}(V^\vee) \longrightarrow X(\Sigma_\varnothing), \qquad L_\mathscr{R}=[f_0:f_1:\cdots:f_n],$$ where $\Sigma_\varnothing$ is the complete fan in $N_{E,\mathbb{R}}$ corresponding to the empty order filter of $\mathscr{P}(E)$. Note that the normal fan of the $n$-dimensional permutohedron $\Sigma_{\mathscr{P}(E)}$ can be obtained from the normal fan of the $n$-dimensional simplex $\Sigma_\varnothing$ by performing a sequence of stellar subdivisions. In other words, there is a morphism between toric varieties $$\pi: X(\Sigma_{\mathscr{P}(E)}) \longrightarrow X(\Sigma_\varnothing),$$ which is the composition of blowups of torus-invariant subvarieties. To be explicit, consider a sequence of order filters of $\mathscr{P}(E)$ obtained by adding a single subset at a time: $$\varnothing,\ldots, \mathscr{P}_{-},\mathscr{P}_+,\ldots,\mathscr{P}(E) \quad \text{with} \quad \mathscr{P}_+=\mathscr{P}_- \cup \{Z\}.$$ The corresponding sequence of $\Sigma$ interpolates between the collections $\Sigma_\varnothing$ and $\Sigma_{\mathscr{P}(E)}$: $$\Sigma_\varnothing \rightsquigarrow \ldots \rightsquigarrow \Sigma_{\mathscr{P}_{-}}\rightsquigarrow \Sigma_{\mathscr{P}_{+}} \rightsquigarrow \ldots \rightsquigarrow \Sigma_{\mathscr{P}(E)}.$$ The modification in the middle replaces the cones of the form $\sigma_{Z<\mathscr{F}}$ with the sums of the form $$\sigma_{\varnothing<\{Z\}}+\sigma_{I<\mathscr{F}},$$ where $I$ is any proper subset of $Z$. The wonderful model $Y_\mathscr{R}$ associated to $\mathscr{R}$ is by definition the strict transform of $\mathbb{P}(V^\vee)$ under the composition of toric blowups $\pi$. The torus-invariant prime divisors of $X(\Sigma_{\mathscr{P}(E)})$ correspond to nonempty proper subsets of $E$, and those divisors intersecting $Y_\mathscr{R}$ exactly correspond to nonempty proper flats of $\mathrm{M}$. Therefore, the smooth projective variety $Y_\mathscr{R}$ is contained in the open subset $$X(\Sigma_\mathrm{M}) \subseteq X(\Sigma_{\mathscr{P}(E)}).$$ The inclusion $Y_\mathscr{R} \subseteq X(\Sigma_\mathrm{M})$ is a Chow equivalence [@Feichtner-Yuzvinsky Corollary 2].
The “only if” direction follows from computations in $A^*(\Sigma_\mathrm{M})$ made in the previous subsection. Suppose that there is a Chow equivalence from a smooth projective variety $$f: Y \longrightarrow X({\Sigma_\mathrm{M}}).$$ Proposition \[PropositionGeneration\] and Proposition \[PropositionDegreeMap\] show that $$A^r(Y) \simeq A^r(\Sigma_\mathrm{M})\simeq \mathbb{Z} \quad \text{and} \quad A^k(Y) \simeq A^k(\Sigma_\mathrm{M}) \simeq 0 \ \ \text{for all $k$ larger than $r$.}$$ Since $Y$ is complete, the above implies that the dimension of $Y$ is $r$. Let $g$ be the composition $$\xymatrixcolsep{2.5pc}
\xymatrix{
Y \ar[r]^<<<<<{f} & X(\Sigma_\mathrm{M}) \ar[r]^<<<<<{\pi_\mathrm{M}}
&X(\Sigma_\varnothing) \simeq \mathbb{P}^n,
}$$ where $\pi_\mathrm{M}$ is the restriction of the composition of toric blowups $\pi$. We use Proposition \[PropositionFundamentalClass\] to compute the degree of the image $g(Y)\subseteq \mathbb{P}^n$.
For this we note that, for any element $i \in E$, we have $$\pi_\mathrm{M}^{-1} \{z_i=0\}=\bigcup_{i \in F} D_F,$$ where $z_i$ is the homogeneous coordinate of $\mathbb{P}^n$ corresponding to $i$ and $D_F$ is the torus-invariant prime divisor of $X(\Sigma_\mathrm{M})$ corresponding to a nonempty proper flat $F$. All the components of $\pi_\mathrm{M}^{-1} \{z_i=0\}$ appear with multiplicity $1$, and hence $$\pi_\mathrm{M}^*\ \mathcal{O}_{\mathbb{P}^n}(1)=\alpha_\mathrm{M} \in A^1(\Sigma_\mathrm{M}).$$ Hence, under the isomorphism $f^*$ between the Chow rings, the $0$-dimensional cycle $(g^* \mathcal{O}_{\mathbb{P}^n}(1))^r$ is the image of the generator $$(\pi_\mathrm{M}^{*} \ \mathcal{O}_{\mathbb{P}^n}(1))^r=\alpha_\mathrm{M}^r \in A^r(\Sigma_\mathrm{M})\simeq \mathbb{Z}.$$ By the projection formula, the above implies that the degree of the image of $Y$ in $\mathbb{P}^n$ is $1$. In other words, $g(Y)\subseteq \mathbb{P}^n$ is an $r$-dimensional linear subspace defined over $\mathbb{K}$. We express the inclusion in the form $$L_\mathscr{R}: \mathbb{P}(V^\vee) \longrightarrow \mathbb{P}^n, \qquad L_\mathscr{R}=[f_0:f_1:\cdots:f_n].$$
Let $\mathrm{M}'$ be the loopless matroid on $E$ defined by the set of nonzero vectors $\mathscr{R} \subseteq V/\mathbb{K}$. The image of $Y$ in $X(\Sigma_\mathrm{M})$ is the wonderful model $Y_\mathscr{R}$, and hence $$X(\Sigma_{\mathrm{M}'}) \subseteq X(\Sigma_{\mathrm{M}}).$$ Observe that none of the torus-invariant prime divisors of $X(\Sigma_{\mathrm{M}})$ are rationally equivalent to zero. Since $f$ is a Chow equivalence, the observation implies that the torus-invariant prime divisors of $X(\Sigma_{\mathrm{M}'})$ and $X(\Sigma_{\mathrm{M}})$ bijectively correspond to each other. Since a matroid is determined by its set of nonempty proper flats, this shows that $\mathrm{M}=\mathrm{M}'$.
Poincaré duality for matroids {#SectionDecompositionTheorem}
=============================
{#section-15}
The principal result of this section is an analogue of Poincaré duality for $A^*(\Sigma_{\mathrm{M},\mathscr{P}})$, see Theorem \[PoincareDuality\]. We give an alternative description of the Chow ring suitable for this purpose.
\[DefinitionIntermediateChow\] Let $S_{E \cup \mathscr{P}}$ be the polynomial ring over $\mathbb{Z}$ with variables indexed by $E \cup \mathscr{P}$: $$S_{E \cup \mathscr{P}}:=\mathbb{Z}[x_i,x_F]_{i \in E,F\in \mathscr{P}}.$$ The *Chow ring* of $(\mathrm{M},\mathscr{P})$ is the commutative graded algebra $$A^*(\mathrm{M},\mathscr{P}):=S_{E \cup\mathscr{P}}/(\mathscr{I}_1+\mathscr{I}_2+\mathscr{I}_3+\mathscr{I}_4),$$ where $\mathscr{I}_1$, $\mathscr{I}_2$, $\mathscr{I}_3$, $\mathscr{I}_4$ are the ideals of $S_{E \cup \mathscr{P}}$ defined below.
(1) The ideal $\mathscr{I}_{1}$ is generated by the quadratic monomials $$x_{F_1}x_{F_2},$$ where $F_1$ and $F_2$ are two incomparable flats in the order filter $\mathscr{P}$.
(2) The ideal $\mathscr{I}_2$ is generated by the quadratic monomials $$x_{i}\hspace{0.5mm}x_{F},$$ where $F$ is a flat in the order filter $\mathscr{P}$ and $i$ is an element in the complement $E \setminus F$.
(3) The ideal $\mathscr{I}_3$ is generated by the monomials $$\prod_{i \in I} x_i,$$ where $I$ is an independent set of $\mathrm{M}$ whose closure is in $\mathscr{P} \cup \{E\}$.
(4) The ideal $\mathscr{I}_4$ is generated by the linear forms $$\Big(x_{i}+\sum_{i \in F} x_F\Big) -\Big(x_{j}+ \sum_{j \in F} x_F\Big),$$ where $i$ and $j$ are distinct elements of $E$ and the sums are over flats $F$ in $\mathscr{P}$.
When $\mathscr{P}=\mathscr{P}(\mathrm{M})$, we omit $\mathscr{P}$ from the notation and write the Chow ring by $A^*(\mathrm{M})$.
When $\mathscr{P}$ is empty, the relations in $\mathscr{I}_4$ show that all $x_i$ are equal in the Chow ring, and hence $$A^*(\mathrm{M},\varnothing) \simeq \mathbb{Z}[x]/(x^{r+1}).$$ When $\mathscr{P}$ is $\mathscr{P}(\mathrm{M})$, the relations in $\mathscr{I}_3$ show that all $x_i$ are zero in the Chow ring, and hence $$A^*(\mathrm{M}) \simeq A^*(\Sigma_{\mathrm{M}}).$$ In general, if $i$ is an element whose closure is in $\mathscr{P}$, then $x_i$ is zero in the Chow ring. The square-free monomial relations in the remaining set of variables bijectively correspond to the non-faces of the Bergman complex $\Delta_{\mathrm{M},\mathscr{P}}$, and hence $$A^*(\mathrm{M},\mathscr{P})\simeq A^*(\Sigma_{\mathrm{M},\mathscr{P}}).$$ We show that the Chow ring of $(\mathrm{M},\mathscr{P})$ is also isomorphic to the Chow ring of the reduced Bergman fan $\widetilde{\Sigma}_{\mathrm{M},\mathscr{P}}$.
\[PropositionOpenInclusion\] Let $I$ be a subset of $E$, and let $F$ be a flat in an order filter $\mathscr{P}$ of $\mathscr{P}(\mathrm{M})$.
(1) If $I$ has cardinality at least the rank of $F$, then $$\Big( \prod_{i \in I} x_i\Big) x_F =0 \in A^*(\mathrm{M},\mathscr{P}).$$
(2) If $I$ has cardinality at least $r+1$, then $$\prod_{i \in I} x_i=0 \in A^*(\mathrm{M},\mathscr{P}).$$
In other words, the inclusion of the open subset $X({\widetilde{\Sigma}_{\mathrm{M},\mathscr{P}}})\subseteq X({\Sigma_{\mathrm{M},\mathscr{P}}})$ is a Chow equivalence. Since the reduced Bergman fan has dimension $r$, this implies that $$A^k(\mathrm{M},\mathscr{P})=0 \ \ \text{for $k>r$.}$$
For the first assertion, we use complement relations in $\mathscr{I}_2$ to reduce to the case when $I \subseteq F$. We prove by induction on the difference between the rank of $F$ and the rank of $I$.
When the difference is zero, $I$ contains a basis of $F$, and the desired vanishing follows from a closure relation in $\mathscr{I}_3$. When the difference is positive, we choose a subset $J \subseteq F$ with $$\text{rk}(J)=\text{rk}(I)+1, \quad I \setminus J=\{i\} \ \ \text{and} \ \ J \setminus I=\{j\}.$$ From the linear relation in $\mathscr{I}_4$ for $i$ and $j$, we deduce that $$x_{i}+\sum_{\substack{i \in G \\ j \notin G}} x_G=
x_{j}+\sum_{\substack{j \in G \\ i \notin G}} x_G,$$ where the sums are over flats $G$ in $\mathscr{P}$. Multiplying both sides by $\Big(\prod_{i \in I \cap J} x_i \Big) x_F$, we get $$\Big( \prod_{i \in I} x_i\Big) x_F=\Big( \prod_{j \in J} x_j\Big) x_F.$$ Indeed, a term involving $x_G$ in the expansions of the products is zero in the Chow ring by
(1) an incomparability relation in $\mathscr{I}_1$, if $G \nsubseteq F$,
(2) a complement relation in $\mathscr{I}_2$, if $I \cap J \nsubseteq G$,
(3) the induction hypothesis for $I \cap J \subseteq G$, if otherwise.
The right-hand side of the equality is zero by the induction hypothesis for $J \subseteq F$.
The second assertion can be proved in the same way, by descending induction on the rank of $I$, using the first part of the proposition.
We record here that the isomorphism of Proposition \[PropositionSimplification\] uniquely extends to an isomorphism between the corresponding Chow rings.
The homomorphism $\pi_{\text{PL}}$ induces an isomorphism of graded rings $$\underline{\pi}_{\text{PL}}: A^*(\mathrm{M},\mathscr{P}) \longrightarrow A^*(\overline{\mathrm{M}},\mathscr{P}).$$ The homomorphism $\iota_{\text{PL}}$ induces the inverse isomorphism of graded rings $$\underline{\iota}_{\text{PL}}: A^*(\overline{\mathrm{M}},\mathscr{P}) \longrightarrow A^*(\mathrm{M},\mathscr{P}).$$
Consider the extensions of $\pi_{\text{PL}}$ and $\iota_{\text{PL}}$ to the polynomial rings $$\xymatrix{
S_{E \cup \mathscr{P}} \ar@<0.5ex>[rr]^{\widetilde{\pi}_\text{PL}} &&S_{\overline{E} \cup \mathscr{P}}\ar@<0.5ex>[ll]^{\widetilde{\iota}_\text{PL}}.
}$$ The result follows from the observation that $\widetilde{\pi}_{\text{PL}}$ and $\widetilde{\iota}_{\text{PL}}$ preserve the monomial relations in $\mathscr{I}_1$, $\mathscr{I}_2$, and $\mathscr{I}_3$.
{#SectionMatroidalFlip}
Let $\mathscr{P}_-$ be an order filter of $\mathscr{P}(\mathrm{M})$, and let $Z$ be a flat maximal in $\mathscr{P}(\mathrm{M}) \setminus \mathscr{P}_-$. We set $$\mathscr{P}_+:=\mathscr{P}_- \cup \{Z\} \subseteq \mathscr{P}(\mathrm{M}).$$ The collection $\mathscr{P}_+$ is an order filter of $\mathscr{P}(\mathrm{M})$.
The *matroidal flip* from $\mathscr{P}_-$ to $\mathscr{P}_+$ is the modification of fans $ \Sigma_{\mathrm{M},\mathscr{P}_{-}} \hspace{-1.5mm}\rightsquigarrow \Sigma_{\mathrm{M},\mathscr{P}_{+}}$.
The flat $Z$ will be called the *center* of the matroidal flip. The matroidal flip removes the cones $$\sigma_{I < \mathscr{F}} \ \ \text{with} \ \ \text{cl}_\mathrm{M}(I)=Z \ \ \text{and} \ \ \text{min}\ \mathscr{F}\neq Z,$$ and replaces them with the cones $$\sigma_{I < \mathscr{F}} \ \ \text{with} \ \ \text{cl}_\mathrm{M}(I) \neq Z \ \ \text{and} \ \ \text{min}\ \mathscr{F}=Z.$$ The center $Z$ is necessarily minimal in $\mathscr{P}_+$, and we have $$\begin{aligned}
&\text{star}(\ \sigma_{Z<\varnothing}\ ,\ \Sigma_{\mathrm{M},\mathscr{P}_-}) \simeq \Sigma_{\mathrm{M}_Z}, \\
&\text{star}(\sigma_{\varnothing<\{Z\}},\Sigma_{\mathrm{M},\mathscr{P}_+}) \simeq \Sigma_{\mathrm{M}^Z,\varnothing} \times \Sigma_{\mathrm{M}_Z}.\end{aligned}$$
The matroidal flip preserves the homotopy type of the underlying simplicial complexes $\Delta_{\mathrm{M},\mathscr{P}_-}$ and $\Delta_{\mathrm{M},\mathscr{P}_+}$. To see this, consider the inclusion $$\Delta_{\mathrm{M},\mathscr{P}_+} \subseteq \Delta^{*}_{\mathrm{M},\mathscr{P}_-}:=\text{the stellar subdivision of $\Delta_{\mathrm{M},\mathscr{P}_-}$ relative to $\vartriangle_{Z<\varnothing}$}.$$ We claim that the left-hand side is a deformation retract of the right-hand side. More precisely, there is a sequence of compositions of elementary collapses $$\begin{gathered}
\Delta^{*}_{\mathrm{M},\mathscr{P}_-} =\Delta^{1,1}_{\mathrm{M},\mathscr{P}_-} \ \rightsquigarrow \ \Delta^{1,2}_{\mathrm{M},\mathscr{P}_-}\ \rightsquigarrow \ \cdots \ \rightsquigarrow\ \Delta^{1,\text{crk}(Z)-1}_{\mathrm{M},\mathscr{P}_-} \ \rightsquigarrow \ \\
\Delta^{1,\text{crk}(Z)}_{\mathrm{M},\mathscr{P}_-} =\Delta^{2,1}_{\mathrm{M},\mathscr{P}_-} \ \rightsquigarrow\ \Delta^{2,2}_{\mathrm{M},\mathscr{P}_-} \ \rightsquigarrow\ \cdots \ \rightsquigarrow \ \Delta^{2,\text{crk}(Z)-1}_{\mathrm{M},\mathscr{P}_-} \ \rightsquigarrow \ \hspace{35mm}\\
\Delta^{2,\text{crk}(Z)}_{\mathrm{M},\mathscr{P}_-} =\Delta^{3,1}_{\mathrm{M},\mathscr{P}_-} \ \rightsquigarrow\ \Delta^{3,2}_{\mathrm{M},\mathscr{P}_-} \ \rightsquigarrow \
\cdots \ \rightsquigarrow\ \Delta^{3,\text{crk}(Z)-1}_{\mathrm{M},\mathscr{P}_-}\ \rightsquigarrow\ \cdots \ \rightsquigarrow \ \Delta_{\mathrm{M},\mathscr{P}_+}, \hspace{10mm}\end{gathered}$$ where $\Delta^{m,k+1}_{\mathrm{M},\mathscr{P}_-}$ is the subcomplex of $\Delta^{m,k}_{\mathrm{M},\mathscr{P}_-}$ obtained by collapsing all the faces $\vartriangle_{I<\mathscr{F}}$ with $$\text{cl}_\mathrm{M}(I) = Z, \quad \text{min}\ \mathscr{F}\neq Z, \quad |Z \setminus I|=m,\quad |\mathscr{F}|=\text{crk}_\mathrm{M}(Z)-k.$$ The faces $\vartriangle_{I<\mathscr{F}}$ satisfying the above conditions can be collapsed in $\Delta^{m,k}_{\mathrm{M},\mathscr{P}_-}$ because $$\text{link}(\vartriangle_{I<\mathscr{F}},\Delta^{m,k}_{\mathrm{M},\mathscr{P}_-})=\{\mathbf{e}_Z\}.$$ It follows that the homotopy type of the Bergman complex $\Delta_{\mathrm{M},\mathscr{P}}$ is independent of $\mathscr{P}$. For basics of elementary collapses of simplicial complexes, see [@Kozlov Chapter 6]. The special case that $\Delta_{\mathrm{M},\varnothing}$ is homotopic to $\Delta_\mathrm{M}$ is known in combinatorial topology as the crosscut theorem, see for example [@Kozlov Chapter 13].
We construct homomorphisms associated to the matroidal flip, the *pullback homomorphism* and the *Gysin homomorphism*.
There is a graded ring homomorphism between the Chow rings $$\Phi_Z: A^*(\mathrm{M},\mathscr{P}_-) \longrightarrow A^*(\mathrm{M},\mathscr{P}_+)$$ uniquely determined by the property $$x_F \longmapsto x_F \quad \text{and} \quad x_i \longmapsto \begin{cases} x_i+x_Z & \text{if $i \in Z$,}\\ x_i & \text{if $i \notin Z$.} \end{cases}$$
We call this map the *pullback homomorphism* associated to the matroidal flip from $\mathscr{P}_-$ to $\mathscr{P}_+$. The pullback homomorphism will later shown to be injective, see Theorem \[DecompositionTheorem\].
Consider the homomorphism between the polynomial rings $$\phi_Z: S_{E \cup \mathscr{P}_-} \longrightarrow S_{E \cup \mathscr{P}_+}$$ defined by the same rule determining $\Phi_Z$. We claim that $$\phi_Z(\mathscr{I}_1) \subseteq \mathscr{I}_1, \quad
\phi_Z(\mathscr{I}_2) \subseteq \mathscr{I}_1 + \mathscr{I}_2, \quad
\phi_Z(\mathscr{I}_3) \subseteq \mathscr{I}_2 + \mathscr{I}_3, \quad
\phi_Z(\mathscr{I}_4) \subseteq \mathscr{I}_4.$$ The first and the last inclusions are straightforward to verify.
We check the second inclusion. For an element $i$ in $E \setminus F$, we have $$\phi_Z(x_ix_F) = \begin{cases}x_ix_F+x_Zx_F &\text{if $i \in Z$,}\\ x_ix_F & \text{if $i \notin Z$.} \end{cases}$$ If $i$ is in $Z \setminus F$, then the monomial $x_Z x_F$ is in $\mathscr{I}_1$ because $Z$ is minimal in $\mathscr{P}_+$.
We check the third inclusion. For an independent set $I$ whose closure is in $\mathscr{P}_- \cup \{E\}$, $$\phi_Z\Big(\prod_{i \in I} x_i\Big) =\prod_{i \in I \setminus Z} x_i \ \prod_{i \in I \cap Z} (x_i+x_Z).$$ The term $\prod_{i \in I} x_i$ in the expansion of the right-hand side is in $\mathscr{I}_3$. Since $Z$ is minimal in $\mathscr{P}_+$, there is an element in $I \setminus Z$, and hence all the remaining terms in the expansion are in $\mathscr{I}_2$.
The pullback homomorphism $\Phi_Z$ is an isomorphism when $\text{rk}_\mathrm{M}(Z)=1$.
Let $j_1$ and $j_2$ be distinct elements of $Z$. If $Z$ has rank $1$, then a flat contains $j_1$ if and only if it contains $j_2$. It follows from the linear relation in $S_{E \cup \mathscr{P}_-}$ for $j_1$ and $j_2$ that $$x_{j_1}=x_{j_2} \in A^*(\mathrm{M},\mathscr{P}_-).$$ We choose an element $j \in Z$, and construct the inverse $\Phi_Z'$ of $\Phi_Z$ by setting $$x_Z \longmapsto x_{j}, \quad x_F \longmapsto x_F, \quad \text{and} \quad
x_i \longmapsto \begin{cases} 0 & \text{if $i \in Z$,}\\ x_i & \text{if $i \notin Z$.} \end{cases}$$ It is straightforward to check that $\Phi_Z'$ is well-defined, and that $\Phi_Z'=\Phi_Z^{-1}$.
As before, we identify the flats of $\mathrm{M}_Z$ with the flats of $\mathrm{M}$ containing $Z$, and identify the flats of $\mathrm{M}^Z$ with the flats of $\mathrm{M}$ contained in $Z$.
\[PropositionGysin\] Let $p$ and $q$ be positive integers.
(1) There is a group homomorphism $$\Psi_{Z}^{p,q}: A^{q-p}(\mathrm{M}_{Z}) \longrightarrow A^{q}(\mathrm{M},\mathscr{P}_+)$$ uniquely determined by the property $
x_\mathscr{F} \longmapsto x_{Z}^p \ x_{\mathscr{F}}.
$
(2) There is a group homomorphism $$\Gamma^{p,q}_Z:A^{q-p}(\mathrm{M}^Z) \longrightarrow A^{q}(\mathrm{M})$$ uniquely determined by the property $x_\mathscr{F} \longmapsto x_Z^p \ x_\mathscr{F}$.
We call the map $\Psi^{p,q}_Z$ the *Gysin homomorphism* of type $p,q$ associated to the matroidal flip from $\mathscr{P}_-$ to $\mathscr{P}_+$. The Gysin homomorphism will later shown to be injective when $p<\text{rk}_\mathrm{M}(Z)$, see Theorem \[DecompositionTheorem\].
It is clear that the Gysin homomorphism $\Psi_{Z}^{p,q}$ respects the incomparability relations. We check that $\Psi_{Z}^{p,q}$ respects the linear relations.
Let $i_1$ and $i_2$ be elements in $E \setminus Z$, and consider the linear relation in $S_{E \cup\mathscr{P}_+}$ for $i_1$ and $i_2$: $$\Big(x_{i_1}+\sum_{i_1 \in F} x_F\Big)-\Big(x_{i_1}+\sum_{i_2 \in F} x_F\Big) \in \mathscr{I}_4.$$ Since $i_1$ and $i_2$ are not in $Z$, multiplying the linear relation with $x_Z^p$ gives $$x_Z^p \Big(\sum_{Z \cup\{i_1\} \subseteq F} x_F - \sum_{Z \cup \{i_2\} \subseteq F} x_F \Big) \in \mathscr{I}_1+\mathscr{I}_2+\mathscr{I}_4.$$
The second statement on $\Gamma^{p,q}_Z$ can be proved in the same way, using $i_1$ and $i_2$ in $Z$.
Let $\mathscr{P}$ be any order filter of $\mathscr{P}(\mathrm{M})$. We choose a sequence of order filters of the form $$\varnothing,\mathscr{P}_1,\mathscr{P}_2,\ldots, \mathscr{P},\ldots,\mathscr{P}(\mathrm{M}),$$ where an order filter in the sequence is obtained from the preceding one by adding a single flat. The corresponding sequence of matroidal flips interpolates between $\Sigma_{\mathrm{M},\varnothing}$ and $\Sigma_{\mathrm{M}}$: $$\Sigma_{\mathrm{M},\varnothing} \rightsquigarrow\Sigma_{\mathrm{M},\mathscr{P}_1} \rightsquigarrow \ldots \rightsquigarrow \Sigma_{\mathrm{M},\mathscr{P}} \rightsquigarrow \ldots \rightsquigarrow \Sigma_{\mathrm{M}}.$$
\[DefinitionIntermediateDegree\] We write $\Phi_\mathscr{P}$ and $\Phi_{\mathscr{P}^c}$ for the compositions of pullback homomorphisms $$\Phi_\mathscr{P}:A^*(\mathrm{M},\varnothing) \longrightarrow A^*(\mathrm{M},\mathscr{P}) \quad \text{and} \quad
\Phi_{\mathscr{P}^c}:A^*(\mathrm{M},\mathscr{P}) \longrightarrow A^*(\mathrm{M}).$$
Note that $\Phi_\mathscr{P}$ and $\Phi_{\mathscr{P}^c}$ depend only on $\mathscr{P}$ and not on the chosen sequence of matroidal flips. The composition of all the pullback homomorphisms $\Phi_{\mathscr{P}^c} \circ \Phi_{\mathscr{P}}$ is uniquely determined by its property $$\Phi_{\mathscr{P}^c} \circ \Phi_{\mathscr{P}}\ (x_i) = \alpha_\mathrm{M}.$$
{#section-16}
Let $\mathscr{P}_-$ and $\mathscr{P}_+$ be as before, and let $Z$ be the center of the matroidal flip from $\mathscr{P}_-$ to $\mathscr{P}_+$. For positive integers $p$ and $q$, we consider the pullback homomorphism in degree $q$ $$\Phi^q_Z:A^q(\mathrm{M},\mathscr{P}_-) \longrightarrow A^q(\mathrm{M},\mathscr{P}_+)$$ and the Gysin homomorphism of type $p,q$ $$\Psi_{Z}^{p,q}: A^{q-p}(\mathrm{M}_{Z}) \longrightarrow A^{q}(\mathrm{M},\mathscr{P}_+).$$
\[PropositionSurjective\] For any positive integer $q$, the sum of the pullback homomorphism and Gysin homomorphisms $$\Phi_{Z}^q \oplus \bigoplus_{p=1}^{\text{rk}(Z)-1} \Psi_Z^{p,q}$$ is a surjective group homomorphism.
The proof is given below Lemma \[LemmaBasisReduction\]. In Theorem \[DecompositionTheorem\], we will show that the sum is in fact an isomorphism.
\[CorollaryTopIsomorphism\] The pullback homomorphism $\Phi_Z$ is an isomorphism in degree $r$: $$\Phi_Z^r:A^r(\mathrm{M},\mathscr{P}_-) \simeq A^r(\mathrm{M},\mathscr{P}_+).$$
Repeated application of the corollary shows that, for any order filter $\mathscr{P}$, the homomorphisms $\Phi_\mathscr{P}$ and $\Phi_{\mathscr{P}^c}$ are isomorphisms in degree $r$: $$\Phi_\mathscr{P}^r: A^r(\mathrm{M},\varnothing) \simeq A^r(\mathrm{M},\mathscr{P}) \quad \text{and} \quad \Phi_{\mathscr{P}^c}^r: A^r(\mathrm{M},\mathscr{P}) \simeq A^r(\mathrm{M}).$$
The contracted matroid $\mathrm{M}_Z$ has rank $\text{crk}_\mathrm{M}(Z)$, and hence $$\Psi_Z^{p,q}=0 \ \ \text{when $p <\text{rk}_\mathrm{M}(Z)$ and $q=r$.}$$ Therefore, Proposition \[PropositionSurjective\] for $q=r$ says that the homomorphism $\Phi_Z$ is surjective in degree $r$.
Choose a sequence of matroidal flips $$\Sigma_{\mathrm{M},\varnothing} \rightsquigarrow \ldots \rightsquigarrow \Sigma_{\mathrm{M},\mathscr{P}_{-}}\rightsquigarrow \Sigma_{\mathrm{M},\mathscr{P}_{+}} \rightsquigarrow \ldots \rightsquigarrow \Sigma_{\mathrm{M}},$$ and consider the corresponding group homomorphisms $$\xymatrixcolsep{3pc}
\xymatrix{
A^r(\mathrm{M},\varnothing) \ar[r]^{\Phi_{\mathscr{P}_-}}& A^r(\mathrm{M},\mathscr{P}_-)
\ar[r]^{\Phi_{\mathscr{P}_Z}}& A^r(\mathrm{M},\mathscr{P}_2) \ar[r]^{\Phi_{\mathscr{P}_+^c}}& A^r(\mathrm{M}).
}$$ Proposition \[PropositionSurjective\] applied to each matroidal flips in the sequence shows that all three homomorphisms are surjective. The first group is clearly isomorphic to $\mathbb{Z}$, and by Proposition \[PropositionDegreeMap\], the last group is also isomorphic to $\mathbb{Z}$. It follows that all three homomorphisms are isomorphisms.
Let $\beta_{\mathrm{M}_Z}$ be the element $\beta$ in Definition \[DefinitionAlphaBeta\] for the contracted matroid $\mathrm{M}_Z$. The first part of Proposition \[PropositionGysin\] shows that the expression $x_Z\hspace{0.5mm} \beta_{\mathrm{M}_Z}$ defines an element in $A^*(\mathrm{M},\mathscr{P}_+)$.
\[MinimalFlatRelation\] For any element $i$ in $Z$, we have $$x_ix_Z+x_Z^2+x_Z\beta_{\mathrm{M}_Z} =0\in A^*(\mathrm{M},\mathscr{P}_+).$$
We choose an element $j$ in $E \setminus Z$, and consider the linear relation in $S_{E \cup \mathscr{P}_+}$ for $i$ and $j$: $$\Bigg(x_i+\sum_{\substack{i \in F \\ j \notin F}} x_F\Bigg)-\Bigg(x_j+\sum_{\substack{j \in F \\ i \notin F}} x_F\Bigg) \in \mathscr{I}_4.$$ Since $i$ is in $Z$ and $Z$ is minimal in $\mathscr{P}_+$, multiplying the linear relation with $x_Z$ gives $$x_Zx_i+x_Z^2+\Bigg(\sum_{Z \subsetneq F \subsetneq F \cup \{j\}} x_Zx_F\Bigg) \in \mathscr{I}_1+\mathscr{I}_2+\mathscr{I}_4.$$ The sum in the parenthesis is the image of $\beta_{\mathrm{M}_Z}$ under the homomorphism $\Psi^{1,2}_Z$.
Let $\alpha_{\mathrm{M}^Z}$ be the element $\alpha$ in Definition \[DefinitionAlphaBeta\] for the restricted matroid $\mathrm{M}^Z$. The second part of Proposition \[PropositionGysin\] shows that the expression $x_Z\hspace{0.5mm} \alpha_{\mathrm{M}^Z}$ defines an element in $A^*(\mathrm{M})$.
\[LemmaAlphaRelation\] If $Z$ is maximal among flats strictly contained in a proper flat $\widetilde{Z}$, then $$x_Z x_{\widetilde{Z}}(x_Z+\alpha_{\mathrm{M}^Z})=0 \in A^*(\mathrm{M}).$$ If $Z$ is maximal among flats strictly contained in the flat $E$, then $$x_Z (x_Z+\alpha_{\mathrm{M}^Z})=0 \in A^*(\mathrm{M}).$$
We justify the first statement; the second statement can be proved in the same way.
Choose an element $i$ in $Z$ and an element $j$ in $\widetilde{Z} \setminus Z$. The linear relation for $i$ and $j$ shows that $$\sum_{\substack{i \in F \\ j \notin F}} x_F=\sum_{\substack{j \in F \\ i \notin F}} x_F \in A^*(\mathrm{M}).$$ Multiplying both sides by the monomial $x_Z \ x_{\widetilde{Z}}$, the incomparability relations give $$x_Z^2 \ x_{\widetilde{Z}}+\Big( \sum_{i \in F \subsetneq Z} x_F \ x_Z\Big) \ x_{\widetilde{Z}}=0 \in A^*(\mathrm{M}).$$ The sum in the parenthesis is the image of $\alpha_{\mathrm{M}^Z}$ under the homomorphism $\Gamma^{1,2}_Z$.
\[LemmaIdeal\] The sum of the images of Gysin homomorphisms is the ideal generated by $x_Z$: $$\sum_{p>0} \sum_{q>0} \text{im}\ \Psi_Z^{p,q}=x_Z \ A^*(\mathrm{M},\mathscr{P}_+).$$
It is enough to prove that the right-hand side is contained in the left-hand side. Since $Z$ is minimal in $\mathscr{P}_+$, the incomparability relations in $\mathscr{I}_1$ and the complement relations in $\mathscr{I}_2$ show that any nonzero degree $q$ monomial in the ideal generated by $x_Z$ is of the form $$x_Z^{k} \ \prod_{F\in \mathscr{F}} x_F^{k_F} \ \prod_{i \in I} x_i^{k_i}, \qquad I \subseteq Z < \mathscr{F},$$ where the sum of the exponents is $q$. Since the exponent $k$ of $x_Z$ is positive, Lemma \[MinimalFlatRelation\] shows that this monomial is in the sum $$\text{im}\ \Psi_Z^{k,q}+\text{im}\ \Psi_Z^{k+1,q}+\cdots + \text{im}\ \Psi_Z^{q,q}. \qedhere$$
\[LemmaInclusions\] For positive integers $p$ and $q$, we have $$x_Z \ \text{im}\ \Phi_Z^q \subseteq \text{im}\ \Psi_Z^{1,q+1}
\quad \text{and} \quad
x_Z \ \text{im}\ \Psi_Z^{p,q} \subseteq \text{im}\ \Psi_Z^{p+1,q+1}.$$ If $F$ is a proper flat strictly containing $Z$, then $$x_F \ \text{im}\ \Phi_Z^q \subseteq \text{im}\ \Phi_Z^{q+1}
\quad \text{and} \quad
x_F \ \text{im}\ \Psi_Z^{p,q} \subseteq \text{im}\ \Psi_Z^{p,q+1}.$$
Only the first inclusion is nontrivial. Note that the left-hand side is generated by elements of the form $$\xi=x_Z \ \prod_{F\in \mathscr{F}} x_F^{k_F} \prod_{i \in I \setminus Z} x_i^{k_i} \prod_{i \in I \cap Z} (x_i+x_Z)^{k_i},$$ where $I$ is a subset of $E$ and $\mathscr{F}$ is a flag in $\mathscr{P}_-$. When $I$ is contained in $Z$, Lemma \[MinimalFlatRelation\] shows that $$\xi=x_Z \prod_{F\in \mathscr{F}} x_F^{k_F} \ \prod_{i \in I} (-\beta_{\mathrm{M}_Z})^{k_i} \in \text{im}\ \Psi_Z^{1,q+1}.$$ When $I$ is not contained in $Z$, a complement relation in $S_{E \cup \mathscr{P}_+}$ shows that $\xi=0$.
\[LemmaBasisReduction\] For any integers $k \ge \text{rk}_\mathrm{M}(Z)$ and $q \ge k$, we have $$\text{im}\ \Psi_{Z}^{k,q} \subseteq \text{im} \ \Phi_{Z}^q+ \sum_{p=1}^{k-1} \text{im}\ \Psi_{Z}^{p,q}.$$
By the second statement of Lemma \[LemmaInclusions\], it is enough to prove the assertion when $q=k$: The general case can be deduced by multiplying both sides of the inclusion by $x_\mathscr{F}$ for $Z<\mathscr{F}$.
By the first statement of Lemma \[LemmaInclusions\], it is enough to justify the above when $k=\text{rk}_\mathrm{M}(Z)$: The general case can be deduced by multiplying both sides of the inclusion by powers of $x_Z$.
We prove the assertion when $k=q=\text{rk}_\mathrm{M}(Z)$. For this we choose a basis $I$ of $Z$, and expand the product $$\prod_{i\in I} (x_i+x_Z) \in \text{im}\ \Phi_Z^k.$$ The closure relation for $I$ shows that the term $\prod_{i \in I} x_i$ in the expansion is zero, and hence, by Lemma \[MinimalFlatRelation\], $$\prod_{i \in I} (x_i+x_Z)=(-\beta_{\mathrm{M}_Z})^k-(-x_Z-\beta_{\mathrm{M}_Z})^k \in \text{im}\ \Phi_Z^k.$$ Expanding the right-hand side, we see that $$x_Z^k\in \text{im}\ \Phi_Z^k+ \sum_{p=1}^{k-1} \text{im}\ \Psi_{Z}^{p,k}.$$ Since $\text{im}\ \Psi_{Z}^{k,k} $ is generated by $x_Z^k$, this implies the asserted inclusion.
By Lemma \[LemmaBasisReduction\], it is enough to show that the sum $
\Phi_{Z}^q \oplus \bigoplus_{p=1}^{q} \Psi_Z^{p,q}
$ is surjective. By Lemma \[LemmaIdeal\], the image of the second summand is the degree $q$ part of the ideal generated by $x_Z$.
We show that any monomial is in the image of the pullback homomorphism $\Phi_Z$ modulo the ideal generated by $x_Z$. Note that any degree $q$ monomial not in the ideal generated by $x_Z$ is of the form $$\prod_{F\in \mathscr{F}} x_F^{k_F} \ \prod_{i \in I} x_i^{k_i}, \qquad Z \notin \mathscr{F}.$$ Modulo the ideal generated by $x_Z$, this monomial is equal to $$\Phi_Z\Big( \prod_{F\in \mathscr{F}} x_F^{k_F} \ \prod_{i \in I} x_i^{k_i}\Big)=\prod_{F\in \mathscr{F}} x_F^{k_F} \prod_{i \in I \setminus Z} x_i^{k_i} \prod_{i \in I \cap Z} (x_i+x_Z)^{k_i}. \qedhere$$
We use Proposition \[PropositionSurjective\] to show that the Gysin homomorphism between top degrees is an isomorphism.
\[GysinTopIsomorphism\] The Gysin homomorphism $\Psi_Z^{p,q}$ is an isomorphism when $p=\text{rk}(Z)$ and $q=r$: $$\Psi_Z^{p,q}: A^{\text{crk}(Z)-1}(\mathrm{M}_Z)\simeq A^r(\mathrm{M},\mathscr{P}_+).$$
We consider the composition $$\xymatrixcolsep{3pc}
\xymatrix{
A^{\text{crk}(Z)-1}(\mathrm{M}_Z) \ar[r]^{\Psi^{p,q}_Z} & A^r(\mathrm{M},\mathscr{P}_+) \ar[r]^{\Phi_{\mathscr{P}_+^c}}&A^r(\mathrm{M}),
}
\qquad
x_\mathscr{F} \longmapsto x_Z^{\text{rk}(Z)} \ x_\mathscr{F}.$$ The second map is an isomorphism by Corollary \[CorollaryTopIsomorphism\], and therefore it is enough to show that the composition is an isomorphism.
For this we choose two flags of nonempty proper flats of $\mathrm{M}$: $$\begin{aligned}
\mathscr{Z}_1&=\text{a flag of flats strictly contained in $Z$ with $|\mathscr{Z}_1|=\text{rk}(Z)-1$,}\\[2pt]
\mathscr{Z}_2&=\text{a flag of flats strictly containing $Z$ with $|\mathscr{Z}_2|=\text{crk}(Z)-1$.}\end{aligned}$$ We claim that the composition maps a generator to a generator: $$(-1)^{\text{rk}(Z)-1} \ x_Z^{\text{rk}(Z)} \ x_{\mathscr{Z}_2}= x_{\mathscr{Z}_1} \hspace{0.5mm} x_Z \hspace{0.5mm} x_{\mathscr{Z}_2} \in A^*(\mathrm{M}).$$ Indeed, the map $\Gamma_Z^{1,\text{rk}(Z)}$ applied to the second formula of Proposition \[PropositionFundamentalClass\] for $\mathrm{M}^Z$ gives $$x_{\mathscr{Z}_1}\hspace{0.5mm}x_Z \hspace{0.5mm}x_{\mathscr{Z}_2}=(\alpha_{\mathrm{M}^Z})^{\text{rk}(Z)-1} \ x_Z \hspace{0.5mm}x_{\mathscr{Z}_2}\in A^*(\mathrm{M}),$$ and, by Lemma \[LemmaAlphaRelation\], the right-hand side of the above is equal to $$(-1)^{\text{rk}(Z)-1}\ x_Z^{\text{rk}(Z)} \ x_{\mathscr{Z}_2} \in A^*(\mathrm{M}). \qedhere$$
{#section-17}
Let $\mathscr{P}_-$, $\mathscr{P}_+$, and $Z$ be as before, and let $\mathscr{P}$ be any order filter of $\mathscr{P}(\mathrm{M})$.
\[DecompositionTheorem\] For any positive integer $q$, the sum of the pullback homomorphism and the Gysin homomorphisms $$\Phi_{Z}^q \oplus \bigoplus_{p=1}^{\text{rk}(Z)-1} \Psi_Z^{p,q}$$ is an isomorphism.
\[PoincareDuality\] For any nonnegative integer $q \le r$, the multiplication map $$A^{q}(\mathrm{M},\mathscr{P}) \times A^{r-q}(\mathrm{M},\mathscr{P}) \longrightarrow A^r(\mathrm{M},\mathscr{P})$$ defines an isomorphism between groups $$A^{r-q}(\mathrm{M},\mathscr{P}) \simeq \text{Hom}_\mathbb{Z}(A^{q}(\mathrm{M},\mathscr{P}), A^r(\mathrm{M},\mathscr{P})).$$
In particular, the groups $A^q(\mathrm{M},\mathscr{P})$ are torsion free. We simultaneously prove Theorem \[DecompositionTheorem\] (Decomposition) and Theorem \[PoincareDuality\] (Poincaré Duality) by lexicographic induction on the rank of matroids and the cardinality of the order filters. The proof is given below Lemma \[LemmaInduction\].
\[LemmaUpperTriangular\] Let $q_1$ and $q_2$ be positive integers.
(1) For any positive integer $p$, we have $$\text{im} \ \Psi_Z^{p,q_1} \cdot \text{im} \ \Phi^{q_2}_Z \subseteq \text{im} \ \Psi_Z^{p,q_1+q_2}$$
(2) For any positive integers $p_1$ and $p_2$, we have $$\text{im} \ \Psi_Z^{p_1,q_1} \cdot \text{im} \ \Psi_Z^{p_2,q_2} \subseteq \text{im} \ \Psi_Z^{p_1+p_2,q_1+q_2}.$$
The first inclusion shows that, when $q_1+q_2=r$ and $p$ is less than $\text{rk}(Z)$, $$\text{im} \ \Psi_Z^{p,q_1} \cdot \text{im} \ \Phi^{q_2}_Z = 0.$$ The second inclusion shows that, when $q_1+q_2=r$ and $p_1+p_2$ is less than $\text{rk}(Z)$, $$\text{im} \ \Psi_Z^{p_1,q_1} \cdot \text{im} \ \Psi_Z^{p_2,q_2} = 0.$$
The assertions are direct consequences of Lemma \[LemmaInclusions\].
\[LemmaInduction\] Let $q$ be a positive integer, and $p_1,p_2$ be distinct positive integers less than $\text{rk}(Z)$.
(1) If Poincaré Duality holds for $A^*(\mathrm{M},\mathscr{P}_-)$, then $$\text{ker} \ \Phi^q_Z=0 \quad \text{and} \quad \text{im} \ \Phi^q_Z \cap \sum_{p=1}^{\text{rk}(Z)-1} \hspace{-1mm}\text{im}\ \Psi_Z^{p,q} =0.$$
(2) If Poincaré Duality holds for $A^*(\mathrm{M}_Z)$, then $$\text{ker} \ \Psi_Z^{p_1,q}=\text{ker} \ \Psi_Z^{p_2,q}=0 \quad \text{and} \quad \text{im} \ \Psi_Z^{p_1,q} \cap \text{im} \ \Psi_Z^{p_2,q}=0$$
Let $\xi$ be a nonzero element in the domain of $\Phi_Z^q$. Since $\Phi_Z$ is an isomorphism between top degrees, Poincaré Duality for $(\mathrm{M},\mathscr{P}_-)$ implies that $$\Phi_Z(\xi) \cdot \text{im} \ \Phi^{r-q}_Z \neq 0.$$ This shows that $\Phi_Z^q$ is injective. On the other hand, Lemma \[LemmaUpperTriangular\] shows that $$\Bigg( \sum_{p=1}^{\text{rk}(Z)-1}\text{im}\ \Psi_Z^{p,q} \Bigg) \cdot \text{im} \ \Phi^{r-q}_Z=0.$$ This shows that the image of $\Phi_Z^q$ intersects the image of $ \oplus_{p=1}^{\text{rk}(Z)-1} \Psi_Z^{p,q} $ trivially.
Let $\xi$ be a nonzero element in the domain of $\Psi_Z^{p,q}$, where $p=p_1$ or $p=p_2$. Since $\Psi_Z$ is an isomorphism between top degrees, Poincaré Duality for $\mathrm{M}_Z$ implies that $$\Psi^{p,q}_Z(\xi) \cdot \text{im}\ \Psi^{\text{rk}(Z)-p,r-q}_Z \neq 0.$$ This shows that $\Psi_Z^{p,q}$ is injective. For this assertion on the intersection, we assume that $p=p_1>p_2$. Under the assumption Lemma \[LemmaUpperTriangular\] shows $$\text{im} \ \Psi_Z^{p_2,q} \cdot \text{im} \ \Psi_Z^{\text{rk}(Z)-p,r-q} = 0.$$ This shows that the image of $\Psi_Z^{p_2,q}$ intersects the image of $\Psi_Z^{p_2,q}$ trivially.
We simultaneously prove Decomposition and Poincaré Duality by lexicographic induction on the rank of $\mathrm{M}$ and the cardinality of $\mathscr{P}$ and $\mathscr{P}_-$. Note that both statements are valid when $r=1$, and Poincaré Duality holds when $q=0$ or $q=r$. Assuming that Poincaré Duality holds for $A^*(\mathrm{M}_Z)$, we show the implications $$\begin{gathered}
\Big(\text{Poincar\'e Duality holds for $A^*(\mathrm{M},\mathscr{P}_-)$}\Big) \Longrightarrow \\
\Big(\text{Poincar\'e Duality holds for $A^*(\mathrm{M},\mathscr{P}_-)$ and Decomposition holds for $\mathscr{P}_- \subseteq \mathscr{P}_+$}\Big) \\ \Longrightarrow
\Big(\text{Poincar\'e Duality holds for $A^*(\mathrm{M},\mathscr{P}_+)$}\Big).\end{gathered}$$ The base case of the induction is provided by the isomorphism $$A^*(\mathrm{M},\varnothing) \simeq \mathbb{Z}[x]/(x^{r+1}).$$ The first implication follows from Proposition \[PropositionSurjective\] and Lemma \[LemmaInduction\].
We prove the second implication. Decomposition for $\mathscr{P}_- \subseteq \mathscr{P}_+$ shows that, for any positive integer $q<r$, we have $$\begin{aligned}
A^q(\mathrm{M},\mathscr{P}_+)&= \text{im} \ \Phi_Z^{q} \oplus \text{im} \ \Psi_Z^{1,q} \oplus \text{im} \ \Psi_Z^{2,q} \oplus \cdots \oplus \text{im} \ \Psi_Z^{\text{rk}(Z)-1,q}, \ \ \text{and}
\\[4pt]
A^{r-q}(\mathrm{M},\mathscr{P}_+)&= \text{im} \ \Phi_Z^{r-q}\oplus \text{im} \ \Psi_Z^{\text{rk}(Z)-1,r-q} \oplus \text{im} \ \Psi_Z^{\text{rk}(Z)-2,r-q} \oplus \cdots \oplus \text{im} \ \Psi_Z^{1,r-q}.\end{aligned}$$ By Poincaré Duality for $(\mathrm{M},\mathscr{P}_-)$ and Poincaré Duality for $\mathrm{M}_Z$, all the summands above are torsion free. We construct bases of the sums by choosing bases of their summands.
We use Corollary \[CorollaryTopIsomorphism\] and Proposition \[GysinTopIsomorphism\] to obtain isomorphisms $$A^r(\mathrm{M},\mathscr{P}_-) \simeq A^r(\mathrm{M},\mathscr{P}_+) \simeq A^{\text{crk}(Z)-1}(\mathrm{M}_Z) \simeq \mathbb{Z}.$$ For a positive integer $q<r$, consider the matrices of multiplications $$\begin{aligned}
\mathscr{M}_+&:=\Big(A^{q}(\mathrm{M},\mathscr{P}_+) \times A^{r-q}(\mathrm{M},\mathscr{P}_+) \longrightarrow \mathbb{Z}\Big),
\\[2pt]
\mathscr{M}_-&:=\Big(A^{q}(\mathrm{M},\mathscr{P}_-) \times A^{r-q}(\mathrm{M},\mathscr{P}_-) \longrightarrow \mathbb{Z}\Big),\end{aligned}$$ and, for positive integers $p<\text{rk}(Z)$, $$\hspace{6mm} \mathscr{M}_p:=\Big(A^{q-p}(\mathrm{M}_Z) \times A^{r-q-\text{rk}(Z)+p}(\mathrm{M}_Z) \longrightarrow \mathbb{Z}\Big).$$ By Lemma \[LemmaUpperTriangular\], under the chosen bases ordered as shown above, $\mathscr{M}_+$ is a block upper triangular matrix with block diagonals $\mathscr{M}_-$ and $\mathscr{M}_p$, up to signs. It follows from Poincaré Duality for $(\mathrm{M},\mathscr{P}_-)$ and Poincaré Duality for $\mathrm{M}_Z$ that $$\text{det}\ \mathscr{M}_+=\pm \text{det}\ \mathscr{M}_- \times \prod_{p=1}^{\text{rk}(Z)-1} \text{det}\ \mathscr{M}_p=\pm 1.$$ This proves the second implication, completing the lexicographic induction.
Hard Lefschetz property and Hodge-Riemann relations {#SectionHLHR}
===================================================
{#section-18}
Let $r$ be a nonnegative integer. We record basic algebraic facts concerning the Poincaré duality, the hard Lefschetz property, and the Hodge-Riemann relations.
A graded Artinian ring $R^*$ satisfies the *Poincaré duality of dimension $r$* if
(1) there are isomorphisms $R^0 \simeq \mathbb{R}$ and $R^r \simeq \mathbb{R}$,
(2) for every integer $q>r$, we have $R^q \simeq 0$, and,
(3) for every integer $q \le r$, the multiplication defines an isomorphism $$R^{r-q} \longrightarrow \text{Hom}_\mathbb{R}(R^{q},R^r).$$
In this case, we say that $R^*$ is a *Poincaré duality algebra of dimension $r$*.
In the remainder of this subsection, we suppose that $R^*$ is a Poincaré duality algebra of dimension $r$. We fix an isomorphism, called the *degree map* for $R^*$, $$\text{deg}: R^r \longrightarrow \mathbb{R}.$$
\[PropositionLocalDuality\] For any nonzero element $x$ in $R^d$, the quotient ring $$R^*/\text{ann}(x), \ \ \text{where}\ \ \text{ann}(x):=\{a \in R^* \mid x \cdot a =0\},$$ is a Poincaré duality algebra of dimension $r-d$.
By definition, the degree map for $R^*/\text{ann}(x)$ *induced by $x$* is the homomorphism $$\text{deg}(x \cdot -): R^{r-d}/\text{ann}(x) \longrightarrow \mathbb{R}, \qquad a+\text{ann}(x) \longmapsto \text{deg}(x \cdot a).$$ The Poincaré duality for $R^*$ shows that the degree map for $R^*/\text{ann}(x)$ is an isomorphism.
This is straightforward to check, see for example [@Meyer-Smith Corollary I.2.3].
Let $\ell$ be an element of $R^1$, and let $q$ be a nonnegative integer $\le \frac{r}{2}$.
(1) The *Lefschetz operator* on $R^q$ associated to $\ell$ is the linear map $$L^q_{\ell}: R^q \longrightarrow R^{r-q}, \qquad a \longmapsto \ell^{r-2q} \ a.$$
(2) The *Hodge-Riemann form* on $R^q$ associated to $\ell$ is the symmetric bilinear form $$Q^{q}_{\ell}: R^q \times R^q \longrightarrow \mathbb{R}, \qquad (a_1,a_2) \longmapsto (-1)^q\ \text{deg}\ ( a_1 \cdot L_{\ell}^q (a_2) ).$$
(3) The *primitive subspace* of $R^q$ associated to $\ell$ is the subspace $$P^q_{\ell}:=\{a \in R^q \mid \ell \cdot L_\ell^q(a)=0\} \subseteq R^q.$$
We say that
(1) $R^*$ satisfies $\text{HL}(\ell)$ if the Lefschetz operator $L_{\ell}^q$ is an isomorphism on $R^q$ for all $q \le \frac{r}{2}$, and
(2) $R^*$ satisfies $\text{HR}(\ell)$ if the Hodge-Riemann form $Q_{\ell}^q$ is positive definite on $P_\ell^q$ for all $q \le \frac{r}{2}$.
If the Lefschetz operator $L^{q}_\ell$ is an isomorphism, then there is a decomposition $$R^{q+1}=P_\ell^{q+1} \oplus \ell \hspace{0.5mm} R^{q}.$$ Consequently, when $R^*$ satisfies $\text{HL}(\ell)$, we have the *Lefschetz decomposition* of $R^q$ for $q \le \frac{r}{2}$: $$R^q=P^q_\ell \oplus \ell\hspace{0.5mm} P^{q-1}_\ell \oplus \cdots \oplus \ell^q \hspace{0.5mm}P^0_\ell.$$ An important basic fact is that the Lefschetz decomposition of $R^q$ is orthogonal with respect to the Hodge-Riemann form $Q^q_\ell$: For nonnegative integers $q_1<q_2 \le q$, we have $$Q_\ell^q\Big(\ell^{q_1} a_1,\ell^{q_2} a_2\Big)=(-1)^q \text{deg}\ \Big(\ell^{q_2-q_1} \ell^{r-2q_2} a_1 a_2\Big)=0, \quad a_1 \in P_\ell^{q-q_1}, \ \ a_2 \in P_\ell^{q-q_2}.$$
\[HLCharacterization\] The following conditions are equivalent for $\ell \in R^1$:
(1) $R^*$ satisfies $\text{HL}(\ell)$.
(2) The Hodge-Riemann form $Q_\ell^q$ on $R^q$ is nondegenerate for all $q \le \frac{r}{2}$.
The Hodge-Riemann form $Q_\ell^q$ on $R^q$ is nondegenerate if and only if the composition $$\xymatrixcolsep{2.5pc}
\xymatrix{
R^q \ar[r]^{L^q_\ell\ } & R^{r-q} \ar[r] & \text{Hom}_\mathbb{R}(R^q,R^r)
}$$ is an isomorphism, where the second map is given by the multiplication in $R^*$. Since $R^*$ satisfies Poincaré duality, the composition is an isomorphism if and only if $L^q_\ell$ is an isomorphism.
If $L_\ell^q(a)=0$, then $Q_\ell^q(a,a)=0$ and $a \in P_\ell^q$. Thus the property $\text{HR}(\ell)$ implies the property $\text{HL}(\ell)$.
\[HRCharacterization\] The following conditions are equivalent for $\ell \in R^1$:
(1) $R^*$ satisfies $\text{HR}(\ell)$.
(2) The Hodge-Riemann form $Q_\ell^q$ on $R^q$ is nondegenerate and has signature $$\sum_{p=0}^q (-1)^{q-p} \Big(\text{dim}_\mathbb{R}R^p -\text{dim}_\mathbb{R} R^{p-1}\Big) \ \ \text{for all $q \le \frac{r}{2}$.}$$
If $R^*$ satisfies $\text{HR}(\ell)$, then $R^*$ satisfies $\text{HL}(\ell)$, and therefore we have the Lefschetz decomposition $$R^q=P^q_\ell \oplus \ell\hspace{0.5mm} P^{q-1}_\ell \oplus \cdots \oplus \ell^q \hspace{0.5mm}P^0_\ell.$$ Note that the Lefschetz decomposition of $R^q$ is orthogonal with respect to $Q^q_\ell$, and that there is an isometry $$\big(P_\ell^p,Q^p_\ell\big) \simeq \big(\ell^{q-p} \hspace{0.5mm}P_\ell^p,(-1)^{q-p}Q_\ell^q\big) \ \ \text{for every nonnegative integer $p \le q$.}$$ Therefore, the condition $\text{HR}(\ell)$ implies that $$\begin{aligned}
\Big(\text{signature of $Q^q_\ell$ on $R^q$}\Big)&=\sum_{p=0}^q (-1)^{q-p}\Big(\text{signature of $Q^{p}_\ell$ on $P_\ell^{p}$}\Big)\\
&=\sum_{p=0}^q (-1)^{q-p} \Big(\text{dim}_\mathbb{R}R^{p} -\text{dim}_\mathbb{R} R^{p-1}\Big). \end{aligned}$$
Conversely, suppose that the Hodge-Riemann forms $Q^q_\ell$ are nondegenerate and their signatures are given by the stated formula. Proposition \[HLCharacterization\] shows that $R^*$ satisfies $\text{HL}(\ell)$, and hence $$R^q=P^q_\ell \oplus \ell\hspace{0.5mm} P^{q-1}_\ell \oplus \cdots \oplus \ell^q \hspace{0.5mm}P^0_\ell.$$ The Lefschetz decomposition of $R^q$ is orthogonal with respect to $Q^q_\ell$, and therefore $$\begin{aligned}
\Big(\text{signature of $Q^q_\ell$ on $P^q_\ell$}\Big)&=
\Big(\text{signature of $Q^q_\ell$ on $R^q$}\Big)
-\Big(\text{signature of $Q^{q-1}_\ell$ on $R^{q-1}$}\Big).\end{aligned}$$ The assumptions on the signatures of $Q^q_\ell$ and $Q^{q-1}_\ell$ show that the right-hand side is $$\text{dim}_\mathbb{R}R^{q} -\text{dim}_\mathbb{R} R^{q-1}= \text{dim}_\mathbb{R} P^{q}_\ell.$$ Since $Q^q_\ell$ is nondegenerate on $P^q_\ell$, this means that $Q^q_\ell$ is positive definite on $P^q_\ell$.
{#section-19}
In this subsection, we show that the properties $\text{HL}$ and $\text{HR}$ are preserved under the tensor product of Poincaré duality algebras.
Let $R_1^*$ and $R_2^*$ be Poincaré duality algebras of dimensions $r_1$ and $r_2$ respectively. We choose degree maps for $R_1^*$ and for $R_2^*$, denoted $$\text{deg}_1:R_1^{r_1} \longrightarrow \mathbb{R}, \qquad
\text{deg}_2:R^{r_2}_2 \longrightarrow \mathbb{R}.$$ We note that $R_1 \otimes_\mathbb{R} R_2$ is a Poincaré duality algebra of dimension $r_1+r_2$: For any two graded components of the tensor product with complementary degrees $$\begin{aligned}
\Big( R_1^p \otimes_\mathbb{R} R_2^0\Big) \oplus \Big( R_1^{p-1} \otimes_\mathbb{R} R_2^1\Big) \oplus \cdots \oplus \Big( R_1^0 \otimes_\mathbb{R} R_2^p\Big), \\
\Big( R_1^q \otimes_\mathbb{R} R_2^0\Big) \oplus \Big( R_1^{q-1} \otimes_\mathbb{R} R_2^1\Big) \oplus \cdots \oplus \Big( R_1^0 \otimes_\mathbb{R} R_2^q\Big),\end{aligned}$$ the multiplication of the two can be represented by a block diagonal matrix with diagonals $$\Big( R^{p-k}_1 \otimes_\mathbb{R} R_2^{k}\Big) \times \Big(R^{q-r_2+k}_1 \otimes_\mathbb{R} R_2^{r_2-k} \Big)\longrightarrow R^{r_1}_1 \otimes_\mathbb{R} R_2^{r_2}.$$ By definition, the *induced degree map* for the tensor product is the isomorphism $$\text{deg}_1 \otimes_\mathbb{R} \text{deg}_2:R_1^{r_1} \otimes_\mathbb{R} R_2^{r_2} \longrightarrow \mathbb{R}.$$ We use the induced degree map whenever we discuss the property $\text{HR}$ for tensor products.
\[TensorHR\] Let $\ell_1$ be an element of $R^1_1$, and let $\ell_2$ be an element of $R_2^1$.
(1) If $R_1^*$ satisfies $\text{HL}(\ell_1)$ and $R_2^*$ satisfies $\text{HL}(\ell_2)$, then $R_1^* \otimes_\mathbb{R} R_2^*$ satisfies $\text{HL}(\ell_1\otimes 1 +1 \otimes \ell_2)$.
(2) If $R_1^*$ satisfies $\text{HR}(\ell_1)$ and $R_2^*$ satisfies $\text{HR}(\ell_2)$, then $R_1^* \otimes_\mathbb{R} R_2^*$ satisfies $\text{HR}(\ell_1\otimes 1 +1 \otimes \ell_2)$.
We begin the proof with the following special case.
\[LemmaPP\] Let $r_1\leq r_2$ be nonnegative integers, and consider the Poincaré duality algebras $$R^*_1= \mathbb{R}[x_1]/(x_1^{r_1+1}) \ \ \text{and} \ \
R^*_2= \mathbb{R}[x_2]/(x_2^{r_2+1})$$ equipped with the degree maps $$\begin{aligned}
\text{deg}_1:R^{r_1}_1 \longrightarrow \mathbb{R}, \qquad x_1^{r_1} \longmapsto 1,\\
\text{deg}_2:R^{r_2}_2 \longrightarrow \mathbb{R}, \qquad x_2^{r_2} \longmapsto 1.\end{aligned}$$ Then $R_1^*$ satisfies $\text{HR}(x_1)$, $R_2$ satisfies $\text{HR}(x_2)$, and $R^*_1 \otimes_\mathbb{R} R^*_2$ satisfies $\text{HR}(x_1 \otimes 1+1 \otimes x_2)$.
The first two assertions are easy to check, and the third assertion follows from the Hodge-Riemann relations for the cohomology of the compact Kähler manifold $\mathbb{C}\mathbb{P}^{r_1}\times \mathbb{C}\mathbb{P}^{r_2}$. Below we give a combinatorial proof using the Lindström-Gessel-Viennot lemma.
For the third assertion, we identify the tensor product with $$R^*:=\mathbb{R}[x_1,x_2]/(x_1^{r_1+1},x_2^{r_2+1}), \ \ \text{and set} \ \ \ell:= x_1 + x_2.$$ The induced degree map for the tensor product will be written $$\text{deg}: R^{r_1+r_2} \longrightarrow \mathbb{R}, \qquad x_1^{r_1}x_2^{r_2} \longmapsto 1.$$
For some (equivalently any) choice of basis of $R^q$, we have $$(-1)^{\frac{q(q+1)}{2}}\text{det}\ \big(Q_\ell^q\big)>0 \ \ \text{for all nonnegative integers} \ \ q \le r_1.$$
We show that it is enough to prove the claim. The inequality of the claim implies that $Q^q_\ell$ is nondegenerate for $q \le r_1$, and hence $L^q_\ell$ is an isomorphism for $q \le r_1$. The Hilbert function of $R^*$ forces the dimensions of the primitive subspaces to satisfy $$\text{dim}_\mathbb{R} P_\ell^q=\begin{cases} 1 & \text{for $q \le r_1$,} \\ 0 & \text{for $q>r_1$,}\end{cases}$$ and that there is a decomposition $$R^q=P^q_\ell \oplus \ell P^{q-1}_\ell \oplus \dots\oplus \ell^q P^0_\ell \ \ \text{for} \ \ q \le r_1.$$ Every summand of the above decomposition is $1$-dimensional, and hence $$\Big(\text{signature of $Q_\ell^q$ on $R^q$}\Big)=\pm1 -\Big(\text{signature of $Q_\ell^{q-1}$ on $R^{q-1}$}\Big).$$ The claim on the determinant of $Q_\ell^q$ determines the sign of $\pm 1$ in the above equality: $$\Big(\text{signature of $Q_\ell^q$}\Big)=1-\Big(\text{signature of $Q_\ell^{q-1}$}\Big).$$ It follows that the signature of $Q_\ell^q$ on $P^q_\ell$ is $1$ for $q \le r_1$, and thus $R$ satisfies $\text{HR}(\ell)$.
We now prove the claim on $\text{det}\ (Q^q_\ell)=\text{det}\ ((-1)^q \hspace{0.5mm}Q^q_\ell)$ for $q \le r_1$. We use the monomial basis $$\Big\{x_1^ix_2^{q-i}\mid i=0,1,\dots, q\Big\} \subseteq R^q.$$ The matrix $[a_{ij}]$ which represents $(-1)^q\hspace{0.5mm} Q^q_\ell$ has binomial coefficients as its entries: $$[a_{ij}]
:=\Bigg[\text{deg}\Big((x_1+x_2)^{r_1+r_2-2q}x_1^{i+j}x_2^{q-i+q-j}\Big)\Bigg]
=\Bigg[\binom{r_1+r_2-2q}{r_1-i-j}\Bigg].$$ We determine the sign of the determinant of $[a_{ij}]$ using the Lindström-Gessel-Viennot lemma, see [@Aigner Section 5.4] for an exposition and similar examples.
Consider the grid graph in the plane with vertices $\mathbb{Z}^2$ and edges directed in the positive $x$-directions and the positive $y$-directions. We place the starting points $\mathcal{P}_0,\ldots,\mathcal{P}_q$ and the ending points $\mathcal{Q}_0\ldots,\mathcal{Q}_q$ on two parallel diagonal lines, $\mathcal{P}_i$ from northwest to southeast and $\mathcal{Q}_j$ from southeast to northwest: $$\arraycolsep=5pt
\def\arraystretch{1.2}
\begin{array}{cccc}
\mathcal{P}_0=(-r_1,q),& \ \mathcal{P}_1\ =\ (1-r_1,q-1),& \ldots & \mathcal{P}_q=(q-r_1,0),
\\
\mathcal{Q}_q=(-q,r_2),& \mathcal{Q}_{q-1}=(1-q,r_2-1), & \ldots & \mathcal{Q}_0=(0,r_2-q).
\end{array}$$ Note that there are exactly $a_{ij}$ distinct lattice paths from $\mathcal{P}_i$ to $\mathcal{Q}_j$.
The Lindström-Gessel-Viennot lemma says that $$\text{det}\ [a_{ij}]=\sum_\sigma \text{sign}(\sigma),$$ where the sum is over tuples of non-intersecting lattice paths $
\sigma:\{\mathcal{P}_0,\ldots,\mathcal{P}_q\} \to \{\mathcal{Q}_0,\ldots,\mathcal{Q}_q\}
$ and $\text{sign}(\sigma)$ is the sign of the induced permutation of $\{0,1,\ldots,q\}$. In our case, any tuple of non-intersecting lattice paths $\sigma$ as above should go from $\mathcal{P}_i$ to $\mathcal{Q}_{q-i}$, and hence $$\text{sign}(\sigma)=(-1)^{q(q+1)/2}.$$ It is clear that there is at least one such tuple of non-intersecting lattice paths; for example, for all $\mathcal{P}_i$ one may first go east $r_1-q$ times and then go north $r_2-q$ times to arrive at $\mathcal{Q}_{q-i}$. This gives $$(-1)^{q(q+1)/2}\text{det}\ [a_{ij}]>0. \qedhere$$
Now we reduce Proposition \[TensorHR\] to the case of Lemma \[LemmaPP\]. We first introduce some useful notions to be used in the remaining part of the proof.
Let $R^*$ be a Poincaré duality algebra of dimension $r$, and let $\ell$ be an element of $R^1$.
Let $V^*$ be a graded subspace of $R^*$. We say that
(1) $V^*$ satisfies $\text{HL}(\ell)$ if $Q^q_\ell$ restricted to $V^q$ is nondegenerate for all nonnegative $q \le \frac{r}{2}$.
(2) $V^*$ satisfies $\text{HR}(\ell)$ if $Q^q_\ell$ restricted to $V^q$ is nondegenerate and has signature $$\sum_{p=0}^q (-1)^{q-p} \Big(\text{dim}_\mathbb{R} V^p - \text{dim}_\mathbb{R} V^{p-1}\Big) \ \ \text{for all nonnegative $q \le \frac{r}{2}$}.$$
Propositions \[HLCharacterization\] and \[HRCharacterization\] show that this agrees with the previous definition when $V^*=R^*$.
Let $V_1^*$ and $V_2^*$ be graded subspaces of $R^*$. We write $$V_1^* \perp_{\text{PD}} V_2^*$$ to mean that $V_1^* \cap V_2^* =0$ and $V_1^{r-q} \ V_2^{q}=0$ for all nonnegative integers $q \le r$, and write $$V_1^* \perp_{Q^*_\ell} V_2^*$$ to mean that $V_1^* \cap V_2^* =0$ and $Q^q_\ell(V_1^q,V_2^q)=0$ for all nonnegative integers $q \le \frac{r}{2}$.
We record here basic properties of the two notions of orthogonality. Let $S^*$ be another Poincaré duality algebra of dimension $s$.
\[LemmaOrthogonality\] Let $V_1^*, V_2^* \subseteq R^*$ and $W_1^*, W_2^* \subseteq S^*$ be graded subspaces.
(1) If $V_1^* \perp_{Q^*_\ell} V_2^*$ and if both $V_1^*$, $V_2^*$ satisfy $\text{HL}(\ell)$, then $V_1^* \oplus V_2^*$ satisfy $\text{HL}(\ell)$.
(2) If $V_1^* \perp_{Q^*_\ell} V_2^*$ and if both $V_1^*$, $V_2^*$ satisfy $\text{HR}(\ell)$, then $V_1^* \oplus V_2^*$ satisfy $\text{HR}(\ell)$.
(3) If $V_1^* \perp_{\text{PD}} V_2^*$ and if $\ell \hspace{0.5mm}V_1^* \subseteq V_1^*$, then $V_1^* \perp_{Q^*_\ell} V_2^*$.
(4) If $V_1^* \perp_{\text{PD}} V_2^*$, then $(V^*_1 \otimes_\mathbb{R} W_1^*) \perp_{\text{PD}} (V^*_2 \otimes_\mathbb{R} W_2^*)$.
The first two assertions are straightforward. We justify the third assertion: For any nonnegative integer $q \le \frac{r}{2}$, the assumption on $V_1^*$ implies $L_\ell^q V_1^q \subseteq V_1^{r-q}$, and hence $$Q_\ell^q(V_1^q,V_2^q) \subseteq \text{deg}(V_1^{r-q} V_2^q)=0.$$ For the fourth assertion, we check that, for any nonnegative integers $p_1,p_2,q_1,q_2$ whose sum is $r+s$, $$V^{p_1}_1V_2^{p_2} \otimes_\mathbb{R} W_1^{q_1}W_2^{q_2} =0.$$ The assumption on $V_1^*$ and $V_2^*$ shows that the first factor is trivial if $p_1+p_2 \ge r$, and the second factor is trivial if otherwise.
Suppose that $R_1^*$ satisfies $\text{HR}(\ell_1)$ and that $R_2^*$ satisfies $\text{HR}(\ell_2)$. We set $$R^*:=R^*_1 \otimes_\mathbb{R} R_2^*, \qquad \ell:=\ell_1 \otimes 1 + 1 \otimes \ell_2.$$ We show that $R^*$ satisfy $\text{HR}(\ell)$. The assertion on $\text{HL}$ can be proved in the same way.
For every $p \le \frac{r_1}{2}$, choose an orthogonal basis of $P_{\ell_1}^p \subseteq R_1^p$ with respect to $Q_{\ell_1}^p$: $$\Big\{v^p_{1},v^p_2,\ldots,v^p_{m(p)}\Big\} \subseteq P_{\ell_1}^p.$$ Similarly, for every $q \le \frac{r_2}{2}$, choose an orthogonal basis of $P_{\ell_2}^q \subseteq R_2^q$ with respect to $Q_{\ell_2}^q$: $$\Big\{w^q_{1},w^q_2,\ldots,w^q_{n(q)}\Big\} \subseteq P_{\ell_2}^q.$$ Here we use the upper indices to indicate the degrees of basis elements. To each pair of $v^p_i$ and $w^q_j$, we associate a graded subspace of $R^*$: [ $$\begin{gathered}
B^*(v_i^p,w_j^q):=B^*(v_{i}^{p}) \otimes_\mathbb{R} B^*(w_{j}^{q}), \ \ \text{where}\\
B^*(v_i^p):= \langle v_i^p\rangle \hspace{0.5mm}\oplus \hspace{0.5mm}\ell_1 \langle v_i^p\rangle\hspace{0.5mm} \oplus \cdots \oplus\hspace{0.5mm} \ell_1^{r_1-2p} \langle v_i^q\rangle \subseteq R^*_1,\\
B^*(w^q_j):= \langle w_j^q\rangle \oplus \ell_2 \langle w_j^q\rangle \oplus \cdots \oplus \ell^{r_2-2q} \langle w_j^q\rangle \subseteq R_2^*,\end{gathered}$$ ]{} Let us compare the tensor product $B^*(v_i^p,w_j^q)$ with the truncated polynomial ring $$S^*_{p,q}:=\mathbb{R}[x_1,x_2]/(x_1^{r_1-2p+1},x_2^{r_2-2q+1}).$$ The properties $\text{HR}(\ell_1)$ and $\text{HR}(\ell_2)$ show that, for every nonnegative integer $k
\le \frac{r_1+r_2-2p-2q}{2}
$, there is an isometry $$\Big( B^{k+p+q}(v_{i}^{p},w_{j}^{q}), \ Q_\ell^{k+p+q} \Big)
\simeq
\Big(S^k_{p,q}, \ (-1)^{p+q} \hspace{0.5mm}Q_{x_1+x_2}^k\Big).$$ Therefore, by Lemma \[LemmaPP\], the graded subspace $B^*(v_i^p,w_j^q) \subseteq R^*$ satisfies $\text{HR}(\ell)$.
The properties $\text{HL}(\ell_1)$ and $\text{HL}(\ell_2)$ imply that there is a direct sum decomposition $$R^*=\bigoplus_{p,q,i,j} B^*(v_i^p,w_j^q).$$ It is enough to prove that the above decomposition is orthogonal with respect to $Q_{\ell}^*$:
Any two distinct summands of $R^*$ satisfy $
B^*(v,w) \perp_{Q_\ell^*} B^*(v',w').
$
For the proof of the claim, we may suppose that $w \neq w'$. The orthogonality of the Lefschetz decomposition for $R_2^*$ with respect to $Q_{\ell_2}^*$ shows that $$B(w) \perp_{\text{PD}} B(w').$$ By the fourth assertion of Lemma \[LemmaOrthogonality\], the above implies $$B^*(v,w) \perp_{\text{PD}} B^*(v',w').$$ By the third assertion of Lemma \[LemmaOrthogonality\], this gives the claimed statement.
{#section-20}
Let $\Sigma$ be a unimodular fan in $N_\mathbb{R}$. For our purposes, it will be enough to assume that $\Sigma$ is simplicial.
We say that $\Sigma$ satisfies the *Poincaré duality of dimension $r$* if $A^*(\Sigma)_\mathbb{R}$ is a Poincaré duality algebra of dimension $r$.
In the remainder of this subsection, we suppose that $\Sigma$ satisfies the Poincaré duality of dimension $r$. We fix an isomorphism, called the *degree map* for $\Sigma$, $$\text{deg}: A^r(\Sigma)_\mathbb{R} \longrightarrow \mathbb{R}.$$ As before, we write $V_\Sigma$ for the set of primitive ray generators of $\Sigma$.
Note that for any nonnegative integer $q$ and $\mathbf{e}\in V_\Sigma$ there is a commutative diagram $$\xymatrixcolsep{5pc}
\xymatrixrowsep{3pc}
\xymatrix{
A^{q}(\Sigma) \ar[r]^{\textrm{p}_{\mathbf{e}}} \ar[dr]_{x_\mathbf{e} \cdot -}& A^q(\text{star}(\mathbf{e},\Sigma)) \ar[d]^{ x_\mathbf{e} \cdot -}\\
&A^{q+1}(\Sigma),
}$$ where $\mathrm{p}_\mathbf{e}$ is the pullback homomorphism $\mathrm{p}_{\mathbf{e}\in \Sigma}$ and $x_\mathbf{e}\cdot -$ are the multiplications by $x_\mathbf{e}$. It follows that there is a surjective graded ring homomorphism $$\pi_{\mathbf{e}}: A^*(\text{star}(\mathbf{e},\Sigma)) \longrightarrow A^*(\Sigma)/\text{ann}(x_\mathbf{e}).$$
The star of $\mathbf{e}$ in $\Sigma$ satisfies the Poincaré duality of dimension $r-1$ if and only if $\pi_{\mathbf{e}}$ is an isomorphism: $$A^*(\text{star}(\mathbf{e},\Sigma)) \simeq A^*(\Sigma)/\text{ann}(x_\mathbf{e}).$$
The ”if” direction follows from Proposition \[PropositionLocalDuality\]: The quotient $A^*(\Sigma)/\text{ann}(x_\mathbf{e})$ is a Poincaré duality algebra of dimension $r-1$.
The ”only if” direction follows from the observation that any surjective graded ring homomorphism between Poincaré duality algebras of the same dimension is an isomorphism.
Let $\Sigma$ be a fan that satisfies Poincaré duality of dimension $r$. We say that
(1) $\Sigma$ satisfies the *hard Lefschetz property* if $A^*(\Sigma)_\mathbb{R}$ satisfies $\text{HL}(\ell)$ for all $\ell \in \mathscr{K}_\Sigma$,
(2) $\Sigma$ satisfies the *Hodge-Riemann relations* if $A^*(\Sigma)_\mathbb{R}$ satisfies $\text{HR}(\ell)$ for all $\ell \in \mathscr{K}_\Sigma$, and
(3) $\Sigma$ satisfies the *local Hodge-Riemann relations* if the Poincaré duality algebra $$A^*(\Sigma)_\mathbb{R}/\text{ann}(x_\mathbf{e})$$ satisfies $\text{HR}(\ell_\mathbf{e})$ with respect to the degree map induced by $x_\mathbf{e}$ for all $\ell \in \mathscr{K}_\Sigma$ and $\mathbf{e} \in V_\Sigma$.
Hereafter we write $\ell_\mathbf{e}$ for the image of $\ell$ in the quotient $A^*(\Sigma)_\mathbb{R}/\text{ann}(x_\mathbf{e})$.
\[lHRtoHL\] If $\Sigma$ satisfies the local Hodge-Riemann relations, then $\Sigma$ satisfies the hard Lefschetz property.
By definition, for $\ell \in \mathscr{K}_\Sigma$ there are positive real numbers $c_\mathbf{e}$ such that $$\ell= \sum_{\mathbf{e} \in V_\Sigma} c_\mathbf{e} \hspace{0.5mm} x_\mathbf{e} \in A^1(\Sigma)_\mathbb{R}.$$ We need to show that the Lefschetz operator $L^q_\ell$ on $A^q(\Sigma)_\mathbb{R}$ is injective for all $q \le \frac{r}{2}$. Nothing is claimed when $r=2q$, so we may assume that $r-2q$ is positive.
Let $f$ be an element in the kernel of $L_\ell^q$, and write $f_\mathbf{e}$ for the image of $f$ in the quotient $A^q(\Sigma)_\mathbb{R}/\text{ann}(x_\mathbf{e})$. Note that the element $f$ has the following properties:
(1) For all $\mathbf{e} \in V_\Sigma$, the image $f_\mathbf{e}$ belongs to the primitive subspace $P_{\ell_\mathbf{e}}^q$, and
(2) for the positive real numbers $c_\mathbf{e}$ as above, we have $$\sum_{\mathbf{e} \in V_\Sigma} c_\mathbf{e} \hspace{0.5mm} Q_{\ell_\mathbf{e}}^q(f_\mathbf{e},f_\mathbf{e})=Q_\ell^q(f,f)=0.$$
By the local Hodge-Riemann relations, the two properties above show that all the $f_\mathbf{e}$ are zero: $$x_\mathbf{e} \cdot f =0 \in A^*(\Sigma)_\mathbb{R} \ \ \text{for all $\mathbf{e} \in V_\Sigma$.}$$ Since the elements $x_\mathbf{e}$ generate the Poincaré duality algebra $A^*(\Sigma)_\mathbb{R}$, this implies that $f=0$.
\[ForSomeForAll\] If $\Sigma$ satisfies the hard Lefschetz property, then the following are equivalent:
(1) $A^*(\Sigma)_\mathbb{R}$ satisfies $\text{HR}(\ell)$ for some $\ell \in \mathscr{K}_\Sigma$.
(2) $A^*(\Sigma)_\mathbb{R}$ satisfies $\text{HR}(\ell)$ for all $\ell \in \mathscr{K}_\Sigma$.
Let $\ell_0$ and $\ell_1$ be elements of $\mathscr{K}_\Sigma$, and suppose that $A^*(\Sigma)_\mathbb{R}$ satisfies $\text{HR}(\ell_0)$. Consider the parametrized family $$\ell_t:= (1-t)\hspace{0.5mm} \ell_0+t\hspace{0.5mm} \ell_1, \qquad 0 \le t \le 1.$$ Since $\mathscr{K}_\Sigma$ is convex, the elements $\ell_t$ are ample for all $t$.
Note that $Q_{\ell_t}^q$ are nondegenerate on $A^q(\Sigma)_\mathbb{R}$ for all $t$ and $q \le \frac{r}{2}$ because $\Sigma$ satisfies the hard Lefschetz property. It follows that the signatures of $Q_{\ell_t}^q$ should be independent of $t$ for all $q \le \frac{r}{2}$. Since $A^*(\Sigma)_\mathbb{R}$ satisfies $\text{HR}(\ell_0)$, the common signature should be $$\sum_{p=0}^q (-1)^{q-p} \Big(\text{dim}_\mathbb{R} \ A^p(\Sigma)_\mathbb{R} -\text{dim}_\mathbb{R} \ A^{p-1}(\Sigma)_\mathbb{R} \Big).$$ We conclude by Proposition \[HRCharacterization\] that $A^*(\Sigma)_\mathbb{R}$ satisfies $\text{HR}(\ell_1)$.
Proof of the main theorem
=========================
{#section-21}
As a final preparation for the proof of the main theorem, we show that the property $\text{HR}$ is preserved by a matroidal flip for particular choices of ample classes.
Let $\mathrm{M}$ be as before, and consider the matroidal flip from $\mathscr{P}_-$ to $\mathscr{P}_+$ with center $Z$. We will use the following homomorphisms:
(1) The pullback homomorphism $\Phi_Z:A^*(\mathrm{M},\mathscr{P}_-) \longrightarrow A^*(\mathrm{M},\mathscr{P}_+)$.
(2) The Gysin homomorphisms $\Psi_{Z}^{p,q}: A^{q-p}(\mathrm{M}_{Z}) \longrightarrow A^{q}(\mathrm{M},\mathscr{P}_+)$.
(3) The pullback homomorphism $\mathrm{p}_{{Z}}: A^*(\mathrm{M},\mathscr{P}_-) \longrightarrow A^*(\mathrm{M}_Z)$.
The homomorphism $\mathrm{p}_{Z}$ is the graded ring homomorphism $\mathrm{p}_{\sigma_{Z<\varnothing}\in \Sigma_{\mathrm{M},\mathscr{P}_-}}$ obtained from the identification $$\text{star}(\sigma_{Z<\varnothing},\Sigma_{\mathrm{M},\mathscr{P}_-}) \simeq \Sigma_{\mathrm{M}_Z}.$$ In the remainder of this section, we fix a strictly convex piecewise linear function $\ell_-$ on $\Sigma_{\mathrm{M},\mathscr{P}_-}$. For nonnegative real numbers $t$, we set $$\ell_+(t):=\Phi_Z(\ell_-)-tx_Z \in A^1(\mathrm{M},\mathscr{P}_+) \otimes_\mathbb{Z} \mathbb{R}.$$ We write $\ell_Z$ for the pullback of $\ell_-$ to the star of the cone $\sigma_{Z<\varnothing}$ in the Bergman fan $\Sigma_{\mathrm{M},\mathscr{P}_-}$: $$\ell_Z:=\mathrm{p}_Z(\ell_-) \in A^1(\mathrm{M}_Z) \otimes_\mathbb{Z} \mathbb{R}.$$ Proposition \[PropositionAmplePullback\] shows that $\ell_Z$ is the class of a strictly convex piecewise linear function on $\Sigma_{\mathrm{M}_Z}$.
$\ell_+(t)$ is strictly convex for all sufficiently small positive $t$.
It is enough to show that $\ell_+(t)$ is strictly convex around a given cone $\sigma_{I<\mathscr{F}}$ in $\Sigma_{\mathrm{M},\mathscr{P}_+}$.
When $Z \notin \mathscr{F}$, the cone $\sigma_{I<\mathscr{F}}$ is in the fan $\Sigma_{\mathrm{M},\mathscr{P}_-}$, and hence we may suppose that $$\text{$\ell_-$ is zero on $\sigma_{I<\mathscr{F}}$ and positive on the link of $\sigma_{I<\mathscr{F}}$ in $\Sigma_{\mathrm{M},\mathscr{P}_-}$.}$$ It is straightforward to deduce from the above that, for all sufficiently small positive $t$, $$\text{$\ell_+(t)$ is zero on $\sigma_{I<\mathscr{F}}$ and positive on the link of $\sigma_{I<\mathscr{F}}$ in $\Sigma_{\mathrm{M},\mathscr{P}_+}$}.$$ More precisely, the statement is valid for all $t$ that satisfies the inequalities $$0<t<\sum_{i \in Z \setminus I} \ell_-(\mathbf{e}_i).$$ Note that $Z \setminus I$ is nonempty and each of the summands in the right-hand side is positive.
When $Z \in \mathscr{F}$, the cone $\sigma_{Z<\mathscr{F} \setminus \{Z\}}$ is in the fan $\Sigma_{\mathrm{M},\mathscr{P}_-}$, and hence we may suppose that $$\text{$\ell_-$ is zero on $\sigma_{Z<\mathscr{F} \setminus \{Z\}}$ and positive on the link of $\sigma_{Z<\mathscr{F} \setminus \{Z\}}$ in $\Sigma_{\mathrm{M},\mathscr{P}_-}$.}$$ Let $J$ be the flat $\text{min} \hspace{0.5mm} \mathscr{F} \setminus \{Z\}$, and let $m(t)$ be the linear function on $N_E$ defined by setting $$\mathbf{e}_i \longmapsto
\left\{\begin{array}{cl}
\frac{t}{|Z \setminus I|} & \text{if $i \in Z \setminus I$}, \\
\frac{-t}{|J \setminus Z|} & \text{if $i \in J \setminus Z$},\\
0 & \text{if otherwise.}
\end{array}\right.$$ It is straightforward to deduce from the above that, for all sufficiently small positive $t$, $$\text{$\ell_+(t)+m(t)$ is zero on $\sigma_{I<\mathscr{F}}$ and positive on the link of $\sigma_{I<\mathscr{F}}$ in $\Sigma_{\mathrm{M},\mathscr{P}_+}$.}$$ More precisely, the latter statement is valid for all $t$ that satisfies the inequalities $$0<t< \text{min}\Big\{\ell_-(\mathbf{e}_F), \ \text{$\mathbf{e}_F$ is in the link of $\sigma_{Z<\mathscr{F} \setminus \{Z\}}$ in $\Sigma_{\mathrm{M},\mathscr{P}_-}$} \Big\}.$$ Here the minimum of the empty set is defined to be $\infty$.
We write “$\text{deg}$” for the degree map of $\mathrm{M}$ and of $\mathrm{M}_Z$, and fix the degree maps $$\begin{aligned}
\text{deg}_+: A^r(\mathrm{M},\mathscr{P}_+) \longrightarrow \mathbb{Z}, \qquad a \longmapsto \text{deg}\big(\Phi_{\mathscr{P}^c_+}(a)\big), \\
\text{deg}_-: A^r(\mathrm{M},\mathscr{P}_-) \longrightarrow \mathbb{Z}, \qquad a \longmapsto \text{deg}\big(\Phi_{\mathscr{P}^c_-}(a)\big), \end{aligned}$$ see Definition \[DefinitionIntermediateDegree\]. We omit the subscripts $+$ and $-$ from the notation when there is no danger of confusion. The goal of this subsection is to prove the following.
\[BlowupHR\] Let $\ell_-$, $\ell_Z$, and $\ell_+(t)$ be as above, and suppose that
(1) the Chow ring of $\Sigma_{\mathrm{M},\mathscr{P}_-}$ satisfies $\text{HR}(\ell_-)$, and
(2) the Chow ring of $\Sigma_{\mathrm{M}_Z}$ satisfies $\text{HR}(\ell_Z)$.
Then the Chow ring of $\Sigma_{\mathrm{M},\mathscr{P}_+}$ satisfies $\text{HR}(\ell_+(t))$ for all sufficiently small positive $t$.
Hereafter we suppose $\text{HR}(\ell_-)$ and $\text{HR}(\ell_Z)$. We introduce the main characters appearing in the proof of Proposition \[BlowupHR\]:
(1) A Poincaré duality algebra of dimension $r$: $$\hspace{-30mm}
A^*_+:=\bigoplus_{q=0}^r\ A^q_+, \qquad A_+^q:=A^q(\mathrm{M},\mathscr{P}_+) \otimes_\mathbb{Z} \mathbb{R}.$$
(2) A Poincaré duality algebra of dimension $r$: $$\hspace{-35mm}
A^*_- := \bigoplus_{q =0}^r\ A^q_-, \qquad A^q_-:= \Big(\text{im}\ \Phi_Z^q\Big) \otimes_\mathbb{Z} \mathbb{R}.$$
(3) A Poincaré duality algebra of dimension $r-2$: $$T^*_Z:=\bigoplus_{q=0}^{r-2} T^q_Z, \ \qquad T^q_Z:=\Big(\mathbb{Z}[x_Z]/(x_Z^{\text{rk}(Z)-1}) \otimes_\mathbb{Z} A^*(\mathrm{M}_Z)\Big)^q \otimes_\mathbb{Z} \mathbb{R}.$$
(4) A graded subspace of $A^*_+$, the sum of the images of the Gysin homomorphisms: $$\hspace{-25mm}G^*_Z:=\bigoplus_{q = 1}^{r-1} G^q_Z, \qquad G^q_Z:=\bigoplus_{p=1}^{\text{rk}(Z)-1} \Big(\text{im}\ \Psi_Z^{p,q}\Big) \otimes_\mathbb{Z} \mathbb{R}.$$
The truncated polynomial ring in the definition of $T^*_Z$ is given the degree map $$(-x_Z)^{\text{rk}(Z)-2} \longmapsto 1,$$ so that the truncated polynomial ring satisfies $\text{HR}(-x_Z)$. The tensor product $T^*_Z$ is given the induced degree map $$(-x_Z)^{\text{rk}(Z)-2} x_\mathscr{Z}\longmapsto 1.$$ It follows from Proposition \[TensorHR\] that the tensor product satisfies $\text{HR}( 1 \otimes \ell_Z -x_Z \otimes 1)$.
For nonnegative $q \le \frac{r}{2}$, we write the Poincaré duality pairings for $A^*_-$ and $T^*_Z$ by $$\begin{aligned}
&\big\langle -,- \big\rangle^q_{A^*_-}: A_-^q \times A^{r-q}_- \longrightarrow \mathbb{R}, \\
&\big\langle -,- \big\rangle^{q-1}_{T^*_Z}: T^{q-1}_Z \times T^{r-q-1}_Z \longrightarrow \mathbb{R}.\end{aligned}$$ We omit the superscripts $q$ and $q-1$ from the notation when there is no danger of confusion.
Theorem \[DecompositionTheorem\] shows that $\Phi_Z$ defines an isomorphism between the graded rings $$A^*(\mathrm{M},\mathscr{P}_-) \otimes_\mathbb{Z} \mathbb{R} \simeq A_-^*,$$ and that there is a decomposition into a direct sum $$A_+^* = A_-^* \oplus G_Z^*.$$ In addition, it shows that $x_Z \cdot -$ is an isomorphism between the graded vector spaces $$T_Z^{*} \simeq G_Z^{*+1}.$$ The inverse of the isomorphism $x_Z \cdot -$ will be denoted $x_Z^{-1} \cdot -$.
We equip the above graded vector spaces with the following symmetric bilinear forms.
Let $q$ be a nonnegative integer $\le \frac{r}{2}$.
(1) $\big(A^q_+,Q^q_- \oplus Q^q_Z\big)$: $Q^q_- $ and $Q^q_Z$ are the bilinear forms on $A^q_-$ and $G^q_Z$ defined below.
(2) $\big(A_-^q,Q^q_- \big)$: $Q^q_-$ is the restriction of the Hodge-Riemann form $Q^q_{\ell_+(0)}$ to $A_-^q$.
(3) $\big(T_Z^q,Q_\mathscr{T}^q\big)$: $Q_\mathscr{T}^q$ is the Hodge-Riemann form associated to $\mathscr{T}:=\Big(1 \otimes \ell_Z - x_Z \otimes 1\Big) \in T^1_Z$.
(4) $\big(G_Z^q,Q_Z^q\big)$: $Q_Z^q$ is the bilinear form defined by saying that $x_Z \cdot -$ gives an isometry $$\Big(T_Z^{q-1}, Q^{q-1}_{\mathscr{T}}\Big) \simeq \Big(G^{q}_Z\hspace{0.5mm},\hspace{0.5mm} Q_Z^{q}\Big).$$
We observe that $Q^q_- \oplus Q_Z^q$ satisfies the following version of Hodge-Riemann relations:
\[SumOfSignatures\] The bilinear form $Q^q_- \oplus Q_Z^q$ is nondegenerate on $A^q_+$ and has signature $$\sum_{p=0}^q (-1)^{q-p}\Big(\text{dim}_\mathbb{R} A^p_+-\text{dim}_\mathbb{R} A^{p-1}_+ \Big) \ \ \text{for all nonnegative $q \le \frac{r}{2}$.}$$
Theorem \[DecompositionTheorem\] shows that $\Phi_Z \otimes_\mathbb{Z} \mathbb{R}$ defines an isometry $$\Big( A^q(\mathrm{M},\mathscr{P}_-)_\mathbb{R} \hspace{0.5mm},\hspace{0.5mm} Q^q_{\ell_-}\Big) \simeq \Big(A^q_- \hspace{0.5mm},\hspace{0.5mm} Q^q_{-}\Big).$$ It follows from the assumption on $\Sigma_{\mathrm{M},\mathscr{P}_-}$ that $Q^q_-$ is nondegenerate on $A^q_-$ and has signature $$\sum_{p=0}^q (-1)^{q-p}\Big(\text{dim}_\mathbb{R} A^p_- - \text{dim}_\mathbb{R} A^{p-1}_- \Big).$$ It follows from the assumption on $\Sigma_{\mathrm{M}_Z}$ that $Q^q_Z$ is nondegenerate on $G^q_Z$ and has signature $$\begin{aligned}
\sum_{p=0}^{q-1} (-1)^{q-p-1}\Big(\text{dim}_\mathbb{R} T^p_Z- \text{dim}_\mathbb{R} T^{p-1}_Z \Big)&=
\sum_{p=0}^{q-1} (-1)^{q-p-1}\Big(\text{dim}_\mathbb{R} G^{p+1}_Z- \text{dim}_\mathbb{R} G^{p}_Z \Big)\\
&=\sum_{p=0}^{q} (-1)^{q-p}\Big(\text{dim}_\mathbb{R} G^p_Z- \text{dim}_\mathbb{R} G^{p-1}_Z \Big).\end{aligned}$$ The assertion is deduced from the fact that the signature of the sum is the sum of the signatures.
We now construct a continuous family of symmetric bilinear forms $Q^q_t$ on $A^q_+$ parametrized by positive real numbers $t$. This family $Q^q_t$ will shown to have the following properties:
(1) For every positive real number $t$, there is an isometry $$\Big(A_+^q\hspace{0.5mm} ,\hspace{0.5mm} Q^q_t\Big) \simeq \Big(A_+^q\hspace{0.5mm} ,\hspace{0.5mm} Q^q_{\ell_+(t)}\Big).$$
(2) The sequence $Q^q_t$ as $t$ goes to zero converges to the sum of $Q^q_{-}$ and $Q^q_{Z}$: $$\lim_{t \to 0} Q^q_t=Q^q_{-} \oplus Q^q_{Z}.$$
For positive real numbers $t$, we define a graded linear transformation $$S_t: A_+^* \longrightarrow A_+^*$$ to be the sum of the identity on $A_-^*$ and the linear transformations $$\Big(\text{im}\ \Psi_Z^{p,q}\Big) \otimes_\mathbb{Z} \mathbb{R} \longrightarrow \Big(\text{im}\ \Psi_Z^{p,q} \Big) \otimes_\mathbb{Z} \mathbb{R}, \qquad a \longmapsto t^{-\frac{\text{rk}(Z)}{2}+p} \ a.$$ The inverse transformation $S_t^{-1}$ is the sum of the identity on $A^*_-$ and the linear transformations $$\Big(\text{im}\ \Psi_Z^{p,q}\Big) \otimes_\mathbb{Z} \mathbb{R} \longrightarrow \Big(\text{im}\ \Psi_Z^{p,q} \Big) \otimes_\mathbb{Z} \mathbb{R}, \qquad a \longmapsto t^{\frac{\text{rk}(Z)}{2}-p} \ a.$$
The symmetric bilinear form $Q^q_t$ is defined so that $S_t$ defines an isometry $$\Big(A_+^q\hspace{0.5mm} ,\hspace{0.5mm} Q^q_t \Big)\simeq \Big(A_+^q\hspace{0.5mm} ,\hspace{0.5mm} Q^q_{\ell_+(t)}\Big) \ \ \text{for all nonnegative integers $q \le \frac{r}{2}$.}$$ In other words, for any elements $a_1,a_2 \in A_+^q$, we set $$Q^q_t(a_1,a_2):=(-1)^q \hspace{0.5mm} \text{deg} \big( S_t(a_1) \cdot \ell_+(t)^{r-2q} \cdot S_t(a_2)\big).$$
The first property of $Q^q_t$ mentioned above is built into the definition. We verify the assertion on the limit of $Q^q_t$ as $t$ goes to zero.
\[PropositionLimit\] For all nonnegative integers $q \le \frac{r}{2}$, we have $$\lim_{t \to 0} Q^q_t=Q^q_{-} \oplus Q^q_{Z}.$$
We first construct a deformation of the Poincaré duality pairing $A_+^q \times A^{r-q}_+ \longrightarrow \mathbb{R}$: $$\big\langle a_1,a_2 \big\rangle^q_t:= \text{deg}\big(S_t(a_1),S_t(a_2)\big), \qquad t>0.$$ We omit the upper index $q$ when there is no danger of confusion.
For any $ b_1,b_2 \in A^*_-$ and $c_1,c_2 \in G^*_Z$ and $a_1=b_1+c_1,a_2=b_2+c_2 \in A^*_+$, $$\big\langle a_1,a_2\big\rangle_0:=\lim_{t \to 0}\big\langle a_1,a_2\big\rangle_t= \big\langle b_1,b_2\big\rangle_{A^*_-} - \big\langle x_Z^{-1} c_1, x_Z^{-1} c_2\big\rangle_{T^*_Z}.$$
We write $z:=\text{rk}(Z)$ and choose bases of $A_+^q$ and $A^{r-q}_+$ that respect the decompositions $$\begin{aligned}
A^q_+&=A^q_- \oplus\Big( \text{im} \ \Psi_Z^{1,q} \oplus \text{im} \ \Psi_Z^{2,q} \oplus \cdots \oplus \text{im} \ \Psi_Z^{z-1,q}\Big) \otimes_\mathbb{Z} \mathbb{R}, \ \ \text{and}
\\[4pt]
A^{r-q}_+&= A_-^{r-q}\oplus\Big( \text{im} \ \Psi_Z^{z-1,r-q} \oplus \text{im} \ \Psi_Z^{z-2,r-q} \oplus \cdots \oplus \text{im} \ \Psi_Z^{1,r-q}\Big) \otimes_\mathbb{Z} \mathbb{R}.\end{aligned}$$ Let $\mathscr{M}_-$ be the matrix of the Poincaré duality pairing between $A^q_-$ and $A^{r-q}_-$, and let $\mathscr{M}_{p_1,p_2}$ is the matrix of the Poincaré duality pairing between $\text{im}\ \Psi_Z^{p_1,q} \otimes_\mathbb{Z}\mathbb{R}$ and $ \text{im}\ \Psi_Z^{p_2,r-q} \otimes_\mathbb{Z}\mathbb{R}$. Lemma \[LemmaUpperTriangular\] shows that the matrix of the deformed Poincaré pairing on $A^*_+$ is $$\left[\begin{array}{cccccc}
\mathscr{M}_- & 0 &0 & 0& \cdots & 0 \\
0 & \mathscr{M}_{1,z-1} & t \mathscr{M}_{2,z-1} & t^2 \mathscr{M}_{3,z-1} & \cdots & t^{z-2} \mathscr{M}_{z-1,z-1} \\
0 &0& \mathscr{M}_{2,z-2} & t \mathscr{M}_{3,z-2} & & t^{z-3} \mathscr{M}_{z-1,z-2}\\
0 & 0 &0 & \mathscr{M}_{3,z-3} & \cdots & t^{z-4} \mathscr{M}_{z-1,z-3} \\
\vdots &\vdots&\vdots&\vdots& \ddots & \vdots\\
0 & 0 &0& 0& 0 & \mathscr{M}_{z-1,1}\\
\end{array}\right].$$ The claim on the limit of the deformed Poincaré duality pairing follows. The minus sign on the right-hand side of the claim comes from the following computation made in Proposition \[GysinTopIsomorphism\]: $$\text{deg}\big(x_Z^{\text{rk}(Z)}x_\mathscr{Z}\big)=(-1)^{\text{rk}(Z)-1}.$$
We use the deformed Poincaré duality pairing to understand the limit of the bilinear form $Q^q_t$. For an element $a$ of $A^1_+$, we write the multiplication with $a$ by $$M^a:A^*_+\longrightarrow A^{*+1}_+,\qquad x \longmapsto a \cdot x,$$ and define its deformation $M^{a}_t:=S_t^{-1} \circ M^a \circ S_t$. In terms of the operator $M^{\ell_+(t)}_t$, the bilinear form $Q^q_t$ can be written $$\begin{aligned}
Q^q_t(a_1,a_2)&=(-1)^q \hspace{0.5mm}\text{deg}\Big(S_t(a_1)\cdot M^{\ell_+(t)} \circ \cdots \circ M^{\ell_+(t)} \circ S_t \ (a_2)\Big)\\
&=(-1)^q \hspace{0.5mm}\text{deg}\Big(S_t(a_1)\cdot S_t \circ M^{\ell_+(t)}_t \circ \cdots \circ M^{\ell_+(t)}_t \ (a_2)\Big)\\
&=(-1)^q \hspace{0.5mm} \Big\langle a_1 \hspace{0.5mm} ,\hspace{0.5mm} M^{\ell_+(t)}_t \circ \cdots \circ M^{\ell_+(t)}_t \ (a_2) \Big\rangle_t\end{aligned}$$ Define linear operators $M^{1 \otimes \ell_Z}$, $M^{x_Z \otimes 1}$, and $M^\mathscr{T}$ on $G^*_Z$ by the isomorphisms $$\begin{aligned}
&\Big(G^*_Z,M^{1 \otimes \ell_Z}\Big) \simeq \Big(T^{*-1},1 \otimes \ell_Z \cdot -\Big),\\
&\Big(G^*_Z,M^{x_Z \otimes 1}\Big) \simeq \Big(T^{*-1},x_Z \otimes 1 \cdot -\Big),\\
&\Big(G^*_Z,M^{\mathscr{T}}\Big) \simeq \Big(T^{*-1},\mathscr{T} \cdot -\Big).\end{aligned}$$ Note that the linear operator $M^\mathscr{T}$ is the difference $M^{1 \otimes \ell_Z}-M^{x_Z \otimes 1}$.
The limit of the operator $M^{\ell_+(t)}_t$as $t$ goes to zero decomposes into the sum $$\Big(A^*_+\hspace{0.5mm} ,\hspace{0.5mm} \lim_{t\to 0} M^{\ell_+(t)}_t \Big)=\Big(A^*_- \oplus G^*_Z \hspace{0.5mm} ,\hspace{0.5mm} M^{\ell_+(0)} \oplus M^\mathscr{T} \Big).$$
Assuming the second claim, we finish the proof as follows: We have $$\lim_{t\to 0} Q_t^q(a_1,a_2)=(-1)^q \hspace{0.5mm} \lim_{t \to 0} \Big\langle a_1 \hspace{0.5mm} ,\hspace{0.5mm} M^{\ell_+(t)}_t \circ \cdots \circ M^{\ell_+(t)}_t \ (a_2) \Big\rangle_t$$ and from the first and the second claim, we see that the right-hand side is $$(-1)^q \hspace{0.5mm} \Big\langle a_1 \hspace{0.5mm} ,\hspace{0.5mm} (M^{\ell_+(0)} \oplus M^\mathscr{T} ) \circ \cdots \circ (M^{\ell_+(0)} \oplus M^\mathscr{T} ) \ (a_2) \Big\rangle_0 \\
=Q^q_-(b_1,b_2) + Q^q_Z(c_1,c_2),$$ where $a_i=b_i+c_i$ for $ b_i \in A^*_-$ and $c_i \in G^*_Z$. Notice that the minus sign in the first claim cancels with $(-1)^{q-1}$ in the Hodge-Riemann form $$\Big(T_Z^{q-1}, Q^{q-1}_{\mathscr{T}}\Big) \simeq \Big(G^{q}_Z\hspace{0.5mm},\hspace{0.5mm} Q_Z^{q}\Big).$$
We now prove the second claim made above. Write $M_t^{\ell_+(t)}$ as the difference $$M_t^{\ell_+(t)}= S_t^{-1} \circ M^{\ell_+(t)} \circ S_t=S_t^{-1} \circ \Big(M^{\ell_+(0)}- M^{tx_Z}\Big) \circ S_t=M_t^{\ell_+(0)}- M_t^{tx_Z}.$$ By Lemma \[LemmaUpperTriangular\], the operators $M^{\ell_+(0)}$ and $S_t$ commute, and hence $$\Big( A^*_+\hspace{0.5mm} ,\hspace{0.5mm} M^{\ell_+(0)}_t\Big)=\Big(A^*_+\hspace{0.5mm} ,\hspace{0.5mm} M^{\ell_+(0)}\Big)=\Big( A^*_- \oplus G^*_Z\hspace{0.5mm} ,\hspace{0.5mm} M^{\ell_+(0)} \oplus M^{1 \otimes \ell_Z}\Big).$$ Lemma \[LemmaUpperTriangular\] shows that the matrix of $M^{x_Z}$ in the chosen bases of $A^q_+$ and $A^{q+1}_+$ is of the form $$\left[
\begin{array}{cccccc}
0& 0 &0 & \cdots & 0 & B_0\\
A &0& 0 & \cdots & 0 & B_1\\
0 &\text{Id} &0& \cdots & 0 & B_2\\
\vdots &\vdots & \ddots &\ddots& \vdots & \vdots \\
0 &0 &\cdots &\text{Id} &0 & B_{z-2} \\
0 &0 &\cdots & 0 &\text{Id} &B_{z-1}
\end{array}
\right],$$ where “$\text{Id}$” are the identity matrices representing $$A^{q-p}(\mathrm{M}_Z)_\mathbb{R} \simeq \text{im} \ \Psi_Z^{p,q} \longrightarrow \text{im} \ \Psi_Z^{p+1,q+1} \simeq A^{q-p}(\mathrm{M}_Z)_\mathbb{R}.$$ Note that the matrix of the deformed operator $M^{tx_Z}_t$ can be written $$\left[
\begin{array}{cccccc}
0& 0 &0 & \cdots & 0 & t^{\frac{\text{rk}(Z)}{2}} B_0\\
t^{\frac{\text{rk}(Z)}{2}}A &0& 0&\cdots & 0 & t^{\text{rk}(Z)-1}B_1\\
0 &\text{Id} &0&\cdots &0 & t^{\text{rk}(Z)-1}B_2\\
\vdots & \vdots & \ddots &\ddots& \vdots & \vdots \\
0 & 0&\cdots &\text{Id} &0 &t^2B_{z-2} \\
0 &0 & \cdots & 0 &\text{Id} &tB _{z-1}
\end{array}
\right].$$ At the limit $t= 0$, the matrix represents the sum $0\oplus M^{x_Z\otimes 1}$, and therefore $$\begin{aligned}
\Big(A^*_+\hspace{0.5mm} ,\hspace{0.5mm} \lim_{t\to 0} M^{\ell_+(t)}_t\Big)&=\Big( A^*_- \oplus G^*_Z\hspace{0.5mm} ,\hspace{0.5mm} M^{\ell_+(0)} \oplus M^{1 \otimes \ell_Z}\Big)-
\Big( A^*_- \oplus G^*_Z\hspace{0.5mm} ,\hspace{0.5mm} 0\oplus M^{x_Z\otimes 1} \Big) \\
& =\Big( A^*_- \oplus G^*_Z\hspace{0.5mm} ,\hspace{0.5mm} M^{\ell_+(0)}\oplus M^{\mathscr{T}}\Big). \end{aligned}$$ This completes the proof of the second claim.
By Proposition \[SumOfSignatures\] and Proposition \[PropositionLimit\], we know that $\lim_{t \to 0} Q^q_t$ is nondegenerate on $A^q_+$ and has signature $$\sum_{p=0}^q (-1)^{q-p}\Big(\text{dim}_\mathbb{R} A^p_+-\text{dim}_\mathbb{R} A^{p-1}_+ \Big) \ \ \text{for all nonnegative $q \le \frac{r}{2}$.}$$ Therefore the same must be true for $Q^q_t$ for all sufficiently small positive $t$. By construction, there is an isometry $$\Big(A_+^q\hspace{0.5mm} ,\hspace{0.5mm} Q^q_t \Big)\simeq \Big(A_+^q\hspace{0.5mm} ,\hspace{0.5mm} Q^q_{\ell_+(t)}\Big),$$ and thus $A^*_+$ satisfies $\text{HR}(\ell_+(t))$ for all sufficiently small positive $t$.
{#section-22}
We are now ready to prove the main theorem. We write “$\text{deg}$” for the degree map of $\mathrm{M}$ and, for an order filter $\mathscr{P}$ of $\mathscr{P}_\mathrm{M}$, fix an isomorphism $$A^r(\mathrm{M},\mathscr{P}) \longrightarrow \mathbb{Z}, \qquad a \longmapsto \text{deg}\big(\Phi_{\mathscr{P}^c}(a)\big).$$
\[MainTheoremBody\] Let $\mathrm{M}$ be a loopless matroid, and let $\mathscr{P}$ be an order filter of $\mathscr{P}_\mathrm{M}$.
(1) The Bergman fan $\Sigma_{\mathrm{M},\mathscr{P}}$ satisfies the hard Lefschetz property.
(2) The Bergman fan $\Sigma_{\mathrm{M},\mathscr{P}}$ satisfies the Hodge-Riemann relations.
When $\mathscr{P}=\mathscr{P}_\mathrm{M}$, the above implies Theorem \[MainTheoremIntroduction\] in the introduction because any strictly submodular function defines a strictly convex piecewise linear function on $\Sigma_\mathrm{M}$.
We prove Theorem \[MainTheoremBody\] by lexicographic induction on the rank of $\mathrm{M}$ and the cardinality of $\mathscr{P}$. Set $\mathscr{P}=\mathscr{P}_+$, and consider the matroidal flip from $\mathscr{P}_-$ to $\mathscr{P}_+$ with center $Z$.
By Proposition \[PropositionSimplification\] and Proposition \[CombinatorialGeometry\], we may replace $\mathrm{M}$ by the associated combinatorial geometry $\overline{\mathrm{M}}$. Thus we may assume that $\mathrm{M}$ has no rank $1$ flat of cardinality greater than $1$. In this case, Proposition \[Star\] shows that the star of every ray in $\Sigma_{\mathrm{M},\mathscr{P}}$ is a product of at most two smaller Bergman fans to which the induction hypothesis applies.
By Proposition \[TensorHR\] and the induction hypothesis applied to the stars, $\Sigma_{\mathrm{M},\mathscr{P}}$ satisfies the local Hodge-Riemann relations. By Proposition \[lHRtoHL\], this implies that $\Sigma_{\mathrm{M},\mathscr{P}}$ satisfies the hard Lefschetz property.
Lastly, we show that $\Sigma_{\mathrm{M},\mathscr{P}}$ satisfies the Hodge-Riemann relations. Since $\Sigma_{\mathrm{M},\mathscr{P}}$ satisfies the hard Lefschetz property, Proposition \[ForSomeForAll\] shows that it is enough to prove $\text{HR}(\ell)$ for some $\ell \in \mathscr{K}_{\mathrm{M},\mathscr{P}}$. This follows from Proposition \[BlowupHR\] and the induction hypothesis applied to $\Sigma_{\mathrm{M},\mathscr{P}_-}$ and $\Sigma_{\mathrm{M}_Z}$.
We remark that the same inductive approach can be used to prove the following stronger statement (see [@Cattani] for an overview of the analogous facts in the context of convex polytopes and compact Kähler manifolds). We leave details to the interested reader.
\[TheoremMixed\] Let $\mathrm{M}$ be a loopless matroid on $E$, and let $\mathscr{P}$ be an order filter of $\mathscr{P}_\mathrm{M}$.
(1) The Bergman fan $\Sigma_{\mathrm{M},\mathscr{P}}$ satisfies the *mixed hard Lefschetz theorem*: For any multiset $$\mathscr{L}:=\big\{\ell_1,\ell_2, \ldots, \ell_{r-2q} \big\} \subseteq \mathscr{K}_{\mathrm{M},\mathscr{P}},$$ the linear map given by the multiplication $$L_\mathscr{L}^q: A^q(\mathrm{M},\mathscr{P})_{\mathbb{R}} \longrightarrow A^{r-q}(\mathrm{M},\mathscr{P})_{\mathbb{R}}, \qquad a \longmapsto \big( \ell_1 \ell_2 \cdots \ell_{r-2q} \big) \cdot a$$ is an isomorphism for all nonnegative integers $q \le \frac{r}{2}$.
(2) The Bergman fan $\Sigma_{\mathrm{M},\mathscr{P}}$ satisfies the *mixed Hodge-Riemann Relations*: For any multiset $$\mathscr{L}:=\big\{\ell_1,\ell_2, \ldots, \ell_{r-2q} \big\} \subseteq \mathscr{K}_{\mathrm{M},\mathscr{P}} \ \ \text{and any} \ \ \ell \in \mathscr{K}_{\mathrm{M},\mathscr{P}},$$ the symmetric bilinear form given by the multiplication $$Q^q_\mathscr{L}: A^q(\mathrm{M},\mathscr{P})_{\mathbb{R}} \times A^q(\mathrm{M},\mathscr{P})_{\mathbb{R}} \longrightarrow \mathbb{R}, \qquad (a_1,a_2) \longmapsto (-1)^q \text{deg}\ \big( a_1 \cdot L_\mathscr{L}^q(a_2)\big)$$ is positive definite on the kernel of $\ell \cdot L^q_\mathscr{L}$ for all nonnegative integers $q \le \frac{r}{2}$.
Log-concavity conjectures {#SectionLCConjectures}
=========================
{#section-23}
Let $\mathrm{M}$ be a loopless matroid of rank $r+1$ on the ground set $E=\{0,1,\ldots,n\}$. The *characteristic polynomial* of $\mathrm{M}$ is defined to be $$\chi_\mathrm{M}( \lambda)=\sum_{I \subseteq E} (-1)^{|I|}\ \lambda^{\text{crk}(I)},$$ where the sum is over all subsets $I \subseteq E$ and $\text{crk}(I)$ is the corank of $I$ in $\mathrm{M}$. Equivalently, $$\chi_\mathrm{M}( \lambda) = \sum_{F \subseteq E} \mu_\mathrm{M}(\varnothing,F)\ \lambda^{\text{crk}(F)},$$ where the sum is over all flats $F \subseteq E$ and $\mu_\mathrm{M}$ is the Möbius function of the lattice of flats of $\mathrm{M}$. Any one of the two descriptions clearly shows that
(1) the degree of the characteristic polynomial is $r+1$,
(2) the leading coefficient of the characteristic polynomial is $1$, and
(3) the characteristic polynomial satisfies $\chi_\mathrm{M}(1)=0$.
See [@Zaslavsky; @AignerEncyclopedia] for basic properties of the characteristic polynomial and its coefficients.
The *reduced characteristic polynomial* $\overline{\chi}_\mathrm{M}( \lambda)$ is $$\overline{\chi}_\mathrm{M}( \lambda):=\chi_\mathrm{M}( \lambda)/ ( \lambda-1).$$ We define a sequence of integers $\mu^0(\mathrm{M}),\mu^1(\mathrm{M}),\ldots,\mu^r(\mathrm{M})$ by the equality $$\overline{\chi}_\mathrm{M}( \lambda)=\sum_{k=0}^r (-1)^k\mu^k(\mathrm{M}) \hspace{0.5mm} \lambda^{r-k}.$$
The first number in the sequence is $1$, and the last number in the sequence is the absolute value of the Möbius number $\mu_\mathrm{M}(\varnothing,E)$. In general, $\mu^k(\mathrm{M})$ is the alternating sum of the absolute values of the coefficients of the characteristic polynomial $$\mu^k(\mathrm{M})=w_k(\mathrm{M})-w_{k-1}(\mathrm{M})+\cdots+(-1)^k w_0(\mathrm{M}).$$ We will show that the Hodge-Riemann relations for $A^*(\mathrm{M})_\mathbb{R}$ imply the log-concavity $$\mu^{k-1}(\mathrm{M})\mu^{k+1}(\mathrm{M}) \le \mu^k(\mathrm{M})^2 \quad \text{for} \quad 0<k<r.$$ Because the convolution of two log-concave sequences is log-concave, the above implies the log-concavity of the sequence $w_k(\mathrm{M})$.
Let $\mathscr{F}=\{F_1\subsetneq F_2\subsetneq\cdots\subsetneq F_k\}$ be a $k$-step flag of nonempty proper flats of $\mathrm{M}$.
(1) The flag $\mathscr{F}$ is said to be *initial* if $r(F_m)=m$ for all indices $m$.
(2) The flag $\mathscr{F}$ is said to be *descending* if $\text{min}(F_1)>\text{min}(F_2)>\cdots>\text{min}(F_k)>0$.
We write $\text{D}_k(\mathrm{M})$ for the set of initial descending $k$-step flags of nonempty proper flats of $\mathrm{M}$.
For inductive purposes it will be useful to consider the *truncation* of $\mathrm{M}$, denoted $\text{tr}(\mathrm{M})$. This is the matroid on $E$ whose rank function is defined by $$\text{rk}_{\text{tr}(\mathrm{M})}(I):=\text{min}(\text{rk}_\mathrm{M}(I),r).$$ The lattice of flats of $\text{tr}(\mathrm{M})$ is obtained from the lattice of flats of $\mathrm{M}$ by removing all the flats of rank $r$. It follows that, for any nonnegative integer $k<r$, there is a bijection $$\text{D}_k(\mathrm{M}) \simeq \text{D}_k(\text{tr}(\mathrm{M})),$$ and an equality between the coefficients of the reduced characteristic polynomials $$\mu^k(\mathrm{M})=\mu^k(\text{tr}(\mathrm{M})).$$ The second equality shows that all the integers $\mu^k(\mathrm{M})$ are positive, see [@Zaslavsky Theorem 7.1.8].
\[LemmaShelling\] For every positive integer $k \le r$, we have $$\mu^k(\mathrm{M})=|\text{D}_k(\mathrm{M})|.$$
The assertion for $k=r$ is the known fact that $\mu^r(\mathrm{M})$ is the number of facets of $\Delta_\mathrm{M}$ that are glued along their entire boundaries in its lexicographic shelling; see [@Bjorner Proposition 7.6.4]. The general case is obtained from the same equality applied to repeated truncations of $\mathrm{M}$. See [@Huh-Katz Proposition 2.4] for an alternative approach using Weisner’s theorem.
We now show that $\mu^k(\mathrm{M})$ is the degree of the product $\alpha_\mathrm{M}^{r-k} \hspace{0.5mm} \beta_\mathrm{M}^k$. See Definition \[DefinitionAlphaBeta\] for the elements $\alpha_\mathrm{M}, \beta_\mathrm{M} \in A^1(\mathrm{M})$, and Definition \[MatroidDegreeMap\] for the degree map of $\mathrm{M}$.
\[LemmaBetaPower\] For every positive integer $k \le r$, we have $$\beta^k_\mathrm{M}=\sum_{\mathscr{F}} x_\mathscr{F} \in A^*(\mathrm{M}),$$ where the sum is over all descending $k$-step flags of nonempty proper flats of $\mathrm{M}$.
We prove by induction on the positive integer $k$. When $k=1$, the assertion is precisely that $\beta_{\mathrm{M},0}$ represents $\beta_{\mathrm{M}}$ in the Chow ring of $\mathrm{M}$: $$\beta_\mathrm{M} =\beta_{\mathrm{M},0}= \sum_{0 \notin F} x_F \in A^*(\mathrm{M}).$$ In the general case, we use the induction hypothesis for $k$ to write $$\beta^{k+1}_\mathrm{M}= \sum_\mathscr{F} \beta_\mathrm{M}\hspace{0.5mm} x_\mathscr{F},$$ where the sum is over all descending $k$-step flags of nonempty proper flats of $\mathrm{M}$. For each of the summands $\beta_\mathrm{M}\hspace{0.5mm} x_\mathscr{F}$, we write $$\mathscr{F}=\big\{F_1\subsetneq F_2\subsetneq\cdots\subsetneq F_k\big\}, \ \ \text{and set} \ \ i_\mathscr{F}:= \text{min}(F_1).$$ By considering the representative of $\beta_\mathrm{M}$ corresponding to the element $i_\mathscr{F}$, we see that $$\beta_\mathrm{M}\hspace{0.5mm} x_\mathscr{F}=\Big(\sum_{i_\mathscr{F} \notin F} x_F \Big)x_\mathscr{F}=\sum_{\mathscr{G}} x_{\mathscr{G}},$$ where the second sum is over all descending flags of nonempty proper flats of $\mathrm{M}$ of the form $$\mathscr{G}=\big\{F \subsetneq F_1 \subsetneq \cdots \subsetneq F_k\big\}.$$ This complete the induction.
Combining Lemma \[LemmaShelling\], Lemma \[LemmaBetaPower\], and Proposition \[PropositionFundamentalClass\], we see that the coefficients of the reduced characteristic polynomial of $\mathrm{M}$ are given by the degrees of the products $\alpha_\mathrm{M}^{r-k}\hspace{0.5mm} \beta_\mathrm{M}^k$:
\[PropositionIntersectionDegree\] For every nonnegative integer $k \le r$, we have $$\mu^k(\mathrm{M})=\deg(\alpha_\mathrm{M}^{r-k} \hspace{0.5mm}\beta_\mathrm{M}^k).$$
{#section-24}
Now we explain why the Hodge-Riemann relations imply the log-concavity of the reduced characteristic polynomial. We first state a lemma involving inequalities among degrees of products:
\[lem:ampleinequality\] Let $\ell_1$ and $\ell_2$ be elements of $A^1(\mathrm{M})_\mathbb{R}$. If $\ell_2$ is nef, then $$\text{deg}(\ell_1 \hspace{0.5mm} \ell_1 \hspace{0.5mm} \ell_2^{r-2})\ \text{deg}(\ell_2\hspace{0.5mm} \ell_2 \hspace{0.5mm} \ell_2^{r-2})\leq \text{deg}(\ell_1\hspace{0.5mm} \ell_2 \hspace{0.5mm} \ell_2^{r-2})^2.$$
We first consider the case when $\ell_2$ is ample. Let $Q^1_{\ell_2}$ be the Hodge-Riemann form $$Q^1_{\ell_2}:A^1(\mathrm{M})_\mathbb{R} \times A^1(\mathrm{M})_\mathbb{R} \longrightarrow \mathbb{R}, \qquad (a_1,a_2) \longmapsto -\text{deg}(a_1\hspace{0.5mm}\ell_2^{r-2} \hspace{0.5mm} a_2 ).$$ Since the Chow ring $A^*(\mathrm{M})$ satisfies $\text{HL}(\ell_2)$, we have the Lefschetz decomposition $$A^1(\mathrm{M})_\mathbb{R}=\langle \ell_2 \rangle \oplus P^1_{\ell_2}(\mathrm{M}).$$ Note that the decomposition is orthogonal with respect to the Hodge-Riemann form $Q^1_{\ell_2}$. The property $\text{HR}(\ell_2)$ shows that $Q^1_{\ell_2}$ is negative definite on $\langle \ell_2 \rangle$ and positive definite on its orthogonal complement $P^1_{\ell_2}(\mathrm{M})$.
We consider $Q^1_{\ell_2}$ restricted to the subspace $\langle \ell_1,\ell_2 \rangle \subseteq A^1(\mathrm{M})_\mathbb{R}$. If $\ell_1$ is not a multiple of $\ell_2$, then the restriction of $Q^1_{\ell_2}$ is neither positive definite nor negative definite, and hence $$\text{deg}(\ell_1 \hspace{0.5mm} \ell_1 \hspace{0.5mm} \ell_2^{r-2})\ \text{deg}(\ell_2\hspace{0.5mm} \ell_2 \hspace{0.5mm} \ell_2^{r-2})< \text{deg}(\ell_1\hspace{0.5mm} \ell_2 \hspace{0.5mm} \ell_2^{r-2})^2.$$
Next consider the case when $\ell_2$ is nef. By Proposition \[Subfan\], there is an element $\ell$ in the ample cone of $\mathrm{M}$. Since $\ell_2$ is nef, we have $$\ell_2(t):=\ell_2+t \hspace{0.5mm} \ell \in \mathscr{K}_\mathrm{M} \ \ \text{for all positive real numbers $t$.}$$ Therefore for all positive real numbers $t$ we have $$\text{deg}(\ell_1 \hspace{0.5mm} \ell_1 \hspace{0.5mm} \ell_2(t)^{r-2})\ \text{deg}(\ell_2(t)\hspace{0.5mm} \ell_2(t) \hspace{0.5mm} \ell_2(t)^{r-2})\leq \text{deg}(\ell_1\hspace{0.5mm} \ell_2(t) \hspace{0.5mm} \ell_2(t)^{r-2})^2.$$ By taking the limit $t \to 0$, we obtain the desired inequality.
Let $\mathrm{M}$ be a loopless matroid.
(1) The element $\alpha_\mathrm{M}$ is the class of a convex piecewise linear function on $\Sigma_\mathrm{M}$.
(2) The element $\beta_\mathrm{M}$ is the class of a convex piecewise linear function on $\Sigma_\mathrm{M}$.
In other words, $\alpha_\mathrm{M}$ and $\beta_\mathrm{M}$ are nef.
For the first assertion, it is enough to show that $\alpha_\mathrm{M}$ is the class of a nonnegative piecewise linear function that is zero on a given cone $\sigma_{\varnothing < \mathscr{F}}$ in $\Sigma_\mathrm{M}$. For this we choose an element $i$ not in any of the flats in $\mathscr{F}$. The representative $\alpha_{\mathrm{M},i}$ of $\alpha_\mathrm{M}$ has the desired property.
Similarly, for the second assertion, it is enough to show that $\beta_\mathrm{M}$ is the class of a nonnegative piecewise linear function that is zero on a given cone $\sigma_{\varnothing < \mathscr{F}}$ in $\Sigma_\mathrm{M}$. For this we choose an element $i$ in the flat $\text{min}\ \mathscr{F}$. The representative $\beta_{\mathrm{M},i}$ of $\beta_\mathrm{M}$ has the desired property.
\[PropositionLogConcave\] For every positive integer $k < r$, we have $$\mu^{k-1}(\mathrm{M})\mu^{k+1}(\mathrm{M}) \le \mu^{k}(\mathrm{M})^2.$$
We prove by induction on the rank of $\mathrm{M}$. When $k$ is less than $r-1$, the induction hypothesis applies to the truncation of $\mathrm{M}$. When $k$ is $r-1$, Proposition \[PropositionIntersectionDegree\] shows that the assertion is equivalent to the inequality $$\text{deg}(\alpha_\mathrm{M}^2\hspace{0.5mm} \beta_\mathrm{M}^{r-2})\text{deg}(\beta_\mathrm{M}^2\hspace{0.5mm} \beta_\mathrm{M}^{r-2})\leq\text{deg}(\alpha^1_\mathrm{M} \hspace{0.5mm}\beta^{r-1}_\mathrm{M})^2.$$ This follows from Lemma \[lem:ampleinequality\] applied to the nef classes $\alpha_\mathrm{M}$ and $\beta_\mathrm{M}$.
As an implication of Proposition \[PropositionLogConcave\], we conclude with the proof of the announced log-concavity results.
\[ReducedLogConcave\] Let $\mathrm{M}$ be a matroid, and let $G$ be a graph.
(1) The coefficients of the reduced characteristic polynomial of $\mathrm{M}$ form a log-concave sequence.
(2) The coefficients of the characteristic polynomial of $\mathrm{M}$ form a log-concave sequence.
(3) The number of independent subsets of size $i$ of $\mathrm{M}$ form a log-concave sequence in $i$.
(4) The coefficients of the chromatic polynomial of $G$ form a log-concave sequence.
The second item proves the aforementioned conjecture of Heron [@Heron], Rota [@Rota], and Welsh [@WelshBook]. The third item proves the conjecture of Mason [@Mason] and Welsh [@Welsh]. The last item proves the conjecture of Read [@Read] and Hoggar [@Hoggar].
It follows from Proposition \[PropositionLogConcave\] that the coefficients of the reduced characteristic polynomial of $\mathrm{M}$ form a log-concave sequence. Since the convolution of two log-concave sequences is a log-concave sequence, the coefficients of the characteristic polynomial of $\mathrm{M}$ also form a log-concave sequence.
To justify the third assertion, we consider the free dual extension of $\mathrm{M}$. It is defined by taking the dual of $\mathrm{M}$, placing a new element $p$ in general position (taking the free extension), and again taking the dual. In symbols, $$\mathrm{M} \times p := (\mathrm{M}^* + p)^*.$$ The free dual extension $\mathrm{M} \times p$ has the following property: The number of independent subsets of size $k$ of $\mathrm{M}$ is the absolute value of the coefficient of $\lambda^{r-k}$ of the reduced characteristic polynomial of $\mathrm{M}$. We refer to [@Lenz] and also to [@Brylawski; @Constructions] for these facts. It follows that the number of independent subsets of size $k$ of $\mathrm{M}$ form a log-concave sequence in $k$.
For the last assertion, we recall that the chromatic polynomial of a graph is given by the characteristic polynomial of the associated graphic matroid [@WelshBook]. More precisely, we have $$\chi_G(\lambda) = \lambda^{n_G} \cdot \chi_{\mathrm{M}_G}(\lambda),$$ where $n_G$ is the number of connected components of $G$. It follows that the coefficients of the chromatic polynomial of $G$ form a log-concave sequence.
[BDP90]{}
Karim Adiprasito and Raman Sanyal, *Log-concavity of Whitney numbers via measure concentration*, preprint.
Karim Adiprasito and Anders Björner, *Filtered geometric lattices and Lefschetz Section Theorems over the tropical semiring*, arXiv:1401.7301.
Karim Adiprasito, June Huh, Eric Katz, *Hodge theory for simplicial fans*, in preparation.
Martin Aigner, *Whitney numbers*, Combinatorial Geometries, 139–160, Encyclopedia of Mathematics and its Applications [**29**]{}, Cambridge University Press, Cambridge, 1987.
Martin Aigner, *A course in enumeration*, Graduate Texts in Mathematics [**23**]{}, Springer-Verlag, Berlin, 2007.
Federico Ardila and Caroline Klivans, *The Bergman complex of a matroid and phylogenetic trees*, Journal of Combinatorial Theory Series B [**96**]{} (2006), no. 1, 38–49.
Farhad Babaee and June Huh, *A tropical approach to the strongly positive Hodge conjecture*, arXiv:1502.00299.
Victor Batyrev and Mark Blume, *The functor of toric varieties associated with Weyl chambers and Losev-Manin moduli spaces*, Tohoku Mathematical Journal [**63**]{} (2011), no. 4, 581–604.
Emili Bifet, Corrado De Concini, and Claudio Procesi, *Cohomology of regular embeddings*, Advances in Mathematics [**82**]{} (1990), no. 1, 1–34.
Louis Billera, *The algebra of continuous piecewise polynomials*, Advances in Mathematics [**76**]{} (1989), no. 2, 170–183.
George Birkhoff, *A determinant formula for the number of ways of coloring a map*, Annals of Mathematics [**14**]{} (1912), no. 1, 42–46.
Anders Björner, *The homology and shellability of matroids and geometric lattices*, Matroid Applications, 226-283, Encyclopedia of Mathematics and its Applications [**40**]{}, Cambridge University Press, Cambridge, 1992.
Michel Brion, *Piecewise polynomial functions, convex polytopes and enumerative geometry*, Parameter spaces (Warsaw, 1994), 25-44, Banach Center Publications [**36**]{}, Polish Academy of Sciences, Warsaw, 1996.
Thomas Brylawski, *The broken-circuit complex*, Transactions of the American Mathematical Society [**234**]{} (1977), no. 2, 417–433.
Thomas Brylawski, *Constructions*, Theory of Matroids, 127–223, Encyclopedia of Mathematics and its Applications [**26**]{}, Cambridge University Press, Cambridge, 1986.
Eduardo Cattani, *Mixed Lefschetz theorems and Hodge-Riemann bilinear relations*, International Mathematics Research Notices, Vol. 2008, no. 10, Article ID rnn025, 20 pp.
Mark de Cataldo and Luca Migliorini, *The hard Lefschetz theorem and the topology of semismall maps*, Annales Scientifiques de l’école Normale Supérieure [**35**]{} (2002), no. 5, 759-772.
Vladimir Danilov, *The geometry of toric varieties*, Russian Mathematical Surveys [**33**]{} (1978), 97–154.
Corrado De Concini and Claudio Procesi, *Wonderful models of subspace arrangements*, Selecta Mathematica. New Series [**1**]{} (1995), no. 3, 459–494.
Dimitry Kozlov, *Combinatorial algebraic topology*, Algorithms and Computation in Mathematics [**21**]{}, Springer-Verlag Berlin, 2008.
Eva Maria Feichtner and Sergey Yuzvinsky, *Chow rings of toric varieties defined by atomic lattices*, Inventiones Mathematicae [**155**]{} (2004), no. 3, 515–536.
William Fulton, *Introduction to Toric Varieties*, Annals of Mathematics Studies, [**131**]{}, Princeton University Press, Princeton, NJ, 1993.
William Fulton and Bernd Sturmfels, *Intersection theory on toric varieties*, Topology [**36**]{} (1997), no. 2, 335–353.
Israel Gel’fand and Robert MacPherson, *A combinatorial formula for the Pontrjagin classes*, Bulletin of the American Mathematical Society (New Series) [**26**]{} (1992), no. 2, 304–309.
Angela Gibney and Diane Maclagan, *Lower and upper bounds on nef cones*, International Mathematics Research Notices (2012), no. 14, 3224–3255.
Andrew Heron, *Matroid polynomials*, Combinatorics (Proc. Conf. Combinatorial Math., Math. Inst., Oxford, 1972), Inst. of Math. and its Appl., Southend-on-Sea, 1972, pp. 164–202.
Stuart Hoggar, *Chromatic polynomials and logarithmic concavity*, Journal of Combinatorial Theory Series B [**16**]{} (1974), 248–254.
June Huh, *Milnor numbers of projective hypersurfaces and the chromatic polynomial of graphs*, Journal of the American Mathematical Society [**25**]{} (2012), 907–927.
June Huh and Eric Katz, *Log-concavity of characteristic polynomials and the Bergman fan of matroids*, Mathematische Annalen [**354**]{} (2012), no. 3, 1103–1116.
Ehud Hrushovski, *Unimodular minimal structures*, Journal of the London Mathematical Society, II. [**46**]{} (1992), no.3, 385-396.
Ilya Itenberg, Ludmil Katzarkov, Grigory Mikhalkin and Ilya Zharkov, *Tropical homology*, in preparation.
Eric Katz and Sam Payne, *Realization spaces for tropical fans*, Combinatorial Aspects of Commutative Algebra and Algebraic Geometry, 73–88, Abel Symposium [**6**]{}, Springer, Berlin, 2011.
Joseph Kung, *The geometric approach to matroid theory*, Gian-Carlo Rota on Combinatorics, 604–622, Contemporary Mathematicians, Birkhäuser Boston, Boston, MA, 1995.
Robert Lazarsfeld, *Positivity in Algebraic Geometry I*, Ergebnisse der Mathematik und ihrer Grenzgebiete [**48**]{}, Springer-Verlag, Berlin, 2004.
Carl Lee, *P.L.-spheres, convex polytopes, and stress*, Discrete and Computational Geometry [**15**]{} (1996), no. 4, 389–421.
Matthias Lenz, *The f-vector of a representable-matroid complex is log-concave*, Advances in Applied Mathematics [**51**]{} (2013), no. 5, 543–545.
John Mason, *Matroids: unimodal conjectures and Motzkin’s theorem*, Combinatorics (Proc. Conf. Combinatorial Math., Math. Inst., Oxford, 1972), pp. 207–220. Inst. Math. Appl., Southend-on-Sea, 1972.
Peter McMullen, *On simple polytopes*, Inventiones mathematicae [**113**]{} (1993), no. 2, 419–444.
Dagmar Meyer and Larry Smith, *Poincaré duality algebras, Macaulay’s dual systems, and Steenrod operations*, Cambridge Tracts in Mathematics [**167**]{}, Cambridge University Press, Cambridge, 2005.
Ezra Miller and Bernd Sturmfels, *Combinatorial commutative algebra*, Graduate Texts in Mathematics [**227**]{}, Springer-Verlag, New York, 2005.
John von Neumann, *Continuous geometry*, Princeton Landmarks in Mathematics Princeton University Press, Princeton, NJ (1960).
James Oxley, *Matroid Theory*, Oxford Science Publications, Oxford University Press, New York, 2011.
Anand Pillay, *Geometric stability theory*, Oxford Logic Guides [**32**]{}, The Clarendon Press, Oxford University Press, New York, 1996.
Ronald Read, *An introduction to chromatic polynomials*, Journal of Combinatorial Theory [**4**]{} (1968), 52–71.
Gian-Carlo Rota, *On the foundations of combinatorial theory. I. Theory of Möbius functions.* Zeitschrift für Wahrscheinlichkeitstheorie und Verwandte Gebiete [**2**]{} (1964), 340–368.
Gian-Carlo Rota, *Combinatorial theory, old and new*, Actes du Congrès International des Mathématicians (Nice, 1970), Tome 3, Gauthier-Villars, Paris, 1971, pp. 229–233.
Richard Stanley, *An introduction to hyperplane arrangements*, Geometric combinatorics, 389–496, IAS/Park City Mathematics Series, [**13**]{}, American Mathematical Society, Providence, RI, 2007.
Dominic Welsh, *Combinatorial problems in matroid theory*, Combinatorial Mathematics and its Applications (Proc. Conf., Oxford, 1969) pp. 291–306, Academic Press, London, 1971.
Dominic Welsh, *Matroid Theory*, London Mathematical Society Monographs, [**8**]{}, Academic Press, London-New York, 1976.
Neil White, *Combinatorial geometries*, Encyclopedia of Mathematics and its Applications [**29**]{}, Cambridge University Press, Cambridge, 1987.
Hassler Whitney, *A logical expansion in mathematics*, Bulletin of the American Mathematical Society [**38**]{}, no. 8 (1932), 572–579.
Hassler Whitney, *On the abstract properties of linear dependence.* American Journal of Mathematics [**57**]{} (1935), no. 3, 509–533.
Thomas Zaslavsky, *The Möbius function and the characteristic polynomial*, Combinatorial geometries, 114–138, Encyclopedia of Mathematics and its Applications [**29**]{}, Cambridge University Press, Cambridge, 1987.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We produce an edge-coloring of the complete 3-uniform hypergraph on $n$ vertices with $e^{O(\sqrt{\log \log n})}$ colors such that the edges spanned by every set of five vertices receive at least three distinct colors. This answers the first open case of a question of Conlon-Fox-Lee-Sudakov [@CFLS] who asked whether such a coloring exists with $(\log n)^{o(1)}$ colors.'
author:
- 'Dhruv Mubayi[^1]'
title: Coloring triple systems with local conditions
---
Introduction
============
A $k$-uniform hypergraph $H$ ($k$-graph for short) with vertex set $V(H)$ is a collection of $k$-element subsets of $V(H)$. Write $K_n^k$ for the complete $k$-graph with vertex set of size $n$. A $(p,q)$-coloring of $K_n^k$ is an edge-coloring of $K_n^k$ that gives every copy of $K_p^k$ at least $q$ colors. Let $f_k(n,p,q)$ be the minimum number of colors in a $(p,q)$-coloring of $K_n^k$. This paper deals only with $k=3$.
Conlon-Fox-Lee-Sudakov [@CFLS] asked whether $f_3(n,p,p-2)=(\log n)^{o(1)}$ for $p \ge 3$ (the case $p=4$ is easy). In this note we answer the first open case with a substantially smaller bound.
$$f_3(n,5,3)= e^{O(\sqrt{\log\log n})}.$$
The problem of determining $f_k(n,p,q)$ for fixed $k,p,q$ has a long history, beginning with its introduction by Erdős and Shelah [@E1; @E2], and subsequent investigation (for graphs) by Erdős and Gyárfás [@EG]. Studying $f_k(n,p,q)$ when $q=2$ is equivalent to studying classical Ramsey numbers, and most of the effort on these problems has therefore been for $q>2$. The simplest nontrivial case in this regime is $f_2(n,4,3)$, which was shown to be $n^{o(1)}$ in [@M] and later $\Omega(\log n)$ (see [@FS; @KM]). The same upper bound was shown for $f(n,5,4)$ in [@EM]. Conlon-Fox-Lee-Sudakov [@CFLS2] recently extended this construction considerably by proving that $f_2(n,p,p-1)=n^{o(1)}$ for all fixed $p \ge 4$. Their result is sharp in the sense that $f_2(n,p,p)=\Omega(n^{1/(p-2)})$.
The first nontrivial hypergraph case is $f_3(n,4,3)$ and has tight connections to Shelah’s breakthrough proof [@S] of primitive recursive bounds for the Hales-Jewett numbers. Answering a question of Graham-Rothschild-Spencer [@GRS], Conlon et. al. [@CFLS] recently proved that $f_3(n,4,3)= n^{o(1)}$. They also posed a variety of basic questions about $f_3(n,p,q)$, including the one we address in this note.
Our construction uses an extension of the coloring in [@M] together with the stepping up technique of Erdős and Hajnal. It is quite possible that, similar to the situation for graphs, other hypergraph cases will eventually be addressed by the ideas introduced here.
The Construction
================
We begin by defining an edge-coloring $\sigma$ of the complete graph $K_n$ whose vertices are ordered.
[**Construction of $\sigma$:**]{} Given integers $t<m$ and $n={m \choose t}$, let $V(K_n)$ be the set of 0/1 vectors of length $m$ with exactly $t$ 1’s. Write $v=(v(1), \ldots, v(m))$ for a vertex. The vertices are naturally ordered by the integer they represent in binary, so $v<w$ iff $v(i)=0$ and $w(i)=1$ where $i$ is the first position (minimum integer) in which $v$ and $w$ differ. By considering vertices as characteristic vectors of sets, we may assume that $V(K_n)= {[m] \choose t}$ whenever convenient. For each $B \in {[m]\choose t}$, let $f_B:2^B\rightarrow [2^t]$ be a bijection. Given vectors $v<w$ that are characteristic vectors of sets $S<T$, let $c_1(vw)=\min \{i: v(i)=0, w(i)=1\}$, $c_2(vw)=\min \{j: j>i, v(i)=1, w(i)=0\}$, $c_3(vw)=f_S(S \cap T)$ and $c_4(vw)=f_T(S \cap T)$. Finally, define $$\sigma(vw)=(c_1(vw), c_2(vw), c_3(vw), c_4(vw)).$$ If $n$ is not of the form ${m \choose t}$, then let $n'\ge n$ be the smallest integer of this form, color ${[n']\choose 2}$ as described above, and restrict the coloring to ${[n]\choose 2}$.
It is known [@M; @M2] that $\sigma$ is both a $(3,2)$ and $(4,3)$-coloring of $K_n$ (we only need the first and fourth coordinates of color vectors for this) and, for suitable choice of $m$ and $t$ it uses $e^{O(\sqrt{\log n})}$ colors for all $n$. We need the following additional properties.
The coloring $\sigma$ satisfies the following properties.
1\) If $v<w<x$, then $\sigma(vw)\ne \sigma(wx)$.
2\) If $v<w<\min\{x,y\}$, and $\sigma(vw)=\sigma(vx)$, then $\sigma(vy)\ne \sigma(wx)$.
3\) If $v<w<x<y$ with $\sigma(vw)=\sigma(xy)$, then $\sigma(vx)\ne \sigma(vy)$.
It suffices to consider the first coordinate $c_1$ of $\sigma$ to prove the first two properties. For 1), observe that $i=c_1(vw)$ implies that $w(i)=1$, while $i=c_1(wx)$ implies that $w(i)=0$. For 2), let $i=c_1(vw)=c_1(vx)$ and suppose for contradiction that $i'=c_1(vy)=c_1(wx)$ so that $v(j)=y(j)$ for $j<i'$. Assume first that $i<i'$ . Then $y(i)=v(i)=0$, while $w(i)=1$. This implies that $w>y$, a contradiction. Now assume that $i>i'$ ($i=i'$ is impossible since $w(i)=1$ while $w(i')=0$). Then $0=v(i')=x(i')=1$ due to $c_1(vy)=i', c_1(vx)=i>i'$ and $c_1(wx)=i'$.
We now prove 3) so assume we are given $v<w<x<y$ with $c_1(vw)=c_1(xy)=i<j=c_2(vw)=c_2(xy)$. Then $v(j)=x(j)=1$ and $y(j)=0$. Suppose that $v,w,x,y$ are characteristic vectors of $V, W, X, Y$ respectively. Then $c_3(vx)=c_3(VX)=f_V(V \cap X)$ while $c_3(vy)=c_3(VY)=f_V(V \cap Y)$. If $c_3(vx)=c_3(vy)$, then $f_V(V \cap X)=f_V(V \cap Y)$ and since $f_V$ is a bijection, $V \cap X=V \cap Y$. But this is impossible as $j \in (V \cap X)\setminus Y$.
We are now ready to describe the edge-coloring $\chi$ of $K_n^3$ that we will use.
[**Construction of $\chi$:**]{} Given a copy of $K_n$ on $[n]$ and the edge-coloring $\sigma$, we produce an edge-coloring $\chi$ of the 3-graph $H$ on $\{0,1\}^n$ as follows. Order the vertices of $H$ according to the integer that they represent in binary. Given vertices $x< y$ in $V(H)$, let $\gamma_{xy}$ be the first coordinate where $x$ and $y$ differ. Given vertices $x<y<z$, let $\delta_{xyz}$ equal 1 if $\gamma_{xy}<\gamma_{yz}$ and $-1$ otherwise. For an edge $uvw$ with $u<v<w$, let $$\chi(uvw)=(\sigma(\gamma_{uv} \gamma_{vw}), \delta_{uvw}). \quad \qed$$
Since $\sigma$ is an edge-coloring of $K_n$ with $e^{O(\sqrt{\log n})}$ colors, $\chi$ is an edge-coloring of $K_N^3$ ($N=2^n$) with $e^{O(\sqrt{\log\log N})}$ colors as promised. Moreover, extending this construction to all $N$ is trivial by considering the smallest $N'\ge N$ which is a power of 2, coloring ${[N']\choose 2}$ and restricting to ${[N]\choose 2}$. We are left with showing that $\chi$ is a $(5,3)$-coloring of $K_N^3$.
[**Proof that $\chi$ is a $(5,3)$-coloring:**]{} Suppose, for contradiction, that $X=\{x_1, \ldots, x_5\}$ where $x_1<x_2<x_3<x_4<x_5$ are five vertices of $H$ forming a 2-colored $K_5^3$. Let $\gamma_{i}=\gamma_{x_ix_{i+1}}$. Let $\gamma=\min \gamma_j$ and assume this minimum is achieved by $\gamma_p$. Note that this minimum is uniquely achieved, and $\gamma_i\ne \gamma_{i+1}$ for all $i$.
[**Case 1:**]{} $p\in \{1,4\}$. The arguments for both cases are almost identical so we only consider the case $p=1$. By assumption we have $\gamma_1<\gamma_2$. First assume that $\gamma_3>\gamma_2$. If $\gamma_4>\gamma_3$, then the $K_4$ on $\{\gamma_i: i \in [4]\}$ has three colors since $\sigma$ is a $(4,3)$-coloring and this gives at least three colors to the edges in $X$. If $\gamma_4<\gamma_3$ then the $K_3$ on $\{\gamma_i: i \in [3]\}$ has two colors since $\sigma$ is a $(3,2)$-coloring and this gives two colors to the edges of $H$ within $\{x_i: i \in [4]\}$ with positive $\delta$-coordinate. On the other hand $\delta_{x_3x_4x_5}=-1$, so we again have three colors on $X$. We now suppose that $\gamma_3<\gamma_2$. If $\gamma_4<\gamma_3$, then the $K_3$ on $\{\gamma_2, \gamma_3, \gamma_4\}$ has two colors since $\sigma$ is a $(3,2)$-coloring and this gives two colors to the edges of $H$ within $\{x_2, x_3, x_4, x_5\}$ with negative $\delta$-coordinate. On the other hand $\delta_{x_1x_2x_3}=1$, so we again have three colors on $X$. Finally, we may assume that $\gamma_1<\gamma_3<\min\{\gamma_2, \gamma_4\}$. Now $\sigma(\gamma_1\gamma_3)\ne \sigma(\gamma_3, \gamma_4)$ due to property 1) of $\sigma$, hence $\chi(x_1x_3x_4)\ne \chi(x_3x_4x_5)$ and both have positive $\delta$-coordinates. But $\delta_{x_2x_3x_4}=-1$, so $\chi(x_2x_3x_4)$ is the third color on $X$.
[**Case 2:**]{} $p\in\{2,3\}$. The arguments for both cases are almost identical so we only consider the case $p=2$. We have $\gamma_3> \gamma_2$. If in addition $\gamma_4>\gamma_3$, then we get two colors among $\{x_2, x_3, x_4, x_5\}$ with positive $\delta$-coordinate while $\delta_{x_1x_2x_3}=-1$. So we may assume that $\gamma_2<\gamma_4<\gamma_3$. Now $\chi(x_2x_3x_4)$ and $\chi(x_2x_4x_5)$ both have positive $\delta$ coordinates while $\delta_{x_3x_4x_5}=-1$. Hence we have three colors unless $\sigma(\gamma_2\gamma_3)=\sigma(\gamma_2\gamma_4)$ which we may assume. Certainly $\delta_{x_1x_2x_3}=-1$, so we are done unless $\sigma(\gamma_2\gamma_1)=\sigma(\gamma_4\gamma_3)$ which we also assume. If $\gamma_1=\gamma_4$, then $\sigma(\gamma_2\gamma_4)=\sigma(\gamma_4\gamma_3)$ and hence $\{\gamma_2, \gamma_4, \gamma_3\}$ is a monochromatic triangle, contradiction. If $\gamma_1> \gamma_4$, then $\gamma_2<\gamma_4< \min\{\gamma_1,\gamma_3\}$ with $\sigma(\gamma_2\gamma_4)=\sigma(\gamma_2\gamma_3)$ and $\sigma(\gamma_2\gamma_1)=\sigma(\gamma_4\gamma_3)$. This contradicts property 2). If $\gamma_1< \gamma_4$, then $\gamma_2<\gamma_1<\gamma_4<\gamma_3$ with $\sigma(\gamma_2\gamma_1)=\sigma(\gamma_4\gamma_3)$ and $\sigma(\gamma_2\gamma_4)=\sigma(\gamma_2\gamma_3)$. This contradicts property 3) and completes the proof.
**Acknowledgment.** I am grateful to David Conlon and Choongbum Lee for carefully reading an earlier draft of this note and giving comments that helped improve the presentation.
[99]{}
D. Conlon, J. Fox, C. Lee, B. Sudakov, On the grid Ramsey problem and related questions, Int. Math. Res. Not., to appear.
D. Conlon, J. Fox, C. Lee and B. Sudakov, The Erdős-Gy' arfás problem on generalized Ramsey numbers, Proc. London Math. Soc., to appear.
P. Erdős, Problems and results on finite and infinite graphs, in Recent advances in graph theory (Proc. Second Czechoslovak Sympos., Prague, 1974), 183–192, Academia, Prague, 1975.
P. Erdős, Solved and unsolved problems in combinatorics and combinatorial number theory, in Proceedings of the twelfth southeastern conference on combinatorics, graph theory and comput-ing, Vol. I (Baton Rouge, La., 1981), Congr. Numer. 32 (1981), 49–62.
P. Erdős and A. Gyárfás, A variant of the classical Ramsey problem, Combinatorica 17 (1997), 459–467.
D. Eichhorn and D. Mubayi, Edge-coloring cliques with many colors on subcliques, Combinatorica 20 (2000), 441–444.
J. Fox and B. Sudakov, Ramsey-type problem for an almost monochromatic $K_4$, SIAM J. Discrete Math. 23 (2008), 155–162.
R. L. Graham, B. L. Rothschild and J. H. Spencer, Ramsey theory, second ed., Wiley Interscience Series in Discrete Mathematics and Optimization, John Wiley & Sons Inc., New York, 1990.
A. Kostochka and D. Mubayi, When is an almost monochromatic $K_4$ guaranteed?, Combin. Probab. Comput. 17 (2008), 823–830.
D. Mubayi, Edge-coloring cliques with three colors on all 4-cliques, Combinatorica 18 (1998), 293–296.
D. Mubayi, An explicit construction for a Ramsey problem, Combinatorica, 24 (2004), no. 2, 313–324
S. Shelah, Primitive recursive bounds for van der Waerden numbers, J. Amer. Math. Soc. 1 (1989), 683–697.
[^1]: Department of Mathematics, Statistics, and Computer Science, University of Illinois, Chicago, IL, 60607 USA. Research partially supported by NSF grant DMS-1300138. Email: [[email protected]]{}
|
{
"pile_set_name": "ArXiv"
}
|
The difficulties and interest in treating strongly correlated electron systems, and the consequences of correlation effects on magnetic behavior in the transitional 4[*f*]{} or 5[*f*]{} localization regime, provide one of the central problems of condensed matter physics.$^{1-3}$ The transitional regime behavior is neither atomiclike nor itinerant. This gives rise to an extremely interesting range of phenomena, but also causes very great difficulties in treating the theory of these phenomena adequately, especially in a way providing the ability to predict the behavior of specific materials.$^{1-3}$ An adequate treatment requires treating the interelectronic coulomb interaction, i.e. the correlation effects, as constrained by exchange symmetry.$^{4-6}$ In this letter, we demonstrate an approach for treating these difficulties in predicting the interesting and complex behavior of an important series of cerium compounds.
The isostructural (rock-salt structure) series of the cerium monopnictides CeX (X = P, As, Sb, Bi) and monochalcogenides (X=S, Se, Te) have become prototype model systems for study, because of their unusual magnetic properties.$^{7-13}$ This series of strongly correlated electron systems offers the opportunity to vary systematically, through chemical pressure, the lattice constant and the cerium-cerium separation on going down the pnictogen or chalcogen column, and hence tailor the degree of 4[*f*]{} localization from the strongly correlated limit in the heavier systems to the weakly correlated limit in the lighter systems.$^{7-13}$ The calculated single-impurity Kondo temperature, T$_K$, presented below, is much smaller than the magnetic ordering temperature in these systems, and hence this series lies in the [*magnetic*]{} regime of the Kondo phase diagram.$^{14}$ Nevertheless, in this work we demonstrate that the sensitivity of the hybridization, coulomb exchange, and crystal-field interactions with the chemical environment gives rise to a variety of unusual and interesting magnetic properties across the series, in agreement with experiment, [*including the occurence of a non-Kondo magnetic moment collapse*]{}.
This class of cerium systems exhibits large magnetic anisotropy which changes from the $<001>$ direction in the pnictides to the $<111>$ direction in the chalcogenides. The low-temperature ordered magnetic moment increases with increasing lattice constant for the pnictides from 0.80$\mu_B$ in CeP to 2.1$\mu_B$ in CeSb and CeBi,$^{7-8}$ while it decreases with increasing lattice constant for the chalcogenides from 0.57$\mu_B$ in CeS to 0.3$\mu_B$ in CeTe.$^{7-9}$ The [*magnetic moment collapse*]{} from CeSb to CeTe, with both systems having about the same lattice constant, is indicative of the sensitivity of the magnetic interactions to chemical environment. The experimentally observed low-temperature structure in CeBi and CeSb is the $<001>$ antiferromagnetic type IA ($\uparrow \uparrow \downarrow \downarrow)$, whereas in CeAs and CeP the structure is the $<001>$ antiferromagnetic type I ($\uparrow \downarrow$).$^{7,15}$ The ordering temperature increases from 8K in CeP to 26K in CeBi for the pnictides, whereas it decreases from 8.4K in CeS to an unusually low 2.2K in CeTe.$^{7-11}$ Another unusual feature of this series of cerium compounds is the large [*suppression*]{} of the crystal field (CF) splitting of the Ce$^{3+}$ free-ion 4$f_{5/2}$ multiplet from values expected from the behavior of the heavier isostructural rare-earth pnictides or chalcogenides.$^{16}$ This can be understood$^{17}$ as arising from band-[*f*]{} hybridization effects. In both the cerium monopnictides and monochalcogenides, the CF splitting between the ${\Gamma_{7}}$ doublet and the ${\Gamma_{8}}$ quartet decreases with increasing anion size, from 150 K for CeP to 10 K in CeBi and from 130 K for CeS to 30 K for CeTe, and it is about the same for the same row in both series, a rather surprising result in view of the additional valence electron on the chalcogen ion.$^{18}$ Neutron scattering experiments have shown$^{19}$ that the ${\Gamma_{7}}$-doublet is the CF ground state in all the cerium pnictides and chalcogenides.
In this paper we present material-predictive results from two [*ab initio*]{} based methods to study the change of magnetic properties across this series of cerium systems. The first, [*ab initio*]{} based, method gives clear insight into the role of the three pertinent interactions: 1) The band-[*f*]{} hybridization-induced inter-cerium magnetic coupling; 2) the corresponding effects of band-[*f*]{} coulomb exchange; and 3) the crystal-field interaction. This approach allows us also to understand the interplay between these interactions as the degree of 4[*f*]{} localization is varied across the series. The predictive calculations give results for the magnetic moments, magnetic structure, and ordering temperatures in excellent agreement with experiment. Thus, this approach allows to understand and predict a number of key features of observed behavior. First, is the very low moment and low ordering temperature of the antiferromagnetism observed in CeTe, an incipient heavy Fermion system. (For a review of theory and experimental behavior of heavy Fermion systems see references 2,3,20,21.) This [*ab initio*]{}-based method, described below, predicts the [*magnetic moment and ordering temperature collapse*]{} from CeSb to CeTe, both systems having about the same lattice constant but CeTe having an additional [*p*]{} electron. The origin of the moment collapse is of non-Kondo origin. The earlier work of Sheng and Cooper$^5$ showed that this magnetic ordering reduction is accurately predicted without including any crystal-field effects. An erroneous statement appears in the recent review article by Santini [*et al.*]{}$^{22}$ stating that crystal-field effects played an important role in the calculated results of Sheng and Cooper.$^5$ This is incorrect, since crystal-field effects were [*not*]{} included in these calculations. We show in this paper that including the crystal-field effects modifies this behavior only quantitatively. Second, our results demonstrate that, while the band-[*f*]{} coulomb exchange mediated interatomic 4[*f*]{}-4[*f*]{} interactions dominate the magnetic behavior for the heavier systems, which are more localized because of the larger Ce-Ce separation, the opposite is true for the lighter, more delocalized systems, where the hybridization-mediated coupling dominates the magnetic behavior. This reflects the great sensitivity of the relative importance of hybridization and coulomb exchange effects on magnetic ordering depending on the degree of 4[*f*]{} localization. Third, we show that for the lighter more delocalized systems the crystal-field interactions are much larger than the inter-cerium interactions and hence dominate the magnetic behavior. Finally, we predict the experimentally observed change of the ground-state magnetic structure from the $<001>$ antiferromagnetic type IA ($\uparrow \uparrow \downarrow \downarrow)$ in CeBi and CeSb to the $<001>$ antiferromagnetic type I ($\uparrow \downarrow$) in CeAs and CeP. On the other hand, the second [*ab initio*]{} method, based on density functional theory within the local density approximation (LDA),$^{23,24}$ fails to predict, even qualitatively, the trend of magnetic properties in this series of strongly correlated electron systems.
The first, [*ab initio*]{} based, method employs the degenerate Anderson lattice model which incorporates explicitly the hybridization and the coulomb exchange interactions on an equal footing$^{4,5}$
$$\begin{aligned}
H &=&\sum_{k} \epsilon_{k} c_{k}^{+} c_{k}
+ \sum_{Rm} \epsilon_{m} f_{m}^{+}(R) f_{m}(R) \\
& & + \frac{U}{2} \sum_{R,m \ne m'} n_{m}(R) n_{m'}(R) \\
& & +\sum_{kmR} [ V_{km} e^{-i{\bf k \cdot R}} c_{k}^{+} f_{m}(R)
+ H.C. ] \\
& & - \sum_{kk'} \sum_{mm'R} J_{mm'}({\bf k,k'})
e^{-i{\bf (k-k') \cdot R}} c_{k}^{+} f_{m}^{+}(R) c_{k'}
f_{m'}(R).\end{aligned}$$ $$$$ The parameters entering the model Hamiltonian, i.e., the band energies $\epsilon_k$, the [*f*]{}-state energy $\epsilon_m$, the on-site coulomb repulsion U, the hybridization matrix elements, V$_{km}$, and the band-[*f*]{} coulomb exchange $J_{m m'}({\bf k, k'}) =
\left< \phi_{k}^{*}({\it r}_{1}) \psi_{m}^{*}({\it r}_{2})
\left|\frac{1}{{\it r}_{12}}\right|
\psi_{m'}({\it r}_{1})\phi_{k'}({\it r}_{2})\right>$ are evaluated on a wholly [*ab initio*]{} basis from non-spin polarized full potential linear muffin tin orbital$^{23}$ (FPLMTO) calculations. Here, ${\it r}_{12}$ stands for $|{\it r}_{1}-{\it r}_{2}|$; $\phi_{k}$ are the non-[*f*]{} basis states of the FPLMTO, and $\psi_{m}$ are the localized [*f*]{} states. Because the size of both the hybridization and coulomb exchange matrix elements are much smaller ($\sim$ 0.1 eV) than the intraatomic coulomb interaction U (6eV), one can apply perturbation theory and evaluate the anisotropic two-ion 6X6 interaction matrices$^{4,5}$, $E_{m_{1}m'_{1}}^{m_{2}m'_{2}}{\bf (R_{2}-R_{1})}$, which couple the two [*f*]{}-ions. The exchange interactions have three contributions: the wholly band-[*f*]{} coulomb exchange mediated interaction proportional to $J_{mm'}^{2}({\bf k,k'})$, the wholly hybridization-mediated exchange interaction proportional to $V_{km}^{4}$, and the cross term proportional to $V_{km}^{2} J_{mm'}({\bf k,k'})$. With the two-ion interactions having been determined, the low-temperature magnetic moment and the ordering temperature can be determined by use of a mean field calculation.$^{4,5,8}$ We have previously applied this [*ab initio*]{} based method to investigate the effect of hybridization-induced cerium-cerium interactions$^{4,17}$ and the combined effect of both the hybridization and coulomb induced interactions$^{5}$ on the magnetic properties of the heavier cerium pnictides and chalcogenides (CeBi, CeSb, and CeTe). However, these calculations did not take into account the crystal field interaction and employed a warped muffin-tin LMTO calculation for the parameters entering the model. The excellent agreement found$^{5}$ with experiment for the low-temperature magnetic moment and ordering temperature is relatively unaffected by the CF interaction, because the CF interaction in the heavier cerium systems is smaller than the two-ion exchange interactions.
The second method employs [*ab initio*]{} spin polarized electronic structure calculations based on the FPLMTO method$^{23}$ using 1) only spin polarization, with the orbital polarization included only through the spin-orbit coupling, and 2) both the spin and orbital polarization polarization.$^{24}$ In these calculations the 4[*f*]{} states are treated as band states. The orbital polarization is taken into account by means of an eigenvalue shift$^{24}$, $\Delta V_{m} = - E^{3} L_{z} m_{l}$, for the 4[*f*]{} atom. Here, L$_{z}$ is the z-component of the cerium total orbital moment, m$_{l}$ is the magnetic quantum number, and E$^{3}$ is the Racah parameter evaluated self-consistently at each iteration.
The crystalline field, which was neglected in the previous calculations,$^{4,5}$ is expected to affect the magnetic behavior considerably, if it is large. It is important to emphasize that since in the first method the 4[*f*]{} states are treated as core states, they interact only with the spherical component of the effective one-electron potential. Thus, the interaction of the [*atomic-like*]{} 4f state with the [*non-spherical*]{} components of the potential, giving rise to the CF splitting, $\Delta_{CF} = \epsilon_{\Gamma _8} -
\epsilon_{\Gamma _7}$, is not included in the calculation of the model Hamiltonian parameters. In this paper, we generalize the first, [*ab initio*]{} based, method to include both the interatomic 4[*f*]{}-4[*f*]{} coupling and the crystal-field interactions on an equal footing and to employ a full potential LMTO evaluation of the model Hamiltonian parameters. While the effect of the full potential on both the hybridization and coulomb exchange interactions is small, including the CF interaction will be shown to play a role as important as the interatomic 4[*f*]{}-4[*f*]{} interactions for understanding and predicting the [*overall trend*]{} in the unusual magnetic properties, as as one chemically tunes the degree of 4[*f*]{} localization across this series of strongly correlated electron systems. The resultant Hamiltonian is$^{4,5}$
$$\begin{aligned}
{\it H}& =& -\sum_{i,j}\sum_{~^{\mu,\nu}_{\epsilon,\sigma}}
\xi^{\epsilon \sigma}_{\mu \nu}(\theta_{ij})
e^{-{\it i}(\mu-\nu+\epsilon-\sigma)\phi_{ij}}
c^{\dag}_{\epsilon}(j)c_{\sigma}(j)
c^{\dag}_{\mu}(i)c_{\nu}(i) \\
&& + B_{4}\sum_{i}\left(O_{4}^{0}(i)+5O_{4}^{4}(i)\right) , \end{aligned}$$ $$$$ where the ${\xi}^{\epsilon \sigma}_{\mu \nu}(\theta_{ij})$ are the two-ion 4[*f*]{}-4[*f*]{} interaction matrices rotated to a common crystal-lattice axis, and the $O_{4}^{0}$ and $O_{4}^{4}$ are the Stevens operators equivalents acting on the Ce$^{3+}$ free-ion 4$f_{5/2}$ multiplet.$^{25}$ The CF splitting is $\Delta_{CF}$ = 360$B_4$; a positive $B_4$ value gives the $\Gamma_7$ ground state, which is experimentally observed.$^{19}$ While our work in progress is aimed at evaluating the CF splitting on a wholly [*ab initio*]{} basis, in the absence of an [*ab initio*]{} value of the CF interaction in this class of strongly correlated cerium systems, the $\Delta_{CF}$ is set to the experimental values listed in Table 3.$^{10,19}$
In Table \[taba\], we present the calculated values of the zero-temperature cerium magnetic moment from the FPLMTO electronic structure calculations. Listed in the table are values both with and without the orbital polarization correction taken into account. Note, the importance of including the orbital polarization in these 4f correlated electron systems. As expected, in all cases, the orbital polarization is found to be opposite to the spin polarization. Comparison of the total energies predicts that the magnetic anisotropy changes from the $<001>$ direction in the pnictides to the $<111>$ in the chalcogenides, in agreement with experiment. On the other hand, except perhaps for the lighter chalcogenides (CeS and CeSe), comparison of the [*ab initio*]{} and experimental values for the magnetic moment indicates the [*failure*]{} of the LDA calculations to treat properly the correlation effects of the 4[*f*]{} states (treated as valence states) within the LDA as the degree of 4[*f*]{} correlations increases in the heavier pnictide systems. Furthermore, the [*ab inito*]{} calculations fail to predict the large [*moment collapse*]{} from CeSb to CeTe, the latter being described as an incipient heavy Fermion system.$^{2,3,20,21}$
------ ----------- ----------- ------- ----------- ----------- ------- -------
EXPT
$\mu_{S}$ $\mu_{L}$ $\mu$ $\mu_{S}$ $\mu_{L}$ $\mu$ $\mu$
CeS -1.00 0.91 -0.09 -1.24 1.99 0.75 0.57
CeSe -1.08 1.02 -0.06 -1.26 2.07 0.81 0.57
CeTe -1.15 1.28 0.07 -1.31 2.29 0.98 0.30
CeP -0.80 0.55 -0.25 -0.85 1.27 0.43 0.80
CeAs -0.84 0.64 -0.20 -0.85 1.42 0.57 0.80
CeSb -0.86 0.74 -0.12 -0.91 1.61 0.70 2.06
CeBi -0.86 0.74 -0.12 -0.95 1.69 0.74 2.10
------ ----------- ----------- ------- ----------- ----------- ------- -------
: Values of the calculated and experimental $^{7-11}$ low-temperature ordered magnetic moments for the cerium chalcogenides and pnictides in units of $\mu_{B}$. Listed are the LMTO values for the spin moment $\mu_{S}$, the orbital moment $\mu_{L}$, and total moment $\mu$, for the spin polarized only calculation and for the calculation with spin polarization and orbital polarization correction.
\[taba\]
In Table \[tabb\], we list the values of the $m$ = $m'$ =1/2 matrix elements (characteristic matrix elements of the 6X6 exchange interaction matrix) for the first three nearest-neighbor shells for the light (CeP and CeS) and the heavier compounds (CeSb and CeTe). Listed separately in this table are the three contributions to the interatomic 4[*f*]{}-4[*f*]{} interactions arising from band-[*f*]{} hybridization (V$^4$), band-[*f*]{} coulomb exchange (J$^2$), and the cross term. It is important to note that while the coulomb exchange mediated interactions dominate the magnetic behavior for the heavier, more localized, 4[*f*]{} systems, the opposite is true for the lighter, more delocalized, systems where the hybridization mediated interactions dominate the magnetic behavior. This change of behavior of the interatomic 4[*f*]{}-4[*f*]{} interactions is a result of the sensitivity of the hybridization and coulomb exchange to the degree of 4[*f*]{} localization. Equally important, is that while both first and second nearest-neighbor 4[*f*]{}-4[*f*]{} interactions are ferromagnetic for CeSb, there is an [*interplay*]{} between ferromagnetic first nearest-neighbor and antiferromagnetic second nearest-neighbor interactions for CeTe.(These interactions are mediated via scattering of conduction electrons). This results in a saturated ordered moment for CeSb and in the ordered [*magnetic moment collapse*]{} for CeTe (see Table III).
In order to determine whether the magnetic moment collapse might be of Kondo origin, we have evaluated the single-impurity Kondo temperature,$^{26}$ k$_B$T$_K$ = De$^{-\frac{1}{2\rho(E_F)|J(E_F)|}}$, across the series. Here, D is the bandwidth of the conduction electron states, $\rho(E_F)$ is the density of states of the conduction electrons at the Fermi energy, and $J(E_F)$ is the conduction electron-[*f*]{} exchange interaction at the Fermi energy, which has contributions both from the coulomb exchange interaction in Eq. (2), provided that it is negative, and the hybridization-induced exchange interaction $|J_{hyb}(E_F)| = \frac{V^2(E_F)U}{(|E_f - E_F|)(|E_f - E_F| + U)}$, where $J_{hyb}(E_F) < 0$. We find that the coulomb exchange interaction in Eq. (2) evaluated at the Fermi energy is positive across the entire series and hence cannot give rise to the Kondo effect. Thus, only the hybridization-induced exchange interaction, $J_{hyb}(E_F)$, can give rise to the Kondo effect.$^{26}$ Using the [*ab initio*]{} values of the parameters entering the expression for T$_K$, we find that T$_K \ll$ T$_{ord}$ across the entire series(T$_K <$ 10$^{-4}$K). The Kondo temperatures in the monopnictide series is smaller than that in the chaclogenides, due to the fact that in the pnictides the Fermi energy lies in the pseudogap, resulting in low $\rho(E_F)$. These results suggest that the moment collapse from CeSb to CeTe is of non-Kondo origin. Rather, it results from an interplay of ferromagnetic and antiferromagnetic interatomic 4[*f*]{}-4[*f*]{} interactions which arises purely from differences in the underlying electronic structure.
--------------------------------------- ------------- -------------- ------------- ------------- -------------- -------------
$E_{V^{4}}$ $E_{V^{2}J}$ $E_{J^{2}}$ $E_{V^{4}}$ $E_{V^{2}J}$ $E_{J^{2}}$
R = ( $\frac{1}{2}$ $\frac{1}{2}$ 0 ) 2.23 0.64 1.53 0.85 -0.40 1.50
R = ( 1 0 0 ) 6.39 0.27 1.65 -1.60 0.04 -0.80
R = ( 1 $\frac{1}{2}$ $\frac{1}{2}$ ) -0.08 -0.02 0.16 0.38 -0.16 0.13
--------------------------------------- ------------- -------------- ------------- ------------- -------------- -------------
: Values of the $m$ = $m'$ =1/2 matrix elements (characteristic matrix elements of the 6X6 interatomic 4[*f*]{}-4[*f*]{} interaction matrix $E^{m_b m'_b}_{m_a m'_a}({\bf R_{b}-R_{a}})$, for the first, second, and third nearest-neighbor shells in degrees Kelvin. Listed are the values of the hybridization induced ($E_{V^{4}}$), cross terms ($E_{V^{2}J}$), and pure coulomb exchange ($E_{J^{2}}$) contributions.
--------------------------------------- ------------- -------------- ------------- ------------- -------------- -------------
$E_{V^{4}}$ $E_{V^{2}J}$ $E_{J^{2}}$ $E_{V^{4}}$ $E_{V^{2}J}$ $E_{J^{2}}$
R = ( $\frac{1}{2}$ $\frac{1}{2}$ 0 ) 0.70 0.34 7.30 0.17 -0.19 2.90
R = ( 1 0 0 ) 2.07 0.07 10.21 -0.19 0.04 -1.69
R = ( 1 $\frac{1}{2}$ $\frac{1}{2}$ ) -0.02 -0.03 0.40 0.04 -0.06 -0.01
--------------------------------------- ------------- -------------- ------------- ------------- -------------- -------------
: Values of the $m$ = $m'$ =1/2 matrix elements (characteristic matrix elements of the 6X6 interatomic 4[*f*]{}-4[*f*]{} interaction matrix $E^{m_b m'_b}_{m_a m'_a}({\bf R_{b}-R_{a}})$, for the first, second, and third nearest-neighbor shells in degrees Kelvin. Listed are the values of the hybridization induced ($E_{V^{4}}$), cross terms ($E_{V^{2}J}$), and pure coulomb exchange ($E_{J^{2}}$) contributions.
\[tabb\]
Listed in Table \[tabc\] are the calculated zero-temperature ordered moment and ordering temperature, T$_N$, from the first, [*ab initio*]{} based, method, with and without the CF interaction. It is clear that for the heavier systems (CeBi, CeSb, CeTe) the effect of the CF interaction on the magnetic moments is small, and it is slightly more pronounced on the ordering temperatures. This is due to the fact that for the more localized systems the CF interaction is smaller than the two-ion interactions. This is the reason that the previous calculations,$^5$ neglecting the CF interaction, gave results in very good agreement with experiment. On the other hand, for the lighter more delocalized systems the CF interactions are much larger than the interatomic 4[*f*]{}-4[*f*]{} interactions, and hence dominate the magnetic behavior. The overall decrease of the magnetic moments in the presence of the CF interaction in all systems, arises from the mixing of the off-diagonal angular momentum states $|\pm5/2>$ and $|\mp3/2>$ states from the CF interaction with $\Gamma_7$ ground state. Overall, we find that the first, [*ab initio*]{} based, approach which takes into account all three pertinent interactions (hybridization, coulomb exchange, and CF interactions) on an equal footing, yields results for both the zero-temperature moment and the ordered temperature (a more stringent test for the theory) in excellent agreement with experiment.
A final corroboration of the success of the first [*ab initio*]{} based method is that it predicts the experimentally observed change of the ground-state magnetic structure from the $<001>$ antiferromagnetic type IA ($\uparrow \uparrow \downarrow \downarrow)$ in CeBi and CeSb to the $<001>$ antiferromagnetic type I ($\uparrow \downarrow$) in CeAs and CeP.$^{7,15}$ More specifically, the sign \[ferromagnetic (F) or antiferromagnetic (AF)\] of the $|\pm 5/2>$ matrix elements of the 6X6 exchange matrix determines the [*interplanar*]{} interaction between successive (001) Ce planes. We find, that for the heavier compounds (CeBi and CeSb) the $|\pm 5/2>$ matrix elements of the [*coulomb exchange*]{} matrix are FM and hence favor the $\uparrow \uparrow \downarrow \downarrow$ type, while in the lighter systems (CeAs and CeP) the $|\pm 5/2>$ matrix elements of the [*hybridization*]{}-induced two-ion matrix are AF, and hence they favor the $\uparrow \downarrow$ type.
[|d|d|c|c|c|c|c|c|d|]{} & $\Delta_{CF}$ & & & & &\
& & no CF & CF & exp & & no CF & CF & exp\
CeS & 140 & 1.80 & 0.73 & 0.57 & & 1.0 & 11.0 & 8.4\
CeSe & 116 & 1.10 & 0.79 & 0.57 & & 2.5 & 14.0 & 5.7\
CeTe & 32 & 0.60 & 0.46 & 0.30 & & 8.0 & 5.0 & 2.2\
CeP & 150 & 2.10 & 0.73 & 0.81 & & 14 & 11 & 8\
CeAs & 137 & 2.10 & 0.74 & 0.85 & & 16 & 13 & 8\
CeSb & 37 & 2.10 & 1.80 & 2.06 & & 20 & 18 & 17\
CeBi & 8 & 2.10 & 2.10 & 2.10 & & 40 & 40 & 26\
\[tabc\]
In conclusion, we have applied two different, [*ab inito*]{} based and [*ab initio*]{} LDA, methods to study the dramatic change of magnetic properties across a series of strongly correlated electron systems which offer the opportunity to chemically tailor the different type of interactions (band-[*f*]{} hybridization, band-[*f*]{} coulomb exchange, and CF interactions), pertinent to the unusual magnetic behavior. The first, [*ab initio*]{} based, approach which explicitly takes into account the interplay of the three pertinent interactions, gives results in excellent agreement with experiment for all compounds in the series, including the [*moment collapse*]{} from CeSb to CeTe and the trend of moments and ordering temperatures across the series. The remaining problem of determining on a wholly [*ab initio*]{} basis the [*suppressed*]{} crystal-field interactions in this class of systems poses a theoretical challenge for future theoretical work. On the other hand, the second, fully [*ab initio*]{} LDA, method gives good results for the lighter chalcogenide systems, but it entirely fails to give, even qualitatively, the trend of the unusual magnetic behavior.
The research at California State University Northridge (CSUN) was supported by the National Science Foundation under Grant No. DMR-9531005, by the US Army Grant No. DAAH04-95, and the CSUN Office of Research and Sponsored Projects, and at West Virginia University by the NSF under Grant No. DMR-9120333.
, eds. A. Gonis, N. Kioussis, and M. Ciftan, (Kluwer Academic/Plenum, New York, 1999).
D. W. Hess, P.S. Riseborough, and J.L. Smith [*Heavy Fermion Phenomena*]{}, Encyclopedia of Applied Physics, edited by G. Trigg, (VCH Publishers Inc., New York, 1993), Vol. 71, p. 435.
N. Grewe and F. Steglich, [*Heavy Fermions*]{}, pp. 343-479, in Handbook on the Physics and Chemistry of the Rare Earths, eds. K. A. Grchneidner Jr. and L. Eyring, (Elsevier Science Publishers, Amsterdam, 1991), Vol. 14, pp. 343.
N. Kioussis, B. R. Cooper, and J. M. Wills, Phys. Rev. B [**44**]{}, 10003 (1991).
Q. G. Sheng and B. R. Cooper, J. Appl. Phys. [**69**]{}, 5472 (1991); Phys. Rev. B [**50**]{}, 965 (1994).
B. R. Cooper, O. Vogt, Q. G. Sheng, and Y. Lin, Phil. Mag. B [**79**]{}, 683 (1999).
J. Rossat-Mignod, P. Burlet, S. Quezel, J. M. Effantin, D. Delacôte, H. Bartholin, O. Vogt, and D. Ravot, J. Magn. Magn. Mater. [**31-34**]{}, 398 (1983).
B. R. Cooper, R. Siemann, D. Yang, P. Thayamballi, and A. Banerjea, in [*The Handbook of the Physics and Chemistry of the Actinides*]{}, edited by A. J. Freeman and G. H. Lander (North-Holland, Amsterdam, 1985), Vol. 2, Chap. 6, pp. 435-500.
A. Dönni, A. Furrer, P. Fischer, and F. Hulliger, Physica B [**186-188**]{}, 541 (1993).
F. Hulliger, B. Natterer, and H. R. Ott, J. Magn. Magn. Mater. [**8**]{}, 87 (1978).
H. R. Ott, J. K. Kjems, and F. Hulliger, Phys. Rev Lett. [**42**]{}, 1378 (1979).
J. Rossat-Mignod, J. M. Effantin, P. Burlet, T. Chattopadhyay, L. P. Regnault, H. Bartholin, C. Vettier, O. Vogt, D. Ravot, and J. C. Achart, J. Magn. Magn. Mater. [**52**]{}, 111 (1985).
N. Môri, Y. Okayama, H. Takahashi, Y. Haga, and T. Suzuki, Physica B [**186-188**]{}, 444 (1993).
S. Doniach, Physica B [**91**]{}, 231 (1977).
Y. Okayama, Y. Ohara, S. Mituda, H. Takahashi, H. Yoshizawa, T. Osakabe, M. Kohgi, Y. Haga, T. Suzuki and N. Mōri, Physica B [**186-188**]{} 531 (1993).
R. J. Birgeneau, E. Bucher, J. P. Maita, L. Passell, and K. C. Turberfield, Phys. Rev. B [**8**]{}, 5345 (1973).
J. M. Wills and B. R. Cooper, Phys. Rev. B [**36**]{}, 3809 (1987).
H. R. Ott, J. K. Kjems, and F. Hulliger, Phys. Rev. Lett. [**42**]{}, 1378 (1979); F. Hulliger and H.R. Ott, J. Phys. (Paris) Colloq. [**40**]{}, C5, 128 (1979).
J. Rossat-Mignod, J. M. Effantin, P. Burlet, T. Chattopadhyay, L. P. Regnault, H. Bartholin, C. Vettier, O. Vogt, D. Ravot, and J.C. Achart, J. Magn. Magn. Mater. [**52**]{} 111 (1985).
P. A. Lee, T. M. Rice, J. W. Serene, L. J. Sham, and J. W. Wilkins, Condens. Matter Phys. [**12**]{}, 99 (1986).
For an experimental review see, G. R. Stewart, Rev. Mod. Phys. [**56**]{}, 755 (1984).
P. Santini, R. Lemanski, and P. Erdos, Advances in Physics, [**48**]{}, 537 (1999).
D. L. Price and B.R. Cooper, Phys. Rev. B [**39**]{}, 4945 (1989).
M. S.S. Brooks and P. J. Kelly, Phys. Rev. Lett. [**51**]{}, 1708 (1983); O. Eriksson, M. S. S. Brooks, and B. Johansson, Phys. Rev. B [**41**]{}, 7311 (1990).
K. R. Lea, M. J. M. Leask, and W. P. Wolf, J. Phys. Chem. Solids [**23**]{}, 1381 (1962).
Peter Fulde, [*Electron Correlations in Molecules and Solids*]{}, (Springer-Verlag, Berlin, 1991), pp. 277-281.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'An implicit fundamental assumption in relativistic perturbation theory is that there exists a parametric family of spacetimes that can be Taylor expanded around a background. The choice of the latter is crucial to obtain a manageable theory, so that it is sometime convenient to construct a perturbative formalism based on two (or more) parameters. The study of perturbations of rotating stars is a good example: in this case one can treat the stationary axisymmetric star using a slow rotation approximation (expansion in the angular velocity $\Omega$), so that the background is spherical. Generic perturbations of the rotating star (say parametrized by $\lambda$) are then built on top of the axisymmetric perturbations in $\Omega$. Clearly, any interesting physics requires non–linear perturbations, as at least terms $\lambda\Omega$ need to be considered. In this paper we analyse the gauge dependence of non–linear perturbations depending on two parameters, derive explicit higher order gauge transformation rules, and define gauge invariance. The formalism is completely general and can be used in different applications of general relativity or any other spacetime theory.'
address:
- ' Institute of Cosmology and Gravitation, Mercantile House, University of Portsmouth, Portsmouth PO1 2EG, Britain'
- '§ Dipartimento di Fisica ”G. Marconi”, Università di Roma ”La Sapienza” and Sezione INFN ROMA 1, piazzale Aldo Moro 2, I–00185 Roma, Italy'
author:
- 'Marco Bruni Leonardo Gualtieri§ and Carlos F. Sopuerta'
title: 'Two–parameter non–linear spacetime perturbations: gauge transformations and gauge invariance'
---
Introduction
============
An implicit fundamental assumption in relativistic perturbation theory is that there exists a parametric family of spacetimes such that the perturbative formalism is built as a Taylor expansion of this family around a background. The perturbations are then defined as the derivative terms of this series, evaluated on this background [@wald]. In most cases of interest one deals with an expansion in a single parameter, which can either be a formal one, as in cosmology [@BMMS; @MMB; @BMT] or in the study of quasi–normal modes of stars and black holes [@chandrabook; @kokkomodes], or can have a specific physical meaning, as in the study of binary black hole mergers via the close limit approximation [@price; @gleisclose], or in the study of quasi–normal mode excitation by a physical source (see [@kokkomodes; @my123] and references therein). Typically the perturbative expansion stops at the first order, but recent interesting developments deal with second order perturbations [@BMMS; @MMB; @BMT; @cl; @gleiser].
In some physical applications it may be instead convenient to construct a perturbative formalism based on two (or more) parameters, because the choice of background is crucial in having a manageable theory. The study of perturbations of stationary axisymmetric rotating stars (see [@kojimarev; @laf; @rsk] and references therein) is a good example. In this case, an analytic stationary axisymmetric solution is not known, at least for reasonably interesting equations of state. A common procedure is to treat axisymmetric stars using the so–called slow rotation approximation, so that the background is a star with spherical symmetry [@hartle; @hartlethorne]. In this approach the first order in $\Omega$ discloses frame dragging effects, with the star actually remaining spherical; $\Omega^2$ terms carry the effects of rotation on the fluid. This is intuitive from a Newtonian point of view, as rotational kinetic energy goes like $\Omega^{2}$. This approximation is valid for angular velocities $\Omega$ much smaller than the mass shedding limit $\Omega_K\equiv\sqrt{M/R_{star}^3}$, with typical values for neutron stars $\Omega_K\sim 10^3Hz$. Therefore the slow rotation approximation, despite the name, can still be valid for large angular velocities. In practice, the perturbative approach up to $\Omega^2$ is accurate for most astrophysical situations, with the exception of newly born neutron stars (see [@STER] and references therein).
Given that the differential operators appearing in the perturbative treatment of a problem are those defined on the background, the theory is considerably simplified when the latter is spherical. Generic time dependent perturbations of the rotating star (parametrized by a dummy parameter $\lambda$ and describing oscillations) are then built on top of the stationary axisymmetric perturbations in $\Omega$. Clearly, in this approach any interesting physics requires non–linear perturbations, as at least terms of order $\lambda\Omega$ need to be considered. A similar approach could be used to study perturbations of the slowly rotating collapse, even if in specific cases [@CPM; @seidel; @gundlach] the perturbative expansion depends by one parameter only.
Classical studies in the literature have not analysed in full the gauge dependence and gauge invariance of the non-linear perturbation theory. For example, in [@CPM] the second order perturbations are treated in a gauge invariant fashion on top of the first order perturbation in a given specific gauge. The perturbation variables used are therefore non gauge invariant under a complete second order gauge transformation [@BMMS; @FW], but only invariant under “first order transformations acting at second order” [@CPM]. While this may be perfectly satisfactory from the point of view of obtaining physical results, one may wish to convert results in a given gauge to a different one [@BMMS; @MMB], to compare results obtained in two different gauges, or to construct a fully gauge invariant formalism. To this end one needs to know the gauge transformation rules and the rules for gauge invariance, either [*up to*]{} order $n$ [@BMMS] or [*at*]{} order $n$ [*only*]{}, as in [@CPM]. The situation is going to be more complicated in the case of two parameters, as we shall see.[^1]
In this paper we keep in mind the above practical examples, but we do not make any specific assumption on the background spacetime and the two–parameter family it belongs to. As in [@BMMS; @SB; @BS], we do not even need to assume that the background is a solution of Einstein’s field equations: the formalism is completely general and can be applied to any spacetime theory. We analyse the gauge dependence of perturbations in the case when they depend on two parameters, $\lambda$ and $\Omega$, derive explicit gauge transformation rules up to fourth order, i.e. including any term $\lambda^k\Omega^{k'}$ with $k+k'\leq 4$, and define gauge invariance. This choice of keeping fixed the total perturbative order is due to the generality of our approach. In practical applications one would be guided by the physical characteristics of the problem in deciding where to truncate the perturbative expansion. For example, in the case of a rotating star one could consider first order oscillations, parametrized by $\lambda$, on top of a stationary axisymmetric background described up to $\Omega^2$, neglecting therefore $\lambda^2\Omega$ terms. Or instead, one could decide that $\lambda^2\Omega$ terms are more interesting than the $\lambda\Omega^2$ ones in certain cases.
From a practical point of view, our aim is to derive the effects of gauge transformations on tensor fields $T$ up to order $k+k'=4$. It is indeed reasonable to assume that in a practical example like that of rotating stars, at most one will want to consider second order oscillations $\sim\lambda^2$ on top of a slowly rotating background described up to $O(\Omega^2)$, in order to take into account large oscillations and fluid deformations due to rotation.
We will show that the coordinate form of a two–parameter gauge transformation can be represented by: \^ & = & x\^+\^\_[(1,0)]{}+ \^\_[(0,1)]{}\
& & +(\^\_[(2,0)]{}+\^\_[(1,0)]{} \^\_[(1,0),]{}) +(\^\_[(0,2)]{}+\^\_[(0,1)]{} \^\_[(0,1),]{})\
& & +(\^\_[(1,1)]{}+\_0\^\_[(1,0)]{} \^\_[(0,1),]{}+\_1\^\_[(0,1)]{} \^\_[(1,0),]{})+O\^3(,), \[coord2\] where the full expression is given in Eq. (\[coordtransf\]). Here $\xi^\mu_{(1,0)}$, $\xi^\mu_{(0,1)}$, $\xi^\mu_{(2,0)}$, $\xi^\mu_{(1,1)}$, and $\xi^\mu_{(0,2)}$ are independent vector fields and $(\epsilon_0,\epsilon_1)$ are any two real numbers satisfying $\epsilon_0+\epsilon_1=1$. Coupling terms like the $\lambda\Omega$ in (\[coord2\]) are the expected new features of the two–parameter case, cf. [@BMMS; @SB]. Our main results are the explicit transformation rules for the perturbations of a tensor field $T$ and the conditions for the gauge invariance of these perturbations.
The paper is organized as follows: in Section \[taylorexp\] we develop the necessary mathematical tools, deriving Taylor expansion formulae for two–parameter groups of diffeomorphisms and for general two–parameter families of diffeomorphisms. In Section \[gauges\] we set up an appropriate geometrical description of the gauge dependence of perturbations in the specific case of two–parameter families of spacetimes. In Section \[gaugeiandt\] we apply the tools developed in Section \[taylorexp\] to the framework introduced in Section \[gauges\], in order to define gauge invariance and formulas for gauge transformations, up to fourth order in the two–parameter perturbative expansion. Section \[conclusions\] is devoted to the conclusions. We follow the notation used previously in [@BMMS; @SB; @BS] for the case of one parameter perturbations.
Taylor expansion of tensor fields {#taylorexp}
=================================
In order to consider the issues of gauge transformations and gauge invariance in two–parameter perturbation theory we need first to introduce some mathematical tools concerning the two–parameter Taylor expansion of tensor fields. Since Taylor expansions are aimed to provide the value of a quantity at some point in terms of its value, and the value of its derivatives, at another point, a Taylor expansion of tensorial quantities can only be defined through a mapping between tensors at different points of the manifold under consideration. In this section we consider the cases where such a mapping is given by a two–parameter family of diffeomorphisms of $\cM$, starting from the simplest case in which such a family constitutes a group.
Two–parameter groups of diffeomorphisms {#groupdif}
---------------------------------------
Given a differentiable manifold $\cM$, a two–parameter group of diffeomorphisms $\phi$ of $\cM$ can be represented as follows
[cccc]{} :&\^2 & & \^2\
&(p,,)&|& (\_[,]{}(p),,).\
For the purpose of introducing two-parameter perturbation theory, where perturbing first with respect to the parameter $\lambda$ and afterwards with respect to the parameter $\Omega$ should be equivalent to the converse operation, we will assume that $\phi_{\lambda,\Omega}$ is such that it satisfies the following property[^2] \_[\_1,\_1]{}\_[\_2,\_2]{}= \_[\_1+\_2,\_1+\_2]{}, ,. By one hand, this property implies that the two-parameter group is Abelian. On the other hand, it allow us to make the following useful decomposition of $\phi_{\lambda,\Omega}$ into two one–parameter groups of diffeomorphisms (flows) that remain implicitly defined by the equalities \_[,]{}=\_[,0]{}\_[0,]{} =\_[0,]{}\_[,0]{}. \[philambda\] The action of the flows $\phi_{\lambda,0}$ and $\phi_{0,\Omega}$ is generated by two vector fields, $\eta$ and $\zeta$ respectively, acting on the tangent space of $\cM\times\RR^2$. The Lie derivatives of a generic tensor $T$ with respect to $\eta$ and $\zeta$ are \_T&: = \_[0]{} (\^\*\_[,0]{}T-T) = &\_[=0]{} =\_,\
\_T&:= \_[0]{} (\^\*\_[0,]{}T-T) = &\_[=0]{} =\_, where the superscript ${}^*$ denotes the pull–back map associated with the corresponding diffeomorphism [@BMMS]. Because the group is Abelian the vector fields $\eta$ and $\zeta$ must commute =0. The Taylor expansion of the pull–backs $\phi^*_{\lambda,0} T,\phi^*_{0,\Omega} T$ is given by (see [@BMMS]) \^\*\_[,0]{}T&=\_[k=0]{}\^ \_[=0]{}= &\_[k=0]{}\^\^k\_T ,\
\^\*\_[0,]{}T&=\_[k=0]{}\^ \_[=0]{}= &\_[k=0]{}\^ \^k\_T. From this, using (\[philambda\]), we can derive the Taylor expansion of the two–parameter group of pull–backs $\phi^*_{\lambda,\Omega}T$: \^\*\_[,]{}T = \_[k,k’=0]{}\^ \_ = \_[k,k’=0]{}\^ \^k\_\^[k’]{}\_T.\[Taylorgroup\]
Two–parameter families of diffeomorphisms {#familydiffeo}
-----------------------------------------
Let us now consider the general case of a two–parameter family of diffeomorphisms $\Phi$:
[cccc]{} :&\^2 & & \^2\
&(p,,)&|& (\_[,]{}(p),,) .\
In this generic case the diffeomorphisms $\Phi_{\lambda,\Omega}$ do not form in general a group. In particular, \_[\_1,\_1]{}\_[\_2,\_2]{}\_[\_1+\_2,\_1+\_2]{}, which means that we cannot decompose $\Phi_{\lambda,\Omega}$ as in the case of a two–parameter group of diffeomorphisms. The Taylor expansion of the pull–back of $\Phi_{\lambda,\Omega}$ is formally given by \^\*\_[,]{} T= \_[k,k’=0]{}\^ \_. \[Taylorfamily\] Since the diffeomorphisms $\Phi_{\lambda,\Omega}$ do not form a group we cannot write this expansion directly in terms of Lie derivatives as in the previous case, Eq. (\[Taylorgroup\]). Nevertheless, in order to study the characteristics and properties of the gauge transformations to be derived in Section \[gaugeiandt\], we would like to find an alternative way of expressing the expansion (\[Taylorfamily\]) in terms of suitable Lie derivatives, in a similar way as it was done in [@BMMS; @BS] in the one–parameter case. To this end, new objects called [*knight diffeomorphisms*]{} were introduced in [@BMMS; @BS]. Broadly speaking, they are a composition of one-parameter groups of diffeomorphisms that can reproduce the action of a given family of diffeomorphisms, in such a way that the more groups we compose the better the approximation is. When the number of composed groups tends to infinity the reproduction of the family is exact (see [@BMMS; @SB; @BS] for a detailed account on knight diffeomorphisms). The main aim for using these objects was to show [@BMMS; @BS] that it is possible to expand a one-parameter family of diffeomorphism, at orden $n$, using Lie derivatives with respect to a finite number $n$ of vector fields. Knight diffeomorphisms constitute an elegant formulation of this kind of expansions. In order to apply these ideas to our case, we can think of translating the idea of knight diffeomorphims from the one-parameter to the two-parameter case. We have studied this question and we have found that in the two–parameter case there are several ways of formulating knight diffeomorphisms, and they lead to different formal expansions of a two-parameter family of diffeomorphisms. Actually, we have found that some of these formulations can be inconsistent in the sense that the differential operators that come out from them do not satisfy the Leibnitz rule. Then, we cannot associated to any of these operators a vector field whose Lie derivative will describe the action of that operator. Despite of these facts, we find that the main goal for introducing knight diffeomorphisms can be achieved in the two-parameter case. That is, we can still expand a two-parameter family of diffeomorphisms, at order $n$, using Lie derivatives with respect to a finite number of vector fields.
Then, here we are going to show how to approximate a given family of diffeomorphisms $\Phi$, up to order $\lambda^k\Omega^{k'}$ with $k+k'=4$, in terms of some differential operators that we will show later can be identified with Lie derivative operators with respect to some vector field. In other words, we are looking for an expansion of the following type: \^\_[,]{}T & & = \_
[c]{}k,k’=0\
k+k’=4
\^4 \^k\^[k’]{}()\^\_[k,k’]{}T. \[mainidea\] We have performed this expansion term by term. For the sake of brevity we present here only the most relevant developments and results. We start by introducing the set of differential operators that we have used to build such an expansion by giving their actions on a general tensorial quantity $T$:[^3] \_[(1,0)]{}T & := & \_ , \[operator10\]\
\_[(0,1)]{}T & := & \_ ,\
\_[(2,0)]{}T & := & \_[ ]{}-\^2\_[(1,0)]{}T,\
\_[(1,1)]{}T & := & \_ -(\_0\_[(1,0)]{}\_[(0,1)]{}+ \_1\_[(0,1)]{}\_[(1,0)]{} )T\[L11\],\
\_[(0,2)]{}T & := & \_[ ]{}-\^2\_[(0,1)]{}T,\
\_[(3,0)]{}T & := & \_[ ]{}-3\_[(1,0)]{}\_[(2,0)]{}T-\^3\_[(1,0)]{}T,\
\_[(2,1)]{}T & := & \_\
& & -2\_[(1,0)]{}\_[[(1,1)]{}]{}T-\_[[(0,1)]{}]{}\_[[(2,0)]{}]{}T- 2\_2\_[[(1,0)]{}]{}\_[[(0,1)]{}]{}\_[[(1,0)]{}]{}T\
& & -(\_1-\_2)\_[[(0,1)]{}]{}\^2\_[[(1,0)]{}]{}T -(\_0-\_2)\^2\_[[(1,0)]{}]{}\_[[(0,1)]{}]{}T \[L21\],\
\_[(1,2)]{}T & := & \_\
& & -2\_[[(0,1)]{}]{}\_[[(1,1)]{}]{}T-\_[[(1,0)]{}]{}\_[[(0,2)]{}]{}T- 2\_3\_[[(0,1)]{}]{}\_[[(1,0)]{}]{}\_[[(0,1)]{}]{}T\
& & -(\_0-\_3)\_[[(1,0)]{}]{}\^2\_[[(0,1)]{}]{}T -(\_1-\_3)\^2\_[[(0,1)]{}]{}\_[[(1,0)]{}]{}T \[L12\],\
\_[(0,3)]{}T & := & \_[ ]{}-3\_[(0,1)]{}\_[(0,2)]{}T-\^3\_[(0,1)]{}T,\
\_[(4,0)]{}T & := & \_[ ]{}\
& & -4\_[(1,0)]{}\_[(3,0)]{}T-3\^2\_[(2,0)]{}T -6\^2\_[(1,0)]{}\_[(2,0)]{}T-\^4\_[(1,0)]{}T,\
\_[(3,1)]{}T & := & \_\
& & -3\_[[(1,0)]{}]{}\_[[(2,1)]{}]{}T -\_[[(0,1)]{}]{}\_[[(3,0)]{}]{}T-3\_4\_[[(2,0)]{}]{}\_[[(1,1)]{}]{}T -3\_5\_[[(1,1)]{}]{}\_[[(2,0)]{}]{}T\
& & -3\^2\_[[(1,0)]{}]{}\_[[(1,1)]{}]{}T -3(\_0\_[[(1,0)]{}]{}\_[[(0,1)]{}]{}T +\_1\_[[(0,1)]{}]{}\_[[(1,0)]{}]{})\_[[(2,0)]{}]{}T\
& & -(\_1-\_2-\_6)\_[[(0,1)]{}]{}\^3\_[[(1,0)]{}]{}T -3\_6\_[[(1,0)]{}]{}\_[[(0,1)]{}]{}\^2\_[[(1,0)]{}]{}T\
& & -3(\_2-\_6)\^2\_[[(1,0)]{}]{}\_[[(0,1)]{}]{}\_[[(1,0)]{}]{}T -(\_0-2\_2+\_6)\^3\_[[(1,0)]{}]{}\_[[(0,1)]{}]{}T ,\
\_[(2,2)]{}T & := & \_\
& & -2\_[[(1,0)]{}]{}\_[[(1,2)]{}]{}T+2\_[[(0,1)]{}]{}\_[[(2,1)]{}]{}T- 2\^2\_[[(1,1)]{}]{}T\
& & -\_7\_[[(2,0)]{}]{}\_[[(0,2)]{}]{}T -\_8\_[[(0,2)]{}]{}\_[[(2,0)]{}]{}T - \^2\_[[(1,0)]{}]{}\_[[(0,2)]{}]{}T-\^2\_[[(0,1)]{}]{}\_[[(2,0)]{}]{}T\
& & -4(\_0\_[[(1,0)]{}]{}\_[[(0,1)]{}]{}T+ \_1\_[[(0,1)]{}]{}\_[[(1,0)]{}]{})\_[[(1,1)]{}]{}T\
& & +(\_3+\_2-\_1+\_9) \^2\_[[(0,1)]{}]{}\^2\_[[(1,0)]{}]{}T +(\_3+\_2-\_0-\_9) \^2\_[[(1,0)]{}]{}\^2\_[[(0,1)]{}]{}T\
& & -2(\_3+\_2-\_0\_1-\_9) \_[[(1,0)]{}]{}\_[[(0,1)]{}]{}\_[[(1,0)]{}]{}\_[[(0,1)]{}]{}T\
& & -2(\_3+\_2-\_0\_1+\_9) \_[[(0,1)]{}]{}\_[[(1,0)]{}]{}\_[[(0,1)]{}]{}\_[[(1,0)]{}]{}T\
& & +2(\_3-\_0\_1) \_[[(1,0)]{}]{}\^2\_[[(0,1)]{}]{}\_[[(1,0)]{}]{}T +2(\_2-\_0\_1) \_[[(0,1)]{}]{}\^2\_[[(1,0)]{}]{}\_[[(0,1)]{}]{}T ,\
\_[(1,3)]{}T & := & \_\
& & -3\_[[(0,1)]{}]{}\_[[(1,2)]{}]{}T -\_[[(1,0)]{}]{}\_[[(0,3)]{}]{}T-3\_[10]{}\_[[(0,2)]{}]{}\_[[(1,1)]{}]{}T -3\_[11]{}\_[[(1,1)]{}]{}\_[[(0,2)]{}]{}T\
& & -3\^2\_[[(0,1)]{}]{}\_[[(1,1)]{}]{}T -3(\_0\_[[(1,0)]{}]{}\_[[(0,1)]{}]{}T+ \_1\_[[(0,1)]{}]{}\_[[(1,0)]{}]{})\_[[(0,2)]{}]{}T\
& & -(\_0-\_3-\_[12]{})\_[[(1,0)]{}]{}\^3\_[[(0,1)]{}]{}T -3\_[12]{}\_[[(0,1)]{}]{}\_[[(1,0)]{}]{}\^2\_[[(0,1)]{}]{}T\
& & -3(\_3-\_[12]{})\^2\_[[(0,1)]{}]{}\_[[(1,0)]{}]{} \_[[(0,1)]{}]{}T-(\_1-2\_3+\_[12]{})\^3\_[[(0,1)]{}]{} \_[[(1,0)]{}]{}T ,\
\_[(0,4)]{}T & := & \_[ ]{}\
& & -4\_[(0,1)]{}\_[(0,3)]{}T-3\^2\_[(0,2)]{}T -6\^2\_[(0,1)]{}\_[(0,2)]{}T-\^4\_[(0,1)]{}T, \[operator04\] where the quantities $\epsilon_A$ ($A=0,\dots,12$) are real numbers which must satisfy the following conditions \_0+\_1 = 1, \_4+\_5 = 1, \_7+\_8 = 1, \_[10]{}+\_[11]{} = 1. \[constraints\] These constants reflect the fact that, as already mentioned above, there is not a unique way of constructing an expansion of the type (\[mainidea\]), actually the number of possibilities is infinite. In this paper we have restricted ourselves to expansions in which every single term has the form $\cL_{(p_1,q_1)}\cdots\cL_{(p_n,q_n)}T$ with $p_1+q_1 \leq \cdots \leq p_n+q_n$. Then, the constants $\epsilon_A$ express the freedom that we have in constructing the expansion (\[mainidea\]) with this criteria. This freedom and the non-uniqueness of the construction above is not a problem: the operators $\cL_{(p,q)}$ and the corresponding vectors (see below) that we are going to define should be though of as a “basis" for the construction in terms of Lie derivative of each of the terms in the expansion (\[Taylorfamily\]). This “basis" is not unique, but the result at each order is.
On the other hand, it is straightforward to check that these operators are linear and satisfy the Leibnitz rule and hence they are derivatives. We can also check that they satisfy the rest of conditions of the theorem stated in \[sebastiano\]. Therefore, for each of them there is a vector field $\xi_{(p,q)}$ such that \_[\_[(p,q)]{}]{}T := \_[(p,q)]{}T (p,q). \[lieder\]
In the particular case when $\Phi$ is a group of diffeomorphisms we recover the previous case (subsection \[groupdif\]), and $\cL_{(p,q)}=0$ if $p+q>1$.
Using the differential operators we have just introduced we can express the Taylor expansion (\[Taylorfamily\]) of $\Phi_{\lambda,\Omega}$, up to fourth order in $\lambda$ and $\Omega$, in terms of the Lie derivatives associated with the vector fields $\xi_{(p,q)}$ (\[lieder\]): \^\_[,]{}T & & = T + \_[\_[(1,0)]{}]{}T + \_[\_[(0,1)]{}]{}T\
& & + {\_[\_[(2,0)]{}]{} +\^2\_[\_[(1,0)]{}]{}}T+{\_[\_[(0,2)]{}]{} +\^2\_[\_[(0,1)]{}]{}}T\
& & + {\_[\_[(1,1)]{}]{}+\_0 \_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}+\_1\_[\_[(0,1)]{}]{} \_[\_[(1,0)]{}]{}}T\
& & + {\_[\_[(3,0)]{}]{}+3\_[\_[(1,0)]{}]{} \_[\_[(2,0)]{}]{}+\^3\_[\_[(1,0)]{}]{}}T\
& & + {\_[\_[(2,1)]{}]{} +2\_[\_[(1,0)]{}]{}\_[\_[(1,1)]{}]{}+\_[\_[(0,1)]{}]{}\_[\_[(2,0)]{}]{} +2\_2\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}.\
& & . +(\_1-\_2)\_[\_[(0,1)]{}]{}\^2\_[\_[(1,0)]{}]{} +(\_0-\_2)\^2\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}}T\
& & + {\_[\_[(1,2)]{}]{}+2 \_[\_[(0,1)]{}]{}\_[\_[(1,1)]{}]{}+\_[\_[(1,0)]{}]{}\_[\_[(0,2)]{}]{} +2\_3\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{}.\
& & . +(\_0-\_3)\_[\_[(1,0)]{}]{}\^2\_[\_[(0,1)]{}]{} +(\_1-\_3)\^2\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}}T\
& & + {\_[\_[(0,3)]{}]{}+3\_[\_[(0,1)]{}]{} \_[\_[(0,2)]{}]{}+\^3\_[\_[(0,1)]{}]{}}T\
& & + {\_[\_[(4,0)]{}]{}+4\_[\_[(1,0)]{}]{} \_[\_[(3,0)]{}]{}+3\^2\_[\_[(2,0)]{}]{}+6\^2\_[\_[(1,0)]{}]{} \_[\_[(2,0)]{}]{}+\^4\_[\_[(1,0)]{}]{}}T\
& & +{\_[\_[(3,1)]{}]{}+3\_[\_[(1,0)]{}]{} \_[\_[(2,1)]{}]{}+\_[\_[(0,1)]{}]{}\_[\_[(3,0)]{}]{}+3\_4 \_[\_[(2,0)]{}]{}\_[\_[(1,1)]{}]{}.\
& & +3\_5\_[\_[(1,1)]{}]{}\_[\_[(2,0)]{}]{} + 3\^2\_[\_[(1,0)]{}]{}\_[\_[(1,1)]{}]{} +3(\_0\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}+ \_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{})\_[\_[(2,0)]{}]{}\
& & + (\_1-\_2-\_6)\_[\_[(0,1)]{}]{} \^3\_[\_[(1,0)]{}]{}+3\_6\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{} \^2\_[\_[(1,0)]{}]{}\
& & . +3(\_2-\_6)\^2\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{} \_[\_[(1,0)]{}]{}+(\_0-2\_2+\_6)\^3\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{}}T\
& & + { \_[\_[(2,2)]{}]{}+2\_[\_[(1,0)]{}]{}\_[\_[(1,2)]{}]{}+2\_[\_[(0,1)]{}]{} \_[\_[(2,1)]{}]{}+2\^2\_[\_[(1,1)]{}]{} .\
& & + \_7\_[\_[(2,0)]{}]{}\_[\_[(0,2)]{}]{} +\_8\_[\_[(0,2)]{}]{}\_[\_[(2,0)]{}]{}+\^2\_[\_[(1,0)]{}]{} \_[\_[(0,2)]{}]{}+\^2\_[\_[(0,1)]{}]{}\_[\_[(2,0)]{}]{}\
& & + 4(\_0\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}+ \_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{})\_[\_[(1,1)]{}]{}\
& & . -(\_3+\_2-\_1+\_9) \^2\_[\_[(0,1)]{}]{}\^2\_[\_[(1,0)]{}]{}-(\_3+\_2- \_0-\_9)\^2\_[\_[(1,0)]{}]{}\^2\_[\_[(0,1)]{}]{}.\
& & +2(\_3+\_2-\_0\_1-\_9) \_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}\
& & +2(\_3+\_2-\_0\_1+\_9) \_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}\
& & .-2(\_3-\_0\_1) \_[\_[(1,0)]{}]{}\^2\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{} -2(\_2-\_0\_1) \_[\_[(0,1)]{}]{}\^2\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{} }T\
& & + {\_[\_[(1,3)]{}]{} +3\_[\_[(0,1)]{}]{}\_[\_[(1,2)]{}]{}+\_[\_[(1,0)]{}]{}\_[\_[(0,3)]{}]{} +3\_[10]{}\_[\_[(0,2)]{}]{}\_[\_[(1,1)]{}]{} .\
& & +3\_[11]{}\_[\_[(1,1)]{}]{}\_[\_[(0,2)]{}]{} + 3\^2\_[\_[(0,1)]{}]{}\_[\_[(1,1)]{}]{} +3(\_0\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}+ \_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{})\_[\_[(0,2)]{}]{}\
& & +(\_0-\_3-\_[12]{})\_[\_[(1,0)]{}]{} \^3\_[\_[(0,1)]{}]{}+3\_[12]{}\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{} \^2\_[\_[(0,1)]{}]{}\
& & . +3(\_3-\_[12]{})\^2\_[\_[(0,1)]{}]{} \_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}+(\_1-2\_3+\_[12]{}) \^3\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}}T\
& & + {\_[\_[(0,4)]{}]{}+4\_[\_[(0,1)]{}]{} \_[\_[(0,3)]{}]{}+3\^2\_[\_[(0,2)]{}]{}+6\^2\_[\_[(0,1)]{}]{} \_[\_[(0,2)]{}]{}+\^4\_[\_[(0,1)]{}]{}}T\
& & + O\^5(,). \[Taylorexpression\] In this expression we can see the way in which the parameters $\{\epsilon_A\}$ \[real constants subject to the conditions (\[constraints\])\] describe the arbitrariness we have in the [*reconstruction*]{} of the Taylor expansion of a two–parameter family of diffeomorphisms in terms of Lie derivative operators.
To finish this section we will show how to recover the one-parameter case from the two-parameter case. The case when one of $\lambda$ or $\Omega$ vanishes is trivial and it can be recovered from the above expressions just setting either $\lambda=0$ or $\Omega=0$. Let then consider the only other case, which arises when the two parameters $\lambda$ and $\Omega$ are not longer independent, e.g. $\Omega=\Omega(\lambda)$. Then the specific way of recovering the single parameter ($\lambda$) case will depend on the specific function $\Omega(\lambda)$. Here we illustrate the simplest case of a linear relation $\Omega = a \lambda$ ($a\neq 0$). To arrive to the one-parameter expansion (see [@BMMS; @SB; @BS]), we need to study the consequences of the dependence between the two parameters. We can do that by looking at the definitions of the operators $\cL_{(p,q)}$ in equations (\[operator10\]-\[operator04\]), and at their association with Lie derivative operators, described by equation (\[lieder\]). The result is a set of relations between the vector fields $\xi_{(p,q)}$ that can be summarized in the following relation: \_[(p,q)]{} = \_[(p+q,0)]{} . Then, rescaling the vector fields $\xi_{(p,0)}$ in the following way: $\xi_{(p,0)} \rightarrow (p+1)^{-1}\xi_{(p,0)}$, we arrive at exactly the same expansions as in the one-parameter case [@BMMS; @SB; @BS]. In the case in which both parameters $\lambda$ and $\Omega$ have a specific physical meaning, it will be the physics that will impose the functional dependence $\Omega(\lambda)$ in particular sub-cases.
Gauges in perturbation theory and the two–parameter case {#gauges}
========================================================
Let us consider for the moment a spacetime $\{g^{\rm b},\cM_{0}\}$ which we call the background, and a physical spacetime $\{g,\cM\}$ which we attempt to describe as a perturbation of $\{g^{\rm b},\cM_{0}\}$.[^4] In relativistic perturbation theory we are used to write expressions of the form g\_(x) = g\^[b]{}\_(x) +g\_(x), \[metricpert\] relating a perturbed tensor field such as the metric with the background value of the same field and with the perturbation. In doing this, we are implicitly assigning a correspondence between points of the perturbed and the background spacetimes [@bi:schutz]. Indeed through (\[metricpert\]), which is a relation between the images of the fields in ${\rm I\! R}^{m}$ rather than between the fields themselves on the respective manifolds $\cM$ and $\cM_{0}$, we are saying that there is a unique point $x$ in ${\rm I\! R}^{m}$ that is at the same time the image of [*two*]{} points: one (say $q$) in $\cM_{0}$ and one ($o$) in $\cM$. This correspondence is what is usually called a gauge choice in the context of perturbation theory. Clearly, this is more than the usual assignment of coordinate labels to points of a single spacetime [@sachs]. Furthermore, the correspondence established by relations such as (\[metricpert\]) is not unique, but rather (\[metricpert\]) typically defines a set of gauges, unless certain specific restrictions are satisfied by the fields involved (e.g., some metric components vanish). Leaving this problem aside, i.e. supposing that the gauge has been somehow completely fixed, let us look more precisely at the implications of (\[metricpert\]), adopting the geometrical description illustrated in Figure \[fig1\].
If we call ${\mbox{\boldmath $X$}}$ the chart on $\cM_0$ and ${\mbox{\boldmath $Z$}}$ the chart on $\cM$ we see that if we choose $z(o)=x(q)$, i.e. the correspondence between points of $\{g^{\rm b},\cM_{0}\}$ and $\{g,\cM\}$ implicit in (\[metricpert\]), we are implicitly defining a map $\varphi$ between $\cM_0$ and $\cM$, such that $\varphi={\mbox{\boldmath $Z$}}^{-1}\circ{\mbox{\boldmath $X$}}$. Thus from the geometrical point of view a gauge choice is an identification of points of $\cM_0$ and $\cM$. Therefore, we could as well start directly assigning the point identification map $\varphi$ first, calling $\varphi$ itself a gauge, and defining coordinates adapted to it later. This turns out to be a simpler way of proceeding in order to derive the gauge transformations in the following section.
Let us follow this idea in the specific case of two parameters introducing, in the spirit of [@stewart; @BMMS; @SB; @BS], an $(m+2)$–dimensional manifold $\cN$, foliated by $m$–dimensional submanifolds diffeomorphic to $\cM$, so that $\cN=\cM\times\RR^2$. We shall label each copy of $\cM$ by the corresponding value of the parameters $\lambda,\Omega$. The manifold $\cN$ has a natural differentiable structure which is the direct product of those of $\cM$ and $\RR^2$. We can then choose charts on $\cN$ in which $x^{\mu}$ ($\mu=0,1,\dots m-1$) are coordinates of each leaf $\cM_{\lambda,\Omega}$ and $x^m=\lambda,\,x^{m+1}=\Omega$.
In this construction we are assuming that the perturbed spacetimes have the same manifold as the background one. With this we are not allowing the possibility of addressing questions like how can perturbations affect the differential structure of the background spacetime. These issues would require the use of a much more complicated mathematical apparatus, in particular the notion of limits of spacetimes introduced by Geroch [@GEROCH] in order to define the background manifold as a limit $\lambda,\Omega\rightarrow 0$ of a family of [*perturbed*]{} manifolds $\cM_{\lambda,\Omega}$. Instead, we consider perturbations as fields [*living*]{} on the background (as in [@stewart; @BMMS; @SB; @BS]), a standard approach in which these issues do not appear. More sofisticated structures of the extended manifold $\cN$ have been considered in [@MAIA], in an attempt to give the background metric the status of a gauge invariant quantity.
Coming back to our construction, if a tensor $T_{\lambda,\Omega}$ is given on each $\cM_{\lambda,\Omega}$, we have that a tensor field $T$ is automatically defined on $\cN$ by the relation $T(p,\lambda,\Omega):= T_{\lambda,\Omega}(p)$, with $p\in\cM_{\lambda,\Omega}$.[^5] In particular, on each $\cM_{\lambda,\Omega}$ one has a metric $g_{\lambda,\Omega}$ and a set of matter fields $\tau_{\lambda,\Omega}$, satisfying the set of field equations \[g\_[,]{},\_[,]{}\]=0. Correspondingly, the fields $g$ and $\tau$ are defined on $\cN$.
We now want to define the perturbation in any tensor $T$, therefore we must find a way to compare $T_{\lambda,\Omega}$ with $T_0$: this requires a prescription for identifying points of $\cM_{\lambda,\Omega}$ with those of $\cM_0$. This is easily accomplished by assigning a diffeomorphism $\varphi_{\lambda,\Omega}:\cN\rightarrow\cN$ such that $\left.\varphi_{\lambda,\Omega}\right|_{\cM_0}:\cM_0
\rightarrow\cM_{\lambda,\Omega}$. Clearly, $\varphi_{\lambda,\Omega}$ can be regarded as the member of a two–parameter group of diffeomorphisms $\varphi$ on $\cN$, corresponding to the values of $\lambda,\Omega$ of the group parameter. Therefore, we could equally well give the vector fields $^{\varphi}\eta,\,^{\varphi}\zeta$ that generate $\varphi$. In the chart introduced above, $^{\varphi}\eta^m=1$, $^{\varphi}\eta^{m+1}=0$, $^{\varphi}\zeta^m=0$, $^{\varphi}\zeta^{m+1}=1$ but, except for these conditions, $^{\varphi}\eta,\,^{\varphi}\zeta$ remain arbitrary. For convenience, we shall also refer to such a pair of vector fields as a gauge. It is always possible to take the chart above defined such that $\,^{\varphi}\eta^{\mu}=\,^{\varphi}
\zeta^{\mu}=0$. So, in this chart, point of different submanifolds $\cM_{\lambda,\Omega}$ connected by the diffeomorphism $\varphi$ have the same $\cM$–coordinates $x^0,\dots,x^{m-1}$, and differ only by the value of the coordinates $\lambda,\,\Omega$. We call such a chart “adapted to the gauge $\varphi$”: this is what is always used in practice.
The perturbation in $T$ can now be defined simply as \^\_0 T\^\_[,]{} := .\^\*\_[,]{}T|\_[\_0]{} -T\_0. \[defperturbation\] The first term on the right–hand side of (\[defperturbation\]) can be Taylor–expanded using (\[Taylorgroup\]) to get \^\_0 T\^\_[,]{}=\_[k,k’=0]{}\^ \^[(k,k’)]{}\_T-T\_0, \[perturbationexpansion\] where \^[(k,k’)]{}\_T:=\_[=0,=0,\_0]{} = .\^k\_[\^]{} \^[k’]{}\_[\^]{}T|\_[\_0]{}, \[defperturbationk\] which defines the perturbation of order $(k,k')$ of $T$ (notice that $\delta^{(0,0)}_\varphi T=T_0$). It is worth noticing $\Delta_0 T^\varphi_{\lambda,\Omega}$ and $\delta^{(k,k')}_\varphi T$ are defined on $\cM_0$; this formalizes the statement one commonly finds in the literature that “perturbations are fields living in the background”. It is important to appreciate that the parameters $\lambda,\,\Omega$ labelling the various spacetime models also serve to perform the expansion (\[perturbationexpansion\]), and therefore determine what one means by “perturbations of order $(k,k')$”.
Gauge invariance and gauge transformations {#gaugeiandt}
==========================================
Let us now suppose that two gauges $\varphi$ and $\psi$, described by pairs of vector fields $(^{\varphi}\eta,\,^{\varphi}\zeta)$ and $(^{\psi}\eta,\,^{\psi}\zeta)$ respectively, are defined on $\cN$, such that in the chart discussed above[^6]
[lclcl]{} \^\^[m]{} & = & \^\^m & = & 1 ,\
\^\^[m]{}& = & \^\^[m]{} & = & 0 ,\
\^\^[m+1]{} & = & \^\^[m+1]{} & = & 0 ,\
\^\^[m+1]{} & = & \^\^[m+1]{} & = & 1 . \[1010\]
Correspondingly, the integral curves of $(^{\varphi}\eta,\,^{\varphi}\zeta)$ and $(^{\psi}\eta,\,^{\psi}\zeta)$ define two two–parameter groups of diffeomorphisms $\varphi$ and $\psi$ on $\cN$, that connect any two leaves of the foliation. Thus, $(^{\varphi}\eta,\,^{\varphi}\zeta)$ and $(^{\psi}\eta,\,^{\psi}\zeta)$ are everywhere transverse to $\cM_{\lambda,\Omega}$ and points lying on the same integral surface of either of the two are to be regarded [*as the same point*]{} within the respective gauge: $\varphi$ and $\psi$ are both point identification maps, i.e. two different gauge choices.
The pairs of vector fields $(^{\varphi}\eta,\,^{\varphi}\zeta)$ and $(^{\psi}\eta,\,^{\psi}\zeta)$ can both be used to pull back a generic tensor $T$ and therefore to construct two other tensor fields $\varphi^*_{\lambda,\Omega}T$ and $\psi^*_{\lambda,\Omega}T$, for any given value of $(\lambda,\Omega)$. In particular, on $\cM_0$ we now have three tensor fields, i.e. $T_0$ and T\^\_[,]{}:= .\^\*\_[,]{} T|\_[\_0]{}, T\^\_[,]{}:= .\^\*\_[,]{} T|\_[\_0]{}. Since $\varphi$ and $\psi$ represent gauge choices for mapping a perturbed manifold $\cM_{\lambda,\Omega}$ into the unperturbed one $\cM_0$, $T^\varphi_{\lambda,\Omega}$ and $T^\psi_{\lambda,\Omega}$ are the representations, in $\cM_0$, of the perturbed tensor according to the two gauges. We can write, using (\[defperturbation\])–(\[defperturbationk\]) and the expansion (\[Taylorgroup\]), T\^\_[,]{} & = & \_[k=0]{}\^ \^[(k,k’)]{}\_T=\_[k,k’=0]{}\^ \^k\_[\^]{}\^[k’]{}\_[\^]{}T=T\_0+ \_0\^ T\_[,]{},\[defexpX\]\
T\^\_[,]{} & = & \_[k=0]{}\^ \^[(k,k’)]{}\_T = \_[k,k’=0]{}\^ \^k\_[\^]{}\^[k’]{}\_[\^]{}T=T\_0+\_0\^ T\_[,]{}\[defexpY\] where $\delta^{(k,k')}_\varphi T$, $\delta^{(k,k')}_\psi T$ are the perturbations (\[defperturbationk\]) in the gauges $\varphi$ and $\psi$ respectively, i.e. \^[(k,k’)]{}\_T & = & . \^k\_[\^]{} \^[k’]{}\_[\^]{}T|\_[\_0]{} ,\
\^[(k,k’)]{}\_T & = & . \^k\_[\^]{} \^[k’]{}\_[\^]{}T|\_[\_0]{}.
Gauge invariance {#gaugeinvariance}
----------------
If $T^\varphi_{\lambda,\Omega}=T^\psi_{\lambda,\Omega}$, for any pair of gauges $\varphi$ and $\psi$, we say that $T$ is [*totally gauge invariant*]{}. This is a very strong condition, because then (\[defexpX\]) and (\[defexpY\]) imply that $\delta^{(k,k')}_\varphi T=
\delta^{(k,k')}_\psi T$, for all gauges $\varphi$ and $\psi$ and for any $(k,k')$. In any practical case, however, one is interested in perturbations up to a fixed order. It is thus convenient to weaken the definition above, saying that $T$ is [*gauge invariant up to order $(n,n')$*]{} iff for any two gauges $\varphi$ and $\psi$ \^[(k,k’)]{}\_T =\^[(k,k’)]{}\_T (k,k’) kn,k’n’.\[charatGI\]
We have that a tensor field $T$ is gauge invariant to order $(n,n')$ iff in a given gauge $\varphi$ we have that $\cL_{\xi}\delta^{(k,k')}_\varphi T=0$, for any vector field $\xi$ defined on $\cM$ and for any $(k,k')<(n,n')$.
To prove this statement, let us first show that it is true for $(n,n')=(1,0)$. In fact, if $\delta^{(1,0)}_\varphi T = \delta^{(1,0)}_\psi T$ for two arbitrary gauges $\varphi,\psi$, we have $\cL_{\,^{\varphi}\!\eta-^{\psi}\!\eta}T|_{\cM_0}
=0$. But since $\varphi$ and $\psi$ are arbitrary gauges, it follows that $\,^{\varphi}\!\eta-\,^{\psi}\!\eta$ is an arbitrary field $\xi$, and $\xi^m=\xi^{m+1}=0$ because $\,^{\varphi}\!\eta^m=\,{}^{\psi}\!\eta^m=1,\,
\,^{\varphi}\!\eta^{m+1}=\,^{\psi}\!\eta^{m+1}=0$, so $\xi$ is tangent to $\cM$. In the same way one proves the statement for $(n,n')=(0,1)$. Now let us suppose that the statement is true for some $(n,n')$. Then, if one also has $\delta^{(n+1,n')}_\varphi T|_{\cM_0}=\delta^{(n+1,n')}_\psi T|_{\cM_0}$, it follows that $\cL_{\,^{\varphi}\!\eta-^{\psi}\!\eta}\delta^{(n,n')}_\varphi T=0$, while if $\delta^{(n,n'+1)}_\varphi T|_{\cM_0}=\delta^{(n,n'+1)}_\psi T|_{\cM_0}$, it follows that $\cL_{\,^{\varphi}\!\zeta-^{\psi}\!\zeta}\delta^{(n,n')}_\varphi T=0$, and we establish the result by induction over $(n,n')$.
As a consequence, $T$ is gauge invariant to order $(n,n')$ iff $T_0$ and all its perturbations of order lower than $(n,n')$ are, in any gauge, either vanishing or constant scalars, or a combination of Kronecker deltas with constant coefficients. Thus, this generalizes to an arbitrary order $(n,n')$ and to the two–parameter case the results of [@BMMS; @sachs; @stewart]. Further, it then follows that $T$ is totally gauge invariant iff it is a combination of Kronecker deltas with coefficients depending only on $\lambda,\Omega$.
Gauge transformations
---------------------
If a tensor $T$ is not gauge invariant, it is important to know how its representation on $\cM_0$ changes under a gauge transformation. To this purpose it is natural to introduce, for each value of $(\lambda,\Omega)\in\RR^2$, the diffeomorphism $\Phi_{\lambda,\Omega}
:\cM_0\rightarrow\cM_0$ defined by \_[,]{}:=\_[-,-]{}\_[, ]{}. Given that from the geometrical point of view adopted here $\varphi$ and $\psi$ are two gauges, $\Phi$ represents the gauge transformation.
The action of $\Phi_{\lambda,\Omega}$ is illustrated in Fig. \[fig:GT\]. We must stress that $\Phi:\cM_0\times\RR^2\rightarrow\cM_0$ thus defined, [*is not*]{} a two–parameter group of diffeomorphisms in $\cM_0$. In fact, $\Phi_{\lambda_1,\Omega_1}\circ\Phi_{\lambda_2,\Omega_2}\neq
\Phi_{\lambda_1+\lambda_2,
\Omega_1+\Omega_2}$, essentially because the fields $(^{\varphi}\eta,\,^{\varphi}\zeta)$ and $(^{\psi}\eta,\,^{\psi}\zeta)$ have, in general, a non–vanishing commutator. However, it can be Taylor expanded, using the results of section (\[familydiffeo\]).
The tensor fields $T^\varphi_{\lambda,\Omega}$ and $T^\psi_{\lambda,\Omega}$, defined on $\cM_0$ by the gauges $\varphi$ and $\psi$, are connected by the linear map $\Phi_{\lambda,\Omega}^*$: T\^\_[,]{} & = & . \^\*\_[,]{}T|\_[\_0]{}= . (\^\*\_[,]{}\^\*\_[-,-]{} \^\*\_[,]{}T)|\_[\_0]{}\
& = & . \^\*\_[,]{}(\^\*\_[,]{}T )|\_[\_0]{} = \^\*\_[,]{}T\^\_[,]{}. Thus, the gauge transformation to an arbitrary order $(n,n')$ is given by the Taylor expansion of the pull–back $\Phi^*_{\lambda,\Omega}T$, whose terms are explicitly given in section \[familydiffeo\]. Up to fourth order, we have explicitly from (\[Taylorexpression\]) T\^\_[,]{} & & = T\^\_[,]{} +\_[\_[(1,0)]{}]{}T\^\_[,]{}+ \_[\_[(0,1)]{}]{}T\^\_[,]{}\
& & + {\_[\_[(2,0)]{}]{} +\^2\_[\_[(1,0)]{}]{}}T\^\_[,]{} +{\_[\_[(0,2)]{}]{}+\^2\_[\_[(0,1)]{}]{}} T\^\_[,]{}\
& & + {\_[\_[(1,1)]{}]{} +\_0\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}+ \_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}} T\^\_[,]{}\
& & + {\_[\_[(3,0)]{}]{}+3\_[\_[(1,0)]{}]{} \_[\_[(2,0)]{}]{}+\^3\_[\_[(1,0)]{}]{}} T\^\_[,]{}\
& & + {\_[\_[(2,1)]{}]{} +2\_[\_[(1,0)]{}]{}\_[\_[(1,1)]{}]{}+\_[\_[(0,1)]{}]{}\_[\_[(2,0)]{}]{} +2\_2\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{} \_[\_[(1,0)]{}]{}.\
& & .+(\_1-\_2)\_[\_[(0,1)]{}]{}\^2\_[\_[(1,0)]{}]{} +(\_0-\_2)\^2\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}} T\^\_[,]{}\
& & + {\_[\_[(1,2)]{}]{}+2\_[\_[(0,1)]{}]{}\_[\_[(1,1)]{}]{} +\_[\_[(1,0)]{}]{}\_[\_[(0,2)]{}]{}+2\_3\_[\_[(0,1)]{}]{} \_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}.\
& & .+(\_0-\_3)\_[\_[(1,0)]{}]{}\^2\_[\_[(0,1)]{}]{} +(\_1-\_3)\^2\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}} T\^\_[,]{}\
& & + {\_[\_[(0,3)]{}]{}+3\_[\_[(0,1)]{}]{} \_[\_[(0,2)]{}]{}+\^3\_[\_[(0,1)]{}]{}} T\^\_[,]{}\
& & + {\_[\_[(4,0)]{}]{} +4\_[\_[(1,0)]{}]{}\_[\_[(3,0)]{}]{} +3\^2\_[\_[(2,0)]{}]{}+6\^2\_[\_[(1,0)]{}]{}\_[\_[(2,0)]{}]{} +\^4\_[\_[(1,0)]{}]{}}T\^\_[,]{}\
& & + {\_[\_[(3,1)]{}]{} +3\_[\_[(1,0)]{}]{}\_[\_[(2,1)]{}]{} +\_[\_[(0,1)]{}]{}\_[\_[(3,0)]{}]{}+3\_4\_[\_[(2,0)]{}]{} \_[\_[(1,1)]{}]{} .\
& & +3\_5\_[\_[(1,1)]{}]{} \_[\_[(2,0)]{}]{} + 3\^2\_[\_[(1,0)]{}]{}\_[\_[(1,1)]{}]{} +3(\_0\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}+ \_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{})\_[\_[(2,0)]{}]{}\
& & +(\_1-\_2-\_6)\_[\_[(0,1)]{}]{} \^3\_[\_[(1,0)]{}]{}+3\_6\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{} \^2\_[\_[(1,0)]{}]{}\
& & . +3(\_2-\_6)\^2\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}+(\_0-2\_2+\_6) \^3\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}}T\^\_[,]{}\
& & + {\_[\_[(2,2)]{}]{}+2\_[\_[(1,0)]{}]{}\_[\_[(1,2)]{}]{} +2\_[\_[(0,1)]{}]{}\_[\_[(2,1)]{}]{}+ 2\^2\_[\_[(1,1)]{}]{} .\
& & + \_7\_[\_[(2,0)]{}]{}\_[\_[(0,2)]{}]{} +\_8\_[\_[(0,2)]{}]{}\_[\_[(2,0)]{}]{} + \^2\_[\_[(1,0)]{}]{}\_[\_[(0,2)]{}]{} +\^2\_[\_[(0,1)]{}]{}\_[\_[(2,0)]{}]{}\
& & + 4(\_0\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}+ \_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{})\_[\_[(1,1)]{}]{}\
& & . -(\_3+\_2-\_1+\_9) \^2\_[\_[(0,1)]{}]{}\^2\_[\_[(1,0)]{}]{} -(\_3+\_2-\_0-\_9) \^2\_[\_[(1,0)]{}]{}\^2\_[\_[(0,1)]{}]{}.\
& & +2(\_3+\_2-\_0\_1-\_9) \_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}\
& & +2(\_3+\_2-\_0\_1+\_9) \_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}\
& & .-2(\_3-\_0\_1) \_[\_[(1,0)]{}]{}\^2\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{} -2(\_2-\_0\_1)\_[\_[(0,1)]{}]{}\^2\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{} }T\^\_[,]{}\
& & + {\_[\_[(1,3)]{}]{} +3\_[\_[(0,1)]{}]{}\_[\_[(1,2)]{}]{}+\_[\_[(1,0)]{}]{}\_[\_[(0,3)]{}]{} +3\_[10]{}\_[\_[(0,2)]{}]{}\_[\_[(1,1)]{}]{} .\
& & +3\_[11]{}\_[\_[(1,1)]{}]{}\_[\_[(0,2)]{}]{} +3\^2\_[\_[(0,1)]{}]{}\_[\_[(1,1)]{}]{} +3(\_0\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}+ \_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{})\_[\_[(0,2)]{}]{}\
& & +(\_0-\_3-\_[12]{})\_[\_[(1,0)]{}]{} \^3\_[\_[(0,1)]{}]{}+3\_[12]{}\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{} \^2\_[\_[(0,1)]{}]{}\
& & . +3(\_3-\_[12]{})\^2\_[\_[(0,1)]{}]{} \_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}+(\_1-2\_3 +\_[12]{})\^3\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{} }T\^\_[,]{}\
& & + {\_[\_[(0,4)]{}]{} +4\_[\_[(0,1)]{}]{}\_[\_[(0,3)]{}]{}+3\^2\_[\_[(0,2)]{}]{} +6\^2\_[\_[(0,1)]{}]{}\_[\_[(0,2)]{}]{}+\^4\_[\_[(0,1)]{}]{}} T\^\_[,]{}\
& & + O\^5(,) , \[totalgaugetransf\] where the $\xi_{(p,q)}$ are now the generators of the gauge transformation $\Phi_{\lambda,\Omega}$.
We can now relate the perturbations in the two gauges. To order $(n,n')$ with $n+n'\leq 4$, these relations can be derived by substituting (\[defexpX\]), (\[defexpY\]) in (\[totalgaugetransf\]): \^[(1,0)]{}\_T-\^[(1,0)]{}\_T = \_[\_[(1,0)]{}]{}T\_0, \[gi10\] \^[(0,1)]{}\_T-\^[(0,1)]{}\_T = \_[\_[(0,1)]{}]{}T\_0, \[gi01\] \^[(2,0)]{}\_T-\^[(2,0)]{}\_T = 2\_[\_[(1,0)]{}]{} \^[(1,0)]{}\_T +{\_[\_[(2,0)]{}]{}+\_[\_[(1,0)]{}]{}\^2}T\_0, \[gi20\] \^[(1,1)]{}\_T-\^[(1,1)]{}\_T = \_[\_[(1,0)]{}]{} \^[(0,1)]{}\_T + \_[\_[(0,1)]{}]{}\_\^[(1,0)]{}T\
+ {\_[\_[(1,1)]{}]{}+\_0\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{}+\_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{} }T\_0 , \[gi11\] \^[(0,2)]{}\_T-\^[(0,2)]{}\_T = 2\_[\_[(0,1)]{}]{} \^[(0,1)]{}\_T +{\_[\_[(0,2)]{}]{}+\_[\_[(0,1)]{}]{}\^2}T\_0, \[gi02\] \^[(3,0)]{}\_T-\^[(3,0)]{}\_T = 3\_[\_[(1,0)]{}]{} \_\^[(2,0)]{}T + 3{\_[\_[(2,0)]{}]{}+\^2\_[\_[(1,0)]{}]{}}\_\^[(1,0)]{}T\
+ {\_[\_[(3,0)]{}]{}+3\_[\_[(1,0)]{}]{}\_[\_[(2,0)]{}]{} +\^3\_[\_[(1,0)]{}]{}}T\_0 , \[gi30\] \^[(2,1)]{}\_T-\^[(2,1)]{}\_T = 2\_[\_[(1,0)]{}]{} \_\^[(1,1)]{}T + \_[\_[(0,1)]{}]{}\_\^[(2,0)]{}T + {\_[\_[(2,0)]{}]{} + \^2\_[\_[(1,0)]{}]{}}\_\^[(0,1)]{}T\
+ 2{\_[\_[(1,1)]{}]{}+\_0\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{}+\_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}} \_\^[(1,0)]{}T\
+ {\_[\_[(2,1)]{}]{}+2\_[\_[(1,0)]{}]{}\_[\_[(1,1)]{}]{} +\_[\_[(0,1)]{}]{}\_[\_[(2,0)]{}]{}+2\_2\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{} \_[\_[(1,0)]{}]{}.\
.+(\_1-\_2)\_[\_[(0,1)]{}]{}\^2\_[\_[(1,0)]{}]{} +(\_0-\_2)\^2\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}}T\_0, \[gi21\] \^[(1,2)]{}\_T-\^[(1,2)]{}\_T = 2\_[\_[(0,1)]{}]{} \_\^[(1,1)]{}T + \_[\_[(1,0)]{}]{}\_\^[(0,2)]{}T+ {\_[\_[(0,2)]{}]{}+\^2\_[\_[(0,1)]{}]{}}\_\^[(1,0)]{}T\
+ 2{\_[\_[(1,1)]{}]{}+\_0\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{}+\_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}} \_\^[(0,1)]{}T\
+ {\_[\_[(1,2)]{}]{}+2\_[\_[(0,1)]{}]{}\_[\_[(1,1)]{}]{} +\_[\_[(1,0)]{}]{}\_[\_[(0,2)]{}]{}+2\_3\_[\_[(0,1)]{}]{} \_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}.\
.+(\_0-\_3)\_[\_[(1,0)]{}]{}\^2\_[\_[(0,1)]{}]{} +(\_1-\_3)\^2\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}}T\_0 , \[gi12\] \^[(0,3)]{}\_T-\^[(0,3)]{}\_T = 3\_[\_[(0,1)]{}]{} \_\^[(0,2)]{}T + 3{\_[\_[(0,2)]{}]{}+\^2\_[\_[(0,1)]{}]{}}\_\^[(0,1)]{}T\
+ {\_[\_[(0,3)]{}]{}+3\_[\_[(0,1)]{}]{} \_[\_[(0,2)]{}]{}+\^3\_[\_[(0,1)]{}]{}}T\_0, \[gi03\] \^[(4,0)]{}\_T-\^[(4,0)]{}\_T = 4\_[\_[(1,0)]{}]{} \_\^[(3,0)]{}T + 6{\_[\_[(2,0)]{}]{}+\^2\_[\_[(1,0)]{}]{}} \_\^[(2,0)]{}T\
+ 4{\_[\_[(3,0)]{}]{}+3\_[\_[(1,0)]{}]{}\_[\_[(2,0)]{}]{} +\^3\_[\_[(1,0)]{}]{}}\_\^[(1,0)]{}T\
+ {\_[\_[(4,0)]{}]{}+4\_[\_[(1,0)]{}]{}\_[\_[(3,0)]{}]{} +3\^2\_[\_[(2,0)]{}]{}+6\^2\_[\_[(1,0)]{}]{}\_[\_[(2,0)]{}]{} +\^4\_[\_[(1,0)]{}]{}}T\_0 , \[gi40\] \^[(3,1)]{}\_T-\^[(3,1)]{}\_T = 3\_[\_[(1,0)]{}]{} \_\^[(2,1)]{}T +\_[\_[(0,1)]{}]{}\_\^[(3,0)]{}T + 3{\_[\_[(2,0)]{}]{} +\^2\_[\_[(1,0)]{}]{}}\_\^[(1,1)]{}T\
+ 3{\_[\_[(1,1)]{}]{}+\_0\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{}+\_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}} \_\^[(2,0)]{}T\
+ {\_[\_[(3,0)]{}]{}+3\_[\_[(1,0)]{}]{} \_[\_[(2,0)]{}]{}+\^3\_[\_[(1,0)]{}]{}}\_\^[(0,1)]{}T\
+ 3{\_[\_[(2,1)]{}]{}+2\_[\_[(1,0)]{}]{}\_[\_[(1,1)]{}]{} +\_[\_[(0,1)]{}]{}\_[\_[(2,0)]{}]{}+2\_2\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}.\
.+(\_1-\_2)\_[\_[(0,1)]{}]{}\^2\_[\_[(1,0)]{}]{} +(\_0-\_2)\^2\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}} \_\^[(1,0)]{}T\
+ {\_[\_[(3,1)]{}]{}+ 3\_[\_[(1,0)]{}]{} \_[\_[(2,1)]{}]{}+\_[\_[(0,1)]{}]{}\_[\_[(3,0)]{}]{}+3\_4 \_[\_[(2,0)]{}]{}\_[\_[(1,1)]{}]{} +3\_5\_[\_[(1,1)]{}]{} \_[\_[(2,0)]{}]{}.\
+ 3\^2\_[\_[(1,0)]{}]{}\_[\_[(1,1)]{}]{} +3(\_0\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}+ \_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{})\_[\_[(2,0)]{}]{}\
+ (\_1-\_2-\_6)\_[\_[(0,1)]{}]{} \^3\_[\_[(1,0)]{}]{} + 3\_6\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{} \^2\_[\_[(1,0)]{}]{}\
. + 3(\_2-\_6)\^2\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}+(\_0-2\_2+\_6) \^3\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}}T\_0 , \[gi31\] \^[(2,2)]{}\_T-\^[(2,2)]{}\_T = 2\_[\_[(1,0)]{}]{} \_\^[(1,2)]{}T +2\_[\_[(0,1)]{}]{}\_\^[(2,1)]{}T\
+ {\_[\_[(2,0)]{}]{}+\^2\_[\_[(1,0)]{}]{}} \_\^[(0,2)]{}T + {\_[\_[(0,2)]{}]{}+\^2\_[\_[(0,1)]{}]{}} \_\^[(2,0)]{}T\
+ 4{\_[\_[(1,1)]{}]{}+\_0\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{}+\_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}} \_\^[(1,1)]{}T\
+ 2{\_[\_[(2,1)]{}]{}+2\_[\_[(1,0)]{}]{}\_[\_[(1,1)]{}]{} +\_[\_[(0,1)]{}]{}\_[\_[(2,0)]{}]{}+2\_2\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}.\
.+(\_1-\_2)\_[\_[(0,1)]{}]{}\^2\_[\_[(1,0)]{}]{} +(\_0-\_2)\^2\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}} \_\^[(0,1)]{}T\
+ 2{\_[\_[(1,2)]{}]{}+2\_[\_[(0,1)]{}]{}\_[\_[(1,1)]{}]{} +\_[\_[(1,0)]{}]{}\_[\_[(0,2)]{}]{}+2\_3\_[\_[(0,1)]{}]{} \_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}.\
.+(\_0-\_3)\_[\_[(1,0)]{}]{}\^2\_[\_[(0,1)]{}]{} +(\_1-\_3)\^2\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}} \_\^[(1,0)]{}T\
+ {\_[\_[(2,2)]{}]{}+2\_[\_[(1,0)]{}]{}\_[\_[(1,2)]{}]{} +2\_[\_[(0,1)]{}]{}\_[\_[(2,1)]{}]{}+2\^2\_[\_[(1,1)]{}]{} .\
+\_7\_[\_[(2,0)]{}]{}\_[\_[(0,2)]{}]{} +\_8\_[\_[(0,2)]{}]{}\_[\_[(2,0)]{}]{} + \^2\_[\_[(1,0)]{}]{} \_[\_[(0,2)]{}]{}+\^2\_[\_[(0,1)]{}]{}\_[\_[(2,0)]{}]{}\
+4(\_0\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}+ \_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{})\_[\_[(1,1)]{}]{}\
. -(\_3+\_2-\_1+\_9) \^2\_[\_[(0,1)]{}]{}\^2\_[\_[(1,0)]{}]{} -(\_3+\_2-\_0-\_9) \^2\_[\_[(1,0)]{}]{}\^2\_[\_[(0,1)]{}]{}.\
+2(\_3+\_2-\_0\_1-\_9) \_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}\
+2(\_3+\_2-\_0\_1+\_9) \_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}\
.-2(\_3-\_0\_1) \_[\_[(1,0)]{}]{}\^2\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{} -2(\_2-\_0\_1)\_[\_[(0,1)]{}]{}\^2\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{} }T\_0, \[gi22\] \^[(1,3)]{}\_T-\^[(1,3)]{}\_T = 3\_[\_[(0,1)]{}]{} \_\^[(1,2)]{}T + \_[\_[(1,0)]{}]{}\_\^[(0,3)]{}T+ 3{\_[\_[(0,2)]{}]{}+\^2\_[\_[(0,1)]{}]{}}\_\^[(1,1)]{}T\
+ 3{\_[\_[(1,1)]{}]{}+\_0\_[\_[(1,0)]{}]{} \_[\_[(0,1)]{}]{}+\_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}} \_\^[(0,2)]{}T\
+ {\_[\_[(0,3)]{}]{}+3\_[\_[(0,1)]{}]{} \_[\_[(0,2)]{}]{}+\^3\_[\_[(0,1)]{}]{}}\_\^[(1,0)]{}T\
+ 3{\_[\_[(1,2)]{}]{}+2\_[\_[(0,1)]{}]{}\_[\_[(1,1)]{}]{} +\_[\_[(1,0)]{}]{}\_[\_[(0,2)]{}]{}+2\_3\_[\_[(0,1)]{}]{} \_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}.\
.+(\_0-\_3)\_[\_[(1,0)]{}]{}\^2\_[\_[(0,1)]{}]{} +(\_1-\_3)\^2\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}} \_\^[(0,1)]{}T\
+ {\_[\_[(1,3)]{}]{}+3\_[\_[(0,1)]{}]{}\_[\_[(1,2)]{}]{} +\_[\_[(1,0)]{}]{}\_[\_[(0,3)]{}]{}+3\_[10]{}\_[\_[(0,2)]{}]{} \_[\_[(1,1)]{}]{}+3\_[11]{}\_[\_[(1,1)]{}]{} \_[\_[(0,2)]{}]{}.\
+ 3\^2\_[\_[(0,1)]{}]{}\_[\_[(1,1)]{}]{} +3(\_0\_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}+ \_1\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{})\_[\_[(0,2)]{}]{}\
+(\_0-\_3-\_[12]{})\_[\_[(1,0)]{}]{} \^3\_[\_[(0,1)]{}]{}+3\_[12]{}\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{} \^2\_[\_[(0,1)]{}]{}\
.+3(\_3-\_[12]{})\^2\_[\_[(0,1)]{}]{} \_[\_[(1,0)]{}]{}\_[\_[(0,1)]{}]{}+(\_1-2\_3+\_[12]{}) \^3\_[\_[(0,1)]{}]{}\_[\_[(1,0)]{}]{}}T\_0, \[gi13\] \^[(0,4)]{}\_T-\^[(0,4)]{}\_T = 4\_[\_[(0,1)]{}]{} \_\^[(0,3)]{}T+6{\_[\_[(0,2)]{}]{}+\^2\_[\_[(0,1)]{}]{}} \_\^[(0,2)]{}T\
+ 4{\_[\_[(0,3)]{}]{}+3\_[\_[(0,1)]{}]{}\_[\_[(0,2)]{}]{} +\^3\_[\_[(0,1)]{}]{}}\_\^[(0,1)]{}T\
+ {\_[\_[(0,4)]{}]{}+4\_[\_[(0,1)]{}]{}\_[\_[(0,3)]{}]{} +3\^2\_[\_[(0,2)]{}]{}+6\^2\_[\_[(0,1)]{}]{}\_[\_[(0,2)]{}]{} +\^4\_[\_[(0,1)]{}]{}}T\_0. \[gi04\] This result is, of course, consistent with the characterization of gauge invariance given in subsection \[gaugeinvariance\]. Equations (\[gi10\]) and (\[gi01\]) imply that $T_{\lambda,\Omega}$ is gauge invariant to the order $(1,0)$ or $(0,1)$ iff $\cL_{\xi}T_0=0$, for any vector field on $\cM_0$. Equation (\[gi20\]) implies that $T_{\lambda,\Omega}$ is gauge invariant to the order $(2,0)$ iff $\cL_{\xi}T_0=0$ [*and*]{} $\cL_{\xi}\delta^{(1,0)}_\varphi T=0$, for any vector field on $\cM_0$, and so on for all the orders.
It is also possible to find the explicit expressions for the generators $\xi_{(p,q)}$ of the gauge transformation $\Phi$ in terms of the gauge vector fields $({}^\varphi\eta,{}^\varphi\zeta)$ and $({}^\psi\eta,{}^\psi\zeta)$. We write here their expressions up to second order: \_[(1,0)]{} &=& \^- \^,\
\_[(0,1)]{} &=& \^- \^,\
\_[(2,0)]{} &=& \[\^,\^\] ,\
\_[(1,1)]{} &=& \_0\[\^,\^\]+ \_1\[\^,\^\] ,\
\_[(0,2)]{} &=& \[\^,\^\].
Coordinate transformations
--------------------------
Up to now, we have built a two–parameter formalism using a geometrical, coordinate–free language. However, in order to carry out explicit calculations in a practical case, one has to introduce systems of local coordinates. In this respect, all our expressions are immediately translated into components simply by using the expression of the components of the Lie derivative of a tensor. Nonetheless, much of the literature on the subject is written using coordinate systems, and gauge transformations are most often represented by the corresponding coordinate transformations. For this reason, we devote this subsection to describe how to establish the translation between the two languages, giving in particular the explicit transformation of coordinates (further details are in [@MMB]).
Let us consider the situation described in Fig. \[fig:GTcoord\]. We have considered two gauge choices, represented by the groups of diffeomorphisms $\varphi$ and $\psi$, under which the point $o$ in the physical manifold $\cM_{\lambda,\Omega}$ corresponds to two different points in the background manifold $\cM_0$, namely $q=\varphi^{-1}_{\lambda,\Omega}(o)$ and $p=\psi_{\lambda,\Omega}^{-1}(o)$. The transformation relating these two gauge choices is described by the two–parameter family of diffeomorphisms $\Phi_{\lambda,\Omega}=
\varphi^{-1}_{\lambda,\Omega}\circ\psi_{\lambda,\Omega}$, so that $\Phi_{\lambda,\Omega}(p)=q$. This gauge transformation maps a tensor field $T$ on $q\in\cM_0$ to the tensor field $(\Phi^*T)(p)=\Phi^*(T(q))$ on $p\in\cM_0$.
Now, let us consider a chart $({\cal U},{\mbox{\boldmath $X$}})$ on an open subset ${\cal U}$ of $\cM_0$. The gauges $\varphi_{\lambda,\Omega}$ and $\psi_{\lambda,\Omega}$ define two maps from $\cM_{\lambda,\Omega}$ to $\RR^m$:
[lccc]{} \^[-1]{}\_[,]{}:& \_[,]{}&&\^m\
& o &|&x(q(o)),\
[lccc]{} \^[-1]{}\_[,]{}:& \_[,]{}&&\^m\
& o &|&x(p(o)).\
Then, we can look at the gauge transformation $\Phi$ in two different ways: from the [*active*]{} point of view or from the [*passive*]{} point of view. In the first case, one considers a diffeomorphism which changes the point on the background $\cM_0$. To these points one associates different values of the coordinates in the chart $({\cal U},{\mbox{\boldmath $X$}})$. So the coordinate change is given by x\^(p) x\^(q) or, defining the pull–back of $x$ as $\tilde{x}^{\mu}(p):=x^{\mu}(\Phi(p))$, x\^(p) [x]{}\^(p). If we instead consider the passive point of view, we need to introduce a new chart $({\cal U}',y)$: := \_[,]{}\^[-1]{}, such that the two sets of coordinates are related by y\^(q)=x\^(p) , so we can say that the gauge transformation does not change the point on $\cM_0$, but it changes the chart from $({\cal U},{\mbox{\boldmath $X$}})$ to $({\cal U}',{\mbox{\boldmath $Y$}})$, i.e. the labels of the point of $\cM_0$. The coordinate transformation is then x\^(q) y\^(q). Now, let us consider the transformation of a vector field $V$. From the active point of view, the components of $V$ in the chart $({\cal U},{\mbox{\boldmath $X$}})$, $V^\mu$, are related with the ones of the transformed vector field ${\tilde V}$, ${\tilde V}^{\mu}$, by \^= (\_\*)\^= (\_\*\_[,]{}\^\*V)\^. From the passive point of view, we can use the properties relating the pull–back and push–forward maps associated with diffeomorphisms: \_\*\_[,]{}\^\*V= \_\*\^[-1]{}\_[\*,]{}V= \_\*V, so we get the well known result that the components of the transformed vector ${\tilde V}$ in the coordinate system ${\mbox{\boldmath $X$}}$ are defined in terms of the components of the vector $V$ in the new coordinate system ${\mbox{\boldmath $Y$}}$: \^(x(p))= (\_\*V(q))\^= V\^(y(q))= . ()|\_[x(q)]{}V\^(x(q)). \[vectorGTcoords\]
In order to write down explicit expressions, we will apply the expansion of the pull–back of $\Phi^*$ \[See equation (\[Taylorexpression\])\] to the coordinate functions $x^{\mu}$. Then, the [*active*]{} coordinate transformation is given by \^(p)&=&x\^(q) = (\^\*x\^)(p)\
& = & x\^(p)+\^\_[(1,0)]{}+\^\_[(0,1)]{}\
& + & (\^\_[(2,0)]{}+\^\_[(1,0)]{} \^\_[(1,0),]{})+(\^\_[(0,2)]{} +\^\_[(0,1)]{}\^\_[(0,1),]{})\
& + & (\^\_[(1,1)]{}+\_0\^\_[(1,0)]{} \^\_[(0,1),]{}+\_1\^\_[(0,1)]{}\^\_[(1,0),]{})\
& + & (\^\_[(3,0)]{}+3\^\_[(1,0)]{} \^\_[(2,0),]{}+\^\_[(1,0)]{}\^\_[(1,0),]{} \^\_[(1,0),]{})\
& + & (\^\_[(2,1)]{}+2\^\_[(1,0)]{} \^\_[(1,1),]{}+\^\_[(0,1)]{}\^\_[(2,0),]{}+2\_2 \^\_[(1,0)]{}\^\_[(0,1),]{}\^\_[(1,0),]{}.\
& + & . (\_1-\_2)\^\_[(0,1)]{} \^\_[(1,0),]{}\^\_[(1,0),]{} +(\_0-\_2)\^\_[(1,0)]{}\^\_[(1,0),]{} \^\_[(0,1),]{})\
& + & (\^\_[(1,2)]{}+2\^\_[(0,1)]{} \^\_[(1,1),]{}+\^\_[(1,0)]{}\^\_[(0,2),]{}+2\_3 \^\_[(0,1)]{}\^\_[(1,0),]{}\^\_[(0,1),]{}.\
& + & . (\_0-\_3)\^\_[(1,0)]{} \^\_[(0,1),]{}\^\_[(0,1),]{}+(\_1-\_3) \^\_[(0,1)]{}\^\_[(0,1),]{}\^\_[(1,0),]{} )\
& + & (\^\_[(0,3)]{}+3\^\_[(0,1)]{} \^\_[(0,2),]{}+\^\_[(0,1)]{}\^\_[(0,1),]{} \^\_[(0,1),]{})\
& + & (\^\_[(4,0)]{}+ 4\^\_[(1,0)]{}\^\_[(3,0),]{} +3\^\_[(2,0)]{}\^\_[(2,0),]{}.\
& + & . 6\^\_[(1,0)]{}\^\_[(1,0),]{} \^\_[(2,0),]{}+\^\_[(1,0)]{}\^\_[(1,0),]{} \^\_[(1,0),]{}\^\_[(1,0),]{}))\
& + & (\^\_[(3,1)]{}+ 3\^\_[(1,0)]{}\^\_[(2,1),]{}+\^\_[(0,1)]{} \^\_[(3,0),]{}+3\_4\^\_[(2,0)]{}\^\_[(1,1),]{} .\
& + & 3\_5\^\_[(1,1)]{}\^\_[(2,0),]{} +3\^\_[(1,0)]{}\^\_[(1,0),]{}\^\_[(1,1),]{} +3(\_0\^\_[(1,0)]{}\^\_[(0,1),]{} +\_1\^\_[(0,1)]{}\^\_[(1,0),]{}) \^\_[(2,0),]{}\
& + & (\_1-\_2-\_6)\^\_[(0,1)]{} \^\_[(1,0),]{}\^\_[(1,0),]{}\^\_[(1,0),]{} +3\_6\^\_[(1,0)]{}\^\_[(0,1),]{} \^\_[(1,0),]{}\^\_[(1,0),]{}\
& + & . 3(\_2-\_6)\^\_[(1,0)]{} \^\_[(1,0),]{}\^\_[(0,1),]{}\^\_[(1,0),]{} +(\_0-2\_2+\_6) \^\_[(1,0)]{} \^\_[(1,0),]{} \^\_[(1,0),]{}\^\_[(0,1),]{})\
& + & (\^\_[(2,2)]{}+ 2\^\_[(1,0)]{}\^\_[(1,2),]{}+2\^\_[(0,1)]{} \^\_[(2,1),]{}+2\^\_[(1,1)]{}\^\_[(1,1),]{}.\
& + & \_7\^\_[(2,0)]{}\^\_[(0,2),]{} +\_8\^\_[(0,2)]{}\^\_[(2,0),]{} +\^\_[(1,0)]{}\^\_[(1,0),]{}\^\_[(0,2),]{} +\^\_[(0,1)]{}\^\_[(0,1),]{}\^\_[(2,0),]{}\
& + & 4(\_0\^\_[(1,0)]{}\^\_[(0,1),]{} +\_1\^\_[(0,1)]{}\^\_[(1,0),]{}) \^\_[(1,1),]{}\
& - & (\_3+\_2-\_1+\_9) \^\_[(0,1)]{}\^\_[(0,1),]{} \^\_[(1,0),]{}\^\_[(1,0),]{}\
& - & (\_3+\_2-\_0-\_9) \^\_[(1,0)]{}\^\_[(1,0),]{} \^\_[(0,1),]{}\^\_[(0,1),]{}\
& + & 2(\_3+\_2-\_0\_1-\_9) \^\_[(1,0)]{}\^\_[(0,1),]{} \^\_[(1,0),]{}\^\_[(0,1),]{}\
& + & 2(\_3+\_2-\_0\_1+\_9) \^\_[(0,1)]{}\^\_[(1,0),]{} \^\_[(0,1),]{}\^\_[(1,0),]{}\
& - & . 2(\_3-\_0\_1) \^\_[(1,0)]{}\^\_[(0,1),]{} \^\_[(0,1),]{}\^\_[(1,0),]{} -2(\_2-\_0\_1) \^\_[(0,1)]{}\^\_[(1,0),]{} \^\_[(1,0),]{}\^\_[(0,1),]{})\
& + & (\^\_[(1,3)]{}+ 3\^\_[(0,1)]{}\^\_[(1,2),]{}+\^\_[(1,0)]{} \^\_[(0,3),]{}+3\_[10]{}\^\_[(0,2)]{} \^\_[(1,1),]{}.\
& + & 3\_[11]{}\^\_[(1,1)]{}\^\_[(0,2),]{} +3\^\_[(0,1)]{}\^\_[(0,1),]{}\^\_[(1,1),]{} +3(\_0\^\_[(1,0)]{}\^\_[(0,1),]{} +\_1\^\_[(0,1)]{}\^\_[(1,0),]{}) \^\_[(0,2),]{}\
& + & (\_0-\_3-\_[12]{})\^\_[(1,0)]{} \^\_[(0,1),]{}\^\_[(0,1),]{}\^\_[(0,1),]{} +3\_[12]{}\^\_[(0,1)]{}\^\_[(1,0),]{} \^\_[(0,1),]{}\^\_[(0,1),]{}\
& + & . 3(\_3-\_[12]{})\^\_[(0,1)]{} \^\_[(0,1),]{}\^\_[(1,0),]{}\^\_[(0,1),]{} +(\_1-2\_3+\_[12]{}) \^\_[(0,1)]{} \^\_[(0,1),]{}\^\_[(0,1),]{}\^\_[(1,0),]{} )\
& + & (\^\_[(0,4)]{}+ 4\^\_[(0,1)]{}\^\_[(0,3),]{} +3\^\_[(0,2)]{}\^\_[(0,2),]{}.\
& + & . 6\^\_[(0,1)]{}\^\_[(0,1),]{} \^\_[(0,2),]{}+\^\_[(0,1)]{}\^\_[(0,1),]{} \^\_[(0,1),]{}\^\_[(0,1),]{}) , \[coordtransf\] where the vector fields $\xi^\mu_{(p,q)}$ and their derivatives are evaluated in $x(p)$. This expression gives the relation between the coordinates, in the chart $({\cal U},{\mbox{\boldmath $X$}})$, of the two points $p$ and $q$ of $\cM_0$.
On the other hand, the [*passive*]{} coordinate transformation is found by inverting (\[coordtransf\]): y\^(q):=x\^(p)=x\^(q)-\^\_[(1,0)]{}(x(p))- \^\_[(0,1)]{}(x(p))+O\^2(,) , and then by expanding $x(p)$ around $x(q)$. We obtain in this way an expression of the form y\^(q)=x\^(q)-\^\_[(1,0)]{}(x(q))- \^\_[(0,1)]{}(x(q))+O\^2(,) , which gives the relation between the coordinates of any arbitrary point $q\in\cM_0$ in the two charts $({\cal U},{\mbox{\boldmath $X$}})$ and $({\cal U}',{\mbox{\boldmath $Y$}})$. Such a relation is needed to find the transformation of the components of a tensor field, by using (\[vectorGTcoords\]), as it is usually done in textbooks for first order gauge transformations [@dinverno; @weinberg]. However, in order to determine these transformation rules it is much simpler to apply directly the expressions (\[gi10\]-\[gi04\]), computing explicitly the Lie derivatives of the tensor field.
Conclusions
===========
Many astrophysical systems (in particular, oscillating relativistic rotating stars) can be well described by perturbation theory depending on two parameters. A well–founded description of two–parameter perturbations can be very useful for such applications, specially in order to handle properly perturbations at second order and beyond. For example, one may wish to compare results derived in different gauges.
In this paper we have studied the problem of gauge dependence of non–linear perturbations depending on two parameters, considering perturbations of arbitrary order in a geometrical perspective, and generalizing the results of the one–parameter case [@BMMS; @SB] to the case of two parameters. We have constructed a geometrical framework in which a [*gauge choice*]{} is a two–parameter [*group*]{} of diffeomorphisms, while a [*gauge transformation*]{} is a two–parameter [*family*]{} of diffeomorphisms. We have shown that any family of diffeomorphisms can be expanded in terms of Lie derivatives with respect to vectors $\xi^{\mu}_{(p,q)}$. In terms of this expansion, which can be deduced order by order, we have derived general expressions for transformations of coordinates and tensor perturbations, and the conditions for gauge invariance of tensor perturbations. We have computed these expressions up to fourth order in the perturbative expansion, i.e. up to terms $\lambda^k\Omega^{k'}$ with $k+k'=4$.
The way in which the expansion of a two-parameter family of diffeomorphisms was derived in this paper is order by order, constructing derivative operators that can be rewritten as Lie derivatives with respect some vector fields. The development of an underlying geometrical structure, analogous to the knight diffeomorphisms introduced in the one–parameter case [@BMMS], would be interesting for two reasons: first, in order to have a deeper mathematical understanding of the theory, and second, in order to derive a close formula, valid at all orders, for gauge transformations and gauge invariance conditions. The present paper has been devoted to the derivation of the useful formulae for practical applications. In particular, our expressions will be useful to compare results derived in different gauges, and can form the basis for the construction of a gauge invariant theory of two-parameter systems in the line of works done for the one-parameter case like as for example [@moncrief; @gerlach; @bardeen; @be; @cb]. We leave the development of a more formal framework for future work.
Proof of the statement (\[lieder\]) {#sebastiano}
===================================
The aim of this Appendix is to give a proof of a theorem that allows us to make the statement contained in equation (\[lieder\]).
[**Theorem:**]{} “Let $\cal L$ be a derivative operator acting on the set of all the tensor fields defined on a differentiable manifold $\cal M$ and satisfying the following conditions: (i) It is linear and satisfies the Leibniz rule; (ii) it is tensor-type preserving; (iii) it commutes with every contraction of a tensor field; and (iv) it commutes with the exterior differentiation d. Then, there exists a vector field $\xi$ such that ${\cal L}$ is equivalent to the Lie derivative operator with respect to $\xi$, that is, $\pounds_\xi$.”
First of all, notice that the operators introduced in equations (\[operator10\]-\[operator04\]) satisfy the conditions of the theorem. In particular, properties (iii) and (iv) follow from the fact that $\Phi^{\ast}$ commutes with contractions and the exterior derivative (see [@thirring]). For more details on this question see, e.g., [@kobano].
The proof of the theorem is as follows: When acting on functions, $\cal L$ defines a vector field $\xi$ through the relation $${\cal L}\,f=:\xi(f), \qquad \forall\; f\in {\cal F}({\cal M})\,,
\label{xi}$$ where ${\cal F}({\cal M})$ denotes the algebra of $C^\infty$ functions on ${\cal M}$. What we want to prove is that on an arbitrary tensor field $T$, $${\cal L}\,T=\pounds_\xi\,T\,.\label{29}$$ Clearly (\[29\]) holds for an arbitrary tensor field T iff it holds for an arbitrary vector field V. For the latter equation (\[29\]) is equivalent to the following expression V=\[,V\]. Applying this to any function $f$ we obtain ([L]{}V)(f)=-V\[(f)\], f ([M]{}). \[vecf\] Therefore, to prove (\[29\]) is equivalent to prove (\[vecf\]). To this end, let us consider the action of the operator $\cal L$ on the function $V(f)$. Using (\[xi\]) we have $${\cal L}[V(f)]=\xi[V(f)] \,. \label{1}$$ On the other hand, using the properties (i)-(iv) of ${\cal L}$ we have \[V(f)\] & = & [L]{}([d]{}f(V))=[L]{}\[[C]{} ([d]{}fV)\]= [C]{}\[[d]{}([L]{}f)V + [d]{}fV\]\
& = & [d]{}([L]{}f)V + [d]{}f([L]{}V) = V([L]{}f) + ([L]{}V)(f). Then, this in combination with (\[1\]), and using (\[xi\]), leads to equation (\[vecf\]), which is what we wanted to prove.
Acknowledgments {#acknowledgments .unnumbered}
===============
We thank Sebastiano Sonego for his help in the proof of equation (\[lieder\]). We thank the anonymous referees for suggesting changes that have improved the manuscript. MB and LG thank Valeria Ferrari for useful comments and remarks. MB thanks the Dipartimento di Fisica “G. Marconi” (Università di Roma “La Sapienza”) and LG the Institute of Cosmology and Gravitation (University of Portsmouth) for hospitality during some stages in the realization of this work. CFS is supported by the EPSRC. This work has been supported by the EU programme ‘Improving the Human Research Potential and the Socio–Economic Knowledge Base’ (Research Training Network Contract HPRN–CT–2000–00137).
References {#references .unnumbered}
==========
[99]{}
Wald R M 1984 [*General Relativity*]{} (Chicago: The University of Chicago Press)
Bruni M, Matarrese S, Mollerach S and Sonego S 1997 [*Class. Quantum Grav.* ]{} [**14**]{} 2585
Matarrese S, Mollerach S and Bruni M 1998 [*Phys. Rev.*]{} [**D58**]{} 043504
Bruni M, Mena F C and Tavakol R 2002 [*Class. Quantum Grav.*]{} [**19**]{} L23
Chandrasekhar S 1983 [*The mathematical theory of black holes*]{} (Oxford: Clarendon Press)
Kokkotas, K D and Schmidt B G 1999 [*Quasi-Normal Modes of Stars and Black Holes*]{}, [*Living Rev. Relativity*]{} [**2**]{}, 2
: [http://www.livingreviews.org/Articles/Volume2/1999-2kokkotas/]{}
Price R H and Pullin J 1994 [*Phys. Rev. Lett.*]{} [**72**]{} 3297
Gleiser R J, Nicasio C O, Price R H and Pullin J 1998 [*Phys. Rev. Lett.*]{} [**77**]{} 4483
Khanna G, Baker J, Gleiser R J, Laguna P, Nicasio C O, Nollert H, Price R H and Pullin J 1999 [*Phys. Rev. Lett.*]{} [**83**]{} 3581
Gualtieri L, Berti E, Pons J A, Miniutti G and Ferrari V 2001 [*Phys. Rev.*]{} [**D64**]{} 104007
Pons J A, Berti E,Gualtieri L, Miniutti G and Ferrari V 2001 [*Phys. Rev.*]{} [**D65**]{} 104021
Miniutti L, Pons J A, Berti E, Gualtieri L, and Ferrari V, astro-ph/0206142
Campanelli M and Lousto C O 1999 [*Phys. Rev.*]{} [**D59**]{} 124022
Gleiser R J, Nicasio C O, Price R H and Pullin J 1996 [*Class. Quantum Grav.*]{} [**13**]{} L117
Gleiser R J, Nicasio C O, Price R H and Pullin J 2000 [*Phys. Rept.*]{} [**325**]{} 41
Kojima Y 1997 [*Prog. Theor. Phys. Suppl.*]{} [**128**]{} 251
Lockitch K H, Andersson N and Friedman J 2001 [*Phys. Rev.*]{} [**D63**]{} 024019
Ruoff J, Stavridis A and Kokkotas K D, gr-qc/0109065
Hartle J B 1967 [*Astrophys. J.*]{} [**150**]{} 1005
Hartle J B and Thorne K S 1968 [*Astrophys. J.*]{} [**153**]{} 807
Stergioulas, N 1998 [*Rotating Stars in Relativity*]{}, [*Living Rev. Relativity*]{} [**1**]{} 8 \[Online article\]: [http://www.livingreviews.org/Articles/Volume1/1998-8stergio/]{}
Cunningham C T, Price R H and Moncrief V 1980 [*Astrophys. J.*]{} [**236**]{} 674
Seidel E and Moore T 1987 [*Phys. Rev.*]{} [**D35**]{} 2287
Seidel E 1990 [*Phys. Rev.*]{} [**D42**]{} 1884
Gundlach C and Martín–García J M 2000 [*Phys. Rev.*]{} [**D61**]{} 084024
Flanagan E and Wald R 1996 [*Phys. Rev. D*]{} [**54**]{} 6233
Maia M D 1987 647
Maia M D 1987 651
Stewart J M and Walker M 1974 [*Proc. R. Soc.London A*]{} [**341**]{} 49
Stewart J M 1990 [*Class. Quantum Grav.*]{} [**7**]{} 1169
Stewart J 1991 [*Advanced General Relativity*]{} (Cambridge: Cambridge University Press)
Sonego S and Bruni M 1998 [*Commun. Math. Phys.*]{} [**193**]{} 209
Bruni M and Sonego S 1998 [*Class. Quant. Grav.*]{} 1999 [**16**]{} L29
Schutz B F 1984 The use of perturbation and approximation methods in general relativity [*Relativistic Astrophysics and Cosmology*]{} ed X Fustero and E Verdaguer (Singapore: World Scientific)
Sachs R K 1964 Gravitational radiation [*Relativity, Groups, and Topology*]{} ed C DeWitt and B DeWitt (New York: Gordon and Breach)
Geroch R 1969 [*Commun. Math. Phys.*]{} [**13**]{} 180
D’Inverno R 1992 [*Introducing Einstein’s Relativity*]{} (Oxford: Clarendon Press)
Weinberg S 1972 [*Gravitation and Cosmology*]{} (New York: Wiley)
Moncrief V 1974 [*Ann. Phys. N.Y.*]{} [**88**]{} 323
Gerlach U H and Sengupta U K 1979 [*Phys. Rev.*]{} [**D19**]{} 2268
Bardeen J M 1980 [*Phys. Rev.*]{} [**D8**]{} 1882
Ellis G F R and Bruni M 1989 [*Phys. Rev.*]{} [**D40**]{} 1804
Clarkson C A and Barrett R K 2002, gr-qc/0209051
Thirring W 1978 [*A Course in Mathematical Physics I: Classical Dynamical Systems*]{} (New York: Springer), pp. 54 and 57.
Kobayashi S and Nomizu K 1963 [*Foundations of differential geometry. Volume I*]{} (New York: John Wiley & Sons)
[^1]: The concept of perturbation theory with more than one parameter has already been introduced, for first-order perturbations, in [@MAIA], where the standard definition of spacetime perturbations [@stewart] is extended by using a (4+n)-dimensional flat space in which space-times are embedded. The main aim of these works is to re-examine the gauge invariance of the metric.
[^2]: We adopt the convention that the first label in $\phi$ corresponds to the flow generated by $\eta$ and parametrized by $\lambda$, while the second label corresponds to the flow generated by $\zeta$ and parametrized by $\Omega$. Therefore $\phi_{\lambda,\Omega}\neq
\phi_{\Omega,\lambda}$.
[^3]: The subscripts $(p,q)$ denote the lowest order in the expansion (\[mainidea\]) at which these operators will appear for the first time. See equation (\[Taylorexpression\]) below.
[^4]: As manifolds $\cM_0$ and $\cM$ are the same; for generality we assume that they are $m$–dimensional.
[^5]: Tensor fields on $\cN$ constructed in this way are “tangent” to $\cM$, in the sense that their components $m$ and $m+1$ in the charts we have defined vanish identically.
[^6]: In general, if the chart is adapted to the gauge $\varphi$, i.e. $\,^{\varphi}\eta^{\mu}=\,^{\varphi}\zeta^{\mu}=0$, it is not adapted to the gauge $\psi$, so $\,^{\psi}\eta^{\mu}\neq 0,\,^{\psi}\zeta^{\mu}\neq0$.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
The past few years have seen a surge of interest in the field of probabilistic logic learning and statistical relational learning. In this endeavor, many probabilistic logics have been developed. ProbLog is a recent probabilistic extension of Prolog motivated by the mining of large biological networks. In ProbLog, facts can be labeled with probabilities. These facts are treated as mutually independent random variables that indicate whether these facts belong to a randomly sampled program. Different kinds of queries can be posed to ProbLog programs. We introduce algorithms that allow the efficient execution of these queries, discuss their implementation on top of the YAP-Prolog system, and evaluate their performance in the context of large networks of biological entities.\
*To appear in Theory and Practice of Logic Programming (TPLP)*
author:
- |
Angelika Kimmig, Bart Demoen and Luc De Raedt\
Departement Computerwetenschappen, K.U. Leuven\
Celestijnenlaan 200A - bus 2402, B-3001 Heverlee, Belgium\
\
- |
Vítor Santos Costa and Ricardo Rocha\
CRACS $\&$ INESC-Porto LA, Faculty of Sciences, University of Porto\
R. do Campo Alegre 1021/1055, 4169-007 Porto, Portugal\
\
bibliography:
- 'tplp.bib'
title: On the Implementation of the Probabilistic Logic Programming Language ProbLog
---
\#1
Introduction {#sec:intro}
============
In the past few years, a multitude of different formalisms combining probabilistic reasoning with logics, databases, or logic programming has been developed. Prominent examples include PHA and ICL [@Poole:93; @Poole00], PRISM [@SatoKameya:01], SLPs [@Muggleton96], ProbView [@Lakshmanan], CLP($\cal BN$) [@Costa03:uai], CP-logic [@Vennekens], Trio [@Trio], probabilistic Datalog (pD) [@Fuhr00], and probabilistic databases [@DalviS04]. Although these logics have been traditionally studied in the knowledge representation and database communities, the focus is now often on a machine learning perspective, which imposes new requirements. First, these logics must be simple enough to be learnable and at the same time sufficiently expressive to support interesting probabilistic inferences. Second, because learning is computationally expensive and requires answering long sequences of possibly complex queries, inference in such logics must be fast, although inference in even the simplest probabilistic logics is computationally hard.
In this paper, we study these problems in the context of a simple probabilistic logic, ProbLog [@DeRaedt07], which has been used for learning in the context of large biological networks where edges are labeled with probabilities. Large and complex networks of biological concepts (genes, proteins, phenotypes, etc.) can be extracted from public databases, and probabilistic links between concepts can be obtained by various techniques [@Sevon06]. ProbLog is essentially an extension of Prolog where a program defines a distribution over all its possible non-probabilistic subprograms. Facts are labeled with probabilities and treated as mutually independent random variables indicating whether or not the corresponding fact belongs to a randomly sampled program. The success probability of a query is defined as the probability that it succeeds in such a random subprogram. The semantics of ProbLog is not new: it is an instance of the distribution semantics [@Sato:95]. This is a well-known semantics for probabilistic logics that has been (re)defined multiple times in the literature, often in a more limited database setting; cf. [@Dantsin; @Poole:93; @Fuhr00; @Poole00; @DalviS04]. Sato has, however, shown that the semantics is also well-defined in the case of a countably infinite set of random variables and formalized it in his well-known distribution semantics [@Sato:95]. However, even though relying on the same semantics, in order to allow efficient inference, systems such as PRISM [@SatoKameya:01] and PHA [@Poole:93] additionally require all proofs of a query to be mutually exclusive. Thus, they cannot easily represent the type of network analysis tasks that motivated ProbLog. ICL [@Poole00] extends PHA to the case where proofs need not be mutually exclusive. In contrast to the ProbLog implementation presented here, Poole’s AILog2, an implementation of ICL, uses a meta-interpreter and is not tightly integrated with Prolog.
We contribute exact and approximate inference algorithms for ProbLog. We present algorithms for computing the success and explanation probabilities of a query, and show how they can be efficiently implemented combining Prolog inference with Binary Decision Diagrams (BDDs) [@Bryant86]. In addition to an iterative deepening algorithm that computes an approximation along the lines of [@Poole93:jrnl], we further adapt the Monte Carlo approach used by [@Sevon06] in the context of biological network inference. These two approximation algorithms compute an upper and a lower bound on the success probability. We also contribute an additional approximation algorithm that computes a lower bound using only the $k$ most likely proofs.
The key contribution of this paper is the tight integration of these algorithms in the state-of-the-art YAP-Prolog system. This integration includes several improvements over the initial implementation used in [@DeRaedt07], which are needed to use ProbLog to effectively query Sevon’s Biomine network [@Sevon06] containing about 1,000,000 nodes and 6,000,000 edges, as will be shown in the experiments.
This paper is organised as follows. After introducing ProbLog and its semantics in Section 2, we present several algorithms for exact and approximate inference in Section 3. Section 4 then discusses how these algorithms are implemented in YAP-Prolog, and Section 5 reports on experiments that validate the approach. Finally, Section 6 concludes and touches upon related work.
ProbLog {#sec:problog}
=======
A ProbLog program consists of a set of labeled facts $p_i::c_i$ together with a set of definite clauses. Each ground instance (that is, each instance not containing variables) of such a fact $c_i$ is true with probability $p_i$, that is, these facts correspond to random variables. We assume that these variables are mutually independent.[^1] The definite clauses allow one to add arbitrary *background knowledge* (BK).
Figure \[fig:Ex\] shows a small probabilistic graph that we shall use as running example in the text. It can be encoded in ProbLog as follows: $$\begin{array}{lllll}
0\ldotp8 :: \mathtt{edge(a,c)\ldotp} & ~~~~ & 0\ldotp7 :: \mathtt{edge(a,b)\ldotp} & ~~~~ & 0\ldotp8 :: \mathtt{edge(c,e)\ldotp} \\
0\ldotp6 :: \mathtt{edge(b,c)\ldotp} & ~~~~ & 0\ldotp9 :: \mathtt{edge(c,d)\ldotp} & ~~~~ & 0\ldotp5 :: \mathtt{edge(e,d)\ldotp}
\end{array}$$ Such a probabilistic graph can be used to sample subgraphs by tossing a coin for each edge. Given a ProbLog program $T=\{p_1::c_1,\cdots,p_n::c_n\} \cup BK$ and a finite set of possible substitutions $\{\theta_{j1}, \ldots \theta_{ji_j}\}$ for each probabilistic fact $p_j::c_j$, let $L_T$ denote the maximal set of *logical* facts that can be added to $BK$, that is, $L_T=\{c_1\theta_{11}, \ldots , c_1\theta_{1i_1}, \cdots, c_n\theta_{n1}, \ldots , c_n\theta_{ni_n}\}$. As the random variables corresponding to facts in $L_T$ are mutually independent, the ProbLog program defines a probability distribution over ground logic programs $L \subseteq L_T$: $$P(L|T)=\prod\nolimits_{c_i\theta_j\in L}p_i\prod\nolimits_{c_i\theta_j\in L_T\backslash L}(1-p_i)\ldotp$$ Since the background knowledge $BK$ is fixed and there is a one-to-one mapping between ground definite clause programs and Herbrand interpretations, a ProbLog program thus also defines a distribution over its Herbrand interpretations. Sato has shown how this semantics can be generalized to the countably infinite case; we refer to [@Sato:95] for details. For ease of readability, in the remainder of this paper we will restrict ourselves to the finite case and assume all probabilistic facts in a ProbLog program to be ground. We extend our example with the following background knowledge: $$\begin{array}{lll}
\mathtt{path(X,Y)} & \mathtt{:-} & \mathtt{edge(X,Y)\ldotp} \\
\mathtt{path(X,Y)} & \mathtt{:-} & \mathtt{edge(X,Z), path(Z,Y)\ldotp}
\end{array}$$ We can then ask for the probability that there exists a path between two nodes, say *c* and *d*, in our probabilistic graph, that is, we query for the probability that a randomly sampled subgraph contains the edge from *c* to *d*, or the path from *c* to *d* via *e* (or both of these).
![ Example of a probabilistic graph: edge labels indicate the probability that the edge is part of the graph.[]{data-label="fig:Ex"}](graph)
Formally, the *success probability* $P_s(q|T)$ of a query $q$ in a ProbLog program $T$ is the marginal of $P(L|T)$ with respect to $q$, i.e. $$P_s(q|T) = \sum\nolimits_{L\subseteq L_T}P(q|L)\cdot P(L|T)\;, \label{eq:p_suc}$$ where $P(q|L) = 1$ if there exists a $\theta$ such that $L\cup BK\models
q\theta$, and $P(q|L)=0$ otherwise. In other words, the success probability of query $q$ is the probability that the query $q$ is *provable* in a randomly sampled logic program.
In our example, $40$ of the $64$ possible subprograms allow one to prove *path$(c,d)$*, namely all those that contain at least the edge from *c* to *d* or both the edge from *c* to *e* and from *e* to *d*, so the success probability of that query is the sum of the probabilities of these programs: $P_s(path(c,d)|T)=P(\{ab,ac,bc,cd,ce,ed\}|T)+\ldots +P(\{cd\}|T)=0\ldotp94$, where $xy$ is used as a shortcut for *edge$(x,y)$* when listing elements of a subprogram. We will use this convention throughout the paper. Clearly, listing all subprograms is infeasible in practice; an alternative approach will be discussed in Section \[sec:exact\].
A ProbLog program also defines the probability of a *specific* proof $E$, also called *explanation*, of some query $q$, which is again a marginal of $P(L|T)$. Here, an *explanation* is a minimal subset of the probabilistic facts that together with the background knowledge entails $q\theta$ for some substitution $\theta$. Thus, the probability of such an explanation $E$ is that of sampling a logic program $L\cup E$ that contains at least all the probabilistic facts in $E$, that is, the marginal with respect to these facts: $$P(E|T) = \sum\nolimits_{ L\subseteq (L_T\backslash E)} P(L\cup E |T) = \prod\nolimits_{c_i \in E}p_i\label{eq:deriv_px}$$ The *explanation probability* $P_x(q|T)$ is then defined as the probability of the most likely explanation or proof of the query $q$ $$P_x(q|T) = \max\nolimits_{E\in E(q)}P(E|T)
= \max\nolimits_{E\in E(q)} \prod_{c_i \in E}p_i,\label{eq:p_exp}$$ where $E(q)$ is the set of all explanations for query $q$, i.e., all minimal sets $E\subseteq L_T$ of probabilistic facts such that $E \cup BK \models q$ [@Kimmig07].
In our example, the set of all explanations for *path$(c,d)$* contains the edge from *c* to *d* (with probability 0.9) as well as the path consisting of the edges from *c* to *e* and from *e* to *d* (with probability $0\ldotp8\cdot 0\ldotp5=0\ldotp4$). Thus, $P_x(path(c,d)|T)=0\ldotp9$.
The ProbLog semantics is essentially a distribution semantics [@Sato:95]. Sato has rigorously shown that this class of programs defines a joint probability distribution over the set of possible least Herbrand models of the program (allowing functors), that is, of the background knowledge $BK$ together with a subprogram $L \subseteq L_T$; for further details we refer to [@Sato:95]. The distribution semantics has been used widely in the literature, though often under other names or in a more restricted setting; see e.g. [@Dantsin; @Poole:93; @Fuhr00; @Poole00; @DalviS04].
Inference in ProbLog {#sec:inference}
====================
This section discusses algorithms for computing exactly or approximately the success and explanation probabilities of ProbLog queries. It additionally contributes a new algorithm for Monte Carlo approximation of success probabilities.
Exact Inference {#sec:exact}
---------------
Calculating the *success probability* of a query using Equation (\[eq:p\_suc\]) directly is infeasible for all but the tiniest programs, as the number of subprograms to be checked is exponential in the number of probabilistic facts. However, as we have seen in our example in Section \[sec:problog\], we can describe all subprograms allowing for a specific proof by means of the facts that such a program has to contain, i.e., all the ground probabilistic facts used in that proof. As probabilistic facts correspond to random variables indicating the presence of facts in a sampled program, we alternatively denote proofs by conjunctions of such random variables. In our example, query *path(c,d)* has two proofs in the full program: *{edge(c,d)}* and *{edge(c,e),edge(e,d)}*, or, using logical notation, $cd$ and $ce \wedge ed$. The set of all subprograms containing *some* proof thus can be described by a disjunction over all possible proofs, in our case, $cd \vee (ce \wedge ed)$. This idea forms the basis for the inference method presented in [@DeRaedt07], which uses two steps:
1. Compute the proofs of the query $q$ in the logical part of the theory $T$, that is, in $BK \cup L_T$. The result will be a DNF formula.
2. Compute the probability of this formula.
Similar approaches are used for PRISM [@SatoKameya:01], ICL [@Poole00] and pD [@Fuhr00].
The probability of a single given proof, cf. Equation (\[eq:deriv\_px\]), is the marginal over all programs allowing for that proof, and thus equals the product of the probabilities of the facts used by that proof. However, we cannot directly sum the results for the different proofs to obtain the success probability, as a specific subprogram can allow several proofs and therefore contributes to the probability of each of these proofs. Indeed, in our example, all programs that are supersets of *{edge(c,e),edge(e,d),edge(c,d)}* contribute to the marginals of both proofs and would therefore be counted twice if summing the probabilities of the proofs. However, for mutually exclusive conjunctions, that is, conjunctions describing disjoint sets of subprograms, the probability is the sum of the individual probabilities. This situation can be achieved by adding *negated* random variables to a conjunction, thereby explicitly excluding subprograms covered by another part of the formula from the corresponding part of the sum. In the example, extending $ce \wedge ed$ to $ce \wedge ed \wedge \neg cd$ reduces the second part of the sum to those programs not covered by the first: $$P_s(path(c,d)|T)=P(cd \vee (ce\wedge ed)|T)$$$$= P(cd|T)+P(ce\wedge ed\wedge\neg cd|T)$$$$= 0\ldotp9 + 0\ldotp8\cdot0\ldotp5\cdot(1-0\ldotp9)=0\ldotp94$$
However, as the number of proofs grows, disjoining them gets more involved. Consider for example the query *path(a,d)* which has four different but highly interconnected proofs. In general, this problem is known as the *disjoint-sum-problem* or the two-terminal network reliability problem, which is \#P-complete [@Valiant1979].
Before returning to possible approaches to tackle the disjoint-sum-problem at the end of this section, we will now discuss the two steps of ProbLog’s exact inference in more detail.
![SLD-tree for query *path$(c,d)$.*[]{data-label="fig:SLD"}](sld)
Following Prolog, the first step employs SLD-resolution to obtain all different proofs. As an example, the SLD-tree for the query *?- path$(c,d)$.* is depicted in Figure \[fig:SLD\]. Each successful proof in the SLD-tree uses a set of ground probabilistic facts $\{p_1::c_1, \cdots, p_k::c_k\}
\subseteq T$. These facts are necessary for the proof, and the proof is *independent* of other probabilistic facts in $T$.
Let us now introduce a Boolean random variable $b_i$ for each ground probabilistic fact $p_i::c_i \in T$, indicating whether $c_i$ is in a sampled logic program, that is, $b_i$ has probability $p_i$ of being true.[^2] A particular proof of query $q$ involving ground facts $\{p_1::c_1, \cdots, p_k::c_k\}
\subseteq T$ is thus represented by the conjunctive formula $b_1
\wedge \cdots \wedge b_k$, which at the same time represents the set of all subprograms containing these facts. Furthermore, using $E(q)$ to denote the set of proofs or explanations of the goal $q$, the set of all subprograms containing *some* proof of $q$ can be denoted by $\bigvee_{e \in E(q) } \, \bigwedge_{c_i \in e} b_i $, as the following derivation shows: $$\begin{aligned}
\bigvee_{e \in E(q) } \, \bigwedge_{c_i \in e} b_i & = & \bigvee_{e \in E(q) } \left( \bigwedge_{c_i \in e} b_i \wedge \bigwedge_{c_i \in L_T \backslash e} (b_i \vee \neg b_i)\right)\\
& = & \bigvee_{e \in E(q) } \bigvee_{L \subseteq L_T\backslash e } \left( \bigwedge_{c_i \in e} b_i \wedge \left(\bigwedge_{c_i \in L} b_i \wedge \bigwedge_{c_i \in L_T \backslash (L\union e)} \neg b_i\right)\right)\\
& = & \bigvee_{e \in E(q) , L \subseteq L_T\backslash e } \left( \bigwedge_{c_i \in L \union e} b_i \wedge \bigwedge_{c_i \in L_T \backslash (L\union e)} \neg b_i\right)\\
& = & \bigvee_{ L \subseteq L_T, \exists\theta L\union BK \models q\theta } \left( \bigwedge_{c_i \in L } b_i \wedge \bigwedge_{c_i \in L_T \backslash L} \neg b_i\right)\end{aligned}$$ We first add all possible ways of extending a proof $e$ to a full sampled program by considering each fact not in $e$ in turn. We then note that the disjunction of these fact-wise extensions can be written on the basis of sets. Finally, we rewrite the condition of the disjunction in the terms of Equation (\[eq:p\_suc\]). This is possible as each subprogram that is an extension of an explanation of $q$ entails some ground instance of $q$, and vice versa, each subprogram entailing $q$ is an extension of some explanation of $q$. As the DNF now contains conjunctions representing fully specified programs, its probability is a sum of products, which directly corresponds to Equation (\[eq:p\_suc\]): $$\begin{aligned}
\lefteqn{ P(\bigvee_{ L \subseteq L_T, \exists\theta L\union BK \models q\theta } \left( \bigwedge_{c_i \in L } b_i \wedge \bigwedge_{c_i \in L_T \backslash L} \neg b_i\right)) }\\
&=& \sum_{ L \subseteq L_T, \exists\theta L\union BK \models q\theta } \left( \prod_{c_i \in L } p_i \cdot \prod_{c_i \in L_T \backslash L} (1- p_i)\right)\\
&=& \sum_{ L \subseteq L_T, \exists\theta L\union BK \models q\theta } P(L|T)\end{aligned}$$ We thus obtain the following alternative characterisation of the success probability: $$P_s(q|T) = P\left( \bigvee_{e \in E(q) } \, \bigwedge_{c_i \in e} b_i \right)
\label{eq:dnf}$$ where $E(q)$ denotes the set of proofs or explanations of the goal $q$ and $b_i$ denotes the Boolean variable corresponding to ground probabilistic fact $p_i::c_i$. Thus, the problem of computing the success probability of a ProbLog query can be reduced to that of computing the probability of a DNF formula.
However, as argued above, due to overlap between different conjunctions, the proof-based DNF of Equation (\[eq:dnf\]) cannot directly be transformed into a sum of products. Computing the probability of DNF formulae thus involves solving the disjoint-sum-problem, and therefore is itself a \#P-hard problem. Various algorithms have been developed to tackle this problem. The pD-engine HySpirit [@Fuhr00] uses the inclusion-exclusion principle, which is reported to scale to about ten proofs. For ICL, which extends PHA by allowing non-disjoint proofs, [@Poole00] proposes a symbolic disjoining algorithm, but does not report scalability results. Our implementation of ProbLog employs Binary Decision Diagrams (BDDs) [@Bryant86], an efficient graphical representation of a Boolean function over a set of variables, which scales to tens of thousands of proofs; see Section \[sec:BDD\] for more details. PRISM [@SatoKameya:01] and PHA [@Poole:93] differ from the systems mentioned above in that they avoid the disjoint-sum-problem by requiring the user to write programs such that proofs are guaranteed to be disjoint.
On the other hand, as the *explanation probability* $P_x$ exclusively depends on the probabilistic facts used in one proof, it can be calculated using a simple branch-and-bound approach based on the SLD-tree, where partial proofs are discarded if their probability drops below that of the best proof found so far.
Approximative Inference
-----------------------
As the size of the DNF formula grows with the number of proofs, its evaluation can become quite expensive, and ultimately infeasible. For instance, when searching for paths in graphs or networks, even in small networks with a few dozen edges there are easily $O(10^6)$ possible paths between two nodes. ProbLog therefore includes several approximation methods.
### Bounded Approximation
The first approximation algorithm, a slight variant of the one proposed in [@DeRaedt07], uses DNF formulae to obtain both an upper and a lower bound on the probability of a query. It is closely related to work by [@Poole93:jrnl] in the context of PHA, but adapted towards ProbLog. The method relies on two observations.
First, we remark that the DNF formula describing sets of proofs is *monotone*, meaning that adding more proofs will never decrease the probability of the formula being true. Thus, formulae describing subsets of the full set of proofs of a query will always give a lower bound on the query’s success probability. In our example, the lower bound obtained from the shorter proof would be $P(cd|T) = 0\ldotp9$, while that from the longer one would be $P(ce\wedge ed|T) = 0\ldotp4$.
Our second observation is that the probability of a proof $b_1 \wedge \ldots\wedge b_n$ will always be at most the probability of an arbitrary prefix $b_1 \wedge \ldots\wedge b_i, i\leq n$. In our example, the probability of the second proof will be at most the probability of its first edge from $c$ to $e$, i.e., $P(ce|T) = 0\ldotp8 \geq 0\ldotp4$. As disjoining sets of proofs, i.e., including information on facts that are *not* elements of the subprograms described by a certain proof, can only decrease the contribution of single proofs, this upper bound carries over to a set of proofs or partial proofs, as long as prefixes for all possible proofs are included. Such sets can be obtained from an incomplete SLD-tree, i.e., an SLD-tree where branches are only extended up to a certain point.
This motivates ProbLog’s *bounded approximation algorithm*. The algorithm relies on a probability threshold $\gamma$ to stop growing the SLD-tree and thus obtain DNF formulae for the two bounds[^3]. The lower bound formula $d_1$ represents all proofs with a probability above the current threshold. The upper bound formula $d_2$ additionally includes all derivations that have been stopped due to reaching the threshold, as these still *may* succeed. Our goal is therefore to grow $d_1$ and $d_2$ in order to decrease $P(d_2|T)-P(d_1|T)$.
Given an acceptance threshold $\delta_p$, an initial probability threshold $\gamma$, and a shrinking factor $\beta\in(0,1)$, the algorithm proceeds in an iterative-deepening manner as outlined in Algorithm \[alg:delta\]. Initially, both $d_1$ and $d_2$ are set to <span style="font-variant:small-caps;">False</span>, the neutral element with respect to disjunction, and the probability bounds are $0$ and $1$, as we have no full proofs yet, and the empty partial proof holds in any model.
$d_1 = $ <span style="font-variant:small-caps;">False</span>; $d_2 = $ <span style="font-variant:small-caps;">False</span>; $P(d_1|T) =0$; $P(d_2|T) = 1$; $p = $<span style="font-variant:small-caps;">True</span>; Expand current proof $p$ set $d_2 = $ <span style="font-variant:small-caps;">True</span> Compute $P(d_1|T)$ and $P(d_2|T)$ $\gamma := \gamma\cdot\beta$ return $[P(d_1|T),P(d_2|T)]$
It should be clear that $P(d_1|T)$ monotonically increases, as the number of proofs never decreases. On the other hand, as explained above, if $d_2$ changes from one iteration to the next, this is always because a partial proof $p$ is either removed from $d_2$ and therefore no longer contributes to the probability, or it is replaced by proofs $p_1,\ldots , p_n$, such that $p_i = p \land s_i$, hence $P(p_1 \lor \ldots \lor p_n|T) = P(p \land s_1\lor\ldots\lor p\land s_n|T) = P(p \land ( s_1\lor\ldots\lor s_n)|T)$. As proofs are subsets of the probabilistic facts in the ProbLog program, each literal’s random variable appears at most once in the conjunction representing a proof, even if the corresponding subgoal is called multiple times when constructing the proof. We therefore know that the literals in the prefix $p$ cannot be in any suffix $s_i$, hence, given ProbLog’s independence assumption, $P(p \land ( s_1\lor\ldots\lor s_n)|T) = P(p|T)P(s_1\lor\ldots\lor s_n|T) \leq P(p|T)$. Therefore, $P(d_2)$ monotonically decreases.
As an illustration, consider a probability threshold $\gamma =0\ldotp9$ for the SLD-tree in Figure \[fig:SLD\]. In this case, $d_1$ encodes the left success path while $d_2$ additionally encodes the path up to *path$(e,d)$*, i.e., $d_1 = cd$ and $d_2 = cd \vee ce$, whereas the formula for the full SLD-tree is $d = cd \vee (ce \wedge ed)$. The lower bound thus is $0\ldotp9$, the upper bound (obtained by disjoining $d_2$ to $cd \vee (ce\wedge\neg cd)$) is $0\ldotp98$, whereas the true probability is $0\ldotp94$.
Notice that in order to implement this algorithm we need to compute the probability of a set of proofs. This task will be described in detail in Section \[sec:implementation\].
### K-Best
Using a fixed number of proofs to approximate the probability allows better control of the overall complexity, which is crucial if large numbers of queries have to be evaluated, e.g., in the context of parameter learning. [@Gutmann08] therefore introduces the $k$-probability $P_k(q|T)$, which approximates the success probability by using the $k$-best (that is, the $k$ most likely) explanations instead of all proofs when building the DNF formula used in Equation (\[eq:dnf\]): $$P_k(q|T) = P\left( \bigvee_{e \in E_k(q) } \, \bigwedge_{b_i \in var(e)} b_i \right)\label{eq:p_k}$$ where $E_k(q)=\{e \in E(q)|P_x(e)\geq P_x(e_k)\}$ with $e_k$ the $k$th element of $E(q)$ sorted by non-increasing probability. Setting $k=\infty$ leads to the success probability, whereas $k=1$ corresponds to the explanation probability provided that there is a single best proof. The branch-and-bound approach used to calculate the explanation probability can directly be generalized to finding the $k$-best proofs; cf. also [@Poole:93].
To illustrate $k$-probability, we consider again our example graph, but this time with query *path$(a,d)$*. This query has four proofs, represented by the conjunctions $ac\wedge cd$, $ab\wedge bc \wedge
cd$, $ac\wedge ce \wedge ed$ and $ab\wedge bc \wedge ce \wedge ed$, with probabilities $0\ldotp72$, $0\ldotp378$, $0\ldotp32$ and $0\ldotp168$ respectively. As $P_1$ corresponds to the explanation probability $P_x$, we obtain $P_1(path(a,d))=0\ldotp72$. For $k=2$, the overlap between the best two proofs has to be taken into account: the second proof only adds information if the first one is absent. As they share edge $cd$, this means that edge $ac$ has to be missing, leading to $P_2(path(a,d))=P((ac\wedge cd) \vee (\neg ac \wedge ab\wedge bc
\wedge cd))=0\ldotp72+(1-0\ldotp8)\cdot 0\ldotp378=0\ldotp7956$. Similarly, we obtain $P_3(path(a,d))=0\ldotp8276$ and $P_k(path(a,d))=0\ldotp83096$ for $k\geq 4$.
### Monte Carlo {#sec:mc_method}
As an alternative approximation technique, we propose a Monte Carlo method, where we proceed as follows.
Execute until convergence:
1. Sample a logic program from the ProbLog program
2. Check for the existence of some proof of the query of interest
3. Estimate the query probability $P$ as the fraction of samples where the query is provable
We estimate convergence by computing the 95% confidence interval at each $m$ samples. Given a large number $N$ of samples, we can use the standard normal approximation interval to the binomial distribution:
$$\delta \approx 2\times\sqrt{\frac{P.(P-1)}{N}}$$
Notice that confidence intervals do not directly correspond to the exact bounds used in our previous approximation algorithm. Still, we employ the same stopping criterion, that is, we run the Monte Carlo simulation until the width of the confidence interval is at most $\delta_p$.
A similar algorithm (without the use of confidence intervals) was also used in the context of biological networks (not represented as Prolog programs) by [@Sevon06]. The use of a Monte Carlo method for probabilistic logic programs was suggested already by [@Dantsin], although he neither provides details nor reports on an implementation. Our approach differs from the MCMC method for Stochastic Logic Programs (SLPs) introduced by [@Cussens00] in that we do not use a Markov chain, but restart from scratch for each sample. Furthermore, SLPs are different in that they directly define a distribution over all proofs of a query. Investigating similar probabilistic backtracking approaches for ProbLog is a promising future research direction.
Implementation {#sec:implementation}
==============
This section discusses the main building blocks used to implement ProbLog on top of the YAP-Prolog system. An overview is shown in Figure \[fig:problog\_imp\], with a typical ProbLog program, including ProbLog facts and background knowledge (BK), at the top.
![ProbLog Implementation: A ProbLog program (top) requires the ProbLog library which in turn relies on functionality from the tries and array libraries. ProbLog queries (bottom-left) are sent to the YAP engine, and may require calling the BDD library CUDD via SimpleCUDD.[]{data-label="fig:problog_imp"}](implementation "fig:")\[fig:problog\]
The implementation requires ProbLog programs to use the `problog` module. Each program consists of a set of labeled facts and of unlabeled *background knowledge*, a generic Prolog program. Labeled facts are preprocessed as described below. Notice that the implementation requires all queries to non-ground probabilistic facts to be ground on calling.
In contrast to standard Prolog queries, where one is interested in answer substitutions, in ProbLog one is primarily interested in a probability. As discussed before, two common ProbLog queries ask for the most likely explanation and its probability, and the probability of whether a query would have an answer substitution. We have discussed two very different approaches to the problem:
- In exact inference, $k$-best and bounded approximation, the engine explicitly reasons about probabilities of proofs. The challenge is how to compute the probability of each individual proof, store a large number of proofs, and compute the probability of sets of proofs.
- In Monte Carlo, the probabilities of facts are used to sample from ProbLog programs. The challenge is how to compute a sample quickly, in a way that inference can be as efficient as possible.
ProbLog programs execute from a top-level query and are driven through a ProbLog query. The inference algorithms discussed above can be abstracted as follows:
- Initialise the inference algorithm;
- While probabilistic inference did not converge:
- initialise a new query;
- execute the query, instrumenting every ProbLog call in the current proof. Instrumentation is required for recording the ProbLog facts required by a proof, but may also be used by the inference algorithm to stop proofs (e.g., if the current probability is lower than a bound);
- process success or exit substitution;
- Proceed to the next step of the algorithm: this may be trivial or may require calling an external solver, such as a BDD tool, to compute a probability.
Notice that the current ProbLog implementation relies on the Prolog engine to efficiently execute goals. On the other hand, and in contrast to most other probabilistic language implementations, in ProbLog there is no clear separation between logical and probabilistic inference: in a fashion similar to constraint logic programming, probabilistic inference can drive logical inference.
From a Prolog implementation perspective, ProbLog poses a number of interesting challenges. First, labeled facts have to be efficiently compiled to allow mutual calls between the Prolog program and the ProbLog engine. Second, for exact inference, $k$-best and bounded approximation, sets of proofs have to be manipulated and transformed into BDDs. Finally, Monte Carlo simulation requires representing and manipulating samples. We discuss these issues next.
Source-to-source transformation
-------------------------------
We use the `term_expansion` mechanism to allow Prolog calls to labeled facts, and for labeled facts to call the ProbLog engine. As an example, the program: $$\begin{array}{l}
\mathtt{0\ldotp715::edge('PubMed\_2196878','MIM\_609065')\ldotp}\\
\mathtt{0\ldotp659::edge('PubMed\_8764571','HGNC\_5014')\ldotp}\\
\end{array}$$ would be compiled as: $$\begin{array}{lll}
\mathtt{edge(A,B)} &\mathtt{:-} & \mathtt{problog\_edge(ID,A,B,LogProb),}\\
& & \mathtt{grounding\_id(edge(A,B),ID,GroundID),}\\
& & \mathtt{add\_to\_proof(GroundID,LogProb)\ldotp}\\
& & \\
\multicolumn{3}{l}{\mathtt{problog\_edge(0,'PubMed\_2196878','MIM\_609065',-0\ldotp3348)\ldotp}} \\
\multicolumn{3}{l}{\mathtt{problog\_edge(1,'PubMed\_8764571','HGNC\_5014',-0\ldotp4166)\ldotp}} \\
\end{array}$$ Thus, the internal representation of each fact contains an identifier, the original arguments, and the logarithm of the probability[^4]. The `grounding_id` procedure will create and store a grounding specific identifier for each new grounding of a non-ground probabilistic fact encountered during proving, and retrieve it on repeated use. For ground probabilistic facts, it simply returns the identifier itself. The `add_to_proof` procedure updates the data structure representing the current path through the search space, i.e., a queue of identifiers ordered by first use, together with its probability. Compared to the original meta-interpreter based implementation of [@DeRaedt07], the main benefit of source-to-source transformation is better scalability, namely by having a compact representation of the facts for the YAP engine [@DBLP:conf/padl/Costa07] and by allowing access to the YAP indexing mechanism [@jit-index].
Proof Manipulation
------------------
Manipulating proofs is critical in ProbLog. We represent each proof as a queue containing the identifier of each different ground probabilistic fact used in the proof, ordered by first use. The implementation requires calls to non-ground probabilistic facts to be ground, and during proving maintains a table of groundings used within the current query together with their identifiers. Grounding identifiers are based on the fact’s identifier extended with a grounding number, i.e. $5\_1$ and $5\_2$ would refer to different groundings of the non-ground fact with identifier $5$. In our implementation, the queue is stored in a backtrackable global variable, which is updated by calling `add_to_proof` with an identifier for the current ProbLog fact. We thus exploit Prolog’s backtracking mechanism to avoid recomputation of shared proof prefixes when exploring the space of proofs. Storing a proof is simply a question of adding the value of the variable to a store.
As we have discussed above, the actual number of proofs can grow very quickly. ProbLog compactly represents a proof as a list of numbers. We would further like to have a scalable implementation of *sets* of proofs, such that we can compute the joint *probability* of large sets of proofs efficiently. Our representation for sets of proofs and our algorithm for computing the probability of such a set are discussed next.
Sets of Proofs
--------------
Storing and manipulating proofs is critical in ProbLog. When manipulating proofs, the key operation is often *insertion*: we would like to add a proof to an existing set of proofs. Some algorithms, such as exact inference or Monte Carlo, only manipulate complete proofs. Others, such as bounded approximation, require adding partial derivations too. The nature of the SLD-tree means that proofs tend to share both a prefix and a suffix. Partial proofs tend to share prefixes only. This suggests using *tries* to maintain the set of proofs. We use the YAP implementation of tries for this task, based itself on XSB Prolog’s work on tries of terms [@RamakrishnanIV-99], which we briefly summarize here.
Tries [@Fredkin-62] were originally invented to index dictionaries, and have since been generalised to index recursive data structures such as terms. Please refer to [@Bachmair-93; @Graf-96; @RamakrishnanIV-99] for the use of tries in automated theorem proving, term rewriting and tabled logic programs. An essential property of the trie data structure is that common prefixes are stored only once. A trie is a tree structure where each different path through the trie data units, the *trie nodes*, corresponds to a term described by the tokens labelling the nodes traversed. For example, the tokenized form of the term $f(g(a),1)$ is the sequence of 4 tokens: $f/2$, $g/1$, $a$ and $1$. Two terms with common prefixes will branch off from each other at the first distinguishing token.
Trie’s internal nodes are four field data structures, storing the node’s token, a pointer to the node’s first child, a pointer to the node’s parent and a pointer to the node’s next sibling, respectively. Each internal node’s outgoing transitions may be determined by following the child pointer to the first child node and, from there, continuing sequentially through the list of sibling pointers. When a list of sibling nodes becomes larger than a threshold value (8 in our implementation), we dynamically index the nodes through a hash table to provide direct node access and therefore optimise the search. Further hash collisions are reduced by dynamically expanding the hash tables. Inserting a term requires in the worst case allocating as many nodes as necessary to represent its complete path. On the other hand, inserting repeated terms requires traversing the trie structure until reaching the corresponding leaf node, without allocating any new node.
In order to minimize the number of nodes when storing proofs in a trie, we use Prolog lists to represent proofs. For example, a ProbLog proof $[3, 5\_1, 7, 5\_2]$ uses ground fact 3, a first grounding of fact 5, ground fact 7 and another grounding of fact 5, that is, list elements in proofs are always either integers or two integers with an underscore in between.
Figure \[fig:trie\_proofs\] presents an example of a trie storing three proofs. Initially, the trie contains the root node only. Next, we store the proof $[3, 5\_1, 7, 5\_2]$ and six nodes (corresponding to six tokens) are added to represent it (Figure \[fig:trie\_proofs\](a)). The proof $[3, 5\_1, 9, 7, 5\_2]$ is then stored which requires seven nodes. As it shares a common prefix with the previous proof, we save the three initial nodes common to both representations (Figure \[fig:trie\_proofs\](b)). The proof $[3, 4, 7]$ is stored next and we save again the two initial nodes common to all proofs (Figure \[fig:trie\_proofs\](c)).
![Using tries to store proofs. Initially, the trie contains the root node only. Next, we store the proofs: (a) $[3, 5\_1, 7, 5\_2]$; (b) $[3, 5\_1, 9, 7, 5\_2]$; and (c) $[3, 4, 7]$.[]{data-label="fig:trie_proofs"}](trie_proofs)
Binary Decision Diagrams {#sec:BDD}
------------------------
![Binary Decision Diagram encoding the DNF formula $cd \vee (ce \wedge
ed)$, corresponding to the two proofs of query *path(c,d)* in the example graph. An internal node labeled $xy$ represents the Boolean variable for the edge between $x$ and $y$, solid/dashed edges correspond to values true/false and are labeled with the probability that the variable takes this value.[]{data-label="fig:BDD"}](bdd)
To efficiently compute the probability of a DNF formula representing a set of proofs, our implementation represents this formula as a reduced ordered Binary Decision Diagram (BDD) [@Bryant86], which can be viewed as a compact encoding of a Boolean decision tree. Given a fixed variable ordering, a Boolean function $f$ can be represented as a full Boolean decision tree, where each node on the $i$th level is labeled with the $i$th variable and has two children called low and high. Leaves are labeled by the outcome of $f$ for the variable assignment corresponding to the path to the leaf, where in each node labeled $x$, the branch to the low (high) child is taken if variable $x$ is assigned 0 (1). Starting from such a tree, one obtains a BDD by merging isomorphic subgraphs and deleting redundant nodes until no further reduction is possible. A node is redundant if the subgraphs rooted at its children are isomorphic. Figure \[fig:BDD\] shows the BDD for the existence of a path between *c* and *d* in our earlier example.
We use SimpleCUDD[^5] as a wrapper tool for the BDD package CUDD[^6] to construct and evaluate BDDs. More precisely, the trie representation of the DNF is translated to a BDD generation script, which is processed by SimpleCUDD to build the BDD using CUDD primitives. It is executed via Prolog’s shell utility, and results are reported via shared files.
$i := 1$ $S_{\wedge} := \{(C,P)|C $ leaf in $T$ and single child of its parent $P \}$ write $n_i = P\wedge C$ $T := \textsc{Replace}(T,(C,P),n_i)$ $i := i + 1$ $S_{\vee} := \{[C_1,\ldots,C_n]|$ leaves $C_j $ are all the children of some parent $P$ in $T\}$ write $n_i = C_1 \vee \ldots \vee C_n$ $T := \textsc{Replace}(T,[C_1,\ldots,C_n],n_i)$ $i := i + 1$ write $top = n_{i-1}$
During the generation of the code, it is crucial to exploit the structure sharing (prefixes and suffixes) already in the trie representation of a DNF formula, otherwise CUDD computation time becomes extremely long or memory overflows quickly. Since CUDD builds BDDs by joining smaller BDDs using logical operations, the trie is traversed bottom-up to successively generate code for all its subtrees. Algorithm \[alg:trie2bdd\] gives the details of this procedure. Two types of operations are used to combine nodes. The first creates conjunctions of leaf nodes and their parent if the leaf is a single child, the second creates disjunctions of all child nodes of a node if these child nodes are all leaves. In both cases, a subtree that occurs multiple times in the trie is translated only once, and the resulting BDD is used for all occurrences of that subtree. Because of the optimizations in CUDD, the resulting BDD can have a very different structure than the trie. The translation for query *path(a,d)* in our example graph is illustrated in Figure \[fig:trie2bdd\], it results in the following script: $$\begin{aligned}
n1 & = & ce \wedge ed\\
n2 & = & cd \vee n1\\
n3 & = & ac \wedge n2\\
n4 & = & bc \wedge n2\\
n5 & = & ab \wedge n4\\
n6 & = & n3 \vee n5\\
top & = & n6\end{aligned}$$
After CUDD has generated the BDD, the probability of a formula is calculated by traversing the BDD, in each node summing the probability of the high and low child, weighted by the probability of the node’s variable being assigned true and false respectively, cf. Algorithm \[alg:calcprob\]. Intermediate results are cached, and the algorithm has a time and space complexity linear in the size of the BDD.
If $n$ is the 1-terminal then return 1 If $n$ is the 0-terminal then return 0 let $h$ and $l$ be the high and low children of $n$ $prob(h) :=$ call <span style="font-variant:small-caps;">Probability</span>($h$) $prob(l) :=$ call <span style="font-variant:small-caps;">Probability</span>($l$) return $p_n \cdot prob(h) + (1-p_n) \cdot prob(l)$
For illustration, consider again Figure \[fig:BDD\]. The algorithm starts by assigning probabilities $0$ and $1$ to the $0$- and $1$-leaf respectively. The node labeled $ed$ has probability $0\ldotp5\cdot1+0\ldotp5\cdot0=0\ldotp5$, node $ce$ has probability $0\ldotp8\cdot0\ldotp5+0\ldotp2\cdot0=0\ldotp4$; finally, node $cd$, and thus the entire formula, has probability $0\ldotp9\cdot1+0\ldotp1\cdot0\ldotp4=0\ldotp94$.
Monte Carlo {#monte-carlo}
-----------
The Monte Carlo implementation is shown in Algorithm \[alg:mc\]. It receives a query $q$, an acceptance threshold $\delta_p$ and a constant $m$ determining the number of samples generated per iteration. At the end of each iteration, it estimates the probability $p$ as the fraction of programs sampled over all previous iterations that entailed the query, and the confidence interval width to be used in the stopping criterion as explained in Section \[sec:mc\_method\].
$c = 0$; $i = 0$; $p = 0$; $\delta = 1$; Generate a sample $P'$; $c:=c+1;$ $i:=i+1$; $p := c/i$ $\delta := 2\times\sqrt{\frac{p\cdot(p-1)}{i}}$ return $p$
Monte Carlo execution is quite different from the approaches discussed before, as the two main steps are **(a)** generating a sample program and **(b)** performing standard refutation on the sample. Thus, instead of combining large numbers of proofs, we need to manipulate large numbers of different programs or samples.
Our first approach was to generate a complete sample and to check for a proof. In order to accelerate the process, proofs were cached in a trie to skip inference on a new sample. If no proofs exist on a cache, we call the standard Prolog refutation procedure. Although this approach works rather well for small databases, it does not scale to larger databases where just generating a new sample requires walking through millions of facts.
We observed that even in large programs proofs are often quite short, i.e., we only need to verify whether facts from a small fragment of the database are in the sample. This suggests that it may be a good idea to take advantage of the independence between facts and generate the sample *lazily*: we verify whether a fact is in the sample only when we need it for a proof. YAP represents samples compactly as a three-valued array with one field for each fact, where $0$ means the fact was not yet sampled, $1$ it was already sampled and belongs to the sample, $2$ it was already sampled and does not belong to the sample. In this implementation:
1. New samples are generated by resetting the sampling array.
2. At every call to `add_to_proof`, given the current ProbLog literal $f$:
1. if $s[f] == 0 $, $s[f] = sample(f)$;
2. if $s[f] == 1$, succeed;
3. if $s[f] == 2$, fail;
Note that as fact identifiers are used to access the array, the approach cannot directly be used for non-ground facts. The current implementation of Monte Carlo therefore uses the internal database to store the result of sampling different groundings of such facts.
Experiments {#sec:experiments}
===========
We performed experiments with our implementation of ProbLog in the context of the biological network obtained from the Biomine project [@Sevon06]. We used two subgraphs extracted around three genes known to be connected to the Alzheimer disease (HGNC numbers 983, 620 and 582) as well as the full network. The smaller graphs were obtained querying Biomine for best paths of length 2 (resulting in graph <span style="font-variant:small-caps;">Small</span>) or all paths of length 3 (resulting in graph <span style="font-variant:small-caps;">Medium</span>) starting at one of the three genes. <span style="font-variant:small-caps;">Small</span> contains 79 nodes and 144 edges, <span style="font-variant:small-caps;">Medium</span> 5220 nodes and 11532 edges. We used <span style="font-variant:small-caps;">Small</span> for a first comparison of our algorithms on a small scale network where success probabilities can be calculated exactly. Scalability was evaluated using both <span style="font-variant:small-caps;">Medium</span> and the entire <span style="font-variant:small-caps;">Biomine</span> network with roughly 1,000,000 nodes and 6,000,000 edges. In all experiments, we queried for the probability that two of the gene nodes mentioned above are connected, that is, we used queries such as `path(’HGNC_983’,’HGNC_620’,Path)`. We used the following definition of an acyclic path in our background knowledge: $$\begin{array}{lll}
\mathtt{path(X,Y,A)} & \mathtt{:-} & \mathtt{path(X,Y,[X],A)},\\
\mathtt{path(X,X,A,A)\ldotp} & &\\
\mathtt{path(X,Y,A,R)} & \mathtt{:-} & \mathtt{X~\backslash ==~Y}, \\ & & \mathtt{edge(X,Z),} \\ & & \mathtt{absent(Z,A),} \\ & & \mathtt{path(Z,Y,[Z|A],R)\ldotp}\\
\end{array}$$ As list operations to check for the absence of a node get expensive for long paths, we consider an alternative definition for use in Monte Carlo. It provides cheaper testing by using the internal database of YAP to store nodes on the current path under key `visited`: $$\begin{array}{lll}
\mathtt{memopath(X,Y,A)} & \mathtt{:-} & \mathtt{eraseall(visited)}, \\ && \mathtt{memopath(X,Y,[X],A)\ldotp}\\
\mathtt{memopath(X,X,A,A)\ldotp} & &\\
\mathtt{memopath(X,Y,A,R)} & \mathtt{:-} & \mathtt{X~\backslash ==~Y}, \\ & & \mathtt{edge(X,Z),} \\ & & \mathtt{recordzifnot(visited,Z,\_),}\\
& & \mathtt{memopath(Z,Y,[Z|A],R)\ldotp}\\
\end{array}$$ Finally, to assess performance on the full network for queries with smaller probabilities, we use the following definition of paths with limited length: $$\begin{array}{lll}
\mathtt{lenpath(N,X,Y,Path)} & \mathtt{ :-} & \mathtt{lenpath(N,X,Y,[X],Path)\ldotp}\\
\mathtt{lenpath(N,X,X,A,A) } & \mathtt{ :-} & \mathtt{ N >= 0\ldotp}\\
\mathtt{lenpath(N,X,Y,A,P) } & \mathtt{ :-} & \mathtt{ X \backslash == Y},\\
&& \mathtt{ N > 0},\\
&& \mathtt{ edge(X,Z)},\\
&& \mathtt{ absent(Z,A)},\\
&& \mathtt{ NN\ is\ N-1},\\
&& \mathtt{ lenpath(NN,Z,Y,[Z|A],P)\ldotp}
\end{array}$$
All experiments were performed on a Core 2 Duo 2.4 GHz 4 GB machine running Linux. All times reported are in `msec` and do not include the time to load the graph into Prolog. The latter takes 20, 200 and 78140 `msec` for <span style="font-variant:small-caps;">Small</span>, <span style="font-variant:small-caps;">Medium</span> and <span style="font-variant:small-caps;">Biomine</span> respectively. Furthermore, as YAP indexes the database at query time, we query for the explanation probability of `path(’HGNC_620’,’HGNC_582’,Path)` before starting runtime measurements. This takes 0, 50 and 25900 `msec` for <span style="font-variant:small-caps;">Small</span>, <span style="font-variant:small-caps;">Medium</span> and <span style="font-variant:small-caps;">Biomine</span> respectively. We report $T_P$, the time spent by ProbLog to search for proofs, as well as $T_B$, the time spent to execute BDD programs (whenever meaningful). We also report the estimated probability $P$. For approximate inference using bounds, we report exact intervals for $P$, and also include the number $n$ of BDDs constructed. We set both the initial threshold and the shrinking factor to $0\ldotp5$. We computed $k$-probability for $k=1,2,\ldots,1024$. In the bounding algorithms, the error interval ranged between 10% and 1%. Monte Carlo recalculates confidence intervals after $m=1000$ samples. We also report the number $S$ of samples used.
#### Small Sized Sample
----------- ------ ----- ------ ------ ----- ------ ----- ----- ------
path
[**k**]{}
1 0 13 0.07 0 7 0.05 0 26 0.66
2 0 12 0.08 0 6 0.05 0 6 0.66
4 0 12 0.10 10 6 0.06 0 6 0.86
8 10 12 0.11 0 6 0.06 0 6 0.92
16 0 12 0.11 10 6 0.06 0 6 0.92
32 20 34 0.11 10 17 0.07 0 7 0.96
64 20 74 0.11 10 46 0.09 10 38 0.99
128 50 121 0.11 40 161 0.10 20 257 1.00
256 140 104 0.11 80 215 0.10 90 246 1.00
512 450 118 0.11 370 455 0.11 230 345 1.00
1024 1310 537 0.11 950 494 0.11 920 237 1.00
**exact** 670 450 0.11 8060 659 0.11 630 721 1.00
----------- ------ ----- ------ ------ ----- ------ ----- ----- ------
: $k$-probability on <span style="font-variant:small-caps;">Small</span>. []{data-label="tab:1"}
We first compared our algorithms on <span style="font-variant:small-caps;">Small</span>. Table \[tab:1\] shows the results for $k$-probability and exact inference. Note that nodes 620 and 582 are close to each other, whereas node 983 is farther apart. Therefore, connections involving the latter are less likely. In this graph, we obtained good approximations using a small fraction of proofs (the queries have 13136, 155695 and 16048 proofs respectively). Our results also show a significant increase in running times as ProbLog explores more paths in the graph, both within the Prolog code and within the BDD code. The BDD running times can vary widely, we may actually have large running times for smaller BDDs, depending on BDD structure. However, using SimpleCUDD instead of the C++ interface used in [@Kimmig08] typically decreases BDD time by at least one or two orders of magnitude.
Table \[tab:2\] gives corresponding results for bounded approximation. The algorithm converges quickly, as few proofs are needed and BDDs remain small. Note however that exact inference is competitive for this problem size. Moreover, we observe large speedups compared to the implementation with meta-interpreters used in [@DeRaedt07], where total runtimes to reach $\delta=0\ldotp01$ for these queries were 46234, 206400 and 307966 `msec` respectively. Table \[tab:3\] shows the performance of the Monte Carlo estimator. On <span style="font-variant:small-caps;">Small</span>, Monte Carlo is the fastest approach. Already within the first 1000 samples a good approximation is obtained.
The experiments on <span style="font-variant:small-caps;">Small</span> thus confirm that the implementation on top of YAP-Prolog enables efficient probabilistic inference on small sized graphs.
---------- ---------- --------------- ---------------- --------------- ------------- ---------------
path
$\delta$
0.10 0 48 4 \[0.07,0.12\] 10 74 6 \[0.06,0.11\] 0 25 2 \[0.91,1.00\]
0.05 0 71 6 \[0.07,0.11\] 0 75 6 \[0.06,0.11\] 0 486 4 \[0.98,1.00\]
0.01 0 83 7 \[0.11,0.11\] 140 3364 10 \[0.10,0.11\] 60 1886 6 \[1.00,1.00\]
---------- ---------- --------------- ---------------- --------------- ------------- ---------------
: Inference using bounds on <span style="font-variant:small-caps;">Small</span>. []{data-label="tab:2"}
---------- ------- ----- ------ ------- ----- ------ ------ ---- ------
path
$\delta$
0.10 1000 10 0.11 1000 10 0.11 1000 30 1.00
0.05 1000 10 0.11 1000 10 0.10 1000 20 1.00
0.01 16000 130 0.11 16000 170 0.11 1000 30 1.00
---------- ------- ----- ------ ------- ----- ------ ------ ---- ------
: Monte Carlo Inference on <span style="font-variant:small-caps;">Small</span>. []{data-label="tab:3"}
#### Medium Sized Sample
For graph <span style="font-variant:small-caps;">Medium</span> with around 11000 edges, exact inference is no longer feasible. Table \[tab:1a\] again shows results for the $k$-probability. Comparing these results with the corresponding values from Table \[tab:1\], we observe that the estimated probability is higher now: this is natural, as the graph has both more nodes and is more connected, therefore leading to many more possible explanations. This also explains the increase in running times. Approximate inference using bounds only reached loose bounds (with differences $>0\ldotp 2$) on queries involving node `’HGNC_983’`, as upper bound formulae with more than 10 million conjunctions were encountered, which could not be processed.
The Monte Carlo estimator using the standard definition of `path/3` on <span style="font-variant:small-caps;">Medium</span> did not complete the first $1000$ samples within one hour. A detailed analysis shows that this is caused by some queries backtracking too heavily. Table \[tab:3a\] therefore reports results using the memorising version `memopath/3`. With this improved definition, Monte Carlo performs well: it obtains a good approximation in a few seconds. Requiring tighter bounds however can increase runtimes significantly.
----------- ------ ---- ------ ------ --- ------ ------ ------ ------
path
[**k**]{}
1 180 6 0.33 1620 6 0.30 10 6 0.92
2 180 6 0.33 1620 6 0.30 20 6 0.92
4 180 6 0.33 1630 6 0.30 10 6 0.92
8 220 6 0.33 1630 6 0.30 20 6 0.92
16 260 6 0.33 1660 6 0.30 30 6 0.99
32 710 6 0.40 1710 7 0.30 110 6 1.00
64 1540 7 0.42 1910 6 0.30 200 6 1.00
128 1680 6 0.42 2230 6 0.30 240 9 1.00
256 2190 7 0.55 2720 6 0.49 290 196 1.00
512 2650 7 0.64 3730 7 0.53 1310 327 1.00
1024 8100 41 0.70 5080 8 0.56 3070 1357 1.00
----------- ------ ---- ------ ------ --- ------ ------ ------ ------
: $k$-probability on <span style="font-variant:small-caps;">Medium</span>. []{data-label="tab:1a"}
---------- ------- ------- ------ ------- ------- ------ ------ ------ ------
memo
$\delta$
0.10 1000 1180 0.78 1000 2130 0.76 1000 1640 1.00
0.05 2000 2320 0.77 2000 4230 0.74 1000 1640 1.00
0.01 29000 33220 0.77 29000 61140 0.77 1000 1670 1.00
---------- ------- ------- ------ ------- ------- ------ ------ ------ ------
: Monte Carlo Inference using `memopath/3` on <span style="font-variant:small-caps;">Medium</span>. []{data-label="tab:3a"}
#### Biomine Database
The Biomine Database covers hundreds of thousands of entities and millions of links. On <span style="font-variant:small-caps;">Biomine</span>, we therefore restricted our experiments to the approximations given by $k$-probability and Monte Carlo. Given the results on <span style="font-variant:small-caps;">Medium</span>, we directly used `memopath/3` for Monte Carlo. Tables \[tab:1c\] and \[tab:3b\] show the results on the large network. We observe that on this large graph, the number of possible paths is tremendous, which implies success probabilities practically equal to 1. Still, we observe that ProbLog’s branch-and-bound search to find the best solutions performs reasonably also on this size of network. However, runtimes for obtaining tight confidence intervals with Monte Carlo explode quickly even with the improved path definition.
----------- --------- -------- ------ ----------- --------- ------ --------- -------- ------
path
[**k**]{}
1 5,760 49 0.16 8,910 48 0.11 10 48 0.59
2 5,800 48 0.16 10,340 48 0.17 180 48 0.63
4 6,200 48 0.16 13,640 48 0.28 360 48 0.65
8 7,480 48 0.16 15,550 49 0.38 500 48 0.66
16 11,470 49 0.50 58,050 49 0.53 630 48 0.92
32 15,100 49 0.57 106,300 49 0.56 2,220 167 0.95
64 53,760 84 0.80 146,380 101 0.65 3,690 167 0.95
128 71,560 126 0.88 230,290 354 0.76 7,360 369 0.98
256 138,300 277 0.95 336,410 520 0.85 13,520 1,106 1.00
512 242,210 730 0.98 501,870 2,744 0.88 23,910 3,444 1.00
1024 364,490 10,597 0.99 1,809,680 100,468 0.93 146,890 10,675 1.00
----------- --------- -------- ------ ----------- --------- ------ --------- -------- ------
: $k$-probability on <span style="font-variant:small-caps;">Biomine</span>. []{data-label="tab:1c"}
---------- ------ --------- ------ ------ ----------- ------ ------ ----------- ------
memo
$\delta$
0.10 1000 100,700 1.00 1000 1,656,660 1.00 1000 1,696,420 1.00
0.05 1000 100,230 1.00 1000 1,671,880 1.00 1000 1,690,830 1.00
0.01 1000 93,120 1.00 1000 1,710,200 1.00 1000 1,637,320 1.00
---------- ------ --------- ------ ------ ----------- ------ ------ ----------- ------
: Monte Carlo Inference using `memopath/3` on <span style="font-variant:small-caps;">Biomine</span>. []{data-label="tab:3b"}
Given that sampling a program that does not entail the query is extremely unlikely for the setting considered so far, we performed an additional experiment on <span style="font-variant:small-caps;">Biomine</span>, where we restrict the number of edges on the path connecting two nodes to a maximum of 2 or 3. Results are reported in Table \[tab:shortpath\]. As none of the resulting queries have more than 50 proofs, exact inference is much faster than Monte Carlo, which needs a higher number of samples to reliably estimate probabilities that are not close to $1$.
---------- ------- ----------- ------ ------- ----------- ------ ------- --------- ------
len
$\delta$
0.10 1000 21,400 0.04 1000 18,720 0.11 1000 19,150 0.58
0.05 1000 19,770 0.05 1000 20,980 0.10 2000 35,100 0.55
0.01 6000 112,740 0.04 16000 307,520 0.11 40000 764,700 0.55
exact - 477 0.04 - 456 0.11 - 581 0.55
0.10 1000 106,730 0.14 1000 105,350 0.33 1000 45,400 0.96
0.05 1000 107,920 0.14 2000 198,930 0.34 1000 49,950 0.96
0.01 19000 2,065,030 0.14 37000 3,828,520 0.35 6000 282,400 0.96
exact - 9,413 0.14 - 9,485 0.35 - 15,806 0.96
---------- ------- ----------- ------ ------- ----------- ------ ------- --------- ------
: Monte Carlo inference for different values of $\delta$ and exact inference using `lenpath/4` with length at most $2$ (top) or $3$ (bottom) on <span style="font-variant:small-caps;">Biomine</span>. For exact inference, runtimes include both Prolog and BDD time.[]{data-label="tab:shortpath"}
Altogether, the experiments confirm that our implementation provides efficient inference algorithms for ProbLog that scale to large databases. Furthermore, compared to the original implementation of [@DeRaedt07], we obtain large speedups in both the Prolog and the BDD part, thereby opening new perspectives for applications of ProbLog.
Conclusions {#sec:conclusion}
===========
ProbLog is a simple but elegant probabilistic logic programming language that allows one to explicitly represent uncertainty by means of probabilistic facts denoting independent random variables. The language is a simple and natural extension of the logic programming language Prolog. We presented an efficient implementation of the ProbLog language on top of the YAP-Prolog system that is designed to scale to large sized problems. We showed that ProbLog can be used to obtain both explanation and (approximations of) success probabilities for queries on a large database. To the best of our knowledge, ProbLog is the first example of a probabilistic logic programming system that can execute queries on such large databases. Due to the use of BDDs for addressing the disjoint-sum-problem, the initial implementation of ProbLog used in [@DeRaedt07] already scaled up much better than alternative implementations such as Fuhr’s pD engine HySpirit [@Fuhr00]. The tight integration in YAP-Prolog presented here leads to further speedups in runtime of several orders of magnitude.
Although we focused on connectivity queries and Biomine in this work, similar problems are found across many domains; we believe that the techniques presented apply to a wide variety of queries and databases because ProbLog provides a clean separation between background knowledge and what is specific to the engine. As shown for Monte Carlo inference, such an interface can be very useful to improve performance as it allows incremental refinement of background knowledge, e.g., graph procedures. Initial experiments with Dijkstra’s algorithm for finding the explanation probability are very promising.
ProbLog is closely related to some alternative formalisms such as PHA and ICL [@Poole:93; @Poole00], pD [@Fuhr00] and PRISM [@SatoKameya:01] as their semantics are all based on Sato’s distribution semantics even though there exist also some subtle differences. However, ProbLog is – to the best of the authors’ knowledge – the first implementation that tightly integrates Sato’s original distribution semantics [@Sato:95] in a state-of-the-art Prolog system without making additional restrictions (such as the exclusive explanation assumption made in PHA and PRISM). As ProbLog, both PRISM and the ICL implementation AILog2 use a two-step approach to inference, where proofs are collected in the first phase, and probabilities are calculated once all proofs are known. AILog2 is a meta-interpreter implemented in SWI-Prolog for didactical purposes, where the disjoint-sum-problem is tackled using a symbolic disjoining technique [@Poole00]. PRISM, built on top of B-Prolog, requires programs to be written such that alternative explanations for queries are mutually exclusive. PRISM uses a meta-interpreter to collect proofs in a hierarchical datastructure called explanation graph. As proofs are mutually exclusive, the explanation graph directly mirrors the sum-of-products structure of probability calculation [@SatoKameya:01]. ProbLog is the first probabilistic logic programming system using BDDs as a basic datastructure for probability calculation, a principle that receives increased interest in the probabilistic logic learning community, cf. for instance [@Riguzzi; @sato:ilp08].
Furthermore, as compared to SLPs [@Muggleton96], CLP($\cal BN$) [@Costa03:uai], and BLPs [@Kersting08], ProbLog is a much simpler and in a sense more primitive probabilistic programming language. Therefore, the relationship between probabilistic logic programming and ProbLog is, in a sense, analogous to that between logic programming and Prolog. From this perspective, it is our hope and goal to further develop ProbLog so that it can be used as a general purpose programming language with an efficient implementation for use in statistical relational learning [@Getoor07] and probabilistic programming [@DeRaedt-PILPbook]. One important use of such a probabilistic programming language is as a target language in which other formalisms can be efficiently compiled. For instance, it has already been shown that CP-logic [@Vennekens], a recent elegant probabilistic knowledge representation language based on a probabilistic extension of clausal logic, can be compiled into ProbLog [@Riguzzi] and it is well-known that SLPs [@Muggleton96] can be compiled into Sato’s PRISM, which is closely related to ProbLog. Further evidence is provided in [@DeRaedt-NIPSWS08].
Another, related use of ProbLog is as a vehicle for developing learning and mining algorithms and tools [@Kimmig07; @DeRaedt08MLJ; @Gutmann08; @Kimmig09; @DeRaedt-IQTechReport]. In the context of probabilistic representations [@Getoor07; @DeRaedt-PILPbook], one typically distinguishes two types of learning: parameter estimation and structure learning. In parameter estimation in the context of ProbLog and PRISM, one starts from a set of queries and the logical part of the program and the problem is to find good estimates of the parameter values, that is, the probabilities of the probabilistic facts in the program. [@Gutmann08] introduces a gradient descent approach to parameter learning for ProbLog that extends the BDD-based methods discussed here. In structure learning, one also starts from queries but has to find the logical part of the program as well. Structure learning is therefore closely related to inductive logic programming. The limiting factor in statistical relational learning and probabilistic logic learning is often the efficiency of inference, as learning requires repeated computation of the probabilities of many queries. Therefore, improvements on inference in probabilistic programming implementations have an immediate effect on learning. The above compilation approach also raises the interesting and largely open question whether not only inference problems for alternative formalisms can be compiled into ProbLog but whether it is also possible to compile learning problems for these logics into learning problems for ProbLog.
Finally, as ProbLog, unlike PRISM and PHA, deals with the disjoint-sum-problem, it is interesting to study how program transformation and analysis techniques could be used to optimize ProbLog programs, by detecting and taking into account situations where some conjunctions are disjoint. At the same time, we currently investigate how tabling, one of the keys to PRISM’s efficiency, can be incorporated in ProbLog [@Mantadelis09; @Kimmig-SRL09].
### Acknowledgements {#acknowledgements .unnumbered}
We would like to thank Hannu Toivonen, Bernd Gutmann and Kristian Kersting for their many contributions to ProbLog, the Biomine team for the application, and Theofrastos Mantadelis for the development of SimpleCUDD. This work is partially supported by the GOA project 2008/08 Probabilistic Logic Learning. Angelika Kimmig is supported by the Research Foundation-Flanders (FWO-Vlaanderen). Vítor Santos Costa and Ricardo Rocha are partially supported by the research projects STAMPA (PTDC/EIA/67738/2006) and JEDI (PTDC/ EIA/66924/2006) and by Fundação para a Ciência e Tecnologia.
[^1]: If the program contains multiple instances of the same fact, they correspond to different random variables, i.e. $\{p::c\}$ and $\{p::c, p::c\}$ are different ProbLog programs.
[^2]: For better readability, we do not write substitutions explicitly here.
[^3]: Using a probability threshold instead of the depth bound of [@DeRaedt07] has been found to speed up convergence, as upper bounds have been found to be tighter on initial levels.
[^4]: We use the logarithm to avoid numerical problems when calculating the probability of a derivation, which is used to drive inference.
[^5]: <http://www.cs.kuleuven.be/~theo/tools/simplecudd.html>
[^6]: <http://vlsi.colorado.edu/~fabio/CUDD>
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Two types of population models are well known – the continuous and the discrete types.The two have very different characteristics and methods of solutions and analysis.In this note, we point out that an iterative technique when applied to the continuous case mimics, surprisingly the discrete theory. The implication is that techniques and conclusions of the latter theory can now be applied to the former case (and vice versa).'
author:
- 'B.G.Sidharth and B.S.Lakshmi[^1]'
title: '**A Link Between The Continuous And The Discrete Logistic Equations**'
---
Discrete Logistic Equation
==========================
Population growth in nature is seldom as smoothly continuous as a classical logistic curve suggests [@freedman]. In a species with a short annual breeding season whose members live for several breeding seasons and die at any time of the year, a continuous record of population size would undoubtedly show seasonal undulations. For many species in fact population growth is markedly discontinuous. These are, for example species whose members reproduce only once in their lifetime and die before their descendants’ lives begin, that is each generation dies before the eggs are hatched,or the seeds are germinated, to form the successive generations. A continuous-time differential equation is then inappropriate as a representation of population growth.We have to consider instead difference equations. As this will be needed in the sequel, we touch upon the salient features.
First-Order Difference Equation
-------------------------------
One of the simplest systems an ecologist can study is a seasonal breeding population in which generations do not overlap. Many natural populations,particularly among temperate zone insects (including the many economically important crop and orchard pests) are of this kind. In this situation, the observed data will consist of information about the maximum or the average, or the total population in each generation [@hallam]. The studies try to form a relation between the magnitude of the population in the generation **n** represented by $x_{n}$ and the magnitude of the population in generation **n+1** represented by $x_{n+1}$ such a relation may be expressed in the general form $$\label{may1}
x_{n+1}= F(x_{n})$$ The function $ F(x_{n})$ will be what the biologist calls “density dependant", and a mathematician calls non-linear [@may]. Equation (\[may1\]) is thus a first order, non-linear difference equation. Equation (\[may1\]) also describes many other examples in biology apart from population growth.\
For instance:\
1. In genetics, where the equation describes the change in gene frequency in time.\
2. In epidemiology where $x_{n}$ represents the fraction of population infected in time n.\
3. In economics where the relationships between commodity quantity and price are studied.\
4. In social sciences to study the propagation of rumors where $x_{n}$ could be the number of people to have heard the rumor after time t.\
In many of these contexts, and for biological populations in particular, there is a tendency for the variable $ x_{n}$ to increase from one generation to the next when it is small, and for it to decrease when it is large. That is, the nonlinear function $F( x_{n}) $ often has the following properties: $ F(0)=0$ and $F(x)$ increases monotonically as x increases through the range $0<x< A$ with F(x) attaining its maximum value at x = A, and F(x) decreases monotonically as x increases beyond x = A. A specific example is afforded by the equation $$\label{may2}
F( x_{n})=x_{n+1}= x_{n}(a-b x_{n})$$
This is sometimes referred to as the logistic difference equation. Using the substitution $x_{n}=b/a x_{n}$ Equation (\[may2\]) can be written as $$\label{may3}
F( x_{n})=x_{n+1}=a x_{n}(1- x_{n})$$ The behavior of the solutions of (\[may3\]) is a function of the parameter ‘a’ Equation (\[may3\]) has meaningful solutions for $$0 \leq a\leq 4$$ with $x_{n}$ measuring a non-negative quantity.\
Studies have shown [@may] that the very simple nonlinear difference equation can possess an extraordinarily rich spectrum of dynamical behavior, from stable points, stable cycles to ultimately chaotic behavior.Thus the problem is far richer than the continuous case seen earlier. In this form equation (\[may3\]) is a simple nonlinear difference equation.
Examples of chaotic dynamical systems : The logistic map {#chatd}
========================================================
The discrete logistic map described by the single difference equation $$x_{n+1}=a x_{n}(1-x_{n}) \label{roderick1}$$ as mentioned earlier, determines the future value of the variable $x_{n+1}$ at time-step n+1 from the past value at time-step n. The time evolution of $x_{n}$ generated by this algebraic equation exhibits an extraordinary transformation from order to chaos as the parameter a, which measures the strength of nonlinearity is increased [@rod].\
Although nonlinear difference equations of this type have been studied extensively as simple models for turbulence in fluids, they also arise naturally in the study of evolution of biological populations.\
For the purpose of illustration we consider the population of gypsy moths in the northern United States, which exhibits wild and unpredictable fluctuations from year to year. However we could equally well consider the evolution of economic prices determined by a nonlinear web model.\
Writing (\[roderick1\]) in a slightly different form $$x_{n+1}=a x _{n}-a x^{2}_{n}$$ we see that it is a simple quadratic equation, with the first term linear and the second term nonlinear. If the parameter $a>1$, the population increases, if $a<1$, the population decreases. If $a>1$, the population will eventually grow to a large enough value for the nonlinear term $-ax^{2}_{n}$, to become important. Since this term is negative, it represents a nonlinear death rate which dominates when the population is too large. Biologically this nonlinear death rate could be due to the depletion of food supplies or the outbreak of diseases in an over-crowded environment. The dynamics of this map and the dependence on the parameter a which measures the rate of linear growth and the size of the nonlinear term, are best understood using graphical analysis. Consider the graphs of $x_{n}$ versus $x_{n+1}$displayed in the Fig.(\[dis1\]) for four different values of a.
![Graphs of $x_n$ versus $x_{n+1}$ for different values of $a$ []{data-label="dis1"}](discrete.eps)
Equation (\[roderick1\]) defines an inverted parabola with intercepts at $$x_{n}=0 \; and \; 1$$ and a maximum value of $$x_{n+1}=
a/4$$ at $$x_{n}=0.5$$.\
Using these maps we can we can get a quantitative understanding of the dynamics of the logistic map in a quick way.\
Briefly, this graphical analysis tells us that if the normalized population starts out larger than 1, then it immediately goes negative , becoming extinct in one time-step. Moreover if $a>4$, the peak of the parabola will exceed 1, which makes it possible for initial populations near $0.5$ to become extinct in two time-steps. We will therefore restrict our analysis to values of $a$ between 0 and 4.\
For values of $a<1$, the population always decreases to 0, as shown for $a=0.95$ in Fig.(\[dis1\]) The intersection of the parabola with the $45^{\tiny{o}}$ line at $x_{n}=0$ represents a stable fixed point on the map. Because $a$ is small a perturbation can be used to verify that almost all initial populations are attracted to this fixed point and become extinct. However for $a>1$ this fixed point becomes unstable. Instead the parabola now intersects the $45^{\tiny{o}}$ line at $$x= \frac{a-1}{a}$$which corresponds to a new fixed point.\
For values of $a$ between 1 and 3 almost all initial populations evolve to this equilibrium population. Then, as $a$ is increased between 3 and 4 , the dynamics change in remarkable ways. First the fixed point becomes unstable and the population evolves to a dynamic steady state in which it alternates between a large and small population. A time sequence converging to such a period-2 cycle is displayed in Fig.(\[dis1\]) for a= 3.2 : the population cycles between two points on the parabola, $x_{n}\sim
0.5$ and $x_{n}\sim 0.8$, in alternate years. For somewhat larger values of $a$ , this period-2 cycle becomes unstable and is replaced by a period-4 cycle in which the population alternates high-low, returning to its original value every four time-steps. As $a$ is increased the longtime motion converges to period – 8,16,32,64, cycles, finally accumulating to a cycle of infinite period $a=a_{\tiny{inf}}\sim 3.57$\
Having observed a period doubling sequence in numerical experiments Feigenbaum was able to prove that the intervals over which a cycle is stable decreases at a geometric rate of $\sim 4.6692016$. The tremendous significance of this work is that this rate and other properties of the period-doubling bifurcation are universal in the sense that they appear in the dynamics of any system which can be approximately modelled by a nonlinear map with a quadratic extremum. Feigenbaum’s theory has subsequently been confirmed by a wide variety of physical systems such as turbulent fluids, oscillating chemical reactions, nonlinear electrical circuits, and ring lasers.\
The investigation of period doubling in nonlinear dynamical systems provides a superb example of the interplay between numerical experiments and analytical theory. However, this sequence of regular periodic orbits is only the precursor to chaos. Included below is a bifurcation diagram , showing the beginning of Chaos:
A Remarkable Mathematical Equivalence
=====================================
We would now like to deduce a mathematical equivalence between the continuous and discrete cases and will comment on this. Let us consider the equation from the continuous growth $$\label{diseq1}\frac{dP}{dt} = f(P), \hspace{2in} set \; \; P \equiv x$$ which gives $$\label{diseq2} x = \int f(x(t)) dt$$ where the integral is over suitable limits. Here f generalizes the dependence of the right side of $$\label{logisticeq} \frac{d(P(t))}{dt}=r(M-P(t))P(t)
\hspace{2in} r > 0$$ where P(t) is the population at given time t and M is the maximum sustainable population [@r1; @r2] on the population P at that time. Let us solve equation (\[diseq2\]) by the method of successive approximation ie., we try on the right side of (\[diseq2\]) a tentative solution $x_{0}(t)$. (\[diseq2\]) can then be written in the form $$\label{diseq3} x_1 = F(x_0)\equiv \int f(x_0)dt$$ where $x_1$ gives the next level of approximation. Before proceeding further, we remark that (\[diseq1\]) is in the form of the initial value problem,where the Lipschitz condition is necessary for the convergence of the iterative procedure [@coddington]. In a similar manner we get from (\[diseq3\]) the more general equation $$\label{diseq4} x_{n+1} = F(x_{n}).$$ (\[diseq4\]) can immediately identified with the discrete logistic equation for example (\[roderick1\]). It must be stressed however that the discrete equation is based on a completely different foundation that is the subscript $n$ in the discrete case represents the population at the $n^{th}$ generation, whereas in (\[diseq4\]) $x_n$ represents the $n^{th}$ iteration or approximation of the population $x \equiv P$ of the continuous case. Nevertheless, this mathematical equivalence enables us to apply the conclusions of the discrete case including the domain of chaos. Thus chaotic behaviour of $x_n$ of the discrete case would represent the lack of convergence of the iterates of the continuous case. In this specific example if $$F(x_n) = a x_n (1-x_n),\label{diseq5}$$ then for $a$ = 3.57 the above iterative procedure breaks down.In this case, using (\[diseq5\]), it follows from (\[diseq3\]) that $$\int f(x) dt = a x (1-x).$$ Finally we remark that for a more conventional approach to the above problem reference can be made to Krempasky, [@krem].
[99]{} Freedman,H.I.,“ Deterministic Mathematical Models in Population Ecology” Pp.,6 ff Thomas G.Hallam and Simon A.Levin“Mathematical Ecology”,Springer-Verlag,New York 1980 May,R.M,“Simple mathematical models with very complicated dynamics” Nature Vol.,261 June 10 1976. Roderick Jensen,“Classical Chaos”,American Scientist,Vol 75,March-April,1987 W.S.Weigelhofer and Kenneth A.Lindsay “Ordinary Differential Equations and Applications”,Albion Publishing, Chichester,1999 pp 13 – 16. H.I.Freedman “ Deterministic Mathematical Models in Population Ecology” Marcel Dekker Inc.pp–5ff E.A.Coddington“An Introduction to Ordinary Differential Equations”,Prentice-Hall,1961 Englewood Cliffs NJ. http://www.fch.vutbr.cz/udalosti/mol/sbornik98/krempasky/Index.html
[^1]: Address for communication :International Institute for Applicable Mathematics and Information Sciences, B.M. Birla Science Centre, Adarsh Nagar, Hyderabad - 500 063(India)
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Dissipative relativistic fluid-dynamical descriptions of the extended fireball formed in high-energy heavy-ion collisions are quite successful, yet require a prescription for converting the fluid into particles. We present arguments in favour of using a locally anisotropic momentum distribution for the particles emitted from the fluid, so as to smooth out discontinuities introduced by the usual conversion prescriptions. Building on this ansatz, we investigate the effect of the asymmetry on several observables of heavy ion physics.'
author:
- Nicolas Borghini
- Steffen Feld
- Christian Lang
title: 'Kinetic freeze out from an anisotropic fluid in high-energy heavy-ion collisions: particle spectra, Hanbury Brown–Twiss radii, and anisotropic flow'
---
Introduction
============
A large amount of the dynamical properties of the fireball created in high energy collisions of heavy nuclei—be it at the Brookhaven Relativistic Heavy Ion Collider (RHIC) or at the CERN Large Hadron Collider (LHC)—can be described to a good approximation within the framework of relativistic fluid dynamics (see Ref. [@Huovinen:2013wma] for a critical review). The relevant equations of motion have to be supplemented with appropriate initial conditions for the continuous medium and with a recipe for the end of its evolution, namely the conversion of the fluid into particles [@Huovinen:2012is]. Our focus in this work will be on the latter point and on how adopting a new ansatz for the transition can help mitigate a few issues in the usual approaches.
After their emission from the fluid—in which the mean free path is assumed to be very small—, the particles can be left to propagate freely, i.e. they at once acquire an infinitely large mean free path: the conversion step is the so-called (kinetic) freeze out, after which the particle momenta no longer evolve. Alternatively, the emitted particles can be fed into an “afterburner” that implements a set of transport equations for the various particle species and thereby ensures a more gradual change of the mean free path (see Ref. [@Hirano:2012kj] for a recent overview). The fluid–particle transition is then a switch between model descriptions, rather than a physical phenomenon.
Irrespective of the subsequent fate of the particles, their emission from the fluid is often modelled in a similar way. For that reason, we shall generally for the sake of simplicity refer to the conversion process as “freeze out”, although the actual decoupling occurs later when the particles are evolved with a transport code.
Most existing studies follow some version of the Cooper–Frye prescription [@Cooper:1974mv]: the fluid breaks up, more or less suddenly, when reaching a three-dimensional hypersurface $\Sigma$—sometimes replaced by a thin four-dimensional shell, to mitigate the inherent abruptness of the description—defined by some a priori criterion, like a constant temperature or energy density. At each point on the freeze-out hypersurface, particles are emitted with a given phase space distribution $f(\mathsf{x},\mathsf{p})$.[^1] Integrating over the whole hypersurface, the resulting invariant spectrum of the emitted particles of type $i$ reads $$\label{Cooper-Frye}
E_{\bm{p}}\frac{{\mathrm{d}}^3N_i}{{\mathrm{d}}^3\bm{p}} =
\frac{g_i}{(2\pi)^3}\!\int_\Sigma\!f_i(\mathsf{x},\mathsf{p})\, p^\mu{\mathrm{d}}^3\sigma_\mu(\mathsf{x}).$$ In this expression, we have taken into account the fact that the phase space distribution depends on the particle species, especially its bosonic or fermionic nature, and we explicitly factorized out the particle degeneracy factor $g_i$.
The phase space occupation factor in the Cooper–Frye formula is chosen so as to ensure the conservation of energy, momentum, and charges—if any—across the freeze-out hypersurface. Accordingly, $f_i(\mathsf{x},\mathsf{p})$ is usually taken to be the equilibrium thermal distribution—which is appropriate for a perfect fluid—or a near-equilibrium distribution including “correction terms” that match the stress energy tensor of a dissipative fluid. Determining these corrections either from pure theory [@Teaney:2003kp; @Dusling:2007gi; @Denicol:2009am; @Monnai:2009ad; @Dusling:2009df; @Pratt:2010jt; @Dusling:2011fd; @Teaney:2013gca; @Molnar:2014fva] or within more phenomenological data-driven approaches [@Luzum:2010ad; @Lang:2013oba] is an ongoing effort. In any case, it is always implicitly assumed that the decoupling medium is not far from local equilibrium, so that dissipative effects remain small.
Thus, the occupation factors $f_i$ at freeze out considered in the literature are, up to small corrections, isotropic in the fluid local rest frame, reflecting the assumption of (near) local thermal equilibrium. Accordingly, $f_i$ depends on position only through the corresponding dependence of thermodynamic variables, namely the flow velocity $\mathsf{u}(\mathsf{x})$ and its gradients, the freeze-out temperature $T_{\mathrm{f.o.}}(\mathsf{x})$, and possibly the chemical potential $\mu_i(\mathsf{x})$.
In this work, we shall depart from this local isotropy of $f_i$ and assume instead a locally asymmetric momentum distribution at decoupling.
Before proceeding any further, let us mention that the existence of some local momentum anisotropy at freeze out was already considered in Ref. [@Rybczynski:2012ee]. As will become clear in the following section, the anisotropy we are interested in is of a different kind, reflecting the dissimilar underlying motivation. Nevertheless, some of the findings of Ref. [@Rybczynski:2012ee] naturally translate into similar results in our case.
Motivation
==========
The sudden-decoupling scenario embodied in the Cooper–Frye formula aims at gluing together two rather different descriptions. The mismatch of the models is obvious if the fluid freezes out into free-streaming particles, as exemplified by the jump of the Knudsen number from very small to very large values. Even when the Cooper–Frye prescription is used to switch from a dissipative fluid to a collection of interacting hadrons, there remain issues [@Huovinen:2012is; @Hirano:2012kj]. An often mentioned problem is the existence of sectors of the conversion hypersurface $\Sigma$ where ${\mathrm{d}}\sigma_\mu(\textsf{x})\,{\mathrm{d}}\sigma^\mu(\textsf{x})<0$, which can locally lead to negative contributions to the Cooper–Frye integral. Cures to this issue have been proposed (see e.g. Ref. [@Grassi:2004dz] and references therein), which themselves remain incomplete since they introduce discontinuities across $\Sigma$ either in the stress energy tensor or in the velocity. These shocks are however artefacts of the modelling, not physical ones.
Another issue of the usual sudden freeze out recipe is the sensitivity of the observables computed with the emitted particles, in particular their spectra, to the parameters in the Cooper–Frye formula. This is in our eyes a rather crucial point: it means that the matching between a “microscopic” approach and a long-wavelength effective theory thereof, namely the kinetic modelling in terms of particles and the fluid-dynamical description, depends significantly on the parameter that separates them, which makes the whole procedure questionable.
A strong theoretical incentive for developing and investigating new approaches to the modelling of decoupling at the end of the dynamics of heavy-ion collisions is thus to obtain a description which interpolates between the hydrodynamic and ballistic regimes in a smoother manner than the usual prescriptions.
A possible way out of the problem is to drop the assumption of a sudden freeze out in favour of a continuous one . However, in the current implementations of this approach, the particles decoupling from the fluid do not reinteract with each other afterwards. This again implies for each particle a sudden transition from a vanishingly small to an infinitely large mean free path—where the latter is viewed somewhat abusively as the average length that a given particle is likely to travel in its next step—, which is again unsatisfactory, even though this does not happen at once for the whole fluid.
Despite its deficiencies, the “naive” Cooper–Frye formula remains attractive because of its simplicity, which makes it easier to test novel ideas. In order to ensure a better transition between the fluid and particle description, it seems desirable to “twist” one of the models or both, so as to bring them closer to each other. In this spirit, we suggest that anisotropic hydrodynamics [@Florkowski:2010cf; @Martinez:2010sc; @Ryblewski:2013jsa] can improve the smoothness of the transition between the continuous and particle frameworks. As we shall demonstrate in next section, this ansatz helps alleviating the sensitivity to the freeze-out temperature $T_{\mathrm{f.o.}}$: Introducing new control parameters, namely those governing the anisotropy of the phase-space distribution at decoupling, widens the possible range of values for $T_{\mathrm{f.o.}}$. In the remainder of this section, we list a few arguments in favour of distorting the particle distribution at freeze out.
First, in the context of heavy-ion physics there is an obvious analogy with the advocated use of anisotropic hydrodynamics at early stages of the medium evolution, to ease the transition from the locally asymmetric energy-momentum tensor of the fields left by the colliding nuclei to the almost isotropic tensor needed to apply usual hydrodynamics consistently. In the early evolution stage, the phase-space distribution is deformed along the axis of the nucleus–nucleus collision ($z$-axis), while in the case we are interested here we do not expect such a global direction for the anisotropy.
As a matter of fact, our second incentive to resort to a possibly strongly anisotropic freeze-out distribution is the observation of a similar asymmetry, parametrized as two different translation temperatures along the streamlines and perpendicular to them, in hypersonic nonrelativistic flows [@Hamel:1966]. These findings help us specify the kind of anisotropy we want to consider hereafter. Let us for simplicity focus on particle emission around midrapidity, so as to discard any anisotropy along the $z$-direction on symmetry grounds. Far from the fluid, each particle will tend to fly away radially, as implied by the simultaneous conservation of angular momentum and (kinetic) energy. The dispersion of the momentum components transverse to the radial direction will thus be much smaller than that of the radial component.
Eventually, a third argument for assuming a deformed particle distribution is that such an anisotropy was actually found for post-freeze-out distributions arising from the decoupling through time-like portions of freeze-out hypersurfaces [@Gorenstein:1983bu; @Molnar:2005gy].
Accordingly, we conclude that it would be helpful to adopt in the Cooper–Frye picture a freeze-out distribution which is already deformed, with a larger mean squared momentum along the radial direction. That is, adopting the Cartesian (out, side, long) system of femtoscopic studies, we assume a larger pressure along the local “out”-direction than in the sidewards and longitudinal directions.
In the present paper, this asymmetry is admittedly a mere assumption, motivated by the observations in non-relativistic studies in which freeze out happens when the local particle distribution has a sizeable anisotropy in momentum space, and by the incentive to have a smoother transition between the fluid and particle descriptions. The actual functional form of the phase-space distribution at freeze out, as well as the size of the parameters measuring the anisotropy, should emerge from a detailed kinetic description of the decoupling process [@BFL_inprep]. In next section, we shall postulate such a form and examine the change induced by the momentum-space asymmetry on various observables of heavy-ion collisions.
Note that the anisotropy we consider hereafter differs from that considered in Ref. [@Rybczynski:2012ee], in which the distribution is assumed to be distorted along the $z$-axis, as a remnant of the distortion along that direction in the initial state of the nucleus–nucleus collision. Both deformations can naturally be present at once, yet our purpose here is to examine the influence of a larger radial-momentum dispersion, so that we keep the pressures in the side- and long-directions equal.
Effect of the local anisotropy on observables {#s:effect_xi}
=============================================
Let us assume for the phase-space distribution at decoupling of a particle species with mass $m$ a Romatschke–Strickland-like profile [@Romatschke:2003ms], namely $$\label{f_an_LR}
f_{\!\!\!\!\begin{array}{c} \textrm{\scriptsize an.} \\[-2mm] \textrm{\tiny (lrf)} \end{array}\!\!\!\!}(\textsf{x},\textsf{p};\Lambda,\xi) =
\bigg[\exp\!\bigg(\!\frac{\sqrt{m^2+\bm{p}^{\prime 2} +
\xi(\textsf{x})_{} p_{\textrm{out}}^{\prime 2}}}{\Lambda(\textsf{x})}\bigg) \mp 1 \bigg]^{-1},$$ where $p'_{\textrm{out}}$ denotes the out component of the particle momentum $\bm{p}^{\prime}$ with respect to the local rest frame (lrf) of the fluid at position $\textsf{x}$. $\Lambda$, which generalizes the temperature, characterizes the scale over which the particle momentum takes significant values. As hinted at by the notations, both $\Lambda$ and the anisotropy parameter $\xi$ depend a priori on position and the particle type. Hereafter they will be treated as parameters, and for simplicity taken as constant over the freeze-out hypersurface $\Sigma$.
The anisotropy parameter $\xi$ must be larger than $-1$, to ensure the positivity of the expression under the square root. In order to obtain a larger pressure along the radial direction than perpendicular to it, $\xi$ should be negative.
To test the influence of the momentum anisotropy in Eq. , we assume some specific freeze-out flow profile and hypersurface $\Sigma$. We thus let the fluid decouple at a constant proper time $\tau_{\textrm{f.o.}}$ on an longitudinally infinite, azimuthally symmetric cylinder of radius $R$. Taking as coordinates in the laboratory frame the proper time $\tau$, space-time rapidity $\varsigma$, and cylindrical coordinates $r,\phi$, we assume for the fluid velocity on $\Sigma$ a generalised blast wave-like profile, namely [@Siemens:1978pb; @Huovinen:2001cy] $$\label{blastwave}
u^r(r,\phi) = \bar{u}_{\max}\frac{r}{R} \bigg( 1 + 2\sum_n V_n\cos n\phi\bigg)$$ for the radial coordinate, $u^\phi = u^\varsigma = 0$ in the azimuthal and $\varsigma$ directions, and eventually $u^\tau=\sqrt{1+(u^r)^2}$. With this choice, the phase-space occupation factor reads, when expressed in the laboratory frame
$$\label{f_an_lab}
f_{\textrm{an.}}(\textsf{x},\textsf{p};\Lambda,\xi) =
\bigg[\exp\!\bigg(\!\frac{\sqrt{[p^\tau u^\tau(\textsf{x}) - p^r u^r(\textsf{x})]^2 +
\xi[p^r u^\tau(\textsf{x}) - p^\tau u^r(\textsf{x})]^2}}{\Lambda}\bigg) \mp 1 \bigg]^{-1}.$$
Under these assumptions, we can numerically compute the Cooper–Frye integral, from which we can obtain the transverse momentum spectrum, Hanbury Brown–Twiss (HBT) radii $(R_{\rm out}, R_{\rm side}, R_{\rm long})$ [@Bertsch:1988db; @Pratt:1990zq], and the anisotropic flow coefficients $v_n$. We shall focus on pions ($m=140\;$MeV) produced at midrapidity.
We first present results obtained with fixed values of the “effective temperature” $\Lambda= 150\;$MeV and of the parameters of the blast wave profile: $\tau_{\textrm{f.o.\!}} = 7.5\;$fm/$c$, $R=10\;$fm, $\bar{u}_{\max} = 1$, $V_2 = V_3 = 0.05$, except for HBT radii for which all $V_n$ vanish. In contrast, we let the anisotropy parameter $\xi$ vary, giving it values from $-0.5$ to 0 in steps of $0.1$, together with 0.15 and 0.3. According to our argumentation in the previous section, these positive values, which lead to smaller pressure in the radial direction as perpendicular to it, should not be relevant for freeze out; yet we included them for reference sake.
![\[fig:dNdpT\_vs\_xi\] Transverse spectra for fixed $\Lambda$ and varying anisotropy parameter $\xi$.](./dNdpT_vs_xi){width="\linewidth"}
{width="0.495\linewidth"} {width="0.495\linewidth"} {width="0.495\linewidth"} {width="0.495\linewidth"}
Figure \[fig:dNdpT\_vs\_xi\] shows the resulting transverse momentum distributions. As is to be expected, non-zero values of $\xi$ lead to deviations from the almost exponential shape valid in the isotropic case. More precisely, the spectrum becomes harder when $\xi$ goes to increasingly negative values. This clearly reflects the growing radial pressure—or equivalently effective radial temperature $\Lambda/\sqrt{1+\xi}$—obtained by assuming $\xi<0$. In figure \[fig:HBT-R\_vs\_xi\], we display the various HBT radii, together with the ratio $R_{\rm out}/R_{\rm side}$, as functions of the pair transverse momentum $K_T$. To be more precise, the radii $R_{\rm side}^2$ and $R_{\rm long}^2$ are the $f_{\rm an.}$-weighted averages over the freeze-out hypersurface of $y^2=r^2\sin^2\phi$ and $z^2=\tau^2\sinh^2\varsigma$, respectively, while $R_{\rm out}^2$ is the average of $(x-K_T t/E_{\bm{K}})^2$, where $x=r\cos\phi$ and $t=\tau\cosh\varsigma$.
As was just mentioned, negative values of $\xi$ amount to a larger “radial temperature”, and thus to higher thermal velocities in the outwards direction. Since at the same time the emission duration barely changes, this naturally leads to a larger $R_{\rm out}$, as observed in the upper left panel, as well as to a larger ratio $R_{\rm out}/R_{\rm side}$ (lower right panel) In turn, the longitudinal radius $R_{\rm long}$ shown in the lower left panel is to a large extent unaffected by $\xi$; this could be anticipated since the longitudinal part of the occupation factor remains unchanged. On the other hand, the behaviour of the sidewards radius $R_{\rm side}$ with varying $\xi$ seen in the upper right panel of figure \[fig:HBT-R\_vs\_xi\] is more involved, and we did not find a satisfactory explanation describing all its details.
![\[fig:v2\_vs\_xi\] Elliptic flow $v_2(p_t)$ for fixed $\Lambda$ and varying anisotropy parameter $\xi$.](./v2_vs_xi){width="\linewidth"}
The transverse-momentum dependence of elliptic flow $v_2$ for various $\xi$ values is shown in figure \[fig:v2\_vs\_xi\]; triangular flow $v_3$ follows exactly the same trend, so that we do not show it. Thus, anisotropic flow decreases when $\xi$ becomes more negative, that is, as the radial temperature grows. This behaviour reflects the fact that an increase in random thermal motion tends to dilute the effect of directed collective behaviour encoded in the flow velocity and its anisotropies, i.e., it diminishes the $v_n$ values, as seen here.
![\[fig:dNdpT\_for\_various\_T-and-xi\] Transverse spectra for various choices of $\Lambda$ and $\xi$.](./dNdpT_for_various_T-and-xi){width="\linewidth"}
Before going any further, let us note that in a more complete approach, the local anisotropy parametrized in this work by $\xi$ should not be uniform, but rather position-dependent. In particular, $\xi$ (or similar parameters) would normally be function of the azimuthal angle $\phi$, paralleling the corresponding dependence of the velocity profile, as we now argue.[^2] The fluid-particle conversion, whose modelling $\xi$ is supposed to facilitate, roughly happens when the fluid expansion rate $\nabla_\mu u^\mu(\textsf{x})$ becomes comparable to that of elastic scatterings. Since the flow velocity varies with $\phi$, so does the expansion rate, which motivates an azimuthal dependence of $\xi$. On the other hand, the scattering rate depends on particle density, obtained by integrating the occupancy factor over momentum, and on the relative velocity of particles. As follows from a straightforward change of integration variable [@Rybczynski:2012ee], the density is inversely proportional to $\sqrt{1+\xi(\textsf{x})}$, thus a priori $\phi$-dependent. In turn, the typical relative velocity is controlled by the (effective) temperature(s) of the decoupling medium, thus function of $\phi$ as well…All in all, every relevant physical quantity depends on azimuth, so it is non-trivial—and within the scope of this paper rather academic—to determine the actual dependence of $\xi$. In any case, there will be such a dependence, which will affect the anisotropic flow coefficients $v_n$. The results shown in figures \[fig:v2\_vs\_xi\] and \[fig:v2\_for\_various\_T-and-xi\] are thus to be taken with a grain of salt, since they neglect this ingredient.
After having investigated the influence of $\xi$ when all other parameters are fixed, we now want to illustrate the degeneracy introduced by this new parameter, showing that very similar values of the observables can be obtained with different pairs $(\Lambda,\xi)$. Note that we did not attempt to optimize the results we now report by fine tuning the parameters, as will be made apparent by the values of the latter.
![\[fig:HBT-R\_for\_various\_T-and-xi\] HBT radii $R_{\rm out}$ (top panel) and $R_{\rm side}$ (bottom panel) for various choices ($\Lambda$,$\xi$).](./Rout_for_various_T-and-xi){width="\linewidth"}
![\[fig:HBT-R\_for\_various\_T-and-xi\] HBT radii $R_{\rm out}$ (top panel) and $R_{\rm side}$ (bottom panel) for various choices ($\Lambda$,$\xi$).](./Rside_for_various_T-and-xi){width="\linewidth"}
In figure \[fig:dNdpT\_for\_various\_T-and-xi\], we display the transverse momentum spectra for four sets of values of $(\Lambda,\xi)$, with $\Lambda$ varying between 130 and 160MeV and $\xi$ ranging from $-0.5$ to 0.3. In all four cases, the values of all other parameters are the same as above, in particular $\bar{u}_{\max}=1$. All four curves are barely distinguishable below $p_T=1.5\,$GeV, above which that with ($\Lambda=130\,$MeV, $\xi=-0.5$) starts curving up. The spectrum for ($\Lambda=140\,$MeV, $\xi=-0.25$) only starts differing from those with larger $\Lambda$ from about 2GeV onwards, while the remaining two stay very close up to at least 3GeV. In addition, we show in the same figure the spectrum for ($\Lambda=130\,$MeV, $\xi=-0.5$) and a different flow velocity, namely with $\bar{u}_{\max}=0.8$. The change in $\bar{u}_{\max}$ makes the spectrum almost collapse on that for ($\Lambda=150\,$MeV, $\xi=0$), with at most a 15% relative difference over the whole momentum range.
The HBT radii $R_{\rm out}$ and $R_{\rm side}$ and the elliptic flow $v_2$ for the same sets of parameters as in figure \[fig:dNdpT\_for\_various\_T-and-xi\] are respectively shown in figures \[fig:HBT-R\_for\_various\_T-and-xi\] and \[fig:v2\_for\_various\_T-and-xi\]. As in the case of the transverse spectra, the values of $R_{\rm out}$ or $v_2$ for all four pairs $(\Lambda,\xi)$ in the case $\bar{u}_{\max}=1$ are very close to each other, with ($\Lambda=130\,$MeV, $\xi=-0.5$) being most apart from the other three. We also include the result of the computation with $\bar{u}_{\max}=0.8$ which gives a good approximation to the $p_T$-distribution: for $v_2$, it basically makes no difference with respect to the case $\bar{u}_{\max}=1$, whereas the departure is more marked for $R_{\rm out}$.
![\[fig:v2\_for\_various\_T-and-xi\] Elliptic flow for various choices of $\Lambda$ and $\xi$.](./v2_for_various_T-and-xi){width="\linewidth"}
All in all, the results for transverse-momentum distributions, $R_{\rm out}$, and $v_2$ support our claim that introducing an extra parameter opens a much wider range for the “freeze-out temperature”, here $\Lambda$, without affecting drastically the values of the observables.
In contrast, the sidewards HBT radius $R_{\rm side}$ displayed in the bottom panel of figure \[fig:HBT-R\_for\_various\_T-and-xi\] is much more sensitive to the choice of decoupling parameters $(\Lambda,\xi)$. This is actually somewhat reassuring, since femtoscopic measurements are precisely designed to probe the space-time configuration at decoupling [@Lisa:2008gf].
Discussion {#s:discussion}
==========
We have argued that there are two main motivations for resorting to an anisotropic momentum distribution to describe the transition from usual dissipative fluid dynamics to a particle description at the end of the evolution of the fireball created in ultrarelativistic heavy-ion collisions. Firstly, this ansatz is supported by non-relativistic studies of freeze out [@Hamel:1966]. Secondly, this could help diminish the sensitivity of computed observables on the parameters introduced by the decoupling prescription, and thus lead to a smoother matching between models, in the spirit of seeing fluid dynamics emerging as the effective theory of some underlying, more microscopic dynamics.
As a matter of fact, our findings for transverse spectra, $R_{\rm out}$, and $v_2$ (figures \[fig:dNdpT\_for\_various\_T-and-xi\]–\[fig:v2\_for\_various\_T-and-xi\]) support the idea that introducing an extra parameter, which governs the local momentum anisotropy at decoupling, opens a much wider range for the “freeze-out temperature”, here $\Lambda$, without changing significantly the values of the observables. This is admittedly not too surprising, since we introduced one new degree of freedom. Yet at the risk of repeating ourselves, it emphasizes the fact that the “freeze-out temperature” is just a parameter for switching between two models, not a real physical temperature determined by some “critical”—in a loose sense—energy or entropy density for which the medium properties change drastically. Being such a parameter—like say a renormalization scale—, it may not have a dramatic impact on measurable quantities.
Accordingly, it seems possible to find a whole region of parameters to which the “early time” signals like anisotropic flow—which carry information on the properties of the fireball along its whole evolution [@Heinz:2013th], rather than on decoupling itself—are to a large extent insensitive. On the other hand, some sensitivity remains for the observables which are governed by the freeze-out process.
In the present exploratory study, we postulated the asymmetric form of the occupation factor at decoupling $f_{\textrm{an.}}$, and investigated some of the consequences within a toy model. The actual form of $f_{\textrm{an.}}$, together with that of the associated hydrodynamical quantities, still has to be calculated in a more microscopic approach [@BFL_inprep]. This involves at the same time a discussion of the freeze-out hypersurface $\Sigma$, whose position in space-time obviously depends on the amount of momentum anisotropy in the phase-space distribution.
Once this is done, it will be necessary to study how the “improved” prescription can be implemented in practice, i.e. how numerical simulations of dissipative fluid dynamics, anisotropic hydrodynamics, and particle transport can be glued together in a satisfactory manner. An important point will be to check what the shortcomings of the sudden freeze-out scenario, in particular the backflow of particles through $\Sigma$ [@Pratt:2014vja], become in the new approach: if there is more freedom in choosing the decoupling hypersurface, some choices may be more convenient than others. Eventually, it will be interesting to investigate the possible relation of the new prescription, which in essence still assumes a sudden fluid-particle conversion, with continuous emission . For instance, one may wonder if it is possible to mimic the latter within the former, or whether one has to formulate a continuous version of the “anisotropic decoupling” scenario.
[99]{}
P. Huovinen, Int. J. Mod. Phys. E [**22**]{} (2013) 1330029 \[arXiv:1311:1849 \[nucl-th\]\] P. Huovinen, H. Petersen, Eur. Phys. J. A [**48**]{} (2012) 171 \[arXiv:1206.3371 \[nucl-th\]\]. T. Hirano [*et al.*]{}, Prog. Part. Nucl. Phys. [**70**]{} (2013) 108 \[arXiv:1204.5814 \[nucl-th\]\]. F. Cooper, G. Frye, Phys. Rev. D [**10**]{} (1974) 186. D. Teaney, Phys. Rev. C [**68**]{} (2003) 034913 \[nucl-th/0301099\]. K. Dusling, D. Teaney, Phys. Rev. C [**77**]{} (2008) 034905 \[arXiv:0710.5932 \[nucl-th\]\]. G. S. Denicol, T. Kodama, T. Koide, P. Mota, Phys. Rev. C [**80**]{} (2009) 064901 \[arXiv:0903.3595 \[hep-ph\]\]. A. Monnai, T. Hirano, Phys. Rev. C [**80**]{} (2009) 054906 \[arXiv:0903.4436 \[nucl-th\]\]. K. Dusling, G. D. Moore, D. Teaney, Phys. Rev. C [**81**]{} (2010) 034907 \[arXiv:0909.0754 \[nucl-th\]\]. S. Pratt, G. Torrieri, Phys. Rev. C [**82**]{} (2010) 044901 \[arXiv:1003.0413 \[nucl-th\]\]. K. Dusling, T. Schäfer, Phys. Rev. C [**85**]{} (2012) 044909 \[arXiv:1109.5181 \[hep-ph\]\]. D. Teaney, L. Yan, Phys. Rev. C [**89**]{} (2014) 014901 \[arXiv:1304.3753 \[nucl-th\]\]. D. Molnar, Z. Wolff, arXiv:1404.7850 \[nucl-th\]. M. Luzum, J. Y. Ollitrault, Phys. Rev. C [**82**]{} (2010) 014906 \[arXiv:1004.2023 \[nucl-th\]\]. C. Lang, N. Borghini, Eur. Phys. J. C [**74**]{} (2014) 2955 \[arXiv:1312.7763 \[nucl-th\]\]. M. Rybczyński, W. Florkowski, J. Phys. G [**40**]{} (2013) 025103 \[arXiv:1206.6587 \[nucl-th\]\]. F. Grassi, Braz. J. Phys. [**35**]{} (2005) 52 \[arXiv:nucl-th/0412082\]. F. Grassi, Y. Hama, T. Kodama, Phys. Lett. B [**355**]{} (1995) 9; Z. Phys. C [**73**]{} (1996) 153. Yu. M. Sinyukov, S. V. Akkelin, Y. Hama, Phys. Rev. Lett. [**89**]{} (2002) 052301 \[arXiv:nucl-th/0201015\]. W. Florkowski, R. Ryblewski, Phys. Rev. C [**83**]{} (2011) 034907 \[arXiv:1007.0130 \[nucl-th\]\]. M. Martinez, M. Strickland, Nucl. Phys. A [**848**]{} (2010) 183 \[arXiv:1007.0889 \[nucl-th\]\]. For a review, see R. Ryblewski, J. Phys. G [**40**]{} (2013) 093101. B. B. Hamel, D. R. Willis, Phys. Fluids [**9**]{} (1966) 829.
M. I. Gorenstein, Yu. M. Sinyukov, Phys. Lett. B [**142**]{} (1984) 425. E. Molnár [*et al.*]{}, J. Phys. G [**34**]{} (2007) 1901 \[arXiv:nucl-th/0503048\]. N. Borghini, S. Feld, C. Lang, in preparation.
P. Romatschke, M. Strickland, Phys. Rev. D [**68**]{} (2003) 036004 \[arXiv:hep-ph/0304092\]. P. J. Siemens, J. O. Rasmussen, Phys. Rev. Lett. [**42**]{} (1979) 880. P. Huovinen, P. F. Kolb, U. W. Heinz, P. V. Ruuskanen, S. A. Voloshin, Phys. Lett. B [**503**]{} (2001) 58 \[arXiv:hep-ph/0101136\]. G. Bertsch, M. Gong, M. Tohyama, Phys. Rev. C [**37**]{} (1988) 1896. S. Pratt, T. Cs[ö]{}rg[ő]{}, J. Zim[á]{}nyi, Phys. Rev. C [**42**]{} (1990) 2646. M. A. Lisa, S. Pratt, in [*Relativistic Heavy Ion Physics*]{}, Landolt–B[ö]{}rnstein New Series I/23A (Springer, Berlin, 2010) \[arXiv:0811.1352 \[nucl-ex\]\]. U. Heinz and R. Snellings, Ann. Rev. Nucl. Part. Sci. [**63**]{} (2013) 123 \[arXiv:1301.2826 \[nucl-th\]\]. S. Pratt, Phys. Rev. C [**89**]{} (2014) 024910 \[arXiv:1401.0316 \[nucl-th\]\].
[^1]: We denote four-vectors in sans serif font and three-vectors in boldface; for the metric we adopt the mostly-minus convention.
[^2]: Similarly, $\Lambda$ also might depend on $\phi$, yet we leave this possibility aside to simplify the discussion.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We optimize modulation formats for the additive white Gaussian noise channel with a nonnegative input constraint, also known as the intensity-modulated direct detection channel, with and without confining them to a lattice structure. Our optimization criteria are the average electrical and optical power. The nonnegativity input signal constraint is translated into a conical constraint in signal space, and modulation formats are designed by sphere packing inside this cone. Some remarkably dense packings are found, which yield more power-efficient modulation formats than previously known. For example, at a spectral efficiency of 1 bit/s/Hz, the obtained modulation format offers a 0.86 dB average electrical power gain and 0.43 dB average optical power gain over the previously best known modulation formats to achieve a symbol error rate of $10^{-6}$. This modulation turns out to have a lattice-based structure. At a spectral efficiency of 3/2 bits/s/Hz and to achieve a symbol error rate of $10^{-6}$, the modulation format obtained for optimizing the average electrical power offers a 0.58 dB average electrical power gain over the best lattice-based modulation and 2.55 dB gain over the best previously known format. However, the modulation format optimized for average optical power offers a 0.46 dB average optical power gain over the best lattice-based modulation and 1.35 dB gain over the best previously known format.'
author:
-
title: 'Designing Power-Efficient Modulation Formats for Noncoherent Optical Systems'
---
Introduction
============
modulation has attracted significant research interest with its ability to improve the spectral efficiency of communication systems. The enabling technology behind it is the coherent transmission and detection, which gives access to both the carrier amplitude and phase to carry information. However, this increased spectral efficiency comes at the expense of a reduced power efficiency, which is undesirable in systems where power consumption is a constraint. Therefore, designing modulation formats which offer a good trade-off between spectral and power efficiency becomes challenging. Using *lattice codes*, which are a finite set of points selected out of an $N$-dimensional lattice, is one approach which has been extensively used in the construction of multilevel modulation formats for additive white Gaussian noise (AWGN) channels with coherent detection [@Forney1988; @Forney1989; @Conway1999]. In addition, techniques such as constellation shaping and nonequiprobable signaling have been used to minimize the average power[@Forney1989; @Calderbank1990]. The former is done by selecting the set of points in a lattice which have minimum energies, whereas the latter minimizes the average power by reducing the transmission frequency of points with high energies. Another approach is by resorting to *numerical optimization* techniques to find the best possible packing of constellation points as in [@Foschini1974; @Porath2003; @Sloane1995; @Graham1990; @Agrell2009] for different power constraints, whether average or peak power. The drawback is the lack of geometric regularity, which increases the modulator and demodulator complexity. As the number of constellation points increases, the best known packings approach a regular structure such as a lattice [@Conway1999 Ch. 1].
Both approaches for designing power-efficient modulation formats assume that both the amplitude and phase of the carrier can be used to carry information. However, in systems where phase information is absent, different modulation techniques must be considered. Examples of such systems include phase noise limited systems, and noncoherent systems where information is encoded onto the amplitude of the carrier and the envelope of the received signal is detected at the receiver, etc. The latter is prevalent in optical communication systems where the overall cost and complexity is a critical constraint. Such type of noncoherent systems are known as *intensity-modulated direct-detection* (IM/DD) systems and will be the focus of our work. In such systems the information is encoded onto the intensity of the optical carrier, and this intensity must, at all time instances, be nonnegative. Applications using IM/DD are, for example, wireless optical communications [@Barry1994; @Kahn1997; @Hranilovic2004a], and short-haul fiber links including data centers [@Randel2008].
In the absence of optical amplification, IM/DD systems can be modeled as a conventional AWGN channel whose input is constrained to being nonnegative [@Barry1994 Ch. 5], [@Kahn1997; @Hranilovic2004; @Hranilovic2003; @Farid2010; @Lapidoth2009]. Since the optical phase cannot be used to carry information, resorting to multilevel pulse amplitude modulation ($M$-PAM) is a natural low-complexity way of extending the widely spread on-off keying (OOK) to improve spectral efficiency. However, this is different from the conventional PAM since no negative amplitudes can be used [@Barry1994 Eq. (5.8)]. In [@Cunningham2006], an IM/DD link analysis using $4$-PAM signaling was performed. In [@Hranilovic2004], upper and lower bounds on the capacity of $2$-, $4$-, $8$-, and $16$-PAM were derived and in [@Walklin], the power efficiency of $M$-PAM was shown to be low. The $M$-ary pulse-position modulation ($M$-PPM) are known to be power-efficient; however, they have a poor spectral efficiency [@Hranilovic2005; @Kahn1997]. Since any nonnegative electrical waveform satisfies the above channel constraint, it can be communicated successfully over an IM/DD link. This implies that if the information to be transmitted is firstly modulated on a subcarrier (electrical) using any $M$-level modulation format, it can be transmitted on an IM/DD link after adding a direct current (DC) bias to ensure its nonnegativity, i.e., the subcarrier amplitude and phase which carries the information can be retrieved at the receiver. This concept is known as *subcarrier modulation* (SCM) and was described in the wireless infrared communication context [@Barry1994 Ch. 5]. Therefore, the power efficiency compared to $M$-PAM can be improved since SCM allows the use of power-efficient multilevel modulation formats with IM/DD systems. In [@Wiberg2009], the SCM concept is experimentally demonstrated, and in [@Space], a novel transmitter design for the subcarrier quadrature phase-shift keying (QPSK) and 16-ary quadrature amplitude modulation (16-QAM) is presented. The DC bias required to ensure the nonnegativity of the electrical waveform does not carry information [@Barry1994 Ch. 5], [@Wiberg2009; @Space]. Therefore, the improved power efficiency can be achieved by allowing the DC bias to vary on a symbol-by-symbol basis and within the symbol interval as in [@Hranilovic1999] and [@You2001], respectively. By guaranteeing nonnegativity, the investigation of lattice codes for IM/DD with AWGN became feasible and this is explored in [@Shiu1999; @Hranilovic2003]. In [@Hranilovic2003], a signal space model for optical IM/DD channels is presented, where average and peak optical power are considered as design constraints for constructing lattice-based modulation formats. In addition, constellation shaping to reduce the average optical power has been studied in [@Shiu1999] for the case where no amplification is used, and in [@Mao2008] where optical amplifiers are used. In this work, we optimize IM/DD modulation formats with and without confining them to a lattice structure. We propose a set of 4- and 8-level single-subcarrier modulation formats which are optimized for average electrical and optical power. These optimization criteria are both relevant, because the average *electrical* power is the standard power measure in digital and wireless communications [@Simon1995 p. 40] and it helps in assessing the power consumption in optical communications [@Chen1996], while the average *optical* power is an important figure of merit for skin- and eye-safety measures in wireless optical links [@Barry1994 Ch. 5], [@Kahn1997; @Hranilovic2003] and for quantifying the impact of shot noise in fiber-optical communications [@Cox2002 p. 20].
System Model {#sec:sysmodel}
============
\[c\]\[c\]\[1\]\[0\][$u(k)$ ]{} \[c\]\[c\]\[1\]\[0\][Modulator ]{} \[c\]\[c\]\[1\]\[0\][$x(t) \geq 0$ ]{} \[c\]\[c\]\[1\]\[0\][Laser diode ]{} \[c\]\[c\]\[1\]\[0\][$z(t)$ ]{} \[c\]\[c\]\[1\]\[0\][Optical link ]{} \[c\]\[c\]\[1\]\[0\][Photodetector ]{} \[c\]\[c\]\[1\]\[0\][$y(t) $ ]{} \[c\]\[c\]\[1\]\[0\][Demodulator ]{} \[c\]\[c\]\[1\]\[0\][$\hat u(k)$ ]{} \[c\]\[c\]\[1\]\[0\][$n(t)$ ]{}
The system model under study is depicted in Fig. \[fig:model\](a). It consists of a modulator which maps the data symbols $u(k)$ at instant $k$ to a waveform belonging to the signaling set $S=\{s_0(t),s_1(t), \ldots, s_{M-1}(t)\}$, where $M$ is the size of the signaling set. The generated waveform $$\label{sig:xt}
x(t)=\sum_{l=-\infty}^{\infty} s_{\Gamma[l]}(t-l T_s),$$ where $\Gamma[l]$ is an ergodic process uniformly distributed over $\{0,1,\ldots,M-1\}$ and $T_s$ is the symbol period, is constrained to being real and nonnegative. The received signal can be written as $$\label{basebandreceivedsignal}
y(t)= x(t) + n(t),$$ where $n(t)$ is a zero-mean Gaussian process with double-sided power spectral density $N_0/2$. It should be noted that there exists no nonnegativity constraint on the signal $y(t)$. This is then followed by the demodulation of $y(t)$ which yields $\hat u(k)$, an estimate of $u(k)$. The demodulator is a correlator or matched filter receiver, which minimizes the symbol error rate at a given signal-to-noise ratio (SNR) [@Simon1995 Sect. 4.1]. This model is different from the conventional AWGN channel by the fact that the input $x(t)$ is constrained to being nonnegative.
The baseband model in Fig. \[fig:model\](a) has been extensively studied in the optical communications context, since it serves as a good model for intensity-modulated direct-detection (IM/DD) systems [@Barry1994 Ch. 5], [@Kahn1997; @Hranilovic2004; @Hranilovic2003; @Farid2010; @Lapidoth2009]. The passband transceiver for IM/DD systems is depicted in Fig. \[fig:model\](b). In such systems, the electrical nonnegative waveform $x(t)$ directly modulates a light source, such as a laser diode. Therefore, the information is carried on the envelope of the passband signal $z(t)=\sqrt{2 c x(t)} \cos( 2 \pi f_o t +\theta)$, i.e., the intensity of the optical field, where $c$ represents the electro-optical conversion factor in watts per ampere (W/A) [@Cox2002 pp. 1–33], [@Westbergh2009], [@Coldren1999 pp. 32–67], $f_o$ is the optical carrier frequency, and $\theta$ is a random phase, uniformly distributed in $[0,2\pi)$. It then propagates through the optical medium depicted as an optical fiber in Fig. \[fig:model\](b), which could be a free-space optical link in other applications. At the receiver, the photodetector detects the power of $z(t)$. Since the dominant channel impairment in optical IM/DD systems is the thermal noise resulting from the optical-to-electrical conversion [@Mao2008], [@Agrawal2005 p. 155], the received electrical signal can be written as $$\label{ytwithslopeefficiency}
y(t)= r c x(t) + n(t),$$ where $r$ is the responsivity of the opto-electrical converter in A/W. Without loss of generality, we set $rc=1$, which yields (\[basebandreceivedsignal\]).
Signal Space Model {#sec:signalspace}
==================
By defining a set of orthonormal basis functions $\phi_k(t)$ for $k=1,2, \ldots,N$ and $N \leq M$ as in [@Hranilovic2003], each of the signals in $S$ can be represented as $$\label{eachsignal}
s_i(t)= \sum_{k=1}^{N} s_{i,k} \phi_k(t)$$ for $i=0, \ldots, M-1$, where $\mathbf{s}_i=(s_{i,1},s_{i,2}, \ldots, s_{i,N})$ is the vector representation of $s_i(t)$ with respect to the aforementioned basis functions. Therefore, the constellation representing the signaling set $S$ can be written as $\Omega=\{\mathbf{s}_0,\mathbf{s}_1, \ldots, \mathbf{s}_{M-1}\}$.
Single-Subcarrier Modulation Formats {#sec:scm}
------------------------------------
For in-phase and quadrature phase (I/Q) modulation formats to be used on IM/DD channels, a DC bias is required in order for $x(t)$ to be nonnegative. This could be translated geometrically by having a three-dimensional (3d) Euclidean space spanned by the orthonormal basis functions $$\begin{aligned}
\label{basisfunctions}
\phi_1(t) &=& \sqrt{ \frac{1}{T_s} } ~{\mathop{\mathrm{rect}}\nolimits}\left(\frac{t}{T_s}\right) \\ \nonumber
\phi_2(t) &=& \sqrt{ \frac{2}{T_s} }~\cos{(2\pi f t)} ~{\mathop{\mathrm{rect}}\nolimits}\left(\frac{t}{T_s}\right)\\\nonumber
\phi_3(t) &=& \sqrt{ \frac{2}{T_s} }~\sin{(2\pi f t)}~{\mathop{\mathrm{rect}}\nolimits}\left(\frac{t}{T_s}\right),\end{aligned}$$ where $$\begin{aligned}
{\mathop{\mathrm{rect}}\nolimits}(t) =
\begin{cases}
1 & \mbox{if } 0 \leq t \leq 1 \\
0 & \mbox{otherwise }
\end{cases}\end{aligned}$$ and $f$ is the electrical subcarrier frequency [@Hranilovic2003]. The basis function $\phi_1(t)$ represents the DC bias, where $s_{i,1}$ is chosen for each $i=0,\ldots,M-1$ such that $$\min_{t} s_i(t) \geq 0,$$ which guarantees the nonnegativity of $x(t)$ in (\[sig:xt\]). However, $\phi_2(t)$ and $\phi_3(t)$ are the basis functions of the conventional I/Q modulation formats such as $M$-PSK and $M$-QAM. As in [@Barry1994 pp. 115–116] and [@Hranilovic2003], we use $f=1/T_s$, which is the minimum value for which $\phi_1(t)$, $\phi_2(t)$, and $\phi_3(t)$ are orthonormal. In [@Hranilovic2003], IM/DD modulation formats based on these three basis functions are referred to as raised-QAM, and in [@Westbergha2010] as single-cycle SCM.
Performance Measures
--------------------
Two important power performance measures can be extracted from the baseband and passband models in Fig. \[fig:model\]. The first entity is the average electrical power defined as $$\label{aveele}
\bar P_{e}= \lim_{T \to \infty} \frac{1}{2T} \int_{-T}^{T} \! x^2(t) \, dt,$$ which for any basis functions can be simplified to $$\label{aveelesig}
\bar P_{e}= \frac{\bar{E_s}}{T_s} =\frac{1}{T_s}~\mathbb{E}[ \|\mathbf{s}_i \|^2] ,$$ where $\bar{E_s}$ is the average energy of the constellation and $\mathbb{E}[\cdot]$ is the expected value. This entity is an important figure of merit for assessing the performance of digital and wireless communication systems [@Simon1995 p. 40]. Therefore, it is relevant for IM/DD systems for compatibility with classical methods and results [@Channels2005; @SvalutoMoreolo2010]. In addition, it helps in quantifying the impact of relative intensity noise (RIN) in fiber-optical links [@Cox2002 pp. 1–33], and in assessing the power consumption of optical systems [@Chen1996]. In [@Karout2010], $\bar P_{e}$ was used as a performance measure for comparing different intensity modulation formats.
The second measure is the average optical power $\bar P_o$, which has been studied in [@Barry1994; @Kahn1997; @Hranilovic2004; @Hranilovic2003; @Farid2010] for the wireless optical channel. Limitations are set on $\bar P_o$ for skin- and eye-safety standards to be met. In fiber-optic communications, this entity is used to quantify the impact of shot noise on the performance [@Cox2002 p. 20]. It is defined as $$\label{opticalpower}
\bar P_{o}= \lim_{T\to \infty}\frac{1}{2T} \int_{-T}^{T} \! z^2(t) \, dt= \lim_{T \to \infty} \frac{c}{2T} \int_{-T}^{T} \! x(t) \, dt.$$ This measure depends solely on the DC bias required to make the signals nonnegative and can be represented in terms of the symbol period and the constellation geometry as [@Hranilovic2004; @Hranilovic2003] $$\label{opticalpowersig}
\bar P_{o}= \frac{c}{\sqrt{T_s}}~\mathbb{E} [s_{i,1}].$$
In order to have a fair comparison between the different modulation formats, the spectral efficiency defined as $
\eta={R_b/W}
$ (bits/s/Hz) should be taken into account, where $R_b=R_s \log_2M$ is the bit rate, $R_s=1/T_s$ is the symbol rate, and $W$ is the baseband bandwidth defined as the first null in the spectrum of $x(t)$. The term baseband bandwidth is due to the fact that the baseband model in Fig. \[fig:model\](a) hides the high frequency nature of the optical carrier [@Barry1994 pp. 109–116]. At the same symbol rate, modulation formats such as OOK and $M$-PAM have $W=R_s$, whereas the modulation formats belonging to the single-subcarrier family in Sec. \[sec:scm\] occupies $W=2 R_s$; this is due to the intermediate step of modulating the information onto an electrical subcarrier before modulating the optical carrier [@Barry1994 Ch. 5], [@Karout2010].
--------------------------------- --------------------------------- --------------------------------- ---------------------------------
{width="23.00000%"} {width="23.00000%"} {width="23.00000%"} {width="23.00000%"}
--------------------------------- --------------------------------- --------------------------------- ---------------------------------
\[8pointscone\]
Constellation Optimization {#sec:const}
==========================
For designing constellations, the admissible region defined in [@Hranilovic2003 Eq. (10)] as the set of all signals satisfying the nonnegativity constraint has to be taken into account. Given the 3d-Euclidean space defined for SCM in Sec. \[sec:scm\], the admissible region can be written as $$\label{adm:scm}
\Upsilon= \{\mathbf{s} \in \mathbb{R}^3 : s_{i,1}^2 \geq 2(s_{i,2}^2 +s_{i,3}^2)\},$$ which is a 3d-cone with apex angle of $\cos^{-1}(1/3)=70.528^{\circ}$ pointing in the dimension spanned by $\phi_1(t)$.
As done before for the conventional AWGN channel [@Foschini1974; @Porath2003; @Sloane1995; @Graham1990; @Agrell2009], our approach of finding the best constellations can be formulated as a sphere-packing problem with the objective of minimizing a cost function depending on the constraints that might be present in the system model shown in Fig. \[fig:model\]. Thus, the optimization problem, for given constants $M$ and $d_{\text{min}}$, can be written as $$\begin{aligned}
\label{obj}
\label{minline} \text{Minimize~} & & \xi(\Omega) \\
\label{minline1} \text{Subject to}& & |\Omega| =M\\
\label{cons2} & & \Omega \subset \Upsilon\\
\label{cons1} & & d(\Omega) = d_{\text{min}},\end{aligned}$$ where $$d(\Omega)=\min_{ \underset{i \ne j }{\mathbf{s}_i ,\mathbf{s}_j \in \Omega}} \| \mathbf{s}_i -\mathbf{s}_j \|.$$
Choosing the objective function as $\xi(\Omega)=\mathbb{E}[ \|\mathbf{s}_i \|^2]$ results in $\Omega=\mathscr{C}_{\bar P_e,M}$, i.e., a constellation optimized for average electrical power, and $\xi(\Omega)=\mathbb{E} [s_{i,1}]$ results in $\Omega=\mathscr{C}_{\bar P_o,M}$, i.e., a constellation optimized for average optical power. The constraint in (\[cons2\]) guarantees that the signals belong to the admissible region $\Upsilon$, therefore satisfying the nonnegativity criterion of the channel. The minimum distance $d_{\text{min}}$ serves as a good measure of error probability performance in the presence of AWGN at high SNR. Although this optimization problem is well formulated mathematically, it is rather difficult to obtain an analytical solution. Therefore, we resorted to numerical optimization techniques as in [@Foschini1974; @Porath2003; @Sloane1995; @Graham1990; @Agrell2009] to find the best constellations. The optimization problem is nonconvex; therefore, a local solution does not imply that it is global.
A special case of this optimization problem, which might not guarantee the optimal solution, is to confine the possible constellations to have a regular structure such as that of a lattice, denoted by $\Lambda$. In this case, the above optimization problem can be reformulated by replacing (\[cons2\]) with $ \Omega \subset \Upsilon \cap \Lambda$, and dropping (\[cons1\]) since it is directly inferred by (\[cons2\]). By using the face-centered cubic lattice ($A_3$), which provides the densest packing for the 3d-Euclidean space [@Conway1999 p. [16]{} ], we obtain $\mathscr{L}_{\bar P_e,M}$, and $\mathscr{L}_{\bar P_o,M}$, the constellations optimized for average electrical and optical power, respectively.
Optimized Constellations
------------------------
Below is the description of the obtained constellations whose coordinates are included in App. \[bestconstapp\]. We conjecture that all of them are optimal solutions of (\[minline\])–(\[cons1\]).
### $4$-level Constellations
Fig. \[4pointscone\] depicts the 4-level constellation which provides the lowest $\bar P_e$ and $\bar P_o$ while satisfying the optimization constraints. The geometry of this constellation is a regular tetrahedron where all the spheres, or the constellation points lying at the vertices of this regular tetrahedron, are equidistant from each other and normalized to unit $d_{\text{min}}$. This constellation is also the result of $\Upsilon \cap A_3$. Since the obtained constellation is optimized for both $\bar P_e$ and $\bar P_o$, we will refer to it as $\mathscr{C}_4$ or $\mathscr{L}_4$. It is a remarkable fact that the vertex angle of the tetrahedron, defined as the apex angle of the circumscribed cone, is exactly $\cos^{-1}(1/3)$, which is equal to the apex angle of the admissible region $\Upsilon$. Thus, $\mathscr{C}_4$ fits $\Upsilon$ snugly, in the sense that all constellation points, regarded as unit-diameter spheres, touch each other as well as the boundary of $\Upsilon$, which, as we shall see in the next section, makes the modulation format very power-efficient. This modulation consists of a zero level signal and a biased ternary PSK. In [@Karout2010], a power-efficient modulation format called on-off phase-shift keying (OOPSK) was presented. It turns out that it has the same geometry as $\mathscr{C}_4$. Other hybrids between amplitude-shift keying and PSK have been studied in [@Essiambre2010] and [@Gursoy2009]; however, such modulation formats do not satisfy the nonnegativity constraint of IM/DD channels.
### $8$-level Constellations
Fig. \[8pointscone\] also shows the 8-level constellations $\mathscr{C}_{\bar P_e,8}$ and $\mathscr{C}_{\bar P_o,8}$. None of these constellations are lattice-based, but all of them contain $\mathscr{C}_4$ as the lowest four spheres. The highly symmetric and compact constellation $\mathscr{C}_{\bar P_e,8}$ consists of four central spheres arranged in a tetrahedron and four additional spheres, each touching three spheres in the central tetrahedron. Surprisingly, seven of the eight spheres touch the conical boundary of $\Upsilon$. This modulation is a hybrid between 2-PAM and two ternary PSK which are DC biased differently. The constellation $\mathscr{C}_{\bar P_o,8}$ is the same as $\mathscr{C}_{\bar P_e,8}$ but with one sphere moved. On the other hand, when confining the set of points to a lattice structure, the resulting constellations which provide the lowest $\bar P_e$ and $\bar P_o$ are the same, $\mathscr{L}_{8}=\mathscr{L}_{\bar P_e,8}=\mathscr{L}_{\bar P_o,8}$, and is depicted in Fig. \[8pointscone\]. This constellation also contains $\mathscr{C}_4$ as the lowest four spheres.
Previously Known Constellations
-------------------------------
Our investigation encompasses other previously known formats which are presented after being normalized to unit $d_{\text{min}}$. At spectral efficiency $\eta=1$ bits/s/Hz, OOK defined as $\{(0),(1)\}$ in terms of $\phi_1(t)$ will be compared with $\mathscr{C}_4$, and with subcarrier QPSK defined as $\{(1,\pm 1/2,\pm 1/2)\}$ in terms of the basis functions defined in Sec. \[sec:scm\].
At a spectral efficiency $\eta=3/2$ bits/s/Hz, $\mathscr{C}_{\bar P_e,8}$, $\mathscr{C}_{\bar P_o,8}$, and $\mathscr{L}_{8}$ will be compared with subcarrier 8-PSK defined as $(1/\sin(\pi/8))\{(1/ \sqrt{2} , \cos( \pi i/4)/2 , \sin( \pi i/4) / 2 )\}$ for $i=0,\ldots,7$, star-shaped 8-QAM [@Essiambre2010] with a constant bias defined as $\{ ( (1+\sqrt{3})/\sqrt{2},\pm 1/2 ,\pm 1/2 ), ((1+\sqrt{3})/\sqrt{2},0,\pm (1+\sqrt{3})/2),((1+\sqrt{3})/\sqrt{2},\pm (1+\sqrt{3})/2,0) \}$. We also include in our analysis a star-shaped 8-QAM denoted as $\breve 8$-QAM in which the DC bias is allowed to vary from symbol to symbol, thus carrying information, and is defined as $\{ ( 1,\pm 1/2 ,\pm 1/2 ), ((1+\sqrt{3})/\sqrt{2},0,\pm (1+\sqrt{3})/2),((1+\sqrt{3})/\sqrt{2},\pm (1+\sqrt{3})/2,0) \}$.
\[c\]\[c\]\[0.75\]\[0\][$\mathrm{OOK}$]{} \[c\]\[c\]\[0.75\]\[0\][QPSK ]{} \[c\]\[c\]\[0.9\]\[0\][$\mathscr{C}_4$ ]{} \[c\]\[c\]\[0.75\]\[0\][8-QAM]{} \[c\]\[c\]\[0.75\]\[0\][$\mathscr{L}_8$]{} \[c\]\[c\]\[0.9\]\[0\][$\mathscr{C}_{\bar P_e,8}$]{} \[c\]\[c\]\[0.9\]\[0\][$\mathscr{C}_{\bar P_o,8}$]{} \[c\]\[c\]\[0.9\]\[0\][$\mathscr{C}_{\hat P_o,8}$]{}
\[c\]\[c\]\[0.75\]\[0\][$\breve 8$-QAM]{} \[c\]\[c\]\[0.75\]\[0\][8-PSK]{}
\[c\]\[c\]\[1\]\[0\][$P_s$]{} \[t\]\[c\]\[1\]\[0\][$\gamma_{\bar E_b}~\text{[dB]}$]{} \[t\]\[c\]\[1\]\[0\][$\gamma_{\bar P_o}~\text{[dB]}$]{} \[t\]\[c\]\[1\]\[0\][$\gamma_{\hat P_o}~\text{[dB]}$]{}
[c]{} ![Simulated (solid) and theoretical (dotted) SER for the various modulation formats vs. $\gamma_{\bar E_b}$ (top), $\gamma_{\bar P_o}$ (bottom).[]{data-label="serebno"}](serebno "fig:")\
\
![Simulated (solid) and theoretical (dotted) SER for the various modulation formats vs. $\gamma_{\bar E_b}$ (top), $\gamma_{\bar P_o}$ (bottom).[]{data-label="serebno"}](seravepo "fig:")\
\
Performance Analysis {#sec:perf}
====================
The symbol error rate (SER) performance of the different modulation schemes will be assessed. The standard union bound found in [@Simon1995 Eq. (4.81)] is used to approximate the theoretical SER. This union bound can be approximated as $$\label{bound}
P_{ s } \approx \frac{2 K }{M }~Q\left( \sqrt{\frac{ d_{\text{min}}^2}{2 N_0}} \right),$$ where $K$ is the number of distinct signal pairs $(s_i(t),s_j(t))$ with $i<j$ for which $\int (s_i(t)-s_j(t))^2 dt = d_{\text{min}}^2$. This approximation is tight at high SNR.
Fig. \[serebno\] (top) shows the simulated and theoretical SER of the studied modulation formats vs. SNR defined as $$\label{snreb}
\gamma_{\bar E_b} =10 \log_{10}\frac{\bar E_b}{N_0}.$$ Apparently, all the modulation formats which are optimized for $\bar P_e$ outperforms the other formats at the same spectral efficiency. For spectral efficiency $\eta=1$, $\mathscr{C}_4$ has a 0.86 dB average electrical power gain over OOK and 2.87 dB gain over QPSK to achieve $P_s= 10^{-6}$. For $\eta=3/2$, $\mathscr{C}_{\bar P_e,8}$ has a 0.3 dB gain over $\mathscr{C}_{\bar P_o,8}$, 0.58 dB gain over $\mathscr{L}_{8}$, 2.55 dB gain over $\breve 8$-QAM, 4.35 dB gain over 8-QAM, and 4.39 dB gain over 8-PSK to achieve $P_s= 10^{-6}$. The modulation formats optimized for $\bar P_e$ and $\bar P_o$ are very close in performance to $\mathscr{L}_{8}$, the lattice-based modulation format.
In order to facilitate the comparison of modulation formats in terms of their average optical power requirements, we define the optical SNR as $$\label{snravePo}
\gamma_{\bar P_o} = 10 \log_{10}\frac{ \bar P_o}{c\sqrt{R_b N_0}}$$ in a similar fashion as in [@Kahn1997 Eq. (5)] . Fig. \[serebno\] (bottom) shows the SER plotted vs. $\gamma_{\bar P_o}$. Quite obviously, the modulation formats optimized for $\bar P_o$ perform better than the rest. For $\eta=1$, $\mathscr{C}_4$ has a 0.43 dB average optical power gain over OOK, and a 2.06 dB gain over QPSK to achieve an SER of $10^{-6}$. For $\eta=3/2$, $\mathscr{C}_{\bar P_o,8}$ has a 0.04 dB gain over $\mathscr{C}_{\bar P_e,8}$, 0.46 dB gain over $\mathscr{L}_8$, 1.35 dB gain over $\breve 8$-QAM, 2.48 dB gain over 8-PSK, and a 2.75 dB gain over 8-QAM to achieve a $P_s= 10^{-6}$. Besides the very close performance of $\mathscr{C}_{\bar P_o,8}$ and $\mathscr{C}_{\bar P_e,8}$, it is clear that they together with the lattice-based modulation perform better than the other modulation formats under study in terms of average optical power performance.
Conclusion {#sec:conclusion}
==========
By relaxing the constraint on the set of points used to design modulation formats for IM/DD channels, we were able to design 4- and 8-level modulation formats which are more power-efficient than known ones. For the 4-level modulation formats, the most power-efficient modulation in terms of average electrical and optical power happens to have a lattice structure even though the number of constellation points is small. This constellation is also a subset of all the obtained higher level constellations. As for the 8-level constellations, power-efficient schemes are obtained by not confining the set of constellation points to a lattice structure. However, this comes at the price of losing the geometric regularity found in the lattice structure. We conjecture that the new modulation formats are optimal for their size and optimization criteria over IM/DD channels.
Obtained Constellations {#bestconstapp}
=======================
Constellations are normalized to unit $d_{\text{min}}$.
$$\begin{aligned}
\mathscr{C}_{\bar P_e,4}= & \mathscr{C}_{\bar P_o,4}=\mathscr{L}_{\bar P_e,4}=\mathscr{L}_{\bar P_o,4}=\{(0 , 0, 0),\\
&(\sqrt{2/3} , 0 , 1/\sqrt{3}),
(\sqrt{2/3}, \pm 1/2 , -\sqrt{3}/6)\}.\\
\mathscr{C}_{\bar P_e,8}= & \mathscr{C}_4 \cup \{
((5/3)\sqrt{2/3} ,0, -5/(3\sqrt{3})),\\
&((5/3)\sqrt{2/3} ,\pm 5/6, 5/(6\sqrt{3})),
(2 \sqrt{2/3} , 0 , 0)
\}.\\
\mathscr{C}_{\bar P_o,8}= & \mathscr{C}_4 \cup \{
((5/3)\sqrt{2/3}, 0, -5/(3\sqrt{3})),\\
&((5/3)\sqrt{2/3}, \pm 5/6 , 5/(6\sqrt{3})),\\
&(1.6293 ,0.9236, -0.6886)
\}.\\
\mathscr{L}_{\bar P_e,8}= & \mathscr{L}_{\bar P_o,8}=\mathscr{C}_4 \cup \{
(2\sqrt{2/3}, \pm1/2, \sqrt{3}/6),\\
&(2\sqrt{2/3}, 0, -1/\sqrt{3}),
( 2\sqrt{2/3}, 1, -1/\sqrt{3})
\}.\end{aligned}$$
Acknowledgment {#acknowledgment .unnumbered}
==============
The authors would like to acknowledge the SSF funding under grant RE07-0026 and LINDO Systems for the free license to use their numerical optimization software.
[10]{} \[1\][\#1]{} url@samestyle \[2\][\#2]{} \[2\][[l@\#1=l@\#1\#2]{}]{}
G. D. Forney, Jr., “[Coset codes–Part I: Introduction and geometrical classification]{},” *IEEE Transactions on Information Theory*, vol. 34, no. 5, pp. 1123–1151, 1988.
G. D. Forney, Jr. and L.-F. Wei, “[Multidimensional constellations–Part I. Introduction, figures of merit, and generalized cross constellations]{},” *IEEE Journal on Selected Areas in Communications*, vol. 7, no. 6, pp. 877–892, 1989.
J. H. Conway and N. J. A. Sloane, *[Sphere Packings, Lattices and Groups]{}*.1em plus 0.5em minus 0.4emSpringer-Verlag, 1999.
A. R. Calderbank and L. H. Ozarow, “[Nonequiprobable signaling on the Gaussian channel]{},” *IEEE Transactions on Information Theory*, vol. 36, no. 4, pp. 726–740, 1990.
G. Foschini, R. Gitlin, and S. Weinstein, “[Optimization of two-dimensional signal constellations in the presence of Gaussian noise]{},” *IEEE Transactions on Communications*, vol. COM-22, no. 1, pp. 28–38, Jan. 1974.
J.-E. Porath and T. Aulin, “Design of multidimensional signal constellations,” *IEE Proceedings - Communications*, vol. 150, no. 5, pp. 317–323, Oct. 2003.
N. J. A. Sloane, R. H. Hardin, T. D. S. Duff, and J. H. Conway, “[Minimal-energy clusters of hard spheres]{},” *Discrete & Computational Geometry*, vol. 14, no. 3, pp. 237–259, 1995.
R. L. Graham and N. J. A. Sloane, “[Penny-packing and two-dimensional codes]{},” *Discrete and Computational Geometry*, vol. 5, no. 1, pp. 1–11, 1990.
E. Agrell and M. Karlsson, “[Power-efficient modulation formats in coherent transmission systems]{},” *Journal of Lightwave Technology*, vol. 27, no. 22, pp. 5115–5126, 2009.
J. R. Barry, *[Wireless Infrared Communications]{}*.1em plus 0.5em minus 0.4emNorwell, MA, USA: Kluwer Academic Publishers, 1994.
J. M. Kahn and J. R. Barry, “[Wireless infrared communications]{},” *Proceedings of the IEEE*, vol. 85, no. 2, pp. 265–298, 1997.
S. Hranilovic, *[Wireless Optical Communication Systems]{}*.1em plus 0.5em minus 0.4emNew York: Springer, 2004.
S. Randel, F. Breyer, and S. C. J. Lee, “[High-speed transmission over multimode optical fibers]{},” in *Optical Fiber Communication Conference and Exposition and The National Fiber Optic Engineers Conference*.1em plus 0.5em minus 0.4emOptical Society of America, 2008, p. OWR2.
S. Hranilovic and F. R. Kschischang, “[Capacity bounds for power- and band-limited optical intensity channels corrupted by Gaussian noise]{},” *IEEE Transactions on Information Theory*, vol. 50, no. 5, pp. 784–795, 2004.
——, “[Optical intensity-modulated direct detection channels: Signal space and lattice codes]{},” *IEEE Transactions on Information Theory*, vol. 49, no. 6, pp. 1385–1399, 2003.
A. A. Farid and S. Hranilovic, “[Capacity bounds for wireless optical intensity channels with Gaussian noise]{},” *IEEE Transactions on Information Theory*, vol. 56, no. 12, pp. 6066–6077, 2010.
A. Lapidoth, S. M. Moser, and M. A. Wigger, “[On the capacity of free-space optical intensity channels]{},” *IEEE Transactions on Information Theory*, vol. 55, no. 10, pp. 4449–4461, Oct. 2009.
J. E. Cunningham, D. Beckman, X. Zheng, D. Huang, T. Sze, and A. V. Krishnamoorthy, “[PAM-4 signaling over VCSELs with 0.13[$\mu$]{}m CMOS chip technology]{},” *Optics Express*, vol. 14, no. 25, pp. 12028–12038, 2006.
S. Walklin and J. Conradi, “[Multilevel signaling for increasing the reach of 10 Gb/s lightwave systems]{},” *Journal of Lightwave Technology*, vol. 17, no. 11, pp. 2235–2248, 1999.
S. Hranilovic, “[On the design of bandwidth efficient signalling for indoor wireless optical channels]{},” *International Journal of Communication Systems*, vol. 18, no. 3, pp. 205–228, 2005.
A. O. J. Wiberg, B.-E. Olsson, and P. A. Andrekson, “[Single cycle subcarrier modulation]{},” in *Optical Fiber Communication Conference, OSA Technical Digest*, 2009, p. OTuE1.
B.-E. Olsson and M. Sköld, “[QPSK transmitter based on optical amplitude modulation of electrically generated QPSK signal]{},” in *Optical Fiber Communication and Optoelectronic Exposition and Conference*, 2008, p. SaA3.
S. Hranilovic and D. A. Johns, “[A multilevel modulation scheme for high-speed wireless infrared communications]{},” in *IEEE International Symposium on Circuits and Systems*, 1999, pp. 338–341.
R. You and J. M. Kahn, “[Average power reduction techniques for multiple-subcarrier intensity-modulated optical signals]{},” *IEEE Transactions on Communications*, vol. 49, no. 12, pp. 2164–2171, 2001.
D.-S. Shiu and J. M. Kahn, “[Shaping and nonequiprobable signaling for intensity-modulated signals]{},” *IEEE Transactions on Information Theory*, vol. 45, no. 7, pp. 2661–2668, 1999.
W. Mao and J. M. Kahn, “[Lattice codes for amplified direct-detection optical systems]{},” *IEEE Transactions on Communications*, vol. 56, no. 7, pp. 1137–1145, 2008.
M. K. Simon, S. M. Hinedi, and W. C. Lindsey, *[Digital Communication Techniques: Signal Design and Detection]{}*.1em plus 0.5em minus 0.4emEnglewood Cliffs, NJ: Prentice-Hall, 1995.
L. P. Chen and K. Y. Lau, “[Regime where zero-bias is the low-power solution for digitally modulated laser diodes]{},” *IEEE Photonics Technology Letters*, vol. 8, no. 2, pp. 185–187, 1996.
W. S. C. Chang, *[RF Photonic Technology in Optical Fiber Links]{}*. 1em plus 0.5em minus 0.4emCambridge University Press, 2002.
P. Westbergh, J. S. Gustavsson, [Å]{}. Haglund, A. Larsson, F. Hopfer, G. Fiol, D. Bimberg, and A. Joel, “[32 Gbit/s multimode fibre transmission using high-speed, low current density 850 nm VCSEL]{},” *Electronics Letters*, vol. 45, no. 7, pp. 366–368, 2009.
L. A. Coldren and E. R. Hegblom, *[Vertical-Cavity Surface-Emitting Lasers: Design, Fabrication, Characterization, and Applications]{}*.1em plus 0.5em minus 0.4emCambridge University Press, 1999.
G. P. Agrawal, *[Lightwave Technology: Telecommunication Systems]{}*. 1em plus 0.5em minus 0.4emNew Jersey: John Wiley & Sons, Inc., 2005.
K. Szczerba, B.-E. Olsson, P. Westbergh, A. Rhodin, J. S. Gustavsson, M. Karlsson, A. Larsson, and P. A. Andrekson, “[37 Gbps transmission over 200 m of MMF using single cycle subcarrier modulation and a VCSEL with 20 GHz modulation bandwidth]{},” in *European Conference And Exhibition On Optical Communication*, 2010, p. We.7.B.2.
K.-P. Ho, “[Exact evaluation of the capacity for intensity-modulated direct-detection channels with optical amplifier noises]{},” *IEEE Photonics Technology Letters*, vol. 17, no. 4, pp. 858–860, Apr. 2005.
M. S. Moreolo, R. Mu[ñ]{}oz, and G. Junyent, “[Novel power efficient optical OFDM based on Hartley transform for intensity-modulated direct-detection systems]{},” *Journal of Lightwave Technology*, vol. 28, no. 5, pp. 798–805, Mar. 2010.
J. Karout, E. Agrell, and M. Karlsson, “Power efficient subcarrier modulation for intensity modulated channels,” *Optics Express*, vol. 18, no. 17, pp. 17913–17921, Aug. 2010.
R.-J. Essiambre, G. Kramer, P. J. Winzer, G. J. Foschini, and B. Goebel, “[Capacity limits of optical fiber networks]{},” *Journal of Lightwave Technology*, vol. 28, no. 4, pp. 662–701, Feb. 2010.
M. C. Gursoy, “[Error rate analysis for peaky signaling over fading channels]{},” *IEEE Transactions on Communications*, vol. 57, no. 9, pp. 2546–2550, Sep. 2009.
[Michael Shell]{} Biography text here.
[John Doe]{} Biography text here.
[Jane Doe]{} Biography text here.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
We present calculations of magnetic exchange interactions and critical temperature $T_c$ in , and . The local spin density approximation is combined with a linear-response technique to map the magnetic energy onto a Heisenberg hamiltonion, but no significant further approximations are made. Special quasi-random structures in large unit cells are used to accurately model the disorder. $T_c$ is computed using both a spin-dynamics approach and the cluster variation method developed for the classical Heisenberg model.
We show the following: $(i)$ configurational disorder results in large dispersions in the pairwise exchange interactions; $(ii)$ the disorder strongly reduces $T_c$; $(iii)$ clustering in the magnetic atoms, whose tendency is predicted from total-energy considerations, further reduces $T_c$. Additionally the exchange interactions $J(R)$ are found to decay exponentially with distance $R^3$ on average; and the mean-field approximation is found to be a very poor predictor of $T_c$, particularly when $J(R)$ decays rapidly. Finally the effect of spin-orbit coupling on $T_c$ is considered. With all these factors taken into account, $T_c$ is reasonably predicted by the local spin-density approximation in MnGaAs without the need to invoke compensation by donor impurities.
author:
- 'J. L. Xu'
- 'M. van Schilfgaarde'
- 'G. D. Samolyuk'
title: 'Role of Disorder in Mn:GaAs, Cr:GaAs, and Cr:GaN'
---
Dilute magnetic semiconductors (DMS), i.e. semiconductors doped with low concentrations of magnetic impurities (usually Cr, Mn, or Co), have attracted much interest because of their potential application to spintronics[@Ohno96; @Ohno99] . is the most widely studied DMS, and it continues to attract interest because it is one of the few DMS where it is generally agreed that the magnetism is carrier-mediated. (This is important in spintronics because the magnetic state can be manipulated by electrical or optical means.)
In recent years Curie temperatures in have risen steadily, reaching $\sim$170K for $x$$\sim$0.08 when grown in thin films annealed at low temperature [@Chiba03; @Ku03; @Edmonds04]. It is generally believed defects (probably Mn interstitials) migrate out of the as-deposited films during the anneal, largely eliminating donor defects that hamper ferromagnetism. Since most practical applications of spintronics require room-temperature operation, a crucial question is then, what is the ultimate limit to $T_c$ in the DMS compounds, and in in particular?
This question was first addressed by Dietl in his now classic paper[@Dietl00], where he predicted a wide range of $T_c$ in tetrahedrally coordinated alloys. This stimulated a great deal of interest, although there is a growing consensus that most of the claims of that paper were artifacts of the assumptions in his original model. On the other hand, Akai[@Akai98] first used the local spin-density approximation (LSDA) to estimate $T_c$ within the Coherent Potential Approximation (CPA) in (In,Mn)As; he argued that a double exchange mechanism was a more appropriate description of the magnetism than the $pd$ exchange assumed by Dietl. Since then LSDA calculations of exchange interactions have been performed by a variety of groups [@mark01; @Sandratskii02; @Bouzerar03; @Kudrnovsky04; @Erwin03; @Sato03], usually extracting exchange parameters by calculating total energies of a fixed atomic but multiple-spin configurations, or by a linear-response technique within the CPA.
To date, disorder has almost always been neglected or treated within some mean-field (MF) approximation (MFA), either in the computation of the exchange parameters themselves, or in the subsequent analysis of magnetization $M(T)$ at finite-temperature, or both (though better treatments within $k\cdot p$ theory has been reported [@Schliemann01]). The LSDA+MF predict a rather high $T_c$ for (typically 350$\sim$400 K for $x$$\sim$0.08 [@Sato03]). The large discrepancy with experiment (at least in Mn:GaAs) is usually attributed to the very large numbers of compensating defects in real samples, which reduce $T_c$ [@Chiba03; @Ku03; @Edmonds04]. The situation remains somewhat uncertain because the number of defects still remaining in the best samples to date is not known.
This Letter addresses the issue of the ultimate limit to $T_c$ in some DMS alloys (focusing on Mn:GaAs) by adopting relatively rigorous approach to the calculation of the magnetic exchange interactions and $T_c$. Random alloys are approximated by large (128-250 atom) supercells where special quasirandom structures (SQS) [@zunger90] are used for the cation sublattice. Using a linear-response technique within the LSDA and the linear-muffin-tin orbitals method[@licht87; @mark99], the magnetic energy is mapped onto a Heisenberg form[@notea] $$H = -\sum_{ij} J(R_{ij}) \, {\hat e_i}\cdot{\hat e_j}
\label{eq:heisenberg}$$ where the sum is over all pairs $ij$ of magnetic atoms. To model $M(T)$ and $T_c$, Eq.(\[eq:heisenberg\]) is treated classically and integrated using a spin-dynamics (SD) technique[@sdyn96]; alternatively $M(T)$ is estimated by the cluster variation method (CVM)[@kikuchi] adapted[@notecvm] to solve Eq.(\[eq:heisenberg\]). Thus it is evaluated without recourse to empirical parameters or to the MFA. We show that the widely used MFA turns out be a very poor predictor of $M(T)$ in these disorded, dilute alloys, dramatically overestimating $T_c$.
With SQS we can rather precisely mimic a fully random configuration, but it is also possible to consider configurations that deviate from random. This can be important because LSDA predicts a strong attractive interaction between magnetic elements [@mark01], which implies a tendency towards clustering. In brief, we show that
- the disorder induces large fluctuations in $J_{ij}\equiv{}J(R_{ij})$ for every connecting vector $R_{ij}$;
- The fluctuations in $J_{ij}$ [*[reduce]{}*]{} $T_c$ relative to the configurationally averaged $\overline
J_{ij}=\left<J_{ij}\right>$;
- clustering [*[reduces]{}*]{} $T_c$, while ordering of the magnetic elements [*[increases]{}*]{} $T_c$.
![Pair exchange interactions $J(R_{ij})$, in mRy, for Mn:GaAs and Cr:GaN at two different concentrations as a function of $R_{ij}^3$. $R_{ij}$ is measured in units of the lattice constant $a$.[]{data-label="fig:jij"}](jij.eps){width="8.5cm"}
Fig. \[fig:jij\] shows $J_{ij}$ computed for an ensemble of 108-cation (216-atom) random supercells following the method of Ref.[@mark99], for and alloys at $x$=4.6% and $x$=8.3%. $3\times3\times3$ $k$-points were used, enabling the calculation of $J$ to very distant neighbors. We chose these two alloys because they are approximately representative of limiting cases. For Cr:GaN, the GaN host has a wide bandgap, and the Cr $t_2$ level falls near midgap. It broadens into an impurity band with 1/3 occupancy, and is believed to be responsible for the ferromagnetic exchange. For Mn:GaAs, most of the weight of the Mn $t_2$-derived state falls below the valence band maximum. A second $t_2$ impurity band about 0.1 eV above the valence band maximum is mainly responsible for the ferromagnetic exchange coupling in this case; the strength of $J(R)$ depends critically on the amount of Mn character in this band[@Mahadevan04]. Katayama-Yoshida used the $x$-dependence of $J_0=\sum_R\overline J(R)$ (computed within the CPA) to identify the ferromagnetism obtained from LSDA with model theories[@Sato03]. Within the CPA, $J_0\sim{}x^{1/2}$ for Cr:GaN, which corresponds to a double-exchange model, while Mn:GaAs displays character intermediate between $J_0\sim{}x^{1/2}$ and the $pd$ exchange ($J_0\sim{}x$) usually assumed by $k\cdot p$ models [@MacDonald99; @Dietl00].
Comparing Cr:GaN to Mn:GaAs, Cr:GaN shows substantially stronger nearest-neighbor (NN) interactions, owing to its small lattice constant; however $\overline J(R_{ij})$ decays much more rapidly with $R_{ij}$. This is because the wave function overlap between transition metal $d$ states decays much more rapidly for midgap states than near band-edge states. Evident also is the large dispersion in $J_{ij}$ for fixed $R_{ij}$ (note $J$ is drawn on a log scale): the root-mean square fluctuations $\Delta J_{ij} =
\sqrt{\left<J_{ij}^2-\bar{J}_{ij}^2\right>}$ are roughly comparable to ${\overline J}$. However $\Delta J_{ij}$ [*increases*]{} with $x$, and is substantially larger for the wide-gap case (Cr:GaN). Note that there is little evidence in either Cr:GaN or Mn:GaAs for oscillatory RKKY-like behavior, which in the simplest approximation predicts $J(R)\sim \cos(2k_F{}R)/R^3$. Instead, ${\overline{J}(R)}$ decays roughly exponentially in $R^3$, corresponding to a Fermi surface with imaginary wave number, as would obtain if the coupling were described by tunneling via a disordered impurity band[@Berciu01].
We now apply Eq.(\[eq:heisenberg\]) to compute $M(T)$, focusing on $T_c$. Mean-field theory, which estimates the effective field at each site from the average field contributed by other sites, predicts $T_c$ well above room temperature both in Mn:GaAs and Cr:GaN[@Newman03; @Sato03]. In spite of the rather strong differences in the form of $J(R)$ (Fig. \[fig:jij\]), mean-field theory predicts that Mn:GaAs and Cr:GaN have roughly similar $T_c$ for $x$$\sim$0.08[@Newman03]. This is because the NN interaction in the latter case is strongest, but the $J$ decays faster with $R$, leading to a comparable mean-field[@notemf] estimate $\overline{T}^{MFA}_c$.
But it should be evident from Fig. \[fig:jij\] that the MFA is of questionable reliability. First, it is well known that for dilute alloys there is a percolation threshold for the onset of ferromagnetism. (The threshold in the present case cannot be readily mapped to known models because $J(R)$ is nonneglible for a rather large number of neighbors.) Moreover, the large fluctuations $\Delta J(R)$ may strongly affect $T_c$, especially since $\Delta J(R)$ itself is purely a function of the environment[@mark01], and consequently of the local percolation path.
To obtain a precise estimate for $M(T)$ and $T_c$, we adopt a spin-dynamics approach[@sdyn96]. A 200 atom SQS structure (250 atom for the 4% alloy) was used to mimic the random alloy. From the TM atoms in the SQS structure, a supercell containing $\sim 2000$ Mn or Cr atoms was constructed to make a simulation cell for prosecuting spin-dynamical simulations. Following the method described in Ref.[@sdyn96], the Landau-Lifshitz (L-L) equation was integrated numerically at a fixed temperature allowing the system to equilibrate, followed by a simulation for $\sim 2\times10^6$ atomic units. The L-L equations were integrated with the Bulirsch-Stoer method. As the L-L equation is a first-order equation, global deamons were used for the heat bath[@sdyn96], to ensure ergodic behavior. The average magnetization $\overline{M}(T)$ was computed as a function of temperature, and $T_c$ was estimated from the inflection point in $\overline{M}(T)$. Owing to finite-size effects and the stochastic character of the simulation, $T_c$ could be determined to a precision of $\sim$5%.
Also we employed a CVM approach recently adapted to the classical Heisenberg hamiltonion[@notecvm]. This relatively simple scheme has been found to be accurate in simple 3$d$ magnets, overestimating $T_c$ by $\sim$5% (similar to the usual CVM for the Ising hamiltonion[@vaks99]). We can check the validity both methods in the DMS case by comparing their predictions of $T_c$. Fig \[fig:tc\] shows $T_c$ determined by both methods for and : agreement between the two methods is $\sim$10%, which is quite satisfactory considering the complexity of the $J_{ij}$. $\overline{T}^{MFA}_c$ is also shown: evidently the MFA rather badly overestimates $T_c$. $\overline{T}^{MFA}_c>T_c$ by $\sim$200K in the Mn:GaAs alloy, and by a somewhat larger amount in Cr:GaAs. The discrepancy is still more dramatic in Cr:GaN (not shown); we find $T_c<50$K for all concentrations studied while $\overline{T}^{MFA}_c\sim
600$K[@Newman03]. Indeed we have found this generally to be the case when $J(R)$ decays rapidly or when $\Delta J(R)/J(R)$ is not small.
These results stand in stark contrast to the $\sim$15% discrepancy between $T^{MFA}_c$ and $T_c$ typically found in simple metals. The reason is easily understood by considering the effective field a mean-field atom sees, $\vec
H^{eff}_i=\sum_jJ_{ij}{\hat e}_j$. From the exponential decay of $J(R)$, it is evident that $H_i$ will be dominated by the nearest neighbors. But for dilute alloys, near-neighbors are not sufficent to form a percolation path. This is immediately evident in the extreme case of a NN pair of magnetic atoms well separated from any other magnetic atoms: the contribution to $T^{MFA}_c$ from this pair would be high, even though the pair would actually contribute nothing to ferromagnetism.
In Ref. [@Bouzerar03] a small discrepancy between $T^{MFA}_c$ and a more sophisticated calculation for $T_c$ was reported. In that calculation the CPA was used to construct an average $\overline{J}_{ij}$ and $M(T)$ modeled by constructing a fcc lattice of magnetic atoms, using concentrated-weighted $\overline{J}_{ij}$ for the exchange parameters. It would seem that their conclusions are an artifact of the neglect of configurational disorder (except in the computation of $\overline{J}_{ij}$). Better would be to estimate $\overline{J}_{ij}$ within the CPA, and then construct a [*disordered*]{} simulation cell using the $\overline{J}_{ij}$ to estimate $M(T)$. Still this approach neglects fluctuations $\Delta J$, which as we have seen are comparable to $\overline{J}_{ij}$ itself. To assess the effect of fluctuations, we repeated the calculation for $T_c$ within the CVM, replacing the environment-specific ${J}_{ij}$ with the configurationally averaged $\overline{J}_{ij}$. For at $x$=0.08, the effect of disorder ($\overline{J}_{ij}\to{J}_{ij}$) was to reduce $T_c$ by 50 K. (It is interesting that the MFA predicts the [*opposite*]{} trend, because of an artificial tendency for $M^{MFA}(T)$ to track whichever site $i$ has the largest $\vec H^{eff}_i$. Then $T^{MFA}_c-\overline{T}^{MFA}_c$ is positive[@notemf] and increases with $\Delta J/J$. This explains why a tight-binding+MF analysis[@Berciu01] predicted that disorder [*increases*]{} $T_c$.)
![Dependence of $T_c$ (K) on $x$ in and . Solid lines: $T_c$ computed from the MF $\overline{T}^{MFA}_c$[@notemf]. Dotted line: $T_{c}$ extracted from spin-dynamics simulations of Eq. \[eq:heisenberg\]. Diamonds: $T_{c}$ computed from the Heisenberg Cluster Variation Method. \[fig:tc\] ](tc.eps){width="6.0cm"}
We next consider the effects of nonrandomness. As noted above, real DMS alloys should exhibit some clustering owing to the attractive interaction between magnetic elements[@mark01]. The true situation is complicated by the nonequilibrium growth required to stabilize the alloy in the zincblende structure. Nevertheless the Mn-Mn or Cr-Cr binding energy is calculated[@mark01] to be an order of magnitude larger than the growth temperature ($\sim$250K), and some pairing or other clustering should be expected, particularly since films must be annealed to obtain good $T_c$. There is some experimental evidence for a tendency to cluster[@Sullivan03].
The effect of clustering on $T_c$ in Ga$_{0.92}$Mn$_{0.08}$As was studied by a simple model. To characterize the configurational disorder we adopt the standard Ising formalism, and assign $\sigma=\pm 1$ to each cation site (+1 for Mn and $-$1 for Ga). The random (SQS) configuration was constructed by searching for configurations which best approximate the ideal random configuration for pair correlation functions $P_{R_{ij}}=\left<\sigma_i\sigma_j\right>$ (and some higher-order correlation functions) up to some fixed distance. For a random configuration, $P_R=(2x-1)^2$ independent of $R$. To parameterize the clustering in a simple manner, we adopted the NN pair correlation function $P_1$ as a measure of clustering. Starting from an initial SQS configuration, a simulated annealing cycle was performed by generating a set of site configurations with increasing $P_1$, corresponding to longer annealing times (For simplicity, $P_n(n>1)$ was optimized to be $(2x-1)^2$ for each configuration.) $J_{ij}$ and $T_c$ were computed by the CVM and MFA[@notemf] as a function of $P_1$; see Fig. \[fig:tccluster\]. $T_c$ is rather strongly [*reduced*]{} with increasing $P_1$. This is perhaps not surprising since increased clustering implies more distant average separation between atoms, which is deleterious to links in the percolation path. Even within the MFA $T_c$ changes slightly, albeit for a different reason. In that case, there is an increase in NN pairs, which would increase $T_c$, but at the same time there is some increase in the likelihood of [*three-*]{} and higher body neighbors. The presence of a third neighbor has the effect of [*reducing*]{} the pairwise $J_{ij}$ [@mark01], and is the origin of the factor-of-three variations in the NN $J$ in Fig. \[fig:jij\].
We also considered the [*ordered*]{} limit, by putting 1 Mn in a 24-atom unit cell, corresponding to $x$=0.083. In this case $P_1$ decreases to 2/3, and $T_c$ increases to 350K (see Fig. \[fig:tccluster\]). Thus we conclude that ordering [*increases*]{} $T_c$, while clustering [*decreases*]{} $T_c$. Perhaps not suprisingly, the MFA $T_c$ approaches the CVM result in the ordered case, since percolation is less critical.
![Dependence of $T_c$ (K) on the pair correlation function $P_1$ in Ga$_{0.92}$Mn$_{0.08}$As. The random (SQS) configuration corresponds to $P_1$=0.7056. (Two SQS structures were calculated.) Diamonds show $T_c$ computed with CVM; circles show $\overline{T}^{MFA}_c$, and triangles show ${T}^{MFA}_c$. The point at $P_1=2/3$ corresponds to the ordered compound. []{data-label="fig:tccluster"}](tc-cluster.eps){width="6.5cm"}
To conclude, we have shown that ferromagnetism is very sensitive to configurational disorder in DMS alloys, and that with proper treatment of disorder $T_c$ is reasonably predicted by the LSDA for , without needing to invoke compensating defects. We briefly consider two important sources of error from elements missing in the theory. First, spin-orbit coupling strongly reduces $T_c$ in $k\cdot{}p$ models. We estimated its effect by computing the change in $\overline{T}^{MFA}_c$ when the $L\cdot{}S$ coupling is added to the LSDA hamiltonion. For Ga$_{0.92}$Mn$_{0.08}$As, $\overline{T}^{MFA}_c$ was reduced by $\sim$10%. Finally, the LSDA itself will overestimate $T_c$ somewhat [@Mahadevan04]. In a future work we will present a reliable parameter-free theory that corrects the principal errors in LSDA—most importantly the Mn $d$ character at $E_F$—and quantify the extent to which the LSDA overestimates $T_c$. Finally, we conclude that the present calculations represent a rather strict upper bound to $T_c$, and that for random or clustered alloys, $T_c>250$ K is unlikely.
This work was supported by the Office of Naval Research.
[28]{} natexlab\#1[\#1]{}bibnamefont \#1[\#1]{}bibfnamefont \#1[\#1]{}citenamefont \#1[\#1]{}url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[2\]\[\][[\#2](#2)]{}
, , , , , , , ****, ().
, , , , , , ****, ().
, , , , ****, ().
, , , , , , , , , , , ****, ().
, , , , , , , , , , , ****, ().
, , , , , ****, ().
, ****, ().
, ****, ().
, ****, ().
, , , , ****, ().
, , , , , , , ****, ().
, ****, ().
, , , ****, ().
, , , , ****, ().
, , , , ****, ().
, , , , ****, ().
, ****, ().
.
, , , , , ****, ().
, ****, ().
, ****, ().
, , , , ****, ().
, ****, ().
, , , , , , , , , ****, ().
, ****, ().
, , , , , , ****, ().
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
We present new H$\alpha$ narrow band imaging of the HII regions in eight Sculptor Group dwarf irregular (dI) galaxies. The H$\alpha$ luminosities of the detected HII regions range from some of the faintest detected in extragalactic HII regions ($\sim$ 10$^{35}$ erg s$^{-1}$ in SC 24) to some of the most luminous ($\sim$ 10$^{40}$ erg s$^{-1}$ in NGC 625). The total H$\alpha$ luminosities are converted into current star formation rates (SFR). Comparing the Sculptor Group dIs to the Local Group dIs, we find that the Sculptor Group dIs have, on average, lower values of SFR when normalized to either galaxy luminosity or gas mass (although there is considerable overlap between the two samples). The range for both the Sculptor Group and Local Group samples is large when compared to that seen for the sample of gas-rich quiescent low surface brightness (LSB) dIs from van Zee et al. (1997) and the sample of isolated dIs from van Zee (2000, 2001). This is probably best understood as a selection effect since the nearby group samples have a much larger fraction of extremely low luminosity galaxies and the smaller galaxies are much more liable to large relative variations in current SFR. The Sculptor Group and LSB samples are very similar with regard to mean values of both $\tau_{gas}$ and $\tau_{form}$, and the Local Group and isolated dI samples are also similar to each other in these two quantities. Presently, the Sculptor Group lacks dI galaxies with elevated normalized current SFRs as high as the Local Group dIs IC 10 and GR 8.
The properties of “transition” (dSph/dIrr) galaxies in Sculptor and the Local Group are also compared and found to be similar. The transition galaxies are typically among the lowest luminosities of the gas rich dwarf galaxies. Relative to the dwarf irregular galaxies, the transition galaxies are found preferentially nearer to spiral galaxies, and are found nearer to the center of the mass distribution in the local cloud. While most of these systems are consistent with normal dI galaxies which currently exhibit temporarily interrupted star formation, the observed density-morphology relationship (which is weaker than that observed for the dwarf spheroidal galaxies) indicates that environmental processes such as “tidal stirring” may play a role in causing their lower SFRs.
author:
- 'Evan D. Skillman'
- Stéphanie Côté
- 'Bryan W. Miller'
title: 'Star Formation in Sculptor Group Dwarf Irregular Galaxies and the Nature of “Transition” Galaxies'
---
Introduction
============
Low-mass dwarf irregular galaxies provide an important testing ground for several fundamental questions about star formation, star formation rates (SFRs), galactic evolution, and cosmology. Due, in large part, to attempts to understand the possible evolutionary connections between the dwarfs with negligible or extremely low present SFRs (the dSph and dE galaxies, hereafter dE galaxies) and the dwarfs with obvious signs of present star formation (the dIrrs, blue compact dwarfs, HII galaxies, hereafter dI galaxies), many theorists are turning their attention to the problem of dwarf galaxy evolution. Environmental effects are turning out to be a key parameter (e.g., van den Bergh 1994b; Klypin et al. 1999; Moore et al. 1999; Gnedin 2000; Mayer et al. 2001a,b; Carraro et al. 2001). An especially interesting question in this regard is the effect of reionization on the suppression of dwarf galaxy formation (Efstathiou 1992; Babul & Rees 1992; Quinn, Katz, & Efstathiou 1996; Barkana & Loeb 1999; Bullock et al. 2000). By comparing the properties of dwarf galaxies in different environments we may be able to isolate key environmental variables (e.g., local density, companionship, group vs. cluster membership) in order to constrain these theories.
The Sculptor group is the closest group of galaxies beyond our Local Group. Originally thought to be at a distance of 2.5 Mpc (de Vaucouleurs 1975), the main members of the group are 5 late-type spiral galaxies (NGC 55, NGC 247, NGC 253, NGC 300, and NGC 7793) with distances ranging between 1.6 and 3.4 Mpc (Puche & Carignan 1988 and references therein) and a dwarf spiral (NGC 45) at a distance of 4.4 Mpc. Additionally, NGC 24, another dwarf spiral, lies close to NGC 45 in radial velocity and projected location and is usually included as a group member (e.g., Giuricin et al.2000) although some distance estimates put it as far away as 11 Mpc (Puche & Carignan 1988). Jerjen, Freeman, & Binggeli (1998) studied surface brightness fluctuations in 5 Sculptor group dEs and determined a spread in distances of dE members of the group from 1.7 to 4.4 Mpc (but note that Jerjen & Rejkuba 2001 claim that some of these distances should be revised). A well defined relationship between radial velocity (corrected to the galactocentric standard of rest) and distance was discovered for all members of the Sculptor group (Jerjen et al. 1998), indicating that radial velocity may be a good distance indicator for Sculptor Group members. Jerjen et al. (1998) propose that the Sculptor Group and the Local Group are part of the same Supergalactic structure (a part of the Coma - Sculptor Cloud delineated by Tully & Fisher 1987). A dynamical picture of the Sculptor Group can be found in Whiting (1999).
Miller (1994), Côté (1995), and Jerjen et al. (2000) independently searched for and found new dwarf galaxy members of the Sculptor Group. The Sculptor Group is now known to contain 16 dwarf irregular galaxies (Côté et al. 1997), some of which are amongst the lowest luminosity dwarf irregulars known. The present Sculptor Group membership situation can be found in Jerjen et al. (1998; 2000).
Miller (1996) noted that several Sculptor Group dIs were undetected in H$\alpha$ emission at reasonably sensitive levels. This has several implications for the present study. First, as Miller (1996) pointed out, the Sculptor Group dIs may have relatively low average SFRs when compared to Local Group dIs. Second, although some of the Miller (1996) sample have now been detected in H$\alpha$ at low levels (see §2.2), there remain three dwarf galaxies in the Sculptor Group (SDIG, DDO 6, and UGCA 438) with the properties of “transition type” galaxies (no detectable HII regions and M$_{HI}$/L$_B$ values in the range of dI galaxies). Thus, the Sculptor Group sample allows us to investigate this less common type of galaxy (see, e.g., Miller et al. 2001). Third, H$\alpha$ surveys identifying HII regions are a necessary first step in studying the ISM abundances of galaxies. Since relatively high surface brightness HII regions are usually required for an accurate abundance analysis, this means that we may not be able to accurately measure the chemical abundances in all of the known Sculptor Group dIs.
Here we present deep CCD H$\alpha$ imaging of eight Sculptor group dIs obtained at the “Danish” 1.5m telescope located at the European Southern Observatory. We have detected HII regions in all eight dIs observed, in addition to the two dIs detected in H$\alpha$ by Miller (1996). We present coordinates and fluxes for the HII regions and estimated SFRs for these galaxies. Additionally, we have used the CTIO 4-m to obtain optical spectra of ten HII regions located in five of these dIs and these observations are presented in a companion paper. By comparing observations of the dwarf galaxies in the next nearest group to other well defined samples, we hope to better understand basic questions such as: what are the average properties of the dwarf galaxies? and, are there observable signatures of environmental influences on galaxy evolution?
Observations
============
Target Selection
----------------
All galaxies observed were chosen from the list of Côté et al. (1997). Table 1 lists some of the properties of the observed galaxies. Based on the observation by Jerjen et al. (1998) that the scatter in the distance – velocity relationship for the Sculptor Group members is “remarkably small”, we have derived new distances for these galaxies using the recessional velocities listed in Côté et al. (1997) and the formula given in Jerjen et al. (1998): $$v_{\rm GSR}\ [km s^{-1}] = 119\, (\pm 7)\, D\ [Mpc]\, - 136\, (\pm 14).$$ For the nine galaxies (6 spirals and 3 dwarfs) used to define the relationship, the average difference between the measured distance and the estimated distance is 10%. Eight of the nine galaxies show differences of less than 20% (the exception being NGC 45, with a difference of 21%). Dolphin (private communication) has calculated distances from Hubble Space Telescope observations of the red giant branch tip for five (1 spiral and 4 dwarfs) Sculptor group galaxies with known recession velocities (DDO 6, DDO 226, NGC 253, ESO 245-G005 and ESO 294-G010), and these five galaxies show good agreement with the relationship derived by Jerjen et al. (1998); all five galaxy distances are within 20% of that predicted by the relationship. The relationship should be updated with inclusion of these new data, but, based on the above, the anticipated revision will be small (of order the uncertainty in the relationship), and we will use the relationship as published. The estimated errors in distances are small enough that we can begin to examine positional relationships between galaxies. Note that the finding by Jerjen & Rejkuba (2001) that the distances estimated for some of the dEs via the method of surface brightness fluctuations could be significantly in error does not directly imply that the formula relating distance and recessional velocity needs to be revised since most of the dEs do not have measured recession velocities and were not used in calibrating the relationship.
This equation was derived for galaxies covering the range in corrected recessional velocity from $\sim$ 70 to $\sim$ 490 km s$^{-1}$, and thus may not be strictly valid for the three galaxies observed here with corrected recessional velocities in excess of 500 km s$^{-1}$. The fact that several of the dIs have recessional velocities in excess of 500 km s$^{-1}$ implies that a significant number of dI galaxies lie on the far side of the Sculptor group. (Note that the spiral galaxy NGC 24, with a recession velocity of 555 km s$^{-1}$ is also in this region.) In comparison with the Local Group, this is not unexpected, as several of the Local Group dIs lie in the low density periphery of the Local Group at distances in excess of 1 Mpc from the Milky Way and M31 (Mateo 1998). The absolute B magnitudes in Table 1 were calculated from the total B magnitudes and extinctions listed in Côté (1995). From a comparison of the values in Côté (1995) and other values in the literature, the uncertainty in M(B) from the photometry is probably a little less than 0.2 magnitudes, and, when considering the error in the distance estimates, the total error is probably a little larger than 0.2 magnitudes for the typical Sculptor Group member. The error could be larger for those galaxies with velocities in excess of 500 km s$^{-1}$. Nonetheless, these errors are small relative to the dynamic range of the luminosities of the galaxies and allow us to investigate relationships between intrinsic properties of galaxies.
H$\alpha$ Imaging
-----------------
The search for HII regions in the Sculptor dIs was performed on October 24 and 25, 1995 with DFOSC at the 1.5m “Danish" telescope at the European Southern Observatory. Additional data for one galaxy (ESO 473-G24) were gathered on November 4, 1995, courtesy of Chris Lidman. The detector used was the Loral 2048 $\times$ 2048, giving a scale of 0.39 pixel$^{-1}$ and field of view of 13.6 $\times$ 13.6. The CCD was read out in the high gain mode yielding a readout noise of 7.2 e$^-$ r.m.s. with a gain of 1.31 e$^-$/ADU. The galaxies were imaged through a narrowband H$\alpha $ filter with a FWHM of 62.1 Å centered at $\sim $6561 Å (ESO\#693), and continuum off-band images were taken in Gunn [*i*]{}. The observations usually consisted of 2 $\times$ 1200s for the H$\alpha $ images and 2 $\times$ 600s for the [*i*]{}-band images. The seeing varied between $\sim $1 and 1.3, but the conditions were not photometric.
The data were reduced following the usual procedures with IRAF. After registering the different images, the two H$\alpha $ and two [*i* ]{} frames were co-added. To optimize the continuum subtraction, we used a small kernel Gaussian filter to smooth the [*i* ]{} image in order to match as closely as possible the point-spread functions of the [*i* ]{} and H$\alpha $ images. Half a dozen isolated bright stars were used to determine a scaling factor between the [*i* ]{} and H$\alpha $ frames. The final continuum subtracted image was then produced by subtracting the appropriately scaled [*i* ]{} image from the H$\alpha $ image. Figures 1a-d show the original H$\alpha $ and the continuum-subtracted H$\alpha $ images.
Although many HII regions are fairly well isolated, many are assembled in larger complexes, or have more complex morphologies like loops or filaments, as is commonly seen in late-type galaxies. Hence it was necessary to determine the boundaries of each HII region by eye, using POLYPHOT, selecting each distinct emission peak as a separate HII region. The borders of the HII regions were set to a constant H$\alpha $ surface brightness level for each galaxy but because the nights were not photometric these levels are estimated to be varying around $5 \pm 2 \times
10^{-17}$ erg cm$^{-2}$ s$^{-1}$ arcsec$^{-2}$. The positions and fluxes of the HII regions are given in Table 2. Errors in the fluxes are estimated to be of order 20% due to photometric uncertainties, except for SC 18, SC 24, and NGC 625; these galaxies were observed at the end of the second night when the conditions had improved, and the errors are estimated to be 10%. In fact, for NGC 625, which is the only galaxy in our sample with previous H$\alpha $ measurements in the literature, our fluxes agree within 4% with that of Marlowe et al. (1997). On the other hand, for AM 0106-382 the conditions had deteriorated so badly that the galaxy was not even detectable in the continuum [*i*]{} exposures. However, our H$\alpha $ images were compared with the $i$ data obtained by Côté (1995), and it appears that the emission peaks seen in our H$\alpha $ image do not have any counterparts in the $i$ image. This indicates that they could be genuine HII regions, so they are listed as well in Table 2. None of the fluxes given in Table 2 have been corrected for \[NII\] contamination; these dwarf galaxies have typically very low nitrogen abundances (see companion paper), so this introduces an additional $\sim $6% flux uncertainty. For deriving accurate positions for the HII regions, HST Guide Star Reference Frame scans from the STScI Digitized Sky Survey were used, and were compared with positions obtained similarly by deriving astrometric plate solutions using bright stars in the Automatic Plate Measuring (APM) catalog (cf. van Zee 2000). In each case these positions agreed to within less than 2.
The HII regions in the two faintest galaxies (SC 18 and SC 24) were just at the limit of detectability. In both cases, coincidence with continuum sources cast doubt on whether these are true H$\alpha$ sources or possible artifacts of an imperfect continuum subtraction. Further inspection and experimentation led to confidence that these are indeed real H$\alpha$ sources. Nonetheless, spectroscopic observations are required for confirmation. For the present paper, we will treat these H$\alpha$ sources as bona fide HII regions. Note that in the case of the Local Group dI DDO 210, van Zee et al. (1997) found a similar single H$\alpha$ detection to show broad Balmer lines and no forbidden lines, and they speculate that this source may be a luminous blue variable star.
Miller (1996) observed and detected HII regions in the Sculptor group dIs UGCA 442 ($=$ ESO 471-G06) and ESO 245-G05 ($=$ A 143), but did not detect H$\alpha$ emission from six other Sculptor group dIs. Note that we did [*not*]{} re-observe any of the Sculptor group dIs from Miller’s sample. Miller’s non-detection of HII regions in the Sculptor Dwarf Irregular Galaxy ($=$ SDIG, ESO 349-G31) has been confirmed by Heisler et al. (1997). van Zee (2000) has observed DDO 6 ($=$ UGCA 15) and did not detect any HII regions, but did detect a small amount of diffuse H$\alpha$ emission. van Zee (2000) also observed DDO 226 ($=$ UGCA 9) and did detect the presence of faint HII regions. Jerjen et al. (1998) detected a faint HII region in ESO 294-G010. ESO 410-G05 is an HI non-detection, and, based on HST imaging, Karachentsev et al. (2000) have determined that this is a dSph galaxy. In light of the results of the more recent deeper spectroscopy, it might be interesting to re-observe the other non-detection by Miller (1996), i.e., UGCA 438 ($=$ ESO 407-G18). Nonetheless, the claim by Miller (1996) that the average current star formation for dIs in the Sculptor group is suppressed relative to other nearby groups appears to be confirmed (see discussion in §3).
The HII Regions and the Star Formation Rates the Sculptor Group Dwarf Irregular Galaxies
========================================================================================
The HII Region Luminosities
---------------------------
As typical for dI galaxies, the majority of HII regions in these dwarfs have a random asymmetric distribution throughout the galaxies (Brosch et al. 1998; van Zee 2000). However, the most luminous HII regions are found at or near the center (in NGC 625, NGC 59 and ESO 347-G17). H$\alpha$ luminosities were calculated from the H$\alpha$ fluxes using the distances of Table 1 and assuming a Galactic extinction correction of the form: A(H$\alpha $) = 2.32 $E(B-V)$ (Miller & Hodge 1994), using reddening values from Schlegel et al. (1998). Figure 2 shows a histogram of the HII region luminosities for all the dwarfs. The bulk of them are mostly low luminosity, around 10$^{37.5}$ erg s$^{-1}$, very similar to what is found in other nearby irregulars (Youngblood & Hunter 1999) and isolated dIs (van Zee 2000). The lowest luminosity HII region, of 10$^{35.4}$ erg s$^{-1}$, is found in SC 24, the nearest dwarf in our sample and also the lowest luminosity dwarf of the group. For comparison, the Orion nebula has an H$\alpha$ luminosity of 10$^{37}$ erg s$^{-1}$ and 30 Doradus in the LMC has an H$\alpha$ luminosity of 1.5 $\times$ 10$^{40}$ erg s$^{-1}$ (Kennicutt 1984).
The most luminous HII regions are found in NGC 625, which is obviously undergoing an active phase of star formation, with its central region covered with bright HII regions. Its most extreme region (\#5 in Figure 1d) is actually a blend of at least 2 peaks (barely resolved with our image quality of 1), and the H$\alpha$ luminosity of the whole complex (4.5 $\times$ 10$^{39}$) is approaching that of the supergiant HII region 30 Doradus. HII regions with these high luminosities are characteristic of nearby BCDs and are even found in a few irregulars as well (e.g., Youngblood & Hunter 1999). In NGC 625, several arcs and filaments, typical signs of active star formation, are seen around the larger brighter complexes (regions \#4 and \#8 for example). Most interesting are the two spectacular ‘chimney’-like radial filaments, extending out of the plane of the galaxy perpendicular to the major axis (regions \#11 and \#12). These structures probably trace the conduits of outflows of hot gas upward into the halo from the inner star-forming regions. The two ‘chimney’-filaments extend to about 300 pc in length, in a galaxy with a total diameter (at $\mu _B$=25 mag arcsec$^{-2}$) of 3.6 kpc. This is larger than similar chimneys traced in other galaxies with sites of intense star-formation, e.g., the irregular NGC 4449 (Bomans et al. 1997) or those seen in the Galaxy (Normandeau et al. 1996). It is expected that in low mass galaxies such as NGC 625, the conditions for massive outflows of hot gas from galactic winds should be optimized, and, indeed, Bomans & Grant (1998) detected x-ray emission from NGC 625 with the ROSAT satellite observatory. One would expect a non-negligible fraction of the ISM to be driven out of the galaxy. High-resolution HI maps of NGC 625 show very unusual HI kinematics, with the gas in large-scale rotation around the major axis rather than the minor one (Côté et al. 2000), although no corresponding features at the location of the chimneys are seen (with the beam resolution of 15). Marlowe et al. (1997) did [*not*]{} detect line-splitting in H$\alpha$ (a signature of expanding bubbles), but their slit position did not cover the filaments labeled regions \#11 and \#12 (and it may be that the gas is preferentially expanding along the morphological minor axis, which is perpendicular to the line of sight).
Global Star Formation Rates and Timescales
------------------------------------------
Following Kennicutt, Tamblyn, & Congdon (1994), H$\alpha$ luminosities were converted to current SFRs with: $$SFR(total) = {{L(H\alpha )}\over{1.26\times 10^{41} erg\ s^{-1}}}\
M_\odot\ yr^{-1}$$ which has been derived for normal spiral galaxies with a modified Salpeter IMF. This is slightly different ($\sim$ 10% lower) from the conversion proposed by Kennicutt (1983) and is the conversion adopted by Kennicutt (1998) in his grand synthesis of global SFRs in galaxies. We calculated the rates in our dwarf galaxies in the same way although no systematic adjustment was made for internal extinction (as for spiral galaxies in Kennicutt 1998) since there is a wealth of evidence that extinction is usually quite small in these low metallicity systems.
Note that using a conversion from H$\alpha$ luminosity to SFR derived for normal spiral galaxies and applying it to dwarf galaxies is liable to both increased uncertainties and possible biases. The main sources of uncertainty arise from metallicity dependences of both the IMF and the production of ionizing photons by the stars. However, our main purpose here is to compare the present sample of dwarf galaxies to other samples of dwarf galaxies, and thus, it is most important to be consistent in the derivation of SFRs so that we can compare directly to other studies like that of van Zee (2001). The possible larger uncertainties and biases for the present sample will be similar to those of other studies of dwarf galaxies. Additionally, H$\alpha$ measurements of star formation rates are measuring massive star formation rates and carry an implicit assumption of a constant IMF.
Table 3 shows the results, where the H$\alpha$ luminosities used in Equation (2) are the sum of all the HII region luminosities in each galaxy. Summing over just the HII regions may underestimate the total H$\alpha$ luminosities by neglecting diffuse H$\alpha$ emission that “leaks out” of the HII regions (Hunter, Hawley, & Gallagher 1993; Ferguson et al. 1996). Youngblood & Hunter (1999) and van Zee (2000) note that, on average, the H$\alpha$ fluxes from the sums of individual HII regions underestimate the total H$\alpha$ emission by roughly a factor of two. However, van Zee (2000) points out that in low surface brightness (LSB) galaxies with low HII region coverage fractions, large aperture H$\alpha$ measurements are susceptible to significant errors with even small continuum subtraction errors. Thus, in the present paper, we have chosen to simply report the sum of the H$\alpha$ emission from the HII regions.
These SFRs vary over several orders of magnitude but most of them are extremely low, as the dwarfs of Sculptor are only experiencing modest star-forming activity. The SFR is known to be a strong function of absolute magnitude, for example, in the dwarf galaxies of the M81 Group (Miller & Hodge 1994), and thus we have calculated SFRs normalized to blue luminosities in Table 3. Normalizing to L(B) significantly decreases the range, and the extreme cases on the low end (ESO 348-G09 and SC 24) and the high side (ESO 473-G24 and NGC 625) are now easily recognized. The rates for the Sculptor dIs are consistent with those of the M81 dIs for a given magnitude, and our values here for extremely faint dwarfs like SC 18 and SC 24 extend this trend to fainter levels.
An interesting way to characterize the present SFR is to compare it to the total gas mass and estimate a gas depletion time scale, ($\tau_{gas}$), the number of years that a galaxy may continue to form stars at the current rate. HI gas masses and values of $\tau_{gas}$ are listed in Table 3. As pointed out by Kennicutt et al. (1994), $\tau_{gas}$ is really a lower limit because it does not account for the gas returned to the ISM by natural stellar evolution processes. Nonetheless, the quantity $\tau_{gas}$ is an interesting one, and the range in $\tau_{gas}$ for the Sculptor Group dIs is quite impressive. The currently star bursting NGC 625 shows a gas consumption time scale of only 3 Gyr, while several of the galaxies have consumption time scales on order or in excess of 10 times the present age of the Universe.
Our first calculation of the gas consumption time scale for the extremely quiescent galaxy SC 24 resulted in a value of 4800 Gyr, but we suspected that this was most likely due to an overestimate of its total HI gas mass. The width of the HI emission line is listed by Côté et al. (1997) as $\Delta$V$_{20}$ $=$ 92 km s$^{-1}$, which would be an unusually large line width for such a low luminosity galaxy. The value of M(HI)/L(B) (23.2) was also unusually large. Inspection of the 21 cm spectrum published by Côté et al. (1997) showed that the detection is close to the range of Galactic emission (with a heliocentric velocity of only 79 km s$^{-1}$) and much of the HI flux could be due to curvature in the continuum baseline that is not well fit. We inspected the HIPASS data base (Barnes et al. 2001) and found an HI detection at the position of SC 24 at a velocity of $\approx$ 80 km s$^{-1}$, but with a $\Delta$V$_{20}$ $\approx$ 25 km s$^{-1}$, and thus a flux integral of only 3.2 Jy km s$^{-1}$ (roughly one quarter that of the original Côté et al. (1997) value). Here we will use the HIPASS value, but interferometric HI observations of SC 24 will be necessary to truly resolve this issue. Note that with the new, lower value for the flux integral, M(HI)/L(B) ($=$ 6.29) is still high, and the gas consumption time scale of 1300 Gyr is still very large.
Figure 3 shows histogram comparisons of the values of $\tau_{gas}$ for the Sculptor Group dIs with three comparison groups, the Local Group dIs (from Mateo 1998), the gas-rich LSB galaxies studied by van Zee et al. (1997), and the isolated dIs of van Zee (2000, 2001). In comparison to the Local Group dIs, the distribution of Sculptor Group dIs shows a shift to larger values of $\tau_{gas}$ (although there is considerable overlap between the two samples). The range for both the Sculptor Group and Local Group samples is large when compared to that seen for the gas-rich quiescent LSB dIs and the isolated dIs. This is due, at least in part, to selection effects. The two nearby group samples have a higher proportion of extremely low luminosity galaxies, and the smaller galaxies are much more liable to large relative variations in current SFR, thus resulting in a large range in $\tau_{gas}$. Large gas content played a primary role in the selection of the LSB dIs, and a secondary role in the selection of the isolated dIs, so a bias toward larger values of $\tau_{gas}$ in these samples is expected.
Are the Sculptor Group dIs really that different from the Local Group dIs? The Local Group has two galaxies with values of log ($\tau_{gas}$) $\le$ 9, and the Sculptor Group has three galaxies with log ($\tau_{gas}$) $\ge$ 11.5, so, at least statistically these two distributions could be found to be different. Using the Kolmogorov-Smirnov (hereafter K-S) test (specifically [*kstwo*]{} of Press et al. 1992) to determine the significance of the differences between the distributions in Figure 3, we find that the probability that the Sculptor Group and Local Group galaxies are drawn from the same distributions is 30%. This emphasizes the large degree of overlap in the two samples. Interestingly, the K-S test yields a probability of 89% that the Sculptor Group and LSB samples are drawn from the same distribution. However, this reflects the fact that the K-S test has a higher sensitivity to the median values of distributions than to the spreads in distributions (Press et al. 1992); the standard deviations of the two samples differ by a factor of two. The F-test statistic ([*ftest*]{} from Press et al. 1992) indicates that the Sculptor Group and the LSB dIs have significantly different variances (significance of 99%). All other K-S test comparisons of the samples shown in Figure 3 yield probabilities of less than 10%. Note that since the current star formation rates represent a snapshot in time, we have no guarantee that these distributions might not have looked very different just a few 10$^8$ years ago.
Another important caveat should be added at this point; missing from Figure 3 are gas-rich galaxies with current star formation rates of zero (no H$\alpha$ emission). The Local Group has five of these dIrr/dSph or “transition” galaxies (Mateo 1998) and the Sculptor Group has three. Although not plotted in Figure 3, the consideration of these galaxies would increase the degree of overlap in the two samples. The transition galaxies are discussed in more detail in §4.
There are different physical reasons for a galaxy to have a very large value of $\tau_{gas}$. For example, van Zee et al. (1997) and Kennicutt & Skillman (2001) found values in excess of 200 Gyr for the LSB dI DDO 154. This appears to be mainly due to the very large HI halo surrounding this galaxy which contains 90% of the galaxy’s HI. In this case, most of this gas is probably unavailable for star formation. In the case of episodic star formation, it is possible to observe a galaxy with a current SFR that is well below its average SFR. This results in an artificially large value of $\tau_{gas}$. With detailed recent star formation histories like those available from imaging studies of the young stellar population (e.g., Dohm-Palmer et al. 1997, 1998) it is possible to get a better measure of the [*average*]{} recent SFR.
Often the H$\alpha$ and total luminosities of galaxies are used in order to investigate how the current SFR compares to some measure of the past average SFR. This ratio has been referred to as both the “star formation timescale” (Roberts 1963, Hodge 1993) which is the ratio of the mass of the stars present to the current rate of star formation and the “birthrate parameter” (Scalo 1986, Kennicutt et al. 1994) which is the ratio of the past average SFR to the current SFR. Deriving the past average SFR requires estimating the total mass of the stars formed over the lifetime of the galaxy. This can be done by converting the color of the stellar population into a representative mass-to-light ratio (e.g., Miller 1996). Alternatively, Kennicutt et al. (1994) convert the H$\alpha$ equivalent width directly into a birthrate parameter. Both methods depend on assumptions of stellar and galaxy evolution, and making assumptions about galaxy evolution in the study of galaxy evolution may produce misleading results (see discussion in van Zee 2001). Often these models are constructed with large, solar metallicity spiral galaxies in mind, so the direct applicability to low metallicity dwarfs carries an additional uncertainty.
Here we will follow Hodge (1993) and make the simple assumption of a mass-to-blue light ratio of 1 so that the ratio of the current SFR to the past average SFR ($\tau_{form}$) is directly comparable to the values listed in Hodge (1993). Assuming a single value of M/L for all of the galaxies is partially supported by the small range in colors observed for these galaxies. From the photometry of Côté (1995), we find that all but two of the galaxies listed in Table 3 have values of B$-$R between 0.7 and 1.0 (with ESO 471-G06 slightly bluer at 0.6 and ESO 245-G05 even bluer at 0.3). Thus, $\tau_{form}$ is simply L(B)/SFR with both given in solar units. This is listed in Table 3, column 4, which is simply the inverse of the values in column 3. Hodge (1993) studied the sample of galaxies in Kennicutt (1983) and found average values of $\tau_{form}$ of about 60 Gyr for early type spirals, 15 Gyr for later type spirals and 8 Gyr for irregular galaxies. The large values of $\tau_{form}$ found here indicate that the typical dI in the Sculptor Group is currently forming stars at a much lower rate than it has in the past, in support of the earlier finding by Miller (1996).
Figure 4 shows histogram comparisons of the values of $\tau_{form}$ for the Sculptor Group dIs with the same three comparison groups used in Figure 3. As in Figure 3 where $\tau_{gas}$ is compared, the range for $\tau_{form}$ in the Sculptor Group dI sample is shifted to larger values relative to the Local Group dI sample (again with considerable overlap between the two samples). The Local Group sample stands out with three galaxies with relatively low values of $\tau_{form}$ (IC 10, GR 8, and NGC 6822). Once again, this is due, in part, to selection effects. The gas-rich quiescent LSB dI sample and the isolated dI sample excluded galaxies with exceptionally high values of current SFR (HII galaxies or blue compact dwarf galaxies). It is also interesting to note the suggestion that IC 10 should be considered to be a blue compact galaxy (Richer et al. 2001 and references therein).
Are the Sculptor Group and Local Group dI distributions different in Figure 4? Here the K-S test gives a probability of only 10% that the Local Group and Sculptor Group samples are drawn from the same distribution. This is due, in part, to the presence of IC 10 and GR 8 in the Local Group sample. If these two galaxies are deleted from the Local Group sample, then the K-S test probability rises to 22%, on the margin of a significant difference. However, IC 10 and GR 8 do exist, so it makes no sense to delete them from the sample. The K-S test yields a 71% probability for the Sculptor Group sample and the LSB sample to be drawn from the same distribution. The Sculptor Group and LSB samples are very similar with regard to mean values of both $\tau_{gas}$ and $\tau_{form}$, while the the Local Group and isolated dI samples are also similar to each other in these two quantities.
Interestingly, the K-S test of the Figure 4 data reveals that the Local Group dI sample has a 51% probability of being drawn from the same distribution as the isolated dI sample, and a 15% probability when compared to the LSB dI sample. These are significantly higher than the respective $\tau_{gas}$ comparisons (8% and 5% respectively). In fact, the Local Group and isolated dI sample distributions in Figure 4 do not look that similar, and the high significance once again reflects the fact that the K-S test has a higher sensitivity to the median values of distributions than to the spreads in distributions.
In Figure 5 we compare the two timescale values ($\tau_{gas}$ and $\tau_{form}$) to each other for the four samples. Figure 5 shows a general correlation of $\tau_{gas}$ with $\tau_{form}$ (as should be expected since both axes share a common denominator). The main reason for the relatively good correlation in Figure 5 is that the gas-rich dwarf galaxies generally show a limited range in M(HI)/L(B) (Skillman 1996). The dotted line in Figure 5 represents an equality between $\tau_{gas}$ and $\tau_{form}$, or, equivalently, M(HI)/L(B) $=$ 1.0. The conclusions of Figure 3 and 4 are supported. While the Sculptor Group dIs and the Local Group dIs tend to fall into the region of the graph that is well populated by the larger LSB and isolated dI samples, it is interesting that the Sculptor group dIs lie at preferentially higher values of log($\tau_{gas}$) with the majority above 10.5, while the Local Group dIs lie at preferentially lower values of log($\tau_{gas}$) with the majority below 10.5. It is interesting that, when normalized by SFR, the LSB galaxies do not deviate far from the trend. Naively, one might expect them to deviate to the upper right (higher gas content and lower luminosity), but, in fact, they do not appear to distinguish themselves in this diagnostic diagram. This is simply due to the fact that the dynamic range in the SFR is much larger than the dynamic range in M(HI)/L(B).
Finally, in Figure 6, we have plotted the SFR versus gas mass with both normalized to the galaxy luminosity (SFR/L$_B$ versus M$_{HI}$/L$_B$). In this diagnostic diagram excursions are much more prominent. The high relative SFR galaxies from the Local Group are not unusually gas rich. The Sculptor Group galaxies tend toward larger gas fractions and lower values of SFR (note the large number of them in the lower right quadrant). The isolated dIs show a large range in values, reaching toward the extremes seen in the two group samples. It would appear that the isolated dIs sample of van Zee (2000, 2001) covers the range of properties of dIs with the possible exception of the lowest luminosity systems.
Comparing “Transition” Galaxies and dI Galaxies
===============================================
Of course, the very large values of $\tau_{gas}$ seen in several of the Sculptor Dwarf dIs do not represent an upper bound; there are several Sculptor Group dIs with HI detected but no H$\alpha$ emission detected, and these would have $\tau_{gas}$ values of infinity. Such galaxies are not limited to the Sculptor Group; in the Local Group we have LGS-3, Antlia, and DDO 210 as examples of dIs with detectable HI but no detectable H$\alpha$ emission (Mateo 1998). These galaxies are often referred to as “transition” galaxies or dIrr/dSph galaxies. Mateo (1998) also includes Pegasus and Phoenix in this category. Pegasus has two detected HII regions (Skillman, Bomans, & Kobulnicky 1997) and the formal calculation of $\tau_{gas}$ for Pegasus yields a value of 3220 Gyr. Until recently, an optical radial velocity was unavailable for Phoenix, and it was not clear whether the HI detected in the direction of Phoenix was directly associated with it (Carignan, Demers, & Côté 1991; Oosterloo et al. 1996, Young & Lo 1997, St-Germain et al. 1999). However, Gallart et al. (2001) have now provided an optical radial velocity of $-$52 $\pm$ 6 km s$^{-1}$, which is relatively close to the HI cloud separated from Phoenix by 6$\arcmin$ with a velocity of $-$23 km s$^{-1}$, and they conclude that the properties of this HI cloud are consistent with having been recently lost by Phoenix. The recent measurement of the optical radial velocity of Phoenix of $-$13 $\pm$ 9 km s$^{-1}$ by Irwin & Tolstoy (2002) strengthens the connection between the HI cloud and the galaxy. Note that the stellar population study by Holtzman, Smith, & Grillmair (2000) indicates that Phoenix has experienced star formation up until roughly 100 million years ago which implies that Phoenix must have had some gas until very recently.
For the galaxies with the very low values of SFR (or high values of $\tau_{gas}$ and $\tau_{form}$), the precise value of the current SFR is probably not very meaningful. The conversion from H$\alpha$ flux to SFR calculated by Kennicutt et al. (1994) is based on a fully populated IMF, and with so few HII regions, it is clear that the whole range of massive stars is not represented. Since the presence or absence of a few HII regions can move an extremely low luminosity galaxy between the dI and transition categories, they may have much in common. One also has to consider the uncertainty introduced by the differences between the H$\alpha$ fluxes calculated from the sum of the HII regions and those calculated from the entire image (including the diffuse component) which can be significant for these very quiescent galaxies.
There are at least two different evolutionary paths that a dI galaxy can take to become a transition galaxy. One possibility is for the galaxy to lose enough of its cold gas to halt present star formation. A second possibility is to have sufficient gas for star formation, but to be simply “in between” episodes of star formation. Phoenix may be a local example of the first case (as discussed by Gallart et al. 2001), while Antlia, DDO 210, LGS-3, and Pegasus may be examples of the second case. Typically, dI galaxies have M$_{HI}$/L ratios of about one in solar units (Skillman 1996). Thus, galaxies with similar M$_{HI}$/L but no current star formation could be what would normally be called dIs, but are simply between episodes of galaxy formation. Simply based on the number of stars and the average lifetimes of HII regions, the smaller the dI galaxy, the higher the likelihood that it could be found in such a phase. Note that all five of the Local Group transition galaxies have absolute V magnitudes near the low end of the dI luminosity function (DDO 210 $=$ $-$10.0, Phoenix $=$ $-$10.1, LGS-3 $=$ $-$10.5, Antlia $=$ $-$10.8, and Pegasus $=$ $-$12.9; Mateo 1998). Of the five Local Group transition galaxies, all have significant HI contents with values of M$_{HI}$/L$_V$ close to one (DDO 210 $=$ 2.35, Phoenix $=$ 0.21, LGS-3 $=$ 0.32, Antlia $=$ 0.56, and Pegasus $=$0.45; Mateo 1998); the value for Phoenix represents the candidate HI cloud at $-$23 km s$^{-1}$. Thus, four of the transition galaxies in the Local Group are consistent with the evolutionary scenario of temporarily interrupted star formation.
In a detailed study of LGS-3, Miller et al. (2001) have shown how the SFR in the outer parts of LGS-3 has decreased faster than the SFR in the inner parts. If the “active” star forming area of a galaxy decreases with time, then the likelihood that it will experience a “transition” phase with no current massive star formation will increase with time. Nonetheless, it may retain sufficient gas for future episodes of star formation.
The Sculptor Group, with three galaxies with clear transition type properties, provides us with additional case studies of transition type galaxies. SDIG, DDO 6, and UGCA 438 are all HII region non-detections with HI detections and M$_{HI}$/L$_B$ values close to one (0.88, 1.4, and 2.4 respectively). ESO 294-G10 is an HI non-detection from both Côté et al. (1997) and the HIPASS database, but has an HII region observed (Jerjen et al. 1998), so it might be interesting to re-observe it in HI. Additionally, four of the galaxies in Table 3 have values of $\tau_{gas}$ larger than 100 Gyr with values of M$_{HI}$/L$_B$ typical for dIs (ESO 471-G06, ESO 348-G09, SC 18, and SC 24).
Although there has not been a great deal of research on transition type galaxies, the fact that they (and galaxies with similar properties) turn up frequently in the Local Group (5 examples) and Sculptor Group (3 examples) indicates that this type of galaxy may be very common among the lowest luminosity, gas-rich, star forming galaxies. It is then an interesting question whether the spatial distribution of transition type galaxies in our Local Group and the Sculptor Group is different from the spatial distribution of normal dIs. In Figure 7 we plot the spirals, dIs, and transition galaxies from the Local and Sculptor groups in Supergalactic coordinates (cf. Jerjen et al. 1998, Figure 9). The distances for the spirals are taken from Jerjen et al. (1998), but we have added NGC 24 at a distance corresponding to Equation (1). The dEs, which tend to cluster near the spiral galaxies, are left out for clarity. Inspection of Figure 7 shows no obvious pattern to distinguish the dIs from the transition galaxies. If the low SFRs in the transition galaxies were due to the influence of the large spiral galaxies, we might expect to see evidence of a morphology density relationship like that observed for the dE galaxies (e.g., Binggeli, Tarenghi, & Sandage 1990 and references therein), but none is immediately evident in Figure 7.
However, Figure 8 shows the same galaxies in another projection, that of the Supergalactic X,Z plane. By coincidence, the Sculptor Group is nearly centered on the Supergalactic Y-axis (de Vaucouleurs 1958, 1975). Thus, if the Local Group – Sculptor Group concentration is modeled as a cylinder like filament aligned along the Supergalactic Y-axis, this projection shows the typical displacement from the center of that concentration. Figure 8 gives the appearance that the transition galaxies show a concentration to the center of this filament similar to that seen by the spirals, while the dIs appear to have larger radial distances.
Figure 9 is a histogram of the radial distances from this axis sorted by galaxy type. Average radial distances and standard deviations have been calculated for each of the types. Here we are including the dE galaxies for the first time. The distances for the Local Group dE galaxies are taken from Mateo (1998) and van den Bergh (2000). The distances of the six known Sculptor Group dEs come from Jerjen et al. (1998: NGC 59, SC 22, ESO 294-G010, ESO 540-G030), Karachentsev et al. (2000: ESO 410-G005), and Jerjen & Rejkuba (2001: ESO 540-G032). The histogram confirms the impression that the radial distributions of the transition galaxies are very similar to those of the spiral galaxies and dE galaxies and distinctly different from those of the dIs. If the galaxies are separated into the Local and Sculptor groups, the offset is still present. The main difference is the significant population of dIs more than 1 Mpc from the cloud axis. The bimodality of the dE distribution simply reflects the fact that most of these galaxies are either satellites of M 31 (with an offset from the Supergalactic Y-axis of 0.7 Mpc) or the Milky Way (with no offset).
Figure 10 shows the results of the calculation of the distances of the dI, transition, and dE galaxies to the nearest spiral galaxy inspired by the results of Figures 8 and 9. In this figure, the Sculptor Group dwarfs are noted by dots in the histograms. The main difference between the dI and transition galaxies in Figure 10 is the lack of transition galaxies at large distances, with the result that the average distance for the dIs is almost a factor of two larger than that of the transition galaxies. This is a marginally stronger result from that shown in Figure 9 (0.82/0.58 $=$ a factor of 1.4 for average radial distance from the Cloud axis versus 0.92/0.50 $=$ a factor of 1.9 for average distance to the nearest spiral). The K-S test indicates that all three distributions in Figure 10 are significantly different. Comparing the transition galaxies to the other two samples yields only a 15% probability that the transition galaxies come from the same sample as the dIs and only an 8% probability of coming from the same sample as the dEs.
The distances calculated for Figure 10 carry a larger degree of uncertainty than those in Figure 9 since the distances to most of the Sculptor dwarfs are based on the distance-velocity relationship in Equation (1). Nonetheless, it is natural to ask the question whether the transition galaxies are found preferentially near to the spiral galaxies or to the axis of the Local Group – Sculptor Cloud that is defined by the spiral galaxies. On its own, the near identity of the means and standard deviations of the transition galaxy distances to the group axis (0.58, 0.24) and the nearest spiral (0.50, 0.34) prevent us from distinguishing which effect may be more relevant for the evolutionary status of these galaxies. Of course, the results for the nearest spiral galaxy distances may change as better distances become available for the bulk of the sample.
In comparing the transition galaxies to the dE galaxies we find a significant difference between the distributions in Figures 9 and 10. The ratio of the average distance to the nearest spiral (2.3 times larger for the transition galaxies relative to the dE galaxies) is much larger than the ratio of the average distance to the group axis (1.5 times larger for the transition galaxies). However, this is mainly reflecting the offset of M 31 (and therefore its companion dE galaxies) from the group axis. Although the samples are admittedly small, the impression is that the three dwarf morphological types are separable by mean distance to nearest neighbor spiral galaxy. This would support the hypothesis that a key factor affecting dwarf morphological type is environmental.
There is, of course, the possibility that Figure 10 is susceptible to the biases of selection effects. It is a safe bet that there remain low luminosity, LSB galaxies undetected in the Sculptor Group. Recently, most of the new low luminosity dEs in the Local Group have been found by searching in the vicinity of M 31 (e.g., Armandroff, Jacoby, & Davies 1999), and this might skew the distribution of Local Group dEs towards higher proximity. Additionally, the Sculptor dEs do not show the same degree of proximity to the spiral galaxies as the Local Group dEs. In the histogram in Figure 10, the Sculptor Group dEs (marked by dots) account for 4 of 23 of the galaxies in the main clump and 2 of 4 of the outliers. On the other hand, the distances to the Sculptor Group dEs are much more uncertain, and this may introduce some scatter. For example, for ESO 540-G030 and ESO 540-G032, the calculated nearest spiral neighbor (NGC 247) distances are 560 kpc and 670 kpc respectively, but the minimum projected distances are 120 kpc and 50 kpc. Ironically, the surface brightness fluctuation distance for ESO 540-G032 placed it very close to NGC 247, but the newer tip of the red giant branch measurement moved it over 1 Mpc further away, so that its nearest neighbor changed to NGC 253. It could be that the distances are still quite uncertain, and that all of the Sculptor dEs could lie in the main group in the histogram determined by the Local Group dEs. Hubble Space Telescope distances from tip of the red giant branch measurements for all of these galaxies are required to place this comparison on firmer ground.
The Nature of “Transition” Galaxies
===================================
The present data indicate that, in the local cloud, transition galaxies are differentiated from dIs by their, on average, closer proximity to spiral galaxies. Since transition galaxies are preferentially found to be lower luminosity systems than the average dI galaxy, a luminosity – distance relationship for the dIs might cause such difference. Indeed, Armandroff et al. (1999) show that the extreme low luminosity dSph galaxies in the Local Group are found preferentially close to our Milky Way galaxy. However, this does not appear to be the case for the Local Group dIs, since the Local Group has several relatively low luminosity dIs at large distances ($\ge$ 1 Mpc) from the nearest spiral galaxies. This points to an underlying cause more closely related to environment (something is happening to these galaxies) as opposed to initial conditions (it is not true that smaller dI galaxies are formed closer to spirals).
What can be made of the morphology-density relationship for the transition galaxies? The transition galaxies do [*not*]{} show the same degree of morphology-density relationship as the dwarf spheroidals (i.e., the typical transition galaxy is not a close ($\le$ 200 kpc) companion to a large spiral galaxy). Therefore, explanations of the morphology-density relationship applicable to the dwarf spheroidals are not necessarily appropriate for the transition galaxies. Because there is a large degree of overlap in the spatial distributions of the transition galaxies and the dIs, it is likely that the typical transition galaxy represents the extreme in the distribution of low SFRs of quiescent dI galaxies. The observation that the transition galaxies tend toward the center of the filamentary group structure, and lie closer, on average, to the spiral galaxies, may be telling us that environmental effects have driven otherwise normal dI galaxies to the observed low values of SFR.
Perhaps these transition galaxies have a higher probability for “tidal stirring” or “harassment” than the typical dI, resulting in their partial evolution to the dwarf spheroidal state (Mayer et al. 2001a,b). Although such interactions result in a temporarily elevated SFR, the main long-term effect of these interactions is to strip the galaxy of gas, resulting in lower SFRs. Given the small samples involved, and the limited range in environment that has been sampled here, definitive statements cannot be made. It is not clear whether galaxy luminosity (small galaxies are most easily found in a quiescent state), position relative to the “cloud” structure (galaxies near the centers of large scale structures are most likely to have the properties of transition galaxies), or position relative to the nearest spiral galaxy (galaxies nearest to spiral galaxies are most likely to be tidally stirred) is the dominant variable in determining that a gas-rich dwarf has the properties of a transition galaxy. Similar studies of other environments should provide additional constraints on this problem. Additionally, clearer observational predictions from the models would be helpful. Can tidally stirred galaxies be differentiated from isolated galaxies by bursty star formation histories or strong stellar population radial gradients? Can a burst of star formation triggered by an external influence be distinguished from a burst with an internal trigger?
Finally, while our conclusion that the transition galaxies have more in common with the dI galaxies than with the dE galaxies may be debatable, it is clear that transition galaxies are not dE galaxies. The similarities between transition galaxies and dI galaxies in gas content and spatial distribution may make it natural to group the transition galaxies with the dIs in making comparative family studies. Note that this is the opposite of the assumption of Blitz & Robishaw (2000) in their study of “gas-rich dwarf spheroidals.” By including the transition galaxies in with the dwarf spheroidals, they concluded that a large fraction of the dwarf spheroidals in the Local Group are associated with HI gas. We would argue that the transition galaxies should be excluded from this sample, leading to the opposite conclusion, that only a very small fraction (if any) of dwarf spheroidals in the Local Group are associated with HI gas.
Conclusions
===========
We have presented H$\alpha$ imaging of dI galaxies in the nearby Sculptor Group. Comparing the Sculptor Group dIs to the Local Group dIs, in support of Miller (1996), we find that the Sculptor Group dIs have, on average, lower values of SFR when normalized to either galaxy luminosity or gas mass (although there is considerable overlap between the two samples). The range for both the Sculptor Group and Local Group samples is large when compared to that seen for the sample of gas-rich quiescent LSB dIs from van Zee et al. (1997) and the sample of isolated dIs from van Zee (2000, 2001). This is probably best understood as a selection effect since the nearby group samples have a much larger fraction of extremely low luminosity galaxies and the smaller galaxies are much more liable to large relative variations in current SFR. The Sculptor Group and LSB samples are very similar with regard to mean values of both $\tau_{gas}$ and $\tau_{form}$, and the the Local Group and isolated dI samples are also similar to each other in these two quantities.
The properties of “transition” galaxies in Sculptor and the Local Group are also compared and found to be similar. The transition galaxies are preferentially among the lowest luminosities of the gas rich dwarf galaxies. Relative to the dwarf irregular galaxies, the transition galaxies are found preferentially nearer to spiral galaxies, and are found nearer to the center of the mass distribution in the local cloud. However, the transition galaxies are not found as close to the nearest neighbor spiral galaxies as the dE galaxies. Most of these systems are consistent with normal dI galaxies which currently exhibit temporarily interrupted star formation. The observed density-morphology relationship indicates that environmental processes such as “tidal stirring” may play a role in causing their lower SFRs.
Special thanks are extended to Chris Lidman for acquiring the ESO 473-G24 H$\alpha$ observations, and to Liese van Zee for the use of her astrometry package to derive the HII region positions. We wish to thank G. Bothun, R. Kennicutt, E. Tolstoy, and L. van Zee for many helpful conversations. John Cannon, Henry Lee, and Liese van Zee proofread and provided valuable comments on this manuscript. We also thank the referee for a prompt and careful reading of the manuscript and several valuable comments. This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. This research has made use of NASA’s Astrophysics Data System Abstract Service. EDS acknowledges partial support from a NASA LTSARP grant No. NAG5-9221 and the University of Minnesota. BWM is supported by the Gemini Observatory, which is operated by the Association of Universities for Research in Astronomy, Inc., on behalf of the international Gemini partnership of Argentina, Australia, Brazil, Canada, Chile, the United Kingdom, and the United States of America. We would like to dedicate this paper to the memory of Robert A. Schommer, whose many contributions to the astronomical community, and the Chilean astronomical community in particular, will always be appreciated.
Armandroff, T. E., Jacoby, G. H., & Davies, J. E. 1999, , 118, 1220
Babul, A. & Rees, M. J. 1992, , 255, 346
Barkana, R., & Loeb, A. 1999, ApJ, 523, 54
Barnes, D. G. et al. 2001, , 322, 486
Binggeli, B., Tarenghi, M., & Sandage, A. 1990, A&A, 228, 42
Blitz, L. & Robishaw, T. 2000, , 541, 675
Bomans, D.J., Chu, Y.H., & Hopp, U. 1997, , 113, 1678
Bomans, D.J., & Grant, M.-B. 1998, Astron. Nach., 319, 26
Brosch, N., Heller, A., & Almoznino, E. 1998, , 300, 1091
Bullock, J.S., Kravtsov, A.V., & Weinberg, D.H. 2000, ApJ, 539, 517
Carraro, G., Chiosi, C., Girardi, L., & Lia, C. 2001, , 327, 69
Carignan, C., Demers, S. & Côté, S. 1991, ApJ, 381, L13
Côté, S. 1995, Ph.D. Thesis, Australian National University
Côté, S., Carignan, C., & Freeman, K.C. 2000, , 120, 3027
Côté, S., Freeman, K. C., Carignan, C., & Quinn, P. 1997, AJ, 114, 1313
de Vaucouleurs, G. 1958, AJ, 63, 253
de Vaucouleurs, G. 1975, in Stars and Stellar Systems 9, Galaxies and the Universe, ed. A. Sandage, M. Sandage, & J. Kristian (Chicago: Univ. Chicago Press), 557
de Vaucouleurs, G., de Vaucouleurs, A., Corwin, H. G. Jr., Buta, R. J., Paturel, G., & Foqué, P. 1991, Third Reference Catalog of Bright Galaxies, (New York: Springer) (RC3)
Dohm-Palmer, R. C. et al. 1997, , 114, 2527
Dohm-Palmer, R. C. et al. 1998, , 116, 1227
Efstathiou, G. 1992, , 256, 43P
Ferguson, A. M. N., Wyse, R. F. G., Gallagher, J. S., & Hunter, D. A. 1996, , 111, 2265
Gallagher, J. S., Tolstoy, E., Dohm-Palmer, R. C., Skillman, E. D., Cole, A., Hoessel, J., Saha, A., & Mateo, M. 1998, AJ, 115, 1869
Gallart, C., Martinez-Delgado, D., Gomez-Flechoso, M.A., Mateo, M. 2001, AJ, 121, 2572
Gnedin, N. 2000, ApJ, 535, L75
Giuricin, G., Marinoni, C., Ceriani, L., & Pisani, A. ApJ, 543, 178
Heisler, C.A., Hill, T.L., McCall, M.L., Hunstead, R.W. 1997, , 285, 374
Hodge, P. 1993, in Star Formation, Galaxies, and the Interstellar Medium, eds. J. Franco, F. Ferrini, & G. Tenorio-Tagle, Cambridge University Press, 294
Holtzman, J. A., Smith, G. H., & Grillmair, C. 2000, , 120, 3060
Hunter, D. A., Hawley, W. N., & Gallagher, J. S. 1993, AJ, 106, 1797
Irwin, M. & Tolstoy, E. 2002, , 336, 643
Jerjen, H., Freeman, K. C., & Binggeli, B. 1998, AJ, 116, 2873
Jerjen, H., Binggeli, B., & Freeman, K. C. 2000, AJ, 119, 593
Jerjen, H., & Rejkuba, M. 2000, A&A, 371, 487
Karachentsev, I. D. et al. 2000, , 542, 128
Kennicutt, R.C. Jr. 1983, , 272, 54
Kennicutt, R.C. Jr. 1984, , 287, 116
Kennicutt, R.C. Jr. 1998, , 498, 541
Kennicutt, R.C. Jr., & Hodge, P.W. 1986, , 306, 130
Kennicutt, R.C. Jr., & Skillman, E.D. 2001, , 121, 1461
Kennicutt, R.C. Jr., Tamblyn, P., & Congdon, C.W. 1994, , 435, 22
Klypin, A., Kravtsov, A. V., Valenzuela, O,, & Prada, F. 1999, ApJ, 522, 82
Knezek, P. M., Sembach, K. R., & Gallagher, J. S., III 1999, ApJ, 514, 119
Marlowe, A. T., Meurer, G. R., Heckman, T. M., & Schommer, R. 1997, , 112, 285
Mateo, M. 1998, ARA&A, 36, 435
Mayer, L., Governato, F., Colpi, M., Moore, B., Quinn, T., Wadsley, J., Stadel, J., & Lake, G. 2001a, ApJ, 547, L123
Mayer, L., Governato, F., Colpi, M., Moore, B., Quinn, T., Wadsley, J., Stadel, J., & Lake, G. 2001b, , 559, 754
Miller, B. W. 1994, Ph.D. Thesis, University of Washington
Miller, B. W. 1996, AJ, 112, 991
Miller, B. W., Dolphin, A. E., Lee, M. G., Kim, S. C., & Hodge, P. 2001 , 562, 713
Miller, B.W., & Hodge, P. 1994, , 427, 656
Moore, B., Ghigna, S., Governato, F., Lake, G., Quinn, T., Stadel, J., & Tozzi, P. 1999, ApJ, 524, L19
Normandeau, M., Taylor, A.R., & Dewdney, P.E. 1996, Nature, 380, 687
Oosterloo, T., Da Costa, G.S., & Staveley-Smith, L. 1996, AJ, 112, 1969
Press, W. H., Teukolsky, S. A., Vetterling, W. T., & Flannery, B. P. 1992, Numerical Recipes in Fortran, Cambridge University Press
Puche, D., & Carignan, C. 1988, AJ, 95, 1025
Quinn, T., Katz, N., & Efstathiou, G. 1996, , 278, L49
Richer, M. G. et al. 2001, , 370, 34
Roberts, M.S. 1963, ARA&A, 1, 149
Sandage, A., & Hoffman, G.L. 1991, ApJ, 379, 45
Scalo, J.M. 1986, Fund. Cos. Phys., 11, 1
Schlegel, D.J., Finkbeiner, D.P., & Davis, M. 1998, , 500, 525
Skillman, E. D. 1996, ASP Conf. Ser. 106: The Minnesota Lectures on Extragalactic Neutral Hydrogen, 208
Skillman, E. D., Bomans, D. J., & Kobulnicky, H. A. 1997, ApJ, 474, 205
St-Germain, J., Carignan, C., Côté, S., & Oosterloo, T. 1999, AJ, 118, 1235
Tully, R.B., & Fisher, J.R. 1987, Nearby Galaxies Atlas, Cambridge University Press
van den Bergh, S. 1994a, AJ, 107, 1328
van den Bergh, S. 1994b, ApJ, 428, 617
van den Bergh, S. 2000, PASP, 112, 529
van Zee, L. 2000, , 119, 2757
van Zee, L. 2001, , 121, 2003
van Zee, L., Haynes, M. P., & Salzer, J. J. 1997a, AJ, 114, 2479
Whiting, A. B. 1999, AJ, 117, 202
Young, L.M., & Lo, K.Y. 1997, ApJ, 490, 710
Youngblood, A.J., & Hunter, D.A. 1999, , 519, 55
[lcccccc]{} ESO 347-G17 & &23:26:57 &$-$37:20:47&695 &6.99 &$-$14.79\
ESO 471-G06 &UGCA 442 &23:43:47 &$-$31:57:09&281 &3.51 &$-$14.06\
ESO 348-G09 & &23:49:23 &$-$37:46:19&640 &6.52 &$-$13.75\
SC 18 & &00:00:59 &$-$41:09:18&119 &2.14 &$-$9.20\
NGC 59 &ESO 539-G04&00:15:25 &$-$21:26:36&392 &4.39 &$-$15.30\
ESO 473-G24 & &00:31:20 &$-$22:46:02&571 &5.94 &$-$12.68\
SC 24 & &00:36:38 &$-$32:34:28& 61 &1.66 &$-$8.39\
DDO 226 &UGCA 9 &00:43:04 &$-$22:14:49&386 &4.39 &$-$13.49\
&IC 1574 &\
&ESO 474-G18&\
DDO 6 &UGCA 15 &00:49:49 &$-$21:00:54&318 &3.82 &$-$12.24\
&ESO 540-G31&\
AM 0106-382 & &01:08:22 &$-$38:12:33&593 &6.13 &$-$12.47\
NGC 625 &ESO 297-G05&01:35:06 &$-$41:26:05&332 &3.93 &$-$16.31\
ESO 245-G05 &A 143 &01:45:04 &$-$43:35:53&310 &3.75 &$-$15.59\
[lcccc]{} ESO 347-G17 \#1 & 23:26:58.87 & -37:20:55.2 & 19 $\pm$ 4 & 38.1\
ESO 347-G17 \#2 & 23:26:58.34 & -37:20:56.0 & 2.1 $\pm$ 0.4 & 37.1\
ESO 347-G17 \#3 & 23:26:58.00 & -37:20:54.3 & 3.1 $\pm$ 0.6 & 37.3\
ESO 347-G17 \#4 & 23:26:57.80 & -37:20:49.4 & 1.6 $\pm$ 0.3 & 37.0\
ESO 347-G17 \#5 & 23:26:57.13 & -37:20:49.4 & 59 $\pm$ 12 & 38.6\
ESO 347-G17 \#6 & 23:26:56.41 & -37:20:52.1 & 2.3 $\pm$ 0.5 & 37.1\
ESO 347-G17 \#7 & 23:26:56.44 & -37:20:39.3 & 10.3 $\pm$ 2.1 & 37.8\
ESO 347-G17 \#8 & 23:26:55.88 & -37:20:49.6 & 4.9 $\pm$ 1.0 & 37.5\
ESO 347-G17 \#9 & 23:26:55.61 & -37:20:50.6 & 1.6 $\pm$ 0.3 & 37.0\
ESO 347-G17 \#10 & 23:26:55.15 & -37:20:51.0 & 25 $\pm$ 5 & 38.2\
ESO 347-G17 \#11 & 23:26:53.36 & -37:20:51.9 & 2.8 $\pm$ 0.6 & 37.2\
& & &\
ESO 348-G9 \#1 & 23:49:27.48 & -37:46:23.8 & 1.0 $\pm$ 0.2 & 36.7\
ESO 348-G9 \#2 & 23:49:27.07 & -37:46:24.9 & 0.4 $\pm$ 0.1 & 36.3\
ESO 348-G9 \#3 & 23:49:26.52 & -37:46:16.9 & 4.7 $\pm$ 0.9 & 37.4\
ESO 348-G9 \#4 & 23:49:21.67 & -37:46:21.3 & 0.7 $\pm$ 0.1 & 36.6\
& & &\
SC 18 \#1 & 00:00:58.02 & -41:09:24.2 & 5.3 $\pm$ 0.5 & 36.5\
& & &\
NGC 59 \#1 & 00:15:25.67 & -21:26:40.1 & 26 $\pm$ 5 & 37.8\
NGC 59 \#2 & 00:15:25.25 & -21:26:40.9 & 130 $\pm$ 26 & 38.5\
NGC 59 \#3 & 00:15:24.93 & -21:26:39.7 & 4.9 $\pm$ 1.0 & 37.1\
NGC 59 \#4 & 00:15:25.03 & -21:26:45.9 & 1.9 $\pm$ 0.4 & 36.7\
& & &\
ESO 473-G24 \#1 & 00:31:23.12 & -22:46:05.7 & 9.4 $\pm$ 1.9 & 37.6\
ESO 473-G24 \#2 & 00:31:22.16 & -22:46:14.6 & 12.3 $\pm$ 2.5 & 37.7\
ESO 473-G24 \#3 & 00:31:22.02 & -22:46:18.2 & 8.9 $\pm$ 1.8 & 37.6\
ESO 473-G24 \#4 & 00:31:21.28 & -22:46:14.1 & 7.0 $\pm$ 1.4 & 37.5\
& & &\
SC 24 \#1 & 00:36:38.35 & -32:34:45.9 & 0.8 $\pm$ 0.1 & 35.4\
& & &\
AM 0106-382 \#1 & 01:08:21.87 & -38:12:33.7 & &\
AM 0106-382 \#2 & 01:08:22.57 & -38:12:35.6 & &\
AM 0106-382 \#3 & 01:08:22.33 & -38:12:42.5 & &\
AM 0106-382 \#4 & 01:08:21.23 & -38:12:42.6 & &\
AM 0106-382 \#5 & 01:08:21.13 & -38:12:45.9 & &\
& & &\
NGC 625 \#1 & 01:35:07.36 & -41:26:11.2 & 30 $\pm$ 3 & 37.8\
NGC 625 \#2 & 01:35:07.23 & -41:25:52.5 & 1.4 $\pm$ 0.1 & 36.4\
NGC 625 \#3 & 01:35:06.78 & -41:26:00.1 & 15.3 $\pm$ 1.5 & 37.5\
NGC 625 \#4 & 01:35:06.86 & -41:26:05.9 & 193 $\pm$ 19 & 38.6\
NGC 625 \#5 & 01:35:06.68 & -41:26:13.0 & 2100 $\pm$ 210 & 39.6\
NGC 625 \#6 & 01:35:06.53 & -41:26:19.8 & 51 $\pm$ 5 & 38.0\
NGC 625 \#7 & 01:35:06.34 & -41:25:52.8 & 14.1 $\pm$ 1.4 & 37.4\
NGC 625 \#8 & 01:35:06.11 & -41:26:04.2 & 23.0 $\pm$ 2.3 & 37.6\
NGC 625 \#9 & 01:35:05.85 & -41:26:11.2 & 470 $\pm$ 47 & 39.0\
NGC 625 \#10 & 01:35:05.86 & -41:25:59.6 & 22.4 $\pm$ 2.2 & 37.6\
NGC 625 \#11 & 01:35:05.73 & -41:25:44.3 & 12.0 $\pm$ 1.2 & 37.4\
NGC 625 \#12 & 01:35:05.20 & -41:25:49.4 & 7.5 $\pm$ 0.8 & 37.2\
NGC 625 \#13 & 01:35:05.36 & -41:26:07.5 & 19.3 $\pm$ 1.9 & 37.6\
NGC 625 \#14 & 01:35:05.54 & -41:26:15.5 & 32 $\pm$ 3 & 37.8\
NGC 625 \#15 & 01:35:05.26 & -41:26:12.4 & 32 $\pm$ 3 & 37.8\
NGC 625 \#16 & 01:35:05.16 & -41:26:15.1 & 14.1 $\pm$ 1.4 & 37.4\
NGC 625 \#17 & 01:35:04.59 & -41:26:18.1 & 21.1 $\pm$ 2.1 & 37.6\
NGC 625 \#18 & 01:35:04.80 & -41:26:09.3 & 161 $\pm$ 16 & 38.5\
NGC 625 \#19 & 01:35:04.64 & -41:26:01.4 & 4.7 $\pm$ 0.5 & 37.0\
NGC 625 \#20 & 01:35:04.42 & -41:26:00.9 & 2.6 $\pm$ 0.3 & 36.7\
NGC 625 \#21 & 01:35:03.99 & -41:26:07.3 & 7.6 $\pm$ 0.8 & 37.2\
NGC 625 \#22 & 01:35:01.96 & -41:26:25.8 & 23.0 $\pm$ 2.3 & 37.6\
NGC 625 \#23 & 01:35:01.47 & -41:26:25.9 & 19.3 $\pm$ 1.9 & 37.6\
[lcccccc]{} ESO 347-G17 & $6.3\times 10^{-3}$ & $5.3\times 10^{-11}$ & 19 & 120 & 1.00 & 25\
ESO 471-G06 & $2.0\times 10^{-3}$ & $3.3\times 10^{-11}$ & 30 & 163 & 2.66 & 110\
ESO 348-G09 & $2.8\times 10^{-4}$ & $0.6\times 10^{-11}$ & 167 & 84.3 & 1.83 & 400\
SC 18 & $2.4\times 10^{-5}$ & $3.4\times 10^{-11}$ & 29 & 5.0 & 7.13 & 270\
NGC 59 & $3.1\times 10^{-3}$ & $1.6\times 10^{-11}$ & 63 & 16.7 & 0.87 & 7.1\
ESO 473-G24 & $1.3\times 10^{-3}$ & $7.6\times 10^{-11}$ & 13 & 63.8 & 3.71 & 65\
SC 24 & $2.1\times 10^{-6}$ & $0.6\times 10^{-11}$ & 167 & 2.1 & 6.29 & 1300\
DDO 226 & $1.3\times 10^{-3}$ & $3.6\times 10^{-11}$ & 28 & 33.9 & 0.93 & 36\
DDO 6 & $2.9\times 10^{-4}$ & $2.5\times 10^{-11}$ & 40 & 15.4 & 1.34 & 70\
NGC 625 & $5.0\times 10^{-2}$ & $10.2\times 10^{-11}$ & 9.8 & 118 & 0.24 & 3.1\
ESO 245-G05 & $1.0\times 10^{-2}$ & $3.9\times 10^{-11}$ & 26 & 289 & 1.15 & 38\
to7.2in
------------------------------------------------------------------------
A complete gzipped .ps file containing Figure 1 can be obtained via anonymous ftp from ftp.astro.umn.edu /pub/users/skillman/sculptor in paper1.ps.gz
to7.2in
------------------------------------------------------------------------
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
This is a set of lecture notes for the 2019 EMS School in Applied Mathematics held in Kácov, Czech Republic.
Conserved or dissipated quantities, like energy or entropy, are at the heart of the study of many classes of time-dependent PDEs in connection with fluid mechanics. This is the case, for instance, for the Euler and Navier-Stokes equations, for systems of conservation laws, and for transport equations. In all these cases, a formally conserved quantity may no longer be constant in time for a weak solution at low regularity. The delicate interplay between regularity and conservation of the respective quantity relates to renormalisation in the DiPerna-Lions theory of transport and continuity equations, and to Onsager’s conjecture in the realm of ideal incompressible fluids. We will review the classical commutator methods of DiPerna-Lions and Constantin-E-Titi, and then proceed to more recent results.
address: '*Emil Wiedemann:* Institute of Applied Analysis, Universität Ulm, Helmholtzstr. 18, 89081 Ulm, Germany'
author:
- Emil Wiedemann
title: Conserved quantities and regularity in fluid dynamics
---
Introduction
============
Quite unsurprisingly, conserved or dissipated quantities play a fundamental role in about any evolution differential equation related to continuum mechanics. On the analytical level, these quantities are often (as in the Navier-Stokes equations) the only source of a priori estimates to yield sufficient compactness for the existence of weak solutions. In the theory of transport equations, conserved quantities allow to show uniqueness and stability through the renormalisation theory of DiPerna-Lions [@dipernalions] discussed below.
On the physical level, quantities that are formally shown to be conserved can, in fact, be observed to be dissipated. This kind of anomalous dissipation occurs, e.g., in the incompressible Euler equations at low regularity due to turbulent energy transfer to high scales, as predicted by Onsager [@ON] based on Kolmogorov’s phenomenological theory of turbulence; hence, at low regularity, one may observe that the kinetic energy $$\label{globaleuler}
\frac12\int |u(x,t)|^2dx$$ is actually decreasing in time, although by formal computation (as it is justified for smooth solutions) it would be constant. In fact, energy conservation formally holds in the stronger local sense $$\label{localeuler}
\partial_t \frac{|u|^2}{2}+\operatorname{div}\left(\left(\frac{|u|^2}{2}+p\right)u\right)=0,$$ which implies the total conservation of energy upon integration in time and using the Divergence Theorem.
The past decades have seen increasingly sophisticated constructions giving rigorous examples of weak solutions to the Euler equations with non-conserved energy [@Scheffer; @shnirel1; @shnirel2; @euler1; @euler2; @eulerinvent; @isett16; @buckmasteretal17], culminating in a complete proof of Onsager’s Conjecture on the threshold regularity up to which anomalous dissipation can occur. This regularity is essentially at $C^{1/3}$ (the Hölder space with exponent $1/3$), and below we will at least show that $1/3$ is an upper bound.
A more classical phenomenon in hyperbolic conservation laws is the formation of shock waves, which also leads to a decrease of the (mathematical) entropy. For instance, the inviscid Burgers equation $$\label{localburgers}
\partial_t u+\partial_x(u^2)=0$$ is easily seen to satisfy the equality $\partial_t\frac{u^2}{2}+\partial_x\frac{2u^3}{3}=0$, and thus to conserve the so-called entropy $\frac12\int u^2 dx$, as long as the solution remains smooth, but it is equally easy to explicitly find an example of a weak solution that becomes discontinuous in finite time and then has strictly decreasing entropy. More precisely, given the Lipschitz initial datum $$u_0(x):=\begin{cases}
1 & \text{if $x\leq0$,}\\
1-x & \text{if $0\leq x\leq 1$,}\\
0 & \text{if $x>1$},
\end{cases}$$
for any time $t<1$ a solution is given by $$u(x,t)=\begin{cases}
1 & \text{if $x\leq t$,}\\
\frac{1-x}{1-t} & \text{if $t\leq x\leq 1$,}\\
0 & \text{if $x>1$},
\end{cases}$$ which can be extended to $t\geq 1$ by
$$u(x,t)=\begin{cases}
1 & \text{if $x<\frac{t+1}{2}$,}\\
0 & \text{if $x>\frac{t+1}{2}$},
\end{cases}$$
and it holds true that, after $t=1$, there are smooth test functions ${\varphi}\geq0$ such that $$\int\int(\partial_t{\varphi}\frac{u^2}{2}+\partial_x{\varphi}\frac{2u^3}{3})dxdt>0,$$ which means the entropy conservation is violated (in accordance with the Second Law of Thermodynamics).
But even when a certain quantity is seen to be dissipative already on the formal level, it is still important to know whether it satisfies a predicted balance as an equality or not. For example, the incompressible Navier-Stokes equations formally satisfy $$\label{NSEbalance}
\frac 12\partial_t\int |u|^2 dx + \nu\int_0^t\int |\nabla u|^2 dxds=\frac 12\int |u_0|^2dx,$$ where $\nu>0$ is the viscosity of the modelled fluid; but it can only be shown to hold as an inequality ($=$ replaced by $\leq$) for generic weak solutions of Leray-Hopf type. In contrast to the Euler equations and hyperbolic conservation laws, however, there seems to be no mathematical or physical reason to believe that should hold with strict inequality.
Let us condense this somewhat loose collection of observations into some general ideas that will form the focus of this survey:
- A great variety of partial differential equations related to continuum dynamics exhibit quantities that can be easily seen, by formal calculation invoking the chain rule of differential calculus, to be conserved in time. For smooth solutions, these calculations are easily justified rigorously.
- These conservation laws can be formulated in a local (e.g., , ) or in a global (e.g., , ) way.
- On the other hand, less regular solutions may not conserve these quantities. This relates to physically observable effects of anomalous energy dissipation due to turbulence, or to increase of physical entropy due to the Second Law of Thermodynamics.
- It is therefore worth investigating the threshold regularity below which such dissipative effects can occur.
Our discussion here mainly focuses on inviscid models. We begin with the arguably simplest case of linear scalar conservation laws, i.e., linear transport equations, which possess infinitely many conserved quantities. The question of threshold regularity can be interpreted in several ways, depending on whether one wishes to impose regularity conditions on the coefficients alone, or on the coefficients and the solution combined. The first approach leads to the DiPerna-Lions theory of renormalisation, while the second one motivates the commutator estimates of Constantin-E-Titi [@CET]. This is presented in Section \[renormsec\].
The mentioned techniques give bounds from above for the sought threshold regularity (i.e., sufficient conditions for conservation), and can be viewed as restoring the chain rule in regimes of low regularity. On the other side, the construction of examples of non-conservation, and therefore of breakdown of the chain rule, can be much harder. We have seen that classical shocks provide such examples in the context of hyperbolic equations, but in incompressible models, shocks are not available. Instead, convex integration has recently become the method of choice to construct dissipative solutions of the Euler equations. In Section \[chainrule\], we outline the method of convex integration in the comparatively simple setting of steady transport, and thus provide counterexamples to renormalisation.
The conservative part of Onsager’s Conjecture for the incompressible Euler system forms the topic of Section \[onsagersec\], including a discussion of recent results concerning bounded domains and the vanishing viscosity limit. The subsequent sections are devoted to various recent extensions of the commutator method to statistical solutions, general conservation laws, and degenerate cases such as the compressible Euler and Navier-Stokes equations with possible vacuum.
Renormalisation of Transport Equations {#renormsec}
======================================
The first partial differential (PDE) considered in the textbook [@evans] of L. C. Evans is also the supposedly simplest one: the *transport equation* $$\label{transporteq}
\partial_t \rho+u\cdot\nabla \rho=0.$$ For further simplicity, let us consider the case of periodic boundary conditions. Then, if ${\mathbb{T}}^d:={\mathbb{R}}^d/{\mathbb{Z}}^d$ is the flat torus, one usually considers the vector field $u:{\mathbb{T}}^d\times [0,T]\to{\mathbb{R}}^d$ as given and divergence-free. The scalar field $\rho:{\mathbb{T}}^d\times[0,T]\to{\mathbb{R}}$ is the unknown, which could be subject to an initial condition $\rho(\cdot,0)=\rho^0$.
The transport equation can be interpreted, for instance, as follows: The given field $u$ can be thought of as the known velocity field of an incompressible flow (hence the divergence-free condition), such as water on the surface of the ocean (in which case $d=2$). The scalar $\rho$ then gives the concentration of, say, a chemical dissolved on the ocean surface, and the chemical is transported by the given flow. As, in this very simple model, the chemical has no effect on the dynamics of the transporting flow, $\rho$ is sometimes called a *passive scalar* or a *passive tracer*.
A remarkable property of the transport equation is that it can be *renormalised*: Suppose an arbitrary $C^1$ function $\eta:{\mathbb{R}}\to{\mathbb{R}}$ is given, then can be multiplied by $\eta'(\rho)$ to yield, by means of the chain rule, $$\label{renorm}
\partial_t\eta(\rho)+u\cdot\nabla \eta(\rho)=0,$$ so that in fact arbitrary functions of a solution become a solution of the same equation. Moreover, integration in space yields $\frac{d}{dt}\int_{{\mathbb{T}}^d}\eta(\rho)dx=0$, and therefore from this renormalisation procedure we obtain infinitely many conserved quantities for .
However, the chain rule (we used it in the form $\partial_t \eta(\rho)=\eta'(\rho)\partial_t \rho$ and $\nabla\eta(\rho)=\eta'(\rho)\nabla \rho$) is only justified if $\rho$ is $C^1$, or at least Lipschitz. So the question arises:
*Under what conditions on $u$ and/or $\rho$ are solutions of the transport equation renormalised?*
First of all, how does the transport even make sense if $\rho$ is not $C^1$? As usual, one considers a distributional concept of solution, using the fact that $u\cdot\nabla \rho=\operatorname{div}(\rho u)$ by virtue of the divergence-free condition on $u$. Therefore, a function $\rho\in L^{\infty}({\mathbb{T}}^d\times [0,T])$ is called a *weak solution* of with initial data $\rho^0\in L^\infty({\mathbb{T}}^d)$ if, for every ${\varphi}\in C_c^1({\mathbb{T}}^d\times[0,T))$, we have $$\int_0^T\int_{{\mathbb{T}}^d} \partial_t{\varphi}\rho+\rho\nabla{\varphi}\cdot u dxdt=\int_{{\mathbb{T}}^d}{\varphi}(x,0)\rho^0(x)dx.$$ Note this definition makes sense as long as $u\in L^1_{loc}({\mathbb{T}}^d\times[0,T))$.
Renormalisation plays an important role not only for the study of or of ordinary differential equations, but also for larger systems of PDEs that contain or the closely related *continuity equation* $$\partial_t \rho+\operatorname{div}(\rho u)=0,$$ where $u$ is no longer assumed divergence-free. Consider two examples: The *isentropic compressible Navier-Stokes system* reads $$\label{compNSE}
\begin{aligned}
\partial_t(\rho u)+\operatorname{div}(\rho u\otimes u)+\nabla p(\rho)&=\operatorname{div}\mathbb{S}(\nabla u),\\
\partial_t \rho+\operatorname{div}(\rho u)&=0,
\end{aligned}$$ where the scalar density $\rho$ is now assumed non-negative, $p$ is a given function of density, and ${\mathbb{S}}$ denotes the Newtonian stress tensor. The velocity $u$ is a vector field ${\mathbb{T}}^d\times[0,T]\to{\mathbb{R}}^d$.
Let us ignore the details of the constitutive theory for the moment and imagine ${\mathbb{S}}$ as the identity, so that the right hand side can be thought of simply as $\Delta u$. It is well-known that these equations satisfy a priori bounds given by the energy inequality $$\int_{{\mathbb{T}}^d}\frac12\rho(x,t)|u(x,t)|^2+P(\rho(x,t))dx + \int_0^t\int_{{\mathbb{T}}^d}{\mathbb{S}}(\nabla u):\nabla u dx ds \leq \int_{{\mathbb{T}}^d}\frac12\rho^0|u^0|^2+P(\rho^0)dx.$$
Here $P$ is the so-called pressure potential given by $$\label{presspot}
P(r)=r\int_1^r \frac{p(s)}{s^2}ds.$$ Since ${\mathbb{S}}(\nabla u):\nabla u\geq c|\nabla u|^2$, we obtain a bound for $u$ in $L^2(0,T;H^1({\mathbb{T}}^d))$, whereas for $\rho$ we obtain only integrability, but no regularity properties. In fact, this is to be expected since is parabolic in the momentum equation but only hyperbolic in the mass equation.
The question whether the density can be renormalised plays an important role, e.g., in the Lions-Feireisl theory of weak solutions for , see [@lions; @feireisl]. As we just saw, we obtain a priori information on the regularity of the transporting velocity, but not on the transported scalar. In this context, one thus asks: Under what (Sobolev) regularity assumptions on the transporting velocity field $u$ is every bounded weak solution $\rho$ of renormalised (i.e. every smooth function of $\rho$ is again a weak solution of )? This is the subject of the famous DiPerna-Lions theory that we will outline shortly.
As another example, consider an *active scalar equation* of the form $$\begin{aligned}
\partial_t \rho+u\cdot \nabla\rho &=0,\\
\operatorname{div}u&=0,\\
u=T[\rho],
\end{aligned}$$ where $T$ is a Fourier multiplier operator of order zero. For instance, if (for $d=2$) the symbol of $T$ is given by $i\frac{\xi^\perp}{|\xi|}$, we obtain the well-known surface quasi-geostrophic (SQG) equation.
The only thing of interest at the moment, anyway, is the fact that $T$ is bounded from $L^p$ to $L^p$ ($1<p<\infty$) and therefore, the active scalar $\rho$ and the velocity $u$ have the same (Lebesgue, Besov, Hölder etc.) regularity. In contrast to the Navier-Stokes equations, therefore, we can not (or do not have to) distinguish between the regularities of the scalar and the vector field.
A similar example is given by the *incompressible Euler equations* $$\label{inceuler}
\begin{aligned}
\partial_t u+(u\cdot \nabla)u+\nabla p&=0\\
\operatorname{div}u&=0,
\end{aligned}$$ where in some sense the velocity is transported by itself (this is the effect of *advection*). Again, there is no way to distinguish between the regularities of the transporting and the transported quantity. We will get back to the Euler equations later.
DiPerna-Lions commutators
-------------------------
Again let us simply consider the transport equation and assume that $\rho, u$ form a weak solution (where $u$ is still divergence-free and $\rho$ is bounded for simplicity). In cases where some regularity is known for $u$ (loosely speaking, at least one full distributional space derivative), but none for $\rho$, the theory of DiPerna-Lions is useful.
Let $\chi:{\mathbb{T}}^d\to{\mathbb{R}}$ be a standard mollifier, i.e. a smooth non-negative radially symmetric function with compact support in $B_1(0)$ and $\int_{B_1(0)}\chi dx=1$. Set $\chi_{\varepsilon}(x)={\varepsilon}^{-d}\chi\left(\frac x{\varepsilon}\right)$, which is supported on $B_{\varepsilon}(0)$ and still has unit integral. For a function $f$, We write $f_{\varepsilon}:=f*\chi_{\varepsilon}$.
Mollifying , we obtain (ignoring issues of time differentiability) $$\label{molltransport}
0=\partial_t\rho_{\varepsilon}+\operatorname{div}(\rho u)_{\varepsilon}=\partial_t\rho_{\varepsilon}+\operatorname{div}(\rho_{\varepsilon}u)+R_{\varepsilon},$$ where $$R_{\varepsilon}=\operatorname{div}(\rho u)_{\varepsilon}-\operatorname{div}(\rho_{\varepsilon}u)$$ is the *commutator*. Let now $\eta\in C^1$ and multiply by $\eta'(\rho_{\varepsilon})$. Since $\rho_{\varepsilon}$ is smooth, we can apply the chain rule to obtain $$\partial_t\eta(\rho_{\varepsilon})+ \operatorname{div}(\eta(\rho_{\varepsilon}) u)=-\eta'(\rho_{\varepsilon})R_{\varepsilon}$$ (recall that $u$ is divergence-free, so that $\operatorname{div}(\rho_{\varepsilon}u)=u\cdot\nabla \rho_{\varepsilon}$). First, it is clear that the terms on the left hand side converge, in the sense of distributions as ${\varepsilon}\to0$, to $\partial_t\eta(\rho)+ \operatorname{div}(\eta(\rho) u)$. Hence, as $\eta'(\rho_{\varepsilon})$ is bounded uniformly in ${\varepsilon}$, it suffices to show that $R_{\varepsilon}\to 0$ in $L^1({\mathbb{T}}^d\times[0,T])$.
To this end, we compute $$\begin{aligned}
R_{\varepsilon}(x,t)&=\operatorname{div}(\rho u)*\chi_{\varepsilon}(x,t)-\operatorname{div}((\rho*\chi_{\varepsilon}) u(x,t))\\
&=-\rho u*\nabla\chi_{\varepsilon}(x,t)+\rho*\nabla\chi_{\varepsilon}\cdot u(x,t)\\
&=\int_{B_{\varepsilon}(x)} \rho(y,t)(u(x,t)-u(y,t))\cdot \nabla\chi_{\varepsilon}(x-y)dy\\
&={\varepsilon}^{-d-1}\int_{B_{\varepsilon}(x)} \rho(y,t)(u(x,t)-u(y,t))\cdot \nabla\chi\left(\frac{x-y}{{\varepsilon}}\right)dy\\
&=-\int_{B_1(0)} \rho(x+{\varepsilon}z,t)\frac{u(x+{\varepsilon}z,t)-u(x,t)}{{\varepsilon}}\cdot \nabla\chi\left(z\right)dz,
\end{aligned}$$ where in the end we used the transformation $z=\frac{y-x}{{\varepsilon}}$.
Suppose that $u\in L^1(0,T;W^{1,1}({\mathbb{T}}^d))$, then, by standard difference quotient lemmas, $\frac{u(x+{\varepsilon}z,t)-u(x,t)}{{\varepsilon}}$ converges, as ${\varepsilon}\to0$, to the directional derivative $\partial_z u(x)$ in $L^1({\mathbb{T}}^d\times (0,T))$ for fixed $z$; moreover it is bounded in $L^1$ uniformly in $z$ and ${\varepsilon}$. Since $\rho(x+{\varepsilon}z)$ is in $L^\infty$, uniformly in ${\varepsilon}$ and $z$, and converges in $L^1$ to $\rho(x)$ for fixed $z$, we obtain the strong $L^1$ convergence $$\begin{aligned}
R_{\varepsilon}(x,t)&\to-\rho(x,t)\int_{B_1(0)} \partial_zu(x,t)\cdot \nabla\chi\left(z\right)dz\\
&=-\rho(x,t)\partial_j u_i(x,t)\int_{B_1(0)} z_j\partial_i\chi\left(z\right)dz\\
&=\rho(x,t)\delta_{ij}\partial_ju_i(x,t)=\rho(x,t)\operatorname{div}u(x,t)=0,
\end{aligned}$$ as desired.
As the convergence argument may not be obvious, let us give it in more detail: Write $D_{\varepsilon}(x,z,t):=\frac{u(x+{\varepsilon}z,t)-u(x,t)}{{\varepsilon}}$ and $\rho_{\varepsilon}(x,z,t):=\rho(x+{\varepsilon}z,t)$, then $$\begin{aligned}
&\int_{{\mathbb{T}}^d\times(0,T)}\left|R_{\varepsilon}(x,t)+\int_{B_1(0)}\partial_zu(x,t)\cdot \nabla\chi\left(z\right)dz\right|dxdt\\
&\leq \|\nabla\chi\|_\infty\left(\int_{{\mathbb{T}}^d\times(0,T)}\int_{B_1(0)}|\rho_{\varepsilon}(x,z,t)D_{\varepsilon}(x,z,t)-\rho_{\varepsilon}(x,z,t)\partial_zu(x,t)|dzdxdt\right.\\
&\hspace{2cm}+\left.\int_{{\mathbb{T}}^d\times(0,T)}\int_{B_1(0)}|\rho_{\varepsilon}(x,z,t)\partial_z u(x,t)-\rho(x,t)\partial_z u(x,t)|dzdxdt\right)\\
&\leq \|\nabla\chi\|_\infty\sup_{{\varepsilon}, z}\|\rho_{\varepsilon}\|_\infty\int_{B_1(0)}\int_{{\mathbb{T}}^d\times(0,T)}|D_{\varepsilon}-\partial_z u|dxdz\\
&\hspace{2cm}+ \|\nabla\chi\|_\infty\sup_{z}\|\partial_z u\|_{L^1}\int_{B_1(0)}\int_{{\mathbb{T}}^d\times(0,T)}|\rho_{\varepsilon}-\rho|dx dz.
\end{aligned}$$ As mentioned above, for each $z$, $\int_{{\mathbb{T}}^d\times(0,T)}|D_{\varepsilon}-\partial_z u|dx$ converges to zero as ${\varepsilon}\to0$, and the same is true for $\int_{{\mathbb{T}}^d\times(0,T)}|\rho_{\varepsilon}-\rho|dx$. Since these expressions are dominated, respectively, by $$2\sup_z\int_{{\mathbb{T}}^d\times(0,T)}|\partial_z u(x)|dx,\quad 2\int_{{\mathbb{T}}^d\times(0,T)}|\rho(x)|dx,$$ which are constant and hence integrable in $z$, we conclude by the Dominated Convergence Theorem.
We have thus proved:
If $u\in L^1(0,T;W^{1,1}({\mathbb{T}}^d))$, then every bounded weak solution of is renormalised in the sense of .
Of course this result holds in much greater generality (the assumptions $\rho\in L^\infty$ and $\operatorname{div}u=0$ can be substantially relaxed). In fact it follows easily from renormalisation that the Cauchy problem for admits a unique weak solution.
Constantin-E-Titi commutators
-----------------------------
Let us present a commutator argument that looks very similar to the previous one, but leads to very different conclusions. Once again we mollify equation in space, $$0=\partial_t\rho_{\varepsilon}+\operatorname{div}(\rho u)_{\varepsilon}=\partial_t\rho_{\varepsilon}+\operatorname{div}(\rho_{\varepsilon}u_{\varepsilon})+S_{\varepsilon},$$ with $$S_{\varepsilon}=\operatorname{div}(\rho u)_{\varepsilon}-\operatorname{div}(\rho_{\varepsilon}u_{\varepsilon}).$$ Note the only difference compared to is that we chose to mollify also $u$. Multiplying again by $\eta'(\rho_{\varepsilon})$, we obtain (noting that $u_{\varepsilon}$ is still divergence-free) $$\partial_t\eta(\rho_{\varepsilon})+\operatorname{div}(\eta(\rho_{\varepsilon})u_{\varepsilon})=-\eta'(\rho_{\varepsilon})S_{\varepsilon},$$ so we obtain renormalisation provided we can show the the right hand side converges to zero, in the sense of distributions, as ${\varepsilon}\to0$.
To this end, let ${\varphi}\in C_c^1({\mathbb{T}}^d\times(0,T))$, so that integration by parts yields $$\begin{aligned}
-\int_0^T\int_{{\mathbb{T}}^d}{\varphi}\eta'(\rho_{\varepsilon})S_{\varepsilon}dxdt &= \int_0^T\int_{{\mathbb{T}}^d}\eta'(\rho_{\varepsilon})\nabla{\varphi}\cdot ((\rho u)_{\varepsilon}-\rho_{\varepsilon}u_{\varepsilon}) dxdt\\
& + \int_0^T\int_{{\mathbb{T}}^d}{\varphi}\eta''(\rho_{\varepsilon})\nabla\rho_{\varepsilon}\cdot ((\rho u)_{\varepsilon}-\rho_{\varepsilon}u_{\varepsilon}) dxdt.
\end{aligned}$$ We only treat the second integral, as the first one is easier. Assume to this end that $\eta''$ is bounded, and consider the pointwise identity $$(\rho u)_{\varepsilon}-\rho_{\varepsilon}u_{\varepsilon}= -(\rho_{\varepsilon}-\rho)(u_{\varepsilon}-u)+\int_{B_{\varepsilon}(0)} \chi_{\varepsilon}(y)(\rho(\cdot-y)-\rho)(u(\cdot-y)-u)dy.$$ Thus, one part of the desired estimate is obtained by $$\int_0^T\int_{{\mathbb{T}}^d}|\nabla\rho_{\varepsilon}| |\rho_{\varepsilon}-\rho||u_{\varepsilon}-u| dxdt\leq \|\nabla\rho_{\varepsilon}\|_{L^p}\|\rho-\rho_{\varepsilon}\|_{L^p}\|u-u_{\varepsilon}\|_{L^q}$$ for some exponents satisfying $\frac2p+\frac1q\leq1$. The other part can be estimated similarly, so we ignore it.
The question arises under what conditions these norms converge to zero as ${\varepsilon}\to0$. To this end, suppose that $$\label{BesovVMO}
\lim_{{\varepsilon}\to0} \int_0^T\int_{{\mathbb{T}}^d}\int_{B_{\varepsilon}(0)}\frac{|\rho(x)-\rho(x-y)|^p}{{\varepsilon}^{d+\alpha p}}dydxdt=0.$$ We aim to show that, under this assumption, $\lim_{{\varepsilon}\to0}{\varepsilon}^{-\alpha}\|\rho-\rho_{\varepsilon}\|_{L^p}=0$. Indeed, using Jensen’s inequality and the definition of $\chi_{\varepsilon}$, $$\begin{aligned}
{\varepsilon}^{-\alpha p}\|\rho-\rho_{\varepsilon}\|_{L^p}^p&=\int_0^T\int_{{\mathbb{T}}^d}\left|\int_{B_{\varepsilon}(0)}\frac{\rho(x)-\rho(x-y)}{{\varepsilon}^\alpha}\chi_{\varepsilon}(y)dy\right|^pdxdt\\
&\leq \int_0^T\int_{{\mathbb{T}}^d}\int_{B_{\varepsilon}(0)}\frac{|\rho(x)-\rho(x-y)|^p}{{\varepsilon}^{d+\alpha p}}\chi\left(\frac y{\varepsilon}\right)dydxdt,
\end{aligned}$$ which converges to zero by virtue of assumption . Likewise, we have $\lim_{{\varepsilon}\to0}{\varepsilon}^{-\beta}\|u-u_{\varepsilon}\|_{L^q}=0$ provided $$\label{BesovVMO2}
\lim_{{\varepsilon}\to0} \int_0^T\int_{{\mathbb{T}}^d}\int_{B_{\varepsilon}(0)}\frac{|u(x)-u(x-y)|^q}{{\varepsilon}^{d+\beta q}}dydxdt=0.$$ Finally, it is not difficult to get the estimate $\lim_{{\varepsilon}\to0}{\varepsilon}^{1-\alpha}\|\nabla\rho_{\varepsilon}\|_{L^p}=0$. Assume now that $2\alpha+\beta\geq 1$, then $$\begin{aligned}
\|\nabla\rho_{\varepsilon}\|_{L^p}&\|\rho-\rho_{\varepsilon}\|_{L^p}\|u-u_{\varepsilon}\|_{L^q}\leq\\
&{\varepsilon}^{1-\alpha}\|\nabla\rho_{\varepsilon}\|_{L^p}{\varepsilon}^{-\alpha}\|\rho-\rho_{\varepsilon}\|_{L^p}{\varepsilon}^{-\beta}\|u-u_{\varepsilon}\|_{L^q}\to0,
\end{aligned}$$ and so we arrive at the following result:
\[transportthm\] Let $\frac 2p+\frac 1q\leq 1$ and $2\alpha+\beta\geq 1$, and $\rho\in L^p({\mathbb{T}}^d\times(0,T))$ and $u\in L^q({\mathbb{T}}^d\times(0,T))$ be a weak solution of satisfying and . If $\eta\in W^{2,\infty}({\mathbb{R}})$, then the renormalised equation holds in the sense of distributions: $$\partial_t\eta(\rho)+u\cdot \nabla \eta(\rho)=0.$$
Several remarks are in order. This commutator method relies on the ideas in [@CET] in the context of Onsager’s conjecture (which we will get back to later). In the case of Euler, the transporting field and the transported quantity are identical, which forces $p=q\geq 3$ and $\alpha=\beta\geq \frac13$. Mixed regularities for $\rho$ and $u$ were introduced in [@LSh; @FGSW]. Theorem \[transportthm\] was proved in [@AkWi], but in a slightly more restrictive functional framework. The paper [@AkWi] also relaxes the condition $\eta\in W^{2,\infty}({\mathbb{R}})$, and gives an application to active scalar equations. Conditions and were first formulated in [@FjWi] and thus generalised previous works like [@CET; @duchonrobert; @CCFS08]. They are implied, for instance, by $\rho\in L^p((0,T);C^{\alpha'}({\mathbb{T}}^d))$ or $\rho\in L^p((0,T);B_{p,\infty}^{\alpha'}({\mathbb{T}}^d))$ for any $\alpha'>\alpha$, where $C^{\alpha'}$ and $B_{p,\infty}^{\alpha'}$ denote the respective Hölder and Besov spaces, and similarly for $u$.
Counterexamples to Renormalisation via Convex Integration {#chainrule}
=========================================================
It may seem strange that renormalisation could fail for a simple linear equation like . But in fact, there is a great deal of counterexamples known, starting from the original paper of DiPerna-Lions and extended in various ways in [@depauw; @shvydkoy; @AlBiCr; @IsVi; @CGSW2; @CGSW1; @MSz; @MSz2; @MSa]. A powerful instrument to produce rough and often pathological solutions to certain PDEs is known as *convex integration*. We present here the main ideas from [@CGSW1] (full details can be found in that paper). The result discussed here is by no means optimal in terms of regularity of the velocity field, but it does show that renormalisation can fail in basically any conceivable way, and it gives an idea of the general technique of convex integration in a comparatively simple setting.
We consider the stationary problem in 3D, and we aim to show the following:
Let $f\in\mathcal{D}'({\mathbb{T}}^3;{\mathbb{R}})$ be a distribution such that there exists a bounded continuous solution of the equation $\operatorname{div}w=f$. Then, there exist a bounded vectorfield $u\in L^\infty({\mathbb{T}}^3;{\mathbb{R}}^3)$ and a bounded scalar field $\rho\in L^\infty({\mathbb{T}}^3)$ such that $$\label{original}
\begin{aligned}
\operatorname{div}(\rho u)&=0\\
\operatorname{div}(u)&=0\\
\operatorname{div}(\rho^2 u)&=f.
\end{aligned}$$
Note once more that if the chain rule were valid, we would have $\operatorname{div}(\rho^2 u)=0$. Therefore, $f$ is called the *renormalisation defect*.
We begin by vastly relaxing the problem to the study of triplets $(m,u,w)$ of vectorfields that satisfy $$\label{linear}
\begin{aligned}
\operatorname{div}(m)&=0\\
\operatorname{div}(u)&=0\\
\operatorname{div}(w)&=f.
\end{aligned}$$ Setting $$\label{nonlinear}
\begin{aligned}
K_{C}&:=\left\{(m,u,w)\in{\mathbb{R}}^{3\times3}: \frac{1}{C}\leq|u|\leq C \right.\\
&\left.\text{ and there is $\frac{1}{C}\leq\rho\leq C$ such that } m=\rho u, w=\rho^2u\right\},
\end{aligned}$$ we realise that a solution of will be a solution of the original problem if $(m,u,w)(x)\in K_C$ for almost every $x\in{\mathbb{T}}^3$, for some $C>1$.
Our strategy is now roughly as follows:
1. Start with the “subsolution” $U^0:=(0,0,w)$, where $w$ has the property $\operatorname{div}w=f$.
2. For almost every $x\in{\mathbb{T}}^3$, represent $(0,0,w)$ as a convex combination of the form $\sum_{i=1}^N\lambda_i(m_i,u_i,w_i)$ in a way that is compatible with , and such that the $(m_i,u_i,w_i)$ are in $K_C$.
3. Find a new triplet of functions $(m^1,u^1,w^1)$ that still satisfies , and that approximately takes the value $(m_i,u_i,w_i)$ on a volume fraction $\lambda_i$ near the point $x$.
4. Iterate the process to obtain a sequence $(m^n,u^n,w^n)_{n\in{\mathbb{N}}}$ of functions satisfying , and show that the sequence converges to a solution of .
Step 1 can be done by assumption. For Step 2, it surely should be clarified what is meant by “compatibility” of the convex combination with constraint . Clearly, without a differential constraint like , it is easy to find a continuous map $U^1:=(m^1,u^1,w^1):{\mathbb{T}}^3\to {\mathbb{R}}^{3\times 3}$ such that, close to any point $x\in{\mathbb{T}}^3$, $U^1$ takes values close to $U_i:=(m_i,u_i,w_i)$ on a set of volume fraction $\lambda_i$; suppose for simplicity that the $\lambda_i$ and $U_i$ are constant, then such a map could simply be given as (a suitable mollification of) the piecewise constant map $$U(x)=U_n \quad\text{if $x_1\in\left(\sum_{i=1}^{n-1}\lambda_i, \sum_{i=1}^{n}\lambda_i\right)$}.$$ Now of course $\lambda_i$ and $U_i$ do depend on $x$, but as this is the case continuously, we can cover the domain by small cubes, consider $\lambda_i$ and $U_i$ constant on each such cube, and glue the various pieces together.
However, we want $U-(0,0,w)$ to be divergence-free, so that remains valid. Suppose, at a given point $x\in{\mathbb{T}}^3$, $U^0(x)$ can be represented as a convex combination of only two matrices, $U^0(x)=\lambda U_1+(1-\lambda)U_2$. Does there exist a divergence-free matrix field $U$ that oscillates between $U_1$ and $U_2$?
To this end, we define the *wave cone* for the divergence-free condition as $$\label{lambdaconvex}
\Lambda:=\{\bar{U}\in{\mathbb{R}}^{3\times 3}: \text{there exists $\xi\neq 0$ such that $\operatorname{div}[h(x\cdot \xi)\bar{U}]=0$ for any $h:{\mathbb{R}}\to{\mathbb{R}}$}\}.$$ Since $\bar{U}\in\Lambda$ is equivalent to $h'(x\cdot\xi)\bar{U}\xi=0$ for all $h$ and a non-trivial $\xi$, we get that $\bar{U}\in\Lambda$ if and only if $\det\bar{U}=0$ or, in other words, the rank of $\bar{U}$ is at most 2. Therefore, there exists an oscillation between $U_1$ and $U_2$ if and only if $\operatorname{rank}(U_2-U_1)\leq 2$.
Inductively, one can then find a condition for divergence-free oscillations between more than two matrices:
\[rk2laminate\] Suppose $\lambda_i>0$ for $i=1,\ldots,n$, $\sum_{i=1}^{n}\lambda_i=1$, and $U_i\in{\mathbb{R}}^{3\times3}$ for $i=1,\ldots,n$. The family of pairs $(\lambda_i,U_i)_{i=1}^n$ satisfies the (inductively defined) [**$H_n$-condition**]{} if
- $\operatorname{rank}(U_2-U_1)\leq2$ in the case $n=2$;
- after a relabeling of indices, if necessary, we have $\operatorname{rank}(U_2-U_1)\leq2$ and the family $(\tau_i,V_i)_{i=1}^{n-1}$ satisfies the $H_{n-1}$-condition, where $$\begin{aligned}
\tau_1=\lambda_1+\lambda_2, \hspace{0.2cm}\tau_i=\lambda_{i+1}\hspace{0.2cm}\text{for $i=2,\ldots,n-1$}
\end{aligned}$$ and $$\begin{aligned}
V_1=\frac{\lambda_1}{\tau_1}U_1+\frac{\lambda_2}{\tau_1}U_2, \hspace{0.2cm}V_i=U_{i+1}\hspace{0.2cm}\text{for $i=2,\ldots,n-1$}
\end{aligned}$$ in the case $n>2$.
Moreover we adopt the convention that every pair of the form $(1,U)$ satisfies the $H_1$-condition.
With this definition, we can thus give a precise meaning to Step (2) above: For almost every $x\in{\mathbb{T}}^3$, represent $U^0:=(0,0,w(x))$ as a convex combination of the form $\sum_{i=1}^N\lambda_i(x)U_i(x)$, where $(\lambda_i(x),U_i(x))_{i=1}^N$ satisfy the $H_N$-condition, and $U_i(x)\in K_C$.
That this can actually be done is the content of the following lemma. Note that, since $w$ is assumed bounded and continuous, we may always assume $|w|\geq 1$, as adding a constant will not affect the property $\operatorname{div}w=f$.
\[geom1\] Let $U=(0,0,w)\in{\mathbb{R}}^{3\times3}$ such that $|w|\geq1$. Then there exist $(\lambda_i,{U_i})_{i=1}^3$ satisfying the $H_3$-condition, such that $$U=\sum_{i=1}^3\lambda_iU_i,$$ and a number $C>1$ such that $$U_i\in K_{C} \quad\text{for $i=1,2,3$.}$$
We split $(0,0,w)$ into $$(0,0,w)=\frac{1}{2}\left(-w,-w,w\right)+\frac{1}{2}\left(w,w,w\right).$$ If we call the matrices corresponding to the two triplets on the right hand side $U_-$ and $U_+$, respectively, we first observe that $U_-$ and $U_+$ are rank-2 connected since $(U_--U_+)e_3=0$. Secondly, $U_+\in K_{C}$ for any $C$ such that $$C\geq|w|.$$
Next, let us further decompose $U_-$. We make the ansatz $$\label{decomp}
\left(-w,-w,w\right)=\frac{1}{2}(\rho_1v_1,v_1,\rho_1^2v_1)+\frac{1}{2}(\rho_2v_2,v_2,\rho_2^2v_2)$$ with $$\label{decomp2}
v_1=w,\hspace{0.2cm}v_2=-3w.$$ Then clearly is a rank-2 decomposition (in fact even rank-1), and and result in the conditions $$\label{rhosystem}
\begin{aligned}
-\rho_1+3\rho_2&=2\\
-\rho_1^2+3\rho_2^2&=-2.
\end{aligned}$$ A direct calculation shows that these requirements can be satisfied by two numbers $\rho_1,\rho_2>1$, and the proof is finished.
We proceed to Step (3). This consists in finding a continuous divergence-free map $U^1=(m^1,u^1,w^1):{\mathbb{T}}^3\to{\mathbb{R}}^{3\times 3}$ that approximately “recovers” the values $U_i(x)$ from the previous step. Given a family $(\lambda_i,U_i)_{i=1}^N$ satisfying the $H_N$-condition, we call the probability measure $$\nu:=\sum_{i=1}^N\lambda_i\delta_{U_i}\in\mathcal{P}({\mathbb{R}}^{3\times 3})$$ a *rank-2 laminate*, and if the $\lambda_i$ and $U_i$ depend on $x$, then we obtain a probability measure $\nu_x$ that depends on $x\in{\mathbb{T}}^3$. (Such measures are often called *parametrised probability measures* or *Young measures*).
If $K\in{\mathbb{R}}^{3\times 3}$ is compact, we denote by $K^{2lc}\supset K$ the *rank-2 lamination convex hull* of $K$, that is, the set of barycentres of rank-2 laminates supported on $K$. In other words, if $U\in K^{2lc}$, then $U$ is the convex combination of some matrices in $K$ that satisfy an $H_N$-condition.
The recovery lemma can then be stated as follows:
\[approx\] Let $K\subset{\mathbb{R}}^{3\times 3}$ be compact and $(\nu_x)_{x\in\Omega}$ be a family of probability measures such that
- the measure $\nu_x$ is a rank-2 laminate for almost every $x\in\Omega$,
- $\operatorname{supp}\nu_x\subset K$ for almost every $x$.
Assume further that $\psi\in C({\mathbb{R}}^{3\times3};{\mathbb{R}})$ is a non-negative function that vanishes on $K$. Then the barycentre $\bar{\nu}_x=\int_K V d\nu_x(V)$ is well-defined for almost every $x\in\Omega$, and for every ${\varepsilon}>0$ there exists a matrix-valued function $U$ such that
- $\operatorname{div}U=\operatorname{div}\bar{\nu}$ in the sense of distributions,
- $$\int_{\Omega}\psi(U(x))dx<{\varepsilon},$$
- $${\|\operatorname{dist}(U(x),K^{2lc})\|}_{L^{\infty}(\Omega)}<{\varepsilon},$$
- $$\label{expectationclose}
\int_{\Omega}\left|U(x)-\bar{\nu}_x\right|dx<\int_{\Omega}\int_{{\mathbb{R}}^{3\times3}}\left|V-\bar{\nu}_x\right|d\nu_x(V)dx+{\varepsilon}.$$
Moreover, if $\bar{\nu}\in C(\bar{\Omega})$, then $U$ can be chosen to satisfy $U\in C(\bar{\Omega})$.
The proof of this lemma is largely standard. By translation and localisation arguments, one reduces to the situation where $\nu$ is independent of $x$, and uses nested oscillations along rank-2 lines, as provided by . Note that all of this requires a localisation technique, i.e. a way to replace a divergence-free matrix field with another one of compact support, such that it is close to the original one on its support. It would be tempting to multiply a given field $U$ by a compactly supported cutoff function ${\varphi}:{\mathbb{T}}^3\to{\mathbb{R}}$, but then ${\varphi}U$ might not be divergence-free; instead, since $\operatorname{div}U=0$, there exists a potential $\Psi:{\mathbb{T}}^3\to{\mathbb{R}}^{3\times 3}$ such that $\operatorname{curl}\Psi=U$, and one localises considering $\tilde{U}:=\operatorname{curl}({\varphi}\Psi)$. We omit details.
In this way, we get a new matrix field $U^1=(m^1,u^1,w^1)$ that is much closer to $K_C$ that the original $U^0=(0,0,w)$. However, as Lemma \[approx\] still unavoidably contains errors of order ${\varepsilon}$, we have to repeat the procedure (Step (4)). Since $U^1$ is no longer of the special form $(0,0,w)$, we need a more general version of Lemma \[geom1\]:
\[geom2\] Let ${\varepsilon}>0$ and $\tilde{C}>1$. There exists a strictly increasing continuous function $h:[0,\infty)\to[0,\infty)$, depending only on $\tilde{C}$, with $h(0)=0$, and a number $\delta>0$, depending only on $\tilde{C}$ and ${\varepsilon}$, such that for every $1<C<\tilde{C}-{\varepsilon}$ and every $U\in{\mathbb{R}}^{3\times3}$ such that $\operatorname{dist}(U,K^{2lc}_{C})<\delta$, there exists a rank-2 laminate $\nu=\sum_{i=1}^n\lambda_i\delta_{U_i}$ such that $$\label{expectation}
U=\sum_{i=1}^n\lambda_iU_i,$$ $$\label{L1estimate}
\sum_{i=1}^n\lambda_i|U_i-U|\leq h\left(\operatorname{dist}(U,K_{C})\right),$$ and $$\operatorname{supp}\nu\subset K_{C+{\varepsilon}}.$$
The proof of this lemma is actually at the core of the whole construction, but we shall not discuss it here since the proof of Lemma \[geom1\] hopefully already gives a taste for the geometric arguments involved.
To finish up the proof, we need to define a sequence $(U^n)_{n\in{\mathbb{N}}}$ of divergence-free matrix fields whose average distance to $K_C$ converges to zero, for some $C>1$.
To this end, let $C_0>1$ be as required by Lemma \[geom1\] applied to $U^0(x)$ for *all* $x\in\bar{\Omega}$ (this is possible since $U^0$ is bounded). Next, pick a sequence $(C_n)_{n\geq0}$ that is strictly increasing such that $C_n\nearrow C_0+1=:C$ as $n\to\infty$. We also set ${\varepsilon}_n:=C_{n+1}-C_n$. Then, $({\varepsilon}_n)$ is a sequence of positive numbers converging to zero.
By Lemma \[geom1\] there exists for almost every $x\in{\mathbb{T}}^3$ a rank-2 laminate $\nu^0_x$ of finite order whose expectation is $U^0(x)$ and whose support is contained in $K_{C_0}$. This completes the definition of $U^0$ and $\nu^0$.
Suppose now that $U^n$ and $\nu^{n}$ have already been constructed for some $n\geq0$ in such a way that $\operatorname{supp}\nu^n\subset K_{C_n}$ and , , are satisfied, that is: $$\operatorname{div}(U^n)=(0,0,f)^T,$$ $$\label{nexpectation}
U^{n}(x)=\bar{\nu}_x^n,$$ $$\int_{{\mathbb{R}}^{3\times3}}|V-U^n(x)|d\nu_x^n(V)\leq h\left(\operatorname{dist}(U^n,K_{C_{n-1}})\right).$$ The last estimate is claimed only for $n\geq1$. By Lemma \[geom2\], where we set ${\varepsilon}={\varepsilon}_{n+1}$ and $\tilde{C}=C+1$, there exists $\delta_{n+1}=\delta({\varepsilon}_{n+1})$ such that whenever $$\operatorname{dist}(U,K^{2lc}_{C_n})<\delta_{n+1},$$ then there exists a rank-2 laminate whose expectation is $U$ and whose support is contained in $$\label{n+1}
K_{C_n+{\varepsilon}_{n+1}}\subset K_{C_{n+1}}.$$ Therefore we apply Lemma \[approx\] to $(\nu^n_x)$ with $K_{C_n}$, ${\varepsilon}=\delta_{n+1}$, and $$\psi=h\left(\operatorname{dist}({\,{\begin{picture}(1,1)(-0.5,-2.5)\circle*{2}\end{picture}}\,},K_{C_n})\right).$$ This yields a matrix field $U_{n+1}$ satisfying $$\operatorname{div}(U_{n+1})=\operatorname{div}\left(\bar{\nu}^n_x\right)=\operatorname{div}(U_n)=(0,0,f)^T,$$ $$\label{Kapprox}
\int_{\Omega}h\left(\operatorname{dist}(U_{n+1}(x),K_{C_n})\right)dx<\delta_{n+1},$$ and $$\label{distn+1}
{\|\operatorname{dist}(U_{n+1}(x),K_{C_n}^{2lc})\|}_{L^{\infty}(\Omega)}<\delta_{n+1}.$$ Therefore, by , we can indeed find, for every $x$, a rank-2 laminate $\nu^{n+1}_x$ with support in $K_{C_{n+1}}$ satisfying and . This completes the construction of the sequence $(U^n)$.
It is then not hard to show that $U^n\to U$ in $L^1({\mathbb{T}}^3;{\mathbb{R}}^{3\times 3})$, and that consequently $U(x)\in K_C$ for almost every $x$. Note that, in the construction of our sequence $(U^n)$, the admissible error ${\varepsilon}$ for Lemma \[approx\] is chosen as $\delta_{n+1}$, which typically is much smaller than the previous ${\varepsilon}$. This choice forces $U^{n+1}-U^n$ to oscillate at a much higher frequency that $U^n$ itself. This separation of frequencies, as one might call it, is typical of any convex integration type argument.
Onsager’s Conjecture {#onsagersec}
====================
Onsager’s conjecture on ${\mathbb{T}}^3$
----------------------------------------
Consider again the incompressible Euler equations, $$\begin{aligned}
\partial_t u+(u\cdot\nabla)u+\nabla p&=0\\
\operatorname{div}u&=0.
\end{aligned}$$ Here, $u:{\mathbb{T}}^3\times[0,T]\to{\mathbb{R}}^3$ denotes the velocity and $p:{\mathbb{T}}^3\times[0,T]\to{\mathbb{R}}$ the pressure of an ideal (i.e. inviscid) incompressible fluid, like approximately water. Multiplying this system by its velocity and integrating in space gives $$\frac{d}{dt}\int_{{\mathbb{T}}^3}|u|^2dx+\int_{{\mathbb{T}}^3}(u\cdot\nabla)u\cdot udx+\int_{{\mathbb{T}}^3}u\cdot\nabla p dx=0.$$ The last integral vanishes since $u$ is divergence-free, and for the middle one we compute, by integration by parts, $$\int_{{\mathbb{T}}^3}(u\cdot\nabla)u\cdot udx=\int_{{\mathbb{T}}^3}u_j\partial_ju_iu_i dx=-\int_{{\mathbb{T}}^3}\partial_ju_ju_iu_i dx-\int_{{\mathbb{T}}^3}u_ju_i\partial_ju_i dx,$$ and since the first integral on the right hand side again vanishes due to the divergence-free condition, it follows that $\int_{{\mathbb{T}}^3}(u\cdot\nabla)u\cdot udx=0$, so we are left with the conservation of energy, $$\label{globalen}
\frac{d}{dt}\int_{{\mathbb{T}}^3}|u|^2dx=0.$$ In fact, a more careful computation (that omits integration in space) yields the *local energy equality* $$\label{localen}
\partial_t\frac{|u|^2}{2}+\left(\left(\frac{|u|^2}{2}+p\right)u\right)=0,$$ which is the precise analogue (for $\eta(u)=\frac12|u|^2$) of the renormalised equation in the context of transport equations.
Note once more that the chain rule (or, on a related note, integration by parts) requires $u$ to have at least one full derivative. On the other hand, there are examples known of weak solutions that grossly violate and even , the first one being due to Scheffer [@Scheffer], whose solutions are no better than $L^2_{loc}$. The question thus arises whether there exists a threshold regularity that distinguishes dissipative from conservative solutions. In 1949, L. Onsager made the following conjecture [@ON]:
- If $u$ is a weak solution of the incompressible Euler equations with $u\in C^{\alpha}$ for an $\alpha>\frac{1}{3}$, then the energy is conserved.
- For every $\alpha<\frac{1}{3}$ there exists a weak solution $u\in C^\alpha$ that dissipates energy.
It may seem strange that Onsager, a physicist, would worry about the validity of the chain rule for non-differentiable functions at a time where no counterexamples to the chain rule were known in the first place. In fact, he didn’t; rather, his motivation stems from phenomenological turbulence theories (particularly the one of Kolmogorov) that make predictions on the energy spectrum of a fully turbulent fluid, which in turn lead to certain regularity properties of the velocity. This also means that non-conservative solutions of Euler are not always mathematical pathologies, but they are expected in turbulence theory and can be experimentally observed. We refer to [@frisch; @eyinksurvey; @shvydkoysurvey] for more physically oriented overviews.
Onsager’s conjecture has meanwhile largely been proved. Part b) was completed only recently in [@isett16; @buckmasteretal17], based on convex integration techniques whose development started with [@euler1]. Part a) is more classical – it was initially solved in [@GEY; @CET] and then refined in [@duchonrobert; @CCFS08; @FjWi].
To formulate Part a) more precisely, we have the following result:
\[eulerthm\] Let $(u,p)\in L^3\times L^{3/2}({\mathbb{T}}^3\times(0,T))$ be a weak solution of such that $$\label{EulerBesov}
\lim_{{\varepsilon}\to0} \int_0^T\int_{{\mathbb{T}}^3}\int_{B_{\varepsilon}(0)}\frac{|u(x)-u(x-y)|^3}{{\varepsilon}^{4}}dydxdt=0.$$ Then, the local energy inequality is satisfied in the sense of distributions.
To prove this, one proceeds exactly as in the proof of Theorem \[transportthm\], where the velocity now plays the role of both $\rho$ and $u$, thus forcing $p=q=3$ and $\alpha=\beta=\frac13$.
As before, note that condition is implied by $u\in L^3(0,T;C^{\alpha}({\mathbb{T}}^3))$ for some $\alpha>\frac13$, or by $u\in L^3(0,T;B_{3,\infty}^{\alpha}({\mathbb{T}}^3))$ for some $\alpha>\frac13$ (this is the space considered in [@CET]), or by $u\in L^3(0,T;B_{3,c_0}^{1/3}({\mathbb{T}}^3))$ (this is the space considered in [@CCFS08]). Another important remark is that the result is actually true in any space dimension (so the exponent $1/3$ has nothing to do with 3D space!). In contrast, Part b) of Onsager’s conjecture is still open in two dimensions.
Bounded domains
---------------
Let now $\Omega\in{\mathbb{R}}^3$ be a smooth bounded domain. Recall that the strategy for Theorems \[transportthm\] and \[eulerthm\] was to mollify the equation in space and thus to obtain $$\label{eulermolly}
\partial_t u_{\varepsilon}+\operatorname{div}(u_{\varepsilon}\otimes u_{\varepsilon})+\nabla p_{\varepsilon}=R_{\varepsilon}$$ for the commutator $R_{\varepsilon}=\operatorname{div}(u_{\varepsilon}\otimes u_{\varepsilon}-(u\otimes u)_{\varepsilon})$. It seems like an issue that the mollification of a function cannot be defined in an obvious way on a bounded domain. However, choosing a compactly supported test function ${\varphi}\in C^1_c(\Omega;{\mathbb{R}})$, $u_{\varepsilon}$ is well-defined on the support of ${\varphi}$ as long as ${\varepsilon}<\operatorname{dist}(\operatorname{supp}{\varphi},\partial\Omega)$. Therefore, testing against ${\varphi}$, everything is well-defined. Suppose now is satisfied on $\operatorname{supp}{\varphi}$ for each ${\varphi}\in C^1_c(\Omega)$; this means that holds locally in $\Omega$, but not necessarily uniformly up to the boundary. Then, just as in the proof of Theorem \[transportthm\] we obtain $u_{\varepsilon}\cdot R_{\varepsilon}\to0$ in the sense of distributions, so that the local energy equality holds in the sense of distributions on $\Omega$.
The idea is that the local energy equality is stronger than the global one , since one can deduce the latter from the former simply by integrating in space. On bounded domains, though, one should be a bit more careful.
Given , integration in space would amount to testing with ${\varphi}\equiv1$. But we are only allowed to test with ${\varphi}\in C^1_c(\Omega)$. So instead, let $\xi:{\mathbb{R}}_0^+\to{\mathbb{R}}_0^+$ be a smooth cutoff function that is zero near $s=0$ and one for $s\geq1$, and let $\xi_\delta(s):=\xi\left(\frac s\delta\right)$. Set ${\varphi}_\delta(x):=\xi_\delta(\operatorname{dist}(x,\partial\Omega))$, so that ${\varphi}_\delta\in C^1_c(\Omega)$, and test against ${\varphi}_\delta$:
$$\frac{d}{dt}\int_\Omega{\varphi}_\delta\frac{|u|^2}{2}dx=\int_\Omega \nabla{\varphi}_\delta\cdot u\left(\frac{|u|^2}{2}+p\right)dx.$$
The left hand side converges to $\frac{d}{dt}\int_\Omega\frac{|u|^2}{2}dx$ as $\delta\to0$. If we want to hold, therefore, we must show that the right hand side converges to zero.
Since $\partial\Omega$ is smooth, there exists a neighbourhood $\Gamma_0$ of $\partial\Omega$ where the orthogonal projection of $x\in\Gamma_0$ to $\partial\Omega$ is uniquely defined. Let’s call this projection $\sigma(x)\in\partial\Omega$, and denote by $n(\sigma(x))$ the outer unit normal to $\partial\Omega$ at $\sigma(x)$.
Then, $\nabla{\varphi}_\delta(x)$ is parallel to $n(\sigma(x))$ with absolute value bounded by $\frac C\delta$, and on the other hand, $\nabla{\varphi}_{\delta}$ is supported on $\Gamma_\delta:=\{\operatorname{dist}(x,\partial\Omega)<\delta\}$, where the volume of $\Gamma_\delta$ is comparable to $\delta$.
Hence, we have $$\left|\int_\Omega \nabla{\varphi}_\delta\cdot u\left(\frac{|u|^2}{2}+p\right)dx\right|\leq \frac C\delta\int_{\Gamma_\delta}\left|\frac{|u|^2}{2}+p\right||u(x)\cdot n(\sigma(x))|dx.$$ From this, it is easy to read off some sufficient conditions for global energy conservation: For instance, one could require $u$ and $p$ to be bounded in some neighbourhood of $\partial\Omega$, $x\mapsto u(x)\cdot n(\sigma(x))$ to be continuous in a neighbourhood of $\partial\Omega$, and $u\cdot n=0$ on $\partial\Omega$, which is the natural slip boundary condition for Euler anyway.
To summarise:
\[eulerboundary\] Let $(u,p)$ be a weak solution of such that $u$ satisfies locally in $\Omega$. Suppose $(u,p)$ is bounded in some neighbourhood of $\partial\Omega$, $x\mapsto u(x)\cdot n(\sigma(x))$ is continuous in a neighbourhood of $\partial\Omega$, and $u\cdot n=0$ on $\partial\Omega$. Then, the global energy equality holds in the sense of distributions.
This result is taken from [@BTW18], where the theorem is stated under much weaker assumptions (in particular on the pressure). Similar results, with some differences on the technical level, were obtained independently in [@DN18]. Earlier results on Onsager’s conjecture in domains with boundaries were obtained in [@RRS18; @BT18].
A useful feature of this result is that, near the boundary, it puts conditions only on the *normal* component of the velocity, but not on the tangential one. This gives rise to an application to the viscosity limit.
Recall the Navier-Stokes equations with viscosity $\nu>0$, $$\label{NSE}
\begin{aligned}
\partial_t u_\nu+(u_\nu\cdot\nabla)u_\nu+\nabla p_\nu&=\nu \Delta u_\nu\\
\operatorname{div}u_\nu&=0,
\end{aligned}$$ which are known to admit (for given initial data in $L^2$) global weak (so-called Leray-Hopf) solutions in $\Omega$, subject to the no-slip boundary condition $u=0$ on $\partial\Omega$. It is commonly expected that, for small $\nu$, the Navier-Stokes flow will behave like an Euler flow except on a boundary layer of thickness of order $\sqrt\nu$, where it decays steeply to $u=0$ at $\partial\Omega$. Thus, one expects the normal velocity component to be very small in the boundary layer, but the tangential velocity component to have a gradient of magnitude $\nu^{-1/2}$. Since Theorem \[eulerboundary\] makes no assumption on the tangential component, the result is consistent with the formation of such a boundary layer. More precisely:
Let $(u_\nu)_{\nu>0}$ be a family of Leray-Hopf weak solutions of with viscosity $\nu$ and initial data $u^0\in L^2(\Omega)$, and suppose this family satisfies the conditions of Theorem \[eulerboundary\] uniformly in $\nu$. Then, there exists a subsequence $\nu_k\to0$ such that $u_\nu\to u$ strongly in $L^\infty(0,T;L^2(\Omega))$, where $u$ is a weak solution of the Euler equations that conserves energy.
It is interesting to compare this result to [@constvic; @DN18a], where purely interior assumptions are made on the solution, and anomalous energy dissipation is therefore not excluded.
Statistical solutions
=====================
As phenomenological theories of turbulence are of a statistical nature, it is natural to place the fundamental PDEs of fluid mechanics in a probabilistic framework. A classical way to do this is given by DiPerna’s measure-valued solutions [@DiPerna], formulated for the incompressible Euler equations by DiPerna and Majda [@dipernamajda]. To illustrate the idea, consider again the incompressible Euler equations . Usually one wants to find a solution which is a vectorfield $u:\mathbb{T}^3\times[0,T]\to{\mathbb{R}}^3$, that is, at each (or at least almost every) point in space and time, one gives the velocity. In contrast, suppose the velocity at a point in space-time is not known exactly, but only as a probability distribution: This can be modelled by a map ${\mathbb{T}}^3\times[0,T]\to\mathcal P({\mathbb{R}}^3)$ from space-time into the set of probability measures on the phase space ${\mathbb{R}}^3$. Thus, at each point $(x,t)$, there is a probability measure $\mu_{x,t}$ such that the probability that the velocity is in a (measurable) subset $U\subset{\mathbb{R}}^3$ is given by $\mu_{x,t}(V)$.
Writing $\bar u(x,t):= \int_{{\mathbb{R}}^3}\xi d\mu_{x,t}(\xi)$ and $\overline{u\otimes u}(x,t):= \int_{{\mathbb{R}}^3}\xi\otimes\xi d\mu_{x,t}(\xi)$, we say that $\mu$ is a measure-valued solution of the Euler equations if $$\begin{aligned}
\partial_t \bar u+\overline{u\otimes u}+\nabla p&=0\\
\operatorname{div}\bar u&=0
\end{aligned}$$ in the sense of distributions. (Observe that $(u\cdot\nabla)u=\operatorname{div}(u\otimes u)$, which allows to write the equations in divergence form.) We deliberately ignore here the issue of possible concentrations requiring a generalised measure-valued setting, and refer to [@dipernamajda].
Although such solutions are easily shown to exist for any initial datum, and have other useful properties (see e.g. [@weakstrongsurvey]), they are felt by some to contain too little information. Indeed, it is not possible to represent with them two-point correlations, i.e. expressions of the form “the probability that the velocity at point $(x,t)$ is in $U_1$ and the velocity at $(y,t)$ is in $U_2$”. On a related note, it is not obvious how to make sense of a Besov condition like for a measure-valued solution.
To describe the statistics of a flow, Fjordholm et al. [@FjLaMi] propose a new notion of statistical solutions, where solutions to conservation laws are given as *correlation Young measures*. In the context of incompressible Euler, this was studied in [@FjWi]. The idea is to describe the two-point statistics of a fluid in terms of a parametrised measure $\mu_{x,y,t}\in\mathcal{P}({\mathbb{R}}^4\times{\mathbb{R}}^4)$, where $x,y\in {\mathbb{T}}^3$ are points from the space domain. The measure $\mu_{x,y}(du_1,dp_1,du_2,dp_2)$ can then be interpreted as the joint probability that the velocity and pressure at point $x$ are in $du_1\times dp_1$ and the velocity and pressure at point $y$ are in $du_2\times dp_2$.
More precisely, such a correlation measure is a pair $(\mu^1,\mu^2)$, where $\mu^1_{x,t}$ is a probability measure on ${\mathbb{R}}^4$ and $\mu^2:{x,t}$ is a probability measure on ${\mathbb{R}}^4\times {\mathbb{R}}^4$ for almost every $(x,t)$, such that the following conditions are satisfied:
1. *Symmetry:* If $f\in C_0({\mathbb{R}}^3\times{\mathbb{R}}^3)$ then $$\int_{{\mathbb{R}}^4\times{\mathbb{R}}^4} f(\xi,\eta)d\mu^2_{x,y,t}(\xi, \eta) = \int_{{\mathbb{R}}^3\times{\mathbb{R}}^3} f(\eta,\xi)d\mu^2_{y,x,t}(\eta,\xi)$$ for a.e. $x,y\in {\mathbb{R}}^3$;
2. *Consistency:* If $f\in C_0({\mathbb{R}}^4\times{\mathbb{R}}^4)$ is of the form $f(\xi,\eta) = g(\xi)$ for some $g\in C_0({\mathbb{R}}^4)$, then $$\int_{{\mathbb{R}}^4\times{\mathbb{R}}^4} f(\xi,\eta)d\mu^2_{x,y,t}(\xi, \eta)=\int_{{\mathbb{R}}^4}g(\xi)d\mu^1_{x,t}(\xi)$$ for almost every $(x,y)\in {\mathbb{R}}^3\times {\mathbb{R}}^3$.
One can now give a natural definition of $\mu$ to be a solution of the Euler equations: Indeed, if $(u,p)$ is a smooth solution and $u^i$ denotes the $i$-th velocity component, then $$\begin{split}
\partial_t\left(u^i(x)u^j(y)\right) + \sum_k \partial_{x^k}\left(u^i(x)u^k(x)u^j(y)\right) + \sum_k\partial_{y^k}\left (u^i(x)u^k(y)u^j(y)\right) \\
~ + \partial_{x^i}\left(p(x)u^j(y)\right) + \partial_{y^j} \left(u^i(x)p(y)\right) = 0,
\end{split}$$ as can be seen by adding the equation evaluated at $x$ and multiplied by $u^j(y)$ to the equation evaluated at $y$ and multiplied by $u^i(x)$. Thus, replacing the occurrences of the velocity and pressure by the correlation measure, one is led to the equation $$\begin{split}
\partial_t\langle\mu^2_{x,y}u_1^iu_2^j\rangle + \sum_k \partial_{x^k}\langle\mu^2_{x,y}u_1^iu_1^ku_2^j\rangle + \sum_k\partial_{y^k}\langle\mu^2_{x,y}u_1^iu_2^ku_2^j\rangle \\
~ + \partial_{x^i}\langle\mu^2_{x,y}p_1u_2^j\rangle + \partial_{y^j} \langle\mu^2_{x,y}u_1^ip_2\rangle = 0,
\end{split}$$ where we wrote $\langle\mu^2_{x,y}u_1^iu_2^j\rangle=\int_{{\mathbb{R}}^4\times{\mathbb{R}}^4}u_1^iu_2^jd\mu_{x,y}(u_1,p_1,u_2,p_2)$ etc. (hence the dummy variables $u_1, u_2, p_1, p_2$ stand for $u(x), u(y), p(x), p(y)$, respectively).
The natural extension of the Besov-type assumption is then $$\lim_{{\varepsilon}\to0} \int_0^T\int_{{\mathbb{T}}^3}\int_{B_{\varepsilon}(0)}\frac{\langle \mu^2_{x,x-y}; |u_1-u_2|^3\rangle}{{\varepsilon}^{4}}dydxdt=0,$$ which is the measure-valued analogue of .
In [@FjWi] we show that the energy is (locally) conserved under this assumption.
General conservation laws
=========================
A crucial observation made in [@FGSW] is that the commutator estimates remain valid even when the nonlinearities involved are not quadratic. For instance, for the isentropic compressible Euler system $$\label{compE}
\begin{aligned}
\partial_t(\rho u)+\operatorname{div}(\rho u\otimes u)+\nabla p(\rho)&=0,\\
\partial_t \rho+\operatorname{div}(\rho u)&=0,
\end{aligned}$$ one formally has the local conservation of energy: $$\partial_t\left(\frac{\rho |u|^2}{2}+P(\rho)\right)+\operatorname{div}\left[\left(\frac{\rho |u|^2}{2}+P(\rho)+p(\rho)\right)v\right]= 0,$$ where $P$ is the pressure potential as defined in . To prove this rigorously for weak solutions with a commutator argument, one needs to estimate, e.g., $$p(\rho)_{\varepsilon}-p(\rho_{\varepsilon}).$$ If $\rho\mapsto p(\rho)$ is twice differentiable in the closure of the range of $\rho$, Taylor expansion yields $$p(\rho_{\varepsilon})\sim p(\rho) + p'(\rho)(\rho_{\varepsilon}-\rho) + \frac12 p''(\rho)(\rho_{\varepsilon}-\rho)^2$$ as well as $$p(\rho(y))\sim p(\rho(x)) + p'(\rho(x))(\rho(y)-\rho(x)) + \frac12 p''(\rho(x))(\rho(y)-\rho(x))^2.$$ Multiplication of the latter by $\chi_{\varepsilon}(x-y)$, integration w.r.t. $y$, and subtraction of both equations yields $$|p(\rho_{\varepsilon})-p(\rho)_{\varepsilon}|\lesssim \|p\|_{C^2}(\rho_{\varepsilon}-\rho)^2,$$ so we have reduced the problem again to a quadratic nonlinearity. In this way, one arrives at the following result:
\[compressibleonsager\] Let $\rho$, $v$ be a solution of in the sense of distributions. Assume $\rho$ and $\rho v$ both satisfy and $v$ satisfies , where also a time shift is considered[^1]\[footnote\]. Suppose further $$0 \leq \underline{\rho} \leq \rho \leq \overline{\rho} \ \mbox{a.a. in} (0,T)\times\mathbb{T}^d,$$ for some constants $\underline{{\varrho}}$, $\overline{{\varrho}}$, and
$$2\alpha+\beta>1,\quad \alpha+2\beta>1, \quad p=q=3.$$
Assume moreover that $p \in C^2[\underline{{\varrho}}, \overline{{\varrho}}]$, and, in addition $$p'(0) = 0 \ \mbox{as soon as}\ \underline{{\varrho}} = 0.$$ Then the energy is locally conserved, i.e. $$\partial_t\left(\frac{1}{2}\rho|u|^2+P(\rho)\right)+\operatorname{div}\left[\left(\frac{1}{2}\rho|u|^2+p(\rho)+P(\rho)\right)u\right]=0$$ in the sense of distributions on $(0,T)\times\mathbb{T}^d$.
Some remarks are in order. The result bears resemblance to Theorem \[transportthm\], the most important difference arguably being the symmetry between the regularity and integrability indices $\alpha, \beta, p, q$; this comes from the fact that, unlike for the transport equation, there are now several commutators to control, in some of which the terms $\rho, \rho u$ appear twice and $u$ appears once, and vice versa. Secondly, we need to control the Besov-type regularity now also in time, since there is the nonlinear term $\rho u$ under the time derivative. (A possible way to avoid this is to write in conservative variables, thus replacing $\rho u$ by $m$, and to obtain the energy equality upon testing with $\frac{m_{\varepsilon}}{\rho_{\varepsilon}}$ rather than $u_{\varepsilon}$. This idea, carried out in [@LSh] for the inhomogeneous incompressible Euler equations, succeeds in avoiding any assumption on time regularity, but leads to trouble with vacuum states.)
The assumption $p\in C^2$ in the closure of the range of $\rho$ will typically require *absence of vacuum*: A common choice for the pressure is the polytropic law $p(\rho)=\rho^\gamma$, $\gamma>1$, and often $\gamma\leq 5/3$ (the exponent $5/3$ corresponds to a monoatomic gas). As such a function $p$ is twice differentiable only away from $\rho=0$, the assumption to justify the Taylor expansion boils down to absence of vacuum states. Note that this property (namely, $\rho\geq c>0$) is not necessarily propagated in time: Suppose the initial density satisfies $\rho_0\geq c>0$, then the maximum principle for transport equations [@dipernalions] implies that $\rho$ will remain bounded away from zero if the divergence of the velocity is bounded. This, however, is not necessarily satisfied for weak solutions of (not even for the compressible Navier-Stokes system). We will get back to the vacuum problem in the next section.
More generally, the Taylor expansion strategy applies to essentially any system of conservation laws that possesses an entropy [@GwMiSw; @BTWP218; @BGSTW]. To illustrate the point, let $$\partial_t u+\partial_x f(u)=0$$ be a scalar conservation law in one dimension, so that $u:{\mathbb{R}}\times[0,T]\to{\mathbb{R}}$ is the unknown and $f$ is a given smooth flux function. Let $\eta:{\mathbb{R}}\to{\mathbb{R}}$ be any convex function (an *entropy*) and $q:{\mathbb{R}}\to{\mathbb{R}}$ a corresponding *entropy flux*, meaning that $q'=f'\eta'$. Then it is easy to see that, again by the chain rule, the *entropy equality* $$\partial_t\eta(u)+\partial_xq(u)=0$$ holds, at least if $u$ is smooth. (The entropy equality is really the same thing as the renormalised equation for transport equations, or the local energy equality for the Euler equations.)
Using the arguments outlined above, one can then show that this is the case if holds. Thus, the exponent $1/3$ appears universally, simply because the leading order in the Taylor expansion of a smooth function which does not commute with a mollification is the second order. Remarkably, shocks provide an easy example that the 1/3-condition is optimal.
Degenerate cases
================
The reduction of an arbitrary nonlinearity to a quadratic one relies crucially on the boundedness of second derivatives for the nonlinearities and entropies involved. There are at least two interesting cases when this condition fails: First, when a transport equation is to be renormalised with $\eta(\rho)=|\rho|^p$ with $p<2$, and secondly when one considers the compressible Euler system with possible vacuum, for pressure laws $p(\rho)=\rho^\gamma$ with $1<\gamma<2$ (these are the physically interesting ones).
These problems have been studied in [@AkWi] and [@AkDeSkWi], respectively. We give here a brief outline of the latter. Recall the discussion of the compressible Euler system in the previous section. One idea is to approximate $p$ locally uniformly by a function $p^\delta$ with bounded second derivatives (up to $\rho=0$ of course). The main error term introduced by this additional layer of approximation takes the form $$\int_{\mathbb{T}^3}u_{\varepsilon}\cdot\nabla[p^\delta(\rho)_{\varepsilon}-p(\rho)_{\varepsilon}]dx=-\int_{\mathbb{T}^3}\operatorname{div}u_{\varepsilon}[p^\delta(\rho)_{\varepsilon}-p(\rho)_{\varepsilon}]dx,$$ and thus converges to zero, uniformly in ${\varepsilon}$, provided $\operatorname{div} u$ is a bounded measure. While for the Euler equations this condition (which is quite popular in the theory of hyperbolic conservation laws, see [@ChenFrid]) cannot be guaranteed a priori, for the compressible Navier-Stokes equations it follows directly from the energy estimate. We therefore obtain that the compressible Navier-Stokes equations conserve energy, even with possible vacuum, under the assumptions of Theorem \[compressibleonsager\] except the $C^2$ condition on the pressure law. This yields a nice complement to the reuslts in [@ChengYu].
If one is not prepared to make such an assumption on $\operatorname{div}u$, one needs to consider two commutator terms involving the pressure, which are therefore sensitive to the condition that $p''$ be bounded. These two commutators, which appear in the course of the computation, are $$R^1_{\varepsilon}:=\int_{\mathbb{T}^3}\operatorname{div} u_{\varepsilon}(p(\rho_{\varepsilon})-p(\rho)_{\varepsilon})dx,$$ and $$R^2_{\varepsilon}:=\int_{\mathbb{T}^3}\operatorname{div} (\rho_{\varepsilon}u_{\varepsilon}-(\rho v)_{\varepsilon}) P'(\rho_{\varepsilon})dx.$$
It is not difficult to show $R^1_{\varepsilon}\to0$ as long as $\rho$, $u$ satisfy , (in the space-time sense of footnote \[footnote\]) with $p=q=3$ and $$\gamma\alpha+\beta>1,$$ which is a stronger version of the previous condition $2\alpha+\beta>1$. (To show convergence of $R^1_{\varepsilon}$ under this condition, one uses “Taylor expansion to order $\gamma$”, see [@AkDeSkWi Lemma 4.2].)
The other commutator $R^2_{\varepsilon}$ is more delicate. We compute $$\begin{aligned}
R^2_{\varepsilon}&=\int_{\mathbb{T}^3}\operatorname{div} (\rho_{\varepsilon}u_{\varepsilon}-(\rho u)_{\varepsilon}) P'(\rho_{\varepsilon})dx\\
&=- \int_{\{\rho_{\varepsilon}>0\}} (\rho_{\varepsilon}u_{\varepsilon}-(\rho u)_{\varepsilon}) \cdot P''(\rho_{\varepsilon})\nabla\rho_{\varepsilon}dx\\
&\sim \int_{\{\rho_{\varepsilon}>0\}} (\rho_{\varepsilon}-\rho) (u_{\varepsilon}- u) \cdot \rho_{\varepsilon}^{\gamma-2}\nabla\rho_{\varepsilon}dx,
\end{aligned}$$ then split the domain of integration into $B^{\varepsilon}:=\{0<\rho_{\varepsilon}<{\varepsilon}^\alpha\}$ and $C^{\varepsilon}:=\{\rho_{\varepsilon}\geq{\varepsilon}^\alpha\}$, and only consider integration over $B^{\varepsilon}$ here (the $C^{\varepsilon}$ part is easier):
$$\label{333}
\begin{aligned}
&\left|\int_{B^{\varepsilon}} (\rho_{\varepsilon}-\rho) (u_{\varepsilon}- u) \cdot \rho_{\varepsilon}^{\gamma-2}\nabla\rho_{\varepsilon}dx\right|\\
&\leq \int_{B^{\varepsilon}} \left|\frac{\rho_{\varepsilon}-\rho}{\rho_{\varepsilon}}\right| |u_{\varepsilon}- u| \left|\rho_{\varepsilon}^{\gamma-1}\right||\nabla\rho_{\varepsilon}| dx\\
&\leq {\varepsilon}^{\alpha(\gamma-1)}\int_{\mathbb{T}^3} \left|\frac{\rho_{\varepsilon}-\rho}{\rho_{\varepsilon}}\right| |u_{\varepsilon}- u| |\nabla\rho_{\varepsilon}| dx\\
&\leq {\varepsilon}^{\alpha(\gamma-1)}\|u_{\varepsilon}-u\|_{L^3}\|\nabla\rho_{\varepsilon}\|_{L^3}\left\|\frac{\rho_{\varepsilon}-\rho}{\rho_{\varepsilon}}\right\|_{L^3}\\
&\lesssim {\varepsilon}^{\gamma\alpha+\beta-1}\left\|\frac{\rho_{\varepsilon}-\rho}{\rho_{\varepsilon}}\right\|_{L^3}.
\end{aligned}$$
Now all is well as long as $\left\|\frac{\rho_{\varepsilon}-\rho}{\rho_{\varepsilon}}\right\|_{L^3}$ is bounded uniformly in ${\varepsilon}$; however, in general, this may be false: If $p>1$, then there are smooth non-negative functions $\rho$ such that $\left\|\frac{\rho_{\varepsilon}-\rho}{\rho_{\varepsilon}}\right\|_{L^p}\to\infty$ as ${\varepsilon}\searrow 0$ [@AkDeSkWi Subsection 4A]. Only for $p=1$ are we able to control this term [@AkDeSkWi Lemma 4.3].
This suggests we should use a Hölder $\infty-\infty-1$ estimate instead of a $3-3-3$ estimate in ; for this, in turn, we need to require $\rho$ and $v$ to be Hölder continuous. We thus arrive at the following result:
Replace $2\alpha+\beta$ with $\gamma\alpha+\beta$ as well as and with $\rho\in C^\alpha(\mathbb{T}^3\times(0,T))$ and $u\in C^\beta(\mathbb{T}^3\times(0,T))$ in the assumptions of Theorem \[compressibleonsager\], and allow for $1<\gamma<2$, $\rho\geq0$. Then the energy is conserved.
[10]{}
I. Akramov and E. Wiedemann. Renormalization of active scalar equations. :254–269, 2019.
I. Akramov, T. Dbiec, J. Skipper, and E. Wiedemann. Energy conservation for the compressible Euler and Navier-Stokes equations with vacuum. .
G. Alberti, S. Bianchini, and G. Crippa. Structure of level sets and Sard-type properties of Lipschitz maps: results and counterexamples. :863–902, 2013.
C. Bardos and E. Titi. Onsager’s conjecture for the incompressible Euler equations in bounded domains. :197–207, 2018.
C. Bardos, E. Titi, E. Wiedemann. Onsager’s conjecture with physical boundaries and an application to the vanishing viscosity limit. :291–310, 2019.
C. Bardos, P. Gwiazda, A. Świerczewska-Gwiazda, E. S. Titi, and E. Wiedemann. On the extension of Onsager’s conjecture for general conservation laws. :501–510, 2019. C. Bardos, P. Gwiazda, A. Świerczewska-Gwiazda, E. S. Titi, and E. Wiedemann. Onsager’s conjecture in bounded domains for the conservation of entropy and other companion laws. :20190289, 18pp., 2019.
T. Buckmaster, C. De Lellis, L. Székelyhidi, Jr., and V. Vicol. Onsager’s Conjecture for admissible weak solutions. :229–274, 2019.
G.-Q. Chen and H. Frid. Divergence-measure fields and hyperbolic conservation laws. :89–118, 1999.
A. Cheskidov, P. Constantin, S. Friedlander, and R. Shvydkoy. Energy conservation and [O]{}nsager’s conjecture for the [E]{}uler equations. :1233–1252, 2008.
P. Constantin, W. E, and E. S. Titi. Onsager’s conjecture on the energy conservation for solutions of [E]{}uler’s equation. :207–209, 1994.
P. Constantin and V. Vicol. Remarks on high Reynolds numbers hydrodynamics and the inviscid limit. :711–724, 2018.
G. Crippa, N. Gusev, S. Spirito, and E. Wiedemann. Non-uniqueness and prescribed energy for the continuity equation. :1937–1947, 2015.
G. Crippa, N. Gusev, S. Spirito, and E. Wiedemann. Failure of the chain rule for the divergence of bounded vector fields. :1–18, 2017.
C. De Lellis and L. Székelyhidi, Jr. The Euler equations as a differential inclusion. :1417–1436, 2009.
C. De Lellis and L. Székelyhidi, Jr. On admissibility criteria for weak solutions of the Euler equations. :225–260, 2010.
C. De Lellis and L. Székelyhidi, Jr. Dissipative continuous Euler flows. :377–407, 2013.
N. Depauw. Non unicité des solutions bornées pour un champ de vecteurs ${B}{V}$ en dehors d’un hyperplan. :249–252, 2003.
R. J. DiPerna. Measure-valued solutions to conservation laws. :227–270, 1985.
R. J. DiPerna and A. J. Majda. Oscillations and concentrations in weak solutions of the incompressible fluid equations. :667–689, 1987.
R. J. DiPerna and P.-L. Lions. Ordinary differential equations, transport theory and Sobolev spaces. :511–547, 1989.
T. D. Drivas and H. Q. Nguyen. Onsager’s conjecture and anomalous dissipation on domains with boundary. :4785–4811, 2018.
T. D. Drivas and H. Q. Nguyen. Remarks on the emergence of weak Euler solutions in the vanishing viscosity limit. :709–721, 2019.
J. Duchon and R. Robert. Inertial energy dissipation for weak solutions of incompressible [E]{}uler and [N]{}avier-[S]{}tokes equations. :249–255, 2000.
L. C. Evans. Partial differential equations. Second edition. Graduate Studies in Mathematics, 19. 2010. xxii+749 pp.
G. L. Eyink. Energy dissipation without viscosity in ideal hydrodynamics, I. Fourier analysis and local energy transfer, [*Phys. D*]{} [**78**]{}:222–240, 1994.
G. L. Eyink and K. R. Sreenivasan. Onsager and the theory of hydrodynamic turbulence. :87–135, 2006.
E. Feireisl. On compactness of solutions to the compressible isentropic Navier-Stokes equations when the density is not square integrable. :83–98, 2001.
E. Feireisl, P. Gwiazda, A. [Ś]{}wierczewska-Gwiazda, and E. Wiedemann. Regularity and [E]{}nergy [C]{}onservation for the [C]{}ompressible [E]{}uler [E]{}quations. :1375–1395, 2017.
U. S. Fjordholm, S. Lanthaler, and S. Mishra. Statistical solutions of hyperbolic conservation laws: foundations. :809–849, 2017.
U. S. Fjordholm and E. Wiedemann. Statistical solutions and Onsager’s conjecture. :259–265, 2018.
U. Frisch. Turbulence. The legacy of A. N. Kolmogorov. 1995.
P. Gwiazda, M. Michálek, and A. Świerczewska-Gwiazda. A note on weak solutions of conservation laws and energy/entropy conservation. :1223–1238, 2018.
P. Isett and V. Vicol. Hölder continuous solutions of active scalar equations. , Art. 2, 2015.
P. Isett. A proof of Onsager’s Conjecture. :871–963, 2018.
T. M. Leslie and R. Shvydkoy. The energy balance relation for weak solutions of the density-dependent Navier-Stokes equations. :3719–3733, 2016.
P.-L. Lions. Mathematical topics in fluid mechanics. Vol. 2. Compressible models. Oxford Lecture Series in Mathematics and its Applications, 10. Oxford Science Publications. 1998. xiv+348 pp.
S. Modena and L. Székelyhidi, Jr. Non-uniqueness for the transport equation with Sobolev vector fields. , Art. 18, 2018.
S. Modena and L. Székelyhidi, Jr. Non-renormalized solutions to the continuity equation. 2018.
S. Modena and G. Sattig. Convex integration for the transport equation with full dimensional concentration. 2019.
L. Onsager. Statistical hydrodynamics. [*Nuovo Cimento*]{} [**6**]{}:279–287, 1949.
J. Robinson, J. L. Rodrigo, and J. W. D. Skipper. Energy conservation in the 3D Euler equations on ${\mathbb{T}}^2\times{\mathbb{R}}_+$ for weak solutions defined without reference to the pressure. :185–202, 2018.
V. Scheffer. An inviscid flow with compact support in space-time. :343–401, 1993.
A. Shnirelman. On the nonuniqueness of weak solution of the Euler equation. :1261–1286, 1997.
A. Shnirelman. Weak solutions with decreasing energy of incompressible Euler equations. (3):541–603, 2000.
R. Shvydkoy. Lectures on the Onsager conjecture. :473–496, 2010.
R. Shvydkoy. Convex integration for a class of active scalar equations. :1159–1174, 2011.
E. Wiedemann. Weak-strong uniqueness in fluid dynamics. 289–326, London Math. Soc. Lecture Note Ser., 452, [*Cambridge Univ. Press, Cambridge,*]{} 2018.
C. Yu. Energy conservation for the weak solutions of the compressible Navier-Stokes equations. :1073–1087, 2017.
[^1]: This means that turns into $$\lim_{{\varepsilon}\to0} \int_0^T\int_{{\mathbb{T}}^d}\int_{B_{\varepsilon}(0)}\frac{|\rho(x,t)-\rho(x-y,t-\tau)|^p}{{\varepsilon}^{d+1+\alpha p}}dyd\tau dxdt=0,$$ where the ball with radius ${\varepsilon}$ is considered in space-time. Analogously, one needs to alter .
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In this paper, we study the asymptotic (large time) behavior of a selection-mutation-competition model for a population structured with respect to a phenotypic trait, when the rate of mutation is very small. We assume that the reproduction is asexual, and that the mutations can be described by a linear integral operator. We are interested in the interplay between the time variable $t$ and the rate $\var$ of mutations. We show that depending on $\alpha > 0$, the limit $\var \to 0$ with $t = \var^{-\alpha}$ can lead to population number densities which are either Gaussian-like (when $\alpha$ is [[small]{}]{}) or Cauchy-like (when $\alpha$ is [[large]{}]{}).'
address:
- 'Departament de Matemàtiques, Universitat Autònoma de Barcelona, 08193 Bellaterra (Barcelona), Spain'
- 'Departament de Matemàtiques, Universitat Autònoma de Barcelona, 08193 Bellaterra (Barcelona), Spain'
- 'Univ. Paris Diderot, Sorbonne Paris Cité, Institut de Mathématiques de Jussieu - Paris Rive Gauche, UMR 7586, CNRS, Sorbonne Universités, UPMC Univ. Paris 06, F-75013, Paris, France.'
- 'CMAP, Ecole Polytechnique, CNRS, Université Paris-Saclay, Route de Saclay, 91128 Palaiseau cedex, France.'
author:
- Àngel Calsina
- Sílvia Cuadrado
- Laurent Desvillettes
- Gaël Raoul
title: 'Asymptotic profile in selection-mutation equations: Gauss versus Cauchy distributions'
---
Introduction
============
Selection-mutation-competition models {#subsec:sel-mut-comp}
-------------------------------------
[[The phenotypic diversity of a species impacts its ability to evolve. In particular, the importance of the variance of the population along a phenotypic trait is illustrated by the *fundamental theorem of natural selection* [@Fisher], and the *breeder’s equation* [@Lush]: the evolution speed of a population along a one dimensional fitness gradient (or under artificial selection) is proportional to the variance of the initial population. Recently, the phenotypic variance of populations has also come to light as an important element to describe the evolutionary dynamics of ecosystems (where many interacting species are considered) [@Violle; @Bolnick; @Vellend].]{}]{}
[[Over the last decade, the thematic of *Evolutionary Rescue* has emerged as an important question [@Bell; @Carlson; @Gonzales] (see also the seminal work of Luria and Delbrück [@Luria]), and led to a new interest in the phenotypic distribution of populations, beyond phenotypic variance]{}]{}. Evolutionary Rescue is concerned with a population living in an environment that changes suddenly. The population will survive either if some individuals in the population carry an unusual trait that turns out to be successful in the new environment, or if new mutants able to survive in the new environment appear before the population goes extinct (see [@Martin] for a discussion on the relative effect of *de novo mutations* and *standing variance* in Evolutionary Rescue). In any case, the fate of the population will not be decided by the properties of the bulk of its density, but rather by the properties of the tail of the initial distribution of the populations, close to the favourable traits for the new environment. A first example of such problem comes from emerging disease [@Gandon]: Animal infections sometimes are able to infect humans. This phenomena, called zoonose, is the source of many human epidemics: HIV, SARS, Ebola, MERS-CoV, etc. A zoonose may happen if a pathogen that reaches a human has the unusual property of being adapted [[to this new human host.]{}]{} A second example comes from the emergence of microbes resistant to an antimicrobial drug that is suddenly spread in the environment of the microbe. This second phenomenon can easily be tested experimentally [@Bell; @Toprak], and [[has major public health implications]{}]{} [[[@Canton]]{}]{}.
[[Most papers devoted to the genetic diversity of populations structured by a continuous phenotypic trait describe the properties of mutation-selection equilibria. It is however also interesting to describe the genetic diversity of population that are not at equilibrium (*transient dynamics*):]{}]{} pathogen populations for instance are often in transient situations, either invading a new host, or being eliminated by the immune system. We refer to [@Hastings] for a review on transient dynamics in ecology. For asexual populations [[structured by a continuous phenotypic trait]{}]{}, several models exist, corresponding to different biological assumptions [@Champagnat]. If the mutations are modeled by a diffusion, the steady populations [[(for a model close to , but where mutations are modelled by a Laplacian)]{}]{} are Gaussian distributions [@Kimura; @Burger]. [[Furthermore, [@Alfaro; @Coville] have considered some transient dynamics for this model. In the model that we will consider (see ), the mutations are modelled by a non-local term. It was shown in [@Burger2] (see also [@Burger]) that mutation-selection equilibria are then Cauchy profiles (under some assumptions), and this result has been extended to more general mutation kernels in [@Calsina], provided that the mutation rate is small enough.]{}]{} Finally, let us notice that the case of sexual population is rather different, since recombinations by themselves can imply that a *mutation-recombination equilibrium* exists, even without selection. We refer to the infinitesimal model [@Bulmer], and to [@Turelli] for some studies on the phenotypic distribution of sexual species in a context close to the one presented here for asexual populations.
In this article, we consider a population consisting of individuals structured [[by]{}]{} a quantitative phenotypic trait $x \in I$ ($I$ open interval of ${\mathbb{R}}$ containing $0$), and denote by $f : = f(t,x) \ge 0$ its density. Here, the trait $x$ is fully inherited by the offspring (if no mutation occurs), so that $x$ is indeed rather a breeding value than a phenotypic trait (see [@Mather]). We assume that the individuals reproduce with a rate $1$, and die at a rate $$x^{2} + \int_{I}f(t,y)\,\mbox{d}y.$$ This means that the individuals with trait $x=0$ are those who are best adapted to their environment, and that the fitness decreases like a parabola around this optimal trait (this is expected in the surroundings of a trait of maximal fitness). It also means that the strength of the competition modeled by the logistic term is identical for all traits. When an individual of trait $x\in I$ gives birth, we assume that the offspring will have the trait $x$ with probability $1-\varepsilon$, and a different trait $x'$ with probability $\var\in(0,1)$. $\var$ is then the probability that a mutation affects the phenotypic trait of the offspring. We can now define the growth rate of the population of trait $x$ (that is the difference between the rate of *births without mutation*, minus the death rate) as $$r_\var(t,x) = 1-\varepsilon -x^{2} - \int_{I}f(t,y)\,\mbox{d}y.$$ When a mutation affects the trait of the offspring, we assume that the trait $x'$ of the mutated offspring is drawn from a law over the set of phenotypes $I\subset \mathbb R$ with a density $\gamma := \gamma (x)\in L^1(I)$. The function $\gamma$ then satisfies $$\gamma(x)\ge 0,\quad \int_I \gamma(x)\, dx = 1,$$ and we assume moreover that $\gamma$ is bounded, $C^{1}$, with bounded derivative and strictly positive on $I$. The main assumption here is that the law of the trait of a mutated offspring does not depend of the trait of its parent. This classical assumption, known as *house of cards* is not the most realistic, but it can be justified when the mutation rate is small [@Burger] (see also [@Calsina][[)]{}]{}. All in all, we end up with the following equation: $$\label{eqq0}
\frac{\partial f_\var(t,x)}{\partial t}= r_\var(t,x) \, f_\var(t,x) +\varepsilon \,\gamma(x)\,\int_{I}f_\var(t,y)\,\mbox{d}y.
$$ This paper is devoted to the study of the asymptotic behaviour of the solutions of equation when $\varepsilon$ is small and $t$ large and it is organized as follows. In the rest of Section 1 the main results are quoted, first in an informal way, and then as rigourous statements. Section 2 contains the proof of Theorem \[theorem1\] and its corollary and finally, in Section 3, Theorem \[thm:smallt\] is proved.
Asymptotic study of the model {#subsec:asymptotics}
-----------------------------
When we consider the solutions of , two particular profiles naturally appear:
- *A Cauchy profile:* For a given mutation rate $\var>0$ small enough, one expects that $f_\var(t,x)$ will converge, as $t$ goes to infinity, to the unique steady-state of , wich is the following Cauchy profile $$\label{eqq2}
f_\var(\infty,x) := \frac{\var\, \gamma(x)\, \mathcal I_\var(\infty)}{\mathcal I_\var(\infty) - (1-\var) + x^2} ,$$ where $\mathcal I_\var(\infty)$ is such that $\int_I f_\var(\infty,x)\,dx=\mathcal I_\var(\infty)$. This steady-state of is the so-called *mutation-selection equilibrium* of the *House of cards* model , which has been introduced in [@Burger2] (we also refer to [@Burger] for a broader presentation of existing results).
- *A Gaussian profile:* If $\var=0$, the solution of (\[eqq0\]) can be written $$\label{eqq1}
f_0(t,x) = f(0,x) \, e^{- \int_0^t \mathcal I_0(s)\, ds + t - t\,x^2} ,$$ where $\mathcal I_0(t):=\int_I f_0(t,x)\,dx$, so that a Gaussian-like behavior (with respect to $x$) naturally appears in this case. Surprisingly, we are not aware of any reference to this property in the population genetics literature.
We will show that, as suggested by the above arguments, we can describe the phenotypic distribution of the population, that is $x\mapsto f_\var(t,x)$, when either $t\gg 1$ (large time for a given mutation rate $\varepsilon>0$), or $0\leq\var\ll 1$ (small mutation rate, for a given time interval $t\in [0,T]$). Before providing the precise statements of our results (see Subsection \[subsec:rigorous\]), we will briefly describe them here, and illustrate them with numerical simulations. The numerical simulations presented in Fig. \[fig1\] and Fig. \[fig2\] are obtained thanks to a finite difference scheme (explicit Runge-Kutta in time), and we illustrate our result with a single simulation of with $\var=10^{-2}$, $I=[-3/2,3/2]$, $\gamma(x)=\frac 1{40\pi}e^{\frac{-x^2}{20}}$ and $f_\varepsilon(0,x)=\Gamma_2(\var,x-1)$ (see the definition of $\Gamma_2$ in eq. (\[eqq4\]) below). The initial condition corresponds to a population at the mutation-selection equilibrium which environment suddenly changes (the optimal trait originally in $x=1$ moves to $x=0$ at $t=0$). This example is guided by the Evolutionary Rescue experiments described in Subsection \[subsec:sel-mut-comp\], where the sudden change is obtained by the addition of e.g. salt or antibiotic to a bacterial culture.
We describe two phases of the dynamics of the population:
- *Large time: Cauchy profile.* We show that $f_\var(t,x)$ is asymptotically (when the mutation rate $\var>0$ is small) close to $$\label{eqq4}
\Gamma_2(\var,x) =\frac{\var\, \gamma(0)}{\gamma(0)^2\pi^2\,\var^2 + x^2},$$ provided $t\gg \var^{-4}$[[. The population is then a time-independent Cauchy distribution for large times]{}]{}. This theoretical result is coherent with [[numerical]{}]{} results: we see in Fig. \[fig1\] that $f_\var(t,\cdot)$ is well described by $\Gamma_2(\var,\cdot)$, as soon as $t\geq 10^5$, which is confirmed by the value of $\|f_\var(t,\cdot)-\Gamma_2(\var,\cdot)\|_{L^1(I)}$ for $t\geq 10^5$ given by Fig. \[fig2\].
- *Short time: Gaussian profile.* We also show that $f_\var(t,x)$ is asymptotically (when the mutation rate $\var>0$ is small) close to $$\label{eqq3}
\Gamma_1(t,\var,x) = \frac{f(0,x)\, \sqrt t }{f(0,0)\int_I e^{-x^2}\,dx}e^{-x^2 \,t},$$ provided $1\ll t\ll \var^{-2/3}$[[. The population has then a Gaussian-type distribution for]{}]{} short (but not too short) times. This theoretical result is coherent with [[numerical]{}]{} results: we see in Fig. \[fig1\] that $f_\var(t,\cdot)$ is well described by $\Gamma_1(t,\var,\cdot)$ for $t\in[10^2,10^4]$, which is confirmed by the value of $\|f_\var(t,\cdot)-\Gamma_2(\var,\cdot)\|_{L^1(I)}$ for $t\in[10^2,10^4]$ given by Fig. \[fig2\].
![The different graphs correspond to different time points, from $t=0$ to $t=175\,000$, of the same simulation of for $\varepsilon=10^{-2}$ (see in the text for a complete description). In each of these plots, the blue (resp. red, black) line represents $x\mapsto f_\var(t,x)$ (resp. $x\mapsto\Gamma_1(t,\var,x)$, $x\mapsto\Gamma_2(\var,x)$). Note that in this figure, the scales of both axis change from one graph to the other, to accommodate with the dynamics of the solution $f(t,\cdot)$.[]{data-label="fig1"}](b0.jpg "fig:")![The different graphs correspond to different time points, from $t=0$ to $t=175\,000$, of the same simulation of for $\varepsilon=10^{-2}$ (see in the text for a complete description). In each of these plots, the blue (resp. red, black) line represents $x\mapsto f_\var(t,x)$ (resp. $x\mapsto\Gamma_1(t,\var,x)$, $x\mapsto\Gamma_2(\var,x)$). Note that in this figure, the scales of both axis change from one graph to the other, to accommodate with the dynamics of the solution $f(t,\cdot)$.[]{data-label="fig1"}](b10.jpg "fig:")![The different graphs correspond to different time points, from $t=0$ to $t=175\,000$, of the same simulation of for $\varepsilon=10^{-2}$ (see in the text for a complete description). In each of these plots, the blue (resp. red, black) line represents $x\mapsto f_\var(t,x)$ (resp. $x\mapsto\Gamma_1(t,\var,x)$, $x\mapsto\Gamma_2(\var,x)$). Note that in this figure, the scales of both axis change from one graph to the other, to accommodate with the dynamics of the solution $f(t,\cdot)$.[]{data-label="fig1"}](b100.jpg "fig:")![The different graphs correspond to different time points, from $t=0$ to $t=175\,000$, of the same simulation of for $\varepsilon=10^{-2}$ (see in the text for a complete description). In each of these plots, the blue (resp. red, black) line represents $x\mapsto f_\var(t,x)$ (resp. $x\mapsto\Gamma_1(t,\var,x)$, $x\mapsto\Gamma_2(\var,x)$). Note that in this figure, the scales of both axis change from one graph to the other, to accommodate with the dynamics of the solution $f(t,\cdot)$.[]{data-label="fig1"}](b1000.jpg "fig:")![The different graphs correspond to different time points, from $t=0$ to $t=175\,000$, of the same simulation of for $\varepsilon=10^{-2}$ (see in the text for a complete description). In each of these plots, the blue (resp. red, black) line represents $x\mapsto f_\var(t,x)$ (resp. $x\mapsto\Gamma_1(t,\var,x)$, $x\mapsto\Gamma_2(\var,x)$). Note that in this figure, the scales of both axis change from one graph to the other, to accommodate with the dynamics of the solution $f(t,\cdot)$.[]{data-label="fig1"}](b10000.jpg "fig:")![The different graphs correspond to different time points, from $t=0$ to $t=175\,000$, of the same simulation of for $\varepsilon=10^{-2}$ (see in the text for a complete description). In each of these plots, the blue (resp. red, black) line represents $x\mapsto f_\var(t,x)$ (resp. $x\mapsto\Gamma_1(t,\var,x)$, $x\mapsto\Gamma_2(\var,x)$). Note that in this figure, the scales of both axis change from one graph to the other, to accommodate with the dynamics of the solution $f(t,\cdot)$.[]{data-label="fig1"}](b30000.jpg "fig:")![The different graphs correspond to different time points, from $t=0$ to $t=175\,000$, of the same simulation of for $\varepsilon=10^{-2}$ (see in the text for a complete description). In each of these plots, the blue (resp. red, black) line represents $x\mapsto f_\var(t,x)$ (resp. $x\mapsto\Gamma_1(t,\var,x)$, $x\mapsto\Gamma_2(\var,x)$). Note that in this figure, the scales of both axis change from one graph to the other, to accommodate with the dynamics of the solution $f(t,\cdot)$.[]{data-label="fig1"}](b100000.jpg "fig:")![The different graphs correspond to different time points, from $t=0$ to $t=175\,000$, of the same simulation of for $\varepsilon=10^{-2}$ (see in the text for a complete description). In each of these plots, the blue (resp. red, black) line represents $x\mapsto f_\var(t,x)$ (resp. $x\mapsto\Gamma_1(t,\var,x)$, $x\mapsto\Gamma_2(\var,x)$). Note that in this figure, the scales of both axis change from one graph to the other, to accommodate with the dynamics of the solution $f(t,\cdot)$.[]{data-label="fig1"}](b175000.jpg "fig:")
![Simulation of with $\varepsilon=10^{-2}$ (see in the text for a complete description). The red line represents $\|f_\var(t,\cdot)-\Gamma_1(t,\var,\cdot)\|_{L^1(I)}$, while the black line represents $\|f_\var(t,\cdot)-\Gamma_2(\var,\cdot)\|_{L^1(I)}$.[]{data-label="fig2"}](distances.jpg)
Another way to look at these results is to [[consider]{}]{} $t\geq 0$ and $\var>0$ as two parameters, and to see the approximations presented above as approximations of $f_\var(t,\cdot)$ for some set of parameters: $f_\var(t,\cdot)\sim_{\var\to 0} \Gamma_2(\var,\cdot)$ for $(t,\var)\in\{(\tilde t,\tilde\var);\tilde t\gg \tilde{\var}^{-4}\}$, while $f_\var(t,\cdot)\sim_{\var\to 0} \Gamma_1(t,\var,\cdot)$ for $(t,\var)\in\{(\tilde t,\tilde\var);1\ll\tilde t\ll \tilde{\var}^{-2/3}\}$. We have represented these sets in Fig \[fig3\].
![Representation of the set $\{(\tilde t,\tilde\var);\tilde t\gg \tilde{\var}^{-4}\}$ (in blue), where the approximation $f_\var(t,\cdot)\sim_{\var\to 0} \Gamma_2(\var,\cdot)$ holds provided that $\var>0$ is small enough; and of the set $\{(\tilde t,\tilde\var);1\ll\tilde t\ll \tilde{\var}^{-2/3}\}$ (in red), where the approximation $f_\var(t,\cdot)\sim_{\var\to 0} \Gamma_1(t,\var,\cdot)$ holds provided that $\var>0$ is small enough.[]{data-label="fig3"}](set3.jpg)
As described in the Subsection \[subsec:sel-mut-comp\], the phenotypic distribution of species is involved in many ecological and epidemiological problematics. Our study is a general analysis of this problem and we do not have a particular application in mind. An interesting and (to our knowledge) new feature described by our study is that the tails of the traits distribution in a population can change drastically between “short times”, that is $1\ll t\ll \var^{-2/3}$ and “large times”, that is $t\gg \var^{-4}$: the distribution is initially close to a Gaussian distribution, with small tails, and then converges to a thick tailed Cauchy distribution. [[This result could have significant consequences in for *evolutionary rescue*: the tails of the distribution then play an important role. Quantifying the effect of this property of the tails of the distributions would however require further work, in particular on the impact of stochasticity (the number of pathogen is typically large, but finite). The plasticity of the pathogen (see [@Chevin]) may also play an important role.]{}]{}
Rigorous statements {#subsec:rigorous}
-------------------
Here we state the two main theorems of the paper[[, each of them followed by a corollary]{}]{}. To do so we start by defining the linear operator
$$(A_{\varepsilon}f) (x):=(1-\varepsilon)f(x)
-x^{2}\,f(x)+\varepsilon \gamma(x)\,\int_{I}f(y)\mbox{d}y$$ and denoting by $\lambda_{\varepsilon}$ the dominant eigenvalue of $A_{\varepsilon}$ and by $\psi_{\varepsilon}(x)=\frac{\varepsilon \gamma(x)}{\lambda_{\varepsilon}-(1-\varepsilon)+x^2}$ the corresponding eigenvector (see Proposition \[proposition1\]).
\[theorem1\] Let us assume that the initial datum $f_0\ge 0$ is integrable on $I$ ($I= ]a,b[$, $-\infty \le a < b \le +\infty$), and $f_0$ is not identically (i.-e. a.e.) $0$.
Then the initial value problem for (\[eqq0\]) with $f(0,x)=f_0(x)$ has a unique (global for positive times) mild solution. Moreover, for $\varepsilon>0$ small enough, and any $\rho_{\varepsilon} < (\gamma(0)\pi\varepsilon)^2$, there exists a constant $C_{\varepsilon}>0$ (depending on $f_0$ and $\var$) such that $$\left\|f(t,\cdot)-\lambda_{\varepsilon}\,\psi_{\varepsilon} \right\|_{L^1(I)} \leq C_{\varepsilon}\,
e^{-\rho_{\varepsilon}\,t}.$$ Furthermore, taking $\rho_{\varepsilon}=\frac{\alpha_{\varepsilon}}{2}=\frac{\lambda_{\varepsilon}-(1-\varepsilon)}{2}$, the following more explicit (in terms of dependence w.r.t $\var$) estimate holds $$\left\|f(.,t)-\lambda_{\varepsilon}\,\psi_{\varepsilon} \right\|_{L^1(I)} \leq
K\,
\varepsilon^{\frac{-\hat{K}}{\varepsilon^2}}\, e^{\frac{-\alpha_{\varepsilon}t}{2}},$$ where $K, \hat{K}>0$ depend on $f_0$ but not on $\var$.
\[cor1\] Under the same hypotheses, there exists positive constants $K, \hat{K}$ and $\tilde{K}$ (independent of $\varepsilon$) such that $$\left\|f(t,\cdot) - \frac{\varepsilon
\gamma(0)}{(\gamma(0)\pi\varepsilon)^2+x^2}\right\|_{L^{1}(I)} \leq K\,
\varepsilon^{\frac{-\hat{K}}{\varepsilon^2}}\,
e^{-\hat{K}\varepsilon^2 t}+ \tilde{K}\varepsilon
\ln{\left(\frac{1}{\varepsilon}\right)}.$$
\[thm:smallt\] Let $\gamma\in C^0(I)\cup L^\infty(I)$ such that $\int_I \gamma(x)\,dx=1$. Let $f(0,\cdot)\in W^{1,\infty}(I)$ satisfying $f(0,0)>0$ and $\int_I f(0,x)\,dx<1$. There exists a constant $C>0$ such that the solution $f\in C^1(\mathbb R_+\times I)$ of satisfies [[$$\label{est:final-smallt}
\forall t\geq 0,\quad \Bigg\| x \mapsto f(t,x)-\frac{f(0,x)\sqrt t e^{-x^2 t}}{f(0,0)\,\int_I e^{-y^2}\,dy}\Bigg\|_{L^1(I)}\leq C \left(\frac 1{\sqrt t}+\varepsilon\, t^{\frac 3 2}\,e^{C\,\varepsilon\, t}\right).$$]{}]{}
\[rem:C\] As can be seen from the proof, the constant $C$ appearing in indeed only depends on some upper bounds on $\|\gamma\|_{L^\infty}$, $\|f(0,\cdot)\|_{W^{1,\infty}}$ and a lower bound on $f(0,0)$, and on $|a-b|$.
In particular, Theorem \[thm:smallt\] implies the following description of the population’s phenotypic diversity during transitory times, that is times $t$ satisfying $1\ll t\ll\varepsilon^{-\frac 23}$:
\[cor:smallt\] Let $\gamma\in C^0(I)\cup L^\infty(I)$ such that $\int_I \gamma(x)\,dx=1$. Let $f(0,\cdot)\in W^{1,\infty}(I)$ satisfying $f(0,0)>0$ and $\int_I f(0,x)\,dx<1$. There exists $C>0$ such that for $\kappa>0$ small enough, as soon as $\varepsilon<\kappa$, the solution $f\in C^1(\mathbb R_+\times I)$ of satisfies $$\forall t\in\left[\kappa^{-2},\kappa^{\frac 23}\varepsilon^{-\frac 23}\right],\quad\Bigg\|f(t,x)-\frac{f(0,x)\sqrt t e^{-x^2 t}}{f(0,0)\,\int_I e^{-y^2}\,dy}\Bigg\|_{L^1}\leq C\kappa.$$
These results hold for models which are slightly more general than equation (\[eqq0\]). In fact, in both theorems one can assume that the competition term is a weighted population instead of the total population number. In Theorem \[thm:smallt\], one could also assume that the mutation kernel depends on the parents trait.
Proof of Theorem \[theorem1\] and of Corollary \[cor1\]
=======================================================
We start here the proof of Theorem \[theorem1\]. We recall that $I=]a,b[$, $-\infty \leq a <0< b \leq \infty$, and $\gamma: = \gamma(x)$ is a bounded, $C^1$ function with bounded derivative, such that $\gamma(x) > 0$ and $\int_{I}
\gamma(x) \,dx=1$. We begin with the study of the linear operator associated to eq. (\[eqq0\]).
Spectrum of the linear operator
-------------------------------
Let us recall that $$\label{eq2} (A_{\varepsilon}f) (x):=(1-\varepsilon)f(x)
-x^{2}\,f(x)+\varepsilon \gamma(x)\,\int_{I}f(y)\,dy$$ is the operator corresponding to the linear part in eq. (\[eqq0\]). It acts on functions of the variable $x \in I$. We begin with a basic lemma which enables to define the semigroup associated with this operator.
\[lemanou\] The linear operator $A_{\varepsilon}$, defined on $L^1(I)$ and with domain $D(A_{\varepsilon})=\{f \in
L^1(I) : \int_{I} x^{2}\, |f(x)|\, \,dx < \infty \}$, generates an irreducible positive $C^0$-semigroup (denoted from now on by $T_{\varepsilon}(t)$).
The multiplication linear operator $(A_{\varepsilon}^{0}f)(x) := (1-\varepsilon)f(x) -x^{2}\,f(x)$ is the generator of a positive $C^0$-semigroup. Since $\gamma$ is strictly positive, $A_{\varepsilon}-A_{\varepsilon}^0 $ is a positive bounded perturbation whose only invariant closed ideals are ${0}$ and the whole space $L^1(I)$. So $T_{\varepsilon}(t)$ is irreducible (see [@clement], Corollary 9.22).
Next, we present a proposition which gives information about the spectrum of $A_{\varepsilon}$.
\[proposition1\] The linear operator $A_{\varepsilon}$ has only one eigenvalue. It is a strictly dominant algebraically simple eigenvalue $\lambda_{\varepsilon}
>1-\varepsilon$ and a pole of the resolvent, with corresponding normalized positive eigenvector $$\psi_{\varepsilon}(x)=\frac{\varepsilon \gamma(x)}{\lambda_{\varepsilon}-(1-\varepsilon-x^{2})}.$$ Moreover, for $\varepsilon$ small enough, $\lambda_{\varepsilon}<1$.
The rest of the spectrum of the linear operator $A_{\varepsilon}$ is equal to the interval $J =
[\min(1-\varepsilon-a^2,1-\varepsilon-b^2), 1-\varepsilon].$
In the sequel, the norm $||\,\, ||$ is the $L^1$ norm on $I$. Let us first show that any $\lambda$ belonging to the set $J=\text{Range}(1-\varepsilon -x^2)$ belongs to the spectrum of $A_{\varepsilon}$. In order to do this, for $\lambda=1-\varepsilon
-x_0^2, x_0 \in \mathring{I}$, let us define $f_n(x)=\frac{n}{2}\left(\chi_{[x_0,x_0+\frac{1}{n}]}(x)-\chi_{[x_0-\frac{1}{n},x_0]}(x)\right)$ for $n$ such that $[x_0-\frac{1}{n},x_0+\frac{1}{n}]\subset I$. We then have $\|f_n\|=1$ and $\left\|(A_{\varepsilon}-\lambda Id)f_n\right\| =
\frac{n}{2}\int_{x_0-\frac{1}{n}}^{x_0+\frac{1}{n}}|x^2-x_0^2|dx
\rightarrow 0$. So $(\min(1-\varepsilon-a^2,1-\varepsilon-b^2),
1-\varepsilon]$ is contained in the spectrum of $A_{\varepsilon}$. The claim follows from the fact that the spectrum is a closed set.
On the other hand, notice that (for $x_0 \in I$), $1-\varepsilon
-x_0^2$ is not an eigenvalue, since the potential corresponding eigenfunction $\frac{\gamma(x)}{x_0^2-x^2}$ is not an integrable function on $I$ (remember that $\gamma$ does not vanish).
Let us now compute the resolvent operator of $A_{\varepsilon}$, that is, let us try to solve the equation $$\label{eq3}
A_{\varepsilon}f-\lambda f= g \in L^1(I).$$ For $\lambda \notin J$, defining $p:= \int_{I}f(y)\,dy$, (\[eq3\]) gives $$\label{eq4} f(x)=\frac{\varepsilon
\gamma(x)p-g(x)}{\lambda-(1-\varepsilon-x^2)} .$$ Integrating, we get $$\label{eq5}
\bigg(1-\varepsilon
\int_{I}\frac{\gamma(x)}{\lambda-(1-\varepsilon-x^2)}\,dx
\bigg)\,p
=\int_{I}\frac{-g(x)}{\lambda-(1-\varepsilon-x^2)}\,dx ,$$ and $\lambda$ belongs to the resolvent set unless the factor of $p$ on the left hand side vanishes. Therefore $\sigma(A)=J \cup \{
\lambda \in \mathbb{C} :\varepsilon
\int_{I}\frac{\gamma(x)}{\lambda-(1-\varepsilon-x^2)}\,dx=1
\}$.
Since for any real number $\lambda>1-\varepsilon$, the function $F_{\varepsilon}(\lambda):= \varepsilon
\int_{I}\frac{\gamma(x)}{\lambda-(1-\varepsilon-x^2)}\,dx$ is continuous, strictly decreasing, and satisfies $\lim_{\lambda \to
1-\varepsilon}F_{\varepsilon}(\lambda)=+\infty$ (recall that $\gamma(0)>0$) and $\lim_{\lambda \to
+\infty}F_{\varepsilon}(\lambda)=0$, we see that there is a unique real solution of $F_{\varepsilon}(\lambda)=1$ in $(1-\varepsilon,\infty)$. We denote it by $\lambda_{\varepsilon}$.
Taking $g(x)=0$ in (\[eq3\]), we see that $\lambda_{\varepsilon}$ is an eigenvalue with corresponding normalized strictly positive eigenvector $$\psi_{\varepsilon}=\frac{\varepsilon
\gamma(x)}{\lambda_{\varepsilon}-\left(1-\varepsilon-x^2\right)}.$$
Taking $g(x)=\psi_{\varepsilon}$ and $\lambda = \lambda_\varepsilon$ we see that the left hand side in (\[eq5\]) vanishes, whereas the right hand side is strictly negative, so that $A_{\varepsilon}f-\lambda_{\varepsilon}f=\psi_{\varepsilon}$ has no solution and hence $\lambda_{\varepsilon}$ is algebraically simple.\
Indeed, it also follows from (\[eq5\]) that the range of $A_{\varepsilon}-\lambda_{\varepsilon}\, Id$ coincides with the kernel of the linear form defined on $L^1(I)$ by the $L^\infty$ function $\frac{1}{\lambda_\var-(1-\varepsilon)+x^2}$ (which is the eigenvector corresponding to the eigenvalue $\lambda_{\varepsilon}$ of the adjoint operator $A_{\varepsilon}^{*}$) and hence it is a closed subspace of $L^1(I)$. Therefore, $\lambda_{\varepsilon}$ is a pole of the resolvent (see Theorem A.3.3 of [@clement]). Furthermore, since $$F_{\varepsilon}(1) = \varepsilon
\int_{I}\frac{\gamma(x)}{\varepsilon
+x^{2}}\,dx=\int_{I}\frac{\gamma(x)}{1
+\left(\frac{x}{\sqrt{\varepsilon}}\right)^{2}}\,dx
\stackrel{\scriptscriptstyle \varepsilon \rightarrow
0}{\displaystyle \longrightarrow}0,$$ we see that $F_{\varepsilon}(1)< 1$ for $\varepsilon$ small enough, and hence $\lambda_{\varepsilon}<1$.
Substituting $\lambda$ by $a+bi$ in the characteristic equation $$\label{characteristic} 1+\varepsilon
\int_{I}\frac{\gamma(x)}{(1-\varepsilon-x^2-\lambda)}\,dx=0$$ we have that the imaginary part is $-\varepsilon b
\int_{I}\frac{\gamma(x)}{(1-\varepsilon-x^2-\lambda)}\,dx$. Since $\gamma(x)>0$, there are no non real solutions of
Note that $\lim_{\varepsilon \to 0} \lambda_{\varepsilon}=1.$
We now write an expansion of the eigenvalue $\lambda_{\varepsilon}$.
\[proposition2\] Let $\lambda_{\varepsilon}$ be the dominant eigenvalue of the operator $A_{\varepsilon}$. Then $$\left|\lambda_{\varepsilon}-(1-\varepsilon)-\gamma(0)^2\pi^{2}\varepsilon^{2}\right|=O\left(\varepsilon^3\ln{\frac{1}{\varepsilon}}\right)$$
Let us consider the change of variable $x=\nu_{\varepsilon}z$ where $\nu_{\varepsilon}=\sqrt{\lambda_{\varepsilon}-(1-\varepsilon)}$. We have $$1=\varepsilon
\int_{a}^{b}\frac{\gamma(x)}{(\lambda_{\varepsilon}-(1-\varepsilon-x^2))}\,dx=\varepsilon
\int_{\frac{a}{\nu_{\varepsilon}}}^{\frac{b}{\nu_{\varepsilon}}}\frac{\gamma(\nu_{\varepsilon}z)}
{\nu_{\varepsilon}^{2}+(\nu_{\varepsilon}z)^2}
\nu_{\varepsilon}\,dz=\frac{\varepsilon}{\nu_{\varepsilon}}
\int_{\frac{a}{\nu_{\varepsilon}}}^{\frac{b}{\nu_{\varepsilon}}}\frac{\gamma(\nu_{\varepsilon}z)}
{1+z^2}\,dz.$$ Then $$\begin{array}{rcl}
\Big| \frac{\nu_{\varepsilon}}{\varepsilon}- \gamma(0)\pi\Big| &=& \Big|\int_{\frac{a}{\nu_{\varepsilon}}}^{\frac{b}{\nu_{\varepsilon}}}\frac{\gamma(\nu_{\varepsilon}z)}
{1+z^2}\,dz-\gamma(0)\pi\Big|\\ \\
& \leq & \Big|\int_{\frac{a}{\nu_{\varepsilon}}}^{\frac{b}{\nu_{\varepsilon}}}\frac{\gamma(\nu_{\varepsilon}z)}
{1+z^2}\,dz- \int_{\mathbb{R}}\frac{\gamma(\nu_{\varepsilon}z)}
{1+z^2}\,dz\Big|+ \Big| \int_{\mathbb{R}}\frac{\gamma(\nu_{\varepsilon}z)-\gamma(0))}
{1+z^2}\,dz\Big|\\ \\
& \leq & 4 \|\gamma\|_{\infty}\int_{\frac{B}{\nu_{\varepsilon}}}^{+\infty}\frac{\,dz}{1+z^2}+2 \|\gamma'\|_{\infty}\nu_{\varepsilon}\int_{0}^{\frac{A}{\nu_{\varepsilon}}}\frac{z}{1+z^2}\,dz
\end{array}$$ where we have used $$|\gamma(\nu_{\varepsilon}z)-\gamma(0)|\leq \min \left(\|\gamma\|_{\infty},\|\gamma'\|_{\infty} \nu_{\varepsilon}|z|\right)$$ and have denoted $A:= \frac{\|\gamma\|_{\infty}}{\|\gamma'\|_{\infty}}$ and $B:=\min (|a|, b, A)$.\
Since $$4 \|\gamma\|_{\infty}\int_{\frac{B}{\nu_{\varepsilon}}}^{+\infty}\frac{\,dz}{1+z^2} = 4 \|\gamma\|_{\infty} \arctan{\left(\frac{\nu_{\varepsilon}}{B}\right)} \leq 4 \|\gamma\|_{\infty}\frac{\nu_{\varepsilon}}{B}$$ and $$2 \|\gamma'\|_{\infty}\nu_{\varepsilon}\int_{0}^{\frac{A}{\nu_{\varepsilon}}}\frac{z}{1+z^2}\,dz= \|\gamma'\|_{\infty}\nu_{\varepsilon} \ln{\left(1+\frac{A^2}{\nu_{\varepsilon}^2}\right)}$$ we obtain $$\label{ine10}
\Big|\nu_{\varepsilon}-\varepsilon\gamma(0)\pi\Big| \leq \varepsilon \nu_{\varepsilon} \left(\frac{4 \|\gamma\|_{\infty}}{B} + \|\gamma'\|_{\infty}
\ln{\left(1+\frac{A^2}{\nu_{\varepsilon}^2}\right)} \right)$$ which implies $$\begin{aligned}
&\varepsilon \left( \gamma(0)\pi - \nu_{\varepsilon}\left(\frac{4 \|\gamma\|_{\infty}}{B} + \|\gamma'\|_{\infty}
\ln{(1+\frac{A^2}{\nu_{\varepsilon}^2})} \right) \right) \\
&\quad \leq \nu_{\varepsilon} \leq \varepsilon \left( \gamma(0)\pi + \nu_{\varepsilon}\left(\frac{4 \|\gamma\|_{\infty}}{B} + \|\gamma'\|_{\infty}
\ln{\left(1+\frac{A^2}{\nu_{\varepsilon}^2}\right)} \right) \right).\end{aligned}$$ Since $$\nu_{\varepsilon}\left(\frac{4 \|\gamma\|_{\infty}}{B} + \|\gamma'\|_{\infty}
\ln{\left(1+\frac{A^2}{\nu_{\varepsilon}^2}\right)} \right)
\stackrel{\scriptscriptstyle \varepsilon \rightarrow
0}{\displaystyle \longrightarrow}0$$ we have $$\label{ineprop1}
\frac{\gamma(0)\pi\varepsilon}{2}\leq \nu_{\varepsilon} \leq 2\, \gamma(0)\pi\varepsilon.$$ for $\varepsilon$ small enough.\
Therefore, using in we get $$\label{ineprop2}
\Big|\nu_{\varepsilon}-\varepsilon\gamma(0)\pi\Big| \leq \varepsilon^2 2 \gamma(0) \pi \left(\frac{4 \|\gamma\|_{\infty}}{B} + \|\gamma'\|_{\infty}
\ln{\left(1+\frac{4A^2}{\gamma(0)^2 \pi^{2} \varepsilon^{2}}\right)} \right) \leq C \varepsilon^2 \ln{\left(\frac{1}{\varepsilon}\right)}.$$ Finally, by and ,
$$|\lambda_{\varepsilon}-(1-\varepsilon)-\gamma(0)^2\pi^{2}\varepsilon^{2}|=
|\nu_{\varepsilon}+\gamma(0)\pi\varepsilon|\;|\nu_{\varepsilon}-\gamma(0)\pi\varepsilon|\leq 3 \gamma(0) \pi C\varepsilon^3 \ln{\left(\frac{1}{\varepsilon}\right)}.$$
Asymptotic behavior of the nonlinear equation
---------------------------------------------
Let us start this subsection with a lemma in which properties of the spectrum of $\tilde{A}_{\varepsilon}=A_{\varepsilon}-\lambda_{\varepsilon}Id$ are used to study the asymptotic behavior of the semigroup $\tilde{T}_{\varepsilon}(t)$ generated by $\tilde{A}_{\varepsilon}$.
\[lemma2.0\]
- The essential growth bound of the semigroup generated by $\tilde{A}_{\varepsilon}$ is $\omega_{ess}(\tilde{T}_{\varepsilon}) = 1-\varepsilon-\lambda_{\varepsilon}.$
- The growth bound of the semigroup generated by $\tilde{A}_{\varepsilon}$ is $\omega_{0}(\tilde{T}_{\varepsilon}) = 0.$
<!-- -->
- $\tilde{A}_{\varepsilon}$ is a compact (one rank) perturbation of $\tilde{A}_{\varepsilon}^0f:=(1-\varepsilon-x^2-\lambda_{\varepsilon})f.$ Then $\omega_{ess}\left(\tilde{T}_{\varepsilon}\right)=\omega_{ess}\left(\tilde{T}_{\varepsilon}^0\right)$ where $\tilde{T}_{\varepsilon}^0(t)$ is the semigroup generated by $\tilde{A}_{\varepsilon}^0$ (see [@nagel]).
Since $\tilde{A}_{\varepsilon}^0$ is a multiplication operator, $\omega_{ess}\left(\tilde{T}_{\varepsilon}^0\right) =
1-\varepsilon-\lambda_{\varepsilon}$ and the result follows.
- By Proposition \[proposition1\], the spectral bound of $\tilde{A}_{\varepsilon}$ is $0$ and the spectral mapping theorem holds for any positive $C^{0}-$semigroup on $L^{1}$ (see [@clement]).
Let us now write, for a positive non identically zero $f_0$, $\left(\tilde{T}_{\varepsilon}(t)\right)f_{0}(x)=c_{f_{0}}\psi_{\varepsilon}(x)+v(t,x)$ where $\psi_{\varepsilon}(x)$ is the eigenvector corresponding to the eigenvalue 0 of $\tilde{A}_{\varepsilon}$ and $c_{f_{0}}\psi_{\varepsilon}(x)$ is the spectral projection of $f_0$ on the kernel of $\tilde{A}_{\varepsilon}$ (Note that $c_{f_{0}}>0$ since $f_{0}$ is positive and $\tilde{A}_{\varepsilon}$ is the generator of an irreducible positive semigroup). We also define $\varphi(t) := \int_I v(t,x)\, dx$. The following lemma gives the asymptotic behavior of $c_{f_0}$:
\[lemma4\] Let us assume that $f_0$ is a positive integrable function on $I$. Then there exist positive constants $K_1$, $K_2$ (independent of $\varepsilon$ but depending on $f_0$) such that $K_1\,\varepsilon^2 \leq c_{f_0} \leq K_2.$ Moreover, $\lim_{\varepsilon \to 0} c_{f_0} = 0.$
Recall that $c_{f_0}= \langle \psi_{\varepsilon}^{*},f_0 \rangle$ where $\psi_{\varepsilon}^{*}$ is the eigenvector of the adjoint operator $A_{\varepsilon}^{*}$ corresponding to the eigenvalue $\lambda_{\varepsilon}$, normalized such that $\langle
\psi_{\varepsilon}^{*},\psi_{\varepsilon} \rangle=1$. Since $$\psi_{\varepsilon}^{*}= \frac{{{\left(\varepsilon
\int_{I}\frac{\gamma(x)}{(\lambda_{\varepsilon}-(1-\varepsilon-x^2))^2}\,dx\right)^{-1}}}}{\lambda_{\varepsilon}-(1-\varepsilon-x^2)},$$ we see that $$c_{f_0}=
\frac{\int_{I}\frac{f_0(x)}{\lambda_{\varepsilon}-(1-\varepsilon-x^2)}\,dx}{\varepsilon
\int_{I}\frac{\gamma(x)}{(\lambda_{\varepsilon}-(1-\varepsilon-x^2))^2}\,dx}.$$ Let us start by bounding the denominator from above. Using that, by Proposition \[proposition2\], for $\varepsilon$ small enough, $\lambda_{\varepsilon}-(1-\varepsilon)
\geq \frac{(\gamma(0)\pi\varepsilon)^2}{2},$ we obtain the bound $$\label{ine1}
\begin{array}{rcl}
\varepsilon
\int_{I}\frac{\gamma(x)}{\left(\lambda_{\varepsilon}-(1-\varepsilon-x^2)\right)^2}\,dx
& \leq & \varepsilon \sup_{x}\gamma(x)
\,\int_{\mathbb{R}}\frac{1}{\left(\frac{(\gamma(0)\pi\varepsilon)^2}{2}+x^2\right)^2}\,dx
\\ \\ & = &\sup_{x}\gamma(x)\, \frac{\sqrt{2}}{\gamma(0)^3(\pi
\varepsilon)^2}=:\frac{K_0}{\varepsilon^2}.
\end{array}$$ Similarly, since for $\varepsilon$ small enough, $\lambda_{\varepsilon}-(1-\varepsilon) \leq
2(\gamma(0)\pi\varepsilon)^2$, so that $$\label{ine2} \varepsilon
\int_{I}\frac{\gamma(x)}{\left(\lambda_{\varepsilon}-(1-\varepsilon-x^2)\right)^2}\,dx
\geq \varepsilon \min_{[-a,a]}\gamma(x)
\int_{-a}^{a}\frac{\,dx}{\left(2(\gamma(0)\pi\varepsilon)^2+x^2\right)^2}
\geq \frac{K_3}{\varepsilon^2} .$$ For the numerator we have, on the one hand, $$\label{ine3} \varepsilon^{2}
\int_{I}\frac{f_0(x)}{\lambda_{\varepsilon}-(1-\varepsilon-x^2)}\,dx
\leq
\int_{I}\frac{\varepsilon^2}{\frac{(\gamma(0)\pi\varepsilon)^2}{2}+x^2}f_0(x)\,dx,$$ where the right hand side tends to $0$ when $\varepsilon$ goes to $0$ by an easy application of the Lebesgue dominated convergence theorem (note that the integrand is bounded above by $\frac{2}{(\gamma(0)\pi)^2}f_0(x)$).
On the other hand, notice that there exists an interval $J \subset I
$ which does not contain $0$ such that $\int_{J}f_0(x)\,dx>0$. Then, since $$\int_{I}\frac{f_0(x)}{\lambda_{\varepsilon}-\left(1-\varepsilon-x^2\right)}\,dx
\geq
\int_{J}\frac{f_0(x)}{\lambda_{\varepsilon}-\left(1-\varepsilon-x^2\right)}\,dx$$ and $$\lim_{\varepsilon \to 0}
\int_{J}\frac{f_0(x)}{\lambda_{\varepsilon}-\left(1-\varepsilon-x^2\right)}\,dx=\int_{J}\frac{f_0(x)}{x^2}\,dx
>0,$$ there exists a constant $K_4>0$ such that $$\label{ine4}
\int_{I}\frac{f_0(x)}{\lambda_{\varepsilon}-\left(1-\varepsilon-x^2\right)}\,dx
> K_4.$$ By and, $$c_{f_0}=\frac{\varepsilon^{2}
\int_{I}\frac{f_0(x)}{\lambda_{\varepsilon}-(1-\varepsilon-x^2)}\,dx}{\varepsilon^{3}
\int_{I}\frac{\gamma(x)}{\lambda_{\varepsilon}-(1-\varepsilon-x^2)}\,dx}
\leq \frac{\varepsilon^{2}
\int_{I}\frac{f_0(x)}{\lambda_{\varepsilon}-(1-\varepsilon-x^2)}\,dx}{K_3}
\stackrel{\scriptscriptstyle \varepsilon \rightarrow
0}{\displaystyle \longrightarrow}0$$ and by and , and $\varepsilon$ small enough, $$c_{f_0} \geq \frac{K_4}{\frac{K_0}{\varepsilon^2}}=:K_1
\varepsilon^2.$$ This completes the proof.
If $f_0(x)$ is bounded below by a positive number $c$ in a neighbourhood $(-\delta, \delta)$ of $0$, then the lower estimate can be improved using that $$\int_{-\delta}^{\delta}\frac{\varepsilon}{k^2
\varepsilon^2+x^2}\,dx = \frac{2}{k} \arctan\left(\frac{\delta}{k
\varepsilon}\right) \stackrel{\scriptscriptstyle \varepsilon \rightarrow
0^{+} }{\displaystyle \longrightarrow}\frac{\pi}{k}.$$ Indeed, for $\varepsilon$ small enough $$\begin{array}{rcl}
\varepsilon
\int_{I}\frac{f_0(x)}{\lambda_{\varepsilon}-(1-\varepsilon-x^2)}\,dx
& \geq & \varepsilon \int_{I}\frac{f_0(x)}{2 (\gamma(0) \pi
\varepsilon)^2+x^2}\,dx \\ \\ & \geq &c
\int_{-\delta}^{\delta}\frac{\varepsilon}{(\sqrt{2} \gamma(0) \pi)^2
\varepsilon^2+x²}\,dx \stackrel{\scriptscriptstyle \varepsilon
\rightarrow 0^{+}}{\displaystyle \longrightarrow}\frac{c}{\sqrt{2}
\gamma(0)}.
\end{array}$$ So in this case, for $\varepsilon$ small enough, $$c_{f_0} \geq
\frac{\frac{c}{\sqrt{2}\gamma(0)\varepsilon}}{\frac{K_0}{\varepsilon^2}}=:K\varepsilon$$ for some constant $K$ independent of $\varepsilon$.
The next two lemmas enable to estimate $\varphi(t)$ (defined above Lemma \[lemma4\]). In the first one, the dependence w.r.t. $\var$ is not explicit.
\[lemma1\] For $\varepsilon$ small enough and any $\rho_{\varepsilon} < (\gamma(0)\pi\varepsilon)^2$ there exists $K_{\varepsilon}>0$ such that $| \varphi(t)| \leq \|v(t,\cdot)\|
\leq K_{\varepsilon}\,e^{-\rho_{\varepsilon}t}\, \|f_{0}\|.$
Since by Lemma \[lemma2.0\] $\omega_{ess}(\tilde{A}_{\varepsilon})<\omega_{0}(\tilde{A}_{\varepsilon})$, we can apply Theorem $9.11$ in [@clement], and get the estimate $$\|
v(t,\cdot)\|=\|\tilde{T}_{\varepsilon}(t)f_{0}-c_{f_{0}}\psi_{\varepsilon}\|\leq
K_{\varepsilon}e^{-\eta t}\|f_{0}\| \quad \forall \eta <
\lambda_{\varepsilon}-(1-\varepsilon).$$ Proposition \[proposition2\] gives then the statement.
We now give an estimate of the dependence of $K_{\varepsilon}$ on $\varepsilon$, provided that $\rho_{\varepsilon}$ is chosen far enough from its limit value. More precisely, we choose $\rho_{\varepsilon}=\frac{\lambda_{\varepsilon}-(1-\varepsilon)}{2}=:\frac{\alpha_{\varepsilon}}{2}.$
\[lem6\] For $\varepsilon$ small enough there exists a constant $K$ independent of $\varepsilon$ and of $f_0$ such that $$\left\|\tilde{T}_{\varepsilon}(t)f_{0}-c_{f_{0}}\,\psi_{\varepsilon}\right\|\leq
K \,\varepsilon^{-4} \,e^{\frac{-\alpha_{\varepsilon}}{2}\, t}\,
\|f_0\| .$$
Since the proof of this result is quite technical we delay it to the end of this section (subsection \[app\]).
We now rewrite equation (\[eqq0\]) as $$\label{eq6}
\frac{\partial f(t,x)}{\partial t}= \tilde{A}_{\varepsilon}f(t,x)+ \bigg(\lambda_{\varepsilon}-\int_{I}f(t,y)\,dy \bigg)\, f(t,x) .$$ We look for solutions of (\[eq6\]) (with positive initial condition $f_{0} \in L^1(I)$) which can be written as $f(t,x)=h(t)(\tilde{T}_{\varepsilon}(t)f_{0})(x)$, with $h := h(t)$ a function of time such that $h(0)=1$. Substituting in (\[eq6\]), it follows that $f$ is indeed a solution of eq. (\[eqq0\]) if $h(t)$ satisfies the following initial value problem for an ordinary differential equation: $$\label{eq7}
h'(t)=\Big(\lambda_{\varepsilon}-h(t)\int_{I}\left(\tilde{T}_{\varepsilon}(t)f_{0}\right)(x)\,dx\Big)h(t),
\qquad h(0)=1,$$ or equivalently $$\label{eq8}
h'(t)=\Big(\lambda_{\varepsilon}-(c_{f_{0}}+\varphi(t))\,
h(t)\Big)\, h(t), \qquad h(0)=1.$$
The two next lemmas explain the asymptotic behavior of $h(t)$. In the first one, the dependence w.r.t. $\var$ of the constants is not explicit.
\[lemma3\] For $\varepsilon>0$ small enough and any $\rho_{\varepsilon} < (\gamma(0)\pi\varepsilon)^2$, there exists a positive constant $\hat{C}_{\varepsilon}>0$ such that $\left|h(t)-\frac{\lambda_{\varepsilon}}{c_{f_{0}}}\right| \leq
\hat{C}_{\varepsilon}\, e^{-\rho_{\varepsilon}\, t}.$
The solution of is explicitly given by $$h(t)=\frac{e^{\lambda_{\varepsilon}t}}{1+\int_{0}^{t}(c_{f_{0}}+\varphi(s))\,
e^{\lambda_{\varepsilon}s}\,
\,ds}=\frac{1}{e^{-\lambda_{\varepsilon}t}+\frac{c_{f_0}}{\lambda_{\varepsilon}}\,
(1-e^{-\lambda_{\varepsilon}t})+e^{-\lambda_{\varepsilon}t}\int_{0}^{t}
\varphi(s)\,e^{\lambda_{\varepsilon}s}\,{d}s}.$$ Then $$\begin{array}{rcl}
\left|h(t)-\frac{\lambda_{\varepsilon}}{c_{f_0}}\right|&=&\left|\frac{1}{e^{-\lambda_{\varepsilon}t}+\frac{c_{f_0}}{\lambda_{\varepsilon}}(1-e^{-\lambda_{\varepsilon}t})+e^{-\lambda_{\varepsilon}t}\int_{0}^{t}
\varphi(s)e^{\lambda_{\varepsilon}s}\,{d}s}-\frac{1}{\frac{c_{f_0}}{\lambda_{\varepsilon}}}\right|\\ \\
&=&
\frac{\frac{\lambda_{\varepsilon}}{c_{f_0}}\left|e^{-\lambda_{\varepsilon}t}\big(1-\frac{c_{f_0}}{\lambda_{\varepsilon}}\big)+e^{-\lambda_{\varepsilon}t}\int_{0}^{t}
\varphi(s)e^{\lambda_{\varepsilon}s}\,ds\right|}{e^{-\lambda_{\varepsilon}t}+e^{-\lambda_{\varepsilon}t}\int_{0}^{t}(c_{f_{0}}+\varphi(s))e^{\lambda_{\varepsilon}s}\,{d}s}\\ \\
& \leq &\hat{C}_{\varepsilon}e^{-\rho_{\varepsilon}t},
\end{array}$$ where for the last inequality we have used that the denominator is a positive continuous function bounded below (it takes the value $1$ for $t=0$ and its limit is $\frac{c_{f_0}}{\lambda_{\varepsilon}}$ when $t$ goes to infinity). We also used the following estimate for the numerator: since, by Lemma \[lemma1\], $|\varphi(s)| \leq
K_{\varepsilon}\,e^{-\rho_{\varepsilon}s}\|f_0\|$, then $$\begin{array}{rcl}
\left|e^{-\lambda_{\varepsilon}t}\big(1-\frac{c_{f_0}}{\lambda_{\varepsilon}}\big)+e^{-\lambda_{\varepsilon}t}\int_{0}^{t}
\varphi(s)\,e^{\lambda_{\varepsilon}s}\,ds\right|&\leq&
e^{-\lambda_{\varepsilon}t}\left(\left|1-\frac{c_{f_0}}{\lambda_{\varepsilon}}\right|-\frac{K_{\varepsilon}\,\|f_0\|}{\lambda_{\varepsilon}-\rho_{\varepsilon}}\right)+\frac{K_{\varepsilon}
\,\|f_0\|}{\lambda_{\varepsilon}-\rho_{\varepsilon}}\, e^{-\rho_{\varepsilon}t}\\
\\ & \leq & 2K_{\varepsilon}\, e^{-\rho_{\varepsilon}t}\|f_0\| .
\end{array}$$
In order to give an estimate of the dependence of $\hat{C}_{\varepsilon}$ w.r.t. $\varepsilon,$ we need to bound the denominator more precisely and to take a value of $\rho_{\varepsilon}$ separated of its limit value. As in Lemma \[lem6\], we choose $\rho_{\varepsilon}=\frac{\lambda_{\varepsilon}-(1-\varepsilon)}{2}=:\frac{\alpha_{\varepsilon}}{2}.$
\[lem5\] For $\varepsilon>0$ small enough, there exist constants $K_{7}$ and $K_{8}$ (independent of $\varepsilon$) such that $$\left|h(t)-\frac{\lambda_{\varepsilon}}{c_{f_0}}\right| \leq K_8\,
\varepsilon^{\frac{-K_7}{\varepsilon^2}}\,
e^{-\frac{\alpha_{\varepsilon} t}{2}}.$$
Using Lemma \[lemma1\] and the fact that the second term is positive we see that $$\begin{array}{rcl}
e^{-\lambda_{\varepsilon} t} + e^{-\lambda_{\varepsilon} t}
\int_0^t(c_{f_0}+\varphi(s))\,e^{\lambda_{\varepsilon} s} \, ds
&\geq& e^{-\lambda_{\varepsilon} t} + \max \left(0,
c_{f_0}(1-e^{-\lambda_{\varepsilon}
t})-K_{\varepsilon}e^{-\rho_{\varepsilon} t} \right) \\
\\ \geq
e^{-\lambda_{\varepsilon} t_{\varepsilon}}
\end{array}\label{eq666}$$ for any $t_{\varepsilon}$ such that $$\label{eq66} c_{f_0}\,(1-e^{-\lambda_{\varepsilon}
t_{\varepsilon}})-K_{\varepsilon}\, e^{-\rho_{\varepsilon}
t_{\varepsilon}} \geq e^{-\lambda_{\varepsilon} t_{\varepsilon}}.$$ (Notice that the left hand side in is an increasing function of $t_{\varepsilon}$). This indeed happens if $K_{\varepsilon}\, e^{-\rho_{\varepsilon} t_{\varepsilon}} \leq
\frac{c_{f_0}}{2}$ and $(1+c_{f_0})\,e^{-\lambda_{\varepsilon}
t_{\varepsilon}} \leq \frac{c_{f_0}}{2}.$ Since the second condition is weaker than the first one for $\varepsilon$ small enough, holds whenever $t_{\varepsilon}$ is such that $e^{-\rho_{\varepsilon} t_{\varepsilon}} \leq
\frac{c_{f_0}}{2K_{\varepsilon}}$, i.e., $e^{-\lambda_{\varepsilon}
t_{\varepsilon}} \leq \left( \frac{c_{f_0}}{2 K_{\varepsilon}}
\right)^{\frac{\lambda_{\varepsilon}}{\rho_{\varepsilon}}}$ and $\varepsilon>0$ is sufficiently small. So, $\left( \frac{c_{f_0}}{2
K_{\varepsilon}}
\right)^{\frac{\lambda_{\varepsilon}}{\rho_{\varepsilon}}}$ is also a lower bound in , and we finally have $$\left|e^{-\lambda_{\varepsilon} t} + e^{-\lambda_{\varepsilon} t}
\int_0^t(c_{f_0}+\varphi(s))e^{-\lambda_{\varepsilon} s} ds\right| \geq
\left( \frac{c_{f_0}}{2 K_{\varepsilon}}
\right)^{\frac{\lambda_{\varepsilon}}{\rho_{\varepsilon}}}.$$ Using the bound on the numerator given in the proof of Lemma \[lemma3\], the previous estimate and using also Lemma \[lem6\], Lemma \[lemma4\] and Proposition \[proposition2\], we obtain
$$\label{bound1}
\begin{array}{rcl}
\left|h(t)-\frac{\lambda_{\varepsilon}}{c_{f_0}}\right|& \leq &
\frac{2K_{\varepsilon}\, e^{-\rho_{\varepsilon}t}\|f_0\|}{\left(
\frac{c_{f_0}}{2 K_{\varepsilon}}
\right)^{\frac{\lambda_{\varepsilon}}{\rho_{\varepsilon}}}}\\ \\
& \leq &
\frac{2K_{5}\,\varepsilon^{-4}\,e^{-\frac{\alpha_{\varepsilon}t}{2}}\,\|f_0\|}{\left(\frac{K_1
\varepsilon^{2}}{2K_{5}\, \varepsilon^{-4}}\right)^{K_{6}\,
\varepsilon^{-2}}}\\
\\
& = &
2K_{5}\,\left(\frac{2K_5}{K_1}\right)^{-\frac{K_6}{\varepsilon^{2}}}\varepsilon^{-4-
\frac{6K_6}{\varepsilon^2}}\, e^{-\frac{\alpha_{\varepsilon}t}{2}}\,
\|f_0\|
\\
\\ &\leq& K_8 \,\varepsilon^{\frac{-K_7}{\varepsilon^2}}\,
e^{-\frac{\alpha_{\varepsilon}t}{2}}.
\end{array}$$
We are now in position to conclude the proof of Theorem \[theorem1\].
We recall that $h$ satisfies the integral equation $$h(t) = 1 +
\int_0^t
\bigg(\lambda_{\varepsilon}-h(s)\,\int_{I}\left(\tilde{T}_{\varepsilon}f_0\right)(x)\,
dx \, \bigg)\, h(s)\, ds$$ from which the following identity follows $$h(t)\tilde{T}_{\varepsilon}(t)f_0 = \tilde{T}_{\varepsilon}(t)f_0 +
\int_0^t
\tilde{T}_{\varepsilon}(t-s)\bigg(\lambda_{\varepsilon}-h(s)\int_{I}\left(\tilde{T}_{\varepsilon}(s)f_0\right)(x)dx\bigg)h(s)\tilde{T}_{\varepsilon}(s)f_0ds,$$ i.e., $f(x,t)$ is a solution of the variations of constants equation.\
On the other hand, the nonlinear part of the right hand side of (\[eq6\]) is a locally Lipschitz function of $f \in L^1(I)$. From this uniqueness follows, whereas global existence is clear from the previous lemmas.
Finally, a standard application of the triangular inequality and Lemmas \[lemma4\], \[lemma1\] and \[lemma3\] gives
$$\label{triangle}
\begin{array}{rcl}\left\|f(.,t)-\lambda_{\varepsilon}\psi_{\varepsilon}(x) \right\| &\leq &
\left|h(t)-\frac{\lambda_{\varepsilon}}{c_{f_0}}\right|\;\left\|\tilde{T}_{\varepsilon}(t)f_{0}\right\|
+\frac{\lambda_{\varepsilon}}{c_{f_0}}\left\|\tilde{T}_{\varepsilon}(t)f_{0}-c_{f_0}\,\psi_{\varepsilon}(x)\right\| \\ \\
& \leq & \hat{C}_{\varepsilon}\,e^{-\rho_{\varepsilon}t} \left(K_{2}+K_{\varepsilon}\,e^{-\rho_{\varepsilon}t}\, ||f_0||\right)+\frac{1}{K_1 \varepsilon^2}K_{\varepsilon}e^{-\rho_{\varepsilon}t} \\ \\
& \leq & C_{\varepsilon} \,e^{-\rho_{\varepsilon}t}.
\end{array}$$
Using Lemmas \[lem6\] and \[lem5\] in the second inequality of , the last statement of Theorem \[theorem1\] follows.
Proof of Corollary \[cor1\]
---------------------------
By the triangular inequality, $$\begin{array}{rcl}
\left\|f(t,\cdot) - \frac{\varepsilon \gamma(0)}{(\gamma(0)\pi\varepsilon)^2+\cdot^2}\right\|_{L^{1}(I)} &\leq& \left\|f(t,\cdot)-\lambda_{\varepsilon}\psi_{\varepsilon}\right\|_{L^{1}(I)}\\ \\ & +& \left\|\lambda_{\varepsilon}\psi_{\varepsilon}(x)-\frac{\varepsilon \gamma(0)}{(\gamma(0)\pi\varepsilon)^2+\cdot^2}\right\|_{L^{1}(I)}
\end{array} .$$ Hence by Proposition \[proposition2\] and Theorem \[theorem1\], we only need to estimate the last term, for which we have $$\begin{aligned}
&\left\|\frac{\lambda_{\varepsilon}\varepsilon \gamma}{\lambda_{\varepsilon}-(1-\varepsilon)+\cdot^2}- \frac{\varepsilon \gamma(0)}{(\gamma(0)\pi\varepsilon)^2+\cdot^2}\right\|_{L^{1}(I)} \\
& \quad\leq \left\|\frac{\varepsilon (\lambda_{\varepsilon}-1)\gamma}{\lambda_{\varepsilon}-(1-\varepsilon)+ \cdot^2}\right\|_{L^{1}(I)}
+ \left\|\frac{\varepsilon \gamma}{\lambda_{\varepsilon}-(1-\varepsilon)+\cdot^2}- \frac{\varepsilon \gamma}{(\gamma(0)\pi\varepsilon)^2+\cdot^2}\right\|_{L^{1}(I)}\\
&\qquad+\left\|\frac{\varepsilon (\gamma-\gamma(0))}{(\gamma(0)\pi\varepsilon)^2+\cdot^2}\right\|_{L^{1}(I)}.\end{aligned}$$ Let us bound the three terms. For the first one we have, by Proposition \[proposition2\], $$\begin{array}{rcl}
\left\|\frac{\varepsilon (\lambda_{\varepsilon}-1)\gamma}{\lambda_{\varepsilon}-(1-\varepsilon)+\cdot^2}\right\|_{L^{1}(I)}& \leq & (\lambda_{\varepsilon}-1)\|\gamma\|_{\infty}\int_{\mathbb{R}}\frac{\varepsilon \,dx}{\frac{(\gamma(0)\pi\varepsilon)^2}{2}+x^2} \\ \\
& = & (\lambda_{\varepsilon}-1)\|\gamma\|_{\infty}\frac{\sqrt{2}}{\gamma(0)}= O(\varepsilon).
\end{array}$$ For the second one, by Proposition \[proposition2\] and , $$\begin{aligned}
&\left\|\frac{\varepsilon \gamma}{\lambda_{\varepsilon}-(1-\varepsilon)+\cdot^2}- \frac{\varepsilon \gamma}{(\gamma(0)\pi\varepsilon)^2+\cdot^2}\right\|_{L^{1}(I)} \\
&\quad \leq |(\gamma(0)\pi\varepsilon)^2-(\lambda_{\varepsilon}-(1-\varepsilon))| \varepsilon \|\gamma\|_{\infty}\int_{\mathbb{R}}\frac{\,dx}{\Big(\frac{(\gamma(0)\pi\varepsilon)^2}{2}+x^2\Big)^{2}}\\
&\quad = \left|(\gamma(0)\pi\varepsilon)^2-(\lambda_{\varepsilon}-(1-\varepsilon))\right| \frac{K_0}{\varepsilon^2}=O\left(\varepsilon \ln{\frac{1}{\varepsilon}}\right).\end{aligned}$$
For the third one, similarly to the proof of Proposition \[proposition2\], denoting by $A:= \frac{\|\gamma\|_{\infty}}{\|\gamma'\|_{\infty}}$, $$\begin{aligned}
\left\|\frac{\varepsilon (\gamma-\gamma(0))}{(\gamma(0)\pi\varepsilon)^2+\cdot^2}\right\|_{L^{1}(I)}&\leq& 2 \varepsilon \int_{0}^{A} \frac{\|\gamma'\|_{\infty}x}{(\gamma(0)\pi\varepsilon)^2+x^2}\,dx+ 2 \varepsilon\int_{A}^{+\infty} \frac{\|\gamma\|_{\infty}}{(\gamma(0)\pi\varepsilon)^2+x^2}\,dx\\
&=& \varepsilon \|\gamma'\|_{\infty} \ln{(1+\frac{A^2}{(\gamma(0)\pi\varepsilon)^2})}+2 \frac{\|\gamma\|_{\infty}}{\gamma(0)\pi} \arctan{\Big(\frac{\gamma(0)\pi \varepsilon}{A}\Big)}\\
& = & O\left(\varepsilon \ln{\frac{1}{\varepsilon}}\right).\end{aligned}$$
Proof of Lemma \[lem6\] {#app}
------------------------
Let us consider the linear initial value problem $$\left\{\begin{array}{rcl}
\frac{\partial u(t,x)}{\partial t}& = & \tilde{A}_{\varepsilon}u(t,x)=(a_{\varepsilon}(x)-\lambda_{\varepsilon})\, u(t,x) +\varepsilon \gamma(x)\,\int_{I}u(t,y)\,dy,\\
\\ u(0,x)&=&u_{0}(x),
\end{array}\right.$$ where $a_{\varepsilon}(x):=1-\varepsilon-x^2$. Let us recall that $s(\tilde{A}_{\varepsilon})=0$ and $\varepsilon\int_{I}
\frac{\gamma(x)}{\lambda_{\varepsilon}-a_{\varepsilon}(x)}\,dx=1$ (see Proposition \[proposition1\]). Applying the Laplace transform with respect to $t$ to the previous equation, we obtain the identity $$\mu \,\mathcal{L}[u](\mu, x)-u_{0}(x)=
(a_{\varepsilon}(x)-\lambda_{\varepsilon})\,\mathcal{L}[u](\mu,x)+\varepsilon\,
\gamma(x)\,\int_{I}\mathcal{L}[u](\mu,y)\, \,dy,$$ that is $$\label{eq13}
\mathcal{L}[u](\mu,x)=\frac{u_{0}(x)}{\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x)}+\frac{\varepsilon
\,\gamma(x)}{\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x)}\int_{I}\mathcal{L}[u](\mu,
y)\,dy.$$ Integrating (with respect to $x$), we obtain $$\int_{I}\mathcal{L}[u](\mu,x)\,dx=\frac{\int_{I}\frac{u_{0}(x)}{\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x)}\,dx}
{1- \int_{I}\frac{\varepsilon
\gamma(x)}{\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x)}\,dx}=\frac{\int_{I}\frac{u_{0}(x)}{\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x)}\,dx}
{\varepsilon \mu
\int_{I}\frac{\gamma(x)}{(\lambda_{\varepsilon}-a_{\varepsilon}(x))(\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x))}\,dx},$$ where we have used, for the second equality, $\varepsilon\int_{I}
\frac{\gamma(x)}{\lambda_{\varepsilon}-a_{\varepsilon}(x)}=1$. Substituting in , we get $$\label{eq14}
\mathcal{L}[u](\mu,x)=\frac{u_{0}(x)}{\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x)}+
\frac{\int_{I}\frac{u_{0}(x)}{\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x)}\,dx}
{\mu
\int_{I}\frac{\gamma(x)}{(\lambda_{\varepsilon}-a_{\varepsilon}(x))(\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x))}\,dx}\frac{\gamma(x)}{(\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x))}.$$ This Laplace transform is analytic for Re $\mu >0$ (note that $\lambda_{\varepsilon}-a_{\varepsilon}(x)$ is positive and tends to zero when $\varepsilon$ tends to zero). Then, for $s>0$, we know, by the inversion theorem, that $$u(t,x)=\frac{1}{2\pi
i}\int_{s-i\infty}^{s+i\infty}\mathcal{L}[u](\mu,x)\,e^{\mu t}\,
\,d\mu.$$ Using the theorem of residues, we can shift the integration path to the left in order to obtain, for any $s' \in
(1-\varepsilon-\lambda_{\varepsilon},0),$ $$u(t,x)=\text{Res}_{\mu=0} \Big(\mathcal{L}[u](\mu, x)e^{\mu
t}\Big)+\frac{1}{2\pi
i}\int_{s'-i\infty}^{s'+i\infty}\mathcal{L}[u](\mu,x)e^{\mu\, t}\,
\,d\mu,$$ where $$\begin{array}{rcl}
\text{Res}_{\mu=0}\Big(\mathcal{L}[u](\mu, x)e^{\mu
t}\Big)&=& \lim_{\mu \to 0} \mu \mathcal{L}[u](\mu, x)\\ \\
&=& \lim_{\mu \to 0} \left(\frac{\mu
u_{0}(x)}{\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x)}+\frac{
\int_{I}\frac{u_{0}(x)}{\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x)}\,{d}x}
{\int_{I}\frac{\gamma(x)}{(\lambda_{\varepsilon}-a_{\varepsilon}(x))(\mu+\lambda_{\varepsilon}
-a_{\varepsilon}(x))}\,{d}x}\,\frac{\gamma(x)}{\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x)}\right)\\ \\
&=& \frac{\langle u_0, \psi_{\varepsilon}^{*}\rangle}{\langle
\psi_{\varepsilon},
\psi_{\varepsilon}^{*}\rangle}\,\psi_{\varepsilon}(x)=c_{u_0}\,\psi_{\varepsilon}(x)
\end{array}$$
(let us recall that $\psi_{\varepsilon}(x)=\frac{\varepsilon
\gamma(x)}{\lambda_{\varepsilon}-a_{\varepsilon}(x)}$ and $\psi_{\varepsilon}^{*}(x)= \frac{\left(\varepsilon
\int_{I}\frac{\gamma(x)\,dx}{(\lambda_{\varepsilon}-(1-\varepsilon-x^2))^2}\right)^{-1}}{\lambda_{\varepsilon}-(1-\varepsilon-x^2)}$).
Thus, we obtain that, for $s' \in
\left(1-\varepsilon-\lambda_{\varepsilon},0\right)$, $$\label{eq14bis}
u(t,x)=c_{u_0}\psi_{\varepsilon}(x)+ \frac{1}{2\pi
}\int_{-\infty}^{+\infty}\mathcal{L}[u](s'+i\tau, x)\,
e^{(s'+i\tau)\, t}\, \,d\tau .$$ We now define $g_{\varepsilon}(\mu):=\frac{\int_{I}\frac{u_{0}(x)\,dx}{\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x)}}
{\mu
\int_{I}\frac{\gamma(x)\,dx}{(\lambda_{\varepsilon}-a_{\varepsilon}(x))(\mu+\lambda_{\varepsilon}-a_{\varepsilon}(x))}},
$ so that we can write $$\label{eq14bisbis}
\begin{array}{rcl}
\frac{1}{2\pi
}\int_{-\infty}^{+\infty}\mathcal{L}[u](s'+i\tau,x)e^{(s'+i\tau) t}
\,d\tau&=& \frac{1}{2\pi}
u_0(x)e^{s't}\int_{-\infty}^{\infty}\frac{e^{i\tau
t}}{s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)+i
\tau}\,d\tau\\ \\ &+&
\frac{1}{2\pi} \gamma(x)e^{s't}\int_{-\infty}^{\infty}\frac{g_{\varepsilon}(s'+i \tau)e^{i\tau t}}{s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)+i \tau}\,d\tau\\ \\
&=&e^{-(\lambda_{\varepsilon}-a_{\varepsilon}(x))t}u_{0}(x)\\ \\
&+&\frac{1}{2\pi}
\gamma(x)e^{s't}\int_{-\infty}^{\infty}\frac{g_{\varepsilon}(s'+i
\tau)e^{i\tau t}}{s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)+i
\tau}\,d\tau,
\end{array}$$ where we used the estimate $s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)>0$ and the identity $\int_{-\infty}^{\infty}\frac{e^{i \tau t}}{\alpha + i
\tau}\,d\tau=2 \pi e^{-\alpha t}$ (for $\alpha >0$).
We now would like to find a bound for $\left\|\frac{1}{2\pi}
\gamma(x)e^{s't}\int_{-\infty}^{\infty}\frac{g_{\varepsilon}(s'+i
\tau)e^{i\tau t}}{s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)+i
\tau}\,d\tau\right\|_{\infty}$.\
We see that $$\label{eq15}
\begin{array}{rcl}
\left\|
\gamma(x)\,e^{s't}\int_{-\infty}^{\infty}\frac{g_{\varepsilon}(s'+i
\tau)e^{i\tau t}}{s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)+i
\tau}\,d\tau\right\|_{\infty}&\leq& e^{s't}
\|\gamma\|_{\infty}\sup_{x}\Big|\int_{-\infty}^{\infty}\frac{g_{\varepsilon}(s'+i
\tau)e^{i\tau t}}{s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)+i
\tau}\,d\tau \Big|\\ \\ &=& e^{s't}
\sup_{x}\Big|\int_{-\infty}^{\infty}\frac{g_{\varepsilon}(s'+i
\tau)e^{i\tau t}}{s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)+i
\tau}\,d\tau \Big|
\end{array}$$ and $$\label{eq16}
\begin{array}{rcl}
\Big|\int_{-\infty}^{\infty}\frac{g_{\varepsilon}(s'+i \tau)e^{i\tau
t}}{s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)+i \tau}\,d\tau
\Big| &\leq & \int_{-\infty}^{\infty}\frac{|g_{\varepsilon}(s'+i
\tau)|}{|s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)+i
\tau|}\,d\tau\\ \\ & \leq &
\int_{-\infty}^{\infty}\frac{|g_{\varepsilon}(s'+i
\tau)|}{|s'+\lambda_{\varepsilon}-(1-\varepsilon)+i
\tau|}\,d\tau
\end{array}$$ since $|s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)+i \tau| \geq
|s'+\lambda_{\varepsilon}-(1-\varepsilon)+i \tau|$.
Let us then find an upper bound for $g_{\varepsilon}(s'+i\tau)$. For the numerator of $g_{\varepsilon}(s'+i\tau)$ we can estimate $$\left|\int_{I}\frac{u_{0}(x)}{s'+i\tau+\lambda_{\varepsilon}-a_{\varepsilon}(x)}\,dx\right|
\leq \frac{\|u_{0}\|_{1}}{|s'+i \tau +
\lambda_{\varepsilon}-(1-\varepsilon)|}.$$ We now find a lower bound for the denominator of $g_{\varepsilon}(s'+i\tau)$. We use the elementary estimate $|z|
\geq \max (|\text{Re}z|, |\text{Im}z|) $ and we start with the real part. $$\begin{array}{rcl}
\Big|\text{Re}\int_{I}\frac{\gamma(x)}{(\lambda_{\varepsilon}-a_{\varepsilon}(x))(s'+i\tau+\lambda_{\varepsilon}-a_{\varepsilon}(x))}\,dx\Big|&=&
\Big|\int_{I}\frac{\gamma(x)}{(\lambda_{\varepsilon}-a_{\varepsilon}(x))}\frac{s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)}{|s'+i\tau+\lambda_{\varepsilon}-a_{\varepsilon}(x)|^2}\,dx\Big| \\ \\&=&
\int_{I}\frac{\gamma(x)}{(\lambda_{\varepsilon}-a_{\varepsilon}(x))}\frac{s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)}{|s'+i\tau+\lambda_{\varepsilon}-a_{\varepsilon}(x)|^2}\,dx
\\ \\&=&
\int_{I}\frac{\gamma(x)}{(\lambda_{\varepsilon}-a_{\varepsilon}(x))\big(s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)+\frac{\tau^2}{s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)}\big)}\,dx\\ \\
& \geq &
\int_{I}\frac{\gamma(x)}{(\lambda_{\varepsilon_{0}}-(1-{\varepsilon_{0}})+x^2)\big(\lambda_{\varepsilon_{0}}-(1-{\varepsilon_{0}})+x^2+\frac{\tau^2}{x^2}\big)}\,dx
\\ & = &
\int_{I}\frac{x^2\gamma(x)}{(\lambda_{\varepsilon_{0}}-(1-{\varepsilon_{0}})+x^2)((\lambda_{\varepsilon_{0}}-(1-{\varepsilon_{0}})+x^2)x^2+\tau^2)}\,dx\\
\\ & =:&F(\tau),
\end{array}$$ where in the last inequality we used the estimates $s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)<
\lambda_{\varepsilon}-a_{\varepsilon}(x)$, $s'+\lambda_{\varepsilon}-(1-\varepsilon)>0$. We also used that, since $\lambda_{\varepsilon}-(1-\varepsilon)$ is strictly positive and tends to zero when $\varepsilon$ goes to zero, there exists $\varepsilon_0$ such that $\forall \varepsilon < \varepsilon_0$ we have $\lambda_{\varepsilon_{0}}-(1-{\varepsilon_{0}})> \lambda_{\varepsilon}-(1-{\varepsilon})$.\
In a similar way, for the imaginary part, $$\begin{array}{rcl}
\Big|\text{Im}\int_{I}\frac{\gamma(x)}{(\lambda_{\varepsilon}-a_{\varepsilon}(x))(s'+i\tau+\lambda_{\varepsilon}-a_{\varepsilon}(x))}\,dx\Big|&=&
\Big|\int_{I}\frac{\gamma(x)}{(\lambda_{\varepsilon}-a_{\varepsilon}(x))}\frac{-\tau}{(s'+\lambda_{\varepsilon}-a_{\varepsilon}(x))^2+
\tau^2}\,dx\Big|\\ \\ &=& | \tau |
\int_{I}\frac{\gamma(x)}{(\lambda_{\varepsilon}-a_{\varepsilon}(x))\big((s'+\lambda_{\varepsilon}-a_{\varepsilon}(x))^2+
\tau^2\big)}\,dx \\ \\ & \geq & | \tau |
\int_{I}\frac{\gamma(x)}{(\lambda_{\varepsilon_0}-(1-\varepsilon_0)+x^2)\big((\lambda_{\varepsilon_0}-(1-\varepsilon_0)+x^2)^2+
\tau^2\big)}\,dx \\ \\ &=:& G(\tau).
\end{array}$$ Defining $H(\tau):=\max(F(\tau),G(\tau))$ we see that $$\label{eq17}
|g_{\varepsilon}(s'+i \tau)| \leq \frac{\frac{\|u_0\|_{1}}{|s'+i \tau
+\lambda_{\varepsilon}-(1-\varepsilon)|}}{|s'+i \tau| H(\tau)},$$ and then, using , and $$\begin{aligned}
&\left|\left|
\gamma(x)e^{s't}\int_{-\infty}^{+\infty}\frac{g_{\varepsilon}(s'+i
\tau)e^{i\tau t}}{s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)+i
\tau}\,d\tau\right|\right|_{\infty} \\
&\quad \leq e^{s't}\int_{-
\infty}^{+\infty} \frac{\,d\tau}{\sqrt{s'^{2}+ \tau^{2}}|s'+i
\tau + \lambda_{\varepsilon}-(1-\varepsilon)|^2 H(\tau)}\|u_0\|_{1}.\end{aligned}$$
Now, since $F$ and $G$ are strictly positive continuous functions, $F(0)>0$ and $\tau G(\tau)$ tends to a positive limit when $\tau$ goes to $\infty$, there exists a constant $C>0$ (independent of $\varepsilon$) such that $H(\tau) \geq \frac{C}{1+\tau}$. Choosing $s'= -\frac{\alpha_{\varepsilon}}{2}$, where $\alpha_{\varepsilon} =
\lambda_{\varepsilon}-(1-\varepsilon),$ we can write
$$\begin{aligned}
\|\gamma(x)\,e^{s't}\,\int_{-\infty}^{+\infty}\frac{g_{\varepsilon}(s'+i
\tau)e^{i\tau t}}{s'+\lambda_{\varepsilon}-a_{\varepsilon}(x)+i
\tau}\,d\tau\|_{\infty} &\leq& \frac{e^{-
\alpha_{\varepsilon}t}}{C}\int_{0}^{+\infty}\frac{2(1+\tau)}{\big((\frac{\alpha_{\varepsilon}}{2})^2+\tau^2\big)^\frac{3}{2}}\,d\tau\|u_{0}\|_{1}\\
&=&
\frac{e^{-\frac{\alpha_{\varepsilon}t}{2}}}{C}\left(\frac{8}{\alpha_{\varepsilon}^2}+\frac{4}{\alpha_{\varepsilon}}\right)
\|u_{0}\|_1.\end{aligned}$$
Finally, going back to and using , we end up with $$\|u((, \cdot)-c_{u_{0}}\psi_{\varepsilon}\| \leq
\left(1+\frac{1}{\pi
C}\left(\frac{4}{\alpha_{\varepsilon}^2}+\frac{2}{\alpha_{\varepsilon}}\right)
\right)\, e^{-\frac{\alpha_{\varepsilon}t}{2}}\,\|u_{0}\|_1\leq
K_5\,\varepsilon^{-4}e^{-\frac{\alpha_{\varepsilon}t}{2}}\,\|u_{0}\|_1.$$
Proof of Theorem \[thm:smallt\]
===============================
We start here the proof of Theorem \[thm:smallt\]. From now on, $C$ will designate a strictly positive constant depending only on some upper bounds on $\|\gamma\|_{L^\infty}$, $\|f(0,\cdot)\|_{W^{1,\infty}}$, a lower bound on $f(0,0)$ (see Remark \[rem:C\]), and on $|b-a|$.
Thanks to the variation of the constant formula, the solution $f$ of (\[eqq0\]) satisfies: $$\begin{aligned}
f(t,x)&=&f(0,x)\, e^{(1-\varepsilon-x^2)\,t-\int_0^t\int_I f(s,y)\,dy\,ds}\nonumber\\
&&+\, \varepsilon \int_0^t\left(\gamma(x)\int_I f(s,y)\,dy\right)e^{(1-\varepsilon-x^2)(t-s)-\int_s^t\int_I f(\sigma,y)\,dy\,d\sigma}\,ds\nonumber\\
&=&f(0,x)\, e^{(1-\varepsilon-x^2)\,t-\int_0^t{{\mathcal I}}(s)\,ds}\nonumber\\
&&+\,\varepsilon \int_0^t{{\left(\gamma(x)\, \mathcal I(s) \right)}} e^{(1-\varepsilon-x^2)(t-s)-\int_s^t {\mathcal I}(\sigma)\,d\sigma}\,ds,\label{eq:varconst}\end{aligned}$$ where $$\mathcal I(t):=\int_I f(t,y)\,dy.$$ Obtaining a precise estimate on $t\mapsto e^{(1-\varepsilon)(t-s)-\int_s^t\mathcal I(\sigma)\,d\sigma}$ is the key to prove Theorem \[thm:smallt\].
Preliminary estimates
---------------------
If we sum along $x\in\mathbb R$, we get, for $t\geq0$: $$\begin{aligned}
\mathcal I(t)&=&\left(\int_I f(0,x)\, e^{-x^2t}\,dx\right) e^{(1-\varepsilon)t-\int_0^t\mathcal I(s)\,ds}\nonumber\\
&&+\,\varepsilon \int_0^t\left(\int_I\int_I\gamma(x) f(s,y)e^{-x^2(t-s)}\,dx\,dy\right)e^{(1-\varepsilon)(t-s)-\int_s^t\mathcal I(\sigma)\,d\sigma}\,ds\nonumber\\
&=&\frac{z_1(t)}{\sqrt{t}} e^{(1-\varepsilon)t-\int_0^t\mathcal I(s)\,ds}+\varepsilon \int_0^t\frac{z_2(s,t-s)}{\sqrt{t-s}}e^{(1-\varepsilon)(t-s)-\int_s^t\mathcal I(\sigma)\,d\sigma}\,ds,\label{eq:varconst2}\end{aligned}$$ where $$z_1(t):=\sqrt{t} \int_I f(0,x)\,e^{-x^2t}\,dx,\quad z_2(\sigma,\tau)=\sqrt{\tau}\int_I\int_I \gamma(x) \,f(\sigma,y)\, e^{-x^2\tau}\,dx\,dy.$$
If we differentiate $\mathcal I$ with respect to $t$, we get $$\begin{aligned}
\frac{\partial \mathcal I}{\partial t}(t)&=&\mathcal I(t)\left(1-\varepsilon-\mathcal I(t)\right)-\int_I x^2f(t,x)\,dx+\varepsilon\int_I\int_I \gamma(x)f(t,y)\,dx\,dy\\
&\leq&\mathcal I(t)\left(1-\varepsilon -\mathcal I(t)\right)+\varepsilon\, \mathcal I(t)\\
&\leq&\mathcal I(t)\left(1-\mathcal I(t)\right),\end{aligned}$$ which implies, since $\mathcal I(0)\leq 1$, that $$\label{alphabound}
0\leq \mathcal I(t)\leq 1.$$ Thanks to , and the nonnegativity of $z_1,\,z_2$, one gets $$\label{estz1}
\frac{z_1(t)}{\sqrt{t}} e^{(1-\varepsilon)t-\int_0^t\mathcal I(s)\,ds}\leq C,$$ while for some constants $C,\,C'>0$, $$\begin{aligned}
z_1(t)&=&\int_I f\left(0,\frac x {\sqrt{t}}\right)e^{-x^2}\,dx\\
&\geq&\frac 1C\int_{-C'}^{C'}f\left(0,\frac x {\sqrt{t}}\right)\,dx\geq \frac 1{C},\end{aligned}$$ for $t\geq 1$. Note that here we used a lower bound on $f(0,\cdot)$ around $x=0$ ([[we have assumed]{}]{} that $f(0,0)>0$ and that $f(0,\cdot)$ is continuous). [[Thanks to this lower bound,]{}]{} becomes $$\label{estexpalpha}
e^{(1-\varepsilon)t-\int_0^t\mathcal I(s)\,ds}\leq C\,\sqrt t .$$ Thanks to and , we can estimate the second term of as follows: $$\begin{aligned}
w(t)&:=&\varepsilon \int_0^t\frac{z_2(s,t-s)}{\sqrt{t-s}}{{e^{(1-\varepsilon)\,t-\int_0^t\mathcal I(\sigma)\,d\sigma}e^{\varepsilon s+\int_0^s\left(\mathcal I(\sigma)-1\right)\,d\sigma}}}\,ds\nonumber\\
&\leq&C\,\varepsilon\,\sqrt t\, \|z_2\|_{L^\infty(I)}
\int_0^t\frac{e^{C\varepsilon s}}{\sqrt{t-s}}\,ds\nonumber\\
&\leq&C\,\varepsilon\,\sqrt t\,\|z_2\|_{L^\infty(I)}\, e^{C\varepsilon t} \int_0^t\frac{e^{{{-}}C\varepsilon s}}{\sqrt{s}}\,ds\leq C\, \varepsilon t\, \|z_2\|_{L^\infty(I)}\, e^{C\varepsilon t}.\label{def:w}\end{aligned}$$ In order to estimate $\|z_2\|_{L^\infty(I)}$, we proceed as follows: $$\begin{aligned}
z_2(s,\tau)&=&\sqrt{\tau}\int_I\int_I \gamma\left(\frac x{\sqrt \tau}\right)\, f(s,y)\,e^{-x^2\tau}\,dx\,dy\\
&\leq&\frac {C\,\mathcal I(s)}{\sqrt \tau}\,\int_I e^{-x^2\,\tau}\,dx \leq C.\end{aligned}$$ This estimate combined with implies that $w(t)\geq 0$ satisfies $$\label{eq:est4}
{{w(t)\leq C\,\varepsilon \,t \,e^{C\,\varepsilon\, t},}}$$ Since $f(0,\cdot)\in W^{1,\infty}(I)$, we can estimate $$\begin{aligned}
z_1(t)&=&\int_I \left(f(0,0)+\int_0^{\frac x{\sqrt t}}\frac{\partial f}{\partial x}\left(0,z\right)\,dz\right)
\,e^{-x^2}\,dx\nonumber\\
&=&f(0,0)\, \int_I e^{-x^2}\,dx +\lambda(t),\label{def:lambda}\end{aligned}$$ where $$\begin{aligned}
|\lambda(t)|&\leq& \int_I \bigg|\int_0^{\frac x{\sqrt t}}\frac{\partial f}{\partial x}(0,z)\,dz\bigg|e^{-x^2}\,dx\leq\frac{C}{\sqrt t}\int_I |x|e^{-x^2}\,dx\nonumber\\
&\leq&\frac C{\sqrt t}.\label{estlambda}\end{aligned}$$
Estimation of $e^{(1-\varepsilon) t-\int_0^t\mathcal I(s)\,ds}$
---------------------------------------------------------------
Thanks to (and the definition of $\lambda$ and $w$: see and respectively), we see that $$\label{eq:alpha}
\mathcal I(t)=\frac{f(0,0)\, \int_I e^{-x^2}\,dx +\lambda(t)}{\sqrt t}e^{(1-\varepsilon)t-\int_0^t\mathcal I(s)\,ds}+w(t),$$ so that $$\begin{aligned}
e^{\int_0^t\mathcal I(s)\,ds}&=&e^{\int_0^1\mathcal I(s)\,ds}+\int_1^t\frac{d}{ds}\left(e^{\int_0^s\mathcal I(\sigma)\,d\sigma}\right)(s)\,ds\nonumber\\
&=&e^{\int_0^1\mathcal I(s)\,ds}+\int_1^t\frac{f(0,0)\, \int_I e^{-x^2}\,dx }{\sqrt s}e^{(1-\varepsilon)s}\,ds\nonumber\\
&&+\int_1^t\frac{\lambda(s)}{\sqrt s}e^{(1-\varepsilon)s}\,ds+\int_1^tw(s)e^{\int_0^s\mathcal I(\sigma)\,d\sigma}\,ds .
\label{eq:intalpha}\end{aligned}$$ We will now estimate each of the terms on the right hand side of . We start by estimating the third term on the right hand side, thanks to and an integration by parts: $$\begin{aligned}
\left|\int_1^t\frac{\lambda(s)}{\sqrt s}e^{(1-\varepsilon)s}\,ds\right|&\leq& C\int_1^t\frac{e^{(1-\varepsilon)s}}s\,ds\nonumber\\
&\leq& C\left[\frac{e^{(1-\varepsilon)t}}{(1-\varepsilon)t}+\int_1^t\frac{e^{(1-\varepsilon)s}}{(1-\varepsilon)s^2}\,ds\right]\nonumber\\
&\leq& \frac{C}{1-\varepsilon}\left[\frac{e^{(1-\varepsilon)t}}{t}+t\max_{s\in [1,t]}\frac{e^{(1-\varepsilon)s}}{s^2}\right]\nonumber\\
&\leq& \frac{2C}{(1-\varepsilon)t}e^{(1-\varepsilon)t},\label{eq:esttruc}\end{aligned}$$ provided $t>0$ is large enough, and $\varepsilon>0$ is small enough (to ensure that $\max_{s\in [1,t]}\frac{e^{(1-\varepsilon)s}}{s^2}=\frac{e^{(1-\varepsilon)t}}{t^2}$).
We now estimate the second term on the right hand side of , using an integration by parts: $$\begin{aligned}
&\int_1^t\frac{f(0,0)\,\int_I e^{-x^2}\,dx }{\sqrt{s}}e^{(1-\varepsilon)s}\,ds \\
&\quad=f(0,0)\,{{\left(\int_I e^{-x^2}\,dx\right)}}\,\left(\frac{e^{(1-\varepsilon)t}}{(1-\varepsilon)\sqrt{t}}-\frac {e^{1-\varepsilon}}{1-\varepsilon}+\int_1^t\frac{e^{(1-\varepsilon)s}}{2(1-\varepsilon)s^{3/2}}\,ds\right),\end{aligned}$$ and then, applying an estimate similar to the one used to obtain , we get, provided that $t>0$ is large enough, and that $\varepsilon>0$ is small enough, $$\label{eq:est1st-term}
0\leq \int_1^t\frac{e^{(1-\varepsilon)s}}{2\,(1-\varepsilon)s^{3/2}}\,ds\leq \int_1^t\frac{e^{(1-\varepsilon)\,s}}{2\,(1-\varepsilon)\,s}\,ds
\leq\frac{1}{(1-\varepsilon)^2\,t}e^{(1-\varepsilon)\,t}.$$ Finally, we estimate the last term of the right hand side of , thanks to estimates and : $$\begin{aligned}
0\leq \int_1^tw(s)e^{\int_0^s\mathcal I(\sigma)\,d\sigma}\,ds&\leq& \int_1^t {{|w(s)|}} e^{\|\mathcal I\|_{L^\infty(\mathbb R_+)} s}\,ds\nonumber\\
&\leq&C\,\varepsilon\int_1^t s\, e^{C\varepsilon s}\,e^{s}\,ds\nonumber\\
&\leq& C\,\varepsilon\frac{t\,e^{\left(1+C\,\varepsilon\right)\,t}}{1+C\,\varepsilon},\label{eq:est3rd-term}\end{aligned}$$ where we have used an integration by part to obtain the last inequality.
Combining these estimates, estimate becomes: $$\label{eq:expalpha1}
e^{\int_0^t\mathcal I(s)\,ds-(1-\varepsilon)t}=\frac{f(0,0)\,\int_I e^{-x^2}\,dx}{(1-\varepsilon)\sqrt t}+\mu(t),$$ or $$\label{eq:expalpha2}
e^{(1-\varepsilon)t-\int_0^t\mathcal I(s)\,ds}=\left(\frac{f(0,0)\, \int_I e^{-x^2}\,dx}{(1-\varepsilon)\sqrt t}+\mu(t)\right)^{-1},$$ where, thanks to , , and , for $t\geq 1$, $$\label{eq:estmu}
-\frac C t\leq\mu(t)\leq C\left(\frac 1t+\varepsilon t e^{C\varepsilon t}\right).$$
[[Estimation of]{}]{} $\left|e^{(1-\varepsilon)t-\int_0^t\mathcal I(s)\,ds}-\frac{\sqrt t}{f(0,0)\, \int_I e^{-x^2}\,dx}\right|$
--------------------------------------------------------------------------------------------------------------------------------
Thanks to , $$\begin{aligned}
&\left|e^{(1-\varepsilon)t-\int_0^t\mathcal I(s)\,ds}-\frac{\sqrt t}{f(0,0)\,\int_I e^{-x^2}\,dx}\right|\nonumber\\
&\quad =\left|\left(\frac{f(0,0)\, \int_I e^{-x^2}\,dx}{(1-\varepsilon)\sqrt t}+\mu(t)\right)^{-1}-\frac{\sqrt t}{f(0,0)\, \int_I e^{-x^2}\,dx}\right|\\
&\quad=\frac{\sqrt t}{f(0,0)\, \int_I e^{-x^2}\,dx}\left|\left(\frac 1{1-\varepsilon}+\frac{\mu(t)\,\sqrt t}{f(0,0)\, \int_I e^{-x^2}\,dx}\right)^{-1}-1\right|.\label{eq:expalpha3}\end{aligned}$$
We notice that thanks to estimate , $$\label{eq:est1}
f(0,0)\,\int_I e^{-x^2}\,dx+(1-\varepsilon)\mu(t) \sqrt t\geq \frac {f(0,0)\,\int_I e^{-x^2}\,dx}2,$$ as soon as $t\geq T$, for some large time $T>0$. Also, for $t\geq 1$, $$\label{eq:est2}
|\mu(t)|\leq C\left(\frac 1t+\varepsilon \,t\,e^{C\,\varepsilon\, t}\right).$$ Using the bounds and , we can show that as soon as $t\geq T$, $$\begin{aligned}
\left|\left(\frac 1{1-\varepsilon}+\frac{\mu(t)\sqrt t}{f(0,0)\,\int_I e^{-x^2}\,dx}\right)^{-1}-1\right|
&=&\left|\frac{-\varepsilon f(0,0)\,\int_I e^{-x^2}\,dx-(1-\varepsilon)\mu(t) \sqrt t}{f(0,0)\, \int_I e^{-x^2}\,dx+(1-\varepsilon)\mu(t) \sqrt t}\right|\nonumber\\
&\leq&C\left(\frac 1{\sqrt t}+\varepsilon t^{\frac 32}\,e^{C\,\varepsilon \, t}\right),\label{eq:est3}\end{aligned}$$ so that identity leads to the bound $$\left|e^{(1-\varepsilon)t-\int_0^t\mathcal I(s)\,ds}-\frac{\sqrt t}{f(0,0)\,\int_I e^{-x^2}\,dx}\right|\leq C\left(1+\varepsilon \, t^{2}\, e^{C\,\varepsilon\, t}\right). \label{eq:est6}$$ Notice also, as this is going to be useful further on, that for $s\geq 1$, thanks to and , $$\begin{aligned}
\left|e^{\int_0^s\mathcal I(\sigma)\,d\sigma-(1-\varepsilon)s}-\frac{f(0,0)\, \int_I e^{-x^2}\,dx}{\sqrt s}\right|&=&\left|\mu(s)+\varepsilon\frac{f(0,0)\,\int_I e^{-x^2}\,dx}{(1-\varepsilon)\sqrt s}\right|\nonumber\\
&\leq& C\left(\frac 1{s}+\varepsilon \,s\,e^{C\,\varepsilon\, s}\right).\label{eq:est5}\end{aligned}$$
Conclusion of the proof of Theorem \[thm:smallt\]
-------------------------------------------------
In this last part of the proof, we consider times $t\geq T$. We estimate $$\begin{aligned}
&\left\|f(t,x)-\frac{f(0,x)\,\sqrt t \,e^{-x^2 t}}{f(0,0)\,\int_I e^{-x^2}\,dx}\right\|_{L^1(I)}\nonumber\\
&\quad\leq \left\|f(t,x)-\frac{f(0,x)\,e^{-x^2 t}}{\frac{f(0,0)\,\int_I e^{-x^2}\,dx}{(1-\varepsilon)\, \sqrt t}+\mu(t)}\right\|_{L^1(I)}\nonumber\\
&\qquad +\, \Bigg\|\frac{f(0,x)\,e^{-x^2 t}}{\frac{f(0,0)\,\int_I e^{-x^2}\,dx}{(1-\varepsilon)\,\sqrt t}+\mu(t)}-\frac{f(0,x)\,\sqrt t\, e^{-x^2 t}}{f(0,0)\,\sqrt \pi}\Bigg\|_{L^1(I)}.\label{eq:diff}\end{aligned}$$ Let us start by estimating the second term on the right hand side of , thanks to estimate : $$\begin{aligned}
&\Bigg\|\frac{f(0,x)e^{-x^2 t}}{\frac{f(0,0)\,\int_I e^{-x^2}\,dx}{(1-\varepsilon)\sqrt t}+\mu(t)}-\frac{f(0,x)\sqrt t e^{-x^2 t}}{f(0,0)\,\int_I e^{-x^2}\,dx}\Bigg\|_{L^1(I)}\nonumber\\
&\quad\leq \left\|\frac{f(0,x)\sqrt t e^{-x^2 t}}{f(0,0)\,\int_I e^{-x^2}\,dx}\left|\left(\frac 1 {1-\varepsilon}+\frac{\mu(t)\sqrt t}{f(0,0)\,\int_I e^{-x^2}\,dx}\right)^{-1}-1\right|\;\right\|_{L^1(I)}\nonumber\\
&\quad\leq \frac{f(0,x)\sqrt t }{f(0,0)\,\int_I e^{-x^2}\,dx}\left|\left(\frac 1 {1-\varepsilon}+\frac{\mu(t)\sqrt t}{f(0,0)\,\int_I e^{-x^2}\,dx}\right)^{-1}-1\right|\int_I e^{-x^2 t}\,dx\nonumber\\
&\quad \leq C \left(\frac 1{\sqrt t}+\varepsilon t^{\frac 32}\,e^{C\,\varepsilon \,t}\right)\label{eq:estprofil1}\end{aligned}$$ We now rewrite the first term on the right hand side of , using formula and : $$\begin{aligned}
&\left\|f(t,x)-\frac{f(0,x)e^{-x^2 t}}{\frac{f(0,0)\, \int_I e^{-x^2}\,dx}{(1-\varepsilon)\sqrt t}+\mu(t)}\right\|_{L^1(I)}\\
&\quad= \left\|\varepsilon \int_0^t\left(\int_I \gamma(x)f(s,y)\,dy\right)e^{(1-\varepsilon-x^2)(t-s)-\int_s^t\mathcal I(\sigma)\,d\sigma}\,ds\right\|_{L^1}\\
&\quad\leq C\varepsilon \int_I\int_0^t\mathcal I(s)e^{-x^2(t-s)}e^{(1-\varepsilon)t-\int_0^t\mathcal I(\sigma)\,d\sigma} e^{\int_0^s\mathcal I(\sigma)\,d\sigma-(1-\varepsilon)s}\,ds\,dx\end{aligned}$$ [[and then, thanks to , and , $$\begin{aligned}
&\left\|x \mapsto f(t,x)-\frac{f(0,x)e^{-x^2 t}}{\frac{f(0,0)\,\int_I e^{-x^2}\,dx}{(1-\varepsilon)\sqrt t}+\mu(t)}\right\|_{L^1(I)}\nonumber\\
&\quad \leq C\,\varepsilon \int_0^1\left(\int_I e^{-x^2(t-s)}\,dx\right) \left(\frac{\sqrt t}{f(0,0)\,\int_I e^{-x^2}\,dx}+1+\varepsilon\, t^{2}\, e^{C\,\varepsilon\, t}\right)\,ds\nonumber\\
&\qquad +C\varepsilon \int_1^t\left(\int_I e^{-x^2(t-s)}\,dx\right) \left(\frac{f(0,0)\, \int_I e^{-x^2}\,dx}{\sqrt s}+\frac 1{ s}+\varepsilon\, s \,
e^{C\,\varepsilon\, s}\right)\nonumber\\
&\phantom{dqsfesrgqdreg}\,\left(\frac{\sqrt t}{f(0,0)\,\int_I e^{-x^2}\,dx}+1+\varepsilon \,t^2\, e^{C\,\varepsilon \,t}\right)\,ds\nonumber\\
&\quad\leq C\,\varepsilon \,\frac 1{\sqrt t}\left(\sqrt t+1+\varepsilon\, t^{2}\, e^{C\,\varepsilon \,t}\right)\nonumber\\
&\qquad+C\varepsilon \int_1^t\frac 1{\sqrt {t-s}} \left(\frac 1{ \sqrt s}+\varepsilon\, s\, e^{C\,\varepsilon\, s}\right)\,\left(\sqrt t+1+\varepsilon \,t^2\, e^{C\,\varepsilon\, t}\right)\,ds.\nonumber\end{aligned}$$ We estimate $$\int_1^t\frac{s e^{C\varepsilon s}}{\sqrt{t-s}}\,ds\leq t e^{C\varepsilon t}\int_1^t\frac{ds}{\sqrt{t-s}}\leq Ct^{\frac 32} e^{C\varepsilon t},$$ and then $$\begin{aligned}
&\left\| x \mapsto f(t,x)-\frac{f(0,x)e^{-x^2 t}}{\frac{f(0,0)\,\int_I e^{-x^2}\,dx}{(1-\varepsilon)\sqrt t}+\mu(t)}\right\|_{L^1(I)}\nonumber\\
&\quad\leq C\varepsilon \left(1+\frac 1{\sqrt t}+\varepsilon\, t^{\frac 3 2}\, e^{C\,\varepsilon\, t}\right)\nonumber\\
&\qquad+C\,\varepsilon \,\left(1+\varepsilon t^{\frac 32} e^{C\varepsilon t}\right)\, \left(\sqrt t+1+\varepsilon\, t^2 \,e^{C\,\varepsilon\, t}\right)\nonumber\\
&\quad\leq C\left(\varepsilon+\varepsilon\sqrt t+\frac \varepsilon{\sqrt t}+ \left(\varepsilon t^{\frac 32}+\varepsilon^2 t^2+\varepsilon^2t^{\frac 32}+\varepsilon^3t^{\frac 72}\right)e^{C\varepsilon t}\right)\nonumber\\
&\quad\leq C\left(\frac \varepsilon{\sqrt t}+ \varepsilon t^{\frac 32}e^{C\varepsilon t}\right),\label{eq:estprofil2}\end{aligned}$$ where we have used the fact that $\varepsilon t\leq Ce^{C\varepsilon t}$. Thanks to and , becomes: $$\Bigg\|x \mapsto f(t,x)-\frac{f(0,x)\sqrt t e^{-x^2 t}}{f(0,0)\,\int_I e^{-y^2}\,dy}\Bigg\|_{L^1(I)}\leq C \left(\frac 1{\sqrt t}+\varepsilon\, t^{\frac 3 2}\,e^{C\,\varepsilon\, t}\right).$$ Theorem \[thm:smallt\] follows from this estimate.]{}]{}
Proof of Corollary \[cor:smallt\]
---------------------------------
If we assume that $t\in\left[\frac 1{\kappa^2}, \kappa^{\frac 23} \varepsilon^{-\frac 23}\right]$, then becomes
$$\begin{aligned}
\Bigg\| x \mapsto f(t,x)-\frac{f(0,x)\sqrt t e^{-x^2 t}}{f(0,0)\, \int_I e^{-y^2}\,dy}\Bigg\|_{L^1(I)}&\leq& C \left(\kappa+\kappa\,e^{C\, \kappa^{\frac 23} \varepsilon^{\frac 13}}\right),\end{aligned}$$
and if furthermore $\varepsilon\leq \kappa\leq 1$, then $$\begin{aligned}
\Bigg\|x \mapsto f(t,x)-\frac{f(0,x)\,\sqrt t\, e^{-x^2 t}}{f(0,0)\, \int_I e^{-y^2}\,dy}\Bigg\|_{L^1(I)}&\leq& C\,\kappa,\end{aligned}$$ which proves Corollary \[cor:smallt\][[, provided that $\kappa>0$ is small enough.]{}]{}
[**[Acknowledgement]{}**]{}: The research leading to this paper was funded by the French “ANR blanche” project Kibord: ANR-13-BS01-0004, and “ANR JCJC” project MODEVOL ANR-13-JS01-0009. A.C. and S.C. were partially supported by the Ministerio de Ciencia e Innovación, Grants MTM2011-27739-C04-02 and MTM2014-52402-C3-2-P.
[99]{}
M. Alfaro, R. Carles, Explicit solutions for replicator-mutator equations: extinction vs. acceleration. *SIAM J. Appl. Math.* [**74**]{}, 1919–1934 (2014).
W. Arendt, A, Grabosch, G. Greiner, I. Groh, H.P. Lotz, U. Moustakas, R. Nagel, F. Neubrander, U. Schlotterbeck. One-parameter semigroups of positive operators. Lecture Notes in Mathematics, 1184. Springer-Verlag, Berlin, 1986.
G. Bell, A. Gonzalez, Evolutionary rescue can prevent extinction following environmental change. *Ecol. Lett.* [**12**]{}, 942–948 (2009).
D. I. Bolnicke, P. Amarasekare, M. S. Araújo, R. Bürger, J. M. Levine, M. Novak, V. H. W. Rudolf, S. J. Schreiber, M. C. Urban, D. A. Vasseur, Why intraspecific trait variation matters in community ecology. *Trends Ecol. Evol.* 26, 183–192 (2011).
M. G. Bulmer, The mathematical theory of quantitative genetics. Oxford, UK: Clarendon Press (1980).
R. B" urger, I. M. Bomze, Stationary distributions under mutation-selection balance: Structure and properties. *Adv. in Appl. Probab.*, [**28**]{}, 227–251 (1996).
R. B" urger, The mathematical theory of selection, recombination, and mutation. Wiley Series in Mathematical and Computational Biology. John Wiley & Sons, Ltd., Chichester, 2000.
A. Calsina, S. Cuadrado, L. Desvillettes, G. Raoul, Asymptotics of steady states of a selection mutation equation for small mutation rate. *P. Roy. Soc. Edinb. A* [**143**]{}(06), 1123–1146 (2013).
R. Cant' on, M. I. Morosini, Emergence and spread of antibiotic resistance following exposure to antibiotics. *FEMS microbiol. Rev.* [**35**]{}(5), 977–991 (2011).
S. M. Carlson, C. J. Cunningham, P. A. Westley, Evolutionary rescue in a changing world. *Trends Ecol. Evol.* [**29**]{}(9), 521–530 (2014).
L. M. Chevin, R. Gallet, R., Gomulkiewicz, R. D. Holt, S. Fellous, Phenotypic plasticity in evolutionary rescue experiments. Phil. Trans. R. Soc. B: Biological Sciences, [**368**]{}(1610), 20120089 (2013).
J. Coville, Convergence to the equilibrium of positive solution of some mutation-selection model. preprint arXiv:1308.6471.
N. Champagnat, R. Ferrière, S. Méléard, Unifying evolutionary dynamics: from individual stochastic processes to macroscopic models. *Theor. Popul. Biol.* [**69**]{}(3), 297–321 (2006).
Ph. Clément, H. J. A. M. Heijmans, S. Angenent, C. J. van Duijn, B. de Pagter, One-parameter semigroups. CWI Monographs, 5. North-Holland Publishing Co., Amsterdam, 1987.
O. Diekmann, J. A. P. Heesterbeek, J. A. Metz, On the definition and the computation of the basic reproduction ratio $R_0$ in models for infectious diseases in heterogeneous populations. J. Math. Biol. [**28**]{}(4), 365-382 (1990).
R. A. Fisher, The Genetical Theory of Natural Selection, Clarendon Press, Oxford 1930.
S. Gandon, M. E. Hochberg, R. D. Holt, T. Day, What limits the evolutionary emergence of pathogens? Phil. Trans. R. Soc. B [**368**]{}(1610), 20120086 (2013).
A. Gonzalez, O. Ronce, R. Ferriere, M. E. Hochberg, Evolutionary rescue: an emerging focus at the intersection between ecology and evolution. *Phil. Trans. R. Soc. B*, [**368**]{}(1610), 20120404 (2013).
A. Hastings, Transients: the key to long-term ecological understanding? *Trends Ecol. Evol.* [**19**]{}(1), 39–45 (2004).
A. R. Hughes, B. D. Inouye, M. T. J. Johnson, N. Underwood, M. Vellend, Ecological consequences of genetic diversity. *Ecol. Lett.* [**11**]{}, 609–623 (2008).
M. Kimura, A stochastic model concerning the maintenance of genetic variability in quantitative characters. *Proc. Natl. Acad. Sci. U.S.A* [**54**]{}, 731–736 (1965).
J. Kingsolver, D. Pfennig, Patterns and power of phenotypic selection in nature. *Bioscience* [**57**]{}, 561–572 (2007).
J. L. Lush, Animal Breeding Plans. Ames, Iowa: Iowa State Press, 1937.
S. E. Luria, M. Delbr[ü]{}ck, Mutations of bacteria from virus sensitivity to virus resistance. *Genetics* [**28**]{}(6), 491–511 (1943).
K. Mather, J. L. Jinks. Biometrical genetics (2 ed.). London: Chapman & Hall (1971).
G. Martin, R. Aguilée, J. Ramsayer, O. Kaltz, O. Ronce, The probability of evolutionary rescue: towards a quantitative comparison between theory and evolution experiments. *Phil. Trans. R. Soc. B* [**368**]{}, 20120088 (2012).
E. Toprak, A. Veres, J. B. Michel, R. Chait, D. L. Hartl, R. Kishony, Evolutionary paths to antibiotic resistance under dynamically sustained drug selection. *Nat. Genet.*, [**44**]{}(1), 101–105 (2012).
M. Turelli, N. H. Barton, Genetic and statistical analyses of strong selection on polygenic traits: what, me normal? *Genetics* [**138**]{}, 913–941 (1994).
M. Vellend, The consequences of genetic diversity in competitive communities. *Ecology* [**87**]{}, 304–311 (2006).
C. Violle, B. J. Enquist, B. J. McGill, L. Jiang, C. H. Albert, C. Hulshof, V. Jung, J. Messier, The return of the variance: intraspecific variability in community ecology. *Trends Ecol. Evol.* [**27**]{}(4), 244–52 (2012).
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Event sequences can be modeled by temporal point processes (TPPs) to capture their asynchronous and probabilistic nature. We propose an intensity-free framework that directly models the point process distribution by utilizing normalizing flows. This approach is capable of capturing highly complex temporal distributions and does not rely on restrictive parametric forms. Comparisons with state-of-the-art baseline models on both synthetic and challenging real-life datasets show that the proposed framework is effective at modeling the stochasticity of discrete event sequences.'
author:
- |
Nazanin Mehrasa[^1] $ ^{\; 1,3}$, Ruizhi Deng[ ^fnsymbol[1]{}^]{}$^{1,3}$, Mohamed Osama Ahmed$^{3}$, Bo Chang$^{3}$,\
**Jiawei He$^{1,3}$,Thibaut Durand$^{3}$, Marcus Brubaker$^{2,3}$, Greg Mori$^{1,3}$**\
$^{1}$Simon Fraser University $^2$York University $^{3}$Borealis AI\
, ,\
, ,\
\
bibliography:
- 'arxivbib.bib'
title: Point Process Flows
---
Conclusion
==========
In this paper, we propose Point Process Flows (PPF), an intensity-free framework that directly models the point process distribution by utilizing normalizing flows. The proposed model is capable of capturing arbitrary complex time distributions as well as performing stochastic future prediction. The proposed PPF can be optimized by maximizing the exact likelihood using change of variable formula, relaxing the strict tractable likelihood constraint in previous works. Extensive evaluation on both synthetic and challenging real-like datasets shows significant improvement over baseline models.
[^1]: Equal Contribution
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'A search for Supernova Relic Neutrinos $\bar{\nu}_e$’s is first conducted via inverse-beta-decay by tagging neutron capture on hydrogen at Super-Kamiokande-IV. The neutron tagging efficiency is determined to be $(17.74\pm0.04_{stat.}\pm1.05_{sys.})\%$, while the corresponding accidental background probability is $(1.06\pm0.01_{stat.}\pm 0.18_{sys.})$%. Using 960 days of data, we obtain 13 inverse-beta-decay candidates in the range of $E_{\bar{\nu}_e}$ between 13.3 MeV and 31.3 MeV. All of the observed candidates are attributed to background. Upper limits at 90% C.L. are calculated in the absence of a signal.'
author:
- |
\
The Super-Kamiokande Collaboration
title: 'Supernova Relic Neutrino Search with Neutron Tagging at Super-Kamiokande-IV'
---
authors20120519.tex
`N`eutron tagging; Water Cherenkov detector; hydrogen; Supernova Relic Neutrinos
Introduction
============
Neutrinos emitted from all past core-collapse supernovae should form an isotropic flux. Sometimes called the Diffuse Supernova Neutrino Background (DSNB), these neutrinos will be referred to as Supernova Relic Neutrinos (SRN) herein. Many models have been constructed to predict the SRN flux and spectrum [@Totani95; @Totani96; @Malaney97; @Hartmann97; @Kaplinghat00; @Strigari03; @Ando03; @Fukugita03; @Lunardini09; @HBD09; @VP11; @Nakazato13]. Although all six types of neutrinos are emitted from a core-collapse supernova, SRN’s are most likely detected via the inverse beta decay (IBD) reaction $\bar{\nu}_e p \rightarrow e^+ n$ in existing detectors. Super-Kamiokande (SK) has previously carried out searches for SRNs from the expected IBD positrons without requiring the detection of a delayed neutron, placing an integral flux limit for $E_{\bar{\nu}_e}>17.3$ MeV ($E_{\bar{\nu}_e}\approx E_{e^+}+1.3~\mathrm{MeV}$) in the absence of a signal [@relic03; @relic12]. Since the detector cannot directly differentiate electrons from positrons (the positron annihilation signal is below Cherenkov threshold in water), these searches suffer from background of electrons and positrons. Some of these potential backgrounds include atmospheric neutrino $\nu_e/\bar{\nu}_e$ and $\nu_{\mu}/\bar{\nu}_{\mu}$ charged-current interactions and atmospheric neutrino neutral-current interactions. Many background channels either do not produce neutrons or more than one neutron, but they can generate an electron or positron that passes all of the selection criteria, thereby contaminating the candidate samples. Spallation backgrounds have limited the lower anti-neutrino energy threshold in previous SRN searches at SK.
Positive identification of $\bar{\nu}_e$’s by tagging (and counting) neutrons in delayed coincidence will play a critical role in both the suppression of backgrounds in those samples as well as in lowering the energy threshold. Kamland made the first attempt to search for the SRN flux down to 8.3 MeV by detecting IBDs with neutron capture on hydrogen in a one-kiloton liquid scintillator detector [@kamland]. This paper will present a study to detect IBDs with neutron capture on hydrogen at SK, providing an improved search of SRNs from the previous threshold of 17.3 MeV down to the present 13.3 MeV, where greater SRN flux is expected. In addition, this study can be treated as an after-the-fact approach in parallel to the ongoing R$\&$D initiative aimed at detecting IBDs in water with enhanced neutron captures on dissolved gadolinium.
Experimental approaches to detect the neutron
=============================================
To detect the neutron signal, two independent approaches have been proposed to implement this capability in the SK experiment, a large underground water Cherenkov detector containing 50 kilotons of pure water. The detector consists of a cylindrical inner volume viewed by 11,129 inward-facing 50-cm diameter photomultiplier tubes (PMTs), surrounded by an outer annular volume viewed by 1,885 outward-facing 20-cm PMTs. More detailed descriptions of the detector can be found elsewhere [@skintroducion]. The first approach [@gad] involves doping the water with a water-soluble chemical compound of gadolinium, neutron capture on which yields a gamma cascade with a total energy of about 8 MeV. These relatively high energy $\gamma$-rays should be readily seen by SK. The second approach is to detect the single 2.2 MeV $\gamma$ released from neutron capture on hydrogen [@force]. This approach requires a 500-$\mu$s forced trigger scheme following a normal trigger, in order to identify the 2.2 MeV $\gamma$ offline.
The detection of delayed-coincidence 2.2 MeV $\gamma$’s was first successfully demonstrated in SK using forced triggers [@ntag]. In the summer of 2008, SK’s front-end electronics were upgraded, after which began the data-taking period known as SK-IV. A major part of this upgrade was to use a 60 kHz periodic trigger to seamlessly read out all PMTs all the time. In this data stream, events are searched for by a software trigger, which is based on the number of coincident PMT hits within 200 ns. To search for delayed-coincidence 2.2 MeV $\gamma$’s, a new coincidence level was introduced: a super-high-energy (SHE) event requires at least 70 coincident PMT hits corresponding to about 10 MeV (this level was lowered to 58 or about 8 MeV in the summer of 2011). SHE events contain all PMT hits from 5 $\mu$s before the SHE trigger time to 35 $\mu$s afterwards. The 5 $\mu$s of data prior the SHE trigger time provides a chance to catch pre-activity events, e.g. a prompt $\gamma$-ray in a sub-Cherenkov muon background event produced by an atmospheric $\nu_\mu$ interaction with oxygen. SHE events without coincident outer detector activity are always followed by an after trigger (AFT) which contains all PMT hits of the subsequent 500 $\mu$s. The SK-IV data set used in this analysis was taken from November 22, 2008, to December 27, 2011, with a total livetime of 960 days.
The IBD event selection
=======================
The IBD candidate search can be divided into two steps: one to find the prompt signal with an energy ranging from 12 to 30 MeV in the SHE trigger data; the other to tag the IBD signal through the detection of a delayed event shortly after a prompt event is found. The timing window for the delayed event ranges from 2 to 535 $\mu$s following the prompt event. To avoid PMT signal reflection at the SK front-end electronics after an event, the delayed event search starts two $\mu$s after the prompt event time (defined by a GPS-synchronized clock). The prompt events are selected by applying a number of cuts to suppress muon-induced spallation background, atmospheric neutrinos, solar neutrinos, and low energy radioactivities. Details of the selection criteria for the prompt events, such as the spallation cut, pre/post activity cut etc, can be found in [@relic12]. Unlike the analysis in [@relic12], the reconstructed Cherenkov angle is required to be greater than 38 and less than 50 degrees. Also, the solar cosine angle cut of [@relic12] is loosened to 0.9 for energies below 16 MeV because of neutron tagging. The number of remaining solar neutrino events in the sample is estimated to be about two events. Table \[tab:prompt\] gives a summary of the event selection. Most efficiencies are evaluated using Monte Carlo (MC) simulation except for that of the spallation cut. The overall systematic uncertainty of the primary event efficiency is $\sim$3.3% due to uncertainties in the IBD cross section (1.0%), and the data reduction (3.1% as given in [@relic12]), which are added in quadrature.
Cuts $N_{e^+}$ $\epsilon_{e}$ (%)
----------------------- ----------- --------------------
First reduction 49288 99.22$\pm$0.04
Spallation cut 2417 86.02$\pm$0.18
External event cut 2148 82.36$\pm$0.19
Solar cut 1649 81.54$\pm$0.20
Cherenkov angle cut 996 75.17$\pm$0.22
Pre/post activity cut 959 75.06$\pm$0.22
$\pi_{like}$ cut 948 74.32$\pm$0.22
Multi-ring cut 943 73.64$\pm$0.22
$\mu/\pi$ cut 942 73.14$\pm$0.23
: \[tab:prompt\] Summary of the selection criteria for the prompt events ($N_{e^+}$) with energy $E_{e^+}$ ranging from 12 to 30 MeV, the number of events surviving each cut and the evaluated efficiency ($\epsilon_{e}$) from the large mixing angle model [@Ando03]. Errors are statistical only.
An average 2.2 MeV $\gamma$ event in SK produces about seven recorded PMT hits, all of which share a common orientation in both time and space, but most of the hits have various arrival times due to the different travel distances. The flat timing distribution for these signal hits can be sharpened to form a timing peak if the common orientation or vertex is known. However, the continuous 50 MHz dark noise and radioactive background for all the PMTs render a stand-alone reconstruction very difficult because of the long neutron lifetime. Fortunately, since neutrons produced in IBD quickly thermalize and are eventually captured by hydrogen with a mean free path length of $\sim$50 cm, to a good approximation the location for an emission of a 2.2 MeV $\gamma$ can be treated to share a common vertex with the prompt event, for which the reconstructed position resolution is 40$\sim$50 cm. The reconstructed vertex for the prompt event is therefore used to calculate the path length to each hit PMT, in order to subtract the time-of-flight (TOF) from the measured light arrival time. Due to a PMT timing resolution of 3 ns, the hits of real signal events cluster within a 10 ns window, while hits due to PMT dark noise, radioactivity in the surrounding rock, radon contamination events in water, and so on are typically more spread out since the light does not originate from the primary event vertex. A 10 ns sliding window is then applied to search for every timing peak and to give the number of PMT hits (N$_{10}$). Fig. \[fig:n10\] shows the distribution of N$_{10}$ for a 2.2 MeV $\gamma$ signal and background, in which the signal events are from Monte Carlo simulation, while the background events are from the random trigger data.
To remove background PMT hits the following selection criteria are applied:
![Distribution of N$_{10}$ for 2.2 MeV $\gamma$ from MC (dashed) and background (solid) from random trigger data. The arrow indicates the cut for selecting the signal.[]{data-label="fig:n10"}](n10.eps){width="1.0\columnwidth"}
The number of PMT hits in N$_{10}$ is required to be greater than 7. It is noted that about 86% of low energy $\gamma$ background events originating near the detector wall have clustered PMTs in both space and time, and thus survive the N$_{10}$ cut. The number of PMT hits in clusters (N$_{\mbox{cluster}}$) is defined for those within 10 ns and less than $14.1^\circ$ as seen from the positron candidate vertex. A cut on the quantity of N$_{10} - $N$_{\mbox{cluster}}$ is required to be greater than 5. Since most of the light of a 2.2 MeV gamma originates from a single Compton electron, most hits are in the same detector hemisphere, while PMT noise is distributed more uniformly. As a result, a summed vector is calculated for all the hits in N$_{10}$. Angles between the individual vector for each hit and the summed vector are computed, giving the number of hits with an angle greater than $90^\circ$ (N$_{\mbox{back}}$). The delayed event should be N$_{10}-$N$_{\mbox{back}}>$ 6, which removes events with many background hits in the backward hemisphere. Some PMTs are more likely to be illuminated than others for a given vertex. A hit probability of the i-th PMT is defined by $\frac{(\cos \theta_i)_{eff}}{R_i^2}e^{-R_i/L} $, where $R_i$ is the distance from the vertex to the i-th PMT, $\theta_i$ is the incident angle, $(\cos \theta_i)_{eff}$ includes the angular dependence of PMT geometry and reflection/absorption of acrylic case, and $L$ is the attenuation length of Cherenkov light in the SK water. The calculated hit probabilities for all the hits are sorted in decreasing order. The number of hits accounting for the bottom 25-50% of the summed probabilities is defined as N$_{\mbox{low}}$. The fraction varies with the vertex location and is set to 50% when the vertex is close to the wall. A cut on N$_{10} - $N$_{\mbox{low}}$ is required to be greater than 4. The final reduction utilizes a likelihood ratio based on four discriminating variables: number of PMT hits within $\pm$150 ns around N$_{10}$ peak, root mean square of the N$_{10}$ timing peak, root mean square of the azimuth angles for all the PMT hit vectors along the summed vector, and mean value of opening angle between the PMT hit vectors and the summed vector. The likelihood ratio is required to be greater than 0.35.
Table \[tab:delayed\] gives a summary for the selection criteria, the background probability, and the efficiency of the delayed events for each cut. The efficiency of the delayed event is corrected with a factor of $\sim$92% due to the 533 $\mu$s time window. It is observed that most delayed events cannot fire sufficient PMTs to meet the minimum requirement on N$_{10}$.
Cuts Bkg Prob. (%) Efficiency(%)
---------------------------------- ---------------- ----------------
N$_{10}>$ 7 100 30.19$\pm$0.04
N$_{10}-$N$_{\mbox{cluster}}>$ 5 25.48$\pm$0.04 28.27$\pm$0.04
N$_{10} - $N$_{\mbox{back}}>$ 6 21.13$\pm$0.04 26.78$\pm$0.04
N$_{10}- $N$_{\mbox{low}}>$ 4 4.14$\pm$0.02 19.11$\pm$0.04
Likelihood ratio $>$ 0.35 1.06$\pm$0.01 17.74$\pm$0.04
: \[tab:delayed\] Summary of the selection criteria, the probability of accidental background, and the efficiency of finding delayed events. The samples are from the random trigger (background) and Monte Carlo simulation (2.2 MeV $\gamma$ signal). All efficiencies for the delayed events are corrected by $\sim$92% due to the width of the 533 $\mu$s time window. Errors are statistical only. See text for variable name definitions.
Basing the TOF correction on the SRN candidate vertex (rather than the true vertex of the delayed event) changes the efficiency of finding delayed events by at most 2.5% relatively. Uniformity of both the MC signal efficiency and the background probability were studied using 110 positions within the detector. These spatial variations of MC signal efficiency and background probability were found to be 5.9% and 16.8%, respectively. These variations were then assigned to the systematic uncertainties. Therefore, the efficiency and the background probability for the delayed events are $(17.74\pm0.04_{stat.}\pm1.05_{sys.})\%$ and $(1.06\pm0.01_{stat.}\pm 0.18_{sys.})$%, respectively. Combining the primary event efficiency and delayed event efficiency, the IBD detection efficiency ($\epsilon$) is obtained to be $(13.0\pm0.8)$%.
Test with Am/Be source data
===========================
To verify the detection efficiency for the 2.2 MeV $\gamma$’s given in Table \[tab:delayed\], a test was carried out using an Am/Be source embedded in a bismuth germanate (BGO) scintillator during SK-IV. The experimental setup and other details can be found elsewhere [@ntag]. The experimental apparatus was deployed at certain positions in the SK tank, during which the forced trigger gate for catching the 2.2 MeV $\gamma$’s was temporarily enlarged to 800 $\mu$s in order to obtain a more complete neutron lifetime spectrum. To get the time distribution of the source-related background and accidental background, 10 Hz of 800$\mu$s random trigger data was also taken. Fig. \[fig:dtc\] shows the distribution of time differences ($\Delta$T) between the delayed events and the prompt events, which is fitted with an exponential plus a constant with the signal fraction as a free parameter, to give the neutron lifetime in water. In order to verify the neutron lifetime time and examine possible position dependence of detection efficiency, the source was deployed at three different locations: at the center of the tank, close to the wall, and close to the top. All of the resulting lifetime measurements were consistent within one standard deviation. The average neutron lifetime in water was found to be ($203.7\pm2.8$)$\mu$s. The efficiencies measured at the three locations are in agreement within 10%, which also agrees with the estimation of Monte Carlo simulation. The average efficiency of $(19.0\pm0.2)$% in this enlarged 800 $\mu$s window is in good agreement with the value of $(19.2\pm0.1)$% estimated from MC simulation.
![Distribution of $\Delta \textrm{T}$ for the Am/Be data(points). The curve is for the fitting results. The shaded histogram indicates the expected background. Errors are statistical only.[]{data-label="fig:dtc"}](c_ml_bkg_free.eps){width="1.0\columnwidth"}
Analysis and Results
====================
Returning to the low energy SRN search using 960 live days of SK-IV data, after passing the selection criteria for both the prompt events and the delayed events the relevant distributions for the remaining events with at least one neutron candidate are shown in Fig. \[fig:electron\]. There are 13 IBD candidates observed consistent with accidental background events evaluated to be $10\pm1.7$. Two out of these 13 primary events with electron energies around 12 MeV are observed to have two neutron candidates, which indicates they are likely to be from spallation backgrounds with high neutron multiplicity.
A number of studies have been performed to provide insight into the origin of possible background in this energy domain, especially those arising from atmospheric $\bar{\nu}_{\mu}/\nu_{\mu}$ CC interaction, $\pi^{\pm}$ production and NC interactions with water. This is achieved by changing the cut on the Cherenkov angle $\theta_C$ of the primary event, in which an electron/positron is defined with $38^\circ <\theta_C<50^\circ$. The $\mu^{\pm}$ and $\pi^{\pm}$ events are defined with $\theta_C < 38^\circ$, while the NC events are defined with $\theta_C > 50^\circ$. There are 144 primary $\mu^{\pm}$ and $\pi^{\pm}$ candidate events with 22 delayed candidates and 489 NC candidate events with 47 delayed candidates. A clear neutron lifetime curve is observed in both delayed candidate samples, showing that the primary events are indeed accompanied by neutrons. The flat timing offset distribution for the delayed candidates in Fig. \[fig:electron\] does not show significant leakage from these two types of physical backgrounds. The number of atmospheric $\bar{\nu}_e$ events is estimated to be 0.1, while the number of the $\bar{\nu}_\mu$ events is about 1.0. The later is due to the $\bar{\nu}_\mu$ charged-current interaction, which produces a delayed neutron and a positron from an invisible $\mu^+$ Michel decay. In absence of a significant signal the Rolke method [@rolke] is used to convert the number of observed and expected background events $n_{obs}=13$ and $n_{bkg}=10.0\pm1.7$ to a 90% C.L upper limit of 80.1 events in total or 30.5 events/22.5 (kton$\cdot$year), taking into into account the IBD detection efficiency $\epsilon$. Table \[tab:model\] lists the expected number of SRN events in 22.5 kton$\cdot$year for different models. The upper limit on the SRN flux $F_{90}$ can be derived from $N^{\prime}_{90}$ using the following simple relation: $$F_{90}=\frac{N^{\prime}_{90}}{N_P} \times F_M$$ where $F_M$(cm$^{-2}$s$^{-1}$) is the total flux for a certain model and $N_P$ is the predicted annual event rate in the energy range which can be found in Table \[tab:model\]. This table also contains upper limits ($F_{90}$) at 90% C.L. for different models and the predicted annual event rate ($T_P$) after efficiency correction.
SRN model $F_M$ $N_P$ $T_P$ $F_{90}$
------------------------------------------ ------- ------- ------- ----------
Constant SN [@Totani95] 52.3 10.8 1.4 147.5
HBD 6 MeV [@HBD09] 21.8 4.4 0.6 150.9
Chemical evolution [@Hartmann97] 8.5 1.5 0.2 172.6
Heavy metal [@Kaplinghat00; @Strigari03] 31.3 4.7 0.6 201.8
LMA [@Ando03] 28.8 4.2 0.5 208.8
Failed SN [@Lunardini09] 12.0 1.7 0.2 214.9
Cosmic gas [@Malaney97] 5.3 0.7 0.1 230.6
Star formation rate [@Fukugita03] 18.7 1.8 0.2 316.3
Population synthesis [@Totani96] 42.1 1.3 0.2 986.1
: \[tab:model\]Total flux for each SRN model ($F_M$), predicted number of SRN events in 22.5 kton$\cdot$year with a neutrino energy range of 13.3$\sim$31.3 MeV ($N_P$), predicted number of SRN events in 22.5 kton$\cdot$year with a neutrino energy range of 13.3$\sim$31.3 MeV ($T_P$) after IBD efficiency correction and flux upper limit at 90% C.L. ($F_{90}$)(cm$^{-2}$s$^{-1}$).
![Positron energy spectrum of the IBD candidates (points). The histogram represents the expected accidental background. The plot embedded in the upper right shows the timing offset for the delayed candidates. Shown at the bottom of the figure is a plot of the IBD detection efficiency for each energy bin; the jumps at 18 MeV and 24 MeV are due to energy-dependent spallation cuts. Errors are statistical only.[]{data-label="fig:electron"}](electron1.eps){width="1.0\columnwidth"}
Model-independent $\bar{\nu}_e$ differential flux upper limits with one MeV energy bins are also calculated. The 90% C.L upper limits are calculated by $$\phi_{90}=\frac{N_{90}}{T\cdot N_p \cdot \bar{\sigma}}$$ where $N_{90}$ is the upper limit at 90% C.L. in each energy bin, $T$ is livetime in seconds, $N_p$ is the number of free protons, $\bar{\sigma}$ is the average cross section for IBD at the center of each energy bin, and $\epsilon$ is the IBD detection efficiency for each energy bin. Fig. \[fig:uplimit\] shows the upper limits for $\bar{\nu}_e$ in the energy range of 13.3$\sim$31.3 MeV. Limits from KamLAND [@kamland] based on 2343 live-days are also shown for comparison.
The previous SK search for SRN IBD positrons in [@relic12] placed an integral 90% C.L. limit on the SRN flux above 17.3 MeV neutrino energy of 2.9 cm$^{-2}$s$^{-1}$ (LMA model [@Ando03]). In that search, SRN signal and atmospheric neutrino backgrounds were fitted to the energy spectra of the data for three different samples differentiated by the reconstructed Cherenkov angle with an extended unbinned maximum likelihood method. The SRN signal populates the single electron-like sample ($38^\circ <\theta_C<50^\circ$) below 30 MeV (signal region). Various types of atmospheric neutrino background dominate the background region above 30 MeV as well as the other two (background) samples. To compare with the SK-IV differential limits in this paper, the previous SK background spectra as well as the SRN candidate positron spectra above 30 MeV (total energy) were fit to only atmospheric neutrino background contributions. The resulting background fit was extrapolated in the signal region between 16 and 29.5 MeV (total positron energy) taking into account statistical and systematic uncertainties. The data was divided into nine bins of 1.5 MeV. Fig. \[fig:uplimit\] shows the 90% C.L. upper flux limits derived for each bin based on the background expectations with Gaussian uncertainties and the IBD cross section evaluated at the bin center. Below 17.3 MeV spallation background increases exponentially, so SRN detection in that energy range is very difficult without neutron tagging. As the SRN flux per MeV rises with decreasing energy, the region below 17.3 MeV is the most sensitive.
![Model-independent 90% C.L. differential upper limits on SRN $\bar{\nu}_e$ for SK-IV (solid circle). For comparison, both KamLAND result (open square) [@kamland] and previous SK result (solid triangle) are also shown.[]{data-label="fig:uplimit"}](diff2.eps){width="1.0\columnwidth"}
Since this study covers the high end of the solar neutrino spectrum, a solar $\bar{\nu}_e$ upper limit at 90% C.L. of the annual event rate is also calculated, giving an estimate of 21.2 events/22.5 kton$\cdot$year. This corresponds to $4.2\times10^{-4}\times F_{SSM}$, where $F_{SSM}$ is the solar $\nu_e$ flux predicted by the Standard Solar Model [@SSM]. This limit is 20 times more stringent than the previous SK result [@SKI] due to the powerful background reduction provided by neutron tagging. However, note that the limit is an order less stringent than the KamLAND result [@kamland] because of the higher neutrino energy threshold.
Summary and outlook
===================
In summary, a search for SRN $\bar{\nu}_e$ at SK-IV is first conducted via IBDs by tagging neutron capture on hydrogen. The neutron tagging efficiency is determined to be $(17.74\pm0.04_{stat.}\pm1.05_{sys.})\%$, while the corresponding accidental background probability is $(1.06\pm0.01_{stat.}\pm 0.18_{sys.})$%. No appreciable IBD signal in the distribution of neutron lifetime is found using 960 days of data. The number of observed IBD candidates are consistent with the expected accidental background. A model-independent differential flux upper limit at SK is first derived from the previous 17.3 MeV threshold down to 13.3 MeV of the electron anti-neutrino energy.
With more data collected and after further efforts in suppressing spallation background, it is expected that the neutrino energy threshold can be lowered down to 10 MeV and the better SRN flux limit can eventually be obtained with neutron capture on hydrogen at SK-IV. In addition, intense R$\&$D is currently underway towards a gadolinium-enhanced SK. The higher signal detection efficiency and greater background rejection provided by neutron capture on gadolinium, as well as the lowered energy threshold it makes possible, are expected – in the not-too-distant future – to greatly improve SK$^\prime$s sensitivity and ultimately provide the world’s first observation of the SRN signal.
Acknowledgment
==============
The authors gratefully acknowledge the cooperation of the Kamioka Mining and Smelting Company. The Super-Kamiokande detector was built and operated with funds provided by the Japanese Ministry of Education, Culture, Sports and Technology, the United States Department of Energy, and the U.S. National Science Foundation. This work is also supported by the National Natural Science Foundation of China (Grants No.11235006).
References
==========
T. Totani and K. Sato, Astropart. Phys. 3, 367 (1995). T. Totani, K. Sato and Y. Yoshii, Astrophys. J. 460, 303 (1996). R. A. Malaney, Astropart. Phys. 7, 125 (1997). D. H. Hartmann and S. E. Woosley, Astropart. Phys. 7, 137 (1997). M. Kaplinghat, G. Steigman and T. P. Walker, Phys. Rev. D 62, 043001 (2000). L. Strigari, M. Kaplinghat, G. Steigman and T. Walker, JCAP, 0403:007 (2004). S.Ando, K.Sato and T. Totani, Astropart. Phys. 18, 307 (2003). M. Fukugita and M.Kawasaki, Mon.Not.Roy.Astron.Soc. 340, L7 (2003). C. Lunardini, Phys. Rev. Lett. 102, 231101(2009). S. Horiuchi, J.F.Beacom and E.Dwek, Phys. Rev. D 79, 083013 (2009). F. Vissani and G. Pagliaroli, Astron. Astrophys. 528 (2011)L1. K. Nakazato, Phys. Rev. D 88, 083012 (2013). A. Gando [*et al.*]{} (KamLAND Collaboration), Astrophys.J. 745(2012)193. M. Malek [*et al.*]{} (Super-Kamiokande Collaboration), Phys. Rev. Lett. 90, 061101(2003). K. Bays [*et al.*]{} (Super-Kamiokande Collaboration), Phys. Rev. D 85, 052007 (2012). Y. Fukuda [*et al.*]{} (Super-Kamiokande Collaboration), Nucl. Instrum. Meth. A501 (2003) 418-462. J. F. Beacom and M. R. Vagins, Phys. Rev. Lett., 2004, [**93**]{}: 171101. S. Chen and Z. Deng, Nucl. Phys. B Proc. Suppl., 2007, [**166**]{}: 252. H. Watanabe [*et al.*]{} (Super-Kamiokande Collaboration), Astropart. Phys., 2009, [**31**]{}: 320. W. A. Rolke, A. M. Lopez and J. Conrad, Nucl. Instrum. Meth. A551 (2005) 493-503. J. N. Bahcall and M. H. Pinsonneault, Phys.Rev.Lett., 2004, [**92**]{} 121301. Y. Gando [*et al.*]{} (Super-Kamiokande Collaboration), Phys.Rev.Lett., 2003, [**90**]{}: 171302.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The spreading of evolutionary novelties across populations is the central element of adaptation. Unless population are well-mixed (like bacteria in a shaken test tube), the spreading dynamics not only depends on fitness differences but also on the dispersal behavior of the species. Spreading at a constant speed is generally predicted when dispersal is sufficiently short-ranged. However, the case of long-range dispersal is unresolved: While it is clear that even rare long-range jumps can lead to a drastic speedup, it has been difficult to quantify the ensuing stochastic growth process. Yet such knowledge is indispensable to reveal general laws for the spread of modern human epidemics, which is greatly accelerated by aviation. We present a simple iterative scaling approximation supported by simulations and rigorous bounds that accurately predicts evolutionary spread for broad distributions of long distance dispersal. In contrast to the exponential laws predicted by deterministic ’mean-field’ approximations, we show that the asymptotic growth is either according to a power-law or a stretched exponential, depending on the tails of the dispersal kernel. More importantly, we provide a full time-dependent description of the convergence to the asymptotic behavior which can be anomalously slow and is needed even for long times. Our results also apply to spreading dynamics on networks with a spectrum of long-range links under certain conditions on the probabilities of long distance travel and are thus relevant for the spread of epidemics.'
author:
- Oskar Hallatschek
- 'Daniel S. Fisher'
title: 'The acceleration of evolutionary spread by long-range dispersal'
---
Humans have developed convenient transport mechanisms for nearly any spatial scale relevant to the globe. We walk to the grocery store, bike to school, drive between cities or take an airplane to cross continents. Such efficient transport across many scales has changed the way we and organisms traveling with us are distributed across the globe [@ruiz2000global; @suarez2001patterns; @brockmann2006scaling; @gonzalez2008understanding; @rhee2011levy]. This has severe consequences for the spread of epidemics: Nowadays, human infectious diseases rarely remain confined to small spatial regions, but instead spread rapidly across countries and continents by travel of infected individuals [@brockmann2013hidden].
Besides hitchhiking with humans, small living things such as microbes or algae are easily caught by wind or sea currents, resulting in passive transport over large spatial scales [@brown2002aerial; @mccallum2003rates; @d2004mixing; @martin2003phytoplankton; @perlekar2010population]. Effective long distance dispersal is also wide-spread in the animal kingdom, occurring when individuals primarily disperse locally but occasionally move over long distances. And such animals, too, can transport smaller organisms.
These active and passive mechanisms of long-range dispersal are generally expected to accelerate the growth of fitter mutants in spatially extended populations. But how can one estimate the resulting speed-up, and the associated spatio-temporal patterns of growth? When dispersal is only short-range, the competition between mutants and non-mutated (“wild-type”) individuals is local, confined to small regions in which they are both present at the same time. As a consequence, a compact mutant population emerges that spreads at a constant speed, as first predicted by Fisher and Kolmogrov and colleagues [@fisher1937wave; @KPP37]: such selective sweeps are slow and dispersal limited. In the extreme opposite limit in which the dispersal is so rapid that it does not limit the growth of the mutant population, the competition is global and the behavior the same as for a fully-mixed (panmictic) population: mutant numbers grow exponentially fast. It is relevant for our purposes to note that in both the short-range and extreme long-range cases, the dynamics after the establishment of the initial mutant population is essentially deterministic.
When there is a broad spectrum of distances over which dispersal occurs, the behavior is far more subtle than either of the well-studied limits. When a mutant individual undergoes a long distance dispersal event – a jump – from the primary mutant population into a pristine population lacking the beneficial mutation, this mutant can found a new satellite sub-population, which can then expand and be the source of further jumps, as shown in Fig. 1B and 1C. Thus, long-range jumps can dramatically increase the rate of growth of the mutant population. Potentially, even very rare long-range exceptionally long jumps could be important. If this is the case, then the stochastic nature of the jumps that drive the dynamics will be essential.
While evolutionary spread with long-range jumps has been simulated stochastically in a number of biological contexts [@mollison1972rate; @clark2003estimating; @filipe2004effects; @cannas2006long; @marco2011comparing; @brockmann2013hidden], few analytic results have been obtained on the ensuing stochastic dynamics [@mollison1972rate; @biskup2004scaling; @chatterjee2013multiple]. Most analyses have resorted to deterministic approximations [@mollison1977spatial; @kot1996dispersal; @neubert2000demography; @mancinelli2002superfast; @del2003front; @brockmann2007front; @del2009truncation], which are successful for describing both the local and global dispersal limits. Yet, in between these extreme limits, stochasticity drastically changes the spreading dynamics of the mutant population. This is particularly striking when the probability of jumps decays as a power-law of the distance. Just such a distance spectrum of dispersal is characteristic of various biological systems [@levandowsky1997random; @klafter1990microzooplankton; @atkinson2002scale; @fritz2003scale; @ramos2004levy; @dai2007short]. We will show that the behavior is controlled by a balance between the rarity and the potential effectiveness of long distance jumps and the whole spectrum of jump distances can matter. The goal of this paper is to develop the theory of stochastic spreading dynamics when the dispersal is neither short range nor global.
Long distance dispersal can occur either on a fixed network, or more homogeneously in space. For simplicity, we focus on the completely homogeneous case, and then show that many of the results also apply for an inhomogeneous transportation network with hubs between which the long distance jumps occur. For definiteness, we consider for most of the paper the evolutionary scenario of the spread of a single beneficial mutation, but, by analogy, the results can be applied to other contexts, such as the spread of infectious disease or of invasive species.
{width=".95\textwidth"}
Basic model {#sec:minimal-model .unnumbered}
-----------
The underlying model of spatial spread of a beneficial mutant is a population in a $d$-dimensional space with local competition that keeps the population density constant at $\hrho$ and with a probability that any individual jumps to any particular point a distance $r$ away of $\J(r)$ per time per area, per length or per volume. At a very low rate, mutants can appear that have a selective advantage, $s$, over the original population. A lattice version of this model is more convenient for simulations (and for aspects of the analysis): each lattice site represents a “deme" with fixed population size, $\hat{n}\gg \frac{1}{s}$ with the competition only within a deme and the jump migration between demes. Initially, a single mutant occurs and if, as occurs with probability proportional to $s$, it survives stochastic drift to establish, it will take over the local population. When the total rate of migration between demes is much slower than this local sweep time, the spatial spread is essentially from demes that are all mutants to demes that are all the original type.
Short jumps result in a mutant population that spreads spatially at a roughly constant rate. But with long-range jumps, new mutant populations are occasionally seeded far away from the place where they came, and these also grow. The consequences of such long jumps is the key issue that we need to understand. As we shall see, the interesting behaviors occur when the jump rate has a power-law tail at long distances, specifically, with $\J(r)\sim 1/r^{d+\mu}$ (with positive $\mu$ needed for the total jump rate to be finite). Crudely, the behavior can be divided into two types: linear growth of the radius of the region that the mutants have taken over, and faster than linear growth. In Fig. \[fig:illustrate-sweep-dynamics\], these two behaviors are illustrated via simulations on two dimensional lattices. In addition to the mutant-occupied region at several times, shown are some of the longest jumps that occur and the clusters of occupied regions that grow from these. In Fig. \[fig:illustrate-sweep-dynamics\]A, there are no jumps that are of comparable length to the size of the mutant region at the time at which they occur, and the rate of growth of the characteristic linear size $\ell(t)$ of the mutant region — loosely its radius — is roughly constant in time, i.e. $\ell(t)\sim t$. In Fig. \[fig:illustrate-sweep-dynamics\]B and C, $J(r)$ is longer range and very long jumps are observed. These result in faster-than linear growth of the radius of the mutant region, as shown. Before developing analytic predictions for the patterns of evolutionary spread, we report our simulation results in detail.
Results {#sec:results}
=======
Simulated spreading dynamics {#sec:snapshots}
----------------------------
We have carried out extensive simulations of a simple lattice model that can be simulated efficiently. The sites form either a one-dimensional, of length $L$, or two-dimensional $L\times L$ square array with periodic boundary conditions. As it is the spreading dynamics at long times that we are interested in, we assume that the local sweeps in a deme are fast compared to migration. We can then ignore the logistic growth process within demes, so that when jumps occur and establish a new mutant population it is saturated in the new deme by the next time step. Therefore, it is convenient to lump together the probability of an individual to jump, the density of the population from which the jumps occur, and the probability (proportional to $s$) that the mutant establishes a new population: we define $G(r)\equiv s\hrho \J(r)$ so that $d^dr d^dr' G(|\bx-\by|)$ is the rate at which a saturated mutant population near $\bx$ nucleates a mutant population near $\by$. In each computational timestep, we pick a source and target site randomly such that their distance $r$ is sampled from the (discretized) jump distribution $G(r)$ – with the $d^dr$ a lattice site. If the source site is a mutant and the target site a wild type, the identity of the target site is updated to mutant. We measure time in units of $L^d$ timesteps. See SI Sec. S1 for more details on the simulation algorithm.
The growth of mutant populations generated by our simulations is best visualized in a space-time portrait. Fig. \[fig:snapshots\]A and Fig. \[fig:snapshots\]B show the overlaid space-time plots of multiple runs in the regimes $1.5<\mu<2$ and $0.8<\mu<1.1$, respectively. Fig. \[fig:v1-spreading\] shows the growth dynamics of the mutant population over large time and length scales for various values of $\mu$. For $\mu \gtrsim 1.4$, the dynamics clearly approaches a power law. For $\mu \lesssim 0.7$, the simulations are consistent with stretched exponentials. The intermediate regime $0.7
<\mu < 1.3$ is elusive, as one cannot extract a clear asymptotic behavior on the time scales feasible in simulations. The behavior in two-dimensions is qualitatively similar, as shown in the SI Fig. S1.
![Stochastic growth of a mutant population over time in one dimension. Each level of grey represents a single simulation run. A) Power law regime: The values of $\mu$ are $1.5, 1.6, 1.7, 2.0$ in order of increasing darkness. B) Regime of very fast growth with $\mu$ near 1. The values of $\mu$ are $0.8, 0.9, 1.0, 1.1$ in order of increasing darkness. \[fig:snapshots\]](sweeps-L10000-tmax1e+03-pfin1e+00-sigma-15-20.png "fig:"){width="\columnwidth"} ![Stochastic growth of a mutant population over time in one dimension. Each level of grey represents a single simulation run. A) Power law regime: The values of $\mu$ are $1.5, 1.6, 1.7, 2.0$ in order of increasing darkness. B) Regime of very fast growth with $\mu$ near 1. The values of $\mu$ are $0.8, 0.9, 1.0, 1.1$ in order of increasing darkness. \[fig:snapshots\]](label-sweeps-sigma-1.png "fig:"){width="\columnwidth"}
\
To explain these dynamics in detail, we develop an analytical theory that is able to predict not only the asymptotic growth dynamics but also the crucial transients.
Breakdown of Deterministic Approximation {#sec:determ-appr}
----------------------------------------
Traditionally, analyses of spreading dynamics start with a deterministic approximation of the selective and dispersal dynamics — ignoring both stochasticity and the discreteness of individuals. To setup consideration of the actual stochastic dynamics, we first give results in this deterministic approximation and show that these exhibit hints of why they break down. When the jump rate decreases exponentially or faster with distance, the spread is [*qualitatively*]{} similar to simple diffusive dispersal and the extent of the mutant population expands linearly in time. However when the scale of the exponential fall-off is long, the speed, $v$, is faster than the classic result for local dispersal, $v=2\sqrt{Ds}$, which only depends on the diffusion coefficient, $D=\frac{1}{2d}\int r^2 J(r) d^dr$: Instead, there is a characteristic range of jump distances determined by $s$ and $\J(r)$ that dominate the spread. As the dispersal range gets longer for fixed $D$, the dominant jumps occur at lower rates suggesting that the stochasticity may become more important.
If $\J(r)$ has a long tail, in particular, $\J(r)\sim 1/r^{d+\mu}$, the spread in the deterministic approximation becomes exponentially fast with the radius of the region taken over by the mutant population, $\ell(t) \sim \exp(dst/(d+\mu))$. This is almost as fast as in a fully mixed population, with the growth rate of the total mutant population size slower only by a factor of $\frac{d}{d+\mu}$ which approaches unity as $\mu\to 0$, the point at which the spatial structure becomes irrelevant. The origin of this exponential growth for power-law $\J(r)$ is the deterministic feeding of the populations far away by jumps from near the origin: this immediately produces a finite population density at all distances, $r$. After a time of order $1/s$ has passed, the exponential growth of the local population takes off, that is, further jumps to that region no longer matter. As in this time $1/s$ the expected total number of jumps to the [*whole*]{} region further than $r$ from the origin is only of order $1/(sr^\mu)$, the probability that [*any*]{} jumps have occurred is very small for large $r$ and the deterministic approximation must fail [@mollison1972rate].
With local dispersal, the deterministic approximation is a good starting point with only modest corrections to the expansion speed at high population density, the most significant effect of stochasticity being fluctuations in the speed of the front [@van2003front]. At the opposite extreme of jump rate independent of distance, the deterministic approximation is also good with the mutant population growing as $e^{st}$ and fluctuations only causing stochastic variability and a systematic reduction in the coefficient which arise from early times when the population is small. Surprisingly, in the regimes intermediate between these two, the deterministic approximation is not even qualitatively reasonable.
Iterative scaling argument {#sec:self-cons-theory}
--------------------------
We assume that, at long times, most of the sites are filled out to some distance scale $\ell(t)$ and that the density decreases sufficiently steeply for larger distances, such that the total mutant population, $M(t)$, is proportional to $\ell^d(t)$. The validity of the this assumption follows from more accurate analyses given in the Appendix. We call the crossover scale $\ell(t)$ the [*core radius*]{} or “size" of the mutant population.
In the dynamical regimes of interest, the core population grows primarily because it “absorbs” satellite clusters, which themselves were seeded by jumps from the core population. We now show that the rate of seeding of new mutant satellite clusters and the growth of the core populations by mergers with previously seeded clusters has to satisfy an iterative condition that enables us to determine the typical spreading dynamics of the mutant population.
![Sketch of the growth of the compact core of a cluster (red) due to long-range jumps. For the (gray) point at distance $\ell(t)$ to be occupied at time $t$, a seed typically must become established somewhere in the blue space-time region (“target funnel”) by means of a long-range jump (black arrow) from the red “source" region. This schematic leads to the iterative scaling approximation in Eq. . Note that the concavity of the source-funnel geometry, leading to a gap between red and blue regions, is key to our arguments and enables neglecting effects of jumps into the gap region. \[fig:funnel-antifunnel\]](Funnel-InverseFunnel-Extreme2-annotated.png){width=".95\columnwidth"}
It is convenient to illustrate our argument using a space-time diagram, Fig. \[fig:funnel-antifunnel\], in which the growth of the core has the shape of a funnel. Now consider the edge of this funnel at time $T$ (gray circle in the figure). The only way that this edge can become populated is by becoming part of a population sub-cluster seeded by an appropriate long-range jump at an earlier time. To this end, the seed of this sub-cluster must have been established somewhere in the inverted blue funnel in Fig. \[fig:funnel-antifunnel\]. This “target” funnel has the same shape as the space-time portrait of the growing total population, but its stem is placed at $(\ell(T),T)$ and the mouth opens backward in time. Notice that if $\ell(t)$ grows faster than linearly, space-time plots of the growing cluster and the funnel have concave boundaries: this necessitates a jump from the source to the funnel of length much longer than $\ell(T/2)$, as shown.
In $d$ dimensions, the consistency of growth and seeding thus requires $$\label{eq:1}
\int_0^{T}dt \int_{\mathcal{B}_{\ell(t)}}d^d\bx
\int_{\mathcal{B}_{\ell(T-t)}}d^d\by\; G\left(|\ell(T) \bu+\by-\bx| \right) \approx 1 \;,$$ where $\mathcal{B}_\ell$ denotes a $d$ dimensional ball of radius $\ell$ centered at the origin, and we have taken the point of interest to be $\bR=\bu\ell(T)$ with $\bu$ a unit vector in an arbitrary direction. The kernel $G(r)$ represents the rate per $d-$dimensional volume of (established) jumps of size $r$.
Asymptotic results for power law jumps {#sec:power-law-regime}
--------------------------------------
We now show that the asymptotic growth dynamics is essentially constrained by the above iterative scaling argument. Specifically, although the argument is more general, we consider a power law jump distribution $$\label{eq:2}
G(r) \approx \frac{\epsilon}{r^{d+\mu}} \ \ \ {\rm with} \ \ \ \mu<d+1$$ for large enough $r$.
For the [*intermediate-range case*]{} $d<\mu<d+1$, equations and exhibit the asymptotic scaling solution $$\label{eq:plaws}
\ell(t)\sim A_\mu (\epsilon t)^\beta \ \ \ {\rm with} \ \ \ \beta =\frac{1}{\mu-d}\ >1 \;,$$ the [*form*]{} of which could have been guessed by dimensional analysis. Inserting the ansatz into equation determines the pre-factor $A_\mu$ in this iterative scaling approximation, up to an order-unity coefficient: see details in the Appendix. Interestingly, the value $A_\mu$ depends very sensitively on $\mu$ and runs from 0 to $\infty$ as $\mu$ passes through the interval from $d$ to $d+1$. This can be seen in in SI Fig. S2, where $A_\mu$ is plotted as a function of $\mu$ for $d=1$: it drops very steeply, as $A_\mu\sim 2^{-2/(\mu-1)^2}$, for $\mu\searrow
1$, and diverges as $A_\mu\sim 1/(2-\mu)$ for $\mu\nearrow 2$. As we discuss below, these singularities are a manifestation of intermediate asymptotic regimes that lead to very slow convergence to the asymptotic behavior.
We now turn to the (very) [*long-range case*]{}, $0<\mu<d$, for which a direct solution to cannot be found (and the dimensional analysis argument gives nonsense). However, much can be learned by approximating using $G[|\ell(T) \bu+\by-\bx|]\approx G[\ell(T)]$, anticipating the very rapid growth and thus likely smallness of $x$ and $y$ compared to $\ell(T)$: $$\label{eq:approx-self-consist-condition}
G[\ell(T)] \int_0^{T}dt \ell^d(t)\ell^d(T-t) \approx 1 \;,$$ (ignoring two factors from the angular integrations). With $\ell(t)$ growing sub-exponentially, the largest contributions will come from $t\approx
\frac{1}{2}T$. Once we have found the form of $\ell(t)$, the validity of the ansatz can be tested by checking whether the $\ell(T)$ is much larger than $\ell(T/2)$. Indeed, for $\mu<d$ the solution to is a rapidly growing stretched exponential, \[eq:explaws\] (t)\~(B\_t\^) [with]{} = , which can be checked by direct insertion into equation . Notice that as $\mu\searrow 0$, $\eta\nearrow 1$ and $\ell(t)$ grows exponentially as for a flat distribution of long-range jumps that extends out to the size of the system: i.e. the globally-mixed limit. In the opposite limit of $d-\mu$ small, $\eta \sim d-\mu$ and the coefficient $B_\mu$ diverges as shown below. We note that the asymptotic stretched-exponential growth for $\mu<d$ also arises in models of “chemical distance" and certain types of spatial spread for network models with a similar power-law distribution of long distance connections: however the pre-factors in the exponent are different [@biskup2004scaling; @biskup2004graph]. We discuss the connections between these in Sec. \[sec:heterogen\].
For the [*marginal case*]{}, $\mu=d$, the asymptotic behavior is similarly found to be: $$\begin{aligned}
\label{eq:marginal}
\ell(t)\sim \exp\left[\log^2(t)/(4d\log(2))\right] \;,\end{aligned}$$ We now show that this behavior also represents an important *intermediate* asymptotic regime which dominates the dynamics over a wide range of times for $\mu$ close to $d$: this is the source of the singular behavior of the coefficients $A_\mu$ and $B_\mu$ in this regime.
Our source-funnel argument obviously neglects jumps that originate from the not-fully-filled regions outside the core radius $\ell(t)$. An improved version of the funnel argument is presented in the SI Sec. S2 C, which also allows us to estimate the probability of occupancy outside the core region. Further, we present in the Appendix outlines of rigorous proofs of lower and upper bounds for the asymptotic growth laws in one dimension, including the slow crossovers near the marginal case. The linear growth for $\mu>3$ in one dimension has been proven by Mollison [@mollison1972rate]. After the present paper was essentially complete, we became aware of a recent preprint by Chatterjee and Dey [@chatterjee2013multiple] who obtained rigorous bounds in all dimensions for the leading asymptotic behaviors in the three super-linear regimes. Our bounds are somewhat tighter than theirs, including the coefficient and leading corrections to the asymptotic behavior in the marginal case, the absence of logarithmic pre factors in the power-law regime, and the full crossovers for $\mu$ near $d$ in one dimension: comparisons are discussed in the Appendix.
Crossovers and Beyond Asymptopia {#sec:transients}
--------------------------------
Asymptotic laws are of limited value without some understanding of their regime of validity, especially if the approach to the asymptotic behavior is slow. And such knowledge is crucially needed to interpret and make use of results from simulations.
We first consider the short time behavior when the total rate of long jumps is small: i.e., when $\epsilon$, the coefficient of $G(r)\approx
\epsilon/r^{d+\mu}$ is small. Short jumps result in diffusive motion and linear growth $\ell(t) \approx v_0t$ with $v_0$ determined by the details of the selective and diffusive dynamics. Long jumps start to become important after enough time has elapsed that there have been at least some jumps of lengths of order $\ell(t)$: i.e. when $\epsilon
t\ell(t)^d /\ell(t)^{\mu}\gg1$: this occurs after a cross-over time $t_\times\sim [v_0^{\mu-d}/\epsilon]^{1/(d+1-\mu)}$ at which point $\ell\sim \ell_\times \sim [v_0/\epsilon]^{1/(d+1-\mu)}$. At longer times and distances, one can measure lengths and times in units of these crossover scales, defining $\lambda\equiv\ell/\ell_\times$ and time $\theta=t/t_\times$ and expect that the behavior in these units will not depend on the underlying parameters. \[Note that this separation in short-time linear growth and long-time regimes can also be done for more general $G(r)$ although then the behavior will depend on the whole function — the crossover on distances of order $\ell_\times$, and the super-linear behavior on the longer distance form.\]
At times much longer than $t_\times$, there is a slow crossover close to the boundary between the stretched exponential and power law regimes. Thus we must take a closer look at the dynamics in the vicinity of the marginal case, $\mu=d$. As $\mu\searrow d$, the integrand in equation develops a sharp peak at $t=T/2$: half-way between the bounds of the integral. Laplace’s method can then be used to approximate the integral leading to a simplified recurrence relation: $$\label{eq:self-consistency-close-to-marginal-nondim}
\lambda^{d+\mu}(\theta)\sim \theta \lambda\left(\theta /2\right)^{2d} \;,$$ This is really only good to a numerical factor of order unity which can be eliminated by rescaling $\theta$, and to a larger logarithmic factor associated with the narrowness of the range of integration and its dependence on $\theta$ and $\mu-d$: this we analyze in SI Sec. S2 A. Both of these are negligible if we focus on the behavior on logarithmic scales in space and time — natural given their relationships. Defining $\varphi\equiv \log_2(\lambda)$ and $z\equiv
\log_2(\theta)$, and taking the binary logarithm, $\log_2$, of equation , yields a linear recurrence relation that can be solved exactly. Rescaling $\lambda$ can be used to make $\varphi(0)=0$ whence $$\label{eq:solution-recurrence-relation}
\frac{\delta^2}{2d}\varphi(z) =\frac{\delta z}{2d}+\left(1+\frac{\delta}{2d}\right)^{- z}-1 \;,$$ where we introduced the variable $\delta=\mu-d$, which measures the distance to the marginal case.
The asymptotic scaling for $\delta>0$ reproduces the earlier predicted power law regime, , and yields the pre-factor $\log
A_\mu\approx - 2 d \log(2)/\delta^2$, up to correction that is subdominant for small $\delta$ (c.f. SI Fig. S3). For $\delta<0$, the asymptotics yields the stretched exponential in , and fixes the pre-factor $B_\mu\approx 2\log(2)\delta^{-2}$, which could not be obtained from the basic asymptotic analysis carried out above.
The singular pre-factors for $\delta\to0$ give warnings of breakdown of the asymptotic results except at very long times. This peculiar behavior is the consequence of an intermediate asymptotic regime that dominates the dynamics close to the marginal case. This leads to slow convergence to the eventual asymptotic behavior for $\mu$ near $d$. The asymptotic scaling can be observed only on times and length such that $$\label{eq:asymptotic-limit}
\log_2(\theta)\gg 2d/|\delta| \ \ \ {\rm and} \ \ \ \log_2(\lambda)\gg 2d/\delta^2\;.$$ On smaller times, the dynamics is similar to the marginal case, . The rapid divergence of the [*logarithm*]{} of the time after which the asymptotic results obtain, make it nearly impossible to clearly observe the asymptotic limits: in one-dimensional simulations this problem occurs when $|\delta|<0.3$, as us clearly visible in Fig. \[fig:v1-spreading\] and it is likely even harder to observe in natural systems. This underscores the need for the much fuller analysis of the spreading dynamics as via .
While the dynamics at moderate times will be dominated by the initial growth characteristic of the marginal case, we expect to be a good description of the universal dynamics at large $z=\log_2 \theta$ even when $\delta$ is small. The limit $z\to\infty$ while $\zeta\equiv \delta z/2d$ fixed is particularly interesting, as the solution then reduces to a scaling form $$\label{eq:scform}
\frac{\delta^2\varphi}{2d}\approx \chi\left(\frac{\delta
z}{2d}\right)$$ with ()=(-)+-1 . This scaling form allows us to test by simulations our analytic results across all intermediate asymptotic regimes, by plotting data obtained for different $\delta$ in one scaling plot, see Fig. \[fig:scaling-plot-a\]. To make the approximation uniformly valid in both the scaling regime and at asymptotically long times outside of it, we can simply replace, for $\delta<0$, the scaling variable by $\zeta\equiv -\log(\theta)\eta$ (with $\eta$ defined in ): the scaling form will then be valid up to corrections that are small compared to the ones given in all regimes: We thus use this form for the scaling fits in the inset of Fig. \[fig:scaling-plot-a\], plotting data obtained for different $\delta$ in one scaling plot, thereby testing our solution across all intermediate asymptotic regimes.
![Data for different $\mu=d+\delta$ are predicted to collapse on a scaling plot close to the marginal case $\mu=d$ between the stretched-exponential and power-law growth regimes, here demonstrated for the one-dimensional case ($d=1$). The main plot shows the rescaled sizes $\delta^2 \log_2(\lambda)/2$ of the mutant population versus rescaled log-time $\zeta\equiv\delta\log_2(\theta)/2$. The differently colored data sets correspond to $10$ realizations with power law exponents $\mu\in\{0.6, 0.7, 0.8, 0.9, 1.1, 1.2, 1.3, 1.4\}$ (same colors as in Fig. \[fig:v1-spreading\]a). The simulated data collapse reasonably well with the dashed red line, representing the predicted scaling function $\chi(\zeta)=\exp(-\zeta)+\zeta-1$. The inset depicts the same data scaled slightly differently away from the scaling regime such that the horizontal axis for $\delta<0$ shows $\eta \log(\theta)$ (with $\eta$ defined in ), upon which the data collapse improves. Note that $\eta \log(\theta)\approx -\zeta$ as $\delta\to0$ so that the stretched exponential form is recovered for $\zeta$ large and negative. \[fig:scaling-plot-a\]](1d-fits-scalingplotinset-with-inset-annotated.png){width=".9\columnwidth"}
Heterogeneities and Dynamics on Networks {#sec:heterogen}
----------------------------------------
Thus far we have considered spatially uniform systems, in which the jump probability between two points only depends on their separation. However, long distance transport processes may be very heterogeneous. An extreme example is airplane travel which occurs on a network of links between airports with mixtures of short and long distance flights, plus local transportation to and from airports. A simple model is to consider each site to have a number of connections from it, with the probability of a connection between each pair of sites a distance $r$ away being $\C(r)$, independently for each pair: note that although the network is heterogeneous, statistically, the system is still homogeneous as the connection probability does not depend on position. If the rate at which jumps occur across a connection of length $r$ is $\H(r)$, then, averaged over all pairs of sites, the rate of jumps of distance $r$ is $\J(r)=\C(r)\H(r)$. How similar is this to a homogeneous model with the same $\J(r)$, in particular if $\J(r)\sim 1/r^{d+\mu}$? If there are a large number of possible connections along which the key jumps can occur to get from a source region of size $\sim\ell(T/2)$ to a funnel of similar size a distance $\ell(T)$ away, then the fact that these only occur to and from a small fraction of the sites should not matter for the large length scale behavior. The number of such connections is $n_c\sim
\ell(T/2)^{2d}\C(\ell(T))$ with, making the ansatz that, as in the the homogeneous case, $T\ell(T/2)^{2d}\J(\ell(T))\sim 1$ (ignoring subdominant factors) one has $n_c \sim 1/[T\H(\ell(T))]$. Thus the condition for our results to be valid asymptotically is that $\H(r)
\ll 1/ \tau(r)$ with $\tau(r)$ the inverse of the function $\ell(t)$: i.e., in the exponential, marginal, and power-law cases, respectively, that $\H(r) \ll \log(r)^{-1/\eta}$, $\H(r) \ll \exp(-\sqrt{4d\log
2\log r})$, and $\H(r)\ll 1/r^{\mu-d}$.
If there are insufficient number of connections for the heterogeneity of the network to be effectively averaged over, the behavior changes. The extreme situation is when there is a distance-independent rate for jumps along the longest connection out of a site: i.e. $\H(r) \to
const.$. In this case, jumps along the path with the shortest number of steps, $S$, to get from the origin to a point $R$ will reach that point in a time proportional to $S$ i.e. $\tau(R)\sim S(R)$. The geometrical problem of obtaining the statistics of $S(R)$ has been analyzed by Biskup [@biskup2004scaling; @biskup2004graph; @DaSilveira]. For $\mu>d$, $ S\sim R$ and long jumps do not matter, while for $\mu<d$, $S\sim (\log R)^{1/\eta}$ with the same exponent $\eta$ as in the homogeneous case we have analyzed. The difference between this result and ours is only in the power-law-of-$T$ pre-factor of $\ell(T)$ arising from the integral over time: this does not exist in the extreme network limit. In the marginal case, $\mu=d$, $T\sim S \
R^\alpha$ with $\alpha$ dependent on the coefficient of the power-law decay of the connection probability.
If the probability of a jump along a long distance connection decays with distance but more slowly than $1/\tau(r)$, the behavior is similar: for $\mu<d$ again the ubiquitous stretched exponential behavior occurs, while for $\mu>d$ there are too few connections only if $\C(r) < 1/r^{2d}$ in which case the number of steps and the time are both proportional to the distance. The marginal cases we have not analyzed further.
For natural transport processes, the probabilities of long dispersal events will depend on both the source and the destination. If the heterogeneities are weak on large length scales, our results still obtain. But if there are sufficiently strong large scale heterogeneities, either in a spatial continuum or in the network structure (i.e. location of nodes and links and the jump rates along these, or hub-spoke structure with multiple links from a small subset of sites), then the spatial spread will be heterogeneous even on large scales: how this reflects the underlying heterogeneities of the dispersal has to be analyzed on a case by case basis.
Discussion {#sec:discussion}
==========
We have studied the impact of long-range jumps on evolutionary spreading using the example of mutants that carry a favorable genetic variant. To this end, we analyzed a simple model in which long-range jumps lead to the continual seeding of new clusters of mutants, which themselves grow and send out more migrant mutants. The ultimate merging of these satellite clusters limits the overall growth of the mutant population, and it is a balance of seeding and merging of sub-clusters that controls the spreading behavior.
To classify the phenomena emerging from this model, we focussed on jump distributions that exhibit a power law tail. We found that, with power law jumps, four generic behaviors are possible on sufficiently long times: The effective radius of the mutant population grows either at constant speed, as a super-linear power law of time, as a stretched exponential, or simply exponentially depending on the exponent, $d+\mu$, of the power law decay of the jump probability. These predictions are in contrast to deterministic approximations that predict exponential growth for power-law decaying jump kernels [@mollison1977spatial; @kot1996dispersal; @mancinelli2002superfast; @del2003front]. In dimensions more than one, the results also contradict the naive expectation from dynamics of neutral dispersal, that a finite diffusion coefficient is sufficient for conventional behavior (in this context, finite speed of spreading): specifically, for $d+1 > \mu >2$, $D<\infty$ but the spread is super-linear — indeed stretched exponential for $d>\mu>2$. That super-linear dynamics can occur for $\mu<2$ is not surprising as even a migrating individual undergoes a Levy flight: more surprising is that this occurs even when the dynamics of individuals is, on large scales, like a normal random walk. The breakdowns of both deterministic and diffusive expectations are indicative of the importance of fluctuations: the dynamics is dominated by very rare — but not too rare — jumps: roughly, the most unlikely that occur at all up to that time. One of the consequences of this control by the rare jumps is the relatively minor role played by the selective fitness advantage, $s$, of the mutants. With short-range dispersal, the speed is $v\approx \sqrt{Ds}$, and for global dispersal, $\log M \approx st$. By contrast, in the regime of power law growth, the asymptotic growth of the mutant population becomes (to leading order) independent of the speed of the growth of individual clusters although when individual clusters grow more slowly, the asymptotic regime is reached at a later time. In the stretched exponential regime, the growth of sub-clusters sets the cross-over time from linear to stretched exponential, and thus determines the pre-factor in the power law that characterizes the logarithm of the mutant population size.
An important feature of our results is that the approach to the asymptotic laws is very slow in the vicinity of the marginal cases, as illustrated in Fig. \[Fig:phase-diagram\]. Consider for instance the two-dimensional case, where we have asymptotically stretched exponential growth for $\mu<2$: For $\mu=1.8$ ($\mu=1.6$), the epidemics has to run for times $\theta\gg 10^6$ $(10^3)$ to reach the asymptotic regime. By that time, however, the mutant population with $\lambda\gg 10^{30}$ $(10^7)$ would have certainly spread over the surface of the earth. To describe the full cross-over dynamics, we developed an approximate recurrence relation based on the geometric source-funnel argument described in the illustration Fig. \[fig:funnel-antifunnel\]. This predicts a universal cross-over function for the transient dynamics near $\mu=d$ that can be uncovered by simulations for different power law exponents collapsed onto one scaling plot, c.f. Fig. \[fig:scaling-plot-a\]. The good agreement of the predicted scaling function and simulation results provides strong support for the iterative scaling approximation. The rigorous bounds whose proofs are outlined in the SI Sec. S3, provides further support. Understanding this crossover is essential for making sense of, and extrapolating from, simulations as asymptotic behavior is not visible until enormous system sizes even when the exponent $\mu$ is more than $0.2$ from its marginal value.
Another benefit of the ability of the simple iterative scaling argument to capture well non-asymptotic behavior, is that it can be used in cases in which the dispersal spectrum of jumps is not a simple power-law, e.g. with a crossover from one form to another as a function of distance. And the heuristic picture that it gives rise to – an exponential hierarchy of time scales separated by roughly factors of two — is suggestive even in more complicated situations. That such a structure should emerge without a hierarchal structure of the underlying space or dynamics is perhaps surprising.
![Close to the marginal case, $\mu=d$, the spreading dynamics exhibits three behaviors. On asymptotically large times, either stretched exponential growth for $-d<\delta\equiv
\mu-d<0$ or superlinear power law growth for $0<\delta<1$ occurs. However, the approach to the asymptotic regime is extremely slow for $\mu$ close to $d$. For log-times $\log_2(\theta)\ll 2d/|\delta|$, the behavior is controlled by the dynamics of the marginal case. Note that cross-over behavior also obtains near to the borderline between linear and super-linear behavior at $\mu=d+1$, cf. SI Sec. S2 B. []{data-label="Fig:phase-diagram"}](Phasediagram-annotated.png){width=".9\columnwidth"}
Potential applications and dynamics of epidemics
------------------------------------------------
Our primary aim biologically is the qualitative and semi-quantitative understanding that emerges from consideration of the simple models and analyses of these, especially demonstrating how rapid spatial spread of beneficial mutations or other biological novelty can be even with very limited long-range dispersal. As the models do not depend on any detailed information about the biology or dispersal mechanisms, they can be considered as a basic null model for spreading dynamics in [*physical*]{}, rather than more abstract network, space.
The empirical literature suggests that power-law spectra of spatial dispersal are wide-spread in the biological world [@levandowsky1997random; @klafter1990microzooplankton; @atkinson2002scale; @fritz2003scale; @ramos2004levy; @dai2007short]. Although these are surely neither a constant power-law over a wide range of scales, nor spatially homogeneous, our detailed results are not directly applicable. But, as discussed above, our iterative scaling argument is more general and can be applied with more complicated distance dependence, anisotropy, etc. Furthermore, some of the heterogeneities of the dispersal will be averaged out for the overall spread, while affecting when mutants are likely to arrive at particular locations.
For dispersal via hitchhiking on human transport, either of pathogens or of commensals such as fruit flies with food, the apparent heterogeneities are large because of the nature of transportation networks, although data suggest that dispersal of humans can be reasonably approximated by power-law jumps [@brockmann2006scaling]. Whether or not transport via a network with hubs at many scales fundamentally changes the dynamics of an expanding population of hitchhikers from that with more homogeneous jump processes, depends on the nature of how the population expands.
For spread of a human epidemic, there are several possible scenarios. If the human population is reasonably uniform spatially, and the chances that a person travels from, say, their home to another person’s home is primarily a function of the distance between these rather than the specific locations, then whether or not the properties of the transportation network matter depends on features of the disease. If individuals are infectious for the whole time the outbreak lasts and if transmission is primarily at end points of journeys rather than enroute — for example HIV — then the transportation network plays no role except to provide the spatial jumps. At the other extreme but still within an SI model, is if individuals living near hubs are more likely to travel (or even if destinations near hubs are more likely), and, more so, if infections are likely to occur enroute, in which case the structure of the transportation network — as well as of spectra of city sizes, etc — matters a great deal. In between these limits the network (or lack of it in places) may matter for initial local spread but at longer times the network structure may effectively average out and the dynamics be more like the homogeneous models. The two opposite limits and behavior in between these, together with the specific network model we analyzed with jumping probabilities depending on distance even in the presence of a connection — as is true from airports — all illustrate an important point: geometrical properties of networks alone rarely determine their properties: quantitative aspects, such as probabilities of moving along links and what exists at the nodes, are crucial.
More complicated epidemic models can be discussed within the same framework: The model discussed thus far corresponds to an SI model, the most basic epidemic model, which consists of susceptible and infected individuals only. Many important epidemics are characterized by rather short infectious periods, so that one has to take into account the transition from infected to recovered: SIR models. This changes fundamentally the geometry of our space time analysis, illustrated in Fig. \[fig:funnel-antifunnel\]. While the target funnel remains a full funnel, the source funnel becomes hollow: The center of the population consists mostly of fully recovered individuals, whose long-range jumps are irrelevant. The relevant source population of infected individuals is primarily near the boundaries of the funnel. This leads to a break in the time-symmetry of the argument. As a result, the spreading crosses over from the behavior described above to genuine SIR behavior. The SIR dynamics is closely related to the scaling of graph-distance in networks with power-law distributions of link lengths [@biskup2004scaling; @biskup2004graph; @DaSilveira], as recently shown by one- and two-dimensional simulations [@grassberger2013sir; @grassberger2013two]. In particular, the limited time of infectiousness causes wave-like spreading at a constant speed for $\mu>d$. But, importantly, the spreading velocity is controlled by the SI dynamics we have studied until a time of order the infectious period. Analogous crossovers from SI to SIR occur also with longer range jumps. More generally, other complications can be discussed in our framework, and we expect new behaviors depending on how they modify the geometry of the source to target-funnel picture.
As a last note, our analyses naturally provide information on the typical structure of infection (or coalescent) trees. For a given site, the path of jumps by which the site was colonized can be plotted as in Fig. \[fig:infection\_trees\]. Doing this for many sites yields coalescent (or infection) trees, which can reveal the key long-range jumps shared by many lineages. For actual epidemics, combining spatio-temporal sampling of rapidly evolving pathogens with whole genome sequencing is now making it possible to construct such infection trees. For inference purposes, it would therefore be interesting to analyze more about the statistical properties of such infection trees and how they depend on the dispersal properties, network structure, and other features of epidemic models.
{width=".95\textwidth"}
This work was partially supported by the DFG via HA 5163/2-1, NSF via DMS-1120699 and PHY-1305433.
[38]{} natexlab\#1[\#1]{}bibnamefont \#1[\#1]{}bibfnamefont \#1[\#1]{}citenamefont \#1[\#1]{}url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[2\]\[\][[\#2](#2)]{}
, , , and , , ****(), .
, , .
, , , .
, and , , ****(), .
, and , , ****(), .
, , and , , ****(), .
, and , , ****(), .
, , and , , ****(), .
, , ****(), .
, , and , , ****(), .
, and , , .
, , , and , , ****(), .
, and , , , .
, , , , and , , ****(), .
, , , and , , ****(), .
, and , , ****(), .
, , ****(), .
, , and , , ****(), .
, , and , , ****(), .
, , ****(), .
, , ****(), .
, , and , , , .
, , and , , ****, .
, , and , , ****(), .
, , and , , ****.
, , and , , ****, .
, , and , , ****(), .
, , ****(), .
, , and , , ****(), .
, , , .
, *et al.*, , in ** ().
, and , , ****(), .
, , , and , , ****(), .
, , , , , and , , ****(), .
, , , , , and , , ****(), .
, , , , , , and , , ****(), .
, , and , , ****(), .
, , ****(), .
Simulation details {#sec:simulation-details}
==================
Simulation Algorithm {#sec:simulation-algorithm}
--------------------
The state of the population is described by a linear array of $N$ sites with periodic boundary conditions. $N$ is chosen large enough so that end-effects can be ignored (typically between $10^7$ and $10^8$ sites). Each site has the identity of either mutant or wildtype. Initially, the whole population is wildtype except for the central site, which is occupied by mutants.
In each computational timestep, a source site $A$ and target site $B$ are chosen randomly such that their distance $r$ is sampled from a probability density function with a tail $\mu r^{-(1+\mu)}$ at large $r$ (see below). If $A$ is mutant and $B$ a wild type then $B$ turns into mutants and seeds a new mutant cluster. We use the convention that $N$ time-steps — i.e. an average of one jump attempt per site — comprise one unit of time, or effective “generation". The rate of long-range jumps should be thought of as representing the product of the probability to establish a new cluster per jump and the jump rate per generation per site.
Jump size distribution {#sec:jump-size-distr}
----------------------
In our simulations, the distance $X$ of a long-range jump was generated as follows. First, draw a random number $Y$ within $(0,1)$, and calculate the variable $$\label{eq:generatejumps1}
X=\left[Y \left(L^{-\mu}-C^{-\mu}\right) +C^{-\mu} \right]^{-1/\mu}\;.$$ where $C$ is a cutoff (see below) and $L$ is the system size. This generates a continuous PDF $$\label{eq:generatejumps2}
\Pr(X=x)=x^{-(\mu+1)} \frac{\mu (C L)^\mu}{L^\mu-C^\mu}$$ with $x$ values in $(C,L)$. The actual jump distance is obtained from $X$ by rounding down to the next integer. (Note that, because the distribution has a tail $\mu r^{-(1+\mu)}$, we have to choose $\epsilon=
\mu$ in equation (2) of the main text.)
For the one dimensional data in all graphs of this paper, we used $C=1$ and system sizes ranging from $L=10^9$ for $\mu=0.6$ to $L=10^8$ for $\mu=1.4$. For such large systems, the tail of the distribution is well approximated by $p(x)\sim \mu x^{-(\mu+1)}$, as stated in the main text. We also tested variations in the cutoff $C$. Using $C=10$ or $C=100$ only affected the short-time dynamics and had very little influence on the intermediate asymptotic or long distance behavior of the system.
For our two-dimensional simulations, we draw jump sizes from the same distribution as the one described above. We set the lower cutoff to $C=1.5>\sqrt{2}$ to make sure that jumps reach out of the source lattice point. After the jump size is drawn, the jump direction is chosen at random.
Iterative scaling approximation: details and extensions {#sec:Iterative-scaling-approx}
=======================================================
Sub-dominant corrections from time integrals {#sec:laplaces-method}
--------------------------------------------
In analyzing the results from the iterative scaling approximation we have ignored the effects of the range of $t$ around $T/2$ that dominates the probabilities of occupation at time $T$. This is valid for obtaining the leading behaviors of $\log \ell(t)$ in the large time limit, but there are corrections to these that can be larger than those that arise from the short-time small-length-scale crossovers that we discussed in the text and do so further below. For $\mu$ not much smaller than $d+1$, when the growth of $\ell$ is a modest power of time, the factor from the range of the time integral is only of order unity and hence no worse than other factors – including from the stochasticity — that we have neglected. But when $\ell(t)$ grows very rapidly, the range of $\frac{1}{2}T - t$ that dominates is much smaller than $T$ and the corrections are larger.
With rapid growth of $\ell(t)$, a saddle point approximation to the time integral is valid: $\int_0^tdt\ell(t)\ell(T-t)\approx Tc_{T/2}
\ell^2(T/2)$ with the pre-factor given by $$\label{eq:integral-laplace}
c_t \approx \sqrt{\frac{2\pi}{-2t^2\partial_t^2\log \ell(t)}}$$ which, with the second derivative absorbing the $t^2$ factor, the derivative part can be rewritten as $t^2\partial_t^2\log
\ell(t)=-\partial_{\log t}\log \ell +\partial^2_{\log t}\log \ell$. With the asymptotic growth laws we have derived, this gives $c\sim
\sqrt{\mu-d}$ for $\mu-d$ small and positive, $c_t \sim 1/\sqrt{\log
t}$ for $\mu=d$, and $c_t \sim t^{-\eta/2}$ for $\mu<d$. Integrating up the effects of this over the scales yields the following corrections in the various regimes: For $\mu>d$, the coefficient, $A_\mu$, of $t^\beta$ is changed by a multiplicative factor which is much less singular for $\mu \searrow d$ than that already obtained. For $\mu=d$, C\[\^2 t - t t + [O]{}(t)\] with $C=1/4d\log 2$, the second term being new and the smaller correction term including the effects of the small time cross-over. For $0<\mu<d$, B\_t\^- t with the second term for $\mu
\nearrow d$ just that which occurs in the crossover regime analyzed in Sec. ”Crossovers and Beyond Asymptopia” (main text), where we showed that the coefficient $B_\mu$ diverges proportional to $1/(d-\mu)^2$.
Pre-factors in power law regime {#sec:power-law-regime-1}
-------------------------------
In the main text, we mainly focussed on regimes in which the mutant growth is very much faster than linear, i.e. $\mu \lesssim
d+1/2$. This allowed us to approximate the integrals in the iterative scaling approximation of equation (1) (main text) by the use of Laplace’s method. This saddle point approximation yields the correct scaling for all exponents $\mu<d$, but (as we see below) incorrect pre-factors in regimes where the actual growth is close to linear, i.e. in the power law growth regime with $d+1>\mu\gtrsim d$.
To obtain a better estimate of the pre-factors in this power law regime, it is helpful to directly solve for the asymptotics of the iterative scaling argument in equation (1) (main text). Here, we demonstrate how this can be done for $d=1$: Assume that most of the weight in the integral comes from regions, where the jump kernel is well approximated by its power-law tail described in equation (2) (main Text). Given $\mu< 2$ (for $d=1$), this always holds at sufficiently long times. Then, one has $$\label{eq:Oskar}
\epsilon \int_0^{t}dt' H(t') \;,$$ where $$\begin{aligned}
\label{eq:Oskar2}
\mu (\mu-1)H(t')&=& (\ell(t)-\ell(t')-\ell(t-t'))^{1-\mu} \nonumber \\
& &-(\ell(t)-\ell(t')+\ell(t-t'))^{1-\mu} \nonumber \\
& &+(\ell(t)+\ell(t')+\ell(t-t'))^{1-\mu} \nonumber\\
& &-(\ell(t)+\ell(t')-\ell(t-t'))^{1-\mu} \;.\end{aligned}$$ For $1<\mu<2$, equation exhibits an asymptotic power law solution $$\label{eq:plaws-glaw}
\ell(t)=A_\mu (\epsilon t)^{1/(\mu-1)}\;.$$ By inserting this ansatz into Eq.s and , we obtain the following result for the numerical pre-factor $$\label{eq:kappa-OH}
A_\mu^{1/(\mu-1)}=\int_0^1dz \tilde R(z)$$ with $\tilde R(z)$ being equal to Eq. (\[eq:Oskar2\]) with $\ell(t)$ replaced by $z^{1/(\mu-1)}$.
The resulting pre-factor is plotted as a function of $\mu-1$ in Supplementary Fig. \[fig:prefactors\]. Notice that $A_\mu$ strongly depends on the exponent $\mu$. It sharply drops for $\mu$ approaching $1$, where it follows the asymptotics $2^{-2(\mu-1)^{-2}}$. On the other hand, as $\mu$ approaches the other marginal case at $\mu=2$, the pre-factor diverges as $A_\mu\sim (2-\mu)^{-1}$. indicate the importance of intermediate asymptotic regimes, as discussed in the main text Sec. ”Crossovers and Beyond Asymptopia” (main text).
While we have focussed on the marginal case near $\mu=d$ in this article, it is clear that another case of marginal stability controls the cross-overs near $\mu=d+1$. Simulation results reported in supplementary Fig. \[fig:second-marginal-point\] indicate that $\ell(t)/t\sim \log(t)$ for $\mu=2$ in one dimension. This is consistent with our funnel argument: With nearly constant speed, the gap between the funnels remains roughly constant for a time of order $t$. To ensure that the source emits about one jump to the target funnel, we must have that, per unit time, the probability of jumping over the gap is of order $1/t$. Thus, the gap size $\Delta \E$ should be such that $\Delta \E^{-\mu}\sim 1/t$. For $\mu=2$, we thus have $\Delta \E\sim t$, i.e., the key jumps span distances of order $t$. This is ensured when $(\ell(t)-2\ell(t/2))/t\sim$ const., i.e., if $\ell(t)/t\sim\log t$. Note that a rigorous upper bound of this form follows from the arguments presented in Sec. \[Sec:intermediate-range-case\] for the regime $\mu>d$. The jumps of order $O(t)$ that drive the logarithmic increase in spreading velocity might be the “leaps forward” [@mollison1972rate] recognized by Mollison in one of the earliest studies on spreading with long-range jumps.
Occupancy profiles and relevance of secondary seeds {#sec:dens-prof-relev}
---------------------------------------------------
In the main text, we introduced the notion of a nearly occupied core of the population (of size $\ell(t)$) as the source of most of the relevant seeds in the target funnel. However, it is clear that outside of this core there is a region of partial occupancy. This region is potentially broad, in particular for $\mu\to 0$, and may therefore lead to a significant fraction of relevant seeds. An improved theory should account for those secondary seeds, and should also be able to determine the profile of mean occupancy or, equivalently, the probability that a site is occupied. While we give rigorous bounds on how the total population grows in Sec. \[sec:rigorous\], we first use an improved version of our funnel argument to describe the occupancy profiles.
We focus on the probability, $\q(r,t)$, that a time $t$ after a mutant establishes, it will have taken over the population a distance $r$ away. We expect that $\q(r,t)$ will be close to unity out to some core radius $\ell(t)$, and then decrease for larger $r$, with the average total mutant population proportional to $\ell(t)^d$. With only short-range dispersal, $\ell(t)\approx vt$ and the core is clearly delineated but when long jumps are important the crossover from mostly occupied core to sparsely occupied halo will not be sharp. The more important quantity is the average of the total area (in two-dimensions or linear extent or volume in one or three) occupied by the mutant population, we denote this $\M(t) =\int d^dr \q(r,t)$.
To find out when long jumps [*could*]{} be important, we first ask whether there are likely to be [*any*]{} jumps longer than $\ell(t)$ that occur up to time $t$. The average number of such long jumps is of order $t\ell(t)^d \int_{\ell(t)}^\infty r^{d-1}dr G(r) $. If $G(r)$ decreases more rapidly than $1/r^{2d+1}$, this is much less than $t/\ell(t)$ for large $t$. As $\ell(t)$ increases at least linearly in time, the probability that there have been any jumps longer than $\ell(t)$ is very small. The guess that $\ell$ indeed grows as $vt$, and consideration of jumps that could advance the front fast enough to contribute substantially to $v$, leads, similarly, to the conclusion that there is a maximum $t$-independent jump length beyond which the effects of jumps are negligible, indeed, their effect decreases more rapidly than $G(r)$. This reinforces the conclusion that there is only linear growth with $\mu>d+1$: a very strong breakdown of the deterministic approximation which yielded exponential growth for any power law.
When $G$ is longer range, in particular if $G(r) \sim 1/r^{d+\mu}$ with $\mu<d+1$, many jumps longer than $\ell(t)$ will have occurred by time $t$. We now investigate the effects of such long jumps on the density profile. To do so, we investigate the behavior of $\q(R,T)$ for large $R$ and $T$ in terms of the $\{\q(r,t)\}$ at shorter times and — primarily — corresponding distances $r \sim \ell(t) $ which can be much less than $R$. Mutants can get to a chosen point, $\bR$, by one making a long jump at time $t$ from a starting point $\bx$ to an end point, $\by$, and subsequent spread from there to $\bR$ during the remaining time interval of duration $T-t$. The rate (per volume elements) of this occurring is $\q(\x,t)G(|\bx-\by|) \q(|\bR-\by|,T-t)$. In the approximation that these are independent, the probability that this does not occur at any $t<T$ from any $\bx$ to any $\by$ is simply Poisson so that (R,T)1- e\^[-(R,T)]{} \[eq:occupancy-probabili\] with (R,T) \_0\^T dt \_[z>(T/2)]{} d\^dd\^d (,t)G(z) (|--|,T-t) . \[eq:Q-definition\] Here, we substituted the final point $\by=\bx+\bz$ by the sum of the jump start site and a jump vector $\bz$, over which we integrate. Note that a lower cut-off in the $z$ integral is necessary to exclude the many very short jumps that lead to strongly correlated establishments. The cut-off is also necessary to not count mutants that results from growth in the target area, rather then seeding from the source funnel. Our main assumption here is that if a single seed is sufficiently far from other seeds or occupied regions then the growth from the seed is independent of the rest of the system as long as collisions are unlikely.
When the jump integral is strongly peaked at $z=R$, as is the case in or close to the stretched exponential regime, the final results will be independent of this cutoff to leading order. Then, we can approximate $Q(R,t)$ as $$\label{eq:Poisson-variable}
Q(R,t)\approx G(R)\int_0^{t}dt' M(t') M(t-t') \;,$$ where $M(t)$ is the expected total size of a population at a time $t$, $$\label{eq:mass}
M(t)=\int d^dx q(x,t) \;.$$
For a power-law kernel $G(R)=G(1)R^{-(1+\mu)}$, we make the ansatz that Eqs. , and can be approximately solved by a scaling form $$\begin{aligned}
\label{eq:scaling-form-densities}
q(R,t)&=&\Xi\left(\frac{R}{\lambda(t)}\right) \\
{\rm with} \ \ \Xi(\xi)&=&1-\exp\left (-\xi^{-(d+\mu)}\right)\;,\end{aligned}$$ which leads to the condition $$\label{eq:Q-scaled}
Q(\xi\lambda(t),t)=\xi^{-(d+\mu)} \kappa^2 G(1)
\lambda^{-(d+\mu)}\int_0^t dt' \lambda^d(t') \lambda^d(t-t') \;,$$ where $M(t)=\kappa\lambda(t)^d$ and $\kappa_\mu$ is given by $$\label{eq:kappa}
\kappa=\int d\xi^d \Xi(\xi)\;.$$ Thus, the above scaling form is a valid solution if the characteristic scale $\lambda(t)$ satisfies $$\label{eq:lambda}
\kappa^2 G(1)
\lambda^{-(1+\mu)}\int_0^t dt' \lambda(t') \lambda(t-t')=1 \;.$$ The resulting condition is similar to our condition in the main text but differs by the numerical factor $G(1)\kappa_\mu^2$. In one dimension, $$\label{eq:kappa-1d}
\kappa_\mu=2\Gamma\left(\frac{\mu}{1+\mu}\right) \;.$$ The divergence $\kappa_\mu^2 \sim \mu^{-2}$ as $\mu\to0$ indicates the importance of seeds from the tail regions for small $\mu$.
Notice that at long distances, $R\gg\ell(T)$, the length of the jumps that dominate are close to $R$ so that our approximation for $\Q(R,T)$ should be correct even if it is a poor approximation for $R\approx
\ell(T)$. Thus the decrease in $\q$ at large distances is simply proportional to $G(R)$, more specifically, (R,T) \~ (as predicted by the scaling form) so that it is of order unity at $R\sim \ell(T)$. Note that this implies that, since $G(r)$ is integrable, $\int d^dr \q(r,t)$ is indeed dominated by $r\sim \ell(t)$ as we have assumed. This long distance form for the density profile is also found in the analyses of upper and lower bounds in the next sections: thus it can be readily proved along the same lines.
Rigorous bounds {#sec:rigorous}
===============
As most of our results are based on approximate analyses and heuristic arguments, it is useful to supplement these by some rigorous results. We focus on the one-dimensional case: extensions to higher dimensions can be done similarly, although with a few complications that will require some care. We sketch here the arguments that can lead to proofs without all the details filled in.
As via the heuristic arguments, we would like to obtain the behavior at longer times in terms of the behavior at shorter times, in particular times around half as long.
We would like to prove that there exist time dependent length scales, $\LL(t)$ and $\LG(t)$ and functions, $\FL(r,t)$ and $\FG(r,t)$ such that the probability, $\q(r,t)$, that a site at $r$ from the origin is occupied at time $t$, is bounded above and below by (r,t) < (r,t) < (r,t) [with]{} M\_< dr (r,t) (t) & \_> dr (r,t) (t) for all times. Then $\LL$ and $\LG$ are lower and upper bounds for $\ell(t)$ — with some appropriately defined definitions of $\ell(t)$ which differ somewhat, although not significantly, for the upper and lower bounds.
While we would like the upper and lower bounds on $\ell(t)$ to be as close as possible to each other, in practice, we have obtained bounds that are good on a logarithmic scale: i.e. for $\log \ell(t)$, rather than on a linear scale.
Similarly, we would like to have the bounds be close to the actual expected form of $\q$, with $\FL$ very close to unity for $r \ll \LL$ and proportional to $\lb[\frac{\LL}{r}\rb]^{\mu+1}$ for $r \gg \LL$ and similar for the upper bounds. It is often more convenient to consider the typical time-to-occupation as a function of the distance, $\tau(r)$ and derive upper and lower bounds for this, $\TG(r)$ and $\TL(r)$, respectively, such that (t=(r))=r & (t=(r))=r with (r) > (r) > (r) . Because of the faster-than-power-law growth, bounds on $\tau(r)$ are generally much closer than those on $\ell(t)$. As we would like to justify the use of the heuristic iterative scaling arguments more generally, it is especially useful to obtain iterative bounds directly of the form used in those heuristic arguments: $\ell(T)$ in terms of $\{\ell(t)\}$ for $t$ in a range near $T/2$. As the heuristic arguments do, in any case, only give $\ell(T)$ up to a multiplicative coefficient of order unity, we will generally ignore such order-unity coefficients in length scales except for coefficients that diverge or vanish exponentially rapidly as $\mu \to d$, in particular in the intermediate range regime the coefficient, $A_\mu$ in $\ell(t)\sim A_\mu t^{1/(\mu-d)}$ which vanishes as $\log (A_\mu) \approx - \log 4 /(\mu-d)^2$ as $\mu \searrow d$.
Upper bounds
------------
### Simple power-law bound
The simplest bound to obtain is an upper bound for $\ell(t)$ in the short and intermediate range regimes: i.e., in one dimension, $\mu >1$. Define $\E(t)$ to be the right-most edge of the occupied region at time $t$, i.e. $\c(x,t) =0$ for $x>\E(t)$. The probability of a jump that fills a position $y>E(t)$ in $(t,t+dt)$ is less than $\int_{-\infty}^{\E(t)} dx G(y-x) \sim 1/(y-\E(t))^\mu$. For $\mu>1$, the lower extent of the integral can be taken to $-\infty$ as the jumps arise, predominantly, from points that are not too far from the edge. \[In contrast, for $\mu<1$ jumps from the whole occupied region are important and this bound would yield a total jump probability to long distances that diverged when integrated over $y$, and one would have to instead use a lower extent of the integral of $-\E(t)$ for the left edge.\]
The advancement of the edge is bounded by a translationally and temporally invariant process of jumps of the position of the edge by distances, $\Delta\E$, whose distribution has a power-law tail. For $\mu>2$, the mean $\langle \Delta\E \rangle <\infty$, implying that the edge, and hence $\ell(t)$, cannot advance faster than linearly in time. But for the intermediate regime, $\langle \Delta\E \rangle =\infty$ so that $\E(t)$ could advance as fast as a one-sided Levy flight with $\E(t) $ dominated by the largest advance. As this process would yield $\E \sim t^{1/(\mu-1)}$, this implies that $\ell(t)$ is bounded above by the same form as the heuristic result.
Although the simple bound captures some relevant features, in particular the dominance of jumps of length of order $r$ to fill up a point at distance $r$, it is otherwise rather unsatisfactory. First, the coefficient does not vanish rapidly as $\mu \searrow 1$. And second, it suggests that the probability that an anomalously distant point, $r\gg\ell(t)$, is occupied, is, in this crude approximation of full-occupancy out to the edge, simply the probability that $\E(t)>r$ which falls-off only as $1/r^{\mu-1}$ — much more slowly than the actual $\q(r,t) \sim r^{-1-\mu}$.
Nevertheless, for proving better upper-bounds, the Levy-flight approximation for the dynamics of the edge is quite useful.
### Upper bounds from source-jump-target picture
As discussed earlier, one would like to make the heuristic argument of a single long jump from a source region to a target funnel region include also — or provide solid reasons to ignore — the effects of jumps from the partially filled region outside the core of the source. Very loosely, one would like to write the probability that a point, $R$, is not occupied at time $T$, as 1-(R,T) with $x$ in the source region and $y$ in the funnel of $R$. But for any positive $\mu$, the spatial integral is dominated by $y-x$ small, so that this does not properly represent the process: there is a drastic over-counting of short jumps.
One can do much better by trying to separate the long jumps from the short ones, and the source region from the funnel (in the crude approximation these overlap). To do this, we choose, for the $R$ and $T$ of interest, a spatio-temporal source region, $S$, around the origin which has a boundary at distance $\BS(t)$ that loosely reflects the growing source: $d\BS/dt \ge 0$. We then separate the process of the set of jumps that lead to $R$ into three parts. First, jumps solely inside $\S$ which lead to a spatio-temporal configuration of occupied sites, $\{\cs(x,t)\}$; second, [*bridging*]{} jumps from these out of $\S$, say at time $t$ from $x$ in $S$ to a point $y$ in the rest of space-time, $\bar{\S}$, and third all the subsequent dynamics from such seeds in $\bar{\S}$ including inside and outside $\S$ and between these. This over-counts the possible spatio-temporal routes to $R,T$ — especially as returns to inside $\S$ from outside are included — and thus provides an upper bound for $\q(R,T)$. The probability, $p_a$, that a single seed, $a$, to $y_a$ at $t_a$ leads to $R$ being filled by $T$ is $\q(R-y_a, T-t_a)$. But the probability that a second seed, $b$, leads to $R$ filled by $T$ is not independent as the fate of these seeds involves overlapping sets of jumps: indeed, they are positively correlated so that . Since for a given $\{\cs(r,t)\}$, the probability density of a seed at $y,t$ is $dydt \int_{|x|<\BS(t) }\cs(x,t)G(y-x)$, and using the generalization of the above bound to many seeds, we have (R,T) 1- \[q-up-iter\] where $\qs(x,t)\equiv\langle \cs(x,t) \rangle$ and we have used $\langle \exp(X) \rangle \ge \exp(\langle X \rangle)$ for any random variable.
To derive a useful upper bound on $\q(R,T)$ we need to choose appropriately the boundary, $\BS(t)$, of the source region and put a sufficiently stringent upper bound on $\qs(x,t)$.
### Long-range case
For the long-range case, $\mu<1$, the integrals over $x<\BS$ and $y>\BS$ of $G(y-x)$ are dominated by long distances. Thus the short jumps from inside to outside $\S$ do not contribute significantly. We can then simply replace $\qs$ by the larger $\q$ to obtain a slightly weaker bound which is of exactly the form of the naive estimate except for the strict delineation of the source region which prevents the most problematic over-counting of the effects of short jumps. A particularly simple choice is $\BS =\frac{1}{2}R$ independent of $t$.
We now proceed by induction. Take the bound on the scaling function to have the form $\FG(r,t)=1$ for $r<\LG(t)$ while $\FG(t)= \lb[\frac{\LG}{r}\rb]^{\mu+1}$ for $r \gg \LG(t)$ and assume that for some appropriate $\LG(t)$, this is indeed an upper bound for all $t<T$: i.e., $\q(r,t) \le \FG[r/\LG(t)]$. We can now use with $\qs$ and $\q$ both replaced by $\FG$.
When $\LG(t)$ and $\LG(T-t)$ are both much less than $R$, the integrals over $x$ and $y$ will be dominated by the regions near the origin and $R$ respectively, yielding the spatial convolution $ \FG \circ G \circ \FG \sim \frac{\LG(t)\LG(T-t)}{R^{\mu+1}}$. There are small positive corrections to this from two sources: first, from the regions near $0$ and $R$, which, by expanding $y-x$ in $x$ and $R-y$, are seen to be of order $ [\LG(t)^2+\LG(T-t)^2]\LG(t)\LG(T-t)/R^{\mu+3}$; and, second, from $y-x \ll R$, the region near the source boundary, which are of order $\lb[\frac{\LG(t)}{R}\rb]^{\mu+1}\lb[\frac{\LG(T-t)}{R}\rb]^{\mu+1}R^{1-\mu}$ with the last part from the integrals over $x$ and $y$. As the dominant part is exactly of the form in the heuristic treatment, integrating it over time is strongly peaked at $t\approx T/2$ (note that for either $t$ or $T-t$ much smaller than $T$, one of the $\FG$ factors will be close to unity near the boundary, but these ranges of time only contribute weakly). If we use $1-e^{-Q} \le \min(Q,1)$ then $\LG(T)$ can be chosen as the value of $R$ for which $Q=1$, and for $R\gg \LG(T)$ the desired $\FG \sim \lb[\frac{\LG}{R}\rb]^{\mu+1}$ is obtained. Including the small correction factors in the convolutions necessitates slight modifications of the recursion relations for $\LG$ but these are negligible at long times.
### Intermediate-range case {#Sec:intermediate-range-case}
Obtaining an upper bound in the intermediate range case is somewhat trickier. If we again replaced $\qs$ by $\q$, then the integrals over $x$ and $y$ would have a part dominated by both points being near the boundary: with $\BS \sim R$, this contribution to the convolution would be of order $\lb[\frac{\LG(t)}{R}\rb]^{\mu+1}\lb[\frac{\LG(T-t)}{R}\rb]^{\mu+1}$. With $t \sim T/2$ and $R\sim \ell(T)$, all the lengths should be of order $t^\beta$, so that this boundary piece is larger by a factor of $[\ell(T)^\beta]^{\mu-1} \sim T$ than what-should-be the dominant part from $x$ and $y$ near $0$ and $R$, respectively. Thus we need a better upper bound on the restricted-source $\qs(r,t)$ which vanishes as $r \nearrow \BS(t)$.
To bound $\qs$, we can make use of the simple bound for the edge of the occupied region derived above, combined with the restrictive effects of the boundary, $\BS(t)$. Instead of choosing $\BS$ to be constant, we choose it to have constant slope, $U\equiv d\BS/dt$, of order $\ell(T)/T$. As jumps that contribute to $\qs$ are not allowed to cross the boundary, the distribution of jumps of the edge $\E(t)$ is cutoff at $\zeta(t)\equiv\BS(t)-\E(t)$. Since $U>v_0$, the speed of spread in the absence of jumps beyond nearest neighboring sites, typically the gap, $\zeta(t)$, will increase with time, decreasing only by jumps. The sum of all the jumps of $\E$ in a time interval $\Delta t$ is dominated by the largest, which is of order $(\Delta t)^{1/(\mu-1)}$. This would result in the edge moving faster than $U$ except for the cutoff. The typical gap, $\tilde\zeta$, is then obtained by balancing its steady decrease against the dominant jump: $U\Delta t \sim (\Delta t)^{1/(\mu-1)}$ yielding $\Delta t \sim U^{(\mu-1/)(2-\mu)}$ and hence \~U\^[1/(2-)]{} \~\^[1/(2-)]{} \~(T) A\_\^[(-1)/(2-)]{} using $\ell(t) \sim A_\mu t^{1/(\mu-1)}$. In the limit of $\mu \searrow 1$, $\tilde{\zeta}/\ell \sim 4^{-1/(\mu-1)}$, vanishing rapidly — a reflection of the strong failure of the simple edge-bound in this limit but sufficient for our present purposes. In a time $\Delta t \ll T$, the distribution of $\zeta$ in this approximation will reach a steady state. The probability that $\zeta \ll \tilde\zeta$ is controlled by the balance between jumps of $\E$ to near the boundary, and the steady increase in $\zeta$ from the boundary motion: its probability density is hence of order $\zeta/\tilde\zeta$ which, since in this approximation all sites are occupied up to $\E$, implies that $\qs$ vanishes at least quadratically for small gap $\zeta$. Combining this with the trivial bound of $\qs< \q$ and choosing a convenient normalization of $\tilde\zeta$, we thus have (r,t) . It remains to chose $\BS(t)$ so that the bound on $\qs$ remains sufficiently good for $t$ small enough that the steady state distribution of $\zeta(t)$ has not yet been reached. To keep $\E(t)$ typically of order $\tilde\zeta$ from $\BS(t)$, we can simply chose $\BS(0)=\tilde\zeta$ and $U= (R-2\tilde\zeta)/T$.
With our improved bound on $\qs$, for the convolution $\qs \circ G \circ \q$, the small $\BS-x$ parts are no longer dominated by $\BS-x$ of order unity, but by $\BS-x$ near the crossover point between the two bounds on $\qs$. This yields a contribution to the convolution of order $\LG(t)^{\alpha_S}\LG(T-t)^{\alpha_F} / R^{\alpha_S+\alpha_F +\mu -1}$ with $\alpha_F=(\mu+1)(2-\mu)$ and $\alpha_S=\alpha_F (3-\mu)/2$ and a multiplicative coefficient that does not depend exponentially on $1/(\mu-1)$ because the integral over $x$ scales as $1/\tilde{\zeta}^{\mu-1}$. As $\mu \searrow 1$, $\alpha_F\to \alpha_S\to1$, and the boundary contribution is less than the dominant part uniformly in $t$. Note that for $\mu>\mu_B \cong 1.5$, the bound on the near-boundary contribution can be somewhat larger for $t<T/2$ than the dominant parts, but it scales in the same way with $T$ and thus only weakens the upper bound on the coefficient, $A_\mu$, which is in any case of order unity in this regime.
Once the over-counting of short jumps has been sufficiently reduced, as we have now done, the rest of the analysis, in particular the large $R/\LG$ form of $\FG$, follows as in the long-range case.
The [*marginal case*]{} $\mu=1$ can be analyzed similarly to the intermediate-range case, resulting in an additional logarithmic dependence on $R$ of the near-boundary contribution which is, nevertheless, still much smaller than the dominant part.
The upper bounds that we have obtained are, except for modifications at small scales and for $\mu$ not much smaller than 2, essentially the same as given by the heuristic arguments, thus only differing at long scales by order-unity coefficients which, in any case, we did not expect to get correctly. All the crossover behavior near $\mu=1$ is in the upper bounds, although that near $\mu=2$ is not.
Lower bounds
------------
To obtain lower bounds on the growth of the characteristic length scale $\ell(t)$ and the occupation probability, $\q(r,t)$, a different strategy needs to be employed. One of the difficulties is the dependence on the behavior at each time scale on all the earlier time scales: for the filled region to grow typically between time $T$ and $2T$, the stochastic processes that lead to the configuration $\c(x,T)$ must not have been atypically slow or ineffective. As this applies iteratively scale by scale, we must allow for some uncertainty in whether or not the smaller scale regions are typical, leading to some uncertainty at all scales which, nevertheless, we need to bound. Because of the stochastic heterogeneity of $\c(x,t)$ it is better to focus on a coarse-grained version of the occupation profile rather than on $\c(x,t)$ itself, as integrations over $\c$ at time $t$ are what act as the sources of future occupation at larger distances.
### Mostly-filled-in: marginal and long-range regimes
We consider the probability that a region is almost full, in particular, with a seed at the origin, we consider the region to one side of the origin and define (r,t;) with $\Phi$ close to or equal to unity being of particular interest. In order to keep events sufficiently independent, we consider, as for the upper bounds, the probability of events that do not involve any jumps out of some region. In particular, defining $\PS(r,t,\Phi)$ similarly $\PF$, but with the restriction that jumps do not go out of the interval $(0,r)$. For the long-range and marginal cases, we will focus on partial filling, but for the intermediate range case the scale invariance mandates different treatment so we instead analyze full filling – i.e. $\Phi=1$.
The basic strategy is to start with a particular deterministic approximation to $\ell(t)$, $\LT(t)$ with corresponding times $\TT(r)$, and then show that at time not too large a multiple of $\TT(r)$, the region out to $r$ will be nearly filled with high probability: i.e. that $\PS(r,\TG(r);\Phi)$ is close to unity for $\TG(r)/\TT(r)$ sufficiently large. We will be interested in large scales as, in any case, fluctuations at the small scales can only change coefficients by order unity. We can thus be sloppy with some of the bounding inequalities: these could be improved to include the ignored corrections to the large scale effects to make fully rigorous bounds.
As the range of time over which the typical $\ell(t)$ expands significantly plays an important role, it is useful to define (t) \^[-1]{} which is small except for $\mu$ substantially larger than one. The dominant jumps from source to funnel involve an integral over time of $\LT(t)\LT(T-t)$, which is primarily from a range of order $T\sqrt{\DT}$ around $T$ as discussed above. The deterministic-iterative approximation that we use as a base for the lower bounds is the solution to the iterative relation (ambiguous up to an ${\cal O}(1)$ multiplicative factor which we ignore throughout): \^[+1]{}= T\[(T)\]\^2 corresponding to roughly one seed into a funnel of width $\LT(T)$ from a jump of distance $\LT(2T)$ from the source up to time $T$. For convenience, we use only half the source — $x$ from $0$ to $\LT(T)$. The results, $\LT(t)$, of this iterative approximation are, up to numerical factors that arise from these modifications and from other from non-asymptotic effects at small scales, equivalent to the upper bounds, $\LG(t)$ from the above. In particular, the ratio between the corresponding times, $\TT(r)$ and $\TL(r)$ we expect to approach constants that are not singular near the marginal case $\mu=d=1$.
For the lower bounds it is convenient to work with a specific set of length scales, $\LT_n=\LT(\TT_n)$, corresponding to a series of time scales, $\TT_n=2^n$ (dropping a prefactor). To mostly fill out to $\LT_{n+1}$ without jumps going out of $(0,\LT_{n+1})$ from the source of size $\LT_n$, most of the $\K_n = \LT_{n+1}/\LT_n $ bins of size $\LT_n$ must be mostly filled. To get a lower bound on how long this takes and how likely it is, we make several simplifications each of which lead to underestimates of the probability that the desired filling has occurred. First, consider only jumps into each bin that come directly from the source (rather than from other bins as can occur later). Second, ignore all but the first seed jump from the source into the bin (the effects of later jumps are not independent of those of the first). And third, include only jumps that lead from the seed in a bin that do not go outside that bin during the time during which the probability of it being mostly filled is considered. The last two conditions mean that the probability that the bin is filled to a fraction $\Phi$ by a given time, $t$, after the seeding jump, is at least as large as $\PS(\LT_n,t;\Phi)$ since a seed at the edge of the bin, which corresponds to the definition at the source, is less likely to mostly fill the bin than a seed away from the edge.
At large scales for $\mu\le 1$, the number of bins, $\K_n$, grows with scale: $\K_n \sim \sqrt{\TT_n}$ for the marginal case and larger for the long-range case. Thus if the probability that the furthest bin from the source is mostly filled is $f_n$, with the filling of the others being more probable as they are closer, it is likely that the number that are similarly mostly filled is close to $\K_n f_n$, with significant deviations from this being very unlikely at large scales. In order to iterate while not losing too much in filling fraction, we chose a series of partial filling fractions, $\{\phi_n\}$, such that $\Phi_N\equiv\prod_{n=1,N-1} \phi_n $ converges to the desired overall filling fraction, $\Phi$, at large $N$, and chose conditions such that $f_n$ is sufficiently large that the fraction of the $\K_n$ bins filled to $\Phi_n$ is greater than $\phi_n$ with high probability: this then implies that the region from the origin to $\LT_{n+1}$ will be filled to greater than $\Phi_{n+1}$ with high probability. A convenient choice is $\phi_n = 1- \Delta/n^{1+\alpha}$ with any positive $\alpha$ and $\Delta \sum_n n^{-1-\alpha} < 1-\Phi$. For convenience in dropping $\log\Delta$ factors that otherwise appear in many places, we restrict consideration to $\Delta$ not very small, and do not keep careful track of $\alpha$ factors that also appear as one can take $\alpha \to 0$ at the expense of corrections that are down by one extra logarithm.
The filling probability of a bin is at least as large as that obtained from the requirement of the occurrence of both of two independent events: a jump into the bin from the source that occurs before some chosen initial time, $T_I$, and the bin being filled from that single seed by a time, $T_B+T_I$. The probability of a jump into a bin is at least $1-e^{-W_n}$ in terms of a conveniently chosen lower bound, $W_n$, on the expected number of jumps from the source into the furthest away bin, and the probability of the bin being filled from the single seed is at least $\PS(\LT_n,T_B;\Phi_n)$. We will find iterative bounds on $\PS$ that it is convenient to write in the form (r,t;) 1 - e\^[-(r,t;)]{} so that 1-f\_n e\^[-W\_n]{} + e\^[\_B]{} [with]{} \_B(\_n,T\_B;\_n) . For convenience we chose conditions so that $\Lambda_B \ge W_n$ and $1-f_n \le \frac{1}{2}(1-\phi_n)$ which, for $K_n$ large, makes the probability that a fraction $\phi_n$ of the bins are not filled exponentially small. We henceforth ignore this factor in the probability as it does not matter except on small scales: adjustments to take it into account are straightforward. We thus require that \_B W\_n = (1+) n + [O]{}(1) .
To obtain a bound on $\PS(\LT_{n+1},T;\Phi_{n+1})$, we must show that a source that can give rise to an average number at least $W_n$ of jumps into the furthest bin by time $T_I$, occurs with probability that is somewhat larger than the desired bound at the next scale. The expected effective number of jumps out of the source of size $\LT_n$ into a bin of the same size a distance up to $\LT_{n+1}$ away before time $\TT_n$ was assumed in the deterministic iterative approximation to be of order $\sqrt{\DT_n}\TT_n$. In order to ensure that the average number from the actual source is sufficiently large, we can require that it be almost filled by some time $T_S$ and only include jumps that occur between $T_S$ and $T_I$ as the rate of these is bounded below by the filling at $T_S$. The required range is T\_I-T\_S=W\_n\_n/\_n .
We now proceed by induction and show that if (\_n,t;\_n) \_n(t-U\_n\_n) for $t$ in a range such that $\Lambda$ is relatively large — the precise range is not crucial but minor modifications are needed to extend out to arbitrary large $t$ — then a similar bound holds at the next scale with coefficients $\gamma_{n+1}$ and $U_{n+1}$ with both these varying slowly with $n$ at large scales. Note that at the smallest scale the probability that a site is filled by a jump directly from the origin by time $t$ converges exponentially to unity for long $t$, thus at the smallest scales there is a trivial bound of this form. As the scale is increased, $\gamma_n$ will initially change, but once the scale becomes large enough that the width of the distribution of the fraction of the bins mostly filled is small, then $\gamma_n$ saturates and becomes weakly dependent on $n$. In the analysis below, it can be replaced by a constant.
Consider a total time $T$ to mostly fill out to $\LT_{n+1}$. The time for the bins to fill with sufficiently high probability once they have been seeded is $T_B\le U_n\TT_n +W_n/\gamma_n$. With $T_I -T_S$ as above, we have a time for the source to fill T\_S T - \_n\[U\_n +W\_n/\_n)\] -W\_n/\_n . Plugging in the probability that the source is filled in this time gives a bound on $\Lambda(T, \LT_{n+1},\Phi_{n+1})$ of the same form but with, dividing out $\TT_{n+1} = 2\TT_n$, U\_[n+1]{} U\_n + + . \[U-itern\] As $\TT_n$ increases rapidly and $W_n$ only slowly, the last term only contributes at small scales.
For the [*long-range regime*]{}, $\DT_n \sim e^{-\eta \log 2 n}$ so the second term in is also small except at small scales and we conclude that $U$ is bounded above by a $\mu$ dependent constant. Thus the lower bound for $\ell(t)$, and upper bound for $\tau(r)$ have exactly the same form as the opposite bounds, except with the scale of $t$ — i.e. $B_\mu ^{-1/\eta}$ — different.
For the [*marginal case*]{}, $\DT_n\approx \frac{2}{n}$ so that $U_n$ changes slowly at large scales. Integrating up, one sees that U\_n < C n \~ \_n with a coefficient independent of $n$ (but depending on $\Phi$ and $\alpha$). One can now solve for the time scale above which mostly filled is likely, $\TG(r)=U(\TT(r))\TT(r)$, to find a lower bound, $\LL(t)$, on $\ell(t)$: ((t))((t))= \[t - 2 t - [O]{}(t)\] \[low-bnd-marginal\] which is very close to the upper bound derived above, ((t))((t))= \[t - t - [O]{}(1)\] differing only in the coefficient of the correction term.
One of the advantages of this iterative approach is that the [*crossover regime*]{} can be handled similarly by integrating up . The lower bound will be similar to the upper bound throughout this crossover regime and into the asymptotic regimes for the marginal and long-range cases.
### Fluctuations and intermediate-range regime
The reason that the fluctuation effects are relatively small for the marginal and long-range regimes is that at each successive time scale, more and more roughly independent long jumps are involved in filling up to the next length scale: i.e. $K_n$ continues to grow. For the long-range regime, it grows so rapidly that almost all the fluctuations come from early times: this is like what occurs for the fully mixed model. For the marginal case, the fluctuations are dominated by the smallest scales but the cumulative effects of them over the longer scales does make a difference as found in obtaining the lower-bounds.
For the intermediate-range case, the ratio of length scales for each factor of two in time scales saturates (when out of the crossover regime) at $K\approx 2^\beta$. This means that whatever fraction, $\phi_n$, of the bins are to be filled at each scale, the probability that this occurs either decreases with scale if the product of the $\phi_n$ does not go to zero, or saturates to a constant if the $\phi_n$’s do also, in which case the overall filling fraction $\Phi_n$ tends to zero as a power of time. At each scale there are a comparable number of long jumps that are needed, thus one should expect that fluctuation effects will be scale invariant and not decrease with scale.
To get useful lower bounds on $\ell(t)$ via an upper bound on $\tau(r)$, $\TG(r)$, the easiest way is to require that the source be completely full, and that jumps from this completely fill all the bins at the next scale: this avoids the problems with the $\phi_n$. As the probability that all the bins are filled is only (readily) bounded by $ (1-e^{-W} -e^{-\Lambda_B})^K\approx 1-2Ke^{-W}$ if we again chose $\Lambda_B \ge W$, $W$ must be larger by $\log K \approx \beta \log 2$ than for the partially filled analysis above. Carrying through similar analysis, gives for large $\beta$ a coefficient $\gamma_n \approx 2\sqrt{\beta}/(n\TT_n)$ which means that the filling probability decays for large times as roughly the inverse of the typical time — natural as the needed long jumps that occur at rate $\sim 1/\TT_n$ have a distribution of when they occur on the same time scale. Note the contrast to the rapid decay of the not-mostly-filled probability on a time scale of order unity from the partially filled bound derived above. The time beyond which the full filling is likely is only bounded by, in this analysis, $\TT_n U_n \sim \TT_n \sqrt{\beta} n^2 $. This gives a lower bound on $\ell (t) $ proportional to $t^\beta/\log^{2\beta}t$. Although on a logarithmic scale the additional factor is smaller, we would like to do better.
The bound can be improved by considering a source that is somewhat smaller — by a factor of two is sufficient – than $\LT_n$ which increases the probability that it is filled, but means that more extra time, $T_I-T_S$, is needed to produce a mean number of jumps to the furthest bin of at least $W$. Using that $U$ and $\gamma \TT$ vary slowly with scale, one can expand $\Lambda(r,t)$ around $r=\LT_n$ and analyze the changes on the bounds at the next scale. This improves the bounds to $\gamma \sim 1/(\sqrt{\beta}\TT)$ and $U \sim \beta^{5/2}$ with $\beta=1/(\mu-1)$. The resulting lower bound on $\ell(t)$ is (t) > \^[-]{} t\^with the coefficient $\tilde{A_\mu} \sim 4^{\beta^2}$ that from the deterministic iteration which is the same, up to an order unity pre factor, as the upper bound. It is not clear where between the lower and upper bounds on the coefficient will be the typical behavior, nor how broad the fluctuations will be — even on a log scale.
The behavior as $\mu \nearrow 2$ we have not analyzed explicitly, instead focussing on the rapidly growing regime for $\mu \searrow 1$, but the bounds will be of similar form although more care is needed to get upper and lower bounds reasonably close to one another due to the important jumps being only a modest fraction of the size of the already occupied region.
For the marginal case, $\mu=1$, one can find an upper bound on the time at which a the region out to $r$ is likely to be fully filled by similar methods to that for the intermediate-range power-law regime. This yields a bound $\LL^{\Phi=1}(t)$ of the same form as that above for partial filling ($\Phi<1$),, except with the coefficient of the $\log\log t$ term in equal to 6 instead of 2. The convergence of the probability of being fully filled is, however, much slower for this bound on compete filling than for the bound on being mostly filled. While the latter converges for $t>\TG^{\Phi}(r)\sim \TT(r)\sqrt{\log\TT(r)}\log\log\TT(r)$ with a rate of order unity — dominated by the small scales — the former converges as the time increases above $\TG^{\Phi=1}(r)\sim \TT(r) \log^{\frac{5}{2}}\TT(r)$ on a time scale, $1/\gamma$, of order $\TT(r) \sqrt{\log\TT(r)}$ — faster than $\TG^{\Phi=1}(r)$ but not much so.
Note that the convergence of the probability for being mostly filled to more than a fixed filling fraction, $\Phi$, is a hybrid property: the probability of a fixed site being filled by time $t$, $q(r,t)$, is bounded below by (roughly) the product of $\Phi$ and the probability that the region out to $r$ is filled to above $\Phi$. To get the convergence of this to unity, $\Phi$ needs to be adjusted and the thus-far ignored $\log(1-\Phi)$ factors kept track of. This also necessitates treating intermediate scales differently as the number of bins that do not need to be filled, $(1-f_n)K_n$ is not large. However a different approach would provide a better bound: focussing on a specific site being filled with high probability can be done by a method more analogous to the funnel picture of the main text: for the site to be filled, it needs to be in a small-scale bin that is mostly but not-necessarily fully filled with high probability, which needs itself to be in a larger bin similarly, etc. But these can be filled from source regions that are not fully filled: being partly filled with high enough probability is sufficient. We have not carried out such analysis in detail in part because the actual mechanism by which sites that are empty for anomalously long will be filled is more complicated as it will involve filling from nearby regions on a hierarchy of scales that were filled at more typical times.
The analyses here can be immediately extended to give lower bounds on the average density profile at long distances: these will be of the same form as the lower bounds, thus demonstrating that the predicted $\frac{\ell(t)}{r}^{-(1+\mu)}$ is essentially correct.
### Comparisons with results of Chatterjee and Dey
As noted in the main text, when this work was essentially complete, a preprint by Chatterjee and Dey (CD) appeared which derives and proves some results closely related to ours in the context of long-range first passage percolation which is essentially equivalent to the lattice dispersal model, with the jump kernel $G(r) \sim r^{-\alpha}$ equivalent to the $1/r^{-(d+\mu)}$ that we use [@chatterjee2013multiple]. While some of the quantities CD focus on are different, the leading asymptotic scaling behaviors they obtain are essentially the same, and their proofs apply in all dimensions. Our $q(r,t)$ corresponds to the probability that the first passage time $T^F(r)$ is less than $t$ and their diameter, $D(t)$ — the maximum distance between any pair of occupied points at time $t$ — is, with high-probability that decays as a power of $ \ell/D$ — not many times $\ell(t)$, as we both obtain.
However CD’s results are sub-optimal. In particular, for the coefficient, $C$, of $\log\ell(t)/\log^2 t$ in the marginal case, they only obtain upper and lower bounds instead of our exact result $C=1/4d\log 2$. Indeed, in one dimension we obtain upper and lower bounds on the errors: 1-c\_- < < 1- with high probability — in senses that can be made precise from our analysis — with the coefficient $c_-$ either 2 or 6 depending on the definition of $\ell(t)$ used. In the intermediate range power-law growth regime, CD’s theorems do not appear to exclude $\log t$ pre-factors in $\ell(t)$, although their analysis might well do so. But the main difference is our analysis of the whole crossover regime for $\mu$ near $d$, including the divergences and vanishings of coefficients of the asymptotic forms, which they do not consider. These are crucial for comparisons with simulations because of the very long length scales of the crossovers.
To turn our upper and lower bounds into formal proofs in one dimension requires primarily filling in some details associated with the small scale regime. For higher dimensions some additional work is needed, although the strategies should work without major modifications.
Supplementary Figures {#sec:suppl-figur}
=====================
\[ht!\] ![The pre-factor $A_\mu$ of the predicted power law growth in Eq. (3) (main text) in one dimension: $\ell(t)\approx A_\mu
t^\beta$ with $\beta=1/(\mu-1)$ for $1<\mu<2$. The blue curve is obtained numerically from solving equation (1) (main text) with the power-law ansatz; the red-curve represents an analytic approximation derived in “Crossovers and Beyond Asymptopia” (main text). Notice the sharp (non-analytic) drop of the pre-factor as $\mu$ approaches $1$. The reason is very slow cross-over to the power law from an intermediate asymptotic regime controlled by the dynamics of the marginal case. As $\mu$ approaches 2, the pre-factor diverges as $A_\mu\sim (2-\mu)^{-1}$ indicative of another slow crossover at $\mu=2$, see Fig. \[fig:second-marginal-point\]. \[fig:prefactors\]](prefacslabeled.png "fig:"){width="\columnwidth"}
\[ht!\] ![Dynamics of growth in one dimension at the marginal point between superlinear and linear growth, $\mu=2$. The number of mutant sites, $M(t)$, scaled by time, $M(t)/t$, is plotted as a function of time, averaged over 10 realizations (black) and for two individual realizations (red and green). While the averaged data suggests $M(t)\sim t \ln(t)$, the individual realizations indicate strong fluctuations caused by occasional rare jumps, which are of order $t$. These “leaps forward” [@mollison1972rate] are driving the logarithmic increase of the spreading velocity. \[fig:second-marginal-point\]](second-marginal-point-annot.png "fig:"){width="\columnwidth"}
|
{
"pile_set_name": "ArXiv"
}
|
---
address: |
Department of Physics, Tokyo Metropolitan University,\
Hachioji-shi, Tokyo 192-03, Japan
author:
- 'Nobuchika Okada [^1] [^2]'
title: |
A Simple Model of Gauge Mediated Supersymmetry Breaking\
with Composite Messenger Fields
---
2.5cm
[Abstract]{} 0.7cm
We present a simple model of gauge mediated supersymmetry breaking with composite messenger fields. Our model is based on the gauge group $SP(8) \times SU(2)$. By the strong $SP(8)$ dynamics, supersymmetry is dynamically broken and the composite fields with charges under the standard model gauge group appear at low energy. The $U(1)_R$ symmetry breaking mass terms for the composite fields are generated by the strong $SU(2)$ dynamics. Then, the composite fields play a role of the messenger fields. On the other hand, the theoretical bounds on the parameters in our model are discussed. Especially, the lower bound on the dynamical scale of the $SP(8) \times SU(2)$ gauge interaction is roughly $10^{15}$ GeV.
The models of gauge mediated supersymmetry breaking (GMSB) have attractive feature in the minimal supersymmetric standard model (MSSM). Since supersymmetry breaking is mediated to the MSSM sector by the standard model gauge interaction through the messenger fields which are charged under the MSSM gauge group, the superpartners with the same charges in the MSSM get the same soft supersymmetry breaking masses. As a result, the problem of the flavor changing neutral current in the MSSM are resolved naturally.
The pioneering works have been done by Dine, Nelson and co-workers [@dine]. They have constructed explicit models which realized the mediation of supersymmetry breaking to the MSSM sector. Furthermore, it has been shown that the models was phenomenologically viable.
However, the original models were very complicated. This fact originates from the complexity of the dynamical supersymmetry breaking mechanism. In addition, introduction of three separated sectors, the supersymmetry breaking sector, the messenger sector and the MSSM sector, make the models more complicated.
Several attempts to obtain more simple GMSB models have been considered by many authors. A simple mechanism of the dynamical supersymmetry breaking has been proposed by Izawa and Yanagida, and Intriligator and Thomas [@Yanagida], and this mechanism was applied to the supersymmetry breaking sector in the GMSB models [@hotta]. Moreover, new types of the GMSB models in which the messenger sector is unified into the supersymmetry breaking sector have been constructed [@composite].
In this letter, we present a simple GMSB model based on the gauge group $SP(8) \times SU(2)$. Supersymmetry is dynamically broken by the strong $SP(8)$ gauge dynamics. Since the standard model gauge group $SU(5)_{SM}\supset SU(3)_c \times SU(2)_L \times U(1)_Y$ is embedded in the global symmetry $SU(10)$ which the $SP(8)$ gauge dynamics has, the messenger sector is unified into the supersymmetry breaking sector and the messenger fields appear as composite fields at low energy. The strong $SU(2)$ gauge dynamics generates the $U(1)_R$ symmetry breaking mass terms for the messenger fields.
Before discussing our model, let us review the messenger sector. The typical superpotential is simply described by $$\begin{aligned}
W_{mess} = \sum_i \lambda_i Z_i \bar{\Phi} \Phi \; \; ,
\label{mess}\end{aligned}$$ where $\bar{\Phi}$ and $\Phi$ have the vector-like charge under the MSSM gauge group, $Z_i$ is a singlet field under the gauge group, and $\lambda_i$ is a dimensionless coupling constant. If nonzero vacuum expectation values of the $F$-component of at least one $Z_i$ and the scalar component of at least one $Z_j$ are realized, the fields $\bar{\Phi}$ and $\Phi$ can play a role of the messenger fields. Note that $i=j$ is not needed in general.
Our model is based on the gauge group $SP(8) \times SU(2)$ as mentioned above. To make our discussion clear, let us consider only the $SP(8)$ dynamics at first. The particle contents are as follows.
--------------------------------------------------------------------------------------------------------
$~SP(8)~$ $~SU(5)_{SM}~$ $~U(1)_R~$
------------- ------------------------------------------------------------ ---------------- ------------
$ \bar{P} $ **[8]{} & $\bar{\bf{5}}$ & 0\
$ P $ & **[8]{} & **[5]{} & 0\
$ Z $ & **[1]{} & **[1]{} & 2\
$ Z^{\prime}$ & **[1]{} & **[1]{} & 2\
$ \bar{\phi}$ & **[1]{} & $\bar{\bf{10}}$ & 2\
$ \phi $ & **[1]{} & **[10]{} & 2\
$ \bar{A} $ & **[1]{} & **[24]{} & 2\
$ N $ & **[1]{} & **[1]{} & 0****************************
--------------------------------------------------------------------------------------------------------
Note that the standard model gauge group $SU(5)_{SM} \supset SU(3)_c \times SU(2)_L \times U(1)_Y$ is embedded in the global symmetry $SU(10)$ which the $SP(8)$ dynamics has. In this paper we always use the notation of the ordinary $SU(5)$ Grand Unified Theory (GUT), for simplicity. It is trivial to decompose it into the standard model notation.
A renormalizable tree level superpotential which is consistent with all the symmetry is given by [^3] $$\begin{aligned}
W_{tree}= \lambda_Z Z \; \left[\bar{P} P\right]_s
+ \lambda_{Z^\prime} Z^\prime ( \left[\bar{P}P\right]_s - \lambda_N N^2 )
\nonumber \\
+ \lambda_{\bar{\phi}} \bar{\phi} \; \left[PP \right]
+ \lambda_{\phi} \phi \; \left[\bar{P} \bar{P}\right]
+ \lambda_{\bar{A}}\;
\mbox{tr} (\bar{A}\; \left[\bar{P} P\right]_{adj}) \; \; ,
\label{tree1}\end{aligned}$$ where square brackets denote the contraction of the $SP(8)$ indices, and $\left[\; \right]_s$ and $\left[ \; \right]_{adj}$ denote to extract a part of singlet and adjoint representation of $SU(5)_{SM}$ from $\left[ \bar{P} P \right]$, respectively. Here, we assume that the tree level superpotential has no dimensionful parameter. As can be seen in the following, in our model, all of the dimensionful parameters are dynamically generated and originate from strong gauge dynamics.
We can obtain the low energy description of this theory by the method of Seiberg and co-workers [@seiberg]. The moduli space is dynamically deformed to satisfy the condition Pf$V=\Lambda^{10}$, where $V$ is $10 \times 10$ antisymmetric tensor given by $$\begin{aligned}
V = \left[
\begin{array}{cc}
\left[\bar{P} \bar{P}\right] & \left[\bar{P} P \right] \\
\left[P \bar{P}\right] & \left[ P P \right] \\
\end{array} \right]
\sim \Lambda \left[
\begin{array}{cc}
\bar{\Phi} & S + A \\
-S - A & \Phi \\
\end{array} \right] \; \; . \end{aligned}$$ Here, $ \Lambda $ is the dynamical scale of the $SP(8)$ gauge interaction. The fields $S$, $A$, $\bar{\Phi}$ and $\Phi$ are the effective fields as follows.
------- --------- ------------------------------------ ---------------------------------------------------------------------------------------
$ ~SU(5)_{SM}~ $
$ S $ $\sim $ $\left[\bar{P}P\right]_s/\Lambda $ **[1]{}\
$ A $ & $\sim $ &$ \left[\bar{P}P\right]_{adj}/\Lambda$& **[24]{}\
$ \bar{\Phi} $ & $\sim $ & $\left[\bar{P} \bar{P}\right]/\Lambda $ & $\bf{\bar{10}}$\
$ \Phi $ & $\sim $ & $ \left[PP\right]/\Lambda $ & **[10]{}******
------- --------- ------------------------------------ ---------------------------------------------------------------------------------------
Since the condition Pf$V=\Lambda^{10}$ contradicts the supersymmetric vacuum conditions required by the tree level superpotential of eq.(\[tree1\]), supersymmetry is dynamically broken [@Yanagida].
To obtain the effective superpotential at low energy, we should eliminate one of the effective fields by considering the condition Pf$V=\Lambda^{10}$. Using the effective fields, the condition is described by $$\begin{aligned}
S^5 - S^3 \left( \bar{\Phi} \Phi + \frac{1}{2} \mbox{tr}A^2\right)
+ \frac{1}{3}S^2 \mbox{tr}A^3
+ S \left\{ \left(\bar{\Phi} \Phi\right)^2 - \frac{1}{4}
\mbox{tr} A^4 \right\}
-\Phi^2 A {\bar{\Phi}}^2 + \frac{1}{5} \mbox{tr} A^5
= \Lambda^5 \; \; .\end{aligned}$$ Considering small fluctuation of $S$ around $\langle S \rangle = \Lambda$, we can obtain $$\begin{aligned}
S \sim \Lambda + \frac{1}{5 \Lambda}
\left( \bar{\Phi} \Phi + \frac{1}{2} \mbox{tr} A^2 \right) \end{aligned}$$ Eliminating $S$ from eq.(\[tree1\]), the effective superpotential is given by $$\begin{aligned}
W_{eff} &\sim & \lambda_Z Z
\left\{ \Lambda^2+ \frac{1}{5}
\left( \bar{\Phi} \Phi + \frac{1}{2} \mbox{tr} A^2 \right) \right\}
\nonumber \\
&+& \lambda_{Z^\prime} Z^\prime \left\{ \Lambda^2 + \frac{1}{5}
\left( \bar{\Phi} \Phi + \frac{1}{2} \mbox{tr} A^2 \right)
- \lambda_N N^2 \right\} \nonumber \\
& + & \lambda_{\bar{\phi}}\Lambda \; \bar{\phi} \Phi
+ \lambda_{\phi} \Lambda \; \bar{\Phi} \phi
+ \lambda_{\bar{A}} \Lambda \; \mbox{tr} (\bar{A}A) \; \; . \label{eff1}\end{aligned}$$ This effective superpotential is one of the type of O’Raifeartaigh model [@O'R]. For small value of $\lambda_Z$ compared with $\lambda_{\bar{\phi}}$, $\lambda_{\phi}$ and $\lambda_{\bar{A}}$, supersymmetry is broken by $\langle F_Z \rangle = - \lambda_Z \Lambda^2$, where $ F_Z $ is the $F$-component of $Z$.
However, note that the scalar potential derived from eq.(\[eff1\]) has the ‘pseudo-flat’ direction, namely, the potential remains minimum along arbitrary value of $\langle Z \rangle$ [^4]. This ‘pseudo-flat’ direction is lifted up by quantum corrections for the effective potential of $Z$. There are two possibilities where the effective potential has minimum. One is $\langle Z \rangle \sim \Lambda$ which may be expected by the effect of the strong $SP(8)$ interaction [@Yanagida][@hotta]. The other is $\langle Z \rangle =0 $ which is expected only if the Yukawa coupling in eq.(\[eff1\]) is considered [@hqu]. Unfortunately, there is currently no technique to definitely decide which vacuum is chosen. In this letter, we assume that true vacuum lies at $\langle Z \rangle =0$.
Then, the vacuum is realized at $\langle F_Z \rangle \neq 0$, $\langle$other $F$-components$\rangle =0$, $\langle N \rangle = \Lambda / \sqrt{\lambda_N}$, and $\langle$other scalar components$\rangle =0$. Note that there is no $U(1)_R$ symmetry breaking mass term for $\bar{\Phi}$, $\Phi$ and $A$ in the effective superpotential, because of $\langle Z \rangle = \langle Z^\prime \rangle =0$. Therefore, the fields $\bar{\Phi}$, $\Phi$ and $A$ cannot play a role of the messenger fields. For example, the gauginos in the MSSM cannot get their soft supersymmetry breaking masses, since the masses are protected by the $U(1)_R$ symmetry.
In order to generate the $U(1)_R$ symmetry breaking mass terms for the fields $\bar{\Phi}$, $\Phi$ and $A$, we introduce new strong $SU(2)$ gauge interaction with two doublet fields $\bar{Q}$ and $Q$ which are singlets of $SU(5)_{SM}$. In addition to the effective superpotential of eq.(\[eff1\]), let us consider new tree level superpotential $$\begin{aligned}
W^{\prime}_{tree} = \lambda_M N \; \left[ \bar{Q} Q \right] \; \; ,
\label{tree2}\end{aligned}$$ where $\left[ \; \right]$ denotes the contraction of the $SU(2)$ indices by the $\epsilon$-tensor. Although this superpotential is the simplest one to attain our aim, the $U(1)_R$ symmetry is explicitly broken by the $SU(2)$ gauge anomaly. This may suggest that a modification of our model is needed. However, there is no R-axion problem because of this explicit breaking. The vacuum is realized with the same vacuum expectation values of the scaler fields discussed above and $\langle \bar{Q} \rangle = \langle Q \rangle =0$.
However, we should take into account the non-perturbative effect of the strong $SU(2)$ gauge interaction at low energy. When the effect is considered, the effective superpotential is given by [@ads] $$\begin{aligned}
W^{\prime}_{eff}= \lambda_M \Lambda^{\prime} N M
+ \frac{\Lambda^{\prime 4}}{M} \; \; ,
\label{eff2} \end{aligned}$$ where $\Lambda^{\prime}$ is the dynamical scale of the $SU(2)$ gauge interaction, and $M \sim \left[\bar{Q}Q \right] / \Lambda^{\prime}$ is the effective fields. Now we obtain the effective superpotential $\tilde{W}_{eff}= W_{eff}+W^{\prime}_{eff}$ as the total effective superpotential in our model.
Let us investigate where the vacuum is realized. The vacuum is changed and $\langle Z^{\prime} \rangle \neq 0$ occur by the strong $SU(2)$ dynamics. Indeed, from two conditions $\partial \tilde{W}_{eff}/ \partial M = 0$ and $\partial \tilde{W}_{eff}/ \partial N = 0$, we obtain $$\begin{aligned}
\langle M \rangle &=& \sqrt{\frac{1}{\lambda_M \langle N \rangle}}
\Lambda^{\prime 3/2}
= \frac{\lambda_N^{1/4}}{\lambda_M^{1/2}}
\frac{\Lambda^{\prime 3/2}}{\Lambda^{1/2}} \; \; , \nonumber \\
%
\langle Z^{\prime} \rangle &=&
\frac{\lambda_M}{2 \lambda_{Z^\prime} \lambda_N}
\frac{\langle M \rangle}{\langle N \rangle} \Lambda^{\prime}
= \frac{\lambda_M^{1/2}}{2 \lambda_{Z^\prime}\lambda_N^{1/4}}
\frac{\Lambda^{\prime 5/2}}{\Lambda^{3/2}} \; \; .
\label{vac}\end{aligned}$$ Then, the $U(1)_R$ symmetry breaking mass terms for the fields $\bar{\Phi}$, $\Phi$ and $A$ are generated. The effective superpotential corresponding to eq.(\[mess\]) is described by $$\begin{aligned}
W_{mess} = \frac{1}{5} (\lambda_Z Z + \lambda_{Z^\prime} Z^\prime)
\left( \bar{\Phi} \Phi + \frac{1}{2} \mbox{tr} A^2 \right) \; \; .
\label{mess2}\end{aligned}$$ Because of $\langle F_Z \rangle \neq 0 $ and $ \langle Z^\prime \rangle \neq 0 $, the composite fields $\bar{\Phi}$, $\Phi$ and $A$ can play a role of the messenger fields.
The mass spectra of all the superpartners in the MSSM are calculated by this superpotential [@martin] with $\langle F_Z \rangle$ and $ \langle Z^\prime \rangle $. The gauginos get their soft supersymmetry breaking masses through the one-loop radiative correction by the messenger fields $\bar{\Phi}$, $\Phi$ and $A$. For simplicity, let us assume $\lambda_Z \langle F_Z \rangle \ll
( \lambda_{Z^\prime} \langle Z^\prime \rangle)^2 $ and $ \lambda_{\bar{\phi}}\Lambda \sim \lambda_{\phi}\Lambda \sim
\lambda_{\bar{A}}\Lambda \ll \lambda_{Z^\prime} \langle Z^\prime \rangle $. Then, the masses of the gauginos are given by $$\begin{aligned}
m_{\lambda_a} \sim \frac{\alpha_a}{4 \pi}
\frac{\lambda_Z \langle F_Z \rangle}
{\lambda_{Z^\prime} \langle Z^\prime \rangle}
\sum_i n_a(i) \; \; ,
\label{gaugino} \end{aligned}$$ where $a=1,2$ and $3$ correspond to the MSSM gauge interaction, $SU(3)_c$, $SU(2)_L$ and $U(1)_Y$, respectively, and $n_a(i)$ is the Dynkin index for the messenger fields running the loop, which is defined as $n_a(i)=1$ for $i=\bf{N+\bar{N}}$ of $SU(N)$ and $n_1=6/5 Y^2$ for the messenger fields with the hypercharge $Y$ by using the $SU(5)$ GUT normalization. Since the messenger fields have the charge $\bf{10+\bar{10}}$ and $\bf{24}$ of $SU(5)_{SM}$, $\sum_i n_3 = \sum_i n_2 = \sum_i n_1 = 8 $. The scalar partners in the MSSM get their masses through the two-loop radiative correction. They are given by $$\begin{aligned}
\tilde{m}^2 \sim 2 \left( \frac{\alpha_a}{4 \pi} \right)^2
\left( \frac{\lambda_Z \langle F_Z \rangle}
{\lambda_{Z^\prime} \langle Z^\prime \rangle} \right)^2
(\sum_a C_a)\; ( \sum_i n_a(i)) \; \; ,
\label{scalar}\end{aligned}$$ where $C_a$ is the quadratic Casimir invariant for the scalar partners which is defined as $C_3=4/3$, $C_2=3/2$ and $C_1=3/5 Y^2$. If the values of parameters $\lambda_Z$, $\lambda_M$, $\lambda_N$, $\Lambda$ and $\Lambda^\prime$ are fixed, the masses of all the superpartners are fixed by eqs.(\[gaugino\]) and (\[scalar\]).
However, all of the values of these parameters are not allowed. For simplicity, we take $\lambda_M \sim \lambda_N \sim \cal{O}$(1) and $\Lambda = \Lambda^\prime$. Then, the dynamical scale $\Lambda$ has a theoretical lower bound. Since there are many charged particles in addition to the ordinary quarks and leptons in our model, the QCD gauge coupling blows up below the Planck scale, unless the dynamical scale of the $SP(8) \times SU(2)$ gauge interaction is high enough. We define mass scale of the fields $\bar{\phi}$, $\phi$ and $\bar{A}$ as $m^\prime = \lambda_{\bar{\phi}}\Lambda \sim \lambda_{\phi}\Lambda
\sim \lambda_{\bar{A}}\Lambda$, and the messenger scale as $m = \lambda_{Z^\prime} \langle Z^\prime \rangle
\sim 1/2 \; \Lambda$. Let us consider one-loop renormalization group equation (RGE) of the QCD coupling [@abh]. At the scale $M_{SUSY} \leq \mu \leq m^\prime$, the solution to the RGE is given by $$\begin{aligned}
\frac{1}{\alpha_3(M_{SUSY})} - \frac{1}{\alpha_3(\mu)}=
-\frac{3}{2 \pi} \ln (\mu/M_{SUSY}) \; \; ,
\label{RGE1}\end{aligned}$$ where $M_{SUSY} \sim 1$TeV is a typical value of masses of the superpartners in the MSSM. At the scale $m^\prime \leq \mu \leq m$ (remember our assumption $m^\prime \ll m$), the fields $\bar{\phi}$, $\phi$ and $\bar{A}$ contribute to the RGE, and the solution is given by $$\begin{aligned}
\frac{1}{\alpha_3(m^\prime)} - \frac{1}{\alpha_3(\mu)}=
\frac{5}{2 \pi} \ln (\mu/m^\prime) \; \; .
\label{RGE2}\end{aligned}$$ At the scale $m \leq \mu$ where all of the colored fields contribute to the RGE, we obtain $$\begin{aligned}
\frac{1}{\alpha_3(m)} - \frac{1}{\alpha_3(\mu)}=
\frac{13}{2 \pi} \ln (\mu/m) \; \; .
\label{RGE3}\end{aligned}$$ Note that this solution is not changed at $ \Lambda \leq \mu $ where the dynamical degrees of freedom of the messenger fields are replaced by that of the elementary fields $\bar{P}$ and $P$. Let us define the theoretical lower bound on $m=1/2 \; \Lambda$ as $1/ \alpha(M_{Pl}) =0$, where $M_{Pl}=10^{19}$GeV is the Planck scale. From eqs.(\[RGE1\]), (\[RGE2\]) and (\[RGE3\]), the bound is given by $$\begin{aligned}
m = \delta^{-1/2} \; M^{3/16}_{SUSY}\;
M_{Pl}^{13/16} \Bigg/
\exp \left( \frac{\pi}{8 \alpha_3(M_{SUSY})} \right)
\sim \delta^{-1/2} \; 10^{14} \; \; \mbox{GeV} \; \; ,
\label{eq16} \end{aligned}$$ where $\delta$ is defined as $\delta= m^\prime / m$, and we take $1/\alpha_3(M_{SUSY})\sim 12$. If we take $\delta \sim 10^{-2}$, the lower bound on the dynamical scale of the $SP(8) \times SU(2)$ gauge interaction is given by $\Lambda \sim 10^{15}$ GeV.
Next, let us investigate the upper bound on $\lambda_Z$ by implying the naturalness criterion [@thooft]. According to the criterion, the masses of the scalar partners in the MSSM should be less than 1 TeV. From eq.(\[scalar\]), we obtain $$\begin{aligned}
\frac{2 \alpha_3}{\sqrt{3}\pi}\; \frac{\lambda_Z \langle F_Z \rangle}
{\lambda_{Z^\prime}\langle Z^\prime \rangle}
\sim \frac{4 \alpha_3}{\sqrt{3}\pi}\;
\lambda_Z^2 \; \Lambda \leq 1 \; \;
\mbox{TeV} \; \; ,
\label{eq17}\end{aligned}$$ where $C_3=4/3$ and $\sum_i n_3 = 8$ are used. Considering the lower bound on $\Lambda \geq 10^{15}$ GeV, the upper bound on $\lambda_Z \leq 10^{-6}$ is obtained, where we take $\alpha_3 \sim 0.1$. Note that this upper bound is consistent with our assumption $\lambda_Z \langle F_Z \rangle \ll m^2$ used to obtain eqs.(\[gaugino\]) and (\[scalar\]).
Here, we give a comment on the value of $\lambda_Z$. Although the upper bound on $\lambda_Z \leq 10^{-6}$ seems to be unnaturally small, this result is due to our assumption $\Lambda = \Lambda^\prime$, and can be avoided in the case $\Lambda \ll \Lambda^\prime$. Eqs.(\[vac\]) and (\[eq17\]) suggest that the upper bound of $\lambda_Z$ becomes larger as $\Lambda^\prime$ becomes larger than $\Lambda$. For example, if we take $\Lambda=4 \times 10^9$ and $\Lambda^\prime=6 \times 10^{11}$ which satisfy eq.(\[eq16\]), $\lambda_Z \leq {\cal O}(1)$ can be obtained from eqs.(\[vac\]), (\[eq16\]) and (\[eq17\]).
In summary, we present a simple model of the gauge mediated supersymmetry breaking. Our model is based on the gauge group $SP(8) \times SU(2)$. Supersymmetry is dynamically broken by the strong SP(8) dynamics, and the composite fields which would be the messenger fields also appear by this dynamics. At this stage, there is no $U(1)_R$ symmetry breaking mass term for the composite fields. The mass terms are generated by the strong $SU(2)$ dynamics. Then, the composite fields can play a role of the messenger fields. On the other hand, the theoretical bounds on the parameters in our model is discussed. The dynamical scale of the $SP(8) \times SU(2)$ gauge interaction should be more than $10^{15}$ GeV to prevent the QCD coupling from blowing up below the Planck scale. The naturalness criterion requires $\lambda_Z \leq 10^{-6}$ together with the lower bound on the dynamical scale.
Finally, we would like to comment on a possibility of extension of our model. The gauge group $SP(8)$ is minimal one to be able to include fields with the vector-like $\bf{5+\bar{5}}$ representation under the MSSM gauge group into the $SP(8)$ dynamics. It is possible to introduce the vector-like fields, only if the number of flavors is more than five. Therefore, we can extend the gauge group $SP(8)$ to $SP(2N)$ ($N \geq 5$) with $N+1$ flavors in general. On the other hand, the gauge group $SU(2)$ is also minimal one. It is possible to generate the $U(1)_R$ symmetry breaking mass terms for the messenger fields by the same mechanism discussed above, only if $N_f < N_C$, where $N_f$ and $N_C$ are number of flavors and colors of $SU(N_C)$, respectively. Therefore, we can extend the gauge group $SU(2)$ to $SU(N)$ ($N \geq 3$) with $N_f < N$ flavors in general.
The author would like to thank Noriaki Kitazawa for useful comments. This work was supported in part by the Grant in Aid for Scientific Research from the Ministry of Education, Science and Culture and the Research Fellowship of the Japan Society for the Promotion of Science for Young Scientists.
M. Dine and A.E. Nelson, Phys. Rev. D 48 (1993) 1277; M. Dine, A.E. Nelson and Y. Shirman, Phys. Rev. D 51 (1995) 1362; M. Dine, A.E. Nelson, Y. Nir and Y. Shirman, Phys. Rev. D 53 (1996) 2658. Izawa K.-I. and T. Yanagida, Prog. Theor. Phys. 95 (1996) 949; K. Intriligator and S. Thomas, Nucl. Phys. B 473 (1996) 121. T. Hotta, Izawa K.-I. and T. Yanagida, Phys. Rev. D 55 (1997) 415; Izawa K.-I., preprint hep-ph/9704382; Izawa K.-I., Y. Nomura, K. Tobe and T. Yanagida, Phys. Rev. D 56 (1997) 2886. E. Poppitz and S.P. Trivedi, Phys. Rev. D 55 (1997) 5508; H. Murayama, Phys. Rev. Lett 79 (1997) 18; S. Dimopoulos, G. Dvali and R. Rattazzi, preprint hep-ph/9707537; M. Luty and J. Terning, preprint hep-ph/9709306; Y. Shirman, preprint hep-ph/9709383. N. Seiberg, Phys. Lett. B 318 (1993) 469; N. Seiberg, Phys. Rev. D 49 (1994) 6857; N. Intriligator, R.G. Leigh and N. Seiberg, Phys. Rev. D 50 (1994) 1092. L. O’Raifeartaigh, Nucl. Phys. B 96 (1975) 331. M. Hqu, Phys. Rev. D 14 (1976) 3548. I. Affleck, M. Dine and N. Seiberg, Nucl. Phys. B 256 (1985) 557. S.P. Martin, Phys. Rev. D 55 (1997) 3177. U. Amaldi, W.de Boer and H. Fürstenau, Phys. Lett. B 260 (1991) 447. G. ’t Hooft, 1979 Cargese Lectures, published in Recent Developments In Gauge Theories, Proceedings, NATO Advanced Study Institute New York, USA: Plenum (1980).
[^1]: e-mail: [email protected]
[^2]: JSPS Research Fellow
[^3]: We assume that all of the parameters in our model are real and positive, for simplicity.
[^4]: We use the same notation for the superfield itself and the scalar component of the superfield.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In this paper we consider finite volume hyperbolic manifolds $X$ with non-empty totally geodesic boundary. We consider the distribution of the times for the geodesic flow to hit the boundary and derive a formula for the moments of the associated random variable in terms of the orthospectrum. We show that the the first two moments correspond to two cases of known identities for the orthospectrum. We further obtain an explicit formula in terms of the trilogarithm functions for the average time for the geodesic flow to hit the boundary in the surface case, using the third moment.'
address:
- Boston College
- National University of Singapore
author:
- Martin Bridgeman
- Ser Peow Tan
title: Moments of the boundary hitting function for the geodesic flow on a hyperbolic manifold
---
Introduction
============
Let $X$ be a finite volume hyperbolic manifold with non-empty totally geodesic boundary $\partial X$. An [*orthogeodesic*]{} for $M$ is a geodesic arc with endpoints perpendicular to $\partial X$. These were first introduced by Basmajian in [@Bas93] in the study of totally geodesic submanifolds. We denote by $O_X = \{\alpha_{i}\}$ the collection of orthogeodesics of $X$ and let $l_i$ be the length of $\alpha_{i}$. We note that $O_X$ is countable as the elements correspond to a subset of the collection of closed geodesics of the double of $X$ along its boundary. We call the set $L_X = \{l_{i}\}$ (with multiplicities) the [*orthospectrum.*]{}
In [@Bas93], Basmajian derived the following boundary orthospectrum identity; $$\mbox{Vol}(\partial X) = 2 \sum_{l \in L_{X}} V_{n-1}\left(\log\left(\coth{\frac{l}{2}}\right)\right)
\label{basmajian}$$ where $V_n(r)$ is the volume of the ball of radius $r$ in ${\mathbb{H}}^n$. The identity comes from considering the universal cover $\tilde{X} \subseteq {\mathbb{H}}^n$ of $X$. Then $\partial\tilde{X}$ is a countable collection of disjoint hyperbolic hyperplanes which are the lifts of the boundary components of $\partial X$. For each component $C$ of $\partial\tilde{X}$, we orthogonally project each of the other components of $\partial\tilde{X}$ onto $C$ to obtain a collection of disjoint disks on each component $C$ of $\partial\tilde{X}$. These disks form an equivariant family of disks that are full measure in $\partial\tilde{X}$. They descend to a family of disjoint disks in $\partial X$ of full measure. As each orthogeodesic lifts to a perpendicular between two components of $\partial\tilde{X}$, each orthogeodesic corresponds to two disks (one at each end) in the family of disks in $\partial X$ and this gives the above identity.
Using a decomposition of the unit tangent bundle, Bridgeman-Kahn (see [@BK10]) derive the identity; $${\mbox{Vol}}(T_1(X)) = \sum_{l \in L_{X}} H_n(l)
\label{bridgeman}$$ where $H_n$ is some smooth function depending only on the dimension $n$. As ${\mbox{Vol}}(T_1(X)) = {\mbox{Vol}}(X).V_{n-1}$ where $V_{n-1}$ is the volume of the unit sphere in ${\mathbb R}^n$, the above identity can also be written as $${\mbox{Vol}}(X) = \sum_{l \in L_{X}} \overline{H}_n(l)$$ where $\ \overline {H}_n(l) = H_n(l)/V_{n-1}$.In the specific case of dimension two, the function $ H_n$ is given in terms of the Rogers dilogarithm (see [@B11]). In the papers [@Cal10a; @Cal10b] Calegari gives an alternative derivation of the identity in equation \[bridgeman\].
The motivation for this paper was to connect the above two identities in a natural framework. The connection is that they are the first two moments of the Liouville measure. A second motivation was to compute the average time it takes to hit the boundary under the geodesic flow. This can be put into the same framework and it turns out that consideration of the third moment gives a formula for the average time it takes to hit the boundary of $M$ under the geodesic flow in terms of the orthospectrum. It is conceivable that higher moments encode other important geometric invariants of the manifold.
Moments of Liouville measure
============================
We let $G({\mathbb{H}}^n)$ be the space of oriented geodesics in ${\mathbb{H}}^n$. By identifying a geodesic with its endpoints on the sphere at infinity, the space $G({\mathbb{H}}^n) \simeq ({\mathbb{S}}_\infty^{n-1}\times {\mathbb{S}}_\infty^{n-1} - \mbox{Diagonal})$. The [*Liouville measure*]{} $\mu$ on $G({\mathbb{H}}^n)$ is a Mobius invariant measure. In the upper half space model, we identify a geodesic with its endpoints $(x,y) \in \overline{{\mathbb R}^{n-1}}\times\overline{{\mathbb R}^{n-1}}$. Then the Liouville measure $\mu$ has the form $$d\mu_{(x,y)} = \frac{2dV_xdV_y}{|x-y|^{2n-2}}$$ where $dV_x = dx_1dx_2\ldots dx_{n-1}$, for $x = (x_1, x_2,\ldots,x_{n-1}) \in {\mathbb R}^{n-1}$.
If $X$ is a hyperbolic $n$-manifold with totally geodesic boundary, we identify $\tilde{X}$, the universal cover of $X$ as a subset of ${\mathbb{H}}^n$ and $\Gamma \subseteq \mbox{Isom}^{+}({\mathbb{H}}^n)$ such that $X = \tilde{X}/\Gamma$. Then $G(\tilde{X}) \subseteq G({\mathbb{H}}^n)$ is the set of geodesics intersecting $\tilde{X}$. We define $G(X) = G(\tilde{X})/\Gamma$, the space of geodesics in $X$. Then by invariance of Liouville measure $\mu$ descends to a measure on $G(X)$ which we also call $\mu$.
We define the measurable function $L:G(X) \rightarrow [0,\infty]$ by $L(g) := \mbox{Length}(g)$, where the length is taken in $X$. This is the hitting length function for $X$. As the limit set $L_\Gamma$ has measure zero, almost every geodesic hits the boundary of $X$ and therefore for almost every geodesic $g$, $L(g)$ is finite and $g$ is a proper geodesic arc.
We define the pushforward measure $M := L_*(\mu)$ on the real line. This measure is the distribution of lengths of geodesics in $X$. We define its k-th moment to be $$M_k(X) = M(x^k) = \int_0^\infty x^k dM = \int_{G(X)} L^k(g) \ d\mu.$$ In general, the moments of a random variable give a set of measurements that describe distributional properties of the random variable such as the average value and variance. Using a decomposition of $G(X)$, we show that the moments $M_k(X)$ have formulae that extend the identities in equations \[basmajian\],\[bridgeman\].
The main result of the paper is the following:
(Main Theorem)
*There exists smooth functions $F_{n,k}: {\mathbb R}_+ \rightarrow {\mathbb R}_+$ and constants $C_n > 0$ such that if $X$ is a compact hyperbolic n-manifold with totally geodesic boundary $\partial X \neq \emptyset$, then*
1. The moment $M_k(X)$ satisfies $$M_k(X) = \sum_{l \in L_X} F_{n,k}(l)$$
2. $M_{0}(X) = C_n.{\mbox{Vol}}(\partial X)$ and the identity for $M_{0}(X)$ is the identity in equation \[basmajian\].
3. $M_{1}(X) = {\mbox{Vol}}(T_1(X))$ and the identity for $M_1(X)$ is the identity in equation \[bridgeman\].
4. $M_2(X) = 2{\mbox{Vol}}(T_1(X))A(X)$ where $A(X)$ is the average time for a vector in $T_1(X)$ to hit the boundary under geodesic flow. Therefore by the identity for $M_2(X)$ $$A(X) = \frac{1}{2{\mbox{Vol}}(T_1(X))}\sum_{l \in L_X} F_{n,2}(l) = \sum_{l \in L_X} G_{n}(l).$$
In the surface case we obtain an explicit formula for the function $G_2$ and hence $A(X)$ in terms of polylogarithms. Furthermore, besides compact surfaces obtained as quotients of Fuchsian groups, the identity holds more generally for finite area surfaces, which we describe next.
If $S$ is a finite area surface with totally geodesic boundary $\partial S \neq \emptyset$, then the boundary components are either closed geodesics or bi-infinite geodesics with cuspidal endpoints. We define a [*boundary cusp*]{} of $S$ to be an ideal vertex of $\partial S$. We let $C_S$ be the number of boundary cusps of $S$. Then we have the following explicit formula for $A(S)$:
Let $S$ be a finite area hyperbolic surface with non-empty totally geodesic boundary. Then $$A(S) = \frac{1}{8\pi^2|\chi(S)|}\left(\sum_{l \in L_S} F\left(\frac{1}{\cosh^2(l/2)}\right) + 6\zeta(3)C_S\right)$$ where $$\begin{aligned}
F(a) &=& -12\zeta(3)-\frac{4\pi^2}{3}\log(1-a)+6\log^2(1-a)\log(a)-4\log(1-a)\log^2(a)\\
&& \qquad -8\log\left(\frac{a^2}{1-a}\right)Li_2(a)
+24Li_3(a)+12Li_3(1-a),\end{aligned}$$ $Li_k(x)$ is the $k^{th}-$polylogarithm function, and $\zeta$ is the Riemann $\zeta-$function. \[avehit\]
A natural fibering
==================
We have the natural fiber bundle $p:T_1({\mathbb{H}}^n) \rightarrow G({\mathbb{H}}^n)$ such that $v$ is tangent to the oriented geodesic $p(v)$. Let $\Omega$ be the volume measure on $T_1({\mathbb{H}}^n)$. We parametrize $T_1({\mathbb{H}}^n)$ as follows: We first choose a basepoint $b_g$ on each geodesic $g$ (say by taking the point closest to a fixed point p). Then a vector $v \in T_1({\mathbb{H}}^n)$ is given by a triple $(x,y,l) \in \overline{{\mathbb R}^{n-1}}\times\overline{{\mathbb R}^{n-1}}\times{\mathbb R}$ where $v$ is tangent to the geodesic $g$ with endpoints $x,y$ (from $x$ to $y$), and $l$ is the signed length along $g$ from the basepoint $b_g$. In terms of this parametrization the volume form $\Omega$ on $T_1({\mathbb{H}}^n)$ is $$d\Omega_{v} = \frac{2dV_xdV_ydl}{|x-y|^{2n-2}} = d\mu_{(x,y)}dl_{v}$$ where $\mu$ is the Liouville measure (see [@Nic89]). If $X$ is a hyperbolic $n$-manifold with totally geodesic boundary and $\tilde{X}$ the universal cover of $X$, then the fiber bundle $p$ restricts to $T_1(\tilde{X})$ to give the equivariant map $p:T_1(\tilde{X}) \rightarrow G(\tilde{X})$ which descends to a map $\overline{p}:T_1(X) \rightarrow G(X)$. We have the function $\overline{L}: T_1(X) \rightarrow [0,\infty]$ given by $\overline{L} = L\circ \overline{p}$. The measure $N = \overline{L}_*\Omega$ was introduced in [@B11] in order to derive the surface case of the identity \[bridgeman\]. The measures $M,N$ have a simple relation which we now describe.
For $\phi:[0,\infty)\rightarrow{\mathbb R}$ a smooth function with compact support then $$N(\phi) = \Omega(\phi\circ \overline{L}) = \int_{v \in T_1(X)} \phi(\overline{L}(v))d\Omega_v = \int_{g \in G(X)} \int_{v \in p^{-1}(g)}\phi(L(g))d\mu_gdl_v$$ $$= \int_{g \in G(X)}\phi(L(g))\left( \int_{v \in p^{-1}(g)}dl_v\right)d\mu_g = \int_{g \in G(X)}\phi(L(g))L(g) d\mu_g = M(x\phi)$$ It follows that the measures $M, N$ satisfy $ dN = xdM.$ We define the moments of $N$ to be $N_k(X) = N(x^k)$. Then $$N_k(X) = M_{k+1}(X).
\label{2moments}$$ Also as $N_0(X) = {\mbox{Vol}}(T_1(X))$ it follows that $M_1(X) = {\mbox{Vol}}(T_1(X))$.
Because of the above relation between the measures $M,N$, the results described in this paper can be given in terms of either. For the most part we will give our results in terms of the measure $M$ and its moments, but when it is more natural to do so, we will consider the measure $N$.
We let $A(X)$ be the average time for a vector in $T_1(X)$ to hit the boundary under geodesic flow. Then as the time for $v$ and $-v$ sum to $\overline{L}(v)$, $A(X)$ is half the average of the function $\overline{L}$. Then $A(X)$ is given by the first moment of measure $N$ and as $N_1(X) = M_2(X)$, we have the formula $$A(X) = \frac{1}{2}\frac{M_2(X)}{{\mbox{Vol}}(T_1(X))}$$ proving part 4) of the Main Theorem.
Moments are finite
==================
Before we derive summation formulae for the moments $M_k(X)$, we need to first show that they are finite. The proof that $M_0(X), M_1(X)$ are finite will follow from explicit calculation, in particular, from the last section $M_1(X) = {\mbox{Vol}}(T_1(X))$ which is finite by assumption. By equation \[2moments\], $M_{k}(X) = N_{k-1}(X)$, and therefore we need only show that $N_k(X)$ is finite for $k \geq 1$. To prove this, we show that the measure $N$ on the real line decays exponentially, i.e. there exist positive constants $a, C$ such that $dN \leq Ce^{-at}dt$ for $t$ large. Then $N_k(X)$ is finite as the measure $x^ke^{-at}dt$ is finite.
We first recall some background on Kleinian groups (see [@Mas87] for details). A [*Kleinian group*]{} $\Gamma$ is a discrete subgroup of the isometries of ${\mathbb{H}}^n$. The [*limit set*]{} $L_\Gamma = \overline{\Gamma x}\cap {\mathbb{S}}^{n-1}_\infty$ is the accumulation set of an orbit of a point $x$ on the boundary. It is easy to show that $L_\Gamma$ is independent of $x$. The [*convex hull*]{} of $\Gamma$, denoted $H(\Gamma)$, is the smallest convex set containing all geodesics with endpoints in $L_\Gamma$. As $H(\Gamma)$ is invariant under $\Gamma$, the [*convex core*]{} is defined to be $C(\Gamma) = H(\Gamma)/\Gamma$. A Kleinian group is [*convex cocompact*]{} if $C(\Gamma)$ is compact. Also a group is [*geometrically finite*]{} if $N_\epsilon(C(\Gamma))$, the $\epsilon$ -neighborhood of the core, is finite volume.
Let $\Gamma$ be a convex cocompact Kleinian group with $N = {\mathbb{H}}^n/\Gamma$ and $X = H(\Gamma)/\Gamma$ its convex core. We let $\delta(\Gamma)$ be the Hausdorff dimension of the limit set $L_\Gamma$. Let $g_t$ be the geodesic flow on $T_1(N)$ and define $$B(t) = \{ v\in T_1(X)\ | \ g_t(v) \in T_1(X)\} = g_t(T_1(X))\cap T_1(X)$$ The set $B(t)$ is the set of tangent vectors that remain in the convex core under time t flow. We now use a standard counting argument on orbits to bound the volume of the set $B(t)$ (see [@Nic89] for background).
Given $\Gamma$ a convex cocompact Kleinian group, then there exists constants $A, T$ such that $${\mbox{Vol}}(B(t)) \leq Ae^{-(n-1 -\delta(\Gamma))t}$$ for $t > T$. In particular if $L_\Gamma \neq {\mathbb{S}}^{n-1}_\infty$ then ${\mbox{Vol}}(B(t))$ is exponentially decaying.
[**Proof:**]{} We take $0 \in H(\Gamma)$ and consider its orbits under $\Gamma$. Then we let $$O(r) = \{\gamma \in \Gamma\ | d(0,\gamma(0)) < r\}\qquad \mbox{and} \qquad N(r) = \#O(r).$$ By Sullivan (see [@Sul79]), there exists constants $A, r_0$ such that $N(r) \leq Ae^{\delta(\Gamma)r}$ for $r > r_0$. We let $D$ be the diameter of $X$. Given a unit tangent vector $v$, we denote its basepoint by $b(v)$. We let $\tilde{B}(t)$ be a lift of $B(t)$ with basepoints within a distance $D$ of $0$. For $v \in \tilde{B}(t)$ then by the trangle inequality, $g_t(v)$ has basepoint $b(g_t(v))$ such that $t-D < d(b(g_t(v)), 0) <t+D$. Therefore $b(g_t(v))$ has nearest orbit $\gamma(0)$ such that $b(g_t(v)) \in B(\gamma(0),D)$, the ball of radius $D$ around $\gamma(0)$. Also we have that $t-2D < d(\gamma(0),0) <t+2D$. Therefore $$U(t) = \bigcup_{O(t-2D)^c\cap O(t+2D)} B(\gamma(0), D)$$
Let $x$ be a distance at most $D$ from $0$. We want to bound the visual measure of $U(t)$ from $x$. Each $B(\gamma(0),D) \in U(t)$ is a distance between $t-3D,t+3D$ from $x$. Let $T_1 > 3D$, and restrict to $t > T_1$. We radially project each $B(\gamma(0),D) \in U(t)$ onto the $S(x,t)$, the sphere of radius $t$ in hyperbolic space. We label the projection $P(\gamma(0),D)$. Then the area of each projection is bounded above by a constant $C_1>0$. Therefore $$Vis_x(U(t)) \leq \frac{C_1N(t+2D)}{{\mbox{Vol}}(S(x,t))}.$$ We have that ${\mbox{Vol}}(S(x,t)) = S_n\sinh^{n-1}(t)$ where $S_n$ is the volume of the standard Euclidean sphere of dimension $(n-1)$. Thus for $t > T_1$, ${\mbox{Vol}}(S(x,t))\geq L_ne^{(n-1)t}$ for some $L_n > 0$.
Therefore for $$Vis_x(U(t)) \leq \frac{C_1.N(t+2D)}{L_ne^{(n-1)t}} \leq \frac{C_1.Ae^{\delta(\Gamma)(t+2D)}}{L_ne^{(n-1)t}} \leq Ce^{-((n-1)-\delta(\Gamma))t}$$ for some constant $C$.
In order to obtain the bound on ${\mbox{Vol}}(B(t))$ we integrate the visual measure of $U(t)$ over $\tilde{B}(t)$ gives $${\mbox{Vol}}(B(t)) = {\mbox{Vol}}(\tilde{B}(t)) \leq {\mbox{Vol}}(X).Ce^{-((n-1)-\delta(\Gamma))t}$$ for $t > T$, giving our result for $\Gamma$ convex cocompact.
If $X$ is a compact hyperbolic manifold with non-empty totally geodesic boundary then the moments $M_k(X)$ are finite for $k \geq 1$.
[**Proof:**]{} Let $$E(t) = \{ v \in T_1(X) \ |\ \overline{L}(v) \in [t,t+1)\}.$$ Then from the above lemma $E(t) \subset B(t/2)\cup B(-t/2)$. Therefore there are constants $a, K > 0$ such that $Vol(E(t)) \leq Ke^{-at}$ for $t > 2T_0$. Therefore $$N_{k}(X) \leq \sum_{n=0}^\infty (n+1)^{k}. Vol(E(n))$$ is finite for $k \geq 0$ by comparison with the series $\sum n^ke^{-an}.$ Therefore $M_k(X)$ is finite for $k \geq 1$.
Decomposition of the space of geodesics
=======================================
We let $G({\mathbb{H}}^n)$ be the space of oriented geodesics in ${\mathbb{H}}^n$. By identifying a geodesic with its endpoints, the Liouville measure $\mu$ on $G({\mathbb{H}}^n)$ is given by $$d\mu = \frac{2dV_xdV_y}{|x-y|^{2n-2}}$$
If $X$ is a hyperbolic $n$-manifold with totally geodesic boundary then we let $\tilde{X}$ be the universal cover of $X$ as a subset of ${\mathbb{H}}^n$ and $\Gamma \subseteq \mbox{Isom}^{+}({\mathbb{H}}^n)$ such that $X = \tilde{X}/\Gamma$. We let $G(\tilde{X}) \subseteq G({\mathbb{H}}^n)$ be the set of geodesics intersecting $\tilde{X}$. We define $G(X) = G(\tilde{X})/\Gamma$, the space of geodesics in $X$. Then by invariance of Liouville measure $\mu$ descends to a measure on $G(X)$ which we also call $\mu$.
The space $G(X)$ has a simple (full measure) decomposition via orthogeodesics. For $\alpha$ an orthogeodesic we define $$F_\alpha =\{ g \in G(X) \ | \ g \mbox{ is homotopic rel $\partial X$ to } \alpha\}.$$ The set of orthogeodesics is countable, so we index our orthogeodesics $ O_X =\{\alpha_i\}$ and the sets $F_i = F_{\alpha_i}$. As the limit set $L_\Gamma$ has zero measure, $\bigcup F_i$ gives a (full measure) partition of $G(X)$ with respect to $\mu$. We note that the sets $F_i \subseteq {\mathbb{S}}_\infty^{n-1}\times {\mathbb{S}}_\infty^{n-1}$ are of the form $(D_1\times D_2)\cup(D_2\times D_1)$ where $D_1, D_2$ are disjoint round disks. Thus $G(X)$ decomposes (up to full measure) into a countable collection of elementary pieces indexed by the orthogeodesics.
Then $$M_k(X) = M(x^k) = \int_0^\infty x^k dM = \int_{G(X)} L^k(g)\ d\mu = \sum_{i} \int_{F_i} L^k(g)\ d\mu$$ where $L:G(X) \rightarrow {\mathbb R}$ is the length of a geodesic.
We now lift $F_i$ to a set $\tilde{F}_i \subset G({\mathbb{H}}^n)$. In the upper half-space model, we choose two planes $P, Q$ with orthogonal distance $l_i = Length(\alpha_i)$. Then $F_i$ lifts to the set of geodesics which intersect $P$ and $Q$. In particular we can take $P, Q$ to have boundary circles centered at $0$ with radii $1, e^{l_i}$ respectively. Lifting the function $L$ to $\tilde{F}_i$ we see that it only depends on the endpoints $(x, y)$ and the ortholength $l_i$. We denote it by $L(x,y, l_i)$. $$\int_{F_i} L^k(g) d\mu = \int_{\tilde{F}_i} \frac{2L^{k}(x,y) dV_xdV_y}{|x-y|^{2n-2}}$$ Integrating over the length parameter (in both directions) we obtain $$\int_{F_i} L^k(g) d\mu = \int_{|x| <1} \int_{|y| > e^{l_i}} \frac{4L(x,y, l_i)^{k} dV_xdV_y}{|x-y|^{2n-2}} = F_{n,k}(l_i).$$ Therefore we have that $$M_{k}(X) = \sum_i F_{n,k}(l_i)$$ and $F_{n,k}$ is given by the integral formula $$F_{n,k}(t) = \int_{|x| <1} \int_{|y| > e^{t}} \frac{4L(x,y,t)^{k} dV_xdV_y}{|x-y|^{2n-2}}$$ This gives the summation formula for the moment $$M_k(X) = \sum_{l \in L_X} F_{n,k}(l)$$ and proves part 1) of the Main Theorem.
By equation \[2moments\], $M_1(X) = N_0(X) = {\mbox{Vol}}(T_1(X))$. Thus letting $E_i \subseteq T_1(X)$ given by $E_i = \overline{p}^{-1}(F_i)$ then the identity for $M_1(X)$ is given by $$M_1(X) = {\mbox{Vol}}(T_1(X)) = \sum_i \Omega(E_i).$$ In the paper [@BK10], we show that this is the identity in equation \[bridgeman\]. This proves part 3) of the Main Theorem.
Zero Moment identity is Basmajian’s identity
============================================
As $M_0(X) = \mu(G(X))$ the identity for $M_0(X)$ is $$\mu(G(X)) = \sum_i \mu(F_i)$$ We now prove some properties of the Liouville measure $\mu$ needed to evaluate both sides of this identity and show that it gives Basmajian’s identity in equation \[basmajian\].
Let $\mu$ be the Liouville measure on $G({\mathbb{H}}^n)$ and $P$ a plane in ${\mathbb{H}}^n$. For $U \subseteq P$, let $G(U) \subseteq G({\mathbb{H}}^n)$ be the set of geodesics intersecting $U$ transversely. Then the measure $\mu_P$ on $P$ defined by $\mu_P(U) = \mu(G(U))$ is a constant times area measure on $P$. In particular there is a constant $K_n > 0$ depending only on dimension such that $\mu_P(U) = K_n.{\mbox{Vol}}(U)$. \[liouville\]
[**Proof:**]{} In the case of $n=2$ this is a standard property of $\mu$ (see [@Bon88]). In general we see that $\mu_P$ gives a Mobius invariant measure on the hyperbolic plane $P$ and therefore must be a multiple of area measure. Thus $\mu_P (U) = K_n. {\mbox{Vol}}(U)$ where $K_n$ only depends on the dimension $n$. We calculate $K_n$ in a later section
[**A Liouville measure preserving map:**]{} We let $P$ be a plane in ${\mathbb{H}}^n$ and $G_P$ the set of geodesics intersecting $P$ transversely. Then $G_P =(D_-\times D_+) \cup (D_+\times D_-)$ where $D_-,D_+$ are disjoint open disks in ${\mathbb{S}}_\infty^{n-1}$ with $\partial D_- = \partial D_+$. We let $g = (x(g),y(g)) \in G_P$ be the ordered pair of endpoints of $g$ and $m(g) = g \cap P$ the point of intersection with $P$. We project the endpoints of $g$ orthogonally onto $P$ to obtain the ordered pair of points $(p(g), q(g)) \in P\times P$. We note that $m(g)$ is the midpoint of the hyperbolic geodesic arc in $P$ joining $p(g),q(g)$. We now define a map $F:G_P \rightarrow G_P$ as follows; We first define $F:D_-\times D_+ \rightarrow D_-\times D_+$ and extend it to $D_+\times D_-$ by conjugating with the map switching endpoints, i.e. if $i(x,y) = (y,x)$ then $F(g) = i(F(i(g))$ for $g \in D_+\times D_-$. If $g \in D_-\times D_+$ then we define $F(g) = h$ where $h$ is the unique geodesic (in $D_-\times D_+$) such that $m(h) = q(g)$ and $q(h) = m(g)$ (see figure \[involution\]). By construction, $F$ is a involution and if $u$ is an isometry of ${\mathbb{H}}^n$ fixing $P$, then $F$ commutes with $u$. We note that $F$ is not the action of an isometry on the space of geodesics.
![The involution $F$ in the Klein model[]{data-label="involution"}](invol.jpg){width="5in"}
The map $F:G_P \rightarrow G_P$ preserves Liouville measure.
[**Proof:**]{} We show that $F_*(\mu) = \mu$ by showing the Radon-Nikodym derivative $$\nu(x) = \frac{dF_*(\mu)}{d\mu} (x) = 1.$$
Alternately, $\nu(x)$ is the function such that for any $\phi$ smooth compactly supported function on $G_P$ then $$F_*(\mu)(\phi) = \int \phi(x) d(F_*(\mu)) = \int \phi(x) \nu(x) d\mu(x)$$ Therefore $$F_*(\mu)(\phi) = \mu(\phi \circ F) = \int \phi(F(x))d\mu(x) = \int \phi(x) \nu(x) d\mu(x)$$
As $F$ is an involution $F(F(x)) = 1$, then by the change of variables formula we have $$\int \phi(x) d\mu(x) = \int \phi(F(F(x))d\mu(x) = \int \phi(F(x))\nu(x)d\mu(x) = \int \phi(x)\nu(F^{-1}(x))\nu(x)d\mu(x)$$ Therefore $$\nu(x)\nu(F^{-1}(x)) = 1 \qquad \mbox{or}\qquad \nu(x)\nu(F(x)) = 1.$$
Similarly we have if $u$ is a hyperbolic isometry then $u$ preserves the Liouville measure and $u_*(\mu) = \mu$. If $u$ also fixes $P$ then $F$ commutes with $u$ and $F(u(x)) = u(F(x))$. Therefore by the change of variables formula again we have $$\nu((u(x)) = \nu(x).$$
Thus combining the above, if there exists an isometry $u$ fixing $P$ such that $u(x) = F(x)$, then $\nu(x) = \nu(F(x))$. But as $\nu(x)\nu(F(x)) = 1$ we then obtain $\nu(x) = 1$.
To find such an isometry, we note that geodesic for $x \in G_P$, and $y = F(x)$ we have the four points $p(x), m(x) = q(y), q(x) = m(y), p(y)$ all collinear on $P$. We choose the plane $P^*$ perpendicular to $P$ which bisects the hyperbolic interval $[p(x),p(y)]$. Then refection in $P^*$ fixes $P$ and sends $x$ to $y = F(x)$.
Thus $\nu(x) = 1$ for all $x \in G_P$ and therefore $F_*(\mu) = \mu$.
If $\alpha$ is an orthogeodesic then $$\mu(F_\alpha) = K_n.V_{n-1}\left(\log\left(\coth{\frac{l(\alpha)}{2}}\right)\right)$$ \[equiv\]
[**Proof:**]{} We consider disjoint planes $P,Q$ with perpendicular distance equal the ortholength $l(\alpha)$. Then $\mu(F_\alpha) = \mu(S)$ where $S \subseteq G_P$ of geodesics which intersect $Q$. We let $G_P =(D_-\times D_+) \cup (D_+\times D_-)$ where $\partial Q \subseteq D_+$. By the above lemma, $\mu(S) = \mu(F(S))$. Let $B$ be the orthogonal projection of $Q$ onto $P$. Then by elementary hyperbolic geometry, $B$ is a ball of radius $r = \log(\coth{\frac{l(\alpha)}{2}})$. The set $F(S)$ is precisely the set of geodesics in $G_P$ transversely intersecting $B$ which we denote by $G(B)$. To see this, we note that if $g \in S$ then $F(g)$ intersects $P$ in $B$ giving $F(S) \subseteq G(B)$. Similarly if $g \in G(B)$ then $F(g)$ is in $S$. Thus, as $F$ is an involution $F(S) = G(B)$.
Therefore by lemma \[liouville\] $$\mu(F_\alpha) = \mu(S) = \mu(F(S)) = \mu(G(B)) = K_n{\mbox{Vol}}(B) = K_nV_{n-1}\left(\log\left(\coth{\frac{l(\alpha)}{2}}\right)\right)$$
We will now prove the second part of the Main theorem.
By the above corollary \[equiv\] we obtain $$\mu(G(X)) = \sum_{\alpha \in O_X}\mu(F_\alpha) = K_n\sum_{\alpha \in O_X}V_{n-1}\left(\log\left(\coth{\frac{l(\alpha)}{2}}\right)\right)$$
For each boundary component $B_i$ of $\partial X$ we let $B'_i$ in ${\mathbb{H}}^n$ be a hyperplane which is a lift of $B_i$. We further take a fundamental domain $D_i$ on $B'_i$ for the action of $\Gamma$. We let $C_i$ be the set of geodesics which intersect $D_i$ transversely such that the geodesics point into $\tilde{X}$ on $B'_i$. Let $G = \cup C_i$ then we see that $G$ is a lift of $G(X)$ (except for a set of measure zero). To see this note that for almost every $g \in G(X)$, $g$ is a proper arc from a $B_i$ to a $B_j$ where the orientation of $g$ is pointing into $X$ at $B_i$ and out at $B_j$. Therefore $g$ has lift $g'$ in $C_i$. Also for $i\neq j$ if $g \in C_i \cap C_j$ then $g$ points inward on both $B_i$ and $B_j$. Therefore $C_i \cap C_j = \emptyset$ for $i \neq j$. Also if $g_1, g_2 \in C_i$ are lifts of the same element of $G(X)$ then there is a $\gamma \in \Gamma$ with $g_2 = \gamma(g_1)$. As $D_i$ is a fundamental domain for the action of $\Gamma$ on $B'_i$ then $g_1, g_2$ must have endpoints on the boundary of $D_i$ which is measure zero.
Using this we can calculate $\mu(G(X))$. We have $$\mu(G(X)) = \sum_i\mu(C_i)$$ By the above lemma \[liouville\], $\mu(C_i) = (K_n/2){\mbox{Vol}}(B_i)$ where the factor of two comes from $C_i$ containing half the geodesics in the set $G(B_i)$ (those pointing into $\tilde{X}$). Therefore $$\mu(G(X)) = \frac{K_n}{2} \sum {\mbox{Vol}}(B_i) = \frac{K_n}{2}{\mbox{Vol}}(\partial X)$$ Thus $$\frac{K_n}{2}{\mbox{Vol}}(\partial X) = K_n\sum_{\alpha \in O_X}V_{n-1}\left(\log\left(\coth{\frac{l(\alpha)}{2}}\right)\right)$$ giving Basmajian’s identity.
Calculating $K_n$
=================
To calculate the constant $K_n$, we derive the Lebesgue density of $\mu_P$ at a point $p$. Let $p$ be at the origin of the Poincare model of ${\mathbb{H}}^n$ and let $P$ be the horizontal plane through $p$. Let $B = B_{n-1}(p,r)$ be a small $(n-1)$-dimensional ball in $P$ about $p$. Then we have $$d\mu = \frac{2dV_xdV_y}{|x-y|^{2n-2}}$$ where $V$ is volume measure on the unit sphere. If we consider geodesics $G(B)$ then if $(x,y) \in G(B)$, we have $|x-y| \simeq 2$. Also if we let $\theta_x$ be the angle the ray $px$ from $x$ to $p$ makes with the plane $P$, then the set $G(B)_x = \{y \in {\mathbb{S}}^{n-1} \ | \ (x,y) \in G(B)\}$ is a small ball in the unit sphere. In fact the set $G(B)$ is the image of $B$ under stereographic projection from $x$. Therefore $G(B)_x$ is an $(n-1)$-dimensional ellipsoid with the axes of $B$ perpendicular to ray $px$ being approximately $2r$ and other axis approximately $2r\sin(\theta_x)$. Therefore $${\mbox{Vol}}(G(B_x)) \simeq 2^{n-1}\sin(\theta_x).{\mbox{Vol}}(B^s_{n-1}(r))$$ where $B^s_{k}(r)$ is a k-dimensional ball of radius $r$ in the unit sphere. Integrating we get $$\mu(G(B)) = \int_{{\mathbb{S}}^{n-1}} 2.\left( \int_{G(B)_x} \frac{dV_y}{|x-y|^{2n-2}} \right)dV_x\simeq \frac{Vol(B^s_{n-1}(r))}{2^{n-2}} \int_{{\mathbb{S}}^{n-1}} \sin(\theta_x) dV_x$$
Thus if $A$ is area measure on $P$ then $$K_n = \frac{d\mu_P}{dA} = \frac{1}{2^{n-2}} \int_{{\mathbb{S}}^{n-1}} \sin(\theta_x)dV_x$$ The set $S_t = \{x\ | \ \theta_x = t\}$ is an $(n-2)$-dimensional sphere of radius $|\cos(t)|$. Therefore as $dV_x = d\theta dV_{S_t}$ $$K_n = \frac{{\mbox{Vol}}({\mathbb{S}}^{n-2})}{2^{n-2}} \int_{0}^{\pi} \sin(\theta)|\cos^{n-2}(\theta)| d\theta = \frac{{\mbox{Vol}}({\mathbb{S}}^{n-2})}{2^{n-1}} \int_{0}^{\pi/2} \sin(\theta)\cos^{n-2}(\theta)d\theta = \frac{{\mbox{Vol}}({\mathbb{S}}^{n-2})}{2^{n-1}(n-1)}$$ In terms of the Gamma function $\Gamma$ we have $${\mbox{Vol}}({\mathbb{S}}^k) = \frac{2\pi^{\frac{k+1}{2}}}{\Gamma(\frac{k+1}{2})}$$ Giving $$K_n = \frac{\pi^{\frac{n-1}{2}}}{2^{n-1}\Gamma(\frac{n+1}{2})}$$ Note: We have $K_2 = 1$.
Explicit Integral Formulae for $F_{n,k}$
========================================
In [@B11] we derive a formula for $L$ in the surface case. Using this we can write $$F_{2,k}(l) = \frac{1}{2^{k-2}} \int_0^{a}\int_1^\infty \frac{\log\left|\frac{y(y-a)(x-1)}{x(x-a)(y-1)}\right|^{k}}{(y-x)^2}dxdy.$$ where $a = \sech^2(l/2)$.
In [@BK10] we consider the case $n > 2$ where we derive the explicit formula for $L$ and reduce the integral formula to a triple integral via an elementary substitution. Using this we can also reduce the integral of $F_{n,k}$ to a triple integral of the form $$F_{n,k}(l) = \frac{{\mbox{Vol}}({\mathbb{S}}^{n-2}){\mbox{Vol}}({\mathbb{S}}^{n-3})}{2^{k-2}}\int_0^1 \frac{r^{n-3}}{(1-r^2)^{\frac{n-2}{2}}}dr\int^{1}_{-1} du\int^{\infty}_{b} \frac{\log\left(\frac{(v^2-1)(u^2-b^2)}{(v^2-b^2)(u^2-1)}\right)^{k}}{(v-u)^{n}}dv$$ where $b = \sqrt{\frac{e^{2l}-r^2}{1-r^2}}$.
Moment Generating function
==========================
The moment generating function of a random variable $Y$ is the function $f_Y(t) =E[e^{tY}]$ where $E$ is the expected value. We define the moment generating function for measure $M$ $$\zeta^M_X(t) = M(e^{tx}) = \int_0^\infty e^{tx} dM = \int_{G(X)} e^{t.L(g)} \ d\mu.$$ It follows from above that $$\zeta^M_X(t) = \sum_{l \in L_X} F_n(t,l)$$ for some function $F_n$ depending only on the dimension $n$. We similarly can define $\zeta^N_X(t) = N(e^{xt})$. The it follows that the two functions are related by $$\zeta_X^N(t) = \frac{d}{dt}\left(\zeta_X^M(t)\right) .$$
[**A simple example, the ideal triangle:**]{} We consider the case of $X$ being an ideal triangle. In this case, it is more natural to consider measure $N$ (in particular $M$ has infinite mass). In [@BD07] we show that $$dN= \frac{12x^2}{\sinh^2{x}}dx.$$ and that $$N_k(X) = \frac{3(k+2)\\!\zeta(k+2)}{2^{k-1}}$$ where $\zeta$ is the Riemann zeta function. In particular the average time to the boundary is $$A(X) = \frac{N_1(X)}{2{\mbox{Vol}}(T_1(X))} = \frac{9}{2\pi^2}\zeta(3).$$
It follows by integrating that $$\zeta^N_X(t) = \int_0^\infty\frac{12x^2e^{xt}}{\sinh^2{x}}dx = 12 \left(\zeta(2,1 - \frac{t}{2}) + \frac{t}{2} \zeta(3,1 - \frac{t}{2})\right)$$ where $\zeta(s,t)$ is the Hurwitz zeta function $$\zeta(s,t) = \sum_{k=0}^\infty \frac{1}{(k+t)^s}$$
The Surface Case
================
For the surface case, the identities in the Main Theorem can be written in terms of polylogarithm functions.
[**Polylogarithms:**]{} The $k^{th}$ polylogarithm function $\mbox{Li}_k$ is defined by the Taylor series $$\mbox{Li}_{k}(z) = \sum_{i=1}^{\infty} \frac{z^n}{n^k }$$ for $|z| < 1$ and by analytic continuation to ${\mathbb{C}}$. In particular $$\mbox{Li}_{0}(z) = \frac{z}{1-z} \qquad \qquad
\mbox{Li}_{1}(z) = -\log(1-z).$$ Also $$\mbox{Li}'_{k}(z) = \frac{\mbox{Li}_{k-1}(z)}{z} \qquad \mbox{giving} \qquad \mbox{Li}_{k}(z) = \int_{0}^{z} \frac{\mbox{Li}_{k-1}(t)}{t}\ t.$$ Also the functions $\mbox{Li}_{k}$ are related to the Riemann $\zeta$ function by $\mbox{Li}_{k}(1) = \zeta(k)$.
Below we describe some properties of the dilogarithm and trilogarithm function. They can all be found in the 1991 survey “Structural Properties of Polylogarithms” by L. Lewin (see [@Lew91]).
[**Dilogarithm:**]{} The dilogarithm function $\mbox{Li}_2(z)$ is given by $$\mbox{Li}_{2}(z) = -\int_{0}^{z} \frac{\log(1-t)}{t}\ dt.$$ From the power series representation, it is easy to see that the dilogarithm function satisfies the functional equation $$\mbox{Li}_{2}(z) + \mbox{Li}_{2}(-z) = \frac{1}{2}\mbox{Li}_{2}(z^{2}).$$ Other functional relations of the dilogarithm can be best described by normalizing the dilogarithm function. The (extended) Rogers dilogarithm function (see [@Rog07]) is defined by $$R(x) = \mbox{Li}_{2}(x) + \frac{1}{2}\log|x|\log (1-x)\qquad x \leq 1.$$ This function arises in calculating hyperbolic volume as the imaginary part of $R(z)$ is the volume of the hyperbolic tetrahedron with vertices having cross ratio $z$.
Also in terms of the Rogers function, various identities have nice form. Euler’s reflection relations for the dilogarithm are given by $$\begin{aligned}
R(x) + R(1-x) = R(1) = \frac{\pi^2}{6} \qquad 0 \leq x \leq 1 \nonumber \\
R(-x) + R(-x^{-1}) = 2R(-1) = -\frac{\pi^{2}}{6} \qquad x > 0
\label{euler}\end{aligned}$$ Also Landen’s identity is $$R\left(\frac{-x}{1-x}\right) = -R(x) \qquad 0 < x < 1
\label{landen}$$ and Abel’s functional equation is $$R(x) + R(y) = R(xy) + R\left(\frac{x(1-y)}{1-xy}\right) + R\left(\frac{y(1-x)}{1-xy}\right).
\label{abel}$$
In [@B11], we showed that the orthospectra of a hyperbolic surface satisfies the following generalized orthospectrum identity.
[(Bridgeman, [@B11])]{} Let $S$ be a finite area hyperbolic surface with non-empty totally geodesic boundary and $C_S$ boundary cusps. Then $$\sum_{l \in L_S} R\left(\frac{1}{\cosh^2\frac{l}{2}}\right) = \frac{\pi^{2}(6|\chi(S)| -C_{S})}{12}$$ \[Bidentity\]
[**Trilogarithm:**]{}
By definition, the trilogarithm function is given by $$Li_3(z) = \int_0^z \frac{Li_2(t)}{t}\ dt$$
The trilogarithm also satisfies a number of identities. $$Li_3(z) + Li_3(-z) = \frac{1}{4}Li_3(z^2)
\label{tri1}$$ $$Li_3(-z) - Li_3(-z^{-1}) = - \frac{1}{6}\log^3(z) - \frac{\pi^2}{6}\log(z)
\label{tri2}$$ and $$Li_3(z)+Li_3(1-z) - Li_3(1-z^{-1}) = \zeta(3)+ \frac{1}{6}\log^3(z) + \frac{\pi^2}{6}\log(z)-\frac{1}{2}\log^2(z)\log(1-z)
\label{tri3}$$
If $l_i$ is an ortholength of $S$, we define $$a_i =\frac{1}{\cosh^2\left(\frac{l_i}{2}\right)}.$$ We will often use the spectrum $\{a_i\}$ instead of $\{l_i\}$. In the paper [@B11], we studied the measure $N$ and derived the following;
[(Bridgeman, [@B11])]{} There exists a smooth function $\rho:{\mathbb R}_+\times(0,1) \rightarrow{\mathbb R}_+$ such that $$dN = \rho_S(x)dx = \left(\frac{4C_{S}x^{2}}{\sinh^{2}(x)} + \sum_{a_i}\rho(x,a_i)\right) dx.$$ Furthermore $$\int \phi(x) \rho(x,a)dx = \int_0^{a}\int_1^\infty \frac{4\phi(L_a(x,y))L_a(x,y)}{(y-x)^2}dxdy$$ where $$L_a(x,y) = \frac{1}{2}\log\left|\frac{y(y-a)(x-1)}{x(x-a)(y-1)}\right|$$
[**The Moment Identities for Surface:**]{} If we apply the above theorem to the function $\phi(x) = 1$ then we recover the identity in theorem \[Bidentity\].
To find $M_k(S)$, we note that $M_{k}(S) = N_{k-1}(S)$. Therefore by the above, we let $\phi(x) = x^{k-1}$ to get $$M_k(S) = N_{k-1}(S) = \int_0^\infty x^{k-1}\rho_S(x)dx = \sum_{a_i} \left(\int_0^{a_i}\int_1^\infty \frac{4L^{k}_{a_i}(x,y)}{(y-x)^2}dxdy\right)+4C_S\left(\int_0^\infty \frac{x^{k+1}dx}{\sinh^2(x)}\right)$$ We then define $$F_k(a) = \int_0^{a}\int_1^\infty \frac{4L_{a}(x,y)^{k}}{(y-x)^2}dxdy = \frac{1}{2^{k-2}}\int_0^{a}\int_1^\infty \frac{\log\left|\frac{y(y-a)(x-1)}{x(x-a)(y-1)}\right|^{k}}{(y-x)^{2}}dxdy.$$ Integrating we have $$\int_0^\infty \frac{x^{k+1}dx}{\sinh^2(x)} = \frac{(k+1)\\!\zeta(k+1)}{2^{k}}$$ Therefore we obtain $$M_k(S) = \left(\sum_{l\in L_S} F_k\left(\frac{1}{\cosh^2(\frac{l}{2})}\right) + C_S \frac{(k+1)\\!\zeta(k+1)}{2^{k-2}}\right).$$ We note that as $\zeta(1) = \infty$, if $S$ has boundary cusps then $M_0(S)$ is not finite. This corresponds to the fact that $M_0(S) = Length(\partial S)$ which is infinite in the case of boundary cusps.
Functions $F_0, F_1$ are given in terms of simple logarithms and dilogarithms respectively. An induction argument shows that $F_k$ can be written as a sum of polylogarithm functions of order at most $k+1$. We will calculate an explicit formula for $F_2$ in terms of trilogarithms in the next section. This will give us the formula for the average hitting time $A(S)$ for geodesic flow described in theorem \[avehit\].
A Somewhat Brutal Calculation
=============================
We will now obtain an explicit formula for the average hitting time in the surface case in terms of sums of polylogarithms evaluated at ortholengths. We let $F = F_2$ given by the above integral formula. Then $$F(a) = \int_0^{a}\int_1^\infty \frac{\log\left|\frac{y(y-a)(x-1)}{x(x-a)(y-1)}\right|^{2}}{(y-x)^{2}}dxdy.$$ Using Mathematica to calculate the indefinite integral first, gives 7,858 polylogarithm terms which then need to be evaluated at the 4 limits to give a final total of approximately 30,000 terms. Also the terms must be grouped so that evaluation gives a finite limit. As this seems a daunting task, we do the calculation directly using hyperbolic relations to simplify as we go along. The calculation is somewhat tedious but the final answer surprisingly short.
For the reader who would rather skip the long and tedious calculation, evidence for its validity is given by figure \[numerical\], which is a plot of the difference between the polylogarithm formula for $F(a)$ and and its values using numerical integration. As can be seen from the plot, the difference is less than $10^{-6}$ indicating they are the same function.
![Difference Between Numerical Integration of F and Polylogarithm Formula[]{data-label="numerical"}](numerical.jpg){width="5in"}
We have that for $a \in (0,1)$ $$F(a) = \int_0^a \int_1^\infty \frac{\log\left(\frac{y(y-a)(y-1)}{x(x-a)(y-1)}\right)^2}{(y-x)^2}dydx$$ Decomposing into cross-ratios, we have $$F(a) = \int_0^a \int_1^\infty \frac{\left(\log\left|\frac{y(x-1)}{x(y-1)}\right|+\log\left|\frac{y-a}{x-a}\right|\right)^2}{(y-x)^2}dydx =\int_0^a \int_1^\infty \frac{\log\left|\frac{y(x-1)}{x(y-1)}\right|^2 +2.\log\left|\frac{y(x-1)}{x(y-1)}\right|.\log\left|\frac{y-a}{x-a}\right| +\log\left|\frac{y-a}{x-a}\right|^2}{(y-x)^2}dydx$$ Under the mobius transformation $m(z) = a/z$, we let $X= m(x), Y = m(y)$, then by invariance of cross ratios, $$\frac{y(x-1)}{x(y-1)} = \frac{(y-0)(x-1)}{(x-0)(y-1)} = \frac{(m(y)-m(0))(m(x)-m(1)}{(m(x)-m(0)(m(y)-m(1)} = \frac{(Y-\infty)(X-a)}{(X-\infty)(Y-a)} = \frac{X-a}{Y-a}$$ Thus $$\int_0^a \int_1^\infty \frac{\log\left|\frac{y(x-1)}{x(y-1)}\right|^2}{(y-x)^2}dydx = \int_\infty^1 \int_a^0 \frac{\log\left|\frac{X-a}{Y-a}\right|^2}{(Y-X)^2}dYdX = \int_0^a \int_1^\infty \frac{\log\left|\frac{y-a}{x-a}\right|^2}{(y-x)^2}dydx$$ $$F(a) = 2.\int_0^a \int_1^\infty \frac{\log\left|\frac{y-a}{x-a}\right|^2}{(y-x)^2}dydx + 2.\int_0^a \int_1^\infty \frac{\log\left|\frac{y(x-1)}{x(y-1)}\right|.\log\left|\frac{y-a}{x-a}\right|}{(y-x)^2}dydx$$ We write this as $F(a) = 2I_1(a) + 2I_2(a)$, where $I_1, I_2$ are the above integrals.
In order to calculate the above integrals we will need the following integral equations. $$\begin{aligned}
\int \frac{\log(x)}{x-a} dx &=& \log(x)\log(1-x/a)+Li_2(x/a)\\
\int \frac{\log(x)^2}{x-a} dx &=& \log(x)^2\log(1-x/a)+2\log(x)Li_2(x/a)-2Li_3(x/a)\\
\int \frac{\log(x)\log(x-a)}{x} dx &=& \frac{1}{2}\log(x)^2\log(a)-\log(x)Li_2(x/a)+Li_3(x/a)\end{aligned}$$
Integral $I_1$
--------------
$$I_1(a) = -\log(1-a)\log^2(a)+\log^3(1-a)-4\log\left(\frac{a}{1-a}\right)Li_2(a)-6Li_3\left(\frac{a}{a-1}\right)$$
[**Proof:**]{} We decompose $I_1$ to obtain $$I_1 = \int_0^a\int_1^\infty \frac{ \log(y-a)^2 - 2\log(y-a)\log(a-x) + \log(a-x)^2}{(y-x)^2}dydx = J_1 -2J_2 + J_3$$ [**Integral $J_1$:**]{}We have $$J_1 = \int_1^\infty \log(y-a)^2\left(\int_0^a \frac{dx}{(y-x)^2}\right)dy = \int_1^\infty \log(y-a)^2\left(\frac{1}{y-a} - \frac{1}{y}\right)dy.$$ Then $$\int \frac{\log(y-a)^2}{y-a}dy = \frac{\log(y-a)^3}{3}$$ Also by integral equations above $$\int \frac{\log(y-a)^2}{y}dy = \log\left(\frac{y}{a}\right)\log(y-a)^2+2\log(y-a)Li_2\left(1-\frac{y}{a}\right)-2Li_3\left(1-\frac{y}{a}\right)$$ Thus $$J_1 = -\frac{\log(1-a)^3}{3}+ \frac{\pi^2}{3}\log(a)+\frac{1}{3}\log(a)^3+
\log\left(\frac{1}{a}\right)\log(1-a)^2+2\log(1-a)Li_2\left(1-\frac{1}{a}\right)-2Li_3\left(1-\frac{1}{a}\right)$$
[**Integral $J_2$:**]{} By parts we have $$J_2 = \int_0^a \log(a-x)\left(\int_1^\infty \frac{\log(y-a)}{(y-x)^2}dy\right)dx = \int_0^a\log(a-x)\left(\frac{\log(1-a)}{1-x}-\frac{\log(1-a)}{a-x}+\frac{\log(1-x)}{a-x}\right)dx$$ As above we have $$\int _0^a \frac{\log(a-x)}{1-x} dx = \left.\left(-\log(a-x)\log\left(\frac{1-x}{1-a}\right) -Li_2\left(\frac{a-x}{a-1}\right)\right)\right|_0^a = -\log(a)\log(1-a)+Li_2\left(\frac{a}{a-1}\right).$$ Also $$\int\frac{\log(a-x)}{a-x}dx = -\frac{1}{2}\log(a-x)^2$$ $$\int \frac{\log(a-x)\log(1-x)}{a-x}dx =-\frac{1}{2}\log(a-x)^2\log(1-a)+\log(a-x)Li_2\left(\frac{a-x}{a-1}\right) - Li_3\left(\frac{a-x}{a-1}\right)$$ Combining we get $$\int_0^a\frac{\log(a-x)(\log(1-x)-\log(1-a))}{a-x}dx =\left. \log(a-x)Li_2\left(\frac{a-x}{a-1}\right) - Li_3\left(\frac{a-x}{a-1}\right) \right|_0^a = -\log(a)Li_2\left(\frac{a}{a-1}\right)+Li_3\left(\frac{a}{a-1}\right)$$ Thus $$J_2 = -\log(a)\log(1-a)^2-\log\left(\frac{a}{1-a}\right)Li_2\left(\frac{a}{a-1}\right)+Li_3\left(\frac{a}{a-1}\right)$$ [**Integral $J_3$:**]{}
Similarly to $J_1$ we have $$J_3 = \int_0^a \log(a-x)^2 \left(\int_1^\infty\frac{1}{(y-x)^2}dy\right) dx = \int_0^a \frac{\log(a-x)^2}{1-x} dx$$ Therefore $$\begin{aligned}
J_3 =\left. -\log(a-x)^2\log\left(\frac{1-x}{1-a}\right) - 2\log(a-x)Li_2\left(\frac{a-x}{a-1}\right)+2Li_3\left(\frac{a-x}{a-1}\right)\right|_0^a\\
-\log(1-a)\log(a)^2+2\log(a)Li_2\left(\frac{a}{a-1}\right)-2Li_3\left(\frac{a}{a-1}\right)\end{aligned}$$
Thus we have the formula for $I_1 = J_1-2J_2+J_3$ giving $$\begin{aligned}
I_1 &= \left(-\frac{\log(1-a)^3}{3}+ \frac{\pi^2}{3}\log(a) +\frac{1}{3}\log(a)^3-\log(a)\log(1-a)^2+2\log(1-a)Li_2\left(1-\frac{1}{a}\right)-2Li_3\left(1-\frac{1}{a}\right)\right)\\ &-2\left(-\log(a)\log(1-a)^2-\log\left(\frac{a}{1-a}\right)Li_2\left(\frac{a}{a-1}\right)+Li_3\left(\frac{a}{a-1}\right)\right)\\&+\left(-\log(1-a)\log(a)^2+2\log(a)Li_2\left(\frac{a}{a-1}\right)-2Li_3\left(\frac{a}{a-1}\right)\right)\end{aligned}$$ Using the identities \[euler\], \[tri2\], we get $$I_1(a) = -\log(1-a)\log\left(\frac{a}{1-a}\right)^2+4\log\left(\frac{a}{1-a}\right)Li_2\left(\frac{a}{a-1}\right)-6Li_3\left(\frac{a}{a-1}\right)$$ Simplifying further we also get $$I_1(a) = -\log(1-a)\log^2(a)+\log^3(1-a)-4\log\left(\frac{a}{1-a}\right)Li_2(a)-6Li_3\left(\frac{a}{a-1}\right)$$
Integral $I_2$
--------------
$$\begin{aligned}
I_2(a) &=&2\zeta(3)+\frac{2\pi^2}{3}\log(1-a)+\frac{1}{3}\log^3(1-a)- \log^2(1-a)\log(a)-\log^2(a)\log(1-a)\\
&&-4\log(a)Li_2(a)+4Li_3(a)-2Li_3\left(\frac{-a}{1-a}\right)-2Li_3(1-a)\end{aligned}$$
[**Proof**]{} We have $$I_2 = \int_0^a \int_1^\infty\frac{\log\left|\frac{y(x-1)}{x(y-1)}\right|.\log\left|\frac{y-a}{x-a}\right|}{(y-x)^2}dydx$$
Again we decompose into integrals $$J_1 = \int_0^a \int_1^\infty \frac{\log\left(\frac{y}{y-1}\right)\log(y-a)}{(y-x)^2}dydx$$ $$J_2 = \int_0^a \int_1^\infty \frac{\log\left(\frac{1-x}{x}\right)\log(y-a)}{(y-x)^2}dydx$$ $$J_3 = \int_0^a \int_1^\infty \frac{\log\left(\frac{x}{1-x}\right)\log(a-x)}{(y-x)^2}dydx$$ $$J_4 = \int_0^a \int_1^\infty \frac{\log\left(\frac{y-1}{y}\right)\log(a-x)}{(y-x)^2}dydx$$
As before we can integrate to get $$J_1 = \int_1^\infty \log\left(\frac{y}{y-1}\right)\log(y-a)\left(\frac{1}{y-a}-\frac{1}{y}\right)dy$$ and $$J_3 = \int_0^a \frac{\log\left(\frac{x}{1-x}\right)\log(a-x)}{1-x}dx$$ We integrate in $J_2, J_3$ the simple factor to get
$$J_2 = \int_0^a \log\left(\frac{1-x}{x}\right)\left(\int_1^\infty \frac{\log(y-a)}{(y-x)^2}dy\right)dx = \int_0^a \log\left(\frac{1-x}{x}\right)\left(\frac{\log(1-x)-\log(1-a)}{a-x}+\frac{\log(1-a)}{1-x}\right)dx$$ Combining $J_2, J_3$ we get $$J_2 + J_3 = \int_0^a \log\left(\frac{1-x}{x}\right)\left(\frac{\log\left(\frac{1-x}{1-a}\right)}{a-x}-\frac{\log\left(\frac{a-x}{1-a}\right)}{1-x}\right)dx$$
The Roger’s normalized dilogarithm is given by $R(x) = Li_2(x) + \frac{1}{2}\log|x|\log(1-x)$ for $x <1$. We note that $$R'(x) = -\frac{1}{2}\left(\frac{\log(1-x)}{x}+\frac{\log|x|}{1-x}\right)$$ Thus we let $R_1(x) = R(\frac{a-x}{1-x})$ and note that $$R_1'(x) = \frac{1}{2}\left(\frac{\log\left(\frac{a-x}{1-a}\right)}{1-x}-\frac{\log\left(\frac{1-x}{1-a}\right)}{a-x}\right)$$ Thus $$J_2 + J_3 = 2\int_0^a \log\left(\frac{1-x}{x}\right)R_1'(x) dx$$ Considering $J_4$ we have similarly $$J_4 = \int_1^\infty \log\left(\frac{y-1}{y}\right) \left(\int_0^a \frac{(\log(a-x)}{(y-x)^2}dx\right)dy =\int_1^\infty \log\left(\frac{y-1}{y}\right) \left(\frac{\log(y-a)-\log(y)+\log(a)}{y-a}-\frac{\log(a)}{y}\right)dy$$ We then have $$J_1+J_4 = \int_1^\infty \log\left(\frac{y-1}{y}\right) \left(\frac{\log\left(\frac{a}{y}\right)}{y-a}+\frac{\log\left(\frac{y-a}{a}\right)}{y}\right)dy$$ We now let $R_2(y) = R(a/y)$, then $$R_2'(y) = \frac{1}{2}\left(\frac{\log\left(\frac{a}{y}\right)}{y-a}+\frac{\log\left(\frac{y-a}{a}\right)}{y}\right)$$ giving $$J_1+J_4 = 2\int_1^\infty \log\left(\frac{y-1}{y}\right)R_2'(y)dy$$ Letting $x = a/y$ we have
$$J_1 + J_4 = 2\int_a^0 \log\left(\frac{a-x}{a}\right)R'(x)dx = 2\log(a)R(a)- 2\int_0^a\log(a-x)R'(x)dx$$
Similarly let $u = (a-x)/(1-x)$ then $x = (a-u)/(1-u)$ and $$J_2 + J_3 = 2\int_{0}^a \log\left(\frac{1-a}{a-u}\right)R'(u) du = 2\log(1-a)R(a)- 2\int_0^a \log(a-x)R'(x)dx$$ Giving $$I_2 = 2\log(a(1-a))R(a)-4\int_0^a \log(a-x)R'(x)dx = 2\log(a(1-a)R(a) + 2\int_0^a \log(a-x) \left(\frac{\log(1-x)}{x}+\frac{\log(x)}{1-x}\right)dx$$
We note the formula $$\begin{aligned}
G(x,a)&=&\int \frac{\log(a-x)\log(1-x)}{x}dx = \\
&&\log(1-x)\log\left(\frac{a-x}{a}\right)\log\left(\frac{x}{a}\right) + \log(a)\log(1-x)\log(x)+ \frac{1}{2}\log(a)\log^2(1-x)\\
&+&\log(a(1-x))Li_2(1-x) +\log\left(\frac{a-x}{a}\right)Li_2\left(\frac{a-x}{a}\right) + \log\left(\frac{a-x}{a(1-x)}\right)\left(Li_2\left(\frac{a-x}{1-x}\right)- Li_2\left(\frac{a-x}{a(1-x)}\right)\right)\\
&-&Li_3(1-x)-Li_3\left(\frac{a-x}{1-x}\right) - Li_3\left(\frac{a-x}{a}\right) + Li_3\left(\frac{a-x}{a(1-x)}\right)\end{aligned}$$
Thus taking limits we have $$\int_0^a \frac{\log(a-x)\log(1-x)}{x}dx =$$ $$\zeta(3)-\frac{\pi^2}{6}\log(a)+\frac{1}{2}\log(a)\log^2(1-a)+\log(1-a)\log^2(a)+\log(a(1-a))Li_2(1-a) +Li_3(a)-Li_3(1-a)$$ From the above $$\begin{aligned}
H(x,a)&=&\int \frac{\log(a-x)\log(x)}{1-x}dx = -G(1-a,1-x)= \\
&-&\log(x)\log\left(\frac{a-x}{1-a}\right)\log\left(\frac{1-x}{1-a}\right) - \log(1-a)\log(1-x)\log(x)- \frac{1}{2}\log(1-a)\log^2(x)\\
&-&\log((1-a)x)Li_2(x) -\log\left(\frac{a-x}{1-a}\right)Li_2\left(\frac{x-a}{1-a}\right) \\
&-& \log\left(\frac{a-x}{(1-a)x}\right)\left(Li_2\left(\frac{x-a}{x}\right)- Li_2\left(\frac{x-a}{(1-a)x}\right)\right)\\
&+&Li_3(x)+Li_3\left(\frac{x-a}{x}\right) + Li_3\left(\frac{x-a}{1-a}\right) - Li_3\left(\frac{x-a}{(1-a)x}\right)\end{aligned}$$ Thus taking limits we have $$\int_0^a \frac{\log(a-x)\log(x)}{1-x}dx =$$ $$\frac{\pi^2}{6}\log(1-a)-\frac{1}{3}\log^3(1-a) -\log(1-a)\log^2(a)- \log((1-a)a)Li_2(a) +\log\left(\frac{a}{1-a}\right)Li_2\left(\frac{-a}{1-a}\right)+Li_3(a)-Li_3\left(\frac{-a}{1-a}\right)$$ We now combine to obtain $$\begin{aligned}
I_2(a) &=&2\zeta(3)+\frac{2\pi^2}{3}\log(1-a)+\frac{1}{3}\log^3(1-a)- \log^2(1-a)\log(a)-\log^2(a)\log(1-a)\\
&&-4\log(a)Li_2(a)+4Li_3(a)-2Li_3\left(\frac{-a}{1-a}\right)-2Li_3(1-a)\end{aligned}$$
Finally we combine $I_1, I_2$ to get $$\begin{aligned}
F(a)& = 4\zeta(3) +\frac{4\pi^2}{3}\log(1-a)+\frac{8}{3}\log^3(1-a)-4\log(1-a)\log^2(a)-2\log^2(1-a)\log(a)\\
&-8\log\left(\frac{a^2}{1-a}\right)Li_2(a)+8Li_3(a)-4Li_3(1-a)-16Li_3\left(\frac{-a}{1-a}\right)\end{aligned}$$ Using the identity \[tri3\] we get $$\begin{aligned}
F(a) = &&-12\zeta(3)-\frac{4\pi^2}{3}\log(1-a)+6\log^2(1-a)\log(a)-4\log(1-a)\log^2(a)\\
&&-8\log\left(\frac{a^2}{1-a}\right)Li_2(a)+24Li_3(a)+12Li_3(1-a)\end{aligned}$$
The function has boundary values $F(0) = 0$ and $F(1) = 12\zeta(3)$ and is maximized at $a = .754493$ with value $17.9804$.
Below is a graph of $F$.
![Function F(x)[]{data-label="graphF"}](graphF.jpg){width="5in"}
[1]{} A. Basmajian, The orthogonal spectrum of a hyperbolic manifold, , 115, 5, 1139–1159, 1993.
F. Bonahon, The geometry of Teichmüller space via geodesic currents, (1988), 139–162.
Martin Bridgeman, Orthospectra and Dilogarithm Identities on Moduli Space. , Volume 15, Number 2, 2011
Martin Bridgeman, David Dumas, Distribution of intersection lengths of a random geodesic with a geodesic lamination. , 27(4), 2007
Martin Bridgeman, Jeremy Kahn, Hyperbolic volume of n-manifolds with geodesic boundary and orthospectra. , Volume 20(5), 2010
D. Calegari, Bridgeman’s orthospectrum identity, , 38, 173–179, 2011
D. Calegari, Chimneys, leopard spots, and the identities of Basmajian and Bridgeman, , 10(3), 1857–1863, 2010
L. Lewin, (Ed.). Structural Properties of Polylogarithms, [*Mathematical Surveys and Monographs*]{}, AMS, Providence, RI, 1991.
B. Maskit, Kleinian Groups, [*Graduate Texts in Mathematics*]{}, Springer-Verlag, 1987.
Peter J. Nicholls. , volume 143 of [ *London Mathematical Society Lecture Note Series*]{}. Cambridge University Press, Cambridge, 1989.
L.J. Rogers. On Function Sum Theorems Connected with the Series $\sum_{1}^{\infty} \frac{x^n}{n^2}$ [*Proc. London Math. Soc.*]{} 4, 169-189, 1907
D. Sullivan, The density at infinity of a discrete group of hyperbolic motions, [*Publ. Math. IHES,*]{} [**50**]{} (1979), pp. 171-202.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Spatial multiplexing (SM) gains in multiple input multiple output (MIMO) cellular networks are limited when used in combination with ultra-dense small cell networks. This limitation is due to large spatial correlation among channel pairs. More specifically, it is due to i) line-of-sight (LOS) communication between user equipment (UE) and base station (BS) and ii) in-sufficient spacing between antenna elements. We propose to shape transmit signals at adjacent antennas with distinct interpolating filters which introduces pulse shaping diversity eventually leading to improved SINR and throughput at the UEs. In this technique, each antenna transmits its own data stream with a relative offset with respect to adjacent antenna. The delay which must be a fraction of symbol period is interpolated with the pulse shaped signal and generates a virtual MIMO channel that leads to improved diversity and SINR at the receiver. Note that non-integral sampling periods with inter-symbol interference (ISI) should be mitigated at the receiver. For this, we propose to use a fractionally spaced equalizer (FSE) designed based on the minimum mean squared error (MMSE) criterion. Simulation results show that for a $2\times2$ MIMO and with inter-site-distance (ISD) of 50 m, the median received SINR and throughput at the UE improves by a factor of 11 dB and 2x, respectively, which verifies that pulse shaping can overcome poor SM gains in ultra-dense small cell networks.'
author:
- |
Amir H. Jafari[$^{12}$]{}, Vijay Venkateswaran[$^{3}$]{}, David López-Pérez[$^{2}$]{}, Jie Zhang[$^{1}$]{}\
Dept. of Electronic & Electrical Engineering, University of Sheffield, Sheffield S1 3JD, UK\
[$~^{2}$]{}Bell Laboratories Nokia, Dublin, Ireland\
[$~^{3}$]{}Huawei Technologies, Sweden\
bibliography:
- 'references.bib'
title: 'Pulse Shaping Diversity to Enhance Throughput in Ultra-Dense Small Cell Networks'
---
Introduction
============
One of the most promising approaches to meet the data deluge and to enhance network capacity is small cell densification, which benefits from extensive spatial reuse of the spectrum [@2015Lopez]. Having multiple antennas and exploiting MIMO technology can further enhance the network capacity [@1532224] [@LTE]. Applying MIMO tools such as spatial multiplexing in ultra-dense small cell networks introduces new challenges due to different propagation conditions when compared to macro cell scenarios. For instance, spatial multiplexing (SM) gain typically improves when the MIMO spatial channels are uncorrelated. Ultra-dense small cell networks are limited by channels being correlated and SM gains are very limited. Two phenomena contribute to this channel correlation. Firstly, antennas at both user equipment (UE) and base station (BS) are placed very closely to each other ($\sim$ half wavelength). Secondly, due to proximity of both UE and BS, there is a high probability of LOS communication resulting in high spatial correlation. Thus the communication channel tends to be ill-conditioned, lowering the number of independent parallel data streams that can be simultaneously multiplexed and transmitted and, therefore, the throughout is considerably degraded.
### **Central Idea** {#central-idea .unnumbered}
In order to improve throughput in ultra-dense small cell networks, we propose a novel transmission technique using distinct pulse shapes to modulate adjacent antennas’ data streams. We refer to this technique as diversity pulse shaped transmission (DPST).
One way to view DPST is that adjacent antenna element signals are shaped with slightly different band limited pulse shaping filters. This change will introduce delay and ISI in time domain. This delay which must be a *fraction of symbol period* allows the UE receiver to sense multiple delayed replicas of the transmitted data stream. Consider a $2\times2$ MIMO setup for simplicity, the receive antennas would observe antenna 1 transmitting its data stream with symbol period $T_s$, as well as antenna 2 transmitting its data stream; however, with delay $\tau$ with respect to antenna 1 ($0 < \tau < T_{s}$) while being sampled at $T_s$. This implies that in a LOS scenario, a receiver that is synchronized to symbol period $T_s$ (ignoring the bulk delays) would observe a direct path with data stream 1 as well as a delayed path with data stream 2, with the latter corrupted by inter-symbol interference (ISI).
The improved diversity is somewhat inspired from faster than Nyquist (FTN) signaling [@6479673] [@4777625] [@1231648]. In FTN, data streams are sampled and transmitted at a fraction of symbol period, eventually leading to an improvement in communication rates at the cost of complicated receiver to combat ISI. In our case, to account for the increased ISI in the system, we propose to use a fractionally spaced equalizer (FSE) at the UE that operates on the precoded data streams and wireless channel output and eventually leads to improved diversity gain in ultra-dense small cell networks [@489269]. In order to ensure a reasonable estimate of multipath signals, minimum mean squared error (MMSE) criterion is used to design the equalizer. Indeed, DPST can also be seen as an application of FTN and FSE to enhance data rates.
### **Contributions** {#contributions .unnumbered}
In this paper, we present DPST whose arrangement based on pulse shaping diversity and oversampled receiver, improves the overall dimensionality of the transmitted multi-antenna data streams viewed at the receiver and consequently enhance the data rates. Moreover, we evaluate the performance of DPST in a single tier hexagonal multi small cell layout. In more detail, we quantify the degradation of SINR due to spatial correlation for a $2\times2$ MIMO system at different inter-site-distances (ISDs), i.e., 20 m, 50 m and 150 m, and show that the proposed DPST leads to a 50%-tile SINR improvement of 11 dB at an ISD of 50 m. We also show that DPST can almost enhance the UE throughput by 2x at all ISDs.
It is important to highlight that DPST differs from cyclic delay diversity (CDD) [@4656965] [@989781] [@4109390], a diversity technique used in LTE. In CDD, the antennas transmit a cyclically shifted version of the same signal to achieve diversity gain, whereas in DPST each antenna transmits an individual signal with a fractional delay with respect to its former antenna which requires to redesign the precoder as well as the receiver in order to enhance spatial multiplexing gains. We also feel it is necessary to note that while exploiting antenna polarization at the transmitter is an effective technique in MIMO systems, it is usually limited to two transmit antennas [@1192168]. In contrast, DPST can be applied to larger MIMO systems subject to derivation of the optimized fractional amount that is applied to transmission of each transmit antenna.
The rest of this paper is as follows. Section \[sec:part\_DPST\] details the proposed DPST. Section \[sec:DPST\_rec\] explains the DPST receiver design, precoder and equalizer to estimate the transmitted signals. Section \[sec:part\_Simulation\] presents a performance evaluation/comparison of DPST with existing MIMO systems. Section \[sec:part\_Conclusion\] draws the conclusions.
Diversity Pulsed Shaped Transmission {#sec:part_DPST}
====================================
Channel Correlation Model
-------------------------
The MIMO performance is highly dependent on the overall channel taps being spatially uncorrelated. In correlated small cell channel scenarios, different transmit-receive antenna pairs will experience similar channel conditions, and as a result the multipath components corresponding to different pairs may not be resolvable by the UE [@1203167]. Thus correlated channels have reduced degrees of freedom, leading to reduced throughputs as detailed in [@1459054] [@892194].
The Rician multipath fading model used to capture the LOS communication can be used to represent the MIMO channel as $$\textbf{H} = \sqrt{\frac{K}{K+1}} {\textbf{H}_{i}}^{\rm LOS} + \sqrt{\frac{1}{K+1}}{\textbf{H}_{s}}^{\rm NLOS},
\label{eq:LOS_NLOS}$$ where $\rm K$ is the Rician K factor [@2015Jafari] modeled as a function of the UE-BS distance *d*, i.e., $$\rm K=\begin{cases}
32 & \text{if $d<18 m$} \\
140.10 \times \exp(-0.107 \times d) & \text{otherwise},
\end{cases}
\label{eq:Kfactor}$$ and ${\textbf{H}_{i}}^{\rm LOS}$ and ${\textbf{H}_{s}}^{\rm NLOS}$ are the identity LOS MIMO channel matrix and the correlated non-LOS (NLOS) MIMO channel matrix, respectively.
It is important to note that condition number ($\mathcal{K}$) is a metric to denote the number of independent streams of the wireless channel, and corresponds to the ratio of the maximum to minimum singular values of the wireless channel $\bH$ [@Matrix]. $\mathcal{K} \approx 1$ implies no correlation between channel pairs, and as long as $\mathcal{K}$ is less than $10$, the channel is regarded as well-conditioned, and can be leveraged to extract the unitary vectors of the channel for precoding and spatial multiplexing purposes. In ultra-dense small cell networks, $\mathcal{K}$ is considerably larger than 10 [@2015Jafari], and thus the channel becomes ill-conditioned and spatial multiplexing gain suffers.
MIMO Link Model with Pulse Shaping {#ssec:Transmitter}
----------------------------------
For simplicity, we will model $2\times2$ MIMO channel used with DPST. Note that, DPST can be applied to arbitrary sized arrays. In the $2\times2$ MIMO system, DPST works by shaping the data streams of the second antenna with respect to the first antenna such that in downlink communications, the multipath components from second antenna arrive at the UE with a time offset when compared to that of the first antenna. This enhances diversity in *fractionally delayed* multipath components of the closely placed transmit antennas in a MIMO system, and allows to improve the UE throughput. Fig. \[fig:DPST\_Block\] provides a block diagram of the proposed setup showing pulse shaping at the transmit side.
The fractionally delayed signaling can be related to FTN communication, where a non-orthogonal sampling kernel is used to allow for signaling above the Nyquist limit while introducing ISI. In traditional cases, we typically use orthogonal sampling kernel such as a sinc pulse where an integral multiple of the symbol period contains only a single non-zero component of the transmitted signal. In FTN shaping, the signals operate above the Nyquist rate and the ISI that is intentionally introduced through oversampling results in enhanced system capacity [@6479673] [@4777625] [@1231648]. Along similar lines, the *fractional* delay imposed on the transmission of the second antenna in DPST injects additional *deterministic* ISI to the system, which suggests the analogy between the performances of DPST and FTN. Fig. \[fig:FTN\_DPST\] intuitively shows the implication of DPST. Fig. \[fig:FTN\_DPST\]a shows signals sampled at *integral* multiples of symbol period with an orthogonal sampling kernel, while Fig. \[fig:FTN\_DPST\]b shows sampling at *non-integral* multiples of symbol period. Note that in the latter case and in contrast with the former, at each sampling instant, there are multiple non-zero samples viewed by the sampling kernel. As a result, this can be exploited to increase the diversity of the wireless channel seen between the closely placed transmit and receive antennas.
When viewed from the receiver perspective, the pulse shaping operation at the transmitter can be perceived as a fractionally delayed transmission with respect to the first transmit antenna where the delay $\tau$ is a fraction of the transmitted symbol period $T_s$, i.e., $0 < \tau < T_s$, and $T_s$ is assumed to be $1$. The received signal can be expressed as $$\bea{ccc}
\left[\bea{c}
x_{1}(t) \\ x_{2}(t)
\ena\right]
& = &
\left[\bea{cc}
h_{1,1}(t) & h_{1,2}(t)
\\ h_{2,1}(t) & h_{2,2}(t)
\ena\right]
\ast
\left[\bea{c}
s_{1}(t) \\ s_{2}(t+\tau)
\ena\right]
\vspace{0.2cm}
\\
\bx(t) & = & \bH(t) \ast \bs(t)
\ena$$ where $\textbf{H}(t)$ represents a continuous time version of the $2\times2$ correlated MIMO channel $\bH$, $s_{1}(t)$ and $s_{2}(t)$ are the signals transmitted by first and second transmit antennas, respectively, and $\ast$ is the convolution operation. In discrete time domain, the delay $\tau$ imposed to the transmission of the second antenna is alternatively modeled by oversampling/interpolating the transmit data streams as shown in the following
= ( ) \[eq:21\] where $\textbf{\ensuremath{\mathcal{I}}}(0)$ and $\textbf{\ensuremath{\mathcal{I}}}(\tau)$ are interpolation kernels that are applied to the first and second transmit antennas, respectively. The elements of $N\times M$ interpolation matrix are obtained as $$\ensuremath{\mathcal{I}}_{nm} = sinc (\frac{n(\frac{T_{s}}{N})+ \tau -m(\frac{T_{s}}{M})}{\frac{T_{s}}{M}}) \quad
\begin{cases}
\textit{m} = 1,2,...,M \\ \textit{n} = 1,2,...,N
\end{cases}
\label{eq:22}$$ where $M$ is the number of input signal samples and $N = M R$ with $R$ being the oversampling ratio [@5706377]. It is evident that the interpolation matrix corresponding to antenna 1 has no time offset, while the one corresponding to antenna 2 is offset by time $\tau$ to account for the delayed pulse shaping. For more discussion on the modeling of oversampled analog signals refer to [@5706377]. Also, note that the transmitted signals in antenna 1 and antenna 2 are not matched anymore, and thus the corresponding signal degradation is taken into account.
![Analogous performance of DPST and FTN.[]{data-label="fig:FTN_DPST"}](FTN_DATS.pdf)
The delay $\tau$ plays a key role in performance of DPST. A delay of $\tau=0$, results in expression (\[eq:21\]) to collapse into the ill-conditioned wireless channel $\bH$, and does not improve diversity. For a delay that is an integral multiple of symbol period $\tau=k T_s$ ($k = 1,2,...$), the integral DPST provides a cyclic shift of transmitted data streams at the receiver. We can interpret this integral delay as a variation of CDD [@4656965] [@989781] [@4109390] which does not lead to significant improvements in throughput. However, when the delay is a fraction of symbol period $\tau \neq k T_s$ ($k = 1,2,...$), the matrix $\ensuremath{\mathcal{I}}(\tau)$ becomes a block diagonal matrix operating on streams of the input sequence $\{s_2[1], \, s_2[2] \, \cdots s_2[k]\}$. The latter scenario is a consequence of using distinct pulses which generates deterministic ISI, thereby increasing the order of the composite channel between the transmit and receive streams as shown in Fig. \[fig:DPST\_Block\]. Indeed, the signals corresponding to adjacent antenna elements are shaped with pulse shaping filters having different bandwidths. The change in bandwidth will introduce a delay and hence ISI in time domain.
From the receiver perspective, the signals from the two transmit antennas are observed at $kT_{s}$ and $kT_{s}+\tau$ ($k = 1,2,...$). However, this delay can only be observed and the induced diversity can only be extracted when the receiver is operating at a rate much greater than the symbol rate. The next section discusses the DPST receiver design.
Receiver and Precoder design {#sec:DPST_rec}
============================
Receiver Design Considerations {#ssec:Receiver}
------------------------------
The sinc interpolation shown in (\[eq:21\]) can be interchanged and be alternatively represented as
= ( )
Interference is the main challenge in ultra-dense small cell networks (interference-limited) and, therefore, ignoring the noise term, the received signal at antenna 1 can be rewritten as $$x_{1}(t) =
\left[ \bea{cc}
\cI(0) \ast h_{1,1}(t) & \cI(\tau) \ast h_{1,2}(t)
\ena\right]
\left[\bea{c}
s_{1}(t) \\
s_{2}(t)
\ena \right]$$
At the receiver side, DPST requires the receiver to operate at a rate significantly greater than the symbol rate. This is done by sampling the received signals $P$ times (typically $P\geq 2$) within the time interval $t \in \left(kT_s, \, [k+1]T_s\right]$ where $t = (k+\frac{1}{P})T_s$ and is stacked as a $P\times1$ vector, i.e., $$\bea{ccc}
\left[ \bea{@{}c@{}}
x_{1}(t=kT_s) \\ \vdots \\ x_{1}(t=(k+\frac{P-1}{P})T_s)
\ena\right]
= \
\cI_{\rm R}
\left[\bea{@{}cc@{}}
\cI(0) \ast h_{1,1}(t) \\ \cI(\tau) \ast h_{1,2}(t)
\ena\right]^{T}
\left[\bea{@{}cc@{}}
s_{1}(t) \\
s_{2}(t)
\ena \right]
\vspace{0.3cm}
\ena$$
\_[1,os]{}\[k\] & = & \_[1,os]{}\[k\] where $\textbf{s}[k] = \textbf{s}(t = k T_{s})$ and $\textbf{x}_{1,os}$ is the interpolated received signal at the first antenna and is represented as a $P \times 1$ vector. $\cI_{R}$ is the $(N \times P) \times N$ sinc interpolation matrix similar to (\[eq:22\]) exploited at each receive antenna and $()^{T}$ denotes the transpose. Note that operating at $P$ times the symbol rate at the receiver allows the UE to suppress the ISI using an equalizer.
The received signal at antenna 2 is also sampled $P$ times and, therefore, stacking the $P \times 1$ vectors $\textbf{x}_{1,os}[k]$ and $\textbf{x}_{2,os}[k]$ from both receive antennas, we get
\_[os]{}\[k\] = = = \_[os]{} \[eq:27\]
It is realized from (\[eq:27\]) that the resultant channel matrix $\textbf{H}_{\rm os}$ is a tall matrix, which can be a full rank column matrix in case of sufficient pulse diversity. Indeed, due to induced pulse diversity and antenna array phase shifts, the columns of $\textbf{H}_{1,\rm {os}}$ and $\textbf{H}_{2,\rm {os}}$ are independent of each other. This implies that the rank of the composite channel $\textbf{H}_{\rm os}$ is greater than 1, in contrary to the initial correlated channel $\textbf{H}$ with rank 1. This rank improvement is the outcome of
- the introduction of distinct pulse shapes and deterministic fractional delay, and
- higher order channel observed at the receiver due to oversampling.
Moreover, it has to be noted that while SINR at a given sampling instant can be degraded due to pulses interfering with each other, the overall SINR after receiver equalization is still enhanced by exploiting the pulse diversity.
From the receiver point of view, DPST enhances the channel degrees of freedom by virtually reducing the correlation between channel pairs. The overall channel $\textbf{H}_{\rm os}$ can be compactly represented as $$\textbf{H}_{\rm os} = \textbf{\ensuremath{\mathcal{I}}}_{\rm Rx} (\textbf{H} \ast \textbf{\ensuremath{\mathcal{I}}}_{\rm Tx}),
\label{eq:28}$$ where $\textbf{\ensuremath{\mathcal{I}}}_{\rm Tx}$ and $\textbf{\ensuremath{\mathcal{I}}}_{\rm Rx}$ consist of the interpolation matrices for both antennas at the transmitter and receiver, respectively. To downsize $\textbf{H}_{\rm os}$ with respect to $\textbf{H}$, the channel is decomposed using singular value decomposition (SVD) as $\textbf{H}_{\rm os} = \textbf{U}_{\rm os} $$_{\rm os} \textbf{V}_{\rm os}^{\rm {H}}$ and since $\textbf{H}$ is a $2\times2$ matrix, $\textbf{U}_{\rm os}$ and $\textbf{V}_{\rm os}$ only take the first two columns and rows. Therefore, the downsized channel is $$\textbf{H}_{\rm R} = \textbf{U}_{\rm os}(:,1:2)^{\rm {H}} \ \textbf{H}_{\rm os} \ \textbf{V}_{\rm os}(1:2,:)^{\rm {H}}$$ The downsized channel $\textbf{H}_{\rm R}$ is also normalized with respect to $\textbf{H}$ as $$\textbf{H}_{\rm N} = \textbf{H}_{\rm R} \times \frac{\parallel \textbf{H} \parallel_{\rm F}}{\parallel \textbf{H}_{\rm R} \parallel_{\rm F}}.$$ where $\parallel$ $\parallel_{\rm F}$ denotes the Frobenius norm.
Precoding and Detection {#ssec:prec_detec}
-----------------------
As previously discussed, the virtual channel $\textbf{H}_{\rm N}$ due to introduction of DPST is a channel with enhanced diversity. To perform closed loop precoding at the transmitter, we do SVD and extract the first two columns of $\textbf{V}_{\rm N}$ from $\textbf{H}_{\rm N} = \textbf{U}_{\rm N} $$_{\rm N} \textbf{V}_{\rm N}^{\rm {H}}$ to generate the precoding matrix denoted by $\textbf{W}$ where $\textbf{W} = \textbf{V}_{\rm N}(:,1:2)$. The precoding matrix is then subject to power scaling as the precoding must not violate the BS transmission power constraint, i.e., $$\textbf{W} = \sqrt{P_{ \rm BS}} \times \rho \times \textbf{W},$$ where $P_{\rm BS}$ is the BS transmit power and $\rho$ is the power scaling ratio. The precoded channel $\textbf{H}_{\rm eq}$ is then defined as $\textbf{H}_{\rm eq} = \textbf{H}_{\rm N} \textbf{W}$.
At the receiver side, minimum mean squared error (MMSE) equalizer is exploited and hence the receive filter is formulated from $\textbf{H}_{\rm eq}$ as $$\bea{ccc}
\textbf{F}^{\rm {MMSE}} & =& \textbf{H}_{\rm eq}^{\rm {H}}(\textbf{H}_{\rm eq}\textbf{H}_{\rm eq}^{\rm {H}} + \textbf{$\Phi$} + \textit{N}_{0}\textbf{I})^{-1},
\ena$$ where $()^{-1}$ and $()^{H}$ refer to inverse and Hermitian transpose, respectively, $\Phi$ is the inter-cell interference covariance matrix, and $\textit{N}_{0}$ is the noise power. The transmitted data stream is then estimated using $\textbf{F}^{\rm {MMSE}}$ as $$\hat{\bs}[k] = \bF^{\textrm{MMSE}} (\bU_{\textrm{os}}^{\rm H} \bH_{\textrm{os}} \ast \bs[k] + \bw[k])$$ where $\bw[k]$ is the additive noise terms at time instant $t=kT_s$ ($k = 1,2,...$).
Simulation Results {#sec:part_Simulation}
==================
In this section, we compare the performance of DPST with that of existing MIMO systems where all the antennas transmit at the same time instant.
We consider a single tier hexagonal layout of small cell BSs in a $500m \times 500m$ scenario with different ISDs to observe the impact of DPST on various degrees of network densification, which in turn impacts channel correlation. The central cell is designated as the serving cell and the remaining six cells are considered as interferers. The carrier frequency and corresponding bandwidth are 2 GHz and 10 MHz, respectively. Macrocell BSs operate in a different frequency band. Each cell consists of an array of two transmit antennas, and only serves a single UE (single user scenario), which has two antennas forming a $2\times2$ MIMO system. Each signal consists of 10 symbols with an oversampling ratio of 4. Antenna gain, path loss, lognormal shadowing and multipath Rician fast fading are included in SINR computation. The path loss model used is the microcell urban model defined in [@3gpp], which includes both the LOS and NLOS components. Closed-loop precoding is considered. Note that it is assumed that the channel is prone to single LOS component.
With regard to DPST, we assume that second transmit antenna is subject to the deterministic delay of $5$ nano sec with respect to first one. Note that the optimization of the precise amount of delay is left as part of future study. We also compare the performance of DPST with respect to the optimistic channel with condition number ($\mathcal{K}$) equals to 1. This is the optimal channel condition for spatial multiplexing, implying that all channels are orthogonal.
Fig. \[fig:MIMO2\] compares the effective SINR CDF for a $2\times2$ MIMO system where DPST is applied to the single tap LOS channel model (full correlation). Results show that in all tested ISDs, DPST considerably improves the effective SINR with respect to the fully correlated LOS channel, offering a close to optimal performance. In more detail, for ISDs of 20 m, 50 m and 150 m, the 50%-tile effective SINR is increased with respect to the LOS scenario by 6.5 dB, 11 dB and 14 dB, respectively. Performance from the optimum is less than 0.1 dB away. Similarly, Fig. \[fig:MIMO3\] shows the throughput CDF of a $2\times2$ MIMO system when DPST is applied to the LOS channel. For all tested ISDs, DPST enhances the throughput CDF by nearly 2x reaching its upper limit bound defined by optimistic channel condition. This implies that despite the presence of spatial correlation, applying DPST allows the two data streams to be simultaneously transmitted and successfully decoded by the UE receiver. This is in contrary to the correlated LOS channel scenario where the transmitted stream by second antenna can not be decoded by the UE receiver. Table \[tab:channel\_cond\] presents the rank and condition number of the virtual channel when DPST is applied. It is realized that DPST can significantly enhance the virtual channel condition number, verifying its almost optimum performance.
![Comparison of effective SINR CDF of LOS, DPST and optimum channels in a 2x2 MIMO at different ISDs.[]{data-label="fig:MIMO2"}](SINR.pdf)
![Comparison of throughput CDF of LOS, DPST and optimum channels in a 2x2 MIMO at different ISDs.[]{data-label="fig:MIMO3"}](Throughput.pdf)
Conclusion {#sec:part_Conclusion}
==========
In this paper, we discussed that in ultra-dense small cell networks, the presence of spatial channel correlation prevents us from achieving spatial multiplexing gain. We proposed a new technique referred to as diversity pulse shaped transmission (DPST), which exploits pulse shaping diversity in combination with delaying the transmission of one antenna with respect to another by a fraction of symbol period followed by a receiver that operates in the oversampled domain. The combined effect is able to virtually lower the channel correlation, improve the channel diversity from the perspective of the UE receiver. We studied the performance of the proposed technique under MMSE criteria and compared it with existing $2\times2$ MIMO cellular networks. We showed that DPST can significantly enhance the 50%-tile effective SINR by 11 dB and almost double the UE throughput at an ISD of 50 m. We realize that there is a limit on the number of antennas that benefit from DPST processing which depends on delay optimisation, and hence optimising the corresponding delay as a function of the number of antennas shapes the scope of future research.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The mass shift induced by one-loop quantum fluctuations on self-dual ANO vortices is computed using heat kernel/generalized zeta function regularization methods. The quantum masses of super-imposed multi-vortices with vorticity lower than five are given. The case of two separate vortices with a quantum of magnetic flux is also discussed.'
author:
- 'A. Alonso Izquierdo$^{(1)}$, W. Garcia Fuertes$^{(2)}$, M. de la Torre Mayado$^{(3)}$ and J. Mateos Guilarte$^{(4)}$'
title: 'Quantum oscillations of self-dual Abrikosov-Nielsen-Olesen vortices'
---
Introduction
============
In this paper we present some new results on the quantization of the self dual multi-vortex solutions of the Abelian Higgs model. We also take the opportunity to offer a detailed description of the concepts and techniques that allowed us to compute the one-loop quantum correction to the mass of self-dual Abrikosov-Nielsen-Olesen vortices with one quantum of magnetic flux in the Rapid Communication [@AGMT]. The AHM provides a theoretical ground in several fields of physics: it provides shape to interesting truncations of the electroweak or grand unified theories, it also provides the basis for the various phenomenological models for cosmic strings, or it can be used as a Ginzburg-Landau theory for superconductivity.
Interest in this research, developed in the supersymmetric framework in [@sc],[@le], was rekindled two years ago. Non-vanishing quantum corrections to the mass of $N=2$ supersymmetric vortices were recently reported in papers [@Vass] and [@Reb], see also [[^1]]{}. In the second paper, it was found that the central charge of the $N=2$ SUSY algebra receives a non-vanishing one-loop correction that is exactly equal to the one-loop mass shift; thus, one could talk in terms of one-loop BPS saturation. This result fits in a pattern first conjectured in [@Reb1] and then proved in [@Shif] for supersymmetric kinks. Another work by the authors of the Stony Brook/Viena group, [@Reb2] unveils a similar kind of behavior of supersymmetric BPS monopoles in $N=2$ SUSY Yang-Mills theory. In this reference, however, it is pointed out that (2+1)-dimensional SUSY vortices do not behave exactly in the same way as their (1+1)- and (3+1)-dimensional cousins. One-loop corrections in the vortex case are in no way related to an anomaly in the conformal central charge, contrarily to the quantum corrections for SUSY kinks and monopoles.
We shall focus, however, on the purely bosonic Abelian Higgs model and rely on the heat kernel/generalized zeta function regularization method that we developed in reference [@Aai1]. Our approach profits from the high-temperature expansion of the heat function, which is compatible with Dirichlet boundary conditions in purely bosonic theories. In contrast, the application of a similar regularization method to the supersymmetric kink requires SUSY-friendly boundary conditions, see [@Vass1]. In [@Aai1] the kink quantum correction in the $\phi^4$ model is estimated by this method and compared with the correct answer obtained from the Dashen-Hasslacher-Neveu formula, [@DHN] in order to check the reliability of our approach. The relative error found is approximately 0.07%. In [@Aai2] and [@Aai3] we also calculated the quantum mass corrections for kinks arising in two-component scalar models, where second-order small fluctuations are ruled by matrix differential operators. Therefore, we were led to generalize the zeta function method to the matrix case, because the DHN approach, based on a direct computation of the spectral density, is not efficient for matrix differential Schrodinger operators. This step has proved to be crucial, opening the possibility of applying our method to two-dimensional topological defects in the Abelian Higgs model.
In order to accomplish this task we shall encounter more difficulties than for one-dimensional multi-component kinks. As noticed by Vassilevich, the lack of analytical expressions for vortex solutions forces us to perform a numerical analysis already at the classical level to solve the field equations. Also, the high-temperature expansion of the heat trace becomes more involved due to the jump from one to two spatial dimensions; the recurrence relations hold between partial -rather than ordinary- derivatives of the high-T expansion coefficients. We stress that the evaluation of the Seeley coefficients is a very laborious task: fluctuations of the vector, Higgs and Goldstone fields are governed by one $4\times 4$-matrix differential operator, whereas fluctuations of the ghosts are determined by one scalar differential operator acting on $L^2({\Bbb R}^2)$. There is, however, one point where the situation is more favorable as compared to the kink case: the generalized zeta function regularization method provides us directly with a finite quantity, without the need of infinite renormalizations. This fact is peculiar to even spatial dimensions and is probably related to the lack of anomalies when fermions are added. As for kinks, we shall obtain a simple formula for the one-loop quantum mass correction depending on the Seeley coefficients and the number of zero modes.
One remarkable aspect of our results is that the correction found by this method in the bosonic system is essentially twice the correction arising in the supersymmetric case in [@Vass] and [@Reb]. This seems to be in agreement with the relationship between the supersymmetric and non-supersymmetric one-loop corrections to the masses of the sine-Gordon and $\phi^4$ kinks, see [@Schf] and [@Wimm].
The organization of the paper is as follows: In Section §.2 we revise the perturbative sector of the Abelian Higgs model in the Feynman-’t Hooft renormalizable gauge and set the one-loop mass renormalization conventions. Section §.3 is devoted to studying ANO vortex solutions and their fluctuations in a partially analytical, partially numerical manner. The high-temperature expansion of the pertinent heat traces is developed in Section §.4 . Section §.5 explains how quantum oscillations of vortices are accounted for in the framework of generalized zeta function regularization. In Section §.6 the one-loop vortex mass shift formula is applied to cylindrically symmetric self-dual vortices. We also briefly discuss how the shift depends on the distance between centers of a two-vortex solution. Finally, we offer a Summary and Outlook.
The planar Abelian Higgs model
==============================
The model
---------
The AHM describes the minimal coupling between an $U(1)$-gauge field and a scalar field in a phase where the gauge symmetry is spontaneously broken. Defining non-dimensional space-time variables, $x^\mu \rightarrow \frac{1}{ev} x^\mu$, and fields, $\phi \rightarrow v\phi=v(\phi_1+i\phi_2)$, $A_\mu \rightarrow v
A_\mu$, from the vacuum expectation value of the Higgs field $v$ and the $U(1)$-gauge coupling constant $e$, the action for the Abelian Higgs model in (2+1)-dimensions reads: $$S= \frac{v}{e}\int d^3 x \left[ -\frac{1}{4} F_{\mu \nu} F^{\mu
\nu}+\frac{1}{2} (D_\mu \phi)^* D^\mu \phi - U(\phi,\phi^*)
\right]$$ with $$U(\phi,\phi^*)=\frac{\kappa^2}{8} (\phi^* \phi-1)^2
\hspace{0.5cm} .$$ $\kappa^2=\frac{\lambda}{e^2}$ is the only classically relevant parameter and measures the ratio between the square of the masses of the Higgs, $M^2=\lambda v^2$, and vector particles, $m^2=e^2v^2$; $\lambda$ is the Higgs field self-coupling. We choose a system of units where $c=1$, but $\hbar$ has dimensions of length $\times$ mass. Also, we define the metric tensor as: $g_{\mu\nu}={\rm diag}(1,-1,-1), \, \mu,\nu=0,1,2$.
Feynman rules in the R-gauge
----------------------------
The choice of $\phi^V=1$ as the ground state causes spontaneous symmetry breaking of the Abelian gauge invariance. In the Feynman-’t Hooft renormalizable gauge, $$R(A_\mu , G)=\partial_\mu A^\mu - G \qquad \qquad ,$$ the particle spectrum involves a vector particle $A_\mu$, Higgs and Goldstone scalar particles $\phi= 1+H+iG$, and a complex ghost $\chi$. The Feynman rules are read from the action, see Reference [@Velt]: $$\begin{aligned}
S&=&{v\over e}\int \, d^3x \, \left[ -\frac{1}{2} A_\mu
[-g^{\mu\nu}(\partial_\alpha\partial^\alpha +1)]A_\nu \right.\\
&+&\frac{1}{2}\partial_\mu G\partial^\mu G-\frac{1}{2} G^2+
\frac{1}{2}\partial_\mu H\partial^\mu H-\frac{\kappa^2}{2} H^2\\
&+&\partial_\mu\chi^*\partial^\mu \chi- \chi^*\chi -{\kappa^2\over
2}H (H^2+G^2)\\&+& A_\mu (\partial^\mu H
G-\partial^\mu G H)+H (A_\mu A^\mu -\chi^* \chi)\\
&-&\left. \frac{\kappa^2}{8} (H^2+G^2)^2 +\frac{1}{2}(G^2+H^2)
A_\mu A^\mu \right] \qquad .\end{aligned}$$ There are four propagators, plus five third-order and five fourth-order vertices shown in the next two Tables:
[lccc]{}\
*Particle* & *Field* & *Propagator* & *Diagram*\
\
Higgs & $H(x)$ & $\displaystyle\frac{i e \hbar
}{v(k^2-\kappa^2+i\varepsilon)}$ &\
Goldstone & $G(x)$ & $\displaystyle\frac{ie
\hbar}{v(k^2-1+i\varepsilon)}$ &\
Ghost & $\chi$ & $\displaystyle\frac{ie
\hbar}{v(k^2-1+i\varepsilon)}$ &\
Vector Boson & $A_\mu(x)$ & $\frac{-ie \hbar g^{\mu\nu}
}{v(k^2-1+i\varepsilon)}$ &\
[clcl]{}\
*Vertex* & *Weight* & *Vertex* & *Weight*\
\
& $\displaystyle -3i\kappa^2\frac{v}{\hbar e} $ & & $\displaystyle -3i\kappa^2\frac{v}{\hbar e}$\
& $\displaystyle -i\kappa^2\frac{v}{\hbar e}$ & & $\displaystyle -3i\kappa^2\frac{v}{\hbar e}$\
& $\displaystyle 2i\frac{v}{\hbar e} g^{\mu \nu}$ & & $\displaystyle -i\kappa^2\frac{v}{\hbar e}$\
& $\displaystyle -i\frac{v}{\hbar e}$ & & $\displaystyle 2 i\frac{v}{\hbar e}g^{\mu \nu}$\
& $\displaystyle (k^\mu-q^\mu)\frac{v}{\hbar e}$ & & $\displaystyle 2 i \frac{v}{\hbar e} g^{\mu \nu}$\
One-loop renormalization
------------------------
Defining $$I(c^2)=\int \, \frac{d^3k}{(2\pi)^3} \cdot
\frac{i}{k^2-c^2+i\varepsilon}$$ and bearing in mind that $I(\kappa^2)=I(1)+\mbox{finite part}$, the one-loop divergences in the planar Abelian Higgs model can be organized as follows:
- Higgs tadpole
$+$ $+$ $+$ $=$
$$= - 2 i ( \kappa^2 + 1) I(1) + \mbox{finite part}$$
- Higgs propagator
$+$ $+$ $+$ $=$
$$= - 2 i ( \kappa^2+ 1) I(1) + \mbox{finite part}$$
- Goldstone propagator
$+$ $+$ $+$ $=$
$$= - 2 i ( \kappa^2+ 1) I(1) + \mbox{finite part}$$
- Vector boson propagator
$+$ $+$ $=$
$$= 2 i I(1) + \mbox{finite part}$$
There are no more one-loop divergent graphs. Therefore, in a minimal subtraction scheme, we add the diagrams shown in the next Table to cancel the divergences in the one-loop graphs.
*Diagram* *Weight*
------------ -- -------------------------------------
$\displaystyle 2i(\kappa^2+1)I(1) $
$\displaystyle 2i(\kappa^2+1) I(1)$
$\displaystyle 2i(\kappa^2+1) I(1)$
$\displaystyle -2i I(1)$
\[0.5cm\]
\[0.25cm\]
: One-loop counter-terms
This is tantamount to considering that the counter-terms $$\begin{aligned}
{\cal L}_{c.t.}^S &=& \frac{\hbar(\kappa^2+1)}{2} I(1)
\left[|\phi|^2-1 \right] \label{eq:ct1}\\
{\cal L}_{c.t.}^A &=& -{\hbar \over 2}I(1) A_\mu A^\mu
\label{eq:ct2}\end{aligned}$$ enter into the Lagrangian.
All the finite parts are proportional to $I(\kappa^2)-I(1)$ and they vanish in the critical point between Type I and Type II superconductivity, $\kappa^2=1$, to be considered in the sequel. Note that the mass of the elementary particles for this critical value of $\kappa$ is taken as subtraction point so that the counter-terms exactly cancel the divergence due to the Higgs tadpole. Therefore, our renormalization criterion is equivalent to the renormalization condition stipulated in [@Vass] and [@Reb] when $\kappa^2=1$.
ANO self-dual vortices
======================
Abrikosov-Nielsen-Olesen vortices are topological defects satisfying the time-independent field equations: $$\partial_i F_{ij} = J_j \hspace{0.3cm};\hspace{0.2cm}
\frac{1}{2} D_i D_i \phi= \frac{\partial U}{\partial \phi^*}
\qquad , \label{eq:soe}$$ where $J_j= \frac{i}{2} \left( \phi^* D_j \phi - (D_j \phi)^* \,
\phi \right)$ is the electric current. They are static and localized solutions for which the energy $$E= \int d^2 x [\frac{1}{4} F_{ij} F_{ij} + \frac{1}{2} (D_i
\phi)^* D_i \phi + \frac{\kappa}{8} (\phi^* \phi-1)^2 ]
\label{eq:he}$$ is finite. Thus, ANO vortices comply with the boundary conditions on $S_\infty^1$, i.e. when $r=\sqrt{x_1^2+x_2^2}$ tends to $\infty$: $$\left. \phi^{*} \phi \right|_{S_\infty^1} =1 \hspace{0.2cm} ,
\hspace{0.2cm} D_i \phi |_{S_\infty^1} =(\partial_i \phi - i A_i
\phi)|_{S_\infty^1}=0 \hspace{0.2cm} , \label{eq:bc}$$ i.e., $\phi |_{S_\infty^1} = e^{il \theta}$, $l\in{\mathbb Z}$, and $A_i |_{S_\infty^1} =- i \phi^*\partial_i \phi|_{S_\infty^1}$.
First-order equations
---------------------
For the value of the coupling constant $\kappa^2=1$, the energy functional can be arranged as follows $$E= \int \frac{d^2 x}{2} \left( |D_1 \phi \pm i D_2 \phi|^2 + [
F_{12} \pm {\textstyle\frac{1}{2}} (\phi^* \phi-1) ]^2
\right)+\frac{1}{2}|g|$$ where $g= \int d^2 x F_{12}=2{\pi l}$ is the non-dimensional quantized magnetic flux. Solutions satisfying the first-order differential equations $$D_1 \phi \pm i D_2 \phi=0 \hspace{0.3cm};\hspace{0.3cm} F_{12} \pm
\frac{1}{2} (\phi^*\phi-1) =0$$ or, equivalently, $$\begin{aligned}
&&(\partial_1 \phi_1 +A_1 \phi_2)\mp(\partial_2 \phi_2-A_2
\phi_1)=0 \label{eq:foe1} \\
&&\pm(\partial_2 \phi_1 +A_2 \phi_2)+(\partial_1 \phi_2-A_1
\phi_1)=0 \label{eq:foe2}\\
&&F_{12}\pm\frac{1}{2}(\phi_1^2+\phi_2^2-1)=0
\label{eq:foe3}\qquad .\end{aligned}$$ also solve the second-order equations (\[eq:soe\]) and are called ANO self-dual vortices if they also satisfy the boundary conditions (\[eq:bc\]). In what follows, we shall focus on solutions with positive $l$: i.e., we shall choose the upper signs in the first-order equations.
Self-dual vortices with cylindrical symmetry
--------------------------------------------
If $\theta={\rm arctan}\frac{x_2}{x_1}$ is the polar angle, the ansatz $$\begin{aligned}
\phi_1(x_1,x_2) = f(r) {\rm cos}l\theta \quad &,& \quad
\phi_2(x_1,x_2) = f(r) {\rm sin}l\theta \\
A_1(x_1,x_2) =-l \frac{\alpha(r)}{r}{\rm sin}\theta \quad &,&
\quad A_1(x_1,x_2) = l \frac{\alpha(r)}{r}{\rm cos}\theta\end{aligned}$$ plugged into the first-order equations (\[eq:foe1\], \[eq:foe2\], \[eq:foe3\]) leads to: $${1\over r} {d \alpha \over d r}= \mp \frac{1}{2 l} (f^2-1) \quad ,
\quad {d f\over d r} = \pm \frac{l}{r} f(r)[1-\alpha(r)] \quad .
\label{eq:rrfo}$$ Regular solutions of (\[eq:rrfo\]) with the boundary conditions ${\displaystyle \lim_{r\rightarrow\infty}} f(r) = 1$, ${\displaystyle \lim_{r\rightarrow\infty}} \alpha(r) = 1$, zeroes of the Higgs and vector fields at the origin, $f(0) =0$, $\alpha(0)=0$, and integer magnetic flux, $$g= - \oint_{r=\infty} dx_i A_i = -l\oint_{r=\infty}{
[x_2dx_1-x_1dx_2]\over r^2}=2 \pi l \,\, ,$$ exist and can be found by a mixture of analytical and numerical methods.
Following the procedure developed in [@VeSh], we obtain numerical solutions for the vortex equations (\[eq:rrfo\]). Indeed, this approach gives the vortex solution in three different ranges of the radial coordinate. For small values of $r$, a power series is tested in the first-order differential equations (\[eq:rrfo\]), leading to a recurrence relation between the coefficients. Reference [@VeSh] also describes the asymptotic behavior of the solutions. Thus, a numerical scheme can be implemented by setting a boundary condition in a non-singular point of (\[eq:rrfo\]), which is obtained from the power series for small values of $r$. This numerical method provides us with the behavior of the vortex solutions for intermediate distances by means of an interpolating polynomial which passes through the numerical data.
The results are shown in figure 1, where the field profiles $\alpha(r)$ and $f(r)$, the magnetic field $B(r)={l\over
2r}\frac{d\alpha}{dr}$ and the energy density $$\varepsilon(r)={1\over 4}(1-f^2(r))^2+{l^2\over
r^2}(1-\alpha(r))^2 f^2(r)$$ are plotted with respect to $r$ for self-dual ANO vortices with $l=1$, $l=2$, $l=3$, and $l=4$. A three-dimensional view of the energy density in the plane is also shown in figure 2 for $l=1$, $l=2$, $l=3$, and $l=4$ self-dual vortices. Note that the $l=1$ vortex shows a different pattern as compared with flux tubes of several quanta: only in the first case is the energy density maximum at the origin (the center).
{height="2.cm"} {height="2.cm"} {height="2.cm"} {height="2.cm"}\
[Figure 1. *Plots of the field profiles $\alpha(r)$ (a) and $f(r)$ (b), the magnetic field $B(r)$ (c), and the energy density $\varepsilon(r)$ for vortices with $l=1$ (solid line), $l=2$ (broken line), $l=3$ (broken-doted line) and $l=4$ (doted line).*]{}
{height="2.6cm"} {height="2.6cm"} {height="2.6cm"} {height="2.6cm"}\
[Figure 2. *3D graphics of the energy density for $l=1$, $l=2$, $l=3$ and $l=4$ self-dual symmetric ANO vortices.*]{}
Two-vortex solutions with distinct centers
------------------------------------------
To tackle the task of building $l=2$ ANO self-dual solutions formed by two $l=1$ vortices with centers separated by a distance $d$, we follow the work [@JR] by Jacobs and Rebbi. A variational method is implemented in two stages:
In the first stage, trial functions depending only on a single variational parameter $w$ are considered:
$$\begin{aligned}
\phi_\omega (z,z^*)&=& \Phi(z,z^*) \, \left[ \omega \,
f^{(1)}(|z-d/2|)\, f^{(1)}(|z+d/2|)+ (1-\omega)\,
\frac{|z^2-(d/2)^2|}{|z^2|}\, f^{(2)}(|z|)\right] \label{eq:fun1} \\
A^\omega (z,z^*) &=& \omega \left( \frac{i}{z^*-d/2}\,
\alpha^{(1)}(|z-d/2|)+ \frac{i}{z^*+d/2}\,
\alpha^{(1)}(|z+d/2|)\right)+(1-\omega)\, \frac{2 i}{z^*} \,
\alpha^{(2)}(|z|) \label{eq:fun2} \qquad .\end{aligned}$$
Here $$z=x_1+i x_2 \qquad , \qquad
A^\omega(z,z^*)=A_1^\omega(z,z^*)+iA_2^\omega(z,z^*) \qquad ,$$ and $$\Phi=\sqrt{\frac{z^2-(d/2)^2}{z^{*2}-(d/2)^2}}$$ is essentially a phase chosen in such a way that the magnetic flux is equal to $4\pi$. $f^{(1)}$, $\alpha^{(1)}$, $f^{(2)}$ and $\alpha^{(2)}$ stand for the functions $f$ and $\alpha$ associated with self-dual solutions with cylindrical symmetry -obtained in the previous subsection- respectively with vorticity $l=1$ and $l=2$. Evoking (\[eq:fun1\]) and (\[eq:fun2\]) we expect that $\omega=0$ for the case $d=0$ and $\omega=1$ for the case $d>>1$. Plugging (\[eq:fun1\]) and (\[eq:fun2\]) into the energy functional, we obtain a expression $E(\omega )$, which is set to be minimized as a function of $\omega$.
In the second stage the trial functions are refined by adding a deformation such that two requirements are fulfilled: 1) the scalar field vanishes at the two centers. 2) the gauge-invariant quantities associated with the solution are symmetric with respect to the reflection $z\rightarrow z^*$. The invariant ansatz reads:
$$\begin{aligned}
\phi(z,z^*)&=& \phi_\omega(z,z^*)+\Phi(z,z^*)\left|z^2-(d/2)^2
\right| (\cosh |z|)^{-1} \sum_{i=0}^N \sum_{j=0}^i f_{ij}
\frac{(zz^*)^i}{2} \left[ \left( \frac{z}{z^*}\right)^j +\left(
\frac{z^*}{z}\right)^j \right] \\
A(z,z^*)&=& A^\omega (z,z^*)+ \frac{1}{\cosh |z|} \left\{ z
\sum_{i=0}^N \sum_{j=0}^i a_{ij}^{I} \frac{(zz^*)^i}{2} \left[
\left( \frac{z}{z^*}\right)^j +\left( \frac{z^*}{z}\right)^j
\right]+ z^* \sum_{i=0}^N \sum_{j=0}^i a_{ij}^{II}
\frac{(zz^*)^i}{2} \left[ \left( \frac{z}{z^*}\right)^j +\left(
\frac{z^*}{z}\right)^j \right] \right\}\end{aligned}$$
{height="3cm"} {height="3cm"} {height="3cm"}\
[Figure 3. *3D graphics of the energy density for $l=2$ self-dual separate vortices with centers at distances $d=1$, $d=2$, $d=3$.*]{}
These expressions involve $\aleph=3\frac{(N+1)(N+2)}{2}$ variational parameters $f_{ij}$, $a^I_{ij}$, $a^{II}_{ij}$. Finding the minimum of the energy functional as a function of these $\aleph$ variables - a task for Mathematica- a good approximation to the $l=2$ self-dual solution with a distance $d$ between the two $l=1$ vortex centers is obtained. For our purposes setting $N=1$ such that $\aleph=9$ will suffice. The energy density for two-vortex solutions found by this method if $\aleph=9$ is depicted for $d=1$, $d=2$ and $d=3$ in the above figure.
Small fluctuations
------------------
We generically denote the vortex solution fields as $$\phi^V=\psi=\psi_1+i\psi_2 \qquad , \qquad A_k^V=V_k \qquad ,
k=1,2 \qquad .$$ Assembling the small fluctuations around the solution $$\phi (\vec{x})=\psi (\vec{x})+\varphi (\vec{x}) \qquad , \qquad
A_k(\vec{x})=V_k(\vec{x})+a_k(\vec{x})$$ in a four column $\xi(\vec{x})$, ${\rm L}^2$-integrable second-order fluctuations around a given vortex solution are still solutions of the first-order equations with the same magnetic flux if they belong to the kernel of the Dirac-like operator, ${\cal
D}\xi (\vec{x})=0$, [@Wein]
$${\cal D}\xi(\vec{x})= \left(\begin{array}{cccc} -\partial_2 &
\partial_1 & \psi_1 & \psi_2 \\ -\partial_1 & -\partial_2 &
-\psi_2 & \psi_1
\\ \psi_1 & -\psi_2 & -\partial_2+V_1 & -\partial_1-V_2 \\ \psi_2 & \psi_1 &
\partial_1+V_2 & -\partial_2+V_1 \end{array}\right)\left(\begin{array}{c} a_1(\vec{x})\\
a_2(\vec{x})\\
\varphi_1(\vec{x})\\ \varphi_2(\vec{x}) \end{array}\right)$$
The first component of ${\cal D}\xi$ gives the deformation of the vortex equation (\[eq:foe3\]), whereas the third and fourth components are due to the respective deformation of the covariant holomorphy equations (\[eq:foe2\]) and (\[eq:foe1\]). The second component sets the background gauge $$B(a_k,\varphi;\psi)=\partial_k
a_k-(\psi_1\varphi_2-\psi_2\varphi_1)$$ on the fluctuations.
The operator ${\cal H}^+={\cal D}^\dagger{\cal D}$ and its partner ${\cal H}^-={\cal D}{\cal D}^\dagger$ read:
$$\begin{aligned}
{\cal H}^+={\small \left(\begin{array}{cccc} -\bigtriangleup
+|\psi|^2 & 0 & -2\nabla_1\psi_2 & 2\nabla_1\psi_1
\\ 0 & -\bigtriangleup+|\psi|^2 & -2\nabla_2\psi_2 &
2\nabla_2\psi_1 \\ -2\nabla_1\psi_2 & -2\nabla_2\psi_2 &
-\bigtriangleup+{1\over 2}(3|\psi|^2+2V_kV_k-1) & -2V_k\partial_k
\\ 2\nabla_1\psi_1 & 2\nabla_2\psi_1 & 2V_k\partial_k &
-\bigtriangleup+{1\over 2}(3|\psi|^2+2V_kV_k-1)
\end{array}\right)}\end{aligned}$$
$$\begin{aligned}
{\cal H}^-={\small \left(\begin{array}{cccc}
-\bigtriangleup+|\psi|^2 & 0 & 0 & 0 \\ 0 &
-\bigtriangleup+|\psi|^2 & 0 & 0 \\ 0 & 0 &
-\bigtriangleup+{1\over 2}(|\psi|^2+1)+V_kV_k & -2V_k\partial_k
\\ 0 & 0 & 2V_k\partial_k & -\bigtriangleup+{1\over
2}(|\psi|^2+1)+V_kV_k
\end{array}\right)} \qquad .\end{aligned}$$
One can check that ${\cal H}^+$ arises in the small deformation of the second-order equations (\[eq:soe\]) in the background gauge for $\kappa=1$, thus ruling the second-order fluctuations around the vortex solutions. In fact, for $l=0$ one finds that ${\cal
H}^+={\cal H}^-={\cal H}_0$, where $${\cal H}_0={\small \left(\begin{array}{cccc} -\bigtriangleup +1 &
0 & 0 & 0
\\ 0 & -\bigtriangleup+1 & 0 &
0 \\ 0 & 0 & -\bigtriangleup+1 & 0
\\ 0 & 0 & 0 &
-\bigtriangleup+1
\end{array}\right)}$$ is the second-order fluctuation operator around the vacuum in the Feynman-’t Hooft renormalizable gauge: the background gauge in the vacuum sector. Note that the fluctuations in this gauge correspond to a massive vector particle plus scalar Higgs and Goldstone fields. It will be useful in the sequel to write the second-order fluctuation operators around $l\geq 1$ vortices in the form: $${\cal H}^\pm={\cal H}_0
+Q^\pm_k(\vec{x})\partial_k+V^\pm(\vec{x})\qquad ,$$ where $Q^\pm_k(\vec{x})$ and $V^\pm(\vec{x})$ are $4\times 4$ functional matrices.
High-temperature expansion of heat traces
=========================================
Index theorem: moduli space of self-dual vortices
-------------------------------------------------
One easily checks that $\dim \ker {\cal D}^\dagger=0$ because the spectrum of ${\cal H}^-$ is definite positive. Thus, the dimension of the moduli space of self-dual vortex solutions with magnetic charge $l$ is the index of ${\cal D}$: $${\rm ind}\,{\cal D}=\dim \ker {\cal D}- \dim \ker {\cal D}^\dagger
\qquad \qquad .$$ We follow Weinberg [@Wein], using the background instead of the Coulomb gauge, to briefly determine ${\rm ind}\,{\cal D}$. The spectra of the operators ${\cal H}^+$ and ${\cal H}^-$ only differ in the number of eigen-functions belonging to their kernels. For topological vortices, we do not expect pathologies due to asymmetries between the spectral densities of ${\cal H}^+$ and ${\cal H}^-$, and thus ${\rm ind}\, {\cal D}={\rm Tr}\,
e^{-\beta{\cal H}^+}-{\rm Tr}\, e^{-\beta{\cal H}^-}$. For a case in which these asymmetries are important, see the treatment of Chern-Simons-Higgs topological and non-topological vortices given in [@Wein1; @GM99].
The heat trace of a $N\times N$ matrix differential operator $${\cal H}={\cal H}_0 +Q_k(\vec{x})\partial_k+V(\vec{x})$$ -like the ${\cal H}^\pm$ operators- is defined as $${\rm Tr}\,e^{-\beta{\cal H}}={\rm tr}\int_{{\mathbb R}^2} \,
d^2\vec{x} \, K_{{\cal H}}(\vec{x},\vec{x};\beta)$$ where $K_{{\cal H}}(\vec{x},\vec{y};\beta)$ is the $N\times N$ matrix kernel of the heat equation and ${\rm tr}$ is the usual matrix trace. Therefore, $K_{{\cal H}}(\vec{x},\vec{y};\beta)$ solves the heat equation $$\left(\frac{\partial}{\partial\beta}{\mathbb I}+{\cal H}
\right)K_{{\cal H}}(\vec{x},\vec{y};\beta )=0 \label{eq:heateq}$$ with initial condition $$K_{{\cal H}}(\vec{x},\vec{y};0)={\mathbb I}\cdot
\delta^{(2)}(\vec{x}-\vec{y}) \quad .\label{eq:inicond}$$ Because $$K_{{\cal H}_0}(\vec{x},\vec{y};\beta)={e^{-\beta}\over
4\pi\beta}\cdot{\mathbb I}\cdot e^{-\frac{|\vec{x}-\vec{y}|
}{4\beta}}$$ is the heat kernel for the Klein-Gordon operator ${\cal H}_0$, it is convenient to write the heat kernel for ${\cal H}$ in the form: $$K_{{\cal H}}(\vec{x},\vec{y};\beta)=C_{\cal
H}(\vec{x},\vec{y};\beta)K_{{\cal H}_0}(\vec{x},\vec{y};\beta)
\label{eq:tru1}$$ with $C_{\cal H}(\vec{x},\vec{x};0)={\mathbb I}$ [@Roe]. Substituting (\[eq:tru1\]) into (\[eq:heateq\]) we find that $C_{\cal H}(\vec{x},\vec{y};\beta)$ solves the transfer equations: $$\begin{aligned}
\left\{ {\partial\over\partial\beta}{\mathbb
I}+{x_k-y_k\over\beta}(\partial_k{\mathbb I}-{1\over
2}Q_k)-\bigtriangleup{\mathbb I}+ \right.&& \nonumber \\
\left. +Q_k\partial_k+V \rule[-0.1in]{0.in}{0.2in}
\right\}C_{\cal H}(\vec{x},\vec{y};\beta)=0 && \quad
.\label{eq:tran}\end{aligned}$$ The high-temperature expansion $$C_{\cal H}(\vec{x},\vec{y};\beta)=\sum_{n=0}^\infty
c_n(\vec{x},\vec{y};{\cal H})\beta^n$$ trades the PDE (\[eq:tran\]) by the recurrence relations $$\begin{aligned}
[n{\mathbb I}+(x_k-y_k)(\partial_k{\mathbb I}-{1\over
2}Q_k)]c_n(\vec{x},\vec{y};{\cal H}) =&& \nonumber
\\=[\bigtriangleup{\mathbb I}
-Q_k\partial_k-V]c_{n-1}(\vec{x},\vec{y};{\cal H}) &&
\label{eq:rec}\end{aligned}$$ among the local coefficients with $n\geq 1$, with the initial condition $c_0(\vec{x},\vec{x};{\cal H})={\mathbb I}$. Taking into account that $$\begin{aligned}
{\rm Tr}e^{-\beta{\cal H}}&=&{e^{-\beta}\over
4\pi\beta}\sum_{n=0}^\infty\sum_{a=1}^4\int \, d^2x
\,[c_n]_{aa}(\vec{x},\vec{x};{\cal H})\beta^n = \nonumber \\
&=& {e^{-\beta}\over 4\pi\beta} \sum_{n=0}^\infty\beta^n c_n({\cal
H})\qquad ,\end{aligned}$$ where we have defined the Seeley coefficients as $$c_n({\cal H})=\sum_{a=1}^4 \int \, d^2x
\,[c_n]_{aa}(\vec{x},\vec{x};{\cal H}) \qquad ,$$ and that the first local coefficient can be easily computed $$c_1(\vec{x},\vec{x};{\cal H})=-V(\vec{x}) \qquad ,$$ by applying these formulas to the ${\cal H}^\pm$ operators we obtain in the $\beta=0$ -infinite temperature- limit : $$\begin{aligned}
{\rm ind}{\cal D}&=&{1\over 4\pi}\left\{c_1({\cal H}^+)-c_1({\cal
H}^-)\right\}=\\&=&{1\over \pi}\int d^2x \left(\frac{\partial
V_2}{\partial x_1}-\frac{\partial V_1}{\partial x_2}\right)
(\vec{x})=2l,\end{aligned}$$ i.e., the dimension of the self-dual vortex moduli space is $2l$. Physically, this means that there are solutions, if $\kappa=1$, for any location of the $l$-vortex centers in the plane [@jatb]; all static configurations of self-dual $l$-vortices can thus be interpreted as states of neutral equilibrium.
Seeley coefficients
-------------------
Computation of the coefficients of the asymptotic expansion is a difficult task; to start with, the order two local coefficient reads: $$\begin{aligned}
&& c_2(\vec{x},\vec{x};{\cal H})= -{1\over 6}\bigtriangleup
V(\vec{x})+{1\over
12}Q_k(\vec{x})Q_k(\vec{x})V(\vec{x})-\\
&&- {1\over 6}\partial_kQ_k(\vec{x})V(\vec{x})+{1\over
6}Q_k(\vec{x})\partial_kV(\vec{x})+{1\over 2}V^2(\vec{x})\quad .\end{aligned}$$ Complexity increases strongly for high-order local coefficients .
The recurrence relation (\[eq:rec\]) allows us to express $c_n(\vec{x},\vec{y},{\cal H})$ and its derivatives in terms of all the $c_k(\vec{x},\vec{y};{\cal H})$ with $k\leq n$ and their derivatives. One passes from this information to the values of the Seeley coefficients $c_n({\cal H})$ in two steps. First, one must reach the subtle $\vec{y}\rightarrow \vec{x}$ limit. In this analytical manoeuvre the partial derivatives of $c_n(\vec{x},\vec{y};{\cal H})$ at $\vec{y}=\vec{x}$ $${}^{(\alpha_1,\alpha_2)}C_n^{ab}(\vec{x})=\lim_{\vec{y}\rightarrow
\vec{x}} \frac{\partial^{\alpha_1+\alpha_2}
[c_n]_{ab}(\vec{x},\vec{y};{\cal H})}{\partial
x_1^{\alpha_1}\partial x_2^{\alpha_2}}$$ play a prominent r$\hat{{\rm o}}$le. Note also that: $$[{c}_n]_{ab}(\vec{x},\vec{x};{\cal
H})={}^{(0,0)}C_n^{ab}(\vec{x})\qquad .$$ In the $\vec{y}\rightarrow\vec{x}$ limit the recurrence relation (\[eq:rec\]) becomes :
$$\begin{aligned}
(k+\alpha_1&+&\alpha_2+1)
{}^{(\alpha_1,\alpha_2)}C_{k+1}^{ab}(\vec{x})=
{}^{(\alpha_1+2,\alpha_2)}C_{k}^{ab}(\vec{x})+
{}^{(\alpha_1,\alpha_2+2)}C_{k}^{ab}(\vec{x})- \nonumber \\
&&-\sum_{d=1}^N \sum_{r=0}^{\alpha_1}\sum_{t=0}^{\alpha_2}
{\alpha_1 \choose r} {\alpha_2 \choose t} \left[
\frac{\partial^{r+t} Q^{ad}_1}{\partial x_1^r\partial x_2^t}
{}^{(\alpha_1-r+1,\alpha_2-t)}C_{k}^{db}(\vec{x})+\frac{\partial^{r+t}
Q^{ad}_2}{\partial x_1^r\partial x_2^t}
{}^{(\alpha_1-r,\alpha_2-t+1)}C_{k}^{db}(\vec{x})
\right]+ \nonumber \\
&&+\frac{1}{2}\sum_{d=1}^N
\sum_{r=0}^{\alpha_1-1}\sum_{t=0}^{\alpha_2} \alpha_1{\alpha_1-1
\choose r} {\alpha_2 \choose t} \frac{\partial^{r+t}
Q^{ad}_1}{\partial x_1^r\partial x_2^t}
{}^{(\alpha_1-1-r,\alpha_2-t)}C_{k+1}^{db}(\vec{x})+ \label{eq:recur}\\
&&+\frac{1}{2}\sum_{d=1}^N
\sum_{r=0}^{\alpha_2-1}\sum_{t=0}^{\alpha_1} \alpha_2{\alpha_2-1
\choose r} {\alpha_1 \choose t} \frac{\partial^{r+t}
Q^{ad}_2}{\partial x_1^t\partial x_2^r}
{}^{(\alpha_1-t,\alpha_2-1-r)}C_{k+1}^{db}(\vec{x})- \nonumber
\\&&-\sum_{d=1}^N
\sum_{r=0}^{\alpha_2}\sum_{t=0}^{\alpha_1}{\alpha_1 \choose
t}{\alpha_2\choose r}
\frac{\partial^{r+t}
V^{ad}}{\partial x_1^t\partial x_2^r}
{}^{(\alpha_1-t,\alpha_2-r)}C_k^{db}(\vec{x}) \nonumber \qquad .\end{aligned}$$
The initial condition $c_0(\vec{x},\vec{x};{\cal H})={\mathbb I}$ means that all the ${}^{(\beta,\gamma)}C_0^{ab}(\vec{x})$ vanish except ${}^{(0,0)}C_0^{aa}(\vec{x})=1$ for $a=1,2,\cdots,N$. Starting from these conditions one computes all the ${}^{(\beta,\gamma)}C_n^{ab}(\vec{x})$ local coefficients by using (\[eq:recur\]). For instance, in order to obtain ${}^{(0,0)}C_6^{ab}(\vec{x})$ for ${\cal H}^+$ we need ${}^{(\beta,\gamma)}C_5^{ab}(\vec{x})$ for $\beta,\gamma=0,1,2$ as data, which in turn can be calculated from ${}^{(\beta,\gamma)}C_4^{ab}(\vec{x})$ for $\beta,\gamma=0,1,2,3,4$, and so forth. Evaluation of ${}^{(0,0)}C_6^{ab}(\vec{x})$ requires knowledge of 4032 local coefficients !!!. In general, the rule is: knowledge of ${}^{(0,0)}C_n^{ab}(\vec{x})$ amounts to knowledge of $\frac{8}{3}(n+1)(n+2)(4n+3)$ ${}^{(\beta,\gamma)}C_k^{ab}(\vec{x})$ local coefficients with $k\leq n$. The second step is much simpler: simple numerical integration of $\sum_{a=1}^4{}^{(0,0)}C_n^{aa}(\vec{x})$ over the plane.
Quantum oscillations of self-dual vortices
==========================================
Standard lore in the semi-classical quantization of solitons tells us that the one-loop mass shift comes from the Casimir energy plus the contribution of the mass renormalization counter-terms: $\Delta M_V=\Delta M_V^C+\Delta M_V^R$.
Casimir energy and vortex mass renormalization counter-terms
------------------------------------------------------------
By expanding the static energy (\[eq:he\]) of the AHM around self-dual vortex solutions one obtains, up to second-order in $\xi$ in the background gauge: $$\begin{aligned}
E&+&{v^2\over 2}\int \, d^2\vec{x} \, \left[\partial_j
a_j-\psi_1\varphi_2+\psi_2\varphi_1\right]^2 \\&\simeq& \pi |l|
v^2+{1\over 2}\int \, d^2\vec{x} \, \xi^T{\cal H}^+ \xi +{\cal
O}(\xi^3) \qquad .\end{aligned}$$ Also, the ghosts -arising when the quantization procedure is performed in the background gauge- contribute negatively to the energy: $$E^{{\rm Ghost}}={v^2\over 2}\int \, d^2\vec{x} \,
\left[\chi^*\left(-\bigtriangleup+|\psi|^2\right)\chi+\psi^*\varphi\chi^*\chi\right]\quad
.$$ Thus, the vortex Casimir energy is the sum of the Casimir energies of the bosonic $a_1,a_2,\varphi_1,\varphi_2$ fluctuations around the vortex minus the Casimir energy of the fermionic fluctuation $\chi$; the ordinary -non-matrix- Schrodinger operator ruling the ghost fluctuation around the vortex is: $${\cal H}^G=-\bigtriangleup+|\psi|^2 \qquad .$$ $\varphi_2$ is a pure gauge oscillation but its contribution is killed by the negative ghost contribution. The same applies for the vacuum Casimir energy: the Goldstone boson Casimir energy is canceled by the ghost Casimir energy, the trace of the square root of ${\cal H}_0^G=-\bigtriangleup +1$. In sum, the vortex Casimir energy measured with respect to the vacuum Casimir energy is given by the formal formula: $$\begin{aligned}
\Delta M_V^C&=&{\hbar m\over 2}\left[{\rm STr}^*\left({\cal
H}^+\right)^{{1\over 2}}-{\rm STr}\left({\cal H}_0\right)^{{1\over
2}}\right] \\ {\rm STr}^*\left({\cal H}^+\right)^{{1\over
2}}&=&\hspace{0.3cm}{\rm Tr}^* \left({\cal H}^+\right)^{{1\over
2}}-{\rm Tr}\left({\cal H}^G\right)^{{1\over 2}}\\ {\rm
STr}\left({\cal H}_0\right)^{{1\over 2}}&=&\hspace{0.3cm}{\rm
Tr}\left({\cal H}_0\right)^{{1\over 2}}-{\rm Tr}\left({\cal H}^G_0
\right) \qquad .\end{aligned}$$ The star means that the $2l$ zero eigenvalues of ${\cal H}^+$ must be subtracted because zero modes only enter at two-loop order.
In the minimal subtraction renormalization scheme, finite renormalizations are adjusted in such a way that the critical point $\kappa^2=1$ is reached at first-order in the loop expansion. Therefore, (\[eq:ct1\]) and (\[eq:ct2\]) tell us that the contribution of the mass renormalization counter-terms to the vortex mass is: $$\begin{aligned}
\Delta M_V^R&=&\Delta M_{c.t.}^S+\Delta M_{c.t.}^A=\hbar\, m\,I(1)
\, \Sigma (\psi,V_k)\\
\Sigma (\psi,V_k)&=&\int \, dx^2 \, [(1-|\psi|^2)-{1\over 2}
V_kV_k]\hspace{0.2cm},\end{aligned}$$ and the divergent integral $I(1)$ can be written in the form $$I(1)={1 \over 2}\int {d^2 {\vec k}\over (2 \pi)^2} {1\over
\sqrt{\vec k \cdot \vec k +1}}$$ after applying the residue theorem to integration in the complex $k_0$-plane.
Zeta function regularization of Casimir energies and self-energy graphs
-----------------------------------------------------------------------
We regularize both infinite quantities $\Delta M_V^C$ and $\Delta
M_V^R$ by means of generalized zeta functions. From the spectral resolution of a Fredholm operator ${\cal H}$ $${\cal H}\xi_n=\lambda_n\xi_n \qquad ,$$ one defines the generalized zeta function as the series $$\zeta_{\cal H}(s)=\sum_n {1\over\lambda_n^s} \qquad ,$$ which is a meromorphic function of the complex variable $s$ [@Gilkey], [@Roe]. We can then hope that, despite their continuous spectra, our operators fit in this scheme, and write: $$\begin{aligned}
\Delta M_V^C (s)&=&\frac{\hbar\mu}{2}\left({\mu^2\over
m^2}\right)^s\left\{\left(\zeta_{{\cal H}^+}(s)-\zeta_{{\cal
H}^G}(s)\right)+ \right. \\ &&+\left. \left(\zeta_{{\cal
H}_0^G}(s)-\zeta_{{\cal
H}_0}(s)\right)\right\} \\
\Delta M_V^R(s)& = &{\hbar\over m L^2} \zeta_{{\cal H}_0} (s)
\Sigma (\psi,V_k)\end{aligned}$$ where $$\zeta_{{\cal H}_0} (s) = {m^2 L^2 \over 4 \pi} {\Gamma (s-1) \over
\Gamma(s)}$$ and $\mu$ is a parameter of inverse length dimensions. Note that $$\Delta M_V^C=\lim_{s\rightarrow -\frac{1}{2}}\Delta M_V^C(s)\quad
, \quad \Delta M_V^R=\lim_{s\rightarrow \frac{1}{2}}\Delta
M_V^R(s)$$ and $$I(1)=\lim_{s\rightarrow{1\over 2}}{1\over 2m^2L^2}\zeta_{{\cal
H}_0}(s)$$ on a square of area $L^2$.
Together with the high-temperature expansion, the Mellin transform of the heat trace $$\zeta_{{\cal H}}(s)={1\over\Gamma(s)}\int_0^\infty \, d\beta \,
\beta^{s-1} \, {\rm Tr} \, e^{-\beta {\cal H}}$$ shows that $$\zeta_{{\cal H}}(s)={1\over\Gamma(s)}\sum_{n=0}^\infty \int_0^1 \,
d\beta \, \beta^{s+n-2}c_n({\cal
H})e^{-\beta}+{1\over\Gamma(s)}B_{{\cal H}}(s)$$ is the sum of meromorphic and entire –$B_{\cal H}(s)$– functions of $s$. Neglecting the entire parts and keeping a finite number of terms, $N_0$, in the asymptotic series for $\zeta_{\cal H}(s)$, we find the following approximations for the generalized zeta functions concerning the differential operators ${\cal H}^+$ and ${\cal H}^G$ relevant to our problem: $$\begin{aligned}
\zeta_{{\cal H}^+} (s) - \zeta_{{\cal H}_0} (s) & \simeq &
\sum_{n=1}^{N_0} c_n({\cal H}^+) \cdot
{\gamma[s+n-1,1] \over 4 \pi \Gamma(s)} \\
\zeta_{{\cal H}_0^G} (s) -\zeta_{{\cal H}^G} (s)
& \simeq & - \sum_{n=1}^{N_0} c_n ({\cal H}^G) \cdot
{\gamma[s+n-1,1] \over 4 \pi \Gamma(s)} \quad ;\end{aligned}$$ $\gamma[s+n-1,1]=\int_0^1 \, d\beta \, \beta^{s+n-2}e^{-\beta} $ is the incomplete gamma function, with a very well known meromorphic structure.
Regarding one-dimensional kinks, see [@Aai1], [@Aai2], [@Aai3], the contributions of $c_0 ({\cal H}^+)$ and $c_0({\cal H}^G)$ to $\zeta_{{\cal H}^+}(s)$ and $\zeta_{{\cal
H}^G}(s)$ are respectively canceled by $\zeta_{{\cal H}_0}(s)$ and $\zeta_{{\cal H}^G_0}(s)$; i.e., renormalization of zero point vacuum energies takes care of the $c_0({\cal H}^+)$ and $c_0({\cal
H}^G)$ contributions to the vortex Casimir energy. Note, however, that, in contrast to the (1+1)-dimensional case, the value $s=-{1\over 2}$ for which we shall obtain the Casimir energy is not a pole. To compute the vortex Casimir energy one can first take the $s=-{1\over 2}$ limit and then subtract the vacuum Casimir energy regularized by this procedure; a finite answer for the kink Casimir energy is only reached if one first subtracts the vacuum Casimir energy of the one-dimensional system.
One-loop mass shift formula
---------------------------
Writing as $\bar{c}_n=c_n({\cal H}^+)-c_n ({\cal H}^G)$ the difference between the Seeley coefficients of ${\cal H}^+$ and ${\cal H}^G$ for vorticity $l$, we check that the contribution of the first coefficient to the Casimir energy $$\Delta M_V^{(1)C} (s) \simeq {\hbar \over 2} \mu \left( {\mu^2
\over m^2}\right)^s \bar{c}_1 \cdot {\gamma[s,1/2] \over 4 \pi
\Gamma(s)}$$ is finite at the $s\rightarrow -\frac{1}{2}$ limit $$\Delta M_V^{(1)C} (-1/2) \simeq - {\hbar m \over 4 \pi } \Sigma
(\psi,V_k) \cdot {\gamma[-1/2,1] \over \Gamma(1/2)}$$ and exactly cancels the contribution of the mass renormalization counter-terms –also finite for $s={1\over 2}$–: $$\begin{aligned}
\Delta M_V^R (s)& \simeq &{\hbar m \over 4 \pi} \cdot \Sigma
(\psi,V_k) \cdot {\gamma[s-1,1]\over \Gamma(s)} \\
\Delta M_V^R (1/2) & \simeq & {\hbar m \over 4 \pi} \cdot \Sigma
(\psi,V_k) \cdot
{\gamma[-1/2,1]\over \Gamma(1/2)} \, .\end{aligned}$$ Our choice of a minimal subtraction scheme not only arranges finite renormalizations in such a way that self-duality holds for $\kappa=1$ at the one-loop order, but also fits in with the criterion that the mass renormalization counter-terms must kill the contribution to the Casimir energy of the first Seeley coefficients for the heat trace expansions of the operators ${\cal
H}^+$, ${\cal H}^G$, ${\cal H}_0$, ${\cal H}_0^G$. The same cancellation happens for kinks only if the mode number cut-off regularization procedure, see [@Reb1], [@Aai1] and [@vanN], is applied.
Subtracting the contribution of the $2l$ zero modes, $$\begin{aligned}
\Delta M_{V}& =&{\hbar m\over 2}\lim_{s\rightarrow
-\frac{1}{2}}\left[-\frac{2l}{\Gamma(s)}\int_0^1 d\beta
\beta^{s-1}+ \right. \\ && \left. + \sum_{n=2}^{N_0} \bar{c}_n
{\gamma[s+n-1,1] \over 4 \pi \Gamma(s)}\right]\end{aligned}$$ we finally obtain the following formula for the vortex mass shift: $$\Delta M_V= -{\hbar m \over 2} \left[ \frac{1}{8\pi\sqrt{\pi}}
\sum_{n=2}^{N_0} \bar{c}_n \gamma[n-\frac{3}{2},1]+\frac{2
l}{\sqrt{\pi}} \right] \, . \label{eq:vorm}$$
One-loop mass shifts
====================
Local coefficients for cylindrically symmetric vortices
-------------------------------------------------------
We shall apply these formulae to cylindrically symmetric vortices. The heat kernel local coefficients, however, depend on successive derivatives of the solution. This dependence can increase the error in the estimation of these local coefficients because we handle an interpolating polynomial as the numerically generated solution, and the successive derivations with respect to $r$ of such a polynomial introduces inaccuracies. Indeed this operation is plugged into the algorithm that generates the local coefficients in order to speed up this process. It is thus of crucial importance to use the first-order differential equations (\[eq:rrfo\]) in order to eliminate the derivatives of the solution and write the local coefficients as expressions depending only on the fields. We find: $$\begin{aligned}
\frac{\partial\psi_1}{\partial x_1}&=&\frac{l
f(r)}{r}\left[\cos\theta \cos l\theta(1-\alpha(r))+\sin\theta \sin l\theta \right]\\
\frac{\partial\psi_1}{\partial x_2}&=&\frac{l
f(r)}{r}\left[\sin\theta \cos l\theta(1-\alpha(r))- \cos\theta \sin l\theta \right]\\
\frac{\partial\psi_2}{\partial x_1}&=&\frac{l
f(r)}{r}\left[\cos\theta \sin l\theta(1-\alpha(r))- \sin\theta \cos l\theta \right]\\
\frac{\partial\psi_2}{\partial x_2}&=&\frac{l
f(r)}{r}\left[\sin\theta \sin l\theta(1-\alpha(r))+ \cos\theta
\cos l\theta \right] \\ \frac{\partial V_1}{\partial
x_1}&=&\sin\theta \cos
\theta\left[\frac{2l f(r)\alpha(r)}{r}+{1\over 2}(f^2(r)-1)\right]\\
\frac{\partial V_1}{\partial x_2}&=&-l \cos
2\theta\frac{\alpha(r)}{r^2}+{1\over 2}\sin^2\theta(f^2(r)-1)\\
\frac{\partial V_2}{\partial x_1}&=&-l \cos
2\theta\frac{\alpha(r)}{r^2}-{1\over 2} \cos^2\theta(f^2(r)-1)\\
\frac{\partial V_2}{\partial x_2}&=&-\sin\theta \cos
\theta\left[\frac{2l f(r)\alpha(r)}{r}+{1\over 2}(f^2(r)-1)\right]\end{aligned}$$ for self-dual ANO vortices with generic (positive) vorticity $l$.
The recurrence formula now gives the local coefficients of the asymptotic expansion in terms of $f(r)$ and $\alpha(r)$, e.g.,
[$$\begin{aligned}
{\rm tr}
[c_1](\vec{x},\vec{x};{\cal H^+})&=& 5[1-f^2(r)]-\frac{2}{r^2} l^2
\alpha^2
(r) \\
{\rm tr}[c_2](\vec{x},\vec{x};{\cal H^+})&=&
\frac{1}{12r^4}\left\{ 37 r^4 +4 l^4 \alpha^4
(r)+8(7l^2r^2-8r^4)f^2(r)+27r^4 f^4(r)- \right.
\\ && \left. -8lr^2 \alpha(r)[-1+(1+13
l)f^2(r)]+8l^2\alpha^2(r)(-2-3r^2+9r^2f^2(r)) \right\} \\
{\rm tr}[c_3](\vec{x},\vec{x};{\cal H^+})&=& \frac{1}{120 r^6}
\left\{ -4l^6 \alpha^6(r)-4 l^3 r^2 \alpha^3(r) [14 +(-132+167
l)f^2(r)]+4l^4 \alpha^4(r)(20+9r^2+32r^2f^2(r))-2lr^2 \alpha(r)
[-4(16+9r^2)+ \right.
\\ && + (64+96 l-472 l^2+344 l^3+88l^2+243 l r^2)
f^2(r)+(-52+109l)r^2 f^4(r)]+ l^2 \alpha^2(r) [-256-144
r^2-117r^4+ \\ && + 2r^2(88-548 l+516l^2+183r^2)f^2(r) + 99r^4
f^4(r)]+r^2[r^2(-16+151r^2)+(-320l^3+160l^4+32r^2+48lr^2-
\\ && \left.-321 r^4 +
8l^2(20+39r^2))f^2(r)+r^2(-16-48
l+44l^2+199r^2)f^4(r)-29r^4f^6(r)] \right\} \qquad .\end{aligned}$$]{}
We have explicitly given only the first three local coefficients of the heat kernel expansion for ${\cal H}^+$ because the complexity of the expressions increases with $n$ enormously. Additionally, [$$\begin{aligned}
&& c_1(\vec{x},\vec{x};{\cal H}^G)= 1-f^2(r)\\
c_2(\vec{x},\vec{x};{\cal H}^G)&=&\frac{-1}{6r^2}\left\{ [4 l^2+ 5
r^2-8 l^2
\alpha(r) +4 l^2 \alpha^2(r)]f^2(r)+ \right. \\
&& \left. \hspace{0.6cm} -3 r^2-2r^2f^4(r) \right\} \\
c_3(\vec{x},\vec{x};{\cal H}^G)&=& \frac{1}{60 r^4} \left\{ 10
r^4-[-32 l^3+16 l^4+8lr^2+23 r^4+ \right. \\ + 16 l^2 (1+r^2)&-&
8l(-12l^2+8l^3+r^2+4l(1+r^2))\alpha(r)+ \\ + 16
l^2(1-6l+6l^2&+&r^2)\alpha^2(r)+ 32(1-2l)l^3\alpha^3(r)+ \\ +
16l^4\alpha^4(r)]f^2(r)&+& r^2[8l+16l^2+17r^2+16l^2\alpha^2(r)-
\\ &&-8l(1+4l)\alpha(r)\left. ] f^4(r)-4r^4f^6(r)
\right\}\end{aligned}$$]{}
are the first three local coefficients for the heat kernel expansion for the ghost operator ${\cal H}^G$.
Plugging these expressions into the partially analytical partially numerical solution for $f(r)$ and $\alpha(r)$, it is possible to compute the local coefficients and integrate them numerically over the whole plane.
Mass shift for vorticities $l=1$, $l=2$, $l=3$, $l=4$
-----------------------------------------------------
Finally, the one-loop quantum correction of the vortex solution with vorticity $l$ is given by formula (\[eq:vorm\]) $$\Delta M_V= -{\hbar m \over 2} \left[ \frac{1}{8\pi\sqrt{\pi}}
\sum_{n=2}^{N_0} \bar{c}_n \gamma[n-\frac{3}{2},1]+\frac{2
l}{\sqrt{\pi}} \right]$$ Using the Mathematica environment in a modest PC we have obtained the coefficients shown in Tables IV and V,
----- ---------------------- ---------------------- --------------- ----------------------
$n$ $ {c}_n({\cal H}^+)$ $ {c}_n({\cal H}^G)$ $ {c}_n({\cal $ {c}_n({\cal H}^G)$
H}^+)$
2 30.36316 2.60773 61.06679 6.81760
3 12.94926 0.31851 25.61572 1.34209
4 4.22814 0.022887 8.21053 0.20481
5 1.05116 0.0011928 2.02107 0.023714
6 0.20094 0.00008803 0.40233 0.002212
----- ---------------------- ---------------------- --------------- ----------------------
: Seeley Coefficients for $l=1,2$.[]{data-label="table:tabla1"}
----- ---------------------- ---------------------- -------------- ----------------------
$n$ $ {c}_n({\cal H}^+)$ $ {c}_n({\cal H}^G)$ ${c}_n({\cal $ {c}_n({\cal H}^G)$
H}^+)$
2 90.20440 11.51035 118.67540 16.46895
3 36.68235 2.60898 46.01141 4.00762
4 11.69979 0.46721 14.64761 0.77193
5 2.86756 0.067279 3.58906 0.11747
6 0.566227 0.0079269 0.667202 0.01620
----- ---------------------- ---------------------- -------------- ----------------------
: Seeley Coefficients for $l=3,4$.[]{data-label="table:tabla1"}
We remark that formula (\[eq:vorm\]) depends on the number $N_0$ chosen to cut the asymptotic heat kernel expansions. We have no means of determining the optimum value for $N_0$, but in practice we can only cope with a small $N_0$ value; a big $N_0$ would require the computation of an enormous number of local coefficients. Nevertheless, the choice $N_0=6$ is acceptable. The behavior of the asymptotic series in (\[eq:vorm\]) is given in Table VI:
------- -------------------- -------------------- ------------- --------------------
$N_0$ $\Delta M_V (N_0)$ $\Delta M_V (N_0)$ $\Delta M_V $\Delta M_V (N_0)$
(N_0)$
$l=1$ $l=2$ $l=3$ $l=4$
2 -1.02951 -2.03787 -3.01187 -3.97025
3 -1.08323 -2.14111 -3.15680 -4.14891
4 -1.09270 -2.15913 -3.18208 -4.18014
5 -1.09427 -2.16212 -3.18628 -4.18534
6 -1.09449 -2.16257 -3.18690 -4.18606
------- -------------------- -------------------- ------------- --------------------
: Convergence of the asymptotic series in units of $\hbar m$.[]{data-label="table:tabla1"}
The convergence up to the sixth order in the asymptotic expansion is very good. In the case of $\lambda (\phi)^4_2$ kinks we found agreement between the result obtained by this method and the exact result up to the fourth decimal figure, see [@Aai1], by choosing $N_0=10$.
There are reasons to expect this behavior on general analytical grounds. Truncation of the asymptotic expansion of the heat function at order $N_0$ produces an error of order $\beta^{N_0}$, which in turn leads to an error proportional to $\gamma[N_0-{1\over 2},1]\simeq {1\over N_0-{1\over 2}}$, for $N_0$ large, in the computation of the $\zeta_{{\cal
H}^+}(-{1\over 2})$ zeta function, see [@Gilkey] Section 1.10. In fact, the rate of convergence is improved in our problem by the the smallness of the $c_n$ coefficients, see Tables IV and V, for large n. This smallness is due to the fact that, when $n$ increases, higher and higher powers of partial derivatives of the field profiles of increasing order enter in the computation of $c_n$. The vortex solutions, however, are as regular and smooth as allowed by the topology. Therefore, the admitted error by cutting the mass shift formula at $N_0=6$ is especially small for low vorticities.
In Table VII we give the one-loop quantum corrections for the vortex solutions up to $l=4$, whereas we plot the correction in the figure as a function of the magnetic flux. The broken line (linear function) represents the hypothetical situation in which each magnetic flux quantum would contribute with the same correction. Hence, this is almost - within the error margin- the situation that we have found.
$l$ $\Delta M_V/\hbar m$
----- ----------------------
1 -1.09449
2 -2.16257
3 -3.18690
4 -4.18606
: One-Loop Quantum Mass Correction to the vortex with vorticity $l=1,2,3,4$.
--------------------------------------
{height="2.2cm"}
--------------------------------------
: One-Loop Quantum Mass Correction to the vortex with vorticity $l=1,2,3,4$.
These results, however, do not allow us to answer the question of whether or not the classical degeneracy with respect to the vortex centers observed at the classical level also holds at one-loop order. The figure in Table VII seems to suggest that the mass shift of $l$ well separated vortices is equal -modulo errors- to $l$ times the mass shift of a single vortex, but we do not know in what direction the errors run.
Mass shift for solutions with two separate vortices
---------------------------------------------------
We now offer two Tables, VIII and IX, where Seeley coefficients and the quantum corrections are given for two-vortex solutions with intermediate separations $d=1$, $d=2$, and $d=3$ between superimposed vortices, $\omega=0$ in (\[eq:fun1\])-(\[eq:fun2\]), and well separated vortices, $\omega=1$ in (\[eq:fun1\])-(\[eq:fun2\]). The coefficients of the asymptotic expansion are computed only up to third order because much more computation time is required. Also, we stress that in this situation, with no cylindrical symmetry, we expect not so good results because there are two more important sources of errors: first, the variational solutions with two separate vortices are far less exact than the solution with $l=2$ and cylindrical symmetry. Second, even though another numerical method would be used in the search of vortex solution we would run in difficulties; there is no way to avoid the use of partial derivatives in the calculation of the coefficients because the vortex equations alone are not enough.
----- ---------------------- ---------------------- --------------- ---------------------- ---------------------- --------------------
$n$ $ {c}_n({\cal H}^+)$ $ {c}_n({\cal H}^G)$ $ {c}_n({\cal $ {c}_n({\cal H}^G)$ $ {c}_n({\cal H}^+)$ $
H}^+)$ {c}_n({\cal H}^G)$
2 61.0518 6.81277 58.3359 6.46609 57.3420 6.03872
3 25.6137 1.33822 24.5050 1.23466 24.1187 1.02031
----- ---------------------- ---------------------- --------------- ---------------------- ---------------------- --------------------
: Seeley Coefficients for $d=1,2,3$.[]{data-label="table:tabla11"}
------- ---------------------------- ---------------------------- ----------------
$N_0$ $\Delta M_V (N_0)/\hbar m$ $\Delta M_V (N_0)/\hbar m$ $\Delta M_V
(N_0)/\hbar m$
$d=1$ $d=2$ $d=3$
2 -2.03770 -1.99798 -1.98848
3 -2.14095 -2.09695 -2.08672
------- ---------------------------- ---------------------------- ----------------
: Convergence of the asymptotic series.[]{data-label="table:tabla12"}
Summary and outlook
===================
The one-loop mass shifts of superimposed vortices with low magnetic fluxes are: $$\begin{aligned}
M_V^{l=1}&=&m\left(\frac{\pi v}{e}-1.09427
\hbar\right)+o(\hbar^2)\\
M_V^{l=2}&=&2 m\left(\frac{\pi v}{e}-1.08106
\hbar\right)+o(\hbar^2)\\
M_V^{l=3}&=&3 m\left(\frac{\pi v}{e}-1.06230
\hbar\right)+o(\hbar^2) \\
M_V^{l=4}&=& 4 m\left(\frac{\pi
v}{e}-1.04651\hbar\right)+o(\hbar^2).\end{aligned}$$ Much less precise results are also provided for two-vortices with separate centers. This is to be compared with the supersymmetric result: $${\cal M}_V^{l}= |l| m \left(\frac{\pi v}{e}-0.5000
\hbar\right)+o(\hbar^2) \qquad ,$$ see [@Vass] and [@Reb]. We notice that the one-loop correction due to bosonic fluctuations of self-dual vortices is almost twice the correction arising in the supersymmetric system coming only from mass renormalization counterterms when proper SUSY-preserving boundary conditions are imposed. The same proportion holds between one-loop corrections to sine-Gordon and $\phi^4$ kink masses in the non-supersymmetric and supersymmetric frameworks, see [@Schf] and [@Wimm].
It seems plausible that a similar method can successfully be applied to compute the one-loop mass shift for self-dual Chern-Simons-Higgs vortices, see [@Wein1]-[@GM99]. A Hamiltonian formalism in the topological sectors of the first-order CSH Lagrangian system should be first developed. More ambitious, generalized zeta functions of $12\times 12$ matrix PDE operators in three variables are essential in computing the one-loop mass shift to BPS monopoles. Thus, our procedure opens a door to calculate quantum corrections to BPS monopole masses in a ${\cal N}=0$ bosonic setting to be contrasted with the ${\cal
N}=2$ and ${\cal N}=4$ supersymmetric results of [@Reb2] and [@Reb3].
Acknowledgements {#acknowledgements .unnumbered}
================
Partially financed by the Spanish Ministerio de Educacion y Ciencia under grant: BFM2003-00936
[99]{} A. Alonso Izquierdo, W. Garcia Fuertes, J. Mateos Guilarte and M. de la Torre Mayado, Phys. Rev. [**D70**]{} (2004) 061702(R).
J.R. Schmidt, Phys. Rev. [**D46**]{} (1992) 1839.
B.H. Lee and H. Min, Phys. Rev. [**D51**]{} (1995) 4458.
D.V. Vassilevich, Phys. Rev. [**D68**]{} (2003) 045005.
A. Rebhan, P. van Nieuwenhuizen, and R. Wimmer, Nucl. Phys. [**B679**]{} (2004) 382.
H. Nastase, M. Stephanov, A. Rebhan and P. van Nieuwenhuizen, Nucl. Phys. [**B542**]{} (1999) 471.
M. Shifman, A. Vainshtein and M. Voloshin, Phys. Rev. [**D59**]{}(1999) 045016.
A. Rebhan, P. van Nieuwenhuizen, and R. Wimmer, Phys. Lett. [**B594**]{} (2004) 234.
A. Alonso Izquierdo, W. Garcia Fuertes, M. A. Gonzalez Leon and J. Mateos Guilarte, Nucl. Phys. [**B635**]{} (2002) 525.
M. Bordag, A. Goldhaber, P. van Nieuwenhuizen and D. Vassilevich, Phys. Rev. [**D66**]{} (2002) 125014.
R. Dashen, B. Hasslacher and A. Neveu, Phys. Rev. [**D10**]{} (1974) 4130.
A. Alonso Izquierdo, W. Garcia Fuertes, M. A. Gonzalez Leon and J. Mateos Guilarte, Nucl. Phys. [**B638**]{} (2002) 378.
A. Alonso Izquierdo, W. Garcia Fuertes, M. A. Gonzalez Leon and J. Mateos Guilarte, Nucl. Phys. [**B681**]{} (2004) 163.
J.F. Schonfeld, Nucl. Phys [**B161**]{} (1979) 125.
R. Wimmer, [*Quantization of supersymmetric solitons*]{}, hep-th/0109119.
M. Veltman, [*Reflections on the Higgs system*]{}, CERN Yellow Report 97-05, 1997
H. J. de Vega and F. A. Schaposnik, Phys. Rev. [**D14**]{} (1976) 1100.
L. Jacobs and C. Rebbi, Phys. Rev. [**B19**]{} (1979) 4486
E. Weinberg, Phys. Rev. [**D19**]{} (1979) 3008.
R. Jackiw, K. Lee and E.J. Weinberg, Phys. Rev. D42 (1990) 3488.
W. Garcia Fuertes and J. Mateos Guilarte, Eur. Phys. J. [**C9**]{} (1999) 167.
P. Gilkey, [*Invariance theory, the heat equation and the Atiyah-Singer index theorem*]{}, Publish or Perish, Delaware, (1984).
J. Roe, [*Elliptic operators, topology and asymptotic methods*]{} Longman Scientific and Technical, New York (1988).
A. Jaffe and C.H. Taubes, [*Vortices and monopoles*]{}, Birkhauser, (1980).
A. Rebhan and P. van Nieuwenhuizen, Nucl. Phys. [**B508**]{} (1997) 449.
A. Rebhan, R. Schoefbeck, P. van Nieuwenhuizen, R. Wimmer, [*BPS saturation of the ${\cal N}=4$ monopole by infinite composite-operator renormalization*]{}, hep-th/0502221
[^1]: In [@Vass] there is the following footnote: Since no analytic form for the profile functions of the ANO vortices is available, calculations of the mass shift in a non-supersymmetric case is a rather complicated problem".
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- Jan Hermann
- Zeno Schätzle
- Frank Noé
bibliography:
- 'preprint.bib'
title: Deep neural network solution of the electronic Schrödinger equation
---
[^1]
Acknowledgments {#acknowledgments .unnumbered}
---------------
Funding is acknowledged from the European Commission (ERC CoG 772230 “Scale-Cell”), Deutsche Forschungsgemeinschaft (CRC1114/A04, GRK2433 DAEDALUS/P04), the MATH$^{+}$ Berlin Mathematics research center (AA1x6, EF1x2). J. H. would like to thank K.-R. Müller for support and acknowledge funding from TU Berlin (Project No. 10032745).
[^1]: $^*$Emails: [email protected], [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
We show that for every Tychonoff space $X$ and Hausdorff operation $\mathbf\Phi$, the class $\mathbf\Phi(\mathscr F\cap\mathscr G_\delta,X)$ generated from closed $\mathscr G_\delta$-sets in $X$ by $\mathbf\Phi$ has the reduction or separation property if the corresponding class $\mathbf\Phi(\mathscr G_\delta,\mathbb R)$ of sets of reals has the same property.
In particular, under Projective Determinacy, these properties of such projective sets in $X$ have the same pattern as the First Periodicity Theorem states for projective sets of reals: the classes $\mathbf\Sigma^{1}_{2n}(\mathscr F\cap\mathscr G_\delta,X)$ and $\mathbf\Pi^{1}_{2n+1}(\mathscr F\cap\mathscr G_\delta,X)$ have the reduction property while $\mathbf\Pi^{1}_{2n}(\mathscr F\cap\mathscr G_\delta,X)$ and $\mathbf\Sigma^{1}_{2n+1}(\mathscr F\cap\mathscr G_\delta,X)$ have the separation property.
author:
- 'Denis I. Saveliev'
title: On reduction and separation of projective sets in Tychonoff spaces
---
[^1]
[^2]
[^3]
In the sequel, $\mathscr F$ denotes the class of closed sets, $\mathscr G$ of open sets, $\mathscr K$ of compact sets, $\mathscr Z$ of zero-sets, i.e., pre-images of the point $0$ of the closed segment $[0,1]$ of the real line under continuous maps; finally, $\mathscr S$ denotes an unspecified class of subsets. These classes are treated as operators applied to a given topological space $X$ so $\mathscr F(X)$ consists of all closed sets in $X$, etc. For arbitrary class $\mathscr S$ we let $\mathscr S(X)=\mathscr S\cap\mathscr P(X)$. Let also $\mathscr S(Y){\upharpoonright}X=\{S\cap X:S\in\mathscr S(Y)\}$. As well-known, $\mathscr Z\subseteq\mathscr F\cap\mathscr G_\delta$, moreover, $\mathscr Z=\mathscr F\cap\mathscr G_\delta$ for normal spaces (see, e.g., [@Engelking], 1.5.11; this fails for Tychonoff spaces), and $\mathscr K\subseteq\mathscr F$ for Hausdorff spaces.
Here we consider Hausdorff operations only of countable arity. Such a [*Hausdorff operation*]{} (or $\delta s$-*operation*) $\mathbf{\Phi}$ applied to a family $(A_n)_{n<\omega}$ of sets $A_n$ has the form $${\mathbf{\Phi}}(A_n)_{n<\omega}=
\bigcup_{f\in S}\bigcap_{n<\omega}A_{f{\upharpoonright}n}$$ for some $S\subseteq\omega^\omega$, called the *base* of ${\mathbf{\Phi}}$, where sets $A_{f{\upharpoonright}n}$ are identified with the sets $A_n$ under a fixed bijection of $\omega$ onto $\omega^{<\omega}$. A ${\mathbf{\Phi}}$-[*set*]{} is a set obtained by ${\mathbf{\Phi}}$. We let ${\mathbf{\Phi}}(\mathscr S,X)$ to denote the class of ${\mathbf{\Phi}}$-sets generated by sets in $\mathscr S(X)$, i.e., $$\mathbf{\Phi}(\mathscr S,X)=
\{{\mathbf{\Phi}}(A_n)_{n<\omega}:
(A_n)_{n<\omega}\in\mathscr S(X)^\omega\}.$$ By ${\mathbf{\Phi}}(\mathscr S)$ we mean the union of ${\mathbf{\Phi}}(\mathscr S,X)$ for all $X$.
In particular, we consider the Borel and projective hierarchies generated by sets in $\mathscr S(X)$, and denote by $\mathbf\Sigma^{0}_{\alpha}(\mathscr S,X)$, $\mathbf\Pi^{0}_{\alpha}(\mathscr S,X)$, and $\mathbf\Delta^{0}_{\alpha}(\mathscr S,X)$ the $\alpha$th additive, multiplicative, and self-dual classes of the resulting Borel hierarchy, and by $\mathbf\Sigma^{1}_{n}(\mathscr S,X)$ $\mathbf\Pi^{1}_{n}(\mathscr S,X)$, and $\mathbf\Delta^{1}_{n}(\mathscr S,X)$ the $n$th additive, multiplicative, and self-dual classes of the resulting projective hierarchy. So $\mathscr G_\delta(X)$ is $\mathbf\Pi^{0}_{2}(\mathscr G,X)$ and $\mathscr F_\sigma(X)$ is $\mathbf\Sigma^{0}_{2}(\mathscr F,X)$.
We use also the following notation: $-\mathscr S(X)=\{X\setminus S:S\in\mathscr S(X)\}$ and $
\mathbf{\Delta}(\mathscr S,X)=
\mathscr S(X)\cap-\mathscr S(X).
$ Given a map $F:X\to Y$ and sets $A\subseteq X$ and $B\subseteq Y$, let $FA=\{F(x):x\in A\}$ and $F^{-1}B=\{x:\exists y\in B\;F(x)=y\}.$ Moreover, $F\mathscr S=\{FS:S\in\mathscr S\}$ and $F^{-1}\mathscr S=\{F^{-1}S:S\in\mathscr S\}$.
Sets $A,B$ are [*reduced*]{} by sets $C,D$ iff $C\cap D=\emptyset$ and $C\cup D=A\cup B$, and [*separated*]{} by a set $C$ iff $A\subseteq C$ and $B\cap C=\emptyset$ (in the latter case $A,B$ should be disjoint). Two following properties of classes of sets are the main subject of this note: $\mathscr S(X)$ has
- the [*reduction*]{} property iff every $A,B\in\mathscr S(X)$ are reduced by some $C,D\in\mathscr S(X)$, and
- the [*separation*]{} property iff every disjoint $A,B\in\mathscr S(X)$ are separated by some $C\in\mathbf{\Delta}(\mathscr S,X)$.
\[l: red in subspace\] Let $\mathscr S$ be a class and $X,Y$ some sets.
- If $\mathscr S(X)$ has reduction then $-\mathscr S(X)$ has separation, and conversely.
- If $\mathscr S(Y)$ has reduction (separation) then $\mathscr S(Y){\upharpoonright}X$ has the same property.
Clear.
\[l: hausd distr\] Let ${\mathbf{\Phi}}$ be a Hausdorff operation. Then:
- finite intersections and unions distribute over ${\mathbf{\Phi}}$,
- ${\mathbf{\Phi}}(\mathscr S(Y){\upharpoonright}X)={\mathbf{\Phi}}(\mathscr S,Y){\upharpoonright}X$ for any $\mathscr S$ and $X,Y$.
(i). If $S\subseteq\omega^\omega$ is a base of ${\mathbf{\Psi}}$, we have $$\begin{aligned}
{\mathbf{\Phi}}(B_n)_{n\in\omega}\cap X=
(\bigcup_{f\in S}\bigcap_{n\in\omega}B_{f{\upharpoonright}n})\cap X=
\bigcup_{f\in S}\bigcap_{n\in\omega}(B_{f{\upharpoonright}n}\cap X)=
{\mathbf{\Phi}}(B_n\cap X)_{n\in\omega},\end{aligned}$$ and similarly for unions.
(ii). Immediate from (i).
\[l: hausd in subspace\] For any Hausdorff operation ${\mathbf{\Phi}}$, class $\mathscr S$, and sets $X\subseteq Y$,
- if $\mathscr S(X)\subseteq\mathscr S(Y){\upharpoonright}X$ then ${\mathbf{\Phi}}(\mathscr S,X)\subseteq{\mathbf{\Phi}}(\mathscr S,Y){\upharpoonright}X$,
- if $\mathscr S(Y){\upharpoonright}X\subseteq\mathscr S(X)$ then ${\mathbf{\Phi}}(\mathscr S,Y){\upharpoonright}X\subseteq{\mathbf{\Phi}}(\mathscr S,X)$.
As ${\mathbf{\Phi}}$ is monotone, i.e., $\mathscr S\subseteq\mathscr T$ implies ${\mathbf{\Phi}}(\mathscr S)\subseteq{\mathbf{\Phi}}(\mathscr T)$, this follows from Lemma \[l: hausd distr\](ii).
\[c: hausd in subspace\] Let ${\mathbf{\Phi}}$ be a Hausdorff operation, and let $\mathscr S$ and $X\subseteq Y$ be such that $\mathscr S(X)=\mathscr S(Y){\upharpoonright}X.$ Then:
- ${\mathbf{\Phi}}(\mathscr S,X)={\mathbf{\Phi}}(\mathscr S,Y){\upharpoonright}X$,
- if ${\mathbf{\Phi}}(\mathscr S,Y)$ has reduction (separation) then ${\mathbf{\Phi}}(\mathscr S,X)$ has the same property.
(i). Lemma \[l: hausd in subspace\].
(ii). Follows from (i) and Lemma \[l: red in subspace\](ii).
The assumption of Corollary \[c: hausd in subspace\] holds, e.g., is $\mathscr S$ is any of $\mathscr F,\mathscr G,\mathscr K$.
Note that for $\mathscr S(Y)$ closed under finite intersections, if $X\in\mathscr S(Y)$, then $\mathscr S(Y){\upharpoonright}X\subseteq\mathscr S(Y)$, and so the assumption gives $\mathscr S(X)\subseteq\mathscr S(Y)$. However, in Theorem \[t: red tych\] where Corollary \[c: hausd in subspace\] will be used, Tychonoff spaces $X$ will be considered as arbitrary subspaces of $Y=[0,1]^\kappa$ without a guarantee of being a member of $\mathscr S(Y)$.
Two results below, Corollary \[c: hausd vs preimage\] and Corollary \[c: hausd vs image\], provide conditions under which classes of ${\mathbf{\Phi}}$-sets are preserved under maps in the image and pre-image direction, respectively, for arbitrary ${\mathbf{\Phi}}$.
\[l: hausd vs preimage\] Let ${\mathbf{\Phi}}$ be a Hausdorff operation. For any sets $X,Y$, map $F:X\to Y$, and $(B_n)_{n\in\omega}$ in $\mathscr P(Y)$, we have $
F^{-1}{\mathbf{\Phi}}(B_n)_{n\in\omega}=
{\mathbf{\Phi}}(F^{-1}B_n)_{n\in\omega}.
$
Let $S\subseteq\omega^\omega$ be a base of ${\mathbf{\Phi}}$. Since pre-images distribute over arbitrary unions and intersections, we have: $$\begin{aligned}
F^{-1}{\mathbf{\Phi}}(B_n)_{n\in\omega}=
F^{-1}\bigcup_{f\in S}\bigcap_{n\in\omega}B_{f{\upharpoonright}n}=
\bigcup_{f\in S}\bigcap_{n\in\omega}F^{-1}B_{f{\upharpoonright}n}=
{\mathbf{\Phi}}(F^{-1}B_n)_{n\in\omega},\end{aligned}$$ as required.
Given $\mathscr S$ and $F:X\to Y$, we say that $F$ [*preserves $\mathscr S$*]{} iff $A\in\mathscr S(X)$ implies $FA\in\mathscr S(Y)$, and $F^{-1}$ [*preserves $\mathscr S$*]{} iff $B\in\mathscr S(Y)$ implies $F^{-1}B\in\mathscr S(X)$. As usual, $F$ is [*closed*]{} iff it preserves $\mathscr F$, [*open*]{} iff it preserves $\mathscr G$, [*continuous*]{} iff $F^{-1}$ preserves $\mathscr F$ (or $\mathscr G$), and [*proper*]{} iff $F^{-1}$ preserves $\mathscr K$.
\[c: hausd vs preimage\] Let ${\mathbf{\Phi}}$ be a Hausdorff operation and $F:X\to Y$. If $F^{-1}$ preserves $\mathscr S$, then $F^{-1}$ preserves ${\mathbf{\Phi}}(\mathscr S)$, i.e., $F^{-1}{\mathbf{\Phi}}(\mathscr S,Y)\subseteq{\mathbf{\Phi}}(\mathscr S,X)$.
Lemma \[l: hausd vs preimage\].
E.g., if $F$ is continuous then $F^{-1}$ preserves each of ${\mathbf{\Phi}}(\mathscr F)$, ${\mathbf{\Phi}}(\mathscr G)$, ${\mathbf{\Phi}}(\mathscr Z)$, and if $F$ is proper then $F^{-1}$ preserves ${\mathbf{\Phi}}(\mathscr K)$.
The purpose of the next lemmas is to construct special maps with prescribed sets as pre-images. For $F:X\to Y$, we consider its [*kernel*]{} $\ker F=\{F^{-1}\{y\}:y\in Y\}$ and [*algebra of pre-images*]{} ${\mathop{\mathrm {alg\,}}\nolimits}F=\{F^{-1}B:B\subseteq Y\}.$
\[l: algebra of preimages\] For any $F:X\to Y$ we have $${\mathop{\mathrm {alg\,}}\nolimits}F=
\{A\subseteq X:F^{-1}FA=A\}.$$ Moreover, ${\mathop{\mathrm {alg\,}}\nolimits}F$ is a complete subalgebra of $\mathscr P(X)$ generated by $\ker F$ and thus isomorphic to $\mathscr P(\ker F)$. Consequently, ${\mathop{\mathrm {alg\,}}\nolimits}F$ is closed under Hausdorff operations.
Clear.
Given maps $F_i:X\to Y_i$, $i\in I$, their [*diagonal product*]{} is the map ${\bigtriangleup}_{i\in I}F_i:X\to\prod_{i\in I}Y_i$ defined by letting for all $x\in X$, $${\bigtriangleup}_{i\in I}F_i(x)=
(F_i(x))_{i\in I}.$$ The diagonal product of continuous maps $F_i$ is continuous (w.r.t. the standard product topology on $\prod_{i\in I}Y_i$), and moreover, it is perfect whenever so is at least one of them, say, $F_j$, and the spaces $Y_i$ for all $i\ne j$ are Hausdorff (see [@Engelking], Theorem 3.7.9).
\[l: algebra vs diag prod\] If $A\in{\mathop{\mathrm {alg\,}}\nolimits}F_j$ for some $j\in I$, then $A\in{\mathop{\mathrm {alg\,}}\nolimits}({\bigtriangleup}_{i\in I}F_i)$.
Let $F={\bigtriangleup}_{i\in I}F_i$. If $A=F^{-1}_jB$ for some $B\subseteq Y_j$ then $A=F^{-1}(B\times\prod_{i\in I\setminus\{j\}}Y_i)$.
As usual, a class $\mathscr Y$ of topological spaces is [*closed under $\kappa$ products*]{} iff $(Y_\alpha)_{\alpha<\kappa}\in\mathscr Y^\kappa$ implies $\prod_{\alpha<\kappa}Y_\alpha\in\mathscr Y$. E.g., the class of Polish spaces is closed under $\omega$ products, the class of spaces of density $\lambda\ge\omega$ is closed under $2^\lambda$ products (see [@Engelking], 2.3.15), and $\mathscr K$ is closed under arbitrary products. Similarly, a class $\mathscr M$ of maps is [*closed under $\kappa$ diagonal products*]{} iff $(F_\alpha)_{\alpha<\kappa}\in\mathscr M^\kappa$ implies ${\bigtriangleup}_{\alpha<\kappa}F_\alpha\in\mathscr M$. E.g., the classes of continuous and of perfect maps are closed under arbitrary products.
\[p: algebra vs diag prod\] Let $\mathscr Y$ be closed under $\kappa$ products, $\mathscr M$ a class of maps closed under $\kappa$ diagonal products, and let $\mathscr S$ be such that for any $S\in\mathscr S(X)$ there exist $Y\in\mathscr Y$ and $F\in\mathscr M\cap Y^X$ such that $S\in{\mathop{\mathrm {alg\,}}\nolimits}F$. Then for any $(S_\alpha)_{\alpha<\kappa}\in\mathscr S(X)^\kappa$ there exist $Y\in\mathscr Y$ and $F\in\mathscr M\cap Y^X$ such that $S_\alpha\in{\mathop{\mathrm {alg\,}}\nolimits}F$ for all $\alpha<\kappa$.
For each $\alpha<\kappa$ pick $Y_\alpha\in\mathscr Y$ and $F_\alpha\in\mathscr M\cap{Y_\alpha}^X$ with $S_\alpha\in{\mathop{\mathrm {alg\,}}\nolimits}F_\alpha$. Let $Y=\prod_{\alpha<\kappa}Y_\alpha$ and $F={\bigtriangleup}_{\alpha<\kappa}F_\alpha$. Then $Y\in\mathscr Y$ since $\mathscr Y$ is closed under $\kappa$ products, $F\in\mathscr M\cap Y^X$ since $\mathscr M$ is closed under $\kappa$ diagonal products, and moreover, $S_\alpha\in{\mathop{\mathrm {alg\,}}\nolimits}F$ for all $\alpha<\kappa$ by Lemma \[l: algebra vs diag prod\].
The following Proposition \[p: zero sets vs diag prod\] is essentially a variant of Proposition \[p: algebra vs diag prod\] where we have $\mathscr S=\mathscr Z$, $\kappa=\omega$, $\mathscr Y=\{[0,1]^\omega\}$, and $\mathscr M$ consists of continuous maps witnessing that sets $A_n$ are in $\mathscr Z(X)$.
\[p: zero sets vs diag prod\] Let ${\mathbf{\Phi}}$ be a Hausdorff operation, $X$ a topological space, and let $(A_n)_{n<\omega}\in\mathscr Z(X)^\omega$. Then there exists a continuous map $F:X\to[0,1]^\omega$ such that ${\mathbf{\Phi}}(A_n)_{n<\omega}\in{\mathop{\mathrm {alg\,}}\nolimits}F$. Moreover, the same remains true for every countable set of Hausdorff operations.
For each $n<\omega$ pick a continuous $F_n:X\to[0,1]$ with $A_n=F^{-1}_n\{0\}$ (which is possible since $A_n$ is in $\mathscr Z(X)$), and thus $A_n\in{\mathop{\mathrm {alg\,}}\nolimits}F_n$. Then $F={\bigtriangleup}_{n<\omega}F_n:X\to[0,1]^\omega$ is continuous, $A_n\in{\mathop{\mathrm {alg\,}}\nolimits}F$ by Lemma \[l: algebra vs diag prod\], and so ${\mathbf{\Phi}}(A_n)_{n<\omega}\in{\mathop{\mathrm {alg\,}}\nolimits}F$ by Lemma \[l: algebra of preimages\].
We turn to the problem of when classes of $\mathbf\Phi$-sets are preserved under maps in the images direction. Easily, the images of a map $F$ distribute over (even binary) intersections iff $F$ is one-to-one. Below we observe that the situation is less trivial if we consider intersections of families of sets directed by the converse inclusion.
A map $F:X\to Y$ is [*finite-to-one*]{} iff $\ker F\subseteq\mathscr{P}_\omega(X)$, and [*compact-to-one*]{} iff $\ker F\subseteq\mathscr{K}(X)$. Let us also say that $F$ is [*compact-to-one on compact sets*]{} iff $F^{-1}\{y\}\cap A\in\mathscr K(X)$ for all $y\in Y$ and $A\in\mathscr K(X)$. Finally, $F$ is [*perfect*]{} iff it is continuous, closed, and proper.
Trivially, any finite-to-one or proper $F$ is compact-to-one. Also, if $X$ is Hausdorff then any compact-to-one $F$ is compact-to-one on compact sets, if $Y$ is locally compact Hausdorff then any continuous closed compact-to-one $F$ is perfect, and if $X$ is compact and $Y$ is Hausdorff then any continuous $F$ is perfect.
Given a partially ordered set $(I,<)$, we shall say that a family $(A_i)_{i\in I}$ of sets is [*decreasing*]{} iff $A_i\supseteq A_j$ for all $i\le j$. Considering below $\omega$ and $\omega^{<\omega}$ as sets of indices, we imply the natural orderings $\le$ and $\subseteq$ of them.
The following result provide conditions under which images distribute over intersections of directed decreasing families.
\[p: intersection vs image\] Let $F:X\to Y$. The equality $
F\bigcap_{i\in I}A_i
=\bigcap_{i\in I}FA_i
$ holds for all directed $(I,<)$ and
- all decreasing $(A_i)_{i\in I}$ in $\mathscr P(X)$ if $F$ is finite-to-one,
- all decreasing $(A_i)_{i\in I}$ in $\mathscr K(X)$ if $F$ is compact-to-one on compact sets.
Since the inclusion $
F\bigcap_{i\in I}A_i
\subseteq\bigcap_{i\in I}FA_i
$ holds always, we prove the converse inclusion.
(i). If $F$ is finite-to-one, let $(I,<)$ be a directed set and $(A_i)_{i\in I}$ a family of nonempty sets such that $A_i\supseteq A_j$ if $i\le j$. Fix any $y\in\bigcap_{i\in I}FA_i$, i.e., $y$ such that $F^{-1}\{y\}\cap A_i\ne\emptyset$ for all $i\in I$, and show that $y\in F\bigcap_{i\in I}A_i$, i.e., that $F^{-1}\{y\}\cap\bigcap_{i\in I}A_i\ne\emptyset$. Since $F$ is finite-to-one, $|F^{-1}\{y\}|<\omega$, say, $F^{-1}\{y\}=\{x_k\}_{k<n}$ for some $n\in\omega$. Toward a contradiction, assume $F^{-1}\{y\}\cap\bigcap_{i\in I}A_i=\emptyset$, so for any $k<n$ there is $i_k\in I$ such that $x\notin A_{i_k}$. Since $(A_i)_{i\in I}$ is decreasing, and so $\supseteq$-directed, there exists $i\in I$ such that $A_i\subseteq\bigcap_{k<n}A_{i_k}$. But then for every $k<n$ we have $x_k\notin A_i$, thus showing $F^{-1}\{y\}\cap A_i\ne\emptyset$; a contradiction.
(ii). If $F$ is compact-to-one on compact sets, let $(I,<)$ be a directed set and $(A_i)_{i\in I}$ a family of nonempty compact sets such that $A_i\supseteq A_j$ if $i\le j$. If $y\in\bigcap_{i\in I}FA_i$ then the intersections $B_i=F^{-1}\{y\}\cap A_i$ are nonempty for all $i\in I$. Moreover, $B_i$ are compact (since $F$ is compact-to-one on compact sets) and form a $\supseteq$-directed family (as $B_i$ are nonempty and $A_i$ form a $\supseteq$-directed family). Any $\supseteq$-directed family of nonempty compact sets has a nonempty intersection, so pick an $x\in\bigcap_{i\in I}B_i$. We have $x\in F^{-1}\{y\}\cap\bigcap_{i\in I} A_i$ and hence $y\in F\bigcap_{i\in I}A_i.$
\[l: hausd vs image\] Let ${\mathbf{\Phi}}$ be a Hausdorff operation and $F:X\to Y$. The equality $
F{\mathbf{\Phi}}(A_s)_{s\in\omega^{<\omega}}=
{\mathbf{\Phi}}(FA_s)_{s\in\omega^{<\omega}}
$ holds
- for all decreasing $(A_s)_{s\in\omega^{<\omega}}$ in $\mathscr P(X)$ if $F$ is finite-to-one,
- for all decreasing $(A_s)_{s\in\omega^{<\omega}}$ in $\mathscr K(X)$ if $F$ is compact-to-one on compact sets.
Let $S\subseteq\omega^\omega$ be a base of ${\mathbf{\Phi}}$. Since the images of $F$ distribute over arbitrary unions and, by Proposition \[p: intersection vs image\], over intersections of decreasing families of (compact) sets if $F$ is finite-to-one (compact-to-one on compact sets), we have: $$\begin{aligned}
F{\mathbf{\Phi}}(A_s)_{s\in\omega^{<\omega}}=
F\bigcup_{f\in S}\bigcap_{n\in\omega}A_{f{\upharpoonright}n}
&=
\bigcup_{f\in S}F\bigcap_{n\in\omega}A_{f{\upharpoonright}n}
\\
&=
\bigcup_{f\in S}\bigcap_{n\in\omega}FA_{f{\upharpoonright}n}
=
{\mathbf{\Phi}}(FA_s)_{s\in\omega^{<\omega}},\end{aligned}$$ as required.
\[c: hausd vs image\] Let ${\mathbf{\Phi}}$ be a Hausdorff operation, $\mathscr S$ a class of sets closed under finite intersections, and $F:X\to Y$. Then:
- if $F$ preserves $\mathscr S$ and is finite-to-one, then $F$ preserves ${\mathbf{\Phi}}(\mathscr S)$, i.e., $F{\mathbf{\Phi}}(\mathscr S,X)\subseteq{\mathbf{\Phi}}(\mathscr S,Y)$,
- if $F$ preserves $\mathscr S\cap\mathscr K$ and is compact-to-one on compact sets, then $F$ preserves ${\mathbf{\Phi}}(\mathscr S\cap\mathscr K)$, i.e., $
F{\mathbf{\Phi}}(\mathscr S\cap\mathscr K,X)
\subseteq{\mathbf{\Phi}}(\mathscr S\cap\mathscr K,Y).
$
If $\mathscr S(X)$ is closed under finite intersections, then every $(A_s)_{s\in\omega^{<\omega}}$ in $\mathscr S(X)$ can be replaced with a decreasing $(B_s)_{s\in\omega^{<\omega}}$ in $\mathscr S(X)$ so that $$\mathbf\Phi(A_s)_{s\in\omega^{<\omega}}=
\mathbf\Phi(B_s)_{s\in\omega^{<\omega}}$$ by letting $B_{f{\upharpoonright}n}=\bigcap_{k\le n}A_{f{\upharpoonright}k}$. Now the claim follows from Lemma \[l: hausd vs image\].
E.g., as each of $\mathscr F,\mathscr G,\mathscr Z$, and $\mathscr K$ for Hausdorff spaces, is closed under finite intersections, we see: if $F$ is closed and finite-to-one then it preserves ${\mathbf{\Phi}}(\mathscr F)$; if $F$ is open and finite-to-one then it preserves ${\mathbf{\Phi}}(\mathscr G)$; if $X,Y$ are Hausdorff, $X$ is compact, and $F$ is continuous, then it preserves ${\mathbf{\Phi}}(\mathscr S)$ where $\mathscr S$ is each of $\mathscr F,\mathscr G,\mathscr Z,\mathscr K$.
Now we combine our previous results to transfer the reduction and separation properties in the pre-image direction.
\[p: preimage vs red\] Let ${\mathbf{\Phi}}$ be a Hausdorff operation and $\mathscr S$ a class closed under finite intersections and such that for any $(A_n,B_n)_{n\in\omega}$ in $\mathscr S(X)$ there are $Y$ and $F:X\to Y$ such that
- $F^{-1}$ preserves $\mathscr S$,
- $(A_n,B_n)_{n\in\omega}$ is in ${\mathop{\mathrm {alg\,}}\nolimits}F$, and
- $
F{\mathbf{\Phi}}(A_n)_{n\in\omega},
F{\mathbf{\Phi}}(B_n)_{n\in\omega}
$ are reduced (separated) by sets in ${\mathbf{\Phi}}(\mathscr S,Y)$.
Then ${\mathbf{\Phi}}(\mathscr S,X)$ has the reduction (separation) property.
Prove, e.g., reduction. Pick any $A,B$ in ${\mathbf{\Phi}}(\mathscr S,X)$ and $(A_n)_{n\in\omega}$, $(B_n)_{n\in\omega}$ such that $A={\mathbf{\Phi}}(A_n)_{n\in\omega}$, $B={\mathbf{\Phi}}(B_n)_{n\in\omega}$. Let $Y$ and $F:X\to Y$ be such that $F$ preserves $\mathscr S$, all the sets $A_n,B_n$ are in ${\mathop{\mathrm {alg\,}}\nolimits}F$, and the sets $FA=F{\mathbf{\Phi}}(A_n)_{n\in\omega}$, $FB=F{\mathbf{\Phi}}(B_n)_{n\in\omega}$ are reduced by some sets in ${\mathbf{\Phi}}(\mathscr S,Y)$, i.e., there exist $C,D$ in ${\mathbf{\Phi}}(\mathscr S,Y)$ such that $$C\cap D=\emptyset
\;\text{ and }\;
C\cup D=(FA)\cup(FB).$$ As $F^{-1}$ preserves $\mathscr S$, it preserves ${\mathbf{\Phi}}(\mathscr S)$ by Corollary \[c: hausd vs preimage\], so $F^{-1}C,F^{-1}D$ are in ${\mathbf{\Phi}}(\mathscr S,X)$. Moreover, we have: $F^{-1}C\cap F^{-1}D=\emptyset$ and $$\begin{aligned}
F^{-1}C\cup F^{-1}D=
F^{-1}(C\cup D)
&=
F^{-1}((FA)\cup(FB))
\\
&=
F^{-1}F(A\cup B)=
A\cup B.\end{aligned}$$ Here the first three equalities uses only that pre-images and images distribute over unions, while the last equality uses that the $A_n,B_n$ are in ${\mathop{\mathrm {alg\,}}\nolimits}F$, and so by Lemma \[l: algebra of preimages\], $A\cup B$ is also in ${\mathop{\mathrm {alg\,}}\nolimits}F$.
This proves reduction in ${\mathbf{\Phi}}(\mathscr S,X)$, as required.
\[p: image preimage vs red\] Let ${\mathbf{\Phi}}$ be a Hausdorff operation and $\mathscr S$ a class of sets.
- If $\mathscr S$ is closed under finite intersections and such that for any $(A_n)_{n\in\omega}$ in $\mathscr S(X)$ there are $Y$ and a finite-to-one $F:X\to Y$ such that
- $F$ and $F^{-1}$ preserve $\mathscr S$,
- $(A_n)_{n\in\omega}$ is in ${\mathop{\mathrm {alg\,}}\nolimits}F$, and
- ${\mathbf{\Phi}}(\mathscr S,Y)$ has the reduction (separation) property,
then ${\mathbf{\Phi}}(\mathscr S,X)$ has the same property.
- The same remains true assuming $\mathscr S\subseteq\mathscr K$ and that such $F$ are (not necessarily finite-to-one but) compact-to-one on compact sets.
As $F$ preserves $\mathscr S$ and is finite-to-one (compact-to-one on compact sets), it preserves ${\mathbf{\Phi}}(\mathscr S)$ by Corollary \[c: hausd vs image\], so $FA,FB$ are in ${\mathbf{\Phi}}(\mathscr S,Y)$, and so by reduction in ${\mathbf{\Phi}}(\mathscr S,Y)$, they are reduced by some sets in ${\mathbf{\Phi}}(\mathscr S,Y)$. Now we are in position to apply Proposition \[p: preimage vs red\] thus getting the same conclusion.
\[t: red tych\] Let $X$ be a Tychonoff space and ${\mathbf{\Phi}}$ a Hausdorff operation. If ${\mathbf{\Phi}}(\mathscr G_\delta,\mathbb R)$ has the reduction (separation) property, then ${\mathbf{\Phi}}(\mathscr F\cap\mathscr G_\delta,X)$ has the same property.
First show that the claim is true if $X$ is any Tychonoff cube $[0,1]^\kappa$. For $\kappa=\omega$ this is trivial since $[0,1]^\omega$ is Polish. For arbitrary $\kappa$, let us verify that the assumptions of Proposition \[p: image preimage vs red\](ii) are met with $\mathscr S=\mathscr F\cap\mathscr G_\delta$ and $Y=[0,1]^\omega$ common for all $(A_n)_{n\in\omega}$ in $\mathscr S(Y)$.
Indeed, since the space $[0,1]^\kappa$ is normal, $
(\mathscr F\cap\mathscr G_\delta)([0,1]^\kappa)=
\mathscr Z([0,1]^\kappa),
$ and if $(A_n)_{n\in\omega}$ is in $\mathscr Z([0,1]^\kappa)$, then Proposition \[p: zero sets vs diag prod\] gives a continuous map $F:[0,1]^\kappa\to[0,1]^\omega$ such that ${\mathbf{\Phi}}(A_n)_{n\in\omega}\in{\mathop{\mathrm {alg\,}}\nolimits}F$. But again since $[0,1]^\kappa$ is compact, $
\mathscr Z([0,1]^\kappa)\subseteq
\mathscr K([0,1]^\kappa),
$ and moreover, $F$ is perfect (as a map of a compact space into a Hausdorff space). Hence, once we have reduction (separation) in ${\mathbf{\Phi}}(\mathscr F\cap\mathscr G_\delta,[0,1]^\omega)$, or equivalently, in ${\mathbf{\Phi}}(\mathscr G_\delta,\mathbb R)$, we are able to apply Proposition \[p: image preimage vs red\](ii), thus getting the same property in ${\mathbf{\Phi}}(\mathscr F\cap\mathscr G_\delta,[0,1]^\kappa)$.
Now let $X$ be an arbitrary Tychonoff space. As well-known (see, e.g., [@Engelking], 2.3.23), letting $\kappa=w(X)$, we can identify $X$ with a subspace of $[0,1]^\kappa$. Then we have $
(\mathscr F\cap\mathscr G_\delta)(X)=
(\mathscr F\cap\mathscr G_\delta)([0,1]^\kappa){\upharpoonright}X,
$ and now Corollary \[c: hausd in subspace\](ii) gives the required conclusion.
The proof is complete.
In particular, Borel and projective classes in Tychonoff spaces generated from their closed $\mathscr G_\delta$-sets have the same pattern of reduction and separation as they do in the real line:
\[c: period tych\] Let $X$ be a Tychonoff space. Then:
- for all $\alpha<\omega_1$, $
\mathbf\Sigma^{0}_{\alpha}
(\mathscr F\cap\mathscr G_\delta,X)
$ have the reduction property while $
\mathbf\Pi^{0}_{\alpha}
(\mathscr F\cap\mathscr G_\delta,X)
$ have the separation property,
- under ${\mathrm{PD}}$, for all $n<\omega$, $
\mathbf\Sigma^{1}_{2n}
(\mathscr F\cap\mathscr G_\delta,X)
$ and $
\mathbf\Pi^{1}_{2n+1}
(\mathscr F\cap\mathscr G_\delta,X)
$ have the reduction property while $
\mathbf\Sigma^{1}_{2n+1}
(\mathscr F\cap\mathscr G_\delta,X)
$ and $
\mathbf\Pi^{1}_{2n}
(\mathscr F\cap\mathscr G_\delta,X)
$ have the separation property.
As well-known, if $X$ is $\mathbb R$ (or another Polish space), both items hold. In fact, all Borel classes $\mathbf\Sigma^{0}_{\alpha}$ have the pre-well-ordering property, so all they have reduction while the dual classes $\mathbf\Pi^{0}_{\alpha}$ have separation (see [@Moschovakis], p. 37), and under ${\mathrm{PD}}$, all projective classes $\mathbf\Sigma^{1}_{2n}$ and $\mathbf\Pi^{1}_{2n+1}$ have the pre-well-ordering property (the fact known as the First Periodicity Theorem), so all they have reduction while the dual classes $\mathbf\Pi^{1}_{2n}$ and $\mathbf\Sigma^{1}_{2n+1}$ have separation (see [@Kechris], [@Moschovakis], or [@Kanamori], 29.14). Now apply Theorem \[t: red tych\].
[111]{}
R. Engelking. [*General Topology*]{}. PWN, Warszawa, 1977.
A. Kanamori. [*The higher infinite*]{}. Second edition, Springer, 2003.
A.S. Kechris. [*Classical descriptive set theory.*]{} Springer, 1994.
Y.N. Moschovakis. [*Descriptive set theory.*]{} Second edition, Mathematical surveys and monographs 155, 2009.
+2em
[The Russian Academy of Sciences, Institute for Information Transmission Problems, Bolshoy Karetny lane 19, build. 1, Moscow 127051 Russia ]{}\
[*E-mail address:*]{} [email protected], [email protected]
[^1]: This work was partially supported by grant 17-01-00705 of Russian Foundation for Basic Research and carried out at Institute for Information Transmission Problems of Russian Academy of Sciences.
[^2]: [*MSC 2010*]{}: Primary 03E15, 03E60, 54H05, Secondary 54C10, 54D30.
[^3]: [*Keywords*]{}: descriptive set theory, reduction, separation, First Periodicity Theorem, Tychonoff space, determinacy, projective set, Borel set, zero set, perfect map, Hausdorff operation.
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- '[**Jinho Baik,**]{}[^1] [^2] [**Percy Deift**]{} [^3] [^4] and [**Eric M. Rains**]{}[^5]'
bibliography:
- 'paper9.bib'
title: |
[**A Fredholm Determinant Identity and\
the Convergence of Moments\
for Random Young Tableaux**]{}
---
[^1]: Department of Mathematics, Princeton University, Princeton, New Jersey, 08544, [email protected]
[^2]: Institute for Advanced Study, Princeton, New Jersey 08540
[^3]: University of Pennsylvania, Philadelphia, 19104, [email protected]
[^4]: Courant Institute of Mathematical Sciences, New York 10012
[^5]: AT&T Research, New Jersey, Florham Park, New Jersey 07932, [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Classification and regression in which the inputs are graphs of arbitrary size and shape have been paid attention in various fields such as computational chemistry and bioinformatics. Subgraph indicators are often used as the most fundamental features, but the number of possible subgraph patterns are intractably large due to the combinatorial explosion. We propose a novel efficient algorithm to jointly learn relevant subgraph patterns and nonlinear models of their indicators. Previous methods for such joint learning of subgraph features and models are based on search for single best subgraph features with specific pruning and boosting procedures of adding their indicators one by one, which result in linear models of subgraph indicators. In contrast, the proposed approach is based on directly learning regression trees for graph inputs using a newly derived bound of the total sum of squares for data partitions by a given subgraph feature, and thus can learn nonlinear models through standard gradient boosting. An illustrative example we call the Graph-XOR problem to consider nonlinearity, numerical experiments with real datasets, and scalability comparisons to naïve approaches using explicit pattern enumeration are also presented.'
author:
- Ryo Shirakawa
- Yusei Yokoyama
- Fumiya Okazaki
- Ichigaku Takigawa
bibliography:
- 'all.bib'
title: Jointly learning relevant subgraph patterns and nonlinear models of their indicators
---
Introduction
============
Graphs are fundamental data structures for representing combinatorial objects. However, precisely because of their combinatorial nature, it is usually difficult to understand the underlying trends in large datasets of graphs. The rapid increase in data in recent years also includes data represented as graphs, and thus supervised learning in which the inputs are graphs of arbitrary size and shape has gained considerable attention. This problem commonly arises in diverse fields such as cheminformatics [@Kashima:2003; @Tsuda:2007; @Saigo:2008a; @Saigo:2009; @Mahe:2009; @Vishwanathan:2010; @Shrvashidze:2011; @Takigawa:2013; @takigawa:2017], and bioinformatics [@Borgwardt:2005; @Karklin:2005; @Takigawa:2011b] as well as wide computer-science applications such as computer vision [@Harchaoui:2007; @Nowozin:2007; @Barra:2013; @Bai:2014a], and natural language processing [@Kudo:2005]. The present paper investigates the supervised learning of a function $f: {\mathcal{G}}\to {\mathcal{Y}}$ from finite pairs of input graphs and output values, where ${\mathcal{G}}$ is a set of graphs and ${\mathcal{Y}}$ is a label space such as $\{-1,+1\}$ and $\mathbb{R}$. In general settings, the most fundamental and widely used features are indicators of subgraph patterns. Since the number of possible subgraph patterns are intractably large due to the combinatorial explosion, we need to use a heuristically limited class of subgraph patterns or to search for relevant patterns during the learning phase.
In addition to extensive studies on *graph kernels* [@Kashima:2003; @Borgwardt:2005; @Harchaoui:2007; @Mahe:2009; @Vishwanathan:2010; @Shrvashidze:2011; @Barra:2013; @Bai:2014a], joint learning of relevant subgraph patterns and classification/regression models by their indicators has also been developed [@Kudo:2005; @Nowozin:2007; @Saigo:2008a; @Saigo:2009; @takigawa:2017]. This approach would not overlook any important features, but need some technical tricks to efficiently search for relevant subgraph patterns from combinatorially huge candidates. The previous methods use $\ell_1$ regularization for *linear* models of all possible subgraph indicators, and thus can select relevant subgraph patterns. In contrast, any practical graph kernels are based on all subgraphs in a predefined class, and do not try to select some relevant subsets of subgraph features. Note that the *all-subgraphs* kernel is known to be theoretically hard[@Gartner:2003].
In the present paper, we investigate *nonlinear* models with all possible subgraph indicators. The following are the contributions of the present study:
- We present two lesser-recognized facts to make sure the difference between linear and nonlinear models of substructural indicators: (1) For a closely related problem of supervised learning from itemsets, the hypothesis space of the nonlinear model of all possible sub-itemset indicators is equivalent to that of the linear model; (2) Nevertheless, for the indicators of connected subgraphs, the hypothesis space of the nonlinear model is strictly larger than that of the linear model. [**(Section \[sec:linearNonlinear\])**]{}
- We develop a novel efficient supervised learning algorithm for joint learning of all relevant subgraph features and a nonlinear models of their indicators. Unlike existing approaches based on $\ell_1$-regularized linear models, the proposed algorithm is based on gradient tree boosting with base regression trees selecting each splitter out of all subgraph indicators with an efficient pruning based on the new bound in Theorem \[thm:bound\]. [**(Section \[sec:proposed\])**]{}
- We empirically demonstrate that (i) for the Graph-XOR dataset, the proposed nonlinear method actually outperforms several linear methods, which implies the existence of problems requiring nonlinear hypotheses, (ii) For several real datasets, we also observe similar superiority of the nonlinear models for some datasets, while it also turns out that the performance of linear models is fairly comparable for some datasets. [**(Section \[sec:experi\])**]{}
Related Research and Our Motivation {#sec:previousModel}
-----------------------------------
Although not discussed explicitly, most previous studies [@Kudo:2005; @Nowozin:2007; @Saigo:2008a; @Saigo:2009; @takigawa:2017] yielded linear models with respect to subgraph indicators as Boolean variables. However, this would not be obvious at first glance because these studies were based on *boosting* such as Adaboost [@Kudo:2005] and LPBoost [@Saigo:2009], which are usually expected to produce nonlinear models. However, this is not the case because these studies used *decision stumps* with respect to a single subgraph feature as base learners.
The research of the present paper starts with our observation that replacing the decision stumps in these existing methods with decision trees is far from straightforward. This is because the previous methods are based on efficient pruning with specifically derived bounds to find a single best subgraph pattern, and use the indicator as a base learner at each iteration.
One naïve method to obtain nonlinear models of subgraph indicators is to enumerate some candidate subgraphs from training graphs, explicitly construct 0-1 indicator-feature vectors of test graphs by solving subgraph-isomorphism directly, and apply a general nonlinear supervised learning to those feature vectors. The performance with all small-size subgraphs occurred in the given graphs is known to be comparable for cheminformatics datasets[@Wale:2008]. However these approaches would not scale well as we see later in Section \[sec:scalability\]. Another good known heuristic idea is to use $r$-neighborhood subgraphs with radius $r$ at each node as seen in ECFP[@Rogers:2010] and graph convolutions[@Kearnes2016; @gilmer:2017]. Unfortunately, the complete enumeration would not scale well either in this case, and usually requires some tricks such as feature hashing, feature folding, or feature embedding through neural nets, all of which are very interesting approaches but beyond the scope of this paper.
Note that it is not difficult to use the number of occurrences of subgraph $g$ in $G$ as features instead of just 0-1 subgraph indicators. Although not discussed herein, this case can be investigated as a weighted version of indicators, and similar properties would hold.
Preliminaries
=============
Notations
---------
Let $[n]$ be $\{1,2,\dots,n\}$, and let ${\mathbb{I}}(P)$ denote the indicator of $P$, i.e., ${\mathbb{I}}(P)=1$ if $P$ is true, else $0$. We denote as $G \sqsupseteq g$ the subgraph isomorphism that $G$ contains a subgraph that is isomorphic to $g$ and its negation as $G \not\sqsupseteq g$. Thus, a subgraph indicator ${\mathbb{I}}(G \sqsupseteq g) = 1$ if $G \sqsupseteq g$, otherwise 0. We also denote the training set of input graphs $G_i \in {\mathcal{G}}$ and output responses $y_i \in {\mathcal{Y}}$ as $$\label{eq:train_data}
{\mathcal{D}}= \{(G_1,y_1),(G_2,y_2),\dots,(G_N,y_N)\},$$ where ${\mathcal{G}}$ is a set of all finite-size, connected, discretely-labeled, undirected graphs. We denote ${\mathcal{G}}_N = \{G_i\mid i \in [N]\}$, and the set of all possible connected subgraphs as ${\mathcal{S}}_N = \bigcup_{G \in {\mathcal{G}}_N} \{g \mid G \sqsupseteq g\}$.
Search Space for Subgraphs {#sec:subgraphMining}
--------------------------
In supervised learning from graphs, we represent each input graph $G_i \in
{\mathcal{G}}_N$ by the characteristic vector $({\mathbb{I}}(G_i \sqsupseteq g) \mid g \in
{\mathcal{S}}) $ with a set ${\mathcal{S}}$ of relevant subgraph features. However, since ${\mathcal{S}}$ is not explicitly available when the learning phase starts, we need to jointly search and construct ${\mathcal{S}}$ during the learning process. In order to define an efficient search space for ${\mathcal{S}}_N$, i.e., any subgraphs occurring in ${\mathcal{G}}_n$, the techniques for *frequent subgraph mining*, which enumerates all subgraphs that appear in more than $m$ input graphs for a given $m$, are useful. Note that any subgraph feature $g \in {\mathcal{S}}_N$ can occur multiple times at multiple locations in a single graph, but ${\mathbb{I}}(G_i \sqsupseteq g) = 1$.
In the present paper, we use the search space of the gSpan algorithm [@Yan:2002], which performs a depth-first search on the tree-shaped search spaces on ${\mathcal{S}}_N$, referred to collectively as an *enumeration tree*, as shown in Figure \[fig:enum\]. Each node of the enumeration tree holds a subgraph feature $g'$ that extends the subgraph feature $g$ at the parent node by one edge, namely, $ g' \sqsupseteq g $.
![An enumeration tree[]{data-label="fig:enum"}](graph_search_tree.eps){width="0.6\linewidth"}
The following *anti-monotone* property of subgraph isomorphism over the enumeration tree on ${\mathcal{S}}_N$ can be used to derive the efficient search-space pruning of the gSpan algorithm: $$\label{eq:propSubgraph}
G_i \not\sqsupseteq g \Rightarrow G_i \not\sqsupseteq g'
\quad \text{for} \quad
g' \sqsupseteq g . $$
Gradient Tree Boosting {#sec:GTB}
----------------------
Gradient tree boosting (GTB) [@friedman2001greedy; @mason2000boosting] is a general algorithm for supervised learning to predict a response $y$ from a predictor $\bm{x}$. For a given hypothesis space ${\mathcal{H}}$, the goal is to minimize the empirical risk $L(f) = N^{-1}\sum_{i \in [N]} \ell(y_i,f(\bm{x}_i))$ of $f \in
{\mathcal{H}}$, which is the average of a loss function $\ell(y,f(\bm{x}))$ over the training data $\{(\bm{x}_i, y_i)\}_{i \in[N]}$.
GTB is an additive ensemble model of regression trees $T_i(\bm{x})$ of the form for fixed-stepsize cases: $$f_k(\bm{x}) = T_0 + \eta \sum _{i \in [k]} T_i(\bm{x})$$ where $T_0$ is the mean of response variables in the training data, $\eta$ is the stepsize, and $T_i(\bm{x})$ is the $i$-th regression tree as a base learner. To fit the model to the training data, GTB performs the following gradient-descent-like iterations as a boosting procedure: $$\begin{aligned}
f_0(\bm{x}) \leftarrow \arg \min_c \sum_{i \in [N]} \ell(y_i, c), \\
\quad\text{and}\quad
f_k(\bm{x}) \leftarrow f_{k-1}(\bm{x}) + \eta \, T_k(\bm{x}),\end{aligned}$$ where $T_k(\bm{x})$ is a regression tree to best approximate the values of negative functional gradient $-\nabla_{f_{k-1}} L(f_{k-1}) $ at $f_{k-1}$ obtained by fitting a regression tree to the data $\left\{\left(
\bm{x}_i, r_i
\right)\right\}_{i \in [N]}$, where $$r_i =
-\left[\frac{\partial\, \ell(y_i,
f(\bm{x}_i))}{\partial
f(\bm{x}_i)}\right]_{f(\bm{x})=f_{k-1}(\bm{x}_i)}$$ Our experiments focus on binary classification tasks with $y \in
\{-1,+1\}$, and thus we use the logistic loss $\ell(y, \mu) = \log
(1+\exp(-2 y \mu))$. Note that even for classification, GTB must fit regression trees instead of classification trees in order to approximate real-valued functions.
The primary hyperparameters of GTB that we consider are the following three parameters: a max tree-depth $d$, a stepsize $\eta$, and the number of trees $k$.
Regression Trees
----------------
The internal regression-tree fitting is performed by the recursive partitioning below:
1. Each node in the regression tree receives a subset ${\mathcal{D}}' \subseteq {\mathcal{D}}$ from the parent node.
2. If a terminal condition is satisfied, the node becomes a leaf decision node with a prediction value by the average of the response values in ${\mathcal{D}}'$.
3. Otherwise, the node becomes an internal node that tries to find the best partition of ${\mathcal{D}}'$ to ${\mathcal{D}}_1$ and ${\mathcal{D}}_0 = {\mathcal{D}}' \setminus {\mathcal{D}}_1$ that minimizes the total sum of squares of residual error $r$: $${\min_{\scalebox{1.0}{${\mathcal{D}}_1, {\mathcal{D}}_0$}}} \bigl[ {\mathrm{TSS}}({\mathcal{D}}_1) + {\mathrm{TSS}}({\mathcal{D}}_0) \bigr] . $$ The subsets ${\mathcal{D}}_1$ and ${\mathcal{D}}_0$ are further sent to the child nodes, and Step (1) is then recursively applied to each subset at the child nodes.
${\mathrm{TSS}}$ here is the total sum of squares of residual error $r$: $$\label{eq:tss}
{\mathrm{TSS}}({\mathcal{D}}) = \frac{1}{2} \sum_{i \in [N]}
(r_i - \bar{r})^2, \,\, \bar{r} = \frac{1}{N} \sum_{i \in [N]}
r_i. $$
Problem Setting and Challenges
==============================
Our goal is learning a nonlinear model $f$ over all possible subgraph indicators ${\mathbb{I}}(G \sqsupseteq g)$ for $g \in {\mathcal{S}}_N$. As we will see in Section \[sec:linearNonlinear\], arbitrary functions of subgraph indicators have a unique multi-linear polynomial form $$f(G) = \sum_{S \subseteq {\mathcal{S}}_N} c_S \prod_{g \in S} {\mathbb{I}}(G \sqsupseteq g) . $$ Input graphs are implicitly represented as a bag of subgraph features, and hence as feature vectors in which the elements are an intractably large number of binary variables of each subgraph indicator. The main challenge is how to learn the relevant features $g$ from such combinatorially large space ${\mathcal{S}}_N$ with also jointly learning the classifier $f$ over those features ${\mathbb{I}}(G \sqsupseteq g)$ for $g \in {\mathcal{S}}_N$. Other technical challenges are (1) feature vectors are binary valued and takes finite discrete values only at the vertices of a very high-dimensional Boolean hypercube; (2) feature vectors are strongly correlated due to subgraph isomorphism.
Pseudo-Boolean Functions of Substructural Indicators {#sec:linearNonlinear}
====================================================
We first investigate the difference between linear and nonlinear models. Any subgraph indicator ${\mathbb{I}}(G \sqsupseteq g)$ is a 0-1 Boolean variable, and thus the hypothesis space that we can consider with respect to these variables is a family of *pseudo-Boolean functions*, regardless of whether they are linear or nonlinear. A real-valued function $f: \{0,1\}^d \to \mathbb{R}$ on the Boolean hypercube $\{0,1\}^d$ is called pseudo-Boolean.
In this section, we explain the inequivalence of linear and nonlinear models of all possible subgraph indicators. Theorem \[thm:linear\_nonlinear\], contrasting the difference from closely related problems for itemsets, suggests an advantage of the proposed nonlinear approach, and an illustrative example we call Graph-XOR indicating that linear models cannot learn is presented in Section \[sec:artiExperi\].
\[thm:linear\_nonlinear\] (1) The hypothesis space of the nonlinear model of all possible sub-itemset indicators is equivalent to that of the linear model. (2) The hypothesis space of the nonlinear model of all possible connected subgraph indicators is strictly larger than that of the linear model.
This result is based on the following fundamental property of pseudo-Boolean functions.
\[le:pseudoBooleanFunc\] [@Hammer1963; @hammer1968boolean] Every pseudo-Boolean function $f : \{0,1\}^d \to \mathbb{R}$ has a unique multi-linear polynomial representation: $$f(x_1, \dots, x_d) = \sum_{S \subseteq [d]} c_S \prod_{j \in S} x_j,
\quad x_j \in \{0,1\},\,\, c_S \in \mathbb{R}.$$
Sub-Itemset Indicators
----------------------
Let $x_j \in \{0,1\}$ be a Boolean variable defined by $x_j = {\mathbb{I}}(j \in
I)$ for an item $j \in [d]$ in a itemset $I \subseteq [d]$. Then we can see that linear and nonlinear models of *sub-itemset indicators* ${\mathbb{I}}(S \subseteq I), S \subseteq [d]$ are equivalent as a hypothesis space on itemsets. For any function $f: 2^{[d]} \to \mathbb{R}$, we have $$\begin{aligned}
f(I) &= \sum_{P \subseteq 2^{[d]}} c_P \prod_{S \in P} {\mathbb{I}}(S \subseteq I)
= \sum_{U \in 2^{[d]}} c_U {\mathbb{I}}(U \subseteq I)\\
&= \sum_{U \subseteq [d]} c_U \prod_{j \in U}{\mathbb{I}}(j \in I) \end{aligned}$$ where $U = \bigcup_{S \in P} S $. Theorem \[thm:linear\_nonlinear\] (1) follows from this simple fact. Note that including the negation terms, as in decision tree learning, does not change the hypothesis space because it can be represented as ${\mathbb{I}}(S \not\subseteq I) = 1 - {\mathbb{I}}(S \subseteq I)$.
Connected-Subgraph Indicators {#sec:ConnectedGraph}
-----------------------------
As for subgraph indicators ${\mathbb{I}}(G \sqsupseteq g)$, the standard setting implicitly assumes that subgraph feature $g$ is a *connected* graph. This would be primarily because the complete search for subgraph patterns, including disconnected graphs, is practically impossible, given that even a set of all connected graphs ${\mathcal{S}}_N$ in ${\mathcal{G}}_N$, is already intractably huge in practice.
The difference between linear model ${f_L}(G)$ and nonlinear model ${f_{NL}}(G)$ is not constantly zero: $$\begin{aligned}
\label{eq:ConnectedGraphLinear}
{f_L}(G) &=& c_0 + \sum_{g \in {\mathcal{S}}_N} c_g {\mathbb{I}}(G \sqsupseteq g) \\
{f_{NL}}(G) &=& c_0 + \sum_{g \in {\mathcal{S}}_N} c_g {\mathbb{I}}(G \sqsupseteq g) \\
~ &~& + \sum_{S \subseteq {\mathcal{S}}_N, |S| \geqslant 2} c_S \prod_{g \in S} {\mathbb{I}}(G \sqsupseteq g)\end{aligned}$$ from which Theorem \[thm:linear\_nonlinear\] (2) follows. This also implies that if we consider *connected-subgraph-set indicators* for the co-occurrence of several connected subgraphs, then the linear model is equivalent to the nonlinear model as a hypothesis space, and more importantly it is identical to ${f_{NL}}(G)$, which is the hypothesis space covered by the proposed algorithm in Section \[sec:proposed\]. Note that connected-subgraph-set indicators differ from the indicators of general subgraphs, including disconnected-subgraph-set indicators, because any subgraph feature $g \in {\mathcal{S}}$ can occur multiple times at different partially overlapped locations in a single graph. The hypothesis space by general subgraph indicators is beyond the scope of the present paper, and, in practice, the complete search for such indicators is computationally too challenging.
Proposed Method {#sec:proposed}
===============
In this section, we present a novel efficient method to produce a nonlinear prediction model based on gradient tree boosting with all possible subgraph indicators. Existing boosting-based methods [@Kudo:2005; @Nowozin:2007; @Saigo:2009] are based on simple but efficiently searchable base-learners of decision stumps (equivalent to subgraph indicators, as demonstrated previously) and construct an efficient pruning algorithm for this single best subgraph search at each iteration. In contrast, the proposed approach involves this subgraph search at finding an optimal split at each internal node of regression trees, while keeping the other outer loops the same as in GTB, as explained in Section \[sec:GTB\]. More specifically, we need to efficiently perform the following optimization over all possible subgraphs in ${\mathcal{S}}_N$: $$\label{eq:internal}
{\min_{\scalebox{1.0}{$g \in {\mathcal{S}}_N$}}} \Bigl[ {\mathrm{TSS}}({\mathcal{D}}_1(g)) + {\mathrm{TSS}}({\mathcal{D}}_0(g)) \Bigr]$$ where ${\mathrm{TSS}}$ is the total sum of squares defined as , ${\mathcal{D}}_1(g) = \{ (G_i, r_i) \in {\mathcal{D}}\mid G \sqsupseteq g \} $ and ${\mathcal{D}}_0(g) = \{ (G_i, r_i) \in {\mathcal{D}}\mid G \not\sqsupseteq g \} $.
Here, $|{\mathcal{S}}_N|$ is too intractably huge to solve (\[eq:internal\]) by exhaustively testing subgraph $g \in {\mathcal{S}}_N$ in order, and thus we perform a branch and bound search over the enumerate tree on ${\mathcal{S}}_N$ with the following lower bound for the total sum of squares of expanded subgraphs:
\[thm:bound\] Given ${\mathcal{D}}_1(g)$ and ${\mathcal{D}}_0(g)$, for any subgraph $g'
\sqsupseteq g$, $$\begin{gathered}
\label{eq:bound}
{\mathrm{TSS}}({\mathcal{D}}_1(g')) + {\mathrm{TSS}}({\mathcal{D}}_0(g')) \geq \\
{\min_{\scalebox{1.0}{$(\diamond,k)$}}} \Big[ {\mathrm{TSS}}({\mathcal{D}}_1(g) \setminus S_{\diamond, k}) + {\mathrm{TSS}}({\mathcal{D}}_0(g) \cup S_{\diamond, k}) \Big]
\end{gathered}$$ where $ (\diamond, k) \in \{ \leq, > \} \times \{ 2, \dots, |{\mathcal{D}}_1(g) - 1| \} $, and $S_{\diamond, k} \subset {\mathcal{D}}_1(g)$, such that $S_{\leq, k}$ is a set of $k$ pair $(G_i, r_i)$ selected from ${\mathcal{D}}_1(g)$ in descending order of residual error $r_i$, and $S_{>, k}$ is that in increasing order. Note that $\setminus$, $\cup$ are set difference and set union respectively.
The result follows from the property . See Appendix \[sec:boundLinear\] for details.
The entire procedure of the proposed algorithm is illustrated in Alg. \[alg:gtb\] and \[alg:rt\]. The novel algorithm for the optimal subgraph search of using the bound is described in detail in Alg. \[alg:proposed\]. In order to solve , the proposed algorithm uses a depth-first search on the enumerate tree over ${\mathcal{S}}_N$. The procedure at each subgraph is as follows:
1. Calculate the total sum of squares $\gets {\mathrm{TSS}}({\mathcal{D}}_1(g) +
{\mathcal{D}}_0(g))$ of subgraph $g$.
2. Update by if $>$ .
3. Calculate $\bound$ and if $<$ $\bound$, then prune all child nodes of .
In the entire procedure, the most time-consuming part is the subgraph search (Alg. \[alg:proposed\]), which is repeatedly called during the learning process. Hence, introducing memorization, whereby we store expensive calls and return the cached results when the same pattern occurs again, can considerably speed up the entire process. First, we can store the result of minimization ${\min_{\scalebox{1.0}{$g \in {\mathcal{S}}_N$}}} ({\mathrm{TSS}}({\mathcal{D}}_1(g)) + {\mathrm{TSS}}({\mathcal{D}}_0(g)))$ for each already checked $g$ and ${\mathcal{D}}$. Second, the subgraph search can be entirely skipped until we need to check any subgraph that has not been checked in any previous iterations.
Numerical Experiments {#sec:experi}
=====================
The Graph-XOR Problem {#sec:artiExperi}
---------------------
The linear separability has long been discussed using the XOR (or parity in general) example, and we present the same key example for graphs, referred to as *Graph-XOR*, where linear models cannot learn the target rule even when noiseless examples are provided.
The Graph-XOR dataset includes 1,035 graphs of seven nodes and six edges, where 506 are positives with $y=+1$ and 529 negatives with $y = -1$. As illustrated in Figure \[fig:artificialGraph\], each graph is generated by connecting two subgraphs by one node Ď. The component subgraphs are selected from the 18 types shown in Figure \[fig:parts\], where all three-node path graphs with candidate nodes {Ǎ, B, Č}, and are randomly classified into two groups. Note that ${\textcircled{\raisebox{-0.7pt}{\footnotesize{A}}}}{\text{--}}{\textcircled{\raisebox{-0.7pt}{\footnotesize{B}}}}{\text{--}}{\textcircled{\raisebox{-0.7pt}{\footnotesize{C}}}}$ is isomorphic to ${\textcircled{\raisebox{-0.7pt}{\footnotesize{C}}}}{\text{--}}{\textcircled{\raisebox{-0.7pt}{\footnotesize{B}}}}{\text{--}}{\textcircled{\raisebox{-0.7pt}{\footnotesize{A}}}}$ and this duplicate redundancy due to the graph isomorphism is removed. The response value $y$ of a graph is $-1$ if two subgraphs are selected from the same group, otherwise $+1$.
Table \[tbl:resultArti\] shows the performance results for the Graph-XOR data by two-fold cross validations. We use the proposed nonlinear method and two linear methods, namely, the proposed algorithm with maximum tree-depth ($d$) $=1$, i.e., with decision stumps, and a state-of-the-art (but linear) method for graphs, gBoost [@Saigo:2009]. The hyperparameter tuning is performed for the ranges described in Table \[tbl:paramForArti\], and the best parameters are also listed in Table \[tbl:resultArti\]. Figures \[fig:artiByd\] and \[fig:artiByx\] show the accuracy and loss changes for the test data with regard to the max tree depth ($d$) and the max subgraph size ($x$). Here “subgraph size” means the number of edges.
The results shown in Table \[tbl:resultArti\] clearly demonstrate that the linear models, including our model with $d=1$, fail, but the nonlinear methods work well. This is also theoretically supported by Theorem \[thm:linear\_nonlinear\]. Figure \[fig:artiByd\] also shows that only the behavior of $d=1$ differ from those of the other depths. Moreover, note that this problem at least requires subgraph features of size 2 (i.e., two edges), but searching excessively large subgraphs results in overfitting, as we see for $x$ $\geqslant 4$ in Figure \[fig:artiByx\].
![Subgraph groups[]{data-label="fig:parts"}](aaa1bbb1.eps "fig:"){width="80.00000%"}\
$y = +1$
![Subgraph groups[]{data-label="fig:parts"}](ccc2aab3.eps "fig:"){width="80.00000%"}\
$y = +1$
![Subgraph groups[]{data-label="fig:parts"}](abc3abc2.eps "fig:"){width="80.00000%"}\
$y = -1$
![Subgraph groups[]{data-label="fig:parts"}](aaa2bcc2.eps "fig:"){width="80.00000%"}\
$y = -1$
\
[ll]{} Group 1 & Group 2\
&\
&\
&\
&\
&\
&\
&\
&\
&\
[llcl]{}\
& $x$ & 2, 3, 4, 5, $\infty$\
\
\
Proposed & max tree depth & $d$ & 1, 2, 3, 4, 5\
& stepsize & $\eta$ & 1, 0.7, 0.4, 0.1, 0.01\
& \# trees & $k$ & 1-500\
gBoost & regularization & $\nu$ & ß[0.6, 0.5, 0.4, 0.3, ]{}[0.2, 0.1, 0.01]{}\
[lllll]{} &\
Proposed & Proposed ($d1$) & gBoost\
[[**100.0**]{}]{} & 64.3 & 70.0\
$x2~d2~\eta0.7~k221$ & $x6~d1~\eta0.7~k26$ & $x6~\nu0.01$\
![Test accuracy and loss with tree depth $d$[]{data-label="fig:artiByd"}](fig5_panel.eps){width="45.00000%"}
![Test accuracy and loss with subgraph size $x$[]{data-label="fig:artiByx"}](fig4_panel.eps){width="45.00000%"}
QSAR with Molecular Graphs
--------------------------
We also evaluate the performance based on the most typical benchmark for graph classification on real datasets: the quantitative structure-activity relationship (QSAR) results with molecular graphs. We select four binary-classification datasets (CPDB, Mutag, NCI1, NCI47) in Table \[tbl:dataset\]: two data (CPDB, Mutag) for mutagenicity tests and two data (NCI1, NCI47) for tumor growth inhibition tests from PubChem BioAssay[^1]. NCI1 and NCI47 are balanced by randomly sampling negatives of the same size as the positives in order to avoid imbalance difficulty in evaluation. All chemical structures are encoded as molecular graphs using RDKit[^2], and some structures in the raw data are removed by chemical sanitization[^3]. We simply apply a node labeling by the RDKit default atom invariants (edges not labeled), i.e., atom type, \# of non-H neighbors, \# of Hs, charge, isotope, and inRing properties. These default atom invariants use connectivity information similar to that used for the well-known ECFP family of fingerprints[@Rogers:2010]. See [@Kearnes2016] for more elaborate encodings.
Tables \[tbl:acc\] and \[tbl:bestParam\] show the performance results obtained by 10-fold cross validations using the same three methods used for the Graph-XOR cases with different hyperparameter settings in Table \[tbl:param\]. We can observe that nonlinear methods often outperform the linear methods. At the same time, we can also observe, in some cases, that the linear methods work fairly well for the real datasets. The real datasets would not have explicit classification rules compared to noiseless problems such as the Graph-XOR cases. Thus, it is necessary to tolerate some noises and ambiguity. Although they may seem limited, linear hypothesis classes are known to be very powerful in such cases, because they are quite stable estimators and the input features can themselves include nonlinear features of data as implied in Theorem \[thm:linear\_nonlinear\].
We also provide the normalized feature importance scores from GTB and the search space size in Figure \[fig:FInSS\] for the CPDB dataset. In Figure \[fig:FInSS\], [*searched*]{} corresponds to the searched subgraphs, and [*selected*]{} to the subgraph selected as internal nodes. This would also implies that (i) the proposed approach can provide information on selected relevant subgraph features and (ii) searches and uses only a portion of the entire search space.
[lccccc]{} Dataset & ß[Graph-]{}[XOR]{} & CPDB & Mutag & NCI1 & NCI47\
\# data & 1035 & 600 & 187 & 4252 & 4202\
\# nodes & 7 & 13.7 & 17.9 & 26.3 & 26.3\
\# edges & 6 & 14.2 & 19.7 & 28.4 & 28.4\
[llcl]{}\
& $x$ & 4, 6, 8\
\
\
Proposed & max tree depth & $d$ & 1, 3, 5\
& stepsize & $\eta$ & 1.0, 0.7, 0.4, 0.1\
& \# trees & $k$ & 1-500\
gBoost & regularization & $\nu$ & ß[0.6, 0.5, 0.4, 0.3,]{}[ 0.2, 0.1, 0.01]{}\
[lcccccccc]{} & CPDB & & Mutag & & NCI1 & & NCI47 & \
& ACC & AUC & ACC & AUC & ACC & AUC & ACC & AUC\
\
ß[Proposed]{}[ ]{} & & ß[84.5]{}[($\pm$3.6)]{} & ß[87.8]{}[($\pm$6.6)]{} & ß[91.6]{}[($\pm$6.3)]{} & & & &\
\
ß[Proposed (d1)]{}[ ]{} & & ß[83.9]{}[($\pm$3.3)]{} & ß[87.8]{}[($\pm$6.6)]{} & ß[91.6]{}[($\pm$6.3)]{} & ß[83.1]{}[($\pm$1.6)]{} & ß[89.8]{}[($\pm$1.3)]{} & ß[82.8]{}[($\pm$1.4)]{} & ß[88.9]{}[($\pm$1.1)]{}\
ß[gBoost]{}[ ]{} & ß[77.1]{}[($\pm$2.7)]{} & ß[73.6]{}[($\pm$4.9)]{} & & & ß[82.7]{}[($\pm$2.2)]{} & ß[83.9]{}[($\pm$2.2)]{} & ß[81.3]{}[($\pm$1.4)]{} & ß[81.8]{}[($\pm$2.6)]{}\
\
L1-LogReg [@takigawa:2017] & 78.3 & - & - & - & - & - & - & -\
MGK [@Saigo:2009] & 76.5 & 75.6 & 80.8 & 90.1 & - & - & - & -\
freqSVM [@Saigo:2009] & 77.8 & 84.5 & 80.8 & 90.6 & - & - & - & -\
gBoost [@Saigo:2009] & 78.8 & [[**85.4**]{}]{} & 85.2 & 92.6 & - & - & - & -\
WL shortest path [@Shrvashidze:2011] & - & - & 83.7 & - & 84.5 & - & - & -\
Random walk [@Shrvashidze:2011] & - & - & 80.7 & - & 64.3 & - & - & -\
Shortest path [@Shrvashidze:2011] & - & - & 87.2 & - & 73.4 & - & - & -\
[lllll]{} & CPDB & Mutag & NCI1 & NCI47\
Proposed & $x4~d5~\eta0.1~k120$ & $x4~d1~\eta1~k22 $ & $x4~d5~\eta0.1~k452$ & $x4~d3~\eta0.4~k308$\
Proposed(d1) & $x8~d1~\eta0.4~k128$ & $x4~d1~\eta1~k22 $ & $x4~d1~\eta0.4~k499$ & $x4~d1~\eta0.4~k499$\
gBoost & $x8~\nu0.5 $ & $x7~\nu0.1 $ & $x8~\nu0.3 $ & $x8~\nu0.4 $\
![Feature importance and search space for CPDB[]{data-label="fig:FInSS"}](fig6_panel.eps){width="50.00000%"}
Scalability comparison to Naïve approach {#sec:scalability}
----------------------------------------
As previously mentioned in Section \[sec:previousModel\], there exists a simple naïve two-step approach to obtain nonlinear models of subgraph indicators. Figure \[fig:scalability\] shows the scalability of this “enumerate & learn” approach by first enumerating all small-size subgraphs and applying general supervised learning to their indicators. The values in the figure are the average values to process each fold in 10-fold cross validation on a single PC with Pentium G4560 3.50GHz and 8GB memory. We enumerate all subgraphs with limited subgraph size[^4], and feed their indicator features to GradientBoostingClassifier with 100 trees (depth $\leqslant$ 5) of scikit-learn[^5]. The proposed method is also tested with the same setting (100 trees, d5). Since this case both use GTB and thus the performance is the same in principle up to implementation details (empirically both 0.75-0.77 for this setting), we focus on scalability comparisons using the fixed hyperparameters. Because the number of subgraph patterns to be enumerated increases exponentially, off-the-shelf packages such as scikit-learn cannot handle them at some point even when pattern enumeration can be done. In Figure \[fig:scalability\], we can observe pattern enumeration can be done for max subgraph size = 1 to 12 (green line, left), but the 2nd scikit-learn step fails for max subgraph size $\geqslant 10$ (green line, right). In this CPDB examples, the numbers of subgraphs, i.e., the dimensions of feature vectors, were 66336.1, 145903.7, 275422.3, 512904.1, 874540.0 for max subgraph size = 8, 9, 10, 11, 12, respectively, and scikit-learn was only feasible for max subgraph size up to 9. Note that we also need to solve a large number of subgraph isomorphism known to be NP-complete.
![Scalability comparison to naïve approach[]{data-label="fig:scalability"}](fig7_panel.eps){width="49.00000%"}
Conclusions
===========
In summary, we investigated nonlinear models with all possible subgraph indicators and provide a novel efficient algorithm to learn from the nonlinear hypothesis space. We demonstrated that this hypothesis space is identical to the (pseudo-Boolean) functions of these subgraph indicators, which are, in general, strictly larger than those of the linear models. This is also empirically confirmed through our Graph-XOR example. Although most existing studies focus only on real datasets, this would also promote interest in whether graph-theoretic classification problems can be approximated in a supervised learning manner. At the same time, the experimental results of the present study also strongly suggest that we need a nonlinear hypothesis space for the QSAR problems based on some real datasets, which would also support a standard cheminformatics approach of applying nonlinear models, such as random forests and neural networks, to 0-1 feature vectors, referred to as *molecular fingerprints*, by the existence of substructural features.
Since research on classification and regression trees originates from the problem of *automatic interaction detection* [@MorganSonquist:1963; @Kass:1975; @breiman84], our approach can provide insights on the question whether such higher-order interactions between input features exist. In this sense, our methods and findings would also be informative to consider a recent hot topic of detecting such interactions in combinatorial data [@terada2013statistical; @sugiyama2015significant; @nakagawa2015safe].
Proof of Theorem \[thm:bound\] {#sec:boundLinear}
==============================
Given ${\mathcal{D}}_1(g)$ and ${\mathcal{D}}_0(g)$, $$\begin{aligned}
\bound &= {\min_{\scalebox{1.0}{$g'$}}} \bigl[ {\mathrm{TSS}}({\mathcal{D}}_1(g')) + {\mathrm{TSS}}({\mathcal{D}}_0(g')) \bigr] \notag \\
\label{eq:boundSubset}
&= {\min_{\scalebox{1.0}{$S \subset {\mathcal{D}}_1(g)$}}} \bigl[ {\mathrm{TSS}}({\mathcal{D}}_1(g) \setminus S) + {\mathrm{TSS}}({\mathcal{D}}_0(g) \cup S) \bigr] \\
\label{eq:linearBound}
&= {\min_{\scalebox{1.0}{$(\diamond,k)$}}} \Big[ {\mathrm{TSS}}({\mathcal{D}}_1(g) \setminus S_{\diamond,k}) + {\mathrm{TSS}}({\mathcal{D}}_0(g) \cup S_{\diamond,k}) \Big]
\end{aligned}$$ where $ (\diamond, k) \in \{ \leq, > \} \times \{ 2, \dots, |{\mathcal{D}}_1(g) - 1| \} $. From the anti-monotone property , we have ${\mathcal{D}}_1(g') \subseteq {\mathcal{D}}_1(g)$ for $g' \sqsupseteq g$ for the training set ${\mathcal{D}}$ from which the equation directly follows. Thus, we show in detail. For simplicity, let $A =,\{ a_1, \dots, a_n \mid a_i \in \mathbb{R} \}$ denote ${\mathcal{D}}_1(g)$, and $B = \{ b_1, \dots, b_m \mid b_i \in \mathbb{R} \}$ denote ${\mathcal{D}}_0(g)$. Then, the goal of is to minimize the total sum of squares ${\mathrm{TSS}}(A \setminus S) + {\mathrm{TSS}}(B \cup S)$ by tweaking $S = \{ s_1, \dots, s_k \} \subset A$. Let ${\bar{a}}$, ${\bar{a}}_{-S}$, ${\bar{b}}$, and ${\bar{b}}_{+S}$ be the means of $A$, $A \setminus S$, $B$, and $B \cup S$, respectively. The key fact is that ${\mathrm{TSS}}(A \setminus S) + {\mathrm{TSS}}(B \cup S)$ can be regarded as a quadratic equation of $\sum_{i=1}^k s_i$ when the size of $S$ is fixed to $k$. More precisely, Therefore, ${\mathrm{TSS}}(A \setminus S) + {\mathrm{TSS}}(B \cup S)$ is minimized when $\sum_{i=1}^k s_i$ is maximized or minimized. In other words, becomes minimum when the mean of $S \subset
{\mathcal{D}}_1(g)$ is maximized or minimized.
[^1]: <https://pubchem.ncbi.nlm.nih.gov/bioassay/>$\langle\text{AID}\rangle$ (AID numbers are 1 and 47, respectively)
[^2]: <http://www.rdkit.org/>
[^3]: Due to this pre-processing, the number of datasets differs from that in the simple molecular graphs in the literature, where the nodes are labeled by atom type, and the edges are labeled by bond type.
[^4]: Small-size subgraphs are known to be more appropriate for this supervised-learning purpose than frequent subgraphs [@Wale:2008].
[^5]: <http://scikit-learn.org>
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Molecular Dynamics simulations of water confined in nanometer sized, hydrophobic channels show that water forms localized cavities for pore diameter $\gtrsim2.0\, nm$. The cavities present non-spherical shape and lay preferentially adjacent to the confining wall inducing a peculiar form to the liquid exposed surface. The regime of localized cavitation appears to be correlated with the formation of a vapor layer, as predicted by the Lum-Chandler-Weeks theory, implying partial filling of the pore.'
author:
- 'John Russo, Simone Melchionna, Francesco De Luca'
- Cinzia Casieri
title: 'Water confined in nanopores: spontaneous formation of microcavities'
---
Introduction
============
Nanotubes [@iijima] are important building blocks of nanocomposite materials and nanomachinery. When immersed in ionic solutions, nanometer-sized pores can be used for the detection and analysis of electrolytes and charged polymers, such as DNA [@kasianowicz], and to understand ion transport and selectivity in biological channels [@hille]. A full understanding of the phase behavior of confined water is a pre-requisite to interpret adsorption and conductivity data. In particular, experiments and simulations have shown that, for channels of sub-nanometer radius water exhibits an intermittent filling/drying transition resulting from an underlying bi-stable free-energy landscape separated by a thermally activated barrier, a behavior specific to water and not observed in a monoatomic fluid [@hummer; @allen]. The switch between empty/filled states upon ion translocation has been advocated as a gating mechanism in passive biological channels [@dzubiella; @sansom].
Cavitation of water under confinement is related to the long-ranged attractive forces exerted by hydrophobic bodies. Two alternative mechanisms have been proposed for explaining the occurrence of these forces: on one hand, the presence of air-filled nanobubbles at the solid surface drives the attraction via a bubble-driven mechanism; on the other hand, long-range correlations in the critical region or near the spinodal line could induce an extended density depletion at the surfaces which results in the attraction [@attard-patey-1; @attard-patey-2; @israelachvili-pnas].
The first line of reasoning relies on the pre-existence of air-filled cavities at the solid surface [@attard]. However, the mechanism for their stabilization is not fully understood since the Laplace-Young equation prescribes an internal pressure of the order of $10^{2}$ atm for nanometer-sized bubbles [@barrat-hansen]. One question pertains how such large internal pressure can sustain stable or metastable cavities, and if these are independently nucleated at the solid, corrugated surface. It has been suggested that line tension can act to stabilize such small cavities [@pompe], by reducing the curvature of the bubble base on the solid surface. As a matter of fact, micron-sized air bubbles are commonly observed in proximity of hydrophobic corrugated surfaces and, in the recent literature, there is growing evidence for the presence of bubbles on corrugated surfaces at the nanoscale. In particular, recent AFM experiments on silicon oxyde wafer surfaces of controlled roughness reported on bubble formation at the solid-water interface [@yang], while syncrotron x-ray reflectivity measurements reported on a small depletion layer in lieu of localized cavitation [@granick].
The second interpretation is based on the formation of a vapor layer in contact with the solid surface. Such depletion layer would involve a high entropic cost, growing with the extension of the exposed surface, but compensated by volume-dependent forces arising from molecular reorganization. The Lum-Chandler-Weeks (LCW) theory [@lcw] formulates the hydrophobic effect in microscopic terms based on the competition between interfacial and bulk forces which therefore depend on the solute surface/volume ratio. The molecular mechanism underlying the force balance is commonly ascribed to the distortion of the hydrogen bond network. In particular, for small spherical solutes water arranges in a clathrate structure, while for larger size the ensemble of hydrogen bond vectors, O-H$^{...}$H, points preferentially towards the solute. Therefore, if the curvature of the embedded body is low, the distortion is large and entropic effects prevail over the enthalpic ones, and vice versa for small solutes.
The cylindrical pore represents an opposite case to the spherical solute, where now the confining surface is concave. However, a similar enthalpic/entropic competition is expected and should depend on the pore diameter. Thus, a relevant information pertains if, for pore diameter larger than that characteristic of intermittent behavior, the liquid phase is completely stabilized or if some peculiar behavior still takes place. The occurrence of cavitation in water confined by cylindrical pores has been recently discussed in view of understanding the hysteresis involved in capillary evaporation [@barrat2].
We report here the results of Molecular Dynamics (MD) simulations undertaken on cylindrical, hydrophobic pores of finite length. In section II the simulation set-up and numerical details are described and in Section III the Molecular Dynamics data are illustrated. In Section IV the LCW theory is described and the numerical solutions are compared with the simulation results. The last Section draws some conclusions and analyzes recent experimental observations.
Simulation details
==================
The geometry of the simulation set-up is depicted in Fig. \[fig:setup\]. The two pore mouths are connected to the same, periodically folded, reservoir of quasi-bulk water. The pore diameter $d$ and pore length $L$ are varied to the following pairs of values $(d,L)=(1.5,4.0)$, $(2.0,4.0)$, $(2.5,4.0)$, $(2.5,8.0)$, $(3.0,4.0)$ and $(3.5,4.0)\, nm$, as used to label the runs. Simulations become rapidly costly with the pore size and we did not attempt to extend our simulations to larger geometries. Besides the nominal values, the effective cylinder diameter and length were evaluated via the criterion that the repulsive wall-oxygen potential be less than $k_{B}T$, giving a reduction of the effective diameter and increase of effective length by $0.46\, nm$ [@allen].
The simulation box is periodic in all three directions with dimensions depending on the pore geometry. The number of water molecules for all simulations ranges between $852$ and $2464$ units. Water is represented via the SPC/E computational model [@spc]. The confining oxygen-wall potential is modeled as a smooth surface generated by carbon atoms distributed uniformly over the surface. The wall-water interaction acts between each water oxygen and a smooth Lennard-Jones potential integrated over the dark region of the wall in Fig. \[fig:setup\]. Each carbon atom carries a Lennard-Jones potential with parameters $\sigma=0.345\, nm$ and $\epsilon=0.7294\, kJmol^{-1}$ [@allen] and standard Lorentz-Berthelot rules are used to construct oxygen-wall interactions. Electrostatic interactions between charged oxygen and hydrogen atoms are computed with the Ewald method via the Smooth Particle Mesh Ewald implementation [@darden]. Moreover, the confining medium is taken as non polarizable.
Special care is taken to keep the reservoir density under control, such that upon emptying/filling of the channel the reservoir maintains a bulk-like behavior with an average mass density of $1\, gr/cm^{3}$ away from the pore. This is obtained by varying over time the length of the reservoir, in direction parallel to the pore axis, by a Berendsen type of piston [@berendsen], with a characteristic coupling time of $10\, ps$, during the equilibration and production runs. The feedback sets the average density in a stripe of thickness $t=10\, nm$ placed at distance $l=7\, nm$ away from the pore (see Fig. \[fig:setup\]). The system temperature is controlled via a Nosè-Hoover thermostat [@frenkelsmit] in order to avoid anomalous temperature drifts during both the equilibration process and the subsequent production runs. The system is simulated at $300\, k$ while some auxiliary simulations are made at $280$ and $320\, K$. The evolution of the system is followed over times of $100\, ps$ (equilibration) and, subsequently, of $1\, ns$ (production). Within the simulation time window, stationarity is monitored by following the number of molecules populating the channel and the formation and size distribution of cavities. In order to monitor the effective stationarity of the equilibrated system, the run of the $(2.5,4.0)\, nm$ geometry has been further extended to $3\, ns$, without observing any departure from stationarity.
In order to detect the cavities we have used a coarse-graining procedure by tesselating the space with cubic cells of edge $0.02\, nm$. An empty cell is defined by having the distance from any oxygen atom greater than $0.3\, nm$. The wall position is defined by the largest radial distance of an oxygen atom from the pore axis and adding an offset of $0.1\, nm$. A cavity is defined as the cluster of continguous empty cells which do not belong to the wall region. The distinction among clusters is made via a graph algorithm and sorted according to size. In this way, the identity of cavities is clearly established.
Molecular Dynamics results
==========================
The simulation data present two distinct behaviors: for the geometries $(1.5,4.0)\, nm$ at $T=300\, K$ and $(2.5,4.0)\, nm$ at $T=280\, K$ we observe emptying of the pore, with a large vapor region extending between the two pore mouths; on the other hand, for the geometries $(2.0,4.0)$, $(2.5,4.0)$, $(2.5,8.0)$, $(3.0,4.0)$ and $(3.5,4.0)\, nm$, water persists inside the pore at density close to liquid state, with formation of nanobubbles in proximity of the confining wall.
At first we report on the process of emptying of the channel, as depicted in Fig. \[fig:longi-empty\]. Given an initial, apparently stable liquid inside the channel, a number of distinct lateral cavities are formed in proximity of the cylinder wall. Subsequently, one of such bubbles grows in size, rapidly extending over the whole pore volume. The final state appears to be stable over the $1\, ns$ time scale. The homogeneous growth of the vapor region is consistent with the recently proposed model of capillary evaporation in hydrophobic pores [@barrat2].
Intermittent emptying/filling of the channel has been previously reported at $T=300\, K$ for $d\lesssim1.5\, nm$ and $L\lesssim1.0\, nm$ (and eventually for larger diameters at lower temperatures) by using MD simulations with the same computational model as used here [@allen]. In the previous work, it was found that the time scale of emptying/filling oscillations is of the order of $\sim100\, ps$. In the current case, the direct observation of intermittency is prohibitive in terms of CPU time. In fact, given the larger pore diameters, intermittency could take place over timescales longer by one order of magnitude, or more [@sansom]. Therefore, we preferred to focus on the interfacial structuring of the liquid in the filled state.
It is interesting to note that the geometry $(2.4,4.0)\, nm$ exhibits pore emptying for $T=280\, K$ and pore filling for $T=300\, K$. The fact that emptying/filling of the pore is sensitive to differences in temperature as small as $\Delta T=20\, k$ is an indication that the underlying bulk phase diagram, with the rather narrow region of bulk liquid water and the close-by liquid-vapor coexistence, is an important driving factor for the confined fluid.
For the filled channel, the liquid interface displays the patterns shown in Figs. \[fig:longi-filled\] and \[fig:trasverse\]. The interface is rather corrugated by the presence of cavities appearing in proximity of the confining wall (lateral cavities), and rarely observed as spherical shape located around the cylinder axis. The cavities resemble spherical caps or rounded cones, with volumes much larger than the molecular size ($v_{w}\simeq0.027\, nm^{3}$). The average volumes range between $\langle v\rangle/v_{w}=5.7$ for the $(d,L)=(2.0,4.0)\, nm$ geometry and $\langle v\rangle/v_{w}=21.0$ for $(d,L)=(3.5,4.0)\, nm$. The peculiar pattern of the liquid exposed surface is rather representative of the highly cohesive character of water. It should be noticed that the shape of the vapor cavities along the transversal section does not appear to be flattened against the wall, while in the longitudinal direction visual analysis does not allow to draw a clear conclusion. We infer that line tension effects can be important only along the longitudinal direction but not on the transverse direction. In terms of dynamics, the cavities present high mobility with fast reshaping on the $1\, ps$ time scale.
The distribution of cavity size is illustrated in the normalized histograms of Fig. \[fig:vPv\]. A systematic increase of cavities of larger volume with the pore diameter is visible, accompanied by a depletion of smaller, molecular-sized ones. The distribution becomes more shifted to the right as the diameter of the cylinder grows. In contrast, water in proximity of a planar hydrophobic surface exhibits cavities with reduced volumes, basically indistinguishable from fluctuations of size of the molecular volume $v_{w}$.
In fig. \[fig:vPv\_reduced\] we report the same histograms but we filter out the contributions arising from cavities of volume $v<v_{w}$. By rescaling the abscissa by the cylinder surface area, the histograms become peaked around the same value $v/(v_{w}\pi dL)\simeq2\, nm^{-2}$, and width increasing with the pore diameter. However, the integrated cavity volume per unit surface is mostly contributed by the right tails of the distributions, which appear to be equal for the $(2.5,4.0)$ and $(3.5,4.0)$ geometries but significantly shifted to the left for the $(1.5,4.0)$ case. Therefore, cavitation does not seem to grow further as the diameter is larger than $2.5\, nm$.
The longer channel does not appear to affect the histogram, indicating that the bubble formation does not depend on interfacial effects arising from the finite pore length. In principle, the finite length could affect the liquid/vapor balance and the drying transition. In fact, by using an elementary macroscopic argument [@barrat-hansen], the difference in free-energy density between the liquid and vapor states is approximated by $\Delta\Omega/\pi dL=\gamma_{lw}-\gamma_{lv}d/2L$ where $\gamma_{lw}$ and $\gamma_{lv}$ are the liquid-wall and liquid-vapor surface tensions under ambient conditions, respectively, and bulk contributions to the free energy have been neglected. In other words, if finite channel effects are in place, by making the channel longer, one should move away from the vapor branch. By taking $\gamma_{lw}=7.4\, k_{B}T/nm^{2}$ and $\gamma_{lv}=26\, k_{B}T/nm^{2}$ it is seen that the geometry $(d,L)=(2.5,4.0)\, nm$ is close to the region where the free-energy difference changes sign [@allen]. Our simulation at room temperature for $(d,L)=(2.5,8.0)\, nm$ showed that the channel remains filled as much as the $(2.5,4.0)\, nm$ geometry, without appreciable differences in the distribution of cavity volumes. Conversely, the sensitivity to the pore length appears in the fluctuations of the number of adsorbed water molecules, which approximately drops by a factor two in the longer channel. We interpret the weak size dependence of cavitation as due to the corrugated, largely exposed surface of the liquid with respect to the simple macroscopic model, that renders finite size effects negligible already at $L=4\, nm$.
Lum-Chandler-Weeks predictions
===============================
In this section we compare our data to the predictions of the LCW theory for an infinitely long pore. Schematically, the theory can described as follows. Let us first consider the decomposition of the microscopic density $n(r)$ into a slowly varying component, $n_{s}(r)$ , and a fast component, $n(r)-n_{s}(r)$. The LCW theory builds upon the well-known square-gradient theory for liquid-vapor coexistence [@hans] by taking into account self-consistently the fast oscillations in density. According to the square gradient approximation, the local grand potential $-W(n)$ is related to the laplacian of the density via$$\left(\frac{dW}{dn}\right)=m\nabla^{2}n(r)\label{eq:squaregradient}$$ where $m$ is an effective parameter derived from the underlying interatomic potential $v(r)$, $m=-\frac{\pi}{6}\int_{0}^{\infty}drr^{4}v(r)$ (Random Phase Approximation [@hans]). By applying a coarse graining procedure, the local density is replaced by the function $n\rightarrow\bar{n}=n+\frac{\lambda^{2}}{2}\nabla^{2}n$ where $\lambda$ is the characteristic length of the coarse graining procedure. Therefore, the starting equation of the LCW theory is derived from eq.(\[eq:squaregradient\]) rewritten in terms of $\bar{n}$, and asserting that this equations holds for the slow component $n_{s}$, thus
$$\left(\frac{dW}{dn}\right)_{n_{s}}=\frac{2m}{\lambda^{2}}[\bar{n}-n_{s}]\label{eq:LCW2}$$
for which we choose $\lambda=0.3\, nm$ and $m=919.9\times10^{21}\, Jnm^{5}mol^{-2}$. To implement eq. (\[eq:LCW2\]) one needs prior knowledge of the bulk equation of state of water near the liquid-vapor coexistence. Following previous studies [@huang], we used a quartic dependence plus a linear correction term which models the proximity to coexistence,$$W(n)=\alpha\left(n-n_{l}\right)^{2}\left(n-n_{g}\right)^{2}+\beta(n-n_{l})\label{eq:free-en}$$ where $\alpha=2.89\times10^{-24}\, Jnm^{9}$, $\beta=3.04\times10^{-29}J$ and the liquid and vapor densities are $n_{l}=32.94\, nm^{-3}$ and $n_{g}=7.7\times10^{-4}\, nm^{-3}$, respectively [@huang]. By solving the equations in cylindrical coordinates and forbidding any angular symmetry breaking, we did not attempt to observe the formation of cavities of given shape. This choice was motivated by the narrow range of numerical stability found during the solution of the LCW equations.
The link between the slowly varying and the complete density fields is established within the gaussian density fluctuations approximation, by writing [@chandler]$$n({\bf r})=n_{s}({\bf r})-\int d{\bf r}'c({\bf r}')\chi({\bf r},{\bf r}')\label{eq:fredholm}$$ where the response function $\chi({\bf r},{\bf r}')\equiv\langle\delta n({\bf r})\delta n({\bf r}')\rangle$ is approximated by $\chi({\bf r},{\bf r}')\simeq n_{s}({\bf r})\delta({\bf r}-{\bf r}')+n_{s}({\bf r})n_{s}({\bf r}')h(|{\bf r}-{\bf r}'|)$ and $h(|{\bf r}-{\bf r}'|)-1$ is the bulk pair correlation function. Moreover, $c({\bf r})$ is the water-wall direct correlation function. Given $n_{s}({\bf r})$, eq.(\[eq:fredholm\]) is solved to obtain $c({\bf r})$ and $n({\bf r})$ via the Nystrom numerical procedure [@press]. The two coupled integro-differential equations (\[eq:LCW2\]) and (\[eq:fredholm\]) are solved iteratively. We did not include the correction to the LCW theory due to the attractive part of the wall-oxygen potential [@huang]. Vice versa, the effective diameter of MD and the diameter used in LCW calculations, $\bar{d}$, were considered as equivalent control parameters.
We have found that for diameter $\bar{d}<2.6\, nm$ the LCW equations predict complete pore emptying, in agreement with previous simulation and experimental results, but below our MD data which exhibit emptying (or eventually intermittency) for effective diameter $\lesssim2.0\, nm$. However, in this geometry the LCW predictions do not distinguish between intermittent and cavitating behavior. In the range $2.6<\bar{d}<3.8\, nm$ the theory predicts partial filling of the pore, as illustrated by the density profiles in Fig. 4, where the micro-phase coexistence is accompanied by a significant density depletion near the wall. Moreover, the LCW equations predict a gradual increase of the vapor layer as the diameter lowers, whilst the MD results show a maximum which varies slowly with the pore diameter. Following previous authors [@huang], we attribute this difference to the presence of the weak attractive tail in the water-wall potential. Finally, at larger diameters, the LCW curves show weakly structured profiles, with a characteristic non-wetting shape near contact. In essence, the LCW theory predicts a cross-over between empty and partially filled states for $\bar{d}=2.6\, nm$, larger than the MD results where the cross-over appears in the $1.5:2.0\, nm$ interval. This discrepancy may be attributed to the one-dimensional solution of the LCW equations and to the missing attractive tail in our theoretical treatment, causes which artificially stabilize the vapor phase, or to the undetermined location of the confining surface, whose diameter can vary by about $0.5\, nm$.
Notwithstanding the negligible correlations emerging from the LCW solutions, we can compare MD and LCW data regarding the number of water molecules filling a channel of length $4\, nm$ as a function of the channel effective diameter $\bar{d}$. From Fig. \[fig:Water-count\] it is apparent that the LCW model shows systematically lower values than the MD data in the region where the theory exhibits partial filling. The MD data are intermediate between the LCW values and the simple model of uniform filling of the pore at the water liquid density. As for the discrepancy in the cross-over diameter, the lack of longitudinal symmetry breaking and the weak attractive tail in the simulated water-wall potential, might explain the smaller number of water molecules predicted by the LCW theory.
Conclusions
===========
The behavior of water in contact with an extended hydrophobic surface has attracted considerable attention on account of the intrinsic thermodynamic problem, which has significant implications for protein folding [@huang], for ionic transport in biological channels [@hille] and for the boundary conditions for fluid flow in microchannels [@degennes].
Our study clearly demonstrates the spontaneous formation of vapor microcavities on the nanometer scale for water in contact with a concave cylindrical surface of diameter $d\gtrsim2.0\, nm$. The cavities are localized both in the angular and, more importantly, in the longitudinal directions. To our knowledge, this surprising result is the first observation for water confined in such geometry.
The results are particularly interesting for the on-going debate on the hydrophobic effect, in which a number of different interpretations have been put forward, such as entropic effects due to molecular rearrangement, electrostatic effects and spontaneous cavitation due to the metastability of the fluid [@israelachvili-pnas]. Recent experiments have focused on analyzing water in contact with a hydrophobic surface and found contrasting results. In particular, either cavitation [@yang; @ishida] or an extended low-density depletion layer [@granick] have been observed. The fact that water around a concave, but smooth, hydrophobic surface forms short-ranged islands of vapor, modulated by the surface curvature, suggests that the contrasting experimental observations arise from the sensitive structural response of water to the roughness of the solid surface.
Moreover, our study sheds some light on the experimental observation of hysteresis in water intrusion/extrusion cycles in pores and the interpretation based on homogenous nucleation [@barrat2]. The simulation data underline the metastable character of highly confined water. However, the rich structural patterns exhibited by the interface are absent in a simple fluid and, therefore, are unlikely to be explained in terms of a macroscopic approach. When comparing the simulation data with a quasi-microscopic treatment, namely the Lum-Chandler-Weeks theory, the density profiles and the cross-over diameter between empty and partially filled states of the latter were found to agree only qualitatively with the simulation data. The discrepancies were explained on the basis of the one-dimensional solution of the LCW equations.
Finally, we wish to comment on some recent measurements on ionic conductance in nanopores of diameter $d\simeq10\, nm$ showing that transport displays an anomalous response, with a five orders of magnitude reduction in the current spectral density power and a strongly noisy response with respect to bulk behavior [@dekker]. This has been attributed to the presence of spherically shaped air bubbles trapped inside the nanopores such that the translocating ions encounter a two-phase filled region. Although the differences in pore diameter between the presently simulated and the experimental systems might seem large, one might expect that water cavitation is still effective in the wider pores or that cavitation is locally enhanced by the translocating ion. The effect of such cavitation would induce non trivial wall-ion dielectric interactions, modulated by the imperfect screening of water, and non trivial hydrodynamic forces. Such scenario would imply a coupled ion-bubble transport mediated by the microscopic liquid-vapor coexistence.
Acknowledgements
================
We wish to thank Jean-Pierre Hansen and Francesco Sciortino for critical reading of the manuscript and Mauro Chinappi for help with the simulation setup.
Figure Captions {#figure-captions .unnumbered}
===============
[10]{} S. Iijima, Nature, 56, 354 (1991).
J.J. Kasianowicz, E. Brandin, D. Branton, D.W. Deamer, *Proc. Natl. Acad. Sci. U.S.A.,* **93**, 13770, (1996).
B.Hille, *Ionic channels of excitable membranes* (Sinauer Associated, Sunderland, MD, 1992).
G. Hummer, J.C. Rasaiah, J.P. Nowortya, *Nature (London)*, **414**, 188 (2001).
R. Allen, S. Melchionna, J-P. Hansen, *Phys.Rev.Lett.*, **89**, 175502 (2002). *J.Chem. Phys.,* **120**, 4927 (2004).
J. Dzubiella, J.-P. Hansen, *J.Chem.Phys*. **122**, 234706 (2005).
O. Beckstein, K. Tai, and M. S. P. Sansom, *J. Am. Chem. Soc.,* **126**, 14694 (2004).
P. Attard, C.P. Ursenbach, G.N. Patey, *Phys. Rev.A*, **45**, 7621 (1992).
D.R. Bérard, P. Attard, G.N. Patey, *J. Phys. Chem.*, **98**, 7236 (1992).
E.E. Meyer, K.J. Rosenberg, J. Israelachvili, *Proc. Natl. Acad. Sci.* USA, 10.1073/pnas.0606422103 (2006).
P. Attard, *Adv. Colloid Interface Sci.*, **104**, 75 (2003).
J.-L. Barrat, J.-P. Hansen, *Basic concepts for simple and complex liquids*, Cambridge University Press (2003).
T. Pompe, S. Herminghaus, *Phys.Rev.Lett.* **85**, 1930 (2000).
J. Yang, H. Duan, D. Fornasiero, J. Ralston, *J.Phys.Chem.B* **107**, 6139 (2003).
N. Ishida, R. Inoue, M. Miyahara, K. Higashitani, *Langmuir*, **16**, 6377 (2000).
A. Poynor, L. Hong, I.K. Robinson, S. Granick, Z.Zhang, P.A. Fenter, *Phys.Rev.Lett.* **97**, 266101 (2006).
K. Lum, D. Chandler, J.D. Weeks, *J. Phys. Chem. B*, **103**, 4570 (1999).
B. Lefevre, A. Saugey, J.-P. Barrat, L. Bocquet, E. Charlaix, P.F. Gobin, and G. Vigier, *J.Chem.Phys.*, **120**, 4927 (2004).
U. Essmann, L. Perera, M.L. Berkowitz, T. Darden, H. Lee, L.G. Pedersen. *J.Chem.Phys.* **103**, 8577 (1995).
H.J.C. Berendsen, J.P.M. Postma, W.F. van Gunsteren, A. Di Nola, J.R. Haak, *J.Chem.Phys.*, **81**, 3684, (1984).
D. Frenkel and B. Smit, *Understanding molecular simulation* (Academic Press, London, 1996).
W.L. Jorgensen, J. Chandrasekhar, J.D. Madura, R.W. Impey, M.L. Klein, *J. Chem. Phys.* **79**, 926 (1983).
D.M. Huang, D. Chandler, *J. Phys. Chem. B*, **106**, 2047 (2002)
D. Chandler, *Phys.Rev. E*, **48**, 2898 (1993)
J.P. Hansen and I.R. McDonald, *Theory of simple liquids*, 3rd edition, Academic Press (2006).
W.H. Press et al., *Numerical Recipes in C: the art of scientific computing*, Cambridge University Press.
W. Humphrey, A. Dalke, K. Schulten, *J. Molec. Graphics*, **14**, pp. 33 (1996).
P.-G. de Gennes, *Langmuir* **18**, 3413 (2002).
R.M.M. Smeets, U.F. Keyser, M.Y. Wu, N.H. Dekker, C. Dekker, *Phys.Rev.Lett.*, **97**, 088101 (2006).
{width="0.7\columnwidth"}
Figure 1, Russo et al.
{width="0.7\columnwidth"}
Figure 2, Russo et al.
{width="0.7\columnwidth"}
Figure 3, Russo et al.
{width="0.8\columnwidth"}
Figure 4, Russo et al.
{width="0.8\columnwidth"}
Figure 5, Russo et al.
{width="0.8\columnwidth"}
Figure 6, Russo et al.
{width="0.9\columnwidth"}
Figure 7, Russo et al.
{width="0.9\columnwidth"}
Figure 8, Russo et al.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
Le problème des arcs de Nash pour les singularités normales de surfaces affirme qu’il y aurait autant de familles d’arcs sur un germe de surface singulier $(S,O)$ que de diviseurs essentiels sur $(S,O)$. Il est connu que ce problème se réduit à étudier les singularités quasi-rationnelles. L’objet de cet article est de répondre positivement au problème de Nash pour une famille d’hypersurfaces quasi-rationnelles non rationnelles. On applique la même méthode pour répondre positivement à ce problème dans les cas de singularités de type ${\ensuremath{\mathbb{E}}}_6$ et ${\ensuremath{\mathbb{E}}}_7$ et pour fournir une nouvelle preuve dans le cas de singularités de type ${\ensuremath{\mathbb{D}}}_n$, $n\geq 4$.
0.5
[A[BSTRACT]{}.]{} [**A solution to the Nash problem on arcs for a family of quasi-rational hypersurfaces.**]{} The Nash problem on arcs for normal surface singularities states that there are as many arc families on a germ $(S,O)$ of a singular surface as there are essential divisors over $(S,O)$. It is known that this problem can be reduced to the study of quasi-rational singularities. In this paper we give a positive answer to the Nash problem for a family of non-rational quasi-rational hypersurfaces. The same method is applied to answer positively to this problem in the case of ${\ensuremath{\mathbb{E}}}_6$ and ${\ensuremath{\mathbb{E}}}_7$ type singularities, and to provide new proof in the case of ${\ensuremath{\mathbb{D}}}_n$, $n\geq 4$, type singularities.
address: 'Université Grenoble I, Institut Fourier, UMR 5582 CNRS-UJF, BP 74, 38402 St. Martin d’Hères cédex, France'
author:
- 'Maximiliano LEYTON-ALVAREZ'
bibliography:
- 'Nash\_toulouse.bib'
title: 'Résolution du problème des arcs de Nash pour une famille d’hypersurfaces quasi-rationnelles'
---
[^1]
Introduction {#sec:int}
============
Soient ${\ensuremath{\mathbf{k}}}$ un corps algébriquement clos de caractéristique nulle, $V$ une variété algébrique normale sur ${\ensuremath{\mathbf{k}}}$ et $\pi:X\rightarrow V$ une désingularisation divisorielle de $V$, c’est-à-dire $\pi$ est une désingularisation ($\pi$ est un morphisme propre et birationnel tel que le morphisme $\pi: X\backslash \pi^{-1}({\ensuremath{\operatorname{Sing}}}V)\rightarrow V\backslash {\ensuremath{\operatorname{Sing}}}V$ est un isomorphisme, où ${\ensuremath{\operatorname{Sing}}}V$ est le lieu singulier de $V$ et $X$ est lisse) telle que les composantes irréductibles de la fibre exceptionnelle $\pi^{-1}({\ensuremath{\operatorname{Sing}}}V)$ sont de codimension $1$ dans $X$. On sait, d’après le théorème de résolution des singularités d’Hironaka, que cette désingularisation existe.
Si $\pi':X'\rightarrow V$ est une autre désingularisation de $V$, alors $(\pi')^{-1}\circ\pi:X\dashrightarrow X'$ est une application birationnelle. Soit ${\ensuremath{\operatorname{E}}}$ une composante irréductible de la fibre exceptionnelle de $\pi$. Comme ${\ensuremath{\operatorname{E}}}$ est un diviseur et $X$ est une variété algébrique normale ($X$ est une variété lisse), il existe un ouvert ${\ensuremath{\operatorname{E}}}^0$ de ${\ensuremath{\operatorname{E}}}$ sur lequel l’application birationnelle $(\pi')^{-1}\circ\pi$ est bien définie. Le diviseur ${\ensuremath{\operatorname{E}}}$ est appelé [*diviseur essentiel sur*]{} $V$ si pour toute désingularisation $\pi':X'\rightarrow V$ de $V$ l’adhérence de $(\pi')^{-1}\circ\pi(E^0)$ dans $X'$ est une composante irréductible de la fibre exceptionnelle du morphisme $\pi'$. On note ${\ensuremath{\operatorname{Ess}}}(V)$ l’ensemble de diviseurs essentiels sur $V$.
On remarque que, si $V$ est une surface algébrique normale sur ${\ensuremath{\mathbf{k}}}$, les diviseurs essentiels sur $V$ sont les composantes irréductibles de la fibre exceptionnelle de la résolution minimale de $V$.\
Soit $K$ un corps d’extension de ${\ensuremath{\mathbf{k}}}$. Un morphisme ${\ensuremath{\operatorname{Spec}}}K[t]/(t^{m+1}) \rightarrow V$ (resp. ${\ensuremath{\operatorname{Spec}}}K[[t]]\rightarrow V$) est appelé $(K,m)$-jet (resp. $K$-arc).\
Soit ${\mathcal{S}ch}/{\ensuremath{\mathbf{k}}}$ la catégorie des schémas sur ${\ensuremath{\mathbf{k}}}$ et ${\mathcal{E}ns}$ la catégorie des ensembles. On considère le foncteur contravariant suivant:
$F_m:{\mathcal{S}ch}/{\ensuremath{\mathbf{k}}}\longrightarrow {\mathcal{E}ns}$, $Y \mapsto {\ensuremath{\operatorname{Hom}}}(Y\times_{{\ensuremath{\mathbf{k}}}}{\ensuremath{\operatorname{Spec}}}{\ensuremath{\mathbf{k}}}[t]/(t^{m+1}),V).$
Ce foncteur est représentable de forme canonique par un schéma $V_m$ de type fini sur ${\ensuremath{\mathbf{k}}}$, c’est-à-dire on a:
${\ensuremath{\operatorname{Hom}}}(Y\times_{{\ensuremath{\mathbf{k}}}} {\ensuremath{\operatorname{Spec}}}{\ensuremath{\mathbf{k}}}[t]/(t^{m+1}),V)\cong {\ensuremath{\operatorname{Hom}}}(Y,V_{m}),$
où $Y$ est un schéma quelconque sur ${\ensuremath{\mathbf{k}}}$. Le schéma $V_m$ est appelé [*l’espace de m-jet sur*]{} $V$.
L’homomorphisme surjectif canonique ${\ensuremath{\mathbf{k}}}[t]/(t^{m+1})\rightarrow {\ensuremath{\mathbf{k}}}[t]/(t^{m})$ induit un morphisme affine ${\ensuremath{\operatorname{p}}}_m: V_m\rightarrow V_{m-1}$. Les morphismes ${\ensuremath{\operatorname{p}}}_{m\;n}:V_m\rightarrow V_n$, où $n<m$ et ${\ensuremath{\operatorname{p}}}_{m\;n}:={\ensuremath{\operatorname{p}}}_{n+1}\circ\cdots\circ {\ensuremath{\operatorname{p}}}_m$, forment un système projectif. Comme les morphismes ${\ensuremath{\operatorname{p}}}_m: V_m\rightarrow V_{m-1}$ sont affines (voir [@Ish07] ou [@EiMu09]), la limite projective existe. On note $V_{\infty}$ cette limite projective, c’est-à-dire $V_{\infty}:= \displaystyle \lim_{\leftarrow m} V_m$. La limite projective $V_{\infty}$ est un schéma qui n’est pas en général de type fini sur ${\ensuremath{\mathbf{k}}}$. Le schéma $V_{\infty}$ est appelé [*l’espace d’arcs sur* ]{} $V$.\
L’espace d’arcs $V_{\infty}$ a [*la propriété fonctorielle*]{} suivante (voir [@IsKo03]):Le foncteur $Y\rightarrow {\ensuremath{\operatorname{Hom}}}(Y\widehat{\times}_{{\ensuremath{\mathbf{k}}}}{\ensuremath{\operatorname{Spec}}}{\ensuremath{\mathbf{k}}}[[t]], V)$, où $Y$ est un schéma quelconque sur ${\ensuremath{\mathbf{k}}}$ et $Y\widehat{\times}_{{\ensuremath{\mathbf{k}}}}{\ensuremath{\operatorname{Spec}}}{\ensuremath{\mathbf{k}}}[[t]]$ est le complété formel du schéma $ Y\times_{{\ensuremath{\mathbf{k}}}}{\ensuremath{\operatorname{Spec}}}{\ensuremath{\mathbf{k}}}[[t]] $ le long du sous-schéma $ Y\times_{{\ensuremath{\mathbf{k}}}}{\ensuremath{\operatorname{Spec}}}{{\ensuremath{\mathbf{k}}}} $, est représentable par le schéma $V_{\infty}$.\
D’après la propriété fonctorielle de l’espace d’arcs $V_{\infty}$, les $K$-points de $V_{\infty}$ sont en correspondance bijective avec les $K$-arcs sur $V$. Par abus de notation, pour $\alpha\in V_{\infty}$, on note $\alpha$ son ${\ensuremath{\mathbf{k}}}_{\alpha}$-arc correspondant, où ${\ensuremath{\mathbf{k}}}_{\alpha}$ est le corps résiduel du point $\alpha$.\
Soit ${\ensuremath{\operatorname{p}_{\infty}}}:V_{\infty}\rightarrow V$ la projection canonique $\alpha\mapsto \alpha(0)$, où $0$ est le point fermé de ${\ensuremath{\operatorname{Spec}}}{\ensuremath{\mathbf{k}}}_{\alpha}[[t]]$. On note $V_{\infty}^{s}:= {\ensuremath{\operatorname{p}_{\infty}}}^{-1}({\ensuremath{\operatorname{Sing}}}V)$, où ${\ensuremath{\operatorname{Sing}}}V$ est le lieu singulier de $V$, et $\mathcal{CN}(V)$ l’ensemble de composantes irréductibles de $V_{\infty}^{s}$.\
Nash a démontré que l’application suivante est bien définie et injective (voir [@Nas95]):
$\mathcal{N}_{V}:\mathcal{CN}(V)\rightarrow {\ensuremath{\operatorname{Ess}}}(V), C\mapsto \overline{\{\widehat{\alpha}(0)\}}$,
où le ${\ensuremath{\mathbf{k}}}_{\alpha}$-arc $\alpha$ est le point générique de $C\in \mathcal{CN}(V)$, le ${\ensuremath{\mathbf{k}}}_{\alpha}$-arc $\widehat{\alpha}$ est le relèvement à $X$ du ${\ensuremath{\mathbf{k}}}_{\alpha}$-arc $\alpha$, c’est-à-dire $\widehat{\alpha}$ est l’unique morphisme tel que $\pi\circ\widehat{\alpha}=\alpha$, et $\overline{\{\widehat{\alpha}(0)\}}$ l’adhérence du point $\widehat{\alpha}(0)$. Cette application est appelée [*l’application de Nash associée à*]{} $V$.\
Le problème de Nash consiste à étudier la surjectivité de l’application de Nash $\mathcal{N}_{V}$.\
Dans plusieurs cas la surjectivité de cette application a été prouvée. Par exemple pour les singularités ${\ensuremath{\mathbb{A}}}_n$ ([@Nas95]), les singularités ${\ensuremath{\mathbb{D}}}_n$ ([@Ple08]), les surfaces sandwichs ([@LeRe99]), les variétés toriques ([@IsKo03]), les hypersurfaces quasi-ordinaires ([@Gon07]), les variétés toriques stables ([@Pet09]), et d’autres cas qu’on peut trouver dans les articles suivants ([@Ish05], [@Ish06], [@Gon07], [@Gole97], [@Lej80],[@Mor08], [@PlPo06], [@PlPo08], [@Reg95]).\
Dans l’article [@IsKo03] Ishii et Kollar montrent que l’application de Nash associée à l’hypersurface de ${\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^5$ donnée par l’équation $x_1^3+x_2^3+x_3^3+x_4^3+x_5^6=0$ n’est pas surjective. Or, en dimension deux ou trois il n’y a pas d’exemple publié où l’application de Nash n’est pas surjective.\
Soit ${\ensuremath{\operatorname{E}}}$ un diviseur essentiel sur $V$ (${\ensuremath{\operatorname{E}}}\in {\ensuremath{\operatorname{Ess}}}(V)$). On note $N_{{\ensuremath{\operatorname{E}}}}$ l’adhérence dans $V_{\infty}$ de l’ensemble suivant:
$\{\alpha\in V_{\infty}\backslash ({\ensuremath{\operatorname{Sing}}}V)_{\infty}\mid \widehat{\alpha}(0)\in E\}$,
où $({\ensuremath{\operatorname{Sing}}}V)_{\infty}$ est le sous-ensemble fermé de $V_{\infty}$ des arcs qui sont concentrés en ${\ensuremath{\operatorname{Sing}}}V$. L’ensemble $N_{{\ensuremath{\operatorname{E}}}}$ est irréductible et on a:
$V_{\infty}^{s}=\bigcup_{{\ensuremath{\operatorname{E}}}}N_{{\ensuremath{\operatorname{E}}}}$,
voir [@Reg06]. On note $\alpha_{{\ensuremath{\operatorname{E}}}}$ le point générique de $N_{{\ensuremath{\operatorname{E}}}}$.\
Un morphisme $\omega:{\ensuremath{\operatorname{Spec}}}K[[s,t]]\rightarrow V$ est appelé $K$-[*wedge sur*]{} $V$. D’après la propriété fonctorielle de l’espace d’arcs $V_{\infty}$, les $K$-wedges sont en correspondance bijective avec les $K[[s]]$-points de $V_{\infty}$. L’image du point fermé (resp. du point générique) de ${\ensuremath{\operatorname{Spec}}}K[[s]]$ dans $V_{\infty}$ est appelé le centre (resp. l’arc générique) du $K$-wedge $\omega$.\
Un $K$-wedge $\omega$ est appelé [*$K$-wedge admissible centré en*]{} $N_{{\ensuremath{\operatorname{E}}}}$ si le centre (resp. l’arc générique) de $\omega$ est le point générique de $N_{{\ensuremath{\operatorname{E}}}}$ (resp. appartient à $V_{\infty}^{s}$). Dans ce cas, par définition, le corps $K$ est forcément un corps d’extension du corps résiduel ${\ensuremath{\mathbf{k}}}_{\alpha_{{\ensuremath{\operatorname{E}}}}}$ du point $\alpha_{{\ensuremath{\operatorname{E}}}}$. Le $K$-wedge $\omega$ peut être interprété comme une déformation à un paramètre des coefficients du comorphisme $\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}$ de l’arc $\alpha_{{\ensuremath{\operatorname{E}}}}$.
Dans [@Reg06] (dans le cas de surfaces voir [@Lej80]) l’auteur montre que ${\ensuremath{\operatorname{E}}}$ appartient à l’image de l’application de Nash $\mathcal{N}_V$ si et seulement si tout $K$-wedge admissible $\omega$ centré en $N_{{\ensuremath{\operatorname{E}}}}$ se relève à $X$, où $K$ est un corps d’extension quelconque du corps $K_{\alpha_{{\ensuremath{\operatorname{E}}}}}$, c’est-à-dire s’il existe un $K$-wedge $\widehat{\omega}$ sur $X$ tel que $\pi\circ \ \widehat{\omega}=\omega$.\
On considère l’hypersurface ${\ensuremath{\operatorname{S(p,h_{q})}}}$ de ${\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ donnée par l’équation $z^p+{\ensuremath{\operatorname{h_q}( x,y)}}=0$, où ${\ensuremath{\operatorname{h_q}}}$ est un polynôme homogène de degré $q$ sans facteurs multiples, $p\geq 2$, $q\geq 2$ deux entiers premiers entre eux. Par exemple si $
{\ensuremath{\operatorname{h_q}( x,y)}}=x^q+y^q$, ${\ensuremath{\operatorname{S(p,h_{q})}}}$ est une surface de [*Pham-Brieskorn*]{}. Les surfaces ${\ensuremath{\operatorname{S(p,h_{q})}}}$ sont toutes quasi-rationnelles (voir [@FlZa03]) et rationnelles si et seulement si $q=2$ ou $(p,q)=(2,3)$. On remarque que ${\ensuremath{\operatorname{S(p,h_{2})}}}$ est une singularité du type ${\ensuremath{\mathbb{A}}}_{p-1}$. Le résultat principal de cet article est le théorème suivant.
\[th:nashBP\] Pour tous les entiers $p\geq 2$, $q\geq 2$ premiers entre eux, l’application de Nash $\mathcal{N}_{{\ensuremath{\operatorname{S(p,h_{q})}}}}$ associée à ${\ensuremath{\operatorname{S(p,h_{q})}}}$ est bijective.
Une note sur la preuve de ce résultat a été publiée dans [@Ley11].\
On remarque que les critères de [@Mor08] et [@PlPo06] ne s’appliquent pas en général aux familles ${\ensuremath{\operatorname{S(p,h_{q})}}}$.\
En utilisant la même méthode que dans la preuve du Théorème \[th:nashBP\] on obtient le résultat suivant:
\[th:nashDP\] Si $S$ est une singularité du type ${\ensuremath{\mathbb{E}}}_{6}$ ou ${\ensuremath{\mathbb{E}}}_{7}$, l’application de Nash $\mathcal{N}_{S}$ associée à $S$ est bijective.
On obtient aussi une preuve simple de la bijectivité de l’application de Nash pour plusieurs cas connus, par exemple ${\ensuremath{\mathbb{D}}}_{n}$, où $n$ est un entier, $n\geq 4$ (voir [@Ple08]).\
Récemment sur ArXiv une preuve de la bijectivité de l’application de Nash pour la singularité de type ${\ensuremath{\mathbb{E}}}_6$ (resp. pour les surfaces quotients) a été publiée, voir [@PlSp10] (resp. voir [@Pe10])). Mais les méthodes de démonstration sont différentes de celle que nous utilisons.\
Cet article est organisé de la façon suivante: dans première section on prouve quelques résultats qui vont être utilisés dans toutes les sections de cet article et on démontre le Théorème \[th:nashBP\]; le Théorème \[th:nashDP\] est le sujet de la deuxième section, on donne la preuve du cas ${\ensuremath{\mathbb{E}}}_6$ en détails et un résumé pour le cas ${\ensuremath{\mathbb{E}}}_7$; dans la dernière section on donne une preuve simple de la bijectivité de l’application de Nash pour les singularités du type ${\ensuremath{\mathbb{D}}}_n$, $n\geq 4$ différente de celle de l’article [@Ple08].
Remerciements {#remerciements .unnumbered}
=============
Je tiens à remercier Gérard Gonzalez-Sprinberg pour les nombreuses discussions et son constant encouragement. Je remercie également Marcel Morales pour l’intérêt qu’il a porté à mon travail.
Le problème Nash pour les hypersurfaces quasi-rationelles ${\ensuremath{\operatorname{S(p,h_{q})}}}$
====================================================================================================
On rappelle que ${\ensuremath{\operatorname{S(p,h_{q})}}}$ est l’hypersurface de ${\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ donnée par l’équation $z^p+{\ensuremath{\operatorname{h_q}( x,y)}}=0$, où ${\ensuremath{\operatorname{h_q}}}$ est un polynôme homogène de degré $q$ sans facteurs multiples et $p\geq 2$, $q\geq 2$ sont deux entiers premiers entre eux.\
Nash a démontré que l’application $\mathcal{N}_{{\ensuremath{\mathbb{A}}}_{m}}$ associée à la surface de type ${\ensuremath{\mathbb{A}}}_{m}$, $m\geq 2$, est bijective (voir [@Nas95]). On suppose donc que $q\geq 3$, car l’hypersurface ${\ensuremath{\operatorname{S(p,h_{2})}}}$ est une singularité du type ${\ensuremath{\mathbb{A}}}_{p-1}$.\
Désingularisation des hypersurfaces ${\mbox{\boldmath{${\ensuremath{\operatorname{S(p,h_{q})}}}$}}}$
----------------------------------------------------------------------------------------------------
Un système de coordonnées affines $\{x,y,z\}$ de ${\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ étant fixé, notons $ O $ l’origine de ${\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$. On décrit la désingularisation de ${\ensuremath{\operatorname{S(p,h_{q})}}}$ en utilisant les constellations toriques de points infiniment voisins de $O$ (voir [@CGL96]), les éventails de Newton (voir [@GoLe91]) et les $G$-[*désingularisations*]{} (voir [@BoGo95]).\
Soient ${\ensuremath{\operatorname{N}}}={\ensuremath{\mathbb{Z}}}^3$ muni de sa base standard $\{{\ensuremath{\operatorname{e}}}_1, {\ensuremath{\operatorname{e}}}_2,{\ensuremath{\operatorname{e}}}_3\}$ et ${\ensuremath{\operatorname{M}}}={\ensuremath{\operatorname{Hom}}}_{{\ensuremath{\mathbb{Z}}}}(N,{\ensuremath{\mathbb{Z}}})$ le dual de ${\ensuremath{\operatorname{N}}}$. On note $\{{\ensuremath{\operatorname{e}}}_1^{\star}, {\ensuremath{\operatorname{e}}}_2^{\star},{\ensuremath{\operatorname{e}}}_3^{\star}\}$ la base duale de $\{{\ensuremath{\operatorname{e}}}_1, {\ensuremath{\operatorname{e}}}_2,{\ensuremath{\operatorname{e}}}_3\}$. On identifie la ${\ensuremath{\mathbf{k}}}$-algèbre ${\ensuremath{\mathbf{k}}}[{\ensuremath{\operatorname{M}}}]$ avec la ${\ensuremath{\mathbf{k}}}$-algèbre ${\ensuremath{\mathbf{k}}}[x,x^{-1},y,y^{-1},z,z^{-1}]$ par l’isomorphisme qui envoie le caractère $\chi^{{\ensuremath{\operatorname{e}}}_1^{\star}}$ (resp. $\chi^{{\ensuremath{\operatorname{e}}}_2^{\star}}$, $\chi^{{\ensuremath{\operatorname{e}}}_3^{\star}}$) sur $x$ (resp. $y$, $z$).
Soit $\Sigma$ un éventail en ${\ensuremath{\operatorname{N}}}_{{\ensuremath{\mathbb{R}}}}:={\ensuremath{\operatorname{N}}}\otimes_{{\ensuremath{\mathbb{Z}}}}{\ensuremath{\mathbb{R}}}$. On note $X(\Sigma)$ la variété torique associée à l’éventail $\Sigma$ et munie de l’action du tore algébrique $({\ensuremath{\mathbf{k}}}^{\star})^3={\ensuremath{\operatorname{Spec}}}{\ensuremath{\mathbf{k}}}[x,x^{-1},y,y^{-1},z,z^{-1}]$.
Soit $X_0= {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$. Une constellation torique de points infiniment voisins de $O$ est un ensemble fini de points ${\ensuremath{\mathcal{C}}}=\{Q_0=O,Q_1,...,Q_{m}\}$, où chaque $Q_{i}$, $1\leq i\leq m$, est une orbite de dimension $0$ de la variété torique $X_i$ obtenue par l’éclatement $\varsigma_i:X_{i}\rightarrow X_{i-1}$ de centre $Q_{i-1}$. On note $X({\ensuremath{\mathcal{C}}})=X_{m+1}$.
On dit que $Q_j$ se projette sur $Q_i$, noté $Q_j\geq Q_i$, si le point $Q_j\in X_j$ est obtenu à partir de $Q_i\in X_i$ par une suite d’éclatements de points. La relation $\geq$ est une relation d’ordre partiel sur les points de ${\ensuremath{\mathcal{C}}}$. Si $\geq$ est un ordre total, on dit que ${\ensuremath{\mathcal{C}}}$ est une constellation en chaîne.\
Supposons que ${\ensuremath{\mathcal{C}}}$ est une constellation torique en chaîne de points infiniment voisins de $O$. Soient ${\ensuremath{\operatorname{\mathcal{B}}}}:=\{{\ensuremath{\operatorname{e}}}_1,{\ensuremath{\operatorname{e}}}_2,{\ensuremath{\operatorname{e}}}_3\}$ la base ordonnée de ${\ensuremath{\operatorname{N}}}$ et $\Delta:=\langle {\ensuremath{\operatorname{\mathcal{B}}}}\rangle$ le cône régulier engendré par ${\ensuremath{\operatorname{\mathcal{B}}}}$. Notons $\Sigma_i$ l’éventail associé à la variété torique $X_i$. L’éventail $\Sigma_1$ est obtenu par la subdivision élémentaire de $\Delta$ centrée en $u=e_1+e_2+e_3$. Pour chaque entier $j$, $1\leq j\leq 3$, soit ${\ensuremath{\operatorname{\mathcal{B}}}}_j$ la base ordonnée de $N$ obtenue en remplaçant ${\ensuremath{\operatorname{e}}}_j$ par $u$ en la base ${\ensuremath{\operatorname{\mathcal{B}}}}$. Soit $\Delta_j:=\langle {\ensuremath{\operatorname{\mathcal{B}}}}_j\rangle$ le cône régulier engendré par ${\ensuremath{\operatorname{\mathcal{B}}}}_j$, pour $1\leq j \leq 3$. Le choix du point $Q_1>Q_0$ équivaut à choisir un entier $a_1$, $1\leq a_1\leq 3$, qui détermine un cône $\Delta_{a_1}$ de l’éventail $\Sigma_1$. La subdivision $\Sigma_2$ de $\Sigma_1$ est obtenue en remplaçant $\Delta_{a_1}$ en $\Sigma_1$ par les cônes $\Delta_{a_1j}:=\langle {\ensuremath{\operatorname{\mathcal{B}}}}_{a_1j}\rangle$, où ${\ensuremath{\operatorname{\mathcal{B}}}}_{a_1j}$ est la base ordonnée de ${\ensuremath{\operatorname{N}}}$ obtenue en remplaçant le $j$-ième vecteur de ${\ensuremath{\operatorname{\mathcal{B}}}}_{a_1}$ par $\sum_{u\in {\ensuremath{\operatorname{\mathcal{B}}}}_{a_1}}u$. Le choix du point $Q_2>Q_1$ équivaut à choisir un entier $a_2$, $1\leq a_2\leq 3$, qui détermine un cône $\Delta_{a_1a_2}$ de l’éventail $\Sigma_2$. Par récurrence, on obtient une codification de la constellation en chaîne ${\ensuremath{\mathcal{C}}}$. Cette codification est notée $Q_j=Q_0(a_1a_2\cdots a_j)$ pour $1\leq j\leq m$.
Soit $b_1,b_2,\cdots,b_k$ une suite d’entiers, où $1\leq b_i\leq 3$, $1\leq i \leq k \leq m$, et telle que le cône $\Delta_{b_1b_2\cdots b_k}$ appartient à l’éventail $\Sigma_{m+1}$. Notons $U_{ b_1b_2\cdots b_k}$ l’ouvert torique affine qui est en correspondance avec le cône $\Delta_{b_1b_2\cdots b_k}$.
Soit ${\ensuremath{\mathcal{C}}}_{n}=\{Q_0=O,Q_1,...,Q_{n-1}\}$, $n\geq 1$, une chaîne torique de points infiniment voisins de $O$ donnée par la codification $Q_j=Q_0(3^j)$ ($j$ fois l’entier $3$) pour $1\leq j\leq n-1$. On note $\sigma_{n}:X({\ensuremath{\mathcal{C}}}_n)\rightarrow {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ le morphisme torique induit par ${\ensuremath{\mathcal{C}}}_n$ et $S_{\mathcal C}$ le transformé strict de ${\ensuremath{\operatorname{S(p,h_{q})}}}$.
\[pr:const\] Soient $p>q$ et $p=nq+r$ la division entière, $1\leq r< q$. Si $r=1$, alors $S_{\mathcal C}$ est la résolution minimale de ${\ensuremath{\operatorname{S(p,h_{q})}}}$ et la fibre exceptionnelle $\sigma_{n}^{-1}(O)\cap S_{\mathcal C}$ est la réunion de $nq$ courbes rationnelles. Si $r> 1$, alors $S_{\mathcal C}$ a un unique point singulier $s$, et de plus, $(S_{\mathcal C},s)$ est isomorphe au germe $({\ensuremath{\operatorname{S(r,h_{q})}}},O)$.
La proposition se démontre par récurrence sur l’entier $n\geq 1$. Mais d’abord démontrons le lemme suivant.\
On considère l’éclatement de ${\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ de centre $O$, $\varsigma_1:X_1\rightarrow {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$. On note $x_i$, $y_i$, $ z_i$ les coordonnées canoniques de l’ouvert torique affine $U_i$, pour l’entier $1\leq i\leq 3$.
Soit $S$ le transformé strict de ${\ensuremath{\operatorname{S(p,h_{q})}}}$ induit par le morphisme $\varsigma_1:X_1\rightarrow {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$. Par abus de notation, on note $\varsigma_1:S\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$ la restriction du morphisme $\varsigma_1:X_1\rightarrow {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ à $S$.\
\[le:prconst\] Les ouverts de $S$, $U_1\cap S$ et $U_2\cap S$ sont lisses et $U_3\cap S=\{(x_3,y_3,z_3)\in U_3\mid z_3^{p-q}+{\ensuremath{\operatorname{h_q}( x_3,y_3)}}=0\}$.
La fibre exceptionnelle du morphisme $\varsigma_1:S\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$ est la réunion de $q$ courbes rationnelles. Les courbes ne s’intersectent qu’en le point fermé de $U_3$ fixé par l’action du tore algébrique $({\ensuremath{\mathbf{k}}}^{\star})^3$.
Si $p-q=1$, alors $S$ est la résolution minimale de ${\ensuremath{\operatorname{S(p,h_{q})}}}$.
Les restrictions du morphisme $\varsigma_1:X_1\rightarrow {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ aux ouverts $U_i$ sont données de la façon suivante:
- $U_1\rightarrow {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$, $(x_1,y_1, z_1)\mapsto (x_1,x_1y_1,x_1z_1)$;
- $U_2\rightarrow {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$, $(x_2,y_2, z_2)\mapsto (y_2x_2,y_2,y_2z_2)$;
- $U_3\rightarrow {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$, $(x_3,y_3, z_3)\mapsto (z_3x_3,z_3y_3,z_3)$.
Ainsi, on obtient que
- $U_1\cap S=\{(x_1,y_1,z_1)\in U_1\mid z_1^{p}x_1^{p-q}+{\ensuremath{\operatorname{h_q}( 1,y_1)}}=0\}$;
- $U_2\cap S=\{(x_2,y_2,z_2)\in U_2\mid z_2^{p}y_2^{p-q}+{\ensuremath{\operatorname{h_q}( x_2,1)}}=0\}$;
- $U_3\cap S=\{(x_3,y_3,z_3)\in U_3\mid z_3^{p-q}+{\ensuremath{\operatorname{h_q}( x_3,y_3)}}=0\}$.
Par un calcul direct, on montre que $U_1\cap S$ et $U_2\cap S$ sont lisses. Ceci achève la preuve de la première partie du lemme.\
Soit $F$ la fibre exceptionnelle du morphisme $\varsigma_1:S\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$. Alors on a
$U_3\cap F=\{(x_3,y_3,z_3)\in U_3\mid z_3=0,\;{\ensuremath{\operatorname{h_q}( x_3,y_3)}}=0\}$.
Comme ${\ensuremath{\operatorname{h_q}}}$ est un polynôme homogène de degré $q$ sans facteurs multiples, l’ensemble $U_3\cap F$ est la réunion de $q$ courbes rationnelles qui s’intersectent en l’origine de $U_3$.
On remarque que les deux ensembles suivants sont de cardinalité $q$.
- $(X_1\backslash U_3)\cap F\cap U_1 =\{(x_1,y_1,z_1)\in U_1\mid z_1=0,\;x_1=0,\; {\ensuremath{\operatorname{h_q}( 1,y_1)}}=0\}$;
- $(X_1\backslash U_3)\cap F\cap U_2 =\{(x_2,y_2,z_2)\in U_2\mid z_2=0,\;y_2=0,\; {\ensuremath{\operatorname{h_q}( x_2,1)}}=0\}$.
Par conséquent, la fibre exceptionnelle est la réunion d’exactement $q$ courbes rationnelles qui s’intersectent en l’origine de $U_3$. Ceci achève la preuve de la deuxième parti du lemme.\
Si $p-q=1$, $S$ est une surface lisse. On considère la fonction régulière de ${\ensuremath{\mathbb{A}}}^{3}_{{\ensuremath{\mathbf{k}}}}$, $g(x,y,z)=x$. Alors $g^{\star}=g\circ \varsigma_1$ est une fonction régulière de $X_1$.
Soit $C$ le diviseur principal de ${\ensuremath{\operatorname{S(p,h_{q})}}}$ associé à la restriction de $g$ à ${\ensuremath{\operatorname{S(p,h_{q})}}}$, c’est-à-dire $C:={\ensuremath{\operatorname{Div(g\mid_{{\ensuremath{\operatorname{S(p,h_{q})}}}})}}}$. On remarque que $C$ est une courbe (l’intersection schématique de ${\ensuremath{\operatorname{Div(g)}}}$ et ${\ensuremath{\operatorname{S(p,h_{q})}}}$ est irréductible et réduit). Notons $C'$ le transformé strict de $C$ par le morphisme $\varsigma_1:S\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$. Par un calcul direct, on peut montrer que $C'$ intersecte $F$ en l’origine de $U_3$.
On considère la restriction de $g^{\star}$ à l’ouvert $U_3$. Alors, on a
$g^{\star}=z_3x_3$.
Par conséquent, le diviseur principal de $S$ associé à la restriction de $g^{\star}$ à $S$ est le suivant:
${\ensuremath{\operatorname{Div(g^{\star}\mid_{S})}}}=C'+\sum_{i=1}^{q}F_i$,
où les $F_i$ sont les composantes irréductibles de $F$. Comme ${\ensuremath{\operatorname{Div(g^{\star}\mid_{S})}}}\cdot F_i=0$ pour tout $i\in\{1,2,...,q\}$, on obtient:
$F_i\cdot F_i=-C'\cdot F_i-\sum_{j\neq i}F_j\cdot F_i=-q\leq -3$.
Ceci achève la preuve du lemme.
Raisonnons par récurrence sur l’entier $n$.\
Si $n=1$, on a $p=q+r$ et ${\ensuremath{\mathcal{C}}}_1=\{Q_0=O\}$. La preuve de ce cas résulte du Lemme \[le:prconst\].\
Maintenant, on suppose que $p=nq+r$.\
D’après le Lemme \[le:prconst\], on a $U_3\cap S={\ensuremath{\operatorname{S((n-1)q+r,h_{q})}}}$ et la fibre exceptionnelle de $\varsigma_1:S\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$ est la réunion de $q$ courbes rationnelles. En appliquant l’hypothèse de récurrence sur $U_3\cap S$, on montre:
- la fibre exceptionnelle de $\sigma_n:S_{\mathcal{C}}\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$ est la réunion de $nq$ courbes rationnelles;
- si $r>1$, alors $S_{\mathcal{C}}$ a un unique point singulier de type ${\ensuremath{\operatorname{S(r,h_{q})}}}$;
- si $r=1$, alors $S_{\mathcal{C}}$ est lisse.\
Pour achever la preuve de la proposition, il faut montrer que si $r=1$, alors le morphisme $\sigma_n:S_{\mathcal{C}}\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$ est la résolution minimale de ${\ensuremath{\operatorname{S(p,h_{q})}}}$.\
Soient $F$ la fibre exceptionnelle du morphisme $\varsigma_1:S\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$ et $F_i$ une composante irréductible de $F$, $1\leq i\leq q$. On note $F'_i$ le transformé strict de $F_i$ dans $S_{\mathcal C}$ (on rappelle que le morphisme $\varsigma_1$ factorise $\sigma_n$). En utilisant l’hypothèse de récurrence, pour montrer que $S_{\mathcal{C}}$ est la résolution minimale de ${\ensuremath{\operatorname{S(p,h_{q})}}}$, il suffit de montrer que $F'_i\cdot F'_i\leq -2$ pour tout $1\leq i\leq q$.\
Pour chaque $Q_i\in {\ensuremath{\mathcal{C}}}_n$, on note ${\ensuremath{\operatorname{B}}}_i$ le diviseur exceptionnel $\varsigma_i^{-1}(Q_{i})$ en $X_{i+1}$ et ${\ensuremath{\operatorname{D}}}_i$ (resp. ${\ensuremath{\operatorname{D}}}_i^{\star}$) le transformé strict (resp. total) de ${\ensuremath{\operatorname{B}}}_i$ dans $X({\ensuremath{\mathcal{C}}}_n)$.\
On rappelle que la constellation ${\ensuremath{\mathcal{C}}}_n$ est donnée par la codification $Q_j=Q_{0}(3^j)$ pour $1\leq j \leq n-1$. En vertu du lemme $1.3$ de [@CGL96], on a
${\ensuremath{\operatorname{D}}}_i={\ensuremath{\operatorname{D}}}_i^{\star}- {\ensuremath{\operatorname{D}}}_{i+1}^{\star}$,
d’où ${\ensuremath{\operatorname{D}}}_i^{\star}=\sum_{j=i}^{n-1}{\ensuremath{\operatorname{D}}}_j$.\
Avant d’achever la preuve de la Proposition \[pr:const\], on introduit les notions suivantes.\
Soient $\mathcal{I}$, $\mathcal{J}$, $\mathcal{P}$ trois idéaux. Le $\star$-produit de $\mathcal{I}$ et $\mathcal{J}$, noté $\mathcal{I}\star\mathcal{J}$, est la clôture intégrale du produit $\mathcal{I}\mathcal{J}$. On suppose que $\mathcal{P}$ est un idéal complet non-trivial. L’idéal $\mathcal{P}$ est $\star$-simple si $\mathcal{P}$ n’a pas de $\star$-factorisation non-triviale.
On suppose que $\mathcal{I}$ est un idéal tel que $\mathcal{I}\mathcal{O}_{X({\ensuremath{\mathcal{C}}}_{n})}$ soit un faisceau d’idéaux inversible. On définit par récurrence le vecteur ${\ensuremath{\operatorname{m}}}(\mathcal{I}) =({\ensuremath{\operatorname{m}}}_0,...,{\ensuremath{\operatorname{m}}}_{n-1})$: $\mathcal{I}_{0}=\mathcal{I}$, ${\ensuremath{\operatorname{m}}}_0={\ensuremath{\operatorname{Ord}}}_{Q_0}\mathcal{I}_0$ et ${\ensuremath{\operatorname{m}}}_{i}={\ensuremath{\operatorname{Ord}}}_{Q_i} \mathcal{I}_{i}$, $1\leq i\leq n-1$, où $\mathcal{I}_{i}=x^{-{\ensuremath{\operatorname{m}}}_{i-1}}\mathcal{I}_{i-1}\mathcal{O}_{X_{i},Q_{i}}$ et $x=0$ est l’équation locale de ${\ensuremath{\operatorname{B}}}_{i-1}$ en $Q_i$. Dans [@Lip88] l’auteur montre qu’il existe un unique idéal $\star$-simple, noté $\mathcal{P}_{Q_{n-1}}$, tel que le vecteur ${\ensuremath{\operatorname{m}}}(\mathcal{P}_{Q_{n-1}})$ est minimal pour l’ordre lexicographique inverse et ${\ensuremath{\operatorname{m}}}_{n-1}=1$. L’idéal $\mathcal{P}_{Q_{n-1}}$ est appelé [*l’idéal $\star$-simple spécial*]{} associé à $Q_{n-1}$.\
Reprenons la démonstration de la Proposition \[pr:const\].\
Soient $\mathcal{P}_{Q_{n-1}}$ [*l’idéal $\star$-simple spécial*]{} associé à $Q_{n-1}$ et ${\ensuremath{\operatorname{D}}}'$ le diviseur de $X({\ensuremath{\mathcal{C}}}_n)$ tel que
$\mathcal{P}_{Q_{n-1}}\mathcal{O}_{X({\ensuremath{\mathcal{C}}}_n)}=\mathcal{O}_{X({\ensuremath{\mathcal{C}}}_n)}(-{\ensuremath{\operatorname{D}}}')$.
D’après le Lemme $2.16$ de [@CGL96], on a
${\ensuremath{\operatorname{D}}}'=\sum_{i=0}^{n-1}{\ensuremath{\operatorname{D}}}_i^{\star}$.
Par conséquent, on a
${\ensuremath{\operatorname{D}}}'=\sum_{i=0}^{n-1}(i+1){\ensuremath{\operatorname{D}}}_i$.
Soit $g$ un élément général de l’idéal $\mathcal{P}_{Q_{n-1}}$. Alors $g^{\star}=g\circ\sigma_n$ est une fonction régulière de $X({\ensuremath{\mathcal{C}}}_n)$. On note $C$ le diviseur associé à $g$ et $C'$ le transformé strict de $C$ dans $X({\ensuremath{\mathcal{C}}}_n)$. Alors, on a:
${\ensuremath{\operatorname{Div(g^{\star})}}}=C'+\sum_{i=0}^{n-1}(i+1){\ensuremath{\operatorname{D}}}_i$.
Pour un diviseur $Z$ de $X({\ensuremath{\mathcal{C}}}_n)$, tel que son support ne contienne pas $S_{\mathcal{C}}$, notons $Z\cdot S_{\mathcal{C}}$ le diviseur de $S_{\mathcal{C}}$ obtenu par la somme formelle des composantes irréductibles de $Z\cap S_{\mathcal{C}}$ pondérées par leurs multiplicités en $S_{\mathcal{C}}$. Comme $g$ est un élément général de $\mathcal{P}_{Q_{n-1}}$ on peut supposer que le support de $C'$ ne contient pas $S_{\mathcal{C}}$. Ainsi, on obtient que
${\ensuremath{\operatorname{Div(g^{\star} \mid_{S_{\mathcal{C}}})}}}=C'\cdot S_{\mathcal{C}}+\sum_{i=0}^{n-1}(i+1){\ensuremath{\operatorname{D}}}_i\cdot S_{\mathcal{C}}$.
On remarque que ${\ensuremath{\operatorname{D}}}_0\cdot S_{\mathcal{C}}= \sum_{i=1}^{q}F'_i$. En effet, il suffit de considérer l’ouvert affine $U_1$. On a donc $U_1\cap S_{\mathcal{C}}=\{(x_1,y_1,z_1)\in U_1\mid z_1^{p}x_1^{p-q}+{\ensuremath{\operatorname{h_q}( 1,y_1)}}=0\}$ et ${\ensuremath{\operatorname{D}}}_0\cap U_1= \{(x_1,y_1,z_1)\in U_1\mid x_1=0\}$. Par conséquent, $(D_0\cdot S_{\mathcal{C}})\cap U_1= \sum_{i=1}^{q}(F'_i\cap U_1)$, d’où ${\ensuremath{\operatorname{D}}}_0\cdot S_{\mathcal{C}}= \sum_{i=1}^{q}F'_i$.\
On a donc
${\ensuremath{\operatorname{Div(g^{\star} \mid_{S_{\mathcal{C}}})}}}=C'\cdot S_{\mathcal{C}}+\sum_{i=1}^{n-1}(i+1){\ensuremath{\operatorname{D}}}_i\cdot S_{\mathcal{C}} +\sum_{i=1}^{q}F'_i$.
On remarque que l’intersection ${\ensuremath{\operatorname{Div(g^{\star} \mid_{S_{\mathcal{C}}})}}}\cdot F'_i$ est nulle pour tout $1\leq i\leq q$, car ${\ensuremath{\operatorname{Div(g^{\star} \mid_{S_{\mathcal{C}}})}}}$ est un diviseur principal. On remarque aussi que $(\sum_{i=1}^{n-1}(i+1){\ensuremath{\operatorname{D}}}_i\cdot S_{\mathcal{C}})\cdot F'_i\geq 2$, car il existe au moins un $ 1\leq i\leq n-1$ tel que $((i+1){\ensuremath{\operatorname{D}}}_i\cdot S_{\mathcal{C}})\cdot F'_i\geq (i+1)\geq 2$. Par conséquent, on obtient que $F'_i\cdot F'_i\leq -2$ pour tout $1\leq i\leq q$, d’où la proposition.
Pour un polynôme ${\ensuremath{\operatorname{g}}}=\sum c_{e}x^{e_1}y^{e_2}z^{e_3}$ dans ${\ensuremath{\mathbf{k}}}[x,y,z]$, où $e=(e_1,e_2,e_3)$ et $c_{e} \in {\ensuremath{\mathbf{k}}}$, on note $\mathcal{E}({\ensuremath{\operatorname{g}}})$ l’ensemble des exposants $e\in {\ensuremath{\mathbb{Z}}}^{3}_{\geq 0}$ dont le coefficient $c_{e}$ est non nul, c’est-à-dire $\mathcal{E}({\ensuremath{\operatorname{g}}}):=\{e\in{\ensuremath{\mathbb{Z}}}^{3}_{\geq 0}\mid c_e\neq 0\}$. Soient $\Gamma_{+}({\ensuremath{\operatorname{g}}})$ l’enveloppe convexe de l’ensemble $\{e+{\ensuremath{\mathbb{R}}}^{3}_{\geq 0}\mid e\in \mathcal{E}({\ensuremath{\operatorname{g}}})\}$ et $\Gamma({\ensuremath{\operatorname{g}}})$ la réunion des faces compactes de $\Gamma_{+}({\ensuremath{\operatorname{g}}})$. On note $\mathcal{I}({\ensuremath{\operatorname{g}}})$ l’idéal monomial de ${\ensuremath{\mathbf{k}}}[x,y,z]$ engendré par les monômes $x^{e_1}y^{e_2}z^{e_3}$ tels que $(e_1,e_2,e_3)\in \Gamma({\ensuremath{\operatorname{g}}})\cap {\ensuremath{\mathbb{Z}}}^3$, c’est-à-dire $\mathcal{I}({\ensuremath{\operatorname{g}}}):=(\{x^{e_1}y^{e_2}z^{e_3}\mid (e_1,e_2,e_3)\in \Gamma({\ensuremath{\operatorname{g}}})\cap {\ensuremath{\mathbb{Z}}}^3\})$. L’éventail de Newton ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})}}}$ associé à ${\ensuremath{\operatorname{g}}}$ est la subdivision de ${\ensuremath{\mathbb{R}}}^{3}_{\geq 0}$ correspondant à l’éclatement normalisé de ${\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ de centre l’idéal $\mathcal{I}(g)$. Pour plus de détails voir [@GoLe91] ou [@KKMS73].\
\[re:x-et-y-no-div-h\] Dans toute la suite, à automorphisme linéaire de ${\ensuremath{\mathbb{A}}}^3_{{\ensuremath{\mathbf{k}}}}$ près, $x$ et $y$ ne divisent pas ${\ensuremath{\operatorname{h_q}( x,y)}}$.
La Figure \[fi:PN\] représente le polyèdre de Newton $\Gamma({\ensuremath{\operatorname{f}}})$ associe à ${\ensuremath{\operatorname{f}}}:=z^p+{\ensuremath{\operatorname{h_q}( x,y)}}$.
(330:3) – (330:5); (90:3) – (90:5); (210:3) – (210:5) ; (330:3) – (90:3) – (210:3) – (330:3);
(90:3) – (330:3) – (210:3); (210:3) –(330:3) – (330:5) – (210:5) ; (210:3) – (90:3) – (90:5) – (210:5) ; (90:3) – (330:3) – (330:5) – (90:5);
(342:3.5) node\[inner sep=-1pt,below=-1pt\] [[$(0,q,0)$]{}]{};
(198:3.5) node\[inner sep=-1pt,below=-1pt\] [[$(q,0,0)$]{}]{};
(75:3.3) node\[inner sep=-1pt,below=-1pt\] [[$(0,0,p)$]{}]{};
Soit $H$ un plan de ${\ensuremath{\mathbb{R}}}^3$ qui ne contient pas l’origine de ${\ensuremath{\mathbb{R}}}^3$ et tel que l’intersection de $H$ et ${\ensuremath{\mathbb{R}}}^3_{\geq 0}$ soit un ensemble compact. La Figure \[fi:EN\] représente l’intersection de $H$ avec la subdivision ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$ de ${\ensuremath{\mathbb{R}}}^3_{\geq 0}$. Chaque sommet du diagramme est identifié avec le [*vecteur extrémal*]{} (autrement dit, vecteur primitif d’un cône de dimension $1$ de l’éventail ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$) correspondant. On note $\tau_1$ (resp. $\tau_2$, $\tau_3$) le cône engendré par les vecteurs $(1,0,0)$ (resp. $(0,1,0)$, $(0,0,1)$) et $(p,p,q)$.
(0,0) – (330:3) (0,0) – (90:3) (0,0) – (210:3); (330:3) – (90:3) – (210:3) – (330:3); (0,-0.4) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(p,p,q)$]{}]{}; (335:3.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(0,1,0)$]{}]{}; (340:1.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [$\tau_2$]{}; (205:3.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(1,0,0)$]{}]{}; (200:1.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [$\tau_1$]{}; (90:3.3) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(0,0,1)$]{}]{}; (80:1.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [$\tau_3$]{}; (0,0) circle(0.7mm) (330:3) circle(0.7mm) (90:3) circle(0.7mm) (210:3) circle(0.7mm);
La proposition suivante résulte d’un calcul direct.
\[pr:tau12\] Le cônes $\tau_1$ et $\tau_2$ sont réguliers.
Soit ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}}$ une $G$-[*subdivision régulière*]{} de ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$, c’est-à-dire une subdivision régulière de chaque cône $\tau\in{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$ n’ayant comme arêtes que celles qui portent les vecteurs du système générateur minimal du semi-groupe $\tau\cap {\ensuremath{\mathbb{Z}}}^3$. D’après [@BoGo95], cette subdivision existe.\
On note $\pi_{\mathcal{N}}:X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}})\rightarrow {\ensuremath{\mathbb{A}}}^{3}_{{\ensuremath{\mathbf{k}}}}$ (resp. $\pi_{\mathcal{G}}:X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}})\rightarrow X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}})$) le morphisme torique induit par la subdivision ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$ de ${\ensuremath{\mathbb{R}}}^3_{\geq 0}$ (resp. ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}}$ de ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$) et $S_{\mathcal{G}}$ le transformé strict de ${\ensuremath{\operatorname{S(p,h_{q})}}}$ associé au morphisme $\pi:=\pi_{\mathcal{G}}\circ \pi_{\mathcal{N}}$ Par abus de notation, on note $\pi:S_{\mathcal{G}}\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$ la restriction du morphisme $\pi:X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}})\rightarrow {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ à $S_{\mathcal{G}}$.\
Soit $k$, un entier $k\geq 1$. Pour un ensemble d’entiers $m_i\geq 2$, $1\leq i\leq k$, on note $[m_1;m_2;\cdots ;m_k]$ la fraction continue définie de la façon suivante:
$[m_k]:=m_k$, $[m_{k-1};m_{k}]:= \displaystyle m_{k-1}-\frac{1}{m_k}$ et $[m_1;m_2;\cdots ;m_k]: =\displaystyle m_1 - \frac{1}{\displaystyle [m_2;\cdots ;m_k]}$.
La proposition suivante est une application directe du Théorème $6.1$ de [@Oka87].
\[pr:Gdes1\] $S_{\mathcal{G}}$ est une bonne résolution de ${\ensuremath{\operatorname{S(p,h_{q})}}}$ et son graphe dual pondéré est une étoile à $q$ branches identiques. Le diagramme de chaque branche est le suivant:
(0:6.1) node\[inner sep=0pt,below=2pt,rectangle,fill=white\] [$ \tiny{ -m_{2}}$]{} (0:3) node\[inner sep=0pt,below=2pt,rectangle,fill=white\] [$ {\small -m_{k}}$]{};
(0:9.1) node\[inner sep=0pt,below=2pt,rectangle,fill=white\] [$ \tiny{ -m_{1}}$]{};
(180:1) node\[inner sep=0pt,below=-1pt,rectangle,fill=white\] [$ \tiny{ E}^{2}_{0}$]{};
(0, 0) – (0:4) (0:4.35) – (0:4.4) (0:4.75) – (0:4.8) (0:5.15) – (0:5.2) (0:5.6) –(0:9.1);
(0,0) circle(0.7mm) (0:6.1) circle(0.7mm) (0:3) circle(0.7mm) (0:9.1) circle(0.7mm);
,
où ${\ensuremath{\operatorname{E}}}_0$ le diviseur associé au sommet central du graphe et les entiers $m_i\geq 2$ (resp. l’entier $k$) sont définis (resp. est défini) de la façon suivante:
- si $q>p$ et $q=np+r$ la division entière, $1\leq r<p$, alors on a
$\displaystyle\frac{p}{p-r}=[m_1;m_2;\cdots ;m_k]$;
- si $p>q$ et $p=nq+r$, la division entière $1\leq r<q$ (resp. $n\geq 1$), alors on a
$\displaystyle\frac{p}{p-q}=[m_1;m_2;\cdots ;m_k]$.
De plus, cette résolution est minimale si et seulement si $p\not \equiv 1 \mod{q}$.
\[co:Gdes.r=1\] Si $p\equiv 1\mod{q}$, seul le diviseur ${\ensuremath{\operatorname{E}}}_0$ n’est pas un diviseur essentiel sur ${\ensuremath{\operatorname{S(p,h_{q})}}}$.
Soit $n\geq 1$ tel que $p=nq+1$. En vertu de la Proposition \[pr:Gdes1\], on a
$\displaystyle \frac{p}{p-q} = \displaystyle \frac{nq +1 }{(n-1)q+1} =\displaystyle 2 - \frac{1}{\displaystyle 2 -\frac{\ddots}{\displaystyle 2-\frac{1}{\displaystyle q+1}}}.$
En particulier l’entier $k$ est égal à $n$, d’où le graphe dual de la résolution $\pi:{\ensuremath{\operatorname{S_{\mathcal{G}}}}}\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$ a $nq +1$ sommets. D’après la Proposition \[pr:const\], il n’y a que un diviseur de la fibre exceptionnelle de $\pi$ qui n’est pas un diviseur essentiel. D’après [*le critère de contraction de Castelnuovo*]{}, ce diviseur a une auto-intersection égale à $-1$.
Comme les branches du graphe dual de la résolution $\pi:{\ensuremath{\operatorname{S_{\mathcal{G}}}}}\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$ sont identiques, forcement le diviseur ${\ensuremath{\operatorname{E}}}_0$ a une auto-intersection égal à $-1$, d’où le corollaire.
Un polynôme ${\ensuremath{\operatorname{g}}}=\sum c_{e}x^{e_1}y^{e_2}z^{e_3}$, où $e=(e_1,e_2,e_3)$ et $c_{e} \in {\ensuremath{\mathbf{k}}}$, est appelé [*non*]{}-[*dégénéré par rapport à la frontière de Newton*]{} si pour toute face compacte $\gamma$ de ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})}}}$ le polynôme ${\ensuremath{\operatorname{g}}}_{\gamma}:=\sum_{e\in \gamma}c_ex^{e_1}y^{e_2}z^{e_3}$ est non singulier sur le tore $T:={\ensuremath{\operatorname{N}}}\oplus_{{\ensuremath{\mathbb{Z}}}}{\ensuremath{\mathbf{k}}}$, c’est-à-dire les polynômes ${\ensuremath{\operatorname{g}}}_{\gamma}$, $\partial_x{\ensuremath{\operatorname{g}}}_{\gamma}$, $\partial_y{\ensuremath{\operatorname{g}}}_{\gamma}$, $\partial_z{\ensuremath{\operatorname{g}}}_{\gamma}$ n’ont pas de zéro commun en dehors de l’ensemble $xyz=0$.\
Dans la proposition suivante on suppose que $S$ est une hypersurface normale de ${\ensuremath{\mathbb{A}}}^{3}_{{\ensuremath{\mathbf{k}}}}$, donnée par l’équation ${\ensuremath{\operatorname{g}}}=0$, où ${\ensuremath{\operatorname{g}}}$ est un polynôme irréductible non-dégénéré par rapport à la frontière de Newton $\Gamma({\ensuremath{\operatorname{g}}})$. De plus, on suppose que $O$ est l’unique point singulier de $S$.\
On considère une $G$-subdivision régulière ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$ de l’éventail de Newton ${\ensuremath{\operatorname{\Gamma^{\star}(g)}}}$. On note $\pi':X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}})\rightarrow {\ensuremath{\mathbb{A}}}^{3}_{{\ensuremath{\mathbf{k}}}}$ le morphisme torique induit par la subdivision ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$ du cône ${\ensuremath{\mathbb{R}}}^{3}_{\geq 0}$ et $X$ le transformé strict de $S$ dans $X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}})$. Par abus de notation, on note $\pi':X\rightarrow S$ la restriction du morphisme $\pi':X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}})\rightarrow {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ à $X$.\
La proposition suivante résulte des Lemmes $10.2$ et $10.3$ de [@Var76] (pour avoir plus de détails, voir [@Mer80]) ou du Théorème principal et de la Remarque [*a*]{}) de la Section $4$ de [@GoLe91].\
\[pr:cr\_nor-g\] Le morphisme $\pi':X\rightarrow S$ est une désingularisation de $S$.
Si l’hypersurface $S$ ne contient pas de $T$-orbite de dimension $1$, alors le morphisme $\pi':X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}})\rightarrow {\ensuremath{\mathbb{A}}}^3_{{\ensuremath{\mathbf{k}}}}$ est une résolution plongée de $S$, c’est-à-dire $\pi':X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}})\rightarrow {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ est un morphisme propre et birationnel, $\pi':X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}})\backslash(\pi')^{-1}(O)\rightarrow {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3\backslash \{O\}$ est un isomorphisme et $(\pi')^{-1}(S)$ est un diviseur à croisements normaux.
On remarque que le polynôme ${\ensuremath{\operatorname{f}}}:=z^p+{\ensuremath{\operatorname{h_q}( x,y)}}$ est non-dégénéré par rapport à la frontière de Newton et que l’hypersurface ${\ensuremath{\operatorname{S(p,h_{q})}}}$ ne contient pas de $T$-orbite de dimension $1$ (voir la Remarque \[re:x-et-y-no-div-h\]).
\[co:cr\_nor\] le morphisme $\pi:X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}})\rightarrow {\ensuremath{\mathbb{A}}}^3_{{\ensuremath{\mathbf{k}}}}$ est une résolution plongée de ${\ensuremath{\operatorname{S(p,h_{q})}}}$
La proposition suivante établit une relation entre le morphisme $\pi:{\ensuremath{\operatorname{S_{\mathcal{G}}}}}\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$ et le morphisme $\sigma_n:S_{\mathcal C}\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$ (voir les Propositions \[pr:const\] et \[pr:Gdes1\]).
\[pr:Gdes2\] Si $p>q$, le morphisme $\sigma_n:S_{\mathcal C}\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$ factorise le morphisme $\pi:{\ensuremath{\operatorname{S_{\mathcal{G}}}}}\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$, c’est-à-dire il existe un morphisme $\pi_0:{\ensuremath{\operatorname{S_{\mathcal{G}}}}}\rightarrow S_{\mathcal C}$ tel que $\pi=\sigma_n \circ \pi_0$.
Dans la preuve de cette proposition, on peut appliquer le Lemme \[le:prconst\] car ses hypothèses sont vérifiées.
On remarque que
$(1,1,1)=\dfrac{(p,p,q)+(p-q)(0,0,1)}{p}$
et que le cône engendré par les vecteurs $(0,0,1)$ et $(1,1,1)$ est régulier. Par conséquent, le vecteur $(1,1,1)$ appartient au système générateur minimal du semi-groupe $\tau_3\cap{\ensuremath{\mathbb{Z}}}^3$, où $\tau_3$ est le cône engendré par les vecteurs $(0,0,1)$ et $(p,p,q)$ (voir la Figure \[fi:EN\]).\
On remarque aussi que l’éventail obtenu par l’éclatement de Newton de ${\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ associé à ${\ensuremath{\operatorname{f}}}=z^p +{\ensuremath{\operatorname{h_q}( x,y)}}$ suivi de la subdivision élémentaire centrée en $(1,1,1)$, coïncide avec celui obtenu par l’éclatement de ${\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ de centre le point $O$ suivi de l’éclatement de Newton de $U_3$ associé à $z_3^{p-q} +{\ensuremath{\operatorname{h_q}( x_3,y_3)}}$ (voir le Lemme \[le:prconst\]).\
Soit $p=nq+r$ la division entière, $1\leq r<q$. En utilisant la remarque ci-dessus et le Lemme \[le:prconst\], la proposition résulte d’une récurrence sur l’entier $n\geq 1$.
Dans la proposition suivante, on suppose que $S$ est une hypersurface quasi-homogène de ${\ensuremath{\mathbb{A}}}^{3}_{{\ensuremath{\mathbf{k}}}}$ donnée par l’équation ${\ensuremath{\operatorname{g}}}=0$, où ${\ensuremath{\operatorname{g}}}$ est un polynôme quasi-homogène et irréductible, que $O$ est l’unique point singulier de l’hypersurface $S$ et que $S$ ne contient pas de $T$-orbite de dimension $1$.\
D’après la proposition \[pr:cr\_nor-g\], le morphisme $\pi':X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}) \rightarrow {\ensuremath{\mathbb{A}}}^{3}_{{\ensuremath{\mathbf{k}}}}$ est une résolution plongée de $S$. On note $X$ le transformé strict de $S$ dans $X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}})$. Par abus de notation, on note $\pi':X\rightarrow S$ la restriction du morphisme $\pi':X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}})\rightarrow {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ à $X$.\
Pour $\rho$ un [*vecteur extrémal*]{} de ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$ (c’est-à-dire $\rho$ est un vecteur primitif d’un cône de dimension $1$ qui appartient à l’éventail ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$), on note ${\ensuremath{\operatorname{D}}}_{\rho}$ l’orbite fermée associée à $\rho$. Notons $E{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}}$ l’ensemble des vecteurs extrémaux de ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$, et $S_{2}{\ensuremath{\operatorname{\Gamma^{\star}(g)}}}$ le $2$-squelette de ${\ensuremath{\operatorname{\Gamma^{\star}(g)}}}$ (c’est-à-dire $S_{2}{\ensuremath{\operatorname{\Gamma^{\star}(g)}}}$ est la réunion des cônes de dimension $2$ qui appartiennent à l’éventail $S_{2}{\ensuremath{\operatorname{\Gamma^{\star}(g)}}}$).\
La proposition suivante donne des équations locales pour les composantes irréductibles de la fibre exceptionnelle de la désingularisation $\pi':X\rightarrow S$.
Dans le cas $S={\ensuremath{\operatorname{S(p,h_{q})}}}$, on rappelle que ${\ensuremath{\operatorname{E}}}_0$ est le diviseur associé au sommet central du graphe induit par $\pi$.
\[pr:sygemi\] Soit $\rho$ un vecteur extrémal de ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$ ($\rho \in E{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$). Alors, on a:
- l’intersection ${\ensuremath{\operatorname{D}}}{\rho}\cap X$ n’est pas vide si et seulement si $\rho$ appartient au 2-squelette de l’éventail de Newton ${\ensuremath{\operatorname{\Gamma^{\star}(g)}}}$, c’est-à-dire $\rho\in S^{2}{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})}}}\cap E{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$;
- les composantes irréductibles de ${\ensuremath{\operatorname{D}}}{\rho}\cap X$ sont diviseurs exceptionnels du morphisme $\pi':X\rightarrow S$ si et seulement si de plus $\rho\in {\ensuremath{\mathbb{Z}}}_{>0}^3$;
- une composante irréductible ${\ensuremath{\operatorname{E}}}$ de la fibre exceptionnelle de $\pi'$ étant donnée, il existe un unique $\rho\in E{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$ tel que ${\ensuremath{\operatorname{E}}}\subset {\ensuremath{\operatorname{D}}}{\rho}$;
- si $S={\ensuremath{\operatorname{S(p,h_{q})}}}$, alors ${\ensuremath{\operatorname{E}}}_0={\ensuremath{\operatorname{D}}}{(p,p,q)}\cap{\ensuremath{\operatorname{S_{\mathcal{G}}}}}$;
- si $S={\ensuremath{\operatorname{S(p,h_{q})}}}$, alors l’ensemble formé par le vecteur $(0,0,1)$ et les vecteurs $\rho \in E{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}}$ tels que les composantes irréductibles de ${\ensuremath{\operatorname{D}}}{\rho}\cap {\ensuremath{\operatorname{S_{\mathcal{G}}}}}$ sont diviseurs exceptionnels de $\pi$ est le système générateur minimal du semi-groupe $\tau\cap {\ensuremath{\mathbb{Z}}}^3$, où $\tau$ est le cône engendré par les vecteurs $(0,0,1)$ et $(p,p,q)$.
Les résultats de l’article [@Oka87] reposent sur la construction d’une subdivision régulière $\Sigma$ de l’éventail ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})}}}$. Cette construction est longue à définir, or quand le polynôme ${\ensuremath{\operatorname{g}}}$ est quasi-homogène, on peut supposer que $\Sigma$ est une $G$-subdivision régulière de l’éventail ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})}}}$.
Les points [*i),ii),iii)*]{} et [*iv)*]{} de la proposition résultent de la Remarque $4.3$ et du Lemme $4.7$ de [@Oka87]. Le point [*v)*]{} résulte des points [*i)*]{}, [*ii)*]{} et de la Proposition \[pr:tau12\].
Preuve du Théorème 1.1.
-----------------------
Dans cette section, on montre la bijectivité de l’application de Nash pour les hypersurfaces quasi-rationnelles ${\ensuremath{\operatorname{S(p,h_{q})}}}$, ce qui équivaut à montrer que tous les wedges admissibles se relèvent à la résolution minimale de ${\ensuremath{\operatorname{S(p,h_{q})}}}$ (voir [@Reg06]). Notre but, dans toute la suite de cette section, est de montrer que pour chaque diviseur essentiel ${\ensuremath{\operatorname{E}}}$ (${\ensuremath{\operatorname{E}}}\in {\ensuremath{\operatorname{Ess}}}({\ensuremath{\operatorname{S(p,h_{q})}}})$) tous les $K$-wedges admissibles centrés en $N_{{\ensuremath{\operatorname{E}}}}$ se relèvent à la résolution minimale de ${\ensuremath{\operatorname{S(p,h_{q})}}}$. De plus, on profite de démontrer quelques résultats qu’on utilise dans toutes les sections de cet article.\
Avec le théorème suivant on réduit le nombre de cas à étudier. Une preuve de ce résultat, dans le cas des singularités de surfaces rationnelles, se trouve dans [@Ple05].
\[th:faremi\] Soient $V$ une surface algébrique normale sur ${\ensuremath{\mathbf{k}}}$ et $\pi:Y\rightarrow V$ la résolution minimale de $V$. Supposons qu’il existe un morphisme propre et birationnel $\pi':V'\rightarrow V$, où $V'$ est une surface algébrique normale sur ${\ensuremath{\mathbf{k}}}$, tel que $\pi'$ factorise $\pi$. Alors, si l’application de Nash ${\ensuremath{\mathcal{N}}}_{V}$ associée à $V$ est bijective, l’application de Nash ${\ensuremath{\mathcal{N}}}_{V'}$ associée à $V'$ l’est aussi.
On remarque que $Y$ est la résolution minimale de $V'$. Si $\omega$ est un $K$-wedge sur $V'$, alors $\pi'\circ \omega$ est un $K$-wedge admissible sur $V$. Par conséquent, $\pi'\circ \omega$ se relève à $Y$, d’où le Théorème.
En vertu des résultats \[pr:const\], \[pr:Gdes2\] et \[th:faremi\], on a le corollaire suivant.
Si l’application de Nash ${\ensuremath{\mathcal{N}}}_{{\ensuremath{\operatorname{S(p,h_{q})}}}}$ associée à l’hypersurface ${\ensuremath{\operatorname{S(p,h_{q})}}}$ est bijective pour tous les entiers $p>q\geq 3$, premiers entre eux, alors l’application de Nash ${\ensuremath{\mathcal{N}}}_{{\ensuremath{\operatorname{S(p,h_{q})}}}}$ est bijective pour tous les entiers $p\geq 2$ $q\geq 2$, premiers entre eux.
\[re:p>q\] Dans toute la suite, on suppose que $p>q\geq 3$.
Dans la proposition suivante, $S$ désigne l’hypersurface de la Proposition \[pr:sygemi\], ${\ensuremath{\operatorname{E}}}$ désigne un diviseur essentiel sur $S$ et $\alpha_{{\ensuremath{\operatorname{E}}}}$ désigne le point générique de $N_{{\ensuremath{\operatorname{E}}}}$. On pose
$(\mu_x,\mu_y,\mu_z):=({\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(x),{\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(y), {\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(z))\in {\ensuremath{\mathbb{Z}}}^3_{>0}$.
\[pr:musygemi-g\] Le vecteur $(\mu_x,\mu_y,\mu_z)$ appartient à l’intersection de l’ensemble $E{\ensuremath{\operatorname{\Gamma^{\star}(g)_{\mathcal{G}}}}}$ des extrémaux de ${\ensuremath{\operatorname{\Gamma^{\star}(g)_{\mathcal{G}}}}}$ avec le $2$-squelette $S_2{\ensuremath{\operatorname{\Gamma^{\star}(g)}}}$ de l’éventail de Newton ${\ensuremath{\operatorname{\Gamma^{\star}(g)}}}$, c’est-à-dire $(\mu_x,\mu_y,\mu_z)\in E{\ensuremath{\operatorname{\Gamma^{\star}(g)_{\mathcal{G}}}}}\cap S_2{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})}}}$.
Dans la démonstration, on utilise les notations usuelles de variétés toriques (voir [@KKMS73]).
Soit ${\ensuremath{\operatorname{E}}}\in {\ensuremath{\operatorname{Ess}}}(S)$. En vertu de la Proposition \[pr:sygemi\], il existe un unique vecteur extrémal $\rho_1$ appartenant à $S_{2}{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})}}}\cap E{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$ tel que ${\ensuremath{\operatorname{E}}}\subset {\ensuremath{\operatorname{D}}}_{\rho_1}$. Soient $\rho_2$ et $\rho_3$ deux vecteurs extrémaux de ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$ adjacents à $\rho_1$, c’est-à-dire il existe un cône $\sigma\in {\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$ de dimension $3$ tel que les vecteurs $\rho_1$, $\rho_2$ et $\rho_3$ sont vecteurs extrémaux de $\sigma$. On remarque que le point générique de ${\ensuremath{\operatorname{E}}}$ n’est pas contenu dans ${\ensuremath{\operatorname{D}}}_{\rho_2}$ ou ${\ensuremath{\operatorname{D}}}_{\rho_3}$.\
Pour un vecteur $m=(a,b,c)\in {\ensuremath{\mathbb{Z}}}^3$, on note $\chi^{m}(t_1,t_2,t_3)=t_1^{a}t_2^{b}t_3^{c}$ le caractère associé à $m$. Soient $U_{\sigma}$ l’ouvert torique de $X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}})$ associé à $\sigma$ et $\chi^{m_i}$ le caractère qui définit une équation de ${\ensuremath{\operatorname{D}}}_{\rho_i}\cap U_{\sigma}$, pour $i\in\{1,2,3\}$. Alors, on a $m_i\cdot \rho_j=\delta_{ij}$, où $\delta_{ij}$ est le symbole de Kronecker. Quitte à remplacer les vecteurs $\rho_2$, $\rho_3$, on peut supposer que $U_{\sigma}\cap E\neq \emptyset$.\
On considère l’unique relèvement $\widehat{\alpha}_{{\ensuremath{\operatorname{E}}}}$ à $X$ du point générique $\alpha_{{\ensuremath{\operatorname{E}}}}$ de $N_{{\ensuremath{\operatorname{E}}}}$ (on rappelle que $X$ est le transformé strict de $S$ dans $X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}})$). On remarque que $\widehat{\alpha}_{{\ensuremath{\operatorname{E}}}}(0)$ est le point générique de ${\ensuremath{\operatorname{E}}}$ et que le ${\ensuremath{\mathbf{k}}}_{\alpha}$-arc $\widehat{\alpha}_{{\ensuremath{\operatorname{E}}}}$ est transverse à ${\ensuremath{\operatorname{E}}}$, c’est-à-dire ${\ensuremath{\operatorname{Ord}_{t} f}}\circ \widehat{\alpha}_{{\ensuremath{\operatorname{E}}}}=1$, où $f$ est une équation locale de ${\ensuremath{\operatorname{E}}}$.\
D’après la Proposition \[pr:cr\_nor-g\], le morphisme $\pi':X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}})\rightarrow {\ensuremath{\mathbb{A}}}^3_{{\ensuremath{\mathbf{k}}}}$ est une résolution plongée de $S$. En particulier $X$ est transverse au diviseur ${\ensuremath{\operatorname{D}}}_{\rho_1}$.Comme le ${\ensuremath{\mathbf{k}}}_{\alpha}$-arc $\widehat{\alpha}_{{\ensuremath{\operatorname{E}}}}$ est transverse à ${\ensuremath{\operatorname{E}}}$, le ${\ensuremath{\mathbf{k}}}_{\alpha}$-arc $\widehat{\alpha}_{{\ensuremath{\operatorname{E}}}}$ est transverse à ${\ensuremath{\operatorname{D}}}_{\rho_1}$. Par conséquent, on obtient que $ m_i\cdot (\mu_x,\mu_y,\mu_z)=\delta_{i 1}$, car $\widehat{\alpha}_{{\ensuremath{\operatorname{E}}}}(0)$ n’est pas contenu dans ${\ensuremath{\operatorname{D}}}_{\rho_2}$ ou ${\ensuremath{\operatorname{D}}}_{\rho_3}$. Ceci implique que $(\mu_x,\mu_y,\mu_z)=\rho_{1}$, d’où la proposition.
\[co:musygemi\] On conserve les hypothèses et notations de la Proposition \[pr:musygemi-g\] et on suppose que $S={\ensuremath{\operatorname{S(p,h_{q})}}}$. Alors, le vecteur $(\mu_x,\mu_y,\mu_z)$ appartient au système générateur minimal du semi-groupe $\tau\cap {\ensuremath{\mathbb{Z}}}^{3}_{\geq 0}$, où $\tau$ est le cône engendré par les vecteurs $(0,0,1)$ et $(p,p,q)$. En particulier, on a $\mu_x=\mu_y\leq p$, $\mu_z\leq q$ et $p\mu_z-q\mu_x\geq 0$.
Le corollaire résulte des Propositions \[pr:sygemi\] et \[pr:musygemi-g\].
Dans la proposition suivante on suppose que $S$ est une hypersurface normale de ${\ensuremath{\mathbb{A}}}^{3}_{{\ensuremath{\mathbf{k}}}}$, donnée par l’équation ${\ensuremath{\operatorname{g}}}=0$, où ${\ensuremath{\operatorname{g}}}$ est un polynôme irréductible non-dégénéré par rapport à la frontière de Newton $\Gamma({\ensuremath{\operatorname{g}}})$. De plus, on suppose que $O$ est l’unique point singulier de $S$.\
On considère un diviseur essentiel ${\ensuremath{\operatorname{E}}}$ sur $S$ et un $K$-wedge admissible, $\omega:{\ensuremath{\operatorname{Spec}}}K[[s,t]]\rightarrow S$, centré en $N_{{\ensuremath{\operatorname{E}}}}$. On pose
$(\eta_x,\eta_y,\eta_z):=({\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\omega^{\star}}}}}(x),{\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\omega^{\star}}}}}(y), {\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\omega^{\star}}}}}(z)).$
On peut écrire le comorphisme de $\omega$ de la façon suivante:
${\ensuremath{\omega^{\star}}}(x)=t^{\eta_x}\chi$; ${\ensuremath{\omega^{\star}}}(y)=t^{\eta_y}\varphi$; ${\ensuremath{\omega^{\star}}}(z)=t^{\eta_z}\psi$,
où $\chi, \varphi$ et $\psi$ sont des séries formelles dans $K[[s,t]]$ qui ne sont pas divisibles par $t$.\
Maintenant, on donne la proposition clé pour la preuve du théorème \[th:nashBP\].
\[pr:invserBP\] Si les séries formelles $\chi$,$\varphi$ et $\psi$ sont inversibles, alors le $K$-wedge admissible $\omega$ centré en $N_{{\ensuremath{\operatorname{E}}}}$ se relève à la résolution minimale de $S$.
On considère une $G$-subdivision régulière ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$ de l’éventail de Newton ${\ensuremath{\operatorname{\Gamma^{\star}(g)}}}$ et on note $\pi':X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}})\rightarrow {\ensuremath{\mathbb{A}}}^{3}_{{\ensuremath{\mathbf{k}}}}$ le morphisme torique induit par la subdivision ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$ du cône ${\ensuremath{\mathbb{R}}}^{3}_{\geq 0}$. On note $X$ le transformé strict de $S$ dans $X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}})$. D’après la proposition \[pr:cr\_nor-g\], le morphisme $\pi':X\rightarrow S$ est une désingularisation de $X$.\
En vertu de la [*version torique du Lemme de Chow*]{} (voir [@Sum74]), il existe une subdivision $\Sigma$ d’éventail ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})_{\mathcal{G}}}}}$ tel que le morphisme torique $\pi'':X(\Sigma)\rightarrow {\ensuremath{\mathbb{A}}}^3_{{\ensuremath{\mathbf{k}}}}$, (le morphisme $\pi''$ est induit par la subdivision $\Sigma$ du cône ${\ensuremath{\mathbb{R}}}^{3}_{\geq 0}$) est un morphisme projectif, la variété $X(\Sigma)$ est quasi-projective et $\pi'$ factorise $\pi''$. Par conséquent, il existe un idéal monomial $\mathcal{I}\subset {\ensuremath{\mathbf{k}}}[x,y,z]$ tel que $X(\Sigma)$ est l’éclatement de ${\ensuremath{\mathbb{A}}}^3_{{\ensuremath{\mathbf{k}}}}$ de centre l’idéal $\mathcal{I}$.\
Comme les séries formelles $\chi$,$\varphi$ et $\psi$ sont inversibles et l’idéal $\mathcal{I}$ est monomial, l’idéal $\omega^{-1}\mathcal{I}\cdot K[[s,t]]$ est inversible. En vertu de la [*propriété universelle de l’éclatement*]{}, le morphisme $\omega$ se relève à $X(\Sigma)$. Par conséquent $\omega$ se relève à $X$. Ceci implique que le $K$-wedge $\omega$ se relève à résolution minimale de $S$.
Maintenant, on donne quelques notions et résultats techniques qui nous permettent de montrer, dans le cas $S={\ensuremath{\operatorname{S(p,h_{q})}}}$, que les séries formelles $\chi$,$\varphi$ et $\psi$ sont inversibles.\
Pour une série non nulle $\phi:=\sum c_{(e_1,e_2)}s^{e_1}t^{e_2}$, où $c_{(e_1,e_2)} \in K$, on définit les applications suivantes:\
- $\nu:{\ensuremath{\mathbb{R}}}^{2}_{>0}\rightarrow {\ensuremath{\mathbb{R}}}_{\geq 0},v\mapsto \nu_v\phi:=\min\{v\cdot e\mid e \in\mathcal{E}(\phi)\}$, où $ \mathcal{E}(\phi)= \{(e_1,e_2)\mid c_{(e_1,e_2)}\neq 0\}$;\
- ${\ensuremath{\operatorname{PPr}}}:{\ensuremath{\mathbb{R}}}^{2}_{>0}\rightarrow K[s,t],v\mapsto\phi_v:=\sum\limits_{{\tiny \mbox{$e\cdot v= \nu_v\phi$}}} c_{(e_1,e_2)}s^{e_1}t^{e_2}$;\
- ${\ensuremath{\operatorname{FI}}}:K[[s,t]]\backslash\{0\}\rightarrow {\ensuremath{\mathbb{Z}}}_{\geq 0}$, où ${\ensuremath{\operatorname{FI}}}(\phi)$ est le nombre de facteurs irréductibles de $\phi$ comptés avec multiplicité.
Un vecteur $v\in {\ensuremath{\mathbb{R}}}^2_{>0}$ définit une graduation positive sur l’anneau $K[[s,t]]$. Cette graduation est appelée $v$-[*graduation*]{}. Pour une série formelle $\phi$, le polynôme $\phi_v$ est la partie principale de $\phi$ pour la $v$-graduation. Le polynôme $\phi_v$ est appelé la $v$-[*partie principale*]{} de $\phi$.\
Soient $\phi$, $\phi'\in K[[s,t]]$ deux séries formelles non nulles. Les séries formelles $\phi$ et $\phi'$ sont associées (resp. non associées) s’il existe (resp. s’il n’existe pas) une série formelle $I\in K[[s,t]]$ inversible tel que $\phi=I\phi'$.\
\[pr:FI-g\] On conserve les hypothèses et notations de la Proposition \[pr:invserBP\]. Alors, il existe un vecteur $v\in {\ensuremath{\mathbb{Q}}}^{2}_{>0} $ tel que:
${\ensuremath{\operatorname{FI}}}(\chi)\leq {\ensuremath{\operatorname{Deg}_{t} \chi_v}}= \nu_v\chi =\mu_x-\eta_x$;
${\ensuremath{\operatorname{FI}}}(\varphi)\leq {\ensuremath{\operatorname{Deg}_{t} \varphi_v}}= \nu_v\varphi= \mu_y-\eta_y$;
${\ensuremath{\operatorname{FI}}}(\psi)\leq {\ensuremath{\operatorname{Deg}_{t} \psi_v}}= \nu_v\psi= \mu_z-\eta_z$.
De plus, $\chi$ (resp. $\varphi$, $\psi$) est inversible si et seulement si $\mu_x-\eta_x=0$ (resp. $\mu_y-\eta_y=0$, $\mu_z-\eta_z=0$).
Soit $\phi\in K[[s,t]]$ une série formelle non nulle et on suppose que
$\phi=I \phi_1^{m_1}\cdots\phi_{n}^{m_n}$, $n\geq 1$,
où les entiers $m_i$ sont strictement positifs et les $\phi_i$ sont des séries formelles irréductibles deux à deux non associées. Alors, on a:
$\phi_v=((\phi_1)_v)^{m_1}\cdots((\phi_{n})_v)^{m_n}$, pour tout $v\in{\ensuremath{\mathbb{R}}}^{2}_{>0}$.
Par conséquent ${\ensuremath{\operatorname{FI}}}(\phi)\leq {\ensuremath{\operatorname{FI}}}(\phi_v)$ pour tout $v\in {\ensuremath{\mathbb{R}}}^{2}_{>0}$. On remarque que les $(\phi_i)_v$, $1\leq i\leq n$, ne sont pas nécessairement irréductibles.\
Dans la suite on cherche un vecteur $v\in {\ensuremath{\mathbb{Q}}}^{2}_{>0} $ tel que
${\ensuremath{\operatorname{FI}}}(\chi_v)={\ensuremath{\operatorname{Deg}_{t} \chi_v}}= \nu_v\chi =\mu_x-\eta_x$;
${\ensuremath{\operatorname{FI}}}(\varphi_v)= {\ensuremath{\operatorname{Deg}_{t} \varphi_v}}= \nu_v\varphi= \mu_y-\eta_y$;
${\ensuremath{\operatorname{FI}}}(\psi_v)={\ensuremath{\operatorname{Deg}_{t} \psi_v}}= \nu_v\psi= \mu_z-\eta_z$.
On rappelle les notations suivantes:\
- $\omega:{\ensuremath{\operatorname{Spec}}}K[[s,t]]\rightarrow S$ est un $K$-wedge admissible centré en $N_{{\ensuremath{\operatorname{E}}}}$;
- $\alpha_{{\ensuremath{\operatorname{E}}}}$ est le point générique de $N_{{\ensuremath{\operatorname{E}}}}$ et $(\mu_x,\mu_y,\mu_z):=({\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(x),{\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(y), {\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(z))$;
- ${\ensuremath{\omega^{\star}}}(x)=t^{\eta_x}\chi$; ${\ensuremath{\omega^{\star}}}(y)=t^{\eta_y}\varphi$; ${\ensuremath{\omega^{\star}}}(z)=t^{\eta_z}\psi$, où $\chi, \varphi$ et $\psi$ sont des séries formelles dans $K[[s,t]]$ qui ne sont pas divisibles par $t$.\
On remarque qu’on peut écrire le comorphisme du $K$-wedge $\omega$ de la façon suivante:
$ (\star) \left \{\begin{array}{ccc}
\omega^{\star}(x) & =&\displaystyle \sum \limits_{\eta_x\leq i<\mu_x} a_i s^{l_i}t^i+\displaystyle \sum \limits_{\mu_x\leq i} a_it^i;\\
\omega^{\star}(y)&=&\displaystyle \sum \limits_{\eta_y\leq j<\mu_y} b_js^{m_j}t^j+\displaystyle \sum \limits_{\mu_y\leq j} b_jt^j;\\
\omega^{\star}(z)&=&\displaystyle \sum \limits_{\eta_z\leq k<\mu_z} c_ks^{n_k}t^k+\displaystyle \sum \limits_{\mu_x\leq k} c_kt^k,
\end{array} \right . $
où les exposants $l_i$ (resp. $ m_j,n_k $) sont strictement positifs et les séries formelles $a_{i}, b_{j}, c_{k} \in K[[s]]$ sont inversibles pour $(i,j,k)\in \{(\eta_x,\eta_y,\eta_z),(\mu_x,\mu_y,\mu_z)\}$ et inversibles ou nulles pour $\eta_x< i< \mu_{x}$, $\eta_y<j < \mu_{y}$, $\eta_z<k< \mu_{z}$. En effet, soit $\lambda_0:{\ensuremath{\operatorname{Spec}}}K[[t]]\rightarrow {\ensuremath{\operatorname{Spec}}}K[[s,t]]$ le morphisme induit par l’homomorphisme canonique $K[[s,t]]\rightarrow K[[s,t]]/(s)=K[[t]]$. On pose $\alpha:=\omega\circ \lambda_0$. Comme $\omega$ est un $K$-wedge admissible centré en $N_{{\ensuremath{\operatorname{E}}}}$ et d’après la propriété fonctorielle de l’espace d’arcs $S_{\infty}$, on a $\alpha=\alpha_{{\ensuremath{\operatorname{E}}}}\circ \lambda_1$, où $\lambda_1:{\ensuremath{\operatorname{Spec}}}K[[s,t]]\rightarrow {\ensuremath{\mathbf{k}}}_{\alpha_{{\ensuremath{\operatorname{E}}}}}[[s,t]]$ est un morphisme induit par une inclusion ${\ensuremath{\mathbf{k}}}_{\alpha_{{\ensuremath{\operatorname{E}}}}}\hookrightarrow K$. Comme $\alpha=\alpha_{{\ensuremath{\operatorname{E}}}}\circ \lambda_1$, on a $({\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha^{\star}}}}}(x),{\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha^{\star}}}}}(y), {\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha^{\star}}}}}(z))=(\mu_x,\mu_y,\mu_z)$, d’où les séries formelles ($\star$).\
Soit $v=(u,1)\in {\ensuremath{\mathbb{Q}}}^{2}_{>0}$. Si $u$ est “assez grand”, alors $\chi_v=at^{\mu_x-\eta_x}$, $\varphi_v=b t^{\mu_y-\eta_y}$ et $\psi_v=ct^{\mu_z-\eta_z}$, où $a$, (resp. $b$ , $c$) est le terme constant de la série formelle inversible $a_{\mu_x}$ (resp. $b_{\mu_y}$, $c_{\mu_z}$). Ceci achève la preuve de la Proposition \[pr:FI-g\].\
On remarque que les série formelle $\chi$ (resp. $\varphi$, $\psi$) est inversible si et seulement si $\chi_v\in K\backslash \{0\}$ (resp. $\varphi_v\in K\backslash \{0\}$, $\phi_v\in K\backslash \{0\}$). Par conséquent $\chi$ (resp. $\varphi$, $\psi$) est inversible si et seulement si $\mu_x-\eta_x=0$ (resp. $\mu_y-\eta_y=0$, $\mu_z-\eta_z=0$).
Dans la proposition suivante, on considère l’hypersurface ${\ensuremath{\operatorname{S(p,h_{q})}}}$ et on majore, en termes des entiers $p$ et $q$, le nombre des facteurs irréductibles comptés avec multiplicité des séries formelles $\chi$, $\varphi$ et $\psi$ qui sont associées au $K$-wedge admissible $\omega:{\ensuremath{\operatorname{Spec}}}K[[s,t]]\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$ centré en $N_{{\ensuremath{\operatorname{E}}}}$.
\[pr:mu-eta\_BP\] On conserve les hypothèses et notations des Propositions \[pr:invserBP\] et \[pr:FI-g\]. De plus, on suppose que $S={\ensuremath{\operatorname{S(p,h_{q})}}}$. Alors, on a:
$\mu_x-\eta_x\leq p-1$, $\mu_y-\eta_y\leq p-1$ et $\mu_z-\eta_z\leq q-1$.
En particulier, on a ${\ensuremath{\operatorname{FI}}}(\chi)\leq p-1$, ${\ensuremath{\operatorname{FI}}}(\varphi)\leq p-1$ et ${\ensuremath{\operatorname{FI}}}(\psi)\leq q-1$.
D’après la Proposition \[pr:FI-g\], si $\mu_x-\eta_x\leq p-1$, $\mu_y-\eta_y\leq p-1$ et $\mu_z-\eta_z\leq q-1$, alors ${\ensuremath{\operatorname{FI}}}(\chi)\leq p-1$, ${\ensuremath{\operatorname{FI}}}(\varphi)\leq p-1$ et ${\ensuremath{\operatorname{FI}}}(\psi)\leq q-1$.\
En vertu du Corollaire \[co:musygemi\], le vecteur $(\mu_x,\mu_y,\mu_z)$ appartient au système générateur minimal du semi-groupe $\tau\cap {\ensuremath{\mathbb{Z}}}^{3}_{\geq 0}$, où $\tau$ est le cône engendré par les vecteurs $(0,0,1)$ et $(p,p,q)$. Par conséquent, on a $\mu_x\leq p$, $\mu_y\leq p$ et $\mu_z\leq q$.
Comme $\omega$ est un $K$-wedge admissible centré en $N_{{\ensuremath{\operatorname{E}}}}$, l’arc générique du $K$-wedge $\omega$ appartient à ${\ensuremath{\operatorname{S(p,h_{q})}}}^{s}_{\infty}$. Par conséquent, on obtient que $\eta_x\geq 1$, $\eta_y\geq 1$ et $\eta_z\geq 1$, d’où $\mu_x-\eta_x\leq p-1$, $\mu_y-\eta_y\leq p-1$ et $\mu_z-\eta_z\leq q-1$.
Dans la proposition suivante, $S$ est l’hypersurface de la Proposition \[pr:invserBP\], c’est-à-dire $S$ est une hypersurface normale de ${\ensuremath{\mathbb{A}}}^{3}_{{\ensuremath{\mathbf{k}}}}$ ayant $O$ comme unique point singulier et qui est donnée par l’équation ${\ensuremath{\operatorname{g}}}=0$, où ${\ensuremath{\operatorname{g}}}$ est un polynôme irréductible non-dégénéré par rapport à la frontière de Newton $\Gamma({\ensuremath{\operatorname{g}}})$.\
On considère un diviseur essentiel ${\ensuremath{\operatorname{E}}}$ sur $S$ et un $K$-wedge $\omega:{\ensuremath{\operatorname{Spec}}}K[[s,t]]\rightarrow S$ admissible centré en $N_{{\ensuremath{\operatorname{E}}}}$. On note
$(\eta_x,\eta_y,\eta_z):=({\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\omega^{\star}}}}}(x),{\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\omega^{\star}}}}}(y), {\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\omega^{\star}}}}}(z)).$
\[pr:etacone-g\] Le vecteur $(\eta_x,\eta_y,\eta_z)$ appartient à l’intersection de ${\ensuremath{\mathbb{Z}}}_{>0}^3$ avec le $2$-squelette $S_2{\ensuremath{\operatorname{\Gamma^{\star}(g)}}}$ de l’éventail ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})}}}$.
Comme $\omega$ est un $K$-wedge admissible centré en $N_{{\ensuremath{\operatorname{E}}}}$, l’arc générique de $\omega$ appartient à $S_{\infty}^s$. Par conséquent, $(\eta_x,\eta_y,\eta_z)$ appartient à ${\ensuremath{\mathbb{Z}}}_{>0}^3$. Il suffit donc de montrer que $(\eta_x,\eta_y,\eta_z)\in S_2{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})}}}$.\
Pour un réel $u\in {\ensuremath{\mathbb{R}}}$, on définit le vecteur suivant:
$\nu_{(u,1)}\omega =(\nu_{(u,1)}\omega^{\star}(x), \nu_{(u,1)}\omega^{\star}(y), \nu_{(u,1)}\omega^{\star}(y))\in {\ensuremath{\mathbb{R}}}^3_{>0}$.
Ce vecteur définit une graduation sur l’anneau ${\ensuremath{\mathbf{k}}}[x,y,z]$. Soit ${\ensuremath{\operatorname{g}}}_u\in {\ensuremath{\mathbf{k}}}[x,y,z]$ la partie principale de ${\ensuremath{\operatorname{g}}}$, par rapport à cette graduation.
Le $K$-wedge $\omega$ doit satisfaire l’équation ${\ensuremath{\operatorname{g}}}=0$, c’est-à-dire on a
${\ensuremath{\operatorname{g}}}(\omega^{\star}(x),\omega^{\star}(y),\omega^{\star}(z))=0$,
ce qui implique que
${\ensuremath{\operatorname{g}}}_u((\omega^{\star}(x))_{(u,1)},(\omega^{\star}(x))_{(u,1)},(\omega^{\star}(x))_{(u,1)})=0$.
Par conséquent, ${\ensuremath{\operatorname{g}}}_{u}$ n’est pas un monôme. Ceci implique que le vecteur $\nu_{(u,1)}\omega$ appartient au $2$-squelette $S^{2}{\ensuremath{\operatorname{\Gamma^{\star}(g)}}}$.
On remarque que $\displaystyle\lim_{u\rightarrow 0} \nu_{(u,1)}\omega=(\eta_x,\eta_y,\eta_z)$. Soit $n> >0$ un entier “assez grand” tel que $\eta_x < n,\;\; \eta_y<n,\;\;\eta_z<n$. Il existe alors un réel $u_0>0$ tel que
$\nu_{(u,1)}\omega \in K_n:=S_{2}{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{g}}})}}}\cap \{(\lambda_1,\lambda_2,\lambda_3)\in {\ensuremath{\mathbb{R}}}_{\geq 0}^3\mid\; \lambda_j\leq n\;\mbox{pour}\; j\in \{1,2,3\}\}$,
pour tout $u\leq u_0$. Comme $K_n$ est compact, on a $ (\eta_x,\eta_y,\eta_z)\in S_{2}{\ensuremath{\operatorname{\Gamma^{\star}(f)}}}$.
\[co:etacone\] On conserve les hypothèses et notations des Propositions \[pr:FI-g\] et \[pr:etacone-g\]. De plus, on suppose que $S={\ensuremath{\operatorname{S(p,h_{q})}}}$. Alors, le vecteur $(\eta_x,\eta_y,\eta_z)$ appartient au semi-groupe $\tau\cap {\ensuremath{\mathbb{Z}}}^{3}_{\geq 0}$, où $\tau$ est le cône engendré par les vecteurs $(0,0,1)$ et $(p,p,q)$. En particulier $\eta_x=\eta_y$ et $p\eta_z-q\eta_x\geq 0$. De plus, si les séries formelles $\chi$,$\varphi$ et $\psi$ ne sont pas simultanément inversibles, alors on a $p\eta_z-q\eta_x> 0$.
Soit $S_2{\ensuremath{\mathbb{R}}}^{3}_{\geq 0}$ le $2$-squelette du cône ${\ensuremath{\mathbb{R}}}^{3}_{\geq 0}$. On rappelle que $\tau_1\in {\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$ (resp. $\tau_2\in {\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$, $\tau_3\in {\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$) est le cône engendré par les vecteurs $(1,0,0)$ (rep. $(0,1,0)$, $(0,0,1)$) et $(p,p,q)$ (voir la Figure \[fi:EN\]). Remarquons que $S_2{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}=\bigcup_{i=1}^{3}\tau_i\cup S_2{\ensuremath{\mathbb{R}}}^{3}_{\geq 0}$. En vertu de la Proposition \[pr:etacone-g\], on a $(\eta_x,\eta_y,\eta_z)\in S_2{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}\cap {\ensuremath{\mathbb{Z}}}_{>0}^{3}$. Ce qui implique que le vecteur $(\eta_x,\eta_y,\eta_z)$ appartient à l’ensemble $\bigcup_{i=1}^{3}\tau_i$.\
D’après la Proposition \[pr:tau12\], le cône $\tau_1$ (resp. $\tau_{3}$) est régulier. Par conséquent, le semi-groupe $\tau_1\cap {\ensuremath{\mathbb{Z}}}^3$ (resp. $\tau_2\cap {\ensuremath{\mathbb{Z}}}^3$) est engendré par les vecteurs $(1,0,0)$ (resp. $(0,1,0)$) et $(p,p,q)$. En particulier, si $(a,b,c)\in \tau_1\cap {\ensuremath{\mathbb{Z}}}_{>0}^3$ (resp. $(a,b,c)\in \tau_2\cap {\ensuremath{\mathbb{Z}}}_{>0}^3$), alors $p\leq a$, $p\leq b$ et $q\leq c$.\
On rappelle que le vecteur $(\mu_x,\mu_y,\mu_z)$ appartient au système générateur minimal du semi-groupe $\tau_3\cap {\ensuremath{\mathbb{Z}}}^3$ (voir le Corollaire \[co:musygemi\]). Par conséquent, on a $\mu_x \leq p$, $\mu_y\leq p$ et $\mu_z\leq q$. Comme $\eta_x\leq \mu_x\leq p$, $\eta_y\leq \mu_y\leq p$ et $\eta_z\leq \mu_z\leq p$, on obtient que le vecteur $(\eta_x,\eta_y,\eta_z)$ appartient au cône $\tau=\tau_3$.\
Si les séries formelles $\chi$,$\varphi$ et $\psi$ ne sont pas simultanément inversibles, alors $p\eta_z-q\eta_x\neq 0$, car si $p\eta_z-q\eta_x=0$, alors $(\eta_x,\eta_y,\eta_z)=(\mu_x,\mu_y,\mu_z)=(p,p,q)$, d’où les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles (voir la Proposition \[pr:FI-g\]).
Dans toute la suite de cette section, on se restreint au cas des hypersurfaces ${\ensuremath{\operatorname{S(p,h_{q})}}}$, où les entiers $p>q\geq 3$ (voir la Remarque \[re:p>q\]) sont premiers entre eux, c’est-à-dire, dans toute la suite, on a:\
- ${\ensuremath{\operatorname{E}}}$ est un diviseur essentiel de ${\ensuremath{\operatorname{S(p,h_{q})}}}$, $p>q\geq 3$;
- $\omega:{\ensuremath{\operatorname{Spec}}}K[[s,t]]\rightarrow {\ensuremath{\operatorname{S(p,h_{q})}}}$ est un $K$-wedge admissible centré en $N_{{\ensuremath{\operatorname{E}}}}$;
- $\alpha_{{\ensuremath{\operatorname{E}}}}$ est le point générique de $N_{{\ensuremath{\operatorname{E}}}}$ et $(\mu_x,\mu_y,\mu_z):=({\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(x),{\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(y), {\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(z))$;
- ${\ensuremath{\omega^{\star}}}(x)=t^{\eta_x}\chi$; ${\ensuremath{\omega^{\star}}}(y)=t^{\eta_y}\varphi$; ${\ensuremath{\omega^{\star}}}(z)=t^{\eta_z}\psi$, où $\chi, \varphi$ et $\psi$ sont des séries formelles dans $K[[s,t]]$ qui ne sont pas divisibles par $t$.\
Soit $\Gamma_{(\mu_x,\mu_z)}$ (resp. $\Gamma_{(p,q)}$) l’enveloppe convexe de $\tau'\cap{\ensuremath{\mathbb{Z}}}_{> 0}^2$, où $\tau'$ est le cône engendré par $(0,1)$ et $(\mu_x,\mu_z)$ (resp. $(0,1)$ et $(p,q)$). La figure suivante donne une idée intuitive de la forme du polyèdre $\Gamma_{(\mu_x,\mu_z)}$.
(0,0) grid (8.5,3);
(-0.2,0) – (8.7,0) node\[right\]; (0,-0.2) – (0,3.2) node\[above\]; (0pt,2pt) – (0pt,-2pt) node\[below\] [$1$]{}; (0pt,2pt) – (0pt,-2pt) node\[below\] [$2$]{};
(0pt,2pt) – (0pt,-2pt) node\[below\] [$\mu_x$]{};
in [1.5, 2,...,3.5]{} (0pt,2pt) – (0pt,-2pt) node\[below\] [$\cdot$]{}; in [4.5, 5,...,8]{} (0pt,2pt) – (0pt,-2pt) node\[below\] [$\cdot$]{}; (0pt,2pt) – (0pt,-2pt) node\[below\] [$p$]{};
(2pt,0pt) – (-2pt,0pt) node\[left\] [$1$]{}; (2pt,0pt) – (-2pt,0pt) node\[left\] [$\cdot$]{}; (2pt,0pt) – (-2pt,0pt) node\[left\] [$\cdot$]{}; (2pt,0pt) – (-2pt,0pt) node\[left\] [$\mu_z$]{};
(2pt,0pt) – (-2pt,0pt) node\[left\] [$\cdot$]{};
(2pt,0pt) – (-2pt,0pt) node\[left\] [$q$]{};
(0,0) – (4,1.5); (4,1.5) – (8,3); (0,0.5) – (1,0.5); (1,0.5) – (4,1.5); (4,1.5) – (8,3)–(0,3) –(0,0.5) – (1,0.5); (0,0)–(0,0.5)–(1,0.5)– (4,1.5); (4,1.5) circle(0.4mm) (0,0.5) circle(0.4mm) (1,0.5) circle(0.4mm) (0,0.5) circle(0.4mm); (2,2) node\[inner sep=-1pt,below=-1pt\] [[ $\Gamma_{(\mu_x,\mu_z)}$]{}]{};
\[pr:Gammapentpos\] Soient $(a_1,b_1)$ et $(a_2,b_2)$, $a_1\leq a_2$, les coordonnés des sommets d’une face compacte du polyèdre $\Gamma_{(\mu_x,\mu_z)}$ et $L$ la droite qui joint les points $(a_1,b_1)$ et $(a_2,b_2)$. Alors la pente de la droite $L$ est positive et strictement plus petite que $\frac{q}{p}$. De plus $\Gamma_{(\mu_x,\mu_z)}=\tau'\cap \Gamma_{(p,q)}$, où $\tau'$ est le cône engendré par les vecteurs $(0,1)$ et $(\mu_x,\mu_z)$.
D’abord on suppose que $(\mu_x,\mu_z)=(p,q)$. On note $\Gamma=\Gamma_{(p,q)}$. Comme $p>q$, le vecteur $(1,1)$ appartient au système générateur minimal du semi-groupe $\tau''\cap {\ensuremath{\mathbb{Z}}}^{2}_{\geq 0}$, où $\tau''$ est le cône engendré par $(0,1)$ et $(p,q)$ (voir la Figure \[fi:Gamma\]). En particulier le vecteur $(1,1)$ appartient à une face compacte du polyèdre $\Gamma$. Comme la pente de la droite qui joint les points $(0,1)$ et $(1,1)$ est nulle et $1< q$, on obtient que la pente de la droite $L$ est un nombre réel positif, car $\Gamma$ est convexe.
En raisonnant par l’absurde, on suppose que
$\dfrac{b_2-b_1}{a_2-a_1}\geq \dfrac{q}{p}$,
d’où $p(b_2-b_1)-q(a_2-a_1)\geq 0$. Ceci implique que le vecteur $(a_2-a_1,b_2-b_1)$ appartient au cône $\tau''$. Or $(a_2,b_2)= (a_2-a_1,b_2-b_1)+(a_1,b_1)$, d’où une contradiction, car le vecteur $(a_2,b_2)$ appartient au système générateur minimal du semi-groupe $\tau''\cap {\ensuremath{\mathbb{Z}}}^{2}_{\geq 0}$.\
On a $\Gamma_{(\mu_x,\mu_z)}=\tau'\cap \Gamma_{(p,q)}$, car la pente de la droite qui joint les points $(0,0)$ et $(\mu_x,\mu_z)$ est plus grande que $\frac{q}{p}$ et les pentes des droites qui définissent les faces compactes de $\Gamma_{(p,q)}$ sont strictement plus petites que $\frac{q}{p}$.
On définit l’application suivante:
${\ensuremath{\operatorname{m}}}:\Gamma_{(\mu_x,\mu_z)} \rightarrow {\ensuremath{\mathbb{R}}}, (u,v)\mapsto pv-qu$.
\[pr:minGamma\] Il existe un vecteur $(u_0,v_0)\in \Gamma_{(\mu_x,\mu_z)}$ tel que
${\ensuremath{\operatorname{m}}}(u_0,v_0)=\inf\{{\ensuremath{\operatorname{m}}}(u,v)\mid (u,v)\in \Gamma_{(\mu_x,\mu_z)}\}$.
De plus on a:
- si $(\mu_x,\mu_z)=(p,q)$, alors $(u_0,v_0)$ appartient au rayon engendré par le vecteur $(p,q)$;
- si $(\mu_x,\mu_z)\neq(p,q)$, alors $(u_0,v_0)=(\mu_x,\mu_z)$.
Le vecteur $(\mu_x,\mu_z)$ étant fixé, on note $\Gamma=\Gamma_{(\mu_x,\mu_z)}$.
Géométriquement, lorsque $c$ croît depuis $-\infty$, les droites $L_c:=\{(u,v)\in {\ensuremath{\mathbb{R}}}^2\mid\;(-q,p)\cdot (u,v)=c\}$ finissent par toucher le bord du polyèdre $\Gamma$ en un point $(u_0,v_0)$. On remarque que ${\ensuremath{\operatorname{m}}}(u_0,v_0)=\inf\{{\ensuremath{\operatorname{m}}}(u,v)\mid (u,v)\in \Gamma\}$ et que l’ensemble $M:=\{(u,v)\in \Gamma \mid {\ensuremath{\operatorname{m}}}(u,v)={\ensuremath{\operatorname{m}}}(u_0,v_0)\}$ est une face ou un sommet de $\Gamma$.\
On rappelle que pour tout vecteur $(u,v)\in \Gamma$ on a $pv-qu\geq 0$. Si $(\mu_x,\mu_z)=(p,q)$, alors $L_0\cap \Gamma$ est la face non compacte engendrée par le vecteur $(p,q)$, d’où le point $i)$ de la proposition. Maintenant, on suppose que $(\mu_x,\mu_z)\neq(p,q)$. La pente de la droite $L'$ qui joint les points $(0,0)$ et $(\mu_x,\mu_z)$ est strictement plus grande que $\frac{q}{p}$, donc toute droite $L_c$ pour $c\in {\ensuremath{\mathbb{R}}}$ intersecte en exactement un point la droite $L'$. Par conséquent, le couple $(u_0,v_0)$ appartient à une face compacte de $\Gamma$. D’après la Proposition \[pr:Gammapentpos\], si $L$ est une droite engendrée par une face de $\Gamma$, alors la pente de $L$ est strictement plus petite que $\frac{q}{p}$. Comme la pente des droites $L_c$, $c\in {\ensuremath{\mathbb{R}}}$, est $\frac{q}{p}$, on obtient que l’ensemble $M$ est un sommet de $\Gamma$.
Soit $n_{\Gamma}$ le nombre de faces compactes de $\Gamma$. On rappelle que $\Gamma$ est l’enveloppe convexe de $\tau'\cap{\ensuremath{\mathbb{Z}}}^{2}_{>0}$, où $\tau'$ est le cône engendré par les vecteurs $(0,1)$ et $(\mu_x,\mu_z)$. Alors, il existe deux suites d’entiers
$ 0=a_0<a_1\cdots <a_{n_{\Gamma}}=\mu_x$ et $ 1=b_0<b_1\cdots <b_{n_{\Gamma}}=\mu_z$,
tels que les couples $(a_i,b_i)$, $0\leq i\leq n_{\Gamma}$, sont les coordonnées des sommets consécutifs de $\Gamma$.
Pour $1\leq i\leq n_{\Gamma}$, on note $L_i$ la droite qui joint les points $(a_{i-1},b_{i-1})$ et $(a_i,b_i)$ et posons $c_i$ (resp $c_0$) le réel tel que la droite $L_{c_i}$ (resp. $L_{c_0}$) intersecte la droite $L_i$ (resp. $L_1$) en le point $(a_i,b_i)$ (resp. $(0,1)$).
En vertu de la Proposition \[pr:Gammapentpos\], on a $c_{i}<c_{i-1}$, pour tout $1\leq i\leq n_{\Gamma}$ (la pente de la droite $L_{i}$, $0\leq i\leq n_{\Gamma}$, est positive et strictement plus petite que $\frac{q}{p}$). Comme ${\ensuremath{\operatorname{m}}}(a_i,b_i)=c_i$, pour tout $0\leq i \leq n_{\Gamma}$, on obtient que ${\ensuremath{\operatorname{m}}}(\mu_x,\mu_z)<{\ensuremath{\operatorname{m}}}(a_i,b_i)$, pour tout $0\leq i \leq n_{\Gamma}-1$. Ceci achève la preuve de la proposition.
\[pr:etaGamma\] Le vecteur $(\eta_x,\eta_z)$ appartient au polyèdre $\Gamma_{(\mu_x,\mu_z)}$.
En vertu du Corollaire \[co:etacone\] le vecteur $(\eta_x,\eta_z)$ appartient au cône $\tau''$ engendré par les vecteurs $(0,1)$ et $(p,q)$. Par conséquent, le vecteur $(\eta_x,\eta_z)$ appartient au polyèdre $\Gamma_{(p,q)}$.
En raisonnant par l’absurde si le vecteur $(\eta_x,\eta_z)$ n’appartient pas au polyèdre $\Gamma_{(\mu_x,\mu_z)}$, alors ce vecteur appartient à l’intersection $\Omega$ du cône $\tau''$ et l’intérieur du triangle définit par les vecteurs $(0,0)$, $(\mu_x,\mu_z)$ et $(\mu_x,0)$ car $\eta_x\leq \mu_x$, $\eta_z\leq \mu_z$ et $\Gamma_{(\mu_x,\mu_z)}$ est l’enveloppe convexe de $\tau'\cap{\ensuremath{\mathbb{Z}}}_{> 0}^2$, où $\tau'$ est le cône engendré par les vecteurs $(0,1)$ et $(\mu_x,\mu_z)$. Mais d’après la Proposition \[pr:Gammapentpos\], l’intersection du polyèdre $\Gamma_{(p,q)}$ et l’ensemble $\Omega$ est vide, d’où une contradiction.
On rappelle qu’on veut montrer que les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles (Proposition \[pr:invserBP\]). Le résultat suivant est une réduction du problème.
\[pr:etaGammapsi\] Si $\chi$ ou $\varphi$ est une série formelle inversible, alors les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles
D’après les résultats \[co:musygemi\], \[pr:FI-g\] et \[co:etacone\], on obtient que la série $\chi$ est inversible si et seulement si $\varphi$ est inversible. Supposons que $\chi$ soit inversible, on a donc $\mu_x=\eta_x$ (Proposition \[pr:FI-g\]). Comme on a $\eta_z \leq \mu_z$ et $\eta_x=\mu_x$, on a $p\eta_z-q\eta_x\leq p\mu_z-q\mu_x$. En vertu des Propositions \[pr:minGamma\] et \[pr:etaGamma\], on obtient que $\mu_z=\eta_z$. La proposition résulte de la Proposition \[pr:FI-g\].
Soit ${\ensuremath{\operatorname{h_q}( x,y)}}=\prod _{i=1}^{q}(a_ix+b_iy)$ la décomposition en facteurs irréductibles de ${\ensuremath{\operatorname{h_q}}}$. Le $K$-wedge $\omega$ doit satisfaire l’équation $z^p=-{\ensuremath{\operatorname{h_q}( x,y)}}$ donc:
$t^{p\eta_{z}-q\eta_x}\psi^p=-{\ensuremath{\operatorname{h_q}( \chi,\varphi)}}=-\prod _{i=1}^{q}\gamma_{i},$ où $\gamma_{i}:=a_i\chi+b_i\varphi$.
Les combinaisons linéaires de $\chi$ et $\varphi$ données par les $\gamma_i$ plus l’hypothèse sur les facteurs irréductibles de ${\ensuremath{\operatorname{h_q}}}$ permettent de montrer le lemme suivant.
\[le:pgcdninv\] Soit $\lambda:={\ensuremath{\operatorname{p.g.c.d(\gamma_{1},\gamma_{2})}}}$. Alors ${\ensuremath{\operatorname{p.g.c.d(\gamma_{i},\gamma_{j})}}}=\lambda I_{ij}$, où $I_{ij}$ est inversible pour tous les entiers $1\leq i<j\leq q$. De plus, si $\lambda$ est inversible, alors $\chi$, $\varphi$ et $\psi$ sont inversibles.
Soit $\lambda_0:={\ensuremath{\operatorname{p.g.c.d(\gamma_{i_0},\gamma_{j_0})}}}$, où $i_0$ et $j_0$ sont deux entiers tels que $1\leq i_0<j_0\leq q$. Si $\lambda_0$ n’est pas inversible, alors $\lambda_0$ divise $\chi$ et $\varphi$. Par conséquent, $\lambda_0$ divise ${\ensuremath{\operatorname{p.g.c.d(\gamma_{i},\gamma_{j})}}}$ pour tous les entiers $i$, $j$ tels que $1\leq i<j\leq q$. Ce qui achève la première partie du lemme.\
Pour la deuxième partie de la proposition on suppose que la série formelle $\lambda$ est inversible.
Raisonnons par l’absurde. En vertu de la Proposition \[pr:etaGammapsi\], les séries $\chi$ et $\varphi$ ne sont pas inversibles. Par conséquent, la série $\gamma_i$, $1\leq i\leq q$, n’est pas inversible. On rappelle que les séries formelles $\chi$, $\varphi$ et $\psi$ satisfont la relation suivante:
$t^{p\eta_{z}-q\eta_x}\psi^p=-{\ensuremath{\operatorname{h_q}( \chi,\varphi)}}=-\prod _{i=1}^{q}\gamma_{i},$ où $\gamma_{i}:=a_i\chi+b_i\varphi$.
Comme $\chi$ et $\varphi$ ne sont pas divisibles par $t$, $t$ divise $\gamma_i$ si et seulement si $t$ ne divise pas $\gamma_j$ pour tout $j\neq i$. Quitte à re-numéroter les $\gamma_i$, on peut supposer que $t^{p\eta_z-q\eta_x}$ divise $\gamma_1$. Soit $\gamma_{1}=-t^{p\eta_{z}-q\eta_x}\gamma'_{1}$.
Comme ${\ensuremath{\operatorname{p.g.c.d(\gamma_{i},\gamma_{j})}}}$ est inversible pour tout $2\leq i<j\leq q$, la série formelle $\gamma'_1\prod _{i=2}^{q}\gamma_{i}$ a au moins $q-1$ facteurs irréductibles deux à deux non associés. Comme $\psi^p= \gamma'_1\prod _{i=2}^{q}\gamma_{i}$, la série formelle $\gamma'_1\prod _{i=2}^{q}\gamma_{i}$ est le produit de $q-1$ puissances de séries formelles irréductibles deux à deux non-associées, car ${\ensuremath{\operatorname{FI}}}(\psi)\leq q-1$ (Proposition \[pr:mu-eta\_BP\]). Par conséquent, on obtient que $\gamma'_1$ est inversible, que la série $\gamma_i$, $2\leq i\leq q$, est une puissance d’une série formelle irréductible et que $\psi=\prod_{i=1}^{q-1}\psi_i$, où les $\psi_{i}$ sont des séries formelles irréductibles deux à deux non associées. Par conséquent, on peut supposer que $\gamma_{i+1}=\psi_i^pI_i$ pour $1\leq i\leq q-1$, où les séries formelles $I_i$ sont inversibles. Comme $\gamma_i=a_i\chi+b_i\varphi$, $1\leq i\leq q$, et $q\geq 3$, il existe deux constantes $a, b\in K$ telles que
$t^{p\eta_{z}-q\eta_x}\gamma'_{1}=aI_{1}\psi_{1}^{p}+bI_{2}\psi_{2}^{p}$.
On rappelle qu’une série formelle dans $K[[s,t]]$ est inversible si et seulement si elle l’est dans $\overline{K}[[s,t]]$, où $\overline{K}$ est la clôture algébrique de $K$. Soient $J_1$ et $J_2$ deux séries formelles inversibles dans $\overline{K}[[s,t]]$ telles que $J_1^p=aI_1$ et $J_2^p=bI_2$. Ainsi, on obtient que:
$t^{p\eta_{z}-q\eta_x}\gamma'_{1}=\prod_{i=1}^{p}(J_1\psi_1+w_iJ_2\psi_2)$,
où les $w_i$ sont les racines $p$-ièmes de l’unité. Mais $\gamma'_1$ est inversible et $p\eta_{z}-q\eta_x>0$ (Proposition \[co:etacone\]), donc $\psi_{1}$ et $\psi_{2}$ sont inversibles ou divisibles par $t$, ce qui est absurde.
La proposition suivante achève la preuve du Théorème \[th:nashBP\] (voir la Proposition \[pr:invserBP\]).
\[pr:NashBP\] Les séries formelles $\chi$, $\varphi$, $\psi$ sont inversibles.
En raisonnant par l’absurde, on suppose les séries formelles $\chi$, $\varphi$ et $\psi$ non toutes inversibles. D’après le corollaire \[co:etacone\], on a $p\eta_z-q\eta_x>0$.\
On rappelle que les séries formelles $\chi$, $\varphi$ et $\psi$ satisfont la relation suivante:
$t^{p\eta_{z}-q\eta_x}\psi^p=-{\ensuremath{\operatorname{h_q}( \chi,\varphi)}}=-\prod _{i=1}^{q}\gamma_{i},$ où $\gamma_{i}:=a_i\chi+b_i\varphi$.
Comme $\chi$ et $\varphi$ ne sont pas divisibles par $t$, $t$ divise $\gamma_i$ si et seulement si $t$ ne divise pas $\gamma_j$ pour tout $j\neq i$. Quitte à re-numéroter les $\gamma_i$, on peut supposer que $t^{p\eta_z-q\eta_x}$ divise $\gamma_1$.\
D’après le Lemme \[le:pgcdninv\], il existe des séries formelles $\gamma'_i$ telles que:
$\gamma_{1}=-t^{p\eta_{z}-q\eta_x}\gamma'_{1}\lambda$ et $\gamma_{i}=\gamma'_{i}\lambda$ pour $2\leq i\leq q$,
où $\lambda$ n’est pas inversible et ${\ensuremath{\operatorname{p.g.c.d(\gamma'_{i},\gamma'_{j})}}}$ est inversible pour $1\leq i<j\leq q$.\
Le lemme suivant est le résultat clé pour la preuve de la Proposition \[pr:NashBP\].\
Dans toute la suite $v$, désigne le vecteur de la Proposition \[pr:FI-g\].
\[le:lambda\] Le $v$-ordre de $\lambda$ est $\mu_{x}-\eta_{x}$, c’est-à-dire $\nu_v{\lambda}=\mu_{x}-\eta_{x}$.
D’abord finissons la preuve de la Proposition \[pr:NashBP\]. On rappelle que $\eta_y=\eta_x\leq \mu_x=\mu_y\leq p$, $\eta_z\leq \mu_z\leq q $ (voir le Corollaire \[co:musygemi\]) et que $(\eta_x,\eta_z)$ appartient à l’enveloppe convexe $\Gamma_{(\mu_x,\mu_z)}$ (voir la Proposition \[pr:etaGamma\]) de l’ensemble $\tau'\cap{\ensuremath{\mathbb{Z}}}_{>0}^2$, où $\tau'$ est le cône engendré par $(0,1)$ et $(\mu_x,\mu_z)$. Soit $\gamma'=\prod_{i=1}^{q} \gamma'_{i}$, d’où $\psi^p=\gamma'\lambda^q$. D’après le Lemme \[le:lambda\], on a $\nu_v\gamma'=p\mu_{z}-q\mu_{x} -(p\eta_{z}-q\eta_{x})$. Par définition $\nu_v\gamma'\geq 0.$ Or $\nu_v\gamma'\geq 0$ si et seulement si $(\eta_x,\eta_z)=(\mu_x,\mu_z)$ (Proposition \[pr:minGamma\]), d’où une contradiction. En effet, si $(\eta_x,\eta_z)=(\mu_x,\mu_z)$ , alors les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles (voir les Propositions \[pr:FI-g\] et \[pr:etaGammapsi\]).\
[*Démonstration du Lemme \[le:lambda\]*]{} On rappelle que $\gamma_{i}=a_i\chi+b_i\varphi$ pour $1\leq i\leq q$. Alors, il existe au plus un $1\leq i_0\leq q$ tel que les $v$-parties principales $\chi_v$, $\varphi_v$ satisfont la relation $a_{i_0}\chi_v+b_{i_0}\varphi_v=0$. En particulier on a $\nu_{v}\gamma_{i}=\mu_x-\eta_x$ pour tout $1\leq i\leq q$ tel que $i\neq i_{0}$. Par conséquent, $\nu_v \lambda\leq \mu_x-\eta_x$.\
Si $\nu_v \lambda< \mu_x-\eta_x$, alors $\nu_v\gamma'_{i}>0$, pour tout $2\leq i\leq q$. Ceci implique que les $\gamma'_{i}$, pour $2\leq i\leq q$, ne sont pas inversibles.
Comme ${\ensuremath{\operatorname{p.g.c.d(\gamma'_{i},\gamma'_{j})}}}$ est inversible pour tous les entiers $i$, $j$ tels que $2\leq i<j\leq q$, la série formelle $\gamma'=\prod_{i=1}^{q} \gamma'_{i}$ a au moins $q-1$ facteurs irréductibles deux à deux non associés. Comme on a $\psi^p=\gamma'\lambda^q$, la série formelle $\gamma'$ est le produit de $q-1$ puissances de séries formelles irréductibles non-associées, car ${\ensuremath{\operatorname{FI}}}(\psi)\leq q-1$ (Proposition \[pr:mu-eta\_BP\]). Par conséquent, on obtient que $\gamma'_1$ est inversible, que la série $\gamma'_i$, $2\leq i\leq q$, est une puissance d’une série formelle irréductible, que $\psi=\prod_{i=1}^{q-1}\psi_i$, où les $\psi_{i}$ sont des séries formelles irréductibles deux à deux non associées, et que $\mu_z-\eta_z=q-1$. De plus, on a $(\mu_x,\mu_y,\mu_z)=(p,p,q)$. Ceci implique que $p\not\equiv 1\mod{q}$, car si $p\equiv 1\mod{q}$, alors le diviseur ${\ensuremath{\operatorname{E}}}_{0}={\ensuremath{\operatorname{D}}}_{(p,p,q)}\cap{\ensuremath{\operatorname{S_{\mathcal{G}}}}}$ n’est pas un diviseur essentiel (voir la Proposition \[pr:sygemi\] et le Corollaire \[co:Gdes.r=1\]).\
On rappelle qu’une série formelle dans $K[[s,t]]$ est inversible si et seulement si elle est inversible dans $\overline{K}[[s,t]]$, où $\overline{K}$ est la clôture algébrique de $K$. Dans la suite on suppose que le corps $K$ est algébriquement clos.\
On fixe un entier $1\leq i\leq q-1$ quelconque. On peut donc supposer que $\psi^p_{i}=\gamma'_{i+1}\lambda_i^{q}$, où $\xi\lambda=\prod_{j=1}^{q-1}\lambda_j$, $\xi^q=\gamma'_1$ et ${\ensuremath{\operatorname{p.g.c.d(\lambda_j,\lambda_{j'})}}}$ est inversible pour tous les entiers $j$, $j'$ tels que $1\leq j<j'\leq q-1$. Comme $\psi_i$ est irréductible et $\gamma'_{i+1}$ n’est pas inversible, il existe deux entiers $l_i\geq 1$ et $m_i\geq 0$ tels que $\gamma'_{i+1}=I_i\psi^{l_i}$ et $\lambda_i=I_i^{-1}\psi_i^{m_i}$, où $I_i$ est une série formelle inversible. Comme $\nu_v\psi_i=1$, on a $\nu_v\lambda_i=m_i$, $\nu_v\gamma'_{i+1}=l_i$ et $l_i+qm_i=p$.\
Comme $(\mu_x,\mu_y,\mu_z)=(p,p,q)$, on a $t^{p\eta_z-q\eta_x}\psi_v=-{\ensuremath{\operatorname{h_q}( \chi_{ {\it v}},\varphi_{{\it v}})}}$ (Proposition \[pr:FI-g\]), d’où $\nu_v\gamma_{j}=\mu_x-\eta_x$ pour tout $1\leq j\leq q$. Par conséquent, pour tout $1\leq j\leq q-1$, on a $l_j=\nu_v\gamma'_{j+1}=\nu_v \gamma'_2=l_1$, car $\gamma_{j+1}=\gamma'_{j+1}\lambda$. En particulier, on a $m_j=m_1$, pour tout $1\leq j\leq q-1$, car $l_1+qm_j=p$. Ainsi, on obtient que $\gamma_{j+1}=I_{j}\psi_{j}^{p-m_1q}\lambda$, pour $1\leq j\leq q-1$, où les $I_{j}$ sont inversibles. Comme $q\geq 3$ et $\gamma_{j}:=a_j\chi+b_j\varphi$, il existe deux constantes $a, b\in K$ telles que
$t^{p\eta_{z}-q\eta_x}\gamma'_{1}=aI_{1}\psi_{1}^{p-m_1q}+bI_{2}\psi_{2}^{p-m_1q}$.
Soient $J_1$ et $J_2$ deux séries formelles inversibles telles que $J_1^{p-m_1q}=aI_1$ et $J_2^{p-m_1q}=bI_2$. Ainsi, on obtient que:
$t^{p\eta_{z}-q\eta_x}\gamma'_{1}=\prod_{i=1}^{p-m_1q}(J_1\psi_1+w_iJ_2\psi_2)$,
où les $w_i$ sont les racines $(p-m_1q)$-ièmes de l’unité. Mais $\gamma'_1$ est inversible, $p\eta_{z}-q\eta_x>0$ et $p\not\equiv 1 \mod q$, donc $\psi_{1}$ et $\psi_{2}$ sont inversibles ou divisibles par $t$, ce qui est absurde.
Preuve de la bijectivité de l’application de Nash pour les singularités de type ${\ensuremath{\mathbb{E}}}_6$ et ${\ensuremath{\mathbb{E}}}_7$
===============================================================================================================================================
La singularité de type ${\ensuremath{\mathbb{E}}}_6$
----------------------------------------------------
Dans cette section, on démontre la bijectivité de l’application de Nash pour la singularité de type ${\ensuremath{\mathbb{E}}}_6$, ce qui équivaut à montrer que tous les wedges admissibles se relèvent à la résolution minimale de ${\ensuremath{\mathbb{E}}}_6$ (voir [@Reg06]).\
Soit $S$ l’hypersurface normale de $ {\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ donnée par l’équation $x^2+y^3+z^4=0$. L’hypersurface $S$ a un unique point singulier de type ${\ensuremath{\mathbb{E}}}_6$ à l’origine de ${\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$.\
Notons ${\ensuremath{\operatorname{f}}}=x^2+y^3+z^4$; on considère l’éventail de Newton ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$ associé à ${\ensuremath{\operatorname{f}}}$. Soit $H$ un plan de ${\ensuremath{\mathbb{R}}}^3$ qui ne contient pas l’origine de ${\ensuremath{\mathbb{R}}}^3$ et tel que l’intersection de $H$ et ${\ensuremath{\mathbb{R}}}^3_{\geq 0}$ soit un ensemble compact. La Figure \[fi:ENE6\] représente l’intersection de $H$ avec la subdivision ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$ de ${\ensuremath{\mathbb{R}}}^3_{\geq 0}$. Chaque sommet du diagramme est identifié avec le [*vecteur extrémal*]{} correspondant. On note $\tau_1$ (resp. $\tau_2$, $\tau_3$) le cône engendré par les vecteurs $(1,0,0)$ (resp. $(0,1,0)$, $(0,0,1)$) et $(6,4,3)$.
(0,0) – (330:3) (0,0) – (90:3) (0,0) – (210:3); (330:3) – (90:3) – (210:3) – (330:3); (0,-0.4) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(6,4,3)$]{}]{}; (335:3.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(0,1,0)$]{}]{}; (340:1.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [$\tau_2$]{}; (205:3.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(1,0,0)$]{}]{}; (200:1.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [$\tau_1$]{}; (90:3.3) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(0,0,1)$]{}]{}; (80:1.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [$\tau_3$]{}; (0,0) circle(0.7mm) (330:3) circle(0.7mm) (90:3) circle(0.7mm) (210:3) circle(0.7mm);
Soit ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}}$ une $G$-[*subdivision régulière*]{} de ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$. On note $\pi_{\mathcal{N}}:X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}})\rightarrow {\ensuremath{\mathbb{A}}}^{3}_{{\ensuremath{\mathbf{k}}}}$ (resp. $\pi_{\mathcal{G}}:X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}})\rightarrow X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}})$) le morphisme torique induit par la subdivision ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$ de ${\ensuremath{\mathbb{R}}}^3_{\geq 0}$ (resp. ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}}$ de ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$) et $S_{\mathcal{G}}$ le transformé strict de $S$ associé au morphisme $\pi:=\pi_{\mathcal{G}}\circ \pi_{\mathcal{N}}$.\
La proposition suivante est un analogue de la Proposition \[pr:Gdes1\].
\[pr:desingE6\] $S_{\mathcal{G}}$ est la résolution minimale de $S$.
Soient ${\ensuremath{\operatorname{E}}}$ un diviseur essentiel sur $S$ (${\ensuremath{\operatorname{E}}}\in {\ensuremath{\operatorname{Ess}}}(S))$ et $\alpha_{{\ensuremath{\operatorname{E}}}}$ le point générique de $N_{{\ensuremath{\operatorname{E}}}}$. On note
$(\mu_x,\mu_y,\mu_z):=({\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(x),{\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(y), {\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(z))$,
où ${\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}$ est le comorphisme de $\alpha_{{\ensuremath{\operatorname{E}}}}$.\
Pour un cône $\tau$ dans ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$ on note $G\tau$ le système générateur minimal du semi-groupe $\tau\cap {\ensuremath{\mathbb{Z}}}^3$.\
La proposition suivante est un analogue du Corollaire \[co:musygemi\].
\[pr:musygemiE6\] Le vecteur $(\mu_x,\mu_y,\mu_z)$ appartient à l’union de $G\tau_2$ et $G\tau_3$, où $\tau_2$ (resp. $\tau_3$) est le cône engendré par les vecteurs $(0,1,0)$ (resp. $(0,0,1)$) et $(6,4,3)$ (voir la figure \[fi:ENE6\]). Autrement dit, $(\mu_x,\mu_y,\mu_z)\in \{(2,2,1),(3,2,2),(4,3,2),(6,4,3)\}$.
Dans la suite, on montre que pour chaque diviseur essentiel ${\ensuremath{\operatorname{E}}}$ tous les $K$-wedges admissibles centrés en $N_{{\ensuremath{\operatorname{E}}}}$ se relèvent à la résolution minimale de $S$.\
Soit ${\ensuremath{\operatorname{E}}}\in {\ensuremath{\operatorname{Ess}}}(S)$ et on considère un $K$-wedge $\omega:{\ensuremath{\operatorname{Spec}}}K[[s,t]]\rightarrow S$ admissible centré en $N_{{\ensuremath{\operatorname{E}}}}$. On pose:
$(\eta_x,\eta_y,\eta_z): =({\ensuremath{\operatorname{Ord}_{t} \omega}}^{\star }(x),{\ensuremath{\operatorname{Ord}_{t} \omega}}^{\star }(y),{\ensuremath{\operatorname{Ord}_{t} \omega}}^{\star }(z))$.
On peut écrire le comorphisme de $\omega$ de la façon suivante:
$\omega^{\star }(x)=t^{\eta_x}\chi,\;\omega^{\star }(y)=t^{\eta_y}\varphi,\; \omega^{\star }(z)=t^{\eta_z}\psi,$
où les séries formelles $\chi$, $\varphi$, $\psi$ ne sont pas divisibles par $t$. On rappelle que $(\eta_x,\eta_y,\eta_z)\in{\ensuremath{\mathbb{Z}}}^{3}_{>0}$, car $\omega$ est un $K$-wedge admissible.\
\[re:221eta643\] En vertu de la Proposition \[pr:musygemiE6\], on a $\mu_x\leq 6$, $\mu_y\leq 4$ et $\mu_z\leq 3$. En particulier, on a $\eta_x\leq 6$, $\eta_y\leq 4$ et $\eta_z\leq 3$, car $\eta_x\leq \mu_x$, $\eta_y\leq \mu_y$ et $\eta_z\leq \mu_z$.
Le $K$-wedge $\omega$ doit satisfaire l’équation $x^2+y^3+z^{4}=0$, d’où la relation suivante:
$$t^{2\eta_x}\chi^2+t^{3\eta_y}\varphi^3+t^{4\eta_z}\psi^{4}=0.$$
Ce qui implique que $2\leq \eta_x\leq 6$, $2\leq \eta_y\leq 4$ et $1\leq \eta_z\leq 3$.
\[re:plan-dem-E6\] D’après la Proposition \[pr:invserBP\], si les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles, alors le $K$-wedge admissible $\omega$ centré en $N_{{\ensuremath{\operatorname{E}}}}$ se relève à la résolution minimale de $S$. En raisonnant par l’absurde, si au moins l’une d’elles n’est pas inversible on obtient deux cas pour le vecteur $(\eta_x,\eta_y,\eta_z)$ (voir la Proposition \[pr:etaconeE6\]), ensuite on considère chaque cas séparément pour obtenir des contradictions (voir les Propositions \[pr:2e6\] et \[pr:1e6\]), d’où le cas ${\ensuremath{\mathbb{E}}}_6$ du Théorème \[th:nashDP\].\
On rappelle les définitions suivantes:\
Pour une série non nulle $\phi:=\sum c_{(e_1,e_2)}s^{e_1}t^{e_2}$, où $c_{(e_1,e_2)} \in K$, on définit les applications suivantes:\
- $\nu:{\ensuremath{\mathbb{R}}}^{2}_{>0}\rightarrow {\ensuremath{\mathbb{R}}}_{\geq 0},v\mapsto \nu_v\phi:=\min\{v\cdot e\mid e \in\mathcal{E}(\phi)\}$, où $ \mathcal{E}(\phi)= \{(e_1,e_2)\mid c_{(e_1,e_2)}\neq 0\}$;\
- ${\ensuremath{\operatorname{PPr}}}:{\ensuremath{\mathbb{R}}}^{2}_{>0}\rightarrow K[s,t],v\mapsto\phi_v:=\sum\limits_{{\tiny \mbox{$e\cdot v= \nu_v\phi$}}} c_{(e_1,e_2)}s^{e_1}t^{e_2}$;\
- ${\ensuremath{\operatorname{FI}}}:K[[s,t]]\backslash\{0\}\rightarrow {\ensuremath{\mathbb{Z}}}_{\geq 0}$, où ${\ensuremath{\operatorname{FI}}}(\phi)$ est le nombre de facteurs irréductibles de $\phi$ comptés avec multiplicité.\
Le réel $\nu_v\phi$ (resp. Le polynôme $\phi_v$) est appelé le $v$-[*ordre*]{} (resp. la $v$-[*partie principale*]{}) de $\phi$.\
D’après la Proposition \[pr:musygemiE6\] et la Remarque \[re:221eta643\], on a:\
$\mu_x-\eta_x\leq 4$, $\mu_y-\eta_y\leq 2$ et $\mu_z-\eta_z\leq 2$.\
En vertu de la Proposition \[pr:FI-g\], on peut majorer le nombre de facteurs irréductibles comptés avec multiplicité des séries formelles $\chi$, $\varphi$ et $\psi$ à l’aide des $v$-ordres. Plus précisément, il existe un vecteur $v\in {\ensuremath{\mathbb{Q}}}^{2}_{>0} $ tel que:\
- ${\ensuremath{\operatorname{FI}}}(\chi)\leq {\ensuremath{\operatorname{Deg}_{t} \chi_v}}= \nu_v\chi =\mu_x-\eta_x\leq 4$;
- ${\ensuremath{\operatorname{FI}}}(\varphi)\leq {\ensuremath{\operatorname{Deg}_{t} \varphi_v}}= \nu_v\varphi= \mu_y-\eta_y\leq 2$;
- ${\ensuremath{\operatorname{FI}}}(\psi)\leq {\ensuremath{\operatorname{Deg}_{t} \psi_v}}= \nu_v\psi= \mu_z-\eta_z\leq 2$.\
Sauf mention du contraire, dans toute la suite le vecteur $v\in {\ensuremath{\mathbb{Q}}}_{>0}^2$ satisfait la propriété ci-dessus.\
On remarque qu’une série formelle $\phi \in K[[s,t]]$ est inversible dans $K[[s,t]]$ si et seulement si elle est inversible dans $\overline{K}[[s,t]]$, où $\overline{K}$ est la clôture algébrique de $K$. Dans toute la suite on suppose que le corps $K$ est algébriquement clos.\
Maintenant, on prouve la première des trois propositions qu’on a anticipé dans la Remarque \[re:plan-dem-E6\].
\[pr:etaconeE6\] S’il existe au moins une série formelle parmi les séries $\chi$, $\varphi$, $\psi$ qui n’est pas inversible, alors le vecteur $(\eta_x,\eta_y,\eta_z)\in{\ensuremath{\mathbb{Z}}}^{3}_{>0}$ satisfait une des relations suivantes:
- $2\eta_x=3\eta_y$ et $\eta_x<2\eta_z$;
- $\eta_x=2\eta_z$ et $2\eta_x<3\eta_y$.
Soit $S_2{\ensuremath{\mathbb{R}}}^3_{\geq 0}$ le $2$-squelette du cône ${\ensuremath{\mathbb{R}}}^3_{\geq 0}$. On rappelle que $\tau_1\in {\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$ (resp. $\tau_2\in {\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$, $\tau_3\in {\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$) est le cône engendré par les vecteurs $(1,0,0)$ (resp. $(0,1,0)$, $(0,0,1)$) et $(6,4,3)$ (voir la Figure \[fi:ENE6\]). Remarquons que $S_2{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}=\bigcup_{i=1}^{3}\tau_i\cup S_2{\ensuremath{\mathbb{R}}}^3_{\geq 0}$.\
En vertu de la Proposition \[pr:etacone-g\], on a $(\eta_x,\eta_y,\eta_z)\in S_{2}{\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}\cap {\ensuremath{\mathbb{Z}}}^{3}_{>0}$. Ce qui implique que le vecteur $(\eta_x,\eta_y,\eta_z)$ appartient à l’ensemble $\bigcup_{i=1}^{3}\tau_i$.\
On remarque que $(\eta_x,\eta_y,\eta_z)\neq (6,4,3)$. En effet, si $(\eta_x,\eta_y,\eta_z)=(6,4,3)$ alors $(\eta_x,\eta_y,\eta_z)=(\mu_x,\mu_y,\mu_z)$ (voir la Remarque \[re:221eta643\]), d’où les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles (Proposition \[pr:FI-g\]). Ceci rentre en contradiction avec les hypothèses de la Proposition \[pr:etaconeE6\]. Par conséquent, le vecteur $(\eta_x,\eta_y,\eta_z)$ appartient à l’ensemble $\bigcup_{i=1}^{3}\tau_i^0$, où $\tau_i^0$ est l’intérieur relatif du cône $\tau_i$, $1\leq i\leq 3$.\
Par un calcul direct, on montre que le cône $\tau_1$ est régulier. Par conséquent, le semi-groupe $\tau_1\cap {\ensuremath{\mathbb{Z}}}^{3}$ est engendré par les vecteurs $(1,0,0)$ et $(6,4,3)$. En particulier, si $(a,b,c)\in \tau_1\cap {\ensuremath{\mathbb{Z}}}^{3}_{>0}$, alors $6\leq a$, $4\leq b$ et $3\leq c$. Ce qui implique que le vecteur $(\eta_x,\eta_y,\eta_z)$ appartient à l’ensemble $\tau^0_2\cup \tau^0_3$.\
Si $(\eta_x,\eta_y,\eta_z)$ appartient à l’ensemble $\tau^0_2$ (resp. $\tau^0_3$), alors $\eta_x=2\eta_z$ et $2\eta_x<3\eta_y$ (resp. $2\eta_x=3\eta_y$ et $\eta_x<2\eta_z$), d’où la proposition.
Maintenant, on démontre quelques résultats techniques.\
On rappelle que le $K$-wedge $\omega$ doit satisfaire l’équation $x^2+y^3+z^{4}=0$, d’où la relation suivante:
$$t^{2\eta_x}\chi^2+t^{3\eta_y}\varphi^3+t^{4\eta_z}\psi^{4}=0. \label{eq:1}\\$$
Cette relation est utilisée dans plusieurs endroits de la démonstration du Théorème \[th:nashDP\].
\[le:643\] Si le vecteur $(\mu_x,\mu_y,\mu_z)$ est égal au vecteur (6,4,3), alors les $v$-ordres des séries formelles $t^{\eta_x}\chi+it^{2\eta_z}\psi^2$ et $t^{\eta_x}\chi-it^{2\eta_z}\psi^2$ sont égaux à $6$, c’est-à-dire on a:
$$\nu_v(t^{\eta_x}\chi+it^{2\eta_z}\psi^2)=\nu_v(t^{\eta_x}\chi-it^{2\eta_z}\psi^2)=6.$$
Dans ce cas, on a $\nu_v(t^{2\eta_x}\chi^2)=\nu_v(t^{3\eta_y}\varphi^3)=\nu_v(t^{4\eta_z}\psi^4)=12$, donc les $v$-parties principales $\chi_v$, $\varphi_v$ et $\psi_v$ satisfont la relation suivante (voir la Relation (\[eq:1\])):
$$t^{3\eta_y}\varphi_v^3+(t^{\eta_x}\chi_v+it^{2\eta_z}\psi_v^2)(t^{\eta_x}\chi_v-it^{2\eta_z}\psi_v^2)=0.$$
La relation ci-dessus implique que $(t^{\eta_x}\chi_v+it^{2\eta_z}\psi_v^2)(t^{\eta_x}\chi_v-it^{2\eta_z}\psi_v^2)\neq0$ et par conséquent, on obtient que les $v$-ordres des séries formelles $t^{\eta_x}\chi+it^{2\eta_z}\psi^2$ et $t^{\eta_x}\chi-it^{2\eta_z}\psi^2$ sont égaux à $6$, car les $v$-parties principales $t^{\eta_x}\chi_v+it^{2\eta_z}\psi_v^2$ et $ t^{\eta_x}\chi_v-it^{2\eta_z}\psi_v^2$ sont différentes de zéro et les $v$-ordres des séries formelles $t^{\eta_x}\chi$ et $t^{2\eta_z}\psi^{2}$ sont égaux à $6$.
\[le:pre0E6\] S’il existe une série formelle irréductible $\lambda$ qui divise $\chi$,$\varphi$, et $\psi$, alors $\lambda^2$ divise $\varphi$.
On suppose que $\lambda$ divise $\chi$, $\varphi$, et $\psi$. Soient $\chi=\lambda\chi_1$, $\varphi=\lambda\varphi_1$ et $\psi=\lambda\psi_1$. Alors au moyen de la Relation (\[eq:1\]) on obtient la relation suivante:
$$t^{2\eta_x}\chi_1^2+t^{3\eta_y}\lambda\varphi_1^3 +t^{4\eta_z}\lambda^2\psi_1^{4}=0,$$ ce qui implique que $\lambda$ divise $\chi_1$. En particulier $\lambda$ divise $\varphi_1$.
\[le:varinv\] Si la série formelle $\varphi$ est inversible, alors les séries formelles $\chi$ et $\psi$ sont inversibles.
Raisonnons par l’absurde. On suppose que $\varphi$ est inversible et que $\chi$ ou $\psi$ n’est pas inversible.\
D’après la Proposition \[pr:etaconeE6\], on a deux possibilités pour le vecteur $(\eta_x,\eta_y,\eta_z) $:
- $\eta_x=2\eta_z$ et $2\eta_x<3\eta_y$;
- $2\eta_x=3\eta_y$ et $\eta_x<2\eta_z$.\
Cas 1). On suppose que: $\eta_x=2\eta_z$ et $2\eta_x<3\eta_y$.\
D’après la Relation (\[eq:1\]), on a:
$$t^{3\eta_y-2\eta_x}\varphi^3+(\chi+i\psi^2)(\chi-i\psi^2)=0.$$
On remarque que $t$ divise $\chi+i\psi^{2}$ si et seulement si $t$ ne divise pas $\chi-i\psi^{2}$ car $t$ ne divise pas $\chi$ et $ \varphi$. On peut donc sans perte de généralité supposer que $t^{3\eta_y-2\eta_x}$ divise $\chi+i\psi^2$.\
Si la série formelle $\varphi$ est inversible, alors la relation $ t^{3\eta_y-2\eta_x}\varphi^3+(\chi+i\psi^2)(\chi-i\psi^2)=0$ équivaut au système de relations suivant:
$$t^{3\eta_y-2\eta_x}I_1=\chi+i\psi^2,\; I_2=-\chi+i\psi^2,$$ où $I_1,$ et $I_2$ sont deux séries formelles inversibles de $K[[s,t]]$ telles que $I_1I_2=\varphi^3$. Alors, les séries formelles $\chi$ et $\psi$ sont inversibles, d’où la contradiction.\
Cas 2). On suppose que: $2\eta_x=3\eta_y$ et $\eta_x<2\eta_z$.\
Comme $2\eta_x=3\eta_y$, on obtient que $\eta_x$ (resp. $\eta_y$) est divisible par $3$ (resp. $2$). Par conséquent, on a $(\eta_x,\eta_y)=(3,2)$ et $2\leq \eta_z\leq 3$, car $2\leq \eta_x\leq 6$, $2\leq \eta_y\leq 4$, $1\leq \eta_z\leq 3$ (voir la Remarque \[re:221eta643\]) et $\eta_x<2\eta_z$.\
On rappelle que $(\mu_x,\mu_y,\mu_z)\in \{(2,2,1),(3,2,2),(4,3,2),(6,4,3)\}$. Comme la série formelle $\chi$ ou la série formelle $\psi$ n’est pas inversible, le $v$-ordre $\nu_v\chi$ ou le $v$-ordre $\nu_v\psi$ n’est pas nul (voir Proposition \[pr:FI-g\]). Alors, on a $(\mu_x,\mu_y,\mu_z)\in \{(4,3,2),(6,4,3)\},$ d’où $1\leq \mu_y-\eta_y\leq2$. Par conséquent, la série formelle $\varphi$ n’est pas inversible, ce qui est une contradiction.
Maintenant, on considère le cas $i)$ de la Proposition \[pr:etaconeE6\].
\[pr:2e6\] On suppose que $2\eta_x=3\eta_y$ et $\eta_x<2\eta_z$. Alors, les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles.
Raisonnons par l’absurde. En vertu du Lemme \[le:varinv\] on suppose que $\varphi$ n’est pas inversible. D’après la Proposition \[pr:FI-g\], on a $\mu_y-\eta_y\geq 1$.\
Comme $2\eta_x=3\eta_y$, on obtient que $\eta_x$ (resp. $\eta_y$) est divisible par $3$ (resp. $2$). Par conséquent, on a $(\eta_x,\eta_y)=(3,2)$ et $2\leq \eta_z\leq 3$, car $2\leq \eta_x\leq 6$, $2\leq \eta_y\leq 4$, $1\leq \eta_z\leq 3$ (voir la Remarque \[re:221eta643\]) et $\eta_x<2\eta_z$.\
On rappelle que $(\mu_x,\mu_y,\mu_z)\in \{(2,2,1),(3,2,2),(4,3,2),(6,4,3)\}$. Comme $\eta_y=2$ et $\mu_y-\eta_y\geq 1$, on obtient que $(\mu_x,\mu_y,\mu_z)\in \{(4,3,2),(6,4,3)\}$. En particulier, $1 \leq \mu_y-\eta_y\leq 2$.
D’après la Proposition \[pr:FI-g\], le nombre de facteurs irréductibles comptés avec multiplicité de $\varphi$ est inférieur ou égal à $2$.\
Au moyen de la Relation (\[eq:1\]) on obtient la relation suivante:
$$\varphi^3+(\chi+it^{2\eta_z-\eta_x}\psi^2)(\chi-it^{2\eta_z-\eta_x}\psi^2)=0.$$
On remarque que la série formelle $\chi+it^{2\eta_z-\eta_x}\psi^2$ est inversible si et seulement si la série formelle $\chi-it^{2\eta_z-\eta_x}\psi^2$ l’est. Par conséquent, si $\chi+it^{2\eta_z-\eta_x}\psi^2$ est inversible, alors $\varphi$ l’est. On obtient donc que les séries formelles $\chi+it^{2\eta_z-\eta_x}\psi^2$ et $\chi-it^{2\eta_z-\eta_x}\psi^2$ ne sont pas inversibles.
La série formelle $\varphi$ n’est pas irréductible, car si $\varphi$ est irréductible, alors $\varphi$ divise $\chi$ et $\psi$, ce qui rentre en contradiction avec le Lemme \[le:pre0E6\]. On a donc $\mu_y-\eta_y=2$ (voir la Proposition \[pr:FI-g\]), ce qui implique que $(\mu_x,\mu_y,\mu_z)=(6,4,3)$, car $\eta_y=2$. On rappelle que $2\leq \eta_z\leq 3$ et $\mu_z=3$, d’où la série formelle $\psi$ est irréductible ou inversible, parce que $\mu_z-\eta_z\leq 1$ (voir la Proposition \[pr:FI-g\]).\
Comme $\varphi$ n’est pas irréductible, le nombre de facteurs irréductibles de $\varphi$ est égal à $2$. On a donc deux cas:\
- La série formelle $\varphi$ est le produit de deux séries formelles irréductibles associées.
- La série formelle $\varphi$ est le produit de deux séries formelles irréductibles non associées.\
Maintenant, on va montrer que dans ces deux cas on arrive à des contradictions, ce qui démontre que les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles.\
Cas 1). On suppose que $\varphi=\varphi_1\varphi_2$, où $\varphi_1$ et $\varphi_2$ sont deux séries formelles irréductibles associées.\
Comme le corps $K$ est algébriquement clos, on peut supposer que $\varphi_1=\varphi_2$.\
D’après la Relation (\[eq:1\]), on a:
$$\varphi_1^6+(\chi+it^{2\eta_z-\eta_x}\psi^2)(\chi-it^{2\eta_z-\eta_x}\psi^2)=0,$$ ce qui équivaut au système de relations suivant:
$$\varphi_1^3 I=-\chi - it^{2\eta_z-\eta_x}\psi^2,\;\; \varphi_1^{3}I^{-1}=\chi - it^{2\eta_z-\eta_x}\psi^2,$$ où $I$ est une série formelle inversible de $K[[s,t]]$. En effet, le vecteur $(\mu_x,\mu_y,\mu_z)$ est égal à $(6,4,3)$, donc le $v$-ordre $\nu_{v}(\chi + it^{2\eta_z-\eta_x}\psi^2)$ est égal au $v$-ordre $\nu_{v}(\chi - it^{2\eta_z-\eta_x}\psi^2)$ (voir le Lemme \[le:643\]), ce qui implique le système de relations ci-dessus. Par conséquent, on a la relation suivante:
$2it^{2\eta_z-\eta_z}\psi^2=-(I+ I^{-1})\varphi_1^3.$
On rappelle que la série formelle $\psi$ est irréductible ou inversible, parce que $\mu_z-\eta_z\leq 1$. Alors, la série formelle irréductible $\varphi_1$ divise $t^{2\eta_y-\eta_z}$, d’où une contradiction.\
Cas 2). On suppose que $\varphi=\varphi_1\varphi_2$ où $\varphi_1$ et $\varphi_2$ sont séries formelles irréductibles non associées.\
D’après la Relation (\[eq:1\]), on a :
$$\varphi_1^3\varphi_2^3+(\chi+it^{2\eta_z-\eta_x}\psi^2)(\chi-it^{2\eta_z-\eta_x}\psi^2)=0.$$
Au moyen des Lemmes \[le:643\] et \[le:pre0E6\] on obtient que cette relation équivaut au système de relations suivant:
$\varphi_1^3 I_1=-\chi - it^{2\eta_z-\eta_x}\psi^2,\;\; \varphi_2^{3}I_2=\chi - it^{2\eta_z-\eta_x}\psi^2,$
où $I_1$, $I_2$ sont deux séries formelles inversibles de $K[[s,t]]$ telles que $I_1I_2=1$, d’où\
$-2it^{2\eta_z-\eta_x}\psi^2=\varphi_1^3I_1+\varphi_2^3I_2=\prod_{i=1}^3(\varphi_1J_1+w_i\varphi_2J_2),$
où les $w_i$ sont les racines i-ièmes de l’unité et $J_1$, $J_2$ sont deux séries formelles inversibles telles que $J_1^3=I_1$ et $J_2^3=I_2$.
On rappelle que la série formelle $\psi$ est irréductible ou inversible. Si $\psi$ est inversible, alors les séries formelles $\varphi_1$ et $\varphi_2$ sont inversibles ou divisibles par $t$. On a donc une contradiction dans les deux cas. Ainsi, on obtient que la série formelle $\psi$ est irréductible.
Comme $t$ ne divise pas $\varphi_1$ et $\varphi_2$, la série formelle $\psi$ divise $\varphi_1$ et $\varphi_2$. Par conséquent, la série $\psi$ divise $t$, d’où une contradiction. Ceci achève la démonstration de la proposition.
Les deux lemmes suivants sont très importants dans la démonstration.
\[le:1.1\] On suppose que $(\mu_x,\mu_y,\mu_z)=(4,3,2)$. Alors, les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles.
Raisonnons par l’absurde. En vertu du Lemme \[le:varinv\] on suppose que $\varphi$ n’est pas inversible. Par conséquent, $\mu_y-\eta_y\geq 1$ (voir la Proposition \[pr:FI-g\]).\
En vertu de la Proposition \[pr:etaconeE6\], le vecteur $(\eta_x,\eta_y,\eta_z)$ satisfait une des relations suivantes:
$\eta_x=2\eta_z$ et $2\eta_x<3\eta_y$; $\;\;\;\;$ $2\eta_x=3\eta_y$ et $\eta_x<2\eta_z$.\
D’après la proposition \[pr:2e6\], si on a $2\eta_x=3\eta_y$ et $\eta_x<2\eta_z$, alors les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles. Par conséquent, le vecteur $(\eta_x,\eta_y,\eta_z)$ satisfait la relation $\eta_x=2\eta_z\;\mbox{et}\; 2\eta_x<3\eta_y$. En particulier, $2$ divise $\eta_x$. Ainsi, on obtient que $\eta_x\in \{2,4\}$, car $2\leq \eta_x\leq \mu_x=4$ (voir la Remarque \[re:221eta643\]). On rappelle que $\mu_y-\eta_y\geq 1$. Comme on a $ \eta_y\geq 2$ (Remarque \[re:221eta643\]) et $\mu_y=3$, on obtient que $\eta_y=2$. On a donc $\eta_x<3$, car $2\eta_x<3\eta_y$. Ainsi, on obtient que $(\eta_x,\eta_y,\eta_z)=(2,2,1)$. Par conséquent, on a les $v$-ordres suivants:
$$\nu_v\chi=2,\; \nu_v\varphi=1,\;\mbox{et}\; \nu_v\psi=1.$$
Comme les $v$-ordres $\nu_v\chi$, $\nu_v\varphi$ et $\nu_v\psi$ sont strictement positifs, les séries formelles $\chi$, $\varphi$ et $\psi$ ne sont pas inversibles. De plus, les séries formelles $\varphi$ et $\psi$ sont irréductibles (voir la Proposition \[pr:FI-g\]).\
D’après la Relation (\[eq:1\]), on a:
$$t^{2}\varphi^3+(\chi+i\psi^2)(\chi-i\psi^2)=0.$$
On remarque que $t$ divise $\chi+i\psi^{2}$ si et seulement si $t$ ne divise pas $\chi-i\psi^{2}$, car $t$ ne divise pas $\chi$ et $ \varphi$. On peut donc supposer, sans perte de généralité, que $t^{2}$ divise $\chi+i\psi^2$.\
Si la série formelle $\varphi$ divise $\chi+i\psi^2$ et $\chi-i\psi^2$, alors $\varphi$ divise $\chi$ et $\psi$, ce qui rentre en contradiction avec le Lemme \[le:pre0E6\]. Comme les séries formelles $\chi$ et $\psi$ ne sont pas inversibles et par hypothèse $t^2$ divise $\chi+i\psi^2$, la relation $ t^{2}\varphi^3+(\chi+i\psi^2)(\chi-i\psi^2)=0$ équivaut au système de relations suivant:
$$t^2I=\chi+i\psi^2,\;\varphi^3 I^{-1}=-\chi+i\psi^2,$$ où $I\in K[[s,t]]$ est un série formelle inversible. Ainsi, on obtient la relation suivante:
$$2i\psi^2=t^2I+\varphi^3 I^{-1}.$$
Le corps $K$ est algébriquement clos, donc il existe $I_1\in K[[s,t]]^{\star}$ tel que $I_1^2=I$. On a donc:
$$(\kappa \psi -tI_1t)(\kappa \psi+tI_1)=\varphi^3I^{-1},$$ où $\kappa^2=2i$. Comme les séries formelles $\kappa \psi -tI_1$ et $\kappa \psi +tI_1$ ne sont pas inversibles et la série formelle $\varphi$ est irréductible, $\varphi$ divise $\kappa \psi -tI_1$ et $\kappa \psi +tI_1$ ce qui implique que $\varphi$ divise la série formelle $tI_1$, d’où une contradiction car $\varphi$ n’est pas divisible par $t$.
\[le:1.2\] On suppose que $(\mu_x,\mu_y,\mu_z)=(6,4,3)$, alors les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles.
Raisonnons par l’absurde. En vertu du Lemme \[le:varinv\] on suppose que $\varphi$ n’est pas inversible.\
En vertu de la Proposition \[pr:etaconeE6\], le vecteur $(\eta_x,\eta_y,\eta_z)$ satisfait une des relations suivantes:
$\eta_x=2\eta_z$ et $2\eta_x<3\eta_y$; $\;\;\;\;$ $2\eta_x=3\eta_y$ et $\eta_x<2\eta_z$.\
D’après la Proposition \[pr:2e6\], si on a $2\eta_x=3\eta_y$ et $\eta_x<2\eta_z$, alors les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles. Par conséquent, le vecteur $(\eta_x,\eta_y,\eta_z)$ satisfait la relation $\eta_x=2\eta_z\;\mbox{et}\; 2\eta_x<3\eta_y$.\
On rappelle que $2\leq \eta_x\leq 6$, $2 \leq \eta_y\leq 4$ et $1\leq \eta_z\leq 3$ (voir la Remarque \[re:221eta643\]). Comme la série formelle $\varphi$ n’est pas inversible, on a $2\leq \eta_y\leq 3$, car $\eta_y<\mu_y=4$ (voir la Proposition \[pr:FI-g\]).\
Si on a $\eta_y=3$, alors on a $\mu_y-\eta_y=1$. Par conséquent, la série formelle $\varphi$ est irréductible.\
D’après la Relation (\[eq:1\]), on a:
$$t^{9-2\eta_x}\varphi^3+(\chi+i\psi^2)(\chi-i\psi^2)=0.$$
On remarque que $t$ divise $\chi+i\psi^{2}$ si et seulement si $t$ ne divise pas $\chi-i\psi^{2}$, car $t$ ne divise pas $\chi$ et $ \psi$. On peut donc supposer, sans perte de généralité, que $t^{9-2\eta_x}$ divise $\chi+i\psi^2$. On a donc le système de relations suivant:
$$t^{9-2\eta_x}\varphi^j I= \chi + i\psi^2,\;\; \varphi^{3-j}I^{-1}=\chi - i\psi^2,$$ où $0\leq j \leq 3$ et $I$ est un élément inversible de $K[[s,t]]$.\
Si on a $j=3$, alors la série formelle $\chi - i\psi^2$ est inversible, d’où on a $\nu_v(\chi - i\psi^2)=0$. Le Lemme \[le:643\] et la propriété multiplicative du $v$-ordre montrent que le $v$-ordre $\nu_v(\chi + i\psi^2)$ est égal à zéro, ce qui implique que le $v$-ordre de $t^{9-2\eta_x}\varphi^3$ est égal à zéro, d’où une contradiction.\
Si on a $1\leq j\leq 2$, alors $\varphi$ divise les séries formelles $\chi$ et $\psi$ ce qui rentre en contradiction avec le Lemme \[le:pre0E6\]. On a donc le système de relations suivant:
$$t^{9-2\eta_x} I= \chi + i\psi^2,\;\; \varphi^{3}I^{-1}=\chi - i\psi^2.$$
Par conséquent, le $v$-ordre $\nu_v(\chi + i\psi^2)$ (resp. le $v$-ordre $\nu_v(\chi- i\psi^2)$) est égal à $9-2\eta_x$ (resp. est égal à 3). Le Lemme \[le:643\] et la propriété multiplicative du $v$-ordre montrent qu’on a $\nu_v(\chi + i\psi^2)= \nu_v(\chi - i\psi^2)$, ce qui implique que $\eta_x=3$, d’où la contradiction car $\eta_x=2\eta_z$. Forcément, on a donc $\eta_y=2$.\
Comme on a $\eta_x=2\eta_z\;\mbox{et}\; 2\eta_x<3\eta_y$ et $\eta_y=2$, on obtient que $(\eta_x,\eta_y,\eta_z)=(2,2,1)$, d’où
$$t^{2}\varphi^3+(\chi+i\psi^2)(\chi-i\psi^2)=0.$$
Comme on a $\nu_v\varphi=\mu_y-\eta_y=2$, la série formelle $\varphi$ est au plus le produit de deux séries formelles irréductibles comptés avec multiplicité, voir la Proposition \[pr:FI-g\].\
Alors, pour la série formelle $\varphi$ on obtient le trois cas suivants:\
- la série formelle $\varphi$ est irréductible;
- la série formelle $\varphi$ est le produit de deux séries formelles non associées;
- la série formelle $\varphi$ est le produit de deux séries formelles associées.\
Maintenant, on va montrer que dans chaque cas ci-dessus on obtient une contradiction, ce qui implique que $\varphi$ est inversible, d’où le lemme.\
On rappelle que dans le trois cas on a la relation suivante:
$$t^{2}\varphi^3+(\chi+i\psi^2)(\chi-i\psi^2)=0.$$
On remarque que $t$ divise $\chi+i\psi^{2}$ si et seulement si $t$ ne divise pas $\chi-i\psi^{2}$ car $t$ ne divise pas $\chi$ et $ \varphi$. On peut donc supposer, sans perte de généralité, que $t^{2}$ divise $\chi+i\psi^2$.\
Cas 1). On suppose que $\varphi$ est une série formelle irréductible.\
Si la série formelle $\varphi$ est irréductible, alors la relation $ t^{2}\varphi^3+(\chi+i\psi^2)(\chi-i\psi^2)=0$ équivaut au système de relations suivant:
$$-t^2\varphi^j I=\chi + i\psi^2,\;\; \varphi^{3-j}I^{-1}=\chi - i\psi^2\mbox{,}$$ où $0\leq j \leq 3$ et $I$ est une série formelle inversible de $K[[s,t]]$.\
Le Lemme \[le:643\] et la propriété multiplicative du $v$-ordre montrent qu’on a $\nu_v(\chi + i\psi^2)= \nu_v(\chi - i\psi^2)$, ce qui implique que $2(1+j)=2(3-j)$ (on rappelle que le $v$-ordre $\nu_v\varphi$ est égal à $2$), d’où $j=1$. Alors, la série formelle $\varphi$ divise les séries formelles $\chi$ et $\psi$, ce qui rentre en contradiction avec le Lemme \[le:pre0E6\].\
Cas 2). On suppose que $\varphi=\varphi_1\varphi_2$ où $\varphi_1$ et $\varphi_2$ sont deux séries formelles irréductibles non associées.\
Dans ce cas, les $v$-ordres $\nu_v\varphi_1$ et $\nu_v\varphi_2$ sont égaux à $1$, car $\nu_v\varphi=2$.\
De la même façon que dans le cas précédent, on a le système de relations suivant:
$$t^2\varphi_1^j \varphi_2^kI=-(\chi + i\psi^2),\;\; \varphi_1^{3-j}\varphi_2^{3-k}I^{-1}=(\chi - i\psi^2),$$ où $0\leq j\leq 3,\;0\leq k\leq 3$ et $I$ est un élément inversible de $K[[s,t]]$.\
Le Lemme \[le:643\] et la propriété multiplicative du $v$-ordre montrent qu’on a $\nu_v(\chi + i\psi^2)= \nu_v(\chi - i\psi^2)$, ce qui implique que $2+j+k=6-j-k$, d’où $j+k=2$. Pour tous les $j$ et $k$ tels que $j+k=2$, on rentre en contradiction avec le Lemme \[le:pre0E6\].\
Cas 3). On suppose que $\varphi=\varphi_1\varphi_2$, où $\varphi_1$ et $\varphi_2$ sont des séries formelles irréductibles associées.\
Comme le corps $K$ est algébriquement clos, on peut supposer que $\varphi_1=\varphi_2$. Dans ce cas, le $v$-ordre $\nu_v\varphi_1$ est égal à $1$.\
La relation $ t^{2}\varphi^3+(\chi+i\psi^2)(\chi-i\psi^2)=0$ équivaut à la relation:
$$t^2\varphi_1^6+(\chi+i\psi^2)(\chi-i\psi^2)=0,$$ et par conséquent, on obtient le système de relations suivant:
$$-t^2\varphi_1^j I=\chi + i\psi^2,\;\; \varphi_1^{6-j}I^{-1}=\chi - i\psi^2,$$ où $0\leq j\leq 6$ et $I$ est un élément inversible de $K[[s,t]]$. Le Lemme \[le:643\] et la propriété multiplicative du $v$-ordre montrent qu’on a $\nu_v(\chi + i\psi^2)= \nu_v(\chi - i\psi^2)$, ce qui implique que $2+j=6-j$, d’où $j=2$. On a donc:
$$2i\psi^2=-(t^2I+\varphi_1^2 I^{-1})\varphi_1^2 .$$
Comme $K$ est algébriquement clos, il existe une série formelle inversible $I_1$ appartenant à $K[[s,t]]$ telle que $I_1^2=I$. Alors, on a:
$$2i\psi^2=-(tI_1+i\varphi_1 I_1^{-1})(tI_1-i\varphi_1 I_1^{-1})\varphi_1^2 .$$
Soient $r:=(tI_1+i\varphi_1 I_1^{-1})$ et $q:=(tI_1-i\varphi_1 I_1^{-1})$. Les séries formelles $r$ et $q$ ne sont pas inversibles car les séries formelles $tI_1$ et $\varphi_1$ ne le sont pas.\
Comme on a $(\eta_x,\eta_y,\eta_z)=(2,2,1)$ et $(\mu_x,\mu_y,\mu_z)=(6,4,3)$, le $v$-ordre $\nu_v\psi$ est égal à $2$ (on remarque que $\mu_z-\eta_z=2$), ce qui implique que $\psi$ est au plus le produit de deux séries formelles irréductibles comptées avec multiplicité (voir la Proposition \[pr:FI-g\]). Par conséquent, la série formelle $\psi^2$ est au plus le produit de quatre séries formelles irréductibles comptées avec multiplicité, ce qui implique que les séries formelles $r$ et $q$ sont irréductibles.\
Les séries formelles $r$ et $q$ ne sont pas associées à la série formelle $\varphi_1$ parce que $t$ ne divise pas $\varphi_1$.\
On obtient donc: $$(tI_1+i\varphi_1 I_1^{-1})=J(tI_1-i\varphi_1 I_1^{-1}),$$ où $J$ est une série formelle inversible. En effet, la série formelle $r$ divise $\psi^2$, alors $r$ divise $\psi$, d’où $r^2$ divise $\psi^2$. Comme $r$ ne divise pas $\varphi_1$, $r$ divise $q$. Par conséquent $r$ et $q$ sont deux séries irréductibles associées.\
Ainsi, on obtient la relation suivante:\
$$t(J-1)I_1=i\varphi_1(J+1)I_1^{-1}.$$
On remarque que $J-1$ ou $J+1$ est inversible. Si $J-1$ est inversible, alors la série $\varphi_1$ est inversible ou $t$ divise $\varphi_1$; ce qui est absurde. Si $J+1$ est inversible, alors $t$ divise $\varphi_1$, ce qui est aussi une contradiction. Ceci achève la démonstration du lemme.
Maintenant, on considère le cas [*ii*]{}) de la proposition \[pr:etaconeE6\].
\[pr:1e6\] On suppose que $\eta_x=2\eta_z$ et $2\eta_x<3\eta_y$. Alors, les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles.
Raisonnons par l’absurde. En vertu du Lemme \[le:varinv\], on suppose que $\varphi$ n’est pas inversible.\
On rappelle que $2\leq \eta_x\leq 6$, $2 \leq \eta_y\leq 4$ et $1\leq \eta_z\leq 3$ (voir la Remarque \[re:221eta643\]). Comme on a $\eta_x=2\eta_z$ et $2\eta_x<3\eta_y$, on obtient que $(\eta_x,\eta_z)\in\{(2,1),(4,2)\}$ et que $2\leq \eta_y \leq 4$.\
On rappelle que $(\mu_x,\mu_y,\mu_z)\in \{(2,2,1),(3,2,2),(4,3,2),(6,4,3)\}$. Comme $\varphi$ n’est pas inversible et $2\leq \eta_y \leq 4$, le vecteur $(\mu_x,\mu_y,\mu_z)\in \{(4,3,2),(6,4,3) \}$. En vertu des Lemmes \[le:1.1\] et \[le:1.2\], si le vecteur $(\mu_x,\mu_y,\mu_z)$ appartient à l’ensemble $ \{(4,3,2),(6,4,3) \}$, alors les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles, d’où une contradiction.
La démonstration de la bijectivité de l’application de Nash pour la singularité de type ${\ensuremath{\mathbb{E}}}_6$ résulte des propositions \[pr:invserBP\], \[pr:etaconeE6\], \[pr:2e6\] et \[pr:1e6\].
La singularité de type ${\ensuremath{\mathbb{E}}}_7$
----------------------------------------------------
La preuve de la bijectivité de l’application de Nash pour le cas de la singularité de type ${\ensuremath{\mathbb{E}}}_7$ (Théorème \[th:nashDP\]) repose sur la construction d’une $G$-désingularisation de l’éventail de Newton et sur des propriétés, pour ${\ensuremath{\mathbb{E}}}_7$, des séries formelles $\chi$, $\varphi$ et $\psi$ analogues aux séries définies pour ${\ensuremath{\operatorname{S(p,h_{q})}}}$ ou ${\ensuremath{\mathbb{E}}}_6$. Dans la suite, on donne un résumé de la preuve.\
Soit $S$ l’hypersurface normale de ${\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ donnée par l’équation $x^2+y(y^2+z^3)=0$. L’hypersurface $S$ a un unique point singulier de type ${\ensuremath{\mathbb{E}}}_7$ à l’origine de ${\ensuremath{\mathbb{A}}}^3$.
On considère l’éventail de Newton ${\ensuremath{\operatorname{\Gamma^{\star}(f)}}}$ associé à ${\ensuremath{\operatorname{f}}}:=x^2+y(y^2+z^3)$ et on note $\tau_1$ (resp. $\tau_2$, $\tau_3$) le cône engendré par les vecteurs $(1,0,0)$ (resp. $(1,2,0)$, $(0,0,1)$) et $(9,6,4)$ (voir la Figure \[fi:ENE7\]).
(0,0) – (90:3) (0,0) – (210:3) (0,0) – (310.89:1.984); (330:3) – (90:3) – (210:3) – (330:3); (310.89:2.2) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(1,2,0)$]{}]{}; (0.5,0.2) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[(9,6,4)]{}]{}; (335:3.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(0,1,0)$]{}]{}; (325:1.2) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [$\tau_2$]{}; (205:3.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(1,0,0)$]{}]{}; (200:1.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [$\tau_1$]{}; (90:3.3) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(0,0,1)$]{}]{}; (100:1.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [$\tau_3$]{}; (0,0) circle(0.7mm) (330:3) circle(0.7mm) (90:3) circle(0.7mm) (210:3) circle(0.7mm) (90:3) circle(0.7mm) (310.89:1.984) circle(0.7mm);
Soit ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}}$ une $G$-[*subdivision régulière*]{} de ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$. On note $\pi_{\mathcal{N}}:X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}})\rightarrow {\ensuremath{\mathbb{A}}}^{3}_{{\ensuremath{\mathbf{k}}}}$ (resp. $\pi_{\mathcal{G}}:X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}})\rightarrow X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}})$) le morphisme torique induit par la subdivision ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$ de ${\ensuremath{\mathbb{R}}}^3_{\geq 0}$ (resp. ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}}$ de ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$) et $S_{\mathcal{G}}$ le transformé strict de $S$ associé au morphisme $\pi:=\pi_{\mathcal{G}}\circ \pi_{\mathcal{N}}$.
De manière analogue à la Proposition \[pr:desingE6\], le morphisme $\pi:S_{\mathcal{G}}\rightarrow S$ est la résolution minimale de $S$.\
Montrer la bijectivité de l’application de Nash $\mathcal{N}_{S}$ équivaut à montrer que, pour chaque diviseur essentiel ${\ensuremath{\operatorname{E}}}$, tous les $K$-wedges admissibles centrés en $N_{{\ensuremath{\operatorname{E}}}}$ se relèvent à la résolution minimale de $S$. Dans la suite, on donne une idée de pourquoi un $K$-wedge admissible centré en $N_{{\ensuremath{\operatorname{E}}}}$ se relève à ${\ensuremath{\operatorname{S_{\mathcal{G}}}}}$.\
Soient ${\ensuremath{\operatorname{E}}}\in {\ensuremath{\operatorname{Ess}}}(S)$ et $\omega:{\ensuremath{\operatorname{Spec}}}K[[s,t]]\rightarrow S$ un $K$-wedge admissible centré en $N_{{\ensuremath{\operatorname{E}}}}$. On rappelle que $\alpha_{{\ensuremath{\operatorname{E}}}}$ est le point générique de $N_{{\ensuremath{\operatorname{E}}}}$. On considère les vecteurs suivants:
$(\mu_x,\mu_y,\mu_z):=({\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(x),{\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(y), {\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(z))$;
$(\eta_x,\eta_y,\eta_z): =({\ensuremath{\operatorname{Ord}_{t} \omega}}^{\star }(x),{\ensuremath{\operatorname{Ord}_{t} \omega}}^{\star }(y),{\ensuremath{\operatorname{Ord}_{t} \omega}}^{\star }(z))$,
où ${\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}$ (resp. ${\ensuremath{\omega^{\star}}}$) est le comorphisme de $\alpha_{{\ensuremath{\operatorname{E}}}}$ (resp. de $\omega$). On peut donc écrire le comorphisme de $\omega$ de la façon suivante:
$\omega^{\star }(x)=t^{\eta_x}\chi,\;\omega^{\star }(y)=t^{\eta_y}\varphi,\; \omega^{\star }(z)=t^{\eta_z}\psi,$
où les séries formelles $\chi$, $\varphi$, $\psi$ ne sont pas divisibles par $t$. Le $K$-wedge $\omega$ doit satisfaire l’équation $x^2+y(y^2+z^{3})=0$, d’où la relation suivante:
$$t^{2\eta_x}\chi^2+t^{3\eta_y}\varphi^3+t^{\eta_y+3\eta_z}\varphi\psi^{3}=0. \label{eq:E7}$$
En vertu des Propositions \[pr:invserBP\], si les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles, alors le $K$-wedge admissible $\omega$ centré en $N_{{\ensuremath{\operatorname{E}}}}$ se relève à ${\ensuremath{\operatorname{S_{\mathcal{G}}}}}$.\
D’après la Proposition \[pr:FI-g\], on peut majorer le nombre de facteurs irréductibles comptés avec multiplicité des séries formelles $\chi$, $\varphi$ et $\psi$ à l’aide des $v$-ordres de la façon suivant:
${\ensuremath{\operatorname{FI}}}(\chi)\leq {\ensuremath{\operatorname{Deg}_{t} \chi_v}}= \nu_v\chi =\mu_x-\eta_x\leq 6$;
${\ensuremath{\operatorname{FI}}}(\varphi)\leq {\ensuremath{\operatorname{Deg}_{t} \varphi_v}}= \nu_v\varphi= \mu_y-\eta_y\leq 4$;
${\ensuremath{\operatorname{FI}}}(\psi)\leq {\ensuremath{\operatorname{Deg}_{t} \psi_v}}= \nu_v\psi= \mu_z-\eta_z\leq 3$.
La proposition suivante est une propriété importante du vecteur $(\mu_x,\mu_y,\mu_z)$. Pour un cône $\tau$ dans ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$, on note $G\tau$ le système générateur minimal du semi-groupe $\tau\cap {\ensuremath{\mathbb{Z}}}^3$.
\[pr:musygemiE7\] Le vecteur $(\mu_x,\mu_y,\mu_z)$ appartient à l’union $G\tau_1\cup G\tau_2\cup G\tau_3$, où $\tau_1$ (resp. $\tau_2$ $\tau_3$) est le cône engendré par les vecteurs $(1,0,0)$ (resp. $(1,2,0)$, $(0,0,1)$) et $(9,6,4)$ (voir la figure \[fi:ENE7\]). Autrement dit, $(\mu_x,\mu_y,\mu_z)\in \{(3, 2, 2), (6, 4, 3), (9, 6, 4), (7, 5, 3),$ $(5, 4, 2), (3, 3, 1), (5, 3, 2) \}$.
Si on suppose les séries $\chi$, $\varphi$ et $\psi$ non simultanément inversibles, on obtient la proposition suivante:
\[pr:etaE7\] S’il existe au moins une série formelle parmi les séries $\chi$, $\varphi$, $\psi$ qui n’est pas inversible, alors le vecteur $(\eta_x,\eta_y,\eta_z)$ satisfait une des relations suivantes:
- $2\eta_x=\eta_y+3\eta_z$ et $2\eta_x<3\eta_y$;
- $2\eta_x=3\eta_y$ et $2\eta_y<3\eta_z$;
- $2\eta_y=3\eta_z$ et $3\eta_y<2\eta_x$.
La Proposition \[pr:etaE7\] donne trois possibilités pour le vecteur $(\eta_x,\eta_y,\eta_z)$. Dans la suite, on considère chaque cas séparément pour obtenir des contradictions.\
Dans chaque cas du vecteur $(\eta_x,\eta_y,\eta_z)$, on peut simplifier la Relation (\[eq:E7\]) et on peut la récrire comme la somme de deux termes. Ceci permet d’établir des relations entre les facteurs irréductibles des décompositions possibles des séries $\chi$, $\varphi$ et $\psi$. L’idée est de trouver des contradictions dans les différents cas de décomposition obtenus.\
D’abord, énonçons le lemme suivant qui est obtenu en utilisant la Relation (\[eq:E7\]).
\[le:varphi-no-irrE7\] Si on suppose que $\varphi=\varphi_1\varphi_2$, où $\varphi_1$ est une série formelle irréductible, alors la série formelle $\varphi_1$ divise la série formelle $\varphi_2$. En particulier, la série formelle $\varphi$ n’est pas irréductible.
Maintenant, on considère le premier cas de la Proposition \[pr:etaE7\].
\[pr:cas1etaE7\] On suppose que $2\eta_x=\eta_y+3\eta_z$ et $2\eta_x<3\eta_y$. Alors, les séries $\chi$, $\varphi$, $\psi$ sont inversibles.
On remarque que d’après la relation (\[eq:E7\]), on obtient la relation suivante: $$\label{eq:C1E7}
\chi^2+t^{3\eta_y-2\eta_x}\varphi^3+\varphi\psi^{3}=0.$$
D’abord, on montre que la série $\varphi$ est inversible, ce qui équivaut à montrer que $\mu_y-\eta_y=0$. En utilisant les relations $2\eta_x=\eta_y+3\eta_z$ et $2\eta_x<3\eta_y$, on obtient que $\eta_y\geq 3$. En vertu du Lemme \[le:varphi-no-irrE7\] et la Proposition \[pr:musygemiE7\], on obtient que $\mu_y-\eta_y\in \{0,2,3\}$.\
Si on suppose que $\mu_y-\eta_y=2$, on obtient que $3\eta_y-2\eta_x\in \{2, 3\}$. D’après le lemme \[le:varphi-no-irrE7\], on a $\varphi=\varphi_1^2$, où $\varphi_1$ est irréductible.\
Dans le cas $3\eta_y-2\eta_x=2$, on peut écrire la Relation (\[eq:C1E7\]) comme la somme de deux termes qui nous permet d’établir des relations entre les facteurs irréductibles des séries $\chi$, $\varphi$ et $\psi$ et de trouver une contradiction.
Dans le cas $3\eta_y-2\eta_x=3$, on ne peut pas écrire de façon évidente la Relation (\[eq:C1E7\]) comme la somme de deux termes. Cependant, on peut considérer le changement de variable $t=u^2$ et étudier la Relation (\[eq:C1E7\]) dans l’anneau $K[[s,u]]$. Dans cet anneau, on peut écrire cette relation comme la somme de deux termes et on traite ce cas comme le précédent.\
Si on suppose que $\mu_y-\eta_y=3$, on obtient $3\eta_y-2\eta_x=3$. D’après le lemme \[le:varphi-no-irrE7\], on a $\varphi=\varphi_1^3$, où $\varphi_1$ est irréductible. Dans ce cas, on peut écrire la Relation (\[eq:C1E7\]) comme la somme de deux termes et on le traite comme les cas précédents.\
D’après la Proposition \[pr:musygemiE7\], le vecteur $(\mu_x,\mu_y,\mu_z)$ satisfait une des relations suivantes:
- $2\mu_x=3\mu_y$ et $2\mu_y < 3\mu_z$;
- $2\mu_x=3\mu_z$ et $3\mu_y < 2\mu_x$;
- $2\mu_x=\mu_y+3\mu_z$ et $2\mu_x\leq 3\mu_y$.
En utilisant que $\mu_y-\eta_y=0$, on obtient le lemme suivant:
\[le:vmu\] Le vecteur $(\mu_x,\mu_y,\mu_z)$ ne satisfait pas les relations suivantes:
- $2\mu_x=3\mu_y$ et $2\mu_y < 3\mu_z$;
- $2\mu_x=3\mu_z$ et $3\mu_y < 2\mu_x$.
En vertu du Lemme \[le:vmu\], on peut supposer que le vecteur $(\mu_x,\mu_y,\mu_z)$ satisfait la relation suivante: $2\mu_x=\mu_y+3\mu_z$ et $2\mu_x\leq 3\mu_y$.
Maintenant, en raisonnant par l’absurde, on suppose que la série $\chi$ ou la série $\psi$ n’est pas inversible. On peut donc établir une liste de cas possibles pour les vecteurs $(\mu_x,\mu_y,\mu_z)$ et $(\eta_x,\eta_y,\eta_z)$. En utilisant la Relation (\[eq:C1E7\]), la Proposition \[pr:musygemiE7\] et les majorants du nombre de facteurs irréductibles comptés avec multiplicité des séries formelles $\chi$ et $\psi$, on obtient une contradiction dans chaque cas de la liste, d’où la proposition.
La preuve des propositions suivantes est analogue à celle de la proposition \[pr:cas1etaE7\].
\[pr:cas2etaE7\] On suppose que $2\eta_x=3\eta_y$ et $2\eta_y<3\eta_z$. Alors, les séries $\chi$, $\varphi$, $\psi$ sont inversibles.
\[pr:cas3etaE7\] On suppose que $2\eta_y=3\eta_z$ et $3\eta_y<2\eta_x$. Alors, les séries $\chi$, $\varphi$, $\psi$ sont inversibles.
La démonstration du cas ${\ensuremath{\mathbb{E}}}_7$ du Théorème \[th:nashDP\] résulte des propositions \[pr:etaE7\], \[pr:cas1etaE7\], \[pr:cas2etaE7\] et \[pr:cas3etaE7\].
Une nouvelle preuve de la bijectivité de l’application de Nash pour les singularités de type ${\ensuremath{\mathbb{D}}}_n$
==========================================================================================================================
Dans l’article [@Ple08] l’auteur démontre la bijectivité de l’application de Nash $\mathcal{N}_{{\ensuremath{\mathbb{D}}}_n}$ associée à une singularité de type ${\ensuremath{\mathbb{D}}}_n$, $n\geq 4$. Dans cette section, en utilisant les mêmes méthodes que dans les preuves des Théorèmes \[th:nashBP\] et \[th:nashDP\], on donne une démonstration de la bijectivité de l’application $\mathcal{N}_{{\ensuremath{\mathbb{D}}}_n}$ différente de celle de [@Ple08].\
L’entier $n\geq 4$ étant fixé, soit $S$ l’hypersurface normale de ${\ensuremath{\mathbb{A}}}_{{\ensuremath{\mathbf{k}}}}^3$ donnée par l’équation $x^2+z(y^2+z^{n-2})=0$. L’hypersurface $S$ a un unique point singulier de type ${\ensuremath{\mathbb{D}}}_n$.\
Comme dans la preuve des Théorèmes \[th:nashBP\] ou \[th:nashDP\] pour prouver la bijectivité de l’application de Nash associée à $S$, on a besoin de quelques résultats sur la résolution minimale de $S$.\
On considère l’éventail de Newton ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$ associé à ${\ensuremath{\operatorname{f}}}:=x^2+z(y^2+z^{n-2})$. Soit $H$ un plan de ${\ensuremath{\mathbb{R}}}^3$ qui ne contient pas l’origine de ${\ensuremath{\mathbb{R}}}^3$ et tel que l’intersection de $H$ et ${\ensuremath{\mathbb{R}}}^3_{\geq 0}$ soit un ensemble compact. La Figure \[fi:ENDn\] représente l’intersection de $H$ avec la subdivision ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$ de ${\ensuremath{\mathbb{R}}}^3_{\geq 0}$. Chaque sommet du diagramme est identifié avec le [*vecteur extrémal*]{} correspondant. On note $\tau_1$ (resp. $\tau_2$, $\tau_3$) le cône engendré par les vecteurs $(1,0,0)$ (resp. $(0,1,0)$, $(0,0,1)$) et $(n-1,n-2,2)$.\
(0,0) – (330:3) (0,0) – (210:3) (0,0) – (-0.6495,1.8750); (330:3) – (90:3) – (210:3) – (330:3); (-1.2,1.8750) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(1,0,2)$]{}]{}; (0.85,0.2) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[(n-1,n-2,2)]{}]{}; (335:3.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(0,1,0)$]{}]{}; (340:1.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [$\tau_2$]{}; (205:3.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(1,0,0)$]{}]{}; (200:1.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [$\tau_1$]{}; (90:3.3) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [[$(0,0,1)$]{}]{}; (100:1.5) node\[inner sep=-1pt,below=-1pt,rectangle,fill=white\] [$\tau_3$]{}; (0,0) circle(0.7mm) (330:3) circle(0.7mm) (90:3) circle(0.7mm) (210:3) circle(0.7mm) (90:3) circle(0.7mm) (-0.6495,1.8750) circle(0.7mm);
Pour un cône $\tau$ dans ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$, on note $G\tau$ le système générateur minimal du semi-groupe $\tau\cap {\ensuremath{\mathbb{Z}}}^3$. Par un calcul direct, on obtient le résultat suivant:
\[pr:GtauDn\] Le système générateur minimal du semi-groupe $\tau_3\cap {\ensuremath{\mathbb{Z}}}^3$ est l’ensemble $G\tau_3=\{(j,j-1,2)\mid j\in \{1,2,...,n-1\}\}$. De plus, soit $k\geq 2$ un entier, on a:
- Si $n=2k$, alors le système générateur minimal du semi-groupe $\tau_1\cap {\ensuremath{\mathbb{Z}}}^3$ est l’ensemble $G\tau_1=\{(1,0,0),(k,k-1,1),(n-1,n-2,2)\}$ et $\tau_2$ est un cône régulier.
- Si $n=2k-1$, alors le système générateur minimal du semi-groupe $\tau_2\cap {\ensuremath{\mathbb{Z}}}^3$ est l’ensemble $G\tau_2=\{(0,1,0),(k-1,k-1,1),(n-1,n-2,2)\}$ et $\tau_1$ est un cône régulier.
Soit ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}}$ une $G$-[*subdivision régulière*]{} de ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$. On note $\pi_{\mathcal{N}}:X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}})\rightarrow {\ensuremath{\mathbb{A}}}^{3}_{{\ensuremath{\mathbf{k}}}}$ (resp. $\pi_{\mathcal{G}}:X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}})\rightarrow X({\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}})$) le morphisme torique induit par la subdivision ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$ de ${\ensuremath{\mathbb{R}}}^3_{\geq 0}$ (resp. ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})_{\mathcal{G}}}}}$ de ${\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$) et $S_{\mathcal{G}}$ le transformé strict de $S$ associé au morphisme $\pi:=\pi_{\mathcal{G}}\circ \pi_{\mathcal{N}}$.\
La proposition suivante est un analogue de la Proposition \[pr:Gdes1\].
\[pr:desingE6\] $S_{\mathcal{G}}$ est la résolution minimale de $S$.
Soient ${\ensuremath{\operatorname{E}}}$ un diviseur essentiel sur $S$ (${\ensuremath{\operatorname{E}}}\in {\ensuremath{\operatorname{Ess}}}(S))$ et $\alpha_{{\ensuremath{\operatorname{E}}}}$ le point générique de $N_{{\ensuremath{\operatorname{E}}}}$. On note
$(\mu_x,\mu_y,\mu_z):=({\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(x),{\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(y), {\ensuremath{\operatorname{Ord}_{t} {\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}}}(z))$,
où ${\ensuremath{\alpha_{{\ensuremath{\operatorname{E}}}}^{\star}}}$ est le comorphisme de $\alpha_{{\ensuremath{\operatorname{E}}}}$.\
La proposition suivante est un analogue du Corollaire \[co:musygemi\].
\[pr:musygemiDn\] Soit $k$ un entier, $k\geq 2$.
- Si $n=2k$, alors le vecteur $(\mu_x,\mu_y,\mu_z)$ appartient à l’union de $G\tau_1$ et $G\tau_3$.
- Si $n=2k-1$, alors le vecteur $(\mu_x,\mu_y,\mu_z)$ appartient à l’union de $G\tau_2$ et $G\tau_3$.
Démontrer que l’application de Nash bijective pour les singularités de type ${\ensuremath{\mathbb{D}}}_n$, $n\geq 4$, équivaut à montrer que tous les wedges admissibles se relèvent à la résolution minimale de ${\ensuremath{\mathbb{D}}}_n$ (voir [@Reg06]). Dans la suite, on montre que pour chaque diviseur essentiel ${\ensuremath{\operatorname{E}}}$ tous les $K$-wedges admissibles centrés en $N_{{\ensuremath{\operatorname{E}}}}$ se relèvent à la résolution minimale de $S$.\
Soit ${\ensuremath{\operatorname{E}}}\in {\ensuremath{\operatorname{Ess}}}(S)$ et on considère un $K$-wedge $\omega:{\ensuremath{\operatorname{Spec}}}K[[s,t]]\rightarrow S$ admissible centré en $N_{{\ensuremath{\operatorname{E}}}}$. On pose:
$(\eta_x,\eta_y,\eta_z): =({\ensuremath{\operatorname{Ord}_{t} \omega}}^{\star }(x),{\ensuremath{\operatorname{Ord}_{t} \omega}}^{\star }(y),{\ensuremath{\operatorname{Ord}_{t} \omega}}^{\star }(z))$.
On peut écrire le comorphisme de $\omega$ de la façon suivante:\
$\omega^{\star }(x)=t^{\eta_x}\chi,\;\omega^{\star }(y)=t^{\eta_y}\varphi,\; \omega^{\star }(z)=t^{\eta_z}\psi,$
où les séries formelles $\chi$, $\varphi$, $\psi$ ne sont pas divisibles par $t$.\
La proposition suivante est un analogue du Corollaire \[co:etacone\]. On note $\tau^0$ l’intérieur du cône $\tau\in {\ensuremath{\operatorname{\Gamma^{\star}({\ensuremath{\operatorname{f}}})}}}$.
\[pr:etaconeDn\] Soit $k$ un entier, $k\geq 2$. S’il existe au moins une série formelle parmi les séries $\chi$, $\varphi$, $\psi$ qui n’est pas inversible, alors on a:
- si $n=2k$, alors le vecteur $(\eta_x,\eta_y,\eta_z)$ appartient à l’union de $\tau_1^{0}$ et $\tau_3^{0}$;
- si $n=2k-1$, alors le vecteur $(\eta_x,\eta_y,\eta_z)$ appartient à l’union de $\tau_2^{0}$ et $\tau_3^{0}$.
En vertu de la Proposition \[pr:musygemiDn\], on a $\mu_x\leq n-1$, $\mu_y\leq n-2$ et $\mu_z\leq 2$. Comme $\eta_x\geq 1$, $\eta_y\geq 1$ et $\eta_z\geq 1$, on a $\mu_x-\eta_x\leq n-2$, $\mu_y-\eta_y\leq n-3$ et $\mu_z-\eta_z\leq 1$.\
D’après la Proposition \[pr:invserBP\], si les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles, alors le $K$-wedge admissible $\omega$ centré en $N_{{\ensuremath{\operatorname{E}}}}$ se relève à la résolution minimale de $S$.
En vertu de la Proposition \[pr:FI-g\], on peut majorer le nombre de facteurs irréductibles comptés avec multiplicité des séries formelles $\chi$, $\varphi$ et $\psi$ à l’aide des $v$-ordres. Plus précisément, il existe un vecteur $v\in {\ensuremath{\mathbb{Q}}}^{2}_{>0} $ tel que:\
- ${\ensuremath{\operatorname{FI}}}(\chi)\leq {\ensuremath{\operatorname{Deg}_{t} \chi_v}}= \nu_v\chi =\mu_x-\eta_x\leq n-2$;
- ${\ensuremath{\operatorname{FI}}}(\varphi)\leq {\ensuremath{\operatorname{Deg}_{t} \varphi_v}}= \nu_v\varphi= \mu_y-\eta_y\leq n-3$;
- ${\ensuremath{\operatorname{FI}}}(\psi)\leq {\ensuremath{\operatorname{Deg}_{t} \psi_v}}= \nu_v\psi= \mu_z-\eta_z\leq 1$.\
Sauf mention du contraire, dans toute la suite le vecteur $v\in {\ensuremath{\mathbb{Q}}}_{>0}^2$ satisfait la propriété ci-dessus. Le résultat suivant est un corollaire de la Proposition \[pr:FI-g\]
\[co:FIDn\] La série formelle $\psi$ est irréductible ou inversible.
Maintenant, on donne notre preuve de la bijectivité de l’application de Nash pour les singularités de type ${\ensuremath{\mathbb{D}}}_n$.
L’application de Nash $\mathcal{N}_S$ associée à $S$ est bijective. \[nashDn\]
En vertu de la Proposition \[pr:invserBP\], pour montrer que le $K$-wedge admissible $\omega$ se relève à la résolution minimale de $S$, il suffit de montrer que les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles. En raisonnant par l’absurde, on suppose qu’il y a au moins l’une d’elles n’est pas inversible.
On remarque qu’une série formelle $\phi \in K[[s,t]]$ est inversible dans $K[[s,t]]$ si et seulement si elle est inversible dans $\overline{K}[[s,t]]$, où $\overline{K}$ est la clôture algébrique de $K$. Dans toute la suite, on suppose que le corps $K$ est algébriquement clos.\
Le $K$-wedge $\omega$ doit satisfaire l’équation $x^2+z(y^2+z^{n-2})=0$, d’où la relation suivante:
$$t^{2\eta_x}\chi^2+t^{2\eta_y+\eta_z}\varphi^2\psi+t^{(n-1)\eta_z}\psi^{n-1}=0 \label{eq:dn}$$
En vertu de la Proposition \[pr:etaconeDn\], on a les cas suivants:\
- le vecteur $(\eta_x,\eta_y,\eta_z)$ appartient à l’intérieur $\tau_3^0$ du cône $\tau_3$;
- l’entier $n$ est égal à $ 2k$, où $k$ est un entier supérieur ou égal à $2$, et le vecteur $(\eta_x,\eta_y,\eta_z)$ appartient à l’intérieur $\tau_1^0$ du cône $\tau_1$;
- l’entier $n$ est égal à $ 2k-1$, où $k$ est un entier supérieur ou égal à $2$, et le vecteur $(\eta_x,\eta_y,\eta_z)$ appartient à l’intérieur $\tau_2^0$ du cône $\tau_2$.\
Dans chaque cas ci-dessus, on va obtenir une contradiction, ce qui achève la preuve du Théorème.\
Cas 1). On suppose que $(\eta_x,\eta_y,\eta_z)\in\tau_3^0$. Dans ce cas, on a $\eta_z=2$ et $\eta_x=\eta_y+1$. Au moyen de la Relation (\[eq:dn\]), on obtient la relation suivante: $$\chi^2+\varphi^2\psi=-t^{2(n-2-\eta_y)}\psi^{n-1}.$$ On remarque que $n-2-\eta_y>0$, car le vecteur $(\eta_x,\eta_y,\eta_z)$ appartient à l’intérieur du cône $\tau_3$.
D’après les Propositions \[pr:GtauDn\] et \[pr:musygemiDn\], on a $\mu_z\leq 2$. Par conséquent, $\mu_z-\eta_z=0$, ce qui implique que la série formelle $\psi$ est inversible (voir la proposition \[pr:FI-g\]).\
Comme $\psi$ est inversible et $K$ algébriquement clos, il existe une série formelle inversible $\psi_0$ tel que $\psi=-\psi_0^2$, d’où $$(\chi+\psi_0\varphi)(\chi-\psi_0\varphi)=\mbox{(-1)}^{n}t^{2(n-2-\eta_y)}\psi_0^{2(n-1)}.$$
Par conséquent, $t$ divise $\chi$ et $\varphi$ ou $\chi$ et $\varphi$ sont inversibles. Ce sont des contradictions.\
Cas 2). On suppose que $n=2k$, où $k$ est un entier supérieur ou égal à $2$ et $(\eta_x,\eta_y,\eta_z)\in \tau_1^0$.\
Comme $(\eta_x,\eta_y,\eta_z)$ appartient à $\tau_1^0$ et $n=2k$, le vecteur $(\eta_x,\eta_y,\eta_z)$ satisfait les relations suivantes:
$(k-1)\eta_z=\eta_y$ et $2\eta_x>(2k-1)\eta_z$.
Comme $1\leq \eta_x\leq \mu_x\leq 2k-1$ (voir les Propositions \[pr:GtauDn\] et \[pr:musygemiDn\]), on obtient que $\eta_z=1$, $\eta_y=2k-1$ et $\eta_x\geq k$. On a donc $\mu_z=2$. En effet, si $\mu_z=1$, alors $(\mu_x,\mu_y,\mu_z)=(k,k-1,1)$. D’après la Proposition \[pr:FI-g\], les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles, ce qui est une contradiction, car on a supposé que parmi ces séries formelles il y en a au moins une qui n’est pas inversible.\
Au moyen de la Relation (\[eq:dn\]), on obtient la relation suivante:
$t^{2\eta_x+1-n}\chi^2+\varphi^2\psi=-\psi^{n-1}.$
On remarque que $\psi$ est irréductible car $\nu_v\psi=\mu_z-\eta_z=1$ (voir la Proposition \[pr:FI-g\] et le Corollaire \[co:FIDn\]) Au moyen de la relation ci-dessus, on obtient que $\chi=\chi_0\psi^{k}$ et $\varphi=\varphi_0\psi^{k-1}$, où $\chi_0$ et $\varphi_0$ sont deux séries formelles qui satisfont la relation suivante:
$t^{2\eta_x+1-n}\chi_0^2\psi+\varphi_0^2=-1.$
Comme $\nu_v\psi=1$, on obtient que $\nu_v\chi=\nu_v\chi_0\psi^k=\nu_v\chi_0+k$ . Comme $\mu_x\leq 2k-1$ et $\eta_x\geq k$, on a $\nu_v\chi=\mu_x-\eta_x\leq k-1$ (voir la Proposition \[pr:FI-g\]). Par conséquent, on a $\nu_v\chi_0 \leq -1$, ce qui est une contradiction.\
La preuve du Cas 3) est analogue à celle du Cas 2).\
Cas 3). On suppose que $n=2k-1$, où $k$ est un entier supérieur ou égal à $2$ et $(\eta_x,\eta_y,\eta_z)\in \tau_2^0$.\
Comme $(\eta_x,\eta_y,\eta_z)$ appartient à $\tau_2^0$ et $n=2k-1$, le vecteur $(\eta_x,\eta_y,\eta_z)$ satisfait les relations suivantes:
$(k-1)\eta_z=\eta_x$ et $2\eta_y>(2k-3)\eta_z$.
Comme $1\leq \eta_y\leq \mu_y\leq 2k-3$ (voir les Propositions \[pr:GtauDn\] et \[pr:musygemiDn\]), on obtient que $\eta_z=1$, $\eta_x=k-1$ et $\eta_y\geq k-1$. On a donc $\mu_z=2$. En effet, si $\mu_z=1$, alors $(\mu_x,\mu_y,\mu_z)=(k-1,k-1,1)$. D’après la Proposition \[pr:FI-g\], les séries formelles $\chi$, $\varphi$ et $\psi$ sont inversibles, ce qui est une contradiction car on a supposé que parmi ces séries formelles il y en a au moins une qui n’est pas inversible.\
Au moyen de la Relation (\[eq:dn\]), on obtient la relation suivante: $$\chi^2+t^{2\eta_y+2-n}\varphi^2\psi=-\psi^{n-1}.$$
On remarque que $\psi$ est irréductible car $\nu_v\psi=\mu_z-\eta_z=1$ (voir la Proposition \[pr:FI-g\] et le Corollaire \[co:FIDn\]) Au moyen de la relation ci-dessus, on obtient que $\chi=\chi_0\psi^{k-1}$ et $\varphi=\varphi_0\psi^{k-1}$, où $\chi_0$ et $\varphi_0$ sont deux séries formelles qui satisfont la relation suivante:
$\chi_0^2+t^{2\eta_y+2-n}\varphi_0^2\psi=-1$.
Comme $\nu_v\psi=1$, on obtient que $\nu_v\varphi=\nu_v\varphi_0\psi^{k-1}=\nu_v\varphi_0+k-1$ . Comme $\mu_y\leq 2k-3$ et $\eta_y\geq k-1$, on a $\nu_v\varphi=\mu_y-\eta_y\leq k-2$ (voir la Proposition \[pr:FI-g\]). Par conséquent, on a $\nu_v\varphi_0 \leq -1$, ce qui est une contradiction.\
Dans le trois cas précédent, on a obtenu une contradiction, d’où le théorème.
[^1]:
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
We consider 3D consistent systems of six independent quad-equations assigned to the faces of a cube. The well-known classification of 3D consistent quad-equations, the so-called ABS-list, is included in this situation. The extension of these equations to the whole lattice ${\mathbb{Z}}^{3}$ is possible by reflecting the cubes. For every quad-equation we will give at least one system included leading to a Bäcklund transformation and a zero-curvature representation which means that they are integrable.
PACS number: 02.30.Ik
author:
- 'Raphael Boll[^1]'
bibliography:
- 'Quellen.bib'
title: 'Classification of 3D consistent quad-equations'
---
Introduction
============
One of the definitions of integrability of lattice equations, which becomes increasingly popular in the recent years, is based on the notion of multidimensional consistency. For two-dimensional lattices, this notion was clearly formulated first in [@NW], and it was proposed to use as a synonym of integrability in [@quadgraphs; @Nijhoff]. The outstanding importance of 3D consistency in the theory of discrete integrable systems became evident no later than with the appearance of the well-known ABS-classification of integrable equations in [@ABS1]. In that article Adler, Bobenko and Suris used a definition of 3D consistency only allowing equations on the faces of a cube, which differ only by the parameter values assigned to the edges of a cube. In [@Atk1] and [@Todapaper] appeared a lot of systems of quad-equations not satisfying this strict definition of 3D consistency. However, these systems can also be seen as families of Bäcklund transformations and they lead to zero curvature representations of participating quad-equations in the same way (see [@quadgraphs]).
As already done in [@ABS2] the definition of 3D consistency can be extended: In contrast to the restriction, that all faces of a cube must carry the same equation up to parameters assigned to edges of the cube, we will allow different equations on all faces of a cube. The classification in that article is restricted to so-called equations of type Q, i.e. those whose biquadratics are all non-degenerate (we will give a precise definition in the next section). The present paper is devoted to systems containing equations which are not necessarily of type Q. Our classification will cover systems appearing in [@Atk1] and all systems in [@Todapaper], as well as equations which are equivalent to the Hietarinta equation [@Hietarinta], to the ÒnewÓ equation in [@HV] and to the equation in [@LY]. Moreover, it will contains also many novel systems.
In addition, every system in this classification can be extended to the whole lattice ${\mathbb{Z}}^{3}$ by reflecting the cubes.
The outline of our approach is the following: In Section \[faces\] we will present a complete classification of a single quad-equation modulo Möbius transformations acting independently on the fields at the four vertices of an elementary quadrilateral. In Section \[cubes\] we will give a classification of 3D consistent systems of quad-equations possessing the so-called tetrahedron property modulo Möbius transformations acting independently on the fields at the eight vertices of an elementary cube. In Section \[embedding\] we will show how to embed our systems in the lattice ${\mathbb{Z}}^{3}$ and how to derive Bäcklund transformations and zero curvature representations from our systems. This will include the idea of embedding considered in [@XP] as a special case and can be seen as a justification for the extended definition of 3D consistency to yield a definition of integrability.
Quad-Equations on Single Quadrilaterals {#faces}
=======================================
At the beginning we will introduce some objects and notations. We will start with the most important one, the *quad-equation* $Q\left(x_{1},x_{2},x_{3},x_{4}\right)=0$, where $Q\in{\mathbb{C}}\left[x_{1},x_{2},x_{3},x_{4}\right]$ is an irreducible multi-affine polynomial.
Very useful tools for characterizing quad-equations, are the *biquadratics*. We define them for every permutation $\left(i,j,k,\ell\right)$ of $\left(1,2,3,4\right)$ as follows $$Q^{i,j}=Q^{i,j}\left(x_{i},x_{j}\right)=Q_{x_{k}}Q_{x_{\ell}}-QQ_{x_{k},x_{\ell}}.$$ A biquadratic $h\left(x,y\right)$ is called *non-degenerate* if no polynomial in its equivalence class with respect to Möbius transformations in $x$ and $y$ is divisible by a factor $x-\alpha_{1}$ or $y-\alpha_{2}$ (with $\alpha_{i}\in{\mathbb{C}}$). Otherwise, $h$ is called *degenerate* and factors $x-\alpha_{1}$ and $y-\alpha_{2}$ with $\left(x-\alpha_{1}\right)\mid h$ and $\left(y-\alpha_{2}\right)\mid h$ are called *factors of degeneracy*. Moreover, if $x\mid x^{2}\cdot h\left(1/x,y\right)$, we write $\left(x-\infty\right)\mid h\left(x,y\right)$.
The Theorem 2 in [@ABS2] and earlier (in a different context) [@biquad] gives a complete classification of biquadratics up to Möbius transformations. In particular, it can be shown that a biquadratic is degenerate if and only if $i_{3}=0$, where for a biquadratic $h\left(x,y\right)$ its *relative invariant* $i_{3}$ is defined by $$i_{3}\left(h,x,y\right)=\frac{1}{4}\det\begin{pmatrix}h&h_{x}&h_{xx}\\h_{y}&h_{xy}&h_{xxy}\\h_{yy}&h_{xyy}&h_{xxyy}\end{pmatrix}.$$ A multi-affine polynomial $Q$ is of *type Q* if all its biquadratics are non-degenerate. Otherwise it is of *type [H^4^]{}* if four out of six biquadratics are degenerate and of *type [H^6^]{}* if all six biquadratics are degenerate. According to Lemmas \[gegenueber\] and \[nicht2\] which we will prove later there are no other possibilities for $Q$.
For every permutation $\left(i,j,k,\ell\right)$ of $\left(1,2,3,4\right)$ the quartic polynomial $$r^{i}=r^{i}\left(x_{i}\right)=\left(Q^{j,k}_{x_{\ell}}\right)^{2}-2Q^{j,k}Q^{j,k}_{x_{\ell},x_{\ell}}$$ is called a corresponding *discriminant*. This polynomial turns out to be independent on permutations of $\left(j,k,\ell\right)$
Let ${\mathcal{P}}_{n}^{m}$ denote the set of polynomials in $n$ variables which are of degree $m$ in each variable. We consider the following action of Möbius transformations on polynomials $f\in{\mathcal{P}}_{n}^{m}$: $$M\left[f\right]\left(x_{1},\ldots,x_{n}\right)=\left(c_{1}x_{1}+d_{1}\right)^{m}\cdots\left(c_{n}x_{n}+d_{n}\right)^{m}f\left(\frac{a_{1}x_{1}+b_{1}}{c_{1}x_{1}+d_{1}},\ldots,\frac{a_{n}x_{n}+b_{n}}{c_{n}x_{n}+d_{n}}\right),$$ where $a_{i}d_{i}-b_{i}c_{i}\neq0$. The group $\left({\text{M\"ob}}\right)^{4}$ acts on quad-equations by Möbius transformations on all fields independently.
We will now present a complete classification of quad-equations on single quadrilaterals. We will not give the complete proofs here, because they are to long. However, in Section \[tools\] we give an overview of the most important ingredients of this proofs.
Quad-Equations of Type Q
------------------------
Quad-equations of type Q were already classified in [@ABS2]. Every quad-equation of type Q is equivalent modulo $\left({\text{M\"ob}}\right)^{4}$ to one of the following quad-equations characterized by the quadruples of discriminants:
- $\left(\delta, \delta, \delta, \delta\right)$: $$Q=\alpha_{2}\left(x_{1}-x_{2}\right)\left(x_{3}-x_{4}\right)-\alpha_{1}\left(x_{1}-x_{4}\right)\left(x_{2}-x_{3}\right)+\delta\alpha_{1}\alpha_{2}\left(\alpha_{1}+\alpha_{2}\right)\tag{$Q_{1}$}$$
- $\left(x_{1},x_{2},x_{3},x_{4}\right)$: $$\begin{gathered}
Q=\alpha_{2}\left(x_{1}-x_{2}\right)\left(x_{3}-x_{4}\right)-\alpha_{1}\left(x_{1}-x_{4}\right)\left(x_{2}-x_{3}\right)\\
+\alpha_{1}\alpha_{2}\left(\alpha_{1}+\alpha_{2}\right)\left(x_{1}+x_{2}+x_{3}+x_{4}\right)
-\alpha_{1}\alpha_{2}\left(\alpha_{1}+\alpha_{2}\right)\left(\alpha_{1}^{2}+\alpha_{1}\alpha_{2}+\alpha_{2}^{2}\right)\tag{$Q_{2}$}\end{gathered}$$
- $\left(x_{1}^{2}-\delta,x_{2}^{2}-\delta,x_{3}^{2}-\delta,x_{4}^{2}-\delta\right)$: $$\begin{gathered}
Q=\left(\alpha_{1}-\alpha_{1}^{-1}\right)\left(x_{1}x_{2}+x_{3}x_{4}\right)+\left(\alpha_{2}-\alpha_{2}^{-1}\right)\left(x_{1}x_{4}+x_{2}x_{3}\right)\\
-\left(\alpha_{1}\alpha_{2}-\alpha_{1}^{-1}\alpha_{2}^{-1}\right)\left(x_{1}x_{3}+x_{2}x_{4}\right)+\frac{\delta}{4}\left(\alpha_{1}-\alpha_{1}^{-1}\right)\left(\alpha_{2}-\alpha_{2}^{-1}\right)\left(\alpha_{1}\alpha_{2}-\alpha_{1}^{-1}\alpha_{2}^{-1}\right)\tag{$Q_{3}$}\end{gathered}$$
- $\left(\left(x_{1}-1\right)\left(k^{2}x_{1}^{2}-1\right),\left(x_{2}-1\right)\left(k^{2}x_{2}^{2}-1\right),\left(x_{3}-1\right)\left(k^{2}x_{3}^{2}-1\right),\left(x_{4}-1\right)\left(k^{2}x_{4}^{2}-1\right)\right)$: $$\begin{gathered}
Q=\operatorname{sn}\left(\alpha_{1}\right)\operatorname{sn}\left(\alpha_{2}\right)\operatorname{sn}\left(\alpha_{1}+\alpha_{2}\right)\left(k^{2}x_{1}x_{2}x_{3}x_{4}+1\right)-\operatorname{sn}\left(\alpha_{1}\right)\left(x_{1}x_{2}+x_{3}x_{4}\right)\\
-\operatorname{sn}\left(\alpha_{2}\right)\left(x_{1}x_{4}+x_{2}x_{3}\right)+\operatorname{sn}\left(\alpha_{1}+\alpha_{2}\right)\left(x_{1}x_{3}+x_{2}x_{4}\right)\tag{$Q_{4}$}\end{gathered}$$
Quad-Equations of Type H4 and H6
--------------------------------
A complete classification of type [H^4^]{} and of type [H^6^]{} quad-equations did not appear in the literature before. Every quad-equation of type [H^4^]{} is equivalent modulo $\left({\text{M\"ob}}\right)^{4}$ to one of the following quad-equations characterized by the quadruples of discriminants:
- $\left(\epsilon,0,\epsilon,0\right)$: $$Q=\left(x_{1}-x_{3}\right)\left(x_{2}-x_{4}\right)+\left(\alpha_{2}-\alpha_{1}\right)\left(1+\epsilon x_{2}x_{4}\right)\tag{$H_{1}^{\epsilon}$}$$
- $\left(\epsilon x_{1}+1,1,\epsilon x_{3}+1,1\right)$: $$\begin{gathered}
Q=\left(x_{1}-x_{3}\right)\left(x_{2}-x_{4}\right)+\left(\alpha_{2}-\alpha_{1}\right)\left(x_{1}+x_{2}+x_{3}+x_{4}\right)+\alpha_{2}^{2}-\alpha_{1}^{2}\\
+\epsilon\left(\alpha_{2}-\alpha_{1}\right)\left(2 x_{2}+\alpha_{1}+\alpha_{2}\right)\left(2 x_{4}+\alpha_{1}+\alpha_{2}\right)+\epsilon\left(\alpha_{2}-\alpha_{1}\right)^{3}\tag{$H_{2}^{\epsilon}$}\end{gathered}$$
- $\left(x_{1}^{2}+\delta\epsilon,x_{2}^{2},x_{3}^{2}+\delta\epsilon,x_{4}^{2}\right)$: $$Q=\alpha_{1}\left(x_{1}x_{2}+x_{3}x_{4}\right)-\alpha_{2}\left(x_{1}x_{4}+x_{2}x_{3}\right)+\left(\alpha_{1}^{2}-\alpha_{2}^{2}\right)\left(\delta+\frac{\epsilon x_{2}x_{4}}{\alpha_{1}\alpha_{2}}\right)\tag{$H_{3}^{\epsilon}$}$$
All these equations were already mentioned in [@ABS2].
Every quad-equation of type [H^6^]{} is equivalent modulo $\left({\text{M\"ob}}\right)^{4}$ to one of the following quad-equations characterized by the quadruples of discriminants:
- $\left(0,0,0,0\right)$: $$Q=x_{1}+x_{2}+x_{3}+x_{4}$$
- $\left(1,0,1,\delta\right)$: $$Q=x_{1}+x_{3}+x_{2}\left(x_{4}+\delta x_{1}\right)$$
- $\left(x_{1}^{2},x_{2}^{2},x_{3}^{2},x_{4}^{2}\right)$: $$Q=x_{1}x_{3}+x_{2}x_{4}+\delta_{1}x_{2}x_{3}+\delta_{2}x_{3}x_{4}$$
Ingredients of the proofs {#tools}
-------------------------
We will now present some ingredients of the proofs needed for the classification of quad-equations. At this point we will repeat two formulas already given in [@ABS2]: $$\label{hformel} 4i_{3}\left(Q^{1,2},x_{1},x_{2}\right)Q^{1,4}=\det\begin{pmatrix}Q^{1,2}&Q^{1,2}_{x_{1}}&\ell\\Q^{1,2}_{x_{2}}& Q^{1,2}_{x_{1}x_{2}}&\ell_{x_{2}}\\Q^{1,2}_{x_{2}x_{2}}&Q^{1,2}_{x_{1}x_{2}x_{2}}&\ell_{x_{2}x_{2}}\end{pmatrix},$$ where $$\ell=Q^{2,3}_{x_{3}x_{3}}Q^{3,4}-Q^{2,3}_{x_{3}}Q^{3,4}_{x_{3}}+Q^{2,3}Q^{3,4}_{x_{3}x_{3}}$$ and $$\label{Qformel} \frac{2Q_{x_{1}}}{Q}=\frac{Q^{1,2}_{x_{1}}Q^{3,4}-Q^{1,4}_{x_{1}}Q^{2,3}+Q^{2,3}Q^{3,4}_{x_{3}}-Q^{2,3}_{x_{3}}Q^{3,4}}{Q^{1,2}Q^{3,4}-Q^{1,4}Q^{2,3}}.$$ The following Lemma gives some informations about the relation between non-degenerate biquadratics of a quad-equation:
\[gegenueber\] Biquadratics on opposite edges (we consider the two diagonals as opposite edges, too) are either both degenerate or both non-degenerate.
A biquadratic $Q^{i,j}$ is degenerate if and only if $i_{3}=0$ holds. Due to [@ABS2] $i_{3}$ is equal for biquadratics on opposite edges.
Moreover, it follows that the number of non-degenerate biquadratics of a quad-equation is even. Another restriction for biquadratics of a quad-equation comes along with the next Lemma:
\[nicht2\] There do not exist any quad-equation with exactly two degenerate biquadratics.
*Assumption:* $Q$ is such a quad-equation. Change variables in a way, that the two degenerate biquadratics are $Q^{1,3}$ and $Q^{2,4}$. Then, all biquadratics on edges are non-degenerate. According to [@ABS2] $Q$ is of Type Q and all biquadratics are non-degenerate. Contradiction!
In addition, one can show, that $Q^{i,j}\not\equiv0$:
Every biquadratic of a quad-equation is not the zero polynomial.
By a simple calculation one can show $$\label{2biquads} Q\left(Q_{x_{2}}^{1,2}-Q_{x_{3}}^{1,3}\right)=2\left(Q_{x_{3}}Q^{1,3}-Q_{x_{2}}Q^{1,2}\right).$$ Let $Q^{1,2}=0$. Then, due to the classification of biquadratics we get $r^{1}=r^{2}= 0$. Assume that $Q^{1,3}\neq0$. We have to consider two cases:
- If $Q^{1,3}$ is non-degenerate, suitable Möbius transformations in $x_{1}$ and $x_{3}$ lead to $Q^{1,3}=\left(x_{1}-x_{3}\right)^{2}$. In the same manner, we get $Q^{2,4}=\left(x_{2}-x_{4}\right)^{2}$, and using we get $Q^{3,4}=0$. Now, one can apply and arrive $Q=\left(x_{1}-x_{3}\right)\left(x_{2}-x_{4}\right)$.
- Otherwise, up to Möbius transformation in $x_{3}$ we get $Q_{x_{3}}^{1,3}=0$ according to the classification of biquadratics. Using we get $Q_{x_{3}}=0$.
Both cases are not possible because $Q$ is irreducible. Therefore, all biquadratics must be zero polynomials. Consider $$\left(\log Q\right)_{x_{1}x_{2}}=\left(\frac{Q_{x_{1}}}{Q}\right)_{x_{2}}=\frac{QQ_{x_{1}x_{2}}-Q_{x_{1}}Q_{x_{2}}}{Q^{2}}=-\frac{h^{3,4}}{Q^{2}}=0$$ and in the same manner $$\left(\log Q\right)_{x_{1}x_{3}}=\left(\log Q\right)_{x_{1}x_{4}}=\left(\log Q\right)_{x_{2}x_{3}}=\left(\log Q\right)_{x_{2}x_{4}}=\left(\log Q\right)_{x_{3}x_{4}}=0.$$ Therefore, $$\log Q=\phi_{1}\left(x_{1}\right)+\phi_{2}\left(x_{1}\right)+\phi_{3}\left(x_{3}\right)+\phi_{4}\left(x_{4}\right)$$ and, furthermore, $$Q=\phi_{1}\left(x_{1}\right)\phi_{2}\left(x_{1}\right)\phi_{3}\left(x_{3}\right)\phi_{4}\left(x_{4}\right)$$ with $\phi_{i}\left(x_{i}\right)=\alpha_{i}x_{i}+\beta_{i}$ which is reducible. Contradiction!
Moreover, we also have the following lemma concerning vanishing biquadratics:
\[110\] There is no solution $\left(x_{1},x_{2},x_{3},x_{4}\right)$ of $Q\left(x_{1},x_{2},x_{3},x_{4}\right)=0$ with $Q^{1,2}\neq0$, $Q^{1,3}\neq0$ and $Q^{1,4}=0$.
Let $\left(x_{1},x_{2},x_{3},x_{4}\right)$ of $Q\left(x_{1},x_{2},x_{3},x_{4}\right)=0$ with $Q^{1,2}\neq0$, $Q^{1,3}\neq0$ and $Q^{1,4}=0$. Then, $Q^{1,2}\neq0$ leads to $Q_{x_{3}}\neq0$ and $Q^{1,3}\neq0$ leads to $Q_{x_{2}}\neq0$. This is a contradiction to $Q_{x_{2}}Q_{x_{3}}=0$ which is equivalent to $Q^{1,4}=0$.
Now, we are able to proof the following lemma:
\[facdeg\] Every factor of degeneracy is a factor of at least two biquadratics, that means if $\left(x-\alpha\right)\mid h^{1,2}$ then $$\left(x_{1}-\alpha\right)\mid h^{1,3}\ \text{or}\ \left(x_{1}-\alpha\right)\mid h^{1,4}.$$
We have to consider the following cases:
1. $\deg_{x_{1}}h^{1,4}\in\left\{0,1\right\}$
Considering the Möbius transformation $x_{1}\mapsto x_{1}+\alpha$ we have to show: If $x_{1}\mid h^{1,2}$, then $x_{1}\mid h^{1,3}$ or $x_{1}\mid h^{1,4}$.
*Assumption:* $x_{1}\mid h^{1,2}$ but $x_{1}\nmid h^{1,3}$ and $x_{1}\nmid h^{1,4}$.
We set $N_{i}:=\left\{x\in{\mathbb{CP}}^{1}:h^{1,i}\left(0,x\right)=0\right\}$ with $i\in\left\{3,4\right\}$. Obviously, $\left|N_{3}\right|,\left|N_{4}\right|<\infty$. Due to Lemma \[110\] there exists no solution $\left(x_{1},x_{2},x_{3},x_{4}\right)$ of $Q\left(x_{1},x_{2},x_{3},x_{4}\right)=0$ with $x_{1}=0$, $x_{3}\notin N_{3}$ and $x_{4}\notin N_{4}$.
$Q$ can be written as $Q=px_{1}+q$ with $p,q\in{\mathbb{C}}\left[x_{2},x_{3},x_{4}\right]$. $q_{x_{2}}=0$ (if not, $q=\tilde{p}x_{2}+\tilde{q}$ with $\tilde{p},\tilde{q}\in{\mathbb{C}}\left[x_{3},x_{4}\right]$ and $\tilde{p}\neq0$. Therefore, there would be $x_{3},x_{4}\in{\mathbb{C}}\setminus\left(N_{3}\cup N_{4}\right)$, such that $\tilde{p}\left(x_{3},x_{4}\right)\neq0$ and therefore, $$Q\left(0,-\frac{\tilde{q}\left(x_{3},x_{4}\right)}{\tilde{p}\left(x_{3},x_{4}\right)},x_{3},x_{4}\right)=0).$$ In the same manner one can show, that $q_{x_{3}}=q_{x_{4}}=0$ and therefore without restriction $q=1$.
Then, $h^{1,4}=p_{x_{2}}x_{1}p_{x_{3}}x_{1}-p_{x_{2},x_{3}}x_{1}\left(px_{1}+1\right)$. From $\deg_{x_{1}}h^{1,4}\leq1$ there follows, $p_{x_{2}}p_{x_{3}}-p_{x_{2},x_{3}}p=0$ and therefore $h^{1,4}=-p_{x_{2}x_{3}}x_{1}$. Contradiction!
2. $h^{1,4}=\left(x_{1}-\epsilon\right)^{2}X$ with $X\in{\mathbb{C}}\left[x_{4}\right]$
If $\epsilon\neq\alpha$, we reach the first case using the Möbius transformation $x_{1}\mapsto \frac{\epsilon x_{1}+1}{x_{1}}$ which leads to $h^{1,4}=X$.
3. $h^{1,4}=\left(x_{1}-\epsilon\right)\left(x_{1}-\hat{\epsilon}\right)X$ with $X\in{\mathbb{C}}\left[x_{4}\right]$ and $\epsilon\neq\hat{\epsilon}$
If $\epsilon\neq\alpha$ and $\hat{\epsilon}\neq\alpha$, we reach the first case using the Möbius transformation $x_{1}\mapsto \frac{\epsilon x_{1}+1}{x_{1}}$ or $x_{1}\mapsto \frac{\hat{\epsilon} x_{1}+1}{x_{1}}$ which leads to $h^{1,4}=\left(x_{1}\pm\frac{1}{\hat{\epsilon}-\epsilon}\right)X$.
These Lemmas are the necessary tools for the classification of quad-equations.
Quad-Equations on the Faces of a Cube {#cubes}
=====================================
We will now consider systems of the type $$\begin{aligned}
\label{system} A\left(x,x_{1},x_{2},x_{12}\right)&=0,&
\bar{A}\left(x_{3},x_{13},x_{23},x_{123}\right)&=0,\notag\\
B\left(x,x_{2},x_{3},x_{23}\right)&=0,&
\bar{B}\left(x_{1},x_{12},x_{13},x_{123}\right)&=0,\\
C\left(x,x_{1},x_{3},x_{13}\right)&=0,&
\bar{C}\left(x_{2},x_{12},x_{23},x_{123}\right)&=0\notag\end{aligned}$$ where the equations $A,\ldots,\bar{C}$ are quad-equations assigned to the faces of a cube in the manner demonstrated in Figure \[fig:cube\]. Such a system is *3D consistent* if the three values for $x_{123}$ (calculated by using $\bar{A}=0$, $\bar{B}=0$ or $\bar{C}=0$) coincide for arbitrary initial data $x$, $x_{1}$, $x_{2}$ and $x_{3}$. It possesses the *tetrahedron property* if there exist two polynomials $K$ and $\bar{K}$ such that the equations $$\begin{aligned}
&K\left(x,x_{12},x_{13},x_{23}\right)=0& \text{ and}&
&\bar{K}\left(x_{1},x_{2},x_{3},x_{123}\right)=0\end{aligned}$$ are satisfied for every solution of the system. It can be shown that the polynomials $K$ and $\bar{K}$ are multi-affine and irreducible. For this proof we use the following Lemma:
\[3biquads\]Consider a 3D consistent system and $$\begin{aligned}
F\left(x,x_{1},x_{2},x_{3},x_{123}\right)&=\bar{A}_{x_{13},x_{23}}BC-\bar{A}_{x_{23}}BC_{x_{13}}-\bar{A}_{x_{13}}B_{x_{23}}C+\bar{A}B_{x_{23}}C_{x_{13}},\\
G\left(x,x_{1},x_{2},x_{3},x_{123}\right)&=\bar{B}_{x_{12},x_{13}}CA-\bar{B}_{x_{13}}CA_{x_{12}}-\bar{B}_{x_{12}}C_{x_{13}}A+\bar{B}C_{x_{13}}A_{x_{12}},\\
H\left(x,x_{1},x_{2},x_{3},x_{123}\right)&=\bar{C}_{x_{12},x_{23}}AB-\bar{C}_{x_{12}}AB_{x_{23}}-\bar{C}_{x_{23}}A_{x_{12}}B+\bar{C}A_{x_{12}}B_{x_{23}}.\end{aligned}$$ Then, $$F=G=H=0$$ and $$F_{x_{1}}F_{x_{2}}-F_{x_{1},x_{2}}F=B^{0,3}C^{0,3}\bar{A}^{3,123}.$$
We get the equations $$F=G=H=0$$ by eliminating $x_{12}$, $x_{13}$ and $x_{23}$ in the system , and we have simply to factorize $F_{x_{1}}F_{x_{2}}-F_{x_{1},x_{2}}F$ to get the second statement.
This allows us to proof the following Lemma:
\[Kirreducible\] Consider a 3D consistent system possessing the tetrahedron property described by the two equations $$\begin{aligned}
K\left(x,x_{12},x_{13},x_{23}\right)&=0\\
\bar{K}\left(x_{1},x_{2},x_{3},x_{123}\right)&=0.\end{aligned}$$ Then, $K$ and $\bar{K}$ are multi-affine, irreducible polynomials.
Consider the system . The elimination of $x_{12}$, $x_{13}$ and $x_{23}$ leads to $$\begin{aligned}
F\left(\overset{2}{x},\overset{1}{x}_{1},\overset{1}{x}_{2},\overset{3}{x}_{3},\overset{1}{x}_{123}\right)&=\bar{A}_{x_{13},x_{23}}BC-\bar{A}_{x_{23}}BC_{x_{13}}-\bar{A}_{x_{13}}B_{x_{23}}C+\bar{A}B_{x_{23}}C_{x_{13}}=0,\\
G\left(\overset{2}{x},\overset{3}{x}_{1},\overset{1}{x}_{2},\overset{1}{x}_{3},\overset{1}{x}_{123}\right)&=\bar{B}_{x_{12},x_{13}}CA-\bar{B}_{x_{13}}CA_{x_{12}}-\bar{B}_{x_{12}}C_{x_{13}}A+\bar{B}C_{x_{13}}A_{x_{12}}=0,\\
H\left(\overset{2}{x},\overset{1}{x}_{1},\overset{3}{x}_{2},\overset{1}{x}_{3},\overset{1}{x}_{123}\right)&=\bar{C}_{x_{12},x_{23}}AB-\bar{C}_{x_{12}}AB_{x_{23}}-\bar{C}_{x_{23}}A_{x_{12}}B+\bar{C}A_{x_{12}}B_{x_{23}}=0\end{aligned}$$ where the numbers over the arguments of $F$, $G$ and $H$ indicate their degrees in the corresponding variables. These degrees are in the projective sense, that is in agreement with the action of Möbius transformations. Therefore the polynomials $F$, $G$ and $H$ must factorize as: $$\begin{aligned}
F&=f\left(x,\overset{2}{x}_{3}\right)L,&
G&=g\left(x,\overset{2}{x}_{1}\right)L,&
H&=h\left(x,\overset{2}{x}_{2}\right)L,&
L=L\left(x,\overset{1}{x}_{1},\overset{1}{x}_{2},\overset{1}{x}_{3},\overset{1}{x}_{123}\right)\end{aligned}$$ and therefore $L=k\left(x,x_{1},x_{2},x_{3},x_{123}\right)\bar{K}$. Therefore, $\bar{K}$ is multi-affine.
Assume, that $\bar{K}$ is reducible. Then, without restriction $\bar{K}=x_{1}d\left(x_{2},x_{3},x_{123}\right)$ or $\bar{K}=\left(x_{1}-x_{3}\right)d\left(x_{2},x_{123}\right)$ (otherwise, change the labeling and apply some Möbius transformations). In both cases $\bar{K}^{3,123}=0$.
Then, due to Lemma \[3biquads\] $$0=f^{2}k^{2}\bar{K}^{3,123}=F_{x_{1}}F_{x_{2}}-F_{x_{1},x_{2}}F=B^{0,3}C^{0,3}\bar{A}^{3,123}$$ which is a contradiction to $B^{0,3},C^{0,3},\bar{A}^{3,123}\not\equiv0$. Analogously, we can proof the same for $K$.
The group $\left({\text{M\"ob}}\right)^{8}$ acts on such a system by Möbius transformations on all vertex fields independently.
We classify all 3D consistent systems with a tetrahedron property, whereas the classification of the case, that all quad-equations are of type Q, was already done in [@ABS2]. Note, that in many cases the tetrahedron property is a consequence of the other assumptions.
There are two essential ideas which allow for this classification. The first one was already used in [@ABS2] and deals with the coincidence of biquadratics assigned to an edge but belonging to different faces. We will adapt the results from [@ABS2] to our situations. The second one is completely new and it can be interpreted as flipping certain vertices of a cube. We will present three theorems, two devoted to the first idea, one to the second one.
\[coincide1\] Consider a 3D consistent system with $B^{0,3}$ and $C^{0,3}$ are non-degenerate and
- $\bar{A}^{3,123}$ is non-degenerate or
- the discriminants of $B$ and $C$ corresponding to the vertices of $x$ and $x_{3}$ are not equal to zero.
Then:
1. possesses the tetrahedron property.
2. For any edge of the cube, the two biquadratics corresponding to this edge coincide up to a constant factor.
3. The product of this factors around one vertex is equal to $-1$; for example $$A^{0,1}B^{0,2}C^{0,3}+A^{0,2}B^{0,3}C^{0,1}=0.$$
The elimination of $x_{12}$, $x_{13}$ and $x_{23}$ leads to $$\begin{aligned}
F\left(\overset{2}{x},\overset{1}{x}_{1},\overset{1}{x}_{2},\overset{3}{x}_{3},\overset{1}{x}_{123}\right)&=\bar{A}_{x_{13},x_{23}}BC-\bar{A}_{x_{23}}BC_{x_{13}}-\bar{A}_{x_{13}}B_{x_{23}}C+\bar{A}B_{x_{23}}C_{x_{13}}=0,\\
G\left(\overset{2}{x},\overset{3}{x}_{1},\overset{1}{x}_{2},\overset{1}{x}_{3},\overset{1}{x}_{123}\right)&=\bar{B}_{x_{12},x_{13}}CA-\bar{B}_{x_{13}}CA_{x_{12}}-\bar{B}_{x_{12}}C_{x_{13}}A+\bar{B}C_{x_{13}}A_{x_{12}}=0,\\
H\left(\overset{2}{x},\overset{1}{x}_{1},\overset{3}{x}_{2},\overset{1}{x}_{3},\overset{1}{x}_{123}\right)&=\bar{C}_{x_{12},x_{23}}AB-\bar{C}_{x_{12}}AB_{x_{23}}-\bar{C}_{x_{23}}A_{x_{12}}B+\bar{C}A_{x_{12}}B_{x_{23}}=0\end{aligned}$$ where the numbers over the arguments of $F$, $G$ and $H$ indicate their degrees in the corresponding variables. Therefore the polynomials $F$, $G$ and $H$ must factorize as: $$\begin{aligned}
F&=f\left(x,\overset{2}{x}_{3}\right)L,&
G&=g\left(x,\overset{2}{x}_{1}\right)L,&
H&=h\left(x,\overset{2}{x}_{2}\right)L,&
L=L\left(x,\overset{1}{x}_{1},\overset{1}{x}_{2},\overset{1}{x}_{3},\overset{1}{x}_{123}\right).\end{aligned}$$ Then, due to Lemma \[3biquads\] $$f^{2}\left(L_{x_{1}}L_{x_{2}}-L_{x_{1},x_{2}}L\right)=F_{x_{1}}F_{x_{2}}-F_{x_{1},x_{2}}F=B^{0,3}C^{0,3}\bar{A}^{3,123}.$$ Consider first the case that $\bar{A}^{3,123}$ is non-degenerate. Then, $\bar{A}^{3,123}\mid\left(L_{x_{1}}L_{x_{2}}-L_{x_{1},x_{2}}L\right)$. Since $B^{0,3}$ and $C^{0,3}$ are non-degenerate, too, $B^{0,3}=f=C^{0,3}$ up to constant factors. In the other case $B^{0,3}$ and $C^{0,3}$ are not complete squares. Therefore, since $B^{0,3}$ and $C^{0,3}$ are non-degenerate $B^{0,3}C^{0,3}\mid f^{2}$ or $B^{0,3}C^{0,3}\mid\left(L_{x_{1}}L_{x_{2}}-L_{x_{1},x_{2}}L\right)$ which is not possible because of $\deg_{x}L=1$. Therefore, $B^{0,3}=f=C^{0,3}$ up to constant factors. Then, $\deg_{x}f=2$ and therefore $\deg_{x}L=0$, so the tetrahedron property is valid. According to [@ABS1] this is equivalent to $$A^{0,1}B^{0,2}C^{0,3}+A^{0,2}B^{0,3}C^{0,1}=0.$$ Therefore, $A^{0,1}/C^{0,1}$ can only depend on $x$ and not on $x_{1}$. Since for symmetry reasons also $$A^{0,1}\bar{B}^{1,12}C^{1,13}+A^{1,12}\bar{B}^{1,13}C^{0,1}=0$$ holds, $A^{0,1}/C^{0,1}$ is constant. This completes the proof.
\[coincide2\] Consider a 3D consistent system with
- all discriminants on diagonals of faces are non-degenerate and
- all discriminants not equal to zero.
Then:
1. For any edge of the cube, the two biquadratic polynomials corresponding to this edge coincide up to a constant factor.
2. If in addition system possesses the tetrahedron property, the product of this factors around one vertex is equal to $-1$; for example, $$A^{0,1}B^{0,2}C^{0,3}+A^{0,2}B^{0,3}C^{0,1}=0.$$
The elimination of $x_{12}$, $x_{13}$ and $x_{23}$ leads to $$\begin{aligned}
F\left(\overset{2}{x},\overset{1}{x}_{1},\overset{1}{x}_{2},\overset{3}{x}_{3},\overset{1}{x}_{123}\right)&=\bar{A}_{x_{13},x_{23}}BC-\bar{A}_{x_{23}}BC_{x_{13}}-\bar{A}_{x_{13}}B_{x_{23}}C+\bar{A}B_{x_{23}}C_{x_{13}}=0,\\
G\left(\overset{2}{x},\overset{3}{x}_{1},\overset{1}{x}_{2},\overset{1}{x}_{3},\overset{1}{x}_{123}\right)&=\bar{B}_{x_{12},x_{13}}CA-\bar{B}_{x_{13}}CA_{x_{12}}-\bar{B}_{x_{12}}C_{x_{13}}A+\bar{B}C_{x_{13}}A_{x_{12}}=0,\\
H\left(\overset{2}{x},\overset{1}{x}_{1},\overset{3}{x}_{2},\overset{1}{x}_{3},\overset{1}{x}_{123}\right)&=\bar{C}_{x_{12},x_{23}}AB-\bar{C}_{x_{12}}AB_{x_{23}}-\bar{C}_{x_{23}}A_{x_{12}}B+\bar{C}A_{x_{12}}B_{x_{23}}=0\end{aligned}$$ where the numbers over the arguments of $F$, $G$ and $H$ indicate their degrees in the corresponding variables. Therefore the polynomials $F$, $G$ and $H$ must factorize as: $$\begin{aligned}
F&=f\left(x,\overset{2}{x}_{3}\right)L,&
G&=g\left(x,\overset{2}{x}_{1}\right)L,&
H&=h\left(x,\overset{2}{x}_{2}\right)L,&
L=L\left(x,\overset{1}{x}_{1},\overset{1}{x}_{2},\overset{1}{x}_{3},\overset{1}{x}_{123}\right).\end{aligned}$$ Then, due to Lemma \[3biquads\] $$f^{2}\left(L_{x_{1}}L_{x_{2}}-L_{x_{1},x_{2}}L\right)=F_{x_{1}}F_{x_{2}}-F_{x_{1},x_{2}}F=B^{0,3}C^{0,3}\bar{A}^{3,123}$$ and since $\bar{A}^{3,123}$ is non-degenerate and $B^{0,3}$ and $C^{0,3}$ are not of Type $\left(0,0\right)$, i.e. $B^{0,3}$ and $C^{0,3}$ are not complete squares, we have $B^{0,3}=f\cdot k_{1}\left(x\right)$ and $C^{0,3}=f\cdot k_{2}\left(x\right)$ with some polynomials $k_{1}$ and $k_{2}$, so that $B^{0,3}/C^{0,3}$ can depend on $x$ only. Analogously, the elimination of $x_{1}$, $x_{2}$ and $x_{123}$ leads to $$\bar{f}^{2}\left(\bar{L}_{x_{13}}\bar{L}_{x_{23}}-\bar{L}_{x_{13},x_{23}}\bar{L}\right)=B^{0,3}C^{0,3}A^{0,12}.$$ Since $A^{0,12}$ is non-degenerate, too, we have $B^{0,3}=\bar{f}\cdot \bar{k}_{1}\left(x_{3}\right)$ and $C^{0,3}=\bar{f}\cdot \bar{k}_{2}\left(x_{3}\right)$ with some polynomials $\bar{k}_{1}$ and $\bar{k}_{2}$, so that $B^{0,3}/C^{0,3}$ can depend on $x_{3}$ only. Therefore, $B^{0,3}=\hat{k}C^{0,3}$ with $\hat{k}\in{\mathbb{C}}$.
In [@ABS1] it is shown, that the tetrahedron property is equivalent to $$A^{0,1}B^{0,2}C^{0,3}+A^{0,2}B^{0,3}C^{0,1}=0.$$ This completes the proof.
\[tetrahedronUse\] Consider a 3D consistent system possessing the tetrahedron property described by the two equations $$\begin{aligned}
K\left(x,x_{12},x_{13},x_{23}\right)&=0\\
\bar{K}\left(x_{1},x_{2},x_{3},x_{123}\right)&=0.\end{aligned}$$ Then, the system $$\begin{aligned}
\label{dualsystem} K\left(x,x_{12},x_{13},x_{23}\right)&=0,&
\bar{K}\left(x_{1},x_{2},x_{3},x_{123}\right)&=0,\notag\\
B\left(x,x_{2},x_{3},x_{23}\right)&=0,&
\bar{B}\left(x_{1},x_{12},x_{13},x_{123}\right)&=0,\\
C\left(x,x_{1},x_{3},x_{13}\right)&=0,&
\bar{C}\left(x_{2},x_{12},x_{23},x_{123}\right)&=0,\notag\end{aligned}$$ which can be assigned to a cube in the manner demonstrated in Figure \[fig:cube2\] on page , is 3D consistent and possesses the tetrahedron property. 3D consistency of is understood as the property of the initial value problem with initial date $x$, $x_{3}$, $x_{13}$ and $x_{23}$.
Let $x$, $x_{3}$, $x_{13}$ and $x_{23}$ be the initial data for the system . Then, we can calculate $x_{1}$ using $C\left(x,x_{1},x_{3},x_{13}\right)=0$, $x_{2}$ using $B\left(x,x_{2},x_{3},x_{23}\right)=0$ and $x_{12}$ using $K\left(x,x_{12},x_{13},x_{23}\right)=0$.
Furthermore, one can proof that $A\left(x,x_{1},x_{2},x_{12}\right)=0$ for this values of $x_{1}$, $x_{2}$ and $x_{12}$: Assume that $A\left(x,x_{1},x_{2},x_{12}\right)=0$ is not satisfied and let $x_{1}$ and $x_{2}$ be fixed. Then, we would get another value $\bar{x}_{12}$ for $x_{12}$ using $A\left(x,x_{1},x_{2},x_{12}\right)=0$. Since possesses the tetrahedron property, $K\left(x,\bar{x}_{12},x_{13},x_{23}\right)=0$ would hold, but due to Lemma \[Kirreducible\] this is a contradiction to $K\left(x,x_{12},x_{13},x_{23}\right)=0$.
With this values of $x_{1}$, $x_{2}$ and $x_{12}$ use the equations $\bar{B}\left(x_{1},x_{12},x_{13},x_{123}\right)=0$ and $\bar{C}\left(x_{2},x_{12},x_{23},x_{123}\right)=0$ to calculate two values for $x_{123}$ which are equal because of the 3D consistency of . Moreover, $\bar{A}\left(x_{3},x_{13},x_{23},x_{123}\right)=0$ is satisfied due to the 3D consistency of and $\bar{K}\left(x_{1},x_{2},x_{3},x_{123}\right)=0$ is satisfied, because possesses the tetrahedron property.
We will now present the classification. The proofs will only be given for the first two sections. The other proofs are quite analogous.
Six Equations of Type H4, First Case
------------------------------------
In this section we consider systems (\[system\]) with
- $A,\bar{A},\ldots,\bar{C}$ of type [H^4^]{} and
- all non-degenerate biquadratics on diagonals of faces.
Below is the list of all 3D consistent systems modulo $\left({\text{M\"ob}}\right)^{8}$ with these properties and with the tetrahedron property. It turns out that their tetrahedron property follows from the above assumptions except for the system characterized by the quadruple $\left(\epsilon,0,0,\epsilon\right)$.
\[Hvierfirst\] Every 3D consistent system (\[system\]) satisfying the properties of this section and possessing the tetrahedron property is equivalent modulo $\left({\text{M\"ob}}\right)^{8}$ to one of the following three systems. They are written in terms of two polynomials $A\left(x,x_{1},x_{2},x_{12};\alpha,\beta\right)$ and $K\left(x,x_{12},x_{13},x_{23};\epsilon\right)$ as $$\begin{aligned}
\bar{A}&=A\left(x_{13},x_{3},x_{123},x_{23};\alpha,\beta\right),&
B&=A\left(x,x_{2},x_{3},x_{23};\beta,\gamma\right),\\
\bar{B}&=A\left(x_{12},x_{1},x_{123},x_{13};\beta,\gamma\right),&
C&=A\left(x,x_{1},x_{3},x_{13};\alpha,\gamma\right),\\
\bar{C}&=A\left(x_{12},x_{2},x_{123},x_{23};\alpha,\gamma\right),&
\bar{K}&=K\left(x_{1},x_{2},x_{3},x_{123};0\right).\end{aligned}$$ The polynumials $A$ and $K$ can be characterized by the quadruples of discriminants of $A$:
- $\left(\epsilon,0,0,\epsilon\right)$: $$\begin{aligned}
A=&\left(x-x_{12}\right)\left(x_{1}-x_{2}\right)-\left(\alpha-\beta\right)\left(1+\epsilon x_{1}x_{2}\right),\\
\begin{split}K=&\left(\beta-\gamma\right)\left(x-x_{12}\right)\left(x_{13}-x_{23}\right)-\left(\alpha-\beta\right)\left(x-x_{23}\right)\left(x_{12}-x_{13}\right)\\
&-\epsilon\left(\alpha-\beta\right)\left(\beta-\gamma\right)\left(\alpha-\gamma\right)\end{split}\end{aligned}$$
- $\left(\epsilon x+1,1,1,\epsilon x_{12}+1\right)$: $$\begin{aligned}
\begin{split}A=&\left(x-x_{12}\right)\left(x_{1}-x_{2}\right)-\left(\alpha-\beta\right)\left(x+x_{1}+x_{2}+x_{12}\right)-\alpha^{2}+\beta^{2}\\
&-\epsilon\left(\alpha-\beta\right)\left(2x_{1}+\alpha+\beta\right)\left(2x_{2}+\alpha+\beta\right)-\epsilon\left(\alpha-\beta\right)^{3},\end{split}\\
\begin{split} K=&\left(\beta-\gamma\right)\left(x-x_{12}\right)\left(x_{13}-x_{23}\right)-\left(\alpha-\beta\right)\left(x-x_{23}\right)\left(x_{12}-x_{13}\right)\\
&+\left(\alpha-\beta\right)\left(\beta-\gamma\right)\left(\alpha-\gamma\right)-2\epsilon\left(\alpha-\beta\right)\left(\beta-\gamma\right)\left(\alpha-\gamma\right)\left(x+x_{12}+x_{13}+x_{23}\right)\\
&-4\epsilon^{2}\left(\alpha-\beta\right)\left(\beta-\gamma\right)\left(\alpha-\gamma\right)\left(\left(\alpha-\beta\right)^{2}+\left(\alpha-\beta\right)\left(\beta-\gamma\right)+\left(\beta-\gamma\right)^{2}\right)\end{split}\end{aligned}$$
- $\left(x^{2}+\delta\epsilon,x_{1}^{2},x_{2}^{2},x_{12}^{2}+\delta\epsilon\right)$: $$\begin{aligned}
A=&\alpha\left(xx_{1}+x_{2}x_{12}\right)-\beta\left(xx_{2}+x_{1}x_{12}\right)
+\left(\alpha^{2}-\beta^{2}\right)\left(\delta+\frac{\epsilon x_{1}x_{2}}{\alpha\beta}\right),\\
\begin{split}K=&\gamma\left(\alpha^{2}-\beta^{2}\right)\left(xx_{12}+x_{13}x_{23}\right)-\beta\left(\alpha^{2}-\gamma^{2}\right)\left(xx_{13}+x_{12}x_{23}\right)\\
&+\alpha\left(\beta^{2}-\gamma^{2}\right)\left(xx_{23}+x_{12}x_{13}\right)+\frac{\delta\epsilon\left(\alpha^{2}-\beta^{2}\right)\left(\alpha^{2}-\gamma^{2}\right)\left(\beta^{2}-\gamma^{2}\right)}{\alpha\beta\gamma}\end{split}\end{aligned}$$
We will start with systems characterized by $\left(\epsilon,0,0,\epsilon\right)$. In this case we suppose that the tetrahedron property holds. Due Section \[faces\] we have $$A=\left(x-x_{12}\right)\left(x_{1}-x_{2}\right)-\alpha\left(1+\epsilon x_{1}x_{2}\right)$$ up to Möbius transformations in $x$, $x_{1}$, $x_{2}$ and $x_{12}$. We have the biquadratics $$\begin{aligned}
&A^{0,1}=\alpha\left(1+\epsilon x_{1}^{2}\right),&
&A^{0,2}=-\alpha\left(1+\epsilon x_{2}^{2}\right),&
&A^{2,12}=\alpha\left(1+\epsilon x_{2}^{2}\right).\end{aligned}$$ The biquadratics $A^{0,2}$ and $B^{0,2}$ coincide up to a constant factor because of the tetrahedron property. Therefore, up to Möbius transformations in $x_{3}$ and $x_{23}$ we have $$B=\left(x-x_{23}\right)\left(x_{2}-x_{3}\right)-\beta\left(1+\epsilon x_{2}x_{3}\right)$$ with biquadratics $$\begin{aligned}
&B^{0,2}=\beta\left(1+\epsilon x_{2}^{2}\right),&
&B^{0,3}=-\beta\left(1+\epsilon x_{3}^{2}\right),&
&B^{2,23}=-\beta\left(1+\epsilon x_{2}^{2}\right).\end{aligned}$$ We have to keep in mind that Möbius transformations $x\mapsto \mu x+\nu$ and, if $\epsilon=0$, also $x_{2}\mapsto \mu_{2} x+\nu_{2}$ do not change $B^{0,2}$ up to a constant factor. However, the influence of these transformations on $B$ can be eliminated by $x_{23}\mapsto \mu x_{23}+\nu$ and $\beta\mapsto\mu\beta$ or, if $\epsilon=0$, by $x_{3}\mapsto \mu_{2}x_{3}+\nu_{2}$, $x_{23}\mapsto \mu x_{23}+\nu$ and $\beta\mapsto\mu\mu_{2}\beta$. Furthermore, the biquadratics $A^{0,1}$ and $C^{0,1}$ as well as $B^{0,3}$ and $C^{0,3}$ coincide up to a constant factor and, moreover, we have $$A^{0,1}B^{0,2}C^{0,3}+A^{0,2}B^{0,3}C^{0,1}=0.$$ Therefore, up to Möbius transformation in $x_{13}$ we have $$C=\left(x-x_{13}\right)\left(x_{1}-x_{3}\right)-\gamma\left(1+\epsilon x_{1}x_{3}\right)+\tilde{\gamma}\left(x-x_{13}\right)$$ with $\tilde{\gamma}=0$, if $\epsilon\neq0$, and biquadratics $$\begin{aligned}
&C^{0,1}=\gamma\left(1+\epsilon x_{1}^{2}\right),&
&C^{0,3}=\gamma\left(1+\epsilon x_{3}^{2}\right).\end{aligned}$$ Again, we have to keep in mind that Möbius transformations $x\mapsto \mu x+\nu$ and, if $\epsilon=0$, also $x_{1}\mapsto \mu_{1}x_{1}+\nu_{1}$ and $x_{3}\mapsto \mu_{1}x_{3}+\nu_{3}$ do not change $C^{0,1}$ and $C^{0,3}$ up to a common constant factor. However, the influence of these transformations on $C$ can be eliminated by $x_{13}\mapsto\mu x_{13}+\nu$ and $\gamma\mapsto\mu\gamma$ or, if $\epsilon=0$, $x_{13}\mapsto\mu x_{13}+\nu$, $\gamma\mapsto\mu\mu_{1}\gamma$ and $\tilde{\gamma}\mapsto \mu_{1}\tilde{\gamma}-\nu_{1}+\nu_{3}$.
From $A$, $B$ and $C$ one can derive $K$. However, $K$ is multi-affine and independent on $x_{1}$, $x_{2}$ and $x_{3}$ only if $\gamma=\alpha+\beta$ and $\tilde{\gamma}=0$ hold. We get $$K=\alpha\left(x-x_{23}\right)\left(x_{12}-x_{13}\right)-\beta\left(x-x_{12}\right)\left(x_{13}-x_{23}\right)+\epsilon\alpha\beta\left(\alpha+\beta\right)$$ with the biquadratic $$K^{12,23}=\alpha\beta\left(\left(x_{12}-x_{23}\right)^{2}+\epsilon\left(\alpha+\beta\right)^{2}\right).$$ In the same way as for $C$, we get, up to Möbius transformation in $x_{123}$, $$\bar{C}=\left(x_{2}-x_{123}\right)\left(x_{12}-x_{23}\right)-\gamma_{2}\left(1+\epsilon x_{12}x_{23}\right)+\tilde{\gamma}_{2}\left(x_{2}-x_{123}\right)$$ with the biquadratic $$\bar{C}^{12,23}=-\left(\left(x_{12}-x_{23}+\bar{\gamma}_{2}\right)^{2}+\epsilon\gamma_{2}^{2}\right).$$ Due to Theorem \[tetrahedronUse\] we have $\gamma_{2}=\alpha+\beta$ and $\bar{\gamma}_{2}=0$. $\bar{A}$, $\bar{B}$ and $\bar{K}$ can now easily derived from the other equations. After transformations $\alpha\mapsto\alpha-\beta$ and $\beta\mapsto\beta-\gamma$ we get the above system.
Now, we consider the systems characterized by $\left(\epsilon x+1,1,1,\epsilon x_{12}+1\right)$. In this case we do not suppose the tetrahedron property. We will show, that it follows from the above assumptions. Due to Section \[faces\] we have $$\begin{gathered}
B=\left(x-x_{23}\right)\left(x_{2}-x_{3}\right)-\left(\beta-\gamma\right)\left(x+x_{2}+x_{3}+x_{23}\right)-\beta^{2}+\gamma^{2}-\\\epsilon\left(\beta-\gamma\right)\left(2x_{2}+\beta+\gamma\right)\left(2x_{3}+\beta+\gamma\right)-\epsilon\left(\beta-\gamma\right)^{3}\end{gathered}$$ up to Möbius transformations in $x$, $x_{2}$, $x_{3}$ and $x_{23}$. We have the following biquadratics $$\begin{aligned}
B^{0,2}&=2\left(\beta-\gamma\right)\left(x+x_{2}+\beta+2\epsilon\left(x_{2}+\beta\right)^{2}\right),\\
B^{3,23}&=2\left(\beta-\gamma\right)\left(x_{3}+x_{23}+\beta+2\epsilon\left(x_{3}+\beta\right)^{2}\right),\\
B^{0,3}&=-2\left(\beta-\gamma\right)\left(x+x_{3}+\gamma+2\epsilon\left(x_{3}+\gamma\right)^{2}\right),\\
B^{2,23}&=-2\left(\beta-\gamma\right)\left(x_{2}+x_{23}+\gamma+2\epsilon\left(x_{2}+\gamma\right)^{2}\right).\end{aligned}$$ Due to Theorem \[coincide2\] the biquadratics $A^{0,2}$ and $B^{0,2}$ coincide up to a constant factor. Therefore, we have, up to Möbius transformations in $x_{1}$ and $x_{12}$, $$\begin{gathered}
A=\left(x-x_{12}\right)\left(x_{1}-x_{2}\right)-\left(\alpha-\beta\right)\left(x+x_{1}+x_{2}+x_{12}\right)-\alpha^{2}+\beta^{2}-\\\epsilon\left(\alpha-\beta\right)\left(2x_{1}+\beta+\alpha\right)\left(2x_{2}+\beta+\alpha\right)-\epsilon\left(\alpha-\beta\right)^{3}\end{gathered}$$ with biquadratics $$\begin{aligned}
A^{0,1}&=2\left(\alpha-\beta\right)\left(x+x_{1}+\alpha+2\epsilon\left(x_{1}+\alpha\right)^{2}\right),\\
A^{2,12}&=2\left(\alpha-\beta\right)\left(x_{2}+x_{12}+\alpha+2\epsilon\left(x_{2}+\alpha\right)^{2}\right),\\
A^{0,2}&=-2\left(\alpha-\beta\right)\left(x+x_{2}+\beta+2\epsilon\left(x_{2}+\beta\right)^{2}\right),\\
A^{1,12}&=-2\left(\alpha-\beta\right)\left(x_{1}+x_{12}+\beta+2\epsilon\left(x_{1}+\beta\right)^{2}\right).\end{aligned}$$ Furthermore, the biquadratics $A^{0,1}$ and $C^{0,1}$ as well as $B^{0,3}$ and $C^{0,3}$ coincide up to a constant factor, and therefore, we have, up to Möbius transformation in $x_{13}$, $$\begin{gathered}
C=\left(x-x_{13}\right)\left(x_{3}-x_{1}\right)-\left(\gamma-\alpha\right)\left(x+x_{1}+x_{3}+x_{13}\right)-\gamma^{2}+\alpha^{2}-\\\epsilon\left(\gamma-\alpha\right)\left(2x_{1}+\gamma+\alpha\right)\left(2x_{3}+\gamma+\alpha\right)-\epsilon\left(\gamma-\alpha\right)^{3}\end{gathered}$$ with biquadratics $$\begin{aligned}
C^{0,1}&=-2\left(\gamma-\alpha\right)\left(x+x_{1}+\alpha+2\epsilon\left(x_{1}+\alpha\right)^{2}\right),\\
C^{3,13}&=-2\left(\gamma-\alpha\right)\left(x_{3}+x_{13}+\alpha+2\epsilon\left(x_{3}+\alpha\right)^{2}\right),\\
C^{0,3}&=2\left(\gamma-\alpha\right)\left(x+x_{3}+\gamma+2\epsilon\left(x_{3}+\gamma\right)^{2}\right),\\
C^{1,13}&=2\left(\gamma-\alpha\right)\left(x_{1}+x_{13}+\gamma+2\epsilon\left(x_{1}+\gamma\right)^{2}\right).\end{aligned}$$ Moreover, the biquadratics $A^{2,12}$ and $\bar{C}^{2,12}$ as well as $B^{2,23}$ and $\bar{C}^{2,23}$ coincide up to a constant factor. Therefore, we have, up to Möbius transformation in $x_{123}$, $$\begin{gathered}
\bar{C}=\left(x_{2}-x_{123}\right)\left(x_{23}-x_{12}\right)-\left(\gamma-\alpha\right)\left(x_{2}+x_{12}+x_{23}+x_{123}\right)-\gamma^{2}+\alpha^{2}-\\\epsilon\left(\gamma-\alpha\right)\left(2x_{2}+\gamma+\alpha\right)\left(2x_{123}+\gamma+\alpha\right)-\epsilon\left(\gamma-\alpha\right)^{3}\end{gathered}$$ with biquadratics $$\begin{aligned}
\bar{C}^{2,12}&=-2\left(\gamma-\alpha\right)\left(x_{2}+x_{12}+\alpha+2\epsilon\left(x_{2}+\alpha\right)^{2}\right),\\
\bar{C}^{23,123}&=-2\left(\gamma-\alpha\right)\left(x_{23}+x_{123}+\alpha+2\epsilon\left(x_{123}+\alpha\right)^{2}\right),\\
\bar{C}^{2,23}&=2\left(\gamma-\alpha\right)\left(x_{2}+x_{23}+\gamma+2\epsilon\left(x_{2}+\gamma\right)^{2}\right),\\
\bar{C}^{12,123}&=2\left(\gamma-\alpha\right)\left(x_{12}+x_{123}+\gamma+2\epsilon\left(x_{123}+\gamma\right)^{2}\right).\end{aligned}$$ In addition, the biquadratics $\bar{A}^{3,13}$ and $C^{3,13}$, $\bar{A}^{3,23}$ and $B^{3,23}$ as well as $\bar{A}^{23,123}$ and $\bar{C}^{23,123}$ coincide up to a constant factor and therefore, we have $$\begin{gathered}
\bar{A}=\left(x_{3}-x_{123}\right)\left(x_{13}-x_{23}\right)-\left(\alpha-\beta\right)\left(x_{3}+x_{13}+x_{23}+x_{123}\right)-\alpha^{2}+\beta^{2}-\\\epsilon\left(\alpha-\beta\right)\left(2x_{3}+\beta+\alpha\right)\left(2x_{123}+\beta+\alpha\right)-\epsilon\left(\alpha-\beta\right)^{3}\end{gathered}$$ with biquadratics $$\begin{aligned}
\bar{A}^{3,13}&=2\left(\alpha-\beta\right)\left(x_{3}+x_{13}+\alpha+2\epsilon\left(x_{3}+\alpha\right)^{2}\right),\\
\bar{A}^{23,123}&=2\left(\alpha-\beta\right)\left(x_{23}+x_{123}+\alpha+2\epsilon\left(x_{123}+\alpha\right)^{2}\right),\\
\bar{A}^{3,23}&=-2\left(\alpha-\beta\right)\left(x_{3}+x_{23}+\beta+2\epsilon\left(x_{3}+\beta\right)^{2}\right),\\
\bar{A}^{13,123}&=-2\left(\alpha-\beta\right)\left(x_{13}+x_{123}+\beta+2\epsilon\left(x_{123}+\beta\right)^{2}\right).\end{aligned}$$ Nevertheless, $A^{1,12}$ and $\bar{B}^{1,12}$, $\bar{B}^{1,13}$ and $C^{1,13}$, $\bar{A}^{13,123}$ and $\bar{B}^{13,123}$ as well as $\bar{B}^{12,123}$ and $\bar{C}^{12,123}$ coincide up to a constant factor. Therefore, we have $$\begin{gathered}
\bar{B}=\left(x_{1}-x_{123}\right)\left(x_{12}-x_{13}\right)-\left(\beta-\gamma\right)\left(x_{1}+x_{12}+x_{13}+x_{123}\right)-\beta^{2}+\gamma^{2}-\\\epsilon\left(\beta-\gamma\right)\left(2x_{1}+\beta+\gamma\right)\left(2x_{123}+\beta+\gamma\right)-\epsilon\left(\beta-\gamma\right)^{3}\end{gathered}$$ with biquadratics $$\begin{aligned}
\bar{B}^{1,12}&=2\left(\beta-\gamma\right)\left(x_{1}+x_{12}+\beta+2\epsilon\left(x_{1}+\beta\right)^{2}\right),\\
\bar{B}^{13,123}&=2\left(\beta-\gamma\right)\left(x_{13}+x_{123}+\beta+2\epsilon\left(x_{123}+\beta\right)^{2}\right),\\
\bar{B}^{1,13}&=-2\left(\beta-\gamma\right)\left(x_{1}+x_{13}+\gamma+2\epsilon\left(x_{1}+\gamma\right)^{2}\right),\\
\bar{B}^{12,123}&=-2\left(\beta-\gamma\right)\left(x_{12}+x_{123}+\gamma+2\epsilon\left(x_{123}+\gamma\right)^{2}\right).\end{aligned}$$ Now, one can easily compute $K$ and $\bar{K}$ from the above equations. Therefore, the tetrahedron property holds for this system.
The next case will consider on systems characterized by $\left(x^{2},x_{1}^{2},x_{2}^{2},x_{12}^{2}\right)$ where all biquadratics on edges have four factors of degeneracy. In this case we suppose the tetrahedron property. Due to Section \[faces\] $$B=\beta\left(xx_{2}+x_{3}x_{23}\right)-\left(xx_{3}+x_{2}x_{23}\right)$$ up to Möbius transformations in $x$, $x_{2}$, $x_{3}$ and $x_{23}$. We have the following biquadratics: $$\begin{aligned}
B^{0,2}&=-\left(\beta^{2}-1\right)xx_{2},&
B^{0,3}&=\left(\beta^{2}-1\right)xx_{3},&
B^{2,23}&=\left(\beta^{2}-1\right)x_{2}x_{23}.\end{aligned}$$ The biquadratics $A^{0,2}$ and $B^{0,2}$ coincide up to a constant factor because of the tetrahedron property. Therefore, up to Möbius transformations in $x_{1}$ and $x_{12}$ we have $$A=\alpha\left(xx_{1}+x_{2}x_{12}\right)-\left(xx_{2}+x_{1}x_{12}\right)$$ with biquadratics $$\begin{aligned}
A^{0,1}&=-\left(\alpha^{2}-1\right)xx_{1},&
A^{0,2}&=\left(\alpha^{2}-1\right)xx_{2},&
A^{2,12}&=-\left(\alpha^{2}-1\right)x_{2}x_{12}.\end{aligned}$$ We have to keep in mind that Möbius transformations $x\mapsto x^{-1}$ and $x_{2}\mapsto x_{2}^{-1}$ do not change $A^{0,2}$ up to a constant factor. However, the influence of these transformations on $A$ can be eliminated by $x_{1}\mapsto x_{1}^{-1}$ and $x_{12}^{-1}$. Furthermore, the biquadratics $A^{0,1}$ and $C^{0,1}$ as well as $B^{0,3}$ and $C^{0,3}$ coincide up to a constant factor and, moreover, we have $$A^{0,1}B^{0,2}C^{0,3}+A^{0,2}B^{0,3}C^{0,1}=0.$$ Therefore, up to Möbius transformation in $x_{13}$ we have $$C=\gamma\left(xx_{3}+x_{1}x_{13}\right)-\left(xx_{1}+x_{3}x_{13}\right)$$ with biquadratics $$\begin{aligned}
C^{0,1}&=-\left(\gamma^{2}-1\right)xx_{1},&
C^{0,3}&=\left(\gamma^{2}-1\right)xx_{3}.\end{aligned}$$ Again, we have to keep in mind that Möbius transformations $x\mapsto x^{-1}$ as well as simultaneously $x_{1}\mapsto x_{1}^{\pm1}$ ans $x_{3}\mapsto x_{3}^{\pm1}$ do not change $C^{0,1}$ and $C^{0,3}$ up to a common constant factor. However, the influence of these transformations on $C$ can be eliminated by $x_{13}\mapsto x_{13}^{-1}$ as well as $\gamma\mapsto\gamma{\mp1}$.
From $A$, $B$ and $C$ one can derive $K$. However, $K$ is multi-affine and independent on $x_{1}$, $x_{2}$ and $x_{3}$ only if $\gamma=1/\left(\alpha\beta\right)$ hold. We get $$K=\beta\left(\alpha^{2}-1\right)\left(xx_{12}+x_{13}x_{23}\right)+\alpha\left(\beta^{2}-1\right)\left(xx_{23}+x_{12}x_{13}\right)-\left(\alpha^{2}\beta^{2}-1\right)\left(xx_{13}+x_{12}x_{23}\right)$$ with the biquadratic $$K^{12,23}=\left(\alpha^{2}-1\right)\left(\beta^{2}-1\right)\left(\alpha\beta x_{12}-x_{23}\right)\left(x_{12}-\alpha\beta x_{23}\right)$$ In the same way as for $C$, we get, up to Möbius transformation in $x_{123}$, $$\bar{C}=\gamma_{2}\left(x_{2}x_{23}+x_{12}x_{123}\right)-\left(x_{2}x_{12}+x_{23}x_{123}\right)$$ with the biquadratic $$\bar{C}^{12,23}=-\left(\gamma_{2} x_{12}-x_{23}\right)\left(x_{12}-\gamma_{2} x_{23}\right).$$ Due Theorem \[tetrahedronUse\] we have $\gamma_{2}=\alpha\beta$. $\bar{A}$, $\bar{B}$ and $\bar{K}$ can now easily derived from the other equations. After transformations $\alpha\mapsto\alpha/\beta$ and $\beta\mapsto\beta/\gamma$ we get the above system with $\delta=\epsilon=0$.
Our last case will consider systems characterized by $\left(x^{2}+\delta\epsilon,x_{1}^{2},x_{2}^{2},x_{12}^{2}+\delta\epsilon\right)$ where all biquadratics have at most two factors of degeneracy. In this case we will also not suppose the tetrahedron property because it follows from the above assumptions. Due to Section \[faces\] $$B=\beta\left(xx_{2}+x_{3}x_{23}\right)-\gamma\left(xx_{3}+x_{2}x_{23}\right)+\left(\beta^{2}-\gamma^{2}\right)\left(\delta+\frac{\epsilon x_{2}x_{3}}{\beta\gamma}\right)$$ up to Möbius transformations in $x$, $x_{2}$, $x_{3}$ and $x_{23}$ with $\delta\neq0$ or $\epsilon\neq0$. We have the following biquadratics: $$\begin{aligned}
B^{0,2}&=-\left(\beta^{2}-\gamma^{2}\right)\left(xx_{2}+\delta\beta+\frac{\epsilon x_{2}^{2}}{\beta}\right),&
B^{0,3}&=\left(\beta^{2}-\gamma^{2}\right)\left(xx_{3}+\delta\gamma+\frac{\epsilon x_{3}^{2}}{\gamma}\right),\\
B^{3,23}&=-\left(\beta^{2}-\gamma^{2}\right)\left(x_{3}x_{23}+\delta\beta+\frac{\epsilon x_{3}^{2}}{\beta}\right),&
B^{2,23}&=\left(\beta^{2}-\gamma^{2}\right)\left(x_{2}x_{23}+\delta\gamma+\frac{\epsilon x_{2}^{2}}{\gamma}\right).\end{aligned}$$ Due to Theorem \[coincide2\] the biquadratics $A^{0,2}$ and $B^{0,2}$ coincide up to a constant factor. Therefore, we have, up to Möbius transformations in $x_{1}$ and $x_{12}$, $$A=\alpha\left(xx_{1}+x_{2}x_{12}\right)-\beta\left(xx_{2}+x_{1}x_{12}\right)+\left(\alpha^{2}-\beta^{2}\right)\left(\delta+\frac{\epsilon x_{1}x_{2}}{\alpha\beta}\right)$$ with biquadratics $$\begin{aligned}
A^{0,1}&=-\left(\alpha^{2}-\beta^{2}\right)\left(xx_{1}+\delta\alpha+\frac{\epsilon x_{1}^{2}}{\alpha}\right),&
A^{0,2}&=\left(\alpha^{2}-\beta^{2}\right)\left(xx_{2}+\delta\beta+\frac{\epsilon x_{2}^{2}}{\beta}\right),\\
A^{2,12}&=-\left(\alpha^{2}-\beta^{2}\right)\left(x_{2}x_{12}+\delta\alpha+\frac{\epsilon x_{2}^{2}}{\alpha}\right),&
A^{1,12}&=\left(\alpha^{2}-\beta^{2}\right)\left(x_{1}x_{12}+\delta\beta+\frac{\epsilon x_{1}^{2}}{\beta}\right).\end{aligned}$$ Furthermore, the biquadratics $A^{0,1}$ and $C^{0,1}$ as well as $B^{0,3}$ and $C^{0,3}$ coincide up to a constant factor, and therefore, we have, up to Möbius transformation in $x_{13}$, $$C=\gamma\left(xx_{3}+x_{1}x_{13}\right)-\alpha\left(xx_{1}+x_{3}x_{13}\right)+\left(\gamma^{2}-\alpha^{2}\right)\left(\delta+\frac{\epsilon x_{1}x_{3}}{\alpha\gamma}\right)$$ with biquadratics $$\begin{aligned}
C^{0,1}&=-\left(\alpha^{2}-\gamma^{2}\right)\left(xx_{1}+\delta\alpha+\frac{\epsilon x_{1}^{2}}{\alpha}\right),&
C^{0,3}&=\left(\alpha^{2}-\gamma^{2}\right)\left(xx_{3}+\delta\gamma+\frac{\epsilon x_{3}^{2}}{\gamma}\right),\\
C^{3,13}&=-\left(\alpha^{2}-\gamma^{2}\right)\left(x_{3}x_{13}+\delta\alpha+\frac{\epsilon x_{3}^{2}}{\alpha}\right),&
C^{1,13}&=\left(\alpha^{2}-\gamma^{2}\right)\left(x_{1}x_{13}+\delta\gamma+\frac{\epsilon x_{1}^{2}}{\gamma}\right).\end{aligned}$$ Moreover, the biquadratics $A^{2,12}$ and $\bar{C}^{2,12}$ as well as $B^{2,23}$ and $\bar{C}^{2,23}$ coincide up to a constant factor. Therefore, we have, up to Möbius transformation in $x_{123}$, $$\bar{C}=\gamma\left(x_{2}x_{23}+x_{12}x_{123}\right)-\alpha\left(x_{2}x_{12}+x_{23}x_{123}\right)+\left(\gamma^{2}-\alpha^{2}\right)\left(\delta+\frac{\epsilon x_{2}x_{123}}{\alpha\gamma}\right)$$ with biquadratics $$\begin{aligned}
\bar{C}^{2,12}&=-\left(\alpha^{2}-\gamma^{2}\right)\left(x_{2}x_{12}+\delta\alpha+\frac{\epsilon x_{2}^{2}}{\alpha}\right),\\
\bar{C}^{23,123}&=-\left(\alpha^{2}-\gamma^{2}\right)\left(x_{23}x_{123}+\delta\alpha+\frac{\epsilon x_{123}^{2}}{\alpha}\right),\\
\bar{C}^{2,23}&=\left(\alpha^{2}-\gamma^{2}\right)\left(x_{2}x_{23}+\delta\gamma+\frac{\epsilon x_{2}^{2}}{\gamma}\right),\\
\bar{C}^{12,123}&=\left(\alpha^{2}-\gamma^{2}\right)\left(x_{12}x_{123}+\delta\gamma+\frac{\epsilon x_{123}^{2}}{\gamma}\right).\end{aligned}$$ In addition, the biquadratics $\bar{A}^{3,13}$ and $C^{3,13}$, $\bar{A}^{3,23}$ and $B^{3,23}$ as well as $\bar{A}^{23,123}$ and $\bar{C}^{23,123}$ coincide up to a constant factor and therefore, we have $$\bar{A}=\alpha\left(x_{3}x_{13}+x_{23}x_{123}\right)-\beta\left(x_{3}x_{23}+x_{13}x_{123}\right)+\left(\alpha^{2}-\beta^{2}\right)\left(\delta+\frac{\epsilon x_{3}x_{123}}{\alpha\beta}\right)$$ with biquadratics $$\begin{aligned}
\bar{A}^{3,13}&=-\left(\alpha^{2}-\beta^{2}\right)\left(x_{3}x_{13}+\delta\alpha+\frac{\epsilon x_{3}^{2}}{\alpha}\right),\\
\bar{A}^{23,123}&=-\left(\alpha^{2}-\beta^{2}\right)\left(x_{23}x_{123}+\delta\alpha+\frac{\epsilon x_{123}^{2}}{\alpha}\right),\\
\bar{A}^{3,23}&=\left(\alpha^{2}-\beta^{2}\right)\left(x_{3}x_{23}+\delta\beta+\frac{\epsilon x_{3}^{2}}{\beta}\right),\\
\bar{A}^{13,123}&=\left(\alpha^{2}-\beta^{2}\right)\left(x_{13}x_{123}+\delta\beta+\frac{\epsilon x_{123}^{2}}{\beta}\right).\end{aligned}$$ Nevertheless, $A^{1,12}$ and $\bar{B}^{1,12}$, $\bar{B}^{1,13}$ and $C^{1,13}$, $\bar{A}^{13,123}$ and $\bar{B}^{13,123}$ as well as $\bar{B}^{12,123}$ and $\bar{C}^{12,123}$ coincide up to a constant factor. Therefore, we have $$\bar{B}=\beta\left(x_{1}x_{12}+x_{13}x_{123}\right)-\gamma\left(x_{1}x_{13}+x_{12}x_{123}\right)+\left(\beta^{2}-\gamma^{2}\right)\left(\delta+\frac{\epsilon x_{1}x_{123}}{\beta\gamma}\right)$$ with biquadratics $$\begin{aligned}
\bar{B}^{1,12}&=-\left(\beta^{2}-\gamma^{2}\right)\left(x_{1}x_{12}+\delta\beta+\frac{\epsilon x_{1}^{2}}{\beta}\right),\\
\bar{B}^{13,123}&=-\left(\beta^{2}-\gamma^{2}\right)\left(x_{13}x_{123}+\delta\beta+\frac{\epsilon x_{123}^{2}}{\beta}\right),\\
\bar{B}^{1,13}&=\left(\beta^{2}-\gamma^{2}\right)\left(x_{1}x_{13}+\delta\gamma+\frac{\epsilon x_{1}^{2}}{\gamma}\right),\\
\bar{B}^{12,123}&=\left(\beta^{2}-\gamma^{2}\right)\left(x_{12}x_{123}+\delta\gamma+\frac{\epsilon x_{123}^{2}}{\gamma}\right).\end{aligned}$$ Now, one can easily compute $K$ and $\bar{K}$ from the above equations. Therefore, the tetrahedron property holds for this system.
Equivalent systems appeared in [@ABS1; @ABS2] without classification results for such systems.
Two Equations of Type Q and Four Equations of Type H4
-----------------------------------------------------
In this section we consider systems (\[system\]) with
- $B$ and $\bar{B}$ of type Q,
- $A$, $\bar{A}$, $C$ and $\bar{C}$ of type [H^4^]{} and
- the non-degenerate biquadratics of $\bar{A}$, $C$ and $\bar{C}$ on edges neighboring $B$ or $\bar{B}$.
Below is the list of all 3D consistent systems modulo $\left({\text{M\"ob}}\right)^{8}$ with these properties. It turns out that their tetrahedron property follows from the above assumptions.
Every 3D consistent system (\[system\]) satisfying the properties of this section is equivalent modulo $\left({\text{M\"ob}}\right)^{8}$ to one of the following three systems. They are written in terms of the two polynomials $A\left(x,x_{1},x_{2},x_{12};\alpha,\beta\right)$ and $B\left(x,x_{2},x_{3},x_{23};\epsilon\right)$ as $$\begin{aligned}
\bar{A}&=A\left(x_{3},x_{13},x_{23},x_{123};\alpha,\beta\right),&
\bar{B}&=B\left(x_{1},x_{12},x_{13},x_{123};0\right),\\
C&=A\left(x,x_{1},x_{3},x_{13};\alpha,\gamma\right),&
\bar{C}&=A\left(x_{2},x_{12},x_{23},x_{123};\alpha,\gamma\right),\\
K&=A\left(x,x_{12},x_{23},x_{13};\beta,\gamma\right),&
\bar{K}&=A\left(x_{2},x_{1},x_{3},x_{123};\beta,\gamma\right).\end{aligned}$$ The polynomials $A$ and $K$ can be characterized by the quadruples of discriminants of $A$:
- $\left(\epsilon,0,\epsilon,0\right)$: $$\begin{aligned}
A=&\left(x-x_{2}\right)\left(x_{1}-x_{12}\right)-\left(\alpha-\beta\right)\left(1+\epsilon x_{1}x_{12}\right),\\
\begin{split}B=&\left(\beta-\gamma\right)\left(x-x_{2}\right)\left(x_{3}-x_{23}\right)-\left(\alpha-\beta\right)\left(x-x_{23}\right)\left(x_{2}-x_{3}\right)\\
&-\epsilon\left(\alpha-\beta\right)\left(\beta-\gamma\right)\left(\alpha-\gamma\right)\end{split}\end{aligned}$$
- $\left(\epsilon x+1,1,\epsilon x_{2}+1,1\right)$: $$\begin{aligned}
\begin{split}A=&\left(x-x_{2}\right)\left(x_{1}-x_{12}\right)-\left(\alpha-\beta\right)\left(x+x_{1}+x_{2}+x_{12}\right)\\
&-\alpha^{2}+\beta^{2}-\epsilon\left(\alpha-\beta\right)\left(2x_{1}+\alpha+\beta\right)\left(2x_{12}+\alpha+\beta\right)-\epsilon\left(\alpha-\beta\right)^{3},\end{split}\\
\begin{split}B=&\left(\beta-\gamma\right)\left(x-x_{2}\right)\left(x_{3}-x_{23}\right)-\left(\alpha-\beta\right)\left(x-x_{23}\right)\left(x_{2}-x_{3}\right)\\
&+\left(\alpha-\beta\right)\left(\beta-\gamma\right)\left(\alpha-\gamma\right)-2\epsilon\left(\alpha-\beta\right)\left(\beta-\gamma\right)\left(\alpha-\gamma\right)\left(x+x_{2}+x_{3}+x_{23}\right)\\
&-4\epsilon^{2}\left(\alpha-\beta\right)\left(\beta-\gamma\right)\left(\alpha-\gamma\right)\left(\left(\alpha-\beta\right)^{2}+\left(\alpha-\beta\right)\left(\beta-\gamma\right)+\left(\beta-\gamma\right)^{2}\right)\end{split}\end{aligned}$$
- $\left(x^{2}+\delta\epsilon,x_{1}^{2},x_{2}^{2}+\delta\epsilon,x_{12}^{2}\right)$: $$\begin{aligned}
A=&\alpha\left(xx_{1}+x_{2}x_{12}\right)-\beta\left(xx_{12}+x_{1}x_{2}\right)+\left(\alpha^{2}-\beta^{2}\right)\left(\delta+\frac{\epsilon x_{1}x_{12}}{\alpha\beta}\right),\\
\begin{split}B=&\gamma\left(\alpha^{2}-\beta^{2}\right)\left(xx_{2}+x_{3}x_{23}\right)-\beta\left(\alpha^{2}-\gamma^{2}\right)\left(xx_{3}+x_{2}x_{23}\right)\\
&+\alpha\left(\beta^{2}-\gamma^{2}\right)\left(xx_{23}+x_{2}x_{3}\right)+\frac{\delta\epsilon\left(\alpha^{2}-\beta^{2}\right)\left(\alpha^{2}-\gamma^{2}\right)\left(\beta^{2}-\gamma^{2}\right)}{\alpha\beta\gamma}\end{split}\end{aligned}$$
Due to Theorem \[coincide1\] all systems we have to consider possesses the tetrahedron property. Therefore, using Theorem \[tetrahedronUse\] we get our systems by the flips $x_{2}\leftrightarrow x_{12}$ and $x_{3}\leftrightarrow x_{13}$ in the systems of Theorem \[Hvierfirst\].
Equivalent systems appeared in [@ABS2; @Atk1; @Todapaper] without classification results for such systems.
Six Equations of Type H4, Second Case
-------------------------------------
In this section we consider systems (\[system\]) with
- $A,\bar{A},\ldots,\bar{C}$ of type [H^4^]{},
- the non-degenerate biquadratics of $A$ and $\bar{A}$ on diagonals of faces and
- the non-degenerate biquadratics of $B$, $\bar{B}$, $C$ and $\bar{C}$ on edges not neighboring $A$ and $\bar{A}$.
Below is the list of all 3D consistent systems modulo $\left({\text{M\"ob}}\right)^{8}$ with these properties. It turns out that their tetrahedron property follows from the above assumptions.
\[Hviersecond\] Every 3D consistent system (\[system\]) satisfying the properties of this section is equivalent modulo $\left({\text{M\"ob}}\right)^{8}$ to one of the following three systems. They are written in terms of three polynomials $A\left(x,x_{1},x_{2},x_{12}\right)$, $B\left(x,x_{2},x_{3},x_{23};\beta\right)$ and $K\left(x,x_{12},x_{13},x_{23}\right)$ as $$\begin{aligned}
\bar{A}&=A\left(x_{3},x_{13},x_{23},x_{123}\right),&
\bar{B}&=B\left(x_{12},x_{1},x_{123},x_{13};\beta\right),&
C&=B\left(x,x_{1},x_{3},x_{13};\alpha\right),\\
\bar{C}&=B\left(x_{12},x_{2},x_{123},x_{23};\alpha\right),&
\bar{K}&=K\left(x_{3},x_{123},x_{1},x_{2}\right).\end{aligned}$$ The polynomials $A$, $B$ and $K$ can be characterized by the quadruples of discriminants of $A$:
- $\left(\epsilon,0,0,\epsilon\right)$: $$\begin{aligned}
&A=\left(x-x_{12}\right)\left(x_{1}-x_{2}\right)-\left(\alpha-\beta\right)\left(1+\epsilon x_{1}x_{2}\right),\\
&B=\left(x-x_{3}\right)\left(x_{2}-x_{23}\right)+\gamma\left(1+\epsilon x_{2}x_{23}\right),\\
&K=\left(x-x_{12}\right)\left(x_{13}-x_{23}\right)-\left(\alpha-\beta\right)\left(1+\epsilon x_{13}x_{23}\right)\end{aligned}$$
- $\left(\epsilon x+1,1,1,\epsilon x_{12}+1\right)$: $$\begin{aligned}
\begin{split}A=&\left(x-x_{12}\right)\left(x_{1}-x_{2}\right)-\left(\alpha-\beta\right)\left(x+x_{1}+x_{2}+x_{12}\right)-\alpha^{2}+\beta^{2}\\
&-\epsilon\left(\alpha-\beta\right)\left(2 x_{1}+\alpha+\beta\right)\left(2 x_{2}+\alpha+\beta\right)-\epsilon\left(\alpha-\beta\right)^{3},\end{split}\\
\begin{split}B=&\left(x-x_{3}\right)\left(x_{2}-x_{23}\right)+\gamma\left(x+x_{2}+x_{3}+x_{23}\right)+\gamma^{2}+2\beta\gamma\\
&+\epsilon \gamma\left(2 x_{2}+2\beta+\gamma\right)\left(2 x_{23}+2\beta+\gamma\right)+\gamma^{3}\epsilon,\end{split}\\
\begin{split}K=&\left(x-x_{12}\right)\left(x_{13}-x_{23}\right)-\left(\alpha-\beta\right)\left(x+x_{12}+x_{13}+x_{23}\right)-\alpha^{2}\beta^{2}\\
&-2\gamma\left(\alpha-\beta\right)-\epsilon\left(\alpha-\beta\right)\left(2x_{13}+\alpha+\beta+2\gamma\right)\left(2x_{23}+\alpha+\beta+2\gamma\right)-\epsilon\left(\alpha-\beta\right)^{3}\end{split}\end{aligned}$$
- $\left(x^{2}+\delta\epsilon,x_{1}^{2},x_{2}^{2},x_{12}^{2}+\delta\epsilon\right)$: $$\begin{aligned}
&A=\alpha\left(xx_{1}+x_{2}x_{12}\right)-\beta\left(xx_{2}+x_{1}x_{12}\right)+\left(\alpha^{2}-\beta^{2}\right)\left(\delta+\frac{\epsilon x_{1}x_{2}}{\alpha\beta}\right),\\
\begin{split}B=xx_{2}+x_{3}x_{23}-\gamma\left(xx_{23}+x_{2}x_{3}\right)-\left(\gamma^{2}-1\right)\left(\delta \beta +\frac{\epsilon x_{2}x_{23}}{\beta\gamma}\right),\end{split}\\
&K=\alpha\left(xx_{13}+x_{12}x_{23}\right)-\beta\left(xx_{23}+x_{12}x_{13}\right)+\left(\alpha^{2}-\beta^{2}\right)\left(\delta\gamma+\frac{\epsilon x_{13}x_{23}}{\alpha\beta\gamma}\right)\end{aligned}$$
These systems did not appear in the literature before.
Four Equations of Type H4 and Two Equations of Type H6, First Case
-------------------------------------------------------------------
No 3D consistent system (\[system\]) exists with the tetrahedron property and with
- $A$ and $\bar{A}$ of type [H^6^]{},
- $B$, $\bar{B}$, $C$ and $\bar{C}$ of type [H^4^]{},
- the non-degenerate biquadratics of $B$, $\bar{B}$, $C$ and $\bar{C}$ on diagonals.
Four Equations of Type H4 and Two Equations of Type H6, Second Case
--------------------------------------------------------------------
In this section we consider systems (\[system\]) with
- $A$ and $\bar{A}$ of type [H^6^]{},
- $B$, $\bar{B}$, $C$ and $\bar{C}$ of type [H^4^]{} and
- the non-degenerate biquadratics of $B$, $\bar{B}$, $C$ and $\bar{C}$ on edges not neighboring $A$ and $\bar{A}$.
Below is the list of all 3D consistent systems modulo $\left({\text{M\"ob}}\right)^{8}$ with these properties and with the tetrahedron property. It turns out that in the last case the tetrahedron property follows from the above assumptions.
\[Hsechs\]Every 3D consistent system (\[system\]) satisfying the properties of this section and possessing the tetrahedron property is equivalent modulo $\left({\text{M\"ob}}\right)^{8}$ to one of the following four systems which can be characterized by the quadruples of discriminants of $A$:
- $\left(0,0,0,0\right)$: $$\begin{aligned}
&A\left(x,x_{1},x_{2},x_{12}\right)=x+x_{1}+x_{2}+x_{12},\\
&B\left(x,x_{2},x_{3},x_{23};\alpha\right)=\left(x-x_{3}\right)\left(x_{2}-x_{23}\right)+\alpha,\end{aligned}$$ $$\begin{aligned}
\bar{A}&=A\left(x_{3},x_{13},x_{23},x_{123}\right),&
\bar{B}&=B\left(x_{1},x_{12},x_{13},x_{123};\alpha\right),\\
C&=B\left(x,x_{1},x_{3},x_{13};-\alpha\right),&
\bar{C}&=B\left(x_{2},x_{12},x_{23},x_{123};-\alpha\right),\\
K&=A\left(x,x_{12},x_{13},x_{23}\right),&
\bar{K}&=A\left(x_{1},x_{2},x_{3},x_{123}\right).\end{aligned}$$
- $\left(1,1,0,\delta_{1}\delta_{2}\right)$: $$\begin{aligned}
&B\left(x,x_{2},x_{3},x_{23};\delta_{2}\right)=\left(x-x_{3}\right)\left(x_{2}-x_{23}\right)+\delta_{1}\alpha\left(x_{2}+x_{23}\right)+\delta_{2}\alpha\left(x+x_{3}\right)+\delta_{1}\delta_{2}\alpha^{2},\\
\begin{split}\bar{B}\left(x_{1},x_{12},x_{13},x_{123};\delta_{1}\right)=\left(x_{1}-x_{13}\right)\left(x_{12}-x_{123}\right)+2\left(\delta_{1}-1\right)\alpha\\
+\left(\delta_{1}\delta_{2}+\delta_{1}-\delta_{2}\right)\alpha\left(x_{12}+x_{123}\right)+2\delta_{1}\delta_{2}\alpha x_{12}x_{123},\end{split}\\
&K\left(x,x_{13},x_{23},x_{12};\alpha\right)=x+x_{13}+\left(\delta_{1}-1\right)x_{23}+\delta_{1}\alpha+x_{12}\left(\delta_{2}x+\delta_{1}x_{23}+\delta_{1}\delta_{2}\alpha\right),\end{aligned}$$ $$\begin{aligned}
A&=K\left(x,x_{1},x_{2},x_{12};0\right),&
\bar{A}&=K\left(x_{3},x_{13},x_{23},x_{123}\right),\\
C&=B\left(x,x_{1},x_{3},x_{13};\delta_{1}+\delta_{2}-\delta_{1}\delta_{2}\right),&
\bar{C}&=\bar{B}\left(x_{2},x_{12},x_{23},x_{123};0\right),\\
\bar{K}&=K\left(x_{3},x_{1},x_{2},x_{123};\alpha\right).\end{aligned}$$
If $\delta_{1}=\delta_{2}=0$, $A$ is reducible.
- $\left(x^{2},x_{1}^{2},x_{12}^{2},x_{2}^{2}\right)$:
- There are two non-equivalent systems with this quadruple of determinants: $$\begin{aligned}
&A\left(x,x_{1},x_{2},x_{12};\delta_{1},\delta_{2}\right)=xx_{12}+x_{1}x_{2}+\delta_{1}x_{1}x_{12}+\delta_{2}x_{2}x_{12},\\
&B\left(x,x_{2},x_{3},x_{23};\alpha\right)=xx_{23}+x_{2}x_{3}-\alpha\left(xx_{2}+x_{3}x_{23}\right)-\delta_{2}\left(\alpha^{2}-1\right)x_{2}x_{23},\end{aligned}$$ $$\begin{aligned}
\bar{A}&=A\left(x_{3},x_{13},x_{23},x_{123};\delta_{1},\delta_{2}\right),&
\bar{B}&=B\left(x_{1},x_{12},x_{13},x_{123};\alpha\right),\\
C&=B\left(x,x_{1},x_{3},x_{13};\alpha^{-1}\right),&
\bar{C}&=B\left(x_{2},x_{12},x_{23},x_{123};\alpha^{-1}\right),\\
K&=A\left(x,x_{13},x_{23},x_{12};\delta_{1}\alpha^{-1},\delta_{2}\alpha\right),&
\bar{K}&=A\left(x_{3},x_{1},x_{2},x_{123};\delta_{1}\alpha^{-1},\delta_{2}\alpha\right)\end{aligned}$$
- and $$\begin{aligned}
&A\left(x,x_{1},x_{2},x_{12};\delta_{1}\right)=xx_{2}+x_{1}x_{12}+\delta_{1}x_{1}x_{2}+\delta_{2}x_{2}x_{12},\\
&B\left(x,x_{2},x_{3},x_{23};\delta_{1}\delta_{2}\right)=xx_{23}+x_{2}x_{3}-\alpha\left(xx_{2}+x_{3}x_{23}\right)+\delta_{1}\delta_{2}\left(\alpha^{2}-1\right)x_{2}x_{23},\end{aligned}$$ $$\begin{aligned}
\bar{A}&=A\left(x_{3},x_{13},x_{23},x_{123};\delta_{1}\right),&
\bar{B}&=B\left(x_{123},x_{13},x_{12},x_{1};0\right),\\
C&=B\left(x,x_{1},x_{3},x_{13};-\delta_{1}\right),&
\bar{C}&=B\left(x_{123},x_{23},x_{12},x_{2};-\delta_{1}\right),\\
K&=A\left(x,x_{13},x_{23},x_{12};\delta_{1}\alpha\right),&
\bar{K}&=A\left(x_{3},x_{1},x_{2},x_{123};\delta_{1}\alpha\right).\end{aligned}$$
These systems did not appear in the literature before. Special cases of equations $A=0$ in the last two systems are equivalent modulo $\left({\text{M\"ob}}\right)^{4}$ to the equations whose integrability was shown in [@Hietarinta; @LY] and the “new” integrable equation from [@HV].
Two Equations of Type H4 and Four Equations of Type H6
-------------------------------------------------------
In this section we consider systems (\[system\]) with
- $A$, $\bar{A}$, $C$ and $\bar{C}$ of type [H^6^]{},
- $B$ and $\bar{B}$ of type [H^4^]{} and
- the non-degenerate biquadratics of $B$ and $\bar{B}$ on diagonals.
Below is the list of all 3D consistent systems modulo $\left({\text{M\"ob}}\right)^{8}$ with these properties and with the tetrahedron property.
Every 3D consistent system (\[system\]) satisfying the properties of this section is equivalent modulo $\left({\text{M\"ob}}\right)^{8}$ to one of the following six systems which can be characterized by the quadruples of discriminants of $A$:
- $\left(0,0,0,0\right)$: $$\begin{aligned}
&A\left(x,x_{1},x_{2},x_{12}\right)=x+x_{1}+x_{2}+x_{12},\\
&B\left(x,x_{2},x_{3},x_{23};\alpha\right)=\left(x-x_{23}\right)\left(x_{2}-x_{3}\right)+\alpha,\end{aligned}$$ $$\begin{aligned}
\bar{A}&=A\left(x_{3},x_{13},x_{23},x_{123}\right),&
\bar{B}&=B\left(x_{1},x_{12},x_{13},x_{123};\alpha\right),\\
C&=A\left(x,x_{1},x_{3},x_{13}\right),&
\bar{C}&=A\left(x_{2},x_{12},x_{23},x_{123}\right),\\
K&=B\left(x,x_{12},x_{13},x_{23};-\alpha\right),&
\bar{K}&=B\left(x_{1},x_{2},x_{3},x_{123};-\alpha\right).\end{aligned}$$
- $\left(1,1,0,\delta_{1}\delta_{2}\right)$:
- There are two non-equivalent systems with this quadruple of determinants: $$\begin{aligned}
\begin{split}B\left(x,x_{2},x_{3},x_{23};\delta_{2}\right)=\left(x-x_{23}\right)\left(x_{2}-x_{3}\right)+\delta_{1}\alpha\left(x_{2}+x_{3}\right)+\delta_{2}\alpha\left(x+x_{23}\right)\\
+\delta_{1}\delta_{2}\alpha^{2},\end{split}\\
\begin{split}\bar{B}\left(x_{1},x_{12},x_{13},x_{123};\delta_{1}\right)=\left(x_{1}-x_{123}\right)\left(x_{12}-x_{13}\right)+2\left(\delta_{1}-1\right)\alpha\\
+\left(\delta_{1}\delta_{2}+\delta_{1}-\delta_{2}\right)\alpha\left(x_{1}+x_{123}\right)+2\delta_{1}\delta_{2}\alpha x_{1}x_{123},\end{split}\\
&C\left(x,x_{1},x_{3},x_{13};\alpha\right)=x+x_{13}+\left(\delta_{1}-1\right)x_{3}+\delta_{1}\alpha+x_{1}\left(\delta_{2}x+\delta_{1}x_{3}+\delta_{1}\delta_{2}\alpha\right),\end{aligned}$$ $$\begin{aligned}
A&=C\left(x,x_{1},x_{2},x_{12};0\right),&
\bar{A}&=C\left(x_{23},x_{123},x_{3},x_{13};0\right),\\
\bar{C}&=C\left(x_{23},x_{123},x_{2},x_{12};\alpha\right),&
K&=B\left(x,x_{12},x_{13},x_{23};\delta_{1}+\delta_{2}-\delta_{1}\delta_{2}\right),\\
\bar{K}&=\bar{B}\left(x_{1},x_{2},x_{3},x_{123};0\right)\end{aligned}$$
- and $$\begin{aligned}
&C\left(x,x_{1},x_{3},x_{13};\alpha\right)=x+x_{3}+\left(\delta_{1}-1\right)x_{13}+\delta_{1}\alpha+x_{1}\left(\delta_{2}x+\delta_{1}x_{13}+\delta_{1}\delta_{2}\alpha\right),\\
\begin{split}K\left(x,x_{12},x_{13},x_{23};\delta_{2}\right)=\left(x-x_{23}\right)\left(x_{12}-x_{13}\right)+\delta_{1}\alpha\left(x_{12}+x_{13}\right)\\
+\delta_{2}\alpha\left(x+x_{23}\right)+\delta_{1}\delta_{2}\alpha^{2},\end{split}\\
\begin{split}\bar{K}\left(x_{1},x_{2},x_{3},x_{123};\delta_{1}\right)=\left(x_{1}-x_{123}\right)\left(x_{2}-x_{3}\right)+2\left(\delta_{1}-1\right)\alpha\\
+\left(\delta_{1}\delta_{2}+\delta_{1}-\delta_{2}\right)\alpha\left(x_{1}+x_{123}\right)+2\delta_{1}\delta_{2}\alpha x_{1}x_{123},\end{split}\end{aligned}$$ $$\begin{aligned}
A&=C\left(x,x_{1},x_{2},x_{12};0\right),&
\bar{A}&=C\left(x_{23},x_{123},x_{3},x_{13};0\right),\\
B&=K\left(x,x_{2},x_{3},x_{23};\delta_{1}+\delta_{2}-\delta_{1}\delta_{2}\right),&
\bar{B}&=\bar{K}\left(x_{1},x_{12},x_{13},x_{123};0\right),\\
\bar{C}&=C\left(x_{23},x_{123},x_{2},x_{12};\alpha\right).\end{aligned}$$
If $\delta_{1}=\delta_{2}=0$, $A$ is reducible.
- $\left(x^{2},x_{1}^{2},x_{12}^{2},x_{2}^{2}\right)$:
- There are three non-equivalent systems with this quadruple of determinants: $$\begin{aligned}
&A\left(x,x_{1},x_{2},x_{12};\delta_{1},\delta_{2}\right)=xx_{1}+x_{2}x_{12}+\delta_{1}x_{1}x_{12}+\delta_{2}x_{1}x_{2},\\
&B\left(x,x_{2},x_{3},x_{23};\alpha\right)=xx_{3}+x_{2}x_{23}-\alpha\left(xx_{2}+x_{3}x_{23}\right)-\delta_{2}\left(\alpha^{2}-1\right)x_{2}x_{3},\end{aligned}$$ $$\begin{aligned}
\bar{A}&=A\left(x_{23},x_{123},x_{3},x_{13};\delta_{1},\delta_{2}\right),&
\bar{B}&=B\left(x_{12},x_{1},x_{123},x_{13};\alpha\right),\\
C&=A\left(x,x_{1},x_{3},x_{13};\delta_{1}\alpha^{-1},\delta_{2}\alpha\right),&
\bar{C}&=A\left(x_{23},x_{123},x_{2},x_{12};\delta_{1}\alpha^{-1},\delta_{2}\alpha\right),\\
K&=B\left(x,x_{12},x_{13},x_{23};\alpha^{-1}\right),&
\bar{K}&=B\left(x_{2},x_{1},x_{123},x_{3};\alpha^{-1}\right),\end{aligned}$$
- moreover, $$\begin{aligned}
&A\left(x,x_{1},x_{2},x_{12};\delta_{1}\right)=xx_{2}+x_{1}x_{12}+\delta_{1}x_{2}x_{12}+\delta_{2}x_{1}x_{2},\\
&B\left(x,x_{2},x_{3},x_{23};\delta_{1}\delta_{2}\right)=xx_{3}+x_{2}x_{23}-\alpha\left(xx_{2}+x_{3}x_{23}\right)+\delta_{1}\delta_{2}\left(\alpha^{2}-1\right)x_{2}x_{3},\end{aligned}$$ $$\begin{aligned}
&\bar{A}=A\left(x_{23},x_{123},x_{3},x_{13};\delta_{1}\right),&
&\bar{B}=B\left(x_{1},x_{12},x_{13},x_{123};0\right),\\
&C=A\left(x,x_{1},x_{3},x_{13};\delta_{1}\alpha\right),&
&\bar{C}=A\left(x_{23},x_{123},x_{2},x_{12};\delta_{1}\alpha\right),\\
&K=B\left(x,x_{12},x_{13},x_{23};-\delta_{1}\right),&
&\bar{K}=B\left(x_{1},x_{2},x_{3},x_{123};-\delta_{1}\right)\end{aligned}$$
- and last but not least $$\begin{aligned}
&A\left(x,x_{1},x_{2},x_{12};\delta_{1}\right)=xx_{12}+x_{1}x_{2}+\delta_{1}x_{2}x_{12}+\delta_{2}x_{1}x_{12},\\
&B\left(x,x_{2},x_{3},x_{23};-\delta_{1}\right)=xx_{3}+x_{2}x_{23}-\alpha\left(xx_{2}+x_{3}x_{23}\right)-\delta_{1}\left(\alpha^{2}-1\right)x_{2}x_{3},\end{aligned}$$ $$\begin{aligned}
&\bar{A}=A\left(x_{23},x_{123},x_{3},x_{13};\delta_{1}\right),&
&\bar{B}=B\left(x_{1},x_{12},x_{13},x_{123};-\delta_{1}\right),\\
&C=A\left(x,x_{1},x_{3},x_{13};\delta_{1}\alpha\right),&
&\bar{C}=A\left(x_{23},x_{123},x_{2},x_{12};\delta_{1}\alpha\right),\\
&K=B\left(x,x_{12},x_{13},x_{23};\delta_{1}\delta_{2}\right),&
&\bar{K}=B\left(x_{1},x_{2},x_{3},x_{123};0\right).\end{aligned}$$
Except two special cases (see [@Atk1]) these systems did not appear in the literature before.
Six Equations of Type H6
------------------------
In this section we consider systems (\[system\]) with $A,\bar{A},\ldots,\bar{C}$ of type [H^6^]{}. Below is the list of all those systems possessing the tetrahedron property.
Every 3D consistent system (\[system\]) satisfying the properties of this section is equivalent modulo $\left({\text{M\"ob}}\right)^{8}$ to one of the following five systems which can be characterized by the quadruples of discriminants of $A$:
- $\left(0,0,0,0\right)$: $$K\left(x,x_{12},x_{13},x_{23};\alpha,\beta\right)=\left(\alpha-\alpha\beta+\beta\right)x-\alpha\beta x_{12}-\alpha x_{13}+\beta x_{23},$$ $$\begin{aligned}
&A=K\left(x,x_{1},x_{2},x_{12};-1,1\right),&
&\bar{A}=K\left(x_{13},x_{3},x_{123},x_{23};\alpha\beta-\alpha-\beta,1\right),\\
&B=K\left(x,x_{2},x_{3},x_{23};-\alpha,1\right),&
&\bar{B}=K\left(x_{12},x_{1},x_{123},x_{13};\left(\alpha\beta-\alpha-\beta\right)/\alpha,1\right),\\
&C=K\left(x,x_{1},x_{13},x_{3};1,-\beta\right),&
&\bar{C}=K\left(x_{12},x_{2},x_{23},x_{123};1,\left(\alpha\beta-\alpha-\beta\right)/\beta\right),\\
&\bar{K}=K\left(x_{123},x_{3},x_{2},x_{1};\alpha,\beta\right).\end{aligned}$$
- $\left(0,0,1,1\right)$:
- There are two non-equivalent systems with this quadruple of determinants: $$\begin{aligned}
&A\left(x,x_{1},x_{2},x_{12}\right)=xx_{1}-x_{2}-x_{12},&
&\bar{A}\left(x_{3},x_{13},x_{23},x_{123}\right)=\left(x_{3}+x_{13}\right)x_{23}+x_{123},\end{aligned}$$ $$\begin{aligned}
&B=A\left(x,x_{23},x_{2},x_{3}\right),&
&\bar{B}=\bar{A}\left(x_{12},x_{13},x_{1},x_{123}\right),&
&C=A\left(-x,x_{1},x_{3},x_{13}\right),\\
&\bar{C}=\bar{A}\left(x_{2},x_{12},x_{23},-x_{123}\right),&
&K=A\left(-x,x_{23},x_{12},x_{13}\right),&
&\bar{K}=\bar{A}\left(x_{2},x_{3},x_{1},-x_{123}\right)\end{aligned}$$
- and $$\begin{aligned}
&A\left(x,x_{1},x_{2},x_{12}\right)=xx_{1}-x_{2}-x_{12},\\
&\bar{A}\left(x_{3},x_{13},x_{23},x_{123}\right)=\left(x_{3}+x_{13}\right)x_{123}+x_{3}-x_{23},\end{aligned}$$ $$\begin{aligned}
&B=\bar{A}\left(x_{2},x_{3},-x_{23},\alpha^{-1}x\right),&
&\bar{B}=A\left(\alpha x_{1},x_{123}+1,-x_{12},-x_{13}\right),\\
&C=A\left(x+\alpha,x_{1},-x_{3},-x_{13}\right),&
&\bar{C}=\bar{A}\left(x_{2},x_{12},-x_{23},x_{123}\right),\\
&K=\bar{A}\left(x_{12},x_{13},x_{23},\alpha^{-1}x\right)&
&\bar{K}=A\left(\alpha x_{1},x_{123},x_{2},x_{3}\right).\end{aligned}$$
- $\left(x^{2},x_{1}^{2},x_{12}^{2},x_{2}^{2}\right)$:
- There are two non-equivalent systems with this quadruple of determinants: $$A\left(x,x_{1},x_{2},x_{12};\delta_{1},\delta_{2}\right)=xx_{12}+x_{1}x_{2}+\delta_{1}x_{1}x_{12}+\delta_{2}x_{2}x_{12},$$ $$\begin{aligned}
&\bar{A}=A\left(x_{13},x_{3},x_{123},x_{23};\delta_{1},\delta_{2}\right),&
&B=A\left(x,x_{2},x_{3},x_{23};\delta_{2},0\right),\\
&\bar{B}=A\left(x_{13},x_{1},x_{123},x_{12};\delta_{1}\delta_{3},\delta_{2}\right),&
&C=A\left(x_{13},x_{3},x,x_{1},\delta_{1},-\delta_{3}\right),\\
&\bar{C}=A\left(x_{123},x_{23},x_{2},x_{12},\delta_{1},-\delta_{3}\right),&
&K=A\left(x_{13},x,x_{23},x_{12};-\delta_{3},-\delta_{1}\delta_{2}\right),\\
&\bar{K}=A\left(x_{3},x_{1},x_{123},x_{2};-\delta_{3},0\right)&\end{aligned}$$
- and $$A\left(x,x_{1},x_{2},x_{12};\delta_{1},\delta_{2}\right)=xx_{1}+x_{2}x_{12}+\delta_{1}x_{1}x_{12}+\delta_{2}x_{1}x_{2},$$ $$\begin{aligned}
&\bar{A}=A\left(x_{13},x_{3},x_{123},x_{23};\delta_{1},\delta_{2}\right),&
&B=A\left(x,x_{3},x_{2},x_{23};\delta_{2},0\right),\\
&\bar{B}=A\left(x_{13},x_{1},x_{123},x_{12};\delta_{1}\delta_{3},\delta_{2}\right),&
&C=A\left(x_{12},x_{1},x_{3},x;-\delta_{3},-\delta_{1}\delta_{2}\right),\\
&\bar{C}=A\left(x_{123},x_{12},x_{23},x_{2};-\delta_{3},0\right),&
&K=A\left(x_{13},x_{12},x,x_{23};\delta_{1},-\delta_{3}\right),\\
&\bar{K}=A\left(x_{123},x_{1},x_{2},x_{3};\delta_{1},-\delta_{3}\right).\end{aligned}$$
All systems except for the one characterized by the quadruple $\left(0,0,0,0\right)$ did not appear in the literature before. The latter is equivalent to a special case of the one of the systems from [@Atk2] (in that paper the tetrahedron property is not assumed).
Embedding in the lattice Z[\^]{}3 {#embedding}
=================================
The main conceptual message of [@quadgraphs; @ABS1] is that 3D consistency is synonymous with integrability. In the situations considered there, where equations on opposite faces of the cube are shifted versions of one another, it was demonstrated how to derive Bäcklund transformations and zero curvature representations from a 3D consistent system.
It might be not immediately obvious whether these integrability attributes can still be derived for our systems, where the equations on opposite faces of one elementary cube happen to be completely different. We show now this is the case, indeed.
![Embedding in a planar lattice[]{data-label="fig:flat"}](figure_1)
We start with the composing an integrable system on ${\mathbb{Z}}^{2}$ from a non-symmetric multi-affine polynomial $Q$ from one of our lists. The polynomials will be assigned to the faces as demonstrated in Figure \[fig:flat\]. For an equation $$Q\left(x,x_{1},x_{2},x_{12}\right)=0$$ we define $$\begin{aligned}
&\left|Q\right.:=Q\left(x_{1},x,x_{12},x_{2}\right),&
&\underline{Q}:=Q\left(x_{2},x_{12},x,x_{1}\right)\quad \text{and}&
&\left|\underline{Q}\right.:=Q\left(x_{12},x_{2},x_{1},x\right).\end{aligned}$$ This can be interpreted as reflections at the axis implied by the notation. So, the basic elements of our embedding are not as usual faces but quadruples of faces as marked by the bold lines in Figure \[fig:flat\] and the embedding is not one-periodic as usual but two-periodic in each direction. In the cases considered in [@XP] the equation $\left|\underline{Q}\right.$ is a shifted version of $Q$ as well as $\underline{Q}$ is a shifted version of $\left|Q\right.$.
We show now that this 2D system, with an elementary $2\times 2$ building-block, is integrable: One can find (properly generalized) Bäcklund transformations and zero curvature representations for these systems.
Let us start with Bäcklund transformations. In the symmetric case, i.e. for systems of the ABS-list, we have the picture like in Figure \[fig:cube3\]. A Bäcklund transformation can be interpreted as one layer of the system in the three dimensional lattice. We have a solution $f:V\left({\mathcal{D}}\right)\to{\mathbb{C}}$ on a quad-graph ${\mathcal{D}}$ of $$Q\left(f,f_{1},f_{2},f_{12}\right)=0$$ on the ground floor, its Bäcklund transformation $f^{+}:V\left({\mathcal{D}}^{+}\right)\to{\mathbb{C}}$ on a copy ${\mathcal{D}}^{+}$ of ${\mathcal{D}}$ with $$Q\left(f^{+},f_{1}^{+},f_{2}^{+},f_{12}^{+}\right)=0$$ on the first floor and the Bäcklund parameter assigned to the vertical edges. A more detailed demonstration of this situation can be found for example in [@ddg]. In the non-symmetric case, i.e. for our systems, we have to consider a picture which is a little bit more extensive, as demonstrated in Figure \[fig:cube4\]. In this case a Bäcklund transformation can be seen as two layers of our lattice. We start again with a solution $f:V\left({\mathcal{D}}\right)\to{\mathbb{C}}$ of $$Q\left(f,f_{1},f_{2},f_{12}\right)=0$$ on the ground floor and get a transformation $g:V\left(\bar{{\mathcal{D}}}\right)\to{\mathbb{C}}$ on a copy $\bar{{\mathcal{D}}}$ of ${\mathcal{D}}$ with the equation on the opposite face of the cube $$\bar{Q}\left(g,g_{1},g_{2},g_{12}\right)=0$$ and a parameter $\lambda_{1}$ assigned to the vertical edges. Every parameter of the system we consider which do not appear in $Q$ and $\bar{Q}$ can be chosen as $\lambda_{1}$. Then, starting from $g$ we get a Bäcklund transformation of $f$ called $f^{+}:V\left({\mathcal{D}}\right)\to{\mathbb{C}}$ on the second floor with $$Q\left(f^{+},f_{1}^{+},f_{2}^{+},f_{12}^{+}\right)=0$$ and a parameter $\lambda_{2}$ assigned to the vertical edges.
Zero curvature representations can be derived for non-symmetric systems, too. We will first consider briefly the idea how to derive zero curvature representations of the symmetric systems. In the symmetric case we have again the picture like in Figure \[fig:cube5\]. In this case a transition matrix of a zero curvature representation of an equation on the ground floor can be interpreted as a Möbius transformation (in the standard matrix notation) from one vertex of the first floor to another one connected by an edge e.g., $$f_{1}^{+}=L\left(f,f_{1},\lambda\right)\left[f^{+}\right]$$ with $L$ a $2\times 2$ transition matrix dependent on the spectral parameter $\lambda$. One can derive the Möbius transformation from the equation of the corresponding face. For informations in more details we again refer to [@ddg]. In the non-symmetric case we also need just one layer to derive a zero curvature equation (see Figure \[fig:cube6\]). Also in this case a transition matrix of a zero curvature representation of an equation on the ground floor can be interpreted as a Möbius transformation from one vertex of the first floor to another one conneted by an edge in the standard matrix notation, e.g., $$g_{1}=L\left(f,f_{1},\lambda_{1}\right)\left[g\right]$$ with $L$ a $2\times 2$ transition matrix dependent on the spectral parameter $\lambda_{1}$.
Concluding Remarks
==================
Due to Section \[embedding\] the Theorems \[Hvierfirst\], \[Hviersecond\] and \[Hsechs\] gives us a Bäcklund transformation and a zero curvature representation for every quad-equation of type [H^4^]{} and of type [H^6^]{} in every arrangement of fields to the vertices of an elementary square of ${\mathbb{Z}}^{2}$. Therefore, they are all integrable.
Moreover, our classification includes all known 3D consistent systems except two systems mentioned in [@Atk1], the system mentioned in [@Hietarinta] and two systems only containing linear equations (see [@Atk2; @ABS2]). Of course, all these system do not possess the tetrahedron property.
Acknowledgments {#acknowledgments .unnumbered}
===============
The author is supported by the Berlin Mathematical School and is indebted to Yuri B. Suris for his continued guidance.
[^1]: Institut für Mathematik, MA 7-2, Technische Universität Berlin, Str. des 17. Juni 136, 10623 Berlin, Germany; e-mail: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In this paper we derive a refined asymptotic expansion, near an isolated singularity, for conformally flat metrics with constant positive $Q$-curvature and positive scalar curvature. The condition that the metric has constant $Q$-curvature forces the conformal factor to satisfy a fourth order nonlinear partial differential equation with critical Sobolev growth, whose leading term is the bilaplacian. We model our results on a similar asymptotic expansion for conformally flat, constant scalar curvature metrics proven by Korevaar, Mazzeo, Pacard, and Schoen. Along the way we analyze the linearization of the $Q$-curvature equation about the Delaunay metrics recently discovered by Frank and König, which may be of independent interest.'
author:
- 'Jesse Ratzkin [^1]'
title: 'On constant $Q$-curvature metrics with isolated singularities'
---
Introducion
===========
Let $(M,g)$ be a Riemannian manifold of dimension $n \geq 5$. In this paper we investigate the behavior of a conformal metric $\widetilde
g = u^{\frac{4}{n-4}} g$ near an isolated singularity, subject to curvature conditions, namely constant and positive $Q$-curvature and positive scalar curvature.
To begin we fix some notation. Let $R_g$ and ${\operatorname{Ric}}_g$ denote the scalar and Ricci curvature of $g$, and let $\Delta_g$ denote the Laplace-Beltrami operator. The (fourth-order) $Q$-curvature of $g$ is $$\label{q_defn1}
Q_g = - \frac{1}{2(n-1)} \Delta_g R_g + \frac{n^3-4n^2+16n-16}
{8(n-1)^2(n-2)^2} R_g^2 - \frac{2}{(n-2)^2} |{\operatorname{Ric}}_g|^2.$$ We can simplify this expression using the Schouten tensor $$\label{schouten_defn}
A_g = \frac{1}{n-2} \left ( {\operatorname{Ric}}_g - \frac{R_g g}{2(n-1)} \right ),
\qquad J_g = \operatorname{tr}_g (A_g) = \frac{R_g}{2(n-1)},$$ so that becomes $$\label{q_defn2}
Q_g = -\Delta_g J_g - 2 |A_g |^2 + \frac{n}{2} J_g^2 .$$ Associated to $Q_g$ we find the fourth order differential operator $$\label{paneitz_op_defn}
P_g (u) = (-\Delta_g)^2 u + \operatorname{div}
\left ( 4 A_g (\nabla u, \cdot) - (n-2) J_g \nabla u \right ) +
\frac{n-4}{2} Q_g,$$ which enjoys the transformation rule $$\label{conf_trans_rule1}
\widetilde g = u^{\frac{4}{n-4}} g \Rightarrow
P_{\widetilde g} (v) = u^{-\frac{n+4}{n-4}} P_g(uv).$$ Substituting $v=1$ into we find $$\label{conf_trans_rule2}
\widetilde g = u^{\frac{4}{n-4}} \Rightarrow
Q_{\widetilde g} = \frac{2}{n-4} u^{-\frac{n+4}{n-4}}
P_g(u).$$
S. Paneitz [@Pan1] first introduced the operator and studied its transformation properties. Later T. Branson [@Bran1; @Bran2] extended this operator to differential forms and studied a related sixth-order operator, as well as the $Q$-curvature. The reader can find summaries of the current understanding of $Q$-curvature in the survey articles [@BG], [@CEOY], and [@HY].
An aside on scalar curvature and the Yamabe problem
---------------------------------------------------
A this point we pause to discuss the Yamabe problem, which provides a guide for much of the investigation of the properties of $Q$-curvature. One can define the conformal Laplacian $$\label{conf_lap}
\mathcal{L}_g (u) = -\Delta_g (u) + \frac{4(n-1)}{n-2} R_g u,$$ which enjoys the transformation rule $$\label{conf_trans_rule3}
\widetilde g = u^{\frac{4}{n-2}} g \Rightarrow
\mathcal{L}_{\widetilde g} (v) = u^{-\frac{n+2}{n-2}}
\mathcal{L}_g(uv).$$ Substituting $v=1$ into we obtain $$\label{conf_trans_rule4}
\widetilde g = u^{\frac{4}{n-2}} g \Rightarrow
R_{\widetilde g} = \frac{n-2}{4(n-1)} u^{-\frac{n+2}{n-2}}
\mathcal{L}_g(u).$$
This last transformation rule allows us to define the conformal invariant $$\begin{aligned}
\label{yamabe_inv1}
\mathcal{Y} ([g]) & = & \inf \left \{ \frac{ \int_M
R_{\widetilde g} d\mu_{\widetilde g}}
{(\operatorname{Vol}_{\widetilde g}(M))^{\frac{n-4}{n}}} : \widetilde g =
u^{\frac{4}{n-2}} g \in [g] \right \} \\ \nonumber
& =& \inf \left \{ \frac{n-2}{4(n-1)}\frac{ \int_M u \mathcal{L}_g(u)
d\mu_g} {\left ( \int_M u^{\frac{2n}{n-2}} d\mu_g \right )^{\frac{n-2}{n}}} :
u \in \mathcal{C}^\infty (M), u>0 \right \}\end{aligned}$$ A straightforward computation shows that critical points of the functional $$\widetilde g\in [g] \mapsto \frac{\int_M R_{\widetilde g}
d\mu_{\widetilde g}}{(\operatorname{Vol}_{\widetilde g}
(M))^{\frac{n-4}{n}}}$$ are precisely the constant scalar curvature metrics in the conformal class $[g]$.
In [@Y] Yamabe proposed finding a constant scalar curvature metric in a given conformal class $[g]$ through the family of related variational problems $$\label{yamabe_inv_p}
\mathcal{S}_p ([g]) =
\inf \left \{ \frac{n-2}{4(n-1)} \frac{\int_M u \mathcal{L}_g(u)
d\mu_g}{\left ( \int_M u^p d\mu_g \right )^{2/p}} :
u \in \mathcal{C}^\infty(M), u>0 \right \}$$ for $p$ such that $1 < p < \frac{2n}{n-2}$. By Rellich’s compactness theorem the infimum $\mathcal{S}_p ([g])$ is a minimum, and realized by a smooth metric. Yamabe’s strategy was to find a constant scalar curvature metric as a limit of minimizers for $\mathcal{S}_p([g])$ as $p \nearrow \frac{2n}{n-2}$.
Trudinger [@Tru] first pointed out the difficulties in extracting this limit, primarily that Rellich’s compactness fails exactly for $p = \frac{2n}{n-2}$. Later Aubin [@Aub] resolved these difficulties in many cases, and Schoen [@Sch] completed Yamabe’s original program. Schoen’s resolution of the Yamabe problem culminated the work of many people over 25 years, and continues to inspire new research.
One can see the lack of compactness directly in the case of the round metric $g_0$ on the sphere ${\mathbf{S}}^n$. Using stereographic projection we can write the round metric as $$\label{scal_curv_sph1}
g_0 = \frac{4}{(1+|x|^2)^2} \delta = \left ( \left ( \frac{1+|x|^2}{2}
\right )^{\frac{2-n}{2}}\right )^{\frac{4}{n-2}} \delta =
u_1^{\frac{4}{n-2}} \delta$$ where $\delta$ is the flat metric. Dilating the Euclidean coordinates by a factor of $\lambda>0$ one finds the conformal factor $$\label{scal_curv_sph2}
u_\lambda =\left ( \frac{1+\lambda^2 |x|^2}{2\lambda}
\right )^{\frac{2-n}{2}}.$$ However, as $\lambda \rightarrow \infty $ we see that $u_\lambda
\rightarrow 0$ outside of any fixed neightborhood of $0$, while $u_\lambda(0) \rightarrow \infty$. Geometrically, this blow-up behavior concentrates the entirety of the sphere in a small neighborhood of the south pole, which corresponds to the origin in Euclidean coordinates, shrinking the complement of this neighborhood to be vanishingly small.
The blow-up described above motivates one to understand the asymptotics of constant scalar curvature metrics with isolated singularities. In the special case that the background metric $g$ is locally conformally flat one can (locally) write a constant scalar curvature metric in the conformal class $[g]$ as $u^{\frac{4}{n-2}} \delta$. A theorem of Schoen and Yau [@SY] implies the value of the scalar curvature must be a positive constant, which we normalize to be $n(n-1)$. In this case case becomes $$\label{scal_curv_pde}
-\Delta u = \frac{n(n-2)}{4} u^{\frac{n+2}{n-2}} .$$ A computation shows that $u_\lambda$, defined in solves , and indicates that one should study the possible asymptotic behavior of positive solutions of . In [@CGS] Caffarelli, Gidas and Spruck proved that any positive solution of in the punctured ball ${\mathbf{B}}_1(0) \backslash \{ 0 \}$ satisfies $$u(x) = \bar{u}(|x|) (1+ \mathcal{O}(|x|)), \qquad
\bar{u} (r) = \frac{1}{r^{n-1} |{\mathbf{S}}^{n-1}|} \int_{|x| = r}
u(x) d\sigma(x).$$ Later Korevaar, Mazzeo, Pacard and Schoen [@KMPS] refined this asymptotic expansion, obtaining the next term in the expansion and giving the terms in this expansion a more geometric interpretation.
Motivation for studying constant $Q$-curvature metrics with isolated singularities
----------------------------------------------------------------------------------
Returning to our discussion of $Q$-curvature, we can follow the well-established model of scalar curvature and define the conformal invariant $$\begin{aligned}
\label{paneitz_inv1}
\mathcal{Y}_4^+ ([g])& = & \inf\left \{ \frac{\int_M
Q_{\widetilde g} d\mu_g}{(\operatorname{Vol}_{\widetilde
g} (M))^{\frac{n-4}{n}} } : \widetilde g = u^{\frac{4}{n-4}} g \in
[g] \right \} \\ \nonumber
& = & \inf \left \{ \frac{2}{n-4} \frac{\int_M u P_g (u)
d\mu_g}{ \left ( \int_M u^{\frac{2n}{n-4}} d\mu_g
\right )^{\frac{n-4}{n}} } : u \in \mathcal{C}^\infty
(M) , u> 0\right \}. \end{aligned}$$ Once again, a straight-forward computation implies critical points of the functional $$u \mapsto \frac{\int_M u P_g (u) d\mu_g}{\left ( \int_M
u^{\frac{2n}{n-4}} d\mu_g \right )^{\frac{n-4}{n}}}$$ give exactly the constant $Q$-curvature metrics within the conformal class $[g]$. In the case that this constant is positive, we normalize it to be $\frac{n(n^2-4)}{8}$, which is the value attained by the round metric on the sphere.
In constrast to the situation with scalar curvature, the existence and properties of minimizers and higher order critical points is poorly understood. However, the conformal class of the round metric on the sphere still provides an illustrative example. We once again write the round metric as a conformally flat metric using stereographic projection, so that $$\label{sph_soln1}
g_0 = \frac{4}{(1+|x|^2)^2} \delta = \left ( \left ( \frac{1+|x|^2}
{2} \right )^{\frac{4-n}{2}} \right )^{\frac{4}{n-4}} \delta =
U_1(x)^{\frac{4}{n-4}}\delta.$$ Once again we may apply a conformal dilation, which gives the solution $$\label{sph_soln2}
U_\lambda = \left ( \frac{1+\lambda^2 |x|^2}{2\lambda}
\right )^{\frac{4-n}{2}}$$ for any $\lambda >0$.
We can seek metrics $g = u^{\frac{4}{n-4}} g_0$ in the conformal class of the round metric $g_0$ with constant $Q$-curvature $\frac{n(n^2-4)}{8}$. In steregraphic coordinates reduces to $$\label{paneitz_pde1}
\Delta^2 u = \frac{n(n-4)(n^2-4)}{16} u^{\frac{n+4}{n-4}}.$$ By the above discussion $U_\lambda$ solves for each $\lambda > 0$. In fact, a theorem of C. S. Lin [@Lin] states that any positive solution of must have the form $U_\lambda (x-x_0)$ for some $\lambda > 0$ and $x_0 \in {\mathbf{R}}^n$.
As before, we examine the behavior of $U_{\lambda, x_0} =
U_\lambda (\cdot + x_0)$ as $\lambda \rightarrow \infty$, finding that $U_{\lambda, x_0} \rightarrow 0$ outside any fixed neighborhood of $x_0$, while it blows up at $x_0$. This behavior illustrates the natural lack of compactness in the infimum , and explains why we should study solutions of with isolated singularities.
Main results
------------
We concentrate on solututions of in the unit ball ${\mathbf{B}}_1(0)$ with an isolated singularity at the origin.
Recently Jin and Xiong [@JX] proved that if $u \in \mathcal{C}^\infty
({\mathbf{B}}_1 (0) \backslash \{ 0 \})$ is a positive solution of which also satisfies $-\Delta u >0$ then $$\label{JX_asymp}
u(x) = \bar u(|x|) (1+ \mathcal{O} (x)), \qquad \bar u (r) =
\frac{1}{r^{n-1} |{\mathbf{S}}^{n-1}|} \int_{|x| = r} u(x) d\sigma (x).$$
To understand our results we introduce some special solutions of . We change to cylindrical coordinates, letting $t = -\log |x|$, $\theta =
x/|x|$, and $$\label{cyl_coord_change}
v: (0,\infty) \times {\mathbf{S}}^{n-1} \rightarrow (0,\infty), \qquad
v(t,\theta) = e^{\left ( \frac{4-n}{2}\right )t} u(e^{-t} \theta) .$$ Under this change of coordinates becomes $$\begin{aligned}
\label{paneitz_pde2}
\frac{n(n-4)(n^2-4)}{16} v^{\frac{n+4}{n-4}} & = & \frac{\partial^4 v}{\partial t^4}
- \left ( \frac{n(n-4)+8}{2} \right )\frac{\partial^2 v}{\partial t^2} +
\frac{n^2 (n-4)^2}{16} v \\ \nonumber
&& + \Delta_\theta^2 v + 2 \Delta_\theta \frac{\partial^2 v}{\partial t^2}
- \frac{n(n-4)}{2}\Delta_\theta v, \end{aligned}$$ where $\Delta_\theta$ is the Laplace-Beltrami operator on the round sphere ${\mathbf{S}}^{n-1}$, and the condition $-\Delta u >0$ becomes $$\label{pos_lap_cyl_coords}
-\ddot v + 2\dot v + \frac{n(n-4)}{4} v - \Delta_\theta v > 0.$$
In [@Lin], Lin also proved that any global solution of must be a function of $t$ alone, and so it must satisfy the ordinary differential equation (ODE) $$\label{paneitz_ode1}
\frac{n(n-4)(n^2-4)}{16} v^{\frac{n+4}{n-4}} = \ddddot v - \left (
\frac{n(n-4)+8}{2} \right ) \ddot v + \frac{n^2(n-4)^2}{16} v .$$ One immediately sees two solutions of : the constant solution $$\label{cyl_soln}
v_{cyl} = \left ( \frac{n(n-4)}{n^2-4} \right )^{\frac{n-4}{8}}$$ and the spherical solution $$u(x) = U_1(x) = \left ( \frac{1+|x|^2}{2} \right )^{\frac{4-n}{2}}
\Leftrightarrow v_{sph} = (\cosh t)^{\frac{4-n}{2}} .$$
Frank and König [@FK] classified all global, positive solutions of . First they demonstrate the existence of a unique periodic solution $v_\epsilon$ of attaining its minimum value of $\epsilon$ at $t=0$ for each $\epsilon \in
(0,v_{cyl}]$. We call $v_\epsilon$ the [**Delaunay solution**]{} with necksize $\epsilon$. In the same paper they prove that each global, positive solution of must either have the form $(\cosh (\cdot + T))^{\frac{4-n}{2}}$ or have the form $v_\epsilon
(\cdot + T)$ for some $\epsilon \in (0, v_{cyl}]$ and $T \in {\mathbf{R}}$.
We are now ready to state our main theorem.
\[refined\_asymp\_thm\] Let $v\in \mathcal{C}^\infty((0,\infty) \times {\mathbf{S}}^{n-1})$ be a positive solution of which also satisfies . Then either $\limsup_{t \rightarrow \infty} v(t,\theta) = 0$ or there exists parameters $\epsilon \in (0,v_{cyl}]$, $T \in {\mathbf{R}}$ and $a \in {\mathbf{R}}^n$, and positive constants $C$ and $\beta > 1$ such that $$\left | v(t,\theta) - v_\epsilon (t+T)- e^{-t} \langle \theta, a \rangle
\left ( -\dot v_\epsilon (t+T) + \frac{n-4}{2} v_\epsilon (t+T) \right )
\right | \leq C e^{-\beta t} .$$
We will see below in and that one obtains $$v_\epsilon + e^{-t} \langle \theta, a \rangle \left ( -\dot v_\epsilon
+ \frac{n-4}{2} v_\epsilon \right )$$ by applying a certain translation to the Delaunay solution $v_\epsilon$.
\[further\_asymp\_rmk\] We speculate that one can adapt our proof below to prove a further refinement of our estimate in Theorem \[refined\_asymp\_thm\], following the estimate of Han, Li and Li [@HLL]. In we define the increasing set of indicial roots $$\Gamma_\epsilon = \{ \dots, -\gamma_{\epsilon,2},
-\gamma_{\epsilon,1} = -1, 0, \gamma_{\epsilon,1} = 1,
\gamma_{\epsilon,2}, \dots \}, \qquad \gamma_{\epsilon,j}
\rightarrow \infty.$$ One should be able to prove an estimate of the form $$\left | v(t,\theta) - v_\epsilon (t+T) - \sum_{i=1}^m \sum_{j=0}^{m-1}
c_{ij} (t,\theta) t^j e^{-\gamma_{\epsilon, j} t} \right | \leq
C t^m e^{-\gamma_{\epsilon, m+1}}$$ where the coefficient functions $c_{ij} (t,\theta)$ are bounded.
We recast Theorem \[refined\_asymp\_thm\] in geometric terms.
Let $n \geq 5$ and let $g = u^{\frac{4}{n-4}} \delta$ be a conformally flat metric on ${\mathbf{B}}_1(0) \backslash \{ 0 \}$ with positive scalar curvature and $Q_g = \frac{n(n^2-4)}{8}$. Then either $g$ extends to a smooth metric on ${\mathbf{B}}_1(0)$ or there exist parameters $\epsilon \in (0,v_{cyl}]$, $T \in {\mathbf{R}}$ and $a \in {\mathbf{R}}$ and $\beta >1$ such that $$u(x) = |x|^{\frac{4-n}{2}} \left ( v_\epsilon (-\log |x| + T)
+ \langle x,a \rangle \left ( -\dot v_\epsilon (-\log |x|+T) + \frac{n-4}{2}
v_\epsilon (-\log |x| + T) \right ) + \mathcal{O} (|x|^\beta) \right )$$
We have already seen that $Q_g = \frac{n(n^2-4)}{8}$ is equivalent to . By $$R_g = \frac{n-2}{4(n-1)}u^{-\left ( \frac{n+2}{n-4} \right )}
\left ( -\Delta u^{\frac{n-2}{n-4}} \right ) ,$$ so that $$\label{pos_scal_curv}
R_g > 0 \Leftrightarrow -\Delta u^{\frac{n-2}{n-4}} > 0 \Leftrightarrow
-\Delta u > \frac{2}{n-4} \frac{|\nabla u|^2}{u}.$$ In particular, $R_g >0$ implies $-\Delta u >0$. Changing to cylindrical coordinates we obtain a function $$v \in \mathcal{C}^\infty ((0,\infty) \times {\mathbf{S}}^{n-1}), \qquad
v(t,\theta) = e^{\left ( \frac{4-n}{2} \right ) t} u(e^{-t} \theta)$$ which satisfies and .
We constrast our methods with that of Jin and Xiong [@JX]. They use Green’s identity to transform into an integral formula $$\label{paneitz_int_form1}
u(x) = \int_{{\mathbf{B}}_1(0)} \frac{(u(y))^{\frac{n+4}{n-4}}}{|x-y|^{n-4}}
d\mu(y) + h(x) , \qquad u \in L^{\frac{n+4}{n-2}} ({\mathbf{B}}_1(0))
\cap \mathcal{C}^1({\mathbf{B}}_1(0) \backslash \{ 0 \})$$ where $h \in \mathcal{C}^1({\mathbf{B}}_1(0))$. In this setting the condition $-\Delta u > 0$ implies $h>0$, which allows Jin and Xiong to apply the method of moving spheres/planes. They first prove [*a priori*]{} upper and lower bounds for $u$, and then prove their asymptoic estimate using moving spheres.
Our proof below follows the techniques in [@KMPS], extracting a limit from a slide-back sequence. We start with $v:(0,\infty) \times {\mathbf{S}}^{n-1}
\rightarrow (0,\infty)$ which satisfies and a sequence $\tau_k \rightarrow \infty$. and let $v_k (t,\theta) =
v(t+\tau_k, \theta)$. The [*a priori*]{} estimates of Jin and Xiong allow us to extract a convergent subsequence, which is defined on all of ${\mathbf{R}}\times {\mathbf{S}}^{n-1}$. By the uniqueness theorem of Frank and König this limit must have the form $v_\epsilon (t + T)$ for some $\epsilon \in (0, v_{cyl}]$ and $T \in {\mathbf{R}}$. The main task in proving simple asymptotics involves proving $\epsilon$ and $T$ do not depend on the choice of the sequence $\tau_k
\rightarrow \infty$ or the choice of the convergent subsequence. Our proof that the parameters $\epsilon$ and $T$ are independent of all choices relies heavily on a careful analysis of the linearization of the PDE about a Delaunay solution, including an asymptotic expansion of solutions of the linearized equation. Once we prove the simple asymptotics, we obtain the refined asymptotics from the asymptotic expansion of the linearized operator.
The two techniques are complementary. The proof of Jin and Xiong is more general, and applies to any solution of the integral equation $$u(x) = \int_{{\mathbf{B}}_1(0)} \frac{(u(y))^{\frac{n+2\sigma}{n-2\sigma}}}
{|x-y|^{n-2\sigma}} d\mu(y) + h(x), \qquad h>0, \quad 0 < \sigma
< \frac{n}{2},$$ whereas at this time one can apply our technique only in the cases $\sigma=1$ and $\sigma =2$. On the other hand, their technique does not give the refined asymptotic expansion.
Previously González [@Gonz] proved a similar asymptotics theorem for $\sigma_k$-curvature, which is a fully-nonlinear genearlization of scalar curvature whose associated PDE is second order and fully nonlinear. More recently, Caffarelli, Jin, Sire, and Xiong [@CJSX] prove an asymptotic result for positive solutions of the nonlocal equation $$(-\Delta)^s u = u^{\frac{n+2s}{n-2s}}$$ for any $s \in (0,1)$. Also Baraket and Rebhi [@BR] and Y.-J. Lin [@YJLin] construct many examples of constant $Q$-curvature metrics.
The rest of this paper proceeds as follows. Section \[prelim\_sec\] we collect some preliminary computations, most of which exist already in the literature. Within this section we include a detailed description of the Delaunay solutions. Our analysis begins in earnest in Section \[lin\_anal\_sec\] where we study the linearization of the PDE . Most importantly we prove the linear stability of the Delaunay solutions using the Fourier-Laplace transform as developed by Mazzeo, Pollack and Uhlenbeck [@MPU]. In this section we also define the indicial roots mentioned above in Remark \[further\_asymp\_rmk\], which give the exponential growth rates of the solutions of the linearization of when linearized about a Delaunay solution. In Section \[asymp\_sec\] we present an alternative proof of the simple asymptotics Jin and Xiong prove in [@JX], and in Section \[refined\_sec\] we derive the refined asymptotics.
[Acknowledgements:]{} This research was partially supported by the National Research Foundation of South Africa. I would like to thank Rupert Frank, Andrea Malchiodi, and Rafe Mazzeo for helpful conversations during the conference “Recent Advances in Nonlocal and Nonlinear Analysis" at ETH Zürich in June 2014, as well as the organizers of this conference for providing a fruitful venue.
Preliminaries {#prelim_sec}
=============
We collect some preliminary computations related to solutions of and constant $Q$-curvature metrics.
Symmetries of the PDE
---------------------
In this section we recall some of the symmetries of , which essentially arise from the transformation rule . One can of course translate solutions to obtain a new solution, but two more interesting symmetries reflect the scale invariance and conformal invariance outlined above.
We first discuss scale invariance. Let $u>0$ solve and let $\lambda>0$. We seek $a>0$ such that $u_\lambda(x) =
\lambda^a u(\lambda x)$ is also a solution. Evaluating, we find $$(-\Delta)^2 u_\lambda = \lambda^{a+4} (-\Delta)^2 u (\lambda x), \qquad
\frac{n(n-4)(n^2-4)}{16} u_\lambda^{\frac{n+4}{n-4}} = \lambda^{\frac{(n+4)a}{n-4}}
\frac{n(n-4)(n^2-4)}{16} (u(\lambda x))^{\frac{n+4}{n-4}},$$ which coincide precisely when $a = \frac{n-4}{2}$. In other words we see $$\label{scaling_law1}
u \textrm{ solves \eqref{paneitz_pde1} } \Rightarrow
u_\lambda(x) = \lambda^{\frac{n-4}{2}} u(\lambda x)
\textrm{ solves \eqref{paneitz_pde1} for all } \lambda >0.$$
This symmetry is much simpler in the cylindrical coordinates. Letting $T = -\log \lambda$ we see $$v_\lambda (t,\theta) = e^{\frac{(4-n)t}{2}} u_\lambda (e^{-t} \theta) =
\lambda^{\frac{n-4}{2}} e^{\frac{(4-n)t}{2}} u(\lambda e^{-t} \theta)
= e^{\frac{(4-n)(t+T)}{2}} u(e^{-(t+T)} \theta),$$ so that $$v \textrm{ solves \eqref{paneitz_pde2} } \Rightarrow
v_T(t,\theta) = v(t+T,\theta)
\textrm{ solves \eqref{paneitz_pde2} for all } T\in {\mathbf{R}},$$ which is readily apparent directly from the PDE.
The second symmetry reflects the invariance under reflections through spheres, and one can write write it explicitly by defining the Kelvin transforms $$\label {kelvin_trans}
\mathbb{K}_{x_0} (u) (x) = |x|^{2-n} u \left ( \frac{x}{|x|^2} + x_0
\right ), \qquad \widehat{\mathbb{K}}_{x_0} (u)(x) = |x|^{4-n} u \left
( \frac{x}{|x|^2} + x_0 \right ) .$$ Observe that the functions $\mathbb{K}_{x_0}(u)$ and $\widehat {\mathbb{K}}_{x_0}
(u)$ are now defined on different domains. For instance, if $u \in
\mathcal{C}^\infty ({\mathbf{B}}_1(0) \backslash \{ 0 \} )$ then $\mathbb{K}_{x_0} (u), \widehat {\mathbb{K}}_{x_0} \in
\mathcal{C}^\infty ({\mathbf{R}}^n \backslash {\mathbf{B}}_1 (x_0))$.
One can find the usual Kelvin transformation law $$\label{orig_kelvin_law}
\Delta (\mathbb{K}_{x_0}(u))(x) = |x|^{-4} \mathbb{K}_0(\Delta u)(x)$$ in many textbooks, and the transformation law $$\label{4th_order_kelvin}
\Delta^2 (\widehat{\mathbb{K}}_{x_0}(u))(x) = |x|^{-8} \widehat{\mathbb{K}}_{x_0}
(\Delta^2 u)(x),$$ appears (for instance) in Lemma 3.6 of [@Xu1].
In the case $x_0 =0$ this transformation looks particularly simple in cylindrical coordinates, namely $v(t,\theta)$ gets transformed to $\widehat{\mathbb{K}}_0(v)(t,\theta)
= v(-t,\theta)$. On the other hand, the transformation is much more complicated in cylindrical coordinates when the center is not $0$.
Delaunay solutions
------------------
We have already introduced the Delaunay solutions, and in this section we give a more detailed description.
The family of Delaunay solutions account for all of the positive solutions of on the whole cylinder ${\mathbf{R}}\times {\mathbf{S}}^{n-1}$. Recall the cylindrical (constant) and spherical solutions $$v_{cyl} = \left ( \frac{n(n-4)}{n^2-4} \right )^{\frac{n-4}{8}},
\qquad v_{sph} = (\cosh t)^{\frac{4-n}{2}}.$$ It is convenient to observe that, because $n > 4$ we have $$n^2-4 -n(n-4) = 4n-4 >0 \Rightarrow 0<v_{cyl} < 1,$$ as well as $$v_{sph} (0) = 1, \qquad \dot v_{sph} (t) > 0 \textrm{ for }
t<0, \qquad \dot v_{sph}(t) < 0 \textrm{ for }t>0, \qquad
\lim_{t \rightarrow \pm \infty} v_{sph} (t) = 0.$$ Frank and König proved that for each $\epsilon \in (0,v_{cyl}]$ there exists a unique positive solution $v_\epsilon$ of attaining its minimum value of $\epsilon$ at $t=0$. Furthermore they prove that any positive, global solution of must be either $v_{sph} (\cdot +T)$ for some $T \in {\mathbf{R}}$ or $v_\epsilon (\cdot +T)$ for some $T \in {\mathbf{R}}$ and $\epsilon \in (0,v_{cyl}]$.
Each $v_\epsilon$ is periodic with period $T_\epsilon$, has local minima at exactly $k T_\epsilon$ for each $k \in \mathbf{Z}$, and local maxima at exactly $\left ( \frac{2k+1}{2} \right ) T_\epsilon$ for each $T_\epsilon$, and no other critical points. For our purposes we take the period of the cylindrical solution to be $T_{cyl}$, given in . Please see the text surrounding for our reasoning. One can also show each $v_\epsilon$ is symmetric about each of its critical point and that $T_\epsilon$ is a decreasing function of $\epsilon$ with $\lim_{\epsilon \searrow 0} T_\epsilon = \infty$. In this context one should think of $v_{sph}$ as the limit of $v_\epsilon (\cdot + T_\epsilon /2)$ as $\epsilon \searrow 0$.
It appears now that the Delaunay metrics occur in a two-parameter family, with the necksize $\epsilon$ and translation parameter $T$ as parameters. However, it is useful to enlarge this family to include a $2n$-dimension family of ambient translations, which we first describe in Euclidean coordinates and then transform to cylindrical coordinates. In preparation we transform the Delaunay solution $v_\epsilon$ to Euclidean coordinates, obtaining $$u_\epsilon(x) = |x|^{\frac{4-n}{2}} v_\epsilon (-\log |x|).$$ The first translation is $$\begin{aligned}
\bar u_{\epsilon,a}(x) & = & u_\epsilon(x-a) = |x -a|^{\frac{4-n}{2}}
v_\epsilon( -\log |x-a|) \\
& = & |x|^{\frac{4-n}{2}} \left | \frac{x}{|x|}
- \frac{a}{|x|} \right |^{\frac{4-n}{2}} v_\epsilon \left (
-\log|x| -\log\left | \frac{x}{|x|} -\frac{a}{|x|} \right | \right ),\end{aligned}$$ where $a \in {\mathbf{R}}^n$. Tranforming this back to cylindrical coordinates we then obtain $$\label{trans_del_soln1}
\bar v_{\epsilon,a} = |\theta - e^t a|^{\frac{4-n}{2}}
v_\epsilon(t-\log|\theta - e^t a|).$$ The function $\bar v_{\epsilon,a}$ is of course not a smooth global solution, and it has a singular point when $$\theta = e^t a \Leftrightarrow t= -\log |a|, \quad \theta
= \frac{a}{|a|}.$$
We obtain the remaining translations using the Kelvin transform defined in . Given $a\in{\mathbf{R}}^n$ we define $$\begin{aligned}
u_{\epsilon, a} (x) & = & \widehat{\mathbb{K}}_0 (\widehat{\mathbb{K}}_0
(u_\epsilon (\cdot - a)) (x) = \widehat{\mathbb{K}}_0
\left ( |\cdot - a|^{4-n} u_\epsilon \left (
\frac{\cdot - a}{|\cdot - a|^2} \right ) \right ) (x) \\
& =& |x|^{4-n} \left | \frac{x}{|x|^2} - a \right |^{4-n}u_\epsilon
\left ( \frac{\frac{x}{|x|^2} - a}{\left | \frac{x}{|x|^2} - a \right |^2}
\right ) = \left |\frac{x}{|x|} - |x|a\right |^{4-n} u_\epsilon \left (
|x|^{-1} \left |\frac{x}{|x|} - |x| a \right |^{-1} \right ) \\
& = & \left | \frac{x}{|x|} - |x| a \right |^{\frac{4-n}{2}} |x|^{\frac{n-4}{2}}
\left | \frac{x}{|x|} - |x| a \right |^{\frac{n-4}{2}} v_\epsilon \left (
-\log \left ( |x|^{-1} \left | \frac{x}{|x|} - |x| a \right |^{-1} \right )
\right ) \\
& = & |x|^{\frac{n-4}{2}} \left | \frac{x}{|x|} - |x| a\right |^{\frac{4-n}{2}}
v_\epsilon \left ( -\log |x| - \log \left | \frac{x}{|x|} - |x| a \right | \right ) ,\end{aligned}$$ which in turn gives us $$\label{trans_del_soln2}
v_{\epsilon,a}(t,\theta) = |\theta - e^{-t}a|^{\frac{4-n}{2}}
v_\epsilon(t+\log|\theta - e^{-t}a|).$$ This function has a singular point when $t=\log|a|$ and $\theta = a/|a|$.
Using the Taylor expansions $$\left | \frac{x}{|x|} - |x| a \right |^{\frac{4-n}{2}} =
1 + \frac{(n-4)}{2} \langle x,a \rangle + \mathcal{O}(|x|^2)$$ and $$\log \left | \frac{x}{|x|} - a |x| \right | = - \langle a, x \rangle
+ \mathcal{O} (|x|^2)$$ we expand $u_{\epsilon,a}(x)$ as $$\begin{aligned}
u_{\epsilon,a} (x) & = & |x|^{\frac{4-n}{2}} \left ( 1 + \frac{n-4}{2}
\langle x,a \rangle + \mathcal{O}(|x|^2) \right ) \left (
v_\epsilon(-\log |x|) - \langle a,x \rangle \dot v_\epsilon(-\log |x|)
+ \mathcal{O}(|x|^2) \right ) \\
& = & |x|^{\frac{4-n}{2}} \left ( v_\epsilon(-\log |x|)
+ \langle x,a\rangle \left ( - \dot v_\epsilon(-\log |x|) + \frac{n-4}{2}
v_\epsilon(-\log |x|) \right ) + \mathcal{O}(|x|^2) \right ) \\
& = & u_\epsilon(x) + |x|^{\frac{4-n}{2}} \langle x,a\rangle
\left ( -\dot v_\epsilon + \frac{n-4}{2} v_\epsilon \right ) +
\mathcal{O}(|x|^{\frac{8-n}{2}}) \end{aligned}$$ as $|x| \rightarrow 0$. Rewriting this in cylindrical coordinates gives $$\label{trans_del_soln3}
v_{\epsilon, a}(t,\theta) = v_\epsilon(t) + e^{-t} \langle \theta, a \rangle
\left ( - \dot v_\epsilon (t) + \frac{n-4}{2} v_\epsilon(t) \right ) +
\mathcal{O}(e^{-2t})$$ as $t \rightarrow \infty$. Unfortunately, the same expansion for $\bar v_{\epsilon,a}$ reveals $$\bar v_{\epsilon,a}(t,\theta) = e^t \left ( \dot v_\epsilon (t) + \frac{n-4}{2}
v_\epsilon (t) \right ) + \mathcal{O}(1).$$ In fact, one expects this behavior, as the motion generating the family $\bar v_{\epsilon, a}$ translates the origin in Euclidean coordinates, which translates the end in cylindrical coordinates in which $t \rightarrow \infty$.
It will be convenient for our later computations to observe that $$\begin{aligned}
\label{trans_nodal_domains1}
&\langle a,\theta\rangle > 0 \Rightarrow v_{\epsilon,a} (t,\theta) > v_\epsilon(t) , \qquad
\langle a,\theta \rangle < 0 \Rightarrow v_{\epsilon, a} (t,\theta) < v_\epsilon(t) &
\\ \nonumber
& \langle a,\theta \rangle > 0 \Rightarrow \bar v_{\epsilon,a} (t,\theta) >
v_\epsilon (t) , \qquad \langle a,\theta \rangle < 0 \Rightarrow \bar v_{\epsilon,a}
(t,\theta) < v_\epsilon (t) .& \end{aligned}$$
One can find a first integral for the ODE . Indeed, differentiating once shows $$\label {delaunay_energy}
\mathcal{H}_\epsilon = -\dot v_\epsilon \dddot v_\epsilon
+ \frac{1}{2} \ddot v_\epsilon^2 + \left ( \frac{n(n-4) + 8}{4} \right )
\dot v_\epsilon^2 - \frac{n^2(n-4)^2}{32} v_\epsilon^2 + \frac{(n-4)^2
(n^2-4)}{32} v_\epsilon^{\frac{2n}{n-4}}$$ is constant function for each Delaunay solution $v_\epsilon$. Evaluating this energy on the cylindrical solution gives $$\label{cyl_energy}
\mathcal{H}_{cyl} = -\frac{(n-4)(n^2-4)}{8} \left ( \frac{n(n-4)}{n^2-4}
\right )^{\frac{n}{4}} < 0$$ and evaluating on the spherical solution gives $$\begin{aligned}
\label{sph_energy}
\mathcal{H}_{sph} & = & \frac{(n-4)^2}{4} (\cosh t)^{-n} \left (
\frac{n(n-2)}{4} \sinh^4 t+ \left ( \frac{4-n}{2} + 2-n\right ) \cosh^2 t \sinh^2 t
- \frac{1}{2} \cosh^4 t \right . \\ \nonumber
&& \qquad \qquad + \frac{(n-2)}{4}\cosh^2 t\sinh^2 t - \frac{(n-2)^2}{8}
\sinh^4 t - \left ( \frac{n(n-4)+8}{4} \right ) \cosh^2 t\sinh^2 t \\ \nonumber
&& \qquad \qquad \left. + \frac{n^2}{8}
\cosh^4 t - \frac{(n^2-4)}{8} \right ) \\ \nonumber
& = & \frac{(n-4)^2}{4}(\cosh t)^{-n} \left ( -\frac{(n^2-4)}{8} + \frac{(n^2-4)}{8}
\cosh^4 t + \frac{(4-n^2)}{4} \cosh^2 t \sinh^2 t + \frac{(n^2-4)}{8}
\sinh^4 t \right ) \\ \nonumber
& = & 0. \end{aligned}$$
It is a consequence of Proposition 6 of [@FK] that $\mathcal{H}_\epsilon$ is a strictly decreasing function of $\epsilon$.
Linearization of the PDE
-------------------------
Let $u$ solve and consider the slightly perturbed function $u+ \epsilon w$. We have $$\label{linear_rhs1}
\frac{n(n-4)(n^2-4)}{16} (u+\epsilon w)^{\frac{n+4}{n-4}}
= \frac{n(n-4)(n^2-4)}{16} u^{\frac{n+4}{n-4}} + \epsilon
\frac{n(n+4)(n^2-4)}{16} w u^{\frac{8}{n-4}} + \mathcal{Q}(\epsilon w),$$ where $$\label {linear_rhs2}
\mathcal{Q}(\epsilon w) = \frac{n(n-4)(n^2-4)}{16} \left [
(u+\epsilon w)^{\frac{n+4}{n-4}} - u^{\frac{n+4}{n-4}}
- \epsilon \frac{n+4}{n-4} w u^{\frac{8}{n-4}} \right ] = \mathcal{O}(\epsilon^2
\| w\|^2).$$ Combining with and we have $$\Delta^2 u + \epsilon \Delta^2 w = \frac{n(n-4)(n^2-4)}{16}
u^{\frac{n+4}{n-4}} + \epsilon \frac{n(n+4)(n^2-4)}{16} w
u^{\frac{8}{n-4}}
+ \mathcal{Q}(\epsilon w),$$ where $\mathcal{Q}(\epsilon w)$ is of order $\epsilon^2 \| w \|^2$. Selecting the terms of order $\epsilon$ from the last equation we obtain the linearization $$\label{linearization1}
\Delta^2 w = \frac{n(n+4)(n^2-4)}{16} u^{\frac{8}{n-4}} w.$$ We refer to a solution $w$ of as a [**Jacobi field**]{} associated to the solution $u$.
Finally, it is useful to rewrite this last linearized equation in cylindrical coordinates, obtaining $$\frac{\partial^4 w}{\partial t^4} - \left ( \frac{n(n-4)
+8}{2} \right )\frac{\partial^2 w}{\partial t^2} + \frac{n^2(n-4)^2}
{16} w + 2 \Delta_\theta \frac{\partial^2 w}{\partial t^2}
- \frac{n(n-4)}{2} \Delta_\theta w + \Delta_\theta^2 w
= \frac{n(n+4)(n^2-4)}{16} w v^{\frac{8}{n-4}},$$ which we can rearrange to read $$\begin{aligned}
\label{linearization2}
0 & = & \frac{\partial^4 w}{\partial t^4} - \left ( \frac{n(n-4)
+ 8}{2} \right ) \frac{\partial^2 w}{\partial t^2} +
\left ( \frac{n^2(n-4)^2}{16} - \frac{n(n+4)(n^2-4)}{16}
v^{\frac{8}{n-4}} \right ) w \\ \nonumber
&& + 2 \Delta_\theta \frac{\partial^2 w}{\partial t^2}
- \frac{n(n-4)}{2} \Delta_\theta w + \Delta_\theta^2 w. \end{aligned}$$ Again, we refer to a solution $w$ of as a Jacobi field associated to the solution $v$.
Integral identities
-------------------
The following is essentially a special case of Proposition 4.2 in [@DMO], and also a special case of Proposition A.2 of [@JX].
\[poho\_prop\] Let $u \in \mathcal{C}^\infty((0,\infty) \times \mathbf{S}^{n-1})$ solve and let $0<T_1< T_2$. Then $$\begin{aligned}
\label{poho1}
& \int_{\{ T_1\} \times \mathbf{S}^{n-1}} \frac{\partial v}
{\partial t} \frac{\partial^3 v}{\partial t^3} - \frac{1}{2}
\left ( \frac{\partial^2 v}{\partial t^2} \right )^2 - \left (
\frac{n(n-4) + 8}{4} \right ) \left (\frac{\partial v}{\partial t}
\right )^2 + \frac{n^2(n-4)^2}{32} v^2 & \\ \nonumber
& \qquad - \frac{(n-4)^2(n^2-4)}{32} v^{\frac{2n}{n-4}} +
\frac{1}{2} (\Delta_\theta v)^2 + \frac{n(n-4)}{4} |\nabla_\theta
v|^2 - \left | \nabla_\theta \frac{\partial v}{\partial t}
\right |^2 d\sigma \\ \nonumber
= & \int_{\{ T_2\} \times \mathbf{S}^{n-1}} \frac{\partial v}
{\partial t} \frac{\partial^3 v}{\partial t^3} - \frac{1}{2}
\left ( \frac{\partial^2 v}{\partial t^2} \right )^2 - \left (
\frac{n(n-4) + 8}{4} \right ) \left (\frac{\partial v}{\partial t}
\right )^2 + \frac{n^2(n-4)^2}{32} v^2 & \\ \nonumber
& \qquad - \frac{(n-4)^2(n^2-4)}{32} v^{\frac{2n}{n-4}} +
\frac{1}{2} (\Delta_\theta v)^2 + \frac{n(n-4)}{4} |\nabla_\theta
v|^2 - \left | \nabla_\theta \frac{\partial v}{\partial t}
\right |^2 d\sigma \end{aligned}$$
Multiplying by $\frac{\partial v}{\partial t}$ and integrating over the sphere $\{ t \} \times \mathbf{S}^{n-1}$ we obtain $$\begin{aligned}
0 & = & \int_{\{ t\} \times \mathbf{S}^{n-1}}
\frac{\partial v}{\partial t} \frac{\partial^4 v}{\partial
t^4} - \frac{n(n-4) + 8}{2} \frac{\partial^2 v}{\partial t^2}
\frac{\partial v}{\partial t} + \frac{n^2(n-4)^2}{16} v \frac
{\partial v}{\partial t} \\
&& \qquad + \frac{\partial v}{\partial t} \Delta^2_\theta v-
\frac{n(n-4)}{2} \frac{\partial v}{\partial t} \Delta_\theta v
+ 2 \frac{\partial v}{\partial t} \Delta_\theta \frac{\partial^2 v}
{\partial t^2} - \frac{n(n-4)(n^2-4)}{16} v^{\frac{n+4}{n-4}}
\frac{\partial v}{\partial t} d\sigma \\
& = & \int_{\{ t \} \times \mathbf{S}^{n-1}} \frac{\partial}{\partial t}
\left [ \frac{\partial v}{\partial t}\frac{\partial^3 v}{\partial t^3}
- \frac{1}{2} \left ( \frac{\partial^2 v}{\partial t^2} \right )^2
- \left ( \frac{n(n-4) + 8}{4} \right ) \left (\frac{\partial v}
{\partial t} \right )^2 + \frac{n^2(n-4)^2}{32} v^2 \right . \\
& & \qquad \qquad - \frac{(n-4)^2(n^2-4)}{32} v^{\frac{2n}
{n-4}} - \frac{1}{2} (\Delta_\theta v)^2 + \frac{n(n-4)}{4}
|\nabla_\theta v|^2 \\
&& \qquad \qquad - \left . \frac{n(n-4)}{2} \operatorname{div}_\theta
\left ( \frac{\partial v}{\partial t} \nabla_\theta v \right )
- \left | \nabla_\theta \frac{\partial v}{\partial t} \right |^2
+ 2 \operatorname{div}_\theta \left ( \frac{\partial v}{\partial t}
\nabla_\theta \frac{\partial^2 v}{\partial t^2} \right ) \right ] d\sigma \\
& = & \frac{d}{dt} \int_{\{ t \} \times \mathbf{S}^{n-1}}
\frac{\partial v}{\partial t}\frac{\partial^3 v}{\partial t^3}
- \frac{1}{2} \left ( \frac{\partial^2 v}{\partial t^2} \right )^2
- \left ( \frac{n(n-4) + 8}{4} \right ) \left (\frac{\partial v}
{\partial t} \right )^2 + \frac{n^2(n-4)^2}{32} v^2\\
& & \qquad \qquad - \frac{(n-4)^2(n^2-4)}{32} v^{\frac{2n}
{n-4}} - \frac{1}{2} (\Delta_\theta v)^2 + \frac{n(n-4)}{4}
|\nabla_\theta v|^2
- \left | \nabla_\theta \frac{\partial v}{\partial t} \right |^2
d\sigma. \end{aligned}$$
Following this integral identity, we define $$\begin{aligned}
\label{radial_poho}
\mathcal{H}_{rad}(v) & = & \int_{ \{ t\} \times {\mathbf{S}}^{n-1}}
-\frac{\partial v}{\partial t} \frac{\partial^3 v}{\partial t^3} +
\frac{1}{2} \left ( \frac{\partial^2 v}{\partial t^2} \right )^2 +
\left ( \frac{n(n-4)+8}{4} \right )\left ( \frac{\partial v}{\partial t} \right )^2
\\ \nonumber
&& \quad - \frac{n^2(n-4)^2}{32} v^2
+ \frac{(n-4)^2(n^2-4)}{32} v^{\frac{2n}{n-4}} - \frac{1}{2}
(\Delta_\theta v)^2 - \frac{n(n-4)}{4} |\nabla_\theta v|^2 + \left |
\nabla_\theta \frac{\partial v}{\partial t} \right |^2 d\sigma ,\end{aligned}$$ which does not depend on $t$ by .
If $v_\epsilon$ is the Delaunay solution of described above then $$\label{poho2}
\mathcal{H}_{rad} (v_\epsilon)
= n \omega_n \left ( \frac{1}{2} \ddot v_\epsilon^2 (0)+ \frac{(n-4)^2}{32}
\epsilon^2 [(n^2-4) \epsilon^{\frac{8}{n-4}} - n^2] \right )
=n \omega_n \mathcal{H}_\epsilon < 0 .$$ where $\omega_n$ is the volume of a unit ball in ${\mathbf{R}}^n$.
A priori estimates
------------------
Jin and Xiong [@JX] prove the following.
Let $u\in \mathcal{C}^\infty ({\mathbf{B}}_1(0) \backslash
\{ 0 \})$ be a positive solution of such that $-\Delta u > 0$. Then either $u$ admits a continuous extension to ${\mathbf{B}}_1(0)$ or there exist positive constants $C_1< C_2$ such that $$\label{apriori_bounds2}
C_1 |x|^{\frac{4-n}{2}} \leq u(x) \leq C_2 |x|^{\frac{4-n}{2}}.$$
The example of the Delaunay solutions demonstrate that the constant $C_1$ in must depends on the choice of the solution $u$. However, the constant $C_2$ in the upper bound is universal.
Let $v \in \mathcal{C}^\infty ((0, \infty) \times {\mathbf{S}}^{n-1})$ be a positive solution of which also satisfies . Then either $\limsup_{t
\rightarrow \infty} v(t,\theta) = 0$ or there exist positive constants $C_1 < C_2$ such that $$C_1 < v(t,\theta) < C_2 .$$ Moreover, in this case $\mathcal{H}_{rad} (v) \neq 0$.
The bounds imply $C_1 < v(t,\theta)< C_2$. Choose a sequence $\tau_i \nearrow
\infty$ and define $v_i (t,\theta) = v(t+\tau_i, \theta)$. This sequence is uniformly bounded, so we may extract a subsequence, still denoted by $\{ v_i\}$, which converges uniformly on compact subsets of ${\mathbf{R}}\times {\mathbf{S}}^{n-1}$ to a solution $\bar v$ of . However, the limit $\bar v$ must be a Delaunay solution $v_\epsilon$, and so (using Proposition \[poho\_prop\]) we have $$\begin{aligned}
\mathcal{H}_{\textrm{rad}} (v) & = & \int_{ \{ 1 \} \times {\mathbf{S}}^{n-1}}
- \frac{\partial v}{\partial t}
\frac{\partial^3 v}{\partial t^3}+ \frac{1}{2} \left ( \frac{\partial^2 v}
{\partial t^2} \right )^2 + \left ( \frac{n(n-4) + 8}{4} \right ) \left ( \frac
{\partial v}{\partial t} \right )^2 \\
&& \quad + \frac{n^2(n-4)^2}{32} v^2 + \frac{(n-4)^2(n^2-4)}{32}
v^{\frac{2n}{n-4}} - \frac{1}{2} (\Delta_\theta v)^2- \frac{n(n-4)}{4}
\left | \nabla_\theta \frac{\partial v}{\partial t} \right |^2 d\sigma \\
&= & \int_{ \{ \tau_i +1 \} \times {\mathbf{S}}^{n-1}}
- \frac{\partial v}{\partial t}
\frac{\partial^3 v}{\partial t^3}+ \frac{1}{2} \left ( \frac{\partial^2 v}
{\partial t^2} \right )^2 + \left ( \frac{n(n-4) + 8}{4} \right ) \left ( \frac
{\partial v}{\partial t} \right )^2 \\
&& \quad + \frac{n^2(n-4)^2}{32} v^2 + \frac{(n-4)^2(n^2-4)}{32}
v^{\frac{2n}{n-4}} - \frac{1}{2} (\Delta_\theta v)^2- \frac{n(n-4)}{4}
\left | \nabla_\theta \frac{\partial v}{\partial t} \right |^2 d\sigma \\
& = & \mathcal{H}_{\textrm{rad}} (v_i) \rightarrow
\mathcal{H}_{\textrm{rad}} (v_\epsilon) = n \omega_n \mathcal{H}_\epsilon
\neq 0.\end{aligned}$$
Linear analysis {#lin_anal_sec}
===============
In this section we study the mapping properties of the linear operator , concentrating on the linearization about a Delaunay metric.
Definitions
-----------
Linearizing about a Delaunay solution $v_\epsilon$ we obtain the operator $$\begin{aligned}
\label{del_linearization1}
L_\epsilon & = & \frac{\partial^4}{\partial t^4} + \Delta_\theta^2 + 2 \Delta_\theta
\frac{\partial^2}{\partial t^2} - \frac{n(n-4)}{2} \Delta_\theta \\ \nonumber
&& - \left ( \frac{n(n-4) + 8}
{2} \right ) \frac{\partial^2 }{\partial t^2} + \left ( \frac{n^2 (n-4)^2}{16} -
\frac{n(n+4)(n^2-4)}{16} v_\epsilon^{\frac{8}{n-4}} \right ) .\end{aligned}$$ We refer to solutions of the equation $L_\epsilon (w)$ as [**Jacobi fields**]{} associated to the Delaunay solution $v_\epsilon$. More generally, if $v$ satisfies and $w$ satisfies then we call $w$ a Jacobi field associated to the solution $v$. We are interested in the mapping properties of $L_\epsilon$, for instance as the map $$L_\epsilon : W^{4,2}((0,\infty) \times {\mathbf{S}}^{n-1}) \rightarrow L^2
((0, \infty) \times {\mathbf{S}}^{n-1}).$$ It turns out the operator written above does not have closed range (see pg. 21, pg. 216, and Theorem 5.40 of [@Mel]), so we will to define certain weighted function spaces to accomodate $L_\epsilon$.
Let $\gamma \in {\mathbf{R}}$ and when $u \in L^2((0,\infty) \times {\mathbf{S}}^{n-1})_{loc}$ define $$\label{cyl_weight_norm}
\| u\|^2_{L^2_\gamma} = \int_0^\infty \int_{{\mathbf{S}}^{n-1}}
e^{-2\gamma t} | u(t, \theta)|^2 d\theta dt .$$ The space $L^2_\gamma((0,\infty) \times {\mathbf{S}}^{n-1})$ is the space of functions with finite norm, as defined above. One can similarly define the Sobolev spaces $W^{k,2}_\gamma$ of functions with $k$ weak derivatives in $L^2$ having finite weighted norms.
One can also define weighted Hölder spaces.
Let $\gamma\in {\mathbf{R}}$ and $\alpha \in (0,1)$. For $u \in \mathcal{C}^{0,\alpha}_{loc}((0,\infty) \times {\mathbf{S}}^{n-1})$ define $$\| u \|_{\mathcal{C}^{0,\alpha}_\gamma} =
\sup_{t_0 > 1} \sup \left \{ \frac{ e^{-\gamma t_1} u(t_1,\theta_1) -
e^{-\gamma t_2} u(t_2, \theta_2)}{d((t_1, \theta_1), (t_2, \theta_2))^\alpha} :
(t_1, \theta_1), (t_2, \theta_2) \in (t_0-1, t_0+1) \times {\mathbf{S}}^{n-1} \right \}.$$ One can similarly define weighted Hölder spaces with more derivatives.
Heuristically, a function in a weighted function space with weight $\gamma$ is bounded from above by a multiple of $e^{\gamma t}$ as $t \rightarrow \infty$. Observe then that when $\gamma_1 < \gamma_2$ we have the inclusions $$W^{k,2}_{\gamma_1}((0,\infty) \times {\mathbf{S}}^{n-1}) \subset
W^{k,2}_{\gamma_2}((0,\infty) \times {\mathbf{S}}^{n-1}), \qquad
\mathcal{C}^{k,\alpha}_{\gamma_1}((0,\infty) \times {\mathbf{S}}^{n-1}) \subset
\mathcal{C}^{k,\alpha}_{\gamma_2}((0,\infty) \times {\mathbf{S}}^{n-1}) .$$
The fact that $L_\epsilon$ the leading order terms of $L_\epsilon$ is $\frac{\partial ^4}{\partial t^4} + \Delta_\theta^2$ tells us the following.
For any $\gamma \in {\mathbf{R}}$ the operators $$L_\epsilon : W^{k+4, 2}_\gamma((0, \infty) \times {\mathbf{S}}^{n-1}) \rightarrow W^{k,2}_\gamma
((0, \infty) \times {\mathbf{S}}^{n-1})$$ and $$L_\epsilon : \mathcal{C}^{k+4, \alpha}_\gamma((0, \infty) \times {\mathbf{S}}^{n-1}) \rightarrow
\mathcal{C}^{k,\alpha}_\gamma ((0, \infty) \times {\mathbf{S}}^{n-1})$$ are bounded, linear, and elliptic.
We complete our understanding of $L_\epsilon$ by identifying the weights for which $$L_\epsilon : W^{k+4, 2}_\gamma((0, \infty) \times {\mathbf{S}}^{n-1}) \rightarrow W^{k,2}_\gamma
((0, \infty) \times {\mathbf{S}}^{n-1})$$ is Fredholm, injective, and/or surjective. Indeed, this is a nontrivial task.
We make our analysis easier by decomposing $w$ in spherical harmonics, writing $$w(t,\theta) = \sum_{j \in \mathbf{Z}} w_j (t) \phi_j,$$ where $\phi_j$ is a normalized eigenfunction of $\Delta_\theta$ on ${\mathbf{S}}^{n-1}$, [*i.e.*]{} $$\label{spherical_eigenfunc}
\Delta_\theta \phi_j = - \lambda_j \phi_j, \qquad \int_{{\mathbf{S}}{n-1}}
\phi_j \phi_k d\theta = \delta_{jk}.$$ The eigenvalues $\lambda_j$ of the $(n-1)$-dimensional sphere have the form $\lambda_j = k(n-2+k)$ for some $k=0,1,2,3,\dots$.
Under this decomposition the Fourier coefficient $w_j$ solves the ODE $$\begin{aligned}
\label{paneitz_ode3}
0 & = & L_{\epsilon, j} w_j \\ \nonumber
& = & \ddddot w_j - \left ( \frac{n(n-4) + 8 + 4\lambda_j}{2} \right ) \ddot w_j
+ \left ( \frac{n^2(n-4)^2}{16} - \frac{n(n+4)(n^2-4)}{16} v_\epsilon^{\frac{8}{n-4}}
+ \frac{n(n-4)}{2} \lambda_j+ \lambda_j^2 \right ) w_j. \end{aligned}$$ It immediately follows that $$\label{spec_decomp}
\operatorname{spec}(L_\epsilon) = \bigcup_{j=0}^\infty \operatorname{spec}
(L_{\epsilon,j}).$$
Low Fourier modes
-----------------
We can explicitly identify some of the ODE solutions when $|j|$ is small. For instance, we have $\lambda_0 = 0$ and so becomes $$\label{paneitz_ode4}
\ddddot w_0 - \left ( \frac{n(n-4) +8}{2} \right ) \ddot w_0 +
\left ( \frac{n^2(n-4)^2}{16} - \frac{n(n+4)(n^2-4)}{16} v_\epsilon^{\frac{8}
{n-4}} \right )w_0 =0,$$ which is the derivative of , Thus $$\label{low_paneitz_mode1}
w_0^+ (t) = \dot v_\epsilon(t) , \qquad w_0^- = \frac{d}{d\epsilon}
v_\epsilon (t)$$ both solve .
\[zero\_ind\_roots\] The function $w_0^+$ is periodic with period $T_\epsilon$ while the function $w_0^-$ grows linearly.
Differentiating the equation $v_\epsilon(t) = v_\epsilon(t+T_\epsilon)$ with respect to $t$ gives $w_0^+(t+T_\epsilon) = w_0(t)$. Differentiating $v_\epsilon(t+T_\epsilon) = v_\epsilon(t)$ with respect to $\epsilon$ gives $$w_0^-(t+T_\epsilon) \frac{d T_\epsilon}{d\epsilon} = w_0^-
(t),$$ and so $w_0^-$ grows linearly.
We can also explicitly identify the Fourier modes when $j=1,2,\dots, n$ and $\lambda_j = n-1$. To do this we first let $e_j$ be the standard basis of ${\mathbf{R}}^n$ and observe $\phi_j = \langle e_j, \theta \rangle$ is the eigenfunction associated to $\lambda_j = n-1$. Substituting $a= \tau e_j$ into and we find $$\begin{aligned}
v_{\epsilon, \tau e_j} & = & |\theta - e^{-t} \tau e_j |^{\frac{4-n}{2}}
v_\epsilon (t +\log |\theta - e^{-t} \tau \theta|) \\
& = & v_\epsilon(t) + \tau e^{-t} \langle \theta, e_j \rangle \left (
-\dot v_\epsilon (t) + \frac{n-4}{2} v_\epsilon (t) \right ) + \mathcal{O}
(e^{-2t}) \\
& = & v_\epsilon (t) + \tau e^{-t} \phi_j \left ( -\dot v_\epsilon (t)
+ \frac{n-4}{2} v_\epsilon (t) \right ) + \mathcal{O} (e^{-2t}). \end{aligned}$$ Differentiating with respect to $\tau$ we obtain $$w_j^- (t) \phi_j(\theta) = \left. \frac{d}{d\tau} \right |_{\tau = 0}
v_{\epsilon, \tau e_j} = e^{-t} \left ( -\dot v_\epsilon (t) + \frac{n-4}{2}
v_\epsilon (t) \right ) \phi_j(\theta) + \mathcal{O}(e^{-2t}) ,$$ or $$\label{low_paneitz_mode2}
w_j^- = e^{-t} \left ( -w_0^+ + \frac{n-4}{2} v_\epsilon \right )
+ \mathcal{O}(e^{-2t}) = e^{-t} \left (- \dot v_\epsilon + \frac{n-4}{2} v_\epsilon
\right ) + \mathcal{O} (e^{-2t} ).$$
However, $v_{\epsilon, \tau e_j}$ satisfies for each $\tau$. Differentiating this relation and using $-\Delta_\theta
\phi_j = (n-1)\phi_j = \lambda_j \phi_j$ we find $$\begin{aligned}
\frac{n(n+4)(n^2-4)}{16} v_\epsilon^{\frac{8}{n-4}} w_j^- \phi_j & = &
\left ( \frac{\partial^4}{\partial t^4} - \left ( \frac{n(n-4)+8}{2} \right ) \frac{\partial^2}
{\partial t^2} + \frac{n^2(n-4)^2}{16} \right ) w_j^- \phi_j \\
&& + \left ( \Delta_\theta^2 + 2 \Delta_\theta \frac{\partial^2}{\partial t^2} -
\frac{n(n-4)}{2} \Delta_\theta \right ) w_j^- \phi_j \\
& = & \left ( \ddddot w_j^- - \left ( \frac{n(n-4)+8+4\lambda_j}{2} \right ) \ddot w_j^-
+ \left ( \frac{n^2 (n-4)^2}{16} + \lambda_j^2 + \frac{n(n-4)}{2} \lambda_j \right )
w_j^- \right ) \phi_j, \end{aligned}$$ which we can rearrange to give $L_{\epsilon, j} (w_j^-) = 0$.
A similar calculation, starting from $$w_j^+ (t) \phi_j(\theta) = \left. \frac{d}{d\tau} \right |_{\tau = 0}
\bar v_{\epsilon, \tau e_j},$$ gives the expansion $$\label{low_paneitz_mode3}
w_j^+ = e^t \left ( w_0^+ + \frac{n-4}{2} v_\epsilon \right )
+ \mathcal{O}(1) = e^t \left ( \dot v_\epsilon + \frac{n-4}{2}
v_\epsilon \right ) + \mathcal{O}(1) .$$
Observe that implies $$\label{trans_nodal_domains2}
w_j^+ > 0, \qquad w_j^- > 0 .$$
It is not surprising that $w_j^\pm(t)$ all agree for $j=1,2,\dots, n$, as each translation is geometrically the same. It is also not surprising that $w_j^+$ grows exponentially while $w_j^-$ decays exponentially. The ambient motion generating $w_j^+$ translates the origin in Euclidean coordinates, which in cylindrical coordinates moves the end corresponding to $t \rightarrow \infty$, whereas the ambient motion generating $w_j^-$ moves the end corresponding to $t\rightarrow -\infty$.
Indicial roots {#indicial_sect}
--------------
We further analyze the ODE for general values of $j$, that is $$\begin{aligned}
0 & = & L_{\epsilon, j} w \\
& = & \ddddot w - \left ( \frac{n(n-4) + 8 + 4\lambda_j}{2} \right ) \ddot w +
\left ( \frac{n^2(n-4)^2}{16} - \frac{n(n+4)(n^2-4)}{16} v_\epsilon^{\frac{8}{n-4}}
+\lambda_j^2 + \frac{n(n-4)}{2} \lambda_j \right ) w .\end{aligned}$$ The coefficients of this ODE are all periodic with period $T_\epsilon$, so there is a (constant) $4 \times 4$ matrix $A_{\epsilon,j}$ such that $$\label{indicial1}
\left ( \begin {array}{c} w(t+T_\epsilon)\\ \dot w(t+T_\epsilon) \\
\ddot w(t+T_\epsilon) \\ \dddot w(t+T_\epsilon) \end {array} \right ) =
A_{\epsilon,j} \left ( \begin {array}{c} w(t) \\ \dot w(t) \\ \ddot w(t) \\
\dddot w(t) \end {array} \right ).$$ Now let $w_1, w_2,w_3,w_4$ be four solutions of , and let $$\label{wronskian1}
W(t) = \det \left ( \begin {array}{cccc} w_1 & w_2 & w_3 & w_4 \\
\dot w_1 & \dot w_2 & \dot w_3 & \dot w_4 \\ \ddot w_1 & \ddot w_2
& \ddot w_3 & \ddot w_4 \\ \dddot w_1 & \dddot w_2 & \dddot w_3
& \dddot w_4 \end {array} \right ) (t)$$ be the associated Wronskian determinant. By Abel’s identity, $\frac{dW}{dt} = 0$ and so $W$ is constant. Combining and we see that $\det A_{\epsilon,j} = 1$. Moreover, the matrix $A_{\epsilon,j}$ has real coefficients, so its eigenvalues occur in conjugate pairs. Suppressing the dependence on $\epsilon$ for the moment, we denote these eigenvalues as $$\label{indicial2}
\mu_j^\pm = e^{\pm i \xi_{\epsilon,j}}, \qquad
\tilde \mu_j^\pm = e^{\pm i \tilde \xi_{\epsilon,j}} .$$ Here we define the [**indicial roots**]{} of the operator $L_{\epsilon,j}$ as all the real numbers $\gamma$ such that $\gamma = \Im (\xi)$ where $\mu=e^{i\xi}$ is an eigenvalue of $A_{\epsilon,j}$. For convenience later on, we collect these numbers as $$\label{indicial_defn}
\Gamma_{\epsilon,j} = \{ \gamma \in {\mathbf{R}}:
\gamma = \Im(\xi) \textrm{ and } \mu =e^{i\xi}
\textrm{ is an eigenvalue of }A_{\epsilon,j} \}, \qquad
\Gamma_\epsilon = \bigcup_{j=0}^\infty \Gamma_{\epsilon,j}.$$ Observe that $\Gamma_{\epsilon,j}$ has at most four elements, so in particular $\Gamma_\epsilon$ is countable. Moreover, by construction $\Gamma_{\epsilon,j}$ is even for each $j$, [*i.e.*]{} $\gamma \in \Gamma_{\epsilon, j}$ if and only if $-\gamma \in \Gamma_{\epsilon,j}$.
The fact that the eigenvalues of $A_{\epsilon,j}$ occur in conjugate pairs implies we can always write the eigenvalues as $\mu^\pm = e^{\pm i \xi}$ and $\tilde \mu^\pm =
e^{\pm i \tilde \xi}$. Writing $\xi = \eta + i \nu$ and $\tilde
\xi = \tilde \eta + i \tilde \nu$ we see that $|\mu^\pm| = e^{\pm
\nu}$ and $|\tilde \mu^\pm| = e^{\pm \tilde \nu}$. In other words, the indicial roots precisely determine the rates of exponential growth of the solutions of .
For each $\epsilon \in (0,v_{cyl}]$ we have $0 \in
\Gamma_{\epsilon,0}$ (with multiplicity $2$) and $\{
-1,1 \} \subset \Gamma_{\epsilon,j}$ for $j=1,2,\dots, n$.
Lemma \[zero\_ind\_roots\] implies $0 \in \Gamma_{\epsilon,0}$ while and together imply $\{ \pm 1\}
\subset \Gamma_{\epsilon, j}$ for $1\leq j \leq n$.
Now we explicitly compute the Jacobi fields and the indicial roots for the special case of the cylindrical metric. We let $\epsilon_n$ denote the Delaunay parameter of the cylindrical solution, that is $$\label{cyl_eps}
\epsilon_n = \left ( \frac{n(n-4)}{n^2-4} \right ) ^{\frac{n-4}{8}},$$ so that $$\label{cyl_fourier_odes}
L_{\epsilon_n, j} = \frac{d^4}{dt^4} - \left ( \frac{n(n-4) + 8
+ 4\lambda_j}{2} \right ) \frac{d^2}{dt^2} + \left ( -\frac{n^2(n-4)}{2}
+ \frac{n(n-4)}{2} \lambda_j + \lambda_j^2 \right ) .$$ Fortunately we can explicity solve the ODEs $L_{\epsilon_n, j} w = 0$. Substituting $$w(t) = c_+ e^{\mu_j t} + c_- e^{-\mu_j t} + \widetilde c_+
e^{\widetilde \mu_j t} + \widetilde c_- e^{-\widetilde \mu t}$$ and using $\lambda_j = k(n-2+k)$ for some nonnegative integer $k$ we find $$\begin{aligned}
\label{cyl_fourier_exp1}
\mu_j^2 & = & \frac{1}{2} \left ( \frac{n(n-4)+8+4\lambda_j}{2} +
\sqrt{\frac{n^4}{4} - 16(n-1-\lambda_j)} \right ) \\ \nonumber
& = & \frac{1}{2} \left ( \frac{(n+2(k-1))^2+ 4}{2} + \sqrt{ \frac{n^4}{4}
+ 16(k-1)(n+k-1)} \right )\end{aligned}$$ and $$\begin{aligned}
\label{cyl_fourier_exp2}
\widetilde \mu_j^2 & = & \frac{1}{2} \left ( \frac{n(n-4)+8+4\lambda_j}{2} -
\sqrt{\frac{n^4}{4} - 16(n-1-\lambda_j)} \right ) \\ \nonumber
& = & \frac{1}{2} \left ( \frac{(n+2(k-1))^2+ 4}{2} - \sqrt{ \frac{n^4}{4}
+ 16(k-1)(n+k-1)} \right ) , \end{aligned}$$ which then gives all solutions after taking square roots. We remark on some properties of $\mu_j$ and $\widetilde
\mu_j$. First observe that, because $n > 4$, $$\frac{n^4}{4} - 16n +16+16\lambda_j \geq n^4-16n +16 > 16\left (
\frac{n^2}{4} -n+1 \right ) >0,$$ so that, in particular, $\mu_j^2$ and $\widetilde \mu_j^2$ are real numbers. Next we observe that $\mu_j^2 > 0$ for each integer $j$. After taking a positive and a negative square root gives us one Jacobi field which grows exponentially and one exponentially decaying Jacobi field, implying each $\Gamma_{\epsilon_n, j}$ contains one positive and one negative index. On the other hand, $\widetilde \mu_0^2<0$ while $\widetilde \mu_j^2 > 0$ for $j>0$. Furthermore, we can explicitly compute these indices when $j=1,2,
\dots, n$, in which case $k=1$ and $\lambda_j = n-1$ and $$\mu_j ^2 = \frac{n^2+2}{2}, \qquad \widetilde \mu_j^2 = 1.$$ Moreover, the fundamental period of the Jacobi field associated to $\widetilde \mu_0$ is $$\label{cyl_period}
T_{cyl} = T_{\epsilon_n} = \frac{2\pi}{\widetilde \mu_0} , \qquad
\widetilde \mu_0 = \frac{1}{2} \sqrt{ \sqrt{n^4 - 64n +64}
- n(n-4)+8}.$$ To summarize, we have proved the following lemma.
We have $$\label{cyl_indicial_roots1}
\Gamma_{\epsilon_n,0} = \left \{ 0, \pm \frac{1}{2}
\sqrt{n(n-4)+8 + \sqrt{n^4-64n+64}} \right \} ,$$ and for $j > 0$ $$\begin{aligned}
\label{cyl_indicial_roots2}
\Gamma_{\epsilon_n, j} & = & \left \{ \pm \frac{1}{2} \sqrt{(n+2(k-1))^2+4 -
\sqrt{n^4+64(k-1)(n+k-1)}}, \right . \\ \nonumber
&& \left. \pm \frac{1}{2} \sqrt{ (n+2(k-1))^2+4
+\sqrt{n^4+64(k-1)(n+k-1)}}\right \}, \end{aligned}$$ where $k$ is the positive integer corresponding to $\lambda_j =
k(n-2+k)$. In particular, $$\label{cyl_indicial_roots3}
\Gamma_{\epsilon_n, 1} = \cdots = \Gamma_{\epsilon_n, n}
= \left \{ \pm 1, \pm \sqrt{\frac{n^2+2}{2}} \right \}.$$
It follows from and that as $k \nearrow \infty$ we have $\gamma_{\epsilon_n, j} \simeq \sqrt{2} (k-1) + 2\sqrt{k-1}$ and $\widetilde \gamma_{\epsilon_n, j} \simeq \sqrt{2} (k-1)
- 2\sqrt{k-1}$. However, we will not need this information later.
For future calculations we will write the set of indicial roots as $$\Gamma_\epsilon = \{\dots, -\gamma_{\epsilon,2}, -\gamma_{\epsilon,1}=-1,
0, \gamma_{\epsilon,1}=1, \gamma_{\epsilon_2}, \dots \}$$ where $\gamma_{\epsilon,j} < \gamma_{\epsilon, j+1} \rightarrow \infty$. We will justify later the fact that $\Gamma_\epsilon$ has no accumulation points.
The Fourier-Laplace transform
-----------------------------
The following transform, defined in [@MPU], plays a key role in our understanding of the mapping properties of $L_\epsilon$ and $\{ L_{\epsilon,j} \}$.
Let $\gamma \in {\mathbf{R}}$ and let $w \in W^{k,2}_\gamma((0,\infty) \times {\mathbf{S}}^{n-1})$. Extend $w$ to be $0$ in the half-space $\{ t< 0 \}$ and define $$\label{fourier_laplace_defn}
\mathcal{F}_\epsilon(w)(t,\xi,\theta) = \widehat w(t, \xi, \theta) = \sum_{k=-\infty}^\infty
e^{-i\xi k} w (t+k T_\epsilon, \theta ).$$ Here $\xi \in \{ \eta +i\nu \in {\mathbf{C}}: \nu <-\gamma T_\epsilon\}$.
\[fourier\_laplace\_conv\] The sum in converges uniformly and absolutely when $w \in
W^{k,2}_\gamma((0,\infty) \times {\mathbf{S}}^{n-1})$ and $\nu = \Im(\xi) <
-\gamma T_\epsilon$. Equivalently, $$w \in W^{k,2}_\gamma ((0,\infty) \times {\mathbf{S}}^{n-1}) \Rightarrow
\mathcal{F}_\epsilon(w) \in \mathcal{C}^\omega (
\{ \Im(\xi) <-\gamma T_\epsilon \}, W^{k,2}
((0,\infty) \times {\mathbf{S}}^{n-1}).$$
We have seen that $w \in W^{k,2}_\gamma ((0,\infty) \times {\mathbf{S}}^{n-1})$ implies $|w(t,\theta)| = \mathcal{O} (e^{\gamma t})$. Writing $\xi = \eta + i\nu$, with $\eta, \nu \in {\mathbf{R}}$, we have $$\begin{aligned}
| \mathcal{F}_\epsilon (w) (t,\xi,\theta) | & \leq &
\sum_{k=-\infty}^\infty \left | e^{-i(\eta + i\nu) k} w(t+ kT_\epsilon ,\theta)
\right | = \sum_{k=-\infty}^{\infty} e^{k \nu}|w(t+kT_\epsilon,\theta)| \\
& \leq & C e^{\gamma t} \sum_{k=-\infty}^{\infty} e^{k(\nu +\gamma T_\epsilon)}.\end{aligned}$$ First observe that, since we have extended $w$ to be $0$ in the region $\{ t< 0\}$, each choice of $\xi = \eta + i \nu$ only gives finitely many nonzero terms with $k<0$, and so we only must resolve the convergence when $k \rightarrow
\infty$. In this case, all exponents are negative precisely when $\nu <
-\gamma T_\epsilon$.
Heuristically, the parameter $\xi$ (more specifically $\nu = \Im(\xi)$) allows us to move the weight as a parameter in the function space to one in the operator.
One can invert this transform, but (as expected) one must choose a branch in of the inversion.
Let $w \in W^{k,2}_\gamma((0,\infty) \times {\mathbf{S}}^{n-1})$ and let $\nu < -\gamma T_\epsilon$. For each $t$ choose $l \in \mathbf{Z}$ and $\tilde t \in [0,T_\epsilon)$ so that $t = \tilde t + l T_\epsilon$. Then $$\label{fourier_laplace_inv}
w(t,\theta) = \frac{1}{2\pi} \int_{\eta = 0}^{2\pi}
e^{ilT_\epsilon (\eta + i\nu)}
\widehat w(\tilde t, \eta + i\nu, \theta) d\eta .$$
Writing $\xi = \eta + i\nu$ we have $$\begin{aligned}
\frac{1}{2\pi} \int_{\eta = 0}^{2\pi}
e^{il \xi} \widehat w(\tilde t, \xi, \theta) d\eta
& = & \frac{1}{2\pi} \int_{\eta = 0}^{2\pi}
e^{il \xi} \sum_{k=-\infty}^\infty
e^{-ik \xi} w(\tilde t + k T_\epsilon, \theta) d\eta \\
& = & \sum_{k=-\infty}^\infty \frac{1}{2\pi} \int_{\eta = 0}^{2\pi}
e^{i(\eta + i\nu)(l-k)} w(\tilde t + kT_\epsilon, \theta) d\eta \\
& = & \frac{w(\tilde t + l T_\epsilon, \theta)} {2\pi}
\int_{\eta = 0}^{2\pi} d\eta \\
& = & w(t, \theta). \end{aligned}$$ Here we have used the fact that $\nu < -\gamma T_\epsilon$ to allow us to interchange the sum and the integral.
In fact, we can treat $\nu$ as a parameter in this inversion, and we see that changing $\nu$ alters the weight of the transformed function. We make this explicit with a version of the Parseval-Plancherel identity.
For each $\theta \in {\mathbf{S}}^{n-1}$ and $\nu \in {\mathbf{R}}$ we have $$\label{parseval1}
\| \widehat w(\cdot, \cdot + i\nu, \theta) \|^2_{L^2([0,T_\epsilon]
\times [0,2\pi])} \simeq 2\pi \| w(\cdot, \theta) \|_{L^2_{\nu/T_\epsilon}
({\mathbf{R}})},$$ where $a \simeq b$ means both $a = \mathcal{O} ( b)$ and $b =
\mathcal{O} (a)$.
We compute $$\begin{aligned}
\int_0^{T_\epsilon} \int_0^{2\pi} |\widehat w(t,\eta+i\nu,\theta)|^2 d\eta
dt & = & \int_0^{T_\epsilon} \int_0^{2\pi} \left ( \sum_{k=-\infty}^\infty
e^{-ik\eta} e^{k\nu} w(t+kT_\epsilon,\theta) \right ) \left ( \sum_{l=-\infty}
^\infty e^{il\eta} e^{l\nu} w(t+lT_\epsilon,\theta) \right ) d\eta dt \\
& = & \int_0^{T_\epsilon} \int_0^{2\pi} \sum_{k=-\infty}^\infty
\sum_{l=-k}^k \binom{k}{l} e^{i(l-k)\eta} e^{(k+l)\nu} w(t+kT_\epsilon, \theta)
w(t+lT_\epsilon,\theta) d\eta dt \\
& = & \int_0^{T_\epsilon} \int_0^{2\pi} \sum_{k=-\infty}^\infty
e^{2\nu k} (w(t+kT_\epsilon, \theta))^2 d\eta dt \\
& \simeq & 2\pi \int_{\mathbf{R}}(e^{\nu t/T_\epsilon} w(t,\theta))^2 dt. \end{aligned}$$ Observe that the integrals of all the cross-terms in the sum all vanish because $\int_0^{2\pi} e^{ikt} dt = 0$ for each $k \in \mathbf{Z}
\backslash \{ 0 \}$.
Evaluating the computation above with the choice $\nu=0$ we find
For each $\theta \in {\mathbf{S}}^{n-1}$ we have $$\label{parseval2}
\|\widehat w (\cdot, \cdot,\theta) \|^2_{L^2([0,T_\epsilon] \times
[0,2\pi])} = 2\pi \| w \|^2_{L^2({\mathbf{R}})}.$$
Take $\nu=0$ in .
Furthermore one can reindex the sum in to obtain $$\begin{aligned}
\label{fourier_laplace_period}
\widehat w(t +T_\epsilon, \xi, \theta) & = & \sum_{k=-\infty}^\infty e^{-ik \xi}
w(t+T_\epsilon + k T_\epsilon, \theta) \\ \nonumber
& = & \sum_{k=-\infty}^\infty e^{-ik\xi} w(t+ (k+1)T_\epsilon,\theta) \\ \nonumber
& = & \sum_{l=-\infty}^\infty e^{-i (l-1)\xi} w(t+lT_\epsilon,\theta) \\ \nonumber
& = & e^{i\xi} \widehat w(t,\xi, \theta) , \end{aligned}$$ which we can write either as $\widehat w(t,\xi,\theta) = e^{-i\xi}
\widehat w(t+T_\epsilon, \xi,\theta)$ or as $w(t+T_\epsilon,\theta) =
\mathcal{F}_\epsilon^{-1} (e^{i\xi}\mathcal{F}_\epsilon(w)) (t,\theta)$. In more geometric/invariant language, this last formula states $\widehat w$ is a section of the flat bundle ${\mathbf{S}}^1 \times {\mathbf{S}}^{n-1}$ with holonomy $\xi$ around the ${\mathbf{S}}^1$ loop.
The Fourier-Laplace transform gives a direct integral decomposition $$\label{integral_decomposition}
L^2 ({\mathbf{R}}\times {\mathbf{S}}^{n-1} ) = \int_{\eta \in [0,2\pi]}^\oplus
L^2_\eta ([0,T_\epsilon] \times {\mathbf{S}}^{n-1}) d\eta,$$ where $L^2_\eta([0,T_\epsilon] \times {\mathbf{S}}^{n-1})$ is the $L^2$-completion of $$\left \{ w \in \mathcal{C}^0 ([0, T_\epsilon] \times {\mathbf{S}}^{n-1}):
w(T_\epsilon, \theta) = e^{i T_\epsilon \eta}
w(0,\theta) \right \}.$$
Combine with and .
Spectral bands of the Jaocbi operator of the Delaunay metrics
-------------------------------------------------------------
At this point we use the decomposition to prove a spectral gap result for the Jacobi operator of a Delaunay metric. Much of this discussion borrows from [@MP_bifur].
We restrict attention in this section to the space of quasi-periodic functions. For each $\eta \in {\mathbf{R}}$ and nonnegative integer $k$ define $W^{k,4}_\eta ([0, T_\epsilon])$ to be the $W^{k,4}$-closure of the space of smooth functions on $[0, T_\epsilon]$ subject to the boundary conditions $$\label{qp_bndry_cond}
\frac{d^l w}{dt^l} (T_\epsilon) = e^{iT_\epsilon \eta}
\frac{d^l w}{dt^l} (0), \qquad l=0,1,\dots, k-1 .$$ and denote by $L_{\epsilon, j, \eta}$ the restriction $$L_{\epsilon, j, \eta} = L_{\epsilon,j} : W^{4,2}_\eta ([0, T_\epsilon])
\rightarrow L^2_\eta ([0,T_\epsilon ]).$$
In order to use the decomposition we define the following twisted operator. To begin we define $\widehat L_\epsilon (\xi)$ by $\widehat L_\epsilon (\xi) (\widehat v) = \widehat {L_\epsilon (v)}$, or $\widehat L_\epsilon (\xi) = \mathcal{F}_\epsilon \circ
L_\epsilon \circ \mathcal{F}_\epsilon^{-1}$. Using we see $$\begin{aligned}
\widehat L_\epsilon (\xi) (e^{i\xi} \widehat w)
(t,\xi,\theta) & = & \widehat L_\epsilon (\xi) (\widehat w)(t+T_\epsilon, \xi,\theta)
= \widehat {L_\epsilon (w)} (t+T_\epsilon,\xi,\theta) \\
& = & e^{i\xi} \widehat {L_\epsilon v} (t,\xi,\theta)
= e^{i\xi} \widehat L_\epsilon (\xi) (\widehat v)(t,\xi,\theta) , \end{aligned}$$ which we can rearrange to read $$e^{-i \xi} \widehat L_\epsilon (\xi) (e^{i \xi} \widehat v)
= \widehat L_\epsilon (\xi) (\widehat v) .$$ This last transformation rule allows us to define the twisted operator $$\label{twisted_op_defn}
\widetilde L_\epsilon (\xi) (\widehat v)
= e^{i\xi t} \mathcal{F}_\epsilon
\circ L_\epsilon \circ \mathcal{F}_\epsilon^{-1} (e^{-i\xi t}
\widehat v) ,$$ which is now a well-defined operator $$\widetilde L_\epsilon (\xi) : W^{k+4,2} ({\mathbf{S}}^1 \times {\mathbf{S}}^{n-1})
\rightarrow W^{k,2} ({\mathbf{S}}^1 \times {\mathbf{S}}^{n-1}),$$ for each value of the paramater $\xi \in {\mathbf{C}}$. Here we identify ${\mathbf{S}}^1
= {\mathbf{R}}/ T_\epsilon \mathbf{Z}$. Our key point here is that $\widetilde L_\epsilon
(\xi)$ act on the [*same*]{} function space for each value of $\xi$.
Observe that $\widehat L_\epsilon$ has the same coordinate expression as $L_\epsilon$. We can again decompose $\widehat L_\epsilon$ and $\widetilde L_\epsilon$ into Fourier components, obtaining $\widehat L_{\epsilon,j}$ and $\widetilde L_{\epsilon,j}$. In particular, by the restriction of $\widehat L_{\epsilon,j}(\eta)$ to the interval $[0,T_\epsilon]$ is exactly the operator $L_{\epsilon,j,\eta}$ defined above.
For each $\epsilon$, $j$, and $\eta$ the operator $L_{\epsilon, j, \eta}$ is a fourth order ordinary differential operator and we denote its eigenvalues by $\sigma_k(\epsilon,j,\eta)$ for $k=0,1,2,\dots$. Furthermore $L_{\epsilon,j, 0}
= L_{\epsilon, j, 2\pi}$ for each $\epsilon$ and $j$, so we may think of $$\sigma_k (\epsilon, j, \cdot) : {\mathbf{S}}^1 \rightarrow {\mathbf{R}}.$$ We denote the image of this eigenvalue maps by $$\label{jth_spec_band_defn}
B_k(\epsilon, j) = \{ \sigma \in {\mathbf{R}}: \sigma = \sigma_k (\epsilon,j,\eta)
\textrm{ for some }\eta \in [0,2\pi/T_\epsilon] \}$$ as the $j$th [**spectral band**]{} of $L_{\epsilon,j}$.
Each band $B_k(\epsilon,j)$ is a nondegenerate interval.
Each $L_{\epsilon,j}$ is a fourth order ordinary differential operator, and so the ODE $L_{\epsilon,j} v = \sigma v$ has a four-dimensional solution space. If the function $\sigma_k(\epsilon,j, \cdot)$ is constant on the interval $[0,2\pi]$ then $L_{\epsilon,j} v = \sigma v$ must have an infinite dimensional solution space for $\sigma \in B_k(\epsilon,j)$. which is impossible. We conclude that no band $B_k (\epsilon,j)$ may collapse to a single point.
The eigenfunction $w$ corresponding to the eigenvalue $\sigma_k(\epsilon,j,\eta)$ satisfies $w(t+ 2\pi/T_\epsilon) = e^{i\eta} w(t) = e^{(2\pi-\eta) i} w(t)$ and so $\bar w(t+2\pi)
= e^{-i\eta} \bar w(t)$. However, the coefficients of the ordinary differential operator $L_{\epsilon,j}$ are real, so $$\sigma_k \left (\epsilon, j , 2\pi - \eta \right ) = \sigma_k (\epsilon,
j, \eta)$$ and we may as well restrict $\sigma$ to the half-circle corresponding to $0 \leq \eta \leq \pi$.
It follows from Floquet theory [@MW] that the band functions $B_{2k}$ are nondecreasing for each $k \in \mathbf{Z}$ while $B_{2k+1}$ are all nonincreasing, so that for each $\epsilon$, $j$ and $k$ we have $$\label{band_ordering1}
\sigma_0(\epsilon,j,0) \leq \sigma_0(\epsilon,j, \pi)
\leq \sigma_1 (\epsilon,j,\pi) \leq \sigma_1 (\epsilon,
j,0) \leq \dots.$$ This in turn implies the bands all have the structure $$\begin{aligned}
\label{band_ordering2}
B_{2k}(\epsilon,j) (\epsilon,j) & = & [\sigma_{2k} (\epsilon,j,0), \sigma_{2k}
(\epsilon, j, \pi)], \\ \nonumber
B_{2j+1} (\epsilon,j) & = & [\sigma_{2k+1} (\epsilon,j, \pi),
\sigma_{2k+1} (\epsilon, j, 0)].\end{aligned}$$
We can related to bands $B_k(\epsilon,0)$ to the bands $B_k(\epsilon,j)$ using the identity $$\label{higher_bands1}
L_{\epsilon,j} = L_{\epsilon,0} - 2\lambda_j \frac{d^2}{dt^2} +
\frac{n(n-4)}{2} \lambda_j + \lambda_j^2.$$ Let $w$ be an eigenvalue of $L_{\epsilon,j, \eta}$, so that implies $$\label{higher_bands2}
\sigma_k(\epsilon,j,\eta) w = L_{\epsilon,j} w = L_{\epsilon,0} w
- 2\lambda_j \ddot w + \frac{n(n-4)}{2} \lambda_j w + \lambda_j^2 w.$$ Writing $w = \sum_{l=0}^\infty \alpha_l w_l$ where $L_{\epsilon,0} w_l = \sigma_l(\epsilon,0,\eta) w_l$ we rewrite as $$\sum_l \alpha_l \sigma_k(\epsilon,j,\eta) w_l =
\sum_l \alpha_l\left ( \sigma_l (\epsilon,0,\eta) w_l
- 2\lambda_j \ddot w_l + \frac{n(n-4)}{2} \lambda_j w_l
+ \lambda_j^2 w_l\right ),$$ which in turn gives us $$\label{higher_bands3}
2\lambda_j \ddot w_l = -\left ( \sigma_k(\epsilon,j,\eta) -
\sigma_l (\epsilon, 0,\eta) - \frac{n(n-4)}{2} \lambda_j -
\lambda_j^2 \right )w_l.$$ This last eigenvalue equation admits quasi-periodic solutions only if $$\label{higher_bands4}
\sigma_k(\epsilon,j,\eta) > \sigma_l (\epsilon,0,\eta) + \frac{n(n-4)}{2}
\lambda_j + \lambda_j^2.$$ We have just proved the following lemma.
For any positive integer $j$ we have the lower bound $$\label{higher_bands5}
\sigma_k(\epsilon,j,0) > \sigma_0 (\epsilon,0,\eta) + \frac{n(n-4)}{2}
\lambda_j + \lambda_j^2 \geq \sigma_0(\epsilon,0,\eta) + \frac{(n-1)}{2}
(n^2-2n-2) .$$
Our main characterization of the spectral bands is the following Proposition.
For each $\epsilon \in (0, \epsilon_n]$ we have $$\label{bottom_bands1}
- \frac{n(n^2-4)}{2} \left ( \frac{1}{T_\epsilon} \int_0
^{T_\epsilon} v_\epsilon^{\frac{2n}{n-4}} dt \right )^{4/n} \leq
\sigma_0(\epsilon,0,0) < 0$$ and $$\label{bottom_bands2}
\textrm{either }\sigma_1(\epsilon,0,0) = 0 \textrm{ or }
\sigma_2(\epsilon,0,0) = 0.$$
Observe that $\dot v_\epsilon$ is a periodic solution of the ODE $L_{\epsilon,0} (\dot v_\epsilon) = 0$, so it must be an eigenfunction with associated eigenvalue $0$, subject to periodic boundary conditions, [*i.e.*]{} $\eta = 0$. This eigenfunction has precisely two modal domains within the interval $[0,
T_\epsilon]$, so it must correspond either to $\sigma_1(\epsilon,0)$ or to $\sigma_2(\epsilon, 0)$. We don’t have enough information at this point to distinguish these two cases.
The function $v_\epsilon$ is also $T_\epsilon$-periodic, and so is an appropriate test function for $\sigma_0(\epsilon,0,0)$. We have $$\begin{aligned}
L_{\epsilon,0} (v_\epsilon) & = & \ddddot{v_\epsilon} - \left ( \frac{n(n-4)+8}{2}
\right ) \ddot v_\epsilon + \frac{n^2(n-4)^2}{16} v_\epsilon -
\frac{n(n+4)(n^2-4)}{16} v^{\frac{n+4}{n-4}} \\
& = & \ddddot v_\epsilon + \left ( \frac{n(n-4)+8}{2} \right ) \ddot v_\epsilon
+ \frac{n^2(n-4)^2}{16} v_\epsilon - \frac{n(n-4)(n^2-4)}{16}
v_\epsilon^{\frac{n+4}{n-4}} - \frac{n(n^2-4)}{2} v_\epsilon^{\frac{n+4}{n-4}} \\
& = & -\frac{n(n^2-4)}{2} v_\epsilon^{\frac{n+4}{n-4}} < 0 ,\end{aligned}$$ and so $$\sigma_0(\epsilon, 0,0) \leq \frac{\int_0^{T_\epsilon} v_\epsilon L_{\epsilon,0}
(v_\epsilon) dt}{\int_0^{T_\epsilon} v_\epsilon^2 dt} = -\frac{n(n^2-4)}{2}
\frac{\int_0^{T_\epsilon} v_\epsilon^{\frac{2n}{n-4}} dt}{\int_0^{T_\epsilon}
v_\epsilon^2 dt} < 0,$$ which gives the upper bound in .
On the other hand, combining the uniqueness theorem of [@FK] and the variational characterization of the Deleaunay solution in Section 5 of [@JX] we we that (up to translations) $v_\epsilon$ is the unique minimizer of the functional $$W^{4,2}_0([0,T_\epsilon]) \ni v \mapsto \frac{\int_0^{T_\epsilon}
\ddot v^2 + \left ( \frac{n(n-4)+8}{2} \right ) \dot v^2 + \frac{n^2(n-4)^2}{16}
v^2 dt }{\left ( \int_0^{T_\epsilon} v^{\frac{2n}{n-4}} dt \right )^{\frac{n-4}{n}}} .$$ By we then have $$\label{bottom_bands3}
\frac{\int_0^{T_\epsilon} \ddot v^2 + \left (\frac{n(n-4)+8}{2}
\right )\dot v^2 + \frac{n^2(n-4)^2}{16} v^2 dt} {\left (
\int_0^{T_\epsilon} v^{\frac{2n}{n-4}} dt \right )^{\frac{n-4}{n}}}
\geq \frac{n(n-4)(n^2-4)}{16} \left ( \int_0^{T_\epsilon}
v_\epsilon^{\frac{2n}{n-4}} dt \right )^{4/n}$$ for each $v \in W^{4,2}_0([0,T_\epsilon])$, which then gives $$\begin{aligned}
\label{bottom_bands4}
\int_0^{T_\epsilon} v L_{\epsilon,0} (v) dt & = & \int_0^{T_\epsilon}
\ddot v^2 + \left ( \frac{n(n-4)+8}{2} \right )\dot v^2 + \frac{n^2(n-4)^2}
{16} v^2 - \frac{n(n+4)(n^2-4)}{16} v^{\frac{2n}{n-4}} dt \\ \nonumber
& \geq & \frac{n(n-4)(n^2-4)}{16} \left ( \int_0^{T_\epsilon} v_\epsilon^{\frac{2n}{n-4}}
dt \right )^{4/n} \left ( \int_0^{T_\epsilon} v^{\frac{2n}{n-4}} dt
\right )^{\frac{n-4}{n}} - \frac{n(n+4)(n^2-4)}{16} \int_0^{T_\epsilon}
v^{\frac{2n}{n-4}} dt \\ \nonumber
& = & \left ( \int_0^{T_\epsilon} v^{\frac{2n}{n-4}} dt \right )^{\frac{n-4}{n}}
\left ( \frac{n(n-4)(n^2-4)}{16} \left ( \int_0^{T_\epsilon} v_\epsilon
^{\frac{2n}{n-4}} dt \right )^{4/n} - \frac{n(n+4)(n^2-4)}{16} \left (
\int_0^{T_\epsilon} v^{\frac{2n}{n-4}} dt \right )^{4/n} \right ) .\end{aligned}$$ Hölder’s inequality with exponents $\frac{n}{n-4}$ and $n/4$ implies $$\int_0^{T_\epsilon} v^2 dt \leq T_{\epsilon}^{4/n} \left ( \int_0^{T_\epsilon}
v^{\frac{2n}{n-4}} dt \right )^{\frac{n-4}{n}},$$ which we combine with to see $$\begin{aligned}
\label{bottom_bands5}
\int_0^{T_\epsilon} v L_{\epsilon,0} (v) dt & \geq &
\frac{n(n-4)(n^2-4)}{16} T_\epsilon^{-4/n} \int_0^{T_\epsilon}
v^2 dt
\left ( \int_0^{T_\epsilon} v_\epsilon
^{\frac{2n}{n-4}} dt \right )^{4/n} \\ \nonumber
&& - \frac{n(n+4)(n^2-4)}{16} T_\epsilon^{-4/n} \int_0^{T_\epsilon}
v^2 dt \left (
\int_0^{T_\epsilon} v^{\frac{2n}{n-4}} dt \right )^{4/n} \end{aligned}$$ for each $v \in W^{4,2}_0([0,T_\epsilon])$. Finally, we are free to choose a scale for our test function $v$, and we normalize so that $\int_0^{T_\epsilon} v^{\frac{2n}{n-4}} dt = \int_0^{T_\epsilon}
v_\epsilon^{\frac{2n}{n-4}} dt$. Using this normalization becomes $$\begin{aligned}
\int_0^{T_\epsilon} v L_{\epsilon,0}(v) dt & \geq & \left (
\frac{n(n-4)(n^2-4)}{16} - \frac{n(n+4)(n^2-4)}{16} \right )
T_\epsilon^{-4/n} \left ( \int_0^{T_\epsilon} v^2 dt \right )
\left ( \int_0^{T_\epsilon}
v_\epsilon^{\frac{2n}{n-4}} dt \right )^{4/n} \\
& = & -\frac{n(n^2-4)}{2} T_\epsilon^{-4/n} \left ( \int_0^{T_\epsilon}
v^2 dt \right ) \left ( \int_0^{T_\epsilon}
v_\epsilon^{\frac{2n}{n-4}} dt \right )^{4/n} ,
\end{aligned}$$ which gives the lower bound in .
We have $B_k(\epsilon,0) \subset (0,\infty)$ for each $k \geq 3$ and $B_k(\epsilon,0) \subset [0,\infty)$ for each $k \geq 2$.
This follows from the previous proposition and .
For each positive integer $j$ we have $B_k(\epsilon,j) \subset
(0,\infty)$.
When $j >n$ we have $\lambda_j \geq 2n$, and in this case gives $$\sigma_k (\epsilon,j,0) > \sigma_0(\epsilon,0,0) + n^3$$ for all $k$. On the other hand, $0<v_\epsilon <1$, so the lower bound in gives us $$\sigma_0 (\epsilon,0,0) \geq -\frac{n(n^2-4)}{2} \left ( \frac{1}{T_\epsilon}
\int_0^{T_\epsilon} v_\epsilon^{\frac{2n}{n-4}} dt \right )^{4/n} \geq
-\frac{n(n^2-4)}{2} ,$$ which implies $$\sigma_k(\epsilon,j,0) > \sigma_0(\epsilon,0,0) + n^3 \geq n^3-\frac{n(n^2-4)}{2}
> 0.$$ and the corollary follows.
The case of $1 \leq j \leq n$ requires some more attention. Luckily, we know from and that $$L_{\epsilon,j} (w_j^\pm) = 0, \qquad
w_j^\pm = e^{\pm t} \left ( \pm \dot v_\epsilon + \frac{n-4}{2}
v_\epsilon \right ) + \mathcal{R}_\pm$$ where $\mathcal{R}_+ = \mathcal{O}(1)$ and $\mathcal{R}_- = \mathcal{O}(e^{-2t})$. By these are positive, periodic solutions, and so must correspond to the bottom of the spectrum of $L_{\epsilon, j, \eta}$.
The following lemma relates the spectral bands $B_k(\epsilon, j)$ and the set of indicial roots $\Gamma_{\epsilon,j}$ of the operator $L_{\epsilon,j}$.
We have $0 \in B_k(\epsilon,j)$ for some $k$ if and only if the ODE $L_{\epsilon,j} w = 0$ admits a quasi-periodic solution.
Let $\widehat v$ satisfy $$0 = L_{\epsilon,j,\eta} \widehat v = e^{i\eta t}\mathcal{F}_\epsilon( L_{\epsilon,j}
(\mathcal{F}_\epsilon^{-1} (e^{-i\eta t} \widehat v)))\Rightarrow
L_{\epsilon,j} (\mathcal{F}_\epsilon^{-1} (e^{-i\eta t} \widehat v)) = 0.$$ In particular, $v = \mathcal{F}_\epsilon^{-1} (e^{-i\eta t} \widehat v)$ satisfies $L_{\epsilon,j} (v) = 0$. In addition, implies $v(t+T_\epsilon,\theta) = e^{i\eta} v(t,\theta)$, and so $v$ must be quasi-periodic.
We summarize the important conclusions of this section with the following Corollary.
\[low\_indicial\_roots\_cor\] The indicial root $0$ is isolated, [*i.e.*]{} there exists $\delta>0$ such that no other indicial roots lie in the interval $(-\delta, \delta)$. Moreover, any Jacobi field with sub-exponential growth ([*i.e.*]{} tempered) must be a linear combination of $w_0^+$ and $w_0^-$, the Jacobi fields generated by translations along the axis and changes of the Delaunay parameter.
Mapping properties of the linearized operator
---------------------------------------------
We have already seen that for each $\delta \in {\mathbf{R}}$ and $k \in \mathbf{N}$ the mapping $$L_\epsilon: W^{k+4,2}_\delta ((0, \infty) \times {\mathbf{S}}^{n-1}) \rightarrow
W^{k,2}_\delta ((0, \infty) \times {\mathbf{S}}^{n-1})$$ is a linear, elliptic operator and it has bounded coefficients.
\[nondegeneracy\] The operator $$L_\epsilon : W^{k+4,2}_\delta ((0,\infty) \times {\mathbf{S}}^{n-1}) \rightarrow
W^{k,2}_\delta ((0,\infty) \times {\mathbf{S}}^{n-1})$$ is Fredholm provided $\delta \not \in \Gamma_\epsilon$, where $\Gamma_\epsilon$ is given in .
Our proof follows that of Proposition 4.8 of [@MPU].
We recall the twisted operator define in depending on a parameter $\xi \in {\mathbf{C}}$, namely $$\widetilde {L}_\epsilon (\xi) : W^{k+4,2} ({\mathbf{S}}^1 \times {\mathbf{S}}^{n-1})
\rightarrow W^{k,2} ({\mathbf{S}}^1 \times {\mathbf{S}}^{n-1}), \qquad
\widetilde{L}_\epsilon (\xi) (\widehat v) = e^{i\xi t} \mathcal{F}_\epsilon
\circ L_\epsilon \circ \mathcal{F}_\epsilon^{-1} (e^{-i\xi t} \widehat v).$$ We then use the analytic Fredholm theorem to prove the twisted operator is Fredholm away from a discrete set of poles in ${\mathbf{C}}$. Unwinding definition, we then translate the Fredholm property of the twisted operator to corresponding properties of $L_\epsilon$ on a weighted function space.
The operator $\widetilde L_{\epsilon}(\xi)$ is linear, bounded, and elliptic for each choice of $\xi$, and depends on $\xi$ holomorphically. Thus, by the analytic Fredholm theorem (see Section 5.3 of [@Mel]) $\widetilde L_{\epsilon}(\xi)$ is either never Fredholm for any value $\xi$ or it is Fredholm for $\xi$ not in a certain discrete set. We choose $\xi = \eta \in (0,2\pi)$ and suppose there exists $\widehat v$ such that $\widetilde L_\epsilon (\eta) (\widehat v) =0$, then $L_\epsilon (v) = 0$ where $v = \mathcal{F}_\epsilon ^{-1}
(e^{-i\eta t} \widehat v)$. Then $v$ is quasi-periodic, in that $v(t+T_\epsilon, \theta) = e^{i\eta } v(t,\theta)$, and so in particular $v$ is bounded. However, by Corollary \[low\_indicial\_roots\_cor\] any bounded Jacobi field must be a multiple of $w_0^+$, which is not quasi-periodic. Thus $\widetilde L_\epsilon (\eta)$ is injective. However, this operator is formally self-adjoint, so it is also surjective, and in particular $\widetilde L_\epsilon (\eta)$ is Fredholm.
Therefore we may safely apply the analytic Fredholm theorem to conclude there exists a discrete set $\mathcal{P} \subset {\mathbf{C}}$ and a meromorphic operator $$\widetilde G_\epsilon (\xi) : W^{k,2} ({\mathbf{S}}^1 \times {\mathbf{S}}^{n-1})
\rightarrow W^{k+4,2} ({\mathbf{S}}^1 \times {\mathbf{S}}^{n-1})$$ such that $\widehat v = \widetilde G_\epsilon (\xi) \circ \widetilde
L_\epsilon (\xi)(\widehat v)$ so long as $\xi \not \in \mathcal{P}$. Moreover, by our construction $$\Gamma_\epsilon = \{ \nu \in {\mathbf{R}}: \nu = \Im (\xi) \textrm{ for
some } \xi \in \mathcal{P} \}.$$
We unravel this relation to find the Greens operator $G_\epsilon$. Again we let $v = \mathcal{F}_\epsilon^{-1}(e^{-i\xi t} \widehat v)$, so that $$\begin{aligned}
e^{i\xi t} \mathcal{F}_\epsilon (v) & = & \widehat v
= \widetilde G_\epsilon (\widetilde L_\epsilon (\widehat v)) \\
& = & \widetilde G_\epsilon (e^{i\xi t} \mathcal{F}_\epsilon
(L_\epsilon (\mathcal{F}_\epsilon^{-1} ( e^{-i\xi t}
e^{i \xi t} \mathcal{F}_\epsilon (v))))) \\
& = & \widetilde G_\epsilon ( e^{i \xi t} \mathcal{F}_\epsilon
(L_\epsilon (v))). \end{aligned}$$ We thus conclude $$\label{greens_op}
G_\epsilon (\phi) = \mathcal{F}_\epsilon^{-1} (e^{-i\xi T_\epsilon t}
(\widetilde G_\epsilon (e^{i\xi T_\epsilon t} (\mathcal{F}_\epsilon (\phi)))).$$ By our construction we have $v = G_\epsilon (\phi) \in W^{k+4,2}_{-\Im(\xi)}
((0,\infty) \times {\mathbf{S}}^{n-1})$, so that $$\delta = \Im(\xi) \not \in \Gamma_\epsilon \Rightarrow
\textrm{ there exist a Greens operator } G_\epsilon :
W^{k,2}_\delta ((0,\infty) \times {\mathbf{S}}^{n-1}) \rightarrow W^{k+4,2}_\delta
((0,\infty) \times {\mathbf{S}}^{n-1}),$$ and by the Fredholm alternative we have completed our proof.
The set of indicial roots $\Gamma_\epsilon$ does not have any accumulation points.
Recall that each $\gamma \in \Gamma_\epsilon$ is the imaginary part of a pole of $\widetilde G_\epsilon$, which forms a discrete set in ${\mathbf{C}}$. Furthermore, the operator $\widetilde L_\epsilon(\xi)$ is unitarily equivalent to $\widetilde L_\epsilon (\xi + 2\pi k)$ for each $k \in \mathbf{Z}$, and so $\xi$ is a pole of $\widetilde G_\epsilon$ if and only if $\xi + 2\pi k$ is as well for each $k \in \mathbf{Z}$. Thus $\widetilde
G_\epsilon$ can only have finitely many poles in each horizontal strip.
For each $\delta \in (0,1)$ the operator $$L_\epsilon : W^{k+4,2}_{-\delta} ((0,\infty) \times {\mathbf{S}}^{n-1}) \rightarrow
W^{k,2}_{-\delta} ((0,\infty) \times {\mathbf{S}}^{n-1})$$ is injective and $$L_\epsilon : W^{k+4,2}_\delta ((0,\infty) \times {\mathbf{S}}^{n-1}) \rightarrow
W^{k,2}_\delta ((0,\infty) \times {\mathbf{S}}^{n-1})$$ is surjective.
In the proof of Proposition \[nondegeneracy\] we showed that $\widetilde L_\epsilon (\xi)$ is injective for each $\xi$ with $-1 < \Im(\xi) < 0$,which in turn implies $L_\epsilon : W^{k+4,2}_{-\delta} ((0,\infty) \times {\mathbf{S}}^{n-1})
\rightarrow W^{k,2}_{-\delta} ((0,\infty) \times {\mathbf{S}}^{n-1})$ is injective. The surjectivity statement now follows from the injectivity statement and duality because $L_\epsilon:
W^{k+4,2}_0 ((0,\infty) \times {\mathbf{S}}^{n-1}) \rightarrow W^{k,2}_0
((0,\infty) \times {\mathbf{S}}^{n-1})$ is formally self-adjoint.
We can extract more information from the Fourier-Laplace transform by examining how it behaves when the contour we are using to define its inverse crosses a pole of $\widetilde G_\epsilon$. We use Proposition 4.14 of [@MPU] as a model for the following.
\[lin\_soln\_asymp\_expansion\] Let $\phi \in \mathcal{C}^\infty_0 ((0,\infty) \times {\mathbf{S}}^{n-1})$ and let $\delta \in (0,1)$ and $v \in W^{4,2}_{-\delta} ((0,\infty) \times {\mathbf{S}}^{n-1})$ satisfy $L_\epsilon (v) = \phi$. Then $v$ has an asymptotic expansion $v = \sum_{k=1}^\infty v_j$ with $L_\epsilon (v_j) = 0$ and $v_j \in W^{4,2}_{-\nu}$ for any $\nu < \gamma_{\epsilon,j}$.
We begin by transforming the equation. Choose $\xi \in {\mathbf{C}}$ with $\Im (\xi) > \delta>0$ and let $$\widetilde v = e^{i\xi t} \widehat v, \qquad
\widetilde \phi = e^{i \xi t} \widehat \phi,$$ so that $\widetilde L_\epsilon (\xi) (\widetilde v) =
\widetilde \phi$. Applying the Greens operator $\widetilde G_\epsilon (\xi)$ we have $\widetilde v =
\widetilde G_\epsilon (\xi)(\widetilde \phi)$. That $\phi \in \mathcal{C}_0^\infty ((0,\infty) \times {\mathbf{S}}^{n-1})$ implies $\widetilde \phi$ is entire in the $\xi$ variable and smooth in the $(t,\theta)$ variables. On the other hand, in the half-plane $\Im (\xi) > \delta$ the poles of $\widetilde G_\epsilon$ occur at the same points as the zeroes of $\widetilde \phi$ (with the same degrees), so $\widetilde v$ is analytic in this half-plane. In fact, because $\gamma_{\epsilon,1} = 1$, for any $\delta' \in (\delta, 1)$ the operator $\widetilde G_\epsilon$ has no poles in the strip $\delta' < \Im (\xi) < \delta$, we can shift the contour integral defining $\mathcal{F}_\epsilon^{-1}$ up to $\Im(\xi) = \delta'$ for any $\delta' < 1$, and so we may take $v \in W^{k+4,2}_{-\nu}((0,\infty) \times {\mathbf{S}}^{n-1})$ for any $\nu <
1$.
\[h\]
\(0) at (2,2); (0,3) – (8,3) coordinate\[label = [below:$x$]{}\] (xmax); (4,-2) – (4,4) coordinate\[label = [right:$y$]{}\] (ymax); (2,0.5) – (4,0.5); (4, 0.5) – (6, 0.5); (2,1.5) – (4,1.5); (4,1.5) – (6,1.5); (6,0.5) – (6, 1.0); (6,1.0) – (6,1.5); (2,0.5) – (2,1.0); ( 2,1.0) – (2,1.5); at (4,1) [$*$]{}; at (5.6,-0.8) [pole at $-i$]{}; (5.5,-0.5) – (4.05,1);
We complete the proof by shifting the contour integral across a pole of $\widetilde G_\epsilon$. We sketch this contour in Figure \[contour\_fig\]. Choose $\delta'' \in
(1, \gamma_{\epsilon,2})$ and $\xi''$ such that $\Im(\xi'') = \delta''$, and let $\widetilde v'' =
\widetilde G_\epsilon (\xi'')$. Applying $\mathcal{F}_\epsilon^{-1}$ along the two contours $\Im(\xi) = \delta$ and $\Im(\xi) = \delta''$ we thus have $v = v' + v''$, where $v'' \in W^{k+4,2}_{-\delta''}
((0,\infty) \times {\mathbf{S}}^{n-1})$ and $L_\epsilon (v') = 0$ and $v' \in W^{k+4,2}_{-\nu}((0,\infty) \times {\mathbf{S}}^{n-1})$ for each $\nu \in (0,1)$. In fact the difference $\widetilde v - \widetilde v'' = (\widetilde G_\epsilon (\xi) -
\widetilde G_\epsilon (\xi'')) (\widetilde \phi)$ is the residue of a meromorphic function with a pole at height $\Im(\xi) = -1$ about a rectangular contour of width $2\pi$ and height $\delta'' - \delta$. We do not see the result of the contour integral along the vertical sides of the rectangle because all our transformed functions are periodic in the real direction with period $2\pi$.
The following is a special case of Proposition \[lin\_soln\_asymp\_expansion\], corresponding to moving the contour integral across only the first pole.
\[lin\_decomp\_lemma\] Let $\delta \in (0,1)$, let $\phi \in \mathcal{C}^\infty ((0, \infty)
\times {\mathbf{S}}^{n-1}) \cap L^2_{-\delta} ((0,\infty) \times {\mathbf{S}}^{n-1})$ and let $v \in W^{4,2}_\delta ((0,\infty) \times {\mathbf{S}}^{n-1})$ such that $L_\epsilon (v) = \phi$. Then there exist $w \in W^{4,2}_{-\delta}
((0,\infty) \times {\mathbf{S}}^{n-1})$ and $ \psi \in \operatorname{Span}
\{ w_0^+, w_0^-\}$ such that $v = w+\psi$.
Simple convergence to a radial solution {#asymp_sec}
=======================================
In this section we prove an asymptotic estimate.
Let $v \in \mathcal{C}^\infty ((0,\infty) \times
{\mathbf{S}}^{n-1})$ be a positive solution of which also satisfies . Then either $\limsup_{t \rightarrow \infty} v(t,\theta) = 0$ or there exists a Delaunay parameter $\epsilon \in (0,v_{cyl}]$ and translation parameter $T \in [0,T_\epsilon )$ and positive constants $C$ and $\alpha$ such that $$\label{asymp_symm1}
|v(t,\theta) - v_\epsilon (t+T)| \leq C e^{-\alpha t} .$$
Strictly speaking, the proof below is not entirely necessary, as one should be able to show the estimate is equivalent to the simple asymptotics Jin and Xiong derive in [@JX]. However, we believe the proof is different and interesting enough to include.
The transformed function $v$ satisfies . We let $\tau_k \rightarrow \infty$ and define $$v_k: (-\tau_k, \infty) \times {\mathbf{S}}^{n-1} \rightarrow \infty, \qquad
v_k(t,\theta) = v(t+ \tau_k, \theta).$$ By the bounds there are constants $0<c_1< c_2$ (which depend on the solutions $v$) such that $c_1 < v_k(t,\theta) < c_2$ for all $k$. Moreover, elliptic estimates imply the sequence $\{ |\nabla v_k |\}$ is also uniformly bounded, so in fact the family $\{ v_k\}$ is also equicontinuous. Thus a subsequence converges uniformly on compact sets to a solution $\bar v =
\lim_{l \rightarrow \infty} v_{k_l} \rightarrow \bar v$. Since $\tau_k \rightarrow \infty$ we now have a global solution $\bar v : {\mathbf{R}}\times {\mathbf{S}}^{n-1} \rightarrow (0,\infty)$ of the PDE . By the classification theorem in [@FK] we must have $\bar v(t,\theta) = v_\epsilon (t+T)$ for some $\epsilon
\in (0,\epsilon_n]$ and $t \in [0,T_\epsilon)$.
It remains to show that $\epsilon$ and $T$ do not depend on the choice of sequence $\tau_k \rightarrow \infty$ or the choice of subsequence $v_{k_l}$.
By for each $k$ we have $$\begin{aligned}
n \omega_n \mathcal{H}_{\textrm{rad}}(v_\epsilon)
& = & \lim_{k \rightarrow \infty} \mathcal{H}_{rad} (v_k) \\
& = & \lim_{k \rightarrow \infty} \int_{\{ 1 \} \times {\mathbf{S}}^{n-1}} - \frac{\partial v_k}{\partial t}
\frac{\partial^3 v_k}{\partial t^3}+ \frac{1}{2} \left ( \frac{\partial^2 v_k}
{\partial t^2} \right )^2 + \left ( \frac{n(n-4) + 8}{4} \right ) \left ( \frac
{\partial v_k}{\partial t} \right )^2 \\
&& \quad + \frac{n^2(n-4)^2}{32} v_k^2 + \frac{(n-4)^2(n^2-4)}{32}
v_k^{\frac{2n}{n-4}} - \frac{1}{2} (\Delta_\theta v_k)^2- \frac{n(n-4)}{4}
\left | \nabla_\theta \frac{\partial v_k}{\partial t} \right |^2 d\sigma \\
& = & \lim_{k \rightarrow \infty} \int_{\{ \tau_k \} \times {\mathbf{S}}^{n-1}} - \frac{\partial v}{\partial t}
\frac{\partial^3 v}{\partial t^3}+ \frac{1}{2} \left ( \frac{\partial^2 v}
{\partial t^2} \right )^2 + \left ( \frac{n(n-4) + 8}{4} \right ) \left ( \frac
{\partial v}{\partial t} \right )^2 \\
&& \quad + \frac{n^2(n-4)^2}{32} v^2 + \frac{(n-4)^2(n^2-4)}{32}
v^{\frac{2n}{n-4}} - \frac{1}{2} (\Delta_\theta v)^2- \frac{n(n-4)}{4}
\left | \nabla_\theta \frac{\partial v}{\partial t} \right |^2 d\sigma \\
& = & \int_{\{ 1 \} \times {\mathbf{S}}^{n-1}} - \frac{\partial v}{\partial t}
\frac{\partial^3 v}{\partial t^3}+ \frac{1}{2} \left ( \frac{\partial^2 v}
{\partial t^2} \right )^2 + \left ( \frac{n(n-4) + 8}{4} \right ) \left ( \frac
{\partial v}{\partial t} \right )^2 \\
&& \quad + \frac{n^2(n-4)^2}{32} v^2 + \frac{(n-4)^2(n^2-4)}{32}
v^{\frac{2n}{n-4}} - \frac{1}{2} (\Delta_\theta v)^2- \frac{n(n-4)}{4}
\left | \nabla_\theta \frac{\partial v}{\partial t} \right |^2 d\sigma \\
& = & \mathcal{H}_{rad} (v), \end{aligned}$$ so $\epsilon$ does not depend on any choices.
We complete the proof with the help of the following lemmas.
\[helper\_lemma1\] Let $\theta \in {\mathbf{S}}^{n-1}$ and let $v: (0,\infty)
\times {\mathbf{S}}^{n-1} \rightarrow (0,\infty)$ solve . Then $$\label{ang_decay1}
\lim_{t\rightarrow\infty} \frac{\partial v}{\partial \theta} (t,\cdot) = 0$$ uniformly.
If this were not the case then there exist $(\tau_j,
\theta_j)\in (0,\infty) \times {\mathbf{S}}^{n-1}$ and $C>0$ such that $\tau_j \rightarrow \infty$ and $$\left | \frac{\partial v}{\partial \theta} (\tau_j, \theta_j) \right |
\geq C > 0.$$ Now translate by $-\tau_j$ and rotate by $-\theta_j$ to obtain a new sequence of solutions $v_j$, for which $$\left | \frac{\partial v_j}{\partial \theta} (0,\theta_0) \right | \geq C
> 0.$$ However, by the reasoning above the sequence $v_j$ must converge to a Delaunay solution, which does not depend on $\theta$ at all, contradicting the lower bound displayed above.
\[helper\_lemma2\] Let $\theta \in {\mathbf{S}}^{n-1}$ and let $v$ solve . The Jacobi field $\phi = \partial_\theta v$. decays exponentially in $t$.
Let $\tau_j \rightarrow \infty$ and let $$v_j (t,\theta) = v(t+\tau_j, \theta), \quad A_j = \sup \left \{ \left |
\frac{\partial v_j}{\partial \theta} \right | : t>0 \right \}.$$ Further suppose there exist $(s_j, \theta_j)$ such that $|\partial_\theta
v_j (s_j,\theta_j)| = |\partial_\theta v(s_j + \tau_j,\theta_j)| = A_j$. If the sequence $\{ s_j\}$ is unbounded then we can translate further to obtain $\phi_j (t,\theta) = \partial_\theta v_j(t+s_j, \theta) =
\partial_\theta v(t+s_j+\tau_j,\theta)$. Letting $j \rightarrow \infty$ we obtain $\phi: {\mathbf{R}}\times {\mathbf{S}}^{n-1} \rightarrow {\mathbf{R}}$, which solves the linearized equation about $v_\epsilon (\cdot + T)$. By our construction $\phi$ is not identically zero and bounded. However, by our construction $\phi \not \in \operatorname{Span}\{ w_0^+ =
\dot v_\epsilon, w_0^- = \frac{d}{d\epsilon} v_\epsilon \}$, so by Corollary \[low\_indicial\_roots\_cor\] $\phi$ must grow exponentially either as $t \rightarrow \infty$ or $t \rightarrow -\infty$. We have already shown $\phi(t,\theta)$ must decay at some rate as $t\rightarrow \infty$, so there must exist $c>0$ such that $|\phi(t,\theta)| \leq c e^{- t}$.
A consequence of Lemma \[helper\_lemma2\] is that any choice of $s_j>0$ such that $$v_j(s_j, \theta_j) = A_j = \sup \left \{ \left | \frac{\partial v_j}
{\partial \theta} \right | : t>0\right \}$$ remains bounded. In particular, there exists a positive integer $N$ such that $N T_\epsilon > \sup_{j\in \mathbf{N}} s_j$, where $T_\epsilon$ is the period of $v_\epsilon$. We now define the intervals $I_N = [0,NT_\epsilon]$ and $J_N = [NT_\epsilon, 2NT_\epsilon]$, and observe that the limit Jacobi field $\phi$ we have just constructed is bounded for $t>0$ and attains its supremum in $I_N$.
\[helper\_lemma3\] Let $\phi$ be the Jacobi field constructed in Lemma \[helper\_lemma2\]. Then there exists $c$ independent of all choices such that $|\phi(t,\theta)|
\leq c e^{- t}$.
Expand the Jacobi field $\phi$ we have just constructed in Fourier modes. Let $\{ w_j^{\pm}, \widetilde{w}_j^{\pm} \}$ span the solutions space of and let $$E' = \operatorname{Span} \{ w_j^{\pm}, \widetilde{w}_j^{\pm}:
j=0,1,\dots, n\}, \qquad E'' = \operatorname{Span} \{ w_j^{\pm} ,
\widetilde{w}_j^{\pm} : j \geq n+1 \},$$ and write $$\phi = \phi' + \phi'' , \qquad \phi' \in E', \qquad \phi'' \in
E''.$$ We claim that there exists $c>0$ independent of any choices we have made such that $|\phi(t,\theta)| \leq c e^{- t}$. We write $$\phi' = \sum_{j=0}^n
(c_j^\pm w_j^\pm + \tilde c_j \widetilde{w}_j^\pm), \qquad
\phi'' = \sum_{j=n+1}^\infty (c_j^\pm w_j^\pm + \widetilde{c}_j^\pm
\widetilde{w}_j^\pm ).$$ By construction $\phi$ decays at some rate as $t \rightarrow \infty$, so $c_0^\pm = 0$ and $\widetilde{c}_0^\pm = 0$. Also by construction $|\phi(t,\theta)| \leq 1$ for $t\geq 0$, which implies $|c_j^\pm|$ and $|\widetilde{c}_j^\pm|$ are all bounded independent of all choices for $j=1,\dots, n$. We conclude $|\phi'(t,\theta)| \leq c' e^{- t}$ for some $c'>0$ which is independent of all choices. We complete the proof of the claim by showing $|\phi''| \leq c'' e^{- t}$ for some $c''>0$ not depending on any choices. Suppose otherwise, then there would exist a sequence $\phi_i'' \in E''$ such that $$A_i =\sup_{t>0} e^{ t} |\phi_i'' (t,\theta)|
\rightarrow \infty, \qquad \phi_i''(t_i,\theta_i) =
e^{- t_i} A_i.$$ Define $$\widetilde {\phi}_i : [-t_i,\infty) \times {\mathbf{S}}^{n-1} \rightarrow {\mathbf{R}}, \qquad
\widetilde{\phi}_i(t,\theta) = \frac{e^{ t_i}}{A_i}
\phi_i''(t+t_i,\theta) \in E'' .$$ The PDE is uniformly elliptic on any bounded cylinder, so the sequence $\{ t_i\}$ cannot be bounded. Thus $t_i
\rightarrow \infty$ and (after passing to a subsequence) we obtain a limit a $\widetilde \phi \in E''$ on the whole cylinder ${\mathbf{R}}\times {\mathbf{S}}^{n-1}$. Moreover, for $t \geq -t_i$ we have $$\widetilde{\phi}_i(t,\theta) = \frac{e^{ t_i}}{A_i}
\phi_i''(t+t_i,\theta) \leq \frac{e^{t_i}}{A_i}
e^{- (t+t_i)} A_i = e^{- t},$$ so that in the limit we obtain a global solution $\widetilde{\phi} \in E''$ on the whole cylinder ${\mathbf{R}}\times {\mathbf{S}}^{n-1}$ which satisfies the bound $|\widetilde \phi(t,\theta)| \leq e^{- t}$. However, this contradicts the definition of $E''$, as all functions in this space must grow as fast as $e^{\gamma_{\epsilon,2}|t|}$ either as $t
\rightarrow \infty$ or $t \rightarrow -\infty$.
\[helper\_lemma4\] Let $v: (0,\infty) \times {\mathbf{S}}^{n-1} \rightarrow (0,\infty)$ solve and let $\tau_j$ be a sequence such that $\tau_j \rightarrow \infty$. Define $v_j(t,\theta) =
v(t+\tau_j, \theta)$ and $$w_j (t,\theta) = v(t+\tau_j,\theta) - v_\epsilon(t+T), \qquad
\alpha_j = \sup_{0\leq t \leq N T_\epsilon} |w_j(t,\theta)|, \qquad
\phi_j = \frac{1}{\alpha_j} w_j .$$ Here $v_\epsilon (\cdot + T)$ is the limit extracted from the sequence $\{ v_j\}$ using [*a priori*]{} upper and lower bounds. The sequence $\{ \phi_j\}$ converges to a Jacobi field $\phi$ for $v_\epsilon ( \cdot + T)$. Moreover, $\phi$ is bounded for $t\geq 0$.
Let $\Delta_{\textrm{cyl}}^2$ denote $\Delta^2$ written in cylindrical coordinates, [*i.e.*]{} $$\label{cyl_bilap}
\Delta_{\textrm{cyl}}^2 = \frac{\partial^4}{\partial t^4} - \left (
\frac{n(n-4)+8}{2} \right ) \frac{\partial^2}{\partial t^2} +
\frac{n^2(n-4)^2}{16} + \Delta_\theta^2 + 2\Delta_\theta
\frac{\partial^2}{\partial t^2} - \frac{n(n-4)}{2} \Delta_\theta.$$ Then $$\begin{aligned}
\Delta_{\textrm{cyl}}^2 w_j & =& \frac{n(n-4)(n^2-4)}{16}
\left ( v_j^{\frac{n+4}{n-4}} - v_\epsilon ^{\frac{n+4}{n-4}} \right ) \\
& = & \frac{n(n-4)(n^2-4)}{16} \left ( \frac{n+4}{n-4}
\right ) v_\epsilon^{\frac{8}{n-4}} w_j + \mathcal{O}
(\| v_j - v_\epsilon \|^2) \end{aligned}$$ which proves the sequence $\{ w_j\}$ converges to a Jacobi field. It then follows that $\{ \phi_j\}$ also converges to a Jacobi field $\phi$.
Next we prove $\phi$ is bounded for $t \geq 0$. Decompose $\phi$ into Fourier modes, writing $$\phi = \phi'+ \phi'' = a_+w_{\epsilon}^+ + a_-
w_0^- + \phi'',$$ where $\phi''$ is the sum of all the nonzero Fourier modes. We may further decompose $\phi'' = \phi''_+ + \phi''_-$, where $\phi''_+$ is the sum of all the Fourier modes which grow exponentially as $t \rightarrow + \infty$ and $\phi''_-$ is the sum of Fourier modes which decay exponentially as $t \rightarrow +\infty$. Observe that $\phi''$ is bounded if and only if $\phi''_+ = 0$. Now fix $\theta \in
{\mathbf{S}}^{n-1}$ and observe $\partial_\theta \phi''_+$ grows exponentially, at least as fast as $e^{\gamma_{\epsilon, 1} t}$, while $\partial \phi''_-$ decays exponentially, at least as fast as $e^{- t}$. Thus $\phi''$ is bounded in the half-cylinder $t\geq 0$ if and only if $\partial_\theta \phi''$ is bounded.
We can write a similar Fourier decomposition for $\phi_j = \phi_j ' +
\phi_j ''$ for each $j$. Observe that each $\phi_j '$ does not depend at all on $\theta$, so that $$\partial_\theta \phi_j'' = \partial _\theta \phi_j = \frac{1}{\alpha_j}
\partial_\theta w_j = \frac{1}{\alpha_j} \partial_\theta v_j.$$ We have just shown in Lemma \[helper\_lemma3\] that $\partial_\theta v_j$ converges to a Jacobi field $\psi$ such that $|\psi (t,\theta)|\leq c e^{-t}$ for $t>0$, where $c$ does not depend on any choices. Now let $$A_j = \sup \{ |\partial_\theta v_j(t,\theta)| = |\partial_\theta
v(t+\tau_j,\theta)| : 0 \leq t \leq N T_\epsilon \}.$$ If there exist $C_1>0$ and $C_2>0$ such that $$C_1 A_j \leq \alpha_j \leq C_2 A_j,$$ [*i.e.*]{} if $A_j$ and $\alpha_j$ are commensurate, then $\partial_\theta \phi'' = \partial_\theta \phi$ is commensurate with the Jacobi field $\psi$, and hence decays exponentially. If no uniform upper bound $C_2$ exists, then the sequence $$\left \{ \frac{1}{\alpha_j} \partial_\theta v_j \right \}$$ is unbounded and cannot converge, contradicting the convergence $\phi_j \rightarrow \phi$. Additionally, if no lower bound $C_1$ exists then $\phi_j \rightarrow 0$ uniformly on $[0,N T_\epsilon ] \times
{\mathbf{S}}^{n-1}$, which is also a contradictions. Thus $\{A_j\}$ and $\{ \alpha_j\}$ must be commensurate and so $\phi''$ must decay exponentially.
Lastly we show $a_- = 0$, completing the proof that $\phi$ is bounded for $t \geq 0$. We have $$v_j (t,\theta) = v_\epsilon (t+T) + \alpha_j \phi + o(\alpha_j),$$ which implies $$\label{changing_poho}
\mathcal{H}_{\textrm{rad}} (v_j) = \mathcal{H}_{\textrm{rad}}
(v_\epsilon) + a_-\frac{d}{d\epsilon} \mathcal{H}_{\textrm{rad}}
(v_\epsilon) + o(\alpha_j).$$ However, we have already shown $\mathcal{H}_{\textrm{rad}}
(v_j) = \mathcal{H}_{\textrm{rad}} (v_\epsilon)$ for each $j$, so is only possible if $a_- = 0$.
We have now shown $\phi = a_+ w_0^+ + \phi''$, where $\phi''$ decays exponentially. However $w_0^+$ is periodic and hence bounded, and so $\phi$ is bounded as well.
\[helper\_lemma5\] Let $v: (0,\infty) \times {\mathbf{S}}^{n-1} \rightarrow (0,\infty)$ satisfy and , let $\tau \geq 0$, and let $B>0$. Let $$w_\tau(t,\theta) = v_\tau(t,\theta) - v_\epsilon (t+T) = v(t+\tau,\theta)
- v_\epsilon(t+T)$$ and for $N \in \mathbf{N}$ let $$\eta(\tau) = \sup_{0\leq t \leq NT_\epsilon} |w_\tau(t,\theta)|.$$ If $\tau$ is sufficiently large and $\eta(\tau)$ is sufficiently small then there exists $s$ such that $|s| \leq B \eta(\tau)$ such that $$\label{pre_exp_decay1}
\eta (\tau + N T_\epsilon + s) \leq \frac{1}{2} \eta (\tau).$$
Fix $B >0$ and $N \in \mathbf{N}$. If does not hold then there must exist $\tau_j \rightarrow \infty$ such that $\eta_j
= \eta (\tau_j) \rightarrow 0$ but for any $s$ such that $|s| \leq B\eta_j$ it holds $\eta (\tau_j + N T_\epsilon + s) \geq \frac{1}{2} \eta_j$. Let $\phi_j
= \frac{1}{\eta_j} w_{\tau_j}$. By Lemma \[helper\_lemma4\] the sequence $\{ \phi_j\}$ converges uniformly on compact subsets to a Jacobi field $\phi$. Moreover, $\phi$ is bounded for $t \geq 0$. By hypothesis $$0 \leq t \leq N T_\epsilon \Rightarrow |\phi(t,\theta)| \geq \frac{1}{2}$$ so $\phi$ is not identically zero. In the half-cylinder $(0,\infty) \times {\mathbf{S}}^{n-1}$ we expand $\phi = a w_0^+ + \widetilde \phi$ where $\widetilde \phi$ is a sum of Fourier modes all of which decay at least like $e^{-\gamma_{\epsilon, 1} t}$. The coefficients of the other Fourier modes must all be zero because $\phi$ is bounded for $t \geq 0$.
Next we adjust $v_\epsilon(\cdot + T)$ by the translation parameter $s_j = - \eta_j a$. Without loss of generality $B > |a|$, so that $|s_j| \leq
B \eta_j$. Thus $$w_{\tau_j + s_j} (t,\theta) = v(t+\tau_j - \eta_j a, \theta) - v_\epsilon(t+T)
= w_{\tau_j}(t,\theta) - a \eta_j w_0^+(t) + o(\eta_j),$$ which implies $$w_{\tau_j+ s_j} = \eta_j \widetilde \phi + o(\eta_j).$$ The function $\widetilde \phi$ decays at least like $e^{- t}$ and $\sup_{0 \leq t \leq NT_\epsilon} |\widetilde \phi(t,\theta)| \leq 1$, so we may choose $N$ sufficiently large such that $$\begin{aligned}
\sup_{0 \leq t \leq NT_\epsilon} |w_{\tau_j + s_j + N T_\epsilon} (t,\theta)|
& = & \sup_{N T_\epsilon \leq t \leq 2N T_\epsilon} |w_{\tau_j + s_j}
(t,\theta)| \\
& = & \eta_j \sup_{N T_\epsilon \leq t \leq 2N T_\epsilon} |\widetilde
\phi(t,\theta) + o(\eta_j) \\
& \leq & \frac{1}{4} \eta_j. \end{aligned}$$ However, this contradicts the hypothesis $\eta(\tau_j + N T_\epsilon + s)
> \frac{1}{2} \eta(\tau_j)$, completing the proof.
We complete our proof of by showing we can choose $\sigma$ such that $w_\sigma \rightarrow 0$. Again we choose $B> 0$ and $N \in \mathbf{N}$. Letting $t> 0$ be sufficiently large we may assume $B \eta(0) \leq \frac{1}{2} N T_\epsilon$. Let $\tau_0 = 0$ and choose $s_0$ such that holds.
Next we choose the sequences $\tau_j$, $s_j$, and $\sigma_j$ by $$\tau_j = \tau_{j-1} + s_{j-1} + N T_\epsilon, \qquad
\sigma_j = \sum_{i=0}^{j-1} s_i,$$ where $s_j$ satisfies with the choice $\tau = \tau_j$. Iterating we see $$\eta(\tau_j) \leq 2^{-j} \eta(0) \Rightarrow |s_j| \leq 2^{-j-1} N T_\epsilon
\Rightarrow \sigma = \lim_{j \rightarrow \infty} \sigma_j =
\sum_{i=0}^\infty s_j \leq N T_\epsilon.$$ Finally we must show that $\sigma$ is indeed the correct translation parameter. For any $t > 0$ write $t = j N T_\epsilon + [t]$ where $0 \leq [t]< N T_\epsilon$. Then $$\begin{aligned}
w_\sigma (t,\theta) & = & v(t+\sigma, \theta) - v_\epsilon (t+T) \\
& = & v(t+\sigma_j, \theta) - v_\epsilon (t+T) + v(t+\sigma, \theta) -
v(t+\sigma_j, \theta) \\
& = & w_{\tau_j} ([t],\theta) + \mathcal{O}(2^{-j}). \end{aligned}$$ Our bound on $\eta(\tau_j)$ then implies $$v(t+\sigma, \theta) - v_\epsilon (t+T) =
w_\sigma(t,\theta) \leq C 2^{-j},$$ which is exactly the exponential decay we claimed.
Refined asymptotics {#refined_sec}
===================
We finally derive a refined asymptotic expansion of solutions of with an isolated singularity, essentially writing out the next term in the Taylor expansion.
Let $v:(0,\infty) \times {\mathbf{S}}^{n-1} \rightarrow (0,\infty)$ be a smooth solution of such that the associated metric $g_{ij} = v^{\frac{4}{n-4}} (dt^2 + d\theta^2)$ has positive scalar curvature. Then either $\limsup_{t \rightarrow \infty} v(t,\theta)
= 0$, in which case $v$ is a spherical solution, or there exist $\epsilon
\in [\epsilon_n,1)$, $T\in [0,T_\epsilon)$, $a \in {\mathbf{R}}^n$, $C>0$ and $\beta>1$ such that $$\label{refined_asymp_estimate}
|v(t,\theta) - v_{\epsilon, a} (t+T, \theta)| \leq C e^{-\beta t}.$$
By there exist $\epsilon$, $T$, $C_1$, and $\alpha$ such that $$|v(t,\theta) - v_\epsilon (t+T) | \leq C_1 e^{-\alpha t}.$$ In other words $$w(t,\theta) = v(t,\theta) - v_\epsilon (t+T) \in
\mathcal{C}^\infty_{-\alpha} ((0,\infty) \times {\mathbf{S}}^{n-1}).$$ However, $$\begin{aligned}
L_\epsilon (w) & = & \Delta^2_{\textrm{cyl}} (v - v_\epsilon (\cdot + T))
- \frac{n(n+4)(n^2-4)}{16} v_\epsilon^{\frac{8}{n-4}} w \\ \nonumber
& = & \frac{n(n-4)(n^2-4)}{16} \left ( v^{\frac{n+4}{n-4}} -
v_\epsilon^{\frac{n+4}{n-4}} \right )
- \frac{n(n+4)(n^2-4)}{16} v_\epsilon^{\frac{8}{n-4}} w \\ \nonumber
& = & \mathcal{Q}_{\textrm{cyl}} (w) \in \mathcal{C}^\infty_{-2\alpha}
((0,\infty) \times {\mathbf{S}}^{n-1})\end{aligned}$$ where $\Delta^2_{\textrm{cyl}}$ is given in . We combine this with Proposition \[lin\_soln\_asymp\_expansion\] to see $w \in \mathcal{C}^\infty_{-2\alpha} ((0,\infty) \times {\mathbf{S}}^{n-1})$. We iterate this finitely many times to obtain $w \in \mathcal{C}^\infty_{-\delta}
((0,\infty) \times {\mathbf{S}}^{n-1})$ for some $\delta \in (1/2, 1)$, and so we can apply Proposition \[lin\_soln\_asymp\_expansion\] once more to see $$w \in \mathcal{C}^\infty_{-\beta} ((0,\infty) \times {\mathbf{S}}^{n-1}) \oplus
\operatorname{Span} \{w_j^\pm, j=0,1,2,\dots, n\}, \qquad
\beta = \min \{ 2\delta, \gamma_{\epsilon,2} \}.$$ However, combining and we have $$v_{\epsilon, a} (t+T,\theta) = v_\epsilon (t+T) + e^{-t-T} \langle \theta, a
\rangle w_1^+(t+T) + \mathcal{O} ( e^{-2t}) ,$$ which completes the proof.
[999]{}
T. Aubin. [*Équations différentielles non linéaires et problème de Yamabe concernant la courbure scalaire.*]{} J. Math. Pures Appl. [**55**]{} (1976), 269–296.
S. Baraket and S. Rebhi. [*Construction of dipole type signular solutions for a biharmonic equation with critical Sobolev exponent.*]{} Adv. Nonlinear Stud. [**2**]{} (2002), 459–476.
T. Branson. [*Differential operators canonically associated to a conformal structure.*]{} Math. Scandinavia. [**57**]{} (1985), 293–345.
T. Branson. [*Group representations arising from Lorentz conformal geometry.*]{} J. Funct. Anal. [**74**]{} (1987), 199–291.
T. Branson and A. R. Gover. [*Origins, applications and generalisations of the $Q$-curvature.*]{} Acta Appl. Math. [**102**]{} (2008), 131–146.
L. Caffarelli, B. Gidas, and J. Spruck. [*Asymptotic symmetry and local behavior of semilinear elliptic equations with critical Sobolev growth.*]{} Comm. Pure Appl. Math. [**42**]{} (1989), 271–297.
L. Caffarelli, T. Jin, Y. Sire, and J. Xiong. [*Local analysis of fractional semi-linear elliptic equations with isolated singularities.*]{} Arch. Ration. Mech. Anal. [**213**]{} (2014), 245–268.
S.-Y. A. Chang, M. Eastwood, B. Ørsted, and P. Yang. [*What is $Q$-curvature?*]{} Acta Appl. Math. [**102**]{} (2008), 119–125.
Z. Djadli, A. Malchiodi, and M. Ould Ahmedou. [*Prescribing a fourth order conformal invariant on the standard sphere II: blowup analysis and applications.*]{} Ann. Scuola Norm. Sup. Pisa [**5**]{} (2002), 387–434.
R. Frank and T. König. [*Classification of positive solutions to a nonlinear biharmonic equation with critical exponent.*]{} Anal. PDE [**12**]{} (2019), 1101–1113.
M. González. [*Singular sets of a class of locally conformally flat manifolds.*]{} Duke Math. J. [**129**]{} (2005), 551–572.
Q. Han, X. Li and Y. Li. [*Asymptotic expansions of solutions of the Yamabe equation and the $\sigma_k$-Yamabe equation near isolated singular points.*]{} preprint, [arxiv:1909.07466]{}
F. Hang and P. Yang. [*Lectures on the fourth order $Q$-curvature equation.*]{} Geometric analysis around scalar curvature, Lect. Notes Ser. Inst. Math. Sci. Natl. Univ. Singap. [**31**]{} (2016), 1–33.
T. Jin and J. Xiong. [*Asymptotic symmetry and local behavior of solutions of higher order conformall invariant equations with isolated singularities.*]{} preprint, [arxiv:1901.01678]{}
N. Korevaar, R. Mazzeo, F. Pacard, and R. Schoen. [*Refined asymptotics for constant scalar curvature metrics with isolated singularities.*]{} Invent. Math. [**135**]{} (1999), 233–272.
C. S. Lin. [*A classification of solutions of a conformally invariant fourth order equation in ${\mathbf{R}}^n$.*]{} Comment. Math. Helv. [**73**]{} (1998), 206–231.
Y.-J. Lin. [*Connected sum construction of constant $Q$-curvature manifolds in higher dimensions.*]{} Differential Geom. Appl. [**40**]{} (2015), 290–320.
W. Magnus and S. Winkler. [*Hill’s Equation.*]{} Dover Press, 1979.
R. Mazzeo and F. Pacard. [*Bifurcating nodoids.*]{} in Topology and geometry: commemorating SISTAG, Contemp. Math. [**314**]{} (2002), 169–186.
R. Mazzeo, D. Pollack, and K. Uhlenbeck. [*Moduli spaces of singular Yamabe metrics.*]{} J. Amer. Math. Soc. [**9**]{} (1996), 303–344.
R. Melrose. [*The Atiyah-Patodi-Singer index theorem.*]{} A. K. Peters, 1993.
S. Paneitz. [*A quartic conformally covariant differential operator for arbitrary pseudo-Riemannian manifolds.*]{} SIGMA Symmetry Integrability Geom. Methods Appl. [**4**]{} (2008), 3 pages (preprint from 1983).
R. Schoen. [*Conformal deformation of a Riemannian metric to constant scalar curvature.*]{} J. Diff. Geom. [**20**]{} (1984), 479–495.
R. Schoen and S. T. Yau. [*Conformally flat manifolds, Kleinian groups, and scalar curvature.*]{} Invet. Math. [**92**]{} (1988), 47–71.
N. Trudinger. [*Remarks concerning the conformal deformation of Riemannian structures on compact manifolds.*]{} Ann. Scuola Norm. Sup. Pisa [**22**]{} (1968), 265–274.
X. Xu. [*Uniqueness theorem for the entire positive solutions of biharmonic equations in ${\mathbf{R}}^n$.*]{} Proc. Royal Soc. Edinburgh. [**130A**]{} (2000), 651–670.
H. Yamabe. [*On the deformation of Riemannian structures on a compact manifold.*]{} Osaka Math. J. [**12**]{} (1960), 21–37.
[^1]: Institut für Mathematik, Universität Würzburg, [[email protected]]{}
|
{
"pile_set_name": "ArXiv"
}
|
[**A moment approach to compute quantum-gravity effects in the primordial universe**]{}
David Brizuela [^1] and Unai Muniain[^2]
> **Abstract.**An approach to compute quantum-gravity corrections to the scalar and tensorial power spectra of the inflationary perturbations is presented. The analysis of the Wheeler-DeWitt equation is performed by a decomposition of the wave function into its infinite set of moments, which must obey certain system of (first-class) constraints. Considering a semiclassical approximation, the system is truncated at second order in moments and an appropriate gauge-fixing condition is introduced, which allows us to interpret the scale factor of the universe as an internal time. The evolution of the different fluctuations and correlations is then explicitly considered for a de Sitter universe. An approximate analytical solution is obtained for the corrections of the power spectra, which produces an enhancement of power for large scales. Remarkably, the result is in agreement with previous studies in the literature that made use of very different semiclassical approximations. Finally, the numerical implementation of the system is also considered to verify the validity of the analytical solution.
Introduction
=============
Even if during the last decades a large amount of approaches have been attempted to construct a consistent theory that describes the quantum behavior of the gravitational interaction [@kieferquantumgravity], the ultimate theory is yet unknown. In particular, the lack of experimental data is one of the issues that makes progress in this research area complicate. Since quantum gravity effects are suppressed by the value of the Planck mass, they are indeed very difficult to be measured. Nonetheless, these effects could have considerable consequences on some astrophysical scenarios, such as the primordial inflationary universe. Although the radiation from such early times is not directly observable, quantum fluctuations present at those eras result into anisotropies on the Cosmic Microwave Background (CMB), for which quite accurate experimental data is already available [@Ade:2015xua].
In this context, it is very usual to apply the approximation of quantum field theory (QFT) on fixed classical backgrounds to study the evolution of such fluctuations. In this approximation the matter and geometric perturbations are treated as quantum entities, whereas the background universe is considered to follow a completely classical behavior. In fact, very good results are obtained by following this method (see, e.g., the reviews [@TASI1; @TASI2; @review3] or the textbook [@peter-uzan]), which agree with the commented data. Nevertheless it is expected that, since these quantum fluctuations were originated at the very beginning of the universe, the quantum behavior of the background might indeed produce some nontrivial (quantum-gravity) corrections to the standard results of QFT on classical backgrounds.
In the present work we will analyze corrections to the power spectra of the gauge-invariant scalar and tensorial inflationary perturbations based on the canonical approach to quantum gravity that leads to the Wheeler-DeWitt equation. Even if this theory might not be the fundamental theory of quantum gravity, it is a very conservative approach and it is expected to be correct at least at a semiclassical level. There are several works in the literature that have already analyzed these corrections, by using different Born-Oppenheimer-like approximations. On the one hand, in [@brizuelakieferkramer1; @brizuelakieferkramer2; @BEKKP13; @kieferkramer] an expansion with respect to the Newton’s gravitational constant $G_N$ was considered. In particular, the term of order $\mathcal{O}(G^0_N)$ provides exactly the QFT approximation and quantum-gravity corrections appear at next order $\mathcal{O}(G^{2}_N)$. On the other hand, in [@KTV13; @KTV14] a decomposition closer to the standard Born-Oppenheimer approximation used in molecular physics was followed, by separating the wave function of the fast degrees of freedom, which correspond to the perturbations, and of the slow ones, given by the background universe. In this case, the fast Hamiltonian reproduces the result obtained within the QFT approximation, while quantum corrections are encoded in the slow sector. Both approaches have agreed qualitatively in an enhancement of power for large scales.
The approach that will be used here is very different from these previous ones. We will use a formalism based on quantum moments for constrained systems, as introduced in [@BT09; @BT10; @BSST09], which does not require a classical deparametrization. More precisely, the wave function will be first decomposed into its infinite moments, such as its expectation values, fluctuations and correlations. The Wheeler-DeWitt equation will then be rewritten as an infinite set of (first-class) constraint equations for these moments. Since a moment of order $n$ scales as $\hbar^{n/2}$, for semiclassical peaked states one can safely truncate the system at a given finite order and get reliable results. In our case, in order to obtain the main quantum-gravity corrections to the inflationary power spectra, it will be enough to truncate the equations at second order in moments, or order $\hbar$. After the truncation, a gauge condition will be imposed to solve the constraint system, that will allow us to interpret the scale factor as an internal time. The obtained equations of motion will be solved analytically, making use of certain approximation. In addition, the numerical implementation of the system will also be considered, which will validate the approximate analytical solution. The well-known results of the QFT on classical background approximation will be easily recovered just by dropping the quantum moments corresponding to the background degrees of freedom.
Finally, let us briefly mention that in the context of loop quantum cosmology, which is also formulated by following a canonical quantization of the system, there has been an intense activity during the last years to obtain similar quantum-gravity corrections for the inflationary power spectra [@AgMo15; @SBBGLM16; @BlOl16; @CBMO17; @ABS18; @ABBMS18; @MLB18; @BJMM18; @Agu18]. Even if here we will not consider this theory, we point out that the method used and developed on this paper could also be applied to that framework.
The rest of the paper is organized as follows. In Section 2 we will briefly review the classical canonical description of an inflationary universe, which will be used to carry out the quantization in Section 3. In this section two different quantizations will be presented: the one corresponding to the approximation of QFT of classical background, and the full quantization of the system, that will lead to the Wheeler-DeWitt equation. We note that, in order to perform this last quantization, two approximations will be used: only the presence of one perturbative mode will be considered, and the frequency of the mode will not be quantized. These assumptions will be detailed and motivated in Section 3.2. In Section 4 we will apply the formalism of quantum moments for both quantization schemes and the expression of the power spectra will be obtained. In Section 5 we will consider the particular case of a de Sitter universe. The usual result of QFT for the power spectrum will be derived and corrections for that formula will be obtained by solving analytically and numerically the equations for the moments corresponding to the Wheeler-DeWitt quantization. Our main results and conclusions will be summarized in Section 6.
The canonical formulation of the classical model
================================================
In this section we briefly summarize the standard canonical formulation of linear perturbations of a homogeneous and isotropic universe.
Background homogeneous and isotropic universe
---------------------------------------------
Inflationary dynamics is very well described by considering the perturbations of a scalar matter field $\phi$ that propagates on a homogeneous and isotropic background. This background is described by the Friedmann-Lemaitre-Robertson-Walker metric $$ds^2 = -N(t)^2 dt^2 + a(t)^2 \left(\frac{dr^2}{1-{\rm k} r^2}+r^2(d\theta^2 + \sin^2\theta d\varphi^2)\right).$$ In this expression there are only two time-dependent variables: the scale factor $a(t)$ and the lapse function $N(t)$. Assuming that the matter field is minimally coupled with gravity and its self-interaction is given by a potential $V(\phi)$, the action for the background is given as [@bojowaldcanonicalgravity] $$S[a,N,\phi]
= L^3 \int dt \left[ -\frac{1}{2G}\left(\frac{a \dot{a}^2}{N}-{\rm k} N\right) + \frac{a^3}{2N}\dot{\phi}^2-Na^3 V(\phi) \right], %= \int L dt,$$ where we have defined the reduced gravitational coupling constant as $$G=\frac{4\pi G_N}{3},$$ $G_N$ being Newton’s constant, and $L^3$ stands for the integrated spatial volume. For the spatially compact cases $L^3$ is the total finite spatial volume of the universe; whereas for the flat case (${\rm k}=0$), which will be considered in the rest of the paper, it is in principle infinite. Nonetheless, in order to regularize the above expression, it is possible to absorb this volume by rescaling the different objects as follows, $$\label{rescaling}
a\rightarrow \frac{a}{L},\quad t\rightarrow L t,\quad N\rightarrow \frac{N}{L}.$$ After applying this rescaling, and by performing a Legendre transformation, the Hamiltonian of this system is obtained in a straightforward way, $$\mathcal{H}_0=N \left(-\frac{G\pi_a^2}{2 a} + \frac{\pi_\phi^2}{2a^3} + a^3 V(\phi)\right). \label{backgroundhamiltonian}$$ As it is well known, the lapse function is not a dynamical variable, but a Lagrange multiplier. The variation of the action with respect to it provides the constraint $$\mathcal{H}_0=0,$$ which implies that the Hamiltonian of the system must be vanishing. From this point on, we will choose the lapse as $N=a$, which defines the conformal time $t=\eta$.
Cosmological perturbations
--------------------------
The perturbed line element has the general form, $$ds^2 = a^2(\eta) \{ -(1-2A)d\eta^2 + 2(\partial_i B) dx^i d\eta + [(1-2\psi)\delta_{ij}+2\partial_i \partial_j E + h_{ij}]dx^i dx^j \}. \label{perturb}$$ The perturbative variables can be classified in three different sectors: the scalar, vectorial and tensorial ones. At linear level different sectors evolve independently and they can thus be treated separately.
The functions $A$, $B$, $E$ and $\psi$, in combination with the perturbation of the field $\delta\phi$, are scalar perturbations. Vector perturbations have not been considered when writing the above form of the metric since, in the absence of vectorial matter, they are purely gauge. Furthermore, this type of perturbations are known to rapidly decay during the accelerated expansion of the universe. Finally, $h_{ij}$ is the tensorial part that describes the gravitational waves.
Tensorial perturbations are invariant under gauge transformations, but the scalar perturbations are not. Nevertheless, they can be combined in a unique gauge-invariant quantity that encodes the complete physical information of the scalar sector. This quantity is called the Mukhanov-Sasaki master variable $v(\eta,\textbf{x})$, and is defined as [@MFB92] $$v(\eta,\textbf{x}) := \frac{a}{a'}\left[
a'\delta\phi + 2 a'\phi'(B-E')
+ a \phi'(A + B'-E'')\right],$$ where the prime denotes the derivative with respect to the conformal time $\eta$. Since the equations of motion are linear, it is useful to introduce the Fourier transform as $$v(\eta,\textbf{x})= \frac{1}{(2\pi)^{3/2}} \int_{\mathbb{R}^3} d^3\textbf{k} \: v_\textbf{k}(\eta) \: e^{i\textbf{k}\cdot\textbf{x}}.\label{vFourier}$$
In order to obtain the dynamics of the variable $v$, the action of the system $S$ is expanded up to second order in the perturbations, $\delta^2 S$. In terms of the Fourier components the scalar part of the action then reads [^3] $$\delta^2 S = \frac{1}{2} \int d\eta \int d^3\textbf{k} \left(v'_\textbf{k} v'^{*}_\textbf{k} + v_\textbf{k} v^{*}_\textbf{k}\left[ \frac{(a\sqrt{\epsilon})''}{a\sqrt{\epsilon}}-k^2 \right]\right),$$ where $\epsilon$ is known as the first slow-roll parameter and is defined as $$\epsilon := -\frac{H'}{a H^2}, \label{epsilondefinition}$$ $H:=a'/a^2$ being the Hubble parameter. From this action, it is straightforward to obtain the canonical momentum for each $\textbf{k}$, $$\pi_\textbf{k} = \frac{\partial L}{\partial v'_\textbf{k}} = v'^*_\textbf{k}.$$ Performing then a Legendre transformation the Hamiltonian can be finally deduced: $$\mathcal{H}_S = \frac{1}{2} \int d^3 \textbf{k} \; \left(\pi_\textbf{k} \pi^{*}_\textbf{k} + v_\textbf{k} v^{*}_\textbf{k}\left[k^2 - \frac{(a\sqrt{\epsilon})''}{a\sqrt{\epsilon}} \right]\right). \label{scalarhamilton}$$ This Hamiltonian corresponds to a system of decoupled harmonic oscillators, with a time-dependent frequency given by $$\omega^2_{S} (\eta,k) := k^2 - \frac{(a\sqrt{\epsilon})''}{a\sqrt{\epsilon}}. \label{scalarfreq}$$
Regarding tensorial perturbations, as commented above, they are described by the tensor $h_{ij}$. Besides being symmetric, this tensor is also traceless $h^i{}_i=0$ and divergence-free $\partial^i h_{ij}=0$. Therefore it just contains two physical degrees of freedom, which correspond to the two polarizations of the gravitational wave: $h^+$ and $h^\times$. In particular, if the Cartesian $z$ coordinate is chosen to be the direction of the propagation of the transverse wave, these polarization terms are given by $h^+ = h_{xx} = -h_{yy}$ and $h^\times = h_{xy} = h_{yx}$.
The derivation of the Hamiltonian for the tensorial part is completely analogous to the process explained above for scalars. In order to write it down, let us define the master variable $$v^s_\textbf{k} := \frac{a}{2\sqrt{3 G}}h^s_\textbf{k},$$ where $s$ stands for the two polarizations, $+$ and $\times$. After expanding the action, transforming to Fourier components and performing the Legendre transformation, the tensorial Hamiltonian is obtained: $$\mathcal{H}_T = \int d^3 \textbf{k} \sum_s \left( \frac{1}{2}\pi^s_\textbf{k}\pi^{s*}_\textbf{k} + \frac{\omega^2_{T}(\eta, k)}{2} v^s_\textbf{k} v^{s*}_\textbf{k} \right).\label{tensorhamilton}$$ Note that for these tensor modes the sum for the two polarizations $s$ must be considered. The Hamiltonian corresponds again to a system of decoupled harmonic oscillators, but in this case the frequency is given by $$\omega_{T}^2 := k^2-\frac{a''}{a}.\label{tensorfreq}$$
Even if both $v_{\bf k}$ and $\pi_{\bf k}$ are complex, in the Hamiltonian they always appear in quadratic combination with their corresponding complex conjugate. In order to proceed to the quantization, for each of the variables one in principle should define a new set of two real variables, for instance by taking their real and imaginary parts as done, e.g., in [@MVP12]. However, since these variables never appear outside the commented quadratic combination, and the result will thus be unaffected by this consideration, in order to enlighten the notation we will just use the variables $v_{\bf k}$ and $\pi_{\bf k}$ as if they were real.
The quantization of the model
=============================
In this section two different quantizations of the classical system presented above will be considered. The first one will be the standard one given by quantum field theory on fixed classical backgrounds. In this case only the perturbative variables $v_{\bf k}$ will be quantized, whereas the background will be left classical. In the second quantization the full model (background plus perturbations) will be quantized.
Quantum field theory on classical backgrounds
---------------------------------------------
In this approach each mode of the tensorial and scalar perturbations can be considered independently, and one just promotes to an operator their corresponding Hamiltonian, either (\[scalarhamilton\]) or (\[tensorhamilton\]), in order to obtain the functional Schrödinger equation, $$\label{schroedingerequation}
\hat{\cal H}_\mathbf{k} \Psi(\eta, v_\textbf{k}):=
\left(\frac{\hat{\pi}_\textbf{k}^2}{2} + \frac{\omega^2(\eta, k) \hat{v}_\textbf{k}^2}{2}\right) \Psi(\eta, v_\textbf{k}) = i\hbar\frac{\partial \Psi(\eta, v_\textbf{k})}{\partial \eta}.$$ In this representation the momentum operator acts as a derivative with respect to $v_\mathbf{k}$, that is $\hat{\pi}_\textbf{k}:=-i\hbar\partial/\partial v_\mathbf{k}$, whereas the position operator acts by multiplication. The dependence on the background universe is entirely inside the classical function $\omega^2(\eta, k)$, which has the form given by (\[scalarfreq\]) or (\[tensorfreq\]) depending whether $v_\textbf{k}$ corresponds respectively to a scalar or tensorial mode.
Full quantization of the model: the Wheeler-DeWitt master equation
------------------------------------------------------------------
In order to perform a quantization of the full model, one needs to consider the total action that describes completely the background and perturbative dynamics. This is given by $$S_{total} = S + \delta^2 S.\label{totalaction}$$ Starting from this action for a general lapse $N(t)$, the total Hamiltonian is obtained, which is just the sum of the background and perturbative Hamiltonians: $$\mathcal{H}_{total} = N(\mathcal{H}_0 + \delta^2 \mathcal{H}) = N(\mathcal{H}_0 + \mathcal{H}_S + \mathcal{H}_T).$$ In this expression, the lapse $N$ multiplies the whole Hamiltonian and the background constraint $\mathcal{H}_0=0$ is not satisfied anymore but it gets quadratic corrections in the perturbative variables. The Hamiltonian constraint now reads $$\mathcal{H}_0 + \mathcal{H}_S + \mathcal{H}_T = 0.$$ More explicitly, by choosing the lapse $N(t)=a$, which corresponds to the conformal time $\eta$, the classical Hamiltonian constraint takes the following form, $$\begin{aligned}
\mathcal{H}_{total} = -\frac{G}{2} \pi_a^2 + \frac{\pi_\phi^2}{2a^2} + a^4 V(\phi) + \frac{1}{2} \int d^3 \textbf{k} \;\left(\pi_\textbf{k}^2 + \omega^2_{S} v_\textbf{k}^2 \right)
% \nonumber\\
% &+&
+\frac{1}{2}\int d^3 \textbf{k} \sum_{s} \left((\pi^s_\textbf{k})^2
+ \omega^2_{T} (v^s_\textbf{k})^2 \right) = 0.\label{hamiltontotal}\end{aligned}$$
Once this Hamiltonian is at hand, the quantization procedure is performed by promoting different variables into operators that satisfy the canonical commutation relations, $$[\hat{a},\hat{\pi}_a] = i\hbar, \qquad [\hat{\phi}, \hat{\pi}_\phi] = i\hbar, \qquad [\hat{v}_\textbf{k},\hat{\pi}_\textbf{p}] = i\hbar\delta(\textbf{k}-\textbf{p}),\label{commutators}$$ where $v_\textbf{k}$ stands now for both scalar and tensorial components. The action of these operators on a wave function on the position representation $\Psi(a,\phi,\{ v_\textbf{k}\})$ is as follows, $$\hat{a}\Psi = a\Psi, \qquad \hat{\pi}_a\Psi = -i\hbar\frac{\partial \Psi}{\partial a}, \qquad
\hat{\phi}\Psi = \phi\Psi, \qquad \hat{\pi}_\phi\Psi = -i\hbar\frac{\partial \Psi}{\partial \phi} \qquad \hat{v}_\textbf{k}\Psi = v_\textbf{k}\Psi, \qquad \hat{\pi}_\textbf{k}\Psi = -i\hbar\frac{\partial \Psi}{\partial v_\textbf{k}}. \label{canonicalquantback}$$
The master Wheeler-DeWitt equation is then obtained just by requesting that the action of the Hamiltonian operator on the wave function is vanishing, $$\label{WdWfull}
\hat{\cal H}_{total}\Psi=0.$$ This equation is very difficult to deal with and, following [@kieferkramer], one can further simplify it by assuming a product ansatz for the wave function $\Psi$ and dropping cross terms that involve different modes. In this way, one obtains an individual Wheeler-DeWitt equation for each mode, $$\hat{\mathcal{H}}\Psi_\textbf{k} := \left(-G\hat{\pi}_a^2+\frac{\hat{\pi}_\phi^2}{\hat{a}^2}+2\hat{a}^4V(\hat{\phi})+\hat{\pi}_\textbf{k}^2+\omega^2(\eta, k) \hat{v}_\textbf{k}^2\right)\Psi_\textbf{k} = 0.\label{WheelerDeWittequation}$$ Physically, considering this last equation instead of the full one corresponds to assuming just the presence of one perturbative mode in the classical model. Therefore, for a given mode $\bf k$, we are neglecting possible effects from other modes on its dynamics, but only analyzing how the quantum behavior of the background will affect its evolution. This is a natural extension of the assumption performed at a classical level (and considered also in the QFT approach) that the perturbations can be linearized, which decouples different modes, and thus their dynamics is only influenced by the background. If now the background dynamics is modified, because of its quantum behavior, it is reasonable to assume that each mode will be affected more importantly from the change of the background and not from the quantum dynamics of other modes. In fact, the effects of the other modes might also come from nonlinearities (second and higher-order terms in perturbation theory), and not only from the quantization of the background. In this way, we focus on the essential ingredient that is neglected in the QFT approach: the quantum behavior of the background.
The equation above is valid for both a scalar or a tensorial mode and that is why we have not written any of the subindices $S$ or $T$ in the frequency $\omega$. In fact, from this point on, all expressions will be valid for both scalar and tensor modes, and thus we will use the symbol $\omega$ indistinctly.
Finally, note that in this quantization procedure the frequency $\omega$ has been left as a classical function of time, given by its form (\[scalarfreq\]) or (\[tensorfreq\]). This is a usual approximation in the literature, see e. g. [@brizuelakieferkramer1; @KTV13]. The main reason not to perform the quantization of the frequency in the general case is that it leads to nonlocal operators because the classical expression of this quantity contains inverse of background momenta. One could argue that one should follow this path and consider this nonlocality as an inherent and fundamental feature of the model. Nevertheless, we note that the inverse of the momenta that appear inside the frequency come from performing certain canonical transformations at a classical level in order to define and decouple the gauge-invariant master variable from the pure-gauge and constrained variables. Nonetheless, the Hamiltonian of the perturbations that is defined just by considering the second-order perturbation of the usual Hamiltonian of general relativity, is linear in the momenta. Therefore, apart from the technical difficulty that this nonlocality would add to the model, because of the commented reasons, it might not be an actual fundamental property of the theory and might lead to spurious results.
In addition, the frequency appears in the Hamiltonian constraint multiplying the square of the perturbative variable $v_{\bf k}$. Therefore, when expanding the Hamiltonian at second-order in moments, the fluctuations and correlations of the frequency would also be multiplied either by $v_{\bf k}$ or $v^2_{\bf k}$. In the QFT approach, by choosing adiabatic initial states , the variable $v_{\bf k}$ is vanishing all along evolution. For the de Sitter case, we will show that this also happens in our framework \[see equation \]. Therefore, at second-order in moments, the quantum effects of the frequency will, in general, be canceled out by the vanishing of the variable $v_{\bf k}$.
Quantum moments and the power spectra for the perturbations {#sec_moments}
===========================================================
In this section we will analyze the two quantizations performed in the previous section by decomposing the corresponding wave function into its infinite set of moments.
Quantum moments for quantum field theory on classical backgrounds {#moments_classical_back}
-----------------------------------------------------------------
In order to analyze the system given in the approximation of QFT on curved backgrounds that is described by the Schrödinger equation , we will define the following quantum moments, $$\Delta (v_{\bf k}^n\pi_{\bf k}^m):=
\langle (\hat{v}_{\bf k}-v_{\bf k})^n
(\hat{\pi}_{\bf k}-\pi_{\bf k})^m
\rangle_{Weyl}, \label{quantummoments}$$ where the Weyl subscript means that the expectation value is evaluated for the totally symmetric ordered set of operators, and the expectation values $v_{\bf k}:= \langle \hat{v}_{\bf k} \rangle$ and $\pi_{\bf k}:= \langle \hat{\pi}_{\bf k} \rangle$ have been defined. This is indeed a slight abuse of notation, as we are using the same symbols to denote the classical variables and these expectation values. Nonetheless, we consider it is not worth introducing a more complicate notation since the distinction will be clear from the context and, in particular, from this point on $v_{\bf k}$ and $\pi_{\bf k}$ will always stand for the expectation values of their corresponding operators, and not for the classical variables. The value $n+m$ will be referred as the order of the quantum moment. In addition, as we will be dealing with each **k** mode separately, we will consider that $v_\mathbf{k}$ and $\pi_\mathbf{k}$ are nonfield quantum variables, satisfying $[\hat{v}_\textbf{k},\hat{\pi}_\textbf{k}] = i\hbar$, and formally regularize the Dirac deltas as $\delta(0)=1$.
Following this definition, the wave function is replaced by its infinite set of quantum moments. In fact, the knowledge of all moments (including the expectation values $v_{\bf k}$ and $\pi_{\bf k}$) is completely equivalent to the full expression of the wave function. The advantage of this formalism is that one works directly with observable quantities. Furthermore, a moment of order $n$ has the dimensions of $\hbar^{\frac{n}{2}}$. Therefore, for peaked semiclassical states ($\hbar \ll 1$), one can make the assumption that higher-order moments are negligible, truncate the infinite set of moments and obtain physically reliable results.
The dynamics of these quantum moments will be given by the expectation value of the Hamiltonian $\hat{\cal H}_k$, which can be written in terms of the above moments by performing an expansion around the expectation values $v_{\bf k}$ and $\pi_{\bf k}$, $$\langle \hat{\mathcal{H}}_k(\hat v_{\bf k},\hat\pi_{\bf k})\rangle =
{\mathcal{H}}_k( v_{\bf k},\pi_{\bf k})+\sum_{n+m\geq 2}\frac{1}{n!m!}\frac{\partial^{n+m}\mathcal{H}_k}{\partial v_{\bf k}^n
\partial \pi_{\bf k}^m}
\Delta (v_{\bf k}^n\pi_{\bf k}^m)=
\frac{1}{2}\left(\pi_\textbf{k}^2 + (\Delta \pi_\textbf{k})^2 + \omega^2 v_\textbf{k}^2 + \omega^2 (\Delta v_\textbf{k})^2\right).\label{qfthamiltonian}$$ (Here, and in the rest of the paper, in order to follow the standard notation, the fluctuations have been denoted as $(\Delta X)^2:=\Delta(X^2)$.) As can be seen, in the general case the expectation value of the Hamiltonian would be given by an infinite sum in moments. Nonetheless, when the Hamiltonian is [*harmonic*]{} (defined as containing only up to quadratic combinations of basic variables), as it is the case, only up to second-order moments appear in the expression of its expectation value. These harmonic Hamiltonians are of a very special kind and their properties in terms of this moment formalism are explained in detail, for instance, in [@Bri14]. One of their properties is that the equations of motion for different orders in moments decouple. In this paper we will be interested in computing the power spectra for the perturbations which, as will be explained below, are given by the fluctuation of the perturbative master variable $(\Delta v_{\bf k})^2$. Therefore, in order to obtain the exact evolution of that variable given by this harmonic Hamiltonian, it will be enough to consider the system up to second order.
The equations of motion of all variables are obtained by computing the Poisson brackets with the above Hamiltonian. These brackets can be defined in terms of the commutators by the standard relation $\{ \langle \hat{X} \rangle, \langle \hat{Y} \rangle \} = -i/\hbar \langle [\hat{X},\hat{Y}] \rangle$. In particular, the evolution for the expectation values and second-order moments are given by, $$\begin{aligned}
\label{qfteq1}
v_\textbf{k}' &=& \pi_\textbf{k},\\
\pi_\textbf{k}' &=& -\omega^2 v_\textbf{k},
\\\label{qfteq3}
\left( (\Delta v_\textbf{k})^2 \right)' &=& 2\Delta(v_\textbf{k} \pi_\textbf{k}),
\\\label{qfteq4}
\left(\Delta(v_\textbf{k} \pi_\textbf{k}) \right)' &=& (\Delta \pi_\textbf{k})^2 - \omega^2 (\Delta v_\textbf{k})^2,
\\
\left( (\Delta \pi_\textbf{k})^2 \right)' &=& -2\omega^2 \Delta(v_\textbf{k} \pi_\textbf{k}).\label{qfteq5}\end{aligned}$$ For future convenience, we note that the equations of motion for second-order moments can be rewritten as a unique equation for the fluctuation $(\Delta v_\textbf{k})^2$, $$\left( (\Delta v_\textbf{k})^2 \right)''' + 4\,\omega^2 \left( (\Delta v_\textbf{k})^2 \right)'+ 4\,\omega\,\omega'\, (\Delta v_\textbf{k})^2 = 0.\label{deltav2general}$$
As can be seen, the expectation values evolve exactly as in the classical theory. In order to solve these equations is usual to find a region where the frequency $\omega$ tends to a constant. In inflationary scenarios this happens at the beginning of inflation ($k\eta \rightarrow -\infty$) since the mode is well-inside the horizon and it does not feel the effects of the curvature. In that limit the frequency tends to the wave number $\omega\rightarrow k$ and the mode behaves as a free mode evolving on a Minkowski background. The state of the mode is then assumed to be stationary. This stationarity condition can be implemented just by setting equal to zero all time derivatives in the equations above –; which implies that the expectation values must be vanishing, $$\label{vpiconditions}
v_\textbf{k}=\pi_\textbf{k}=0,$$ as well as the correlation, $$\label{condition1}
\Delta(v_\textbf{k} \pi_\textbf{k})=0.$$ In addition, from equation , the following relation between fluctuations must hold, $$\label{condition2}
(\Delta \pi_\textbf{k})^2 = k^2 (\Delta v_\textbf{k})^2,$$ where the frequency has been replaced by its constant value $k$. Note that, no matter what is the evolution of the frequency, the stationarity conditions for the expectation values will be obeyed during the whole evolution. And, as long as the frequency $\omega$ is kept constant, the conditions - on second-order moments will also be obeyed. But, the evolution of the universe makes the frequency time-dependent and thus, when the mode begins to feel the effects of the curvature, these conditions will cease to be valid and the values of the fluctuations and the correlation will not be constant anymore.
Finally the above initial conditions – are just 4 relations for our 5-variable system –. Therefore, one freedom is left to be fixed, which is related to the energy of the initial state at the beginning of inflation. It is usual, though not necessary and in fact several works can be found in the literature analyzing an excited initial state, see for instance [@Arm07; @Gan11; @ALP13], to impose that the state is on its fundamental (less-energetic) level. This choice is usually known as the Bunch-Davies vacuum [@BD78]. Replacing conditions – in the Hamiltonian and taking into account that the Heisenberg uncertainty relation, $$(\Delta v_\textbf{k})^2(\Delta \pi_\textbf{k})^2 - (\Delta(v_\textbf{k}\pi_\textbf{k}))^2 \geq \frac{\hbar^2}{4},\label{uncertainty}$$ must be obeyed, it is easy to obtain the initial value for the fluctuations of the basic variables, $$\label{bdvacuum}
(\Delta \pi_\textbf{k})^2 = \frac{\hbar k}{2},\qquad
(\Delta v_\textbf{k})^2 =\frac{\hbar}{2 k}.$$ This obviously corresponds to the fundamental state of the harmonic oscillator and its energy is $\hbar k/2$. As a side remark, let us mention that in this framework this is the only place where the Planck constant $\hbar$ enters the computations (it does not appear in the Hamiltonian or in the equations of motion –). Therefore, the appearance of the Planck constant in the result will be just due to this choice of initial state.
Finally, let us note that the dynamics given by the present Hamiltonian conserves the combination that appears on the left-hand side of the uncertainty relation , which is another generic property of the harmonic Hamiltonians [@Bri14]. This happens because the evolution of the uncertainty relation is given by third-order derivatives of the Hamiltonian and, thus, if those are vanishing this combination is conserved. Therefore, any state saturating initially the uncertainty relation will saturate it during the whole evolution. Even so, in the general case, each moment independently will not be constant and thus the state will be deformed through evolution.
Quantum moments for the Wheeler-DeWitt equation
-----------------------------------------------
In the full quantum model, the background degrees of freedom, described by the couples $(a,\pi_a)$ and $(\phi,\pi_\phi)$, are also promoted to operators. Therefore, in order to describe the complete system, their corresponding moments have also to be taken into account. The notation will be similar to the one used in the previous section. We will only deal with up to second-order quantities. Therefore the fluctuation of a given operator $\hat X$ will be denoted as $(\Delta X)^2$, whereas the correlation between two operators $\hat X$ and $\hat Y$, will be denoted as $\Delta (X Y)$, where the operators will be symmetrically ordered and thus $\Delta (X Y)=\Delta (YX)$.
As opposed to the case of QFT on classical backgrounds of the previous section, in the full quantum picture there is not an evolution equation of the form , rather the constraint equation (\[WheelerDeWittequation\]) governs the dynamics of the system. Furthermore, in this case the Hamiltonian is not harmonic and different orders in moments couple. In particular, higher-order moments enter into the equations of motion for the expectation values, and thus these do not follow the classical trajectories anymore. As a first approximation, we will assume a peaked semiclassical state and introduce a truncation of the system at second order in moments.
In order to rewrite the Wheeler-DeWitt equation in terms of the moments we will follow [@BT09] and request that, since the action of the Hamiltonian on any physical wave function must be vanishing, the expectation value of the Hamiltonian constraint multiplied from the left-hand side by any product of basic operators should be vanishing as well. More precisely, the following set of infinite relations should be obeyed $$\langle (\hat{\pi}_a-\pi_a)^{n_1}(\hat{a}-a)^{n_2}(\hat{\pi}_\phi-\pi_\phi)^{n_3} (\hat{\phi}-\phi)^{n_4} (\hat{\pi}_\textbf{k}-\pi_\textbf{k})^{n_5} (\hat{v}_\textbf{k}-v_\textbf{k})^{n_6} \hat{\mathcal{H}} \rangle = 0.\label{towerconstraints}$$ This is in fact a tower of constraints that should be solved order by order. As commented, we will truncate the system at second order. In this way, only 7 constraints will have to be satisfied, which are the following: [$$\begin{aligned}
C &:=& \langle \hat{\mathcal{H}} \rangle = -G\left(\pi_a^2+ (\Delta \pi_a)^2\right) + \frac{1}{a^4}\left(a^2 \pi_\phi^2 + a^2 (\Delta \pi_\phi)^2-4 a \pi_\phi \Delta(a \pi_\phi) + 3\pi_\phi^2 (\Delta a)^2\right)
\notag \\
&+&2a^2 V(\phi)\,\left(a^2+6 (\Delta a)^2\right)
+ 8a^3 V'(\phi) \Delta(a \phi)+a^4 V''(\phi) (\Delta \phi)^2
+ \pi_\textbf{k}^2 + (\Delta \pi_\textbf{k})^2
\notag \\
&+& \omega^2(v_\textbf{k}^2 + (\Delta v_\textbf{k})^2) = 0,\label{quantumconstraint1}\\
C_{\pi_a} &:= &\frac{1}{2} \langle(\hat{\pi}_a-\pi_a)\hat{\mathcal{H}}\rangle = -G\pi_a (\Delta \pi_a)^2 + \frac{\pi_\phi}{a^2} \Delta(\pi_a \pi_\phi)+
\frac{1}{2 a^3}\left( 4a^6 V(\phi)-\pi_\phi^2\right)
\left(2 \Delta(\pi_a a)-i\hbar\right)\notag \\
&+& a^4 V'(\phi)\,\Delta(\pi_a \phi) + \pi_\textbf{k}\, \Delta(\pi_a \pi_\textbf{k}) + \omega^2 v_\textbf{k}\, \Delta(\pi_a v_\textbf{k})=0, \label{quantumconstraint2}\\
C_{a} &:= & \frac{1}{2} \langle(\hat{a}-a)\hat{\mathcal{H}} \rangle = -\frac{G}{2}\pi_a (2 \Delta(\pi_a a)+i \hbar) + \frac{\pi_\phi}{a^2} \Delta(a \pi_\phi)+ \frac{1}{a^3}\left( 4a^6 V(\phi)-\pi_\phi^2\right) (\Delta a)^2\notag \\\label{quantumconstraint3}
&+& a^4 V'(\phi)\,\Delta(a \phi ) + \pi_\textbf{k} \,\Delta(a \pi_\textbf{k}) + \omega^2 v_\textbf{k}\, \Delta(a v_\textbf{k})=0, \\
C_{\pi_\phi} &:= & \frac{1}{2} \langle(\hat{\pi}_\phi-\pi_\phi)\hat{\mathcal{H}} \rangle = -G\pi_a \Delta(\pi_a \pi_\phi) + \frac{\pi_\phi}{a^2} (\Delta \pi_\phi)^2+\frac{1}{a^3}\left( 4a^6 V(\phi)-\pi_\phi^2\right)\Delta(a \pi_\phi)\notag \\
&+& \frac{a^4}{2} V'(\phi)\,(2\Delta(\pi_\phi \phi)-i\hbar) + \pi_\textbf{k} \,\Delta(\pi_\phi \pi_\textbf{k}) + \omega^2 v_\textbf{k}\, \Delta(\pi_\phi v_\textbf{k})=0, \\
C_{\phi} &:= & \frac{1}{2} \langle(\hat{\phi}-\phi)\hat{\mathcal{H}} \rangle = -G\pi_a \Delta(\pi_a \phi) + \frac{\pi_\phi}{2 a^2} (2 \Delta(\pi_\phi \phi)+i\hbar)+\frac{1}{a^3}\left( 4a^6 V(\phi)-\pi_\phi^2\right)\Delta(a \phi)\notag \\
&+& a^4 V'(\phi)\,(\Delta \phi)^2 + \pi_\textbf{k} \,\Delta(\phi \pi_\textbf{k}) + \omega^2 v_\textbf{k}\, \Delta(\phi v_\textbf{k})=0 \\
C_{\pi_{\rm k}} &:= & \frac{1}{2} \langle(\hat{\pi}_\textbf{k}-\pi_\textbf{k})\hat{\mathcal{H}} \rangle = -G\pi_a \Delta(\pi_a \pi_\textbf{k}) + \frac{\pi_\phi}{a^2} \Delta(\pi_\phi \pi_\textbf{k})+\frac{1}{a^3}\left( 4a^6 V(\phi)-\pi_\phi^2\right)\Delta(a \pi_\textbf{k})\notag \\
&+& a^4 V'(\phi)\,\Delta(\phi \pi_\textbf{k}) + \pi_\textbf{k}\, (\Delta \pi_\textbf{k})^2 + \frac{\omega^2}{2} v_\textbf{k}\, (2 \Delta(\pi_\textbf{k} v_\textbf{k})-i\hbar)=0, \\
C_{v_{\rm k}} &:= & \frac{1}{2} \langle(\hat{v}_\textbf{k}-v_\textbf{k})\hat{\mathcal{H}} \rangle = -G\pi_a \Delta(\pi_a v_\textbf{k}) + \frac{\pi_\phi}{a^2} \Delta(\pi_\phi v_\textbf{k})+\frac{1}{a^3}\left( 4a^6 V(\phi)-\pi_\phi^2\right)\Delta(a v_\textbf{k})\notag \\
& + & a^4 V'(\phi)\,\Delta(\phi v_\textbf{k}) + \frac{\pi_\textbf{k}}{2}(2 \Delta(\pi_\textbf{k} v_\textbf{k})+i\hbar) + \omega^2 v_\textbf{k}\, (\Delta v_\textbf{k})^2=0. \label{quantumconstraint7}\end{aligned}$$ ]{} Note that, for convenience, the Hamiltonian operator has been assumed to be Weyl ordered. In order to impose such ordering, for both the inverse of the scale-factor operator $\hat a$ and the generic potential of the field $V(\hat\phi)$ that appear inside , a Taylor expansion around their corresponding expectation values has been considered.
In the Appendix A all Poisson brackets between these different constraints are explicitly computed up to linear order in moments or order $\hbar$. As can be seen, all of them are a linear combination of constraints and thus vanish on shell. Therefore, all the above constraints are first class. In summary, we have a 27-dimensional space, composed by the 6 expectation values $(a,\pi_a,\phi,\pi_\phi,v_\mathbf{k},\pi_\mathbf{k})$, their 6 corresponding fluctuations, and their 15 independent correlations. The evolution of the system will be described by curves on that space and we have the gauge freedom to choose the time that will parameterize those curves. In principle there are no restrictions for such a choice and any combination of the expectation values, or higher-order moments, could do the job. Nonetheless, considering the physical system we are describing, it is natural to choose one of the background variables as time, so that the deparametrized system still makes sense for the purely background case without perturbations. There are 4 such variables $(a,\pi_a,\phi,\pi_\phi)$ but, in the next section, this formalism will be applied to the de Sitter universe, where the matter field $\phi$ is a constant and not a dynamical variable anymore. (In such a case, an evolution in $\phi$ would describe the nonphysical evolution between different de Sitter universes.) Hence, we will impose a gauge so that $a$ will be our internal time variable and its canonical conjugate momentum $\pi_a$ will play the role of a physical Hamiltonian. Let us explain in more detail how this is actually derived from the above constraint equations.
Since we have decided to use $a$ as our time variable, the above system of 7 equations must be understood to be solved for all 7 objects related to its momentum $\pi_a$; namely, $\pi_a$ itself, its fluctuation $(\Delta \pi_a)^2$, and its 5 correlations with the rest of the variables $\Delta(X\pi_a)_{X \neq \pi_a}$. In particular, we note that this system is linear in the commented fluctuations and correlations. Therefore, it is straightforward to obtain the solution for the commented second-order moments $\Delta(X\pi_a)$ from – and replace it in the expectation value of the Hamiltonian constraint . In this way, one ends up with a fourth-order polynomial equation for the momentum $\pi_a$: $$\label{eqpia}
- G^2 \pi_a^4+ G A_2 \pi_a^2+i \hbar G A_1\pi_a+A_0=0,$$ where the coefficients $A_i$ are real functions of expectation values and moments unrelated to $\pi_a$, that is, moments of the form $\Delta(XY)_{X\neq \pi_a, Y\neq \pi_a}$. In particular, the only imaginary term appears as a multiplicative factor of the linear term in $\pi_a$, and its corresponding coefficient $A_1$ has the form, $$A_1=-\frac{\pi_\phi^2}{a^3}+4 a^3 V(\phi).$$ Note that this coefficient is given in terms of expectation values and does not contain any higher-order moment. In fact, it is proportional to the “time” ($a$-) derivative of the classical constraint ${\cal H}_0$. That is, if the Hamiltonian was time ($a$-) independent, this term would not be present and the equation to solve for $\pi_a$ would be real. The appearance of imaginary parts for the time variable is a generic feature of this approach, as has been analyzed in [@BHT11; @BHT11a; @HKT12; @BoHa18]. Nonetheless, as will be explained below, in our case the evolution of the universe will make the imaginary terms to be decreasing and thus will be safe to drop them. Furthermore, the other two coefficients of the polynomial equation are given by, $$\begin{aligned}
A_0&=& 2 a \Delta(a\phi) V'(\phi ) \left[\pi_\phi^2-4 a^6 V(\phi )\right]
-\frac{(\Delta a)^2}{a^6}\left[\pi_\phi^2-4 a^6 V(\phi
)\right]^2-a^8 (\Delta \phi)^2 V'(\phi )^2
\\
&+&\frac{\Delta(a\pi_\phi)}{a^5}\left[2 \pi_\phi^3-8 a^6 \pi_\phi
V(\phi )\right]-\frac{\pi_\phi^2}{a^4}(\Delta \pi_\phi)^2-2 a^4 \pi_{\bf k}
\Delta(\phi\pi_{\bf k}) V'(\phi )-2 a^4 \omega ^2 v_\mathbf{k} \Delta(\phi v_{\bf k}) V'(\phi )
\\
&-&\frac{2\pi_\phi}{a^2}\left[ \pi_{\bf k} \Delta(\pi_\phi\pi_{\bf k})
-\omega ^2 v_\mathbf{k} \Delta(\pi_\phi v_{\bf k})\right]
- 2 a^2 \pi_\phi \Delta(\phi\pi_\phi) V'(\phi )
\\
&+&\frac{2}{a^3}[\pi_{\bf k} \Delta(a\pi_{\bf k})
+\omega ^2 v_{\bf k}\Delta(a v_{\bf k})] \left[\pi_\phi^2-4 a^6 V(\phi )\right]
-\pi_{\bf k}^2 (\Delta \pi_{\bf k})^2
-2 \pi_{\bf k} \omega^2 v_{\bf k} \Delta(v_{\bf k}\pi_{\bf k})
-\omega ^4 v^2 (\Delta v_{\bf k})^2,\\
A_2&=&
\frac{\pi_\phi^2}{a^2}+2 a^4 V(\phi)+\pi_\textbf{k}^2+\omega ^2 v_\textbf{k}^2
+\frac{3 \pi_\phi^2}{a^4} (\Delta a)^2+\frac{1}{a^2} (\Delta \pi_\phi)^2-\frac{4\pi_\phi}{a^3} \Delta (a \pi_\phi)+ 12 a^2 V(\phi )(\Delta a)^2
\nonumber\\
&+&a^4
(\Delta \phi)^2 V''(\phi )
+8 a^3 \Delta(a \phi) V'(\phi)
+\omega^2 (\Delta v_\textbf{k})^2+(\Delta \pi_\textbf{k})^2.\end{aligned}$$ Note that if $A_0,A_1\ll A_2$, the momentum would take the form $\pi_a\approx \sqrt{A_2/G}$. That is in fact the solution one would obtain from the expectation value of the Hamiltonian constraint by considering that the system was on a “momentum eigenstate” and thus its fluctuation was vanishing $(\Delta \pi_a)^2=0$. Therefore, in this sense, the coefficients $A_0$ and $A_1$ contain the effects of considering the quantum behavior of this variable.
In fact, in our case, the term $A_1$ can be neglected. We will consider the de Sitter case with an initial Bunch-Davies vacuum and, in principle, it is expected that the system will not depart too much from its classical behavior. Therefore, the Hubble factor $H$ would be approximately constant and the scale factor will be given by an exponential of the cosmological time $t$ as $a\approx e^{H t}$. The rest of the variables classically go as $\pi_a\approx e^{2 H t}$, $\pi_\phi\approx 0$, $V(\phi)\approx H^2$, $v_\textbf{k} \approx \pi_\textbf{k}\approx 0$. With this information at hand, we obtain that the two first terms of the equation behave as $$\pi_a^4\approx A_2 \pi_a^2\approx e^{8 H t},$$ whereas the linear term also increases in time, but much slower, $A_1\pi_a\approx e^{5 H t}$. Therefore, in this case, one can neglect this last term. As a side remark, we will keep $A_0$, even if its approximate behavior is given by $A_0\approx e^{6 H t}(\Delta a)^2$ and, once $a$ is chosen as time and its fluctuations requested to be vanishing, it will be quite small. In summary, by dropping the linear term $A_1=0$, relation is a quadratic equation for $\pi_a^2$, which can readily be solved, $$\label{quadraticeqpia}
\pi_a^2=\frac{1}{2G}\left(
A_2\pm \sqrt{A_2^2+4 A_0}
\right).$$
In order to choose the correct sign inside the square root one can replace this solution, along with the form for the correlations $\Delta(X\pi_a)$, obtained from –, in and solve that relation for the fluctuation $(\Delta\pi_a)^2$. It can be checked that, by choosing the minus sign in the expression above, and linearizing the result in moments, the fluctuation $(\Delta\pi_a)^2$ turns out to be divergent; whereas for the plus sign the form of this fluctuation is given by an expression linear in other second-order moments. Therefore, in order to be consistent with the semiclassical expansion performed in this formalism, it is necessary to choose the plus sign. In addition, the global sign that comes from taking the square root of both sides in is chosen as negative in order to get below a positive physical Hamiltonian, which will be defined as $-\pi_a$. In this way, one ends up with the solution, $$\pi_a=-\frac{1}{\sqrt{2G}}\left(
A_2+ \sqrt{A_2^2+4 A_0}
\right)^{1/2} =: -{\cal H}_a.\label{quantumhamiltonian}$$
At this point we have already solved the constraints for the moments related to the momentum $\pi_a$, but we have not imposed any gauge and thus there is still some freedom left. In order to discuss and fix this gauge freedom, let us rewrite the constraints – in the following form: $$\begin{aligned}
C_1 &:= & \pi_a + \mathcal{H}_a = 0,\\
C_2 &:= & \Delta(\pi_a a) - f_2 = 0,\\\label{constraintc3}
C_3 &:= & (\Delta \pi_a)^2 - f_3 = 0,\\
C_4 &:= & \Delta(\pi_a \pi_\phi) - f_4 = 0,\\
C_5 &:= & \Delta(\pi_a \phi) - f_5 = 0,\\
C_6 &:= & \Delta(\pi_a \pi_\textbf{k}) - f_6 = 0,\\\label{constraintc7}
C_7 &:= & \Delta(\pi_a v_\textbf{k}) - f_7 = 0,\end{aligned}$$ where the functions $f_i$ only depend on the expectation values and moments not related with $\pi_a$. Note that for writing $C_1$ a fourth-order polynomial equation has been solved and the corresponding solution chosen as explained above. Nonetheless, the rest of the constraints $C_2-C_7$ corresponds to linear combinations of constraints – and are thus completely equivalent to those. Since that system of constraints was proven to be first-class up to linear order with respect to second-order quantum moments, the same applies to the last system $C_1-C_7$.
Therefore, we have a system of 7 first-class constraints, and following the standard approach to constraint systems, the generalized Hamiltonian is given as $$\label{totalham}
C_{total} = \sum_{i=1}^7 \mu_i C_i = 0,$$ with arbitrary Lagrange multipliers $\mu_i$. Now, we need to impose a gauge and choose the Lagrange multipliers $\mu_i$ that keeps the gauge conditions invariant.
The above constraints are meant to be solved for the 7 moments related to $\pi_a$, but there are still other 6 variables related to the degree of freedom $(a,\pi_a)$; namely the expectation value $a$ itself and its fluctuation $(\Delta a)^2$, as well as its correlations $\Delta(a \pi_\phi)$, $\Delta(a \phi)$, $\Delta(a \pi_\textbf{k})$, $\Delta(a v_\textbf{k})$. These are the 6 gauge parameters of our system, and in order to reduce them just to one, it is necessary to impose 5 gauge conditions. Since we want to interpret $a$ as time, it is natural to request that it is a parameter with vanishing fluctuations and correlations. Therefore, the most natural gauge choice is given by the conditions, $$\begin{aligned}
\label{gaugefixing1}
\phi_1 &=& (\Delta a)^2=0,\\
\phi_2 &=& \Delta(a \pi_\phi) =0,\\
\phi_3 &=& \Delta(a \phi) =0,\\
\phi_4 &=& \Delta(a \pi_\textbf{k})=0,\\\label{gaugefixinglast}
\phi_5 &=& \Delta(a v_\textbf{k})=0.\end{aligned}$$ It can be shown that, once constraints and gauge conditions are imposed, the gauge conditions $\phi_i$ commute among themselves and with the physical variables (expectation values and all moments that do not involve any $\pi_a$ or $a$). Therefore, they also commute with $C_1$, so that it remains first class. In addition, the constraint $C_2$ also remains first class but it acts trivially on the physical variables, so it does not generate any flow on the surface defined by the constraints and gauge conditions. This means that the Hamiltonian $\eqref{totalham}$ with Lagrange multipliers $\mu_1=1$ and $\mu_i=0$ for $i=2,...,7$, conserves these gauge conditions throughout evolution.
Furthermore, in the Appendix B we show that taking the constraints $C_3-C_7$ and the gauge-fixing conditions – as a new whole set of constraints, their commutation matrix that is formed by their Poisson brackets and it is necessary to construct the Dirac brackets, is invertible in the general case. This proves the adequacy of the chosen gauge.
In summary, we end up with the Hamiltonian $(\pi_a+{\cal H}_a)$, where gauge conditions – should be imposed. Furthermore, $a$ plays the role of the evolution parameter and the physical variables are the 4 expectation values $\phi$, $\pi_\phi$, $v_\textbf{k}$ and $\pi_\textbf{k}$, together with their 4 fluctuations and the 6 correlations between them. In fact, since these physical variables commute with $\pi_a$ we can define the object ${\cal H}_a$ as the physical Hamiltonian of our system.
Power spectra
-------------
The power spectrum for the variable $v$ can be defined as, $$P_v(k)=\frac{k^3}{2\pi^2}(\Delta v_\textbf{k})^2.\label{powerspectrumfluctuation}$$ Nonetheless, since this quantity is not observed directly at the CMB, it is convenient to write the power spectra of quantities related to measurable magnitudes. On the one hand, scalar perturbations are present in temperature anisotropies on the CMB. They are related to the comoving curvature perturbation $$\mathcal{R}:= \frac{H}{\phi'}v_\textbf{k}.$$ Therefore, from (\[powerspectrumfluctuation\]), the power spectrum of scalar perturbations can be written as $$P_S(k) := P_\mathcal{R}(k) = \frac{H^2}{\phi'^2} P_v(k) = \frac{H^2}{\phi'^2} \frac{k^3}{2\pi^2}(\Delta v_\textbf{k})^2
= \frac{3 G}{2 \epsilon}\frac{k^3}{\pi^2 a^2}(\Delta v_\textbf{k})^2.\label{powerspectrumscalar}$$ In the last equality we have used the relation between the slow-roll parameter $\epsilon$ and the derivative of the scalar field $\phi$ that can be obtained from Friedmann equations. Tensor perturbations, on the other hand, are expressed with the quantity $$h_\textbf{k} := \frac{2\sqrt{3G}}{a} v_\textbf{k}.$$ Due to the fact that tensor perturbations have two polarizations, the tensorial power spectrum will have a factor of 2 with respect to the power spectrum of $h_\textbf{k}$, $$P_T(k):= 2P_h(k) = 2 \frac{12 G}{ a^2} P_v(k) = 12 \frac{G k^3}{\pi^2 a^2}(\Delta v_\textbf{k})^2.\label{powerspectrumtensor}$$ These power spectra will be evaluated at late times ($|k\eta| \rightarrow 0$ or equivalently $a\rightarrow \infty$) [@Kin05]. As can be seen both scalar and tensorial power spectra are proportional to the dimensionless ratio, $$\label{ratio}
{\cal P}:=\frac{G k^3 (\Delta v_\textbf{k})^2}{a^2}.$$ Therefore, in the next section we will analyze the behavior of this object for the case of a de Sitter universe.
de Sitter universe
==================
Classical de Sitter background
------------------------------
Since we are interested in computing quantum corrections to the de Sitter power spectra, we will first obtain the result in the approximation of QFT on classical backgrounds. This is a well-known result but, for clarity, we think it is interesting to make it explicit in terms of this moment formalism.
The de Sitter universe is characterized by a constant value of the Hubble factor $H$. This universe can be recovered from the model presented above just by imposing a constant value for the potential $V(\phi)=H^2/(2 G)$, which fixes the energy of inflation. The evolution of the scale factor in terms of the conformal time $\eta$ is given by $$\label{desittera}
a(\eta) = \frac{-1}{H\eta}.$$
Since in the de Sitter case the slow-roll parameter $\epsilon$ vanishes, the frequency of the scalar sector coincides with the one of the tensorial sector , and takes the following form, $$\omega_S^2 = \omega_T^2 = \frac{a''}{a} = k^2 - \frac{2}{\eta^2} = k^2 \left(1 - \frac{2}{\xi}\right),$$ where we have introduced the dimensionless time $\xi:=-k\eta$ in order to simplify expressions below. By replacing this form of the frequency in the equation of the fluctuation $(\Delta v_\textbf{k})^2$ , one gets $$\xi^3 \frac{d^3 (\Delta v_\textbf{k})^2}{d\xi^3} + 4\xi (\xi^2-2) \frac{d (\Delta v_\textbf{k})^2}{d\xi}+8 (\Delta v_\textbf{k})^2 = 0.\label{deltav2desitter}$$ This equation can be solved analytically, which yields the following form of the fluctuation of the perturbative master variable, $$(\Delta v_\textbf{k})^2= \xi \left(A (J_{3/2}(\xi))^2 + B J_{3/2}(\xi)Y_{3/2}(\xi) + C (Y_{3/2}(\xi) )^2\right), \label{generalsoldesitter}$$ where $J_{3/2}$ and $Y_{3/2}$ are the first- and second-type Bessel functions respectively, which can be written in terms of elementary trigonometric functions, $$J_{3/2}(\xi) = \sqrt{\frac{2}{\pi\xi}} \left(\frac {\sin(\xi)}{\xi} - \cos(\xi)\right),$$ $$Y_{3/2}(\xi) = -\sqrt{\frac{2}{\pi \xi}} \left( \frac{\cos(\xi)}{\xi} + \sin(\xi)\right).$$
In order to fix the integration constants $A$, $B$ and $C$, we consider the Bunch-Davies vacuum – as initial condition which, as explained above, corresponds to the state with the minimum energy at the beginning of inflation ($\xi \rightarrow \infty$). With this initial state, the evolution of the three second-order moments is then given by $$\begin{aligned}
(\Delta v_\textbf{k})^2 (\xi) &=&
\frac{\hbar}{2k} \frac{1+\xi^2}{\xi^2},\label{classicalms1}
\\\label{classicalms2}
(\Delta \pi_\textbf{k})^2(\xi) &=&
\frac{k \hbar}{2}\frac{\xi^4-\xi^2+1}{\xi^4},
\\\label{classicalms3}
\Delta (v_\textbf{k} \pi_\textbf{k})(\xi) &=& \frac{\hbar}{2\xi^3}.\end{aligned}$$ In particular, from these expressions, it can be explicitly seen that the uncertainty principle is saturated for all $\xi$. For later convenience, since in the quantum-background approach the time parameter will be the scale factor $a$, we reparametrize the fluctuation of the perturbative master variable by using the expression , $$\label{deltvaqft}
(\Delta v_\textbf{k})^2 (a) = \frac{\hbar a^2 H^2}{2k^3} + \frac{\hbar}{2k}.$$ In the late-time limit $a \rightarrow \infty$, the second term above is then suppressed and the standard result for the power spectrum is obtained, $$\label{psqftdesitter}
{\cal P}=\frac{G k^3}{a^2}(\Delta v_\textbf{k})^2=\frac{\widetilde{H}^2}{2},$$ where we have defined the dimensionless Hubble parameter $\widetilde{H}:=l_p H$, with the reduced Planck length $l_p:=\sqrt{G\hbar}=(4\pi\hbar G_N/3)^{1/2}$. One of the main properties of this power spectrum is its scale invariance, since it does not depend on the wave number $k$.
Quantum de Sitter background
----------------------------
The quantum-gravity corrections for the power spectrum will be now analyzed, where the dynamics of the variables is governed by the physical Hamiltonian ${\cal H}_a$, as defined in , with gauge conditions –. The amount of physical variables is now 14: the 4 expectation values ($\phi$, $\pi_\phi$, $v_\textbf{k}$, $\pi_\textbf{k}$), their 4 fluctuations and the 6 correlations between each couple. The evolution is then given by the solutions of the system of 14 differential equations of the form $dX/da=\{X,\mathcal{H}_a\}$. In the general case the equations turn out to be quite complicate but, as will be shown below, in the de Sitter case it will be possible to find an approximate analytic solution.
As in the previous case we impose the de Sitter conditions $V=H^2/(2 G)$ and $\omega^2=k^2-2a^2H^2$. In addition, for the expectation values $v_{\bf k}$, $\pi_{\bf k}$ and $\pi_\phi$, we will assume that initially they take the same value as in the previous section, that is, $v_{\bf k}=0$, $\pi_{\bf k}=0$ and $\pi_\phi=0$. The first two values come from the stationary-state analysis discussed in Section \[moments\_classical\_back\], whereas the last one is motivated from the classical behavior of this variable. It turns out that our quantum equations of motion preserve those values. That is, once replaced all the commented conditions, one gets that at this order the expectation values $\phi$, $\pi_\phi$, $v_{\bf k}$, and $\pi_{\bf k}$ are constants of motion. Therefore, concerning these variables, the stationary point and the classical de Sitter behavior are respected. This is a nontrivial result, since the mentioned equations of motion might well have been modified by different moments in such a way that, for instance, ($v_{\bf k}=0$, $\pi_{\bf k}=0$) was not a stationary point of the system anymore.
In summary, once imposed the above conditions, out of the 14 equations of motion, 5 are trivial. The expectation values of the field $\phi$ and of the perturbative variables $v_{\bf k}$, as well as of their momenta $\pi_\phi$ and $\pi_{\bf k}$, are constants of motion: $$\begin{aligned}
\frac{d}{da} \phi&=&0,\\
\frac{d}{da} \pi_\phi&=&0,\\\label{v0constant}
\frac{d}{da} v_{\bf k}&=&0,\\
\frac{d}{da} \pi_{\bf k}&=&0.
\end{aligned}$$ Furthermore, the fluctuation of the momentum of the field is also conserved through evolution, $$\frac{d}{da} (\Delta \pi_\phi)^2=0.$$ Finally, the other 9 nontrivial equations take the following form: $$\begin{aligned}
\label{quantumeq1}
G{\cal H}_a \frac{d}{da}((\Delta \phi)^2)&=&\frac{2}{a^2} \Delta(\phi\pi_\phi)\\
G{\cal H}_a \frac{d}{da}(\Delta(\phi\pi_\phi))&=&\frac{1}{a^2}(\Delta \pi_\phi)^2,
\\
G{\cal H}_a \frac{d}{da}(\Delta(\phi v_{\bf k}))&=&\Delta(\phi \pi_{\bf k})
+\frac{1}{a^2}\Delta(\pi_\phi v_{\bf k}),
\\G{\cal H}_a
\frac{d}{da}(\Delta(\phi \pi_{\bf k}))&=&
\Delta(\phi v_{\bf k}) \left(2 a^2 H^2-k^2\right)+\frac{1}{a^2}\Delta(\pi_\phi \pi_{\bf k}),
\\
G{\cal H}_a \frac{d}{da}(\Delta(\pi_\phi v_{\bf k}) )&=&\Delta(\pi_\phi \pi_{\bf k} ),
\\\label{quantumeqlast}
G{\cal H}_a \frac{d}{da}(\Delta(\pi_\phi \pi_{\bf k}))&=&- \Delta(\pi_\phi v_{\bf k}) \left(k^2-2 a^2 H^2\right),
\\\label{quantumms1}
G{\cal H}_a \frac{d}{da}((\Delta v_{\bf k})^2)&=&2 \Delta(v_{\bf k}\pi_{\bf k}),\\\label{quantumms2}
G{\cal H}_a\frac{d}{da}((\Delta \pi_{\bf k})^2)&=&-2 \Delta(v_{\bf k}\pi_{\bf k}) \left(k^2-2 a^2 H^2\right),
\\\label{quantumms3}
G{\cal H}_a \frac{d}{da}(\Delta(v_{\bf k}\pi_{\bf k}))&=&(\Delta v_{\bf k})^2 \left(2 a^2 H^2-k^2\right)+(\Delta \pi_{\bf k})^2,\end{aligned}$$ with the physical Hamiltonian given by, $$\label{physicalham}
{\cal H}_a= \frac{1}{\sqrt{G}} \left[\frac{a^4}{G} H^2+(\Delta v_{\bf k})^2
\left(k^2-2 a^2 H^2\right)+(\Delta \pi_{\bf k})^2
+\frac{1}{a^2}(\Delta \pi_\phi)^2\right]^{1/2}.$$ This is a highly coupled and nonlinear system of differential equations, in particular due to the form of this physical Hamiltonian, and analytical solutions are extremely difficult to get. Looking at the equations, we note that it would be natural to define a time $\tau$, in such a way that $G{\cal H}_a d\tau= da$, so that the left-hand side of all equations would be just the derivative of the corresponding variable with respect to this time.
The limit of QFT on classical backgrounds is easily recovered from the system above just by dropping all moments that involve any background object. In particular, equations – are trivially obeyed in that limit. Furthermore, to obtain that limit, the physical Hamiltonian should also be taken as a classical variable (since it is the momentum $\pi_a$), so one should drop all moments from its definition, which leads to ${\cal H}_a=a^2 H/G$. In this case the natural time $\tau$ is the standard conformal time $\tau=\eta=-1/(H a)$ and equations – for the fluctuations of the perturbative variables reduce to their form –.
It is interesting to note that, since the fluctuation of the momentum of the field $(\Delta \pi_\phi)^2$, which appears inside the Hamiltonian, is a constant of motion, equations – for the moments related to the perturbative degree of freedom $(v_{\bf k},\pi_{\bf k})$ decouple from the rest. In particular, it is possible to write the generalization of equation for the fluctuation of the perturbative master variable as follows, $$\label{quantumeqdeltav}
{\cal H}_a^2 \frac{d^3(\Delta v_{\bf k})^2}{da^3} +3
{\cal H}_a \frac{{d\cal H}_a}{da} \frac{d^2(\Delta v_{\bf k})^2}{da^2}+
\left[4 \frac{\omega^2}{G^2}+{\cal H}_a
\frac{{d^2\cal H}_a}{da^2} +\left( \frac{{d\cal H}_a}{da} \right)^2\right]\frac{d(\Delta v_{\bf k})^2}{da} -\frac{8}{G^2} a H^2 (\Delta v_{\bf k})^2=0.$$ Nevertheless, contrary to the QFT version of this equation , this full-quantum equation cannot be solved independently from the rest, since it is still coupled to the fluctuation of the momentum $(\Delta \pi_{\bf k})^2$ and their correlation $\Delta(v_{\bf k}\pi_{\bf k})$ through the Hamiltonian ${\cal H}_a$. Finally, if the dependence of the Hamiltonian on $a$ was known, one could write that equation in terms of the time $\tau$ in a very compact way, $$\frac{d^3}{d\tau^3}(\Delta v_{\bf k})^2+4\omega\frac{d}{d\tau}(\omega(\Delta v_{\bf k})^2)=0.$$ Although this might seem a simple equation to deal with, all the complexity is inside the frequency $\omega=\omega(\tau)$ and analytic solutions are very hard to find.
### Approximate analytical solution for the power spectrum
Since one expects to find small changes to the result of the previous section, a first approximation to obtain the evolution of the fluctuation $(\Delta v_{\bf k})^2$ would be to assume the behavior derived in the approximation of QFT on classical backgrounds to compute the evolution of the physical Hamiltonian ${\cal H}_a(a)$. Plugging this form into will decouple this equation from the rest and one can then solve it to obtain the approximate dynamics of the fluctuation $(\Delta v_{\bf k})^2$.
Therefore, replacing the form of the moments - in the definition of the Hamiltonian one gets, $${\cal H}_a=
\frac{a^2 H}{G}\sqrt{1
+\frac{\widetilde{H}^2}{2 k^3} \left(2 \xi ^4-2 \xi^2-1\right)
+\frac{\widetilde{H}^4}{k^6} \sigma^2 \xi^6}.\label{physicalhamiltonapprox}$$ In order to absorb some of the fundamental constants in this expression we have made use of the dimensionless Hubble parameter $\widetilde{H}:=l_p H$ and the time parameter $\xi:= k/(aH)$, as already defined above; whereas the constant fluctuation of the momentum $(\Delta \pi_\phi)^2$ has been described by the quantity $\sigma$ as follows: $(\Delta \pi_\phi)^2=:G\hbar^2 \sigma^2$. In this way, it is shown very explicitly that $\widetilde{H}$ is the parameter that measures the strength of quantum-gravity effects. Since this parameter is known to be quite small [@Ade:2015xua], one can perform a power expansion of the square root above and write the Hamiltonian in the following way, $$\begin{aligned}
\label{expham}
{\cal H}_a&=&\frac{a^2 H}{G}\sum_{n=0}^\infty\left( \frac{\widetilde H^2}{k^3}\right)^n f_n(\xi)
\\\nonumber
&= &
\frac{a^2 H}{G}\left[ 1 +\frac{\widetilde{H}^2}{4 k^3} \left(2 \xi ^4-2 \xi ^2-1\right)
-\frac{\widetilde{H}^4}{32 k^6} \left[4 \xi ^8-8 \xi ^6 \left(2 \sigma ^2+1\right)+4 \xi^2+1\right]+{\cal O}\left(\frac{\widetilde H^6}{k^9}\right)\right],\end{aligned}$$ where the series only contains even powers of $\widetilde{H}/k^{3/2}$. It is remarkable that the quantity $\sigma$ only enters at order $\widetilde H^4$, which means that the actual state of the fluctuations of the matter field is not the most dominant part among the quantum-gravity corrections.
One can then assume a similar power expansion for the fluctuation $(\Delta v_{\bf k})^2$, around its value in the QFT approximation , as follows $$\label{expv}
(\Delta v_{\bf k})^2=\frac{\hbar}{k}\left[\frac{\xi^2+1}{2 \xi ^2} +\sum_{n=1}^\infty\left( \frac{\widetilde H^2}{k^3}\right)^n\delta_n(\xi) \right],$$ with unknown functions $\delta_n(\xi)$. Replacing the last two expressions – in equation , and taking into account the definition of the dimensionless time $\xi=k/(a H)$, one can solve for the different functions $\delta_n(\xi)$ iteratively at each order in $\widetilde H^2$.
The power spectrum can be computed just by replacing the above form of $(\Delta v_{\bf k})^2$ in its definition and it takes the form $$\label{psexpansion}
{\cal P}=\frac{G k^3}{ a^2}(\Delta v_{\bf k})^2=\frac{\widetilde{H}^2}{2}\left[1+\xi^2+2 \sum_{n=1}^\infty\left(\frac{\widetilde H^2}{k^3}\right)^n \xi^2\delta_n(\xi)\right].$$ In order to obtain its exact numerical value, the different $\delta_n$ functions must be obtained as indicated above. Nonetheless, from this expression one can already see the explicit dependence of the power spectrum on the different parameters, like the Hubble factor $H$ and the wave number $k$. In particular we see that the quantum-gravity corrections are given by a power series in the parameter $\widetilde H^2/k^3$, which makes it scale dependent and more relevant for large scales, which are related with small values of the mode number $k$. We would like to emphasize that the very same form for the first term of this series has been obtained in several other studies with very different approximation methods, see for instance [@brizuelakieferkramer1; @kieferkramer; @KTV13].
In order to know whether these correction terms imply an enhancement or a suppression of power, we need to study the behavior of the $\delta_n$ functions in the asymptotic limit of super-Hubble scales. Since the expansion parameter is very small, we will just focus on the first correction term given by $\delta_1$, which obeys the following equation, $$\xi ^5 \delta_1'''(\xi )+4 \xi ^3 (\xi^2 -2) \delta_1'(\xi )+8 \xi ^2 \delta_1(\xi )+4 \xi ^2+6=0.$$ The analytical solution of this equation is obtained and the constants are fixed by requesting that the solution is not oscillating at the beginning of inflation ($\xi\rightarrow \infty$). In this way, we find the following form for the first corrective term to the evolution of the fluctuation of the perturbative variable $(\Delta v_{\bf k})^2$: $$\delta_1\!=\!\frac{1}{12\xi^2}\left\{11+
4 \text{Ci}(2 \xi ) \left[\left(\xi ^2-1\right) \cos (2 \xi )-2 \xi \sin (2
\xi )\right]\!-\!2\left[\left(\xi ^2-1\right) \sin (2 \xi )+2 \xi \cos (2 \xi ) \right] (\pi -2 \text{Si}(2 \xi ))\right\}\!,\label{delta1}$$ $\text{Si}$ and $\text{Ci}$ being respectively the sine and cosine integral functions. The asymptotic limit for super-Hubble scales ($\xi\rightarrow 0$) for the coefficient that appears multiplying the correction term of order $\widetilde H^2/k^3$ in is given by, $$\label{delta1sol}
2 \xi^2 \delta_1=\frac{1}{6} (11-4\gamma_E-4 \ln 2\xi)+{\cal O}(\xi),$$ $\gamma_E$ being the Euler-Mascheroni constant. This expression still contains certain time dependence in the last logarithmic term, which slowly diverges. In order to give an estimate, it is usual to evaluate this expression at horizon crossing ($\xi=1$), which gives the following numerical value, $$2 \xi^2 \delta_1\approx 0.986.$$ Hence, up to the considered order, the power spectrum takes the form $${\cal P}\approx\frac{\widetilde{H}^2}{2}\left[1+\frac{\widetilde{H}^2}{k^3}\right]+{\cal O}(\widetilde H^4),\label{spectrumanalytical}$$ where the above numerical value has been rounded up to one. This is one of the main results of this paper and implies an enhancement of the power spectrum for large scales. In fact, the numerical value of the corrections obtained in this analysis is in agreement with the one computed in [@brizuelakieferkramer1], where a very different semiclassical approximation scheme was used. This is certainly an important indication of the robustness of the presented result.
### Numerical analysis of the system
In order to show the validity of the approximate analytical solution obtained above, in this subsection we will solve numerically the full system of quantum equations –. In fact, as already commented above, since the fluctuation of the momentum of the field $(\Delta \pi_\phi)^2=G \hbar^2\sigma^2$ is a constant of motion, equations – for the moments related to the perturbative degree of freedom $(v_{\bf k},\pi_{\bf k})$ decouple from the rest. Since, in order to compute the power spectrum, we are interested in the behavior of the fluctuation $(\Delta v_{\bf k})^2$, it is enough to consider those three equations. For the rest of this section we will choose units with $\hbar=1$ and $G=1$.
The main objective is to compare the correction of the power spectrum for both the approximate analytical and numerical approaches. We thus define the relative correction between the quantum and classical background results as, $$\beta(a) = \frac{(\Delta v_\mathbf{k})^2_{QG}(a)-(\Delta v_\mathbf{k})^2_{QFT}(a)}{(\Delta v_\mathbf{k})^2_{QFT}(a)},$$ where $(\Delta v_\mathbf{k})^2_{QG}$ stands for the solution of the full quantum equations – (obtained either analytically as in – or numerically in this subsection) and $(\Delta v_\mathbf{k})^2_{QFT}$ corresponds to the solution of QFT on classical backgrounds. In fact, within the decomposition performed in the previous subsection, this object $\beta$ is related to the $\delta_n$ functions. More precisely $\beta$ is equal to the third-term inside the square brackets of equation , that provides the relative correction to the power spectrum, and its explicit form within the approximation considered there is given by, $$\label{approxbeta}
\beta\approx \frac{2\widetilde H^2}{k^3}\xi^2\delta_1,$$ with $\delta_1$ given by . In the following we will compare this result with the numerical solution of the full system.
One problem in the numerical approach is that the system – is singular at $a=0$. Thus that point can not be used to impose the initial conditions and there is no other preferred choice for the initial value of the scale factor $a_0>0$. We have therefore used two different methods to impose the initial state numerically. On the one hand, in the first method we have assumed the Bunch-Davies vacuum – at an initial value of the scale factor $a_0$. On the other hand, in the second method we have used the analytical expression for the quantum variables obtained above – (where the Bunch-Davies vacuum has been exactly implemented for $a_0 = 0$) as initial conditions for the numerical simulation at a finite value of $a_0$.
![The relative correction of the power spectrum $\beta$, as computed by the numerical solution of the system – and the approximate analytical solution –, for $H = 10^{-5}$, $\sigma = 1/10$, $k=1$ and an initial value of the scale factor $a_0 = 10^{-15}$. The numerical evolution shown in the plot on the left-hand side corresponds to the first method (an initial Bunch-Davies vacuum), whereas the numerical evolution shown on the right-hand side corresponds to the second method (initial state given by the approximate analytical solution). In both cases it can be observed that for large scales the numerical evolution of $\beta$ is qualitatively the same as the evolution given by the approximate solution. In fact, their difference is small and approximately constant in time. []{data-label="relativeerror"}](relativeerror1.png "fig:") ![The relative correction of the power spectrum $\beta$, as computed by the numerical solution of the system – and the approximate analytical solution –, for $H = 10^{-5}$, $\sigma = 1/10$, $k=1$ and an initial value of the scale factor $a_0 = 10^{-15}$. The numerical evolution shown in the plot on the left-hand side corresponds to the first method (an initial Bunch-Davies vacuum), whereas the numerical evolution shown on the right-hand side corresponds to the second method (initial state given by the approximate analytical solution). In both cases it can be observed that for large scales the numerical evolution of $\beta$ is qualitatively the same as the evolution given by the approximate solution. In fact, their difference is small and approximately constant in time. []{data-label="relativeerror"}](relativeerror2.png "fig:")
In summary, for both methods, although the evolution is sensitive to the initial value of $a_0$, we observe that for $a_0 \ll 1$ the solutions are robust and have the same asymptotic behavior for large scales. In particular, we have performed tests with $a_0 = 10^{-10}$, $a_0 = 10^{-15}$ and $a_0 = 10^{-20}$, and the evolution of all variables remains invariant for large values of the scale factor. In addition, for both methods, the value of $\beta$ for large scales is approximately the same. Furthermore, for small values of the fluctuation of the field $\sigma\lessapprox 1$, the asymptotic value of the relative correction $\beta$ coincides with the one obtained with the approximate solution with high accuracy, which validates the analytical result presented in the previous subsection.
More precisely, in Fig. \[relativeerror\] the evolution of the relative correction $\beta$ as computed with the approximate analytical and with the numerical solution is shown. The plot on the left corresponds to the first method of imposing the initial conditions, and the plot on the right to the second one. In both cases $a_0 = 10^{-15}$, $k=1$, $\sigma=1/10$, and $H = 10^{-5}$ have been chosen. This value of the Hubble factor is in accordance with the experimental data [@Ade:2015xua]. In the numerical evolutions two different regimes can be clearly identified. First, for small values of $a$, the function $\beta$ is oscillatory. The difference between the first and the second method is more relevant in this regime, as the amplitude of the oscillations and their average value is smaller in the latter case. Then, at certain point, $\beta$ stops oscillating and it becomes a monotonically increasing function. In this regime both methods provide very similar results, with the same qualitative evolution as the one predicted by the approximate analytical solution. In this second regime, it is possible to see that the numerical value of $\beta$ differs slightly from the one obtained analytically, but this difference is small and it remains approximately constant in time. In the particular case shown in Fig. \[relativeerror\], at the beginning of this second regime (at around $a \approx 2\cdot10^5$) the relative difference between the numerically computed and analytically approximated $\beta$ is around $10\%$ in the left plot and $5\%$ in the right plot.
![The relative correction of the power spectrum $\beta$ as a function of $\sigma$, for $H = 10^{-5}$ and $k=1$, evaluated at $a=10^{30}$. Each plot shows different ranges of values for $\sigma$.[]{data-label="sigmaplot"}](sigmaplot1.png "fig:") ![The relative correction of the power spectrum $\beta$ as a function of $\sigma$, for $H = 10^{-5}$ and $k=1$, evaluated at $a=10^{30}$. Each plot shows different ranges of values for $\sigma$.[]{data-label="sigmaplot"}](sigmaplot2.png "fig:")
Finally, in the analytical discussion it has been observed that the dependence of the quantum-gravity corrections on $\sigma$, which describes the quantum state of the scalar field, are of order $\widetilde H^4$. Therefore, as long as $\sigma$ is not large, its effects should be negligible. In order to see if the numerical solution validates this conclusion, in Fig. \[sigmaplot\] the relative correction $\beta$ evaluated at a large value of $a=10^{30}$ is plotted as a function of $\sigma$. It can be seen that $\beta(\sigma)$ follows an oscillatory behavior, with average at $\beta(\sigma = 0)$, and a growing amplitude for larger values of $\sigma$. In particular, for $\sigma \simeq 10$, as shown in the plot on the right, the amplitude of $\beta$ is of the same order of magnitude as $\beta$ itself, and hence not negligible. However, for small $\sigma$, the change in $\beta$ due to different values of $\sigma$ is one or two orders of magnitude smaller than $\beta$ itself. Therefore, the conclusions derived with the analytical method are correct and, as long as the fluctuation of the momentum of the scalar field $\sigma$ is not very large, its actual value is not relevant to compute the quantum-gravity corrections of the power spectrum.
Conclusions
===========
In this work we have analyzed the behavior of the power spectra for both scalar and tensorial inflationary perturbations in two different quantization schemes. To that end, we have made use of a formalism that is based on a decomposition of the corresponding wave function in terms of its moments. On the one hand, the usual approximation of QFT on fixed backgrounds has been considered. In this case, the background degrees of freedom are treated classically and the wave function that describes the perturbative sector obeys a Schrödinger equation. The effective Hamiltonian that provides the evolution of the different moments is quadratic and thus different orders in moments decouple. In particular the power spectrum is related to the fluctuation of the master perturbative variable $(\Delta v_{\bf k})^2$, which is a second-order moment. Therefore, in this case a cutoff is not needed and the exact study of the dynamics at second order is enough to obtain the well-known result for the scale-invariant power spectrum for the de Sitter universe .
On the other hand, the Wheeler-DeWitt quantization has been implemented, where the background degrees of freedom have also been quantized. As opposed to the previous case, the main (Wheeler-DeWitt) equation is not an evolution equation, but rather a constraint equation. The information on the Wheeler-DeWitt equation is translated to an infinite set of constraint equations obeyed by the moments. In addition, the corresponding effective Hamiltonian is not quadratic in all the variables and different orders in moments do couple. Hence, for this scheme, it is necessary to apply a truncation. In our case we have truncated the system at second order in moments, which is assumed to be a good approximation for peaked semiclassical states. Once the truncation has been performed, one gets a finite system of (first-class) constraint equations –. We have then considered appropriate gauge-fixing conditions –, so that the scale factor plays the role of time, and hence (minus) its conjugate momentum is the physical Hamiltonian ${\cal H}_a$ . The evolution of this system has then been analyzed in detail for the case of a de Sitter universe. In particular, we have been able to obtain an analytical approximate solution for the evolution of the fluctuation of the master perturbative variable $(\Delta v_{\bf k})^2$, given by and , which has been validated by the subsequent numerical analysis.
In order to construct such approximate solution, we have considered the behavior given by the approximation of QFT on fixed backgrounds for the moments that appear inside the physical Hamiltonian. In this way, the third-order equation for the fluctuation of the perturbative variable decouples and can be solved separately from the rest. At this point, it has been observed that this approximate physical Hamiltonian (as well as the fluctuation of the perturbative variable) can be expanded as a power series on the parameter $l_p^2 H^2/k^{3}$. As expected, the order zero leads to the well-known result of QFT on classical backgrounds analyzed above, whereas the first order provides the main quantum-gravity corrections. After computing analytically the solution of the system, it has been derived that the sign of the corrections is positive, which implies an enhancement of power. Equation provides the form of the power spectrum, and it is the main result of the paper. It can be seen that the strength of the quantum-gravity corrections is tuned by the square of the dimensionless Hubble parameter $(l_p H)^2$ and, more importantly, that these corrections break the scale-invariance due to the $k^{-3}$ dependence, which makes them more relevant for large scales (small $k$). Remarkably, within this Wheeler-DeWitt quantization scheme, identical results have been obtained by making use of very different semiclassical approximation techniques [@brizuelakieferkramer1; @KTV13].
In the last subsection, by considering the numerical implementation of the full system –, we have verified the validity of the approximate solution. In particular it has been shown that, concerning the fluctuation of the perturbative variable $(\Delta v_{\bf k})^2$, for small values of the scale factor $a$ the approximate and numerical solutions have different behavior, as the former is constant through evolution whereas the latter one is oscillating. This issue might not be something fundamental, but rather related to the chosen initial conditions. Nonetheless, for large scales the numerical and approximate evolutions become qualitatively identical and very close in value. Since the relevant value of the power spectrum, that serves as a seed for the anisotropies measured at the CMB, is obtained at super-Hubble scales, this numerical approach shows the validity of the analytical solution discussed above.
Finally, another remarkable property that is derived from the approximate solution is that the dependence of the power spectrum on the actual state of the background matter degrees of freedom (the inflaton field $\phi$) is of the order of $(l_p^2 H^2\sigma)^2$, $\sigma^2$ being the dimensionless fluctuation of the momentum of the inflaton $(\Delta \pi_\phi)^2$, which is a constant of motion. This means that, as long as the value of $\sigma$ is not too large, the result is expected not to be much affected by the state of the matter. In fact, the explicit dependence of the power spectrum on $\sigma$ has been studied within the numerical setting and, as can be seen in Fig. \[sigmaplot\], this expectation has been confirmed.
Poisson brackets between quantum constraints
============================================
In this appendix we provide the Poisson brackets between the constraints defined in (\[quantumconstraint1\])-(\[quantumconstraint7\]) up to linear order in moments, or order $\hbar$. It turns out that all the brackets are a linear combination of constraints, which proves that all of them are first class. In principle, with these 7 constraints, 49 different Poisson brackets can be constructed $\{C_I,C_J\}$. However, due to the antisymmetry, 7 are identically zero, $\{C_I,C_I\}=0$, and from the other 42, they are related by pairs $\{C_I,C_J\} = -\{C_J,C_I\}$. Consequently, there are 21 independent nontrivial brackets, which take the following form: [$$\begin{aligned}
\{C,C_{\pi_a}\} &=& \frac{6}{a^4}\left(\pi_\phi^2+4a^6V(\phi)\right)C_a-\frac{4}{a^3}\pi_\phi C_{\pi_\phi}+8a^3V'(\phi)C_\phi, \\
\{C,C_a\}&=&2 G C_{\pi_a},\\
\{C,C_{\pi_\phi}\} &=& 8a^3V'(\phi)C_a + 2a^4V''(\phi)C_{\pi_\phi},\\
\{C,C_{\phi}\}&=&\frac{4\pi_\phi}{a^3}C_a - \frac{2}{a^2}C_{\pi_\phi},\\
\{C,C_{\pi_{\bf k}}\}&=&2\omega^2 C_{v_{\bf k}},\\
\{C,C_{v_{\bf k}}\} &=& -2C_{\pi_{\bf k}},\\
\{C_{\pi_a},C_a\} &=&4 G \pi_a C_{\pi_a} + \frac{4}{a^3}\left(\pi_\phi^2-4a^6V(\phi)\right)C_a + \frac{2\pi_\phi}{a^2}C_{\pi_\phi}-2a^4V'(\phi)C_\phi-2\pi_\mathbf{k} C_{\pi_{\bf k}}-2\omega^2 v_\mathbf{k} C_{v_{\bf k}},\\
\{C_{\pi_a},C_{\pi_\phi}\} &=& 2a^4 V'(\phi) C_{\pi_a} + \frac{2}{a^3}\left(\pi_\phi^2-4a^6V(\phi)\right) C_{\pi_\phi},
\\
\{C_{\pi_a},C_{\phi}\} &=& - \frac{2\pi_\phi}{a^2}C_{\pi_a}+\frac{2}{a^3}\left(\pi_\phi^2-4a^6V(\phi)\right)C_\phi, \\
\{C_{\pi_a},C_{\pi_{\bf k}}\}&=&2\omega^2 v_\mathbf{k} C_{\pi_a}+\frac{2}{a^3}\left(\pi_\phi^2-4a^6V(\phi)\right)C_{\pi_{\bf k}},\\
\{C_{\pi_a},C_{v_{\bf k}}\}&=&-2\pi_\mathbf{k} C_{\pi_a}+\frac{2}{a^3}\left(\pi_\phi^2-4a^6V(\phi)\right)C_{ v_{\bf k}},
\\
\{C_a,C_{\pi_\phi}\}&=&2a^4V'(\phi)C_a - 2 G \pi_aC_{\pi_\phi},\\
\{C_a,C_{\phi}\}&=&-\frac{2\pi_\phi}{a^2} C_a - 2 G\pi_aC_{\phi},\\
\{C_a,C_{\pi_{\bf k}}\}&=&2\omega^2 v_\mathbf{k} C_a - 2 G \pi_a C_{\pi_{\bf k}},\\
\{C_a,C_{v_{\bf k}}\}&=&-2\pi_\mathbf{k} C_a - 2 G \pi_aC_{v_{\bf k}},\\
\{C_{\pi_\phi},C_\phi\} &=& 2 G \pi_a C_{\pi_a} + \frac{2}{a^3}\left(\pi_\phi^2-4a^6V(\phi)\right)C_a + \frac{4\pi_\phi}{a^2}C_{\pi_\phi}-4a^4V'(\phi)C_\phi-2\pi_\mathbf{k} C_{\pi_{\bf k}}-2\omega^2 v_\mathbf{k} C_{v_{\bf k}},\\
\{C_{\pi_\phi},C_{\pi_{\bf k}}\} &=& 2\omega^2 v_\mathbf{k} C_{\pi_\phi}-2a^4V'(\phi)C_{\pi_{\bf k}},\\
\{C_{\pi_\phi},C_{v_{\bf k}}\} &=& 2\pi_\mathbf{k} C_{\pi_\phi}-2a^4V'(\phi)C_{ v_{\bf k}},\\
\{C_\phi,C_{\pi_v}\}&=&2\omega^2 v_\mathbf{k} C_\phi + \frac{2\pi_\phi}{a^2}C_{\pi_v},\\
\{C_\phi,C_{v_{\bf k}}\}&=&-2\pi_\mathbf{k} C_\phi + \frac{2\pi_\phi}{a^2}C_{v_{\bf k}},\\
\{C_{\pi_{\bf k}},C_{v_{\bf k}}\} &=& 2 G \pi_a C_{\pi_a} + \frac{2}{a^3}\left(\pi_\phi^2-4a^6V(\phi)\right)C_a + \frac{2\pi_\phi}{a^2}C_{\pi_\phi}-2a^4V'(\phi)C_\phi-4\pi_\mathbf{k} C_{\pi_{\bf k}}-4\omega^2 v_\mathbf{k} C_{v_{\bf k}}.\end{aligned}$$ ]{}
The Dirac brackets
==================
In this appendix we verify that the matrix that must be defined to construct the Dirac brackets when imposing the gauge conditions – on the first-class constraint system – is invertible in the general case. Let us define the whole set of constraints by $\Phi_I=\phi_I$, for $I=1,2,3,4,5$, and $\Phi_I=C_{I-3}$, for $I=6,7,8,9,10$. Then the Poisson bracket matrix is defined by $M_{IJ}:=\{\Phi_I,\Phi_J\}$, and its determinant is given by, $$\begin{aligned}
det(M)&=& -\frac{1}{64} \left\{\hbar^5+4 \hbar^3
\left[(\Delta(\pi_\textbf{k} v_\textbf{k}))^2+(\Delta(\pi_\phi \phi))^2-(\Delta \pi_\textbf{k})^2 (\Delta v_\textbf{k})^2+2 \Delta(\pi_\phi v_\textbf{k})
\Delta(\phi \pi_\textbf{k})
\right.\right. \\ &-&\left.\left.2 \Delta(\phi v_\textbf{k})
\Delta(\pi_\phi \pi_\textbf{k})
% \right.\right.\nonumber\\ &\left.\left.
- (\Delta \phi)^2(\Delta \pi_\phi)^2\right]
+16 \hbar
\left[(\Delta(\pi_\phi \phi))^2
(\Delta(\pi_\textbf{k} v_\textbf{k}))^2
\right.\right.\nonumber\\ &-& \left.\left.
(\Delta \phi)^2 (\Delta \pi_\phi)^2
(\Delta(\pi_\textbf{k} v_\textbf{k}))^2
+2 (\Delta \pi_\phi)^2 \Delta(\phi \pi_\textbf{k}) \Delta(\phi v_\textbf{k}) \Delta(\pi_\textbf{k} v_\textbf{k})
\right.\right.\nonumber\\ &-& \left.\left.
2
\Delta(\phi v_\textbf{k}) \Delta(\pi_\phi \pi_\textbf{k}) \Delta(\pi_\phi \phi)
\Delta(\pi_\textbf{k} v_\textbf{k})
+(\Delta(\phi v_\textbf{k}))^2
(\Delta(\pi_\phi \pi_\textbf{k}))^2
\right.\right.\nonumber\\ &-& \left.\left.
(\Delta \pi_\textbf{k})^2 (\Delta \pi_\phi)^2
(\Delta(\phi v_\textbf{k}))^2+(\Delta(\pi_\phi v_\textbf{k}))^2
\left((\Delta(\phi \pi_\textbf{k}))^2
-(\Delta \pi_\textbf{k})^2
(\Delta \phi)^2\right)
\right.\right.\nonumber\\ &-& \left.\left.
(\Delta v_\textbf{k})^2
\left((\Delta \phi)^2 (\Delta(\pi_\phi \pi_\textbf{k}))^2-2
\Delta(\phi \pi_\textbf{k}) \Delta(\pi_\phi \phi)
\Delta(\pi_\phi \pi_\textbf{k})
% \right.\right.\right.\nonumber\\ \left.\left.\left.
+(\Delta \pi_\textbf{k})^2(\Delta(\pi_\phi \phi))^2
\right.\right.\right.\nonumber\\ &+&\left.\left.\left.
(\Delta \pi_\phi)^2
\left((\Delta(\phi \pi_\textbf{k}))^2-(\Delta \pi_\textbf{k})^2
(\Delta \phi)^2\right)\right)
-2 \Delta(\pi_\phi v_\textbf{k})
((\Delta(\pi_\textbf{k} v_\textbf{k}) \Delta(\phi \pi_\textbf{k})
\right.\right.\nonumber\\ &-&\left.\left.
(\Delta v_\textbf{k})^2
\Delta(\phi v_\textbf{k})) \Delta(\pi_\phi \phi)+\Delta(\pi_\phi \pi_\textbf{k})
% \right.\right.\nonumber\\\nonumber & \left.\left.
(\Delta(\phi \pi_\textbf{k}) \Delta(\phi v_\textbf{k})-\Delta(\pi_\textbf{k} v_\textbf{k})
(\Delta \phi)^2))\right]\right\}^2.\end{aligned}$$ This is a complicate polynomial expression, which will be nonvanishing for generic values of the different moments. In this general case, the Dirac brackets can be properly defined and the chosen gauge is thus valid.
Nonetheless, this expression might be vanishing for certain special combination of moments. It is not possible to find all the roots of the determinant, but there is one particular and interesting case, for which this determinant vanishes. This happens for the case of a product state that saturates the Heisenberg uncertainty principle of every degree of freedom. In terms of the moments, this implies that the different degrees of freedom are independent, and thus the correlations $\Delta(\phi \pi_\textbf{k})$, $\Delta(\phi v_\textbf{k})$, $\Delta(\pi_\phi \pi_\textbf{k})$, $\Delta(\pi_\phi v_\textbf{k})$ are zero; and, in addition, the fluctuations and correlations of each degree of freedom must be related by, $$(\Delta \phi)^2(\Delta \pi_\phi)^2- \Delta (\phi \pi_\phi)^2=\frac{\hbar^2}{4},\qquad
(\Delta \phi_{\bf k})^2(\Delta v_{\bf k})^2- \Delta(\phi_{\bf k} v_{\bf k})^2=\frac{\hbar^2}{4}.$$ Nevertheless, in this case the gauge above is also valid since, it can be shown that the null space of the matrix $M$ acts trivially on the physical variables and thus does not generate any Poisson flow. The degeneracy of the matrix in this special case of saturation of the uncertainty principle seems to be a generic feature of this formalism, since it also happens for other models studied in the literature, see for instance [@BT09], [@BHT11]. In fact, in [@BT09] there is a typo and the determinant computed in Section 3.2. is also vanishing for saturation of the uncertainty relation, but this does not affect the general conclusion since the same argument as here applies for the validity of the imposed gauge.
Acknowledgments {#acknowledgments .unnumbered}
===============
The authors would like to thank M. Bojowald and A. Tsobanjan for correspondence, and L. Chataignier, C. Kiefer, and M. Krämer for comments on the manuscript. This work has been supported by Project FIS2017-85076-P (MINECO/AEI/FEDER, UE) and Basque Government Grant No. IT956-16.
[99]{} C. Kiefer, “Quantum Gravity”, Oxford University Press (2012).
P. A. R. Ade [*et al.*]{} \[Planck Collaboration\], “Planck 2015 results. XIII. Cosmological parameters”, Astron. Astrophys. [**594**]{} (2016) A13.
D. Baumann, “TASI lectures on inflation”, \[arXiv:0907.5424v2 \[hep-th\]\].
W. H. Kinney, “TASI lectures on inflation”, \[arXiv:0902.1529v2 \[astro-ph.CO\]\].
J. Martin, “Inflation and precision cosmology”, Braz. J. Phys. **34** (2004) 1307.
P. Peter and J.-P. Uzan, “Primordial cosmology”, Oxford University Press (2009).
D. Brizuela, C. Kiefer, and M. Krämer, “Quantum-gravitational effects on gauge-invariant scalar and tensor perturbations during in inflation: the de Sitter case”, Phys. Rev. D **93** (2016) 104035.
D. Brizuela, C. Kiefer, and M. Krämer, “Quantum-gravitational effects on gauge-invariant scalar and tensor perturbations during in inflation: the slow-roll approximation”, Phys. Rev. D **94** (2016) 123527.
D. Bini, G. Esposito, C. Kiefer, M. Krämer, and F. Pessina, “On the modification of the cosmic microwave background anisotropy spectrum from canonical quantum gravity”, Phys. Rev. D **87** (2013) 104008.
C. Kiefer and M. Krämer, “Quantum gravitational contributions to the cosmic microwave background anisotropy spectrum”, Phys. Rev. Lett. **108** (2012) 021301.
A. Y. Kamenshchik, A. Tronconi, and G. Venturi, “Inflation and quantum gravity in a Born-Oppenheimer context”, Phys. Lett. B **726** (2013) 518.
A. Y. Kamenshchik, A. Tronconi, and G. Venturi, “Signatures of quantum gravity in a Born-Oppenheimer context”, Phys. Lett. B **734** (2014) 72.
M. Bojowald and A. Tsobanjan, “Effective constraints for relativistic quantum systems”, Phys. Rev. D [**80**]{} (2009) 125008.
M. Bojowald and A. Tsobanjan, “Effective constraints and physical coherent states in quantum cosmology: a numerical comparison”, Class. Quant. Grav. [**27**]{} (2010) 145004.
M. Bojowald, B. Sandhofer, A. Skirzewski, and Artur Tsobanjan, “Effective constraints for quantum systems”, Rev. Math. Phys. [**21**]{} (2009) 111.
I. Agullo and N. A. Morris, “Detailed analysis of the predictions of loop quantum cosmology for the primordial power spectra”, Phys. Rev. D **92** (2015) 124040.
S. Schander, A. Barrau, B. Bolliet, J. Grain, L. Linsefors, and J. Mielczarek, “Primordial scalar power spectrum from the Euclidean big bounce”, Phys. Rev. D **93** (2016) 023531.
D. M. de Blas and J. Olmedo, “Primordial power spectra for scalar perturbations in loop quantum cosmology”, JCAP **1606** (2016) 029.
L. Castelló Gomar, D. M. de Blas, G. A. Mena Marugán, and J. Olmedo, “Hybrid loop quantum cosmology and predictions for the cosmic microwave background”, Phys. Rev. D **96** (2017) 103528.
I. Agullo, B. Bolliet, and V. Sreenath, “Non-Gaussianity in loop quantum cosmology”, Phys. Rev. D **97** (2018) 066021.
E. Alesci, A. Barrau, G. Botta, K. Martineau, and G. Stagno, “Phenomenology of quantum reduced loop gravity in the isotropic cosmological sector”, Phys. Rev. D **98** (2018) 106022.
J. Mielczarek, L. Linsefors, and A. Barrau, “Silent initial conditions for cosmological perturbations with a change of space-time signature”, Int. J. Mod. Phys. D **27** (2018) 1850050.
A. Barrau, P. Jamet, K. Martineau, and F. Moulin, “Scalar spectra of primordial perturbations in loop quantum cosmology”, Phys. Rev. D **98** (2018) 086003.
I. Agullo, “Primordial power spectrum from the Dapor-Liegener model of loop quantum cosmology”, Gen. Rel. Grav. **50** (2018) 91.
M. Bojowald, “Canonical gravity and applications”, Cambridge University Press (2011).
V. F. Mukhanov, H. A. Feldman, and R. H. Brandenberger, “Theory of cosmological perturbations”, Phys. Rep. **215** (1992) 203.
J. Martin, V. Vennin, and P. Peter, “Cosmological inflation and the quantum measurement problem”, Phys. Rev. D **86** (2012) 103524.
D. Brizuela, “Statistical moments for classical and quantum dynamics: formalism and generalized uncertainty relations”, Phys. Rev. D **90** (2014) 085027.
C. Armendariz-Picon, “Why should primordial perturbations be in a vacuum state?”, JCAP [**02**]{} (2007) 031.
J. Ganc, “Calculating the local-type fNL for slow-roll inflation with a non-vacuum initial state”, Phys.Rev. D [**84**]{} (2011) 063514.
A. Aravind, D. Lorshbough, and S. Paban, “Non-Gaussianity from excited initial inflationary states”, JHEP [**07**]{} (2013) 076.
T. S. Bunch and P. C. W. Davies, “Quantum field theory in de Sitter space: renormalization by point splitting”, Proc. Roy. Soc. Lond. **A360** (1978) 117.
M. Bojowald, P. A. Hoehn, and A. Tsobanjan, “An effective approach to the problem of time”, Class. Quant. Grav. [**28**]{} (2011) 035006.
M. Bojowald, P. A. Hoehn, and A. Tsobanjan, “Effective approach to the problem of time: general features and examples”, Phys. Rev. D [**83**]{} (2011) 125023.
P. A. Hoehn, E. Kubalova, and A. Tsobanjan, “Effective relational dynamics of a nonintegrable cosmological model”, Phys. Rev. D [**86**]{} (2012) 065014.
M. Bojowald and T. Halnon, “Time in quantum cosmology”, Phys. Rev. D [**98**]{} (2018) 066001.
W. H. Kinney, “Horizon crossing and inflation with large $\eta$”, Phys. Rev. D **72** (2005) 023515.
[^1]: E-mail address: [email protected]
[^2]: E-mail address: [email protected]
[^3]: Since the time variable $\eta$ has been rescaled above to absorb the volume of the universe $L^3$, one also needs to rescale the perturbative variable $v_{\bf k}$ (for both scalar and tensorial sectors) and the wave number $k$ as $v_{\bf k}\rightarrow L^2 v_{\bf k}$ and $k\rightarrow k/L$ so that $L$ does not appear in these expressions. From this point on, all the variables are rescaled as commented. In this way, $\eta$ and $k$ are dimensionless, whereas $a$ has dimensions of length and $v_{\bf k}^2$ has dimensions of action. For a more explicit discussion about this rescaling transformation see, e.g., Section III of [@brizuelakieferkramer1].
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We present a manifestly Lorentz-covariant description of the phase space of general relativity with the Immirzi parameter. This formulation emerges after solving the second-class constraints arising in the canonical analysis of the Holst action. We show that the new canonical variables give rise to other Lorentz-covariant parametrizations of the phase space via canonical transformations. The resulting form of the first-class constraints in terms of new variables is given. In the time gauge, these variables and the constraints become those found by Barbero.'
author:
- Merced Montesinos
- Jorge Romero
- Mariano Celada
bibliography:
- 'references.bib'
title: 'Manifestly Lorentz-covariant variables for the phase space of general relativity'
---
Introduction
============
Loop quantum gravity [@rovelli2004quantum; @AshLewcqg2115; @thiemann2007modern; @RovelliLR; @*Rovellicag2815] is one of the most promising candidates to successfully achieve a nonpertubative and background-independent quantization of the gravitational field. It is based on the Ashtekar-Barbero variables for general relativity [@Barbero], which emerge from the canonical analysis–with a partial gauge fixing–of the Holst action [@Holst]. Although this action gives rise to the same gravitational dynamics dictated by Einstein’s equations, it contains a free parameter[^1] that turns out to be meaningful at the quantum level, since it shows up in the spectra of quantum observables [@RoveSmolinNucP442; @*AshtLewacqg14] and in the black hole entropy [@RovPRLt.77.3288; @*AshBaezcqg105; @*Meisscqg21; @*Agulloetcprl100; @*EngleNuiPRL105]. It is believed that the presence of the so-called Immirzi parameter [@Immirzicqg1410] may be due to the fact that the Ashtekar-Barbero variables are obtained through the use of the time gauge (a different gauge fixing has been recently studied in [@Liu_Noui]), which breaks the Lorentz invariance down to rotational invariance in order to simplify the construction of the associated quantum theory. Because of this, there have been several attempts to construct a Lorentz-covariant canonical description of the phase space of general relativity seeking to resolve the Immirzi ambiguity [@Alexcqg1720; @*AlexVassprd644; @*AlexLivprd674; @Cianfraniprl; @NouiSIGMA72011; @*Nouiprd84044002] (see also [@Nouiprd9110] for a lower-dimensional model).
The Lorentz-covariant canonical analysis of general relativity features the presence of second-class constraints. They can be equivalently dealt with either by introducing the Dirac bracket [@Alexcqg1720; @*AlexVassprd644; @*AlexLivprd674] or by solving them in an explicit manner [@Barros; @NouiSIGMA72011; @*Nouiprd84044002; @CelMontesRom]. In this paper we follow the latter direction. It is worth mentioning that, although the approach derived in [@Barros; @CelMontesRom] certainly is Lorentz covariant, it is not manifestly Lorentz covariant, since it splits the Lorentz group into boosts and rotations. Given that we would like to maintain untouched the classical symmetries of general relativity as much as possible, we ask whether it is plausible to solve the second-class constraints in a manifestly Lorentz-covariant fashion. The answer given in this paper is in the affirmative, showing that we can describe the phase space of general relativity using several canonical pairs, one of which is made up of Lorentz vectors. It turns out that the different canonical pairs are related to one another by canonical transformations that can be regarded as the Lorentz-covariant generalization of Barbero’s canonical transformation. As expected, the canonical variables found in this paper lead to the Ashtekar-Barbero variables in the time gauge. For that reason, the new canonical variables reported here certainly constitute a Lorentz-covariant extension of the Ashtekar-Barbero variables.
Hamiltonian action
==================
Our notation is as follows. Internal (Lorentz) indices are denoted by $I,J,\dots$, and take the values $\{0,i\}$, where $i=1,2,3$. Likewise, $a,b,\dots$ label spatial coordinates. The internal indices are raised and lowered with the metric $\eta_{IJ}=\rm{diag}(\sigma,1,1,1)$, where $\sigma=-1\ (=+1)$ for Lorentzian (Euclidean) signature. The internal group, denoted by SO$(\sigma)$, corresponds to the Lorentz group SO$(1,3)$ for $\sigma=-1$ or the rotation group SO(4) for $\sigma=+1$. The weight of a tensor is sometimes indicated with the presence of a tilde over or below it, whereas the time derivative is represented by a dot over the corresponding variable. The internal tensor $\epsilon_{IJKL}$ is totally antisymmetric and such that $\epsilon_{0123}=+1$. Similarly, the spatial tensor density $\underaccent{\tilde}{\eta}_{abc}$ ($\tilde{\eta}^{abc}$) is totally antisymmetric and satisfies $\underaccent{\tilde}{\eta}_{123}=1$ ($\tilde{\eta}^{123}=+1$). The symmetrizer and antisymmetrizer are defined correspondingly by $A_{(\alpha\beta)}:=(A_{\alpha\beta}+A_{\beta\alpha})/2$ and $A_{[\alpha\beta]}:=(A_{\alpha\beta}-A_{\beta\alpha})/2$. Furthermore, for any antisymmetric quantity $A^{IJ}$ we define its internal (Hodge) dual as $\star A^{IJ}:=(1/2)\epsilon^{IJ}{}_{KL}A^{KL}$ and also the corresponding $\gamma$-valued quantity $\stackrel{(\gamma)}{A} {}^{IJ}:=P^{IJ}{}_{KL}A^{KL}=A^{IJ}+(1/\gamma)\star A^{IJ}$, where $\gamma$ is the Immirzi parameter. In the first-order formalism, general relativity with the Immirzi parameter can be described either by the Holst action or by a $BF$-type action supplemented with constraints [@CMPR2001; @*cqgrevBF] (the $BF$-type formulations play a fundamental role in the spin-foam approach to quantum gravity [@alexandrov2012spin; @*perez2013]). By performing the $3+1$ decomposition of the Holst action in an SO($\sigma$)-covariant fashion [@Barros] (we assume that the spacetime has a topology $\mathbb{R}\times \Omega$, with $\Omega$ a spacelike three-dimensional manifold without a boundary), it takes the simplified form $$\label{Act3+1}
S=\int_\mathbb{R} dt\int_\Omega d^3x\left(\stackrel{(\gamma)}{\tilde{\Pi}} {}^{aIJ}\dot{\omega}_{aIJ}-\tilde{H}\right),$$ where $(\omega_{aIJ},\stackrel{(\gamma)}{\tilde{\Pi}}{}^{aIJ})$ are canonical coordinates[^2] and $\tilde{H}$ is the Hamiltonian (density), which is given by $$\label{Hamact}
\tilde{H}=\underaccent{\tilde}{N}\tilde{\tilde{\mathcal{H}}} + N^a\tilde{\mathcal{V}}_a + \xi_{IJ}\tilde{\mathcal{G}}^{IJ}+\underaccent{\tilde}{\varphi}{}_{ab}\tilde{\tilde{\Phi}}{}^{ab}+\psi_{ab}\Psi^{ab}.$$ Here $\underaccent{\tilde}{N},N^a,\xi_{IJ},\underaccent{\tilde}{\varphi}{}_{ab}$, and $\psi_{ab}$ (of weight -2, so that $\Psi^{ab}$ has weight +3) are Lagrange multipliers imposing the constraints
$$\begin{aligned}
&&\tilde{\mathcal{G}}^{IJ}:= D_a \stackrel{(\gamma)}{\tilde{\Pi}}\!\!{}^{aIJ} \approx 0, \label{Gauss}\\
&&\tilde{\mathcal{V}}_a := \frac{1}{2}\tilde{\Pi}^{bIJ}\stackrel{(\gamma)}{F}\!\!{}{_{baIJ}} \approx 0,\label{Vector} \\
&&\tilde{\tilde{\mathcal{H}}}:=\frac{1}{2} \tilde{\Pi}^{aIK}\tilde{\Pi}^b{}_K{}^J\stackrel{(\gamma)}{F}\!\!{}{_{abIJ}} + \sigma\Lambda g \approx 0 ,\label{Scalar}\\
&&\tilde{\tilde{\Phi}}^{ab} := -2\sigma\star\tilde{\Pi}^a{}_{IJ}\tilde{\Pi}^{bIJ} \approx 0, \label{phi}\\
&&\Psi^{ab} := \epsilon_{IJKL}\tilde{\Pi}^{(a|IM}\tilde{\Pi}^c{}_M{}^JD_c \tilde{\Pi}^{|b)KL}\approx 0\label{psi}, \end{aligned}$$
where $F_{abIJ}:=2\left(\partial_{[a}\omega_{b]IJ}+\omega_{[a|IK}\omega_{|b]}{}^K{}_J\right)$ is the curvature of $\omega_{aIJ}$, $D_a$ is the spatial component of the SO$(\sigma)$-covariant derivative, $g:=\det(g_{ab})$ is the determinant of the spatial metric $g_{ab}$ (the induced metric on $\Omega$) whose inverse is defined by $gg^{ab}:=(\sigma/2)\tilde{\Pi}^{aIJ}\tilde{\Pi}^{b}{}_{IJ}$, and $\Lambda$ is the cosmological constant. As a result, the constraints $\tilde{\mathcal{G}}^{IJ}$, $\tilde{\mathcal{V}}_a$, and $\tilde{\tilde{\mathcal{H}}}$ (they are called Gauss, vector, and scalar constraints, respectively) are first class, and generate the gauge symmetries of the theory. Sometimes, instead of $\tilde{\mathcal{V}}_a$, we consider the diffeomorphism constraint $\tilde{\mathcal{D}}_a:=\tilde{\mathcal{V}}_a+(1/2)\omega_{aIJ}\tilde{\mathcal{G}}^{IJ}$. On the other hand, the constraints $\tilde{\tilde{\Phi}}^{ab}$ and $\Psi^{ab}$ are second class. These constraints arise from the implementation of the Dirac procedure and are necessary to obtain the correct physical degrees of freedom of general relativity. It is worth realizing that, although the expressions for the constraints (\[Scalar\]) and (\[psi\]) differ from the corresponding ones reported in [@CelMontcqg2920], they are actually equivalent to them, since we have the relation $\tilde{\eta}^{abc}g_{cd}\tilde{\Pi}^d{}_{IJ}=\pm \sigma\epsilon_{IJKL}\tilde{\Pi}^{aKM}\tilde{\Pi}^b{}_M{}^L$ when the constraint (\[phi\]) holds, the sign having to do with the sign ambiguity in the solution of the simplicity constraint in the $BF$ formalism.
Solution of the second-class constraints
========================================
We now solve the second-class constraints, and we proceed in such a way that we keep the explicit SO($\sigma$) covariance of the theory. The constraint (\[phi\]) amounts to a total of six restrictions on the 18 variables $\tilde{\Pi}^{aIJ}$. Its solution is then given in terms of 12 independent variables $\tilde{B}^{aI}$ as (see [@ashtekar1991lectures; @peld1994115]) $$\label{sol2nd}
\tilde{\Pi}^{aIJ}=\epsilon\tilde{B}^{a[I}m^{J]},$$ where $\epsilon=\pm1$ (since the constraint is quadratic in $\tilde{\Pi}^{aIJ}$) and $$\label{intm}
m_I:=\frac{1}{6\sqrt{h}}\epsilon_{IJKL}\underaccent{\tilde}{\eta}_{abc}\tilde{B}^{aJ}\tilde{B}^{bK}\tilde{B}^{cL}$$ for $h:=\det(h^{ab})$, with $h^{ab}:=\tilde{B}^{aI}\tilde{B}^b{}_{I}$. Notice that $m^I$ satisfies the identities $m_Im^I=\sigma$ and $\tilde{B}^{aI}m_I=0$. Moreover, we have the relation $4gg^{ab}=h^{ab}$ (thus, $h^{ab}$ can be regarded as the densitized metric). From now on, the inverse of $h^{ab}$, of weight -2, is denoted $h_{ab}$. This then implies the important relation $q^I{}_J:=h_{ab}\tilde{B}^{aI}\tilde{B}^b{}_J=\delta^I_J-\sigma m^Im_J$, which embodies the projector on the orthogonal plane to $m^I$.
It remains to solve the constraint (\[psi\]). This constraint imposes six restrictions on the 18 components $\omega_{aIJ}$, meaning that the general solution of (\[psi\]) takes the form $$\label{wMN}
\omega_{aIJ}= M_a{}^b{}_{IJK}C_{b}{}^{K} + N_{aIJ},$$ where the first and the second terms on the right are the homogeneous and particular solutions of (\[psi\]), respectively. Here, the 12 variables $C_{aI}$ parametrize the homogeneous solution. To determine $M_a{}^b{}_{IJK}$ and $N_{aIJ}$, we demand the independent variables $\tilde{B}^{aI}$ and $C_{aI}$ to be canonically conjugate to each other. This is allowed, since the solution induces a reduction of the symplectic structure in , $$\label{ss}
\stackrel{(\gamma)}{\tilde{\Pi}} {}^{aIJ}\dot{\omega}_{aIJ}=\tilde{B}^{aI}\dot{C}_{aI},$$ where we have defined $$\label{C}
C_{aI} := \epsilon \left( \stackrel{(\gamma)}{\omega}\!\!{}_{aIJ}m^J + m_I \stackrel{(\gamma)}{\omega}\!\!{}_{bJK} h_{ac} \tilde{B}^{cJ}\tilde{B}^{bK}\right).$$ Solving jointly and , we obtain $$\begin{aligned}
M_a{}^b{}_{IJK} &=& \epsilon \sigma \left[ - \delta_{a}^{b} m_{[I} \eta_{J] K} + \delta_{a}^{b} \left( P^{-1}\right)_{I J K L} m^{L} \phantom{\dfrac{1}{2}}\right. \nonumber \\
& & - \left( P^{-1}\right)_{I J L M} h_{a c} \tilde{B}^{cL} \tilde{B}^{bM} m_{K} \nonumber \\
& & \left. + \dfrac{1}{\gamma}\star \left( P^{-1}\right)_{IJLM} h_{a c} \tilde{B}^{bL}m^{M} \tilde{B}^{c}{}_{K} \right]\label{Mom}, \\
\label{N}
N_{aIJ} & = & \underaccent{\tilde}{ \lambda}_{ab} \left( -\sigma \epsilon_{IJKL}\tilde{B}^{bK}m^L +\dfrac{2}{\gamma} \tilde{B}^{b}{}_{[I}m_{J]} \right)\label{Nom},\end{aligned}$$ with $\left( P^{-1}\right)_{IJ}{}^{KL}$ being the inverse of $P^{IJ}{}_{KL}$ that satisfies $ P^{IJ}{}_{KL}(P^{-1})^{KL}{}_{MN}=\delta^I_{[M}\delta^J_{N]}$ and $$\underaccent{\tilde}{\lambda}_{ab} := \dfrac{\sigma}{2} \epsilon_{IJKL}\left( h_{ab}h_{cd} - 2 h_{c(a}h_{b)d}\right)\tilde{B}^{cI}\tilde{B}^{fJ}m^{L}\partial_f \tilde{B}^{dK}.$$ In short, the expression , together with and , is the solution of (\[psi\]). Notice that the quantities $N_{aIJ}$ (or $\underaccent{\tilde}{ \lambda}_{ab}=\underaccent{\tilde}{ \lambda}_{ba}$), which can be thought of as the components of the connection not showing up in the symplectic structure , are the ones getting fixed by the solution of (\[psi\]).
Up to now, we have gotten rid of the second-class constraints, leaving in the process a phase space parametrized by the canonical pair $(C_{aI},\tilde{B}^{aI})$ subject to first-class constraints only. It is then necessary to rewrite these constraints in terms of the new canonical variables. To carry out this, let us first introduce the covariant derivative compatible with $\tilde{B}^{aI}$ satisfying $$\label{covder}
\nabla_a \tilde{B}^{bI}:=\partial_a\tilde{B}^{bI}+\Gamma_a{}^I{}_J\tilde{B}^{bJ}+\Gamma^b{}_{ac}\tilde{B}^{cI}-\Gamma^c{}_{ac}\tilde{B}^{bI}=0,$$ where $\Gamma_{aIJ}$ ($=-\Gamma_{aJI}$) takes values in the Lie algebra of SO($\sigma$) and $\Gamma^b{}_{ac}=\Gamma^b{}_{ca}$. These 36 equations allow us to completely fix the 36 quantities $\Gamma_{aIJ}$ and $\Gamma^b{}_{ac}$. In fact, $\Gamma^b{}_{ac}$ is nothing but the Christoffel symbol for the metric $g_{ab}$, that is, the Levi-Civita connection compatible with the spatial metric.
The introduction of $\Gamma^{a}{}_{bc}$ and $\Gamma_{aIJ}$, together with the solutions for $\tilde{\Pi}^{aIJ}$ and $\omega_{aIJ}$, allows us to express the first-class constraints (\[Gauss\])-(\[Scalar\]) as
\[constr2\] $$\begin{aligned}
&&\hspace{-1.5mm}\tilde{\mathcal{G}}^{IJ}= \tilde{B}^{a[I}C_a{}^{J]}+2\epsilon P^{IJ}{}_{KL}\tilde{B}^{a[M}m^{K]}\Gamma_a{}^L{}_M\approx 0, \label{GaussC}\\
&&\hspace{-1.5mm}\tilde{\mathcal{V}}_a = \nabla_{[b}\left(\tilde{B}^{bI}C_{a]I}\right)+\epsilon \tilde{B}^{b[I}m^{K]}\stackrel{(\gamma)}{\Gamma}_{aIJ}\Gamma_b{}^J{}_K\nonumber\\
&&\hspace{7.5mm}-\epsilon\sigma\tilde{\mathcal{G}}^{IJ}\Bigl(C_{aI}-\epsilon \stackrel{(\gamma)}{\Gamma}_{aIK}m^K\Bigr)m_J\approx 0,\label{VectorC}\\
&&\hspace{-1.5mm}\tilde{\tilde{\mathcal{H}}}= -\frac{\sigma}{8}\tilde{B}^{aI}\tilde{B}^{bJ}R_{abIJ} +\frac{1}{4}\tilde{B}^{a[I|}\tilde{B}^{b|J]}\Biggl[C_{aI}C_{bJ}\nonumber\\
&&-2\epsilon C_{aI}\stackrel{(\gamma)}{\Gamma}_{bJK}m^K+\left(\Gamma_{aIL}+\frac{2}{\gamma}\star\Gamma_{aIL}\right)\Gamma_{bJK}m^Km^L\nonumber \\
&&+\frac{1}{\gamma^2}q^{KL}\Gamma_{aIK}\Gamma_{bJL}\Biggr]-\frac{\epsilon}{2}\tilde{B}^{aI} m^J\nabla_a\tilde{\mathcal{G}}_{IJ} +\frac{\sigma\Lambda}{8} \sqrt{h}\approx 0,\nonumber\\
\label{ScalarC}
\end{aligned}$$
where $R_{abIJ}$ is the curvature of $\Gamma_{aIJ}$ and the terms proportional to $\tilde{\mathcal{G}}^{IJ}$ squared have been dropped. It is important to note that although the constraints look rather complicated, they collapse to the ones of the Ashtekar-Barbero’s formulation once we take the time gauge (see below), meaning that the variables $(C_{aI},\tilde{B}^{aI})$ may be regarded as the explicitly SO($\sigma$)-covariant version of the Ashtekar-Barbero variables. Nonetheless, at the fully covariant level, we can explore different parametrizations of the phase space of the theory.
Other Lorentz-covariant parametrizations of the phase space
===========================================================
Let us consider a change of coordinates in which the momentum variables $\tilde{B}^{aI}$ remain unchanged, while the configuration variables take the form
\[transfcan\] $$\begin{aligned}
\hspace{-7mm}C_{aI}=&&\epsilon\left(\Gamma_{aIJ}m^J+h_{ab}\tilde{B}^{bJ}\tilde{B}^{cK}\Gamma_{cJK}m_I\right)+K_{aI}\label{transfcan1}\\
=&&\epsilon\left(\stackrel{(\gamma)}{\Gamma}_{aIJ}m^J\hspace{-.5mm}+h_{ab}\tilde{B}^{bJ}\tilde{B}^{cK}\stackrel{(\gamma)}{\Gamma}_{cJK}m_I\hspace{-.5mm}\right)\hspace{-.5mm}+Q_{aI}.\label{transfcan2}\end{aligned}$$
These transformations give rise to the canonical pairs $(K_{aI}, \tilde{B}^{aI})$ and $(Q_{aI}, \tilde{B}^{aI})$. Indeed, a direct substitution of in results in
\[canvar\] $$\begin{aligned}
\tilde{B}^{aI}\dot{C}_{aI}\label{canvar1} &=& \partial_a\left(\epsilon\dot{\tilde{B}}{}^{aI}m_I\right)+\tilde{B}^{aI}\dot{K}_{aI}\label{canvar2}\\
&=&\partial_a\left(\epsilon\dot{\tilde{B}}{}^{aI}m_I-\frac{\epsilon\sigma}{2\gamma}\sqrt{h}\tilde{\eta}^{abc}h_{bd}h_{ce}\dot{\tilde{B}}^{dI}\tilde{B}^e{}_I\right)\nonumber\\
& &+\tilde{B}^{aI}\dot{Q}_{aI}.\label{canvar3}\end{aligned}$$
We see that, in any case, the symplectic structures in differ from one another by a divergence (which does not contribute if the spatial manifold has no boundary or if suitable boundary conditions are imposed when it has a boundary); this shows that the transformations associated to (\[transfcan\]) are canonical. Alternatively, given that the pair $(C_{aI}, \tilde{B}^{aI})$ is canonical, it can be shown by using the Poisson brackets that the relations (\[transfcan\]) induce canonical transformations among the different pairs considered. Actually, the only complicated bracket is the one involving the new configuration variable with itself, but, by following a procedure close to that of [@ashtekar1991lectures; @thiemann2007modern], it can be shown that it vanishes because the terms between parentheses in (\[transfcan\]) can be derived from a potential in each case.
For the sake of completeness, we display in the following lines the form of the constraints in the new sets of canonical variables. In the canonical coordinates $(K_{aI},\tilde{B}^{aI})$, the constraints read
\[constr3\] $$\begin{aligned}
&&\hspace{-3mm}\tilde{\mathcal{G}}^{IJ}= \tilde{B}^{a[I}K_a{}^{J]}+\frac{\epsilon}{\gamma} \epsilon^{IJ}{}_{KL}\tilde{B}^{a[M}m^{K]}\Gamma_a{}^L{}_M\approx 0 , \label{GaussK}\\
&&\hspace{-3mm}\tilde{\mathcal{V}}_a = \nabla_{[b}\left(\tilde{B}^{bI}K_{a]I}\right)+\frac{\epsilon}{\gamma} \tilde{B}^{b[I}m^{K]}\star\Gamma_{aIJ}\Gamma_b{}^J{}_K\nonumber\\
&&\hspace{5mm}-\epsilon\sigma\tilde{\mathcal{G}}^{IJ}\left(K_{aI}-\frac{\epsilon}{\gamma}\star\Gamma_{aIK}m^K\right)m_J\approx 0,\label{VectorK} \\
&&\hspace{-3mm}\tilde{\tilde{\mathcal{H}}}= -\frac{\sigma}{8}\tilde{B}^{aI}\tilde{B}^{bJ}R_{abIJ} +\frac{1}{4}\tilde{B}^{a[I|}\tilde{B}^{b|J]}\Biggl[K_{aI}K_{bJ}\nonumber\\
&&\hspace{5mm}-\frac{2\epsilon}{\gamma} K_{aI}\star\Gamma_{bJK}m^K+\frac{1}{\gamma^2}q^{KL}\Gamma_{aIK}\Gamma_{bJL}\Biggr]\nonumber \\
&&\hspace{5mm}-\frac{\epsilon}{2}\tilde{B}^{aI} m^J\nabla_a\tilde{\mathcal{G}}_{IJ} +\frac{\sigma\Lambda}{8} \sqrt{h}\approx 0,\label{ScalarK}
\end{aligned}$$
whereas for the set $(Q_{aI},\tilde{B}^{aI})$, the constraints are simply
\[constr1\] $$\begin{aligned}
&&\tilde{\mathcal{G}}^{IJ}= \tilde{B}^{a[I}Q_a{}^{J]} \approx 0, \label{Gauss1}\\
&&\tilde{\mathcal{V}}_a = \nabla_{[b}\left(\tilde{B}^{bI}Q_{a]I}\right)-\epsilon\sigma\tilde{\mathcal{G}}^{IJ}Q_{aI}m_J\approx 0,\label{Vector1} \\
&&\tilde{\tilde{\mathcal{H}}}= -\frac{\sigma}{8}\tilde{B}^{aI}\tilde{B}^{bJ}R_{abIJ} +\frac{1}{4}\tilde{B}^{a[I|}\tilde{B}^{b|J]}Q_{aI}Q_{bJ}\nonumber\\
&&\hspace{8.5mm}-\frac{\epsilon}{2}\tilde{B}^{aI} m^J\nabla_a\tilde{\mathcal{G}}_{IJ} +\frac{\sigma\Lambda}{8} \sqrt{h} \approx 0.\label{Scalar1}
\end{aligned}$$
As can be inferred from (\[Gauss1\]), both $\tilde{B}^{aI}$ and $Q_{aI}$ transform as SO($\sigma$) vectors. Meanwhile, in the other cases we observe that $\tilde{B}^{aI}$ still transforms as an SO($\sigma$) vector but, according to (\[GaussC\]) and (\[GaussK\]), the properties of $C_{aI}$ and $K_{aI}$ under SO($\sigma$) transformations are nontrivial (but they contain a vector part). Furthermore, notice that the diffeomorphism constraint can be expressed as $$\label{diff}
\tilde{\mathcal{D}}_a=\tilde{B}^{bI}\partial_{[b}U_{a]I}+\frac{1}{2}U_{aI}\partial_b\tilde{B}^{bI},$$ for $U_{aI}=C_{aI},\ K_{aI}$ or $Q_{aI}$, which implies that these variables transform as 1-forms under spatial diffeomorphisms.
It is worth realizing that the Immirzi parameter does not explicitly appear in the constraints (\[constr1\]) (it however affects the constants in front of some of the terms proportional to $\tilde{\mathcal{G}}^{IJ}$ squared, which we have neglected). Hence, for the canonical variables $(Q_{aI},\tilde{B}^{aI})$, the Immirzi parameter remains undetectable at the classical level. The constraints (\[constr1\]) actually have the same form as those obtained for the case of the Palatini action (with a cosmological constant) alone (take the limit $\gamma\rightarrow\infty$; see also [@peld1994115]). Therefore, the canonical transformation (\[transfcan\]) allows us to connect the Hamiltonian form of the Holst action with that of Palatini’s, reinforcing the classical equivalence of them.
Time gauge
==========
Here we show how the previous SO($\sigma$)-covariant variables lead to the Ashtekar-Barbero ones. To that end, we adopt the time gauge, which allows us to reduce SO($\sigma$) to its compact subgroup SO(3) by fixing the boosts. In the present framework, this is accomplished by setting $\tilde{B}^{a0}=0$, which is equivalent to $m^i=0$ for a nondegenerate $\tilde{B}^{ai}$ (assumed in what follows). This gauge condition in turn implies that the boost constraint $\tilde{\mathcal{G}}_{0i}$ (in any case) must be solved at once, since $\{\tilde{B}^{a0}(x),\tilde{\mathcal{G}}^{0i}(y)\}=(\sigma/2)\tilde{B}^{ai}\delta^3(x,y)$ defines a nonsingular matrix and so $\tilde{B}^{a0}$ and $\tilde{\mathcal{G}}^{0i}$ form a second-class pair. From , the solution of $\tilde{\mathcal{G}}^{0i}$ reads $C_{a0}=\sigma m^0 \tilde{B}^{bi}\partial_{b}\underaccent{\tilde}{B}_{ai}$ (for the other pairs of variables we obtain $K_{a0}=0$, $Q_{a0}=0$), with $\underaccent{\tilde}{B}_{ai}$ being the inverse of $\tilde{B}^{ai}$. In consequence, the remaining internal symmetry is SO(3), whose infinitesimal generator is the constraint $\tilde{\mathcal{G}}^{ij}$.
In the time gauge, we have, from (\[covder\]), that $\Gamma_{a0i}=0$, whereas $\Gamma_{ai}:=(1/2)\epsilon_{ijk}\Gamma_a{}^{ij}$ ($\epsilon_{ijk}:=\epsilon_{0ijk}$) becomes the spin connection compatible with $\tilde{B}^{ai}$, $$\label{spinconn}
\Gamma_{ai}=\epsilon_{ijk}\left(\partial_{[b}\underaccent{\tilde}{B}_{a]}{}^j+\underaccent{\tilde}{B}_a{}^{[l|}\tilde{B}^{c|j]}\partial_b\underaccent{\tilde}{B}_{cl}\right)\tilde{B}^{bk}.$$ The canonical transformation (\[transfcan\]) then takes the form $$\label{tgct}
A_{ai}=-\epsilon m^0Q_{ai}+\frac{1}{\gamma}\Gamma_{ai},$$ with $A_{ai}:=-\epsilon m^0C_{ai}\ (=-\epsilon m^0K_{ai})$. Since $Q_{ai}$ is a vector and $\Gamma_{ai}$ is a connection, $A_{ai}$ is a connection, the gauge group being in this case SO(3). The canonical transformation (\[tgct\]) is nothing but Barbero’s canonical transformation [@Barbero] (Barbero picks $\gamma=-1$ to rewrite the constraints), which here was derived from (\[transfcan\]). Thus, the latter corresponds to the SO($\sigma$)-covariant version of the canonical transformation implemented by Barbero in order to obtain his canonical description of the phase space of general relativity. Accordingly, the quantity $-\epsilon m^0Q_{ai}$ can be related to the extrinsic curvature in the SO(3) ADM formalism [@AshBalJo; @ashtekar1991lectures], while $\gamma A_{ai}$ corresponds to the Ashtekar-Barbero connection.
Let us introduce the densitized triad $\tilde{E}^{ai}$ through $\tilde{B}^{ai}=:-2\epsilon m^0 \tilde{E}^{ai}$. The phase space is now parametrized by the pair $(A_{ai},\tilde{E}^{ai})$ satisfying the canonical commutation relation $\{A_{ai}(x),\tilde{E}^{bj}(y)\}=(1/2)\delta_a^b\delta_i^j\delta^3(x,y)$. Notice that since (\[spinconn\]) is invariant under constant rescalings, $\Gamma_{ai}$ takes exactly the same form in terms of the densitized triad. Since $A_{ai}$ is an SO(3) connection ($\gamma A_{ai}$ to be more precise), let $F_{abi}:=2\partial_{[a}A_{b]i}-\gamma\epsilon_{ijk}A_{a}{}^j A_{b}{}^k$ be its field strength. Using (\[tgct\]), the next identity arises right away, $$\label{relcurvs}
F_{abi}=-2\epsilon m^0\nabla_{[a}Q_{b]i}+\frac{1}{\gamma}R_{abi}-\gamma\epsilon_{ijk}Q_{a}{}^j Q_{b}{}^k,$$ where $R_{abi}:=(1/2)\epsilon_{ijk}R_{ab}{}^{jk}=2\partial_{[a}\Gamma_{b]i}-\epsilon_{ijk}\Gamma_{a}{}^j \Gamma_{b}{}^k$. With the previous expression at hand, the constraints (\[Gauss1\])-(\[Scalar1\]) can be given, in the time gauge, the form
\[barbero\] $$\begin{aligned}
&&\hspace{-4mm}\tilde{\mathcal{G}}_i:=\frac{1}{2}\epsilon_{ijk}\tilde{\mathcal{G}}^{jk}=\frac{1}{\gamma}\left(\partial_a\tilde{E}^a{_i}-\gamma\epsilon_{ijk}A_a{}^j\tilde{E}^{ak}\right) \approx 0, \label{Gauss2}\\
&&\hspace{-4mm}\tilde{\mathcal{V}}_a = \tilde{E}^{bi}F_{bai} +\left(\gamma A_{ai}-\Gamma_{ai}\right)\tilde{\mathcal{G}}^i\approx 0,\label{Vector2} \\
&&\hspace{-4mm}\tilde{\tilde{\mathcal{H}}}= -\frac{1}{2\gamma}\epsilon_{ijk}\tilde{E}^{ai}\tilde{E}^{bj}\left[F_{ab}{}^k+\left(\sigma\gamma-\gamma^{-1}\right)R_{ab}{}^k\right]\nonumber\\
&&\hspace{4.5mm}+\frac{1}{\gamma} \tilde{E}^{ai}\nabla_a \tilde{\mathcal{G}}_i +\sigma\Lambda |\tilde{\tilde{E}}| \approx 0 ,\label{Scalar2}
\end{aligned}$$
with $\tilde{\tilde{E}}:=\det\tilde{E}^{ai}$. These are precisely the constraints of the Ashtekar-Barbero formulation of general relativity. It is worth stressing that although we followed a path resembling the one walked by Barbero, the previous result can be achieved regardless of the canonical pair considered. In particular, since we may think of the constraints as the constraints with the canonical transformation already implemented, we can obtain directly from in the time gauge. Notice that in our approach the Immirzi parameter plays the role of a coupling constant for the local SO(3) symmetry.
Conclusions
===========
In this paper we have solved the second-class constraints arising in the Hamiltonian analysis of first-order general relativity with the Immirzi parameter in a manifestly SO($\sigma$)-covariant fashion (recall that $\sigma=-1$ corresponds to the Lorentz group). As a result, we obtained a description of the phase space involving only first-class constraints that exhibits a dependence on the Immirzi parameter. The associated canonical variables, which we called $(C_{aI},\tilde{B}^{aI})$, turn out to be related to other two sets of SO($\sigma$)-covariant variables $(K_{aI},\tilde{B}^{aI})$ and $(Q_{aI},\tilde{B}^{aI})$ by means of the canonical transformations determined by . In contrast to both $K_{aI}$ and $C_{aI}$, whose transformation law under local SO($\sigma$) transformations is nontrival \[as can be deduced from (\[GaussC\]) and (\[GaussK\])\], the configuration variable $Q_{aI}$ transforms as an SO($\sigma$) vector (because of the form of the canonical transformation, $\tilde{B}^{aI}$ is an internal vector in any case). In terms of the canonical pair $(Q_{aI},\tilde{B}^{aI})$, the constraints take the same form as those resulting from the canonical analysis of the Palatini action, thus eliminating the Immirzi parameter from the canonical theory. Remarkably, in the time gauge the canonical transformation becomes the transformation implemented by Barbero in order to obtain his canonical formulation of general relativity [@Barbero], the spatial components $C_{ai}$ (or $K_{ai}$) being related to the Ashtekar-Barbero connection. In consequence, the canonical variables $(C_{aI},\tilde{B}^{aI})$ \[or $(K_{aI},\tilde{B}^{aI})$\] can be regarded as the SO($\sigma$)-covariant extension of the Ashtekar-Barbero variables.
To sum up, here we have clarified the origin of the Ashtekar-Barbero variables and their relation to the Holst action. More interesting is the Hamiltonian formulation embodied in the constraints (\[constr2\]), (\[constr3\]) or (\[constr1\]), which explicitly exhibits the SO($\sigma$) invariance of the original theory (on the other hand, the Ashtekhar-Barbero formulation breaks it by resorting to the time gauge). This feature is appealing since the Lorentz symmetry is thought to be one of the fundamental symmetries of nature, and so it would be desirable to preserve it to the utmost in a quantum theory of gravity. Thus, the Hamiltonian formulation presented in this paper could be an interesting starting point for new developments to approach the quantization of gravity, something we think will provide meaningful results. In particular, we expect that this formulation might help to determine the fate of the Immirzi parameter in quantum gravity once for all.
We thank Giorgio Immirzi, Karim Noui, Alejandro Perez, Carlo Rovelli, José David Vergara, and José A. Zapata for their valuable comments. This work was supported in part by Consejo Nacional de Ciencia y Tecnología (CONACyT), México, Grant No. 237004-F.
[^1]: This parameter does not modify the classical dynamics (on shell), but becomes relevant as long as we work off shell. See for instance [@RefGSGR].
[^2]: We can also use the canonical variables $(\stackrel{(\gamma)}{\omega}{}_{aIJ},\tilde{\Pi}^{aIJ})$.
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- 'L. Fossati'
- 'N. Castro'
- 'M. Schöller'
- 'S. Hubrig'
- 'N. Langer'
- 'T. Morel'
- 'M. Briquet [^1]'
- 'A. Herrero'
- 'N. Przybilla'
- 'H. Sana'
- 'F. R. N. Schneider'
- 'A. de Koter'
- the BOB collaboration
title: 'B fields in OB stars (BOB): low-resolution FORS2 spectropolarimetry of the first sample of 50 massive stars[^2]'
---
[ Within the context of the collaboration “B fields in OB stars (BOB)”, we used the FORS2 low-resolution spectropolarimeter to search for a magnetic field in 50 massive stars, including two reference magnetic massive stars. Because of the many controversies of magnetic field detections obtained with the FORS instruments, we derived the magnetic field values with two completely independent reduction and analysis pipelines. We compare and discuss the results obtained from the two pipelines. We obtaind a general good agreement, indicating that most of the discrepancies on magnetic field detections reported in the literature are caused by the interpretation of the significance of the results (i.e., 3–4$\sigma$ detections considered as genuine, or not), instead of by significant differences in the derived magnetic field values. By combining our results with past FORS1 measurements of HD46328, we improve the estimate of the stellar rotation period, obtaining P=2.17950$\pm$0.00009days. For HD125823, our FORS2 measurements do not fit the available magnetic field model, based on magnetic field values obtained 30 years ago. We repeatedly detect a magnetic field for the O9.7V star HD54879, the HD164492C massive binary, and the He-rich star CPD$-$573509. We obtain a magnetic field detection rate of 6$\pm$4%, while by considering only the apparently slow rotators we derive a detection rate of 8$\pm$5%, both comparable with what was previously reported by other similar surveys. We are left with the intriguing result that, although the large majority of magnetic massive stars is rotating slowly, our detection rate is not a strong function of the stellar rotational velocity. ]{}
Introduction {#sec:introduction}
============
Magnetic fields play an important role in the structure and evolution of stars, and systematic surveys aiming at the detection and characterisation of magnetic fields in massive stars, have only recently started to be carried out [@wade2014; @morel2014; @morel2015]. Their most evident achievement is the great increase in the number of detected magnetic massive stars, leading for example to the determination of a magnetic field incidence of $\sim$7%, made on the basis of a sample of hundreds of stars [@wade2014]. Recently, the detection of rather weak magnetic fields opened the possibility that the incidence may be higher, calling for deeper observations for the brightest stars [@fossati2015].
Despite these achievements, the number of known magnetic massive stars is still relatively small, particularly with respect to the wide variety of detected phenomena and features in their spectra and light curves. The detection of more magnetic massive stars is therefore a necessary step for further advances.
This work is part of the collaboration “B fields in OB stars” (BOB), whose primary aim is to characterise the incidence of large-scale magnetic fields in slowly rotating (i.e., $\lesssim$100) main-sequence massive stars (i.e., early B- and O-type stars), to test whether the slow rotation is primarily caused by the presence of a magnetic field. The observations are being performed with the high-resolution HARPSpol polarimeter [@snik2011; @piskunov2011], feeding the HARPS spectrograph [@mayor2003] attached to the ESO 3.6m telescope in LaSilla (Chile), and the FORS2 low-resolution spectropolarimeter [@app1992] attached to the Cassegrain focus of the 8m Antu telescope of the ESO Very Large Telescope of the Paranal Observatory. More details about the BOB collaboration can be found in @morel2014 [@morel2015]. We present here the results obtained from the first set of 50 stars, while the results of a subsequent sample will be presented in a forthcoming paper (Schöller et al., in preparation).
Target selection {#sec:targ_selection}
================
The target selection was performed considering the stellar [*i*]{}) spectral type (O- and early B-type stars), [*ii*]{}) luminosity class (dwarfs and giants; V$\rightarrow$III), and [*iii*]{}) projected rotational velocity ($\leq$100). As main sources of information we used @howarth1997, the UVES Paranal Observatory Project spectral library [because of the availability of high-resolution spectra, which would in particular complement the low-resolution FORS2 observations; @uvespop], the GOSSS survey [@gosss Barba priv. communication], and the IACOB database [@iacob]. We also checked the catalogue compiled by @bychkov2009 for previous magnetic field measurements, while we gathered information about possible binarity from the surveys cited above. As shown by @babel1997, the interaction of the stellar wind of magnetic massive stars with their magnetosphere can be a strong source of hard X-rays, which may be detectable if the stars are close enough. For this reason, we included in our target list previously identified hard X-ray sources, using available X-ray catalogues and archival X-ray data, with values up to 120.
The selected sample of stars also includes two known magnetic reference stars: HD46328 [@hubrig2006] and HD125823 [@wolff1974; @borra1983]. We tried to limit the observations of supergiants (luminosity class I) because for these we cannot exclude that even a non-magnetic wind [@langer1998] might have spun them down. The compiled target list was then split according to stellar magnitude, so that stars with $V\gtrsim$7.5mag have been preferentially observed with FORS2 and the remaining with HARPSpol.
Observations {#sec:observations}
============
FORS2 is a multi-mode optical instrument capable of imaging, polarimetry, and long-slit and multi-object spectroscopy. The polarimetric optics, previously mounted on FORS1 [@app1998], have been moved to FORS2 in March 2009. During the first run, performed between the 7 and 9 of April 2013, we observed 24 stars, while during the second run, performed between the 6 and 8 of February 2014, we observed 28 stars (HD102475 and HD144470 were observed during both runs). The observing log of both runs is given in Table \[tab:obs.log\].
For the first run, we used the 2k$\times$4k E2V CCDs (pixel size 15$\mu$m$\times$15$\mu$m) which are optimised for observations in the blue spectral region (i.e., $<$4500Å), while for the second run we used the 2k$\times$4k MIT CCDs (pixel size 15$\mu$m$\times$15$\mu$m)[^3]. All observations were performed using a single narrow slit width of 0.4$\arcsec$, to reach a high spectral resolution and to minimise spurious effects of seeing variations [see e.g. @fossati2015b], the 200kHz/low/1$\times$1 readout mode, to minimise overheads and increase the dynamic range, and the GRISM600B. Each spectrum covers the 3250–6215Å spectral range which includes all Balmer lines, except H$\alpha$, and a number of He lines. Using the emission lines of the wavelength calibration lamp we measured an average (across the covered wavelength range) resolving power of 1700. Each star was observed with a sequence of spectra obtained by rotating the quarter waveplate alternatively from $-$45$^{\circ}$ to $+$45$^{\circ}$ every second exposure (i.e., $-$45$^{\circ}$, $+$45$^{\circ}$, $+$45$^{\circ}$, $-$45$^{\circ}$, $-$45$^{\circ}$, $+$45$^{\circ}$, etc.). The adopted exposure times and obtained signal-to-noise ratios (S/N) per pixel calculated around 4950Å of Stokes $I$ are listed in Table \[tab:obs.log\].
Data reduction and analysis {#reduction.analysis}
===========================
Because of the several controversies present in the literature about magnetic field detections in intermediate- and high-mass stars performed with the FORS spectropolarimeters [see e.g. @wade2007; @silvester2009; @shultz2012; @bagnulo2012; @bagnulo2013], the data were independently reduced by two different groups (one based in Bonn and one based in Potsdam) using a set of completely independent tools and routines. The first reduction and analysis (Bonn) was performed with a set of IRAF[^4] [@tody] and IDL routines (hereafter called Bonn pipeline) developed following most of the technique and recipes presented by @bagnulo2012 [@bagnulo2013], while the second reduction and analysis (hereafter called Potsdam pipeline) was based on the tools described in @hubrig2004a [@hubrig2004b], with the recent update described in @steffen2014.
The surface-averaged longitudinal magnetic field was measured using the following relation [@angel1970; @landstreet1975]: $$\label{eq:bz}
V(\lambda)=-g_{\rm eff}C_z\lambda^2\frac{1}{I(\lambda)}\frac{{\rm d}I(\lambda)}{{\rm d}\lambda}\langle\,B_z\,\rangle$$ and the least-squares technique, originally proposed by @bagnulo2002 and further refined by @bagnulo2012. In Eq. \[eq:bz\] $V(\lambda)$ and $I(\lambda)$ are the Stokes $V$ and $I$ profiles, respectively, $g_{\rm eff}$ is the effective Landé factor, which was set to 1.25 except for the region of the hydrogen Balmer lines where $g_{\rm eff}$ was set to 1.0, and $$\label{eq:cz}
C_z=\frac{e}{4 \pi m_ec^2}$$ where $e$ is the electron charge, $m_e$ the electron mass, and $c$ the speed of light ($C_z\simeq4.67\,\times\,10^{-13}$Å$^{-1}$G$^{-1}$). See @bagnulo2012 for a detailed discussion of the physical limitations of this technique.
In the remainder of this section, we thoroughly describe the routines and settings adopted within the two pipelines. We also schematically summarise the main similarities and differences.
Bonn pipeline
-------------
Within the Bonn pipeline, we applied a bias subtraction, but no flat-field correction[^5]. We performed an average extraction, as recommended by @bagnulo2012, using a fixed extraction radius of 25 pixels, without background subtraction. The adopted extraction radius allowed us to avoid the spectrum of the parallel beam being contaminated by a strong instrumental internal reflection, which would otherwise irreparably affect the Stokes profiles in the region around H$\delta$. Within each night, each parallel or perpendicular beam was wavelength calibrated using the parallel or perpendicular beam of one wavelength calibration lamp obtained in the morning following the night of observation. The wavelength calibration was performed manually to ensure that the same set of arc lines and fitting functions were used for both beams [@bagnulo2013]. The pipeline finally bins the spectra according to the natural sampling of the instrument/grism of 0.75Å/pix.
We combined the profiles to obtain Stokes $I$, $V$, and the diagnostic $N$ parameter [@donati1992] using the difference method following the formalism of @bagnulo2009[^6]. We rectified each Stokes $V$ profile using a fourth-order polynomial and applied a sigma clipping to filter out all data points where the $N$ profile deviated more than 3$\sigma$ from the average value ($\overline{N}$), where $\sigma$ is the standard deviation of the $N$ profile. The value of was calculated using either the hydrogen lines, the metallic lines, or the whole spectrum in the 3710–5870Å spectral region. The Stokes $I$ spectra were inspected to remove all spectral regions contaminated by emission lines. The field was calculated minimising $$\label{eq:chi}
\chi^2=\sum_i\frac{(V(\lambda_i)-\langle\,B_z\,\rangle\,x_i-b)^2}{\sigma_i^2}$$ where $x_i$=$-g_{\rm eff}C_z\lambda_i^2(1/I(\lambda)\times{\rm d}I(\lambda)/{\rm d}\lambda)_i$, $i$ indicates each spectral point, and $b$ is a constant that accounts for possible spurious continuum polarisation left after the rectification [see @bagnulo2002; @bagnulo2012 for more details]. Finally, the code provides the values of and (the magnetic field calculated from the $N$ profile), their standard uncertainty, and their $\chi^2$-scaled uncertainty [ and – see Sect. 3.4 of @bagnulo2012]. Optionally, the IDL routine allows one to extract with a $\chi^2$ minimisation routine that takes into account the uncertainties on both axes, using the [astrolib fitexy.pro]{}[^7] routine based on a routine that is part of the numerical recipes [@press1992]. In this work, we always adopted the $\chi^2$-scaled uncertainties, taking into account only the error bars on Stokes $V$. By adopting the $\chi^2$-scaled uncertainties, we also compensated for variations of the CCD gain from the nominal value, which was adopted for the spectral extraction. Using the bias and flat-field calibration frames collected during our runs, we consistently measured a CCD gain slightly lower than the adopted nominal value. This is confirmed, for example, by the fact that for the $N$ profile we constantly obtained an average uncertainty smaller than the standard deviation.
Potsdam pipeline
----------------
{width="180mm"}
{width="45.00000%"} {width="45.00000%"} {width="32.00000%"} {width="32.00000%"} {width="32.00000%"}
Within the Potsdam pipeline, the parallel and perpendicular beams were extracted from the raw FORS2 data using a pipeline written in the MIDAS environment by T. Szeifert. This pipeline reduction by default includes background subtraction and no flat-fielding. A unique wavelength calibration frame was used for each night. The spectra were resampled with a spectral bin size of 0.1Å/pix.
Stokes $V$ and $I$ were combined in the same way as for the Bonn pipeline. The $V/I$ spectra were rectified using a linear function in the way described by @hubrig2014b. The diagnostic null spectra, $N$, were calculated as pairwise differences from all available $V$ spectra. From these, 3$\sigma$-outliers were identified and used to clip the $V$ spectra. Following these steps, a visual inspection of all resulting spectra is necessary to ensure that no spurious signals have gone undetected.
Given the Stokes $I$ and $V$ spectra, the mean longitudinal magnetic field is derived for the wavelength region 3645–5880Å by linear regression. In the past, the Potsdam pipeline followed the same path as the Bonn pipeline, using Eq. \[eq:chi\] and applying the $\chi^2$-correction to the resulting error, if the $\chi^2$ was larger than 1. Since we used 0.1Å/pix as spectral bin size, we had to multiply the resulting error by a factor $\sqrt{7.5}$. Now, we relied on the bootstrapping technique, first introduced by @rivinius2010 for the magnetic field measurements. For this, we generated $M = 250\,000$ statistical variations of the original dataset and analysed the resulting distribution $P(\left<B_{\rm z}\right>)$ of the M regression results, where Eq. \[eq:chi\] was applied to each of the statistical variations. Mean and standard deviation of this distribution were identified with the most likely mean longitudinal magnetic field and its 1$\sigma$ error, respectively. The main advantage of this method is that it provides an independent error estimate.
Comparison
----------
Table \[tab:pipelines\] summarises the main nominal similarities and differences between the two pipelines. Although both pipelines applied a sigma-clipping algorithm and a normalisation of the Stokes $V$ spectrum and of the $N$ profile, these operations were performed in significantly different ways. The Bonn pipeline used a polynomial to rectify the final co-added Stokes $V$ spectrum and applied the same function to the $N$ profile, while the Potsdam pipeline used a linear function to rectify each single Stokes $V$ spectrum obtained from each pair of frames (i.e., $-$45$^{\circ}$, $+$45$^{\circ}$), with the $N$ profile being the difference of already rectified Stokes $V$ spectra. The Potsdam pipeline applied a sigma clipping algorithm based on deviations from the $N$ profile, similarly to the Bonn pipeline, but because of the oversampling, it also rejected the ten points next to the deviating ones. We considered that for the brightest stars there might be an additional difference in the number of frames considered for the analysis, because of the differences in identifying and discarding saturated frames within the two pipelines, with the Bonn pipeline having a more severe criterion (i.e., a frame is removed when 20 or more neighbouring pixels have a number of counts larger than 60000, each). Another substantial difference is in the wavelength ranges selected for the analysis of the spectra using hydrogen lines (or metallic lines) that were manually selected on a star-by-star basis by the users of each pipeline.
Results
=======
Magnetic field detection rate {#Bfield}
-----------------------------
Table \[tab:mag.field\] lists the magnetic field values obtained using the two pipelines. Following @bagnulo2012, the BOB collaboration decided to consider a magnetic field to be detected only above the 5$\sigma$ level and with a value consistent with zero. The average S/N of the spectra is about 2500 with an average uncertainty of about 80G (considering the measurements conducted on the hydrogen lines), in agreement with the empirical S/N-uncertainty relation given by @bagnulo2015.
The whole sample is composed of 50 stars (28 O-type stars, 19 B-type stars, 1 A-type supergiant, and 2 F-type stars; note that the spectra of the two stars classified in Simbad as F-type suggest instead an earlier spectral type), two of them being the magnetic reference stars HD46328 and HD125823. The sample comprises at least three spectroscopic binaries (HD164492C, HD117357, and HD92206c; no high-resolution spectra are available for most of the observed stars, hence only limited information on possible binarity is available), five likely post-main-sequence stars (HD168607, HD168625, HD92207, HD72754, and HD48279AB), and one known chemically peculiar He-rich star (CPD$-$573509). Ten stars have a value above $\sim$100.
On the basis of this sample, and excluding the two magnetic reference stars, we detected three magnetic stars: HD54879, HD164492C, and CPD$-$573509. The corresponding detection rate is therefore of 6$\pm$4%, consistent with that obtained by the Magnetism in Massive Stars (MiMeS) survey [@wade2014]. By only considering the slow rotators instead, we derive a slightly higher magnetic field detection rate of 8$\pm$5%, still consistent with that given by the MiMeS survey. Thus, the detection rate amongst slow rotators is apparently only slightly enhanced. This is surprising, given that the bimodal distribution of massive stars [e.g., @dufton2013; @oscar2013; @iacob] may suggest that about 25% of the O- and B-type stars show a below 100, but about 80% of the 64 magnetic O- and B-type stars discussed by @petit2013 have a projected rotational velocity below this threshold. Both numbers together lead to an expected detection rate of about 20% amongst the slow rotators.
The reason for this discrepancy remains unclear at present, but biases could lead to this situation; several magnetic stars have been selected from secondary magnetic field indicators (spectral variability, X-ray emission, etc.), for instance, before their field has been determined, which could imply that the non-biased detection rate is lower than the reported one. Moreover, unlike the intermediate-mass stars, the massive stars appear not to show a magnetic desert [@fossati2015], meaning that many of them could have relatively weak fields that remained undetected. To resolve this puzzle is left to future investigations.
For three stars, HD102475, HD118198, and HD144470, we obtained a measurement of the magnetic field at the 3–4$\sigma$ level using both pipelines, but either from hydrogen lines or the entire spectrum, but never both. Although further FORS2 observations led to clear non-detections, it would be important to observe these stars with a high-resolution spectropolarimeter to perform a deeper search for a magnetic field.
Standard stars: HD46328 and HD125823
------------------------------------
Figures \[fig:hd46328\_bonn\] and \[fig:hd46328\_potsdam\] illustrate the results obtained for the analysis of the hydrogen lines of the magnetic standard star HD46328 from the Bonn and Potsdam pipelines, respectively. Figure \[fig:hd125823\] illustrates the results of the Bonn pipeline for the analysis of the hydrogen lines of the magnetic standard star HD125823.
The star HD46328 ($\xi^1$CMa) is a $\beta$Cep star [@saesen2006] for which the presence of a magnetic field has first been reported by @hubrig2006 and @hubrig2009. This was further confirmed by high-resolution spectropolarimetry [@silvester2009; @four2011; @shultz2012]. @hubrig2011 used the FORS1 measurements to model the magnetic field of HD46328, assuming a dipolar configuration of the magnetic field. They obtained a rotation period of P=2.17937$\pm$0.00012days, a dipolar magnetic field strength B$_{\mathrm d}$ of 5.3$\pm$1.1kG, and an obliquity $\beta$ of 79.1$^\circ$$\pm$2.8$^\circ$. As shown in Table \[tab:mag.field\], both pipelines led to the measurement of a positive longitudinal magnetic field (at the $\sim$7$\sigma$ level) of about 400G, as expected on the basis of the previous FORS1 measurements.
Taking advantage of the longer time-base, we used the FORS1 and FORS2 measurements of , obtained from the analysis of the whole spectrum, to improve the estimate of the stellar rotation period. To be consistent with the FORS1 measurements, we used the FORS2 results of the Potsdam pipeline for this analysis. We derived the stellar rotation period adopting the frequency analysis and mode identification for asteroseismology (FAMIAS) package [@zima2008] and the phase dispersion minimization (PDM) method [@j1971; @s1978], consistently obtaining a period of P=2.17950$\pm$0.00009days. Following @breger1993 we find this period to be significant. On the basis of Musicos and ESPaDOnS high-resolution spectropolarimetric observations, @shultz2015 suggested a rotation period longer than 40years. Their measurement of the period is mostly constrained by Musicos observations made at very high airmass, which led to negative values of . We can only report here that the FORS observations conducted in the past years always led to positive values of , and that only further observations obtained in the next 2–5 years will allow unambiguously distinguishing between the two solutions.
Figure \[fig:phase\_plot\_hd46328\] shows the phase plot obtained using the FORS1 and FORS2 measurements, and the results of the magnetic field modelling given by @hubrig2009. The results obtained with both pipelines fit the expected behaviour of the longitudinal magnetic field well. This is most likely because the two sets of measurements were obtained with essentially the same instrument (the polarimetric optics of FORS1 were moved to FORS2 after the FORS1 decommissioning) and using similar (almost identical in the case of the Potsdam pipeline) analysis techniques.
![Phase plot of the values obtained for HD46328 from the FORS1 [black asterisks; @hubrig2009] and FORS2 (red rhombs: Bonn pipeline, blue triangles: Potsdam pipeline; using the whole spectrum) data, and the sine wave function calculated using the magnetic field model given by @hubrig2011. A slight phase shift has been applied between our two sets of FORS2 measurements for visualisation purposes.[]{data-label="fig:phase_plot_hd46328"}](./figures/phase_plot_hd46328.ps){width="90mm"}
The star HD125823 (aCen) is a Bp star with a rotation period of 8.817744$\pm$0.000019days [@catalano1996]. @borra1983 detected a magnetic field ranging between $-$470G and $+$430G. We used the stellar magnetic field model by @bychkov2005 to compare the FORS2 measurements (from both pipelines) with that of @borra1983. We note that @bychkov2005 considered a period of 8.8171days, which is slightly different from that given by @catalano1996. The phase plot is shown in Fig. \[fig:phase\_plot\_hd125823\]. The FORS2 measurements do not fit the magnetic field model well that was obtained by @bychkov2005 using the results of @borra1983. This could be due to a systematic shift (of $\sim$400G) between the two datasets due to the use of different instruments, setups, and wavelength regions for the magnetic field measurements [@landstreet2014], and/or more likely to small errors in the magnetic model that, given the long time-span between the two sets of observations, led to a significant discrepancy (e.g., a phase shift of $\sim$0.3).
![Phase plot of the values obtained for HD125823 from the measurements of @borra1983 (black asterisks) and FORS2 (red rhombs: Bonn pipeline, blue triangles: Potsdam pipeline; using the whole spectrum) data, and the sine wave function calculated using the magnetic field model given by @bychkov2005. A slight phase shift has been applied between the two sets of FORS2 measurements for visualisation purposes.[]{data-label="fig:phase_plot_hd125823"}](./figures/phase_plot_hd125823.ps){width="90mm"}
New detections: HD54879, HD164492C, and CPD$-$573509
----------------------------------------------------
The star HD54879 is a single, slowly rotating O9.7V star [@sota2011] and a probable member of the CMaOB1 association [@claria1974]. The discovery of the magnetic field was presented by @castro2015. Figure \[fig:hd54879\] shows the outcome of the Bonn pipeline indicating the clear detection of the magnetic field at the $\sim$9$\sigma$ level, already reported by @castro2015. The stellar photospheric spectrum does not present any morphological peculiarity, typical for example of Of?p stars, and its analysis did not reveal any chemical peculiarity. The only distinctive feature in the spectrum of HD54879 is a prominent H$\alpha$ emission that @castro2015 attributed to circumstellar material, as the comparison of the H$\alpha$ line profile with that of the star defining the O9.7V spectral type excludes the stellar wind as the cause of the emission. The star HD164492C is a massive star in the centre of the Trifid nebula. @hubrig2014 reported the detection of a rather strong magnetic field on the basis of FORS2 and HARPSpol data. Figure \[fig:hd164492C\] illustrates the clear detection of the magnetic field at the $\sim$9$\sigma$ level, already reported by @hubrig2014[^8]. The high-resolution HARPSpol observations and further high-resolution UVES spectra revealed that HD164492C is in fact a multiple system, composed of at least two stars. More details about this system and the UVES observations will be given in a follow-up paper (Gonz[á]{}lez et al., in prep.). The star CPD$-$573509 is a He-rich B2 star member of the $\sim$10Myr old open cluster NGC3293. We observed the star with FORS2 twice during the run in February 2014. Figure \[fig:cpd-573509\] reveals the detection of the magnetic field (at the $\sim$5$\sigma$ level) obtained from the data collected on 7 February 2014. Following the FORS2 measurements, we observed the star with the HARPSpol high-resolution spectropolarimeter confirming the presence of a magnetic field. Our measurements of the magnetic field are suggestive of the presence of a rather strong and rapidly varying magnetic field. A preliminary analysis confirms the He-rich nature of the star (about three times solar). Its membership in the NGC3293 open cluster allows us to conclude that the star has evolved throughout about one third of its main-sequence lifetime. This makes CPD$-$573509 one of the most evolved He-rich stars with a tight age constraint, promising to provide information on the evolution of stars with magnetically confined stellar winds. More details will be given in a dedicated paper (Przybilla et al., in prep.).
Discussion {#sec:discussion}
==========
{width="185mm"}
{width="185mm"}
One of the characteristics of the BOB collaboration is that the reduction and analysis of the spectropolarimetric data is independently carried out by two teams using different and independent tools and pipelines. This gives us the possibility to directly compare the results on a statistically large sample of stars.
To make a more thorough comparison, we also applied a mixed reduction and analysis of the data: we derived the and values using the Bonn pipeline for the data reduction (i.e., bias subtraction, spectral extraction, wavelength calibration) and the Potsdam pipeline for the spectral analysis (i.e., derivation of the Stokes parameters and of the magnetic field values), and vice versa. The results of this test are presented in Table \[tab:cross.check\].
Figures \[fig:plottoneV\] and \[fig:plottoneN\] show the comparison between the results obtained by reducing and analysing the spectra (hydrogen lines or whole spectrum) with the Bonn and Potsdam pipelines, or the mixed reduction and analysis. We consider here 102 sets of measurements, each set composed of four measurements (i.e., and obtained from the analysis of the hydrogen lines or of the whole spectrum), and obtained in four different ways with six possible comparisons (i.e., BrPa, PrBa, and PrPa compared to BrBa; BrPa and PrBa compared to PrPa; BrPa compared to PrBa – the meaning of each acronym can be found in the header of Tables \[tab:mag.field\] and \[tab:cross.check\]), for a total of 2448 direct comparisons.
Figures \[fig:plottoneV\] and \[fig:plottoneN\] display a general good agreement among the four sets of results, and for most cases ($\sim$96.73%) the differences are within 2$\sigma$. In about 1.6% of the cases the difference between the various sets of and values is above 3$\sigma$. This is close to the expectations of Gaussian statistics. In addition, @bagnulo2012 showed that even slight changes in just one step in the data reduction or analysis procedure may lead to variations in the and values of 2–3$\sigma$. We note that the comparison of the uncertainties shown in Fig. \[fig:plottoneV\] and \[fig:plottoneN\] is slightly affected by the fact that the Potsdam pipeline calculates the uncertainties using the nominal CCD gain, while the uncertainties calculated with the Bonn pipeline, because of the $\chi^2$ scaling, account for deviations from the nominal value of the CCD gain.
The best agreement is found when comparing the results of the two pipelines separately (i.e., BrBa vs. PrPa) and of each pipeline with what is obtained from the mixed Bonn pipeline reduction and Potsdam pipeline analysis (i.e., BrBa vs. BrPa and PrPa vs. BrPa) with $<$2% of the cases having a difference larger than 2$\sigma$. For the other three comparisons (i.e., BrBa vs. PrBa, PrPa vs. PrBa, and PrBa vs. BrPa), in 5–8% of the cases the difference is larger than 2$\sigma$, about what expected by random noise. These results do not seem to display a regular pattern that would allow one to conclude anything about the relative importance of the adopted reduction or analysis procedure in the final results.
The largest differences ($\geq$4$\sigma$) instead follow a clear pattern as they are found almost exclusively among the measurements conducted for the magnetic stars. This is probably because, for the non-magnetic stars, both and measure noise, for which one may expect a Gaussian behaviour, which therefore leaves limited room for large deviations. On the other hand, for the magnetic stars, uncertainties are generally small and differences in the data reduction or analysis procedure may indeed modify the Stokes $V$ signatures, which therefore leads to significant differences. This suggests that the optimal data reduction and analysis procedure may therefore be sought by considering magnetic (standard) stars [see also @landstreet2014] in addition to the analysis of large samples [see e.g., @bagnulo2012; @bagnulo2015]. The identification of the exact reduction step(s) leading to the observed differences is beyond the scope of this work.
On the basis of our analysis, we conclude that except for a few cases [e.g., HD92207; @bagnulo2013], the several discrepancies reported in the literature are mostly due to the interpretation of the significance of the results, that is, whether 3–4$\sigma$ detections are considered as genuine or not.
Conclusion {#sec:conclusion}
==========
Within the context of the BOB collaboration, whose primary aim is characterising the incidence of magnetic fields in slowly rotating massive stars, we obtained FORS2 spectropolarimetric observations of a set of 50 massive stars selected considering their spectral type, luminosity class, and projected rotational velocity. Within this sample, we also observed two massive stars that were previously known to host a magnetic field and that we used as standards (HD46328 and HD125823). The observations were performed in April 2013 and February 2014.
We derived the longitudinal magnetic field values using two fully independent reduction and analysis pipelines to compare the results and decrease the probability of spurious detections. We detected the magnetic field for both HD46328 and HD125823. We used previous FORS1 measurements, in addition to our FORS2 results, to further constrain the rotation period of HD46328, obtaining a best fit of P=2.17950$\pm$0.00009days. We did not find evidence for a long rotation period ($>$40years), as recently suggested by @shultz2015, but only further observations obtained in the next years will allow unambiguously distinguishing between the two solutions. Our FORS2 results are also a good fit to the magnetic field model of HD46328 presented by @hubrig2011. In contrast, our measurements do not fit the magnetic field model of HD125823 well that was reported by @bychkov2005 on the basis of measurements obtained by @borra1983, possibly because of systematic shifts between the two datasets [see e.g., @landstreet2014] and/or of small errors in the magnetic field model that would be magnified when considering measurements so much spread in time.
Within the remaining sample of 50 stars, we detected a magnetic field for three of them: HD54879, HD164492C, and CPD$-$573509. For the chemically normal O9.7V star HD54879 we detected a longitudinal magnetic field with a maximum strength of about 1kG [see @castro2015 for more details]. HD164492C is a massive binary system in the centre of the Trifid nebula for which we detected a magnetic field of about 600G, although it is unclear which of the stars composing this system is magnetic [see @hubrig2014 for more details]. The star CPD$-$573509 is a He-rich B2 star member of the NGC3293 open cluster. We detected a rapidly varying longitudinal magnetic field of about 700G, further confirmed by follow-up HARPSpol high-resolution spectropolarimetric observations (Przybilla et al., in prep.).
Considering the whole sample of observed stars, but excluding HD46328 and HD125823, we obtained a magnetic field detection rate of 6$\pm$4%, while by considering only the apparently slow rotators we reached a slightly higher detection rate of 8$\pm$5%. Both numbers are comparable to the magnetic field incidence rate of O- and B-type stars of 7% reported by @wade2014. Given that the vast majority of magnetic massive stars rotate slowly, we expected to find a higher magnetic fraction (about 20%) from our sample of slow rotators. That this is not so may hint at biases in the magnetic stars sample and might imply that a large number of massive stars contain magnetic fields that are too weak to be detected at present [@fossati2015].
Finally, we compared the magnetic field values obtained from the two reduction and analysis pipelines. We obtained a general good agreement, and for only about 1% of the cases, the difference is above 3$\sigma$, the majority of those being for the magnetic stars. Our results indicate that most discrepancies on magnetic field detections reported in the literature are mainly caused by the interpretation of the significance of the results, that is, it depends on whether 3–4$\sigma$ detections are considered as genuine, or not.
LF acknowledges financial support from the Alexander von Humboldt Foundation. TM acknowledges financial support from Belspo for contract PRODEX GAIA-DPAC. LF thanks Stefano Bagnulo and Konstanze Zwintz for fruitful discussions. SH and MS thank Thomas Szeifert for providing the pipeline for the FORS spectra extraction. We thank the referee, Gautier Mathys, for his useful comments. This research has made use of the SIMBAD and ViZieR databases, and of the WEBDA database, operated at the Department of Theoretical Physics and Astrophysics of the Masaryk University.
Angel, J. R. P. & Landstreet, J. D. 1970, , 160, L147 Appenzeller, I. & Rupprecht, G. 1992, The Messenger, 67, 18 Appenzeller, I., Fricke, K., F[ü]{}rtig, W., et al. 1998, The Messenger, 94, 1 Auri[è]{}re, M., Wade, G. A., Silvester, J., et al. 2007, , 475, 1053 Babel, J. & Montmerle, T. 1997, , 485, L29 Bagnulo, S., Szeifert, T., Wade, G. A., Landstreet, J. D. & Mathys, G. 2002, , 389, 191 Bagnulo, S., Jehin, E., Ledoux, C., et al. 2003, The Messenger, 114, 10 Bagnulo, S., Landolfi, M., Landstreet, J. D., et al. 2009, , 121, 993 Bagnulo, S., Landstreet, J. D., Fossati, L. & Kochukhov, O. 2012, , 538, A129 Bagnulo, S., Fossati, L., Kochukhov, O. & Landstreet, J. D. 2013, , 559, A103 Bagnulo, S., Fossati, L., Landstreet, J. D. & Izzo, C. 2015, , submitted Borra, E. F., Landstreet, J. D. & Thompson, I. 1983, , 53, 151 Breger, M., Stich, J., Garrido, R., et al. 1993, , 271, 482 Bychkov, V. D., Bychkova, L. V. & Madej, J. 2005, , 430, 1143 Bychkov, V. D., Bychkova, L. V. & Madej, J. 2009, , 394, 1338 Castro, N., Fossati, L., Hubrig, S., et al. 2015, , in press (arXiv: 1507.03591) Catalano, F. A. & Leone, F. 1996, , 311, 230 Clari[á]{}, J. J. 1974, , 37, 229 Donati, J.-F., Semel, M. & Rees, D. E. 1992, , 265, 669 Drilling, J. S. 1981, , 250, 701 Dufton, P. L., Langer, N., Dunstall, P. R., et al. 2013, , 550, A109 Eikenberry, S. S., Chojnowski, S. D., Wisniewski, J., et al. 2014, , 784, L30 Fossati, L., Castro, N., Morel, T., et al. 2015a, , 574, A20 Fossati, L., Bagnulo, S., Landstreet, J. D. & Kochukhov, O. 2015b, in Physics and evolution of magnetic and related stars, ed. Y. Y. Balega, I. I. Romanyuk, & D. O. Kudryavtsev (San Francisco: ASP), ASP Conf. Ser., 494, 63 (arXiv: 1502.00779) Fourtune-Ravard, C., Wade, G. A., Marcolino, W., et al. 2011, in Active OB stars: structure, evolution, mass loss, and critical limits, Proc. International Astronomical Union (Cambridge: CUP), IAU Symp., 272, 180 Howarth, I. D., Siebert, K. W., Hussain, G. A. J. & Prinja, R. K. 1997, , 284, 265 Hubrig, S., Kurtz, D. W., Bagnulo, S., et al. 2004a, , 415, 661 Hubrig, S., Szeifert, T., Schöller, M., et al. 2004b, , 415, 685 Hubrig, S., Briquet, M., Sch[ö]{}ller, M., t al. 2006, , 369, L61 Hubrig, S., Briquet, M., De Cat, P., et al. 2009, Astronomische Nachrichten, 330, 317 Hubrig, S., Ilyin, I., Sch[ö]{}ller, M., et al. 2011, , 726, L5 Hubrig, S., Fossati, L., Carroll, T. A., et al. 2014a, , 564, L10 Hubrig, S., Sch[ö]{}ller, M. & Kholtygin, A. F. 2014b, , 440, 1779 Jurkevich, I. 1971, , 13, 154 Landstreet, J. D., Borra, E. F., Angel, J. R. P. & Illing, R. M. E. 1975, , 201, 624 Landstreet, J. D., Bagnulo, S. & Fossati, L. 2014, , 752, A113 Langer, N. 1998, , 329, 551 Ma[í]{}z Apell[á]{}niz, J., Pellerin, A., Barb[á]{}, R. H., et al. 2012, in ASP Conf. Ser., 465, ed. L. Drissen, et al., 484 Mayor, M., Pepe, F., Queloz, D., et al. 2003, The Messenger, 114, 20 Morel, T., Castro, N., Fossati, L., et al. 2014, The Messenger, 157, 27 Morel, T., Castro, N., Fossati, L., et al. 2015, IAU Symposium, 307, 342 Netopil, M., Paunzen, E., Maitzen, H. M., North, P. & Hubrig, S. 2008, , 491, 545 Petit, V., Owocki, S. P., Wade, G. A., et al. 2013, , 429, 398 Piskunov, N., Snik, F., Dolgopolov, A., et al. 2011, The Messenger, 143, 7 Press, W. H., Teukolsky, S. A., Vetterling, W. T. & Flannery, B. P. 1992, Numerical recipes in C. The art of scientific computing, Cambridge: University Press, Vol. 6, No. 3 Ram[í]{}rez-Agudelo, O. H., Sim[ó]{}n-D[í]{}az, S., Sana, H., et al. 2013, , 560, A29 Rivinius T., Szeifert T., Barrera L., et al. 2010, , 405, L46 Saesen, S., Briquet, M., & Aerts, C. 2006, Communications in Asteroseismology, 147, 109 Shultz, M., Wade, G. A., Grunhut, J., et al. 2012, , 750, 2 Shultz, M., Wade, G., Rivinius, T., Marcolino, W., Henrichs, H. & Grunhut, J. 2015, [$\xi$]{}$^{1}$ CMa: An Extremely Slowly Rotating Magnetic B0.7 IV Star, Proceedings of IAUS 307: New windows on massive stars: asteroseismology, interferometry and spectropolarimetry, 399 Silvester, J., Neiner, C., Henrichs, H. F., et al. 2009, , 398, 1505 Sim[ó]{}n-D[í]{}az, S. & Herrero, A. 2014, , 562, A135 Snik, F., Kochukhov, O., Piskunov, N., et al. 2011, in ASP Conf. Ser., 437, ed. J. R. Kuhn, et al., 237 Sota, A., Ma[í]{}z Apell[á]{}niz, J., Walborn, N. R., et al. 2011, , 193, 24 Steffen, M., Hubrig, S., Todt, H., et al. 2014, , 570, A88 Stellingwerf, R. F. 1978, , 224, 953 Tody, D. 1993, in Astronomical Data Analysis Software and Systems II, ed. R. J. Hanisch, R. J. V. Brissenden, & J. Barnes (San Francisco: ASP), ASP Conf. Ser., 52, 173 Wade, G. A., Bagnulo, S., Drouin, D., Landstreet, J. D. & Monin, D. 2007, , 376, 1145 Wade, G. A., Ma[í]{}z Apell[á]{}niz, J., Martins, F., et al. 2012, , 425, 1278 Wade, G. A., Grunhut, J., Alecian, E., et al. 2014, Proceedings of IAUS 302: Magnetic fields throughout stellar evolution, 265 Wolff, S. C. & Morrison, N. D. 1974, , 86, 935 Zboril, M. & North, P. 1998, Contributions of the Astronomical Observatory Skalnate Pleso, 27, 371 Zima, W. 2008, Communications in Asteroseismology, 157, 387
[^1]: F.R.S.-FNRS Postdoctoral Researcher, Belgium
[^2]: Based on observations made with ESO Telescopes at the La Silla Paranal Observatory under programme ID191.D-0255(A,C).
[^3]: The E2V CCDs have a nominal gain (conversion from counts to electrons) of 2.20 and a readout noise (in electrons) of 4.20, while the MIT CCDs have a nominal gain of 1.25 and a readout noise of 2.70.
[^4]: Image Reduction and Analysis Facility (IRAF – [http://iraf.noao.edu/]{}) is distributed by the National Optical Astronomy Observatory, which is operated by the Association of Universities for Research in Astronomy (AURA) under cooperative agreement with the National Science Foundation.
[^5]: For polarisation measurements, from a mathematical point of view the flat-field correction has no influence on the results. However, @bagnulo2012 showed that in practice this is not the case, most likely because of fringing, but it is not possible to clearly identify the best option.
[^6]: Optionally, the IDL routine allows one to calculate the uncertainty of Stokes $V$ using the simplified formulation given in Eq. A6 of @bagnulo2009, which is valid for low polarisation values.
[^7]: http://idlastro.gsfc.nasa.gov/
[^8]: Note that there is a slight difference between the and measurements reported here (Table \[tab:mag.field\]) and that given by @hubrig2014 because of a more recent update in the Bonn pipeline.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We study the rapid variability in Cyg X-1 observed with the Rossi Timing Explorer (RXTE) on Jan 17 and 20 of 1997. The power spectra below 200 Hz can be characterized by two “shot noise” components and a peaked-noise centered at 0.2 Hz. This is similar to those “shot noise” observed in another black hole candidate GRO J0422+32 and a neutron star X-ray binary 1E 1724-3045. The similarity suggests that the generation mechanism of “shot noise” and peaked-noise is probably similar in both black hole systems and neutron star systems, and in both high mass X-ray binaries (HMXBS) and low mass X-ray binaries (LMXBs). We have also analyzed $\sim$ 500 brightest shots selected from the Cyg X-1 light curves. The time scale of the spectral variation around the shot can be as long as $\sim$ a few seconds, consistent with Ginga results. The difference between the superposed shot profile in different energy bands can be attributed to the time lag and the different shot width between different energy bands. The shot width defined from the auto-correlation function of shot profile does not show a bimodal distribution.'
address: |
Laboratory for Cosmic Ray and High Energy Astrophysics\
Institute of High Energy Physics\
Chinese Academy of Sciences\
P.O.Box: 918-3\
Beijing, 100039\
P. R. China\
author:
- Wenfei Yu and Tipei Li
title: 'Similar Shot Noise in Cyg X-1, GRO J0422+32 and 1E 1724-3045'
---
[Email: [email protected], [email protected]]{}
Introduction
============
The aperiodic X-ray variability of the black hole X-ray binary Cyg X-1 and other stellar black hole candidates has been described by the phenomenological “shot noise” models (e.g. [@ref1][@ref2] [@ref3][@ref4] and references therein). Such rapid X-ray variability or flickering is more pronounced during their hard state [@ref5][@ref6][@ref7]. In hard states, the power density spectra show a flat top followed by a power law at a certain break frequency. The higher amplitude of the rapid variability is, the lower the break frequency is. This suggests that the shot properties vary with time in the hard states.
In the framework of “shot noise” models, the properties of the shots and superposed shot profiles in Cyg X-1 have been studied with data obtained from Uhuru, HEAO 1 A-2, EXOSAT/ME, Ginga, and RXTE/PCA (see [@ref2] and references therein;[@ref8][@ref9][@ref10]). In summary, the shot properties are as follows: (1) The shot has nearly time symmetric rise and decay lasting for up to a few seconds; (2) The energy spectrum of the shot changes with time. (3) The shot duration changes with the states.
Recently, striking timing similarities of the power density spectra between the black hole candidate GRO J0422+32 (Nova Persei 1992) [@ref11][@ref12] and the X-ray burster 1E 1724-3045[@ref13] were reported. In this paper, we show similar aperiodic variability in Cyg X-1 in hard state and compare the shot noise properties with those of GRO J0422+32 and 1E 1724-3045. We also present our results from the study of about $\sim$ 500 shots observed in Cyg X-1 with RXTE/PCA.
RXTE/PCA Observation of Cyg X-1
===============================
We analyze the data obtained from RXTE/PCA observations conducted on 1997 Jan 17 and 20. The entire PCA observation lasted for about 12.5 hours. The average count rates in the entire PCA band for the two days are 4330 and 3880 cps, respectively. High time resolution ($\sim$ 2$^{-12}$ s) [*Single-Bit*]{} mode data in the energy range 1.0-5.1 keV, 5.1-8.7 keV, 8.7-18.3 keV and 18.3-98.5 keV were used in studying the aperiodic variability. We combine the data in the 4 bands to calculate the average power density spectra (PDS). The average PDS of Jan 20 is plotted in Fig.1. The PDS displays a flat top below a low frequency break around 0.03 Hz, a peaked-noise component centered at 0.2 Hz with a FWHM $\sim$ 0.2 Hz and a second break frequency around 3 Hz.
Similar to the previous study of GRO J0422+32[@ref11][@ref12] and 1E 1724-3045[@ref13], we apply a model consisting of two “shot noise” components characterized by two Lorentzian functions in the frequency range below 0.1 Hz and above 1.0 Hz (dashed line and dash-dot line), and fit the residual noise power in the frequency range 0.1–1.0 Hz with a model composed of a linear rise with an exponential decay (solid line in the inset panel). The model is shown as solid line in the Figure 1.
Comparison with Observations of GRO J0422+32 and 1E 1724-3045
=============================================================
The PDS of OSSE observation to GRO J0422+32[@ref11][@ref12] and the PDS of 1E 1724-3045 observed with RXTE/PCA[@ref13] can also be characterized by two “shot noise” components and one peaked-noise feature. The characteristic duration of the two “shot noise” components ($\tau_1$ or $\tau_2$) is represented by the Half Width of Half Maximum (HWHM) of each Lorentzian as $$\tau_{1,2}=\frac{1}{2\pi~HWHM}$$ The parameters of the noise components of the three sources are compared in Table 1.
X-ray Shots of Cyg X-1 in Hard State
====================================
To investigate the above interpretations in terms of “shot noise”, we study the X-ray shots observed with RXTE/PCA in the energy range $\sim$ 2–60 keV. The X-ray Shots were selected in the light curve. The criteria are that their peak X-ray counts in 0.125 s bin should be larger than 1250, and should be the maximum within the neighboring 5.0 s on both sides. In the entire observation, we have found 513 shots which meet the requirement.
#### Shot Width
The width of each shot is derived from the auto-correlation coefficients of 10 s light curve around each shot, $A(i)$. They are defined as follows: $$A(i)=\frac{\sum_{k=0}^{N-i-1}(X_{k}-\bar{X})(X_{k+i}-\bar{X})}
{\sum_{k=0}^{N-1}(X_{k}-\bar{X})^2}$$ where $\delta{t}$ is $\sim$ 2.44 ms, the time resolution of the light curves, and $A(i)$ ($i=0,...,N-1$) the auto-correlation coefficient at $i\delta{t}$. Then we define the average shot width as $${T_{width}}=2.0\times\sqrt{\frac{\sum_{k=1}^{M}k^2A(k)+(0.25)^2A(0)}{\sum_{k=0}^{M}A(k)}}\delta{t}$$ where 0.25 represents the average time shift of the central bin A(0), and $M$ is the maximum of $i$ with $A(i)$ no less than 0.0 in the main peak of the auto-correlation function. In Fig.2, we show the profile of one of the brightest shot. The profile was obtained from the [*Standard 1*]{} data mode. The corresponding auto-correlation coefficients are shown in Fig.3. They were obtained from [*Single-Bit* ]{} data and bined to a time resolution of $\sim$ 2.44 ms. The coefficients used in the calculation of shot width are shaded in the figure. The distribution of shot width for the 513 shots is shown in Fig.4.
#### Peak-aligned Shot Profile and Spectral Variation
We superposed the 513 shots by aligning them at the center of their 0.125 s peak bins (combine all 4 energy channels). The time resolution used in the alignment is $\sim$ 2.44 ms. In the upper panel of Fig.5, we show the peak-aligned shot profiles of the soft band (channel 1+2) and the hard band (channel 3+4), respectively. They were normalized to their peak counts. In the lower panel of the same figure, we plot the residuals of the profile subtraction (3+4)–(1+2). In general, the peak-aligned profiles in Fig.5 shows: (1)The time scale of the shot rise in the hard band is smaller than that in the soft band; (2)The shot decay in the hard band is slower than the decay in the soft band; (3)There are more hard photons after the shot peak than those before the shot peak, indicating a spectral hardening.
Three factors would introduce the difference between the profiles in the soft band and those in the hard band and the observed residuals in the lower panel of Fig.5. One is the time lag of the shot rise between the soft band and the hard band. The other is the slower decay in the hard band compared with that of the soft band. The third is that the shot peak in the hard band is narrower than that in the soft band. To study the spectral variation, we plot the ratio between the peak-aligned profile in the hard band and that in the soft band in Fig.6. Both sides around the peak were fit to a 5-degree polynomial. The spectra after the shot peak is harder than that before the shot peak, as shown in Fig.6. This is consistent with Ginga results[@ref8].
Summary
=======
In summary, we have obtained the following results:
- We have found that the aperiodic X-ray variability in Cyg X-1 in hard state is similar to those observed in another black hole candidate GRO J0422+32 and a neutron star X-ray binary 1E 1724-3045. Based on the mass of companion star and the mass of the central object in the three X-ray binaries, we conclude that the generation mechanism of the X-ray shots is probably independent on the mass of the companion star, the mass of the central compact object (BH or NS), and the type of accretion.
- The spectral evolution around the shot peak could last for as long as a few seconds, and there is a spectral hardening after the shot peak. These are consistent with the previous study of the X-ray shots in Cyg X-1 observed with Ginga[@ref8].
- The duration of $\sim$ 513 bright shots in Cyg X-1, defined from their auto-correlation coefficients, ranging from $\sim$ 0.1 s to $\sim$ 2.0 s, is not bimodally distributed. This does not support the assumption that there are two kinds of shots with different duration as inferred from the power spectra. Thus the attribution of each of the noise component to a group of shots with a certain characteristic duration is probably wrong.
Source Name GRO J0422+32 1E 1724-3045 Cyg X-1
---------------------------------------- ----------------------------------- ---------------- ------------------
Binary Type LMXB LMXB HMXB
Accretion Type Disk Disk Wind + Disk
Compact Object BH NS BH
Central Mass (M$\odot$)$^{(4)}$ 3.2-3.9 or $>$ 9 1.4-2.0 16$\pm$5
Source State Hard Island, Low Hard
Rms Amplitude 30% (35-60 keV), 40% (75-175 keV) 25% (2-20 keV) 39% (1-98.5 keV)
Characteristic Duration $\tau_{1}$ (s) $\sim$ 2.2 $\sim$ 0.7 $\sim$ 2.2
Characteristic Duration $\tau_{2}$ (s) $\sim$ 0.050 $\sim$ 0.017 $\sim$ 0.070
Peaked-Noise Frequency (Hz) $\sim$ 0.2 $\sim$ 0.8 $\sim$ 0.2
\
[*Note:*]{}\
(1) [@ref11]; (2) [@ref13]; (3) Yu, W. 1998, Ph.D thesis; (4) [@ref14] and references therein.
Acknowledgments {#acknowledgments .unnumbered}
===============
WY appreciate various supports and helpful discussions and comments by Prof. J. Van Paradijs, Dr. C. Kouveliotou and Dr. M. Finger at NASA/MSFC.
References {#references .unnumbered}
==========
[99]{}
N.J. Terrell, ApJ, 174, L35 (1972)
J.C. Lochner, J.H. Swank, & A.E. Szymkowiak, ApJ, 375, 295 (1991)
T. Manmoto et al. ApJ, 464, L135 (1996)
M.A. Abramowicz et al. ApJ, 489, 819 (1997)
S. Miyamoto et al. ApJ, 391, L21 (1992)
T. Belloni & G. Hasinger, A&A, 230, 103 (1990)
D.J. Crary et al. ApJ, 462, L71 (1996)
H. Negoro, S. Miyamoto, & S. Kitamoto, 1994, ApJ, 423, L127 (1994)
H. Negoro et al. ApJ, 452, L49 (1995)
Y.X. Feng, T.P. Li & L. Chen, ApJ, in press (astro-ph/9810122) (1999)
J.E. Grove et al., in C. Fichtel, N. Gherels and J. Norris (Eds), [*The Second Compton Symposium*]{}, AIP No. 304, 192, AIP Press (1993)
J.E. Grove et al., ApJ, 502, L45 (astro-ph/9805256) (1998)
J.F. Olive et al., A&A, 333, 942 (astro-ph/9802053) (1998)
J. van Paradijs, to appear in R. Buccheri, J. van Paradijs, M.A. Alpar (Eds), [*The Many Faces of Neutron Stars*]{}, Kluwer Academic Publishers, astro-ph/9802177 (1998)
|
{
"pile_set_name": "ArXiv"
}
|
6.5in 9.0in
=cmbx10 scaled 2[February, 2002]{}
1.5 cm
**Implications of Recent $\ov B^0\to D^{(*)0}
X^0$ Measurements**
**Hai-Yang Cheng**
Institute of Physics, Academia Sinica
Taipei, Taiwan 115, Republic of China
and
Physics Department, Brookhaven National Laboratory
Upton, New York 11973
**Abstract**
5 cm
PACS numbers: 13.25.-k
Introduction
============
For some time $B\to\J K$ and $B\to \J K^*$ remain to be the only color-suppressed $B$ meson two-body decay modes that have been measured experimentally. Recently, the long awaited color-suppressed decay modes $\ov B^0\to D^{(*)0}\pi^0$ are finally measured by both Belle [@Belle] and CLEO [@CLEO] with the $D^0\pi^0$ branching ratio larger than the upper limit previously reported [@CLEOa]. The channels $\ov B^0\to
D^{(*)0}\eta$ and $\ov B^0\to D^{(*)0}\omega$ are also observed by Belle [@Belle]. We shall see below that the theoretical predictions based on the factorization approach in general are too small to account for the observed decay rates of color-suppressed modes $D^{(*)0}X^0$ with $X=\pi,\eta,\omega$. This has important implications for final-state interactions (FSIs).
Under the factorization hypothesis, the nonleptonic decay amplitudes are approximated by the factorized hadronic matrix elements multiplied by some universal, process-independent effective coefficients $a_i$. Based on the factorization assumption, one can catalog the decay processes into three classes. For class-I decays, the decay amplitudes, dominated by the color-allowed external $W$-emission, are proportional to $a_1\la O_1\ra_{\rm fact}$ where $O_1$ is a charged current–charged current 4-quark operator. For class-II decays, the decay amplitudes, governed by the color-suppressed internal $W$-emission, are described by $a_2\la O_2\ra_{\rm fact}$ with $O_2$ being a neutral current–neutral current 4-quark operator. The decay amplitudes of the class-III decays involve a linear combination of $a_1\la O_1\ra_{\rm fact}$ and $a_2\la O_2\ra_{\rm
fact}$. If factorization works, the effective coefficients $a_i$ in nonleptonic $B$ or $D$ decays should be channel by channel independent.
What is the relation between the coefficients $a_i$ and the Wilson coefficients in the effective Hamiltonian approach ? Under the naive factorization hypothesis, one has \[nf\] a\_1()=c\_1()+[1N\_c]{}c\_2(), a\_2()=c\_2()+[1N\_c]{}c\_1(), for decay amplitudes induced by current-current operators $O_{1,2}(\mu)$, where $c_{1,2}(\mu)$ are the corresponding Wilson coefficients and $N_c$ is the number of colors. In the absence of QCD corrections, $c_1=1$ and $c_2=0$, and hence class-II modes governed by $a_2=1/N_c$ are obviously “color-suppressed". However, this naive factorization approach encounters two principal difficulties: (i) the coefficients $a_i$ given by Eq. (\[nf\]) are renormalization scale and $\gamma_5$-scheme dependent, and (ii) it fails to describe the color-suppressed class-II decay modes. For example, the ratio $R=\Gamma(D^0\to\ov
K^0\pi^0)/\Gamma(D^0\to K^-\pi^+)$ is predicted to be only of order $3\times 10^{-4}$ due to the smallness of $a_2$ in the naive factorization approach, while experimentally it is measured to be $0.55\pm 0.06$ [@PDG]. It is known that the decay $D^0\to\ov
K^0\pi^0$ is enhanced by two mechanisms. First, $a_2$ receives a large nonfactorizable correction. Second, the weak decay $D^0\to
K^-\pi^+$ followed by the inelastic rescattering $K^-\pi^+\to\ov
K^0\pi^0$ can raise $\B(D^0\to\ov K^0\pi^0)$ dramatically by lowering $\B(D^0\to K^-\pi^+)$.
Beyond naive factorization the parameters $a_{1,2}$ have the general expression a\_[1,2]{}= c\_[2,1]{}()+[c\_[1,2]{}()N\_c]{} +[nonfactorizable corrections]{}, where nonfactorizable corrections include vertex corrections, hard spectator interactions involving the spectator quark of the heavy meson, and FSI effects from inelastic rescattering, resonance effects, $\cdots$, etc. In the generalized factorization approach of [@Ali; @CT98], one includes the vertex corrections which will compensate the renormalization scale and $\gamma_5$-scheme dependence of the Wilson coefficients to render $a_{1,2}$ scale and scheme independent. Contrary to the naive one, the improved generalized factorization scheme assumes that nonfactorizable effects are incorporated in a process independent form. Since not all nonfactorizable effects are calculable by perturbative QCD, one will treat $a_1$ and $a_2$ as free parameters in the generalized factorization approach and extract them from experiment. The phenomenological analysis of two-body decay data of $D$ and $B$ mesons will tell us if the generalized factorization hypothesis works reasonably well by studying the variation of the parameters $a_{1,2}$ from channel to channel.
The experimental measurement of $B\to\J K$ leads to $|a_2(\J
K)|=0.26\pm 0.02$ [@a1a2]. This seems to be also supported by the study of $B\to D\pi$ decays: Assuming no relative phase between $a_1$ and $a_2$, the result $a_2\sim {\cal O}(0.20-0.30)$ [@a1a2; @ns] is inferred from the data of $\ov B^0\to
D^{(*)+}\pi^-$ and $B^-\to D^{(*)0}\pi^-$. However, as we shall show below, the above value of $a_2$ leads to too small decay rates for $\ov B^0\to D^{(*)0}\pi^0$ when compared to recent measurements. In order to account for the observation, one needs a larger $a_2(D\pi)$ with a non-trivial phase relative to $a_1$. The importance of FSIs has long been realized in charm decay since some resonances are known to exist at energies close to the mass of the charmed meson. We shall see in this work that, just as $D^0\to\bar K^0\pi^0$, both nonfactorizable effects and FSIs are also needed to explain the data of $\ov B^0\to D^{(*)0}\pi^0$, though these two effects in $B$ decays are naively expected to be not as dramatic as in the charm case.
The color-suppressed mode is a very suitable place for studying the effect of FSIs (especially the soft one) in weak decays. The ratio of the color-suppressed decay amplitudes with and without FSIs is $R_{\ov K\pi} \equiv|A(D^0\to \ov K^0\pi^0)/A(D^0\to \ov
K^0\pi^0)_{\rm without~FSIs}|\approx 2.0$ and the relative phase between $D^0\to \ov K^0\pi^0$ and $D^0\to K^-\pi^+$ is about $150^\circ$. It is expected that for $\ov B\to D\pi$ decay, $R_{D\pi}$ and the relative phase among decay amplitudes will become smaller. The recent measurement of the $\ov B^0\to
D^0\pi^0$ mode allows us to determine the above two quantities. We shall see that although the relative phase among $\ov B\to D\pi$ decay amplitudes becomes smaller, $R_{D^{(*)}\pi}$ does not decrease in a significant way from charm to bottom case. The implications and related physics will be discussed below in details.
Factorization
=============
We begin with by considering the branching ratios of the color-suppressed modes $\ov B^0\to D^{(*)0}X^0$ $(X=\pi,\eta,\omega)$ within the framework of the factorization approach. The $\ov B^0\to D^0\pi^0$ amplitude is given by \[D0pi0amp\] A(B\^0D\^0\^0) &=& [1]{}(-+), where $\C,~\E$ are color-suppressed internal $W$-emission and $W$-exchange amplitudes, respectively. In terms of the factorized hadronic matrix elements, they read \[CE\] [C]{} &=& i [G\_F]{}V\_[cb]{}V\_[ud]{}\^\*a\_2(D)(m\_B\^2-m\_\^2)f\_D F\_0\^[B]{}(m\_D\^2),\
[E]{} &=& i [G\_F]{}V\_[cb]{}V\_[ud]{}\^\*a\_2(D)(m\_D\^2-m\_\^2)f\_B F\_0\^[0D]{}(m\_B\^2), where $a_2(D\pi)$ is a parameter to be determined from experiment. The annihilation form factor $F_0^{0\to D\pi}(m_B^2)$ is expected to be suppressed at large momentum transfer, $q^2=m_B^2$, corresponding to the conventional helicity suppression. Based on the argument of helicity and color suppression, one may therefore neglect short-distance (hard) $W$-exchange contributions. However, it is not clear if the long-distance contribution to $W$-exchange is also negligible. Likewise, A(B\^0D\^0) &=& i[G\_F]{}V\_[cb]{}V\_[ud]{}\^\*a\_2(D)(m\_B\^2-m\_\^2)f\_D F\_0\^[B]{}(m\_D\^2),\
A(B\^0D\^[\*0]{}\^0) &=& -[G\_F]{}V\_[cb]{}V\_[ud]{}\^\*a\_2(D\^\*)m\_[D\^\*]{}f\_[D\^\*]{} F\_1\^[B]{}(m\_[D\^\*]{}\^2),\
A(B\^0D\^0) &=& [G\_F]{}V\_[cb]{}V\_[ud]{}\^\*a\_2(D)2m\_ f\_D A\_0\^[B]{}(m\_D\^2), and A(B\^0D\^[\*0]{}) &=& -i[G\_F]{}V\_[cb]{}V\_[ud]{}\^\*a\_2 (D\^\*)f\_m\_. Here factorization implies a universal $a_2$, namely, $a_2(D^*\omega)=a_2(D\omega)=a_2(D\eta)=a_2(D^*\pi)=a_2(D\pi)$. In naive factorization, $a_2$ is not only small, of order 0.10, but also renormalization scale and scheme dependent. In the generalized factorization approach, the scale- and scheme-independent $a_2$ can be extracted from experiment and the factorization hypothesis is tested by studying $a_2$ to see if it is process independent or insensitive.
To proceed, we shall consider four distinct form-factor models: the Neubert-Rieckert-Stech-Xu (NRSX) model [@NRSX], the relativistic light-front (LF) quark model [@cch], the Neubert-Stech (NS) model [@ns], and the Melikhov-Stech (MS) model based on the constituent quark picture [@Melikhov]. The NRSX model takes the Bauer-Stech-Wirbel (BSW) model [@bsw] results for the form factors at zero momentum transfer but makes a different ansatz for their $q^2$ dependence, namely, a dipole behavior is assumed for the form factors $F_1,~A_0,~A_2,~V$, motivated by heavy quark symmetry, and a monopole dependence for $F_0,A_1$, where we have followed the definition of form factors given in [@bsw]. For reader’s convenience, the values of relevant form factors are listed in Table I (see [@a1a2] for some details about the NS model).
The form factors for $B\to\eta$ and $B\to\eta'$ transitions have been calculated by BSW [@bsw] in a relativistic quark model. However, in their relativistic quark model calculation of $B\to
\eta^{(')}$ transitions, BSW considered only the $u\bar u$ component of the $\eta$ and $\eta'$; that is, the form factors calculated by BSW are actually $F_0^{B\eta_{u\bar u}}$ and $F_0^{B\eta'_{u\bar u}}$ induced from the $b\to u$ transition. It is thus more natural to consider the flavor basis of $\eta_q$ and $\eta_s$ defined by \_q=[1]{}(u|u+d|d), \_s=s|s. The wave functions of the $\eta$ and $\eta'$ are given by ()=()(), where $\phi=\theta+{\rm arctan}\sqrt{2}$, and $\theta$ is the $\eta\!-\!\eta'$ mixing angle in the octet-singlet basis. The physical form factors then have the simple expressions: \[Beta\] F\_[0,1]{}\^[B]{}=[1]{}F\_[0,1]{}\^[B\_[u|u]{}]{}, && F\_[0,1]{}\^[B’]{}=[1]{} F\_[0,1]{}\^[B’\_[u|u]{}]{}. Using $F_0^{B\eta_{u\bar u}}(0)=0.307$ and $F_0^{B\eta'_{u\bar
u}}(0)=0.254$ obtained from [@bsw] and the mixing angle $\phi=39.3^\circ$ (or $\theta=-15.4^\circ$) [@Kroll] we find $F_0^{B\eta}(0)=0.168$ and $F_0^{B\eta'}(0)=0.114$ in the BSW model and hence the NRSX model. For other form-factor models,[^1] we shall apply the relation based on isospin-quartet symmetry \[isospinquart\] F\_[0,1]{}\^[B\_[u|u]{}]{}=F\_[0,1]{}\^[B’\_[u| u]{}]{}=F\_[0,1]{}\^[B]{} and Eq. (\[Beta\]) to obtain the physical $B-\eta$ and $B-\eta'$ transition form factors.
$F_0^{B\pi}(m_D^2)$ $F_1^{B\pi}(m^2_{D^*})$ $F_0^{B\eta}(m_D^2)$ $F_1^{B\eta}(m_\D^2)$ $F_0^{BD}(m_\pi^2)$ $A_0^{B\D}(m_\pi^2)$ $A_0^{B\omega}(m_D^2)$ $A_1^{B\omega}(m_\D^2)$ $A_2^{B\omega}(m_\D^2)$ $V^{B\omega}(m_\D^2)$
------ --------------------- ------------------------- ---------------------- ----------------------- --------------------- ---------------------- ------------------------ ------------------------- ------------------------- ----------------------- --
NRSX 0.37 0.45 0.19 0.23 0.69 0.62 0.26 0.23 0.27 0.32
LF 0.34 0.39 0.18 0.22 0.70 0.73 0.25 0.16 0.17 0.28
MS 0.32 0.36 0.17 0.20 0.67 0.69 0.26 0.20 0.21 0.28
NS 0.27 0.32 0.15 0.18 0.63 0.64 0.22 0.20 0.22 0.22
: Form factors in various form-factor models. Except for the NRSX model, the relations $A_i^{B\omega}(q^2)=A_i^{B\rho^0}(q^2)$ $(i=0,1,2)$ and $V^{B\omega}(q^2)=V^{B\rho^0}(q^2)$ are assumed in all the form-factor models. The pion in the $B-\pi$ transition is referred to the charged one.
As mentioned in the Introduction, in the absence of a relative phase between $a_1$ and $a_2$, a value of $a_2$ in the range of 0.20 to 0.30 is inferred from the data of $\ov B^0\to
D^{(*)+}\pi^-$ and $B^-\to D^{(*)0}\pi^-$. For definiteness, we shall use the representative value $a_2=0.25$ for the purpose of illustration. The calculated branching ratios for $\ov B^0\to
D^{(*)0}X^0$ are shown in Table II for $f_D=200$ MeV and $f_\D=230$ MeV. Evidently, the predicted rates for color suppressed modes are too small compared to recent measurements. It should be stressed that if there is no relative phase between $a_1$ and $a_2$, then one cannot increase $a_2$ arbitrarily to fit the data as this will enhance the decay rate of the $\Delta I=3/2$ mode $B^-\to D^{(*)0}\pi^-$ and destroy the agreement between theory and experiment for the charged mode. For example, fitting $a_2$ to the data of $D^0\pi^0$ without FSIs will yield $a_2=0.45$ in the MS model, which in turn implies $\B(B^-\to
D^0\pi^-)=7.9\times 10^{-3}$ and this is obviously too large compared to the experimental value $(5.3\pm0.5)\times 10^{-3}$ [@PDG]. In this case, one needs FSIs to convert $D^+\pi^-$ into $D^0\pi^0$. In contrast, if $a_2$ is of order 0.45, then a relative strong phase between $a_1$ and $a_2$ will be needed in order not to over-estimate the $D^0\pi^-$ rate. In either case, we conclude that FSIs are the necessary ingredients for understanding the data.
------------------------------ ------------------------ ---------------------- ---------------------- ---------------------- ----------------------------- --------------------------------
\[0cm\]\[0cm\][Decay mode]{} \[0cm\]\[0cm\][NRSX]{} \[0cm\]\[0cm\][LF]{} \[0cm\]\[0cm\][MS]{} \[0cm\]\[0cm\][NS]{} Belle [@Belle] CLEO [@CLEO]
$\ov B^0\to D^0\pi^0$ 1.13 0.93 0.82 0.58 $3.1\pm0.4\pm0.5$ $2.74^{+0.36}_{-0.32}\pm 0.55$
$\ov B^0\to D^{*0}\pi^0$ 1.57 1.20 1.01 0.80 $2.7^{+0.8+0.5}_{-0.7-0.6}$ $2.20^{+0.59}_{-0.52}\pm0.79$
$\ov B^0\to D^0\eta$ 0.55 0.53 0.48 0.34 $1.4^{+0.5}_{-0.4}\pm 0.3$
$\ov B^0\to D^{*0}\eta$ 0.76 0.68 0.58 0.46 $2.0^{+0.9}_{-0.8}\pm0.4$
$\ov B^0\to D^0\omega$ 0.76 0.71 0.76 0.54 $1.8\pm0.5^{+0.4}_{-0.3}$
$\ov B^0\to D^{*0}\omega$ 1.60 1.16 1.75 1.35 $3.1^{+1.3}_{-1.1}\pm0.8$
------------------------------ ------------------------ ---------------------- ---------------------- ---------------------- ----------------------------- --------------------------------
: Predicted branching ratios (in units of $10^{-4}$) of $\ov B^0\to D^{(*)0}X^0$ $(X=\pi,\eta,\omega)$ in the generalized approach with various form-factor models for $a_2=0.25$, $f_D=200$ MeV and $f_\D=230$ MeV.
Extraction of $\lowercase{a_1}$ and $\lowercase{a_2}$
=====================================================
In this section we will extract the parameters $a_1$ and $a_2$ in two different approaches. In the first approach, the topological amplitudes are assumed to incorporate all the information of strong interactions. Therefore, $a_{1,2}$ thus determined already include the effects of FSIs. In the second approach, one will assume that quark-diagram topologies in their original forms do not include FSIs from the outset.
Direct analysis
---------------
In terms of the quark-diagram topologies $\T$, $\C$ and $\E$, where $\T$ is the color-allowed external $W$-emission amplitude, the other $\ov B\to D\pi$ amplitudes can be expressed as \[Dpiamp\] A(B\^0D\^+\^-) &=& +,\
A(B\^-D\^0\^-) &=& +, and they satisfy the isospin triangle relation A(B\^0D\^+\^-)=A(B\^0D\^0\^0)+A(B\^-D\^0\^-). In writing Eqs. (\[D0pi0amp\]) and (\[Dpiamp\]) it has been assumed that the topologies $\T,~\C,~\E$ include the information of all strong interactions for physical $\ov B\to D\pi$ amplitudes (for an earlier discussion of quark-diagram amplitudes, see [@CC]). Now since all three sides of the $\ov B\to D\pi$ triangle are measured, we are able to determine the relative phases among the decay amplitudes. Using the data [@PDG] (B\^0D\^+\^-)=(3.00.4)10\^[-3]{}, && (B\^-D\^0\^-)=(5.30.5)10\^[-3]{},\
(B\^0D\^[\*+]{}\^-)=(2.760.21)10\^[-3]{}, && (B\^-D\^[\*0]{}\^-)=(4.60.4)10\^[-3]{}, \[data\] and the combined value of Belle and CLEO for the neutral modes (see Table II) (B\^0D\^0\^0)=(2.920.46)10\^[-4]{}, (B\^0D\^[\*0]{}\^0)=(2.470.67)10\^[-4]{}, we find (only the central values for phase angles are shown here) \[CET\] .[-+]{}|\_[D]{}=(0.440.05) e\^[i59\^]{}, && .[-+]{}|\_[D]{}=(0.340.03) e\^[i37\^]{},\
.[-+]{}|\_[D\^\*]{}=(0.420.06) e\^[i63\^]{}, &&.[-+]{}|\_[D\^\*]{}=(0.340.05) e\^[i44\^]{}, where we have employed the $B$ meson lifetimes given in [@PDG].
The same phases also can be obtained from the isospin analysis. Decomposing the physical amplitudes into their isospin amplitudes yields \[isospin\] A(B\^0D\^+\^-) &=& A\_[1/2]{}+A\_[3/2]{},\
A(B\^0D\^0\^0) &=& A\_[1/2]{} -A\_[3/2]{},\
A(B\^-D\^0\^-) &=& A\_[3/2]{}. The isospin amplitudes are related to the topological quark-diagram amplitudes via \[isospinrel\] A\_[1/2]{}=[1]{}(2-+3), A\_[3/2]{}=[1]{}(+). Intuitively, the phase shift difference between $A_{1/2}$ and $A_{3/2}$, which is of order $90^\circ$ for $D\to \ov K\pi$ modes (see below), is expected to play a minor role in the energetic $B\to D\pi$ decay, the counterpart of $D\to\ov K\pi$ in the $B$ system, as the decay particles are moving fast, not allowing adequate time for final-state interactions. Applying the relations (see e.g. [@ns]) |A\_[1/2]{}|\^2&=& |A(B\^0D\^+\^-)|\^2+|A(B\^0 D\^0\^0)|\^2-[13]{}|A(B\^-D\^0\^-)|\^2,\
|A\_[3/2]{}|\^2&=&[13]{}|A(B\^-D\^0\^-)|\^2,\
(\_[1/2]{}-\_[3/2]{}) &=& [3|A(B\^0 D\^+\^-)|\^2-2|A\_[1/2]{}|\^2-|A\_[3/2]{}|\^22 |A\_[1/2]{}||A\_[3/2]{}|]{}, we obtain \[A13\] .[A\_[1/2]{}A\_[3/2]{}]{}|\_[D]{}=(0.700.10)e\^[i29\^]{}, .[A\_[1/2]{} A\_[3/2]{}]{}|\_[D\^\*]{}=(0.740.07)e\^[i29\^]{}. Similar results are also obtained before in [@Xing; @NP] using the preliminary Belle and CLEO measurements. It is easy to check that the ratio $(\C-\E)/(\C+\E)$ in Eq. (\[CET\]) follows from Eqs. (\[isospinrel\]) and (\[A13\]). It is also interesting to compare the above results with that for $D\to \ov K^{(*)}\pi$ decays [@PDG]: .[A\_[1/2]{}A\_[3/2]{}]{}|\_[K]{}= (2.700.14)e\^[i90\^]{}, .[A\_[1/2]{}A\_[3/2]{}]{}|\_[K\^\*]{}= (3.970.25)e\^[i104\^]{}. The smaller isospin phase shift difference in $B$ decays is in accord with expectation. Notice that while $\Delta I=1/2$ and 3/2 amplitudes in $\ov B\to D^{(*)}\pi$ are of the same size, the $D\to\ov K\pi$ decays are dominated by the isospin $\Delta I=1/2$ amplitude. In the heavy quark limit, the ratio of $A_{1/2}/(\sqrt{2}A_{3/2})$ approaches to unity [@NP]. Evidently, the charm system exhibits a more deviation than the $B$ system from the heavy quark limit, as expected.
The ratio of $a_2/a_1$ can be extracted from Eq. (\[CET\]) or Eq. (\[A13\]). Noting that the factorized color-allowed tree amplitude reads \[T\] = i[G\_F]{}V\_[cb]{}V\_[ud]{}\^\*a\_1(D)(m\_B\^2-m\_D\^2)f\_F\_0\^[BD]{}(m\_\^2), and neglecting $W$-exchange contributions, we get \[a21Dpi\] .[a\_2a\_1]{}|\_[D]{}&=& (0.440.05)e\^[i59\^]{}[m\_B\^2-m\_D\^2 m\_B\^2-m\_\^2]{}[F\_0\^[BD]{}(m\_\^2) F\_0\^[B]{}(m\_D\^2)]{}\
&=& [1-.[A\_[1/2]{}A\_[3/2]{}]{}|\_[D]{}+.[A\_[1/2]{}A\_[3/2]{}]{}|\_[D]{} ]{}[f\_f\_D]{}[m\_B\^2-m\_D\^2 m\_B\^2-m\_\^2]{}[F\_0\^[BD]{}(m\_\^2) F\_0\^[B]{}(m\_D\^2)]{}. Likewise, for the $\ov B\to D^*\pi$ decays \[a21D\*pi\] .[a\_2a\_1]{}|\_[D\^\*]{}&=& (0.420.06)e\^[i63\^]{}[A\_0\^[BD\^\*]{}(m\_\^2) F\_1\^[B]{}(m\_\^2)]{} = [1-.[A\_[1/2]{}A\_[3/2]{}]{}|\_[D\^\*]{}+.[A\_[1/2]{}A\_[3/2]{}]{}|\_[D\^\*]{} ]{}[f\_ f\_]{}[A\_0\^[BD\^\*]{}(m\_\^2) F\_1\^[B]{}(m\_\^2)]{}. With the form factors given in various models, we are ready to extract $a_1$ and $a_2$ from the experimental data. The results are shown in Table III and the parameter $a_2$ falls into the range of $|a_2(D\pi)|\sim 0.35-0.60$ and $|a_2(D^*\pi)|\sim
0.25-0.50$. Note that the phases of $a_2/a_1$, $59^\circ$ for the $D\pi$ system and $63^\circ$ for $D^*\pi$, are slightly different from that given in [@NP] based on the preliminary Belle and CLEO data. We see that although $|a_2(D\pi)|$ and $|a_2(D^*\pi)|$ agree to within one standard deviation, there is a tendency that the former is slightly larger than the latter. Hence, nonfactorizable effects could be process dependent, recalling that the experimental value for $B\to\J K$ is $|a_2(\J K)|=0.26\pm
0.02$ [@a1a2].
Model $|a_1(D\pi)|$ $|a_2(D\pi)|$ $a_2(D\pi)/a_1(D\pi)$ $|a_1(\D\pi)|$ $|a_2(\D\pi)|$ $a_2(\D\pi)/a_1(\D\pi)$
--------- --------------- --------------- --------------------------------------- ---------------- ---------------- ---------------------------------------
NRSX $0.85\pm0.06$ $0.40\pm0.05$ $(0.47\pm0.05)\,{\rm exp}(i59^\circ)$ $0.94\pm0.04$ $0.31\pm0.04$ $(0.33\pm0.04)\,{\rm exp}(i63^\circ)$
LF $0.84\pm0.06$ $0.44\pm0.06$ $(0.53\pm0.06)\,{\rm exp}(i59^\circ)$ $0.80\pm0.03$ $0.36\pm0.05$ $(0.45\pm0.06)\,{\rm exp}(i63^\circ)$
MS $0.88\pm0.06$ $0.47\pm0.06$ $(0.53\pm0.06)\,{\rm exp}(i59^\circ)$ $0.85\pm0.03$ $0.389\pm0.05$ $(0.46\pm0.06)\,{\rm exp}(i63^\circ)$
NS $0.93\pm0.06$ $0.56\pm0.07$ $(0.60\pm0.07)\,{\rm exp}(i59^\circ)$ $0.91\pm0.03$ $0.44\pm0.06$ $(0.48\pm0.06)\,{\rm exp}(i63^\circ)$
: Extraction of the parameters $a_1$ and $a_2$ from the measured $B\to D^{(*)}\pi$ rates by assuming a negligible $W$-exchange contribution. Note that $a_2(D\pi)$ and $a_2(D^*\pi)$ should be multiplied by a factor of (200 MeV/$f_D$) and (230 MeV/$f_\D$), respectively.
Ideally, the parameters $a_1$ and $a_2$ will be more precisely determined if the topologies $\T,~\C$ and $\E$ can be individually extracted from experiment. Indeed, this is the case for charm decays where $\T,~\C$ and $\E$ can be determined from $D\to \ov
K\pi$, $D\to \ov K\eta$ and $D\to\ov K\eta'$ decays based on SU(3) flavor symmetry and it is found that $|\T|:|\C|:|\E|\sim 1.7: 1.3:
1.0$ [@Rosner]. Hence, the $W$-exchange amplitude that receives short-distance and long-distance contributions is not negligible at all in charm decay.[^2] Unfortunately, one cannot extract those three quark-diagram amplitudes for $B$ decays since the decay amplitudes of $\ov B^0\to D^0(\eta,\eta')$ are proportional to $(\C+\E)$, while $D^0\pi^0$ is governed by $(-\C+\E)$ \[see Eq. (\[CE\])\]. Therefore, the quark-diagram amplitudes $\C$ and $\E$ cannot be disentangled. Nevertheless, an accurate measurement of $D^0(\eta,\eta')$ will enable us to test the importance of $W$-exchange in $\ov B\to D\pi$ decays.
In principle, $a_1$ can be determined in a model-independent way from the measurement of the ratio of the decay rate of color-allowed modes to the differential semileptonic distribution at the appropriate $q^2$ [@Bjorken]: \[S\] S\_h\^[(\*)]{}=6\^2a\_1\^2f\_h\^2 |V\_[ij]{}|\^2Y\_h\^[(\*)]{}, where $V_{ij}$ is the relevant CKM matrix element and the expression of $Y_h^{(*)}$ can be found in [@ns]. Since the ratio $S_h^{(*)}$ is independent of $V_{cb}$ and form factors, its experimental measurement can be utilized to fix $a_1$ in a model-independent manner, provided that $Y_h^{(*)}$ is also independent of form-factor models. Based on the earlier CLEO data, it is found that $a_1(D\pi)=0.93\pm 0.10$ and $a_1(D^*\pi)=1.09\pm
0.07$ [@a1a2]. Needless to say, the forthcoming measurements from BaBar, Belle and CLEO will enable us to extract the model-independent $a_1$ more precisely. Note that QCD factorization predicts $a_1(D^{(*)}\pi)\approx 1.05$ in the heavy quark limit [@BBNS].
Assuming $a_2(D^{(*)}\eta^{(')})=a_2(D^{(*)}\pi)$ we see from Table IV that the predicted branching ratios of $\ov B^0\to
D^{(*)0}\eta$ are consistent with experiment. Note that the predicted rates of $D^{(*)0}(\eta, \eta')$ are the same for LF, MS and NS models since $a_2(D\pi)F_0^{B\pi}(m_D^2)$ is model independent \[see Eq. (\[CE\])\] and the form factors $F_0^{B\eta_{0}}$ and $F_0^{B\eta_8}$ are assumed to be proportional to $F_0^{B\pi}$ in these models.
[Decay mode]{} NRSX LF MS NS Experiment \[1\]
-------------------------- --------------- --------------- --------------- --------------- ---------------------------- --
$\ov B^0\to D^0\eta$ $1.43\pm0.24$ $1.69\pm0.28$ $1.69\pm0.28$ $1.69\pm0.28$ $1.4^{+0.5}_{-0.4}\pm 0.3$
$\ov B^0\to D^{*0}\eta$ $1.20\pm0.29$ $1.41\pm0.35$ $1.41\pm0.35$ $1.41\pm0.35$ $2.0^{+0.9}_{-0.8}\pm0.4$
$\ov B^0\to D^0\eta'$ $0.89\pm0.15$ $1.05\pm0.18$ $1.05\pm0.18$ $1.05\pm0.18$
$\ov B^0\to D^{*0}\eta'$ $0.72\pm0.18$ $0.85\pm0.21$ $0.85\pm0.21$ $0.85\pm0.21$
: Predicted branching ratios (in units of $10^{-4}$) of $\ov B^0\to D^{(*)0}(\eta,\eta')$ in various form-factor models by assuming $a_2(D^{(*)}\eta^{(')})=a_2(D^{(*)}\pi)$.
Effective parameters $a_1^{\rm eff}$ and $a_2^{\rm
eff}$
--------------------------------------------------
Thus far we have assumed that quark-diagram topologies include all strong-interaction effects including FSIs. It is equally well to take a different point of view on the quark-diagram topologies, namely, their original forms do not include FSIs from the outset. In this case, there is no relative strong phase between the isospin amplitudes $A_{1/2}$ and $A_{3/2}$ given by Eq. (\[isospinrel\]). Next, one puts isospin phase shifts into Eq. (\[isospin\]) to get \[isospinfsi\] A(B\^0D\^+\^-)\_[FSI]{} &=& A\_[1/2]{}e\^[i\_[1/2]{}]{}+A\_[3/2]{}e\^[i\_[3/2]{}]{},\
A(B\^0D\^0\^0)\_[FSI]{} &=& A\_[1/2]{}e\^[i\_[1/2]{}]{} -A\_[3/2]{}e\^[i\_[3/2]{}]{},\
A(B\^-D\^0\^-)\_[FSI]{} &=& A\_[3/2]{}e\^[i\_[3/2]{}]{}, where the subscript “FSI" indicates that the physical amplitudes take into account the effects of FSIs. This is motivated by comparing the experimental results with the calculated isospin amplitudes under the factorization approximation. Neglecting inelastic scattering, one can then extract the coefficients $a_{1,2}^{\rm eff}$ from a comparison of the measured and calculated isospin amplitudes [@ns]. It is straightforward to show that \[ampfsi\] A(B\^0D\^0\^0)\_[FSI]{} &=& A(B\^0D\^0\^0)+[2[T]{}- [C]{}+33]{}( e\^[i(\_[1/2]{}-\_[3/2]{})]{}-1),\
A(B\^0D\^+\^-)\_[FSI]{} &=& A(B\^0D\^+\^-)+[2[T]{}- [C]{}+33]{}( e\^[i(\_[1/2]{}-\_[3/2]{})]{}-1), where we have dropped the overall phase $e^{i\delta_{3/2}}$. The quark-diagram amplitudes $\T,~\C,~\E$ in Eq. (\[ampfsi\]) have the same expressions as before except that $a_{1,2}$ in Eqs. (\[CE\]) and (\[T\]) are replaced by the real parameters $a_{1,2}^{\rm eff}$. The latter do not contain FSI effects and are defined for $\delta_{1/2}=\delta_{3/2}=0$ [@Xing].[^3] In other words, the parameters $a_{1,2}^{\rm eff}$ are defined when FSIs are not imposed to the topological quark diagram amplitudes.
The isospin phase difference in Eq. (\[ampfsi\]) is $29^\circ$ for both $\ov B\to D\pi$ and $\ov B\to D^*\pi$. It is easily seen that $a_2^{\rm eff}/a_1^{\rm eff}$ is determined from the second line of Eqs. (\[a21Dpi\]) and (\[a21D\*pi\]) but without a phase for the ratio $A_{1/2}/(\sqrt{2}A_{3/2})$. For example, $a_2^{\rm eff}/a_1^{\rm eff}$ for $\ov B\to D\pi$ is given by .[a\_2\^[eff]{}a\_1\^[eff]{}]{}|\_[D]{} = [1-|[A\_[1/2]{}A\_[3/2]{}]{}|\_[D]{}+|[A\_[1/2]{}A\_[3/2]{}]{}|\_[D]{} ]{}[f\_f\_D]{}[m\_B\^2-m\_D\^2 m\_B\^2-m\_\^2]{}[F\_0\^[BD]{}(m\_\^2) F\_0\^[B]{}(m\_D\^2)]{}. The results are shown in Table V. Obviously $a^{\rm eff}_2/a^{\rm
eff}_1$ and $a^{\rm eff}_2$ are smaller than the previous solution.
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Model $a_1^{\rm eff}(D\pi)$ $a_2^{\rm eff}(D\pi)$ $a_2^{\rm eff}(D\pi) $a_1^{\rm eff}(\D\pi)$ $a_2^{\rm eff}(\D\pi)$ $a_2^{\rm eff}(\D\pi)/a_1^{\rm eff}(\D\pi)$
/a_1^{\rm eff}(D\pi)$
------- ----------------------- ----------------------- ------------------------ ------------------------ ------------------------ ---------------------------------------------
NRSX $0.88\pm0.06$ $0.23\pm0.08$ $0.26\pm0.09$ $0.97\pm0.04$ $0.16\pm0.04$ $0.17\pm0.04$
LF $0.87\pm0.06$ $0.25\pm0.09$ $0.29\pm0.10$ $0.83\pm0.03$ $0.18\pm0.05$ $0.22\pm0.05$
MS $0.91\pm0.06$ $0.27\pm0.10$ $0.30\pm0.10$ $0.87\pm0.03$ $0.20\pm0.05$ $0.23\pm0.06$
NS $0.96\pm0.06$ $0.32\pm0.12$ $0.34\pm0.11$ $0.94\pm0.04$ $0.22\pm0.06$ $0.24\pm0.06$
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
: Extraction of the parameters $a^{\rm eff}_1$ and $a^{\rm
eff}_2$ from the measured $B\to D^{(*)}\pi$ rates. Note that $a_2^{\rm eff}(D\pi)$ and $a_2^{\rm eff}(D^*\pi)$ should be multiplied by a factor of (200 MeV/$f_D$) and (230 MeV/$f_\D$), respectively.
Comparison
----------
We are ready to compare the above two different types of approaches. In the type-I solution, $D^{(*)0}\pi^0$ rates are accommodated because of an enhanced $|a_2(D^{(*)0}\pi)|$. The branching ratio of $D^{(*)0}\pi^-$ is not over-estimated owing to a relative strong phase between $a_1$ and $a_2$. In the type-II solution, although $a_2^{\rm eff}$ is smaller than the magnitude of $a_2$, the $D^{(*)0}\pi^0$ states gain a feedback from $D^{(*)+}\pi^-$ via FSIs. [^4] More precisely, elastic FSIs will enhance the decay rate of $D^0\pi^0$ by a factor of about 3 and suppress $D^+\pi^-$ slightly.
It has been realized that the isospin analysis proves to be useful only if a few channels are open as the case of two-body nonleptonic decays of kaons and hyperons. The isospin phases there (or decay amplitude phases) are related to strong-interaction eigenphases (for a recent discussion, see [@Suzuki]). For example, one can identify the isospin phase shift in $K\to \pi\pi$ with the measured $\pi\pi$ strong-interaction phase at the energy $\sqrt{s}=m_K^2$. However, when there are many channels open and some channels coupled, as in $D$ and especially $B$ decays, the decay phase is no longer the same as the eigenphase in the $S$-matrix. Indeed, the $S$-matrix in general contains a parameter describing inelasticity. Consider the decay $\ov B^0\to D^+\pi^-$ as an example. The state $D^+\pi^-$ couples to not only $D^0\pi^0$, but also $D^0\eta,\,D^0\eta'$, $D\pi\pi\pi$ channels, $\cdots$, etc. It has been argued that in the heavy quark limit the $B$ decay is dominated by multiparticle inelastic rescattering [@Donoghue]. As a consequence, even if elastic $D^{(*)}\pi$ scattering is measured at energies $\sqrt{s}=m_B$, the isospin phases appearing in (\[isospin\]) or (\[isospinfsi\]) cannot be identified with the measured strong phases. Moreover, the isospin amplitudes are not conserved by inelastic FSIs. Therefore, the isospin analysis presented before should be regarded as an intermediate step for describing physical decay amplitudes.
Nevertheless, the isospin decomposition of $\ov B\to D\pi$ amplitudes in Eq. (\[isospin\]) or (\[isospinfsi\]) is still valid. The isospin analysis is useful in some aspects. First, it provides an independent check on the relative phases among three decay amplitudes. Second, the deviation of $|A_{1/2}/(\sqrt{2}A_{3/2})|$ from unity measures the degree of departure from the heavy quark limit [@NP]. Third, the deviation of $a_2$ from $a_2^{\rm eff}$ characterizes the importance of (soft) FSI contributions to the color-suppressed quark diagram, recalling that $a_{1,2}^{\rm eff}$ are defined for the topologies without FSIs. This point will be elucidated more below.
As stressed in [@CC], the topological quark graphs are meant to have all strong interactions included. Hence, they are [*not*]{} Feynman graphs. For example, the genuine $W$-exchange topology in $\ov B\to D\pi$ decay consists of not only the short-distance $W$-exchange diagram but also the rescattering graph in which $\ov B^0\to D^+\pi^-$ is followed by the strong interaction process: $(D^+\pi^-)_{I=1/2}\to$ scalar resonances $\to D^0\pi^0$. Likewise, the process with inelastic rescattering from the leading $\T$ amplitude into $D^0\pi^0$ via quark exchange has the same topology as the color-suppressed tree diagram $\C$ [@Neubert]. Therefore, color-suppressed tree and $W$-exchange topologies receive short-distance and long-distance contributions.
From Tables III and V we see that $R_{D\pi}=|a_2(D\pi)/a_2^{\rm
eff}(D\pi)|\approx 1.75$ and $R_{D^*\pi}=|a_2(D^*\pi)/a_2^{\rm
eff}(D^*\pi)|\approx 1.95$. The corresponding quantities in $D\to\ov K\pi$ decays are $R_{\ov K\pi}\approx 2.0$ and $R_{\ov
K^*\pi}\approx 1.7$, respectively. Therefore, although the relative phase $59^\circ$ ($63^\circ$) between $B^0\to
D^{0(*)}\pi^0$ and $B^0\to D^{+(*)}\pi^-$ is significantly reduced from the phase $150^\circ$ between $D^0\to\ov K^{0(*)}\pi^0$ and $D^0\to K^{-(*)}\pi^+$ [@Rosner], the ratio $R$ does not decrease sizably from charm to bottom and, in contrast, it increases for the $VP$ case. It is thus anticipated that in both $D\to \ov K\pi$ and $\ov B\to D\pi$ decays, the soft FSI contributions to the color-suppressed topology $\C$ are dominated by inelastic rescattering [@Donoghue].[^5] Since $\eta$ and $\omega$ are isospin singlets, the conventional isospin analysis of FSIs is no longer applicable to the final states involving $\eta$ or $\omega$. The fact that the predicted $\ov B^0\to D^{(*)0}\eta$ rates based on the assumption $a_2(D^{(*)}\eta)=a_2(D^{(*)}\pi)$ are consistent with experiment (see Table IV) supports the notion that FSIs in $B$ decay are indeed highly inelastic.
Discussion and Conclusion
=========================
Beyond the phenomenological level, it is desirable to have a theoretical estimate of $a_2(D\pi)$. Unfortunately, contrary to the parameter $a_1(D\pi)$, $a_2(D\pi)$ is not calculable in the QCD factorization approach owing to the presence of infrared divergence caused by the gluon exchange between the emitted $D^0$ meson and the $(\ov B^0\pi^0)$ system. In other words, the nonfactorizable contribution to $a_2$ is dominated by nonperturbative effects. Nevertheless, a rough estimate of $a_2$ by treating the charmed meson as a light meson while keeping its highly asymmetric distribution amplitude yields $a_2(D\pi)\approx
0.25\exp(-i40^\circ)$ [@BBNS]. Evidently, large power corrections from long-distance FSI effects are needed to account for the discrepancy between theory and experiment for $a_2(D\pi)$. The rescattering contribution via quark exchange, $D^+\pi^-\to
D^0\pi^0$, to the topology $\C$ in $\ov B^0\to D^0\pi^0$ has been estimated in [@Blok] using $\rho$ trajectory Regge exchange. It was found that the additional contribution to $D^0\pi^0$ from rescattering is mainly imaginary: $a_2(D\pi)/a_2(D\pi)_{\rm
without~FSIs}=1+0.61{\rm exp}(73^\circ)$. This analysis suggests that the rescattering amplitude can bring a large phase to $a_2(D\pi)$ as expected.
In QCD factorization, $a_2(\pi\pi)$ or $a_2(K\pi)$ is found to be of order 0.20 with a small strong phase (see e.g. [@BBNS1]). The fact that the magnitude of $a_2(D\pi)$ is larger than the short-distance one, $a_2(K\pi)$ or $a_2^{\rm eff}(D\pi)$, should not be surprising because the former includes all possible FSIs, while the latter is defined without long-distance FSIs. In other words, $a_2(D\pi)$ include many possible long-distance effects. In the language of isospin analysis, we see from Eq. (\[ampfsi\]) that a\_2(D)=a\_2\^[eff]{}(D)-[2ha\_1\^[eff]{}(D)-a\_2\^[eff]{} (D) 3]{}(e\^[i(\_[1/2]{}-\_[3/2]{})]{}-1), where we have neglected $W$-exchange and h=[f\_f\_D]{}[m\_B\^2-m\_D\^2 m\_B\^2-m\_\^2]{}[F\_0\^[BD]{}(m\_\^2) F\_0\^[B]{}(m\_D\^2)]{}. It follows from Eq. (\[A13\]) and Table V that $a_2(D\pi)/a_2^{\rm eff}(D\pi)\approx 1.65\,{\rm exp}(56^\circ)$. It is worth remarking that $a_2(\J K)$ in $B\to \J K$ decay is calculable in QCD factorization; the theoretical result $|a_2(\J
K)|=0.19^{+0.14}_{-0.12}$ [@CYJpsiK] is consistent with the data $0.26\pm 0.02$ [@a1a2]. Hence it remains to understand why $|a_2(D\pi)|$ is larger than $|a_2(D^*\pi)|$ and $|a_2(\J K)|$ or why (soft) final-state interaction effects are more important in $D\pi$, $D^*\pi$ than in $\J K$ final states.
To conclude, the recent measurements of the color-suppressed modes $\ov B^0\to D^{(*)0}\pi^0$ imply non-vanishing relative FSI phases among various $\ov B\to D\pi$ decay amplitudes. Depending on whether or not FSIs are implemented in the topological quark-diagram amplitudes, two solutions for the parameters $a_1$ and $a_2$ are extracted from data using various form-factor models. It is found that $a_2$ is not universal: $|a_2(D\pi)|\sim
0.40-0.55$ and $|a_2(D^*\pi)|\sim 0.30-0.45$ with a relative phase of order $60^\circ$ between $a_1$ and $a_2$. If FSIs are not included in quark-diagram amplitudes from the outset, we have $a_2^{\rm eff}(D\pi)\sim 0.23-0.32,~a_2^{\rm eff} (D^*\pi)\sim
0.16-0.22\,.$ The large value of $|a_2(D\pi)|$ compared to $a_2^{\rm eff}(D\pi)$ or naive expectation implies the importance of long-distance FSI contributions to color-suppressed internal $W$-emission via final-state rescatterings of the color-allowed tree amplitude.
2 cm We would like to thank Hsiang-nan Li, Alexey A. Petrov, Zhi-zhong Xing and Kwei-Chou Yang for delighting discussions. We also wish to thank Physics Department, Brookhaven National Laboratory for its hospitality. This work was supported in part by the National Science Council of R.O.C. under Grant No. NSC90-2112-M-001-047.
[99]{}
[Belle]{} Belle Collaboration, K. Abe [*et al.,*]{} , 052002 (2002).
[CLEO]{} CLEO Collaboration, T.E. Coan [*et al.,*]{} , 062001 (2002).
[CLEOa]{} CLEO Collaboration, B. Nemati [*et al.,*]{} , 5363 (1998).
[PDG]{} Particle Data Group, D.E. Groom [*et al.,*]{} [*Eur. Phys. J*]{}, [**C15**]{}, 1 (2000).
A. Ali and C. Greub, , 2996 (1998).
H.Y. Cheng and B. Tseng, , 094005 (1998); Y.H. Chen, H.Y. Cheng, B. Tseng, and K.C. Yang, , 094014 (1999).
[a1a2]{} H.Y. Cheng and K.C. Yang, , 092004 (1999).
M. Neubert and B. Stech, in [*Heavy Flavours*]{}, 2nd edition, ed. by A.J. Buras and M. Lindner (World Scientific, Singapore, 1998), p.294 \[hep-ph/9705292\].
M. Neubert, V. Rieckert, B. Stech, and Q.P. Xu, in [*Heavy Flavours*]{}, 1st edition, edited by A.J. Buras and M. Lindner (World Scientific, Singapore, 1992), p.286.
H.Y. Cheng, C.Y. Cheung, and C.W. Hwang, , 1559 (1997).
[Melikhov]{} D. Melikhov and B. Stech, , 014006 (2001).
M. Wirbel, B. Stech, and M. Bauer, , 637 (1985); M. Bauer, B. Stech, and M. Wirbel, Z. Phys. C [**34**]{}, 103 (1987).
[Kroll]{} T. Feldmann, P. Kroll, and B. Stech, , 114006 (1998); , 339 (1999).
[Deandrea]{} A. Deandrea and A. D. Polosa, hep-ph/0107234.
[CC]{} L.L. Chau and H.Y. Cheng, , 137 (1987); L.L. Chau, H.Y. Cheng, W.K. Sze, B. Tseng, and H. Yao, , 2176 (1991).
[Xing]{} Z.Z. Xing, hep-ph/0107257.
[NP]{} M. Neubert and A.A. Petrov, , 50 (2001).
[Rosner]{} J.L. Rosner, , 114026 (1999).
J.D. Bjorken, Nucl. Phys. B (Proc. Suppl.) [**11**]{}, 325 (1989).
[BBNS]{} M. Beneke, G. Buchalla, M. Neubert, and C.T. Sachrajda, , 313 (2000).
[CHY]{} C.K. Chua, W.S. Hou, and K.C. Yang, hep-ph/0112148.
[Suzuki]{} M. Suzuki, , 111504 (1998).
[Donoghue]{} J.F. Donoghue, E. Golowich, A.A. Petrov, and J.M. Soares, , 2178 (1996).
[Neubert]{} M. Neubert, , 152 (1998).
[Zen]{} P. Żenczykowski, [*Acta Phys. Polon.*]{} [**B28**]{}, 1605 (1997) \[hep-ph/9601265\].
[CT]{} H.Y. Cheng and B.Tseng, , 014034 (1998); [*Chin. J. Phys.*]{} [**39**]{}, 28 (2001).
[Blok]{} B. Blok and I. Halperin, , 324 (1996).
[BBNS1]{} M. Beneke, G. Buchalla, M. Neubert, and C.T. Sachrajda, , 245 (2001).
[CYJpsiK]{} H.Y. Cheng and K.C. Yang, , 074011 (2001); J. Chay and C. Kim, hep-ph/0009244.
[^1]: The form factors $F_0^{B\eta}(m_D^2)=0.28$ and $F_1^{B\eta}(m_\D^2)=0.33$ for the NS model obtained in [@Deandrea] are larger than ours by about a factor of 2.
[^2]: From [@Rosner] one can deduce that $xa_2/a_1=\C/\T=(0.73\pm0.05)\,{\rm
exp}(i152^\circ)$ for $D\to PP$ decays without making any assumption on $W$-exchange, to be compared with the value $(1.05\pm0.05){\rm exp}(i149^\circ)$ obtained in [@NP] by neglecting $W$-exchange.
[^3]: The distinction of hard and soft FSI phases in principle cannot be done in a systematical way. For example, a sizable “hard" strong-interaction phase for $a_2$ in $B\to\pi\pi$ decay is calculable in the QCD factorization approach. However, $a_2$ is not computable for $\ov B\to D\pi$ and hence its strong phase is most likely soft.
[^4]: Recently, it has been suggested in [@CHY] that quasi-elastic scatterings of $D^{(*)}P\to D^{(*)}P$ and $DV\to DV$, for example, $DP=D^+\pi^-,
D^0\pi^0,D^0\eta_8,D_s^+K^-$, can explain the enhancement of not only $D^0\pi^0$ but also $D^0\eta$ via inelastic rescattering from the class-I mode $\ov B^0\to D^+\pi^-$.
[^5]: The quark diagram $W$-exchange in $D\to \ov PP$ decays and its phase relative to the topological amplitude $\T$ are dominated by nearby resonances in the charm mass region [@Zen], as shown explicitly in [@CT].
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The purpose of this study is the identification of young ($1< age < 100$ Myr), nearby ($d \leqslant100$ pc) moving groups (YNMGs) through their kinematic signature. YNMGs could be the result of the recent dispersal of young embedded clusters, such that they still represent kinematically cold groups, carrying the residual motion of their parental cloud. Using the fact that a large number ($\sim$ 14000) of the RAVE sources with evidence of chromospheric activity, also present signatures of stellar youth, we selected a sample of solar type sources with the highest probability of chromospheric activity to look for common kinematics. We made use of radial velocity information from RAVE and astrometric parameters from GAIA DR2 to construct a 6-dimension position-velocity vector catalog for our full sample. We developed a method based on the grouping of stars with similar orientation of their velocity vectors, which we call the Cone Method Sampling. Using this method, we detected 646 sources with high significance in the velocity space, with respect to the average orientation of artificial distributions made from a purely Gaussian velocity ellipsoid with null vertex deviation. We compared this sample of highly significant sources with a catalog of YNMGs reported in previous studies, which yield 75 confirmed members. From the remaining sample, about 50% of the sources have ages younger than 100 Myr, which indicate they are highly probable candidates to be new members of identified or even other YNMGs in the solar neighborhood.'
author:
- 'Valeria G. Ramírez-Preciado'
- 'Carlos G. Román-Zúñiga'
- Luis Aguilar
- Genaro Suárez
- Juan José Downes
title: Kinematic Identification of Young Nearby Moving Groups from a sample of Chromospherically Active Stars in the RAVE catalog
---
Introduction
============
Young star clusters are typically found in star forming regions within Giant Molecular Clouds (GMC) while still embedded in their parental gas [e.g. @Lada2003; @Allen2007; @Pelu2012]. Young stellar aggregations are usually bright in infrared wavelengths because many of their members display dusty circumstellar disks and the regions are normally associated to bright nebulosity features, all clear indicators of youth. Moreover, the dusty parental gas acts as a screen against the background population making clusters easier to spot on images. This is valid for relatively nearby associations where foreground population is minimum and contamination can be easily removed. But once a young cluster is no longer embedded, its detection gets complicated. Typically, after 10 Myr most of the parental gas is evacuated, along with the dispersal of most circumstellar disks and the dynamical relaxation of the system, which diminishes its density [@Lada2003]. Then, the components of most of the clusters mix with the Galactic disk population, which challenge the identification of the cluster members. At that evolutionary stage, emission line youth indicators such as the Li 6707 ${\mbox{\normalfont\AA}}$ line [e.g. @Sicilia2005] or X-ray emission [e.g. @Stelzel] are invoked for membership confirmation, but these diagnostics are difficult to implement in large samples. Moreover, if the cluster is close enough (10-$10^2$ pc), it cannot be distinguished as an overdensity in the sky because it covers very large areas.
A feature that can help to identify a dispersed cluster is the common kinematics still imprinted in their members; at ages of one to a few tens of Myr, young groups still stand out from the local velocity ellipsoid and have small velocity dispersions [@Antoja2012]. Known at this point as Young Nearby Moving Groups [YNMG; for a detailed review on the topic see @Torres2008], they can still be identified as they keep moving together away from their parental cloud before entering the Galactic Disk highway.\
The motivation of this work is to present a new method and procedure to identify emerging groups of young stars in the disk of the Galaxy through their kinematic signature. Finding such groups can be useful to understand the early evolution of unbound stellar groups, particularly how they disperse and integrate to the population of the galactic disk. Also, we look for a suitable technique for the identification of YNMG members, taking advantage of the increasing availability of data to provide the necessary parameters to construct full position-velocity vectors. Moreover, studying these kind of stars in the Solar Neighborhood can contribute to the understanding of star formation on the disk, as well as to its kinematic evolution.
Current identification methods are largely based on proper motions [e.g. @Aguilar1999; @Gagne2018] and positions [e.g. @kop2018], but this limited kinematical information makes group identification inconclusive, or unreliable. [@Riedel2017] developed a statistical method for the identification of YNMGs through their position and partial kinematical information, but as we mentioned before, position is not a reliable parameter for the identification of such disperse groups of stars. That is why it is necessary to implement other methods for its identification.
In this paper we present a simple and effective method to identify candidate members of YNMGs when their full 6D position-velocity vectors are known. The method can be applied to any sample that combines reliable proper motions, radial velocities and distances. Our test dataset is the RAdial Velocity Experiment (RAVE) catalog [@kunder2017], known to contain solar-type young star candidates.
This paper is organized as follows: we describe our sample selection in section 2, followed by a description of our methology in section 3. The results of applying our method to the selected dataset are presented in section 4. Finally, section 5 contains a discussion and summary of this work.
Sample \[s:sample\]
===================
The RAVE catalog contains stars with F, G and K spectral types, distributed in brightness within 8 and 12 magnitudes in the near-infrared $I$ band across the Southern hemisphere sky. The current (Data Release 5) RAVE catalog contains parameters for 520,630 individual stellar spectra, which are classified by means of a method called Local Linear Embedding [LLE, @Matijevic2012] which is based on a dimensionality reduction algorithm [@LLE]. The classification consists on applying the method directly to stellar spectra, reducing the number of dimensions (defined from a set of spectral features across previously defined spectral bins) that are needed to make the classification of a certain type of star. These dimensions are defined as spectral features common to an specific type of source. Each observed spectrum is compared against a grid of synthetic spectra in order to define a comparison sub-sample (5000 stars) from the previous data release. If the observed spectrum presents, with a certain level of confidence, features similar to those of other previously classified spectra in one or more dimensions, it is assigned a “flag” value [@Matijevic2010] in the corresponding dimension, indicating the nearest classification type. For each source listed, twenty flags representing the spectral feature dimensions are listed and each flag contains a letter representing the closest among eleven different stellar spectral classes of sources [@Matijevic2012]. The first three flags are those that have the highest weight in the classification. If these three first flags coincide, then the star has a high probability of belonging to that class.
For our study, we were interested in chromospherically active stars (CAS) in the RAVE catalog because it is estimated that about 40% of the CAS in RAVE coincide with high H$\alpha$ emitters from the ESO-GAIA catalog [@Zerjal2013; @zwitter2015]. The other $\sim$60% left could be young stars with not strong emission in that line or could be contaminants such as giants (see Section \[s:diagrams-HR\]). For the selection of our sample we considered stars for which the 3 first flags coincided with type *“chromospherically active”*. With this criterion we obtained a sample of 3128 stars with a high level of confidence of being CAS, from which many of them are expected to be young stars.
GAIA DR2 kinematic parameters
-----------------------------
The RAVE catalog provides radial velocities for all sources in our sample [RAVE provides good measurements for radial velocities, with uncertainties of 5 km/s or less]{}). In order to obtain the kinematic parameters for our main analysis, we complemented the RAVE CAS sample with astrometric parameters obtained from the GAIA Collaboration Data release 2 [hereafter GAIA DR2 @GAIADR2]. From GAIA DR2 we obtained RA and DEC positions, parallaxes and proper motions in RA and DEC, for all the sources in the sample. Using the `TOPCAT` tool version 4.6-1 [@topcat], we applied the method of @bj2018 to convert parallaxes into distance estimates using a local density exponential decrease prior with a scale parameter $h=500$ pc. With this information we were able to provide 6-dimensional position-velocity vectors for all the sources in the sample.
Data Analysis\[s:analysis\]
===========================
Velocity ellipsoids projections \[s:analysis:ss:kinematics\]
------------------------------------------------------------
The velocity vector of a star is fully determined by the radial velocity and the two components of its tangential velocity on the celestial sphere in addition to a parallax or distance to the source. In the 6-dimensional position-velocity space (hereafter PV6), each star is defined by six observational parameters: two coordinates (e.g. $l,b$ in the Galactic coordinate system), the corresponding two proper motion components $\mu_{l}$, $\mu_{b}$, a parallax $\varpi$ from which a distance can be estimated, and the radial velocity, usually calculated from Doppler shift measurements on a spectrum. From this, a PV6 vector ($X$, $Y$, $Z$, $U$, $V$, $W$) can be determined.
We use a heliocentric frame and following the canonical scheme [e.g. @schonrich2012], $X$ and $U$ are positive toward the galactic center, $Y$ and $V$ are positive along the direction of the galactic rotation, while $Z$ and $W$ are positive toward the galactic north pole. $U$, $V$ and $W$ are defined by equations \[u1\],\[u2\] and \[u3\] below (see Figure \[schon12\], based on Figure 1 of @schonrich2012 2012):
$$\begin{aligned}
\label{u1}
U = v_r (\cos{l} \cos{b}) - v_l (\sin{l}) - v_b (\sin{b} \cos{l})\\
\label{u2}
V = v_r (\sin{l} \cos{b}) + v_l (\cos{l}) - v_b (\sin{b} \sin{l}) \\
\label{u3}
W = v_r (\sin{b}) + v_b (\cos{b})\end{aligned}$$
where $v_r$ is the radial velocity, $v_l$ and $v_b$ are the tangential components of the velocity.\
The U, V, W velocity components give us the velocity distribution for a given population, usually described to first approximation as a *velocity ellipsoid*, and scatter plots of any two such components give us projections of this ellipsoid. These projections can be used to look for overdensities that may represent YNMG of stars [e.g. @Antoja2009; @Antoja2012].
Lack of structure in the velocity ellipsoid projections
-------------------------------------------------------
We constructed the velocity ellipsoid projections (see Figure \[UV-maps\]) for our sample of RAVE CAS using 2-dimensional histograms. We look for substructure on top of the smooth ellipsoidal distribution, as signal of moving groups. Our criteria for these plots were *a)* a resolution of 3 km/s per bin, and *b)* to consider only those stars with velocity moduli smaller than 600 $\mathrm{km/s}$. The latter was chosen from a histogram for $(U,V,W)$ values, indicating that more than 80% of our original sample was smaller than 200 $\mathrm{km/s}$.
![Graphical description of the $(U,V,W)$ coordinates frame according to the galactic coordinates as expressed in equations \[u1\], \[u2\] y \[u3\].[]{data-label="schon12"}](f1.pdf){width="50.00000%"}
All three projections show an ellipsoidal-like distribution for our sample. The left panel of Figure \[UV-maps\] shows a clear vertex deviation with an angle of approximately 15 $^\circ$, and an indication of an overdensity near $(U,V)=(-15,-20)$ plus a few small lumps near the center. The $VW$ and $UW$ projections are mostly featureless. However, we conclude that the signal to noise of these histogram images may not be high enough to resolve substructure clearly with our relative small sample.\
Increasing the resolution of these 2D histograms did not improve our results either, because the noise level also increased, making it actually more difficult to distinguish any overdensity features.
{width="0.46\linewidth"} {width="0.46\linewidth"} {width="0.46\linewidth"} {width="0.072\linewidth"}
Figure \[wave\] shows a further effort to highlight and identify structures within the $UV$ projection. We maintaned the resolution to a projection bin of 3 km/s and we applied a wavelet filtering method to the resultant image in order to highlight overdensity regions while removing extended low density structure. For this purpose we used an algorithm (B. Vandame, personal communication) based on the Multi-scale Vision Model (MVM) by @Rue1997. This filtering process allowed to highlight some possibly significant over-densities in the UV projection, as shown in Figure \[wave\]. Our wavelet image shows four main lumps with sizes of about 10 km/s. The sizes and the separations between groups in our wavelet filtered map are consistent with the velocity ellipsoid projections for moving groups in the solar vicinity by @Antoja2012. The largest lump near (U,V)=(-10,-20) is actually close with the Hyades, and the central lump is close to Coma Berenice, as reported in that work, but the coincidences are not exact, and the other two small lumps in our diagram are not directly related to any of their groups. On one hand, the differences could be explained by our use of GAIA DR2 parameters, which may be refining some values and highlighting distinct features. But on the other hand, our CAS sample is very different from theirs and we cannot venture to claim a real coincidence. Moreover, on Antoja work, they also used a higher resolution in their maps, but we cannot reach that resolution due to the size of our sample, which complicates the use of the velocity projections to identify additional structure. As we expect more than four YNMGs in our sample, we implemented an additional method to try to identify them.
![Density map of the UV projection of the velocity ellipsoid for RAVE CAS after applying the wavelet filtering process based on MVM. Although the substructure of left panel in Figure 2 is more evident here, the resolution of this image (3 km/s) is still not sufficient to carry out a satisfactory separation of individual moving groups.The colorbar indicates source density.[]{data-label="wave"}](f3.pdf){width="54.00000%"}
The Cone Method \[s:analysis:ss:cone\]
--------------------------------------
As shown in the velocity ellipsoid projection figures (Figures \[UV-maps\] and \[wave\]), some substructure becomes apparent when using some contrast enhancing techniques, like the wavelet filter. The basic problem we have here is that we are dealing with projections, which diminishes the contrast of 3D substructure. As we are searching for stars that belong to a YNMG,they should constitute a kinematically “cold” group. This means that their velocity vectors, when seen from a reference frame away from its own barycenter, should all point roughly in the same direction. This is the basis of the method we introduce here. While this is still based on a projection, it is entirely different in its construction.
![The panel shows the configuration of the cone on each star. It shows how the direction of the velocity vector fall within the cone. Thus, all stars that have the same configuration of $\theta$ and $\phi$ are grouped.[]{data-label="cone-method"}](f4.pdf){width="0.9\linewidth"}
First, a cone with vertex at the velocity frame origin is defined (see Figure \[cone-method\]). The angles $\theta$ and $\phi$ define its orientation in this space ($0 < \theta \leqslant 360^\circ$ along the $U$-$V$ plane, measured from the first toward the second axis; $-90^\circ \leqslant \phi \leqslant 90^\circ$ perpendicular to the first axis and null for $W = 0$) and the angle $\alpha$ denotes its aperture. The unit vector $\hat{L}$ indicates the cone symmetry axis:
$$\label{cono2}
\hat{L}=[\cos{(\theta)} \cos{(\phi)}, \sin{(\theta)} \cos{(\phi)},\sin{(\phi)}].$$
Then, we identify all stars in the sample whose velocity vector $\vec{v}$ falls within this cone, i.e. the following condition is satisfied: $$\label{constriction}
\frac{\hat{L} \cdot \vec{v}}{|\vec{v}|} < \cos{(\alpha)}.$$
Notice that this condition is set in velocity space, not in configuration space, which means that the star position in configuration space is irrelevant.\
All stars that fall within the cone are assigned to this particular combination of $(\theta, \phi)$ values. We then establish a $(\theta, \phi)$ grid on the unit sphere and compute the number of stars assigned to each such gridpoint, calling it the Cone Method Sampling (CMS). The grid is designed so that each cell subtends equal solid angles. We can then generate star count maps on the unit sphere where local peaks indicate groupings of stars that share the same velocity vector orientations (within an angle $\alpha$). Reducing the cone opening makes the criterion more stringent, allowing to identify colder clumps, but reduces the number of stars within a group. It is also obvious that $\alpha$ should not be reduced below the uncertainty in velocity orientations resulting from the errors in the observables.\
After we identified substructure using this procedure, it is necessary to assign to each peak a statistical significance, i.e. a measure of its probability that it is not a mere chance fluctuation. For this, we need a null-hypothesis (NH) that all substructure is merely due to fluctuations due to counting of discrete events in a mesh. In our case, we use as NH a trivariate Gaussian distribution whose centroid and extent are given by the individual means and standard deviation of the individual $U$, $V$ and $W$ distributions of the sample. No correlation between the individual components is assumed, so no vertex deviation is considered. We then performed Monte Carlo sampling of the NH to construct 5,000 synthetic samples of equal size to the real one. At the end we built $(\theta, \phi)$ maps of the expected median density of sources at each gridpoint under the hypothesis that no substructure really exists. We used these values, gridpoint by gridpoint, to establish the statistical significance of the peaks in the real sample, as we describe below.
{width="65.00000%"}
Structure separation
--------------------
We applied the CMS, as previously described (see Section \[s:analysis:ss:cone\]), to our sample of RAVE CAS, in order to obtain the distribution of stars that satisfied the dot product condition in inequality \[constriction\]. For our purposes, we chose the value $\alpha=3^\circ$, and our grid was constructed as follows: first, we made a $\cos(b)$ correction on latitude, to assure uniform coverage; then we used Nyquist sampling in order to reduce the step to half the resolution of the grid so no stars are left out of the counting. A map of the distribution of our CAS sample in the $(\theta, \phi)$ grid is shown in Figure \[ncount\].
{width="0.45\linewidth"} {width="0.5\linewidth"} {width="0.5\linewidth"}
In order to obtain a more reliable identification of possible YNMG from the CMS, we developed and implemented a method to identify structures within the map described above, based on the statistical significance of the counts at each position.\
Using the Monte Carlo realizations, we used the median of the count density at each position in the mesh for the NH case as a central estimator. We used the median because it is a more robust indicator for the distributions in each mesh point, which are in most cases positively skewed. Then, we obtained a deviation by calculating the absolute difference between the median and the value corresponding to the 90 percentile, also at each position of the mesh. The corresponding median and deviation maps are shown in the top-right and top-left panels of Figure \[med-dev\].
We define a **reliability** range at each position in the mesh as: $$S = \frac{observed \ data \ - \ median \ (NH)}{deviation \ (NH)}
\label{signif}$$
This way, we define high significance on all points as $S > 1.5$ and low significance as $1<S<1.5$.\
A map that shows the final distribution of CMS counts using the significance estimator, is shown in the bottom panel of Figure \[med-dev\]. We found a total of 646 stars located in mesh positions with high significance. These sources represent a new sample of solar type CAS candidates to be members of recently disaggregated young star clusters.
Uncertainty cone
----------------
It was necessary to corroborate that our choice for the cone opening value, $\alpha$, was adequate for the detection of groups with similar velocity vector orientation in the RAVE CAS sample. This is, we need to make sure that $\alpha$ is consistently wide enough compared to the uncertainty in the angle between $\hat{L}$ and $\vec{v}$. For this purpose, we constructed an error cone from the uncertainties in the $U$, $V$ and $W$ components. Using the following transformation expressions from the velocity to the cone space:
$$\label{conoerror11}
\theta = \arctan{(\frac{W}{\sqrt{U^2 + V^2 }})},$$
$$\label{conoerror21}
\phi = \arctan{(\frac{V}{U})},$$
we determined the error propagation to first order, which provides $\delta \theta$ and $\delta \phi$. This way we can determine $\delta \alpha$ as the equivalent radius of the area $\delta \theta \times \delta \phi$, which corresponds to the opening of the cone formed with the uncertainties. A detailed derivation of $\delta \alpha$ is included on Appendix \[App1\]. If $\delta\alpha$ is consistently smaller than the characteristic aperture $\alpha$ of the map (in our case $3^\circ$.), we can trust that the aperture used is adequate for finding common kinematics between stars in our sample. We confirmed that more than 90% of the stars in our sample have an opening of the uncertainly cone smaller than $\alpha$, indicating that the $\alpha$ value we use is reliable.
Results \[s:results\]
=====================
Moving Groups \[s:mg\]
----------------------
After analyzing the significant regions with the method described above, we need to know if our selected stars have been already identified as members of YNMGs. For this purpose, we cross matched our list with a compilation of known members of YNMG including results from: @Torres2006 [@Torres2008] ($\beta$ Pictoris, Columba, Tucana-Horologium), @Dawson2012 [@Dawson2013] ($\eta$ Chameleontis), @Ducourant2014 (TW Hydrae), @Elliott2014 (AB Doradus, Argus, $\beta$ Pictoris, Carina, Columba, $\eta$ Chameleontis, Octans, TW Hydrae), @Galvez2010 [@Galvez2014] (Castor), @Malo2013 [@Malo2014] ($\beta$ Pictoris, TW Hydrae, Tucana-Horologium, Columba, Carina, Argus, AB Doradus), @Riedel2014 ($\eta$ Chameleontis, TW Hydrae, $\beta$ Pictoris, Octans, Tucana Horologium, Columba, Carina, Argus, AB Doradus), @Gagne2015 [@Gagne2018] (Argus, Columba, $\beta$ Pictoris, AB Doradus, Carina, TW Hydrae, Tucana Horologium), @Desilva2009 [@Desilva2013] ($\eta$ Chameleontis, TW Hydrae, $\beta$ Pictoris, Octans, Tucana-Horologium, Columba, Carina, Argus, AB Doradus), @Cruz2009 (AB Doradus, $\beta$ Pictoris, TW Hydrae), @Kraus2014 (Tucana-Horologium), @Makarov2000 (Carina), @Moor2013 (Columba, Carina, Argus, AB Doradus, $\beta$ Pictoris), @Murphy2015 (Octans), @Shkolnik2012 (AB Doradus, $\beta$ Pictoris, Carina, Castor, $\eta$ Chameleontis, Columba, TW Hydrae, Tucana-Horologium) and @zuckerman2004 (AB Doradus, $\eta$ Chameleontis). The catalog contains information for over 2300 members of associations at distances smaller than $\sim200$ $\mathrm{pc}$, with revised positions and information about which YNMG they each belong. We found a total of 75 matches with 10 known YNMGs, which are listed in Table 1 of Appendix \[App3\]. In Figure \[moving-groups\] we show a map, in Galactic coordinates, of these stars. The remaining 571 sources have no matches with recent literature on YNMG.
To corroborate the reliability of our technique, we made the same analysis for a sample of 275 stars from BANYAN IV [@Malo2014]. We found that with our method we recovered 173 stars of the initial sample which have $S>1.5$ and 270 sources with $S>1$. Determination of membership for each candidate using typical methods [e.g. spectroscopy; see for instance @Binks2015] is beyond the scope of this study. In the ideal case, we should produce a table with individual membership probabilities based on known properties of YMNG groups, like average radial velocity, average distance, etc. However, this kind of information is difficult to recollect in a consistent way for most YNMGs. We only were able to make an acceptable comparison with the Beta Pictoris group [distance=18-40 pc, $\mathrm{<v_r>=60}$km/s, extension = 40 pc. @Malo2013; @Moor2013], were 80 stars in our remaining candidate sample coincide within the uncertainties with the characteristics of this group. For this reason, we chose instead to analyze the location of our CMS significant CAS in the HR diagram, focusing on the ages of the candidates, in order to highlight possibly young sources.\
{width="0.8\linewidth"} {width="0.12\linewidth"}
{width="0.7\linewidth"}
HR diagrams \[s:diagrams-HR\]
-----------------------------
The remaining 571 stars identified with our method, may still have kinematics in common even though they are not associated to any known YNMG. For instance, a number of these stars could be unidentified members or new groups. However, the list may also contain a fraction of contaminant sources. This mainly comes from the fact that chromospheric activity is not exclusive of young stars; some types of evolved sources in the sub-giant and giant branch [@Oz2018], and spectroscopic binaries [@Fekel2002] may present chromospheric emission and be selected as RAVE CAS. This occurs because chromospherical activity can be present before and after the main sequence phase [@Frasca2015]. As mentioned in Section 1, our main goal is to select members of recently dispersed young clusters, so we need to know which sources are consistent with young ages (10-10$^2$ Myr).
The original goal of this work is the identification of YNMG member candidates through their kinematic signature, but estimating the individual ages of our candidates to isolate probably young stars, helps us to depurate our sample and to reinforce the results.
Combining the $T_{eff}$ and $A_{v}$ from the RAVE DR5 catalog with the distances from [@Luri2018] and using the optical $V$ magnitudes from APASS DR9 [@Henden2016], we constructed the HR diagram for the CMS candidate sample. This diagram allowed us to estimate the ages and masses of the candidates working with a method similar to that used by [@Suarez2017], which basically interpolates the $\mathrm{T_{eff}}$ and $\mathrm{L_{bol}}$ into the stellar models from [@Bressan2012] and [@Marigo2017] to estimate masses and ages, as well as their uncertainties. For our purposes, in order to remove most of the contamination by evolved sources, we limited our candidate sample to those stars with $\mathrm{M_V}\geqslant 4.5$ mag. After this cutoff, we selected those stars younger than 100 Myr. The resulting clean sample contains 290 candidates, which represents the $\sim$50% of the sample of YNMG member candidates. From the remain 280 removed stars, 70 stars are $H\alpha$-emitters. This percentage is consistent with the fraction of CAS in RAVE with high H$\alpha$ emission (see Section \[s:sample\]). This show that the use of HR diagrams allows to depurate the YNMG member candidate selection.
In Figure \[HR\] we show the HR diagram for the members and candidates of YNMGs. We can see that all of member and candidates lie between the 1 and 100 Myr isochrones from [@Marigo2017] and between the 0.5 and 1.6 $M_{\odot}$ evolutionary tracks from [@Bressan2012].
Discussion and Summary \[s:discussion\]
=======================================
From our analysis, we found that the kinematic identification of YNMGs in the RAVE CAS sample, directly from the velocity ellipsoid projections, was not satisfactory. This is likely because the sample is not robust enough or, perhaps because the number of CAS sources that we can identify as known members of individual YNMGs is small. It may be possible that for a larger sample of stars, that included sources located in a larger volume, such method could be implemented successfully and its velocity ellipsoid could distinguish between groups with different kinematic signature.
Precisely this problem was the origin of our idea of the alternative method, the CMS we present in this paper. Our method uses directly the orientation of the velocity vectors in the velocity space to identify groups of stars with a common kinematic signature. This made our method distinc from others typically used.
While the idea behind the CMS is relatively simple, we showed that it can be a reliable tool, useful for the identification of kinematically cold groups. Moreover, the CMS appears to work well with small groups in relatively small samples like the RAVE CAS. In this sense, our method is not exclusive for groups of young stars. Our CMS, in principle, can be applied to any catalog of sources that have reliable observational parameters to construct PV6 vectors. This makes our method very well suited for other applications. We applied our method to samples of known groups and detected those overdensities on the $(\theta, \phi)$ map.
With our method, we successfully identified 75 members of YNMGs previously reported in the literature (see appendix \[App3\]). For the remaining sources in the sample, our analysis of the HR diagram, indicates that a significant number of sources are consistent with ages between 1 and 100 Mys, indicating that the RAVE CAS sample possibly contains dozens of young stars that belong to known or even new YNMG in the solar neighborhood. Follow-up work should focus on determination of membership for the stars in Table \[tab:results\], based on youth signatures from spectroscopy.
ACKNOWLEDGMENTS.
VRP and GS acknowledge support from a graduate studies fellowship from CONACYT/UNAM Mexico. VRP, CRZ and GS acknowledges support from program UNAM-DGAPA-PAPIIT IN108117, Mexico. LA acknowledges support from UNAM-DGAPA-PAPIITT IG100125 project.
This paper makes use of RAVE data. Funding for RAVE has been provided by: the Australian Astronomical Observatory; the Leibniz-Institut fuer Astrophysik Potsdam (AIP); the Australian National University; the Australian Research Council; the French National Research Agency; the German Research Foundation (SPP 1177 and SFB 881); the European Research Council (ERC-StG 240271 Galactica); the Istituto Nazionale di Astrofisica at Padova; The Johns Hopkins University; the National Science Foundation of the USA (AST-0908326); the W. M. Keck foundation; the Macquarie University; the Netherlands Research School for Astronomy; the Natural Sciences and Engineering Research Council of Canada; the Slovenian Research Agency; the Swiss National Science Foundation; the Science & Technology Facilities Council of the UK; Opticon; Strasbourg Observatory; and the Universities of Groningen, Heidelberg and Sydney. The RAVE web site is at https://www.rave-survey.org.
This publication makes use of data products from the Two Micron All Sky Survey, which is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology, funded by the National Aeronautics and Space Administration and the National Science Foundation.
This work has made use of data from the European Space Agency (ESA) mission [*Gaia*]{} (<https://www.cosmos.esa.int/gaia>), processed by the [*Gaia*]{} Data Processing and Analysis Consortium (DPAC, <https://www.cosmos.esa.int/web/gaia/dpac/consortium>). Funding for the DPAC has been provided by national institutions, in particular the institutions participating in the [*Gaia*]{} Multilateral Agreement.
Some figures in this paper, as well as distance estimates were obtained with the fabulous TOPCAT software, which is described in [@topcat].
Allen, L., Megeath, S. T., Gutermuth, R., et al. 2007, Protostars and Planets V, 361
, T., [Valenzuela]{}, O., [Pichardo]{}, B., [et al.]{} 2009, , 700, L78
, T., [Helmi]{}, A., [Figueras]{}, F., & [Romero-G[ó]{}mez]{}, M. 2012, in European Physical Journal Web of Conferences, Vol. 19, European Physical Journal Web of Conferences, 05002
, T. L., & [Bailer-Jones]{}, C. A. L. 2016, , 832, 137
, C. A. L. 2015, , 127, 994
Bailer-Jones, C. A. L., Rybizki, J., Fouesneau, M., Mantelet, G., & Andrae, R. 2018, arXiv:1804.10121
Binks, A. S., Jeffries, R. D., & Maxted, P. F. L. 2015, , 452, 173
Bressan, A., Marigo, P., Girardi, L., et al. 2012, , 427, 127
Cruz, K., Looper, D., Prato, L., & Kirkpatrick, J. D. 2009, NOAO Proposal,
Dawson, P., Scholz, A., Ray, T. P., et al. 2013, , 429, 903
da Silva, L., Torres, C. A. O., de La Reza, R., et al. 2009, , 508, 833
De Silva, G. M., D’Orazi, V., Melo, C., et al. 2013, , 431, 1005
Dawson, P., Scholz, A., Ray, T. P., et al. 2012, arXiv:1211.4484
Ducourant, C., Teixeira, R., Galli, P. A. B., et al. 2014, , 563, A121
Elliott, P., Bayo, A., Melo, C. H. F., et al. 2014, VizieR Online Data Catalog, 356,
Fekel, F. C., Henry, G. W., Eaton, J. A., Sperauskas, J., & Hall, D. S. 2002, , 124, 1064
Frasca, A., Biazzo, K., Lanzafame, A. C., et al. 2015, , 575, A4
Gagn[é]{}, J., Faherty, J. K., Cruz, K. L., et al. 2015, , 219, 33
Gagn[é]{}, J., Mamajek, E. E., Malo, L., et al. 2018, , 856, 23
Gaia Collaboration, Brown, A. G. A., Vallenari, A., et al. 2016, , 595, A2
Gaia Collaboration, Brown, A. G. A., Vallenari, A., et al. 2018, arXiv:1804.09365
G[á]{}lvez-Ortiz, M. C., Clarke, J. R. A., Pinfield, D. J., et al. 2010, , 409, 552
G[á]{}lvez-Ortiz, M. C., Kuznetsov, M., Clarke, J. R. A., et al. 2014, , 439, 3890
Henden, A. A., Templeton, M., Terrell, D., et al. 2016, VizieR Online Data Catalog, 2336,
Hoogerwerf, R., & Aguilar, L. A. 1999, , 306, 394
Kraus, A. L., Shkolnik, E. L., Allers, K. N., & Liu, M. C. 2014, , 147, 146
Koppelman, H. H., Virginiflosia, T., Posti, L., Veljanoski, J., & Helmi, A. 2018, arXiv:1804.07530
, A., [Kordopatis]{}, G., [Steinmetz]{}, M., [et al.]{} 2017, , 153, 75
, C. J., & [Lada]{}, E. A. 2003, , 41, 57
Luri, X., Brown, A. G. A., Sarro, L. M., et al. 2018, arXiv:1804.09376
Malo, L., Doyon, R., Lafreni[è]{}re, D., et al. 2013, , 762, 88
Malo, L., Doyon, R., Feiden, G. A., et al. 2014, , 792, 37
Makarov, V. V., & Urban, S. 2000, , 317, 289
Marigo, P., Girardi, L., Bressan, A., et al. 2017, , 835, 77
, G., [Zwitter]{}, T., [Munari]{}, U., [et al.]{} 2010, , 140, 184
, G., [Zwitter]{}, T., [Bienaym[é]{}]{}, O., [et al.]{} 2012, , 200, 14
Mo[ó]{}r, A., Szab[ó]{}, G. M., Kiss, L. L., et al. 2013, , 435, 1376
Murphy, S. J., & Lawson, W. A. 2015, VizieR Online Data Catalog, 744,
zdarcan, O., & Dal, H. A. 2018, arXiv:1801.06087
Pelupessy, F. I., & Portegies Zwart, S. 2012, , 420, 1503
Riedel, A. R., Finch, C. T., Henry, T. J., et al. 2014, , 147, 85
Riedel, A. R., Blunt, S. C., Lambrides, E. L., et al. 2017, , 153, 95
Roweis, S. T., & Saul, L. K. 2000, science, 290, 2323
, F., & [Bijaoui]{}, A. 1997, Experimental Astronomy, 7, 129
Sch[ö]{}nrich, R. 2012, , 427, 274
Sicilia-Aguilar, A., Hartmann, L. W., Hern[á]{}ndez, J., Brice[ñ]{}o, C., & Calvet, N. 2005, , 130, 188
Shkolnik, E. L., Anglada-Escud[é]{}, G., Liu, M. C., et al. 2012, , 758, 56
Stelzer, B., & Neuh[ä]{}user, R. 2000, , 361, 581
Su[á]{}rez, G., Downes, J. J., Rom[á]{}n-Z[ú]{}[ñ]{}iga, C., et al. 2017, , 154, 14
Taylor, M. B. 2005, Astronomical Data Analysis Software and Systems XIV, 347, 29
Traven, G., Zwitter, T., van Eck, S., et al. 2015, VizieR Online Data Catalog, 358,
Torres, C. A. O., Quast, G. R., da Silva, L., et al. 2006, , 460, 695
Torres, C. A. O., Quast, G. R., Melo, C. H. F., & Sterzik, M. F. 2008, Handbook of Star Forming Regions, Volume II, 5, 757
, M., [Zwitter]{}, T., [Matijevi[č]{}]{}, G., [et al.]{} 2013, , 776, 127
Zuckerman, B., & Song, I. 2004, , 42, 685
Zwitter, T., Kos, J., [Ž]{}erjal, M., & Traven, G. 2016, ASPC, 507, 201
Uncertainty cone {#App1}
================
The following error propagation is a linear approximation to obtain the suitable $\alpha$ aperture for the CMS. Starting from the definitions of $\theta$ and $\phi$ from the transformation of the velocity space $$\theta = \arctan{(\frac{W}{\sqrt{U^2 + V^2 }})},$$
$$\phi = \arctan{(\frac{V}{U})},$$
we calculated the error propagation at first order: $$\delta \phi = \delta U (\frac{UW}{(\sqrt{U^2 + V^2}) (U^2 + V^2 + W^2)}) + \delta V (\frac{VW}{(\sqrt{U^2 + V^2}) (U^2 + V^2 + W^2)}) + \delta W (\frac{\sqrt{U^2 + V^2}}{U^2 + V^2 + W^2})$$ and $$\delta \theta = \delta U (\frac{U}{\sqrt{U^2 + V^2}}) + \delta V (\frac{U}{\sqrt{U^2 + V^2}})$$ Considering that $\delta \alpha$ is the equivalent radius of the segment formed by $\delta \theta \times \delta \phi$, then $\delta \alpha$ is calculated: $$\delta \alpha = \sqrt{\frac{|\delta \theta| \times |\delta \phi|}{\pi}}$$\
YNMG detected members in the CAS RAVE Sample {#App3}
============================================
[lccccccccccccc]{} 23093711-0225551 & 23:09:37.10 & -02:25:55.0 & 10.933 & 8.567 & 60.842 & -45.963 & -11.21 & 4000.0 & 0.215 & 0.734 & 12.27 & CAR\
23215251-6942118 & 23:21:52.50 & -69:42:12.0 & 10.008 & 8.657 & 41.368 & -32.023 & 5.6 & 5248.0 & 0.816 & 1.045 & 19.08 & Unknown\
05023042-3959129 & 05:02:30.40 & -39:59:13.0 & 10.654 & 8.727 & 35.104 & -23.752 & 26.329 & 4561.0 & 0.218 & 0.752 & 39.81 & ABDMG\
22463348-3928451 & 22:46:33.50 & -39:28:45.0 & 9.528 & 8.21 & 75.105 & -3.24 & -1.64 & 5175.0 & 0.839 & 1.08 & 16.49 & Unknown\
05332558-5117131 & 05:33:25.60 & -51:17:13.0 & 11.805 & 8.986 & 42.755 & 26.101 & 18.825 & 4000.0 & 0.153 & 0.706 & 35.37 & THA\
03241504-5901125 & 03:24:15.00 & -59:01:13.0 & 12.108 & 9.547 & 43.729 & 7.913 & 19.376 & 4000.0 & 0.239 & 0.734 & 10.04 & COL\
05451623-3836491 & 05:45:16.30 & -38:36:49.0 & 11.016 & 9.561 & 10.622 & 7.65 & 30.688 & 5230.0 & 1.256 & 1.245 & 10.85 & COL\
08430040-5354076 & 08:43:00.40 & -53:54:08.0 & 11.099 & 9.755 & -23.293 & 22.849 & 12.148 & 5249.0 & 0.91 & 1.092 & 16.57 & ARG\
13544209-4820578 & 13:54:42.10 & -48:20:58.0 & 11.024 & 9.285 & -31.868 & -23.39 & 0.338 & 5000.0 & 0.683 & 1.047 & 16.27 & Unknown\
11594226-7601260 & 11:59:42.30 & -76:01:26.0 & 11.139 & 9.14 & -41.025 & -6.19 & 14.435 & 3996.0 & 0.465 & 0.735 & 3.491 & ECh\
\[tab:members\]
YNMG Candidates in the CAS RAVE Sample {#App2}
======================================
[lcccccccccccc]{} 23123243-0240516 & 23:12:32.46 & -02:40:51.9 & 12.915 & 10.8 & 17.44 & -4.568 & 11.995 & 4000.0 & 0.21 & 0.732 & 13.05\
21323568-5558015 & 21:32:35.70 & -55:58:01.6 & 12.473 & 10.987 & -18.23 & -12.874 & 69.72 & 5750.0 & 1.924 & 1.238 & 14.73\
23581157-3850073 & 23:58:11.58 & -38:50:07.5 & 11.689 & 10.104 & -28.779 & -30.763 & 4.097 & 4964.0 & 0.432 & 0.879 & 29.25\
00220533-4050257 & 00:22:05.34 & -40:50:25.7 & 13.19 & 11.444 & -5.68 & -19.514 & 4.565 & 5000.0 & 0.727 & 1.072 & 15.12\
02523097-5447531 & 02:52:30.99 & -54:47:53.3 & 13.295 & 11.19 & 0.693 & -30.811 & -0.92 & 4500.0 & 0.191 & 0.717 & 52.92\
00263489-6545359 & 00:26:34.90 & -65:45:36.0 & 11.289 & 9.611 & 27.489 & 4.21 & -14.657 & 4755.0 & 0.321 & 0.829 & 32.38\
05213171-3641084 & 05:21:31.73 & -36:41:08.5 & 12.919 & 10.749 & -1.924 & 17.613 & -11.656 & 4000.0 & 0.341 & 0.731 & 5.331\
10573417+0048243 & 10:57:34.16 & +00:48:24.2 & 13.533 & 11.348 & -40.667 & 16.654 & -28.338 & 4465.0 & 0.178 & 0.706 & 60.11\
12530218-1549546 & 12:53:02.15 & -15:49:54.6 & 14.267 & 11.37 & -3.745 & 17.96 & -5.038 & 3814.0 & 0.128 & 0.704 & 29.81\
21480570-0127397 & 21:48:05.71 & -01:27:39.9 & 11.642 & 10.017 & 14.8 & 12.746 & -1.185 & 5003.0 & 0.336 & 0.8 & 43.78\
\[tab:results\]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Some interesting properties of almost Cohen-Macaulay rings are investigated and a Serre type property connected with this class of rings is studied.'
author:
- |
Cristodor Ionescu\
Institute of Mathematics *Simion Stoilow*\
of the Romanian Academy\
P.O. Box 1-764\
RO 014700 Bucharest\
Romania\
email: [email protected]
title: 'More properties of almost Cohen-Macaulay rings'
---
Introduction
============
A flaw in the chapter dedicated to Cohen-Macaulay rings in the first edition of [@Mat] was corrected in the second edition. This led to the study of the so-called almost Cohen Macaulay rings, first by Y. Han [@Han] and later by M.-C. Kang [@K1], [@K2]. Since the first of these papers is written in Chinese, the others two are the main reference for the subject.
\[fordef\] Let $A$ be a commutative noetherian ring, $P\in{{\rm Spec}}(A)$ and $M\neq 0$ a finitely generated $A$-module. Then ${{\rm depth}}_P(M)\leq{{\rm depth}}_{PA_P}M_P.$
\[maindef\] [(cf. [@Han], [@K1])]{} Let $A$ be a commutative noetherian ring. A finitely generated $A$-module $M\neq 0$ is called almost Cohen-Macaulay if ${{\rm depth}}_PM={{\rm depth}}_{PA_P}M_P$, for any $P\in{{\rm Supp}}(M).$ $A$ is called an almost Cohen-Macaulay ring if it is an almost Cohen-Macaulay $A$-module, that is if for any $P\in{{\rm Spec}}(A),$ ${{\rm depth}}_PA={{\rm depth}}_{PA_P}A_P.$
Several properties of almost Cohen-Macaulay rings are proved in [@K1] and several interesting examples are given in [@K2]. In the following we are trying to complete the results in [@K1] and to introduce a Serre-type condition, that we call $(C_k),$ for any $ k\in\mathbb{N},$ condition that is to be to almost Cohen-Macaulay rings what the classical Serre condition $(S_k)$ is to Cohen-Macaulay rings.
Properties of almost Cohen-Macaulay rings
=========================================
All rings considered will be commutative and with unit. We start by reminding some basic properties of almost Cohen-Macaulay rings.
\[obs1\] Let $A$ be a noetherian ring. Then:
a\) $A$ is almost Cohen-Macaulay iff ${{\rm ht}}(P)\leq 1+{{\rm depth}}_PA, \forall\ P\in{{\rm Spec}}(A)$ [([@K1], 1.5)]{};
b\) $A$ is almost Cohen-Macaulay iff $A_P$ is almost Cohen-Macaulay for any $P\in{{\rm Spec}}(A)$ iff $A_Q$ is almost Cohen-Macaulay for any $Q\in{{\rm Max}}(A)$ iff ${{\rm ht}}(Q)\leq 1+{{\rm depth}}A_Q$ for any $Q\in{{\rm Max}}(A)$ [([@K1], 2.6)]{};
c\) If $A$ is local, it follows from b) that $A$ is almost Cohen-Macaulay if and only if $\dim(A)\leq 1+{{\rm depth}}(A).$
Our first result is a stronger formulation of [@K1], 2.10 and deals with the behaviour of almost Cohen-Macaulay rings with respect to flat morphisms.
\[flat\] Let $u:(A,m)\to(B,n)$ be a local flat morphism of noetherian local rings.
a\) If $B$ is almost Cohen-Macaulay, then $A$ and $B/mB$ are almost Cohen-Macaulay.
b)If $A$ and $B/mB$ are almost Cohen-Macaulay and one of them is Cohen-Macaulay, then $B$ is almost Cohen-Macaulay.
*Proof:* a) We have $$\dim(A)=\dim(B)-\dim(B/mB)\leq 1+{{\rm depth}}B-\dim(B/mB)\leq$$ $$\leq 1+{{\rm depth}}B-{{\rm depth}}(B/mB)=1+{{\rm depth}}A.$$ We have also $$\dim(B/mB)-{{\rm depth}}(B/mB)=(\dim(B)-{{\rm depth}}B)-(\dim(A)-{{\rm depth}}A)\leq$$ $$\leq 1-(\dim(A)-{{\rm depth}}A)\leq 1.$$
b\) Since $u$ is flat we have $$\dim(B)=\dim(A)+\dim(B/mB)\leq 1+{{\rm depth}}(A)+{{\rm depth}}(B/mB)=$$ $$=1+{{\rm depth}}(B).$$
\[prob\]
We don’t know of any example of a local flat morphism of noetherian local rings $u:(A,m)\to(B,n)$ such that $A$ and $B/mB$ are almost Cohen-Macaulay and $B$ is not almost Cohen-Macaulay.
\[compl\] Let $A$ be a noetherian local ring, $I\neq A$ be an ideal contained in the Jacobson radical of $A$ and $\hat{A}$ the completion of $A$ in the $I$-adic topology. Then $A$ is almost Cohen-Macaulay if and only if $\hat{A}$ is almost Cohen-Macaulay.
*Proof:* Since $I$ is contained in the Jacobson radical of $A,$ the canonical morphism $A\to\hat{A}$ is faithfully flat and ${{\rm Max}}(A)\cong{{\rm Max}}(\hat{A}).$ Moreover, if $m\in{{\rm Max}}(A)$ and $\hat{m}$ is the corresponding maximal ideal of $\hat{A},$ the closed fiber of the morphism $A_m\to\hat{A}_{\hat{m}}$ is a field. Now apply \[flat\].
\[series\] [(see [@K1], 1.6)]{} Let $A$ be a noetherian ring and $n\in\mathbb{N}.$ Then $A$ is almost Cohen-Macaulay if and only if $A[[X_1,\ldots,X_n]]$ is almost Cohen-Maculay.
*Proof:* Suppose that $A$ is almost Cohen-Macaulay. We may clearly assume that $A$ is local and $n=1.$ By [@K1], 1.3 we get that $A[X]_{(X)}$ is almost Cohen-Macaulay. Now apply \[compl\]. The converse is clear.
For the next corollary we need some notations.
\[notatie\] If **P** is a property of noetherian local rings, we denote by $\textbf{P}(A):=\{Q\in{{\rm Spec}}(A)\mid A_Q\ \text{has the property}\ \textbf{P}\}$ and by $\textbf{NP}(A):=\{Q\in{{\rm Spec}}(A)\mid A_Q\ \text{has not the property}\ \textbf{P}\}={{\rm Spec}}(A)\setminus\textbf{P}(A).$
\[acm\] Let $A$ be a noetherian ring. According to \[notatie\], the set $${{\rm {\bf aCM}}}(A):=\{P\in{{\rm Spec}}(A)\ \vert\ A_P\ \text{is almost Cohen-Macaulay}\}$$ is called the almost Cohen-Macaulay locus of $A.$
\[spec\] Let $u:A\to B$ be a morphism of noetherian local rings and $\varphi:{{\rm Spec}}(B)\to {{\rm Spec}}(A)$ the induced morphism on the spectra. If the fibers of $u$ are Cohen-Macaulay, then $\varphi^{-1}({{\rm {\bf aCM}}}(A))={{\rm {\bf aCM}}}(B).$
*Proof:* Obvious from \[flat\].
In Cohen-Macaulay rings chains of prime ideals behave very well, in the sense that Cohen-Macaulay rings are universally catenary(see [@Mat]). This is no more the case for almost Cohen-Macaulay rings.
\[uncat\] There exists a local almost Cohen-Macaulay ring which is not catenary.
*Proof:* Indeed, by [@K1], Ex. 2, any noetherian normal integral domain of dimension 3 is almost Cohen-Macaulay. In [@Og] such a ring which is not catenary is constructed.
The next result shows that some of the formal fibres of almost Cohen-Macaulay rings are almost Cohen-Macaulay. A stronger fact will be proved in \[ffiber\].
\[brh\] Let $A$ be a noetherian local almost Cohen-Macaulay ring, $P\in{{\rm Spec}}(A), Q\in{{\rm Ass}}(\hat{A}/P\hat{A}).$ Then $\hat{A}_Q/P\hat{A}_Q$ is almost Cohen-Macaulay.
*Proof:* We have $$\dim(\hat{A}_Q/P\hat{A}_Q)=\dim\hat{A}_Q-\dim A_P\leq{{\rm depth}}\hat{A}_Q+1-\dim A_P\leq$$ $$\leq{{\rm depth}}\hat{A}_P+1-\dim A_P={{\rm depth}}(\hat{A}_Q/P\hat{A}_Q)+1.$$
The following result shows that the almost Cohen-Macaulay property is preserved by tensor products and finite field extensions.
\[tensor\] Let $k$ be a field, $A$ and $B$ be two $k-$algebras such that $A\otimes_k B$ is a noetherian ring. If $A$ and $B$ are almost Cohen-Macaulay, then $A\otimes_k B$ is almost Cohen-Macaulay.
*Proof:* Let $P\in{{\rm Spec}}(A).$ We have a flat morphism $B\to B\otimes_k k(P)$ and let $Q\in{{\rm Spec}}(B).$ Set $T:=A/P\otimes_k B/Q=A\otimes_k B/(P\otimes_k B+A\otimes_k Q).$ Then $k(P)\otimes_k k(Q)$ is a ring of fractions of $T,$ hence noetherian by assumption. By [@TY], 1.5, it follows that $k(P)\otimes_k k(Q)$ is locally a complete intersection. Let now $Q\in{{\rm Spec}}(B)$ and $P=Q\cap A.$ By the above the flat local morphism $A_P\to (B\otimes_kk(P))_Q$ has a complete intersection closed fiber, hence the ring $(B\otimes_k k(P))_Q$ is almost Cohen-Macaulay by \[flat\]. Now consider the flat morphism $A\to A\otimes_k B,$ let $Q\in{{\rm Spec}}(A\otimes_k B)$ and $P=Q\cap A.$ Then the flat local morphism $A_P\to (A\otimes_k B)_Q$ has a complete intersection closed fiber, whence $(A\otimes_k B)_Q$ is almost Cohen-Macaulay.
\[geomacm\] Let $k$ be a field, $A$ a noetherian $k-$algebra which is almost Cohen-Macaulay and $L$ a finite field extension of $k.$ Then $A\otimes_k L$ is almost Cohen-Macaulay.
As for the Cohen-Macaulay property, the formal fibres of factorizations of almost Cohen-Macaulay rings are almost Cohen-Macaulay.
\[ffiber\] Let $B$ be a local almost Cohen-Macaulay ring, $I$ an ideal of $B$ and $A=B/I.$ Then the formal fibers of $A$ are almost Cohen-Macaulay.
*Proof:* We have $\hat{A}=\hat{B}\otimes_B A=\hat{B}/I\hat{B},$ hence the formal fibers of $A$ are exactly the formal fibers of $B$ in the prime ideals of $B$ containing $I.$ Let $P$ be such a prime ideal, let $S = B\setminus P$ and let $C := S^{-1}(\hat{B} /I\hat{B}).$ Let also $Q\in{{\rm Spec}}(C).$ There exists $Q'\in{{\rm Spec}}(\hat{B})$ such that $Q=Q'C$ and $Q'\cap B=P.$ Thus we have a local flat morphism $B_Q\to \hat{B}_{Q'}.$ But $B$ is almost Cohen-Macaulay, hence $\hat{B}_{Q'}$ and consequently $C_Q\cong\hat{B}_{Q'}/P\hat{B}_{Q'}$ are almost Cohen-Macaulay, by \[flat\].
The property $(C_n)$
====================
Recall that given a natural number $n,$ a noetherian ring $A$ is said to have Serre property $(S_n)$ if ${{\rm depth}}(A_P)\geq\min({{\rm ht}}{P},n)$ for any prime ideal $P\in{{\rm Spec}}(A).$ Moreover, $A$ is Cohen-Macaulay if and only if $A$ has the property $(S_n)$ for any $n\in\mathbb{N}$ (see [@Mat], (17.I)). We will try to characterize almost Cohen-Macaulay rings in a similar way.
\[tn\] Let $n\in\mathbb{N}$ be a natural number. We say that a noetherian ring $A$ has the property $(C_n)$ if ${{\rm depth}}(A_P)\geq\min({{\rm ht}}{P},n)-1, \forall\ P\in{{\rm Spec}}(A).$
\[tnsn\] a) It is clear that $(C_n)\Rightarrow (C_{n-1})$ and that $(S_n)\Rightarrow (C_n), \forall n\in\mathbb{N}.$
b\) It is also clear that if $A$ has $(C_n),$ then $A_P$ has $(C_n), \forall P\in{{\rm Spec}}(A).$
\[acmtn\] A noetherian ring $A$ is almost Cohen-Macaulay if and only if $A$ has the property $(C_n)$ for every $n\in\mathbb{N}.$
*Proof:* Assume that $A$ is almost Cohen-Macaulay and let $P\in{{\rm Spec}}(A).$ Then $A_P$ is almost Cohen-Macaulay, hence ${{\rm depth}}(A_P)\geq{{\rm ht}}(P)-1.$ If $n\geq{{\rm ht}}(P),$ then $\min({{\rm ht}}(P),n)={{\rm ht}}(P),$ hence ${{\rm depth}}(A_P)\geq\min(n,{{\rm ht}}(P))-1.$ If $n<{{\rm ht}}(P),$ then $\min(n,{{\rm ht}}(P))=n,$ so that ${{\rm depth}}(A_P)\geq{{\rm ht}}(P)-1>n-1=\min({{\rm ht}}(P),n)-1.$
For the converse, let $P\in{{\rm Spec}}(A), {{\rm ht}}(P)=l.$ Then $${{\rm depth}}(A_P)\geq\min(l,{{\rm ht}}(P))-1={{\rm ht}}(P)-1.$$
\[loctk\] Let $k\in\mathbb{N}.$ A noetherian ring $A$ has the property $(C_k)$ if and only if $A_P$ is almost Cohen-Macaulay for any $P\in{{\rm Spec}}(A)$ with ${{\rm depth}}(A_P)\leq k-2.$
*Proof:* Let $P\in{{\rm Spec}}(A)$ such that $\min(k,{{\rm ht}}(P))-1\leq{{\rm depth}}(A_P)\leq k-2.$ If ${{\rm ht}}(P)\leq k,$ then ${{\rm depth}}(A_P)\geq{{\rm ht}}(P)-1.$ And if ${{\rm ht}}(P)>k,$ then it follows that $k-2>{{\rm depth}}(A_P)\geq k-1.$ Contradiction!
Conversely, let $P\in{{\rm Spec}}(A).$ If ${{\rm depth}}(A_P)\leq k-2,$ then $A_P$ is almost Cohen-Macaulay, hence ${{\rm ht}}(P)-1\leq{{\rm depth}}(A_P)\leq k-2.$ Thus $\min({{\rm ht}}(P),k)={{\rm ht}}(P),$ whence ${{\rm depth}}(A_P)\geq\min(k,{{\rm ht}}(P).$ If $k-2<{{\rm depth}}(A_P),$ then ${{\rm ht}}(P)>k-2,$ hence ${{\rm depth}}(A_P)\geq\min(k,{{\rm ht}}(P))-1.$
\[regel\] Let $A$ be a noetherian ring, $k\in\mathbb{N}$ and $x\in A$ a non zero divisor. If $A/xA$ has the property $(C_k)$, then $A$ has the property $(C_k).$
*Proof:* Let $Q\in{{\rm Spec}}(A)$ such that ${{\rm depth}}(A_Q)=n\le k-2.$ If $x\in Q,$ then ${{\rm depth}}(A/xA)_Q=n-1\leq k-3.$ Then ${{\rm ht}}(Q/xA)\leq n-1+1=n,$ hence ${{\rm ht}}(Q)\leq n+1={{\rm depth}}A_Q+1.$ If $x\notin Q,$ let $P\in{{\rm Min}}(Q+xA).$ Then $(P+xA)A_Q$ is $QA_Q$-primary and ${{\rm depth}}(A_P)\leq{{\rm depth}}(A_Q)+1=n+1.$ Then ${{\rm depth}}(A/xA)_Q=n-1,$ hence ${{\rm ht}}(P/xA)\leq n.$ It follows that ${{\rm ht}}(P)\leq n+1={{\rm depth}}(A_P)+1.$
\[nagcr\] We say that a property **P** of noetherian local rings satisfies Nagata’s Criterion (NC) if the following holds: if $A$ is a noetherian ring such for every $P\in \textbf{P}(A),$ the set $\textbf{P}(A/P)$ contains a non-empty open set of ${{\rm Spec}}(A/P)$, then $\textbf{P}(A)$ is open in ${{\rm Spec}}(A).$
An interesting study of Nagata Criterion is performed in [@Mas].
\[Tknagata\] Let $k\in\mathbb{N}.$ The property $(C_k)$ satisfies (NC).
*Proof:* Let $Q\in C_k(A).$ Then ${{\rm depth}}(A_Q)\geq\min(k,{{\rm ht}}(Q))-1.$
Case a): ${{\rm ht}}(Q)\leq k.$ Then $\min(k,{{\rm ht}}(Q))={{\rm ht}}(Q),$ hence ${{\rm depth}}(A_Q)+1\geq{{\rm ht}}(Q)$ and $A_Q$ is almost Cohen-Macaulay. Let $f\in A\setminus Q$ such that $$\dim(A_P)=\dim(A_Q)+\dim(A_P/QA_P)$$ and $${{\rm depth}}(A_P)={{\rm depth}}(A_Q)+{{\rm depth}}(A_P/QA_P)$$ for any $P\in D(f)\cap V(Q)\cap NT_k(A).$ Then ${{\rm depth}}(A_P)\ngeqq\min(k,{{\rm ht}}(P))-1.$
Case a1): ${{\rm ht}}(P)\leq k.$ Then $\min(k,{{\rm ht}}(P))={{\rm ht}}(P),$ hence ${{\rm depth}}(A_P)+1<{{\rm ht}}(P).$ Then $${{\rm depth}}(A_P/QA_P)+1={{\rm depth}}(A_P)-{{\rm depth}}(A_Q)+1<$$ $$<{{\rm ht}}(P)-{{\rm depth}}(A_Q)\leq{{\rm ht}}(P)-{{\rm ht}}(Q)+1.$$ Then ${{\rm depth}}(A_P/QA_P)<\dim(A_P/QA_P)=\dim(A_P)-\dim(A_Q)$ and it follows that $A_P/QA_P$ is not $(C_k).$
Case a2): ${{\rm ht}}(P)> k.$ Then $\min(k,{{\rm ht}}(P))=k,$ hence ${{\rm depth}}(A_P)<k-1.$ It follows that $${{\rm depth}}(A_P/QA_P)={{\rm depth}}(A_P)-{{\rm depth}}(A_Q)<$$ $$<k-1+1-{{\rm ht}}(Q)=k-{{\rm ht}}(Q).$$ This implies that $A_P/QA_P$ is not $(C_k).$
Case b): ${{\rm ht}}(Q)>k.$ Then $\min(k,{{\rm ht}}(Q))=k$ and ${{\rm depth}}(A_Q)+1\geq k.$ Since ${{\rm ht}}(P)>k,$ it follows that $\min(k,{{\rm ht}}(P))=k$ and ${{\rm depth}}(A_P)+1<k.$ Let $x_1,\ldots,x_r$ be an $A_Q$-regular sequence. Then there exists $f\in A\setminus Q$ such that $x_1,\ldots,x_r$ is $A_f$-regular. If $P\in D(f)\cap V(Q),$ it follows that $A_P$ is $(C_k).$
\[acmnagata\] The property almost Cohen-Macaulay satisfies (NC).
\[open\] Let $A$ be a quasi-excellent ring and $k\in\mathbb{N}.$ Then $C_k(A)$ and ${{\rm {\bf aCM}}}(A)$ are open in the Zariski topology of ${{\rm Spec}}(A).$
*Proof:* Let $P\in{{\rm Spec}}(A).$ Then ${{\rm {\bf aCM}}}(A/P)$ and $C_k(A/P)$ contain the non-empty open set ${\bf Reg}(A/P)=\{ P\in{{\rm Spec}}(A)\ \vert\ A_P\ \text{is regular}\ \}.$ Now apply \[Tknagata\] and \[acmnagata\].
\[opencom\] Let $A$ be a complete semilocal ring and $k\in\mathbb{N}.$ Then $C_k(A)$ and $aCM(A)$ are open in the Zariski topology of ${{\rm Spec}}(A).$
\[cmfib\] Let $A$ be a noetherian local ring with Cohen-Macaulay formal fibers. Then ${{\rm {\bf aCM}}}(A)$ is open.
*Proof:* Follows from \[opencom\] and \[spec\].
\[flattk\] Let $u:A\to B$ be a flat morphism of noetherian rings and $k\in\mathbb{N}.$ If $B$ has $(C_k),$ then $A$ has $(C_k).$
*Proof:* We may assume that $A$ and $B$ are local rings and that $u$ is local. Let $P\in{{\rm Spec}}(A)$ and $Q\in{{\rm Min}}(PB).$ Then $\dim(B_Q/PB_Q)=0,$ hence $${{\rm depth}}(A_P)={{\rm depth}}(B_Q)\geq\min(k,\dim(B_Q))-1=$$ $$=\min(k,\dim(A_P))-1.$$
\[flattk2\] Let $u:A\to B$ be a flat morphism of noetherian rings and $k\in\mathbb{N}.$
a\) If $A$ has $(C_k)$ and all the fibers of $u$ have $(S_k),$ then $B$ has $(C_k).$
b\) If $A$ has $(S_k)$ and all the fibers of $u$ have $(C_k),$ then $B$ has $(C_k).$
*Proof:* a) Let $Q\in{{\rm Spec}}(B), P=Q\cap A.$ Then by flatness we have $$\dim(B_Q)=\dim(A_P)+\dim(B_Q/PB_Q),$$ $${{\rm depth}}(B_Q)={{\rm depth}}(A_P)+{{\rm depth}}(B_Q/PB_Q).$$ By assumption we have $${{\rm depth}}(A_P)\geq\min(k,{{\rm ht}}(P))-1,$$ $${{\rm depth}}(B_Q/PB_Q)\geq\min(k,\dim(B_Q/PB_Q).$$ Hence we have $${{\rm depth}}(B_Q)={{\rm depth}}(A_P)+{{\rm depth}}(B_Q/PB_Q)\geq$$ $$\geq\min(k,{{\rm ht}}(P))-1+\min(k,\dim(B_Q/PB_Q))=\min(k,{{\rm ht}}(B_Q))-1.$$
b\) The proof is the same.
As a corollary we get a new proof of a previous result.
\[flatacm\] Let $u:A\to B$ be a flat morphism of noetherian rings.
a\) If $B$ is almost Cohen-Macaulay, then $A$ is almost Cohen-Macaulay.
b\) If $A$ is almost Cohen-Macaulay and the fibers of $u$ are Cohen-Macaulay, then $B$ is almost Cohen-Macaulay.
\[exemplu\] Let $k$ be a field and let $X_0,X_1,X_2,Y_1,Y_2$ be indeterminates. Set $B=k[[X_0,X_1,X_2]]/(X_0)\cap(X_0,X_1)^2\cap(X_0,X_1,X_2)^3$ and $A:=B[[Y_1,Y_2]].$ It is easy to see that $A$ is a noetherian local ring with $\dim(A)=5, {{\rm depth}}(A)=2.$ It is also not difficult to see that $A$ has the property $(C_3)$ and not the property $(C_4).$ Similar other examples can easily be constructed.
\[altex\] Let $k$ be a field, $X,Y$ indeterminates and consider the ring $A=k[[X,Y]]/(X^2,XY).$ Then $A$ has $(C_2)$ and not $(S_2).$
[99]{} Han, Y.(1998), D-rings. *Acta Math. Sinica* 4:1047–1052. Kang, M.C.(2001), Almost Cohen-Macaulay modules. *Comm. Algebra* 29:781–787. Kang, M.C.(2002), Addendum to “Almost Cohen-Macaulay modules”. *Comm. Algebra* 30:1049–1052. Massaza, C.(1981), Su alcuni aspetti assiomatici del criterio di Nagata. *Rend. Circ. Mat. Palermo* XXX:365-377. Matsumura, H.(1980), Commutative Algebra, 2nd Edition, Benjamin Cummings, 1980.
Ogoma, T.(1980), Non-catenary pseudo-geometric normal rings. *Japan J. Math.* 6:147–163. Tousi M. and Yassemi S.(2003), Tensor products of some special rings. *J. Algebra* 268:672–675.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We’ve all heard how academia evaluates our work: papers, papers, and... more papers. While academic publishing is an essential part of disseminating research results, it is only one activity amongst many that make up a career (or for that matter, a life). The following essay is based on an invited talk for the NSF AAPF Symposium, given at the 2018 Winter AAS Meeting. Here, as in the original talk, I offer some practical advice for thinking about one’s work within the larger frame of your personal values and goals. While I will draw examples from my own career, I hope to offer readers guidance in articulating what is important to them, aligning their career choices with those values, and establishing metrics that go beyond the h-index.'
author:
- 'Lucianne M. Walkowicz'
title: |
Choose Your Own Adventure:\
Developing A Values-Oriented Framework for Your Career
---
Background
==========
When I was asked to speak at the NSF AAPF Symposium back in the fall of 2017, the request was very open-ended. It happened to come at an interesting time for me, in that I had just moved to work at the Library of Congress for a year, where I could devote myself mostly full time to studying the ethics of Mars exploration, which I have been thinking about in my non-copious spare time for a while. As such, I had stepped a bit away from the research I was focused on in Chicago, but I was not yet ready to give a full talk on the work I’m doing at the Library. The NSF AAPF Symposium is also somewhat unique, in that it’s not often that one has a chance to speak with an audience oriented towards postdocs and organized specifically around a postdoctoral fellowship, so I thought about what my postdoc self would have liked to hear a talk about. What follows is a slightly edited version of the talk I decided to give.
As corny as it might sound, this essay is about success– but success that is defined by you, rather than by the external measuring sticks that those of us in academia are used to measuring ourselves with. A better word for success in this context might be “fulfillment”. “Fulfillment” (as I use it here) is a condition that describes the path you are on, rather than a destination you can reach– and practically speaking, it is only partially in your control, because it is created both by your choices, as well as by circumstances beyond you.
My argument is that fulfillment stems from aligning your career path with your personal values, and that a great deal of unhappiness comes from a mismatch in the way you are evaluated versus the way you see yourself generating value in the world. Naturally, I can’t tell you what your values are, or what choices you should make– but I hope to provide a way of thinking about choices: a framework for making decisions and prioritizing what you do that you can apply to your own life, if and where applicable.
First, a short story: when I was in my first postdoc at Berkeley, I went to a colloquium given by Matthew Bailes of Swinburne University of Technology. He began the talk by telling us that he was really enjoying his visit, that we all seemed really bright, and then he essentially told us that most of us would no longer be astronomers in a few years. He ran through the numbers of how many grads and postdocs were in the department, how many tenure track faculty jobs there are out there, and made a pretty convincing case that we would mostly not be getting the faculty jobs we were ostensibly trying to get. Matthew has a really great, dry sense of humor, so it was a lightly delivered blow, but literally no one wants to hear that, even if you suspect or know it to be true!
Once he’d spent the first five minutes of colloquium crushing our dreams, Matthew went on to tell us what happened when he found himself with a career not working out the way he’d planned– and how he eventually came to found the department he is currently in, which he served as director of for over a decade. His talk really stuck with me, because it had always seemed to me that everyone who’d arrived at a stage in their careers where they were giving colloquia at Berkeley had just breezily sailed their way through to there. I’d never heard anyone so much as acknowledge that they had struggled, I’d never heard anyone express vulnerability, and I’d never heard any discussion of what one does when those struggles arise. After the talk, we postdocs debated the relative roles of choice and chance in Matthew’s eventual success, and it seemed to me (and still does) to have been a combination of the two.
Shortly after that colloquium, I applied for and got the TED Fellowship. Many people have heard of TED Talks, but if you don’t know, those TED talks come from a big conference, not unlike the AAS in size, but attended by various movers and shakers from the worlds of Technology, Entertainment, and Design (which is what its acronym stands for). However, it is several orders of magnitude more expensive to attend TED than the AAS, so TED started the TED Fellows program to bring early career people who would never be able to afford it to the conference for free. TED was a great experience for me in a variety of ways, but possibly the biggest outcome was that they teamed me up with a career coach, Jen Sellers, who has patiently spent the last 5-6 years walking me through the dark forest of uncertainty that is being a living human being, and most especially, a postdoc. Just as a good research talk distills down some of what one has learned about the universe, I hope to pay Jen’s generosity forwards a bit by distilling some of what I’ve learned by working with her through my postdoc years and into early faculty years.
Starting Points
===============
This essay is considerably more personal than any other professional writing I have done to date. Since the guidance I am providing is based in my own personal experiences, I’d like to acknowledge that this will not be a cure-all or universal solution. I am starting from a place I call “So Far, So Good”: this is what works for me, but it doesn’t mean that my choices are “correct”, or that they should be your choices, or even that these choices will be good forever. They are intended to be illustrative rather than definitive. Like I said: so far, so good.
I would also be remiss not to acknowledge that I am speaking from privilege, in a variety of ways: one being that I am a decade out of grad school and several years out of being a postdoc, so I have achieved some form of success: I am still in astronomy, I have a position that pays me to be in astronomy, and people ask me to give talks. I have other obvious privileges as well– I am white in a mostly white discipline, and though we are also in a predominantly male field, white, cis-sex women have gained more entry to the field of astronomy than our colleagues from other marginalized backgrounds. While I struggle with both mental and physical disability, both are mostly well-controlled through treatment and both are usually not visible, so in sum, I have it easier than a lot of people.
However, one of the things that has been a challenge for me I think actually has helped me deal with a life in academia, particularly post-grad school, and that’s that I grew up with a lot of instability. I come from a family where I was expected to go to college, but I also went to college to get away from an abusive home, and after that I went to grad school not only because I wanted to be an astronomer, but because I had no plan B. I have been without a safety net for most of my life, which for better or worse, one gets used to.
In reflecting on some of the choices I’ve made in my life as well as my career, I think there are two events that shaped my framework for decision making more than others: the first of these is that when I was four, my father dropped dead. He went to work in the morning, felt sick at work, had an aneurysm, and the next time I saw him was in a coffin. Though I was really young, I remember that time very well, and I think this forever cemented in my mind that work might be what you are doing on the last day of your life.
The second major event is that, in my freshman year of college, I went in for a routine health check up and ended up being diagnosed with a pituitary tumor, which is located smack in the middle of your brain. I won’t go into details of what that was like, but much like my dad’s death altered my path forever, I learned that circumstances can change much faster than you anticipate. It took me much longer to learn the accompanying lesson that this event should have taught me, which is to prioritize and safeguard your health.
Determining Your Values
=======================
That concludes the death and illness portion of the essay– on to sunnier topics, like being fulfilled in your career! As I’ve said, fulfillment is only partly up to you– so how are we to wrestle with something that isn’t completely ours to direct? I began by telling you that it is essential to align your actions and values, so one way to start is by consciously interrogating the values you hold– which [**are**]{} up to you– and evaluating decisions based on those values.
As you read the following, keep in mind these two disclaimers: making your decisions in this way may not result in a career in astronomy. Furthermore, aligning your career path with your values may also not be compatible with a career in astronomy, and as a corollary to that: astronomy may ultimately not be the dream of fulfillment you hope for.
However, I think that engaging in this way may be helpful for the following reason: unless you create them for yourself, there are relatively few formal opportunities to think about your decisions in a larger framework. In academic astronomy, the main time we do this kind of exercise is in planning our path to completing grad school: we choose a research topic, look at what actions we must take to address that topic, and there is a clearly defined marker of success (completing grad school with the terminal degree of your choice). However, once one leaves grad school, the goal posts move. As a postdoc, you know you are supposed to publish– because that will make you competitive for a second postdoc, or perhaps for a faculty job. If you’re in your second postdoc, you’re also supposed to publish, for the same reasons. If you pass go and move on to being junior faculty, you are once again supposed to publish, but now you’re also supposed to get grants, teach classes, and mentor students. In some ways it’s no wonder that publication record becomes a proxy for productivity– it’s a number, something quantifiable that one can quickly determine. The impact of other activities, by comparison, may seem “squishier” and more complicated.
It is also true that some activities in astronomy are more valued in hiring than others– a very common complaint I hear is that many junior astronomers are interested in outreach and public communication of their science, but that these activities are either not valued, or worse, public-facing activities are actively discouraged because they are considered a “distraction”. Another common issue is that people are not valued for software development, even if that software is as (or more!) widely used than a hardware instrument like a spectrograph. The valuing of software has made some progress, but only by being shoehorned into something that is publishable and citable. The relative valuing of various activities is an example of an external measuring stick, a measure that is imposed upon us regardless of whether it aligns with our values. You cannot prevent people from measuring you according to their values, but you can make sure what you do aligns with your own. Remember: work may be what you are doing on the last day of your life.
Practical Strategies for Building Your Framework
================================================
So with all that context, let’s move onto some practical strategies. It is impossible to consciously align your actions with your values if you don’t know what your values are. If you are anything like I was as a grad student and postdoc, you are perhaps vaguely aware of having a set of values, and you may intuitively have a sense of whether you are acting within them or not. Very few people, unless prompted, have to specifically articulate what their values are– but doing so can be a very useful exercise. For those of you with teaching experience, it’s a little like the understanding of a scientific topic you gain when you’re required to explain it to someone else, where putting words and specificity to something can bring clarity, or highlight areas that need to be clarified.
Write a Mission Statement
-------------------------
Identifying and articulating your values is a process, but one can use that process to create a helpful beacon: the mission statement. We usually encounter mission statements in the context of organizations– for example, here’s the Adler’s:
> [*The mission of the Adler Planetarium is to inspire exploration and understanding of our Universe.*]{}
And here is that of the Library of Congress:
> [*The Library of Congress’s mission is to support Congress in fulfilling its constitutional duties, and to further the progress of knowledge and creativity for the benefit of the American people.*]{}
Since we are used to hearing about mission statements in the context of institutions, it might sound weird for you to have one, just for you– but it’s really nothing more than taking a close look at what’s important to you, and writing it down. There are a lot of guides for writing personal mission statements online, but I find most of them fairly cheesy, so I’ll distill the process into prompts that I find helpful.
### Questions for Reflection
One place to start is to think of who you admire, and what qualities you admire in them. Maybe they’re some famous scientist, maybe they’re a family member, friend, or colleague. What do they embody that you admire? These don’t have to be perfect people, they don’t have to embody who you want to be in every aspect of their being– the important thing is to think about [**why**]{} you admire them.
What do you see as your purpose? What is important to you? When you do something that feels like an accomplishment, why does it feel that way? In all of this, it is helpful to be as honest and non-judgmental with yourself as possible– for example, perhaps you like giving public talks because you feel as though you are helping people understand the universe, but perhaps you also like being on stage and performing for a crowd. One of these is driven by an urge to help others, one by the needs of one’s own ego, but it’s important to understand both if they are both at play.
Brainstorm your aptitudes: what do you think you are good at? What do you really enjoy doing? What were you doing the last time you felt “good-tired” instead of just “tired-tired”?
Lastly (and here’s where we return to death for a moment): what do you want to be known for? Sometimes, this is phrased as your “legacy”, which definitely implies you have died but never mind that for a moment– what would you want your colleagues, friends and family to say about you?
Now the hard part: much like deciding what your thesis topic is going to be, you now have to take the results of this exercise and turn it into a specific articulation of your intent. It helps to be as succinct as possible, but it’s okay if your mission is a couple different statements that complement each other. Think 1-3 sentences, not a page. Sometimes phrasing can help: I personally found it too intimidating to write those sentences with “My mission is...” so instead, mine start with “I want to...” Phrase it however works for you.
Finally, the really hard part: get someone else to read it. Your mission statement is for you, and you don’t have to declare it publicly or put it on the internet or anything like that– but again, it resembles teaching in that the more clearly you can explain it to someone else, the more clearly you yourself will understand it. So share what you’ve written with a close friend, or even make a pact with a good colleague to work on this together. Can you explain what you mean to them? Do you understand it yourself?
One final word on mission statements: they can change, and they can evolve. Always let what you’ve written guide you without constraining you. Your mission statement should be a lighthouse, not an anchor.
Make Conscious Choices
----------------------
Life as a postdoc is a time of opportunity, decisions, and the massive amount of stress that accompanies both. You probably have a lot more freedom than you did as a grad student– you have more choice in what you work on, who you work with, what conferences you attend, and ultimately, what jobs you apply to. The downside is that choices can be overwhelming, especially when you are bogged down with trying to get a lot of work done in the limited duration of your postdoc.
In my experience, towards the end of grad school and in early postdoc was also a time of some desperation– we all hear stories about how hard it is to get a job in the field, and so when I was earlier in my career I definitely took the “spray of buckshot” approach to job applications: I applied to anything and everything I could remotely be considered qualified for. It stressed me out, and worse, it wasted my time. I can also tell you now, having hired a postdoc myself, that I am not the only person that does this– part of the reason that astronomy jobs have pools of 300+ applicants is that people are submitting the same application to a lot of jobs they have not closely evaluated. You can’t blame anyone for doing that, because seemingly no one is taught how to decide whether a job is suitable for them or not– but don’t do that. The people reading your applications can tell.
There are good and bad aspects to casting a wide net, just like there are good and bad things about having opportunities. When you are starting out, it can be helpful to take a “many seeds in the ground” approach, because sometimes you can’t tell which will germinate and eventually grow into something more. However, the flip side of that is that you can get stuck in a cycle of unfocused pursuit, going after any and all things with equal attention.
### Questions for Evaluating Opportunities
To help manage your energy, it can be useful to come up with a series of questions you ask yourself when trying to decide what to pursue. These questions will differ from person to person, and even for individuals there will be times when these questions may not be applicable. However, if you don’t have a framework to evaluate your decisions, it can be hard to cross-compare competing decisions, and it can be difficult after the fact to articulate [**why**]{} you made those decisions. As an example, here are the questions that I came up with:
- [**How well does this opportunity fit with my mission and values?**]{} This question is the big one– the things you spend your time on should serve your mission and goals related to it.
- [**What is new and different about this opportunity?**]{} Does it overlap with or duplicate any existing things I’m doing? What do I hope to gain as a result of taking this opportunity? Be specific: let’s say you are invited to a conference. If you are looking for exposure for a new science result, consider who will be in your audience. If you are looking to start new collaborations, look for overlap with people you are hoping to speak with. As a postdoc there can be value in getting yourself out and seen as much as possible, but there’s also a limit to that. A related point here is that your hiring will eventually depend on people outside your subfield, so while specialty meetings that are discipline specific are useful for starting new collaborations or identifying future colleagues/employers, meetings that are broader, like ones around a specific telescope or facility, help expose your work to a wider swath of the population.
- [**How long do I think it will take to fulfill this commitment, or what is the time demand of this opportunity?**]{} Again, be specific: if you join a users’ panel or advisory committee and the duration of your commitment to them is three years, that can be made more granular by looking at how many meetings you will have, whether those meetings are a day long or a couple days long, and whether you will have to travel or not.
Then, at least in my case: [**multiply that time estimate by a factor of 2-3**]{}. Know yourself well enough to know when you have regular mistakes you make, like underestimating how long things take, which is something I do all the time.
- [**What am I willing to give up for this opportunity?**]{} This question is a rough one, but it’s one you must ask. As a grad student, then postdoc, then faculty, you will repeatedly have more things on your to do list than you can handle. When you take on a new opportunity or commitment, there will be a cost– most of the time, most people don’t articulate what that is when they take something on, and they end up bearing the cost in some way they didn’t anticipate. It could be something simple, like you will have to work an hour later each day, but often these simple things are not what they seem: does that mean an hour less spent with your friends, partner, or family? Does it mean an hour less sleep? These can be difficult things to quantify and hard to look at when you do, but it’s important that you are honest and open-eyed.
Applying Your Framework
-----------------------
Once you have identified your values, and created a list of questions that can help you analyze opportunities that come along, you’re ready to start applying this framework. Below, I discuss two brief examples of using the evaluation questions in the previous question to specific situations that arose for me (and may arise for the reader).
### What jobs should I apply for?
As I previously mentioned, my first few rounds of the job slog, I applied for literally everything under the sun. I spent a lot of my own time writing applications for places of dubious fit for me, and perhaps worse I also wasted the time of my letter writers, and of the people on the other end reading the applications. It wasn’t until my last time on the job market that I decided to approach my search with a different tactic. My last round of applications was in the second year of my second postdoc, a year [*before*]{} I absolutely needed a job. When I ran through the list of questions I’d come up with, I saw that what was different about applying then was that I wasn’t absolutely required to apply, so I could be a bit more selective, thus limiting the amount of time I’d have to spend (and thus things I’d have to sacrifice) in Questions 2 and 3. I therefore chose my applications based on a single criterion: I only applied to a job if I could, after reading the job ad (and maybe doing a minimal amount of research on the department, if I wasn’t already familiar with them) express in a sentence or two why I was a good fit for the job, and why it was a good fit for me. Job applications, after all, are arguments– you are arguing for your suitability to a particular position. If I felt like I couldn’t make that argument right away, I passed on applying for that job.
### How do I evaluate a faculty offer?
That job application cycle was successful for me, in part because of my approach, but also (and perhaps more significantly) I was fortunate that there were a number of opportunities for which I felt I could make a good argument. Again, there are factors out of your control, and I am writing from a position of privilege. During the process of writing down why I thought certain jobs were a good fit, I realized that I thought I would be happy in a fairly wide variety of settings, from teaching-heavy, undergrad-only institutions, to traditional research universities, to something focused on public communication (like the Adler). I interviewed at a sampling of each of these kinds of institutions, and ultimately got two offers, one from the Adler, and one from a very traditional physics and astronomy department at a research university, which I will refer to as Research University.
I’d like to take a moment to acknowledge here that one factor beyond your control are the norms of astronomical society. When I was in grad school, and I think this is still true now, getting a tenure-track faculty position at a research university was supposed to be the pinnacle of one’s existence as an astronomer. Nobody outright told me that, of course, but I had gleaned as much from spending time in astronomy environments. Creating and asking myself these evaluation questions helped me pay attention to my needs and desires, rather than the clamoring voice of societal pressure.
Both opportunities were new and different, because they were both faculty offers, so that wasn’t a great distinguishing criterion. Time-wise, I knew they were both going to absorb all of my time in one way or another, because they essentially represented a phase change in my career. However, they presented fairly different sacrifices: at Research University, I would be expected to do what traditional faculty do: conduct research, mentor graduate students, teach classes, apply for grants, and probably service work for the university in one way or another. After I got the offer from them, I visited Research University, and asked how success (presumably defined in relationship to getting tenure) was defined there. The response (and I’m paraphrasing here but not by a lot) was “do a lot of research and publish a lot of papers, don’t mess up teaching, and when you have tenure \[presumably in 6 years\] you can do outreach”. That meant to me that I would need to spend a minimum of 6 years [**not**]{} doing the public communication that had become very important to me, and I kept thinking, “what if I get hit by a bus in Year Five?”
The Adler, on the other hand, had its own sacrifices: while the positions are not term limited, they are also not tenure track. It’s also a non-profit institution and on a 12 month, rather than 9-month, salary– which means the job is less stable and the pay is lower than what I would have the potential to earn at a university. In general, your starting salary is one of the biggest influences on your lifetime earnings, so that also means lower lifetime earnings overall.
So: was I willing to sacrifice the part of me that likes spreading scientific knowledge beyond the experts, or the part that wants to never be in danger of eviction again? In the words of A Tribe Called Quest, “Riding on the train with no dough, sucks”– but ultimately, the decision came down to values. I decided that I would rather devote the majority of my time convincing as many people as possible that science is accessible to them if they want it, than to focus on creating new astronomers by shepherding graduate students through research programs. The Adler clearly offered me the opportunity to not only speak with the broader public, but specifically to reach young people of color via the Chicago Public School system field trips, and our outreach programs throughout the city. At Research University, I would largely be reaching kids who had already made it to college, or to grad school. Perhaps most importantly of all, my success at the Adler would be evaluated along similar criteria to what I considered important for myself, demonstrating the alignment of the institution’s values with my own.
Evaluation Metrics: Measuring Your Progress Based On Your Values
----------------------------------------------------------------
As I mentioned at the beginning of this talk, it isn’t really all that surprising that publications have become the de facto metric by which success in the field is determined: it’s a number, something that can quantify the impact of your research output, something simple enough that it can be calculated by Google Scholar or any number of online tools, and then used to cross-compare between astronomers in very different subfields. As a side note, astronomy has a particularly high publication rate compared to other sciences, which makes it an easy metric too.
One of the things I have struggled with is that academic publishing is probably my least favorite thing about astronomy– I think it’s important, I like doing research, and I like writing, but I am not great at taking papers to completion (fortunately, I have great collaborators who are better at this than I am). So, I am perhaps a bit biased– but also, I find it to be a poor proxy for my own output in the field: a H- or i-index doesn’t capture things like writing or speaking for non-specialist audiences, creating public science programs, running workshops, managing collaborations, mentoring students, or really most of what I actually spent my time on. Even though I could recognize that this was true, however, I spent many years feeling like a failure, not only as a postdoc but also as faculty, because I didn’t think I published “enough”. On the other hand, I was also enjoying and feeling fulfilled in the things I actually spent time on, but always with guilt and fear hanging over me.
In the past year or so, it’s come to me that while I had a way of making decisions about what to do within that value-oriented framework at the beginning of an activity, I lacked a way of evaluating whether my continuing actions continued to be in line with that framework. A rising H-index, by comparison, is supposed to tell you that you are doing well– so in short, how do [**I**]{} know I’m doing well?
Coming up with metrics for your progress is difficult, in part because success and fulfillment are qualities of evolving pathways, rather than destinations or fixed outcomes. An evaluation metric therefore [**can**]{} be a concrete outcome– as in, you could aim to publish a certain number of papers, or give a certain number of talks– but perhaps more importantly, it could be any kind of indicator that is the outward manifestation of your value system. Thinking of traditional metrics of success works within this framework too– beyond publishing, you could have a certain number of grant applications, or if you’re bold, a certain grant acceptance rate could be a metric for you. The idea here though is to expand upon those to be inclusive of non-traditional metrics as well.
Example (Personal) Metrics
--------------------------
To provide some concrete examples, here is a sampling of the metrics I personally work with, along with what they tell me.
- [**Referrals of people in need (usually students)**]{}: One of the things that has started to happen in the past few years is that people have started to come to me for advice. Indeed, that is where this essay comes from. A significant career health indicator for me, then, is how often people not only come to me for help, but how often they refer others to me as well. What this is an indicator of to me is the broader astronomical community valuing my expertise– it is an acknowledgement that I have a contribution to make.
- [**Student mentoring**]{}: When I originally took my position at the Adler, I knew that I was sacrificing, in large part, the opportunity to work closely with undergrad and grad students. However, that didn’t turn out to be the case completely– I ended up working with a graduate student from a local institution (Daniel Giles of Illinois Institute of Technology)– and following that, ended up founding the LSSTC Data Science Fellowship Program for graduate students. I also serve as an unofficial auxiliary mentor via the kinds of referrals I mentioned before, which sometimes evolve into continuing mentoring relationships, and sometimes those referrals are for people facing specific crises. If I am working with students, especially to solve problems, I can be sure that I am providing functional, concrete help to those people– which is important to me.
- [**Spaces created for enfranchisement– of both non-scientists and scientists**]{}: A big motivating factor for me in my work is to help increase accessibility and participation in STEM for diverse peoples, including those who don’t necessarily want to be scientists themselves. I therefore think of public engagement– in any form, whether in the form of a public program through the Adler, or in terms of giving a talk– as creating a space that empowers people to ask their own questions. Many of us in astronomy focus on disseminating information to people in a way which is accessible, which is a good thing to do, and which is why the primary mode of “outreach” is public talks. However, one-way forms of communication don’t necessarily leave people feeling empowered to question the world around them. One-way forms of communication also don’t give us, as scientists, opportunities to listen and learn what people actually need. Part of my goal is also to better understand how to make astronomy– and broadly, STEM– better serve people who have been traditionally marginalized. Instead of asking “how do we get more diversity in STEM”, we should be asking what we offer, instead of assuming the value is there. I therefore look at how often I am able to create space for someone to feel empowered as a health indicator for my work.
- [**Audience reach**]{}: Audience reach is an easy example of a seemingly “squishy” aspect of communication that actually is easily quantifiable: the number of people you reach with your efforts, whether they be focused on peer-to-peer communication, or focused on reaching a broader audience. Your audience is a quantifiable proxy for the potential reach of your message, whatever that happens to be.
In addition to audience, many of the above seemingly “squishy” things can be quantified by asking simple questions: “how many?” “how often?” “for who?” Thinking of each metric in this way can not only help you understand how well you’re doing, it can also help you communicate your actions to others who might be more used to measuring your “output” by more traditional metrics. On top of that, you can choose from a world full of options by considering impact ahead of time: think about the likely ultimate outcome of an action in terms of the metrics that matter.
A Few Final Thoughts
====================
Of the afflictions common to academic life, feeling out of control and/or like your efforts will never be enough surely rank near epidemic proportions. Having said that, it is important to remember that you do have agency, and more importantly, that your value is not equivalent to your productivity. Playing the game of academia by someone else’s rules will not necessarily result in a job– so you may as well play by your own. Also, the normative “rules” of the game can change: if you mold yourself solely by the rules that are normative now, you deny the fact that our community values can, should, and are changing, and that change is driven by [*people like you*]{}.
I will also point out that an essential part of this process is knowing your relationship to risk: beyond values, a lot of what these decisions come down to is understanding the level of risk you consider acceptable in life. In my own experience, very few things are as safe or stable as they seem, but your mileage may vary. After all, it is also possible to fail: you can come up with a mission statement, personal metrics for success, etc, and you still might not meet your goals despite your best efforts. That is OK– it might suck, but it happens. If that happens, it bears remembering that this framework is flexible, one that can and should be revisited and revised over time.
One of the powers here is that having a personal framework also helps you identify fit [**beyond**]{} the field of astronomy, too– whether that is in seeking a job industry, or in communicating what you actually do all day to friends and family outside astronomy. While within the field, we often strongly emphasize specialization, the truth is that you are likely more flexible and adaptable in your skills and abilities than your research training would lead you to believe.
Many thanks to the organizers of the NSF AAPF Symposium– Cameron Hummels, Abby Crites, and Phil Rosenfield– for inviting me to speak and thus prompting the creation of this work. This essay (and my whole life) has deeply benefitted from conversations with and advice from friends, many of whom read and gave me feedback on the ideas presented here. I would particularly like to thank Jedidah Isler, Renée Hložek, Jeff Oishi, Nicole Cabrera Salazar, and Adam Miller, for their friendship, thought-provoking discussion, and insightful comments.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We perform $N$-body simulations on a multiple massive black hole (MBH) system in a host galaxy to derive the criteria for successive MBH merger. The calculations incorporate the dynamical friction by stars and general relativistic effects as pericentre shift and gravitational wave recoil. The orbits of MBHs are pursed down to ten Schwarzschild radii ($\sim 1$AU). As a result, it is shown that about a half of MBHs merge during 1 Gyr in a galaxy with mass $10^{11}M_{\odot}$ and stellar velocity dispersion $240$ km s$^{-1}$, even if the recoil velocity is two times as high as the stellar velocity dispersion. The dynamical friction allows a binary MBH to interact frequently with other MBHs, and then the decay of the binary orbits leads to the merger through gravitational wave radiation, as shown by @Tanikawa11. We derive the MBH merger criteria for the masses, sizes, and luminosities of host galaxies. It is found that the successive MBH mergers are expected in bright galaxies, depending on redshifts. Furthermore, we find that the central stellar density is reduced by the sling-shot mechanism and that high-velocity stars with $\sim 1000$ km s$^{-1}$ are generated intermittently in extremely radial orbits.'
author:
- |
A. Tanikawa$^{1,2,3}$[^1] and M. Umemura$^{1}$\
$^{1}$Center for Computational Sciences, University of Tsukuba, 1-1-1, Ten-nodai, Tsukuba, Ibaraki 305-8577, Japan\
$^{2}$School of Computer Science and Engineering, University of Aizu, Tsuruga, Ikki-machi, Aizu-Wakamatsu, Fukushima 965-8580, Japan\
$^3$RIKEN Advanced Instistute for Computational Science, 7-1-26, Minatojima-minami-machi, Chuo-ku, Kobe, Hyogo, 650-0047, Japan
date: 'Accepted 1988 December 15. Received 1988 December 14; in original form 1988 October 11'
title: Merger Criteria of Multiple Massive Black Holes and the Impact on the Host Galaxy
---
\[firstpage\]
black hole physics — galaxies: nuclei — methods: numerical
Introduction
============
Massive black holes (hereafter MBHs) with the mass of more than $10^6$ solar mass ($M_{\odot}$) have been found in the centres of galaxies. The mass of MBHs is correlated with the properties of the spheroidal components of their host galaxies, with respect to the mass [@Kormendy95; @Magorrian98; @Marconi03], the velocity dispersions [@Ferrarese00; @Tremaine02; @Gultekin09], and the number of globular clusters [@Burkert10; @Harris11]. The origin of MBHs is an open issue of great significance.
In the last decade, quasars (QSOs) that possess $\sim 10^9~M_{\odot}$ MBHs have been found at high redshifts of $z \gtrsim 6$ [e.g., @Fan01], that is, at the cosmic age of $\lesssim
1$ Gyr. Conservatively speaking, the seeds of the MBHs could be stellar mass black holes as massive star remnants. In particular, the remnants of first stars are one of plausible candidates, since first stars are likely to be as massive as a few hundred solar mass [@Abel00; @Nakamura01; @Bromm02; @Yoshida06], several tens solar mass [@Clark11], or about $50M_{\odot}$ [@Hosokawa11]. However, in order for first star remnants to grow up to $\sim 10^9~M_{\odot}$ in $1$ Gyr, the Eddington ratio of mass accretion rate should be larger than unity [e.g., @Umemura01; @Greene12 and references therein]. Super-Eddington accretion is one of possible solutions for the MBH growth [e.g. @Abramowicz88; @Kawaguchi03; @Ohsuga05]. On the other hand, the integration of the QSO luminosity function is concordant with the integrated mass function of MBHs in the local universe, as long as the Eddington ratios are between 0.1 and 1.7 [@Soltan82; @Yu02; @Marconi04]. This implies that supermassive black holes acquire the bulk of mass through gas accretion in the late evolutionary stages and the mass accretion rates are not highly super-Eddington. Also, the gas accretion onto the seeds should be intermittent, and on average could be lower than the Eddington accretion rate [@Milosavljevic09a; @Milosavljevic09b]. If the merger of multiple black holes precedes the growth via gas accretion, the merged MBH can be a seed of a supermassive black hole, and therefore the constraint for the BH growth can be weaker.
In the cold dark matter cosmology, larger galaxies form hierarchically through mergers of smaller galaxies. Hence, many MBHs are assembled in a larger galaxy, if smaller galaxies already possess MBHs. Furthermore, MBHs could be born in hyper-massive star clusters formed by galaxy collisions [@Matsui11]. Thus, galaxy merger remnant can contain many MBHs, even if precursory galaxies have no MBHs.
Observationally, multiple active galactic nucleus (AGN) systems have been discovered recently. They include a triple AGN in the galaxy SDSS J1027+1749 at $z = 0.066$ [@Liu11], three rapidly growing MBHs of $10^6-10^7 M_\odot$ in a clumpy galaxy at $z = 1.35$ [@Schawinski11], a firstly-discovered physical quasar triplet QQQ J1432–0106 within the projected separation of $30-50$kpc at $z =
2.076$ [@Djorgovski07], and a second quasar triplet QQQ J1519+0627 within the projected separation of $200$kpc, which is likely to be harboured in a yet-to-be-formed massive system at $z=1.51$ [@Farina13]. According to the hierarchical merger history, galaxies with many MBHs are likely to form at higher redshifts. Although the galaxy merger proceeds through the violent relaxation, the merger of MBHs has difficulty. As pointed out in @Begelman80, two MBHs in a galaxy are likely to form a binary, but unlikely to merge directly due to the so-called loss cone depletion (the depletion of stars on orbits that intersect the binary MBH). A binary MBH cannot reach sub-parsec separation due to the loss cone depletion, which is called the final parsec problem [e.g. @Merritt04]. A possible way to evade the loss cone depletion is the nonaxisymmetric potential of the host galaxy [@Merritt04; @Berczik06; @Khan11; @Khan12], which is the natural consequence of the galaxy merger. A binary MBHs can merge in the nonaxisymmetric potential in $10$ Gyr or $0.3$ Gyr, when the galaxy contains stars with $10^9M_{\odot}$ or $10^{11}M_{\odot}$, respectively [@Khan11]. However, since this timescale is comparable to or longer than the galactic dynamical timescale, other galaxies harbouring MBHs can intrude before two MBHs merge. This is likely to occur at higher redshifts of $z \gtrsim 6$, at which the universe age is less than $1$ Gyr. If there are more than two MBHs in a galaxy, the dynamical relaxation of MBHs is significantly controlled by the gravity of MBHs themselves, especially by three-body interaction. When a third MBH intrudes into a binary MBH, one of the three MBHs carries away angular momentum from the rest two MBHs, reducing the binary separation, and eventually the binary merges [e.g. @Iwasawa06].
So far, galaxy structures have not been investigated when the galaxies contain more than three MBHs, although they have been investigated in the cases of two MBHs [e.g. @Khan11; @Khan12] and three MBHs [e.g. @Iwasawa08]. [@Tanikawa11] (hereafter, paper I) scrutinised a system of multiple MBHs in a galaxy by high-resolution $N$-body simulations, and found that multiple MBHs produce one dominant MBH through successive mergers. Binary MBHs lose their angular momentum owing to sling-shot mechanism, which induces the decay of the binary orbits through gravitational wave (GW) radiation. In paper I, we investigated one model of a galaxy containing multiple MBHs. In this paper, we explore the evolution of multiple MBHs in galaxies with different 3-dimensional stellar velocity dispersions to derive the criteria of the MBH merger. We also consider the effect of the recoil by anisotropic GW radiation at the MBH merger. Since the recoil velocity typically reaches several hundred km s$^{-1}$ [@Kesden10], it could suppress the MBH growth. Furthermore, we investigate the impact by the MBH merger on the galaxy structure.
The paper is organised as follows. In section \[sec:model\], we describe the simulation model. In section \[sec:results\], we show numerical results. In section \[sec:criteria\], the results are translated to derive the criteria for MBH merger, which are applied for high and low redshift galaxies. In section \[sec:back-reaction\], the back-reaction to a host galaxy is discussed with respect to the galactic structure and the production of high velocity stars. In section \[sec:summary\], we summarise this paper.
Model {#sec:model}
=====
Initial conditions
------------------
We consider a model galaxy that initially contains ten MBHs of equal mass. The effect by the inequality of MBH mass has been explored by several authors [e.g. @Iwasawa11; @Khan12]. In the present simulation, an unequal mass binary forms as a consequence of the MBH merger. The case in which unequal mass MBHs are set up initially will be investigated elsewhere. Stars in a galaxy are treated as superparticles. The number of stars is $N=512k$ ($1k=1024$). The stars are initially distributed according to the Hernquist’s profile, where the mass density distribution is given by $$\rho (r) = \frac{M_{\rm g}}{6 \pi r_{\rm g}^3} \frac{1}{(r/r_{\rm
g})\left[ (r/r_{\rm g}) + 1/3 \right]^3},$$ where $M_{\rm g}$ and $r_{\rm g}$ are respectively the total mass and virial radius of the host galaxy. Here, $r_{\rm g}$ is given by $$r_{\rm g} = \frac{GM_{\rm g}}{2v_{\rm g}^2}, \label{eq:virial}$$ with the gravitational constant $G$ and the 3-dimensional stellar velocity dispersion $v_{\rm g}$.
The mass of MBH is set to be $0.01$ % of the galaxy mass. So, the total mass of ten MBHs is $0.1$ % of the galaxy mass. We realise the distribution of MBHs as follows. The distribution function is supposed to be the same as that of stars within one-third of $r_{\rm g}$ (see Paper I for the dependence on the spread). First, we generate positions and velocities for stars according to the above distributions. Next, we convert ten stars into ten MBHs; we choose randomly ten stars from the stars within one-third of $r_{\rm g}$.
The 3-dimensional velocity dispersion, $v_{\rm g}$, is one of key parameters in the present simulations. Note that the velocity dispersion in this paper is 3-dimensional unless otherwise noted. We consider several galaxy models with different velocity dispersions. The assumed models are shown in Table \[tab:mainresult\]. In models A, the velocity dispersion is $v_{\rm g}=350$ km s$^{-1}$, where A$_{0,1}$, A$_{0,2}$, and A$_{0,3}$ are based on different sets of random numbers. In models B, C, and D, $v_{\rm g}=240$, 180, and 120 km s$^{-1}$, respectively. In models A$_1$, B$_1$, and C$_1$, the recoil velocity, $v_{\rm recoil}$, is added after the MBH merger. Also, for the comparison to models with ten MBHs, we perform simulations for galaxies without MBHs (model BH0) and with two MBHs (model BH2).
In the present simulations, we adopt the standard $N$-body units, where $G=M_{\rm g}=r_{\rm g}=1$. Then, $v_{\rm g}=1/\sqrt{2}$ from Equation (\[eq:virial\]). The speed of light, $c$, is required to be redefined in the present units, since we include Post-Newtonian (PN) corrections (described later). The speed of light changes as $c=6.06 \times 10^2$, $8.84 \times 10^2$, $1.18 \times 10^3$, or $1.77
\times 10^3$ for models A, B, C, or D, respectively.
After we determine the velocity dispersion, $v_{\rm g}$, we still have one free parameter, although $M_{\rm g}/r_{\rm g}$ is fixed for each $v_{\rm g}$ (see Equation (\[eq:virial\])). Setting either of the galaxy mass $M_{\rm g}$ or the galactic virial radius $r_{\rm g}$, we can transform the code units to physical units. When we set $M_{\rm
g}$, we express $r_{\rm g}$ and the dynamical time at $r_{\rm g}$ as follows: $$r_{\rm g} \simeq 1.76 \left( \frac{M_{\rm g}}{10^{11}~M_{\odot}}
\right) \left( \frac{v_{\rm g}}{350~\mbox{km~s$^{-1}$}} \right)^{-2}
\mbox{[kpc]}, \label{eq:rv}$$ and $$t_{\rm dy,g} = \frac{r_{\rm g}}{\sqrt{2}v_{\rm g}} \simeq 3.47
\left( \frac{M_{\rm g}}{10^{11}~M_{\odot}} \right) \left(
\frac{v_{\rm g}}{350~\mbox{km~s$^{-1}$}} \right)^{-3}
\mbox{[Myr]}. \label{eq:tdy}$$ We show $M_{\rm g}$ when $r_{\rm g}=1$ kpc in the rightmost column of Table \[tab:mainresult\], which is derived from Equation (\[eq:rv\]). The average mass density inside $r_{\rm g}$ is given as $$\rho_{\rm g} = \frac{27M_{\rm g}}{64\pi r_{\rm g}^3} \simeq 2.48
\left( \frac{M_{\rm g}}{10^{11} M_{\odot}} \right)^{-2} \left(
\frac{v_{\rm g}}{350~\mbox{km~s$^{-1}$}} \right)^6
\mbox{[$M_\odot~\mbox{pc}^{-3}$]}.$$
Equation of motion
------------------
The equations of motion for field stars and MBHs are respectively given by $$\begin{aligned}
\frac{d^2\bm{r}_{{\rm f},i}}{dt^2} &= \sum_{j \neq i}^{N_{\rm
f}} \bm{a}_{{\rm ff},ij} + \sum_{j}^{N_{\rm B}}
\bm{a}_{{\rm fB},ij}, \\
\frac{d^2\bm{r}_{{\rm B},i}}{dt^2}
&= \sum_{j}^{N_{\rm f}} \bm{a}_{{\rm Bf},ij} + \sum_{j \neq
i}^{N_{\rm B}} \bm{a}_{{\rm BB},ij},\end{aligned}$$ where $\bm{r}_{{\rm f},i}$ and $\bm{r}_{{\rm B},i}$ are the position vectors of $i$-th field star and $i$-th MBH, $N_{\rm f}$ and $N_{\rm B}$ are the numbers of field stars and MBHs, $\bm{a}_{{\rm
ff},ij}$ and $\bm{a}_{{\rm fB},ij}$ are the accelerations by $j$-th field star and $j$-th MBH on $i$-th field star, and $\bm{a}_{{\rm Bf},ij}$ and $\bm{a}_{{\rm BB},ij}$ are the accelerations by $j$-th field star and $j$-th MBH on $i$-th MBH, respectively. Excepting the MBH-MBH interaction, the accelerations are given by Newtonian gravity: $$\begin{aligned}
\bm{a}_{{\rm ff},ij} &= - Gm_{{\rm f},j}\frac{\bm{r}_{{\rm
f},i} - \bm{r}_{{\rm f},j}}{(|\bm{r}_{{\rm f},i} -
\bm{r}_{{\rm f},j}|^2+\epsilon^2)^{3/2}} \\ \bm{a}_{{\rm
fB},ij} &= - Gm_{{\rm B},j}\frac{\bm{r}_{{\rm f},i} -
\bm{r}_{{\rm B},j}}{|\bm{r}_{{\rm f},i} - \bm{r}_{{\rm
B},j}|^3} \\ \bm{a}_{{\rm Bf},ij} &= - Gm_{{\rm
f},j}\frac{\bm{r}_{{\rm B},i} - \bm{r}_{{\rm
f},j}}{|\bm{r}_{{\rm B},i} - \bm{r}_{{\rm f},j}|^3},\end{aligned}$$ where $m_{{\rm f},j}$ and $m_{{\rm B},j}$ are respectively the masses of $j$-th field star and $j$-th MBH, and the softening parameter ($\epsilon=10^{-3}$) is introduced only in star-star interactions.
The acceleration between two MBHs is composed of the Newtonian gravity and PN corrections, such as $$\bm{a}_{{\rm BB},ij} = - Gm_{{\rm B},j}\frac{\bm{r}_{{\rm
B},i}-\bm{r}_{{\rm B},j}}{|\bm{r}_{{\rm
B},i}-\bm{r}_{{\rm B},j}|^3} + \bm{a}_{{\rm
PN},ij}. \label{eq:aBB}$$ We explain the second term below.
Relativistic effects
--------------------
We incorporate the general relativistic effects on the orbits of MBHs, that is, the pericentre shift, GW radiation, and GW recoil. We model the pericentre shift and GW radiation by including the second term ($\bm{a}_{{\rm PN},ij}$) in Equation (\[eq:aBB\]) up to 2.5PN term. The pericentre shift corresponds to 1PN and 2PN terms, and the GW radiation does to 2.5PN term [@Damour81; @Soffel89; @Kupi06]. We employ Equations (1), (2), (3), and (4) in [@Kupi06] for the general relativistic corrections.
Also, we model the GW recoil as follows. At the moment when two MBHs merge, we add recoil velocities to the merged MBHs. Their absolute values are fixed in each simulation. Their direction is determined by the Monte-Carlo method, assuming the isotropic probability. In practice, the absolute values of recoil velocities widely range from several ten km s$^{-1}$ to several thousand km s$^{-1}$, and their directions are determined by the mass ratio and spins of two MBHs [@Campanelli07; @Lousto10]. However, if their spins are aligned before their merger due to relativistic spin precession [@Kesden10], then the recoil velocity decreases to a few $100$ km s$^{-1}$. The recoil velocity $v_{\rm recoil}$ in each simulation is summarised in Table \[tab:mainresult\]. We set the recoil velocity to be equal to or more than $200$ km s$^{-1}$, excepting models without the recoil.
Merger condition
----------------
We assume that two MBHs merge, when the separation between two MBHs is less than ten times the sum of their Schwarzschild radii: $$\left|\bm{r}_{{\rm B},i} - \bm{r}_{{\rm B},j} \right| < 10
\left( r_{{\rm sch},i} + r_{{\rm sch},j} \right),$$ where $r_{{\rm sch},i}$ is the Schwarzschild radius of $i$-th MBH that is $2Gm_{{\rm B},i}/c^2$ for the MBH mass $m_{{\rm B},i}$ with the speed of light $c$ .
Numerical scheme
----------------
Since our numerical scheme is the same as that in paper I, we describe its outline here. We adopt a fourth-order Hermite scheme with individual timestep scheme [@Makino92] for time integration method for an MBH and a star.
For compact binary MBHs, we transform their motions to their relative motion and the centre-of-mass motion. The relative motions are integrated in the same way as single MBHs and stars. In calculating tidal forces on the binary MBHs, we consider the other MBHs and nearby stars as perturbers, and ignore perturbation by the rest of stars. A perturber of a binary MBH is defined as a particle whose distance from the binary MBH is smaller than $200$ times of the semi-major axis of the binary MBH. For the centre-of-mass motion, we adopt Hermite Ahmad-Cohen scheme [@Makino92] for time integration.
We perform $N$-body simulations with the FIRST simulator [@Umemura08] at University of Tsukuba. We use 64 nodes of the FIRST simulator. Each node is equipped with one Blade-GRAPE, which is one of GRAPEs: a special purposed accelerator for a collisional $N$-body system [@Sugimoto90; @Makino03; @Fukushige05]. We compute gravitational forces exerting on a given particle in parallel, which is the so-called $j$-parallel algorithm.
Numerical Results {#sec:results}
=================
Model dependence
----------------
We have calculated a system of ten MBHs in one galaxy during about $140\, t_{\rm dy,g}$, which corresponds to about 1 Gyr in physical units if we adopt $M_{\rm g}=10^{11} M_{\odot}$ and $v_{\rm
g}=240$ km s$^{-1}$. The models and results are summarised in Table \[tab:mainresult\]. The first and second columns indicate the model name and the number of MBHs, respectively. Models A$_{0,1}$, A$_{0,2}$, and A$_{0,3}$ corresponds to models A$_1$, A$_{2}$, and A$_{3}$ in Paper I, respectively. The third column is the stellar velocity dispersion. In the fourth and fifth columns, we show the assumed GW recoil velocity and the ratio of the recoil velocity to the velocity dispersion, respectively. As numerical results, we show the mass of the heaviest MBH ($m_{\rm B,p}$) and the second heaviest MBH ($m_{\rm B,s}$) in the sixth and seventh columns, respectively. Their mass is scaled by the initial MBH mass. If the heaviest or second heaviest MBHs are ejected from the galaxy centre, we attach “(e)” beside their mass. The number of MBHs ejected from the galaxy centre ($N_{\rm B,ej}$) is shown in the eighth column. We define the ejected MBHs to be far by more than $r_{\rm g}$ from the galaxy centre. The galaxy centre is obtained from the density centre of stars, which is calculated in the same way as [@Casertano85].
Fig. \[fig:mmax\] shows the time evolution of the mass of the heaviest MBH in each model. In each of model A, B, and C$_0$, one dominant MBH grows in a galaxy. They are formed through mergers of $4$ – $6$ MBHs. In some of these models, other MBHs become heavier than the initial ones. However, they are ejected from the galaxy through sling-shot mechanism by three MBH interaction. They are not ejected by the GW recoil, since the GW recoil is set to be at most $2v_{\rm
g}$, while the escape velocity of our galaxy models is about $4v_{\rm g}$ (described below in detail). In these models, about a half of MBHs successively merge in $140\, t_{\rm dy,g}$.
![Time evolution of the mass of the heaviest MBH in each model. Models A$_{0,1}$, A$_{0,2}$, and A$_{0,3}$ are brought together as “A$_{0,x}$”. Models A are shown in the top panel, and the other in the bottom panel.[]{data-label="fig:mmax"}](fig01.eps)
On the other hand, in models C$_1$ and D$_0$, only three MBHs merge in $140\, t_{\rm dy,g}$. A heavier MBH might form, if we follow the evolution of the MBHs beyond $140\, t_{\rm dy,g}$. However, we do not follow their evolution, since artificial two-body relaxation may affect the merger for the present number of particles.
Model $N_{B}$ $v_{\rm g}/\mbox{km s}^{-1}$ ${v_{\rm recoil}}/\mbox{km s}^{-1}$ $v_{\rm recoil}/v_{\rm g}$ $m_{\rm B,p}$ $m_{\rm B,s}$ $N_{\rm B,ej}$ $M_{\rm g}/(10^{10}M_{\odot})$ $[r_{\rm g}=\mbox{1kpc}]$
----------- --------- ------------------------------ ------------------------------------- ---------------------------- --------------- --------------- ---------------- ----------------------------------------------------------
A$_{0,1}$ $10$ $350$ $0$ $0$ $4$ $3$(e) $2$ $5.7$
A$_{0,2}$ $10$ $350$ $0$ $0$ $4$ $1$ $3$ $5.7$
A$_{0,3}$ $10$ $350$ $0$ $0$ $6$ $1$ $1$ $5.7$
A$_1$ $10$ $350$ $500$ $1.4$ $6$ $1$ $1$ $5.7$
B$_0$ $10$ $240$ $0$ $0$ $6$ $1$ $1$ $2.7$
B$_1$ $10$ $240$ $500$ $2.1$ $4$ $2$(e) $2$ $2.7$
C$_0$ $10$ $180$ $0$ $0$ $5$ $1$ $1$ $1.5$
C$_1$ $10$ $180$ $200$ $1.1$ $3$ $2$ $1$ $1.5$
D$_0$ $10$ $120$ $0$ $0$ $3$ $1$ $2$ $0.67$
BH0 $0$ $240$ – – – – – $2.7$
BH2 $2$ $240$ – – – – – $2.7$
Merger dynamics
---------------
Here, we see the merger process in detail, using the result of model A$_1$, which includes the GW recoil. The merger process is similar to that in the models without the GW recoil, i.e., models A$_{0,1}$, A$_{0,2}$, and A$_{0,3}$, in which the merger process is shown in Paper I. As seen in the top panel of Fig. \[fig:bbh\], one MBH grows by the merger, and other MBHs do not grow by the terminal time of our simulation. Two MBHs temporarily merge at $t_{\rm dy,g} =54$. However, the merged MBH is swallowed by the heaviest MBH at $t_{\rm
dy,g} = 80$.
The result that only one MBH predominantly grows comes from the following three facts. Two MBHs merge only via a phase of a binary MBH (fact 1). The binary MBH tends to contain the heaviest MBH (fact 2). Furthermore, the binary MBH is unique in the galaxy at any time (fact 3). Fact 1 can be verified in the second top panel of Fig. \[fig:bbh\]. Binary MBHs have semi-major axes of $10^{-5}r_{\rm
g}$ – $10^{-4}r_{\rm g}$ for a long time until they merge. Fact 2 can be seen in the second bottom panel. The heaviest MBH is contained in a binary MBH through most of time (except during $t_{\rm dy,g}=46$ – $54$). This is because a binary MBH often experiences three MBH interactions, through which a heavier MBH is more easily retained in the binary MBH. Fact 3 can be confirmed in the bottom panel. For most of time, the number of binary MBHs in the galaxy is one or zero.
For the mergers of MBHs, the dynamical friction plays a key role. (see also Fig. 2 in Paper I). The dynamical friction by field stars allows MBHs to gather near the galaxy centre. Thus, two MBHs can compose a binary MBH, and subsequently another MBH can intrude the binary MBH. Then, the single MBHs interact with the binary MBH repeatedly and consequently the semi-major axis and eccentricity of the binary MBH are changed owing to the angular momentum loss. Such repeated interactions occur before most of mergers. However, the crucial impact is brought by one strong interaction, and thereby the distance of the binary MBH at the pericentre ($r_{\rm p}$) shrinks significantly, so that the GW radiation works effectively to lose the energy, eventually causing the merger of the binary.
Such merger process is not affected by the GW recoil, if the recoil velocity is of the order of the stellar velocity dispersion, $v_{\rm
g}$. A merged MBH is retained in the inner region of the galaxy, in which the stellar density is high. In this region, the dynamical friction effectively loses angular momenta of the merged MBH. Hence, the merged MBH falls again toward the galactic centre, and form a binary MBH with another single MBH. The binary MBH can interact again a third MBH.
Also, we have found that the secular angular momentum loss of a binary MBH through the Kozai mechanism [@Kozai62] is not effective. The Kozai mechanism can work through eccentricity oscillation, if the semi-major axis ratio is small [@Blaes02; @Berentzen09]. But, in our simulations, the semi-major axis ratio is too large to allow eccentricity oscillation. Instead, the relativistic pericentre shift (1PN and 2PN) is dominant. In fact, if we do not include the 1PN and 2PN terms, the Kozai mechanism works for a binary to merge. The suppression of the Kozai mechanism is also demonstrated in the case of stellar-sized black holes [@Miller02] and in the planetary orbits [@Fabrycky07].
![Time evolution of MBHs in model A$_1$. The top panel shows the masses of the heaviest and second heaviest MBHs at each time. We indicate parameters of the binary MBH with the smallest semi-major axis at each time, i.e. its semi-major axis (second top), the distance at the pericentre (middle), and component masses (second bottom). The bottom panel shows the number of binary MBHs whose semi-major axes are less than $10^{-3}r_{\rm
g}$. Pairs of integers in parentheses in the second top panel show the labels of MBHs composing the binary MBHs, where the heaviest MBH is labelled with “H” and the second heaviest one is “S”. We attached labels only to binary MBHs which are long-lived, or merge eventually. In the second top and middle panels, filled circles indicate the moments when MBHs merge and crosses denote those when binary components are exchanged. []{data-label="fig:bbh"}](fig02.eps)
Criteria for Successive Mergers {#sec:criteria}
===============================
Constraints for galaxy mass and size
------------------------------------
From the above numerical results, we conclude that the lower limit of stellar velocity dispersion is $v_{\rm g} \sim
180~\mbox{km~s$^{-1}$}$, when the GW recoil is $200~\mbox{km~s$^{-1}$}$. Regardless of whether the GW recoil is exerted or not, one dominant MBH grows in model B$_0$ and B$_1$, in both of which the stellar velocity dispersion is more than $240~\mbox{km~s$^{-1}$}$. On the other hand, the growth of one dominant MBH depends on the GW recoil in models C in which galaxies have the stellar velocity dispersion of $180$ km s$^{-1}$. The dependence of the MBH growth on the stellar velocity dispersion can be understood as follows. Using Equation (\[eq:virial\]), we express the ratio of a Schwarzschild radius of an MBH to the virial radius of the galaxy as $$\frac{r_{{\rm sch},i}}{r_{\rm g}} = 4 \left( \frac{m_{{\rm
B},i}}{M_{\rm g}} \right) \left( \frac{v_{\rm g}}{c}
\right)^2. \label{eq:rschoverrv}$$ This means that the MBH horizon size is smaller compared to the galaxy size if the stellar velocity dispersion is smaller. Therefore, a larger amount of angular momenta should be extracted for a binary MBH to merger, so that the resultant largest MBH becomes less massive.
Here, we estimate the constraints for galaxy mass and size to allow the MBH merger. In Fig. \[fig:lmt\], we show the mass and size of galaxies in which MBHs successively merge during $140 t_{\rm
dy,g}$. They have the stellar velocity dispersion of more than $180$ km s$^{-1}$. Using Equation (\[eq:rv\]), we relate their masses to their sizes as $$\left( \frac{M_{\rm g}}{10^{10}M_{\odot}} \right) \gtrsim
1.5 \left( \frac{r_{\rm g}}{1\mbox{kpc}}
\right). \label{eq:vgcnstrn}$$ Such regions are above the solid lines in the top panels of Fig. \[fig:lmt\].
We also impose the conditions on which the successive mergers of MBHs occurs within $1$ Gyr or within $10$ Gyr. If the merger timescale is $140 t_{\rm dy,g}$, the galaxies should have their dynamical time of less than $7$ Myr for $1$ Gyr case or $70$ Myr for $10$ Gyr case. Using Equation (\[eq:rv\]) and (\[eq:tdy\]), we obtain $$\left( \frac{M_{\rm g}}{10^9 M_{\odot}} \right) = 1.1 \left(
\frac{t_{\rm dy,g}}{14 \mbox{Myr}} \right)^{-2} \left( \frac{r_{\rm
g}}{1 \mbox{kpc}} \right)^{3}.$$ Therefore, we can write the relation between their masses and sizes: $$\left( \frac{M_{\rm g}}{10^9~M_{\odot}} \right) > \left\{
\begin{array}{lc}
\displaystyle 4.4 \left( \frac{r_{\rm g}}{1\mbox{kpc}} \right)^3 &
\cdots \mbox{$1$~Gyr case}\\ \displaystyle 0.044 \left(
\frac{r_{\rm g}}{1\mbox{kpc}} \right)^3 & \cdots \mbox{$10$~Gyr
case}
\end{array}
\right.
\label{eq:tdycnstrn_hz}$$ These regions are upper sides of the dashed lines in the top panels of Fig. \[fig:lmt\]. As a result, the shaded regions in the top panels of Fig. \[fig:lmt\] are the allowed regions for the masses and sizes of galaxies in which MBHs can successively merge.
These constraints can be translated into those for the mass ($M_{\rm
h}$) and size ($r_{\rm h}$) of a dark matter halo, using a simplified model. A dark matter halo is assumed to be six times more massive than that of a galactic stellar component, according to the ratio of dark matter to baryon in the universe [@Komatsu11]. However, stellar components are more concentrated than the dark matter components due to cooling when stars are formed. Hence, the dark matter halo does not seem to make a significant effect on the merger dynamics. Actually, dark matter mass at the central region is much less than or at most comparable to stellar mass [e.g. @Forman85; @Saglia92]. In the present analysis, the stellar velocity dispersion is assumed to be twice of the velocity dispersion in the dark matter halo. This is justified by the difference between observed velocity dispersions at effective radii and those at several effective radii in elliptical galaxies [@Coccato09]. Then, the size of the dark matter halo is $24$ times larger than that of the stellar component from virial theorem. Using these relations, we can obtain the regions of the masses and sizes of dark matter haloes in which MBHs successively merge, which are the shaded regions in the bottom panels of Fig. \[fig:lmt\].
The formation epoch (redshift) can be assessed depending on the masses and sizes of dark matter haloes, in the same way as [@Mo98]. We equate the size of a dark matter halo ($r_{\rm h}$) to the radius inside which the mean mass density is $200$ times the critical density at a given redshift $z$, and then derive the mass of a dark matter halo ($M_{\rm h}$) inside $r_{\rm h}$. The virial mass and radius are related as $$M_{\rm h} = 100 G^{-1} H(z)^2 r_{\rm h}^3. \label{eq:Mhalo0}$$ We can rewrite Equation (\[eq:Mhalo0\]) as $$\left( \frac{M_{\rm h}}{10^{11}~M_{\odot}} \right) = 12 \left[
\frac{H(z)}{H(10)} \right]^2 \left( \frac{r_{\rm h}}{30\mbox{kpc}}
\right)^3. \label{eq:Mhalo1}$$ The function $H(z)$ is expressed as $$H(z) = H_0 \bigl[ \Omega_{\rm \Lambda} + (1 - \Omega_{\rm \Lambda} -
\Omega_{\rm m}) (1+z)^2 + \Omega_{\rm m} (1+z)^3 \bigr]^{1/2},$$ where $H_0$ is the Hubble constant, and $\Omega_{\rm \Lambda}$ and $\Omega_{\rm m}$ are the lambda parameter and the matter density parameter, respectively. In Equation (\[eq:Mhalo1\]), we adopt $(\Omega_{\rm \Lambda}, \Omega_{\rm m}) = (0.7, 0.3)$, and hereafter we also adopt these values and $h=0.7$, where the Hubble constant is $H_0 = 100h$ km s$^{-1}$ Mpc$^{-1}$. From these equations, we can draw the relation between masses and sizes of dark matter haloes at formation redshifts, which is shown by dashed-dotted lines in the bottom panels of Fig. \[fig:lmt\].
From the bottom panels of Fig. \[fig:lmt\], we can estimate the minimum mass of a dark matter halo which allows MBH successive mergers at a given redshift. In order for the merger to occur during $1$ Gyr, dark matter haloes formed at redshift $z=7$ should have more than $4
\times 10^{10} M_{\odot}$, which corresponds to the stellar component mass of about $6.7 \times 10^{9} M_{\odot}$. In the case of mergers during $10$ Gyr, dark matter haloes formed at redshift $z=3$ should have more than $7 \times 10^{10} M_{\odot}$.
If there were only one binary MBH in a nonaxisymmetric galactic potential, the timescale for the merger is about $10$ Gyr or $0.3$ Gyr respectively for the stellar component of $10^9M_{\odot}$ or $10^{11}M_{\odot}$ [@Khan11]. Therefore, at redshifts of $z
\gtrsim 7$, when the cosmic age is less than $1$ Gyr, another MBH may intrude before a binary MBH merges. Our results show that even if multiple MBHs exist in a galaxy with $4 \times 10^{10} M_{\odot}$ at redshift $z=7$, the MBHs can successively merge.
Merger criteria for galactic luminosity
---------------------------------------
In the above, we have derived the constraints for the masses and sizes of galaxies and their parent dark haloes, in which the successive merger of MBHs can occur. Here, we compare the luminosity function based on the Press-Schechter formalism. The luminosity function of galaxies is obtained as follows. We can give an ultraviolet (UV) magnitude of a galaxy embedded in a halo with mass $M_{\rm h}$ as $$M_{\rm UV} = M_{{\rm UV},\odot} - \frac{2.5}{\log{10}} \log \left[
\left( \frac{M_{\rm h}}{M_{\odot}} \right) \left(
\frac{\Omega_{\rm b}}{\Omega_{\rm m}} \right) \Upsilon_{{\rm
UV},\odot}^{-1} \right],$$ where $M_{{\rm UV},\odot}$ is UV magnitude of the Sun, $\Upsilon_{{\rm
UV},\odot}$ is the mass-to-UV luminosity ratios scaled by that of the Sun. We set $M_{{\rm UV},\odot}=5.6$. Here, we assume that a galaxy mass $M_{\rm g}$ is equal to $(\Omega_{\rm m}/\Omega_{\rm
b})^{-1}M_{\rm h}$, and that $\Omega_{\rm m}/\Omega_{\rm b}=6$. We denote the number density of haloes by $n$. Note that $n$ can be regarded as the number density of galaxies, since we assume a halo has one galaxy. Then, Press-Schechter mass function of dark matter haloes (shown in the top left panel of Fig. \[fig:psmf\]) can be transformed into the luminosity function at UV band as $$\begin{aligned}
\frac{dn}{dM_{\rm UV}} &= - \frac{d\left(M_{\rm
h}/M_{\odot}\right)}{dM_{\rm UV}} \frac{dn}{d(M_{\rm
h}/M_{\odot})} \\ &= \frac{\log{10}}{2.5} \left( \frac{M_{\rm
h}}{M_{\odot}} \right) \frac{dn}{d(M_{\rm
h}/M_{\odot})}.\end{aligned}$$
Observationally, the mass-to-luminosity ratios for high-redshift Ly $\alpha$ emitters (LAEs), $\Upsilon_{{\rm UV},\odot}$, can range from 0.3 to 10 [@Fernandez08]. For low-redshift galaxies, the mass-to-luminosity ratios range from 2 to 10 in normal galaxies, but reach $\sim 100$ in dwarf galaxies [@Hirashita98; @Strigari08]. Considering the observed mass-to-luminosity ratios, we draw the UV luminosity function $dn/d(M_{\rm UV}/M_{\odot})$ in Fig. \[fig:psmf\] for the cases of redshift $z=7$ (top right), $3$ (bottom left), and $0$ (bottom right). In each panel, we show the UV luminosity function $dn/d(M_{\rm UV}/M_{\odot})$ with different $\Upsilon_{{\rm UV},\odot}$. The values of $\Upsilon_{{\rm UV},\odot}$ are indicated by numbers beside the curves. The critical luminosity of galaxies for the successive mergers is shown by vertical dashed lines attached with each curve. The successive mergers happen in galaxies brighter than the critical luminosities.
Observed luminosity functions of high-redshift LAEs [@Ouchi09] and those of Lyman break galaxies (LBGs) [@Jiang11] seem to match well the curves with $\Upsilon_{{\rm UV},\odot} \sim 1$. Thus, the successive MBH merger is expected for LAEs or LBGs brighter than $M_{\rm UV}\simeq - 19$ (see the top right and bottom left panels of Fig. \[fig:psmf\]). Since $\Upsilon_{{\rm UV},\odot} = 2$ – $10$ in low-redshift galaxies, the successive merger is expected for low-redshift galaxies brighter than $M_{\rm UV}\simeq - 18$ (see the bottom right panel of Fig. \[fig:psmf\]). Note that if $\Upsilon_{{\rm UV},\odot}$ is as large as 100 in dwarf galaxies, the critical UV magnitude can be overestimated by about a factor of two. These dwarf galaxies would lose an amount of baryons through their evolution. Although we adopt $\Omega_{\rm m}/\Omega_{\rm b}=6$ for all galaxies, $\Omega_{\rm m}/\Omega_{\rm b}$ should be set to a larger value for these dwarf galaxies. If we do so, the curve of $\Upsilon_{{\rm UV},\odot}=100$ in the bottom right panel of Fig. \[fig:psmf\] will shift leftward.
Back-reaction to a Host Galaxy {#sec:back-reaction}
==============================
Galaxy structure
----------------
Hereafter, we focus on the simulation results of model B$_0$. If necessary, we can compare a simulation including the GW recoil, model B$_1$. Fig. \[fig:evl\_rho\] shows the evolution of mass density profile of stars (the top panel). The mass density inside $r/r_{\rm g}
= 0.05$ decreases gradually. This is because MBHs give their kinetic energy to stars as a back reaction of dynamical friction and sling-shot mechanism. Until $t/t_{\rm dy,g}=80$, six MBHs merge. We can see in the top panel of Fig. \[fig:evl\_rho\] that the mass density profile is roughly proportional to $r^{-0.5}$ in the range from $r/r_{\rm g}=5\times 10^{-3}$ to $r/r_{\rm g}=0.05$. Such a density slope is consistent with those in a galaxy with two MBHs and three MBHs [@Nakano99; @Iwasawa08]. In Fig. \[fig:evl\_mr\], we see an enclosed mass of the galaxy within $r/r_{\rm g}=0.05$ (vertical dashed line) is $10^{-2}M_{\rm g}$, which is ten times higher than the total mass of MBHs. Hence, the present simulation shows that MBHs can affect the galactic structure of the central regions that include about ten times the total mass of MBHs.
![Mass density profile of stars in a galaxy with ten MBHs (top) or without MBHs (bottom). In both panels, dashed-dotted, dotted, dashed, and solid curves indicate the mass density at $t/t_{\rm dy,g}=0$, $10$, $45$, and $80$, respectively. A solid line in the top panel shows the relation of $\rho \propto
r^{-0.5}$.[]{data-label="fig:evl_rho"}](fig05.eps)
![Mass of stars within each radius in model B$_0$ with ten MBHs. Dashed-dotted, dotted, dashed, and solid curves indicate the mass at the time $t/t_{\rm dy,g}=0, 10, 45$, and $80$. The vertical dashed line indicates $r/r_{\rm g}=0.05$.[]{data-label="fig:evl_mr"}](fig06.eps)
We also compare the structure of the galaxy containing ten MBHs to that containing two MBHs. The total masses of MBHs are the same, that is, $0.1$ % of the galaxy mass in both of the models. The top panel of Fig. \[fig:rho\] shows the mass density profile of the galaxy with ten MBHs at $t/t_{\rm dy,g}=79, 80$ and $81$. The mass density is not fluctuated on the dynamical timescale in the range from $r/r_{\rm
g}=0.01$ to $0.05$. In the middle panel of Fig. \[fig:rho\], we show the mass density profile of the galaxy with two MBHs at $t/t_{\rm
dy,g}=30, 40$, and $50$. During $20$ dynamical time, the mass density profile is not changed in the case of the galaxy with two MBHs. We expect that the mass density profile is never changed after $t/t_{\rm dy,g}=50$.
In the bottom panel of Fig. \[fig:rho\], we compare the mass density profile of the galaxy containing ten MBHs at $t/t_{\rm dy,g}=80$ with those containing two MBHs at $t/t_{\rm dy,g}=50$. Both of the mass density slopes are proportional to $r^{-0.5}$. However, the mass density of the galaxy with ten MBHs is lower by a factor of $1.5$ than that with two MBHs. This difference results from the sling-shot mechanism in the galaxy with ten MBHs. Owing to the sling-shot mechanism among three MBHs, MBHs receive kinetic energy, and transfer their kinetic energy to stars through the dynamical friction. Such picture is consistent with a galaxy with three MBHs [@Iwasawa08].
![Mass density profile of a galaxy with ten MBHs (top), that with two MBHs (middle), and both (bottom). In all the panels, the dotted curve shows the profile at the initial time. In the top panel, the solid curves indicate the profile at the time $t/t_{\rm
dy,g}=79, 80$, and $81$. In the middle panel, the dashed curves indicate the profile at the time $t/t_{\rm dy,g}=30, 40$, and $50$. In the bottom panel, the solid and dashed curves are the profiles of a galaxy with ten MBHs at $t/t_{\rm dy,g}=80$, and a galaxy with two MBHs at $t/t_{\rm dy,g}=50$, respectively.[]{data-label="fig:rho"}](fig07.eps)
Here, we verify that the central density of a galaxy in model B$_0$ are decreased by MBH dynamics, not by artificial two-body relaxation. The bottom panel of Fig. \[fig:evl\_rho\] shows the evolution of mass density profile of a galaxy without any MBH, in which only two-body relaxation decreases the central mass density of the galaxy. Comparing mass densities in the top and bottom panels of Fig. \[fig:evl\_rho\], we can see that the central density in model B$_0$ is decreased much more rapidly than that in model BH0. Therefore, the central density of a galaxy in model B$_0$ are dominantly decreased by MBH dynamics.
High-velocity stars
-------------------
We investigate stars which are ejected from a galaxy with high speed. Such stars are generated through the sling-shot mechanism induced by a binary MBH. We focus on the simulation results of model B$_0$. If necessary, we can compare a simulation including the GW recoil, model B$_1$, in which MBHs also successively merge.
We compare the velocity distributions of stars as a function of $\theta$ at the time $t/t_{\rm dy,g}=0$ and $80$ in the cases of models with and without MBHs. We illustrate the relation of $\theta$ to the position and velocity vectors in Fig. \[fig:theta\]. Then, the $\theta$ is expressed as $$\theta = \cos^{-1}\left(\frac{\bm{r}_{\rm f} \cdot
\bm{v}_{\rm f}}{r_{\rm f} v_{\rm f}}\right),$$ where $\bm{r}_{\rm f}$ and $\bm{v}_{\rm f}$ are respectively the position and velocity vectors of a star, and $r_{\rm
f}=|\bm{r}_{\rm f}|$ and $v_{\rm f}=|\bm{v}_{\rm f}|$. The origin of the position vector is set to the galaxy centre.
We define high-velocity stars as stars whose velocities are more than $2\sqrt{2}v_{\rm g}$. Fig. \[fig:ang\_hvs\] shows the resultant velocity distributions of stars. The presence of high-velocity stars is an outstanding feature of model B$_0$ at the time $t_{\rm dy,g}=80$ (the second top panel). We also find such high-velocity stars in a galaxy with two MBHs, model BH2 (the second bottom panel). We can see that some stars have velocities higher than $v_{\rm f}/v_{\rm
g}=10$. Furthermore, they have extremely radial orbits around $\theta=0$. This is because they are generated at the galactic centre through the sling-shot mechanism by a binary MBH, and directly go away outside the galaxy. Note that no high-velocity star is generated in a galaxy without MBHs, model BH0 (see the bottom panel of Fig. \[fig:ang\_hvs\]).
![Illustration of the position vector ($\bm{r}_{\rm f}$), velocity vector ($\bm{v}_{\rm f}$) of a star, and $\theta$.[]{data-label="fig:theta"}](fig08.eps)
![Velocity distributions of stars as a function of $\theta$ in models B$_0$, B$_1$, BH2, and BH0 at the initial time (top), in model B$_0$ at $t/t_{\rm dy,g}=80$ (second top), in model B$_1$ at $t/t_{\rm dy,g}=75$ (middle), in model BH2 at $t/t_{\rm dy,g}=65$ (second bottom), and in model BH0 at $t/t_{\rm dy,g}=80$ (bottom).[]{data-label="fig:ang_hvs"}](fig09.eps)
We investigate the difference between properties of high-velocity stars in the cases of galaxies with ten MBHs and with two MBHs. $37$ high-velocity stars have been generated at $t/t_{\rm dy,g}=80$ in model B$_0$, in contrast to $188$ high-velocity stars at $t/t_{\rm
dy,g}=50$ in model BH2. The generation rate of high-velocity stars in model B$_0$ is ten times lower than that in model BH2. This is because a galaxy with ten MBHs does not always have a binary MBH with small semi-major axis, while a binary MBH stays in the central region of the galaxy in model BH2 (see the top panel of Fig. \[fig:cfunc\_hvs\]). It is difficult for a binary MBH to produce high-velocity stars, unless its semi-major axis is as small as $\sim
10^{-5} r_{\rm g}$. This is readily estimated with numerical results by [@Quinlan96]. A binary MBH typically gives a kick velocity of the order of $v_{\rm kick} = \sqrt{2.5G\mu/a}$ to a field star through the sling-shot mechanism, where $\mu$ and $a$ are the reduced mass and semi-major axis of a binary MBH. Supposing that a binary MBH consists of two MBHs with the initial mass, a binary with the semi-major axis of $a<3.3 \times 10^{-5} r_{\rm g}$ can produce high-velocity stars with $v_{\rm kick}>2\sqrt{2}v_{\rm g}$.
Another possible reason is that more stars interact with a binary MBH in model BH2, since the total mass of the binary MBH in model BH2 is larger than that in model B$_0$.
The total number of high-velocity stars increases in different ways between models B$_0$ and BH2. As seen in the bottom panel of Fig. \[fig:cfunc\_hvs\], the number of high-velocity stars increases at a roughly constant rate in model BH2 from the time $t/t_{\rm
dy,g}=10$ to $50$. On the other hand, the generation rate of high-velocity stars is largely changed in model B$_0$ from the time $t/t_{\rm dy,g}=0$ to $80$ (see the solid curve in the bottom panel of Fig. \[fig:cfunc\_hvs\]). The generation rate is low during the time $t/t_{\rm dy,g}=40$ – $60$, and during $t/t_{\rm dy,g}=70$ – $90$, while it is high during $t/t_{\rm dy,g}=10$ – $40$ and during $t/t_{\rm dy,g}=60$ – $70$. This feature is similar to high-velocity stars in model B$_1$.
In a galaxy with ten MBHs, such as models B$_0$ and B$_1$, high-velocity stars are generated intermittently because of the occasional absence of a binary MBH whose semi-major axis is favourable to eject stars, $\sim 10^{-5} r_{\rm g}$. This can be verified in models B$_0$ and BH2. During $t_{\rm dy,g}=10$ – $40$, and $60$ – $70$, high-velocity stars are generated at a high rate in model B$_0$ (see the bottom panel of Fig. \[fig:cfunc\_hvs\]). At this time, there is a binary MBH with semi-major axis of about $10^{-5}r_{\rm g}$ (see the top panel of Fig. \[fig:cfunc\_hvs\]). The generation rate of high-velocity stars is low during $t_{\rm dy,g}=10$ – $20$, $40$ – $60$, and $70$ – $90$. Except $t_{\rm dy,g}=70$ – $80$, there is a binary MBH with semi-major axis much larger than $10^{-5} r_{\rm
g}$, or no binary MBH. Therefore, stars are not ejected through sling-shot mechanism. During $t_{\rm dy,g}=70$ – $80$, there is a binary MBH with semi-major axis much less than $10^{-5} r_{\rm
g}$. Such a binary MBH cannot interact with stars due to small cross section. On the other hand, there is a binary MBH with semi-major axis $\sim 10^{-5} r_{\rm g}$ in model BH2.
![Time evolution of the minimum semi-major axis of a binary MBH at each time (top), and the number of high-velocity stars scaled by its final number (bottom). High-velocity stars are defined as those whose velocities are more than $2\sqrt{2}v_{\rm
g}$.[]{data-label="fig:cfunc_hvs"}](fig10.eps)
The feature of the intermittent generation rate can be a useful probe to constrain the formation mechanism of a single merged MBH, or a binary MBH at the galaxy centre at the present time.
Summary {#sec:summary}
=======
We have performed $N$-body simulations to investigate successive mergers of MBHs in galaxies with different masses and radii. We have found that about a half of multiple MBHs successively merge to one bigger MBH within $140t_{\rm dy,g}$ in galaxies with the velocity dispersion larger than $\sim 180$ km s$^{-1}$. The merger of MBHs is promoted, such that the loss cone of binary MBHs is refilled by MBHs losing their angular momenta due to dynamical friction. GW recoil does not affect the merger process, if the recoil velocity is of the order of the stellar velocity dispersion. Galaxies which allow multiple MBHs to merge should reside in dark matter haloes with the mass more than $4 \times 10^{10}M_{\odot}$, if these dark matter haloes form at high redshifts. These galaxies could correspond to LAEs or LBGs brighter than the UV magnitude $M_{\rm UV} \simeq -19$ at high redshifts. On the other hand, an MBH which has experienced the successive merger can inhabit low-redshift galaxies brighter than $M_{\rm UV} \simeq -18$.
We have also investigated the evolution of the galactic structure and the generation of high-velocity stars as the back-reaction by the successive merger of MBHs. We have found that the dynamics of MBHs affects the central regions of galaxy that contain about ten times the total mass of MBHs. The mass density profile is transformed to $\rho
\propto r^{-0.5}$, which is the same as the mass density profile in the case of a galaxy with two and three MBHs. The mass density in the central regions is $1.5$ times smaller than in the case of the galaxy with two MBHs. In a galaxy with ten MBHs, high-velocity stars are generated intermittently, while they are generated at a constant rate in the case of a galaxy with two MBHs. Such features should enable us to constrain the merger mechanism of MBHs in a galaxy.
Acknowledgements {#acknowledgements .unnumbered}
================
We thank Kohji Yoshikawa for fruitful advice on cosmological arguments, Masaki Iwasawa for useful comments about our simulation method, and Yuichi Matsuda for stimulating discussion. Numerical simulations have been performed with computational facilities at the Center for Computational Sciences in the University of Tsukuba. This work was supported in part by the FIRST project based on the Grants-in-Aid for Specially Promoted Research by MEXT (16002003), and Grant-in-Aid for Scientific Research (S) by JSPS (20224002).
[99]{} Abel T., Bryan G. L., Norman M. L., 2000, [ApJ]{}, 540, 39 Abramowicz M. A., Czerny B., Lasota J. P., Szuszkiewicz E., 1988, [ApJ]{}, 332, 646 Begelman M. C., Blandford R. D., Rees M. J., 1980, [Nature]{}, 287, 307 Berczik P., Merritt D., Spurzem R., Bischof H., 2006, [ApJ]{}, 642, 21 Berentzen I., Preto M., Berczik P., Merritt D., Spurzem R., 2009, [ApJ]{}695, 455 Blaes O., Lee M. H., Socrates A., 2002, [ApJ]{}, 578, 775 Bromm V., Coppi P. S., Larson R. B., 2002, [ApJ]{}, 564, 23 Burkert A., Tremaine S., 2010, [ApJ]{}, 720, 516 Campanelli M., Lousto C., Zlochower Y., Merritt D., 2007, [ApJ]{}, 659, 5 Casertano S., Hut P., 1985, [ApJ]{}, 298, 80 Clark P. C., Glover S. C. O., Klessen R. S., Bromm V. 2011, [ApJ]{}, 727, 110 Coccato L. et al., 2009, [MNRAS]{}, 394, 1249 Damour T., Dervelle N., 1981, Physics letters A, 87, 81 Djorgovski S. G., Courbin F., Meylan G., Sluse D., Thompson D., Mahabal A., Glikman E., 2007, [ApJ]{}, 662, L1 Fabrycky D., Tremaine S., 2007, [ApJ]{}, 669, 1298 Fan X. et al., 2001, [AJ]{}, 122, 2833 Farina E. P., Montuori C., Decarli R., Fumagalli M., 2013, [MNRAS]{}, 431, 1019 Fernandez E. R., Komatsu E., 2008, [MNRAS]{}, 384, 1363 Ferrarese L., Merritt D., 2000, [ApJ]{}, 539, L9 Forman W., Jones C., Tucker W., 1985, [ApJ]{}, 293, 102 Fukushige T., Makino J., Kawai A., 2005, [PASJ]{}, 57, 1009 Greene J. E., 2012, [Nature]{} Communications, 3, 1304 G[ü]{}ltekin K., Richstone D. O. et al., 2009, [ApJ]{}, 698, 198 Harris G. L. H., Harris W. E., 2011, [MNRAS]{}, 410, 2347 Hirashita H., Takeuchi T. T., Tamura, N., 1998, [ApJ]{}, 504, L83 Hosokawa T., Omukai K., Yoshida N., Yorke H., 2011, [Science]{}, 334, 1250 Iwasawa M., Funato Y., Makino J., 2006, [ApJ]{}, 651, 1059 Iwasawa M., Funato Y., Makino J., 2008, preprint (arXiv:0801.0859) Iwasawa M., An S., Matsubayashi T., Funato Y., Makino J., 2011, [ApJ]{}, 731, L9 Jiang L. et al., 2011, [ApJ]{}, 743, 65 Kawaguchi T., 2003, [ApJ]{}, 593, 69 Kesden M., Sperhake U., Berti E., 2010, [ApJ]{}, 715, 1006 Khan F. M., Just A., Merritt D., 2011, [ApJ]{}, 732, 89 Khan F. M., Preto M., Berczik P., Berentzen I., Just A., Spurzem R., 2012, [ApJ]{}, 749, 147 Komatsu E. et al., 2011, [ApJS]{}, 192, 18 Kormendy J., Richstone D., 1995, [ARA&A]{}, 33, 581 Kozai Y., 1962, [AJ]{}, 67, 591 Kupi G., Amaro-Seoane P., Spurzem R., 2006, [MNRAS]{}, 371, 45 Liu X., Shen Y., Strauss M. A., 2011, [ApJ]{}, 736, L7 Lousto C. O., Campanelli M., Yosef Z., Nakano H., 2010, Classical and Quantum Gravity, 27, 114006 Makino J. Aarseth S., 1992, [PASJ]{}, 44, 141 Makino J., Fukushige T., Koga M., Namura K., 2003, [PASJ]{}, 55, 1163 Magorrian J. et al. 1998, [ApJ]{}, 115, 2285 Marconi A., Hunt L. K., 2003, [ApJ]{}, 589, L21 Marconi A. et al., 2004, [MNRAS]{}, 351, 169 Matsui H. et al., 2012, [ApJ]{}, 746, 26 Merritt D., Poon M. Y., 2004, [ApJ]{}, 606, 788 Miller M. C., Hamilton D. P., 2002, [ApJ]{}, 576, 894 Milosavljevic M, Couch S. M., Bromm V., 2009, [ApJ]{}, 696, 146 (Milosavljevic et al. 2009a) Milosavljevic M, Bromm V., Couch S. M., Oh S. P., 2009, [ApJ]{}, 698, 766 (Milosavljevic et al. 2009b) Mo H. J., Mao S., White S. D. M., 1998, [MNRAS]{}, 295, 319 Nakamura F., Umemura M., 2001, [ApJ]{}, 548, 19 Nakano T., Makino J., 1999, [ApJ]{}, 510, 155 Ohsuga K., Mori M., Nakamoto T., Mineshige S., 2005, [ApJ]{}, 628, 368 Ouchi M. et al., 2009, [ApJ]{}, 706, 1136 Quinlan G. D., 1996, [NewA]{}, 1, 35 Saglia R. P., Bertin G., 1992, [ApJ]{}, 384, 433 Schawinski K., Urry M., Treister E., Simmons B., Natarajan P., Glikman E., 2011, [ApJ]{}, 743, L37 Soffel M. H., 1989, Relativity in Astrometry, Celestial Mechanics and Geodesy. Berlin Springer Soltan A., 1982, [MNRAS]{}, 200, 1155 Strigari L. E. et al., 2008, [Nature]{}, 454, 1096 Sugimoto D., Chikada Y., Makino J., Ito T., Ebisuzaki T., Umemura M., 1990, [Nature]{}, 345, 33 Tanikawa A., Umemura M., 2011, [ApJ]{}, 728, L31 (paper I) Tremaine S. et al. 2002, [ApJ]{}, 574, 740 Umemura M., 2001, [ApJ]{}, 560, L29 Umemura M., Susa H., Suwa T., Sato D., eds. Proc. AIP. Conf. 990, First Stars III, Melville, NY:AIP, p. 386 Yoshida N., Omukai K., Hernquist L., Abel T., 2006, [ApJ]{}, 652, 6 Yu Q., Tremaine S. 2002, [MNRAS]{}, 335, 965
\[lastpage\]
[^1]: E-mail: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In this paper, we consider compressive sensing (CS)-based recovery of delays and Doppler frequencies of targets in high resolution radars. We propose a novel sub-Nyquist sampling method in the Fourier domain based on difference sets (DS), called DS-sampling, to create dictionaries with highly incoherent atoms. The coherence of the dictionary reaches the Welch minimum bound if the DS-sampling is employed. This property let us to implement sub-Nyquist high resolution radars with minimum number of samples[^1]. We also develop a low complexity recovery method, based on structured CS and propose a new waveform, called difference set–frequency coded modulated (DS-FCM) waveform, to boost the recovery performance of the sub-Nyquist radar in noisy environments. The proposed method solves some of the common problems in many CS-based radars and overcome disadvantages of the conventional Nyquist processing, *i.e.* matched filtering in high resolution radar systems. The proposed method allows us to design sub-Nyquist radars, which require less than $2\%$ of Nyquist samples and recover targets without resolution degradation in comparison to the conventional Nyquist processing.'
author:
- 'Iman Taghavi, Mohamad F. Sabahi, Farzad Parvaresh and Mohsen Mivehchy [^2]'
bibliography:
- 'ref.bib'
title: High Resolution Compressed Sensing Radar using Difference Set Codes
---
Compressive sensing (CS), difference set (DS), Welch bound, coherence, structured CS, Doppler focusing, high resolution radar.
Introduction {#sec:Introduction}
============
Matched filtering and fast Fourier transform (FFT) are the two conventional processing techniques used in monostatic pulse-Doppler radars, which are employed to estimate delays and Doppler frequencies of targets, respectively. In the conventional processing, the delay and Doppler resolution of a radar relatively depend on the bandwidth of the radar waveform and the number of pulses that the radar transmits and receives from the targets [@Skolnik2003]. Hence, high resolution radar systems use waveforms with large bandwidth, and according to the Shannon-Nyquist theorem [@Jerri1977], these radars require to sample the received signals in the radars receivers at high rates. Another disadvantage of matched filtering in the radars receivers is producing relatively large side-lobes for stronger targets, which may cover nearby weaker ones [@Skolnik2003]. In this paper, we propose a novel sub-Nyquist sampling scheme for the radar receiver, which relaxes the dependency of the sampling rate in the radar receiver to the radar waveform bandwidth and consequently the radar resolution.
Compressed sensing or compressive sampling (CS), which rigorously was firstly studied by Donoho [@Donoho2006a] and Candes [*et. al.*]{} [@candes2006compressive], is an emerging field that has attracted considerable amount of research over the past decade. One of the main goals of CS is sampling wideband signals with *known properties*, such as sparsity, at rates significantly lower than the Shannon-Nyquist rate without losing the information that is required to reconstruct the signals from the samples. In practice, sampling a signal at a sub-Nyquist rate means that a lower rate analog to digital converters (ADCs) is required, which leads to less power consumption, heat dissipation and cost in the receiver circuit. Moreover, sampling at sub-Nyquist rate requires less memory for data processing [@Eldar2012; @duarte2011structured]. Several previous works study CS-based radar systems. The sparse estimation methods in [@Ender2010a] and [@Herman2009] require a huge dictionary-size proportional to the product of the delay and Doppler grid sizes. The large dictionary-size in CS-based radar systems increases the computational complexity of the recovery algorithms and makes them infeasible for real-time target detection in CS-based radars. In [@Demissie2014], the authors propose a CS recovery algorithm for high resolution estimation of delays and Doppler frequencies, however, this work does not address the sample rate reduction. The authors of [@Bajwa2011] have used recovery algorithms based on spectral estimation tools to recover delays and Doppler frequencies directly from the low rate samples. Finally, the recently proposed approach in [@Bar-Ilan2014], called Doppler focusing, uses a low complexity and robust algorithm to estimate delays and Doppler frequencies of targets. It is worth mentioning that the sub-Nyquist rate sampling schemes used in [@Bajwa2011] and [@Bar-Ilan2014] degrade the delay resolution of radar compared to the conventional Nyquist processing radars.
In this paper, we focus on designing an optimal CS-dictionary and finding a low complexity recovery method for detecting targets in high resolution radars. We also introduce an appropriate radar waveform for this recovery algorithm.
Our contributions are as follows:
- A low complexity structured CS model and an efficient recovery algorithm for a high resolution radar with sub-Nyquist sampling is introduced.
- An optimal sub-Nyquist sampling technique based on difference sets called DS-sampling is proposed, which guarantees high resolution recovery of radar targets using CS algorithms.
- A frequency coded modulated waveform based on difference sets called DS-FCM waveform is presented, which improves the recovery performance of the CS-based radar significantly in noisy conditions.
- A modified version of the Doppler focusing (modified-DF) approach [@Bar-Ilan2014], based on the DS-sampling and DS-FCM waveform, is developed to increase resolution of Doppler focusing approach and make it more robust in noisy conditions.
Difference sets are well studied in combinatorics [@pott2012difference; @hall1998combinatorial]. Difference set codes have various applications in sequence design [@pott2012difference], communication systems [@weng2001perfect; @Ipatov2004], array sensors [@kopilovich2008square; @oliveri2009linear] and error correction systems [@Kato]. The main idea of our work comes from [@Xia2005], which uses difference sets to design complex code-books that achieve the Welch bound with equality.
The rest of the paper is organized as follows: In [Section \[sec:Radar Measurement Model\]]{}, we describe the radar measurement models. We explain the properties of the sampling dictionary in [Section \[sec:Coherence Analysis\]]{}, we prove that highly incoherent dictionaries may be obtained via DS-sampling. The recovery algorithms for the proposed structured model is developed in [Section \[sec:Recovery Algorithms\]]{}. We introduce DS-FCM waveform in [Section \[sec.DSFCM\]]{}. In [Section \[sec:SimulationResults\]]{} we present the simulation results, and finally, we conclude the paper in [Section \[sec:Conclusion\]]{}.
Throughout the paper, we use the following notation. Matrices and vectors are denoted by bold letters, where we use lowercase letters, such as $\mathbf{x}$, for vectors and uppercase letters, such as $\mathbf{X}$, for matrices. The $i^{\text{th}}$ element of a vector ${\mathbf{x}}$ is denoted by $x_i$ and $X_{ij}$ denotes the element in the matrix ${\mathbf{X}}$ that is in the $i^{\text{th}}$ row and $j^{\text{th}}$ column. The $i^{\text{th}}$ column of the matrix ${\mathbf{X}}$ is denoted by the vector ${\mathbf{x}}_i$. Also, we use the notation $\tilde{{\mathbf{x}}}_i$ to denote the $i^{{\text{th}}}$ column of the matrix ${\mathbf{X}}^T$, *i.e.* the transpose of the $i^{{\text{th}}}$ row of the matrix ${\mathbf{X}}$. The cardinality of a set $\mathcal{K}$ is denoted by $|\mathcal{K}|$. $\mathbb{Z}_N = \{0, 1, ..., N-1\}$ denotes the set of integers modulo $N$. The $l_p$ norm of a vector $\mathbf{x}$ with length $n$ for $p \in [1,\infty)$ is defined as $\|{\mathbf{x}}\|_p=(\sum_{i=1}^n |{\mathbf{x}}_i|^p)^\frac{1}{p}$. The $\operatorname{rect}(\cdot)$ function represents a square pulse signal with height and width equal to one. Mathematically, $\operatorname{rect}(x) = 1$ if $|x| {\leqslant}1/2$ and $\operatorname{rect}(x) = 0$ otherwise.
Radar Measurement Models {#sec:Radar Measurement Model}
========================
The transmitted signal of a radar is assumed to be a train of $P$ pulses with pulse repetition interval (PRI) of $\tau$ $$\label{Eq.12}
x_T(t) = \sum_{p=0}^{P-1}{h(t -p\tau)}, \: \: \: 0{\leqslant}t< P\tau,$$ where $h(t)$ is a baseband pulse with the continuous-time Fourier transform (CTFT) $H(\omega)$ and two-sided spectrum bandwidth of $B_h$.
The common assumption in the radar signal processing is that the unknown parameters of targets (*i.e.* attenuation factors, delays and Doppler frequencies) are approximately remain constant during transmitting and receiving the $P$ pulses, so this time interval is called the coherent processing interval (CPI) of the radar [@Skolnik2003]. Based on this assumption, the received signal can be written as $$\label{Eq.received_signal}
y_R(t) = \sum_{p=0}^{P-1} \sum_{s=1}^S {a_s \, h(t - t_s -p\tau)} \, e^{j2\pi f_s t},$$ where $S$ is the number of targets and $\{a_s, t_s, f_s\}_{s=1}^S$ are unknown attenuation factors, delays and Doppler frequencies of the targets, respectively. Furthermore, we assume that the delays and Doppler frequencies belong to the unambiguous delay-Doppler region of the radar, meaning that $t_s \in [0, \tau)$ and $f_s \in [-1/2\tau, 1/2\tau )$ for $s=1,2,\ldots,S$.
Fourier coefficients of the received signal in the $p^{\text{th}}$ PRI are $$\label{Eq.14}
\begin{split}
&Y_p[k] = \frac{1}{\tau}\int_{0}^{\tau} \sum_{s=1}^S a_s \, h(t-t_s) \, e^{j2\pi f_s (t+p\tau)} \, e^{-j\omega_k t}{\mathop{}\!\mathrm{d}}t \\
&=\frac{1}{\tau}\sum_{s=1}^S a_s e^{j2\pi f_s (t_s + p\tau)}\int_{0}^{\tau} h(t-t_s) e^{j2\pi f_s (t-t_s)}e^{-j\omega_k t} {\mathop{}\!\mathrm{d}}t,
\end{split}$$ where $\omega_k = 2 \pi k/ \tau$ and the indices $k$ are integers that belong to $\mathbb{Z}$. Using the time and frequency shifting properties of the Fourier transform we may rewrite as $$\label{Eq.15}
Y_p[k] = \frac{1}{\tau}\sum_{s=1}^S a_s \, H(\omega_k - 2\pi f_s) \, e^{j2\pi f_s p \tau} \, e^{-j(\omega_k - 2\pi f_s)t_s}.$$
Bandwidth of the radar waveform is typically much greater than the Doppler frequencies of targets. This implies that for most $k \in \mathbb{Z}$ we have $\omega_k \gg 2\pi f_s$. Using this fact we may approximate by $$\label{Eq.16}
Y_p[k] \approx \frac{1}{\tau}H(\omega_k)\sum_{s=1}^S a_s \, e^{j2\pi f_s p \tau} \, e^{-j\omega_k t_s}.$$ The approximation for the Fourier coefficients of the received signal, given in (\[Eq.16\]), is known as the Fourier coefficient of a semi-periodic finite rate of innovation (FRI) signal in the CS literature [@Eldar2012], and it has been used for radar signal modeling in previous works [@Bajwa2011; @Bar-Ilan2014].
Note that the radar waveform is a baseband pulse with the two-sided bandwidth of $B_h$. Therefore, assuming that the waveform has a symmetric spectrum, only the Fourier coefficients with indices $k$ belonging to the set $$\label{Eq.Fset}
\mathcal{I} = \left\{{-}\!\left\lceil\frac{\tau B_h}{2} \right\rceil, \ldots, -1, 0, 1, \ldots, \left\lfloor \frac{\tau B_h}{2} \right\rfloor \right\},$$ have non-zero values. Using this fact, we can give an interpretation of the Shannon-Nyquist theorem in the Fourier domain; We can reconstruct the received signal exactly, using $|\mathcal{I}| = \lceil\tau B_h\rceil$ Fourier coefficients per PRI. Commonly, $B_h$ (related to both the positive and negative frequencies of $h(t)$) is called the *Nyquist frequency* or the *Nyquist rate* of the signal. In practice, we know that the targets are small objects in the space that is scanned by the radar. Thus, if we represent the pair of unknown delays and Doppler frequencies of all targets with points $\{(t_s, f_s) \}_{s=1}^S$, then, these points are distributed sparsely over the unambiguous delay-Doppler plane of the radar. CS theory predicts that high dimensional *sparse* signals can be recovered from highly incomplete measurements using efficient algorithms [@candes2006compressive]. So, it is very promising to employ CS theory in order to recover sparse targets directly from sub-Nyquist samples.
The first step in the receiver of a CS-based radar is sampling the received signal with a rate that is far lower than the Nyquist rate. For this reason, we only acquire the Fourier coefficients in for indices $k \in \mathcal{K} \subset \mathcal{I}$ such that $|\mathcal{K}| \ll |\mathcal{I}|$. For simplicity, let’s denote the size of the set $\mathcal{K}$ by $K=|\mathcal{K}|$. Entries of the set $\mathcal{K}$ may be chosen deterministically or randomly from the set $\mathcal{I}$. Practically, the Fourier coefficients in the radar receiver may be acquired using different hardwares, such as sampling kernels [@Tur2011; @Unser2000] or multichannel sampling schemes [@Gedalyahu2011; @Kusuma2006; @Olkkonen2008]. Let us denote the incomplete measurements obtained from the $p^{\text{th}}$ PRI by a $K \times 1$ vector ${\mathbf{y}}_p = [{\bar{Y}}_p[\kappa_1], {\bar{Y}}_p[\kappa_2], \ldots, {\bar{Y}}_p[\kappa_K]]^T$, where $\kappa_i$ for $i=1,2,\ldots, K$ denotes the $i^{{\text{th}}}$ element of the set $\mathcal{K}$ and we define ${\bar{Y}}_p[k] = \tau Y_p[k]/H(\omega_k)$. The measurement matrix ${\mathbf{Y}}=[{\mathbf{y}}_0,{\mathbf{y}}_1,\ldots,{\mathbf{y}}_{P-1}]$ is constructed by collecting the measurement vectors ${\mathbf{y}}_p$ for $p=0,1, \ldots, P-1$, $$\label{MM}
{\mathbf{Y}}=
\begin{bmatrix}
{\bar{Y}}_0[\kappa_1] & {\bar{Y}}_1[\kappa_1] & \cdots & {\bar{Y}}_{P-1}[\kappa_1] \\
{\bar{Y}}_0[\kappa_2] & {\bar{Y}}_1[\kappa_2] & \cdots & {\bar{Y}}_{P-1}[\kappa_2] \\
\vdots & \vdots & \ddots & \vdots \\
{\bar{Y}}_0[\kappa_{K}] & {\bar{Y}}_1[\kappa_{K}] & \cdots & {\bar{Y}}_{P-1}[\kappa_{K}] \\
\end{bmatrix}.$$
Next, we need a mathematical relation between the low-rate measurement samples ${\mathbf{Y}}$ and the sparse delay-Doppler plane of the radar. To do this, first consider a quantized delay-Doppler plane with $N$ uniform delay grids over $[0, \tau)$ and $M$ uniform Doppler frequency grids over $[-1/2\tau, 1/2\tau)$, as illustrated in [Fig. \[fig:DDDP\]]{}. So the $n^{{\text{th}}}$ delay grid indicates the delay of $t_n = \frac{(n-1)\tau}{N}$ and the $m^{{\text{th}}}$ Doppler frequency grid represents the Doppler frequency of $f_m = -\frac{1}{2\tau} + \frac{m-1}{M\tau}$. We denote this quantized delay-Doppler plane by an $M \times N$ matrix ${\mathbf{X}}$, which has only $S$ non-zero elements. Here, we consider several approaches that relates the measurement matrix ${\mathbf{Y}}$ to the sparse delay-Doppler plane matrix ${\mathbf{X}}$.
![A quantized delay-Doppler plane with $N$ uniform delay grids and $M$ uniform Doppler grids[]{data-label="fig:DDDP"}](figures/fig_QDDP){width=".4\textwidth"}
The Standard Models
-------------------
The models proposed in [@Ender2010a] and [@Herman2009], which we call them standard models, use a standard CS algorithm with huge dictionary-size to recover sparse targets in the delay-Doppler plain. Here, we introduce a new formulation for the standard models, which helps us develop simple models and recovery algorithms when we discuss our dictionary construction.
Consider the column vectors ${\mathbf{y}}$ and ${\mathbf{x}}$, which are constructed by concatenating all columns of the measurement matrix ${\mathbf{Y}}$ and sparse matrix ${\mathbf{X}}$, respectively. Assuming that targets lie exactly on the delay-Doppler grids, one can show may be written in the matrix form $$\label{standard_model}
{\mathbf{y}}= ({\mathbf{\Phi}}\otimes {\mathbf{\Psi}}) {\mathbf{x}},$$ where $\otimes$ denotes the Kronecker product, vector ${\mathbf{x}}$ is an $S$-sparse vector (*i.e.* it has at most $S$ non-zero elements), ${\mathbf{\Phi}}=[\Phi_{kn}]$ is a $K \times N$ matrix where $\Phi_{kn} = e^{-j\omega_k t_n}$ and ${\mathbf{\Psi}}=[\Psi_{pm}]$ is a $P \times M$ matrix where $\Psi_{pm} = e^{j2\pi f_m (p-1)\tau}$. We call ${\mathbf{\Phi}}$ and ${\mathbf{\Psi}}$ the delay and Doppler dictionaries, respectively.
Let denote the Kronecker product of the delay and Doppler dictionaries by ${\mathbf{A}}= {\mathbf{\Phi}}\otimes {\mathbf{\Psi}}$. Then, we can almost always recover the sparse vector ${\mathbf{x}}$ by solving the following optimization problem [@candes2006robust] $$\label{eq.l1min}
\min_{{\mathbf{x}}}{\|{\mathbf{x}}\|_1} \quad \text{subject to} \quad \mathbf{y = A x},$$ assuming that $S$ is small enough compared to the number of measurements.
Note that in reality, our measurements are contaminated by noise. In this case, instead of solving the optimization , we can solve the following optimization $$\label{eq.l1min_noisy}
\min_{{\mathbf{x}}}{\|{\mathbf{x}}\|_1} \quad \text{subject to} \quad \mathbf{\|y - A x\|_2 {\leqslant}\varepsilon},$$ where $\varepsilon$ is a bound on the noise magnitude chosen appropriately [@candes2006robust].
Numerous tools and algorithms in convex optimization [@Boyd2004] may be employed to solve and . Greedy algorithms are an alternative approach to obtain the sparse vector ${\mathbf{x}}$. These algorithms are iterative in nature and select columns of matrix ${\mathbf{A}}$ according to their correlation with the measurement ${\mathbf{y}}$. There are several greedy algorithms proposed for the sparse recovery. The orthogonal matching pursuit (OMP) [@Pati1993] and the iterative hard thresholding (IHT) [@blumensath2009iterative] are the two simple and well-known greedy methods commonly used in the most CS applications. The setting that uses only the sparsity of the unknown vector to regularize the problem of finding ${\mathbf{x}}$ in is called the standard CS [@duarte2011structured].
When ${\mathbf{x}}$ is recovered, it is easy to obtain the delay-Doppler matrix ${\mathbf{X}}$, where the indices of its non-zero elements show indices of the delay and Doppler grids. Note that each non-zero element of ${\mathbf{X}}$ is a complex number that corresponds to one target and its value is equal to the attenuation factor $a_s$ of that target. However, in real world scenarios, targets do not lie exactly on the delay-Doppler grid. Due to this fact, even in the noiseless setting, CS algorithms recover the values of ${\mathbf{X}}$ with different leakage attenuation based on the position of the targets. This leakage effect can be reduced by increasing the number of delay and Doppler grids in high resolution radars.
Unfortunately, the standard model uses a dictionary with the size of $PK\times MN$, which is proportional to the number of delay-Doppler grid size. The huge dictionary-size in high resolution radars results in recovery algorithms with very high computational complexity. In the following, we develop two models by breaking the huge dictionary of the standard model to smaller dictionaries. This reduction helps us employing recovery algorithms with much lower computational complexity.
The Proposed Structured Method {#subsec.structured_model}
------------------------------
\[sec:proposed\_algorithm\]
In addition to sparsity, we sometimes have extra knowledge about the structure of the measured signal. Structured CS uses this knowledge in order to reduce the sampling rate or improve the recovery performance compared to the standard CS [@duarte2011structured]. The proposed structured method uses a similar procedure as the model used in [@Bajwa2011]. However, our method uses a CS-dictionary to recover the delay-Doppler plain, while the proposed method in [@Bajwa2011] uses complicated spectral estimation tools for recovery.
To model radar measurements based on structured CS, first, we recover the delays of targets. In this case, the measurements corresponding to the $p^{{\text{th}}}$ PRI can be written as $$\label{eq.Delay-recovery-model}
{\mathbf{y}}_p = {\mathbf{\Phi}}{\mathbf{b}}_p,$$ where ${\mathbf{\Phi}}$ is the delay dictionary and ${\mathbf{b}}_p$ is an $S$-sparse vector with non-zero elements at indices of the delay grids corresponding to target delays. One can easily show that the non-zero element of ${\mathbf{b}}_p$ resulting from the $s^{{\text{th}}}$ target have the value proportional to $e^{j2\pi f_s (p-1)\tau}$. Furthermore, all sparse vectors $\{{\mathbf{b}}_p\}_{p=1}^P$ have the same support, because we assume that the unknown parameters of the received signal are constant during each CPI.
Consequently, the multiple measurements of radar can be modeled as $$\label{Jointly-Delay-recovery-model}
{\mathbf{Y}}= {\mathbf{\Phi}}{\mathbf{B}},$$ where ${\mathbf{Y}}$ is the measurement matrix, defined in , and ${\mathbf{B}}=[{\mathbf{b}}_1, {\mathbf{b}}_2, \ldots, {\mathbf{b}}_P]$ denotes an $N \times P$ matrix where the $p^{{\text{th}}}$ column of ${\mathbf{B}}$ is equal to ${\mathbf{b}}_p$.
The analogue of standard CS recovery problem in the multiple measurements case is [@cotter2005sparse; @fornasier2008recovery] $$\label{eq.l0minMMV}
\min_{{\mathbf{B}}}{\|{\mathbf{B}}\|_{1,q}} \quad s.t. \quad \mathbf{Y = \Phi B},$$ for some $q {\geqslant}1$ ($q=1,2$ and $\infty$ have been used in literature), where we define $\|\cdot\|_{i,q}$ as $$\label{eq.lpqNorm}
\|{\mathbf{B}}\|_{i,q} = \bigg(\sum_{n=1}^N \|\tilde{{\mathbf{b}}}_n\|_i^q \bigg)^\frac{1}{q}.$$ Similar to standard CS, there are various algorithms, based on the greedy pursuit [@tropp2006algorithmsgreed; @mishali2008reduce] and convex optimization [@tropp2006algorithmsconvex], that use the joint sparsity knowledge of the measured signal to recover the matrix ${\mathbf{B}}$ from the multiple measurement vectors (MMV). After successful recovery of the delays of the targets, we keep the delay indices of targets on the delay grid in the set $\mathbf{\Omega}$. Next, Doppler frequencies of each target can be estimated by solving the following $S$ low complexity standard CS problems: $$\label{Doppler-recovery}
\tilde{{\mathbf{b}}}_n = {\mathbf{\Psi}}{\mathbf{x}}_n, \quad n \in \mathbf{\Omega},$$ where $\tilde{{\mathbf{b}}}_n$ denotes the $n^{{\text{th}}}$ column of ${\mathbf{B}}^T$, ${\mathbf{\Psi}}$ is the Doppler dictionary and ${\mathbf{x}}_n$ is the $n^{{\text{th}}}$ column of the delay-Doppler matrix ${\mathbf{X}}$.
An alternative approach to estimate the Doppler frequency corresponding to $\tilde{{\mathbf{b}}}_n$ is using FFT, similar to the conventional Nyquist processing. It is enough to find the peak of FFT$(\tilde{{\mathbf{b}}}_n)$ and the frequency corresponds to this peak. The above proposed algorithm may be utilized to recover the delay-Doppler matrix ${\mathbf{X}}$, by breaking the solving procedure of the previous standard model, given in , with the dictionary size of $PK\times MN$ to two steps. Step one is solving one simultaneous delay recovery problem, given in , with the dictionary size of $K \times N$, and step two includes $|\mathbf{\Omega}|$ (the number of recovered delays) standard Doppler recovery problems, given in , with the dictionary size of $P \times M$. We can see that the complexity of solving the introduced subproblems for the recovery of delay-Doppler plane is much lower than the complexity of solving the original standard model.
The problem of delay-Doppler recovery may be solved via the Doppler focusing approach [@Bar-Ilan2014] by solving $M$ standard recovery problem with the dictionary size of $K \times N$. We discuss this method in the next subsection.
Doppler Focusing {#subsec.DF_model}
----------------
Similar to the proposed algorithm in Section \[sec:proposed\_algorithm\], Doppler focusing is also a recovery approach with low computational complexity, which has recently been proposed for sub-Nyquist radars [@Bar-Ilan2014]. Actually, this approach first recovers the Doppler frequencies of the targets by using FFT (or filter bank with non-uniform Doppler grids) and later finds the delays of the targets. In Doppler focusing the order of recovering delays and Doppler frequencies are reversed compared to the proposed method in Section \[sec:proposed\_algorithm\].
In the Doppler focusing, first, we calculate $M$ focused measurements by filtering measurements of $P$ pulses around a Doppler grid as follows: $$\label{Focused_Measurements}
d_m[k] = \sum_{p=0}^{P-1}{\bar{Y}}_p[k]e^{-jf_m p\tau}, \quad k \in \mathcal{K}, \ 1 {\leqslant}m {\leqslant}M,$$ where $d_m[k]$ denotes the focused Fourier coefficient with index $k$, around the $m^{{\text{th}}}$ Doppler grid with frequency of $f_m$. Note that, if Doppler grids are chosen uniformly, then is actually the definition of the discrete Fourier transform (DFT) of length $P$. It is shown that using focused measurements improves SNR proportional to the number of pulses $P$ [@Bar-Ilan2014]. Let vector $\mathbf{d}_m$ denote the focused measurement vector, where $k^{{\text{th}}}$ element of $\mathbf{d}_m$ is $d_m[k]$. We can recover delays of targets for each focused area by solving the standard CS problems as below $$\label{DF_model}
\mathbf{d}_m = {\mathbf{\Phi}}\tilde{{\mathbf{x}}}_m, \quad 1 {\leqslant}m {\leqslant}M,$$ where $\tilde{{\mathbf{x}}}_m$ denotes the $m^{{\text{th}}}$ column of ${\mathbf{X}}^T$.
To compare, we summarize the number of problems and dimension of each problem for different measurement models in Table \[tab0\].
\[h!\]
----------------------- --------------------- ------------------------------------------ ----------------
Measurements Model Number of Dictionary Dimensions
Problems
Standard Model $1$ ${\mathbf{\Phi}}\otimes {\mathbf{\Psi}}$ $KP \times NM$
Structured Model $1$ ${\mathbf{\Phi}}$ $K \times N$
$|\mathbf{\Omega}|$ ${\mathbf{\Psi}}$ $P \times M$
Doppler Focusing (DF) $M$ ${\mathbf{\Phi}}$ $K \times N$
----------------------- --------------------- ------------------------------------------ ----------------
: Comparison of dimensions and number of problems for each model
\[tab0\]
In the next section, we investigate the necessary conditions on the measurement dictionaries to guarantee uniqueness of recovery.
Dictionary Coherence Analysis {#sec:Coherence Analysis}
=============================
The CS-dictionary should be designed such that the number of measurements is reduced as much as possible, and the uniqueness of measurement for any particular signal ${\mathbf{x}}$ is guaranteed as well. Coherence is one of the easily computable properties of ${\mathbf{\Phi}}$ that guarantees distinct signals ${\mathbf{x}}$ and $\tilde{{\mathbf{x}}}$, lead to different measurement vectors ${\mathbf{\Phi}}{\mathbf{x}}$ and ${\mathbf{\Phi}}\tilde{{\mathbf{x}}}$. This property is defined as follows [@tropp2005signal]
The coherence of a dictionary ${\mathbf{\Phi}}=[\mathbf{\phi}_1,\phi_2, \ldots, \phi_N]$, denoted by $\mu({\mathbf{\Phi}})$, is the largest absolute inner product between any two columns ${\mathbf{\phi}}_i$, ${\mathbf{\phi}}_j$ of ${\mathbf{\Phi}}$
$$\label{Eq.coh}
\mu({\mathbf{\Phi}}) = \max_{1{\leqslant}i < j {\leqslant}N}{\frac{| \langle{\mathbf{\phi}}_i, {\mathbf{\phi}}_j\rangle |}{\|{{\mathbf{\phi}}_i}\|_2 \|{{\mathbf{\phi}}_j\|_2}}}.$$
In the other words, the coherence is the maximum cross-correlation between normalized columns (*i.e.* atoms) of the dictionary. It has been shown that $\mu({\mathbf{\Phi}})$ is lower bounded by $\sqrt{\frac{N-K}{K(N-1)}}$ [@Welch1974], which is known as the *Welch bound*.
[Lemma\[thm:cohcon\]]{} gives the necessary condition on the measurement matrix ${\mathbf{\Phi}}$ that guarantees uniqueness of recovery.
\[thm:cohcon\] For a matrix ${\mathbf{\Phi}}$, if $$\label{Eq.cohcon}
S < \frac{1}{2} \left(1+\frac{1}{\mu({\mathbf{\Phi}})}\right),$$ then for each measurement vector ${\mathbf{y}}$ there exists at most one $S$-sparse signal ${\mathbf{x}}$ such that $\mathbf{y = \Phi x}$ *[@Donoho2003; @tropp2004greed]*.
In this section, we look for the best sampling index set $\mathcal{K}$ that minimizes coherence of the delay dictionary, without decreasing the radar resolution. By substituting values of $\omega_k = 2\pi \kappa_k/\tau$ and delay grids $t_n = (n-1)\tau/N$ in the delay dictionary ${\mathbf{\Phi}}$, defined in , we may rewrite the dictionary as $$\label{delay dictionary}
{\mathbf{\Phi}}=
\left(
\begin{array}{cccc}
1 & e^{-j \frac{2\pi}{N} \kappa_1} & \cdots & e^{-j \frac{2\pi}{N} \kappa_1 (N-1)} \\
1 & e^{-j \frac{2\pi}{N} \kappa_2} & \cdots & e^{-j \frac{2\pi}{N} \kappa_2 (N-1)} \\
\vdots & \vdots & \ddots & \vdots \\
1 & e^{-j \frac{2\pi}{N} \kappa_{K}} & \cdots & e^{-j \frac{2\pi}{N} \kappa_{K} (N-1)} \\
\end{array}
\right),$$ where $\kappa_k$ denotes the $k^{{\text{th}}}$ element of the set $\mathcal{K}$ and $N$ is the number of delay grids.
If we denote the normalized cross-correlation between the $\ell^{{\text{th}}}$ and the $(\ell+u)^{{\text{th}}}$ columns of the dictionary ${\mathbf{\Phi}}$ by $\mu(u)$, the coherence of ${\mathbf{\Phi}}$, according to Definition 1, is given by $$\label{Eq.coh2}
\mu({\mathbf{\Phi}}) = \max_{1 {\leqslant}u {\leqslant}N}{\mu(u)} = \max_{1 {\leqslant}u {\leqslant}N}{\frac{1}{K}
\left\lvert
{\sum_{k=1}^{K}{e^{-j2\pi\frac{u}{N}\kappa_k}}}
\right\rvert
}.$$ As mentioned in the previous section, the entries of the set $\mathcal{K}$ may be chosen deterministically or randomly from the set of the all Fourier indices $\mathcal{I}$.
Here we discuss different sampling schemes of the set $\mathcal{I}$:
Consecutive Sampling
--------------------
The simplest sampling scheme uses a sampling index set $\mathcal{K}$ which the elements of $\mathcal{K}$ are consecutive elements of $\mathcal{I}$. It is obvious that in this case is the summation of a geometric sequence, which is equal to $$\label{Eq.coh_consec_sampling}
\mu({\mathbf{\Phi}}) = \max_{1 {\leqslant}u {\leqslant}N}{\mu(u)} = \max_{1 {\leqslant}u {\leqslant}N}{\frac{1}{K}
\left\rvert
\frac{\sin(\pi u K/N)}{\sin(\pi u/N)}
\right\rvert
}.$$ We conclude from that
- The adjacent columns in the delay dictionary, related to adjacent delay grids, have the most correlation and $\mu(u)$ maximizes for $u=1$.
- Increasing the number of grids can make adjacent atoms to become more correlated.
- It is possible to design a dictionary with orthogonal columns, if and only if $N = K$. In this case, the number of samples determines the grid resolution.
This means that the consecutive sampling does not have any advantage over the Nyquist sampling, because the radar resolution is proportional to the sampling rate. This diminishes the main goal of CS, *i.e.* reducing the sampling rate without losing resolution. Furthermore, one can limit the bandwidth of the received signal with an appropriate filter and use Nyquist processing to achieve the same performance with a recovery algorithm that has a lower computational complexity. This is the main problem of consecutive sampling that is observed in [@Bar-Ilan2014] and [@Bajwa2011].
Random Sampling
---------------
In this case, the measurements of each pulse is acquired by randomly choosing the entries of the set $\mathcal{K}$ from the set $\mathcal{I}$. Random sampling is a favorite scheme in the CS literature as it improves the mean coherence of the dictionary. It is shown that random partial Fourier matrices, with $O(S \log^5{N})$ measurements, guarantee unique recovery of an $S$-sparse vector with high probability [@rudelson2008sparse]. However, in some radar applications, it is very important to have a reliable result in the worst case for each target estimation, and not have a reliable estimation on average. Moreover, random sampling requires more complex and sometimes impractical sampling hardware. So we are interested in optimal deterministic sampling schemes for CS-based radars.
Optimal Sampling
----------------
In order to achieve the highest possible resolution for the delay recovery and minimize the number of samples required for unique recovery, we must design ${\mathbf{\Phi}}$ with the minimum coherence, *i.e.* the coherence of ${\mathbf{\Phi}}$ should become equal to the Welch bound. This problem is formulated mathematically as $$\label{Eq.welch_problem}
\text{find } \mathcal{K}\subseteq \mathcal{I} \quad s.t. \quad \mu({\mathbf{\Phi}}) = \sqrt{\frac{N-K}{K(N-1)}}.$$
It is shown that the difference sets may be employed to create complex code-books that have coherence equal to the Welch bound [@Xia2005]. This result may be extended to CS theory and we can prove that the coherence of the delay dictionary, which is constructed via the difference set $\mathcal{K}$ is equal to the Welch bound. Before continuing, let us define a difference set.
\[dif.diffset\] A subset $\mathcal{K} = \{\kappa_1, \kappa_2, \ldots, \kappa_{K}\}$ of $\mathbb{Z}_N$ is called an $(N,K,\lambda)$ difference set if set of the $K(K-1)$ modulo $N$ differences of the form $$\label{Eq.diffset}
\mathcal{D}_N = \bigcup_{k = 1}^K\bigcup\limits_{\substack{l=1 \\ l\neq k}}^K\bigg\{(\kappa_k - \kappa_l)\ mod\ N\bigg\}$$ contains every non-zero values of $\mathbb{Z}_N$ exactly $\lambda$ times.
According to this definition, the three parameters $(N,K,\lambda)$ are not independent and they satisfy $$\label{Eq.diffsetparam}
\lambda(N-1) = K(K-1).$$
\[thm:DS\_sampling\] The coherence of a $K \times N$ delay dictionary ${\mathbf{\Phi}}$, defined in , is equal to the Welch bound if the sampling index set $\mathcal{K}$ is an $(N,K,\lambda)$ difference set.
One solution to the problem is obtained by considering a special case that all atoms have the same cross-correlation with each others. In other words, we find atoms of a dictionary such that $$\label{Eq.welch_problem_special_case}
\text{find } \mathcal{K} \subseteq \mathcal{I} \quad s.t. \quad \mu(u) = \sqrt{\frac{N-K}{K(N-1)}}, \ 1 {\leqslant}u {\leqslant}N.$$
We follow the steps of [@Xia2005] to show that the coherence of ${\mathbf{\Phi}}$ is equal to the Welch bound. From , we have $$\begin{aligned}
\label{Eq.mu}
\nonumber
\mu^2(u) &= \frac{1}{K^2}
\Bigg(
\sum_{k=1}^K e^{-j2\pi u \kappa_k/N}
\Bigg)
\Bigg(
\sum_{l=1}^K e^{j2\pi u \kappa_l/N}
\Bigg) \\
\nonumber
&= \frac{1}{K} + \frac{1}{K^2}
\sum_{k=1}^K\sum\limits_{\substack{l=1 \\ l\neq k}}^K e^{j2\pi u (\kappa_l - \kappa_k)/N} \\
\nonumber
&= \frac{N-K}{K(N-1)} + \frac{K-1}{K(N-1)} \\
& \quad + \frac{1}{K^2}
\sum_{k=1}^K\sum\limits_{\substack{l=1 \\ l\neq k}}^K e^{j2\pi u (\kappa_l - \kappa_k)/N}.\end{aligned}$$ By defining $$\label{eq.defz}
z[u] \triangleq K^2. \left(\mu^2(u) - \frac{N-K}{K(N-1)}\right),$$ and $$\lambda_0 \triangleq \frac{K(K-1)}{N-1},$$ we can rewrite as $$\label{Eq.mu2}
z[u] = \lambda_0 + \sum_{k=1}^K\sum_{\substack{l=1 \\ l\neq k}}^K e^{j2\pi u (\kappa_l - \kappa_k)/N}.$$ Consider the set $\mathcal{D}_N = \{\kappa_l - \kappa_k \ | \ 1 {\leqslant}k, l {\leqslant}K, k \neq l\}$. Assume that number of elements in $\mathcal{D}_N$ with the value $d$ is equal to $\lambda_d$, for $d = 1,...,N-1$. Then, can be written as $$\label{Eq.times}
z[u] = \sum_{d=0}^{N-1}\lambda_d \, e^{j2\pi u d/N}.$$ Notice that in the sequence $\{z[u]\ : \ 1 {\leqslant}u {\leqslant}N\}$ is the inverse discrete Fourier transform (IDFT) of the sequence $\{\lambda_d \ : \ 1 {\leqslant}d {\leqslant}N\}$. According to and , it can be seen that the set $\mathcal{K}$ is the solution of if $$\label{Eq.zcondition}
z[0] = \frac{K(K-1)N}{N-1}, \text{ and } z[1] = 0, z[2] = 0, \ldots, z[N-1] = 0,$$ which leads to $$\label{Eq.timescondition}
\lambda_0 = \lambda_1 = \ldots = \lambda_{N-1} = \frac{K(K-1)}{(N-1)}.$$ This is due to the fact that the delta function and the constant value function are a Fourier transform pair. This means that, all elements of the set $\mathcal{D}_N$ are repeated exactly $\lambda = \frac{K(K-1)}{(N-1)}$ times. So, the only solution of the problem according to must be an $(N,K,\lambda)$ difference set.
We can extend the result of this theorem to CS-based radar models by following lemmas.
\[lem:DSstructured\] Consider a CS-based radar with the structured measurement model (introduced in Section \[subsec.structured\_model\]) and sampling index set $\mathcal{K}$, where $\mathcal{K} \subseteq \mathcal{I}$ and the set $\mathcal{I}$ is defined in . If $\mathcal{K}$ is an $(N,K,\lambda)$ difference set with $K \ll N$, then CS algorithms can recover at most $\lfloor \frac{1}{2} \bigr( 1 + \sqrt{K} \bigl)\rfloor$ unknown delays with delay resolution of $\tau/N$.
In the structured model, first, we recover delays of targets by estimating the jointly sparse matrix ${\mathbf{B}}$ in or sparse vectors $\{{\mathbf{b}}_p\}_{p=1}^P$ in . Considering $K \ll N$, based on [Theorem\[thm:DS\_sampling\]]{}, the coherence of the delay dictionary ${\mathbf{\Phi}}$ can be approximated by $\mu({\mathbf{\Phi}}) \approx \frac{1}{\sqrt{K}}$. So, according to [Lemma\[thm:cohcon\]]{}, we can uniquely recover $S$-sparse vectors with sparsity level of $$\label{Eq.DScohcon}
S < \frac{1}{2}\big(1+\sqrt{K}\big) \approx O(\sqrt{K}),$$ using appropriate CS algorithms. The delay dictionary includes highly incoherent atoms, which are related to delay grids. Therefore, the delay resolution of the recovered unique $S$-sparse vector is determined by the delay grids resolution, which is $\tau/N$. Note that the number of delay grids $N$ is determined by the order of the difference set and it is bounded by $N {\leqslant}\tau B_h$, because $\mathcal{K} \subseteq \mathcal{I}$ and according to we have $|\mathcal{I}| = \tau B_h$.
\[lem:DSdopplerfocusing\] (Modified Doppler focusing) Consider a CS-based radar with the Doppler focusing model (introduced in Section \[subsec.DF\_model\]) and sampling index set $\mathcal{K}$, where $\mathcal{K} \subseteq \mathcal{I}$ and the set $\mathcal{I}$ is defined in . If $\mathcal{K}$ is an $(N,K,\lambda)$ difference set with $K \ll N$, then CS algorithms can recover at most $\lfloor \frac{P}{2} \bigr( 1 + \sqrt{K} \bigl)\rfloor$ targets with the delay resolution of $\tau/N$.
The Doppler focusing approach, first, separates Doppler frequencies of targets using an appropriate filter bank. So, the Doppler resolution for a uniform Doppler grid is determined similar to a conventional Doppler recovery using FFT, *i.e.* $1/P\tau$ [@Skolnik2003]. According to [Theorem\[thm:cohcon\]]{}, for all separated Doppler frequencies we can uniquely recover $\tilde{{\mathbf{x}}}_m$ in with the sparsity level of at most $\lfloor \frac{1}{2} \bigr( 1 + \sqrt{K} \bigl)\rfloor$. Therefore, all the delay-Doppler plain can be recovered with the sparsity level of at most $\lfloor \frac{P}{2} \bigr( 1 + \sqrt{K} \bigl)\rfloor$. Similar to [Lemma\[lem:DSstructured\]]{}, the delay resolution is $\tau/N$, because these models use the same delay dictionary ${\mathbf{\Phi}}$.
The search for finding new difference sets is an active area of research. However, we can find some of the well studied difference sets in [@pott2012difference] and [@hall1998combinatorial]. Also a comprehensive repository of difference sets is available in [@Gordon.20151June]. For illustration purposes, we list several useful difference sets for high resolution CS-based radars in Tabel \[tab1\]. It is more efficient, if we use equivalent difference sets that have values around zero as listed in third column of Table \[tab1\]. Thus, the resulting indices can conform with the Fourier coefficients which are located around zero. We call the sampling scheme based on these equivalent difference sets *DS-sampling*. To obtain an equivalent difference set we can replace any element $\kappa$ of an $(N,K,\lambda)$ difference set by $\kappa + kN$, where $k$ is an arbitrary integer number, because based on all differences are calculated modulo $N$.
\[!th\]
$(N,K,\lambda)$ Difference Set DS-sampling indices
----------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
$(91, 10, 1)$ {0, 1, 3, 9, 27, 49, 56, 61, 77, 81} {-42, -35, -30, -14, -10, 0, 1, 3, 9, 27}
$(993, 32, 1)$ {0, 1, 33, 86, 90, 132, 148, 168, 191, 213, 241, 251, 260, 262, 265, 446, 490, 507, 586, 615, 650, 656, 663, 690, 774, 792, 800, 872, 887, 926, 938, 963} {-486, -407, -378, -343, -337, -330, -303, -219, , -121, -106, -67, -55, -30, 0, 1, 33, 86, 90, 132, 148, 168, 191, 213, 241, 251, 260, 262, 265, 446, 490}
(2863,54,1) {0, 1, 18, 90, 101, 354, 429, 490, 514, 612, 620, 622, 671, 731, 753, 797, 809, 849, 911, 1054, 1074, 1083, 1087, 1171, 1178, 1199, 1236, 1306, 1387, 1458, 1622, 1637, 1669, 1672, 1714, 1837, 1843, 1868, 1873, 1916, 1942, 1983, 2010, 2029, 2063, 2086, 2149, 2213, 2347, 2361, 2516, 2555, 2571, 2609} {-1405, -1241, -1226, , -1191, -1149, -1026, -1020, -995, -990, -947, -921, -880, -853, -834, -800, -777, -714, -650, -516, -502, -347, -308, -292, -254, 0, 1, 18, 90, 101, 354, 429, 490, 514, 612, 620, 622, 671, 731, 753, 797, 809, 849, 911, 1054, 1074, 1083, 1087, 1171, 1178, 1199, 1236, 1306, 1387}
\[1ex\]
: Examples of Appropriate Difference Sets for CS-based radars.
\[tab1\]
To make it more clear, here we compare coherence of different sampling schemes numerically. Consider the case that the delay dictionary ${\mathbf{\Phi}}$ is constructed using $N = 2863$ uniform delay grids, and $K = 54$ indices are chosen consecutively, randomly or based on the difference set represented in the last row of Table \[tab1\]. The cross-correlation of two atoms correspond to two delay grids versus distance $u$ is illustrated in [Fig. \[fig:coh\]]{}. As we have shown analytically, increasing the number of delay grids in consecutive sampling causes more correlation between adjacent columns of the dictionary. It is obvious that random sampling decreases correlation of adjacent atoms dramatically. Finally, as expected by the result of [Theorem\[thm:DS\_sampling\]]{}, the DS-sampling causes all atoms to have cross correlations equal to the Welch bound. To compare the difference between DS-sampling and random sampling we depict histogram of coherence of the dictionary ${\mathbf{\Phi}}$ in [Fig. \[fig:pdf\]]{} for $10^5$ dictionary construction. It is obvious that for the same number of delay grids, the coherence of delay dictionary constructed via DS-sampling is much lower than the random dictionaries.
![Cross-correlation between atoms with distance $|u|$.[]{data-label="fig:coh"}](figures/fig_coherence){width="21pc"}
![Histogram of coherence of delay dictionary for different sampling schemes ($K = 54$, $N=2863$)[]{data-label="fig:pdf"}](figures/fig_pdf){width="21pc"}
All results expressed in this section may be developed for the Doppler dictionary ${\mathbf{\Psi}}$, defined in , by exchanging parameters $N$ and $K$ with $M$ and $P$, respectively. However, normally the number of pulses $P$ in the radar signal is much smaller in comparison to Fourier samples of each PRI. So compression in this dimension is not necessary.
Recovery Algorithms {#sec:Recovery Algorithms}
===================
The structured method introduced in [Section \[sec:Radar Measurement Model\]]{}-B requires a jointly sparse recovery algorithm to solve the problem to obtain the joint sparse matrix ${\mathbf{B}}$. As mentioned previously, various algorithms are developed for jointly sparse approximation problems (see [@duarte2011structured; @mishali2008reduce; @tropp2006algorithmsgreed; @tropp2006algorithmsconvex]). After successful delay recovery, corresponding columns of the delay-Doppler matrix may be obtained by solving the standard CS problems . [Algorithm\[alg:greedystructuredmodel\]]{} is used for the delay-Doppler recovery in the proposed structured method. In this algorithm, first, we use simultaneous orthogonal matching pursuit (SOMP) [@tropp2006algorithmsgreed] to recover the joint sparse matrix ${\mathbf{B}}$ as well as the support of its non-zero rows (indexed by the set $\mathbf{\Omega}$). Then, the Doppler frequencies correspond to the recovered delays are determined by finding columns of the Doppler dictionary that have maximum cross-correlation with rows of ${\mathbf{B}}$, which are indexed by $\mathbf{\Omega}$. The stopping criterion of [Algorithm\[alg:greedystructuredmodel\]]{} is set to the bound on the sparsity level, *i.e.* $\sqrt{K}$, because according to [Lemma\[lem:DSstructured\]]{} we know that the number of targets must be lower than this bound to guarantee unique recovery. However, similar to any greedy algorithm, which is employed for the sparse approximation, the stopping criterion of [Algorithm\[alg:greedystructuredmodel\]]{} can be chosen to be a bound on the norm of the residual matrix, too.
**Input:** Measurement Matrix ${\mathbf{Y}}_{K \times P}$, Delay Dictionary ${\mathbf{\Phi}}_{K \times N}$, Doppler Dictionary ${\mathbf{\Psi}}_{P \times M}$\
**Initialization:** $ {{\mathbf{X}}}_{M \times N} = \mathbf{0}$, $ {{\mathbf{B}}}_{N \times P} = \mathbf{0}, \mathbf{R} = {\mathbf{Y}}, \mathbf{\Omega = \varnothing}$\
Find an index $n$ such that $$n = \operatorname{argmax}_{j \notin \mathbf{\Omega}}{\|\phi_j^H \mathbf{R} \|_2};$$ where $\mathbf{\phi}_j$ is the $j$’th column of $\mathbf{\Phi}$. Update the support: $$\mathbf{\Omega = \Omega} \cup \{n\};$$ Update the estimate: $$\mathbf{{B}_\Omega} = {\mathbf{\Phi}}_\mathbf{\Omega}^\dagger \mathbf{R};$$ where $\dagger$ is the Moore-Penrose pseudoinverse operator [@golub2012matrix] and ${\mathbf{\Phi}}_\mathbf{\Omega}$ denotes the restriction of ${\mathbf{\Phi}}$ to the columns indexed by $\mathbf{\Omega}$. Update the residual: $$\mathbf{R} = \mathbf{Y} - {\mathbf{\Phi}}_\mathbf{\Omega}\mathbf{{B}}_\mathbf{\Omega};$$ Find index $m$ such that $$m = \operatorname{argmax}_j{\psi_j^H \tilde{{\mathbf{b}}}_{\Omega_i}};$$ where $\psi_j$ is the $j$’th column of ${\mathbf{\Psi}}$ and $\tilde{{\mathbf{b}}}_{\Omega_i}$ denotes the $i$’th column of ${\mathbf{B}}_\mathbf{\Omega}^T$. Estimate delay-Doppler Matrix: $${\mathbf{x}}_{\Omega_i} = \operatorname{argmin}_\mathbf{z}{\|\tilde{{\mathbf{b}}}_{\Omega_i} -\psi_m^\dagger \mathbf{z} \|_2};$$ where ${\mathbf{x}}_{\Omega_i}$ denotes the $i$’th column of ${\mathbf{X}}_\mathbf{\Omega}$.
**Output:** Estimated Delay-Doppler Matrix ${{\mathbf{X}}}$\
After recovery of the delay-Doppler matrix, targets are detected by finding local maxima of ${\mathbf{X}}$ (or a function of ${\mathbf{X}}$), which are greater than a certain threshold. Threshold adjustment has close relation to the detector structure. In this paper, we do not consider the construction of the detector. For simulation purposes, we assume that the number of targets, $S$, is known. Thus, by finding the $S$ largest local maxima of ${\mathbf{X}}$, the delay and Doppler frequency of each target can be found.
DS-Frequency Coded Modulated Waveform {#sec.DSFCM}
=====================================
Pulse compression is the common signal processing technique used in radars to increase the delay resolution as well as the signal to noise ratio. This is achieved by employing modulated waveforms [@Levanon2004]. In [Section \[sec:Coherence Analysis\]]{}, we show that CS-based radars can achieve the highest possible resolution using DS-sampling. Now, we look for the appropriate modulated waveform to make CS-based radars more robust in noisy conditions. CS-based radars, which perform compressed sampling in the Fourier domain, have low recovery performance in the noisy conditions in comparison to Nyquist processing methods (see the simulation results of [@Bajwa2011] and [@Bar-Ilan2014]). The reason is that the signal energy is spread throughout the radar spectrum. When we reduce the Fourier domain samples, the signal energy of unsampled spectrum is discarded. To address this problem, we should employ a waveform that focuses energy around the selected Fourier domain samples. Here, we propose a frequency coded modulated waveform, called “DS-FCM” waveform, where its energy concentrates around the DS-samples.
We define the DS-FCM waveform with the pulse width of $T_p$ as $$\label{Eq.DSFCM.timedomain}
h(t) = \frac{1}{K} \operatorname{rect}\! \left(\frac{t}{T_p}\right)\sum_{k\in\mathcal{K}}{e^{j2\pi k t/\tau}},$$ where $\mathcal{K}$ is an $(N,K,\lambda)$ difference set. To focus the pulse energy on DS-samples in the Fourier domain, we must increase the pulse width (*i.e.* increasing the support of signal in the time domain). So, the two-sided spectrum bandwidth of is determined by its complex envelope and we may approximate $B_h$ by the largest and smallest entries of the difference set $\mathcal{K}$ as follows $$\label{Eq.DSFCM.bw}
B_h \approx \frac{k_{\max} - k_{\min}}{\tau},$$ where $k_{\max}$ and $k_{\min}$ are the largest and the smallest entries of the difference set $\mathcal{K}$, respectively.
![Power spectrum of the Rectangular, LFM and DS-FCM waveforms. All waveforms have unit power and two-sided spectrum bandwidth of $300$ MHz.[]{data-label="fig:spec"}](figures/fig_spectrum){width="21pc"}
The power spectrum of the rectangular waveform, linear frequency modulated (LFM) waveform and DS-FCM waveform, which is based on a $(2863,54,1)$ difference set, are represented in the last row of Table \[tab1\], are shown in [Fig. \[fig:spec\]]{}. All waveforms have equivalent unit power and two-sided bandwidth of $300$ MHz. It is obvious that the energy of DS-FCM waveform is focused around the Fourier samples rather than the entire spectrum.
Simulation Results {#sec:SimulationResults}
==================
The performance of the proposed sub-Nyquist sampling and processing methods have been evaluated by means of simulation. To compare, results obtained from the conventional Nyquist processing and Doppler focusing approach with the consecutive sampling are also presented. The following specifications and parameters are used in the simulations:
- Radar PRI[^3]: $\tau = 10\ \mu s$
- Receiver bandwidth: $B_h = 300$ MHz (two-sided spectrum bandwidth of the pulse $h(t)$)
- Number of pulses: $P = 20$
- Nyquist sampling rate: $B_h = 300$ MSPS $= 3000$ samples per PRI
- $128$-point FFT is used in Nyquist processing and the corresponding Doppler dictionary in sub-Nyquist methods include $128$ Doppler grids.
- A rectangular waveform with the two-sided spectrum bandwidth of $B_h$ is used for the Nyquist processing.
- Sub-Nyquist methods use DS-FCM waveform, unless it is specified.
- DS-sampling and DS-FCM waveform, which are used in the sub-Nyquist methods, are based on the $(2863,54,1)$ difference set, given in the last row of Table \[tab1\]. This sampling scheme uses only $54$ samples per PRI, which is less than $2\%$ of the Nyquist samples.
- Number of targets $S = 5$, unless it is specified.
- Target delays and Doppler frequencies are distributed uniformly at random over the unambiguous delay-Doppler region and the attenuation factors associated with each target is taken to have a unit amplitude and a random phase.
We consider the case that all received signals contaminated by zero mean complex additive white Gaussian noise (AWGN) with variance $\sigma_w^2$. In this setting, the signal to noise ratio (SNR) of the received signal from the $s^{{\text{th}}}$ target is $$\label{eq.snr.continuous}
{\text{SNR}}_s = \frac{\frac{1}{T_p}\int_0^{T_p}{|a_s h(t)|^2 {\mathop{}\!\mathrm{d}}t}}{\sigma_w^2 B_h},$$ where $T_p$ is the pulse width of the transmitted signal. In the numerical simulations, where an analog signal is modeled with its high rate sampled discrete equivalent, the SNR of the $s^{{\text{th}}}$ target can be calculated as $$\label{eq.snr.discrete}
{\text{SNR}}_s = \frac{\|a_s \mathbf{h} \|_2^2}{L\sigma_w^2},$$ where the $L$-length vector $\mathbf{h}$ is the discrete equivalent of the continuous signal $h(t)$. Notice that, if the entries of $\mathbf{h}$ are the Nyquist samples of $h(t)$, then $L = \lfloor T_p B_h \rfloor$.
The conventional Nyquist processing delay and Doppler frequency resolution (Nyquist bins) are defined as $\delta_t = 1/B_h$ and $\delta_f = 1/P\tau$, respectively. Since we are interested in comparison of sub-Nyquist methods with the conventional Nyquist processing, the following performance metrics based on the Nyquist bins will be useful.
1. **Probability of Detection:** A “*true detection*” is defined as a delay-Doppler estimation with the error less than Nyquist bins. In other words, we have a true detection for the $s^{{\text{th}}}$ target if its estimated delay and Doppler frequency satisfy $$|\hat{t}_s - t_s| < \delta_t \quad \text{and} \quad |\hat{f}_s - f_s| < \delta_f,$$ where $t_s$ and $\hat{t}_s$ denote the true and the estimated delay of the $s^{{\text{th}}}$ target, respectively. Similarly, the true and the estimated Doppler frequency of the $s^{{\text{th}}}$ target are denoted by $f_s$ and $\hat{f}_s$, respectively.
2. **Normalized RMS Error:** For all true detections, the normalized RMS error of estimated delays and Doppler frequencies are defined as $$e_t \ {\mbox{$\stackrel{\rm def}{=}$}}\ \sqrt{\frac{1}{T}\sum_{s=1}^{T}{\bigg(\frac{\hat{t}_s - t_s}{\delta_t}\bigg)^2}}$$ and $$e_f \ {\mbox{$\stackrel{\rm def}{=}$}}\ \sqrt{\frac{1}{T}\sum_{s=1}^{T}{\bigg(\frac{\hat{f}_s - f_s}{\delta_f}\bigg)^2}},$$ respectively, where $T$ denotes the total number of true detections.
3. **Probability of Separate Detection:** We define “*Separate Detection*” as detection of two close targets with the delay and Doppler errors less than half of their true delay and Doppler frequency spacing in the delay-Doppler plain, respectively. To be more specific, let $t_{ij} = |t_i - t_j|$ and $f_{ij} = |f_i - f_j|$ denote the true delay and Doppler frequency spacing of the $i^{{\text{th}}}$ and the $j^{{\text{th}}}$ targets, respectively. If the estimated delays and Doppler frequencies satisfy $$\forall{s \in \{i,j\}}: |\hat{t}_s - t_s| < \frac{t_{ij}}{2} \quad \text{and} \quad |\hat{f}_s - f_s| < \frac{f_{ij}}{2},$$ then we say these two targets are detected separately.
Recovery Performance
--------------------
We first study the effect of radar waveform, the number of samples and radar measurement model on recovery performance in the noisy settings. The results of this experiment are shown in [Fig. \[fig:robust\]]{}, which plots the probability of detection as a function of SNR. It can be seen that the recovery performance of sub-Nyquist methods degrades with reduction of sampling rate in radars with rectangular waveform. However, this performance degradation is compensated using modulated waveform that focuses energy around the sub-Nyquist samples. As can be seen, using DS-FCM waveform combined with DS-sampling considerably improves the performance of CS-based radar. It is clear that the modified-DF approach has better detection performance than other methods for SNRs below $-12$ dB while the structured method have better performance for higher SNRs. The standard model leads to lower performance, because this method only uses sparsity of targets while two other method use joint sparsity and focused measurements to improve recovery performance. Furthermore, standard model uses a huge dictionary including $366,464$ atoms and the processor must perform a heavy combinational search over these atoms to find all targets.
![Probability of detection for the conventional Nyquist processing and sub-Nyquist processing methods based on standard model, structured model, Doppler focusing (DF) approach [@Bar-Ilan2014], and modified-DF. Sub-Nyquist sample rate is $2\%$ and $10\%$ of Nyquist rate.[]{data-label="fig:robust"}](figures/fig_robust){width="21pc"}
![Probability of detection versus SNR for different Doppler focusing approaches. Low bandwidth-DF uses a waveform with $30$ MHz bandwidth, while other three methods use $300$ MHz waveforms.[]{data-label="fig:DFcompare"}](figures/fig_DFcomparison){width="21pc"}
The performance of Doppler focusing approach [@Bar-Ilan2014], which uses consecutive sampling and rectangular waveform, is much less than the conventional Nyquist processing method. Moreover, increasing bandwidth of radar waveform decrease energy of each Fourier samples. To solve this problem, the authors of [@Bar-Ilan2014] pass the signal through a low pass filter and readjust amplitude of the resultant signal so that the target SNR remains constant (see Fig. 8. of [@Bar-Ilan2014] and its comments). In the other word, it is proposed that only a waveform with reduced bandwidth, proportional to sample rate reduction, be transmitted. One can show that such assumption leads to less radar resolution. By the way, we compare all Doppler focusing approaches with the Nyquist processing in [Fig. \[fig:DFcompare\]]{}. It can be seen that performance of low bandwidth-DF is increased in comparison to Doppler focusing with original bandwidth waveform.
The normalized RMS error of delay and Doppler frequency estimation for high performance recovery methods are illustrated in [Fig. \[fig:RMS\]]{}. It can be seen that the structured model has less RMS error than others in Doppler recovery and RMS error of its delay recovery is similar to Modified-DF.
![Normalized RMS error of delay and Doppler frequency recovery.[]{data-label="fig:RMS"}](figures/fig_RMS){width="21pc"}
\[b\]
--------------------------------- ----------------- ---------------------
Recovery Method Number of Processing time
Samples per PRI
Conventional Nyquist Processing $3000$ $1,293\ \text{ms}$
Doppler Focusing $300$ $6,686\ \text{ms}$
Low bandwidth-DF
Standard Model $54$ $16,072\ \text{ms}$
Modified-DF $54$ $472.8\ \text{ms}$
The Structured Model $54$ $70.3\ \text{ms}$
\[1ex\]
--------------------------------- ----------------- ---------------------
: Processing times of various recovery methods.
\[tab2\]
Recovery algorithms of sub-Nyquist methods are more complicated than the conventional Nyquist processing. However, low computational complexity algorithms can achieve faster recovery than the conventional processing, because sub-Nyquist methods in these simulations use only $2\%$ of Nyquist samples. To compare the recovery speed, we list the required processing time for an “*Intel(R) Core(TM) i7 - 4790k @ 4.00GHz*” processor to perform each recovery method (see Table \[tab2\]). It is evident that by using the structured model we can achieve faster recovery time. Note that, the standard model requires large number of computations and consequently needs long time to perform recovery. Therefore, using the standard model is infeasible for high resolution radar applications, specially for long range radars. The processing time is proportional to the size of CS problem for each model as listed in TABLE \[tab0\].
Recovery Resolution
-------------------
Recall from [Section \[sec:Coherence Analysis\]]{} that we prove analytically, DS-sampling scheme leads to a delay dictionary with minimum coherence. As each atom in the delay dictionary belongs to a specific delay grid, we expect to achieve the highest possible resolution, limited by difference set order in our sub-Nyquist sampling scheme. In these simulations, we use a difference set with order $N = 2863$ to create a delay dictionary with delay grid resolution of $\tau/N = 1.048 \times \delta_t$. The probability of separate detection is a good performance metric to represent the resolution of radar in terms of the ability to separate recovery of two closely spaced targets. We assume that the received signal from these two targets have the same SNR, which is high enough for the true detection. For example, based on [Fig. \[fig:robust\]]{} considering SNR = $5$ dB for the Nyquist processing and structured model with DS-sampling is enough for the true detection of targets with high probability. In order to evaluate delay resolution, we consider the case that two close targets have Doppler frequency spacing less than or equal to a Doppler Nyquist bin such that the ability of separating two targets using different Doppler frequencies can not affect the delay resolution. As illustrated in [Fig. \[fig:res\]]{}, sub-Nyquist methods using DS-sampling can achieve better delay resolution than sub-Nyquist methods with random sampling. Furthermore, we can distinguish close targets with higher probability using sub-Nyquist methods with DS-sampling, while it uses only $2\%$ of Nyquist samples. To compare, we also represent the resolution of sub-Nyquist methods with consecutive sampling [@Bajwa2011; @Bar-Ilan2014], which uses $10\%$ and $15\%$ of Nyquist samples. As we expected from the discussion presented in [Section \[sec:Coherence Analysis\]]{}, consecutive sampling does not exploit all information about targets, specially when the targets are close together. This property of consecutive sampling contradicts with the goal of CS theory, *i.e.* sub-sampling without loosing information.
![Probability of separate detection of two close targets with different delay spacing on delay-Doppler plane. The sub-Nyquist methods, based on the proposed DS-sampling, random sampling and consecutive sampling [@Bajwa2011; @Bar-Ilan2014], are presented.[]{data-label="fig:res"}](figures/fig_res){width="21pc"}
The Effect of Number of Pulses
------------------------------
Conventional Nyquist processing methods can achieve higher recovery performance in the noisy condition by increasing the number of transmitted pulses. In this experiment, we investigate the effect of increasing the number of pulses on sub-Nyquist methods. The results of this numerical experiment are reported in [Fig. \[fig:numbpulse\]]{}. It can easily be seen from the figure that the behavior of sub-Nyquist methods are also improved in the noisy condition by increasing the number of pulses.
![Probability of detection versus SNR for different recovery methods, where the solid lines show recovery methods with $P=40$ pulses, and the dashed lines indicate recovery methods with $P=20$ pulses.[]{data-label="fig:numbpulse"}](figures/fig_numbpulse){width="21pc"}
The Effect of Number of Targets
-------------------------------
Our final numerical experiment studies the effect of increasing the number of targets on the recovery performance. [Figure \[fig:numbtarg\]]{} shows that increasing the number of targets causes degradation in the recovery performance of sub-Nyquist method. As predicted by [Lemma\[lem:DSstructured\]]{} and [Lemma\[lem:DSdopplerfocusing\]]{}, recovery performance of the Modified-DF decreases slower by an increasing number of targets in comparison to the structured model.
![Probability of detection versus the number of targets for different recovery methods, where the solid lines show recovery methods with $P=20$ pulses, and the dashed lines indicate recovery methods with $P=10$ pulses (SNR = $0$ dB).[]{data-label="fig:numbtarg"}](figures/fig_numbtarg){width="21pc"}
Conclusion {#sec:Conclusion}
==========
This paper presents a novel sub-Nyquist sampling scheme based on difference set codes, called “DS-sampling”. It is shown analytically and numerically that DS-sampling can reduce the sampling rate of wideband radars significantly without any reduction in the radar resolution. It is shown that the radar waveform has an important role in sub-Nyquist methods. We also introduce a new modulated waveform based on difference sets, called “DS-FCM” waveform, which can highly boost the recovery performance of sub-Nyquist methods in noisy conditions. The proposed structured model reduces complexity and processing time of the delay-Doppler recovery, however, its performance decreases for a large number of targets. To solve this problem, we develop a modified version of the Doppler focusing approach based on the DS-sampling and the DS-FCM waveform. Modified-DF has ability of recovering more targets at the cost of more processing time. Finally, we show that our sampling and processing method addresses the common problems of many CS-based radars.
[^1]: Notice that for an optimal sub-Nyquist radar we expect that the number of samples to be $K = O(S \log^c N/S)$, where $S$ is the number of targets and $N$ is the number of Nyquist samples. However, for a dictionary construction based on the coherence, one can show that for successful recovery $K = O(S^2)$, and our construction, in this sense, is optimal.
[^2]: The authors are with the Department of Electrical Engineering, University of Isfahan, Isfahan, Iran (e-mail: [email protected], [email protected], [email protected], [email protected]).
[^3]: This parameter is limited by the simulation running time and the availbale computer physical memory. We use this value similar to [@Bar-Ilan2014] for a fair comparison.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Electroweak corrections associated with the instability of the top quark to the next-to-next-to-leading logarithmic (NNLL) total top pair threshold cross section in $e^+e^-$ annihilation are determined. Our method is based on absorptive parts in electroweak matching conditions of the NRQCD operators and the optical theorem. The corrections lead to ultraviolet phase space divergences that have to be renormalized and lead to NLL mixing effects. Numerically, the corrections can amount to several percent and are comparable to the known NNLL QCD corrections.'
author:
- 'André H. Hoang and Christoph J. Reißer'
title: Electroweak Absorptive Parts in NRQCD Matching Conditions
---
Introduction {#sectionintroduction}
============
The line-shape scan of the threshold top pair production cross section $\sigma(e^+e^-\to \gamma^*,Z^*\to t\bar t)$ constitutes a major part of the top quark physics program at the International Linear Collider (ILC) project that is currently being initiated. Because in the Standard Model the top quark width $\Gamma_t\approx 1.5$ GeV is much larger than the typical hadronization energy $\Lambda_{\rm QCD}$, it is expected that the line-shape of the total cross section is a smooth function of the c.m.energy, and that non-perturbative effects are strongly suppressed[@Kuehn1; @Fadin1]. The determination of the top quark mass (in a threshold mass scheme [@Habilitation; @CKMworkshop]) is the most important measurement that can be obtained from the threshold scan since an uncertainty of only around 100 MeV is expected[@TTbarsim; @synopsis]. This prospect is quite robust, from the theoretical as well as from the experimental point of view, since it relies mostly on the determination of the c.m.energy where the cross section rises. Because the $t\bar t$ pair is produced predominantly in an S-wave state the rise of the cross section is quite rapid and easily measurable even in the presence of beam effects [@TTbarsim]. In addition it will also be possible to determine the strong coupling $\alpha_s$, the total top quark width $\Gamma_t$ and, if the Higgs boson is light, the top Yukawa coupling $g_{\rm tth}$. However, the latter measurements are sensitive to the form and the normalization of the line-shape. Since the observable cross section is a convolution of the theory prediction with the partly machine-dependent luminosity spectrum arising from QED effects [@TTbarsim; @Cinabro1], high demands are imposed on theoretical predictions and experimental analyses to make these measurements possible. In particular, theoretical predictions need to have a precision at the level of only a few percent.
The common theoretical tool to make computations for top threshold observables is non-relativistic QCD (NRQCD).[^1] It provides an economic and systematic treatment for the non-relativistic expansion and for QCD radiative corrections coming from high and low energies. It has also become evident that it is advantageous to use renormalization group methods [@LMR] to resum logarithms of the top quark velocity $v$ to all orders of QCD perturbation theory in order to avoid large normalization uncertainties of at least $20$ % that are obtained in fixed-order predictions [@synopsis]. Concerning QCD effects at next-to-leading-logarithmic (NLL) order for the total cross section only the $t\bar t$ production current has a non-trivial running, which is fully known[@LMR; @HoangStewartultra; @Manohar1; @Pineda1]. At next-to-next-to-leading-logarithmic (NNLL) order for the total cross section the QCD evolution of almost all required couplings is known [@HoangStewartultra; @Pineda2; @Hoang3loop; @Peninc1] except for missing subleading mixing effects in the running of the heavy quark pair production current. Theoretical analyses for the total cross section at NNLL order were given in [@hmst; @hmst1; @HoangEpiphany]. Currently the normalization uncertainties of the total cross section from QCD effects are estimated to be around $6\%$ [@HoangEpiphany].
While the major focus in recent work was directed on a better understanding of QCD corrections at NNLL order, a systematic treatment of electroweak effects at the same level has not yet been accomplished. Electroweak corrections are responsible for a variety of different physical effects. At leading order, the three basic electroweak effects are the $e^+e^-$ annihilation process that leads to top pair production by virtual photon and Z exchange, the finite top lifetime, which can be implemented by the replacement rule $E\to E+i\Gamma_t$ [@Fadin1] ($E=\sqrt{s}-2m_t$ being the c.m.energy with respect to the top threshold and $\Gamma_t$ being the on-shell top quark width), and the luminosity spectrum mentioned above. All three effects have so far been treated independently. It has become the convention that only the first two effects are included into theoretical predictions while the luminosity spectrum is accounted for in the experimental simulations [@TTbarsim].
At the subleading level a coherent treatment of electroweak effects has not yet been achieved, although previous partial analyses have indicated that they can reach the level of a few percent [@GuthKuehn; @HoangTeubnerdist]. It is also evident that at the subleading level an independent treatment of various different electroweak effects will, eventually, be impossible. Since a systematic treatment relies on the consistent separation of off-shell (non-resonant) and close-to-mass-shell (resonant) fluctuations, the concept of effective theories appears again to be a highly efficient tool to make progress [@Beneke1]. For the $t\bar t$ threshold such a framework is already provided by the effective theory NRQCD itself, since it automatically achieves an expansion in the off-shellness of the top quark through the non-relativistic expansion in $v$. The NRQCD effective theory formalism can be extended to account for electroweak corrections. For example, including the electroweak radiative corrections to the top quark two-point function in the NRQCD matching conditions one obtains the additional heavy quark bilinear terms[^2] $$\begin{aligned}
\label{Gtop}
\delta {\cal L} = \sum_{{\mbox{\scriptsize\boldmath $p$}}} \psip{{\mbox{\scriptsize\boldmath $p$}}}^\dagger \: \frac{i}{2} \Gamma_t\:
\psip{{\mbox{\scriptsize\boldmath $p$}}} \ +\
\sum_{{\mbox{\scriptsize\boldmath $p$}}} \chip{{\mbox{\scriptsize\boldmath $p$}}}^\dagger\: \frac{i}{2} \Gamma_t\: \chip{{\mbox{\scriptsize\boldmath $p$}}} \,\end{aligned}$$ in the effective Lagrangian, where $\psip{p}$ and $\chip{p}$ represent Pauli spinor field operators destroying top and antitop quarks, respectively. For simplicity color indices are suppressed throughout this paper. The terms in Eq.(\[Gtop\]) reproduce the replacement rule $E\to E+i\Gamma_t$. They render the effective theory Lagrangian non-hermitian because they describe an absorptive on-shell process that has been integrated out from the theory. Nevertheless, they allow for a correct determination of the total cross section from the forward scattering amplitude using the optical theorem and the unitarity of the underlying theory. In fact, this issue is analogous to the so-called strong phases in QCD amplitudes that are the basis for the search for CKM CP-violation in a number of B meson decays [@BaBarBook].
In this paper we extend this approach and investigate the role of absorptive parts related to the top quark decay in electroweak loop corrections to the NRQCD matching conditions that contribute to the NNLL total cross section.[^3] For simplicity we neglect the bottom quark mass and set $V_{tb}=1$, and approximate the W boson and bottom quark as stable particles. We demonstrate that these electroweak corrections properly account for the interference of the dominant double-resonant process $e^+e^-\to t\bar
t\to bW^+\bar b W^-$ with the $v^2$-suppressed single-resonant amplitudes for $e^+e^-\to b W^+\bar t\to bW^+\bar b W^-$ and $e^+e^-\to t\bar b W^-\to bW^+\bar b W^-$. We also show that absorptive parts that do not contribute to the $bW^+\bar b W^-$ final state, and are therefore also not accounted for in line-shape measurements, can be excluded in a gauge invariant way. This requires to include also fields for the electrons and positrons from the initial state into the effective theory, which act like classic fields for QCD interactions. The new corrections show interesting features. They slightly modify the form of the cross section line-shape and lead to UV phase space divergences that are directly related to the fact that the top quarks are unstable. The divergences lead to anomalous dimensions of $(e^+e^-)(e^+e^-)$ operators that also contribute to the absorptive part of the forward scattering amplitude. In the total cross section these mixing effects contribute at NLL order and represent a novel NLL effect that has remained unnoticed in previous work. Numerically, the size of the new corrections ranges up to 5% and partly compensates for the large QCD corrections found recently in [@Hoang3loop].
The program of this paper is as follows. In Sec.\[sectionpowercounting\] we discuss the power counting for the total cross section with respect to electroweak effects associated with the top quark decay. We introduce the $(e^+e^-)(t\,\bar t)$ effective theory operators needed to account for the electroweak absorptive parts that arise in top pair production or annihilation. In particular, we show that up to NNLL order there are no contributions from interference effects originating from (ultrasoft) gluons carrying momenta of order $m_t v^2$. In Sec.\[sectionmatchingconditions\] the electroweak absorptive parts of the matching conditions for the $(e^+e^-)(t \,\bar t)$ effective theory operators relevant for the $bW^+\bar bW^-$ final state are computed. In Sec.\[sectionrenormalization\] the resulting NNLL corrections for the total cross section are determined and the renormalization of the $(e^+e^-)(e^+e^-)$ effective theory operators needed to account for the phase space divergences is discussed. In particular, we compute and solve the anomalous dimensions and determine their contribution to the total cross section. Sec.\[sectionanalysis\] contains a brief numerical analysis and in Sec.\[sectionconclusion\] we conclude.
Power Counting and Matching Conditions {#sectionpowercounting}
======================================
In this work we are interested in the total cross section and not in any differential information on the top decay final states. We therefore include all effects related to the top quark decay as non-hermitian matching conditions of effective theory operators that describe the non-relativistic top and antitop dynamics and their interactions with soft and ultrasoft gluons. We employ gauge invariant operators, and the matching conditions are computed for on-shell external lines. This allows to maintain gauge invariance in a transparent way.
To illustrate the power counting needed to classify the order at which these electroweak effects can contribute let us recall the matching conditions for the bilinear quark field operators. They are obtained by matching top or antitop 2-point functions in the effective theory to those in the full electroweak and QCD theory. The result up to NNLL order reads $$\begin{aligned}
\label{Lke}
{\mathcal L}_{\rm bilinear}(x) &=& \sum_{{\mbox{\scriptsize\boldmath $p$}}}
\psip{{\mbox{\scriptsize\boldmath $p$}}}^\dagger(x) \biggl\{ i D^0 - {(\itbf{p}-i{\mathbf D})^2 \over 2 m_t}
+\frac{{\itbf{p}}^4}{8m_t^3}
+ \frac{i}{2} \Gamma_t \bigg( 1 - \frac{{\itbf{p}}^2}{2 m_t^2} \bigg)
- \delta m_t \biggr\} \psip{{\mbox{\scriptsize\boldmath $p$}}}(x)
\nonumber\\ & & \qquad
+ (\psip{{\mbox{\scriptsize\boldmath $p$}}} \to\chip{{\mbox{\scriptsize\boldmath $p$}}})\,, \end{aligned}$$ and includes the terms shown in Eq.(\[Gtop\]). Here, the fields $\psip{{\mbox{\scriptsize\boldmath $p$}}}$ and $\chip{{\mbox{\scriptsize\boldmath $p$}}}$ destroy top and antitop quarks with momentum ${\itbf{p}}$, $D^\mu=(D^0,-{\mathbf D})=\partial^\mu + i g A^\mu$ is the ultrasoft gauge covariant derivative and $\Gamma_t$ is the top quark width defined at the top quark pole. At order $g^2$, $g$ being the SU(2) gauge coupling, the width has the form $$\begin{aligned}
\label{Gamma}
\Gamma_t & = &
\frac{\alpha |V_{tb}|^2 m_t}{16 s_w^2 x}\,(1-x)^2(1+2x)
\,,\end{aligned}$$ where $s_w$ ($c_w$) is the sine (cosine) of the weak mixing angle, $\alpha$ the fine structure constant and $$\begin{aligned}
x\equiv \frac{M_W^2}{m_t^2}
\,.\end{aligned}$$ We use the usual $v$-counting $D^0\sim m_t v^2\sim\Gamma_t\sim m_t g^2$, which leads to the scaling relation $$\begin{aligned}
v\,\sim\,\alpha_s\,\sim\,g\,\sim\,g^\prime\end{aligned}$$ for the SU(2) and U(1) gauge couplings $g$ and $g^\prime$. Because the weak mixing is of order one we apply the same counting to the SU(2) and U(1) gauge couplings. The term $\delta m_t$ is a residual mass term of order $v^2$ that arises if a threshold mass scheme [@synopsis] is used. The LL order terms in Eq.(\[Lke\]) lead to the top/antitop propagator $$\begin{aligned}
\frac{i}{p^0 - {\itbf{p}}^2/(2m_t) + i\Gamma_t/2 -\delta m_t} \,.
\label{toppropagator}\end{aligned}$$ Although the exchange of time-like ultrasoft $A^0$ gluons contributes at LL order according to the $v$-counting, their contribution at LL order can be removed from the particle-antiparticle sector of the theory by a redefinition of the top and antitop fields related to static Wilson lines [@Korchemsky1; @Bauer1].[^4] The time dilatation term $\propto
\Gamma_t({\mathbf p}^2/2m_t^2)$ originates from the momentum-dependence of the full theory spinors and contributes at NNLL order.
The $t\bar t$ pair is produced by an electroweak process. As long as electroweak effects are only treated at leading order in the $v$-expansion it is sufficient to describe $t\bar t$ production by a bilinear quark-antiquark current. However, as shown below it is necessary to include the initial-state $(e^+e^-)$ fields to ensure electroweak gauge invariance at subleading order in the $v$-expansion. The dominant operators that have to be used describing $t\bar t$ spin-triplet production have the form $$\begin{aligned}
{\cal O}_{V,{\mbox{\scriptsize\boldmath $p$}}} & = &
\big[\,\bar e\,\gamma_j\,e\,\big]\,{\cal O}_{{\mbox{\scriptsize\boldmath $p$}},1}^j
\,,
\\[2mm]
{\cal O}_{A,{\mbox{\scriptsize\boldmath $p$}}} & = &
\big[\,\bar e\,\gamma_j\,\gamma_5\,e\,\big]\,{\cal O}_{{\mbox{\scriptsize\boldmath $p$}},1}^j
\,,\end{aligned}$$ where $$\begin{aligned}
{\cal O}_{{\mbox{\scriptsize\boldmath $p$}},1}^j & = &
\Big[\,\psi_{{\mbox{\scriptsize\boldmath $p$}}}^\dagger\, \sigma_j (i\sigma_2)
\chi_{-{\mbox{\scriptsize\boldmath $p$}}}^*\,\Big]
\,.\end{aligned}$$ They give the contribution $\Delta {\cal L} = \sum_{{\mbox{\scriptsize\boldmath $p$}}} \left(C_V {\cal O}_{V,{\mbox{\scriptsize\boldmath $p$}}} + C_A
{\cal O}_{A,{\mbox{\scriptsize\boldmath $p$}}} \right) + \mbox{h.c.}$ to the effective theory Lagrangian where the hermitian conjugation is referring to the operators only. The index $j=1,2,3$ is summed. The corresponding operators describing $t\bar t$ annihilation are obtained by the hermitian conjugation. Since in this work we only focus on the electroweak effects related to the top quark decay and, in particular, neglect QED radiative corrections (including QED binding and the beam effects mentioned above) the electron and positron fields act like classic fields in the effective theory.[^5] In a more complete treatment of electroweak effects, however, their interactions with photons have to be accounted for [@Beneke1]. The leading order matching conditions of the operators ${\cal O}_{V/A,{\mbox{\scriptsize\boldmath $p$}}}$ obtained from the full theory Born diagrams with photon and Z exchange are of order $g^2$ and read $$\begin{aligned}
C_V^{\rm born}(\nu=1) & = &
\frac{\alpha\pi}{m_t^2(4c_w^2-x)}\,\bigg[\,
Q_e Q_t(4-x) + Q_t-Q_e-\frac{1}{4s_w^2}
\,\bigg]
\,,
\label{Cvborn}
\\[2mm]
C_A^{\rm born}(\nu=1) & = &
-\,\frac{\alpha\pi}{m_t^2(4c_w^2-x)}\,\bigg[\,
Q_t-\frac{1}{4s_w^2}
\,\bigg]
\,,
\label{Caborn}\end{aligned}$$ where $\nu$ is the vNRQCD renormalization scaling parameter.[^6] Except for the QED beam effects, which we will not consider here, the electroweak effects in Eqs.(\[Gamma\],\[Cvborn\],\[Caborn\]) are the only ones at leading order. In particular, at this order there are no electroweak effects contributing to the Coulomb potential acting between the $t\bar t$ pair, $$\begin{aligned}
{\cal L}_{\rm pot} & = &
-\sum_{{\mbox{\scriptsize\boldmath $p$}},{\mbox{\scriptsize\boldmath $p$}}^\prime} \frac{{\cal V}_c^{(s)}(\nu)}{({\mathbf p}-{\mathbf p}^\prime)^2}\,
\psip{{\mbox{\scriptsize\boldmath $p$}}^\prime}^\dagger \psip{{\mbox{\scriptsize\boldmath $p$}}} \chip{-{\mbox{\scriptsize\boldmath $p$}}^\prime}^\dagger\chip{-{\mbox{\scriptsize\boldmath $p$}}}
\,,
\label{Lpot}\end{aligned}$$ where ${\cal V}_c^{(s)}(\nu)=-4\pi C_F \alpha_s(m_t\nu)$ is the LL Coulomb Wilson coefficient for a color singlet heavy quark pair.
For the NLL order approximation it has been frequently stated that there are no new operators that can contribute and that power counting tells that we only need to consider ${\cal O}(\alpha_s)$ QCD corrections to the LL matching conditions in Eq. (\[Lke\]) to account for all electroweak effects [@Fadin2; @Melnikov1]. However, it was also noted in [@HoangTeubnerdist] that the mismatch between the $t\bar t$ phase space in the full and the effective theory leads to additional NLL matching corrections for unstable top quarks. We come back to the role played by these specific NLL contributions in Sec.\[sectionrenormalization\] and ignore them for the following considerations. Thus, concerning NLL effects related to the top decay only the one-loop QCD corrections to the on-shell top decay width have to be accounted for [@Jezabek1]. In particular, there are no QCD interference effects from gluon radiation off the top/antitop quark or its decay products[@Fadin2; @Melnikov1]. In our approach, which only aims at the total cross section, one can show that such QCD interference effects do not even contribute at NNLL order in the non-relativistic expansion.
To discuss the QCD interference contributions we have to consider ultrasoft gluons, which carry momenta of order $m_t v^2\sim\Gamma_t$, because they can interact with a resonant top quark without kicking it off-shell. For the time-like $A^0$ gluons we already mentioned that their leading interaction with the quarks can be removed by a field redefinition related to static Wilson lines [@Korchemsky1; @Bauer1]. Moreover, QCD gauge invariance ensures that the dominant electroweak matching corrections to the $A^0$ interaction vertex vanish because we can set the ultrasoft gluon momentum to zero. Radiative corrections can, however, generate an anomalous interaction in analogy to the $g-2$ in QED. Yet this interaction is suppressed by a factor $1/m_t^2$ and cannot contribute to the cross section matrix elements at NNLL level without even having accounted for additional powers of the coupling constants. It remains to discuss the space-like ultrasoft ${\mathbf A}$ gluons which couple to the quarks with the ${\mathbf p}.{\mathbf A}/m_t$ coupling. In pure QCD space-like ultrasoft gluon exchange contributes to the cross section matrix elements at NNLL order and also to the renormalization group running of the $t\bar t$ production operators at the NLL level [@LMR] through operator mixing. Accounting for the $g^2$-suppression from an additional electroweak loop correction to the interaction vertex then also leads to a contribution beyond the NNLL order.
At NNLL order let us first consider whether one needs to account for any operator in addition to those present already at the LL level. Ultrasoft gluon interactions have been discussed above. Soft gluon operators first contribute at the NLL level in pure QCD (for example as corrections to the Coulomb potential [@amis; @amis2]), so $g^2$ electroweak corrections to these operators are beyond NNLL order. Such corrections can also not contribute at NNLL order through mixing since the Coulomb potential does not cause UV divergences. It remains to discuss four-quark operators. Because an electroweak loop would require a factor $g^4$ in addition to the $1/m_t^2$ suppression for dimensional reasons, such an operator could not contribute at NNLL order either [@Beneke2].
It remains to discuss $g^2$ corrections to the operators contributing at the LL level. For the bilinear quark operators in the effective theory Lagrangian one obtains the terms shown in Eq.(\[Lke\]). Concerning the instability of the top quark only the time dilatation correction is obtained, and the ${\cal O}(\alpha_s^2)$ and one-loop electroweak corrections to the on-shell top quark width [@Blokland1; @Denner1] have to be accounted for. The ${\cal O}(\alpha_s^2)$ correction to the top width is easy to implement together with the Born and one-loop QCD results in the width term $\Gamma_t$ and will not be discussed further in this work. On the other hand, for the Coulomb potential all dominant $g^2$ corrections cancel due to SU(3) gauge invariance because in the first approximation one can neglect the gluon momentum flowing into the vertex correction [@Modritsch1; @WiseTTbarSusy]. The mechanism is equivalent to the gauge cancellation discussed above for the time-like $A^0$ gluon. The order $g^4$ matching conditions of the production operators ${\cal O}_{V,{\mbox{\scriptsize\boldmath $p$}}}$ and ${\cal O}_{A,{\mbox{\scriptsize\boldmath $p$}}}$, on the other hand, do not cancel and have to be determined from matching to the one-loop Standard Model amplitudes for the process $e^+e^-\to\gamma,Z\to t\bar t$. At this order the matching computation can be carried out for an on-shell top-antitop pair at rest. The full set of one-loop electroweak corrections was determined in Ref.[@GuthKuehn]. For the examinations in this work we have to account only for the $bW^+$ and $\bar b W^-$ cuts because, as shown in Sec.\[sectionrenormalization\], only these cuts are relevant for the $bW^+\bar b W^-$ final state that can interfere with top pair production. Because in Ref.[@GuthKuehn] only the sum of all contributions was presented, including the $b\bar b$ and $W^+W^-$ cuts, we rederive the results for the $bW^+$ and $\bar b W^-$ cuts in the next section.
Absorptive Matching Conditions {#sectionmatchingconditions}
==============================
The top pair production diagrams in the full theory that need to be considered to determine the absorptive $bW^+$ and $\bar b W^-$ cuts are shown in Fig.\[figcuts\]. The external (on-shell) top quarks can be taken to be at rest.
![ Full theory diagrams in Feynman gauge that have to be considered to determine the electroweak absorptive parts in the Wilson coefficients $C_A$ and $C_V$ related to the physical $bW^+$ and $\bar b W^-$ intermediate states. Only the $bW^+$ cut is drawn explicitly. \[figcuts\] ](figures/fig1.ps){width="14cm"}
The results for the cuts in the full theory amplitude have the form $$\begin{aligned}
{\cal A} & = &
i\,\Big[\,\bar v_{e^+}(k^\prime)\,\gamma^\mu
(i C_V^{\rm bW,abs}+i C_A^{\rm bW,abs}\,\gamma_5)\, u_{e^-}(k)\,\Big]\,
\Big[\,\bar u_t(p)\,\gamma_\mu\,v_{\bar t}(p)\,\Big]
\,,
\label{eett}\end{aligned}$$ where $k+k^\prime=2p=(2m_t,0)$ and $$\begin{aligned}
i C_V^{\rm bW,abs} & = &
-i\,\frac{\alpha^2 \pi |V_{tb}|^2}{12 m_t^2 s_w^2 x(4c_w^2-x)(1+x)}\,
\bigg[\,
\frac{3x(1+x)}{(1-x)}\bigg(1+\frac{x-4}{4s_w^2}\bigg)\,\ln\Big(\frac{2-x}{x}\Big)
\nonumber\\[2mm] & &
+\,Q_e Q_t (1-x)(4-x)(1+2x)(1+x+x^2)
\nonumber\\[3mm] & &
+\, Q_e(x-1)(1+4x+2x^2+2x^3)
\, +\, Q_t(1-x)(1+2x)(1+x+x^2)
\nonumber\\[2mm] & &
-\,\frac{1}{2}(1+12x+9x^2+2x^3)
\,+\,\frac{1}{8s_w^2}(2+41x+28x^2-x^3+2x^4)
\,\bigg]
\,,
\nonumber\\[3mm]
i C_A^{\rm bW,abs} & = &
i\,\frac{\alpha^2 \pi |V_{tb}|^2}{12 m_t^2 s_w^2 x(4c_w^2-x)(1+x)}\,
\bigg[\,
\frac{3x(1+x)}{(1-x)}\bigg(1+\frac{x-4}{4s_w^2}\bigg)\,\ln\Big(\frac{2-x}{x}\Big)
\nonumber\\[3mm] & &
+\, Q_t(1-x)(1+2x)(1+x+x^2)
\nonumber\\[2mm] & &
-\, \frac{1}{2}(1+12x+9x^2+2x^3)
\,+\,\frac{1}{8s_w^2}(2+41x+28x^2-x^3+2x^4)
\,\bigg]
\,.\end{aligned}$$ The results for the charge conjugated process describing top pair annihilation, on the other hand, read $$\begin{aligned}
\bar{\cal A} & = &
i\,\Big[\,\bar u_{e^-}(k)\,\gamma^\mu
(i C_V^{\rm bW,abs}+i C_A^{\rm bW,abs}\,\gamma_5)\, v_{e^+}(k^\prime)\,\Big]\,
\Big[\,\bar v_{\bar t}(p)\,\gamma_\mu\,u_t(p)\,\Big]
\,.
\label{eettadj}\end{aligned}$$ We used the cutting equations to obtain the result and checked electroweak gauge invariance by carrying out the computation in unitary and Feynman gauge. In both cases we computed the cut $W$ lines with physical polarizations as well as with unphysical ones including also the charged Goldstone exchange. We note that the contributions that arise from off-shell corrections in the top self-energy graphs are necessary for electroweak gauge invariance. Since the $b\bar b$ and $W^+W^-$ cuts lead to different distinct phase space factors, we found that it is possible to identify the results also from the formulae given in [@GuthKuehn].
It is an important fact that the sign of the imaginary part of the amplitude does not change in the charge conjugated amplitude. As for the quark field bilinear terms discussed before in Eq.(\[Lke\]) this is related to the unitarity of the underlying theory. It is straightforward to match the amplitudes for the operators ${\cal O}_{V/A,{\mbox{\scriptsize\boldmath $p$}}}$ and ${\cal O}^\dagger_{V/A,{\mbox{\scriptsize\boldmath $p$}}}$ to the full theory results in Eqs.(\[eett\]) and (\[eettadj\]). The resulting matching conditions at the hard scale read $$\begin{aligned}
C_V(\nu=1) & = & C_V^{\rm born} + i\,C_V^{\rm bW,abs}
\,,
\nonumber
\\
C_A(\nu=1) & = & C_A^{\rm born} + i\,C_A^{\rm bW,abs}
\,,
\label{Cvaabs}\end{aligned}$$ where we have included also the Born level contributions from Eqs.(\[Cvborn\]) and (\[Caborn\]). We emphasize again that these matching conditions are valid for the operators ${\cal O}_{V/A,{\mbox{\scriptsize\boldmath $p$}}}$ and ${\cal O}^\dagger_{V/A,{\mbox{\scriptsize\boldmath $p$}}}$. In a full treatment of electroweak and QCD effects the coefficients $C_{V/A}$ also include the real parts of the full set of electroweak one-loop diagrams indicated in Figs.\[figcuts\] and the QCD matching corrections known from previous work [@hmst1; @c12loop; @Hoangc12loopQED]. These corrections lead to an energy-independent multiplicative modification of the cross section normalization which is, however, not subject of the investigations in this work. The results for the real parts of the full set of electroweak one-loop diagrams were given in [@GuthKuehn].
Time-Ordered Product and Renormalization {#sectionrenormalization}
========================================
Using the optical theorem the NNLL order corrections to the total cross section that come from the absorptive one-loop electroweak matching conditions for the operators ${\cal O}_{V/A,{\mbox{\scriptsize\boldmath $p$}}}$ and from the time dilatation corrections can be computed from the imaginary part of the $(e^+e^-)(e^+e^-)$ forward scattering amplitude, $$\begin{aligned}
\sigma_{\rm tot} & \sim &
\frac{1}{s}\,
\mbox{Im}\left[\,\Big(C_V^2(\nu)+C_A^2(\nu)\Big)\,L^{lk}\,{\cal A}_1^{lk}\,\right]
\,,\end{aligned}$$ where ($k+k^\prime=(\sqrt{s},0)$ and ${\bf \hat e}={\mathbf k}/|{\mathbf k}|$) $$\begin{aligned}
L^{lk} &=&
\frac{1}{4}\,\sum\limits_{e^\pm\rm spins}\,
\Big[\,\bar v_{e^+}(k^\prime)\,\gamma^l\,(\gamma_5)\,u_{e^-}(k)\,\Big]
\,\Big[\,\bar u_{e^-}(k)\,\gamma^k\,(\gamma_5)\,v_{e^+}(k^\prime)\,\Big]
\nonumber
\\[2mm]
&=& \frac{1}{2}\,(k+k^\prime)^2\,(\delta^{lk}-\hat e^l \hat e^k)\end{aligned}$$ is the spin-averaged lepton tensor and ($\hat{q}\equiv(\sqrt{s}-2m_t,0)$) $$\begin{aligned}
{\cal A}_1^{lk} &=&
i\, \sum\limits_{\mbox{\scriptsize\boldmath $p$},\mbox{\scriptsize\boldmath $p'$}} \int\! d^4x\:
e^{-i\hat q \cdot x}\:
\Big\langle\, 0\,\Big|\,T\,
{{\cal O}_{\mbox{\scriptsize\boldmath $p$},1}^l}^{\!\!\!\dagger} (0)\,
{\cal O}_{\mbox{\scriptsize\boldmath $p'$},1}^k (x)\Big|\,0\,\Big\rangle
\nonumber\\[2mm] & = &
2\,N_c\,\delta^{lk}\,G^0(a,v,m_t,\nu)\end{aligned}$$ is the time-ordered product of the $t\bar t$ production and annihilation operators ${\cal O}_{{\mbox{\scriptsize\boldmath $p$}},1}^j$ and ${{\cal
O}_{{\mbox{\scriptsize\boldmath $p$}},1}^j}^{\!\!\!\dagger}$ [@hmst1]. In dimensional regularization the result reads $$\begin{aligned}
\Delta\sigma_{\rm tot}^{\Gamma,1} & = &
2\,N_c\,\mbox{Im}\bigg\{\,
2i\,\big[\,C_V^{\rm born}\,C_V^{\rm bW,abs}+C_A^{\rm
born}\,C_A^{\rm bW,abs}\,\big]\,G^0(a,v,m_t,\nu)
\nonumber\\[2mm] & &
+\,\big[\,(C_V^{\rm born})^2+(C_A^{\rm born})^2\,\big]\,
\delta G^0_\Gamma(a,v,m_t,\nu)
\,\bigg\}
\,,
\label{dsignaNNLL}\end{aligned}$$ where $a\equiv -{\cal V}_c^{(s)}(\nu)/4\pi=C_F\alpha_s(m_t\nu)$. The term $G^0$ is the zero-distance S-wave Green function of the non-relativistic Schrödinger equation which is obtained from the LL order terms in the Lagrangian shown in Eqs.(\[Lke\]) and (\[Lpot\]). In dimensional regularization it has the form [@hmst1] $$\begin{aligned}
G^0(a,v,m_t,\nu) & = &
\frac{m_t^2}{4\pi}\left\{\,
i\,v - a\left[\,\ln\left(\frac{-i\,v}{\nu}\right)
-\frac{1}{2}+\ln 2+\gamma_E+\psi\left(1\!-\!\frac{i\,a}{2\,v}\right)\,\right]
\,\right\}
\nonumber \\ & &
+\,\frac{m_t^2\,a}{4 \pi}\,\,\frac{1}{4\,\epsilon}
\,,
\label{deltaGCoul}\end{aligned}$$ where $$\begin{aligned}
v & = &
\sqrt{\frac{\sqrt{s}-2(m_t+\delta m_t)+i\Gamma_t}{m_t}}
\,,\end{aligned}$$ $\sqrt{s}$ being the c.m.energy. The term $\delta G^0_\Gamma$ represents the corrections originating from the time dilatation correction in Eq.(\[Lke\]) and reads $$\begin{aligned}
\delta G^0_\Gamma(a,v,m_t,\nu) & = &
-i\,\frac{\Gamma_t}{2m_t}\,\bigg[\,
1+\frac{v}{2}\frac{\partial}{\partial v} + a\frac{\partial}{\partial a}
\,\bigg]\,G^0(a,v,m_t,\nu)
\,.
\label{deltaGGamma}\end{aligned}$$ Note that the Wilson coefficients $C_{V/A}$ do not run at LL order, so only the matching conditions at $\nu=1$ appear in Eq.(\[dsignaNNLL\]).
It is straightforward to check that the terms proportional to $C_{V/A}^{\rm bW,abs}$ in Eq.(\[dsignaNNLL\]) are in agreement with the full theory matrix elements from the interference between the double-resonant amplitudes for the process $e^+e^-\to t\bar t\to bW^+\bar b W^-$ (Fig.\[figinterfer\]a) and the single-resonant amplitudes describing the processes $e^+e^-\to t+\bar b W^-\to
bW^+\bar b W^- $ and $e^+e^-\to bW^+\bar t\to bW^+\bar b W^-$ (Figs.\[figinterfer\]b-i) in the $t\bar t$ threshold limit for $m_t\to\infty$.
![ Full theory Feynman diagrams describing the process $e^+e^-\to bW^+\bar b W^-$ with one or two intermediate top or antitop quark propagators. The circle in diagram (a) represents the QCD form factors for the $t\bar t$ vector and axial-vector currents. \[figinterfer\] ](figures/fig2.ps){width="14cm"}
Note that diagram (a) dominates in the non-relativistic limit due to two resonant top/antitop lines, while diagrams (b-i) are $v^2$-suppressed having only one resonant top/antitop line. Diagram (a) also contains a subleading $v^2$-suppressed contribution that has to be accounted for. Diagrams with no top/antitop line are suppressed by $v^4$ and do not need to be considered. This also means that pure background diagrams containing no intermediate top quark can be neglected at this order. We also note that to find literal agreement between full and effective theory matrix elements one has to replace the $i\epsilon$ terms in the resonant full theory top propagators by the Breit-Wigner term $i m_t\Gamma_t/2$. The circle shown in Fig.\[figinterfer\]a represents the QCD form factors for the $t \bar t$ vector and axial-vector currents. In the non-relativistic limit they reduce to the insertions of Coulomb potentials described by the higher order terms in Eq.(\[deltaGCoul\]). Due to the cancellation of the QCD interference effects caused by gluons with ultrasoft momenta there are no further QCD corrections in the non-relativistic limit.
An interesting new conceptual aspect of the corrections shown in Eq.(\[dsignaNNLL\]) is that they have UV $1/\epsilon$-divergences that arise from a logarithmic high energy behavior of the top-antitop effective theory phase space integration for matrix elements containing a single insertion of the Coulomb potential. In the forward scattering amplitude these UV divergences arise because the imaginary parts of the matching conditions of Eqs.(\[Cvaabs\]) lead to a dependence on the real part of $G^0$ (see Eq. (\[deltaGCoul\])). In the full theory this logarithmic behavior is regularized by the top quark mass. While phase space logarithms are known in the literature and can be resummed with renormalization group techniques [@Bauer2], the divergences here are specific since they would not exist if the top quark were approximated as being stable. In particular, the UV divergences from the time dilatation corrections arise from the Breit-Wigner-type high energy behavior of the effective theory top propagator in Eq.(\[toppropagator\]) which differs from the one for a stable particle. Likewise, the interference effects described by the absorptive electroweak matching conditions for the operators ${\cal O}_{V,{\mbox{\scriptsize\boldmath $p$}}}$ and ${\cal O}_{A,{\mbox{\scriptsize\boldmath $p$}}}$ would not have to be taken into account if the top quarks were stable particles. UV divergences of the same kind have already been observed and described before in the NNLL relativistic corrections to the S-wave zero-distance Green function if the unstable propagator in Eq.(\[toppropagator\]) is used [@hmst1; @HoangTeubnerdist; @HoangTeubner]. For the P-wave zero-distance Green function, which is generated by $t\bar t$ production through an axial-vector current from the Z-exchange and that contributes only at NNLL order, a similar UV divergence arises already at leading order in the non-relativistic expansion. Like for the case of the time dilatation corrections these divergences originate from the modified high energy behavior of the unstable top propagator. We believe it is evident that these divergences do not represent a deficiency of the effective theory, because the concept of separating resonant and non-resonant fluctuations appears to be the only practical way to make systematic predictions involving unstable particles. Thus these UV divergences should be handled with the renormalization techniques known from effective theories for stable particles. The only difference is that the renormalization procedure will involve operators having non-hermitian Wilson coefficients.
The operators that are renormalized by the UV divergences displayed in Eq.(\[dsignaNNLL\]) are the two $(e^+e^-)(e^+e^-)$ operators $$\begin{aligned}
\tilde{\cal O}_V & = &
-\,\big[\,\bar e\,\gamma^\mu\,e\,\big]\,
\big[\,\bar e\,\gamma_\mu\,e\,\big]
\,,
\\[2mm]
\tilde{\cal O}_A & = &
-\,\big[\,\bar e\,\gamma^\mu\,\gamma_5 \, e\,\big]\,
\big[\,\bar e\,\gamma_\mu\,\gamma_5\,e\,\big]
\,,\end{aligned}$$ which give the additional contribution $\tilde\Delta {\cal L} = \tilde C_V \tilde{\cal O}_V + \tilde C_A
\tilde{\cal O}_A$ to the effective theory Lagrangian, $\tilde C_{V/A}$ being the Wilson coefficients. Because in this work we neglect QED effects, the electron and positron act as classic fields and therefore $\tilde C_{V}$ and $\tilde
C_{A}$ run only through mixing due to UV divergences such as in Eq.(\[dsignaNNLL\]). Since only the imaginary parts of the coefficients $\tilde
C_{V/A}$ can contribute to the total cross section through the optical theorem we neglect the real contributions in the following. Using the standard $\msb$ subtraction procedure the (non-hermitian) counterterms of the renormalized $\tilde{\cal O}_{V/A}$ operators read $$\begin{aligned}
\delta \tilde C_V & = &
i\,\frac{N_c m_t^2}{32 \pi^2 \epsilon}\,
\bigg[ (C_V^{\rm born})^2 \frac{\Gamma_t}{m_t}
+2 C_V^{\rm born} C_V^{\rm bW,abs}
\bigg]\,
{\cal V}_c^{(s)}(\nu)
\nonumber \\[2mm] & &
+\,i\,\frac{N_c m_t^2}{32 \pi^2 \epsilon}\,
(C_V^{\rm born})^2 \frac{\Gamma_t}{m_t} \,
\Big[ \Big( 2c_2(\nu) -1\Big){\cal V}_c^{(s)}(\nu) + {\cal V}_r^{(s)}(\nu) \Big]
\nonumber \\[2mm] & &
+\,i\,\frac{N_c m_t^2}{48 \pi^2 \epsilon}\,
(C_V^{\rm ax})^2 \frac{\Gamma_t}{m_t} \,{\cal V}_c^{(s)}(\nu)
\,,
\nonumber\\[4mm]
\delta \tilde C_A & = &
i\,\frac{N_c m_t^2}{32 \pi^2 \epsilon}\,
\bigg[ (C_A^{\rm born})^2 \frac{\Gamma_t}{m_t}
+2 C_A^{\rm born} C_A^{\rm bW,abs}
\bigg]\,
{\cal V}_c^{(s)}(\nu)
\nonumber \\[2mm]& &
+\,i\,\frac{N_c m_t^2}{32 \pi^2 \epsilon}\,
(C_A^{\rm born})^2 \frac{\Gamma_t}{m_t} \,
\Big[ \Big( 2c_2(\nu) -1\Big){\cal V}_c^{(s)}(\nu) + {\cal V}_r^{(s)}(\nu) \Big]
\nonumber \\[2mm] & &
+\,i\,\frac{N_c m_t^2}{48 \pi^2 \epsilon}\,
(C_A^{\rm ax})^2 \frac{\Gamma_t}{m_t} \,{\cal V}_c^{(s)}(\nu)
\,,
\label{Otildecounter}\end{aligned}$$ where the respective first term on the RHS’s subtract the $1/\epsilon$ divergences shown in Eq.(\[dsignaNNLL\]) and the other terms account for the UV divergences in the P-wave Green function and the NNLL corrections of the S-wave Green function computed in Ref. [@hmst1]. Here, $C_{V/A}^{\rm ax}$ are Born level Wilson coefficients of operators describing top pair production in a P-wave (originating from pure Z exchange), ${\cal V}_r^{(s)}$ is the color singlet coefficient of the potential $({\mathbf p}^2+{\mathbf p}^{\prime\,2})/(2m_t({\mathbf p}-{\mathbf p}^\prime)^2)$ [@HoangStewartultra] and $c_2$ the coefficient of the ${\mathbf p}^2$-suppressed S-wave production current [@hmst1]. The explicit formulae read $$\begin{aligned}
C_{V}^{\rm ax} & = &
\frac{\alpha\pi}{m_t^2(4c_w^2-x)}\,\bigg[\, Q_e+\frac{1}{4s_w^2} \,\bigg]
\,,
\nonumber\\[2mm]
C_{A}^{\rm ax} & = &
-\frac{\alpha\pi}{4 s_w^2 m_t^2(4c_w^2-x)}
\,,
\nonumber\\[2mm]
{\cal V}_r^{(s)}(\nu) & = & -4\pi C_F \alpha_s(m_t) z\,
\left[ 1+\frac{8C_A}{3\beta_0}\ln(2-z)
\right]
\,,
\nonumber\\[2mm]
c_2(\nu) & = & -\frac{1}{6}
- \frac{8
C_F}{3\beta_0}\ln\Big(\frac{z}{2-z}\Big)
\,,
\nonumber\\[2mm]
z & \equiv & \frac{\alpha_s(m_t\nu)}{\alpha_s(m_t)}
\,.\end{aligned}$$ The resulting renormalization group equations for the Wilson coefficients $\tilde C_{V/A}$ have the form $$\begin{aligned}
\frac{d \tilde C_V(\nu)}{d\ln\nu} & = &
i\,\frac{N_c m_t^2}{8\pi^2}\,\bigg\{\,
(C_V^{\rm born})^2\,\frac{\Gamma_t}{m}\,\Big(
2 c_2(\nu) {\cal V}^{(s)}_c(\nu) + {\cal V}_r^{(s)}(\nu)\Big)
+ 2 C_V^{\rm born} C_V^{\rm bW,abs} {\cal V}^{(s)}_c(\nu)
\,\bigg\}
\nonumber\\[2mm] & &
+\,i\,\frac{N_c m_t^2}{12\pi^2}\,\bigg\{\,
(C_V^{\rm ax})^2\,\frac{\Gamma_t}{m_t}\,{\cal V}^{(s)}_c(\nu)
\,\bigg\}
\,,
\nonumber\\[4mm]
\frac{d \tilde C_A(\nu)}{d\ln\nu} & = &
i\,\frac{N_c m_t^2}{8\pi^2}\,\bigg\{\,
(C_A^{\rm born})^2\,\frac{\Gamma_t}{m}\,\Big(
2 c_2(\nu) {\cal V}^{(s)}_c(\nu) + {\cal V}_r^{(s)}(\nu)\Big)
+ 2 C_A^{\rm born} C_A^{\rm bW,abs} {\cal V}^{(s)}_c(\nu)
\,\bigg\}
\nonumber\\[2mm] & &
+\,i\,\frac{N_c m_t^2}{12\pi^2}\,\bigg\{\,
(C_A^{\rm ax})^2\,\frac{\Gamma_t}{m_t}\,{\cal V}^{(s)}_c(\nu)
\,\bigg\}
\,,\end{aligned}$$ and the solutions for scales below $m_t$ ($\nu<1$) read $$\begin{aligned}
\tilde C_V(\nu)& = & \tilde C_V(1) +
i\,\frac{2 N_c m_t^2 C_F}{3\beta_0}\,\bigg\{
\bigg[ \Big( (C_V^{\rm born})^2+ (C_V^{\rm ax})^2 \Big)\,\frac{\Gamma_t}{m_t}
+ 3 C_V^{\rm born} C_V^{\rm bW,abs}\bigg]\,\ln(z)
\nonumber\\[2mm] & &\qquad
- \frac{4C_F}{\beta_0}\,\frac{\Gamma_t}{m_t}\,(C_V^{\rm born})^2\ln^2(z)
+ \frac{4(C_A+2C_F)}{\beta_0}\,\frac{\Gamma_t}{m_t} \,(C_V^{\rm born})^2\rho(z)
\bigg\}
\,,
\nonumber\\[4mm]
\tilde C_A(\nu)& = & \tilde C_A(1) +
i\,\frac{2 N_c m_t^2 C_F}{3\beta_0}\,\bigg\{
\bigg[ \Big( (C_A^{\rm born})^2+ (C_A^{\rm ax})^2 \Big)\,\frac{\Gamma_t}{m_t}
+ 3 C_A^{\rm born} C_A^{\rm bW,abs}\bigg]\,\ln(z)
\nonumber\\[2mm] & &\qquad
- \frac{4C_F}{\beta_0}\,\frac{\Gamma_t}{m_t}\,(C_A^{\rm born})^2\ln^2(z)
+ \frac{4(C_A+2C_F)}{\beta_0}\,\frac{\Gamma_t}{m_t} \,(C_A^{\rm born})^2\rho(z)
\bigg\}
\,,
\label{tildeCav}\end{aligned}$$ where $$\begin{aligned}
\rho(z) & = & \frac{\pi^2}{12}-\frac{1}{2}\ln^22 + \ln2 \ln(z) - {\rm
Li}_2\left(\frac{z}{2}\right)
\,,\end{aligned}$$ and the $\tilde C_{V/A}(1)$ are the hard matching conditions, which are presently unknown. Finally, the contribution of the operators $\tilde {\cal O}_{V/A}$ to the total cross section reads $$\begin{aligned}
\Delta\sigma_{\rm tot}^{\Gamma,2} & = &
\mbox{Im}\Big[\,{\tilde C_V} +{\tilde C_A}\,\Big]
\,.\end{aligned}$$ Parametrically $\Delta\sigma_{\rm tot}^{\Gamma,2}$ is of order $g^6$. Comparing this with the LL cross section which counts as $g^4 v\sim
g^5$ we see that it constitutes a NLL contribution. This is also evident from the fact that the corresponding UV divergences were generated in NNLL order effective theory matrix elements. The correction $\Delta\sigma_{\rm tot}^{\Gamma,2}$ is energy-independent, but it is scale-dependent and compensates the logarithmic scale-dependence in the NNLL order matrix elements. As mentioned before, the matching conditions $\tilde C_{V/A}(\nu=1)$ are presently unknown and we therefore set them to zero in the numerical analysis presented below. We note, however, that it was shown in [@HoangTeubnerdist] that the difference between the full theory phase space (which is cut off by the large, but finite $m_t$) and the effective theory phase space (which is infinite in the computation of the forward scattering amplitude) contributes to $\tilde C_{V/A}(\nu=1)$ and also represents a NLL effect.
Numerical Analysis {#sectionanalysis}
==================
\
\
0.0cm
In Fig.\[figanalysis\] we have plotted $\Delta\sigma_{\rm tot}^{\Gamma,1}$ and $\Delta\sigma_{\rm tot}^{\Gamma,2}$ in picobarn in the 1S mass scheme [@HoangTeubnerdist; @Hoangupsilon] for $M_{\rm 1S}=175$ GeV, $\alpha=1/125.7$, $s_w^2=0.23120$, $V_{tb}=1$ and $M_W=80.425$ GeV with the renormalization scaling parameter $\nu=0.1$ (solid curves), $0.2$ (dashed curves) and $0.3$ (dotted curves). The divergences in $\Delta\sigma_{\rm tot}^{\Gamma,1}$ are subtracted minimally. For the QCD coupling we used $\alpha_s(M_Z)=0.118$ as an input and employed 4-loop renormalization group running. Note that in the 1S scheme $\delta m_t=M_{\rm 1S}({\cal V}_c^{(s)}(\nu)/4\pi)^2/8$. In Fig.\[figanalysis\]a the sum of $\Delta\sigma_{\rm tot}^{\Gamma,1}$ and $\Delta\sigma_{\rm tot}^{\Gamma,2}$ is shown while in Fig.\[figanalysis\]b both contributions are presented separately. For the top quark width we adopted the value $\Gamma_t=1.43$ GeV.[^7] We find that the sum of the corrections is negative and shows a moderate $\nu$ dependence. Compared to the most recent NNLL QCD predictions for the total cross section [@HoangEpiphany] the corrections are around $-10\%$ for energies below the peak, between $-2\%$ and $-4\%$ close to the peak and about $-2\%$ above the peak. Their magnitude is comparable to the NNLL QCD corrections. Interestingly, they partly compensate the sizeable positive QCD corrections found in [@Hoang3loop; @HoangEpiphany]. The peculiar energy dependence of the corrections, caused by the dependence on the real part of the Green function $G^0$, also leads to a slight displacement of the peak position. Relative to the peak position of the LL cross section one obtains a shift of $(30,35,47)$ MeV for $\nu=(0.1,0.2,0.3)$. This shift is comparable to the expected experimental uncertainties of the top mass measurements from the threshold scan [@TTbarsim].
Conclusion {#sectionconclusion}
==========
We have determined electroweak corrections to the NNLL top pair threshold cross section in $e^+e^-$ annihilation related to the instability of the top quark. Our approach is closely related to the treatment of absorptive processes in the optical theory. It includes the effects of the top instability by accounting for the absorptive parts in electroweak corrections to the effective theory matching conditions that are related to the observable $bW^+\bar b W^-$ final state. The matching conditions render the NRQCD Lagrangian non-hermitian, but they allow for the determination of the total cross section using the $e^+e^-\to e^+e^-$ forward scattering matrix element and the optical theorem. We have shown that the absorptive parts of the electroweak matching conditions for the $t\bar t$ production and annihilation operators describe the interference of the double-resonant amplitude for $e^+e^-\to t\bar t\to bW^+\bar b W^-$ with the single-resonant (and $v^2$-suppressed) amplitudes for $e^+e^-\to b W^+\bar t\to bW^+\bar b W^-$ and $e^+e^-\to t\bar b W^-\to bW^+\bar b W^-$. We have also shown that at NNLL order there are no further interference effects caused by the exchange and radiation of ultrasoft gluons.
The novel feature of the NNLL corrections is that they lead to new UV divergences. These divergences originate from the logarithmic high energy behavior of the $t\bar t$ phase space in the effective theory forward scattering amplitudes, that is caused by the interferences and by the modified propagators of an unstable top quark. The divergences renormalize $(e^+e^-)(e^+e^-)$ operators that contribute to the forward scattering amplitude already at NLL order. The corrections determined in this work slightly modify the cross section shape and are comparable to the known NNLL QCD corrections. The size of the corrections shows that a complete treatment of all NNLL electroweak effects is desirable.
A.H. thanks the Aspen Center for Physics where this work was initiated and I.Stewart for collaboration during the initial stages of the project. We thank A. Manohar for comments to the manuscript.
I. I. Bigi, Y. L. Dokshitzer, V. Khoze, J. Kühn and P. Zerwas, Phys. Lett. B [**181**]{}, 157 (1986).
V. S. Fadin and V. A. Khoze, JETP Lett. [**46**]{}, 525 (1987).
A. H. Hoang, arXiv:hep-ph/0204299.
M. Battaglia [*et al.*]{}, arXiv:hep-ph/0304132.
M. Martinez and R. Miquel, Eur. Phys. J. C [**27**]{}, 49 (2003) \[arXiv:hep-ph/0207315\].
A. H. Hoang [*et al.*]{}, in Eur. Phys. J. direct C [**2**]{}, 1 (2000) \[arXiv:hep-ph/0001286\].
D. Cinabro, arXiv:hep-ex/0005015.
M. Luke, A. Manohar and I. Rothstein, Phys. Rev. D [**61**]{}, 074025 (2000) \[arXiv:hep-ph/9910209\].
A. H. Hoang and I. W. Stewart, Phys. Rev. D [**67**]{}, 114020 (2003) \[arXiv:hep-ph/0209340\].
A. V. Manohar and I. W. Stewart, Phys. Rev. D [**63**]{}, 054004 (2001) \[arXiv:hep-ph/0003107\].
A. Pineda, Phys. Rev. D [**66**]{}, 054022 (2002) \[arXiv:hep-ph/0110216\].
A. Pineda, Phys. Rev. D [**65**]{}, 074007 (2002) \[arXiv:hep-ph/0109117\].
A. H. Hoang, Phys. Rev. D [**69**]{}, 034009 (2004) \[arXiv:hep-ph/0307376\].
A. A. Penin, A. Pineda, V. A. Smirnov and M. Steinhauser, Nucl. Phys. B [**699**]{}, 183 (2004) \[arXiv:hep-ph/0406175\].
A. H. Hoang, A. V. Manohar, I. W. Stewart and T. Teubner, Phys. Rev. Lett. [**86**]{}, 1951 (2001) \[arXiv:hep-ph/0011254\].
A. H. Hoang, A. V. Manohar, I. W. Stewart and T. Teubner, Phys. Rev. D [**65**]{}, 014014 (2002) \[arXiv:hep-ph/0107144\].
A. H. Hoang, Acta Phys. Polon. B [**34**]{}, 4491 (2003) \[arXiv:hep-ph/0310301\].
R. J. Guth and J. H. Kühn, Nucl. Phys. B [**368**]{}, 38 (1992).
A. H. Hoang and T. Teubner, Phys. Rev. D [**60**]{}, 114027 (1999) \[arXiv:hep-ph/9904468\]. M. Beneke, A. P. Chapovsky, A. Signer and G. Zanderighi, Phys. Rev. Lett. [**93**]{}, 011602 (2004) \[arXiv:hep-ph/0312331\]; M. Beneke, A. P. Chapovsky, A. Signer and G. Zanderighi, Nucl. Phys. B [**686**]{}, 205 (2004) \[arXiv:hep-ph/0401002\].
P. F. Harrison and H. R. Quinn \[BABAR Collaboration\], SLAC-R-0504 [*Papers from Workshop on Physics at an Asymmetric B Factory (BaBar Collaboration Meeting), Rome, Italy, 11-14 Nov 1996, Princeton, NJ, 17-20 Mar 1997, Orsay, France, 16-19 Jun 1997 and Pasadena, CA, 22-24 Sep 1997*]{}
G. P. Korchemsky and A. V. Radyushkin, Phys. Lett. B [**279**]{}, 359 (1992) \[arXiv:hep-ph/9203222\].
C. W. Bauer, D. Pirjol and I. W. Stewart, Phys. Rev. D [**65**]{}, 054022 (2002) \[arXiv:hep-ph/0109045\].
A. V. Manohar and I. W. Stewart, Phys. Rev. Lett. [**85**]{}, 2248 (2000) \[arXiv:hep-ph/0004018\]. V. S. Fadin, V. A. Khoze and A. D. Martin, Phys. Rev. D [**49**]{}, 2247 (1994).
K. Melnikov and O. I. Yakovlev, Phys. Lett. B [**324**]{}, 217 (1994) \[arXiv:hep-ph/9302311\]. M. Jezabek and J. H. Kühn, Nucl. Phys. B [**314**]{}, 1 (1989). A.V. Manohar and I.W. Stewart, Phys. Rev. D [**62**]{}, 014033 (2000) \[arXiv:hep-ph/9912226\]. A.V. Manohar and I.W. Stewart, Phys. Rev. D [**62**]{}, 074015 (2000) \[arXiv:hep-ph/0003032\]. M. Beneke and G. Buchalla, Phys. Rev. D [**53**]{}, 4991 (1996) \[arXiv:hep-ph/9601249\].
I. Blokland, A. Czarnecki, M. Slusarczyk and F. Tkachov, Phys. Rev. Lett. [**93**]{}, 062001 (2004) \[arXiv:hep-ph/0403221\].
A. Denner and T. Sack, Nucl. Phys. B [**358**]{}, 46 (1991).
W. Mödritsch and W. Kummer, Nucl. Phys. B [**430**]{}, 3 (1994).
S. f. Su and M. B. Wise, Phys. Lett. B [**510**]{}, 205 (2001) \[arXiv:hep-ph/0104169\].
A. Czarnecki and K. Melnikov, Phys. Rev. Lett. [**80**]{}, 2531 (1998) \[arXiv:hep-ph/9712222\]; M. Beneke, A. Signer and V. A. Smirnov, Phys. Rev. Lett. [**80**]{}, 2535 (1998) \[arXiv:hep-ph/9712302\].
A. H. Hoang, Phys. Rev. D [**56**]{}, 7276 (1997) \[arXiv:hep-ph/9703404\].
C. W. Bauer, A. F. Falk and M. E. Luke, Phys. Rev. D [**54**]{}, 2097 (1996) \[arXiv:hep-ph/9604290\].
A. H. Hoang and T. Teubner, Phys. Rev. D [**58**]{}, 114023 (1998) \[arXiv:hep-ph/9801397\]. A. H. Hoang, Z. Ligeti and A. V. Manohar, Phys. Rev. Lett. [**82**]{}, 277 (1999) \[arXiv:hep-ph/9809423\]; A. H. Hoang, Z. Ligeti and A. V. Manohar, Phys. Rev. D [**59**]{}, 074017 (1999) \[arXiv:hep-ph/9811239\].
[^1]: We use the term NRQCD to refer to a generic low-energy effective theory which describes nonrelativistic $t\bar t$ pairs and bound state effects and not for a theory valid only for scales $m_t > \mu > m_t v$. For the presentation we use the conventions and notations of vNRQCD established in [@LMR; @HoangStewartultra], but we emphasize that our results are generally true.
[^2]: Note that in the effective theory we use positive energy spinors for the antiparticles.
[^3]: In a complete treatment of all electroweak effects one can integrate out all electroweak effects associated with the massive W, Z and Higgs bosons at the scale $m_t$. Below the scale $m_t$ gluons, photons and quarks remain as dynamical degrees of freedom.
[^4]: In an explicit computation, quark pair production and quark-antiquark scattering diagrams involving the time-like gluons cancel at LL order. In Coulomb gauge all leading order diagrams with time-like gluons are dimensionless and individually zero in dimensional regularization [@LMR].
[^5]: For a treatment of nonrelativistic QED effects within vNRQED see Ref. [@ManoharQED].
[^6]: In vNRQCD the renormalization scales for soft and ultrasoft fluctuations, $\mu_S$ and $\mu_U$, are correlated through the heavy quark equation of motion, $\mu_U=\mu_S^2/m_t$. The correlated running from the hard scale down to the soft and ultrasoft scales is described by the dimensionless scaling parameter $\nu$ defined by $\mu_S=m_t\nu$ and $\mu_U=m_t\nu^2$. Thus $\nu=1$ corresponds to the hard matching scale.
[^7]: In a complete analysis of electroweak effects the top quark width depends on the input parameters given above and is not an independent parameter. For the purpose of the numerical analysis in this work, however, our treatment is sufficient.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In this article, we proposed a new probability distribution named as power Maxwell distribution (PMaD). It is another extension of Maxwell distribution (MaD) which would lead more flexibility to analyze the data with non-monotone failure rate. Different statistical properties such as reliability characteristics, moments, quantiles, mean deviation, generating function, conditional moments, stochastic ordering, residual lifetime function and various entropy measures have been derived. The estimation of the parameters for the proposed probability distribution has been addressed by maximum likelihood estimation method and Bayes estimation method. The Bayes estimates are obtained under gamma prior using squared error loss function. Lastly, real-life application for the proposed distribution has been illustrated through different lifetime data.'
author:
- |
[Abhimanyu Singh Yadav$^{1}$[^1], Hassan S. Bakouch $^2$, Sanjay Kumar Singh$^3$ and Umesh Singh$^3$]{}\
$^1$[Department of Statistics, Central University of Rajasthan, Ajamer, India]{}\
$^2$ [Department of Mathematics, Faculty of Science, Tanta University, Tanta, Egypt]{}\
$^3$[Department of Statistics and DST-CIMS, BHU, Varanasi, India]{}
title: ' **Power Maxwell distribution: Statistical Properties, Estimation and Application**'
---
**Keywords:** Maxwell distribution, Power Maxwell distribution, moments, stochastic order, entropy, Classical and Bayes estimation.
Introduction
============
The Maxwell distribution has broad application in statistical physics, physical chemistry, and their related areas. Besides Physics and Chemistry it has good number of applications in reliability theory also. At first, the Maxwell distribution was used as lifetime distribution by Tyagi and Bhattacharya (1989). The inferences based on generalized Maxwell distribution has been discussed by Chaturvedi and Rani (1998). Bekker and Roux (2005) consider the estimation of reliability characteristics under for Maxwell distribution under Bayes paradigm. Radha and Vekatesan (2005) discuss the prior selection procedure in case of Maxwell probability distribution. Shakil et al. (2008) studied the distributions of the product $|XY|$ and ratio $|X/Y|$ when X and Y are independent random variables having the Maxwell and Rayleigh distributions. Day and Maiti (2010) proposed the Bayesian estimation of the parameter for the Maxwell distribution. Tomer and Panwar (2015) discussed the estimation procedure for the parameter of Maxwell distribution in the presence of progressive type-I hybrid censored data. After this, Modi (2015), Saghir and Khadim (2016), proposed lengths biased Maxwell distribution and discussed its various properties. Furthermore, several generalizations based on Maxwell distribution are advocated and statistically justified. Recently, two more extensions of Maxwell distribution has been introduced by Sharma et al. (2017a), (2017b) and discussed the classical as well as Bayesian estimation of the parameter along with the real-life application.\
A random variable $Z$ follows Maxwell distribution with scale parameter $\alpha$, denoted as $Z \sim
MaD(\alpha )$, if its probability density function (pdf) and cumulative distribution function (cdf) are given by $$f(z,\alpha)=\dfrac{4}{\sqrt{\pi}} \alpha^{\frac{3}{2}} z^{2}e^{-\alpha z^{2}}~~~~z\ge 0,\alpha>0$$ and $$F(z,\alpha)=\dfrac{2}{\sqrt{\pi}}\Gamma\left( \frac{3}{2},\alpha z^{2}\right)$$ respectively, where $\Gamma (a, z) =\int_{0}^{z}p^{a-1}e^{-p}dp$ is the incomplete gamma function.\
In this article, we proposed PMaD as a new generalization of the Maxwell distribution and discussed its various statistical properties and application. The objective of this article is to study the statistical properties of the PMaD distribution, and then estimate the unknown parameters using classical and Bayes estimation methods. Other motivations regarding advantages of the PMaD distribution comes from its flexibility to model the data with nono-monotone failure rate. Thus, it can be taken as an excellent alternative to several inverted families of distributions. The uniqueness of this study comes from the fact that we provide a comprehensive description of mathematical and statistical properties of this distribution with the hope that it will attract more extensive applications in biology, medicine, economics, reliability, engineering, and other areas of research.\
The rest of the paper has been shaped in the following manner. The introduction of the proposed study including the methodological details is given in Section and Subsection of 1. Section 2 provides some statistical properties related to the proposed model. Residual and reverse residual lifetime function for PMaD is derived in Section 3. In Section 4, order statistics have been obtained. The MLEs and Bayes estimation procedure have been discussed in Section 5. In Section 6, simulation study is carried out to compare the performance of maximum likelihood estimates (MLEs) and Bayes estimates. In Section 7, we illustrate the application and usefulness of the proposed model by applying it to four data sets. Finally, Section 8 offers some concluding remarks.
Power Maxwell Distribution and Statistical Properties
=====================================================
In statistical literature, several generalizations based on certain baseline probability distribution have been advocated regarding the need of the study. These generalized model accommodate the various nature of hazard rate and seems to be more flexible. Here, this paper provides another generalization of the MaD using power transformation of Maxwell random variates. Let us consider a transformation $X=Z^{\frac{1}{\beta}}$ where $Z\sim MaD(\alpha)$. Then the resulting distribution of $X$ is called as power maxwell distribution with parameter $\alpha$ and $\beta$ respectively. From now, it is denoted by $X\sim PMaD(\alpha,\beta)$, where, $\alpha$ and $\beta$ are the scale and shape parameter of the proposed distribution. The probability density function and cumulative distribution function of the PMaD are given by $$f(x,\alpha,\beta)=\dfrac{4}{\sqrt{\pi}} \alpha^{\frac{3}{2}}\beta x^{3\beta-1}e^{-\alpha x^{2\beta}}~~~~x\ge 0,\alpha,\beta>0$$ $$F(x,\alpha,\beta)=\dfrac{2}{\sqrt{\pi}}\gamma\left( \frac{3}{2},\alpha x^{2\beta}\right)$$ respectively. The different mathematical and statistical properties such as moments, reliability, hazard, median, mode, the coefficient of variation, mean deviation, conditional moments, Lorentz curve, stochastic ordering, residual life, entropy measurements, of PMaD have been derived in following subsections.
{width="6.5in" height="4in"}
Asymptotic behaviour
--------------------
This subsection, described the symptotic nature of density and survival functions for the proposed distribution. To illustrate assymptoic behaviour, at first, we will show that $\lim\limits_{x\rightarrow 0}f(x,\alpha,\beta)=0$ and $\lim\limits_{x\rightarrow \infty}f(x,\alpha,\beta)=0$ . Therefore, using (2.1) $$\begin{split}
\lim\limits_{x\rightarrow 0}f(x,\alpha,\beta)&=\dfrac{4}{\sqrt{\pi}} \alpha^{\frac{3}{2}}\beta\lim\limits_{x\rightarrow 0} x^{3\beta-1}e^{-\alpha x^{2\beta}}\\&=\dfrac{4}{\sqrt{\pi}} \alpha^{\frac{3}{2}}\beta\times 0=0
\end{split}$$ $\implies$ $\lim\limits_{x\rightarrow \infty}f(x,\alpha,\beta)=0$\
and $$\begin{split}
\lim\limits_{x\rightarrow \infty}f(x,\alpha,\beta)&=\dfrac{4}{\sqrt{\pi}} \alpha^{\frac{3}{2}}\beta\lim\limits_{x\rightarrow \infty} x^{3\beta-1}\lim\limits_{x\rightarrow \infty}e^{-\alpha x^{2\beta}}\\&=\dfrac{4}{\sqrt{\pi}} \alpha^{\frac{3}{2}}\beta\times \infty\times 0=0
\end{split}$$ $\implies$ $\lim\limits_{x\rightarrow \infty}f(x,\alpha,\beta)=0$\
Similarly, the asymptotic behaviour of survival function can also be shown and found that $\lim\limits_{x\rightarrow 0}S(x)=1$ and $\lim\limits_{x\rightarrow \infty}S(x)=0$.
Reliability and hazard functions
--------------------------------
The characteristics based on reliability function and hazard function are very useful to study the pattern of any lifetime phenomenon. The reliability and hazard function of the proposed distribution have been derived as;
- The reliability function $R(x,\alpha,\beta)$ is given by $$R(x)=1-\dfrac{2}{\sqrt{\pi}}\gamma\left( \frac{3}{2},\alpha x^{2\beta}\right)$$
- The mean time to system failure (MTSF) is given as $$Mt(x)=\dfrac{2}{\sqrt{\pi}}\left( \frac{1}{\alpha}\right) ^{\frac{1}{2\beta}}\Gamma\left( \dfrac{3\beta+1}{2\beta}\right)$$
- The hazard function $H(x)$ is given as $$H(x)=\dfrac{4\alpha^{\frac{3}{2}}\beta x^{3\beta-1}e^{-\alpha x^{2\beta}}}{\sqrt{\pi}-2\gamma\left( \frac{3}{2},\alpha x^{2\beta}\right) }$$
{width="6.5in" height="4in"}
- The reverse hazard rate $h(x)$ is obatined as $$h(x)=\dfrac{2\alpha^{\frac{3}{2}}\beta x^{3\beta-1}e^{-\alpha x^{2\beta}}}{\gamma\left( \frac{3}{2},\alpha x^{2\beta}\right) }$$
- The odds function is defined as; $$O(x)=\dfrac{2\gamma\left( \frac{3}{2},\alpha x^{2\beta}\right) }{\sqrt{\pi}-2\gamma\left( \frac{3}{2},\alpha x^{2\beta}\right) }$$
Moments
-------
Let $x_1,x_2,\cdots x_n$ be the random observation from PMaD$(\alpha,\beta)$. The $r^{th}$ moment $\mu_r^{'}$ about origin is defined as $$\begin{split}
\mu_r^{'}&=E(x^{r})=\int_{x=0}^{\infty}x^{r}f(x,\alpha,\beta)\quad dx\\&=\dfrac{2}{\sqrt{\pi}}\left( \dfrac{1}{\alpha}\right)^{\frac{r}{2\beta}}\Gamma\left( \dfrac{3\beta+r}{2\beta}\right)
\end{split}$$ The first, second, third and fourth raw moment about origin are obtained by putting $r=1, 2 ,\cdots,4$ in above expression. If $r=1$ then we get mean of the proposed distribution. Thus, $$\mu_1^{'}=\dfrac{2}{\sqrt{\pi}}\left( \frac{1}{\alpha}\right) ^{\frac{1}{2\beta}}\Gamma\left( \dfrac{3\beta+1}{2\beta}\right)$$ for $r=2,3 \, \& 4$ $$\mu_2^{'}=\dfrac{2}{\sqrt{\pi}}\left( \frac{1}{\alpha}\right) ^{\frac{1}{\beta}}\Gamma\left( \dfrac{3\beta+2}{2\beta}\right)$$ $$\mu_3^{'}=\dfrac{2}{\sqrt{\pi}}\left( \frac{1}{\alpha}\right) ^{\frac{3}{2\beta}}\Gamma\left( \dfrac{3\beta+3}{2\beta}\right)$$ and $$\mu_4^{'}=\dfrac{2}{\sqrt{\pi}}\left( \frac{1}{\alpha}\right) ^{\frac{2}{\beta}}\Gamma\left( \dfrac{3\beta+4}{2\beta}\right)$$ The respective central moment can be evaluated by using the following relations. $$\mu_2=\mu_2^{'}-\left( \mu_1^{'}\right)^{2}=\dfrac{4}{\pi}\left( \dfrac{1}{\alpha}\right)^{\frac{1}{\beta}}\left[ \sqrt{\frac{\pi}{4}} \Gamma\left( \dfrac{3\beta+2}{2\beta}\right) -\left\lbrace \Gamma\left( \dfrac{3\beta+1}{2\beta}\right)\right\rbrace^{2} \right]$$ $$\mu_3=\mu_3^{'}-3\mu_2^{'}\mu_1^{'}+2\left( \mu_1^{'}\right) ^{3}$$ $$\mu_4=\mu_4^{'}-4\mu_3^{'}\mu_1^{'}+6\mu_2^{'}\left( \mu_1^{'}\right) ^{2}-3\left( \mu_1^{'}\right) ^{4}$$
Coefficient of Skewness and Kurtosis
------------------------------------
The coefficient of skewness and kurtosis measure nd convexity of the curve and its shape. It is obtained by moments based relations suggested by Pearson and given by; $$\beta_1=\dfrac{\left[ \mu_3^{'}-3\mu_2^{'}\mu_1^{'}+2\left( \mu_1^{'}\right) ^{3}\right] ^{2}}{\left[\mu_2^{'}-\left( \mu_1^{'}\right)^{2} \right]^{3} }$$ and $$\beta_2=\dfrac{\mu_4^{'}-4\mu_3^{'}\mu_1^{'}+6\mu_2^{'}\left( \mu_1^{'}\right) ^{2}-3\left( \mu_1^{'}\right) ^{4}}{\left[\mu_2^{'}-\left( \mu_1^{'}\right)^{2} \right]^{2} }$$ These values are calculated in Table 1 for different combination of model parameters and it is observed that the shape of PMaD is right skewed and almost symmetrical for some choices of $\alpha, \beta$. Also, it can has the nature of platykurtic, mesokurtic and leptokurtic, thus PMaD may be used to model skewed and symmetric data as well.
Coefficient of variation
------------------------
The coefficient of variation (CV) is calculated by $$CV=\dfrac{\sqrt{\mu_2^{'}-\left( \mu_1^{'}\right)^{2}}}{\mu_1^{'}}\times 100$$
Mode and Median
---------------
The mode $(M_0)$ for PMaD $(\alpha,\beta)$ is obtained by solving the following expression $$\dfrac{d}{dx}f(x,\alpha,\beta)|_{M_0}=0$$ which yield $$M_0=\left( \dfrac{3\beta-1}{2\alpha\beta}\right)^{\frac{1}{2\beta}}$$ The median $(M_{d})$ of the proposed distribtuion can be calculated by using the empirical relation amung mean, median and mode. Thus, the median is, $$M_{d}=\frac{1}{3} M_{0}+\frac{2}{3}\mu_1^{'}=\frac{1}{3} \left[ \left( \dfrac{3\beta-1}{2\alpha\beta}\right)^{\frac{1}{2\beta}}+ \dfrac{4}{\sqrt{\pi}}\left( \frac{1}{\alpha}\right) ^{\frac{1}{2\beta}}\Gamma\left( \dfrac{3\beta+1}{2\beta}\right)\right]$$
[ccccccc]{}\
\[0\][\*]{}[$ \alpha,~ \beta $]{} & $\mu_1^{'}$ & $\mu_2$ & $\beta_1$& $\beta_2$ & $x_0$ & CV\
&\
0.5, 0.5 & 3.0008 & 5.9992 & 2.6675 & 7.0010 & 1.0000 & 0.8162\
0.5, 1.0 & 1.5962 & 0.4530 & 0.2384 & 3.1071 & 1.4142 & 0.4217\
0.5, 1.5 & 1.3376 & 0.1499 & 0.0102 & 2.7882 & 1.3264 & 0.2894\
0.5, 2.5 & 1.1780 & 0.0445 & 0.0481 & 2.7890 & 1.2106 & 0.1792\
0.5, 3.5 & 1.1204 & 0.0211 & 0.1037 & 2.4351 & 1.1533 & 0.1298\
\
0.5, 0.75 & 1.9392 & 1.1443 & 0.7425 & 3.8789 & 1.4057 & 0.5516\
1.0, 0.75 & 1.2216 & 0.4541 & 0.7425 & 3.8789 & 0.8855 & 0.5516\
1.5, 0.75 & 0.9323 & 0.2645 & 0.7425 & 3.8789 & 0.6758 & 0.5516\
2.5, 0.75 & 0.6632 & 0.1338 & 0.7425 & 3.8789 & 0.4807 & 0.5516\
3.5, 0.75 & 0.5299 & 0.0855 & 0.7425 & 3.8789 & 0.3841 & 0.5516\
\
1, 1 & 1.1287 & 0.2265 & 0.2384 & 3.1071 & 1.0000 & 0.4217\
2, 2 & 0.8723 & 0.0372 & 0.0102 & 2.7895 & 0.8891 & 0.2212\
3, 3 & 0.8484 & 0.0163 & 0.0831 & 2.6907 & 0.8736 & 0.1506\
4, 4 & 0.8509 & 0.0094 & 0.1069 & 1.9643 & 0.8750 & 0.1140\
5, 5 & 0.8586 & 0.0062 & 0.0677 & 0.1072 & 0.8805 & 0.0915\
\[tab:addlabel\]
Mean Deviation
--------------
The mean deviation (MD) about mean $(\mu^{'}_{1}=\mu)$ is defined by $$\begin{split}
MD&=\int_{x}|x-\mu| f(x,\alpha,\beta) dx\\&=\int_{x=0}^{\mu}(\mu-x)f(x,\alpha,\beta) dx+\int_{x=\mu}^{\infty}(x-\mu) f(x,\alpha,\beta) dx
\end{split}$$ After simplification, we get $$\begin{split}
MD&=2\mu F(\mu)-2\mu+2\int_{\mu}^{\infty} f(x,\alpha,\beta) dx\\&=(\mu-1)\left[ \dfrac{4}{\sqrt{\pi}}\gamma\left( \frac{3}{2},\alpha \mu^{2\beta}\right)-2\right]
\end{split}$$
Generating Functions
--------------------
In distribution theory, the role of generating functions are very usefull to generate the respective moments of the distribution and also these functions are uniquely determine the distribution. The different generating function for PMaD $(\alpha,\beta)$ have been caluculated as follows;
- Moment generating function (mgf) $M_X(t)$ for a random variable $X$ is obatined as $$M_X(t)=E(e^{tx})=\dfrac{2}{\sqrt{\pi}}\sum_{i=0}^{\infty}\frac{1}{j!} \left( \dfrac{t}{\alpha^{2\beta}}\right)^{r} \Gamma\left( \dfrac{3\beta+r}{2\beta}\right)$$
- Characteristics function (chf) $\phi_X(t)$ for random variable $X$ is obtained by replacing $t$ by $jt$, $$\phi_X(t)=E(e^{jtx})=\dfrac{2}{\sqrt{\pi}}\sum_{i=0}^{\infty}\frac{1}{j!} \left( \dfrac{jt}{\alpha^{2\beta}}\right)^{r} \Gamma\left( \dfrac{3\beta+r}{2\beta}\right)$$ where, $j^2=-1$.
- The kumulants generating function (KGF) is obtained as $$K_X(t)=\ln \left[ \dfrac{2}{\sqrt{\pi}}\sum_{i=0}^{\infty}\frac{1}{j!} \left( \dfrac{t}{\alpha^{2\beta}}\right)^{r} \Gamma\left( \dfrac{3\beta+r}{2\beta}\right)\right]$$
Conditional Moment and MGF
--------------------------
Let $X$ be a random variable from PMaD$ (\alpha,\beta)$, then conditional moments $E(X^r|X>k)$ and conditional mgf $E(e^{tx}|X>k)$ are evaluated in following expressions; $$\begin{split}
E(X^r|X>k)&=\dfrac{\int_{x>k}x^r f(x,\alpha,\beta) dx}{\int_{x>k} f(x,\alpha,\beta) dx}\\&=\dfrac{2\left(\frac{1}{\alpha}\right)^{\frac{r}{2\beta}}\gamma\left( \dfrac{3\beta+r}{2\beta},\alpha x^{2\beta}\right)}{\sqrt{\pi}-2\gamma\left( \frac{3}{2},\alpha x^{2\beta}\right)}
\end{split}$$ $$\begin{split}
E(e^{tx}|X>k)&=\dfrac{\int_{x>k}e^{tx}f(x,\alpha,\beta) dx}{\int_{x>k} f(x,\alpha,\beta) dx}\\&=\dfrac{2\sum_{i=0}^{\infty}\dfrac{t^i}{i!}\left(\frac{1}{\alpha}\right)^{\frac{r}{2\beta}}\gamma\left( \dfrac{3\beta+r}{2\beta},\alpha x^{2\beta}\right)}{\sqrt{\pi}-2\gamma\left( \frac{3}{2},\alpha x^{2\beta}\right)}
\end{split}$$ respectively.
Bonferroni and Lorenz Curves
----------------------------
In economics to measure the income and poverty level, Bonferroni and Lorenz curves are frequently used. These two have good linkup to each other and has more comprehensive applications in actuarial as well as in demography. It was initially proposed and studied by Bonferroni (1920), matthematically, it is defined as; $$\zeta(\nu)_{b}=\dfrac{1}{\nu \mu}\int_{0}^{q}xf(x,\alpha,\beta)dx$$ $$\zeta(\nu)_{l}=\dfrac{1}{ \mu}\int_{0}^{q}xf(x,\alpha,\beta)dx$$ respectively. where $q=F^{-1}(\nu)$ and $\mu=E(X)$. Hence using eqn (2.1), the above two equations are reduces as $$\zeta(\nu)_{b}=\dfrac{\sqrt{\alpha}~IG\left( \frac{1+2\beta}{2\beta},\alpha q^{2\beta}\right) }{\nu\Gamma\left( \frac{3\beta+1}{2\beta}\right) }$$ $$\zeta(\nu)_{l}=\dfrac{\sqrt{\alpha}~IG\left( \frac{1+2\beta}{2\beta},\alpha q^{2\beta}\right) }{\Gamma\left( \frac{3\beta+1}{2\beta}\right) }$$
Stochastic Ordering
-------------------
A random variable $X$ is said to be stochastically greater $(Y \le_{st} X)$ than $Y$ if $F_X(x)\le F_Y (x)$ for all $x$. In the similar way, $X$ is said to be stochastically greater $(X \le_{st} Y )$ than $Y$ in the
- hazard rate order $(X \le_{hr} Y )$ if $h_X(x) \ge h_Y (x)$ $\forall x$.
- mean residual life order $(X \le_{mrl} Y )$ if $m_X(x) \ge m_Y (x)$ $\forall x$.
- likelihood ratio order $(X \le_{lr} Y )$ if $\left[\dfrac{f_{X}(x)}{f_{Y}(x)}\right]$ decreases in x.
From the above relations, we can veryfied that; $$(X \le_{lr} Y )\Rightarrow (X \le_{hr} Y )\Downarrow (X \le_{st} Y )\Rightarrow (X \le_{mrl} Y )$$ The PMaD is ordered with respect to the strongest likelihood ratio ordering as shown in the following theorem.\
**Theorem:** Let $X\sim PMaD(\alpha_1,\beta_1)$ and $Y\sim PMaD(\alpha_2,\beta_2)$. Then $(X \le_{lr} Y )$ and hence $(X \le_{hr} Y )$, $(X \le_{mrl} Y )$ and $(X \le_{st} Y )$ for all values of $\alpha_i,\beta_i$; $i=1, 2$.\
**Proof:** The likelihood ratio is $\left[\dfrac{f_{X}(x)}{f_{Y}(x)}\right]$ i.e. $$\begin{split}
\Phi=\dfrac{f_{X}(x)}{f_{Y}(x)}&=\left( \dfrac{\alpha_1}{\alpha_2}\right)^{\frac{3}{2}} \left( \dfrac{\beta_1}{\beta_2}\right) x^{3(\beta_1-\beta_2)} e^{-(\alpha_1 x^{2\beta_1}+\alpha_2 x^{2\beta_2})}
\end{split}$$ Therefore, $$\Phi^{'}=\log \left( \dfrac{f_{X}(x)}{f_{Y}(x)}\right) =\dfrac{1}{x}\left[ 3(\beta_1-\beta_2)-(\alpha_1 x^{2\beta_1}+\alpha_2 x^{2\beta_2})\right]$$ If $\beta_1=\beta_2=\beta (say)$, then $\Phi^{'}<0$, which shows that $(X \le_{lr} Y )$. The remaining ordering behaviour can be proved in same way. Also, if $\alpha_1=\alpha_2=\alpha (say)$ and $\beta_1<\beta_2$ then again $\Phi^{'}<0$, which shows that $(X \le_{lr} Y )$. The remaining ordering can be proved in same way.
Residual Lifetime
=================
In survival analysis, the term residual lifetime often used to describe the remaining lifetime associated with any particular system. Here, we derived the expression of residual life and reversed residual life for PMaD. The residual lifetime function is defined by $ R_{t}= P[x -t|x > t], t \ge0$ and the reversed residual life is described as $\bar{R }_{t} = P[t -x|x \le t]$ which denotes the time elapsed from the failure of a component given that its life less or equal to $t$.
- **Residual life time function**\
The survival function of the residual lifetime is given by $$S_{R_t}(x)=\dfrac{S(t+x)}{S(t)}= \dfrac{\sqrt{\pi}-2\gamma\left[ \frac{3}{2},\alpha (x+t)^{2\beta}\right] }{\sqrt{\pi}-2\gamma\left( \frac{3}{2},\alpha x^{2\beta}\right) }~~~~~ ;x> t$$ The corresponding probability density function is $$f_{R_t}(x)= \dfrac{4 \alpha^{\frac{3}{2}}\beta (x+t)^{3\beta-1}e^{-\alpha (x+t)^{2\beta}}}{\sqrt{\pi}-2\gamma\left( \frac{3}{2},\alpha x^{2\beta}\right) }$$ Thus, hazard function is obtained as $$h_{R_t}(x)=\dfrac{4 \alpha^{\frac{3}{2}}\beta (x+t)^{3\beta-1}e^{-\alpha (x+t)^{2\beta}}}{\sqrt{\pi}-2\gamma\left[ \frac{3}{2},\alpha (x+t)^{2\beta}\right]}$$
- **Reversed residual lifetime function**\
The survival function for MOEIED is given by $$S_{\bar{R}_t}=\dfrac{F(t-x)}{F(t)}= \dfrac{\gamma\left[ \frac{3}{2},\alpha (t-x)^{2\beta}\right] }{\gamma\left( \frac{3}{2},\alpha x^{2\beta}\right) }~~~~~ ;0\le x<t$$ The associated pdf is evaluated as; $$f_{\bar{R}_t}=\dfrac{2 \alpha^{\frac{3}{2}}\beta (t-x)^{3\beta-1}e^{-\alpha (t-x)^{2\beta}}}{\gamma\left( \frac{3}{2},\alpha x^{2\beta}\right) }$$ Hence, the hazard function based on reversed residual lifetime is obtained as $$h_{\bar{R}_t}=\dfrac{2 \alpha^{\frac{3}{2}}\beta (t-x)^{3\beta-1}e^{-\alpha (t-x)^{2\beta}}}{\gamma\left[ \frac{3}{2},\alpha (t-x)^{2\beta}\right]}$$
Entropy Measurements
====================
In information theory, entropy measurement plays a vital role to study the uncertainty associated with the probability distribution. In this section, we discuss the different measure of change. For more detail about entropy measurement, see Reniyi (1961).
Renyi Entropy
-------------
Renyi entropy of a r.v. $x$ is defined as $$\begin{split}
R_E&=\dfrac{1}{(1-\in)}\ln\left[ \int_{x=0}^{\infty}f_{w}^{\in}(x,\alpha,\beta)dx\right]\\&=\dfrac{1}{(1-\in)}\ln \left[ \int_{x=0}^{\infty}\left\lbrace \dfrac{4}{\sqrt{\pi}} \alpha^{\frac{3}{2}}\beta x^{3\beta-1}e^{-\alpha x^{2\beta}} \right\rbrace^{\in} dx \right]
\end{split}$$ Hence, after solving the internal, we get the following [$$\begin{split}
R_E&=\dfrac{1}{(1-\in)}\left[\lambda \ln 4-\frac{\lambda}{2}\ln\pi+\lambda\ln\beta-\dfrac{1-\lambda-2\beta}{2\beta}\ln\alpha-\dfrac{3\lambda\beta-\lambda+1}{2\beta}\ln \lambda+\ln \left( \dfrac{3\beta\lambda-\lambda+1}{2\beta}\right) \right]
\end{split}$$]{}
$\Delta$-Entropy
----------------
The $\beta$-entropy is obtained as follows $$\Delta_E=\dfrac{1}{\Delta-1}\left[1-\int_{x=0}^{\infty}f^{\Delta}(x,\alpha,\beta)dx \right]$$ Using pdf (1.4) and after simplification the expression for $\beta$-entropy is given by; $$\Delta_E=\dfrac{1}{\Delta-1}\left[1-\left( \dfrac{4}{\sqrt{\pi}}\right) ^{\Delta}\beta^{\Delta} \left( \dfrac{1}{\alpha}\right)^{\dfrac{1-\Delta-2\beta}{2\beta}} \left( \dfrac{\Gamma\left(\dfrac{3\Delta\beta-\Delta+1}{2\beta} \right) }{\Delta^{\dfrac{3\Delta\beta-\Delta+1}{2\beta}}}\right) \right]$$
Generalized Entropy
-------------------
The generalized entropy is obtained by; $$G_E=\dfrac{\nu_\lambda\mu^{-\lambda}-1}{\lambda(\lambda-1)}~~~~~;\lambda\ne 0,1$$ where, $\nu_\lambda=\int_{x=0}^{\infty}x^{\lambda}f_w(x,\alpha,\theta^{})dx$ and $\mu=E(X)$. The value of $\nu_\lambda$ is calculated as
$$\begin{split}
\nu_\lambda&=\dfrac{2}{\sqrt{\pi}}\left( \dfrac{1}{\alpha}\right)^{\frac{\lambda}{2\beta}}\Gamma\left( \dfrac{3\beta+\lambda}{2\beta}\right)
\end{split}$$
After using (4.6) and (2.8), we get $$G_E=\left( \dfrac{4}{\pi}\right)^{\frac{1-\lambda}{2}} \left[ \dfrac{\Gamma\left( \dfrac{3\beta+\lambda}{2\beta}\right) \left\lbrace \Gamma\left( \dfrac{3\beta+1}{2\beta}\right) \right\rbrace ^{-\lambda}}{\lambda(\lambda-1)}\right] ~~~~~;\lambda\ne0,1$$
Parameter Estimation
====================
Here, we describe maximum likelihood estimation method and Bayes estimation method for estimating the unknown parameters $\alpha, \beta$ of the PMaD. The estimators obtained under these methods are not in nice closed form; thus, numerical approximation techniques are used to get the solution. Further, the performances of these estimators are studied through Monte Carlo simulation.
Maximum Likelihood Estimation
-----------------------------
The most popular and efficient method of classical estimation of the parameter (s) is maximum likelihood estimation. The estimators obtained by this method passes several optimum properties. The maximum likelihood estimation theory required formulation of the likelihood function. Thus, let us suppose that $X_1,X_2,\cdots,X_n$ are the iid random sample of size $n$ taken from PMaD $(\alpha,\beta)$. The likelihood function is written as; $$\begin{split}
L(\alpha,\theta)&=\prod_{i=1}^{n}\dfrac{4}{\sqrt{\pi}} \alpha^{\frac{3}{2}}\beta x_i^{3\beta-1}e^{-\alpha x_i^{2\beta}}=\dfrac{4^{n}}{\pi^{n/2}} \alpha^{\frac{3n}{2}}\beta^{n} e^{-\alpha\sum_{i=1}^{n} x_i^{2\beta}}\left( \prod_{i=1}^{n}x_i^{3\beta-1}\right)
\end{split}$$ Log-likelihood function is written as; $$\ln L(\alpha,\theta)=l=n\ln 4-\frac{n}{2}\ln \pi+\dfrac{3n}{2}\ln \alpha+n\ln\beta
-\alpha\sum_{i=1}^{n}x_i^{2\beta}+(3\beta-1)\sum_{i=1}^{n}\ln x_i$$ for MLEs of $\alpha$ and $\beta$, $$\dfrac{\partial l}{\partial\alpha}=0~~~\&~~~\dfrac{\partial l}{\partial\beta}=0$$ which yield, $$\dfrac{3n}{2\alpha}-\sum_{i=1}^{n}x_i^{2\beta}=0$$ $$\dfrac{n}{\beta}-2\alpha\sum_{i=1}^{n} x_i^{2\beta}\ln x_i+3\sum_{i=1}^{n}\ln x_i =0$$ The MLE’s of the parameters are obtained by solving the above two equations simultaneously. Here, we used non-linear maximization techniques to get the solution.
### Uniqueness of MLEs
The uniqueness of MLEs discussed in previous section can be checked by using following propositions.\
**Proposition 1:** If $\beta$ is fixed, then $\hat{\alpha}$ exist and it is unique.\
**Proof:** Let $L_{\alpha}=\dfrac{3n}{2\alpha}-\sum_{i=1}^{n}x_i^{2\beta}$, since $L_{\alpha}$ is continuous and it has been verified that $\lim\limits_{\alpha\rightarrow 0}L_{\alpha}=\infty$ and $\lim\limits_{\alpha\rightarrow \infty}L_{\alpha}=-\sum_{i=1}^{n}x_i^{2\beta}<0$. This implies that $L_{\alpha}$ will have atleast one root in interval $(0,\infty)$ and hence $L_{\alpha}$ is a decreasing function in $\alpha$. Thus, $L_{\alpha}=0$ has a unique solution in $(0,\infty)$.\
**Proposition 2:** If $\alpha$ is fixed, then $\hat{\beta}$ exist and it is unique.\
**Proof:** Let $L_{\beta}=\dfrac{n}{\beta}-\alpha\sum_{i=1}^{n} x_i^{2\beta}\ln x_i+3\sum_{i=1}^{n}\ln x_i $, since $L_{\beta}$ is continuous and it has been verified that $\lim\limits_{\beta\rightarrow 0}L_{\beta}=\infty$ and $\lim\limits_{\beta\rightarrow \infty}L_{\beta}=-2\sum_{i=1}^{n}\ln {x_i}<0$. This implies in same as above $\hat{\beta}$ exists and it will be unique.
### Fisher Information Matrix
Here, we derive Fisher information matrix for constructing 100$(1-\Psi)\%$ asymptotic confidence interval for the parameters using large sample theory. The Fisher information matrix can be obtained by using equation (5.2) as $$I(\hat{\alpha},\hat{\beta})=-E\begin{pmatrix}
l_{\alpha\alpha} & l_{\alpha\beta}\\
\\
l_{\beta\alpha} & l_{\beta\beta}
\end{pmatrix}_{(\hat{\alpha},\hat{\beta})} \eqno{(5.2.1)}$$ where, $$l_{\alpha\alpha}=-\dfrac{3n}{2\alpha^2},~~l_{\alpha\beta}=-2\sum_{i=1}^{n}x_i^{2\beta}\ln x_i, ~~
l_{\beta\beta}=-\dfrac{n}{\beta^2}-4\alpha\sum_{i=1}^{n}x_i^{2\beta}(\ln x_i)^{2}$$
The above matrix can be inverted and diagonal elements of $I^{-1}(\hat{\alpha},\hat{\beta})$ provide asymptotic variance of $\alpha$ and $\beta$ respectively. Now, two sided $100(1-\Psi)\%$ asymptotic confidence interval for $\alpha$, $\beta$ has been obtained as $$[\alpha_l,\alpha_u] \in [\hat{\alpha}\mp Z_{1-\frac{\Psi}{2}}\sqrt{var(\hat{\alpha})}]$$ $$[\beta_l,\beta_u] \in [\hat{\beta}\mp Z_{1-\frac{\Psi}{2}}\sqrt{var(\hat{\beta})}]$$ respectively.
Bayes Estimation
----------------
In this subsection, the Bayes estimation procedure for the PMaD has been developed. In this estimation technique, as we all know that the unknown parameter treated as the random variable and this randomness of the parameter quantify in the form of prior distribution. Here, we took two independent gamma prior for both shape and scale parameter. The considered prior is very flexible due to its flexibility of assuming different shape. Thus, the joint prior $g(\alpha,\beta)$ is given by;
$$g(\alpha,\beta)\propto \alpha^{a-1}\beta^{c-1}e^{-b\alpha-d\beta}~~;~~~\alpha, \beta>0$$
where, a, b, c & d are the hyperparmaters of the considered priors. Using (5.1) and (5.5), the joint posterior density function $\pi(\alpha,\beta|x)$ is derived as $$\begin{split}
\pi(\alpha,\beta|x)&= \dfrac{L(x|\alpha,\beta)g(\alpha,\beta)}{\int_{\alpha}\int_{\beta} L(x|\alpha,\beta)g(\alpha,\beta) d\alpha\,d\beta}\\&= \dfrac{\alpha^{\frac{3n}{2}+a-1}\beta^{n+c-1} e^{-\alpha \left(b+\sum_{i=1}^{n} x_i^{2\beta}\right) }e^{-d\beta}\left( \prod_{i=1}^{n}x_i^{3\beta-1}\right) }{\int_{\alpha}\int_{\beta}\alpha^{\frac{3n}{2}+a-1}\beta^{n+c-1} e^{-\alpha \left(b+\sum_{i=1}^{n} x_i^{2\beta}\right) }e^{-d\beta}\left( \prod_{i=1}^{n}x_i^{3\beta-1}\right) \quad d\alpha\quad d\beta}
\end{split}$$ In the Bayesian analysis, the specification of proper loss function plays an important role. Here, we took most frequently used square error loss function (SELF) to obtain the estimates of the parameters. It is defined as; $$L(\phi,\hat{\phi})\propto \left( \phi-\hat{\phi}\right)^{2}$$ where, $\hat{\phi}$ is estimate of $\phi$. Bayes estimates under SELF is the posterior mean and evaluated by $$\hat{\phi}_{SELF}=\left[ E(\phi|x)\right]$$ provided the expectation exist and finite. Thus, the Bayes estimator based on equation no. (5.6) under SELF are given by $$\hat{\alpha}_{bs}=E_{\alpha,\beta|x}(\alpha|\beta,x)=\eta \int_{\alpha}\int_{\beta}\alpha^{\frac{3}{2}+a}\beta^{n+c-1} e^{-\alpha \left(b+\sum_{i=1}^{n} x_i^{2\beta}\right) }e^{-d\beta}\left( \prod_{i=1}^{n}x_i^{3\beta-1}\right) ~ d\alpha~ d\beta$$ and $$\hat{\beta}_{bs}=E_{\alpha,\beta|x}(\beta|\alpha,x)=\eta \int_{\alpha}\int_{\beta}\alpha^{\frac{1}{2}+a}\beta^{n+c} e^{-\alpha \left(b+\sum_{i=1}^{n} x_i^{2\beta}\right) }e^{-d\beta}\left( \prod_{i=1}^{n}x_i^{3\beta-1}\right) ~ d\alpha~ d\beta$$ where, $\eta=\int_{\alpha}\int_{\beta}\alpha^{\frac{3n}{2}+a-1}\beta^{n+c-1} e^{-\alpha \left(b+\sum_{i=1}^{n} x_i^{2\beta}\right) }e^{-d\beta}\left( \prod_{i=1}^{n}x_i^{3\beta-1}\right) ~d\alpha~ d\beta$\
From equation number (5.9), (5.10) it is easy to observed that the posterior expectations are appearing in the form of the ratio of two integrals. Thus, the analytical solution of these expetations are not presumable. Therefore, any numerical approximation techniques may be implemented to secure the solutions. Here, we used one of the most popular and quite effective approximation technique suggested by Lindley (1980). The detailed description can be seen in below; $$\begin{aligned}
\label{eq20}
(\hat{\alpha},\hat{\beta})_{Bayes}&=&\dfrac{\int_{\alpha}\int_{\beta}u(\alpha,\beta)e^{\rho(\alpha,\beta)+l}\quad d\alpha d\beta }{\int_{\alpha}\int_{\beta}e^{\rho(\alpha,\beta)+l}\quad d\alpha d\beta}\\&=&(\hat{\alpha},\hat{\beta})_{ml}+\dfrac{1}{2}[(u_{\alpha\alpha}+ 2 u_\alpha\rho_\alpha)\tau_{\alpha\alpha}+(u_{\alpha\beta}+2u_\alpha\rho_\beta)\tau_{\alpha\beta}+(u_{\beta\alpha}+ 2u_\beta\rho_\alpha)\tau_{\beta\alpha}\nonumber\\
&+&(u_{\beta\beta}+2u_\beta\rho_\beta)\tau_{\beta\beta}]+\dfrac{\alpha}{\beta}[(u_\alpha\tau_{\alpha\alpha}+u_\beta\tau_{\alpha\beta})(l_{111}\tau_{\alpha\alpha}+2l_{21}\tau_{\alpha\beta}+l_{12}\tau_{\beta\beta})\nonumber\\
&+&(u_\alpha\tau_{\beta\alpha}+u_\beta\tau_{\beta\beta})(l_{21}\tau_{\alpha\alpha}+2 l_{12}\tau_{\beta\alpha}+l_{222}\tau_{\beta\beta})]
\end{aligned}$$ where, $u(\alpha,\beta)=(\alpha,\beta)$, $\rho(\alpha,\beta)=\ln g(\alpha,\beta)$ and $l=\ln L(\alpha,\beta|\underbar x)$, $$\begin{aligned}
l_{ab}=\dfrac{\partial^{3}l}{\partial\alpha^{a}\partial\beta^{b}},\quad a,b=0,1,2,3\quad a+b=3,\quad
\rho_\alpha=\dfrac{\partial \rho}{\partial\alpha},\quad\rho_\beta=\dfrac{\partial \rho}{\partial\beta}\\ u_\alpha=\dfrac{\partial u}{\partial\alpha},\quad u_\beta=\dfrac{\partial u}{\partial \beta},\quad u_{\alpha\alpha}=\dfrac{\partial^2 u}{\partial\alpha^2},\quad u_{\beta\beta}=\dfrac{\partial^2 u}{\partial\beta^2},\quad u_{\alpha\beta}=\dfrac{\partial^2 u}{\partial\alpha\partial\beta},
\end{aligned}$$ $$\tau_{\alpha\alpha}=\dfrac{1}{l_{20}},\,\tau_{\alpha\beta}=\dfrac{1}{l_{11}}=\tau_{\beta\alpha},\, \tau_{\beta\beta}=\dfrac{1}{l_{02}}$$
since, $u(\alpha,\beta$ is the function of $\alpha,\beta$ both. Therefore,
- If $u(\alpha,\beta)=\alpha$ in (5.12) then; $$\begin{aligned}
u_\alpha&=1,\quad u_\beta=0, \quad u_{\alpha\alpha}=u_{\beta\beta}=0,\quad u_{\alpha\beta}=u_{\beta\alpha}=0
\end{aligned}$$
- If $u(\alpha,\beta)=\beta$ in (5.12) then; $$\begin{aligned}
u_\beta&=1,\quad u_\alpha=0, \quad u_{\alpha\alpha}=u_{\beta\beta}=0,\quad u_{\alpha\beta}=u_{\beta\alpha}=0\\
\end{aligned}$$
and the rest derivatives based on likelihood function are as obtained as; $$l_{30}=\dfrac{3n}{\alpha^3},~~l_{11}=-2\sum_{i=1}^{n}x_i^{2\beta}\ln x_i, ~~
l_{03}=\dfrac{2n}{\beta^3}-8\alpha\sum_{i=1}^{n}x_i^{2\beta}(\ln x_i)^{3}$$ $$l_{12}=-4\sum_{i=1}^{n}x_i^{2\beta}(\ln x_i)^{2}=l_{21}$$ Using these derivatives the Bayes estimates of $(\alpha,\beta)$ are obtained by following expressions $$\begin{split}
\hat{\alpha}_{bl}=&\hat{\alpha}_{ml}+\dfrac{1}{2}[(2u_\alpha\rho_\alpha)\tau_{\alpha\alpha}+(2u_\alpha\rho_\beta)\tau_{\alpha\beta}]+\dfrac{1}{2}[(u_\alpha\tau_{\alpha\alpha})(l_{30}\tau_{\alpha\alpha}+2l_{21}\tau_{\alpha\beta}+l_{12}\tau_{\beta\beta})\\&+(u_\alpha\tau_{\beta\alpha})(l_{21}\tau_{\alpha\alpha}+2l_{12}\tau_{\beta\alpha}+l_{03}\tau_{\beta\beta})]
\end{split}$$ $$\begin{split}
\hat{\beta}_{bl}&=\hat{\beta}_{ml}+\dfrac{1}{2}[(2u_\beta\rho_\alpha)\tau_{\beta\alpha}+(2u_\beta\rho_\beta)\tau_{\beta\beta}]+\dfrac{1}{2}[(u_\beta\tau_{\alpha\beta})(l_{30}\tau_{\alpha\alpha}+2l_{21}\tau_{\alpha\beta}+l_{12}\tau_{\beta\beta})\\&+(u_\beta\tau_{\beta\beta})(l_{21}\tau_{\alpha\alpha}+2l_{12}\tau_{\beta\alpha}+l_{03}\tau_{\beta\beta})]
\end{split}$$
Simulation Study
================
In this section, Monte Carlo simulation study has been performed to assess the performance of the obtained estimators in terms of their mean square error (MSEs). The maximum likelihood estimates of the parameters are evaluated by using $nlm () $ function, and MLEs of reliability characteristics are obtained by using invariance properties. The Bayes estimates of the parameter are evaluated by Lindley’s approximation technique. The hyper-parameters values are chosen in such a way that the prior mean is equal to the true value, and prior variance is taken as very small, say 0.5. All the computations are done by $R 3.4.1$ software. At first, we generated 5000 random samples from PMaD $(\alpha,\beta)$ using Newton-Raphson algorithm for different variation of sample sizes as $n=10$ (small), $n=20, 30$ (moderate), $n=50$ (large) for fixed $(\alpha=0.75, \beta=0.75)$ and secondly for different variation of $(\alpha, \beta)$ when sample size is fixed say $(n=20)$ respectively. Average estimates and mean square error (MSE) of the parameters and reliability characteristics are calculated for the above mentioned choices, and the corresponding results are reported in Table 2. The asymptotic confidence interval (ACI) and asymptotic confidence length (ACL) are also obtained and presented in Table 3. From this extensive simulation study, it has been observed that the precision of MLEs and Bayes estimator are increasing when the sample size is increasing while average ACL is decresing. Further, the Bayes estimators are more precise as compared ML estimators for all considered cases.
[ccccccccc]{}\
n & $\alpha, \beta$ & $\alpha_{ml}$& $\beta_{ml}$ & $MTTF_{ml}$ &$R (t)_{ml}$& $H(t)_{ml}$ & $\alpha_{bl}$ & $\beta_{bl}$\
\[0\][\*]{}[10]{} & \[0\][\*]{}[0.75,0.75]{} & 0.5070 & 1.1598 & 1.5119 & 0.9691 & 0.1663 & 0.5063 & 1.1028\
& & 0.0631 & 0.2588 & 0.0164 & 0.0049 & 0.0947 & 0.0631 & 0.2027\
\[0\][\*]{}[20]{} & & 0.6560 & 0.8848 & 1.4922 & 0.9343 & 0.2965 & 0.6521 & 0.8647\
& & 0.0098 & 0.0326 & 0.0093 & 0.0014 & 0.0703 & 0.0105 & 0.0263\
\[0\][\*]{}[30]{} & & 0.7096 & 0.8064 & 1.4883 & 0.9163 & 0.3504 & 0.7058 & 0.7951\
& & 0.0022 & 0.0103 & 0.0071 & 0.0004 & 0.0010 & 0.0025 & 0.0087\
\[0\][\*]{}[50]{} & & 0.7542 & 0.7453 & 1.4869 & 0.8988 & 0.3968 & 0.7514 & 0.7397\
& & 0.0003 & 0.0031 & 0.0046 & 0.0001 & 0.0003 & 0.0003 & 0.0031\
\
\[0\][\*]{}[20]{} & \[0\][\*]{}[0.5,0.75]{} & 0.6603 & 0.6832 & 1.7380 & 0.9044 & 0.3400 & 0.6574 & 0.6716\
& & 0.0261 & 0.0125 & 0.0585 & 0.0017 & 0.0099 & 0.0252 & 0.0117\
& \[0\][\*]{}[0.5, 1.5]{} & 0.7290 & 0.3033 & 4.6222 & 0.7871 & 0.3556 & 0.7258 & 0.3229\
& & 0.0528 & 1.4330 & 11.9171 & 0.0402 & 0.1139 & 0.0513 & 1.3866\
& \[0\][\*]{}[1.5, 0.5]{} & 0.5090 & 2.9297 & 1.1531 & 0.9983 & 0.0207 & 0.5517 & 2.8634\
& & 0.9907 & 6.6465 & 0.0242 & 0.1274 & 26.0695 & 0.9087 & 6.3006\
& \[0\][\*]{}[2.5,2.5]{} & 1.0448 & 0.5958 & 1.4084 & 0.7953 & 0.6393 & 1.2825 & 0.6727\
& & 2.1402 & 3.6573 & 0.3860 & 0.0373 & 0.3553 & 1.5058 & 3.3715\
\[tab:addlabel\]
[cccccccc]{}\
n & $\alpha, \beta$ & $\alpha_L$ & $\alpha_U$ & $ACL_\alpha$ & $\beta_L$& $\beta_U$ & $ACl_\beta$\
10 & 0.75,0.75 & 0.0874 & 0.9266 & 0.8393 & 0.5711 & 1.7485 & 1.1775\
20 & 0.75,0.75 & 0.3209 & 0.9911 & 0.6703 & 0.5525 & 1.2171 & 0.6646\
30 & 0.75,0.75 & 0.4263 & 0.9928 & 0.5665 & 0.5555 & 1.0574 & 0.5019\
50 & 0.75,0.75 & 0.5290 & 0.9794 & 0.4505 & 0.5631 & 0.9275 & 0.3644\
\
\[0\][\*]{}[20]{} & 0.5, 0.75 & 0.3255 & 0.9951 & 0.6696 & 0.4142 & 0.9523 & 0.5381\
& 0.5, 1.5 & 0.3794 & 1.0785 & 0.6991 & 0.4819 & 1.7425 & 1.2429\
& 1.5, 0.5 & 0.4206 & 1.7812 & 0.76058 & 0.2260 & 1.8334 & 1.3807\
& 2.5, 2.5 & 0.5804 & 2.9509 & 0.9788 & 0.54133 & 2.7783 & 1.1365\
\[tab:addlabel\]
Real Data Illustration
======================
This section, demonstrate the practical applicability of the proposed model in real life scenario especially for the survival/relibaility data taken from diffierent sources. The proposed distribution is compared with Maxwell distribution (MaD) and its different generalizations, such as length biased maxwell distribution (LBMaD), area biased maxwell distribution (ABMaD), extended Maxwell distribution (EMaD) and generalized Maxwell distribution (EMaD). For these models the estimates of the parameter (s) are obtained by method of maximum likelihood and the compatibility of PMaD has been discussed using model selection tools such as log-likelihood (-log L), Akaike information criterion (AIC), corrected Akaike information criterion (AICC), Bayesian information criterion (BIC) and Kolmogorov Smirnov (K-S) test. In general, the smaller values of these statistics indicate, the better fit to the data. The point estimates of the parameters and reliability function and hazard function for each data set are reported in Table 6. The interval estimate of the parameter and corresponding asimptotic confidence length are also evaluated and presented in Table 7.
**Data Set-I (Bladder Cancer Data):** This data set represents the remission times (in months) of a 128 bladder cancer patients, and it was initially used by Lee and Wang (2003). The same data set is used to show the superiority of extended maxwell distribution by Sharma et al (2017).\
**Data Set-II : Item Failure Data**\
This dataset is taken from Murthy et al. (2004). It shows 50 items put into use at t = 0 and failure items are recorded in weeks.\
**Data Set-III:** The data set was initially considered by Chhikara and Folks (1977). It represent the 46 repair times (in hours) for an airborne communication transceiver.\
**Data Set-IV: Flood data**\
The data are the exceedances of flood peaks (in m3/s) of the Wheaton River near Carcross in Yukon Territory, Canada. The data consist of 72 exceedances for the years 1958–1984, rounded to one decimal place. This data was analyzed by Choulakian and Stephens (2011).
[cccccccc]{}\
\
Model & $ \hat{\alpha} $ & $ \hat{\beta} $ & -logL & AIC & AICC & BIC & K-S\
**PMaD** & **0.7978** & **0.1637** & **366.3820** & **736.7639** & **732.8599** & **742.4680** & **0.3675**\
MaD & 0.0076 &– & 1014.4440 & 2030.8870 & 2028.9190 & 2033.7400 & 0.4144\
LBMaD & 98.6386 & – & 669.3668 & 1340.7340 & 1338.7650 & 1343.5860 & 0.4906\
ABMaD & 78.9109 & – & 767.8122 & 1537.6240 & 1535.6560 & 1540.4770 & 0.5608\
ExMaD & 0.8447 & 1.4431 & 412.1232 & 828.2464 & 824.3424 & 833.9504 & 0.8265\
GMaD & 0.7484 & 527.2314 & 426.6019 & 857.2037 & 853.2997 & 862.9078 & 0.7086\
\
Model & $ \hat{\alpha} $ & $ \hat{\beta} $ & -logL & AIC & AICC & BIC & K-S\
**PMaD** & **0.8339** & **0.1820** & **135.8204** & **275.6407** & **271.8961** & **279.4648** & **0.2625**\
MaD & 0.0104 & – & 367.8528 & 737.7056 & 735.7890 & 739.6177 & 0.4268\
LBMaD & 72.1146 & – & 315.1624 & 632.3248 & 630.4081 & 634.2368 & 0.5112\
ABMaD & 57.6917 & – & 374.1247 & 750.2494 & 748.3328 & 752.1615 & 0.5825\
ExMaD & 0.6186 & 1.0139 & 151.2998 & 306.5996 & 302.8550 & 310.4237 & 0.7327\
GMaD & 0.5400 & 534.1569 & 151.2643 & 306.5287 & 302.7840 & 310.3527 & 0.3920\
\
Model & $ \hat{\alpha} $ & $ \hat{\beta} $ & - logL & AIC & AICC & BIC & K-S\
**PMaD** & **0.8735** & **0.2709** & **101.9125** & **207.8249** & **204.1040** & **211.4822** & **0.2136**\
MaD & 0.0406 & – & 245.1383 & 492.2766 & 490.3675 & 494.1052 & 0.5027\
LBMaD & 18.4603 & – & 237.4945 & 476.9890 & 475.0799 & 478.8176 & 0.5771\
ABMaD & 14.7683 & – & 284.7017 & 571.4034 & 569.4943 & 573.2320 & 0.6324\
ExMaD & 0.7290 & 0.8672 & 103.3052 & 210.6104 & 206.8895 & 214.2677 & 0.2989\
GMaD & 0.6015 & 122.7666 & 110.8521 & 225.7042 & 221.9833 & 229.3615 & 0.4392\
\
Model & $ \hat{\alpha} $ & $ \hat{\beta} $ & - logL & AIC & AICC & BIC & K-S\
**PMaD** & 0.805185 & 0.1504145 & 212.8942 & 429.7884 & 425.9623 & 434.3418 & 0.2760\
MaD & 0.005032 & – & 610.9235 & 1223.847 & 1221.904 & 1226.124 & 0.3821\
LBMaD & 149.0315 & – & 426.3076 & 854.6153 & 852.6724 & 856.8919 & 0.4113\
ABMaD & 119.2252 & – & 493.3271 & 988.6543 & 986.7114 & 990.9309 & 0.4529\
ExMaD & 0.697471 & 1.306933 & 251.9244 & 507.8487 & 504.0226 & 512.4021 & 0.7487\
GMaD & 0.648149 & 919.7356 & 251.2767 & 506.5534 & 502.7273 & 511.1068 & 0.4998\
\[tab:addlabel\]
Data Min Q1 Q2 Mean Q3 Max Kurtosis Skewness
------ ------- ------- ------- -------- -------- -------- ---------- ----------
I 0.080 3.348 6.395 9.366 11.838 79.050 18.483 3.287
II 0.013 1.390 5.320 7.821 10.043 48.105 9.408 2.306
III 0.200 0.800 1.750 3.607 4.375 24.500 11.803 2.888
IV 0.100 2.125 9.500 12.204 20.125 64.000 5.890 1.473
: Summary of the data sets
\[tab:addlabel\]
Data $ \alpha_{ml}$ $ \beta_{ml}$ $ MTTF_{ml} $ $ R(t)_{ml} $ $ H(t)_{ml} $ $ \alpha_{bl}$ $ \beta_{bl}$
------ ---------------- --------------- --------------- --------------- --------------- ---------------- ---------------
I 0.7978 0.1637 28.2109 0.7019 0.2827 0.7962 0.1639
II 0.8339 0.1820 15.3594 0.6953 0.3224 0.8292 0.1821
III 0.8735 0.2709 4.0773 0.7212 0.4326 0.8675 0.2703
IV 0.8052 0.1504 42.8622 0.6923 0.2696 0.8023 0.1506
: Real data estimates
\[tab:addlabel\]
Data $ \alpha_L $ $ \alpha_U $ $ ACL_\alpha $ $ \beta_L $ $ \beta_U $ $ ACL_\beta $
------ -------------- -------------- ---------------- ------------- ------------- ---------------
I 0.6545 0.9411 0.2866 0.1373 0.1902 0.0529
II 0.5962 1.0717 0.4754 0.1376 0.2263 0.0888
III 0.6202 1.1269 0.5067 0.2081 0.3337 0.1256
IV 0.6126 0.9978 0.3852 0.1186 0.1822 0.0636
: Interval estimates based on real data
\[tab:addlabel\]
{width="6.5in" height="3in"}
{width="6.5in" height="3in"}
{width="6.5in" height="3in"}
{width="6.5in" height="3in"}
From the tabulated value of different model selection tools, -LogL, AIC, AICC, BIC, and K-S values it has been noticed that PMaD has least -LogL, AIC, AICC, BIC, and K-S. The empirical cumulative distribution function and Q-Q plots are also given in Figure 5, 6 & 7 respectively. Therefore, PMaD can be recommended as a good alternative to the existing family of Maxwell distribution. Summary of the considered data sets is given in Table 2 and seen that skewness is positive for all data sets which indicates that it has positive skewness which appropriately suited to the proposed model.
Conclusion
==========
This article proposed power Maxwell distribution (PMaD) as an extension of Maxwell distribution and studied its different mathematical and statistical properties, such as reliability characteristics, moments, median, mode, mean deviation, generating functions, stochastic ordering, residual functions, entropy etc. We also study the skewness and kurtosis of the PMaD and found that it is capable of modeling the positively skewed as well as symmetric data sets. The unknown parameters of the PMaD are estimated by maximum likelihood estimation and Bayes estimation method. The MLEs of the reliability function and hazard function are also obtained by using invariance property. The 95% asymptotic confidence interval for the parameter are constructed using Fisher information matrix. The MLEs and Bayes estimators are compared through the Monte Carlo simulation and observed that Bayes estimators are more precise under informative prior. Finally, medical/reliability data have been used to show practical utility of the power Maxwell distribution, and it is observed that PMaD provides the better fit as compared to other Maxwell family of distributions. Thus, it can be recommended as an alternative model for the non-monotone failure rate model.
[20]{} Maxwell, J., 1860. On the dynamical theory of gases, presented to the meeting of the british association for the advancement of science. Scientific Letters I, 616. Gupta, R. C., Gupta, R. D. and Gupta, P. L. (1998): Modeling failure time data by Lehman alternatives. Communication in Statistics-Theory and Methods, 27 (4), 887?-904. Lee, E. T. and Wang, J. W. (2003): Statistical Methods for Survival Data Analysis. Wiley, New York, DOI:10.1002/0471458546. Vikas Kumar Sharma, Hassan S. Bakouch & Khushboo Suthar (2017) An extended Maxwell distribution: Properties and applications, Communications in Statistics - Simulation and Computation, 46:9, 6982-7007, DOI: 10.1080/03610918.2016.1222422. Vikas Kumar Sharma, Sanku Dey, Sanjay Kumar Singh & Uzma Manzoor (2017): On Length and Area biased Maxwell distributions, Communications in Statistics - Simulation and Computation, DOI: 10.1080/03610918.2017.1317804. Bekker, A., and J. J. Roux. 2005. Reliability characteristics of the Maxwell distribution: A Bayes estimation study. Communications in Statistics - Theory and Methods 34:2169-78. doi:10.1080/STA-200066424. Chaturvedi, A., and U. Rani. 1998. Classical and Bayesian reliability estimation of the generalized Maxwell failure distribution. Journal of Statistical Research 32:113-20. Lindley, D. V. Approximate Bayes method, Trabajos de estadistica, Vol. 31, 223-237, 1980. Dey, S., and S. S. Maiti. 2010. Bayesian estimation of the parameter of Maxwell distribution under different loss functions. Journal of Statistical Theory and Practice 4:279–87. doi:10.1080/ 15598608.2010.10411986. Modi, K. 2015. Length-biased weighted Maxwell distribution. Pakistan Journal of Statistics and Operation Research 11:465–72. doi:10.18187/pjsor.v11i4.1008. Saghir, A., and A. Khadim. 2016. The mathematical properties of length biased Maxwell distribution. Journal of Basic and Applied Research International 16:189–95. Bonferroni C. E. ( 1930), Elementi di Statistica General, Seeber, Firenze. Gross A. J. and Clark V. A. (1975) Survival distributions: reliability application in the biomedical sciences. New Yark: John Wiley and Sons. Tyagi, R. K., and S. K. Bhattacharya. 1989. A note on the MVU estimation of reliability for the Maxwell failure distribution. Estadistica 41:73–79. Lin C., Duran B. S. and Lewis T. O. (1989), Inverted gamma as a life distribution. Microelectron. Reliab. 29 (4):619-626. Renyi A. (1961). On measures of entropy and information, in: Proceedings of the 4th Berkeley Symposium on Mathematical Statistics and Probability, University of California Press, Berkeley. Tomer, S. K., and Panwar, M. S. 2015. Estimation procedures for Maxwell distribution under type I progressive hybrid censoring scheme. Journal of Statistical Computation and Simulation 85:339–56.
[^1]: Corresponding author E-mail: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Inspired by the classical category theorems of Halmos and Rohlin for discrete measure preserving transformations, we prove analogous results in the abstract setting of unitary and isometric $C_0$-semigroups on a separable Hilbert space. More presicely, we show that the set of all weakly stable unitary groups (isometric semigroups) is of first category, while the set of all almost weakly stable unitary groups (isometric semigroups) is residual for an appropriate topology.'
address:
- 'Tanja Eisner Mathematisches Institut, Universität TübingenAuf der Morgenstelle 10, D-72076, Tübingen, Germany'
- 'András Serény Department of Mathematics and its Applications, Central European University,Nádor utca 9, H-1051 Budapest, Hungary '
author:
- Tanja Eisner
- András Serény
title: Category theorems for stable semigroups
---
[^1]
Introduction
============
In 1944 Halmos [@halmos:1944] showed that a “typical” dynamical system is weakly mixing, or, more precisely, the set of all weakly mixing transformations on a measure space is residual and hence of second category. Four years later, Rohlin [@rohlin:1948] showed that the set of all strongly mixing transformation is of first category (see also Halmos [@halmos:1956 pp. 77–80]). In this way they proved the existence of weakly but not strongly mixing systems (see e.g. Petersen [@petersen:1983 Section 4.5] for later concrete examples and for a method to construct such systems). In the time continuous case, i.e., for measure preserving flows, analogous results are not stated in the standard literature. However, Bartoszek and Kuna [@bartoszek/kuna:2006] recently proved a category theorem for Markov semigroups on the Schatten class $C_1$, while an analogous result for stochastic semigroups can be found in Lasota, Myjak [@lasota/myrjak:1992]. In this paper we consider the time continuous case in the more general setting of unitary and isometric semigroups on a Hilbert space and prove corresponding category results.
Recall that a (continuous) dynamical system $(\Omega, \mu, (\varphi_t)_{t\geq 0})$ induces a unitary/ isometric $C_0$-(semi)group $(T(t))_{t\geq 0}$ on the Hilbert space $H:=L^2(\Omega,\mu)$ via the formula $T(t)f(\omega)=f(\varphi_t (\omega))$. Moreover, the decomposition into invariant subspaces $H=\langle \mathbf{1}\rangle \oplus H_0$ holds for $H_0=\{f\in H: \int_\Omega f d\mu=0\}$. Recall further that the flow $(\varphi_t)$ is strongly mixing if and only if the restricted semigroup $(T_0(t))_{t\geq 0}$ on $H_0$ is weakly stable, i.e., $\lim_{t\to\infty}\langle T_0(t)x,y\rangle=0$ for every $x,y\in H_0$. Analogously, weak mixing of the flow corresponds to so-called *almost weak stability* of the semigroup $T_0(\cdot)$, i.e., $$\lim\langle T_0(t)x,y\rangle=0, \quad t\to\infty,\ t\in M, \quad \text{for every } x,y\in H_0$$ for a set $M\subset {\mathbb{R}}_+$ with (asymptotical) density one, i.e., with $d(M):=\lim_{t\to\infty}\frac{\lambda(M\cap[0,t])}{t}=1$ for the Lebesgue measure $\lambda$. We refer to e.g. Cornfeld, Fomin, Sinai [@cornfeld/fomin/sinai:1982 Section 1.7] for details.
We now consider an arbitrary Hilbert space $H$ and a contractive $C_0$-semigroup $T(\cdot)$ on $H$ with generator $A$. By the classical decomposition theorem of Jacobs–Glicksberg–de Leeuw (see Glicksberg, De Leeuw [@glicksberg/deleeuw:1961] or Engel, Nagel [@engel/nagel:2000 Theorem V.2.8]), $H$ decomposes into the so-called reversible and stable parts $$\begin{aligned}
H_r&:=&\overline{\text{lin}}\{x: Ax=i\alpha x \text{ for some } \alpha
\in{\mathbb{R}}\}, \\
H_s&:=&\{x: 0 \text { is a weak accumulation point of } \{T(t)x,t\geq 0\}\}.\end{aligned}$$ (Note that the Jacobs–Glicksberg–de Leeuw decomposition is valid for every bounded semigroup on a reflexive Banach space or, more generally, for every relatively weakly compact semigroup.) A result of Hiai [@hiai:1978] being a continuous version of discrete results of Nagel [@nagel:1974] and Jones, Lin [@jones/lin:1976] shows that the semigroup $T(\cdot)$ restricted to $X_s$ is almost weakly stable as defined above. (For a survey on weak and almost weak stability see Eisner, Farkas, Nagel, Serény [@EFNS].)
Weak mixing (almost weak stability) can be characterised by a simple spectral condition, while the more natural property of strong mixing (weak stability) *“is, however, one of those notions, that is easy and natural to define but very difficult to study...”*, see the monograph of Katok, Hasselblatt [@katok/hasselblatt:1995 p. 748].
In this paper we extend the results of Halmos and Rohlin and show that weak and almost weak stability differ fundamentally. More precisely, we prove that for an appropriate (and natural) topology the set of all weakly stable unitary groups on a separable infinite-dimensional Hilbert space is of the first category and the set of all almost weakly stable unitary groups is residual and hence of the second category (Section 2). An analogous result holds for isometric semigroups (Section 3). In Section 4 we discuss the case of contractive semigroups.
For analogous results on the time discrete case, see Eisner, Serény [@eisner/sereny:2006].
Unitary case {#section:unitary-gr}
============
Consider the set ${\mathcal{U}}$ of all unitary $C_0$-groups on an infinite-dimensional Hilbert space $H$. The following density result on periodic unitary groups is a first step in our construction.
\[prop:periodic\] For every $N\in{\mathbb{N}}$ and every unitary group $U(\cdot)$ on $H$ there is a sequence $\{V_n(\cdot)\}_{n=1}^\infty$ of periodic unitary groups with period greater than $N$ such that $\lim_{n\to\infty}\|U(t)-V_n(t)\|=0$ uniformly on compact time intervals.
Take $U(\cdot)\in {\mathcal{U}}$ and $n\in {\mathbb{N}}$. By the spectral theorem (see, e.g., Halmos [@halmos:1963]), $H$ is isomorphic to $L^2(\Omega, \mu)$ for some locally compact space $\Omega$ and measure $\mu$ and $U(\cdot)$ is unitary equivalent to a multiplication group $\tilde{U}(\cdot)$ with $$(\tilde{U}(t)f)(\omega)=e^{itq(\omega)}f(\omega),\quad \forall
\omega\in\Omega,\ t\in {\mathbb{R}}, \ f\in L^2(\Omega, \mu)$$ for some measurable $q:\Omega \to {\mathbb{R}}$.
We approximate the (unitary) group $\tilde{U}(\cdot)$ as follows. Take $n>N$ and define $$q_n(\omega):= \frac{2\pi j}{n},\quad \forall \omega\in q^{-1}\left(\left[\frac{2\pi
j}{n}, \frac{2\pi (j+1)}{n}\right]\right),\ j\in{\mathbb{Z}}.$$ Denote now by $\tilde{V}_n(t)$ the multiplication operator with $e^{itq_n(\cdot)}$ for every $t\in{\mathbb{R}}$. The unitary group $\tilde{V}_n(\cdot)$ is periodic with period greater than or equal to $n$ and therefore $N$. Moreover, $$\begin{aligned}
\Vert \tilde{U}(t)-\tilde{V}_n(t) \Vert \leq \sup_{\omega} |e^{itq(\omega)} - e^{itq_n(\omega)}|
\leq |t| \sup_{\omega} |q(\omega)-q_n(\omega)| \leq \frac{2\pi |t|}{n} \underset{n\to\infty}{\longrightarrow} 0
$$ uniformly in $t$ on compact intervals and the proposition is proved.
By a modification of the proof of Proposition \[prop:periodic\] one can show that for every $N\in{\mathbb{N}}$ the set of all periodic unitary groups with period greater than $N$ with bounded generators is dense in ${\mathcal{U}}$ with respect to the strong operator topology uniform on compact time intervals.
For the second step we need the following lemma. From now on we assume the Hilbert space $H$ to be separable.
\[lemma:discrete-appr-of-I\] Let $H$ be a separable infinite-dimensional Hilbert space. Then there exists a sequence $\{U_n(\cdot )\}_{n=1}^\infty$ of almost weakly stable unitary groups with bounded generator satisfying $\lim_{n\to\infty}\Vert U_n(t) - I\Vert= 0$ uniformly in $t$ on compact intervals.
By the isomorphism of all separable infinite-dimensional Hilbert spaces we can assume without loss of generality that $H=L^2({\mathbb{R}})$ with respect to the Lebesgue measure. Take $n\in{\mathbb{N}}$ and define $U_n(\cdot)$ on $L^2({\mathbb{R}})$ by $$(U_n(t)f)(s):= e^{\frac{itq(s)}{n}}f(s), \ \ s\in{\mathbb{R}},\ \ f\in L^2({\mathbb{R}}),$$ where $q:{\mathbb{R}}\to (0,1)$ is a strictly monotone increasing function.
Then all $U_n(\cdot)$ are almost weakly stable by the theorem of Jacobs–Glicksberg–de Leeuw and we have $$\Vert U_n(t) - I \Vert = \sup_{s\in{\mathbb{R}}} |e^{\frac{itq(s)}{n}} -
1| \leq [\text{for } t\leq \pi n]\leq |e^{\frac{it}{n}} - 1| \leq \frac{2t}{n} \to 0, \quad n\to \infty,$$ uniformly on t in compact intervals.
The metric we introduce now on the space ${\mathcal{U}}$ is given by $$d(U(\cdot),V(\cdot)):= \sum_{n,j=1}^\infty
\frac{\sup_{t\in[-n,n]}\|U(t)x_j -V(t)x_j\|}
{2^{j+n} \|x_j\|}\quad \text{for } U,V\in {\mathcal{U}},$$ where $\{x_j\}_{j=1}^\infty$ some fixed dense subset of $H$ with all $x_j\neq 0$. Note that the topology coming from this metric is a continuous analogue of the so-called strong\* operator topology for operators, see, e.g., Takesaki [@takesaki p. 68]), and the corresponding convergence is the strong convergence uniform on compact time intervals of (semi)groups and their adjoints. This metric makes ${\mathcal{U}}$ a complete metric space. We denote by $\mathcal{S_U}$ the set of all weakly stable unitary groups on $H$ and by $\mathcal{W_U}$ the set of all almost weakly stable unitary groups on $H$. The following proposition shows the density of $\mathcal{W_U}$ which will play an important role later.
\[prop:almweakstab\] The set $\mathcal{W_U}$ of all almost weakly stable unitary groups with bounded generators is dense in ${\mathcal{U}}$.
By Proposition \[prop:periodic\] it is enough to approximate periodic unitary groups by almost weakly stable unitary groups. Let $U(\cdot)$ be a periodic unitary group with generator $A$ and period $\tau$. Take $\varepsilon>0$, $n\in{\mathbb{N}}$, $x_1,\ldots, x_n\in H$ and $t_0>0$. We have to find an almost weakly stable unitary group $T(\cdot)$ with $\Vert U(t)x_j - T(t)x_j \Vert \leq \varepsilon$ for all $j=1,\ldots, n$ and $t\in[-t_0,t_0]$.
By Engel, Nagel [@engel/nagel:2000 Theorem IV.2.26] we have the orthogonal space decomposition $$\label{eq:periodic-decomp}
\displaystyle H=\overline{\bigoplus_{k\in{\mathbb{Z}}}
\ker\left(A-\frac{2\pi
ik}{\tau}\right)}.$$ Assume first that $\{x_j\}_{j=1}^n$ is an orthonormal system of eigenvectors of $A$.
Our aim is to use Lemma \[lemma:discrete-appr-of-I\]. For this purpose we first construct a unitary group $V(\cdot)$ which coincides with $U(\cdot)$ on every $x_j$ and has infinite-dimensional eigenspaces only. Define the $U(\cdot)$-invariant subspace $H_0:=\text{lin}\{x_1, \ldots,
x_n\}$ and the unitary group $V_0(\cdot):=U(\cdot)|_{H_0}$ on $H_0$. Since $H$ is separable, we can decompose $H$ in an orthogonal sum $$\displaystyle H=\bigoplus_{k=0}^\infty H_k \quad \text{with } \dim H_k=\dim H_0 \text{ for every } k\in{\mathbb{N}}.$$ Denote by $P_k$ an isomorphism from $H_k$ to $H_0$ and define $V_k(\cdot):=P_k^{-1}V_0(\cdot)P_k$ on $H_k$ being copies of $V_0(\cdot)$. Consider now the unitary group $\displaystyle V(\cdot):=\bigoplus_{k=0}^\infty V_k (\cdot)$ on $H$ which is periodic by Engel, Nagel [@engel/nagel:2000 Theorem IV.2.26]. (Note that its generator is even bounded.)
For the periodic unitary group $V(\cdot)$ the space decomposition analogous to (\[eq:periodic-decomp\]) holds. Moreover there are only finitely many eigenspaces (less or equal to $n$, depending on $x_1,...,x_n$) and they all are infinite-dimensional by the construction. Applying Lemma \[lemma:discrete-appr-of-I\] to each eigenspace we find an almost weakly stable unitary group $T(\cdot)$ with $\Vert U(t)x_j - T(t)x_j \Vert \leq \varepsilon$ for all $j=1,\ldots, n$ and $t\in[-t_0,t_0]$.
Take now arbitrary $x_1,\ldots,x_n\in H$. Take further an orthonormal basis of eigenvectors $\{y_k\}_{k=1}^\infty$. Then there exists $N\in {\mathbb{N}}$ such that $x_j=\sum_{k=1}^N a_{jk} y_k + o_j$ with $\|o_j\|<\frac{\varepsilon}{4}$ for every $j=1,\ldots,n$.
We can apply the arguments above to $y_1,\ldots,y_N$ and find an almost weakly stable unitary group $T(\cdot)$ with $\|U(t)y_k-T(t)y_k\|< \frac{\varepsilon}{4N M}$ for $M:=\max_{ k=1,\ldots,N, j=1,\ldots,n}{|a_{jk}|}$ and every $k=1,\ldots,N$ and $t\in[-t_0,t_0]$. Hence $$\|U(t)x_j-T(t)x_j\|\leq \sum_{k=1}^N |a_{jk}|\|U(t)y_k-T(t)y_k\|+2\|o_j\|< \varepsilon$$ for every $j=1,\ldots,n$ and $t\in[-t_0,t_0]$, and the proposition is proved.
We are now ready to prove a category theorem for weakly and almost weakly unitary groups. For this purpose we extend the argument used in the proof of the category theorems for operators induced by measure preserving transformation in ergodic theory (see Halmos [@halmos:1956 pp. 77–80]).
\[thm:unitary\] The set $\mathcal{S_U}$ of weakly stable unitary groups is of first category and the set $\mathcal{W_U}$ of almost weakly stable unitary groups is residual in ${\mathcal{U}}$.
We first prove the first part of the theorem. Fix $x \in H$ with $\|x\|=1$ and consider the sets
$$M_t:=\left\{U(\cdot)\in {\mathcal{U}}:\ |\langle U(t) x,x\rangle| \leq \frac{1}{2} \right\}.$$
Note that all sets $M_t$ are closed.
For every weakly stable $U(\cdot)\in {\mathcal{U}}$ there exists $t>0$ such that $U\in M_s$ for all $s\geq t$, i.e., $\displaystyle U(\cdot)\in N_t:=\cap_{s\geq t} M_t$. So we obtain $$\mathcal{S_U} \subset \bigcup_{t>0} N_t.$$ Since all $N_t$ are closed, it remains to show that ${\mathcal{U}}\setminus N_t$ is dense for every $t$.
Fix $t>0$ and let $U(\cdot)$ be a periodic unitary group. Then $U(\cdot)\notin M_s$ for some $s\geq t$ and therefore $U(\cdot)\notin N_t$. Since by Proposition \[prop:periodic\] periodic unitary groups are dense in ${\mathcal{U}}$, $\mathcal{S}$ is of first category.
To show that $\mathcal{W_U}$ is residual we take a dense subspace $D=\{x_j\}_{j=1}^\infty$ of $H$ and define $$W_{jkt}:=\left\{U(\cdot)\in {\mathcal{U}}:\ |\left<U(t) x_j,x_j\right>| < \frac{1}{k} \right\}.$$ All these sets are open. Therefore the sets $\displaystyle W_{jk}:=\bigcup_{t>0} W_{jkt}$ are open as well.
We show that $$\label{W}
\mathcal{W_U}=\bigcap_{j,k=1}^\infty W_{jk}$$ holds.
The inclusion “$\subset$” follows from the definition of almost weak stability. To prove the converse inclusion we take $U(\cdot)\notin \mathcal{W_U}$. Then there exists $x\in H$ with $\|x\|=1$ and $\varphi \in
{\mathbb{R}}$ such that $U(t)x=e^{it\varphi} x$ for all $t>0$. Take $x_j\in D$ with $\|x_j-x\|\leq \frac{1}{4}$. Then $$\begin{aligned}
|\left<U(t) x_j,x_j \right>| &=& |\left<U(t) (x-x_j),x-x_j \right> + \left<U(t) x,x \right>
- \left<U(t) x,x-x_j \right> - \left<U(t) (x-x_j),x \right>| \\
&\geq& 1-\|x-x_j\|^2 -2\|x-x_j\|>\frac{1}{3}\end{aligned}$$ for every $t>0$. So $U(\cdot)\notin W_{j3}$ which implies $U(\cdot)\notin \cap_{j,k=1}^\infty W_{jk}$. Therefore equality (\[W\]) holds. Combining this with Proposition \[prop:almweakstab\] we obtain that $\mathcal{W_U}$ is residual as a dense countable intersection of open sets.
Isometric case {#section:isometric-sgr}
==============
In this section we extend the result from the previous section to isometric semigroups on $H$ which we again assume to be infinite-dimensional and separable.
Denote by $\mathcal{I}$ the set of all isometric $C_0$-semigroups on $H$. On $\mathcal{I}$ we consider the metric given by the formula $$d(T(\cdot),S(\cdot)):= \sum_{n,j=1}^\infty \frac{\sup_{t\in[0,n]}\|T(t)x_j -S(t)x_j\|}{2^{j+n} \|x_j\|}\quad \text{for } T(\cdot),S(\cdot) \in \mathcal{I},$$ where $\{x_j\}_{j=1}^\infty$ is a fixed dense subset of $H$. This corresponds to the strong convergence uniform on compact time intervals. The space $\mathcal{I}$ endowed with this metric is a complete metric space.
Analogously to Section \[section:unitary-gr\] we denote by $\mathcal{S_I}$ the set of all weakly stable and by $\mathcal{W_I}$ the set all almost weakly stable isometric semigroups on $H$.
The key for our results in this section is the following classical structure theorem on isometric semigroups on Hilbert spaces.
\[thm:Wold\]([*Wold decomposition*]{}, see [@sznagy/foias Theorem III.9.3].) Let $V(\cdot)$ be an isometric semigroup on a Hilbert space $H$. Then $H$ can be decomposed into an orthogonal sum $H=H_0 \oplus H_1$ of $V(\cdot)$-invariant subspaces such that the restriction of $V(\cdot)$ on $H_0$ is a unitary (semi)group and the restriction of $V(\cdot)$ on $H_1$ is a continuous unilateral shift, i.e., $H_1$ is unitarily equivalent to $L^2({\mathbb{R}}_+, Y)$ for some Hilbert space $Y$ such that the restriction of $V(\cdot)$ to $H_1$ is equivalent to the right shift semigroup on $L^2({\mathbb{R}}_+, Y)$.
We further need the following lemma, see also Peller [@peller] for the discrete version.
\[lemma:shift-appr\] Let $Y$ be a Hilbert space and let $R(\cdot)$ be the right shift semigroup on $H:=L^2({\mathbb{R}}_+, Y)$. Then there exists a sequence $\{U_n(\cdot)\}_{n=1}^\infty$ of periodic unitary (semi)groups on $H$ converging strongly to $R(\cdot)$ uniformly on compact time intervals.
For every $n\in{\mathbb{N}}$ we define $U_n(\cdot)$ by $$(U_n(t)f)(s):=
\begin{cases}
f(s), \quad &s\geq n;\\
R_n(t) f(s), \quad &s\in [0,n],
\end{cases}$$ where $R_n(\cdot)$ denotes the $n$-periodic right shift on the space $L^2([0,n],Y)$. Then every $U_n(\cdot)$ is a $C_0$-semigroup on $L^2({\mathbb{R}}_+,Y)$ which is isometric and $n$-periodic, and therefore unitary.
Fix $f\in L^2({\mathbb{R}}_+,Y)$ and take $0\leq t<n$. Then we have $$\begin{aligned}
\Vert U_n(t) f - R(t) f\Vert^2
= \int_{n-t}^n \|f(s)\|^2 ds +\int_n^\infty \|f(s)-f(s-t)\|^2 ds \leq 2 \int_{n-t}^\infty \|f(s)\|^2 ds \underset{n\to\infty}{\longrightarrow} 0
$$ uniformly in $t$ on compact intervals and the lemma is proved.
As a consequence of the Wold decomposition and Lemma \[lemma:shift-appr\] we obtain the following density result for periodic unitary (semi)groups in $\mathcal{I}$.
\[prop:periodic-i\] The set of all periodic unitary semigroups is dense in $\mathcal{I}$.
Let $V(\cdot)$ be an isometric semigroup on $H$. Then by Theorem \[thm:Wold\] the orthogonal decomposition $H=H_0\oplus
H_1$ holds, where the restriction $V_0(\cdot)$ of $V(\cdot)$ to $H_0$ is unitary, $H_1$ is unitarily equivalent to $L^2({\mathbb{R}}_+, Y)$ for some $Y$ and the restriction $V_1(\cdot)$ of $V(\cdot)$ on $H_1$ corresponds by this equivalence to the right shift semigroup on $L^2({\mathbb{R}}_+, Y)$. By Proposition \[prop:periodic\] and Lemma \[lemma:shift-appr\] we can approximate both semigroups $V_0(\cdot)$ and $V_1(\cdot)$ by unitary periodic ones and the assertion follows.
We further have a density result for almost weakly stable isometric semigroups.
\[prop:almweakstab-i\] The set $\mathcal{W_I}$ of almost weakly stable isometric semigroups is dense in $\mathcal{I}$.
Let $V$ be an isometry on $H,\ H_0,\ H_1$ the orthogonal subspaces from Theorem \[thm:Wold\] and $V_0$ and $V_1$ the corresponding restrictions of $V$. By Lemma \[lemma:shift-appr\] the operator $V_1$ can be approximated by unitary operators on $H_1$. The assertion now follows from Proposition \[prop:almweakstab\].
By using the same arguments as in the proof of Theorem \[thm:unitary\] we obtain with the help of Propositions \[prop:periodic-i\] and \[prop:almweakstab-i\] the following category result for weakly and almost weakly stable isometries.
\[thm:isometry\] The set $\mathcal{S_I}$ of all weakly stable isometric semigroups is of first category and the set $\mathcal{W_I}$ of all almost weakly stable isometric semigroups is residual in $\mathcal{I}$.
A remark on the contractive case
================================
It is not clear how to prove an analogue to Theorems \[thm:unitary\] and \[thm:isometry\] for contractive semigroups, while this is done in the discrete case in Eisner, Serény [@eisner/sereny:2006]. We point out one of the difficulties.
Let $\mathcal{C}$ denote the set of all contraction semigroups on $H$ endowed with the metric $$d(T(\cdot),S(\cdot)):= \sum_{n,i,j=1}^\infty \frac{\sup_{t\in[0,n]}|\left<T(t)x_i,x_j\right> -\left< S(t)x_i,x_j\right>|}{2^{i+j+n} \|x_i\| \|x_j\|}\quad \text{for }\ T,S \in \mathcal{C},$$ where $\{x_j\}_{j=1}^\infty$ is a fixed dense subset of $H$. The corresponding convergence is the weak convergence of semigroups uniform on compact time intervals. Note that this metric is a continuous analogue of the metric used in Eisner, Serény [@eisner/sereny:2006].
Since for $H:=l^2$ there exists a Cauchy sequence $\{T_n(\cdot)\}_{n=1}^\infty$ in $\mathcal{C}$ of semigroups (with bounded generators) such that the pointwise limit $S(\cdot)$ does not satisfy the semigroup law (see Eisner, Serény [@eisner/sereny:2007a]), the metric space $\mathcal{C}$ is *not* complete (or compact) in general, therefore the standard Baire category theorem cannot be applied. It is an open question whether $\mathcal{C}$ is at least Čech complete.
**Acknowledgement.** The authors are very grateful to Rainer Nagel and the referee for valuable comments.
[10]{}
W. Bartoszek and B. Kuna, *Strong mixing Markov semigroups on $C_1$ are meager*, Colloq. Math. **105** (2006), 311–317.
I. P. Cornfeld, S. V. Fomin, and Ya. G. Sinai, *Ergodic [T]{}heory*, Grundlehren der mathematischen Wissenschaften 245, Springer-Verlag, 1982.
K. de Leeuw and I. Glicksberg, *Applications of almost periodic compactifications*, Acta Math. **105** (1961), 63–97.
T. Eisner, B. Farkas, R. Nagel, and A. Serény, *Weakly and almost weakly stable $C_0$-semigroups*, Int. J. Dyn. Syst. Diff. Eq. **1** (2007), 44–57.
T. Eisner, A. Serény, *Category theorems for stable operators on Hilbert spaces*, Acta Sci. Math. (Szeged), to appear.
T. Eisner, A. Serény, *On a weak analogue of the Trotter-Kato theorem*, submitted, 2007.
K.-J. Engel and R. Nagel, *One-parameter [S]{}emigroups for [L]{}inear [E]{}volution [E]{}quations*, Graduate Texts in Mathematics, vol. 194, Springer-Verlag, New York, 2000.
S. R. Foguel, *Powers of a contraction in [H]{}ilbert space*, Pacific J. Math. **13** (1963), 551–562.
P. R. Halmos, *In general a measure preserving transformation is mixing*, Ann. Math. [**45**]{} (1944), 786–792.
P. R. Halmos, *Lectures on [E]{}rgodic [T]{}heory*, Chelsea Publishing Co., New York, 1956.
P. R. Halmos, *What does the spectral theorem say?*, Amer. Math. Monthly **70** (1963), 241–247.
F. Hiai, *Weakly mixing properties of semigroups of linear operators*, Kodai Math. J. **1** (1978), 376–393.
L. Jones, M. Lin, *Ergodic theorems of weak mixing type*, Proc. Am. Math. Soc. [**57**]{} (1976), 50–52.
A. Katok and B. Hasselblatt, *Introduction to the [M]{}odern [T]{}heory of [D]{}ynamical [S]{}ystems*, Encyclopedia of Mathematics and its Applications, vol. 54, Cambridge University Press, Cambridge, 1995.
A. Lasota and J. Myjak, *Generic properties of stochastic semigroups*, Bull. Polish Acad. Sci. Math. **40** (1992), 283–292.
R. Nagel, *Ergodic and mixing properties of linear operators*, Proc. Roy. Irish Acad. Sect. A. **74** (1974), 245–261.
V. V. Peller, *Estimates of operator polynomials in the space $L^p$ with respect to the multiplicative norm*, J. Math. Sciences [**16**]{} (1981), 1139–1149.
K. Petersen, *Ergodic [T]{}heory*, Cambridge Studies in Advanced Mathematics, Cambridge University Press, 1983.
V. A. Rohlin, *A “general” measure-preserving transformation is not mixing*, Doklady Akad. Nauk SSSR [**60**]{} (1948), 349–351.
B. Sz.-Nagy and C. Foia[ş]{}, *Harmonic [A]{}nalysis of [O]{}perators on [H]{}ilbert [S]{}pace*, North-Holland Publ. Comp, Akad[é]{}miai Kiad[ó]{}, Amsterdam, Budapest, 1970.
M. Takesaki, *Theory of [O]{}perator [A]{}lgebras I*, Springer-Verlag, 1979.
[^1]: The second author was supported by the DAAD-PPP-Hungary Grant, project number D/05/01422.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We explore the application of volumetric reconstruction from structured-light sensors in cognitive neuroscience, specifically in the quantification of the *size-weight illusion*, whereby humans tend to systematically perceive smaller objects as heavier. We investigate the performance of two commercial structured-light scanning systems in comparison to one we developed specifically for this application. Our method has two main distinct features: First, it only samples a sparse series of viewpoints, unlike other systems such as the Kinect Fusion. Second, instead of building a distance field for the purpose of points-to-surface conversion directly, we pursue a first-order approach: the distance function is recovered from its gradient by a *screened Poisson reconstruction*, which is very resilient to noise and yet preserves high-frequency signal components. Our experiments show that the quality of metric reconstruction from structured light sensors is subject to systematic biases, and highlights the factors that influence it. Our main performance index rates estimates of volume (a proxy of size), for which we review a well-known formula applicable to incomplete meshes. Our code and data will be made publicly available upon completion of the anonymous review process.'
author:
- 'J. Balzer, M. Peters, and S. Soatto'
title: Volumetric Reconstruction Applied to Perceptual Studies of Size and Weight
---
Introduction {#sec:introduction}
============
\
Motivation
----------
We like to believe that our sensory systems provide us with precise and accurate information about objects within the environment, but our perception is often subject to systematic errors, or [*illusions*]{} (Figure \[fig:intro\]). These can also occur between sensory modalities, often with visual information influencing haptic (touch) estimates of properties such as size or weight. For example, a curious experience occurs when we lift two objects of equal weight but different size; systematically and repeatably, the smaller object feels heavier than the larger. This *size-weight illusion* (SWI) [@Charpentier91] cannot be explained by simple motor force error (i.e., it is not simply due to the production of more lifting or grip force for the larger object) [@Flanagan00; @Grandy06], and so carries important implications for the dynamics of sensory integration between vision and haptics. Likewise, altered visual appearance of an object (e.g. through stereoscopic goggles [@Ernst02] or optical distortion with prisms [@Rock64]) can significantly impact haptically-judged estimates of its size. Simply put, when an object looks bigger than it really is, it feels bigger, too – and any mismatch between vision and touch often goes completely unnoticed.
In order to establish a solid quantitative empirical assessment of these illusions, we have developed methodologies to examine the relationship between true size and perceived size. Previous investigations have uncovered evidence that the relationship between an object’s true volume and its perceived volume often follows a power function with an average exponent of $0.704$ ($\sigma=0.08$) [@Frayman81]. However, these prior investigations have predominantly used objects which are geometric, symmetrical, and convex – properties which alone cannot adequately capture the range of objects regularly encountered in everyday environments. Thus, to systematically and comprehensively explore the relationship between true volume and perceived volume so as to better understand this percept’s contribution to visual-haptic integration and, consequently, perception in general, we have developed dedicated methods to capture an ecologically valid set of stimuli.
![We compare the performance of two scanning systems with the one proposed here at hand of a set of geometric primitives, whose volume can be easily measured by hand. Although shown as a collection, all items above have been reconstructed and texture-mapped individually.[]{data-label="fig:cubes"}](cubes_all.jpg "fig:"){height="4cm"} ![We compare the performance of two scanning systems with the one proposed here at hand of a set of geometric primitives, whose volume can be easily measured by hand. Although shown as a collection, all items above have been reconstructed and texture-mapped individually.[]{data-label="fig:cubes"}](cubes_rec "fig:"){height="4cm"}
Our goal is to build a data base of digital models of our specimens, which would allow us to infer volume and any other desired geometric properties. A method to create such models should meet the following list of criteria:
- It is mandatory that the sensing modality of choice be contactless: Cell phones and other hand-held consumer electronics, e.g., are among the classes of objects relevant to our psychological studies. They cannot simply be sunk in a fluid leveraging Archimedes’ principle, neither can anything which is permeated by the fluid, as doing so would not provide the desired *visible* volume.
- The underlying sensor should be inexpensive and easy to use for non-experts; e.g., researchers in psychology and neuroscience. This rules out dedicated lab equipment, e.g., for white-light interferometry and the like.
- The resulting models should exhibit some topological structure: For volume computations, it must at least be possible to distinguish interior and exterior. Therefore, point clouds alone are insufficient in this regard.
- Given the complexity of everyday objects – and that they frequently depart from the cubic, spherical, or cylindrical – we target an improvement of accuracy over back-of-the-envelope estimates.
- A rich spectrum of object classes is covered in terms of admissible geometry and reflectance properties. Specular surfaces, e.g., would need to be coated with powder to make them amenable to laser scanning. But this would contravene the first criterion and thus eliminates laser scanning from the list of candidates.
In light of these requirements, we opt for triangulation based on structured-light encoding as the primary sensing modality. The principle behind this method has been known for decades but has seen a renaissance in computer vision ever since Primesense introduced a fully functional color-range (“RGBD”) sensor unit in integrated-circuit design. This system-on-chip later became the core component of Microsoft’s Kinect, which subsequently had an considerable impact in geometry reconstruction, tracking, occlusion detection, and action recognition, among other applications.
Contribution and overview
-------------------------
We develop a system for structured-light scanning of small- to medium-scale objects, which we dub *Yet Another Scanner*, or YAS. Naturally, the question arises why we would need yet another scanner when several systems and commercial products are already available, e.g., Kinect Fusion [@Newcombe2011], ReconstructMe[^1], Artec Studio[^2], KScan3D[^3], Scanect[^4], Scenect[^5], and Fablitec’s 3d scanner[^6]; in particular, when most of these generate visually highly-pleasing results.
The main reason is that, albeit visually pleasing, the reconstructions provided by these methods are subject to biases that make them unsuitable for scientific investigation. The analysis, which is presented in Sect. \[subsec:results\], compares the performance of YAS with that of two competing state-of-the-art implementations. While the reconstruction algorithm described in Sects. \[subsec:alignment\] and \[subsec:poisson\] itself is not novel, we carefully justify all choices to be made in its design w.r.t. above-listed requirements. Additionally, we address the issue that an aligned series of range images suffers from incompleteness precisely where the ground plane supports the object. Sect. \[subsec:volumeestimation\] proposes a strategy to circumvent this problem in volume estimation which avoids complicated hole-filling algorithms. We believe that a tool which outperforms commercial software but is accessible for further scientific development may be of interest to the computer vision community as well. Hence, as the final contribution, we will distribute the source through the repository at <https://bitbucket.org/jbalzer/yas>.
System description
==================
Data acquisition and calibration
--------------------------------
In all our experiments studies, we used Microsoft’s Kinect and Primesense’s Carmine 1.09. Both devices are shipped with a factory calibration of depth and RGB camera intrinsics as well as the coordinate transformation between their local reference frames. Initial visual assessment (by the naked human eye) approves of the default calibration simply because the point clouds computed from the range image seem to be accurately colored by the values of the RGB image. Extensive tests, however, have shown that – in the spirit of our introductory remarks – such an evaluation is misleading, and significant metric improvements through manual re-calibration are possible. For this purpose, we rely on the toolbox accompanying the paper [@DanielHerreraC2012] to estimate all aforementioned parameters plus a depth uncertainty pattern, which varies both spatially and with depth itself.
View alignment {#subsec:alignment}
--------------
A calibrated sensor immediately delivers physically plausible depth data. The integration of measurements from different vantage points into a common 3-d model can thus be seen as the core challenge here. A comprehensive overview of the state of the art in scan alignment is found in the recent survey [@Tam2013]. Essentially, one can distinguish between two approaches: *tracking* and *wide-baseline matching*. The former is at the heart of Kinect Fusion [@Newcombe2011] and the majority of commercially available software. Its main motivation stems from the fact that correspondence is easier to establish when two images haven been acquired closely in time – supposing, of course, that the motion the camera has undergone between each image acquisition and the next meets certain continuity constraints. We believe, however, that for the purpose of small-scale object reconstruction, the disadvantages of tracking predominate. First and foremost, there is the question of redundancy: How do we deal with the stream of depth data when operating an RGBD camera at frame rates up to $30~\mathrm{fps}$? On the one hand, redundancy is desirable because single depth images may not cover the entire surface of the unknown object, e.g., due to occlusions or radiometric disturbances of the projected infrared pattern. On the other hand, integration of overcomplete range data into a common 3-d model puts high demands on the quality of alignment. Most feature trackers operate on a reduced motion model[^7] and are thus prone to drift. Such deviations in combination with the uncertainty in the raw depth data can lead to a *stratification* of points in regions appearing in more than a single image. This effect is illustrated in Fig. \[fig:misalign1\], which becomes more severe with higher numbers of processed images.
Kinect Fusion [@Newcombe2011] deals with redundancy by instantaneously merging the depth stream into an implicit surface representation over a probabilistic voxel grid. The extra dimension, however, raises memory consumption – even in implementations utilizing truncation or an efficient data structure such as an octree. Also, without a-priori knowledge of the specimen’s size, it is difficult to gauge the interplay between the spatial resolutions of 3-d grid and raw depth data, the latter being left exploited only sub-optimally. Last but not least, a system based on tracking is little user-friendly: It requires the operator to move the sensor as steadily as possible. Otherwise, temporal under-sampling or motion blur can lead to a total breakdown of the alignment process.
\
Here, we closely follow the wide-baseline matching procedure developed by the robotics community, notably in the work of Henry et al. [@Henry2012]. It follows two quasi canonical steps: First, a set of local descriptors around interest points in each RGBD image is computed as well as a set of tentative matches between them. Such descriptors incorporate radiance and depth information either exclusively or in combination. Second, subsets of cardinality three are selected from all matches at random. Each subset admits a hypothesis about the rigid motion that transforms one of the point clouds into the other. The winning hypothesis is taken to be the one supporting the most matches, i.e., generating the highest number of *inliers* [@Fischler1981]. We review these initial two steps in the following sections.
### Sampling {#subsubsec:sampling}
Let us formally consider the case of two views $l=0,1$, i.e., we look for two rigid motions $g_l\in\operatorname{SE}(3)$, $g_l:\bm{x}\mapsto\mathbf{R}_l\bm{x}+\bm{t}_l$, with $\mathbf{R}_l\in\operatorname{SO}(3)$ and $\bm{t}_l\in\mathbb{R}^3$. Without loss of generality, one can assume that $g_0$ coincides with the world reference frame, i.e., $\mathbf{R}_0=\mathbf{I}$ and $\bm{t}_0=\bm{0}$, which leads to a simpler notation of the unknowns $\mathbf{R}_1=\mathbf{R}$ and $\bm{t}_1=\bm{t}$.
A number of interest points $\{\bm{p}^i_0\}, \{\bm{p}^j_1\}$ with high response is extracted from each of the two RGB images corresponding to $g_{0,1}$ by means of the SIFT detector. The points are combined into a set of putative correspondences $\mathcal{C}=\{(\bm{p}^{i_k}_0,\bm{p}^{j_k}_1)\;|\; k=1\ldots,n\in\mathbb{N}\}$ by thresholded forward-backward comparison of the distances between associated SIFT descriptors [@Lowe1999]. The search for nearest neighbors can be sped up by a locality-sensitive hashing technique or similar. However, we found in all of our experiments that the time consumed by a brute-force search was within acceptable limits. Next, we repeatedly draw a sample of three matches from $\mathcal{C}$ and obtain a set of triples $\mathcal{H}=\{(k_1,k_2,k_3)\;|\; 1\leq k_1,k_2,k_3 \leq n,k_1 \neq k_2 \neq k_3 \}$.
### Consensus {#subsubec:consensus}
Implicitly, each of the elements of $\mathcal{H}$ determines a hypothesis about the transformation we are looking for: Suppose we already know $g_1\in\operatorname{SE}(3)$, then the geometric least-squares error for some $(k_1,k_2,k_3)\in\mathcal{H}$ is given by $$\label{eq:lse}
e(k_1,k_2,k_3)=\sum\limits_{k\in\{k_1,k_2,k_3\}}\frac{1}{2}\|\bm{x}^{i_{k}}_0-\mathbf{R}_k\bm{x}^{j_k}_1-\bm{t}_k\|^2.$$ Here, the six points $\bm{x}_0^{i_{k}},\bm{x}_1^{j_{k}}\in\mathbb{R}^3$ equal the backprojections of the three matches $(\bm{p}^{i_k}_0,\bm{p}^{j_k}_1)$ forming the current hypothesis. Given the intrinsic camera parameters, they can be easily computed from the data delivered by the calibrated depth sensor. Conversely, given a triple $(k_1,k_2,k_3)\in\mathcal{H}$, we can find a global minimizer $g_1^*$ of the convex function in the following way: Denote by $\bar{\bm{x}}_0$ the mean of $\bm{x}_0^{i_{k}}$ over $k$ and define $\bar{\bm{x}}_0^{i_{k}}=\bm{x}_0^{i_{k}}-\bar{\bm{x}}_0$. The quantities $\bar{\bm{x}}_1$ and $\bar{\bm{x}}_1^{j_{k}}$ are defined analogously. A minimizer in the *entire general linear group* of matrices $\operatorname{GL}(3)$ is $$\mathbf{H}=\sum\limits_{k\in\{k_1,k_2,k_3\}}\bar{\bm{x}}_0^{i_{k}}(\bar{\bm{x}}_1^{j_{k}})^{\top},$$ cf. [@Williams2001]. One needs to make sure that the optimal $g_1^*$ involves a genuine *rotation matrix* by projecting $\mathbf{H}$ onto $\operatorname{SO}(3)$. This is commonly achieved by *Procrustes analysis*, essentially consisting of a singular-value decomposition: Write $\mathbf{H}$ as the product $\mathbf{U}\mathbf{\Sigma}\mathbf{V}^{\top}$ with two orthogonal factors $\mathbf{U},\mathbf{V}\in\mathbb{R}^{3\times 3}$, then $\mathbf{R}_k^*=\mathbf{V}\mathbf{U}^{\top}$. Once $\mathbf{R}_k^*$ is known, the optimal translation vector can be computed as $\bm{t}_k^*=\bar{\bm{x}}_0-\mathbf{R}_k^*\bar{\bm{x}}_1$. The transformation for the element of $\mathcal{H}$ attaining the highest consensus among *all* matches in $\mathcal{C}$ constitutes the solution to the global alignment problem [@Fischler1981]. The result is refined based on the inlier correspondences with the iterative closest-point (ICP) method [@Besl1992]. This also ensures a *geometrically* continuous alignment, which is not guaranteed because $\mathcal{H}$ was generated merely based on *photometry*.
Surface reconstruction {#subsec:poisson}
----------------------
The common point cloud obtained after merging all aligned depth maps carries no information about the topological relationship between its elements, but as we will see shortly, such information plays a crucial part in volume estimation. There exists a wealth of algorithms for point-to-surface conversion, most of which depend on the signed or unsigned Euclidean distance field $\varphi:\mathbb{R}^3\to\mathbb{R}$ induced by the point cloud. Kinect Fusion, e.g., computes $\varphi$ directly. Alternatively, when the point cloud is oriented, i.e., each point $\bm{x}$ is endowed with an estimate of the normal vector $\bm{n}$ the surface should have at that location, one can search for the function $\varphi$ minimizing $$\label{eq:dirichlet}
\int\limits_D\frac{1}{2}\|\nabla\varphi-\bm{n}\|^2 \mathrm{d}\bm{x}\rightarrow\min.$$ Here, with slight abuse of notation, $\bm{n}$ refers to an (arbitrary) continuation of the normal field from the point set to some sufficiently large rectangular domain $D\subset\mathbb{R}^3$. Since the gradient of any scalar function is orthogonal to its level sets, this gives a family of *integral surfaces* $$\label{eq:isosurface}
\Gamma = \{\bm{x}\in\mathbb{R}^3\,|\,\varphi(\bm{x})=C\}$$ of $\bm{n}$. A minimizer of is found as the solution of the Euler-Lagrange equation $$\label{eq:poisson}
\Delta \varphi = \operatorname{div}\bm{n}$$ under natural boundary conditions (here of Neumann type). Eq. is the well-known Poisson equation and eponymous for the *Poisson reconstruction* algorithm proposed in [@Kazhdan2006].
The motivation for increasing the order of differentiation as compared to the direct approach (i.e., that followed in Kinect Fusion) is twofold: First, Eq. is a variant of the Dirichlet energy, which implies that small holes in the point cloud, i.e., areas where $\bm{n}=\bm{0}$, will automatically be in-painted harmonically. Second, for a solution to exist in the strong sense $\nabla\varphi=\bm{n}$, the normal field must be *integrable* or *curl-free*. Noise, which is very common in RGBD images, is responsible for most of the non-integrability in a measured normal field $\bm{n}$. In the variational setting , however, $\bm{n}$ is implicitly replaced by the next-best gradient (its so-called *Hodge projection*, cf. [@Cantarella2002]), which makes the approach very resilient to stochastic disturbances but at the same time destroys fine details. The smoothing effect can be mitigated by imposing Dirichlet conditions on at a sparse set of salient points. This so-called *screened Poisson reconstruction* has recently been introduced in [@Kazhdan2013].
The point cloud is easily oriented exploiting the known topological structure of the pixel lattice: Given a depth parametrization of the surface $z(x,y)$ over the two orthogonal camera coordinate directions $x$ and $y$, the normal can be written as $\bm{n}=(-\partial_x z, -\partial_y z,1)^{\top}$. The partial derivatives of $z$ w.r.t. camera and image coordinates $(x,y)^{\top}$ respectively $(u,v)^{\top}$ are related by the chain rule: $$\label{eq:normal}
\frac{\partial z}{\partial x} = \frac{\partial z}{\partial u}\frac{\partial u}{\partial x}=\frac{f_u}{z}\frac{\partial z}{\partial u},\quad \frac{\partial z}{\partial y} = \frac{\partial z}{\partial v}\frac{\partial v}{\partial y}=\frac{f_v}{z}\frac{\partial z}{\partial v}.$$ Here, $f_u, f_v$ are the focal lengths of the pinhole depth camera, and finite-differencing provides an approximation to the gradient of $z(u,v)$. For the details of numerically solving and selecting the constant $C$ in appropriately, we refer the reader to the original paper [@Kazhdan2006].
Volume estimation {#subsec:volumeestimation}
-----------------
### Closed surfaces {#subsubsec:volcont}
Suppose for the moment that $\Gamma$ given by is compact and closed. The volume $V$ of the domain $\Omega\subset\mathbb{R}^3$ it encompasses is defined as the integral of the characteristic function $\chi_{\Omega}$ of $\Omega$: $$\label{eq:volume}
V=|\Omega|=\int\limits_{\mathbb{R}^3}\chi_{\Omega}\mathrm{d}\Omega=\int\limits_{\Omega}1 \mathrm{d}\Omega.$$ Unfortunately, an evaluation of this integral is not very practical for two reasons. First, doing so would require a regular grid over $\Omega$, which introduces undesirable artefacts where it interacts with a discrete version of $\Gamma$. Second, an expensive nearest-neighbor problem would need to be solved[^8] to determine whether a point is inside or outside of $\Omega$. The following trick is based on the classic Gauss divergence theorem, cf. [@Mirtich1996], which relates the flow of *any* continuously differentiable vector field $\bm{v}:\mathbb{R}^3\to\mathbb{R}^3$ through the boundary $\Gamma=\partial\Omega$ of $\Omega$ with its source density or *divergence* in the interior: $$\label{eq:gauss}
\int\limits_{\Omega}\operatorname{div}\bm{v}\mathrm{d}\Omega=\int\limits_{\Gamma}\langle\bm{v},\bm{n}\rangle \mathrm{d}\Gamma.$$ The left-hand side of this equation does not quite resemble the right-hand side of , yet. However, this can be achieved by a clever choice of $\bm{v}$, e.g., $\bm{v}:=(x,0,0)^{\top}$, but note that several variants will work equally well and that $\bm{v}$ is *not* unitary. We will return to this point later in Sect. \[subsubsec:holefilling\]. We have $\operatorname{div}\bm{v}=1$ so that combining and yields $$\label{eq:volintegral}
V=\int\limits_{\Gamma}\langle\bm{v},\bm{n}\rangle \mathrm{d}\Gamma.$$
Let us look at the discrete case: Here, a level set $\Gamma_h$ of is extracted by the marching cubes in the form of a triangular mesh [@Lorensen1987]. Such a piece-wise linear representation of the geometry provides a likewise locally-linear approximation of any function $f$ whose values $f_i$ are known at the vertices $\bm{x}_i\in\Gamma_h$, $i\in\mathbb{N}$. A Gauss-Legendre quadrature rule for $f$ with linear precision defined over the triangle $T$ is $$\int\limits_{T}f\mathrm{d}\bm{x}\approx A(T)\sum_{i_k\in\mathcal{I}(T)}^3 \frac{1}{3}f_{i_k},$$ where $A(T)$ equals the area of $T$, and $\mathcal{I}(T)$ enumerates its three corner vertices. Now substitute $f_i$ by the flow $\langle\bm{v}_i,\bm{n}_i\rangle$. The vertex normals $\bm{n}_i$ are usually taken to be the normalized mean of the face normals in a one-ring neighborhood of $\bm{x}_i$. Altogether, we finally obtain the following approximation of Eq. $$V\approx\sum_{T\in \Gamma_h}\frac{A(T)}{3}\sum_{i_k\in\mathcal{I}(T)}^3 \langle\bm{v}_{i_k},\bm{n}_{i_k}\rangle.$$
### Surfaces with boundary {#subsubsec:holefilling}
As explained in Sect. \[subsec:poisson\], Poisson reconstruction accounts for most smaller holes in the aligned point clouds. The support of the object, i.e., its “bottom” or the area where it is in contact with the ground plane, however, remains usually unfilled. At the beginning of Sect. \[subsubsec:volcont\], we demanded that $\Gamma$ be compact and closed because only then the volume of $\Omega$ is well-defined. We can lift this assumption in parts simply by a coordinate transformation: Remember that we chose $\bm{v}$ to point in the direction of the $x$-axis of the world coordinate system. Consequently, the flow through any of the planes $y=\operatorname{const}$ or $z=\operatorname{const}$ vanishes. As shown in Fig. \[fig:vol\], all we have to do is align the support of the model with one of these planes. Without loss of generality, we choose $\{(x,y,z)^{\top}\in\mathbb{R}^3\,|\, z=0 \}$. In our scanning scenario, it is reasonable to assume that the specimens to be measured are spatially isolated enough that the depth images capture a significant portion of the ground plane surrounding the object, which can thus be detected fully automatically. To this end, we again invoke a RANSAC-type procedure, which samples triplets of points, calculates their common plane as a putative solution, and evaluates each such hypothetical plane by how many other points in the cloud it contains.
Experiments {#sec:experiments}
===========
Implementation {#subsec:implementation}
--------------
We created a C++ implementation of most of the reconstruction pipeline, including raw data acquisition, coarse and fine registration as well as detection of the ground plane. Ease of use is of premier priority in view of the interdisciplinary nature of this project. Therefore, the number of dependencies was kept as small as possible: The OpenCV library supplies us with all functionality for feature matching (Sect. \[subsubsec:sampling\]). Our implementation of the ICP method requires fast nearest-neighbor lookup which is based on the kd-tree data structure from the ANN library. We also created a graphical QT frontend which is showcased in the video included in the supplemental material. Poisson reconstruction (Sect. \[subsec:poisson\]) is currently done in Meshlab but will be integrated into our code in upcoming releases.
We compare our method to the Kinect Fusion algorithm and Scenect, which is one of the few commercial software packages without hindering functionality restraints in the trial version. To warrant a fair comparison, all participating systems should be operated with the same sensor and the same intrinsic calibration. This proved to be somewhat difficult: Both Scenect and YAS access devices through the OpenNI framework driver supporting all of Primesense’s products and the Xtion by Asus among others. The Point Cloud Library provides an open-source version of the Kinect Fusion algorithm which could potentially function with the Carmine 1.09 as well, but our experiences with it were little encouraging. For this reason, we had to resort to Microsoft’s own implementation Kinect Fusion Explorer, which works exclusively with proprietary hardware, the Kinect.
Results {#subsec:results}
-------
---------- ----------------------------------------- --------------------------------------- ----------------------------- ----------------------------------------- --------------------------------- --------------------------------------- -------------------------------
No. $V_{\mathrm{manual}}\atop [10^{-3}m^3]$ $V_{\mathrm{YAS}}\atop [10^{-3} m^3]$ $E_{\mathrm{YAS}}\atop[\%]$ $V_{\mathrm{Scenect}}\atop[10^{-3}m^3]$ $E_{\mathrm{Scenect}}\atop[\%]$ $V_{\mathrm{KinFu}}\atop[10^{-3}m^3]$ $E_{\mathrm{KinFu}}\atop[\%]$
1 $1.26$ $1.26$ $0.11$ $1.26$ $0.6$ $1.47$ $16.6$
2 $2.82$ $2.72$ $-3.38$ $2.68$ $-5.1$ $3.61$ $28.4$
3 $1.29$ $1.21$ $-6.07$ $1.21$ $-6.17$ $1.63$ $27.0$
4 $3.07$ $3.07$ $-0.02$ $2.83$ $-7.57$ $4.23$ $38.0$
5 $2.18$ $2.06$ $-5.44$ $2.1$ $-3.79$ $2.85$ $31.1$
6 $6.26$ $6.44$ $2.86$ $5.76$ $-7.07$ $7.69$ $22.9$
7 $1.66$ $1.79$ $8.22$ $1.75$ $5.65$ $1.67$ $0.55$
8 $1.78$ $1.89$ $6.0$ $1.67$ $-6.28$ $2.15$ $20.3$
9 $2.75$ $2.75$ $0.05$ $2.68$ $2.41$ $3.84$ $39.8$
10 $24.6$ $24.8$ $1.05$ $18.6$ $-24.4$ $28.8$ $17.1$
$\mu$ $-0.34$ $-5.74$ $24.2$
$\sigma$ $4.59$ $7.76$ $11.5$
---------- ----------------------------------------- --------------------------------------- ----------------------------- ----------------------------------------- --------------------------------- --------------------------------------- -------------------------------
: Volume estimates and their relative error w.r.t. the value obtained by manual measurement.[]{data-label="tab:volumes"}
The set of 20 specimens can be divided into two equally-sized groups: The first group contains objects of simple geometry like cubes and cylinders, whose basic dimensions can be measured manually with a tape measure or ruler, see Fig. \[fig:cubes\]. Free-forms of sizes ranging from just a few centimeters (fifth row of Fig. \[fig:examplerec\]) to the size of a human upper body (first row of Fig. \[fig:examplerec\]) make up the second group. The “ground truth” volumes for the first group are listed in the first column of Tab. \[tab:volumes\]. Needless to say these are afflicted by their own uncertainty, given that they were determined through measurements with a ruler. Our reconstructions of the cube data set are depicted in Fig. \[fig:cubes\]. We obtain the best average relative volume error $(V_{\mathrm{YAS}}-V_{\mathrm{Manual}})/V_{\mathrm{Manual}}$ of $-0.34$ $\%$. The performance of Scenect is comparable, which is somewhat surprising in view of Fig. \[fig:scenect\]. The meshes created by the Kinect Fusion explorer are of inferior topological quality: they contain a high number of non-manifold simplices. This, however, does not seem to affect the volume estimates negatively. Also it can be said that the sensitivity of volumes w.r.t. the ground plane parameters is relatively low.
As can be seen from the last two columns of Tab. \[tab:volumes\], the Kinect Fusion explorer systematically overestimates the ground truth volume by a significant margin. We conjecture that the issue is rooted in calibration. In fact, an important lesson learned during our experimental studies was that a good calibration can make a difference in error of an order of magnitude. Indeed, Scenect provides a calibration program, but Kinect Fusion Explorer does not. A visualization of Tab. \[tab:volumes\] is plotted in Fig. \[fig:plots\].
Results for the second group of objects are shown in Fig. \[fig:examplerec\]. Scenect performs worst among the three compared methods. It must be said, though, that Scenect does not offer mesh reconstruction feature, and the point clouds it exports are not oriented. Normals can be computed by singular value decomposition considering the nearest neighbors of a point, which is probably less accurate than the finite-differences approximations of . The lack of loop-closure whose effect can be better identified in the point cloud in Fig. \[fig:misalign2\] carries over to the triangular mesh in row of Fig. \[fig:scenect\]. Premature termination of the tracker is responsible for the poor reconstruction of the penguin in the last row of Fig. \[fig:scenect\].
Although it behaved unreliably during volume estimation, Kinect Fusion produced high quality models in real-time, which justifies the tremendous success it had since its inception. Still, the scale bias shows, and a lot of details appear to be missing – details which are present in our YAS reconstructions despite the fact that the Poisson algorithm is known to possess the characteristics of a low-pass filter (see the discussion in Sect. \[subsec:poisson\]). The precision setting in Kinect Fusion is quite rigid since the dimensions of the Cartesian grid have to be fixed *before* the reconstruction process even starts. We found that the maximal resolution of $512\times 512\times 512$ voxels rendered the tracking unstable and/or led to incomplete meshes.
Discussion
----------
We establish correspondence based on photometry, hence our approach fails in the absence of sufficiently exciting texture. This however does not necessarily need to be on the target object itself but can be found in the background, which may be “enriched” since after all, in our application, it is not of interest. The majority of competing approaches, including the two we compare against, can cope with the issue. The main reason is the trade-off between a sparse sampling of viewpoints and aforementioned need for sufficient texture: All previous method implicitly leverage on geometry in the motion estimation stage, which is only made possible by the small baseline between adjacent frames, i.e., through tracking, the disadvantages of which have been discussed in Sect. \[subsec:alignment\]. In fact, ICP measures the similarity between two points by their distance, hence endows each of them with a geometry descriptor, although a primitive one, too primitive to support wide-baseline matching. More suitable descriptors are available, but we are not yet considering them here, for already the process of salient point detection let alone the computation of informative geometry descriptors is extremely challenging on noisy, occlusion-ridden, and incomplete depth data such as from RGBD sensors.
The system has no real-time capabilities. We believe this is not necessary for our target application of small-scale object scanning (unlike e.g. navigation, map-building, or odometry). Considering the low resolution of RGB images, the matching process is a matter of seconds. This bottleneck could be removed by an approximate nearest-neighbors search. Our impression is that our system requires the same or even less overall acquisition time compared to e.g. Kinect Fusion, which requires slow and steady motion around the object (and sometimes even a complete reset).
Conclusion
==========
We described a system for integrating a set of RGBD images of small-scale objects into a geometrically faithful 3-d reconstruction. The system is intended to support researchers in the field of cognitive neuroscience who will use it for acquiring ground truth data for their own experimental studies. To assess the suitability of our 3-d models and those obtained by comparable algorithms, we performed an in-depth analysis of theoretical and empirical kind. There are two main conclusions we would like to draw here: First, the quality of a set of calibration parameters or metric reconstruction can be deceptive. Only quantitative analysis enables veridical information. Second, while the uncertainties of hand-held structured-light scanners may be tremendous in the eyes of the optical metrologist, they help improving studies in the cognitive neurosciences, where manual measurement is still common practice.
[10]{}=-1pt
P. Besl and H. McKay. . , 14(2):239–256, 1992.
J. Cantarella, D. DeTurck, and H. Gluck. . , 109(5):409–442, 2002.
A. Charpentier. . , 3:122–135, 1891.
M. Ernst and M. Banks. . , 415:429–433, 2002.
M. Fischler and R. Bolles. . , 24(6):381–395, 1981.
J. Flanagan and M. Beltzner. . , 3:737–741, 2000.
B. Frayman and W. Dawson. . , 29:56–62, 1981.
M. Goodale. . , 51(13):1567–1587, 2011.
M. Grandy and D. Westwood. . , 95:3887–3892, 2006.
P. Henry, M. Krainin, E. Herbst, X. Ren, and D. Fox. . , 31(5):647–663, 2012.
D. Herrera, J. Kannala, and J. Heikkilä. , 34(10):2058–64, 2012.
M. Kazhdan, M. Bolitho, and H. Hoppe. . , 1:61–70, 2006.
M. Kazhdan and H. Hoppe. Screened poisson surface reconstruction. , 32(1):1–13, 2013.
W. Lorensen and H. Cline. . , 21(4):163–169, 1987.
D. Lowe. . , 1:1150–1157, 1999.
B. Mirtich. . , 1(2):31–50, 1996.
R. Newcombe, D. Molyneaux, D. Kim, P. Koli, A. Davison, J. Shotton, S. Hodges, and A. Fitzgibbon. . , 1:127–136, 2011.
I. Rock and J. Victor. . , 143(3606):594–596, 1964.
G. K. L. Tam, Z.-Q. Cheng, Y.-K. Lai, F. C. Langbein, Y. Liu, D. Marshall, R. R. Martin, X.-F. Sun, and P. L. Rosin. , 19(7):1199–217, 2013.
J. Williams and M. Bennamoun. Simultaneous registration of multiple corresponding point sets. , 81(1):117–142, 2001.
[^1]: <http://reconstructme.net/>
[^2]: <http://www.artec3d.com/software/>
[^3]: <http://www.kscan3d.com/>
[^4]: <http://skanect.manctl.com/>
[^5]: <http://www.faro.com/scenect/>
[^6]: <http://www.fablitec.com/>
[^7]: E.g., the Lucas-Kanade method assumes pure translational motion.
[^8]: But could be remedied by re-visiting the signed distance function $\varphi$ of $\Gamma$ from Sect. \[subsec:poisson\].
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We present a nonrelativistic transition potential for the weak strangeness-changing reaction $\Lambda N \to NN$. The potential is based on a one meson exchange model (OME), where, in addition to the long-ranged pion, the exchange of the pseudoscalar $K, \eta$, as well as the vector $\rho, \omega, K^*$ mesons is considered. Results obtained for different hypernuclear decay observables are compared to the available experimental data.'
address:
- |
Institute for Nuclear Theory, University of Washington, Box 351550\
Seattle, WA 98195-1550
- |
Center of Nuclear Studies, Department of Physics,\
The George Washington University, Washington, D. C. 20052
- |
Departament d’Estructura i Constituents de la Matèria, Facultat de Física,\
Diagonal 647, E-08028 Barcelona
author:
- 'A. Parreño[^1] , C. Bennhold and A. Ramos'
title: Decay of Hypernuclei
---
Introduction
============
As is well known, the mesonic decay ($\Lambda \to \pi N$) of a $\Lambda$ particle in the nuclear medium is highly suppressed due to the Pauli blocking effect acting on the outgoing nucleon. In contrast, the nonmesonic (NM) decay mode ($\Lambda N \to NN$), where the mass difference between the initial hyperon and nucleon is converted into kinetic energy for the outgoing nucleons, is the dominant decay mode for medium to heavy hypernuclei.
During the last thirty years, many theoretical works have described the NM decay of hypernuclei by mainly using two different approaches: a description in terms of quark degrees of freedom or/and the use of a OME model. A recent review of the present status of the theoretical and experimental situation can be found in Ref. [@RO98].
The OME approach is the one that will be presented in this contribution. In order to draw conclusions regarding the weak dynamics, nuclear structure details have to be treated with as few approximations and ambiguities as possible. Working in a shell-model framework, spectroscopic factors are employed to describe the initial hypernuclear and final nuclear structure. To reduce the uncertainties regarding initial and final short-range correlations we use realistic $\Lambda N$ and $NN$ strong interactions based on the Nijmegen baryon-baryon potential. More details about this calculation can be found in Ref. [@PRB97].
The Meson Exchange Potential
============================
In a OME model, the transition $\Lambda N \to N N$ is assumed to proceed via the exchange of virtual mesons belonging to the ground-state pseudoscalar and vector meson octets. The nonrelativistic reduction of the free space Feynman amplitude involving a weak and a strong baryon-baryon-meson (BBM) vertex, gives the nonrelativistic potential in momentum space [@PRB97]. The $\Delta I = 1/2$ rule, which is experimentally known to hold in the free $\Lambda$ decay, has been assumed when computing the NM amplitudes. How to account for possible violations of this rule and their consequences is discussed below. For pseudoscalar mesons, the transtion potential reads: $$V_{ps}({\bf q}) = - G_F m_\pi^2
\frac{g}{2M} \left(
{\hat A} + \frac{{\hat B}}{2\overline{M}}
\mbox{\boldmath $\sigma$}_1 {\bf q} \right)
\frac{\mbox{\boldmath $\sigma$}_2 {\bf q} }{{\bf q}^2+\mu^2} \ ,
\label{eq:pspot}$$ where $G_F m_\pi^2= 2.21\times 10^{-7}$ is the Fermi weak coupling constant, ${\bf q}$ is the momentum carried by the meson, $g = g_{\rm {\scriptscriptstyle BBM} }$ the strong coupling constant for the BBM vertex, $\mu$ the meson mass, $M$ the nucleon mass and $\overline M$ the average between the nucleon and $\Lambda$ masses. The operators ${\hat A}$ and ${\hat B}$ contain the weak parity violating (PV) and parity conserving (PC) coupling constants respectively, as well as the isospin dependence of the potential. For vector mesons the potential has the form: $$\begin{aligned}
{V_{v}}({\bf q}) &=&
G_F m_\pi^2
\left( F_1 {\hat \alpha} - \frac{({\hat \alpha} + {\hat \beta} )
( F_1 + F_2 )} {4M \overline{M}}
(\mbox{\boldmath $\sigma$}_1 \times {\bf q})
(\mbox{\boldmath $\sigma$}_2 \times {\bf q}) \right. \nonumber \\
& & \phantom { G_F m_\pi^2 A }
\left. - i \frac{{\hat \varepsilon} ( F_1 + F_2 )} {2M}
(\mbox{\boldmath $\sigma$}_1 \times
\mbox{\boldmath $\sigma$}_2 ) {\bf q}\right)
\frac{1}{{\bf q}^2 + \mu^2} \
\label{eq:rhopot}\end{aligned}$$ with $F_1 = g^{\rm {\scriptscriptstyle V}}_{\rm{\scriptscriptstyle BBM}}$ and $F_2 = g^{\rm {\scriptscriptstyle T}}_{\rm {\scriptscriptstyle BBM}}$ the vector and tensor strong couplings. The (PC) ${\hat \alpha}$, ${\hat
\beta}$ and (PV) ${\hat \varepsilon}$ operators contain the isospin structure together with the corresponding weak couplings. In the case of isovector mesons ($\pi$,$\rho$) the isospin factor is $\mbox{\boldmath $\tau$}_1
\mbox{\boldmath $\tau$}_2$, and for isoscalar mesons ($\eta$,$\omega$) this factor is just $\hat{1}$ for all spin structure pieces of the potential. In the case of isodoublet mesons ($K,K^*$) there are contributions proportional to $\hat{1}$ and to $\mbox{\boldmath
$\tau$}_1 \mbox{\boldmath $\tau$}_2$ that depend on the coupling constants and, therefore, on the spin structure piece of the potential.
In order to account for the finite size and structure of baryons and mesons, a monopole form factor $F({\bf q}^2)= (\Lambda^2-\mu^2)/(\Lambda^2+{\bf q} ^2)$ is considered in both the strong and weak vertices, where the value of the cut-off, $\Lambda$, different for each meson, is taken from the Jülich $YN$ interaction[@juelich]. To incorporate the effects of the $NN$ interaction, we solve a T-matrix scattering equation in momentum space, which describes the relative motion of two nucleons moving under the influence of the strong interaction. For the strong $NN$ interaction we use the updated version of the Nijmegen $NN$ potential[@stoks]. For the bound $\Lambda N$ state, we replace the uncorrelated shell-model $\Lambda N$ wave function (for which we take harmonic oscillator solutions) by a correlated $\Lambda N$ wave function that contains the effect of the strong $\Lambda N$ interaction. This wave function is obtained by taking, as a guide, microscopic finite-nucleus $G$-matrix calculations[@halder] using the soft-core and hard-core Nijmegen models[@nijme]. It has been shown [@sitges] that multiplying the uncorrelated $\Lambda N$ wave function with the spin-independent correlation function $$f(r)=\left( 1 - {\rm e}^{-r^2 / a^2} \right)^n + b r^2 {\rm
e}^{-r^2 / c^2} \ ,
\label{eq:cor}$$ with $a=0.5 $, $b=0.25 $, $c= 1.28$, $n= 2$, yields decay rates which lie in between those using the hard and soft Nijmegen $\Lambda N$ potential models. Since the deviations were at most 10% the above parametrization can be used as a good approximation to the full correlation function.
The strong BBM couplings are taken from either the Nijmegen[@nijme] or the Jülich [@juelich] potentials. In the weak sector, only the couplings corresponding to decays involving pions are experimentally known. The weak couplings involving mesons heavier than the pion are obtained following Refs. [@holstein; @delatorre]. The PV amplitudes for the nonleptonic decays $B \rightarrow B' + M$ involving pseudoscalar mesons are derived using the soft-meson reduction theorem and SU(3) symmetry, which allows us to relate the physical amplitudes of the nonleptonic hyperon decays into a pion plus a nucleon or a hyperon, $B \rightarrow B' + \pi$, with the unphysical amplitudes of the other members of the meson octet ($K, \eta$). On the other hand, SU(6$)_w$ permits relating the amplitudes involving pseudoscalar mesons with those of the vector mesons. The pole model, where the pole terms due to the $(\frac{1}{2})^+$ ground state (singular in the SU(3) soft meson limit) are assumed to be the dominant contribution, is used for obtaining the PC coupling constants.
Results
=======
In Table \[tab:rate\] we explore the effect of including all the mesons on the weak decay observables for the $^{12}_\Lambda $C hypernucleus. These observables include: the NM decay rate in units of the free $\Lambda$ decay rate, $\Gamma_\Lambda = 3.8 \times 10^9 s^{-1}$ (second column), the fraction between the neutron ($\Lambda n \to nn$) and the proton ($\Lambda p \to np$) induced decays (third column) and the intrinsic $\Lambda$ asymmetry parameter, $a_\Lambda$, characteristic of the elementary $\vec{\Lambda} N \to NN$ reaction taking place in the nuclear medium (last column). This last parameter is related to the observed angular asymmetry in the distributions of protons coming from the decay of polarized hypernuclei. The numbers between parenthesis are obtained when the Jülich constants are used in the strong vertex instead of the Nijmegen ones. Note that, through the pole model, this choice of strong coupling constants would also affect the couplings in the PC weak vertices. However, we keep them fixed to the values obtained using the Nijmegen model and modify only the strong vertex. This allows us to assess the effect of using different sets of strong couplings derived from $YN$ potentials that fit the hyperon-nucleon scattering data equally well.
Even if not shown here, the pion-exchange (OPE) contribution dominates the rate not only in magnitude but in range, a consequence of the pion being the lightest meson. This rate is especially sensitive to the inclusion of the strange mesons, while including the $\rho$-meson has almost no effect. Note that the contribution of each isospin-like pair \[$(\pi,\rho)$, ($K,K^*$), $(\eta,\omega)$\] interferes destructively, so, the reduction caused by the $K$ meson is mostly compensated for by adding the $K^*$. A similar situation is observed between the $\eta$ and $\omega$ mesons, consequently their combined effect on the rate is negligible. The final result for the rate is 15% smaller or greater than the pion-only one, depending on the choice of couplings in the strong sector. This sensitivity is unfortunate since it will certainly complicate the task of extracting weak couplings from this reaction. Improved $YN$ potentials which narrow the range of the strong coupling constants are required to reduce this uncertainty. Table \[tab:rate\] also shows the results obtained when the $NNK$ weak coupling constants derived with one-loop corrections to the leading order in $\chi$PT[@spring] are used (last row). Due to the smaller value of the coupling constants the effect of the $K$ meson is reduced and thus the total rate is increased by about 10%.
It has been known for a long time that the large tensor transition ($^3S_1 \to ^3D_1$) in the OPE mechanism, where only $T=0$ $np$ pairs can occur, is the reason for the small value of [$\Gamma_n/\Gamma_p \,$]{}given by pion exchange alone[@mckellar]. For many years, it was believed that the inclusion of additional mesons would dramatically increase this observable. But here, we find the opposite to be true. This ratio is, as expected, quite sensitive to the isospin structure of the exchanged mesons, and it is again the inclusion of the two strange mesons that dramatically modifies this partial ratio. Including the $K$-exchange which interferes destructively with the pion amplitude in the neutron-induced channel leads to a reduction of the ratio by more than a factor of three. The $K^*$, on the other hand, adds contructively. Using the Nijmegen strong couplings constants leads to a final ratio that is 34% smaller than the pion-only ratio, while using the Jülich couplings leaves this ratio unchanged, due mostly to the larger $K^*$ and $\omega$ couplings. Employing the weak $NNK$ couplings calculated with $\chi$PT increases the $\Gamma_n/\Gamma_p$ ratio by 17% with Nijmegen couplings while the ratio remains unchanged for the Jülich model.
The intrinsic asymmetry parameter, $a_\Lambda$, is also found to be very sensitive to the different mesons included in the model. This is the only observable which is changed dramatically by the inclusion of the $\rho$, reducing the pion-only value by more than a factor of two. Adding the other mesons increases $a_\Lambda$, leading to a result about 30% or 50% larger than for $\pi$-exchange alone, again depending on the type of strong couplings used.
Our final results for various hypernuclei are compared with experimental data in Table \[tab:hyps\]. We find overall agreement between our results for the nonmesonic rate and the experimental values, especially when the $\chi$PT weak couplings for the $K$ meson are used. The situation for the [$\Gamma_n/\Gamma_p \,$]{}ratio is different, and the theoretical values greatly underestimate the newer central experimental ones, although the large experimental error bars do not permit any definite conclusions at this time. On the other hand, the proton-induced rate which has errors of the same magnitude as the total rate is overpredicted by our calculations by up to a factor of two. Even though [$\Gamma_n \,$]{}and [$\Gamma_p \,$]{} appear in disagreement with the data, their sum conspires to give a total rate which reproduces the measurements.
Motivated by the hope that $\Delta I=1/2$ violations would affect the [$\Gamma_n/\Gamma_p \,$]{}ratio, we studied the effect of allowing $\Delta I=3/2$ transitions in the vector meson exchange potential[@PLB98]. The new weak couplings were derived in the factorization approximation and, in order to take into account the limitations imposed by this derivation, we allowed them to vary by up to a factor of $\pm 3$. While the total decay rate changed by at most 6%, the [$\Gamma_n/\Gamma_p \,$]{}ratio could be enhanced by a factor of 2, for different combinations of the multiplying factors. These modifications were almost due to changes in [$\Gamma_n \,$]{}; [$\Gamma_p \,$]{}was barely afected. Even though the estimates based on the factorization approximation are rather crude, the variation on the results for the [$\Gamma_n/\Gamma_p \,$]{}ratio and for the asymmetry parameter (which can be altered by a factor of 7) clearly indicates that one cannot assume the validity of the $\Delta I=1/2$ rule. Experiments measuring partial decay rates of light hypernuclei are desirable in order to clarify the validity of such rule.
Regarding the asymmetry parameter, comparison with experiment can only be made at the level of the measured proton asymmetry, which is determined as a product of $a_\Lambda$, characteristic of the weak decay process, and the polarization of the $\Lambda$ inside the hypernucleus, $p_\Lambda$,characteristic of the strong formation process and which must be determined theoretically. In order to avoid the need for theoretical input and access $A_p$ directly, a new experiment at KEK has measured the decay of polarized $^{5}_\Lambda$He extracting the pion asymmetry from the mesonic channel, ${\cal A}_{\pi^-}$[@ajim98]. The asymmetry parameter $a_{\pi^-}$ of the pionic channel has been estimated to be very similar to that of the free $\Lambda$ decay[@motoba], and, therefore, the hypernuclear polarization is obtained from the relation $P_y={\cal A}_{\pi^-}/a_{\pi^-}$. This value, together with the measurement of the proton asymmetry from the nonmesonic decay has allowed to get the value of $a_\Lambda$ for $^{5}_\Lambda$He, which has been quoted as $0.22
\pm 0.20$[@ajim99], in disagreement with the present theoretical predictions.
Summary and Outlook
===================
Total decay rates evaluated with the full weak OME potential fall within 15% of the value obtained with pion exchange only and reproduce the experimental data. This is partly due to the destructive interference between the contributions of the heavier mesons whose individual influence on the decay rate can be substantial. The importance of kaon exchange makes it possible to see the effects of modifying the weak $NNK$ couplings by one-loop corrections to the leading order in $\chi$PT. Including these loop graphs leads to a reduction of the $NNK$ couplings from their tree-level value up to 50%, which in turn modifies the decay rate by up to 20%. Future experiments should be able to verify this effect. We found the influence of strange mesons to be even more pronounced in the partial rates and their ratio. Furthermore, this ratio turns out to be sensitive to the choice of strong coupling constants as well. This finding indicates the need for improved $YN$ potentials with better determined strong couplings at the hyperon-nucleon-meson vertices. In contrast to the previous observables we found the proton asymmetry to be very sensitive to the $\rho$-exchange while the influence of the kaon is more moderate. This polarization observable is therefore an important addition to the set of observables since its sensitivities are different from the total and partial rates.
Within the one-meson exchange picture it would be desirable to use weak coupling constants obtained from more sophisticated approaches. A beginning has been made by Savage and Springer[@spring] however, an understanding of the weak $\Lambda {\rm N} \pi$ and $\Sigma {\rm N} \pi$ couplings within the framework of chiral lagrangians is still missing. Furthermore, due to the importance of the $K^*$-meson it would be desirable to recalculate its weak $NNK^*$ couplings in improved models as well.
In an attempt to solve the [$\Gamma_n/\Gamma_p \,$]{}puzzle, the authors of Ref.[@ramos4] studied the 3N emission channel ($\Lambda N N \to N N N$), where the virtual pion emitted at the weak vertex is absorbed by a pair of nucleons which are correlated through the strong force, including final state interactions of the three nucleons on their way out of the nucleus via a Monte Carlo simulation. It was shown that the new channel influences the analysis of the ratio [$\Gamma_n/\Gamma_p \,$]{} increasing its experimental error bars and leading to a value compatible with the predictions of the OPE model. However, the same calculation shows that a comparison of the proton spectrum with the experimental one favors values of $\Gamma_n/\Gamma_p=$2–3. It is therefore imperative, before speculating further about the deficiencies of the present models in reproducing this ratio, to carry out more precise experiments such as the measurement of the number of protons emitted per $\Lambda$ decay.
We have seen that the consideration of the exchange of mesons heavier than the pion in the NM decay of hypernuclei does not change the total and partial rates in a dramatic way. A different approach to the problem can be found in Ref.[@OTIS98], where the authors compare the Direct Quark (DQ) potential with a conventional OPE one, and calculate the NM decay rates of light hypernuclei. In order to fix the relative phase between both contributions, the weak $\Lambda N \pi$ coupling is related to a baryon matrix element of the weak hamiltonian for quarks by using soft-pion relations. They found that the DQ contributes the most in $J=0$ transitions, enhancing [$\Gamma_n \,$]{}and therefore the neutron-to-proton ratio, and that the $\Delta I =3/2$ components are significant in the $J=0$ transitions. On the other hand, their nuclear matter results show that a softer cut-off for the pion ($\approx$ 800 MeV) compared to the Born-like one used in the present contribution (1.3 GeV), seems more appropriate to reproduce experimental values of [$\Gamma_p \,$]{}[@Ikek98].
The need for improved $YN$ interaction models has also been pointed out in Ref. [@PRKB98]. Motivated by future measurements [@kishi98] of cross sections and different polarization observables [@kishi99] for the $pn \to p \Lambda$ reaction near threshold, Ref. [@PRKB98] gives a theoretical prediction using the same model as presented here, and taking advantage of the lack of nuclear structure ingredients. This study has shown that this reaction is not only sensitive to the weak ingredients of the model, but also to the strong $YN$ interaction. The cross sections have been found to be of the order of $10^{-12} mb$, a good challenge for the improved experimental devices. The experiment should shed some light onto the understanding of the weak $\Delta S = 1$ hadronic interaction.
The hypernuclear weak decay studies are being extended to double-$\Lambda$ hypernuclei[@PRBfutur]. Very few events involving these exotic objects — whose very existence would place stringent constraints on the existence of the elusive H-dibaryon — have been reported. Studying the weak decay of these objects would open the door to a number of new exotic $\Lambda$-induced decays: $\Lambda \Lambda \to \Lambda N$ and $\Lambda
\Lambda \to \Sigma N$. Both of these decays would involve hyperons in the final state and should be distinguishable from the ordinary $\Lambda N \to
NN$ mode. Especially the $\Lambda \Lambda \to \Lambda N$ channel would be intruiging since the dominant pion exchange is forbidden, thus this reaction would have to occur mostly through kaon exchange. One would therefore gain access to the weak $\Lambda \Lambda K$ vertex.
The promising hypernuclear program at KEK, after an improved measurement of the $^{5}_\Lambda$He decay, the continuing program at BNL, which recently proved the existence of the $^4_\Sigma$He hypernucleus, and the hypernuclear physics program (FINUDA) at DA$\Phi$NE, represent excellent opportunities to obtain new valuable information that will shed light onto the still unresolved problems of the weak decay of hypernuclei.
[9]{}
E. Oset and A. Ramos, Prog. Part. Nucl. Phys. [**41**]{} (Pergamon, 1998) 191. A. Parreño, A. Ramos, and C. Bennhold, Phys. Rev. C [**56**]{}, 339 (1997); A. Parreño, Ph.D. Thesis, Univ. of Barcelona (1997): http://int.phys.washington.edu/ parreno. B. Holzenkamp, K. Holinde, and J. Speth, Nucl. Phys. [**A500**]{}, 485 (1989). V.G. Stoks, R.A.M. Klomp, C.P.F. Terheggen and J.J. de Swart, Phys. Rev. C [**49**]{}, 2950 (1994). D. Halderson, Phys. Rev. C [**48**]{}, 581 (1993). M.N. Nagels, T.A. Rijken, and J.J. de Swart, Phys. Rev. D [**15**]{}, 2547 (1977); P.M.M. Maessen, Th. A. Rijken and J.J. de Swart, Phys. Rev. C [**40**]{}, 2226 (1989). A. Parreño, A. Ramos, C. Bennhold, and D. Halderson, in [*Dynamical Features of Nuclei and Finite Fermi Systems*]{} (World Scientific, Singapore, 1994) 318. J.F. Dubach, G.B. Feldman, B.R. Holstein, L. de la Torre, Ann. Phys. (N.Y.) [**249**]{}, 146 (1996). L. de la Torre, Ph.D. Thesis, Univ. of Massachusetts (1982). M.J. Savage and R.P. Springer, Phys. Rev. C [**53**]{}, 441 (1996). Erratum: [*ibid.*]{} [**54**]{}, 2786 (1996). B.H.J. McKellar and B.F. Gibson, Phys. Rev. C [**30**]{}, 322 (1984). A. Parreño, A. Ramos, C. Bennhold, K. Maltman, Phys. Lett. [**B435**]{} 1-8 (1998). S. Ajimura et al, Phys. Rev. Lett. [**80**]{} 3471 (1998). T. Motoba and K. Itonaga, Nucl. Phys. [**A577**]{}, 293c (1994). S. Ajimura, talk in this Symposium. A. Ramos, M.J. Vicente-Vacas and E. Oset, Phys. Rev. C. [**55**]{}, 735 (1997). T. Inoue, M. Oka, T. Motoba and K. Itonaga, Nucl. Phys. [**A633**]{} (1998) 312; M. Oka, Y. Tani, T. Inoue and K. Sasaki, e-print archive: nucl-th/981103. T. Inoue, talk in this Symposium. A. Parreño, A. Ramos, N.G. Kelkar, C. Bennhold, to appear in Phys. Rev. C, e-print archive: nucl-th/9809053. T. Kishimoto, Nucl. Phys. [**A629**]{}, 369c (1998). N. Nabetani, T. Ogaito, T. Sato, T. Kishimoto, e-print archive: nucl-th/9812072. C. Bennhold, A. Parreño, A. Ramos, M. Zarei (in preparation). J.J. Szymanski et al., Phys. Rev. C [**43**]{}, 849 (1991). H. Noumi et al., Phys. Rev. C [**52**]{}, 2936 (1995). A. Montwill et al., Nucl. Phys. [**A234**]{}, 413 (1974).
0.1 in
$\Gamma^{nm}/\Gamma_\Lambda$ $\Gamma_n/\Gamma_p$ $a_\Lambda$
------------ ------------------------------ --------------------- -------------------
$\pi$ 0.89 (0.89) 0.10 (0.10) $-0.24$ $(-0.24)$
$+ \rho$ 0.86 (0.83) 0.10 (0.10) $-0.10$ $(-0.05)$
$+ K$ 0.50 (0.51) 0.03 (0.03) $-0.14$ $(-0.07)$
$+ K^*$ 0.76 (0.90) 0.05 (0.07) $-0.18$ $(-0.20)$
$+ \eta$ 0.68 (0.90) 0.06 (0.07) $-0.20$ $(-0.20)$
$+ \omega$ 0.75 (1.02) 0.07 (0.11) $-0.32$ $(-0.37)$
0.84 (1.10) 0.08 (0.11) $-0.30$ $(-0.35)$
: Nonmesonic decay observables of ${}^{12}_\Lambda$C. The values in parentheses have been calculated using the Jülich-B coupling constants at the strong vertex.
\[tab:rate\]
0.1 in
$^5_\Lambda$He $^{11}_\Lambda$B $^{12}_\Lambda$C
--------------------------- ----------------------------- ------------------------------------------ ------------------------------------------
$\Gamma/\Gamma_\Lambda$ 0.41 (0.47) 0.61 (0.69) 0.75 (0.84)
EXP: $0.41\pm 0.14$ [@szymanski] $0.95 \pm 0.13 \pm 0.04$ [@noumi] $1.14\pm 0.2$ [@szymanski]
$0.89 \pm 0.15 \pm 0.03$ [@noumi]
$\Gamma_n/\Gamma_p$ 0.07 (0.09) 0.08 (0.10) 0.07 (0.08)
EXP: $0.93\pm 0.55$[@szymanski] $1.04^{+0.59}_{-0.48}$ [@szymanski] $1.33^{1.12}_{-0.81}$ [@szymanski]
$2.16 \pm 0.58^{+0.45}_{-0.95}$ [@noumi] $1.87 \pm 0.59^{+0.32}_{-1.00}$ [@noumi]
$0.70\pm 0.3$[@montwill] $0.70\pm 0.3$[@montwill]
$0.52\pm 0.16$[@montwill] $0.52\pm 0.16$[@montwill]
$\Gamma_p/\Gamma_\Lambda$ 0.39 (0.43) 0.56 (0.62) 0.71 (0.78)
EXP: $0.21 \pm 0.07$[@szymanski] $0.30^{+0.15}_{-0.11}$[@noumi] $0.31^{+0.18}_{-0.11}$[@noumi]
$a_\Lambda$ $-0.27$ ($-0.27$) $-0.39$ ($-0.38$) $-0.32$ ($-0.30$)
EXP: 0.22 $\pm$ 0.20[@ajim99]
${\cal A}(0^{\rm o})$ $-0.12$ ($-0.12$) $-0.03$ ($-0.03$)
EXP: $-0.20\pm 0.10$ [@ajim98] $-0.01\pm 0.10$ [@ajim98]
: Weak decay observables for various hypernuclei. The values in parentheses have been calculated using the $NNK$ weak coupling constants obtained when including one-loop corrections to the leading order in $\chi$PT[@spring].
\[tab:hyps\]
[^1]: Talk presented at the KEK-Tanashi International Symposium on Physics of Hadrons and Nuclei, Tokyo, Japan, December 14-17, 1998. In honor of Prof. K. Yazaki.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We consider the problem of estimating confidence intervals for the mean of a random variable, where the goal is to produce the smallest possible interval for a given number of samples. While minimax optimal algorithms are known for this problem in the general case, improved performance is possible under additional assumptions. In particular, we design an estimation algorithm to take advantage of side information in the form of a control variate, leveraging order statistics. Under certain conditions on the quality of the control variates, we show improved asymptotic efficiency compared to existing estimation algorithms. Empirically, we demonstrate superior performance on several real world surveying and estimation tasks where we use the output of regression models as the control variates.'
bibliography:
- 'main.bib'
---
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The single integral form of Pasquier inversion representation of Khuri-Treiman (KT) equation presents great advantages for describing final state interaction of three-body decay or production processes. However, the original form of Pasquier inversion representation is only given in decay region and regions below. For the regions above, analytic continuation of original form is required. Because of non-trivial nature of analytic continuation procedure, it is the purpose of this work to obtain a well-defined Pasquier inversion representation of KT equation for all the energy range.'
author:
- Peng Guo
title: ' Analytic continuation of Pasquier inversion representation of Khuri-Treiman equation '
---
Introduction {#intro}
============
The theoretical framework for describing low energy hadronic three-body interaction has attracted significant attentions in the past, different approaches have been developed, such as field theory based Faddeev and Bethe-Salpeter type equations [@Faddeev:1960su; @Faddeev:1965su; @Taylor:1966zza; @Basdevant:1966zzb; @Gross:1982ny], and dispersion relation orientated Khuri-Treiman (KT) equation [@Khuri:1960zz; @Bronzan:1963xn]. In processes, such as $\eta \rightarrow 3 \pi$, three-body final state interaction has been reported to play a important role in explaining the discrepancy of Dalitz plot expansion parameters between experimental measurements and theoretical calculations [@Gasser:1984pr; @Kambor:1995yc; @Anisovich:1996tx; @Bijnens:2002qy; @Bijnens:2007pr; @Colangelo:2009db; @Zdrahal:2009cp; @Schneider:2010hs].
Among different methods, dispersion approach based KT equation shows some advantages because of it’s simplicity of formalism and analogue to naive isobar model approximation [@Goradia:1975ec; @Ascoli:1975mn]. Since first proposed in [@Khuri:1960zz], KT equation has been further developed by many authors [@Bronzan:1963xn; @Aitchison:1965kt; @Aitchison:1965zz; @Aitchison:1966kt; @Pasquier:1968zz; @Pasquier:1969dt; @Guo:2014vya]. Original form of KT equation is written in a form of double integrals dispersion equation, one integral comes from the dispersion integration and another is related to partial wave projection. By using Pasquier inversion technique [@Pasquier:1968zz; @Aitchison:1978pw; @Guo:2014vya], the order of two integrals can be exchanged, and it results in a single integral representation of KT equation that is more suitable for numerical computation [@Aitchison:1965kt; @Aitchison:1965zz; @Aitchison:1966kt; @Pasquier:1968zz; @Pasquier:1969dt; @Guo:2014vya]. Unfortunately, original form of Pasquier inversion representation of KT equation is not well-defined for all the energy range, in fact, the original form is only given in the physical decay region and regions below. For other energy regions, analytic continuation of Pasquier inversion representation of KT equation has to be carried out deliberately to avoid singularities generated by contour integrations. As will be discussed in this work, the energy range above two-particle threshold is divided by a complex contour into three parts: decay, unphysical and scattering regions. Unphysical region is disconnected from decay and scattering regions, in this region, original form of KT equation has to be modified and an extra term is needed to keep solution of KT equation staying on physical sheet. Due to non-trivial procedure of analytic continuation, we describe some details of analytic continuation in this work, and present a well-defined form of Pasquier inversion representation of KT equation in all energy regions.
The paper is organized as follows. The original form of Pasquier inversion representation of KT equation is briefly introduced in Section \[formalism\]. The procedure of analytic continuation is described in Section \[analyticpasquier\]. The summary and conclusion are given in Section \[summary\].
Subenergy dispersion approach to three-body final state interaction {#formalism}
===================================================================
A general amplitude for a particle with spin-$J$ decays into three spinless particles, such as in $J/\psi $ decays [@Guo:2010gx; @Guo:2011aa], reads $$\langle 1 2 3 , \mbox{out}| J(\lambda) , \mbox{in} \rangle= i (2\pi)^{4} \delta^{4} (\sum_{i=1,2,3} p_{i}-P) T_{\lambda},$$ where we denote the four momenta by $p_{i},P$ for i-th final state particle and initial decay particle, and $\lambda$ is the spin projection of the initial state along a fixed axis. Suppressing the isospin coupling among initial and final states, the amplitude $T_{\lambda}$ is given by, $$\begin{aligned}
\label{decayamp}
T_{\lambda}(s,t,u)= & \sum_{S,L,\mu} N_{SL \mu} D^{J *}_{\lambda, \mu} (r_{s}) d^{S}_{\mu,0}(\theta_{s}) a^{(s)}_{SL} (s)\nonumber \\
& + (s \rightarrow t) + (s \rightarrow u),\end{aligned}$$ where the invariants are defined by , and , and they are constrained by relation, ($m_{i}$’s are final state particles masses and $M$ is mass of initial particle), and . The spin of pair $(12)$ is denoted by $S$, and the relative orbital angular momentum between $(12)$ and the third particle is given by $L$. $ \theta_{s}$ is polar angle of particle-1 in pair $(12)$ rest frame. The rotation $r_{s}$, which is given by three Euler angles [@Guo:2010gx; @Guo:2011aa], rotates the standard configuration in $(12)3$ coupling scheme to the actual one. In the standard configuration of $(12)3$ coupling (rest frame of three-particle), third particle moves along negative $z$ axis while particle-1 and -2 move in the $xz$ plane. The amplitudes in $(23)1$ and $(31)2$ coupling schemes (denoted by $t$- and $u$-channel respectively) are defined in a similar way as in $(12)3$ coupling (denoted by $s$-channel). The dynamics of decay process are described by scalar functions $a^{(s,t,u)}_{SL} $, which only depend on subenergy of isobar pairs and possess only unitarity cut in subenergy by assumption [@Aitchison:1965zz; @Aitchison:1966kt; @Pasquier:1968zz].
For simplicity, in the following discussion, we consider the decay of a scalar particle, , and truncate the partial waves to include only $S$-wave: . Masses of final particles are assumed identical: , and sub-channels are assumed symmetric: . Thus, the decay amplitude is simply given by sum of three terms, $$\begin{aligned}
\label{decayampswave}
T(s, t, u) = & a (s) + a(t) + a(u) .\end{aligned}$$
Khuri-Treiman equation and Pasquier inversion representation
-------------------------------------------------------------
The discontinuity of decay amplitude crossing unitarity cut in a subenergy, such as $s$, is given by $$\begin{aligned}
\triangle T(s,t,u) =& \frac{T(s+i \epsilon ,t, u ) - T(s-i \epsilon, t, u)}{2i} \nonumber \\
=& \rho(s) f^{* }(s) \frac{1}{2} \int_{-1}^{1} d z_{s} T (s, t, u), \label{fdis1}\end{aligned}$$ where , and $f(s)$ denotes $S$-wave two-body elastic scattering amplitude and is parametrized by phase shift of two-body scattering, . is given by , where . A diagrammatic representation of discontinuity relations in Eq.(\[fdis1\]) is shown in Fig.\[isobarplot\].
![ A diagrammatic representation of discontinuity relations in Eq.(\[fdis1\]). \[isobarplot\]](isobar_model_pasq.pdf){width="48.00000%"}
![ The path of $t_{\pm}(s)$ in $t$ complex plane as $s$ increased from $4 m^{2}$ to $\infty$. The black waggle lines represent right hand cuts of $g(t)$ function. The points labeled by correspond to (a), (b), (c), (d), (e), (f), (g), (h), and (i), respectively. \[fig:2\]](angular_path_pasq_stu.pdf){width="48.00000%"}
![ The path of $s_{\pm}(t)$ in $s$ complex plane for . The arrows indicate the directions that invariants follow along the path of integrations. The black wiggle lines represent cuts attached to two branch points: $(M\pm m)^{2}$ in $s$ plane respectively. The points labeled by correspond to (a), (b), (c), (d), (e), (f), (g), (h), and (i)$s_{+}(-\infty)= \infty$, respectively. \[fig:3\]](path_pasq_stu.pdf){width="48.00000%"}
By assumption, $a$’s possess only unitarity cuts, thus, , and $$\begin{aligned}
\triangle & a(s) = \rho(s) f^{* }(s) \left [ a(s ) + \frac{2}{\rho(s) k(s)} \int_{t_{-}(s)}^{t_{+}(s)} \!\!\!\!\!\!\!\!\ \!\!\!\!\!\! (\Gamma) \ d t \, a( t ) \right ], \label{fdisinvar}\end{aligned}$$ where the factor $2$ in front of contour integral takes into account the contribution for $u$-channel. As discussed in [@Aitchison:1965zz; @Aitchison:1966kt; @Pasquier:1968zz; @Guo:2014vya], the angular projection in Eq.(\[fdis1\]) is replaced by a contour integration in complex plane according to perturbation theory [@Bronzan:1963xn; @Bronzan:1964zz], contour $\Gamma$ is given in Fig.\[fig:2\]. The boundaries of Dalitz plot, $t_{\pm}(s)$, are given by the solutions of , where , the analytic continuation of $t_{\pm}(s)$ in $s$ is specified by $\Gamma$, see Fig.\[fig:2\]. The scalar function $a$ then is determined by subenergy dispersion relation, $$\begin{aligned}
\label{disprelF}
a(s) &= \frac{1}{\pi} \int_{4 m^{2}}^{\infty} d s' \frac{1}{s'-s }\triangle a(s') .\end{aligned}$$
As discussed in [@Guo:2014vya], usually, it is useful to parameterize $a$ as a product of a known function and a reduced amplitude. For instance, we may choose parameterization, , thus, the discontinuity relation for the reduced amplitude $g$ is given by [@Guo:2014vya], $$\begin{aligned}
\label{discangg}
\triangle g (s)
= &- \theta(s_{L}-s) \frac{ \mbox{Im} f (s) }{f^{*}(s)} g (s) \nonumber \\
&+ \theta(s - 4 m^{2}) \frac{2}{ k(s)} \int_{t_{+}(s)}^{t_{-}(s)} \!\!\!\!\!\!\!\!\ \!\!\!\!\!\! (\Gamma) \ d t \, f(t) g(t) ,\end{aligned}$$ where $s_{L}$ labels branch point of left hand cut in $f(s)$, and $$\begin{aligned}
\label{disprelg}
& g(s) = \frac{1}{\pi} \int_{-\infty}^{\infty} d s' \frac{1}{s'-s } \triangle g (s') .\end{aligned}$$ By using Pasquier inversion technique [@Aitchison:1965zz; @Aitchison:1966kt; @Pasquier:1968zz; @Guo:2014vya], also see Appendix \[pasquier\], we may obtain a single integral equation for $g$, $$\begin{aligned}
\label{pasqdisprelg}
g(s) = &- \frac{1}{\pi} \int_{-\infty}^{s_{L}} d s' \frac{1}{s'-s} \frac{ \mbox{Im} f (s') }{f^{*}(s')} g(s') + g_{R}(s) ,\end{aligned}$$ where $$\begin{aligned}
g_{R}(s) = \frac{2}{\pi} \int_{- \infty}^{(M-m)^{2}} \!\! d t \, f(t)g(t) \left [ \theta(t) \Delta(s, t ) -\theta(-t) \Sigma(s, t ) \right ] . \label{physgR}\end{aligned}$$ The kernel functions $\Delta$ and $\Sigma$ are given by $$\begin{aligned}
\Delta ( s,t) &= \int_{s_{-}(t)}^{s_{+}(t)} \!\!\!\!\!\!\!\!\ \!\!\!\!\!\! (C') \ d s' \frac{ 1 }{ U(s') } \frac{1}{s'-s } , \\
\Sigma ( s,t) &= \int_{s_{+} (t)}^{ s_{+} (\infty)} \!\!\!\!\!\!\!\!\!\!\!\!\!\! (C') \ d s' \frac{ 1 }{ U(s') } \frac{1}{s'-s } ,\end{aligned}$$ where square root function is defined in complex-$s$ plane, the phase convention for $U(s)$ is chosen by for , respectively. Thus, the square root $k(s)$ is given by the value of $U(s)$ right below two cuts attached to , . The contour $C'$ is given in Fig.\[fig:3\], and are specified by solutions of and contour $C'$.
The Pasquier inversion representation of $g(s)$ in Eq.(\[pasqdisprelg\]-\[physgR\]) is initially defined in the range (on left and upper side of contour $C'$). As will be made clear in section \[analyticpasquier\], contour $C'$ in kernel functions, $\Delta$ and $\Sigma$, is singular and divides $s$ plane into several isolated regions. Therefore, Eq.(\[pasqdisprelg\]-\[physgR\]) can only hold for a complex $s$ that stays at the same side of contour $C'$ and does not cross contour $C'$. When $s$ is taken to cross contour $C'$ to reach the region on the other side, for Pasquier inversion representation of KT equation to stay on physical sheet, $C'$ has to be deformed and an extra piece is picked up as the consequence of deformation of contour. In follows, we present procedure of analytic continuation of Pasquier inversion representation of KT equation into regions.
![ Analytic continuation of a function of type, $g_{R}(s)$ in Eq.(\[orderexchange\]), is followed by the motion of $s$ (red dashed curve). The physical sheet of $g_{R}(s)$ is defined in the upper half plane that is divided by contour $C'$. Lower half plane can be reached by crossing $C'$, when it does, a discontinuity has to be picked up to keep $g_{R}(s)$ on physical sheet. \[pasqtriangle\]](pasquier_analytic_contiuation.pdf){width="45.00000%"}
Analytic continuation of Pasquier inversion representation for $s\in[(M-m)^{2}, \infty]$ {#analyticpasquier}
========================================================================================
As mentioned previously, $g(s)$ given by Eq.(\[pasqdisprelg\]-\[physgR\]) is originally defined for . The analytic continuation of first term on right hand side of Eq.(\[pasqdisprelg\]) shows no difficulty, therefore we will only focus on the second term on right hand side of Eq.(\[pasqdisprelg\]), $g_{R}(s)$, in following discussion. The $s$ dependence of $g(s)$ on second term, $g_{R}(s)$, is through kernel functions $\Delta $ and $\Sigma$, $\Delta $ and $\Sigma$ on physical sheet for are given by the value of $s$ running along the black wiggle line attached to $(M-m)^{2}$ in Fig.\[fig:3\]. Therefore, the strategy of analytic continuation is that we start from here and then increase $s$ continuously until a singularity is encountered. Unfortunately for $g_{R}(s)$, contour $C'$ presents a cut in complex-$s$ plane which stops us to naively use Eq.(\[pasqdisprelg\]-\[physgR\]) in nearby region . To illustrate this point, we first use the technique presented in Appendix \[pasquier\], and rewrite $g_{R}(s) $ to, $$\begin{aligned}
&g_{R}(s) = \frac{2}{\pi} \int_{\Gamma'} d t \, a(t) \int_{s_{\Gamma'}(t)}^{ \infty} \!\!\!\!\!\!\!\!\!\! (C') \ d s' \frac{1}{s'-s } \frac{ 1 }{ U(s') } , \end{aligned}$$ where contour $\Gamma'$ is given in Fig.\[collapsepath\], the location of $s_{\Gamma'}(t)$ on $C'$ is specified by the location of $t$ on $\Gamma'$, see more details in Appendix \[pasquier\]. Exchanging the order of two integrals leads to, $$\begin{aligned}
& g_{R}(s) =\frac{2}{\pi} \int_{C'} d s' \frac{1}{s'-s } \frac{ 1 }{ U(s') } \int_{0}^{ t_{C'}(s')} \!\!\!\!\!\!\!\!\!\!\!\!\! (\Gamma') \ \ d t \, a(t) , \label{orderexchange}\end{aligned}$$ where $t_{C'}(s')$ is inverse of $s_{\Gamma'}(t)$, and Eq.(\[orderexchange\]) is similar to Eq.(\[pathinteg\]) but with contours $C'$ and $\Gamma'$ instead. Now, we can clearly see the cut structure on $s$ generated by contour $C'$ in Eq.(\[orderexchange\]). As $s$ is moved from left hand side of $C'$ in region to reach region by crossing contour $C'$ (motion of $s$ is demonstrated in Fig.\[pasqtriangle\] by red dashed curve), $C'$ has to be deformed to keep $g_{R}(s)$ on physical sheet. For a example, at a point $s_{A}$ in Fig.\[pasqtriangle\], which sits right next to the inside circle of $C'$ in complex plane, then, $g_{R}(s_{A})$ on physical sheet is given by, $$\begin{aligned}
g_{R}(s_{A}) =& \frac{2}{\pi} \int_{C'} d s' \frac{1}{s'-s_{A} } \frac{ 1 }{ U(s') } \int_{0}^{ t_{C'}(s')} \!\!\!\!\!\!\!\!\!\!\!\!\! (\Gamma') \ \ d t \, a(t) \nonumber \\
+& \frac{ 4 i }{ U(s_{A}) } \int_{0}^{ t_{C'}(s_{A})} \!\!\!\!\!\!\!\!\!\!\!\!\!\!\! (\Gamma') \ \ d t \, a(t). \label{collapsegR}\end{aligned}$$ Next, $s$ is moved away from $s_{A}$ to a point on real axis in region , such as in Fig.\[pasqtriangle\], thus $C'$ is further deformed to follow the motion of $s$. When $s$ reach real axis, $C'$ in second term on the right hand side of Eq.(\[collapsegR\]) collapse onto real axis and $\Gamma'$ opens up accordingly into $\Gamma$, thus, for on the real axis, we obtain, $$\begin{aligned}
g_{R}(s) =& \frac{2}{\pi} \int_{- \infty}^{(M-m)^{2}} \!\! d t \, a(t) \left [ \theta(t) \Delta(s, t ) -\theta(-t) \Sigma(s, t ) \right ] \nonumber \\
+& \frac{ 4 i }{ U(s) } \int_{0}^{ t_{+}(s)} \!\!\!\!\!\!\!\!\!\!\! (\Gamma) \ d t \, a(t), \nonumber \\
& \quad \quad \quad \ \ \mbox{for} \ \ s \in [(M-m)^{2}, (M+m)^{2}].\end{aligned}$$ At last, the analytic continuation of $g_{R}(s)$ from to region , where $s$ runs along black wiggle line attached to $(M+m)^{2}$ in Fig.\[fig:3\], does not encounter any singularities and so it does not require the deformation of contour $C'$, see the motion of red dashed curve in Fig.\[pasqtriangle\], therefore $g_{R}(s)$ in Eq.(\[physgR\]) remains unchanged for .
On the other hand, we may also perform the analytic continuation of Pasquier inversion representation of KT equation through a triangle diagram. Using Eq.(\[disprelF\]), we first rewrite Eq.(\[pasqdisprelg\]-\[physgR\]) to $$\begin{aligned}
\label{cauchypasqdisprelg}
g(s) = &- \frac{1}{\pi} \int_{-\infty}^{s_{L}} d s' \frac{1}{s'-s } \frac{ \mbox{Im} f (s') }{f^{*}(s')} g(s') \nonumber \\
&+ \frac{2}{\pi} \int_{4 m^{2}}^{\infty} d t' \triangle a(t') \mathcal{G} (s,t'), \nonumber \\
& \quad \quad \quad \quad \quad \quad \quad \quad \mbox{for} \ \ s <(M-m)^{2},\end{aligned}$$ where $\mathcal{G} $ is given by, $$\begin{aligned}
\label{trianglediagram}
\mathcal{G}(s,t') = & \frac{1}{\pi} \int_{- \infty}^{(M-m)^{2}} \!\! d t \frac{1}{t'-t} \left [ \theta(t) \Delta(s, t ) -\theta(-t) \Sigma(s, t) \right ] , \nonumber \\
&\quad \quad \quad \quad \mbox{for} \ \ t'>4m^{2}, s <(M-m)^{2}.\end{aligned}$$ $\mathcal{G} $ is identified as the Pasquier inversion representation of a triangle diagram in region . The analytic continuation of $\mathcal{G} $ in different representations is presented in Appendix \[triangle\], the Pasquier inversion representation of $\mathcal{G} $ for is given by Eq.(\[fullpastrianglediagram\]), $$\begin{aligned}
\mathcal{G}(s,t')
&= \frac{1}{\pi} \int_{- \infty}^{(M-m)^{2}} d t \frac{1}{t'-t} \left [ \theta(t) \Delta(s, t) -\theta(-t) \Sigma(s, t) \right ] \nonumber \\
&+ 2 i \theta \left (s - (M-m)^{2} \right ) \theta \left ((M+m)^{2} -s \right ) \nonumber \\
& \quad \times \left [ \frac{1 }{ U(s)} \int_{0}^{t_{+}(s)} \!\!\!\!\!\!\!\!\! \!\! (\Gamma) \ \frac{ d t}{t'-t} + \theta(t') \theta(4 m^{2} -t') \frac{2\pi i}{U(s)} \right ] , \nonumber \\
&\quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad \mbox{for} \ \ (s,t') \in [-\infty, \infty]. \nonumber\end{aligned}$$ The $s$ dependence of $g(s)$ in second term in Eq.(\[cauchypasqdisprelg\]) is all through triangle diagram $\mathcal{G} $, thus, analytic continuation of $\mathcal{G} $ completes the analytic continuation of Pasquier inversion representation of $g(s)$. Plugging Eq.(\[fullpastrianglediagram\]) back into Eq.([\[cauchypasqdisprelg\]]{}), we once again obtain the Pasquier inversion representation of $g(s)$ for , $$\begin{aligned}
\label{analyticpasqdisprelg}
g(s) &= - \frac{1}{\pi} \int_{-\infty}^{s_{L}} d s' \frac{1}{s'-s } \frac{ \mbox{Im} f (s') }{f^{*}(s')} g(s') \nonumber \\
&+ \frac{2}{\pi} \int_{- \infty}^{(M-m)^{2}} \!\! d t f( t ) g(t) \left [ \theta(t) \Delta(s, t ) -\theta(-t) \Sigma(s, t ) \right ] \nonumber \\
&+ 4 i \theta \left (s- (M-m)^{2} \right ) \theta \left ((M+m)^{2} -s \right ) \nonumber \\
& \quad \quad \quad \quad \quad \quad \quad \quad \quad \times \frac{1 }{ U(s)} \int_{0}^{t_{+}(s)} \!\!\!\!\!\!\!\!\! \!\! (\Gamma) \ d t f( t ) g(t) , \nonumber \\
&\quad \quad \quad \quad \quad \quad \quad \quad \quad \mbox{for} \ \ s \in [-\infty, \infty].\end{aligned}$$
![ The real (top plot) and imaginary (bottom plot) parts of $g(s)$ by solving both dispersion representation Eq.(\[discangg\]-\[disprelg\]) (black circles) and Pasquier inversion representation Eq.(\[analyticpasqdisprelg\]) (solid red curves). Blue dashed curves are the solution of $g(s)$ from Eq.(\[pasqdisprelg\]-\[physgR\]) without proper analytic continuation, purple dashed curves represent contribution from extra term picked up by analytic continuation in Eq.(\[analyticpasqdisprelg\]), $4 i /U(s) \int_{0}^{t_{+}(s)} (\Gamma) d t a(t) $. The input model of $f(s)$ is taken from Eq.(28) in [@Guo:2014vya], with fixed parameters: , , , and . The left hand cut of $f(s)$ is placed at and $g(s)$ is normalized to . \[fig:gcont\]](gcont_re.pdf "fig:"){width="54.00000%"} ![ The real (top plot) and imaginary (bottom plot) parts of $g(s)$ by solving both dispersion representation Eq.(\[discangg\]-\[disprelg\]) (black circles) and Pasquier inversion representation Eq.(\[analyticpasqdisprelg\]) (solid red curves). Blue dashed curves are the solution of $g(s)$ from Eq.(\[pasqdisprelg\]-\[physgR\]) without proper analytic continuation, purple dashed curves represent contribution from extra term picked up by analytic continuation in Eq.(\[analyticpasqdisprelg\]), $4 i /U(s) \int_{0}^{t_{+}(s)} (\Gamma) d t a(t) $. The input model of $f(s)$ is taken from Eq.(28) in [@Guo:2014vya], with fixed parameters: , , , and . The left hand cut of $f(s)$ is placed at and $g(s)$ is normalized to . \[fig:gcont\]](gcont_im.pdf "fig:"){width="54.00000%"}
The analytic continuation of Pasquier inversion representation of $g(s)$ given by Eq.(\[analyticpasqdisprelg\]) is tested numerically by using a toy model for $f(s)$, model of $f(s)$ is taken from [@Guo:2014vya]. The comparison of $g(s)$’s by solving Pasquier inversion representation Eq.(\[analyticpasqdisprelg\]) and dispersion representation Eq.(\[discangg\]-\[disprelg\]) is shown in Fig.\[fig:gcont\]. We also shown the results by solving Eq.(\[pasqdisprelg\]-\[physgR\]) without proper analytic continuation compared to the contribution of extra term that is picked up by analytic continuation, . As demonstrated in Fig.\[fig:gcont\], solution of analytic continuation of Pasquier inversion representation of $g(s)$ is consistent with dispersion representation of $g(s)$. Solution of Pasquier inversion representation of $g(s)$ without proper analytic continuation jumps in unphysical region , extra term, , is needed to keep $g(s)$ continuous and staying on physical sheet.
At last, similarly, if we parametrize [@Guo:2014vya], where $D(s)=N(s)/f(s)$ contains only unitarity cut of scattering amplitude and all other cuts are absorbed into function $N(s)$ [@Chew:1960iv; @Frye:1963zz], thus we obtain, $$\begin{aligned}
G(s) =& \frac{2}{\pi} \int_{- \infty}^{(M-m)^{2}} \!\! d t \frac{G(t)}{D(t)} \left [ \theta(t) \Delta_{G}(s, t ) -\theta(-t) \Sigma_{G}(s, t ) \right ] \nonumber \\
+& 4 i \theta \left (s- (M-m)^{2} \right ) \theta \left ((M+m)^{2} -s \right ) \nonumber \\
& \quad \quad \quad \quad \quad \quad \quad \quad \times \frac{N(s) }{ U(s)} \int_{0}^{t_{+}(s)} \!\!\!\!\!\!\!\!\! \!\! (\Gamma) \ d t \frac{G(t)}{D(t)} , \nonumber \\
&\quad \quad \quad \quad \quad \quad \quad \quad \mbox{for} \ \ s \in [-\infty, \infty], \label{analycontdispersionrep}\end{aligned}$$ where the kernel functions $\Delta_{G}$ and $ \Sigma_{G}$ are given by Eq.(\[deltaG\]) and (\[deltaG\]) respectively.
Summary
=======
We presented the analytic continuation procedure of Pasquier inversion representation of KT equation, and a well-defined Pasquier inversion representation of KT equation for an arbitrary $s$ on real axis is given by Eq.(\[analyticpasqdisprelg\]) and Eq.(\[analycontdispersionrep\]).
Comparing the Pasquier inversion representation of KT equation in Eq.(\[analyticpasqdisprelg\]) to dispersion representation of KT equation in Eq.(\[discangg\]-\[disprelg\]), as has been also discussed in [@Guo:2014vya], the single integral form of Pasquier inversion representation in Eq.(\[analyticpasqdisprelg\]) indeed present a significant advantage on numerical computation in regions and . However, in unphysical region , dispersion representation in Eq.(\[discangg\]-\[disprelg\]) requires no extra efforts, but analytic continuation of Pasquier inversion representation becomes non-trivial and need an extra term to keep solution, $g(s)$, staying on physical sheet. At last, we solved Pasquier inversion representation of KT equation in Eq.(\[analyticpasqdisprelg\]) numerically by using a toy model of $f(s)$, solutions with and without proper analytic continuation compared to the solution of dispersion representation are illustrated in Fig.\[fig:gcont\].
ACKNOWLEDGMENTS
===============
We thank Adam P. Szczepaniak for many fruitful discussion. We acknowledge supports in part by the U.S. Department of Energy under Grant No. DE-FG0287ER40365 and the Indiana University Collaborative Research Grant. We also acknowledge support from U.S. Department of Energy contract DE-AC05-06OR23177, under which Jefferson Science Associates, LLC, manages and operates Jefferson Laboratory.
Pasquier inversion technique {#pasquier}
============================
![ The path of $t_{\pm}(s)$ (solid purple curves), contour $C$ (solid black lines) and $\gamma$ (solid blue lines). The arrows indicate the directions that invariants follow in double integrations, Eq.(\[doubleintef\]) and (\[inverseintef\]). The blue wiggle line represents the unitarity cut in $t$ plane, and black wiggle lines represent the cuts in $s$ plane. \[doublepath\]](cauchy_path_new_pasq_stu.pdf){width="48.00000%"}
![ The contour $\Gamma'$ (solid purple lines), $C'$ (solid black curves) and $\gamma'$ (solid blue lines) in Eq.(\[collapseintef\]). The arrows indicate the directions that invariants follow along the path of integrations. The purple and black wiggle lines represent unitarity cut in $s$ plane and cuts attached to two branch points: $(M\pm m)^{2}$ in $s$ plane respectively. \[collapsepath\]](inverse_path_cauchy_new_pasq_stu.pdf){width="48.00000%"}
For completeness, we present the Pasquier inversion technique [@Pasquier:1968zz; @Aitchison:1978pw] in this section. Considering a double integrals equation of type, $$\label{doubleintef}
I(s)= \int_{ 4 m^{2}}^{\infty} d s' \frac{1}{s'-s} \frac{N(s') }{k(s')} \int_{t_{-}(s')}^{t_{+}(s')} \!\!\!\!\!\!\!\!\!\!\! (\Gamma) \ d t \, a(t),$$ where contour $\Gamma$ followed by $t$ integration is defined to avoid unitarity cut in $a$, see Fig.\[fig:2\], and the integration path of $s'$ is defined on the real axis, the physical value of $I(s)$ is given by $s$ running above real axis.
As described in [@Pasquier:1968zz; @Aitchison:1978pw], we first split $ t$ integral into two pieces and rewrite the double integrals in Eq.(\[doubleintef\]) to $$\label{splitintef}
I(s)= \int_{4 m^{2}}^{\infty} d s' \frac{1}{s'-s} \frac{N(s') }{k(s')} \left [ \int_{0}^{t_{+}(s')} \!\!\!\!\!\!\!\!\!\!\!\! (\Gamma) \ - \int_{0}^{t_{-}(s')} \!\!\!\!\!\!\!\!\!\!\!\! (\Gamma) \ \ \right ] d t \, a(t).$$ Then, for first term in bracket in Eq.(\[splitintef\]), the path of $s'$ integration is shifted to above real axis, and for second term in bracket in Eq.(\[splitintef\]), the path of $s'$ integration is shifted to below real axis. Note that kinematic function $k(s')$ as a function of $s'$ has two branch points: $(M \pm m)^{2}$. Two cuts may be attached to these two points, one runs from $-\infty$ up to $(M -m)^{2}$ and another runs from $(M + m)^{2}$ up to $\infty$, see black wiggle lines in Fig.\[doublepath\]. As we have mentioned before, $U$ is defined as the continuation of $k(s')$ for a complex argument, the physical value of $k(s')$ is given by taking the branch of $U(s')$ below two cuts attached to , . These two kinematic cuts are placed above both real axis and shifted $s'$ integration paths described previously, see Fig.\[doublepath\], therefore, operation of shifting $s'$ integration paths is valid and $s'$ integration paths do not interfere with cuts in $U$. Thus, we can safely rewrite the double integrals in Eq.(\[splitintef\]) to, $$\label{pathinteg}
I(s)= \int_{C} d s' \frac{1}{s'-s} \frac{N(s') }{U(s')} \int_{0}^{t_{C}(s')} \!\!\!\!\!\!\!\!\!\!\!\! (\Gamma) \ \ d t \, a(t),$$ where the path $C$ of $s'$ integral is shown in Fig.\[doublepath\], whether $t_{C}(s')$ is depends on which portion of path $C$ the invariant $s'$ is on. $t_{+}(t_{-})$ is assigned to $s'$ on the portion of $C$ above(below) real axis. In Eq.(\[pathinteg\]), the physical value of $I(s)$ is given by $s$ running above contour $C$. Next, we exchange the order of two integrals, so that Eq.(\[pathinteg\]) becomes, $$\label{inverseintef}
I(s)= \int_{\Gamma} d t \, a(t) \left [ \int_{s_{\Gamma} (t)}^{\infty} \!\!\!\!\!\!\!\!\! (C) \ d s' \frac{1}{s'-s} \frac{N(s') }{U(s')} \right ],$$ where $t$ integration on contour $\Gamma$ runs from $0$ to $- \infty$ by looping around threshold $(M-m)^{2}$, see Fig.\[doublepath\], and $s'$ integration runs from $s_{\Gamma}(t)$ up to $\infty$ along path $C$, $s_{\Gamma}(t)$ is given by the inverse of $t_{C}(s')$. By assumption, $a $ has only unitarity cut, using Cauchy’s theorem, we can write a equation, $$a(t) = \frac{1}{2 \pi i} \int_{\gamma} d t' \frac{a(t')}{t'-t},$$ where contour $\gamma$ loops around the unitarity cut but avoiding interference with $\Gamma$, see in Fig.\[doublepath\], the convergence of integration has been assumed valid so that the circle of contour $\gamma$ at infinity can be dropped. Thus, we obtain, $$\begin{aligned}
\label{cauchyintef}
I(s)=&\frac{1}{2 \pi i} \int_{\gamma} d t' \, a(t') \left [ \int_{\Gamma} \frac{d t }{t'-t} \int_{s_{\Gamma} (t)}^{\infty} \!\!\!\!\!\!\!\! (C) \ d s' \frac{1 }{s'-s} \frac{N(s') }{U(s')} \right ].\end{aligned}$$ When $N$ function is replaced by a constant, the function in bracket in Eq.(\[cauchyintef\]) may be associated to a triangle diagram, see Appendix \[triangle\]. The next step is to deform the contour $\gamma$ onto real axis toward but avoid both unitarity cut in $a$ and the singularities from the expression in bracket in Eq.(\[cauchyintef\]). By construction of $\Gamma$, unitarity cut of $a$ sits along the blue wiggle line in Fig.\[doublepath\], and $a(t)$ for $t$ running above unitarity cut is defined physical. Therefore, as long as deformation of $\gamma$ and $\Gamma$ toward negative real axis does not interfere with unitarity cut of $a$, $a$ remains on physical sheet all the time. Singularities of the function in bracket in Eq.(\[cauchyintef\]) have been extensively studied by authors in [@Aitchison:1965zz; @Pasquier:1968zz; @Aitchison:1964ta; @Aitchison:1964nc; @Kacser:1966ta] from perturbation theory perspective, they are branch points at and $(M - m)^{2}$. One may attach branch cuts to those branch points running toward negative real axis [@Aitchison:1965zz; @Pasquier:1968zz; @Aitchison:1964ta; @Aitchison:1964nc; @Kacser:1966ta], therefore, the contour $\gamma'$ may be chosen to loop around the threshold $(M - m)^{2}$ toward negative real axis. The deformation of contour $\gamma$ also drags the contour $\Gamma$ going with it back onto real axis, the correspondent contour $C$ must then be opened up accordingly. Simultaneously, in order for $I(s)$ staying on physical sheet, some $s$ are also dragged by the deformation of $C$ into complex plane, and physical value of $I(s)$ is now given by a $s$ that sits on the same side of $C$ when it opens up into complex plane. The only $t'$-dependent singularities come from factor $1/(t'-t)$ in bracket in Eq.(\[cauchyintef\]), so that, when contour $\gamma$ is collapsed onto real axis, the discontinuity of this factor $1/(t'-t)$ along the cut from $-\infty$ to $(M- m)^{2}$ is picked up. Equivalently, we may replace $ \int_{\gamma} d t' /(t'-t)$ by in Eq.(\[cauchyintef\]). Therefore, Eq.(\[cauchyintef\]) becomes, $$\begin{aligned}
\label{collapseintef}
I(s)=& \int_{ \Gamma'} d t \, a(t) \left [ \int_{s_{\Gamma'} (t)}^{\infty} \!\! \!\!\!\!\!\!\!\! (C') \ d s' \frac{1 }{s'-s} \frac{N(s') }{U(s')} \right ] .\end{aligned}$$ The contours $C'$ and $\Gamma'$ have to be chosen to avoid the singularities in integrands. Examining Eq.(\[collapseintef\]), we note that integrands of contour integration over invariant $s'$ are only the product of $1/(s'-s)$, kinematic function $U$ and left hand cut function $N$. As we have mentioned early, kinematic function $U$ as a function of $s'$ has two branch points: $(M \pm m)^{2}$, two cuts are attached to these two points, one runs toward $-\infty$ and another runs toward $\infty$ respectively, see black wiggle lines in Fig.\[collapsepath\]. Therefore, the contour $C'$ may be chosen to avoid two cuts attached to $(M\pm m)^{2}$, as plotted in Fig.\[collapsepath\]. With this choice, the function in bracket in Eq.(\[collapseintef\]) may be associated to the discontinuities of triangle diagram defined in Eq.(\[cauchyintef\]) along the cut attached to branch point $(M- m)^{2}$ in $t$ plane. Similar to $\gamma'$, the contour $\Gamma'$ also loops around the point $(M- m)^{2}$ and is placed above unitarity cut in $a$. Whether $s_{\Gamma'} (t) $ is $s_{+} (t) $ or $s_{-} (t) $ depends on whether $t$ is above or below the cut attached to $(M-m)^{2}$ in $t$ plane respectively, see Fig.\[collapsepath\].
As mentioned early, the physical value of $I(s) $ was chosen by $s$ running above contour $C$ in Eq.(\[pathinteg\]), when $C$ is deformed to $C'$, see Fig.\[doublepath\] and Fig.\[collapsepath\]. In order for $I(s)$ to stay on physical sheet, $s$ is not allowed to cross contour, as a result, some $s$ are forced to follow the deformation of contour into complex plane. Specifically, (1) for , the physical value of $I(s)$ is given by $s$ running along the black wiggle line attached to , (2) for , now physical $I(s)$ is trapped into the value of $s$ running along balck wiggle line attached to between sections and on $C'$, and (3) the form of $I(s)$ in Eq.(\[collapseintef\]) for on real axis is no longer on the physical sheet, physical $I(s)$ now is given by a complex $s$ running on the upper side of arc on $C'$. To reach physical $I(s)$ for on real axis, the analytic continuation is required, the procedure is described in section \[analyticpasquier\].
At last, by splitting $s'$ integration path, in Eq.(\[collapseintef\]) (subscript $+/-$ of integration limits denotes the path of integration lying above or below the cut attached to branch point $(M-m)^{2}$ in $t$ plane, see Fig.\[collapsepath\]), we obtain, $$\begin{aligned}
\label{pasqinvert}
& I(s)= \int_{ -\infty }^{ (M - m)^{2} } \!\!\!\! d t \, a(t) \left [ \theta(t) \Delta_{G} (s, t) - \theta(-t) \Sigma_{G} (s, t) \right ] ,\end{aligned}$$ where $$\begin{aligned}
\Delta_{G} ( s,t) &= \int_{s_{-}(t)}^{s_{+}(t)} \!\!\!\!\!\!\!\!\ \!\!\!\!\!\! (C') \ d s' \frac{1}{s'-s } \frac{ N(s') }{ U(s') } \label{deltaG}, \\
\Sigma_{G} ( s,t) &= \int_{s_{+} (t)}^{ \infty} \!\!\!\!\!\!\!\!\ \!\!\! (C') \ d s' \frac{1}{s'-s } \frac{ N(s') }{ U(s') } \label{sigmaG} .\end{aligned}$$ For $s$ on real axis, value of $\Delta_{G} $ and $\Sigma_{G} $ on physical sheet is only defined in regions, and . For , $\Delta_{G} $ and $\Sigma_{G} $ given by Eq.(\[deltaG\]-\[sigmaG\]) without proper analytic continuation are on unphysical sheet. For the case , the corresponding kernels are denoted as $\Delta $ and $\Sigma $.
Kernel functions $\Delta $ and $\Sigma $ can be expressed in terms of elementary functions. For real $s$ and $t$, the value of $\Delta $ and $\Sigma $ given below by Eq.(\[deltaphy\]-\[sigmaunphy\]) are simply corresponding to the limit and , and again, Eq.(\[deltaunphy\]) and (\[sigmaunphy\]) are defined on unphysical sheet, $$\begin{aligned}
\Delta (s, t) & = \frac{1}{U(s )} \ln | \frac{R (s, t) + U(s ) U(t )}{R (s, t) - U(s ) U(t ) } | - \theta \left ( \phi (s, t) \right ) \frac{ i \pi}{U(s )} , \nonumber \\
& \mbox{for} \ \ t\in[0, (M-m)^{2}], \nonumber \\
& \quad \ \ s\in [-\infty, (M-m)^{2}] \& [(M+m)^{2}, \infty], \label{deltaphy}\end{aligned}$$ and $$\begin{aligned}
& \Delta (s, t) = \frac{1}{U(s )} \ln \frac{R (s, t) + U(s ) U(t )}{R (s, t) - U(s ) U(t ) } \nonumber \\
& \quad \quad \quad \quad - \theta\left (m(m+M)-t \right ) \theta \left ( s_{R}(t)-s \right ) \frac{ 2 i \pi}{U(s )} , \nonumber \\
& \mbox{for} \ \ t\in[0, (M-m)^{2}], s\in [ (M-m)^{2}, (M+m)^{2}], \label{deltaunphy}\end{aligned}$$ where and $s_{R}(t)$ is given by the solution of . $$\begin{aligned}
&\Sigma (s, t) = \frac{1}{U(s )} \ln | L(s,t) |- \theta \left( s- s_{+}(t) \right) \frac{ i \pi}{U(s )} , \nonumber \\
& \mbox{for} \ \ t <0, s\in [-\infty, (M-m)^{2}] \& [(M+m)^{2}, \infty], \label{sigmaphy}\end{aligned}$$ and $$\begin{aligned}
& \Sigma (s, t) = \frac{1}{U(s )} \ln L(s,t) - \theta \left( \mbox{Im}L (s, t) \right) \frac{ 2 i \pi}{U(s )} , \nonumber \\
& \quad \quad \quad \mbox{for} \ \ t <0, s\in [ (M-m)^{2}, (M+m)^{2}], \label{sigmaunphy}\end{aligned}$$ where $$\begin{aligned}
&L (s, t) \nonumber \\
& = \frac{ \left [ s_{+}(t )-s \right ] \left [ s- M^{2} -m^{2} + U(s) \right ] }{ \left [ s_{+}(t )-s \right ] (s- M^{2} -m^{2} ) + U^{2} (s) - U(s) U ( s_{+} (t) )} . \nonumber\end{aligned}$$
![ A triangle diagram with a fixed internal mass $\sqrt{t}$ in $(23)$ sub-channel. \[trianglediagram\]](triangle_pasq.pdf){width="24.00000%"}
Different representations of a triangle diagram {#triangle}
===============================================
From perturbation theory, the Feynman parametrization of a triangle diagram in Fig.\[trianglediagram\] is given by [@Bronzan:1963xn], $$\begin{aligned}
\label{feynmantriangle}
& \mathcal{G}(s,t) = \frac{1}{\pi} \int_{0}^{1} d \alpha_{1} d \alpha_{2} d \alpha_{3} \nonumber \\
&\times \frac{\delta(1-\alpha_{1} - \alpha_{2} - \alpha_{3})}{\alpha_{1} t + (1- \alpha_{1} -\alpha_{1} \alpha_{2}) m^{2} - \alpha_{3} (\alpha_{1} M^{2} + \alpha_{2} s) -i \epsilon},\end{aligned}$$ where $t$ denotes the invariant mass square of pair $(23)$ propagator. The analytic continuation of Feynman parametrization representation of $\mathcal{G}$ as a function of complex arguments $(s,t)$ is carried out by $i \epsilon$ prescription [@Bronzan:1963xn].
In follows, we present the analytic continuation of both dispersion representation and Pasquier inversion representation of $\mathcal{G}$, the strategy is that we start at a region where a representation of $\mathcal{G}$ is defined on physical sheet and consistent with perturbation theory result Eq.(\[feynmantriangle\]), then, $\mathcal{G}$ is continued to other regions by using perturbation theory result Eq.(\[feynmantriangle\]) as a reference.
\(1) The dispersion representation of a triangle diagram for has been discussed in [@Bronzan:1963xn], $$\begin{aligned}
\mathcal{G}(s,t) =& \frac{1}{\pi} \int_{4 m^{2}}^{\infty} d s' \frac{1}{s'-s } \left [ \frac{1 }{ k(s')} \int_{t_{-}(s')}^{t_{+}(s')} \!\!\!\!\!\!\!\!\! \!\!\! (\Gamma) \ \ d t' \frac{1}{t-t'} \right ] , \label{disptrikernel} \nonumber \\
& \quad \quad \quad \quad \quad \quad \quad \quad \mbox{for} \ \ t> 4 m^{2}.\end{aligned}$$
\(2) The Pasquier inversion representation of a triangle diagram for is given by [@Guo:2014vya], $$\begin{aligned}
\label{pastrianglediagram}
\mathcal{G}(s,t) = & \frac{1}{\pi} \int_{- \infty}^{(M-m)^{2}} \!\! \frac{d t' }{t-t'} \left [ \theta(t') \Delta(s, t' ) -\theta(-t') \Sigma(s, t' ) \right ] , \nonumber \\
& \quad \quad \quad \quad \quad \quad \quad \quad \mbox{for} \ \ s < (M-m)^{2}.\end{aligned}$$
![ Analytic continuation of a function in $t$ of type, $\mathcal{G}(s,t)$ in Eq.(\[tridispexchange\]), is followed by the motion of $t$ (black dashed curve). The physical sheet of $\mathcal{G}(s,t)$ in $t$ is defined in outside space of $\Gamma$. $t\in[0,4m^{2}]$ inside of $\Gamma$ region can be reached by crossing $\Gamma$, then a discontinuity has to be picked up to keep $\mathcal{G}(s,t)$ on physical sheet. \[disptriangle\]](disp_analytic_continuation.pdf){width="45.00000%"}
Analytic continuation of dispersion representation of triangle diagram
----------------------------------------------------------------------
We first perform analytic continuation of dispersion representation of $\mathcal{G}$ in Eq.(\[disptrikernel\]). Note that the overlapping region for both dispersion representation and Pasquier inversion representation of $\mathcal{G}$ on physical sheet is and . As described in Appendix \[pasquier\], exchanging order of double integrals encounters no extra singularities in this region, so, we start from here and rewrite Eq.(\[disptrikernel\]) to, see Eq.(\[doubleintef\]-\[inverseintef\]), $$\begin{aligned}
\mathcal{G}(s,t) =& \frac{1}{\pi} \int_{\Gamma} d t' \frac{1}{t-t'} \left [ \int_{s_{\Gamma} (t')}^{\infty} \!\!\!\!\!\!\!\!\!\! (C) \ d s' \frac{1}{s'-s} \frac{N(s') }{U(s')} \right ] , \nonumber \\
& \quad \quad \quad \quad \mbox{for} \ \ t> 4 m^{2}, s <(M-m)^{2}. \label{tridispexchange}\end{aligned}$$ The cut in $t$ generated by contour $\Gamma$ is now explicitly given by , we start with running along the black wiggle line in Fig.\[fig:2\], where $ \mathcal{G}$ is defined on physical sheet. As long as the motion of $t$ in complex plane does not interfere with the contour $\Gamma$, $ \mathcal{G}$ remains on physical sheet, thus, Eq.(\[disptrikernel\]) still holds for , see the motion of $t$ represented by black dashed curve in Fig.\[disptriangle\]. However, when $t$ is moved to cross contour $\Gamma$, the contour $\Gamma$ has to be deformed to keep $ \mathcal{G}$ on physical sheet. To reach region , we can first move $t$ to $t_{A}$ which is a point sits right inside circle of $\Gamma$, see Fig.\[disptriangle\]. Thus, the deformation of $\Gamma$ leads to $$\begin{aligned}
\mathcal{G}(s,t_{A}) =& \frac{1}{\pi} \int_{\Gamma} d t' \frac{1}{t_{A}-t'} \left [ \int_{s_{\Gamma} (t')}^{\infty} \!\!\!\!\!\!\!\!\!\! (C) \ d s' \frac{1}{s'-s} \frac{N(s') }{U(s')} \right ] \nonumber \\
+& 2 i \int_{s_{\Gamma} (t_{A})}^{\infty} \!\!\!\!\!\!\!\!\!\!\! (C) \ d s' \frac{1}{s'-s} \frac{N(s') }{U(s')} , \label{disprepGcomplex} \nonumber \\
& \quad \quad \quad \quad \quad \quad \mbox{for} \ \ s <(M-m)^{2}.\end{aligned}$$ When $t_{A}$ is moved to on real axis, contour $\Gamma$ in second piece on the right hand side of Eq.(\[disprepGcomplex\]) is dragged by the motion of $t$ to collapse onto real axis, see in Fig.\[collapsepath\], accordingly, $C$ has to be opened up to $C'$. Thus, we obtain, $$\begin{aligned}
\mathcal{G}(s,t_{B}) =&\frac{1}{\pi} \int_{4 m^{2}}^{\infty} d s' \frac{1}{s'-s } \left [ \frac{1 }{ k(s')} \int_{t_{-}(s')}^{t_{+}(s')} \!\!\!\!\!\!\!\!\! \!\!\! (\Gamma) \ \ d t' \frac{1}{t_{B}-t'} \right ] \nonumber \\
+& 2 i \int_{s_{-} (t_{B})}^{\infty} \!\!\!\!\!\!\!\!\!\!\! (C') \ d s' \frac{1}{s'-s} \frac{N(s') }{U(s')} ,\nonumber \\
& \quad \quad \quad \quad \quad \quad \quad \mbox{for} \ \ s <(M-m)^{2}. \label{disprepGs}\end{aligned}$$ So continuation in $t$ is complete. Next, we need to continue $s$ to the region , the continuation of first term on the right hand side of Eq.(\[disprepGs\]) shows no difficulty and encounters no extra singularities. However, as we can see in Fig.\[pasqtriangle\], $s$ on real axis is divided by contour $C'$ into three sections, thus, for , a pole contribution, $-4\pi/U(s)$, is picked up by second term on the right hand side of Eq.(\[disprepGs\]). In the end, analytic continuation of dispersion representation of $\mathcal{G}$ is given by $$\begin{aligned}
\mathcal{G}(s,t) =& \frac{1}{\pi} \int_{4 m^{2}}^{\infty} d s' \frac{1}{s'-s } \left [\frac{1 }{ k(s')} \int_{t_{-}(s')}^{t_{+}(s')} \!\!\!\!\!\!\!\!\! \!\!\!\! (\Gamma) \ \ d t' \frac{1}{t-t'} \right ] \nonumber \\
+& 2 i \theta(t) \theta(4 m^{2} -t) \bigg [ \int_{s_{-} (t)}^{ s_{+} (\infty)} \!\!\!\!\!\!\!\!\ \!\!\! \!\!\!\! \!\! (C') \ \ d s' \frac{1}{s'-s } \frac{ 1 }{ U(s') } \nonumber \\
& \ + \theta \left (s-(M-m)^{2} \right ) \theta \left ((M+m)^{2}-s \right ) \frac{2\pi i}{U(s)} \bigg ] , \nonumber \\
& \quad \quad \quad \quad \quad \quad \quad \mbox{for} \ \ (s,t) \in [-\infty, \infty]. \label{fulldisptriangle} \end{aligned}$$
Analytic continuation of Pasquier inversion representation of triangle diagram
------------------------------------------------------------------------------
For the analytic continuation of Eq.(\[pastrianglediagram\]), similarly, we start from region . We first write Eq.(\[pastrianglediagram\]) to, see Eq.(\[collapseintef\]-\[pasqinvert\]), $$\begin{aligned}
\mathcal{G}(s,t) = & \frac{1}{\pi} \int_{\Gamma'} \frac{d t' }{t-t'} \int_{s_{\Gamma'}(t)}^{ \infty} \!\!\!\!\!\!\!\!\!\! (C') \ d s' \frac{1}{s'-s } \frac{ 1 }{ U(s') } , \nonumber \\
& \quad \quad \quad \quad \mbox{for} \ \ t> 4 m^{2}, s < (M-m)^{2}.\end{aligned}$$ By exchanging the order of two integrals, we obtain, $$\begin{aligned}
& \mathcal{G}(s,t) =\frac{1}{\pi} \int_{C'} d s' \frac{1}{s'-s } \frac{ 1 }{ U(s') } \int_{0}^{ t_{C'}(s')} \!\!\!\!\!\!\!\!\!\!\!\!\! (\Gamma') \ \frac{d t' }{t-t'} , \nonumber \\
& \quad \quad \quad \quad \quad \quad \quad \quad \mbox{for} \ \ t> 4 m^{2}, s < (M-m)^{2}. \label{pasqrepgexchange}\end{aligned}$$ As we can see in Eq.(\[pasqrepgexchange\]) and also described previously in section \[analyticpasquier\], $s$ plane is divided by contour $C'$. Only for the region , $ \mathcal{G}$ need to pick up an extra term $ 2i/ U(s) \int_{0}^{ t_{C}(s)} (\Gamma) d t' /(t-t')$ to stay on physical sheet, thus, the analytic continuation in $s$ leads to, $$\begin{aligned}
& \mathcal{G}(s,t) = \frac{1}{\pi} \int_{- \infty}^{(M-m)^{2}} \!\! \frac{d t' }{t-t'} \left [ \theta(t') \Delta(s, t' ) -\theta(-t') \Sigma(s, t' ) \right ] \nonumber \\
& + \theta \left (s-(M-m)^{2} \right ) \theta \left ((M+m)^{2}-s \right ) \frac{ 2i }{ U(s) } \int_{0}^{ t_{C}(s)} \!\!\!\!\!\!\!\!\!\!\!\!\! (\Gamma) \ \frac{d t' }{t-t'}, \nonumber \\
& \quad \quad \quad \quad \quad \quad \quad \quad \mbox{for} \ \ t> 4 m^{2}, s \in [-\infty, \infty]. \label{pasqrepgt}\end{aligned}$$ Next, we continue $t$ to below $4m^{2}$, again, the first term on the right hand side of Eq.(\[pasqrepgt\]) shows no difficulty of continuation and remains the same. From Fig.\[disptriangle\], we can see, $t$ plane is divided by contour $\Gamma$, thus, only second term on the right hand side of Eq.(\[pasqrepgt\]) for $t\in[0,4m^{2}]$ need to pick up a pole contribution, , to stay on physical sheet. In the end, analytic continuation of Pasquier inversion representation of $\mathcal{G}$ is given by, $$\begin{aligned}
\label{fullpastrianglediagram}
\mathcal{G}(s,t)
& = \frac{1}{\pi} \int_{- \infty}^{(M-m)^{2}} \!\! \frac{d t' }{t-t'} \left [ \theta(t') \Delta(s, t' ) -\theta(-t') \Sigma(s, t' ) \right ] \nonumber \\
&+ 2 i \theta \left (s-(M-m)^{2} \right ) \theta \left ((M+m)^{2}-s \right ) \nonumber \\
& \quad \quad \times \left [ \frac{1 }{ U(s)} \int_{ 0}^{t_{+}(s)} \!\!\!\!\!\!\!\!\! \!\! (\Gamma) \ \frac{d t' }{t-t'} + \theta(t) \theta(4 m^{2} -t) \frac{2\pi i}{U(s)} \right ], \nonumber \\
& \quad \quad \quad \quad \quad \quad \quad \quad \mbox{for} \ \ (s,t) \in [- \infty, \infty].\end{aligned}$$
[99]{}
L. D. Faddeev, Zh. Eksp. Teor. Fiz. [**39**]{}, 1459 (1960) \[Sov. Phys. -JETP [**12**]{}, 1014(1961)\].
L. D. Faddeev, [*Mathematical Aspects of the Three-Body Problem in the Quantum Scattering Theory*]{} (Israel Program for Scientific Translation, Jerusalem, Israel, 1965).
J. G. Taylor, Phys. Rev. [**150**]{}, 1321 (1966).
J. -L. Basdevant and R. E. Kreps, Phys. Rev. [**141**]{}, 1398 (1966).
F. Gross, Phys. Rev. C [**26**]{}, 2226 (1982).
N. N. Khuri and S. B. Treiman, Phys. Rev. [**119**]{}, 1115 (1960).
J. B. Bronzan and C. Kacser, Phys. Rev. [**132**]{}, 2703 (1963).
J. Gasser and H. Leutwyler, Nucl. Phys. B [**250**]{}, 539 (1985).
J. Kambor, C. Wiesendanger and D. Wyler, Nucl. Phys. B [**465**]{},215 (1996).
A. V. Anisovich and H. Leutwyler, Phys. Lett. B [**375**]{},335 (1996).
J. Bijnens and J. Gasser, Phys. Scripta [**T 99**]{} 34 (2002).
J. Bijnens and K. Ghorbani, JHEP [**11**]{}, 030 (2007).
G. Colangelo, S. Lanz and E. Passemar, PoS (CD09) 047 \[arXiv:0910.0765\].
M. Zdrahal, K. Kampf, M. Knecht and J. Novotny, PoS (CD09) 122. \[arXiv:0910.1721\].
S. P. Schneider, B. Kubis and C. Ditsche, JHEP [**02**]{}, 028 (2011).
Y. Goradia and T. A. Lasinski, Phys. Rev. D [**15**]{}, 220 (1977). G. Ascoli and H. W. Wyld, Phys. Rev. D [**12**]{}, 43 (1975).
I. J. R. Aitchison, II Nuovo Cimento [**35**]{}, 434 (1965).
I. J. R. Aitchison, Phys. Rev. [**137**]{}, B1070 (1965); Phys. Rev. [**154**]{}, 1622 (1967).
I. J. R. Aitchison and R. Pasquier, Phys. Rev. [**152**]{}, 1274 (1966).
R. Pasquier and J. Y. Pasquier, Phys. Rev. [**170**]{}, 1294 (1968).
R. Pasquier and J. Y. Pasquier, Phys. Rev. [**177**]{}, 2482 (1969).
P. Guo, I. V. Danilkin and A. P. Szczepaniak, arXiv:1409.8652\[hep-ph\].
I. J. R. Aitchison and J. J. Brehm, Phys. Rev. D [**17**]{}, 3072 (1978).
P. Guo, R. Mitchell, and A. P. Szczepaniak, Phys. Rev. D [**82**]{}, 094002 (2010).
P. Guo, R. Mitchell, M. Shepherd, and A. P. Szczepaniak, Phys. Rev. D [**85**]{}, 056003 (2012).
J. B. Bronzan, Phys. Rev. [**134**]{}, B687 (1964).
G. F. Chew and S. Mandelstam, Phys. Rev. [**119**]{}, 467 (1960).
G. Frye and R. L. Warnock, Phys. Rev. [**130**]{}, 478 (1963).
I. J. R. Aitchison and C. Kacser, Phys. Rev. [**133**]{}, B1239 (1964).
C. Kacser, J. Math. Phys. [**7**]{}, 2008 (1966).
I. J. R. Aitchison and C. Kacser, II Nuovo Cimento A [**40**]{}, 576 (1965).
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
We present a new directly-observable statistic which uses sky position ($x, y$) and proper motion ($v_x, v_y$) of stars near the Galactic center massive black hole to identify populations with high orbital eccentricities. It is most useful for stars with large orbital periods for which dynamical accelerations are difficult to determine.
We apply this statistic to a data set of B-stars with projected radii $0.1 \arcsec\! < p \!< 25 \arcsec$ ($\sim0.004 - 1 {\,\mathrm{pc}}$) from the massive black hole in the Galactic center. We compare the results with those from $N$-body simulations to distinguish between scenarios for their formation. We find that the scenarios favored by the data correlate strongly with particular $K$-magnitude intervals, corresponding to different zero-age main-sequence (MS) masses and lifetimes. Stars with $14 \lesssim m_K \lesssim 15$ ($15 - 20 {M_{\odot}}$, $t_{\rm MS} = 8-13 {{\,\rm Myr}}$) match well to a disk formation origin, while those with $m_K \geq 15$ ($<15 {M_{\odot}}$, $t_{\rm MS} >13 {{\,\rm Myr}}$), if isotropically distributed, form a population that is more eccentric than thermal, which suggests a Hills binary-disruption origin.
author:
- 'Ann-Marie Madigan$^{1,2,3}$, Oliver Pfuhl$^{4}$, Yuri Levin$^{3,5}$, Stefan Gillessen$^{4}$, Reinhard Genzel$^{4,6}$ and Hagai B. Perets$^{7}$'
title: 'On the origin of the B-stars in the Galactic center'
---
Introduction
============
The nuclear star cluster (NSC) within the central few parsecs of our galaxy contains a massive black hole (MBH), SgrA\*, and $\sim 10^7 {M_{\odot}}$ in stellar populations of various ages. The bulk of the stars are old, $\sim 80\%$ forming more than 5 Gyr ago, possibly at the same time as the galactic bulge. After a period of reduced star formation, the star formation rate increased during the last $200 - 300 {{\,\rm Myr}}$ [@Blu03; @Pfu11]. While the late type stars may be too old to retain memory of their initial orbital configuration, and hence formation mechanism, the kinematics of the early type stars should reflect their original distribution. One can resolve these stars individually due to our proximity to the Galactic center (GC)[^1], and use their phase space parameters to constrain formation scenarios. This is a useful way to understand stellar mass accumulation in NSCs in general, which is very likely connected to the formation and growth of MBHs at their centers [@Hop10a; @Hop10b].
There are two fundamental scenarios for the accumulation of stars in NSCs. The first scenario is the merger of multiple star clusters following migration towards the center of a galaxy via dynamical friction with background stars [@Tre75; @Cap93; @Aga11; @Ant12a; @Ant13a]. The second scenario is in-situ formation in nuclear stellar disks as a result of gas migration into the center of galaxies [@Mil04b]. Observational results and theoretical arguments suggest that both mechanisms are necessary to explain the morphology, kinematics and complex star formation history of NSC stellar populations [@Har11; @Lei12; @Ant12a].
Massive young stars in the central parsec of the GC provide evidence of the second mechanism, that is formation in a nuclear stellar disk. Roughly half of the brightest young stars – a population of O- and Wolf-Rayet (WR) stars, $\sim 6 \pm 2 {{\,\rm Myr}}$ old[^2], with masses $\gtrsim 20 {M_{\odot}}$ [@Pau06] – form a thin, clockwise(CW)-rotating disk with projected radii $0.8 \arcsec \-- 12\arcsec$ [@Lev03; @Gen03a; @Lu09; @Bar09], though more young star candidates have been detected at larger distances [@Bar10; @Nis12]. This disk is thought to have formed in-situ from the fragmentation of in-falling or colliding gas clumps at the GC [@Mor93; @San98; @Lev03; @Nay05b; @Lev07; @War08; @War12]. We refer to this structure as the “young CW disk”.
One also observes a population of fainter B-stars, $m_{\rm k} \gtrsim 14$ ($m_K=14$ corresponds to a B0V star), which are not obviously associated with this disk. They appear more isotropically distributed than the brighter stars, though a number may be members of the young CW disk [@Bar10]. They are not truncated in projected radius at the disk inner edge, but continue inwards to the MBH. Those that lie within the central $0.8 \arcsec$ are collectively referred to as the “S-stars”; their kinematics reveal randomly-inclined and near-thermal eccentricity orbits [@Ghe05; @Eis05; @Gil09a].
The orbits of the B-stars further out have not yet been determined and it is unclear whether or not the S-stars and the outermost B-stars form distinct populations. Their ages range from a spectroscopically confirmed $<\! 10 {{\,\rm Myr}}$ for the S-star S2/S0-2 [@Ghe03a; @Eis05; @Mar08], to an upper-limit of $\sim \! 100 - 200 {{\,\rm Myr}}$ on the main sequence lifetime of the lower-mass B-stars. This upper limit does not preclude the B-stars forming contemporaneously with the young CW stars, but they may well derive from an older starburst or even form a continuous distribution in age.
The proximity of the S-stars to the MBH, which prohibits in-situ star formation due to its immense tidal force [@Mor93], combined with their young ages imply a “paradox of youth” [@Ghe03a]. Arguably, the most plausible theory for their origin is the tidal capture by SgrA\* of in-falling B-star binaries by Hills mechanism [@Hil88; @Hil91; @Gou03], following dynamical relaxation by massive perturbers such as giant molecular clouds [@Per07] within the central $10 \-- \!100 {\,\mathrm{pc}}$. In this theory, the S-stars are captured on orbits of very high orbital eccentricity whilst their binary companions may be ejected as hypervelocity stars [see e.g., @Bro12b and references therein]. @Ant13b show that post-capture dynamical evolution via resonant relaxation [@Rau96] can bring the highly-eccentric population of S-stars close to their observed near-thermal eccentricity distribution [@Gil09a] within $50 {{\,\rm Myr}}$ for models of the GC with relaxed NSC, or $\sim 10 {{\,\rm Myr}}$ for models with a dense cluster of $10 {M_{\odot}}$ black holes [see also @Per09b]. It is possible that the B-stars outside the central arcsecond also formed via Hills mechanism, as proposed by @Per10 [hereafter PG10]. However, as the latter authors point out, the initial high-eccentricity distribution must persist, since neither two-body nor resonant relaxation will be able to significantly change the orbital eccentricities of the B-stars at large radii within their lifetimes. This sets up a prediction which observational data from this population can verify or refute.
Outside of the central arcsecond, the accelerations of the B-stars are too small to be reliably detected within $\sim 10 {{\,\rm yr}}$ of observations. Thus we do not get a full orbital solution for each star. In this paper we devise a statistic which uses only the star’s sky position and proper motion velocity and is particularly sensitive in identifying distributions with high orbital eccentricities. We present this high-eccentricity statistic in Section \[S:h-stat\]. In Section \[S:simulations\] we explore another mechanism for dynamically relaxing the orbital eccentricity distribution of B-stars – the formation and gravitational influence of the young CW stellar disk – and investigate two scenarios with $N$-body simulations: the massive perturber plus binary disruption scenario [@Per07], and one based on the proposed model by @Set06 of episodic in-situ star formation, wherein the B-stars formed in a nuclear stellar disk $\sim \! 100 {{\,\rm Myr}}$ ago. We examine the resulting orbital eccentricities of the B-stars after $6 {{\,\rm Myr}}$ of interaction with the young CW disk. In Section \[S:obs\] we introduce our observations and use direct observables and the high-eccentricity statistic to compare them with simulations in Section \[S:comp\]. We discuss our findings in Section \[S:discussion\]. In a follow-up paper we will expand our current analysis on the orbital parameters of the B-stars including radial velocity information.
The high-eccentricity statistic {#S:h-stat}
===============================
The basic idea for identifying stars with high orbital eccentricities is straightforward: a radial orbit in three spacial dimensions also appears as a radial orbit in projection on the sky. This was noted by @Gen03a and revisited by @Pau06 and @Bar09. These authors use the $j$ versus $p$ diagram, where $j$ is the normalized angular momentum along the line-of-sight (positive $z$-axis), $$\begin{split}
j & = \frac{j_z}{j_{\rm z(max)}} \\
&= \frac{x v_y - y v_x }{p v_p},
\end{split}$$ and $$p = (x^2 + y^2)^{1/2}$$ is the projected radius from the MBH. The positive $x$-axis points west and the positive $y$-axis points north. $v_x,v_y$ are the right ascension and declination velocities of a star at $(x,y)$ on the sky such that the projected velocity (i.e., proper motion) is $$v_p= (v_x^2 + v_y^2)^{1/2}.$$ The quantity $j$ is $\sim\! 1, \sim\! 0, \sim\! -1$ if the stellar orbit projected on the sky is mainly clockwise (CW) tangential, radial, or counterclockwise (CCW) tangential. @Gen03a define three $j$ ranges: CW tangential ($j \ge 0.6$), CCW tangential ($j \le -0.6$) and radial ($|j| \le 0.3$). Though a useful tool for classification, the quantity $j$ is not optimally sensitive to high-eccentricity ($j \sim 0$) orbits. This is because stars on radial orbits spend the majority of their orbital period near apoapsis with low $v_p$ with respect to the circular velocity at their projected radii $p$; this increases their value of $j$ and imparts a more tangential orbit in projection. In its place, we propose to use a new high-eccentricity statistic, $h$: $j_z$ normalized to the maximum angular momentum at projected radius $p$ (i.e. replacing $v_p$ with circular velocity at $p$): $$\begin{split}
h & = \frac{j_z}{J_p} \\
&= \frac{x v_y - y v_x }{\sqrt{G {M_{\bullet}}p}},
\end{split}$$ where we use the Kepler circular velocity such that $v_{\rm circ}(p) = (G{M_{\bullet}}/p)^{1/2}$. As with $j$, the quantity $h$ is $\sim\! 1, \sim\! 0, \sim\! -1$ depending on whether the stellar orbit projected on the sky is mainly CW tangential, radial, or CCW tangential, but radial orbits are now confined to low $|h|$-values. We show this in Figure \[fig:inc\_ecc\_h\_j\] where we initialize a cluster of $10^4$ isotropically arranged stars with a thermal distribution of orbital eccentricities to see how well their $j$- and $h$-values constrain their original orbital eccentricity. Each star is distributed randomly in its orbital phase and their positions and velocities are projected onto the plane of the sky to get values for $j$ and $h$. We plot the inclination of the stellar orbits[^3] (ranging from $0 \arcdeg - 90 \arcdeg$ as the distribution in $|j|$ and $|h|$ is symmetric about this range) as a function of orbital eccentricity. The left (right) plot shows stars color-coded according to $|j|$ ($|h|$)-values.
The $h$-statistic differentiates well between tangential and radial orbits. Although low $|h|$-values can correspond to stars that have high-eccentricity and/or inclined orbits (i.e., edge-on with respect to line-of-sight), high eccentricity orbits are not contaminated by high $|h|$-values; $|h|$-values are sharply defined as a function of orbital eccentricity and inclination.
In contrast, high $|j|$-values (yellow dots) are scattered throughout the inclination and eccentricity plane. Many high eccentricity orbits are represented by high $|j|$-values which makes them hard to isolate as a group. We refer the reader to the Appendix for statistical constraints on orbital eccentricity and inclination in different $h$-ranges, the maximum value of $|h|$ for a bound orbit and the effect of the stellar gravitational potential on its estimate.
$N$-body simulations {#S:simulations}
====================
We perform $N$-body simulations of two formation scenarios for the population of large-radii B-stars in the GC — a disk origin, and a binary disruption origin. The stars in the two formation scenarios differ only in orbital angular momentum distribution. We investigate whether their original orbital eccentricities are preserved over $6 {{\,\rm Myr}}$, having been subjected to gravitational torquing from the young CW disk, and calculate the resulting $h$-values to be compared with observations.
We use a special-purpose $N$-body integrator, which is described in detail in @Mad11b. Our integrator is based on a mixed-variable symplectic algorithm [@Wis91; @Kin91; @Sah92] and designed to accurately integrate the equation of motion of a particle in a near-Keplerian potential. We use direct $N$-body particles which move in Kepler elements along ellipses under the influence of the central object, and calculate perturbations to their orbits in Cartesian co-ordinates from surrounding $N$-body particles [@Dan92]. We define semi-major axis, $a$, and eccentricity, $e$, of stellar orbits with respect to a stationary MBH, $$a = - \frac{G {M_{\bullet}}}{2 E},$$ and $$e = \left( 1 - \frac{J^2}{G {M_{\bullet}}a} \right)^{1/2},$$ where ${M_{\bullet}}$ is the mass of the MBH, and $J$ and $E$ are the specific orbital angular momentum and energy of a star. The periapsis of the stellar orbits precess with retrograde motion due to Newtonian mass precession from the additional smooth potential from surrounding cluster of stars.
The time to precess by $2\pi$ radians is $$\label{e:prec}
t^{\rm cl}_{\rm prec} = \pi(2 - \alpha)\dfrac{{M_{\bullet}}}{N(<a)m} P(a) f(e, \alpha),$$ where $N(<a)$ is the number of stars within a given $a$, $m$ is the mass of a single star, $f(e,\alpha)$ is a function which depends on the eccentricity of the orbit and the power-law density index $\alpha$ of the surrounding cluster of stars [@Iva05; @Mad12], and $P(a) = 2\pi(a^3/G{M_{\bullet}})^{1/2}$ is the orbital period of a star with semi-major axis $a$. We include the first post-Newtonian general relativistic effect, that is prograde apisidal precession with a timescale, $$\label{e:prec_GR}
t^{\rm GR}_{\rm prec} = \dfrac{1}{3} (1 - e^2) \dfrac{a c^2}{G{M_{\bullet}}} P(a).$$
Our simulations have four main components chosen to represent the Galactic center $\sim 6 {{\,\rm Myr}}$ in the past:
1. A MBH of mass ${M_{\bullet}}= 4.3 \times 10^6 {M_{\odot}}$ [@Ghe08a; @Gil09b].
2. A smooth stellar cusp with power-law density profile $n(r) \propto r^{-\alpha}$, $\alpha = (0.5, 1.75)$, normalized with a mass of $1.5 \times 10^6 M_{\odot}$ within 1 pc [@Sch07; @Tri08; @Sch09]. We use a smooth gravitational potential for the cusp in our simulations as this greatly decreases the required computation time. Traditional two-body gravitational relaxation has little impact on the stellar orbits as its characteristic timescale is $\mathcal{O}(1 {{\,\rm Gyr}})$ [@Mer10; @Bar12]. We model two-body relaxation and resonant relaxation (which occurs on a shorter timescale), using the ARMA code described in detail in @Mad11b to derive initial conditions for our simulations (see Figure \[fig:arma\_bd\]). @Mad09 and @Gua12 find that in the case of coherently-eccentric disks, significant angular momentum changes due to self-gravity of the disk occurs on timescales $\lesssim 1 {{\,\rm Myr}}$.
3. An eccentric stellar disk, $e = (0.3, 0.6$), representing the young CW disk with surface density profile $\Sigma(a) \propto a^{-2}$. It consists of $N$-body particles with equal masses of $100 {M_{\odot}}$, total mass $M_{\rm CW} = (1,2,4) \times 10^4 {M_{\odot}}$ and semi-major axes $0.03 {\,\mathrm{pc}}\!\le\! a \!\le\! 0.5 {\,\mathrm{pc}}$. In our basic model, the young CW disk is formed instantaneously, i.e., fully formed at $t = 0$, with an opening angle of $1 \arcdeg$. We also run a number of simulations wherein we model its formation using a “switch-on” multiplicative function for the disk mass, such that the mass of a single star is $$\label{e:switch}
m(t) = \dfrac{M_{\rm disk}}{N_{\rm disk}} \tanh \left(\dfrac{t-t_0}{\tau}\right),$$ where $t_0$ is $\mbox{-} 1$ years (so that the young CW disk has mass at t = 0) and the growth timescale $\tau$ is $ 1 \times 10^5$ years [@BoR08]. Secular gravitational interactions with the young CW disk will change the angular momenta of the B-stars. We anticipate the greatest orbital eccentricity change for stars at similar radii to the inner edge of the disk – the torques are much greater at these radii, as $$\label{e:tau}
\tau \sim \left( \dfrac{G M_{\rm disk} e_{\rm disk} e}{a} \right) \delta \phi,$$ where $\tau$ is specific torque on a stellar orbit, $e_{\rm disk}$ is a typical orbital eccentricity of a star in the disk and $\delta \phi$ is the angle between them.
4. A population of B-stars with semi-major axes between $0.03 {\,\mathrm{pc}}\!\le\! a \!\le\! 0.7 {\,\mathrm{pc}}$, consisting of 100 $N$-body particles with equal masses of $100 {M_{\odot}}$. Though the individual masses of the stars are high with respect to real B-star masses, the orbits will respond to the gravitational potential of the young CW disk in the same way, regardless of their mass, due to the equivalence principle. We run convergence tests to confirm this; see appendix. The stars are initialized with a surface density profile $\Sigma(a) \propto a^{-2}$ while their angular momentum distribution depends on the scenario we are simulating; we describe them both here.
[*Binary Disruption (BD) Scenario*]{}: In this scenario, the B-stars form in binaries outside the central parsec. Due to enhanced relaxation from massive perturbers [@Per07], they are propelled onto near-radial orbits where they are disrupted by the MBH via Hills’ mechanism and become bound to the MBH as their partners are ejected into the halo at high velocities. They have a spatially isotropic distribution and are initialized on Kepler orbits with very high eccentricities, $e = 1 - r_t/a \sim 1- (m_{\rm bin}/{M_{\bullet}})^{1/3}$ [see, e.g., @Pfa05], where $r_t = a_{\rm bin} ({M_{\bullet}}/m_{\rm bin})^{1/3}$ is the tidal radius of the MBH, and $m_{\rm bin}$ and $a_{\rm bin}$ are the mass and semi-major axis of the binary.
We use our ARMA code [@Mad11b] to simulate the evolution of orbital eccentricities for B-stars in the [*binary disruption*]{} scenario under the dynamical influence of two-body and resonant relaxation. We confirm the result by PG10 that stellar orbits remain at very high eccentricities outside $\sim 0.1 {\,\mathrm{pc}}$. We find higher mean orbital eccentricity values at all radii however, for both steep ($\alpha = 1.75$) and shallow ($\alpha = 0.5$) cusp profiles; see Figure \[fig:arma\_bd\] in which we plot mean orbital eccentricities as a function of semi-major axis at $60$ and $100 {{\,\rm Myr}}$. As PG10 use full $N$-body simulations, they do not include the entire stellar cusp and hence precession rates for stars are lower than in our simulations, contributing to a higher resonant relaxation rate and hence larger orbital eccentricity changes. We simulate both a [*burst*]{} scenario in which all B-stars begin at $t = 0$, and a [*continuous*]{} scenario in which they are randomly initialized between $t = 0$ and $t = t_{\rm max}$; the latter best reflects binary disruptions due to massive perturbers but the former can be directly compared with the simulations of PG10. We use a fit of the resulting orbital eccentricity distribution after $100 {{\,\rm Myr}}$ in the [*continuous*]{} case as our initial conditions for the [*binary disruption*]{} scenario.
[cccccccccccccc]{} $^{a}$ & $\langle e \rangle^{b}$ & $\alpha^{c}$ & $\langle e \rangle_{\rm CW }^{d}$ & $M_{\rm CW }^{e}$ & $sw^{f}$ & $\theta^{g}$ & $\langle i \rangle^{h}$ & & &&&&&&& & $\langle |h| \rangle^{j}$ & $\sigma_{\langle |h| \rangle}^{k}$ & $s_e^{l}$& $\langle |h| \rangle^{j}$ & $\sigma_{\langle |h| \rangle}^{k}$ & $s_e^{l}$\
BD$_1$ & $0.97$ & $0.5$ & $0.6 $ c & 1 & $0$ & - & - & 0.095 & 0.063 & 0.006 & 0.096 & 0.060 & 0.006 BD$_2$ & $0.97$ & $0.5$ & $0.6 $ c & 1 & $1$ & - & - & 0.098 & 0.063 & 0.007 & 0.097 & 0.060 & 0.006 BD$_3$ & $0.97$ & $0.5$ & $0.6 $ & 1 & $0$ & - & - & 0.097 & 0.061 & 0.010 & 0.095 & 0.056 & 0.011 BD$_4$ & $0.97$ & $0.5$ & $0.3$ c & 1 & $0$ & - & - & 0.099 & 0.058 & 0.012 & 0.096 & 0.055 & 0.010 BD$_5$ & $0.93$ & $1.75$ & $0.6$ c & 1 & $0$ & - & - & 0.130 & 0.068 & 0.010 & 0.125 & 0.065 & 0.009 BD$_6$ & $0.93$ & $1.75$ & $0.3$ c & 1 & $0$ & - & - & 0.127 & 0.072 & 0.014 & 0.123 & 0.068 & 0.011 BD$_7$ & $0.97$ & $0.5$ & $0.6 $ c & 2 & $0$ & - & - & 0.114 & 0.077 & 0.007 & 0.102 & 0.059 & 0.005 BD$_8$ & $0.97$ & $0.5$ & $0.6 $ c & 4 &$0$ & - & - & 0.141 & 0.102 & 0.009 & 0.129 & 0.095 & 0.011 DD$_{1}$ & $0.71$ & $0.5$ & $0.6$ c & 1 & $0$ & $40$ & 80 & 0.325 & 0.236 & 0.048 & 0.313 & 0.247 & 0.051 DD$_{2}$ & $0.66$ & $0.5$ & $0.6$ c & 1 &$0$ & $40$ & 25 & 0.352 & 0.244 & 0.033 & 0.349 & 0.248 & 0.034 DD$_{3}$ & $0.68$ & $0.5$ & $0.6$ c & 1 &$0$ & $40$ & 121 & 0.283 & 0.227 & 0.038 & 0.250 & 0.188 & 0.029 DD$_{4}$ & $0.66$ & $0.5$ & $0.6$ c & 1 &$0$ & $40$ & 163 & 0.348 & 0.201 & 0.043 & 0.350 & 0.203 & 0.045 DD$_{5}$ & $0.67$ & $0.5$ & $0.6$ & 1 &$0$ & $40$ & 80 & 0.301 & 0.210 & 0.034 & 0.262 & 0.191 & 0.035 DD$_{6}$ & $0.30$ & $0.5$ & $0.6$ c & 1 &$0$ & $40$ & 80 & 0.573 & 0.279 & 0.049 & 0.563 & 0.270 & 0.067 DD$_{7}$ & $0.69$ & $1.75$ & $0.6$ c & 1 & $0$ & $40$ & 80 & 0.277 & 0.212 & 0.035 & 0.235 & 0.178 & 0.035 DD$_{8}$ & $0.67$ & $0.5$ & $0.6$ c & 1 & $0$ & $60$ & 80 & 0.311 & 0.235 & 0.021 & 0.290 & 0.230 & 0.031 DD$_{9}$ & $0.67$ & $0.5$ & $0.6$ c & 1 & $1$ & $40$ & 80 & 0.315 & 0.263 & 0.028 & 0.251 & 0.215 & 0.038 DD$_{10}$ & $0.62$ & $0.5$ & $0.3$ c & 1 &$0$ & $40$ & 80 & 0.382 & 0.246 & 0.028 & 0.364 & 0.235 & 0.033
[*Dissolved Disk (DD) Scenario*]{}: In this scenario, the B-stars form in-situ around the MBH in a nuclear stellar disk during an earlier episode ($\sim 100 {{\,\rm Myr}}$) of gas infall and fragmentation: we call this the [*dissolved disk*]{} scenario as the disk structure should have ‘puffed up’ due to gravitational interactions between stars, in particular due to vector resonant relaxation [@Koc11] and the gravitational influence of the circum-nuclear disk [@Sub09]. We distribute the B-stars in a stellar disk with their initial eccentricities, disk opening angle and inclination with respect to the young CW disk varying with simulation number. Inclinations are selected such that co-rotating and counter-rotating cases with both large and small angles between the two disks are explored. The basic model, DD$_{1}$, draws eccentricities from a thermal distribution.
Our simulations explore a wide range of parameters relevant for the GC as listed in Table \[tab:sim\_params\]. An important variation in the models is the power-law stellar density index of which we choose two values – a steep stellar cusp $\alpha\!=\!1.75$, and a shallow stellar cusp $\alpha\!=\!0.5$. This variation reveals itself in the precession rate of stars at different radii, and hence the persistence of stellar torques. Another variation is the mean eccentricity of the B-star and the young CW disk orbits, which affects their eccentricity evolution through the strength of the torques between the two groups. We run simulations with both coherently-eccentric (or lopsided) young CW disks, in which the eccentricity vectors of the stellar orbits initially overlap, and non-coherent ones.
Evolution of orbital eccentricities
-----------------------------------
We follow the change in orbital angular momentum in the B-stars over $6 {{\,\rm Myr}}$ in response to the young CW disk. The B-stars with small semi-major axes not only experience a greater torque, $\tau \propto 1/a$, but have lower angular momentum, $J$, and hence the relative change in angular momentum is high. This brings about a rapid change in their orbital eccentricities. In contrast, B-stars with large semi-major axes retain memory of their initial orbital eccentricities; these stars can best constrain their formation scenario. This is as PG10 found for resonant relaxation, but the change in eccentricities of B-stars over $6 {{\,\rm Myr}}$ under the persistent torques of the young CW disk proves to be more rapid than resonant relaxation over $100 {{\,\rm Myr}}$. We show this effect in Figure \[fig:eccVa\], plotting the eccentricity, $e$, of B-star orbits in two simulations (DD$_6$, BD$_1$) as a function of semi-major axis, $a$. The [*binary disruption*]{} scenario produces the most dramatic signature as stars with large semi-major axes retain their high orbital eccentricities. In the [*dissolved disk*]{} scenario, the B-stars with large semi-major axes have similar eccentricities as their initial input values.
Simulated $h$-values and difference in initial conditions
---------------------------------------------------------
We contrast the $j$ and $h$ versus $p$ diagrams for both scenarios using an arbitrary viewing angle in Figure \[fig:jh\]. The advantage of the $h$-statistic is not obvious in the [*dissolved disk*]{} scenario (left) as there are few very high eccentricity orbits. However, in the [*binary disruption*]{} scenario (right), the B-stars at large radii retain their high eccentricities over the $6 {{\,\rm Myr}}$ simulation, and their $h$ values are centered about zero.
We plot the mean value of $|h|$ as a function of projected radius $p$ after $6 {{\,\rm Myr}}$ in Figure \[fig:h\_mean\]. The mean value of $|h|$ is derived from combining $h$-values from three different viewing directions for each simulation. The most distinguishing feature between the two scenarios is the disparate $|h|$-signature at large projected radii. The simulations that model the young CW disk formation by employing the switch-on function of Eq. \[e:switch\] show no significant differences with respect to the basic models.
We find an inverse relation in the [*binary disruption*]{} simulations between the value of the power-law index of the stellar cusp, $\alpha$, and the slope of the distribution of $\langle |h| \rangle$-values with distance from the MBH. Steep stellar cusps result in flatter $\langle |h| \rangle$-$r$ distributions and vice versa. This is due to the coherence time over which stellar torques can act on the individual B-star orbits. Simulations BD$_{1-4}$ which have a shallow stellar cusp ($\alpha = 0.5$) result in a steep $\langle |h| \rangle$-distribution across projected radii. The B-star orbits at small radii precess relatively slowly and hence secular changes in $J$ are efficient at changing their eccentricities. This results in large $\langle |h| \rangle$-values at small radii. B-star orbits at large radii precess relatively quickly and secular changes in $J$ are less efficient as the coherence time is short. These B-stars retain their high orbital eccentricities, and have low $\langle |h| \rangle$-values. Simulations BD$_{5-6}$ which have a steep stellar cusp ($\alpha = 1.75$) result in a flatter $\langle |h| \rangle$-$r$ distribution. The stellar orbits at small radii precess relatively quickly and hence secular changes in $J$ are inefficient relative to that experienced by stars in simulations BD$_{1-4}$. This results in lower $\langle |h| \rangle$-values at small radii. The stellar orbits at large radii precess relatively slowly and secular changes in $J$ are more efficient as the coherence time is longer. $\langle |h| \rangle$-values at large radii are consequently larger than for BD$_{1-4}$. One can in principle constrain the mass distribution of the underlying stellar cusp using observations of stars which have been disrupted from a binary by the MBH from the relation between the stellar cusp profile and $\langle |h| \rangle$-values as a function of distance from the MBH.
Simulations BD$_{7-8}$ (larger mass young CW disk) show the most evolution in B-star eccentricities. $\langle |h| \rangle$-values are high relative to the simulations with $M_{\rm CW } = 10^4{M_{\odot}}$ but still lower than in the [*dissolved disk*]{} simulations.
The DD$_6$ simulation, in which the B-stars are initialized in an $e = 0.3$ disk, and DD$_{10}$ with an $e = 0.3$ young CW disk, result in the lowest eccentricities of the [*dissolved disk*]{} simulations and hence the highest $\langle |h| \rangle$-values. DD$_7$ which has a steep stellar cusp ($\alpha = 1.75$) shows the most evolution in B-star eccentricity at large radii, reaching the lowest $\langle |h| \rangle$-values of the [*dissolved disk*]{} simulations. In the DD$_{2}$ simulation, where the initial mean inclination angle between the two disks is $\langle i \rangle = 25 \arcdeg$, the B-star angular momentum vectors overlap with those of the young CW disk by the end of the simulation. The same overlap is observed in the DD$_{1}$ simulation ($\langle i \rangle = 80 \arcdeg$) but to a lesser extent. Simulations DD$_{3,4}$, in which the B-star and young CW disks are counter-rotating with respect to one another, $\langle i \rangle > 90 \arcdeg$, show a substantial number of stars from each disk with reversed signs of angular momentum but little overlap. The counter-rotating instability [e.g. @Tou12] is suppressed due to precession resulting from the presence of a stellar cusp.
The distribution in angular momentum vectors of the young CW disk stars can become very spread out, in contrast to the observed dispersion angle [@Pau06; @Bel06; @Lu09; @Bar09]. This is due to secular gravitational torquing of the stellar orbits between the two disks. We find that small angles between the disks cause the most spreading as the torques are stronger when they are closer together (see Equation \[e:tau\]). Without the second disk, the orbits of the young CW disk are less spread out in angular momentum. A steep density profile in the stellar cusp also hinders spreading as the rapid orbital precession time decreases the timescale over which orbits can feel coherent torques. The inclination angle between the B-star disk and the young CW disk affects the resulting $\langle |h| \rangle$-values; the DD$_{3}$ simulation with $\langle i \rangle = 121 \arcdeg$ shows the lowest $\langle |h| \rangle$-values at large radii.
For each simulation, we chose ten random viewing directions to calculate a mean $|h|$-value for each star. In Table \[tab:sim\_params\] we show $\langle |h| \rangle$ and one standard deviation, $\sigma_h$, for all stars with projected radii $p \geq 7 \arcsec$ and $p \geq 10 \arcsec$. The [*binary disruption*]{} simulations produce low $\langle |h| \rangle$-values with a range ($0.095 - 0.141$) for $p \geq 7 \arcsec$ and ($0.095 - 0.129$) for $p \geq 10 \arcsec$ with $\sigma_h \sim 0.06$ on most values.\
We also show the standard error on the mean, $s_e$, of $\langle |h| \rangle$ for the ten different viewing directions. [*Binary disruption*]{} simulations have small $s_e$ values, reflecting the fact that the B-stars are isotropically distributed and as a consequence the different viewing directions produce similar results.
The [*dissolved disk*]{} simulations produce higher mean $\langle |h| \rangle$-values with a range ($0.277 - 0.573$) for $p \geq 7 \arcsec$ and ($0.251 - 0.563$) for $p \geq 10 \arcsec$ with $\sigma_h \sim 0.2$ on most values. [*Dissolved disk*]{} simulations have large $s_e$ values as the different viewing directions produce different results, depending on the angle between the viewing direction and the plane of the B-star disk. To further demonstrate this, in Figure \[fig\_DD\_inc\_cum\] we plot the cumulative $|h|$-distribution function as a function of viewing angle with respect to the B-star disk in [*dissolved disk*]{} simulation, DD$_1$. The left plot shows data from the start of simulation in which the disk has an opening angle of $30 \arcdeg$. Larger viewing angles with respect to the disk plane produce larger $|h|$-values. The right plot shows the data after $6 {{\,\rm Myr}}$. Here the viewing directions produce more similar results to each other as the disk opening angle has increased and the orbits are more isotropically distributed. The disk structure is still non-isotropic however; the $\langle |h| \rangle$-values increase from $0\arcdeg - 90\arcdeg$.
Observational data {#S:obs}
==================
Observations and data reduction
-------------------------------
The observational data comprise 207 spectroscopically identified early-type stars within $25\arcsec$ ($\sim 1{\,\mathrm{pc}}$) from SgrA\*. The data set incorporates those used in the studies of [@Bar09], [@Bar10], [@Pfu11] and additional stars found in new fields which will be reported in Fritz et al. (in prep). We refer the reader to the [@Bar10] for a detailed description of the data reduction process.
The corresponding observations have taken place between 2002 and 2010 at the VLT in Cerro Paranal Chile (ESO programs 075.B-0547, 076.B-0259, 077.B-0503, 179.B-0261 and 183.B-0100). The imaging data were obtained with the adaptive optics camera NACO [@Rou03; @Har03]. The photometric reference images were taken on the 29th of April 2006 and on the 31st of March 2010. Most of the images were obtained, using the $K'$-band filter ($2.17\, \rm \mu m$) together with the 27$\rm{mas/pixel}$ camera of NACO. Each image was processed in the same way, using sky-subtraction, bad-pixel and flat-field correction as described in [@Tri08]. The spectroscopic data were obtained with the adaptive optics assisted integral field spectrograph SINFONI [@Eis03; @Bon04]. The data output of SINFONI consists of cubes with two spatial axes and one spectral axis. Depending on the plate scale, an individual cube covers $3.2\arcsec \times3.2\arcsec$ or $8\arcsec \times8\arcsec$; the spectral resolution varies between 2000 and 4000 depending on the chosen bandpass and the field-of-view.\
A major challenge for the identification of stars in the Galactic Center is stellar crowding. The bulk of the resolved stellar population are low-mass giants later than K0III. Due to the extreme extinction towards the Galactic Center ($A_K\approx2.7$), foreground stars are easily excluded by their blue color. The patchiness of the extinction, however, prevents a photometric distinction between evolved giants ($\rm T\sim 4000\, \rm K$) and young early-type stars ($\rm T>15,000\,\rm K$; spectral types between B9V up to WR$/$O) in the cluster. $K$-band spectra of the stars classify them as late-type if CO absorption features (2.29 -2.40$\rm \,\mu m$) are present, or early-type if Br$\gamma$ (2.166$\rm \,\mu m$) or HeI (2.058$\rm \,\mu m$) absorption lines are present. In this way, the combination of imaging and spectroscopy allows a clear identification of the young and massive early-type stars amongst the equally bright but much more numerous old low mass giants.
Astrometry
----------
Similar to the method of [@Tri08] the stars in the individual images were detected using the algorithm FIND [@Ste87]. The individual positions in each image were retrieved using Gaussian fits with formal fit errors of the order $\approx300 \, \rm \mu as$. About 560 bright isolated stars in the field served as an astrometric reference frame. The proper motions were computed by fitting a linear function to the astrometric star positions with time. Typical errors were $\rm \approx0.13\,mas$/${{\,\rm yr}}$ ($\rm 5\,km/s$).
Stellar masses and main-sequence lifetimes
------------------------------------------
The early-type stars in our sample contain main-sequence B-stars up to evolved WR/Ofpn stars. An individual classification of the spectral type however is non-trivial with $K$-band spectra alone. One difficulty is the absence of a simple correlation between Br$\gamma$ (HeI) strength and spectral type. Only the most massive WR$/$O stars can be recognized due to their strong wind emission lines, but the large mass-loss rates make a current mass estimate even more difficult and these stars must be modeled in detail to derive temperatures and masses [e.g., @Mar07]. Only stars more massive than $\sim20-25 {M_{\odot}}$ reach the WR phase however [@Mae04]. The mass of the less luminous B-stars can be deduced from their observed luminosity [using e.g., @Ber94 isochrones]. Although the spectral information alone provides not a unique classification, the known absolute magnitude of the stars allows to constrain their masses quite accurately. Unlike any other OB clusters, the distance to the Galactic Center is known to better than 5% ($\rm 8.33\pm0.35\,kpc$). Together with precise extinction measurements [@Fri11 $A_K\approx2.7$], the absolute magnitudes of the stars are known to $\approx$ 0.2mag. We use magnitude, infrared color, temperature and mass calibrations from [@Cox00], and ages from [@Sal06]. An $m_K=14.1$ star at the distance of the GC corresponds to a B0V dwarf with an initial mass of $M_{\rm MS}=17\,M_{\odot}$ and a main-sequence lifetime of $t_{\rm MS}=8\rm {{\,\rm Myr}}$, which is of the same order as the young CW disk age. Fainter B-dwarfs with $m_K=15.5$ ($M_{\rm MS}=11\,M_{\odot}$, $t_{\rm MS}=25\rm {{\,\rm Myr}}$) and $m_K=16.5$ ($M_{\rm MS}=6\,M_{\odot}$, $t_{\rm MS}=120\rm {{\,\rm Myr}}$) can be significantly older than the young CW disk.
[ccccccccccc]{} & $m_{\rm K}$ & x & y & p & ex & ey & vx & vy & evx & evy & (mag) & (arcsec) & (arcsec) & (arcsec) & (arcsec) & (arcsec) & (mas ${\,{\rm yr^{-1}}}$) & (mas ${\,{\rm yr^{-1}}}$) & (mas ${\,{\rm yr^{-1}}}$) & (mas ${\,{\rm yr^{-1}}}$) 1 & 16.7283 & 0.0463 & -0.1040 & 0.1139 & 0.0009 & 0.0008 & -7.2066 & -27.1863 & 0.4076 & 0.3534 2 & 15.8604 & -0.0673 & 0.1182 & 0.1360 & 0.0015 & 0.0006 & 42.2079 & 26.0934 & 0.4189 & 0.1749 3 & 17.4537 & 0.1403 & -0.0311 & 0.1437 & 0.0020 & 0.0017 & -2.3484 & -25.4647 & 0.7190 & 0.6107 4 & 14.2067 & 0.0281 & 0.1458 & 0.1485 & 0.0002 & 0.0006 & -8.8333 & 25.1138 & 0.0793 & 0.2235 5 & 16.9512 & -0.1984 & -0.0702 & 0.2105 & 0.0004 & 0.0006 & -17.2400 & -18.2476 & 0.1970 & 0.2380 6 & 15.6632 & 0.1873 & 0.1345 & 0.2306 & 0.0005 & 0.0004 & 23.1651 & 18.5350 & 0.1129 & 0.0978 7 & 15.7113 & 0.2264 & 0.1111 & 0.2522 & 0.0007 & 0.0005 & -5.2460 & -6.1800 & 0.2829 & 0.1908 8 & 14.7296 & 0.0231 & -0.2547 & 0.2558 & 0.0001 & 0.0001 & 22.6434 & -18.2185 & 0.0470 & 0.0564 9 & 14.5520 & 0.3052 & 0.1216 & 0.3285 & 0.0002 & 0.0001 & 12.8644 & -0.6664 & 0.0561 & 0.0324 10 & 16.5369 & 0.1906 & -0.2697 & 0.3302 & 0.0002 & 0.0003 & 22.7268 & -6.9537 & 0.1068 & 0.1335
$h$-values in different $K$-magnitude ranges
--------------------------------------------
In Figure \[fig\_h\_plane\_magbins\] we plot $h$-values of stars as a function of projected radius, $p$, for different $K$-magnitude ranges. Error bars are determined through error propagation from position and proper motion uncertainties given in Table \[tab:properMotions\], and an MBH mass uncertainty of $0.36 \times 10^6 {M_{\odot}}$ [@Gil09b]. Stars in our sample have, on average, lower $|h|$-values as their $K$-magnitude increases. In Figure \[fig\_cum\_data\] we plot the cumulative $|h|$-distribution function for stars in these different $K$-magnitude ranges. Stars with $m_K \geq 15$ have lower $|h|$-values than those with $m_K < 15$, and if isotropically distributed, form a population more eccentric than a thermal distribution, $N(e) de \sim e de$. We compare different distributions to each other using the one-dimensional two-sample Kolmogorov-Smirnov (KS) test, under the null hypothesis that the samples are drawn from the same distribution. We refer to the test statistic as $p_{\rm ks}$ to distinguish it from the projected radius of a star, $p$. We reject the null hypothesis if $p_{\rm ks}$ is smaller than or equal to the significance level, $\alpha = 0.05$. A KS test between stars with $14 \leq m_K < 15$ and those with $m_K \geq 15$ yields $p_{\rm ks} = 0.007$, which suggests they are not drawn from the same population, and between stars with $15 \leq m_K < 16$ and $m_K \geq 16$ yields $p_{\rm ks} = 0.965$.
In Figures \[figHistAllData\] - \[figHist1415\] we plot histograms of $h$-values for observational data, binned in $K$-magnitude and radius, compared to that for a simulated thermal eccentricity distribution. The $m_K \ge 14$ population contains fractionally more CCW orbits and low $|h|$-values than the $m_K < 14$ population. Stars with $14 \leq m_K < 15$ have a flatter $h$-distribution than those with $m_K \ge 15$ and are missing the central low-$|h|$ peak. We plot $h$-values for stars with projected radii less than and greater than $7\arcsec$ as the azimuthal coverage drops at his radius (corresponds to middle black circle in Figure 1 of @Bar10). The $m_K \ge 14$ stars with $p > 7\arcsec$ have low $|h|$-values; this is not seen in the $m_K < 14$ population. The are more $m_K \ge 14$ stars with $p > 7\arcsec$ on CCW orbits than CW orbits but this result is not statistically significant.
In Figure \[figHistOnOffDisk\] we plot a histogram of $h$-values for $m_K < 14$ stars on and off the young CW disk using a re-analysis of @Bar10 data (this re-analysis revealed 43 stars to be on the CW disk, instead of 45). The off-disk population have slightly lower $|h|$-values than the on-disk population, indicative of a more eccentric or inclined population. As the CW disk is more edge-on than face-on, a population of stars with the same eccentricities but more isotropically distributed should have higher $|h|$-values. This suggests the off-disk population is more eccentric. They also have more $h < 0$ values or CCW orbits. In Figure \[fig\_cum\_data\_CWdisk\] we plot the cumulative $|h|$-distribution function for the same data. The relatively high $|h|$-values of the young CW disk stars suggests, given their inclination, that they are actually quite a low-eccentricity population.
$|h|$-values as function of projected radius, $p$
-------------------------------------------------
In Figure \[fig:p\_gtr\_8\] we plot the cumulative $|h|$-distribution function for data in different $K$-magnitude ranges, binned according their projected radius, either less or greater than $p = 0.8 \arcsec$. All stars in our sample with $p < 0.8 \arcsec$ have $m_K \geq 14.2$; these are the ‘S-stars’. They have a $|h|$-distribution that, if isotropically distributed, is just slightly more eccentric than a thermal eccentricity distribution. A KS-test between the two populations (S-stars and thermal distribution) yields $p_{\rm ks} = 0.48$. Stars with $m_K \geq 16$, if isotropically distributed, form a more eccentric group at large radii, $p \geq 0.8 \arcsec$, than at small radii ($p_{\rm ks} = 0.017$ between the two populations). Stars between $14 \leq m_K < 15$ appear more eccentric at smaller projected radii, $p < 0.8 \arcsec$, but the two groups do not differ in a statistically significant way ($p_{\rm ks} = 0.16$). In Figure \[fig\_hist\_fn\_p\] we show a histogram of $|h|$-values as a function of projected distance for stars in different magnitude ranges. While the binning is arbitrary, this visually confirms that stars with $m_K \geq 15$ have lower $|h|$-values at larger projected radii (i.e., become more eccentric/inclined to $xy$-plane), while the opposite is true for stars with $14 \leq m_K < 15$. In Table \[tab:p\_gtr\_7\] we compare the mean $|h|$-values for stars with $p \geq 7 \arcsec, 10 \arcsec$. Stars with higher $K$-magnitude have lower mean $|h|$-values at large radii.
[ccccccc]{} Sample & & & Num & $\langle |h| \rangle^{b}$ & $\sigma_{|h|}^{c}$ & Num & $\langle |h| \rangle^{b}$ & $\sigma_{|h|}^{c}$ all & 82 & 0.390 & 0.261 & 39 & 0.300 & 0.204 $14 \leq m_K < 15$ & 10 & 0.412 & 0.182 & 7 & 0.333 &0.142 $15 \leq m_K < 16$ & 13 & 0.181 & 0.149 & 9 & 0.200 & 0.170 $m_K \geq 15$ & 19 &0.158 & 0.131 & 13 & 0.176 & 0.148 $m_K \geq 16$ & 6 & 0.109 & 0.066 & 4 & 0.126 & 0.076
\[tab:p\_gtr\_7\]
Comparison between observations and simulations {#S:comp}
===============================================
We compare the observational data with simulation results from two formation scenarios of B-stars in the Galactic center: the [*binary disruption*]{} scenario in which stars begin their lives near the MBH on high-eccentricity ($e \simeq 0.97$) orbits, and the [*dissolved disk*]{} scenario in which they form in a nuclear stellar disk of lower eccentricity ($0.3 \lesssim e \lesssim 0.7$). In Figure \[fig\_h\_plane\_comp\_sims\_data\] we show $h$-values of both simulated stars (at the end of the simulation, $t = 6 {{\,\rm Myr}}$) and data in different magnitude ranges, as a function of their projected radii, $p$. The higher magnitude stars, $m_K \geq 15$, qualitatively match the [*binary disruption*]{} simulations well while the lower magnitude stars, $14 \leq m_K < 15$, match better the [*dissolved disk*]{} simulations.
We compare the range in $\langle |h| \rangle$-values for stars at large projected radii in our simulations to the single values taken from observational data binned in $K$-magnitude (see Table \[tab:p\_gtr\_7\]), in Figure \[fig\_h\_grt\_7\_10\]. For stars with $p \geq 7 \arcsec$ and $14 \leq m_K < 15$, $\langle |h| \rangle = 0.412$, which places this population well outside the range found in the [*binary disruption*]{} simulations ($0.095 - 0.141$), whereas for $m_K \geq 16$, $\langle |h| \rangle = 0.109$, which lies inside the correct range for [*binary disruption*]{} simulations, not for [*dissolved disk*]{} simulations, for which $0.277 \leq \langle |h| \rangle \leq 0.573$.
KS-testing, correcting for incompleteness and incorporating observational errors
--------------------------------------------------------------------------------
[cccc]{} Radial Bin & $m_K < 14$ & $14 \leq m_K < 15$ & $15 \leq m_k < 17$ (arcsec) & (mag) & (mag) & (mag) 0 - 5 & 0.564400 & 0.454493 & 0.155868 5 - 10 & 0.390942 & 0.237714 & 0.144753 10 - 15 & 0.249561 & 0.106667 & 0.066129 15 - 20 & 0.164209 & 0.043871 & 0.026194
\[tab:completenessCorrection\]
To include the dependence of $|h|$-values on projected radius, we compare our simulations with the data using radial completeness corrections derived from from observations in @Bar10. We show the corrections used in radial and magnitude bins in Table \[tab:completenessCorrection\]. We calculate projected $|h|$ and $p$ values of the stars in our simulations and then sample according to the completeness at their projected radii and $K$-magnitude. For illustration, we plot the cumulative $|h|$-distributions of all the [*dissolved disk*]{} simulations and [*binary disruption*]{} simulations in Figure \[fig:cum\_h\_bd\_sims\_cc\]. $|h|$-values are sampled from a single random viewing direction, using the observational completeness correction for stars with $14 \leq m_K < 15$ in the [*dissolved disk*]{} simulations, and for stars with $15 \leq m_K < 17$ in the [*binary disruption*]{} simulations. We note that the results from the [*dissolved disk*]{} simulations are highly-dependent on the chosen viewing angle, in contrast to the [*binary disruption*]{} simulations. In Figure \[fig\_HISTBD1DD6\] we plot a histogram of the $h$-distributions of B-stars in DD$_6$ and BD$_1$ simulations.
To incorporate the errors in the observed $h$-values, we take each star in a magnitude-selected range and Monte Carlo sample 100 times from a Gaussian distribution with mean $h$ and standard deviation $e_h$. We run a KS-test between the resulting cumulative distribution functions and those from completeness-corrected simulations. For the [*dissolved disk*]{} simulations we select ten random viewing directions with respect to the B-star disk in the simulation and generate new $|h|$-values before sampling according to $p$-values. Hence each [*dissolved disk*]{} simulation will have $10 \times 100$ $p_{\rm ks}$ values. In Figure \[fig:aitoff\] we plot the angular momentum vectors of young CW disk stars and B-stars in an Aitoff projection in the [*dissolved disk*]{} simulations DD$_1$, DD$_2$, DD$_3$ and DD$_4$, indicating the ten randomly-selected viewing directions.
In Figure \[fig:scatter\_ks\_sims\] we plot the average $p_{\rm ks}$ value between the completeness corrected simulation (listed on $x$-axis) and stars from a particular magnitude bin (title) as a function of the viewing direction. The $p_{\rm ks}$ values are color-coded into two ranges for simplicity: orange for $p_{\rm ks} < 0.05$, blue for $p_{\rm ks} \ge 0.05$, where we reject the null hypothesis that two samples are drawn from the same distribution if $p_{\rm ks} < 0.05$. The first four plots show that, for most viewing directions, the [*dissolved disk*]{} simulations are compatible with observations in the lowest magnitude range, $14 \leq m_K < 15$. Stars in the range $16 \leq m_K < 17$ are also mostly compatible with [*dissolved disk*]{} simulations. There are however few stars in this range, and both simulations and observations are close to a thermal distribution over all semi-major axes, though they are not so similar at large $p$. The fifth plot shows results for [*binary disruption*]{} simulations. The viewing direction is unimportant as the B-stars are distributed isotropically. In this plot the $y$-axis shows the lower value on the range of the magnitude cut taken for the observations. The range spans one magnitude in total, $\Delta m_K = 1$ (i.e., $m_K = 14$ spans the range $14 \leq m_K < 15$). This plot shows that lower magnitude stars and the binary disruption scenario are incompatible. A change occurs in the simulations BD$_{1-6}$ (young CW disk mass of $10^4 {M_{\odot}}$) for the range $14.8 \leq m_K < 15.8$ and above. Here $p_{\rm ks} \geq 0.05$ for all simulations. This transition is also seen in the [*dissolved disk*]{} simulations but in the opposite direction: mostly $p_{\rm ks} \geq 0.05$ for $14 \leq m_K < 14.8$, and $p_{\rm ks} < 0.05$ for $m_K \ge 14.8$.
Samples $p_{\rm ks}$
---------------------------- --------------
$14 \leq m_K < 15$, BD$_8$ 0.039
$14 \leq m_K < 15$, DD$_1$ 0.953
$15 \leq m_K < 16$, BD$_8$ 0.253
$15 \leq m_K < 16$, DD$_1$ 0.001
$m_K \geq 16$, BD$_8$ 0.997
$m_K \geq 16$, DD$_1$ 0.094
: Example results of 1D two-sample KS testing between completeness corrected simulations (random viewing direction) and observational data.
\[tab:pks\]
In Table \[tab:pks\] we provide a sample of $p_{\rm ks}$-values between the data binned in $K$-magnitude and completeness-corrected simulations. The population with $14 \leq m_K < 15$ is inconsistent with the basic [*binary disruption*]{} scenario (BD$_8$). In contrast, a comparison with the [*dissolved disk*]{} scenario (DD$_1$) yields a high $p_{\rm ks}$-value. Again, [*dissolved disk*]{} scenarios give more ambiguous predictions as the $h$-values are dependent on the viewing direction with respect to the B-star disk. Fainter ($m_K \geq 16$) stars have lower $|h|$-values and hence are more eccentric or edge-on in inclination. A KS test between this population and BD$_8$ yields a high $p_{\rm ks}$ value $= 0.997$. In Figure \[fig\_cum\_data\_with\_sims\_cc\] we plot the cumulative $|h|$-distribution function for observational data binned in $K$-magnitude and for the simulations that match well with observations — DD$_1$ and BD$_8$ simulations sampled with completeness corrections.
Discussion {#S:discussion}
==========
We present a new, directly-observable statistic, $h$, which uses the position of stars on the sky ($x$, $y$) and their proper motion ($v_x$, $v_y$) to recognize groups of high-eccentricity orbits. It is particularly useful for stars with long-period orbits for which dynamical accelerations, and hence orbital parameters, are difficult to determine. We use a Monte Carlo ARMA code and $N$-body simulations to evolve stellar orbits in two formation scenarios for the B-stars in the GC; a [*dissolved disk*]{} scenario based on the model proposed by @Set06, and a [*binary disruption*]{} scenario due to enhanced stellar relaxation from massive perturbers by @Per07. We investigate the change in the B-star orbital parameters after $6 {{\,\rm Myr}}$ of gravitational interaction with the young CW disk and compare the results to observational data using the $h$-statistic. We summarize our results here:
1. Although the gravitational potential of the young CW disk can effectively exert torques on the orbits of the surrounding cluster stars within a few ${{\,\rm Myr}}$, for a disk mass of $\sim 10^4 {M_{\odot}}$, stars with semi-major axes greater than $\sim 0.2 {\,\mathrm{pc}}$ retain memory of their origin through their eccentricity distribution. The more massive the young CW disk, the greater torque it exerts and larger the eccentricity evolution of surrounding stars over $6 {{\,\rm Myr}}$. This result does not qualitatively change if the young CW disk is younger – such as the $\sim4 {{\,\rm Myr}}$ as found by @Do13 and @Lu13 – but the high-eccentricity signature of the B-stars in the [*binary disruption*]{} scenario will be even more prominent as the stars have less time to interact with the young CW disk.
2. Simulations in which the B-star and young CW disks have small angles with respect to one another produce a large spread in angular momentum vectors of the young CW disk stars, in contrast to the observed dispersion angle [@Pau06; @Bel06; @Lu09; @Bar09]. The concentration seen in the data for the young CW disk hints at a low initial eccentricity, a cusp rather than a core in stellar density and/or no second disk structure. However it is difficult to keep a distinct concentration of angular momentum vectors and, at the same time, produce off-disk and counter-rotating orbits with a secular mechanism.
3. The [*binary disruption*]{} scenario leaves a signature of decreasing mean values, and scatter, in $|h|$ with increasing radii. The [*dissolved disk*]{} scenario results in a broad range of $\langle |h| \rangle$-values with a large scatter due to the dependence on viewing angle with respect to the initial disk.
4. The B-stars in our data set have lower $|h$|-values with increasing $K$-magnitude intervals. If their orbits are isotropically distributed, this means that the lower mass, potentially much-older B-stars are more eccentric than their more massive, younger companions. If the B-stars are preferentially aligned with the young CW disk, incomplete azimuthal coverage at large radii and sampling along the disk can lower the $|h|$-distribution and mimic a high eccentricity signature. However the $m_K < 14$ stars are sampled with the same azimuthal coverage as the $m_K \ge 14$ population so this cannot explain the difference between the low and high $K$-magnitude stars.
5. The cumulative $|h|$-distribution function for the S-stars is similar to, but slightly lower than, that of an isotropic, thermal eccentricity distribution. As we know that they are isotropically distributed, this tells us that they form a population that is slightly more eccentric than thermal. This matches the distribution found from orbital fitting of individual stars [@Gil09a].
6. Stars with $14 \leq m_K < 15$ and those with $m_K \geq 15$ have different cumulative $|h|$-distributions with a KS value of $p_{\rm KS} = 0.007$. This suggests that they are not drawn from the same ($e,i$) population. The difference is important for the interpretation of the K-band luminosity function (KLF) and initial mass function (IMF) of the young CW disk. If this dynamical information relates to a different origin mechanism, then then the KLF slope of the young CW disk must get flatter and the IMF is even more top-heavy than previously reported [@Bar10; @Do13; @Lu13].
7. The stars with the highest $K$-magnitudes in our sample, $m_K \geq 16$, have similar $|h|$-distributions to those with $15 \leq m_K < 16$ with a KS value of $p_{\rm KS} = 0.965$. They have lower $|h|$-values at large radii, $ p \geq 0.8 \arcsec$. If they are isotropically distributed, they are more eccentric than those located closer in projected distance to the MBH. Stars with $m_K \geq 16$ and $p \geq 0.8 \arcsec$ do not appear to be drawn from the same distribution as those with $m_K \geq 16$ and $p < 0.8 \arcsec$ ($p_{\rm KS} = 0.02$). In context of the [*binary disruption*]{} scenario, this can be explained by the decreasing relaxation times as orbits get closer to the MBH, which rapidly changes their low angular momenta and hence $|h|$-values. We would like to increase our sample of $m_K \geq 16$ stars to confirm this result.
8. In comparing the observed B-star data with simulations of the two formation scenarios, we find the following: Stars with $14 \leq m_K < 15$ have higher $|h|$-values than expected for a thermal distribution and hence fit better with the [*dissolved disk*]{} scenario. Given the short lifetimes for these stars ($t_{\rm MS} \lesssim 13 {{\,\rm Myr}}$), they are most likely members of the recent star formation episode which created with young CW disk. Fainter B-stars with $m_K \geq 15$, and hence longer lifetimes, have lower $|h|$-values than expected for a thermal distribution and for that reason fit better with Hills [*binary disruption*]{} scenario, though the data are not as eccentric (if isotropically distributed) as in most of the simulations. The best match to the data involves a steep stellar cusp such that coherence times for stellar torques are high at large radii, and/or a larger mass of the young CW disk ($\sim 4 \times 10^4 {M_{\odot}}$).
An alternative scenario for the origin of the B-stars is formation in the same star formation episode that formed the young CW disk [@Pau06; @Do13; @Lu13]. To explain the low $|h|$-values of the high-magnitude stars however, there must exist a mechanism which differentiates between low- and high-mass stars in orbital eccentricity and/or inclination, and it must act on a short timescale. @Ale07 show that, for an initially circular disk, energy relaxation between stars of different masses can significantly change the velocity dispersion of different populations of stars of different masses, and hence change their orbital eccentricities. We repeat their analysis and find that, though important for low and moderately eccentric disks ($e \lesssim 0.7$), this mechanism cannot account for the magnitude of the difference in $h$-values of the low and high $K$-magnitude stars.
It is important to obtain more observations of B-star positions and proper motions, particularly those at high $K$-magnitude and large projected radii, to increase the sample-size that can be used in comparison with simulations. Recent observations show that the oldest stars observable in the GC, red giants, do not form a cusp within $0.5 {\,\mathrm{pc}}$ of the MBH, in contradiction with theoretical predictions [see e.g., @Do13b]. Whether or not this result is specific to the red giant population or true also of the less luminous stellar distribution is important for NSC formation and evolution theory. If the high magnitude B-stars originate from the [*binary disruption*]{} scenario, the shape of the $\langle |h| \rangle$-$r$ relation can be a probe of the mass distribution of the dark stellar cluster.\
Constraints on $\lowercase{h}$-statistic for bound stars
========================================================
The maximum value of $|h|$ for a bound stellar orbit is calculated by limiting the velocity of a star to the escape velocity at its projected radius. The maximum angular momentum of a stellar orbit on the plane of sky is $$\begin{split}
j_z =& v_{\rm esc} . p \\
=& \sqrt{2G{M_{\bullet}}p},
\end{split}$$ and therefore, $$h = \frac{j_z}{J_p} \le \sqrt{2}.$$ For a bound Kepler orbit, $h \le \sqrt{2}$. One can use this constraint to find unbound stars and/or stars affected by confusion with incorrect proper motion values.\
A star with $1 < |h| \le \sqrt{2}$ has a $j_z$ value larger than the circular angular momentum at its projected radius. This requires that $e > 0$, but moreover that $p < a$, i.e., that the star’s semi-major axis is larger than its projected radius. This star is traveling on the inner part of its orbit, closer to periapsis than apoapsis. This may provide an extra constraint on $z$, the position of a star along the line-of-sight, when estimating stellar orbital parameters [@Bar09; @Lu09].
Effect of massive stellar potential on value of $\lowercase{h}$-statistic
=========================================================================
The $h$-statistic is defined for a Kepler orbit ($h = j_z/J_p$). In a real NSC, the mass contained within a stellar orbit is due to both the black hole mass and the enclosed mass of the stellar cusp. If we do not take the latter mass into account, our theoretical value of $J_p$, the maximum angular momentum at $p$, will be smaller than the true value. Hence, $|h|$ will appear artificially larger (i.e. a stellar orbit will appear less eccentric or more ‘face-on’) than its true value. Due to the increasing enclosed stellar mass with radius, this effect will increase with radius. We quantify this in Figure \[fig\_stellarmass\] where we plot the fraction of two evaluations of $h$ (one takes stellar mass into account in calculating $J_p$, $h_{\rm cusp}$, the other does not, $h_{\rm Kepler}$) for a face-on circular orbit ($e = i = 0$). We assume a stellar mass of $1.5 \times 10^6 {M_{\odot}}$ within $1 {\,\mathrm{pc}}$ and vary the power-law index $\alpha$, where the mass within radius $r$ scales as $m(<r) \propto r^{\alpha}$. From this simple analysis we see that a star with a face-on circular orbit at $p=0.4 {\,\mathrm{pc}}\sim 10 \arcsec$, will have a $h$-value that is fractionally larger by ($0.034, 0.043, 0.067$) than for a Kepler potential for this particular power-law density profile with $\alpha = (1.75, 1.5, 1.0)$. This plot can inform us on the magnitude of the expected error on $h$ due to the stellar cusp.
Statistical constraints on orbital eccentricity and inclination from the $\lowercase{h}$-statistic
==================================================================================================
There is a degeneracy in the value of $h$ with respect to orbital eccentricity and inclination. However, if we assume a known distribution for one of the parameters we can place constraints on the mapping of the other to $h$. For example, we can take a cut in inclination (see Figure \[fig\_inc\]) and see how the range in $h$-values map to orbital eccentricity.
[ccccccc]{} $|h|^{1}$ & $\mu$ & $\sigma^2$ & $\beta$ & $\gamma$ & $\delta$ & $\epsilon$ $< 0.1$ & $0.96 $ & $0.17$ & $-1.06$ & $-1.22$ & $1.20$ & $1.10$ $< 0.3$ & $1.06$ & $0.01$ & $4.22$ & $0.15$ & $-0.16$ & $-4.09$$< 0.6$ & $1.66$ & $0.16$ & 5.43 & $0.09$ & $-0.10$ & $-5.26$$> 0.1$ & $0.34$ & $0.13$ & 5.75 & $0.29$ & $ -0.45$ & $-6.52$$> 0.3$ & $0.71$ & $0.07$ & -0.62 & $0.01$ & $ -0.05$ & $1.00$$> 0.6$ & $0.51$ & $0.05$ & -2.80 & $-0.54$ & $ 0.51$ & $3.15$
Statistical constraints are placed on the mapping of orbital eccentricities to $|h|$ for an isotropically distributed thermal cluster of stars. We plot the cumulative distribution of orbital eccentricities in specific $|h|$ ranges in Figure \[fig:cum\_e\_iso\]. $50 \%$ of stars with $|h| < 0. 1 (0.3)$ have an orbital eccentricity $e > 0.87 (0.84)$, while $90 \%$ of stars with $|h| > 0.6$ have an orbital eccentricity $e < 0.73$. We fit the cumulative distributions shown in Figure \[fig:cum\_e\_iso\] with the following formula $$\label{eq:fit}
\begin{split}
f(x) &= \dfrac{1}{2} \left[ 1 + {\rm erf} \left( \dfrac{x - \mu}{\sqrt{2\sigma^2}} \right) \right] \\
&+ \beta x + \gamma \sqrt{1 - x^2} + \delta \cos(x) + \epsilon \sin(x),
\end{split}$$ where $\mu$ and $\sigma^2$ are the mean and variance of a cumulative Gaussian distribution, $\beta$, $\gamma$, $\delta$ and $\epsilon$ are coefficients and $x = e$. The trigonometric terms are chosen ad-hoc; they are necessary to fit the extreme distributions $|h| >0.6$, $|h| < 0.1$. The fitted values for each line in Figure \[fig:cum\_e\_iso\] are listed in Table \[tab:cdf\_params\]; the residuals from subtracting the functions from the distributions are plotted in the lower panel.
Statistical constraints are also placed on the mapping of orbital inclinations to $h$ for an isotropically distributed thermal cluster of stars. In Figure \[fig\_inc\_cum\], we plot the cumulative distribution function of the stellar inclinations for different $h$-values. Those with $|h| > 1$ (right) have orbital inclinations such that their angular momenta are aligned close to the $z$-axis (face-on orbits). Stars with $|h| > 1.2$ have almost face-on orbits. Stars with very low values of $|h|$ are likely to have angular momenta that are highly inclined to the $z$-axis. Almost all stars with $|h| < 0.01$ have inclinations of $\sim 90\arcdeg$ (edge-on orbit).
Convergence Testing
===================
We run two [*binary disruption*]{} simulations with BD$_{1}$ parameters and individual stellar masses of $20 {M_{\odot}}$ (BD$_{20}$) and $50 {M_{\odot}}$ (BD$_{50}$) for convergence testing. For the $20 {M_{\odot}}$ simulation, $\langle |h| \rangle = 0.105 (0.103)$, $\sigma_h = 0.057 (0.053)$, $s_e = 0.003 (0.004)$ for stars with projected radii $p \geq 7(10) \arcsec$. For the $50 {M_{\odot}}$ simulation, $\langle |h| \rangle = 0.112 (0.110)$, $\sigma_h = 0.068 (0.061)$, $s_e = 0.007 (0.008)$ for $p \geq 7(10) \arcsec$. These values demonstrate that simulations with smaller stellar masses produce the same results (c.f. Table \[tab:sim\_params\]). In Figure \[fig\_inc\_cum\] we compare the cumulative distribution functions of $|h|$-values drawn from a random viewing direction for each simulation. KS testing between the distributions yield $p_{\rm ks} = 0.84, 0.79, 0.98$ for BD$_{1}$ and BD$_{20}$, BD$_{1}$ and BD$_{50}$, BD$_{20}$ and BD$_{50}$ respectively. These results are insensitive to the chosen viewing direction.
[77]{}
, M. & [Milosavljevi[ć]{}]{}, M. 2011, , 729, 35
, R. D., [Begelman]{}, M. C., & [Armitage]{}, P. J. 2007, , 654, 907
, F. 2013, , 763, 62
, F., [Capuzzo-Dolcetta]{}, R., [Mastrobuono-Battisti]{}, A., & [Merritt]{}, D. 2012, , 750, 111
, F. & [Merritt]{}, D. 2013, , 763, L10
, B., [Kupi]{}, G., & [Alexander]{}, T. 2013, , 764, 52
, H., [Martins]{}, F., [Fritz]{}, T. K., et al. 2009, , 697, 1741
, H., [Martins]{}, F., [Trippe]{}, S., et al. 2010, , 708, 834
, A. M., [Levin]{}, Y., [Eisenhauer]{}, F., et al. 2006, , 648, 405
, G., [Bressan]{}, A., [Chiosi]{}, C., [Fagotto]{}, F., & [Nasi]{}, E. 1994, , 106, 275
, R. D., [Ram[í]{}rez]{}, S. V., [Sellgren]{}, K., & [Olsen]{}, K. 2003, , 597, 323
, I. A. & [Rice]{}, W. K. M. 2008, Science, 321, 1060
, H., [Abuter]{}, R., [Baker]{}, A., et al. 2004, The Messenger, 117, 17
, W. R., [Cohen]{}, J. G., [Geller]{}, M. J., & [Kenyon]{}, S. J. 2012, , 754, L2
, R. 1993, , 415, 616
, A. N. 2000, [Allen’s astrophysical quantities]{} (New York: AIP Press; Springer, 4th ed.)
, J. M. A. 1992, [Fundamentals of celestial mechanics]{} (Richmond: Willman-Bell, |c1992, 2nd ed.)
, T., [Lu]{}, J. R., [Ghez]{}, A. M., et al. 2013, , 764, 154
, T., [Martinez]{}, G. D., [Yelda]{}, S., et al. 2013, , 779, L6
, F., [Sch[ö]{}del]{}, R., [Genzel]{}, R., et al. 2003, , 597, L121
, F. [et al.]{} 2005, , 628, 246
, T. K., [Gillessen]{}, S., [Dodds-Eden]{}, K., et al. 2011, , 737, 73
, R., [Eisenhauer]{}, F., & [Gillessen]{}, S. 2010, Reviews of Modern Physics, 82, 3121
, R., [Sch[ö]{}del]{}, R., [Ott]{}, T. [et al.]{} 2003, , 594, 812
, A. M., [Salim]{}, S., [Hornstein]{}, S. D., et al. 2005, , 620, 744
, A. M., [Salim]{}, S., [Weinberg]{}, N. N., et al. 2008, , 689, 1044
, A. M. [et al.]{} 2003, , 586, L127
, S., [Eisenhauer]{}, F., [Fritz]{}, T. K.,et al. 2009, , 707, L114
, S., [Eisenhauer]{}, F., [Trippe]{}, S., et al. 2009, , 692, 1075
, A. & [Quillen]{}, A. C. 2003, , 592, 935
, A., [Mapelli]{}, M., & [Perets]{}, H. B. 2012, , 427, 1793
, M., [Debattista]{}, V. P., [Seth]{}, A., [Cappellari]{}, M., & [Quinn]{}, T. R. 2011, , 418, 2697
, M., [Lenzen]{}, R., [Hofmann]{}, R., et al. 2003, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Vol. 4841, Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, ed. [M. Iye & A. F. M. Moorwood]{}, 425–436
, J. G. 1988, , 331, 687
—. 1991, , 102, 704
, P. F. & [Quataert]{}, E. 2010, , 407, 1529
—. 2010, , 405, L41
, P. B., [Polnarev]{}, A. G., & [Saha]{}, P. 2005, , 358, 1361
, H., [Yoshida]{}, H., & [Nakai]{}, H. 1991, Celestial Mechanics and Dynamical Astronomy, 50, 59
, B. & [Tremaine]{}, S. 2011, , 412, 187
, N., [B[ö]{}ker]{}, T., & [Knigge]{}, C. 2012, , 424, 2130
, Y. 2007, , 374, 515
, Y. & [Beloborodov]{}, A. M. 2003, , 590, L33
, J. R., [Do]{}, T., [Ghez]{}, A. M., et al. 2013, , 764, 155
, J. R., [Ghez]{}, A. M., [Hornstein]{}, S. D., et al. 2009, , 690, 1463
, A.-M., [Hopman]{}, C., & [Levin]{}, Y. 2011, , 738, 99
, A.-M. & [Levin]{}, Y. 2012, , 754, 42
, A.-M., [Levin]{}, Y., & [Hopman]{}, C. 2009, , 697, L44
, A. & [Meynet]{}, G. 2004, in IAU Symposium, Vol. 215, Stellar Rotation, ed. A. [Maeder]{} & P. [Eenens]{}, 500
Martins, F., Genzel, R., Hillier, D. J., et al. 2007, A&A, 468, 233
, F., [Gillessen]{}, S., [Eisenhauer]{}, F., et al. 2008, , 672, L119
, D. 2010, , 718, 739
, M. 2004, , 605, L13
, M. 1993, , 408, 496
, S. & [Cuadra]{}, J. 2005, , 437, 437
, S. & [Sch[ö]{}del]{}, R. 2013, , 549, A57
, T., [Genzel]{}, R., [Martins]{}, F., et al. 2006, , 643, 1011
, H. B. & [Gualandris]{}, A. 2010, , 719, 220
, H. B., [Gualandris]{}, A., [Kupi]{}, G., [Merritt]{}, D., & [Alexander]{}, T. 2009, , 702, 884
, H. B., [Hopman]{}, C., & [Alexander]{}, T. 2007, , 656, 709
, E. 2005, , 626, 849
, O., [Fritz]{}, T. K., [Zilka]{}, M., et al. 2011, , 741, 108
, K. P. & [Tremaine]{}, S. 1996, New Astronomy, 1, 149
, G., [Lacombe]{}, F., [Puget]{}, P., et al. 2003, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Vol. 4839, Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, ed. [P. L. Wizinowich & D. Bonaccini]{}, 140–149
, P. & [Tremaine]{}, S. 1992, , 104, 1633
, M. & [Cassisi]{}, S. 2006, [Evolution of Stars and Stellar Populations]{}
, R. H. 1998, , 294, 35
, R., [Eckart]{}, A., [Alexander]{}, T., et al. 2007, , 469, 125
, R., [Merritt]{}, D., & [Eckart]{}, A. 2009, , 502, 91
, A. C., [Dalcanton]{}, J. J., [Hodge]{}, P. W., & [Debattista]{}, V. P. 2006, , 132, 2539
, P. B. 1987, , 99, 191
, J. R. & [Sridhar]{}, S. 2012, , 423, 2083
, S. D., [Ostriker]{}, J. P., & [Spitzer]{}, Jr., L. 1975, , 196, 407
, S., [Gillessen]{}, S., [Gerhard]{}, O. E., et al. 2008, , 492, 419
, L. ., [Schovancov[á]{}]{}, J., & [Kroupa]{}, P. 2009, , 496, 695
, M. & [Yusef-Zadeh]{}, F. 2008, , 683, L37
—. 2012, , 750, L38
, J. & [Holman]{}, M. 1991, , 102, 1528
[^1]: We adopt a distance to SgrA\* of 8.3 kpc [@Eis03; @Ghe08a; @Gil09a; @Gil09b] for which $1\arcsec \sim 0.04 {\,\mathrm{pc}}$, an average extinction of $A_{K_s} = 2.7$ [@Fri11], and refer the reader to @Gen11 for a complete review of the GC.
[^2]: @Lu13 analyze the entire population of young stars as a single starburst cluster and find an age between $2.5-5.8 {{\,\rm Myr}}$ with $95\%$ confidence. Though a younger age than commonly adopted for the O/WR stars, it is consistent within the uncertainty range reported by @Pau06.
[^3]: The inclination, $i$, of a stellar orbit is calculated from the angle between its angular momentum vector and the positive $z$-axis (line-of-sight). Hence $i = 0 \arcdeg$ ($i = 90 \arcdeg$) corresponds to a face-on (edge-on) orbit.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'These notes are based on the mini-course “On the Graham Higman group”, given at the Erwin Schrödinger Institute in Vienna, January 20, 22, 27 and 29, 2016, as a part of the Measured Group Theory program.[^1] The main purpose is to describe p-quotients of the Higman group $H(k)$ for $p|(k-1)$. (One may check that the condition $p|(k-1)$ is necessary for the existence of such quotients.)'
author:
- 'L. Glebsky'
title: '$p$-quotients of the G.Higman group'
---
Higman group
============
Consider the Higman group $H(k)=\ls{a_0,\dots,a_{3}\;|\;\{a^{-1}_ia_{i+1}a_i=a_{i+1}^k,\;i=0,...,3\}}$, $i+1$ is taking $\mod 4$ here. It may be constructed as successive amalgamated free products, starting from Baumslag-Solitar group BS(1,k)=$\langle a_0,a_1\;|\;a^{-1}_0a_{1}a_0=a_{1}^k \rangle$ : $$B_3\langle a_0,a_1,a_2\rangle=\langle a_0,a_1\;|\;a^{-1}_0a_{1}a_0=a_{1}^k \rangle
\mathop{*}\limits_{a_1\leftrightarrow a_1}\langle a_1,a_2\;|\;a^{-1}_1a_{2}a_1=a_{2}^k \rangle.$$ Similarly, one may construct $B_3\langle a_2,a_3,a_0\rangle$ and $$H_k=B_3\langle a_0,a_1,a_2\rangle\mathop{*}_{a_0\leftrightarrow a_0,a_2\leftrightarrow a_2}
B_3\langle a_2,a_3,a_0\rangle$$ The group $H(2)$ was introduced by Graham Higman in [@Higman] as an example of group without finite quotients. Still $H(2)$ has a lot of quotients, moreover, it is SQ-universal, [@Lindon]. Actually the proof of [@Lindon] works for $H(k)$, $k\geq 2$, so, $H(k)$ is SQ-universal for any $k\geq 2$. Some other techniques that were used for $H(2)$ seem to be applicable for $H(k)$, see [@Hoff; @Mart]. But $H(k)$ for $k>2$ have another, compared with $H(2)$, behavior with respect to finite quotients. Particularly, $H(k)$ has an arbitrary large $p$-quotient for $p|(k-1)$.Moreover, the intersection of the kernels of these quotient maps intersects trivially with the Baumslag-Solitar subgroups $B(1,k)=\langle a_i,a_{i+1}\rangle<H(k)$. Using [@Hoff] it implies the following statement:
Let $p|(k-1)$ be a prime. Then for any $\varepsilon>0$ there is an $n\in{\mathbb Z}$ and a bijection $f : {\mathbb Z}/p^n{\mathbb Z}\to {\mathbb Z}/p^n{\mathbb Z}$ such that $f(x + 1) = kf(x)$ for at least $(1 - \varepsilon)p^n$ elements $x$ of ${\mathbb Z}/p^n{\mathbb Z}$ and $f (f (f (f (x)))) = x$ for all $x\in{\mathbb Z}/p^n{\mathbb Z}$.
The interesting property of this $f$ is that it behaves “almost like” a modular exponent ($x\to ak^x\mod p^n$), but all it’s cycles are of the length $4$. Precisely, $f(x)=a(x)k^x\mod p^n$ where $a(x)=a(x+1)$ for almost all $x$ ($a(x)$ is “almost a constant”). Maybe, the existence of such functions explains the difficulty in proving estimates for the number of small cycles in repeated modular exponentiation, [@GS]. The case $k=2$ is not follows from this notes. So, the existence of $f$ for $k=2$ is an open question.
Let $X$ be a group (or other algebraic system), $x\in X$ and $\phi:X\to Y$ be a homomorphism. We systematically, abusing notations, will write $x$ to denote $\phi(x)$ if it is clear from the context that we are dealing with an element of $Y$. If there are $\phi_j:X\to Y_j$ we may say “$x$ of $Y_j$” to denote $\phi_j(x)$.
$p$-quotients of a group and it’s $p$-central series. {#sec_p-quotient}
=====================================================
Let $G$ be a group. For $S\subseteq G$ let $\langle S\rangle$ denote the subgroup of $G$ generated by $S$. For $g,h\in G$ let $[g,h]=g^{-1}h^{-1}gh$ denote the commutator of $g$ and $h$. For $H_1,H_2<G$ let $[H_1,H_2]=\langle \{[g,h]\;|\;g\in H_1,\;h\in H_2\}\rangle <G$ denote the commutator subgroup of $H_1$ and $H_2$. The $p$-central series $G_1,G_2,\dots$ of a group $G$ is defined as $$G_1=G,\;\;G_{i+1}=G_i^p[G_i,G].$$ It is clear by the definition that $G/G_{i+1}$ is the maximal $p$-quotient of $G$ of $p$-class at most $i$. There is another, equivalent, definition of $G_k$. Let $G_{[i]}$ be the lower central series for $G$: $$G_{[1]}=G,\;\; G_{[i+1]}=[G_{[i]},G].$$
[**Exercise 1.**]{} Show that $G_n=\langle G_{[i]}^{p^j},\;i+j=n\rangle$. Hint. Using the commutator identities (Theorem 5.1 (Witt-Hall identities) of [@Mag1]) show that $$[u,v^p]=[u,v]^p\cdot [[u,v],v]\cdot [[u,v^2],v]\cdot \dots \cdot [[u,v^{p-1}],v].$$ Particularly, this implies that $[G_{[i]},G_{[j]}^{p^k}]\subseteq [G_{[i]},G_{[j]}^{p^{k-1}}]^p[G_{[i+j]},G_{[j]}^{p^{k-1}}]$. Show that $[G,G^{p^k}_{[i]}]\subseteq \langle G^{p^j}_{[r]},\; r+j=i+k+1\rangle$. Then apply induction on $n$.
Calculating of $p$-quotients.
=============================
Let ${\mathbb Z}_{p^n}={\mathbb Z}/p^n{\mathbb Z}$. Consider the non-commutative ring ${\mathbb Z}_{p^n}[\bar x]$ of polynomials with non-commutative (but associative) variables $\bar x =(x_0,\dots,x_m)$ over ${\mathbb Z}_{p^n}$. The ring ${\mathbb Z}_{p^n}[\bar x]$ contains finite subring ${\mathbb Z}_{p^n}[p\bar x]$ of polynomials $f(p\bar x)$. It is clear that each monomial of $f(p\bar x)$ of order $k$ is divisible by $p^k$. Inside of ${\mathbb Z}_{p^n}[p\bar x]$ there is a group $\Gamma=\ls{(1+px_0),\dots,(1+px_m)}$, generated by $(1+px_i)$. Notice, that $$(1+px)^{-1}=\sum_{j=0}^{n-1}(-p)^jx^j$$
\[lm\_power\] $(1+px_i)^{p^n}=1$ in ${\mathbb Z}_{p^n}[\bar x]$. In other words $j\to (1+px_i)^j$ is a function ${\mathbb Z}_{p^n}\to{\mathbb Z}_{p^n}[\bar x]$.
\[lm\_free\] $\Gamma$ is isomorphic to $F/F_{n}$ where $F$ is a free group of rank $m+1$.
We prove this lemma in Section \[sec\_lm\_free\]. Let $I$ be a (two-sided) ideal in ${\mathbb Z}_{p^n}[\bar x]$ and $\phi:{\mathbb Z}_{p^n}[\bar x]\to {\mathbb Z}_{p^n}[\bar x]/I$ be a natural map. Let $\Gamma_I=\phi(\Gamma)$. We are going to use $\Gamma_I$ for calculating $G/G_n$ as follows. Let $$G=\ls{a_0,\dots,a_m\;|\; u_i(a_0,\dots,a_m)=w_i(a_0,\dots,a_m),\;i=0,\dots,k}.$$ Let $g_i=p^{-\alpha_i}(u_i(1+px_0,\dots,1+px_k)-w_i(1+px_0,\dots,1+px_k))$, where $p^{\alpha_i}$ devides all coefficients of $u_i(1+px_0,\dots,1+px_k)-w_i(1+px_0,\dots,1+px_k)$ and being maximal with this property. Consider $I=I(g_i,\;i=0,\dots,k)$, an ideal in ${\mathbb Z}_{p^n}[\bar x]$ generated by $g_i$.
\[lm\_G\] $\Gamma_I$ is a homomorphic image of $G/G_{n}$.
By construction $\Gamma_I$ is a homomorphic image of $G$ under $\phi:a_i\to (1+px_i)$. It is easy to check that $\phi(G_{n})=\{1\}$.
When $\Gamma_I$ is isomorphic to $G/G_n$? Probably the answer is the following: if $p\neq 2$ then $\Gamma_I$ is isomorphic to $G/G_n$; for $p=2$ there are $G$ such that $\Gamma_I$ is not isomorphic to $G/G_n$. Probably, it is well known. Otherwise, one may try to use similar technique as in the proof of Lemma \[lm\_free\] (see, Section \[sec\_lm\_free\]) taking into account the solution of the dimension subgroup problem. See [@Gupta] and the bibliography therein for the dimension subgroup problem.
$p$-quotients of $H(k)$, $p|(k-1)$.
===================================
$H(k)=\ls{a_0,\dots,a_{3}\;|\;\{a_{i+1}a_i=a_ia_{i+1}^k,\;i=0,...,3\}}$, here $i+1$ is taken $\mod 4$. This is a presentation of the Higman group without inversion. Let $p|(k-1)$ then substituting $a_i=(1+px_i)$ leads $$\label{eq_g}
g_i=\frac{1}{p^2}(a_{i+1}a_i-a_ia_{i+1}^k)=
x_{i+1}x_i-x_ix_{i+1}+Q_0(x_{i+1})+pQ_1(x_i,x_{i+1}).$$ Our aim is to study $I=I(g_0,g_1,g_2,g_3)$ in ${\mathbb Z}_{p^n}[x_0,\dots,x_3]$, or precisely, ${\mathbb Z}_{p^n}[x_0,\dots,x_3]/I$ and $\Gamma_I$. To this end we introduce some notions.
\[def\_algebra\] A (non-commutative) ring $A$ is called to be an algebra over ${\mathbb Z}_{p^n}$ if
- $A$ has the unity $1\in A$.
- Multiplication is ${\mathbb Z}_{p^n}$-bilineal.
An algebra $A$ is tame if $A$ is a free ${\mathbb Z}_{p^n}$-modul with a free basis $\cA\ni 1$.
All algebras over ${\mathbb Z}_{p^n}$ we deal with are tame. So, in what follows we use just “algebra” to denote “tame algebra”.
\[def\_zappa-szep\_alg\] Let $A$ (resp. $B$) be a ${\mathbb Z}_{p^n}$-algebra and $\cA\ni 1$ (resp. $\cB\ni 1$) be a set of it’s free generators as a ${\mathbb Z}_{p^n}$-modul. A Zappa-Szep product $C=A\bowtie B$ is a ${\mathbb Z}_{p^n}$ algebra such that
- $C$ contains isomorphic copies of $A$ and $B$ such that $A\cap B={\mathbb Z}_{p^n}\cdot 1$ (we have fixed such a copies of $A$ and $B$ and denote them by the same letters).
- The set $\{ab\;|\;a\in\cA,\;b\in\cB\}$, as well as the set $\{ba\;|\;a\in\cA,\;b\in\cB\}$, forms a free basis of $C$ as a ${\mathbb Z}_{p^n}$-modul.
[**Exercise 2.**]{} It looks that by definition one should say that $C=A\bowtie B$ with respect to $\cA$ and $\cB$. Show that $C=A\bowtie B$ with respect to any free bases $\cA'\ni 1$ and $\cB'\ni 1$.
\[def\_zappa\_szep\_group\] Let $K,L$ be a groups. A Zappa-Szep product $G=K\bowtie L$ of groups $K$ and $L$ is a group such that
- $G$ contains isomorphic copies of $K$ and $L$. We fix such copies and assume that $K,L<G$.
- $K\cap L=\{1\}$ in $G$.
- $G=KL$. (This easily implies that $G=LK$.)
[**Remark.**]{} Of course, the definitions do not imply that a Zappa-Szep product is uniquely defined by a pair of algebras o groups. Really, in order to define $X\bowtie Y$ uniquely (up to isomorphism) one needs a function $com:X\times Y\to Y\times X$ which describe how the elements of $X$ commute with elements of $Y$. So, in some sense, $Z=X\bowtie Y$ is an abuse of notation. In any case, when we use $Z=X\bowtie Y$, the structure of $Z$ will be described.
\[th\_algebra\] Let $I=I(g_0,g_1,g_2,g_3)$ be an ideal in ${\mathbb Z}_{p^n}[\bar x]$ generated by $g_i$ of Eq.(\[eq\_g\]). Then ${\mathbb Z}_{p^n}[\bar x]/I= {\mathbb Z}_{p^n}[x_0,x_2]\bowtie{\mathbb Z}_{p_n}[x_1,x_3]$.
Recall, that $G_i$ denotes the $i$-th term of the $p$-central series of group $G=G_0$. Also we suppose that $p|(k-1)$.
\[cor\_group\] There is a surjective homomorphism $H(k)/H_n(k)\to F/F_n\bowtie F/F_n$, where $F$ is a free group of rank $2$.
By Lemma \[lm\_G\] $H(k)/H_n(k)$ surjects into $\Gamma_I$. By Theorem \[th\_algebra\] and Lemma \[lm\_free\] there are $S,\tilde S<\Gamma_I$ such that
- $S=\ls{a_0=(1+px_0),a_2=(1+px_2)}$, $\tilde S=\ls{a_1=(1+px_1),a_3=(1+px_3)}$;
- $S$ and $\tilde S$ are isomorphic to $F/F_n$;
- $S\cap\tilde S=\{1\}$;
It follows that $|S\tilde S|=|S|\cdot|\tilde S|$ and $S\tilde S\leq\Gamma_I$. So, we have to prove that $\Gamma_I\subset S\tilde S$. To this end it suffices to show that in $\Gamma_I$ there exist relations removing appearance of $a_1^{m}a_0^{r}$, $a_1^{m}a_2^{r}$, $a_3^{m}a_0^{r}$ and $a_3^{m}a_2^{r}$. By Lemma \[lm\_power\] we may assume that $m,r\in{\mathbb Z}_{p^n}$. The relation $a_1 a_0=a_0 a_1^k$ implies $a^m_1 a_0^r=a_0^r a_1^{mk^r}$. The relation $a_2 a_1=a_1 a_2^k$ implies $a_1^m a_2^r=a_2^{rk^{-m}}a_1^m$. Considerations for the other indexes are the same. Notice here, that the condition $p|(k-1)$ implies that $r\to k^r$ is a well definite function ${\mathbb Z}_{p^n}\to{\mathbb Z}_{p^n}$.
What is going on
================
Consider $w\in H_k$ as a word $a_{i_1}^{n_1}a_{i_2}^{n_2}\dots$ but now suppose that $n_j$ are in some commutative ring, in our example, $n_j\in{\mathbb Z}_{p^n}$. We get a new group $\tilde H_k$ wich is a homomorphic image of $H_k$. Now, the relation $a_1^k=a_0^{-1}a_1a_0$ implies as well the relation $a_1^{{1}/{k}}=a_0a_1a_0^{-1}$. We also need that $j\to k^j$ is well defined $\mod p^n$. After that any element of $\tilde H_k$ may be written as $wu$, where $w=a_0^{n_1}a_2^{m_1}\dots a_2^{m_j}$, $n_i,m_i\in{\mathbb Z}_{p^n}$ and, similarly, $u=u(a_1,a_3)$. So, we have $\tilde H_k=({\mathbb Z}_{p^n}*{\mathbb Z}_{p^n})\bowtie({\mathbb Z}_{p^n}*{\mathbb Z}_{p^n})$ The whole problem is to show that the natural homomorphism ${\mathbb Z}_{p^n}*{\mathbb Z}_{p^n}\to F/F_n$ is compatible with the $\bowtie$ structure. ( Here $F$ is a rank 2 free group). Which, probably, may be done another way as well...
Let restate all it more formally. On can easy to check that if $G_1=H_1\bowtie K$ and $G_2=H_2\bowtie K$ then $G_1\mathop{*}\limits_{K=K}G_2=(H_1*H_2)\bowtie K$. For $H_k$, $p|(k-1)$ we have the following:
- $BS(1,k)\to BS_{p^n}(1,k)={\mathbb Z}_{p^n}\ltimes_k{\mathbb Z}_{p^n}$,
- $B_3\to BS_{p^n}\mathop{*}\limits_{Z_{p^n}} BS_{p^n}=({\mathbb Z}_{p^n}*{\mathbb Z}_{p^n})\bowtie{\mathbb Z}_{p^n}$, where we amalgamate the different factors of $BS_{p^n}$.
- Finally we get $H_k\to ({\mathbb Z}_{p^n}*{\mathbb Z}_{p^n})\bowtie ({\mathbb Z}_{p^n}*{\mathbb Z}_{p^n})$.
For the case $k=2$ we may do similar things. We need a unitary commutative ring $R$ such that function $r\to 2^r$ is defined in $R$. Such a ring exists, for example the real numbers $\R$. So, we may embed $BS(1,2)\hookrightarrow \R\ltimes_2\R$, where for $(\alpha_i,\beta_i)\in \R\ltimes_2\R$ the multiplication is defined as $(\alpha_1,\beta_1)(\alpha_2,\beta_2)=(\alpha_1+\alpha_2,2^{\alpha_2}\beta_1+\beta_2)$. Similarly, there is a homorphism $B_3\to (\R*\R)\bowtie\R$, which is not injective, but nontrivial. And finally we obtain $H_2\to H_{\R}=(\R*\R)\bowtie(\R*\R)$. (Any element of $(\R*\R)\bowtie(\R*\R)$ is of the form $wu$, $w=a_{0}^{\alpha_1}a_2^{\beta_1}\dots a_2^{\beta_k}$, $\alpha_i\beta_i\in \R$ and, similarly, $u=u(a_1,a_3)$.) Actually, $H_k$ has a nontrivial homomorphic image in $H_{\R}$ for any $k\in{\mathbb Z}$.
What is the structure of $H_\R$? Changing $a_i\to a_i^\delta$ we may write $$H_\R=\langle a_i^\alpha,\; i=0,\dots,3,\;\alpha\in\R\;|\;a_i^{-\alpha}a_{i+1}^{\beta}a_i^{\alpha}=
a_{i+1}^{\beta\exp(\alpha)}\rangle.$$ It is not hard to show that $\langle a_0^\alpha,a_1^\alpha,a_2^\alpha,a_3^\alpha\rangle<H_\R$ is residually finite for all $\alpha\not\in Y$ for a countable set $Y\subset\R$.
Proof Theorem \[th\_algebra\]
=============================
The algebra ${\mathbb Z}_{p^n}[\bar x]/I$ may be constructed as amalgamated free products, similar to the construction of the Higman group itself.
Let $A$, $B$ be ${\mathbb Z}_{p^n}$-algebras, $\cA\ni 1$, $\cB\ni 1$ their corresponding free generators as ${\mathbb Z}_{p^n}$-modules. A free product $A*B$ is a ${\mathbb Z}_{p^n}$-algebra that
- Generated by $1$ and alternating words of letters from $\cA'=\cA\setminus \{1\}$ and $\cB'=\cB\setminus\{1\}$ as a free ${\mathbb Z}_{p^n}$-module. A word $w=w_1w_2\dots w_r$ is alternating if $(w_i,w_{i+1})\in \cA'\times\cB'\cup\cB'\times\cA'$.
- It suffices to define product for two alternating words $w=w_1\dots w_r$ and $u=u_1\dots u_k$. So, $wu=w_1\dots w_ru_1\dots u_k$ if $(w_r,u_1)$ alternating and $wu=w_1\dots w_{r-1}(w_r\cdot u_1)u_2\dots u_r$ otherwise. Here $w_r\cdot u_1$ is the product in $A$ or $B$.
For example, ${\mathbb Z}_{p^n}[x]*{\mathbb Z}_{p^n}[x]={\mathbb Z}_{p^n}[x_0,x_1]$.
[**Exercise 3.**]{} Check that $A*B$ is well defined, independent of $\cA$ and $\cB$, contains isomorphic copies of $A$ and $B$ with intersection equals to ${\mathbb Z}_{p^n}\cdot 1$ and satisfies the universal property of free product of algebras.
Let $C_1=A\bowtie V$ and $C_2=B\bowtie V$. We define $C=C_1\mathop{*}\limits_{V}C_2$ as $C=(A*B)\bowtie V$. Let $\cA$, $\cB$ and $\cV$ be free bases of $A$, $B$, and $V$, correspondingly. Any element is a ${\mathbb Z}_{p^n}$-combination of $wv$, where $w$ is alternating word of letters from $\cA'$, $\cB'$ and $v\in\cV$. In order to define multiplication on $C$ it suffices to define $vw$ as a combination of $w_iv_i$. It could be done using Zappa-Zsep product structure of $C_1$ and $C_2$. Suppose, for example that $w=a_1b_1\dots a_kb_k$ then $$va_1b_1\dots a_kb_k=\sum_{i_1}a^1_{i_1}v_{i_1}b_1\dots a_kb_k=
\sum_{i_1,j_1}a^1_{i_1}b^1_{i_1,j_1}v_{i_1,j_1}\dots a_kb_k=$$ $$\sum_{i_1,j_1,\dots,i_k,j_k}a^1_{i_1}b^1_{i_1,j_1}\dots a^k_{i_1,\dots j_{k-1},i_k}b^k_{i_1,\dots,j_k}
v_{i_1,\dots,j_k}$$
[**Exercise 4.**]{} Check that $C_1\mathop{*}\limits_VC_2$ is well defined, contains isomorphic copies of $C_1$ and $C_2$ such that $C_1\cap C_2=V$. Check that $C_1\mathop{*}\limits_VC_2$ satisfies the universal property of amalgamated free products of algebras.
Consider $A_0={\mathbb Z}_{p^n}[x_0,x_1]/I(g_0)$, where $I(g_0)$ is the ideal generated by $g_0$ of Eq.(\[eq\_g\]).
\[prop\_A0\] $A_0={\mathbb Z}_{p^n}[x_0]\bowtie{\mathbb Z}_{p^n}[x_1]$
We prove the proposition in Section \[sub\_propA0\]. Now let as show how Theorem \[th\_algebra\] follows from Proposition \[prop\_A0\]. Let $A_1={\mathbb Z}[x_1,x_2]/I(g_1)$. Clearly, $A_1$ is isomorphic to $A_0$, so $A_1={\mathbb Z}_{p^n}[x_1]\bowtie{\mathbb Z}_{p^n}[x_2]$. Consider $A_{01}=A_0\mathop{*}_{{\mathbb Z}_{p^n}[x_1]} A_1$. Notice, that the roles of ${\mathbb Z}_{p^n}[x_1]$ in $A_0$ and $A_1$ are different, precisely the isomorphism from $A_0$ to $A_1$ maps ${\mathbb Z}_{p^n}[x_1]$ to ${\mathbb Z}_{p^n}[x_2]$ not to ${\mathbb Z}_{p^n}[x_1]$ of $A_1$.
\[lm\_A01\] $A_{01}={\mathbb Z}_{p^n}[x_0,x_2]\bowtie{\mathbb Z}_{p^n}[x_1]$ is isomorphic to ${\mathbb Z}_{p^n}[x_0,x_1,x_2]/I(g_0,g_1)$.
By construction $g_0=g_1=0$ in $A_{01}$. So, the map $x_i\to x_i$ prolongs to a surjective homomorphism $\phi:{\mathbb Z}_{p^n}[x_0,x_1,x_2]/I(g_0,g_1)\to A_{01}$. Using the universal property of $A_{01}$ define $\psi:A_{01}\to {\mathbb Z}_{p^n}[x_0,x_1,x_2]/I(g_0,g_1)$. Notice, that $\psi$ is surjective as well (${\mathbb Z}_{p^n}[x_0,x_1,x_2]/I(g_0,g_1)$ is generated by $x_i$). Check that $\psi \phi=id:A_{01}\to A_{01}$ by the universal property.
Similarly, the algebras $A_2={\mathbb Z}_{p^n}[x_2,x_3]/I(g_2)$, $A_3={\mathbb Z}_{p^n}[x_3,x_0]/I(g_3)$, and $A_{23}=A_2\mathop{*}\limits_{{\mathbb Z}_{p^n}[x_3]}A_3={\mathbb Z}[x_2,x_0]\bowtie{\mathbb Z}[x_3]$ may be constructed. Notice, that there exist isomorphism $A_{01}\to A_{23}$ that sends $x_0\to x_2$, $x_2\to x_0$, $x_1\to x_3$. Now, we may construct $A=A_{01}\mathop{*}_{{\mathbb Z}_{p^n}[x_0,x_2]} A_{23}$ (we make isomorphism ${\mathbb Z}[x_0,x_2]$ of $A_{01}$ with ${\mathbb Z}[x_2,x_0]$ of $A_{23}$ sending $x_0\to x_0$ and $x_2\to x_2$). Theorem \[th\_algebra\] follows from the lemma.
$A={\mathbb Z}_{p^n}[x_0,x_2]\bowtie{\mathbb Z}_{p^n}[x_1,x_3]$ is isomorphic to ${\mathbb Z}_{p^n}[x_0,x_1,x_2,x_3]/I(g_0,\dots,g_3)$.
Proof of Proposition \[prop\_A0\] {#sub_propA0}
=================================
The proof is based on the fact that a polynomial $g_0$ forms a kind of a Grobner basis (over non-commutative polynomials). We will not define what a Grobner basis is for non-commutative polynomials. Instead, we directly apply a Knuth-Bendix algorithm [@Baader] to $\{g_0\}$.
Let $m(y_1,\dots,y_k)$ be a non-commutative monomial or, the same, a word in alphabet $\{y_1,\dots,y_k\}$, that is, $m(y_1,\dots,y_k)=z_1z_2\dots z_r$, $z_i\in\{y_1,\dots,y_k\}$. The product of two monomials is just the concatenation. A non-commutative polynomial $f$ over ${\mathbb Z}_{p^n}$ is a “linear” combination of monomials $f=\sum a_im_i$, $a_i\in{\mathbb Z}_{p^n}$. The product $f_1f_2$ of polynomials is defined using the product of monomials by linearity.
Let us return to the study of $A_0={\mathbb Z}_{p^n}[x_0,x_1]/I(g_0)$. We call a polynomial $f\in {\mathbb Z}_{p^n}[x_0,x_1]$ left (resp. right) reduced if $f=\sum a_{i,j}x_0^ix_1^j$ (resp. $f=\sum a'_{i,j}x_1^ix_0^j$). Proposition \[prop\_A0\] is equivalent to the claim.
\[cl\_1\] For any $f\in {\mathbb Z}_{p^n}[x_0,x_1]$ there exists a left (resp. right) reduced polynomial $\tilde f$ such that $f-\tilde f\in I$. If $f\in I$ is left (right) reduced then $f=0$.
From this point we restrict ourselves to the left case. The right case may be considered similarly. Define the one step reduction based on equalities $g_0=0$: $$x_1x_0\sredr x_0x_{1}-Q_0(x_{1})-pQ_1(x_0,x_{1})$$
Let $f,\tilde f\in {\mathbb Z}_{p^n}[x_0,x_1]$.
- $\tilde f$ is a one step reduction of $f$ ($f\sredr\tilde f$) if an appearance of $x_1x_0$, in some monomial of $f$ is changed according to the above described rule; $\tilde f$ is the resulting polynomial (after applying associativity and linearity).
- $f$ is said to be terminal if no monomial of $f$ contains $x_1x_0$. It means that we are unable to apply $\sredr$ to $f$.
- We write $f\redr\tilde f$ ($\tilde f$ is a reduction of $f$) if there is a sequence $f_0=f,f_1,\dots,f_k=\tilde f$ such that $f_i\sredr f_{i+1}$ for $i=0,\dots,k-1$.
- We write $f\redr\tilde f+$ if $f\redr\tilde f$ and $\tilde f$ is terminal.
\[prop\_reduction\]
- $f$ is terminal if and only if $f$ is left reduced.
- There is no infinite sequence $f_0\sredr f_1\sredr f_2\dots$
- For any non-terminal $f$ there exist a unique $\tilde f$ such that $f\redr\tilde f+$.
Two last items mean that any sequence $f\sredr f_1\sredr\dots$ of one step reduction terminates and the terminal polynomial depends only on $f$.
We prove the proposition in Subsection \[sub\_prop\]. Now let us show how Proposition \[prop\_reduction\] implies Claim \[cl\_1\]. We use notation $f=\sredr f'$ (resp. $f=\redr f'$) to denote $f\sredr f'$ or $f=f'$ (resp. $f\redr f'$ or $f=f'$).
\[lm\_lin\] The map $f=\redr\tilde f+$ is linear, that is, if $f_1=\redr\tilde f_1+$ and $f_2=\redr\tilde f_2+$ then $a_1f_1+a_2f_2=\redr a_1\tilde f_1+a_2\tilde f_2$, where $a_1,a_2\in{\mathbb Z}_{p^n}$.
Let $a_1f_1+a_2f_2\sredr f'$. It means that we apply reduction to a monomial $m$ of $a_1f_1+a_2f_2$. The monomial $m$ may appear in $f_1$, $f_2$, or in the both polynomials. In any case there exist $f'_1$ and $f'_2$ such that $f_1=\sredr f_1'$, $f_2=\sredr f'_2$ and $f'=a_1f'_1+a_2 f'_2$. It follows by induction that if $a_1f_1+a_2f_2\redr f'$ then $f'=a_1f'_1+a_2f'_2$ for some $f'_1, f'_2$ such that $f_1=\redr f'_1$ and $f_2=\redr f'_2$. Now suppose that $a_1f'_1+a_2f'_2$ is terminal but, say, $f'_1$ is not terminal. There are two possibilities:
1. $a_1f'_1$ is terminal. In this case, collecting terminal monomials, we may write $f'_1=\alpha+\beta$ with $\alpha$ terminal and $a_1\beta=0$. In this case further reduction of $f'_1$ does not change $a_1f'_1$. So, w.l.g. we may assume $f'_1$ to be terminal.
2. $a_1f'_1$ is not terminal. In this case we may write $f'_i=\alpha_i+\beta_i$ with $\alpha_i$ terminal and $a_1\beta_1+a_2\beta_2=0$. Now, apply the same reduction to $\beta_1$ and $\beta_2$, keeping the sum $a_1f'_1+a_2f'_2$ unchanged.
We are done by Proposition \[prop\_reduction\].
\[lm\_ideal\] $f\redr 0+$ if and only if $f\in I(g_0,\dots,g_3)$.
[**Only if.**]{} It is clear by construction that $f\redr\tilde f$ implies that $f-\tilde f\in I$.\
[**If.**]{} Let $f\in I$. It means that $f=\sum \alpha_ig_0\beta_i$. Applying associativity we may write $f=\sum a_im_ig_0m'_i$, where $m_i$ and $m'_i$ are monomials and $a_i\in{\mathbb Z}_{p^n}$. By construction, there is a one step reduction $m_ig_0m'_i\sredr 0$. We are done by Lemma \[lm\_lin\] and Propostion \[prop\_reduction\].
Proof of Proposition \[prop\_reduction\]. {#sub_prop}
-----------------------------------------
The first item of Proposition \[prop\_reduction\] is straightforward. So we start with the proof of the second item of the proposition. It is the most difficult part of the proposition and will be used for the proof of the third item.
### Proof of the second item.
Let $t=a_jm(x_0,x_1)$ be a term (a monomial with a coefficient). We are going to measure how an application of one step reduction makes a term more close to a left reduced polynomial. To this end we define:
- $|t|=\min\{k\in{\mathbb N}\;|\;p^kt=0\}$.
- $n_0(t)$ – number of $x_0$ in $t$, for example, $n_0(x_1^ix_0^j)=j$.
- $def(t)$ – the defect of $t$, the total number of pairs where $x_1$ appears before $x_0$, for example, $def(x_1^jx_0^kx_1^rx_0^m)=jk+jm+rm$.
To each term we associate the ordered triple $(|t|,n_0(t),def(t))$. On the set of triple we consider lexicographical order: $(\alpha,\beta,\gamma)<(\alpha',\beta',\gamma')$ iff
- $\alpha<\alpha'$; or
- $\alpha=\alpha'$ and $\beta<\beta'$; or
- $\alpha=\alpha'$, $\beta=\beta'$, and $\gamma<\gamma'$.
Now one may check that $(|t|,n_0(t),def(t))>(|t_j|,n_0(t_j),def(t_j))$ if $t\redr\sum_j t_j$. We need the following result.
Any decreasing (with respect to lexicographic order) sequences in ${\mathbb N}^3$ is finite.
Consider now the reduction process of $t$ as a tree: To each vertex we associate a term in such a way that in any reduction step the resulting polynomial is a sum of terms of the leafs of the tree. With root we associate $t$. For each reduction of term $t'$ in a leaf $l$ we connect the leaf $l$ with new leafs with all terms appearing in the reduction. Any descending path in this tree is finite by Dickson Lemma. This tree is $k$-regular by construction, so the tree is finite and the reduction process terminates.
### Proof of the third item.
This uses the Newman’s lemma, or Diamond lemma for reduction processes.
Suppose that on a set $X$ a reduction process $\cdot\sred{*}\cdot$ (just a relation on $X$) is defined. Denote by $\red{*}$ it’s transitive closure. We say that $x$ is terminal if there are no $y\in X$ such that $x\red{*}y$ As before, let $x\red{*}y+$ denotes $x\red{*}y$ and $y$ is terminal.
Let $\sred{*}$ satisfies the following properties:
- Any sequence $x_1\sred{*}x_2\sred{*}...$ is finite.
- $\sred{*}$ is locally confluent, that is, for any $x$, $y_1$ and $y_2$ such that $x\sred{*} y_1$ and $x\sred{*} y_2$ there exists $z\in X$ such that $y_1\red{*}z$ and $y_2\red{*} z+$.
Then $\red{*}$ is globally confluent, that is, for any non-terminal $x$ there exists unique $y$ such that $x\red{*} y$.
So, in order to show Proposition \[prop\_A0\] it suffices to check the second condition of the Diamond Lemma for $\sredr$. Let $f\sredr f_1$ and $f\sredr f_2$. If the reduction applies to a different terms then existence of $f_3$, $f_1\sredr f_3$ and $f_2\sredr f_3$ is trivial. It suffices to consider $f=ax_{i_1}...x_{i_m}$. Suppose w.l.g., that $f\sredr f_1$ is an application of reduction to $x_{i_j}x_{i_{j+1}}=x_1x_0$ and $f\sredr f_2$ $x_{i_k}x_{i_{k+1}}=x_1x_0$ for $k>j+1$. Then $f_1=ax_{i_1}\dots x_{i_{j-1}}qx_{i_{j+2}}\dots x_{i_m}$ and $f_2=ax_{i_1}\dots x_{i_{k-1}}qx_{i_{k+2}}\dots x_{i_m}$, where $q= x_0x_{1}-Q_0(x_{1})-pQ_1(x_0,x_{1})$. One may check that $f_1\redr f_3$ and $f_2\redr f_3$ for $f_3=ax_{i_1}\dots x_{i_{j-1}}qx_{i_{j+2}}\dots x_{i_{k-1}}qx_{i_{k+2}}\dots x_{i_m}$.
Proof of Lemma \[lm\_free\] {#sec_lm_free}
===========================
Let ${\mathbb Z}(\bar x)$ be an algebra of power series with noncomutative (but associative) variables $\bar x=x_0,x_1,\dots,x_m$ over ${\mathbb Z}$. For $a,b\in{\mathbb Z}(\bar x)$ let $\lceil a,b\rceil =ab-ba$ and $\Lambda[\bar x]$ be a submodule of ${\mathbb Z}[\bar x]$ generated by $\lceil \cdot,\cdot\rceil$ starting from $\bar x$. Let $\Lambda^j \subset\Lambda[\bar x]$ consist of uniform polynomials of order $j$. So, $$\Lambda[\bar x]=\cup_{j=0}^\infty \Lambda^j.$$ Let $I=I(\bar x)$ be a (two-sided) ideal in ${\mathbb Z}(\bar x)$ generated by $\bar x$. Clearly, this ideal consists of polynomials without constant term. Let $G$ be a group. Notations $G_n$ and $G_{[n]}$ are defined in Section \[sec\_p-quotient\].
\[th\_Magnus\]
- Let $a_i=1+x_i$. The group $F=\langle a_i\rangle$ is a free group, freely generated by $a_i$.
- $(1+I^n)\cap F=F_{[n]}$.
- If $w \in F_{[n]}$ then $w=1+d+z$, where $d\in\Lambda^n$ and $z$ does not contain terms of order $\leq$ $n$.
- For any $d\in\Lambda^n$ there exists $z\in{\mathbb Z}[\bar x]$ without terms of order $\leq$ $n$ such that $1+d+z\in F_{[n]}$.
Consider homomorphism $\pi:{\mathbb Z}(\bar x)\to {\mathbb Z}(\bar x)$, defined by $\pi(x_i)=px_i$. Clearly, $\pi({\mathbb Z}(\bar x))={\mathbb Z}(p\bar x)$. Also, $\pi(F)$ is an inclusion of a free group $F$ into ${\mathbb Z}(p\bar x)$. For a two sided ideal $J$ of ${\mathbb Z}(p\bar x)$ let $N_j=\{w\in\pi(F)\;|\;w-1\in J\}$.
\[lm\_ideal\_normal\] $N_j\triangleleft\pi(F)$.
Clearly, ${\mathbb Z}(\bar x)/p^n{\mathbb Z}(\bar x)\equiv {\mathbb Z}_{p^n}(\bar x)$. Denote $(p^n)=p^n{\mathbb Z}(\bar x)\cap {\mathbb Z}(\bar px)$. Notice, that ${\mathbb Z}_{p^n}(p\bar x)={\mathbb Z}_{p^n}[p\bar x]$. Now, Lemma \[lm\_free\] is a consequence of the following theorem.
\[th\_Jacobson\] $N_{(p^n)}=\pi(F_n)$
We present here the Jacobson proof (see [@Jacobson]) which is a reduction to Theorem \[th\_Magnus\]. In [@Jacobson] the definition of $(p^n)$ is different and not equivalent of ours. But the proof in [@Jacobson] is, actually, for our definition of $(p^n)$.
Notice, that $(1+p^m w\dots)^p=1+p^{m+1}\dots$ and $[(1+p^mw\dots),(1+p^ku\dots)]=1+p^{m+k}(wu-uw)\dots$ where the omitted terms are of higher $p$-order. This implies that $N_{(p^m)}^p\subseteq N_{(p^{m+1})}$ and $[N_{(p^m)},N_{(p^k)}]\subseteq N_{(p^{m+1})}$. Consequently, we have $\pi(F_n)\subseteq N_{(p^n)}$ and $N_{(p^{n+1})}\subseteq N_{(p^n)}$.
According to [@Jacobson] we show the equality $N_{(p^n)}=\pi(F_n)$ by induction. By definition, $N_{(p^1)}=\pi(F_1)=\pi(F)$. Suppose, that $N_{(p^n)}=\pi(F_n)$. Then we know that $\pi(F_{n+1})\subseteq N_{(p^{n+1})}\subseteq \pi(F_n)$. So, it suffices to show that $\pi(F_n)\cap\pi(F_{n+1})\supseteq N_{(p^{n+1})}\cap\pi(F_n)$, or, the same, to prove that if $w\in\pi(F_n)\setminus\pi(F_{n+1})$ then $w\not\in N_{(p^{n+1})}$. Let $w\in\pi(F_n)\setminus\pi(F_{n+1})$. There exists a unique $i$ such that $w\in \pi(F_{[i]})\setminus\pi(F_{[i+1]})$. By Theorem \[th\_Magnus\] $w=1+p^jd_i(p\bar x)+z$ where $d_i\in \Lambda^i$, $z$ has $\bar x$-order more than $i$. Also we have that $i+j\geq n$ (as $w\in N_{(p^n)}$). Applying once again Theorem \[th\_Magnus\] we find $u=1+d_i(p\bar x)+z'\in\pi(F_{[i]})$, where the $\bar x$-order of $z'$ is more than $i$. So, $w=u^{p^j}w_1$, where $w_1\in \pi(F_{[i']})$ for $i'>i$. Repeating this procedure one gets $w=u_1^{p^{j_1}}u_2^{p^{j_2}}\dots u_k^{p^{j_k}}w_k$, where $w_k\in \pi(F_{[n+1]})$ and $u_r=(1+d_{i_r}(p\bar x)\dots)$ with $d_{i_r}\in\Lambda^{i_r}$ (term of higher order in $\bar x$ are omitted) and $i_r+j_r\geq n$. Now, $u_r\in \pi(F_{[i_r]})$ and, consequently, $u_r^{p^{j_r}}\in \pi(F_{[i_r]}^{p^{j_r}}\subseteq F_{i_r+j_r})$. If $\forall r\;\;i_r+j_r>n$ then $w\in \pi(F_{n+1})$, so, by our assumptions, $i_r+j_r=n$ for some $r$. It implies that $w=u_1^{p^{j_1}}\dots u_k^{p^{j_k}}w_k=
(1+p^{j_1}d_{i_1}(p\bar x)+p^{j_2}d_{i_2}(p\bar x)\dots)\not\in N_{(p^{n+1})}$.
Let $F=\langle a_0,\dots,a_m\rangle$ be a free group on $\{a_0,\dots,a_m\}$. Let ${\mathbb Z}_{p^n}[F/F_n]$ be a group algebra of $F/F_n$ over ${\mathbb Z}_{p^n}$. Theorem \[th\_Jacobson\] implies that there exists unique homomorphism $\phi:{\mathbb Z}_{p^n}[F/F_n]\to {\mathbb Z}_{p^n}[p\bar x]$ such that $\phi(a_i)=1+px_i$. (Here we, abusing notation, denote by the same symbol $a_i$ its image in $F/F_n$.) Moreover, $(\ker(\phi)+1)\cap F/F_n=\{1\}$. Still $\ker(\phi)$ is not trivial. For example, if $w\in F/F_n$ and $w^{p^j}=1$ then $p^j(w-1)\in\ker(\phi)$. What is the structure of $\ker(\phi)$? For example, is it true that $\ker(\phi)$ is generated by $\{p^j(w-1)\;|\;w\in F/F_n,\;w^{p^j}=1\}$?
[99]{} Franz Baader, Tobias Nipkov. , Cambridge University Press 1998. xii+301 pp. ISBN: 0-521-45520-0; 0-521-77920-0 .
Lev Glebsky and Igor E. Shparlinski. Short cycles in repeated exponentiation modulo a prime. 56(1):35–42, 2010.
Gupta, N. Lectures on Dimension Subgroups. , 1996, Vol.2, No.3, 263-273
Harald A. Helfgott and Kate Juschenko. Soficity, short cycles and the Higman group. arXiv:1512.02135
Graham Higman. A finitely generated infinite simple group. 26:61–64, 1951.
Jacobson, N.(1-YALE) Magnus’ method in the theory of free groups. 1 (1992), no. 1,
Magnus, Wilhelm; Karrass, Abraham; Solitar, Donald Dover Publications, Inc., Mineola, NY, 2004. xii+444 pp. ISBN: 0-486-43830-9 Reprint of the 1976 second edition.
Alexandre Martin, On the cubical geometry of Higman’s group. arXiv:1506.02837
Schupp, Paul E. Small cancellation theory over free products with amalgamation. 193 (1971), 255–264
[^1]: This work was partially supported by the European Research Council (ERC) grant no. 259527 of G. Arzhantseva, part of this work was done in the Nizhny Nivgorod University and supported by the RSF (Russia) grant 14-41-00044. The stay in Vienna was supported by ERC grant no. 259527 of G. Arzhantseva.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Consequences of the basic and most evident consistency requirement—that measured events cannot happen and not happen at the same time—are shortly reviewed. Particular emphasis is given to event forecast and event control. As a consequence, particular, very general bounds on the forecast and control of events within the known laws of physics result. These bounds are of a global, statistical nature and need not affect singular events or groups of events.'
author:
- |
Karl Svozil\
[Institut für Theoretische Physik,]{} [Technische Universität Wien ]{}\
[Wiedner Hauptstraße 8-10/136,]{} [A-1040 Vienna, Austria ]{}\
[e-mail: [email protected]]{}
date:
title: |
Forecast and event control\
On what is and what cannot be possible\
Part I: Classical case
---
### Principle of self-consistency {#principle-of-self-consistency .unnumbered}
An irreducible, atomic physical phenomenon manifests itself as a click of some detector. There can either be a click or there can be no click. This yes-no scheme is experimental physics in-a-nutshell (at least according to a theoretician). From that kind of elementary observation, all of our physical evidence is accumulated.
Such irreversibly observed events (whatever the relevance or meaning of those terms are [@wigner:mb; @wheeler; @greenberger2; @hkwz]) are subject to the primary condition of [*consistency*]{} or [*self-consistency*]{}: [*“Any particular irreversibly observed event either happens or does not happen, but it cannot both happen and not happen.”*]{}
Indeed, so trivial seems the requirement of consistency that David Hilbert polemicised against “another author” with the following words [@hilbert-26], “...for me, the opinion that the \[\[physical\]\] facts and events themselves can be contradictory is a good example of thoughtlessness.”
Just as in mathematics, inconsistency, i.e., the coexistence of truth and falseness of propositions, is a fatal property of any physical theory. Nevertheless, in a certain very precise sense, quantum mechanics incorporates inconsistencies in a very subtle way, which assures overall consistency. For instance, a particle wave function or quantum state is said to “pass” a double slit through both slits at once, which is classically impossible. (Such considerations may, however, be considered as mere trickery quantum talk, devoid of any operational meaning.) Yet, neither a particle wave function nor quantum states are directly associable with any sort of irreversible observed event of physical reality. We shall come back to a particular quantum case in the second part of this investigation.
And just as in mathematics and in formal logic it can be argued that too strong capacities of intrinsic event forecast and intrinsic event control renders the system overall inconsistent. This fact may indeed be considered as one decisive feature in finite deterministic (“algorithmic”) models [@svozil-93]. It manifests itself already in the early stages of Cantorian set theory: any claim that it is possible to enumerate the real numbers yields, via the diagonalization method, to an outright contradiction. The only consistent alternative is the acceptance that no such capacity of enumeration exists. Gödel’s incompleteness theorem [@godel1] states that any formal system rich enough to include arithmetic and elementary logic could not be both consistent and complete. Turing’s theorem on the recursive unsolvability of the halting problem [@turing-36], as well as Chaitin’s $\Omega$ numbers [@chaitin:92] are formalizations of related limitations in formal logics, the computer sciences and mathematics.
In what follows we shall proceed along very similar lines. We shall first argue that any capacity of total forecast or event control—even in a totally deterministic environment—is contradicting the (idealistic) idea that decisions between alternatives are possible; stated differently: that there is free will. Then we shall proceed with possibilities of forecast and event control which are consistent both with free will and the known laws of physics.
It is also clear that some form of forecast and event control is evidently possible—indeed, that is one of the main achievements of contemporary natural science, and we make everyday use of it, say, by switching on the light. These capacities derived from the standard natural sciences are characterized by a high chance of reproducibility, and therefore do not depend on single events.
In what follows, we shall concentrate on very general bounds of capacities of forecast and event control; bounds which are imposed upon them by the requirement of consistency. These considerations should be fairly general and do not depend on any particular physical model. They are valid for all conceivable forms of physical theories; classical, quantum and forthcoming alike.
### Strong forecasting {#strong-forecasting .unnumbered}
Let us consider forecasting the future first. Even if physical phenomena occur deterministically and can be accounted for (“computed”) on a higher level of abstraction, from within the system such a complete description may not be of much practical, operational use [@toffoli:79; @svozil-unev].
Indeed, suppose there exists free will. Suppose further that an agent could predict [*all*]{} future events, without exceptions. We shall call this the [*strong form of forecasting.*]{} In this case, the agent could freely decide to counteract in such a way as to invalidate that prediction. Hence, in order to avoid inconsistencies and paradoxes, either free will has to be abandoned, or it has to be accepted that complete prediction is impossible.
Another possibility would be to consider strong forms of forecasting which are, however, not utilized to alter the system. Effectively, this results in the abandonment of free will, amounting to an extrinsic, detached viewpoint. After all, what is knowledge and what is it good for if it cannot be applied and made to use?
It should be mentioned that the above argument is of an ancient type [@martin]. As has already been mentioned, it has been formalized recently in set theory, formal logic and recursive function theory, where it is called “diagonalization method.”
In doing this, we are inspired by the recent advances in the foundations of quantum (information) theory. There, due to complementarity and the impossibility to clone generic states, single events may have important meanings to some observers, although they make no sense at all to other observers. One example for this is quantum cryptography. Many of these events are stochastic and are postulated to satisfy all conceivable statistical laws (correlations are nonclassical, though). In such frameworks, high degrees of reproducibility cannot be guaranteed, although single events may carry valuable information, which can even be distilled and purified.
### Strong event control {#strong-event-control .unnumbered}
A very similar argument holds for event control and the production of “miracles” [@frank]. Suppose there exists free will. Suppose further that an agent could entirely control the future. We shall call this the [*strong form of event control.*]{} Then this observer could freely decide to invalidate the laws of physics. In order to avoid a paradox, either free will or some physical laws would have to be abandoned, or it has to be accepted that complete event control is impossible.
### Weak forecast and event control {#weak-forecast-and-event-control .unnumbered}
From what has already been said, it should be clear that it is reasonable to assume that [*forecast and event control should be possible only if this capacity cannot be associated with any paradox or contradiction.*]{}
Thus the requirement of consistency of the phenomena seems to impose rather stringent conditions on forecasting and event control. Similar ideas have already been discussed in the context of time paradoxes in relativity theory (cf. [@friedetal] and [@nahin p. 272], [*“The only solutions to the laws of physics that can occur locally $\ldots$ are those which are globally self-consistent”*]{}).
There is, however, a possibility that the forecast and control of future events [*is*]{} conceivable for [*singular*]{} events within the statistical bounds. Such occurrences may be “singular miracles” which are well accountable within the known laws of physics. They will be called [*weak forms of forecasting and event control.*]{}
It may be argued that, in order to obey overall consistency, such a framework should not be extendable to any forms of strong forecast or event control, because, as has been argued before, this could either violate global consistency criteria or would make necessary a revision of the known laws of physics.
The relevant laws of statistics (e.g., all recursively enumerable ones) impose rather lax constraints especially on finite sequences and do not exclude local, singular, improbable events. For example, a binary sequence such as $11111111111111111111111111111111$ is just as probable as the sequences $11100101110101000111000011010101$ and $01010101010101010101010101010101$ and its occurrence in a test is equally likely, although the “meaning” an observer could ascribe to it is rather different. These sequences may be embedded in and be part of much longer stochastic sequences. If short finite regular (or “meaningful”) sequences are padded into long irregular (“meaningless”) ones, those sequences become statistically indistinguishable for all practical purposes from the previous sequences. Of course, the “meaning” of any such sequence may vary with different observers. Some of them may be able to decipher a sequence, others may not be capable of this capacity.
It is quite evident that per definition any finite regularity in an otherwise stochastic environment should exclude the type of high reproducability which one has gotten used to in the natural sciences. Just on the contrary: single “meaningful” events which are hardly reproducible might indicate a new category of phenomena which is dual to the usual “lawful” and highly predictable ones.
Just as it is perfectly all right to consider the statement “This statement is true” to be true, it may be perfectly reasonable to speculate that certain events are forecasted and controlled within the domain of statistical laws. But in order to be within the statistical laws, any such method [*needs not to be guaranteed*]{} to work at all times.
To put it pointedly: it may be perfectly reasonable to become rich, say, by singular forecasts of the stock and future values or in horse races, but such an ability must necessarily be not extendible, irreproducible and secretive; at least to such an extend that no guarantee of an overall strategy and regularity can be derived from it.
The associated weak forms of forecasting and event control are thus beyond any global statistical significance. Their importance and meaning seems to lie mainly on a very subjective level of singular events. This comes close to one aspect of what Jung imagined as the principle of “synchronicity” [@jung1], and is dual to the more reproducible forms one is usually accustomed to.
### Against the odds {#against-the-odds .unnumbered}
This final paragraph reviews a couple of experiments which suggest themselves in the context of weak forecast and event control. All are based on the observation whether or not an agent is capable to forecast or control correctly future events such as, say, the tossing of a fair coin.
In the first run of the experiment, no consequence is derived from the agent’s capacity despite the mere recording of the data.
The second run of the experiment is like the first run, but the [*meaning*]{} of the forecasts or controlled events are different. They are taken as outcomes of, say gambling, against other individuals (i) with or (ii) without similar capacities, or against (iii) an anonymous “mechanic” agent such as a casino or a stock exchange.
As a variant of this experiment, the partners or adversaries of the agent are informed about the agent’s intentions.
In the third run of experiments, the experimenter attempts to counteract the agent’s capacity. Let us assume the experimenter has total control over the event. If the agent predicts or attempts to bring about to happen a certain future event, the experimenter causes the event not to happen and so on.
It might be interesting to record just how much the agent’s capacity is changed by the setup. Such an expectation might be defined from a dichotomic observable $$e(A,i) =\left\{
\begin{array}{cl}
+1&\qquad {\rm correct\; guess}\\
-1&\qquad {\rm incorrect\; guess}\\
\end{array}
\right.$$ where $i$ stands for the $i$’th experiment and $A$ stands for the agent $A$. An expectation function can then be defined as usual by the average over $N$ experiments; i.e., $$E(A) = {1\over N}\sum_{i=1}^N e(A,i).$$
From the first to the second type of experiment it should become more and more unlikely that the agent operates correctly, since his performance is leveled against other agents with more or less the same capacities. The third type of experiment should produce a total anticorrelation. Formally, this should result in a decrease of $E$ when compared to the first round of experiment.
Another, rather subtle, deviation from the probabilistic laws may be observed if [*correlated*]{} events are considered. Just as in the case of quantum entanglement, it may happen that individual components of correlated systems may behave totally at random exhibit more disorder than the system as a whole [@nielsen-kem-2001].
If once again one assumes two dichotomic observables $e(A,i),e(B,i)$ of a correlated subsystem, then the correlation function $$C(A,B) = {1\over N}\sum_{i=1}^N e(A,i)e(B,i)$$ and the associated probabilities may give rise to violations of the Boole-Bell inequalities—Boole’s [*“conditions of possible \[\[classical\]\] experience”*]{} [@Boole-62; @Hailperin; @pitowsky; @Pit-94] and may even exceed [@svozil-krenn] the Tsirelson bounds [@cirelson:80; @cirelson:87; @cirelson] for “conditions of possible \[\[quantum\]\] experience.” There, the agent should concentrate on influencing the [*coincidences*]{} of the event rather than the single individual events. In such a case, the [*individual*]{} observables may behave perfectly random, while the associated [*correlations*]{} might be nonclassical and even stronger-than-quantum and might give rise to highly nonlocal phenomena. As long as the individual events cannot be controlled, this needs not even violate Einstein causality. (But even then, consistent scenarios remain [@svozil-relrel].)
In summary it can be stated that, although total forecasting and event control are incompatible with free will, more subtle forms of these capacities remain conceivable even beyond the present laws of physics; at least as long as their effects upon the “fabric of phenomena” are consistent. These capacities are characterized by singular events and not on the reproducible patterns which are often encountered under the known laws of physics. Whether or not such capacities exist at all remains an open question. Nevertheless, despite the elusiveness of the phenomenology involved, it appears not unreasonable that the hypothesis might be testable, operationalizable and even put to use in certain contexts.
[10]{}
Eugene P. Wigner. Remarks on the mind-body question. In I. J. Good, editor, [*The Scientist Speculates*]{}, pages 284–302. Heinemann and Basic Books, London and New York, 1961. Reprinted in [@wheeler-Zurek:83 pp. 168-181].
John A. Wheeler. Law without law. In John A. Wheeler and W. H. Zurek, editors, [*Quantum Theory and Measurement*]{}, pages 182–213. Princeton University Press, Princeton, 1983. [@wheeler-Zurek:83].
Daniel B. Greenberger and A. YaSin. “[H]{}aunted” measurements in quantum theory. , 19(6):679–704, 1989.
Thomas J. Herzog, Paul G. Kwiat, Harald Weinfurter, and Anton Zeilinger. Complementarity and the quantum eraser. , 75(17):3034–3037, 1995.
David Hilbert. ber das [U]{}nendliche. , 95:161–190, 1926.
Karl Svozil. . World Scientific, Singapore, 1993.
Kurt G[ö]{}del. ber formal unentscheidbare [Sä]{}tze der [P]{}rincipia [M]{}athematica und verwandter [S]{}ysteme. , 38:173–198, 1931. English translation in [@godel-ges1], and in [@davis].
A. M. Turing. On computable numbers, with an application to the [E]{}ntscheidungsproblem. , 42 and 43:230–265 and 544–546, 1936-7 and 1937. Reprinted in [@davis].
Gregory J. Chaitin. . World Scientific, Singapore, 1992.
T. Toffoli. The role of the observer in uniform systems. In G. Klir, editor, [*Applied General Systems Research*]{}. Plenum Press, New York, London, 1978.
Karl Svozil. Undecidability everywhere? In [*Boundaries and Barriers. On the Limits to Scientific Knowledge*]{}, pages 215–237. Addison-Wesley, Reading, MA, 1996.
A. R. Anderson. St. [P]{}aul’s epistle to [T]{}itus. In R. L. Martin, editor, [*The Paradox of the Liar*]{}. Yale University Press, New Haven, 1970. The Bible contains a passage which refers to Epimenides, a Crete living in the capital city of Cnossus: [*“One of themselves, a prophet of their own, said, ‘Cretans are always liars, evil beasts, lazy gluttons.’ ”*]{},— St. Paul, Epistle to Titus I (12-13).
Philip Frank. . Springer, Vienna, 1932.
John Friedman, Michael S. Morris, Igor D. Novikov, Fernando Echeverria, Gunnar Klinkhammer, Kip S. Thorne, and Ulvi Yurtsever. Cauchy problem in spacetimes with closed timelike curves. , D42(6):1915–1930, 1990.
Paul J. Nahin. . AIP Press and Springer, New York, 1998.
Carl Gustav Jung. Synchronizit[ä]{}t als ein [P]{}rinzip akausaler [Z]{}usammenh[ä]{}nge. In Carl Gustav Jung and Wolfgang Pauli, editors, [ *Naturerkl[ä]{}rung und Psyche*]{}. Rascher, Z[ü]{}rich, 1952.
M. A. Nielsen and J. Kempe. Separable states are more disordered globally than locally. , 86(22):5184–5187, 2001.
George Boole. On the theory of probabilities. , 152:225–252, 1862.
Theodore Hailperin. . North-Holland, Amsterdam, 1976.
Itamar Pitowsky. . Springer, Berlin, 1989.
Itamar Pitowsky. eorge [B]{}oole’s ‘conditions od possible experience’ and the quantum puzzle. , 45:95–125, 1994.
G. Krenn and K.Svozil. Stronger-than-quantum correlations. , 28(6):971–984, 1998.
B. S. Cirel’son. Quantum generalizations of [B]{}ellos inequality. , 4:93–100, 1980.
B. S. Tsirel’son. Qantum analogues of the [B]{}ell inequalities. [T]{}he case of two spatially separated domains. , 36(4).
B. S. [Cirel’son (=Tsirelson)]{}. Some results and problems on quantum [B]{}ell-type inequalities. , 8:329–345, 1993.
Karl Svozil. Relativizing relativity. , 30(7):1001–1016, 2000.
John Archibald Wheeler and Wojciech Hubert Zurek. . Princeton University Press, Princeton, 1983.
Kurt G[ö]{}del. In S. Feferman, J. W. Dawson, S. C. Kleene, G. H. Moore, R. M. Solovay, and J. van Heijenoort, editors, [*Collected Works. Publications 1929-1936. Volume [I]{}*]{}. Oxford University Press, Oxford, 1986.
M. Davis. . Raven Press, New York, 1965.
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- 'Rudy Romain, Daniel Hennequin and Philippe Verkerk'
title: 'Phase-space description of the magneto-optical trap'
---
Introduction
============
The magneto-optical trap (MOT) is the primary tool to cool atoms. The development of this technique led to spectacular breakthroughs in experimental quantum physics: MOTs are the first step in the realization of optical lattices [@lattices], cold molecules [@molecules], or Bose-Einstein condensates [@BEC]. But the MOT is also an interesting object *per se*. It produces a cloud of cold atoms, the physics of which is complex. In particular, spatio-temporal instabilities of this cloud are commonly observed [@wilkowski2000; @labeyrie2006]. Several models with different approaches have been proposed to describe these dynamics, and to identify the mechanisms leading to instabilities. Unfortunately, none of these models gives a satisfying description of the observed dynamics. In [@wilkowski2000], a very simple model allowed to describe experimentally observed instabilities. This model, improved in [@hennequin2004], predicted another type of instabilities, which were effectively observed in [@distefano2003; @distefano2004]. However, the agreement between the model and the experiments was only qualitative. Moreover, it concerned the particular case where the counterpropagating trap beams are obtained by retro-reflection, i.e. a global asymmetry is introduced in the trap.
Recently, it has been proposed to describe the symmetric MOT, with all trap beams which are independent, as a weakly damped plasma [@labeyrie2006; @pohl2006]. Indeed, the cloud of cold atoms in a MOT is a confined dilute object with long-range interactions, as in plasmas [@walker1990; @pruvost2000]. This model predicted the existence of instabilities above a relatively high threshold, so that instabilities should exist only in large MOTs. This seems in contradiction with the observations related in [@hennequin2004]. Moreover, no direct comparison with the experimental temporal regimes allowed validating this model. More recently, a more complete description was derived using the methods of waves and oscillations in plasmas, leading to interesting predictions [@mendonca2008]. But as in the previous case, this study is based upon intermediate well-established results, valid only in specific cases (e.g. a low beam intensity or a negligible viscosity). These conditions do not correspond in general to the experimental situations, and indeed, these results were not compared to experimental results.
It appears from these numerous works that a reference model for MOT atom clouds lacks. Such a model should be as general as possible, and should at least describe the usual experimental situations. In particular, such approximations as the low saturation limit should not be done *a priori*. Another interest of such a model is that it could help in determining precisely the analogies between MOT atom clouds and other systems, such as plasmas.
The present work is a first step towards such a model. Its aim is to build a model, with the least possible hypotheses and approximations, of a 1D symmetric MOT. The resulting set of equations describes as precisely as possible the dynamics of atoms inside a 1D MOT, and constitutes a basis model. If simplifications are necessary, approximations should be applied on this model, *a posteriori.*
The paper is organized as follows: section 2 defines the bases of the model, while in section 3, we show that the dynamics of the MOT phase space density can be described by a Vlasov-Fokker-Planck equation with two relaxation processes. In section 4, we derive the different terms of this equation of evolution as a function of the usual experimental parameters. In section 5, we establish the equations of propagation of the trapping beams. We obtain a set of coupled equations fully describing our system, in the general case. This set of equation can be solved numerically to obtain solutions of the general system. However, to have a better understanding of the MOT, it is interesting to go further in the analytical approach, simplifying *a posteriori* the model, as discussed above. This is done in section 6, where we examine an approximation for the atomic response.
Definition of the model
=======================
As discussed above, we consider here a 1D configuration. Two counterpropagating laser beams with opposite circular polarizations interact with the atoms, as sketched on Figure \[fig:Sketch\]a. The beam with the $\sigma^{-}$ polarization comes from the negative abscissa, and is denoted by the minus sign (intensity $I_{-}(x,t)$, wave vector $k_{-}$). In the same way, the beam with the $\sigma^{+}$ polarization comes from the positive abscissa (intensity $I_{+}(x,t)$, wave vector $k_{+}$). Forces originate from the exchange of momentum between the atoms and the electromagnetic field. We consider here that the atoms are the simplest ones for which the magneto-optical trapping is possible. The laser frequency $\omega_{L}$ is tuned in the vicinity of a $J=0\rightarrow J'=1$ transition with a frequency $\omega_{0}$ (Fig. \[fig:Sketch\]b).
![\[fig:Sketch\]Sketch of the 1D system considered here. a) the two contrapropagating beams have opposite circular polarizations. b) The laser beams interact with “three-level” atoms on a $J=0\rightarrow J'=1$ transition, the degeneracy of which is lifted by an inhomogeneous magnetic field. $\Omega_{\pm}$ are the Rabi frequencies associated with the beams; $\Delta_{\pm}$ are the effective detunings (see section 4.1).](fig1){width="7.5cm"}
In the 2D phase space, a point has the coordinates $\left(x,p\right)$, where $x$ is the position and $p$ is the momentum. To describe the cloud of cold atoms, we introduce the phase space density $\rho\left(x,p,t\right)$. Formally, the complete atomic system is described by a phase space density including both the hot and the cold atoms. However, laser cooling acts only in a limited region of space, typically the intersection of the laser beams, and for moderate velocities. As a consequence, the surrounding hot atoms are considered as a large reservoir which remain in a thermal equilibrium at room temperature. In the following, we neglect the action of the cold atoms on the hot background. These approximations lead to a huge simplification in the description of the collisions. Thus, to describe the cold atom dynamics, we have just to derive the equation of evolution of $\rho\left(x,p,t\right)$ from the basic principles of atomic physics. This approach allows us to go beyond the Wieman description of multiple scattering in terms of absorption cross sections [@walker1990; @sesko1991]. $\rho\left(x,p,t\right)$ influences the propagation of the trapping beams, and reciprocally the modification in the beam intensities changes the evolution of the phase space density. Thus, we expect to obtain a system of coupled nonlinear differential equations. As our aim is to build a theoretical frame in which most of the experimental situations can be explored, we limit as much as possible the initial approximations. In particular, we do not restrict our study to the low saturation limit, as in [@mendonca2008].
To derive the system of coupled equations, we proceed in two steps. We first assume that we know the intensities of the beams everywhere in the sample, and we derive the equation of evolution of the density in phase space (section 3 et 4). Then, we write the equations of propagation of the beams assuming that we know the atomic density in phase space (section 5).
Evolution of the phase space density
====================================
To derive the equation of evolution of the phase space density $\rho\left(x,p,t\right)$, we consider an elementary cell centered in $\left(x_{0},p_{0}\right)$, with dimensions $\delta x$ and $\delta p$. The number of atoms contained in this cell is $N(t)=\rho\left(x_{0},p_{0},t\right)\delta x\,\delta p$, where we assume that $\delta x$ and $\delta p$ have been chosen small enough to neglect higher order corrections. The variations of $N$ between $t$ and $t+\delta t$ are governed by three distinct phenomena: (i) an atom is kicked in or out by a collision with the hot background gas or (ii) an atom crosses the border of the cell because of the evolution of either its position or (iii) its momentum. The simultaneous change in position and in momentum would lead to second order terms in $\delta t$, which are neglected.
The collisional processes lead to two terms, one for losses and one for gains : $$\delta N_{coll}=-\frac{N\,\delta t}{\tau}+\lambda\,\delta x\,\delta p\,\delta t\label{eq:dNcoll}$$ where $\tau$ is the mean time interval between two collisions with the hot atoms. The source term $\lambda$ is due to the collisions between two hot atoms that tend to restore the Maxwell-Boltzmann distribution for the phase space density (uniform in $x$, gaussian in $p$). A contribution from the collisions between cold atoms could also be considered but, in the conditions where instabilities are observed in a MOT, this last contribution can be neglected.
The second mechanism is the variation of $N$ due to the velocity of the atoms. As depicted in Fig. \[fig:Cell\]a, the atoms that cross the borders in $x_{0}\pm\delta x/2$ with a velocity $p_{0}/m$, where $m$ is the atomic mass, between $t$ and $t+\delta t$ are those in the close vicinity of that border. The area of these zones is simply $\delta p\, p_{0}\,\delta t/m$. The variation $\delta N_{x}$ in $N$ due to the velocity of the atoms can then be written:$$\begin{aligned}
\delta N_{x} & = & \frac{\delta p\, p_{0}\,\delta t}{m}\left[\rho\left(x_{0}-\frac{\delta x}{2},p_{0},t\right)-\rho\left(x_{0}+\frac{\delta x}{2},p_{0},t\right)\right]\nonumber \\
& = & -\left(\delta x\,\delta p\,\delta t\right)\,\frac{p_{0}}{m}\,\frac{\partial}{\partial x}\rho\left(x_{0},p_{0},t\right)\label{eq:dNx}\end{aligned}$$
![\[fig:Cell\]Modification of the number of atoms contained in the elementary cell due to the evolution of (a) the position and (b) the momentum. The dashed regions contribute to losses, while dotted regions contribute to gains. In (a), all the atoms in the indicated regions will cross a boundary (we assume that $p_{0}>0$), and the areas are proportional to $\delta t$. In (b), a small fraction of the atoms, proportional to $\delta t$, will receive a momentum kick. We have to consider separately the momentum kicks with $q>0$ and those with $q<0$.](fig2){width="8cm"}
The last mechanism that changes $N$ is due to the changes in atomic momentum, which is more tricky to evaluate. The atoms undergo cycles where one photon is absorbed and another one is emitted. We consider that all the underlying physics is described by a probability per unit of time $\mathcal{P}\left(x,p,t,q\right)$: the probability to change the atomic momentum by a quantity $q$ between $t$ and $t+\delta t$ is simply $\delta t\,\mathcal{P}\left(x,p,t,q\right)$. We assume that the time interval $\delta t$ is small enough so that an atom can undergo at most one photon scattering event. In such a case, $q=\hbar\left(k_{a}-k_{s}\right)$ where $k_{a}$ and $k_{s}$ are respectively the wavevector of the absorbed and emitted photon. As a consequence, $q$ is bounded to the range $\left[-2\hbar k,2\hbar k\right]$. A fraction, proportional to $\delta t$, of the atoms contained in this region will change their momentum by $q$. On the side $p_{+}=p_{0}+\delta p/2$, the atom number variation $\delta N_{p+}$ is the difference between the incoming atoms ($q<0)$ and the lost atoms ($q>0$): $$\begin{aligned}
\delta N_{p+} & = & \left(\delta x\,\delta t\right)\int_{0}^{2\hbar k}dq\int_{0}^{q}dq'\left(\mathcal{N}_{+}-\mathcal{N}_{-}\right)\label{eq:dNp1}\\
\mathrm{with\quad}\mathcal{N}_{\pm} & = & \mathcal{P}\left(x,p_{+}\pm q',t,\mp q\right)\,\rho\left(x,p_{+}\pm q',t\right)\nonumber \end{aligned}$$ To simplify this expression, we assume that the product $\mathcal{P}\left(x,p,t,q\right)\,\rho\left(x,p,t\right)$ varies slowly with $p$. Then, the integrand can be approximated by its Taylor expansion in the vicinity of . To recover the momentum diffusion process responsible for the non-zero temperature of the trapped atoms, we have to expand the integrand up to first order in $q'$. The inner integration is then straightforward because the integrand becomes a linear function of $q'$. The same calculation has to be done on the opposite side in $p_{-}=p_{0}-\delta p/2$, which gives an analogous expression for $\delta N_{p-}$.
It appears natural to introduce the mean values of $q$ and $q^{2}$: $$\begin{aligned}
F\left(x,p,t\right) & = & \int_{-2\hbar k}^{2\hbar k}\, q\, dq\,\mathcal{P}\left(x,p,t,q\right)\label{eq:Force}\\
D\left(x,p,t\right) & = & \int_{-2\hbar k}^{2\hbar k}\,\frac{q}{2}^{2}\, dq\,\mathcal{P}\left(x,p,t,q\right)\label{eq:Diffusion}\end{aligned}$$ These quantities can be interpreted as the mean force and the momentum diffusion coefficient. The variation $\delta N_{p}=\delta N_{p+}+\delta N_{p-}$ in $N$ due to the change in momentum can be written as:$$\delta N_{p}=-\left(\delta x\,\delta p\,\delta t\right)\left[\frac{\partial}{\partial p}\left(F\rho\right)-\frac{\partial^{2}}{\partial p^{2}}\left(D\rho\right)\right]\label{eq:dNp}$$ Collecting all the contributions (\[eq:dNcoll\]), (\[eq:dNx\]) and (\[eq:dNp\]) and going to the limit $\delta t\rightarrow0$, we obtain the equation of evolution for the atomic density$$\frac{\partial}{\partial t}\rho+\frac{p}{m}\,\frac{\partial}{\partial x}\rho+\frac{\partial}{\partial p}\left(F\rho\right)-\frac{\partial^{2}}{\partial p^{2}}\left(D\rho\right)=-\frac{\rho}{\tau}+\lambda\label{eq:drhodt}$$
The three left terms of this equation are characteristic of a Vlasov type kinetic model, except for the third term, as here the force depends on the velocity of the particles. Thus this last term is a drift term, and together with the fourth term, which is a relaxation term, they denote a Fokker-Planck description. So the motion of cold atoms in a MOT appears to be described by a Vlasov-Fokker-Planck (VFP) equation. Thus MOTs are part of a large class of systems described by the VFP equations, as e.g. plasmas [@plasma], stars [@star], granular media [@benedetto2004] or electrons in a storage ring [@warnock2006]. The different systems are characterized by the dependence of the force $F$ on the phase space density $\rho$. For example, for a plasma without magnetic fields, the Vlasov-Poisson-Fokker-Planck is used.
In these systems, the Fokker-Planck terms denote the immersion of the particles in a thermal bath, which corresponds for the cold atoms to the laser light. But in the case of cold atoms, a second bath, namely the hot atoms of the residual gas, produces a second relaxation term, as well as a source term (the right hand terms of Eq. \[eq:drhodt\]): indeed, the standard MOT is an open system, where the total population can vary. This last point is crucial and cannot be neglected, as the collision processes between hot atoms are the one and only source of velocity redistribution allowing for a high MOT population.
Because of the two relaxation processes involved in the MOTs, the question of the mechanism in which originate the instabilities is far from being trivial. Indeed, it is well known that the VFP equation may have, for adequate parameters, unstable solutions. But as in [@distefano2003], instabilities lead to variations of population, it is clear that the loss and source terms can also generate unstable solutions. Therefore, instabilities may originate in two different mechanisms, and it would be interesting to search if one of them prevails. An interesting case is that of the lower MOT in a double cell system. In this case, the gas pressure in the cell is so low that the collision losses are negligible. If the loading process by the upper MOT is stopped, Eq. becomes fully equivalent to plasmas equations, and only one relaxation process remains. Thus it would be interesting to study experimentally such a system, as the fact that instabilities persist or not in this case could help in understanding the origin of these instabilities.
To go further, we need to evaluate the force and the diffusion coefficient for a specific atom. To do so, we have to calculate the probability $\mathcal{P}\left(x,p,t,q\right)$ associated with the momentum exchange $q=\hbar\left(k_{a}-k_{s}\right)$. However, as the probability to emit a spontaneous photon in the direction $k_{s}$ is always identical to that for the opposite direction $-k_{s}$, the contribution of spontaneous emission cancels in $F$, while in $D$ the cross-term in $q^{2}$ vanishes. As a consequence, it is not necessary to evaluate completely $\mathcal{P}\left(x,p,t,q\right)$, but only the absorption rate for a photon of wavevector $k_{a}.$
\[sub:Radiative-force-and\]Radiative forces and coefficients of diffusion
=========================================================================
As stated above, we consider the simple scheme of a laser frequency $\omega_{L}$ tuned in the vicinity of a $J=0\rightarrow J'=1$ transition with a frequency $\omega_{0}$. These atoms are not suited for sub-Doppler cooling mechanisms [@dalibard1989], but these subtle processes are not needed to describe the temperatures measured experimentally in the cold atomic cloud obtained with high intensities and small detunings for the trapping beams [@distefano2003]. In the case of retroreflected trapping beams, this simple atom allows to develop a model which convincingly reproduces the experimental observations of instabilities, both deterministic [@distefano2004] and stochastic [@hennequin2004].
As usual in problems of laser cooling, we will extensively use the fact that the time scales for the different processes are quite different. The first time scale is given by the time a photon needs to go through the sample. This time scale is so short that we can consider that the light follows immediately any change in the sample. A second time scale is given by the atomic response time $\Gamma^{-1}$, which is in turn shorter than the third one related to the evolution of the external degrees of freedom (the atomic position and velocity). This inequality allows us to consider that the atomic internal state has always reached its steady state.
In the following, we consider an atom in $X$ with a velocity $V$, where we use capital letters to distinguish the position and velocity of this “real atom” from the coordinates $\left(x,p\right)$ of a point in the phase space. The atoms are excited on a $J=0\rightarrow J'=1$ transition by two counterpropagating laser beams with opposite circular polarizations. They also see a “bath” of photons which are spontaneously emitted by all the other atoms in the sample. These photons propagate in all directions of the real $3D$ space and they have a broad spectrum. Although it is not always true, we consider here that the effect of the bath of spontaneous photons can be treated as a perturbation. Then, we split the force and the diffusion coefficient into two parts: one due to the trapping beams and the second one due to multiple scattering: $F=F_{t}+F_{m}$ and $D=D_{t}+D_{m}$.
Effect of the trapping beams
----------------------------
The trap consists in two counterpropagating beams with the same frequency $\omega_{L}$ and opposite polarizations (Fig. \[fig:Sketch\]a). The $\sigma^{-}$ beam has a wavevector $k_{-}=k\equiv\omega_{L}/c$ and has an intensity $I_{-}\left(X,t\right)$, while the $\sigma^{+}$ beam has a wavevector $k_{+}=-k$ and an intensity $I_{+}\left(X,t\right)$. The frequency $\omega_{L}$ is slightly lower than the frequency $\omega_{0}$ of the atomic transition to insure an efficient Doppler cooling. The sample is submitted to an inhomogeneous magnetic field aligned on the propagation axis of the beams. For the sake of simplicity, we assume that the magnetic field varies linearly: $B_{x}=bX$. The Zeeman effect together with the cooling beams lead to a restoring force that gathers the atoms in the vicinity of $X=0$.
In the atomic rest frame, the apparent frequencies of the trapping beams are Doppler shifted: $\omega_{\pm}=\omega_{L}\pm kV$. As the two beams have opposite circular polarizations, the relevant detunings are $\Delta_{\pm}=\Delta\pm\delta$, where we have introduced the detuning $\Delta=\omega_{L}-\omega_{0}$ for an atom at rest with no magnetic field, and the sum $\delta=kV+\mu_{B}bX/\hbar$ of the Doppler shift $kV$ and the Zeeman shift $\mu_{B}B_{x}/\hbar$. As the two beams are circularly polarized, the excited state $\left|0\right\rangle $ plays no role in the present configuration. The coupling between the ground state $\left|g\right\rangle $ and the excited states $\left|\pm\right\rangle $ is described by the Rabi frequencies $\left|\Omega_{\pm}\right|=\Gamma\sqrt{I_{\pm}/2I_{s}}$, where $I_{s}$ is the saturation intensity. It is then possible to find exactly the steady state of the density matrix, by solving the master equation. We are not interested here in the explicit form of the populations or of the coherences, which are given in a previous article [@hennequin2004]. The only important point is that the stationary populations $\Pi_{\pm}$ in the excited states $\left|\pm\right\rangle $ can be written as a function of the relevant parameters $\Delta$, $\delta$, $I_{+}$ and $I_{-}$. Note that these stationary populations do not depend on the position or the velocity of the atom explicitly, but only through $\delta$, $I_{+}$ and $I_{-}$.
The key point is that the simplicity of the atomic structure means that $\gamma_{\pm}=\Gamma\Pi_{\pm}$ are the emission rates of spontaneous $\sigma^{\pm}$ polarized photons. The exact expressions of $\Pi_{\pm}$ can be found in [@hennequin2004]. Again because of the simple $J=0\rightarrow J'=1$ transition, a photon with the same polarization has to be absorbed before each spontaneous emission process. As a consequence, we obtain a very simple expression for the radiative force due to the trapping beams, together with the expression for the diffusion coefficient due to the trapping beams:
\[eq:ftdt\]$$\begin{aligned}
F_{t} & = & \hbar k\left(\gamma_{-}-\gamma_{+}\right)\label{eq:ft}\\
D_{t} & = & \frac{7}{5}\hbar^{2}k^{2}\left(\gamma_{-}+\gamma_{+}\right)\label{eq:dt}\end{aligned}$$
where a contribution $2k^{2}/5$ in $D_{t}$ comes from the mean value of the square of the projected component of the spontaneous wavevector. To evaluate this average, we consider the diagram of emission of a circular electric dipole ($3/10$ in the plane and $2/5$ in the orthogonal direction).
Effect of multiple scattering
-----------------------------
The contribution of multiple scattering is trickier to evaluate. We have to consider our atom in a bath of incoherent photons which propagate in all directions. Because of their physical origin, photons do not differ in spectrum wherever the scatterer is and whatever their direction of propagation is. Therefore, the rate of absorption of photons going in a given direction is simply proportional to the flux of photons that travel in this direction. The contribution to that flux of photons scattered just once is easy to evaluate. The flux of photons scattered more than once is more difficult to estimate, but it is in general much smaller than the previous one, and we neglect it in the following. The flux $\Psi_{+}$(resp. $\Psi_{-}$) of photons, travelling to the right (resp. to the left) and impinging on an atom in $X$, is just half of the total number of photons scattered by the atoms on the left (resp. on the right) of $X$.
\[eq:Psi\] $$\begin{aligned}
\Psi_{+}\left(X,t\right) & = & \frac{1}{2}\int_{-\infty}^{X}dx'\int dp\,\rho\left(x',p,t\right)\,\left(\gamma_{-}+\gamma_{+}\right)\label{eq:Psi+}\\
\Psi_{-}\left(X,t\right) & = & \frac{1}{2}\int_{X}^{+\infty}dx'\int dp\,\rho\left(x',p,t\right)\,\left(\gamma_{-}+\gamma_{+}\right)\label{eq:Psi-}\end{aligned}$$
In these expressions, we do not differentiate the polarization of the scattered photons. We just sum up the contribution $\gamma_{+}$ of the mainly $\sigma^{+}$ polarized photons and the contribution $\gamma_{-}$ of the mainly $\sigma^{-}$ polarized photons. It is interesting to note that the sum $\Psi_{+}+\Psi_{-}$ of these two fluxes does not depend on $X$. We shall see in the next section that these fluxes have much simpler expressions.
To evaluate the force exerted by the scattered photons on the atoms, we have to know the fraction of photons which are absorbed and the momentum carried by each photon. This momentum is smaller than $\hbar k$, because we have to keep here only the on-axis component for photons travelling in all directions of the real $3D$ space. Considering the emission diagram of a circular dipole, we obtain $\left\langle k_{a}\right\rangle =9k/16$ and $\left\langle k_{a}^{2}\right\rangle =2k^{2}/5$. This leads to the following expressions for $F_{m}$ and $D_{m}$:
\[eq:fmdm\]$$\begin{aligned}
F_{m} & = & \sigma_{R}\,\frac{9}{16}\,\hbar k\left(\Psi_{+}-\Psi_{-}\right)\\
D_{m} & = & \sigma_{R}\,\frac{4}{5}\,\hbar^{2}k^{2}\left(\Psi_{-}+\Psi_{+}\right)\end{aligned}$$
where we have introduced re-absorption cross-section $\sigma_{R}$.
In the limit of very low intensities ($\Omega\ll\Gamma$), we can give an estimate of $\sigma_{R}$. In this case, it is well-known that for a 2-level atom, the emission spectrum is mainly elastic [@cohen]. This result holds for the $J=0\rightarrow J'=1$ transition considered here and the scattered photons have the same frequency $\omega_{L}$ as the trapping lasers. To be rigorous, Doppler broadening should be taken into account, as the scatters are moving and the emitted photons do not propagate in the direction of the trapping beams. However, we neglect this broadening here, because it is smaller than both the detuning and the natural linewidth. We also neglect the Zeeman shift and the Doppler shift for the absorber. The absorption cross-section is thus reduced by a factor $\Gamma^{2}/\left(4\Delta^{2}+\Gamma^{2}\right)$ with respect to the cross-section at resonance $\sigma_{0}=3\lambda^{2}/2\pi$ . In the very low intensity limit, we have:$$\begin{aligned}
\sigma_{R} & =\sigma_{0} & \frac{\Gamma^{2}}{4\Delta^{2}+\Gamma^{2}}\end{aligned}$$
For higher but still modest intensities, a new contribution shows up : the blue sideband of the Mollow triplet [@mollow], which excites resonantly the atomic transition, grows up as $\Omega^{2}/\left(4\Delta^{2}+\Gamma^{2}\right)$. As soon as $\Omega\gg\Gamma$, this new resonant contribution dominates, even in the low saturation regime ($1\gg\Omega^{2}/\left(4\Delta^{2}+\Gamma^{2}\right)$).
For even higher intensities, as we know the exact steady state of the density matrix, we can, in principle, compute the spectrum of the scattered light and the absorption cross-section, following what is usually done for 2-level atoms [@pruvost2000]. However, contrary to the previous publications [@pruvost2000; @sesko1991], we do not evaluate $\sigma_{R}$ for a 2-level atom, to remain consistent with the model of a $J=0\rightarrow J'=1$ transition where at least one of the excited state sub-levels does not interact with the trapping beams. As the exact calculation is quite heavy, we shall not make here the full calculation in the general case, but we shall restrict ourselves to the simpler situation considered in section \[sec:Approximations\].
Propagation of the trapping beams\[sub:Propagation\]
====================================================
In this section, we derive the equation of propagation for the trapping beams assuming that we know the atomic density $\rho\left(x,p,t\right)$. We are not interested in the phase of the laser beams, but just in their intensities. We have seen in the previous section that the rates $\gamma_{\pm}=\Gamma\Pi_{\pm}$ are the absorption rates of $\sigma^{\pm}$ polarized photons for an atom in $X$ moving with velocity $V$. To know how many photons are absorbed between $x$ and $x+\delta x$, we just have to sum the contributions of all atoms in $X=x$ with all possible velocities $V=p/m$. These photons are taken from the trapping beam with the appropriate polarization, so we get :
\[eq:dl\]$$\begin{aligned}
\frac{\partial I_{+}}{\partial x} & = & +\hbar\omega_{L}\int_{-\infty}^{+\infty}dp\,\rho\left(x,p,t\right)\,\gamma_{+}\label{eq:dI+}\\
\frac{\partial I_{-}}{\partial x} & = & -\hbar\omega_{L}\int_{-\infty}^{+\infty}dp\,\rho\left(x,p,t\right)\,\gamma_{-}\label{eq:dI-}\end{aligned}$$
The plus sign in (\[eq:dI+\]) comes from the fact that the $\sigma^{+}$ polarized beam propagates backwards. So $I_{+}\left(x,t\right)$ increases with $x$, while $I_{-}\left(x,t\right)$ decreases.
Plugging (\[eq:dl\]) in the fluxes (\[eq:Psi\]) and performing formally the integration on positions, we get :
$$\begin{aligned}
\Psi_{+}\left(x,t\right) & = & \frac{I_{+}\left(x,t\right)-I_{-}\left(x,t\right)-I_{+}\left(-\infty,t\right)+I_{-}\left(-\infty,t\right)}{2\hbar\omega_{L}}\\
\Psi_{-}\left(x,t\right) & = & \frac{I_{+}\left(\infty,t\right)-I_{-}\left(\infty,t\right)-I_{+}\left(x,t\right)+I_{-}\left(x,t\right)}{2\hbar\omega_{L}}\end{aligned}$$
The intensities $I_{-}\left(-\infty,t\right)$ and $I_{+}\left(\infty,t\right)$ are the incoming laser intensities, which are assumed to be constant. As soon as we know the atomic phase space density $\rho\left(x,p,t\right)$, we can solve the equations to get the spatial evolution of the intensities. However, the evolution of the phase space density depends on the laser intensities through the radiative forces and the diffusion coefficients. Thus, the equations and form a set of coupled equations which have to be solved together.
It is important to note that Eqs introduce in the force a term which, in general, is not present in other systems described by the VFP equation. This difference has to be taken into account if we want to apply results obtained for plasmas to the dynamics of MOTs. Indeed, in plasmas, the particles do not act, by definition, on the thermal bath. On the contrary, in MOTs, atoms act on the beams (the equivalent of the thermal bath, as seen above), through the absorption and the so-called shadow effect. Thus it leads to another indirect dependence of the force on the density.
Approximation for the atomic response\[sec:Approximations\]
===========================================================
In the general case, the equations and describing the MOT are highly non linear, and have no simple solution. Numerical simulations could give some solutions, but they could not give any physical insight of the phenomenon. However, starting from this general model, we can now consider some approximations allowing going further in the understanding of the MOT dynamics. Through these approximations, we aim to determine the expressions of the trapping beam contribution and of the multiple scattering contribution to the mean force and diffusion coefficient of the MOT.
Trapping forces
---------------
Concerning the trapping beam components, Eqs shows that we need essentially to explicit the scattering rates $\gamma_{\pm}=\Gamma\,\Pi_{\pm}$, introduced as the emission rates of spontaneous $\sigma^{\pm}$ polarized photons, and also as the absorption rates of the corresponding trapping photons. In the case of a $J=0\rightarrow J'=1$ transition, these rates are exactly computable, as in [@hennequin2004]. However, the obtained expressions are very heavy and difficult to interpret. To have a better physical insight, we will simplify these expressions through approximations valid in our domain of interest.
The usual approach consists in considering the low saturation limit. In this case, each wave acts on the atoms independently, and the forces and the diffusion coefficient can be computed. But experiments are seldom realized in this low intensity limit. For slightly higher intensities, the calculation of the next order of perturbation (fourth order in field) must be considered. But this calculation is not straightforward. Indeed, the third order of perturbation builds up Zeeman coherence between the excited sublevels, which lead to non-trivial modifications of the populations at the fourth order. The calculation can be done, but the cross-terms do not allow any simple interpretation. Moreover, to evaluate the effect of multiple scattering, we need to know the absorption cross section for the bath of photons that have been scattered elsewhere. It requires one more step of perturbation by an extra weak field, which is not easy to implement.
This first approximation is usually followed by another approximation, namely the low velocity limit, where the total trapping force is linearized in $V$. It naturally splits in a term due to the intensity imbalance and in a friction term. In the same way, the gradient of the magnetic field introduces a restoring force, linear in $X$. The condition of validity for this approximation is $\left|\delta\right|\ll max\left(\left|\Delta\right|,\Gamma\right)$, where $\delta$ is the sum of the Doppler and Zeeman shifts, as introduced in section \[sub:Radiative-force-and\].
Let us consider here another approach, where the approximation $\delta$ is applied before going to the low saturation limit. The total shift $\delta$ is considered as a perturbation, that is expanded to first order only. At a first glance, this choice seems quite surprising, as the “perturbation” is diagonal in the natural basis $\left\{ \left|g\right\rangle ,\left|+\right\rangle ,\left|-\right\rangle \right\} $. However, the good basis to do the calculation is not the natural one, and then the shift $\delta$ leads to off-diagonal terms.
At zeroth order, the case $\delta=0$ corresponds to an atom interacting with a field which has both $\sigma^{\pm}$ components. It is thus natural to introduce the coupled $\left|C\right\rangle $ and non-coupled $\left|N\right\rangle $ states: $$\begin{aligned}
\left|C\right\rangle & = & \frac{\Omega_{+}\left|+\right\rangle +\,\Omega_{-}\left|-\right\rangle }{\Omega}\\
\left|N\right\rangle & = & \frac{\Omega_{-}^{*}\left|+\right\rangle -\,\Omega_{+}^{*}\left|-\right\rangle }{\Omega}\end{aligned}$$ where $\Omega_{\pm}$ are the generalized (C-number) Rabi frequencies that include the complex phase of the fields and $\Omega=\sqrt{\left|\Omega_{+}\right|^{2}+\left|\Omega_{-}\right|^{2}}$ describes the coupling between the ground state and the coupled one. Then the problem reduces to that of a 2-level atom interacting with a single laser field. The steady state population of the coupled state is simply:$$\Pi_{C}^{(0)}=\frac{\Omega^{2}}{4\Delta^{2}+2\Omega^{2}+\Gamma^{2}}$$ which gives the value of the scattering rates :$$\gamma_{\pm}^{(0)}=\Gamma\,\frac{\Omega_{\pm}^{2}}{4\Delta^{2}+2\Omega^{2}+\Gamma^{2}}$$ The radiative force due to the beam imbalance and the coefficient of diffusion follows:$$\begin{aligned}
F_{t}^{(0)} & = & \frac{\hbar\, k\,\Gamma}{4\Delta^{2}+2\Omega^{2}+\Gamma^{2}}\left(\left|\Omega_{-}\right|^{2}-\left|\Omega_{+}\right|^{2}\right)\\
D_{t}^{(0)} & = & \frac{7}{5}\,\hbar^{2}k^{2}\,\Gamma\frac{\Omega^{2}}{4\Delta^{2}+2\Omega^{2}+\Gamma^{2}}\end{aligned}$$ It is interesting to note that, in these expressions, we get the same result if we replace, in the expressions obtained in the low saturation regime, the usual denominator $\left(4\Delta^{2}+\Gamma^{2}\right)$ by $\left(4\Delta^{2}+2\Omega^{2}+\Gamma^{2}\right)$.
At first order, when $\delta\neq0$, we introduce the same coupled and non-coupled states as above, and a first order perturbation in $\delta$ gives the friction and restoring forces. The diagonal (in the natural basis) part of the hamiltonian $H_{A}$ writes: $$\begin{aligned}
H_{A} & = & -\left(\Delta_{+}\left|+\right\rangle \left\langle +\right|+\,\Delta_{-}\left|-\right\rangle \left\langle -\right|\right)\\
& = & -\Delta_{N}\left|N\right\rangle \left\langle N\right|-\,\Delta_{C}\left|C\right\rangle \left\langle C\right|\\
& & -2\frac{\delta}{\Omega^{2}}\left(\Omega_{+}\Omega_{-}\left|N\right\rangle \left\langle C\right|+\Omega_{+}^{*}\Omega_{-}^{*}\left|C\right\rangle \left\langle N\right|\right)\end{aligned}$$ with $$\begin{aligned}
\Delta_{C} & = & \Delta+\delta\frac{\left|\Omega_{+}\right|^{2}-\left|\Omega_{-}\right|^{2}}{\Omega^{2}}\\
\Delta_{N} & = & \Delta-\delta\frac{\left|\Omega_{+}\right|^{2}-\left|\Omega_{-}\right|^{2}}{\Omega^{2}}\end{aligned}$$ The slight change from $\Delta$ to $\Delta_{C}$ introduces a first order correction in the population of the excited coupled state.$$\Pi_{C}^{(1)}=\frac{8\,\Delta\,\delta\left(\left|\Omega_{-}\right|^{2}-\left|\Omega_{+}\right|^{2}\right)}{\left(4\Delta^{2}+2\Omega^{2}+\Gamma^{2}\right)^{2}}\label{eq:pop_C1}$$ The off-diagonal perturbation terms induce, at the first order, coherences between the non-coupled state and the two other states. But only the coherences in the excited state are needed to evaluate the populations $\Pi_{\pm}$$$\begin{aligned}
\left\langle N\left|\sigma^{(1)}\right|C\right\rangle & = & \frac{-2\,\delta\,\Omega_{+}\Omega_{-}}{4\Delta^{2}+2\Omega^{2}+\Gamma^{2}}\,\frac{4\,\Gamma}{4\Gamma\Delta-i\,\left(2\Gamma^{2}+\Omega^{2}\right)}\label{eq:coh_NC1}\\
\left\langle C\left|\sigma^{(1)}\right|N\right\rangle & = & \left\langle N\left|\sigma^{(1)}\right|C\right\rangle ^{*}\label{eq:coh_CN1}\end{aligned}$$ The correction in the scattering rates is deduced from the expressions of the population and of the coherences and :$$\begin{aligned}
\gamma_{\pm} & = & \frac{\mp8\,\Delta\,\delta\,\Gamma}{\left(4\Delta^{2}+2\Omega^{2}+\Gamma^{2}\right)^{2}}\\
& & \times\left(\left|\Omega_{\pm}\right|^{2}+2\left|\Omega_{-}\right|^{2}\left|\Omega_{+}\right|^{2}\frac{4\Gamma^{2}-\Omega^{2}}{16\Gamma^{2}\Delta^{2}+\left(2\Gamma^{2}+\Omega^{2}\right)^{2}}\right)\end{aligned}$$ The evaluation of the friction force is then straightforward: $$\begin{aligned}
F_{t}^{(1)} & = & \frac{\hbar\, k\,\Gamma\,8\,\Delta\,\delta}{\left(4\Delta^{2}+2\Omega^{2}+\Gamma^{2}\right)^{2}}\\
& & \times\left(\Omega^{2}+4\left|\Omega_{-}\right|^{2}\left|\Omega_{+}\right|^{2}\frac{4\Gamma^{2}-\Omega^{2}}{16\Gamma^{2}\Delta^{2}+\left(2\Gamma^{2}+\Omega^{2}\right)^{2}}\right)\\
D_{t}^{(1)} & = & \frac{7}{5}\,\hbar^{2}k^{2}\,\Gamma\frac{8\,\Delta\,\delta\left(\left|\Omega_{-}\right|^{2}-\left|\Omega_{+}\right|^{2}\right)}{\left(4\Delta^{2}+2\Omega^{2}+\Gamma^{2}\right)^{2}}\end{aligned}$$ It is easy to check that the force calculated in this way is rather a friction force, as soon as $\Delta<0$. The first term of the parenthesis, $\Omega^{2}$, together with the prefactor, corresponds to what is guessed from the unsaturated expression: the original denominator $\left(4\Delta^{2}+\Gamma^{2}\right)$ becomes $\left(4\Delta^{2}+2\Omega^{2}+\Gamma^{2}\right)$. However, a second term is needed to take into account properly the cross-saturation effects. This term is often forgotten by the authors, as for instance in [@pohl2006], whereas it can modify significantly the spring constant of the trap, as it is of the same order of magnitude as $\Omega^{2}$. In particular, the minus sign means that, depending on the parameters, the spring constant is increased or decreased significantly. On the contrary, the correction $D_{t}^{(1)}$ to the coefficient of diffusion is usually very small, because $\left|\Omega_{-}\right|^{2}\simeq\left|\Omega_{+}\right|^{2}$.
Multiple scattering
-------------------
To evaluate the effect of multiple scattering, we neglect, as usual, the Doppler broadening due to the motion of the emitter and the Doppler and Zeeman shifts for the re-absorber. Thus, we consider that $\delta$ vanishes. Eqs shows that we have to evaluate the re-absorption cross-section $\sigma_{R}$, taking into account the fluorescence spectrum and the absorption spectrum, which are affected by the intense trapping field.
When the laser field is intense enough, the dressed atom in the secular limit allows a simple evaluation of the fluorescence spectrum and of the re-absorption cross-section [@pruvost2000; @cohen]. However, instead of considering a 2-level atom, we remain consistent with the model of a $J=0\rightarrow J'=1$ transition where we have three sublevels in the excited state. We have also to consider that the polarization of the photon scattered by an atom somewhere in the sample will not match exactly the polarization of the local field. For the sake of simplicity, let us consider here the case where the trapping beams have the same intensity, resulting in a linear polarization for the total trapping field. The consequence of the polarization mismatch is that, on average, one half of the scattered photons has the polarization of the local field while the other half has the orthogonal polarization. The detailed calculation is done in the appendix. The energy levels of the dressed atom are sketched in Fig. \[fig:At\_hab\]. In Fig. \[fig:At\_hab\]a, we have evidenced the fluorescence transitions that lead to the Mollow triplet. On the other hand, the absorption lines are drawn on Fig. \[fig:At\_hab\]b, considering all the possible polarizations for the incident photon. Schematically, we have to consider the overlap of the four components of the Mollow triplet with the four absorption lines of the 3-level atom, as represented in Fig. \[fig:At\_hab\]c. Some care has to be taken to compare the various contributions in the different regimes. Let us now examine the four main situations:
![\[fig:At\_hab\]Sketch of the energy levels of the dressed-atom. a) Spontaneous emission lines leading to the Mollow triplet. b) Absorption lines. c) Overlap of the fluorescence spectrum (solid line) with the absorption lines of the 2-level system (dotted lines) and with those due to the third level (dashed lines).](fig3){width="8cm"}
\(i) $\Omega,\left|\Delta\right|\gg\Gamma$. In general, for large detunings and intensities and large light-shifts ($\delta_{r}=\frac{1}{2}\left(\sqrt{\Delta^{2}+\Omega^{2}}-\left|\Delta\right|\right)\gg\Gamma$), the resonant term of the 2-level system dominates the sum, and all the other terms should be dropped to remain consistent with the secular approximation done before. The resulting re-absorption cross-section has thus the following simple expression:$$\sigma_{R}=\frac{\sigma_{0}}{4}\,\frac{\Delta^{2}\Omega^{2}}{4\Delta^{4}+8\Delta^{2}\Omega^{2}+3\Omega^{4}}\label{eq:eta_resonant}$$ This result is consistent, within a factor of 2, with the limit values given for $\Omega\gg\left|\Delta\right|\gg\Gamma$ and $\left|\Delta\right|\gg\Omega\gg\Gamma$ in the appendix of Ref. [@pruvost2000].
\(ii) $\left|\Delta\right|\gg\Omega$. The contribution due to the non-coupled level is never exactly resonant. However, in this case, the light-shift, $\delta_{r}\simeq\Omega^{2}/4\left|\Delta\right|$, can be arbitrarily small, leading to an almost resonant behaviour if $\delta_{r}\lesssim\Gamma$. In the case where $\left|\Delta\right|\gg\Omega\gg\Gamma$ and $\Gamma\gg\Omega^{2}/4\left|\Delta\right|$, we take into account the quasi resonant contribution of the third level:$$\sigma_{R}=\frac{\sigma_{0}}{8}\,\frac{\Omega^{2}}{\Delta^{2}}\label{eq:eta _quasiresonant}$$ Note that is not the limit of when $\left|\Delta\right|\gg\Omega$, although it has the same form.
\(iii) $\Omega\ll\Gamma$. When the intensity is very small, the resonant or quasi-resonant contributions go to zero and the leading term corresponds to the reabsorption of the elastic component:$$\sigma_{R}=\frac{\sigma_{0}}{4}\,\frac{\Gamma^{2}}{\Delta^{2}}\label{eq:eta_faible_int}$$
\(iv) $\left|\Delta\right|\ll\Gamma$. For small detunings, the reabsorption by a 2-level atom goes to zero because the transition is strongly saturated. Then, the contribution of the third level, although it is non-resonant, has to be taken into account. We have:$$\sigma_{R}=\sigma_{0}\frac{25}{48}\,\frac{\Gamma^{2}}{\Omega^{2}}\label{eq:eta_resonance}$$
The above results allow us to write in the different regimes the contribution of the multiple scattering to the MOT dynamics. These expressions are rather simple as compared to those found in the literature. It is interesting to note that $\sigma_{R}$ is always larger than the absorption cross-section for the trapping beams. This is very often assumed in previous papers, as stated in [@sesko1991]. We have demonstrated here that, in the secular regime for three level atoms, it is always true. The above expressions of $\sigma_{R}$, together with the expression of the trapping force derived in the previous paragraph, allow us to write a full model of the MOT, which can be used as a basis for future analyses.
Conclusion
==========
In this paper, we built a model for a 1D Magneto-Optical Trap as general as possible. We show that such a trap is described by a Vlasov-Fokker-Planck equation with a second relaxation term and a source term, both originating in the bath of hot atoms of the atomic vapor. This VFP equation is coupled to a set of two differential equations describing the beam propagation in the cold atoms. This system could be considered as relatively similar to plasmas, where the role of the thermal bath is played by the trapping beams. However, it appears that the MOT differs from plasmas on two important points: the second thermal bath, formed by the hot atoms, induces new interactions as compared to plasmas; the trapping beams are not a “thermal” bath, as the atoms act on them through the absorption.
In the last section of this paper, we derive in a more detailed way the equations established in the previous sections, and compare the results with those found in the litterature. We show that important correction terms must be taken into account in the evaluation of the trapping forces, and we establish the expression of the multiple scattering in different situations.
These equations aimed to be a basis for future developments. They should contribute to obtain a better agreement between the experimental observations and the theoretical predictions. Let us remember that the previous existing models were only in qualitative agreement with the experimental observations, in particular for the situations out of equilibrium. Using this model to study the instabilities in the cold atoms should give a better insight of the mechanism in which they originate. Thus a natural continuation of the present work would be to extend this model to the 3D traps.
Appendix
========
In this appendix, we present the detailed calculations of the re-absorption cross section, resulting from the overlap of the fluorescence spectrum of an atom somewhere in the sample with the absorption spectrum of another atom elsewhere. In the $J=0\rightarrow J'=1$ transition considered here, we have three sublevels in the excited state, and the polarization of the scattered photon will not match exactly the polarization of the local field. For the sake of simplicity, let us consider here the case where the trapping beams have the same intensity. In real experiments, it cannot be true everywhere in the sample because of the shadow effect, but the deviation remains on the order of 10%. In this simple case, the resulting polarization of the total trapping field is everywhere linear, but its direction follows an helix. The laser wavelength is the scale for changing the relative phase between $\Omega_{+}$ and $\Omega_{-}$, because the two beams are contrapropagating. As soon as we are not interested in what happens at length scales smaller than the laser wavelength, we can estimate that one half of the photons in the bath will interact with the transition between the ground state and the coupled one, while the other half of them can excite the atom to the non-coupled level. The total re-absorption cross-section, $\sigma_{R}$, is thus the average of the usual cross-section, $\sigma_{R}^{(2)}$, calculated with a 2-level atom [@pruvost2000] and of a new contribution, $\sigma_{R}^{(3)}$, coming from the presence of an excited level which is not coupled to the trapping field. The empty non-coupled state allows a strong absorption of the scattered light, while the absorption by the coupled state can be saturated by the trapping field.
As usual, when one works with the dressed atom model, the calculations are much simpler in the secular approximation. This approximation is valid as soon as the energy splitting is larger than the typical width of the levels, which is expressed by the condition: $$\sqrt{\Delta^{2}+\Omega^{2}}\gg\Gamma\label{eq:seculaire}$$ Please note that this relation does not require that both the detuning and the intensity are large, and some interesting limits can be examined with either $\Omega\lesssim\Gamma$ or $\left|\Delta\right|\lesssim\Gamma$, in the secular limit. First, we calculate the emission spectral density, $S\left(\omega\right)$, for an atom illuminated with the two trapping beams. Then, we evaluate the absorption spectra of another atom, also interacting with the trapping field. Two contributions appear in the absorption process: on one hand, the coupled excited level can still absorb light, leading to a contribution which is the one of a 2-level atom, $\sigma_{A}^{(2)}\left(\omega\right)$, and on the other hand, the non-coupled level has to contribute with $\sigma_{A}^{(3)}\left(\omega\right)$. Finally, we evaluate the total re-absorption cross-section, $\sigma_{R}$, with:$$\begin{aligned}
\sigma_{R} & = & \frac{1}{2}\left(\sigma_{R}^{(2)}+\sigma_{R}^{(3)}\right)\label{eq:sigma_R}\\
\sigma_{R}^{(2)} & = & \int\,\sigma_{A}^{(2)}\left(\omega\right)\, S\left(\omega\right)\, d\omega\nonumber \\
\sigma_{R}^{(3)} & = & \int\,\sigma_{A}^{(3)}\left(\omega\right)\, S\left(\omega\right)\, d\omega\nonumber \end{aligned}$$ All these calculations are done using the formalism of the dressed atom, in the secular limit (\[eq:seculaire\]). The normalized fluorescence spectrum is given by:$$\begin{aligned}
S\left(\omega\right) & = & \frac{\left(c^{2}-s^{2}\right)^{2}}{c^{4}+s^{4}}\delta\left(\omega-\omega_{L}\right)+\frac{\left(2c^{2}s^{2}\right)^{2}}{c^{4}+s^{4}}L\left(\omega-\omega_{L},\Gamma_{p}\right)\\
& & +c^{2}s^{2}L\left(\omega-\omega_{L}-\Omega_{G},\Gamma_{c}\right)\\
& & +c^{2}s^{2}L\left(\omega-\omega_{L}+\Omega_{G},\Gamma_{c}\right)\end{aligned}$$ where $\Omega_{G}=\sqrt{\Delta^{2}+\Omega^{2}}$ is the generalized Rabi frequency. The transformation from the bare basis to the dressed one is given by $c=cos\left(\theta\right)$, $s=sin\left(\theta\right)$, with the angle $\theta$ defined by $tg\left(2\theta\right)=-\frac{\Omega}{\Delta}$. $\delta\left(\omega-\omega_{L}\right)$ is the Dirac delta function, and $L\left(\omega,\Gamma\right)$ is the normalized Lorentzian:$$L\left(\omega,\Gamma\right)=\frac{1}{\pi}\,\frac{\Gamma}{\omega^{2}+\Gamma^{2}}$$ and the relaxation rates for the populations and the coherences are:$$\begin{aligned}
\Gamma_{p} & = & \Gamma\left(1-2c^{2}s^{2}\right)\label{eq:Gamma_pop}\\
\Gamma_{c} & = & \frac{\Gamma}{2}\left(1+2c^{2}s^{2}\right)\label{eq:Gamma_coh}\end{aligned}$$ The first term in $S\left(\omega\right)$ corresponds to the elastic scattering, while the three other terms are inelastic components. The last two terms are the sidebands of the Mollow triplet, centered in $\omega_{L}\pm\Omega_{G}$, which are proportional to the intensity when $\left|\Delta\right|\gg\Omega$.
The absorption spectra are given by:$$\begin{aligned}
\sigma_{A}^{(2)}\left(\omega\right) & = & \sigma_{0}\frac{\pi\Gamma}{2}\frac{c^{2}-s^{2}}{c^{4}+s^{4}}c^{4}L\left(\omega-\omega_{L}-\Omega_{G},\Gamma_{c}\right)\\
& & -\sigma_{0}\frac{\pi\Gamma}{2}\frac{c^{2}-s^{2}}{c^{4}+s^{4}}s^{4}L\left(\omega-\omega_{L}+\Omega_{G},\Gamma_{c}\right)\\
\sigma_{A}^{(3)}\left(\omega\right) & = & \sigma_{0}\frac{\pi\Gamma}{2}c^{6}L\left(\omega-\omega_{L}-\Omega_{G}+\delta_{r},\Gamma_{2}\right)\\
& & +\sigma_{0}\frac{\pi\Gamma}{2}s^{6}L\left(\omega-\omega_{L}+\delta_{r},\Gamma_{1}\right)\end{aligned}$$ where $\delta_{r}$ is the light-shift, and $\Gamma_{1,2}$ are the relaxation rates for the coherences between the non-coupled level and the dressed states. $$\begin{aligned}
\delta_{r} & = & \frac{1}{2}\left(\Delta+\sqrt{\Delta^{2}+\Omega^{2}}\right)\label{eq:depl_lum}\\
\Gamma_{1} & = & \frac{\Gamma}{2}\left(1+c^{2}\right)\label{eq:Gamma1}\\
\Gamma_{2} & = & \frac{\Gamma}{2}\left(1+s^{2}\right)\label{eq:Gamma2}\end{aligned}$$ The absorption spectrum of the 2-level part consists in two lines centred in $\omega_{L}\pm\Omega_{G}$, which means that they are resonantly excited by the two sidebands of the Mollow triplet. The two absorption lines due to the third level are centred in $\omega_{L}+\Omega_{G}-\delta_{r}$ and in $\omega_{L}-\delta_{r}$, then the excitation by the fluorescence light is not resonant if $\delta_{r}\gg\Gamma$. Because of the secular approximation we made for the calculation, it is incorrect to write $\sigma_{R}$ as the sum of 16 terms, coming from the overlap of the four components of the fluorescence with the four absorption lines. We have to consider separately the different cases.
In general, for large detunings and intensities ($\Omega,\left|\Delta\right|\gg\Gamma$) and large light-shifts ($\delta_{r}\gg\Gamma$), the resonant term of the 2-level system dominates the sum, and one should drop all the other terms to be consistent with the secular approximation done before. The resulting cross-section has thus the following simple expression:$$\sigma_{R}=\frac{\sigma_{0}}{4}\,\frac{\Delta^{2}\Omega^{2}}{4\Delta^{4}+8\Delta^{2}\Omega^{2}+3\Omega^{4}}\label{eq:resonant}$$ As already mentioned, the contribution due to the non-coupled level is never exactly resonant. However, if $\left|\Delta\right|\gg\Omega$, the light-shift, $\delta_{r}\simeq\Omega^{2}/4\left|\Delta\right|$, can be arbitrarily small, leading to an almost resonant behaviour if $\delta_{r}\lesssim\Gamma$. In the case where $\left|\Delta\right|\gg\Omega\gg\Gamma$ and $\Gamma\gtrsim\Omega^{2}/4\left|\Delta\right|$, we take into account the quasi resonant contribution of the third level:$$\sigma_{R}=\frac{\sigma_{0}\,\Omega^{2}}{16\,\Delta^{2}}\,\left\{ 1+\frac{16\,\Delta^{2}\,\Gamma^{2}}{\Omega^{4}+16\,\Delta^{2}\,\Gamma^{2}}\right\} \label{eq:quasiresonant}$$
[20]{} L. Guidoni and P. Verkerk, J. Opt. B: Quantum Semiclass. Opt. **1** R23 (1999)
*Cold Molecules: Theory, Experiment, Applications*, edited by Roman Krems, Bretislav Friedrich and William C Stwalley (CRC Press, Boca Raton, 2009)
M. H. Anderson, J. R. Ensher, M. R. Matthews, C. E. Wieman and E. A. Cornell, Science **269** 198-201 (1995)
D. Wilkowski, J. Ringot, D. Hennequin and J. C. Garreau, Phys. Rev. Lett. **85** 1839-1842 (2000)
G. Labeyrie, F. Michaud and R. Kaiser, Phys. Rev. Lett. **96**, 023003 (2006)
D. Hennequin, Eur. Phys. J. D **28** 135-147 (2004)
A. di Stefano, M. Fauquembergue, P. Verkerk and D. Hennequin, Phys. Rev. A **67** 033404 (2003)
A. di Stefano, P. Verkerk and D. Hennequin, Eur. Phys. J. D **30** 243-258 (2004)
T. Pohl, G. Labeyrie, and R. Kaiser, Phys. Rev. A **74**, 023409 (2006)
T.G. Walker, D.W. Sesko, and C. Wieman, Phys. Rev. Lett. **64**, 408 (1990)
L. Pruvost, I. Serre, H. T. Duong, and J. Jortner, Phys. Rev. A **61**, 053408 (2000)
J. T. Mendonça, R. Kaiser, H. Terças, and J. Loureiro, Phys. Rev. A **78**, 013408 (2008)
D.W. Sesko, T.G. Walker, and C. Wieman, J. Opt. Soc. Am. B **8** 946 (1991)
H. Neunzert, M. Pulvirenti and L. Triolo, Math. Methods Appl. Sci. **6** 527-538 (1984)
Javier Ramos-Caro and Guillermo A González, Class. Quantum Grav. **25** 045011 (2008)
Dario Benedetto, Emanuele Caglioti, François Golse, and Mario Pulvirenti, Comm. Math. Sci. **2**, 121 (2004)
R. Warnock, Nucl. Instrum. Methods Phys. Res., Sect. A **561**, 186 (2006).
J. Dalibard and C. Cohen-Tannoudji, J. Opt. Soc. Am. B **6**, 2023 (1989)
C. Cohen-Tannoudji, J. Dupont-Roc, and G. Grynberg, *Atom-Photon Interactions* (Wiley, New York, 2004)
B.R. Mollow, Phys. Rev. **188**, 1969 (1969)
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'RR Lyr is one of the most studied variable stars. Its light curve has been regularly monitored since the discovery of the periodic variability in 1899. Analysis of all observed maxima allows us to identify two primary pulsation states defined as pulsation over a long ($P_0$ longer than 0.56684 d) and a short ($P_0$ shorter than 0.56682 d) primary pulsation period. These states alternate with intervals of 13-16 yr, and are well defined after 1943. The 40.8 d periodical modulations of the amplitude and the period (i.e. Blazhko effect) were noticed in 1916. We provide homogeneous determinations of the Blazhko period in the different primary pulsation states. The Blazhko period does not follow the variations of $P_0$ and suddenly diminished from 40.8 d to around 39.0 d in 1975. The monitoring of these periodicities deserved and deserves a continuous and intensive observational effort. For this purpose we have built dedicated, transportable and autonomous small instruments, Very Tiny Telescopes (VTTs), to observe the times of maximum brightness of RR Lyr. As immediate results the VTTs recorded the last change of $P_0$ state in mid-2009 and extended the time coverage of the [*Kepler*]{} observations, thus recording a maximum O-C amplitude of the Blazhko effect at the end of 2008, followed by the historically smallest O-C amplitude in late 2013. This decrease is still ongoing and VTT instruments are ready to monitor the expected increase in the next few years.'
author:
- |
J. F. Le Borgne,$^{1,2,3}$[^1] E. Poretti,$^{1,2,3,4}$ A. Klotz,$^{1,2,3}$ E. Denoux,$^{3}$ H. A. Smith,$^{5}$ K. Kolenberg,$^{6,7}$ R. Szabó,$^{8}$ S. Bryson,$^{9}$ M. Audejean,$^{10}$ C. Buil ,$^{11}$ J. Caron,$^{12}$ E. Conseil,$^{13}$ L. Corp,$^{3,14}$ C. Drillaud,$^{13}$ T. de France,$^{14}$ K. Graham,$^{14}$ K. Hirosawa,$^{15}$ A.N. Klotz,$^{3}$ F. Kugel,$^{12}$ D. Loughney,$^{16}$ K. Menzies,$^{14}$ M. Rodríguez,$^{17}$ and P. M. Ruscitti$^{18}$\
\
$^{1}$Université de Toulouse; UPS-OMP; IRAP; Toulouse, France\
$^{2}$CNRS; IRAP; 14, avenue Edouard Belin, F-31400 Toulouse, France\
$^{3}$GEOS (Groupe Européen d’Observations Stellaires), 23 Parc de Levesville, 28300 Bailleau l’Evêque, France\
$^{4}$INAF-Osservatorio Astronomico di Brera, Via E. Bianchi 46, 23807, Merate (LC), Italy\
$^{5}$Michigan State University, Department of Physics and Astronomy, East Lansing, MI 48824, USA\
$^{6}$Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge MA 02138, USA\
$^{7}$Instituut voor Sterrenkunde, K.U. Leuven, Celestijnenlaan 200D, 3001 Heverlee, Belgium\
$^{8}$Konkoly Observatory, MTA CSFK, Konkoly-Thege Miklós út 15-17, H-1121 Budapest, Hungary\
$^{9}$NASA Ames Research Center, Moffett Field, Mountain View, CA 94035, USA\
$^{10}$Observatoire de Chinon, Astronomie en Chinonais, Mairie, Place du Général de Gaulle, 37500, Chinon, France\
$^{11}$Observatoire de Castanet-Tolosan, 6 place Clémence Isaure, 31320, Castanet-Tolosan, France\
$^{12}$Observatoire Chante-Perdrix, Dauban, 04150 Banon, France\
$^{13}$AFOEV (Association Française des Observateurs d’Etoiles Variables), Observatoire de Strasbourg 11, rue de l’Université, 67000 Strasbourg, France\
$^{14}$AAVSO (American Association of Variable Star Observers), 49 Bay State Rd., Cambridge, MA 02138, USA\
$^{15}$Variable Star Observers League in Japan (VSOLJ), 405-1003 Matsushiro, Tsukuba, Ibaraki 305-0035, Japan\
$^{16}$The British Astronomical Association, Variable Star Section (BAA VSS), Burlington House, Piccadilly, London, W1J 0DU, United Kingdom\
$^{17}$Alberdi 42 2F, 28029 Madrid, Spain\
$^{18}$Osservatorio Astronomico B. Occhialini, Via G. Garibaldi 17, 67041 Aielli (AQ), Italy\
date: 'Accepted ..... Received ....; in original form ....'
title: |
The historical vanishing of the Blazhko effect of RR Lyr\
from GEOS and [[*Kepler*]{}]{} surveys
---
\[firstpage\]
techniques: photometric – stars: individual: RR Lyrae – stars: oscillations – stars: variables: RR Lyrae.
Introduction
============
The modulation of the amplitude in luminosity and the pulsation period, known as Blazhko effect, is observed in numerous RR Lyr stars [@blagal]. Alhough its theoretical explanation has not yet been determined, a considerable breakthrough in its interpretation has been realized because of recent space observations. As a matter of fact, CoRoT [@793; @eli1] and [[*Kepler*]{}]{} [@eli2] results have shown that the Blazhko mechanism is not acting as a clock, thus undermining both the competing models based on a strict regularity: the oblique pulsator and the resonant nonradial pulsator. In our previous work on galactic Blazhko stars [@blagal], we did not consider the eponym of the class, i.e., HD 182989$\equiv$RR Lyr. We know that its pulsational period shows some apparent erratic changes [@pervar]. Indeed, we thought that the observation of [RR Lyr]{} through the ages deserved a detailed study and, furthermore, dedicated projects.
The variability of [RR Lyr]{} was discovered by Mrs. Willemina P. Fleming [see @cannon for a short but complete biography] on the photographic plates of the Henry Draper Memorial in 1899 [@pickering]. @wendell reports that the first observation dates back to July 20, 1899 and the first maximum to September 23, 1899 [HJD 2414921.675, calculated by @schutte]. @prager and @shapley were the first to show the correlated photometric and spectroscopic variations. Fig. 5 of Shapley (1916) constitutes the [*ante litteram*]{} representation of the Blazhko effect – note that this definition was introduced later in the astronomical literature – because it shows the oscillation of the median magnitude of the ascending branch with a period of 40 d and an amplitude of 37 min.
@detre43 compiled the first long list of observed maxima and reported a detailed investigation of the Blazhko effect [@balazs]. @preston65 found that spectroscopic and photometric parameters were strongly variable in the Blazhko cycles of 1961 and 1963, but essentially constant in those of 1962. The idea of irregularities in the Blazhko period gradually grew and later extended photoelectric observations suggested a 4-year cycle in the Blazhko effect, with a minimum variability in 1971 June and a new maximum variability in 1972 April [@detreibvs]. A hundred years after the discovery of the [RR Lyr]{} variability, @budapest have reported that attempts at representing the Blazhko variations by longer periods have failed because they are not strictly repetitive. Extensive photoelectric and CCD observations obtained over a 421-day interval in 2003-2004 fixed a shorter Blazhko period of 38.8$\pm$0.1 d [@kk2006]. Finally, [RR Lyr]{}$\equiv$KIC7198959 was included in the field of view of the [[*Kepler*]{}]{} space telescope [@borucki]. Thus, high-precision, continuous observations could be secured, first in the long cadence (29.4 min) mode and then in the short cadence (1.0 min) observing mode. @kk2011 have reported the results in the long cadence mode, while @molnar and @pdm have reported the results in the short cadence mode. [RR Lyr]{} shows no particularities in metallicity, abundances and physical characteristics when compared with the other [[*Kepler*]{}]{} RR Lyrae stars, Blazhko and non-Blazhko, observed by means of homogeneous high-resolution spectroscopy [@nemec].
It is clear that the full comprehension of the [RR Lyr]{} variability is very time demanding since the star has to be observed over decades with time series having a high temporal resolution, in order to survey the pulsation period ([$P_0$]{}) of 13 h, the Blazhko period ([$P_B$]{}) of 40 d, and the long-term changes of both.
--------------------------- -------------------------- -----------
Maurice Audejean Reflector 320mm CCD
Christian Buil Reflector 280mm CCD$^{a}$
Emmanuel Conseil Reflector 150mm DSLR
Laurent Corp Photographic lens CCD
Eric Denoux VTT CCD
Eric Denoux Reflector 280mm CCD
Christian Drillaud Refractor 70mm DSLR
Thibault de France Refractors 60mm and 80mm CCD
Thibault de France Reflector 130mm CCD
Keith Graham Reflector 200mm CCD
Kenji Hirosawa Photographic lens DSLR
Alain and Adrien N. Klotz VTT CCD
F. Kugel and J. Caron Photographic lens CCD
F. Kugel and J. Caron Refractor 80mm CCD
Jean-François Le Borgne VTT CCD
Des Loughney Photographic lens DSLR
Kenneth Menzies Reflector 317mm CCD
Miguel Rodríguez Refractor 60mm CCD
Paolo Maria Ruscitti Reflector 130mm DSLR
Horace A. Smith and coll. Reflector 600mm$^{b}$ CCD
--------------------------- -------------------------- -----------
: Observers and observing instruments.[]{data-label="observers"}
$^{a}$Synthetic photometry from low resolution spectra using Shelyak Alpy 600 spectrograph.\
$^{b}$Telescope at the Michigan State University campus Observatory (East Lansing, Michigan, USA) operated by H. Smith, with the help of Michigan State University students Charles Kuhn, James Howell, Eileen Gonzales, and Aron Kilian.
Observations
============
With the goal of making the ground-based survey of the Blazhko effect of RR Lyr as effective as possible, we decided to devise small, autonomous, and transportable photometric instruments. The basic idea was that these instruments are able to follow [RR Lyr]{} continuously, in order to obtain a reliable time of maximum brightness ([T$_{\it max}$]{}) on as many clear nights as possible. Since the calibration of a photometric system would have required a major refiniment and would have increased the costs of the instrument (e.g. filters, standard stars observations, cooling, etc.), both ill-suited for the requirements of simplicity and duplicability, the determination of a standard, calibrated magnitude of the maximum was not pursued as a goal. The first observations were performed in 2008 and these are expected to continue for many years to come. The instruments are composed of a commercial equatorial mount (Sky-Watcher HEQ5 Pro Goto), an AUDINE CCD camera (512x768 kaf400 chip) and a photographic 135-mm focal, f/2.8 lens with a field of view of 2$^\circ$x3$^\circ$. We gave them the nickname, Very Tiny Telescopes (VTTs). Three such instruments have been built and used mainly in 3 different places near Toulouse, Castres and Caussades (Région Midi-Pyrénées, France). However they have been occasionally moved to several other places in France, Spain, and Italy.
The observations of the VTTs are controlled from a computer using the program AudeLA. VTT images were obtained with no filter with an exposure time of 30 s. Science images have been corrected with mean dark images obtained during the night to compensate the absence of cooling. A dark frame was obtained every five images on the target and the mean dark image of five individual dark frames was subtracted from the 25 target images concerned. From June 2008 to November 2013, 332 maxima were measured by VTTs with about 360000 images collected on 829 nights. The photometry of [RR Lyr]{} and of the comparison star HD 183383 was performed with the SExtractor software [@bertin]. The times of maximum brightness were determined by means of cubic spline functions with a non-zero smoothing parameter which depends on the number of measurements to fit and on their scatter [@reinsch]. The smoothing parameter was chosen to be large enough to avoid local maximums and so that the fitting curve goes through the points with zero mean residuals over a characteristic time interval of 5 min. The uncertainty on the time of the maximum was the difference between the two times corresponding to the intersection of the spline function with the line $y=m_{max} + \sigma/\sqrt{N-1}$, where $m_{max}$ is the instrumental magnitude at maximum, $\sigma$ is the standard deviation of the fit and $N$ is the number of measurements used in the light curve.
Alhough VTT observations are most numerous during the 2008-2013 campaign (829 night runs compared to 938 in total), observations with other intruments are also included in the present study (Table \[observers\]). Most of these additional observations where carried out with classical telescopes, refractors or reflectors, with diameters from 55 mm to 320 mm, equipped with CCD cameras. Among these, [ *BVI*]{} measurements were obtained with the 60-cm telescope at the Michigan State University campus observatory. However, some observations were carried out with digital photographic cameras (DSLR). As a new approach to spectrophotometry, synthetic photometry was performed on low-resolution spectra using a Shelyak Alpy 600 spectrograph mounted on a 280-mm diameter reflector. Low-resolution spectra ($R$=600) were obtained through a wide slit and calibrated in flux by means of spectrophotometric standard stars. Photometry is then performed by integrating the spectra in Johnson filter bandpasses.
GEOS data base
==============
The Groupe Européen d’Observations Stellaires (GEOS) RR Lyr data base [^2] (Le Borgne et al. 2007) is a collection of published maxima of galactic RR Lyr stars which contains 2245 maxima of RR Lyr itself (up to 2013 December 5). In Appendix A, we give the list of references used to build the GEOS data base for RR Lyr. In our analysis, we did not consider uncertain maxima and those noted by the authors as normal, created from observations drawn from many different individual maxima. This is because they could be referred to a wrong epoch if an inaccurate [$P_0$]{} value were to be used. The normal maximum could also be the arbitrary epoch of an ephemeris. This is the case of the first ephemerides of [RR Lyr]{} [@prager; @shapley; @sanford] that were calculated using JD 2414856, i.e., the date of the first observation, rather than JD 2414921, i.e., the date of the first maximum. Moreover, a normal maximum masks the Blazhko effect because it averages observations on a large interval of time.
We added 692 new photographic and photoelectric maxima observed by L. Detre in the years 1944-1981. The list of $\sim$7000 measurements reported by @Szeidl97 was scanned and digitalized by means of a semi-automatic procedure. These measurements are now available on electronic form on the Konkoly Observatory web site. We also evaluated the differences between 278 [T$_{\it max}$]{} observed simultaneously in $B$ and $V$ filters. No systematic effect was detected. 67 per cent of the [T$_{\it max}$]{} differences are within the interval from $-0.0016$ to $+0.0016$ d, symmetrically distributed with respect to 0.000 d. The frequency analysis did not detect any periodicity in the [T$_{\it max}$]{} differences. Finally, we measured 25 [T$_{\it max}$]{} values from the original observations collected at Michigan University in the framework of the 2003-04 campaign [@kk2006].
[$P_0$]{} variations of RR Lyr over more than one century
=========================================================
We analysed 3975 [T$_{\it max}$]{} (obvious outliers were removed) spanning 114 yr and we calculated the linear ephemeris from a least-squares fitting of all them: $${\rm HJD\,\,\, Max}\ = 2414921.7746 + 0.566835616 \cdot {\rm E}
\label{cst}$$
In Appendix B, we give the list of [T$_{\it max}$]{}used to calculate equation \[cst\] and used in the subsequent analysis. The purpose of using the above ephemeris was to detect large changes in the [$P_0$]{} value. Indeed, the plot of the O-C (observed minus calculated [T$_{\it max}$]{}) values clearly pointed them out (Fig. \[cstperiod\]). Therefore, we subdivided the 114 yr of observations of [RR Lyr]{} into several time intervals, following the [$P_0$]{} changes regardless of the observing technique. Table \[history\] lists the actual [$P_0$]{} values in each interval.
The Blazhko period contributes greatly to an increase in the O-C scatter beyond that attributable to the observing technique alone. The varying thickness in Fig. \[cstperiod\] also suggests a variable amplitude. To study the behaviour of [$P_0$]{}, we calculated a linear ephemeris in each interval of Table \[history\] this time by dividing the [T$_{\it max}$]{}on the basis of the observing technique (visual, photographic, photoelectric, or CCD) or instrument (VTTs, [[*Kepler*]{}]{}). These subsets supplied independent values of [$P_0$]{} (Table \[history2\]) in good agreement with those of the whole time interval (Table \[history\]). The uncertainties on [$P_0$]{} (as well as those on the Blazhko period [$P_B$]{}, see below) are the formal error bars derived from the least-squares fittings. About the first long time interval, we preferred to use the visual maxima instead of the photographic maxima because the visual technique implied the survey of the star for several hours, while the photographic technique often recorded a few measurements only. The visual maxima are very useful to fill the long gap between 1982 and 2000, not covered by photoelectric observations before many amateur astronomers upgraded their instrumentation to CCD detectors. In addition to the time intervals listed in Table \[history2\], we note that visual maxima provided [$P_0$]{} values in excellent agreement with the photoelectric and CCD ones in the intervals JD 2439000-2442500, 2442500-2445000, and 2455000-2456200.
[ll r l]{} & & &\
& & &\
1899-1908 & 14921-18000 & 5 & 0.5668 (1)\
1908-1946 & 18000-32000 & 138 & 0.5668 481(3)\
1946-1958 & 32000-36500 & 70 & 0.5668 116(9)\
1959-1965 & 36500-39000 & 164 & 0.5668 721(17)\
1965-1975 & 39000-42500 & 418 & 0.5668 204(4)\
1975-1981 & 42500-45000 & 324 & 0.5668 585(10)\
1982-1989 & 45000-47800 & 205 & 0.5668 224(12)\
1989-1995 & 47800-50000 & 103 & 0.5668 427(17)\
1995-2003 & 50000-53000 & 132 & 0.5668 174(10)\
2004-2009 & 53000-55000 & 188 & 0.5668 619(14)\
2009-2013 & 55000-57000 &2228 & 0.5667 975(4)\
[l l r ll c]{} & & & & &\
& & & & &\
19635-27313 & visual & 75 & 0.5668 470(5) & 40.89(3) & 0.018(3)\
32062-33455 & photog. & 27 & 0.5668 193(45)& 40.86(28) & 0.014(3)\
33505-36457 & photoel. & 39 & 0.5668 200(9) & 40.94(28) & 0.005(2)\
36674-38996 & photoel. & 160 & 0.5668 734(17)& 40.88(10) & 0.018(2)\
39008-42405 & photoel. & 342 & 0.5668 205(4) & 41.15(3) & 0.014(1)\
42504-44822 & photoel. & 162 & 0.5668 583(13)& 38.92(12) &0.013(2)\
45493-47779 & photoel. & 14 & 0.5668 271(26) &\
45131-47982 & visual & 193 & 0.5668 239(13) & 39.02(20) & 0.007(3)\
48012-50000 & visual & 98 & 0.5668 427(18) & 39.03(10) & 0.018(3)\
50224-52926 & visual & 126 & 0.5668 1703(10) & 39.06(11) & 0.012(2)\
52915-54733 & CCD & 40 & 0.5668 621(28) & 39.00(5) & 0.024(2)\
54652-55000 & VTTs & 69 & 0.5668 589(14) & 39.39(5) & 0.026(2)\
55276-56390 & [*Kepler*]{} & 1815 & 0.5667 953(4) & 38.84(2) & down to 0.009\
55000-56624 & VTTs & 264 & 0.5668 024(11) & 38.91(7) & down to 0.006\
After an initial change, which the few observed maxima place around 1910, the period of [RR Lyr]{} was constant for about 36 yr. Then, around 1946, it started a series of suddden changes, on a timescale of few years. The O-C pattern shows jumps from long values (O-C’s from negative to positive values, with maxima values reached on 1946.5, 1965.7, 1982.1, 1994.4, and 2009.5) to short values (O-C’s from positive to negative values, with minima values reached on 1958.8, 1975.2, 1989.8, and 2004.0). Table \[history\] lists the computed values of [$P_0$]{} after any observed change and the last three digits of [$P_0$]{} are also noted in Fig. \[cstperiod\]. The minimum difference between a long and a short value is between JD 2445000 and 2450000. One of the changes of state is well covered by VTT observations: when considering the [T$_{\it max}$]{}before JD 2455000 the period is a long one (0.56686 d), after a short one (0.56680 d, see Table \[history2\]).
We also note that there is a sort of semiregular cadence separating two consecutive O-C maxima or two consecutive O-C minima, about 13–16 years and that the [$P_0$]{} is still decreasing: if we consider the maxima after the change at JD 2455000 only, the period is below 0.5668000 d (Tables \[history\] and \[history2\]).
[$P_B$]{} variations of RR Lyr over more than one century
=========================================================
Because the pulsation period of [RR Lyr]{} was undergoing changes from two different states, the question of whether these changes effect the Blazhko effect immediately arose. This point could be very important in understanding the relation between the two periods. It required the analyses of all the previous observations by a homogenous procedure, because often the [$P_B$]{} values were determined in different ways or also simply assumed from previous works.
Therefore, we performed the frequency analysis of the O-C values obtained from the linear fits used to determine the [$P_0$]{} in each subset (Table \[history2\]). We used the iterative sine-wave fitting method [@vani] and we present the results as amplitude spectra for the sake of clarity. Figure \[spectra\] shows the application of the method to several subsets covering a time interval of one century. We can clearly see that the amplitude is variable and that [$P_B$]{} jumps from left to right of the mark at 0.025 d$^{-1}$ (i.e., 40.0 d) around 1975. The [$P_B$]{} values obtained from the highest peaks were refined by means of the MTRAP code [@mtrap] and the final values are listed together with error bars and amplitudes in the last two columns of Table \[history2\]. It is quite evident that [$P_0$]{} and [$P_B$]{} changed in a way completely uncorrelated with each other (see Sect. \[disc\]).
The [[*Kepler*]{}]{} data
=========================
[[*Kepler*]{}]{} data allowed us to combine the analysis of [T$_{\it max}$]{} variations with another specific Blazhko characteristic, i.e., the changes of the magnitude at the maximum brightness ([$K_{p,max}$]{}). We determined [T$_{\it max}$]{} and [$K_{p,max}$]{} from the original [[*Kepler*]{}]{} data by means of the same procedure used for VTT data. We used the Q5-Q16 short-cadence data acquired from 2010 March 20 to 2013 April 3. The analysis of the almost continuous succession of observed maxima already pointed out a totally new and prominent feature, i.e., the alternation of higher and lower maxima [i.e. period doubling; see Fig. 4 in @szabo2010 for a clear example]. A theoretical background has been proposed for this new phenomenon [@molnar2011]. We investigated the regularity of this effect all along the time interval of the [[*Kepler*]{}]{} observations (Fig. \[doubling\]). The scatter due to the period doubling effect is always noticeable (top panel). The amplitudes are variable and the largest amplitudes are not related to a particular phase of the Blazhko effect, because the related large scatter is observed at both the maximum and at the minimum values of [$K_{p,max}$]{}. Moreover, there are Blazhko cycles where the period doubling effect is always very noticeable, as that from BJD 2455710 to 2455750. There is also a damping of the effect toward the end of observing time, when also [$K_{p,max}$]{} variations have a small amplitude.
As a new contribution to the characterization of the period doubling effect, we calculated the differences between the [$K_{p,max}$]{} value of an even (2$n$) epoch and that of an odd (2$n-1$) epoch. These differences are both positive and negative (middle panel) and this implies that the highest [$K_{p,max}$]{} changes from an odd epoch to an even one. In this plot the highest maxima or the deepest minima are separated by a time interval corresponding to the characteristic period of the switching from an odd epoch of high [$K_{p,max}$]{} to an even one. Moreover, more rapid fluctuations are also visible. It is worth analysing these time series to search for periodicities in the switching process. The iterative sine-wave fitting method [@vani] is well suited to disentangle such periodicities because it allows the detection of the components of the light curve one by one. Only the values of the detected frequencies (known constituents) are introduced in each new search, while their amplitudes and phases are recalculated for each new trial frequency. In such a way, the exact amount of signal for any detected frequency is always subtracted. In the first power spectrum of the [$K_{p,max}$]{} differences (bottom-left panel), the highest peak was at a low frequency, $f_1$=0.018 d$^{-1}$, corresponding to $P$=55.6 d, i.e., about 98 [$P_0$]{}. After introducing it as a known constituent, we could identify a higher frequency, $f_2$=0.062 d$^{-1}$ (bottom-right panel), corresponding to $P$=16.1 d, i.e., about 28.5 [$P_0$]{}. The peak close to $f$=0.0 d$^{-1}$ appearing in both spectra is a result of the very long-term effect (see below).
The [[*Kepler*]{}]{} data make it possible for us to study in detail the cycle-to-cycle variations of the Blazhko effect in [RR Lyr]{}. To do this, we calculated a least-squares fit of the observed O-C and [$K_{p,max}$]{} values on sliding boxes of 56 d shifted from each other by 8 d. This procedure returns the values of the amplitudes of the O-C and [$K_{p,max}$]{} variations for each box (Fig. \[patata\]). The general trend for both amplitudes is a slow decrease. In particular, the O-C curve does not seem to have reached the final minimum at the end of [[*Kepler*]{}]{} observations (middle panel), while the [$K_{p,max}$]{} amplitude curve seems to start to increase again after a shallow minimum at JD 2459000 (bottom-left panel). The extreme changes in the Blazhko effects are sketched by the shrinking of the close curve connecting O-C and [$K_{p,max}$]{} values (bottom-right panel). The alternation of low and high maxima also twists the regular shape of the close curve, adding a new model to the already variegated collection [@blagal]. Modulations are clearly visible both in the O-C and in the [$K_{p,max}$]{} amplitudes (Fig. \[patata\]). Combined with the long-term trend, they produce the cycle-to-cycle variations of the Blazhko effect. We performed the frequency analysis of the time series of the O-C and [$K_{p,max}$]{} amplitudes to search for periodicities by means of a sinusoidal fit and a parabolic trend. The two power spectra are characterized by broad structures with different highest peaks at low frequencies. The inconsistency between the two results does not support a reliable identification of real periodicities in the changing shape of the Blazhko effect.
VTT data
========
The first VTT [T$_{\it max}$]{} was observed on 2008 July 4 simultaneously with two instruments. Since then the regular survey of [RR Lyr]{} has yielded 55, 78, 42, 56, 75, and 27 [T$_{\it max}$]{} from 2008 to 2013, respectively. To analyse VTT data we performed a least-squares fit of the O-C values determined each year. Indeed, ground-based observations of RR Lyr are concentrated in a few months of each year, covering about two consecutive Blazhko cycles. The variations between the Blazhko cycles in a given year are very small and consequently the folded curves of O-C over the Blazhko cycle are quite representative of the behavior in each year. Indeed, the continuous decline in amplitude and the changing shape of the O-C curve is well reproduced and variations have already become noticeable from one year to the next already (Fig. \[ocampli\]). In particular, note how the curve becomes more sinusoidal with the decreasing O-C amplitude.
By adding the VTT points to [[*Kepler*]{}]{} ones (Fig. \[patata\], middle panel), we can state that the maximum O-C amplitude has been reached well before the space observations began. The CCD determinations preceeding the VTT observations (40 [T$_{\it max}$]{} between JD 2452915 and 2454733) supply an O-C amplitude of 0.024 d (Fig. \[spectra\]), while the visual [T$_{\it max}$]{} s collected before CCD ones supply smaller amplitudes, i.e., 0.018$\pm$0.003 d and 0.012$\pm$0.002 d (Table \[history2\]). From these values we can infer that a minimum O-C amplitude occurred when only visual [T$_{\it max}$]{} s were available, in 1984-85 (tentatively around JD 2446000) followed by an increase that reached its maximum (0.031 d) at the beginning of the VTT observations, near the end of 2008 (JD 2454800, see Fig. \[ocampli\], top panels). The subsequent decrease is still ongoing (bottom panels). Indeed, the O-C amplitude recorded by the VTTs until the end of 2013 (0.006 d) is even smaller than that of the last [[*Kepler*]{}]{} data (Fig. \[patata\], middle panel).
Discussion {#disc}
==========
The VTT observations started 624 d before the first [[*Kepler*]{}]{} observation in short-cadence mode, and they are still continuing, while the last [[*Kepler*]{}]{} [T$_{\it max}$]{} was obtained in 2013, in early April. The re-analysis of the [T$_{\it max}$]{} epochs listed in the GEOS data base allowed us to reconstruct the changes in the pulsational period of [RR Lyr]{}. We could establish the existence of two states characterized by the pulsation over a long [$P_0$]{} (longer than 0.56684 d) and over a short [$P_0$]{} (shorter than 0.56682 d). The history begins with a long [$P_0$]{} status that lasted from 1910 to 1943. After this, the two states alternate more frequently and usually long states last much less than the short states. Since 1943 the same state seems to re-appear after a time interval of 13-16 yr (Fig. \[cstperiod\]). The frequency analysis of the O-C values since 1950 shows the highest peak at 14 yr. The last cycle began in 2003 with a long [$P_0$]{} and the VTTs recorded that it switched into a short (the shortest, actually) one in 2009, and that is still running. The cyclic shift of [$P_0$]{} amounts on average to $\Delta P_0=4\cdot10^{-5}$ d (Table \[history\]) and hence $\Delta P_0/P_0=7\cdot10^{-5}$.
We also determined the periods of the Blazhko effect by means of an homogenous technique. We can provide a new reliable chronological set of values since the beginning of the twentieth century, replacing the values reported by each author on the basis of different methods of analysis or simply adopting values reported in the literature [see Table 6 in @kk2006 for a detailed list]. The alternate states of [$P_0$]{} do not have a counterpart in the variations of [$P_B$]{} (Fig. \[correl\]). Actually, when comparing the whole set of the new determinations of [$P_B$]{} we can argue that [$P_B$]{} suddenly changed in 1975 (around JD 2442500, Fig. \[correl\]), much earlier than that reported by @kk2006. It was around 41 d until 1975 and since then shortened to 39 d; the shift from one side to the other of the 40 d mark is also visible in Fig. \[spectra\]. The corresponding rate $\Delta P_B/P_B=0.05$ is three orders of magnitude larger than that observed for [$P_0$]{}. Correlated and anticorrelated changes of [$P_0$]{} and [$P_B$]{} were observed in Blazhko RRab stars: RW Dra [@firma], XZ Dra [@xzdra], XZ Cyg [@xzcyg], RR Gem [@rrgem], RV UMa [@rvuma], M5 stars [@m5], RZ Lyr [@rzlyr], and Z CVn [@blagal]. [RR Lyr]{} shows alternate states of long and short [$P_0$]{} combined with the decreasing [$P_B$]{}. Taking into account that changes of [$P_0$]{} and [$P_B$]{} also occurred at different epochs, it seems that [RR Lyr]{} adds another kind of relation between the two periods describing the light curves of Blazhko stars.
The combination of [[*Kepler*]{}]{} and VTT data supplies us with a clear picture of the vanishing of the Blazhko effect. The space telescope continuously monitored the monotonic long-term decrease, proving that small-scale modulations, lasting from 2 to 4 [$P_B$]{}, are also visible in the O-C values. The VTTs have allowed us to assess that the decline in amplitude started in 2008 and it is still ongoing. The plot of the O-C amplitude (Fig. \[patata\], middle panel) covers about 5 yr and it shows the continuous decrease. Hence, it does not support the action of a 4-yr modulation cycle of the Blazhko effect [@detreibvs]. We also note that the minimum full-amplitudes of the [T$_{\it max}$]{} and [$K_{p,max}$]{} variations observed with VTTs and [[*Kepler*]{}]{} (0.012 d and 0.04 mag, respectively) are about half those recorded in the 1971 minimum (0.020 d and 0.07 mag). Therefore, it seems evident that we are observing the historical minimum level of the Blazhko effect. Such a small value was observed perhaps only during the sharp decrease after the O-C maximum in 1950 (Fig. \[cstperiod\]), but the event is poorly covered because of the small numbers of [T$_{\it max}$]{}s. Another minimum O-C amplitude was perhaps observed around 1985, but on this occasion we only obtain very scattered visual [T$_{\it max}$]{}. We note that these three minimum O-C amplitudes occurred when [$P_0$]{} was in the short state.
Combined with the Blazhko effect, the period doubling is making [RR Lyr]{} still more intriguing. The analysis of the [[*Kepler*]{}]{} short-cadence data was helpful in understanding this new effect. We can verify that this effect does not seem to be related to any particular Blazhko phase and can be observed at any time in the data. We find two clear periodicities describing a long-time (55.6 d) and a short-time (16.1 d) switch between the epochs (odd or even) of the higher [$K_{p,max}$]{}. Both these periodicities are not obviously related with [$P_B$]{}. We can just note that 55.6/38.8=1.43, roughly similar to the occurrence of half-integer values, but the 1.5 value is not matched. Moreover, here we are dealing with periods, while the period doubling effect can be represented by means of half-integer values of the pulsational frequency ($f/2, 3/2f, 5/2f$, ...).
Conclusions
===========
The most promising mechanism that can explain the Blazhko effect is the $9P_9$=2[$P_0$]{} resonance between the ninth overtone and the fundamental mode, also capable of producing the period doubling effect [@buchler]. A recent new explanation is based on the transient excitation of the first overtone radial mode [@gillet]. The signature of this mode has already been found in the Q5-Q6 [[*Kepler*]{}]{} data [@molnar] and the analysis of other datasets is ongoing. The results described here supply a complete overview of the behaviour of the Blazhko effect and of the pulsation content of [RR Lyr]{} since its discovery 114 yr ago, thus putting time constraints on the Blazhko mechanism. In particular, the completely different behaviours of the [$P_0$]{} and [$P_B$]{} changes suggest that they are not coupled in a direct way.
The VTT monitoring complemented the [[*Kepler*]{}]{} one and allowed us to follow the historical minimum amplitude of the Blazhko effect. The previously suggested 4-yr cycle does not seem effective in fitting the cycle-to-cycle and the long-term variations. The alternation of the long and short [$P_0$]{} with a semiregular timescale of 14 yr stresses the necessity to collect long series of light maxima in a continuous way. Because of the new [[*Kepler*]{}]{} orientation, VTTs are probably the only instruments that can monitor the expected regrowth of the Blazhko effect and to measure the new [$P_0$]{} in the coming years.
These insights lead new contributions to the description of the pulsation of [RR Lyr]{}. They corroborate our decisions to maintain an updated data base of [T$_{\it max}$]{} of [RR Lyr]{} stars [@pervar], to monitor Blazhko stars with modern instruments [@blagal], and to have started the project to continuously monitor [RR Lyr]{} itself. These facts support us in our aim to continue and to improve the VTT project for several more years.
Acknowledgments {#acknowledgments .unnumbered}
===============
Funding for the [[*Kepler*]{}]{} Discovery Mission is provided by NASA’s Science Mission Directorate. The [[*Kepler*]{}]{} Team and the [[*Kepler*]{}]{} Guest Observer Office are recognized for helping to make this mission and these data possible. EP acknowledges Observatoire Midi-Pyrénées for the two-month grant allocated in 2013 October and November which allowed him to work at the Institut de Recherche en Astrophysique et Planétologie in Toulouse, France. The present study has used the SIMBAD data base operated at the Centre de Données Astronomiques (Strasbourg, France) and the GEOS RR Lyr data base hosted by IRAP (OMP-UPS, Toulouse, France). This study has been supported by the Lendület-2009 Young Researchers Program of the Hungarian Academy of Science, the Hungarian OTKA grant K83790 and the KTIA URKUT\_10-1-2011-0019 grant. The research leading to these results has received funding from the European Community’s Seventh Framework Programme (FP7/2007-2013) under grant agreement no. 269194 (IRSES/ASK). RSz was supported by the János Bolyai Research Scholarship of the Hungarian Academy of Sciences. The basic ideas of the VTT project have been sketched during several GEOS meetings and fruitful discussions with R. Boninsegna, M. Dumont, J. Fabregat, F. Fumagalli, D. Husar, J. Remis, J. Vandenbroere, and J.M. Vilalta are gratefully acknowledged.
[99]{} Balázs J., Detre L., 1943, Communications of the Konkoly Observatory, No. 18, p. 125 Bertin E., Arnouts S., 1996, A&A, 117, 393 Borucki W.J. et al., 2010, Science, 327, 977 Buchler J.R., Kolláth Z., 2011, ApJ, 731, 24 Cannon A.J., 1911, ApJ, 34,314 Carpino M., Milani A., Nobili A.M., 1987, A&A, 181, 182 Detre L., 1943, Communications of the Konkoly Observatory, No. 17, p. 1 Detre L., Szeidl B., 1973, Inf. Bull. Variable Stars, No. 764 Firmanyuk B.N., 1978, Astron. Tsirkulyar, 1019, 3 Gillet D., 2013, A&A, 554, A46 Guggenberger E., Kolenberg K., Chapellier E., Poretti E., Szabó R., Benkő J.M., Paparó M., 2011, MNRAS, 415, 1577 Guggenberger E. et al., 2012, MNRAS, 424, 649 Hurta Zs., Jurcsik J., Szeidl B., Sódor Á., 2008, AJ, 135, 957 Jurcsik J., Benkő J.M., Szeidl B., 2002, A&A, 396, 539 Jurcsik J. et al., 2012, MNRAS, 423, 993 Kolenberg K. et al., 2011, MNRAS, 411, 878 Kolenberg K. et al., 2006, A&A, 459, 577 Kolláth Z., Molnár L., Szabó R., 2011, MNRAS, 414, 1111 LaCluyzé A. et al., 2004, AJ, 127, 1653 Le Borgne J.F. et al., 2007, A&A, 476, 307 Le Borgne J.F. et al., 2012, AJ, 144, 39 Molnár L., Kolláth Z., Szabó R., Bryson S., Kolenberg K., Mullally F., Thompson S.E., 2012, ApJ, 757, L13 Nemec J.M., Cohen J.G., Ripepi V., Derekas A., Moskalik P., Sesar B., Chadid M., Bruntt H., 2013, ApJ, 773, 181 Pickering E. C., Colson H. R., Fleming W. P., Wells L. D., 1901, ApJ, 13, 226 (also Harvard Circ., 54) Poretti E. et al., 2010, A&A, 520, A108 Prager R., 1916, Sitzb. Ak. Berlin, 8, 216 Preston G. W., Smak J., Paczynski B., 1965, ApJs, 12, 99 Reinsch H., 1967, Numerische Mathematik 10, 177 Sanford R.F., 1928, ApJ, 67, 319 Schütte K., 1923, Astron. Nachr., 218, 170 Shapley H., 1916, ApJ, 43, 217 Sódor Á., Szeidl B., Jurcsik J., 2007, A&A, 469, 1033 Stellingwerf R.F., Nemec J.M., Moskalik P., 2013, arXiv:1310.0543, in Kinemuchi K., Kuehn C.A., De Lee N., Smith H.A., eds., 40 Years of Variable Stars: A Celebration of Contributions by Horace A. Smith conference, arXiv1310.0149 Szabó R. et al., 2010, MNRAS, 409, 1244 Szeidl B., Guinan E. F., Olah K., Szabados L., 1997, Comunications of the Konkoly Observatory, Hungary, 99, 1 Szeidl B., Hurta Zs., Jurcsik J., Clement C., Lovas M., 2011, MNRAS, 411, 1763 Szeidl B., Kolláth Z., 2000, in Szabados L., Kurtz D.W., eds, ASP Conf. Series Vol. 203, The impact of Large-Scale Surveys on Pulsating Star Research, Astron. Soc. Pac., San Francisco, p. 281 Vaniĉek P., 1971, ApSS, 12, 10 Wendell O.C., 1909, Annals of Harvard College Observatory, vol. 69, p. 1
References to published times of maximum
========================================
The historical study of the variations of the pulsating period and of the Blazhko effect period of RR Lyr uses determinations of times of individual maxima reported in numerous publications since the early twentieth century. We list here the references of the publications available, to the best of our knowledge. As said in the text, the reference @Szeidl97 does not contain determinations of times of maximum, but the measurements from which we determined the individual times of maxima.
Berdnikov L.N., 1973, Astronomical Circular 777, 8 Berdnikov L.N., 1977, Variable Stars Supp. 3, 329 Bogdanov M.B., 1972, Variable Stars Supp. 1, 309 Braune W., Hübscher J., 1967, Astron. Nachr. 290, 105 (BAV Mitt. 18) Braune W., Hübscher J., 1987, BAV Mitt. 46 Braune W., Hübscher J., Mundry E., 1970, Astron. Nachr. 292,185 (BAV Mitt. 23) Braune W., Hübscher J., Mundry E., 1972, Astron. Nachr. 294,123 (BAV Mitt. 25) Braune W., Hübscher J., Mundry E., 1977, Astron. Nachr. 298,121 (BAV Mitt. 28) Braune W., Hübscher J., Mundry E., 1979, Astron. Nachr. 300,165 (BAV Mitt. 29) Braune W., Hübscher J., Mundry E., 1981, Astron. Nachr. 302, 53 (BAV Mitt. 31) Braune W., Mundry E., 1973, Astron. Nachr. 294, 225 (BAV Mitt. 26) Busch H., 1975, MVS 7, 2, 32 Busch H., 1986, MVS 11, 1, 19 Datchenko A.I., 1982, Variable Stars Supp. 4, 275 Detre L., 1943, Communications of the Konkoly Observatory, Hungary, 17 Fedotov Y.T., Gladyr A.V., 1989, Astronomical Circular 1537, 25 Ferrand S., Vandenbroere J., 2013, GEOS Circular RR52 Firmanyuk B.N., 1977, Astronomical Circular 979, 7 Firmanyuk B.N., Derevyagin V.G., Lysova L.E., Movchan A.I., Fedotov Y.T., 1987, Astronomical Circular 1508, 7 Fitch W.S., Wisniewski W.Z., Johnson H.L., 1966, Comm. Lunar and Planetary Lab., vol. 5, part 2, (no71) Florya N.F., 1936, Variable Stars 5, 55 Gordon K., Kron G., 1947, ApJ 106, 318 Hertzsprung E., 1922, BAN 1, 139, obs. 45 209 Hirosawa K., 2012, VSOLJ 52 Hirosawa K., 2013, VSOLJ 54 Hübscher J., 2000, BAV Mitt. 131 Hübscher J., 2001, BAV Mitt. 143 Hübscher J., 2002, BAV Mitt. 154 Hübscher J., 2003, BAV Mitt. 157 Hübscher J., 2005a, BAV Mitt. 171 Hübscher J., 2005b, IBVS 5643, (BAV Mitteilungen No. 172) Hübscher J., 2005c, BAV Mitt. 174 Hübscher J., Agerer F., 1996, BAV Mitt. 93 Hübscher J., Agerer F., Frank P., Wunder E., 1994, BAV Mitt. 68 Hübscher J., Agerer F., Hassforther B., Wunder E., 1997, BAV Mitt. 101 Hübscher J., Agerer F., Hassforther B., Wunder E., 1998, BAV Mitt. 113 Hübscher J., Agerer F., Wunder E., 1991, BAV Mitt. 59 Hübscher J., Agerer F., Wunder E., 1992, BAV Mitt. 60 Hübscher J., Agerer F., Wunder E., 1993, BAV Mitt. 62 Hübscher J., Agerer F., Wunder E., 1995, BAV Mitt. 79 Hübscher J., Lichtenknecker D., 1988, BAV Mitt. 50 Hübscher J., Lichtenknecker D., Mundry E., 1985, BAV Mitt. 39 Hübscher J., Lichtenknecker D., Meyer J., 1986, BAV Mitt. 43 Hübscher J., Lichtenknecker D., Wunder E., 1989, BAV Mitt. 52 Hübscher J., Lichtenknecker D., Wunder E., 1990, BAV Mitt. 56 Hübscher, J., Mundry E., 1984, BAV Mitt. 38 Hübscher J., Paschke A., Walter F., 2006, IBVS 5731, (BAV Mitteilungen No. 178) Hübscher J., Steinbach H.-M., Walter F., Vohla F., 2008, Open European Journal on Variable Stars 97 (BAV Mitteilungen No. 202) Hübscher J., Steinbach H.-M., Walter F., 2009, IBVS 5874, (BAV Mitteilungen No. 201) Hübscher J., Steinbach H.-M., Vohla F., Walter F., 2010, Open European Journal on Variable Stars 120 (BAV Mitteilungen No. 204) Kasimirskij N.B., 1981, Astronomical Circular 1204, 7 Kukarkin B.V., 1928, Variable Stars 1, num. 1 Kukarkin B.V., 1930, Variable Stars 3, 20 Kukarkin B.V., 1934, Variable Stars 4, 269 Kusakin A.V., 1977, Variable Stars Supp. 3, 337 Lange G.A., Motrich V.D., Firmanyuk D.N., Tsesevich, V.P., 1976, Astronomical Circular 900, 5 Lange G.A., 1967, Astronomical Circular 436, 4 Lange G.A., 1969, Astronomical Circular 503, 7 Latyshev I.N., 1969, Variable Stars 17, 64 Le Borgne J.F., 2004, GEOS Circular RR22 Le Borgne J.F., Vandenbroere J., Henden A.A., Butterworth N., Dvorak S., 2008, IBVS 5854 Mironov A.V., Moshkliov V.G., Kolykhalova O.M., 1981, Variable Stars Supp. 4, 7 Murnikova V.P., 1977, Variable Stars Supp. 3, 221 Murnikova V.P., 1981, Variable Stars Supp. 4, 1 Preston G. W., Smak J., Paczynski B., 1965, ApJS, 12, 99 Romanov Y.S., Fedotov Y.T., Movchan A.I., 1982, Astronomical Circular 1205, 4 Samolyk G., 2011, eJAAVSO 39, 1, 23 Samolyk G., 2013, eJAAVSO 41, 1, 85 Schoneich W., Lange D., 1978, IBVS 1557 Siegel M. J., 1982, PASP 94, 122 Szeidl B., Guinan E.F., Olah K., Szabados L., 1997, Commmunications of the Konkoly Observatory, Hungary, 99, 1 Vandenbroere J., 1997, GEOS Circular RR14 Vandenbroere J., 1998, GEOS Circular RR15 Vandenbroere J., 2001, GEOS Circular RR17 Vandenbroere J., 2003, GEOS Circular RR19 Vandenbroere J., 2005, GEOS Circular RR23 Vandenbroere J., 2006, GEOS Circular RR27 Vandenbroere J., Denoux E., 2007, GEOS Circular RR30 Vandenbroere J., Dumont M., 2008, GEOS Circular RR32 Vandenbroere J., Ferrand S., 2013, GEOS Circular RR50 Vandenbroere J., Hambsch J., 2009, GEOS Circular RR40 Vandenbroere J., Hambsch J., 2011, GEOS Circular RR47 Vandenbroere J., Kleidis S., 2008, GEOS Circular RR36 Vandenbroere J., Le Borgne J.F., 2012, GEOS Circular RR49 Vandenbroere J., Salmon G., 2009, GEOS Circular RR38 Vandenbroere J., Salmon G., 2010, GEOS Circular RR45 Wendell O.C., 1909, HA 69, part I, 45 Wendell O.C., 1914, HA 69, part II, 124 Wenzel W., Editor, 1984, MVS 10, 4, 104 Wenzel W., Editor, 1985, MVS 10, 7, 163 Yilmaz M., Basturk O., Alan N. et al., 2009, IBVS 5887
Historical maximum list
=======================
Table \[maxima\] lists the times of maximum of RR Lyr used to determine the historical linear ephemeris.\
The columns give the following:\
HJD = Heliocenric Julian Day\
Uncert. = Estimated uncertainty on the time of maximum\
O-C : Observed time of maximum minus calculated time of maximum\
E : Cycle number used in the linear ephemeris to obtain the calculated time of maximum\
Reference : Paper where the tabulated time of maximum is reported\
Observer: Name of the observer, if specified in the Reference\
Method of observation: visual (vis), photographic (pg), pe (photolectric), ccd (CCD), dslr (digital photographic camera)\
Comment : supplementary information.\
The complete table is available online as supporting information.
-------------- ------ --------- ------ ---------------------------- ---------------- ----- --
2414921.6750 0.01 -0.0996 0 Wendell,1909, Wendell,1914 O.C. Wendell vis
2414925.6350 0.01 -0.1074 7 Wendell,1909, Wendell,1914 O.C. Wendell vis
2414938.6410 -0.1387 30 Wendell,1909, Wendell,1914 O.C. Wendell vis
2414984.5600 -0.1334 111 Wendell,1909, Wendell,1914 O.C. Wendell vis
2415184.6460 -0.1403 464 Wendell,1909, Wendell,1914 O.C. Wendell vis
2418919.4580 0.01 -0.2082 7053 Hertzsprung,1922 E. Hertzsprung pg
2418944.4270 0.01 -0.1800 7097 Hertzsprung,1922 E. Hertzsprung pg
-------------- ------ --------- ------ ---------------------------- ---------------- ----- --
: lists the times of maximum of RR Lyr used to determine the historical linear ephemeris.[]{data-label="maxima"}
\[lastpage\]
[^1]: E-mail: [email protected]
[^2]: http://rr-lyr.irap.omp.eu/dbrr/dbrr-V1.0\_08.php?RR%20Lyr
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In stochastic portfolio theory, a relative arbitrage is an equity portfolio which is guaranteed to outperform a benchmark portfolio over a finite horizon. When the market is diverse and sufficiently volatile, and the benchmark is the market or a buy-and-hold portfolio, functionally generated portfolios introduced by Fernholz provide a systematic way of constructing relative arbitrages. In this paper we show that if the market portfolio is replaced by the equal or entropy weighted portfolio among many others, no relative arbitrages can be constructed under the same conditions using functionally generated portfolios. We also introduce and study a shaped-constrained optimization problem for functionally generated portfolios in the spirit of maximum likelihood estimation of a log-concave density.'
address: |
Department of Mathematics\
University of Washington\
Seattle, WA 98195
author:
- 'Ting-Kam Leonard Wong'
bibliography:
- 'infogeo.bib'
title: Optimization of relative arbitrage
---
\[section\] \[theorem\][Lemma]{} \[theorem\][Proposition]{} \[theorem\][Corollary]{}
\[theorem\][Definition]{} \[theorem\][Assumption]{} \[theorem\][Notation]{} \[theorem\][Problem]{}
\[theorem\][Remark]{} \[theorem\][Example]{} \[theorem\][Claim]{}
[^1]
Introduction {#sec:intro}
============
A major aim of stochastic portfolio theory (see [@F02] and [@FKSurvey] for an introduction) is to uncover relative arbitrage opportunities under minimal and realistic assumptions on the behavior of equity markets. Consider an equity market with $n$ stocks. The market weight $\mu_i(t)$ of stock $i$ at time $t$ is the market capitalization of stock $i$ divided by the total capitalization of the market. The vector $\mu(t) = (\mu_1(t), ..., \mu_n(t))$ of market weights takes value in the open unit simplex $\Delta^{(n)}$ in ${\Bbb R}^n$ defined by $$\Delta^{(n)} = \left\{p = (p_1, ..., p_n): p_i > 0, \sum_{i = 1}^n p_i = 1 \right\}.$$ For each $t$, the portfolio manager chooses a portfolio vector in $\overline{\Delta^{(n)}}$, where $\overline{\Delta^{(n)}}$ is the closure of $\Delta^{(n)}$. Its components represent the proportions of the current capital invested in each of the stocks. We assume that the portfolio is self-financing and all-long, so short selling is prohibited. The [*market portfolio*]{} is the portfolio whose portfolio weight at time $t$ is $\mu(t)$. It is a buy-and-hold portfolio since no trading is required after its installment. In general trading is required to maintain the target portfolio weights. A [*relative arbitrage*]{} with respect to the market portfolio over the horizon $[0, t_0]$ is a portfolio which is guaranteed to outperform the market portfolio at time $t_0$.
We say that the market is [*diverse*]{} if $\max_{1 \leq i \leq n} \mu_i(t) \leq 1 - \delta$ for some $\delta > 0$ and for all $t$, or more generally if $\mu(t) \in K$ for all $t$ where $K$ is an appropriate subset of $\Delta^{(n)}$. The market is [*sufficiently volatile*]{} if the cumulated volatility of the market weight grows to infinity in a suitable sense. Assuming the market is diverse and sufficiently volatile, it is possible to construct relative arbitrages with respect to the market portfolio over a finite (but possibly long) horizon; see for example [@FKSurvey], [@PW13] and the references therein. In fact, it is possible to construct relative arbitrages whose portfolio weights are deterministic functions of the current market weights. In particular, forecasts of expected returns and the covariance matrix are not required. These portfolios, first introduced in [@F99], are said to be [*functionally generated*]{}. This is in accordance with the observation by many academics and practitioners (see for example [@FGH98], [@DGU09] and [@PPA12]) that simple portfolio rules such as the equal and diversity weighted portfolios often beat the market over long periods. Intuitively, these portfolios work by capturing market volatility while controlling the maximum drawdown relative to the market portfolio (the main ideas will be reviewed in Section \[sec:Fernholz\]). In [@PW14] we proved the converse: a relative arbitrage portfolio (more precisely a [*pseudo-arbitrage*]{}, see below) depending deterministically on the current market weights must be functionally generated. We emphasize that a relative arbitrage portfolio is supposed to perform well [*for all*]{} possible realizations of the market weight process satisfying diversity and sufficient volatility. This observation is utilized in [@PW14] to allow a geometric, pathwise approach without assuming any stochastic model for the market weight process.
There are two important questions that are not fully addressed by the existing theory. First, what happens if the market portfolio is replaced by another benchmark? In [@S12] the concept of functionally generated portfolio and the key ‘master equation’ (see Lemma \[lem:FernholzDecomp\] below) are extended to arbitrary benchmark portfolios. However, little is known about the existence of relative arbitrage under general conditions such as diversity and sufficient volatiltiy. For example, can we beat the equal-weighted portfolio by a functionally generated portfolio in a diverse and sufficiently volatile market, in the same way a functionally generated portfolio beats the market portfolio? More generally, does there exist an infinite hierarchy of relative arbitrages? Is there a ‘maximal portfolio’ which cannot be beaten if only diversity and sufficient volatility are assumed?
Second, is there a sound and applicable optimization theory for relative arbitrages and functionally generated portfolios? Such a theory is clearly of great interest and this problem was raised already in Fernholz’s monograph [@F02 Problems 3.1.7-8]. To the best of our knowledge limited progress has been made to optimization of functionally generated portfolios. See [@PW13] for an attempt in the two asset case and [@PW14] for an approach using optimal transport. On the theoretical side, if the market model is given it is sometimes possible to characterize the highest return relative to the market or a given trading strategy that can be achieved using nonanticipative investment rules over a given time horizon. See [@FK10] for the case of Markovian markets, [@FK11] for a more general setting which allows uncertainty regarding the drift and diffusion coefficients, and [@R11] which expresses optimal relative arbitrages with respect to Markovian trading strategies as delta hedges. For optimization of functionally generated portfolios, a major difficulty is that the class of functionally generated portfolios is a function space and the optimization has to be nonparametric. Ideally, given historical data or a stochastic model of the market weight process, we want to pick an optimal functionally generated portfolio subject to appropriate constraints.
The present paper attempts to give answers to both questions. In this paper we interpret relative arbitrage by what we call [*pseudo-arbitrage*]{} in [@PW13]. This is a model-free concept and the precise definition will be stated in Section \[sec:prelim\]. We only consider portfolios which are deterministic functions of the current market weight, so a portfolio is represented by a map $\pi: \Delta^{(n)} \rightarrow \overline{\Delta^{(n)}}$. This means that the portfolio manager always chooses $\pi(p)$ when the current market weight is $\mu(t) = p \in \Delta^{(n)}$, regardless of previous price movements. Following [@PW14], in this paper time is discrete and the market is represented by a deterministic sequence $\{\mu(t)\}_{t = 0}^{\infty}$ with state space $\Delta^{(n)}$. No underlying probability space is required.
Regarding the hierarchy of relative arbitrages, we first define a partial order among portfolios. If $\pi$ is a portfolio, we let $V_{\pi}(t)$ be the ratio of the growth of $\$1$ invested in the portfolio to that of $\$1$ invested in the market portfolio, and call it the [*relative value process*]{}. Let $\pi, \tau: \Delta^{(n)} \rightarrow \overline{\Delta^{(n)}}$ be portfolios. We say that $\tau$ [*dominates $\pi$ on compacts*]{} (written $\tau \succeq \pi$) if for any compact set $K \subset \Delta^{(n)}$, there exists a constant $\varepsilon = \varepsilon(\pi, \tau, K) > 0$ such that $V_{\tau}(t) / V_{\pi}(t) \geq \varepsilon$ for all $t$ and for all sequences of market weight $\{\mu(t)\}_{t = 0}^{\infty}$ taking values in $K$. That is, the maximum drawdown of $\tau$ relative to $\pi$ is uniformly bounded regardless of the market movement in that region. Since the compact set $K$ is arbitrary, this is a global property and defines a partial order among portfolios. If ${\mathcal S}$ is a family of portfolios, we say that a portfolio $\pi \in {\mathcal S}$ is [*maximal*]{} in ${\mathcal S}$ if there is no portfolio, other than $\pi$ itself, which dominates $\pi$ on compacts, i.e., $\tau \in {\mathcal S}$ and $\tau \succeq \pi$ implies $\tau = \pi$. In Section \[subsec:pseudo\] we will relate this partial order with pseudo-arbitrage. Here we note that if $\tau$ is a relative or pseudo-arbitrage with respect to $\pi$ in all diverse and sufficiently volatile markets, it is necessarily the case that $\tau$ dominates $\pi$ on compacts.
Let $\pi: \Delta^{(n)} \rightarrow \overline{\Delta^{(n)}}$ be a portfolio and $\Phi$ be a positive concave function on $\Delta^{(n)}$. We say that $\pi$ is [*functionally generated*]{} with generating function $\Phi$ if for all $p \in \Delta^{(n)}$, the vector of coordinatewise ratios $\pi(p)/p$ defines a supergradient of the concave function $\log \Phi$ at $p$ (see Definition \[def:fgp\] below for the rigorous definition). If $\Phi$ is $C^2$ (twice continuously differentiable), then $\pi$ is necessarily given by $$\label{eqn:fgweight}
\pi_i(p) = p_i \left(1 + D_{e(i) - p} \log \Phi(p) \right), \quad i = 1, ..., n, \quad p \in \Delta^{(n)}.$$ Here $D_{e(i) - p}$ is the directional derivative in the direction $e(i) - p$, where $e(i)$ is the vertex of $\overline{\Delta^{(n)}}$ in the $i$-th direction. For example, the market portfolio is generated by the constant function $\Phi(p) \equiv 1$. We say that $\Phi$ is a [*measure of diversity*]{} if it is $C^2$ and symmetric (invariant under permutations of the coordinates). Let $\overline{e} = \left(\frac{1}{n}, ..., \frac{1}{n}\right)$ be the barycenter of $\Delta^{(n)}$. For portfolios that are continuously differentiable, the following theorem gives a sufficient condition for a portfolio to be maximal.
\[thm:main\] Let $\pi$ be a portfolio generated by a measure of diversity $\Phi$. If $$\label{eqn:integralcondition}
\int_0^1 \frac{1}{\Phi(te(1) + (1 - t)\overline{e})^2} \mathrm{d}t = \infty,$$ then $\pi$ is maximal in the class of portfolios $\tau: \Delta^{(n)} \rightarrow \overline{\Delta^{(n)}}$ that are continuously differentiable.
This sufficient condition is satisfied by the equal and entropy weighted portfolios (see Table \[tab:benchmark\] in Section \[sec:benchmark\] for the definitions) among many others. For the market portfolio the generating function is constant and so the integral in converges. In Section \[sec:benchmark\] we will show if $\pi$ is functionally generated and $\tau$ dominates $\pi$ on compacts, then $\tau$ must be functionally generated. Thus we may rephrase Theorem \[thm:main\] by saying that if holds then $\pi$ is maximal in the family of functionally generated portfolios with $C^2$ generating functions. A consequence of Theorem \[thm:main\] is the following.
\[cor:main\] Under the setting of Theorem \[thm:main\], suppose $\tau$ is a $C^1$ portfolio not equal to $\pi$. Then there is a compact set $K \subset \Delta^{(n)}$ and a market weight sequence $\{\mu(t)\}_{t \geq 0}$ taking values in $K$, such that the portfolio value of $\tau$ relative to $\pi$ tends to zero as $t$ tends to infinity.
One can interpret Corollary \[cor:main\] by saying that if $\pi$ is maximal and $\tau \neq \pi$, it is possible to find a diverse and sufficiently volatile market in which $\pi$ beats $\tau$ in the long run. In this sense, for a portfolio $\pi$ satisfying , it is impossible to find a (deterministic) portfolio which is a relative arbitrage with respect to $\pi$ in all diverse and sufficiently volatile markets. Theorem \[thm:main\] will be proved by comparing the relative concavities of portfolio generating functions.
Regarding optimization of functionally generated portfolios, we formulate a [*shape-constrained optimization problem*]{} in the spirit of maximum likelihood estimation of a log-concave density. For the statistical theory we refer the reader to [@DR09], [@CSS10], [@CS10], [@KM10] and [@SW10]. Following [@PW14], we associate to each functionally generated portfolio an [*L-divergence functional*]{} $T\left(\cdot \mid \cdot\right)$ defined on $\Delta^{(n)} \times \Delta^{(n)}$ (see Definition \[def:discreteenergy\]). Intuitively, $T\left(q \mid p\right)$ measures the potential profit from volatility captured when the market weight jumps from $p$ to $q$ in $\Delta^{(n)}$. Let ${\Bbb P}$ be an intensity measure over the jumps $(p, q)$ which can be defined in terms of data or a given model (examples will be given in Section \[sec:optimization\]). We maximize $$\int T\left( q \mid p \right) \mathrm{d}{\Bbb P}$$ over all functionally generated portfolios with or without constraints. This optimization problem is shape-constrained because the generating function of a functionally generated portfolio is concave. We prove that the optimization problem is well-posed and is in a suitable sense [*consistent*]{} when interpreted as a statistical estimation problem. In this paper we implement this optimization for the case of two assets (analogous to univariate density estimation) and a general algorithm will be the topic of future research. We illustrate a typical application in portfolio management with a case study.
The paper is organized as follows. In Section \[sec:prelim\] we set up the notations and recall the definitions of pseudo-arbitrage and functionally generated portfolio. In Section \[sec:benchmark\] we extend the framework of [@PW14] to benchmark portfolios that are functionally generated. Using a relative concavity lemma given in [@CDO07], we prove Theorem \[thm:main\] and Corollary \[cor:main\] in Section \[sec:concavity\]. Optimization of functionally generated portfolios is studied in Section \[sec:optimization\] and an empirical case study is presented in Section \[sec:empirical\]. Several proofs of a more technical nature are gathered in Appendex \[sec:appendix\].
Pseudo-arbitrage and functionally generated portfolio {#sec:prelim}
=====================================================
Portfolio and pseudo-arbitrage {#subsec:pseudo}
------------------------------
We work under the discrete time, deterministic set-up of [@PW14] which we briefly recall here. Let $n \geq 2$ be the number of stocks or assets in the market. We endow the open unit simplex $\Delta^{(n)}$ with the Euclidean metric. The open ball in $\Delta^{(n)}$ centered at $p$ with radius $\delta$ is denoted by $B(p, \delta)$. A tangent vector of $\Delta^{(n)}$ is a vector $v = (v_1, ..., v_n) \in {\Bbb R}^n$ satisfying $\sum_{i = 1}^n v_i = 0$. We denote the vector space of tangent vectors of $\Delta^{(n)}$ by $T\Delta^{(n)}$. For $i = 1, ..., n$, we let $e(i) = (0, ..., 0, 1, 0, ..., 0)$ be the vertex of $\Delta^{(n)}$ in the $i$-th direction. If $a$ and $b$ are vectors in ${\Bbb R}^n$, we let $\langle a, b \rangle$ be the Euclidean inner product. The Euclidean norm is denoted by $\|\cdot\|$. If $b$ has nonzero entries, $a / b$ is the vector of the componentwise ratios $a_i / b_i$.
Throughout this paper time is discrete ($t = 0, 1, 2, ...$). Extensions to continuous time will be discussed briefly in Section \[sec:continuoustime\]. Let $X_i(t) > 0$ be the market capitalization of stock $i$ at time $t$. The total capitalization of the market is then $X_1(t) + \cdots + X_n(t)$. The market weight of stock $i$ is defined by $$\mu_i(t) = \frac{X_i(t)}{X_1(t) + \cdots + X_n(t)}, \quad i = 1, ..., n.$$ The vector $\mu(t) = (\mu_1(t), ..., \mu_n(t))$ takes values in $\Delta^{(n)}$ and represents the relative sizes of the firms. As the stock prices move the market weights fluctuate accordingly.
As in [@PW14], the stock market is modeled as a deterministic sequence $\{\mu(t)\}_{t \geq 0}$ taking values in $\Delta^{(n)}$, so an underlying probability space is not required. Our approach is analogous to that of universal prediction (see for example [@CL06]) where it is not assumed that the data is generated by a stochastic model. Only structural properties such as diversity and sufficient volatility will be imposed on the sequences.
We consider a small investor in this market who cares about the value of his or her portfolio relative to that of the entire market. We restrict ourselves to portfolios which are deterministic functions of the current market weights. Short sales are not allowed and we assume there is no transaction cost.
A portfolio is a Borel measurable map $\pi: \Delta^{(n)} \rightarrow \overline{\Delta^{(n)}}$. The market portfolio $\mu$ is the identity map $p \mapsto p$ and we do not distinguish it from the market weight process $\{\mu(t)\}$. Given a portfolio $\pi$, its relative value process $\{V_{\pi}(t)\}_{t \geq 0}$ is defined by $V_{\pi}(0) = 1$ and $$\label{eqn:relativevalue}
\frac{V_{\pi}(t+1)}{V_{\pi}(t)} = 1 + \left\langle \frac{\pi(\mu(t))}{\mu(t)}, \mu(t + 1) - \mu(t) \right\rangle, \quad t \geq 0.$$ The weight ratio of the portfolio at $p \in \Delta^{(n)}$ is the vector $\frac{\pi(p)}{p} = \left(\frac{\pi_1(p)}{p_1}, ..., \frac{\pi_n(p)}{p_n}\right)$.
The relative value $V_{\pi}(t)$ can be interpreted as the ratio of the growth of $\$1$ invested in the portfolio to that of $\$1$ invested in the market portfolio. If $V_{\pi}(t_1) > V_{\pi}(t_0)$, the portfolio outperforms the market portfolio over the (discrete) time interval $[t_0, t_1]$. As mentioned in [@PW14], it is helpful to think of the weight ratio $p \mapsto \frac{\pi(p)}{p}$ as a vector field on $\Delta^{(n)}$. From , the portfolio outperforms the market over $[t, t + 1]$ if the inner product between the displacement $\mu(t + 1) - \mu(t)$ of the market weight and the weight ratio is positive. This means on average the portfolio puts more weight on the assets which perform well relative to the rest of the market.
In the first part of the paper we will study the hierarchy of portfolios defined by the relation ‘domination on compacts’.
\[def:pseudoarbitrage\] Let $\pi$ and $\tau$ be portfolios. We say that $\tau$ dominates $\pi$ on compacts (written $\tau \succeq \pi$) if for any compact subset $K$ of $\Delta^{(n)}$, there exists a constant $C = C(\pi, \tau, K) \geq 0$ such that for any path $\{\mu(t)\}_{t \geq 0} \subset K$, we have $$\label{eqn:lowerbound}
\log\frac{V_{\tau}(t)}{V_{\pi}(t)} \geq -C, \quad t \geq 0.$$
Thus, if $\tau \succeq \pi$, the value of $\pi$ cannot grow at a rate faster than that of $\tau$ under the diversity condition $\mu(t) \in K$, for any compact subset $K$. The relation $\tau \succeq \pi$ defines a partial order among the class of portfolio maps. We include the logarithm in as this formulation is more convenient when we discuss functionally generated portfolios. This definition is closely related to that of pseudo-arbitrage introduced in [@PW14]. The definition given below is extended slightly to allow for an arbitrary benchmark portfolio.
\[def:pseudoarbitrage2\] Let $\pi$ and $\tau$ be portfolios, and $K$ be a subset of $\Delta^{(n)}$, not necessarily compact. We say that $\tau$ is a pseudo-arbitrage with respect to $\pi$ on $K$ if the following properties hold:
1. There exists a constant $C = C(\pi, \tau, K) \geq 0$ such that holds for any sequence $\{\mu(t)\}_{t \geq 0} \subset K$.
2. There exists a sequence $\{\mu(t)\}_{t \geq 0} \subset K$ along which $\lim_{t \rightarrow \infty} \log V(t) = \infty$.
We refer the reader to [@PW14] for more discussion of the definition. Here we note that the requirement $\{\mu(t)\}_{t \geq 0} \subset K$ in (i) is a diversity condition which is portfolio-specific, and (ii) refers to the presence of sufficient volatility. The following is an easy consequence of the definitions.
\[lem:easy\] Let $\pi$ and $\tau$ be portfolios. Suppose $\tau$ is a pseudo-arbitrage relative to $\pi$ on $K_j$ for all $j$, where $\{K_j\}$ is a compact exhaustion of $\Delta^{(n)}$. Then $\tau$ dominates $\pi$ on compacts.
Let ${\mathcal S}$ be a family of portfolios and $\pi \in {\mathcal S}$. We say that $\pi$ is maximal in ${\mathcal S}$ if there is no portfolio in ${\mathcal S}$, other than $\pi$ itself, which dominates $\pi$ on compacts.
Note that a maximal portfolio may not exist and may not be unique in the given class. In Section \[sec:concavity\] we will study the maximal portfolios where ${\mathcal S}$ is the class of portfolios with $C^2$ generating functions. By Lemma \[lem:easy\], if $\pi$ is maximal there is no portfolio which is a pseudo-arbitrage with respect to $\pi$ on all sufficiently large compact subsets of $\Delta^{(n)}$. In this sense a maximal portfolio is one which is impossible to beat assuming only diversity and sufficient volatility.
The relation ‘domination on compacts’ refers to global properties of portfolios. Even if $\pi$ is maximal, for a [*fixed*]{} subset $K \subset \Delta^{(n)}$ it may be possible to find a portfolio $\tau$ (depending on $K$) which beats $\pi$ in the long run whenever $\{\mu(t)\} \subset K$. For example, when $n = 2$, it can be shown that the entropy-weighted portfolio beats the equal-weighted portfolio in the long run if $\{\mu(t)\}$ is sufficiently volatile and stays in a certain neighborhood of $\left(\frac{1}{2}, \frac{1}{2}\right)$. This, however, requires that $K$ is known in advance. Maximality of $\pi$ requires that there is no [*single*]{} $\tau$ which beats $\pi$ on [*all*]{} compact sets $K \subset \Delta^{(n)}$.
Functionally generated portfolio {#sec:fgp}
--------------------------------
Functionally generated portfolio was first introduced in a general form in [@F99]. We will follow the intrinsic treatment in [@PW14 Section 2] which emphasizes the relationship with convex analysis. Throughout the paper we will rely heavily on results from convex analysis and a standard reference is [@R70].
\[Functionally generated portfolios\] \[def:fgp\] [ ]{} Let $\pi$ be a portfolio and $\Phi: \Delta^{(n)} \rightarrow (0, \infty)$ be a concave function. We say that $\pi$ is generated by $\Phi$ if the inequality $$\label{eqn:superdiff}
1 + \left\langle \frac{\pi(p)}{p}, q - p \right\rangle \geq \frac{\Phi(q)}{\Phi(p)}$$ holds for all $p, q \in \Delta^{(n)}$. We call $\Phi$ the generating function of $\pi$. We denote by ${\mathcal{FG}}$ the collection of all functionally generated portfolios $(\pi, \Phi)$ where $\pi$ is generated by the concave function $\Phi$.
It is known (see [@PW14 Proposition 5]) that the generating function is unique up to a positive multiplicative constant, so the use of ‘the’ in the above definition is justified (up to the constant). On the other hand, by Lemma \[lem:superdiff\](ii) below a non-smooth concave function $\Phi$ generates multiple portfolios but they differ only on the set where $\Phi$ is not differentiable (i.e., the superdifferential $\partial \log \Phi(p)$ has more than one element), and this set has Lebesgue measure zero (relative to $\Delta^{(n)}$) by [@R70 Theorem 25.5]. Note that here the generating function is concave by definition, while in [@F02] non-concave generating functions are allowed. See Theorem \[thm:PW14\] and Proposition \[prop:MCMfgp\] below for a justification of our definition.
Let $\Phi$ be a concave function on $\Delta^{(n)}$ and $p \in \Delta^{(n)}$. The [*superdifferential*]{} of $\Phi$ at $p$ is the set $\partial \Phi(p)$ defined by $$\label{eqn:superdiffdef}
\partial \Phi(p) = \{\xi \in T\Delta^{(n)}: \Phi(p) + \langle \xi, q - p \rangle \geq \Phi(q) \ \forall q \in \Delta^{(n)}\}.$$ If $\Phi$ is concave and positive, it can be shown that $\log \Phi$ is also a concave function, and $$\label{eqn:supdiffequal}
\partial \log \Phi(p) = \frac{1}{\Phi(p)} \partial \Phi(p) = \left\{\frac{1}{\Phi(p)} \xi: \xi \in \partial \Phi(p)\right\}.$$
[@PW14 Proposition 6] \[lem:superdiff\] Let $\Phi$ be a positive concave function on $\Delta^{(n)}$.
1. Let $\pi$ be a portfolio generated by $\Phi$. Then for $p \in \Delta^{(n)}$, the tangent vector $v = (v_1, ..., v_n)$ defined by $$\label{eqn:definev}
v_i = \frac{\pi_i(p)}{p_i} - \frac{1}{n} \sum_{j = 1}^n \frac{\pi_j(p)}{p_j}, \quad i = 1, ..., n,$$ belongs to $\partial \log \Phi(p)$.
2. Conversely, if $v \in \partial \log\Phi(p)$, then the vector $\pi = (\pi_1, ..., \pi_n)$ defined by $$\label{eqn:definepi}
\frac{\pi_i}{p_i} = v_i + 1 - \sum_{j = 1}^n p_jv_j, \quad i = 1, ..., n,$$ is an element of $\overline{\Delta^{(n)}}$. In particular, any measurable selection of $\partial \log \Phi$ (a Borel measurable map $\xi: \Delta^{(n)} \rightarrow T\Delta^{(n)}$ such that $\xi(p) \in \partial \log \Phi(p)$ for all $p \in \Delta^{(n)}$) defines via a portfolio generated by $\Phi$. (By [@RW98 Theorem 14.56], there is always a measurable selection of $\partial \log \Phi$.)
Moreover, the operations $\pi \mapsto v$ and $v \mapsto \pi$ defined by and are inverses of each other.
From , it can be seen that Fernholz’s definition (see [@F02 Theorem 3.1.5]) is consistent with ours. If $\pi$ is generated by $\Phi$, the weight ratio vector field $\frac{\pi}{p}$ is [*conservative*]{} on $\Delta^{(n)}$ and its potential function is given by the logarithm of the generating function $\Phi$. Here is a precise statement and the details can be found in the proof of [@PW14 Theorem 8]. Let $\pi$ be a portfolio. If $\gamma: [0, 1] \rightarrow \Delta^{(n)}$ is a piecewise linear path in $\Delta^{(n)}$, we let $$\label{eqn:lineintegral}
I_{\pi}(\gamma) := \int_{\gamma} \frac{\pi}{p} \mathrm{d}p \equiv \int_0^1 \sum_{i = 1}^n \frac{\pi_i(\gamma(t))}{p_i(\gamma(t))}\gamma'_i(t)\mathrm{d}t$$ be the line integral of the weight ratio along $\gamma$. If $\pi$ is functionally generated, the weight ratio $\frac{\pi}{p}$ is conservative in the sense that this line integral is zero whenever $\gamma$ is closed, i.e., $\gamma(0) = \gamma(1)$. Moreover, for any $p, q \in \Delta^{(n)}$ we have $$\label{eqn:lineintegral2}
\log \Phi(q) - \log \Phi(p) = I_{\pi}(\gamma),$$ where $\gamma$ is any piecewise linear path from $p$ to $q$. In classical terminology, $\log \Phi$ is then the potential function of the weight ratio vector field. Fernholz’s decomposition (see Lemma \[lem:FernholzDecomp\] below) shows that the log relative value $\log V_{\pi}(t)$ can be decomposed as the sum of the increment of $\log \Phi(\mu(t))$ and a non-decreasing process related to market volatility.
The concavity of the generating function will be measured in terms of the L-divergence introduced in [@PW14].
\[def:discreteenergy\] Let $\pi$ be a portfolio generated by a concave function $\Phi: \Delta^{(n)} \rightarrow (0, \infty)$. The L-divergence functional of the pair $(\pi, \Phi)$ is the function $T: \Delta^{(n)} \times \Delta^{(n)} \rightarrow [0, \infty)$ defined by $$\label{eqn:discreteenergy}
T\left(q \mid p \right) = \log \left(1 + \left\langle \frac{\pi(p)}{p}, q - p \right\rangle \right) - \log \frac{\Phi(q)}{\Phi(p)}, \quad p, q \in \Delta^{(n)}.$$
Using , it can be shown that $T\left(q \mid p \right) \geq 0$ and $T\left(q \mid p \right) = 0$ only if $\Phi$ is affine on the line segment containing $p$ and $q$. $T\left(\cdot \mid \cdot\right)$ is a logarithmic version (hence the ‘L’) of [*Bergman divergence*]{} used in information geometry (see [@AC10]) and should be thought of as a measure of the concavity of $\Phi$.
With these definitions, the main results of [@PW14] can be summarized as follow.
[@PW14 Theorem 1, Theorem 2] \[thm:PW14\] A portfolio $\pi$ is a pseudo-arbitrage relative to the market portfolio $\mu$ on a convex subset $K \subset \Delta^{(n)}$ if and only if $\pi$ is generated by a concave function $\Phi: \Delta^{(n)} \rightarrow (0, \infty)$ which is bounded below on $K$ and $T \left( \cdot \mid \cdot \right)$ is not identically zero on $K \times K$. Moreover, these portfolios correspond to solutions of an optimal transport problem.
In Section \[sec:concavity\] we will focus on functionally generated portfolios with $C^2$ generating functions.
\[def:fg\] \[defn:C2fgp\] [ ]{}
1. We denote by ${\mathcal{FG}}^2$ the collection of functionally generated portfolios whose generating functions are $C^2$ and concave. An element of ${\mathcal{FG}}^2$ is denoted by either $\pi$, $\Phi$ or $(\pi, \Phi)$ where $\pi$ is generated by $\Phi$. In this case $\pi$ is necessarily given by .
2. A positive $C^2$ concave function $\Phi$ on $\Delta^{(n)}$ is called a measure of diversity if it is symmetric, i.e., $$\Phi(p_1, ..., p_n) = \Phi(p_{\sigma(1)}, ..., p_{\sigma(n)})$$ for all $p \in \Delta^{(n)}$ and any permutation $\sigma$ of $\{1, ..., n\}$.
Measure of diversity was introduced by Fernholz in [@F99 Section 4]. Some examples are given in Table \[tab:benchmark\] and more can be found in [@F02 Section 3.4]. A measure of diversity gives a numerical measure of the concentration of the capital distribution $\mu(t) = \left(\mu_1(t), ..., \mu_n(t)\right)$ and also generates a portfolio.
Benchmarking a functionally generated portfolio {#sec:benchmark}
===============================================
Fix a portfolio $\pi$ generated by a concave function $\Phi: \Delta^{(n)} \rightarrow (0, \infty)$ and call it the [*benchmark portfolio*]{}. Some examples we have in mind are given in Table \[tab:benchmark\]. All of these portfolios are generated by measures of diversity.
[lll]{} Name & Portfolio weights & Generating function\
Market & $\pi_i(p) = p_i$ & $\Phi(p) = 1$\
Diversity-weighted ($0 < r < 1$) & $\pi_i(p) = \frac{p_i^r}{\sum_{j = 1}^n p_j^r}$ & $\Phi(p) = \left( \sum_{j = 1}^n p_j^r \right)^{\frac{1}{r}}$\
Equal-weighted & $\pi_i(p) = \frac{1}{n}$ & $\Phi(p) = \left(p_1 p_2 \cdots p_n \right)^{\frac{1}{n}}$\
Entropy-weighted & $\pi_i(p) = \frac{-p_i \log p_i}{\sum_{j = 1}^n -p_j \log p_j}$ & $\Phi(p) = \sum_{j = 1}^n -p_j \log p_j$\
As mentioned in the introduction, it can be proved that many functionally generated portfolios (including the three nontrivial examples above) outperform the market over sufficiently long periods under the assumptions of diversity and sufficient volatility. As these hypotheses appear to hold empirically, many functionally generated portfolios outperform the market over long periods. See [@F02 Chapter 6] for several case studies using data of the US stock market. Since these portfolios contain no proprietary modeling, behave reasonably well and are easily replicable, they also serve as alternative benchmarks as discussed in practitioner papers such as [@FGH98] and [@HCKL11]. It is natural to ask whether we can construct relative or pseudo-arbitrages with respect to these portfolios.
Fernholz’s decomposition {#sec:Fernholz}
------------------------
The relative value process of a functionally generated portfolio satisfies an elegant decomposition formula. It is a direct consequence of and and can be motivated by the vector field interpretation discussed in Section \[sec:fgp\].
\[lem:FernholzDecomp\] [@F99 Theorem 3.1] [@PW14 Lemma 7] If $\pi$ is generated by a concave function $\Phi$, the relative value process $V_{\pi}$ has the decomposition $$\label{eqn:FernholzDecomp}
\log V_{\pi}(t) = \log \frac{\Phi(\mu(t))}{\Phi(\mu(0))} + A(t),$$ where $A(t) = \sum_{k = 0}^{t-1} T\left(\mu(k+1) \mid \mu(k)\right)$ is non-decreasing. We call $A(t)$ the drift process of the portfolio.
![Hypothetical performance of a functionally generated portfolio. If the market weight $\mu(t)$ stays within a subset $K \subset \Delta^{(n)}$, the relative value process will stay within the dashed curves which are vertical translations of the drift process $A(t)$. The width of the ‘sausage’ is given by the oscillation of $\log \Phi$ on $K$ defined by ${{\mathrm{osc}}}_K(\log \Phi) = \sup_{p, q \in K} |\log \Phi(q) - \log \Phi(p)|$.[]{data-label="fig:FernholzDecomp"}](fernholz.pdf)
The key idea of the decomposition is that over any period $[t_0, t_1]$ where $\log \Phi(\mu(t_1))$ and $\log \Phi(\mu(t_0))$ are approximately equal, the portfolio will outperform the market by an amount equal to $A(t_1) - A(t_0)$, see Figure \[fig:FernholzDecomp\] for an illustration. For this reason, the drift process $A(t)$ can be thought of as the cumulative amount of market volatility captured by the portfolio. The condition of sufficient volatility requires that $A(t)$ grows unbounded as $t \rightarrow \infty$. Empirical studies (see for example [@FKSurvey Figure 11.2]) show that $A$ increases at a roughly linear rate depending on the portfolio and market volatility. Thus, as long as the fluctuation of $\log \Phi(\mu(t))$ remains bounded, the drift process will dominate in the long run and the portfolio will outperform the market. The assumption on diversity is imposed to bound $\log \Phi(\mu(t))$. For (say) the entropy-weighted portfolio, $\log \Phi(\mu(t))$ is bounded as long as $\max_{1 \leq i \leq n} \mu_i(t) \leq 1 - \delta$ for some $\delta > 0$, so we can take $K$ in Definition \[def:pseudoarbitrage2\] and Theorem \[thm:PW14\] to be the set $\{p \in \Delta^{(n)}: \max_{1 \leq i \leq n} p_i \leq 1 - \delta\}$ (this is the definition of diversity stated in [@F99] and [@FKSurvey]). For other portfolios such as the equal-weighted portfolio, this condition is not enough and we require that $\mu(t)$ stays within a compact subset of $\Delta^{(n)}$. Thus the set $K$ is portfolio-specific. Fernholz’s decomposition is implemented in the `R` package `RelValAnalysis` (available on `CRAN`) written by the author.
Domination on compacts
----------------------
In [@PW14] pseudo-arbitrages with respect to the market portfolio are characterized in terms of a property called [*multiplicative cyclical monotonicity*]{} (MCM). It is a variant of cyclical monotonicity in convex analysis (see [@R70 Section 24]) and is equivalent to $c$-cyclical monotonicity in optimal transport for a special cost function. Intuitively, this property requires that the portfolio outperforms the market portfolio whenever the market weight goes through a cycle. It is natural to extend the definition as follow.
Let $\pi$ and $\tau$ be portfolios. We say that $\tau$ satisfies multiplicative cyclical monotonicity relative to $\pi$ if over any discrete cycle $$\mu(0), \mu(1), ..., \mu(m), \mu(m+1) = \mu(0)$$ in $\Delta^{(n)}$, we have $$\label{eqn:rmcm}
V_{\tau}(m + 1) \geq V_{\pi}(m + 1).$$
In [@PW14] we proved that functionally generated portfolios are characterized by the MCM property relative to the market portfolio.
[@PW14 Proposition 4] \[prop:MCMfgp\] A portfolio satisfies MCM relative to the market portfolio if and only if it is generated by a positive concave function.
For an arbitrary functionally generated benchmark portfolio, we can generalize Proposition \[prop:MCMfgp\] as follow. This result provides equivalent formulations of the partial order $\succeq$ that are easier to work with. The proof is analogous to those of Proposition 4 and Theorem 1 of [@PW14].
\[prop:MCM\] Let $\pi$ be a portfolio generated by a concave function $\Phi: \Delta^{(n)} \rightarrow (0, \infty)$, and let $\tau$ be a portfolio. The following statements are equivalent.
1. $\tau$ dominates $\pi$ on compacts, i.e., $\tau \succeq \pi$.
2. $\tau$ satisfies MCM relative to $\pi$.
3. $\tau$ is generated by a concave function $\Psi$, and the L-divergence $T_{\tau}\left(\cdot\mid \cdot\right)$ of $(\tau, \Psi)$ dominates $T_{\pi}\left(\cdot\mid \cdot\right)$ of $(\pi, \Phi)$ in the sense that $$\label{eq:divergenceineq}
T_{\tau}\left(q\mid p\right) \geq T_{\pi}\left(q\mid p\right)$$ for all $p, q \in \Delta^{(n)}$.
\(i) $\Rightarrow$ (ii): Suppose $\tau$ dominates $\pi$ on compacts. If $\tau$ does not satisfy MCM relative to $\pi$, we can find a discrete cycle $\{\mu(t)\}_{t = 0}^{m + 1}$ such that $\eta := V_{\tau}(m + 1) / V_{\pi}(m + 1) < 1$. Consider the market weight sequence which goes over this cycle again and again, i.e., $\mu(t) = \mu(t + (m+1))$ for all $t$. Then $$\frac{V_{\tau}(k(m + 1))}{V_{\pi}(k(m + 1))} = \eta^k$$ for all $k \geq 0$ and the ratio tends to $0$ as $k \rightarrow \infty$. This contradicts the hypothesis $\tau \succeq \pi$. Thus if $\tau$ dominates $\pi$ on compacts then $\tau$ satisfies MCM relative to $\pi$.
\(ii) $\Rightarrow$ (iii): Suppose $\tau$ satisfies MCM relative to $\pi$. Since $V_{\mu}(\cdot) \equiv 1$ and $\pi$ satisfies MCM relative to the market portfolio (by Proposition \[prop:MCMfgp\]), $\tau$ satisfies MCM relative to the market portfolio as well. By Proposition \[prop:MCMfgp\] again $\tau$ has a generating function $\Psi$. To prove , let $p, q \in \Delta^{(n)}$ with $p \neq q$. Let $\{q = \mu(1), ..., \mu(m), \mu(m+1) = p\}$ be a partition of the line segment $[q, p]$. Then if $\mu(0) = p$, $\{\mu(k)\}_{k = 0}^{m+1}$ is a cycle which starts at $p$, jumps to $q$ and then returns to $p$ along the partition. Then the RMCM inequality implies $$\label{eq:MCM}
\begin{split}
& \left( 1 + \left\langle \frac{\tau(p)}{p}, q - p \right\rangle \right) \prod_{k = 1}^m \left( 1 + \left\langle \frac{\tau(\mu(k))}{\mu(k)}, \mu(k+1) - \mu(k) \right\rangle \right) \\
& \geq \left( 1 + \left\langle \frac{\pi(p)}{p}, q - p \right\rangle \right) \prod_{k = 1}^m \left( 1 + \left\langle \frac{\pi(\mu(k))}{\mu(k)}, \mu(k+1) - \mu(k) \right\rangle \right).
\end{split}$$ Taking log on both sides, we have $$\begin{split}
& \log\left( 1 + \left\langle \frac{\tau(p)}{p}, q - p \right\rangle \right) + \sum_{k = 1}^m \log \left( 1 + \left\langle \frac{\tau(\mu(k))}{\mu(k)}, \mu(k+1) - \mu(k) \right\rangle \right) \\
& \geq \log \left( 1 + \left\langle \frac{\pi(p)}{p}, q - p \right\rangle \right) + \sum_{k = 1}^m \log \left( 1 + \left\langle \frac{\pi(\mu(k))}{\mu(k)}, \mu(k+1) - \mu(k) \right\rangle \right).
\end{split}$$ By the fundamental theorem of calculus for concave function and Taylor approximation, we can choose a sequence of partitions with mesh size going to zero, along which $$\begin{split}
\sum_{k = 1}^m \log \left( 1 + \left\langle \frac{\pi(\mu(k))}{\mu(k)}, \mu(k+1) - \mu(k) \right\rangle \right) &\rightarrow \int_{\gamma} \frac{\pi}{\mu} \mathrm{d}\mu = \log \frac{\Phi(p)}{\Phi(q)},\\
\sum_{k = 1}^m \log \left( 1 + \left\langle \frac{\tau(\mu(k))}{\mu(k)}, \mu(k+1) - \mu(k) \right\rangle \right)&\rightarrow \int_{\gamma} \frac{\tau}{\mu} \mathrm{d}\mu = \log \frac{\Psi(p)}{\Psi(q)},
\end{split}$$ where $\gamma$ is the line segment from $q$ to $p$. Taking the corresponding limit in , we obtain the desired inequality .
\(iii) $\Rightarrow$ (i): Let $\{\mu(t)\}_{t \geq 0}$ be any market weight sequence. By Lemma \[lem:FernholzDecomp\] we can write $$\log \frac{V_{\tau}(t)}{V_{\pi}(t)} = \log \frac{\Psi(\mu(t)) / \Psi(\mu(0))}{\Phi(\mu(t)) / \Phi(\mu(0))} + \left(A_{\tau}(t) - A_{\pi}(t)\right),$$ where $A_{\tau}$ and $A_{\pi}$ are the drift processes of $\tau$ and $\pi$ respectively. By (iii), $A_{\tau}(t) - A_{\pi}(t)$ is non-decreasing in $t$. Since $\log \frac{\Psi(\mu(t)) / \Psi(\mu(0))}{\Phi(\mu(t)) / \Phi(\mu(0))}$ is bounded as long as $\mu(t)$ stays within a compact subset of $\Delta^{(n)}$, $\tau$ dominates $\pi$ on compacts.
Theorem \[prop:MCM\] reduces the study of the partial order $\tau \succeq \pi$ to comparing the relative concavities of generating functions, where concavity is measured by the $L$-divergence. In this paper we focus on generating functions that are twice continuously differentiable. Then the infinitesimal version of leads to second order differential inequalities.
\[def:driftform\] Let $(\pi, \Phi) \in {\mathcal{FG}}^2$. Its drift quadratic form, denoted by both $H_{\pi}$ and $H_{\Phi}$, is defined by $$H_{\pi}(p)(v, v) := \frac{-1}{2\Phi(p)} {{\mathrm{Hess}} \nobreak\hspace{.16667em plus .08333em} }\Phi(p)(v, v), \quad p \in \Delta^{(n)}, v \in T\Delta^{(n)}.$$ Here ${{\mathrm{Hess}} \nobreak\hspace{.16667em plus .08333em} }\Phi$ is the Hessian of $\Phi$ regarded as a quadratic form. By definition, it is given by $$\label{eqn:Hessian}
{{\mathrm{Hess}} \nobreak\hspace{.16667em plus .08333em} }\Phi(p)(v, v) = \left.\frac{\mathrm{d}^2}{\mathrm{d}t^2} \Phi(p + tv) \right|_{t = 0}.$$
\[lem:drift\] Let $(\pi, \Phi), (\tau, \Psi) \in {\mathcal{FG}}^2$, and let $T_{\pi}$ and $T_{\tau}$ be their corresponding L-divergences. If $\tau \succeq \pi$ and therefore $T_{\tau}\left(q \mid p\right) \geq T_{\pi}\left(q \mid p\right)$ for all $p, q \in \Delta^{(n)}$, then $H_{\tau} \geq H_{\pi}$ in the sense that $$\label{eqn:driftineq}
H_{\tau}(p)(v, v) \geq H_{\pi}(p)(v, v)$$ for all $p \in \Delta^{(n)}$ and $v \in T\Delta^{(n)}$.
The lemma follows immediately from the Taylor approximation $$\label{eqn:taylor}
T_{\pi} \left(p + tv \mid p \right) = \frac{-1}{2\Phi(p)}{{\mathrm{Hess}} \nobreak\hspace{.16667em plus .08333em} }\Phi(p)(tv, tv) + o\left(t^2\right).$$ where $p \in \Delta^{(n)}$, $v$ is a tangent vector, and $t \in {\Bbb R}$ is small.
As a consequence of Lemma \[lem:drift\], in order to show that a portfolio $\pi \in {\mathcal{FG}}^2$ is maximal in ${\mathcal{FG}}^2$, it is enough to show that its drift quadratic form $H_{\pi}$ is not dominated (in the sense of ) by that of some other portfolio. This is the approach we use in Section \[sec:concavity\] to prove Theorem \[thm:main\]. Simple examples show, however, that $H_{\tau} \geq H_{\pi}$ does not imply $T_{\tau} \geq T_{\pi}$.
For $0 < r < 1$, the diversity-weighted portfolio $\pi$ introduced at the beginning of this section is generated by the function $$\Phi(p) = \left(\sum_{j = 1}^n p_j^r\right)^{\frac{1}{r}}.$$ It is easy to show that $\Phi$ is bounded below by $1$. Let $\tau$ be the portfolio generated by $\Psi := \Phi - 1$. Then it can be shown that $\tau \succeq \pi$. To see this, write the L-divergence in the form $$\label{eqn:divergence2}
T_{\pi}\left(q \mid p\right) = \log \frac{\Phi(p) + D_{q - p}\Phi(p)}{\Phi(q)}, \quad p, q \in \Delta^{(n)}.$$ Then $$T_{\tau}\left(q \mid p\right) = \log \frac{\left(\Phi(p) - 1\right) + D_{q - p}\Phi(p)}{\Phi(q) - 1} \geq T_{\pi}\left(q \mid p\right).$$ From , we can show that for a portfolio $(\pi, \Phi)$ to be maximal in ${\mathcal{FG}}^2$, it is necessary that the continuous extension of $\Phi$ to the closure $\overline{\Delta^{(n)}}$ (which exists by [@R70 Theorem 10.3]) vanishes at all the vertices $e(1)$, ..., $e(n)$ (because otherwise we can subtract an affine function from $\Phi$ and make $T$ larger). However this condition is not sufficient for $\pi$ to be maximal in ${\mathcal{FG}}^2$.
Relative concavity and maximal portfolios {#sec:concavity}
=========================================
Two asset case
--------------
In this section we study the maximal portfolios in ${\mathcal{FG}}^2$ and prove Theorem \[thm:main\]. To illustrate the ideas involved we first give a proof of the maximality of the equal-weighted portfolio for $n = 2$. This result is the starting point of this paper.
\[prop:equalweight\] For $n = 2$, the equal-weighted portfolio $\pi \equiv \left(\frac{1}{2}, \frac{1}{2}\right)$ generated by the geometric mean $\Phi(p) = \sqrt{p_1p_2}$ is maximal in ${\mathcal{FG}}^2$.
Let $(\tau, \Psi) \in {\mathcal{FG}}^2$ be a portfolio which dominates $(\pi, \Phi)$ on compacts. Define $u(x) = \Phi(x, 1 - x) = \sqrt{x(1 - x)}$ and let $v(x) = \Psi(x, 1 - x)$, $x \in (0, 1)$. Then $u$ and $v$ are positive $C^2$ concave functions on $(0, 1)$. By Theorem \[prop:MCM\] and Lemma \[lem:drift\], the drift quadratic form of $\tau$ dominates that of $\pi$. Using , we have the differential inequality $$\label{eqn:n=2domination}
\frac{-v''(x)}{v(x)} \geq \frac{-u''(x)}{u(x)} = \frac{1}{4\left(x(1 - x)\right)^2}, \quad x \in (0, 1).$$ We claim that $v$ also generates the equal-weighted portfolio, and so $\tau = \pi$.
We will use a transformation which amounts to a change of numéraire using $y = \log \frac{x}{1 - x}$. See the binary tree model in [@PW13 Section 4] for the motivation of this transformation and related results. Define a function $\tau_1: (0, 1) \rightarrow [0, 1]$ by $$\label{eqn:n=2weight}
\tau_1(x) = x + x(1 - x) \frac{v'(x)}{v(x)} = x \left[1 + (1 - x) (\log v)'(x)\right].$$ By , this is the portfolio weight of stock $1$ generated by $v$ and $\tau_1$ takes value in $[0, 1]$. Let $y = \log \frac{x}{1 - x}$, so $x = \frac{e^y}{1 + e^y}$. Define $q: {\Bbb R} \rightarrow [0, 1]$ by $$q(y) = \tau_1(x) = \frac{e^y}{1 + e^y} + \frac{e^y}{(1 + e^y)^2} \frac{v'(x)}{v(x)}, \quad x = \frac{e^y}{1 + e^y}, \quad y \in {\Bbb R}.$$ For the equal-weighted portfolio the corresponding portfolio weight function is identically $\frac{1}{2}$. It follows from a straightforward computation that $$q(y)(1 - q(y)) - q'(y) = \frac{-e^{2y}}{(1 + e^y)^4} \frac{v''(x)}{v(x)}.$$ Now can be rewritten in the form $$\label{eqn:transformeddrift}
q(y)(1 - q(y)) - q'(y) \geq \frac{1}{4}, \quad y \in {\Bbb R}.$$ The proof is then completed by the following elementary result.
\[lem:diffeqn\] Suppose $q: {\Bbb R} \rightarrow [0, 1]$ is differentiable and $q(1 - q) - q' \geq 1/4$ on ${\Bbb R}$. Then $q \equiv 1/2$.
Since $0 \leq q(y) \leq 1$, we have $$q' \leq q(1 - q) - \frac{1}{4} \leq \frac{1}{4} - \frac{1}{4} = 0,$$ so $q$ is non-increasing. If $q(y_0) = q_0 < \frac{1}{2}$ for some $y_0$, then on $y \in [y_0, \infty]$, $q$ must satisfy the differential inequality $$q'(y) \leq q_0(1 - q_0) - \frac{1}{4} < 0,$$ which contradicts the fact that $q(y) \geq 0$. Similarly, if $q(y_0) = q_0 > \frac{1}{2}$ for some $y_0$, the same inequality is satisfied on $(-\infty, y_0]$, again a contradiction. Thus we get $q(y) \equiv \frac{1}{2}$ for all $y \in {\Bbb R}$.
The main idea of the proof of Proposition \[prop:equalweight\] is that for a portfolio to dominate the equal-weighted portfolio $\pi$ on compacts, it must be more aggressive than $\pi$ [*everywhere*]{} on the simplex. This means buying more and more the underperforming stock at a sufficiently fast rate satisfying , but this is impossible to continue up to the boundary of the simplex. While there is a multi-dimensional analogue of the differential inequality (see [@PW14 Theorem 9]), we are unable to extend this proof to the multi-asset case since the market and portfolio weights can move in many directions. Instead, we will work with portfolio generating functions and use the simple but powerful tools of convex analysis.
Main result {#subsec:relative}
-----------
Before we give the proof of Theorem \[thm:main\] we note that the integral condition is sufficient to capture many important examples. The proof is an exercise in elementary calculus and is left to the reader.
The following portfolios satisfy .
1. The equal-weighted portfolio $\pi \equiv \left(\frac{1}{n}, ..., \frac{1}{n}\right)$ generated by the geometric mean $\Phi(p) = \left(p_1 \cdots p_n\right)^{\frac{1}{n}}$.
2. The entropy-weighted portfolio $\pi_i = -(p_i \log p_i) / \Phi(p)$ generated by the Shannon entropy $\Phi(p) = -\sum_{j = 1}^n p_j \log p_j$.
The main ingredient of the proof of Theorem \[thm:main\] is the following ingenious observation taken from [@CDO07] and [@CDOS09 Lemma 2] (it is called the relative convexity lemma in these references). It can be proved by direct differentiation.
[@CDO07] \[lem:relativeconcavity\] Let $-\infty < a < b \leq \infty$ and $c, C: [a, b) \rightarrow {\Bbb R}$ be continuous. Suppose $u, v: [a, b) \rightarrow (0, \infty)$ are $C^2$ and satisfy the differential equations $$\begin{split}
u''(x) + c(x)u(x) &= 0, \quad x \in [a, b), \\
v''(x) + C(x)v(x) &= 0, \quad x \in [a, b).
\end{split}$$ Define $F: [a, b) \rightarrow [0, \infty)$ by $$F(x) = \int_a^x \frac{1}{u(t)^2} \mathrm{d}t, \quad x \in [a, b).$$ Let $G$ be the inverse of $F$ defined on $[0, \ell)$, where $\ell = \lim_{x \uparrow b} F(x)$. Then the function $$w(y) := \frac{v(G(y))}{u(G(y))}$$ defined on $[0, \ell)$ satisfies the differential equation $$w''(y) = -(C(x) - c(x))u(x)^4w(y), \quad 0 \leq y < \ell, \quad x = G(y).$$ In particular, if $C(x) \geq c(x)$ on $[a, b)$, then $w$ is concave on $[0, \ell)$.
We also need some convex analytic properties of functionally generated portfolios.
\[lem:FGconvex\] Let $\pi^{(1)}, \pi^{(2)} \in {\mathcal{FG}}$ be generated by $\Phi^{(1)}$ and $\Phi^{(2)}$ respectively, and $\lambda \in [0, 1]$. Then the portfolio given by the weighted average $$\pi := \lambda \pi^{(1)} + (1 - \lambda) \pi^{(2)}$$ belongs to ${\mathcal{FG}}$. Indeed, $\pi$ is generated by the geometric mean $$\Phi := \left( \Phi^{(1)} \right)^{\lambda} \left( \Phi^{(2)} \right)^{1 - \lambda}$$ of the two generating functions.
For $C^2$ generating functions this result is stated in [@F02 Page 50]. The same is true in the general case where the generating functions are not necessarily smooth. To prove this, we need to check that $\pi = \lambda \pi^{(1)} + (1 - \lambda) \pi^{(2)}$ satisfies the defining inequality . This is an easy consequence of the AM-GM inequality and the proof is omitted.
\[lem:Driftconcave\] The L-divergence and the drift quadratic form are concave in the portfolio weights in the following sense. Let $(\pi^{(1)}, \Phi^{(1)}), (\pi^{(2)}, \Phi^{(2)}) \in {\mathcal{FG}}$. For $\lambda \in [0, 1]$, let $\pi = \lambda \pi^{(1)} + (1 - \lambda) \pi^{(2)}$ and let $\Phi = \left(\Phi^{(1)}\right)^{\lambda} \left( \Phi^{(2)} \right)^{1 - \lambda}$ be the generating function of $\pi$. Let $T$, $T^{(1)}$ and $T^{(2)}$ be the L-divergences of $(\pi, \Phi)$, $(\pi^{(1)}, \Phi^{(1)})$ and $(\pi^{(2)}, \Phi^{(2)})$ respectively. Then $$\label{eqn:concavediscrete}
T\left(q \mid p \right) \geq \lambda T^{(1)}\left(q \mid p \right) + (1 - \lambda) T^{(2)}\left(q \mid p \right), \quad p, q \in \Delta^{(n)}.$$ If $\Phi^{(1)}$ and $\Phi^{(2)}$ are $C^2$, then $H_{\pi} \geq \lambda H_{\pi^{(1)}} + (1 - \lambda) H_{\pi^{(2)}}$ in the sense that $$\label{eqn:concavecont}
H_{\pi}(p)(v, v) \geq \lambda H_{\pi^{(1)}}(p)(v, v) + (1 - \lambda) H_{\pi^{(2)}}(p)(v, v)$$ for all $p \in \Delta^{(n)}$ and $v \in T\Delta^{(n)}$.
To prove we write the L-divergence $T\left(q \mid p \right)$ of a functionally generated portfolio $(\pi, \Phi)$ in the form $$T\left(q \mid p \right) = \log \left(1 + \left\langle \frac{\pi(p)}{p}, q - p \right\rangle\right) - I_{\pi}(\gamma),$$ where $I_{\pi}(\gamma) = \int_{\gamma} \frac{\pi}{p} dp$ is the line integral of the weight ratio along the line segment from $p$ to $q$ (see ). Since the line integral is linear in $\pi$ and the logarithm is concave, we see that $T\left(q \mid p \right)$ is concave in $\pi$. The statement for the drift quadratic form follows from the Taylor approximation .
We are now ready to prove Theorem \[thm:main\].
Let $\tau: \Delta^{(n)} \rightarrow \overline{\Delta^{(n)}}$ be a $C^1$ portfolio which dominates $\pi$ on compacts. We want to prove that $\tau = \pi$. By Theorem \[prop:MCM\], $\tau$ is generated by a concave function $\Psi: \Delta^{(n)} \rightarrow (0, \infty)$. Since $\tau$ is $C^1$, by [@PW14 Proposition 5(iii)] $\Psi$ is $C^2$, so $\tau \in {\mathcal{FG}}^2$. Thus we may rephrase Theorem \[thm:main\] by saying that $\pi$ is maximal in ${\mathcal{FG}}^2$.
Let $\Psi$ be a generating function of $\tau$. By scaling, we may assume that $\Psi(\overline{e}) = \Phi(\overline{e})$. We will prove that $\Psi$ equals $\Phi$ identically, so $\Psi$ generates $\pi$ and $\tau = \pi$. We divide the proof into the following steps.
[*Step 1 (Symmetrization).*]{} Let $S_n$ be the set of permutations of $\{1, ..., n\}$. For $\sigma \in S_n$, define $\Psi_{\sigma}$ by relabelling the coordinates, i.e., $$\Psi_{\sigma}(p) = \Psi(p_{\sigma(1)}, ..., p_{\sigma(n)}).$$ Since $\tau \succeq \pi$, by Lemma \[lem:drift\] (and relabeling the coordinates) we have $H_{\Psi_{\sigma}} \geq H_{\Phi_{\sigma}}$ for all $\sigma \in S_n$. But $\Phi$ is a measure of diversity, so $\Phi_{\sigma} = \Phi$ by symmetry and we have $H_{\Psi_{\sigma}} \geq H_{\Phi}$ for all $\sigma \in S_n$. Let $$\widetilde{\Psi} = \prod_{\sigma \in S_n} \left(\Psi_{\sigma}\right)^{\frac{1}{n!}}$$ be the symmetrization of $\Psi$. By Lemma \[lem:FGconvex\], $\widetilde{\Psi}$ generates the symmetrized portfolio $$\widetilde{\tau}(p) = \frac{1}{n!} \sum_{\sigma \in S_n} \tau(p_{\sigma(1)}, ..., p_{\sigma(n)}), \quad p \in \Delta^{(n)}.$$ By Lemma \[lem:Driftconcave\], we have $$\label{eqn:symmetrizedineq}
H_{\widetilde{\Psi}} \geq \frac{1}{n!} \sum_{\sigma \in S_n} H_{\Psi_{\sigma}} \geq H_{\Phi}.$$ Thus $H_{\widetilde{\Psi}} \succeq H_{\Phi}$. Clearly $\widetilde{\Psi}$ is a measure of diversity and by symmetry it achieves its maximum at $\overline{e}$.
[*Step 2 ($\widetilde{\Psi} \leq \Phi$).*]{} We claim that $\widetilde{\Psi} \leq \Phi$ on $\Delta^{(n)}$. Let $p \in \Delta^{(n)}$ and consider the one-dimensional concave functions $$\label{eqn:uandv}
\begin{split}
u(t) &= \Phi((1 - t)\overline{e} + tp) \\
v(t) &= \widetilde{\Psi}((1 - t)\overline{e} + tp)
\end{split}$$ defined on $[0, 1]$. We have $u(0) = v(0)$ and $u'(0) = v'(0) = 0$ since both $\Phi$ and $\widetilde{\Psi}$ achieve their maximums at $\overline{e}$. Since $H_{\widetilde{\Psi}} \geq H_{\Phi}$, we have $$\frac{-v''(t)}{v(t)} \geq \frac{-u''(t)}{u(t)}, \quad t \in [0, 1].$$ By the relative concavity lemma (Lemma \[lem:relativeconcavity\]), $$\label{eqn:w}
w(y) = \frac{v(G(y))}{u(G(y))}$$ is a positive concave function on $[0, \ell]$, where $\ell = \int_0^1 \frac{1}{u(t)^2} \mathrm{d}t$, with $w(0) = 1$ and $w'(0) = 0$ (by the quotient rule). Note that $\ell < \infty$ as $\Phi$ is continuous and positive on the line segment $[\overline{e}, p] \subset \Delta^{(n)}$. Also, it is straightforward to see that in this case the relative concavity lemma can be applied to $[0, \ell]$ instead of $[0, \ell)$. This implies that $w$ is non-increasing and so $w(\ell) = \widetilde{\Psi}(p) / \Phi(p) \leq 1$.
[*Step 3 ($\widetilde{\Psi} \equiv \Phi$).*]{} Let $Z = \{p \in \Delta^{(n)}: \widetilde{\Psi}(p) = \Phi(p)\}$ and we claim that $Z = \Delta^{(n)}$. Here we follow an idea in the proof of [@CDOS09 Theorem 3]. Define $u$ and $v$ on $[0, 1)$ by with $p$ replaced by $e(1)$. Then the function $w$ defined as in is positive and concave on $[0, \infty)$ since the integral in (which defines $\ell = \int_0^1 \frac{1}{u(t)^2} \mathrm{d}t$) diverges. Again $w$ satisfies $w(0) = 1$ and $w'(0) = 0$. But since $w$ is defined on an infinite interval, if $w'(y) < 0$ for some $y$, then $w$ must hit zero as $w'$ is non-increasing by concavity. This contradicts the positivity of $w$, and so $w$ is identically one on $[0, \infty)$. It follows that $\widetilde{\Psi} = \Phi$ on the line segment $[\overline{e}, e(1))$. By symmetry, $Z$ contains the segments $[\overline{e}, e(i))$ for all $i$.
Next we show that the set $Z$ is convex. Let $p, q \in Z$. Again we consider the pair of functions $$\label{eqn:uandvpq}
\begin{split}
u(t) &= \Phi((1 - t)p + tq) \\
v(t) &= \widetilde{\Psi}((1 - t)p + tq)
\end{split}$$ on $[0, 1]$. Let $\widetilde{w}(t) = \frac{v(t)}{u(t)}$, $t \in [0, 1]$. By the relative concavity lemma again, we know that $\widetilde{w}$ is concave after a reparameterization. But $\widetilde{w}(t) \leq 1$ by Step 2 and $\widetilde{w}$ equals one at the endpoints $0$ and $1$. By concavity, $\widetilde{w}$ is identically one on $[0, 1]$. Hence if $Z$ contains $p$ and $q$, it also contains the line segment $[p, q]$. Now $Z$ is a convex set containing $[\overline{e}, e(i))$ for all $i$. It is easy to see that $Z$ is then the simplex $\Delta^{(n)}$. Hence $\widetilde{\Psi}$ equals $\Phi$ identically.
[*Step 4 (Desymmetrization).*]{} We have shown that $\widetilde{\Psi} \equiv \Phi$, and so $H_{\widetilde{\Psi}} = H_{\Phi}$. By , we have $$H_{\Phi} = H_{\widetilde{\Psi}} \geq \frac{1}{n!} \sum_{\sigma \in S_n} H_{\Psi_{\sigma}} \geq H_{\Phi}.$$ Since $H_{\Psi_{\sigma}} \geq H_{\Phi}$ for each $\sigma \in S_n$, we have $H_{\Psi_{\sigma}} = H_{\Phi}$ for all $\sigma$. In particular, taking $\sigma$ to be the identity, we have $H_{\Psi} = H_{\Phi}$. It remains to show that $\Psi$ equals $\Phi$ identically (recall that we assume $\Psi(\overline{e}) = \Phi(\overline{e})$).
Fix $i \in \{1, ..., n\}$ and consider $$\begin{split}
u(t) &= \Phi((1 - t)\overline{e} + te(i)) \\
v(t) &= \Psi((1 - t)\overline{e} + te(i))
\end{split}$$ for $t \in [0, 1)$. By the argument in Step 3, if $\left(\frac{v}{u}\right)'(0) \leq 0$, the integral condition implies that $v / u$ is identically one. So $\left(\frac{v}{u}\right)'(0) \leq 0$ implies $\left(\frac{v}{u}\right)'(0) = 0$. For $\sigma \in S_n$ let $$v_{\sigma}(t) = \Psi((1 - t)\overline{e} + te(\sigma(i))).$$ Since $\widetilde{\Psi} = \Phi$, we have $$\prod_{\sigma \in S_n} \left(\frac{v_{\sigma}(t)}{u(t)}\right)^{\frac{1}{n!}} = 1.$$ Taking logarithm on both sides and differentiating, we see that the average of the derivatives $\left(\frac{v}{u}\right)'(0)$ over $i$ is $0$ (recall that $\Phi$ is symmetric). Since all derivatives are non-negative by the above argument, in fact they are all $0$, and so $\Psi = \Phi$ on $[\overline{e}, e(i))$ for all $i$.
Since the vectors $e(i) - \overline{e}$ span the plane parallel to $\Delta^{(n)}$, the graphs of $\Psi$ and $\Phi$ have the same tangent plane at $\overline{e}$. Since $\Phi$ achieves its maximum at $\overline{e}$, we see that $\Psi$ achieves its maximum at $\overline{e}$ as well. Now we may apply the argument in Steps 2 and 3 to conclude that $\Psi$ equals $\Phi$ identically on $\Delta^{(n)}$. Thus $\tau = \pi$ and we have proved that $\pi$ is maximal in ${\mathcal{FG}}^2$.
Let $\tau$ be a $C^1$ portfolio not equal to $\pi$. By the maximality of $\pi$, it is not the case that $\tau \succeq \pi$. By Theorem \[prop:MCM\], $\tau$ does not satisfy MCM relative to $\pi$. Thus, there is a cycle $\{\mu(t)\}_{t = 0}^{m+1}$ (with $\mu(0) = \mu(m + 1)$) over which $$\label{eqn:badcycle}
\frac{V_{\tau}(m+1)}{V_{\pi}(m+1)} < 1.$$ Consider, as in the proof of Theorem \[prop:MCM\], the market weight sequence which goes through this cycle again and again. Clearly $\{\mu(t)\}_{t \geq 0}$ takes values in a finite set $K$ which is compact. From , it is clear that $V_{\tau}(t) / V_{\pi}(t) \rightarrow 0$ as $t \rightarrow \infty$.
Extension to continuous time {#sec:continuoustime}
----------------------------
We discuss briefly how Theorem \[thm:main\] can be generalized to continuous time. In continuous time, we let the market weight process $\{\mu(t)\}_{t \geq 0}$ be a continuous semimartingale with state space $\Delta^{(n)}$. The market weight process of a portfolio $\pi$ satisfies the stochastic differential equation $$\frac{\mathrm{d}V_{\pi}(t)}{V_{\pi}(t)} = \sum_{i = 1}^n \pi_i(\mu(t)) \frac{\mathrm{d}\mu_i(t)}{\mu_i(t)}.$$ Let $(\pi, \Phi), (\tau, \Psi) \in {\mathcal{FG}}^2$. Then we have the decomposition $$\log \frac{V_{\tau}(t)}{V_{\pi}(t)} = \log \frac{\Psi(\mu(t)) / \Psi(\mu(0))}{\Phi(\mu(t)) / \Phi(\mu(0))} + A(t),$$ where the drift process takes the form $A(t) = A_{\tau}(t) - A_{\pi}(t)$, $$\label{eqn:qvariation}
A_{\tau}(t) = \int_0^t H_{\tau}(\mu(s))(\mathrm{d}\mu(s), \mathrm{d}\mu(s)),$$ and the analogous definition holds for $A_{\pi}$. See [@F02 Theorem 3.1.5]. In we use the intrinsic notation of [@EM89] for the quadratic variation of $\{\mu(t)\}$ with respect to the non-negative definite form $H_{\tau}$. It can be shown that $A(t)$ is non-decreasing almost surely for all continuous semimartingales $\{\mu(t)\}$ if and only if $A_{\tau} \geq A_{\pi}$. We may define the relation $\tau \succeq \pi$ (domination on compacts) in the same way as in Definition \[def:pseudoarbitrage\], except that we require for any continuous semimartingale $\{\mu(t)\}$ with values in $K$, holds for all $t \geq 0$ almost surely. Using the results established, one can show in continuous time that $\pi$ is maximal in ${\mathcal{FG}}^2$ if it is generated by a measure of diversity satisfying .
Moreover, in continuous time, [@CDOS09 Theorem 3] shows that the integral condition is also necessary for $(\pi, \Phi)$ to be maximal in ${\mathcal{FG}}^2$ when $n = 2$. Let $u(x) = \Phi(x, 1 - x)$. The idea is that if the integral converges, we can solve the initial value problem $$v''(x) + \left(\frac{-u''(x)}{u(x)} + s(x)\right)v(x) = 0, \quad x \in (0, 1),$$ $$v\left(\frac{1}{2}\right) = u\left(\frac{1}{2}\right), \quad v'\left(\frac{1}{2}\right) = u'\left(\frac{1}{2}\right) = 0,$$ for some appropriately chosen function $s(x)$ such that $s(x) \geq 0$, $s(x) \not\equiv 0$ and $s$ is symmetric about $\frac{1}{2}$. Sturm’s comparison theorem implies that the solution $v(x)$ is positive (and concave) on $(0, 1)$. Let $\Psi(p) = v(p_1)$ and let $\tau$ be the portfolio generated by $\Psi$. Then the corresponding portfolio $\tau$ is not equal to $\pi$ and dominates $\pi$ on compacts, so $\pi$ is not maximal in ${\mathcal{FG}}^2$.
Characterize the maximal portfolios of ${\mathcal{FG}}$.
Optimization of functionally generated portfolios {#sec:optimization}
=================================================
A shape-constrained optimization problem {#sec:optim2}
----------------------------------------
Consider the relative value process of a functionally generated portfolio. If we have a model for the market weight process $\{\mu(t)\}_{t \geq 0}$, a natural optimization problem is to maximize the expected growth rate of the drift process over some horizon. To this end, suppose we are given an [*intensity measure*]{} ${\Bbb P}$ of the increments $(\mu(t), \mu(t + 1))$ modeled as a Borel probability measure on $\Delta^{(n)} \times \Delta^{(n)}$. We assume that ${\Bbb P}$ is either discrete (with countably many masses) or absolutely continuous with respect to the measure $\nu := m \otimes m$ on $\Delta^{(n)} \times \Delta^{(n)}$, where $m$ is the surface measure of $\Delta^{(n)}$ in ${\Bbb R}^n$ (which should be thought of as the Lebesgue measure on $\Delta^{(n)}$). We will abbreviate this by simply saying ${\Bbb P}$ is absolutely continuous. For technical reasons, we assume that ${\Bbb P}$ is supported on $K \times K$ for some compact subset $K$ of $\Delta^{(n)} \times \Delta^{(n)}$.
Given the intensity measure ${\Bbb P}$, we consider the optimization problem $$\label{eqn:newoptim1}
\max_{(\pi, \Phi) \in {\mathcal{FG}}} \int T\left(q \mid p\right) \mathrm{d} {\Bbb P}.$$
First we give some examples of the intensity measure.
\[exm:markov\] Suppose $\{(\mu(t - 1), \mu(t))\}$ is an ergodic Markov chain on $K \times K$. We can take ${\Bbb P}$ to be the stationary distribution of $(\mu(t - 1), \mu(t))$. It is easy to see that an optimal portfolio in maximizes the asymptotic growth rate $\lim_{t \rightarrow \infty} \frac{1}{t} \log V_{\pi}(t)$ of the relative value (the term $\frac{1}{t} \log \frac{\Phi(\mu(t))}{\Phi(\mu(0))}$ vanishes as $t \rightarrow \infty$). This portfolio can be regarded as a [*growth optimal portfolio*]{} (relative to the market portfolio) among the functionally generated portfolios.
\[exm:green\] We model $\{\mu(t)\}_{t \geq 0}$ as a stochastic process. Let $K$ be a compact subset of $\Delta^{(n)}$ containing $\mu(0)$. Let $\tau$ be the first exit time of $K$, i.e., $$\tau = \inf\{t \geq 0: \mu(t) \notin K\}.$$ Consider the measure ${\Bbb G}$ on $K \times K$ defined by $${\Bbb G}(A) := {\Bbb E} \left[\sum_{t = 1}^{\tau - 1} 1_{\{(\mu(t-1), \mu(t)) \in A\}} \right], \quad A \subset K \times K \text{ measurable}.$$ If the process $\{(\mu(t-1), \mu(t))\}$ is Markovian, ${\Bbb G}$ is the [*Green kernel*]{} of the process killed at time $\tau$. Suppose ${\Bbb G}(K \times K) = {\Bbb E} (\tau - 1) < \infty$, i.e., the exit time has finite expectation. Then $${\Bbb P}(\cdot) := \frac{1}{{\Bbb G}(K \times K)} {\Bbb G}(\cdot)$$ is a probability measure on $K \times K$. This intensity measure will be used in the empirical example in Section \[sec:empirical\].
Note that Example \[exm:markov\] deals with infinite horizon while Example \[exm:green\] is concerned with a finite (but random) horizon. The optimization problem is shape-constrained because the generating function is concave by definition. We will first study some theoretical properties of this abstract (unconstrained) optimization problem, and then focus on a discrete special case where numerical solutions are possible and further constraints are imposed. In contrast to classical portfolio selection theory where the portfolio weights are optimized period by period, in we optimize the portfolio weights over a region simultaneously.
Throughout the development it is helpful to keep in mind the analogy between and the maximum likelihood estimation of a log-concave density. In that context, we are given a random sample $X_1, ..., X_N$ from a log-concave density $f_0$ on ${\Bbb R}^d$ (i.e., $\log f_0$ is concave). The log-concave maximum likelihood estimate (MLE) $\widehat{f}$ is the solution to $$\label{eqn:MLE}
\max_f \sum_{j = 1}^N \log f(X_j),$$ where $f$ ranges over all log-concave densities on ${\Bbb R}^d$. It can be shown that the MLE exists almost surely (when $N \geq d + 1$ and the support of $f_0$ has full dimension) and is unique; see [@CSS10] for precise statements of these results. We remark that is more complicated than because the portfolio weights correspond to selections of the superdifferential $\partial \log \Phi$, wheras involves only the values of the density.
Theoretical properties
----------------------
It is easy to check that is a convex optimization problem since the L-divergence is concave in the portfolio weights (Lemma \[lem:Driftconcave\]). First we show that has an optimal solution and study in what sense the solution is unique.
Given an intensity measure ${\Bbb P}$, it can be decomposed in the form $$\label{eqn:conditional}
{\Bbb P}(\mathrm{d}p\mathrm{d}q) = {\Bbb P}_1(\mathrm{d}p) {\Bbb P}_2(\mathrm{d}q | p),$$ where ${\Bbb P}_1$ is the first marginal of ${\Bbb P}$ and ${\Bbb P}_2$ is the conditional distribution of the second variable given $p$. We will need a technical condition for ${\Bbb P}$ which allows jumps in all directions.
Let ${\Bbb P}$ be an absolutely continuous probability measure on $\Delta^{(n)} \times \Delta^{(n)}$ with the decomposition . Write $${\Bbb P}_1(\mathrm{d}p) = f(p) m(\mathrm{d}p),$$ where $f(\cdot)$ is the density of ${\Bbb P}_1$ with respect to $m$. We say that ${\Bbb P}$ satisfies the support condition if for $m$-almost all $p$ for which $f(p) > 0$, for all $v \in T\Delta^{(n)}$, there exists $\lambda > 0$ such that $p + \lambda v$ belongs to the support of ${\Bbb P}_2(\cdot | p)$.
We have the following result which is analogous to [@CSS10 Theorem 1].
\[thm:optim\] Consider the optimization problem where ${\Bbb P}$ is a discrete or absolutely continuous Borel probability measure on $\Delta^{(n)} \times \Delta^{(n)}$ supported on $K \times K$ with $K \subset \Delta^{(n)}$ compact.
1. The problem has an optimal solution.
2. If $\pi^{(1)}$ and $\pi^{(2)}$ are optimal solutions, then $$\label{eqn:uniqueness}
\left\langle \frac{\pi^{(1)}(p)}{p}, q - p \right\rangle = \left\langle\frac{\pi^{(2)}(p)}{p}, q - p \right\rangle$$ for ${\Bbb P}$-almost all $(p, q)$. In particular, if ${\Bbb P}(\mathrm{d}p\mathrm{d}q) = {\Bbb P}_1(\mathrm{d}p) {\Bbb P}_2(\mathrm{d}q | p)$ is absolutely continuous with ${\Bbb P}_1(\mathrm{d}p) = f(p) m(\mathrm{d}p)$ and satisfies the support condition, then $\pi^{(1)} = \pi^{(2)}$ $m$-almost everywhere on $\{p: f(p) > 0\}$.
The proofs of Theorem \[thm:optim\] and Theorem \[thm:optim2\] below are given in Appendix \[sec:appendix\].
Let ${\Bbb P}$ an intensity measure. Suppose $\{{\Bbb P}_N\}_{N \geq 1}$ is a sequence of probability measures converging weakly to ${\Bbb P}$. By definition, this means that $$\lim_{N \rightarrow \infty} \int f \mathrm{d}{\Bbb P}_N = \int f \mathrm{d}{\Bbb P}$$ for all bounded continuous functions on $\Delta^{(n)} \times \Delta^{(n)}$. For example, one may sample i.i.d. observations $\{(p(j), q(j))\}_{j = 1}^N$ from ${\Bbb P}$ and take ${\Bbb P}_N$ to be the empirical measure $\frac{1}{N} \sum_{j = 1}^N \delta_{(p(j), q(j))}$, where $\delta_{(p(j), q(j))}$ is the point mass at $(p(j), q(j))$. From the perspective of statistical inference, the optimal portfolio $(\widehat{\pi}^{(N)}, \widehat{\Phi}^{(N)})$ for ${\Bbb P}_N$ can be regarded as a point estimate of the optimal portfolio $(\pi, \Phi)$ for ${\Bbb P}$. The following result states that the estimator is consistent. See [@CS10 Theorem 4] for an analogous statement in the context of log-concave density estimation.
\[thm:optim2\] Let $(\pi, \Phi)$ be the optimal portfolio in problem for ${\Bbb P}$, where ${\Bbb P}(\mathrm{d}p\mathrm{d}q) = {\Bbb P}_1(\mathrm{d}p) {\Bbb P}_2(\mathrm{d}q | p)$ is absolutely continuous with ${\Bbb P}_1(\mathrm{d}p) = f(p) m(\mathrm{d}p)$, supported on $K \times K$ with $K \subset \Delta^{(n)}$ compact, and satisfies the support condition. Let $\{{\Bbb P}_N\}$ be a sequence of discrete or absolutely continuous probability measures on $K \times K$ such that ${\Bbb P}_N \rightarrow {\Bbb P}$ weakly, and suppose $(\widehat{\pi}^{(N)}, \widehat{\Phi}^{(N)})$ is optimal for the measure ${\Bbb P}_N$, $N \geq 1$. Then $\widehat{\pi}^{(N)} \rightarrow \pi$ $m$-almost everywhere on $\{p: f(p) > 0\}$.
Finite dimensional reduction
----------------------------
Without further constraints, the optimal portfolio weights of may be highly irregular. Now we restrict to the special case where $$\label{eqn:discreteP}
{\Bbb P} = \frac{1}{N} \sum_{j = 1}^N \delta_{(p(j), q(j))}$$ is a discrete measure and $(p(j), q(j)) \in \Delta^{(n)} \times \Delta^{(n)}$ for $j = 1, ..., N$. This presents no great loss of generality because in practice the market weights have finite precision and we can choose the pairs $(p(j), q(j))$ to take values in a grid approximating $\Delta^{(n)} \times \Delta^{(n)}$. Moreover, from Theorem \[thm:optim2\] we expect that when $N$ is large the optimal solution approximates that of the continuous counterpart. Consider the modified optimization problem $$\label{eqn:newoptim2}
\begin{aligned}
& \underset{(\pi, \Phi) \in {\mathcal{FG}}}{\text{maximize}}
& & \int T\left(q \mid p\right)\mathrm{d} {\Bbb P} \\
& \text{subject to}
& & (\pi(p(1)), ..., \pi(p(N))) \in C, \\
\end{aligned}$$ where $C$ is a given closed convex subset of $\overline{\Delta^{(n)}}^N$. Some examples of $C$ are given in Table \[tab:constraints\], where each constraint is a cylinder set of the form $\{\pi(p(j)) \in C_j\}$ with $C_j$ a closed convex set of $\overline{\Delta^{(n)}}$. ‘Global’ constraints on the weights can be imposed, see Section \[sec:empirical\] for an example. It can be verified easily that the proof of Theorem \[thm:optim\] goes through without changes with these constraints, so has an optimal solution. Moreover, if $\pi^{(1)}$ and $\pi^{(2)}$ are optimal solutions, then $$\left\langle \frac{\pi^{(1)}(p(j))}{p(j)}, q(j) - p(j) \right\rangle = \left\langle\frac{\pi^{(2)}(p(j))}{p(j)}, q(j) - p(j) \right\rangle, \quad j = 1, ..., N.$$
[ll]{} Constraint & Interpretation\
$a_i \leq \pi_i(p) \leq b_i$ & Box constraints on portfolio weights\
$m_i \leq \frac{\pi_i(p)}{p_i} \leq M_i$ & Box constraints on weight ratios\
$(\pi(p) - p)' \Sigma (\pi(p) - p) < \varepsilon$ & Constraint on tracking error given a covariance matrix\
For maximum likehood estimation of log-concave density, it is shown in [@CSS10] that the logarithm of the MLE $\widehat{f}$ is [*polyhedral*]{}, i.e., $\log \widehat{f}$ is the pointwise minimum of several affine functions (see [@R70 Section 19]). In particular, there exists a triangulation of the data points over which $\log \widehat{f}$ is piecewise affine. We show that an analogous statement holds for . Let $D = \{p(j), q(j): j = 1, ..., N\}$ be the set of data points.
\[thm:optim3\] Let $(\pi, \Phi)$ be an optimal portfolio for the problem where ${\Bbb P} = \frac{1}{N} \sum_{j = 1}^N \delta_{(p(j), q(j))}$. Let $\overline{\Phi}: \Delta^{(n)} \rightarrow (0, \infty)$ be the smallest positive concave function on $\Delta^{(n)}$ such that $\overline{\Phi}(p) \geq {\Phi}(p)$ for all $x \in D$. Then $\overline{\Phi}$ is a polyhedral positive concave function on $\Delta^{(n)}$ satisfying $\overline{\Phi} \leq \Phi$ and $\overline{\Phi}(p) = \Phi(p)$ for all $p \in D$. Moreover, $\overline{\Phi}$ generates a portfolio $\overline{\pi}$ such that $\overline{\pi}(p(j)) = \pi(p(j))$ for all $j$. In particular, $(\overline{\pi}, \overline{\Phi})$ is also optimal for the problem .
It is a standard result in convex analysis that $\overline{\Phi}$ such defined is [*finitely generated*]{} (see [@R70 Section 19]). By [@R70 Corollary 19.1.2], $\overline{\Phi}$ is a polyhedral concave function. By definition of $\overline{\Phi}$ and concavity of $\Phi$, we have $\overline{\Phi}(p) = {\Phi}(p)$ for all $x \in D$ for all $j$ and $\overline{\Phi} \leq \Phi$. This implies that $\partial \log \Phi(p(j)) \subset \partial \log \overline{\Phi}(p(j))$ for all $j$. By Lemma \[lem:superdiff\](ii), $\overline{\Phi}$ generates a portfolio $\overline{\pi}$ which agrees with $\pi$ on $\{p(1) ,..., p(N)\}$. It follows that (using obvious notations) $$\overline{T}\left(q(j) \mid p(j)\right) = T\left(q(j) \mid p(j) \right)$$ for all $j$, and hence $(\overline{\pi}, \overline{\Phi})$ is optimal for .
Theorem \[thm:optim3\] reduces to a finite-dimensional problem. In the next section we present an elementary implementation for the case $n = 2$ (analogous to univariate density estimation) and illustrate its application in portfolio management with a case study.
Empirical examples {#sec:empirical}
==================
A case study
------------
![The figure on the left shows the growth of $\$1$ for each asset, and the one on the right shows the time series of the market weight $\mu_1(t)$ of US. The vertical dotted line divides the data set into the training and testing periods respectively.[]{data-label="fig:data"}](data.pdf)
In global portfolio management, an important topic is the determination of the aggregate portfolio weights for countries. In this example we consider two countries: US and China. We represent them by the S&P US BMI index (asset 1) and the S&P China BMI index (asset 2) respectively. The ‘market’ consists of these two assets. We collect monthly data from January 2001 to June 2014 using Bloomberg. The benchmark portfolio is taken to be the buy-and-hold portfolio starting with weights $(0.5, 0.5)$ at January 2001. Here the initial market weights $(0.5, 0.5)$ are chosen arbitrarily. The data from January 2001 to December 2010 will be used as the training data to optimize the portfolio which will be backtested in the subsequent period. The market weights at January 2011 are $(0.1819, 0.8191)$. The data is plotted in Figure \[fig:data\].
Let $K \subset \Delta^{(2)}$ be the compact set defined by $$\label{eqn:set}
K = \{p = (p_1, p_2) \in \Delta^{(2)}: 0.1 \leq p_1 \leq 0.3\}.$$ Our objective here is to optimize a functionally generated portfolio to be held as long as the market weights stay within $K$. If the market weight of US approach these boundary points (regarded as a regime change), a new portfolio will be chosen, so $0.1$ and $0.3$ can be thought of as the [*trigger points*]{}.
The intensity measure and constraints {#subsec:example}
-------------------------------------
![Density estimate of ${\Bbb P}_N$ on $K \times K$ in terms of the market weight of US.[]{data-label="fig:density"}](density.pdf)
Suppose $t = 0$ corresponds to January 2011. We model $\{\mu(t)\}_{t \geq 0}$ as a discrete-time stochastic process (time is monthly) where $\mu(0)$ is constant. Let ${\Bbb P}$ be the measure in Example \[exm:green\] where $\tau$ is the first exit time of $K$ given in .
If a stochastic model is given, we may approximate ${\Bbb P}$ by simulating paths of $\{\mu(t)\}$ killed upon exiting $K$. The resulting empirical measure $${\Bbb P}_N = \frac{1}{N} \sum_{j = 1}^N \delta_{(p(j), q(j))}$$ is then taken as the intensity measure of the optimization problem .
Since our main concern is the implementation of the optimization problem , sophisticated modeling of $\{\mu(t)\}$ will not be attempted and we will use a simple method to simulate paths of $\{\mu(t)\}$. Namely, starting at $\mu(0) = (0.1819, 0.8191)$, we simulate paths of $\{\mu(t)\}_{t = 0}^{\tau - 1}$ by [*bootstrapping*]{} the past returns of the two assets and computing the corresponding market weight series. In view of the possible recovery of US, before the simulation we recentered the past returns so that they both have mean zero over the training period. (Essentially, only the difference in returns matter for the evolution of the market weights.) We simulated 50 such paths and obtained $N = 3115$ pairs $(p(j), q(j))$ in $K \times K$. A density estimate of ${\Bbb P}_N$ (in terms of the market weight of US) is plotted in Figure \[fig:density\]. To reduce the number of variables, the market weights are rounded to 3 decimal places, so the market weights of US take values in the set $D = \{0.100, 0.101, ..., 0.299, 0.300\}$.
Next we specify the constraints for $\{\pi(p_1) := \pi(p_1, 1 - p_1): p_1 \in D\}$. (This notation should cause no confusion since the market weight of China is determined by that of US.) First, we require that $\pi_1(p_1)$ is non-decreasing in $p_1$, i.e., $$\pi_1(0.100) \leq \pi_1(0.101) \leq \cdots \leq \pi_1(0.300).$$ This imposes a shape constraint on the portfolio weights which guarantees that the portfolio weights always move in the direction of market movement. To control the concentration of the portfolio we require also that the weight ratio of US satisfies $0.5 \leq \frac{\pi_1(p_1)}{p_1} \leq 2$ for $p_1 \in D$ (since there are only two assets, this implies a weight ratio bound for China). These constraints determine the convex set $C$ in the optimization problem we are about to solve.
Optimization procedure
----------------------
![The portfolio weight and the generating function of the optimized portfolio.[]{data-label="fig:result"}](result.pdf)
By Theorem \[thm:optim3\], it suffices to optimize over generating functions that are piecewise linear over the data points. First we introduce some simplifying notations. Write the set of grid points as $D = \{x_1 < x_2 < \cdots < x_m\}$ and let $x_0 = 0$, $x_{m+1} = 1$ be the endpoints of the interval. Let the decision variables be $$z_j := \pi(x_j, 1 - x_j), \quad j = 1, ..., m,$$ $$\varphi_j := \Phi(x_j, 1 - x_j), \quad j = 0, ..., m + 1.$$ By scaling, we may assume $\varphi_1 = 1$. The constraints on $\{\varphi_j\}$ are $$\label{eqn:nonnegative}
\varphi_j \geq 0, \quad j = 0, ..., m+1, \quad \varphi_1 = 1, \quad \text{(non-negativity)}$$ $$\label{eqn:concavity}
s_0 \geq s_1 \geq \cdots \geq s_m, \quad s_j := \frac{\varphi_{j+1} - \varphi_j}{x_{j+1} - x_j}. \quad \text{(concavity)}$$ We require that $\pi$ is generated by $\Phi$. By and Lemma \[lem:superdiff\], it can be seen that $z_j$ satisfies the inequality $$\label{eqn:generated}
x_j + x_j(1 - x_j) \frac{s_j}{\varphi_j} \leq z_j \leq x_j + x_j(1 - x_j) \frac{s_{j-1}}{\varphi_j}, \quad j = 1, ..., m. \quad \text{($(\pi, \Phi) \in {\mathcal{FG}}$)}$$ We require that $z_j$ is non-decreasing in $j$: $$\label{eqn:monotone}
z_1 \leq z_2 \leq \cdots \leq z_m. \quad \text{(monotonicity)}$$ Finally, we require that the weight ratios are bounded between $0.5$ and $2$: $$\label{eqn:weightratio}
0.5 \leq \frac{z_j}{x_j} \leq 2, \quad j = 1, ..., m. \quad \text{(weight ratios)}$$ With the constraints - we maximize $$\int T\left(q \mid p\right) \mathrm{d}{\Bbb P}_N = \frac{1}{N} \sum_{j = 1}^N T\left( q(j) \mid p(j) \right)$$ over $\{z_j\}$ and $\{\varphi_j\}$. This is a standard non-linear, but smooth, constrained optimization problem (convexity is lost because $\Phi$ is now piecewise linear). We implement this optimization problem using the `fmincon` function in `MATLAB`. The optimal portfolio weights together with the generating function are plotted in Figure \[fig:result\]. It turns out that the optimal portfolio is close to constant-weighted (with weights $(0.2331, 0.7669)$). Note that the constraint on the weight ratio limits the deviation of $\pi_1(p_1)$ from the market weight $p_1$. If the weight ratio constraint was not imposed (while the monotonicity constraint was kept), the optimal portfolio would be the equal-weighted portfolio $\pi \equiv (0.5, 0.5)$, and the reason can be seen from the proof of Lemma \[lem:diffeqn\].
Backtesting the portfolio
-------------------------
![Fernholz’s decomposition of the optimized portfolio over the testing period. The log relative value is $\log V_{\pi}(t)$. The generating function term is $\log \Phi(\mu(t)) - \log \Phi(\mu(0))$, and the drift process is $A(t)$.[]{data-label="fig:backtest"}](backtest.pdf)
Finally, we compute the performance of the optimized portfolio over the testing period January 2011 to June 2014. The result (plotted using the function `FernholzDecomp` of the `RelValAnalysis` package) is shown in Figure \[fig:backtest\]. Over the testing period, the portfolio beats the market by nearly 2% in log scale and its performance has been steady. From the decomposition, about half of the outperformance is attributed to the increase of the generating function (note that the market weight of US becomes closer to $0.2331$ where the generating function attains its maximum), and the rest comes from the drift process. That the optimal portfolio is close to constant-weighted may not be very interesting, but this is a consequence of the data and our choice of constraints and is by no means obvious. Our optimization framework allows many other possibilities especially when there are multiple assets. Other useful constraints and efficient algorithms are natural subjects of further research.
Proofs of Theorem \[thm:optim\] and Theorem \[thm:optim2\] {#sec:appendix}
==========================================================
First we will state and prove some lemmas from convex analysis.
\[lem:compactness\] Let $p_0 \in \Delta^{(n)}$ be fixed and let ${\mathcal C}_0$ be the collection of positive concave functions $\Phi$ on $\Delta^{(n)}$ satisfying $\Phi(p_0) = 1$. Then any sequence in ${\mathcal C}_0$ has a subsequence which converges locally uniformly on $\Delta^{(n)}$ to a function in ${\mathcal C}_0$.
By [@R70 Theorem 10.9], it suffices to prove that ${\mathcal C}_0$ has a uniform upper bound (the lower bound is immediate since functions in ${\mathcal C}_0$ are non-negative). We first derive an upper bound in the one-dimensional case. Let $f$ be a non-negative concave function on the real interval $[a, b]$. Let $x_0 \in (a, b)$ and suppose $f(x_0) = 1$. Let $x \in [a, x_0]$ and write $x_0 = \lambda x + (1 - \lambda)b$ for some $\lambda \in [0, 1]$. By concavity, $$1 = f(x_0) \geq \lambda f(x) + (1 - \lambda) f(b) \geq \lambda f(x).$$ Thus $$f(x) \leq \frac{1}{\lambda} = \frac{b - x}{b - x_0} \leq \frac{b - a}{b - x_0}, \quad x \in [a, x_0].$$ The case $x \in [x_0, b]$ can be handled similarly, and we get $$\label{eqn:concaveupperbound}
f(x) \leq \frac{b - a}{\min\{|x_0 - a|, |x_0 - b|\}}, \quad x \in [a, b].$$ Now let $\Phi \in {\mathcal C}_0$. Applying to the restrictions of $\Phi$ to line segments in $\Delta^{(n)}$ containing $p_0$, we get $$\Phi(p) \leq \frac{\text{diam}\left(\Delta^{(n)}\right)}{\text{dist}\left(p_0, \partial \Delta^{(n)}\right)}, \quad p \in \Delta^{(n)},$$ where $\text{diam}(\Delta^{(n)})$ is the diameter of $\Delta^{(n)}$ and $\text{dist}(p_0, \partial \Delta^{(n)})$ is the distance from $p_0$ to the boundary of $\Delta^{(n)}$. This completes the proof of the lemma.
\[lem:subdiffconvergence\] Let $(\pi, \Phi), (\pi^{(k)}, \Phi^{(k)}) \in {\mathcal{FG}}$, $k \geq 1$. Suppose $\Phi^{(k)}$ converges locally uniformly on $\Delta^{(n)}$ to $\Phi$. Let $p \in \Delta^{(n)}$ be a point at which $\Phi$ is differentiable. Then given $\varepsilon > 0$, there exists $\delta > 0$ and a positive integer $k_0$ such that $\|\pi^{(k)}(q) - \pi(p)\| < \varepsilon$ whenever $k \geq k_0$ and $q \in B(p, \delta)$. In particular, $\pi^{(k)}$ converges $m$-almost everywhere to $\pi$ as $k \rightarrow \infty$.
It is clear that $\log \Phi^{(k)}$ also converges locally uniformly to $\log \Phi$. We will use a well-known convergence result for the superdifferentials of concave functions, see [@HJL96 Theorem 6.2.7]. Indeed, the proof of [@HJL96 Theorem 6.2.7] implies a slightly stronger statement than the theorem. Namely, for any $p \in \Delta^{(n)}$ and any $\varepsilon > 0$, there exists a positive integer $k_0$ and $\delta > 0$ such that $$\label{eqn:claim1}
\begin{split}
\partial \log \Phi^{(k)}(q) &\subset \partial\log \Phi(p) + B(0, \varepsilon), \quad k \geq k_0, \quad q \in B(p, \delta), \\
\partial \log \Phi(q) &\subset \partial \log\Phi(p) + B(0, \varepsilon), \quad q \in B(p, \delta).
\end{split}$$
Suppose $\Phi$ is differentiable at $p$. Then $\partial \log\Phi(p)$ is a singleton. By Lemma \[lem:superdiff\], there are measurable selections $\xi^{(k)}$ and $\xi$ of $\partial \log \Phi^{(k)}$ and $\partial \log \Phi$ respectively such that $$\begin{split}
\pi^{(k)}_i(q) &= q_i \left(\xi^{(k)}_i(q) + 1 - \sum_{j = 1}^n q_j\xi^{(k)}_j(q)\right), \\
\pi_i(q) &= q_i \left(\xi_i(q) + 1 - \sum_{j = 1}^n q_j\xi_j(q)\right), \\
\end{split}$$ for all $q \in \Delta^{(n)}$, $i = 1, ..., n$, and $k \geq 1$.
For each $i = 1, ..., n$, consider the map $G_i$ defined by $$(q, \xi) \in \Delta^{(n)} \times T\Delta^{(n)} \mapsto q_i\left(\xi_i + 1 - \sum_{j = 1}^n q_j \xi_j\right).$$ The map $G = (G_1, ..., G_n)$ is clearly jointly continuous. We have $\pi(q) = G(q, \xi(q))$ and $\pi^{(k)}(q) = G(q, \xi^{(k)}(q))$.
By , for any $\varepsilon > 0$, there exists $k_0$ and $\delta > 0$ such that $$\label{eqn:delta}
\|\xi^{(k)}(q) - \xi(p)\| < \varepsilon, \quad \|\xi(q) - \xi(p)\| < \varepsilon$$ for all $k \geq k_0$ and $q \in B(p, \delta)$. The claim follows from and the joint continuity of $G$ at $(q, \xi(q))$. The last statement follows since a finite concave function on $\Delta^{(n)}$ is differentiable $m$-almost everywhere [@R70 Theorem 25.5].
\(i) The existence of an optimal solution will be proved by a compactness argument. Suppose $(\pi^{(k)}, \Phi^{(k)})$ is a maximizing sequence for . By scaling, we may assume $\Phi^{(k)}(p_0) = 1$ where $p_0 \in \Delta^{(n)}$ is fixed. By Lemma \[lem:compactness\], we may replace it by a subsequence such that $\Phi^{(k)}$ converges locally uniformly on $\Delta^{(n)}$ to a positive concave function $\Phi$ on $\Delta^{(n)}$. By Lemma \[lem:superdiff\](ii), $\Phi$ generates a portfolio $\pi$.
[*Case 1.*]{} ${\Bbb P}$ is absolutely continuous. By Lemma \[lem:subdiffconvergence\], $\pi^{(k)}$ converges $m$-almost everywhere to $\pi$. Let $T^{(k)}$ and $T$ be the L-divergences of $(\pi^{(k)}, \Phi^{(k)})$ and $(\pi, \Phi)$ respectively. Recall that ${\Bbb P}$ is supported on $K \times K$ where $K \subset \Delta^{(n)}$ is compact. For $x \in \overline{\Delta^{(n)}}$ and $p, q \in K$, we have $$\label{eqn:Tupperbound}
1 + \left\langle \frac{x}{p}, q - p \right\rangle = \sum_{i = 1}^n x_i \frac{q_i}{p_i} \leq \sum_{i = 1}^n \frac{x_i}{p_i} \leq \frac{1}{\min_{p \in K, 1 \leq i \leq n} p_i}.$$ Also $\Phi^{(k)} \rightarrow \Phi$ uniformly on $K$. Hence the family of L-divergences $\{T, T^{(1)}, T^{(2)}, ...\}$ is uniformly bounded on $K \times K$. By Lebesgue’s dominated convergence theorem, we have $$\lim_{k \rightarrow \infty} \int T^{(k)}\left(q \mid p \right) \mathrm{d} {\Bbb P} = \int T\left(q \mid p \right) \mathrm{d} {\Bbb P}.$$ Thus $(\pi, \Phi)$ is optimal.
[*Case 2.*]{} ${\Bbb P}$ is discrete and has masses at $(p(j), q(j))$. Since $\overline{\Delta^{(n)}}$ is compact, by a diagonal argument we can extract a further subsequence (still denoted by $\{(\pi^{(k)}, \Phi^{(k)})\}$) such that $\lim_{k \rightarrow \infty} \pi^{(k)}(p(j))$ exists for each $j$. Now we can redefine $\pi$ on $\{p(1), p(2), ...\}$ such that $\pi(p(j)) = \lim_{k \rightarrow \infty} \pi^{(k)}(p(j))$ for each $j$. Since we only modify $\pi$ at countably many points, $\pi$ is still Borel measurable. Now we may apply Lebesgue’s dominated convergence theorem and conclude that $(\pi, \Phi)$ is optimal.
\(ii) Suppose $(\pi^{(1)}, \Phi^{(1)})$ and $(\pi^{(2)}, \Phi^{(2)})$ are optimal solutions. Define $\pi = \frac{1}{2} \pi^{(1)} + \frac{1}{2} \pi^{(2)}$ which is generated by the geometric mean $\Phi = \sqrt{\Phi^{(1)}\Phi^{(2)}}$ (Lemma \[lem:FGconvex\]) . Also let $T$, $T^{(1)}$ and $T^{(2)}$ be the L-divergences of $(\pi, \Phi)$, $(\pi^{(1)}, \Phi^{(2)})$ and $(\pi^{(2)}, \Phi^{(2)})$ respectively. By concavity of the L-divergence (Lemma \[lem:Driftconcave\]), we have $$\label{eqn:uniqueness2}
\int T\left(q \mid p \right) \mathrm{d} {\Bbb P} \geq \frac{1}{2} \left(\int T^{(1)}\left(q \mid p \right) \mathrm{d} {\Bbb P} + \int T^{(2)}\left(q \mid p \right) \mathrm{d} {\Bbb P}\right).$$ Hence $(\pi, \Phi)$ is also optimal. It follows from and the strict concavity of the logarithm that $$\left\langle \frac{\pi^{(1)}(p)}{p}, q - p \right\rangle = \left\langle\frac{\pi^{(2)}(p)}{p}, q - p \right\rangle$$ for ${\Bbb P}$-almost all $(p, q)$.
If ${\Bbb P}$ is absolutely continuous and satisfies the support condition, then for $m$-almost all $p$ for which $f(p) > 0$, we have $$\left\langle \frac{\pi^{(1)}(p)}{p}, v \right\rangle = \left\langle\frac{\pi^{(2)}(p)}{p}, v \right\rangle$$ for all tangent vectors $v$. This and the fact that $\pi^{(1)}(p), \pi^{(2)}(p) \in \overline{\Delta^{(n)}}$ imply that $\pi^{(1)}(p) = \pi^{(2)}(p)$ $m$-almost everywhere on $\{p: f(p) > 0\}$.
\[Proof of Theorem \[thm:optim2\]\] By scaling, we may assume that $\widehat{\Phi}^{(N)}(p_0) = \Phi(p_0) = 1$ for all $N \geq 1$. By Lemma \[lem:compactness\], any subsequence of $\{\widehat{\Phi}^{(N)}\}$ has a further subsequence which converges locally uniformly to a positive concave function $\widehat{\Phi}$ on $\Delta^{(n)}$. Replacing $\{\widehat{\Phi}^{(N)}\}$ by such a convergent subsequence, we may assume that $\widehat{\Phi}^{(N)} \rightarrow \widehat{\Phi}$ locally uniformly on $\Delta^{(n)}$. Let $\widehat{\pi}$ be any portfolio generated by $\widehat{\Phi}$ (which exists by Lemma \[lem:superdiff\](ii)). We claim that $(\widehat{\pi}, \widehat{\Phi})$ is optimal and hence $\widehat{\pi} = \pi$ $m$-almost everywhere on $\{p: f(p) > 0\}$.
Let $\widehat{T}^{(N)}$, $\widehat{T}$ and $T$ be the L-divergences of $(\widehat{\pi}^{(N)}, \widehat{\Phi}^{(N)})$, $(\widehat{\pi}, \widehat{\Phi})$ and $(\pi, \Phi)$ respectively. By the optimality of $(\pi^{(N)}, \Phi^{(N)})$ for the measure ${\Bbb P}_N$, we have $$\label{eqn:finiteNoptimality}
\int \widehat{T}^{(N)}\left(q \mid p \right) \mathrm{d} {\Bbb P}_N \geq \int T\left(q \mid p \right) \mathrm{d} {\Bbb P}_N, \quad N \geq 1.$$
We would like to let $N \rightarrow \infty$ in . The L-divergence $T\left( q \mid p \right)$ is clearly continuous on $K \times K$ (note that $K$ is compact). By the definition of weak convergence, we have $$\lim_{N \rightarrow \infty} \int T\left(q \mid p \right) \mathrm{d} {\Bbb P}_N = \int T\left(q \mid p\right) \mathrm{d} {\Bbb P}.$$ Suppose we can prove that $$\label{eqn:bigclaim}
\lim_{N \rightarrow \infty} \int \widehat{T}^{(N)}\left(q \mid p \right) \mathrm{d} {\Bbb P}_N = \int \widehat{T}\left(q \mid p\right) \mathrm{d} {\Bbb P}.$$ Then letting $N \rightarrow \infty$ in , we have $$\int \widehat{T}\left(q \mid p\right) \mathrm{d} {\Bbb P} \geq \int T\left(q \mid p\right) \mathrm{d} {\Bbb P},$$ so $(\widehat{\pi}, \widehat{\Phi})$ is optimal for the measure ${\Bbb P}$. Since ${\Bbb P}$ satisfies the support condition by assumption, by Theorem \[thm:optim\](ii) $\widehat{\pi}$ and $\pi$ are equal $m$-almost everywhere on $\{p: f(p) > 0\}$.
Thus we only need to prove . Here the technicality lies in the fact that both the integrands and the measures change with $N$, so standard integral convergence theorems do not apply.
The main idea is to use the local uniform convergence property in Lemma \[lem:subdiffconvergence\] and approximate the integrals in by Riemann sums. Let $\varepsilon > 0$ be given. We will construct two partitions $\{A_k\}_{k = 0}^{k_0}$, $\{B_{\ell}\}_{\ell = 1}^{\ell_0}$ of $K$, points $p_k \in A_k$, $q_{\ell} \in B_{\ell}$ and a positive integer $N_0$ with the following properties:
1. $A_k \times B_{\ell}$ is a ${\Bbb P}$-continuity set, i.e., ${\Bbb P}(\partial(A_k \times B_{\ell})) = 0$. Thus, by the Portmanteau theorem (see [@B09]), we have $$\lim_{N \rightarrow \infty} {\Bbb P}_N(A_k \times B_{\ell}) = {\Bbb P}(A_k \times B_{\ell}).$$ So for $N \geq N_0$ where $N_0$ is sufficiently large, we have $$\left|{\Bbb P}_N(A_k \times B_{\ell}) - {\Bbb P}(A_k \times B_{\ell})\right| < \frac{\varepsilon}{k_0\ell_0}$$ for all $k$, $\ell$.
2. ${\Bbb P}(A_0 \times K) < \varepsilon$ and ${\Bbb P}_N(A_0 \times K) < \varepsilon$ for $N \geq N_0$.
3. For $N \geq N_0$, $p \in A_k$, $q \in B_{\ell}$, $1 \leq k \leq k_0$ and $1 \leq \ell \leq \ell_0$, we have $$\left|\widehat{T}^{(N)}\left(q \mid p \right) - \widehat{T}\left(q_{\ell} \mid p_k\right)\right| < \varepsilon, \quad
\left|\widehat{T}\left(q \mid p \right) - \widehat{T}\left(q_{\ell} \mid p_k\right)\right| < \varepsilon.$$
4. $\left|\log \widehat{\Phi}^{(N)}(p) - \log \widehat{\Phi}(p) \right| < \varepsilon$ for $p \in K$ and $N \geq N_0$. (This is immediate since $\widehat{\Phi}^{(N)}$ converges uniformly to $\widehat{\Phi}$ on $K$ and $\widehat{\Phi}$ is positive on $K$.)
Suppose these objects have been constructed. Then for $N \geq N_0$ we can approximate the integrals as follows. By (ii) and (iii), we have $$\label{eqn:estimate1}
\begin{split}
& \left| \int \widehat{T}\left(q \mid p\right)\mathrm{d}{\Bbb P} - \sum_{\ell = 1}^{\ell_0} \sum_{k = 1}^{k_0} \widehat{T}\left(q_{\ell} \mid p_k\right) {\Bbb P}(A_k \times B_{\ell})\right| \\
&\leq \left| \int_{A_0 \times K} \widehat{T}\left(q \mid p\right) \mathrm{d}{\Bbb P} \right| + \sum_{\ell = 1}^{\ell_0} \sum_{k = 1}^{k_0} \int_{A_k \times B_{\ell}} \left|\widehat{T}\left(q \mid p \right) - \widehat{T}\left(q_{\ell} \mid p_k\right)\right| \mathrm{d} {\Bbb P} \\
&\leq \varepsilon \max_{p, q \in K} \widehat{T}\left(q \mid p\right) + \varepsilon.
\end{split}$$ Similarly, we have $$\label{eqn:estimate2}
\begin{split}
& \left| \int \widehat{T}^{(N)}\left(q \mid p\right)\mathrm{d}{\Bbb P}_N - \sum_{\ell = 1}^{\ell_0} \sum_{k = 1}^{k_0} \widehat{T}\left(q_{\ell} \mid p_k\right) {\Bbb P}_N(A_k \times B_{\ell}) \right| \\
&\leq \varepsilon \max_{p, q \in K} \widehat{T}^{(N)}\left(q \mid p\right) + \varepsilon.
\end{split}$$
By and uniform convergence of $\{\widehat{\Phi}^{(N)}\}$ on $K$, we can bound $\max_{p, q \in K} \widehat{T}\left(q \mid p\right)$ and $ \max_{p, q \in K} \widehat{T}^{(N)}\left(q \mid p\right)$ by a constant $C$. Using (i) and (iii), we get $$\label{eqn:estimate3}
\begin{split}
& \left| \sum_{k, \ell} \widehat{T} \left(q_{\ell} \mid p_k \right) {\Bbb P}_N(A_k \times B_{\ell}) -
\sum_{k, \ell} \widehat{T} \left(q_{\ell} \mid p_k \right) {\Bbb P}(A_k \times B_{\ell}) \right| \\
&\leq \sum_{k, \ell} \widehat{T}\left(q_{\ell} \mid p_k\right) \left| {\Bbb P}_N(A_k \times B_{\ell}) - {\Bbb P}(A_k \times B_{\ell})\right| \\
&\leq k_0\ell_0C\frac{\varepsilon}{k_0\ell_0} = C\varepsilon.
\end{split}$$ Combining , and , we have the estimate $$\left|\int \widehat{T}^{(N)}\left(q \mid p\right) \mathrm{d}{\Bbb P}_N - \int \widehat{T}\left(q \mid p\right) \mathrm{d}{\Bbb P}\right| \leq (3C + 2)\varepsilon, \quad N \geq N_0,$$ and so holds.
It remains to construct the sets $\{A_k\}$, $\{B_{\ell}\}$, the points $p_k$, $q_{\ell}$ and $N_0$ satisfying (i)-(iv). Before we begin, we note the fact that the boundary of any convex subset of $\Delta^{(n)}$ has $m$-measure zero [@L86 Theorem 1]. Let $\varepsilon > 0$ be given. By [@R70 Theorem 10.6], the family $\{\widehat{\Phi}, \widehat{\Phi}^{(1)}, \widehat{\Phi}^{(2)}, ...\}$ is uniformly Lipschitz on $K$. Also, it is not difficult to verify that there exists a constant $L > 0$ so that $$\left|\log \left(1 + \left \langle \frac{x}{p}, q - p \right\rangle\right) - \log \left(1 + \left \langle \frac{x}{p'}, q' - p' \right\rangle\right)\right| \leq L\left(\|p - p'\| + \|q - q'\|\right)$$ for all $x \in \overline{\Delta^{(n)}}$ and $p, p', q, q' \in K$. It follows that the family of L-divergences $\{\widehat{T}, \widehat{T}^{(1)}, \widehat{T}^{(2)} ...\}$ is uniformly Lipschitz on $K \times K$. Thus there exists $\delta_0 > 0$ such that if $p, p', q, q' \in \Delta^{(n)}$, then $$\label{eqn:energyconverge}
\left|\widehat{T}^{(N)}\left(q' \mid p' \right) - \widehat{T}^{(N)}\left(q \mid p \right)\right| < \frac{\varepsilon}{2} \quad \text{and} \quad
\left|\widehat{T}\left(q' \mid p' \right) - \widehat{T}\left(q \mid p \right)\right| < \varepsilon$$ whenever $\|q - q'\| < \delta_0$, $\|p - p'\| < \delta_0$.
Let $D$ be the set of points in $K$ at which $\widehat{\Phi}$ is differentiable. Then $K \setminus D$ has $m$-measure zero by [@R70 Theorem 25.5]. Let $\varepsilon' > 0$ be arbitrary. By Lemma \[lem:subdiffconvergence\], for each $p \in D$ there exists $0 < \delta(p) \leq \delta_0$ and a positive integer $N_0(p)$ such that $\left\|\widehat{\pi}^N(q) - \widehat{\pi}(p)\right\| < \varepsilon'$ for all $N \geq N_0(p)$ and $q \in B(p, \delta(p))$.
Since $K$ is compact, it is separable, and so is $D$ as a subset of $K$. The collection $\{B(p, \delta(p))\}_{p \in D}$ forms an open cover of $D$ and hence there exists a countable subcover. By the continuity of measure, for any $\eta > 0$ there exists $p_1, ..., p_{j_0} \in D$ such that $$m(A_0) < \eta, \quad A_0 := K \setminus \bigcup_{j = 1}^{j_0} B(p_j, \delta(p_j)),$$ Since $\partial A_0 \subset \partial K \cup \bigcup_j \partial B(p_j, \delta(p_j))$, $\partial (A_0 \times K)$ has $m$-measure zero and hence $A_0 \times K$ is a ${\Bbb P}$-continuity set. Since ${\Bbb P}$ is absolutely continuous, choosing $\eta > 0$ sufficiently small we have $${\Bbb P}(A_0 \times K) < \varepsilon,$$ and by weak convergence we have ${\Bbb P}_N(A_0 \times K) < \varepsilon$ for $N$ sufficiently large, so (ii) holds. Let $A_1 = B(p_1, \delta(p_1)) \cap K$ and define $A_k = \{p_k\} \cup (B(p_k, \delta(p_k)) \cap K) \setminus (A_1 \cup \cdots \cup A_{k-1})$, $j = 2, ..., k_0$. If $N \geq \max_{1 \leq k \leq k_0} N_0(p_k)$, we have $$\label{eqn:estimate4}
\left\|\widehat{\pi}^N(p) - \widehat{\pi}(p_k)\right\| < \varepsilon', \quad p \in A_k, \quad k = 1, ..., k_0.$$ Next choose $q_1, ..., q_{\ell_0} \in K$ such that $K \subset \bigcup_{\ell = 1}^{\ell_0} B(q_{\ell}, \delta_0)$. Define $B_1 = B(q_1, \delta_0) \cap K$ and $B_{\ell} = \{q_{\ell}\} \cup (B(q_{\ell}, \delta_0) \cap K) \setminus (B_1 \cup \cdots \cup B_{\ell-1})$, $j = 2, ..., \ell_0$. Again it is clear that $\partial(A_k \times B_{\ell})$ has $m$-measure zero and is a ${\Bbb P}$-continuity set. So (i) holds for $N$ sufficiently large. Finally, if we choose $\varepsilon' > 0$ small enough in , we have $$\left|\widehat{T}^{(N)}\left( q \mid p \right) - \widehat{T}\left( q \mid p_k \right)\right| < \frac{\varepsilon}{2}, \quad p \in B(p_k, \delta_0), \quad q \in \Delta^{(n)}$$ for $N$ sufficiently large. This and imply (iii) and the proof of Theorem \[thm:optim2\] is complete.
[^1]: The author would like to thank Soumik Pal for his constant guidance and support during the preparation of the paper, Tatiana Toro for helpful discussions about the proof of Theorem \[thm:main\], and Jiashan Wang for help with numerical optimization. He also thanks the anonymous referee who spotted an error in the original definition of the support condition and suggested the current definition. The referee’s valuable comments improved greatly the presentation of the paper.
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- Manon Marchand
- Frédéric Restagno
- Emmanuelle Rio
- François Boulogne
bibliography:
- 'biblio.bib'
title: '**Roughness-induced friction in liquid foams**'
---
The flow of complex fluids – polymers [@Priestley2005; @Boukany2010], emulsions [@Goyon2008], granular materials [@Siavoshi2006; @Pouliquen1999], bubbly liquids [@Germain2016], foams [@Kabla2003; @Golemanov2008] – near a solid surface displays different behaviors depending on the surface properties. Often, these systems exhibit a slippage on smooth surfaces but deform and flow on rough surfaces. Therefore, asperities are commonly added on the measurement apparatus to suppress the interfacial dissipations [@Barnes1995; @Cloitre2017; @Cohen-Addad2013]. Here, we focus our study on the flow of foams for which the elementary elements, the bubbles, can be observed directly.
The friction of an elongated bubble in a smooth round capillary has been studied in the pioneering work of Bretherton at low capillary numbers $\mathrm{Ca} = \mu V / \gamma$, where $\mu$ and $\gamma$ are the dynamic viscosity and air-liquid surface tension of the solution, and $V$ is the meniscus velocity in the capillary [@Bretherton1961]. It is shown experimentally and theoretically that the stress follows a $\mathrm{Ca}^{2/3}$ power law. This prediction is also valid for a single bubble in a flat cell [@Park1984], and can be extended for foams [@Cantat2013]. The surface rheology of the foaming solution influences the power law, but in this study we work with a solution for which this issue is negligible [@Cantat2013; @Denkov2005]. Also, experimental measurements provide a correction accounting for the liquid fraction of the foam $\varphi_\ell$ [@Raufaste2009].
In rheology, the wall slippage of the probed material must be avoided to ensure that the dissipation is localized in the bulk of the sample. In the case of foams, a commonly adopted rule-of-thumb to prevent slippage consists in adding asperities larger than the bubble size on the walls [@Khan1988], and to assert either visually [@Katgert2010] or by showing that the rheological measurements do not depend on the confinement [@Denkov2005; @Cloitre2017], that there is no remaining slippage. This trial and error process is not always possible and the characterization of the effect of different roughness sizes has been studied for single wet bubbles on walls [@Germain2016], but not for dry liquid foams.
In this Letter, we explore systematically the effect of different sizes of wall asperities $a$ on the flow of dry liquid foams and we identify three friction regimes at zero, high or intermediate roughness. The contribution of the roughness size $a$, of the liquid fraction $\varphi_\ell$, and of the inserting velocity in the foam $V$ to the stresses are rationalized in each regime. The roughness size is shown to change the stresses over one order of magnitude. We show that the roughness factor $a/r_\mathrm{pb}$ is a single criterion for the crossover between wall and bulk dissipation regimes. Interestingly, this parameter is independent of the imposed velocity.
Our experiment consists in inserting horizontally a controlled rough surface in a dry monodisperse foam generated in a transparent container (Fig. \[fig:principle\]a and b). We produce the foam by blowing air through needles in a foaming solution. The obtained bubble radius is $R = 0.70 \pm 0.05$ mm, and the liquid fraction $\varphi_\ell$ of the foam can be varied between $0.03$ and $8$ % by changing the working height above the liquid-foam interface [@Maestro2013] (Fig. S2). For such dry foams, the radius of curvature of the Plateau borders is given by $r_\mathrm{pb} = R \sqrt{\varphi_\ell / 0.33}$ [@Koehler2000]. It varies in the range $0.02-0.4$ mm in our experiment. Model rough surfaces are obtained by gluing glass beads of mean radii $a$ on microscope glass slides (Fig. S1), Fig. \[fig:principle\]a). We insert the surfaces in foams at constant velocities $V$ from $0.5$ to $20$ mm/s, corresponding to capillary numbers $\mathrm{Ca}$ in the range $1.7\times 10^{-5}$ to $6.7 \times 10^{-4}$. Simultaneously, we measure the tangential force $F$ exerted by the foam on the surface (Fig. S3).
![Foam flow dynamics near a solid surface decorated with glass beads. (a) Side view of the surface of roughness size $a$ moving at constant speed $V$ in a foam of bubble radius $R$ and radius of curvature of Plateau borders $r_\mathrm{pb}$. The force $F$ exerted by the foam on the surface is recorded. (b) Photograph corresponding to the schematic (a). The scale bar is $5$ mm. (c) Raw force measurements as a function of time $t$ and position of the surface $p$ for one smooth and two patterned surfaces translating at $V=1$ mm/s from left to right. (d) Visualization of the Plateau borders in contact with the patterned surfaces for three different $a/r_\mathrm{pb}$. The bottom of each picture is the surface moving at $1$ mm/s. We mark the position $p_\mathrm{pb}$ of a Plateau border with a circle. A plastic event is circled in white. The scale bar represents $2$ mm. (e) Position of the Plateau borders $p_\mathrm{pb}$ for the three friction regimes, reported from (d). []{data-label="fig:principle"}](figure1.pdf){width="\linewidth"}
By tracking the position of the Plateau borders $p_\mathrm{pb}$ in contact with the patterned surface during the motion, we identify three distinct friction regimes (Fig. \[fig:principle\]d), which are discriminated by the dimensionless roughness factor $a / r_{\mathrm{pb}}$. The emergence of this parameter can be intuited. When a bead is smaller than the size of the Plateau border ${r_\mathrm{pb}}$, it enters the Plateau border and slides inside it whereas bigger beads experience a pinning force at the liquid air interface (Fig.\[fig:measurements\]e). For negligible roughness sizes compared to the characteristic size of a Plateau border, $a \ll r_\mathrm{pb}$, the Plateau borders slide on the solid surface (Supplementary Movie S1). For $a \sim r_\mathrm{pb}$, Plateau borders are anchored on the surfaces but occasionally jump back (Supplementary Movie S2), while for $a \gtrsim r_\mathrm{pb}$ Plateau borders remain anchored and the stress is released by plastic events between bubbles in the bulk of the foam (Fig. \[fig:principle\]d, e, Supplementary Movie S3). We will respectively refer to these regimes as sliding, stick-slip, and anchored regimes in the following discussion. For $a/ r_\mathrm{pb}$ between $1$ and $3$, the visualizations reveal that the stress is released by mixed slip or plastic events. This provides a first estimate of the transition between the stick-slip regime and the anchored one.
Typical force measurements as a function of time $t$, or equivalently of the position of the surface $p$, are presented for each regime in Fig. \[fig:principle\]c. The linear evolution of the force indicates that the probed phenomena are independent of the penetration depth of the surface, and that a constant stress value $\tau_\mathrm{p}$ can be extracted from $F(t)$ and $p(t)$ for each experiment (Fig. S5).
{width="\linewidth"}
We reproduce the experiment for various roughness sizes $a$, radius of curvature of Plateau borders $r_\mathrm{pb}$, and insertion velocities $V$. In Fig. \[fig:measurements\]a, we plot the stress $\tau_\mathrm{p}$ as a function of the roughness factor $a /r_\mathrm{pb}$. The different point colors represent the three regimes identified from the images. In this representation, the data for a single velocity collapses on a trend curve (Fig. S4). Complementary representations of the stick-slip and the anchored regimes presenting the influence of the asperity size $a$ and the velocity $V$ are provided in Fig. S7. Interestingly, tuning $a / r_\mathrm{pb}$ leads to a stress variation of about an order of magnitude, which highlights the crucial effect of the boundary conditions on stresses. In the following, we examine each regime to rationalize the dynamics.
The sliding regime, for which $a \ll r_\mathrm{pb}$, has already been investigated [@Bretherton1961; @Raufaste2009; @Cantat2013], and these studies validate experimentally that the stress at the solid wall can be written
$$\tau_\mathrm{slip} = 3.8 \frac{\gamma}{R} \varphi_\ell^{-0.25} \mathrm{Ca}^{2/3}.
\label{eq:bretherton}$$
In Fig. \[fig:measurements\]b, the stress measurements are plotted as a function of the prediction given by equation (\[eq:bretherton\]), which shows a good agreement and validates this experimental approach.
Now, we consider the anchored regime for the large roughness factors. The common description of the foam rheology suggests that for shear stresses lower than a yield stress $\tau_\mathrm{ys}$, the foam behaves as an elastic solid, whereas for stresses exceeding the yield stress, the foam flows. Phenomenologically, this behavior is described by a Hershell-Bulkley law $\tau_\mathrm{HB} = \tau_\mathrm{ys} + \eta(\dot{\gamma}) \dot{\gamma}$, where $\dot{\gamma} \sim V / \delta$ is the shear rate that scales as the velocity over a characteristic shear length $\delta$ and $\eta(\dot{\gamma})$ is the effective foam viscosity that depends on $\dot{\gamma}$ [@Denkov2009; @Cohen-Addad2013]. The ratio of the yield stress and the viscous term defines the Bingham number $\textrm{Bi} = \frac{\tau_\mathrm{ys}}{\eta(\dot{\gamma}) \dot{\gamma}}$ [@Bingham1916]. The yield stress value is described by a semi-empirical law $\tau_\mathrm{ys} = K \frac{\gamma}{R} (\varphi_\mathrm{c} - \varphi_\ell)^2$, where $\varphi_\mathrm{c} = 0.26$ is the fraction of gaps remaining in a close-packing of hard spheres [@Maestro2013], and $K$ is a proportionality factor which reported values vary between $0.5$ and $6$ [@Lexis2014; @Rouyer2005]. Since we are working with dry foams, we have $\varphi_\ell \ll \varphi_\mathrm{c}$, which leads to
$$\tau_\mathrm{ys} = K \frac{\gamma}{R}\varphi_\mathrm{c}^2.
\label{eq:yield_stress}$$
In the limit of low velocities, corresponding to $\mathrm{Bi} > 1$, the contribution of the yield stress is dominant (see SI). The Fig. \[fig:measurements\]d shows that the stress applied by the foam on the rough surfaces is nearly independent on the velocity and its value is predicted by the equation (\[eq:yield\_stress\]) with a prefactor $K = 2.9 \pm 0.3$, which is in agreement with previously recorded values [@Lexis2014; @Rouyer2005]. We notice that the measured stress increases slightly for the larger velocities, which is reminiscent of the viscous term in the Hershell-Bulkley model.
![Edible foams and texture sensation. (a) Albumen foam in a spoon. The scale bar is $1$ cm. A drop of food dye is added for visualization. (b) Rough surface with beads of mean radius $a = 225$ $\mu$m. The scale bar is $5$ mm. (c) Close up picture that shows the papillae on a human tongue. The scale bar is $5$ mm. (d) Albumen foam sheared between a table spoon and a glass surface covered with beads of mean radius $a = 225$ $\mu$m. The left side is a wet foam with $\varphi_\ell \simeq 0.7$ %, corresponding to $\tilde{a} \simeq 0.5$. The right side is a dryer foam with a liquid fraction $\varphi_\ell \simeq 0.02$ %, meaning $\tilde{a} \simeq 3$. In those two cases the last bubble on the surface is circled to enlighten the motion of foam on the model tongue. There is a $1$ s interval between two consecutive images. The scale bar is $1$ cm. []{data-label="fig:application"}](figure3.pdf){width="\linewidth"}
Up to this point, we analyzed the two extreme limits of negligible and large dimensionless roughness factor $a/r_{\mathrm{pb}}$. For roughness sizes comparable to the radius of the Plateau borders, we observe a stick-slip regime that has been observed but not characterized [@Buchgraber2012], to the best of our knowledge. Our experiments indicate that the transition between the stick-slip and the anchored regimes is continuous (Fig. \[fig:measurements\]a). This suggests that the stick-slip regime is a combination of sliding, where the energy is released by viscous dissipation in the Plateau borders, and anchoring, where the energy is stored by deforming elastically the foam.
In the stick-slip regime, (Fig. \[fig:measurements\]e) an asperity moving in a foam goes through the liquid air interface of a Plateau border over a duration $t_\mathrm{anchored} \propto a/V$, experiencing a pinning force. Then, the asperity moves inside the Plateau border over a typical duration $t_\mathrm{slip} \propto r_\mathrm{pb} / V$. Therefore, we introduce the ratio of these durations, $\Tilde{a} = a/ k r_\mathrm{pb}$, where the prefactor $k$ represents the value of the transition. This two-step description happens for each asperity on the whole probing surface. The stick-slip events are not occurring simultaneously (Supplementary Movie S2), thus we average the individual behaviors of the Plateau borders. Hence, we construct a leverage rule stating that the total stress is the sum of the contributions of a pinning stress for a relative duration $\Tilde{a}$ and of a viscous stress for the complementary relative duration $(1-\Tilde{a})$. This writes
$$\tau_\mathrm{stick-slip} = \tau_\mathrm{ys} \Tilde{a} + \tau_\mathrm{slip}(1-\Tilde{a}).
\label{eq:stick-slip}$$
To determine more precisely the value of the transition $k$, we optimized equation (\[eq:stick-slip\]) against this parameter and we obtained a transition at $k=1.9 \pm 0.2$ (Fig. S6a). This value of order unity is in agreement with our estimation of the transition from direct visualizations of the Plateau border dynamics (see the orange and gray areas in Fig. \[fig:measurements\]a). As shown in Fig. \[fig:measurements\]c, the proposed equation (\[eq:stick-slip\]) is in excellent agreement with the stress measurements for the different liquid fractions and surface velocities explored in this study. In the vicinity of small velocities, we expect $\tau_\mathrm{stick-slip} \underset{V \to 0}{\rightarrow} \tau_\mathrm{ys} \Tilde{a}$, and this is observed in Fig. \[fig:measurements\]a and Fig. S6b.
Thus, we identified a practical criterion $\tilde{a}$ to discriminate between surface and bulk dissipation for the friction of a foam on a solid surface. The anchored regime is of particular interest for rheological studies focusing on the flow of foams where a no slip condition is necessary to transfer shear to the bulk of the material. It was already known that adding some asperities by gluing sand paper or engraving grooves on a measuring apparatus reduces the wall slippage [@Khan1988], but the size of these asperities had not been tuned systematically. Our study validates this approach, since the stress in the anchored regime are independent of the size of the asperities beyond a critical roughness factor (Fig. \[fig:measurements\]a). When the grit size of the sand paper is insufficient to totally eliminate the wall slippage, the mixed surface and bulk dissipation leads to a difficult interpretation of the data [@Golemanov2008]. Differences in the measured stresses are also reported when the grit size is increased [@Jimenez-Junca2011]. This behavior is likely the stick-slip regime we report in Fig. \[fig:measurements\]c. Our contribution allows one to choose a suitable roughness to be added on the measuring systems regarding the foam properties. The results then assess that the obtained measurements in the anchored regime will be uncorrelated with the size of the roughness.
The stick-slip regime allows, for instance, to understand the mastication of an aerated food (Fig. \[fig:application\]a). Adding air bubbles in a food product affects mouthfeel perception, improves digestibility, and aids mastication [@Campbell1999]. In the case of eating disorders, such as dysphagia, more viscous foods are proven to be safer [@Nishinari2019]. With viscosities typically $10^3$ times bigger than the viscosity of water [@Marze2005], foams could help patients. We make a visualization experiment to illustrate the tunability of perception during the mastication of a foamed food. Human tongue asperities have a fixed size, therefore the remaining adjustable parameters in this case are the bubble size and the liquid fraction of the foam. A glass slide covered with asperities of mean radius $a = 225$ $\mu$m, a size comparable to human papillae [@Essick2003], can be considered as a model human tongue (Fig. \[fig:application\]a, b and c). We produce two edible albumen foams with a bubble size $R=0.7$ mm and two different liquid fractions, using the same technique than soap foams, but with a solution containing $1$ g of egg white powder (Louis François) and $200$ mL of deionized water. The two foams are sheared between a table spoon moving at $10$ mm/s and the static model tongue. In the two cases, the last bubble is circled to illustrate the slippage of the foam (Fig. \[fig:application\]d). For the wetter foam for which $\tilde{a} \simeq 0.5 < 1$, the conditions predict a stick-slip regime and the foam moves on the surface (Supplementary Movie S4), whereas for the dryer foam, for which $\tilde{a} \simeq 3 > 1$, the foam sticks to the surface (Supplementary Movie S5). This illustrates that the results of our controlled experiment are valid in a more general context, with different shearing geometries and surfactants. When the normalized roughness $\tilde{a}$ varies from $0$ to $1$ in the stick-slip regime, we see in Fig. \[fig:measurements\]a that the stress varies between $0.8$ and $9.8$ Pa, a range of stresses detectable by a human tongue [@Foegeding2015; @Linne2017]. Thus, tuning the macroscopic properties of a product, namely the bubble size and the liquid-gas ratio, must lead to different sensations in mouth.
In conclusion, we reveal that the foam flow near surfaces exhibit different behaviors: sliding, anchored, and stick-slip. By recording the stresses exerted during these regimes, we show that there exists a roughness, normalized by the curvature radius of the Plateau borders of the foam, $\tilde{a} = a / k r_\mathrm{pb}$ beyond which the dissipation is transferred from the surface to the bulk. This transition does not depend on the imposed velocity in the explored range. In the stick slip regime, we propose a leverage rule describing the influence of speed, radius of curvature of Plateau borders, and liquid fraction on the stresses exerted on the walls. We measure that these stresses vary on one order of magnitude when the normalized roughness is increased. Therefore, we contribute to an improved description of wall roughness effect on flowing foams, which is of particular interest to design surfaces given the foam properties in industrial applications. In contrast, when surface asperities are unchangeable, as in the case of tongue papillae, our description of friction forces allows to finely tune the properties of edible foams to get creamy or gooey mouth feelings. This is useful for inventing new foamed food products, either for the ease of feeding for dysphagic patients, or for the gourmets’ pleasure.
#### Acknowledgements
M.M. acknowledges EDPIF for funding support. We thank J. Sanchez for programming the computer interface of the experiment, and G. Guillier for support manufacturing the force sensor. We thank B. Dollet and C. Raufaste for discussions.
{width="\textwidth"}
{width="\textwidth"}
{width="\textwidth"}
{width="\textwidth"}
{width="\textwidth"}
{width="\textwidth"}
{width="\textwidth"}
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
A sample of edge-on spiral galaxies aimed at a thorough study of the main structural and photometric parameters of edge-on galaxies both of early and late types is presented. The data were taken from the Two Micron All Sky Survey (2MASS) in the $J$, $H$ and $K_s$ filters. The sources were selected according to their angular size mainly on the basis of the 2MASS-selected Flat Galaxy Catalog (2MFGC). The sample consists of 175 galaxies in the $K_s$-filter, 169 galaxies in the $H$-filter and 165 galaxies in the $J$-filter. We present bulge and disc decompositions of each galaxy image. All galaxies have been modelled with a [Sérsic]{} bulge and exponential disc with the BUDDA v2.1 package. Bulge and disc sizes, profile shapes, surface brightnesses are provided.
Our sample is the biggest up-to-date sample of edge-on galaxies with derived structural parameters for discs and bulges. In this paper we present the general results of the study of this sample. We determine several scaling relations for bulges and discs which indicate a tight link between their formation and evolution. We show that galaxies with bulges fitted by the [Sérsic]{} index $n \la 2$ have quite different distributions of their structural parameters than galaxies with $n \ga 2$ bulges. At a first approximation the [Sérsic]{} index threshold $n \simeq 2$ can be used to identify pseudobulges and classical bulges. Thus, the difference in parameter distributions and scaling relations for these subsamples suggests that two or more processes are responsible for disk galaxy formation. The main conclusions of our general statistical analysis of the sample are:
\(1) The distribution of the apparent bulge axis ratio $q_\mathrm{b}$ for the subsample with $n \la 2$ can be attributed to triaxial, nearly prolate bulges that are seen from different projections, while $n \ga 2$ bulges seem to be oblate spheroids with moderate flattening. Triaxiality of late-type bulges may be due to the presence of a bar that thickened in the vertical direction during its secular evolution.
\(2) For the sample galaxies, the effective radius of the bulge $r_\mathrm{e,b}$, the disc scalelength $h$ and the disc scaleheight $z_0$ are well correlated. However, there is a clear trend for the ratio $r_\mathrm{e,b}/ h$ to increase with $n$. As $n$ is an indicator of the Hubble type, such a trend unambiguously rules out the widely discussed hypothesis of a scale-free Hubble sequence. The found correlation between $z_0$ and $r_\mathrm{e,b}$ is new and was not described earlier.
\(3) There is a hint that the fundamental planes of discs, which links only disc parameters and the maximum rotational velocity of gas, are different for galaxies with different bulges. This may indicate a real difference of discs in galaxies with low and high concentration bulges.
\(4) The most surprising result arises from the investigation of the Photometric Plane of sample bulges. It turns that the plane is not flat and has a prominent curvature towards small values of $n$. For bulges this fact was not noticed earlier.
\(5) The clear relation between the flattening of stellar discs $h/z_0$ and the relative mass of a spherical component, including a dark halo, is confirmed not for bulgeless galaxies but for galaxies with massive bulges.
Many of our results are in good agreement with the results of other authors, several ones are new. Thus, our sample is very useful for further detailed studying and modelling of the edge-on spiral galaxies.
author:
- |
A. V. Mosenkov, N. Ya. Sotnikova[^1] and V. P. Reshetnikov\
St.Petersburg State University, Universitetskij pr. 28, 198504 St.Petersburg, Stary Peterhof, Russia\
and Isaac Newton Institute of Chile, St.Petersburg Branch
date: 'Accepted 2009 September 6. Received September 6; in original form 2009 July 2'
title: '2MASS photometry of edge-on spiral galaxies. I. Sample and general results'
---
\[firstpage\]
galaxies: fundamental parameters — galaxies: general — galaxies: infrared — galaxies: spiral edge-on galaxies — galaxies: photometry.
Introduction
============
The photometric and kinematic study of spiral galaxies provides a method to tackle the problems concerning their formation and evolution. A lot of observational data has been obtained in recent decades, especially with the advent of CCD arrays. Over the last years, significant progress has been made in both observational and theoretical studies aimed at understanding the structure and evolution of galaxies. Modern sky surveys (2MASS, SDSS, etc.) provide information on the characteristics and spatial distribution of millions of galaxies. The large available amount of information has altered the face of modern extragalactic astronomy. Many important tasks, including detailed photometric studies of the structure of nearby galaxies, can be now solved without additional observations by telescopes. Here we focus on the surface photometry of spiral galaxies on the basis of the 2MASS near-infrared survey.
The parametrization of galaxies is a way to describe the multitude of normal galaxies without any peculiarities within their structure. Various fitting functions are used to derive structural parameters of bulges and discs. Traditional fitting functions for stellar discs include an exponential function (de Vaucouleurs 1959, Freeman 1970) and an inner-truncated exponential function (Kormendy 1977). Bulges are well described by the so-called [Sérsic]{} profile that includes the de Vaucouleurs profile as a special case (Sérsic 1968). The de Vaucouleurs law, or $r^{1/4}$-law, was obtained empirically for elliptical galaxies (de Vaucouleurs 1948) and later for bulges of early-type galaxies (de Vaucouleurs 1959). Later it was shown that the [Sérsic]{} law is preferable to fit spheroidal components of galaxies (e.g. Andredakis et al. 1995). The $z$-distribution of surface brightness of edge-on discs is exponential or the so called “sech$^2$-distribution” (van der Kruit & Searle 1981). (All the mentioned functions are described in Section 3 of this paper.)
In order to derive structural parameters of bulges and discs, we must perform photometric bulge/disc decomposition of galaxy images. There are numerous methods of such decomposition. The one-dimensional analysis of the surface-brightness distribution is the most commonly used method (e.g. Boroson 1981; Kent 1985; Baggett, Baggett & Anderson 1998; Bizyaev & Mitronova 2002, hereafter BM02). But as was shown by Byun & Freeman (1995) the one-dimensional decomposition procedure leads to strong systematic errors. The two-dimensional approach provides much better estimates of the structural parameters for each component of a galaxy. This method is powerful when we need to separate the non-axisymmetric structures (bars, rings, spiral arms) from axisymmetric components (disc and bulge).
In recent decades a number of studies concerning structural properties of galaxies has been performed. The results of decompositions of near face-on disc dominated galaxies as well as of early-type galaxies were presented, for instance, by de Jong (1996). It was concluded that there is a correlation between structural parameters of discs and bulges. Möllenhoff & Heidt (2001) found similar correlations and speculated about the formation and evolution of spiral galaxies. Graham (2001) performed a detailed analysis of low-inclination spiral galaxies and described many significant properties of their bulges and discs. Other studies dealt with the correlation between the scalelengths of bulges and discs (e.g. Courteau et al. 1996; de Jong 1996), the presence of additional morphological components such as bars, lenses, rings, inner bars and inner discs (e.g. de Souza, Gadotti & dos Anjos 2004, SGA04 hereafter; Castro-Rodríguez & Garzón 2003; Laurikainen, Salo & Buta 2005; Gadotti 2009) and the revision of morphological classification of galaxies (SGA04). The fundamental plane (FP) of discs as well as the photometric plane of bulges were also widely discussed (e.g. Khosroshahi et al. 2000a,b; Möllenhoff & Heidt 2001; Moriondo, Giovanelli & Haynes 1999).
The main aim of our project is to perform decompositions of a large number of edge-on galaxies both of early and late types into bulges and discs. We chose this type of galaxies because they provide a unique possibility to obtain information about the vertical structure of galactic discs. In the edge-on view non-axisymmetric features such as bars, rings or spiral arms are not seen. On the other hand, the vertical distribution of light in edge-on galaxies allows us to study a thin disc, a thick disc and a stellar halo (see e.g. Yoachim & Dalcanton 2005; Seth, Dalcanton & de Jong 2005).
Unfortunately, there are some difficulties in studying edge-on galaxies. In the optical band we need to take into account the dust extinction in the plane of a disc. The distribution of dust is often unknown and it is hard to obtain true values of disc and bulge parameters. The advantage of NIR observations is that the general extinction is drastically reduced, although we can not exclude it completely. Nevertheless it is better to perform the decomposition of galaxies by using their images in red and infrared bands. Using three IR bands for galaxy decomposition and subsequent statistical analysis of galaxy properties in all three bands encourages us in our suggestion that extinction problems are not very severe (at least for general statistical correlations which are the subject of the present work). However, edge-on galaxies are rare in occurrence in contrast to face-on galaxies or galaxies with intermediate inclination angles. Therefore, in order to create a large sample of edge-on galaxies it is necessary to use special catalogues which include a lot of edge-on spiral galaxies such as the Revised Flat Galaxy Catalog (Karachentsev et al. 1999).
A detailed study of the disks of edge-on galaxies is described in many papers (e.g., van der Kruit & Searle 1981; Reshetnikov & Combes 1997; de Grijs 1998). In a more recent study, Dalcanton & Bernstein (2000, 2002) examined a sample of extremely late-type, edge-on spirals with no apparent bulges and concluded that thick disks are common around galaxy disks of all masses. BM02 also analyzed a sample of late-type edge-on galaxies in the $J$, $H$ and $K_s$-bands to compare the mean ratios $h/z_0$ in these bands. They have noted a strong correlation between the central surface brightness of the disk and the $h/z_0$ ratio: the thinner the galaxy, the lower the central surface brightness reduced to the face-on inclination. Kregel, van der Kruit & de Grijs (2002), Kregel (2003), Kregel, van der Kruit & Freeman (2005) and Zasov et al. (2002) presented a complex study (dynamical properties of the stellar discs, three-dimensional disc structure, stellar kinematics, rotation curves) of several dozens of edge-on spirals. Unfortunately, there are no extended studies of structural parameters of edge-on galaxies both of early and late types which join analysis of their bulges and discs. Our study is intended to fill this gap. We analyse the statistical properties of the sample edge-on galaxies and present correlations between the global structural parameters of [*bulges*]{} and [*discs*]{}, including those that describe their vertical structure, in the near-infrared bands.
This paper is organized as follows. In Section 2 we present the detailed description of our sample of spiral edge-on galaxies which includes the selection criteria and a brief description of the data sources. We also describe the completeness of the sample and its main properties. In Section 3 we use the two-dimensional least-squares algorithm to decompose the galaxy images into a bulge and a disc. We recall some aspects of the code BUDDA (SGA04) which we used as the decomposition tool. In Section 4, we present the results of image decomposition in the $J$, $H$ and $K_s$-bands. We compare our results with previous results from the literature and analyse the reliability of the results of the BUDDA decompositions. In Section 5 we describe general properties of edge-on spiral galaxies and discuss some correlations between galactic subsystems. Some of them are well-known but there are also some new results. In Section 6 we summarize our main conclusions. In the forthcoming papers we are going to discuss properties of edge-on galaxies in more detail.
The Sample
==========
We intended to find edge-on galaxies with a wide range of bulges: from bulge-dominated spirals to galaxies with small bulges. We used the 2MASS-selected Flat Galaxy Catalog (2MFGC) (Mitronova et al. 2003) as a source of objects and the 2MASS survey (Skrutskie et al. 2006) as a source of image data. We also used the Revised Flat Galaxies Catalog (RFGC) by Karachentsev et al. (1999) to add late-type galaxies to our sample.
The Two Micron All Sky Survey is one of the best known digital surveys in a wavelength range close to the optical one. This survey covers the whole sky in the filters $J$ (1.25 $\mu$m), $H$ (1.65 $\mu$m), and $K_s$ (2.16 $\mu$m). Unfortunately, the 2MASS survey has a weak sensitivity to late-type galaxies, especially of low surface brightness (Jarrett 2000). This is due to the high brightness of the night sky in the near-IR range and short exposures. For this reason the periphery of the discs of spiral galaxies is generally unseen beyond the isophotes fainter than $\mu_K=20$ mag arcsec$^{-2}$. We did not study truncation radii for the 2MASS-galaxies and did not take them into account performing the decomposition.
The sample was selected mainly from the 2MFGC. This catalogue contains 18020 disc-like galaxies covering all the celestial sphere. The objects were taken from the Extended Source Catalog of the 2MASS (XSC 2MASS) according to their 2MASS axial ratio $a/b \geq 3$. The 2MFGC gives value of axial ratio $b/a$, fiducial elliptical Kron magnitude, position angle, the $J$-band concentration index ($IC$) and other useful information. Concentration index is the ratio of the radii of circles that contain 3/4 and 1/4 of the measured flux in the same band, respectively. $IC$ together with $b/a$ are known to serve as dividing lines between early- and late-type galaxies (see e.g. Strateva et al. 2001; Nakamura et al. 2003). We decided to select galaxies both of early and late types in almost equal quantities. Our criteria are
- axial ratio for the combined $J+H+K$ image, $sba>0.2$;
- $K$-band fiducial elliptical Kron radius, $R_K\geq30''$;
- Hubble type ranging from S0 to late types;
- non-interacting galaxies;
- concentration index $IC>2.0$.
The latter item means that we exclude galaxies with strongly peculiar surface brightness distributions since for pure exponential disks $IC=2.8$ and for $r^{1/4}$ ellipticals $IC=7$ (Fraser 1972).
From the set of 3167 objects which contains galaxies with pronounced bulges (Sa, Sb types) and galaxies with small bulges (Sc, Scd types) we selected a sufficient number of regular spiral galaxies to make statistical analysis of edge-on galaxies properties. We did not add to the sample galaxies that are clearly lopsided or warped (with some exceptions below). In addition, on the basis of careful visual inspection we rejected non-edge-on and peculiar objects. Also, we added 13 bright galaxies that were absent in the 2MFGC and several edge-on galaxies which were not selected according with foregoing criteria but were included in the sample of RFGC galaxies by BM02. We chose them to compare the results of our decomposition with those obtained by BM02.
The final sample consists of 67 early-type galaxies (S0-Sab), 61 late-type galaxies (Sbc-Sd) and 47 galaxies of intermediate type Sb. The absolute numbers of galaxies of various morphological types (taken from the LEDA) and their percentages are listed in the Table \[Types\]. The full list of the sample galaxies studied here is presented in the Appendix, where we provide the basic information about them. According to LEDA 23 galaxies have bars, 6 galaxies have rings, and 3 objects are interacting galaxies. We did not take into account additional components like bars performing the decomposition, but checking all correlations between structural parameters, we did not find any strong deviation of the above-mentioned objects from derived trends.
We also did not reject 3 interacting galaxies and several other galaxies with some unusual structural features (double exponential discs, rings, etc., which will be described in our resulting tables) to have about 10 unusual spiral edge-on galaxies for further work.
As can be seen in the Table \[Types\] the sample contains a substantial percentage of lenticular galaxies. It is necessary to notice that Hubble classification for edge-on galaxies is very subjective because the spiral arm structure is not seen in this case. The only way to classify edge-on galaxies according to morphological types is the estimation of the ratio of the total luminosities of a bulge and a disc (hereafter $B/D$). We assume that a large fraction of classified S0 galaxies are Sa galaxies in fact. In Fig. \[GALexamples\] the images of typical sample galaxies are shown. According to the LEDA data, our sample galaxies are, on average, true highly-inclined galaxies with the mean inclination $i = 87.^{\circ}8 \pm 4.^{\circ}8$. Of course, this estimate can be too optimistic but a moderate deviation from edge-on orientation does not significantly change the slope of the vertical surface brightness distribution (Barteldrees & Dettmar 1994).
Type Number Percentages
---------- -------- -------------
S0, S0-a 41 23.4
Sa 11 6.3
Sab 15 8.6
Sb 47 26.9
Sbc 21 12.0
Sc 36 20.5
Scd 4 2.3
Total 175 100
![Examples of the sample galaxies in the $K_s$-band. The upper image is a typical galaxy of S0-a type: ES0 311-G012. The middle image is NGC 4013 of Sb type. The bottom image is an example of Sc galaxy: NGC 5907.[]{data-label="GALexamples"}](fig1a.ps "fig:"){width="6cm"} ![Examples of the sample galaxies in the $K_s$-band. The upper image is a typical galaxy of S0-a type: ES0 311-G012. The middle image is NGC 4013 of Sb type. The bottom image is an example of Sc galaxy: NGC 5907.[]{data-label="GALexamples"}](fig1b.ps "fig:"){width="6cm"} ![Examples of the sample galaxies in the $K_s$-band. The upper image is a typical galaxy of S0-a type: ES0 311-G012. The middle image is NGC 4013 of Sb type. The bottom image is an example of Sc galaxy: NGC 5907.[]{data-label="GALexamples"}](fig1c.ps "fig:"){width="6cm"}
The distances of the galaxies are taken from NASA/IPAC Extragalactic Database (NED) with the Hubble constant $H_0=73$ km s$^{-1}$/Mpc, $\Omega_\mathrm{{matter}}=0.27$, and $\Omega_{\mathrm{vacuum}}=0.73$). The luminosity distances are in the range from 4.3 Mpc (NGC 891) to 171 Mpc (ESO 251-G028) with a sample median value of 40 Mpc. Three galaxies (ESO 013-G024, ESO 555-G023, ESO 555-G032) have no measured or estimated distances. We do not take them into account in statistical analysis where the distances are required.
To investigate some kinematical and dynamical properties of the sample galaxies we added information about maximum rotation velocity for each galaxy if it was presented in the LEDA data base (see the section 5.3 for more details).
The sample consists of 175 galaxies in the $K_s$-filter, 169 galaxies in the $H$-filter and 165 galaxies in the $J$-filter. (Several galaxies have prominent dust lanes within their discs and even in the $J$ and $H$ passbands absorption of the dust is significant.)
In order to evaluate the completeness of the sample, we use the method proposed by Schmidt (1968) (the so-called $V/V_\mathrm{max}$ completness test). Following Thuan & Seitzer (1979) we used this method for our angular-size-limited sample. The full sample is incomplete but the subsample of galaxies with angular radius $r\geq 60''$ is complete ($V/V_\mathrm{max}=0.49 \pm 0.03$). This subsample consists of 92 galaxies (47 early-type galaxies and 45 late-type galaxies).
The images used in this work were taken from the database of the All-Sky Release Survey Atlas. The Atlas’ images are FITS standard images in 512 $\times$ 1024 pixel format (with 1.$^{''}$00 pixel$^{-1}$). Description of the images such as the pixel gain, readout noise, seeing FWHM ($\sim2.''5$), sky value and also zero points are available in the 2MASS webpages. This information has been used in making decompositions of galaxy images. Each image was thoroughly examined in order to mask background stars and galaxies and other contaminating sources. All the images were rotated to align the galaxy major axis parallel to the horizontal image borders. These procedures were carried out in three filters so all images of the same galaxy in the $J$, $H$ and $K_s$ bands have the same extension and position angles. The smallest image of our sample is 48$\times$18 pixels for ESO 555-G032 and the largest image is 869$\times$165 pixels for NGC 4565. The median value of the size of galaxy images is 119$\times$27 pixels, that is enough to determine the main photometric parameters of galaxy discs and bulges. The forthcoming analysis was performed with the MIDAS (developed by the European Southern Observatory) and the BUDDA packages.
Two-dimensional bulge/disc decomposition
========================================
BUDDA (Bulge/Disc Decomposition Analysis) is a code devised to perform two-dimensional decomposition of galactic images. This code was produced by R.E. de Souza, D.A. Gadotti and S. dos Anjos (for more information about BUDDA see SGA04). Since the year of 2004 BUDDA has been publicly available to the astronomical community[^2]. We use BUDDA v2.1 to produce the fits, allowing the inclusion of bulges, discs, bars and AGN in the models.
The input data for the code is an image of a galaxy, consisting of two major components: a bulge and a disc. The disc is represented by an exponential distribution of the luminosity density $I(r,z)$ with the central luminosity density $I(0,0)$, the scalelength $h$ and the ‘isothermal’ scaleheight $z_0$: $$I(r,z) = I(0,0) \, e^{-r/h} \, \hbox{sech}^2(z / z_0) \, ,
\label{form1}$$ where $(r,z)$ are the cylindrical coordinates. The disc is assumed to be axisymmetric and transparent.
The surface brightness distribution for the face-on disk ($i = 0^\circ$) can be expressed as follows: $$I_\mathrm{d}(r) = I_\mathrm{0,d} \, e^{-r / h}, \,
\label{form2}$$ where $I_\mathrm{0} = 2 \, z_0 \, I(0,0)$. The same expression in magnitudes per arcsec$^2$ is: $$\mu_\mathrm{d}(r) = S_\mathrm{0,d} + 1.086 \, r/h, \,
\label{form3}$$ where $S_\mathrm{0,d}$ is the central surface brightness of the face-on disc. For edge-on galaxies ($i = 90^\circ$) the following expression is valid: $$I(r,z) = I(0,0) \,
\frac{r}{h} \, K_1\left(\frac{r}{h}\right) \, \hbox{sech}^2(z / z_0)\, ,
\label{form4}$$ where $K_1$ is the modified Bessel function of the first order.
The bulge surface brightness profile is described by the [Sérsic]{} law (Sérsic 1968): $$I_\mathrm{b}(r) = I_\mathrm{0,b} \,
e^{-\nu_\mathrm{n}[(r/r_\mathrm{e,b})^{1/n}]}\, ,$$ where $r_\mathrm{e,b}$ is the effective radius of the bulge, i.e., the radius of a circle that contains 50% of the total galaxy luminosity, $I_\mathrm{0,b}$ is the bulge central surface brightness, $n$ is the [Sérsic]{} index, defining the shape of the profile, and the parameter $\nu_n$ ensures that $r_\mathrm{e,b}$ is the half-light radius. In magnitudes per arcsec$^2$ the expression looks as follows: $$\mu_\mathrm{b}(r) = \mu_\mathrm{0,b} +
\frac{2.5\nu_\mathrm{n}}{\ln 10} \,
\left(\frac{r}{r_\mathrm{e,b}}\right)^{1/n}\, ,
\label{form6}$$ where $\mu_\mathrm{0,b}$ is the bulge effective surface brightness expressed in mag per arcsec$^2$, i.e., the surface brightness at $r_\mathrm{e,b}$. The BUDDA code uses a numerical approximation of $\nu_n \simeq (0.868n-0.142) \, \ln 10$. The [Sérsic]{} index of $n=4$ represents the de Vaucouleurs profile which was very popular to describe the surface brightness distribution of bright elliptical galaxies and of bulges in early-type spirals. The [Sérsic]{} index of $n=1$ represents the exponential profile of bulges in late-type spirals, of galactic discs and of dwarf elliptical galaxies (see e.g. de Vaucouleurs 1948, 1959; Freeman 1970; Graham 2001 and references therein).
It is also useful to determine such parameters as ellipticities ($\epsilon = 1 - b/a$) of a bulge $\epsilon_\mathrm{b}$ and a disc $\epsilon_\mathrm{d}$, as well as their position angles $P.A._\mathrm{b}$ and $P.A._\mathrm{d}$. To account for the effect of seeing there is a Moffat smearing (Moffat 1969) of the brightness profiles in the code controlled by a parameter associated with the seeing radius $a_\mathrm{s}$. There is also included a correction term of the sky level $\Delta_\mathrm{sky}$. The isophotes of bulges and discs can be described by generalized ellipses with the ellipse index parameter $\gamma$ controlling their shape: $$\left(\frac{|x|}{a}\right)^{\gamma + 1} +
\left(\frac{|y|}{b}\right)^{\gamma + 1} = 1,
\label{form36}$$ where $x$ and $y$ are the pixel coordinates of the ellipse points, $b$ and $a$ are the sizes of its semi-major and semi-minor axes, respectively. When $\gamma = 1$ the ellipse is simple, while for $\gamma > 1$ the ellipse is boxy and for $\gamma < 1$ the ellipse is discy.
For edge-on galaxies BUDDA derives the disc scaleheight $z_0$ instead of the disc ellipticity $\epsilon_\mathrm{d}$. Thus, a total number of the main parameters for the bulge/disc model is 11: the center coordinates $x_0$ and $y_0$, the central surface brightnesses of a bulge $I_\mathrm{0,b}$ (or the effective surface brightness of bulge $I_\mathrm{e,b}$ as used in BUDDA) and a disc $I_\mathrm{0,d}$ (both in ADU), $r_\mathrm{e,b}$, $h$, $P.A._\mathrm{b}$, $P.A._\mathrm{d}$, the disc scaleheight $z_0$, the bulge ellipticity $\epsilon_\mathrm{b}$ and the [Sérsic]{} index of a bulge $n$.
Total luminosity of a disc can be expressed as: $$L_\mathrm{D} =
2\pi \, I_\mathrm{0,d} \, h^2\, , \label{form37}$$ while the total luminosity of a bulge is $$L_\mathrm{B} = \frac{2\pi n}{\nu_\mathrm{n}^{2n}} \,
\Gamma (2n) \, I_\mathrm{0,b} \, r_\mathrm{e,b}^2 \,
(1 - \epsilon_\mathrm{b})\,,
\label{form38}$$ where $\Gamma$ means the gamma function.
The code BUDDA allows the user to add some additional components in a galaxy model (bars, AGNs and double exponential discs), but we made decompositions of all our galaxies into a bulge and a disc only. There are some difficulties to determine the presence of a bar or any other substructures in edge-on perspective. The observed surface brightness distribution is a result of luminosity integration along the line of sight and such integration can smooth and hide such subcomponents. It is also hard to make the first approximation of the values of parameters which characterize these additional features.
Fig. \[example\] illustrates typical results of our decomposition for the Sb-type galaxy ESO 240-G011. The size of the image is 228$\times$40 pixels. The sky background, Galactic stars, bright HII regions and nearby companions were removed. The derived parameters are given in Table \[example2\]. As seen in Fig. \[example\] the fit is fairly good.
-------------------- ------------------ ------------------------- ----------------
$D$ (Mpc) 35.7 $\mu_\mathrm{e,b}$ $17.09\pm0.09$
$r_\mathrm{e,b}$ (kpc) $1.34\pm0.09$
$\mu_\mathrm{0,d}$ $16.22 \pm 0.16$ $n$ $2.4 \pm0.3$
$h$ (kpc) $5.1 \pm 0.9$ $q_\mathrm{b}$ $0.69\pm0.04$
$z_0$ (kpc) $0.80\pm0.13$ $B/D$ 0.38
$h/z_0$ $6.4$ $v_\mathrm{rot}$ (km/s) $267.5\pm3.3$
-------------------- ------------------ ------------------------- ----------------
\[Texample\]
{width="5.4cm"} {width="5.4cm"} {width="5.4cm"}
Decomposition and Results
=========================
To start the galaxy decomposition, we must determine the first-order approximations for the input parameters (see previous section). To obtain the first-order approximation for our 2D fits, we have used parameters derived from 1D fits of the major and minor axes profiles.
In almost all cases the ellipse index $\gamma$ of a bulge was taken to be equal $1.0 \pm 0.01$ (for simplicity we consider bulges with elliptical isophotes). A galaxy with complex morphology of a bulge requires more detailed investigation with higher resolution images than those that are given by 2MASS. The result of decompositions into a bulge and a disc is the list of all structural parameters of each object of the sample in 3 photometric bands. The final values of $\mu_\mathrm{e,b}$ and $\mu_\mathrm{0,d}$ were corrected for the Galactic extinction according to Schlegel, Finkbeiner & Davis (1998).
To check the reliability of our 2D decompositions, we subtracted the full apparent magnitudes of our models from the apparent magnitudes presented in the 2MASS catalogue. The mean values and standard deviations of these subtractions for the sample galaxies are: $\Delta J = 0.^m011\pm0.^m057\,,$ $\Delta H = 0.^m021\pm0.^m049\,,$ $\Delta K_s = 0.^m014\pm0.^m053\,.$
Our sample has 30 galaxies in common with the sample of BM02. The method used by these authors is based on several photometric cuts parallel to the minor axis of each galaxy and parallel to its major axis. Their sample includes flat galaxies mainly of late types in the $K_s$ filter. Fig. \[Mitronova\] shows a comparison of radial scalelengths (the upper frame), vertical scale heights (the middle frame) and deprojected central surface brightness of discs (the bottom frame). Open squares are related to the objects which have some peculiarities or features in structure (NGC5965, UGC6012, UGC12533 are with possible bars, IC4202, UGC4517, ESO121-G006 are with strong dust lanes, and UGC1817 is a bulgeless galaxy in a group). These objects do not lie on the diagonal. The mean values and standard deviations of difference between parameters obtained by us and by BM02 are: $$\langle h-h_\mathrm{BM} \rangle = 2.\arcsec46\pm2.\arcsec11\, ,$$ $$\langle z_0-z_\mathrm{0,BM} \rangle = 0.\arcsec41\pm0.\arcsec29\, ,$$ $$\langle S_0-S_\mathrm{0,BM} \rangle = 0.^m28\pm0.^m19\,.$$ As one can see (Fig. \[Mitronova\]), our results and those of BM02 are in good agreement.
De Grijs (1998) presented detailed surface photometry for a sample of edge-on galaxies in the $B$, $I$ and $K$ bands. Fig. \[de Grijs\] shows a comparison of radial scalelengths for 14 joint galaxies ($K$-band). Some systematics are evident which can be attributed to different decomposition procedures.
Table \[Bedregal\] compares decomposition results for 5 galaxies in the present work and in the paper by Bedregal, Aragon-Salamanca & Merrifield (2006). As one can see, the results are in agreement that is typical for such a comparison[^3]. It proves the reliability of the obtained results and makes them suitable in our further statistical analysis.
----------- ----------------- ------------------ ----------------- ------------------------ ------------------------ ------------------------
Name $h$ $r_\mathrm{e,b}$ $n$ $h$ $r_\mathrm{e,b}$ $n$
(kpc) (kpc) (kpc) (kpc)
NGC 5308 $2.80 \pm$ 0.89 $1.53 \pm 0.10$ $3.74 \pm 0.47$ $2.68^{+0.12}_{-0.05}$ $0.62^{+0.11}_{-2.51}$ $2.99^{+0.04}_{-0.06}$
NGC 1184 $4.29 \pm 0.64$ $1.84 \pm 0.06 $ $3.83 \pm 0.21$ $3.80^{+0.06}_{-0.05}$ $1.07^{+0.07}_{-2.43}$ $3.58^{+0.07}_{-0.05}$
NGC 4417 $1.57 \pm 0.35$ $0.86 \pm 0.03 $ $3.25 \pm 0.23$ $1.82^{+0.04}_{-0.05}$ $0.62^{+0.03}_{-0.02}$ $2.79^{+0.07}_{-0.06}$
NGC 1380A $1.70 \pm 0.23$ $0.53 \pm 0.02 $ $3.36 \pm 0.29$ $1.80^{+0.04}_{-0.06}$ $0.98^{+0.32}_{-0.25}$ $3.71^{+0.07}_{-0.10}$
NGC 1381 $2.36 \pm 0.56$ $0.88 \pm 0.03 $ $2.89 \pm 0.20$ $1.88^{+0.07}_{-0.05}$ $0.67^{+0.02}_{-0.01}$ $3.07^{+0.14}_{-0.10}$
----------- ----------------- ------------------ ----------------- ------------------------ ------------------------ ------------------------
Main results
============
Statistical correlations between structural parameters give us a clue for understanding the formation and evolution of spiral galaxies. Here we present only some of the correlations of the main parameters of the galaxies. The detailed analysis of the properties of bulges and discs will be done in a forthcoming article.
Distributions of structural parameters
--------------------------------------
Here we briefly describe the distributions of structural parameters of stellar components. In general, we do not take into account any corrections for the incompleteness and selection effects of the sample but in special cases we give the best fitting parameters obtained for the complete subsample.
### Discs
The distributions of the apparent central surface brightness of discs for edge-on view $\mu_\mathrm{0,d}$ corrected for Galactic extinction in the $JHK_s$ filters are shown in Fig. \[Distrib\]a. The values of this parameter span more than 3$^m$ arcsec$^{-2}$ in each filter. The distributions shift from the filter $J$ to the filter $K_s$ with increasing of mean central surface brightness. It is well seen that LSB galaxies were not included in the sample. Thus, one has to keep in mind that in fact the number of such galaxies could be high at the faint end of the distribution of $\mu_\mathrm{0,d}$ in all three passbands.
Fig. \[Distrib\]b shows a large range in scalelengths $h$. Most of the sample galaxies have $h = 1-6$ kpc but the number of extended galaxies is small. The distributions of the disc scaleheight $z_0$ are shown in Fig. \[Distrib\]c. There is a fairly narrow distribution, that lies in the range $z_0 = 0.4-1.2$ kpc, and a tail of “thick” galaxies with $z_0 \ga 1.2$ kpc. The shapes of distributions of $h$ and $z_0$ in each filter are quite similar and the median values of $h$ and $z_0$ have no systematic differences in the infrared bands under consideration (Table \[MedDisc\]).
The statistical study of the ratio of the radial to the vertical scalelengths for galactic discs, $h/z_0$, gives constraints on kinematical and dynamical models of galaxies. The distribution of this ratio $h/z_0$ is shown in Fig. \[Distrib\]d. It is rather flat. The mean ratio $\langle h/z_0 \rangle$ is about 3.9 for the $H$ and $K_s$-bands (Table \[MedGal\]). Kregel et al. (2002) found the mean ratio of $\langle h/z_0 \rangle=3.7\pm1.1$ (in the $I$-band) which is close to our result. For the sample of 153 galaxies composed by BM02 the mean ratio of $h/z_0$ is about 4.8. Although they studied galaxies in the $K_s$-band, their sample consists of late-type galaxies and the mean value of the ratio $h/z_0$ for these galaxies is larger than the mean ratio found in this paper. Apparently the mean $h/z_0$ ratio is higher in the blue filters than in the near-infrared bands (see for comparison e.g. van der Kruit & Searle 1982a and de Grijs & van der Kruit 1996).
### Bulges
The distributions of the apparent effective surface brightness of bulges are shown in Fig. \[Distrib\]e. The distributions in each filter have a narrow peak in contrast to the distributions of the central surface brightness of discs.
Fig. \[Distrib\]f demonstrates the fairly broad distributions of the effective radii of the bulges of galaxies. The distributions of the [Sérsic]{} indices of galaxies are shown in Fig. \[Distrib\]g. The distributions demonstrate a weak bimodality which may reflect the existence of two families of bulges: bulges with $n \ga 2$ and bulges with $n \la 2$. This bimodality is known for some galaxy samples (e.g. Fisher & Drory 2008) and probably is real for our sample. As was reported earlier (see e.g. Fisher & Drory 2007, 2008) such a bimodality correlates with morphological type of the bulge. Classical bulges have $n \ga 2$ and so-called pseudobulges have $n \la 2$, but this [Sérsic]{} index threshold can be considered only as an approximation to identify pseudobulges. Further we are showing the difference in the photometric planes for these two types of bulges.
In Fig. \[Distrib\]h we demonstrate the distributions of model bulge axis ratio $q_\mathrm{b} = b/a$ for edge-on view. In the case of edge-on galaxies this parameter describes directly the bulge flatness and the intrinsic 3D structure of bulges if they are assumed to be oblate spheroids. The Table \[MedGal\] gives the median value of this parameter $\sim 0.63$, independently of the band. This is in good agreement with Moriondo, Giovanardi & Hunt (1998) and Noordermeer & van der Hulst (2007). Moriondo et al. (1998) derived for their samples of 14 moderately inclined galaxies the median intrinsic axis ratio $q_\mathrm{b} = 0.64$ while Noordermeer & van der Hulst (2007) gave the average value of $\langle q_\mathrm{b} \rangle = 0.55 \pm 0.12$ for their sample of 21 early-type disc galaxies decomposed into contributions from a spheroidal bulge with a [Sérsic]{} profile and a flat disk. Thus, bulges are definitely nonspherical and flattened.
In recent years a great deal of new data has made researchers revise the assumption that bulges are oblate spheroids. However, conclusions about intrinsic 3D shapes of bulges and their possible triaxiality are still somewhat controversial, despite the fact that such conclusions are thought to be crucial for testing different scenarios of galaxy formation. For galaxies at intermediate inclination angles misalignment between the bulge and the disk major axes indicates that bulges are probably triaxial (Bertola, Vietri & Zeiliger 1991). From the observed distribution function of apparent ellipticities and misalignment angles Bertola et al. (1991) determined the probability distribution function (PDF) of intrinsic axial ratios for 32 bulges of their sample. The peak of the PDF falls on the intrinsic equatorial axis ratio[^4] $B/A = 0.95$ and bulge flattening $C/A = 0.65$. About one half of all bulges seems to be close to oblate, with the remainder being triaxial. Fathi and Peletier (2003) measured the bulge deprojected ellipticity in the equatorial plane that was obtained from the ellipse fitting the galaxy isophotes within a bulge region. One can estimate the median of $B/A$ from their distribution function as $0.7-0.8$ with smaller values attributed to more elongated bulges in late-type galaxies. Méndez-Abreu et al. (2008) also focused their attention on the intrinsic equatorial ellipticity of bulges. They reconstructed the PDF for the sample of 148 unbarred S0-Sb galaxies and obtained the value of $\langle B/A \rangle = 0.85$. This is consistent with findings by Bertola et al. (1991) and Fathi and Peletier (2003). However, in contrast to Fathi and Peletier (2003), Méndez-Abreu et al. (2008) did not find any significant differences in the shape of bulges between samples of early- and late-type galaxies.
Our sample of edge-on galaxies allows us to distinguish clearly the difference in the bulge equatorial ellipticities for early- and late-type galaxies. As a [Sérsic]{} index is correlated with the Hubble type (Andredakis, Peletier & Balcells 1995; Graham 2001; Möllenhof 2004) (as well as with the bulge morphology, dividing bulges into classical ones and pseudobulges; Fisher & Drory 2008) we divided our sample bulges between those with [Sérsic]{} index $n \ga 2$ (early-type galaxies or classical bulges) and those with $n \la 2$ (late-type galaxies or pseudobulges) using $K_s$ filter. As one can see in Fig. \[Distrib\_qb\] for bulges with $n \ga 2$ the distribution of $q_\mathrm{b}$ has a rather narrow peak at $q_\mathrm{b} \approx 0.65$. This may reflect the fact that bulges in early-type spirals are nearly oblate spheroids with moderate flattening. The distribution of $q_\mathrm{b}$ for bulges with $n \la 2$ is very wide, spreading from flat bulges up to nearly spherical ones. Such a distribution may be attributed to definitely triaxial, near prolate bulges that are seen from different projections — along the major axis and perpendicular to it. But we can not exclude that triaxial shape of bulges in late-type galaxies may hide the presence of bars that thickened in the vertical direction during secular evolution. We are planning to investigate this question in more detail in a forthcoming paper.
Fig. \[BD\] demonstrates the distributions of the ratio of bulge and disc luminosities $B/D \equiv L_\mathrm{b}/L_\mathrm{d}$. Our sample contains many early-type galaxies with $B/D \ga 0.5$ and late-type galaxies with values of $B/D \la 0.2$. For edge-on galaxies the ratio of $B/D$ is a key parameter in addition to $n$ for classifying spiral galaxies on the Hubble sequence because the spiral arms are not seen in this case (de Jong 1996; Graham 2001). Our sample also demonstrates a correlation between the ratios of $B/D$ and the Hubble morphological types $T$ taken from LEDA (Fig. \[HT\]a) that is in agreement with previous work (e.g., de Jong 1996; Graham 2001; Möllenhoff & Heid 2001; Hunt, Pierini & Giovanardi 2004; Möllenhoff 2004). However for edge-on galaxies morphological types are very subjective. Comparisons of the Hubble types given by different classifiers show an rms uncertainty in the type index of order 2 $T$-units (Lahav et al. 1995). Unfortunately there are errors in the bulge/disc decomposition. Therefore we use both approaches (type $T$ from LEDA and obtained $B/D$) for describing morphological types of edge-on galaxies. In some cases we will distinguish early- and late-type galaxies using the [Sérsic]{} index $n$ because there is a good correlation between $B/D$ and $n$ (Fig. \[HT\]b). This correlation is not new and was presented by many authors (e.g., Andredakis et al. 1995; Graham 2001; Möllenhoff & Heid 2001; Hunt et al. 2004).
Details of the statistical properties of the sample discs and bulges are presented in the Table \[MedDisc\], the Table \[MedBulge\] and the Table \[MedGal\]. The values of $n$ in Table \[MedBulge\] systematically increase going from $J$ to $K_s$ – this may imply that $n$ is systematically underestimated due to extinction affecting the central peak of emission at the shorter near-IR wavelengths. Another possible explanation is a real change of the surface brightness shape with wavelength. But, as one can see in Table \[MedBulge\], the quoted changes are within statistical errors.
---------------- -------------------------- ------------------------ ------------------------ --------------------------
Sp $\mu_{\mathrm{0,d}}$ $h$ $z_0$ $M_\mathrm{d}$
($^m/\mathrm{arcsec}^2$) (kpc) (kpc) $(^m)$
J $17.57^{-0.55}_{+0.43}$ $3.68^{-1.30}_{+1.44}$ $1.04^{-0.34}_{+0.48}$ $-22.53^{-0.77}_{+0.83}$
H $16.87^{-0.43}_{+0.35}$ $3.87^{-1.36}_{+1.70}$ $0.99^{-0.29}_{+0.45}$ $-23.23^{-0.78}_{+0.82}$
K$_\mathrm{s}$ $16.44^{-0.41}_{+0.43}$ $3.66^{-1.27}_{+1.49}$ $0.97^{-0.30}_{+0.45}$ $-23.52^{-0.79}_{+0.75}$
---------------- -------------------------- ------------------------ ------------------------ --------------------------
---------------- -------------------------- ------------------------ ------------------------ --------------------------
Sp $\mu_{\mathrm{e,b}}$ $r_{\mathrm{e,b}}$ $n$ $M_\mathrm{b}$
($^m/\mathrm{arcsec}^2$) (kpc) $(^m)$
J $17.68^{-0.13}_{+0.31}$ $1.23^{-0.54}_{+0.55}$ $2.35^{-1.36}_{+0.70}$ $-21.97^{-0.97}_{+1.85}$
H $17.13^{-0.17}_{+0.36}$ $1.48^{-0.63}_{+0.68}$ $2.65^{-1.06}_{+0.68}$ $-22.96^{-0.95}_{+1.93}$
K$_\mathrm{s}$ $16.69^{-0.17}_{+0.34}$ $1.35^{-0.59}_{+0.61}$ $2.74^{-1.44}_{+0.53}$ $-23.19^{-0.94}_{+2.16}$
---------------- -------------------------- ------------------------ ------------------------ --------------------------
Sp $h/z_0$ $q_\mathrm{b}$ $B/D$
---------------- ------------------------ ------------------------ ------------------------
J $3.54^{-0.93}_{+0.86}$ $0.63^{-0.07}_{+0.06}$ $0.50^{-0.38}_{+0.62}$
H $3.94^{-0.85}_{+0.92}$ $0.63^{-0.08}_{+0.05}$ $0.69^{-0.49}_{+0.79}$
K$_\mathrm{s}$ $3.92^{-0.92}_{+0.95}$ $0.62^{-0.11}_{+0.04}$ $0.64^{-0.44}_{+0.70}$
Scaling relations
-----------------
The relationships between bulge and disc parameters might give an insight in the ages of the bulge and disc formation and help to reconstruct the chronology of these events. Andredakis & Sanders (1994) were the first to report the existence of a correlation between disc and bulge scalelengths for their sample of 34 late-type spirals. Just after that, such a correlation was confirmed and used for speculating about the scale-free Hubble sequence since the relative size of bulge and disc seemed not to depend on the morphological type (de Jong 1996; Courteau et al. 1996). De Jong (1996) found the equation for the least squares fitted line in the $K$ passband: $\lg(r_\mathrm{e,b}) = 0.95\lg(h) - 0.86$ with a standard deviation of 0.17 and correlation coefficient $r \sim 0.8$. This suggests a linear relation between $h$ and $r_\mathrm{e,b}$, i.e. the disc scalelength increases when the bulge effective radius increases.
Furthermore it became clear that the ratio $r_\mathrm{e,b} / h$ appeared to be strongly influenced by the bulge parametrization (Graham & Prieto 1999; Graham 2001). The independence of morphological type and a strong correlation were seen only when galaxies were best modeled with a fixed $n = 1$ bulges for all type galaxies (de Jong 1996; Courteau et al. 1996; Moriondo et. al. 1998; Hunt et al. 2004). According to whether bulges were fitted with the $n = 1$ [Sérsic]{} model or with the de Vaucoulers law, the correlation between scalelengths was clear or disappeared completely (Andredakis & Sanders 1994; de Jong 1996; Hunt et al. 2004).
Using [Sérsic]{} bulges with a free shape parameter $n$ as the best fitting models (Möllenhoff & Heidt 2001; Khosroshahi et al. 2000b; Graham 2001; MacArthur, Courteau & Holtzman 2003) or different but fixed $n$ for early- and late-type galaxies (Graham & Prieto 1999) confirms this correlation with a fairly large scatter and opens the discussion about the variation of the ratio $r_\mathrm{e,b} / h$ with Hubble types (Moriondo et al. 1998; Graham & Prieto 1999; Khosroshahi et al. 2000b; Graham 2001; Scodeggio et al. 2002; MacArthur et al. 2003; Möllenhoff 2004) and about the prematurity of the claims made by de Jong (1996) and Courteau et al. (1996) for a scale-free Hubble sequence.
We also found that there exist fairly strong correlations between structural parameters of discs and bulges. The correlation between $h$ and $r_\mathrm{e,b}$ is shown in Fig. \[Corr1\]a. In this figure and hereafter we mark the complete subsample galaxies with filled squares. The regression line is plotted only for points of the complete subsample. The equations of the linear regression and the correlation coefficients are presented in the Table \[Correlations1\]. These equations are consistent with a linear relation between $h$ and $r_\mathrm{e,b}$.
We found a mean $r_\mathrm{e,b}/ h$ in the $J$-band of $0.39 \pm 0.18$, in the $H$-band of $0.42 \pm 0.20$ and in the $K_s$-band of $0.40 \pm 0.20$, which are in agreement within errors with the values obtained by Méndez-Abreu et al. (2008) for their sample of 148 unbarred S0–Sb galaxies $\langle r_\mathrm{e,b} / h \rangle = 0.36 \pm 0.17$ (in the $J$-band).
There is some controversial evidence for systematic changes with morphological type in the ratio $r_\mathrm{e,b} / h$ (Scodeggio et al. 2002; MacArthur et al. 2003; Hunt et al. 2004; Möllenhoff 2004). Despite being difficult to detect, the dependence with Hubble types might be real in the sense that the sample of predominantly late-type disk galaxies gives $\langle r_\mathrm{e,b} / h \rangle = 0.22 \pm 0.09$ (MacArthur et al. 2003; the $H$-band) and the sample of predominantly early-type galaxies leads to $\langle r_\mathrm{e,b}/ h \rangle = 0.33 \pm 0.17$ (Khosroshahi et al. 2000b; the $K$-band).
For edge-on galaxies the reliable indicator of morphological type is $B/D$ or $n$. Fig. \[Corr1\]b demonstrates the dependence of the ratio $r_\mathrm{e,b}/ h$ on the [Sérsic]{} shape parameter $n$. There is a clear trend for $r_\mathrm{e,b}/ h$ to increase with $n$. The size ratio increases from $n \sim 0.5$ to $n \sim 5$ by a factor of $\sim 3-4$. Almost all bulges of the sample of MacArthur et al. (2003) have $n < 1.2$ and most of the bulges of the sample of Khosroshahi et al. (2000b) have $n > 2.2$. Thus, the difference in the mean value of $r_\mathrm{e,b}/ h$ for these two samples reflects the same tendency that one can see in Fig. \[Corr1\]b. The same increase was declared by Hunt et al. (2004). They found for $n=1$ bulges $\langle r_\mathrm{e,b}/ h \rangle = 0.14 \pm 0.06$, for $n=2$ bulges $\langle r_\mathrm{e,b}/ h \rangle = 0.25 \pm 0.17$, for $n=3$ bulges $\langle r_\mathrm{e,b}/ h \rangle = 0.35 \pm 0.27$, and for $n=4$ bulges $\langle r_\mathrm{e,b}/ h \rangle = 0.54 \pm 0.54$.
Since $n$ is an indicator (in statistical sense) of morphological type, the changes of $r_\mathrm{e,b}/ h$ with the shape of the bulge implies that Hubble sequence is not scale-free. The same conclusion has been recently made by Gadotti (2009) for his sample of nearly 1000 galaxies. Moreover, he was able to distinguish in his sample two subsamples of classical bulges and pseudobulges. Gadotti (2009) have used the Kormendy relation (Kormendy 1977) for identifying pseudobulges and bulges by their loci in the $\langle \mu_\mathrm{e,b} \rangle$[^5] – $r_\mathrm{e,b}$ plane. The corresponding loci are quite different for these two types of bulges. Gadotti (2009) found that $r_\mathrm{e,b}/ h$ for galaxies of both subsamples was clearly correlated with the bulge-to-total luminosity ratio, although the corresponding relations for pseudobulges and classical bulges were offset. As the bulge-to-total luminosity ratio is related to the morphological (Hubble) type of a galaxy, the existence of such relations rule out the hypothesis of a scale-free Hubble sequence. It means that the widely discussed secular evolution scenario of the bulge formation should be improved to include the variation of $r_\mathrm{e,b}/ h$ with $n$ (or other indicators of the type) even in the case of pseudobulges.
The disc flattening can be measured directly only in edge-on galaxies. It is commonly expressed as the ratio of disc scaleheight $z_0$ to disc scalelength $h$. This ratio shows a weak trend with Hubble type (de Grijs 1998), but there is a fairly clear correlation between both scale parameters (e.g. Kregel et al. 2002). The result of Kregel et al. (2002) can be transformed into the mean value of $\langle h/ z_0 \rangle = 3.7 \pm 1.1$ for the volume corrected distribution. Our data also demonstrates such a correlation for galaxies of all types (Fig. \[Corr2\]a). It is remarkable that the line of the linear regression is drawn through the point close to $(0,0)$.
We used the data of photometric parameters $h$ and $z_0$ in BM02 ($K_s$ filter) of their reliable subsample and found the following equation of linear regression: $z_0 = (0.210\pm0.026) \, h - (0.017\pm0.258)$ with correlation coefficient $r = 0.727$, which is very close to our result (Table \[Correlations1\]).
Thus, the galaxies with large disc scalelengths have on average larger disc scaleheights and larger effective radii of their bulges. However, one should keep in mind the fact that the spread in the $h - z_0$ plane is large. This scatter is reflected in a wide and remarkably flat distribution of the ratio $h/z_0$ presented in Fig. \[Distrib\]d. We return to this question in Section \[DHalo\].
As is seen in Fig. \[Corr2\]b the disc scaleheight tends to increase with the bulge effective radius. The correlation between $z_0$ and $r_\mathrm{e,b}$ is new and was not described earlier. We found the following ratios of mean $r_\mathrm{e,b}/z_0$: $\langle r_\mathrm{e,b}/z_0 \rangle = 1.23 \pm 0.48$ for the $J$-band, $\langle r_\mathrm{e,b}/z_0 \rangle = 1.57 \pm 0.62$ for the $H$-band, $\langle r_\mathrm{e,b}/z_0 \rangle = 1.49 \pm 0.60$ for the $K_s$-band.
-------------------------------------------------------------------------------------
$J$:$\lg h = (1.011 \pm 0.143) \lg r_\mathrm{e,b} + (0.596 \pm 0.010)$, $r=0.59$
$H$: $\lg h = (0.936 \pm 0.113) \lg r_\mathrm{e,b} + (0.533 \pm 0.030)$, $r=0.66$
$K_s$: $\lg h = (0.933 \pm 0.104) \lg r_\mathrm{e,b} + (0.555 \pm 0.021)$, $r=0.68$
$J$:$z_0 = (0.237 \pm 0.015) h + (0.010 \pm 0.140)$, $r=0.87$
$H$: $z_0 = (0.208 \pm 0.014) h + (0.030 \pm 0.144)$, $r=0.84$
$K_s$: $z_0 = (0.207 \pm 0.013) h + (0.027 \pm 0.129)$, $r=0.85$
$J$:$z_0 = (0.891 \pm 0.120) r_\mathrm{e,b} + (0.058 \pm 0.356)$, $r=0.62$
$H$: $z_0 = (0.680 \pm 0.100) r_\mathrm{e,b} + (0.045 \pm 0.366)$, $r=0.59$
$K_s$: $z_0 = (0.715 \pm 0.094) r_\mathrm{e,b} + (0.041 \pm 0.325)$, $r=0.61$
-------------------------------------------------------------------------------------
Well-known relationships
------------------------
Here we investigate the general scaling relations for spiral galaxies such as the Tully-Fisher relation, the Fundamental Plane (FP) of discs and the Photometric Plane of bulges (PhP). In this subsection we study the Tully-Fisher relation and the Fundamental Plane of discs and in the next subsection we specially investigate the Photometric Plane of bulges.
### The Tully-Fisher relation
The Tully-Fisher (TF) relation links the luminosity and the maximum rotation velocity $v_\mathrm{rot}$ of galaxies. The small scatter in the Tully-Fisher relation allows us to use it as a method for measurements of distances to spiral galaxies. To determine the slope and the zeropoint of the TF relation we used the total luminosities of the sample galaxies taken from 2MASS ($J_\mathrm{ext}$, $H_\mathrm{ext}$ or $K_\mathrm{ext}$). They were derived from the isophotal magnitudes by using the $K_s$-band 20 mag/arcsec$^2$ elliptical radius ($J_{20}$, $H_{20}$, $K_{20}$) and then were extrapolated to roughly four times the disc scalelength $h$ (for the detailed description see Jarrett et al. 2000 and Cutri et al. 2006).
Unfortunately, measured rotation velocities are not known for all the sample galaxies. In the $J$-band the number of galaxies with known rotational velocities is 107, in the $H$-band there are 110 galaxies and in the $K_s$-band there are 116 objects with known velocities. The maximum rotational velocity corrected for inclination $v_\mathrm{rot}$ was taken from LEDA.
The absolute magnitude in different bands was calculated as $$M_{\lambda} = m_{\lambda} - A_{\lambda}-25-5\lg(D_\mathrm{L}) \,,$$ where $m_{\lambda}$ is the total apparent magnitude in each passband ($J_\mathrm{ext}$, $H_\mathrm{ext}$ or $K_\mathrm{ext}$), $D_\mathrm{L}$ is the luminosity distance and $A_{\lambda}$ is the Galactic extinction taken from Schlegel et al. (1998). The luminosity is expressed as $$\lg L_{\lambda} = 0.4(M^{\lambda}_{\sun}-M_{\lambda}) \,,$$ where $M^J_{\sun}=3.64$ mag, $M^H_{\sun}=3.32$ mag and $M^{K}_{\sun}=3.28$ mag (Binney & Merrifield 1998). We fitted the TF relation coefficients by taking the bisector between the lines of linear regression obtained through the standard LSQ and inverse LSQ methods. The best fitting results are presented in the Table \[Correltions2\]. The $K_s$-band TF relation is plotted in Fig. \[discCorr\]a.
There are many papers with the TF relation in the NIR bands for early- and late-type galaxies. Here we compare our results with those of some works based on 2MASS observations. The results by Karachentsev et al. (2002), Courteau et al. (2007), de Rijcke et al. (2007) and Masters, Springob & Huchra (2008) are presented in Table \[TF\].
As was shown by Masters et al. (2008) the TF relation coefficients depend on galaxy morphology. The relationship is steeper for later type spirals than earlier type spirals. The TF zeropoint is dimmer for late-type galaxies. We did not divide our sample on morphological types to get the TF relation for various galactic types. Our sample contains an almost equal number of early- and late-type galaxies. That is why the coefficients found are somewhat different from those given in other papers.
---------------------------- ------ ------ ------ ---------- ------ ------ -------------------------------------------------------------------------------------
Reference Comments
a b a b a b
This work 3.43 2.67 3.44 2.83 3.55 2.66 all types
Karachentsev et al. (2002) 3.28 3.48 3.61 Based on $J_\mathrm{ext}$, $H_\mathrm{ext}$ and $K_{s,\mathrm{ext}}$, RFGC galaxies
Courteau et al. (2007) 2.5 4.0 2.6 4.0 2.6 3.9 64% Sa galaxies
de Rijcke et al. (2007) 3.46 2.44 early-type galaxies
Masters et al. (2008) 3.63 1.05 3.61 1.30 4.01 0.36 Sc galaxies
Masters et al. (2008) 3.12 2.32 3.12 2.53 3.45 1.76 Sb galaxies
Masters et al. (2008) 2.44 4.11 2.43 4.31 2.78 3.48 Sa galaxies
---------------------------- ------ ------ ------ ---------- ------ ------ -------------------------------------------------------------------------------------
### The Fundamental Plane of discs
Following Moriondo et al. (1999) we fit a relation for disc parameters as: $$\lg h = a \lg v_\mathrm{rot} + b \, S_\mathrm{0,d} + c \, ,
\label{ScaleRel}$$ involving the disc scalelength, the maximum rotational velocity, and the deprojected central surface brightness $S_0$, analogous to the FP of elliptical galaxies. To define coefficients of the linear approximation we used the standard least squares method. From the complete subsample we excluded 7 galaxies: NGC 7232 and NGC 5775 are interacting, NGC 7814 is a pure S0 galaxy with very faint disc, NGC 4222 and NGC 100 are low surface brightness galaxies, NGC 5023 and NGC 4183 have very large errors of disc parameters. The results of the fitting are given in Table \[Correltions2\]. The scatter of points is rather large which may be caused by observational uncertainties and decomposition errors. The coefficients derived in this paper are comparable with those reported by Karachentsev (1989) for the $I$-band: $a=1.4$, $b=0.28$; and by Moriondo et al. (1999) for the $H$-band: $a=1.31 \pm 0.19$, $b=0.25 \pm 0.04$. Our coefficients as well as the coefficients derived by Karachentsev (1989) and Moriondo et al. (1999) are not consistent with what would be expected on the basis of the virial theorem and a universal mass-to-light ratio ($a=2$ and $b=0.4$).
We divided galaxies into galaxies with $n\ga2.2$ and galaxies with $n\la2.2$. In spite of the fact that the shape parameter $n$ is related to the bulge model, the fundamental planes of discs, which tie only disc parameters and the maximum rotational velocity of gas, are different for galaxies with different bulges. Perhaps it is due to observational errors and errors of decomposition for galaxies of different morphological types (for early-type galaxies rotational velocities and parameters of faint discs are less well defined than for late-type galaxies). Another reason may be the real difference of discs in galaxies with low and high density bulges. This effect requires a more careful examination.
---------------------------------------------------------------------------------------------------------
$J$:$\lg L = (3.43 \pm 0.29) \lg v_\mathrm{rot} + (2.67 \pm 1.39)$, $r=0.82$
$H$: $\lg L = (3.44 \pm 0.28) \lg v_\mathrm{rot} + (2.83 \pm 1.32)$, $r=0.82$
$K_s$: $\lg L = (3.55 \pm 0.30) \lg v_\mathrm{rot} + (2.66 \pm 1.41)$, $r=0.81$
For all galaxies of the complete sample with known velocities:
$J$:$\lg h = (1.017 \pm 0.101) \lg v_\mathrm{rot} + (0.173 \pm 0.021)S_\mathrm{0,d}-(5.04 \pm 0.11)$
$H$: $\lg h = (1.044 \pm 0.092) \lg v_\mathrm{rot} + (0.190 \pm 0.020)S_\mathrm{0,d}-(5.28 \pm 0.11)$
$K_s$: $\lg h = (1.077 \pm 0.104) \lg v_\mathrm{rot} + (0.202 \pm 0.023)S_\mathrm{0,d}-(5.53 \pm 0.12)$
For galaxies of the complete sample with $n\ga2.2$:
$J$:$\lg h = (0.729 \pm 0.265) \lg v_\mathrm{rot} + (0.204 \pm 0.054)S_\mathrm{0,d}-(4.92 \pm 0.14)$
$H$: $\lg h = (0.790 \pm 0.149) \lg v_\mathrm{rot} + (0.193 \pm 0.029)S_\mathrm{0,d}-(4.70 \pm 0.11)$
$K_s$: $\lg h = (0.750 \pm 0.149) \lg v_\mathrm{rot} + (0.209 \pm 0.031)S_\mathrm{0,d}-(4.83 \pm 0.12)$
For galaxies of the complete sample with $n\la2.2$:
$J$:$\lg h = (1.218 \pm 0.122) \lg v_\mathrm{rot} + (0.213 \pm 0.027)S_\mathrm{0,d}-(6.29 \pm 0.09)$
$H$: $\lg h = (1.224 \pm 0.149) \lg v_\mathrm{rot} + (0.230 \pm 0.025)S_\mathrm{0,d}-(6.45 \pm 0.08)$
$K_s$: $\lg h = (1.281 \pm 0.146) \lg v_\mathrm{rot} + (0.231 \pm 0.026)S_\mathrm{0,d}-(6.53 \pm 0.08)$
---------------------------------------------------------------------------------------------------------
Photometric Plane and [Sérsic]{} index
--------------------------------------
The [Sérsic]{} index $n$ represents the parameter quantifying the radial concentration of the stellar distribution of bulges. The detailed structure of bulges, their formation and evolutionary status is now widely discussed. Here we present a correlation between bulge absolute magnitude $M_\mathrm{b}$ and $n$. The plot is shown in Fig. \[BulgeCorr\]a. The presented correlation is well-known (e.g. Andredakis et al. 1995; Graham 2001; Hunt et al. 2004; Möllenhoff & Heidt 2001; Möllenhoff 2004) and seen in all bands — from optical to near-infrared. It shows that exponential bulges are fainter than the de Vaucouleurs’ bulges, although the scatter in the plot is large.
There are a lot of trends of galactic structural parameters with the [Sérsic]{} index of a bulge $n$, but the most prominent difference in parameters arises when one divides the sample under investigation into the subsample of classical bulges and pseudobulges. As was shown above, perhaps, the difference between classical bulges and pseudobulges is reflected in the intrinsic flattening and the 3D shape of bulges, in the ratio $r_\mathrm{e,b}/ h$ and in building fundamental planes for discs.
The division of bulges into classical bulges and pseudobulges was described in many works (see Kormendy & Kennicutt 2004 for review). Classical bulges appear to be similar to elliptical galaxies and have similar properties. It is assumed that these systems were built via minor and major merging. The galaxies with pseudobulges are classified by the presence of some features within a bulge, e.g. nuclear bars, nuclear spirals and nuclear rings (Erwin & Spark 2002; Carollo et al. 1997; Fisher & Drory 2008). Pseudobulges are thought to be formed via disc instabilities and secular evolution and have disc-like apparent flattening. One of the main differences between classical bulges and pseudobulges is in the shape parameter $n$. Pseudobulges as usual have $n\la2$ and classical bulges have $n\ge2$ with little-to-no overlap.
The images of galaxies under investigation have low resolution and inner features inside a bulge are smoothed. That is why we were searching for correlations between obtained structural parameters of the bulges with different [Sérsic]{} index $n$, not with morphological features. Here we describe the difference between bulges and pseudobulges while building the photometric plane for them.
Khosroshahi et al. (2000a,b) found a tight correlation of the [Sérsic]{} indices $n$ with the central surface brightness $\mu_\mathrm{0,b}$ and the effective radius of a bulge $r_\mathrm{e,b}$. They called this relationship the Photometric Plane.
Following Khosroshahi et al. (2000a,b) we performed the least-squared fit which expresses $\lg n$ as a linear combination of two other parameters: $\lg n = a \lg r_\mathrm{e} + b \, \mu_\mathrm{0,b} + c$. From the complete subsample we exclude galaxies with very small bulges and 2 bright flat galaxies of late types: NGC 4244 and NGC 4517.
For the complete sample the best fitting results are presented in Table \[FP\]. Khosroshahi et al. (2000a,b) found the Photometric Plane for 26 bulges of predominantly early-type galaxies in the $K$-band: $\lg n = (0.130 \pm 0.040)\lg r_\mathrm{e,b} -
(0.073 \pm 0.011)\mu_\mathrm{0,b} + (1.21 \pm 0.11)$. Möllenhoff & Heidt (2001) obtained for their sample of 41 predominantly late-type spirals in the $JHK$ bands: $\lg n = 0.187 \lg r_\mathrm{e,b} - 0.081\mu_\mathrm{0,b} + 1.34 \pm 0.10$ with the correlation coefficient $r = 0.91$. Finally, Méndez-Abreu et al. (2008) gave $\lg n = 0.17(\pm 0.02)\lg r_\mathrm{e,b} -
0.088(\pm 0.004)\mu_\mathrm{0,b} + 1.48(\pm 0.05)$ for their sample of 148 unbarred S0–Sb galaxies in the $J$-band. Within error bars, the differences between our coefficients for all our complete sample and those of Khosroshahi et al. (2000a,b), Möllenhoff & Heidt (2001) and Méndez-Abreu et al. (2008) are not significant, at least in the $K$-band.
The surprising result was the non-linear correlation of the PhP for galaxies with various values of $n$ (Fig. \[BulgeCorr\]b). From our main sample we extracted the subsample of galaxies with $\lg n \ga 0.2$ ($n \ga 1.58$) and obtained the best-fit Photometric Plane (see Table \[FP\]). It occurred that the galaxies with $\lg n \ga 0.2$ lying on the PhP, that is built for these galaxies, has a small scatter, but the galaxies with $\lg n \la 0.1$ ($n \la 1.26$) do not correspond to this plane and form their own plane (see Table \[FP\]). Inside the window $0.1 <\lg n\la 0.2$ there are points which do not correspond to these two planes. The regression of the PhP is non-linear, but we approximate this dependence as a superposition of two linear regressions. The definite difference between the photometric planes of the bulges with different shape parameters $n$ is shown in Fig. \[BulgeCorr\]b. The error bars are plotted as an indicator of the really different location of bulges and pseudobulges in the photometric planes. At a first glance the errors in the leftmost region of the plot are large, but bulges with very small $n$ are clearly recognized in surface brightness profiles of galaxies. What more, the relationship for bulges in the rightmost region of the plot begins to deviate from a linear trend for values of $n$ significantly above 1. The main difference between our sample and the samples of the above-mentioned authors is the range of the shape parameters $n$. Our sample contains a substantial amount of bulges with $\lg n \la 0.1$. For other samples the low boundary for $n$ lies at $\lg n \ga 0.2$, that is why the curvature of the Photometric Plane towards small values of $n$ for bulges was not noticed earlier.
This result is quite new and surprising. We realise that the points on the lower-left of the figure have bulges with $n < 1$, which are somewhat controversial and need further investigation, which will be done in a future paper. But a similar planar relation associated with the Photometric Plane was found for bright ellipticals, dwarf ellipticals and lenticular galaxies (Khosroshahi et al. 2000a; Khosroshahi et al. 2004; Ravikumar et al. 2006; Barway et al. 2009). If the range of the [Sérsic]{} parameter $n$ for ellipticals stretches to very small values of $n$, the curvature is clearly observed (Khosroshahi et al. 2004; Ravikumar et al. 2006).
The origin of the tight relation betveen $n$ and the linear combination of $\lg r_\mathrm{e}$ and $\mu_\mathrm{0,b}$ as well as the curvature of this relation are not completely clear. Aceves, Velásquez & Cruz (2006) argued that the curvature was related to an intrinsic property of a [Sérsic]{} profile. They wrote an expression for the total luminous matter associated with a [Sérsic]{} profile in log-space, involving $n$, $\lg r_\mathrm{e}$ and $\mu_\mathrm{0,b}$, and showed that for a set of galaxies with equal luminosities there were non-constant terms in this expression. These terms introduce a systematic change in a PhP-like expression.
It is not clear whether two distinct parts of a curve in Fig. \[BulgeCorr\]b reflect the different origin of bulges lying in the corresponding regions. However, the collisionless merger remnants of disc galaxies obtained in $N$-body simulations and fitted by a [Sérsic]{} profile fairly well reproduce the slope of the PhP in the region of $\lg n \ga 0.2$ (Aceves et al. 2006). We put on the data of Aceves et al. (2006) in our Fig. \[BulgeCorr\]b and found the numerical data are consistent with our observational ones.
The curvature of the PhP towards small values of $n$ may reflect the quite different nature of such bulges, formed, for example, via secular evolution of discs. One of the signs of secular evolution is the presence of a bar or other structures (such as rings, double exponential discs). We have made the assumptions about the existence of a bar or other structures in our sample galaxies on the basis of the 1D surface brightness profiles along the major axis and on the residual images where additional components, that were not included in the decomposition, become apparent. The galaxies with or without bars are represented among the galaxies with $\lg n \ga 0.2$ (classical bulges) as well as among the galaxies with $\lg n \la 0.2$ (pseudobulges). So the shape of the PhP may describe the deeper nature of these objects. Márquez et al. (2001) have suggested a theoretical explanation of the PhP of ellipticals. They calculated the specific entropy of elliptical galaxies (Entropic Surface), took into account a scaling relation between the potential energy and mass (Energy-Mass Surface) and showed that the PhP arose as an intersection line of these two planes demonstrating a curvature consistent with the observational data in the limited range of values that has been considered for $n$. All these arguments may be valid for bulges of all types too.
[l]{} For all galaxies of complete sample\
$J$:$\lg n = (0.053 \pm 0.059) \lg r_\mathrm{e,b} - (0.101 \pm 0.004)\mu_\mathrm{0,b} + (1.63 \pm 0.06)$\
$H$: $\lg n = (0.106 \pm 0.047) \lg r_\mathrm{e,b} - (0.084 \pm 0.004)\mu_\mathrm{0,b} + (1.37 \pm 0.05)$\
$K_s$: $\lg n = (0.160 \pm 0.058) \lg r_\mathrm{e,b} - (0.084 \pm 0.005)\mu_\mathrm{0,b} + (1.34 \pm 0.07)$\
\
$\lg n\geq0.2$\
$J$:$\lg n = (0.030 \pm 0.018) \lg r_\mathrm{e,b} - (0.067 \pm 0.002)\mu_\mathrm{0,b} + (1.23 \pm 0.03)$\
$H$: $\lg n = (0.018 \pm 0.014) \lg r_\mathrm{e,b} - (0.062 \pm 0.002)\mu_\mathrm{0,b} + (1.14 \pm 0.02)$\
$K_s$: $\lg n = (0.022 \pm 0.013) \lg r_\mathrm{e,b} - (0.061 \pm 0.002)\mu_\mathrm{0,b} + (1.11 \pm 0.02)$\
\
$\lg n\la0.1$\
$J$:$\lg n = (0.137 \pm 0.134) \lg r_\mathrm{e,b} - (0.142 \pm 0.039)\mu_\mathrm{0,b} + (2.31 \pm 0.67)$\
$H$: $\lg n = (0.201 \pm 0.162) \lg r_\mathrm{e,b} - (0.172 \pm 0.051)\mu_\mathrm{0,b} + (2.76 \pm 0.83)$\
$K_s$: $\lg n = (0.467 \pm 0.214) \lg r_\mathrm{e,b} - (0.204 \pm 0.075)\mu_\mathrm{0,b} + (3.21 \pm 1.19)$\
\
The dark halo and disc flattening {#DHalo}
---------------------------------
In spite of a good correlation between the disc scalelength $h$ and the disc scaleheight $z_0$, the scatter of the ratio $h/z_0$ is fairly large. The ratio $h/z_0$ characterizes a relative thickness of a galaxy, or its flattening. Fig. \[Distrib\]d shows that $h/z_0$ varies from $\sim 8$ for thin galaxies to $1-2$ for plump discs. Disc flattening appears to correlate with the morphological type (e.g. de Grijs 1998) and the global HI content (Reshetnikov & Combes 1997). Late-type and gas-rich galaxies are, on average, thinner than early-type systems, but this correlation is rather weak. Although the discs seem to be larger and thicker with a higher maximum rotation (van der Kruit & de Grijs 1999; Kregel et al. 2002, 2005), the ratio $h/z_0$ is not correlated with the rotation velocity (Zasov et al. 2002; Kregel et al. 2002, 2005).
Zasov et al. (2002) and Kregel et al. (2005) found that the disc flattening shows a definite trend with the ratio of dynamical mass to disc luminosity provided the dynamical mass is defined as the total mass enclosed within the sphere of radius that is equal to four disc scalelengths: $\mathrm{M}_\mathrm{tot} = 4 h v_\mathrm{rot}^2 / G$. The total mass includes the mass of the disc $\mathrm{M}_\mathrm{d}$, the dark halo $\mathrm{M}_\mathrm{h}$ and the bulge $\mathrm{M}_\mathrm{b}$. So, the ratio of dynamical mass to disc luminosity determines the relative mass of a spherical component or the relative mass of the dark halo for bulgeless galaxies. The found correlation implies that relatively thinner discs in bulgeless galaxies tend to be embedded in more massive dark haloes.
Kregel et al. (2005) argued that the analytical collapse model of disc galaxy formation (White & Rees 1978; Fall & Efstathiou 1980; Blumenthal 1986; Mo, Mao & White 1998), the local Toomre’s (1964) stability criterion for a disc and the suggestion about vertical dynamical equilibrium, provided the constant value of the ratio of the vertical to the radial velocity dispersion $\sigma_\mathrm{z} / \sigma_\mathrm{R}$, lead to the correlation between the disc flattening and the ratio of dynamical mass to disc luminosity.
Zasov et al. (2002) suggested that discs are marginally stable against the growth of perturbations in their planes and bending perturbations. This stability against bending perturbations means that the ratio $\sigma_\mathrm{z} / \sigma_\mathrm{R}$ must be larger than some analytical threshold: $\sigma_\mathrm{z} / \sigma_\mathrm{R} \ga 0.3-0.4$ (Polyachenko & Shukhman 1977; Araki 1985). $N$-body simulations give a somewhat larger value for this threshold — $\sigma_\mathrm{z} / \sigma_\mathrm{R} \simeq 0.6-0.8$ (e.g. Sotnikova & Rodionov 2003, 2006). For a disc in vertical dynamical equilibrium Zasov, Makarov & Mikhailova (1991) and Zasov et al. (2002) deduced a simple expression for $h/z_0$. We rewrite it in the following form: $$\frac{h}{z_0}
\sim
\frac{1}{\left[Q (\sigma_\mathrm{z} / \sigma_\mathrm{R})\right]^2}
\frac{\mathrm{M}_\mathrm{tot}}{\mathrm{M}_\mathrm{d}}
\sim
\frac{f_{\lambda}}{\left[Q (\sigma_\mathrm{z} / \sigma_\mathrm{R})\right]^2}
\frac{\mathrm{M}_\mathrm{tot}}{L_\mathrm{d}}\, ,
\label{dark}$$ where $Q$ is Toomre’s (1964) stability parameter, and $f_{\lambda}$ is the mass-to-luminosity ratio in a fixed band. The linear relation between the ratio $h/z_0$ and the ratio of dynamical mass (or dark halo mass) to disc luminosity arises for marginally stable discs with $Q$ and $\sigma_\mathrm{z} / \sigma_\mathrm{R}$ near their thresholds.
To verify a linear trend (\[dark\]), obtained analytically, Zasov et al. (2002) studied two different samples of edge-on bulgeless galaxies with known structural parameters in the $R$ and $K_s$ bands. They used the HI line width $W_{50}$ as a measure of dynamical mass and concluded that discs become thinner with increasing mass fraction of their dark halos. For our sample, which contains a great many galaxies with massive bulges, we performed the same analysis.
Fig. \[DHCorr\]a shows the distribution of the sample galaxies over the ratio of the dynamical mass to the stellar mass $\mathrm{M}_{*} = \mathrm{M}_\mathrm{d} + \mathrm{M}_\mathrm{b}$. We adopted mass-to-light ratios $f_J = 1.5$, $f_H = 1.0$, and $f_{K_s} = 0.8$ $\mathrm{M}_\odot / L_\odot$ (McGaugh et al. 2000). These mass-to-light ratios are consistent with the maximum disk best fitting the rotation curves for bright galaxies (e.g. Bottema 1999). There is a hint of bimodality in Fig \[DHCorr\]a. This may reflect the presence of two different families of galaxies with different bulges in our sample.
The distribution of the ratio $\mathrm{M}_\mathrm{tot}/\mathrm{M}_{*}$ ranges over the values from $\sim 1$ to $\sim 8-10$. The distribution of the ratio $h/z_0$ is varying in the same range (Fig. \[Distrib\]d). As the relation (\[dark\]) deals not only with the contribution of the dark halo to the total mass but with the relative mass of spherical components, including a bulge, we choose the ratio $\mathrm{M}_\mathrm{tot}/\mathrm{M}_\mathrm{d}$ to demonstrate a trend for the disc flattening with the relative mass of a spherical component (Fig. \[DHCorr\]b and Table \[CorrelationsDH\]). We excluded 6 points from fitting of $\mathrm{M}_\mathrm{tot}/\mathrm{M}_\mathrm{d}$ vs. $h/z0$: NGC 7814, NGC 4222, NGC 5965, NGC 4183 and NGC 100, that have faint disc, and NGC 5981 with a prominent bar.
The scatter of points in Fig. \[DHCorr\]b is rather large. It may be caused by uncertainties in velocities and mass-to-light ratio determinations, decomposition errors and disc relaxation processes, that can thicken a disc just above the threshold for marginal stability. But some of this scatter must be real. Sotnikova & Rodionov (2005) concluded that the presence of a compact bulge is enough to suppress the bending instability that leads to the disc thickness increasing. A series of $N$-body simulations with the same total mass of a spherical component (dark halo $+$ bulge) were performed. The final disc thickness was found to be much smaller in the simulations, where a dense bulge is present, than in the simulations with bulgeless systems. The results of $N$-body simulations of discs starting from an unstable state were summarized by Sotnikova & Rodionov (2006). They plotted the ratio $z_0 / h$ versus $(\mathrm{M}_\mathrm{h} + \mathrm{M}_\mathrm{b})/\mathrm{M}_\mathrm{d}$ and showed that there is a clear scatter in this relation, in spite of the same model mass for a spherical component ($\mathrm{M}_\mathrm{h} + \mathrm{M}_\mathrm{b}$). We are planning to compare in details our observational data with the results of simulations in a forthcoming paper.
-------------------------------------------------------------------------------------------------------------
$J$:$h/z_0 = (0.634 \pm 0.100)\mathrm{M}_\mathrm{tot}/\mathrm{M}_\mathrm{d} + (1.63 \pm 1.16)$, $r=0.58$
$H$: $h/z_0 = (0.696 \pm 0.120)\mathrm{M}_\mathrm{tot}/\mathrm{M}_\mathrm{d} + (1.76 \pm 1.51)$, $r=0.51$
$K_s$: $h/z_0 = (0.756 \pm 0.135)\mathrm{M}_\mathrm{tot}/\mathrm{M}_\mathrm{d} + (1.79 \pm 1.92)$, $r=0.46$
-------------------------------------------------------------------------------------------------------------
Summary and conclusions
=======================
We constructed the largest sample of edge-on spiral galaxies with performed 2D bulge/disc decomposition of 2MASS galaxies images in $J$-, $H$- and $K_s$-passband using the BUDDA v2.1 package. We extracted global bulge and disc parameters for all objects. This first paper describes our edge-on galaxy sample. The main point of this work is in the first carefully fitting of edge-on spiral galaxies of early and late morphological types in 3 near-infrared bands.
Edge-on galaxies are of great interest because they provide a unique possibility to obtain information about the vertical structure of a disk and a bulge. The disc scaleheight $z_0$ together with the disc scalelength $h$ determine the relative thickness of a stellar disc for each galaxy. This ratio as well as the bulge flattening $q_\mathrm{b}$ may put constraints on the disc and bulge formation processes and their secular evolution.
The results of the sample parameter analysis, that are specific for edge-on galaxies, can be summarized as follows:
1. Our data demonstrate a clear correlation between the disc scaleheight $z_0$ and the disc scalelength $h$ for all type galaxies. The distributions of the sample galaxies over the ratio of $h/z_0$ have a median value of 3.5 for $J$-band and 3.9 for $H$ and $K_s$-bands. But there is a substantial scatter of this ratio (the distribution is fairly flat). The ratio $h/z_0$ stretches over a large range — from very thin galaxies to plump ones. There is a correlation between the relative thickness of stellar discs $h/z_0$ and the relative mass of a spherical component, including a dark halo. This correlation was known previously for bulgeless galaxies (Zasov et al. 2002) or samples with predominantly bulgeless galaxies (Kregel et al. 2005) and was argued to arise from marginally stable discs. Our sample is much larger than the samples of Zasov et al. (2002) and Kregel et al. (2005) and more reliable for statistical analysis. Here, we confirm the correlation under discussion. What is more, we do this not for bulgeless galaxies but for galaxies with massive bulges. We conclude that the [*total*]{} mass contained in spherical components may be one of the factors that determines the final steady state disc thickness.
2. The bulge flattening in the vertical direction $q_\mathrm{b}$ divides the sample into two different families — triaxial, nearly prolate bulges and close to oblate bulges with moderate flattening. The [Sérsic]{} index threshold $n \simeq 2$ can be used to identify these two bulge types.
3. We found a correlation between $z_0$ and $r_\mathrm{e,b}$. It is new and has not been described previously. It means that the disc flattening is linked with the bulge structure.
Many of our results are in good agreement with the results of other authors, but we found several new relations.
1. In accordance with previous studies we found that the scale parameters of bulges and discs are tightly correlated. The disc scalelength linearly increases when the bulge effective radius increases. However, there is a clear trend for the ratio $r_\mathrm{e,b}/ h$ to increase with $n$. As $n$ is an indicator of the Hubble type, such a trend unambiguously rules out the widely discussed hypothesis of a scale-free Hubble sequence.
2. There is a hint that the fundamental planes of discs, which link only disc parameters and the maximum rotational velocity of gas, are different for galaxies with different bulges. This may indicate a real difference of discs in galaxies with low and high density bulges.
3. The most surprising result arises from the investigation of the Photometric Plane of sample bulges. The bulges with $n\ga2$ populate a narrow strip in their Photometric Plane. However, there is a difference in behavior of this plane for bulges with $n\ga2$ and $n\la2$. The plane is not flat and has a prominent curvature towards small values of $n$. For bulges this fact was not noticed earlier. This result may be due to the physical distinction between classical bulges and pseudobulges. As the Photometric Plane is proposed to be used as a distance indicator, the correct shape of the plane is of no little interest.
Some of results were described only briefly, but it is clear that our sample is very useful for further detailed studying and modelling of edge-on spiral galaxies. In the forthcoming papers we are planning to discuss properties of edge-on galaxies in more detail in the context of their bulge and disc formation and joint evolution.
Acknowledgments {#acknowledgments .unnumbered}
===============
We thank R.E. de Douza, D.A. Gadotti and S. dos Anjos for getting the last version of code BUDDA. We especially thank D.A. Gadotti for the helpful comments to this program. We are indebted to the referee, Phil James, for his constructive comments and insightful suggestions which helped to improve the quality and the presentation of the paper.
This work was supported by the Russian Foundation for Basic Research (grant 09-02-00968) and by a grant from President of the RF for support of Leading Scientific Schools (grant NSh-1318.2008.02).
This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. We made use of the LEDA database (http://leda.univ-lyon1.fr).
[99]{}
Aceves H., Velázquez H., Cruz F., 2006, MNRAS, 373, 632
Andredakis Y.C., Sanders R.H., 1994, MNRAS, 267, 283
Andredakis Y.C., Peletier R.F., Balcells M.J., 1995, MNRAS, 275, 874
Araki S., A theoretical Study of the Stability of Disk Galaxies and Planetary Rings, PhD Thesis, Massachus. Inst. Tech., 1985
Baggett W.E., Baggett S.M., Anderson K.S.J., 1998, AJ, 116, 1626
Barteldrees A., Dettmar R.-J., 1994, A&AS, 103, 475
Bedregal A.G., Aragon-Salamanca A., Merrifield M.R., 2006, MNRAS, 373, 1125
Bertola F., Vietri M., Zeilinger W.W., 1991, ApJ, 374, L13
Binney J., Merrifield M., 1998, Galactic Astronomy, Princeton Univ. Press, Princeton
Bizyaev D., Mitronova S., 2002, A&A, 389, 795 (BM02)
Blumenthal G.R., Faber S.M., Flores R., Primack J.R., 1986, ApJ, 301, 27
Boroson T., 1981, ApJS, 46, 177
Bottema R., 1999, A&A, 348, 77
Byun Y.I., Freeman K.C., 1995, ApJ, 448, 563
Castro-Rodriguez N. Garzón F., 2003, A&A, 411, 55
Courteau S., de Jong R.S., Broeils A.H., 1996, ApJ, 457, 73
Courteau S., Dutton A.A., van den Bosch F.C., Mac Arthur L.A., Dekel A., McIntosh D.H., Dale D.A., 2007, ApJ, 671, 203
Cutri R.M., Skrutskie M.F., Van Dyk S., et al., 2006, Explanatory Supplement to the 2MASS All Sky Data Release and Extended Mission Products, http://www.ipac.caltech,edu/2mass/releases/allsky/doc/
Dalcanton J.J., Bernstein R.A., 2000, AJ, 120, 203
Dalcanton J.J., Bernstein R.A., 2002, AJ, 124, 1328
de Grijs R., 1998, MNRAS, 299, 595
de Grijs R., van der Kruit P.C., 1996, A&AS, 117, 19
de Jong R.S., 1996, A&A, 313, 45
de Rijcke S., Zeilinger W.W., Hau G.K.T., Prugniel P., Dejonghe H., 2007, ApJ, 659, 1172
de Vaucouleurs, 1948, Annales d’Astrophysique, 11, 247
de Vaucouleurs, 1953, MNRAS, 113, 134
de Vaucouleurs, 1959, Hdb. d. Physik, 53, 311
D’Onofrio M., Fasano G., Varela J., Bettoni D., Moles M., Kjærgaard P., Pignatelli E., Poggianti B., Dressler A., Cava A., Fritz J., Couch W.J., Omizzolo A., 2008, ApJ, 685, 875
de Souza R.E., Gadotti D.A., dos Anjos S., 2004, ApJS, 153, 411 (SGA04)
Erwin P., Sparke L.S., 2002, AJ, 124, 65
Faber S.M., Jackson R.E., 1976, ApJ, 204, 668
Fall S.M., Efstatiou G., 1980, MNRAS, 193, 189
Fathi K., Peletier R.F., 2003, A&A, 407, 61
Fisher D.B., Drory N., 2007, ApJ, 664, 640
Fisher D.B., Drory N., 2008, AJ, 136, 773
Fraser C.W., 1972, Observatory, 92, 51
Freeman K.C., 1970, ApJ, 160, 811
Gadotti D.A., 2009, MNRAS, 393, 1531
Graham A.W., 2001, AJ, 121, 820
Graham A.W., Prieto M., 1999, ApJ, 524, L23
Hunt L.K., Pierini D., Giovanardi C., 2004, A&A, 414, 905
Jarrett T.H., 2000, PASP, 112, 1008
Jarrett T.H., Chester T., Cutri R., Schneider S., Skrutskie M., Huchra J.P., 2000, AJ, 119, 2498
Karachentsev I.D., 1989, AJ, 97, 1566
Karachentsev I.D., Karachentseva V.E., Kudrya Yu.N., Sharina M.E., Parnovskij S.L., 1999, Bull. Spec. Astrophys. Obs., 47, 185
Karachentsev I.D., Mitronova S.N., Karachentseva V.E., Kudrya Yu.N., Jarrett T.H., 2002, A&A, 396, 431
Kent S.M., 1985, ApJS, 59, 115
Khosroshahi H.G., Wadadekar Y., Kembhavi A., Mobasher B., 2000a, ApJ, 531, L103
Khosroshahi H.G., Wadadekar Y., Kembhavi A., 2000b, ApJ, 533, 162
Khosroshahi H.G., Raychaudhury S., Ponman T.J., Miles T.A., Forbes D.A., 2004, MNRAS, 349, 527
Kormendy J., 1977, ApJ, 218, 333
Kormendy J., Kennicutt Jr. R.C., 2004, ARAA, 42, 603
Kregel M., Structure and Kinematics of Edge-on Galaxy Disks, PhD Thesis, Univ. Groningen, 2003
Kregel M., van der Kruit P.C., de Grijs R., 2002, MNRAS, 334, 646
Kregel M., van der Kruit P.C., Freeman K.C., 2005, MNRAS, 358, 503
Lahav O., Naim A., Buta R.J., Corwin H.G., de Vaucouleurs G., Dressler A., Huchra J.P., van den Bergh S., Raychaudhury S., Sodre L., Jr., Storrie-Lombardi M.C., 1995, Science, 267, 859
Laurikainen E., Salo H., Buta R., 2005, MNRAS, 362, 1319
MacArthur L.A., Courteau S., Holtzman J.A., 2003, ApJ, 582, 689
Márquez I., Lima Neto G.B., Capelato H., Durret F., Lanzoni B., Gerbal D., 2001, A&A, 379, 767
Masters K.L., Springob C.M., Huchra J.P., 2008, AJ, 135, 1738
McGaugh S.S., Schombert J.M., Bothun G.D., de Blok W.J.G., 2000, ApJ, 533, L99
Mitronova S.N., Karachentsev I.D., Karachentseva V.E., Jarrett T.H., Kudrya Yu.N., 2003, Bull. Spec Astrophys. Obs., 57, 5
Méndez-Abreu J., Aguerri J.A.L., Corsini E.M., Simonneau E., 2008, A&A, 478, 353
Mo H.J., Mao S., White S.D.M., 1998, MNRAS, 295, 319.
Möllenhoff C., 2004, A&A, 415, 63
Möllenhoff C., Heidt J., 2001, A&A, 368, 16
Moffat A.F.J., 1969, A&A, 3, 455
Moriondo G., Giovanardi C., Hunt L.K., 1998, A&AS, 130, 81
Moriondo G., Giovanelli R., Haynes M.P., 1999, A&A, 346, 415
Nakamura O., Fukugita M., Yasuda N., Loveday J., Brinkmann J., Schneider D., Shimasaku K., SubbaRao M., 2003, AJ, 125, 1682
Noordermeer E., van der Hulst J.M., 2007, MNRAS, 376, 1480
Reshetnikov V., Combes F., 1997, A&A, 324, 80
Sérsic J.L., 1968, Atlas de Galaxias Australes, Observatorio Astronomico, Cordoba
Schlegel D.J., Finkbeiner D.P., Davis M., 1998, ApJ, 500, 525
Schmidt M., 1968, ApJ, 151, 393
Scodeggio M., Gavazzi G., Franzetti P., Boselli A., Zibetti S., Pierini D., 2002, A&A, 384, 812
Seth A.C., Dalcanton J.J., de Jong R.S., 2005, AJ, 130, 1574
Skrutskie M.F., Cutri R.M., Stiening R., et al., 2006, AJ, 131, 1163
Sotnikova N.Ya., Rodionov S.A., 2003, Astr. Lett., 29, 321
Sotnikova N.Ya., Rodionov S.A., 2005, Astr. Lett., 31, 15
Sotnikova N.Ya., Rodionov S.A., 2006, Astr. Lett., 22, 649
Strateva I., Ivezić Ž., Knapp G.R., et al., 2001, AJ, 122, 1861
Thuan T.X., Seitzer P.O., 1979, ApJ, 231, 680
Toomre A., 1964, ApJ, 139, 1217
van der Kruit P.C., Searle L., 1981, A&A, 95, 105
van der Kruit P.C., de Grijs R., 1999, A&A, 352, 129
White S.D.M., Rees M.J, 1978, MNRAS, 183, 341
Yoachim P., Dalcanton J.J., 2005, ApJ, 624, 701
Zasov A.V., Makarov D.I., Mikhailova E.A., 1991, Astron. Lett., 17, 374
Zasov A.V., Bizyaev D.V., Makarov D.I., Tyurina N.V., 2002, Astron. Lett., 28, 527
---- --------------- ------- ------ ------- -- -- --
\# Galaxy 2MFGC Type $D$
(Mpc)
(1) (2) (3) (4)
1 UGC 529 633 S0-a 70.4
2 NGC 504 1059 S0 53.2
3 NGC 527 1070 S0-a 74.7
4 NGC 565 1133 Sa 57.0
5 UGC 1120 1193 Sab 58.8
6 ESO 353-G020 1202 S0-a 61.7
7 UGC 1166 1243 S0-a 60.8
8 ESO 013-G024 1422 Sab —
9 UGC 1531 1557 Sb 105.0
10 IC 207 1660 S0-a 62.1
11 NGC 861 1758 Sb 106.0
12 UGC 1938 1930 Sbc 82.0
13 NGC 955 1965 Sab 17.0
14 NGC 960 1984 Sb 63.1
15 MCG-02-07-038 2044 S0 62.2
16 NGC 1029 2102 S0-a 46.0
17 UGC 2304 2260 Sbc 93.7
18 MCG-02-09-013 2605 S0-a 61.2
19 ESO 005-G004 4913 Sb 26.1
20 NGC 1381 2983 S0 22.2
21 NGC 1394 3025 S0 55.5
22 NGC 1401 3030 S0 18.8
23 NGC 4256 9696 Sb 35.7
24 MCG-02-08-012 2274 Sc 68.8
25 NGC 1529 3366 S0 68.7
26 UGC 7388 — SBb 89.8
27 ESO 251-G028 3869 Sb 158.0
28 NGC 4866 10288 S0-a 31.2
29 ESO 555-G023 4860 S0-a —
30 ESO 555-G032 4905 Sb —
31 ESO 087-G004 5097 Sbc 118.0
32 UGC 4012 6163 Sb 130.0
33 UGC 4190 6390 S0 69.9
34 UGC 4507 6803 Sb 141.0
35 UGC 4526 6822 Sab 62.4
36 ESO 433-G012 7223 Sbc 115.0
37 ESO 373-G013 7473 S0-a 40.7
38 NGC 2946 7484 SBb 122.0
39 UGC 5345 7714 Sb 53.5
40 IC 615 8108 Sb 133.0
41 ESO 501-G056 8260 S0 53.0
42 NGC 3431 8449 SABb 75.3
43 MCG-01-28-005 8496 Sab 91.0
44 CGCG 267-039 8658 Sb 88.2
45 NGC 3539 8691 S0-a 132.0
46 NGC 1184 2685 S0-a 30.8
47 NGC 4149 9578 Sb 43.4
48 MCG-02-32-006 9725 SBc 60.0
49 NGC 4686 10067 Sa 69.5
50 MCG-02-33-076 10273 Sab 47.0
51 UGC 8119 10315 Sbc 117.0
52 IC 844 10368 S0 43.6
53 MCG-01-33-076 10374 Sc 58.4
54 IC 5096 16137 Sbc 41.2
55 IC 881 10662 Sa 96.9
56 NGC 5119 10737 S0-a 43.6
57 UGC 8462 10813 Sab 132.0
58 UGC 8704 11102 Sa 140.0
59 NGC 5290 11096 Sbc 37.4
60 NGC 5553 11621 Sa 63.8
---- --------------- ------- ------ ------- -- -- --
: The Sample.[]{data-label="Table1"}
----- -------------- ------- ------ ------- -- -- --
\# Galaxy 2MFGC Type $D$
(Mpc)
(1) (2) (3) (4)
61 ESO 384-G014 11249 S0 57.2
62 IC 1110 12300 Sa 46.0
63 ESO 387-G029 12429 Sab 60.6
64 UGC 11060 14135 Sa 61.1
65 NGC 4235 9679 Sa 37.4
66 UGC 11614 15682 Sb 105.0
67 IC 5181 16757 S0 24.1
68 NGC 7232 16791 S0-a 23.1
69 ESO 471-G024 17877 S0-a 113.0
70 NGC 482 1009 Sab 85.0
71 NGC 522 1102 Sbc 32.8
72 NGC 585 1169 Sa 68.9
73 NGC 653 1286 Sab 69.9
74 NGC 684 1382 Sb 44.1
75 NGC 1596 3625 S0 20.5
76 IC 2085 3680 S0-a 13.4
77 NGC 2295 5413 Sab 26.9
78 ESO 311-G012 6172 S0-a 18.2
79 NGC 3203 7996 S0-a 37.5
80 NGC 3404 8433 SBab 66.8
81 NGC 5047 10588 S0 88.7
82 ESO 321-G010 9597 Sa 46.5
83 ESO 416-G025 2246 Sb 65.0
84 NGC 3390 8406 Sb 45.7
85 IC 127 1149 Sb 23.1
86 NGC 3628 — Sb 16.2
87 NGC 3717 — Sb 28.2
88 NGC 5965 12550 Sb 46.8
89 NGC 4013 9412 Sb 14.5
90 ESO 446-G018 11466 Sb 67.5
91 ESO 512-G012 11898 Sb 49.5
92 NGC 4710 10109 S0-a 19.6
93 NGC 7814 — Sab 9.5
94 ESO 240-G011 — Sc 35.7
95 NGC 4565 — Sb 20.7
96 NGC 5746 — SABb 26.4
97 NGC 891 — Sb 4.3
98 NGC 5775 12067 SBc 25.7
99 NGC 4217 9661 Sb 16.8
100 NGC 5908 — Sb 45.6
101 NGC 4244 — Sc 6.8
102 NGC 4570 9935 S0 28.1
103 NGC 5907 12346 Sc 10.0
104 NGC 3564 — S0 42.5
105 NGC 4266 9707 SBa 25.6
106 IC 4202 10456 Sbc 98.5
107 NGC 4703 10103 Sb 64.5
108 ESO 443-G042 10371 Sb 43.5
109 NGC 5308 11118 S0 29.1
110 NGC 5365A 11257 SBb 40.5
111 IC 1048 11928 Sb 25.2
112 UGC 243 293 Sb 66.6
113 ESO 315-G020 7524 Sb 68.9
114 NGC 669 1340 Sab 59.5
115 NGC 973 2024 Sb 62.4
116 IC 335 2973 S0 20.7
117 IC 1970 2982 Sb 15.9
118 NGC 1380A 2985 S0 20.0
119 ESO 362-G011 4306 Sbc 18.8
120 NGC 1886 4383 Sbc 24.3
----- -------------- ------- ------ ------- -- -- --
: continued.[]{data-label="Table2"}
----- --------------- ------- ------ ------- -- -- --
\# Galaxy 2MFGC Type $D$
(Mpc)
(1) (2) (3) (4)
121 NGC 2310 5487 S0 17.9
122 NGC 2654 6926 SBab 19.8
123 NGC 2862 7315 SBbc 58.9
124 NGC 3126 7847 Sb 73.2
125 NGC 3692 8974 Sb 28.3
126 NGC 3957 9342 S0-a 27.1
127 NGC 4417 9805 S0 15.8
128 NGC 1247 2619 Sbc 50.9
129 NGC 5981 12588 Sbc 24.5
130 NGC 4026 9425 S0 15.3
131 NGC 3501 — Sc 20.0
132 NGC 5023 10525 Sc 8.3
133 NGC 360 761 Sbc 30.1
134 UGC 1817 1825 Scd 47.2
135 MCG-02-10-009 3079 Sc 27.7
136 UGC 3326 4605 Sc 55.1
137 UGC 3474 5237 Sc 49.1
138 NGC 3044 7660 SBc 22.3
139 NGC 3279 8207 Sc 23.7
140 NGC 4222 9670 Sc 7.7
141 ESO 533-G004 16767 Sc 31.2
142 NGC 4330 9747 Sc 25.8
143 ESO 564-G027 7159 Sc 33.8
144 ESO 531-G022 16371 Sbc 42.1
145 IC 4393 11612 Sc 40.5
146 ESO 263-G015 7900 Sc 38.0
147 NGC 4835A 10246 Sc 49.3
148 ESO 288-G025 16587 Sbc 30.7
149 IC 2058 3483 Scd 18.6
150 ESO 201-G022 3389 Sc 55.4
151 IC 4871 15015 SABc 25.1
152 IC 4484 11988 Sc 61.3
153 ESO 340-G008 15435 Sc 36.1
154 UGC 12533 17534 Sb 75.8
155 ESO 509-G019 10818 Sbc 144.0
156 ESO 506-G002 9712 Sbc 57.9
157 NGC 4183 9620 Sc 15.8
158 NGC 4517 9881 Sc 20.1
159 NGC 2357 5811 Sc 33.0
160 UGC 5173 7514 Sb 87.6
161 NGC 100 282 Sc 6.7
162 ESO 121-G006 4933 Sc 17.6
163 NGC 2424 6083 SBb 47.3
164 IC 2207 6207 Sc 66.7
165 ESO 563-G014 6848 SBcd 27.1
166 UGC 6012 8487 Sbc 88.8
167 ESO 507-G007 10056 Sbc 75.2
168 IC 3799 10098 Scd 54.4
169 MCG-03-34-041 10613 Sc 40.2
170 NGC 5073 10645 SBc 41.4
171 IC 4351 — Sb 39.8
172 NGC 5529 11577 Sc 41.4
173 NGC 5714 11872 Sc 32.1
174 ESO 340-G026 15509 Sc 71.2
175 UGCA 150 7144 SABb 29.3
----- --------------- ------- ------ ------- -- -- --
: continued.[]{data-label="Table3"}
\[lastpage\]
[^1]: E-mail: [email protected]
[^2]: see http://www.mpa-garching.mpg.de/$\sim$dimitri/budda.html
[^3]: For NGC 5308 and NGC 1184 Bedregal et al. (2006) give the formal errors for the bulge effective radii that are very large and asymmetric. Unfortunately, they did not provide any explanation of these facts. For both galaxies, the bulge effective radii from our study and those by Bedregal et al. (2006) are quite different.
[^4]: We use capital letters $A$, $B$, $C$ to define the intrinsic semi-axes of a triaxial system and small letters $a$, $b$ for the apparent semi-axes of an image. In axisymmetric case and an edge-on view the ratio $b/a$ coincides with the ratio $C/A$.
[^5]: The mean effective surface brightness within the circle of effective radius $r_\mathrm{e,b}$.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We present here a set of lecture notes on quantum systems with time-dependent boundaries. In particular, we analyze the dynamics of a non-relativistic particle in a bounded domain of physical space, when the boundaries are moving or changing. In all cases, unitarity is preserved and the change of boundaries does not introduce any decoherence in the system.'
address:
- |
Dipartimento di Matematica, Università di Bari, Via Orabona 4\
Bari, I-70125, Italy
- |
Dipartimento di Fisica and MECENAS, Università di Bari, Via Amendola 173\
Bari, I-70126, Italy\
INFN, Sezione di Bari, Via Amendola 173\
Bari, I-70126, Italy
author:
- SARA DI MARTINO
- PAOLO FACCHI
title: 'QUANTUM SYSTEMS WITH TIME-DEPENDENT BOUNDARIES'
---
Introduction
============
We present here the notes of three lectures given by one of us at the International Workshop on Mathematical Structures in Quantum Physics, held in February 2014 in Bangalore at the Centre for High Energy Physics, Indian Institute of Science. The course considers some aspects of quantum systems with time-dependent boundaries, a very active area both from the mathematical point of view, see for instance the works of Yajima [@Yajima87; @Yajima96], Dell’Antonio *et al* [@Dell'Antonio] and Posilicano *et al* [@Posilicano07; @Posilicano11], and from a physical perspective. Notable applications arise in different fields ranging from atoms in cavities [@HarocheRaimond1; @HarocheRaimond2] to ions and atoms in magnetic traps [@atoms_ions], to superconducting quantum interference devices (SQUID) [@reviewSQUID], to the dynamical Casimir effect in microwave cavities [@photons]. Moreover, the role and importance of boundary conditions at a fundamental level have been stressed in an interesting series of articles, see [@Bala; @Wilczek] and references therein, where varying boundary conditions are viewed as a model of space-time topology change.
These notes will be organized as follows. In the first lecture, in Section \[sec-interval\], we will study the problem of a quantum particle moving in a one-dimensional box. In particular we will introduce and describe the problem in a pedagogical way focussing on the mathematical aspects of the physical model. We will try to give all the mathematical tools needed to a deep understanding of the physics of the problem. Moreover, at the end of the section, we will follow Berry [@berry] and show how, quite surprisingly, even an easy spectral problem like this can lead to intricate and rich dynamics. In the following lectures, we will look at the dynamical evolution with time-dependent boundaries. In particular, in the second lecture, in Section \[sec-MW\], we will discuss the case in which the walls of the bounding box are moving [@MW], while in the last lecture, in Section \[sec-change\], we will look at the problem of the change of boundary conditions with time [@AFMP]. Finally, in Section \[sec-conc\] we will draw some conclusions.
Interspersed between the lectures we offer a selection of exercises ranging in difficulty. They are useful to focus the attention on specific points and to test the learning of the reader, therefore we kindly recommend not to skip them.
Lecture 1: A particle in a one-dimensional box {#sec-interval}
==============================================
One of the very first problems involving boundary conditions in quantum mechanics is the study of a non-relativistic particle in a one-dimensional box. Indeed, it is one of the exercises sometimes professors give to students just to test their ability. The problem consists in a massive particle moving freely in an infinitely deep well located, say, at $I=[a,b]$ with $b>a$, from which it cannot escape (Fig. \[fig:box\]).
![A particle moving in a one-dimensional box.[]{data-label="fig:box"}](Fixed.pdf)
We will see that even if the problem is apparently fairly simple, it leads to very interesting results that stress the difference between classical and quantum mechanics. In fact, the very presence of boundaries in general tends to enhance the quantum aspects of the system. The reason is that in quantum mechanics the behavior at the boundary is constrained by the very structure of the theory and in particular by unitarity. This is at variance with classical mechanics, where the behavior at the boundary does not derive from fundamental principles, and is usually treated phenomenologically.
From the mathematical point of view the system is described by a Hamiltonian operator associated to the kinetic energy of the particle, namely the Laplacian, $$\label{eq:ham}
T=\frac{p^2}{2m}=-\frac{\hbar^2}{2m}\frac{\d^2}{\d x^2},$$ acting on square integrable functions (remember that $|\psi(x)|^2$ is a probability density!) with square integrable second (distribution) derivative on the interval $I$ of the box, $\mathcal{H}^2(I)=\{\psi\in L^2(I)\ |\ p^2\psi\in L^2(I)\}$. In the mathematical literature $\mathcal{H}^2(I)$ is known as the second Sobolev space; it is nothing but the maximal domain of definition of the kinetic energy operator $T$.
Equation (\[eq:ham\]) describes the action of $T$ in the bulk of the box. In order to get a well-defined dynamics, Eq. (\[eq:ham\]) should be equipped with suitable *boundary conditions*, which specify the behavior of the particle at the walls. Namely, the wave functions $\psi$ to which we are allowed to apply $T$ must belong to the domain $$D(T)= \{\psi \in \mathcal{H}^2(I), \text{with suitable boundary conditions}\}.$$ The question that here arises naturally is how to choose these boundary conditions. This marks the difference between classical and quantum mechanics: in quantum mechanics the behavior at the boundary, encoded in the domain $D(T)$, derives from basic principles. Indeed, in quantum mechanics the conservation of probability leads to the necessity for the evolution to be represented by unitary operators. But unitarity of the evolution is equivalent to self-adjointness of its generator, as stated in Stone’s theorem [@Kato]:
[(Stone)]{} Let $(U_t)_{t\in\mathbb{R}}$ be a strongly continuous one-parameter group of unitaries on a Hilbert space $\H$. Then there is a unique self-adjoint operator $H$ on $\H$, called the generator of the group, such that $U_t=\e^{-\ii H t/\hbar}$ for all $t$.
According to Stone’s theorem the operator $T$, that generates the dynamics of the particle, $U_t = \e^{-\ii T t/\hbar}$, must be self-adjoint, i.e. $$T=T^\dagger \qquad \text{and}\quad D(T)=D(T^\dagger).$$ In order to find the domain $D(T)$ let us consider $\phi,\psi\in \mathcal{H}^2([a,b])$; then a double integration by part gives (prove it!): $$\begin{aligned}
\label{eq:prod-scal}
-{\langle\psi\vert \phi''\rangle}&=&- \int_{a}^{b}\overline{\psi(x)}\phi''(x)\, \d x = -{\langle\psi''\vert \phi\rangle}+ \Lambda(\psi,\phi),
\\
\Lambda(\psi,\phi)&=&\overline{\psi'(b)}\phi(b)-\overline{\psi'(a)}\phi(a)-\overline{\psi(b)}\phi'(b)+\overline{\psi(a)}\phi'(a).
\label{eq:Lambdadef}\end{aligned}$$ The self-adjointness requirement amounts to finding a maximal subspace $D(T)$ of $\mathcal{H}^2([a,b])$ on which the *boundary form* $\Lambda$ identically vanishes, namely $\Lambda(\psi,\phi)=0$ for all $\phi, \psi \in D(T)$.
The more common boundary conditions you can encounter in the literature are:
1. [*Dirichlet boundary conditions*]{}: $$\label{eq:dir}
\psi(a)=\psi(b)=0;$$
2. [*Neumann boundary conditions*]{}, involving the derivative of the function: $$\label{eq:neu}
\psi'(a)=\psi'(b)=0.$$ (Of course, the above conditions must be satisfied by *both* $\psi$ and $\phi$ in the boundary form $\Lambda(\psi,\phi)$!)
Check that the Dirichlet and the Neumann boundary conditions make $\Lambda$ identically vanish, and thus, modulo a check of maximality, yield a good domain $D(T)$ of self-adjointness for the kinetic energy operator $T$.
There is a one-parameter family of boundary conditions that includes Dirichlet and Neumann as limit cases: *Robin’s boundary conditions*, that connect the function with its derivative at the boundary: $$\label{eq:rob}
\psi'(a)=-\frac{1}{l_0}\tan \frac{\alpha}{2}\, \psi(a), \qquad
\psi'(b)=\frac{1}{l_0}\tan \frac{\alpha}{2}\, \psi(b),$$ where $\alpha\in(-\pi,\pi]$, and $l_0$ is a reference length. From (\[eq:rob\]) we recover Dirichlet and Neumann in the limits $\alpha\to\pi$ and $\alpha\to0$, respectively.
Notice that the choice of a domain of self-adjointness $D(T)$ is not just a mathematical nuisance. Different domains give rise to different physics! Indeed, the choice of boundary conditions strongly depends on the physical behavior of the walls, as we will now show.
Let us concentrate on one wall, e.g. the left one at $a=0$, and look at the scattering process of a plane wave arriving from the left with momentum $-\hbar k <0$. After reflection by the wall a plane wave with momentum $\hbar k$ will arise.
![Reflection at a wall[]{data-label="fig:walls"}](Walls.pdf)
Therefore, the wave function will have the form $$\psi(x)\sim \e^{-\ii kx}+r\,\e^{\ii kx},$$ where $r$ is a reflection coefficient. Since we are dealing with an impenetrable box, and we assume that the particle does not stick at the wall, we are forced to require that $|r|^2=1$, so that the only freedom left is in choosing a phase, namely $r=\e^{\ii \beta}$. The Dirichlet boundary condition, $\psi(0)=0$, corresponds to a phase $\beta=\pi$ for all $k$, while Neumann corresponds to the choice $\beta=0$ for all $k$. Notice that in general the phase $\beta(k)$ has a nontrivial dependence on the wavenumber $k$. The following exercise shows that for a generic $\alpha$ in (\[eq:rob\]) the behavior of the walls depends on the momentum $\hbar k$ of the impinging particle through a nontrivial phase $\beta(k)$.
Prove that Robin’s boundary condition $\psi'(0)=-\frac{1}{l_0}\tan \frac{\alpha}{2}\, \psi(0)$ corresponds to a momentum dependence of the phase shift $\beta(k)$ given by $$\tan \frac{\beta(k)}{2} = \frac{1}{k\, l_0} \tan \frac{\alpha}{2}.$$ Recover, as a limit, Dirichlet and Neumann, and show that they are the only phases independent of $k$.
The general theory of self-adjoint extensions and boundary conditions [@AIM] asserts that all possible boundary conditions that make $T$ self-adjoint can be parametrized by $2\times 2$ unitary matrices $U$ in the following way: $$\ii (\I +U) \Psi' = (\I-U) \Psi, \qquad U\in\cU(2),
\label{eq:bc2}$$ where $$\Psi := \left(\begin{array}{r} \psi(a) \\
\psi(b)
\end{array}\right), \qquad \Psi' := l_0\left(\begin{array}{r}
- \psi'(a) \\
\psi'(b)\end{array}\right),
\label{eq:bc21}$$ and $\I$ is the identity matrix.
Check that if $\psi$ and $\phi$ satisfy (\[eq:bc2\])-(\[eq:bc21\]) for the same unitary $U$, then the boundary form (\[eq:Lambdadef\]) vanishes, $\Lambda(\psi,\phi)=0$.
Robin’s boundary conditions (\[eq:rob\]) correspond to (prove it!) $$U=\e^{-\ii\alpha} \I
=\left(\begin{array}{l l}
\e^{-\ii\alpha} &0 \\
0&\e^{-\ii\alpha}
\end{array}\right),$$ and in particular the Dirichlet boundary conditions correspond to $U=-\I$, while the Neumann ones are given by $U=\I$. More generally, by taking $$U=\left(\begin{array}{l l}
\e^{-\ii\alpha_1} &0 \\
0&\e^{-\ii\alpha_2}
\end{array}\right),
\label{eq:local}$$ with $\alpha_1, \alpha_2 \in (-\pi,\pi]$ we can describe different boundary conditions at the two endpoints, e.g. Dirichlet/Neumann ($\alpha_1=\pi$, $\alpha_2=0$), and include all cases considered above, when $\alpha_1=\alpha_2$. But, of course the unitaries (\[eq:local\]) do not exhaust all allowed boundary conditions, since they form only a two-dimensional submanifold $\mathcal{U}(1)\times \mathcal{U}(1)$ of the total four-dimensional manifold $\mathcal{U}(2)$. Thus, where do all missing boundary conditions come from?
Up to now we have considered only *local* boundary conditions, somewhat deceived by the physics depicted in Fig. \[fig:walls\]: the two walls did not talk each other. Indeed, the unitary matrices in (\[eq:local\]) are diagonal and do not mix the boundary values at $a$ with those at $b$; we missed all nondiagonal unitaries, describing *nonlocal* boundary conditions. Mathematics tells us that unitarity allows also for them. So what physics do they describe, if any?
![Change of topology: from an interval to a circle[]{data-label="fig:bend"}](bending.pdf)
In fact, nondiagonal unitaries are describing a physical situation which is different from that of a box with two walls far apart. In order to interact, the two ends of the interval should come close, as in Fig. \[fig:bend\], so that the interval should be bent and the two walls should become the two sides of a junction. In other words, by changing from a diagonal $U$ to a nondiagonal $U$ we are assisting at a change of topology: from an interval to a circle.
Therefore, the geometry that is able to support all possible boundary conditions in $\mathcal{U}(2)$ is that of a ring with a junction. If the junction is impermeable, i.e. there is total reflection at the walls, we are back to the interval, otherwise there is nonzero transmission across the junction, from one wall to the other. An interesting example is given by the matrix $$U= \left(\begin{array}{ll}0 & \e^{-\ii\alpha} \\ \e^{\ii\alpha} &\ 0 \end{array}\right) =
\cos \alpha\, \sigma_x + \sin\alpha\, \sigma_y,
\label{eq:Uperiodic}$$ which describes pseudo-periodic boundary conditions (prove it!): $$\psi(b)=\e^{\ii \alpha}\, \psi(a), \qquad \psi'(b)=\e^{\ii \alpha}\, \psi'(a).
\label{eq:phiperiodic}$$ By passing through the junction, the wave function acquires a phase $\alpha$. If $\alpha=0$ we get the famous periodic boundary conditions, and the geometry is that of a circle; if $\alpha=\pi$ we get antiperiodic boundary conditions. The phase $\alpha$ encodes the properties of the junction, for example the material it is made of, or its width.
In general, if the unitary $U$ has no $-1$ eigenvalues, the wave function $\psi$ can assume any value at the endpoints of the interval. Only the boundary values of its derivative are constrained in some way. On the other hand, one eigenvalue equal to $-1$ corresponds to one constraint on the values of $\psi$ at the ends, as for example in the first equation in (\[eq:phiperiodic\]). Finally, two $-1$ eigenvalues, i.e. $U=-\I$, correspond to two constraints on the boundary values of the wave function, i.e. Dirichlet at both ends.
Prove that the unitary (\[eq:Uperiodic\]) has always an eigenvalue equal to $-1$. Find its corresponding eigenvector $\xi$ and show that the first equation in (\[eq:phiperiodic\]) is nothing but an orthogonality condition ${\langle\xi\vert \Psi\rangle}=0$. (In Lecture 3 we come back to the geometrical meaning of this condition.)
Hard walls
----------
Let us concentrate now on the textbook case of a particle confined in an infinitely deep well, Fig. \[fig:box\]. The appropriate boundary conditions are Dirichlet’s: $\psi(a)=\psi(b)=0$. The dynamics of the particle is described by the Schrödinger equation $$\ii\hbar \frac{\partial \psi(x,t)}{\partial t}= -\frac{\hbar^2}{2m} \frac{\partial^2 \psi(x,t)}{\partial x^2}.
\label{eq:schr_eq}$$ A separation of variables, $\psi(x,t)=u(x) \exp(-\ii E t/\hbar)$, reduces the problem to the solution of the spatial part of the differential equation, that means to find eigenvectors and eigenvalues of the operator $T$: $$-\frac{\hbar^2}{2m}u''(x)=E\, u(x).$$ The general solution is $$u(x)=c_1\, \e^{+\ii k x}+c_2\,\e^{-\ii k x},$$ with $k = \sqrt{2 m E}/\hbar$ (in principle $k$ can be imaginary, but see below), and $c_1$ and $c_2$ are arbitrary constants that can be fixed (up to a common phase) by imposing the Dirichlet boundary conditions, $$\begin{aligned}
u(a)&=&c_1\, \e^{\ii k a}+c_2\, \e^{-\ii k a}=0\\
u(b)&=&c_1\, \e^{\ii k b}+c_2\,\e^{-\ii k b}=0,\end{aligned}$$ and normalization $${\langleu\vert u\rangle}=1.$$
Prove that the normalized eigenfunctions of $T$, with the Dirichlet boundary conditions, are $$\label{eq:eigen}
u_n(x)=\sqrt{\frac{2}{l}} \sin \left(\frac{n \pi}{l} (x-a)\right),$$ where $l=b-a$, and that the eigenvalues, giving the permitted energy levels are $$\label{eq:energy}
E_n=\frac{\hbar^2}{2m}\frac{n^2\pi^2}{l^2},$$ for $n=1,2,\dots$. See Fig. \[fig:eigen\]
![The eigenfunctions (solid lines) of the operator $T$ in the box at different levels of energy (dashed lines).[]{data-label="fig:eigen"}](EigenfunctionsDef.pdf)
Eq. (\[eq:energy\]) tells us two important things on the bound states of the particle:
1. the energies are quantized;
2. the energy is always strictly positive.
The strictly positivity of the energy is a property of the Dirichlet boundary conditions. Indeed, with the Neumann ones the energy of the ground state is 0, and, more surprisingly, it is even *negative* with Robin’s boundary conditions!
Prove that the ground state of $T$, with the Neumann boundary conditions is $$v_0(x)=\sqrt{\frac{1}{l}},
\label{eq:v0}$$ and has zero energy, $E_0=0$. Then, look at the eigenvalue problem with Robin’s boundary conditions (\[eq:rob\]).
Fractals in a box
-----------------
The textbook exercise of the quantum particle in a box inevitably ends with the evaluation of the eigenvalues (\[eq:energy\]) and the eigenfunctions (\[eq:eigen\]). The result is so simple and intelligible that we all felt a profound satisfaction when we derived it in our first course of quantum mechanics. The simplicity of the spectrum is deceptive and leads us to think that we fully understand the physical problem. In particular, we are convinced that the dynamics, which is the solution to the Schrödinger equation (\[eq:schr\_eq\]), must surely be as much simple. In fact, this belief is false, as showed by Berry [@berry]: the dynamics is instead very intricate.
Let us assume at time $t=0$ that $\psi(x,0)=v_0(x)$, with $v_0$ given by (\[eq:v0\]). This is the simplest conceivable initial condition, corresponding to a flat probability in the box $[a,b]$, with $l=b-a$.
![Graphs of $|\vartheta(\xi,\tau)|^2$ vs $\xi$ at different rational times $\tau$ along the Fibonacci sequence tending to the golden mean, $\phi=(1+\sqrt{5})/2$. See the emergence of a fractal structure.[]{data-label="fig:Fractals"}](Fractals.pdf)
We are interested in the time evolution of this initial wave function. Its $L^2$-expansion in terms of the eigenfunctions (\[eq:eigen\]) of $T$ reads $$v_0(x)=\sum_{n=1}^\infty c_n u_n(x),$$ where $c_n={\langleu_n\vert v_0\rangle}$.
Show that $$c_n=\frac{\sqrt{2}}{n\pi} \left[1-(-1)^n\right],
\label{eq:cn}$$ and, in particular, $c_{2n}=0$, for all $n=1,2,\dots$.
In the same way the quantum evolution, described by the action of the unitary operator $U(t)=\e^{-iTt/\hbar}$, can be written as an $L^2$-convergent series $$\psi(x,t)=(\e^{-\ii T t/\hbar}v_0)(x)=\sum_{n=1}^{+\infty}c_n\ \e^{-\ii E_n t/\hbar}\ u_n(x).$$ We can now use the explicit expressions (\[eq:eigen\])-(\[eq:energy\]) and obtain $$\psi(x,t)=\sqrt{\frac{2}{l}}\ \sum_{n=1}^{+\infty}c_n \sin \left(\frac{n \pi}{l} (x-a)\right)\exp\left(-\frac{\ii \hbar}{2m}\frac{n^2\pi^2 t}{l^2}\right) .$$ In terms of the dimensionless variables $\xi = l^{-1} (x- (a+b)/2) \in [-1/2, 1/2]$, and $\tau= 2 \pi\, t\, \hbar /m l^2 \in\mathbb{R}$, it reads $$\vartheta(\xi,\tau)=\sqrt{\frac{2}{l}}\ \sum_{k=0}^{+\infty}c_{2k+1} \sin \left[2\pi\left(\xi+\frac{1}{2}\right) \left(k+\frac{1}{2}\right) \right]\exp\left[-\ii \pi \tau \left(k+\frac{1}{2}\right)^2 \right] .$$ By writing the sine as the sum of exponentials and by making use of the expression (\[eq:cn\]), we finally get $$\vartheta(\xi,\tau)=\sum_{n=-\infty}^{+\infty} d_n \e^{\ii 2 \pi \xi \left(n+\frac{1}{2}\right) -\ii \pi \tau \left(n+\frac{1}{2}\right)^2}, \qquad \xi\in \left[-\frac{1}{2}, \frac{1}{2}\right],
\label{eq:finalsum}$$ (notice that now the sum runs over all $n\in\mathbb{Z}$).
Derive Equation (\[eq:finalsum\]) and show that $$\label{dn}
d_n= \frac{1}{\pi \sqrt{l}} \frac{(-1)^n}{n+\frac{1}{2}}, \qquad n\in\mathbb{Z}.$$
If we take a closer look at the expression (\[eq:finalsum\]) we notice that it is a Fourier series with quadratic phases. This series is the boundary value of a Jacobi theta function [@Abra], which is defined in the lower complex half plane of $\tau$, and it has a very rich structure investigated at length by mathematicians. For a full immersion in its deep arithmetic properties see the charming “Tata Lectures on Theta” by Mumford [@theta]. A simple property is its quasi-periodicity in the rescaled time $\tau$ (check it!): $$\vartheta(\xi,\tau+1)= \e^{-\ii \pi/4}\, \vartheta(\xi,\tau).$$ Thus at integer times $\tau$ the wave function comes back (up to a phase) to its initial flat form (\[eq:v0\]): these are the *quantum revivals*. More generally, at rational values of $\tau$ the graph of $|\vartheta(\xi,\tau)|^2$ is piecewise constant and there is a partial reconstruction of the initial wave function [@berryklein], see Fig. \[fig:Fractals\]. On the other hand, at irrational times, the wave function is a fractal, with Hausdorff dimension $D_H=3/2$, as shown in the last panel of Fig. \[fig:Fractals\]. In fact, $\vartheta(\xi,\tau)$ can be proved to be a fractal function in space *and* time, and to form a beautifully intricate *quantum carpet*, with different Hausdorff dimensions along different space-time directions [@berryschleich].
Lecture 2: Moving walls! {#sec-MW}
========================
In this lecture we will try to answer the following question: What happens if the walls of our box start moving? Which equation will describe the quantum dynamics of the bouncing particle?
The classical version of this problem was introduced by Fermi [@Fermi] in 1949, and then investigated by Ulam [@Ulam]. It is convenient to parametrize the confining interval as $$I_{l,d}= \left[ d -\frac{l}{2}, d+ \frac{l}{2}\right],$$ so that $l>0$ is the width of the box and $d\in\mathbb{R}$ its center. We will suppose that $l$ and $d$ are regular functions of time, $t\mapsto l(t)$ and $t\mapsto d(t)$, with $l(t)>l_0>0$ so that the interval never shrinks to a point. From now on we will often omit the dependence on $t$.
As we did in the case of still walls, we analyze the dynamics described by the Schrödinger equation $$\ii \hbar \frac{\d }{\d t} \psi(t)=\frac{p^2}{2m}\psi(t),
\label{eq:Mov_SchrEq}$$ where the domain of $p^2$ is $$D_{l,d}=\left\{ \psi\in
\mathcal{H}^2\left(I_{l,d}\right), \; \psi\left(d-\frac{l}{2}\right)=\psi\left(d+\frac{l}{2}\right)=0\right\},
\label{eq:Dld}$$ (Dirichlet’s boundary conditions). Notice that this domain depends on time, so that at different times we work on different spaces. This means that the time derivative, $$\frac{\d }{\d t} \psi(t)=\lim_{\epsilon\to 0} \frac{\psi(t+\epsilon)-\psi(t)}{\epsilon},$$ involves the sum of vectors belonging to different Hilbert spaces, since in general $D_{l(t),d(t)}\neq D_{l(t+\epsilon),d(t+\epsilon)}$. Therefore, we need to take more care in the formulation of the problem and in the interpretation of Eq. (\[eq:Mov\_SchrEq\]).
The correct formulation of the problem can be accomplished by embedding the space of square integrable functions on the interval, $L^2(I_{l,d})$, in the larger Hilbert space $L^2(\mathbb{R})$ on the real line: $$L^2(\mathbb{R})=L^2(I_{l,d})\oplus L^2(I_{l,d}^c),
\label{eq:L2sum}$$ where $X^c=\mathbb{R}\setminus X$ denotes the complement of the set $X$. Thus, every wave function $\psi\in L^2(\mathbb{R})$ can be written as a sum $\psi=\chi +\phi$, where $\chi\in L^2(I_{l,d})$ and $\phi\in L^2(I_{l,d}^c)$. Following this strategy the Hamiltonian of the system, representing the kinetic energy of the particle in the box, in the containing space reads: $$H_0(l,d)=\frac{p_{l,d}^2}{2m}= -\frac{\hbar^2}{2 m} \frac{\mathrm{d}^2}{\mathrm{d} x^2}\oplus_{l,d} 0,
\label{Ham}$$ equipped with the Dirichlet boundary conditions. Notice that the choice of the Hamiltonian to be 0 in the component $L^2(I_{l,d}^c)$, i.e. outside the box, is completely arbitrary and immaterial, since we will be interested in the dynamics inside the box, for a particle with initial (and evolved) wave function with $\phi=0$. Moreover, it is worth noticing that even if the Hamiltonian in Eq. (\[eq:Mov\_SchrEq\]) appeared to be time-independent, the direct sum decomposition in (\[Ham\]), in the case of moving walls, makes its time dependence clear. Indeed, the direct sum in Eq. (\[eq:L2sum\]) depends on time through the functions $l$ and $d$. In other words, the restriction to the interval somehow concealed the time-dependence of the Hamiltonian, which becomes evident once one enlarges the space. Therefore, the Schrödinger equation we are dealing with has a time-dependent Hamiltonian on a time-dependent domain.
Reduce and Conquer
------------------
The attack strategy that we will follow is to find a unitarily equivalent problem where the Schrödinger operators act on a *common fixed domain* of self-adjointness. Let us consider the transformation $$\qquad U(l,d):L^2(\mathbb{R}) \rightarrow L^2(\mathbb{R}),
\label{eq:Uld}$$ that acts as $$(U(l,d)\psi)(\xi)=\sqrt{\frac{l}{l_0}}\, \psi\left(\frac{l}{l_0} \xi+d\right),$$ where $l_0>0$ is a reference length.
Prove that the transformation $U(l,d)$ is unitary and maps the interval $I_{l,d}$ onto the reference interval $$I=I_{l_0,0} = \left[-\frac{l_0}{2},\frac{l_0}{2}\right].$$
In fact, $U(l,d)$ can be rewritten as a composition of two transformations $$U(l,d) = \mathcal{D}\left(\ln \frac{l}{l_0}\right)^\dag \mathcal{T}(d)^\dag = \mathcal{D}\left(-\ln \frac{l}{l_0}\right) \mathcal{T}(-d),
\label{eq:decomp}$$ where $$(\mathcal{T}(d) \psi) (x) = \psi (x-d), \qquad (\mathcal{D}(s) \psi ) (x) = \e^{-s/2} \psi(\e^{-s} x),
\label{eq:decomp1}$$ are two (strongly continuous) one-parameter unitary groups implementing the translations and dilations on $L^2(\mathbb{R})$, see Fig. \[fig:transf\].
Prove the decomposition (\[eq:decomp\]).
![The effect of the transformation on the box can be described as the composition of a translation and a dilation.[]{data-label="fig:transf"}](transf.pdf)
Now notice that the domain (\[eq:Dld\]) is mapped onto the fixed domain $D=U(l,d) D_{l,d}$ $$D=\left\{
\phi\in \mathcal{H}^2(I),\;
\phi\left(-\frac{l_0}{2}\right)=\phi\left(\frac{l_0}{2}\right)=0\right\}\subset L^2(I),
\label{eq:Dirichlet}$$ describing the Dirichlet boundary conditions in a fixed box $I$, and the Hamiltonian (\[Ham\]) is mapped into $$H(l) = U H_0 U^{\dagger}= \left(\frac{l_0}{l}\right)^2
\frac{p^2}{2m} \oplus 0 =-\left(\frac{l_0}{l}\right)^2 \frac{\hbar^2}{2 m}
\frac{\mathrm{d}^2}{\mathrm{d} x^2}\oplus 0.
\label{eq:H(l,d)}$$
Prove that the unitary transformation $U(l,d)$ yields constant boundary conditions only when the latter do not mix the wave function with its derivative.
Henceforth we will denote the wave functions in the frame with moving and fixed walls by $\psi(x)$ with $x\in I_{l,d}$ and by $\phi(\xi)$ with $\xi\in I$, respectively. By deriving the relation $$\phi(t)=U(l(t),d(t))\psi(t),$$ we get the Schrödinger equation in the fixed reference frame from the one in the frame with moving walls: $$\begin{aligned}
\nonumber
\ii\hbar\frac{\d}{\d t}\phi&=&\ii \hbar\left(U(l,d)\dot{\psi}+\dot{U}(l,d)\psi\right)\\
\nonumber&=&\left(U(l,d)H_0(l,d)+\ii\hbar\dot{U}(l,d)\right)\psi\\
&=&\left(U(l,d)H_0(l,d)U^\dagger(l,d)+\ii\hbar\ \dot{U}(l,d)U^\dagger(l,d)\right)U(l,d)\psi.\end{aligned}$$ Notice that now the Schrödinger operator contains not only the transformed Hamiltonian $H(l)$, but also an additional geometrical term $$K(l,d)=\ii\hbar \dot{U}(l,d)U^\dagger(l,d).
\label{eq:K(l,d)}$$ Let us compute this geometric contribution step by step. First of all the action of $\dot{U} = \d U(l(t),d(t))/\d t$ on a test function $\psi$: $$\begin{aligned}
\left(\frac{\d U}{\d t} \psi \right)(\xi)&=&\frac{\d}{\d
t} \left(\sqrt{\frac{l}{l_0}} \, \psi \left(\frac{l}{l_0}\xi+d\right) \right)
\nonumber\\
&=& \frac{\dot{l}}{2\sqrt{l_0 l}}
\psi \left(\frac{l}{l_0}\xi+d\right) +\sqrt{\frac{l}{l_0}}\left(\frac{\dot{l}}{l_0}\xi+\dot{d}\right) \psi' \left(\frac{l}{l_0}\xi+d\right).\end{aligned}$$ Then, since $$(U^\dagger(l,d) \phi)(x) = (\mathcal{T}(d) \mathcal{D}(\ln l/l_0) \phi)(x) = \sqrt{\frac{l_0}{l}}\,
\phi\left(\frac{l_0}{l}(x-d)\right),$$ we have $$\ii \hbar \frac{\d U}{\d t} U^\dagger \phi (\xi) = \ii
\frac{\hbar}{2}\, \frac{\dot{l}}{l}\, \phi (\xi) + \ii \hbar \left( \frac{\dot{l}}{l}\xi+\frac{l_0 }{l} \dot{d}\right) \phi'
(\xi),$$ that is $$\ii \hbar \frac{\d U}{\d t} U^\dagger = - \frac{\dot{l}}{l} \left(x p- \ii
\frac{\hbar}{2}\right) - \frac{l_0 }{l} \dot{d}\, p,$$ with $x$ and $p$ the position and momentum operators. Thus, the geometric generator of the unitary transformation reads $$K(l,d)= \ii \hbar\, \frac{\d U}{\d t} U^\dagger= - \frac{\dot{l}}{l} x \circ p-\frac{l_0 }{l} \dot{d}\, p,
\label{eq:transport}$$ where $A\circ B = (AB+BA)/2$ is the symmetrized (Jordan) product of the operators $A$ and $B$, and the canonical commutation relation $[x, p]=\ii\hbar$ has been used.
Show that the generator of the translation group $\mathcal{T}(d)$ in (\[eq:decomp1\]) is the momentum operator $p$, while the generator of the dilation group $\mathcal{D}(s)$ is the virial operator $x\circ p$. As a consequence, by using the decomposition (\[eq:decomp\]) show that $$U(l,d)= \exp\left(\frac{\ii}{\hbar} \ln \left(\frac{l}{l_0 }\right) x \circ p \right)\, \exp\left(\frac{\ii}{\hbar} d\, p \right),$$ which yields (\[eq:transport\]).
Summing up, we finally come to the Schrödinger equation in the reference frame with fixed walls $$\label{eq:StaticH_SchrEq}
\ii \hbar \frac{\d }{\d t}\phi=
\Big(H(l)+K(l,d) \Big)\phi
=\left(\frac{1}{l^2}\frac{p^2}{2m
}-\frac{\dot{l}}{l} x \circ p-\frac{l_0}{l} \dot{d}\, p\right)\phi\,.$$
Let us now make some considerations about this equation and its solution. We need a definition and a theorem [@Kato].
Let $T$ and $A$ be two operators on a Hilbert space $\mathcal{H}$, such that $D(T)\subset D(A)$. $A$ is said to be *relatively bounded* with respect to $T$ or simply *$T$-bounded*, if there exist two non-negative constants $a$ and $b$ such that: $$\|A \psi \| \leq a \|\psi\|+b \|T \psi\|, \qquad \forall\ \psi\in D(T).$$ Moreover, the infimum of the possible values of $b$ is called the [*$T$-bound*]{} of $A$.
[(Kato-Rellich)]{} Let $T$ be self-adjoint and bounded from below. If $A$ is symmetric and $T$-bounded with $T$-bound smaller than $1$, then $T+A$ is self-adjoint and bounded from below.
Show that $K(l,d)$ is relatively bounded with respect to $H(l)$ with $0$ relative bound.
Applying the Kato-Rellich theorem it is easy to prove that the total Hamiltonian $H(l) + K(l,d)$ with domain $D(H(l)+K(l,d))= D(H(l))=D$ is self-adjoint. Therefore, the Schrödinger equation in (\[eq:StaticH\_SchrEq\]) is well defined for any initial condition $\phi(0)\in D$, and yields a unique unitary propagator. The proof is a corollary of Theorem X.70 in [@Reed], since $t\mapsto H(l(t)) + K(l(t),d(t))$ is a one-parameter family of Schrödinger operators on a common domain of self-adjointness $D$ for any pair of differentiable functions $d(t)$ and $l(t)$, with $l(t) > l_0$, for some $l_0>0$.
Prove that the energy rate equation of the particle is given by $$\begin{aligned}
\frac{\d}{\d t}{\langle\phi(t)\vert H(l(t)) \phi(t)\rangle} &=& -\frac{\hbar^2}{2 m} \left(\frac{l_0}{l(t)}\right)^3 \left[\left( \frac{\dot{l}(t)}{l_0} \xi +\dot{d}(t) \right)
|\phi'(\xi,t)|^2 \right]_{-\frac{l_0}{2}}^{\frac{l_0}{2}} ,
\label{exp_val}\end{aligned}$$ where $\phi'(\xi,t)= \partial_\xi \phi(\xi,t)$. *Hint:* The energy rate can be computed as $$\dot{E}(t)=\frac{\d }{\d t}{\langle\phi\vert H(l)
\phi\rangle}=\frac{\ii}{\hbar}\big({\langle\phi \vert K H\phi\rangle}- {\langleK H
\phi\vert \phi \rangle}\big)+{\langle\phi \vert \dot{H} \phi\rangle},
\label{eq:ev_exp_val}$$ where $K(l,d)$ is the operator in (\[eq:transport\]). Pay attention to the domains.
[\[Rigid translation\]]{} Find the Schrödinger operator in the moving reference frame in the case of a translation without dilation of the walls: $$\dot{l}=0, \qquad d(t)=d_0+vt.$$
An Example: The Accelerating Box
--------------------------------
Let us consider the case in which the box is moving with a constant acceleration $g$, without dilating: $$\dot{l}=0, \qquad d(t)=d_0-\frac{1}{2}gt^2.$$ Since the particle is confined in a accelerating rigid box this is the problem of a particle in a rocket, see Fig. \[fig:elev\]. If we assume $l=l_0$ we find: $$H=\frac{p^2}{2m}+gtp,$$ with Dirichlet boundary conditions.
![The particle in an accelerating box.[]{data-label="fig:elev"}](Missile.pdf)
We can now apply the gauge (unitary) transformation $$G(t): L^2(I) \rightarrow
L^2(I) : \phi \mapsto \chi =
G(t)\phi\,,$$ with $$\chi(\xi) = (G(t)\phi)(\xi)=\e^{\frac{\ii}{\hbar}
\left(mg\xi t-\frac{1}{6}mg^2t^3\right)}\phi (\xi)\,,$$ in such a way that the Hamiltonian becomes $$G(t)H G^\dagger(t)=\frac{p^2}{2m}-\frac{1}{2}mg^2t^2.$$ Therefore, the Schrödinger equation reads $$\ii \hbar \frac{\d}{\d t} \chi(t)=\left(\frac{p^2}{2m
}- mgx\right)\chi(t),$$ describing a particle in a constant gravitational field, in agreement with the equivalence principle.
The eigenfunctions of the operator $\frac{p^2}{2m }-mgx$ belonging to the eigenvalue $E$ are linear combinations of the two Airy functions [@Abra]:
$$\phi(x)=c_1 \mbox{Ai}\left(\frac{x}{l}-\frac{E}{mgl}\right)+c_2 \mbox{Bi}\left(\frac{x}{l}-\frac{E}{mgl}\right).$$
Prove that the permitted values of energy are the solutions of: $$\mbox{Ai}\left(-\frac{1}{2}-\frac{E}{mgl}\right)\mbox{Bi}\left(\frac{1}{2}-\frac{E}{mgl}\right) - \mbox{Ai}\left(\frac{1}{2}-\frac{E}{mgl}\right)\mbox{Bi}\left(-\frac{1}{2}-\frac{E}{mgl}\right)=0,
\label{eq:Airy}$$ and in particular that the first four are ($\epsilon=\frac{E}{mgl}$), see Fig. \[fig:Airy\]:
$\epsilon_1$ $9.86851$
-------------- -----------
$\epsilon_2$ $39.4787$
$\epsilon_3$ $88.8266$
$\epsilon_4$ $157.914$
![The zeros of the function (\[eq:Airy\]) yield the permitted energy levels.[]{data-label="fig:Airy"}](Airy.pdf)
Lecture 3: Changing boundary conditions {#sec-change}
=======================================
In this lecture we will consider a somewhat different situation: we will assume that the walls of the box are fixed, so that $I=[a,b]$ is given, but the way the walls interact with the quantum particle changes in time due to some physical mechanism. That means in our model that the boundary conditions change in time. More generally, the situation we have in mind is that depicted in Fig. \[fig:bend\]: a ring with a junction whose properties are time dependent, so that all possible boundary conditions (\[eq:bc2\]) can in principle be implemented. The evolution of the particle will be given by a Schrödinger equation with a time-dependent Hamiltonian: $$\ii\hbar \frac{\d}{\d t} \psi(t) = T_{U(t)} \psi(t),
\label{eq:ScUt}$$ with $\psi(0)=\psi_0$. Here $T_{U(t)}= p^2/2m$ on $\mathcal{H}^2(I)\subset L^2(I)$ with boundary conditions (\[eq:bc2\]), with $U=U(t)$. Thus the Hamiltonian depends on time through its boundary conditions $U(t)$.
This model can be implemented in a superconducting quantum interference device (SQUID) with a tunable junction, obtained by replacing the junction with an additional flux loop [@Vion; @Cosmelli; @Mooij]. This can be an experimental realization of a continuous change among different topologies [@Wilczek].
We will not study this problem in full generality, but instead we will consider the particular case of boundary conditions rapidly alternating between two values $U$ and $V$ (for example they periodically alternate between Dirichlet and Neumann): $$U(t)=
\begin{cases}
V,& 2k \tau \leq t < (2k+1) \tau, \\
U,& (2k+1) \tau \leq t < (2k+2) \tau,
\end{cases}$$ with $k=0,1,2,\dots$, and $\tau$ the time period.
In this scenario, since the Hamiltonian is constant during the period among the changes, the Schrödinger equation (\[eq:ScUt\]) can be explicitly integrated (the trick is revealed: the reason for the funny time-dependence of $U(t)$ is just this!) and at time $2 N \tau$ we end up with $$\begin{aligned}
\label{eq:2evol}
\psi(2N\tau)&=& \underbrace{\left(\e^{-\ii \tau T_{U}/\hbar} \e^{-\ii \tau T_{V}/\hbar}\right)
\left(\e^{-\ii \tau T_{U}/\hbar} \e^{-\ii \tau T_{V}/\hbar}\right) \dots
\left(\e^{-\ii \tau T_{U}/\hbar} \e^{-\ii \tau T_{V}/\hbar}\right)}_{N \; \mathrm{ times}} \psi_0
\nonumber \\
&=& \quad \left(\e^{-\ii \tau T_{U}/\hbar} \e^{-\ii \tau T_{V}/\hbar}\right)^N \psi_0.\end{aligned}$$ We want to study the behavior of the dynamics in the limit $N\to \infty$, when the time interval between the switches $\tau=t/N$ goes to zero, the number of switches goes to infinite, while the total time of the evolution $2 N\tau= 2t$ is kept constant. Therefore, we are led to the study of the limit of the product formula: $$\label{eq:evollim?}
\lim_{N\to\infty} \left(\e^{-\ii t T_{U}/N \hbar} \e^{-\ii t T_{V}/N\hbar}\right)^N .$$
Prove that, given two matrices $A$ and $B$ the *Lie product formula* holds $$\left( \e^{- \ii A/N} \e^{-\ii B/N}\right)^N \to \e^{-\ii C},
\label{eq:Trotter}$$ with $C=A+B$.
*Hint*: use the telescopic equation $$D^N - E^N=
\sum_{k=0}^{N-1} D^k (D-E) E^{N-1-k},$$ with $D= \e^{- \ii A/N} \e^{-\ii B/N}$ and $E= \e^{-\ii (A+B)/N}$, together with the estimates $$D-E=\e^{- \ii A/N} \e^{-\ii B/N}- \e^{-\ii (A+B)/N} = -\frac{1}{2N^2} [A,B] +O\left(\frac{1}{N^3}\right).$$
Trotter in [@Trotter] proved that Lie’s product formula (\[eq:Trotter\]) is still valid when applied to (unbounded) self-adjoint operators $A$ and $B$, such that their sum $C=A+B$ is still (essentially) self-adjoint on the intersection of their domains, $D(C)= D(A)\cap D(B)$.
Well, this seems to be our case: $t T_U/\hbar$ and $t T_V/\hbar$ are unbounded self-adjoint operators, so the limit of the Lie-Trotter product formula (\[eq:evollim?\]) should be $\exp(-\ii (T_U+T_V)t/\hbar)$. Now we get $$T_{U}+T_{V} = \frac{p^2}{2 m} + \frac{p^2}{2 m} = 2 \frac{p^2}{2m},$$ on the common domain $D(T_U)\cap D(T_V)$. Therefore, $$\label{eq:evollim}
\left(\e^{-\ii t T_{U}/\hbar N} \e^{-\ii t T_{V}/\hbar N}\right)^N \to
\e^{-\ii 2 t T_{W}/\hbar},$$ when $N\to\infty$, and we arrive at the intelligible result that rapidly alternating free evolutions with two boundary conditions $U$ and $V$ yield again a free evolution, possibly with different boundary conditions $W$.
Unfortunately, life is not so easy: Trotter’s theorem does not hold, because the intersection of the the domains $D(T_W)=D(T_U)\cap D(T_V)$ is too small, being defined by too many constraints (those of $U$ and those of $V$). For example, let us consider Dirichlet, $U=-\I$, and Neumann, $V=\I$, boundary conditions. In this case $D(T_{-\I})\cap D(T_{\I})=\{\psi\, | \, \Psi=\Psi'=0\}$ and in this domain the kinetic operator $p^2/2m$ is symmetric but not self-adjoint! In other words, our common sense tells us that we should get $p^2/2m$ with some boundary conditions $W$, but we do not know, even heuristically, which ones, because the boundary conditions $U$ and $V$ fight each other and yield too many constraints!
Math again comes to our aid: The right objects to look at are not the operators $T_U$ and $T_V$, but the *quadratic forms* associated to them, that is their expectation values $t_U(\psi)={\langle\psi\vert T_U \psi\rangle}$ and $t_V(\psi)={\langle\psi\vert T_V \psi\rangle}$, measuring the kinetic energy of the particle in state $\psi$. The crucial fact is that the quadratic form $a(\psi) = {\langle\psi\vert A \psi\rangle}$ associated to an unbounded operator $A$ makes sense on a domain $D(a)$ which is larger than the domain of $A$, namely $D(A)\subset D(a)$. Thus the sum of two quadratic forms $a(\psi)+b(\psi)$ is well defined on wave functions $\psi\in D(a)\cap D(b)$ for which the operator sum $A+B$ is not defined, since $\psi\notin D(A)\cap D(B)$.
The extension of Trotter’s theorem to quadratic forms reads as follows [@Kato-Trotter; @Lapidus]:
[(Lapidus)]{} Let $a$ and $b$ be the quadratic forms associated to $A$ and $B$. If $A$ and $B$ are self-adjoint and bounded below, and $D=D(a)\cap D(b)$ is dense, then Trotter’s formula (\[eq:Trotter\]) holds with $$C=A\dot{+}B,
\label{eq:Kato1}$$ the unique operator associated to the quadratic form $a+b$. The operator $C$ is called the *form sum* of $A$ and $B$.
In our case, we will see that the domains of the quadratic forms of the kinetic energy $t_U(\psi)$ involve *only* the boundary values (\[eq:bc21\]) of the wave function $\Psi$ and *not* the boundary values of its derivative $\Psi'$. This will imply that the sum of the kinetic energies $$t_W(\psi)= \frac{t_U(\psi)+ t_V(\psi)}{2}
\label{eq:sumquad}$$ is always the quadratic form associated to a self-adjoint operator $T_W$, namely $$t_W(\psi)= {\langle\psi\vert T_W \psi\rangle}, \qquad \psi\in D(T_W).$$ For example, in the case of Dirichlet and Neumann we will see that $D(t_{-\I})= \{\psi\, | \, \Psi=0\}$ and $D(t_{\I})= \{\psi\, | \, \text{no bound. conds.}\}$, so that $D(t_{-\I})\cap D(t_{\I})=\{\psi\, | \, \Psi=0\}=D(t_{-\I})$: alternating Dirichlet and Neumann give Dirichlet.
Summing up, we get that the limit (\[eq:evollim\]) holds with $$T_{W}=\frac{T_U\ \dot{+}\ T_V}{2},
\label{eq:recipe}$$ the form sum of $T_U$ and $T_V$.
From operators to quadratic forms {#sec-qforms}
---------------------------------
For any $\psi \in \mathcal{H}^2(I)$, one integration by parts yields $$\begin{aligned}
{\langle \psi\vert \frac{p^2}{2m} \psi \rangle} &=& -\frac{\hbar^2}{2m}\int_a^b \overline{\psi(x)} \psi''(x)\, \d x
\nonumber\\
&=& \frac{\hbar^2}{2m}\int_a^b \overline{\psi'(x)} \psi'(x)\, \d x
- \frac{\hbar^2}{2m} \left(\overline{\psi(b)} \psi'(b) - \overline{\psi(a)} \psi'(a)\right)
\nonumber\\
&=& \frac{\hbar^2}{2m}\left( \| \psi' \|^2 - \langle \Psi | \Psi' \rangle \right),
\label{eq:tU0}\end{aligned}$$ where $\Psi$ and $\Psi'$ are the two-dimensional vectors of boundary data defined in (\[eq:bc21\]).
Prove that Eq. (\[eq:tU0\]) holds.
Notice that, at variance with our starting point, the last line of (\[eq:tU0\]) involves only the first derivative of $\psi$, and thus makes sense on a larger domain. Moreover, by using the boundary conditions (\[eq:bc2\])-(\[eq:bc21\]) we will trade the boundary values of the derivative for the boundary values of the function in (\[eq:tU0\]), obtaining the following expression for the quadratic form associated to $T_U$: $$t_U(\psi) = \frac{\hbar^2}{2m}\left( \| \psi' \|^2 + \Gamma_U(\Psi) \right),
\label{eq:quadraticform}$$ with $\Gamma_U(\Psi)$ a quadratic form of the boundary vector $\Psi$.
The form $t_U(\psi)$ is well defined for any integrable function with square integrable *first* (distribution) derivative, $\mathcal{H}^1(I)=\{\psi\in L^2(I)\ |\ p\psi\in L^2(I)\}$ (the first Sobolev space). In order to get the explicit expression of $\Gamma_U$ and the precise domain $D(t_U)$ we have to distinguish among three possibilities according to the number of eigenvalues $u_1$ and $u_2$ of $U$ equal to $-1$.
1. \[case:K\] If both the eigenvalues of $U$ are different from $-1$, such as in the case of the Neumann boundary conditions, then $(\I+U)$ is invertible, and the boundary values of the derivative can be expressed in terms of the boundary values of the function: $$\Psi' = -\frac{\ii}{l_0} \frac{\I-U}{\I + U} \Psi.
\label{eq:Phi'Phi}$$ Therefore we get $$\begin{aligned}
\Gamma_U (\Psi) = \frac{\ii}{l_0} \langle\Psi | \frac{\I-U}{\I + U} \Psi \rangle,
\label{eq:GammaU1}\end{aligned}$$ and $D(t_U) = \mathcal{H}^1 (I)$, with no constraints on the boundary values $\Psi$.
2. \[case:-1\] If $U$ has only one eigenvalue equal to $-1$, namely $u_1=-1$ and $u_2\neq -1$, then, if we call $\xi$ the normalized eigenvector corresponding to the eigenvalue $-1$, and $\xi^\perp$ its orthogonal, from (\[eq:bc2\]) we get $$\langle \xi | \Psi \rangle = 0, \qquad \langle \xi^\perp| \Psi' \rangle =
-\frac{\ii}{l_0} \frac{1-u_2}{1+u_2}\, \langle \xi^\perp| \Psi \rangle .
\label{eq:perp0}$$ Therefore, $${\left\langle \xi\right\vert} \Psi\rangle = 0, \qquad \Gamma_U(\Psi)= \frac{\ii}{l_0} \frac{1-u_2}{1+u_2}\, |\langle \xi^\perp| \Psi \rangle|^2 .
\label{eq:perp1}$$
Prove Equations (\[eq:perp0\]) and (\[eq:perp1\]).
3. \[case:Dirichlet\] Finally, if $u_1=u_2=-1$, i.e. in the case of the Dirichlet boundary conditions, then $U=-\I$, so that $$\Psi= 0, \qquad \Gamma_{-\I}(\Psi) = 0 .$$
Composition law of boundary conditions
--------------------------------------
Now we can use the quadratic form (\[eq:quadraticform\]) to evaluate the limit of the alternating dynamics (\[eq:evollim\]), according to the recipe (\[eq:recipe\]). By cooking our equations we will prove that the boundary conditions $W$ are given by the composition law: $$\label{eq:2evollim}
W= U\star V = V\star U,$$ where $\star$ is a commutative and associative product on the boundary unitary operators.
The evaluation of the emergent boundary condition $W$ in (\[eq:2evollim\]) requires the computation of the sum of the kinetic energies (\[eq:sumquad\]) and the evaluation of the domain $$D(t_W) = D(t_U) \cap D(t_V).$$ Again, we distinguish various cases according to the number of eigenvalues $-1$:
1. When both $U$ and $V$ have no eigenvalues equal to $-1$, we get $$\Gamma_W = (\Gamma_U + \Gamma_V)/2,$$ with no constraints on the wave-function boundary values $\Psi$. By using the expression (\[eq:GammaU1\]) we obtain (prove it!) $$W = U \star V = \frac{\I-\frac{1}{2}\left(\frac{\I-U}{\I+U} + \frac{\I-V}{\I+V} \right)}
{\I+\frac{1}{2}\left(\frac{\I-U}{\I+U} + \frac{\I-V}{\I+V} \right)} .
\label{eq:regular}$$
2. If $-1$ is a nondegenerate eigenvalue of $U$, and $V$ has no eigenvalues equal to $-1$, then $D(t_W)= D(t_U)$, with the only constraint ${\langle\xi\vert \Psi\rangle} = 0$. Therefore, the boundary forms $\Gamma_U$ and $\Gamma_V$ are nonzero and add up only on the orthogonal subspace, spanned by $\xi^\perp$. It is easy to see that $$W= U\star V = - {\left\vert \xi\right\rangle} {\left\langle \xi\right\vert} + w_2 {\left\vert \xi^\perp\right\rangle} {\left\langle \xi^\perp\right\vert},
\label{eq:-1w2}$$ where $w_2$ is a function of $u_2$, $\xi$, and $V$.
Derive Eq. (\[eq:-1w2\]) and find the explicit expression of $w_2$.
3. If $-1$ is a nondegenerate eigenvalue of both $U$ and $V$, that is $u_1=v_1 = -1$ and $u_2, v_2 \neq -1$, then there are two possibilities:
1. If the eigenvectors of $U$ and $V$ belonging to $-1$ are parallel, that is $U$ commutes with $V$, then $D(t_W)=D(t_U)=D(t_V)$. Thus, the only constraint is ${\left\langle \xi\right\vert}\Psi\rangle=0$ and $W$ has the previous form (\[eq:-1w2\]).
Find how $w_2$ particularizes in this case to a function of $u_2$ and $v_2$.
2. If the eigenvectors $\xi$ of $U$ and $\eta$ of $V$ belonging to $-1$ are not parallel, then they span the whole space. The constraints ${\langle\xi\vert \Psi\rangle}=0$ and ${\langle\eta\vert \Psi\rangle}=0$ imply Dirichlet’s boundary conditions $\Psi=0$, so that $D(t_W)=D(t_{-\I})$ and $$W= U\star V = -\I .
\label{eq:indipUV}$$
4. Finally, in the case $U=-\I$ (or $V=-\I$) then $D(t_W)= D(t_{-\I})$, so that $\Psi=0$ and $$W = (-\I) \star U = U \star (-\I) = -\I.
\label{eq:attractor}$$
Summing up, the emerging boundary conditions are given by the composition law (\[eq:2evollim\]), where the product $\star$ is given by the Cayley transform (\[eq:regular\]) for $U$ and $V$ with no eigenvalues $-1$ (i.e. free ends $\Psi$). On the other hand, eigenspaces with eigenvalues $-1$ are absorbing for the product, that is all constraints on the wave-function boundary values $\Psi$ are inherited by $W$. In particular the Dirichlet boundary conditions $-\I$ play the role of an attractor (\[eq:attractor\]), and when $U$ and $V$ have independent constraints on $\Psi$, then their composition (\[eq:indipUV\]) is Dirichlet, $\Psi=0$.
Conclusions {#sec-conc}
===========
In these lectures we have seen that sometimes examples that are apparently very simple may conceal an unexpected rich structure. And in fact they can help us to build and understand general schemes. Our goal was to communicate the mathematical and the physical ideas in the most simple setting, without all the unnecessary technical complications that a more realistic model inevitably has. We hope to have hit our goal, at least partially. Moreover, we hope that the references we have provided will stimulate the interest of the reader to go beyond this notes and to get directly in touch with a so active field. For instance, the generalization to higher dimensions is nontrivial and introduce new savory ingredients, because in such a case the Hilbert space of the boundary is infinite dimensional. But that’s another story.
Acknowledgments {#acknowledgments .unnumbered}
===============
We thank Manolo Asorey, Andrzej Kossakowski, Beppe Marmo, Ninni Messina, Daniele Militello, and Saverio Pascazio for stimulating discussions. We also thank Giuseppe Florio and Giancarlo Garnero for interesting suggestions and for carefully reading the manuscript. P. F. would like to thank the local organizers, A. P. Balachandran and Sachin Vaidya, for their kindness in inviting him and for the effort they exerted on the organization of the workshop. This work was partially supported by the Italian National Group of Mathematical Physics (GNFM-INdAM), and by PRIN 2010LLKJBX on “Collective quantum phenomena: from strongly correlated systems to quantum simulators”.
[99]{}
K. Yajima, *Commun. Math. Phys.* **110** (1987), 415.
K. Yajima, *Commun. Math. Phys.* **181** (1996), 605.
G. F. Dell’Antonio, R. Figari and A. Teta, *CMS Conf. Proc. Amer. Math. Soc.* **28**, (2000) 99.
A. Posilicano, *Proc. Amer. Math. Soc.* **135**, (2007) 1785.
N. C. Dias, A. Posilicano and J. N. Prata,*CPAA* **10**, (2011) 1687.
S. Haroche and J. M. Raimond, *Exploring the quantum* (Oxford University Press, 2006).
A. Mody, M. Haggerty, J. M. Doyle and E. J. Heller, *Phys. Rev B* **64**, (2001) 085418.
W. Paul, *Rev. Mod. Phys.* **62**, (1990) 531.
J. R. Friedman, V. Patel, W. Chen, S. K. Tolpygo and J. E. Lukens, *Nature* **406**, (2000) 43.
C. M. Wilson, T. Duty, M. Sandberg, F. Persson, V. Shumeiko and P. Delsing, *Phys. Rev. Lett* [**105**]{} (2010) 233907.
M. Asorey, A. P. Balachandran, G. Marmo, I. P. Costa e Silva, A. R. de Queiroz, P. Teotonio-Sobrinho and S. Vaidya, arXiv:1211.6882 \[hep-th\]
A. D. Shapere, F. Wilczek and Z. Xiong, arXiv:1210.3545 \[hep-th\].
M. V. Berry, *J. Phys. A: Math. Gen.* **29**, (1996) 6617.
S. Di Martino, F. Anzà, P. Facchi, A. Kossakowski, G. Marmo, A. Messina, B. Militello and S. Pascazio, *J. Phys. A: Math. Theor.* **46**, (2013) 365301.
M. Asorey, P. Facchi, G. Marmo and S. Pascazio, *J. Phys. A: Math. Theor.* **46**, (2013) 102001.
T. Kato, [*Perturbation theory for linear operators*]{} (Springer-Verlag, Berlin, 1966).
M. Asorey, A. Ibort and G. Marmo, *Int. J. Mod. Phys. A* [**20**]{}, (2005) 1001.
M. Abramowitz and I. A. Stegun, [*Handbook of Mathematical Functions*]{} (Dover, New York, 1972).
D. Mumford, *Tata Lectures on Theta* (Birkhäuser, Boston, 1983).
M. V. Berry and S. Klein *J. Mod. Opt.* **43**, (1996) 2139.
M. V. Berry, I. Marzoli and W. Schleich *Physics World* (June 2001) 39.
E. Fermi, *Phys. Rev.* **75** (1949) 1169.
S. M. Ulam, *Proc. Fourth Berkeley Symp. on Math. Statist. and Prob.* Vol. **3** (1961) 315 (Univ. of California Press).
M. Reed and B. Simon, *Methods of Modern Mathematical Physics, Vol. 2: Fourier Analysis, Self Adjointness* (Academic Press, London, 1975).
D. Vion, A. Aassime, A. Cottet, P. Joyez, H. Pothier, C. Urbina, D. Esteve and M.H. Devoret, *Science* **296**, (2002) 886.
S. Poletto, F. Chiarello, M.G. Castellano, J. Lisenfeld, A. Lukashenko, C. Cosmelli, G. Torrioli, P. Carelli and A.V. Ustinov, *New J. Phys.* **11**, (2009) 013009.
F. G. Paauw, A. Fedorov, C. J. P. M Harmans and J. E. Mooij, *Phys. Rev. Lett.* **102**, (2009) 090501.
H. F. Trotter, *Proc. Amer. Math. Soc.* **10**, (1959) 545.
T. Kato, *Topics in functional analysis*, Adv. in Math. Suppl. Stud. **3**, 185 (Academic Press, New York, 1978).
G. W. Johnson and M. L. Lapidus, *The Feynman integral and Feynman’s operational calculus* (Oxford University Press, New York, 2000).
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We have derived an X-ray luminosity function using parallax-based distance measurements of a set of 12 dwarf novae, consisting of [*Suzaku*]{}, [*XMM-Newton*]{} and [*ASCA*]{} observations. The shape of the X-ray luminosity function obtained is the most accurate to date, and the luminosities of our sample are concentrated between $\sim$ 10$^{30}$–10$^{31}$ erg s$^{-1}$, lower than previous measurements of X-ray luminosity functions of dwarf novae. Based on the integrated X-ray luminosity function, the sample becomes more incomplete below $\sim$ 3 $\times$ 10$^{30}$ erg s$^{-1}$ than it is above this luminosity limit, and the sample is dominated by X-ray bright dwarf novae. The total integrated luminosity within a radius of 200 pc is 1.48 $\times$ 10$^{32}$ erg s$^{-1}$ over the luminosity range of 1 $\times$ 10$^{28}$ erg s$^{-1}$ and the maximum luminosity of the sample (1.50 $\times$ 10$^{32}$ erg s$^{-1}$). The total absolute lower limit for the normalised luminosity per solar mass is 1.81 $\times$ 10$^{26}$ erg s$^{-1}$ M$_{\odot}^{-1}$ which accounts for $\sim$ 16 per cent of the total X-ray emissivity of CVs as estimated by @saz06.'
author:
- |
K. Byckling$^{1}$[^1] K. Mukai$^{2}$, J.R. Thorstensen$^{3}$ and J. P. Osborne$^{1}$\
$^{1}$Department of Physics and Astronomy, University of Leicester, University Road, Leicester, LE1 7RH, UK\
$^{2}$NASA/Goddard Space Flight Center, Greenbelt, MD 20771, USA\
$^{3}$Department of Physics and Astronomy, 6127 Wilder Laboratory, Dartmouth College, Hanover, NH 03755-3528, USA
date: 'Accepted . Received ; in original form'
title: 'Deriving an X-ray luminosity function of dwarf novae based on parallax measurements'
---
\[firstpage\]
cataclysmic variables – stars: dwarf novae – X-rays: stars – X-rays: binaries – stars: distances – stars: luminosity function
Introduction
============
Cataclysmic variables, i.e. CVs consist of an accreting white dwarf primary and a late-type main sequence star, and accrete via Roche lobe overflow. CVs can be divided into several subclasses of which so called dwarf novae (DNe) are the most numerous subclass of CVs in our Galaxy. In these systems, the white dwarf has a weak magnetic field strength [B $\lesssim$ 10$^{6}$ G, @van96] compared to magnetic CVs, such as polars, and thus the formation of an accretion disc is possible. From time to time, the disc is seen to brighten by several magnitudes lasting from days to several weeks. This brightening of the disc, i.e. an outburst, is thought to be caused by disc instabilities, which are described in detail in @las01. In quiescence, DNe are sources of optical emission emanating from the accretion disc and the bright spot where the material from the secondary hits the edge of the disc. Quiescent optical spectra of DNe are characterized by strong Balmer emission lines and weaker He I lines with some heavier elements. Also, DNe are sources of hard X-rays which are thought to originate from an optically thin boundary layer during quiescence. However, during an outburst hard X-rays are quenched as the boundary layer becomes optically thick and thus a source of soft X-rays and EUV emission [@pri77; @pri79].
At the time of the discovery of the Galactic Ridge X-ray Emission (GRXE) in 1982 [@wor82], discrete point sources were thought to be the origin of the GRXE emission. However, the origin has been debated since, but observational evidence gathered to date since the GRXE discovery supports the view that the GRXE is not due to diffuse origin but due to discrete point sources, such as CVs and other accreting binary systems [see recent studies by e.g. @rev07; @rev08]. More supporting evidence was given by the recent [*Chandra*]{} study carried out by @rev09 who resolved over 80 per cent of the GRXE into point sources in the 6–7 keV energy range during an ultra-deep 1 Msec observation.
Based on [*EXOSAT*]{} observations of the X-ray emission in the Galactic Plane, @war85 concluded that if the GRXE is assumed to be originating from discrete point sources, the bulk of the emission observed must be due to a population of low luminosity X-ray sources (L$_{x}$ $<$ 10$^{33.5}$ erg s$^{-1}$), such as CVs. Subsequently, @muk93 suggested that DNe could significantly contribute to the GRXE based on their study of an [*EXOSAT*]{} Medium Energy (ME) DN sample. According to this study, the space density of DNe is sufficiently high to account for a significant fraction of the GRXE. Later on, @ebi01 resolved sources down to 3 $\times$ 10$^{-15}$ erg cm$^{-2}$ s$^{-1}$ in their [*Chandra*]{} observation of the Galactic Ridge, equivalent to $L_{x}$ $>$ 2.3 $\times$ 10$^{31}$ erg s$^{-1}$ at 8 kpc, concluding that the number of resolved point sources above this level is insufficient for them to be the major contributor to the GRXE. Since these previous works have not completely resolved the contribution of CVs to the GRXE, further studies are needed. As was noted by @muk93, unbiased and sensitive surveys with accurate distance measurements of CVs are needed. This way, accurate X-ray luminosity functions (XLFs) can be obtained, and the contribution to the GRXE estimated more precisely.
The motivation for our work was mainly given by the inaccuracies in the XLFs of Galactic CV populations, such as those by @bas05 and @saz06. Baskill et al. derived an XLF using 34 [*ASCA*]{} observations of non-magnetic CVs (including 23 DN observations). Their sample lacked accurate distance measurements as only 10 sources had parallax-based distance measurements. Furthermore, this sample was biased by high X-ray flux sources since [*ASCA*]{} was intended to be a spectroscopic mission and the sources in the studied sample were known to be X-ray bright. Also, the [*ASCA*]{} study was purely archival without any sample selection (e.g. the distance was not limited) as Baskill et al. chose all non-magnetic CV observations in the archive, they did not filter out sources which were in an outburst state, or restrict the study to one type of objects only. The XLF study by @saz06 focused on building up an XLF in the 2–10 keV range combining the [*RXTE*]{} Slew Survey (XSS) and [*ROSAT*]{} All-Sky Survey (RASS) observations of active binaries, CVs and young main sequence stars in the luminosity range $\sim$ 10$^{27.5}$ $<$ $L_{x}$ $<$ 10$^{34}$ erg s$^{-1}$. However, uncertainties in the luminosities in this study were introduced by inadequate accuracies in the distances, for example, many of the intermediate polars (IPs) in their sample had poorly known distances. Only a few sources had parallax measurements from, e.g., the Hipparcos or Tycho catalogues (astrometric uncertainties $\sim$ 1 mas) and ground-based parallax measurements from @tho03. In addition, the RASS luminosities had 50 per cent uncertainties in addition to statistical errors after conversion from the 0.1–2.4 keV to the 2–10 keV range.
The primary aim of this paper is to derive the most accurate shape of the XLF to date by using a carefully selected sample of DNe. In order to achieve this, we aim to minimise the biases seen in other published XLFs by using sample selection criteria described in Section \[sourcesample\]. One of the criteria worth mentioning here is that we only use parallax-based distance estimates. The source sample used in this study does not represent a complete sample of DNe within 200 pc: the sample is more of a ”fair sample” which was not chosen based on the X-ray properties of the sources, and which represents typical DNe within the solar neighbourhood. We will consider the possible effects of the optical selection, inherent in the parallax sample, on the XLF which we derive in Section 6 (Discussion). The motivation for choosing a group of DNe as the sample is based on observations of the Galactic CV populations and previous CV population models. Various authors, such as @how97, @pre07a and @gan09, have pointed out that binaries which are brighter in X-rays and show frequent outbursts, may not represent the true majority of Galactic CV population. The discovery methods of CVs usually favour brighter systems, and thus fainter CVs with low mass accretion rates are likely to be under-represented. However, population models of CVs predict that the majority of CVs are short-period systems ($P_{orb}$ $<$ 2.5 h) and X-ray faint [e.g. @kol93; @how97]. These models are supported by observational evidence, e.g. @pat84 showed that a sample of CVs with a total space density of 6 $\times$ 10$^{-6}$ pc$^{-3}$ was dominated by low mass accretion rate, and thus short period, systems. Also, the SDSS study by @gan09 showed that orbital periods of intrinsically faint Galactic CVs accumulated in the 80–86 min range; they found that 20 out of 30 SDSS CVs in this period range showed characteristics which implied that they are low mass accretion rate WZ Sge type DNe. As has been shown by these studies, less X-ray luminous objects (such as DNe) dominated the studied volumes, and thus we choose to focus on DNe in this paper. It is also worth mentioning the study by @pre07b who carried out the [*ROSAT*]{} North Ecliptic Pole (NEP) survey using a purely X-ray flux limited and a complete sample of 442 X-ray sources above a flux limit of $\sim$ 10$^{-14}$ erg cm$^{-2}$ s$^{-1}$ in the 0.5–2.0 keV band (only five systems were CVs). They concluded that if the overall space density of CVs is as high as 2 $\times$ 10$^{-4}$ pc$^{-3}$, then the dominant CV population must be fainter than 2 $\times$ 10$^{29}$ erg s$^{-1}$.
We have carried out X-ray spectral analysis of our sample of 13 sources and derived an XLF in the 2–10 keV band for 12 of them with reliable distance measurements based on those of by @har04, @tho03 and @tho08. By using sources with accurate distance measurements, we minimise the error on the luminosity. Also, we have carried out timing analysis for 5 sources in the sample which were recently observed by [*Suzaku*]{}. At the time of writing this paper, the Z Cam type star KT Per went into an outburst in January 2009, and thus we also briefly report on the [*Suzaku*]{} observations of KT Per during the outburst in Section \[sec:ktper\].
The selection criteria and the source sample {#sourcesample}
============================================
Since we wanted to obtain accurate luminosities for the sources (and thus an accurate shape for the luminosity function), the first step was to avoid selecting sources randomly from the archive [see e.g. @bas05] or selecting an X-ray flux limited source sample. The aim was to have a [*distance-limited*]{} sample. Thus, sources were not selected based on their X-ray properties, but we chose only those DNe which have accurately measured distances based on trigonometric parallax measurements within $\sim$ 200 pc. Note that by using all available distance measurement techniques, Patterson (priv. comm.) estimates that currently there are 13 DNe within 100 pc, and $\sim$ 33 DNe within 200 pc from the Sun, of which the latter count is clearly incomplete. Above the 200 pc limit, ground-based parallax technique does not give accurate and reliable distance measurements. By using trigonometric parallax-based distance measurements, we are more likely to avoid biases in the distance measurements which are present in the previous, published X-ray luminosity functions. Due to the lack of ground-based parallax measurement programme for the Southern hemisphere, our sample is limited to northern and equatorial objects. However, this selection should not introduce any biases in terms of the optical or X-ray luminosities in our sample.
The distance measurements of the sources chosen for this work are based on astrometric parallaxes obtained by the [*Hubble Space Telescope*]{} (HST) Fine Guidance Sensors (FGSs) [@har04], and trigonometric parallaxes obtained by the ground-based 2.4 m Hiltner Telescope at the MDM Observatory on Kitt Peak, Arizona [@tho03; @tho08] and Thorstensen (in prep.). The first accurate astrometric parallaxes of DNe (SS Cyg, SS Aur and U Gem) were measured in 1999 using the FGSs which can deliver high-precision parallaxes with sub-milliarcsecond uncertainties [@har99]. Trigonometric parallaxes derived by ground-based observations have uncertainties around 1 mas (= 10$^{-3}$ arcsec) or less [@tho03; @tho08] which is almost as good as the uncertainty on the FGS parallax measurements.
The second selection criterion was to restrict the sample to sources which had been observed by X-ray imaging telescopes with CCDs in the energy range 0.2–10 keV. Once we had obtained a list of targets with parallax measurements, we then looked for archival data of pointed imaging X-ray observations of these targets in the energy range $\sim$ 0.2–10 keV. If the chosen targets did not have previous X-ray imaging observations, we requested [*Suzaku*]{} X-ray observations. Finally, we wanted to constrain the sample to those sources which were in their quiescent states during the observations in order to avoid biases in the luminosities, and thus [*AAVSO*]{}[^2] light curves of the selected sources were inspected to confirm that the sources were in quiescence during the X-ray observations.
The final source sample consists of 9 SU UMa (including 2 WZ Sge systems), 3 U Gem and 1 Z Cam type DNe. The main characteristic which separates these classes of DNe is the outburst behaviour: U Gem type DNe outburst mainly in timescales of every few weeks to every few months whereas SU UMa stars show normal, U Gem type DN outbursts and, in addition, superoutbursts with superhumps (variations in the light curves at a period of a few per cent longer than the orbital period) in timescales of several months to years. The extreme cases, WZ Sge stars, only have superoutbursts with outburst timescales of decades without normal DN outbursts. The defining characteristic for Z Cam stars is standstills, i.e., it is possible that after an outburst, they do not return to the minimum magnitude (unlike U Gem stars), but remain between the minimum and maximum magnitudes for 10–40 days.
------------- -------------------- -------------- ----------------- ---------------------------------- -------
Source Inclination P$_{orb}$ M$_{WD}$ Distance Type
deg h M$_{\odot}$ pc
SS Cyg 40 $\pm$ 8 $^{c}$ 6.603 $^{c}$ 1.19 $^{f}$ 165$^{+13}_{-11}$ $^{c}$ UG
V893 Sco 71 $\pm$ 5 $^{a}$ 1.82 $^{a}$ 0.89 $^{d}$ 155$^{+58}_{-34}$ $^{a}$ SU
SW UMa 45 $\pm$ 18 $^{g}$ 1.36 $^{b}$ 0.80 $^{e}$ 164$^{+22}_{-19}$ $^{b}$ SU
VY Aqr 63 $\pm$ 13 $^{a}$ 1.51 $^{a}$ 0.8/0.55 $^{e}$ 97$^{+15}_{-12}$ $^{a}$ SU
SS Aur 40 $\pm$ 7 $^{c}$ 4.39 $^{a}$ 1.03 $^{e}$ 167$^{+10}_{-9}$ $^{c}$ UG
BZ UMa 60–75 $^{e}$ 1.63 $^{b}$ 0.55 $^{e}$ 228$^{+63}_{-43}$ $^{b}$ SU
U Gem 69 $\pm$ 2 $^{c}$ 4.246 $^{c}$ 1.03 $^{e}$ 100 $\pm$ 4 $^{c}$ UG
T Leo 47 $\pm$ 19 $^{a}$ 1.42 $^{a}$ 0.35 $^{e}$ 101$^{+13}_{-11}$ $^{a}$ SU
WZ Sge 76 $\pm$ 6 $^{c}$ 1.36 $^{a}$ 0.90 $^{e}$ 43.5 $\pm$ 0.3 $^{c}$ SU/WZ
HT Cas 81 $\pm$ 1 $^{h}$ 1.77 $^{b}$ 0.8 $^{e}$ 131$^{+22}_{-17}$ $^{b}$ SU
GW Lib 11 $\pm$ 10 $^{a}$ 1.28 $^{a}$ 0.8 $^{e}$ 104$^{+30}_{-20}$ $^{a}$ SU/WZ
Z Cam$^{*}$ 65 $\pm$ 10 $^{a}$ 6.98 $^{a}$ 1.21 $^{e}$ 163$^{+68}_{-38}$ $^{a}$ ZC
ASAS J0025 – 1.37 $^{j}$ – $\sim$ 175$^{+120}_{-40}$ $^{i}$ SU
\[sources\]
------------- -------------------- -------------- ----------------- ---------------------------------- -------
The sources, which were included in the calculation of the X-ray luminosity function and when testing different correlations discussed later in this paper, were observed with [*Suzaku*]{} (BZ UMa, SW UMa, VY Aqr, SS Cyg, SS Aur, V893 Sco, and ASAS J002511+1217.2), [*XMM-Newton*]{} (U Gem, T Leo, HT Cas and GW Lib) and with [*ASCA*]{} (WZ Sge). [*Suzaku*]{} observations of BZ UMa, SW UMa, VY Aqr, SS Aur, V893 Sco and ASAS J0025 were requested as these observations were not in the archive. @muk09 discuss the [*Suzaku*]{} observations of V893 Sco in more detail. We also included Z Cam in the source sample since it has a parallax measurement, but it appeared to be in a transition state during the observations. Thus, we have only reported the results of the spectral analysis for Z Cam, but excluded it when calculating the X-ray luminosity function, and when testing correlations between different parameters. The system parameters for all the 13 sources are given in Table \[sources\].
Observations and data reduction
===============================
The details of the [*Suzaku*]{}, [*XMM*]{}, and [*ASCA*]{} observations are given in Table \[observations\], and the data reduction methods are described in the following sections.
------------------ ------------ -------------- ------------- ------------ ----------- -------
Source ObsID Instrument T$_{start}$ T$_{stop}$ T$_{exp}$ State
ks
SS Cyg 400006010 XIS/Suzaku 2005-11-02 2005-11-02 39 Q
V893 Sco 401041010 XIS/Suzaku 2006-08-26 2006-08-27 18 Q
SW UMa 402044010 XIS/Suzaku 2007-11-06 2007-11-06 17 Q
VY Aqr 402043010 XIS/Suzaku 2007-11-10 2007-11-11 25 Q
SS Aur 402045010 XIS/Suzaku 2008-03-04 2008-03-05 19 Q
BZ UMa 402046010 XIS/Suzaku 2008-03-24 2008-03-25 30 Q
ASAS 403039010 XIS/Suzaku 2009-01-10 2009-01-11 33 Q
J0025
KT Per 403041010 XIS/Suzaku 2009-01-12 2009-01-13 29 OB
U Gem 0110070401 MOS1/XMM 2002-04-13 2002-04-13 23(22.4) Q
0110070401 MOS2/XMM 2002-04-13 2002-04-13 23(22.4) Q
T Leo 0111970701 PN/XMM 2002-06-01 2002-06-01 13(13) Q
HT Cas 0111310101 PN/XMM 2002-08-20 2002-08-20 50(6.9) Q
GW Lib 0303180101 PN/XMM 2005-08-25 2005-08-26 22(6.7) Q
WZ Sge 34006000 GIS,SIS/ASCA 1996-05-15 1996-05-15 85 Q
Z Cam 35011000 GIS,SIS/ASCA 1997-04-12 1997-04-12 41 T
\[observations\]
------------------ ------------ -------------- ------------- ------------ ----------- -------
Suzaku data reduction
---------------------
[*Suzaku*]{} [@mit07], originally [*Astro-E2*]{}, was launched in 2005 and is Japan’s 5th X-ray astronomy mission. In this paper, we will focus on the X-ray Imaging Spectrometer (XIS) data. The XIS consists of four sensors: XIS0,1,2,3 of which three (XIS0,2,3) contain front-illuminated (FI) CCDs, and XIS1 contains a back-illuminated (BI) CCD. The XIS0,2,3 are less sensitive to soft X-rays than XIS1 due to the thin Si and SiO$_{2}$ layers on the front side of the XIS0,2,3 CCDs. Since November 9, 2006, the XIS2 unit has not been available for observations. The [*Suzaku*]{} background is low and hardly affected by soft proton flares often seen in [*XMM*]{} observations.
The unfiltered event lists of SS Cyg and V893 Sco were reprocessed with <span style="font-variant:small-caps;">xispi</span> and screened in <span style="font-variant:small-caps;">xselect</span> with <span style="font-variant:small-caps;">xisrepro</span> since the pipeline version for these observations was older than v.2.1.6.15 which does not include correction for the time- and energy-dependent effects in energy scale calibration. For all the other [*Suzaku*]{} observations, the observations had been processed by more recent pipeline versions and thus reprocessing was not necessary. Pile-up was not a problem for our data since the source count rates were safely below the pile-up limit (12 ct s$^{-1}$) for point sources observed in the ’Normal’ mode using Full Window[^3]. The [*Suzaku*]{} data reduction described below was carried out in a similar manner for all the [*Suzaku*]{} observations. The cleaned event lists were read into <span style="font-variant:small-caps;">xselect</span> in which X-ray spectra were extracted for each source. Light curves were extracted for SW UMa, BZ UMa, SS Aur, ASAS J0025, and VY Aqr for timing analysis studies. To include 99 per cent of the flux and to obtain the most accurate flux calibration, the spectra and light curves were extracted using a source radius of 260” (250 pixels). The backgrounds were taken as an annulus centred on the source excluding the inner 4’ source region. The outer radii of the background annuli were determined according to how close the calibration sources were to the target. The response matrix files (RMFs) and ancillary response files (ARFs) were created and combined within <span style="font-variant:small-caps;">xisresp</span> v.1.10. The XIS0,2,3 source and background spectra and the corresponding response files were summed in <span style="font-variant:small-caps;">addascaspec</span> to create the total XIS0,2,3 source and background spectra, and the total XIS0,2,3 response file. For the light curves, the background areas were scaled to match the source areas, and the scaled background light curves were then subtracted from the source light curves in <span style="font-variant:small-caps;">lcmath</span>.
XMM data reduction
------------------
[*XMM-Newton*]{} [@jan01] is the cornerstone mission of the European Space Agency (ESA). It has been operating since 1999 with three X-ray cameras (EPIC pn, MOS1 and MOS2), the Optical Monitor (OM), and the Reflection Grating Spectrometer (RGS) onboard. The X-ray cameras cover the energy range 0.2–12.0 keV.
The data were obtained from the [*XMM-Newton*]{} Science Archive (XSA) and the [*XMM-Newton*]{} data were reduced and analysed in the standard manner using the [*XMM-Newton*]{} Science Analysis System <span style="font-variant:small-caps;">sas</span> version 8.0.0. Each observation was checked for high background flares in the range 10–12 keV using single pixel events (<span style="font-variant:small-caps;">pattern == 0</span>). The high background flares were cut above 0.35 ct s$^{-1}$ for the MOS data and above 0.40 ct s$^{-1}$ for the pn data. The source and background extraction regions were taken from circular extraction areas avoiding any contaminating background sources. The radii of the source regions were calculated by using the <span style="font-variant:small-caps;">sas</span> task <span style="font-variant:small-caps;">region</span> in order to derive source extraction radii which include $\sim$ 90 per cent of the source flux for each source. The background extraction region (r$_{bg}$ = 130 arcsec) was taken from the same chip as the source extraction region, or from an adjacent chip in case of a crowded source chip. When extracting the X-ray spectra, only well-calibrated X-ray events were selected for all the sources, i.e. for the pn spectra single and double pixel events with <span style="font-variant:small-caps;">pattern</span> $\leq$ 4 were chosen, and in order to reject bad pixels and events close to CCD gaps, <span style="font-variant:small-caps;">FLAG</span> == 0 was used. For the MOS, <span style="font-variant:small-caps;">pattern $\leq$ 12</span> and \#XMMEA\_EM were applied.
For most of the observations, we analysed the pn observations only. Since the total effective area of the two EPIC MOS cameras is nearly equal to the effective area of the EPIC pn, the MOS spectra would not add any significant information to the pn spectra. The only exception was U Gem pn observation which had been obtained in Small Window mode. Thus, we used the MOS1 and 2 data which had been obtained in Large and Small window modes, respectively, and selected the backgrounds from the surrounding CCDs. To form the total MOS spectrum for U Gem, the U Gem MOS1 and 2 spectra were summed in <span style="font-variant:small-caps;">addascaspec</span>. All the observations were checked in case of pile-up by using the [*XMM*]{} <span style="font-variant:small-caps;">sas</span> task <span style="font-variant:small-caps;">epatplot</span>. Pile-up did not occur in any of the observations, but the source PSFs of HT Cas and T Leo were contaminated by out-of-time (OoT) events, introduced by these two sources. Therefore, the background regions in the HT Cas and T Leo observations were taken from the adjacent chip in order to avoid the OoT events. These OoT events were removed from the source X-ray spectra according to the ’<span style="font-variant:small-caps;">sas</span> threads’[^4], i.e. the source spectra extracted from the OoT event lists were subtracted from the source spectra extracted from the original event list.
ASCA data reduction
-------------------
The [*Advanced Satellite for Cosmology and Astrophysics*]{} [[*ASCA*]{}, @tan94] was Japan’s fourth cosmic X-ray astronomy mission operating between February 1993 and July 2000 and was the first X-ray observatory which carried CCD cameras. The main science goal of [*ASCA*]{} was the X-ray spectroscopy of astrophysical plasmas. It carried four X-ray telescopes with two types of detectors located inside them: two CCD cameras, i.e. the Solid-state Imaging Spectrometers (SIS0 and SIS1) with spectral resolution of 2 per cent at 5.9 keV at launch, and two scintillation proportional counters, i.e. the Gas Imaging Spectrometers (GIS2 and GIS3).
The [*ASCA*]{} data reduction was performed in the standard manner by mostly using the standard screening values for the GIS and SIS instruments as described in NASA [*ASCA*]{} online manual[^5]. For both instruments, intervals outside the South Atlantic Anomaly (SAA) were chosen, also including intervals when the attitude control was stable with the upper limit of the angular distance from the target set to <span style="font-variant:small-caps;">ang\_dist</span> $<$ 0.02 degrees. For the SIS instruments, the bright earth angle of <span style="font-variant:small-caps;">br\_earth</span> $>$ 10 was applied excluding the data taken below the 10$^{\circ}$ angle. Times of high background were excluded when the PIXL monitor count rate was 3$\sigma$ above the mean of the observation. Also, the background monitor count rate of <span style="font-variant:small-caps;">rbm\_cont</span> $<$ 500 was applied (the standard screening value is <span style="font-variant:small-caps;">rbm\_cont</span> $<$ 100). Events which occurred before the first Day-Night transition and at least 32 seconds after the Day-Night transition, and also before the passage of the South Atlantic Anomaly (SAA) and at least 32 s after the SAA, were selected (<span style="font-variant:small-caps;">T\_DY\_NT < 0 $\parallel$ T\_DY\_NT > 32</span> && <span style="font-variant:small-caps;">T\_SAA < 0 $\parallel$ T\_SAA > 32</span>).
The source extraction regions for the SIS and GIS were centred on the source. For the GIS, a $\sim$ 6 arcmin circular source extraction region was used for both Z Cam and WZ Sge, while the SIS source extraction regions were smaller so that they could be safely fitted within the chip. Thus, the source extraction radii for Z Cam were 4.4 arcmin (SIS0) and $\sim$ 3.5 arcmin (SIS1) and for WZ Sge $\sim$ 4 arcmin (SIS0) and $\sim$ 3 arcmin (SIS1). The background extraction region for the GIS was taken centred on the detector excluding a circular region of $\sim$ 8 arcmin centred on the source. The background extraction radii for GIS2 and GIS3 were 15.7 and 15.5 arcmin (Z Cam), and 15.4 and 13.8 arcmin (WZ Sge) respectively. For the SIS background, blank-sky background observations were used for Z Cam due to lack of space for a local background region on the CCDs. For WZ Sge, the total area of the two active CCDs excluding a 5.5 arcmin region around the source was used as the background extraction region.
The ancillary response files (ARFs) for the GIS and SIS spectra were created with <span style="font-variant:small-caps;">ascaarf</span> and the SIS response matrix files (RMFs) with <span style="font-variant:small-caps;">sisrmg</span>. Finally, the total SIS and GIS X-ray spectra were created by combining SIS0 and SIS1, and the GIS2 and GIS3 spectra in <span style="font-variant:small-caps;">addascaspec</span>, respectively.
Timing analysis
===============
Since VY Aqr, SS Aur, BZ UMa, SW UMa and ASAS J0025 have not been subject to previous, pointed, imaging X-ray observations before the [*Suzaku*]{} observations, we looked for periodicities from the data of these sources. KT Per has been observed by the [*Einstein*]{} Observatory [@cor84], but no previous X-ray spectral or timing analysis studies have been carried out for it. In order to ensure that these objects are not intermediate polars (IPs) and to look for orbital and spin modulation in the data, the power spectra were calculated by using a Lomb-Scargle periodogram [@sca82] which is used for period analysis of unevenly spaced data. When searching over the frequency range 0.00001–0.03 Hz, no significant periodicities were seen at the 99 per cent confidence level.
Spectral analysis
=================
We carried out X-ray spectral analysis in order to study the underlying spectra of the source sample, and, ultimately, to calculate the fluxes and luminosities of the sources. To employ Gaussian statistics, the X-ray spectra were binned at 20 ct bin$^{-1}$ with <span style="font-variant:small-caps;">grppha</span> and then fitted in <span style="font-variant:small-caps;">Xspec11</span> [@arn96].
In CVs, the power source of X-ray emission is known to be accretion onto the white dwarf. The accreted material is shock-heated to high temperatures [$kT_{max}$ $\sim$ 10–50 keV, @muk03], and this material has to cool before settling onto the white dwarf surface. Thus, the cooling gas flow is assumed to consist of a range of temperatures which vary from the hot shock temperature $kT_{max}$ to the temperature of the optically thin cooling material which eventually settles onto the surface of the white dwarf [@muk97]. Thus, when fitting X-ray spectra of CVs, cooling flow spectral models should represent more physically correct picture of the cooling plasma, unlike single temperature spectral models. Cooling flow models have successfully been applied to CV spectra in previous studies by e.g. @whe96 and @muk03. In this view, the multi-temperature characteristic is our motivation for emphasizing the cooling flow model in the rest of this work. The differential emission measure dEM/dT for an isobaric cooling flow can be described by [@pan05]
$$\label{eq:cool}
\frac{dEM}{dT} = \frac{5k\dot{M}n^{2}}{2\mu m_{p} \epsilon(T,n)},$$
where $m_{p}$ is the mass of a proton, $\mu$ the mean molecular weight ($\sim$ 0.6), $\epsilon$(T,n) total emissivity per volume in units of erg s$^{-1}$ cm$^{-3}$, $\dot{M}$ accretion rate, $n$ particle density, and $k$ the Boltzmann constant. The source of the X-ray emission above the white dwarf illuminates the surface of the white dwarf and thus causes a reflection, which is seen as Fe K$\alpha$ iron fluorescence line at 6.4 keV [@geo91]. According to George & Fabian, an infinite slab reflector subtending a total solid angle of $\Omega$ = 2$\pi$ where the X-ray source is located right above the slab, produces an equivalent width of up to $\sim$ 150 eV for the 6.4 keV Fe K$\alpha$ fluorescence line. The equivalent width of the 6.4 keV iron line depends on the total abundance of the reflector [@don97], the inclination angle between the surface of the reflector and the observer’s line of sight, and the photon index of the spectrum of the X-ray emission source [@ish09].
[ccc]{}\
Name & EW(mekal) & EW(mkcflow)\
& eV & eV\
BZ UMa & 67$^{+42}_{-42}$ & $<$ 79\
\
HT Cas & $<$ 81 & $<$ 91\
\
SS Aur & 73$^{+37}_{-36}$ & 86$^{+52}_{-53}$\
\
SW UMa & 201$^{+124}_{-124}$ & $<$ 141\
\
U Gem & 50$^{+25}_{-24}$ & 60$^{+33}_{-32}$\
\
T Leo & 71$^{+38}_{-38}$ & $<$ 73\
\
V893 Sco & 46$^{+12}_{-11}$ & 45$^{+11}_{-12}$\
\
VY Aqr & $<$ 156 & $<$ 157\
\
WZ Sge & $<$ 140 & $<$ 76\
\
SS Cyg & 75$^{+9}_{-4}$ & 73$^{+6}_{-7}$\
\
Z Cam & 120$^{+42}_{-42}$ & 164$^{+42}_{-43}$\
\
ASAS J0025 & $<$ 220 & $<$ 200\
\[ev\]
Even though we believe that the cooling flow -type multi-temperature model is the correct description of the physics of the cooling gas flow in CVs, previous works have often used single temperature plasma models. Thus, in order to compare the effects of two different spectral models on the spectral fit parameters, we fitted the spectra with 1) a single temperature optically thin thermal plasma model [`mekal,` @mew86; @lie95] and 2) a cooling flow model (`mkcflow`) which was originally developed to describe the cooling flows in clusters of galaxies [@mus88], adding photoelectric absorption [`wabs`, @mor83] to both models. In order to investigate the equivalent width of the 6.4 keV iron emission line, a Gaussian line was added at 6.4 keV with a line width fixed at $\sigma$ = 10 eV. The spectral fits did not necessarily require the 6.4 keV line, e.g., for SS Aur the $\chi^{2}_{\nu}$/$\nu$ = 0.96/629 when a Gaussian line at 6.4 keV was not included.
[cccccccc]{}\
Name & $n_{H_{1}}$ & $n_{H_{2}}$ & CFrac & kT & Ab & $\chi^{2}_{\nu}$/$\nu$ & P$_{null}$\
& 10$^{20}$ cm$^{-2}$ & 10$^{20}$ cm$^{-2}$ & & keV & $Z_{\odot}$ & &\
BZ UMa & $<$0.19 &–&–& 4.17$^{+0.16}_{-0.17}$&0.51$^{+0.07}_{-0.08}$ & 1.02/905 & 0.349\
\
GW Lib &3.16$^{+8.58}_{-3.16}$&–&–&1.62$^{+1.92}_{-0.68}$&0.20$^{+1.23}_{-0.20}$& 1.02/8 & 0.414\
\
HT Cas &–&15.36$^{+2.13}_{-1.08}$ & 0.95$^{+0.04}_{-0.04}$ & 6.43$^{+0.60}_{-0.64}$ & 0.71$^{+0.18}_{-0.17}$ & 1.07/259 & 0.207\
\
SS Aur & $<$0.56 &–&–&6.35$^{+0.40}_{-0.40}$&1.0$^{+0.14}_{-0.15}$ & 1.06/628 & 0.127\
\
SW UMa & $<$0.20 &–&–&2.77$^{+0.12}_{-0.13}$&0.20$^{+0.07}_{-0.05}$ & 1.23/470 & 5.36$\times$10$^{-4}$\
\
U Gem &0.89$^{+0.19}_{-0.20}$&–&–&0.78$^{+0.03}_{-0.01}$&1.05$^{+0.12}_{-0.09}$&1.34/401 & 5.80$\times$10$^{-6}$\
& & & &6.85$^{+0.22}_{-0.23}$& & &\
\
T Leo &1.09$^{+0.21}_{-0.21}$&–&–&3.55$^{+0.10}_{-0.11}$&0.50$^{+0.06}_{-0.06}$ & 1.40/631 & 8.36$\times$10$^{-11}$\
\
V893 Sco &–&80.89$^{+4.18}_{-3.87}$&0.86$^{+0.01}_{-0.01}$&7.99$^{+0.29}_{-0.27}$&0.76$^{+0.04}_{-0.05}$& 1.02/1936 & 0.245\
\
VY Aqr & $<$1.64 &–&–&5.06$^{+0.43}_{-0.50}$&0.66$^{+0.18}_{-0.17}$ & 0.90/445 & 0.942\
\
WZ Sge &8.97$^{+2.41}_{-1.92}$&–&–&4.88$^{+0.55}_{-0.54}$&0.33$^{+0.17}_{-0.19}$&0.84/409 & 0.993\
\
SS Cyg &2.98$^{+0.14}_{-0.25}$&–&–&10.44$^{+0.16}_{-0.17}$&0.51$^{+0.02}_{-0.01}$&1.24/2881 & 2.74$\times$10$^{-17}$\
\
Z Cam &28.21$^{+2.63}_{-2.62}$ &292.41$^{+99.42}_{-68.86}$&0.35$^{+0.04}_{-0.05}$&8.68$^{+0.84}_{-0.79}$& 1.0 & 1.10/769 & 0.03\
\
ASAS & $<$0.84 &–&–& 4.38$^{+0.61}_{-0.53}$&0.56$^{+0.29}_{-0.24}$ & 0.88/366 & 0.958\
J0025 & & & & & & &\
\[mekal\]
[cccccccc]{}\
Name & $n_{H_{1}}$ & $n_{H_{2}}$ & CFrac & kT$_{max}$ & Ab & $\chi^{2}_{\nu}$/$\nu$ & P$_{null}$\
& 10$^{20}$ cm$^{-2}$ & 10$^{20}$ cm$^{-2}$ & & keV & $Z_{\odot}$ & &\
BZ UMa & $<$0.87 &–&–& 13.71$^{+1.38}_{-0.81}$ & 0.57$^{+0.13}_{-0.07}$ & 0.88/904 & 0.994\
\
GW Lib &$<$ 3.76 &–&–&6.96$^{+8.79}_{-3.12}$&1.0& 0.60/8 & 0.782\
\
HT Cas &–&16.74$^{+4.05}_{-2.49}$ & 0.92$^{+0.05}_{-0.04}$ & 23.09$^{+4.15}_{-5.33}$ & 0.78$^{+0.27}_{-0.22}$ & 0.99/258 & 0.525\
\
SS Aur &3.30$^{+1.79}_{-1.51}$&–&–& 23.47$^{+4.01}_{-3.02}$ &1.0& 0.95/628 & 0.832\
\
SW UMa & $<$0.67 &–&–& 8.33$^{+0.62}_{-0.99}$ & 0.41$^{+0.08}_{-0.10}$ & 0.87/469 & 0.978\
\
U Gem &0.76$^{+0.28}_{-0.21}$&–&–&25.82$^{+1.98}_{-1.43}$&1.04$^{+0.13}_{-0.11}$& 1.23/402 & 1.1$\times$10$^{-3}$\
\
T Leo &0.68$^{+0.24}_{-0.21}$&–&–&10.97$^{+0.85}_{-0.67}$ & 0.50$^{+0.07}_{-0.07}$& 1.08/629 & 7.24$\times$10$^{-2}$\
\
V893 Sco &–&103.71$^{+3.98}_{-3.07}$&0.90$^{+0.01}_{-0.01}$&19.32$^{+1.29}_{-1.40}$&0.94$^{+0.05}_{-0.05}$& 0.94/1934 & 0.973\
\
VY Aqr &1.10$^{+3.15}_{-1.10}$&–&–& 16.47$^{+2.68}_{-2.22}$ & 0.69$^{+0.25}_{-0.20}$ & 0.86/444 & 0.984\
\
WZ Sge &11.58$^{+3.96}_{-3.06}$&–&–&13.31$^{+3.01}_{-3.16}$ & 0.23$^{+0.16}_{-0.13}$ & 0.83/408 & 0.996\
\
SS Cyg &2.84$^{+0.11}_{-0.11}$&–&–&41.99$^{+1.20}_{-0.76}$&0.61$^{+0.03}_{-0.02}$ & 1.19/2883 & 4.80$\times$10$^{-12}$\
\
Z Cam &31.92$^{+4.77}_{-5.00}$ & 180.28$^{+53.37}_{-35.14}$&0.47$^{+0.07}_{-0.06}$&25.76$^{+5.16}_{-2.39}$ & 1.0 & 1.08/768 & 0.06\
\
ASAS & $<2.67$ &–&–& 14.43$^{+4.36}_{-2.69}$ & 0.68$^{+0.44}_{-0.29}$ & 0.81/366 & 0.996\
J0025 & & & & & & &\
\[mkcflow\]
The [*Suzaku*]{} XIS1 and XIS0,2,3 spectra were fitted simultaneously for each source as well as the [*ASCA*]{} GIS and SIS spectra of Z Cam and WZ Sge with the models mentioned above. Some data sets required additional components to improve the fits. Three of the sources, HT Cas, V893 Sco and Z Cam, required partial covering absorption model, `pcfabs`, to reduce residuals in the low energy end (between $\sim$ 0.6–2 keV). To reduce residuals around 0.80 keV in the SS Cyg spectrum, we added a Gaussian line at 0.81 keV with a line width of 0.24 keV letting the line energy and width both to vary free. For U Gem, single absorbed optically thin thermal plasma model yielded a $\chi^{2}$/$\nu$ = 2.23/403. Since the fit was not statistically satisfactory, we added a second optically thin thermal plasma component to improve the fit and obtained $\chi^{2}$/$\nu$ = 1.34/401 which was good enough for our analysis.
In the spectral fitting, the parameters of the spectral models were tied between different instrument spectra but let to vary free, apart from the Gaussian line energy at 6.4 keV and the line width $\sigma$ which were fixed. In order to estimate the abundances, the abundance parameter of the models was let to vary free for most data sets. For those sources for which abundance was significantly higher than the solar value, it was fixed at 1.0. An example of a source with a super-solar abundance is Z Cam for which the obtained abundance was 1.46$^{+0.34}_{-0.19}$ $Z_{0}$ with the partial covering + photoelectric absorption combined with the cooling flow model when the abundance was let to vary free.
The measured equivalent widths of the 6.4 keV line for each source are given in Table \[ev\], and the results of the spectral fitting for the optically thin thermal plasma and the cooling flow models are given in Table \[mekal\] and \[mkcflow\], respectively. These results show that in general, better $\chi^{2}_{\nu}$/$\nu$ values are achieved with the cooling flow model. For example, the improvement with the cooling flow model was statistically significant for SW UMa and T Leo. Fig. \[spectra\] illustrates the X-ray spectra of the new [*Suzaku*]{} XIS observations, i.e. VY Aqr, SW UMa, BZ UMa, SS Aur, and ASAS J0025, which have been fitted with the cooling flow model absorbed by photoelectric absorption with an added 6.4 keV Gaussian line component. Most of the X-ray spectra show that the clearest, discrete emission feature seen in the spectra of our source sample is the iron Fe K$\alpha$ complex, except in GW Lib, for which the signal-to-noise at $\sim$ 6 keV is too low for a reliable measurement.
Absorption
----------
Since the studied sources are all within $\sim$ 200 pc, i.e., within the solar neighbourhood, the effect of interstellar absorption should be negligible. Thus, high measured absorption columns would mainly be due to intrinsic absorption, associated with the sources. For most of the sources, the measured absorption columns were typically of the order of a few $\times$ 10$^{20}$ cm$^{-2}$, or even lower (10$^{19}$ cm$^{-2}$) which indicate low intrinsic absorption.
The highest intrinsic absorption columns are found in V893 Sco, Z Cam and HT Cas when compared to the rest of the source sample. All these three sources have partial covering absorbers $n_{H_{2}}$ with values of the order of 10$^{21}$ – 10$^{22}$ cm$^{-2}$ depending on the model. In addition to the partial covering absorber, Z Cam also has a simple absorption component with the highest $n_{H_{1}}$ value, $n_{H_{1}}$ $\sim$ 3 $\times$ 10$^{21}$ cm$^{-2}$, within the source sample. Originally, V893 Sco was found to have high intrinsic absorption by @muk09, and has a partial X-ray eclipse, also discovered by their study. Also, according to the best-fit model of @bas01, Z Cam had large amounts of absorption with $n_{H}$ = 9 $\times$ 10$^{21}$ cm$^{-2}$ during the transition state. Baskill et al. suggested that this absorption was associated with a clumpy disc wind.
Temperatures
------------
The measured shock temperatures $kT_{max}$ seem to be correlated with the white dwarf masses (Fig. \[mass\_temp\]) as one would expect. In Fig. \[mass\_temp\] it has been assumed that the white dwarf mass of VY Aqr is 0.8 M$_{\odot}$ (see Table \[sources\]). ASAS J0025 is not included in Fig. \[mass\_temp\] since the mass estimate is currently unknown. SS Cyg appears to be located in the upper right corner due to its high-mass white dwarf and thus high shock temperature. The shock temperatures, $kT_{max}$, in Fig. \[mass\_temp\] have been derived from the spectral fits of the cooling flow model for each source. The blue dashed line in Fig. \[mass\_temp\] represents the theoretical shock temperatures for given white dwarf masses. The radii, $R_{*}$, of the given white dwarf masses, $M_{1}$, were calculated assuming the mass-radius relation for cold, non-rotating and non-relativistic helium white dwarfs [see @pri75]
$$R_{*} = 7.7 \times 10^{8} x^{0.3767 - 0.00605\, \log(x)}\quad(cm),$$
where $x$ = $\frac{1.44 M_{\odot}}{M_{1}}$ -1. Subsequently, the theoretical shock temperatures, $T_{shock}$, for non-magnetic CVs were calculated according to Eq. \[eq:tshock\] for optically thin gas [@fra02]
$$\label{eq:tshock}
T_{shock} = \frac{3}{16}\frac{GM_{1}\mu m_{H}}{kR_{*}},$$
where $m_{H}$ is the mass of a hydrogen atom, $\mu$ the mean molecular weight, and $k$ the Boltzmann constant. As it appears from Fig. \[mass\_temp\], sources with high shock temperatures and low luminosities are not seen. This is sensible since the X-ray luminosity is proportional to $kT_{max}$ and the mass accretion rate, i.e. the normalization of the cooling flow model (Eq. \[eq:cool\]), thus we would expect to see high shock temperatures and high luminosities. Also, due to this proportionality, we expect to see an anti-correlation between $\dot{M}$ and $kT_{max}$ which indeed is seen for example in SW UMa (Fig. \[2dim\]).
![Mass of the white dwarf versus the shock temperature, $kT_{max}$, of the source sample with 90 per cent uncertainties for $kT_{max}$. The light blue dashed line represents the theoretical shock temperatures for given white dwarf masses. The figure does not include ASAS J0025 since a mass estimate does not currently exist for this source.[]{data-label="mass_temp"}](fig6.eps){width="80mm"}
As it appears from Fig. \[mass\_temp\], the white dwarf mass obtained for T Leo by @urb06 is only 0.35 M$_{\odot}$. This mass estimate may be unreliable, since as @lem93 argue, a low white dwarf mass would not allow superhumps to develop. See also @pat05 who refer to previous superhump studies which have shown that the mass ratio $q$ = $M_{2}/M_{1}$ has a key role in producing superhumps where $q_{crit}$ $\sim$ 0.3, although this value has not been confirmed by observational evidence.
Abundances
----------
We found that for most of the objects in the sample the obtained abundances were sub-solar with both models. In general, the abundances seem to be dependent on the spectral model: abundances are slightly lower when the spectra are fitted with the optically thin thermal plasma model. This is due to the single temperature characteristic of the optically thin thermal plasma model, i.e. it is likely that the abundances are underestimated because the best-fit temperature usually converges close to the peak of the 6.7 keV He-like Fe K$\alpha$ emissivity, whereas the cooling flow model consists of a range of temperatures outside the peak [see @muk09].
X-ray fluxes and luminosities
-----------------------------
The 2–10 and bolometric 0.01–100 keV fluxes and luminosities which were derived using the cooling flow model are given in Table \[fluxes\]. This shows that most of the 2–10 keV X-ray luminosities are concentrated around 10$^{30}$ erg s$^{-1}$. This is also seen in Fig. \[histo\] which shows a histogram of the X-ray luminosities of our sample. Only one object, GW Lib, stands out with a very low luminosity (4 $\times$ 10$^{28}$ erg s$^{-1}$). The measured luminosity of GW Lib is consistent with the results obtained by @hil07. @byc09 showed that GW Lib was still an order of a magnitude brighter (L $\sim$ 10$^{30}$ erg s$^{-1}$) in X-rays during [*Swift*]{} observations two years after the 2007 outburst than in 2005. But since the optical magnitude had not reached the quiescence level (V $\sim$ 18) in 2009, we do not consider the [*Swift*]{} 2009 X-ray luminosity as the quiescent luminosity. Thus, the higher X-ray luminosity measured in the [*Swift*]{} data does not affect the results of this paper.
![A histogram showing the X-ray luminosities of the source sample in 2–10 keV.[]{data-label="histo"}](fig8.eps){width="80mm"}
[ccccc]{}\
Source & F(2–10 keV) & L(2–10 keV) & F(0.01–100 keV) & L(0.01–100 keV)\
& x 10$^{-12}$ erg cm$^{-2}$ s$^{-1}$ & $\times$ 10$^{30}$ erg s$^{-1}$ & x 10$^{-12}$ erg cm$^{-2}$ s$^{-1}$ & $\times$ 10$^{30}$ erg s$^{-1}$\
BZ UMa &2.4$^{+0.1}_{-0.2}$ & 14.9$^{+10.8}_{-5.9}$ & 5.8 & 36.5\
\
GW Lib &0.04$^{+0.03}_{-0.01}$ & 0.05$^{+0.10}_{-0.02}$ & 0.1 & 0.1\
\
HT Cas &2.9$^{+0.7}_{-0.4}$ & 6.1$^{+4.2}_{-2.2}$ & 7.2 & 14.9\
\
SS Aur &2.9$^{+0.3}_{-0.3}$ & 9.6$^{+2.3}_{-1.9}$ & 7.1 & 23.9\
\
SW UMa &1.5$^{+0.2}_{-0.1}$ & 4.9$^{+2.2}_{-1.3}$ & 4.2 & 13.7\
\
U Gem &6.9$^{+0.3}_{-0.3}$ & 8.3$^{+1.0}_{-1.0}$ & 17.1 & 20.6\
\
T Leo &5.2$^{+0.3}_{-0.3}$ & 6.4$^{+2.3}_{-1.7}$ & 13.2 & 16.3\
\
V893 Sco &17.3$^{+1.1}_{-1.1}$ & 50.1$^{+51.9}_{-21.4}$ & 45.7 & 133.0\
\
VY Aqr &1.1$^{+0.2}_{-0.2}$ & 1.3$^{+0.8}_{-0.5}$ & 2.6 & 3.0\
\
WZ Sge &3.1$^{+1.2}_{-0.5}$ & 0.7$^{+0.3}_{-0.1}$ & 7.7 & 1.8\
\
SS Cyg &45.7$^{+0.5}_{-0.8}$ & 150.0$^{+29.0}_{-20.0}$ & 131.7 & 433.0\
\
Z Cam &19.2$^{+1.9}_{-2.6}$ & 61.6$^{+74.4}_{-30.3}$ & 52.3 & 168.0\
\
ASAS J0025 &0.4$^{+0.1}_{-0.1}$ & 1.6$^{+3.8}_{-0.8}$ & 1.1 & 3.9\
\[fluxes\]
One of the sources in our sample, SS Aur, has previously been listed in the [*RXTE*]{} All-Sky Slew Survey catalog where it appears more luminous in X-rays than in our [*Suzaku*]{} observation (the [*RXTE*]{} flux of SS Aur in 2–10 keV is $\sim$ 1.1 $\times$ 10$^{-11}$ erg cm$^{-2}$ s$^{-1}$). We suspect that the higher flux in the [*RXTE*]{} observation is due to other, bright sources in the field which overestimate the flux. E.g., the [*ROSAT*]{} Bright Source Catalogue lists a cluster of galaxies, Abell 553, which is 53’ away from SS Aur and has a WebPIMMS[^6] estimated flux of $\sim$ 9.3 $\times$ 10$^{-12}$ erg cm$^{-2}$ s$^{-1}$ in 2–10 keV [bremsstrahlung kT = 4 keV, Galactic n$_{H}$ = 1.56 $\times$ 10$^{21}$ cm$^{-2}$ as in @ebe96]. Thus, the higher [*RXTE*]{} flux of SS Aur is very likely biased by the background sources and not reliable.
The outburst of KT Per {#sec:ktper}
----------------------
We also analysed the [*Suzaku*]{} outburst data of KT Per obtained in January 2009, and report the results here. KT Per is a Z Cam type dwarf nova, and was seen as an X-ray source by the [*Einstein*]{} satellite in 1979 [@cor84]. We employed the same models which were used for the source sample above, i.e. an absorbed optically thin thermal plasma model and an absorbed cooling flow model with an added 6.4 keV line. Both models yielded acceptable fits: $\chi_{\nu}^{2}/\nu$ = 0.97/838 (thermal plasma) and $\chi_{\nu}^{2}/\nu$ = 0.95/837 (cooling flow). Fig. \[ktper\_fig\] shows the XIS1 and the combined XIS0,3 X-ray spectra of KT Per which have been fitted with an absorbed cooling flow model with a 6.4 keV Gaussian line. The spectral fit parameters for the optically thin thermal plasma and the cooling flow models with fluxes, luminosities and fit statistics are given in Table \[ktper\]. The luminosities given in Table \[ktper\] are calculated for the distance of 180$^{+36}_{-28}$ pc [@tho08]. @bas05 noted that cooling flow models are often a good representation of quiescent X-ray spectra of CVs [see also @muk03], but not outburst spectra. Baskill et al. applied the <span style="font-variant:small-caps;">Xspec</span> multi-temperature model `cevmkl` to their [*ASCA*]{} spectra in order to fit a range of outburst and quiescent spectra with a single simple model. We also investigated how this multi-temperature model combined with photoelectric absorption and a 6.4 keV Gaussian line would fit the outburst data of KT Per, and obtained a statistically acceptable fit: $\chi_{\nu}^{2}/\nu$ = 0.96/836, P$_{0}$ = 0.819.
[ccc]{}\
Parameter & Thermal plasma & Cooling flow\
n$_{H}$ & 14.40$^{+1.60}_{-1.54}$ & 15.70$^{+2.05}_{-1.78}$\
$\times$ 10$^{20}$ cm$^{-2}$ & &\
\
kT & 5.11$^{+0.32}_{-0.31}$ & 12.60$^{+1.47}_{-2.29}$\
(keV)\
\
Abundance & 0.40$^{+0.07}_{-0.07}$ & 0.42$^{+0.09}_{-0.08}$\
\
EW & 52$^{+39}_{-38}$ & 45$^{+44}_{-44}$\
(eV) & &\
\
Flux(2–10 keV) & 2.55 & 2.62\
$\times$ 10$^{-12}$ erg cm$^{-2}$ s$^{-1}$ & &\
\
Flux(0.01–100 keV) & 5.60 & 6.19\
$\times$ 10$^{-12}$ erg cm$^{-2}$ s$^{-1}$ & &\
\
Luminosity(2–10 keV) & 1.0 & 1.03\
$\times$ 10$^{31}$ erg s$^{-1}$\
\
Luminosity(0.01–100 keV)& 2.19 & 2.42\
$\times$ 10$^{31}$ erg s$^{-1}$ & &\
\
$\chi^{2}_{\nu}/\nu$ & 0.97/838&0.95/837\
\
P$_{0}$ &0.730&0.819\
\[ktper\]
Calculating the height of the X-ray emission source above the white dwarf surface
---------------------------------------------------------------------------------
The method which is used for calculating the height of the X-ray emission source above the white dwarf, has been explained by @ish09 for SS Cyg. We have adopted the same method here in our work. As was explained in the beginning of Section 5, an equivalent width of up to $\sim$ 150 eV can be expected for the fluorescent Fe K$\alpha$ line at 6.4 keV. In this work, we have assumed that the reflection originates from the white dwarf surface only, thus the reflection from the accretion disc is $\Omega_{disc}$/2$\pi$ = 0. The equivalent width of 150 eV calculated by @geo91 was assumed under the solar abundance conditions of @mor83 where \[Fe/H\] = 3.2 $\times$ 10$^{-5}$. We have employed the abundances of @and89 which are the default abundance values built in the <span style="font-variant:small-caps;">Xspec</span> cooling flow and optically thin thermal plasma emission models. For the solar abundances of Anders & Grevesse, the \[Fe/H\] composition is 4.68 $\times$ 10$^{-5}$. @ish09, who also employed the Anders & Grevesse abundances, corrected this abundance difference (see their Eq.3) using their measured iron abundance of 0.37 $Z_{0}$. For solar abundance, the observed equivalent width of the 6.4 keV line is
$$\begin{aligned}
\label{eq:ew}
EW_{observed} &= 150 \times \frac{4.68 \times 10^{-5}}{3.2 \times 10^{-5}} \Big(\frac{\Omega_{WD}}{2\pi}\Big)Z\quad(eV)
\nonumber \\
&= 220 \Big(\frac{\Omega_{WD}}{2\pi}\Big)Z\quad(eV),\end{aligned}$$
where $Z$ is the measured elemental abundance in solar units $Z_{\odot}$ and $\Omega_{WD}$ the solid angle of the white dwarf viewed from the plasma of the boundary layer. In our sample, the observed equivalent widths (Table \[ev\]) are mainly below 150 eV and this implies that the X-ray source is located at a height $h$ above the white dwarf surface. In the following, we use the values of $EW_{observed}$ and abundances calculated from the cooling flow model. As an example, the $EW_{observed}$ for SS Aur is 86 eV and the abundance is 1.0, thus Eq. \[eq:ew\] gives $\Omega_{WD}$/2$\pi$ = 0.39. If the X-ray source is point-like, the height $h$ of the X-ray source above the white dwarf of a radius $R_{WD}$ is $h$ $<$ 0.14 $R_{WD}$. As another example, we obtain $\Omega_{WD}$/2$\pi$ = 0.22 and $h$ $<$ 0.64 $R_{WD}$ for V893 Sco ($EW_{observed}$ = 45 eV, $Z$ = 0.94 Z$_{0}$).
Discussion
==========
We have derived an X-ray luminosity function for 12 dwarf novae using archival [*Suzaku*]{}, [*XMM-Newton*]{}, and [*ASCA*]{} observations, and obtained new observations for BZ UMa, SW UMa, VY Aqr, SS Aur, V893 Sco and ASAS J0025 with [*Suzaku*]{} as originally, they were not available in the archive. Our results show that the 2–10 keV luminosities, presented in Table \[fluxes\], span a range between 4 $\times$ 10$^{28}$ and 1.5 $\times$ 10$^{32}$ erg s$^{-1}$, and that most of the source luminosities in the sample are located within 10$^{30}$ erg s$^{-1}$, see Fig. \[histo\], whereas, the X-ray luminosities of the [*ASCA*]{} sample by @bas05 were mainly concentrated on higher luminosities between 10$^{31}$ and 10$^{32}$ erg s$^{-1}$. This difference is most likely due to the fact that we did not apply X-ray selection criteria to our sample. Also, the objects observed by [*ASCA*]{} were known to be X-ray bright, thus the sample of Baskill et al. is very likely biased by sources which are X-ray bright.
In order to derive the integrated X-ray luminosity function (XLF), N($>$ L), for 12 sources within a distance of d = 200 pc, we assumed that the luminosity function is characterized by a power law N($>$ L) = k(L/L$_{t}$)$^{-\alpha}$ (see Fig. \[lum\] where the best-fit parameters $\alpha$ = -0.64 and k = 2.39 $\times$ 10$^{-7}$, corresponding to a threshold luminosity of L$_{t}$ = 3 $\times$ 10$^{30}$ erg s$^{-1}$). The histogram illustrates the cumulative source distribution per pc$^{3}$ in which a break is seen at $\sim$ 3 $\times$ 10$^{30}$ erg s$^{-1}$. This can be due to two possible scenarios: 1) a single $\alpha$ power law describes the luminosity function of DNe, but the sample becomes more incomplete below $\sim$ 3 $\times$ 10$^{30}$ erg s$^{-1}$ than it is above this limit, or 2) the shape of the true XLF of DNe is a broken power law with a break at around 3 $\times$ 10$^{30}$ erg s$^{-1}$. From these two scenarios, the first one is more likely since the sample contains only a few sources below $\sim$ 10$^{30}$ erg s$^{-1}$. Also, as was shown by, e.g., the study of @gan09, more fainter CVs, such as WZ Sge types, are expected to exist. Based on the obtained power law slope, the sample is dominated by the brighter DNe: this is probably caused by the parallax measurement method which favours optically brighter DNe which usually have high X-ray luminosities.
![The cumulative source distribution (histogram) and the integrated power law luminosity function N($>$ L) as a function of X-ray luminosity in log L in the 2–10 keV energy band. The error bar on the top right represents a typical error on the luminosities.[]{data-label="lum"}](fig10.eps){width="90mm"}
When calculating the total, integrated luminosity of the sample, we restricted the calculations to the distance of 200 pc, thus excluding BZ UMa. Integrating between the luminosities of 1 $\times$ 10$^{28}$ and the maximum luminosity of the sample ($L_{max}$ = 1.50 $\times$ 10$^{32}$ erg s$^{-1}$), yields the total integrated luminosity of 1.48 $\times$ 10$^{32}$ erg s$^{-1}$, whereas the integrated luminosity between the threshold luminosity 3 $\times$ 10$^{30}$ and $L_{max}$ is 1.15 $\times$ 10$^{32}$ erg s$^{-1}$. These two results show that there are uncertainties in the integrated luminosities, most likely caused by the small number of sources in the sample. In order to obtain more accurate value for the integrated luminosity, the power law slope ($\alpha$ = -0.64) should be better established. If the obtained slope is not far from the true power law slope of DNe in the solar neighbourhood, estimating the integrated luminosity more accurately and constraining the bright luminosity end (10$^{32}$ erg s$^{-1}$) requires more DNe to be included in the sample. Since the source density at $\sim$ 10$^{32}$ erg s$^{-1}$ is $\sim$ 3 $\times$ 10$^{-8}$ pc$^{-3}$ according to Fig. \[lum\], we would need to survey within a volume of 1 $\times$ 10$^{9}$ pc$^{3}$ to find $\sim$ 30 SS Cyg -type DNe and thus find a statistically significant constraint for the brighter luminosities in the sample. This volume would correspond to a distance of $\sim$ 620 pc with a flux limit of $\sim$ 3.2 $\times$ 10$^{-12}$ erg cm$^{-2}$ s$^{-1}$.
Following this, we estimated how easy it would be to hide typical DN luminosities in the solar neighbourhood. Assuming a typical dwarf nova with a 5 keV bremsstrahlung and a low Galactic n$_{H}$ = 1 $\times$ 10$^{20}$ cm$^{-2}$ in <span style="font-variant:small-caps;">WebPIMMs</span> yields a 2–10 keV flux of 5 $\times$ 10$^{-13}$ erg cm$^{-2}$ s$^{-1}$, corresponding to the [*ROSAT*]{} PSPC count rate of 0.04 ct s$^{-1}$ which is just below the detection limit (0.05 ct s$^{-1}$) of [*ROSAT*]{} PSPC [@vog99]. Thus, luminosities above 2.4 $\times$ 10$^{30}$ erg s$^{-1}$ within 200 pc or above 6 $\times$ 10$^{29}$ erg s$^{-1}$ within 100 pc should have been found by [*ROSAT*]{} and thus should be in the RASS. However, given that sources with luminosities of 10$^{30}$ erg s$^{-1}$ and below at a distance of 100 pc were too faint for the RASS, and that our XLF peaks at $\sim$ 10$^{30}$ erg s$^{-1}$, we conclude that there is no existing X-ray selected sample that we can use for this line of research.
How far is the total luminosity of our sample from accounting for the total CV X-ray emissivity? In order to estimate this, we calculated the absolute lower limit for the luminosity per cubic parsec volume (L$_{x}$/vol). For a distance of r = 200 pc, the volume V = 4/3 $\times$ $\pi$ $\times$ (200 pc)$^{3}$ = 3.3 $\times$ 10$^{7}$ pc$^{3}$, and the total summed luminosity L$_{x}$ of the sample is 2.39 $\times$ 10$^{32}$ erg s$^{-1}$ (without BZ UMa). Thus, the total absolute lower limit L$_{x}$/volume = 7.24 $\times$ 10$^{24}$ erg s$^{-1}$ pc$^{-3}$. Normalising this value to the local stellar mass density 0.04 M$_{\odot}$ pc$^{-3}$ [@jah97] yields 1.81 $\times$ 10$^{26}$ erg s$^{-1}$ M$_{\odot}^{-1}$ in the 2–10 keV range. For comparison, @saz06 obtained (1.1 $\pm$ 0.3) $\times$ 10$^{27}$ erg s$^{-1}$ M$_{\odot}^{-1}$ (2–10 keV) for the total CV X-ray emissivity per unit stellar mass. Thus, our sample would account for $\sim$ 16 per cent of this value.
And finally, how much would our sample account for the GRXE? The Galactic Ridge X-ray emissivity estimated by @rev06 in the 3–20 keV range was L$_{x}$/M $\sim$ (3.5 $\pm$ 0.5) $\times$ 10$^{27}$ erg s$^{-1}$ M$_{\odot}^{-1}$, meaning that our sample would account for 5 per cent of the Galactic Ridge X-ray emissivity. As we estimated the X-ray emissivity of all CVs within 200 pc, we used the exponential vertical density profile
$$\label{density}
\rho(z) = \rho_{0}e^{|z|/h},$$
of CVs with a scale height for short period systems ($h$ = 260 pc) as in @pre07b, where $z$ = $d$ sin $b$ is the perpendicular distance from the Galactic plane and $b$ Galactic latitude. Integrating Eq.\[density\] over a sphere with a radius of 200 pc gives $\sim$ 280 as the total number of DNe within 200 pc. If the space density of DNe follows the space density of CVs as in @pre07b, i.e., $\rho_{0}$ = 1.1$^{+2.3}_{-0.7}$ $\times$ 10$^{-5}$ pc$^{-3}$, and if a typical DN has an X-ray luminosity corresponding to the mean luminosity (2 $\times$ 10$^{31}$ erg s$^{-1}$) of our sample of 11 sources (BZ UMa excluded), then the 2–10 keV X-ray emissivity of all DNe in the solar neighbourhood would be 5.5$^{+11.5}_{-3.5}$ $\times$ 10$^{27}$ erg s$^{-1}$ M$_{\odot}^{-1}$ (these account for the uncertainty on the space density, assuming that this is the dominant source of uncertainty for the X-ray emissivity of DNe). This would account for more than 100 per cent of the GRXE emissivity. If DNe were uniformly distributed in the solar neighbourhood, the X-ray emissivity would be overestimated also in this case (by 20–30 per cent). However, in both cases, one should remember that the calculated X-ray emissivity of all DNe within 200 pc is likely overestimated by the brighter sources in our sample, thus the calculations give excess emission.
Correlations between X-ray luminosity and other parameters {#correlation}
----------------------------------------------------------
In order to understand whether the X-ray luminosity and the various parameters (inclination $i$, orbital period $P_{orb}$, shock temperature $kT_{max}$ and white dwarf mass $M_{WD}$) are correlated, we carried out Spearman’s rank correlation test. Plotting X-ray luminosity versus a few of these parameters ($i$ and $P_{orb}$) shows that GW Lib seems to appear as an outlier compared to the rest of the sample (Fig. \[orb\] and \[incl\]). Thus, to explore how the presence/absence of GW Lib affects the test results, two test cases were used: 1) GW Lib was included, and 2) GW Lib was excluded from the rest of the sample. In addition, we investigated whether a correlation between the white dwarf masses $M_{WD}$ and the shock temperatures $kT_{max}$ (Fig. \[mass\_temp\]) exists, although in this case, GW Lib seems to follow the rest of the sample, thus, carrying out test case 2) was not necessary.
![The X-ray luminosities (2.0–10.0 keV) versus orbital periods of the source sample.[]{data-label="orb"}](fig11.eps){width="70mm"}
![The X-ray luminosities (2.0–10.0 keV) versus inclinations of the source sample.[]{data-label="incl"}](fig12.eps){width="70mm"}
A strong correlation was found at the 99.95 per cent significance level (2.8$\sigma$) between the X-ray luminosities and orbital periods (Fig. \[orb\]) when GW Lib is included in the sample. The correlation still holds when GW Lib is excluded (significance is 99.67 per cent). @bas05 noted that there was a weak correlation between the X-ray luminosities and the orbital periods in their [*ASCA*]{} sample, concluding that the X-ray luminosity probably also correlates with long-term mean accretion rate.
The X-ray luminosity and the inclination $i$ are not correlated in either case (Fig. \[incl\]). The correlation between these parameters was measured when the inclination of BZ UMa was set to 65$^{\circ}$, and altering the inclination between 60$^{\circ}$ and 75$^{\circ}$ did not affect the result. Since no correlation was found, this result is in contrast with the discovery of anti-correlation between the emission measure and inclination found by @van96. It is worth noting that the [*ROSAT*]{} bandpass was very narrow, covering only 0.1–2.4 keV where the softer X-ray emission (and more luminous emission) is probably intrinsically absorbed by the sources. In addition, an anti-correlation between the X-ray luminosity and inclination was also seen by @bas05 in the [*ASCA*]{} sample, although, Baskill et al. noted that the inclinations might be uncertain, and this can also be the case in our sample.
Finally, the white dwarf masses $M_{WD}$ and the shock temperatures $kT_{max}$ correlate with a significance of 98.5 per cent when the mass of VY Aqr is 0.80 M$_{\odot}$, but becomes less significant (97.4 per cent) if the mass is 0.55 M$_{\odot}$. Of the rest of the parameters, i.e. the X-ray luminosity $L_{x}$ versus $kT_{max}$ and $M_{WD}$, $kT_{max}$ showed evidence of correlation with $L_{x}$ at a significance of 97.6 per cent when GW Lib was included in the sample, but $L_{x}$ and $M_{WD}$ had a much lower correlation significance (69 per cent) when including GW Lib. For the latter correlation test ($L_{x}$ versus $M_{WD}$), the result was the same with both $M_{WD}$ values for VY Aqr. Excluding GW Lib decreased the significance to 91 per cent ($L_{x}$ versus $kT_{max}$) and to 63 per cent ($L_{x}$ versus $M_{WD}$).
Conclusions
===========
We have analysed the X-ray spectra of 13 dwarf novae with accurate parallax-based distance estimates, and derived the most accurate shape for the X-ray luminosity function of DNe in the 2–10 keV band to date due to accurate distance measurements and due to the fact that we did not use an X-ray selected sample.
The derived X-ray luminosities are located between $\sim$ 10$^{28}$–10$^{32}$ erg s$^{-1}$, showing a peak at $\sim$ 10$^{30}$ erg s$^{-1}$. Thus, we have obtained peak luminosities which are lower compared to other previous studies of CV luminosity functions. The shape of the X-ray luminosity function of the source sample suggests that the two following scenarios are possible: 1) the sample can be described by a power law with a single $\alpha$ slope, but the sample becomes more incomplete below $\sim$ 3 $\times$ 10$^{30}$ erg s$^{-1}$ than it is above this limit, or, 2) the shape of the real X-ray luminosity function of dwarf novae is a broken power law with a break at around 3 $\times$ 10$^{30}$ erg s$^{-1}$.
The integrated luminosity between 1 $\times$ 10$^{28}$ erg s$^{-1}$ and the maximum luminosity of the sample, 1.50 $\times$ 10$^{32}$ erg s$^{-1}$, is 1.48 $\times$ 10$^{32}$ erg s$^{-1}$. In order to better constrain the integrated luminosity and the slope of the X-ray luminosity function, more dwarf novae need to be included in the sample. Thus, we suggest more future X-ray imaging observations of dwarf novae in the 2–10 keV band with accurate distance measurements. The total X-ray emissivity of the sample within a radius of 200 pc is 1.81 $\times$ 10$^{26}$ erg s$^{-1}$ M$^{-1}_{\odot}$ (2–10 keV). This accounts for $\sim$ 16 per cent of the total X-ray emissivity of CVs as estimated by @saz06, and $\sim$ 5 per cent of the Galactic Ridge X-ray emissivity.
The X-ray luminosities and the inclinations of our sample do not show anti-correlation which has been seen in other previous correlation studies, but a strong correlation is seen between the X-ray luminosities and the orbital periods. Also, evidence for a correlation between the white dwarf masses and the shock temperatures exists. In the future, larger dwarf nova samples are needed in order to confirm these results.
Acknowledgments {#acknowledgments .unnumbered}
===============
This research has made use of data obtained from the [*Suzaku*]{} satellite, a collaborative mission between the space agencies of Japan (JAXA) and the USA (NASA). JO acknowledges support from STFC. Part of this work is based on observations obtained with [*XMM-Newton*]{}, an ESA science mission with instruments and contributions directly funded by ESA Member States and the USA (NASA). We thank the reviewer M. Revnivtsev for his helpful comments on this paper.
[99]{} Arnaud K.A., 1996, ASP Conf. Ser. Vol. 101, Astronomical Data Analysis Software and Systems V. Astron. Soc. Pac., San Francisco, p. 17
Anders E., Grevesse N., 1989, Geochim. Cosmochim. Acta, 53, 197
Baskill D., Wheatley P. J., Osborne J. P., 2001, MNRAS, 328, 71
Baskill D., Wheatley P. J., Osborne J. P., 2005, MNRAS, 357, 626
Byckling K., Osborne J. P., Wheatley P. J., Wynn G. A., Beardmore A., Braito V., Mukai K., West R., 2009, MNRAS, 399, 1576
Córdova F.A., Mason K.O., 1984, MNRAS, 206, 879
Done C., Osborne J. P., 1997, MNRAS, 288, 649
Ebeling H., Voges W., Böhringer H., Edge A. C., Huchra J. P., Briel U. G., 1996, MNRAS, 281, 799
Ebisawa K., Maeda Y., Kaneda H., Yamauchi S., 2001, Science, 293, 1633
Frank J., King A., Raine D., 2002, Accretion Power in Astrophysics, CUP, p. 156
Friend M. T., Connon-Smith, R., Jones, D. H. P., 1990, MNRAS, 246, 654
George I. M., Fabian, A. C., 1991, MNRAS, 249, 352
Gänsicke B. T. et al., 2009, MNRAS, 397, 2170
Harrison T. E., McNamara B. J., Szkody P., McArthur B. E., Benedict G. F., Klemola A. R., Gilliland R. L., 1999, ApJ, 515, 93
Harrison T. E., Johnson, J. J., McArthur, B. E., Benedict, G. F., Szkody, P., Howell, S. B., Gelino, D. M., 2004, AJ, 127, 460
Hilton E.J., Szkody P., Mukadam A., Mukai K., Hellier C., van Zyl L., Homer L., 2007, AJ, 134, 1503
Horne K., Wood J. H., Stiening R. F., 1991, ApJ, 378, 271
Howell S. B., Rappaport S., Politano M., 1997, MNRAS, 287, 929
Ishida M., Okada S., Hayashi T., Nakamura R., Terada Y., Mukai K., Hamaguchi K., 2009, PASJ, 61, S77
Kolb U., 1993., A&A, 271, 149
Jahreiß H., Wielen R., 1997, ESA SP-402: Hipparcos - Venice ’97, 675
Jansen F. et al., 2001, A&A, 365, 1
Lasota J.-P., 2001, NewAR, 45, 449
Lemm K., Patterson J., Thomas G., Skillman D. R., 1993, PASP, 105, 1120
Liedahl D. A., Osterheld A. L., Goldstein W. H., 1995, ApJ, 438, L115
Mason E., Skidmore W., Howell S.B., Mennickent R.E., 2001, ApJ, 563, 351
Mewe R., Lemen J. R., van den Oord G. H. J., 1986, A&A, 65, 511
Mitsuda K. et al., 2007, PASJ, 59, 1
Mukai K., Shiokawa K., 1993, ApJ, 418, 863
Mukai K., Wood J. H., Naylor T., Schlegel E. M., Swank J. H., 1997, ApJ, 475, 812
Mukai K., Kinkhabwala A., Peterson J. R., Kahn S.M., Paerels F., 2003, ApJ, 586, L77
Mukai K., Zietsman E., Still M., 2009, ApJ, 707, 652
Mushotzky R. F., Szymkowiak A. E., 1988, in Cooling Flows in Clusters and Galaxies, ed. A.C. Fabian, (Dordrecht:Kluwer), 53
Morrison R., McCammon D., 1983, ApJ, 270, 119
Pandel D., Córdova F. A., Mason K. O., Priedhorsky W. C., 2005, ApJ, 626, 396
Patterson J., 1984, ApJS, 54, 443
Patterson J., Kemp J., Harvey D. et al., 2005, PASP, 117, 1204
Pretorius M. L., Knigge C., Kolb U., 2007a, MNRAS, 374, 1495
Pretorius M. L., Knigge C., O’Donoghue D., Henry J. P., Gioia I. M., Mullis C. R., 2007b, MNRAS, 382, 1279
Pringle, J. E., Webbink, R. F., 1975, MNRAS, 172, 493
Pringle J. E., 1977, MNRAS, 178, 195
Pringle J. E., Savonije G. J., 1979, MNRAS, 187, 777
Revnivtsev M., Sazonov S., Gilfanov M., Churazov E., Sunyaev R., 2006, A&A, 452, 169
Revnivtsev M., Vikhlinin A., Sazonov S., 2007, A&A, 473, 857
Revnivtsev M., Sazonov S., Krivonos R., Ritter H., Sunyaev R., 2008, A&A, 489, 1121
Revnivtsev M., Sazonov S., Churazov E., Forman W., Vikhlinin A., Sunyaev R., 2009, Nature, 458, 1142
Ritter H., Kolb U., 2003, A&A, 404, 301 (update RKcat7.10)
Sazonov S., Revnivtsev M., Gilfanov M., Churazov E., Sunyaev R., 2006, A&A, 450, 117
Scargle J. D., 1982, ApJ, 263, 835
Tanaka Y., Inoue H., Holt S. S., 1994, PASJ, 46, L37
Templeton M. R. et al., 2006, PASP, 118, 236
Thorstensen J. R., 2003, AJ, 126, 3017
Thorstensen J. R., Lépine S., Shara M., 2008, AJ, 136, 2107
Urban J.A., Sion M.E., 2006, ApJ, 642, 1029
van Teeseling A., Beuermann K., Verbunt F., 1996, ApJ, 315, 467
Voges W. et al., 1999, A&A, 349, 389
Warwick R. S., Turner M. J. L., Watson M. G., Willingale R., 1985, Nature, 317, 218
Wheatley P. J., Verbunt F., Belloni T., Watson M. G., Naylor T., Ishida M., Duck S. R., Pfeffermann E., 1996, A&A, 307, 137
Worrall D. M., Marshall F. E., Boldt E. A., Swank J. H., 1982, ApJ, 255, 111
\[lastpage\]
[^1]: E-mail: [email protected]
[^2]: www.aavso.org
[^3]: http://heasarc.gsfc.nasa.gov/docs/suzaku/analysis/abc/abc.html
[^4]: http://xmm2.esac.esa.int/sas/8.0.0/documentation/threads/\
EPIC\_OoT.html
[^5]: http://heasarc.gsfc.nasa.gov/docs/asca/abc/abc.html
[^6]: http://heasarc.gsfc.nasa.gov/Tools/w3pimms.html
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'A frequent topic in the study of pattern avoidance is identifying when two sets of patterns $\Pi, \Pi''$ are Wilf equivalent, that is, when $|\text{Av}_n(\Pi)| = |\text{Av}_n(\Pi'')|$ for all $n$. In recent work of Dokos et al. the notion of Wilf equivalence was refined to reflect when avoidance of classical patterns preserves certain statistics. In this article, we continue their work by examining $\text{des}$-Wilf equivalence when avoiding certain non-classical patterns.'
address:
- |
Department of Mathematics\
Michigan State University\
East Lansing, MI 48824
- |
School of Mathematics and Statistics\
Xi’an Jiaotong University\
Xi’an, Shaanxi 710049, China
author:
- Caden Bielawa
- Robert Davis
- Daniel Greeson
- Qinhan Zhou
bibliography:
- 'references.bib'
title: 'Descents and $\operatorname{des}$-Wilf Equivalence of Permutations Avoiding Certain Non-classical Patterns'
---
[^1]
Introduction
============
Let ${\mathfrak{S}}_n$ denote the set of permutations of $[n] := \{1,\dots,n\}$, and let ${\mathfrak{S}}= {\mathfrak{S}}_1 \cup {\mathfrak{S}}_2 \cup \dots$ be the set of all permutations of finite length. We write $\sigma \in {\mathfrak{S}}_n$ as $\sigma = a_1a_2\dots a_n$ to indicate that $\sigma(i) = a_i$. A function $\operatorname{st}: {\mathfrak{S}}_n \to {\mathbb{N}}$ is called a *statistic*, and the systematic study of permutation statistics is generally accepted to have begun with MacMahon [@MacMahonCA Volume I, Section III, Chapter V]. Four of the most well-known statistics are the *descent*, *inversion*, *major*, and *excedance* statistics, defined respectively by $$\begin{aligned}
\operatorname{des}(\sigma) &=& | \operatorname{Des}(\sigma)| \\
\operatorname{inv}(\sigma) &=& | \{(i,j) \in [n]^2\ |\ i < j \text{ and } a_i > a_j \}| \\
\operatorname{maj}(\sigma) &=& \sum_{i \in \operatorname{Des}(\sigma)} i \\
\operatorname{exc}(\sigma) &=& |\{ i \in [n]\ |\ a_i > i\}|,\end{aligned}$$ where $\operatorname{Des}(\sigma) = \{i \in [n-1]\ |\ a_i > a_{i+1}\}$. Given any statistic $\operatorname{st}$, one may form the generating function $$F_n^{\operatorname{st}}(q) = \sum_{\sigma \in {\mathfrak{S}}_n} q^{\operatorname{st}\sigma}.$$ A famous result due to MacMahon [@MacMahonCA] states that $F_n^{\operatorname{des}}(q) = F_n^{\operatorname{exc}}(q)$, and that both are equal to the *Eulerian polynomial* $A_n(q)$. Similarly, it is known that $F_n^{\operatorname{inv}}(q) = F_n^{\operatorname{maj}}(q) = [n]_q!$, where $$[n]_q = 1 + q + \dots + q^{n-1} \text{ and } [n]_q! = [n]_q[n-1]_q\dots [1]_q.$$
Let $A \subseteq [n]$, and denote by ${\mathfrak{S}}_A$ the set of permutations of the elements of $A$. The *standardization* of $\sigma = a_1\dots a_{|A|} \in {\mathfrak{S}}_A$ is the element of ${\mathfrak{S}}_{|A|}$ whose letters are in the same relative order as those of $\sigma$; we denote this permutation by $\operatorname{std}(\sigma)$. Now, we say that a permutation $\sigma \in {\mathfrak{S}}_n$ *contains the pattern* $\pi \in {\mathfrak{S}}_k$ if there exists a subsequence $\sigma' = a_{i_1}\dots a_{i_k}$ of $\sigma$ such that $\operatorname{std}(\sigma') = \pi$. If no such subsequence exists, then we say that $\sigma$ *avoids the pattern* $\pi$. Since we will introduce additional notions of patterns, we may call such a pattern a *classical pattern* to avoid confusion. If $\Pi \subseteq {\mathfrak{S}}$, then we say $\sigma$ *avoids* $\Pi$ if $\sigma$ avoids every element of $\Pi$. The set of all permutations of ${\mathfrak{S}}_n$ avoiding $\Pi$ is denoted $\operatorname{Av}_n(\Pi)$. In a mild abuse of notation, if $\Pi = \{\pi\}$, we will write $\operatorname{Av}_n(\pi)$. If $\Pi,\Pi'$ are two sets of patterns and $|\operatorname{Av}_n(\Pi)| = |\operatorname{Av}_n(\Pi')|$ for all $n$, then we say $\Pi$ and $\Pi'$ are *Wilf equivalent* and write $\Pi \equiv \Pi'$.
These ideas may be combined by setting $$F_n^{\operatorname{st}}(\Pi;q) = \sum_{\sigma \in \operatorname{Av}_n(\Pi)} q^{\operatorname{st}\sigma}.$$ This allows one to say that $\Pi,\Pi'$ are *$\operatorname{st}$-Wilf equivalent* if $F_n^{\operatorname{st}}(\Pi;q) = F_n^{\operatorname{st}}(\Pi';q)$ for all $n$, and write this as $\Pi {\overset{\operatorname{st}}{\equiv}}\Pi'$. Thus, $\Pi$ and $\Pi'$ may be Wilf equivalent without being $\operatorname{st}$-Wilf equivalent. As a concrete example, $123$ and $321$ are clearly not $\operatorname{des}$-Wilf equivalent, even though they are Wilf equivalent. It is straightforward to check that $\operatorname{st}$-Wilf equivalence is indeed an equivalence relation on ${\mathfrak{S}}$.
Since it is generally a difficult question to determine whether two sets are nontrivially Wilf equivalent, one should not expect it to be any easier to determine $\operatorname{st}$-Wilf equivalence. However, it is certainly possible to obtain some results; see [@DokosEtAl] for results regarding $F_n^{\operatorname{inv}}$ and $F_n^{\operatorname{maj}}$, and [@Baxter; @CameronKillpatrick] for further results, including a study of enumeration strategies for questions of this nature. In this article, we will study $F_n^{\operatorname{des}}(\Pi;q)$ for certain non-classical patterns, called mesh patterns and barred patterns. Special cases will allow us to identify $\operatorname{des}$-Wilf equivalences. We will also present several conjectural $\operatorname{des}$-Wilf equivalences and provide computational evidence for these.
Acknowledgements
----------------
The authors would like to thank the anonymous referee for his or her thoughtful comments which significantly improved this work.
Pattern Avoidance Background
============================
Classical Patterns
------------------
In order to work most efficiently, it is important to recognize that certain Wilf equivalences are almost immediate to establish. For example, it is obvious that $|\operatorname{Av}_n(123)| = |\operatorname{Av}_n(321)|$, since $a_1\dots a_n \in \operatorname{Av}_n(123)$ if and only if $a_na_{n-1}\dots a_1 \in \operatorname{Av}_n(321)$. This idea can be generalized significantly.
(1,1) grid (6,6); iin [(1,3),(2,4),(3,2),(4,5),(5,1),(6,6)]{} [ icircle \[radius = 3pt\]; ]{}
The $\emph{plot}$ of $\sigma \in {\mathfrak{S}}_n$ is the set of pairs $(i,\sigma(i)) \in {\mathbb{R}}^2$ and will be denoted $P(\sigma)$. The plot of $342516$ is shown in Figure \[fig: plot\]. Let $D_4 = \{R_0,R_{90},R_{180},R_{270},r_{-1},r_{0},r_{1},r_{\infty}\}$, where $R_{\theta}$ is counterclockwise rotation of a plot by an angle of $\theta$ degrees and $r_m$ is reflection across a line of slope $m$. A couple of these rigid motions have easy descriptions in terms of the one-line notation for permutations. If $\pi = a_1a_2 \ldots a_k$ then its [*reversal*]{} is $\pi^r =a_k\ldots a_2a_1 =r_{\infty}(\pi)$, and its [*complement*]{} is $\pi^c = (k+1-a_1)\ (k+1-a_2)\ \dots \ (k+1-a_k)=r_0(\pi)$.
Note that $\sigma \in \operatorname{Av}_n(\pi)$ if and only if $f(\sigma) \in \operatorname{Av}_n(f(\pi))$ for any $f \in D_4$, hence $\pi \equiv f(\pi)$. For this reason, the equivalences induced by the dihedral action on a square are often referred to as the [*trivial Wilf equivalences*]{}.
Using these techniques, it is easy to show that $123$ and $321$ are trivially Wilf equivalent, as are all of $132$, $213$, $231$, and $312$. It is less obvious, however, whether $123$ and $132$ are Wilf equivalent. This question was settled by independent results due to MacMahon [@MacMahonCA] and Knuth [@KnuthVol1], whose combined work showed that $\operatorname{Av}_n(132)$ and $\operatorname{Av}_n(123)$ are enumerated by the *$n^{th}$ Catalan number* $$C_n = \frac{1}{n+1}\binom{2n}{n}.$$ The Catalan numbers are famous for appearing in a multitude of combinatorial situations; see [@StanleyCatalan] for many of them.
One of the most well-known combinatorial objects enumerated by the Catalan numbers are Dyck paths. A *Dyck path of length $2n$* is a lattice path in ${\mathbb{R}}^2$ starting at $(0,0)$ and ending at $(2n,0)$, using steps $(1,1)$ and $(1,-1)$, which never goes below the $x$-axis. See Figure \[fig: dyck\] for an example Dyck path of length $8$.
(0,0) – (8,0);
\(A) at (0,0) ; (B) at (1,1) ; (C) at (2,0) ; (D) at (3,1) ; (E) at (4,2) ; (F) at (5,1) ; (G) at (6,2) ; (H) at (7,1) ; (I) at (8,0) ;
\(A) – (B) – (C) – (D) – (E) – (F) – (G) – (H) – (I);
in [(0,0),(1,1),(2,0),(3,1),(4,2),(5,1),(6,2),(7,1),(8,0)]{} [ circle \[radius=3pt\]; ]{}
Non-classical Patterns
----------------------
In this section, we will define two classes of non-classical patterns and describe what it means for a permutation to contain or avoid them. The definitions of Wilf equivalence and $\operatorname{des}$-Wilf equivalence then extend to these patterns in the same way as classical patterns, so their precise definitions will be omitted.
A *mesh pattern* is a pair $(\pi,M)$ where $\pi \in {\mathfrak{S}}_k$ and $M \subseteq [0,k]^2$. Mesh patterns are a vast generalization of classical patterns and were first introduced by Brändén and Claesson [@BrandenClaessonMesh]. It is convenient to represent a mesh pattern as a grid which plots $\pi$ and shades in the unit squares whose bottom-left corners are the elements of $M$. For example, one may represent the mesh pattern $(\pi_0,M_0) = (4213,\{(0,2),(1,0),(1,1),(3,3),(3,4),(4,3)\})$ as follows: $$(\pi_0,M_0) = { \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {0/2,1/0,1/1,3/3,3/4,4/3}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {4,2,1,3}}+0.99,\xintNthElt{0}{\xintCSVtoList {4,2,1,3}}+0.99);
}
{\foreach[count=\x] \y in {4,2,1,3}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {4,2,1,3}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$$
Containment of mesh patterns is most easily understood by an informal statement but illustrative examples; the formal definition, given in [@BrandenClaessonMesh], shows that the intuition developed this way behaves as expected. We say that $\sigma \in {\mathfrak{S}}_n$ *contains the mesh pattern* $(\pi,M)$ if $\sigma$ contains an occurrence of $\pi$ and the shaded regions of $P(\pi)$ corresponding to this occurrence contain no other elements of $P(\sigma)$. If $\sigma$ does not contain $(\pi,M)$, then we say $\sigma$ *avoids* $(\pi,M)$.
For the illustrative examples, first consider $\sigma = 612435$. Notice that while $6435$ is an occurrence of $4213$ in $\sigma$, it is not an occurrence of the mesh pattern $(\pi_0,M_0)$ given above, since the shaded regions in $P(\sigma)$ dictated by $M_0$ yield $${ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {0/4,1/0,1/1,1/2,1/3,2/0,2/1,2/2,2/3,3/0,3/1,3/2,3/3,5/5,5/6,6/5}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {6,1,2,4,3,5}}+0.99,\xintNthElt{0}{\xintCSVtoList {6,1,2,4,3,5}}+0.99);
}
{\foreach[count=\x] \y in {6,1,2,4,3,5}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {6,1,2,4,3,5}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$$ Now consider $\sigma' = 153624$. In this case, $5324$ is an occurrence of both $4213$ and $(\pi_0,M_0)$ in $\sigma'$, since the shading in this case is $${ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {0/3,1/3,2/0,2/1,2/2,5/4,5/5,5/6,6/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,5,3,6,2,4}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,5,3,6,2,4}}+0.99);
}
{\foreach[count=\x] \y in {1,5,3,6,2,4}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,5,3,6,2,4}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$$
In certain cases, determining which permutations avoid a mesh pattern $(\pi,M)$ with $M$ nonempty is equivalent to determining which permutations avoid $\pi$ as a classical pattern. When this happens, we say that $(\pi,M)$ has *superfluous mesh*, and Tenner [@TennerSuperfluous] identified when exactly a mesh pattern has superfluous mesh. To do this, we first define an *enclosed diagonal* of $(\pi,M)$ to be a triple $((i,j),\varepsilon,\ell)$ where $\varepsilon \in \{-1,1\}$, $\ell \geq 1$, and the following three properties hold:
1. The plot of $\pi$ contains the set $\{(i+d,j+\varepsilon d) \mid 1 \leq d < \ell\}$;
2. The plot of $\pi$ contains neither $(i,j)$ nor $(i+\ell,j+\varepsilon \ell)$;
3. $\{(i+d,j+\varepsilon d) \mid 0 \leq d < \ell\} \subseteq M$.
Note that an enclosed diagonal may consist of a single element, as long as the corresponding box in the mesh pattern contains no element of $P(\pi)$. To illustrate, the following three mesh patterns all have a unique enclosed diagonal: $${ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {2/0}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
} \quad { \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {0/0,1/1}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
} \quad { \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/3,2/2,3/1}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$$ However, none of the following five mesh patterns have any enclosed diagonals: $${ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {2/2}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
} \quad { \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {0/1}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
} \quad { \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/1,1/2}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
} \quad { \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/1,1/2,2/1,2/2}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
} \quad { \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {0/1,1/1,2/1,3/1}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$$ The following theorem gives the characterization of when a pattern has superfluous mesh. As a result, we will not focus on any patterns with superfluous mesh, but we will still use the theorem briefly.
\[thm:superfluous\] A mesh pattern has superfluous mesh if and only if it has no enclosed diagonals.
Mesh patterns also generalize *$1$-barred patterns*, in which a classical pattern is allowed (but not required) to have a bar above one letter. This is a special case of *barred patterns*, in which each letter is allowed to have a bar above it. The bars above letters indicate that certain additional rules are required in order to define containment of the pattern. We will not give the precise definition of containment and avoidance of barred patterns in general, but will observe that if there are two or more bars in the pattern, there is not necessarily a simple translation of the barred pattern into a mesh pattern. In some instances, a barred pattern may be described as a *decorated mesh pattern* [@UlfarssonDecoratedMesh], but this is not always possible. To avoid this difficulty in the statement and proof of Proposition \[prop:barred prop\], we will simply describe here what it means for a permutation to avoid two specific barred patterns.
We say that $\sigma = a_1\dots a_n$ *avoids* $\bar1\bar243$ if, whenever $a_ia_j$ is an occurrence of $21$, then there are some integers $k,l$ such that $k < l < i$ and $a_ka_la_ia_j$ is an occurrence of $1243$. We also say that $\sigma$ *avoids* $\bar132\bar4$ if, whenever $a_ia_j$ is an occurrence of $21$, then there are some integers $k,l$ such that $k < i < j < l$ and $a_ka_ia_ja_l$ is an occurrence of $1324$. As an example, $\sigma = 124635$ avoids $\bar1\bar243$ since all occurrences of $21$, which are $43$, $63$, and $65$, extend to an occurrence of $1243$ by placing $12$ before them. However, $\sigma$ contains $\bar132\bar4$ since $63$, which is an occurrence of $21$, does not play the role of $32$ in any occurrence of $1324$ in $\sigma$.
Main Results
============
We now have all of the tools we need to begin proving results. We begin with a simple application of several known theorems.
If $(132,M_1)$ and $(312,M_2)$ are mesh patterns, neither of which contain an enclosed diagonal, then $$(132,M_1) {\overset{\operatorname{des}}{\equiv}}(312,M_2).$$
By Theorem \[thm:superfluous\], $\operatorname{Av}_n((312,M_2)) = \operatorname{Av}_n(312)$, so $(312,M_2) {\overset{\operatorname{des}}{\equiv}}312$. It then follows directly from [@Reifegerste Remark 2.5(b)] that the number of elements in $\operatorname{Av}_n(312)$ with exactly $k$ descents is $$N_{n,k} := \frac{1}{n}\binom{n}{k}\binom{n}{k+1}.$$ Since the sequence $\{N_{n,k}\}_{k = 0}^{n-1}$ is symmetric for fixed $n$, and since $$\operatorname{des}(\sigma) = n - 1 - \operatorname{des}(\sigma^c),$$ we have $$(312,M_2) {\overset{\operatorname{des}}{\equiv}}312 {\overset{\operatorname{des}}{\equiv}}132.$$ Again by Theorem \[thm:superfluous\], we have $\operatorname{Av}_n(132) = \operatorname{Av}_n((132,M_1))$, so these two patterns are $\operatorname{des}$-Wilf equivalent as well. Connecting the equivalences, the claim follows.
Characterizing the $\operatorname{des}$-Wilf classes for mesh patterns $(\pi,M)$ where $\pi \in {\mathfrak{S}}_4$ is difficult, and we will not attempt to fully characterize the $\operatorname{des}$-Wilf equivalence classes of such patterns. In what follows, we merely wish to present a step toward understanding these in more depth, but first we need two more definitions.
If $A \subseteq [n]$, $f \in D_4$, and $\sigma \in {\mathfrak{S}}_A$, then we let $f^A(\sigma)$ denote the unique element of ${\mathfrak{S}}_A$ whose standardization is $f(\operatorname{std}(\sigma))$. We say that $f^A$ is a dihedral action *relative to $A$*. As a simple example, if $7461 \in {\mathfrak{S}}_{\{1,4,6,7\}}$, then $\operatorname{std}(7461) = 4231$ and $R_{90}^{\{1,4,6,7\}}(\sigma) = 1647$.
We have $${ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0,1/1,1/2,1/3,1/4,2/0,2/1,2/2,2/3,2/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {3,1,2,4}}+0.99,\xintNthElt{0}{\xintCSVtoList {3,1,2,4}}+0.99);
}
{\foreach[count=\x] \y in {3,1,2,4}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {3,1,2,4}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}
{\overset{\operatorname{des}}{\equiv}}{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0,1/1,1/2,1/3,1/4,2/0,2/1,2/2,2/3,2/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {2,3,1,4}}+0.99,\xintNthElt{0}{\xintCSVtoList {2,3,1,4}}+0.99);
}
{\foreach[count=\x] \y in {2,3,1,4}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {2,3,1,4}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}
{\overset{\operatorname{des}}{\equiv}}{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0,1/1,1/2,1/3,1/4,2/0,2/1,2/2,2/3,2/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {2,4,1,3}}+0.99,\xintNthElt{0}{\xintCSVtoList {2,4,1,3}}+0.99);
}
{\foreach[count=\x] \y in {2,4,1,3}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {2,4,1,3}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$$
First consider $(\pi_1,M_1) = { \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0,1/1,1/2,1/3,1/4,2/0,2/1,2/2,2/3,2/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {3,1,2,4}}+0.99,\xintNthElt{0}{\xintCSVtoList {3,1,2,4}}+0.99);
}
{\foreach[count=\x] \y in {3,1,2,4}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {3,1,2,4}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$ and $(\pi_2,M_2) = { \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0,1/1,1/2,1/3,1/4,2/0,2/1,2/2,2/3,2/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {2,3,1,4}}+0.99,\xintNthElt{0}{\xintCSVtoList {2,3,1,4}}+0.99);
}
{\foreach[count=\x] \y in {2,3,1,4}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {2,3,1,4}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$. To prove their $\operatorname{des}$-Wilf equivalence, we will form a $\operatorname{des}$-preserving bijection $$\alpha: {\mathfrak{S}}_n \setminus \operatorname{Av}_n((\pi_1,M_1)) \to {\mathfrak{S}}_n \setminus \operatorname{Av}_n((\pi_2,M_2)),$$ that is, a $\operatorname{des}$-preserving bijection between permutations in ${\mathfrak{S}}_n$ containing $(\pi_1,M_1)$ and those containing $(\pi_2,M_2)$.
Suppose $\sigma = a_1\dots a_n \in {\mathfrak{S}}_n$ contains $(\pi_1,M_1)$. If $\sigma$ contains $(\pi_2,M_2)$, then set $\alpha(\sigma) = \sigma$. Otherwise, let $j$ be the smallest index in which an occurrence of $(\pi_1,M_1)$ begins, and consider $a_ia_{i+1}\dots a_p$, where $$p = \min \{ m \mid m > j+2,\, a_m >a_j\} \text{ and } i = \min \{ m \mid m \leq j,\, a_m,a_{m+1},\dots,a_j < a_p\}.$$
Let $A = \{a_i, a_{i+1}, \dots, a_p\}$, and set $$R_{180}^A(a_i\dots a_p) = b_i\dots b_p,$$ and further set $$\alpha(\sigma) = a_1\dots a_{i-1}b_i\dots b_pa_{p+1}\dots a_n.$$
Since $R_{180}^A$ is aa $\operatorname{des}$-preserving map, we have that for any $k \in \{1,\dots, p-1-i\}$, $i + k \in \operatorname{Des}(\sigma)$ if and only if $p-k \in \operatorname{Des}(\alpha(\sigma))$. Additionally, for any $k \in \{1,\dots,i-1,p,p+1,\dots,n-1\}$, $k \in \operatorname{Des}(\sigma)$ if and only if $k \in \operatorname{Des}(\alpha(\sigma))$. Thus, $\alpha$ is $\operatorname{des}$-preserving.
To show that $\alpha$ is invertible, we will construct a map $$\beta: {\mathfrak{S}}_n \setminus \operatorname{Av}_n((\pi_2,M_2)) \to {\mathfrak{S}}_n \setminus \operatorname{Av}_n((\pi_1,M_1))$$ and show that $\beta \circ \alpha$ is the identity map on ${\mathfrak{S}}_n \setminus \operatorname{Av}_n((\pi_1,M_1))$. If $\sigma' = a_1'\dots a_n'$ contains $(\pi_2,M_2)$, then we create $\beta(\sigma)$ by first testing a construction similar to the one from the previous paragraph. Namely, let $j'$ be the smallest index in which an occurrence of $(\pi_2,M_2)$ begins, and consider $a_i'a_{i+1}'\dots a_p'$, where $$p' = \min \{ m \mid m > j'+2,\, a_m' >a_{j'+1}'\} \text{ and } i' = \min \{ m \mid m \leq j',\, a_m',a_{m+1}',\dots,a_{j'} < a_p'\}.$$ This time, let $A' = \{a_i', a_{i+1}', \dots, a_p'\}$, and set $$R_{180}^{A'}(a_i'\dots a_p') = b_i'\dots b_p'.$$ If $a_1'\dots a_{i-1}'b_i'\dots b_{p'}'a_{p'+1}'\dots a_n'$ contains both $(\pi_2,M_2)$ and $(\pi_1,M_1)$, then set $\beta(\sigma') = \sigma'$. Otherwise, set $\beta(\sigma') = a_1'\dots a_{i-1}'b_i'\dots b_{p'}'a_{p'+1}'\dots a_n'$. The fact that $\beta \circ \alpha$ is the identity map on ${\mathfrak{S}}_n \setminus \operatorname{Av}_n((\pi_1,M_1))$ follows from construction.
Now consider $(\pi_2,M_2)$ and $(\pi_3,M_3) = { \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0,1/1,1/2,1/3,1/4,2/0,2/1,2/2,2/3,2/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {2,4,1,3}}+0.99,\xintNthElt{0}{\xintCSVtoList {2,4,1,3}}+0.99);
}
{\foreach[count=\x] \y in {2,4,1,3}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {2,4,1,3}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$. Suppose $\sigma=a_1a_2\cdots a_n$ and $a_ja_{j+1}a_{j+2}a_p$ is the first copy of $(\pi_3,M_3)$, as identified in the second paragraph in this proof. If $a_p$ is the only $a_l$ for which $l > j+2$ and $a_l>a_j$, then set $\alpha(\sigma)$ to be $\sigma$ with $a_{j+1}$ and $a_p$ transposed. Otherwise, choose $$r = \min \{ l \mid a_j<a_l<a_{j+1},\,l>j+2 \}.$$ Let $S=\{ a_r,a_{r+1},\dots,a_q\}$ where $q$ is the maximum index for which $\{a_r,a_{r+1},\dots,a_q\}$ is increasing and $a_j < a_k < a_{j+1}$ for all $k \in S$. Set $\alpha(\sigma)$ to be $\sigma$ with $a_{j+1}$ and $\max S$ transposed. By choosing the maximum of $S$ we are guaranteeing that $\alpha$ is $\operatorname{des}$-preserving. By construction, $\alpha(\sigma)$ contains an occurrence of $(\pi_2,M_2)$. Using an argument similar to the first part of this proof, $\alpha$ is invertible and is therefore a bijection.
Recall that the *Stirling numbers of the second kind*, denoted $S(n,k)$, record the number of ways to partition $[n]$ into $k$ nonempty blocks. Here, we will begin to find useful the notation $$\operatorname{Av}_n^{\operatorname{des},k}(\Pi) = \{ \sigma \in \operatorname{Av}_n(\Pi) \mid \operatorname{des}(\sigma) = k\}.$$
Let $(\pi,M) = { \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {2/0}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$. For all $n$, we have $$F_n^{\operatorname{des}}\left((\pi,M); q\right) = \sum_{k = 0}^{n-1} S(n,k+1)q^k.$$
Let $\Sigma_{n,k}$ denote the collection of set partitions of $[n]$ into exactly $k$ nonempty blocks. We will create a bijection $$f: \operatorname{Av}_n^{\operatorname{des},k}((\pi,M)) \to \Sigma_{n,k+1},$$ from which the conclusion follows.
First, let $\sigma = a_1\dots a_n \in \operatorname{Av}_n^{\operatorname{des},k}((\pi,M))$. It follows from [@BursteinLankham Theorem 4.1] that any such permutation is the concatenation of substrings $$\begin{aligned}
a_1 &< \dots <& a_{i_0} \\
a_{i_0+1} &< \dots <& a_{i_1} \\
& \vdots & \\
a_{i_k + 1} &< \dots <& a_n,
\end{aligned}$$ where $a_1 < a_{i_j+1} > a_{i_{j+1}+1} $ for all $j$. In particular, the values $a_{i_0},\dots,a_{i_k}$ determine the entire permutation.
Associate to $\sigma$ the set partition $$f(\sigma) = \{\{a_1, \dots, a_{i_0}\}, \{a_{i_0+1}, \dots, a_{i_1}\}, \dots, \{a_{i_k + 1}, \dots, a_n\}\}.$$ Note that if $\sigma = 12\dots n$, then $k = 0$, so this partition consists of only one block. Thus, if $\sigma$ has $k$ descents, then the partition obtained has $k+1$ blocks. Because each choice of the $a_{i_j}$ determines $\sigma$, we know that $f(\sigma) \neq f(\sigma')$ whenever $\sigma' \in \operatorname{Av}_n^{\operatorname{des},k}((\pi,M))$ and $\sigma \neq \sigma'$. That is, $f$ is injective.
Now we will show that $f$ is surjective. Consider a set partition $B = \{B_1,\dots,B_{k+1}\}$ of $[n]$ into $k+1$ blocks. We are free to write the $B_i$ such that $$B_i = \{b_{i,1} < \dots < b_{i,i_l}\} \text{ and } \min B_i < \min B_{i+1}$$ for all $i$. Construct the permutation $$b_{k+1,1}b_{k+1,2}\dots b_{k+1,i_{k+1}}b_{k,1}b_{k,2}\dots b_{k,i_k}\dots b_{1,1}b_{1,2}\dots b_{1,i_1}.$$ We claim that this permutation is an element of $\operatorname{Av}_n^{\operatorname{des},k}((\pi,M))$.
Any occurrence of ${ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$, say, $b_{\alpha}b_{\beta}b_{\gamma}$, implies that $b_{\alpha} \in B_i$, $b_{\beta} \in B_j$, and $b_{\gamma} \in B_k$ for some $i \leq j < k$. Since the sequence of minima of the blocks is decreasing, we know that $\min B_k < b_{\alpha} < b_{\gamma}$. Thus, the string $$b_{\alpha}b_{\beta}(\min B_k)b_{\gamma}$$ is an occurrence of ${ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {2/0}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$. Since the elements of the blocks strictly increase, the minima decrease, and since there are $k+1$ blocks, there are $k$ descents in the permutation. Thus $f$ is surjective, completing the proof.
Consider the permutation $3427156 \in \operatorname{Av}_6^{\operatorname{des},2}\left({ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {2/0}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}\right)$. Our construction in the previous proof associates to this permutation the partition $$\{\{3,4\},\{2,7\},\{1,5,6\}\}.$$ In the other direction, given the set partition $$\{\{5\},\{3,1,4\},\{7,2,6\}\} = \{\{5\},\{2,6,7\},\{1,3,4\}\},$$ we obtain the permutation $5267134$, which the reader may verify is indeed an element of $\operatorname{Av}_7^{\operatorname{des},2}\left({ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {2/0}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}\right)$.
A *Motzkin path of length $n$* is a lattice path from $(0,0)$ to $(n,0)$ using only *up-steps* $(1,1)$, *down-steps* $(1,-1)$, and *horizontal steps* $(1,0)$ such that the path does not go below the $x$-axis. An example is shown in Figure \[fig: Motzkin path\]. We let $\mathcal{M}_{n,k}$ denote the set of Motzkin paths of length $n$ with exactly $k$ up-steps.
The next result we present was first proven in [@ChenDengYang] by writing Motzkin paths according to a “strip decomposition” and by writing permutations according to canonical reduced decompositions. Here, we present a new, simpler proof. To do so, we only need a few more definitions.
(0,0) – (10,0); (A) at (0,0) ; (B) at (1,0) ; (C) at (2,1) ; (D) at (3,2) ; (E) at (4,1) ; (F) at (5,1) ; (G) at (6,0) ; (H) at (7,0) ; (I) at (8,1) ; (J) at (9,1) ; (K) at (10,0) ;
\(A) – (B) – (C) – (D) – (E) – (F) – (G) – (H) – (I) – (J) – (K);
in [(0,0),(1,0),(2,1),(3,2),(4,1),(5,1),(6,0),(7,0),(8,1),(9,1),(10,0)]{} [ circle \[radius=3pt\]; ]{}
If $i$ is a descent of $\sigma = a_1\dots a_n$, then we call $a_i$ a *descent top* and $a_{i+1}$ a *descent bottom*. Let $\operatorname{Destop}(\sigma)$ denote the set of descent tops of $\sigma$ and let $\operatorname{Desbot}(\sigma)$ denote the set of descent tops of $\sigma$. A *valley* in $\sigma$ is an element $i$ for which $a_{i-1} > a_i < a_{i+1}$.
Let $\Pi = \left\{{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {3,2,1}}+0.99,\xintNthElt{0}{\xintCSVtoList {3,2,1}}+0.99);
}
{\foreach[count=\x] \y in {3,2,1}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {3,2,1}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}, { \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {2,3,1}}+0.99,\xintNthElt{0}{\xintCSVtoList {2,3,1}}+0.99);
}
{\foreach[count=\x] \y in {2,3,1}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {2,3,1}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}\right\}$. For all $n$, $$F_n^{\operatorname{des}}(\Pi; q) = \sum_{k=0}^n |\mathcal{M}_{n,k}|q^k$$
We will form a bijection $$\mu: \operatorname{Av}_n^{\operatorname{des},k}\left(\Pi\right) \to \mathcal{M}_{n,k}.$$ For $\sigma = a_1\dots a_n \in \operatorname{Av}_n^{\operatorname{des},k}\left(\Pi\right)$, let $\mu(\sigma)$ be the lattice path obtained by making step $a_i$ a down-step if $a_i$ is a descent bottom, an up-step if $a_i$ is a descent top, and a horizontal step if $a_i$ is neither.
First, we need to check that $\mu$ is well-defined. Note that no letter of $\sigma$ can be both a descent top and a descent bottom, since this would imply $\sigma$ contains an instance of ${ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {3,2,1}}+0.99,\xintNthElt{0}{\xintCSVtoList {3,2,1}}+0.99);
}
{\foreach[count=\x] \y in {3,2,1}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {3,2,1}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$, which is forbidden. So, since the sets of descent tops and of descent bottoms are disjoint, and these appear in pairs, then we can be certain that the path constructed by $\mu$ has length $n$ and ends at $(n,0)$. Moreover, since a descent top always appears before a descent bottom, at no step of the path can there have been more down-steps than up-steps. This establishes that $\mu(\sigma)$ is a Motzkin path of length $n$. Finally, since there are $k$ descents, there are $k$ descent tops, and $\mu(\sigma)$ will have $k$ up-steps. Hence, $\mu(\sigma) \in \mathcal{M}_{n,k}$.
Next we will show that $\mu$ is injective. To do so, we will determine exactly the structure of the elements in $\operatorname{Av}_n\left(\Pi\right)$. Notice that the descent bottoms of $\sigma$ must appear in increasing order in $\sigma$, since, otherwise, there would be an occurrence of ${ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {3,2,1}}+0.99,\xintNthElt{0}{\xintCSVtoList {3,2,1}}+0.99);
}
{\foreach[count=\x] \y in {3,2,1}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {3,2,1}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$. For the same reason, the descent tops must appear in increasing order in $\sigma$.
Let $\sigma = a_1\dots a_n \in \operatorname{Av}_n^{\operatorname{des},k}\left(\Pi\right)$ and suppose that $i$ is neither a descent top nor descent bottom. Suppose for now that $j$ is the first descent greater than $i$. If $a_{j+1} < a_i < a_j$, then $a_ia_ja_{j+1}$ is an occurrence of $231$. Since $\sigma$ avoids ${ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {2,3,1}}+0.99,\xintNthElt{0}{\xintCSVtoList {2,3,1}}+0.99);
}
{\foreach[count=\x] \y in {2,3,1}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {2,3,1}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}$, there must be some $l$ for which $\sigma$ has the subsequence $a_ia_la_ja_{j+1}$ and $a_l < a_{j+1}$. This implies that some integer $i+1,i+2,\dots,l-1$ is a descent, which contradicts the fact that $j$ is the first descent greater than $i$. So, it must be true that $a_i < a_{j+1} < a_j$. Since $j$ is the first descent greater than $i$, it follows that $a_ia_{i+1}\dots a_{j-1}a_{j+1}$ is an increasing sequence. It follows that the subsequence of $\sigma$ consisting of all letters that are not descent tops is an increasing sequence.
Now we will show that $\mu$ is injective. If $\mu(\sigma_1) = \mu(\sigma_2)$ for $\sigma_1,\sigma_2 \in \operatorname{Av}_n\left(\Pi\right)$, then $\sigma_1$ and $\sigma_2$ have the same descent topsets and the same descent bottomsets, since these are identified by the up-steps and down-steps in the Motzkin path. Our description of elements of $\operatorname{Av}_n\left(\Pi\right)$ shows that once the descent topsets and descent bottomsets have been identified, there is a unique $\sigma$ in the avoidance class with those sets. Therefore, $\sigma_1 = \sigma_2$, and $\mu$ is injective.
Finally, we will show that $\mu$ is surjective. Let $A \in \mathcal{M}_{n,k}$, and label its steps $1,\dots,n$ from left to right. We will construct its preimage in stages. First write down $1,\dots,n$ but excluding the labels on the down-steps. Then insert the label on the $i^{th}$ down-step immediately before the label of the $i^{th}$ up-step. Call the resulting permutation $\sigma_A$. Using the description of elements of $\operatorname{Av}_n\left(\Pi\right)$ from earlier in this proof, we see that $\sigma_A \in \operatorname{Av}_n\left(\Pi\right)$. Additionally, it is clear that $\mu(\sigma_A) = A$ by our construction of $\sigma_A$ and the definition of $\mu$. Therefore, $\mu$ is surjective, completing the proof.
Let $A$ be the Motzkin path in Figure \[fig: Motzkin path\]. Steps $2$, $3$, and $8$ are up-steps, and therefore will be descents bottoms. Steps $4$, $6$, and $10$ are down-steps, so these will be descent tops. The remaining numbers will be neither descent tops nor bottoms.
When the descent tops are removed from $\mu^{-1}(A)$, the result will be an increasing string of numbers: $1235789$. The descent tops are then placed immediately preceding the descent bottoms, to obtain $1426357(10)89$.
For the final result of the section, we make two notes. First, recall that the *Eulerian polynomial* $A_n(q)$ is the polynomial $$\sum_{\sigma \in {\mathfrak{S}}_n} q^{\operatorname{des}(\sigma)} = A_n(q).$$ It should be noted that some authors (e.g. in [@StanleyVol1]) define the Eulerian polynomials using $q^{\operatorname{des}(\sigma) + 1}$ rather than the definition given here. So, one should take care when encountering Eulerian polynomials in the literature. Second, recall from the end of Section $2$ what it means for a permutation to contain and avoid the barred patterns $\bar1\bar243$ or $\bar132\bar4$.
\[prop:barred prop\] For all $n$, $$F_n^{\operatorname{des}}(\bar1\bar243; q) = F_n^{\operatorname{des}}(\bar132\bar4; q) =
\begin{cases}
1 & \text{ if } n = 0,1 \\
A_{n-2}(q) & \text{ if } n \geq 2
\end{cases}$$
We will first show that $F_n(\bar1\bar243;q)$ satisfies the right hand side. The conclusion is clearly true for $n < 2$, so we will restrict our attention to when $n \geq 2$. Choose $\sigma = a_1\dots a_n \in \operatorname{Av}_n(\bar1\bar243)$. Note first that $a_1 < a_2$ since, if $a_1 > a_2$, then $a_1a_2$ would be an occurrence of $u(\bar1\bar243) = 21$ but this cannot extend to an occurrence of $1243$.
Now, suppose $a_2 > 2$. Setting $a_m = \min\{a_i \mid 3 \leq i \leq n \}$we have $a_2 > a_m$, so $a_2a_m$ is an occurrence of $u(\bar1\bar243)$ in $\sigma$. However, there is only letter to the left of $a_2$, so this pattern does not extend to an instance of $1243$. Thus, $a_2 = 2$. Together with the previous paragraph, we know $a_1 = 1$ as well. In particular, $a_1 < a_2 < a_i$ for all $i \geq 3$.
Now, take any occurrence $a_ia_j$ of $21$ in which $2 < i < j$. Clearly, $a_1a_2a_ia_j$ is an extension to $1243$. This holds for any possible permutation of $3,\dots,n$ as the final $n-2$ letters. Since $1$ and $2$ are never descents of these permutations, we have $$F_n^{\operatorname{des}}(\bar1\bar243;q) = A_{n-2}(q),$$ as claimed.
Now we will show that the same formula holds for $\bar132\bar4$. This time, assume $\sigma \in \operatorname{Av}_n(\bar132\bar4)$. If $a_i = 1$ for some $i > 1$, then $a_1a_i$ would be an occurrence of $21$. However, this can never extend to $1324$ since there is no letter to the left of $a_1$. Thus, $a_1 = 1$. An analogous argument shows $a_n = n$.
This allows $a_2\dots a_{n-1}$ to be any arrangement of $2,3,\dots,n-1$, since, whenever $a_ia_j$ is an occurrence of $21$ for $2 \leq i,j \leq n-1$, this extends to $1a_ia_jn$. So, we have the bijection $$a_1a_2\dots a_n \mapsto (a_2-1)(a_3-1)\dots(a_{n-1}-1)$$ with elements of ${\mathfrak{S}}_{n-2}$. Since $1$ and $n$ are never descents in $\operatorname{Av}_n(\bar132\bar4)$, this is a $\operatorname{des}$-preserving bijection. Therefore, $F_n^{\operatorname{des}}(\bar1\bar243;q) = A_{n-2}(q)$.
Conjectures and Further Directions
==================================
In this section, we provide a few conjectures, supporting data, and additional direction in which this work could proceed. In all cases, no closed forms for the functions $F_n^{\operatorname{des}}(\Pi;q)$ are known. We refer the reader to Table \[tab:data\] for all known polynomials $F_n^{\operatorname{des}}(\Pi;q)$ for $4 \leq n \leq 8$, since, for these choices of $\Pi$, $F_n^{\operatorname{des}}(\Pi;q) = F_n^{\operatorname{des}}(\emptyset;q)$ for $n \leq 3$.
The following $\operatorname{des}$-Wilf equivalences hold: $${ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0,1/1,1/2,1/3,1/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,2,4,3}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,2,4,3}}+0.99);
}
{\foreach[count=\x] \y in {1,2,4,3}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,2,4,3}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}
{\overset{\operatorname{des}}{\equiv}}{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0,1/1,1/2,1/3,1/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {3,4,1,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {3,4,1,2}}+0.99);
}
{\foreach[count=\x] \y in {3,4,1,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {3,4,1,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}
\quad\text{and}\quad
{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0,1/1,1/2,1/3,1/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,4,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,4,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,4,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,4,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}
{\overset{\operatorname{des}}{\equiv}}{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0,1/1,1/2,1/3,1/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {2,4,1,3}}+0.99,\xintNthElt{0}{\xintCSVtoList {2,4,1,3}}+0.99);
}
{\foreach[count=\x] \y in {2,4,1,3}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {2,4,1,3}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}.$$
To state our next conjecture, we must discuss a particular sorting of permutations. Let $\sigma = a_1\dots a_n \in {\mathfrak{S}}_n$ and suppose $a_i = n$. Let $\Gamma$ be the operator defined recursively as $$\Gamma(\sigma) = \Gamma(a_1\dots a_{i-1})\Gamma(a_{i+1}\dots a_n)n.$$ We say that $\sigma$ is *West-$t$-stack-sortable* if $\Gamma^t(\sigma)$ is the identity permutation. Note that the $2$-West-stack-sortable permutations [@WestThesis] are exactly those in $$\operatorname{Av}_n\left({ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {2,3,4,1}}+0.99,\xintNthElt{0}{\xintCSVtoList {2,3,4,1}}+0.99);
}
{\foreach[count=\x] \y in {2,3,4,1}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {2,3,4,1}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
},{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {3,2,4,1}}+0.99,\xintNthElt{0}{\xintCSVtoList {3,2,4,1}}+0.99);
}
{\foreach[count=\x] \y in {3,2,4,1}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {3,2,4,1}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}\right).$$
The following $\operatorname{des}$-Wilf equivalence holds: $$\left\{
{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {2,3,4,1}}+0.99,\xintNthElt{0}{\xintCSVtoList {2,3,4,1}}+0.99);
}
{\foreach[count=\x] \y in {2,3,4,1}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {2,3,4,1}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
},
{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {3,2,4,1}}+0.99,\xintNthElt{0}{\xintCSVtoList {3,2,4,1}}+0.99);
}
{\foreach[count=\x] \y in {3,2,4,1}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {3,2,4,1}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}
\right\}
{\overset{\operatorname{des}}{\equiv}}\left\{
{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {2,4,1,3}}+0.99,\xintNthElt{0}{\xintCSVtoList {2,4,1,3}}+0.99);
}
{\foreach[count=\x] \y in {2,4,1,3}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {2,4,1,3}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
},
{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {2/2}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {3,1,4,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {3,1,4,2}}+0.99);
}
{\foreach[count=\x] \y in {3,1,4,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {3,1,4,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}
\right\}$$
If this conjecture is true, then from [@Bona02] it follows that $$\begin{aligned}
F_n^{\operatorname{des}}\left(\left\{
{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {2,3,4,1}}+0.99,\xintNthElt{0}{\xintCSVtoList {2,3,4,1}}+0.99);
}
{\foreach[count=\x] \y in {2,3,4,1}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {2,3,4,1}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
},
{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {3,2,4,1}}+0.99,\xintNthElt{0}{\xintCSVtoList {3,2,4,1}}+0.99);
}
{\foreach[count=\x] \y in {3,2,4,1}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {3,2,4,1}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}
\right\};q\right) &=& F_n^{\operatorname{des}}\left(\left\{
{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {2,4,1,3}}+0.99,\xintNthElt{0}{\xintCSVtoList {2,4,1,3}}+0.99);
}
{\foreach[count=\x] \y in {2,4,1,3}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {2,4,1,3}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
},
{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {2/2}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {3,1,4,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {3,1,4,2}}+0.99);
}
{\foreach[count=\x] \y in {3,1,4,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {3,1,4,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}
\right\};q\right) \\
&=& \sum_{k=0}^{n-1} \frac{(n+k)!(2n-k-1)!}{(k+1)!(n - k)!(2k + 1)!(2n - 2k - 1)!}q^k.\end{aligned}$$
$$\begin{array}{|c|c|c|}\hline \relax
\Pi & n & F_n^{\operatorname{des}}(\Pi;q) \\ \hline \relax
\multirow{2}{*}{$\left\{{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0,1/1,1/2,1/3,1/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,2,4,3}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,2,4,3}}+0.99);
}
{\foreach[count=\x] \y in {1,2,4,3}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,2,4,3}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}\right\},\left\{{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0,1/1,1/2,1/3,1/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {3,4,1,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {3,4,1,2}}+0.99);
}
{\foreach[count=\x] \y in {3,4,1,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {3,4,1,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}\right\}$} & 4 & 1+10q+ 11q^2 +q^3 \\
& 5 & 1+ 20q+ 57q^2 +26q^3 +q^4 \\
& 6 & 1+ 35q + 204q^2 + 252q^3 + 57q^4 + q^5 \\
& 7 & 1+ 56q+ 581q^2+ 1500q^3+ 969q^4+ 120q^5+ q^6 \\
& 8 & 1+84q+ 1414q^2 + 6588q^3 + 9117q^4 + 3426q^5+ 247q^6+ q^7 \\ \hline \relax
\multirow{2}{*}{$\left\{{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0,1/1,1/2,1/3,1/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {1,3,4,2}}+0.99,\xintNthElt{0}{\xintCSVtoList {1,3,4,2}}+0.99);
}
{\foreach[count=\x] \y in {1,3,4,2}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {1,3,4,2}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}\right\},\left\{{ \raisebox{0.6ex}{
\begin{tikzpicture}[scale={0.25}, baseline=(current bounding box.center)]
{
\foreach \x/\y in {1/0,1/1,1/2,1/3,1/4}
\fill[pattern=north east lines, pattern color=black!75] (\x,\y) rectangle +(1,1);
}
{ \pgfmathsetmacro\circlesize{8+4}
{\draw (0.01,0.01) grid (\xintNthElt{0}{\xintCSVtoList {2,4,1,3}}+0.99,\xintNthElt{0}{\xintCSVtoList {2,4,1,3}}+0.99);
}
{\foreach[count=\x] \y in {2,4,1,3}
\filldraw (\x,\y) circle (8 pt);
}
\foreach \x in {}
\draw (\x,\xintNthElt{\x}{\xintCSVtoList {2,4,1,3}}) circle (\circlesize pt);
}
\end{tikzpicture}}\;
}\right\}$} & 4 & 1 + 10q+ 11q^2 + q^3 \\
& 5 & 1+ 20q+ 56q^2+ 26q^3+ q^4 \\
& 6 & 1+ 35q+ 196q^2+ 241q^3+ 57q^4+ q^5\\
& 7 & 1+ 56q+ 546q^2+ 1361q^3+ 897q^4+ 120q^5 +q^6 \\
& 8 & 1+ 84q+ 1302q^2+ 5675q^3+ 7739q^4+ 3060q^5+ 247q^6+ q^7 \\ \hline
\end{array}$$
Instead of generalizing the patterns being avoided, one may generalize permutations themselves. One way to do this is to consider the *colored permutations* $$G_{r,n} := \{({\varepsilon},\sigma) \mid {\varepsilon}\in {\mathbb{Z}}_r,\, \sigma \in {\mathfrak{S}}_n\}.$$ In this case, we say that $({\varepsilon},\sigma) \in G_{r,n}$ *contains* $(\zeta, \pi) \in G_{s,m}$ if there are elements $1 \leq i_1 < i_2 < \dots < i_s \leq n$ such that $\operatorname{std}(\sigma_{i_1}\dots\sigma_{i_s}) = \pi$ and ${\varepsilon}_{i_j} = \zeta_j$ for all $j$. If no such choice of $i_j$ exist, then we say $({\varepsilon},\sigma)$ *avoids* $(\zeta,\pi)$. For a set of colored permutations $\Pi$, let $$\operatorname{Av}_{r,n}(\Pi) = \{({\varepsilon},\sigma) \in G_{r,n} \mid ({\varepsilon},\sigma) \text{ avoids all } (\zeta,\pi) \in \Pi\}.$$
What can be said about the polynomials $$F_{r,n}^{\operatorname{st}}(\Pi;q) = \sum_{({\varepsilon},\sigma) \in \operatorname{Av}_{r,n}(\Pi)} q^{\operatorname{st}({\varepsilon},\sigma)}?$$
We close by noting that $G_{r,n}$ is the set of elements in the wreath product ${\mathbb{Z}}_r \wr {\mathfrak{S}}_n$, a fact which may be useful when addressing the above questions.
[^1]: The fourth author was supported in part by the Michigan State University Discovering America exchange program.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Independent component analysis (ICA) is a powerful method for blind source separation based on the assumption that sources are statistically independent. Though ICA has proven useful and has been employed in many applications, complete statistical independence can be too restrictive an assumption in practice. Additionally, important prior information about the data, such as sparsity, is usually available. Sparsity is a natural property of the data, a form of diversity, which, if incorporated into the ICA model, can relax the independence assumption, resulting in an improvement in the overall separation performance. In this work, we propose a new variant of ICA by entropy bound minimization (ICA-EBM)—a flexible, yet parameter-free algorithm—through the direct exploitation of sparsity. Using this new SparseICA-EBM algorithm, we study the synergy of independence and sparsity through simulations on synthetic as well as functional magnetic resonance imaging (fMRI)-like data.'
address: |
$^1$University of Maryland, Baltimore County, Dept. of Mathematics and Statistics, Baltimore, MD 21250\
$^2$University of Maryland, Baltimore County, Dept. of CSEE, Baltimore, MD 21250
bibliography:
- './ReferencesICASSP17.bib'
title: |
Enhancing ICA performance by exploiting sparsity: Application to\
fMRI Analysis
---
independent component analysis, sparsity, fMRI
Introduction {#sec:intro}
============
Independent component analysis (ICA) is a data-driven method that provides a unique decomposition of a dataset solely through the assumption that sources are statistically independent and has found wide use in a variety of applications. Although statistical independence is a natural assumption in many cases, there are many practical applications where such a strong assumption is unrealistic. Often in these cases, some important prior information about the data is available and incorporating it into the ICA model will result in better overall separation performance.
A widely used approach for incorporating prior information into the ICA framework is through the use of constrained independent component analysis (C-ICA) [@lu2005approach], which incorporates prior information using equality and inequality constraints under a Lagrangian framework. Such prior information can be about the task in functional magnetic resonance imaging fMRI analysis and can be included as constraints on the mixing matrix columns [@calhoun2005semi; @6960099; @wang2011fixed] or spatial maps [@lin2010semiblind; @soldati2013ica; @lu2005approach]. While this approach is practical, such constraints have to be in an exact functional form, something that is not always available in practice. Another form of prior information that can be considered are natural properties of the data, such as sparsity.
There are many ways to impose sparsity into the ICA model, such as by selecting a density model that favors sparse distributions [@bell1995information; @hyvarinen1999fast] or by using sparsity transformations [@5495320] following ICA. Although, selecting the source distribution would allow the ICA model to enjoy the desirable large sample properties of the ML formulation [@6784026; @Comon_HandbookBSS_2010], the model would be limited to a specific type of sparse distribution [@6784026]. Additionally, sparsity transformations are an indirect way of imposing sparsity and do not allow direct way to control independence versus sparsity.
The main contribution of this work is the development of a new ICA algorithm that takes the sparsity of each individual source into account. We incorporate sparsity into the ICA algorithm, entropy bound minimization (ICA-EBM) [@li2010independent], by introducing a weighting factor to the ICA cost function to balance the contribution of sparsity for each of the individual sources. ICA-EBM is a flexible and parameter-free ICA algorithm that can separate sources from a wide range of distributions. The new SparseICA-EBM algorithm inherits all the advantages of ICA-EBM, namely its flexibility, though with enhanced performance due to the exploitation of the sparsity and allows the user to balance the roles of independence and sparsity.
The remainder of this paper is organized as follows. In Section 2, we provide the necessary background on ICA. Section 3, provides the mathematical development of SparseICA-EBM. In Section 4, we demonstrate the effectiveness of SpaseICA-EBM through sparse simulated data as well as simulated fMRI-like data. The conclusions are presented in Section 5.
Background {#sec:backgr}
==========
Independent Component Analysis
------------------------------
Let $N$ statistically independent sources ${\bf s}(t) = [ s_1(t),\dots,s_N(t)]^{\top}$ be mixed through an unknown invertible mixing matrix ${\bf A}\in \mathbb{R}^{N\times N}$ so that we obtain mixtures ${\bf x}(t) = [ x_1(t),\dots,x_N(t)]^{\top}$, through the linear model $$\nonumber
{\bf x}(t) = {\bf A}{\bf s}(t),~~~ t=1,\dots,T,$$ where $t$ denotes the discrete time index and $(\cdot)^{\top}$ the transpose. The goal of ICA is to estimate a demixing matrix ${\bf W}\in \mathbb{R}^{N\times N}$ to yield maximally independent source estimates ${\bf y}(t) = {\bf W}{\bf x}(t)$. A natural cost function to achieve such a separation is mutual information (MI), which is defined as the Kullback-Leibler (KL)-distance between the joint source density and the product of the marginal estimated source densities. Therefore, the MI cost function is given by $$\label{ICAmutualInfo}
J_{ICA}({\bf W}) = \sum_{n=1}^N H(y_n) - \log|\det({\bf W})| - H({\bf x}),$$ where $y_n = {\bf w}_n^{\top} {\bf x}$ and the terms $H(y_n)$ and $H({\bf x})$ are the (differential) entropy of the source estimates and the mixtures, respectively. Note that the term $H({\bf x})$ is independent of ${\bf W}$ and can be treated as a constant $C$. The minimization of the MI is equivalent to the maximization of the maximum likelihood (ML) cost function, hence, making available all the theoretical advantages associated with the ML theory [@6784026] for large sample sizes. It is impractical to try to exploit prior information in (\[ICAmutualInfo\]) as it requires either complete knowledge of the demixing matrix or of the sources, information that is not usually available. Assuming that the demixing matrix is orthogonal would loosen this strict requirement, but would unecessarily limit the solution space. Moreover, direct implementation of (\[ICAmutualInfo\]) implies that each latent source has the same distribution, which is unrealistic in many practical applications. All of these issues can be avoided by rewriting (\[ICAmutualInfo\]) and its gradient with respect to each row of ${\bf W}$, ${\bf w}_m,\ m=1,\dots N$. Thus, by using this decoupling approach [@li2007nonorthogonal; @li2010independent], the MI cost function can be written as $$\label{ICAmutualInfoDecoupling}
J_{ICA}({\bf w}_m) = \sum_{n=1}^NH(y_n) - \log\left| {\bf h}_m^{\top} {\bf w}_m \right| - C_m,~m = 1,\dotsc,N,$$ where ${\bf h}_m$ is a unit vector that is perpendicular to all row vectors of ${\bf W}$ except ${\bf w}_m$ and $C_m$ is a constant. The gradient of (\[ICAmutualInfoDecoupling\]) can be written in a decoupled form and is given by $$\label{gradcost}
\frac{\partial J_{ICA}({\bf w}_m)}{\partial {\bf w}_m} = -E\left\{ \phi(y_m) {\bf x} \right\} - \frac{{\bf h}_m}{ {\bf h}_m^{\top} {\bf w}_m},\nonumber$$ where $\phi(y_m) = \frac{{\partial \log p(y_m)}}{{\partial y_m}}$ is called the score function and the probability density function (PDF) of the $m$th estimated source, $p(y_m)$, can be adaptively determined for each estimated source independently.
Mathematical Development and Implementation
===========================================
The formal definition for sparsity is given through the $\ell^0$ norm and is defined as the number of non-zero coefficients from a vector $y\in \mathbb{R}^T$ $$\label{SparsityL0Norm}
||y||_0 = \# \{y_i\neq 0;j=1,\dots T\}.$$ Although the incorporation of (\[SparsityL0Norm\]) into the ICA framework is a direct way to impose sparsity, the $\ell^0$ norm is computationally intractable. Instead, the $\ell^1$ norm, defined as the sum of the absolute values of a vector’s coefficients, serves as a surrogate sparsity regularizer of the $\ell^0$ norm in many optimization problems [@candes2005decoding; @schmidt2007fast; @tibshirani1996regression]. We can promote the synergy between independence and sparsity through the addition of the $\ell^1$ regularization term to (\[ICAmutualInfoDecoupling\]), which we expect to improve separation performance when the underlying sources are truly sparse.
The proposed decoupled sparsity promoting ICA cost function is thus given by $$\label{SparseCostFunction}
J({\bf w}_m) = J_{ICA}({\bf w}_m) + \lambda_m f(y_m),~m = 1,\dotsc,N,$$ where $f(y_m) = ||y_m||_1$ is the regularization term and $\lambda_m$ is called the sparsity parameter. The $\ell^1$ norm is a non-differentiable function, so it is replaced by the the sum of multi-quadratic functions [@lee2006efficient], given by $$f(y_m) = \lim_{\epsilon\to 0} \sum_{t=1}^T\sqrt{y_{m_t}^2 + \epsilon},\nonumber$$ where $\epsilon$ is the smoothing parameter. Therefore, the proposed gradient can be written as $$\begin{aligned}
\frac{\partial}{\partial {\bf w}_m} J({\bf w}_m)&=\frac{\partial}{\partial {\bf w}_m} \left(J_{ICA}({\bf w}_m) + \lambda_m \lim_{\epsilon\to 0} f(y_m)\right)\\ \\
%&= \frac{\partial J_{ICA}({\bf w}_m)}{\partial {\bf w}_m} + \lambda \lim_{\epsilon\to 0}\frac{\partial f(y_m)}{\partial {\bf w}_m}\\ \\
&= \frac{\partial J_{ICA}({\bf w}_m)}{\partial {\bf w}_m} + \lambda_m \lim_{\epsilon\to 0}\sum_{t=1}^T \frac{y_{m_t}}{\sqrt{y_{m_t}^2 + \epsilon}}{\bf x}.\end{aligned}$$
Due to its ability to maximize independence in an efficient manner through the use of four measuring functions favoring bimodal, symmetric or skewed, heavy-tailed or not heavy-tailed distributions [@li2010independent], ICA-EBM serves as the algorithm for the direct integration of (\[SparseCostFunction\]). The new SparseICA-EBM not only provides flexible density matching but also yields solutions with variable levels of sparsity.
Experimental Results {#sec:mathdev}
====================
We demonstrate the performance of SparseICA-EBM (\[SparseCostFunction\]), in terms of its separation power, using simulated sparse sources as well as simulated fMRI-like data. We compare the SparseICA-EBM algorithm with the original ICA-EBM algorithm. Additionally, due to its popularity in many applications including fMRI analysis, we also compare SparseICA-EBM with two implementations of the Infomax algorithm [@bell1995information]. One version is based on the natural gradient optimization framework (Infomax-NG) and the other one is based on a quasi-Newton technique Broyden, Fletcher, Goldfarb, and Shanno (BFGS) [@nielsen2000ucminf], which we call Infomax-BFGS. The hardware used in the computational studies is part of the UMBC High Performance Computing Facility (HPCF), for more information see hpcf.umbc.edu.
Simulated Sparse Sources
------------------------
For the first set of experiments, we generate 20 simulated sources, each of which is distributed according to a generalized Gaussian distribution (GGD) with sample size $T=10^3$. The PDF of each source is given by [@nadarajah2005generalized] $$p(x;\beta, \sigma) = \eta\exp\left( -\frac{x}{2\sigma} \right)^{2\beta},\ x\in\mathbb{R} \nonumber$$ where $\eta = \frac{\beta}{2^{\frac{1}{2\beta}}\Gamma(\frac{1}{2\beta})\sigma}$. The shape parameter, $\beta$, controls the peakedness and spread of the distribution as well as its sparsity. If $\beta<1$, the distribution is more peaky than the Gaussian with heavier tails, and if $\beta > 1$, it is less peaky with lighter tails. Thus, as $\beta\to 0$ the distribution becomes more sparse.
To verify the sparse nature of the sources used for the first set of the experiments, we generate 20 sources with sample size $T=10^4$ and shape parameter $\beta$ from the range $[0.1,0.5]$ with a step size of $0.05$. For each specific source, we measure the sparsity level using the Gini Index as described in [@hurley2009comparing] and average over the sources that correspond to a specific $\beta$. In Fig. \[GiniIndex\], we see that as we increase $\beta$, sources become less sparse.
![Average Gini Index as a function of the shape parameter, $\beta$. The Gini Index is normalized and 1 corresponds to very sparse sources while 0 to dense sources.[]{data-label="GiniIndex"}](GiniIndex){width="7cm"}
To evaluate the performance of the algorithms, we use the average-interference-to-signal ratio (ISR) as in [@li2007nonorthogonal]. For SparseICA-EBM, the algorithm parameters are $\lambda = 10^4$ and $\epsilon = 10^{-2}$ and are determined based on a grid search selection. All results are the average of 300 independent runs.
In Fig. \[wrtbeta\], we display the normalized ISR as a function of $\beta$. We observe that for small values of $\beta$, [*i.e*]{}., highly sparse case, SparseICA-EBM exhibits better performance. On the other hand, ICA-EBM starts performing better than the other algorithms as we increase $\beta$, [*i.e.*]{}, decrease sparsity. It is worth mentioning that Infomax-NG often fails to converge as $\beta$ increases revealing its poor performance under this experimental setup. On the other hand, Infomax-BFGS shows reasonable performance especially for small values of $\beta$.
![Performance comparison of four ICA algorithms in terms of the normalized average ISR as a function of shape parameter, $\beta$, for 20 sources with $T=10^3$. Each point is the result of 300 independent runs.[]{data-label="wrtbeta"}](300RunsWrtBetaN1000ICASSP2017){width="7.5cm"}
In Fig. \[wrtSamples\], we display the normalized ISR as a function of the sample size. To study the case where sources are very sparse we generate all sources using $\beta=0.1$. As the sample size increases, SparseICA-EBM and ICA-EBM perform better than the other two algorithms, since the large sample size enables an accurate approximation of the differential entropy of the estimated sources. When the sample size becomes greater than $10^3$, Infomax-BFGS starts providing highly inaccurate results, due to algorithmic issues in the approximation of the inverse of the Hessian matrix.
![Performance comparison of four ICA algorithms in terms of the normalized average ISR as a function of sample size, $T$, for 20 sources with $\beta=0.1$. Each point is the result of 300 independent runs.[]{data-label="wrtSamples"}](WrtSampleSizeBeta01N20ICASSP2017){width="7.5cm"}
Finally, in Fig. \[wrtSources\], we display the normalized ISR as a function of the number of sources where for each source $T=10^3$ and $\beta=0.1$. It is clear from Fig. \[wrtSources\] that SparseICA-EBM shows the best performance. Infomax-BFGS performs well when the number of sources is small since the optimization procedure is performed in a low dimensional space. This reveals the benefit of employing the decoupling approach, since the reduction to a set of vector optimization problems avoids over-complicated surfaces for the cost function.
![Performance comparison of four ICA algorithms in terms of the normalized average ISR as a function of number of sources, $N$, with $T=10^3$ and $\beta=0.1$. Each point is the result of 300 independent runs.[]{data-label="wrtSources"}](300RunsWrtSourcesbeta01T1000ICASSP2017){width="7.5cm"}
Simulated fMRI-like Data
------------------------
For the second set of experiments, we used simulated fMRI-like sources at different contrast-to-noise ratio (CNR) levels. Estimated spatial fMRI sources tend to have sparse distributions [@6268324], leading to the popularity of sparsity favoring algorithms such as Infomax. Since Infomax-NG performed poorly for all the experiments using simulated sources, for this set of experiments, we only compare SparseICA-EBM with the original ICA-EBM and Infomax-BFGS. To generate fMRI-like data we use the fMRI toolbox, SimTB [@erhardt2012simtb], which enables flexible generation of fMRI datasets under a model of spatiotemporal separability.
For our experiment, we generate 20 spatial maps using 10 subjects. Each spatial mask is a $100\times100$ image with a baseline intensity of 800. The length of the experiment is 260 samples. Rician noise is added to each dataset at specified CNR value. The parameters for SparseICA-EBM, $\lambda = 0.31$ and $\epsilon = 10^{-1}$, are determined based on a grid search selection method performed on noiseless data. The first step in processing the fMRI like-data consists of the application of principal component analysis to each dataset individually. Since 20 sources are generated for each dataset, the dimension of each dataset is reduced to 20. After dimension reduction, we apply the ICA algorithms to each dataset, such that we are seeking spatially independent components that correspond to spatial functional connectivity maps shown in Fig. \[Components\]. After obtaining the estimated demixing matrices from each of the algorithms and for each dataset, we estimate the independent components and, together with their associated demixing vectors, pair them with the true sources. In the case where more than one estimated component is paired with a single true source, we use Bertsekas algorithm [@bertsekas1988auction] to find the best assignment. To evaluate the performance of the ICA algorithms, we use the average absolute value of the correlation between the true and the estimated sources.
![Simulated fMRI-like components. Note that each color indicates a different component.[]{data-label="Components"}](Comps){width="4cm"}
As expected, we observe in Fig. \[fMRI\], that SparseICA-EBM provides superior performance compared with the other algorithms. Specifically, we note a nearly $100\%$ improvement in performance for CNR values below 0.1. We conclude that SparseICA-EBM is more robust to noise than classical ICA algorithms as it incorporates true prior information of the sources into the estimation.
![Spatial correlation between the true and the estimated sources as a function of the CNR level. Each point is the result of 300 independent runs.[]{data-label="fMRI"}](300RunsNoiseDataComp20Components){width="7.5cm"}
Conclusion {#sec:conc}
==========
Both sparsity and independence have proven useful in a variety of applications, motivating the development of an algorithm that can effectively balance the contributions of these two forms of diversity. In this paper, we propose a new ICA algorithm, SparseICA-EBM, that takes the sparsity of each individual source directly into account. Experimental results, using synthetic as well as simulated fMRI-like data, confirm the superiority of SparseICA-EBM over traditional ICA algorithms. This, combined with the fact that SparseICA-EBM is robust to noise, make it an attractive ICA algorithm for applications where prior information about the sparsity of the sources is available.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Low-energy nuclear weak-interaction processes play important roles in many astrophysical contexts, and effective field theory is believed to be a highly useful framework for describing these processes in a model-independent manner. I present a brief account of the basic features of the nuclear effective theory approach, and some examples of actual calculations carried out in this method.'
---
.25in
USC(NT)-0403
[**ASTROPHYSICAL WEAK-INTERACTION PROCESSES AND NUCLEAR EFFECTIVE FIELD THEORY**]{}[^1]
[ K. Kubodera]{}[^2]
[*Department of Physics and Astronomy,\
University of South Carolina,\
Columbia, SC 29208, USA\
E-mail: [email protected]*]{}
Introduction
============
Low-energy nuclear weak-interaction processes play important roles in many astrophysical phenomena and also in terrestrial experiments designed to detect the astrophysical neutrinos. Obviously, it is important to have reliable estimates of the cross sections for these processes. I wish to describe here some of the recent developments in our endeavor to obtain such estimates.[^3] My main emphasis will be placed on comparison between the traditional method, to be designated as the [*standard nuclear physics approach*]{} (SNPA), and the newly developed [*nuclear effective field theory*]{} (EFT) approach. I shall advocate the viewpoints that (i) nuclear EFT can indeed be a powerful framework for describing low-energy nuclear electroweak processes and (ii) that, in practical applications, EFT and SNPA can play complementary roles.
These points are nicely illustrated by the following three examples: (i) neutrino-deuteron reactions for solar neutrino energies; (ii) solar pp fusion; (iii) solar Hep fusion. Since the process (iii) and related topics will be discussed in detail by Dr. Tae-Sun Park at this Symposium, I shall concentrate on the first two reactions. Let me start with a brief explanation of why these processes are of particular current interest.
At SNO (Sudbury Neutrino Observatory), a one-kiloton heavy water Cerenkov counter is used to detect the solar neutrinos. SNO can monitor the neutrino-deuteron reactions: $$\begin{aligned}
\nu_e+d &\rightarrow&
e^- + p + p\,,\,\,\,\,\,\,
\nu_x + d \rightarrow
\nu_x+p + n\,, \nonumber\\
\bar{\nu}_e + d
&\rightarrow& e^+ +n+n\,,\,\,\,\,\,
\bar{\nu}_x+d \rightarrow
\bar{\nu}_x+p+n\,,\label{nu-d}\end{aligned}$$ as well as the pure leptonic reaction $\nu_x+e^-\rightarrow \nu_x+e^-$. Here $x$ stands for a neutrino of any flavor ($e$, $\mu$ or $\tau$). The recent SNO experiments [@ahmetal] have established that the total solar neutrino flux (counting all neutrino flavors) agrees with the prediction of the standard solar model [@bahcall], whereas the electron neutrino flux from the sun is significantly smaller than the total solar neutrino flux. The amount of deficit in the electron neutrino flux was found to be consistent with what had been known as the solar neutrino problem. These results of the SNO experiments have given clear evidence for the transmutation of solar electron neutrinos into neutrinos of other flavors. Obviously, a precise knowledge of the $\nu$-$d$ reaction cross sections is important for the in-depth interpretation of the existing and future SNO data.
Meanwhile, the pp fusion reaction $$p+p\rightarrow d+e^++\nu_e \label{eq:pp}$$ is the primary solar thermonuclear reaction that essentially controls the luminosity of the sun, and therefore the exact value of its cross section is a crucial input for any elaborate solar models.
Calculational frameworks
========================
Standard nuclear physics approach (SNPA)
----------------------------------------
The phenomenological potential picture has been highly successful in describing a vast variety of nuclear phenomena. In this picture an A-nucleon system is described by a non-relativistic Hamiltonian of the form $$H\,=\,\sum_i^A t_i + \sum_{i<j}^A V_{ij}
+\sum_{i<j<k}^A V_{ijk}+ \cdots\,, \label{HSNPA}$$ where $t_i$ is the kinetic energy of the $i$-th nucleon, $V_{ij}$ is a phenomenological two-body potential between the $i$-th and $j$-th nucleons, $V_{ijk}$ is a phenomenological three-body potential, and so on. (Since potentials involving three or more nucleons play much less important roles than the two-body interactions, we shall be concerned here mainly with $V_{ij}$.) Once the model Hamiltonian $H$ is specified, the nuclear wave function $|\Psi\!\!>$ is obtained by solving the Shrödinger equation $$H|\Psi\!>\,=\,E|\Psi\!>\,.\label{Sch}$$ It is fortunate that the progress of numerical techniques for solving eq.(\[Sch\]) has reached such a level [@cs98] that the wave functions of low-lying levels for light nuclei can now be obtained with essentially no approximation (once the validity of the model Hamiltonian eq.(\[HSNPA\]) is accepted). This liberates us from the “familar" nuclear physics complications that arise as a result of truncating nuclear Hilbert space down to certain model space (such as shell-model configurations within a limited number of major shells, cluster-model trial functions, etc.)
We note that there is large freedom in selecting possible forms of $V_{ij}$, apart from the well-established requirement that, for a large enough value of the inter-nucleon distance, $V_{ij}$ should agree with the one-pion exchange Yukawa potential. For the model-dependent short-range part of $V_{ij}$, the best we could do is to assume certain functional forms and fix the parameters contained in them by demanding that the solutions of eq.(\[Sch\]) for the A=2 case reproduce the nucleon-nucleon scattering data (typically up to the pion-production threshold energy) as well as the deuteron properties. There are by now several so-called [*modern high-precision*]{} [*phenomenological*]{} N-N potential that can reproduce all the existing two-nucleon data with normalized $\chi^2$ values close to 1. These potentials differ widely in the ways short-range physics is parametrized, and, as a consequence, they exhibit substantial difference in their off-shell behaviors. To what extent this arbitrariness may affect the observables of our concern is an important question, to which I will come back later.
In normal situataions, nuclear responses to external electroweak probes are given, to good approximation, by one-body terms, which are also called the impulse approximation (IA) terms. To obtain higher accuracy, however, we must include exchange current (EXC) terms, which represent nuclear responses involving two or more nucleons. These exchange currents (usually taken to be two-body operators) are derived from one-boson exchange diagrams, and the vertices featuring in the relevant diagrams are determined to satisfy the low-energy theorems and current algebra [@crit]. We refer to a formalism based on this picture as the [*standard nuclear physics approach*]{} (SNPA). (This is also called a potential model in the literature.) Schematically, the nuclear matrix element in SNPA is given by $${\mathcal M}_{fi}^{\mbox{\tiny SNPA}}\,=\,
<\!\Psi_{\!f}^{\mbox{\tiny SNPA}}
\,|\sum_\ell^A{\mathcal O}_\ell^{\mbox{\tiny SNPA}}
+\sum_{\ell<m}^A
{\mathcal O}_{\ell m}^{\mbox{\tiny SNPA}}
\,|\Psi_i^{\mbox{\tiny SNPA}}\!>\,, \label{ME-SNPA}$$ where the initial (final) nuclear wave function, $\Psi_i^{\mbox{\tiny SNPA}}$ ($\Psi_f^{\mbox{\tiny SNPA}}$), is a solution of eq.(\[Sch\]); ${\mathcal O}_\ell^{\mbox{\tiny SNPA}}$ and ${\mathcal O}_{\ell m}^{\mbox{\tiny SNPA}}$ are, respectively, the one-body and two-body transition operators for a given electroweak process.
SNPA has been used extensively to describe nuclear electroweak processes in light nuclei, and generally good agreement between theory and experiment [@cs98] gives a strong indication that SNPA essentially captures much of the physics involved.
Effective field theory (EFT)
----------------------------
Even though SNPA has been extremely successful in correlating and explaining a wealth of nuclear phenomana, it is still important from a fundamental point of view to raise the following issues. First, since the hadrons and hadronic systems (such as nuclei) are governed by quantum chromodynamics (QCD), we should ultimately be able to relate nuclear phenomena with QCD, but SNPA is reticent about this relation. In particular, whereas chiral symmetry is known to be a fundamental symmetry of QCD, the SNPA is largely disjoint from this symmetry. Second, even for describing low-energy phenomena, SNPA starts with a “realistic" phenomenological potential which is tailored to encode short-range (high-momentum) and long-range (low-momentum) physics simultaneously. This mixing of the two different scales seems theoretically unsatisfactory. Third, as we write down a phenomenological Lagrangian for describing the nuclear interaction and nuclear responses to the electroweak currents, SNPA does not offer us a clear guiding principle; it is not obvious whether SNPA is equipped with any identifiable expansion parameter that helps us to control the possible forms of terms in the Lagrangian and that provides a general measure of errors in our calculation. To address these and other related issues, a new approach based on EFT was proposed [@wei90] and it has been studied with great intensity; for reviews, see Refs. [@bkm]$^-$[@br02].
The intuitive picture of EFT is quite simple. In describing phenomena characterized by a typical energy-momentum scale $Q$, we may expect that our Lagrangian need not contain explicitly those degrees of freedom that belong to energy-momentum scales much higher than $Q$. This expectation motivates us to introduce a cut-off scale $\Lambda$ that is sufficiently large compared with $Q$ and classify our fields (to be generically represented by $\phi$) into two groups: high-frequency fields $\phi_{\mbox{\tiny H}}$ whose frequencies are higher than $\Lambda$, and low-frequency fields $\phi_{\mbox{\tiny L}}$ with frequencies lower than $\Lambda$. By eliminating (or [*integrating out*]{}) $\phi_{\mbox{\tiny H}}$, we arrive at an [*effective*]{} Lagrangian that only involves $\phi_{\mbox{\tiny L}}$ as explicit dynamical variables. In terms of path integrals, the effective Lagrangian ${\mathcal L}_{{\rm eff}}$ is related to the original Lagrangian ${\mathcal L}$ as $$\begin{aligned}
\int\![d\phi]
\exp\{{\rm i}\!\!\int \!\!d^4x{\mathcal L}[\phi]\}
&=&\int\![d\phi_{\mbox{\tiny H}}]
[d\phi_{\mbox{\tiny L}}]
\exp\{{\rm i}\!\!\int \!\!d^4x{\mathcal L}
[\phi_{\mbox{\tiny H}},\phi_{\mbox{\tiny L}}]\}\\
&\equiv& \int\![d\phi_{\mbox{\tiny L}}]
\exp\{{\rm i}\!\!\int\!\! d^4x{\mathcal L}_{\rm eff}
[\phi_{\mbox{\tiny L}}]\}\,.\label{EFTdef}\end{aligned}$$
It can be shown that ${\mathcal L}_{{\rm eff}}$ defined by eq.(\[EFTdef\]) inherits the symmetries (and the patterns of symmetry breaking, if there are any) of the underlying Lagrangian ${\mathcal L}$. It also follows that ${\mathcal L}_{{\rm eff}}$ should be the sum of all possible monomials of $\phi_{\mbox{\tiny L}}$ and their derivatives that are consistent with the symmetry requirements of ${\mathcal L}$. Because a term involving $n$ derivatives scales like $(Q/\Lambda)^n$, we can organize terms in ${\mathcal L}_{{\rm eff}}$ into a perturbative series in which $Q/\Lambda$ serves as an expansion parameter. The coefficients of terms in this expansion scheme are called the low-energy constants (LECs). Provided all the LEC’s up to a specified order $n$ can be fixed either from theory or from fitting to the experimental values of relevant observables, ${\mathcal L}_{{\rm eff}}$ serves as a complete (and hence model-independent) Lagrangian to the given order of expansion.
Having sketched the basic idea of EFT, we now discuss the specific aspects of EFT as applied to nuclear physics. The underlying Lagrangian ${\mathcal L}$ in this case is the QCD Lagrangian ${\mathcal L}_{QCD}$, whereas, for a typical nuclear physics energy-momentum scale $Q\ll \Lambda_{\chi}\sim 1$ GeV, the effective degrees of freedom that feature in ${\mathcal L}_{{\rm eff}}$ are the hadrons rather than the quarks and gluons. It is non-trivial to apply the formal definition in eq.(\[EFTdef\]) to derive ${\mathcal L}_{{\rm eff}}$ written in terms of hadrons starting from ${\mathcal L}_{QCD}$, because the hadrons cannot be simply identified with the low-frequency field $\phi_L$ in ${\mathcal L}_{QCD}$. To proceed, we choose to be guided solely by symmetry considerations and the above-mentioned expansion scheme. Chiral symmetry plays an important role here. Chiral symmetry is known to be spontaneously broken, leading to the generation of the pions as Nambu-Goldstone bosons. We can incorporate this feature by assigning suitable chiral transformation properties to the Goldstone bosons and writing down all possible chiral-invariant terms up to a specified chiral order [@geo84]. It is to be noted that the above consideration presupposes exact chiral symmetry in ${\mathcal L}_{QCD}$. In reality, ${\mathcal L}_{QCD}$ contains small but finite quark mass terms, which violate chiral symmetry explicitly and lead to a non-vanishing value of the pion mass $m_\pi$. Again, there is a well-defined framework to determine what terms are needed to represent the effect of explicit chiral symmetry breaking [@geo84]. These considerations lead to an EFT called chiral perturbation theory ($\chi$PT) [@wei79; @gl84]. The successes of $\chi$PT in the meson sector are well known; see, [*e.g.,*]{} Ref. [@bkm].
A difficulty we encounter in extending $\chi$PT to the nucleon sector is that, because the nucleon mass $\mN$ is comparable to the cut-off scale $\Lambda_{\chi}$, a simple application of expansion in $Q/\Lambda$ does not work. We can surmount this obstacle with the use of heavy-baryon chiral perturbation theory (HB$\chi$PT), which essentially consists in shifting the reference point of the nucleon energy from 0 to $\mN$ and integrating out the small component of the nucleon field as well as the anti-nucleon field. Thus an effective Lagrangian in HB$\chi$PT contains as explicit degrees of freedom the pions and the large components of the redefined nucleon field. HB$\chi$PT has as expansion parameters $Q/\Lambda_{\chi}$, $m_\pi/\Lambda_{\chi}$ and $Q/\mN$. Since $\mN\approx \Lambda_{\chi}$, it is convenient to combine chiral and heavy-baryon expansions and introduce the chiral index ${\bar \nu}$ defined by $\bar{\nu}=d+(n/2)-2$. Here $n$ is the number of fermion lines participating in a given vertex, and $d$ is the number of derivatives (with $m_\pi$ counted as one derivative). A similar power counting scheme can be introduced for Feynman diagrams as well. According to Weinberg [@wei90], a Feynman diagram that contains $N_A$ nucleons, $N_E$ external fields, $L$ loops and $N_C$ disjoint parts scales like $(Q/\Lambda)^\nu$, where the chiral index $\nu$ is defined as $$\nu = 2 L + 2 (N_C-1) + 2 - (N_A+N_E) +
\sum_i \bar \nu_i\,,\lab{eq:nu}$$ with the summation running over all the vertices.
Although HB$\chi$PT has been very successful in the one-nucleon sector [@bkm], we cannot apply HB$\chi$PT in a straightforward manner to nuclei, which contain more than one nucleon. This is because nuclei allow very low-lying excited states, and the existence of this small energy scale invalidates chiral counting [@wei90]. Weinberg avoided this difficulty by classifying Feynman diagrams into two groups, irreducible and reducible diagrams. Irreducible diagrams are those in which every intermediate state has at least one meson in flight; all others are categorized as reducible diagrams. The chiral counting rules should only be applied to irreducible diagrams. The contribution of all the two-body irreducible diagrams (up to a specified chiral order) is treated as an effective potential (to be denoted by $V_{ij}^{\mbox{\tiny EFT}}$) that acts on nuclear wave functions. Meanwhile, the contributions of reducible diagrams can be incorporated [@wei90] by solving the Schrödinger equation $$H^{\mbox{\tiny EFT}}
|\Psi^{\mbox{\tiny EFT}}\!>
\,=\,E|\Psi^{\mbox{\tiny EFT}}\!>\,,
\label{Sch-EFT}$$ where $$H^{\mbox{\tiny EFT}}
\,=\,\sum_i^A t_i +
\sum_{i<j}^A V_{ij}^{\mbox{\tiny EFT}}\,,
\label{HEFT}$$ We refer to this two-step procedure as [*nuclear*]{} , or, to be more specific, [*nuclear*]{} in the Weinberg scheme. (This is often called the $\Lambda$-counting scheme [@lep99].)
To apply nuclear to a process that involves (an) external current(s), we derive a nuclear transition operator ${\mathcal T}$ by calculating the contributions of all the irreducible diagrams (up to a given chiral order $\nu$) that involve the relevant external current(s). To maintain consistent chiral counting, the nuclear matrix element of ${\mathcal T}$ must be calculated with the use of nuclear wave functions which are governed by nuclear interactions that represent all the irreducible A-nucleon diagrams up to $\nu$-th order. Thus, a transition matrix in nuclear EFT is given by $${\mathcal M}_{fi}^{\mbox{\tiny EFT}}\,=\,
<\!\Psi_{\!f}^{\mbox{\tiny EFT}}
\,|\sum_\ell^A{\mathcal O}_\ell^{\mbox{\tiny EFT}}
+\sum_{\ell<m}^A
{\mathcal O}_{\ell m}^{\mbox{\tiny EFT}}
\,|\Psi_i^{\mbox{\tiny EFT}}
\!>\,, \label{ME-EFT}$$ where the superscript, “EFT", implies that the relevant quantities are obtained according to EFT as described above. If this program is carried out exactly, it would constitute an [*ab initio*]{} calculation. It is worth noting that EFT tells us exactly at what chiral order three-body operators start to contribute to ${\mathcal T}$, and that, to chiral orders of our present concern, we do not need three-body operators. For this reason we have retained in eq.(\[ME-EFT\]) only one- and two-body operators. This type of unambiguous classification of transition operators according to their chiral orders is a great advantage of EFT, which is missing in eq.(\[ME-SNPA\]).
I should mention that there exists an alternative form of nuclear EFT based on the power divergence subtraction (PDS) scheme. The PDS scheme proposed by Kaplan, Savage and Wise in their seminal papers [@ksw] uses a counting scheme (often called Q-counting) that differs from the Weinberg scheme. An advantage of the PDS scheme is that it maintains formal chiral invariance, whereas the Weinberg scheme loses manifest chiral invariance. In many practical applications, however, this formal problem is not worrisome up to the chiral order under consideration, [*i.e.,*]{} the chiral order up to which our irreducible diagrams are evaluated. Although the PDS scheme has produced many important results (for a review, see Ref. [@beaetal01]), I concentrate here on the Weinberg scheme, because this is the framework in which our own work has been done.
Hybrid EFT
----------
In the above I emphasized the formal merits of nuclear EFT. In actual calculations, however, we face the following two problems. First, it is still a great challenge to generate, strictly within the EFT framework, nuclear wave functions whose accuracy is comparable to that of SNPA wave functions. Second, as mentioned earlier, the chiral Lagrangian, ${\mathcal L}_{{\rm eff}}$, is definite only when the values of all the relevant LECs are fixed, but there may be cases where this requirement cannot be readily met. A pragmatic solution to the first problem is to use in eq.(\[ME-EFT\]) wave functions obtained in SNPA; we refer to this eclectic approach as hybrid EFT. A nuclear transition matrix element in hybrid EFT is given by $${\mathcal M}_{fi}^{hyb-{\mbox{\tiny EFT}}}\,=\,
<\!\Psi_{\!f}^{\mbox{\tiny SNPA}}\,
|\sum_\ell^A{\mathcal O}_\ell^{\mbox{\tiny EFT}}
+\sum_{\ell<m}^A
{\mathcal O}_{\ell m}^{\mbox{\tiny EFT}}
\,|\Psi_i^{\mbox{\tiny SNPA}}\!>\,, \label{ME-hybrid}$$ Because, as mentioned, the NN interactions that generate SNPA wave functions reproduce accurately the entirety of the two-nucleon data, the adoption of eq.(\[ME-hybrid\]) is almost equivalent to using the empirical data themselves to control the initial and final nuclear wave functions. In the context of theoretically deriving the nuclear interactions based on EFT, hybrid EFT may look like “retrogression". But, if our goal is to obtain a transition matrix element as accurately as possible with the maximum help of available empirical input, hybrid EFT is a justifiable approach insofar as the above-mentioned off-shell problem and the contributions of three-body (and higher-body) interactions are properly addressed. These points will be discussed later on.
The calculations reported in Refs. [@pkmr99; @pc00] seem to support hybrid EFT. There, the nuclear matrix elements in the A=2 systems for one-body operators (or IA terms) calculated with the use of EFT-generated wave functions were found to be very close to those calculated with the SNPA wave functions. Thus EFT and hybrid EFT should give practically the same IA matrix elements. Meanwhile, we can generally expect that the ratio of the two-body EXC contributions to those of the IA operators should be much less sensitive to the details of the nuclear wave functions than the absolute values are. It therefore seems reasonable to rely on for deriving transition operators and evaluate their matrix elements using the realistic wave functions obtained in SNPA, and in this sense hybrid EFT is more than a mere expedient.
The issue of possible unknown LECs will be discussed in the next subsection.
MEEFT or EFT\*
--------------
Hybrid EFT can be used for complex nuclei (A = 3, 4, ...) with essentially the same accuracy and ease as for the A=2 system. We should reemphasize in this connection that, in A-nucleon systems (A$\ge$3), the contributions of transition operators involving three or more nucleons are intrinsically suppressed according to chiral counting, and hence, up to a certain chiral order, a transition operator in an A-nucleon system consists of the same EFT-based 1-body and 2-body terms as used for the two-nucleon system. Then, since SNPA provides high-quality wave functions for the A-nucleon system, one can calculate ${\mathcal M}_{fi}^{hyb-{\mbox{\tiny EFT}}}$ with precision comparable to that for the corresponding two-nucleon case.
Now, in most practical cases, the one-body operator, ${\mathcal O}_\ell^{\mbox{\tiny EFT}}$, is free from unknown LECs. So let us concentrate on the two-body operator, ${\mathcal O}_{\ell m}^{\mbox{\tiny EFT}}$, and suppose that ${\mathcal O}_{\ell m}^{\mbox{\tiny EFT}}$ under consideration contains an LEC (call it $\kappa$) that cannot be determined with the use of A=2 data alone. It is possible that an observable (call it $\Omega$) in a A-body system (A$\ge$3) is sensitive to $\kappa$ and that the experimental value of $\Omega$ is known with sufficient accuracy. Then we can determine $\kappa$ by calculating ${\mathcal M}_{fi}^{hyb-{\mbox{\tiny EFT}}}$ responsible for $\Omega$ and adjusting $\kappa$ to reproduce the empirical value of $\Omega$. Once $\kappa$ is fixed this way, we can make [*predictions*]{} for any other observables for any other nuclear systems that are controlled by the same transition operators. When hybrid EFT is used in this manner, we refer to it as MEEFT ([*more effective*]{} EFT) or EFT\*.
MEEFT is the most efficient existing formalism for correlating various observables in different nuclei, using the transition operators controlled by EFT. A further notable advantage of MEEFT is that, since correlating the observables in neighboring nuclei is likely to serve as an additional renormalization, the possible effects of higher chiral order terms and/or off-shell ambiguities can be significantly suppressed by the use of MEEFT.[^4] I will come back to this point later, when I discuss concrete examples.
We need to recall here the important role of momentum cutoff in EFT. As emphasized before, the effective Lagrangian ${\mathcal L}_{eff}$ is, by construction, valid only below the specified cutoff scale $\Lambda$. This basic constraint must be respected in our nuclear EFT calculations; we must ensure that nuclear intermediate states involved in the computation of eq.(\[ME-EFT\]) remain within this constrained regime. It is reasonable to implement this constraint by requiring that the two-nucleon relative momentum should be smaller than $\Lambda$. A possible choice of the cutoff function is the Gaussian form $\exp(-\vec{p}^2/\Lambda^2)$. (The detailed form of the cutoff function should not be very important.) As a reasonable range of the value of $\Lambda$ we may choose: 500 MeV $\lsim \Lambda \lsim$ 800 MeV, where the lower bound is dictated by the requirement that $\Lambda$ should be sufficiently large compared with the pion mass (in order to accommodate pion physics), while the upper bound reflects the fact that our EFT is devoid of the $\rho$ meson.
Numerical results
=================
We now discuss the applications of the above-described calculational methods to the two processes of our concern: pp fusion and the $\nu$-$d$ reaction. These reactions share the common feature that a precise knowledge of the Gamow-Teller (GT) transition matrix elements is crucial in estimating their cross sections. We therefore concentrate on the GT transitions. We will show here, following Refs. [@PMS-pp; @PMS-pphep], that MEEFT can be used very profitably for these reactions.
We can argue (see, [*e.g.,*]{} Ref. [@PMS-pphep]) that 1-body IA operators for the GT transition can be fixed unambiguously from the available 1-body data. As for the 2-body operators, to next-to-next-to-next-to-leading order (N$^3$LO) in chiral counting, there appears one unknown LEC that cannot be at present determined from data for the A=2 systems. This unknown LEC, denoted by $\hat{d}_R$ in Ref. [@pkmr98b], parametrizes the strength of contact-type four-nucleon coupling to the axial current. Park [*et al. *]{}[@PMS-pp; @PMS-pphep] noted that the same LEC, $\hat{d}_R$, also appears as a single unknown parameter in the calculation of the tritium $\beta$-decay rate $\Gamma_{\beta}^t$, and they used MEEFT to place a constraint on $\hat{d}_R$ from the experimental value of $\Gamma_{\beta}^t$. Since $\Gamma_{\beta}^t$(exp) is known with high precision, and since the accurate wave functions of $^3$H and $^3$He are available from a well-developed SNPA calculation [@Metal], we can determine $\hat{d}_R$ with sufficient accuracy for our purposes. Once the value of $\hat{d}_R$ is determined this way, we can carry out parameter-free MEEFT calculations for pp-fusion [@PMS-pp; @PMS-pphep] and the $\nu$-$d$ reactions [@andetal-nud]. I present here a brief summary of the results of these calculations.
For a given value of $\Lambda$ within the above-mentioned range (500 MeV $\lsim \Lambda \lsim$ 800 MeV), $\hat{d}_R$ is adjusted to reproduce $\Gamma_{\beta}^t$(exp), and then the cross sections for pp-fusion and the $\nu d$ reactions are calculated. The results indicate that, although the best-fit value of $\hat{d}_R$ varies significantly as a function of $\Lambda$, the observables (in our case the above two reaction cross sections) exhibit a high degree of stability against the variation of $\Lambda$. This stability may be taken as an indication that the use of MEEFT for inter-correlating the observables in neighboring nuclei [*effectively*]{} renormalizes various effects, such as the contributions of higher-chiral order terms, mismatch between the SNPA and EFT wave functions, etc. This stability is essential in order for MEEFT to maintain its predictive power.
Park [*et al.*]{} [@PMS-pp; @PMS-pphep] used MEEFT to calculate the rate of pp fusion, $pp\!\rightarrow\!e^+\nu_e d$. The result expressed in terms of the threshold $S$-factor is $$S_{pp}(0)=3.94\!\times\!(1\pm0.005)
\times 10^{-25}\,{\rm MeV\, b}\,.\label{Spp}$$ It has been found that $S_{pp}(0)$ changes only by $\sim$0.1% against changes in $\Lambda$, assuring thereby the robustness of the MEEFT prediction. The MEEFT result, eq.(\[Spp\]), is consistent with that obtained in SNPA by Schiavilla [*et al.*]{} [@schetal98]. Meanwhile, the fact that MEEFT allows us to make an error estimate \[as given in eq.(\[Spp\])\] is a notable advantage over SNPA. The details on how we arrive at this error estimate can be found in Refs. [@PMS-pp; @PMS-pphep]. Here I just remark that the error indicated in eq.(\[Spp\]) represents an improvement by a factor of $\sim$10 over the previous results based on a naive hybrid EFT [@pkmr98b].
We now move to the $\nu$-$d$ reactions, eq.(\[nu-d\]), and give a brief survey of all the recent results obtained in SNPA, EFT and MEEFT. Within SNPA a detailed calculation of the $\nu$-$d$ cross sections, $\sigma(\nu d)$, was carried out by Nakamura, Sato, Gudkov and myself [@NSGK],[^5] and this calculation has recently been updated by Nakamura [*et al.*]{} (NETAL) [@NETAL]. As demonstrated in Ref.[@crsw], the SNPA exchange currents for the GT transition are dominated by the $\Delta$-particle excitation diagram, and the reliability of estimation of this diagram depends on the precision with which the coupling constant $g_{\pi N\Delta}$ is known. NETAL fixed $g_{\pi N\Delta}$ by fitting $\Gamma_{\beta}^t$(exp), and proceeded to calculate $\sigma(\nu d)$. Meanwhile, Butler, Chen and Kong (BCK) [@EFT] carried out an EFT calculation of the $\nu$-$d$ cross sections, using the PDS scheme [@ksw]. The results of BCK agree with those of NETAL in the following sense. BCK’s calculation involves one unknown LEC (denoted by $L_{\rm 1A}$), which like $\hat{d}_R$ in Ref.[@PMS-pphep], represents the strength of a four-nucleon axial-current coupling term. BCK determined $L_{\rm 1A}$ by requiring that the $\nu d$ cross sections of NETAL be reproduced by their EFT calculation. With the value of $L_{\rm 1A}$ adjusted this way, $\sigma(\nu d)$’s obtained by BCK show a perfect agreement with those of NETAL for all the four reactions in eq.(\[nu-d\]) and for the entire solar neutrino energy range, $E_\nu\lsim$ 20 MeV. Moreover, the best-fit value, $L_{\rm 1A}=5.6\,{\rm fm}^3$, found by BCK [@EFT] is consistent with its magnitude expected from the naturalness argument (based on a dimensional analysis), $|L_{\rm 1A}|\le 6\,{\rm fm}^3$. The fact that an EFT calculation (with one parameter fine-tuned) reproduces the results of SNPA very well strongly suggests the robustness of the SNPA calculation of $\sigma(\nu d)$.
Even though it is reassuring that the $\nu$-$d$ cross sections calculated in SNPA and EFT agree with each other (in the above-explained sense), it is desirable to carry out an EFT calculation that is free from any adjustable LEC. Fortunately, MEEFT allows us to carry out an EFT-controlled parameter-free calculation of the $\nu$-$d$ cross sections, and such a calculation was carried out by Ando [*et al.*]{} [@andetal-nud]. The $\sigma(\nu d)$’s obtained in Ref. [@andetal-nud] are found to agree within 1% with $\sigma(\nu d)$’s obtained by NETAL using SNPA [@NETAL]. These results show that the $\nu$-$d$ cross sections used in interpreting the SNO experiments [@ahmetal] are reliable at the 1% precision level.
We remark that, as PDS [@ksw] is built on an expansion scheme for transition amplitudes themselves, it does not employ the concept of wave functions. This feature is an advantage in some contexts, but its disadvantage in the present context is that we cannot readily relate the transition matrix elements for an A-nucleon system with those for the neighboring nuclei; in PDS, each nuclear system requires a separate parametrization. This feature underlies the fact that, in the work of BCK [@EFT], $L_{1A}$ remained undetermined, because no experimental data is available to fix $L_{1A}$ within the two-nucleon systems.
Discussion
==========
In introducing hybrid EFT, we replace $|\Psi^{\mbox{\tiny EFT}}\!>$ for the initial and final nuclear states in eq.(\[ME-EFT\]) with the corresponding $|\Psi^{\mbox{\tiny SNPA}}\!>$’s; see eq.(\[ME-hybrid\]). This replacement may bring in a certain degree of model dependence, called the off-shell effect, because the phenomenological NN interactions are constrained only by the on-shell two-nucleon observables.[^6] This off-shell effect, however, is expected to be small for the reactions under consideration, since they involve low momentum transfers and hence are not extremely sensitive to the short-range behavior of the nuclear wave functions. One way to quantify this expectation is to compare a two-nucleon relative wave function generated by the phenomenological potential with that generated by an EFT-motivated potential. Phillips and Cohen [@pc00] made such a comparison in their analysis of the 1-body operators responsible for electron-deuteron Compton scattering, and showed that a hybrid EFT works well up to momentum transfer 700 MeV. A similar conclusion is expected to hold for a two-body operator, so long as its radial dependence has a “smeared-out" structure reflecting the finite momentum cutoff. We can therefore expect that hybrid EFT as applied to low energy should be practically free from the off-shell ambiguities. The off-shell effect should be even less significant in MEEFT, wherein an additional “effective" renormalization is likely to be at work (see subsection 2.4).
Another indication of the stability of the MEEFT results comes from a recently proposed idea of the low-momentum nuclear potential [@kuo]. As mentioned, a “realistic phenomenological" nuclear interaction, $V_{ij}$ in eq.(\[HSNPA\]), is determined by fitting to the full set of two-nucleon data up to the pion production threshold energy. So, physically, $V_{ij}$ should reside in a momentum regime below a certain cutoff, $\Lambda_c$. In the conventional treatment, however, the existence of this cutoff scale is ignored, and eq.(\[Sch\]) is solved in such a manner that the entire momentum range is allowed to participate. Bogner [*et al. *]{} proposed to construct an [*effective low-momentum*]{} potential $V_{low-k}$ by eliminating (or integrating out) from $V_{ij}$ the momentum components higher than $\Lambda_c$, and calculated $V_{low-k}$’s corresponding to a number of well-established of $V_{ij}$’s. It was found that all these $V_{low-k}$’s lead to identical half-off-shell T-matrices, even though the ways short-range physics is encoded in them are highly diverse. This implies that the $V_{low-k}$’s are free from the off-shell ambiguities, and therefore the use of $V_{low-k}$’s is essentially equivalent to employing $V_{ij}^{\mbox{\tiny EFT}}$ (appearing in eq.(\[HEFT\])), which by construction should be model-independent. Now, as mentioned, our MEEFT calculation has a momentum-cutoff regulator built in, and this essentially ensures that the matrix element, ${\mathcal M}_{fi}^{hyb-{\mbox{\tiny EFT}}}$, in eq.(\[ME-hybrid\]) is only sensitive to the half-off-shell T-matrices that are controlled by $V_{low-k}$ instead of $V_{ij}$. Therefore, we can expect that the MEEFT results reported here are essentially free from the off-shell ambiguities.
Summary
=======
After giving a very limited survey of nuclear , I must repeat my disclaimer that I have left out many important topics belonging to nuclear . Among others, I did not discuss very important studies by Epelbaum, Glöckle and Mei[ß]{}ner [@epeetal] to construct a formally consistent framework for applying to complex nuclei. It should be highly informative to apply this type of formalism to electroweak processes and compare the results with those of MEEFT. In this connection I find it noteworthy that the range of the cutoff parameter favored in Ref. [@epeetal] is consistent with the range used by Park [*et al.*]{} [@PMS-pp; @PMS-pphep]
Despite the highly limited scope of topics covered, I hope I have succeeded in demonstrating that MEEFT is a powerful framework for computing the transition amplitudes of low-energy electroweak processes in light nuclei. I also wish to emphasize that, in each of the cases for which both SNPA and MEEFT calculations have been performed, it has been found that the result of MEEFT supports and improves the SNPA result.
Acknowledgments {#acknowledgments .unnumbered}
===============
This talk is based on the work done in collaboration with T.-S. Park, M. Rho, D.-P. Min, F. Myhrer, T. Sato, V. Gudkov and S. Nakamura, and I wish to express my sincere thanks to these colleagues.
[0]{}
K. Kubodera, Proceedings of the First Yamada Conference on Neutrinos and Dark Matter in Nuclear Physics (NDM03), Nara, Japan, June 2003, On-line publication: http://ndm03.phys.sci.osaka-u.ac.jp/proc/htm;\
nucl-th/0308055.
Q. Ahmad [*et al.,*]{} [*Phys. Rev. Lett. *]{} [**87**]{}, 071301 (2001); [*ibid*]{} [**89**]{}, 011301 (2002); [*ibid*]{} [**89**]{}, 011302 (2002).
J. Bahcall, M. Pinsonneault and S. Basu, [*Astrophys. J. *]{} [**555**]{}, 990 (2001).
See, [*e.g.,*]{} J. Carlson and R. Schiavilla, [*Rev. Mod. Phys. *]{} [**70**]{}, 743 (1998).
M. Chemtob and M. Rho, [*Nucl. Phys.*]{} [**A163**]{}, 1 (1971); E. Ivanov and E. Truhlik, [*Nucl. Phys.*]{} [**A316**]{}, 437 (1979); [*ibid*]{} [**A316**]{}, 451 (1979).
S. Weinberg, [*Phys. Lett.*]{} [**B251**]{}, 288 (1990); [*Phys. Lett.*]{} [**B295**]{}, 114 (1992).
See. [*e.g.,*]{} V. Bernard, N. Kaiser and U.-G. Meissner, [*Int. J. Mod. Phys. *]{} [**E4**]{}, 193 (1995).
See, [*e.g.*]{}, U. van Kolck, [*Prog. Part. Nucl. Phys.*]{} [**43**]{}, 337 (1999).
G.P. Lepage, in: Proceedings of the Second Workshop on Effective Field Theories in Nuclear Physics, eds. P. Bedaque [*et al.,*]{} (World Scientific, 1999), p. 353.
S. Beane [*et al.,*]{} in “[*At the Frontier of Particle Physics – Handbook of QCD"*]{}, ed. M Shifman, vol. 1 (World Scientific, Singapore, 2001).
G. Brown and M. Rho, [*Phys. Rep. *]{} [**363**]{}, 85 (2002).
See, [*e.g.,*]{} H. Georgi, [*Weak Interactions and Modern Particle Physics*]{} (Benjamin, 1984).
S. Weinberg, [*Physica*]{}, [**6A**]{}, 327 (1979).
J. Gasser and H. Leutwyler, [*Ann. Phys.*]{} [**158**]{}, 142 (1984); J. Gasser, M. Sainio and S. Švarc, [*Nucl. Phys.*]{} [**B307**]{}, 779 (1988).
D. Kaplan, M. Savage and M. Wise, [*Nucl. Phys.*]{} [**B478**]{}, 629 (1996); [*ibid*]{} [**B535**]{}, 329 (1998).
T.-S. Park, D.-P. Min and M. Rho, [*Phys. Rev. Lett.*]{} [**74**]{}, 4153 (1995).
T.-S. Park, K. Kubodera, D.-P. Min and M. Rho, [*Phys. Rev.*]{} [**C58**]{}, 637 (1998).
T.-S. Park, K. Kubodera, D.-P. Min and M. Rho, [*Astrophys. J.*]{} [**507**]{}, 443 (1998).
T.-S. Park, K. Kubodera, D.-P. Min and M. Rho, [*Nucl. Phys.*]{} [**A646**]{}, 83 (1999).
D.R. Phillips and T.D. Cohen, [*Nucl. Phys. *]{} [**A668**]{}, 45 (2000).
T.-S. Park, K. Kubodera, D.-P. Min and M. Rho, nucl-th/9904053; M. Rho, in: Proceedings of the Second Workshop on Effective Field Theories in Nuclear Physics, ed. P. Bedaque, M. Savage, R. Seki, U. van Kolck (World Scientific, Singapore, 1999), p. 225,
T.-S. Park, K. Kubodera, D.-P. Min and M. Rho, [*Nucl. Phys.*]{} [**A684**]{}, 101c (2001).
T.-S. Park, L. Marcucci, R. Schiavilla, M. Viviani, A. Kievsky, S. Rosati, K. Kubodera, D.-P. Min and M. Rho, nucl-th/0106026.
T.-S. Park, L. Marcucci, R. Schiavilla, M. Viviani, A. Kievsky, S. Rosati, K. Kubodera, D.-P. Min and M. Rho, [*Phys. Rev. *]{} [**C67**]{}, 055206 (2003).
L. Marcucci [*et al.,*]{} [*Phys. Rev. Lett.*]{} [**84**]{}, 5959 (2000); L. Marcucci, R. Sciavilla, M. Viviani, A. Kievsky, S. Rosati and J. Beacom, [*Phys. Rev.*]{} [**C63**]{}, 015801 (2001).
R. Schiavilla [*et al.,*]{} [*Phys. Rev.*]{}, [**C58**]{}, 1263 (1998).
S. Ando, Y.H. Song, T.-S. Park, H. Fearing and K. Kubodera, [*Phys. Lett.* ]{} [**B555**]{}, 49 (2003).
S. Nakamura, T. Sato, V. Gudkov and K. Kubodera, [*Phys. Rev.*]{} [**C63**]{}, 034617 (2001).
K. Kubodera and S. Nozawa, [*Int. J. Mod. Phys.*]{} [**E3**]{}, 101 (1994).
S. Nakamura, T. Sato, S. Ando, T.-S. Park, V. Gudkov, F. Myhrer and K. Kubodera, [*Nucl. Phys. *]{} [**A707**]{}, 561 (2002).
J. Carlson, D.-O. Riska, R. Schiavilla and R. Wiringa, [*Phys. Rev.*]{} [**C44**]{}, 619 (1991).
M. Butler and J.-W. Chen, [*Nucl. Phys.*]{} [**A675**]{}, 575 (2000); M. Butler, J.-W. Chen and X. Kong, [*Phys. Rev.*]{} [**C63**]{}, 035501 (2001).
S. Bogner, T.T.S. Kuo and L. Corragio, [*Nucl. Phys.*]{} [**A684**]{}, 432c (2001); S. Bogner [it et al.,]{} [*Phys. Rev.*]{} [**C65**]{}, 051301 (2002); A. Schwenk, G.E. Brown and B. Friman, [*Nucl. Phys.* ]{}, [**A703**]{}, 745 (2002); S.K. Bogner, T.T.S. Kuo and A. Schwenk, [*Phys. Rep.*]{} [**386**]{}, 1 (2003).
E. Epelbaum, W. Glöckle, A. Krüger and U.-G. Mei[ß]{}ner, [*Nucl. Phys.*]{} [**A645**]{}, 413 (1999); E. Epelbaum [*et al.,*]{} [*Nucl. Phys.*]{} [**A671**]{}, 295 (2000); E. Epelbaum [et al.,]{} [*Phys. Rev. Lett.*]{} [**86**]{}, 4787 (2001); E. Epelbaum [*et al.,*]{} [*Phys. Rev.*]{} [**C66**]{}, 064001 (2002); E. Epelbaum [*et al.,*]{} [*Eur. Phys. J.*]{} [**A15**]{}, 543 (2002); E. Epelbaum, talk at the LOWq-03: 2nd Workshop on Electromagnetic Nuclear Reactions at Low Momentum Transfer, Halifax, Nova Scotia, Canada; July 2003; http://www.ap.stmarys.ca/LOWq-03.
[^1]: Invited talk at the KIAS-APCTP Astro-Hadron Symposium, KIAS, Seoul, Korea, November 10-14, 2003.
[^2]: Work partially supported by the US National Science Foundation, Grant No. PHY-0140214
[^3]: This talk has some overlap with the one I gave at NDM03 [@NDM03].
[^4]: MEEFT should be distinguished from an earlier naive hybrid EFT model in which the short-range terms were dropped altogether using an intuitive argument based on short-range NN repulsion.
[^5]: For a review of the earlier SNPA calculations, see Ref. [@kn94].
[^6]: In a fully consistent theory, physical observables are independent of field transformations that lead to different off-shell behaviors, and therefore the so-called off-shell effect is not really a physical effect. In an approximate theory, observables may exhibit superficial dependence on the off-shell behavior, and it is customary to refer to this dependence as an off-shell “effect".
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We perform an analysis of the Einstein-Skyrme cosmological model in Bianchi-IX background. We analytically describe asymptotic regimes and semi-analytically – generic regimes. It appears that depending on the product of Newtonian constant $\kappa$ with Skyrme coupling $K$, in absence of the cosmological term there are three possible regimes – recollapse with ${\kappa K}< 2$ and two power-law regimes – $\propto t^{1/2}$ for ${\kappa K}=2$ and $\propto t$ for ${\kappa K}> 2$. In presence of the positive cosmological term, power-law regimes turn to the exponential (de Sitter) ones while recollapse regime turn to exponential if the value for $\Lambda$-term is sufficiently large, otherwise the regime remains recollapse. Negative cosmological term leads to the recollapse regardless of ${\kappa K}$. All nonsingular regimes have the squashing coefficient $a(t) \to 1$ at late times, which is associated with restoring symmetry dynamics. Also all nonsingular regimes appear to be linearly stable – exponential solutions always while power-law for an open region of the initial conditions.'
author:
- 'Sergey A. Pavluchenko'
title: 'Dynamics of gravitating hadron matter in Bianchi-IX cosmological model'
---
Introduction
============
One of the most important nonlinear field theories is the sigma model, with its applications covering many aspects of quantum physics (see e.g. [@Manton1] for review), but within this model it is impossible to build static soliton solutions in 3+1 dimensions. To overcome this Skyrme introduced [@Skyrme] term which allows static soliton solutions with finite energy, called [*Skyrmions*]{} (see also [@Manton1; @ped2] for review), to exist. It appears that excitations around Skyrme solutions may represent Fermionic degrees of freedom, suitable to describe baryons (see [@ferm1] for detailed calculations and [@ferm2; @ferm3; @ferm4; @ferm5] for examples). Winding number of Skyrmions is identified with the baryon number in particle physics [@wind]. Apart from particle and nuclear physics, Skyrme theory is relevant to astrophysics [@astro], Bose-Einstein condensates [@EBc], nematic liquids [@nematic], magnetic structures [@magnetic] and condensed matter physics [@cond]. Also, Skyrme theory naturally appears in AdS/CFT context [@adscft].
Due to highly nonlinear character of sigma and Skyrme models, it is very difficult to build exact solution in both of them. So, to make field equations more tractable, one usually adopts certain [*ansatz*]{}. For Skyrme model one of the best known and mostly used is hedgehog [*ansatz*]{} for spherically symmetric systems, which reduces field equations to a single scalar equation. It worth mentioning that recently this [*ansatz*]{} was generalized [@CH] for non-spherically-symmetric cases.
Use of hedgehog [*ansatz*]{} allows study of self-gravitating Skyrme models. In particular, it was demonstrated the potential presence of Skyrme hair for spherically-symmetric black-hole configurations [@hair]. This is the first genuine counterexample to “no-hair” conjecture which appears to be stable [@h1]; its particle-like [@h2] counterparts and dynamical configurations [@h3] have been studied numerically. After that, more realistic spherically- and axially-symmetric black-hole and regular configurations were studied [@recent].
Apart from spherically-symmetric configurations, of particular interest are cosmologically-type solutions. Generalized hedgehog [*ansatz*]{} makes it possible to write down simplified field equations for non-spherically-symmetric configurations which we used to perform analysis of Bianchi-I and Kantowski-Sachs models for Einstein-Skyrme cosmology with $\Lambda$-term [@we14] (particular subcase was studied in [@another]). The paper [@fabr15] was a logical continuation of them, as the particular solution of the Bianchi-IX cosmological model was described. The analysis suggests that, based on the static counterpart of this model, the construction of exact multi-Skyrmion configurations composed by elementary spherically symmetric Skyrmions with non-trivial winding number in four-dimensions is possible [@rest] (see also [@sun] for possible generalization to higher $SU(N)$ models).
In this paper we are going to consider full Bianchi-IX cosmological model in Einstein-Skyrme system. Our study is motivated from both field theory and cosmological point of view. Indeed, this is one of few (if not the only) systems where one can study analytically dynamical and cosmological consequences of the conserved topological charge, which in this particular case is associated with the baryon number. From the cosmological point of view, Bianchi-IX model is well-known and well-studied in cosmology – for instance, for the proof of inevitability of the physical singularity through oscillatory approach to it [@belinski]. So that, if we consider Bianchi-IX model, the results could be translated and compared with the counterparts from our physical Universe.
The structure of the manuscript is as follows: first we review Einstein-Skyrme system and derive basic equations, then we study asymptotic case both with and without $\Lambda$-term. After that we study general case, address linear stability of the obtained solutions and finally discuss and summarize the results.
Equations of motion
===================
The Skyrme action can be constructed in the following way: let be $U$ a $SU(2)$ valued scalar field. We can the define the quantities: $$A_{\mu}^i t_i \equiv A_{\mu}=U^{-1}\nabla_{\mu}U,$$ $$F_{\mu \nu}=[A_{\mu},A_{\nu}].$$ Here the Latin indices correspond to the group indices and the generators $t_i$ of $SU(2)$ are related to the Pauli matrices by $t_i =-i \sigma_i$. The Skyrme action is then defined as $$S_{Skyrme}=\frac{K}{2}\int d^4 x\sqrt{-g}\mathrm{Tr}\left( \frac{1}{2}A_{\mu}A^{\mu}+ \frac{\lambda}{16}F_{\mu \nu}F^{\mu \nu} \right). \label{s1}$$ The case when $\lambda =0$ is called non linear Sigma Model and the term which multiplies $\lambda$ is called the Skyrme term. The total action for a self gravitating Skyrme field reads $$S=\int d^4x \sqrt{-g}\dac{R-2\Lambda}{2\kappa}+S_{Skyrme}, \label{s2}$$ where $\kappa$ is the gravitational constant, $R$ is the Ricci scalar and $\Lambda$ is the cosmological constant. Skyrme field equation reads $$\nabla^{\mu}A_{\mu}+\frac{\lambda}{4}\nabla^{\mu}[A^{\nu},F_{\mu\nu}]=0. \label{skyrme}$$
The topological charge of the Skyrme model is $$w = - \dac{1}{24\pi^2} \int\limits_{t={\mathop{\rm const}\nolimits}} {\mathop{\rm Tr}\nolimits}\[ \epsilon^{ijk} A_i A_j A_k \], \label{top}$$ and physically it represents the baryonic charge.
The $SU(2)$ valued scalar field can be parameterized in a standard way $$U=\mathbf{I} Y^0 + Y^it_i \; \; \; ; \; \; \; U^{-1}= \mathbf{I} Y^0 - Y^it_i, \label{Ufield}$$ with $Y^0=Y^0(x^{\mu})$ and $Y^i=Y^i(x^{\mu})$ must satisfy $(Y^0)^2+Y_iY^i=1$.The most famous and most studied [*ansatz*]{} for searching solutions to the (non-self gravitating) Skyrme theory is so called “hedgehog” which is obtained by choosing $$Y^0 =\cos (\alpha ) \; \; \; ; \; \; \; Y^i= n^i\sin (\alpha ), \label{hedgehog}$$ where $\alpha$ is a radial profile function and $n^i$ is a normal radial vector $$n^1=\sin \Theta \cos \Phi \; \; \; ; \; \; \; n^2 = \sin \Theta \sin \Phi \; \; \; ; \; \; \; n^3=\cos \Theta. \label{normalvector}$$
As mentioned, we work with Bianchi-IX metric
$$ds^2=-dt^2+\frac{\rho^2 (t)}{4}\left[ a^2 (t)(d\gamma +\cos \theta d\varphi )^{2}+d\theta ^{2}+\sin ^{2}\theta d\varphi ^{2}\right] , \label{BIX}$$
where $\rho (t)$ is a global scale factor and $a(t)$ is a squashing coefficient. One can check that (see also [@fabr15]), with unit baryonic charge $w=+1$ (\[top\]), the configuration $$\Phi =\frac{\gamma +\varphi }{2},\ \tan \Theta =\frac{\cot \left( \frac{\theta }{2}\right) }{\cos \left( \frac{\gamma -\varphi }{2}\right) },\ \tan \alpha =\frac{\sqrt{1+\tan ^{2}\Theta }}{\tan \left( \frac{\gamma -\varphi }{2}\right) } \label{conf}$$ identically satisfies the Skyrme field equations (\[skyrme\]) on any background metric of the form (\[BIX\]). Now substituting metric (\[BIX\]) and configuration (\[conf\]) into action (\[s1\]) and (\[s2\]) as well as to the hedgehog [*ansatz*]{}, one can derive equations of motion in the following form (see also [@fabr15]):
$$\begin{array}{l}
2a\rho ^2 (2\rho \dot{a}+3a\dot{\rho})\dot{\rho}-2a^2 \rho^2 (\Lambda \rho^2 +a^2 -4)-\kappa K[(2\rho^2 +\lambda) a^2 +\rho^2 +2\lambda ] =0, \\
2a^2 \rho^2 (2\rho \ddot{\rho}+\dot{\rho}^2)-2a^2 \rho^2 (\Lambda \rho^2 +3a^2 -4)-\kappa K[(2\rho^2 +\lambda)a^2 -\rho^2 -2\lambda] =0, \\
a\rho^3 (\rho \ddot{a}+3\dot{\rho}\dot{a})+(a^2 -1)[\kappa K(\lambda +\rho^2)+4a^2\rho^2] =0.
\end{array} \label{full}$$
Asymptotic $a(t) \equiv 1$ case
===============================
We start from equations for the special case $a(t) \equiv 1$ after substituting it into (\[full\]):
$$\begin{array}{l}
\dot{\rho}^2 =\dac{\Lambda}{3}\rho^2 +\dac{\lambda \kappa K}{2\rho^2}+\dac{\kappa K-2}{2}\ , \\ \\ \ddot{\rho} =\dac{\Lambda }{3}\rho -\dac{\lambda \kappa K}{2\rho^3}\ .
\end{array} \label{spec_a1}$$
Let us first analyze $\Lambda=0$ case. In that case system (\[spec\_a1\]) has exact solution with integration constant which we fix from the condition $\rho\to 0$ as $t\to 0$; the resulting solution is
$$\begin{array}{l}
\rho = \dac{1}{\sqrt{2}} \sqrt{t \( ({\kappa K}- 2)t + 2\sqrt{2\lambda{\kappa K}} \)}.
\end{array} \label{sol_noL}$$
One can see that for ${\kappa K}> 2$ late-time asymptote is $\rho\propto t$ while for ${\kappa K}= 2$ solution (\[sol\_noL\]) reduces to
$$\begin{array}{l}
\rho = \sqrt{2\sqrt{2}\lambda t},
\end{array} \label{sol_noL_kk2}$$
and one can see that its late-time asymptote is $\rho\propto t^{1/2}$. Finally, for ${\kappa K}< 2$ the radicand in (\[sol\_noL\]) eventually becomes negative at some $t$ which corresponds to the recollapse; all three situations are presented in Fig. \[pre01a\]. In black we presented ${\kappa K}< 2$ case, in dashed grey – ${\kappa K}=2$ and in solid grey – ${\kappa K}> 2$ cases.
![Solutions of the $a(t)\equiv 1$ and $\Lambda = 0$ case – ${\kappa K}< 2$ in black, ${\kappa K}= 2$ in dashed grey and ${\kappa K}> 2$ in solid grey (see the text for more details).[]{data-label="pre01a"}](fig1a.eps){width="40.00000%"}
![Phase portrait of the $a(t)\equiv 1$ model with ${\kappa K}< 2$ and $\Lambda > 0$: cases with positive discriminant of (\[spec\_a1\]) (black curve), zeroth (solid grey) and negative (dashed grey) (see the text for more details).[]{data-label="pre01b"}](fig1b.eps){width="40.00000%"}
Now let us turn to $\Lambda\ne 0$ case. In that case we can reduce first of (\[spec\_a1\]) to biquadratic equation with respect to $\rho$ and find condition when its discriminant is negative – in that case $\dot\rho^2 > 0$ always. This happens if
$$\begin{array}{l}
\Lambda \geqslant \Lambda_0 = \dac{3}{8}~\dac{({\kappa K}- 2)^2}{\lambda{\kappa K}}.
\end{array} \label{L1}$$
Now let us plot $\dot\rho (\rho)$ phase portrait; we did it for ${\kappa K}< 2$ in Fig. \[pre01b\] for three cases – with the discriminant of (\[spec\_a1\]) being positive (black curve), zeroth (solid grey) and negative (dashed grey). One can see that the only smooth and nonsingular regime occurs when the discriminant is negative so if (\[L1\]) is fulfilled. In two other cases one faces finite-time future singularity at some finite $t$. So that to have smooth and nonsingular regime for ${\kappa K}< 2$ case we need $\Lambda > \Lambda_0$ from (\[L1\]). For ${\kappa K}= 2$ case, as we can see from (\[L1\]), any $\Lambda > 0$ is sufficient; ${\kappa K}> 2$ case is unaffected by (\[L1\]).
The late-time regime in this case is described by the $\rho(t)\to\infty$ branch from Fig. \[pre01b\] – it could be derived from the first of (\[spec\_a1\]) taking the mentioned limit – dynamical equation reduces to with expanding solution $\rho(t) \propto \exp (\sqrt{\Lambda/3}t)$ – usual exponential solution.
Our claim that the ${\kappa K}> 2$ case is unaffected by (\[L1\]) could be proved as follows: from the first of (\[spec\_a1\]) one can see that for ${\kappa K}\geqslant 2$ we always have $\dot\rho^2 > 0$, given $\lambda,\,\Lambda,\,{\kappa K}> 0$. Of these, $\lambda > 0$ and $K > 0$ come from the Skyrme theory and $\kappa > 0$ since we have gravitational attraction. On contrary, sometimes in different aspects of field theory $\Lambda < 0$ is considered, which gives anti de Sitter in cosmological background. One can immediately see from the first of (\[spec\_a1\]), that in $\Lambda < 0$ case at small $\rho$ we have $\dot\rho^2 > 0$ while at large $\rho$ it is negative, so the dynamics is limited and we have finite-time future singularity at some finite $t$, similar to the ${\kappa K}< 2$, $\Lambda < \Lambda_{cr}$ case. In case of negative $\Lambda$ it is true regardless of ${\kappa K}$, so in the remaining part of the paper we consider $\Lambda > 0$ only.
So to summarize our findings of the $a(t)\equiv 1$ case – if $\Lambda = 0$, there are three regimes, depending on the ${\kappa K}$: if ${\kappa K}< 2$, there is a recollapse, if ${\kappa K}= 2$, the late-time regime is power-law $\rho(t) \propto \sqrt{t}$ and if ${\kappa K}> 2$ – it is another power-law $\rho(t) \propto t$. If $\Lambda$ is nonzero and negative, then we always have recollapse; if $\Lambda > 0$ and ${\kappa K}\geqslant 2$, we always reach exponential regime $\rho(t) \propto \exp (\sqrt{\Lambda/3}t$. Finally, if $\Lambda > 0$ and ${\kappa K}< 2$, then if (\[L1\]) is fulfilled, we have exponential solution and if not – recollapse. Let us note that all these regimes we derived analytically and so they should take place for all initial conditions; our additional numerical analysis support this claim.
General case with dynamical $a(t)$
==================================
In this section we analyze the behavior of the general system (\[full\]) with dynamical $a(t)$. First we numerically analyze system (\[full\]) with $\Lambda=0$ and presented typical behavior for each case in Fig. \[pre01\](a)–(c). In (a) panel we present typical behavior for ${\kappa K}< 2$ case – one can see that it asymptotically tends to $a(t) \equiv 1$ scenario with oscillations around it. And similar to the $a(t)\equiv 1$ counterpart, our dynamical $a(t)$ case has finite-time future singularity. In (b) panel we demonstrate typical ${\kappa K}=2$ dynamics – one can see that, similar to the previous case, we have oscillations around $a(t)\equiv 1$ regime with $\rho (t) \propto t^{1/2}$ asymptotic behavior. And finally in (c) panel we present the ${\kappa K}> 2$ case with oscillations around $a(t)=1$ and $\rho (t) \propto t$ asymptotic behavior. So that we can see that in all $\Lambda = 0$ cases we have oscillatory approach to the corresponding $a(t)\equiv 1$ cases, described in the previous section. Actual evolutions curves depends on the initial conditions a bit (say, period and amplitude of oscillations depend on the initial conditions), but general behavior and late-time asymptotes are the same within the same case.
Final case to consider is general dynamical $a(t)$ with $\Lambda \ne 0$. As we just saw, with $\Lambda = 0$, dynamical $a(t)$ cases tend to their $a(t)\equiv 1$ counterparts through oscillation – the same behavior have dynamical $a(t)$ cases with nonzero $\Lambda$. So, similarly to the $a(t)\equiv 1$ case, negative $\Lambda$ always leads to the recollapse regardless of ${\kappa K}$. As we found in the previous section, $a(t)\equiv 1$ with $\Lambda > 0$ cases have either exponential regime or recollapse as a late-time attractor, and so dynamical $a(t)$ cases have the same attractor as well. So for ${\kappa K}\geqslant 2$ we always have exponential solutions with damping oscillations while for ${\kappa K}< 2$ we have either recollapse or exponential solution depending on $\Lambda$ – the same behavior we described in the previous section for $a(t)\equiv 1$ case. In Fig. \[pre01\](d) we presented typical behavior in the vicinity of separation of these two regimes – the lower regime experience recollapses while the upper reaches exponential regime; both regimes experience oscillations.
![Dynamics of $a(t) \ne 1$ and $\Lambda = 0$ case on (a)–(c) panels: ${\kappa K}< 2$ on (a) panel, ${\kappa K}= 2$ on (b) and ${\kappa K}> 2$ on (c). On (d) panel – dynamics of $a(t) \ne 1$ and $\Lambda \ne 0$ case: exponential (upper curve) and recollapse (lower) behavior (see the text for more details).[]{data-label="pre01"}](fig2_2.eps){width="100.00000%"}
![Contours of equal $\Lambda_{cr}$ on the initial conditions space $\{\rho_0,\,\dot\rho_0\}$ for $a_0 = 0.8$ on (a) panel, $a_0 = 1.0$ on (b) panel and $a_0 = 1.2$ on (c) panel. Example of $\Lambda_{cr}$ behavior with varying $a_0$ on (d) panel (see the text for more details).[]{data-label="pre02"}](fig2_t3.eps){width="100.00000%"}
In the general ${\kappa K}< 2$ case (with dynamical $a(t)$ with $\Lambda > 0$) the value for $\Lambda_{cr}$ which separates recollapse from exponential expansion (see this separation e.g. in Fig. \[pre01\](d)) is actually lower then $\Lambda_0$, given by (\[L1\]). Of course, generally $\Lambda_{cr} \leqslant \Lambda_0$, and actual values we present in Fig. \[pre02\](a)–(c). We provided contours of equal $\Lambda_{cr}$ on the initial conditions space $\{\rho_0,\,\dot\rho_0\}$ for $a_0 = 0.8$ on (a) panel, $a_0 = 1.0$ on (b) panel and $a_0 = 1.2$ on (c) panel. Levels correspond to 0.37, 0.36, 0.35, 0.34, 0.33, 0.3, 0.2 and 0.1 with decreasing blackness (so black is $\Lambda_{cr} \geqslant 0.37$ and white is $\Lambda_{cr}
\leqslant 0.1$). As these contours are plot for $\lambda=1$ and ${\kappa K}=1$ which gives $\Lambda_0 = 0.375$ derived from (\[L1\]), we can see that for $a_0=1$, presented in Fig. \[pre02\](b), $\Lambda_0$ is reached for all $\rho_0$ and $\dot\rho_0$ (so that for each $\rho_0$ exists $\dot\rho_0$ where $\Lambda_0$ is reached and vice versa) – utmost black corresponds to $\Lambda_{cr} \geqslant 0.37$. On contrary, for $a_0$ differ from 1, $\Lambda_0$ is reached for lesser measure of the initial conditions – see Fig. \[pre02\](a) for $a_0 = 0.8$ and Fig. \[pre02\](c) for $a_0 = 1.2$. We can see from these two panels that $\Lambda_0$ is shifted towards higher $\dot\rho_0$ and with growth of $|a_0 - 1|$ difference, the gap between highest $\Lambda_{cr}$ and $\Lambda_0$ also increase – in Fig. \[pre02\](d) we presented one-dimensional scan on $a_0$ – one can see that $\Lambda_{cr}$ could be orders of magnitude below $\Lambda_0$.
And a short summary of this sections findings: we found that the $\Lambda = 0$ case with generic $a(t)$ have three distinct late-time regimes which coincide with those described in the previous $a(t)\equiv 1$ section. So for ${\kappa K}< 2$ we have a recollapse, for ${\kappa K}= 2$, the late-time regime is power-law $\rho(t) \propto \sqrt{t}$ and for ${\kappa K}> 2$ – it is another power-law $\rho(t) \propto t$. In the general $\Lambda > 0$, dynamical $a(t)$ case, again, similar to the $a(t)\equiv 1$ case, we have either exponential solution or recollapse. The former of them takes place for ${\kappa K}\geqslant 2$ while the latter for ${\kappa K}< 2$ and $\Lambda < \Lambda_{cr}$. This $\Lambda_{cr} \leqslant \Lambda_0$ defined from (\[L1\]) and the actual value for $\Lambda_{cr}$ heavily depends on the initial conditions, as presented in Fig. \[pre02\]. One cannot miss the strong dependence of $\Lambda_{cr}$ on $a_0$ – more initial anisotropy – lesser value for $\Lambda$-term is needed to reach exponential expansion.
Linear stability
================
Now let us turn our attention to the stability of the solutions. In the course of paper we saw there are three nonsingular regimes: two power-law – $\rho(t) \propto \sqrt{t}$ and $\rho(t) \propto t$, and exponential $\rho(t) \propto \exp(Ht)$; all three regimes have $a(t) \to 1$. So that we perturb full system (\[full\]) around solution $a(t) = 1$ and with these three different $\rho(t)$. Linear perturbations around $a(t) = 1$ read $a\to 1+\delta a$, $\dot a \to \dot\delta a$, $\ddot a \to \ddot\delta a$, $\rho\to\rho+\delta\rho$, $\dot\rho\to\dot\rho+\dot\delta\rho$, $\ddot\rho\to\ddot\rho+\ddot\delta\rho$ and the equations on perturbations take form
$$\begin{array}{l}
4\rho^3\dot\rho\dot{\delta a} + 12\rho^2\dot\rho\dot{\delta\rho} + (-4\Lambda\rho^4 + 12\rho^2\dot\rho - 4\rho^2{\kappa K}+ 8\rho^2 - 2\lambda{\kappa K}) \delta a + \\ + (-8\Lambda\rho^3 + 12\rho\dot\rho - 6\rho{\kappa K}+ 12\rho)\delta\rho = 0, \\ \\
4\rho^3 \ddot{\delta\rho} + 4\rho^2\dot\rho\dot{\delta\rho} + (-4\Lambda\rho^4 + 8\rho^3\ddot\rho + 4\rho^2\dot\rho^2 - 4\rho^2{\kappa K}- 8\rho^2 - 2\lambda{\kappa K})\delta a + \\ + (-8\Lambda\rho^3 + 12\rho^2\ddot\rho
+ 4\rho\dot\rho^2 - 2\rho{\kappa K}+ 4\rho)\delta\rho = 0, \\ \\
\rho^4 \ddot{\delta a} + 3\rho^3\dot\rho\dot{\delta a} + (2\rho^2 {\kappa K}+ 8\rho^2 + 2\lambda{\kappa K}) \delta a = 0.
\end{array} \label{pert}$$
Last of (\[pert\]) could be solved for stability in $a$-direction. Substitution of exponential solution $\rho(t) = \rho_0 \exp (Ht)$ leads us to
$$\begin{array}{l}
\rho_0^4 \exp (4Ht) \( \ddot\delta a(t) + 3H\dot \delta a(t) \) + 2\rho_0^2 \exp (2Ht) \delta a(t) ({\kappa K}+ 4) + 2{\kappa K}\lambda \delta a(t) = 0.
\end{array} \label{pert_exp_1}$$
Using new variable $y = \dot\delta a(t)/\delta a(t)$, we can rewrite (\[pert\_exp\_1\]) as
$$\begin{array}{l}
\dot y + y^2 + 3Hy + F(t) = 0,~~F(t) = \dac{2{\kappa K}}{\rho_0^2 e^{2Ht}} + \dac{8}{\rho_0^2 e^{2Ht}} + \dac{2\lambda{\kappa K}}{\rho_0^4 e^{4Ht}},
\end{array} \label{pert_exp_2}$$
where $F(t)$ could be treated as a perturbative force acting on a system described by homogeneous equation. The solution of the homogeneous equation from (\[pert\_exp\_2\]) is
$$\begin{array}{l}
y(t) = \dac{3H}{3HC_1 e^{3Ht} - 1},
\end{array} \label{pert_exp_3}$$
and then we can solve it for $\delta a(t)$:
$$\begin{array}{l}
\delta a(t) = C_2 \( 3HC_1 - e^{-3Ht}\).
\end{array} \label{pert_exp_4}$$
The solution of the general (\[pert\_exp\_1\]) equation leads to an expression through M and W Whittaker functions [@spec_func] and generally cannot be expressed through elementary functions. But with an analysis performed in (\[pert\_exp\_2\])–(\[pert\_exp\_4\]) we describe the general behavior as follows: $F(t)$ acts as a perturbative force and generates oscillations around (\[pert\_exp\_3\]) – the solution of the homogeneous equation from (\[pert\_exp\_2\]). One can see that at $t\to\infty$ we have $F(t)\to 0$ so that at late times we can use (\[pert\_exp\_3\]) as an exact solution, which leads to (\[pert\_exp\_4\]) as a solution for original perturbation equation (\[pert\_exp\_1\]). One can note that the amplitude of perturbations does not damp to zero – as $t\to\infty$ we have $\delta a \to 3HC_1 C_2$. The reason behind it is not clear, but as the perturbations do not grow, we treat this case as stable. Our numerical analysis totally supports this description – at the beginning the solution is represented by damping oscillations, but after they decay the asymptote value is not zero but some small constant. This is the same for a wide variety of the initial conditions and the constant is also the same; though it varies for different parameters.
Now let us turn our attention to the power-law regimes. In that case the solution of the last of (\[pert\]) could be written in terms of J and Y Bessel functions and is represented by oscillations with damping amplitude, which directly points to stability, as long as solution itself exists. Solution for $\rho(t) = \rho_0 \sqrt{t/t_0}$ exists iff $\rho_0^4 \geqslant 64\lambda t_0^2$ and solution for $\rho(t) = \rho_0 (t/t_0)$ exists iff $\rho_0^2 \geqslant 2({\kappa K}+ 4)$.
To summarize, we found that the exponential solution behave a bit unusually, but we claim that we could call it stable – the perturbations experience damping oscillations and reach constant value afterwards. As they do not grow, we claim them to be stable. The power-law solutions are stable everywhere within their range of existence.
Discussion
==========
In current paper we considered Bianchi-IX cosmological model in Einstein-Skyrme system (\[full\]). The original system was simplified and considered with growth of complexity, which allows us to build semi-analytical solution. Purely analytical solutions are obtained for the simplest case with $a(t)\equiv 1$ and $\Lambda=0$ – in that case there are three possible solutions – one with recollapse for ${\kappa K}<2$ and two power-laws – $\propto t^{1/2}$ for ${\kappa K}=2$ and $\propto t$ for ${\kappa K}> 2$. All three are presented in Fig. \[pre01a\] and one cannot miss their similarity with three different Friedmann solutions from classical cosmology – with spatial curvature $k=\pm 1$ and $0$. The scales with time are different but the qualitative behavior is the same – in some sense $(2-{\kappa K})$ plays a role similar to the spatial curvature.
Further complications of the system act as modifications of the obtained exact solution. Turning $a(t)$ dynamical (but with still $\Lambda=0$) leads to oscillatory behavior like presented in Fig. \[pre01\](a)–(c). Let us remind that oscillatory behavior is a part of early Bianchi-IX universe, as discovered by Belinskij, Khalatnikov and Lifshits [@belinski]. If one keep $a(t)\equiv 1$ but make $\Lambda > 0$, then power-law regimes turn to exponential while recollapse regime turn to exponential if (\[L1\]) is satisfied; if not, they remain recollapse. Finally, if one combine both – dynamical $a(t)$ with $\Lambda > 0$, the resulting trajectories have oscillations and exponential (de Sitter) late-time asymptote for ${\kappa K}\geqslant 2$; for ${\kappa K}< 2$ one have oscillations and de Sitter if $\Lambda > \Lambda_{cr}$ and recollapse if $\Lambda < \Lambda_{cr}$; the separation between these two cases is presented in Fig. \[pre01\](d). Recollapse behavior is also encountered in anti de Sitter case – when $\Lambda < 0$ – and in this case the result is independent on ${\kappa K}$. The value for $\Lambda_{cr}$ cannot exceed $\Lambda_0$ from (\[L1\]) but could be much less (orders of magnitude), as our numerical investigation suggests. In Fig. \[pre02\] we provided the distribution of $\Lambda_{cr}$ over initial conditions space for three different $a_0$ on (a)–(c) panels and linear cut over $a_0$ on (d).
One can see that all nonsingular regimes have $a(t) \to 1$ at late times. From metric (\[BIX\]) point of view, $a(t) = 1$ solution is the most symmetric one (so that it has more Killing fields then $a(t) \ne 1$ one), so that we can see that all nonsingular regimes have symmetry restoring dynamics, and all these solutions are stable. Singular regimes, which do not possess this feature, are either ${\kappa K}< 2$ cases with $\Lambda < \Lambda_{cr}$ or $\Lambda < 0$ AdS cases; for the latter the value for ${\kappa K}$ is irrelevant.
For more physical analysis we use real values for the Skyrme coupling constants [@real]. Then one can immediately see that ${\kappa K}\lll 1$ and so ${\kappa K}< 2$ is the case. For ${\kappa K}< 2$ from (\[sol\_noL\]) one can derive the “lifetime” – with real values for couplings substituted, this time appear to be of the order of Planck time, which means that without $\Lambda$-term or some other matter sources with sufficient density, Bianchi-IX universe with Skyrme would collapse immediately. On the other hand, on this time scales the space-time cannot be described by classical means and additional investigation with involvement of quantum physics is required. Finally, if we substitute coupling constants into (\[L1\]), the resulting value for the cosmological constant appears to be in agreement with other estimates from quantum field theory, treating it as vacuum energy, and is around 120 orders of magnitude higher than the observed value (so-called “cosmological constant problem”, see e.g. [@weinberg]).
In a sense the results of current paper complement the results of [@we14], where we studied Bianchi-I and Kantowski-Sachs universes in Einstein-Skyrme system. In both papers the cosmological constant (or probably some other matter field) is necessary for viable cosmological behavior. But unlike [@we14], where we demonstrated need for the upper bound on the value of $\Lambda$-term, in current paper we found the lower bound. It is interesting that different topologies in presence of Skyrme source require either not too large or not too low values for the cosmological constant.
This finalize our study of Bianchi-IX Skyrme-Einstein system. We described its dynamics and derived conditions for different regimes to take place. Generally, Einstein-Skyrme systems are very interesting and are not much considered, probably due to their complexity, so each new result improves our understanding of cosmological hadron dynamics. In particular, these systems offer the interesting possibility to study the cosmological consequences to have conserved topological charge. Thus the present analysis is quite relevant as the energy-momentum tensor a Skyrmions of unit topological charge.
This work was supported by FAPEMA under project BPV-00038/16. We are thankful to the referee for their valuable comments which lead to the manuscript improvement.
[99]{}
N. Manton and P. Sutcliffe, [*Topological Solitons*]{}, (Cambridge University Press, Cambridge, 2007).
T. Skyrme, Proc. R. Soc. London [**A 260**]{}, 127 (1961); Proc. R. Soc. London [**A 262**]{}, 237 (1961); Nucl. Phys. [**31**]{}, 556 (1962).
, ed. by David I. Olive and Peter C. West (Cambridge University Press, Cambridge, 1999).
A.P. Balachandran, H. Gomm and R.D. Sorkin, Nucl. Phys. [**B 281**]{}, 573 (1987).
A.P. Balachandran, A. Barducci, F. Lizzi, V.G.J. Rodgers and A. Stern, Phys. Rev. Lett. [**52**]{}, 887 (1984).
A.P. Balachandran, F. Lizzi, V.G.J. Rodgers and A. Stern, Nucl. Phys. [**B 256**]{}, 525 (1985).
G. S. Adkins, C. R. Nappi and E. Witten, Nucl. Phys. [**B 228**]{}, 552 (1983).
E. Guadagnini, Nucl. Phys. [**B 236**]{}, 35 (1984).
E. Witten, Nucl. Phys. [**B 223**]{}, 422 (1983); Nucl. Phys. [**B 223**]{}, 433 (1983).
H. Pais and J. R. Stone, Phys. Rev. Lett. [**109**]{}, 151101 (2012).
U. Al Khawaja and H. Stoof, Nature [**411**]{}, 918 (2001).
J.-I. Fukuda and S. Zumer, Nature Communications [**2**]{}, 246 (2011).
S. Mühlbauer, B. Binz, F. Jonietz, C. Pfleiderer, A. Rosch, A. Neubauer, R. Georgii and P. Böni, Science [**323**]{}, 915 (2009).
A. N. Bogdanov and D. A. Yablonsky, Sov. Phys. JETP [**68**]{}, 101 (1989).
T. Sakai and S. Sugimoto, Prog. Theor. Phys. [**113**]{}, 843 (2005) \[hep-th/0412141\].
F. Canfora and P. Salgado-Rebolledo, Phys. Rev. D [**87**]{}, 045023 (2013); F. Canfora and H. Maeda, Phys. Rev. D [**87**]{}, 084049 (2013).
H. Luckock and I. Moss, Phys. Lett. [**B 176**]{}, 341 (1986); S. Droz, M. Heusler and N. Straumann, Phys. Lett. [**B 268**]{}, 371 (1991).
S. Droz, M. Heusler and N. Straumann, Phys. Lett. [**B 271**]{}, 61 (1991).
N.K. Glendenning, T. Kodama and F.R. Klinkhamer, Phys. Rev. D [**38**]{}, 3226 (1988); B. M. A. G. Piette and G. I. Probert, Phys. Rev. D [**75**]{}, 125023 (2007); G.W. Gibbons, C. M. Warnick and W. W. Wong, J. Math. Phys. [**52**]{}, 012905 (2011); S. Nelmes and B. M. A. G. Piette, Phys. Rev. D [**84**]{}, 085017 (2011).
P. Bizon and T. Chmaj, Phys. Rev. D [**58**]{}, 041501 (1998); P. Bizon, T. Chmaj and A. Rostworowski, Phys. Rev. D [**75**]{}, 121702 (2007); S. Zajac, Acta Phys. Polon. B [**40**]{}, 1617 (2009); [*ibid*]{} [**42**]{}, 249 (2011).
N. Sawado, N. Shiiki, K.-i. Maeda and T. Torii, Gen. Rel. Grav. [**36**]{}, 1361 (2004); H. Sato, N. Sawado and N. Shiiki, Phys. Rev. D [**75**]{}, 014011 (2007); T. Ioannidou, B. Kleihaus and W. Zakrzewski, Phys. Lett. [**B600**]{}, 116 (2004); [*ibid*]{} [**B643**]{}, 213 (2006); [*ibid*]{} [**B635**]{}, 161 (2006); H. Sato and N. Sawado, Phys. Lett. [**B660**]{}, 72 (2008).
F. Canfora, A. Giacomini and S. A. Pavluchenko, Phys. Rev. D [**90**]{}, 043516 (2014).
L. Parisi, N. Radicella, G. Vilasi, Phys.Rev. D [**91**]{}, 063533 (2015).
E. Ayon-Beato, F. Canfora and J. Zanelli, Phys. Lett. B [**752**]{}, 201 (2016).
F. Canfora, Phys. Rev. D [**88**]{}, 065028 (2013); F. Canfora, F. Correa and J. Zanelli, Phys. Rev. D [**90**]{}, 085002 (2014).
F. Canfora, M. D. Mauro, M. A. Kurkov and A. Naddeo, Eur. Phys. J. [**C75**]{}, 443 (2015).
V.A. Belinskij, I.M. Khalatnikov and E.M. Lifshits, Adv. Phys. [**19**]{}, 525 (1970); [*ibid*]{} [**31**]{}, 639 (1982).
M. Abramowitz and I. Stegun, [*Handbook of Mathematical Functions*]{}, New York: Dover Publications (1972); Y. Luke, [*The Special Functions and Their Approximations. Vol 1*]{}, Academic Press (1969).
G. S. Adkins, C. R. Nappi, and E. Witten, Nucl. Phys. B [**228**]{}, 552 (1983).
S. Weinberg, Rev. Mod. Phys. [**61**]{}, 1 (1989).
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- Denis Denisov
- Neil Walton
bibliography:
- 'SAMBA.bib'
title: 'Regret Analysis of a Markov Policy Gradient Algorithm for Multi-arm Bandits'
---
Acknowledgments. {#acknowledgments. .unnumbered}
================
Bandits is a new area for both authors. So we are grateful to Tor Lattimore for references, comments and suggestions on the positioning of this work.
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- |
\
INFN-Sezione di Pisa\
E-mail:
title: 'Long-Range Models of Modified Gravity and Their Agreement with Solar System and Double Pulsar Data'
---
\#1[ \#1 ]{} \#1\#2[\_[\#1]{}\^[\#2]{}]{} \#1[eq. (\[\#1\])]{} \#1\#2[eq. (\[\#1\])-eq. (\[\#2\])]{} \#1[Eq. (\[\#1\])]{} \#1\#2[Eq. (\[\#1\])-eq. (\[\#2\])]{} \#1 \#1\#2[\^[\#2]{}\_[.\#1]{}]{}
\#1\#2 \#1\#2 \#1\#2
|[$$\begin{aligned}
}
\def\ear{\end{aligned}$$]{} \#1\#2[[\#1\#2]{}]{} \#1 \#1[\[\#1\]]{}
\#1[“\#1”]{}
2[$\mathcal{O}(c^{-2})$]{}
\#1
Motivations for Long-Range Modifications of Gravity
===================================================
Historically, the first attempt to modify the laws of gravity commonly accepted at that time was due to Laplace [@Lap] who, in 1805, tried to add a velocity-dependent term to the standard inverse-square law of Newton to account for the finite velocity of propagation of gravity. But this work did not find echo practically until the surroundings of 1880, when a series of works to estimate the gravitational finite propagation speed began. Such attempts to find deviations from the Newtonian inverse-square law of gravitation were performed to explain the anomalous secular precession of Mercury’s perihelion, discovered by Le Verrier [@lev], without invoking undetected (baryonic) matter like the hypothesized planet Vulcan: for example, Hall [@hal] noted that he could account for Mercury’s precession if the law of gravity, instead of falling off as $1/r^2$, actually falls of as $1/r^k$ with $k = 2.00000016$. However, such an idea was not found to be very appealing, since it conflicts with basic conservation laws, e.g., Gauss Law, unless one also postulates a correspondingly modified metric for space. Other historical attempts to modify Newton’s law of gravitation to account for the Mercury’s perihelion behavior yielded velocity-dependent additional terms: for a review of them see Ref. [@gine] and references therein. Such attempts practically ceased after the successful explanation of the perihelion rate of Mercury by Einstein [@ein] in terms of his tensorial general theory of relativity: an exception is represented by Manev [@man] who, with a $1/r^2$ correction to the Newtonian potential, was able to reproduce the anomalous apsidal precession of Mercury.
Moving to more recent times, in the modern framework of the challenge of unifying gravity with the other three fundamental interactions of Nature, it was realized that possible new phenomena could show up just as deviations from the Newtonian inverse-square law of gravitation [@nuovo1; @nuovo2]. In general, they would occur at submillimeter length scales. Concerning general relativity, traditionally, corrections to it, in the form of modifications of the Einstein-Hilbert action in order to include higher-order curvature invariants with respect to the Ricci scalar, were considered to be important only at scales close to the Planck one and, consequently, in the early universe or near black hole singularities[^1]; see, e.g., Ref [@Prima] and references therein. It was not expected that such corrections could affect the gravitational phenomenology at low energies, and consequently at larger, macroscopic scales.
Thus, why considering modifications of the standard laws of gravity at large, astronomical, astrophysical and cosmological as well, as done in recent years? To try to accommodate some recently observed phenomena, occurring at very different scales ranging from solar system to cosmological distances, which, at present, have not yet found fully satisfactorily mundane explanations in terms of conventional physics, gravitational or not [@lamm; @Berto; @megaberto]. Below we list just some of such anomalous effects.
- The flyby anomaly. It consists of unexplained changes of the asymptotic outgoing velocities of some spacecraft (Galileo, NEAR, Cassini, and MESSENGER) that occurred at their closest approaches with the Earth [@flyby1; @flyby2]. Is it due to [ conventional non-conservative effects]{}, or are [ modifications of the laws of gravity]{} responsible of it?
- The [anomalous]{} [perihelion]{} precession of [Saturn]{} [@IorSat] detected by processing almost one century of planetary observations with the inclusion of the latest radiometric data of the Cassini spacecraft [@Pit08]: is it a [[data processing artifact]{}]{} or a genuine [[ physical effect]{}]{}?
- The [Pioneer anomaly]{}. It is an unexplained [acceleration]{} ${ A_{\rm Pio} = (8.74\pm 1.33)\times 10^{-10}} $ [ m s$^{-2}$ ]{} approximately directed towards the Sun affecting the [Pioneer 10/11]{} spacecraft after they passed the [20 AU]{} threshold in the [solar system]{} [@Pio1; @Pio2]. Is it induced by some [mundane non-conservative effects]{}, or is it a sign of [ modifications of the laws of gravity]{}?
- The dark matter issue. In many astrophysical systems like, e.g., spiral galaxies and clusters of galaxies a discrepancy between the observed kinematics of their exterior parts and the predicted one on the basis of the Newtonian dynamics and the matter detected from the emitted electromagnetic radiation (visible stars and gas clouds) was present since the pioneering studies by Zwicky [@Zwi] on the Coma cluster, and by Kahn and Woltjer [@Kahn], Bosma [@Bos], and Rubin and coworkers [@Rub; @Rub83] on individual galaxies. More precisely, such an effect shows up in the galactic velocity rotation curves [@flatt] whose typical pattern after a few kpc from the center differs from the Keplerian $1/\sqrt{r}$ fall-off expected from the usual dynamics applied to the electromagnetically-observed matter. Does the cause of such a phenomenology reside in the action of still undetected (non-baryonic) dark matter whose dynamics is governed by the usual Newtonian laws of gravitation, or have they to be modified?
- The dark energy issue. In recent years, an increasing amount of observational evidence has accumulated pointing toward the fact that the universe has entered a phase of accelerating expansion. Some of such observations are of direct, geometrical nature: standard candles like the supernov[æ]{} SnIa [@sna1; @sna2], gamma ray bursts [@gamma] and standard rulers like the CMB sound horizon [@CMB1; @CMB2]. Other ones are of dynamical nature like the rate of growth of cosmological perturbations [@grow] probed by the redshift distortion factor or by weak lensing [@Fu]. All these observations are converging towards a confirmation of the accelerating expansion of the universe, assumed homogeneous. They are successfully fitted by the simplest cosmological model predicting accelerating cosmic expansion: its ingredients are the the assumptions of flatness, validity of general relativity, the presence of a cosmological constant $\Lambda$, identified with an unknown and still directly undetected form of energy (named dark energy for these reasons), and Cold Dark Matter ($\Lambda$CDM). However, for some puzzles of the $\Lambda$CDM cosmology, see Ref. [@Peri]. Contrary to the $\Lambda$CDM paradigm, is it possible to accommodate the aforementioned observations by invoking modifications of the standard laws of gravity?
Some of the models of modified gravity that have been proposed to address the aforementioned phenomenology are listed below.
- Dvali-Gabadadze-Porrati ([DGP]{}) braneworld model [@DGP]. In it our universe is a (3+1) space-time brane embedded in a five-dimensional Minkowskian bulk. All the particles and fields of our experience are constrained to remain on the brane apart from gravity which is free to explore the empty bulk. Beyond a certain threshold $r_0$, which is a free-parameter of the theory and is fixed by observations to about 5 Gpc, gravity experiences strong modifications with respect to the usual four-dimensional Newton-Einstein picture: they allow to explain the observed acceleration of the expansion of the Universe without resorting to the concept of dark energy.
- MOdified Gravity ([MOG]{}) by Moffat [@MOG]. It is a fully covariant theory of gravity which is based on the existence of a massive vector field coupled universally to matter. MOG yields a Yukawa-like modification of gravity with three constants which, in the most general case, are running; they are present in the theory’s action as scalar fields which represent the gravitational constant, the vector field coupling constant, and the vector field mass. It has used to successfully describe various observational phenomena on astrophysical and cosmological scales without resorting to dark matter [@Mof2].
- MOdified Newtonian Dynamcs ([MOND]{}) [@Mil83a; @Mil83b; @Mil83c]. It is a non-linear theory of gravity which predicts departures from the standard Newtonian inverse-square law at a characteristic acceleration scale [@Bege] $A_0=1.27\times 10^{-10}$ m s$^{-2}$ below which the gravitational acceleration gets a $\approx 1/r$ behavior. MOND was proposed to explain certain features of the motion of ordinary electromagnetically detectable matter in galaxies and of galaxies in galactic clusters without resorting to exotic forms of still undetected dark matter.
- ${f(R)}$ models [@super]. These theories come about by a straightforward generalization of the Lagrangian in the Einstein-Hilbert action in which the Ricci scalar $R$ is replaced by a general analytical function $f(R)$ of $R$. They have mainly been used in cosmological and astrophysics scenarios [@capoz1].
- [Curvature Invariants]{} models. They encompass inverse powers [@Nav1; @Nav2] and logarithm [@Nav3] of some curvature invariants in the Einstein-Hilbert action and have been used for tackling the dark energy-dark matter problems.
- [Yukawa]{}-like models. There are many theoretical frameworks yielding such a modification of the Newtonian inverse-square law [@Kra; @Adel; @megaberto]. Models encompassing Yukawa-type extra-accelerations have been used for a variety of applications ranging from solar system effects like the Pioneer anomaly [@MOG] to astrophysical and cosmological scenarios [@Koch; @Ame; @Sea; @je; @Shi; @Ser06].
- [Hooke]{}-like models. With such a definition we mean models of gravity introducing an additional term proportional to the distance $r$. An important case is given by the Schwarzschild-de Sitter spacetime [@desi] which yields a correction to the Newtonian inverse-square law proportional to $\Lambda r$ [@Rind], where $\Lambda$ is an uniform cosmological constant. Another example is given by the extra-acceleration proposed by Jaekel and Reynaud [@Jekel1] to explain the Pioneer anomaly.
- [Pioneer]{}-like models. The simplest one consists of postulating a constant and uniform acceleration radially directed towards the Sun and having the same magnitude of $A_{\rm Pio}$ existing in the outer regions of the solar system at heliocentric distances $r\geq 20$ AU. Other forms have been postulated for it, both distance-dependent [@Jekel1; @mofPio] and velocity-dependent [@Jekel2; @Jekel3; @Sta08].
Other models of modified gravity that we will not consider here are the Einstein-Aether theory [@Mat], Tensor-Vector-Scalar (TeVeS) [@teves] and braneworld gravity [@brane].
The only motivations for the aforementioned models are just the phenomena themselves for which they have been introduced. Thus, such models must be put on the test in different scenarios by devising independent observational checks. In particular, they must not exhibit discrepancies with the well-tested standard laws of gravity in local astronomical systems like, e.g., our solar system and the double pulsar PSR J0737-3039 [@psr1; @psr2].
Our Method for Testing Modified Models of Gravity in Astronomical Scenarios
===========================================================================
In general, a given Long Range Modified Model of Gravity (LRMMG) yields [predictions]{} ${\mathcal{P}}$ for certain [observable]{} effects ${\mathcal{O}}$ of the form ,where $K\rightarrow 0$ implies no modifications of gravity, and ${g}$ is a function of the geometrical configuration of the system adopted [ characteristic of the LRMMG considered]{}; the fact that the LRMMG parameter $K$ enters as a multiplicative factor in the predicted effects $\mathcal{P}$ will be very important for us, as we will see below. For example, for the solar system’s planetary longitudes of the [perihelia]{} ${\varpi}$, it turns out with $a$ semimajor axis and $e$ eccentricity of the planetary orbit considered. [Corrections]{} ${\Delta\dot\varpi}$ to the usual Newtonian/Einsteinian perihelion precessions have been recently estimated for [ several]{} planets of the solar system by [E.V. Pitjeva]{} [@Pit05; @Pit; @Pit08] by fitting 100 yr of observations of several kinds with the force models of various versions of the [EPM]{} ephemerides (EPM2004 [@PitSS] EPM2006 [@Pit], EPM2008 [@Pit08]). Since they do [ not]{} include any LRMMGs, the corrections $\Delta\dot\varpi$, [by construction]{}, account for them, so that they will be [our]{} ${\mathcal{O}}$: they are listed in Table \[chebolas\].
[@cccc@]{} [Mercury]{} & [Venus]{} & [Earth]{} & [Mars]{}\
$-36\pm 50\ [42]$ & $-4\pm 5\ [1]$ & $-2\pm 4\ [1]$ & $1\pm 5\ [1]$\
By [directly]{} comparing ${\Delta\dot\varpi}$ to the predicted anomalous perihelion precession ${\mathcal{P}=K g(a,e)}$ for [each planet separately]{} allows to put [upper bounds]{} on ${|K|}$ since $\Delta\dot\varpi$ are compatible with zero, according to Table \[chebolas\]. This approach is good for test the hypothesis ${K=0}$. The hypothesis ${K\neq 0}$ can be tested by taking the [ratios]{} ${\Delta\dot\varpi}$ for [different pairs]{} of planets A and B, and comparing them to the [predicted ratios]{} :$\xi_{\rm AB}$, by construction, does [not]{} explicitly depend on ${K}$, but it still retains a functional dependence on $a$ and $e$ [ typical of the LRMMG]{} considered. If within the observational errors, i.e. if ,the LRMMG considered is [severely challenged]{}. Of course, the uncertainty in the ratios $\Pi_{\rm AB}$ has to be evaluated in a realistic and conservative way to reduce the risk of unsound conclusions; see the discussion in Section \[finale\]. We quote the ratios $\Pi_{\rm AB}$ for all the pairs of the inner planets in Table \[chebolas2\].
[@llrr@]{} A & B & $\Pi_{\rm AB}$ & $\delta\Pi_{\rm AB}$\
Mercury & Venus & $9$ & 24\
Venus & Mercury & $0.1$ & 0.3\
Mercury & Earth & $18$ & 61\
Earth & Mercury& $0.06$ & 0.19\
Mercury & Mars & $-36$ & 230\
Mars & Mercury& $-0.03$ & 0.18\
Venus & Earth & $2$ & 6.5\
Earth & Venus& $0.5$ & 1.6\
Venus & Mars & $-4$ & 25\
Mars & Venus & $-0.2$ & 1.6\
Earth & Mars & $-2$ & 14\
Mars & Earth & $-0.5$ & 3.5\
An analogous approach can be followed with the [double pulsar]{} PSR J0737-3039. Now, one observable is ${ \Delta\dot\omega }$, i.e. the difference between the [phenomenologically determined]{} periastron precession ${ \dot\omega_{\rm meas} }$ and the [computed]{} rate ${ \dot\omega_{\rm 1PN} }$ for the general relativistic 1PN rate (it is analogous to the well-known Mercury’s perihelion precession by Einstein). Another observable is ${ \Delta P }$, i.e. the difference between the [phenomenologically determined]{} orbital period ${ P_{\rm b} }$ and the [computed]{} Keplerian one ${ P^{\rm Kep} }$. Thus, the [observation-based]{} ratio can be compared with the [predictions]{} $\mathcal{P}$ for the same ratio by various [LRMMGs]{}; again, if they are [equal]{} within the errors, LRRMMG [ passes]{} the test, [otherwise]{} it is [ challenged]{}.
Results from Local Systems
==========================
Dvali-Gabadadze-Porrati (DGP) Braneworld Model
----------------------------------------------
The [ DGP braneworld]{} model [@DGP], put forth to explain the cosmic acceleration without resorting to dark energy, predicts an anomalous [perihelion precession]{} independent of the semimajor axis [@lue; @DV; @IorDGP] , r\_05 [Gpc]{},where the $\mp$ sign are related to the two different cosmological branches of the model. Thus, for it. By linearly adding the absolute values of the uncertainties in $\Delta\dot\varpi$, it turns out for the pairs A=Mars, B=Mercury and A=Earth, B=Mercury [@IorAHEP] $$\begin{aligned}
\Psi_{{\rm Mars\ Mercury}} &=& {1.0\pm 0.2},\ {\rm ruled\ out\ at}\ {5\sigma}\ {\rm level}\\\nonumber\\
%
\Psi_{{\rm Earth\ Mercury}} &=& {0.9\pm 0.2},\ {\rm ruled\ out\ at}\ {4.5\sigma}\ {\rm level}\end{aligned}$$
In the case of the [double pulsar]{} [@Iorpul], the [predicted]{} ratio ${\mathcal{R}_{\rm DGP}}$ between $\dot\omega_{\rm DGP}$ and $P_{\rm DGP}$ is .Instead, the [observed]{} ratio ${\mathcal{R}_{\rm obs}}$ between $\Delta\dot\omega\equiv \dot\omega_{\rm meas}-\dot\omega_{\rm 1PN}$ and $\Delta P=P_{\rm b}-P^{\rm Kep}$ is .They are [not]{} compatible at ${14\sigma}$ level.
MOdified Gravity (MOG)
----------------------
[MOG]{} [@MOG] predicts a [Yukawa-type]{} extra-acceleration [@mog1; @mog2] ,where $\alpha$ and $\mu$ have been fitted to a set of astrophysical galactic data [@mog2] in the framework of the searches for an explanation of the flat rotation curves of galaxies without resorting to dark matter. For a Sun-planet system we have a perihelion extra-rate (${1/\mu = 33,000}$ [AU]{}) [@IorMOG] .The inner planets yield [@IorMOG] $$\begin{aligned}
\Psi_{{\rm Venus\ Mercury}} &=& {1.3\pm 0.3},\ {\rm ruled\ out\ at\ } {4\sigma}\ {\rm level}\\\nonumber\\
%
\Psi_{{\rm Earth\ Mercury}} &=& {1.6\pm 0.2},\ {\rm ruled\ out\ at\ } {8\sigma}\ {\rm level}\\\nonumber\\
%
\Psi_{{\rm Mars\ Mercury}} &=& {2\pm 0.2},\ {\rm ruled\ out\ at\ } {10\sigma}\ {\rm level}\end{aligned}$$
MOdified Newtonian Dynamics (MOND)
-----------------------------------
MOND [@Mil83a; @Mil83b; @Mil83c] predicts [acceleration-dependent]{} modifications of the Newtonian inverse-square law to explain the [almost flat galactic rotation curves]{} where galaxies data fitting yields [@Bege] ${A_0=1.2\times 10^{-10}\ {\rm m\ s}^{-2} }$ and ${\mu(x)}$ is an [interpolating function]{} whose most widely used forms are with 1, x1,and x, x1. observations are, strictly speaking, tests of the [form]{} of the interpolating function in the [large acceleration limit]{} (${x\gg 1}$).
The anomalous [perihelion precession]{} predicted by MOND is, in the large acceleration limit (${x\gg 1}$), [@Mil83a; @Tal; @Ser] ;${k_0=1,\ h=1}$ correspond to ;${ k_0=1/2,\ h=2}$ correspond to .It turns out [@IorMOND] that the [ratios]{} ${\Pi_{\rm AB}}$ of the perihelia, [independent]{} by construction of ${k_0}$, for the pairs [A=Mars, B=Mercury]{} and [A=Earth, B=Mercury]{} [rule out]{} the MOND perihelion precessions for $ {1\leq h \leq 2}$ at [several sigma]{} level.
Extended $f(R)$ theories
------------------------
Extended theories of gravity with [@capoz2] ,where ${R}$ is the [Ricci scalar]{}, yield [ power-law corrections]{} to the Newton’s law which, for ${\beta = 0.817}$, obtained by fitting the SNeIa Hubble diagram, yields good results in fitting some galactic rotation curves; $\beta$ is related to the exponent $k$ of the Ricci scalar $R$. The induced perihelion precession is [@IorRug] ,where $G(e,\beta)$ is a function of the eccentricity and $\beta$. The resulting $\Psi_{\rm AB}$ are [not]{} compatible with zero for [many]{} pairs A and B of inner and outer planets at [several sigma]{} level [@IorRug].
Inverse powers and logarithm of some curvature invariants models
----------------------------------------------------------------
[ Inverse powers]{} of [curvature invariants]{} in the action lead to [@Nav1; @Nav2] .For ${k=1}$ and the [Sun]{}, ${r_c=10}$ [pc]{}, so that Thus, the [inner]{} planets yield [@IorAHEP] $$\begin{aligned}
\Psi_{{\rm Mars\ Mercury}} &=& {10^5\pm 0.1},\ {\rm ruled\ out\ at\ } {10^6 \sigma}\ {\rm level}\\ \nonumber \\
%
\Psi_{{\rm Mars\ Earth}} &=& {38\pm 3.5},\ {\rm ruled\ out\ at\ } {11\sigma}\ {\rm level} \\ \nonumber \\
%
\Psi_{{\rm Earth\ Mercury}} &=& {10^3\pm 0.2},\ {\rm ruled\ out\ at\ } {10^4 \sigma}\ {\rm level}\end{aligned}$$ [Logarithm]{} of some [curvature invariants]{} in the action, able to modify gravity at [MOND scales]{} so to [jointly]{} treat [Dark Matter]{} and [Dark Energy]{}, induces [@Nav3] ,so that The pairs [A=Mars, B=Mercury]{} and [A=Earth, B=Mercury]{} yield [@IorAHEP] $$\begin{aligned}
\Psi_{{\rm Mars\ Mercury}} &= &{30.7\pm 0.1},\ {\rm ruled\ out\ at\ } {31 \sigma}\ {\rm level}\\ \nonumber \\
%
\Psi_{{\rm Earth\ Mercury}} &=& {10.6\pm 0.2},\ {\rm ruled\ out\ at\ } {53 \sigma}\ {\rm level}\end{aligned}$$
Yukawa-like theories
---------------------
Many theoretical frameworks [@Kra; @Adel; @megaberto] yield a Yukawa-type acceleration .For ${\lambda\gtrsim ae}$, the induced perihelion rate is [@Iyuk] ,so that for a pair of planet A and B .
The pair [A=Earth, B=Mercury]{} allows to obtain [@Iyuk] .Solving for ${\alpha}$ and using $\Delta\dot\varpi$ and $a$ for [ Mars ]{} yields [@Iyuk] ,in which we have used the value for ${\lambda}$ obtained from [Earth]{} and [Mercury]{}.
Hooke-type theories
-------------------
A Hooke-like acceleration is induced, e.g., by a uniform [cosmological constant]{} ${\Lambda}$ in the [Schwarzschild-De Sitter]{} spacetime [@Rind] .The induced perihelion shift is [@Kerr] .Some pairs of planets yield [@IorH] $$\begin{aligned}
\Psi_{{\rm Mars\ Mercury}} &=& {7.8\pm 0.2},\ {\rm ruled\ out\ at\ } {39 \sigma}\ {\rm level},\\ \nonumber\\
%
\Psi_{{\rm Earth\ Mercury}} &=& {4.1\pm 0.2},\ {\rm ruled\ out\ at\ } {20 \sigma}\ {\rm level},\\ \nonumber\\
%
\Psi_{{\rm Jupiter\ Mercury}} &=& {51\pm 12},\ {\rm ruled\ out\ at\ } {4 \sigma}\ {\rm level}\end{aligned}$$
In the case of the [double pulsar]{}, the [predicted]{} ratio ${\mathcal{R}_{\Lambda}}$ between $\dot\omega_{\Lambda}$ and $P_{\Lambda}$ is [@Iorpul] .Instead, the [observed]{} ratio ${\mathcal{R}_{\rm obs}}$ between $\Delta\dot\omega\equiv \dot\omega_{\rm meas}-\dot\omega_{\rm 1PN}$ and $\Delta P=P_{\rm b}-P^{\rm Kep}$ is [@Iorpul] .They are [not]{} compatible at ${11\sigma}$ level.
The Pioneer anomaly
-------------------
If the [Pioneer Anomaly]{} [@Pio1; @Pio2] was of [gravitational origin]{}, the exotic force causing it [should also act on the planets]{} of the Solar System, at least on those moving in the spatial regions in which it manifested itself in its presently known form. A [constant and uniform radial]{} acceleration with the same magnitude of that causing the Pioneer Anomaly would induce orbital effects [too large]{} [@IorGiu] to have escaped so far detection, as shown by the ${\Delta\dot\varpi}$ estimated with the [EPM2006]{} ephemerides [@Pit] for [Jupiter, Saturn and Uranus]{} [@IorMGM; @IorJgP], and from the [residuals]{} of right ascension $\alpha$ and declination $\delta$ computed with the [JPL]{} ephemerides [@IorFoP]. Also some recently proposed [velocity-dependent]{} exotic forces [@Jekel2; @Jekel3; @Sta08] are [ruled out]{} by planetary observations [@IorFoP; @IorIJ]. For other works dealing with the same problem, see Refs. [@Page06; @Tangen; @Wal; @Page09].
The anomalous perihelion precession of Saturn
---------------------------------------------
[E.V. Pitjeva]{} has recently fitted almost one century of planetary data of all kinds, including also [3 years of radiotechnical data from Cassini]{}, with the [EPM2008]{} ephemerides [@Pit08]. She estimated (E.V. Pitjeva, private communication, 2008) a statistically significant non-zero correction ,while the [formal]{} error of the fit is [0.0007 arcsec cy$^{-1}$ ]{}. However, applying the standard re-scaling by a factor 10 of the optical only observations there is also, in principle, the possibility that the uncertainty can be as large as 0.007 arcsec cy$^{-1}$ (E.V. Pitjeva, private communication, 2008). Previous results obtained with the EPM2006 ephemerides [@Pit], which did not include Cassini data, yielded \_[Saturn]{} = -0.920.29 [arcsec cy\^[-1]{}]{} ([formal error]{}).Is it really a genuine physical effect needing explanation, or is it some artifact of the data reduction procedure? [None]{} of the exotic models examined so far [is able to accommodate it]{} [@IorSat].
The general relativistic Lense-Thirring effect
----------------------------------------------
Until now we have only dealt with putative modifications of the standard Newtonian/Einsteinian laws of gravity. In fact, the estimated corrections $\Delta\dot\varpi$ to the usual rates of the perihelia account, by construction, also for a standard general relativistic effect which has not been included in the dynamical force models of the EPM ephemerides used to determine them: it is the gravitomagentic Lense-Thirring effect [@LT] consisting of secular precessions of the form \_[LT]{}=,where $S$ is the Sun’s angular momentum and $I$ is the inclination of the planetary orbital plane to the Sun’s equator which is quite small for all the inner planets. The Lense-Thirring precessions for the inner planets are of the order of $10^{-3}-10^{-4}$ arcsec cy$^{-1}$, and their direct measurability has been discussed in, e.g., Ref. [@POS]. It is interesting to check if the unmodelled Lense-Thirring precessions pass the test of the ratio of the perihelia, with \^[LT]{}\_[AB]{}.By comparing Table \[chebolas2\] with Table \[chebolas3\], it can be seen that $\Psi_{\rm AB}$ is compatible with zero for all the pairs A B of inner planets, contrary to all the models of modified gravity examined so far.
[@lll@]{} A & B & $\xi^{\rm LT}_{\rm AB}$\
Mercury & Venus & $7$\
Venus & Mercury & $0.1$\
Mercury & Earth & $18$\
Earth & Mercury& $0.05$\
Mercury & Mars & $64$\
Mars & Mercury& $0.01$\
Venus & Earth & $3$\
Earth & Venus& $0.4$\
Venus & Mars & $9$\
Mars & Venus & $0.1$\
Earth & Mars & $3.5$\
Mars & Earth & $0.3$\
Discussion and conclusions
==========================
Some technical aspects
----------------------
[If]{} and [when]{} [other]{} teams of astronomers will [independently]{} estimate their own corrections $\Delta\dot\varpi$ with [different]{} ephemerides, it will be possible to fruitfully repeat all the tests presented here. In doing them we [always]{} used the formal errors in $\Delta\dot\varpi$ [re-scaled]{} by a factor ${\approx 2-5}$ for the [ inner]{} planets and up to [10]{} times for the [outer]{} planets for which mainly optical data have been used. Moreover, in view of the [correlations]{} among the estimated $\Delta\dot\varpi$, we always [linearly]{} propagated their [errors]{}, instead of summing them in quadrature, by getting \_[AB]{}|\_[AB]{}|( + ).For example, the correlations between the corrections $\Delta\dot\varpi$ of Mercury and the Earth are as large as $20\%$ (Pitjeva, private communication, 2005). Concerning the uncertainties in $\xi_{\rm AB}$, in principle, they should have been computed by propagating the errors in the semimajor axes $a$ and the eccentricities $e$ of the planets A and B entering them for each model considered. However, they are quite negligible because the relative (formal) uncertainties in the semimajor axes and eccentricities of the inner planets are all of the order of $10^{-12}$ and $10^{-9}-10^{-11}$, respectively, according to Table 3 of Ref. [@Pit]. Processing [ more ranging data from Cassini]{} will help in clarifying if [the perihelion precession of Saturn]{} can really be considered as a [genuine physical effect]{}.
Conclusions
-----------
- [ All]{} the long-range modified models of gravity examined here are [severely challenged]{} either by [the ratios of the perihelia]{} of different pairs of solar system’s planets or by the [double pulsar]{} combined data for the periastron and the orbital period. Both such kinds of ratios cancel out the small multiplicative parameters which directly account for the various exotic models considered, but they still retain a pattern which is characteristic of the model tested through the orbital parameters of the systems used. Only the general relativistic Lense-Thirring effect, not modelled in the EPM ephemerides and, thus, accounted for, in principle, by the estimated corrections $\Delta\dot\varpi$, passes the test of the ratio of the perihelia.
- [None]{} of the exotic models examined [is able to explain the anomalous perihelion precession of Saturn]{}. The analysis of [more Cassini data]{} will help in clarifying if it is really incompatible with zero at some statistically significant level, thus requiring an explanation in terms of some physical phenomena, or if it is some artifact of the data reduction procedure.
[99]{}
P.S. de Laplace, *Traité de Mécanique Céleste*, tome IV, livre X, chapitre 7, Courcier, Paris, 1805.
U. Le Verrier, *Compt. Rendus de l’ Académie des Sciences (Paris)* [**49**]{}, 379 (1859).
A. Hall, *Astron. J.* [**14**]{}, 49 (1894).
J. Giné, *Chaos, Solitons & Fractals* [**38**]{}, 1004 (2008), \[[physics/0510086]{}\].
A. Einstein, *Sitzungsber. Kön. Preuss. Akad. Wiss. (Berlin)* [**47**]{}, 831 (1915).
G. Manev, *Compt. Rendus de l’Acad´emie des Sciences (Paris)* [**190**]{}, 1374 (1930).
E. Fischbach et al., *Metrologia* [**29**]{}, 213 (1992).
Y. Fujii, *Int. J. Mod. Phys. A* [**6**]{}, 3505 (1991).
H.J. Schmidt, *Int. J. Geom. Meth. Mod. Phys* [**4**]{}, 209 (2007), \[[gr-qc/0602017]{}\].
C. Lämmerzahl, O. Preuss and H. Dittus, [*Is the physics within the solar system really understood?*]{}, in H. Dittus, C. Lämmerzahl and S.G. Turyshev (ed.) [*Lasers, Clocks and Drag Free Control: Exploration of Relativistic Gravity in Space*]{}, Springer, Berlin, 2008, pp. 75-104, \[[gr-qc/0604052]{}\].
O. Bertolami, *Int. J. Mod. Phys. D* [**16**]{}, 2003 (2008), \[[astro-ph/0608276]{}\].
O. Bertolami, J. Paramos, S.G. Turyshev, [*General Theory of Relativity: Will it survive the next decade?*]{} in H. Dittus, C. Lämmerzahl and S.G. Turyshev (ed.) [*Lasers, Clocks and Drag Free Control: Exploration of Relativistic Gravity in Space*]{}, Springer, Berlin, 2008, pp. 27-74, \[[gr-qc/0602016]{}\].
J.D. Anderson, J.K. Campbell and M.M. Nieto, *New Astron.* [**12**]{}, 383 (2007), \[[astro-ph/0608087]{}\].
J.D. Anderson, J.K. Campbell, J.E. Ekelund, J. Ellis and J.F. Jordan, *Phys. Rev. Lett.* [**100**]{}, 091102 (2008).
L. Iorio, *Astron. J.* [**137**]{}, 3615 (2009), \[[arXiv:0811.0756]{}\].
E.V. Pitjeva, [*Ephemerides EPM2008: the updated models, constants, data*]{} paper presented at [*Journ$\acute{e}$es and X Lohrmann-Kolloquium 22-24 September 2008, Dresden, Germany*]{}.
J.D. Anderson, P.A. Laing, E.L. Lau, A.S. Liu, M.M. Nieto and S.G. Turyshev, *Phys. Rev. Lett.* [**81**]{}, 2858 (1998), \[[gr-qc/9808081]{}\].
J.D. Anderson, P.A. Laing, E.L. Lau, A.S. Liu, M.M. Nieto and S.G. Turyshev, *Phys. Rev. D* [**65**]{}, 082004 (2002), \[[gr-qc/0104064]{}\].
F. Zwicky, *Helvetica Physica Acta* [**6**]{}, 110 (1933).
F.D. Kahn and L. Woltjer, 1959, *Astrophys. J.* [**130**]{}, 705 (1959).
A. Bosma, *Astron. J.* [**86**]{}, 1791 (1981).
V. Rubin, W. Ford, N. Thonnard and D. Burstein, 1982, *Astroph. J.* [**261**]{}, 439 (1982).
V. Rubin, *Science* [**220**]{}, 1339 (1983).
M. Persic, P. Salucci and F. Stel, *Mon. Not. Roy. Astron. Soc.* [**281**]{}, 27 (1996), \[[astro-ph/9506004]{}\].
A.G. Riess et al., *Astron. J.* [**116**]{}, 1009 (1998),\[[astro-ph/9805201]{}\].
S. Perlmutter et al., *Astrophys. J.* [**517**]{}, 565 (1999), \[[astro-ph/9812133]{}\].
A.S. Friedman and J.S. Bloom, *Astrophys. J.* [**627**]{}, 1 (2005), \[[astro-ph/0408413]{}\].
W.J. Percival et al., *Mon. Not. Roy. Astron. Soc.* [**381**]{}, 1053 (2007), \[[arXiv:0705.3323]{}\].
E. Komatsu et al., *Astrophys. J. Suppl.* [**180**]{}, 330 (2009), \[[arXiv:0803.0547]{}\].
G. Efstathiou et al., *Mon. Not. Roy. Astron. Soc.* [**330**]{}, L29 (2002), \[[astro-ph/0109152]{}\]
L. Fu et al., *Astron. Astrophys.* [**479**]{}, 1 (2008), \[[arXiv:0712.0884]{}\].
L. Perivolaropoulos, Six Puzzles for LCDM Cosmology, in P.M. Lavrov (ed.) *The problems of modern cosmology. A volume in honour of Professor S.D. Odintsov on the occasion of his 50th birthday*, Tomsk State Pedagogical University Press, Tomsk, 2009, pp. 245-254, \[[arXiv:0811.4684]{}\].
G. Dvali, G. Gabadadze and M. Porrati, *Phys. Lett. B* [**485**]{}, 208 (2000), \[[hep-th/0005016]{}\].
J.W. Moffat, *J. Cosmol. Astropart. Phys.* [**2006**]{}, 004 (2006), \[[gr-qc/0506021]{}\].
J.W. Moffat, *Int. J. Mod. Phys. D* [**16**]{}, 2075 (2008), \[[gr-qc/0608074]{}\].
M. Milgrom, *Astrophys. J.* [**270**]{}, 365 (1983).
M. Milgrom, *Astrophys. J.* [**270**]{}, 371 (1983).
M. Milgrom, *Astrophys. J.* [**270**]{}, 384 (1983).
K. Begeman, A. Broeils and R. Sanders, *Mon. Not. Roy. Astron. Soc.* [**249**]{}, 523 (1991).
T.P. Sotiriou and V. Faraoni, \[[arXiv:0805.1726]{}\].
S. Capozziello and M. Francaviglia, *Gen. Relativ. Gravit.* [**40**]{}, 357 (2008), \[[arXiv:0706.1146]{}\].
I. Navarro and K. van Acoleyen, *Phys. Lett. B* [**622**]{}, 1 (2005), \[[gr-qc/0506096]{}\].
I. Navarro and K. van Acoleyen, *J. Cosmol. Astropart. Phys.* [**2006**]{}, 008 (2006), \[[gr-qc/0511045]{}\].
I. Navarro and K. van Acoleyen, *J. Cosmol. Astropart. Phys.* [**2006**]{}, 006 (2006), \[[gr-qc/0512109]{}\].
D.E. Krause and E. Fischbach, Searching for extra-dimensions and new string-inspired forces in the Casimir regime, in C. Lämmerzahl, C.W.F. Everitt and F.W. Hehl (ed.) *Gyros, Clocks, Interferometers. . . : Testing Relativistic Gravity in Space* Springer, Berlin, 2001, pp. 292-309, \[[hep-ph/9912276]{}\].
E.G. Adelberger, B.R. Heckel and A.E. Nelson, *Ann. Rev. Nucl. Part. Sci.* [**53**]{}, 77 (2003), \[[hep-ph/0307284]{}\].
M. White and C.S. Kochanek, *Astrophys. J.* [**560**]{}, 539 (2001), \[[astro-ph/0105227]{}\].
L. Amendola and C. Quercellini, *Phys. Rev. Lett.* [**92**]{}, 181102 (2004), \[[astro-ph/0403019]{}\].
C. Sealfon, L. Verde and R. Jimenez, *Phys. Rev. D* [**71**]{}, 083004 (2005), \[[astro-ph/0404111]{}\].
S. Reynaud and M.-T. Jaekel, *Int. J. Mod. Phys. A* [**20**]{}, 2294 (2005), \[[gr-qc/0501038]{}\].
A. Shirata, T. Shiromizu, N. Yoshida and Y. Suto, *Phys. Rev. D* [**71**]{}, 064030 (2005), \[[astro-ph/0501366]{}\].
M. Sereno and J.A. Peacock, *Mon. Not. Roy. Astron. Soc.* [**371**]{}, 719 (2006), \[[astro-ph/0605498]{}\].
Z. Stuchlík and S. Hledík, *Phys. Rev. D* [**60**]{}, 044006 (1999).
W. Rindler, *Relativity: Special, General, and Cosmological* Oxford University Press, Oxford, 2001.
M.-T. Jaekel and S. Reynaud, *Class. Quantum Grav.* [**23**]{}, 7561 (2006), \[[gr-qc/0610155]{}\].
J.R. Brownstein and J.W. Moffat, *Class. Quantum Grav.* [**23**]{}, 3427 (2006), \[[gr-qc/0511026]{}\].
M.-T. Jaekel and S. Reynaud, *Mod. Phys. Lett. A* [**20**]{}, 1047 (2005), \[[gr-qc/0410148]{}\].
M.-T. Jaekel and S. Reynaud, *Class. Quantum Grav.* [**22**]{}, 2135 (2005), \[[gr-qc/0502007]{}\].
M.-T. Jaekel and S. Reynaud, *Class. Quantum Grav.* [**23**]{}, 777 (2006), \[[gr-qc/0510068]{}\].
E.M. Standish, Planetary and Lunar Ephemerides: testing alternate gravitational theories, in A. Macias, C. Lämmerzahl and A. Camacho (ed.) *AIP Conference Proceedings Volume 977. RECENT DEVELOPMENTS IN GRAVITATION AND COSMOLOGY: 3rd Mexican Meeting on Mathematical and Experimental Physics*, American Institute of Physics, 2008, pp. 254-263.
T. Jacobson and D. Mattingly, *Phys. Rev. D* [**64**]{}, 024028 (2001), \[[gr-qc/0007031]{}\].
J.D. Bekenstein, *Phys. Rev. D* [**70**]{}, 083509 (2004), \[[astro-ph/0403694]{}\].
R. Maartens, *Living Rev. Rel.* [**7**]{}, 7 (2004), \[[gr-qc/0312059]{}\].
M. Burgay, N. D’Amico, A. Possenti, R.N. Manchester, A.G. Lyne, B.C. Joshi, M.A. McLaughlin, M. Kramer, J.M. Sarkissian, F. Camilo, V. Kalogera, C. Kim and D.R. Lorimer, *Nature* [**426**]{}, 531 (2003), \[[astro-ph/0312071]{}\].
M. Kramer, I.H. Stairs, R.N. Manchester, M.A. McLaughlin, A.G. Lyne, R.D. Ferdman, M. Burgay, D.R. Lorimer, A. Possenti, N. D’Amico, J.M. Sarkissian, G.B. Hobbs, J.E. Reynolds, P.C.C. Freire and F. Camilo, *Science* [**314**]{}, 97 (2006), \[[astro-ph/0609417]{}\].
E.V. Pitjeva, *Astron. Lett.* [**31**]{}, (2005).
E.V. Pitjeva, Use of optical and radio astrometric observations of planets, satellites and spacecraft for ephemeris astronomy, in W.J. Jin, I. Platais and M.A.C. Perryman (ed.) *A Giant Step: from Milli- to Micro-arcsecond Astrometry Proceedings IAU Symposium No. 248, 2007*, Cambridge University Press, Cambridge, 2008, pp. 20-22.
E.V. Pitjeva, *Sol. Sys. Res.* [ **39**]{}, 176 (2005).
A. Lue and G. Starkman, *Phys. Rev. D* [**67**]{}, 064002 (2003), \[[astro-ph/0212083]{}\].
G. Dvali, A. Gruzinov and M. Zaldarriaga, *Phys. Rev. D* [**68**]{}, 024012 (2003), \[[hep-ph/0212069]{}\].
L. Iorio, *Class. Quantum Grav.* [**22**]{}, 5271 (2005), \[[gr-qc/0504053]{}\].
L. Iorio, *Adv. High En. Phys.* [**2007**]{}, 90731 (2007), \[[arXiv:0710.0022]{}\].
L. Iorio, *New Astron.* [**14**]{}, 196 (2009), \[[arXiv:0808.0256]{}\].
J.W. Moffat and V.T. Toth, *Astrophys. J.* [**680**]{}, 1158 (2008), \[[arXiv:0708.1935]{}\].
J.W. Moffat and V.T. Toth, *Class. Quantum Grav.* [**26**]{}, 085002 (2009), \[[arXiv:0712.1796]{}\].
L. Iorio, *Schol. Res. Exchange* [**2008**]{}, 238385 (2008), \[[arXiv:0809.3563]{}\].
J. Bekenstein and M. Milgrom, *Astrophys. J.* [**286**]{}, 7 (1984).
B. Famaey and J. Binney, *Mon. Not. Roy. Astron. Soc.* [**363**]{}, 603 (2005), \[[astro-ph/0506723]{}\].
C. Talmadge, J.-P. Berthias, R.W. Hellings and E.M. Standish, *Phys. Rev. Lett.* [**61**]{}, 1159 (1988).
M. Sereno and Ph. Jetzer, *Mon. Not. Roy. Astron. Soc.* [**371**]{}, 626 (2006), \[[astro-ph/0606197]{}\].
L. Iorio, *J. Gravit. Phys.* [**2**]{}, 26 (2008), \[[arXiv:0711.2791]{}\].
S. Capozziello, V.F. Cardone and A. Troisi, *Mon. Not. Roy. Astron. Soc.* [**375**]{}, 1423 (2007), \[[astro-ph/0603522]{}\].
L. Iorio and M.L. Ruggiero, *Schol. Res. Exchange* [**2008**]{}, 968393 (2008), \[[arXiv:0711.0256]{}\].
L. Iorio, *J. High En. Phys.* [**2007**]{}, 041 (2007), \[[arXiv:0708.1080]{}\].
A.W. Kerr, J.C. Hauck and B. Mashhoon, *Class. Quantum Grav.* [**20**]{}, 2727 (2003), \[[gr-qc/0301057]{}\].
L. Iorio, *Adv. Astron.* [**2008**]{}, 268647 (2008), \[[arXiv:0710.2610]{}\].
L. Iorio and G. Giudice, *New Astron.* [**11**]{}, 600 (2006), \[[gr-qc/0601055]{}\].
L. Iorio, THE LENSE-THIRRING EFFECT AND THE PIONEER ANOMALY: SOLAR SYSTEM TESTS, in H. Kleinert, R.T. Jantzen and R.J. Ruffini (ed.) *THE ELEVENTH MARCEL GROSSMANN MEETING On Recent Developments in Theoretical and Experimental General Relativity, Gravitation and Relativistic Field Theories. Proceedings of the MG11 Meeting on General Relativity Berlin, Germany 23-29 July 2006*, Singapore, World Scientific, 2008, pp. 2558-2560, \[[gr-qc/0608105]{}\].
L. Iorio, *J. Gravit. Phys.* [**1**]{}, 1 (2007), \[[arXiv:0712.1273]{}\].
L. Iorio, *Found. Phys.* [**37**]{}, 897 (2007), \[[gr-qc/0610050]{}\].
L. Iorio, *Int. J. Mod. Phys. D* at press, \[[arXiv:0806.3011]{}\].
G. Page, D.S. Dixon, J.F. Wallin, *Astrophys. J.* [**642**]{}, 606 (2006), \[[astro-ph/0504367]{}\].
K. Tangen, *Phys. Rev. D* [**76**]{}, 042005 (2007), \[[gr-qc/0602089]{}\].
J.F. Wallin, D.S. Dixon and G.L. Page, *Astrophys. J.* [**666**]{}, 1296 (2007), \[[arXiv:0705.3408]{}\].
G. Page, J.F. Wallin, D.S. Dixon, *Astrophys. J.* [**697** ]{}, 1226 (2009), \[[arXiv:0905.0030]{}\].
J. Lense and H. Thirring, *Phys. Z.* [**19**]{}, 156 (1918).
L. Iorio, *Proc. of Science* [**ISFTG**]{}, 017 (2009), \[[arXiv:0905.0300]{}\].
[^1]: It was shown that in such ways the non-renormalizability of general relativity became more tractable in the context of quantum field theory.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Using gamma-ray data from observations of the Milky Way, Andromeda (M31), and the cosmic background, we calculate conservative upper limits on the dark matter self-annihilation cross section to a number of final states, over a wide range of dark matter masses. We first constrain annihilation to a pair of monoenergetic gamma rays, $\langle \sigma_A v \rangle_{\gamma \gamma}$, and show that in general our results are unchanged for a broader annihilation spectrum, if at least a few gamma rays are produced with energies within a factor of a few from the dark matter mass. We then place constraints on the self-annihilation cross section to an electron-positron pair, $\langle\sigma_A v \rangle_{e^+e^-}$, using gamma rays produced via internal bremsstrahlung radiative corrections. We also place constraints on annihilation into the other charged leptons. We make conservative assumptions about the astrophysical inputs, and demonstrate how our derived bounds would be strengthened if stronger assumptions about these inputs are adopted.'
author:
- 'Thomas D. Jacques'
date: 'August 14, 2009'
title: Robust Constraints on Dark Matter Annihilation into Gamma Rays and Charged Particles
---
[address=[School of Physics, The University of Melbourne, Victoria 3010, Australia]{}]{}
Introduction
============
For thermal relic Dark Matter (DM), the self-annihilation cross section must be $\langle \sigma_A v \rangle \sim 3 \times 10^{-26}$ cm$^3$ s$^{-1}$ to obtain the observed relic abundance. There are a number of ways in which the self-annihilation rate can be enhanced e.g., [@Hisano:2004ds; @Feng:2008ya; @Cirelli:2008pk; @ArkaniHamed:2008qn; @Pospelov:2008jd; @Nelson:2008hj; @Cholis:2008qq; @Bai:2008jt; @Fox:2008kb], while non-thermal relic DM may allow larger values for $\langle \sigma_A v \rangle$ [@Das:2006ht; @KKT]. Self-annihilation can lead to indirect detection of DM through observation of the annihilation products. We constrain the self-annihilation cross section by considering the gamma rays produced via DM annihilation to a number of final states.
We first focus on annihilation to a pair of monenergetic gamma rays. This annihilation channel has a very clean signature, with $E_\gamma=m_\chi$, and future observation of such a line could provide clear evidence for the DM mass. We compare the calculated signal with gamma-ray observations of the Galactic Center, Andromeda (M31), and the cosmic background, and place upper limits on the partial self-annihilation cross section, $\langle \sigma_A v \rangle_{\gamma \gamma}$. Our limits cover a broad DM mass range, from 20 keV up to 10 TeV. We show our constraints are applicable even if DM annihilates to a broader spectrum of gamma rays, in general constraining the total annihilation cross section. With a conservative choice for the branching ratio, $Br(\gamma\gamma)=10^{-4}$, we constrain the *total* annihilation cross section, $\langle \sigma_A v \rangle_{\rm total}=\langle \sigma_A v \rangle_{\gamma \gamma}/Br(\gamma\gamma)$.
Next we place constraints on the self-annihilation cross section to an electron-positron pair, $\langle\sigma_A v \rangle_{e^+e^-}$. These particles will inevitably be accompanied by gamma rays through a process known as *internal bremsstrahlung* (IB) or *final state radiation*. This is the emission of a gamma-ray from one of the final state charged particles, occurring at the Feynman diagram level. See Refs. [@IB; @BBB; @Birkedal:2005ep; @BergstromRC] for a detailed discussion. We use these gamma rays to constrain $\langle\sigma_A v \rangle_{e^+e^-}$ by again comparing the calculated annihilation signal with the observed gamma-ray flux. We place corresponding constraints on annihilation to the other charged leptons. The IB gamma-ray spectrum per annihilation does not depend on the tree level cross section, so our analysis remains DM model independent. Because emission occurs at the Feynman diagram level, the gamma-ray flux from IB remains independent of uncertain astrophysical phenomenon such as the interstellar magnetic field or radiation field.
The $e^+e^-$ final state is especially interesting in light of the recent PAMELA/ATIC/ Fermi results suggesting an excess in the observed positron flux. There are now numerous DM models that can explain these results, many with a large branching ratio to charged leptons, e.g. [@Cirelli:2008pk; @ArkaniHamed:2008qn; @Pospelov:2008jd; @Nelson:2008hj; @Cholis:2008qq; @Bai:2008jt; @Fox:2008kb]. Our limits on $\langle \sigma_A v \rangle_{e^+e^-}$ directly constrain many models attempting to explain the positron excess.
We show how our results depend on the astrophysical uncertainties in the DM density profile. We are conservative in our analysis methods and input choices, and show how our results would be strengthened by more optimistic choices. We compare our constraints with pre-existing bounds on the total annihilation cross section. We find that our constraints on $\langle\sigma_A v \rangle_{\rm total}$ are particularly strong for small DM masses, while our constraints on $\langle\sigma_A v \rangle_{\gamma\gamma}$ are stronger than those on $\langle\sigma_A v \rangle_{e^+e^-}$ by a factor of $\lesssim 10^2$, as expected from the $\alpha\simeq 1/137$ suppression of the radiative corrections.
Annihilation to $\gamma\gamma$
==============================
Our analysis and results are described in detail in Mack *et al.* [@MJBBY]. The flux of gamma rays from DM annihilation in our halo is given by $$\frac{d\Phi_{\gamma}}{dE} =
\frac{\langle \sigma_A v\rangle}{2}
\frac{{\cal J}_{\Delta \Omega}}{{\rm J_0}}
\frac{ 1 }{4\pi m_\chi^2}
\frac{dN_{\gamma}}{dE}\,,
\label{haloflux}$$ where $dN_\gamma/dE$ is the gamma-ray spectrum per annihilation, $\Delta\Omega$ is the field of view in steradians, and
$${\cal J}_{\Delta \Omega} = \frac{2\pi }{\Delta \Omega} {\rm J_0}\int_0^{\psi}
\int^{\ell_{max}}_0 \rho^2\left(\sqrt{R_{\textrm{sc}}^2 -
2\ell R_{\textrm{sc}}\cos{\psi} +\ell^2}\right)d\ell \sin{\psi }d\psi\,,$$
is proportional to the dark matter density squared, integrated over the line of sight. For annihilation to a gamma-ray pair, the spectrum per annihilation is simply $dN_{\gamma}/dE = 2\delta(m_\chi - E)$.
We collect gamma-ray flux data from a variety of sources, covering a large range of energies. We compare the calculated annihilation flux with the observed flux, integrating each over an energy bin of equal size, and find an upper limit on $\langle \sigma_A v\rangle$ by demanding that the annihilation flux be smaller than the entire observed flux. This is extremely conservative, as in reality, the DM annihilation flux would only be a small fraction of the observed flux. We focus on the Galactic Center flux, using data from COMPTEL and EGRET aboard the CGRO [@COMPTELweb; @EGRETweb], INTEGRAL [@INTweb], and H.E.S.S. [@HESSweb]. We also look at gamma rays arising from annihilation in the M31 (Andromeda) galaxy, and from cosmic annihilation, using the methods described in Mack *et al.* [@MJBBY]. We use M31 data from EGRET, CELESTE [@CELESTEweb] and HEGRA [@HEGRAweb], and diffuse cosmic flux data from INTEGRAL, COMPTEL, EGRET and the Solar Maximum Mission [@SMMweb].
We place constraints on $\langle \sigma_A v\rangle_{\gamma\gamma}=\langle \sigma_A v\rangle_{\rm total} Br(\gamma\gamma)$, for both the conservative Kravtsov [@Kravtsov] and the commonly-adopted NFW DM density profile [@NFW], over an energy range spanning 20 keV to 10 TeV. These results are shown in Fig. \[fig:sigmagg-fig2\].
![ The limits on the partial cross section, $\langle \sigma_A v
\rangle_{\gamma\gamma}$, derived from the various gamma-ray data. Our overall limit is shown as the dark shaded exclusion region. For comparison, the light-shaded region shows the corresponding limits for the NFW (rather than the Kravtsov) profile.[]{data-label="fig:sigmagg-fig2"}](sigmagg-fig2){width="3.25in"}
To find a constraint on the total cross section, we need the branching ratio to this final state. Typical branching ratios are $\sim 10^{-3}$. In Fig. \[fig:sigmagg-fig3\] we show our upper bound on $\langle \sigma_A v\rangle_{\rm total}$ using a conservative choice of $Br(\gamma\gamma)=10^{-4}$. In this plot we use the conservative Kravtsov density profile. For comparison, we show other limits on $\langle \sigma_A v\rangle$, including a bound based on observations of the Neutrino flux, from Yüksel *et al.* [@YHBA], and the standard cross section for thermal relic DM. The KKT and Unitarity bounds are described in Ref. [@MJBBY].
![ The gamma-ray and neutrino limits on the total annihilation cross section, selecting $Br(\gamma\gamma) = 10^{-4}$ as a conservative value. The unitarity and KKT bounds are also shown. The overall bound on the total cross section at a given mass is determined by the strongest of the various upper limits.[]{data-label="fig:sigmagg-fig3"}](sigmagg-fig3){width="3.25in"}
Most of these experiments had modest energy resolution. To be conservative, we assume an analysis bin with a logarithmic energy width of 0.4 in $\log_{10} E$ (i.e., $\Delta(\ln E) \sim 1$) for the Galactic and cosmic diffuse analyses; this is at least as wide as the energy bins reported by the experiments. In effect, our results are what one would obtain for an annihilation gamma-ray spectrum as wide as 0.4 in $\log_{10} E$. The exception is the INTEGRAL line search.
Annihilation to $e^+e^-$
========================
For annihilation to an $e^+e^-$ pair, we follow a similar method to that outlined above. We constrain $\langle \sigma_A v \rangle_{e^+e^-}$ by demanding that the calculated gamma-ray signal from IB emission not exceed the entire observed gamma-ray flux. Our analysis and results are described in detail in Bell and Jacques [@IB]. The IB gamma-ray spectrum per annihilation is $$\frac{dN_\gamma}{dE} =
\frac{1}{\sigma_{\rm tot}}\frac{d\sigma_{\rm IB}}{dE_\gamma}.
\label{Ngamma}$$ where $$\frac{d\sigma_{\rm IB}}{dE_\gamma} = \sigma_{\rm tot} \times
\frac{\alpha}{E_\gamma \pi}
\bigg[ \ln\bigg(\frac{4m_\chi(m_\chi-E_\gamma)}{m_e^2} \bigg) -1 \bigg]\bigg[1+\bigg( \frac{4m_\chi(m_\chi-E_\gamma)}{4m_\chi^2}\bigg)^2\bigg],
\label{sigbrem}$$ and $\sigma_{\rm tot}$ is the tree-level cross section for annihilation to $e^+e^-$. Note that $\sigma_{\rm tot}$ factors out from the IB cross-section. This important feature implies that the IB spectrum is independent of the unknown physics which mediates the lowest order annihilation process. Fig. \[spectra\] shows $E^2 dN_\gamma/dE$ for IB emission for a number of DM masses.
![ Internal bremsstrahlung gamma-ray spectra per $\chi\chi\rightarrow e^+e^-$ annihilation, for $m_\chi=$ 100 GeV, 200 GeV, 500 GeV, 1000 GeV.[]{data-label="spectra"}](spectra.eps){width="3.0in"}
We use Galactic gamma-ray data from COMPTEL, EGRET and H.E.S.S. We also use data from CELESTE observations of M31 to cover the gap between the highest energy EGRET data and the lowest H.E.S.S. data points. The strongest bounds are obtained by choosing an analysis bin where the ratio of the calculated annihilation signal to the observed background is largest. Fig. \[spectra\] shows that $dN_\gamma/dE$ falls with energy. However, the observed gamma-ray flux falls with energy at a steeper rate, so that the strongest results are obtained by choosing an analysis bin with the DM mass as the upper energy limit.
![Upper limit on $\langle \sigma v \rangle_{e^+e^-}$ as a function of DM mass for the Kravtsov (solid), NFW (dashed) and Moore (dot-dashed) profiles.[]{data-label="results1"}](results-1.eps){width="3.5in"}
In Fig. \[results1\] we show the upper limits on $\langle \sigma_A v
\rangle_{e^+e^-}$ as a function of DM mass, using the observational data described above. We give the Galactic Center results for the Kravtsov and NFW profiles mentioned earlier, as well as the steep Moore profile [@Moore]. We show the CELESTE constraint using only the Kravtsov profile.
In Fig. \[results2\] we show the upper bounds on the annihilation cross sections into $e^+e^-$, $\mu^+\mu^-$, and $\tau^+\tau^-$, based upon the IB emission from each final state (all use the conservative Kravtsov profile). For annihilation to $\tau^+\tau^-$, we do not take into account gamma rays arising from the hadronic decay modes. Including this would strengthen our bounds. For comparison, we also show limits on $\langle \sigma_A v \rangle_{\gamma\gamma}$ discussed in the previous section, as well as some earlier bounds on $\langle \sigma_A v\rangle$. For full details see Ref. [@IB].
![Upper limits on $Br(ii)\times\langle \sigma v
\rangle_{total}$ for various final states $ii=e^+e^-$ (solid black; labelled), $\mu^+\mu^-$ (thick dashed; labelled), $\tau^+\tau^-$ (thick dashed; labelled), $\gamma\gamma$ (solid; labelled), and $\bar\nu\nu$ (solid; labelled), using the conservative Kravtsov profile. Also shown are the KKT (thin dashed) and unitarity (thin dot-dashed) limits on the total cross section described in Refs. [@MJBBY; @IB], and the cross section for thermal relic DM (natural scale). The $\bar\nu \nu$ limits are taken from Ref. [@YHBA].[]{data-label="results2"}](results-2.eps){width="3.5in"}
Discussion
==========
While the Kravtsov, NFW and Moore profiles diverge towards the center of the Galaxy, they are similar at large radii. As the INTEGRAL, EGRET and COMPTEL observations encompass relatively large angular scales, the density profile changes have a modest effect. On the other hand, the H.E.S.S. constraints correspond to a much smaller angular region toward the Galactic Center, and vary by orders of magnitude depending on the profile adopted. (See Ref. [@YHBA] for a full discussion of the differences between the profiles for different angular regions.) If DM annihilates to $e^\pm$, photons will be produced not only by IB, but also by energy loss processes including inverse Compton scattering and synchrotron radiation. However, the intensity of these signals depend on uncertain astrophysical parameters, such as magnetic field strength, radiation field intensities, and electron diffusion scales. By contrast, IB is free of these astrophysical uncertainties, and has a fixed spectrum and normalization. Another key difference is the energy of the photons. Synchrotron radiation produces generally low energy photons, while IB provides some hard gamma rays near the endpoint. Since the background flux falls off with energy, these hard gamma rays are extremely useful. The sharp edge in the IB spectrum at $E=m_\chi$ can be used to diagnose the DM mass; this is not possible with synchrotron radiation.
Refs. [@Cirelli:2008pk; @ArkaniHamed:2008qn; @Pospelov:2008jd; @Nelson:2008hj; @Cholis:2008qq; @Bai:2008jt; @Fox:2008kb] and others have recently proposed models in which DM annihilates directly to charged leptons, with cross sections well above that expected for a thermal relic. This may account for anomalies in cosmic ray spectra from PAMELA, ATIC and Fermi, and microwave signals from WMAP, all of which seem to require more electrons and positrons than can be explained otherwise. Our bounds on $\langle \sigma_A v
\rangle_{l^+l^-}$ will directly constrain the allowed parameter space for these types of DM models.
[**Acknowledgments:**]{} This article is based on Ref. [@MJBBY], done in collaboration with Greg Mack, John Beacom, Nicole Bell, Hasan Yuksel, and Ref. [@IB], done in collaboration with Nicole Bell. The author was supported by the Commonwealth of Australia.
[99]{}
J. Hisano, S. Matsumoto, M. M. Nojiri and O. Saito, Phys. Rev. D [**71**]{}, 063528 (2005) \[arXiv:hep-ph/0412403\]. J. L. Feng and J. Kumar, arXiv:0803.4196 \[hep-ph\]. M. Cirelli, M. Kadastik, M. Raidal and A. Strumia, cosmic ray arXiv:0809.2409 \[hep-ph\]. N. Arkani-Hamed, D. P. Finkbeiner, T. Slatyer and N. Weiner, arXiv:0810.0713 \[hep-ph\]. See also N. Arkani-Hamed and N. Weiner, arXiv:0810.0714 \[hep-ph\]. M. Pospelov and A. Ritz, arXiv:0810.1502 \[hep-ph\]. A. E. Nelson and C. Spitzer, arXiv:0810.5167 \[hep-ph\]. I. Cholis, D. P. Finkbeiner, L. Goodenough and N. Weiner, arXiv:0810.5344 \[astro-ph\]. Y. Bai and Z. Han, arXiv:0811.0387 \[hep-ph\]. P. J. Fox and E. Poppitz, arXiv:0811.0399 \[hep-ph\]. S. Das and N. Weiner, \[arXiv:astro-ph/0611353\]. M. Kaplinghat, L. Knox and M. S. Turner, Phys. Rev. Lett. [**85**]{}, 3335 (2000) \[arXiv:astro-ph/0005210\]. N. F. Bell and T. D. Jacques, Phys. Rev. D [**79**]{}, 043507 (2009) arXiv:0811.0821 \[astro-ph\]. J. F. Beacom, N. F. Bell and G. Bertone, Phys. Rev. Lett. [**94**]{}, 171301 (2005) \[arXiv:astro-ph/0409403\]. A. Birkedal, K. T. Matchev, M. Perelstein and A. Spray, arXiv:hep-ph/0507194. L. Bergstrom, T. Bringmann, M. Eriksson and M. Gustafsson, Phys. Rev. Lett. [**94**]{}, 131301 (2005). \[arXiv:astro-ph/0410359\]. G. D. Mack, T. D. Jacques, J. F. Beacom, N. F. Bell and H. Yuksel, Phys. Rev. D [**78**]{}, 063542 (2008) \[arXiv:0803.0157 \[astro-ph\]\]. H. Yuksel, S. Horiuchi, J. F. Beacom and S. Ando, Phys. Rev. D [**76**]{}, 123506 (2007) arXiv:0707.0196 \[astro-ph\]. A. V. Kravtsov, A. A. Klypin, J. S. Bullock and J. R. Primack, Astrophys. J. [**502**]{}, 48 (1998) \[arXiv:astro-ph/9708176\]. J. F. Navarro, C. S. Frenk and S. D. M. White, Astrophys. J. [**462**]{}, 563 (1996) \[arXiv:astro-ph/9508025\]. B. Moore, T. R. Quinn, F. Governato, J. Stadel and G. Lake, Mon. Not. Roy. Astron. Soc. [**310**]{}, 1147 (1999) \[arXiv:astro-ph/9903164\].
http://www.sciops.esa.int/integral
http://cossc.gsfc.nasa.gov/docs/cgro/cgro/comptel.html
http://cossc.gsfc.nasa.gov/docs/cgro/cgro/egret.html
http://www.mpi-hd.mpg.de/hfm/HESS/HESS.html
http://doc.in2p3.fr/themis/CELESTE/
http://wwwiexp.desy.de/groups/astroparticle/de/archive/hegra/
http://heasarc.nasa.gov/docs/heasarc/missions/solarmax.html
|
{
"pile_set_name": "ArXiv"
}
|
Introduction
============
Colorectal cancer (CRC), also known as bowel cancer or colon cancer, is the development of cancer from the colon or rectum and accounts for about 10% of all cancer cases worldwide [@Kuip2015]. Precise diagnosis and prognosis are very important in the choice of the most appropriate treatment and the facilitation of the subsequent clinical management of patients. Consequently, an increasing amount of current research is concerned with improving nuanced disease understanding, and the precision and accuracy of survivability estimation.
The Tumour-Node-Metastasis (TNM) staging system is still widely regarded as one of the best population level predictors of CRC outcome [@fleming2012colorectal]. However, its patient level precision is low [@pages2018international]. In part driven by this observation and the desire to effect improvement, the past five years have witnessed a major growth of the application of various machine learning techniques in patient level cancer prognosis [@chen2014risk; @TunAranCaie2018; @NeoAranHarrCaie2018].
The aim of the present work is to develop an end-to-end framework that predicts the disease specific survivability of CRC patients over a five year period by applying deep learning to unannotated whole slide histopathology images (WSIs). The key challenges addressed are thus as follows:
- Computational impracticality of training classifiers with gigapixel WSIs,\
- Efficient ground-truth labelling of individual image patches,\
- Determination of discriminative patch subsets,\
- Convolutional neural network (CNN) design for patch level outcome prediction, and\
- Fusion of patch level into image level predictions.
Relevant background
===================
In 2018 in the United States, the estimated number of new cases of CRC was 97,220 with 50,630 disease associated deaths [@ACS2018]. The overall death rate of CRC has decreased from 28 per 100,000 (1975) to 14 per 100,000 (2015) as the result of increased screening, decline in incidence, and improvement of treatment [@ACS2018]. Screening can effectively prevent CRC by the identification and subsequent removal of early-stage precancerous growth, with a range of factors employed to identify patients at risk. Some of the notable risk factors include colorectal adenomas [@johns2001systematic], hereditary conditions including Lynch syndrome and adenomatous polyposis [@mork2015high], personal history of long-standing chronic ulcerative colitis [@laukoetter2011intestinal], and alcohol use [@fedirko2011alcohol]. Although colorectal cancer screening guidelines do not distinguish between female and male, the statistics show that the number of new cases of males is 17% greater than that of female [@liang2009cigarette] while female over 65 years old presents higher mortality rate and lower 5-year survival rate of CRC compared to their age-matched male counterparts [@kim2015sex].
However, CRC is found to be highly treatable and often curable when it is confined to the bowel after surgical intervention [@miller2016cancer]. Nevertheless, CRC is generally considered not curable when the cancer cells have spread to other organs, termed metastasis. In this case, appropriate choices of health management, including chemotherapy or targeted therapy, can still help improve the quality and length of life [@miller2016cancer]. Therefore, the early diagnosis and accurate prognostic prediction of the cancer aggressiveness and patient outcome are significant.
CRC staging and prognosis
=========================
In present clinical practice, the main prognostic factors for CRC comprise: (T) depth of tumour penetration through bowel wall, (N) presence or absence of nodal involvement, and (M) presence or absence of distant metastases. These form the basis of the five stage TNM staging system [@gospodarowicz2017tnm]. Stage 0 is least severe, with all the lesions restricted to the mucosa and the lamina propria. Local excision or simple polypectomy with clear margins is the most common treatment option. In Stage I, cancer may have grown into the muscularis mucosa or into the muscularis propria but has not spread deeper into the colon muscle wall, to nearby lymph nodes or other distant sites. Because CRC at this stage is still localized, it also has a high cure rate with wide surgical resection and anastomosis. Stage II characterizes CRC that has spread to to or beyond the serosa and may have grown into nearby tissue or organs, but not to the lymph nodes and has not metastasised. Surgical resection is again the standard treatment, however high-risk patients, such as those with t4 disease may be offered chemotherapy. Stage III is characterized by lymph node involvement and the standard treatments are wide surgical resection and anastomosis, and adjuvant chemotherapy. Stage IV disease is characterized bymetastatic disease. The treatment of CRC at this stage largely depends on the sites of metastatic disease. Liver metastasis makes up approximately 50% of Stage IV and recurrent CRC, and the options for treatment include all the aforementioned ones as well as palliative radiation therapy, palliative chemotherapy, and targeted therapy.
H&E staining
------------
In this work, we utilize digitized whole slide images of archived diagnostic histopathological tissue sections stained with haematoxylin and eosin (H&E); see examples in Fig \[f:normEx\]. Eosin is an acidic dye that stains the basic structures red or pink, such as the proteins within the cytoplasm. Haematoxylin, on the other hand, is a basic dye which stains the acidic structure blue or purple, such as DNA in the nucleus.
![Chromatic normalization examples (left & right: original & normalized tiled strips).[]{data-label="f:normEx"}](norm_chrom_02.png){width=".45\textwidth"}
Within an H&E stained colon section, we can observe nuclei of cells in purplish blue, cytoplasm in red, erythrocytes in cherry red, collagen and mitochondria in pale pink. However, the colour intensity of the stain depends on both the amount of stain applied and the duration of exposure.
Previous work
-------------
The manual reporting of H&E stained tissue sections under the microscope, or WSIs on the computer monitor, for TNM staging is laborious. It is also largely based on the subjective experience-based assessment of the pathologist, which often causes variation across different observers. Motivated by these limitations, there is an increasing amount of research on the use of machine learning for the analysis of WSIs.
As a data-driven and end-to-end approach that learns high-level feature without subjective biases, application of convolutional neural networks (CNNs), computer-aided interpretation tools, can be traced to 1990s when a convolutional artificial neural network which attempts to mimic a radiologist’s analytic process of radiographic images was introduced [@lo1995artificial]. Thereafter, CNN architectures for carotid intima-media thickness measurement in ultrasound images [@shin2016automating], brain tumour segmentation in magnetic resonance imaging scans [@havaei2017brain], neuronal membrane segmentation in electron microscopy images [@ciresan2012deep], and many others have been proposed. For WSI histological analysis, CNNs have been used for challenging problems such as automated nuclear atypia scoring [@lu2017deep] and the discrimination between epithelial and stromal tissues [@bianconi2015discrimination], with promising results.
More recently, a series of works suggested that fine-tuned pre-trained CNNs outperformed CNNs that are trained from scratch while taking much less time as well [@gao2018holistic]. Conventionally, when training from scratch, all the parameters in the architecture of an artificial neural network are randomly initialized. By contrast, when fine-tuning a CNN, the weight and bias values are initialized with the parameters of a pre-trained CNN with the same architecture. It is argued that the early layers of CNN learn the low-level image features which are roughly the same for different vision problems, however, the deeper layers that learn the high-level features are specific with respect to the classification task itself [@tajbakhsh2016convolutional]. Four specific image analysis problems, including colonic polyp or pulmonary embolism detection, colonoscopy frame classification and intima-media boundary segmentation were considered by Tajbakhsh et al. [@tajbakhsh2016convolutional] to demonstrate the potential of knowledge transfer between ‘natural’ and medical images.
### Patch-based CNNs
Although CNNs are widely considered as the state-of-the-art models in various applications of image classification, the analysis of WSIs remains challenging because training a deep CNN model with gigapixel WSIs is still computationally impractical. Hence, most of the aforementioned techniques work with severely down-sampled images. However, this approach inherently effects a loss of discriminative information at finer scales.
Hou et al. [@hou2016patch] train a model on patches of high-resolution images and from these make predictions for entire WSIs, with an expectation maximization based algorithm used to automatically determine discriminative patches. Thereafter, a number of other methods which rely on the use of image patches and CNNs have demonstrated promising results in distinguishing WSIs of tumorous and normal tissues [@jamaluddin2017tumor], as well as the segmentation of precursor lesions [@albayrak2017segmentation]. However, prognosis is an inherently more difficult learning challenge [@mackillop2003importance].
Recently an adaptive sampling method was applied in an end-to-end framework [@zhu2017wsisa] to cluster group of images with different local content. This framework comprises four key stages: (i) adaptive generation of patches from WSIs, (ii) patch clustering according to the phenotypes, (iii) automatic clustering selection, and (iv) aggregation of cluster level predictions. However, due to the low-resolution of down-sampled phenotypes and the lack of an effective way to aggregate patch-wise predictions, the framework only achieves 57% average accuracy. The framework we propose in the present paper overcomes the key limitations of the aforementioned work in part by introducing a novel way of generating phenotypes.
Proposed methodology
====================
The main objective of this work is to predict the survivability of stage I and II CRC patients from whole-slide H&E stained histopathology images. More specifically, the key aim is to predict whether the patient is likely to survive at least five years (the currently used clinically driven salient follow-up time) after surgery.
Data preparation & pre-processing
---------------------------------
The analysis of WSIs is widely conceived as one of the most challenging tasks in the field of medical image analysis due to the following factors: (i) individual image size, (ii) usually extreme class imbalance, (iii) low total slide count (contextually speaking), (iv) scanning and preparation image artefacts, and (v) WSI salient information heterogeneity. We will shortly expand in greater detail but in summary, in order to address these we propose a sequence of steps which include chromatic normalization, patch extraction, data augmentation, and patch clustering.
#### Chromatic normalization
Histopathological tissue sections or WSI are often examined individually by pathologists, who mainly focus on relative colour and pattern differences within a single tissue section. It is rare to compare directly different slides in order to make a diagnosis; each slide is examined to identify particular spatial or pattern characteristics. However, in the application of quantitative analysis and medical statistics for diagnosis and prognosis, different overall absolute colour value can lead to serious bias especially when the slide count is low. The variation in terms of colour distribution is ultimately the difference in the amount of light absorbed; Fig \[f:normEx\] shows examples of slides from different surgeries. As can be seen, the colour profile exhibits great inter-clinic variability. While it is true that the use of greyscale would address this problem, it also effects a loss of valuable pathological information.
We apply the Reinhard normalization [@reinhard2001color] which begins by converting the RGB representation into a perception-based colour space $l\alpha\beta$ with known phosphor chromaticity. This is done by a conversion from RGB to XYZ tristimulus values and then a conversion from XYZ space to LMS which is then followed by principal component analysis (effecting axes rotation).
The first conversion is based on the phosphors of monitor that is used to display an image. A device independent conversion is applied as an approximation that maps the white in the chromaticity diagram to white in RGB space: [$$\begin{aligned}
\left[
\begin{array}{c}
X\\
Y\\
Z
\end{array}\right]=
\left[
\begin{array}{ccc}
0.5141 & 0.3239 & 0.1604\\
0.2651 & 0.6702 & 0.0641\\
0.0241 & 0.1228 & 0.8444
\end{array}\right]
\left[
\begin{array}{c}
R\\
G\\
B
\end{array}\right].\end{aligned}$$]{} Next, mapping to LMS space is performed: [$$\begin{aligned}
\left[
\begin{array}{c}
L\\
M\\
S
\end{array}\right]=
\left[
\begin{array}{ccc}
0.3897 & 0.6890 & −0.0787\\
−0.2298 & 1.1834 & 0.0464\\
0.0000 & 0.0000 & 1.0000
\end{array}\right]
\left[
\begin{array}{c}
X\\
Y\\
Z
\end{array}\right].\end{aligned}$$]{} To eliminate distribution skew in the LMS space a logarithmic transform is applied and a pre-calculated maximal de-correlation matrix used to rotate the axes: [$$\begin{aligned}
\left[
\begin{array}{c}
l\\
\alpha\\
\beta
\end{array}\right]=
\left[
\begin{array}{ccc}
\frac{1}{\sqrt{3}} & 0 & 0\\
0 & \frac{1}{\sqrt{6}} & 0\\
0 & 0 & \frac{1}{\sqrt{2}}
\end{array}\right]
\left[
\begin{array}{ccc}
1 & 1 & 1\\
1 & 1 & -2\\
1 & -1 & 0\\
\end{array}\right]
\left[
\begin{array}{c}
L\\
M\\
S
\end{array}\right].\end{aligned}$$]{} where the $l$ component is the achromatic value, the $\alpha$ and $\beta$ components represent chromatic yellow-blue and red-green channel values. Now the achromatic axis is orthogonal to the equiluminant plane. Finally, the distribution of values in each channel is normalized to be the standard normal.
#### Patch extraction
Patches are extracted after the entire WSI has been down-sampled and normalized. The gigapixel resolution of the WSIs makes the existing approaches in the literature prohibitively computationally demanding. To make the model trainable, herein tiles of size $224 \times 224$ pixels are extracted from the 1/10 resolution image (n.b. $40\times$ magnification level was used in acquisition).
Recall that the patch extraction process in the pioneer work of [@hou2016patch] is still human labour intensive as it requires patches with less than 30% of tissue area or excessive blood content to be labelled and discarded manually. To construct an end-to-end pipeline, we instead approach the problem of relevant patch selection through the use of fully automatic clustering which does not assume or require application of human prior knowledge.
#### Data augmentation
To prevent the model from over-fitting and alleviate problems caused by data imbalance, data augmentation is applied to the training data. Herein we perform augmentation by means of applications of a restricted set of affine transformations, Gaussian blur, and all-channel multiplication.
Specifically, we apply geometric transformations by means of 90$^\circ$ rotations, and vertical and horizontal reflections, in random combinations, effecting an 8-fold increase in the data set size. Gaussian blur with the unitary standard deviation is applied in order to make the model generalize to blurred images since blur is often found in regions of WSIs as a result of locally poor focusing [@lopez2013automated].
{width=".99\textwidth"}
Patch clustering
----------------
A major problem for patch-wise based classification approaches is that there is no ground truth label for each individual patch. In order to overcome this issue, we broadly consider a patch either to be (sufficiently) discriminative or not. This alone does not get one much further as it is very difficult to extract the discriminative subset of patches without expert knowledge and intensive human labour. Therefore, to obtain a collection of discriminative patches, we propose that an unsupervised learning method is used to cluster similar patches into several groups. In particular, we apply the $k$-means algorithm to group patches from a single WSI [@Aran2013c]. To increase the robustness of the result to the random initialization of parameters we perform multiple clusterings using different random starting parameters, and adopt the one associated with the lowest loss, thereby avoiding sub-optimal local minima. Lastly, CNN based classifiers are trained with patches from different clusters and used to determine which clusters are discriminative. In this work we adopt two clustering approaches, described next.
#### Information density clustering
Information density is a simple but efficient way to group the extracted patches. In particular, since peripheral patches tend to contain large uniform areas, they are suited for compression by the DEFLATE algorithm used by the PNG image format. The information ratio is defined here as the inverse of data compression ratio, $IR =\frac{1}{CR}=\frac{S_c}{S_u}$ where $S_u$ is the bit size of an uncompressed image. For a $224\times 224$ pixel RGB 8-bit image, $S_u$ is 150,528 bytes, and $S_c$ is the size of the corresponding losslessly compressed PNG file.
An example is shown in Fig \[f:clusteringCR\]. Observe that most patches fall into one of the three clusters (${CR}^{-1} = 0.1, 0.4, 0.7$) as well as that patches with higher ${CR}^{-1}$ tend to contain more salient (cancer related) information than others. However, it is important to note that this does not necessarily imply that they are more pertinent for *prognosis* i.e. our ultimate task. For example, the spatial arrangement of immune and cancer cells in peripheral regions around the tumour is known to be informative in this regard.
#### Phenotype clustering
We also developed a new *phenotype clustering* approach [@zhu2017wsisa]. The motivation behind phenotype clustering stems from the observation that the extracted patches exhibit significant heterogeneity; see Fig \[f:tiles\]. Because it is computationally expensive to perform clustering in the original 150,528 dimensional space ($224 \times 224 \times 3$), herein (instead of performing simple down-sampling) we used an ImageNet pre-trained CNN to generate phenotypes and then principal component analysis for dimensionality reduction. A summary of the process is shown in Table \[t:clusteringPhenotype\] and visual examples in Fig \[f:clusteringPhenotype\].
[l|c]{} Operation & Input dimension\
Pre-trained CNN & $224 \times 224 \times 3$\
Global average pooling & $7 \times 7 \times 512$\
Dimension reduction (PCA) & $512$\
$k$-means clustering & $50$\
\[t:clusteringPhenotype\]
![Examples of clustering results. Each block of four images comprises the original WSI (top left), and the corresponding information density (top right) and phenotype based ($k=5$ and $k=10$ respectively bottom left and bottom right) clustering labels, colour coded.[]{data-label="f:clusteringPhenotype"}](res_01.png "fig:"){width=".48\textwidth"} ![Examples of clustering results. Each block of four images comprises the original WSI (top left), and the corresponding information density (top right) and phenotype based ($k=5$ and $k=10$ respectively bottom left and bottom right) clustering labels, colour coded.[]{data-label="f:clusteringPhenotype"}](res_02.png "fig:"){width=".48\textwidth"}
Patch-wise CNN prognosis
------------------------
Our CNN design was inspired by the well-known VGG16 network though with some notable changes to its architecture and training, made to best suit the problem at hand. The main of these are: (i) size of the fully connected layer, (ii) learning optimizer, (iii) learning rate, and (iv) use of transfer learning.
We apply the CNN on the patch level and regard the ground truth label of the entire corresponding WSI as its label. The fact that not all patches contain discriminative features (for survival prediction) motivated our choice to train a network for each cluster separately and independently. Then the clusters which correspond to the networks that converge and have high validation accuracy are inferred to be discriminative. Networks of non-discriminative patches clusters may also converge but their high validation error still allows for the automatic inference of their non-discriminative nature.
Aggregation of predictions
--------------------------
The validation accuracy threshold of 65% was chosen for the selection of discriminative clusters. After the patch levels predictions are made, these are aggregated into cluster level predictions. Since each WSI has different numbers of tiles from a given cluster, patch level predictions are represented by normalized histograms, thus effecting a homogeneous representation. A support vector machine (SVM) classifier [@BirkAranHump2017] is trained to learn the cluster level outcome.
Experimental evaluation
=======================
Patient samples and ethics
--------------------------
Whole slide images from patients operated on in NHS Lothian hospitals were used in the present work. Our data set comprises WSIs of tissue sections from each of the diagnostically residual and archived formalin fixed paraffin embedded tissue blocks, from CRC stage I and stage II patients who underwent surgical resection. This work was conducted in accordance with the declaration of Helsinki and no patient identifiable information was provided to the researchers. Ethical approval was obtained after review by the NHS Lothian NRS BioResource, REC approved Research Tissue Bank (REC approval ref: 15/ES/0094), granted by East of Scotland Research Ethics Service. Apart from the WSIs, each patient data sample is accompanied by a demographical description and follow-up information including age at surgery, date of death, and whether this patient dies of CRC (recall that in the present paper we are interested in outcome prognosis i.e. survivability prediction). The original slides were stained using haematoxylin and eosin at the time of treatment, and were scanned using a ZEISS Axio scan Z1 (Zeiss, Oberkochen, DE) whole slide scanner with a $40\times$ objective. The scale of a single pixel represents $0.111\mu m \times 0.111\mu m$ of the actual size. The digital camera used was a Hitachi HVF2025SCL with an exposure time of 200$\mu s$. As our focus is on prognostic classification, all the images in the corpus correspond to positively diagnosed CRC cases.
The entire data set contains 75 WSIs representing one slide from a single patient. The smallest image is 6GB and the largest one is 18GB, with an average size of the WSIs being 8GB, which is approximately 300,000 pixels by 200,000 pixels. The bit depth is 24 with 3 channels. The original data was in the commonly used CZI format and included an identity label for each slide.
Results and discussion
----------------------
We started our analysis by examining the overall prediction results and, in particular, the effect that different clustering and aggregation techniques, and their parameters have. For comparison, in addition to our SVM based aggregation described in the previous section, we also present results for majority voting based aggregation [@AranCipo2004a]. A summary is provided in Table \[t:results\].
[l||cc|cc]{} & &\
& Accuracy & F1 & Accuracy & F1\
ID3-CNN-Vote & 0.60 & 0.67 & 0.50 & 0.67\
ID3-CNN-SVM & 0.60 & 0.67 & 1.00 & 1.00\
Ph5-CNN-Vote & 0.68 & 0.67 & 1.00 & 1.00\
Ph5-CNN-SVM & 0.68 & 0.67 & 1.00 & 1.00\
Ph10-CNN-Vote & 0.70 & 0.81 & 0.50 & 0.67\
Ph10-CNN-SVM & 0.70 & 0.81 & 0.50 & 0.67\
\[t:results\]
There are several important observations that are readily apparent from the table. Firstly, all of the approaches – that is, different combinations of clustering and decision fusion techniques – performed very well indeed already on the level of individual patches. On the patch level phenotype based clustering with larger $k$ performed best. Interestingly, at this stage the manner of decision fusion (majority vote vs. SVM based) made no difference in the context of any of the different algorithms.
Further insight can be gained by looking at cluster level performance, especially when interpreted in the context of the aforementioned results. Here we do observe some advantage of SVM based decision fusion, albeit only when information density based clustering is employed. Another noteworthy observation is that unlike in other cases, cluster level prediction is worse when phenotype based clustering with larger $k$ is used. Our hypothesis, which requires further experiments for validation, is that this is not a ‘true’ trend but rather a stochastic anomaly which emerges from the need of more data for large $k$. Lastly, and most importantly, moving from patch to cluster level prediction using SVM based fusion dramatically improves algorithms with both information density and small $k$ phenotype based clustering, and results in perfect performance. This observation also supports our hypothesis as regards the anomaly noticed for large $k$ phenotype based clustering.
Finally, we sought additional insight and examined the $k=5$ clusters for the best performing method (Ph5-CNN-SVM). What we found was that the semantics of the five clusters were very easy to interpret and can be summarized as containing the following patch types – Cluster 0: outside of the tissue micro-section, Cluster 1: containing blood cells, Cluster 2: containing cancerous and immune cells, Cluster 3: void, and Cluster 4: containing fat cells. Thus, we can conclude that our method not only performs extremely well in terms of the ultimate goal of survival prognosis but also that it does so by learning clinically meaningful problem structure. Ample previous work testifies to the importance of interpretability in the adoption of novel machine learning assistive tools by medical professionals.
Summary and conclusions
=======================
This paper is the first work to address one of the most challenging problems in the emerging sphere of digital pathology – that of using images not previously annotated by a pathologist to develop algorithms that can be applied automatically to generate diagnostic and prognostic information from WSIs. Almost all current applications of CNN require careful annotation of tissue images by a qualified pathologist, and this is a rate limiting step. The novel algorithm we introduced addresses the overwhelming amount of data by automatic, unsupervised discriminative patch selection and the convergence performance of cluster level trained convolutional neural networks, and the inference of prognosis on the level of individual discriminative clusters followed by decision fusion using support vector machines. On a real-world corpus our phenotype based clustering employed in conjunction with the aforementioned techniques achieved perfect performance both in terms of overall accuracy and F1 score.
Acknowledgements {#acknowledgements .unnumbered}
================
The authors gratefully acknowledge the support of NVIDIA Corporation for their donation of the Titan Xp GPU used for this research.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We investigate the holographic superconductors in Gauss-Bonnet gravity with Born-Infeld electrodynamics. We find that the Gauss-Bonnet constant, the model parameters and the Born-Infeld coupling parameter will affect the formation of the scalar hair, the transition point of the phase transition from the second order to the first order, and the relation connecting the gap frequency in conductivity with the critical temperature. The combination of the Gauss-Bonnet gravity and the Born-Infeld electrodynamics provides richer physics in the phase transition and the condensation of the scalar hair.'
author:
- 'Jiliang [Jing]{}[^1]'
- Liancheng Wang
- Qiyuan Pan
- Songbai Chen
title: |
Holographic Superconductors in Gauss-Bonnet gravity\
with Born-Infeld electrodynamics
---
Introduction
============
The AdS/CFT correspondence [@Maldacena; @polyakov; @Witten] relates a weak coupling gravity theory in an anti-de Sitter space to a strong coupling conformal field theory in one less dimensions. Recently it has been applied to condensed matter physics and in particular to superconductivity [@Gubser:2005ih; @GubserPRD78]. In the pioneering papers Gubser [@Gubser:2005ih; @GubserPRD78] suggested that near the horizon of a charged black hole there is in operation a geometrical mechanism parameterized by a charged scalar field of breaking a local $U(1)$ gauge symmetry. Then, the gravitational dual of the transition from normal to superconducting states in the boundary theory was constructed. This dual consists of a system with a black hole and a charged scalar field, in which the black hole admits scalar hair at temperature lower than a critical temperature, but does not possess scalar hair at higher temperatures [@HartnollPRL101]. In this system a scalar condensate can take place through the coupling of the scalar field with the Maxwell field. Much attention has been focused on the application of AdS/CFT correspondence to condensed matter physics since then [@HartnollJHEP12; @HorowitzPRD78; @Nakano-Wen; @Amado; @Koutsoumbas; @Maeda79; @Sonner; @HartnollRev; @HerzogRev; @Ammon:2008fc; @Gubser:2009qm; @CJ0].
Recently, it is of great interest to generalize the investigation to the Einstein-Gauss-Bonnet gravity, which is motivated by the application of the Mermin-Wagner theorem to the holographic superconductors. It was found [@Gregory; @Pan-Wang; @Ge-Wang; @Brihaye; @Gregory1009] that the higher curvature corrections in general make the condensation of the scalar field harder to form and give larger corrections to the so-called Horowitz’s relation $\omega_{g}/T_c\approx8$ for the conductivity. And then the general holographic superconductor models in Einstein-Gauss-Bonnet gravity are constructed and it is observed that different values of Gauss-Bonnet correction term and model parameters can determine the order of phase transitions and critical exponents of second-order phase transitions [@Pan-Wang1]. Very recently, the holographic p-wave superconductor models in the Gauss-Bonnet gravity was introduced in the probe limit and the Gauss-Bonnet correction term can also effect the condensation of the vector field [@Cai-pGB].
On the other hand, non-linear electrodynamics has been a subject of research for many years. Heisenberg and Euler [@euler] noted that quantum electrodynamics predicts that the electromagnetic field behaves non-linearly through the presence of virtual charged particles. Born and Infeld [@born] presented a new classical non-linear theory of electromagnetism which contains many symmetries common to the Maxwell theory despite its non-linearity. It was found that the Born-Infeld electrodynamics is the only possible non-linear version of electrodynamics that is invariant under electromagnetic duality transformations [@gibb]. Thus, the interest of study for Born-Infeld electrodynamics has been arisen in [@Olivera1; @Hoffman-Gibbons-Rasheed; @Oliveira]. The static spherically symmetric black holes for the Born-Infeld electrodynamics coupled to Einstein gravity was derived in Refs. [@Hoffman-Gibbons-Rasheed; @Oliveira]. Within the framework of AdS/CFT correspondence, we studied the effects of the Born-Infeld electrodynamics on the holographic superconductors in the background of a Schwarzschild-AdS black hole spacetime [@Jing-Chen].
Motivated by the recent studies mentioned above and the fact that, within the framework of AdS/CFT correspondence, higher-derivative corrections to either gravitational or electromagnetic action in AdS space are expected to modify the dynamics of the strongly coupled dual theory, in this paper we will investigate the behavior of the holographic superconductors in the Gauss-Bonnet gravity with the Born-Infeld electrodynamics in a five dimensional planar black-hole background, and to see how the combination of the Gauss-Bonnet gravity and the Born-Infeld electrodynamics affect the formation of the scalar hair, the phase transition and Horowitz’s relation.
The paper is organized as follows. In Sec. II, we explore the scalar condensation in the background of the Gauss-Bonnet black hole by introducing a complex charged scalar field coupling with an electric field obeyed to Born-Infeld electrodynamics. In Sec. III, we study the electrical conductivity and find the ratio of the gap frequency in conductivity to the critical temperature. We summarize and discuss our conclusions in the last section.
Scalar condensation
===================
The Einstein-Gauss-Bonnet theory is the most general Lovelock theory in five and six dimensions and the action is described by $$I_{grav}=\frac{1}{16\pi G}\int\limits_{\mathcal{M}}d^{d}x\,\sqrt{-g}\,\left[
R-2\Lambda +\hat{\alpha} \,\left( R^{2}-4R_{\mu \nu }R^{\mu \nu }+R_{\mu \nu
\lambda \sigma }R^{\mu \nu \lambda \sigma }\right) \right] \,,$$where $\Lambda=-(d-1)(d-2)/(2L^2) $ is the cosmological constant, $G$ is the gravitational constant, and $\hat{\alpha} $ is the Gauss-Bonnet coupling constant. The static spacetime of a neutral black hole in $d$ dimensional Einstein-Gauss-Bonnet gravity is [@Boulware-Deser; @Cai-2002; @Charmousis:2002rc] $$\begin{aligned}
\label{BH metric}
ds^2=-f(r)dt^{2}+\frac{dr^2}{f(r)}+r^{2}dx_{i}dx^{i},\end{aligned}$$ with $$\begin{aligned}
f(r)=\frac{r^2}{2\alpha}\left[1-\sqrt{1-\frac{4\alpha}{L^{2}}
\left(1-\frac{ML^{2}}{r^{d-1}}\right)}~\right],\end{aligned}$$ where $\alpha=\hat{\alpha}(d-3)(d-4)$ and the constant $M$ is relate to the black hole horizon by $r_{+}=(ML^{2})^{1/(d-1)}$. In the asymptotic region ($r\rightarrow\infty$), we have $
f(r)\sim\frac{r^2}{2\alpha}\left(1-\sqrt{1-4\alpha/L^2} \right). $ Thus, we can define the effective asymptotic AdS scale by $ L^2_{\rm
eff}=(2\alpha)/(1-\sqrt{1-4\alpha/L^2}). $ The Hawking temperature of the black hole is $$\begin{aligned}
\label{Hawking temperature} T=\frac{(d-1)r_{+}}{4\pi L^{2}}\ .\end{aligned}$$
We now consider the Born-Infeld electrodynamics and the charged scalar field coupled via a generalized Lagrangian $$\begin{aligned}
\label{System}
S=\int d^{d}x\sqrt{-g}\left[
\frac{1}{b^{2}}\left( 1
-\sqrt{1+\frac{b^{2} F^{2}}{2}}\right)
-\frac{1}{2}\partial_{\mu}\tilde{\psi}\partial^{\mu}\tilde{\psi}
-\frac{1}{2}m^2\tilde{\psi}^2-\frac{1}{2}|\mathfrak{F}(\tilde{\psi})|(\partial_{\mu}p-A_{\mu})
(\partial^{\mu}p-A^{\mu}) \right] \ ,\nonumber\\\end{aligned}$$ where $\mathfrak{F}$ is taken as $
\mathfrak{F}(\tilde{\psi})=\tilde{\psi}^{2}
+c_{\gamma}\tilde{\psi}^{\gamma}+c_{4}\tilde{\psi}^{4} $ with the model parameters $c_{\gamma}$, $\gamma$ and $c_{4}$ in order to introduce a general class of gravity duals to superconducting theories that exhibit both first and second-order phase transitions at finite temperature in strongly interacting systems [@FrancoPRD; @Franco], and it reduces to the model considered in [@Jing-Chen] if $c_{\gamma}$ and $c_{4}$ are zero. It should be noted that $b$ is the Born-Bonnet coupling parameter and the Born-Infeld electrodynamics will reduce to the Maxwell case in the weak-coupling limit $b\rightarrow 0$. We can use the gauge freedom to fix $p=0$ and take $\psi\equiv\tilde{\psi}$, $A_{t}=\phi$ where $\psi$, $\phi$ are both real functions of $r$ only. Then the equations of motion are given by $$\begin{aligned}
\label{Psi}
&&\psi^{\prime\prime}+\left(
\frac{f^\prime}{f}+\frac{d-2}{r}\right)\psi^\prime
+\frac{\phi^2}{f^2}\left(\psi+\frac{\gamma}{2}c_\gamma \psi^\gamma+2c_4\psi^3 \right)-\frac{m^2}{f}\psi=0\,,
\\ \label{Phi} &&
\left(\phi^{\prime\prime}+\frac{d-2}{r}\phi^\prime\right)
\left(1-b^2 \phi^{\prime 2}\right)+b^2\phi^{\prime 2}\phi^{\prime \prime}-\left(1-b^2 \phi^{\prime 2}\right)^{\frac{3}{2}}\left(\psi^2+c_\gamma \psi^\gamma+c_4\psi^4 \right)\frac{\phi}{f}=0~,\end{aligned}$$ where a prime denotes the derivative with respect to $r$. At the event horizon $r=r_+$, we must have $$\begin{aligned}
\psi(r_{+})&=&-\frac{(d-1) }{ m^{2} L^2}\psi^\prime(r_{+}),~~~~ \phi(r_{+})=0,\end{aligned}$$ and at the asymptotic AdS region ($r\rightarrow\infty$), the solutions behave like $$\begin{aligned}
\psi=\frac{\psi_{-}}{r^{\lambda_{-}}}+\frac{\psi_{+}}{r^{\lambda_{+}}}\,,\hspace{0.5cm}
\phi=\mu-\frac{\rho}{r^{d-3}}\,, \label{infinity}\end{aligned}$$ with $$\begin{aligned}
\lambda_\pm=\frac{1}{2}\left[(d-1)\pm\sqrt{(d-1)^{2}+4m^{2}L_{\rm
eff}^2}~\right]\,, \label{LambdaZF}\end{aligned}$$ where $\mu$ and $\rho$ are interpreted as the chemical potential and charge density in the dual field theory respectively. We take $\psi_{-}=0$ because we can impose boundary condition that either $\psi_{+}$ or $\psi_{-}$ vanishes [@HartnollPRL101; @HartnollJHEP12], and we will focus on $d=5$ and $m^2L^2=-3$ here. Thus, the scalar condensate is now described by the operator $\langle{\cal O}_{+}\rangle=\psi_{+}$. In what following we will present a detail analysis of the condensation of the operator $\langle{\cal O}_{+}\rangle$ by taking numerical integration of the equations (\[Psi\]) and (\[Phi\]) from the horizon out to the infinity with the boundary conditions mentioned above.
![\[condensate\] (Color online) The condensate $\langle{\cal O}_{+}\rangle$ as a function of temperature with fixed value $\alpha=0.1$ for different values of the model parameters ($c_\gamma,~\gamma, c_4$) and Born-Infeld coupling parameter $b$, which shows that a different values of these parameters not only change the formation of the scalar hair, but also separate the first- and second-order phase transition. ](GBIFconda01c00 "fig:")![\[condensate\] (Color online) The condensate $\langle{\cal O}_{+}\rangle$ as a function of temperature with fixed value $\alpha=0.1$ for different values of the model parameters ($c_\gamma,~\gamma, c_4$) and Born-Infeld coupling parameter $b$, which shows that a different values of these parameters not only change the formation of the scalar hair, but also separate the first- and second-order phase transition. ](GBIFconda01c03 "fig:")\
![\[condensate\] (Color online) The condensate $\langle{\cal O}_{+}\rangle$ as a function of temperature with fixed value $\alpha=0.1$ for different values of the model parameters ($c_\gamma,~\gamma, c_4$) and Born-Infeld coupling parameter $b$, which shows that a different values of these parameters not only change the formation of the scalar hair, but also separate the first- and second-order phase transition. ](GBIFconda01c05 "fig:") ![\[condensate\] (Color online) The condensate $\langle{\cal O}_{+}\rangle$ as a function of temperature with fixed value $\alpha=0.1$ for different values of the model parameters ($c_\gamma,~\gamma, c_4$) and Born-Infeld coupling parameter $b$, which shows that a different values of these parameters not only change the formation of the scalar hair, but also separate the first- and second-order phase transition. ](GBIFconda01c07 "fig:")\
![\[condensate\] (Color online) The condensate $\langle{\cal O}_{+}\rangle$ as a function of temperature with fixed value $\alpha=0.1$ for different values of the model parameters ($c_\gamma,~\gamma, c_4$) and Born-Infeld coupling parameter $b$, which shows that a different values of these parameters not only change the formation of the scalar hair, but also separate the first- and second-order phase transition. ](GBIFconda01y3cy01c00 "fig:")![\[condensate\] (Color online) The condensate $\langle{\cal O}_{+}\rangle$ as a function of temperature with fixed value $\alpha=0.1$ for different values of the model parameters ($c_\gamma,~\gamma, c_4$) and Born-Infeld coupling parameter $b$, which shows that a different values of these parameters not only change the formation of the scalar hair, but also separate the first- and second-order phase transition. ](GBIFconda01y3cy01c01 "fig:")\
![\[condensate\] (Color online) The condensate $\langle{\cal O}_{+}\rangle$ as a function of temperature with fixed value $\alpha=0.1$ for different values of the model parameters ($c_\gamma,~\gamma, c_4$) and Born-Infeld coupling parameter $b$, which shows that a different values of these parameters not only change the formation of the scalar hair, but also separate the first- and second-order phase transition. ](GBIFconda01y35cy01c00 "fig:") ![\[condensate\] (Color online) The condensate $\langle{\cal O}_{+}\rangle$ as a function of temperature with fixed value $\alpha=0.1$ for different values of the model parameters ($c_\gamma,~\gamma, c_4$) and Born-Infeld coupling parameter $b$, which shows that a different values of these parameters not only change the formation of the scalar hair, but also separate the first- and second-order phase transition. ](GBIFconda01y35cy01c01 "fig:")
$ $ $c_{4}=0$ $c_{4}=0.3$ $c_{4}=0.5$ $c_{4}=0.7$ $c_{4}=1.0$
---------------- ----------- ------------------------------------------ ------------------------------------------ ------------------------------------------- ------------------------
$\alpha=0.1~~$ No $b_c=\frac{\sqrt{2}}{10}$ $T_c=0.16253$ $b_c=\frac{1}{10}$ $T_c=0.17456$ $b_c=0$ $T_c=0.18622$ No
$\alpha=0.0~~$ No $b_c=\frac{2}{10}$ $T_c=0.16566$ $b_c=\frac{\sqrt{2}}{10}$ $T_c=0.18118$ $b_c=\frac{1}{10}$ $T_c=0.18973$ No
$\alpha=-0.1$ No $b_c=\frac{\sqrt{7}}{10}$ $T_c=0.16824$ $b_c=\frac{2}{10}$ $T_c=0.18454$ $b_c=\frac{\sqrt{2}}{10}$ $T_c=0.19636$ $b_c=0$ $T_c=0.20636$
: \[Tc-abc\] The critical values of $T_c$ and $b_c$ for different $\alpha$ and $c_{4}$, which can separate the first- and second-order phase transitions for the simple model $\mathfrak{F}
(\psi) = \psi^2+c_4\psi^4$. The word “No" in the table corresponds to the inexistence of the critical point.
We present in Fig. \[condensate\] the influence of the parameters ($c_\gamma,~\gamma,~c_4$) and $b$ on the condensation with fixed values $ m^2 L^2=-3$ and $\alpha=0.1$. In fact, the different choices of $\alpha$ can not qualitatively change our results. We know from the figure that the Born-Infeld coupling parameter and the model parameters have obvious different effects on the critical temperature. If we fix the model parameters ($c_\gamma,~\gamma,~c_4$), we note that the critical temperature becomes smaller as the Born-Infeld coupling parameter $b$ increases for two types of phase transitions, i.e., the scalar hair can be formed harder for the larger $b$. However, the story is completely different if we fix the Born-Infeld coupling parameter $b$. For the cases of second phase transition, the critical temperature keeps as a constant with the increase of ($c_\gamma,~\gamma,~c_4$). That is to say, the formation of the scalar hair does not affect by model parameter ($c_\gamma,~\gamma,~c_4$). But for the cases of first phase transition, the critical temperature is larger with the increase of ($c_\gamma,~c_4$) or decrease of $\gamma$, which means that the scalar hair can be formed easier for the larger model parameter ($c_\gamma,~c_4$) or smaller $\gamma$. It should be pointed out that we define the critical temperature $T_{c}$ just as Franco [*et al*]{} for the first phase transition [@Franco].
From Fig. \[condensate\] we also find that, for a simple model $\mathfrak{F}(\psi)=\psi^{2}+c_{4}\psi^{4}$ with fixed $c_4$ ($c_4>0)$, there is a phase transition from the second order to the first one as we increase value of $b$. In table \[Tc-abc\], we list the critical values of $b_{c}$ and $T_{c}$ which separate the second order and the first order phase transitions for selected $\alpha$ and $c_4$. Note that the word “No" in this table corresponds to the inexistence of the critical values of $b_{c}$ and $T_{c}$, i.e., the phase transition is always of the second order if $c_{4}=0$ with different $\alpha$ but the first order if $c_{4}=1.0$ with $\alpha=0.0$ and $0.1$. We learn from the figure and the table that both $b_c$ and $T_c$ decrease as $\alpha$ increases for fixed $c_{4}$, and $b_c$ decreases but $T_c$ increases as $c_{4}$ increases for fixed $\alpha$. Thus, the Gauss-Bonnet constant $\alpha$, the model parameter $c_4$ and the Born-Infeld coupling parameter $b$ provide richer physics in the phase transition.
Electrical Conductivity
========================
In the study of (2+1) and (3+1)-dimensional superconductors, Horowitz [*et al.*]{} [@HorowitzPRD78] got a universal relation connecting the gap frequency in conductivity with the critical temperature, which is described by $$\begin{aligned}
\frac{\omega_g}{T_c}\approx 8,\end{aligned}$$ with deviations of less than $8\%$. This is roughly twice the BCS value 3.5 indicating that the holographic superconductors are strongly coupled. However, the authors in Refs. [@Pan-Wang; @Gregory] found that this relation is not stable in the presence of the Gauss-Bonnet correction terms. We now examine this relation for the Gauss-Bonnet gravity with the Born-Infeld electrodynamics.
In order to compute the electrical conductivity, we should study the Born-Infeld electromagnetic perturbation in this Gauss-Bonnet black hole background, and then calculate the linear response to the perturbation. In the probe approximation, the effect of the perturbation of metric can be ignored. Assuming that the perturbation of the vector potential is translational symmetric and has a time dependence as $\delta A_x=A_x(r)e^{-i\omega t}$, we find that the motion equation for the Born-Infeld electrodynamics in the Gauss-Bonnet black hole background reads $$\begin{aligned}
&&\left(A_{x}^{\prime\prime}+\frac{f^\prime}{f}A_{x}^\prime
+\frac{d-4}{r}A_{x}^{\prime}+\frac{\omega^2}{f^2}A_{x}\right)\left(1-b^2 \phi^{\prime 2}\right)\nonumber \\
&&+b^2\phi^{\prime }\phi^{\prime \prime}A_{x}^{\prime}
-\left(1-b^2 \phi^{\prime 2}\right)^{\frac{3}{2}}\left(\psi^2+c_\gamma \psi^\gamma+c_4 \psi^4\right)\frac{A_{x}}{f}=0 \; .
\label{Maxwell Equation}\end{aligned}$$ From Eq. (\[Maxwell Equation\]), an ingoing wave boundary condition near the horizon is given by $$\begin{aligned}
A_{x}(r)\sim f(r)^{-\frac{i \omega L^2}{(d-1) r_+}}, \end{aligned}$$ and in the asymptotic AdS region ($r\rightarrow\infty$), the general behavior for $d=5$ should be [@Gregory1009] $$\begin{aligned}
\label{Maxwell boundary}
A_{x}=L_e^{-1/2} A^{(0)}+\frac{L_e^{3/2}}{r^2}\left(A^{(2)}-\frac{1}{2} \ln \frac{r}{L} \partial_t^2 A^{(0)} \right).\end{aligned}$$ Then the conductivity can be expressed as [@Gregory1009] $$\begin{aligned}
\label{GBConductivity}
\sigma=\frac{2A^{(2)}}{i\omega A^{(0)}}+\frac{i\omega}{2} -i \omega \log\frac{L_e}{L} \ ,\end{aligned}$$ where the factor of $L_e^{-1/2}$ ensures that the gauge fields $A^{(n)}_\mu$ have the correct dimensionality. We can obtain the conductivity by solving the motion equation (\[Maxwell Equation\]) numerically for the general forms of function $\mathfrak{F}(\psi)=\psi^{2} +c_{\gamma}\psi^{\gamma}+c_4\psi^4$. Here we also focus our attention on the case for $m^{2}L^{2}=-3$.
-------------------- ----------- ----------- ------------ -------------- --------------- ---------------
b=0.0 b=0.1 b=0.2 b=0.0 b=0.1 b=0.2
$\alpha=0.1~~$ $8.5$ $9.0$ $10.0$ $ 9.3 $ $ 9.7 $ $ 11.6 $
$\alpha=0.0~~$ $ 7.7 $ $8.1 $ $8.7 $ $ 8.4 $ $ 8.7 $ $ 9.7 $
$\alpha=-0.1$ $7.3$ $7.6$ $7.9$ $ 7.8 $ $ 8.1 $ $ 9.0 $
-------------------- ----------- ----------- ------------ -------------- --------------- ---------------
: \[ConductivityTc\] The ratio $\omega_{g}/T_{c}$ for different values of the Gauss-Bonnet constant $\alpha$, the model parameter $c_\gamma$ and the Born-Infeld coupling parameter $b$ with $m^{2}L^2=-3$ and $\gamma=3$.
![\[Conductivity\] (Color online) The conductivity of the superconductors as a function of $\omega/T_c$ for different values of $\alpha$ and $b$ with $c_4=0$, $c_\gamma=0$ and $m^{2}L^{2}=-3$. The blue (bottom) line represents the real part of the conductivity, $Re(\sigma)$, and red (top) line is the imaginary part of the conductivity, $Im(\sigma)$.](GBIFa01y00c00b00 "fig:")![\[Conductivity\] (Color online) The conductivity of the superconductors as a function of $\omega/T_c$ for different values of $\alpha$ and $b$ with $c_4=0$, $c_\gamma=0$ and $m^{2}L^{2}=-3$. The blue (bottom) line represents the real part of the conductivity, $Re(\sigma)$, and red (top) line is the imaginary part of the conductivity, $Im(\sigma)$.](GBIFa01y00c00b01 "fig:")![\[Conductivity\] (Color online) The conductivity of the superconductors as a function of $\omega/T_c$ for different values of $\alpha$ and $b$ with $c_4=0$, $c_\gamma=0$ and $m^{2}L^{2}=-3$. The blue (bottom) line represents the real part of the conductivity, $Re(\sigma)$, and red (top) line is the imaginary part of the conductivity, $Im(\sigma)$.](GBIFa01y00c00b02 "fig:")\
![\[Conductivity\] (Color online) The conductivity of the superconductors as a function of $\omega/T_c$ for different values of $\alpha$ and $b$ with $c_4=0$, $c_\gamma=0$ and $m^{2}L^{2}=-3$. The blue (bottom) line represents the real part of the conductivity, $Re(\sigma)$, and red (top) line is the imaginary part of the conductivity, $Im(\sigma)$.](GBIFa00y00c00b00 "fig:")![\[Conductivity\] (Color online) The conductivity of the superconductors as a function of $\omega/T_c$ for different values of $\alpha$ and $b$ with $c_4=0$, $c_\gamma=0$ and $m^{2}L^{2}=-3$. The blue (bottom) line represents the real part of the conductivity, $Re(\sigma)$, and red (top) line is the imaginary part of the conductivity, $Im(\sigma)$.](GBIFa00y00c00b01 "fig:")![\[Conductivity\] (Color online) The conductivity of the superconductors as a function of $\omega/T_c$ for different values of $\alpha$ and $b$ with $c_4=0$, $c_\gamma=0$ and $m^{2}L^{2}=-3$. The blue (bottom) line represents the real part of the conductivity, $Re(\sigma)$, and red (top) line is the imaginary part of the conductivity, $Im(\sigma)$.](GBIFa00y00c00b02 "fig:")\
![\[Conductivity\] (Color online) The conductivity of the superconductors as a function of $\omega/T_c$ for different values of $\alpha$ and $b$ with $c_4=0$, $c_\gamma=0$ and $m^{2}L^{2}=-3$. The blue (bottom) line represents the real part of the conductivity, $Re(\sigma)$, and red (top) line is the imaginary part of the conductivity, $Im(\sigma)$.](GBIFam01y00c00b00 "fig:")![\[Conductivity\] (Color online) The conductivity of the superconductors as a function of $\omega/T_c$ for different values of $\alpha$ and $b$ with $c_4=0$, $c_\gamma=0$ and $m^{2}L^{2}=-3$. The blue (bottom) line represents the real part of the conductivity, $Re(\sigma)$, and red (top) line is the imaginary part of the conductivity, $Im(\sigma)$.](GBIFam01y00c00b01 "fig:")![\[Conductivity\] (Color online) The conductivity of the superconductors as a function of $\omega/T_c$ for different values of $\alpha$ and $b$ with $c_4=0$, $c_\gamma=0$ and $m^{2}L^{2}=-3$. The blue (bottom) line represents the real part of the conductivity, $Re(\sigma)$, and red (top) line is the imaginary part of the conductivity, $Im(\sigma)$.](GBIFam01y00c00b02 "fig:")\
In Fig. \[Conductivity\] and table \[ConductivityTc\] we present the frequency dependent conductivity obtained by solving the motion equation of the Born-Infeld electrodynamics numerically for different values of $\alpha$, $b$ and $c_\gamma$ with $c_4=0$, $\gamma=3$ and $m^{2}L_{AdS}^{2}=-3$ (we plot the conductivity at temperature $T/T_c\simeq 0.3$). We find that the gap frequency $\omega_{g}$ increases with the increase of the coupling parameter $b$ for fixed $\alpha$ and $c_\gamma$, it decreases as $\alpha$ decreases for fixed $b$ and $c_\gamma$, and it increases as $c_\gamma$ increase for fixed $\alpha$ and $b$. From Figs. \[Conductivity\] and table \[ConductivityTc\], we find that the ratio of the gap frequency in conductivity $\omega_g$ to the critical temperature $T_c$ in the Gauss-Bonnet black hole with the Born-Infeld electrodynamics depends on the Gauss-Bonnet constant, the model parameters and the Born-Infeld coupling parameter.
conclusions
===========
The behaviors of the holographic superconductors in the Gauss-Bonnet gravity have been investigated by introducing a complex charged scalar field coupling with an electric field obeyed to Born-Infeld electrodynamics in a planar black-hole background. We present a detail analysis of the condensation of the operator $\langle{\cal O}_{+}\rangle$ by numerical method. For the interesting simple model $\mathfrak{F}(\psi)=\psi^{2}+c_{4}\psi^{4}$, we know that there is a phase transition from the second order to the first one as we alter the values of the Gauss-Bonnet constant $\alpha$, the model parameter $c_4$ and the Born-Infeld coupling parameter $b$. For the transition point, the relation of $\alpha$, $c_4$ and the critical values $b_c$ and $T_c$ which can separate the first- and second-order behavior is: both $b_c$ and $T_c$ decrease as $\alpha$ increases for fixed $c_{4}$, and $b_c$ decreases but $T_c$ increases as $c_{4}$ increases for fixed $\alpha$. It is interesting to find that the Born-Infeld coupling parameter and model parameters have obvious different effects on the critical temperature for general model $\mathfrak{F}(\psi)=\psi^{2} +c_{\gamma}\psi^{\gamma}+c_4\psi^4$. If we fix the model parameters ($c_\gamma,~\gamma,~c_4$), we note that the critical temperature becomes smaller as the Born-Infeld coupling parameter $b$ increases for two types of phase transitions, i.e., the scalar hair can be formed harder for the larger $b$. However, the story is completely different if we fix the Born-Infeld coupling parameter $b$. For the cases of second phase transition, the formation of the scalar hair does not affect by model parameters ($c_\gamma,~\gamma,~c_4$). But for the cases of first phase transition, the scalar hair can be formed easier for the larger model parameter ($c_\gamma,~c_4$) or smaller $\gamma$. We finally find that the ratio of the gap frequency in conductivity $\omega_g$ to the critical temperature $T_c$ in the Gauss-Bonnet black hole with the Born-Infeld electrodynamics depends on the Gauss-Bonnet constant $\alpha$, model parameters $(c_4,~c_\gamma, ~\gamma)$, and the coupling parameter $b$. Thus, the Gauss-Bonnet constant, model parameters and Born-Infeld coupling parameter provide richer physics in the phase transition and the condensation of the scalar hair.
This work was supported by the National Natural Science Foundation of China under Grant Nos 10875040, 10905020 and 10875041; a key project of the National Natural Science Foundation of China under Grant No 10935013; the National Basic Research of China under Grant No. 2010CB833004, PCSIRT under Grant No IRT0964, and the Construct Program of the National Key Discipline. Thanks the Kavali Institute for Theoretical Physics China for hospitality in the final stages of this work.
[99]{}
J. Maldacena, Adv. Theor. Math. Phys. [**2**]{}, 231 (1998).
S. S. Gubser, I. R. Klebanov and A. M. Polyakov, Phys. Lett. [**B 428**]{}, 105 (1998). \[hep-th/9802109\]
E. Witten, Adv. Theor. Math. Phys. [**2**]{}, 253 (1998).
S. S. Gubser, Class. Quant. Grav. [**22**]{}, 5121 (2005).
S. S. Gubser, Phys. Rev. D [**78**]{}, 065034 (2008).
S. A. Hartnoll, C. P. Herzog, and G. T. Horowitz, Phys. Rev. Lett. [**101**]{}, 031601 (2008).
S. A. Hartnoll, C. P. Herzog, and G. T. Horowitz, J. High Energy Phys. [**0812**]{}, 015 (2008).
G. T. Horowitz and M. M. Roberts, Phys. Rev. D [**78**]{}, 126008 (2008).
E. Nakano and Wen-Yu Wen, Phys. Rev. D [**78**]{}, 046004 (2008).
I. Amado, M. Kaminski, and K. Landsteiner, J. High Energy Phys. [**0905**]{}, 021 (2009).
G. Koutsoumbas, E. Papantonopoulos and G. Siopsis, J. High Energy Phys. [**0907**]{}, 026 (2009).
K. Maeda, M. Natsuume, and T. Okamura, Phys. Rev. D [**79**]{}, 126004 (2009).
Julian Sonner, Phys. Rev. D [**80**]{}, 084031 (2009).
S. A. Hartnoll, Class. Quant. Grav.[**26**]{}, 224002 (2009) \[arXiv: 0903.3246.\]
C. P. Herzog, J. Phys. A [**42**]{}, 343001 (2009).
M. Ammon, J. Erdmenger, M. Kaminski, and P. Kerner, Phys. Lett. B [**680**]{}, 516 (2009). S. S. Gubser, C. P. Herzog, S. S. Pufu, and T. Tesileanu, Phys. Rev. Lett. [**103**]{}, 141601 (2009).
Songbai Chen, Liancheng Wang, Chikun Ding, and Jiliang Jing, Nucl. Phys. B[**836**]{}, 222 (2010). \[arXiv: 0912.2397\]
R. Gregory, S. Kanno, and J. Soda, J. High Energy Phys. [**0910**]{}, 010 (2009).
Q. Y. Pan, B. Wang, E. Papantonopoulos, J. Oliveria, and A.B. Pavan, Phys. Rev. D [**81**]{}, 106007 (2010).
X. H. Ge, B. Wang, S. F. Wu, and G. H. Yang, arXiv:1002.4901 \[hep-th\].
Y. Brihaye and B. Hartmann, Phys. Rev. D [**81**]{}, 126008 (2010); arXiv:1003.5130 \[hep-th\].
L. Barclay, R. Gregory, S. Kanno, and P. Sutcliffe, arXiv:1009.1991\[hep-th\].
Q. Y. Pan, B. Wang, Phys. Lett. B [**693**]{}, 159 (2010).
Rong-Gen Cai, Zhang-Yu Nie, and Hai-Qing Zhang, Phys. Rev. D[**82**]{}, 066007 (2010); arXiv:1007.3321. W. Heisenberg and H. Euler, Z. Phys. 98 (1936)
M. Born and L. Infeld, Proc. R. Soc. London, Ser A 144, 425-451 (1934)
G. W. Gibbons and D. A. Rasheed, Nucl. Phys. **B454**, 185 (1995).
B. Hoffmann, Phys. Rev. **47**, 877 (1935).
H. P. de Oliveira, Class. Quant. Grav. **11**, 1469 (1994).
Olivera Mišković and Rodrigo Olea, arXiv:1009.5763.
Jiliang Jing and Songbai Chen, Phys. Lett. B [**686**]{}, 68 (2010)
D. G. Boulware and S. Deser, Phys. Rev. Lett. [**55**]{}, 2656 (1985).
Rong-Gen Cai, Phys. Rev. D [**65**]{}, 084014 (2002).
C. Charmousis and J. F. Dufaux, Class. Quant. Grav. [**19**]{}, 4671 (2002). S. Franco, A. M. Garcia-Garcia, and D. Rodriguez-Gomez, Phys. Rev. D [**81**]{}, 041901(R) (2010). S. Franco, A. M. Garcia-Garcia, and D. Rodriguez-Gomez, JHEP [**04**]{}, 092 (2010).
[^1]: Electronic address: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
Using the Belle data on the decay $\Upsilon(10860)\to
\Upsilon(1S)f_0(980)\to\Upsilon(1S)\pi^+\pi^-$, we estimate the fraction of the process $\Upsilon(10860)\to\Upsilon(1S)f_0(980)\to
\Upsilon(1S)\eta\pi^0$, caused by the mixing of $a^0_0(980)$ and $f_0(980)$ resonances that breaks the isotopic invariance due to the $K^+$ and $K^0$ meson mass difference. With an instantaneous luminosity of $8\times10^{35}$ cm$^{-2}\cdot$s$^{-1}$ at the SuperKEKB collider, one can collect with the Belle II detector about a hundred $\Upsilon(10860)\to\Upsilon(1S)\eta\pi^0$ events in the narrow region of the $\eta\pi^0$ invariant mass near the $K\bar K$ thresholds in a time roughly spent on the present Belle experiment.
author:
- 'N. N. Achasov and G. N. Shestakov'
title: 'Observation of the isospin breaking decay $\Upsilon(10860)\to\Upsilon(1S) f_0(980)\to\Upsilon(1S)\eta\pi^0$ with the Belle II detector'
---
Recently, the Belle Collaboration performed a full amplitude analysis of three-body $e^+e^-\to\Upsilon(nS)\pi^+\pi^-$ ($n = 1, 2,
3$) transitions at $\sqrt{s}=10.865$ GeV and determined the relative fractions of various quasitwo-body components of the three-body amplitudes as well as the spin and parity of the two observed $Z_b$ states [@Ga15]. They also reported the first observation of the $e^+e^-\to\Upsilon(1S)f_0(980)$ transition. The fraction of the decay $\Upsilon(10860)\to\Upsilon(1S)f_0(980)$ is (see Table VI in Ref. [@Ga15]) $$\label{Eq1}
f_{\Upsilon(1S)f_0(980)}=\frac{BR(\Upsilon(10860)\to\Upsilon(1S)f_0
(980)\to\Upsilon(1S)\pi^+\pi^-)}{BR(\Upsilon(10860)\to\Upsilon(1S)
\pi^+\pi^-)}=\left(6.9\pm1.6^{+0.8}_{-2.8} \right)\%
\,.$$
The resonance $f_0(980)$ can decay into $\eta\pi^0$ via the transition $f_0(980)\to(K^+K^-+K^0\bar K^0)\to a^0_0(980)\to\eta
\pi^0$, i.e., owing to the $a^0_0(980)-f_0(980)$ mixing [@ADS79; @AS17]. Using, as a guide, the central value for the $a^0_0(980)-f_0(980)$ mixing intensity measured by the BESIII Collaboration in the reaction $J/\psi\to\phi f_0(980) \to\phi
a^0_0(980)\to\phi\eta\pi^0$ [@Ab11], $$\label{Eq2}
\xi_{fa}=\frac{BR(f_0(980)\to K\bar K\to a^0_0(980)\to\eta\pi^0)}
{BR(f_0(980)\to\pi^+\pi^-)}\approx0.009,$$ we obtain the following estimate for the isospin breaking decay fraction of the $\Upsilon(10860)$ $$\label{Eq3} f_{\Upsilon(1S)a^0_0(980)}=
\frac{BR(\Upsilon(10860)\to\Upsilon(1S)f_0(980)\to\Upsilon(1S)\eta\pi^0)}
{BR(\Upsilon(10860)\to\Upsilon(1S)\pi^+\pi^-)}\approx6.2\cdot10^{-4}\,.$$ The most characteristic feature of the decay $\Upsilon(10860)\to
\Upsilon(1S)f_0(980)\to\Upsilon(1S)\eta\pi^0$ is the dominance of the narrow resonance structure in the $\eta\pi^0$ mass spectrum in the vicinity of the $K\bar K$ thresholds [@AS17; @AKS16]. The corresponding $\eta\pi^0$ mass spectrum (see Fig. \[Figure1\]) is given by $$\begin{aligned}
\label{Eq4}
\frac{dN(\Upsilon(5S)\to\Upsilon(1S)\eta\pi^0)}{dm}= Cp(m)\left|
\frac{\Pi_{a^0_0f_0}(m)}{D_{a^0_0}(m)D_{f_0}(m)-\Pi^2_{a^0_0f_0}
(m)}\right|^2\frac{2m^2\Gamma_{a^0_0\to\eta\pi^0}(m)}{\pi},\end{aligned}$$ where $\Upsilon(5S)$ is a short notation for $\Upsilon(10860)$, $m$ is the invariant mass of the $\eta\pi^0$ system, $p(m)=\sqrt{m^4_{\Upsilon(5S)}-2m^2_{\Upsilon(5S)}(m^2+m^2_{\Upsilon(1S)
})+(m^2-m^2_{\Upsilon(1S)})^2}\,/(2m_{\Upsilon(5S)})$, and $C$ is the normalization constant.
=7.cm
The $a^0_0(980)-f_0(980)$ mixing amplitude $\Pi_{a^0_0f_0}(m)$ in Eq. (\[Eq4\]), caused by the diagrams shown in Fig. \[Figure2\], has the form
![\[Figure2\] The $K\bar K$ loop mechanism of the $a^0_0(980)-f_0(980)$ mixing.](f0-a0-mixing.eps){width="20pc"}
$$\begin{aligned}
\label{Eq5} %\hspace*{-35pt} &&
&& \Pi_{a^0_0f_0}(m)=\frac{g_{a^0_0K^+K^-}g_{f_0K^+K^-
}}{16\pi}\Biggl[\,i\,\Bigl(\rho_{K^+K^-}(m)%\nonumber\\ \hspace*{-35pt} &&
-\rho_{K^0\bar K^0}(m)\Bigr)\nonumber\\ && -
\frac{\rho_{K^+K^-}(m)}{\pi}\ln\frac{1+\rho_{K^+K^-}(m)}
{1-\rho_{K^+K^-}(m)}%\nonumber\\ \hspace*{-35pt} &&
+\frac{\rho_{K^0 \bar K^0}(m)}{\pi}\ln\frac{1+\rho_{K^0\bar
K^0}(m)}{1-\rho_{K^0\bar K^0}(m)}\,\,\Biggl]\,,\end{aligned}$$
where $m\geq2m_{K^0}$ and $\rho_{K\bar K}(m)=\sqrt{1-4m_K^2/m^2}$; in the region $0\leq m\leq2m_{K^0}$, $\rho_{K^0\bar K^0}(m)$ should be replaced by $i|\rho_{K^0\bar K^0}(m)|$, and in the region $0\leq
m\leq2m_{K^+}$, $\rho_{K^+K^-} (m)$ should be replaced by $i|\rho_{K^+K^-}(m)|$. $D_r(m)$, in Eq. (\[Eq4\]), is the inverse propagator of the unmixed resonance $r$ $[r=a^0_0(980),f_0(980)]$ with the mass $m_r$, $$\begin{aligned}
\label{Eq6}
D_r(m)=m^2_r-m^2+\sum_{ab}[\mbox{Re}\Pi^{ab}_r(m_r)-\Pi^{ab}_r(m)],\end{aligned}$$ $ab=(\eta\pi^0,\,K^+K^-,\,K^0\bar K^0,\,\eta'\pi^0)$ for $r=a^0_0(980)$ and $ab=(\pi^+\pi^-,\,\pi^0\pi^0,\,K^+K^-,\,K^0
\bar K^0,\,\eta\eta)$ for $r=f_0(980)$; $\Pi^{ab}_r(m)$ stands for the diagonal matrix element of the polarization operator of the resonance $r$ corresponding to the contribution of the $ab$ intermediate state. At $m>m_a+m_b$, $$\begin{aligned}
\label{Eq7}\Pi^{ab}_{r}(m)=\frac{g^2_{r
ab}}{16\pi} \left[\frac{m_{ab}^{(+)}m_{ab}^{(-)}}{\pi
m^2}\ln\frac{m_b}{m_a}+\rho_{ab}(m)%\right.\ \nonumber\\ \left.\times
\left(i-\frac{1}{\pi}\,\ln\frac{\sqrt{m^2-m_{ab}^{(-)
\,2}}+\sqrt{m^2-m_{ab}^{(+)\,2}}}{\sqrt{m^2-m_{ab}^{(-)\,2}}-\sqrt{m^2
-m_{ab}^{(+)\,2}}}\right)\right],\end{aligned}$$ where $g_{rab}$ is the coupling constant of $r$ with $ab$, $\rho_{ab}(m)$=$
\sqrt{m^2-m_{ab}^{(+)\,2}} \,\sqrt{m^2-m_{ab}^{(-)\,2}}\,/m^2$, $m_{ab}^{(\pm)}$=$m_a\pm m_b$, and $m_a\geq m_b$; $\mbox{Im}
\,\Pi^{ab}_r(m)=m\Gamma_{r\to ab}(m)=(g^2_{r
ab}/16\pi)\rho_{ab}(m)$. At $m_{ab}^{(-)}<m<m_{ab}^{(+)}$ $$\begin{aligned}
\label{Eq8}\Pi^{ab}_{r}(m)=\frac{g^2_{r
ab}}{16\pi} \left[\frac{m_{ab}^{(+)}m_{ab}^{(-)}}{\pi
m^2}\ln\frac{m_b}{m_a}%\right.\nonumber\\ \left.
-\rho_{ab}(m)\left(1-\frac{2}{\pi}\arctan\frac{\sqrt{
m_{ab}^{(+)\,2}-m^2}}{\sqrt{m^2-m_{ab}^{(-)\,2}}}\right)\right],\end{aligned}$$ where $\rho_{ab}(m)$=$\sqrt{m_{ab}^{(+)\,2}-m^2}
\,\sqrt{m^2-m_{ab}^{(-)\,2}}\,/m^2$. At $m\leq m_{ab}^{(-)}$ $$\begin{aligned}
\label{Eq9}\Pi^{ab}_{r}(m)=\frac{g^2_{r
ab}}{16\pi} \left[\frac{m_{ab}^{(+)}m_{ab}^{(-)}}{\pi
m^2}\ln\frac{m_b}{m_a}%\right.\ \ \nonumber\\ \left.
+\rho_{ab}(m)\frac{1}{\pi}\,\ln\frac{
\sqrt{m_{ab}^{(+)\,2}-m^2}+\sqrt{m_{ab}^{(-)\,2}-m^2}}
{\sqrt{m_{ab}^{(+)\,2}-m^2}-\sqrt{m_{ab}^{(-)\,2}-m^2}}\right],\end{aligned}$$ where $\rho_{ab}(m)$=$\sqrt{m_{ab}^{(+)\,2}-m^2}\,\sqrt{
m_{ab}^{(-)\,2}-m^2}\,/m^2$. Here we use as a guide the numerical estimates of the coupling constants $g^2_{f_0 ab}/(16\pi)$ and $g^2_{a^0_0ab}/(16\pi)$ obtained in Ref. [@AKS16] by analyzing the BESIII data on the $a^0_0(980)-f_0(980)$ mixing [@Ab11] $$\begin{aligned}
\label{Eq10}
\frac{g^2_{f_0\pi\pi}}{16\pi}\equiv\frac{3}{2}\frac{g^2_{f_0\pi^+\pi^-}}
{16\pi}=0.098\mbox{\ GeV}^2,\\ \label{Eq11} \frac{g^2_{f_0 K\bar
K}}{16\pi}\equiv2\frac{g^2_{f_0 K^+K^-}}{16\pi}=0.4\mbox{\ GeV}^2,
\\ \label{Eq12} \frac{g^2_{a^0_0\eta\pi^0}}{16 \pi}=0.2\mbox{\ GeV}^2,
\quad\quad\ \\ \label{Eq13} \frac{g^2_{a^0_0 K\bar K}}{16\pi}
\equiv2\frac{g^2_{a^0_0 K^+K^-}}{16\pi}=0.5\mbox{\ GeV}^2.\end{aligned}$$ As in Ref. [@AKS16], we fix $m_{a^0_0}=0.985$ GeV, $m_{f_0}=
0.985 $ GeV (see also Ref. [@Fn1]) and set $g^2_{a^0_0\eta'\pi^0}=g^2_{a^0_0\eta\pi^0}$ and $g^2_{f_0\eta\eta}=g^2_{f_0 K^+K^-}$ by the $q^2\bar q^2$ model.
In the Belle experiment of 2015 [@Ga15], $2090\pm115$ $\Upsilon(10860)\to\Upsilon(1S)\pi^+\pi^-$ events were collected. Thus, one or two $\Upsilon (10860)\to\Upsilon(1S)\eta\pi^0$ events could be produced in this experiment due to the $a_0(980)-f_0(980)$ mixing.
There is no visible $\eta\pi^0$ background and one can hope that this rare decay will be measured with the use of the Belle II detector at the SuperKEKB with at least 10% statistical accuracy. In the new experiment with a 40 times greater instantaneous luminosity, one can collect about a hundred $\Upsilon (10860)\to
\Upsilon(1S)\eta\pi^0$ events (i.e., from 50 to 150 events) in the narrow region of the $\eta\pi^0 $ invariant mass near the $K\bar K$ thresholds in the time comparable with that spent on the initial Belle experiment [@Ga15]. According to our estimates, the statistics in the experiment [@Ga15] were collected in about three months.
The observation of the decay $\Upsilon (10860)\to\Upsilon(1S)f_0
(980)\to\Upsilon(1S)\eta\pi^0$, together with investigations of the weak hadronic decays $D_s^+\to\eta\pi^0\pi^+$ [@AS17a], $D^0\to
K_S\pi^+\pi^-$ and $D^0\to K_S\eta\pi^0$ [@AS17b], can open a new stage in a thorough study of the $a^0_0(980)-f_0(980)$ mixing phenomenon and the nature of the light scalar mesons.
We are grateful to A.E. Bondar, who drew our attention to the decay under consideration. The present work is partially supported by the Russian Foundation for Basic Research Grant No. 16-02-00065 and the Presidium of the Russian Academy of Sciences Project Grant No. 0314-2015-0011.
[99]{} A. Garmash [*et al.*]{} (Belle Collaboration), Phys. Rev. D [**91**]{}, 072003 (2015). N. N. Achasov, S. A. Devyanin, and G. N. Shestakov, Phys. Lett. [**88B**]{}, 367 (1979). N. N. Achasov and G. N. Shestakov, Nucl. Part. Phys. Proc. [**287–288**]{}, 89 (2017). M. Ablikim [*et al.*]{} (BESIII Collaboration), Phys. Rev. D [**83**]{}, 032003 (2011). N. N. Achasov, A. A. Kozhevnikov, and G. N. Shestakov, Phys. Rev. D [**93**]{}, 114027 (2016). The Belle Collaboration [@Ga15] used the Flatté propagator with the $f_0(980)$ mass $M=0.950$ GeV obtained by A. Garmash [*et al.*]{} (Belle Collaboration), Phys. Rev. Lett. [**96**]{}, 251803 (2006). At first glance it is a misleading result. In fact, we previously noted many times that the Flatté propagator is not adequate one for the $f_0(980)$ and $a_0(980)$ study. See N.N. Achasov and V.V. Gubin, Phys. Lett. B [**363**]{}, 106 (1995); N.N. Achasov and V.V. Gubin, Phys. Rev. D [**56**]{}, 4084 (1997); N.N. Achasov, V.V. Gubin, and V.I. Shevchenko, Phys. Rev. D [**56**]{}, 203 (1997); N.N. Achasov and A.V. Kiselev, Phys. Rev. D [**70**]{}, 111901(R) (2004). The mass renormalization in the Flatté propagator is required, which results in $m_{f_0(980)}=0.979$ GeV. Our propagators, which first appeared in Ref. \[2\], have no such problem. Consequently, we once again recommend their use. N. N. Achasov and G. N. Shestakov, Phys. Rev. D [**96**]{}, 036013 (2017). N. N. Achasov and G. N. Shestakov, Phys. Rev. D [**96**]{}, 016027 (2017).
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We study the Josephson effect through a magnetic molecule with anisotropic properties. Performing calculations in the tunneling regime, we show that the exchange coupling between the electron spin on the molecule and the molecular spin can trigger a transition from the $\pi$ state to the $0$ state, and we study how the spin anisotropy affects this transition. We show that the behavior of the critical current as a function of an external magnetic field can give access to valuable information about the spin anisotropy of the molecule.'
author:
- 'I. A. Sadovskyy,$^{1}$ D. Chevallier,$^{2,3}$ T. Jonckheere,$^{2}$ M. Lee,$^{4}$ S. Kawabata,$^{5}$, and T. Martin$^{2,3}$'
title: Josephson effect through an anisotropic magnetic molecule
---
Introduction {#sec:introduction}
============
The Josephson effect[@Josephson:1962; @Gennes:1964] is a striking manifestation of many body physics and macroscopic quantum coherence in condensed matter systems. While early investigations concerned mainly bulk superconducting junctions separated by an insulating barrier, in the last decades it has become a very active field of study in the context of mesoscopic physics. Indeed the insulating barrier can be replaced by a conductor or a nano-device that can be as small as a quantum dot or a single molecule. In this sense the study of the Josephson current can provide a novel way to investigate the electronic properties of the nano-object, which is sandwiched between the superconducting electrodes. More than a decade ago, it was predicted using the Krein theorem[@Spivak:1991; @Krichevsky:2000; @Benjamin:2007] that when a singly occupied quantum dot in the Coulomb blockade regime is inserted between the superconductors, the Josephson current phase relation acquires a $\pi$ shift, i.e., the critical current has the opposite sign from that of a tunnel junction. A phase diagram of the $\pi$–$0$ phase transition was derived later on for contacts with arbitrary transparency using a combination of Hubbard-Stratonovich and saddle-point approximation.[@Rozhkov:2000] Experimentally (for nanoscale devices) it was measured in superconductor–nanotube–superconductor systems.[@Dam:2006] This picture gets more complicated when the Kondo temperature is lower than the superconducting gap: a $0$-junction state is restored,[@Siano:2004; @Choi:2004] albeit with a different current phase relationship.
In recent years theoretical and experimental studies have addressed transport geometries where a molecule — artificial or otherwise — is inserted between two electrodes.[@Kasumov:1999; @JarilloHerrero:2006; @Cleuziou:2006; @Eichler:2009; @Roch:2008; @Winkelmann:2009] This goes one step beyond the study of transport through quantum dots because the molecule has internal degrees of freedom (such as vibrations and possibly spin). On the one hand, such degrees of freedom have an effect on the electronic current, on the other hand, the current itself can be considered as a probe of the inherent mechanisms of the molecule.
A subfield of molecular electronics is called molecular spintronics: it focuses on molecules which have an intrinsic spin,[@Sessoli:2003; @Petukhov:2004] and it is expected that electron transfer through the molecule can trigger changes in the molecule spin because of the existence of an exchange coupling with the electron spin. Such molecules (such as a buckminsterfullerene doped with a magnetic atom) may have an isotropic spin, or otherwise the spin may have a preferred direction due to the crystalline structure of the molecule (this is the case of Mn$_{12}$ acetate). Recently, there have been some efforts to describe and measure transport through molecular spintronics devices with normal metal or ferromagnetic leads,[@Romeike:2006; @Heersche:2006a; @Grose:2008; @Roch:2011] with an emphasis on master equations approach on the theoretical side. Nevertheless, efforts in the field of molecular spintronics with superconducting electrodes are still at their beginning stage.
![ (Color online) A magnetic molecule (e.g., M@C$_{80}$, Mn$_{12}$, $\ldots$) connecting two superconductors via tunnel barriers $t_{\rm\scriptscriptstyle L}$ and $t_{\rm\scriptscriptstyle R}$. The exchange coupling between the molecular spin and the electronic spin can strongly modify the Josephson current. []{data-label="fig:setup"}](fig1.eps){width="4.5cm"}
A recent theoretical work focused on the Josephson current through an isotropic magnetic molecule, via perturbative calculations in the tunneling Hamiltonian as well as numerical renormalization group calculations.[@Lee:2008a] It allowed us to draw a complete phase diagram of the $\pi$–$0$ phase transition. An equivalent study of supercurrent through molecules which have an anisotropic spin, which magnetization can tunnel, and which are subject to a (weak) external magnetic field is still lacking. This is precisely the focus of the present work. One of the challenges of this work is that we have to deal with a large number of parameters: the exchange coupling $J$ between the dot electron spin and the molecule spin, the anisotropy constant $D$, and the coefficient $B_2$ for quantum tunneling of magnetization of the molecule, the dependence on external magnetic field $B$ as well as the dot level $\epsilon_{\rm d}$, which can be adjusted by a gate voltage. Note that it is now experimentally possible to manipulate the anisotropy parameters of magnetic molecules.[@Parks:2010; @Zyazin:2010] One of our goals is to determine to what extent the measurement of the critical current can provide information of the sign or magnitude of such parameters. For simplicity, we focus on the regime where the superconducting gap is much larger than the Kondo temperature, which allows us to focus on weak coupling (small tunneling Hamiltonian) calculations. Also, we restrict the analysis on the simplest case of a molecule spin $S=1$ to demonstrate the effect where the two main contributions due to spin anisotropy (easy axis anisotropy and quantum tunneling of magnetization) are present.
The outline of the paper is as follows. In Sec. \[sec:model\], we introduce the model for the magnetic molecule connected to two superconducting leads and we compute the expression of the Josephson current through this molecule. In Sec. \[sec:results\], we study the effect of the anisotropic parameters and of the adjustable experimental parameters on the sign of the critical current. Finally, we conclude in Sec. \[sec:conclusion\].
Model {#sec:model}
=====
Hamiltonian
-----------
The total Hamiltonian of the system (see Fig. \[fig:setup\]) consists of the three terms $\mathcal{\hat H} = \mathcal{\hat H}_{\rm d} + \mathcal{\hat H}_{\rm s} + \mathcal{\hat H}_{\rm t}$. The first one is the Hamiltonian of the molecule, $$\mathcal{\hat H}_{\rm d} =
\mathcal{\hat H}_{\rm m}
+ \epsilon_{\rm d} \sum\limits_\sigma {\hat d}_\sigma^\dag {\hat d}_\sigma^{\phantom\dag}
+ U {\hat d}_\uparrow^\dag {\hat d}_\uparrow^{\phantom\dag} {\hat d}_\downarrow^\dag {\hat d}_\downarrow^{\phantom\dag},
\label{eq:Hd}$$ where $\epsilon_{\rm d}$ is the electronic level of the molecule implied in the transport, and $U$ is the Coulomb interaction strength. The ${\hat d}_\sigma^\dag$ and ${\hat d}_\sigma^{\phantom\dag}$ are electronic creation and annihilation operators on the electronic level in the molecule. Since $U$ is typically much larger than the other energies in the system, we consider the limit of infinite Coulomb interaction $U\rightarrow\infty$, thus only one electron is allowed to occupy the dot. With this assumption, the Hamiltonian $\mathcal{\hat H}_{\rm m}$, which characterizes the magnetic properties of the dot, reads $$\mathcal{\hat H}_{\rm m} =
-D {\hat S}_z^2 + B ({\hat S}_z + {\hat s}_z)
- \frac{B_2}{2} ({\hat S}_+^2 + {\hat S}_-^2) +
J {\hat{\bf S}}{\hat{\bf s}},
\label{eq:Hm}$$ where $S_z$ is the molecular spin and $s_z$ the spin of the electron on the molecule (if present). $J$ is the exchange coupling between molecular and electronic spin, $D>0$ is the easy axis anisotropy constant, $B_2$ is the coefficient of quantum tunneling of magnetization (QTM), and $B$ is the external magnetic field. Figure \[fig:spin\_states\] shows how these terms couple the states of the molecule in the case of a spin $S=1$. In order to avoid a too large number of parameters, we have made some simplifying assumptions when writing this Hamiltonian: the anisotropy terms are not affected by the charge of the dot level (this should be the case for systems like M@C$_{80}$, but not for molecules like Mn$_{12}$),[@Heersche:2006b] the magnetic field is taken parallel to the spin anisotropy,[@Timm:2007] and higher order terms $(-B_{2n}/2) ({\hat S}_+^{2n} + {\hat S}_-^{2n})$ are neglected (they are usually small).
![ (Color online) Spin states for $S=1$ spin, and coupling between these states due to the different terms of the Hamiltonian. $B_2$ induces tunneling between $|1\rangle_{\rm m}$ and $|-1\rangle_{\rm m}$ states; $J$ and $B$ induce $|1\rangle_{\rm m} \leftrightarrow |0\rangle_{\rm m}$ and $|0\rangle_{\rm m} \leftrightarrow |-1\rangle_{\rm m}$ tunnelings. []{data-label="fig:spin_states"}](fig2.eps){width="4.2cm"}
The second term corresponds to the superconducting parts, described by the BCS Hamiltonian $$\mathcal{\hat H}_{\rm s} =
\sum\limits_{\ell, k, \sigma} \epsilon_k
{\hat c}_{\ell, k, \sigma}^\dag {\hat c}_{\ell, k, \sigma}^{\phantom\dag}
- \sum\limits_{\ell, k} \Delta_{\ell,k}
\Bigl[ {\hat c}_{\ell, k, \uparrow}^\dag {\hat c}_{\ell, -k, \downarrow}^\dag + {\rm H.c.} \Bigr],
\label{eq:sH}$$ where $\epsilon_k = {\hbar}^2 k^2/2m - E_{\rm\scriptscriptstyle F}$ is the dispersion relation for free electrons, ${\hat c}_{\ell, k, \sigma}^\dag$ and ${\hat c}_{\ell, k, \sigma}^{\phantom\dag}$ are electronic creation and annihilation operators in the superconductors, $\ell$ enumerates left ($\ell = {\rm L}$) and right ($\ell = {\rm R}$) leads, $\Delta_{{\rm\scriptscriptstyle L(R)},k} = \Delta e^{\pm i\varphi/2}$, with $\Delta$ the superconducting gap and $\varphi$ the superconducting phase difference along the junction.
The last term is the tunnel Hamiltonian between the leads and the molecule $$\mathcal{\hat H}_{\rm t} =
\sum\limits_{\ell, k, \sigma}
\Bigl[ t_{\ell, k} {\hat d}_\sigma^\dag {\hat c}_{\ell, k, \sigma}^{\phantom\dag} + {\rm H.c.} \Bigr],
\label{eq:tH}$$ where $t_{\ell, k}$ are the tunneling amplitudes. By performing a gauge transformation for ${\tilde t}_{{\rm\scriptscriptstyle L},k} = t_{{\rm\scriptscriptstyle L},k} e^{i\varphi/4}$, ${\tilde t}_{{\rm\scriptscriptstyle R},k} = t_{{\rm\scriptscriptstyle R},k} e^{-i\varphi/4}$ and simultaneously for ${\tilde {\hat c}}_{{\rm\scriptscriptstyle L}, k, \sigma} = {\tilde {\hat c}}_{{\rm\scriptscriptstyle L}, k, \sigma} e^{-i\varphi/4}$, ${\tilde {\hat c}}_{{\rm\scriptscriptstyle R}, k, \sigma} = {\tilde {\hat c}}_{{\rm\scriptscriptstyle R}, k, \sigma} e^{i\varphi/4}$, one can “move” the dependence on $\varphi$ from $\Delta_{\ell,k}$ to $t_{\ell, k}$ and ${\hat c}_{\ell, k, \sigma}$ in Eqs. (\[eq:sH\]) and (\[eq:tH\]).[@Spivak:1991] We also perform a Bogoliubov transformation[@Gennes:1964] to diagonalize the BCS Hamiltonian, which takes the following form: $$\mathcal{\hat H}_{\rm s} =
\sum\limits_{\ell, k, \sigma} E_k
{\hat\gamma}_{\ell, k, \sigma}^\dag {\hat\gamma}_{\ell, k, \sigma}^{\phantom\dag}
\label{eq:sHbt}$$ and the tunneling Hamiltonian reads $$\mathcal{\hat H}_{\rm t} =
\sum\limits_{\ell, k, \sigma}
\Bigl[
{\tilde t}_{\ell, k} {\hat d}_\sigma^\dag \big(
u_k {\hat\gamma}_{\ell, k, \sigma}^{\phantom\dag} +
\sigma \, v_k {\hat\gamma}_{\ell, k, -\sigma}^{\dag}
\big) +
{\rm H.c.}
\Bigr],
\label{eq:tHbt}$$ where ${\hat\gamma}_{\ell, k, \sigma}^{\dag}$ and ${\hat\gamma}_{\ell, k, \sigma}$ are the quasiparticle creation and annihilation operators, $u_k = \sqrt{(1+\epsilon_k/E_k)/2}$ and $v_k = \sqrt{(1-\epsilon_k/E_k)/2}$ are the electron and hole coefficients, and $E_k = \sqrt{\epsilon_k^2 + \Delta^2}$ is the energy dispersion. In the following calculations we will consider for simplicity the case of symmetric contacts, thus $t_{{\rm\scriptscriptstyle L},k} = t_{{\rm\scriptscriptstyle R},k} = t_k$.
Specific Hamiltonian for the $S=1$ case
---------------------------------------
In the following, we will for simplicity restrict our calculations to the case of a molecular spin with $S=1$, which is the smallest value where easy axis anisotropy (parameter $D$) and QTM (parameter $B_2$) are nontrivial. As the electron occupation of the level is restricted to $0$ or $1$, let us write explicitly the molecule Hamiltonian in each case \[see Eqs. (\[eq:Hd\]) and (\[eq:Hm\])\].
For the empty electronic level, we have $\mathcal{H}_{\rm d}=\mathcal{H}_{{\rm m},0}$, and we use the basis $\{ |0\rangle_{\rm e} |1\rangle_{\rm m}, \; |0\rangle_{\rm e} |0\rangle_{\rm m}, \; |0\rangle_{\rm e} |-1\rangle_{\rm m} \}$, where $|0\rangle_{\rm e}$ represents the empty electronic state and $|S_z\rangle_{\rm m}$ the states of the molecule with spin projections $S_z = 1, 0, -1$. The matrix elements of $\mathcal{H}_{\rm m}$ are $$\mathcal{H}_{\rm m,0} =
\left[ \!
\begin{array}{ccc}
B-D & 0 & -B_2 \\
0 & 0 & 0 \\
-B_2 & 0 & -B-D \\
\end{array}
\! \right] \!.
\label{eq:uH_me}$$ The eigenvalues are noted $E_{0,i}$ ($i=1,2,3$), and the corresponding eigenvectors are ${\bf b}_i$. Below we will use the matrix $b_{ij} = [{\bf b}_1, \; {\bf b}_2, \; {\bf b}_3]$, which consists of columns of eigenvectors (first index enumerates columns, the second enumerates rows), and the inverse matrix ${\tilde b}_{ij} = (b_{ij})^{-1}$.
When the electronic level is occupied by one electron, we have $\mathcal{H}_{{\rm d},1} = \mathcal{H}_{{\rm m},1} + \epsilon_{\rm d}$, and we use the uncoupled spin basis ${ |s\rangle_{\rm e} |S_z\rangle_{\rm m} }$ (with $s=\uparrow,\downarrow$ and $S_z=+1,0,-1$). The matrix representation of $\mathcal{H}_{{\rm m},1}$ can then be decomposed as two independent $3\times 3$ submatrices:[@Huertas-Hernando:2006] $\mathcal{H}_{\rm m,1} = {\rm diag} \{\mathcal{H}_{\rm m,1}^+, \mathcal{H}_{\rm m,1}^- \}$, with $$\mathcal{H}_{\rm m,1}^+ =
\left[ \!
\begin{array}{ccc}
3B/2+J/2-D & 0 & -B_2 \\
0 & -B/2 & J/\sqrt{2} \\
-B_2 & J/\sqrt{2} & -B/2-J/2-D \\
\end{array}
\! \right]
\label{eq:oHp_me}$$ in the basis $\{ |\!\uparrow\rangle_{\rm e} |1\rangle_{\rm m}, \; |\!\downarrow\rangle_{\rm e} |0\rangle_{\rm m}, \; |\!\uparrow\rangle_{\rm e} |-1\rangle_{\rm m} \}$ and $$\mathcal{H}_{\rm m,1}^- =
\left[ \!
\begin{array}{ccc}
B/2-J/2-D & J/\sqrt{2} & -B_2 \\
J/\sqrt{2} & B/2 & 0 \\
-B_2 & 0 & -3B/2+J/2-D \\
\end{array}
\! \right]
\label{eq:oHm_me}$$ in the basis $\{ |\!\downarrow\rangle_{\rm e} |1\rangle_{\rm m}, \; |\!\uparrow\rangle_{\rm e} |0\rangle_{\rm m}, \; |\!\downarrow\rangle_{\rm e} |-1\rangle_{\rm m} \}$. These matrices have eigenvalues $E_{1,i}^+$, $E_{1,i}^-$ and corresponding eigenvectors ${\bf a}_{i}^+$, ${\bf a}_{i}^-$. As previously, we define matrices $a_{ij}^{\pm} = [{\bf a}_1^{\pm}, \; {\bf a}_2^{\pm}, \; {\bf a}_3^{\pm}]$, and inverse matrices ${\tilde a}_{ij}^{\pm} = (a_{ij}^{\pm})^{-1}$.
Josephson current
-----------------
The Josephson current through the molecule can be calculated using perturbation theory in the tunneling Hamiltonian $\mathcal{\hat H}_{\rm t}$;[@Spivak:1991] the first nonvanishing term is given by $$\begin{gathered}
I = \frac{2e}{\hbar} \frac{\partial}{\partial\varphi}
\bigl\langle {\rm gs} \big|
\mathcal{\hat H}_{\rm t}
(E_{\rm gs} - \mathcal{\hat H}_0)^{-1}
\mathcal{\hat H}_{\rm t}
(E_{\rm gs} - \mathcal{\hat H}_0)^{-1} \\
\times \mathcal{\hat H}_{\rm t}
(E_{\rm gs} - \mathcal{\hat H}_0)^{-1}
\mathcal{\hat H}_{\rm t}
\bigr|{\rm gs} \big\rangle,
\label{eq:I_pert}\end{gathered}$$ where $\mathcal{\hat H}_0 = \mathcal{\hat H}_{\rm d} + \mathcal{\hat H}_{\rm s}$. The ground state $|{\rm gs}\rangle$ is the occupied state with lowest energy, thus it has energy $E_{\rm gs} = {\rm min} \{ E_{1,i}^\pm \}$, and $|{\rm gs}\rangle = |{\bf a}_i^\zeta\rangle$, where $i = 1, 2, 3$ specifies the state number and $\zeta=\pm$ is the block index. Note that the dot-lead coupling induces energy shifts for the occupied states of the dot, starting at order 2 in $\mathcal{\hat H}_{\rm t}$. However, we do not need to compute these shifts, as they will be identical for the two single occupied states, and they can be included in the value of $\epsilon_{\rm d}$ (see Ref. for a multilevel case where these shifts have to be computed).
As was shown in Ref. , in the absence of coupling to a molecular spin, the perturbative approach allows us to understand the $\pi$ state due to large Coulomb interaction on the dot: the order of the electrons of a Cooper pair is necessarily reversed during tunneling through the dot, which gives opposite sign for the current due to the singlet nature of the Cooper pair. Here, the exchange coupling between the electron spin and a molecular spin means that the occupied state of the dot is a linear combination of states involving in general both $|\!\uparrow\rangle$ and $|\!\downarrow\rangle$ states of the electron spin. This creates the possibility of spin-flip processes: a spin-up electron tunneling in the dot can tunnel out as a spin-down electron for example. With such a spin flip, it is now possible for a Cooper pair to tunnel through the dot without reversing the order of electrons, thus contributing to positive current. In the presence of exchange coupling with a molecular spin, one can thus expect that, among all the lowest-order processes contributing to the Josephson current, some of them will contribute to negative current, and some others to positive current. The global sign of the current will thus depend on the relative weight of the different processes, which are a function of the parameters of the molecule Hamiltonian.
Expressing in Eq. (\[eq:I\_pert\]) the action of the tunneling Hamiltonian on the eigenstates introduced in the previous section, a lengthy but straightforward calculation gives eventually $$\begin{aligned}
I = -\frac{4e}{\hbar} \sin\varphi
\sum\limits_{k, k'} &
t_{{\rm\scriptscriptstyle L}, k}^2 t_{{\rm\scriptscriptstyle R}, k'}^2
u_k v_k u_{k'} v_{k'} \nonumber \\
\times \sum_j \biggl\{
& \frac{
A_{j,k'}^{\zeta*} B_{j,k}^\zeta +
B_{j,k'}^{\zeta*} A_{j,k}^\zeta
}{
E_k + E_{k'} + E_{1,j}^\zeta - E_{\rm gs}
} \nonumber \\
+ & \frac{
A_{j,k'}^{{\bar\zeta}*} A_{j,k}^{\bar\zeta} +
B_{j,k'}^{{\bar\zeta}*} B_{j,k}^{\bar\zeta}
}{
E_k + E_{k'} + E_{1,j}^{\bar\zeta} - E_{\rm gs}
}
\biggr\}.
\label{eq:I_fin}\end{aligned}$$ Here $$\begin{aligned}
A_{j,k}^{\pm} = &
\pm \frac{
({\tilde a}_{{\rm gs},1} b_{11} + {\tilde a}_{{\rm gs},3} b_{31})
({\tilde b}_{11} a_{1j}^\pm + {\tilde b}_{13} a_{3j}^\pm)
}{
E_k + E_{0,1} - E_{\rm gs} - \epsilon_{\rm d}
} \nonumber \\
& \pm \frac{
({\tilde a}_{{\rm gs},1} b_{13} + {\tilde a}_{{\rm gs},3} b_{33})
({\tilde b}_{31} a_{1j}^\pm + {\tilde b}_{33} a_{3j}^\pm)
}{
E_k + E_{0,3} - E_{\rm gs} - \epsilon_{\rm d}
},
\label{A} \\
B_{j,k}^{\pm} = &
\mp \frac{
{\tilde a}_{{\rm gs},2} a_{2j}^\pm
}{
E_k + E_{0,2} - E_{\rm gs} - \epsilon_{\rm d}
},
\label{eq:B}\end{aligned}$$ where $A_{j,k}^{\pm*} \equiv (A_{j,k}^\pm)^*$ and ${\bar\zeta} = -\zeta$. Equations (\[eq:I\_fin\])–(\[eq:B\]) represent the main results of this paper. Because we have performed a lowest-order tunneling calculation, we get a simple $I = I_{\rm c} \sin\varphi$ dependence of the current. However, the study of value of the critical current $I_{\rm c}$ (in addition to its sign) will give us precious information on the system. At zero temperature the sums over $k$ and $k'$ should be taken over the energy region $\epsilon_k, \epsilon_{k'} > 0$. Both summations over $k$ can be replaced by the integration over energy $\epsilon$: $\sum_k \to \int d\epsilon \rho(\epsilon)$, where $\rho(\epsilon)$ is a density of states.
Our formulas of course contain the known result for the case where there is no molecular spin and no magnetic field ($B = B_2 = J = D = 0$):[@Spivak:1991:comment] we obtain a $\pi$-junction with negative critical current, $$\begin{aligned}
I_{\rm c}^{(0)}
& = -\frac{4e}{\hbar}
\sum\limits_{k, k'}
\frac{
t_{{\rm\scriptscriptstyle L}, k}^2 t_{{\rm\scriptscriptstyle R}, k'}^2
u_k v_k u_{k'} v_{k'}
}{
(E_k + E_{k'}) (E_k - \epsilon_{\rm d}) (E_{k'} - \epsilon_{\rm d})
}
\nonumber \\
& \!\!\!\! = -\frac{4e}{\hbar}
\frac{\Gamma_{\rm\scriptscriptstyle L}\Gamma_{\rm\scriptscriptstyle R} \Delta^2}{4\pi^2} \!
\int\limits_0^\infty \!
\frac{d\epsilon_1 d\epsilon_2}{
E_1 E_2 (E_1 + E_2) (E_1 - \epsilon_{\rm d}) (E_2 - \epsilon_{\rm d})
},
\label{eq:I_0}\end{aligned}$$ where we assume constant density of states $\rho(\epsilon) = \rho_0 = 2m/\pi\hbar^2$, tunneling rates $\Gamma_{\rm\scriptscriptstyle L(R)} = \pi\rho_0 t_{\rm\scriptscriptstyle L(R)}^2$, and $E_{1(2)} = \sqrt{\Delta^2 + \epsilon_{1(2)}^2}$.
![ The dependence of the normalized critical current $i_{\rm c}$ as a function of the exchange coupling $J$ for an isotropic magnetic molecule ($D = B_2 = 0)$ and dot level $\epsilon_{\rm d}/\Delta = -5$ in the absence of magnetic field ($B = 0$). []{data-label="fig:I_J"}](fig3.eps){width="\linewidth"}
In the next section we analyze the dependence of the dimensionless critical current $i_{\rm c} = I_{\rm c} / |I_{\rm c}^{(0)}|$ on the magnetic molecule parameters $J$, $D$, $B_2$, dot energy $\epsilon_{\rm d}$, and external magnetic field $B$. Positive $i_{\rm c}>0$ corresponds to the $0$-junction phase, negative $i_{\rm c}<0$ corresponds to the $\pi$-junction phase.
Results and discussion {#sec:results}
======================
For reference, we start by analyzing Eq. (\[eq:I\_fin\]) as a function of exchange coupling $J$, when no anisotropy is present ($D = B_2 = 0$) and without magnetic field ($B=0$). As shown in Fig. \[fig:I\_J\](a), the current is suppressed both by negative and positive $J$. For negative $J$ (ferromagnetic coupling) the system always remains in the $\pi$ state ($i_{\rm c} < 0$). For positive $J$ (antiferromagnetic coupling) a $\pi$–$0$ transition occurs for $J/\Delta \sim 10$ (the precise value is slowly varying with $\epsilon_{\rm d}$). This behavior can be understood by looking at the formula for the current,[@Lee:2008a] $$\begin{aligned}
I = -\frac{4e}{\hbar} \sin\varphi
\sum\limits_{k, k'} &
t_{{\rm\scriptscriptstyle L}, k}^2 t_{{\rm\scriptscriptstyle R}, k'}^2
u_k v_k u_{k'} v_{k'} \nonumber \\
\times \frac{1}{3 \mathcal{E}_k \mathcal{E}_{k'}} \biggl\{
& \frac{4}{3J/2+E_k+E_{k'}} -
\frac{1}{E_k+E_{k'}}
\biggr\},
\label{eq:I_J}\end{aligned}$$ where $\mathcal{E}_k = E_k + J - \epsilon_{\rm d}$. The first term depicts the transfer of a Cooper pair involving a change of the total coupled spin (electronic and molecule) during the intermediate state \[e.g., see Fig. \[fig:spin\_flips\](a)\], while the second term corresponds to a Cooper pair without change of total spin during the intermediate state \[e.g., see Fig. \[fig:spin\_flips\](b)\]. For large positive $J$, the first term becomes smaller than the second one, and the sign of the current changes, which explains the $\pi$–$0$ transition.
Note that there is no change of ground state associated with this transition occurring for large positive $J$, hence the critical current shows a smooth change from negative to positive value, passing continuously through arbitrary small values. This is to be contrasted with $0$–$\pi$ transition, which is due to the crossing of energy levels leading to a change of ground state,[@Rozhkov:2000; @Novotny:2005; @Sadovskyy:2007] where an abrupt change of the critical current can be observed (see, e.g., Figs. 4 and 6 in Ref. ).
![ (Color online) Illustration of two typical tunneling processes leading to the transfer of a Cooper pair. The presence of strong Coulomb interaction prohibits the double occupation on the dot and the electrons are transferred one by one. Because of the exchange coupling $J$ between the electron spin and the molecular spin, the state of the occupied dot (black circle on the figure) is characterized by the total spin, $S-1/2$ and $S+1/2$. These two levels are separated by an energy $3J/2$. The process where the intermediate state of the occupied dot (a) is different from the initial one and the process where the intermediate state is the same as the initial one (b) contribute with different signs to the Josephson current. The competition between these two processes leads to the existence of the $\pi$–$0$ transition; see Eq. (\[eq:I\_J\]). []{data-label="fig:spin_flips"}](fig4.eps){width="\linewidth"}
We will now consider the effect of the anisotropy ($D$ and $B_2$) and of the magnetic field $B$ on the critical current, especially near the $\pi$–$0$ transition. We assume that the superconducting gap is independent of the magnetic field. Figure \[fig:J\_D\_B2\](a) shows the effect of $D$ and $B_2$ on the transition; the surface shows the values of the parameter for which the current is zero. Above the surface the system is in the $\pi$-junction phase ($i_{\rm c}<0$), while under the surface the system is in the zero phase ($i_{\rm c}>0$). One can see that both $D$ and $B_2$ move the $\pi$–$0$ transition to higher values of $J$. This is confirmed by Figs. \[fig:J\_D\_B2\](b) and \[fig:J\_D\_B2\](c), which correspond to cuts of the three-dimensional (3D) plot for fixed values of $B_2$ and $D$, respectively. On these panels, the different curves correspond to different values of the magnetic field $B$: we see that increasing the magnetic field tends to push the system toward the 0-junction phase (note that the results are insensitive to the sign of $B$). On the 3D plot Fig. \[fig:J\_D\_B2\](a) the effect of the magnetic field $B$ is thus to shift the zero current surface as shown with magenta arrows, and also to somewhat smear the sharp behavior in $B_2$ as shown with blue arrows.
![ (Color online) (a) $\pi$- and $0$-junction regions as a function of the $J$, $D$, and $B_2$. The surface divides 3D space $(J, D, B_2)$ to the top region in a $\pi$-junction regime and to the bottom one with $0$-junction regime; at the surface current is zero. Magnetic field is zero, $B=0$, its increasing leads to “shift” and “smear” of the surface as shown by magenta and blue arrows. (b) $\pi$–$0$ transition diagram in $(J, D)$ space at $B_2 = 0$. Different curves correspond to the different $B$’s: $B/\Delta = 0.0$ (solid), $0.5$ (dashed), and $1.0$ (dotted). (c) $\pi$–0 transition diagram in $(J, B_2)$ space at different $B$ and $D = 0$. The nonzero $D$ and $B_2$ increases the critical $J_{\rm c}$ (see Fig. \[fig:I\_J\]) and magnetic field mainly decreases $J_{\rm c}$. []{data-label="fig:J_D_B2"}](fig5.eps){width="0.95\linewidth"}
Up to now, we have studied the phase diagram of the system as a function of the exchange coupling $J$ and of the anisotropy parameters $D$ and $B_2$. However, for a given molecule, these parameters have usually a fixed value. We will now study the behavior of the critical current when the experimentally adjustable quantities, the external magnetic field $B$ and the dot level $\epsilon_{\rm d}$, are varied. The goal is to understand how the values of the exchange coupling and of the anisotropy parameters will modify the behavior of the current as a function of $B$ and $\epsilon_{\rm d}$. This could be an original way to obtain information on the exchange coupling and on the spin anisotropy in the molecule, by measuring the critical current of the tunnel junction and varying $B$ and $\epsilon_{\rm d}$.
{width="80.00000%"}
The different panels of Fig. \[fig:I\_B\] show the behavior of the critical current as a function of $B$ and for various values of the dot level $\epsilon_{\rm d}$, the exchange coupling $J$ and of the anisotropy parameter $D$ (for simplicity, we have taken $B_2 = 0$). Each column is for a given value of $J$: deep in the $\pi$-junction regime $J/\Delta = -20$ (left), in the intermediate regime $J/\Delta = 11$ (middle), and deep in the $0$-junction regime $J/\Delta = 20$ (right). The top panel of each column is for $D = 0$, while the two bottom panels of each column are for nonzero values of $D$ as indicated. The richer behavior is obtained when the exchange coupling has a value that allows us to observe the $\pi$–$0$ transition, here in the second column for $J/\Delta = 11$. Without anisotropy \[Fig. \[fig:I\_B\](d)\], we see that by sweeping the magnetic field we can observe the $\pi$–$0$ transition. In the presence of small anisotropy \[Fig. \[fig:I\_B\](e)\], we observe a nonmonotonic behavior as a function of $B$, with the modulus of the critical current $|i_{\rm c}|$ decreasing as a function of $B$ for small $B$, but increasing for large $B$. Finally, for larger anisotropy \[Fig. \[fig:I\_B\](f)\], $|i_{\rm c}|$ is everywhere increasing as a function of $B$. Note that, between panel Fig. \[fig:I\_B\](d) ($D = 0$) and panel Fig. \[fig:I\_B\](f) ($D/\Delta = 4$), the order of the curves as a function of $\epsilon_{\rm d}$ has been reversed. When $J$ is much larger than the superconducting gap (right column, with $J/\Delta = 20$), the system is deep in the $0$-junction phase, but the anisotropy has a visible impact on the curves: comparing Fig. \[fig:I\_B\](g) (for $D/\Delta = 0$) with Figs. \[fig:I\_B\](h) and \[fig:I\_B\](i) (for $D/\Delta = 7$ and $14$), we see that when $D$ is large enough, the slope of the critical current is the opposite of the one for small $D$. This is a consequence of the $\pi$–$0$ transition, which happens for larger $D$. Finally, for negative $J$ \[Figs. \[fig:I\_B\](a)–\[fig:I\_B\](c) with $J/\Delta = -20$\], the anisotropy does not bring any qualitative change to the behavior of the current as a function of the magnetic field, and $|i_{\rm c}|$ always decrease with $B$.
From the different curves shown in Fig. \[fig:I\_B\], we can deduce that when $J$ is positive (the antiferromagnetic coupling case), the anisotropy has a visible impact on the behavior of the critical current as a function of $B$, as it can produce a nonmonotonic behavior close to the $\pi$–$0$ transition, and reverse the slope of $|i_{\rm c}|$ as a function of $B$ when $J$ is much larger than the critical value. On the other hand, for negative $J$ (ferromagnetic coupling), the anisotropy does not have a qualitative effect on the critical current, and it merely reduces the value of $|i_{\rm c}|$.
Conclusion {#sec:conclusion}
==========
We have computed the Josephson current through a magnetic molecule in the tunneling regime, studying the effect of the exchange coupling with the molecular spin, and the spin anisotropy of the molecule. Performing a perturbative calculation starting from a Hamiltonian model, we have shown that an antiferromagnetic coupling between the electron spin and the molecular spin can induce a $\pi$–$0$ transition. We have described how the spin anisotropy $D$ and the quantum tunneling of magnetization term $B_2$ affect the transition.
We have shown that by studying the behavior of the critical current as a function of the magnetic field and the level position (which are both experimentally tunable parameters), it is possible to get information on the value of the spin anisotropy $D$, even outside the range of the $\pi$–$0$ transition.
This work could be extended in several directions. The calculations could be performed for a larger molecular spin (albeit at the cost of heavier expressions). One could also use anisotropy parameters that depend on the charge state of the molecule (and thus on the occupation of the dot in our model), which could describe more faithfully molecular magnets like Mn$_{12}$.[@Heersche:2006b] One could also consider the case of an external magnetic field aligned along an arbitrary direction (and not along the anisotropy axis of the molecule), in order to describe experiments where it is not possible to control the anisotropy orientation. Such a magnetic field should have a strong impact on the current, as it will mix efficiently all the molecular states.[@Timm:2007]
Finally, new possibilities could open up if one considers explicitly the Josephson current between type II superconductors. In this case, it could be possible to control the value of the superconducting gap $\Delta$ with the applied magnetic field. Going to very small $\Delta$ would give large values of $J/\Delta$, $D/\Delta$, etc., and a very large parameter range of the system, including the $\pi$–$0$ transition for $J>0$, could be explored. In the same manner, it is possible to enhance the critical temperature $T_{\rm c}$ and the second critical field $H_{\rm c2}$ by decreasing the thickness of the superconductor.[@Meservey:1971] This could allow us to use large values of the magnetic field.
We acknowledge financial support by the CNRS LIA agreements with Landau Institute; Grants No. NSF ECS-0608842, No. ARO W911NF-09-1-0395, No. DARPA HR0011-09-1-0009 (I.A.S.); NRF 2011-0003495 (M.L.); and JST-CREST, the “Topological Quantum Phenomena” (No. 22103002) KAKENHI on Innovative Areas, and a Grant-in-Aid for Scientific Research (No. 22710096) from MEXT of Japan (S.K.).
[99]{}
B. D. Josephson, Phys. Lett. [**1**]{}, 251 (1962).
P. G. De Gennes, Rev. Mod. Phys. [**36**]{}, 225 (1964).
B. I. Spivak and S. A. Kivelson, Phys. Rev. B [**43**]{}, 3740 (1991).
A. Krichevsky, M. Schechter, Y. Imry, Y. Levinson, Phys. Rev. B [**61**]{}, 3723 (2000).
C. Benjamin, T. Jonckheere, A. Zazunov, and T. Martin, Eur. Phys. J. B [**57**]{}, 279 (2007).
A. V. Rozhkov and D. P. Arovas, Phys. Rev. B [**62**]{}, 6687 (2000).
J. A. van Dam, Yu. V. Nazarov, E. P. A. M. Bakkers, S. De Franceschi, and L. P. Kouwenhoven, Nature (London) [**442**]{}, 667 (2006).
F. Siano and R. Egger, Phys. Rev. Lett. [**93**]{}, 047002 (2004).
M.-S. Choi, M. Lee, K. Kang, and W. Belzig, Phys. Rev. B [**70**]{}, 020502 (2004).
A. Yu. Kasumov, R. Deblock, M. Kociak, B. Reulet, H. Bouchiat, I. I. Khodos, Yu. B. Gorbatov, T. V. Volkov, C. Journet, and M. Burghard, Science [**284**]{}, 1508 (1999).
P. Jarillo-Herrero, J. A. van Dam, and L. P. Kouwenhoven, Nature (London) [**439**]{}, 953 (2006).
J.-P. Cleuziou, W. Wernsdorfer, V. Bouchiat, T. Ondaro̧uhu, and M. Monthioux, Nat. Nanotechnol. [**1**]{}, 53 (2006).
A. Eichler, R. Deblock, M. Weiss, C. Karrasch, V. Meden, C. Schönenberger, and H. Bouchiat, Phys. Rev. B [**79**]{}, 161407(R) (2009).
N. Roch, S. Florens, V. Bouchiat, W. Wernsdorfer, and F. Balestro, Nature (London) [**453**]{}, 633 (2008).
C. B. Winkelmann, N. Roch, W. Wernsdorfer, V. Bouchiat, and F. Balestro, Nat. Phys. [**5**]{}, 876 (2009).
D. Gatteschi and R. Sessoli, Angew. Chem. [**42**]{}, 268 (2003).
K. Petukhov, S. Hill, N. E. Chakov, K. A. Abboud, and G. Christou, Phys. Rev. B [**70**]{}, 054426 (2004).
C. Romeike, M. R. Wegewijs, W. Hofstetter, and H. Schoeller, Phys. Rev. Lett. [**96**]{}, 196601 (2006).
H. B. Heersche, Z. de Groot, J. A. Folk, H. S. J. van der Zant, C. Romeike, M. R. Wegewijs, L. Zobbi, D. Barreca, E. Tondello, and A. Cornia, Phys. Rev. Lett. [**96**]{}, 206801 (2006).
J. E. Grose, E. S. Tam, C. Timm, M. Scheloske, B. Ulgut, J. J. Parks, Héctor D. Abruña, W. Harneit, and D. C. Ralph, Nat. Mater. [**7**]{}, 884 (2008).
N. Roch, R. Vincent, F. Elste, W. Harneit, W. Wernsdorfer, C. Timm, and F. Balestro, Phys. Rev. B [**83**]{}, 081407(R) (2011).
M. Lee, T. Jonckheere, and T. Martin, Phys. Rev. Lett. [**101**]{}, 146804 (2008).
J. J. Parks, A. R. Champagne, T. A. Costi, W. W. Shum, A. N. Pasupathy, E. Neuscamman, S. Flores-Torres, P. S. Cornaglia, A. A. Aligia, C. A. Balseiro, G. K.-L. Chan, H. D. Abruna, and D. C. Ralph, Science [**328**]{}, 1370 (2010).
A. S. Zyazin, J. W. G. van den Berg, E. A. Osorio, H. S. J. van der Zant, N. P. Konstantinidis, M. Leijnse, M. R. Wegewijs, F. May, W. Hofstetter, C. Danieli, and A. Cornia, Nano Lett. [**10**]{}, 3307 (2010).
H. B. Heersche, Z. de Groot, J. A. Folk, H. S. J. van der Zant, C. Romeike, M. R. Wegewijs, L. Zobbi, D. Barreca, E. Tondello, and A. Cornia, Phys. Rev. Lett. [**96**]{}, 206801 (2006).
C. Timm, Phys. Rev. B [**76**]{}, 014421 (2007).
D. Huertas-Hernando, F. Guinea, and A. Brataas, Phys. Rev. B [**74**]{}, 155426 (2006).
M. Lee, T. Jonckheere, and T. Martin, Phys. Rev. B [**81**]{}, 155114 (2010).
Note that result Eq. (\[eq:I\_0\]) disagrees with Ref. , where a similar calculation was done; in the answer of this paper the expression $u_k^2 v_{k'}^2 + u_{k'}^2 v_{k}^2$ should be replaced by $u_k v_k u_{k'} v_{k'}$.
T. Novotný, A. Rossini, K. Flensberg, Phys. Rev. B [**72**]{}, 224502 (2005).
I. A. Sadovskyy, G. B. Lesovik, and G. Blatter, Pis’ma v Zh. Eksp. Teor. Fiz. [**86**]{}, 239 (2007) \[JETP Lett. [**86**]{}, 210 (2007)\].
R. Meservey and P. M. Tedrow, J. Appl. Phys. [**42**]{}, 51 (1971).
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'To identify the optimum conditions for the optical trapping of a droplet under microgravity, we theoretically analyzed the efficiency of trapping with counter laser beams. We found that the distance between the two foci is an important parameter for obtaining stable trapping conditions. We also performed an optical trapping experiment with counter laser beams under microgravity. The experimental results correspond well to the theoretical prediction.'
address: 'Department of Physics, Graduate School of Science, Kyoto University, Kyoto 606-8502, Japan & Spatio-Temporal Project, ICORP, JST, JAPAN'
author:
- Akihiro Isomura
- Nobuyuki Magome
- 'Masahiro I. Kohira'
- Kenichi Yoshikawa
bibliography:
- 'microg.bib'
title: Toward the Stable Optical Trapping of a Droplet with Counter Laser Beams under Microgravity
---
,
,
,
Introduction
============
Since the discovery of optical trapping in 1970 by Ashkin[@ashkin_prl_1970], optical tweezers have been actively applied in the fields of biology, physical chemistry, condensed matter physics and so on[@david_sci_2003]. Recently, some reports have mentioned the advantages of such systems on the International Space Station (ISS)[@panda_nasa_2002][@susan_nasa_2004]. The technique of optical trapping is expected to be useful on the ISS for the manipulation of droplets on a ${\rm \mu m}$ to sub-${\rm mm}$ scale, including in crystal growth by avoiding the effect of vessels. However, as far as we know, there has been no report on the optical trapping of a droplet under microgravity conditions in air.
Recently, we reported the optical levitation of a droplet under a linear increase in gravitational acceleration using a single laser[@kohira_cpl_2005]. To attain stable optical trapping with a single beam, it is necessary to use a lens with a high magnification. This means that the working distance, or the distance between the lens surface and the object, is rather small; on the order of a few ${\rm mm}$. In view of such application on the ISS, it is important to find the stable trapping condition with a greater working distance. Toward this end, we adapted an optical system with low-converged counter laser beams for optical trapping with a long working distance.
Theoretical
===========
We consider the counter laser system shown in Fig.1. In this system, the distance between the foci $d$ plays a crucial role. Although the qualitative property of such counter laser systems has been discussed in Ref.[@rossen_optcomm_1977], the effect of the distance $d$ has not yet been discussed in detail. Here, we calculate the trapping force produced by counter lasers with the variables $d$ and $R$ (which is the radius of a spherical object).
![Schematic illustration of the counter laser beams. $f$ is the focus of each laser beam and $d$ is the distance between them. We take the sign of $d$ as shown in the figure.[]{data-label="fig1"}](fig1.eps)
To take into account the effect of the inertia of the object, we start our discussion with a motion equation. To simplify the problem, we consider motion along the $x$-direction as in Fig.1. $$m \ddot{x} = F_l + F_v + F_{others}$$ where $m$ is the mass of the object, $\ddot{x}$ is the acceleration of the object along the $x$-direction, $F_l$ is the force along the optical axis induced by the converged laser, and $F_{others}$ is the other forces (including the effective acceleration force along the $x$-direction and random force induced by air).
As we see in the latter part, the velocity of the trapped object $U$ is at most ${\rm 10^{-3} mm}$. We assume that the scale of the trapped object $L$ is at most ${\rm 10^{-4} m}$, and the kinetic viscosity of air $\nu$ is around ${\rm 10^{-5} m^2/s}$. In this condition, the Reynolds number is: $Re = RU/\nu \simeq 10^{-2} \ll 1$ and we can apply Stokes’ approximation. By calculating $m = 4\pi a^3 \rho /3$ and $F_{v} = 6 \pi \eta R v$, we can verify the relation $mv^2 / 2 \ll F_{v} \cdot \Delta x$; where the radius of the trapped droplet $R$ is on the order of 10 ${\rm \mu m}$, the experimental time resolution $\Delta t$ is ${\rm 0.33\ s}$, the velocity of the trapped droplet $v = \Delta x / \Delta t$, the viscosity of air $\eta = {\rm
1.8 \times 10^{-5}\ Pa \cdot s}$ and the density of the droplet $\rho= {\rm 1.0 \times
10^{3} kg/m^3}$. This estimation indicates that the viscous force is large enough to locate the droplet at an equilibrium position in the optical potential. Therefore, under reasonable assumptions of the viscous limit, we discuss the stability of trapping droplets through $F_l$ and the effective optical potential.
$F_l$ is found as follows. In the case of trapping a large object($R/
\lambda > 10$, where $\lambda$ is the wavelength of laser), we can calculate the trapping force with the ray optics theory. Within the framework of ray optics, a TEM$_{00}$ mode laser can be divided into rays, which are suffixed with $i$, and each power of the ray $P_i$ is related to the beam deviation $\sigma$ and lens size $L$ Each ray hits the surface of the droplet at a different incident angle $\phi_i$ ($0 \leq
\phi_i \leq \phi$), repeatedly reflects and is transmitted in the droplet until the intensity of the rays decreases to the zero limit, and the momentum is given with a certain efficiency along the $x$-axis, $Q_i=Q_i(R,x;\phi_i,n_1,n_2)$: $$\begin{aligned}
Q_{i} = \sin \phi_i \left\{ R_i \sin 2 \theta_i - \frac{T_i^2[\sin(2\theta_i - 2r_i)+R_i\sin2\theta_i]}{1+R_i^2+2R_i\cos2r_i}\right\}\nonumber\\+ \cos \phi_i \left\{ 1+R_i \cos 2\theta_i -\frac{T_i^2[\cos(2\theta_i-2r_i)+R_i\cos2\theta_i]}{1+R_i^2+2R_i\cos2r_i} \right\}
\label{eq:qi}\end{aligned}$$ where $\theta_i$: incident angle, $r_i$: refractive angle, $R_i$:reflection coefficient and $T_i$: transmission coefficient. These parameters are obtained by considering the geometric relation between the droplet and the direction of the beam (see ref.[@ashkin_bpj_1992]). Assuming a spherical droplet with radius $R$ whose center is located at distance $x$ from the center of the foci, we calculate the details of the reflections and transmissions for all paths of the laser beam. The total force $f_l$ is $$f_l = \sum_{i} \frac{n_1 P_i}{{\rm c}}Q_i = \frac{n_1 P}{{\rm c}} Q_x$$ where $Q_x=Q_x(R,x; \phi,\frac{\sigma}{L},n_1,n_2)$: trapping efficiency along the optical axis and $c$: velocity of light. In addition, considering the distances between the foci, the total force $F_l$ is: $$F_l(d,x,R) =f_{l}^{Right} + f_{l}^{Left}$$ Under the assumption of a spherical trapped object, optical trapping exhibits geometrical symmetry and we can use a normalized unit of length. Eventually, the optical potential is expressed by two normalized parameters, $d/R$ and $x/R$.
The results of the calculation for effective trapping potential are shown in Fig.2, indicating that the trapping efficiency is significantly dependent on $d$. An optical configuration with positive $d$ is clearly preferable for optilcal trapping under microgravity. Figure 2(c) shows the change in the potential profile as a function of $d/R$. As shown, not only the distance between the foci but also the radius of the droplet greatly affect the profile of the optical potential. This chart is reasonable for understanding qualitative properties: if the radius $R$ is sufficiently large, the trapped object is insensitive to the distance between the foci $d$ and the sign of $d$ become less important. In contrast, when $R$ is sufficiently small, $d$ becomes non-negligible and we must choose better settings of $d$. Since we sought to define the optimum conditions for trapping large objects (sub-mm scale), this diagram is applicable for designing a counter laser trapping system. For example. if we can precisely adjust the positions of the foci, the setting at $d/R = 4$ is advantageous for trapping the object rather than $d/R = -4$. The potentiality to trap small particles is important in the growth of protein crystals[@giege_pcgcm_1995][@hosokawa_jap_2004] and liquid droplets[@magome_jpc_2003].
Experimental
============
Microgravity conditions were achieved by using a jet airplane(Mitsubishi MU-300, operated by Diamond Air Service Co., Aichi, Japan). In the experiment, the airplane flew along a parabolic flight profile, as shown in Fig.3. During parabolic flight, the effective gravitational acceleration $g$ in the airplane was 0.01 $g_0$ for about 20 s, where the gravitational acceleration on the ground $g_0 = 9.8 {\rm m/s^2}$.
The experimental setup for the optical trapping is shown in Fig.3(b). We selected a counter laser system to press the droplets to the center of the foci.[@rossen_pl_1976][@cizmar_apl_2005] The laser beam from a diode pumped ${\rm Nd:YVO_4}$($\lambda = 532\ {\rm nm}$) was split into two beams by a half mirror, transferred via optical fibers, and focused by achromatic lenses. The convergence angle of the lenses was 17 degrees, so the working distance was on the order of [cm]{}. The laser power was set to [170 mW]{}. We arranged two different optical settings, where the positions of the laser foci are different, either $d$ is positive or negative as shown in Fig.1. The droplets were injected into a glass cell ${\rm (10 mm \times 10 mm \times 50 mm)}$ with an atomizer. The droplets, as visualized by scattering visible laser light, were monitored with a CCD camera from the y-direction in Fig.3(b). Movies were recorded at 30 frames per second. The experiments was carried out at arround 298K.
Results and Discussion
======================
On the ground in air, using the same optical trapping system, the injected droplets immediately fell due to the force of gravity. However, under microgravity in air, most of the droplets didn’t fall and we could observe the motions of the droplets along the laser axes. To trace the droplets’ motion along the laser axes, we converted the movie images to a spatio-temporal diagram. Figure 4 shows the results of the optical trapping of the droplets under microgravity.
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Figure 4 (a1) and (b1) are spatio-temporal diagrams which show the trajectory of the droplets. Figure 4 (a2) and (b2) are snapshots of the droplets. In these figures, the burring on the images correspond to rather fast motion of the droplets. The small white spots are droplets that have departed from the laser axes and are floating in front of or behind the camera foci. In contrast, a trapped droplet is a star-like object(white arrow), since a trapped droplet on the laser axes scatter light intensively.
Figure 4(c) shows typical profiles depicted based on the results of Fig.2(a) and (b), which are associated with the two different optical settings in Fig.4(a2) and (b2). As shown near the trapping region, the gradient force in the case of the solid arrow is much greater than that in the case of the dashed arrow($d/R=-4$). It is clear that our theoretical study reproduces the experimental trends well.
Conclusion
==========
We performed the optical trapping of a water droplet with a counter laser in air under microgravity and investigated the properties of the counter laser trapping system. We showed that the position of the laser foci strongly affected the trapping efficiency. A theoretical calculation based on ray optics reproduced the experimental trend. The present results may contribute to the design of a manipulation system on the International Space Station (ISS), including experiments on protein crystal growth and a container-less micro-reactor.
Acknowledgement
===============
The authors thank Mr. S. Watanabe, Mr. Y. Sumino, and N. Matsuda for their helpful suggestions and Ms. Hayata and Messrs. Fujii, Kawakatsu, and Takahashi for their technical assistance. This research was supported in part by a Grant-in Aid from the Ministry of Education, Science, Sports and Culture of Japan, the Grant-in-Aid for the 21st Century COE ¡ÈCenter for Diversity and Universality in Physics¡É from the Ministry of Education, Culture, Sports, Science and Technology (MEXT) of Japan, and ¡ÈGround-based Research Announcement for Space Utilization¡É promoted by the Japan Space Forum.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In distributed processing, agents generally collect data generated by the *same* underlying unknown model (represented by a vector of parameters) and then solve an estimation or inference task cooperatively. In this paper, we consider the situation in which the data observed by the agents may have risen from two *different* models. Agents do not know beforehand which model accounts for their data and the data of their neighbors. The objective for the network is for all agents to reach agreement on which model to track and to estimate this model cooperatively. In these situations, where agents are subject to data from unknown different sources, conventional distributed estimation strategies would lead to biased estimates relative to any of the underlying models. We first show how to modify existing strategies to guarantee unbiasedness. We then develop a classification scheme for the agents to identify the models that generated the data, and propose a procedure by which the entire network can be made to converge towards the same model through a collaborative decision-making process. The resulting algorithm is applied to model fish foraging behavior in the presence of two food sources.'
author:
- 'Sheng-Yuan Tu, and Ali H. Sayed, [^1]'
bibliography:
- 'IEEEfull.bib'
- 'refs.bib'
title: 'Distributed Decision-Making over Adaptive Networks'
---
Adaptive networks, diffusion adaptation, classification, decision-making, biological networks.
Introduction
============
Self-organization is a remarkable property of biological networks [@Camazine03; @Couzin09], where various forms of complex behavior are evident and result from decentralized interactions among agents with limited capabilities. One example of sophisticated behavior is the group decision-making process by animals [@Sumpter09]. For example, it is common for biological networks to encounter situations where agents need to decide between multiple options, such as fish deciding between following one food source or another [@Couzin11], and bees or ants deciding between moving towards a new hive or another [@Britton02; @Pratt02]. Although multiple options may be available, the agents are still able to reach agreement in a decentralized manner and move towards a common destination (e.g., [@Beekman06]).
In previous works, we proposed and studied several diffusion strategies [@Lopes08; @Cattivelli08; @Cattivelli10; @Chen12; @Sayed13a; @Sayed13b] that allow agents to adapt and learn through a process of in-network collaboration and learning. References [@Sayed13a; @Sayed13b] provide overviews of diffusion techniques and their application to distributed adaptation, learning, and optimization over networks. Examples of further applications and studies appear, e.g., in [@Li09; @Chouvardas11; @Lorenzo13b; @Xia11b; @Takahashi10b; @Chouvardas12]. Diffusion networks consist of a collection of adaptive agents that are able to respond to excitations in real-time. Compared with the class of consensus strategies [@Tsitsiklis86; @Nedic09; @Schizas09; @Mateos09; @Dimakis10; @Kar11; @Kar12], diffusion networks have been shown to remain stable irrespective of the network topology, while consensus networks can become unstable even when each agent is individually stable [@Tu12a]. Diffusion strategies have also been shown to lead to improved convergence rate and superior mean-square-error performance [@Tu12a; @Sayed13b]. For these reasons, we focus in the remainder of this paper on the use of diffusion strategies for decentralized decision-making.
Motivated by the behavior of biological networks, we study distributed decision-making over networks where agents are subject to data arising from two different models. The agents do not know beforehand which model accounts for their data and the data of their neighbors. The objective of the network is for all agents to reach agreement on one model and to estimate and track this *common* model cooperatively. The task of reaching agreement over a network of agents subjected to different models is more challenging than earlier works on inference under a single data model. The difficulty is due to various reasons. First, traditional (consensus and diffusion) strategies will converge to a biased solution (see Eq. (\[eq91\])). We therefore need a mechanism to compensate for the bias. Second, each agent now needs to distinguish between which model each of its neighbors is collecting data from (this is called the *observed* model) and which model the network is evolving to (this is called the *desired* model). In other words, in addition to the learning and adaptation process for tracking, the agents should be equipped with a classification scheme to distinguish between the observed and desired models. The agents also need to be endowed with a decision process to agree among themselves on a common (desired) model to track. Moreover, the classification scheme and the decision-making process will need to be implemented in a fully distributed manner and in real-time, alongside the adaptation process.
There have been useful prior works in the literature on formations over multi-agent networks [@Jadbabaie03; @Olfati04; @Fax04; @Olfati06; @Yildiz10; @Khan10; @Forero10] and opinion formation over social networks [@Castellano09; @Acemoglu10; @Jadbabaie12] using, for example, consensus strategies. These earlier works are mainly interested in having the agents reach an average consensus state, whereas in our problem formulation agents will need to reach one of the models and not the average of both models. Another difference between this work and the earlier efforts is our focus on combining real-time classification, decision-making, and adaptation into a single integrated framework running at each agent. To do so, we need to show how the distributed strategy should be modified to remove the bias that would arise due to the multiplicity of models — without this step, the combined decision-making and adaptation scheme will not perform as required. In addition, in our formulation, the agents need to continuously adjust their decisions and their estimates because the models are allowed to change over time. In this way, reaching a static consensus is not the objective of the network. Instead, the agents need to continuously adjust and track in a dynamic environment where decisions and estimates evolve with time as necessary. Diffusion strategies endow networks with such tracking abilities — see, e.g., Sec. VII of [@Zhao12], where it is shown how well these strategies track as a function of the level of non-stationarity in the underlying models.
Diffusion Strategy
==================
Consider a collection of $N$ agents (or nodes) distributed over a geographic region. The set of neighbors (i.e. neighborhood) of node $k$ is denoted by $\mathcal{N}_k$; the number of nodes in $\mathcal{N}_k$ is denoted by $n_k$. At every time instant, $i$, each node $k$ is able to observe realizations $\{d_k(i),u_{k,i}\}$ of a scalar random process $\boldsymbol{d}_{k}(i)$ and a $1\times M$ *row* random regressor $\boldsymbol{u}_{k,i}$ with a positive-definite covariance matrix, $R_{u,k}=\mathbb{E}\boldsymbol{u}_{k,i}^T\boldsymbol{u}_{k,i}>0$. The regressors $\{\boldsymbol{u}_{k,i}\}$ are assumed to be temporally white and spatially independent, i.e., $\mathbb{E}\boldsymbol{u}^T_{k,i}\boldsymbol{u}_{l,j}=
R_{u,k}\delta_{kl}\delta_{ij}$ in terms of the Kronecker delta function. Note that we are denoting random quantities by boldface letters and their realizations or deterministic quantities by normal letters. The data $\{\boldsymbol{d}_{k}(i),\boldsymbol{u}_{k,i}\}$ collected at node $k$ are assumed to originate from one of two unknown *column* vectors $\{w^\circ_0,w^\circ_1\}$ of size $M$ in the following manner. We denote the generic observed model by $z_k^\circ\in \{w_0^\circ, w_1^\circ\}$; node $k$ does not know beforehand the observed model. The data at node $k$ are related to its observed model $z^\circ_k$ via a linear regression model of the form: $$\label{eq2}
\boldsymbol{d}_k(i) = \boldsymbol{u}_{k,i}z^\circ_k+\boldsymbol{v}_k(i)$$ where $\boldsymbol{v}_k(i)$ is measurement noise with variance $\sigma^2_{v,k}$ and assumed to be temporally white and spatially independent. The noise $\boldsymbol{v}_k(i)$ is assumed to be independent of $\boldsymbol{u}_{l,j}$ for all $\{k,l,i,j\}$. All random processes are zero mean.
The objective of the network is to have *all* agents converge to an estimate for *one* of the models. For example, if the models happen to represent the location of food sources [@Tu11a; @Sayed13a], then this agreement will make the agents move towards one particular food source in lieu of the other source. More specifically, let $\boldsymbol{w}_{k,i}$ denote the estimator for $z^\circ_k$ at node $k$ at time $i$. The network would like to reach an ageement on a common $q$, such that $$\label{eq23}
\boldsymbol{w}_{k,i} \rightarrow w^\circ_q
\text{ for $q=0$ or $q=1$ and for all $k$ as $i\rightarrow\infty$}$$ where convergence is in some desirable sense (such as the mean-square-error sense).
Several *adaptive* diffusion strategies for distributed estimation under a common model scenario were proposed and studied in [@Lopes08; @Cattivelli10; @Cattivelli08; @Chen12; @Sayed13a], following the developments in [@Lopes06; @Sayed07; @Lopes07; @Cattivelli07; @Cattivelli08b] — overviews of these results appear in [@Sayed13a; @Sayed13b]. One such scheme is the adaptive-then-combine (ATC) diffusion strategy [@Cattivelli08b; @Cattivelli10]. It operates as follows. We select an $N\times N$ matrix $A$ with nonnegative entries $\{a_{l,k}\}$ satisfying: $$\label{eq50}
\mathds{1}^T_NA=\mathds{1}^T_N \quad\text{and}\quad
a_{l,k}=0 \text{ if $l\notin\mathcal{N}_k$}$$ where $\mathds{1}_N$ is the vector of size $N$ with all entries equal to one. The entry $a_{l,k}$ denotes the weight that node $k$ assigns to data arriving from node $l$ (see Fig. \[Fig\_1\]). The ATC diffusion strategy updates $\boldsymbol{w}_{k,i-1}$ to $\boldsymbol{w}_{k,i}$ as follows: $$\begin{aligned}
\label{eq49}
\boldsymbol{\psi}_{k,i} &= \boldsymbol{w}_{k,i-1}+
\mu_k\cdot \boldsymbol{u}_{k,i}^T[\boldsymbol{d}_k(i)-
\boldsymbol{u}_{k,i}\boldsymbol{w}_{k,i-1}]\\
\boldsymbol{w}_{k,i} &= \sum_{l\in\mathcal{N}_k}a_{l,k}\boldsymbol{\psi}_{l,i} \label{eq1}\end{aligned}$$ where $\mu_k$ is the *constant* positive step-size used by node $k$. The first step (\[eq49\]) involves local adaptation, where node $k$ uses its own data $\{\boldsymbol{d}_k(i),\boldsymbol{u}_{k,i}\}$ to update the weight estimate at node $k$ from $\boldsymbol{w}_{k,i-1}$ to an intermediate value $\boldsymbol{\psi}_{k,i}$. The second step (\[eq1\]) is a combination step where the intermediate estimates $\{\boldsymbol{\psi}_{l,i}\}$ from the neighborhood of node $k$ are combined through the weights $\{a_{l,k}\}$ to obtain the updated weight estimate $\boldsymbol{w}_{k,i}$. Such diffusion strategies have found applications in several domains including distributed optimization, adaptation, learning, and the modeling of biological networks — see, e.g., [@Sayed13a; @Sayed13b; @Tu11a] and the references therein. Diffusion strategies were also used in some recent works [@Ram10; @Bianchi11; @Srivastava11; @Stankovic11] albeit with diminishing step-sizes ($\mu_k(i)\rightarrow 0$) to enforce consensus among nodes. However, decaying step-sizes disable adaptation once they approach zero. Constant step-sizes are used in (\[eq49\])-(\[eq1\]) to enable continuous adaptation and learning, which is critical for the application under study in this work.
![A connected network where data collected by the agents are influenced by one of two models. The weight $a_{l,k}$ scales the data transmitted from node $l$ to node $k$ over the edge linking them.[]{data-label="Fig_1"}](fig1_network_plot.eps){width="14em"}
When the data arriving at the nodes could have risen from one model or another, the distributed strategy (\[eq49\])-(\[eq1\]) will not be able to achieve agreement as in (\[eq23\]) and the resulting weight estimates will tend towards a biased value. We first explain how this degradation arises and subsequently explain how it can be remedied.
\[ass3\] The network topology is assumed to be strongly connected so that the corresponding combination matrix $A$ is primitive, i.e., there exists an integer power $j>0$ such that $[A^j]_{l,k} > 0$ for all $l$ and $k$.
As explained in [@Sayed13a], Assumption \[ass3\] amounts to requiring the network to be connected (where a path with nonzero weights exists between any two nodes), and for at least one node to have a non-trivial self-loop (i.e., $a_{k,k}>0$ for at least one $k$). We conclude from the Perron-Frobenius Theorem [@Horn85; @Berman94] that every primitive left-stochastic matrix $A$ has a unique eigenvalue at one while all other eigenvalues are strictly less than one in magnitude. Moreover, if we denote the right-eigenvector that is associated with the eigenvalue at one by $c$ and normalize its entries to add up to one then it holds that: $$\label{eq9}
Ac=c,\quad \mathds{1}_N^T c=1,\quad
\text{and}\quad0<c_k<1.$$
Let us assume for the time being that the agents in the network have agreed on converging towards one of the models (but they do not know beforehand which model it will be). We denote the desired model generically by $w_q^\circ$. In Section \[sec2\], we explain how this agreement process can be attained. Here we explain that even when agreement is present, the diffusion strategy (\[eq49\])-(\[eq1\]) leads to biased estimates unless it is modified in a proper way. To see this, we introduce the following error vectors for any node $k$: $$\label{eq29}
\tilde{\boldsymbol{w}}_{k,i}\triangleq
w^\circ_q-\boldsymbol{w}_{k,i}\;\;\text{ and }\;\;
\tilde{z}^\circ_{k}\triangleq w^\circ_q-z^\circ_k.$$ Then, using model (\[eq2\]), we obtain that the update vector in (\[eq49\]) becomes $$\begin{aligned}
\boldsymbol{h}_{k,i}&\triangleq
\boldsymbol{u}^T_{k,i}[\boldsymbol{d}_k(i)-
\boldsymbol{u}_{k,i}\boldsymbol{w}_{k,i-1}]\notag \\
&=\boldsymbol{u}^T_{k,i}\boldsymbol{u}_{k,i}\tilde{\boldsymbol{w}}_{k,i-1}
-\boldsymbol{u}^T_{k,i}\boldsymbol{u}_{k,i}\tilde{z}^\circ_{k}
+\boldsymbol{u}^T_{k,i}\boldsymbol{v}_k(i).\label{eq20}\end{aligned}$$ We collect all error vectors across the network into block vectors: $\tilde{\boldsymbol{w}}_{i}\triangleq
\text{col}\left\{\tilde{\boldsymbol{w}}_{k,i}\right\}$ and $\tilde{z}^\circ\triangleq
\text{col}\left\{\tilde{z}^\circ_{k}\right\}$. We also collect the step-sizes into a block diagonal matrix and introduce the extended combination matrix: $$\label{eq56}
\mathcal{M} = \text{diag}\{\mu_kI_M\}\quad \text{and}\quad
\mathcal{A}\triangleq A\otimes I_M$$ where $I_M$ denotes the identity matrix of size $M$. In (\[eq56\]), the notation $\text{diag}\{\cdot\}$ constructs a diagonal matrix from its arguments and the symbol $\otimes$ denotes the Kronecker product of two matrices. Moreover, the notation $\text{col}\{\cdot\}$ denotes the vector that is obtained by stacking its arguments on top of each other. Then, starting from (\[eq49\])-(\[eq1\]) and using relation (\[eq20\]), we can verify that the global error vector $\tilde{\boldsymbol{w}}_{i}$ of the network evolves over time according to the recursion: $$\label{eq3}
\tilde{\boldsymbol{w}}_{i}=
\boldsymbol{\mathcal{B}}_i \tilde{\boldsymbol{w}}_{i-1}+
\boldsymbol{y}_i$$ where the matrix $\boldsymbol{\mathcal{B}}_i$ and the vector $\boldsymbol{y}_i$ are defined in Table \[tab1\] with $\boldsymbol{\mathcal{R}}_i\triangleq\text{diag}\{
\boldsymbol{u}^T_{k,i}\boldsymbol{u}_{k,i}\}$ and $\boldsymbol{s}_i\triangleq\text{col}\{
\boldsymbol{u}^T_{k,i}\boldsymbol{v}_{k,i}\}$. Note that the matrix $\boldsymbol{\mathcal{B}}_i$ is a random matrix due to the randomness of the regressors $\{\boldsymbol{u}_{k,i}\}$. Since the regressors are temporally white and spatially independent, then $\boldsymbol{\mathcal{B}}_{i}$ is independent of $\tilde{\boldsymbol{w}}_{i-1}$. In addition, since $\boldsymbol{u}_{k,i}$ is independent of $\boldsymbol{v}_{k}(i)$, the vector $\boldsymbol{s}_i$ in $\boldsymbol{y}_i$ has zero mean. Then, from (\[eq3\]), the mean of $\tilde{\boldsymbol{w}}_{i}$ evolves over time according to the recursion: $$\label{eq5}
\mathbb{E}\tilde{\boldsymbol{w}}_{i}=\mathcal{B}\cdot
\mathbb{E}\tilde{\boldsymbol{w}}_{i-1}+y$$ where $\mathcal{B}\triangleq \mathbb{E}\boldsymbol{\mathcal{B}}_{i}$ and $y \triangleq \mathbb{E}\boldsymbol{y}_{i}$ are defined in Table \[tab1\] with $\mathcal{R}\triangleq
\mathbb{E}\boldsymbol{\mathcal{R}}_{i} =\text{diag}\{R_{u,k}\}$. It can be easily verified that a necessary and sufficient condition to ensure the convergence of $\mathbb{E}\tilde{\boldsymbol{w}}_{i}$ in (\[eq5\]) to zero is $$\label{eq22}
\rho(\mathcal{B})<1\quad\text{and}\quad
y=0$$ where $\rho(\cdot)$ denotes the spectral radius of its argument. It was verified in [@Sayed13a; @Tu12a] that a sufficient condition to ensure $\rho({\cal B})<1$ is to select the site-sizes $\{\mu_k\}$ such that $$\label{eq26}
0 < \mu_k < \frac{2}{\rho(R_{u,k})}$$ for all $k$. This conclusion is independent of $A$. However, for the second condition in (\[eq22\]), we note that in general, the vector $y=\mathcal{A}^T\mathcal{M}\mathcal{R}\tilde{z}^\circ$ cannot be zero no matter how the nodes select the combination matrix $A$. When this happens, the weight estimate will be biased. Let us consider the example with three nodes in Fig. \[Fig\_2\] where node 1 observes data from model $w^\circ_0$, while nodes 2 and 3 observe data from another model $w^\circ_1$. The matrix $A$ in this case is shown in Fig. \[Fig\_2\] with the parameters $\{a,b,c,d\}$ lying in the interval $[0,1]$ and $b+c\leq1$. We assume that the step-sizes and regression covariance matrices are the same, i.e., $\mu_k=\mu$ and $R_{u,k}=R_u$ for all $k$. If the desired model of the network is $w^\circ_q=w^\circ_0$, then the third block of $y$ becomes $\mu
R_u(w^\circ_0-w^\circ_1)$, which can never become zero no matter what the parameters $\{a,b,c,d\}$ are. More generally, using results on the limiting behavior of the estimation errors $\{\tilde{\boldsymbol{w}}_{k,i}\}$ from [@Chen13b], we can characterize the limiting point of the diffusion strategy (\[eq49\])-(\[eq1\]) as follows.
\[tab1\]
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
**Diffusion** (\[eq49\])-(\[eq1\]) **Modified diffusion** (\[eq6\])-(\[eq15\])
--------------------------------------------------------------- --------------------------------------------------------------------- -----------------------------------------------------------------------
$\boldsymbol{\mathcal{B}}_i$ $\mathcal{A}^{T}(I_{NM}-\mathcal{M}\boldsymbol{\mathcal{R}}_i)$ $\mathcal{A}_1^{T}(I_{NM}-\mathcal{M}\boldsymbol{\mathcal{R}}_i)
+\mathcal{A}_2^{T}$
$\mathcal{B}\triangleq\mathbb{E}\boldsymbol{\mathcal{B}}_{i}$ $\mathcal{A}^T(I_{NM}-\mathcal{M}\mathcal{R})$ $\mathcal{A}_1^T(I_{NM}-\mathcal{M}\mathcal{R})+\mathcal{A}_2^T$
$\boldsymbol{y}_i$ $\mathcal{A}^T\mathcal{M}\boldsymbol{\mathcal{R}}_i\tilde{z}^\circ- $\mathcal{A}_1^T\mathcal{M}\boldsymbol{\mathcal{R}}_i\tilde{z}^\circ-
\mathcal{A}^T\mathcal{M}\boldsymbol{s}_i$ \mathcal{A}_1^T\mathcal{M}\boldsymbol{s}_i$
$y\triangleq \mathbb{E}\boldsymbol{y}_i$ $\mathcal{A}^T\mathcal{M}\mathcal{R}\tilde{z}^\circ$ $\mathcal{A}_1^T\mathcal{M}\mathcal{R}\tilde{z}^\circ$
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
: [The error vector evolves according to the recursion]{} $\tilde{\boldsymbol{w}}_{i}=\boldsymbol{\mathcal{B}}_i
\tilde{\boldsymbol{w}}_{i-1}+\boldsymbol{y}_i$, [where the variables]{} $\{\boldsymbol{\mathcal{B}}_i,
\boldsymbol{y}_i\}$ [and their respective means are listed below for the conventional and modified diffusion strategies.]{}
![A three-node network. Node 1 observes data from $w^\circ_0$ while nodes 2 and 3 observe data from $w^\circ_1$.[]{data-label="Fig_2"}](fig2_Three_node_example2.eps){width="20em"}
\[lem3\] For the diffusion strategy (\[eq49\])-(\[eq1\]) with $\mu_k=\mu$ and $R_{u,k}=R_u$ for all $k$ and for sufficiently small step-sizes, all weight estimators $\{\boldsymbol{w}_{k,i}\}$ converge to a limit point $w^\circ$ in the mean-square sense, i.e., $\mathbb{E}\|w^\circ-\boldsymbol{w}_{k,i}\|^2$ is bounded and of the order of $\mu$, where $w^\circ$ is given by $$\label{eq91}
w^\circ = \sum_{k=1}^Nc_kz^\circ_k$$ where the vector $c$ is defined in (\[eq9\]).
The result follows from Eq. (25) in [@Chen13b] by noting that the variable $s_k(w^\circ)$ used in [@Chen13b] is given by $R_{u}(z^\circ_k-w^\circ)$.
Thus, when the agents collect data from different models, the estimates using the diffusion strategy (\[eq49\])-(\[eq1\]) converge to a convex combination of these models given by (\[eq91\]), which is different from any of the individual models because $c_k>0$ for all $k$. A similar conclusion holds for the case of non-uniform step-sizes $\{\mu_k\}$ and covariance matrices $\{R_{u,k}\}$.
Modified Diffusion Strategy
===========================
To deal with the problem of bias, we now show how to modify the diffusion strategy (\[eq49\])-(\[eq1\]). We observe from the example in Fig. \[Fig\_2\] that the third entry of the vector $y$ cannot be zero because the neighbor of node $3$ observes data arising from a model that is different from the desired model. Note from (\[eq20\]) that the bias term arises from the gradient direction used in computing the intermediate estimates in (\[eq49\]). These observations suggest that to ensure unbiased mean convergence, a node should not combine intermediate estimates from neighbors whose observed model is different from the desired model. For this reason, we shall replace the intermediate estimates from these neighbors by their previous estimates $\{\boldsymbol{w}_{l,i-1}\}$ in the combination step (\[eq1\]). Specifically, we shall adjust the diffusion strategy (\[eq49\])-(\[eq1\]) as follows: $$\begin{aligned}
\label{eq6}
\boldsymbol{\psi}_{k,i} &= \boldsymbol{w}_{k,i-1}+
\mu_k\cdot\boldsymbol{u}_{k,i}^T
[\boldsymbol{d}_k(i)-\boldsymbol{u}_{k,i}\boldsymbol{w}_{k,i-1}]\\
\boldsymbol{w}_{k,i} &= \sum_{l\in\mathcal{N}_k}\left(a^{(1)}_{l,k}
\boldsymbol{\psi}_{l,i}
+a^{(2)}_{l,k}\boldsymbol{w}_{l,i-1}\right)\label{eq15}\end{aligned}$$ where the $\{a^{(1)}_{l,k}\}$ and $\{a^{(2)}_{l,k}\}$ are two sets of nonnegative scalars and their respective combination matrices $A_1$ and $A_2$ satisfy $$\label{eq111}
A_1+A_2=A$$ with $A$ being the original left-stochastic matrix in (\[eq50\]). Note that step (\[eq6\]) is the same as step (\[eq49\]). However, in the second step (\[eq15\]), nodes aggregate the $\{\boldsymbol{\psi}_{l,i},\boldsymbol{w}_{l,i-1}\}$ from their neighborhood. With such adjustment, we will verify that by properly selecting $\{a^{(1)}_{l,k},a^{(2)}_{l,k}\}$, unbiased mean convergence can be guaranteed. The choice of which entries of $A$ go into $A_1$ or $A_2$ will depend on which of the neighbors of node $k$ are observing data arising from a model that agrees with the desired model for node $k$.
Construction of Matrices $A_1$ and $A_2$
----------------------------------------
To construct the matrices $\{A_1,A_2\}$ we associate two vectors with the network, $f$ and $g_i$. Both vectors are of size $N$. The vector $f$ is fixed and its $k$th entry, $f(k)$, is set to $f(k)=0$ when the observed model for node $k$ is $w_0^\circ$; otherwise, it is set to $f(k)=1$. On the other hand, the vector $g_i$ is evolving with time; its $k$th entry is set to $g_i(k)=0$ when the desired model for node $k$ is $w_0^\circ$; otherwise, it is set equal to $g_i(k)=1$. Then, we shall set the entries of $A_1$ and $A_2$ according to the following rules: $$\begin{aligned}
\label{eq24}
& a^{(1)}_{l,k,i} =
\begin{cases}
a_{l,k}, &\text{if $l\in\mathcal{N}_k$ and $f(l)= g_i(k)$}\\
0, &\text{otherwise}
\end{cases}\\
& a^{(2)}_{l,k,i} =
\begin{cases}
a_{l,k}, &\text{if $l\in\mathcal{N}_k$ and $f(l)\neq g_i(k)$}\\
0, &\text{otherwise}
\end{cases}.\label{eq25}\end{aligned}$$ That is, nodes that observe data arising from the same model that node $k$ wishes to converge to will be reinforced and their intermediate estimates $\{\boldsymbol{\psi}_{l,i}\}$ will be used (their combination weights are collected into matrix $A_1$). On the other hand, nodes that observe data arising from a different model than the objective for node $k$ will be de-emphasized and their prior estimates $\{\boldsymbol{w}_{l,i-1}\}$ will be used in the combination step (\[eq15\]) (their combination weights are collected into matrix $A_2$). Note that the scalars $\{a^{(1)}_{l,k,i},a^{(2)}_{l,k,i}\}$ in (\[eq24\])-(\[eq25\]) are now indexed with time due to their dependence on $g_i(k)$.
Mean-Error Analysis
-------------------
It is important to note that to construct the combination weights from (\[eq24\])-(\[eq25\]), each node $k$ needs to know what are the observed models influencing its neighbors (i.e., $f(l)$ for $l\in\mathcal{N}_k$); it also needs to know how to update its objective in $g_i(k)$ so that the $\{g_i(l)\}$ converge to the same value. In the next two sections, we will describe a distributed decision-making procedure by which the nodes are able to achieve agreement on $\{g_i(k)\}$. We will also develop a classification scheme to estimate $\{f(l)\}$ using available data. More importantly, the convergence of the vectors $\{f,g_i\}$ will occur before the convergence of the adaptation process to estimate the agreed-upon model. Therefore, let us assume for the time being that the nodes know the $\{f(l)\}$ of their neighbors and have achieved agreement on the desired model, which we are denoting by $w_q^\circ$, so that (see Eq. (\[eq127\]) in Theorem \[thm1\]) $$\label{eq30}
g_i(1)=g_i(2)=\cdots=g_i(N)=q,\;\; \text{ for all $i$.}$$
Using relation (\[eq20\]) and the modified diffusion strategy (\[eq6\])-(\[eq15\]), the recursion for the global error vector $\tilde{\boldsymbol{w}}_{i}$ is again given by (\[eq3\]) with the matrix $\boldsymbol{\mathcal{B}}_i$ and the vector $\boldsymbol{y}_i$ defined in Table \[tab1\] and the combination matrices $\mathcal{A}_1$ and $\mathcal{A}_2$ defined in a manner similar to $\mathcal{A}$ in (\[eq56\]). We therefore get the same mean recursion as (\[eq5\]) with the matrix $\mathcal{B}$ and the vector $y$ defined in Table \[tab1\]. The following result establishes asymptotic mean convergence for the modified diffusion strategy (\[eq6\])-(\[eq15\]).
\[thm2\] Under condition (\[eq30\]), the modified diffusion strategy (\[eq6\])-(\[eq15\]) converges in the mean if the matrices $A_1$ and $A_2$ are constructed according to (\[eq24\])-(\[eq25\]) and the step-sizes $\{\mu_k\}$ satisfy condition (\[eq26\]) for those nodes whose observed model is the same as the desired model $w^\circ_q$ for the network.
See Appendix \[appC\].
We conclude from the argument in Appendix \[appC\] that the net effect of the construction (\[eq24\])–(\[eq25\]) is the following. Let $w_q^\circ$ denote the desired model that the network wishes to converge to. We denote by $\mathcal{N}_q$ the subset of nodes that receive data arising from the *same* model. The remaining nodes belong to the set $\mathcal{N}^c_q$. Nodes that belong to the set $\mathcal{N}_q$ run the traditional diffusion strategy (\[eq49\])-(\[eq1\]) using the combination matrix $A$ and their step-sizes are required to satisfy (\[eq26\]). The remaining nodes in $\mathcal{N}_q^c$ set their step-sizes to zero and run only step (\[eq1\]) of the diffusion strategy. These nodes do not perform the adaptive update (\[eq49\]) and therefore their estimates satisfy $\boldsymbol{\psi}_{k,i}=\boldsymbol{w}_{k,i-1}$ for all $k\in \mathcal{N}^c_q$.
![Decision-making process (a) node $k$ receives the desired models from its neighbors (b) node $k$ updates its desired model using (\[eq17\])-(\[eq57\]) with 20 neighbors.[]{data-label="Fig_13"}](fig3_Desired_model2.eps){width="25em"}
Distributed Decision-Making {#sec2}
===========================
The decision-making process is motivated by the process used by animal groups to reach agreement, and which is known as quorum response [@Britton02; @Pratt02; @Sumpter09]. The procedure is illustrated in Fig. \[Fig\_13\] and described as follows. At time $i$, every node $k$ has its previous desired model $\boldsymbol{g}_{i-1}(k)$, now modeled as a random variable since it will be constructed from data realizations that are subject to randomness. Node $k$ exchanges $\boldsymbol{g}_{i-1}(k)$ with its neighbors and constructs the set $$\label{eq98}
\mathcal{N}_{k,i-1}^g =
\{l\;|\;l\in\mathcal{N}_k,\boldsymbol{g}_{i-1}(l)=\boldsymbol{g}_{i-1}(k)\}.$$ That is, the set $\mathcal{N}_{k,i-1}^g$ contains the subset of nodes that are in the neighborhood of $k$ and have the same desired model as node $k$ at time $i-1$. This set changes over time. Let $n^g_k(i-1)$ denote the number of nodes in $\mathcal{N}_{k,i-1}^g$. Since at least one node (node $k$) belongs to $\mathcal{N}_{k,i-1}^g$, we have that $n^g_k(i-1)\geq 1$. Then, one way for node $k$ to participate in the quorum response is to update its desired model $\boldsymbol{g}_i(k)$ according to the rule: $$\label{eq17}
\boldsymbol{g}_i(k) = \begin{cases}
\boldsymbol{g}_{i-1}(k), &\text{with probability $q_{k,i-1}$}\\
1-\boldsymbol{g}_{i-1}(k), &\text{with probability $1-q_{k,i-1}$}
\end{cases}$$ where the probability measure is computed as: $$\label{eq57}
q_{k,i-1} = \frac{[n^g_k(i-1)]^K}{[n^g_k(i-1)]^K+[n_k-n^g_k(i-1)]^K} > 0$$ and the exponent $K$ is a positive constant (e.g., $K=4$). That is, node $k$ determines its desired model in a probabilistic manner, and the probability that node $k$ maintains its desired target is proportional to the $K$th power of the number of neighbors having the same desired model (see Fig. \[Fig\_13\](b)). Using the above stochastic formulation, we are able to establish agreement on the desired model among the nodes.
\[thm1\] For a connected network starting from an arbitrary initial selection for the desired models vector $\boldsymbol{g}_{i}$ at time $i=-1$, and applying the update rule (\[eq98\])-(\[eq57\]), then all nodes eventually achieve agreement on some desired model, i.e., $$\label{eq127}
\boldsymbol{g}_i(1)=\boldsymbol{g}_i(2)=
\ldots=\boldsymbol{g}_i(N),\quad \text{as $i\rightarrow \infty$.}$$
See Appendix \[appD\].
Although rule (\[eq98\])-(\[eq57\]) ensures agreement on the decision vector, this construction is still not a distributed solution for one subtle reason: nodes need to agree on which index (0 or 1) to use to refer to either model $\{w_0^\circ,w_1^\circ\}$. This task would in principle require the nodes to share some global information. We circumvent this difficulty and develop a distributed solution as follows. Moving forward, we now associate with each node $k$ two local vectors $\{f_k,\boldsymbol{g}_{k,i}\}$; these vectors will play the role of local estimates for the network vectors $\{f,\boldsymbol{g}_i\}$. Each node will then assign the index value of one to its observed model, i.e., each node $k$ sets $f_k(k)=1$. Then, for every $l\in\mathcal{N}_k$, the entry $f_k(l)$ is set to one if it represents the same model as the one observed by node $k$; otherwise, $f_k(l)$ is set to zero. The question remains about how node $k$ knows whether its neighbors have the same observed model as its own (this is discussed in the next section). Here we comment first on how node $k$ adjusts the entries of its vector $\boldsymbol{g}_{k,i-1}$. Indeed, node $k$ knows its desired model value $\boldsymbol{g}_{k,i-1}(k)$ from time $i-1$. To assign the remaining neighborhood entries in the vector $\boldsymbol{g}_{k,i-1}$, the nodes in the neighborhood of node $k$ first exchange their desired model indices with node $k$, that is, they send the information $\{\boldsymbol{g}_{l,i-1}(l),\;l\in
\mathcal{N}_k\}$ to node $k$. However, since $\boldsymbol{g}_{l,i-1}(l)$ from node $l$ is set relative to its $f_l(l)$, node $k$ needs to set $\boldsymbol{g}_{k,i-1}(l)$ based on the value of $f_k(l)$. Specifically, node $k$ will set $g_{k,i-1}(l)$ according to the rule: $$\label{eq31}
\boldsymbol{g}_{k,i-1}(l) =
\begin{cases}
\boldsymbol{g}_{l,i-1}(l), &\text{if $f_k(l)=f_k(k)$}\\
1-\boldsymbol{g}_{l,i-1}(l), &\text{otherwise}
\end{cases}.$$ That is, if node $l$ has the same observed model as node $k$, then node $k$ simply assigns the value of $\boldsymbol{g}_{l,i-1}(l)$ to $\boldsymbol{g}_{k,i-1}(l)$.
In this way, computations that depend on the network vectors $\{f,\boldsymbol{g}_{i}\}$ will be replaced by computations using the local vectors $\{f_k,\boldsymbol{g}_{k,i}\}$. That is, the quantities $\{f(l),g_i(l)\}$ in (\[eq24\])-(\[eq25\]) and (\[eq98\])-(\[eq57\]) are now replaced by $\{f_k(l),\boldsymbol{g}_{k,i}(l)\}$. We verify in the following that using the network vectors $\{f,\boldsymbol{g}_{i}\}$ is equivalent to using the local vectors $\{f_k,\boldsymbol{g}_{k,i}\}$.
It holds that $$\begin{aligned}
\label{eq34}
f(l) \oplus \boldsymbol{g}_i(k) &= f_k(l)\oplus \boldsymbol{g}_{k,i}(k) \\
\boldsymbol{g}_i(l) \oplus \boldsymbol{g}_{i}(k) &=
\boldsymbol{g}_{k,i}(l)\oplus \boldsymbol{g}_{k,i}(k)\label{eq35}\end{aligned}$$ where the symbol $\oplus$ denotes the exclusive-OR operation.
Since the values of $\{f_k(l),\boldsymbol{g}_{l,i}(l),\boldsymbol{g}_{k,i}(l)\}$ are set relative to $f_k(k)$, it holds that $$\begin{aligned}
f(k) \oplus f(l) &= f_k(k)\oplus f_k(l) \\
f(k) \oplus \boldsymbol{g}_i(k) &= f_k(k)\oplus \boldsymbol{g}_{k,i}(k) \\
f(k) \oplus \boldsymbol{g}_i(l) &= f_k(k)\oplus \boldsymbol{g}_{k,i}(l)\end{aligned}$$ Then relations (\[eq34\]) and (\[eq35\]) hold in view of the fact: $$(a \oplus b) \oplus (a \oplus e) = b \oplus e$$ for any $a$, $b$, and $e\in\{0,1\}$.
With these replacements, node $k$ still needs to set the entries $\{f_k(l)\}$ that correspond to its neighbors, i.e., it needs to differentiate between their underlying models and whether their data arise from the same model as node $k$ or not. We propose next a procedure to determine $f_k$ at node $k$ using the available estimates $\{\boldsymbol{w}_{l,i-1},\boldsymbol{\psi}_{l,i}\}$ for $l\in\mathcal{N}_k$.
Model Classification Scheme {#sec1}
===========================
To determine the vector $f_k$, we introduce the belief vector $\boldsymbol{b}_{k,i}$, whose $l$th entry, $\boldsymbol{b}_{k,i}(l)$, will be a measure of the belief by node $k$ that node $l$ has the same observed model. The value of $\boldsymbol{b}_{k,i}(l)$ lies in the range $[0,1]$. The higher the value of $\boldsymbol{b}_{k,i}(l)$ is, the more confidence node $k$ has that node $l$ is subject to the same model as its own model. In the proposed construction, the vector $\boldsymbol{b}_{k,i}$ will be changing over time according to the estimates $\{\boldsymbol{w}_{l,i-1},\boldsymbol{\psi}_{l,i}\}$. Node $k$ will be adjusting $\boldsymbol{b}_{k,i}(l)$ according to the rule: $$\label{eq32}
\boldsymbol{b}_{k,i}(l)=
\begin{cases}
\alpha \boldsymbol{b}_{k,i-1}(l) + (1-\alpha), &\text{to increase belief}\\
\alpha \boldsymbol{b}_{k,i-1}(l), &\text{to decrease belief}
\end{cases}$$ for some positive scalar $\alpha\in(0,1)$, e.g., $\alpha=0.95$. That is, node $k$ increases the belief by combining in a convex manner the previous belief with the value one. Node $k$ then estimates $f_k(l)$ according to the rule: $$\label{eq37}
\hat{\boldsymbol{f}}_{k,i}(l) =
\begin{cases}
1, &\text{if $\boldsymbol{b}_{k,i}(l)\geq 0.5$}\\
0, &\text{otherwise}
\end{cases}$$ where $\hat{\boldsymbol{f}}_{k,i}(l)$ denotes the estimate for $f_k(l)$ at time $i$ and is now a random variable since it will be computed from data realizations. Note that the value of $\hat{\boldsymbol{f}}_{k,i}(l)$ may change over time due to $\boldsymbol{b}_{k,i}(l)$.
Since all nodes have similar processing abilities, it is reasonable to consider the following scenario.
\[ass1\] All nodes in the network use the same step-size, $\mu_k=\mu$, and they observe data arising from the same covariance distribution so that $R_{u,k}=R_u$ for all $k$.
Agents still need to know whether to increase or decrease the belief in (\[eq32\]). We now suggest a procedure that allows the nodes to estimate the vectors $\{f_k\}$ by focusing on their behavior in the *far-field* regime when their weight estimates are usually far from their observed models (see (\[eq4\]) for a more specific description). The far-field regime generally occurs during the initial stages of adaptation and, therefore, the vectors $\{f_k\}$ can be determined quickly during these initial iterations.
To begin with, we refer to the update vector from (\[eq20\]), which can be written as follows for node $l$: $$\begin{aligned}
\boldsymbol{h}_{l,i} &=
\mu^{-1}(\boldsymbol{\psi}_{l,i}-\boldsymbol{w}_{l,i-1})\notag
\\ &=\boldsymbol{u}_{l,i}^T\boldsymbol{u}_{l,i}
(z^\circ_l-\boldsymbol{w}_{l,i-1})+
\boldsymbol{u}^T_{l,i}\boldsymbol{v}_l(i).\label{eq36}\end{aligned}$$ Taking expectation of both sides conditioned on $\boldsymbol{w}_{l,i-1}=w_{l,i-1}$, we have that $$\label{eq21}
\bar{h}_{l,i}\triangleq
\mathbb{E}[\boldsymbol{h}_{l,i}\;|\;\boldsymbol{w}_{l,i-1}=w_{l,i-1}]=
R_{u}(z^\circ_l-w_{l,i-1}).$$ That is, the expected update direction given the previous estimate, $w_{l,i-1}$, is a scaled vector pointing from $w_{l,i-1}$ towards $z^\circ_l$ with scaling matrix $R_{u}$. Note that since $R_{u}$ is positive-definite, then the term $\bar{h}_{l,i}$ lies in the same half plane of the vector $z^\circ_l-w_{l,i-1}$, i.e., $\bar{h}^T_{l,i}(z^\circ_l-w_{l,i-1})>0$. Therefore, the update vector provides useful information about the observed model at node $l$. For example, this term tells us how close the estimate at node $l$ is to its observed model. When the magnitude of $\bar{h}_{l,i}$ is large, or the estimate at node $l$ is far from its observed model $z^\circ_l$, then we say that node $l$ is in a *far-field* regime. On the other hand, when the magnitude of $\bar{h}_{l,i}$ is small, then the estimate $w_{l,i-1}$ is close to $z_l^\circ$ and we say that the node is operating in a *near-field* regime. The vector $\bar{h}_{l,i}$ can be estimated by the first-order recursion: $$\begin{aligned}
\label{eq18}
\hat{\boldsymbol{h}}_{l,i} = (1-\nu) \hat{\boldsymbol{h}}_{l,i-1} +
\nu \mu^{-1}(\boldsymbol{\psi}_{l,i}-\boldsymbol{w}_{l,i-1})\end{aligned}$$ where $\hat{\boldsymbol{h}}_{l,i}$ denotes the estimate for $\bar{h}_{l,i}$ and $\nu$ is a positive step-size. Note that since the value of $\bar{h}_{l,i}$ varies with $w_{l,i-1}$, which is updated using the step-size $\mu$, then the value of $\nu$ should be set large enough compared to $\mu$ (e.g., $\mu=0.005$ and $\nu=0.05$ are used in our simulations) so that recursion (\[eq18\]) can track variations in $\bar{h}_{l,i}$ over time. Moreover, since node $k$ has access to the $\{\boldsymbol{w}_{l,i-1},\boldsymbol{\psi}_{l,i}\}$ if node $l$ is in its neighborhood, node $k$ can compute $\hat{\boldsymbol{h}}_{l,i}$ on its own using (\[eq18\]). In the following, we describe how node $k$ updates the belief $\boldsymbol{b}_{k,i}(l)$ using $\{\hat{\boldsymbol{h}}_{k,i},\hat{\boldsymbol{h}}_{l,i}\}$.
![Illustration of the vectors $\bar{h}_{k,i}$ and $\bar{h}_{l,i}$ (a) when both nodes are in far-field and have the same observed model or (b) different observed models.[]{data-label="Fig_3"}](fig4_classification.eps){width="18em"}
During the initial stage of adaptation, nodes $k$ and $l$ are generally away from their respective observed models and both nodes are therefore in the far-field. This state is characterized by the conditions $$\label{eq4}
\|\hat{\boldsymbol{h}}_{k,i}\| > \eta\quad \text{and}
\quad \|\hat{\boldsymbol{h}}_{l,i}\|> \eta$$ for some threshold $\eta$. If both nodes have the same observed model, then the estimates $\hat{\boldsymbol{h}}_{k,i}$ and $\hat{\boldsymbol{h}}_{l,i}$ are expected to have similar direction towards the observed model (see Fig. \[Fig\_3\](a)). Node $k$ will increase the belief $\boldsymbol{b}_{k,i}(l)$ using (\[eq32\]) if $$\begin{aligned}
\label{eq10}
\hat{\boldsymbol{h}}_{k,i}^T\hat{\boldsymbol{h}}_{l,i}>0.\end{aligned}$$ Otherwise, node $k$ will decrease the belief $\boldsymbol{b}_{k,i}(l)$. That is, when both nodes are in the far-field, then node $k$ increases its belief that node $l$ shares the same observed model when the vectors $\hat{\boldsymbol{h}}_{k,i}$ and $\hat{\boldsymbol{h}}_{l,i}$ lie in the same quadrant. Note that it is possible for node $k$ to increase $\boldsymbol{b}_{k,i}(l)$ even when nodes $k$ and $l$ have distinct models. This is because it is difficult to differentiate between the models during the initial stages of adaptation. This situation is handled by the evolving network dynamics as follows. If node $k$ considers that the data from node $l$ originate from the same model, then node $k$ will use the intermediate estimate $\boldsymbol{\psi}_{l,i}$ from node $l$ in (\[eq15\]). Eventually, from Lemma \[lem3\], the estimates at these nodes get close to a convex combination of the underlying models, which would then enable node $k$ to distinguish between the two models and to decrease the value of $\boldsymbol{b}_{k,i}(l)$. Clearly, for proper resolution, the distance between the models needs to be large enough so that the agents can resolve them. When the models are very close to each other so that resolution is difficult, the estimates at the agents will converge towards a convex combination of the models (which will be also close to the models). Therefore, the belief $\boldsymbol{b}_{k,i}(l)$ is updated according to the following rule: $$\label{eq16}
\boldsymbol{b}_{k,i}(l)=
\begin{cases}
\alpha \boldsymbol{b}_{k,i-1}(l) + (1-\alpha), &\text{if $E_1$}\\
\alpha \boldsymbol{b}_{k,i-1}(l), &\text{if $E_1^c$}
\end{cases}$$ where $E_1$ and $E_1^c$ are the two events described by: $$\begin{aligned}
\label{eq116}
E_1&: \|\hat{\boldsymbol{h}}_{k,i}\|>\eta,
\|\hat{\boldsymbol{h}}_{l,i}\|>\eta,\text{ and }
\hat{\boldsymbol{h}}_{k,i}^T\hat{\boldsymbol{h}}_{l,i}>0\\
E_1^c&: \|\hat{\boldsymbol{h}}_{k,i}\|>\eta,
\|\hat{\boldsymbol{h}}_{l,i}\|>\eta,\text{ and }
\hat{\boldsymbol{h}}_{k,i}^T\hat{\boldsymbol{h}}_{l,i} \leq 0. \label{eq118}\end{aligned}$$ Note that node $k$ updates the belief $b_{k,i}(l)$ only when both nodes $k$ and $l$ are in the far-field.
Diffusion Strategy with Decision-Making {#sec5}
=======================================
Combining the modified diffusion strategy (\[eq6\])-(\[eq15\]), the combination weights (\[eq24\])-(\[eq25\]), the decision-making process (\[eq98\])-(\[eq57\]), and the classification scheme (\[eq37\]) and (\[eq16\]) with $\{f(l),g_i(l)\}$ replaced by $\{\hat{\boldsymbol{f}}_{k,i}(l),\boldsymbol{g}_{k,i}(l)\}$, we arrive at the listing shown in the table. It is seen from the algorithm that the adaptation and combination steps of diffusion, which correspond to steps 1) and 8), are now separated by several steps. The purpose of these intermediate steps is to select the combination weights properly to carry out the aggregation required by step 8). Note that to implement the algorithm, nodes need to exchange the quantities $\{\boldsymbol{w}_{k,i-1},\boldsymbol{\psi}_{k,i},\boldsymbol{g}_{k,i-1}(k)\}$ with their neighbors. We summarize the computational complexity and the amount of scalar exchanges of the conventional and modified diffusion strategies in Table II. Note that the modified strategy still requires in the order of $O(M)$ computations per iteration. Nevertheless, the modified diffusion strategy requires about $2n_kM$ more additions and multiplications than conventional diffusion. This is because of the need to compute the terms $\{\hat{\boldsymbol{h}}_{l,i}\}$ in step 2). If the nodes can afford to exchange extra information, then instead of every node connected to node $l$ computing the term $\hat{\boldsymbol{h}}_{l,i}$ in step 2), this term can be computed locally by node $l$ and shared with its neighbors. This reveals a useful trade-off between complexity and information exchange.
For each node $k$, initialize $\boldsymbol{w}_{k,-1}=0$, $\hat{\boldsymbol{h}}_{k,-1}=0$, $\boldsymbol{b}_{k,-1}(l)=0.5$, and $\boldsymbol{g}_{k,-1}(k)=1$.
1\) Perform an adaptation step using the local data $\{\boldsymbol{d}_k(i), \boldsymbol{u}_{k,i}\}$: $$\begin{aligned}
\boldsymbol{\psi}_{k,i} = \boldsymbol{w}_{k,i-1}+
\mu\boldsymbol{u}_{k,i}^T
[\boldsymbol{d}_k(i)-\boldsymbol{u}_{k,i}\boldsymbol{w}_{k,i-1}].\end{aligned}$$ 2) Exchange the vectors $\{\boldsymbol{\psi}_{k,i},\boldsymbol{w}_{k,i-1}\}$ with neighbors and update the\
$\hspace{1.2em}$average update vectors $\{\hat{\boldsymbol{h}}_{l,i}\}$ for $l\in\mathcal{N}_k$: $$\begin{aligned}
\hat{\boldsymbol{h}}_{l,i}=(1-\nu)\hat{\boldsymbol{h}}_{l,i-1}+
\nu \mu^{-1}(\boldsymbol{\psi}_{l,i}-\boldsymbol{w}_{l,i-1}).\end{aligned}$$ 3) Update the beliefs $\{\boldsymbol{b}_{k,i}(l)\}$ for $l\in\mathcal{N}_k\setminus\{k\}$: $$\boldsymbol{b}_{k,i}(l)=
\begin{cases}
\alpha \boldsymbol{b}_{k,i-1}(l) + (1-\alpha), &\text{if $E_1$}\\
\alpha \boldsymbol{b}_{k,i-1}(l), &\text{if $E_1^c$}
\end{cases}$$ $\hspace{1.2em}$where $E_1$ and $E_1^c$ are defined in (\[eq116\])-(\[eq118\]).
4\) Identify the observed models $\{\hat{\boldsymbol{f}}_{k,i}(l)\}$ for $l\in\mathcal{N}_k\setminus\{k\}$: $$\hat{\boldsymbol{f}}_{k,i}(l)=
\begin{cases}
1, &\text{if $\boldsymbol{b}_{k,i}(l) \geq 0.5$}\\
0, &\text{otherwise}
\end{cases}.$$ 5) Collect the desired models $\{\boldsymbol{g}_{k,i-1}(l)\}$ for $l\in\mathcal{N}_k\setminus\{k\}$ and\
$\hspace{1.2em}$construct the set $\mathcal{N}^g_{k,i-1}$ as follows: $$\begin{aligned}
\boldsymbol{g}_{k,i-1}(l) &=
\begin{cases}
\boldsymbol{g}_{l,i-1}(l), &\text{if $\hat{\boldsymbol{f}}_{k,i}(l) = 1$}\\
1-\boldsymbol{g}_{l,i-1}(l), &\text{otherwise}
\end{cases}\\
\mathcal{N}_{k,i-1}^g &=
\{l\;|\;l\in\mathcal{N}_k,\boldsymbol{g}_{k,i-1}(l)=\boldsymbol{g}_{k,i-1}(k)\}.\end{aligned}$$ 6) Update the desired model $\boldsymbol{g}_{k,i}(k)$: $$\label{eq19}
\boldsymbol{g}_{k,i}(k) = \begin{cases}
\boldsymbol{g}_{k,i-1}(k), &\text{w.p. $q_{k,i-1}$}\\
1-\boldsymbol{g}_{k,i-1}(k), &\text{w.p. $1-q_{k,i-1}$}
\end{cases}$$ $\hspace{1.2em}$where the probability $q_{k,i-1}$ is defined in (\[eq57\]).
7\) Adjust the combination weights $\{a^{(1)}_{l,k}\}$ and $\{a^{(2)}_{l,k}\}$: $$\begin{aligned}
& a^{(1)}_{l,k,i} =
\begin{cases}
a_{l,k}, &\text{if $l\in\mathcal{N}_k$ and
$\hat{\boldsymbol{f}}_{k,i}(l)= \boldsymbol{g}_{k,i}(k)$}\\
0, &\text{otherwise}
\end{cases}\\
& a^{(2)}_{l,k,i} =
\begin{cases}
a_{l,k}, &\text{if $l\in\mathcal{N}_k$ and
$\hat{\boldsymbol{f}}_{k,i}(l)\neq \boldsymbol{g}_{k,i}(k)$}\\
0, &\text{otherwise}
\end{cases}.\end{aligned}$$ 8) Perform the combination step: $$\begin{aligned}
\boldsymbol{w}_{k,i}= \sum_{l\in\mathcal{N}_k}\left(a^{(1)}_{l,k,i}
\boldsymbol{\psi}_{l,i}
+a^{(2)}_{l,k,i}\boldsymbol{w}_{l,i-1}\right).\end{aligned}$$
**Diffusion** (\[eq49\])-(\[eq1\]) **Modified diffusion**
---------------------- ------------------------------------ ------------------------
**Multiplications** $(n_k+2)M$ $(3n_k+2)M+n_k-1$
**Additions** $(n_k+1)M$ $(3n_k+1)M+n_k-1$
**Scalar exchanges** $n_kM$ $n_k(2M+1)$
: [Comparison of the number of multiplications and additions per iteration, as well as the number of scalars that are exchanged for each iteration of the algorithms at every node $k$. In the table, the symbol $n_k$ denotes the degree of node $k$, i.e., the size of its neighborhood $\mathcal{N}_k$.]{}
Due to the dependency among the steps of the algorithm, the analysis of its behavior becomes challenging. However, by examining the various steps, some useful observations stand out. Specifically, it is observed that the convergence of the algorithm occurs in three phases as follows (see also Sec. VIII):
1. Convergence of the classification scheme: The first phase of convergence happens during the initial stages of adaptation. It is natural to expect that during this stage, all weight estimates are generally away from their respective models and the nodes operate in the far-field regime. Then, the nodes use steps 2)-5) to determine the observed models $\{\hat{\boldsymbol{f}}_{k,i}(l)\}$ of their neighbors. We explain later in Eq. (\[eq99\]) in Theorem \[thm5\] that this construction is able to identify the observed models with high probability. In other words, the classification scheme is able to converge reasonably well and fast during the initial stages of adaptation.
2. Convergence of the decision-making process: The second phase of convergence happens right after the convergence of the classification scheme, once the $\{\hat{\boldsymbol{f}}_{k,i}(l)\}$ have converged. Because the nodes now have correct information about their neighbor’s observed models, they use steps 5)-6) to determine their own desired models $\{\boldsymbol{g}_{k,i}(k)\}$. The convergence of this step is ensured by Eq. (\[eq127\]) in Theorem \[thm1\].
3. Convergence of the diffusion strategy: After the classification and decision-making processes converge, the estimates $\{\hat{\boldsymbol{f}}_{k,i}(l),\boldsymbol{g}_{k,i}(l)\}$ remain largely invariant and the combination weights in step 7) therefore remain fixed for all practical purposes. Then, the diffusion strategy becomes unbiased and converges in the mean according to Theorem \[thm2\]. Moreover, when the estimates are close to steady-state, those nodes whose observed models are the same as the desired model enter the near-field regime and they stop updating their belief vectors (this will be justified by the future result (\[eq74\])).
Performance of Classification Procedure {#sec3}
=======================================
It is clear that the success of the diffusion strategy and decision-making process depends on the reliability of the classification scheme in (\[eq37\]) and (\[eq16\]). In this section, we examine the probability of error for the classification scheme under some simplifying conditions to facilitate the analysis. This is a challenging task to pursue due to the stochastic nature of the classification and decision-making process, and due to the coupling among the agents. Our purpose in this section is to gain some insights into this process through a first-order approximate analysis.
Now, there are two types of error. When nodes $k$ and $l$ are subject to the same observed model (i.e., $z^\circ_k=z^\circ_l$ and $f_{k}(l)=1$), then one probability of error is defined as: $$\begin{aligned}
P_{e,1}&=\Pr\left(\hat{\boldsymbol{f}}_{k,i}(l)=0
\;|\;f_{k}(l)=1\right)\notag \\
&=\Pr\left(\boldsymbol{b}_{k,i}(l)<0.5
\;|\;z^\circ_k=z^\circ_l\right)\label{eq77}\end{aligned}$$ where we used rule (\[eq37\]). The second type of probability of error occurs when both nodes have different observed models (i.e., when $z^\circ_k\neq z^\circ_l$ and $f_{k}(l)=0$) and refers to the case: $$\begin{aligned}
P_{e,0}&=\Pr\left(\hat{\boldsymbol{f}}_{k,i}(l) = 1
\;|\;f_{k}(l)=0\right)\notag \\
&=\Pr\left(\boldsymbol{b}_{k,i}(l) >
0.5\;|\;z^\circ_k\neq z^\circ_l\right).\label{eq92}\end{aligned}$$ To evaluate the error probabilities in (\[eq77\])-(\[eq92\]), we examine the probability distribution of the belief variable $\boldsymbol{b}_{k,i}$. Note from (\[eq16\]) that the belief variable can be expressed as: $$\label{eq46}
\boldsymbol{b}_{k,i}(l) = \alpha \boldsymbol{b}_{k,i-1}(l)
+(1-\alpha)\boldsymbol{\xi}_{k,i}(l)$$ where $\boldsymbol{\xi}_{k,i}(l)$ is a Bernoulli random variable with $$\label{eq107}
\boldsymbol{\xi}_{k,i}(l) =
\begin{cases}
1, &\text{with probability $p$}\\
0, &\text{with probability $1-p$}
\end{cases}.$$ The value of $p$ depends on whether the nodes have the same observed models or not. When $z^\circ_k=z^\circ_l$, the belief $\boldsymbol{b}_{k,i}(l)$ is supposed to be increased and the probability of detection, $P_d$, characterizes the probability that $\boldsymbol{b}_{k,i}(l)$ is increased, i.e., $$\label{eq100}
P_d = \Pr(\boldsymbol{\xi}_{k,i}(l)=1
\;|\; z^\circ_k=z^\circ_l).$$ In this case, the probability $p$ in (\[eq107\]) will be replaced by $P_d$. On the other hand, when $z^\circ_k\neq z^\circ_l$, the probability of false alarm, $P_f$, characterizes the probability that the belief $\boldsymbol{b}_{k,i}(l)$ is increased when it is supposed to be decreased, i.e., $$\label{eq101}
P_f = \Pr(\boldsymbol{\xi}_{k,i}(l)=1
\;|\; z^\circ_k\neq z^\circ_l)$$ and we replace $p$ in (\[eq107\]) by $P_f$. We will show later (see Lemma \[lem6\]) how to evaluate the two probabilities $P_d$ and $P_f$. In the sequel we denote them generically by $p$.
Expanding (\[eq46\]), we obtain $$\begin{aligned}
\label{eq71}
\boldsymbol{b}_{k,i}(l)=\alpha^{i+1}\boldsymbol{b}_{k,-1}(l)+
(1-\alpha)\sum_{j=0}^{i}\alpha^j\boldsymbol{\xi}_{k,i-j}(l).\end{aligned}$$ Although it is generally not true, we simplify the analysis by assuming that the $\{\boldsymbol{\xi}_{k,i}(l)\}$ in (\[eq107\]) are independent and identically distributed (i.i.d.) random variables. This assumption is motivated by conditions (\[eq116\])-(\[eq118\]) and by the fact that the type of model that is observed by node $k$ is assumed to be independent from the type of model that is observed by node $l$. The assumption is also motivated by the fact that the regression data and noise across all nodes are assumed to be temporally white and independent over space. Now, since $\alpha$ is a design parameter that is smaller than one, after a few iterations, say, $C$ iterations, the influence of the initial condition in (\[eq71\]) becomes small and can be ignored. In addition, the distribution of $\boldsymbol{b}_{k,i}(l)$ can be approximated by the distribution of the following random variable, which takes the form of a random geometric series: $$\label{eq117}
\boldsymbol{\zeta}_k(l) \triangleq(1-\alpha)
\sum_{j=0}^C\alpha^j\boldsymbol{\xi}_{k,j}(l)$$ where we replaced the index $i-j$ in (\[eq71\]) by $j$ because the $\{\boldsymbol{\xi}_{k,i}(l)\}$ are assumed to be i.i.d. There have been several useful works on the distribution function of random geometric sequences and series [@Hill73; @Smith91; @Bovier93]. However, it is generally untractable to express the distribution function in closed form. We instead resort to the following two inequalities to establish bounds for the error probabilities (\[eq77\])-(\[eq92\]). First, for any two generic events $E_1$ and $E_2$, if $E_1$ implies $E_2$, then the probability of event $E_1$ is less than the probability of event $E_2$ [@Papoulis02], i.e., $$\label{eq60}
\Pr(E_1)\leq \Pr(E_2)\quad\text{if}\quad E_1\subseteq E_2.$$ The second inequality is the Markov inequality [@Papoulis02], i.e., for any nonnegative random variable $\boldsymbol{x}$ and positive scalar $\delta$, it holds that $$\label{eq61}
\Pr(\boldsymbol{x} \geq \delta)=\Pr(\boldsymbol{x}^2 \geq \delta^2)
\leq \frac{\mathbb{E}\boldsymbol{x}^2}{\delta^2}.$$ To apply the Markov inequality (\[eq61\]), we need the second-order moment of $\boldsymbol{\zeta}_k(l)$ in (\[eq117\]), which is difficult to evaluate because the $\{\boldsymbol{\xi}_{k,j}(l)\}$ are not zero mean. To circumvent this difficulty, we introduce the change of variable: $$\boldsymbol{\xi}^\circ_{k,j}(l) \triangleq
\frac{\boldsymbol{\xi}_{k,j}(l)-p}{\sqrt{p(1-p)}}.$$ It can be verified that the $\{\boldsymbol{\xi}^\circ_{k,j}(l)\}$ are i.i.d. with *zero* mean and unit variance. Then, we can write (\[eq117\]) as $$\begin{aligned}
\label{eq93}
\boldsymbol{\zeta}_k(l)
= p\left(1-\alpha^{C+1}\right) + \sqrt{p(1-p)}\boldsymbol{\zeta}^\circ_k(l)\end{aligned}$$ where the variable $\boldsymbol{\zeta}^\circ_k(l)$ is defined by $$\boldsymbol{\zeta}^\circ_k(l) \triangleq (1-\alpha)\sum_{j=0}^C\alpha^j
\boldsymbol{\xi}^\circ_{k,j}(l)$$ and its mean is zero and its variance is given by $$\begin{aligned}
\mathbb{E}(\boldsymbol{\zeta}^\circ_k(l))^2
=\frac{1-\alpha}{1+\alpha}\left(1-\alpha^{2(C+1)}\right)
\approx\frac{1-\alpha}{1+\alpha}.\end{aligned}$$ Then, from (\[eq77\]) and (\[eq93\]) and replacing the probability $p$ by $P_d$ and for $C$ large enough so that $1-\alpha^{C+1}\approx 1$, we obtain that $$\begin{aligned}
P_{e,1}
& \approx \Pr(\boldsymbol{\zeta}_k(l) < 0.5\;|\; z^\circ_k=z^\circ_l) \notag\\
& = \Pr\left(\boldsymbol{\zeta}^\circ_k(l) < \frac{-(P_d-0.5)}
{\sqrt{P_d(1-P_d)}}\;|\;z^\circ_k=z^\circ_l\right) \notag \\
& \leq \Pr\left(|\boldsymbol{\zeta}^\circ_k(l)| > \frac{P_d-0.5}
{\sqrt{P_d(1-P_d)}}\;|\;z^\circ_k=z^\circ_l\right) \notag \\
& \leq \frac{1-\alpha}{1+\alpha}\cdot
\frac{P_d(1-P_d)}{(P_d-0.5)^2}\label{eq78}\end{aligned}$$ where we used (\[eq60\]) and the Markov inequality (\[eq61\]) in the last two inequalities. Note that in (\[eq78\]), we assume the value of $P_d$ to be greater than $0.5$. Indeed, as we will argue in Lemma \[lem6\], the value of $P_d$ is close to one. Similarly, replacing the probability $p$ by $P_f$ and assuming that $P_f<0.5$, we obtain from (\[eq92\]) and (\[eq93\]) that $$\begin{aligned}
P_{e,0}\leq \frac{1-\alpha}{1+\alpha}\cdot
\frac{P_f(1-P_f)}{(0.5-P_f)^2}.\label{eq79}\end{aligned}$$ To evaluate the upper bounds in (\[eq78\])-(\[eq79\]), we need the probabilities of detection and false alarm in (\[eq100\])-(\[eq101\]). Since the update of $\boldsymbol{b}_{k,i}(l)$ in (\[eq16\]) depends on $\{\hat{\boldsymbol{h}}_{k,i},\hat{\boldsymbol{h}}_{l,i}\}$, we need to rely on the statistical properties of these latter quantities. In the following, we first examine the statistics of $\hat{\boldsymbol{h}}_{k,i}$ constructed via (\[eq18\]) and then evaluate $P_d$ and $P_f$ defined by (\[eq100\])-(\[eq101\]).
Statistics of $\hat{\boldsymbol{h}}_{k,i}$
------------------------------------------
We first discuss some assumptions that lead to an approximate model for the evolution of $\hat{\boldsymbol{h}}_{k,i}$ in (\[eq75\]) further ahead. As we mentioned following (\[eq18\]), since the step-sizes $\{\mu,\nu\}$ satisfy $\mu\ll \nu$, the variation of $\boldsymbol{w}_{k,i-1}$ can be assumed to be much slower than the variation of $\hat{\boldsymbol{h}}_{k,i}$. For this reason, the analysis in this section will be conditioned on $\boldsymbol{w}_{k,i-1}=w_{k,i-1}$, as we did in (\[eq21\]), and we introduce the following assumption.
\[ass2\] The step-sizes $\{\mu,\nu\}$ are sufficiently small, i.e., $$\label{eq70}
0 < \mu \ll \nu \ll 1$$ so that $w_{k,i} \approx w_{k,i-1}$ for all $k$.
In addition, since the update vector from (\[eq21\]) depends on the covariance matrix $R_u$, we assume $R_u$ is well-conditioned so that the following is justified.
\[ass5\] The regression covariance matrix $R_{u}$ is well-conditioned such that it holds that $$\begin{aligned}
\label{eq72}
&\text{if }\|z^\circ_k-w_{k,i-1}\| \gg 1\text{, then }
\|\bar{h}_{k,i}\| \gg \eta\\
&\text{if }\|z^\circ_k-w_{k,i-1}\| \ll 1\text{, then }
\|\bar{h}_{k,i}\| \ll \eta.\label{eq73}\end{aligned}$$ Moreover, the fourth-order moment of the regression data $\{\boldsymbol{u}_{k,i}\}$ is assumed to be bounded such that $$\label{eq108}
\nu \tau \ll 1$$ where the scalar $\tau$ is a bound for $$\label{eq102}
\frac{
\mathbb{E}\|\boldsymbol{u}_{k,i}^T\boldsymbol{u}_{k,i}
(z_k^\circ-w_{k,i-1})-\bar{h}_{k,i}\|^2}{\|\bar{h}_{k,i}\|^2}
\leq \tau$$ and its value measures the randomness in variables involving fourth-order products of entries of $\boldsymbol{u}_{k,i}$.
Note that condition (\[eq102\]) can be rewritten as $$\frac{(z_k^\circ-w_{k,i-1})^T\mathbb{E}(\boldsymbol{u}_{k,i}^T\boldsymbol{u}_{k,i}
\boldsymbol{u}_{k,i}^T\boldsymbol{u}_{k,i}-R_u^2)(z_k^\circ-w_{k,i-1})}
{(z_k^\circ-w_{k,i-1})^TR_u^2(z_k^\circ-w_{k,i-1})}\leq \tau$$ which shows that (\[eq102\]) corresponds to a condition on the fourth-order moment of the regression data. Combining conditions (\[eq70\]) and (\[eq108\]), we obtain the following constraint on the step-sizes $\{\mu,\nu\}$: $$\label{eq126}
0 \ll \mu \ll \nu \ll \min\{1,1/\tau\}.$$ To explain more clearly what conditions (\[eq72\])-(\[eq73\]) entail, we obtain from (\[eq21\]) that $\|\bar{h}_{k,i}\|^2$ can be written as the weighted square Euclidean norm: $$\label{eq76}
\|\bar{h}_{k,i}\|^2 = \|z^\circ_k-w_{k,i-1}\|^2_{R_u^2}.$$ We apply the Rayleigh-Ritz characterization of eigenvalues [@Horn85] to conclude that $$\lambda_{\min}(R_u) \cdot\|z^\circ_k-w_{k,i-1}\|
\leq \|\bar{h}_{k,i}\| \leq \lambda_{\max}(R_u) \cdot\|z^\circ_k-w_{k,i-1}\|$$ where $\lambda_{\min}(R_u)$ and $\lambda_{\max}(R_u)$ denote the minimum and maximum eigenvalues of $R_u$. Then, condition (\[eq72\]) indicates that whenever node $k$ is operating in the far-field regime, i.e., whenever $\|z^\circ_k-w_{k,i-1}\|\gg 1$, then we would like $$\lambda_{\min}(R_u) \cdot\|z^\circ_k-w_{k,i-1}\| \gg \eta.$$ Likewise, whenever $\|z_k^\circ-w_{k,i-1}\|\ll 1$, then $$\lambda_{\max}(R_u) \cdot\|z^\circ_k-w_{k,i-1}\| \ll \eta.$$ Therefore, the scalars $\lambda_{\min}(R_u)/\eta$ and $\lambda_{\max}(R_u)/\eta$ cannot be too small or too large, i.e., the matrix $R_u$ should be well-conditioned.
We are now ready to model the average update vector $\hat{\boldsymbol{h}}_{k,i}$. From Assumption \[ass2\], since the estimate $w_{k,i-1}$ remains approximately constant during repeated updates of $\hat{\boldsymbol{h}}_{k,i}$, we first remove the time index in $w_{k,i-1}$ and examine the statistics of $\hat{\boldsymbol{h}}_{k,i}$ under the condition $w_{k,i-1} = w_k$. From (\[eq36\]) and (\[eq18\]), the expected value of $\hat{\boldsymbol{h}}_{k,i}$ given $w_{k,i-1} = w_k$ converges to $$\label{eq63}
\lim_{i\rightarrow \infty}\mathbb{E}\hat{\boldsymbol{h}}_{k,i} =
R_{u}(z_k^\circ-w_k)
\triangleq \bar{h}_k.$$ We can also obtain from (\[eq36\]) and (\[eq18\]) that the limiting second-order moment of $\hat{\boldsymbol{h}}_{k,i}$, which is denoted by $\sigma^2_{\hat{h},k}$, satisfies: $$\begin{aligned}
\label{eq39}
\sigma^2_{\hat{h},k}\triangleq \lim_{i\rightarrow \infty}
\mathbb{E}\|\hat{\boldsymbol{h}}_{k,i}-\bar{h}_k\|^2
= (1-\nu)^2 \sigma^2_{\hat{h},k}+ \nu^2 \sigma^2_{h,k}\end{aligned}$$ where $\sigma^2_{h,k}\triangleq
\mathbb{E}\|\boldsymbol{h}_{k,i}-\bar{h}_k\|^2$ is given by $$\begin{aligned}
\sigma^2_{h,k}=\mathbb{E}\|\boldsymbol{u}_{k,i}^T\boldsymbol{u}_{k,i}
(z_k^\circ-w_k)-\bar{h}_k\|^2+
\sigma^2_{v,k}\text{Tr}(R_{u}). \label{eq33}\end{aligned}$$ Note that the cross term on the right-hand side of (\[eq39\]) is zero because the terms $\hat{\boldsymbol{h}}_{k,i-1}-\bar{h}_k$ and $\boldsymbol{h}_{k,i}-\bar{h}_k$ are independent under the constant $w_k$ condition. Note also that $\boldsymbol{h}_{k,i}-\bar{h}_k$ has zero mean. Then, from (\[eq39\]) and Assumption \[ass2\], the variance $\sigma^2_{\hat{h},k}$ is given by $$\begin{aligned}
\label{eq59}
\sigma^2_{\hat{h},k}=\frac{\nu}{2-\nu}\sigma^2_{h,k}
\approx\frac{\nu}{2}\sigma^2_{h,k}.\end{aligned}$$
Since $w_{k,i-1}$ remains approximately constant, the average update vector $\hat{\boldsymbol{h}}_{k,i}$ has mean and second-order moment close to expressions (\[eq63\]) and (\[eq59\]). We then arrive at the following approximate model for $\hat{\boldsymbol{h}}_{k,i}$.
\[ass6\] The estimate $\hat{\boldsymbol{h}}_{k,i}$ is modeled as: $$\label{eq75}
\hat{\boldsymbol{h}}_{k,i} = \bar{h}_{k,i} + \boldsymbol{n}_{k,i}$$ where $\boldsymbol{n}_{k,i}$ is a random perturbation process with zero mean and $$\label{eq40}
\mathbb{E}\|\boldsymbol{n}_{k,i}\|^2 \leq
\frac{\nu[\tau\|\bar{h}_{k,i}\|^2+\sigma^2_{v,k}\mathrm{Tr}(R_{u})]}{2}$$ with the scalar $\tau$ defined by (\[eq108\]).
Note that since the perturbation $\boldsymbol{n}_{k,i}$ is from the randomness of the regressor and noise processes $\{\boldsymbol{u}_{k,i},\boldsymbol{v}_{k}(i)\}$, then it is reasonable to assume that the $\{\boldsymbol{n}_{k,i}\}$ are independent of each other.
Before we proceed to the probability of detection (\[eq100\]) and the probability of false alarm (\[eq101\]), we note that the update of the belief $\boldsymbol{b}_{k,i}(l)$ happens only when both nodes $k$ and $l$ are in the far-field regime, which is determined by the magnitudes of $\hat{\boldsymbol{h}}_{k,i}$ and $\hat{\boldsymbol{h}}_{l,i}$ being greater than the threshold $\eta$. The following result approximates the probability that a node is classified to be in the far-field.
\[lem5\] Under Assumptions \[ass2\]-\[ass6\], it holds that $$\begin{aligned}
\label{eq62}
\Pr(\|\hat{\boldsymbol{h}}_{k,i}\| > \eta\;|\;
\|z^\circ_k-w_{k,i-1}\| \gg 1) &\geq 1-\frac{\nu \tau}{2} \\
\Pr(\|\hat{\boldsymbol{h}}_{k,i}\| > \eta\;|\;
\|z^\circ_k-w_{k,i-1}\| \ll 1) &\leq
\frac{\nu \sigma^2_{v,k}\mathrm{Tr}(R_u)}{2\eta^2}. \label{eq74}\end{aligned}$$
See Appendix \[appA\].
From Assumptions \[ass2\]-\[ass5\], the probability in (\[eq62\]) is close to one and the probability in (\[eq74\]) is close to zero. Therefore, this approximate analysis suggests that during the initial stages of adaptation, the magnitude of $\{\|\hat{\boldsymbol{h}}_{k,i}\|\}$ successfully determines that the nodes are in the far-field state and they update the belief using rule (\[eq16\]). When the estimates approach steady-state, the nodes whose observed models are the same as the desired model satisfy the condition $\|z^\circ_k-w_{k,i-1}\| \ll 1$ and, therefore, they stop updating their belief vectors in view of (\[eq74\]). On the other hand, when both nodes $k$ and $l$ have observed models that are different from the desired model (and, therefore, their estimates are away from their observed models), they will continue to update their beliefs. The proof in Appendix \[appB\] then establishes the following bounds on $P_d$ and $P_f$.
\[lem6\] Under Assumptions \[ass2\]-\[ass6\] and during the far-field regime (\[eq72\]), the probabilities of detection and false alarm defined by (\[eq100\])-(\[eq101\]) are approximately bounded by $$\begin{aligned}
\label{eq68}
P_d \geq 1-\frac{16\nu \tau}{\pi^2} \quad \text{and}\quad
P_f \leq \frac{16\nu \tau}{\pi^2}.\end{aligned}$$
The above result establishes that the probability of detection is close to one and the probability of false alarm is close to zero in view of $\nu \tau \ll 1$. That is, with high probability, node $k$ will correctly adjust the value of $\boldsymbol{b}_{k,i}(l)$. We then arrive at the following bound for error probabilities in (\[eq77\])-(\[eq92\]).
\[thm5\] Under Assumptions \[ass2\]-\[ass6\] and in the far-field regime (\[eq72\]), the error probabilities $\{P_{e,1},P_{e,0}\}$ are approximately upper bounded by $$\label{eq99}
P_u=\frac{1-\alpha}{1+\alpha}\cdot\frac{16\nu \tau}{\pi^2}
\cdot \frac{1-16\nu \tau/\pi^2}{(1/2-16\nu \tau/\pi^2)^2}
=\mathcal{O}(\nu).$$
Let the function $f(p)$ be defined as $p(1-p)/(p-0.5)^2$. It can be verified that the function $f(p)$ is strictly increasing when $p\in[0,0.5)$ and strictly decreasing when $p\in(0.5,1]$. From Lemma \[lem6\], we conclude that $P_d>0.5$ and $P_f<0.5$. Therefore, an upper bound for $P_{e,1}$ can be obtained by replacing $P_d$ in (\[eq78\]) by the lower bound in (\[eq68\]). Similar arguments apply to the upper bound for $P_{e,0}$.
This result reveals that the $\{P_{e,1},P_{e,0}\}$ are upper bounded by the order of $\nu$. In addition, the upper bound $P_u$ also depends on the value of $\alpha$ used to update the belief in (\[eq16\]). We observe that the larger the value of $\alpha$, the smaller the values of the error probabilities. In simulations, we choose $\nu=0.05$ and $\alpha=0.95$, which will give the upper bound in (\[eq99\]) the value $P_u\approx 0.008\tau <
\nu \tau$. This implies that the classification scheme (\[eq37\]) identifies the observed models with high probability.
Rates of Convergence
====================
There are two rates of convergence to consider for adaptive networks running a decision-making process of the form described in the earlier sections. First, we need to analyze the rate at which the nodes reach an agreement on a desired model (which corresponds to the speed of the decision-making process). Second, we analyze the rate at which the estimates by the nodes converge to the desired model (which corresponds to the speed of the diffusion adaptation).
Convergence Rate of Decision-Making Process
-------------------------------------------
From the proof of Theorem \[thm1\] (see Appendix \[appD\]), the decision-making process can be modeled as a Markov chain with $N+1$ states $\{\chi_i\}$ corresponding to the number of nodes whose desired vectors are $w^\circ_1$. The Markov chain has two absorbing states $\{0,N\}$ and its transition probability matrix $P$ can be written as: $$\label{eq81}
P = \begin{bmatrix}
1 & 0 & 0 \\
b & Q & c \\
0 & 0 & 1
\end{bmatrix}$$ where the matrix $Q$ of size $(N-1)\times(N-1)$ is the transition matrix among the transient states $\{1,2,\cdots,N-1\}$, and the vectors $\{b,c\}$ of size $N-1$ are the transition probabilities from the transient states to the absorbing states. The convergence rate of the decision-making process is then determined by the rate at which, starting at any arbitrary transient state, the Markov chain converges to one of the absorbing states. The argument that follows is meant to show that the rate of convergence of the decision making process improves with the parameter $K$ used in (\[eq57\]); the larger the value of $K$ the faster is the convergence.
To arrive at this conclusion, we first remark that to assess the rate of convergence, we need to compute the $j$th power of $P$ from (\[eq81\]) to find that $$P^j = \begin{bmatrix}
1 & 0 & 0 \\
\bar{b} & Q^j & \bar{c} \\
0 & 0 & 1
\end{bmatrix}$$ where $\{\bar{b},\bar{c}\}$ are two $N\times 1$ vectors. Let the Markov chain start from any arbitrary initial state distribution, $y$, of the form $$y^T = \begin{bmatrix}
0 & y^T_Q & 0
\end{bmatrix}$$ where $y_Q$ is a vector of size $N-1$ and its entries add up to one, i.e., $y_Q^T\mathds{1}_{N-1} = 1$. We shall select $y_Q$ in a manner that enables us to determine how the convergence rate depends on $K$. Thus, note that the state distribution after $j$ transitions becomes $$y^TP^j =
\begin{bmatrix}
y^T_Q\bar{b} & y^T_QQ^j & y^T_Q\bar{c}
\end{bmatrix}.$$ Therefore, the convergence rate is measured by the rate at which the matrix $Q^j$ converges to zero, which is determined by the spectral radius of $Q$. Since $Q$ is the sub-matrix of the transition probability matrix, all entries of $Q$ are nonnegative, then by the Perron-Frobenius Theorem [@Horn85], the vector $y_Q$ can be selected to be the left eigenvector of $Q$ corresponding to the eigenvalue $\rho(Q)$, i.e., $y_Q^T Q = \rho(Q)y_Q^T$. Moreover, from (\[eq123\]), the matrix $Q$ is primitive and, therefore, all entries of $y_Q$ are positive. Furthermore, since the transition probability matrix $P$ is right-stochastic (i.e., $P\mathds{1}_{N+1}=\mathds{1}_{N+1}$), from (\[eq81\]) it holds that $$\label{eq82}
b + c + Q\mathds{1}_{N-1} = \mathds{1}_{N-1}.$$ Pre-multiplying the vector $y_Q$ on both sides of (\[eq82\]), we obtain that the convergence rate of the decision-making process can be determined by $$\begin{aligned}
\rho(Q) = y_Q^TQ\mathds{1}_{N-1}
= 1-y_Q^T(b+c).\label{eq52}\end{aligned}$$ We now determine the value of the vector sum $b+c$. We note from (\[eq121\]) that the transition probabilities $\{p_{n,m}\}$ in $Q$ are determined by the probability $q_{k,i-1}$ from (\[eq57\]), so is the spectral radius of $Q$. We further note from (\[eq57\]) that there is a single parameter $K$ dictating the value of $q_{k,i-1}$. In the following, we examine the dependence of the convergence rate $\rho(Q)$ on the parameter $K$. It is generally challenging to develop the relation because the transition probability $p_{n,m}$ needs to be computed in a compounded way where we need to evaluate the summation of the products of $\{q_{k,i-1}\}$. Nevertheless, some useful insights can be obtained by means of the following approximate argument. Suppose the network size is sufficiently large and that the nodes are uniformly distributed in the spatial domain so that each of the nodes in the network has approximately the same number of neighbors collecting data from model $w_{1}^\circ$; likewise, each of the nodes in the network has approximately the same number of neighbors collecting data from the other model. Suppose that there are $\chi_{i-1}=n$ out of $N$ nodes with desired model $w^\circ_1$, then, on average, node $k$ with $n_k$ neighbors will have $n_kn/N$ neighbors whose desired model is $w^\circ_1$ and have $n_k(1-n/N)$ neighbors whose desired model is $w^\circ_0$. Then, from rule (\[eq17\])-(\[eq57\]), node $k$ chooses $w^\circ_1$ as its desired model with probability $$\begin{aligned}
\label{eq110}
q_n &\triangleq\frac{(n_kn/N)^K}{(n_kn/N)^K+(n_k(N-n)/N)^K} \notag \\
&= \frac{n^K}{n^K+(N-n)^K}\end{aligned}$$ which is independent of the node index $k$ and is denoted by $q_n$. Then, the second summation in (\[eq121\]) can be evaluated in a way that there are $m$ out of $N$ nodes choosing $w^\circ_1$ as their desired model and the remaining $N-m$ nodes choosing $w^\circ_0$, which is equal to $$\label{eq125}
\binom{N}{m}q^m_n(1-q_n)^{N-m}.$$ Note that the probability in (\[eq125\]) depends on $g_{i-1}$ only through its sum, which is equal to $n$. Therefore, the transition probability $p_{n,m}$ in (\[eq121\]) has the same form as (\[eq125\]). To evaluate the spectral radius of $Q$ from (\[eq52\]), we need the value of $p_{n,0}+p_{n,N}$ (i.e., the $n$th entry of $b+c$), which is given by: $$\label{eq51}
p_{n,0}+p_{n,N} = \frac{n^{NK}+(N-n)^{NK}}{(n^K+(N-n)^K)^N}.$$ The following result establishes a monotonicity property of the sum in (\[eq51\]).
\[lem1\] Let $f(x)$ be a function of the form $$f(x) = \frac{a^{Nx}+b^{Nx}}{(a^x+b^x)^N}$$ for some positive scalars $\{a,b,N\}$ with $N>1$. Then, $f(x)$ is a non-decreasing function, i.e., $$f'(x) \geq 0$$ with equality if, and only if, $a=b$.
The proof follows from evaluating $f'(x)$.
Since the spectral radius of $Q$ depends on the value of $K$ in (\[eq17\]), we will index the quantities with the parameter $K$. For example, we denote the spectral radius of $Q$ by $\rho[Q(K)]$. The following result relates the convergence rate of the decision-making process to the parameter $K$.
\[thm4\] The spectral radius $\rho[Q(K)]$ is a strictly decreasing function of $K$ for $N>2$, i.e., $$\label{eq128}
\rho[Q(K+1)] < \rho[Q(K)].$$
From (\[eq52\]), the spectral radius $\rho[Q(K)]$ is given by: $$\rho[Q(K)] = 1 -
\sum_{n=1}^{N-1}y_{Q,n}[p_{n,0}(K)+p_{n,N}(K)]$$ where $y_{Q,n}$ is the $n$th entry of $y_Q$ and the sum inside the brackets is shown in (\[eq51\]). From Lemma \[lem1\], we have that $$\label{eq83}
p_{n,0}(K+1)+p_{n,N}(K+1) \geq p_{n,0}(K)+p_{n,N}(K)$$ with equality if, and only if, $n=N/2$. Therefore, if $N> 2$, there exists $n\in{1,2,\cdots,N-1}$ such that strict inequality holds in (\[eq83\]). Moreover, since the matrix $Q$ is primitive, the $\{y_{Q,n}\}$ are positive and we arrive at (\[eq128\]).
We therefore conclude that to improve the convergence rate of the decision-making process, the nodes should use larger values of $K$. Nevertheless, it may not be beneficial for the network to seek fast convergence during the decision making process because the network (e.g., a fish school) may converge to a bad model (e.g., a food source of poor quality). There exists a trade-off between exploration and exploitation, as in the case of multi-armed bandit problem [@Gittins89]. Such trade-off can be taken into account by introducing some weighting scalar $\beta_{k}(i-1)$ that measures the quality of the desired model of node $k$ at time $i-1$ relative to the other model. The higher values of $\beta_{k}(i-1)$, the better the quality of the model and the higher probability that node $k$ will maintain its desired model. Therefore, node $k$ adjusts the probability $q_{k,i-1}$ from (\[eq57\]) to $$q_{k,i-1} = \frac{\left[\beta_k(i-1)n^g_k(i-1)\right]^K}
{\left[\beta_k(i-1)n^g_k(i-1)\right]^K+
[n_k-n^g_k(i-1)]^K}.$$
Convergence Rate of Diffusion Adaptation {#sec4}
----------------------------------------
Using the arguments in Section VI, we assume in the following that the nodes have achieved agreement on the desired model, say, $w^\circ_q$ as in (\[eq30\]). We know from the proof of Theorem \[thm2\] (see Appendix \[appC\]) that a modified diffusion network is equivalent to a network with a mixture of informed and uninformed nodes, as studied in [@Tu13a]. That is, nodes whose observed model is identical to its desired model ($f(l)=q$) are informed; otherwise they are uninformed. The convergence rate of the learning process specifies the rate at which the mean-square error converges to steady-state. Using the results of [@Tu13a], we can deduce that the convergence rate, denoted by $r$, of the modified diffusion strategy (\[eq6\])-(\[eq15\]) is given by: $$\label{eq85}
r = \left[\rho(\mathcal{B})\right]^2$$ where $\mathcal{B}$ is defined in Table \[tab1\]. Note that the value of $r$ depends on the combination matrix $A$. Under Assumptions \[ass1\]-\[ass2\], it was shown that the convergence rate is bounded by [@Tu13a]: $$\label{eq86}
\left(1-\mu\lambda_{\min}(R_u)\right)^2\leq r< 1.$$ To improve the convergence rate, it is desirable for the nodes to select their combination weights so that the network has lower value of $r$. It was shown in [@Tu13a] that for any connected network, the convergence rate (\[eq85\]) can achieve the lower bound in (\[eq86\]) (namely, the network is able to converge to steady-state at the fastest rate) by selecting the combination matrix $A$ according to the following rules:
1. If there are informed nodes (i.e., nodes with positive step-sizes) in the neighborhood of node $k$, then it will assign positive combination weights to those nodes only.
2. Otherwise, node $k$ will assign positive combination weights to neighbors that are closer (with shorter path) to informed nodes.
However, there are two issues with this construction. First, it is difficult to construct the weights in a distributed manner because rule 2) requires spatial distribution of informed nodes. Second, the constructed combination matrix is not primitive (i.e., Assumption \[ass3\] does not hold) because there are no links from uninformed nodes to informed nodes. Therefore, Theorem \[thm2\] would not apply here. In the following, we first propose a way to select combination weights that approximate rule 2) and then show that the approximate weights ensure mean convergence.
Let $\mathcal{N}_{k}^f$ denote the set of nodes that are in the neighborhood of $k$ and whose observed model is the same as the desired model $w^\circ_q$ (i.e., they are informed neighbors) $$\mathcal{N}_{k}^f = \{l\;|\;l\in\mathcal{N}_k,
f(l)=q\}.$$ Also, let $n^f_k$ denote the number of nodes in the set $\mathcal{N}_{k}^f$. The selection of combination weights is specified based on three types of nodes: informed nodes ($f(k)=q$), uninformed nodes with informed neighbors ($f(k)\neq q$ and $n^f_k\neq 0$), and uninformed nodes without informed neighbors ($f(k)\neq q$ and $n^f_k = 0$). The first two types correspond to rule 1) and their weights can satisfy rule 1) by setting $$\label{eq95}
a_{l,k} = \begin{cases}
1/n^f_k, &\text{if $l\in\mathcal{N}^f_k$}\\
0, &\text{otherwise}
\end{cases}.$$ That is, node $k$ places uniform weights on the informed neighbors and zero weights on the others. The last type of nodes corresponds to rule 2). Since these nodes do not know the distribution of informed nodes, a convenient choice for the approximate weights they can select is for them to place zero weights on themselves and uniform weights on the others, i.e., $$\label{eq96}
a_{l,k} = \begin{cases}
1/(n_k-1), &\text{if $l\in\mathcal{N}_k$ and $l\neq k$}\\
0, &\text{otherwise}
\end{cases}.$$ Note that the weights from (\[eq95\])-(\[eq96\]) can be set in a fully distributed manner and in real-time. To show the mean convergence of the modified diffusion strategy using the combination matrix $A$ constructed from (\[eq95\])-(\[eq96\]), we resort to Theorem 1 from [@Tu13a]. It states that the strategy converges in the mean for sufficiently small step-sizes if for any node $k$, there exists an informed node $l$ and an integer power $j$ such that $$\label{eq97}
[A^j]_{l,k} > 0.$$ Condition (\[eq97\]) is clearly satisfied for the first two types of nodes. For any node belonging to the last type, since the network is connected and from (\[eq96\]), there exists a path with nonzero weight from a node of the second type (uninformed with informed neighbors) to itself. In addition, there exist direct links from informed nodes to the nodes of the second type, condition (\[eq97\]) is also satisfied. This implies that the modified diffusion strategy using the combination weights from (\[eq95\])-(\[eq96\]) converges in the mean.
Simulation Results
==================
We consider a network with 40 nodes randomly connected. The model vectors are set to $w^\circ_0=[5;-5;5;5]$ and $w^\circ_1 =
[5;5;-5;5]$ (i.e. $M=4$). Assume that the first 20 nodes (nodes 1 through 20) observe data originating from model $w^\circ_0$, while the remaining nodes observe data originating from model $w^\circ_1$. The regression covariance matrix $R_u$ is diagonal with each diagonal entry generated uniformly from $[1,2]$. The noise variance at each node is generated uniformly from $[-35,-5]$ dB. The step-sizes are set to $\mu=0.005$, $\nu=0.05$, and $\alpha=0.95$. The threshold $\eta$ is set to $\eta =1$. The network employs the decision-making process with $K=4$ in (\[eq57\]) and the uniform combination rule: $a_{l,k} =1/n_k$ if $l\in\mathcal{N}_k$.
In Fig. \[Fig\_5\], we illustrate the network mean-square deviation (MSD) with respect to the two model vectors over time, i.e., $$\text{MSD}_q(i) = \frac{1}{N}\sum_{k=1}^N\mathbb{E}
\|w^\circ_q-\boldsymbol{w}_{k,i}\|^2$$ for $q=0$ and $q=1$. We compare the conventional ATC diffusion strategy (\[eq49\])-(\[eq1\]) and the modified ATC diffusion strategy (\[eq6\])-(\[eq15\]) with decision-making. We observe the bifurcation in MSD curves of the modified ATC diffusion strategy. Specifically, the MSD curve relative to the model $w^\circ_0$ converges to 23 dB, while the MSD relative to $w^\circ_1$ converges to -50 dB. This illustrates that the nodes using the modified ATC diffusion are able to agree on a model and to converge to steady-state (to model $w^\circ_1$ in this case). We also show in Fig. \[Fig\_6\] the evolution of the beliefs $\{b_{k,i}(l)\}$ for a particular node using the update rule (\[eq16\]). The node has two neighbors observing data that originate from the same model and two neighbors observing data from a different model. We observe that, at the initial stage of adaptation, all beliefs increase. Nevertheless, as time evolves, the node is able to differentiate between the two models and the beliefs for the latter two neighbors decrease. Note that the belief converges to one if a node has the same observed model; otherwise, it converges to zero. This indicates that the classification scheme successfully identifies the observed models of neighboring nodes. On the other hand, for the conventional diffusion strategy, the nodes also converge because the MSD curves in Fig. \[Fig\_5\] remain flat. However, the MSD values are large (about 18 dB). This implies that the nodes converge to a common vector that does not coincide with either of the model vectors.
![Transient network MSD over a network using the conventional diffusion strategy (\[eq49\])-(\[eq1\]) and using the modified diffusion strategy (\[eq6\])-(\[eq15\]). The network with decision-making converges to the model $w^\circ_1$ while the network without decision making converges to a vector that is not identical to either of the model vectors.[]{data-label="Fig_5"}](fig5_MSD1.eps){width="20em"}
![Evolution of beliefs using (\[eq16\]) at a particular node. The node has four neighbors; two of them collect data from the same model while the other two collect data from a different model.[]{data-label="Fig_6"}](fig6_Belief.eps){width="20em"}
We also show the dependence of the convergence rate on the parameter $K$. We compare two modified diffusion strategies using decision-making with $K=1$ and $K=4$ in (\[eq57\]). The network MSD curves for these two strategies are shown in Fig. \[Fig\_8\]. We observe that the MSD curves relative to the model $w^\circ_1$ decrease at the same rate and converge to the same steady-state value. However, there is about 75 shift in time between these curves: the MSD$_{1,i}$ with $K=4$ is 75 time steps ahead of the MSD curve with $K=1$. As the analytical result revealed, the decision-making processes adopting larger values of parameter $K$ achieve agreement at faster rate. We also consider the effect of the combination weights on the convergence rate of the adaptation strategies. Figure \[Fig\_9\] illustrates the modified diffusion strategies with different combination weights: one with the uniform combination rule and the other one with the combination rule in (\[eq95\])-(\[eq96\]). We observe that the diffusion strategy using the proposed rule converges at faster rate with some degradation in steady-state MSD. Note that the trade-off between convergence rate and MSD is also indicated in [@Tu13a].
![Transient network MSD over the modified diffusion strategies (\[eq6\])-(\[eq15\]) with decision-making process for $K=1$ and $K=4$ in (\[eq57\]).[]{data-label="Fig_8"}](fig7_MSD2.eps){width="20em"}
![Transient network MSD over the modified diffusion strategy (\[eq6\])-(\[eq15\]) using the uniform combination rule and the proposed rule (\[eq95\])-(\[eq96\]).[]{data-label="Fig_9"}](fig8_MSD3.eps){width="20em"}
We apply the results of this paper to model the fish schooling behavior in the presence of two food sources (located at $w^\circ_0$ and $w^\circ_1$). It is observed in nature that fish move in a harmonious manner so that they align their motion and keep a safe distance from each other [@Jadbabaie03; @Gazi04; @Olfati06; @Lorenzo11; @Zavlanos11]. We apply the motion control mechanism from [@Tu11a] to model mobile agents. Let $x_{k,i}$ denote the location vector of node $k$ at time $i$. Every node $k$ adjusts its location vector according to the rule: $$\label{eq105}
x_{k,i+1} = x_{k,i}+\Delta t \cdot v_{k,i+1}$$ where $\Delta t$ is a positive time step and $v_{k,i+1}$ is the velocity vector at node $k$, which is set according to the rule: $$\label{eq106}
v_{k,i+1} = \lambda \frac{w_{k,i}-x_{k,i}}{\|w_{k,i}-x_{k,i}\|}
+\beta \sum_{l\in\mathcal{N}_k}c_{l,k}v_{k,i}+\gamma \delta_{k,i}$$ where $\{\lambda,\beta,\gamma\}$ are nonnegative scalars and $\delta_{k,i}$ helps the nodes keep a certain distance $d_s$ to each other and is given by $$\delta_{k,i} = \frac{1}{n_k-1}\sum_{l\in\mathcal{N}_k\setminus \{k\}}
\left(\|x_{l,i}-x_{k,i}\|-d_s\right)\frac{x_{l,i}-x_{k,i}}{\|x_{l,i}-x_{k,i}\|}.$$ The nodes employ the diffusion strategy to estimate the location of food sources. This is achieved as follows. We showed in [@Tu11a] that the distance, $d^\circ_k(i)$, between the target located at $w^\circ$ and a node $k$ located at $x_{k,i}$ can be expressed as the inner product (see Fig. \[Fig\_15\]): $d^\circ_k(i)=u^\circ_{k,i}(w^{\circ}-x_{k,i})$ where $u^\circ_{k,i}$ denotes the unit direction vector pointing to $w^\circ$ from $x_{k,i}$. However, the nodes observe a noisy distance $d_k(i)$ and a noisy direction $u_{k,i}$ to the target, which can be related to $w^\circ$ as follows (the same form as (\[eq2\])): $$\begin{aligned}
\hat{d}_k(i) \triangleq d_k(i)+u_{k,i}x_{k,i}
= u_{k,i}w^\circ+v_{k}(i) \label{eq14}\end{aligned}$$ where $v_k(i)$ is the scalar noise term and its variance is proportional to the distance of node $k$ to the target, i.e., $$\sigma^2_{v,k,i}=\kappa\|w^\circ-\boldsymbol{x}_{k,i}\|^2$$ with $\kappa=0.01$. In simulation, there are two targets located at $w^\circ_0=[10,10]$ and $w^\circ_1=[-10,10]$. The nodes then apply Algorithm in Section \[sec5\] to achieve agreement on a desired target. The simulation results are illustrated in Fig. \[Fig\_14\]. The parameters used in (\[eq105\])-(\[eq106\]) are set to $(\Delta t,\lambda,\beta,\gamma,d_s)=(0.1,0.3,0.7,1,3)$. Initially, there are $40$ nodes uniformly distributed in a $20\times 20$ square area around the origin. There are $20$ nodes collecting data that originate from target $w^\circ_0$ and the remaining $20$ nodes collecting data arising from the other target $w^\circ_1$. In Fig. \[Fig\_14\], nodes that would like to move towards $w^\circ_0$ are shown as blue dots and nodes that would like to move towards $w^\circ_1$ are shown as red circles. We observe that the node achieve agreement on a desired target and get to the target (at $w^\circ_1=[40,-40]$ in this case).
![Distance and direction of the target $w^\circ$ from node $k$ at location $x_k$. The unit direction vector $u_k^\circ$ points towards $w^\circ$.[]{data-label="Fig_15"}](fig9_data_model.eps){width="16em"}
![Maneuver of fish schools with two food sources over time (a) $t=0.5$ (b) $t=1$ (c) $t=2$ and (d) $t=20$ sec.[]{data-label="Fig_14"}](fig10_Motion2.eps){width="24em"}
Concluding Remarks
==================
In the presence of distinct models observed by the nodes in a network, conventional distributed estimation strategies will lead to biased solutions. In this paper, we proposed a modified strategy to address this issue. To do so, we allow the nodes to exchange not only intermediate estimates, but also previous estimates. We also developed a classification scheme and a decision-making procedure for the nodes to identify the underlying models that generate data and to achieve agreement among the nodes on the desired objective. It is useful to comment on some features of the proposed framework.
We focused in this work on the case where nodes need to choose between two models. Extension of the techniques to multiple models require additional analysis. The case of two models is not a serious limitation especially since many hypothesis testing problems tend to be formulated as deciding between two choices. In addition, it is natural to expect that convergence of the decision process will occur towards one model or the other in a probabilistic manner since the outcome is influenced by the fraction of nodes that sense data from one model or another. Interestingly, though, the decision-making process and the estimation task are largely independent of each other. This is because there are two tasks that the nodes need to accomplish. First, they need to decide which of the two models to follow and, second, they need to estimate the model. To solve the first task, agents do not need to know the exact model values. An arbitrary node $k$ only needs to know whether a neighboring node $l$ is observing data from the same model or from a different model regardless of the model values. This property enables the initial decision process to converge faster and to be largely independent of the estimation task.
Proof of Theorem \[thm2\] {#appC}
=========================
Without loss of generality, let $w^\circ_0$ be the desired model for the network (i.e., $q=0$ in (\[eq30\])) and assume there are $N_0$ nodes with indices $\{1,2,\ldots,N_0\}$ observing data arising from the model $w^\circ_0$, while the remaining $N-N_0$ nodes observe data arising from model $w^\circ_1$. Then, we obtain from (\[eq29\]), (\[eq24\]), and (\[eq25\]) that $$\begin{aligned}
\tilde{z}^\circ_k &=
\begin{cases}
0, &\text{if $k\leq N_0$}\\
w^\circ_0-w^\circ_1, &\text{if $k>N_0$}
\end{cases}\\
a^{(1)}_{l,k} &= 0 \text{ if $l>N_0$} \quad\text{and}\quad
a^{(2)}_{l,k} = 0 \text{ if $l\leq N_0$}.\end{aligned}$$ Since the matrix $\mathcal{M}\mathcal{R}$ is block diagonal, we conclude that $$y=0\quad\text{and}\quad
\mathcal{B} = \mathcal{A}^T(I_{NM}-\mathcal{M}_e\mathcal{R})$$ where $\mathcal{M}_e$ is an $N\times N$ block diagonal matrix of the form $$\label{eq84}
\mathcal{M}_e \triangleq \text{diag}
\{\mu_1I_M,\cdots,\mu_{N_0}I_M,0,\cdots,0\}.$$ That is, its mean recursion in (\[eq5\]) is equivalent to the mean recursion of a network running the traditional diffusion strategy (\[eq49\])-(\[eq1\]) with $N_0$ nodes (nodes 1 to $N_0$) using positive step-sizes and $N-N_0$ nodes (nodes $N_0+1$ to $N$) having zero step-sizes. Then, according to Theorem 1 of [@Tu13a] and under the assumption that the matrix $A$ is primitive, if the step-sizes $\{\mu_1,\mu_2,\cdots,\mu_{N_0}\}$ are set to satisfy (\[eq26\]), then the spectral radius of $\mathcal{B}$ will be strictly less than one.
Proof of Theorem \[thm1\] {#appD}
=========================
For a given vector $\boldsymbol{g}_{i-1}$, we denote by $\chi_{i-1}$ the number of nodes whose desired model is $w^\circ_1$ at time $i-1$, i.e., $$\label{eq103}
\chi_{i-1} \triangleq \sum_{k=1}^N \boldsymbol{g}_{i-1}(k).$$ From (\[eq98\])-(\[eq57\]), the vector $\boldsymbol{g}_{i}$ depends only on $\boldsymbol{g}_{i-1}$. Thus, the value of $\chi_{i}$ depends only on $\chi_{i-1}$. Therefore, the evolution of $\chi_i$ forms a Markov chain with $N+1$ states corresponding to the values $\{0,1,2,\ldots,N\}$ for $\chi_i$. To compute the transition probability, $p_{n,m}$, from state $\chi_{i-1}=n$ to state $\chi_i=m$, let us denote by $\mathcal{G}_{n}$ the set of vectors $g=\{g(1),g(2),\cdots,g(N)\}$ whose entries are either 1 or 0 and add up to $n$, i.e., $$\label{eq120}
\mathcal{G}_{n} = \left\{g \;|\;
\sum_{k=1}^N g(k)=n\right\}.$$ Then, the $p_{n,m}$ can be written as: $$\begin{aligned}
\label{eq121}
p_{n,m}=\sum_{\boldsymbol{g}_{i-1}\in\mathcal{G}_{n}}
\Pr(\boldsymbol{g}_{i-1})\sum_{\boldsymbol{g}_i\in\mathcal{G}_{m}}
\prod_{l=1}^N
\Pr(\boldsymbol{g}_{i}(l)\;|\; \boldsymbol{g}_{i-1}(l))\end{aligned}$$ where $\Pr(\boldsymbol{g}_{i-1})$ is *a priori* probability and where the probability $\Pr(\boldsymbol{g}_{i}(l)\;|\;
\boldsymbol{g}_{i-1}(l))$ is determined by (\[eq57\]). Note that for a static network, the transition probability $p_{n,m}$ is independent of $i$, i.e., the Markov chain is *homogeneous* [@Lawler06].
Now we assume that $\chi_{i-1}=n\neq0,N$. Since the network is connected, for any $\boldsymbol{g}_{i-1}\in\mathcal{G}_{n}$ at least one node (say, node $k$) has desired model $w^\circ_1$ and has a neighbor with distinct desired model $w_0^\circ$ so that $n^g_{k}(i-1) < n_{k}$ and $1-q_{k,i-1} > 0$ from (\[eq57\]). Since $q_{l,i-1}>0$ for all $l$, we obtain from (\[eq121\]) that $$\begin{aligned}
p_{n,n-1}
&\geq \sum_{\boldsymbol{g}_{i-1}\in\mathcal{G}_{n}}\Pr(\boldsymbol{g}_{i-1})
(1-q_{k,i-1}) \prod_{l\neq k}q_{l,i-1}
>0\notag \\
p_{n,n} &>0\quad\text{and}\quad p_{n,n+1}>0 \label{eq123}\end{aligned}$$ for $n\neq0,N$. When $n=0$ or $n=N$, we have that $p_{0,0}=p_{N,N}=1$. This indicates that the Markov chain has two absorbing states: $\chi_i=0$ (or, $\boldsymbol{g}_i(1)=\boldsymbol{g}_i(2)=\cdots=\boldsymbol{g}_i(N)=0$) and $\chi_i=N$ (or, $\boldsymbol{g}_i(1)=\boldsymbol{g}_i(2)=\cdots=\boldsymbol{g}_i(N)=1$), and for any state $\chi_i$ different from $0$ and $N$, there is a nonzero probability traveling from an arbitrary state $\chi_i$ to state $0$ and state $N$. Therefore, no matter which state the Markov chain starts from, it converges to state $0$ or state $N$ [@Lawler06 p.26], i.e., all nodes reach agreement on the desired model.
Proof of Lemma \[lem5\] {#appA}
=======================
Let $C_1$ denote the far-field condition: $\|z^\circ_k-w_{k,i-1}\|
\gg 1$. We obtain from Assumption \[ass5\] and (\[eq75\]) that $$\begin{aligned}
\Pr(\|\hat{\boldsymbol{h}}_{k,i}\| > \eta\;|\; C_1)
&\overset{(a)}{\geq} \Pr(\|\bar{h}_{k,i}\|-\|\boldsymbol{n}_{k,i}\| > \eta\;|\;C_1)\notag \\
&= 1- \Pr(\|\boldsymbol{n}_{k,i}\| \geq \|\bar{h}_{k,i}\|-\eta\;|\;C_1)\notag \\
&\overset{(b)}{\geq} 1- \frac{\mathbb{E}\|\boldsymbol{n}_{k,i}\|^2}
{(\|\bar{h}_{k,i}\|-\eta)^2}\notag \\
&\overset{(c)}{=} 1- \frac{\nu[\tau\|\bar{h}_{k,i}\|^2+\sigma^2_{v,k}\text{Tr}(R_u)]}
{2(\|\bar{h}_{k,i}\|-\eta)^2} \label{eq41}\end{aligned}$$ where step (a) follows from the triangle inequality of norms and (\[eq60\]), step (b) is by the Markov inequality (\[eq61\]) and Assumption \[ass5\], and step (c) is by (\[eq40\]). Moreover, under conditions (\[eq72\]) and $C_1$, we can ignore the term $\eta$ in the denominator of (\[eq41\]). In addition, from condition $C_1$ and (\[eq76\]), and since the variance $\sigma^2_{v,k}$ is generally small, we may ignore the term $\nu\sigma^2_{v,k}\text{Tr}(R_u)$ in (\[eq41\]) and obtain (\[eq62\]). Similar arguments apply to (\[eq74\]).
Proof of Lemma \[lem6\] {#appB}
=======================
Under condition (\[eq72\]) and from (\[eq16\]), the probability $P_d$ in (\[eq100\]) becomes $$\begin{aligned}
P_d
&= \Pr(\|\hat{\boldsymbol{h}}_{k,i}\| > \eta,\;
\|\hat{\boldsymbol{h}}_{l,i}\| > \eta,\;
\hat{\boldsymbol{h}}_{k,i}^T\hat{\boldsymbol{h}}_{l,i}>0
\;|\;z^\circ_k=z^\circ_l)\notag \\
&\approx \Pr(\hat{\boldsymbol{h}}_{k,i}^T\hat{\boldsymbol{h}}_{l,i}>0
\;|\;z^\circ_k=z^\circ_l)\label{eq45}\end{aligned}$$ where we used the fact that $\hat{\boldsymbol{h}}_{k,i}$ and $\hat{\boldsymbol{h}}_{l,i}$ are independent, as well as the result of Lemma \[lem5\] which ensures that $\|\hat{\boldsymbol{h}}_{k,i}\|>\eta$ with high probability (likewise, for the norm of $\hat{\boldsymbol{h}}_{l,i}$). Note that the event $\hat{\boldsymbol{h}}_{k,i}^T\hat{\boldsymbol{h}}_{l,i}>0$ is equivalent to the fact that the angle between these two vectors is less than $\pi/2$. Let $\boldsymbol{\theta}_{k,i}$ denote the angle between the vectors $\bar{h}_{k,i}$ and $\hat{\boldsymbol{h}}_{k,i}$ due to the noise $\boldsymbol{n}_{k,i}$ (see Fig. \[Fig\_10\](a)). The value of $\boldsymbol{\theta}_{k,i}$ is positive if the vector $\hat{\boldsymbol{h}}_{k,i}$ rotates counter-clockwise relative to $\bar{h}_{k,i}$; otherwise, its value is negative. Then, we have that the angle $\boldsymbol{\theta}_{k,i}$ is upper bounded by (see Fig. \[Fig\_10\](a)): $$\label{eq42}
|\boldsymbol{\theta}_{k,i}|\leq
\sin^{-1}\left(\frac{\|\boldsymbol{n}_{k,i}\|}
{\|\bar{h}_{k,i}\|}\right)
\approx
\frac{\|\boldsymbol{n}_{k,i}\|}
{\|\bar{h}_{k,i}\|}$$ That is, the maximum value of $\boldsymbol{\theta}_{k,i}$ occurs when the vectors $\hat{\boldsymbol{h}}_{k,i}$ and $\boldsymbol{n}_{k,i}$ are perpendicular. The approximation in (\[eq42\]) is from (\[eq72\]) and (\[eq40\]) so that it holds that $$\label{eq66}
\frac{\mathbb{E}\|\boldsymbol{n}_{k,i}\|^2}{\|\bar{h}_{k,i}\|^2}
\leq \frac{\nu[\tau\|\bar{h}_{k,i}\|^2+\sigma^2_{v,k}\text{Tr}(R_u)]}
{2\|\bar{h}_{k,i}\|^2}
\approx \frac{\nu \tau}{2}.$$ Since all nodes start from the same initial estimate (i.e., $w_{k,-1}=0$ for all $k$), the estimates $\{\boldsymbol{w}_{k,i-1}\}$ are close to each other during the initial stages of adaptation and it is reasonable to assume that $\|\boldsymbol{w}_{k,i-1}-\boldsymbol{w}_{l,i-1}\| \ll
\|z^\circ_k-\boldsymbol{w}_{k,i-1}\|$. Therefore, we arrive at the approximation $\bar{h}_{l,i}\approx \bar{h}_{k,i}$ for computing $P_d$. This implies that the vectors $\hat{\boldsymbol{h}}_{k,i}$ and $\hat{\boldsymbol{h}}_{l,i}$ can be modeled as starting approximately at the same location $\boldsymbol{w}_{k,i-1}$ but having deviated by angles $\boldsymbol{\theta}_{k,i}$ and $\boldsymbol{\theta}_{l,i}$, respectively (see Fig. \[Fig\_10\](b)). Therefore, the angle between $\hat{\boldsymbol{h}}_{k,i}$ and $\hat{\boldsymbol{h}}_{l,i}$ is equal to $|\boldsymbol{\theta}_{k,i}-\boldsymbol{\theta}_{l,i}|$. From (\[eq45\]), we obtain that $$\begin{aligned}
P_d &\approx \Pr\left(|\boldsymbol{\theta}_{k,i}-\boldsymbol{\theta}_{l,i}| <
\frac{\pi}{2}\;|\; z^\circ_k=z^\circ_l\right)\notag \\
&\overset{(a)}{\geq} \Pr\left(|\boldsymbol{\theta}_{k,i}|+|\boldsymbol{\theta}_{l,i}| <
\frac{\pi}{2}\;|\; z^\circ_k=z^\circ_l\right)\notag \\
&\overset{(b)}{\geq} \Pr\left(\frac{\|\boldsymbol{n}_{k,i}\|}{\|\bar{h}_{k,i}\|}
+\frac{\|\boldsymbol{n}_{l,i}\|}{\|\bar{h}_{k,i}\|} <
\frac{\pi}{2}\right)\notag \\
&= 1- \Pr(\|\boldsymbol{n}_{k,i}\|+\|\boldsymbol{n}_{l,i}\| \geq
\pi\|\bar{h}_{k,i}\|/2) \label{eq65}\end{aligned}$$ where step (a) is by the triangle inequality of norms and (\[eq60\]) and step (b) is by (\[eq42\]). To evaluate the probability in (\[eq65\]), we resort to the following fact. For any two random variables $\boldsymbol{x}$ and $\boldsymbol{y}$ and for any constant $\eta$, it holds from (\[eq60\]) that $$\begin{aligned}
\Pr(\boldsymbol{x}+\boldsymbol{y}>\eta)
\leq \Pr(\boldsymbol{x} >
\eta/2)+\Pr(\boldsymbol{y}>\eta/2).\label{eq80}\end{aligned}$$ This leads to $$P_d \geq 1-\Pr(\|\boldsymbol{n}_{k,i}\| > \pi\|\bar{h}_{k,i}\|/4)
-\Pr(\|\boldsymbol{n}_{l,i}\| > \pi\|\bar{h}_{k,i}\|/4).$$ We then arrive at (\[eq68\]) because $$\begin{aligned}
P_d\geq 1- \frac{16(\mathbb{E}\|\boldsymbol{n}_{k,i}\|^2+
\mathbb{E}\|\boldsymbol{n}_{l,i}\|^2)}{\pi^2\|\bar{h}_{k,i}\|^2}
\geq 1-\frac{16\nu \tau}{\pi^2}\end{aligned}$$ where we used the Markov inequality (\[eq61\]) and (\[eq66\]). Similar arguments apply to $P_f$ when $z^\circ_k\neq z^\circ_l$ by noting that the vectors $\bar{h}_{k,i}$ and $\bar{h}_{l,i}$ can again be modeled as starting approximately at the same location $w_{k,i-1}$, but pointing towards different directions: $\bar{h}_{k,i}$ towards $z_k$ and $\bar{h}_{l,i}$ towards $z_l$, and the angle between these two vectors now assumes a value close to $\pi$ according to Lemma \[lem3\].
![Illustration of (a) the angle $\boldsymbol{\theta}_{k,i}$ between $\bar{h}_{k,i}$ and $\hat{\boldsymbol{h}}_{k,i}$ due to the noise $\boldsymbol{n}_{k,i}$ and (b) the angle between $\hat{\boldsymbol{h}}_{k,i}$ and $\hat{\boldsymbol{h}}_{l,i}$ when $z^\circ_k=z^\circ_l$.[]{data-label="Fig_10"}](fig11_angle_deviation3.eps){width="22em"}
[^1]: Copyright (c) 2013 IEEE. Personal use of this material is permitted. However, permission to use this material for any other purposes must be obtained from the IEEE by sending a request to [email protected].
This work was supported in part by NSF grant CCF-1011918. An earlier conference version of this work appeared in [@Tu12c]. The authors are with the Department of Electrical Engineering, University of California, Los Angeles (e-mail: [email protected]; [email protected]).
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We analyze the existence and stability of nonlinear localized waves described by the Kronig-Penney model with a nonlinear impurity. We study the properties of such waves in a homogeneous medium, and then analyze new effects introduced by periodicity of the medium parameters. In particular, we demonstrate the existence of a novel type of stable nonlinear band-gap localized states, and also reveal an important physical mechanism of the oscillatory wave instabilities associated with the band-gap wave resonances.'
author:
- 'Andrey A. Sukhorukov and Yuri S. Kivshar'
---
impurity, nonlinearity, linear stability, superlattice, band-gap materials, photonic crystals
Introduction
============
Wave propagation in periodic media is associated with many interesting physical phenomena [@yeh; @book]. Modern technology allows to create different kinds of macro- and mesoscopic periodic and layered structures such as semiconductor superlattices and heterostructures, magnetic multilayers possessing the giant magnetoresistance, multiple-quantum-well structures, optical waveguide arrays, photonic band-gap materials and photonic crystal fibers, etc. The main feature of different periodic structures (which follows from the classical Floquet-Bloch theory) is the existence of forbidden frequency band gaps (or stop bands) where [*linear waves undergo Bragg reflection*]{} from the periodic structure [@yeh]. However, many of the recently fabricated periodic structures exhibit pronounced nonlinear properties that give rise to qualitatively new physical effects such as [*multistability*]{} of a finite nonlinear periodic medium [@multi] and energy localization in the form of [*gap solitons*]{} [@gap]. Such effects are usually analyzed in the framework of the coupled-mode theory [@coupled], and they are associated with the nonlinearity-dependent tuning of the stop-band as the wave intensity is increased.
Another fundamental reason for the wave localization is the presence of impurities. As a matter of fact, artificially introduced inhomogeneities are often used as a powerful means of controlling wave scattering. However, the impurity-induced localization in [*nonlinear and periodic media*]{} is largely an open area of research. We stress that the impurity response is determined by the local field amplitude, and therefore the standard averaging procedure can not be directly applied to this kind of problems.
Recent experiments, e.g. the observation of optical gap solitons [@gap_exp] and the control of coherent matter waves in optical lattices [@kasevich], as well as theoretical results such as the discovery of the oscillatory instability of gap solitons [@gap_inst], call for a systematic analysis of nonlinear effects in periodic structures and band-gap localized states [*beyond the approximation provided by the coupled-mode theory*]{}. As we show below, such an analysis is crucially important for determining [*stability of nonlinear waves in periodic structures*]{} because the wave instabilities can appear due to the mode coupling to other bands.
In this paper, we analyze [*the existence and stability of nonlinear localized waves in a periodic medium with multiple gaps in the transmission spectrum*]{} that is valid beyond the coupled-mode theory. We consider a simple model where waves are localized in a layered medium by an intensity-dependent impurity (or, in other words, they are guided by a thin-layer nonlinear waveguide). Assuming the applicability of our results to a variety of different physical systems (see below), we characterize, in the framework of a unified and systematic approach, the properties of [*two types of nonlinear localized waves*]{} [@ref]. In particular, we reveal an important physical mechanism of wave instability associated with the band-gap resonances. We demonstrate also that several types of [*stable band-gap localized states*]{} can exist in the presence of nonlinearity.
General analysis
================
Model equations
---------------
We consider a general problem in which the dynamics of elementary excitations of a physical system (e.g., phonon, magnons, etc.) is described by an effective equation for the wave-packet envelope $\psi (x,t)$ [@kosevich]. When the density of such quasi-particles becomes high enough, their interaction should be taken into account, e.g. in the framework of the mean-field approximation. In the simplest case, the quasi-particle interaction and collective phenomena in an inhomogeneous medium can be described by the nonlinear Schrödinger (NLS) equation for the wave-packet envelope $\psi (x,t)$, $$\label{eq:nls}
i \frac{\partial \psi}{\partial t}
+ \frac{\partial^2 \psi}{\partial x^2}
+ {\cal F}(I; x) \psi = 0,$$ where $I \equiv |\psi|^2$ is the wave intensity, $t$ is time (or propagation variable), $x$ is the spatial coordinate, and the real function ${\cal F}(I; x)$ describes both [*nonlinear*]{} and [*periodic*]{} properties of the medium. We note that the system (\[eq:nls\]) is Hamiltonian, and for localized solutions the power $P = \int_{-\infty}^{+\infty} I(x)\; dx$ is finite, and it is a conserved quantity.
We assume that the superlattice is linear, and nonlinearity appears only through the intensity dependent response of an embedded localized impurity. Then, if the corresponding width of the wave envelope is much larger than that of the impurity, the inhomogeneity can be modeled by a delta-function and, in the simplest case, we can write $$\label{eq:def_F}
{\cal F}(I; x) = \nu(x) + \delta(x) G(I),$$ where the function $G(I)$ characterizes the properties of the impurity, and $\nu(x) \equiv \nu(x+h)$ describes an effective potential of the superlattice with the spatial period $h$.
Model (\[eq:nls\]),(\[eq:def\_F\]) appears in different physical problems of the macroscopic nonlinear dynamics of solids and nonlinear optical systems. In particular, it describes a special case of a more general problem of the existence and stability of nonlinear guided waves in a layered (or stratified) dielectric medium, where the delta-function impurity corresponds to a very thin layer embedded into a nonlinear medium with a Kerr or non-Kerr response [@optics]; in this case the time variable $t$ stands for the propagation coordinate along the layer, and $x$ is the transverse coordinate. Model (\[eq:nls\]),(\[eq:def\_F\]) can also appear in the problem of the Bose-Einstein condensation in optical lattices [@kasevich].
Localized modes
---------------
We study stationary localized solutions of Eq. (\[eq:nls\]) in the form $\psi(x,t) = u(x) e^{i \omega t}$, where $\omega$ is the normalized frequency (or the propagation constant, in optics), and the real function $u(x)$ satisfies the equation: $$\label{eq:u0_inh}
- \omega u + \frac{d^2 u}{d x^2} + {\cal F}(I; x) u = 0.$$ For a local nonlinearity described by Eq. (\[eq:def\_F\]), the localized waves can be constructed with the help of a matching condition, using the solutions of Eq. (\[eq:u0\_inh\]) with ${\cal F}(I; x) = \nu(x)$, presented in the form of the Bloch-type wave functions [@yeh].
If the effective periodic potential $\nu(x)$ is approximated by a piecewise-constant function (as in the case of the so-called Kronig-Penney model), the solution on both sides of the impurity can be decomposed into a pair of counter-propagating waves with the amplitudes $a(x)$ and $b(x)$, $$u_b(x) = a(x) e^{- {\mu}(x) x}
+ b(x) e^{{\mu}(x) x},$$ where $\mu(x) = \sqrt{\omega - \nu(x)}$ is the local wavenumber. As follows from the Floquet-Bloch theory, for a Bloch-wave solution the reflection coefficient $r(x) = b(x) / a(x)$ is a periodic function, i.e. $r(x) = r(x + h)$, and it is found from the following eigenvalue problem: $$\label{eq:eigen}
T(x) \left( \begin{array}{l} 1 \\ r(x) \end{array} \right)
=
\tau(x)
\left( \begin{array}{l} 1 \\ r(x) \end{array} \right) ,$$ where $T(x)$ is a [*transfer matrix*]{} that describes a change of the wave amplitudes $\{a,b\}$ after one period $(x,x+h)$. An explicit expression for the transfer matrix can be found in Ref. [@our_pre], where it was also proven that ${\rm det}\;T \equiv 1$, and two linearly independent solutions of Eq. (\[eq:eigen\]) correspond to a pair of the eigenvalues $\tau$ and $\tau^{-1}$. Relation $\tau(\omega)$ determines a [*band-gap structure*]{} of the superlattice spectrum: the waves are [*propagating*]{}, if $|\tau| = 1$, and they are [*localized*]{}, if $|\tau| \ne 1$. In the latter case, a nonlinear impurity can support [*nonlinear localized waves*]{}, and the wave amplitude at the impurity is determined from the continuity condition [@our_pre] at $x=0$ \[$I_0 \equiv I(0)$\]: $$\label{eq:zeta}
G_0 \equiv G(I_0) = \zeta(\omega) .$$ Here $\zeta = (\zeta^+ + \zeta^-)_{x=0}$, ${\zeta^\pm} = \mu^\pm {\left( 1 - r^\pm \right)}
{\left( 1 + r^\pm \right)}^{-1}$, and we denote with “$+$” and “$-$” the lattice characteristics on the right and the left side of the nonlinear impurity, respectively, i.e. $\nu(x) = \nu^+(|x|)$, for $x>0$, and $\nu(x) = \nu^-(|x|)$, for $x<0$. Note that the reflection coefficients should correspond to the Bloch-wave solutions with the asymptotics $|u( x \rightarrow \pm \infty )| \rightarrow 0$.
Relation (\[eq:zeta\]) allows us to identify different types of nonlinear localized states. We notice that such localized states, being supported by an attractive nonlinear impurity ($G_0 > 0$), exist in the so-called [*waveguiding*]{} regime, when $\zeta(\omega) > 0$. Additionally, localization can occur at a repulsive nonlinear impurity ($G_0 < 0$) in the [*anti-waveguiding*]{} regime, provided $\zeta(\omega) < 0$.
Stability properties
--------------------
To study [*the linear stability*]{}, we consider the evolution of small-amplitude perturbations of the localized state presenting the solution in the form $$\psi (x,t)
= \left\{ u(x) + v(x) e^{i \gamma t}
+ w^{\ast}(x) e^{-i \gamma^{\ast} t}
\right\} e^{i \omega t} ,$$ and obtain the linear eigenvalue problem for $v(x)$ and $w(x)$, $$\label{eq:Leigen}
\begin{array}{l}
{ \displaystyle
-(\omega+\gamma) v + \frac{d^2 v}{d x^2} + \nu(x) v
+ \delta(x) \left[ G_1 v + (G_1 - G_0) w \right] = 0 ,
} \\*[9pt] { \displaystyle
-(\omega-\gamma) w + \frac{d^2 w}{d x^2} + \nu(x) w
+ \delta(x) \left[ G_1 w + (G_1 - G_0) v \right] = 0 ,
} \end{array}$$ where $G_1 \equiv G_0 + I_0 G^{\prime}(I_0)$, the intensity $I_0$ is calculated for an unperturbed solution, and the prime stands for the derivative.
We find that the localized eigenmode solutions of Eq. (\[eq:Leigen\]) exist only for the particular eigenvalues satisfying the solvability condition, $Y( \gamma ) = 0$, where $Y( \gamma )$ is the so-called [*Evans function*]{}, $$Y( \gamma ) =
\left[ G_1 - \zeta(\omega+\gamma) \right]
\left[ G_1 - \zeta(\omega-\gamma) \right]
- \left( G_1 - G_0 \right)^2 .$$ In general, the localized eigenmodes fall into one of the following categories: (i) [*internal modes*]{} with real eigenvalues describe periodic oscillations (“breathing”) of the localized state, (ii) [*instability modes*]{} correspond to purely imaginary eigenvalues, and (iii) [*oscillatory instabilities*]{} can occur when the eigenvalues are complex. To conduct the further analysis in more details, we need to know the properties of the wave spectrum defined by the type of a linear (homogeneous or periodic) medium.
Impurity-induced localization in a homogeneous medium
=====================================================
In order to provide a link with earlier studies, first we apply the general analytical technique to describe the properties of nonlinear impurity modes in a homogeneous medium, i.e. when $\nu(x) \equiv 0$. In this case, we have $\mu = \sqrt{\omega}$, and the Bloch-type solutions correspond to non-interacting counter-propagating waves, with constant values of $a(x)$ and $b(x)$. Therefore, we have $r^{\pm} \equiv 0$, ${\zeta^\pm} = 2 \mu$, and $\tau^\pm = e^{-\mu h}$. Then, the condition $|\tau| \ne 1$ reduces to a simple relation $\omega > 0$, so that $\mu$ is real, which means that [*wave localization is possible due to the total internal reflection (IR)*]{} only.
To demonstrate the basic properties of the IR mode, we consider a localized impurity possessing a cubic nonlinear response, $$G(I) = \alpha + \beta I.$$ Under the proper scaling, the absolute value of the nonlinear coefficient $\beta$ can be normalized to unity, so that $\beta=+1$ corresponds to [*self-focusing*]{}, and $\beta=-1$ to [*self-defocusing*]{} nonlinearity. Localization depends also on the sign of the linear coefficient $\alpha$, which defines the impurity response at small intensities: [*attractive*]{}, if $\alpha>0$, and [*repulsive*]{}, otherwise. In this case, the matching condition (\[eq:zeta\]) can be readily solved, and we obtain an explicit solution for the profile of the localized waves, $$u(x) = u_0 e^{-\sqrt{\omega} |x|}, \quad
u_0^{2} = \frac{1}{\beta} (2 \sqrt{\omega} - \alpha) ,$$ and find the corresponding expression for the power, $P(\omega) = {u_0^2} / {\sqrt\omega}$. We can also present the Evans function in a simple form, $$\label{eq:Evans_homog}
Y(\gamma) = 2 (\eta - 2 \sqrt{\omega}) (\eta - 2 \sqrt{\omega} + \alpha),$$ where $\eta = \sqrt{\omega+\gamma} + \sqrt{\omega-\gamma}$, and the square root values are taken on a branch with a positive real part.
Our analysis shows that the localized modes can exist in the presence of self-defocusing nonlinearity ($\beta<0$) only for $\alpha>0$. Then, the family of localized solutions bifurcates from the linear limit at the [*cut-off frequency*]{} $\omega_b=\alpha^2/4$, with $P(\omega_b)=0$, and exists for $0<\omega<\omega_b$. It is straightforward to check that the equation $Y(\gamma)=0$, for the Evans function given by Eq. (\[eq:Evans\_homog\]), has solutions only in the trivial case $\gamma=0$ (corresponding to $\eta=2\sqrt{\omega}$), since a formal solution $\eta = (2\sqrt{\omega}-\alpha)=2(\sqrt{\omega}-\sqrt{\omega_b})$ is negative, and thus it can not correspond to any $\gamma$. Therefore, the nonlinear states supported by a nonlinearly repulsive impurity are stable.
In the case of self-focusing nonlinearity ($\beta>0$), the modes appear above the frequency cut-off, $\omega>\omega_b$, if $\alpha>0$, and they exist for $\omega>0$, provided $\alpha<0$. Non-trivial linear eigenvalues exist if $\sqrt{\omega} > \alpha / (2-\sqrt{2})$, and they are given by the following expression, $\gamma = \pm 2 (\sqrt{\omega}-\alpha/2)
[ \alpha (\sqrt{\omega} - \alpha/4)]^{1/2}$. It then follows that $\gamma$ is imaginary, and the modes are [*unstable*]{}, if $\alpha < 0$, and [*stable*]{} if $\alpha>0$, when $\gamma$ is real. We would like to note that for the self-focusing nonlinearity the mode stability can also be determined using the [*Vakhitov-Kolokolov criterion*]{} [@VK], which states that the modes are stable, if $d P / d \omega > 0$, and unstable, otherwise. Indeed, the slope of the dependence $P(\omega)$ is always positive if $\alpha > 0$, as illustrated in Fig. \[fig:pwr-homog\].
Impurity-induced localization in a periodic superlattice
========================================================
We now consider the properties of the localized waves supported by an impurity with a self-focusing nonlinearity ($\beta=+1$), and attractive linear response ($\alpha>0$), in a two-component superlattice. Due to a periodic modulation of the effective potential, $\nu(x)$, such a lattice possesses [*several band gaps*]{}, as demonstrated in Fig. \[fig:bandgap\] (top).
The first (semi-infinite) band gap exists due to [*the internal reflection*]{} (IR) and, therefore, the properties of the corresponding localized waves should be similar to those existing in a homogeneous medium. Indeed, we find that the localized states in the IR band gap resemble conventional impurity modes (or solitary waves) modulated by a periodic structure \[see Fig. \[fig:bandgap\](b)\], and these states exist above the cut-off frequency $\omega_b$ defined by the equality $\zeta(\omega_b^{\rm (IR)})=\alpha$. Since such an IR wave is a fundamental eigenstate of the self-induced waveguide, it can be demonstrated that the conditions of the Vakhitov-Kolokolov (VK) stability theorem [@VK] are satisfied, and the IR states are [*unstable*]{} if and only if $d P / d \omega < 0$. The critical point $d P / d \omega = 0$ corresponds to a “collision” between two internal modes of the localized wave at the origin, as illustrated in Fig. \[fig:pwr-sf-alp\] (middle). It is interesting to note that the high-frequency localized waves can become unstable in a periodic superlattice, although they are linearly stable when a nonlinear impurity with similar characteristics is embedded in a homogeneous medium (cf. Figs. \[fig:pwr-homog\] and \[fig:pwr-sf-alp\]).
Additional band gaps appear at lower frequencies due to the resonant [*Bragg-type reflection*]{} (BR) from a periodic structure. The corresponding localized waves exist in the waveguiding regime (white regions in Fig. \[fig:pwr-sf-alp\], top), above the cut-off frequencies $\omega_b^{\rm (BR)}$, where again $\zeta(\omega_b^{\rm (BR)})=\alpha$. Such localized solutions are somewhat similar to the gap solitons composed of mutually coupled backward and forward propagating waves \[see Fig. \[fig:bandgap\](a)\]. For the BR states, the VK criterion provides only a necessary condition for stability. Indeed, we notice that in the linear limit there always exists an internal mode corresponding to a resonant coupling between the BR and IR band-gaps, since $Y \left( \omega_b^{\rm (IR)}-\omega_b^{\rm (BR)} \right) \equiv 0$. We perform extensive numerical calculations and find that this mode leads to an oscillatory instability of BR waves when the value $(\omega-{\rm Re}\;\gamma)$ moves outside the band gap; it happens when the intensity exceeds a threshold value (see Figs. \[fig:pwr-sf-alp\] and \[fig:imode-sf-alp\]).
Concluding remarks
==================
We have analyzed the existence and stability of nonlinear localized waves in one-dimensional periodic structures (or superlattices) considering the simplest nonlinear generalization of the Kronig-Penney model with a nonlinear impurity. Taking into account many common features of nonlinear guided waves and impurity modes in stratified and disordered media, on one side, and the self-trapped states and solitary waves in homogeneous nonlinear media, on the other side, we expect that many of the results described above will be found in more realistic physical models of nonlinear periodic media. Such cases include the electron self-trapping and locking states in cuprates and semiconductor superlattices [@a1]; nonlinear guided waves in optical superlattices [@a2]; impurity modes in photonic band-gap materials [@a3], magneto-optical periodic structures [@a4], and photonic crystal fibers [@pcf]; coherent matter waves in optical lattices [@a5], etc. In particular, stability of nonlinear localized impurity modes is a crucial issue for creating tunable band-gap materials where gaps could be controlled by changing the input light intensity.
We are indebted to O. Bang for useful collaboration at the initial stage of this project. The work has been partially supported by the Performance and Planning Fund of the Institute of Advanced Studies.
[1]{}
See, e.g., P. Yeh, [*Optical Waves in Layered Media*]{} (John Wiley & Sons, New York, 1988).
See, e.g., [*Confined Electrons and Photons: New Physics and Applications*]{}, Eds. E. Burstein and C. Weisbuch (Plenum Press, New York, 1995).
F. Delyon, Y.-E. Lévy, and B. Souillard, Phys. Rev. Lett. [**57**]{}, 2010 (1986); see also a review paper D. Hennig and G. P. Tsironis, Phys. Rep. [**307**]{}, 333 (1999).
W. Chen and D. L. Mills, Phys. Rev. Lett. [**58**]{}, 160 (1987); D. N. Christodoulides and R. I. Joseph, Phys. Rev. Lett. [**62**]{}, 1746 (1989); N. Aközbek and S. John, Phys. Rev. E [**57**]{}, 2287 (1998).
H. Kogelnik and C. V. Shank, J. Appl. Phys [**42**]{}, 2327 (1972); H. G. Winful, Appl. Phys. Lett. [**46**]{}, 527 (1985).
B. J. Eggleton, R. E. Slusher, C. M. de Sterke, P. A. Krug, and J. E. Sipe, Phys. Rev. Lett. [**76**]{}, 1627 (1996).
B. P. Anderson and M. A. Kasevich, Science [**282**]{}, 1686 (1998).
I. V. Barashenkov, D. E. Pelinovsky, and E. V. Zemlyanaya, Phys. Rev. Lett. [**80**]{}, 5117 (1998); see also A. De Rossi, C. Conti, and S. Trillo, Phys. Rev. Lett. [**81**]{}, 85 (1998).
Using the optics terminology, we notice that such two types of localized states are the guided waves corresponding to the conventional waveguiding, due to the total internal reflection, and to the band-gap states, due to the Bragg-type reflection.
See, e.g., Yu. S. Kivshar, A. M. Kosevich, and O. A. Chubykalo, Zh. Éksp. Teor. Fiz. [**93**]{}, 5968 (1987) \[Sov. Phys. JETP [**66**]{}, 545 (1987)\]; Phys. Lett. A [**125**]{}, 35 (1987); and references therein.
See, e.g., the review papers: G. I. Stegeman, C. T. Seaton, W. H. Hetherington, A. D. Boardman, and P. Egan, in [*Electromagnetic Surface Excitations*]{}, Eds. R. F. Wallis and G. I. Stegeman (Springer-Verlag, Berlin, 1986); F. Lederer, U. Langbein, and H. E. Ponath, in [*Lasers and Their Applications*]{}, Ed. A.Y. Spassov (World Scientific, Singapore, 1987); and references therein.
A. A. Sukhorukov, Yu. S. Kivshar, O. Bang, and C. M. Soukoulis, Phys. Rev. E [**63**]{}, 016615 (2001).
N. G. Vakhitov and A. A. Kolokolov, Izv. Vyssh. Uchebn. Zaved. Radiofiz. [**16**]{}, 1020 (1973) \[Radiophys. Quantum Electron. [**16**]{}, 783 (1973)\]; see also a recent review paper, Yu. S. Kivshar and A. A. Sukhorukov, in [*Spatial Optical Solitons*]{}, Eds. W. Torruellas and S. Trillo (Springer-Verlag, New York) in press, arXiv: .
J. Hader, P. Thomas, and S. W. Koch, Progr. Quantum Electron. [**22**]{}, 123 (1998); O. M. Bulashenko, V. A. Kochelap, and L. L. Bonilla, Phys. Rev. B [**54**]{}, 1537 (1996); Q. Tian and C. Wu, Phys. Lett. A [**262**]{}, 83 (1999); F. V. Kusmartsev and H. S. Dhillon, Phys. Rev. B [**60**]{}, 6208 (1999).
See, e.g., Y. Y. Zhu and N. B. Ming, Opt. Quantum. Electron. [**31**]{}, 1093 (1999), and references therein.
See, e.g., Qiming Li, C. T. Chan, K. M. Ho, and C. M. Soukoulis, Phys. Rev. B [**53**]{}, 15577 (1996); A. Mekis, S. Fan, and J. D. Joannopoulos, Phys. Rev. B [**58**]{}, 4809 (1998); O. Painter, R. K. Lee, A. Scherer, A. Yariv, J. D. O’Brien, P. D. Dapkus, and I. Kim, Science [**284**]{}, 1819 (1999).
See, e.g., M. Inoue, K. Arai, T. Fujii, and M. Abe, J. Appl. Phys. [**83**]{}, 6768 (1998).
R. F. Cregan, B. J. Mangan, J. C. Knight, T. A. Birks, P. St. J. Russell, P. J. Roberts, and D. C. Allan, Science [**285**]{}, 1537 (1999); B. J. Eggleton, P. S. Westbrook, R. S. Windeler, S. Spälter, and T. A. Strasser, Opt. Lett. [**24**]{}, 1460 (1999).
F. Barra, P. Gaspard, and S. Rica, Phys. Rev. E [**61**]{}, 5852 (2000); K.-P. Marzlin and W. Zhang, Eur. Phys. J. D [**12**]{}, 241 (2000); J. C. Bronski, L. D. Carr, B. Deconinck, and J. N. Kutz, arXiv: ; J. C. Bronski, L. D. Carr, B. Deconinck, J. N. Kutz, and K. Promislow, arXiv: .
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Let $X$ and $Y$ be topological spaces, let $Z$ be a metric space, and let $f: X\times Y\to Z$ be a mapping. It is shown that when $Y$ has a countable base $\mathcal B$, then under a rather general condition on the set-valued mappings $X\ni x\to f_x(B)\in 2^Z$, $B\in\mathcal B$, there is a residual set $R\subset X$ such that for every $(a,b)\in R\times Y$, $f$ is jointly continuous at $(a,b)$ if (and only if) $f_a: Y\to Z$ is continuous at $b$. Several new results are also established when the notion of continuity is replaced by that of quasicontinuity or by that of cliquishness. Our approach allows us to unify and improve various results from the literature.'
address:
- 'Département de Mathématiques, Université de Rouen, UMR CNRS 6085, Avenue de l’Université, BP.12, F76801 Saint-Etienne du Rouvray, France'
- '3 Impasse des Avocettes, F76930 Cauville sur Mer, France'
author:
- 'A. Bouziad'
- 'J.-P. Troallic'
title: 'Lower Quasicontinuity, Joint Continuity and Related concepts'
---
introduction
============
Let $f$ be a mapping of the product $X\times Y$ of two topological spaces into a metric space $Z$. Let $\mathcal B$ be a countable collection of subsets of $Y$ and let $B$ be the set of all $y\in Y$ such that $\mathcal B$ includes a neighborhood base at $y$ in $Y$. In this note, we are mainly concerned with the following question (in the spirit of [@CT]): find some general assumptions on the partial mappings $f_x$, $f^y$ ($x\in X$, $y\in Y$) ensuring the existence of a residual set $R\subset X$ such that $f$ is jointly continuous at each point of $R\times B$. In [@CT], it is shown that if $f$ is separately continuous, then such a set $R$ exists. Theorem 3.3 below is the main result of this note; the first of assertions (1), (2) and (3) constituting this theorem is a much more general result than that of [@CT] (although the proof is hardly more difficult); the other two are of the same sort as (1), with the concept of continuity replaced by that of quasicontinuity in (2), and that of cliquishness in (3).
The paper is organized as follows. In Section 2, we introduce and examine the concept of lower quasicontinuity with respect to the variable $x$ (at $(a, b)\in X\times Y$) for the mapping $f: X\times Y\to Z$. (Here, $Z$ may be any topological space.) The statement and the proof of the main Theorem 3.3, as well as some immediate corollaries, occupy Section 3. Finally, the results of Sections 2 and 3 are related in Section 4 to some well-known theorems from the literature.
As usual, $\mathbb N$, $\mathbb Q$ and $\mathbb R$ will denote, respectively, the sets of natural, rational and real numbers. For any set $A$, $2^A$ will denote the set of all nonempty subsets of $A$. For $x\in X$, ${\mathcal V}_X(x)$ will denote the set of all neighborhoods of $x$ in $X$.
Lower quasicontinuity and mappings of two variables
===================================================
In this section, $X$, $Y$, $Z$ are topological spaces and $f: X\times Y\to Z$ is a mapping. First, recall that a mapping $g:X\to Z$ is said to be [*quasicontinuous*]{} at $a\in X$ if for each neighborhood $U$ of $a$ in $X$ and each neighborhood $W$ of $g(a)$ in $Z$, there is an open set $O\subset X$ such that $\emptyset\not= O\subset U$ and $g(O)\subset W$ [@Th; @Ma] (see also [@Ke; @Bl]; the terminology differs in [@Th; @Bl]). Obviously, $g$ is said to be quasicontinuous if it is quasicontinuous at each point of $X$. (From now on, obvious definitions will be omitted.) A set-valued mapping $F: X\to 2^Z$ is said to be [*lower quasicontinuous*]{} at $x_0\in X$ if for each neighborhood $U$ of $x_0$ in $X$ and each open set $W\subset Z$ such that $F(x_0)\cap W\not=\emptyset$, there is an open set $O\subset X$ such that $\emptyset\not= O\subset U$ and $F(x)\cap W\not=\emptyset$ for every $x\in O$ [@Ne; @N] (and [@BG] with another terminology). Given any $W\subset Z$, we put $F^-(W)=\{x\in X : F(x)\cap W\not=\emptyset\}$.
The following concept is formulated in [@MN]: The mapping $f:X\times Y\to Z$ is said to be [*horizontally quasicontinuous*]{} at $(a, b)\in X\times Y$ if for each neighborhood $U$ of $a$ in $X$, each neighborhood $V$ of $b$ in $Y$, and each neighborhood $W$ of $f(a, b)$ in $Z$, there is an open set $O\subset X$ and $y\in V$ such that $\emptyset\not= O\subset U$ and $f(O\times\{y\})\subset W$. Let us note that if $f^{b}$ is quasicontinuous at $a$, then $f$ is horizontally quasicontinuous at $(a,b)$. It is also easy to check the following:
If $a\in X$ and if $f: X\times Y\to Z$ is horizontally quasicontinuous at $(a, y)$ for every $y$ in a nonempty open set $V\subset Y$, then the set-valued mapping $F^V:X\ni x\to f_x(V)\in 2^Z$ is lower quasicontinuous at $a$.
The converse of Proposition 2.1 is false: Let $f$ be the mapping of ${\mathbb R}\times {\mathbb R}$ into the discrete space $\{0,1\}$ defined by $f(x,y)=1$ if and only if $x-y\in\mathbb Q$; then for every nonempty open set $V\subset \mathbb R$, the set-valued mapping ${\mathbb R}\ni x\to f_x(V)\in 2^{\{0,1\}}$ is lower quasicontinuous; however, there is no point of ${\mathbb R}\times {\mathbb R}$ at which $f$ is horizontally quasicontinuous. (Curiously enough, both facts hold for the same reason: For every $x\in\mathbb R$, the sets $x+\mathbb Q$ and $x+({\mathbb R}\setminus {\mathbb Q})$ are dense in $\mathbb R$.)
[We say that $f: X\times Y\to Z$ is [*lower quasicontinuous with respect to the variable $x$*]{} ([*lower $X$-quasicontinuous*]{}, for short) at the point $(a, b)\in X\times Y$ if for each neighborhood $U$ of $a$ in $X$, each neighborhood $V$ of $b$ in $Y$, and each neighborhood $W$ of $f(a,b)$ in $Z$, there is an open set $O\subset X$ such that $\emptyset\not= O\subset U$ and $f(\{x\}\times V)\cap W\neq\emptyset$ for every $x\in O$. Clearly, “$f: X\times Y\to Z$ is lower $X$-quasicontinuous” means exactly that for each nonempty open set $V\subset Y$, the set-valued mapping $F^V:X\ni x\to f_x(V)\in 2^Z$ is lower quasicontinuous. On the other hand, let us point out that if $f$ is horizontally quasicontinuous at $(a,b)$, then $f$ is lower quasicontinuous with respect to the variable $x$ at $(a,b)$. ]{}
The following proposition describes a wide class of mappings for which the results in Section 3 will apply (see Section 4).
Let $V\subset Y$ be a nonempty open set, and suppose that the mapping $f: X\times Y\to Z$ satisfies one of the following:
- $f$ is vertically quasicontinuous at every point of $X\times V$, and, for each $x\in X$, there exists a dense subset $D_x$ of the space $V$ such that $f$ is lower $X$-quasicontinuous at every point of $\{x\}\times D_x$.
- $f$ is lower $Y$-quasicontinuous at every point of $X\times V$, and there exists a dense subset $D$ of the space $V$ such that $f$ is lower $X$-quasicontinuous at every point of $X\times D$.
Then the set-valued mapping $F^V:X\ni x\to f_x(V)\in 2^Z$ is lower quasicontinuous.
Let $U\subset X$, $W\subset Z$ be nonempty open sets such that $U$ meets $(F^V)^-(W)$, and let us show that there is a nonempty open set in $X$ contained in $U$ and $(F^V)^-(W)$. Let us choose an arbitrary point $(x_0,y_0)$ in $(U\times V)\cap f^{-1}(W)$. In case ([1]{}), there is a nonempty open set $V_0\subset V$ and $x_1\in U$ such that $f(\{x_1\}\times V_0)\subset W$; taking $y_1\in V_0\cap D_{x_1}$, one can find a nonempty open set $U_0\subset U$ such that $f(\{x\}\times V)\cap W\neq\emptyset$ for every $x\in U_0$. In case ([2]{}), there is a nonempty open set $V_1\subset V$ such that $f(U\times \{y\})\cap W\neq\emptyset$ for every $y\in V_1$. Take $y_2\in V_1\cap D$ and choose $x_2\in U$ such that $f(x_2,y_2)\in W$; then there is a nonempty open set $U_1\subset U$ such that $f(\{x\}\times V)\cap W\neq\emptyset$ for every $x\in U_1$.
The next statement in terms of the familiar concept of quasicontinuity follows from 2.3; it will be used in Sections 3 and 4 to derive several Hahn or Kempisty type results from Theorem 3.3.
Suppose that for each $x\in X$, $f_x$ is quasicontinuous and there is a dense set $D_x\subset Y$ such that $f^y$ is quasicontinuous at $x$ for every $y\in D_x$. Then $f: X\times Y\to Z$ is lower $X$-quasicontinuous.
\(1) It is easy to see that the mapping $f: X\times Y\to Z$ is lower quasicontinuous with respect to the variable $x$ at $(a,b)\in X\times Y$ if and only if $f(a,b)\in \overline{f((U\cap A)\times V)}$ for any neighborhood $U$ of $a$ in $X$, any neighborhood $V$ of $b$ in $Y$, and any dense subset $A$ of $X$.
\(2) Let $\tau_f$ be the topology on $X\times Y$ generated by $f: X\times Y\to Z$ and the two projections $X\times Y\ni (x,y) \to x\in X$, $X\times Y\ni (x,y) \to y\in Y$. Then, as the proof of Proposition 2.3 shows, the set $Q$ of all $(x,y)\in X\times Y$ at which $f$ is lower $X$-quasicontinuous is closed with respect to $\tau_f$. Thus, any condition ensuring that $Q$ is $\tau_f$-dense in $X\times Y$ will imply that $f$ is lower $X$-quasicontinuous; for instance, the conditions ([1]{}) and ([2]{}) in Proposition 2.3 (when satisfied for all nonempty open subsets of $Y$) are of this sort.
joint continuity and related concepts
=====================================
In order to state our main result in 3.3 below, let us recall some definitions concerning the concept of quasicontinuity and that, weaker, of cliquishness. Let $X$ and $Y$ be two topological spaces and let $(Z, d)$ be a metric space. A mapping $g$ of $X$ into $(Z, d)$ is said to be [*cliquish*]{} at $a\in X$ if for each $\varepsilon > 0$ and each neighborhood $U$ of $a$ in $X$ there is an open set $O\subset X$ such that $\emptyset\not= O\subset U$ and ${\rm diam} (g(O))
\leq\varepsilon$ [@Th; @Ma]. (If $W\in 2^Z$, then ${\rm diam} (W) = \sup\{ d(u, v) : u, v\in W\}$.) A mapping $f$ of $X\times Y$ into $(Z, d)$ is said to be [*quasicontinuous*]{} ([*cliquish*]{}) [*with respect to the variable*]{} $x$ at $(a,b)\in X\times Y$ if for each neighborhood $V$ of $b$ in $Y$ and each $\varepsilon > 0$, there is a neighborhood $U$ of $a$ in $X$ and an open set $O\subset Y$ such that $\emptyset\not= O\subset V$ and $d(f(a, b), f(x, y)) \leq \varepsilon$ for all $x\in U$, $y\in O$ [@M; @P1] (respectively, and $d(f(a,y),f(x,y^{\prime})) \leq\varepsilon$ for all $x\in U$, $y, y^{\prime}\in O$ [@F]). In relation to Theorem 4.4 below, it is worth noticing that [*if $a\in X$ and if $f$ is cliquish with respect to the variable $x$ at $(a,y)\in X\times Y$ for every $y\in Y$, then the set of all $y\in Y$ such that $f$ is continuous at $(a,y)$ is residual in $Y$*]{} [@F].
Recall also that a collection $\mathcal B$ of nonempty open sets in a topological space is called a [*pseudobase*]{} (or [*$\pi$-base*]{}) for this space if any nonempty open set contains some member of $\mathcal B$ [@O].
Let $X$ be a topological space and let $(Z, d)$ be a bounded metric space. Let $F$ be a lower quasicontinuous set-valued mapping of $X$ into $2^Z$. Let $z_0\in Z$. Then the mapping $\phi : X\rightarrow \mathbb R$ defined by $\phi(x) = \sup\{d(z_0, z) :
z\in F(x)\}$ [(]{}$x\in X$[)]{} is cliquish.
Let $U$ be a nonempty open subset of the space $X$, and let $\varepsilon > 0$. Let us put $r = \sup\{ \phi(x) : x\in U\} - \varepsilon$ and let us choose $x_0\in U$ such that $r < \phi(x_0)$. There is $z_1\in F(x_0)$ such that $r < d(z_0, z_1)$; let us put $\rho = d(z_0, z_1) - r$. Let $O$ be a nonempty open subset of $X$ contained in $U\cap F^-( B(z_1, \rho))$; then $r < \phi(x) \leq r + \varepsilon$ for every $x\in O$.
For a set-valued mapping $F:X\to 2^Z$ and $U\subset X$, let $F(U)$ denote the set $\cup_{x\in U}F(x)$.
Let $X$ be a topological space and let $(Z, d)$ be a bounded metric space. Let $F$ be a lower quasicontinuous set-valued mapping of $X$ into $2^Z$. Then $$A = \{x\in X : \forall\varepsilon > 0, \exists U\in {\mathcal V}_X(x), {\rm diam} (F(U)) \leq 2 {\rm diam} (F(x)) + \varepsilon\}$$ is a residual subset of $X$.
Let $\varepsilon > 0$, and let $A_{\varepsilon}$ be the union of all open subsets $U$ of $X$ such that ${\rm diam} (F(U)) \leq 2{\rm diam} (F(x)) + \varepsilon$ for every $x\in U$. Let us show that the open subset $A_{\varepsilon}$ of $X$ is dense in $X$; this will prove the lemma since $\bigcap_{n\in{\mathbb N}} A_{{1}/{n + 1}}\subset A$. Let $O$ be a nonempty open subset of $X$. Then, taking $x_0\in O$ and $z_0\in F(x_0)$, one can find a nonempty open subset $O^{\prime}$ of $X$ contained in $O\cap F^-(B(z_0,{\varepsilon}/4))$. Let $\phi$ be the real-valued mapping on $X$ defined by $\phi(x) = \sup\{d(z_0, z) \mid z\in F(x)\}$ for every $x\in X$. By Lemma 3.1, there is a nonempty open set $O^{{\prime}{\prime}}\subset O^{\prime}$ such that ${\rm diam} (\phi(O^{{\prime}{\prime}})) \leq {\varepsilon}/{4}$. Let us verify that $O^{{\prime}{\prime}}\subset A_{\varepsilon}$, which will imply that $O\cap A_{\varepsilon} \neq \emptyset$ since $O^{{\prime}{\prime}}\subset O$. Let $x^{{\prime}{\prime}}\in O^{{\prime}{\prime}}$. Let $z_i\in F(O^{{\prime}{\prime}})$, and let $x_i\in O^{{\prime}{\prime}}$ such that $z_i\in F(x_i)$ ($i= 1, 2$); then $$d(z_1, z_2) \leq d(z_1, z_0) + d(z_0, z_2) \leq\phi(x_1) + \phi(x_2) \leq 2\phi(x^{{\prime}{\prime}}) + {\varepsilon}/{2};$$ consequently ${\rm diam} (F(O^{{\prime}{\prime}})) \leq 2\phi(x^{{\prime}{\prime}}) + {\varepsilon}/{2}$. Now, let $z^{{\prime}{\prime}}\in F(x^{{\prime}{\prime}})\cap B(z_0, {\varepsilon}/{4})$; then $$\sup\{d(z_0, z) : z\in F(x^{{\prime}{\prime}})\}
\leq d(z_0, z^{{\prime}{\prime}}) + \sup\{ d(z^{{\prime}{\prime}}, z) : z\in F(x^{{\prime}{\prime}})\},$$ and consequently $\phi(x^{{\prime}{\prime}}) \leq {\varepsilon}/{4} + {\rm diam} (F(x^{{\prime}{\prime}}))$. Since $O^{{\prime}{\prime}}$ is a nonempty open subset of $X$ such that ${\rm diam} (F(O^{{\prime}{\prime}})) \leq 2{\rm diam} (F(x^{{\prime}{\prime}})) + {\varepsilon}$ for every $x^{{\prime}{\prime}}\in O^{{\prime}{\prime}}$, the inclusion $O^{{\prime}{\prime}}\subset A_{\varepsilon}$ holds.
Let $X$ and $Y$ be topological spaces, let $(Z, d)$ be a metric space, and let $f : X\times Y\rightarrow Z$ be a mapping. Let ${\mathcal B}$ be a countable collection of nonempty subsets of $Y$. Let us suppose that, for every $V\in {\mathcal B}$, the set-valued mapping $F^V:X\ni x\to f_x(V)\in 2^Z$ is lower quasicontinuous. Then there is a residual set $R\subset X$ such that for each $a\in R$:
- If $f_a$ is continuous at $b\in Y$ and if $b$ has a neighborhood base contained in ${\mathcal B}$, then $f$ is continuous at $(a, b)$.
- If $f_a$ is quasicontinuous at $b\in Y$ and if some neighborhood of $b$ in $Y$ has a pseudobase contained in ${\mathcal B}$, then $f$ is quasicontinuous with respect to the variable $x$ at $(a, b)$.
- If $f_a$ is cliquish at $b\in Y$ and if some neighborhood of $b$ in $Y$ has a pseudobase contained in ${\mathcal B}$, then $f$ is cliquish with respect to the variable $x$ at $(a, b)$.
One can assume that $(Z, d)$ is bounded. By Lemma 3.2, for every $V\in{\mathcal B}$, the set $$R_V = \{x\in X : \forall\varepsilon > 0, \exists U\in {\mathcal V}_X(x), {\rm diam} (F^V(U)) \leq 2 {\rm diam} (F^V(x)) + \varepsilon\}$$ is a residual subset of $X$. Let us put $R = \bigcap_{V\in {\mathcal B}} R_V$; since $R$ is the intersection of a countable family of residual subsets of $X$, $R$ is a residual subset of $X$. Let us consider $a\in R$.
\(1) Let $b\in Y$ such that $f_a$ is continuous at $b$ and ${\mathcal B}$ contains a neighborhood base at $b$. Let $\varepsilon > 0$, and let ${V\in {\mathcal B}}$ be a neighborhood of $b$ in $Y$ such that ${\rm diam} (f_a(V)) \leq {\varepsilon}/4$. Since $a$ belongs to $R$, there exists $U\in {\mathcal V}_X(a)$ such that ${\rm diam} (F^V(U)) \leq 2 {\rm diam} (F^V(a)) + {\varepsilon}/{2}$. Since $U\times V$ is a neighborhood of $(a, b)$ in $X\times Y$, and since $${\rm diam}(f(U\times V)) = {\rm diam} (F^V(U)) \leq 2 {\rm diam} (f_a(V)) + \varepsilon/2\leq \varepsilon,$$ $f$ is continuous at $(a, b)$.
\(2) Let us suppose that $f_a$ is quasicontinuous at $b\in Y$, and let us suppose that ${\mathcal B}$ contains a pseudobase for some neighborhood $V^{\prime}$ of $b$ in $Y$. Let $V\in{\mathcal V}_Y(b)$ and $\varepsilon > 0$. There is a nonempty open set $O^{\prime}$ in $Y$ such that $O^{\prime}\subset V\cap V^{\prime}$ and $ f_a(O^{\prime})\subset B(f_a(b), {\varepsilon}/6)$. Let ${O\in {\mathcal B}}$ contained in $O^{\prime}$ and open in $V^{\prime}$; $O$ is a nonempty open set in $Y$ contained in $V$. Since $a\in R$, there exists $U\in {\mathcal V}_X(a)$ such that ${\rm diam} (F^O(U)) \leq 2 {\rm diam} (F^O(a)) + {\varepsilon}/{6}$. Now, since for any $(x, y)\in U\times O$, $$d(f(a, b), f(x, y))\leq d(f_a(b), f_a(y)) + d(f_a(y), f_x(y))\leq {\varepsilon}/{6} + {\rm diam} (F^O(U) \leq \varepsilon,$$ statement (2) is proved.
\(3) Suppose that $f_a$ is cliquish at $b\in Y$ and that ${\mathcal B}$ contains a pseudobase for some neighborhood $V^{\prime}$ of $b$ in $Y$. Let $V\in{\mathcal V}_Y(b)$ and $\varepsilon > 0$. There is a nonempty open set $O^{\prime}$ in $Y$ such that $O^{\prime}\subset V\cap V^{\prime}$ and ${\rm diam}( f_a(O^{\prime})) \leq {\varepsilon}/4$. Let ${O\in {\mathcal B}}$ contained in $O^{\prime}$ and open in $V^{\prime}$; $O$ is a nonempty open set in $Y$ contained in $V$. Since $a\in R$, there is $U\in {\mathcal V}_X(a)$ such that ${\rm diam} (F^O(U)) \leq 2 {\rm diam} (F^O(a)) + {\varepsilon}/{2}$. It follows from what preceeds that $${\rm diam}(f(U\times O)) = {\rm diam} (F^O(U)) \leq 2 {\rm diam} (f_a(O)) + {\varepsilon}/{2}\leq \varepsilon,$$ which establishes statement (3).
[In point (1) of Theorem 3.3, the topology on $Y$ is quite irrelevant inasmuch as the neighborhood of $b$ used to establish the continuity of $f$ at $(a,b)$ belongs to the collection $\mathcal B$. Similar remarks hold for point (2) and point (3) of that same theorem.]{}
In view of 3.3 (and 2.4, for Corollary 3.7), we can state:
Let $X$ and $Y$ be topological spaces, let $Z$ be a metric space, and let $f: X\times Y\to Z$ be a mapping. Let us suppose that $Y$ has a countable base and that for every $y\in Y$, the mapping $f^y$ is quasicontinuous. Then there is a residual set $R\subset X$ such that for every $(a,b)\in R\times Y$, $f$ is continuous at $(a,b)$ provided that $f_a$ is continuous at $b$.
Let $X$ and $Y$ be topological spaces, let $Z$ be a metric space, and let $f: X\times Y\to Z$ be a mapping. Let us suppose that $Y$ has a countable pseudobase and that for every $y\in Y$, the mapping $f^y$ is quasicontinuous. Then there is a residual set $R\subset X$ such that for every $(a,b)\in R\times Y$, $f$ is quasicontinuous [(]{}cliquish[)]{} with respect to the variable $x$ at $(a,b)$ provided that $f_a$ is quasicontinuous [(]{}respectively, cliquish[)]{} at $b$.
Let $X$ be a topological space, let $Y$ be a topological space with a countable base [(]{}pseudobase[)]{}, and let $f: X\times Y\to Z$ be a mapping of $
X\times Y$ into a metric space $Z$. Suppose that for each $x\in X$, the mapping $f_x$ is continuous [(]{}quasicontinuous[)]{} and there is a dense set $D_x\subset Y$ such that $f^y$ is quasicontinuous at $x$ for every $y\in D_x$. Then there is a residual set $R\subset X$ such that $f$ is continuous [(]{}respectively, quasicontinuous with respect to the variable $x$[)]{} at every point of $R\times Y$.
[One can also obtain Theorem 3.3 by using the following lemma established in [@Ku]: [*Let $X$ be a topological space, let $(Z, d)$ be a metric space, and let $F: X\to 2^Z$ be a lower quasicontinuous set-valued mapping. Then for every $r > 0$ there exists an open dense subset $U$ of $X$ and a continuous mapping $f: U\to Z$ such that $d(f(u), F(u)) < r$ for all $u\in U$.* ]{}]{}
New light on some well-known results
====================================
The aim of this section is to relate the results in Section 3 to some well-known results from the literature. In what follows, $X$ is a Baire space, $Y$ is a topological space, and $f$ is a mapping of $X\times Y$ into a metric space $(Z, d)$.
In [@Mi Theorem 1], Mibu proves that the set of continuity points of $f$ is residual in $X\times Y$ assuming $f_x$ to be continuous for all $x\in X$, $f^y$ continuous for all $y$ in a given dense subset of $Y$, and $Y$ first countable. Theorem 3.3 and Corollary 2.4 above immediately give Mibu’s result (in a somewhat more general form).
Suppose that $Y$ is first countable and that, for each $x\in X$, $f_x$ is continuous and $f^y$ is quasicontinuous at $x$ for every $y$ in a dense set $D_x\subset Y$. Then for each $y\in Y$, there exists a residual set $R_y\subset X$ such that $f$ is continuous at every point of $R_y\times \{y\}$.
In [@F Theorem 3], Fudali concludes that $f$ is cliquish assuming $f_x$ to be cliquish for each $x\in X$, $f^y$ quasicontinuous for each $y\in Y$, and $Y$ locally second countable (a result stronger than Mibu’s Theorem 2 in [@Mi]). The following (slightly more general) statement is deduced from 3.3.
Suppose that $f^y$ is quasicontinuous for every $y\in Y$, and suppose that there is a dense set $D\subset Y$ such that for every $y\in D$:
- There is a dense Baire subspace $Q_y$ of $X$ such that $f_x$ is cliquish at $y$ for every $x\in Q_y$, and
- some neighborhood of $y$ in $Y$ has a countable pseudobase.
Then $f$ is cliquish.
Let $U$ be a nonempty open set in $X$, $V$ a nonempty open set in $Y$, and $\varepsilon>0$. Let us choose $b\in V\cap D$. By Theorem 3.3, there is a residual set $R\subset X$ such that for any $a\in R$, if $f_a$ is cliquish at $b$, then $f$ is cliquish with respect to the variable $x$ at $(a,b)$. Since $Q_b$ is a dense Baire subspace of $X$, $R\cap Q_b$ is dense in $X$; therefore there is $a\in U$ such that $f$ is cliquish with respect to the variable $x$ at $(a,b)$. Let us choose $U_1\in {\mathcal V}_X(a)$ contained in $U$ and a nonempty open set $V_1$ in $Y$ contained in $V$ such that $d(f(a,y),f(x,y'))<{\varepsilon}/{2}$ for all $x\in U_1$ and $y,y'\in V_1$; then $d(f(x,y),f(x',y'))< \varepsilon$ for all $x,x'\in U_1$ and $y,y'\in V_1$.
In [@M Theorem 1], it is proved by Martin that if $f_x$ is quasicontinuous for every $x\in X$, $f^y$ quasicontinuous for every $y\in Y$, and $Y$ second countable, then $f$ is quasicontinuous. Using 3.3 and 2.4 under condition (i), and 3.3 under condition (ii), Martin’s result can be improved as follows.
Suppose that for each $y\in Y$, some neighborhood of $y$ in $Y$ has a countable pseudobase, and suppose that one of the following holds:
- For each $x\in X$, $f_x$ is quasicontinuous and there is a dense set $D_x\subset Y$ such that $f^y$ is quasicontinuous at $x$ for every $y\in D_x$.
- For each $y\in Y$, $f^y$ is quasicontinuous and there is a dense Baire subspace $Q_y\subset X$ such that $f_x$ is quasicontinuous at $y$ for every $x\in Q_y$.
Then $f$ is quasicontinuous.
Let $(a,b)$ in $X\times Y$. Let $U$ be an open neighborhood of $a$ in $X$, $V$ an open neighborhood of $b$ in $Y$, and $\varepsilon>0$.
Case ([1]{}): By Theorem 3.3 and Corollary 2.4, for each $y\in Y$ there exists a residual set $R_y\subset X$ such that $f$ is quasicontinuous with respect to the variable $x$ at every point of $R_y\times\{y\}$. Let $V_1\subset V$ be a nonempty open set such that $d(f(a,y),f(a,b))<\varepsilon$ for every $y\in V_1$. Choose $y_1\in V_1\cap D_a$ and let $U_1\subset U$ be a nonempty open set such that $d(f(x,y_1),f(a,b))<\varepsilon$ for every $x\in U_1$. Now, choosing $x_1\in U_1\cap R_{y_1}$ gives an open neighborhood $U_2$ of $x_1$ contained in $U_1$ and a nonempty open set $V_2\subset V_1$ such that $d(f(x,y),f(a,b))<\varepsilon$ for every $(x,y)\in U_2\times V_2$. Hence the mapping $f$ is quasicontinuous at $(a,b)$.
Case ([2]{}): By Theorem 3.3, for each $y\in Y$ there is a residual set $R_y\subset X$ such that for any $a^{\prime}\in R_y$, if $f_{a^{\prime}}$ is quasicontinuous at $y$, then $f$ is quasicontinuous with respect to the variable $x$ at $(a^{\prime},y)$. Choose a nonempty open set $U_1\subset U$ such that $d(f(x,b),f(a,b))<\varepsilon$ for each $x\in U_1$. Since $Q_b$ is a dense Baire subspace of $X$, $R_b\cap Q_b$ is dense in $X$. Choosing $x_1\in U_1\cap R_b\cap Q_b$ gives an open neighborhood $U_2$ of $x_1$ contained in $U_1$ and a nonempty open set $V_1\subset V$ such that $d(f(x,y),f(a,b))<\varepsilon$ for every $(x,y)\in U_2\times V_1$. Hence the mapping $f$ is quasicontinuous at $(a,b)$.
In [@M Theorem 4], Martin also proves that if $f_x$ is continuous for every $x\in X$, $f^y$ quasicontinuous for every $y\in Y$, and $Y$ first countable, then $f$ is quasicontinuous with respect to the variable $y$ (cf. also [@P1]). The following variant of this theorem is an easy application of 3.3.
Let $b\in Y$ be a point with a countable neighborhood base in $Y$. Suppose that $f^y$ is quasicontinuous for every $y\in Y$, and suppose that $f_x$ is continuous at $b$ for every $x$ in a given dense Baire subspace $Q$ of $X$. Then:
- The mapping $f$ is quasicontinuous with respect to the variable $y$ at each point of $X\times\{b\}$.
- The set of all $x\in X$ such that the mapping $f$ is continuous at $(x, b)$ is residual in $X$.
Let $A$ be the set of all $x\in X$ such that the mapping $f$ is continuous at $(x, b)$. By Theorem 3.3, there is a residual set $R\subset X$ such that $f$ is continuous at each point $(x,b)\in R\times\{b\}$ provided that $f_x$ is continuous at $b$. Remark that $Q$ being a dense Baire subspace of $X$, the subset $R\cap Q$ of $A$ is dense in $X$, and consequently, $A$ is dense in $X$.
\(1) Let $a\in X$. Let $U\in {\mathcal V}_X(a)$ and $\varepsilon>0$. Let $U^{\prime}$ a nonempty open set in $X$ such that $U^{\prime}\subset U$ and $f^b(U^{\prime})\subset B(f^b(a), {\varepsilon}/{2})$; taking $x_0\in U^{\prime}\cap A$ gives a nonempty open set $O\subset U^{\prime}$ and $V\in {\mathcal V}_Y(b)$ such that, for every $(x,y)\in O\times V$, $d(f(x,y),f(x_0,b))<{\varepsilon}/{2}$, and hence such that $d(f(x,y),f(a,b))<{\varepsilon}$.
\(2) The $G_{\delta}$-set $A$ in $X$ being dense in $X$, it is residual in $X$.
As our last application of Theorem 3.3, we will now prove a variant of a recent result [@HH Theorem 2.5]. Following [@HH], a sequence $(f_n)_{n\in\mathbb N}$ of mappings of $X$ into $(Z, d)$ is said to be [*equi-quasicontinuous*]{} at $x\in X$ if, for each neighborhood $U$ of $x$ in $X$ and each $\varepsilon>0$, there exists an open set $O\subset X$ and $n_0\in\mathbb N$ such that $\emptyset\not= O\subset U$ and $d(f_n(x),f_n(y))<\varepsilon$ for every $n\geq n_0$ and $y\in O$.
Let $f_n : X\to Z$, $n\in\mathbb N$, be a sequence of quasicontinuous mappings, and let $f: X\to Z$ be a mapping. Suppose that for each $x\in X$, the sequence $(f_n(x))_{n\in\mathbb N}$ clusters to $f(x)$, and suppose that for each $x$ in a given dense Baire subspace $A$ of $ X$, the sequence $(f_n(x))_{n\in\mathbb N}$ converges to $f(x)$. Then, for each $x\in A$, the following are equivalent:
- The sequence $(f_n)_{n\in\mathbb N}$ is equi-quasicontinuous at $x$.
- The mapping $f: X\to Z$ is quasicontinuous at $x$.
We refer to [@HH] for the implication $(1) \Rightarrow (2)$. To prove that (2) implies (1), let $Y={\mathbb N}\cup\{\infty\}$ equipped with the topology whose nonempty open sets are the subsets $\{m\in{\mathbb N}:m\geq n\}\cup\{\infty\}$ of $Y$ ($n\in\mathbb N$). Let $g: X\times Y\to Z$ be the mapping defined by $g(x,n)=f_n(x)$ and $g(x,\infty)=f(x)$. It is easy to check that for each $n\in\mathbb N$, the set-valued mapping $X\ni x\to \{f_m(x):m\geq n\}\cup\{f(x)\}\in 2^Z$ is lower quasicontinuous. Let $a\in A$ such that $f$ is quasicontinuous at $a$. Let $U\in {\mathcal V}_X(a)$ and $\varepsilon>0$. There is $n_0\in \mathbb N$ such that $d(f_n(a),f(a))<{\varepsilon}/3$ for every $n\geq n_0$. Let $V$ be a nonempty open set in $X$ such that $V\subset U$ and $d(f(a),f(x))<{\varepsilon}/3$ for every $x\in V$. The mapping $g_x$ being continuous at $\infty$ for any $x\in A$, and $A\cap R$ being dense in $X$ for any residual subset $R$ of $X$, it follows from Theorem 3.3 that there is $b\in V$ such that $g$ is continuous at $(b,\infty)$; in particular, there is a nonempty open set $O\subset V$ and $n_1\geq n_0$ such that $d(f(b),f_n(x))<{\varepsilon}/3$ for every $x\in O$ and $n\geq n_1$. For every $x\in O$ and $n\geq n_1$, we have $$d(f_n(x),f_n(a))\leq d(f_n(x),f(b))+d(f(b),f(a))+d(f(a),f_n(a))< \varepsilon.$$
[99]{} W. W. Bledsoe, Neighborly functions, Proc. Amer. Math. Soc. 3 (1952), 114–115.
J. Calbrix, J.-P. Troallic, Applications séparément continues, C. R. Acad. Sci. Paris 288 (1979), 647–648.
L.A. Fudali, On cliquish functions on product spaces, Math. Slovaca 33 (1983), no. 1, 53–58.
J.R. Giles, M.O. Bartlett, Modified continuity and a generalisation of Michael’s selection theorem, Set-Valued Anal. 1 (1993), no. 4, 365–378.
L. Holá, D. Holý, Pointwise convergence of quasicontinuous mappings and Baire spaces (preprint).
S. Kempisty, Sur les fonctions quasicontinues, Fund. Math. 19 (1932), 184–197.
I. Kupka, Quasicontinuous selections for compact-valued multifunctions, Math. Slovaca 43 (1993), no. 1, 69–75.
S. Marcus, Sur les fonctions quasicontinues au sens de S. Kempisty, Colloq. Math. 8 (1961), 47–53.
N.F.G. Martin, Quasi-continuous functions on product spaces, Duke Math. J. 28 (1961), 39–43.
V.K. Maslyuchenko, V.V Nesterenko, Joint continuity and quasicontinuity of horizontally quasicontinuous mappings, (Ukrainian) Ukraïn. Mat. Zh. 52 (2000), no. 12, 1711–1714; translation in Ukrainian Math. J. 52 (2000), no. 12, 1952–1955 (2001). Y. Mibu, On quasi-continuous mappings defined on a product space, Proc. Japan Acad. 192 (1958), 189–192.
T. Neubrunn, On quasicontinuity of multifunctions, Math. Slovaca 32 (1982), no. 2, 147–154.
T. Neubrunn, Quasi-continuity, Real Anal. Exchange 14 (1988/89), no. 2, 259–306. J.C. Oxtoby, Cartesian products of Baire spaces, Fund. Math., 49 (1961), 157-166.
Z. Piotrowski, Continuity points in $\{x\}\times Y$, Bull. Soc. Math. France 108 (1980), 113–115.
H.P. Thielman, Types of functions, Amer. Math. Monthly 60 (1953), 156-161.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
We apply the statistical measure of complexity, introduced by López-Ruiz, Mancini and Calbet to a hard-sphere dilute Fermi gas whose particles interact via a repulsive hard-core potential. We employ the momentum distribution of this system to calculate the information entropy, the disequilibrium and the statistical complexity. We examine possible connections between the particle correlations and energy of the system with those information and complexity measures. The hard-sphere model serves as a test bed for concepts about complexity.
PACS number(s): 05.30.Fk, 89.70.Cf, 05.30.-d, 05.90.+m\
Keywards: Momentum Distribution, Fermi Systems, Nuclear Matter, Hard-Sphere Gas, Information Entropy, Statistical Complexity, Correlations.
author:
- |
Ch.C. Moustakidis$^{1}$, K.Ch. Chatzisavvas$^{1,2}$, N.S. Nikolaidis$^{1,3}$, and C.P. Panos$^{1}$\
\
$^{1}$Department of Theoretical Physics, Aristotle University of Thessaloniki,\
54124 Thessaloniki, Greece\
$^{2}$Informatics and Telecommunications Engineering Department,\
University of Western Macedonia, 50100 Kozani, Greece\
$^{3}$Department of Automation, Faculty of Applied Technology\
Alexander Technological Educational Institute (ATEI) of Thessaloniki\
57400 Thessaloniki, Greece\
title: 'Statistical measure of complexity of hard-sphere gas: applications to nuclear matter'
---
Introduction
============
There has been a remarkable growth in research focused on complexity and in general on information theories in recent years in a variety of fields [@Pines-99] including physics [@Bialynicki75], chemistry [@Seurs-80], biology [@Adami-04], neuroscience [@Borst-99], mathematics [@Cover] and computer science [@Nillsen]. In particular there are various applications in quantum many-body systems e.g. atoms [@Chatzisavvas05; @Chatzisavvas07a; @Chatzisavvas07b], nuclei [@Panos97; @Massen98; @Moustakidis05; @Chatzisavvas-05], atomic clusters [@Massen98], bosonic traps [@Massen98] e.t.c. In fact, they lead to the clarification of basic quantum concepts and provide results about the information content of systems according to various definitions e.g. Shannon information entropy [@Shannon48], Fisher [@Fisher25] information, Onicescu information energy [@Onicescu66] e.t.c. Also they represent a suitable framework, via a probabilistic description, to assess the presence of interactions, correlations with experimentally measurable quantities, the derivation of universal relations e.t.c [@Panos97; @Massen98; @Moustakidis05; @Shannon48; @Fisher25; @Onicescu66; @Kullback59; @Renyi61; @Gadre84; @Ohya93; @BAB95; @Nagy96; @Majer96; @Ghosh84; @Garba06; @Frieden04; @Patil07; @Liu07; @Luzanov07; @Antolin08; @Sagar08]. Thus, traditional methods can be extended by an alternative information-theoretic way and give new insights for the treatment of simple quantum systems as well as more complicated many-body ones. A recent and important advance is to calculate several complexity measures, based on a probabilistic description via previous experience on information entropy in order to quantify statistical indicators of complex behavior in different systems scattered in a broad spectrum of fields [@Chatzisavvas05; @Chatzisavvas07a; @Chatzisavvas07b; @Lopez95; @Catalan02; @Sanudo09; @Landsberg-98; @Shiner-99; @Crutch98; @Crutch00; @Binder00; @Martin03; @Panos-09; @Lopez05; @Yamano04; @Plastino96; @Borgoo07; @Montogomery08; @Calbet-07; @Angulo08; @Angulo08b; @Sanudo08; @Sanudo08b; @Chatzi-09; @Roza-09-a; @Ruiz-09; @Antolin09]. Related research started connecting the above measures with experiment e.g. Fisher information entropy has been found to correlate with the ionization potential and dipole polarizability in atoms [@Chatzisavvas07a] and also complexity in a correlated Fermi gas has been connected with the specific heat $C_V$ e.t.c. [@Moustakidis-010].
The statistical measure of complexity $C_{LMC}$ introduced by López-Ruiz, Calbet and Mancini (LMC) [@Lopez95] is defined in the form of the product $C_{\rm LMC}=S D$. Here, $S$ is the information entropy i.e. the information content of the system, while $D$ is the disequilibrium i.e. the distance to the equilibrium probability distribution. Although complexity is a multi-faced quantity and several definitions of complexity measures have been proposed, the LMC measure has been employed recently in various studies for the following reasons: it exhibits the correct asymptotic properties of a well-behaved measure of complexity, as expected by intuition e.g. it vanishes for the two extreme cases of a perfect crystal (complete order) and ideal gas (complete disorder). In addition it is easily calculable for a quantum system, described by its very nature by probability densities leading to a feasible calculation of its basic factors $S$ and $D$. Other definitions of complexity, although sometimes may be considered that they describe certain aspects of complex systems in a satisfactory way, they have other disadvantages e.g. Kolmogorov’s algorithmic complexity is hard to compute. It is defined as the length of the shortest (optimum) program needed to describe the system, a goal difficult to attain and prove [@Kolmogorov-65]. Our approach is a pragmatic one: we start from the LMC definition, which is relatively easily calculable and hope to improve in the future, by a assessing the obtained results, comparing with other definitions of complexity e.g. the SDL measure according to Shiner, Davison and Landsber [@Shiner-99].
The initial definition of $C_{LMC}$ has been slightly modified in a suitable way by Catalan *et al* [@Catalan02], leading to the form $C=\e^S D$ applicable to systems described by either discrete or continuous probability distributions. In [@Catalan02] it was shown that the results in both, discrete and continuous cases, are consistent: extreme values of $C$ are observed for distributions characterized by a peak superimposed onto a uniform sea. Moreover, $C$ should be minimal, when the system reaches equipartition and the minimum value of $C$ is attained for rectangular (uniform) density probabilities giving the value $C=1$. Additionally, $C$ is not an upper bounded function and can become infinitely large.
The motivation of the present work, is to extend our previous study on complexity measures of uniform Fermi systems [@Moustakidis05], by employing the complexity measure proposed by López-Ruiz et al. [@Lopez95], using probability distributions in momentum space. In uniform systems the density $\rho=N/V$ is a constant and the interaction of the particles is reflected to the momentum distribution which deviates from the $theta$ function form of the ideal Fermi-gas model. Our aim is to connect $C$, a measure based on a probabilistic description and the shape of the corresponding momentum distributions to the phenomenological parameters introducing the inter-particle correlations.
The study of uniform quantum systems (both fermionic or bosonic) in momentum space is very important. Very interesting phenomena such as superfluidity, superconductivity, Bose-Einstein condensation e.t.c are observed and also well defined in momentum space. Thus, it is interesting to concentrate our study on the connection between complexity, defined in momentum space to the correlated behavior of a fermionic (or bosonic ) system, by employing the simple, but effective, hard-sphere model. Our specific application is nuclear matter. The basic model with hard-spheres is a suitable starting point in order to assess the relevance of various concepts and definitions of complexity. The present application in a correlated Fermi system like nuclear matter is facilitated by our previous experience. We use the simplest potential of a hard-sphere interaction with a hard core radius. The outline of the present work is the following: In Sec. 2 we present the momentum distribution and information measures employed to quantify complexity, while Sec. 3 contains our numerical results and discussion. Finally, in Sec. 4 we exhibit our conclusions.
Momentum distribution and information measures
==============================================
We adopt the formalism employed in our previous work [@Moustakidis05; @Moustakidis-010], adjusted here to a hard-sphere gas and specializing in nuclear matter.
Momentum distribution
---------------------
The momentum distribution (MD) of an interacting Fermi system is given in general by the relation $$n(k)=\frac{1}{V_k}\left\{
\begin{array}{rr}
n_{<}(k), \quad \text{for $k>k_F$}\\
n_{>}(k), \quad \text{for $k<k_F$}
\end{array} \right.
\label{nk-1}$$ where $V_k=\frac{4}{3}\pi k_F^3$. The Fermi wave number $k_F$ is related with the constant density $\rho=N \rho_0=3/(4\pi r_0^3)$ as follows $$k_F=\left( \frac{6 \pi^2 \rho}{\nu} \right)^{1/3}= \
\left(\frac{9 \pi}{2 \nu}\frac{1}{r_0^3} \right)^{1/3}. \label{kfermi}$$ The normalization of $n(k)$ obeys the relation $$\left(\frac{4\pi}{3}k_F^3\right)^{-1} \int n(k) d{\bf k}=1.
\label{norm-1}$$ The simplest form for $n(k)$ appears for an ideal Fermi gas. In this case $n(k)$ is just a step function $$n_0(k)=\frac{1}{V_k}\, \theta(k_F-k). \label{MD-IG}$$ The potential of the hard-sphere interaction is defined as follows, $$V(r)=\left\{
\begin{array}{rr}
\infty, \quad \text{for $r<c$}\\
0, \quad \text{for $r>c$},
\end{array} \right.
\label{hsi}$$ where $c$ denotes the hard core radius.
The momentum distribution of a hard-sphere dilute Fermi gas had previously been calculated by Czyz and Gottfried [@Czyz-61] and also by Sartor and Mahaux [@Sartor-80]. The above authors have studied a low density Fermi gas, whose particles interact via a repulsive hard core potential of the form (\[hsi\]). In this model, the quantities of interest can be expanded in powers of the parameter $(k_Fc)$.
The analytical expressions for the dimensionless $n(k)$ in the hard-sphere Fermi gas model have the following form for $k<k_F$ [@Sartor-80] $$\begin{aligned}
n_{<}(x)=&1&-\frac{\nu-1}{3\pi^2 x}(k_Fc)^2 \left[ \frac{}{}(7
\ln2-8)x^3+(10-3\ln 2)x \right. \nonumber \\
&+&\left. 2\ln \frac{1+x}{1-x}-2(2-x^2)^{3/2} \ln
\frac{(2-x^2)^{1/2}+x}{(2-x^2)^{1/2}-x} \right ], \label{nk-l}\end{aligned}$$ where $x=k/k_F$ and $\nu=4$. For $1<x<\sqrt{2}$: $$\begin{aligned}
n_{>}(x)&=&\frac{\nu-1}{6\pi^2 x}(k_Fc)^2\left((7x^3-3x-6)\ln
\frac{x-1}{x+1}+(7x^3-3x+2)\ln2 -8x^3+22x^2+6x
\right. \\
&-&\left. 24+2(2-x^2)^{3/2}\left[
\ln\frac{2+x+(2-x^2)^{1/2}}{2+x-(2-x^2)^{1/2}}+
\ln\frac{1+(2-x^2)^{1/2}}{1-(2-x^2)^{1/2}}-
2\ln\frac{x+(2-x^2)^{1/2}}{x-(2-x^2)^{1/2}}\right]\right).
\nonumber \label{nk-r-1}\end{aligned}$$ For $\sqrt{2}<x<3$: $$\begin{aligned}
n_{>}(x)&=&\frac{\nu-1}{6\pi^2 x}(k_Fc)^2\left((7x^3-3x-6)\ln
\frac{x-1}{x+1}+(7x^3-3x+2)\ln2 -8x^3+22x^2+6x
\right. \\
&-&\left. 4(x^2-2)^{3/2}\left[ \tan^{-1}\frac{x+2}{(x^2-2)^{1/2}}+
\tan^{-1}\frac{1}{(x^2-2)^{1/2}}-
2\tan^{-1}\frac{x}{(x^2-2)^{1/2}}\right]\right). \nonumber
\label{nk-r-2}\end{aligned}$$ For $3<x$: $$\begin{aligned}
n_{>}(x)&=&2\frac{\nu-1}{3\pi^2 x}(k_Fc)^2\left(2\ln
\frac{x+1}{x-1}-2x+(x^2-2)^{3/2}
\right. \\
&\times&\left. \left[2\tan^{-1}\frac{x}{(x^2-2)^{1/2}}-
\tan^{-1}\frac{x-2}{(x^2-2)^{1/2}}-\tan^{-1}\frac{x+2}{(x^2-2)^{1/2}}
\right ]\right). \nonumber \label{nk-r-3}\end{aligned}$$
Another characteristic quantity, used as a measure of the strength of correlations of uniform Fermi systems, is the discontinuity, $Z$, of the momentum distribution at $k/k_F=1$. It is defined as $$Z=n(1^-)-n(1^+).
\label{Z-1}$$
The behavior of the momentum distribution, as a function of $x=k/k_F$ for various values of the correlation parameter $k_Fc$ is shown in Fig. 1. The discontinuity $Z$ is also displayed in each case. For ideal Fermi systems $Z=1$, while for interacting ones $Z<1$. In the limit of very strong interaction where $Z
\rightarrow 0$, there is no discontinuity in the momentum distribution of the system. The quantity ($1-Z$) measures the ability of correlations to deplete the Fermi sea by exciting particles from states below it (hole states) to states above it (particle states) [@Flyn84].
The asymptotic behavior of $n(1^-)$ for $x->1^{-}$ reads [@Sartor-80] $$n(1^-)\approx 1-\frac{2}{3\pi^2}(\nu-1)(k_Fc)^2\left[3\ln
2+1-3(x-1)\ln \mid x-1\mid +(6\ln 2 -\frac{15}{2})(x-1)\right],
\label{nim-1}$$ while for $x->1^{+}$ $$n(1^+)\approx \frac{2}{3\pi^2}(\nu-1)(k_Fc)^2\left[\frac{}{}3\ln
2-1-3(x-1)\ln (x-1) -(6\ln 2 -7)(x-1)\right]. \label{nip-1}$$ It is worthwhile to note the existence of a logarithmic singularity in the function $n(k)$ at $k=k_F$, a general feature of normal Fermi systems.
The discontinuity $Z$, according to Eqs (\[Z-1\]) and (\[nim-1\]), (\[nip-1\]) is given by $$Z=1-\frac{4}{\pi^2}\ln 2(\nu-1)(k_Fc)^2. \label{Z-2}$$ The energy per particle $E$ of the ground state of $\nu=4$-component fermion fluid of hard-spheres, in the low-density expansion, has been derived in Refs. [@Baker-82]. Accordingly, the energy $E$, in units of the ideal gas energy $E_0$, is given by [@Baker-82] $$E/E_0=e(y)\simeq 1+D_1y+D_2y^2+D_3y^3+D_4y^4\ln y,\qquad
y=k_Fc,\quad E_0=\frac{3}{5}\frac{\hbar^2 k_F^2}{2m}. \label{En-1}$$ where the coefficients $D_i$ are given in Table VI of Ref. [@Baker-82]. It is one of the aims of the present work to investigate the connection between the various information measures and complexity with experimental quantities (as the ground state energy and the discontinuity $Z$). In addition, we intend to produce not only qualitative but mostly quantitative results, by connecting the strength of the correlations with the above measures.
Information measures
--------------------
The information entropy in momentum space is given by the relation $$S_k=-\int n(k) \ln{n(k)} \, \d{\bf k} . \label{IE-Sk-1}$$ So, for an ideal Fermi gas, using Eq. (\[MD-IG\]), $S_{k}$ becomes $$S_k=S_0=\ln V_k=\ln \left(\frac{6 \pi^2}{\nu}\frac{1}{r_0^3}
\right). \label{IE-Sk-2}$$ For correlated Fermi systems, $S_k$, can be found from Eq. (\[IE-Sk-1\]) by employing Eq. (\[nk-1\]). $S_k$ is written now [@Moustakidis05] $$S_k=\ln V_k-\frac{4 \pi}{V_k}\left( \int_{0}^{k_F^{-}} k^2
n_{<}(k)\ln n_{<}(k) dk+ \int_{k_F^{+}}^{\infty} k^2 n_{>}(k) \ln
n_{>}(k) dk \right). \label{Cor-Sk-1}$$ The correlated entropy $S_k$ has the form $$S_k=S_0+S_{\rm cor}, \label{Cor-S-1}$$ where $S_0$ is the uncorrelated entropy given by Eq. (\[IE-Sk-2\]) and $S_{\rm cor}$ is the contribution of the particle correlations to the entropy. That contribution can be found from the expression $$S_{\rm cor}=-3\left( \int_{0}^{1^{-}} x^2 n_{<}(x)\ln n_{<}(x) dx+
\int_{1^{+}}^{\infty} x^2 n_{>}(x)\ln n_{>}(x) dx \right),
\label{S-cor}$$ where $x=k/k_F$.
The disequilibrium $D_k$ (or information energy, defined by Onicescu [@Onicescu66]), in momentum space as another functional of a probability distribution, in our case $n(k)$ is given by the relation $$D_k=\int n^2(k) d{\bf k} . \label{D-1}$$ For an ideal Fermi gas, using Eq. (\[MD-IG\]), becomes $$D_k=D_0=\frac{1}{V_k}. \label{D-ideal}$$ In the case of correlated Fermi systems, $D_k$ is written as $$D_k=\frac{1}{V_k} \frac{4\pi}{V_k}\left( \int_{0}^{k_F^{-}} k^2
n^2_{<}(k) dk+ \int_{k_F^{+}}^{\infty} k^2 n^2_{>}(k)dk \right).
\label{D-Cor-1}$$ The correlated disequilibrium $D_k$ is $$D_k=D_0 D_{\rm cor}, \label{D-Cor-2}$$ where $D_0$ is given in Eq. (\[D-ideal\]) and $D_{cor}$ can be found also from the expression $$D_{cor}=3\left( \int_{0}^{1^{-}} x^2 {n}^2_{<}(x)dx+
\int_{1^{+}}^{\infty} x^2 {n}^2_{>}(x) dx \right). \label{D-cor-3}$$ The statistical complexity measure, proposed by Catalan et. al. [@Catalan02], in momentum space, is defined as $$C_{\rm LMC}=C=D_k H_k, \label{C-1}$$ where $H$ represents the information content of the system defined as $$H_k=\e^{S_k}. \label{H-1}$$ It is easy to show that $$C=C_0 C_{\rm cor}=D_{\rm cor} \e^{S_{\rm cor}}, \qquad C_0=D_0
\e^{S_0}=1. \label{C-2}$$ The physical meaning of Eq. (\[C-2\]) is clear. In the case of an ideal Fermi gas (see Eq. (\[MD-IG\])) $C$ is minimal with the value $C_0=1$ (see also [@Catalan02]). Moreover as pointed out in Ref. [@Catalan02], $C$ is not an upper bounded function and can therefore become infinitely large. From the above analysis it is clear that complexity $C$ is an accounter of correlations in an infinite Fermi system. Thus, the next step is to try to find the connection between $C$ and the correlation parameters of the systems. The correlation invoke diffusion of the momentum distribution and we expect this effect to be reflected on the values of $C$.
Results and discussion
======================
The behavior of the momentum distribution, as a function of $k/k_F$, for various values of the wound parameter $k_Fc$ is shown in Fig. 1. The discontinuity $Z$ is also displayed in each case. For ideal Fermi systems $Z=1$, while for interacting ones $Z<1$. In the limit of very strong interaction $Z=0$, there is no discontinuity on the momentum distribution of the system.
The calculated values of $S_{\rm cor}$, $D_{\rm cor}$ and $C$ for nuclear matter versus the correlation parameter $k_Fc$ are displayed in Fig. 2. $S_{\rm cor}$ and $C$ increase with $y$, while $D_{\rm cor}$ decreases. We fitted the numerical values of the above quantities, with simple functions of $y=k_Fc$ and we find respectively the following formulae $$S_{\rm cor}(y)=\alpha y^{\beta}, \qquad \alpha=2.16379, \quad
\beta=1.67053. \label{Scor-fit-1}$$ $$D_{\rm cor}(y)=1+\alpha y^{\beta}, \qquad \alpha=-0.79871, \quad
\beta=1.83155. \label{Dcor-fit-1}$$ $$C(y)=1+\alpha y^{\beta}, \qquad \alpha=1.68358, \quad
\beta=1.67566. \label{Ccor-fit-1}$$ The values of the parameters $\alpha$, $\beta$ and $\gamma$, for each case, have been selected by a least squares fit (LSF) method. It is worthwhile to notice that for an ideal Fermi gas there is an upper limit for $C$ ($C_{max}\simeq 1.5845$) [@Moustakidis-010]. However, for an interacting Fermi gas there is no such constraint (at least to the region under consideration in the present work). It is obvious that the calculated values of complexity reflect the different way that the interaction or temperature affect the trend of the momentum distribution.
The quantity ($1-Z$) measures the ability of correlations to deplete the Fermi sea by exciting particles from states below it (hole states) to states above it (particle states) [@Flyn84]. The dependence of $S_{\rm cor}$, $D_{\rm cor}$ and $C$ on the quantity $(1-Z)$ is shown in Fig. 3. It is seen that $S_{\rm cor}$ and $C$ are increasing functions of $(1-Z)$, while $D_{\rm cor}$ is a decreasing one, as a direct consequence of the dependence of the above quantities on the correlation parameter $k_Fc$. That dependence can be reproduced very well by simple expressions as in Eqs. (\[Scor-fit-1\]), (\[Dcor-fit-1\]) and (\[Ccor-fit-1\]) replacing $k_Fc$ by $(1-Z)$ $$S_{\rm cor}(Z)=\alpha (1-Z)^{\beta}, \qquad \alpha=2.49614, \quad
\beta=0.83527. \label{Scor-fit-2}$$ $$D_{\rm cor}(Z)=1+\alpha (1-Z)^{\beta}, \qquad \alpha=-0.93413,
\quad \beta=0.91574. \label{Dcor-fit-2}$$ $$C(Z)=1+\alpha (1-Z)^{\beta}, \qquad \alpha=1.94305, \quad
\beta=0.83784. \label{Ccor-fit-2}$$
From the above analysis we can conclude that LMC complexity $C$ can be employed as a measure of the strength of correlations in the same way the wound and the discontinuity parameters are used. An explanation of the above behavior of $C$ is the following: The effect of nucleon correlations is the departure from the step function form of the momentum distribution (ideal Fermi gas) to the one with long tail behavior for $k>k_F$. The diffusion of the distribution leads to a decrease of the order of the system (the disequilibrium $D_k$ decreases and the information entropy $S_k$ increases accordingly). In total, the contribution of $S_k$ in $C$ dominates over the contribution of $D_k$ and thus the complexity increases with the correlations (at least in the region under consideration).
It is one of the aims of the present work to connect $C$, a measure of complexity based on a probabilistic description and the shape of the corresponding momentum distributions with other quantities like the ground state energy per particle $E$. In view of the above, we display in Fig. 3 the dependence of $S_{cor}$, $D_{cor}$ and $C$, as well as the energy fraction $e(y)$, on the correlation parameter $k_Fc$. The dependence of $S_{cor}$, $D_{cor}$ and $C$ on $e(y)$, as displayed in Fig. 4, is in a very good approximation linear. The fitted expressions are the following: $${\rm e}=1.0479+1.27353 S_{cor}, \label{s-e}$$ $${\rm e}=4.6861-3.5985 D_{cor}, \label{d-e}$$ $${\rm e}=-0.5847+1.6358 C. \label{c-e}$$ In total we observe an empirical connection of the energy with $S_{cor}$, $D_{cor}$ and $C$ calculated employing information entropy, which, by definition, is not related directly to the energy of the system, in contrast to the traditional concept of thermodynamic entropy. The above results confirm our recent finding, according to similar lines, that there is also a connection between the “energy-like” quantity specific heat $C_V$ of an ideal electron gas with the complexity $C$ [@Moustakidis-010].
Finally, in Fig. 5 we compare the present results with those taken from the Low Order Approximation (LOA) method. Thus the momentum distribution of nuclear matter is evaluated by employing the LOA and the MD takes the form [@Flyn84] $$n_{\rm LOA}(k)=\theta(k_F-k) \left[ 1-k_{\rm dir}+Y(k,8) \right]+
8 \left[ k_{\rm dir}Y(k,2)-[Y(k,4)]^2 \right], \label{mn-mom-1}$$ where $$c_{\mu}^{-1}Y(k,\mu)=
\frac{\e^{-\tilde{k}_{+}^{2}}-\e^{-\tilde{k}_{-}^{2}}}{2\tilde{k}}
+\int_{0}^{\tilde{k}_{+}} \e^{-y^2} \, \d y +
{\rm sgn}(\tilde{k}_{-}) \int_{0}^{\mid \tilde{k}_{-} \mid} \e^{-y^2} \d y,
$$ and $$c_{\mu}=\frac{1}{8\sqrt{\pi}}\left(\frac{\mu}{2}\right)^{3/2},
\quad \tilde{k}=\frac{k}{\beta \sqrt{\mu}} , \quad
\tilde{k}_{\pm}=\frac{k_{F}\pm k}{\beta \sqrt{\mu}}, \quad
\mu=2,4,8.
$$ while ${\rm sgn}(x)=x/\mid x \mid$. The dimensionless wound parameter $k_{\rm dir}$ can serve as a rough measure of correlations and the rate of convergence of the cluster expansion is defined as $$k_{\rm dir}=\rho \int [f(r)-1]^2 \, \d {\bf r}. \label{eq-kdir}$$ The normalization condition for the momentum distribution is $$\int_{0}^{\infty} n_{\rm LOA}(k)k^2 \d k=\frac{1}{3}\, k_{F}^{3}.$$ The following relation between the wound parameter $k_{\rm dir}$ and the correlation parameter $\beta$ $$k_{\rm dir}=\frac{1}{3 \sqrt{2\pi}}\left(\frac{k_F}{\beta}
\right)^3.
$$ It is clear that large values of $k_{\rm dir}$ imply strong correlations and simultaneously poor convergence of the cluster expansion. In the numerical calculations the correlation parameter $\beta$ is in the interval: $1.01 \leq \beta \leq 2.482 $. That range corresponds to $ 0.3 \geq k_{\rm dir} \geq 0.02 $ and this is reasonable, in the case of nuclear matter [@Flyn84]. However, the origin of the two methods is different and as a consequence they influence in a different way the momentum distribution. Nevertheless, we found that $S_{cor}$, $D_{cor}$ and $C$ exhibit a similar trend as a functions of the parameter $1-Z$.
Conlusions
==========
In conclusion, we calculate information and complexity measures of a uniform Fermi system, like nuclear matter, in the framework of the hard-sphere model. The effect of correlations is connected intuitively with the concept of complexity, in a qualitative, and somehow vague way as stated in [@Moustakidis-010] as well. In fact, it turns out that all information measures used by us, show a strong dependence on the correlation parameter $k_Fc$ as well as on the Fermi discontinuity $Z$. The most distinctive feature is the occurrence of an almost linear dependence between information measures and complexity on energy. The above statement is in keeping with the recent finding of the existence of an empirical connection between the specific heat and complexity [@Moustakidis-010]. However, the applicability of our approach is much wider than nuclear matter, since the impenetrable hard spheres (not overlapping in space) can simulate the extremely strong repulsion that atoms and spherical molecules feel at very small distances. Thus, the significance of a suitable quantification of complexity emerges in statistical mechanics of fluids and solids. We do not claim that our approach is the only or more important one, but so far our results are interesting and encouraging. We stress again that the proposed LMC measure of complexity is by its definition an appropriate one and specifically tailored for systems described statistically, through a probability distribution. Since “information is physical” [@DiVincenzo], it is promising to examine how far this quotation goes, in the sense that calculations employing a good measure of information content of a quantum system (and consequently of complexity) are expected to give, at least, interesting results of physical relevance. Landauer dedicated his research on similar ideas. Hence, in our present work and previous ones, we proceed towards a numerical quantification of complexity. One of our goals is to examine, as a first step, whether a particular definition of complexity is reasonable and robust enough e.g. if one increases the value of one parameter (or parameters) describing a quantum system, the corresponding value of complexity increases accordingly. This would satisfy minimally intuition about complexity and validate its definition.
[99]{} G.A. Cowan, D. Pines, and D. Meltzer, [*Complexity: Metaphors, Models and Reality*]{}, (Addison-Wesley, Redwood City, 1999). I. Bialynicki-Birula and J. Mycielski, Commun. Math. Phys. [**44**]{}, (1975) 129. S.B. Sears, [*Applications of information theory in chemical physics*]{}, Ph.D. Thesis, Univiversity of North Caroline at Chapel Hill (1980). C. Adami, [*Information theory in molecular bilogy*]{}, Phys. of Life Reviews, 1 (2004). A. Borst and F.E. Theunissen, Nature Neuroscience [**2**]{}, 947 (1999). T.M. Cover and J.A. Thomas, [*Element of Information Theory*]{} Wiley Series in Telecommunitions. M.A. Nillsen and I.L. Chuang, [*Quantum Computation and Quantum Informtion*]{} Cambridge University Press. K.Ch. Chatzisavvas, Ch.C. Moustakidis, and C.P. Panos, J. Chem. Phys. [**123**]{}, 174111 (2005). K.D. Sen, C.P. Panos, K.Ch. Chatzisavvas, and Ch.C. Moustakidis, Phys. Lett. A [**364**]{}, 286 (2007). C.P. Panos, K.Ch. Chatzisavvas, Ch.C. Moustakidis, and E.G. Kyrkou, Phys. Lett. A [**363**]{}, 78 (2007). C.P. Panos and S.E. Massen, Int.J. Mod. Phys. [**[E 6]{}**]{}, 497 (1997); G.A. Lalazissis, S.E. Massen, C.P. Panos, and S.S. Dimitrova, Int. J. Mod. Phys. [**[E 7]{}**]{}, 485 (1998). S.E. Massen and C.P. Panos, Phys. Lett. A [**246**]{}, 530 (1998); S.E. Massen and C.P. Panos, Phys. Lett. A [**280**]{}, 65 (2001); S.E. Massen, Ch.C. Moustakidis, and C.P. Panos, Phys. Lett A [**299**]{}, 131 (2002); C.P. Panos, Phys. Lett. A [**289**]{}, 287 (2001). Ch.C. Moustakidis and S.E. Massen, Phys. Rev. B [**71**]{}, 045102 (2005). K.Ch. Chatzisavvas and C.P. Panos, Int. J. Mod. Phys. E, [**14**]{}, 653 (2005). C.E. Shannon, Bell Syst. Tech. J. [**27**]{}, 379 (1948). R.A. Fisher, *Proc. Cambridge Phil. Sec.* **22**, 700 (1925). O. Onicescu, Theorie de l’information. Energie informationelle., Vol. [**263**]{} of A, C. R. Acad. Sci. Paris, 1966. S. Kullback, *Statistics and Information theory*, (Wiley, New York, 1959). A. Rényi *On measures of information and entropy*, Proceedings of the 4th Berkeley Symposium on Mathematics, Statistics and Probability 1960, pp. 547-561 (1961). S.R. Gadre, Phys. Rev. A [**30**]{}, 620 (1984); S.R. Gadre, S.B. Sears, S.J. Chakravorty, and R.D. Bendale, Phys. Rev. A [**32**]{}, 2602 (1985); S.R. Gadre and R.D. Bendale, Phys. Rev. A [**36**]{}, 1932 (1987). M. Ohya and P. Petz, “Quantum entropy and its use” (Springer Berlin, 1993). V. Zelevinsky, M. Horoi, and B.A. Brown, Phys. Lett. B [**350**]{}, 141 (1995). V.V. Sokolov, B.A. Brown, and V. Zelevinsky, Phys. Rev E [**58**]{}, 56 (1998). A. Nagy and R.G. Parr, Int. J. Quant. Chem. [**58**]{}, 323 (1996). V. Majernic and T. Opatrny, J. Phys. A [**29**]{}, 2187 (1996). S.K. Ghosh, M. Berkowitz, and R.G. Parr, Proc. Natl. Acad. Sc. USA [**81**]{}, 8028 (1984). P. Garbaczewski, J. Stat. Phys. [**123**]{}, 315 (2006). B.R. Frieden, *Science from Fisher Information*, (Cambridge University Press, 2004). S.H. Patil, K.D. Sen, N.A. Watson, and H.E. Montogomery, J. Phys. B [**40**]{}, 2147 (2007). S.B. Liu, J. Chem. Phys. [**126**]{}, 191107 (2007). A.V. Luzanov and O.V. Prezhdo, Mol. Phys. [**105**]{}, 2879 (2007). J. Antolín and J.C. Angulo, Eur. Phys. J. D [**46**]{}, 21 (2008). R.P. Sagar and N.L. Guevara, J. Mol. Struct. (Theochem) [**857**]{}, 72 (2008). R. López-Ruiz, H.L. Mancini, and X. Calbet, Phys. Lett. A [**209**]{}, 321 (1995). R.G. Catalán, J. Garay, and R. López-Ruiz, Phys. Rev. E [**66**]{}, 011102 (2002). J. Sañudo and A.F. Pacheco, Phys. Lett. A [**373**]{}, 807 (2009). P.T. Landsberg and J.S. Shiner, Phys. Lett. A [**245**]{}, 228 (1998); P.T. Landsberg, Phys. Lett. A [**102**]{}, 171 (1984). J.S. Shiner, M. Davison, and P.T. Landsberg, Phys. Rev. E [**59**]{}, 1459 (1999). D.P. Feldman and J.P. Crutchfield, Phys. Lett. A [**238**]{}, 244 (1998). J.P. Crutchfield, D.P. Feldman, and C.R. Shalizi, Phys. Rev. E [**62**]{}, 2996 (2000). P.M. Binder and N. Perry, Phys. Rev. E [**62**]{}, 2998 (2000). M.T. Martin, A. Plastino, and O.A. Rosso, Phys. Lett. A [**311**]{}, 126 (2003). C.P. Panos, N.S. Nikolaidis, K.Ch. Chatzisavvas, and C.C. Tsouros, Phys. Lett. A [**373**]{}, 2343 (2009). R. López-Ruiz, Biophys. Chem. [**115**]{}, 215 (2005). T. Yamano, J. Math. Phys. [**45**]{}, 1974 (2004). C. Anteneodo and and A.R. Plastino, Phys. Lett. A [**223**]{}, 348 (1996). A. Borgoo, F. De Proft, P. Geerlings, and K.D. Sen, Chem. Phys. Lett. [**444**]{}, 186 (2007). H.E. Montgomery and K.D. Sen, Phys. Lett. A [**372**]{}, 2271 (2008). X. Calbet and R. López-Ruiz, Physica A [**382**]{}, 523 (2007). J.C. Angulo and J. Antolín, J. Chem. Phys. [**128**]{}, 164109 (2008). J.C. Angulo, J. Antolín, and K.D. Sen, Phys. Lett. A [**372**]{}, 670 (2008). J. Sañudo and R. López-Ruiz, Phys. Lett. A [**372**]{}, 5283 (2008). J. Sañudo and R. López-Ruiz, J. Phys. A [**41**]{}, 265303 (2008). K.Ch. Chatzisavvas, V.P. Psonis, C.P. Panos, and Ch.C. Moustakidis, Phys. Lett. A [**373**]{}, 3901 (2009). S. López-Rosa, D. Manzano, and J.S. Dehesa, Physica A [**388**]{}, 3273 (2009). R. López-Ruiz, A. Nagy, E. Romera, and J. Sañudo, Journ. of Mathem Phys. [**50**]{}, 123528 (2009). Ch.C. Moustakidis, V.P. Psonis, K.Ch. Chatzisavvas, C.P. Panos, and S.E. Massen, Phys. Rev. E [**81**]{}, 011104 (2010). J. Antolín, S. Lopéz-Rosa, J.C. Angulo, Chem. Phys. Lett. **474** 233 (2009). A.N. Kolmogorov, Probl. Inf. Transm. [**1**]{}, 3 (1965). W. Czyż and K. Gottfried, Nucl. Phys. [**21**]{}, 676 (1961). R. Sartor and C. Mahaux, Phys. Rev. C [**21**]{}, 1546 (1980); R. Sartor and C. Mahaux, Phys. Rev. C [**25**]{}, 677 (1982). M.F. Flynn, J.W. Clark, R.M. Panoff, O. Bohigas, and S. Stringari, Nucl. Phys. [**A427**]{}, 253 (1984). G.A. Baker Jr., L.P. Benofy, M. Fortes, M. de Llano, S.M. Peltier and A. Plastino, Phys. Rev. A [**26**]{}, 3575 (1982); G.A. Baker Jr., G. Gutierrez, and M.de Llano, Ann. of. Phys. [**153**]{}, 283 (1984); M.A. Solis, M. De Llano, J.W. Clark, and G.A. Baker Jr., Phys. Rev. E [**76**]{}, 031125 (2007). D.P. Di Vincenzo and D. Loss, [*Quantum Information is Physical*]{} ArXiV: cond-mat/9710259.
![(a) The momentum distribution for correlated nuclear matter versus $k/k_F$ for various values of the correlation parameter $k_Fc$ (b) The same, but on a logarithmic scale, for $n(x)$.[]{data-label=""}](fig-1a.eps "fig:"){height="7.0cm" width="7.0cm"}![(a) The momentum distribution for correlated nuclear matter versus $k/k_F$ for various values of the correlation parameter $k_Fc$ (b) The same, but on a logarithmic scale, for $n(x)$.[]{data-label=""}](fig-1b.eps "fig:"){height="7.0cm" width="7.0cm"}
![(a) $S_{cor}$, $D_{cor}$ and $C$ of nuclear matter versus the the correlation parameter $k_Fc$ and (b) the discontinuity parameter $(1-Z)$ []{data-label=""}](fig-2a.eps "fig:"){height="7.0cm" width="7.0cm"}![(a) $S_{cor}$, $D_{cor}$ and $C$ of nuclear matter versus the the correlation parameter $k_Fc$ and (b) the discontinuity parameter $(1-Z)$ []{data-label=""}](fig-2b.eps "fig:"){height="7.0cm" width="7.0cm"}
![$S_{cor}$, $D_{cor}$, $C$ and the energy fraction $e$ (for definition see Eq. (\[En-1\])) versus the correlation parameter $k_Fc$. []{data-label=""}](fig-3.eps){height="7.0cm" width="7.0cm"}
![The energy fraction $e$ versus (a) $S_{cor}$ (b) $D_{cor}$ and (c) $C$. The lines corresponds to the expressions (\[s-e\]), (\[d-e\]) and (\[c-e\]), respectively with the parameters derived by the least squares fit method (LSF). []{data-label=""}](fig-4a.eps "fig:"){height="6.5cm" width="5.5cm"}![The energy fraction $e$ versus (a) $S_{cor}$ (b) $D_{cor}$ and (c) $C$. The lines corresponds to the expressions (\[s-e\]), (\[d-e\]) and (\[c-e\]), respectively with the parameters derived by the least squares fit method (LSF). []{data-label=""}](fig-4b.eps "fig:"){height="6.5cm" width="5.5cm"}![The energy fraction $e$ versus (a) $S_{cor}$ (b) $D_{cor}$ and (c) $C$. The lines corresponds to the expressions (\[s-e\]), (\[d-e\]) and (\[c-e\]), respectively with the parameters derived by the least squares fit method (LSF). []{data-label=""}](fig-4c.eps "fig:"){height="6.5cm" width="5.5cm"}
![$S_{cor}$, $D_{cor}$ and $C$ of nuclear matter versus the the discontinuity parameter $(1-Z)$ calculated by employing two different methods: the Low Order Approximation (LOA) (see text) and the hard sphere interaction (HS).[]{data-label=""}](fig-5.eps){height="7.0cm" width="7.0cm"}
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We present a class of new black hole solutions in $D$-dimensional Lovelock gravity theory. The solutions have a form of direct product $\mathcal{M}^m \times \mathcal{H}^{n}$, where $D=m+n$, $\mathcal{H}^n$ is a negative constant curvature space, and the solutions are characterized by two integration constants. When $m=3$ and $4$, these solutions reduce to the exact black hole solutions recently found by Maeda and Dadhich in Gauss-Bonnet gravity theory. We study thermodynamics of these black hole solutions. Although these black holes have a nonvanishing Hawking temperature, surprisingly, the mass of these solutions always vanishes. While the entropy also vanishes when $m$ is odd, it is a constant determined by Euler characteristic of $(m-2)$-dimensional cross section of black hole horizon when $m$ is even. We argue that the constant in the entropy should be thrown away. Namely, when $m$ is even, the entropy of these black holes also should vanish. We discuss the implications of these results.'
address:
- '$^{a}$ Key Laboratory of Frontiers in Theoretical Physics, Institute of Theoretical Physics, Chinese Academy of Sciences, P.O. Box 2735, Beijing 100190, China\'
- '$^{b}$ Department of Physics, Kinki University, Higashi-Osaka, Osaka 577-8502, Japan'
author:
- 'Rong-Gen Cai$^{a,b,}$[^1], Li-Ming Cao$^{b,}$[^2], Nobuyoshi Ohta$^{b,}$[^3]'
title: Black Holes without Mass and Entropy in Lovelock Gravity
---
Introduction
============
With the development of string theory, supergravity and brane world scenarios, over the past years, gravity theories have been widely studied in higher dimensions. In the low-energy approximation, Einstein general relativity naturally arises from string theories. As corrections from massive states of string theories and from loop expansions in string theories, some higher derivative curvature terms also appear in the low-energy effective action of string theories [@Zwie; @Lowenergylimit]. Therefore it is of great interest to discuss potential roles of those higher derivative terms in various aspects, for example, in black hole physics and early universe. Indeed, there exist a lot of works on higher derivative gravity theories in the literature.
In this paper we focus on a class of special higher derivative gravity theory, namely, Lovelock gravity [@Love], which is a natural generalization of general relativity in higher dimensions in the sense that the equations of motion of Lovelock gravity do not contain more than second order derivatives with respect to metric, as the case of general relativity. The Lagrangian of $D$-dimensional Lovelock gravity consists of the dimensionally extended Euler densities $$\label{1eq1} {\cal L} = \sum^p_{k=0}c_k {\cal L}_k,$$ where $p\leq [(D-1)/2]$($[N]$ denotes the integral part of the number $N$), $c_k$ are arbitrary constants with dimension of $[\mathrm{Length}]^{2k-2}$, and ${\cal L}_k$ are the Euler densities $${\cal L}_k=\frac{1}{2^k}\sqrt{-g}\delta^{\mu_1\cdots
\mu_k\nu_1\cdots\nu_k}_{\lambda_1\cdots
\lambda_k\sigma_1\cdots\sigma_k}R^{\lambda_1\sigma_1}_{~~~~
\mu_1\nu_1}\cdots R^{\lambda_k\sigma_k}_{~~~~ \mu_k\nu_k}\, .
\label{Lk}$$ Here, the generalized delta is totally antisymmetric in both sets of indices. ${\cal L}_0=1$, so the constant $c_0$ is just the cosmological constant. ${\cal L}_1$ gives us the usual curvature scalar term, and for simplicity, we set $c_1=1$, while ${\cal L}_2$ is just the Gauss-Bonnet term. The Gauss-Bonnet term is argued to appear in the low-energy action of heterotic string theory with a positive coefficient [@Zwie]. The equations of motion following from the Lagrangian (\[1eq1\]) have the form $\mathcal{G}_{\mu\nu}=0$, where $$\mathcal{G}^{\mu}_{\nu}=\sum_{k=0}^{p}\frac{1}{2^{k+1}}
c_{k}\delta^{\mu\mu_1\cdots
\mu_k\nu_1\cdots\nu_k}_{\nu\lambda_1\cdots
\lambda_k\sigma_1\cdots\sigma_k}R^{\lambda_1\sigma_1}_{~~~~
\mu_1\nu_1}\cdots R^{\lambda_k\sigma_k}_{~~~~ \mu_k\nu_k} \,
.\label{em}$$ Since the action of Lovelock gravity is the sum of the dimensionally extended Euler densities, there are no more than second order derivatives with respect to metric in its equations of motion. Furthermore, the Lovelock gravity is shown to be free of ghost when expanded on a flat space, evading any problems with unitarity [@Zwie; @Des].
Finding exact analytic solutions of any gravity theory is an issue of long-standing interest. Indeed, there exist a lot of works to discuss exact black hole solutions for Lovelock gravity in the literature. The static, spherically symmetric black hole solutions in the theory have been found in [@Des; @Whee; @Cai1; @Cai4; @Banados; @CaiSoh; @Zaneli] and discussed [@Myers], and topological nontrivial black holes have been studied in [@Cai1; @Cai4; @CaiSoh; @Zaneli]. Some rotating solutions in Gauss-Bonnet theory have been studied in [@rotating; @Anabalon:2009kq]. However, a general rotating solution is still absent even in Gauss-Bonnet gravity. See also [@others; @others1] for some other extensions including perturbative AdS black hole solutions in the gravity theories with second order curvature corrections. For a nice review of black holes in Lovelock gravity, see [@GG].
Recently, Maeda and Dadhich presented a class of exact solutions in Gauss-Bonnet gravity [@Maeda:2006iw; @Maeda:2006hj; @Dadhich:2007xf; @Molina:2008kh] (Some similar black hole solutions have been found in Codimension-2 brane world theory [@CuadrosMelgar:2007jx; @CuadrosMelgar:2008kn]). They assumed the spacetime has a direct product structure (one is a four/three-dimensional spacetime, the other is a negative constant curvature space), and then split the equations of motion into two sets according to the direct product structure of spacetime. For a suitable choice of those coefficients in Gauss-Bonnet gravity so that the set of equations of motion for the four/three-dimensional part is trivially satisfied, the set of the equations of motion for the negative constant curvature space part then reduces to a single equation. Solving the latter yields a class of new exact analytic solutions for Gauss-Bonnet gravity. This class of the solutions has two integration constants; one is argued to be related to the mass of the solutions and the other behaves like a Maxwell charge. The Maxwell charge is called “Weyl charge" due to the existence of the extra negative constant curvature space. This class of solutions is quite different from the normal ones in the sense which will become clear shortly. Quantum properties of those black hole solutions have not yet been studied so far.
In this paper, we consider a general Lovelock theory instead of the Gauss-Bonnet theory, and seek for more general black hole solutions and study their thermodynamics. The outline of the paper is as follows. In Sec. II, we present a class of new black hole solutions in the general Lovelock theory, following [@Maeda:2006iw; @Maeda:2006hj]. In Sec. III, we study thermodynamic properties of those black hole solutions. Sec. IV is devoted to conclusion and discussion.
General Black Hole Solutions
============================
Equations of Motion
-------------------
Consider a $D(=m+n)$-dimensional spacetime $X$, locally homeomorphic to $\mathcal{M}^m\times \mathcal{N}^n$. We assume the metric of this spacetime has the form $$\label{metric} ds^2=g_{ij}dx^i dx^j+r_{0}^2 \gamma_{ab}dy^a dy^b\,
,$$ where $g_{ij}dx^i dx^j$ is the metric on $\mathcal{M}^m$ with the coordinates $\{x^i,i=1,\cdots , m\}$, $r_{0}$ is a constant and $\gamma_{ab}dy^a dy^b$ is the metric on the $n$-dimensional space $\mathcal{N}^n$ with the coordinates $\{y^a,a=,1,\cdots,n\}$. $\mathcal{N}^n$ is a constant curvature space with curvature $\bar{k}=\pm 1,0$. It is easy to find that the components of the Riemann tensor for (\[metric\]) have the form $$\begin{aligned}
\label{riemann1}
R_{ijkl}=\bar{R}_{ijkl},\quad R_{ij}{}{}^{kl}=\bar{R}_{ij}{}{}^{kl}\, ,\nonumber \\
R_{abcd}=\tilde{R}_{abcd},\quad
R_{ab}{}{}^{cd}=\tilde{R}_{ab}{}{}^{cd}\, ,\end{aligned}$$ where $\bar{R}_{ijkl}$ and $\tilde{R}_{abcd}$ denote the components of Riemann tensor on $\mathcal{M}^m$ and $\mathcal{N}^n$ respectively. For $\mathcal{N}^n$, we can write the $\widetilde{R}_{abcd}$ as $$\tilde{R}_{abcd}=\bar{k} r_{0}^2
\left(\gamma_{ac}\gamma_{bd}-\gamma_{ad}\gamma_{bc}\right), \quad
\tilde{R}^{ab}{}{}_{cd}= \frac{\bar{k}}{r_{0}^2} \delta^{ab}_{cd}\,
. \label{riemann2}$$ According to the decomposition of the Riemann tensor (\[riemann1\]), we can decompose the equations of motion (\[em\]) into $m$-dimensional and $n$-dimensional parts: $$\begin{aligned}
\mathcal{G}^{i}{}_{j}&\equiv&\sum_{k=0}^{p}\frac{1}{2^{k+1}}c_{k}\delta^{i
\mu_1\cdots \mu_k\nu_1\cdots\nu_k}_{j\lambda_1\cdots
\lambda_k\sigma_1\cdots\sigma_k}R^{\lambda_1\sigma_1}_{~~~~
\mu_1\nu_1}\cdots R^{\lambda_k\sigma_k}_{~~~~ \mu_k\nu_k}=0 \, , \nonumber \\
\mathcal{G}^{a}{}_{b}&\equiv&\sum_{k=0}^{p}\frac{1}{2^{k+1}}c_{k}\delta^{a
\mu_1\cdots \mu_k\nu_1\cdots\nu_k}_{b\lambda_1\cdots
\lambda_k\sigma_1\cdots\sigma_k}R^{\lambda_1\sigma_1}_{~~~~
\mu_1\nu_1}\cdots R^{\lambda_k\sigma_k}_{~~~~ \mu_k\nu_k}=0 \, .
\label{emsplit}\end{aligned}$$ Other components (such as $\mathcal{G}^{i}{}_{a}$) automatically vanish. Since $i,j$ run only in the range $\{1,\cdots ,
m\}$, and $R_{ijkl}$ can appear in the products “$RR\cdots$" no more than $q=[(m-1)/2]$ times, we have $$\begin{aligned}
\label{2eq5}
\mathcal{G}^{i}{}_{j}&=&\sum_{k=0}^{p}\frac{c_{k}}{2^{k+1}}\delta^{i
\mu_1\cdots \mu_k\nu_1\cdots\nu_k}_{j\lambda_1\cdots
\lambda_k\sigma_1\cdots\sigma_k}R^{\lambda_1\sigma_1}_{~~~~
\mu_1\nu_1}\cdots R^{\lambda_k\sigma_k}_{~~~~ \mu_k\nu_k}\nonumber \\
&=&\sum_{t}^{q}\sum_{k=t}^{p} {k \choose t} \frac{c_{k}}{2^{k+1}}
\delta_{j m_{1} n_{1}\cdots m_{t} n_{t} e_{1} \cdots e_{k-t}
f_{1}\cdots f_{k-t}}^{i k_1 l_1 \cdots k_t l_t c_{1}\cdots c_{k-t}
d_{1} \cdots d_{k-t}}
R_{k_1 l_1}{}{}^{m_{1} n_{1}}\cdots R_{k_t l_t}{}{}^{m_{t} n_{t}}\nonumber \\
&& \times R_{c_{1}d_{1}}{}{}^{e_{1}f_{1}}\cdots
R_{c_{k-t}d_{k-t}}{}{}^{e_{k-t}f_{k-t}}\, .\end{aligned}$$ Substituting Eqs. (\[riemann1\]) and (\[riemann2\]) into (\[2eq5\]) and using the identity $$\delta_{\nu_{1}\cdots \nu_{p-1}\nu_{p}}^{\mu_{1}\cdots \mu_{p-1}
\mu_{p}}\delta_{\mu_{p-1}\mu_{p}}^{\nu_{p-1}\nu_{p}}
=2\left[r-(p-1)\right]\left[r-(p-2)\right]\delta_{\nu_{1}\cdots
\nu_{p-2}}^{\mu_{1}\cdots \mu_{p-2}}\, , \label{idenKD}$$ where $r$ denotes the range of the index ($r=m$ for $\mathcal{M}^m$ and $r=n$ for $\mathcal{N}^n$), we have $$\begin{aligned}
\label{emm}
\mathcal{G}^{i}{}_{j}&=&\sum_{t=0}^{q}\left[\sum_{k=t}^{p} {k
\choose t} c_k \frac{(D-m)!}{(D-m-2(k-t))!}
\left(\frac{\bar{k}}{r_{0}^2 }\right)^{k-t}
\right]\overline{\mathcal{G}}^{i}_{(t)j}\, ,\end{aligned}$$ where $$\overline{\mathcal{G}}^{i}_{(t)j}=\frac{1}{2^{t+1}}\delta^{ik_1 l_1
\cdots k_{t} l_{t}}_{jm_{1} n_{1}\cdots m_{t} n_{t}}R_{k_1
l_1}{}{}^{m_{1} n_{1}}\cdots R_{k_{t} l_{t}}{}{}^{m_{t} n_{t}}\, .$$ Similarly, we have $$\begin{aligned}
\mathcal{G}^{a}{}_{b}&=&\sum_{k=0}^{p}\frac{c_{k}}{2^{k+1}}\delta^{a
\mu_1\cdots \mu_k\nu_1\cdots\nu_k}_{b\lambda_1\cdots
\lambda_k\sigma_1\cdots\sigma_k}R^{\lambda_1\sigma_1}_{~~~~
\mu_1\nu_1}\cdots R^{\lambda_k\sigma_k}_{~~~~ \mu_k\nu_k}\nonumber \\
&=&\sum_{t=0}^{s}\sum_{k=t}^{p} {k \choose t} \frac{c_{k}}{2^{k+1}}
\delta_{b m_{1} n_{1}\cdots m_{t} n_{t} e_{1} \cdots e_{k-t}
f_{1}\cdots f_{k-t}}^{a k_1 l_1 \cdots k_t l_s c_{1}\cdots c_{k-t}
d_{1} \cdots d_{k-t}}
R_{k_1 l_1}{}{}^{m_{1} n_{1}}\cdots R_{k_t l_t}{}{}^{m_{t} n_{t}}\nonumber \\
&&\times R_{c_{1}d_{1}}{}{}^{e_{1}f_{1}}\cdots
R_{c_{k-t}d_{k-t}}{}{}^{e_{k-t}f_{k-t}}\, ,\end{aligned}$$ where $s=[m/2]$. Substituting Eqs. (\[riemann1\]), (\[riemann2\]) and (\[idenKD\]) into the above equation, we can express it as $$\begin{aligned}
\label{enn} \mathcal{G}^{a}{}_{b}&=& \frac{1}{2}\delta^{a}{}_{b}
\Bigg\{ \sum_{t=0}^{s}\left[\sum_{k=t}^{p} {k \choose t} c_k
\frac{(D-m-1)!}{(D-m-1-2(k-t))!} \left(\frac{\bar{k}}{r_{0}^2
}\right)^{k-t}\right]\bar{L}_t \Bigg\}\, ,\end{aligned}$$ where $$\bar{L}_t =\frac{1}{2^t}\delta^{k_1l_1\cdots
k_{t}l_{t}}_{m_1n_1\cdots m_{t}n_{t}}R_{k_1l_1}{}{}^{m_1n_1}\cdots
R_{k_{t}l_{t}}{}{}^{m_{t}n_{t}}\, . \label{Lbart}$$ Note that $\mathcal{G}^a_b$ is always proportional to $\delta^a_b$, which is a crucial point to our discussions below. Let us note that if the following equations are satisfied $$\label{lineq}
0=A_t \equiv \sum_{k=t}^{p} {k \choose t} c_k
\frac{(D-m)!}{(D-m-2(k-t))!} \left(\frac{\bar{k}}{r_{0}^2
}\right)^{k-t},\quad t=0, \cdots, q\, ,$$ then the equations of motion (\[emm\]) are always trivially satisfied. These are $(q+1)$-linear equations for $c_0,\cdots,c_p $. Recall $c_1=1$ and if we consider the case with $$p=q+1=\left[\frac{m-1}{2}\right]+1\, ,$$ then the equations (\[lineq\]) indicate that $c_k (k \neq 1)$ has a unique expression in terms of $\left(\bar{k}/r_{0}^2 \right)$ and dimension $D$. When $\bar{k}=-1$ (which implies that $\mathcal{N}^n$ is a negative constant curvature space. It will be denoted by $\mathcal{H}^n$ in the following discussion.) and $D\geq m+2$, we find that all $c_k$ are positive. Some examples will be given soon.
Now we turn to the equations of $\mathcal{G}_{ab}$. Due to the fact that $\mathcal{G}_{ab}$ is proportional to $\delta_{ab}$, the equations $\mathcal{G}_{ab}=0$ reduce to a single equation $$\label{2eq14}
0=\sum_{i=0}^{s}\alpha_i \bar{L}_i\, ,$$ where the coefficients $\alpha_i$’s are given by $$\alpha_i=\sum_{k=i}^{p} {k \choose i} c_k
\frac{(D-m-1)!}{(D-m-1-2(k-i))!} \left(\frac{\bar{k}}{r_{0}^2
}\right)^{k-i}\, ,$$ and $c_k$ are determined by solutions (\[lineq\]).
Let us further assume the $m$-dimensional metric $g_{ij}$ takes the form $$\label{m-2metric}
g=-f(r)dt^2+\frac{1}{f(r)}dr^2+r^2d\Sigma_{m-2}^2\, ,$$ where $d\Sigma_{m-2}^2$ is the line element of $(m-2)$-dimensional surface with constant scalar curvature $(m-2)(m-3)\delta$. Without loss of generality, $\delta$ can be set to $\pm1$ or zero. It is easy to find that the nonvanishing components of the Riemann tensor are $$R^{tr}{}{}_{tr}=-\frac{f^{''}}{2},\quad
R^{ti}{}{}_{tj}=R^{ri}{}{}_{rj}=-\frac{f^{'}}{2r}\delta_{j}^i,\quad
R^{ij}{}{}_{kl}=\frac{\delta-f}{r^2}\delta_{kl}^{ij}\, ,$$ where the prime stands for derivative with respect to $r$. The Euler density then has the form $$\begin{aligned}
\bar{L}_i&=&
\frac{(m-2)!}{(m-2-2i)!}\left(\frac{\delta-f}{r^2}\right)^{i}+4i
\frac{(m-2)!}{(m-1-2i)!} \left(-\frac{f^{'}}{2r}\right)
\left(\frac{\delta-f}{r^2}\right)^{i-1}
\nonumber \\
&+&2i \frac{(m-2)!}{(m-2i)!}
\left(-\frac{f^{''}}{2}\right)\left(\frac{\delta-f}{r^2}\right)^{i-1}
+4i(i-1)\frac{(m-2)!}{(m-2i)!}\left(-\frac{f^{'}}{2r}\right)^2
\left(\frac{\delta-f}{r^2}\right)^{i-2}\, .\nonumber\\\end{aligned}$$ Defining $$F(r)=\frac{\delta-f(r)}{r^2}\, ,$$ $\bar{L}_i$ can be rewritten as $$\bar{L}_i = \frac{(m-2)!}{(m-2i)!}\frac{1}{r^{m-2}}\left(r^m
F(r)^i\right)^{''}\, .$$ Finally Eq. (\[2eq14\]) becomes $$\label{2eq21}
0=\sum_{i=0}^{s}\widehat{\alpha}_{i}\left(r^m
F(r)^i\right)^{''}\, ,$$ where $$\widehat{\alpha}_{i}=\frac{(m-2)!}{(m-2i)!}\alpha_i\, .$$ The solution to (\[2eq21\]) is determined by the algebraic equation $$\label{Fequ}
\sum_{i=0}^{s}\widehat{\alpha}_{i}F(r)^i=\frac{M}{r^{m-1}}+\frac{Q}{r^{m}}\,
,$$ where $M, Q$ are two integration constants. Naively one may think they are related to the mass and Weyl charge of the solution [@Maeda:2006iw; @Dadhich:2000am; @Shiromizu:1999wj], respectively. But actually the integration constant $M$ has nothing to do with the mass of the solution, which will be shown shortly. The constant $Q$ may be positive, zero and negative. Here some remarks are in order.
\(i) Since $\mathcal{G}^{a}_{b}\sim \delta^{a}_{b}$, the equations $\mathcal{G}^{a}_{b}=0$ reduce to a single equation (\[2eq14\]). This is very different from the normal case in Lovelock gravity, where $\mathcal{G}^{a}_{b}$ is not proportional to $\delta^{a}_{b}$ even under a spherical symmetric assumption. For example, the equation $\mathcal{G}^{t}_{t}=0$ of Lovelock theory will give a first order differential equation like [@Whee; @Cai4] $$0=\sum_{i=0}c_{i}\left(r^{m-1} F(r)^i\right)^{'}\, ,$$ in the static, spherically symmetric case. In that case, there is only one integration constant, which is nothing but the mass of the solution [@Cai4]. In the present case, one has only one traceless-like equation, which is a second order differential equation. There is therefore one more integration constant $Q$ in the present case.
\(ii) If $Q=0$, Eq. (\[Fequ\]) is very similar to the corresponding one for static, spherically symmetric black hole solutions in Lovelock theory [@Whee; @Cai4]. However, there are two obvious differences: one is that here the coefficients $\widehat{\alpha}_i$ are all fixed by $\bar{k}/r_0^2$ and $D$, while in the normal case, those coefficients are free parameters [@Whee; @Cai4]. We will show this below. The other is that in Eq. (\[Fequ\]), the range of $i$ is $[0,\cdots,
s=[m/2]]$, while in the normal case, the range is $ [0,\cdots,
[(m-1)/2]]$. Therefore when $m$ is even, $i$ can take the value $[m/2]$, which will not appear in the normal case.
\(iii) Since all $\widehat{\alpha}_0$ are fixed by Eq. (\[lineq\]) and they are all positive constants, the solutions are not asymptotically flat, but asymptotically AdS spacetimes.
Black hole solution with $M=0$
-------------------------------
The spacetime (\[m-2metric\]) describes a black hole provided $f(r_+)=0$ and $f(r)>0$ with $r>r_+$. Here $r=r_+$ is called black hole horizon. We can see from (\[Fequ\]) that even if the “mass" $M$ of the solution vanishes, black hole horizon can still exist. To show this, let us discuss the solution $F=F_0$ of Eq. (\[Fequ\]) with $M=0$: $$\sum_{i=0}^{s}\widehat{\alpha}_{i}F^i_0=\frac{Q}{r^{m}}$$ with horizons. Assume $f_0(\bar{r})=0$ at some positive $\bar{r}$, and we have $$\bar{r}^2F_0(\bar{r})= \delta \quad \mathrm{or}\quad
F_0(\bar{r})=\frac{\delta}{\bar{r}^2}\, .$$ That is, $\bar{r}$ must satisfy the following equation $$\label{2eq27}
\bar{r}^{m}\sum_{i=0}^{s}\widehat{\alpha}_{i}\left(\frac{\delta}{\bar{r}^2}\right)^i=Q\,
.$$ In order for the equation to hold, the constant $Q$ must satisfy some constraints. We will discuss these constraints in the cases of $\delta=0$ and $\delta=\pm1$, respectively.
(i). $\delta=0$. This case is simple. In this case, only one term in (\[2eq27\]) remains. We have $$\bar{r}^{m}\widehat{\alpha}_{0}=Q\, ,$$ Since $\widehat{\alpha}_{0}>0$, this indicates a positive $\bar{r}$ exists provided $Q>0$, and $\bar{r}$ is just the horizon radius $r_+$.
(ii). $\delta=\pm 1$. If $m$ is odd, one then has $s=(m-1)/2$, and $$\label{2eq29}
\bar{r}^{m}\widehat{\alpha}_{0}\pm\bar{r}^{m-2}\widehat{\alpha}_{1}+\cdots
+(\pm1)^s\bar{r}\widehat{\alpha}_{s}=Q\, .$$ Obviously, because all coefficients $\{\widehat{\alpha}_{0},
\widehat{\alpha}_{1}\cdots \widehat{\alpha}_{s}\}$ are positive, and especially $\widehat{\alpha}_{0}>0$, Eq. (\[2eq29\]) has at least one positive root $\bar{r}$ if $Q>0$. The black hole horizon $r_+$ is just the largest positive root of Eq. (\[2eq29\]).
On the other hand, if $m$ is even, one then has $s=m/2$, and $$\bar{r}^{m}\widehat{\alpha}_{0}\pm\bar{r}^{m-2}\widehat{\alpha}_{1}+\cdots
+(\pm1)^{s-1}\bar{r}^2\widehat{\alpha}_{s-1}+(\pm1)^s\widehat{\alpha}_{s}=Q\,
.$$ From the theory of polynomial, the above equation has at least one negative root and one positive root if $\widehat{\alpha}_{0}\left[(\pm1)^s\widehat{\alpha}_{s}-Q\right]<0$. Recall $\widehat{\alpha}_{0}>0$, and this condition can be always satisfied if $Q>\widehat{\alpha}_{s}$. Namely, black hole horizon exists in this case.
In summary, black hole horizon always exists provided $Q-\widehat{\alpha}_{s}>0$, even when the parameter $M$ vanishes.
Examples of exact solutions
---------------------------
To be more explicit, in this subsection, we give some simple examples of exact solutions given in (\[Fequ\]).
### The case of $m=3$, $p=2$, $D\geq 5$
In this case, Eqs. (\[lineq\]) give $$\label{c0c2r0D} c_0=\frac{1}{2}(D^2-3D-6)r_0^{-2}\, ,\quad
c_2=\frac{1}{2(D-3)(D-4)}r_0^2\, .$$ We then have $$\widehat{\alpha}_0=\frac{2(2D-9)}{3(D-3)}r_0^{-2}\, ,\quad
\widehat{\alpha}_1=\frac{2}{D-3}\, .$$ Equation (\[Fequ\]) for $F$ becomes $$\widehat{\alpha}_0+\widehat{\alpha}_1
F=\frac{M}{r^2}+\frac{Q}{r^3}\, ,$$ which has the solution $$f=\frac{1}{\widehat{\alpha}_1}\left(-M-\frac{Q}{r}+\widehat{\alpha}_0
r^2\right)\, .$$ Here we have used the fact that the $(m-2)$-dimensional constant curvature space is always Ricci flat for $m=3$, i.e., $\delta=0$. Since $\widehat{\alpha}_0$ is always positive, this solution is just a BTZ black hole deformed by the additional charge $Q$. This kind of solution has been obtained in [@Maeda:2006hj].
### The case of $m=4$, $p=2$, $D\geq 6$
In this case, Eq. (\[lineq\]) give $$\label{c0c2r0Dm4} c_0=\frac{1}{2}(D^2-5D-2)r_{0}^{-2},\quad
c_2=\frac{1}{2(D-4)(D-5)}r_{0}^{2}\, .$$ We then have $$\widehat{\alpha}_0=\frac{2D-11}{3(D-4)}r_{0}^{-2}\, ,\quad
\widehat{\alpha}_1=\frac{2}{D-4}\, ,\quad
\widehat{\alpha}_2=\frac{1}{(D-4)(D-5)}r_{0}^{2}\, .$$ The equation for $F$ becomes $$\widehat{\alpha}_0+\widehat{\alpha}_1
F+\widehat{\alpha}_2F^2=\frac{M}{r^3}+\frac{Q}{r^4}\, ,$$ which has the solution $$F(r)=-\frac{\widehat{\alpha}_1}{2\widehat{\alpha}_2}
\left(~1\mp\sqrt{1-\frac{4\widehat{\alpha}_0
\widehat{\alpha}_2}{\widehat{\alpha}_1^2}+\frac{4
\widehat{\alpha}_2M}{\widehat{\alpha}_1^2~r^3}+\frac{4
\widehat{\alpha}_2Q}{\widehat{\alpha}_1^2~r^4}}~\right)\, ,$$ $$\label{m4solution} f(r)=\delta+\frac{\widehat{\alpha}_1
r^2}{2\widehat{\alpha}_2}\left(~1\mp\sqrt{1-\frac{4\widehat{\alpha}_0
\widehat{\alpha}_2}{\widehat{\alpha}_1^2}+\frac{4
\widehat{\alpha}_2M}{\widehat{\alpha}_1^2~r^3}+\frac{4
\widehat{\alpha}_2Q}{\widehat{\alpha}_1^2~r^4}}~\right)\, .$$ This solution with two branches is just the one recently obtained by Maeda and Dadhich in [@Maeda:2006iw]. It is easy to see that for $D\ge 6$, we have $$\widehat{\alpha}_1^2-4\widehat{\alpha}_0\widehat{\alpha}_2=\frac{4}{3(D-4)(D-5)}>0\,
,$$ So the vacuum AdS solution ($M=Q=0$) always exists.
### The case of $m=5$, $p=3$, $D\geq 7$
Equations (\[lineq\]) in this case lead to $$\begin{aligned}
c_0&=&\frac{D^4-10D^3+11D^2+22D+360}{3(D^2-7D+4)}r_{0}^{-2}\, ,\nonumber \\
c_2&=&\frac{1}{(D^2-7D+4)}r_{0}^{2}\, ,\nonumber\\
c_3&=&\frac{1}{3(D^2-7D+4)(D-5)(D-6)}r_{0}^{4}\, .\end{aligned}$$ Then the corresponding $\widehat{\alpha}$ are $$\begin{aligned}
\widehat{\alpha}_0&=&\frac{4(5D^2-67D+225)}{5(D^2-7D+4)(D-5)}r_{0}^{-2}\, ,\nonumber \\
\widehat{\alpha}_1&=&\frac{8(2D-13)}{(D^2-7D+4)(D-5)}\, ,\nonumber \\
\widehat{\alpha}_2&=&\frac{12}{(D^2-7D+4)(D-5)}r_{0}^{2}\, .\end{aligned}$$ The equation for $F$ is still in second order and has the solution $$F(r)=-\frac{\widehat{\alpha}_1}{2\widehat{\alpha}_2}
\left(~1\mp\sqrt{1-\frac{4\widehat{\alpha}_0
\widehat{\alpha}_2}{\widehat{\alpha}_1^2}+\frac{4
\widehat{\alpha}_2M}{\widehat{\alpha}_1^2~r^4}+\frac{4
\widehat{\alpha}_2Q}{\widehat{\alpha}_1^2~r^5}}~\right)\, ,$$ $$f(r)=\delta+\frac{\widehat{\alpha}_1
r^2}{2\widehat{\alpha}_2}\left(~1\mp\sqrt{1-\frac{4\widehat{\alpha}_0
\widehat{\alpha}_2}{\widehat{\alpha}_1^2}+\frac{4
\widehat{\alpha}_2M}{\widehat{\alpha}_1^2~r^4}+\frac{4
\widehat{\alpha}_2Q}{\widehat{\alpha}_1^2~r^5}}~\right)\, .$$ This solution is an example with the third order Lovelock term. Note that when $m=5$, the solution also has two branches. In addition, it is also easy to see that the vacuum AdS solution exists, because $$\widehat{\alpha}_1-4\widehat{\alpha}_0\widehat{\alpha}_1
=\frac{64 (5D-34)}{5(D^2-7D+4)(D-5)}>0$$ for $D\ge 7$.
Thermodynamic properties of black hole solutions
================================================
Naive consideration: $m$-dimensional black holes
------------------------------------------------
The black hole spacetime has a direct product form $\mathcal{M}^m
\times \mathcal{H}^{n}$, where $\mathcal{H}^{n}$ is a negative constant curvature space with a constant radius $r_0$. From the point of view of usual Kaluza-Klein dimensional reduction, the thermodynamics for the whole spacetime is equivalent to that for $m$-dimensional black hole with redefined gravitational constant $G_m=G_{m+n}/{\rm Vol}(n)$. Here $G_m$ and $G_{m+n}$ are gravitational constants in $m$ dimensions and $(m+n)$ dimensions, respectively, while ${\rm Vol}(n)$ is the volume of the constant curvature space $\mathcal{H}^{n}$. In this subsection, we will discuss the black hole thermodynamics from the point of view of $m$ dimensions.
Assume that black hole has a horizon at $r_+$, which is the largest positive root of $f(r)=0$. The horizon radius then must satisfy $$F(r_{+})=\frac{\delta}{r_{+}^2}\, , \quad \mathrm{or} \quad
r_{+}^2F(r_{+})= \delta\, .$$ The Hawking temperature of the black hole can be easily calculated by Euclidean method. To avoid conical singularity at the horizon, the period of Euclidean time should be $ \beta = 4\pi/f'(r_+) $, and the Hawking temperature is just the inverse of the period. This way we get the temperature of the black hole $$\label{temperature}
\mathfrak{T}=\frac{1}{\beta}=\frac{1}{4\pi}f^{'}(r_{+})=-\frac{1}{4\pi}
\left(\frac{2\delta}{r_{+}}+r_{+}^2F^{'}(r_{+})\right)\,
.$$ To get the explicit form of the Hawking temperature in terms of black hole horizon, we have to give the expression of $F^{'}$. Taking derivative on both sides of Eq. (\[Fequ\]) with respect to $r$, one has $$F^{'}(r_{+})=-\frac{(m-1)M r_{+} + m Q}{r_{+}^{m+1}\sum_{i=1}^{s}
i~\widehat{\alpha}_{i}\left(\frac{\delta}{r_{+}^2}\right)^{i-1}}\, .$$ On the other hand, from the equation (\[Fequ\]), we have $$\label{massparameter} M=-\frac{Q}{r_{+}}+
r_{+}^{m-1}\sum_{i=0}^{s}\widehat{\alpha}_{i}\left(\frac{\delta}{r_{+}^2}\right)^i\,
.$$ Thus we can express $F^{'}(r_+)$ as $$F^{'}(r_{+})=-\frac{Q+(m-1)r_{+}^m\sum_{i=0}^{s}
\widehat{\alpha}_{i}\left(\frac{\delta}{r_{+}^2}\right)^{i}}{r_{+}^{m+1}\sum_{i=1}^{s}
i~\widehat{\alpha}_{i}\left(\frac{\delta}{r_{+}^2}\right)^{i-1}}\, ,$$ and the Hawking temperature has the form $$\mathfrak{T}=\frac{1}{4\pi r_{+}\sum_{i=1}^{s}
i~\widehat{\alpha}_{i}\left(\frac{\delta}{r_{+}^2}\right)^{i-1}}
\left(\sum_{i=0}^{s}(m-2i-1)
\widehat{\alpha}_{i}\delta\left(\frac{\delta}{r_{+}^2}\right)^{i-1}
+\frac{Q}{r_{+}^{m-2}}\right)\,
.$$ Clearly, if we choose a suitable $Q$, the Hawking temperature may vanish. This case corresponds to the “extremal" black holes with vanishing Hawking temperature. For example, when $m=4$, choosing $$Q=-3\widehat{\alpha}_0 r_{+}^4-\widehat{\alpha}_1
r_{+}^2\delta+\widehat{\alpha}_2\delta^2$$ will lead to a vanishing temperature.
To get the mass of the black hole, we expand the metric $g_{00}$ in the large $r$ limit, subtract the corresponding one for a suitable reference background solution $F_b$, and then read off the mass with the coefficient in front of some power of the radial coordinate $r$. Here we choose the vacuum AdS solutions with vanishing $M$ and $Q$ as the reference background, i.e., $$\sum_{i=0}^{s}\widehat{\alpha}_i F_b^i = 0 \, .$$ For large $r$, we can expand $F$ as $F=F_b+\Delta F$ with the leading order correction $\Delta F$. We arrive at $$\Delta F\sum_{i=0}^{s} i ~\widehat{\alpha}_iF_b^{i-1}= \Delta
F\widehat{\alpha}=\frac{M}{r^{m-1}} \, .$$ Here the constant $\widehat{\alpha}$ is given by $$\widehat{\alpha}=\sum_{i=0}^{s} i ~\widehat{\alpha}_iF_b^{i-1}\, .$$ For solutions in some branch, $\widehat{\alpha}$ may be negative. However, we only consider the cases with positive $\widehat{\alpha}$ here. So we find the expansion of metric around the background as $$g_{tt}-g_{(b)tt}= -f+f_{b}=r^2\Delta F \approx
\frac{M}{\widehat{\alpha}~ r^{m-3}}=\frac{16\pi
G_{m}\mathfrak{M}}{(m-2)V_{m-2}r^{m-3}}\, .$$ Thus we find that the mass of the black hole $\mathfrak{M}$ has a relation to $M$ as $$\label{massm} \frac{M}{\widehat{\alpha}}=\frac{16\pi
G_{m}\mathfrak{M}}{(m-2)V_{m-2}}\, ,$$ where $V_{m-2}$ is the volume of $(m-2)$-dimensional cross section of horizon surface. The mass can be expressed in terms of horizon radius $r_+$ and $Q$ as $$\mathfrak{M}=\frac{(m-2)V_{m-2}}{16\pi
G_{m}\widehat{\alpha}}\left(-\frac{Q}{r_{+}}+
r_{+}^{m-1}\sum_{i=0}^{s}\widehat{\alpha}_{i}\left(\frac{\delta}{r_{+}^2}\right)^i\right)\,
,$$ and its variation with respect to the horizon radius is $$\left(\frac{\partial \mathfrak{M}}{\partial
r_{+}}\right)_{Q}=\frac{(m-2)V_{m-2}}{16\pi
G_{m}\widehat{\alpha}}r_{+}^{m-4}\left(\sum_{i=0}^{s}(m-2i-1)
\widehat{\alpha}_{i}\delta\left(\frac{\delta}{r_{+}^2}\right)^{i-1}+\frac{Q}{r_{+}^{m-2}}\right)\,
.$$ Since we are dealing with black holes in higher derivative gravity theory, the well-known area formula for black hole entropy breaks down. Let us try to obtain the black hole entropy by integrating the first law of the black hole thermodynamics $$\label{integration}
\mathfrak{S}=\int \mathfrak{T}^{-1} d
\mathfrak{M} =\int^{r_+} \mathfrak{T}^{-1}\left(\frac{\partial
\mathfrak{M}}{\partial r_{+}}\right)_{Q}dr_{+}\, .$$
(i). When $m$ is even, $s$ takes the value $m/2$. The integral gives $$\begin{aligned}
\label{3eq16}
\mathfrak{S}&=&\int \mathfrak{T}^{-1} d \mathfrak{M}
=\int^{r_+} \mathfrak{T}^{-1}\left(\frac{\partial
\mathfrak{M}}{\partial
r_{+}}\right)_{Q}dr_{+}\nonumber\\
&=&\frac{V_{m-2}}{4
G_{m}\widehat{\alpha}}\left[\sum_{i=1}^{s-1}\frac{m-2}{m-2i}
i~\widehat{\alpha}_{i}\delta ^{i-1}r_+^{m-2i} +s(s-1)
\widehat{\alpha}_s~\delta^{s-1}~\mathrm{ln}
\left(r_+^2\right)\right] +\mathfrak{S}_0\, .\end{aligned}$$ The last term $\mathfrak{S}_0$ is an integration constant. Note that here a logarithmic term appears, which comes from the fact that $s$ can take the value $m/2$.
(ii). When $m$ is odd, $s$ is $(m-1)/2$. In this case, the integral gives $$\begin{aligned}
\label{3eq17}
\mathfrak{S}&=&\int \mathfrak{T}^{-1} d \mathfrak{M}
=\int^{r_+} \mathfrak{T}^{-1}\left(\frac{\partial
\mathfrak{M}}{\partial
r_{+}}\right)_{Q}dr_{+}\nonumber\\
&=&\frac{V_{m-2}}{4
G_{m}\widehat{\alpha}}\left[\sum_{i=1}^{s}\frac{m-2}{m-2i}
i~\widehat{\alpha}_{i}\delta ^{i-1}r_+^{m-2i}\right]+ \mathfrak{S}_0
\, .\end{aligned}$$ Here $\mathfrak{S}_0$ is also an integration constant. Note that here the integration constant $\mathfrak{S}_0$ should be set to zero because when the black hole horizon shrinks to zero, the entropy of the black hole should vanish [@Cai1]. However, the integration constant cannot be fixed by the same argument in the case of even $m$, due to the existence of the logarithmic term in the black hole entropy. In addition, let us notice that when the black hole horizon is a Ricci flat surface, namely, $\delta=0$, not only does the logarithmic term disappear in (\[3eq16\]), but also both (\[3eq16\]) and (\[3eq17\]) give an entropy proportional to horizon area. This is also a general feature of black hole entropy in Lovelock gravity [@Cai4; @Cai1].
The entropy expressions (\[3eq16\]) and (\[3eq17\]) look quite similar to the entropy formula of static, spherically symmetric black holes in Lovelock gravity [@Cai4], except for the logarithmic term in (\[3eq16\]). The appearance of the logarithmic term is strange, although such a term appears in the entropy expressions of black holes in Horava-Lifshitz gravity theory [@CCO], while the latter is not a full diffeomorphism invariant theory. For a diffeomorphism invariant gravity theory, Wald showed that black hole entropy is a Noether charge [@Wald:1993nt]; further a well-known entropy formula was developed [@Wald:1993nt; @Iyer:1994ys]. By Wald’s entropy formula, black hole entropy is given by some integral on the black hole horizon. Therefore black hole entropy must be a function of horizon geometry and a logarithmic term will never appear in Wald’s entropy formula. This may cause suspicion whether the results given above are valid or not.
Let us notice that the above way to obtain black hole solution in $m$ dimensions is quite different from the usual Kaluza-Klein dimensional reduction. In the usual Kaluza-Klein theory, with the assumption of direct product of the manifold $\mathcal{M}^m\times \mathcal{H}^n$, one gets reduced action by integrating the total action over the extra space $\mathcal{H}^n$. Certainly here too, with the assumption of direct product structure of the spacetime, one can get reduced action from the total action (\[1eq1\]). This reduced action is an $m$-dimensional version of Lovelock gravity as given below in Eq. (\[reducedaction\]). We can get equations of motion (\[emm\]) only for $m$-dimensional part but not the $n$-dimensional part by variation of this reduced action. Obviously these are nothing but the usual equations of motion of the Lovelock gravity in $m$ dimensions with special coefficients $A_{t}$’s. For example, when $m=4$, these equations are the Einstein equations with cosmological constant $$\label{eins}
A_0 g_{ij} + A_1 E_{ij}=0\, ,$$ where $E_{ij}$ is the Einstein tensor in four dimensions, and $A_0$ and $A_1$ are given in (\[lineq\]). Because our solutions are obtained for $A_0=A_1=0$, Eqs. (\[eins\]) have no information on our solutions. Therefore the field equations for $m$-dimensional part is trivially satisfied, while the nontrivial solutions come from the trace equation of gravitational field for the $n$-dimensional part, which is not obtained from the reduced action. So the reduced action certainly exists, but it does not give solutions in this paper. If we naively omit the extra dimensions in our solutions, the corresponding $m$-dimensional local diffeomorphism-invariant “effective action" is absent, and we have to consider the whole $(m+n)$-dimensional theory. This is very different from the usual Kaluza-Klein theory in which the effective action is just the reduced action. As a result, we cannot simply use Wald’s entropy formula to get the entropy of the $m$-dimensional black hole. On the other hand, Lovelock theory is diffeomorphism invariant and Wald’s entropy formula is applicable in the whole $(m+n)$ dimensions. In the following subsections, we discuss thermodynamics of the black holes by Euclidean action and Wald’s entropy formula in $(m+n)$ dimensions, and find quite different and surprising results.
Entropy of $(m+n)$-dimensional black holes
------------------------------------------
From the viewpoint of the whole $D(=m+n)$ dimensions, to study thermodynamics of these black holes is straightforward. The temperature of the black holes is the same as the one in Eq. (\[temperature\]) because it is determined by the horizon geometry only. In Lovelock gravity, Wald’s entropy formula can be expressed as [@Jacobson:1993xs] $$\label{entropywald} \mathfrak{S}=\sum_{k=1}^{p}4\pi k c_k \int
d^{D-2}x\mathcal{L}_{k-1}(\tilde{h})\, ,$$ where $\mathcal{L}_{k-1}(\tilde{h})$ has the same form as (\[Lk\]) except that metric is replaced by $\tilde{h}$, which is the induced metric on the $(D-2)$-dimensional cross section of the horizon. The induced metric $\tilde h$ is $$\tilde{h}= \tilde{q}_{ij}dz^idz^j + r_{0}^2 \gamma_{ab}dy^a dy^b\, ,$$ where $\tilde{q}_{ij}$ is the induced metric of the cross section of the horizon in the $m$-dimensional part.
By the similar procedure to get (\[emm\]) and (\[enn\]), we have $$\begin{aligned}
\mathfrak{S}&=&\sum_{t=0}^{w}\sum_{k=t+1}^{p}\Bigg{\{}{k-1 \choose
t}4\pi k c_k
\frac{(D-m)!}{(D-m-2(k-1-t))!}\left(\frac{\bar{k}}{r_0^2}\right)^{k-t-1}\nonumber\\
&~~~~~~~\times& \int d^{D-m}yr_0^{D-m} \sqrt{\gamma}\int
d^{m-2}z\mathcal{L}_{t}(\tilde{q})\Bigg{\}}\, ,\end{aligned}$$ where $w=[(m-2)/2]$. Define $\ell=t+1$, we have $$\begin{aligned}
\label{ent}
\mathfrak{S}&=&\sum_{\ell=1}^{w+1}\Bigg{\{}4\pi
\ell\Bigg{[}\sum_{k=\ell}^{p}{k \choose \ell} c_k
\frac{(D-m)!}{(D-m-2(k-\ell))!}\left(\frac{\bar{k}}{r_0^2}\right)^{k-\ell}
\Bigg{]}\nonumber\\
&~~~~~~~\times& \int d^{D-m}yr_0^{D-m} \sqrt{\gamma}\int
d^{m-2}z\mathcal{L}_{\ell-1}(\tilde{q})\Bigg{\}}\, .\end{aligned}$$ Thus the entropy can be expressed as $$\begin{aligned}
\label{3eq22}
\mathfrak{S}=\sum_{\ell=1}^{w+1}\Bigg{\{}4\pi \ell
A_{\ell} \int d^{D-m}yr_0^{D-m} \sqrt{\gamma}\int
d^{m-2}z\mathcal{L}_{\ell-1}(\tilde{q})\Bigg{\}}\, ,\end{aligned}$$ where $A_{\ell}$ is defined in Eq. (\[lineq\]), from which we have $$\label{Alequation} A_{\ell}=0\, ,\quad \ell=1,\cdots ,[(m-1)/2]\, .$$ Thus we find from the entropy (\[3eq22\]) that when $m$ is odd, $\mathfrak{S}=0$, a vanishing entropy!
When $m$ is even, the entropy is $$\label{entropy1} \mathfrak{S}=2\pi m
\left(r_0^{D-m}\Omega_{D-m}\right)c_{m/2}\chi(\Sigma_{m-2})\, .$$ where $\Omega_{D-m}$ is the volume of $\mathcal{H}^n$, and $\chi(\Sigma_{m-2})$ is the integration of Euler characteristic on the $(m-2)$-dimensional cross section of horizon surface, i.e., $$\chi(\Sigma_{m-2})=\int_{\Sigma_{m-2}}
d^{m-2}z\mathcal{L}_{(m/2-1)}(\tilde{q})\, .$$ For the cross section of horizon surface $\Sigma_{m-2}$, which is a constant curvature space, $\chi(\Sigma_{m-2})$ is constant, while in the case of $\Sigma_{m-2}$ being a closed manifold, $\Sigma_{m-2}$ need not be a constant curvature space, and in that case, $\chi(\Sigma_{m-2})$ is the Euler number of $\Sigma_{m-2}$ up to a constant factor. For example, when $m=4$, $D=6$, we have $$\mathfrak{S}= 64\pi^2\cdot \delta\cdot c_2 \cdot(r_0^2 \Omega_2)\, .$$ Clearly we see that when $\Sigma_2$ is a Ricci flat space, i.e. $\delta =0$, the constant entropy vanishes. Here the constant means that it is independent of the horizon radius $r_+$ and charge $Q$.
Now we argue that the constant entropy is meaningless for black hole thermodynamics and should be dropped. One simple reason is that when the cross section of horizon surface $\sigma_{m-2}$ is a negative constant curvature space, $\chi$ is negative, giving a negative entropy which does not make sense in thermodynamics. Another reason is provided by the following example. Consider a four-dimensional Schwarzschild black hole solution. In this case, the Euler density is the Gauss-Bonnet term. If one considers the contribution of the Gauss-Bonnet term to the black hole entropy, besides the usual area entropy, one has an additional constant $\mathfrak{S}= 64\pi^2 c_2$ from (\[entropy1\]), where $c_2$ is the Gauss-Bonnet coefficient. Such a constant term remains even when the black hole horizon goes to zero. Both of these clearly indicate that the constant entropy from the horizon topology should be dropped when black hole thermodynamics is concerned.
Let us illustrate these discussions by two examples of $m=3$ and 4.
### The case of $m=3$, $p=2$, $D\ge 5$
In this case, the metric $\tilde{h}$ is very simple, which is just the metric of the constant curvature space $\mathcal{H}^n$ plus the metric of a circle, i.e., $$\tilde{h}= dz^2 + r_{0}^2 \gamma_{ab}dy^a dy^b\, ,$$ and Wald’s entropy (\[entropywald\]) becomes $$\label{m34entropy} \mathfrak{S}=4\pi \int d^{D-2}x+8\pi c_2\int
d^{D-2}xR(\tilde{h})\, .$$ Here we have set $c_1$ to unity as before, and $R(\tilde{h})$ is the scalar curvature of the metric $\tilde{h}$. It is easy to find $$R(\tilde{h})=(D-3)(D-4)\left(\frac{\bar{k}}{r_0^2}\right)\, ,$$ thus we have $$\mathfrak{S}=4\pi
(r_0^{D-3}\Omega_{D-3})V_1\left[1+2c_2(D-3)(D-4)\left(\frac{\bar{k}}{r_0^2}\right)\right]
\, ,$$ where $V_1$ is the volume of the circle. Considering $\bar{k}=-1$, and the explicit relation among $c_2$, $r_0$ and $D$ in Eq. (\[c0c2r0D\]), the entropy identically vanishes.
Actually, for $m=3$, we have $w=0$, so $\ell$ in Eq. (\[3eq22\]) can take the value $1$ only. Since $A_{1}=0$ by Eq. (\[Alequation\]), we reach a vanishing entropy from (\[3eq22\]), as we have just shown above.
### The case of $m=4$, $p=2$, $D\ge 6$
In this case, the metric $\tilde{h}$ consists of the metric of the constant curvature space $\mathcal{H}^n$ and the metric of a 2-dimensional constant curvature space $\Sigma_2$, i.e., $$\tilde{h}= \tilde{q}_{ij}dz^idz^j + r_{0}^2 \gamma_{ab}dy^a dy^b\, ,$$ and Wald’s entropy is the same as (\[m34entropy\]). Now, the scalar curvature $R(\tilde{h})$ becomes $$R(\tilde{h})= R(\tilde{q}) +
(D-4)(D-5)\left(\frac{\bar{k}}{r_0^2}\right)\, ,$$ where $R(\tilde{q})$ is the scalar curvature of the metric $\tilde{q}$. Thus the entropy (\[ent\]) has the form $$\mathfrak{S}=4\pi
(r_0^{D-4}\Omega_{D-4})V_2\left[1+2c_2(D-4)(D-5)\left(\frac{\bar{k}}{r_0^2}\right)\right]+
8\pi c_2 (r_0^{D-4}\Omega_{D-4}) \int
\sqrt{\tilde{q}}d^2zR(\tilde{q}) \, .$$ Using the explicit relation among $c_2$, $r_0$ and $D$ in equation (\[c0c2r0Dm4\]), and $\bar{k}=-1$, we see that only the last term remains $$\label{entropym4} \mathfrak{S}= 8\pi c_2 (r_0^{D-4}\Omega_{D-4})
\int \sqrt{\tilde{q}}d^2zR(\tilde{q}) \, .$$ In other words, in this case, the entropy is totally determined by the integration of the Euler characteristic $R(\tilde{q})$ on $\Sigma_2$, in agreement with (\[entropy1\]). Note that the the two-dimensional induced horizon $\tilde{q}$ is a constant curvature space with scalar curvature $2\delta$. The entropy can be further expressed as $$\mathfrak{S}= 64\pi^2 \cdot \delta \cdot c_2 \cdot
(r_0^{D-4}\Omega_{D-4})\, .$$
In fact, in the case of $m=4$, one has $w=1$, and $\ell$ in (\[3eq22\]) can take values $1$ and $2$. We have $A_1=0$ from (\[Alequation\]), while $A_2=c_2$. By the general entropy expression (\[3eq22\]), we arrive at the same result as Eq. (\[entropym4\]). As argued above, the constant entropy does not make sense in black hole thermodynamics, we should drop it and conclude that the physical entropy is zero.
Mass and Euclidean action of $m+n$ dimensional black holes
----------------------------------------------------------
In this subsection we show another surprising result that the mass of these black holes also vanishes. To do this, we employ the Euclidean approach to black hole thermodynamics. The Euclidean action $I_E$ of the black holes includes two parts, the bulk and boundary parts, $$I_E=I+B\, ,$$ where $I$ is the bulk action, while $B$ denotes the boundary term. The bulk part is given by $$\label{reducedaction}
I=-\left(r_0^{D-m}\Omega_{D-m}\right)
\left\{\sum_{t=0}^{s}A_t\int\sqrt{g}d^{m}x\bar{L}_{t}\right\}\, ,$$ where $\bar{L}_{t}$ is given by (\[Lbart\]). With the metric (\[m-2metric\]), we get $$\label{bulkaction} I=-\left(r_0^{D-m}\Omega_{D-m}V_{m-2}\right)
\left\{\sum_{t=0}^{s}A_t\frac{(m-2)!}{(m-2t)!}\int d\tau dr
\left(r^m F^t\right)''\right\}\, ,$$ where $V_{m-2}$ is the volume of the $\Sigma_{m-2}$ with unit radius. In general, the boundary term $B$ is a little bit complicated. For simplicity, we only consider here the case that the highest derivative term is the Gauss-Bonnet term, i.e., we deal with the cases with $m=3$ and $m=4$. In that case, the boundary term is given by [@Myers:1987yn; @Davis:2002gnGravanis:2002wy] $$B=-2\int_{\partial X}d^{D-1}v\sqrt{h}\left[K+2c_2
\left(J-2E_{\mu\nu}K^{\mu\nu}\right)\right]\, ,$$ where we have set $c_1=1$ and $h$ is the induced metric on a timelike boundary $\partial X$. The tensor $K^{\mu\nu}$ is the extrinsic curvature of the boundary, and $K$ is its trace. We denote by $J$ the trace of the tensor $$J_{\mu\nu}=\frac{1}{3}\left(2K
K_{\mu\lambda}K^{\lambda}{}_{\nu}+K_{\lambda\sigma}K^{\lambda\sigma}
K_{\mu\nu}-2K_{\mu\lambda}K^{\lambda\sigma}K_{\sigma\nu}-K^2
K_{\mu\nu}\right)\, .$$ Tensor $E_{\mu\nu}$ is the Einstein tensor of the induced metric $h$. With this boundary term, the variation principle is well defined for the Gauss-Bonnet gravity.
In the following calculations, we consider a boundary $\partial X$ with a given $r \gg r_+$, and take the limit of $r \to \infty$ at the end of calculations.
### The case of $m=3$, $p=2$, $D\ge 5$
The case of $m=3$ is quite simple. Both the bulk and boundary terms identically vanish $I=B=0$. In fact, in this case all $A_{t}$ are zero and therefore the bulk term (\[bulkaction\]) vanishes. For the boundary term $B$, after some calculation, it is not hard to find the tensor $J_{\mu\nu}=0$, and $$K=\frac{2f+rf'}{2r\sqrt{f}}\, ,\nonumber$$ $$2c_2(J-2E_{\mu\nu}K^{\mu\nu})=-c_2(D-3)(D-4)\left(\frac{2f+rf'}{rr_0^2\sqrt{f}}\right)\,
.$$ So the boundary term is given by $$B=- \beta
r_0^{D-3}\Omega_{D-3}V_1(2f+rf')\left[1-\frac{2c_2(D-3)(D-4)}{r_0^2}\right]\,
,$$ where $\beta$ is the period of Euclidean time. Again, considering the relation among $c_2$, $r_0$ and $D$ in equation (\[c0c2r0D\]), the boundary term has no contribution to the Euclidean action. Thus the Euclidean action of the black hole solutions is always zero, which leads to the conclusion that the energy and entropy of the black holes always vanish.
In fact, when $m$ is odd, since all $A_t=0$, both the bulk and boundary terms always vanish. Thus the result with vanishing energy and entropy is universal for odd $m$.
### The case of $m=4$, $p=2$, $D\ge 6$
When $m=4$, the bulk action (\[bulkaction\]) reduces to $$I= 4 c_2 \left(r_0^{D-4}\Omega_{D-4}V_2\right)\beta
\left[f'(\delta-f)\right]_{r=\infty} - 4 c_2
\left(r_0^{D-4}\Omega_{D-4}V_2\right)\beta \left[f'(\delta
-f)\right]_{r=r_+}\, ,$$ where $\beta$ is the period of Euclidean time as before. It is easy to find the trace of the extrinsic curvature is given by $$K=\frac{4f+rf'}{2r\sqrt{f}}\, ,$$ and after some calculations, we can obtain $$\begin{aligned}
2c_2(J-2E_{\mu\nu}K^{\mu\nu})&=&
\frac{c_2}{r^3r_0^2\sqrt{f}}\bigg{\{}rf'\left[2\delta r_0^2-(D-4)(D-5)r^2\right]\nonumber\\
&&-2f \left[2(D-4)(D-5)r^2-4r_0^2+rr_0^2f' \right]\bigg{\}}\, .\end{aligned}$$ Therefore, the boundary term $B$ is $$\begin{aligned}
B&=&-2 \left(r_0^{D-4}\Omega_{D-4}V_2\right)\beta \Bigg{[}\left(1
-\frac{2c_2(D-4)(D-5)}{r_0^2}\right)\left(2rf+\frac{1}{2}r^2f'\right)\nonumber\\
&&+2c_2 f'(\delta-f)+\frac{8c_2}{r}f\Bigg{]}_{r=\infty}\, .\end{aligned}$$ Note that the equation (\[c0c2r0Dm4\]) or $A_1=0$ gives $1-2c_2(D-4)(D-5)/r_0^2=0$. Thus the boundary term reduces to $$B=-4c_2 \left(r_0^{D-4}\Omega_{D-4}V_2\right)\beta \left[
f'(\delta-f)+\frac{4}{r}f\right]_{r=\infty}\, .$$ We thus get the total action $$I_E=I+B=-4c_2 \left(r_0^{D-4}\Omega_{D-4}V_2\right)\beta \left[
f'(\delta -f)\right]_{r=r_+}-\left.16\beta c_2
\left(r_0^{D-4}\Omega_{D-4}V_2\right)
\frac{f}{r}\right|_{r=\infty}\, .$$ For our solutions, the second term in the right hand side is divergent when $r \to \infty $. This divergence can be removed by the background subtraction method. By subtracting the contribution from the reference background with vanishing $M$ and $Q$, we find that the second term does not make any contribution to the Euclidean action and only the first term remains.
Note that $\beta=4\pi/f'(r_+)$ and $f=0$ at the horizon. The first term can be expressed as $$\label{3eq36}
I_E= -64\pi^2\cdot \delta\cdot c_2 \cdot\left(r_0^{D-4}\Omega_{D-4}\right)\, .$$ This is a constant independent of temperature. Considering the relation between the Euclidean action $I_E$ and free energy $F$: $I_E=\beta F$, we immediately see that the energy of the black holes always vanishes, while the Euclidean action (\[3eq36\]) gives the constant entropy (\[entropy1\]) found by Wald’s formula.
Thus by calculating Euclidean action and Wald’s entropy, we have shown that mass and entropy of these black hole solutions presented in the previous section vanish identically.
Conclusion and Discussion
=========================
In this paper we have presented a class of black hole solutions in ($m+n$)-dimensional Lovelock gravity. The black hole solutions have a direct product structure $\mathcal{M}^m \times \mathcal{H}^{n}$, where $\mathcal{H}^n$ is a negative constant curvature space with a constant radius. When $m=3$ and $4$, these solutions reduce to those recently found by Maeda and Dadhich [@Maeda:2006iw; @Maeda:2006hj; @Dadhich:2007xf; @Molina:2008kh] in Gauss-Bonnet gravity. We have obtained these black hole solutions in a way as follows. We first decompose the equations of motion into two sets, one for $m$-dimensional part and the other for $n$-dimensional part. Then imposing constraints on the coefficients of higher curvature terms in Lovelock gravity so that the set of equations of motion for the $m$-dimensional part is trivially satisfied, we solve the trace equation for the $n$-dimensional part and obtain the black hole solutions. Since the trace equation is a second order differential equation, integrating the equation gives rise to two integration constants $M$ and $Q$.
We have tried to understand the physical meaning of the two integration constants by studying thermodynamics of these black hole solutions. The black holes we have found are exact solutions in $(m+n)$-dimensional Lovelock gravity theory. Naively considering the solution without the extra dimensions, it appeared that the mass were proportional to parameter $M$ as in Eq. (\[massm\]). By using the first law, we then found that the entropy of the black hole would have logarithmic term when $m$ is even. However, we have argued that this naive result is not valid for our solutions because the equations of motion for the $m$-dimensional part are trivially satisfied and our solutions come from the trace equation for the $n$-dimensional part. As a result the effective action for the $m$-dimensional part does not make any sense for the black hole solutions. We should consider the black hole solutions in the point of view of the whole $(m+n)$-dimensional spacetime. It is not surprising because our theory is intrinsically $(m+n)$-dimensional. We cannot naively neglect the extra dimensions as in the case of the usual Kaluza-Klein theory in which the thermodynamics in higher dimensions and lower dimensions are equivalent.
Then by employing Euclidean action approach to black hole thermodynamics and Wald’s entropy formula, we have found an astonishing result that both mass and entropy of these black holes always vanish identically although there exists a nonvanishing Hawking temperature for these black holes. Here it may be worth mentioning that when $m$ is even, by Wald’s entropy formula, the black hole has a constant entropy coming from the topological structure of the black hole horizon. But we have argued that the constant entropy should be neglected from the point of view of black hole thermodynamics.
Black hole solutions with a nonvanishing temperature and always vanishing mass and entropy look strange. But such a situation has happened in a class of Lifshitz black holes in $R^2$ gravity [@Cai:2009ac]. There these authors got the solution by adjusting the coupling constant of $R^2$ term to a critical value. Note that the same thing happens here in the class of Lovelock black hole solutions since we have chosen a special set of coupling coefficients of higher derivative terms in order to get our solutions.
Let us now try to understand such a phenomenon that a black hole has a nonvanishing temperature, but vanishing mass and entropy. Recall that in the $R^2$ gravity considered in [@Cai:2009ac], the resulting Lifshitz black hole solution satisfied $1+2\alpha R=0$, where $\alpha$ is the coefficient of the term $R^2$, which is a crucial point to give the zero entropy of the black hole. Let us notice that in the $R^2$ theory, the factor $1+2\alpha R$ is nothing but the effective coupling constant for some polarized graviton. From the equations of motion, it is easy to see that the effective gravitational constant turns to be $G_{eff}= G/(1+2\alpha R)$. As a result, $1+2\alpha R=0$ implies that the effective gravitational constant is divergent for the class of solutions with $1+2\alpha
R=0$. Wald’s entropy is equal to a quarter of the horizon area in units of the effective gravitational coupling [@BGH]. This is the reason why the entropy of the Lifshitz black hole has a vanishing entropy. Furthermore, because of $G_{eff} \to \infty$, from the point of view of background fluctuations, kinetic terms of those fluctuations always vanish, and only potential terms remain. This indicates that there is no dynamics for those fluctuations. In other words, there are no excitations of the background spacetime. This might be an interpretation why the black hole has no entropy. While the Hawking temperature (surface gravity) of a black hole is purely determined by black hole geometry in the sense that the Hawking temperature is just the inverse period of the Euclidean time of the black hole, the first law of thermodynamics enforces that a black hole has a vanishing mass (energy) if its entropy is zero.
Let us turn to the black hole spacetime discussed in the present paper. In fact the same happens here. The part of $\mathcal{H}^{n}$ is a trivial negative constant curvature space. The effective gravitational field equations for the part of $m$-dimensional black hole spacetime are trivially satisfied in the sense that those coefficients in front of some gravitational tensors like Einstein tensor are identically zero \[see (\[emm\]) and (\[lineq\])\]. To see this more clearly, one may refer to the simple case with Gauss-Bonnet gravity discussed in [@Maeda:2006iw]. The vanishing coefficients are correspondent to the factor $1+2\alpha R$ discussed above for $R^2$ gravity. Therefore, due to the special reduction used to find the black hole solutions in the present paper, these effective coupling constants from the $m$-dimensional point of view identically vanish. In this sense, the effective gravitational constant $G_{eff}$ diverges as in the case of the $R^2$ gravity. Then the same story goes on as the case of the $R^2$ gravity and these black holes have vanishing entropy and mass.
If our arguments are true, our above discussions and those in Ref. [@Cai:2009ac] both have important consequence on our understanding of the microscopic degrees of freedom of black hole entropy. According to ’t Hooft’s brick wall model [@tHooft], black hole entropy might come from statistical degrees of freedom of quantum fluctuations outside the black hole, namely if there is no such degrees of freedom of quantum fluctuations, there is no contribution to the entropy. The black hole entropy is not merely determined by the geometry of the horizon. In the examples discussed above, the Bekenstein-Hawking geometry entropy of the black holes always vanishes and the zero entropy is found to be closely related to the fact that the effective gravitational coupling constants are infinity such that any fluctuations are forbidden, there are totally no physical degrees of freedom associated with quantum fluctuations. Thus our results provide evidence that black hole entropy comes from statistical degrees of freedom of quantum fluctuations around the black hole. No doubt, it is worthwhile to further investigate this interesting issue.
Acknowledgements {#acknowledgements .unnumbered}
================
RGC is supported partially by grants from NSFC, China (No. 10535060, No. 10821504 and No. 10975168) and a grant from MSTC, China (No. 2010CB833004). LMC and NO were supported in part by the Grants-in-Aid for Scientific Research Fund of the JSPS Nos. 20540283 and 21$\cdot$09225, and also by the Japan-U.K. Research Cooperative Program. This work was finished during RGC’s visit to Kinki University with a JSPS invitation fund.
[99]{}
B. Zwiebach, Phys. Lett. B [**156**]{}, 315 (1985).
D. J. Gross and E. Witten, Nucl. Phys. B [**277**]{}, 1 (1986); R. R. Metsaev and A. A. Tseytlin, Nucl. Phys. B [**293**]{}, 385 (1987); I. Jack and D. R. T. Jones, Nucl. Phys. B [**303**]{}, 260 (1988); K. A. Meissner, Phys. Lett. B [**392**]{}, 298 (1997) \[arXiv:hep-th/9610131\].
D. Lovelock, J. Math. Phys. [**12**]{}, 498 (1971). D. G. Boulware and S. Deser, Phys. Rev. Lett. [**55**]{}, 2656 (1985). J. T. Wheeler, Nucl. Phys. B [**268**]{}, 737 (1986); Nucl. Phys. B [**273**]{}, 732 (1986). R. G. Cai, Phys. Lett. B [**582**]{}, 237 (2004) \[arXiv:hep-th/0311240\]. M. Banados, C. Teitelboim and J. Zanelli, Phys. Rev. D [**49**]{}, 975 (1994) \[arXiv:gr-qc/9307033\]. R. G. Cai and K. S. Soh, Phys. Rev. D [**59**]{}, 044013 (1999) \[arXiv:gr-qc/9808067\]. J. Crisostomo, R. Troncoso and J. Zanelli, Phys. Rev. D [**62**]{}, 084013 (2000) \[arXiv:hep-th/0003271\]; R. Aros, R. Troncoso and J. Zanelli, Phys. Rev. D [**63**]{}, 084015 (2001) \[arXiv:hep-th/0011097\]. R. G. Cai, Phys. Rev. D **65**, 084014 (2002) \[arXiv:hep-th/0109133\]; R. G. Cai and Q. Guo, Phys. Rev. D **69**, 104025 (2004) \[arXiv:hep-th/0311020\]. R. C. Myers and J. Z. Simon, Phys. Rev. D [**38**]{}, 2434 (1988). H. C. Kim and R. G. Cai, Phys. Rev. D [**77**]{}, 024045 (2008) \[arXiv:0711.0885 \[hep-th\]\]; Y. Brihaye and E. Radu, Phys. Lett. B [**661**]{}, 167 (2008) \[arXiv:0801.1021 \[hep-th\]\]; S. Alexeyev, N. Popov, M. Startseva, A. Barrau and J. Grain, J. Exp. Theor. Phys. [**106**]{}, 709 (2008) \[arXiv:0712.3546 \[gr-qc\]\]. A. Anabalon, N. Deruelle, Y. Morisawa, J. Oliva, M. Sasaki, D. Tempo and R. Troncoso, Class. Quant. Grav. [**26**]{}, 065002 (2009) \[arXiv:0812.3194 \[hep-th\]\]. S. Nojiri, S. D. Odintsov and S. Ogushi, Phys. Rev. D **65**, 023521 (2002) \[arXiv:hep-th/0108172\]; M. Cvetic, S. Nojiri and S. D. Odintsov, Nucl. Phys. B **628**, 295 (2002) \[arXiv:hep-th/0112045\]; Y. M. Cho and I. P. Neupane, Phys. Rev. D **66**, 024044 (2002) \[arXiv:hep-th/0202140\]; I. P. Neupane, Phys. Rev. D **67**, 061501 (2003) \[arXiv:hep-th/0212092\]; I. P. Neupane, Phys. Rev. D **69**, 084011 (2004) \[arXiv:hep-th/0302132\]; M. H. Dehghani and R. B. Mann, arXiv:hep-th/0602243; M. H. Dehghani and S. H. Hendi, arXiv:hep-th/0602069. M. H. Dehghani and R. B. Mann, Phys. Rev. D **72**, 124006 (2005) \[arXiv:hep-th/0510083\]. M. H. Dehghani and M. Shamirzaie, Phys. Rev. D **72**, 124015 (2005) \[arXiv:hep-th/0506227\]; Y. S. Myung, Phys. Lett. B **624**, 297 (2005) \[arXiv:hep-th/0506096\]; P. Mora, R. Olea, R. Troncoso and J. Zanelli, JHEP **0406**, 036 (2004) \[arXiv:hep-th/0405267\]; M. Banados, R. Olea and S. Theisen, JHEP **0510**, 067 (2005) \[arXiv:hep-th/0509179\]; R. G. Cai and N. Ohta, Phys. Rev. D [**74**]{}, 064001 (2006) \[arXiv:hep-th/0604088\]; D. Kastor and R. B. Mann, JHEP [**0604**]{}, 048 (2006) \[arXiv:hep-th/0603168\]; G. Giribet, J. Oliva and R. Troncoso, JHEP [**0605**]{}, 007 (2006) \[arXiv:hep-th/0603177\]; Y. S. Myung, Y. W. Kim and Y. J. Park, Eur. Phys. J. C [**58**]{}, 337 (2008) \[arXiv:0806.4452 \[gr-qc\]\]. Z. K. Guo, N. Ohta and T. Torii, Prog. Theor. Phys. [**120**]{}, 581 (2008) \[arXiv:0806.2481 \[gr-qc\]\]. S. H. Mazharimousavi, O. Gurtug and M. Halilsoy, arXiv:0809.3649 \[gr-qc\]; G. Dotti, J. Oliva and R. Troncoso, Int. J. Mod. Phys. A [**24**]{}, 1690 (2009) \[arXiv:0809.4378 \[hep-th\]\]; R. G. Cai, L. M. Cao, Y. P. Hu and S. P. Kim, Phys. Rev. D [**78**]{}, 124012 (2008) \[arXiv:0810.2610 \[hep-th\]\]; A. Khodam-Mohammadi and M. Monshizadeh, Phys. Rev. D [**79**]{}, 044002 (2009) \[arXiv:0811.1268 \[hep-th\]\]; Z. K. Guo, N. Ohta and T. Torii, Prog. Theor. Phys. [**121**]{}, 253 (2009) \[arXiv:0811.3068 \[gr-qc\]\]; H. Maeda, M. Hassaine and C. Martinez, Phys. Rev. D [**79**]{}, 044012 (2009) \[arXiv:0812.2038 \[gr-qc\]\]; C. Bogdanos, Phys. Rev. D [**79**]{}, 107501 (2009) \[arXiv:0902.2703 \[gr-qc\]\]; N. Ohta and T. Torii, Prog. Theor. Phys. [**121**]{}, 959 (2009) \[arXiv:0902.4072 \[hep-th\]\]; M. H. Dehghani and R. Pourhasan, Phys. Rev. D [**79**]{}, 064015 (2009) \[arXiv:0903.4260 \[gr-qc\]\]; M. H. Dehghani and S. H. Hendi, Gen. Rel. Grav. [**41**]{}, 1853 (2009) \[arXiv:0903.4259 \[hep-th\]\]; R. Biswas and S. Chakraborty, arXiv:0905.1801 \[gr-qc\]; A. N. Petrov, Class. Quant. Grav. [**26**]{}, 135010 (2009) \[arXiv:0905.3622 \[gr-qc\]\]; C. Bogdanos, C. Charmousis, B. Gouteraux and R. Zegers, JHEP [**0910**]{}, 037 (2009) \[arXiv:0906.4953 \[hep-th\]\]; M. H. Dehghani, N. Bostani and R. Pourhasan, arXiv:0908.0663 \[gr-qc\]; N. Ohta and T. Torii, arXiv:0908.3918 \[hep-th\]; K. i. Maeda, N. Ohta and Y. Sasagawa, arXiv:0908.4151 \[hep-th\]; C. M. Chen, D. V. Gal’tsov, N. Ohta and D. G. Orlov, arXiv:0910.3488 \[hep-th\]. C. Garraffo and G. Giribet, Mod. Phys. Lett. A [**23**]{}, 1801 (2008) \[arXiv:0805.3575 \[gr-qc\]\]. H. Maeda and N. Dadhich, Phys. Rev. D [**74**]{}, 021501 (2006) \[arXiv:hep-th/0605031\]. H. Maeda and N. Dadhich, Phys. Rev. D [**75**]{}, 044007 (2007) \[arXiv:hep-th/0611188\]. N. Dadhich and H. Maeda, Int. J. Mod. Phys. D [**17**]{}, 513 (2008) \[arXiv:0705.2490 \[hep-th\]\]. A. Molina and N. Dadhich, Int. J. Mod. Phys. D [**18**]{}, 599 (2009) \[arXiv:0804.1194 \[gr-qc\]\]. B. Cuadros-Melgar, E. Papantonopoulos, M. Tsoukalas and V. Zamarias, Phys. Rev. Lett. [**100**]{}, 221601 (2008) \[arXiv:0712.3232 \[hep-th\]\]. B. Cuadros-Melgar, E. Papantonopoulos, M. Tsoukalas and V. Zamarias, Nucl. Phys. B [**810**]{}, 246 (2009) \[arXiv:0804.4459 \[hep-th\]\].
N. Dadhich, R. Maartens, P. Papadopoulos and V. Rezania, Phys. Lett. B [**487**]{}, 1 (2000) \[arXiv:hep-th/0003061\].
T. Shiromizu, K. i. Maeda and M. Sasaki, Phys. Rev. D [**62**]{}, 024012 (2000) \[arXiv:gr-qc/9910076\]. R. M. Wald, Phys. Rev. D [**48**]{}, 3427 (1993) \[arXiv:gr-qc/9307038\]. V. Iyer and R. M. Wald, Phys. Rev. D [**50**]{}, 846 (1994) \[arXiv:gr-qc/9403028\]. T. Jacobson and R. C. Myers, Phys. Rev. Lett. [**70**]{}, 3684 (1993) \[arXiv:hep-th/9305016\]. R. G. Cai, L. M. Cao and N. Ohta, Phys. Rev. D [**80**]{}, 024003 (2009) \[arXiv:0904.3670 \[hep-th\]\]; R. G. Cai, Y. Liu and Y. W. Sun, JHEP [**0906**]{}, 010 (2009) \[arXiv:0904.4104 \[hep-th\]\]; R. G. Cai, L. M. Cao and N. Ohta, Phys. Lett. B [**679**]{}, 504 (2009) \[arXiv:0905.0751 \[hep-th\]\]; R. G. Cai and N. Ohta, arXiv:0910.2307 \[hep-th\]; Y. S. Myung, arXiv:0908.4132 \[hep-th\]; E. Kiritsis and G. Kofinas, arXiv:0910.5487 \[hep-th\].
R. C. Myers, Phys. Rev. D [**36**]{}, 392 (1987).
S. C. Davis, Phys. Rev. D [**67**]{}, 024030 (2003) \[arXiv:hep-th/0208205\]. E. Gravanis and S. Willison, Phys. Lett. B [**562**]{}, 118 (2003) \[arXiv:hep-th/0209076\].
R. G. Cai, Y. Liu and Y. W. Sun, arXiv:0909.2807 \[hep-th\].
R. Brustein, D. Gorbonos and M. Hadad, Phys. Rev. D [**79**]{}, 044025 (2009) \[arXiv:0712.3206 \[hep-th\]\]. G. ’t Hooft, Nucl. Phys. B [**256**]{}, 727 (1985).
[^1]: e-mail address: [email protected]
[^2]: e-mail address: [email protected]
[^3]: e-mail address: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We show that all density operators of 2$\times N$–dimensional quantum systems that remain invariant after partial transposition with respect to the first system are separable. Based on this criterion, we derive a sufficient separability condition for general density operators in such quantum systems. We also give a simple proof of the separability criterion in $2\times 2$–dimensional systems \[A. Peres, Phys. Rev. Lett [**77**]{}, 1413 (1996)\].'
address:
- '$^1$Institut für Theoretische Physik, Universität Hannover, D-30163 Hannover, Germany'
- '$^2$Institut für Theoretische Physik, Universität Innsbruck, A-6020 Innsbruck, Austria'
author:
- 'M. Lewenstein$^1$, J. I. Cirac$^2$, and S. Karnas$^1$'
title: 'Separability and entanglement in 2$\times N$ composite quantum systems'
---
= 10000
Perhaps, entanglement is the most intriguing property of Quantum Mechanics. It arises when two or more systems are a non–separable state; that is, in a state that cannot be prepared locally [@We89]. Apart from having fundamental implications[@Peresbook], such as Bell’s theorem or the existence of decoherence, entanglement is in the realm of several practical applications of Quantum Information [@Di95].
Despite of its importance, we do not know yet how to quantify entanglement. In fact, even in the simple case in which we only have two systems $A$ and $B$ in a mixed state $\rho$, there exists no general criterion that allows us to distinguish whether the state is separable or not; that is, whether it can be written as =\_i |e\_i,f\_ie\_i,f\_i|, where $|e_i,f_i\rangle\equiv |e_i\rangle_A\otimes|f_i\rangle_B$ are product states. An important step forward to clarify this situation was taken by Peres [@Pe96] and the Horodecki family [@Ho96]. They found such a criterion for the cases in which the Hilbert space corresponding to the first subsystem is two–dimensional and the one of the second is either two– or three– dimensional, the so–called $2\times 2$ and $2\times 3$ cases, respectively. Although the criterion is very simple, its proof involves mathematically advanced concepts [@St63], and it is not easily accessible to most physicists working on Quantum Information.
Let us consider the $M\times N$ case, where the Hilbert spaces $\C^M$ and $\C^N$ corresponding to systems $A$ and $B$ have dimension $M$ and $N$, respectively. Given an operator $X$ and an orthonormal basis $|m\rangle_A$ ($m=1,\ldots,M$) for the first system, the partial transpose of $X$ with respect to system $A$ in that basis is defined as \[parttran\] X\^[T\_A]{} = \_[n,m=1]{}\^M \_An||m \_A |m\_An| The separability criterion introduced by Peres [@Pe96] states that for $M=2$ and $N=2,3$ the density operator $\rho$ describing the state of systems is separable iff $\rho^{T_A}\ge 0$, i.e. if its partial transpose is a positive operator. Surprisingly enough, this criterion is not valid for higher values of $N$ or $M$ [@Ho98b]. In particular, for $N=M=3$ (i.e. two three–level systems), Bennett [*et al.*]{} [@Be98] have found families of non–separable states fulfilling $\rho^{T_A}=\rho\ge
0$. These states are related to the existence of sets of product states that cannot be extended with other product states to form a basis [@Be98], and play a crucial role in the understanding of concepts like non–locality [@Be98b] or EPR paradox without entanglement [@Ho98]. One may wonder whether this class of non–separable states exists in other dimensions. In this Letter we show that this is not the case for $2\times N$ systems. More specifically, we show that if $\rho^{T_A}=\rho$ then necessarily $\rho$ must be separable. Based on this fact, we give a sufficient condition for separability in these systems [@Sepcond]. Furthermore, our results also allow to derive a proof based on simple concepts of linear algebra of the Peres criterion in $2\times 2$ systems.
Given the fact that some of the results we have obtained to prove our claims may be interesting in other applications of Quantum Information Theory, we present them in the form of lemmas. This also leads to a more compact and readable structure of this Letter.
We consider a density operator $\rho\ge 0$ acting on $\C^2\otimes \C^N$ and with $\rho^{T_A} \ge 0$. We will denote by $K(\rho)$, $R(\rho)$ and $r(\rho)$ the kernel, range, and rank of $\rho$, respectively. The partial transpose will be taken with respect to a given basis $\{|0\rangle,|1\rangle\}\in \C^2$ of system A. We will denote by $|e^\ast\rangle$ the complex conjugated vector of $|e\rangle$ in that basis; that is, if $|e\rangle=\alpha|0\rangle +\beta|1\rangle$ then $|e^\ast\rangle=\alpha^\ast|0\rangle +\beta^\ast|1\rangle$. We will use the subscript “r” to denote real vectors (e.g. $|e_{\rm
r}\rangle=|e_{\rm r}^\ast\rangle$), and we will denote by $|\hat
e\rangle$ the vector orthogonal to $|e\rangle\in \C^2$, i.e. $\langle
e|\hat e\rangle=0$. Throughout this paper we will make use of the property \[property\] \_Ae\_1||e\_2\_A = \_Ae\^\_2|\^[T\_A]{}|e\_1\^\_A for any pair of vectors $|e_{1,2}\rangle \in \C^2$; this property follows directly from the definition of partial transposition (\[parttran\]). Unless we specify it and in order to simplify the notation, will consider all the states of system $A$ normalized and the ones of system $B$ unnormalized.
In order to carry out our analysis we need to recall a lemma and a corollary introduced in Ref. [@Le98]. For their proofs we refer the reader to this reference.
[**Lemma 1:**]{} If $|e,g\rangle\in R(\rho)$ then $\rho_1\equiv
\rho-(\langle e,g|\rho^{-1}|e,g\rangle)^{-1} |e,g\rangle\langle e,g|$ is positive and $\rho^{-1}|e,g\rangle \in K(\rho_1)$.
[**Corollary 1:**]{} Let $|e,g\rangle\in R(\rho)$, $|e^\ast,g\rangle\in
R(\rho^{T_A})$, $\lambda_1\equiv (\langle
e,g|\rho^{-1}|e,g\rangle)^{-1}$ and $\lambda_2 \equiv (\langle
e^\ast,g|(\rho^{T_a})^{-1}|e^\ast,g\rangle)^{-1}$. Then $\rho_1\equiv
\rho- \lambda |e,g\rangle\langle e,g|$ with $\lambda= {\rm
min}[\lambda_1,\lambda_2]$ fulfills $\rho_1,\rho_1^{T_A}\ge 0$ and $r(\rho_1)=r(\rho)-1$ \[$r(\rho^{T_A}_1)=r(\rho^{T_A})-1$\] if $\lambda_1\le \lambda_2$ \[$\lambda_2\le \lambda_1$\].
We will now consider some results that imply that if there is a product vector in the kernel of $\rho$, then we can reduce the dimensionality of the second system.
[**Lemma 2:**]{} $|e,f\rangle \in K(\rho)$ iff $|e^\ast,f\rangle \in
K(\rho^{T_A})$.
[*Proof:*]{} If $|e,f\rangle \in K(\rho)$ we have $0=\langle
e,f|\rho|e,f\rangle=\langle e^\ast,f|\rho^{T_A}|e^\ast,f\rangle$. Given that $\rho^{T_A}\ge 0$ this implies that $\rho^{T_A}|e^\ast,f\rangle=0$. Similary, we have that if $|e^\ast,f\rangle \in K(\rho^{T_A})$ then $\rho|e,f\rangle=0$.$\Box$
[**Lemma 3:**]{} If $|e,f\rangle \in K(\rho)$, then one of the following statements is true: (i) $|\hat e,f\rangle \in K(\rho)$ and $|\hat
e^\ast,f\rangle \in K(\rho^{T_A})$; (ii) There exists a non-zero $|g\rangle\in \C^N$ such that $\rho|\hat e,f\rangle=|\hat e,g\rangle$ and $\rho^{T_A}|\hat
e^\ast,f\rangle=|\hat e^\ast,g\rangle$.
[*Proof:*]{} Using Lemma 2 we have that for all $|h\rangle\in \C^N$ we can write $0=\langle \hat e^\ast,h|\rho^{T_A}|e^\ast,f\rangle=\langle
e,h|\rho|\hat e,f\rangle$ and therefore either $\rho|\hat e,f\rangle=0$ or $\rho|\hat e,f\rangle=|\hat e,g_1\rangle$ for some $|g_1\rangle\in
\C^N$. In a similar way we have that either $\rho^{T_A}|\hat
e^\ast,f\rangle=0$ or $\rho^{T_A}|\hat e^\ast,f\rangle=|\hat
e^\ast,g_2\rangle$ for some $|g_2\rangle\in \C^N$. Using Lemma 2 we conclude that there are two possibilities: (i) $\rho|\hat e,f\rangle=0$ and $\rho^{T_A}|\hat e^\ast,f\rangle=0$, which coincides with the first statement of the Lemma; (ii) $\rho|\hat e,f\rangle=|\hat e,g_1\rangle$ and $\rho^{T_A}|\hat e^\ast,f\rangle=|\hat e^\ast,g_2\rangle$ for some $|g_1\rangle,|g_2\rangle \in \C^N$. We have $|g_2\rangle=\langle \hat
e^\ast|\rho^{T_A}|\hat e^\ast,f\rangle =\langle
e|\rho|\hat e,f\rangle=|g_1\rangle$ which corresponds to the second statement.$\Box$
[**Lemma 4:**]{} If $\exists |e,f\rangle \in K(\rho)$ then we can write $\rho=\rho_1 + \rho_s$, where $\rho_s$ is separable, $\rho_1$ is supported on $\C^2\otimes \C^{N-1}$ and $\rho_1,\rho_1^{T_A}\ge 0$.
[*Proof:*]{} According to Lemma 3, we can have two situations: (i) $|e,f\rangle,|\hat e,f\rangle\in K(\rho)$ and therefore $\rho$ is already supported on $\C^2\otimes \C^{N-1}$; (ii) $\rho^{-1}|\hat
e,g\rangle=|\hat e,f\rangle$ and $(\rho^{T_A})^{-1}|\hat
e^\ast,g\rangle=|\hat e^\ast,f\rangle$. If we define \[rho1\] \_[1]{}=- |e,ge,g| =- |e,ge,g| then $\rho_{1}^{T_A}=\rho^{T_A} - (\langle \hat
e^\ast,g|(\rho^{T_A})^{-1}|\hat e^\ast,g\rangle)^{-1} |\hat
e^\ast,g\rangle\langle \hat e^\ast,g|$. According to Lemma 1, $\rho_1,\rho_1^{T_A}\ge 0$. Moreover, $\rho_1|\hat e,f\rangle=\rho|\hat
e,f\rangle-(\langle g|f\rangle)^{-1} |\hat e,g\rangle\langle \hat
e,g|\hat e,f\rangle =0$ and $\rho_1|e,f\rangle=\rho|e,f\rangle=0$. Thus $|e,f\rangle,|\hat e,f\rangle\in K(\rho_1)$ and therefore $\rho_1$ is supported on $\C^2\otimes \C^{N-1}$. Note that in this case $r(\rho_1)=r(\rho)-1$. $\Box$
This powerful lemma simply states that if there is a product vector in the kernel of $\rho$, then we can reduce the dimensionality of the second system keeping positive the partial transposed operator. On the other hand, Corolary 1 allows to reduce the rank of $\rho$, i.e. to increase the dimension of the kernel of $\rho$, if we find appropriate product vectors in its range. Thus, the idea is to find out whether there are product vectors in the kernel of $\rho$. If there are, we reduce the dimensionality. If not, then we try to find product vectors in the range until we make the kernel “sufficiently large” to include a product vector in it. Thus, the important question is to find out when we can ensure that there are product vectors in the range or in the kernel. The following lemma states that we can always find product vectors in any subspace of dimension $\ge N$.
[**Lemma 5:**]{} Any subspace $H\subset \C^2\otimes \C^N$ with dim$(H)=M\ge N$ contains at least one product vector. For $M>N$ it contains at least one product vector $|e_{\rm r},f\rangle$.
[*Proof:*]{} Lets us denote by $\{ |\Psi_i\rangle,
i=1,\ldots,2N-M\}$ an orthonormal basis in the orthogonal complement of $H$. We write \[Psi\] |\_i= \_[k=1]{}\^N A\_[i,k]{}\^|0,k+ B\_[i,k]{}\^|1,k, and $A^\dagger A + B^\dagger B=1$. We look for product vectors in $H$ of the form $|\Phi\rangle=(\alpha|0\rangle+|1\rangle)\otimes \sum f_k
|k\rangle$, i.e. \[AB\] (A + B)f =0. If $M>N$ we have more variables than equations, and therefore there is a solution for all $\alpha$, and in particular for $\alpha$ real. For $M=N$ we have that there is always a solution since ${\rm det}(\alpha A + B)$ is a polynomial of degree $N$ in $\alpha$.$\Box$
From the results that we have obtained so far, we have the following
[**Theorem 1:**]{} If $r(\rho)=N$, and $K(\rho)$ does not contain any pair $|e,f\rangle,|\hat e, f\rangle$ \[i.e. $R(\rho)$ is not supported on $\C^2\otimes \C^{N-1}$\], then $\rho$ is separable.
[*Proof:*]{} We use induction. For $N=1$ the statement is true. Let us assume that it is true for $N-1$. If $r(\rho)=N$, then $K(\rho)$ has dimension $N$, and according to Lemma 5 it contains a product vector $|e,f\rangle$. Lemma 3 tells us that $\rho|\hat e,f\rangle=|\hat
e,g\rangle$, with $|g\rangle \ne 0$ and from Lemma 4 we obtain $\rho=\rho_1 + \rho_s$, with $\rho_1\ge 0$ supported on $\C^2\otimes
\C^{N-1}$ and $r(\rho_1)=N-1$. Now we show that $\rho_1$ cannot be supported on $\C^2\otimes \C^{N-2}$, which would complete the proof. If it were, there would be two vectors $|e,h\rangle,|\hat e,h\rangle\in
K(\rho_1)$. Using that $\rho_s\propto |\hat e,g\rangle\langle \hat
e,g\rangle$ we would have $\rho|e,h\rangle=0$ and $\rho|\hat e,h\rangle=
c |\hat e,g\rangle$ where $c$ is a constant. If we define $|f'\rangle=c|f\rangle-|h\rangle$ we have that $|e,f'\rangle,|\hat
e,f'\rangle\in K(\rho)$, contrary to our assumption.$\Box$
Now, we turn to the case $\rho=\rho^{T_A}$, and therefore specialize our previous results to this case. In particular, we formulate a resut similar to Lemma 4.
[**Lemma 4b:**]{} If $\rho=\rho^{T_A}$ and $\exists |e,f\rangle \in
K(\rho)$ then we can write $\rho=\rho_{N-1} + \rho_s$, where $\rho_s$ is separable, $\rho_{N-1}$ is supported on $\C^2\otimes \C^{N-1}$ and $\rho_{N-1}^{T_A}=\rho_{N-1}$.
[*Proof:*]{} First, according to Lemma 2 we have that $|e^\ast,f\rangle\in K(\rho^{T_A})=K(\rho)$ and therefore we obtain that $|e_{\rm r},f\rangle\in K(\rho)$ with $|e_{\rm r}\rangle=|e\rangle+|e^\ast\rangle$ real (if this vector is zero we can take $|e_{\rm r}\rangle=(|e\rangle-|e^\ast\rangle)/i$). Now, using the vector $|e_{\rm r}\rangle$ instead of $|e\rangle$ we can derive the results of Lemmas 3 and 4a but with real vectors $|\hat e_r\rangle$. Thus, according to (\[rho1\]) we have $\rho_{1}^{T_A} =\rho^{T_A}-(\langle g|f\rangle)^{-1}
|\hat e_{\rm r}^\ast,g\rangle\langle \hat e_{\rm r}^\ast,g|=\rho_1$.$\Box$
We are not at the position of proving the main result of this paper:
[**Theorem 2:**]{} If $\rho=\rho^{T_A}$ then $\rho$ is separable.
[*Proof:*]{} We prove it by induction. First, in the case $2\times 1$ (i.e. $N=1$) it is obviously true. Now, let us assume that it is valid for the case $2\times (N-1)$ and let us prove that then it is also valid for the case $2\times N$. In order to do that, we will show that any density operator $\rho$ can be decomposed as $\rho=\rho_1 + \rho_s$ where $\rho_s$ is separable and $\rho_1=\rho_1^{T_A}$ is supported on $\C^2\otimes \C^{N-1}$ and therefore is also separable according to our assumption. We consider two cases: (i) $r(\rho)\le N$: we have dim$[K(\rho)]\ge N$; using Lemma 5, there is a product vector in $K(\rho)$, so that according to Lemma 4b we have obtained the desired decomposition. (ii) $r(\rho)>N$: according to Lemma 5 there is a product vector $|e_{\rm r},g\rangle$ with $|e_{\rm r}\rangle=|e_{\rm
r}^\ast\rangle$. Using Corollary 1 we can use this product vector to reduce the rank of $\rho$. But since $|e_{\rm r},g\rangle =|e_{\rm
r}^\ast,g\rangle$ we have that the resulting operator is also equal to its partial transpose. We can proceed in this way until $r(\rho)=N$ which corresponds to the case (i), and therefore we complete the proof.$\Box$
Note that we could have chosen a different basis for the partial transposition. Taking into account that for any symmetric unitary operator we can always find a basis in which partial transposition is related to the original one by that such operator, we have
[**Corollary 2:**]{} If $\rho=(U_A\otimes 1)
\rho^T (U_A\otimes 1)^\dagger$ for some unitary operator $U_A=U_A^{T_A}$ then $\rho$ is separable [@gener].
Theorem 2 suggests that if $\rho$ is not very different from $\rho^{T_2}$, then it should also be separable. Indeed, one can construct a powerful sufficient separability condition based on that theorem. We will first introduce a lemma that generalizes Lemma 1 and that is useful to determine whether the difference of two positive operators is positive. We will use the operator norm which is defined as usual $\|A\| = \max \| A|x\rangle\|$ with $\||x\rangle=1$.
[**Lemma 6:**]{} Given two hermintian operators $X,Y\ge 0$, $X-Y\ge 0$ iff $Y$ is supported on $R(X)$ and $\|Y^{1/2} X^{-1/2} \|^2 \le 1$.
[*Proof:*]{} If $Y$ is not supported on $R(X)$ we have that $\exists
|\phi\rangle \in R(Y),K(X)$ and therefore $\langle\phi|X-Y|\phi\rangle=-\langle\phi|Y|\phi\rangle<0$. On the contrary, if $Y$ is supported on $R(X)$ we have that $X-Y\ge 0$ iff $\forall |\varphi\rangle\in R(X)$ we have $\langle\varphi|Y|\varphi\rangle/\langle\varphi|X|\varphi\rangle\le 1$. Denoting by $|\psi\rangle=X^{1/2}|\varphi\rangle$ we obtain that $X-Y\ge
0$ iff 1\_[|]{} = Y\^[1/2]{}X\^[-1/2]{}\^2.
In order to give the sufficient condition for separability, we have to introduce some definitions. In $2 \times N$ we can always write $\rho$ as \[dec2\] = + \_s + \^A\_yB, where $2\rho_s=\rho+\rho^{T_A}$, $\sigma^A_y=i(|0\rangle_A\langle
1|-|1\rangle_A\langle 0|)$, and $2B=2B^\dagger={\rm
tr_A}[\sigma_y^A(\rho-\rho^{T_A})]$. This operator $B$ can be decomposed as B=\_[i=1]{}\^K \_i |v\_iv\_i|. In particular, one of such is the spectral decomposition. Given one of such decomposition $\{\lambda_i,|v_i\rangle\}_{i=1}^K$ and a set of real numbers $\{a_i\}_{i=1}^K$ we define the operator C(a,,v)\_[i=1]{}\^K |\_i| (a\_i\^2|00| +a\_i\^[-2]{}|11|)|v\_iv\_i|, which is obviously positive. We have:
[**Theorem 3:**]{} Given a decomposition of $B$ $\{\lambda_i,|v_i\rangle\}_{i=1}^K$ and a set of real numbers $\{a_i\}_{i=1}^K$, if $\|C^{1/2}(a,\lambda,v)\rho_s^{-1/2}\|^2\le 1$, then $\rho$ is separable.
[*Proof:*]{} We define $\tilde\rho_s=\rho_s-C(a,\lambda,v)=
\tilde\rho_s^{T_A}\ge 0$ according to Lemma 6. Using Theorem 2 we have that $\tilde\rho_s$ is separable. Let $|w_i\rangle=a_i|0\rangle
-ia_i^{-1} {\rm sign}(\lambda_i)|1\rangle$. Then, it is easy to check that \[end\] =\_s + \_[i=1]{}\^K|\_i| |w\_i,v\_iw\_i,v\_i|. which shows that $\rho$ is separable.$\Box$
Thus, we can show that a density operator is positive if we can find a decomposition of $B$ and a set of real numbers that fulfill certain conditions. In particular we can take the spectral decomposition of $B$ and $a_i=1$. Using the fact that $\|AB\| \le \|A\| \|B\|$ one can easily prove the following:
[**Corollary 3:**]{} If $\rho+\rho^{T_A}$ is of full range and $\|(\rho +
\rho^{T_A})^{-1}\| \|\rho-\rho^{T_A}\| \le 1$, then $\rho$ is separable. This corollary implies that if $\rho$ is full range and is very close to $\rho^{T_A}$ then it is separable.
The results introduced in the first part of this Letter also allow to prove Peres criterion [@Pe96] for separability in the $2\times 2$ case. In particular, in the following we will show that in these systems if $\rho^{T_A}\ge 0$ then $\rho$ is separable. Note that the converse can be easily proved [@Pe96].
[**Theorem 4:**]{} (Peres, [@Pe96]). If $\rho,\rho^{T_A}\ge 0$ are operators supported on $\C^2\otimes \C^2$, then $\rho$ is separable.
[*Proof:*]{} We just have to consider the case ${\rm dim}[K(\rho)]={\rm
dim}[K(\rho^{T_A})]=1$ in which both $K(\rho)$ and $K(\rho^{T_A})$ contain no product vector. The reason is that: (i) if $K(\rho)$ or $K(\rho^{T_A})$ contain a product vector then using Lemma 4a we can reduce the problem to the $2\times 1$ case, and therefore $\rho$ is separable; (ii) If ${\rm dim}[K(\rho)]=2$ then according to Lemma 5 there is a product vector in $K(\rho)$, so that we are back in (i) and therefore $\rho$ is separable; (iii) similarly, if ${\rm
dim}[K(\rho^{T_A})]=2$ we conclude that $\rho^{T_A}$ and therefore $\rho$ are separable; (iv) If ${\dim K}(\rho^{T_A})=0$ and ${\dim
K}(\rho)<2$ then according to Lemma 5 $R(\rho)$ contains at least one product vector $|e,f\rangle$ and obviously $|e^\ast,f\rangle\in
R(\rho^{T_A})$; we can apply Corolary 1 and substract product vectors from $\rho$ until either ${\rm dim}K(\rho)=2$, in which case $\rho$ is separable, or ${\rm dim}[K(\rho)]={\rm dim}[K(\rho^{T_A})]=1$; (v) Similarly If ${\dim K}(\rho)=0$ and ${\dim K}(\rho^{T_A})<2$ we arrive at the same conclusion. Thus, let us assume that $\rho|\Psi_1\rangle=0$ and $\rho^{T_A}|\Psi_2\rangle=0$, where $|\Psi_{1,2}\rangle$ are of the form (\[Psi\]) and they are not product vectors. We will first show that these vectors can always be writen as \[dec\] |\_1&& |e,f- |e,g,\
|\_2&& |e\^,h- |e\^,f, where both $\{|f\rangle,|g\rangle\}$ ($\{|f\rangle,|h\rangle\}$) are linearly independent since otherwise $|\Psi_1\rangle$ ($|\Psi_2\rangle$) would be a product vector. Then we will show that either $\rho=\rho^{T_A}$ or $|h\rangle=k^2|g\rangle$ where $k^2>0$. In the first case, according to Theorem 2, $\rho$ is separable. In the second case, we have that if we define $|a\rangle=|e\rangle + k |\hat e\rangle$ and $|b\rangle = \langle f|\hat g\rangle
|\hat f\rangle - k \langle g|\hat g\rangle |\hat g\rangle$ then the vector $|a,b\rangle\in R(\rho)$ and $|a^\ast,b\rangle\in R(\rho^{T_A})$, since they are orthogonal to $|\Psi_{1,2}\rangle$ respectively. Thus, according to Corolary 1 we can substract this product vector and increase the dimension of either $K(\rho)$ or $K(\rho^{T_A})$ to 2, which as we showed in (ii) and (iii) implies that $\rho$ is separable.
We show how to obtain the decomposition (\[dec\]). We look for two unnormalized vector $|e\rangle = \alpha|0\rangle + |1\rangle$ and $|\tilde f\rangle=f_0|0\rangle+f_1|1\rangle$ such that $\langle
e,\tilde f|\Psi_1\rangle = \langle \hat e^\ast,\tilde f|\Psi_2\rangle=0$. We obtain two linear equations for $f_0$ and $f_1$ of the form $C(\alpha) \vec
f=0$, where $C$ is a matrix that depends linearly on $\alpha$ (and not on $\alpha^\ast$ [@note]). These equations always have a solution for some given $\alpha$ since the condition ${\rm det}[C(\alpha)]=0$ is a second order equation for $\alpha$.
We finally show that either $\rho=\rho^{T_A}$ or $|h\rangle=k^2|g\rangle$ where $k^2>0$. First, we show that \[state\] f||f= g||h= h||g. To obtain that we use the fact that the vectors $|\Psi\rangle$ are in the kernel of $\rho$ or $\rho^{T_A}$, and therefore $\rho|e,f\rangle=\rho|\hat e,g\rangle$ and $\rho^{T_A}|e^\ast,h\rangle=\rho^{T_a}|\hat e^\ast,f\rangle$. Using (\[property\]) we can write the second equation as $\langle
e|\rho|h\rangle = \langle \hat e|\rho|f\rangle$. Using these equations it is easy to obtain that $\langle e,f|\rho|f\rangle=\langle e,g|\rho|h\rangle$ and $\langle \hat e,f|\rho|f\rangle=\langle \hat
e,g|\rho|h\rangle$ which automatically proves the statement (\[state\]). Now, given that $|f\rangle$ and $|g\rangle$ are linearly independent vectors, we can write $|h\rangle=\alpha|f\rangle+\beta|g\rangle$ with $\beta\ne 0$. If $\alpha=0$, then using (\[state\]) we have that $\beta\equiv k^2>0$ is real and positive. If $\alpha\ne 0$ we have that both $\langle g|\rho|f\rangle$ and $\langle f|\rho|g\rangle$ can be expressed as linear combinations of $\langle f|\rho|f\rangle$ and $\langle g|\rho|g\rangle$. These two last operators are hermitian and they are equal to their transpose in some given basis (if we write $\langle f|\rho|f\rangle\propto 1+\vec n_f\cdot \sigma$ and $\langle g|\rho|g\rangle\propto 1+\vec n_g\cdot \sigma$ then we just have to use the basis such that both $\vec n_f$ and $\vec n_g$ lie in the $x-z$ plane). Thus, $\langle g|\rho|f\rangle$ and $\langle f|\rho|g\rangle$ are also equal to their transposes in that basis. Since $|f\rangle$ and $|g\rangle$ are linearly independent we immediately arrive to the conclusion that if we consider the partial transpose in that basis we have that $\rho=\rho^{T_A}$, which completes the proof.$\Box$
Summarizing, we have demonstrated that all density operators on $2\times
N$ that remain invariant after partial transposition with respect to the first system are separable. Using this fact we have constructed a sufficient separability condition for such systems. We have also given a relatively simple proof of Peres criterion based on these results.
This work has been supported by Deutsche Forshungsgemeinschaft (SFB 407), the Österreichisher Fonds zur Frderung der wissenschaftlichen Forschung (SFB F15), and the European TMR network ERB-FMRX-CT96-0087. J. I. C. thanks the Univesity of Hannover for hospitality. We thank A. Sanpera and G. Vidal for fruitful discussions.
R. F. Werner Phys. Rev. A [**40**]{}, 4277 (1989).
See, for example, A. Peres, [*Quantum Theory: concepts and methods*]{}, (Kluwer Academic Publisheres, Dordrecht, 1993).
See, for example, D. P. DiVincenzo, Science [**270**]{},255 (1995).
A. Peres, Phys. Rev. Lett [**77**]{}, 1413 (1996).
M. Horodecki, P. Horodecki and R. Horodecki, Phys. Lett. A[**223**]{}, 8 (1996).
E. Stromer, Acta Math. [**110**]{}, 233 (1963); S. L. Woronowicz, Rep. Math. Phys. [**10**]{}, 165 (1976).
P. Horodecki, Phys.Lett. A[**232**]{}, 333 (1997); M. Horodecki, P. Horodecki, R. Horodecki, Phys. Rev. Lett. [**80**]{}, 5239 (1998).
C. H. Bennett [*et al.*]{}, quant-ph/9808030; B. Terhal, quant-ph/9810091.
C. H. Bennett [*et al.*]{}, quant-ph/9804053.
R. Horodecki, M. Horodecki, and P. Horodecki, quant-ph/9811004.
There exist very few sufficient conditions for separability in systems other than $2\times 2$ and $2\times 3$: G. Vidal and R. Tarrach, quant-ph/9806094; S. Braunstein [*et al.*]{}, quant-ph/9811018
M. Lewenstein and A. Sanpera, Phys. Rev. Lett. [**80**]{}, 2261 (1998); A. Sanpera, R. Tarrach, and G. Vidal, Phys. Rev. A[**58**]{}, 826 (1998).
This corolary can be generalized using arbitrary nonsingular matrices instead of unitary ones.
Note that if the coefficients depend on $\alpha^\ast$ as well it might happen that the second order polynomial equation does not have solutions. This is the reason why we cannot make use of Lemma 5 directly to show that there exists a vector $|a,b\rangle\in R(\rho)$ and $|a^\ast,b\rangle\in R(\rho^{T_A})$.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
The cross sections of $B_{c}$ absorption by $\pi $ mesons are calculated using hadronic Lagrangian based on SU(5) flavor symmetry. Calculated cross sections are found to be in range 2 to 7 mb and 0.2 to 2 mb for the processes $B_{c}^{+}\pi \rightarrow DB$ and $B_{c}^{+}\pi
\rightarrow D^{\ast }B^{\ast }$ respectively, when the monopole form factor is included. These results could be useful in calculating production rate of $B_{c}$ meson in relativistic heavy ion collisions.
author:
- |
M. A. K. Lodhi$^{a}$[^1], Faisal Akram$^{b}$[^2] and Shaheen Irfan$^{b}$[^3]\
$^{a}$*Department of Physics, MS 1051, Texas Tech University, Lubbock TX 79409, USA*\
$^{b}$*Center for High Energy Physics, Punjab University, Lahore, PAKISTAN*
title: 'Hadronic absorption cross sections of $B_{c}$'
---
PACS number(s): 13.75.Lb, 14.40.Nd, 25.75.-q
Introduction
============
T. Matsui and H. Satz [@Matsui1986] postulated that $J/\psi $ would be dissociated due to color Debye screening in deconfined phase of hadronic matter, called Quark-Gluon Plasma (QGP). Thus suppression of $J/\psi $ could be regarded as a signal for the existence of QGP. NA50 experiment at CERN [@NA50] has observed an anomalously large suppression of events with moderate to large transfer energy form the Pb + Pb collision at $P_{Lab}$ = 158 GeV/c. However, this observed suppression may also occur due to absorption by comoving hadrons. It has been argued by many authors that this phenomenon could be significant if the absorption cross section is in the range of at least few mb [Cassing1997,Armesto1998,Kahana1999,Gale1999,Spieles1999,Sa1999]{} Extensive work has been done to calculate these cross sections using perturbative QCD [@Kharzeev1994], QCD sum-rule approach [@sum-rule], quark potential models [@quark; @models] and hadronic Lagrangian based on flavor symmetry [@Lin2001; @Haglin2000; @Lin2000; @Liu2001].
Bottomonium states analogous to charmonium are also subjected to dissociation due color screening [@Matsui1986], therefore their suppression is also expected in QGP. Recently the most striking observation from CMS (Compact Muon Solenoid experiment) is that weakly bound states of the b-quark are heavily suppressed in Pb+Pb collisions [@cms]. This phenomenon is important for understanding the properties of the QGP. Once again the knowledge of absorption cross section is required to interpret the observed signal [Lin2001,Vogt1997]{}. It has also been suggested that the production rate of heavy mixed flavor hadrons would also be affected in the presence of QGP[Schro2000,CUP2002]{}. In order to calculate production rates one require complete knowledge of production mechanism in the presence of QGP and absorption cross sections by comoving hadrons. In this paper we have focused on $B_{c}$ meson. It is expected that $B_{c}$ production could be enhanced in the presence of QGP. Due to color Debye screening, QGP contains many unpaired $b(\overline{b})$ and $c(\overline{c})$ quarks, which upon encounter could form $B_{c}$ and probably survive in QGP due to relatively large binding energy [@Lodhi2007]. However, observed production rate would also depend upon the absorption cross section by hadronic comovers. $B_{c}$ absorption cross section by nucleons has been calculated in [Lodhi2007]{} using meson-baryon exchange model. This cross section is found to have value on the order of few mb. In this paper, we have calculated $B_{c}$ absorption cross sections by $\pi $ mesons using hadronic Lagrangian based on SU(5) flavor symmetry.
In Sec. II, we define hadronic Lagrangian and derive the interaction term relevant for $B_{c}$ absorption of $\pi $ mesons. In Sec. III, we calculate the absorption cross sections. In Sec. IV, we discuss the numerical values of different couplings used in the calculation. In Sec. V, we present numerical results of the cross sections with and without form factor. Finally, some concluding remarks are made in Sec. VI.
Interaction Lagrangian
======================
The following processes are studied in this work using SU(5) flavor symmetric Lagrangian.$$B_{c}^{+}\pi \rightarrow DB,\text{ }B_{c}^{-}\pi \rightarrow \overline{D}\overline{B},\text{ }B_{c}^{+}\pi \rightarrow D^{\ast }B^{\ast },\text{ }B_{c}^{-}\pi \rightarrow \overline{D}^{\ast }\overline{B}^{\ast } \label{1}$$
First and second processes are charge conjugation of each other and hence have same cross sections. Similarly third and fourth processes are also charge conjugation of each other and have same cross sections.
To calculate cross sections of the above processes, we use SU(5) flavor symmetric Lagrangian density [@Lin2001]. Free SU(5) Lagrangian density is given by,$$\mathcal{L}_{0}=Tr(\partial _{\mu }P^{\dagger }\partial ^{\mu }P)-\frac{1}{2}Tr(F_{\mu
\nu }^{\dagger }F^{\mu \nu }) \label{2}$$
Where, $F_{\mu \nu }=\partial _{\mu }V_{\nu }-\partial _{\nu
}V_{\mu } ,$ $P$ and $V_{\mu }$ denote pseudo-scalar and vector mesons matrices as given in ref. [@Lin2001].
The following minimal substitutions,$$\partial _{\mu }P\rightarrow D_{\mu }P=\partial _{\mu }P-\frac{ig}{2}[V_{\mu
},P] \label{3}$$$$F_{\mu \nu }\rightarrow F_{\mu \nu }-\frac{ig}{2}[V_{\mu },V_{\nu }]
\label{4}$$
produce the following interaction Lagrangian desnity.$$\begin{aligned}
\mathcal{L} &=&\mathcal{L}_{0}+igTr(\partial ^{\mu }p[P,V_{\mu }])-\frac{g^{2}}{4}Tr([P,V_{\mu
}]^{2}) \notag \\
&&+igTr(\partial ^{\mu }V^{\nu }[V_{\mu },V_{\nu }])+\frac{g^{2}}{8}Tr([V_{\mu },V_{\nu }]^{2})\end{aligned}$$
All mass terms, which breaks SU(5) symmetry, are added directly in the above Lagrangian. The Lagrangian density terms relevant for $B_{c}$ absorption by $\pi $ mesons are given by,
\[Lag\] $$\begin{aligned}
\mathcal{L}_{\pi DD^{\ast }} &=&ig_{\pi DD^{\ast }}D^{\ast \mu }\overrightarrow{\tau }\cdot (\overline{D}\partial _{\mu }\overrightarrow{\pi }-\partial _{\mu }\overline{D}\overrightarrow{\pi })+hc \\
\mathcal{L}_{\pi BB^{\ast }} &=&ig_{\pi BB^{\ast }}\overline{B}^{\ast \mu }\overrightarrow{\tau }\cdot (B\partial _{\mu }\overrightarrow{\pi }-\partial
_{\mu }B\overrightarrow{\pi })+hc \\
\mathcal{L}_{B_{c}BD^{\ast }} &=&ig_{B_{c}BD^{\ast }}D^{\ast \mu }(B_{c}^{-}\partial
_{\mu }B-\partial _{\mu }B_{c}^{-}B)+hc \\
\mathcal{L}_{B_{c}B^{\ast }D} &=&ig_{B_{c}B^{\ast }D}\overline{B}^{\ast \mu
}(B_{c}^{+}\partial _{\mu }\overline{D}-\partial _{\mu }B_{c}^{+}\overline{D})+hc \\
\mathcal{L}_{\pi B_{c}D^{\ast }B^{\ast }} &=&-g_{\pi B_{c}D^{\ast }B^{\ast }}B_{c}^{+}\overline{B}^{\ast \mu }\overrightarrow{\tau }\cdot \overrightarrow{\pi }\overline{D}_{\mu }^{\ast }+hc\end{aligned}$$
Where,
$$\begin{aligned}
D &=&\left(
\begin{array}{cc}
D^{0} & D^{+}\end{array}\right) ,\overline{D}=\left(
\begin{array}{cc}
\overline{D}^{0} & D^{-}\end{array}\right) ^{T},D_{\mu }^{\ast }=\left(
\begin{array}{cc}
D_{\mu }^{\ast 0} & D_{\mu }^{\ast +}\end{array}\right) , \notag \\
B &=&\left(
\begin{array}{cc}
B^{+} & B^{0}\end{array}\right) ^{T},B_{\mu }^{\ast }=\left(
\begin{array}{cc}
B_{\mu }^{\ast +} & B_{\mu }^{\ast 0}\end{array}\right) ^{T}, \notag \\
\overrightarrow{\pi } &=&\left( \pi _{1},\pi _{2},\pi _{3}\right) ,\pi ^{\pm
}=\frac{1}{\sqrt{2}}\left( \pi _{1}\mp i\pi _{2}\right)\end{aligned}$$
Here we follow the convention of representing a field by the symbol of the particle which it absorbs. The coupling constants in Eq. (6) are expressed in terms of SU(5) universal coupling constant $g$ as following.$$g_{\pi DD^{\ast }}=g_{\pi BB^{\ast }}=\frac{g}{4},\text{ }g_{B_{c}BD^{\ast
}}=g_{B_{c}B^{\ast }D}=\frac{g}{2\sqrt{2}},\text{ }g_{\pi B_{c}D^{\ast
}B^{\ast }}=\frac{g^{2}}{4\sqrt{2}}$$
It is also noted that SU(5) symmetry also implies the following relation between the couplings.$$g_{\pi B_{c}D^{\ast }B^{\ast }}=2g_{\pi DD^{\ast }}g_{B_{c}B^{\ast
}D}=2g_{\pi BB^{\ast }}g_{B_{c}BD^{\ast }}$$
$B_{c}$ absorption cross section
================================
Feynman diagrams of the process $B_{c}^{+}\pi \rightarrow DB$ are shown in Fig. \[fig1\]
![Feynman Diagrams for $B_{c}$ absorption process $B_{c}^{+}\protect\pi \rightarrow DB$.[]{data-label="fig1"}](fig1.eps){width="50.00000%"}
![Feynman Diagrams for $B_{c}$ absorption process $B_{c}^{+}\protect\pi \rightarrow D^{\ast }B^{\ast }.$[]{data-label="fig2"}](fig2.eps){width="70.00000%"}
Scattering amplitudes of these diagrams are given by,
$$\begin{aligned}
M_{1a} &=&g_{\pi DD^{\ast }}g_{B_{c}BD^{\ast }}(p_{1}+p_{3})_{\mu }\frac{-i}{t-m_{D^{\ast }}^{2}}\left( g^{\mu \nu }-\frac{(p_{1}-p_{3})^{\mu
}(p_{1}-p_{3})^{\nu }}{m_{D^{\ast }}^{2}}\right) (-p_{4}-p_{2})_{\nu } \\
M_{1b} &=&g_{\pi BB^{\ast }}g_{B_{c}B^{\ast }D}(p_{1}+p_{4})_{\mu }\frac{-i}{u-m_{B^{\ast }}^{2}}\left( g^{\mu \nu }-\frac{(p_{1}-p_{4})^{\mu
}(p_{1}-p_{4})^{\nu }}{m_{B^{\ast }}^{2}}\right) (-p_{3}-p_{2})_{\nu }\end{aligned}$$
Total amplitude is given by,
$$M_{1}=M_{1a}+M_{1b} \label{11}$$
Feynman diagrams of the process $B_{c}^{+}\pi \rightarrow D^{\ast
}B^{\ast } $ are shown in Fig. \[fig2\]
Scattering amplitudes of these diagrams are given by,
$$\begin{aligned}
M_{2a} &=&-g_{\pi DD^{\ast }}g_{B_{c}B^{\ast }D}(2p_{1}-p_{3})_{\mu }\frac{i}{t-m_{D}^{2}}(p_{2}-p_{1}+p_{3})_{\nu }\varepsilon _{r}^{\mu
}(p_{3})\varepsilon _{s}^{\nu }(p_{4}) \\
M_{2b} &=&-g_{\pi BB^{\ast }}g_{B_{c}BD^{\ast }}(2p_{1}-p_{4})_{\mu }\frac{i}{u-m_{B}^{2}}(p_{2}-p_{1}+p_{4})_{\nu }\varepsilon _{r}^{\mu
}(p_{3})\varepsilon _{s}^{\nu }(p_{4}) \\
M_{2c} &=&-ig_{\pi B_{c}B^{\ast }D^{\ast }}g_{\mu \nu }\varepsilon _{r}^{\mu
}(p_{3})\varepsilon _{s}^{\nu }(p_{4})\end{aligned}$$
And total amplitude is given by,
$$M_{2}=M_{2a}+M_{2b}+M_{2c} \label{13}$$
Using the total amplitudes given in Eqs. \[11\] and \[13\], we calculate unpolarized but not the isospin averaged cross sections. The isospin factor in this case is simply 2 for the both processes.
Numerical values of input parameters
====================================
Numerical values of all the masses are taken from Particle Data Group [PDG]{}. The coupling constant $g_{\pi DD^{\ast }}=4.4$, is determined from $D^{\ast }$ decay width [@Col1994; @Bel1995]. The coupling $g_{\pi
BB^{\ast }}$ can be fixed by two methods. Heavy quark symmetries [Bel1995,Wise1992,Grin1993]{} imply that $g_{\pi BB^{\ast }}\approx g_{\pi
DD^{\ast }}\frac{m_{B}}{m_{D}}=12.4$ and from light-cone QCD sum rule [Bel1995]{}, we obtain $g_{\pi BB^{\ast }}=10.3$ . In this paper, we use the value obtained from the former method.
The values of the couplings $g_{B_{c}BD^{\ast }}$ and $g_{B_{c}B^{\ast }D}$ are fixed by using $g_{\Upsilon BB}=13.3$, which is obtained using vector meson dominance (VMD) model in ref. [@Lin2001] and SU(5) symmetry result $g_{B_{c}BD^{\ast }}=g_{B_{c}B^{\ast }D}=\frac{2}{\sqrt{5}}g_{\Upsilon BB}$ [@Lodhi2007]. In this way we obtain $g_{B_{c}BD^{\ast
}}=g_{B_{c}B^{\ast }D}=11.9$.
There is no empirically fitted value available for the four-point coupling $g_{\pi B_{c}B^{\ast }D^{\ast }}$, thus we use SU(5) symmetry, which implies $g_{\pi B_{c}D^{\ast }B^{\ast }}=2g_{\pi DD^{\ast }}g_{B_{c}B^{\ast
}D}=2g_{\pi BB^{\ast }}g_{B_{c}BD^{\ast }}$. These two identities give two values of 105 and 295, whereas their mean values in 200. The values of coupling constants used in this paper and methods for obtaining them are summarized in Table \[table1\].
Coupling constant Value Method of Derivation
--------------------------------- ------------ ------------------------
$g_{\pi DD^*}$ 4.4 $D^*$ decay width
$g_{\pi BB^*}$ 12.4 Heavy quark symmetries
$g_{B_cBD^*}$ and $g_{B_cB^*D}$ 11.9 VMD, SU(5) symmetry
$g_{\pi B_cB^*D^*}$ 105 to 295 SU(5) symmetry
: Values of coupling constants used in this paper[]{data-label="table1"}
Results and Discussion
======================
Fig. 3 shows the $B_{c}$ absorption cross sections of the process $B_{c}^{+}\protect\pi \rightarrow DB$ as a function of total center of mass (c.m) energy $\sqrt{s}$. Solid and dashed curves in this figure represent cross sections without and with form factors. Form factors are included to account the finite size of interacting hadrons. We use following monopole form factor at three point vertices.$$f_{3}=\frac{\Lambda ^{2}}{\Lambda ^{2}+\overline{q}^{2}}$$
Where, $\Lambda $ is cutoff parameter and $\overline{q}^{2}$ is squared three momentum transfer in c.m frame. At four point vertex, we use the following form factor.$$f_{4}=\left( \frac{\Lambda ^{2}}{\Lambda ^{2}+\overline{q}^{2}}\right) ^{2}$$
Where, $\overline{q}^{2}=\frac{1}{2}\left[ (\overline{p}_{1}-\overline{p}_{3})^{2}+(\overline{p}_{1}-\overline{p}_{4})^{2}\right] _{c.m}$
![$B_{c}$ absorption cross sections for the process $B_{c}^{+}\protect\pi \rightarrow DB$. Solid and dashed curves represent cross sections without and with form factor respectively. Lower and upper dashed curves are with cutoff parameter $\Lambda =1$ and $2$ GeV respectively. Threshold energy is 7.15 GeV.[]{data-label="fig3"}](fig3.eps){width="70.00000%"}
In general, the value of cutoff parameter used in the form factor could have different values at different vertices. There is no direct way to calculate the values of these parameters. In some cases cutoff parameters can be fixed empirically by studying hadronic scattering data in meson or baryon exchange models. Such empirical fits put the cutoff parameters on the scale of 1 to 2 GeV for the vertices connecting light hadrons ($\pi$, $K$, $\rho$, $N$ etc) [@machleid1987]. However, due to limited information about the scattering data of charmed and bottom hadrons, no empirical values of the related cutoff parameters are known. In this case we can estimate cutoff parameters by relating them with inverse (rms) size of hadrons. Cutoff parameter for meson-meson vertex is determined by the ratio of size of nucleon to pseudoscalar meson in ref. [@yasui2009].
$$\Lambda _{D}=\frac{r_{N}}{r_{D}}\Lambda _{N},\text{ \ \ \ \ \ \ \ }\Lambda
_{B}=\frac{r_{N}}{r_{B}}\Lambda _{N}$$
The values of the ratios $r_{N}/r_{D}=1.35$ and $r_{N}/r_{B}=1.29$ are determined by the quark potential model for $D$ and $B$ mesons respectively [@yasui2009]. Cutoff parameter $\Lambda _{N}$ for nucleon-meson vertex can be determined from empirical data of nucleon-nucleon system. In ref. [@yasui2009] $\Lambda _{N}=0.94$ GeV, is fixed from the empirical value of the binding energy of deuterium. Where as, nucleon-nucleon scattering data gives $\Lambda _{\pi NN}=1.3$ GeV and $\Lambda _{\rho NN}=1.4$ GeV [@holinde1989]. A variation of 0.9 to 1.4 GeV in $\Lambda _{N}$ produces variation of 1.2 to 1.8 GeV in $\Lambda _{D}$ and $\Lambda _{B}$. Based on these results we take all the cutoff parameters same for simplicity and vary them on the scale 1 to 2 GeV to study the uncertainties in cross sections due to cutoff parameter.
Fig. 3 shows that for $B_{c}^{+}\pi \rightarrow DB$ process the cross section roughly varies from 2 to 7 mb, when the cutoff parameter is between 1 to 2 GeV. Suppression due to form factor at cutoff $\Lambda =1$ and 2 GeV is roughly by factor 11 and 3 respectively.
![$B_{c}$ absorption cross sections of the process $B_{c}^{+}\protect\pi \rightarrow D^{\ast }B^{\ast }$ for three different values of four-point coupling, $g_{\protect\pi B_{c}B^{\ast }D^{\ast }}=105, 200, 295$ for dotted, solid and dashed curve respectively (a) without and (b) with form factor. Cutoff parameter is taken 1.5 GeV. []{data-label="fig4"}](fig4a.eps "fig:"){width="45.00000%"} \[fig4a\] ![$B_{c}$ absorption cross sections of the process $B_{c}^{+}\protect\pi \rightarrow D^{\ast }B^{\ast }$ for three different values of four-point coupling, $g_{\protect\pi B_{c}B^{\ast }D^{\ast }}=105, 200, 295$ for dotted, solid and dashed curve respectively (a) without and (b) with form factor. Cutoff parameter is taken 1.5 GeV. []{data-label="fig4"}](fig4b.eps "fig:"){width="45.00000%"} \[fig4b\]
$B_{c}$ absorption cross section of the process $B_{c}^{+}\pi \rightarrow D^{\ast }B^{\ast }$ depends upon the four point contact coupling $g_{\pi B_{c}B^{\ast }D^{\ast }}$, whose values is fixed through SU(5) symmetry. It is noted in the previous section that although SU(5) symmetry uniquely fix it, but difference in the values of the couplings $g_{\pi DD^*}$ and $g_{\pi BB^*}$ produces two values 105 and 295 of the four point contact coupling. In this paper, we treat this variation as uncertainty in the coupling and study its effect on the cross section of the process. Fig. \[fig4\]a, shows how the value of the four point coupling could affect the values of $B_{c}$ absorption cross sections through the process $B_{c}^{+}\pi \rightarrow D^{\ast }B^{\ast }$ without form factor. Both of the cross sections increase very rapidly for the values 105 and 295, which are not realistic. However, if we use the value of 200, the average to two extreme values the variation in the cross section, denoted by solid line is some what a compromise. Fig. \[fig4\]b, shows the effect of uncertainty in the four point contact coupling, on the cross section with form factor. This figure indicates that the value of the contact coupling significantly affects the cross section only near the threshold energy (7.34 GeV). It will be discussed later that this effect is further marginalized in the total absorption cross section.
![$B_{c}$ absorption cross sections for the process $B_{c}^{+}\protect\pi \rightarrow D^{\ast }B^{\ast }$. Solid and dashed curves represents cross sections without and with form factor respectively. Lower and upper dashed curves are with cuttoff parameter $\Lambda =1$ and $2$ GeV respectively and $g_{\pi B_{c}B^{\ast }D^{\ast }}=200$. Threshold energy is 7.34 GeV[]{data-label="fig5"}](fig5.eps){width="70.00000%"}
Fig. 5 shows the $B_{c}$ absorption cross sections of the process $B_{c}^{+}\pi \rightarrow D^{\ast }B^{\ast }$ as a function of total center of mass (c.m) energy $\sqrt{s}$. The cross section of the process roughly varies from 0.2 to 2 mb, when the cutoff parameter is between 1 to 2 GeV and $g_{\pi B_{c}B^{\ast }D^{\ast }}=200$. Suppression due to form factor at cutoff $\Lambda =1$ and 2 GeV is roughly by factor 45 and 7 respectively. Relatively high suppression in this process is mainly due to large values of mass of final particles $D^{\ast }$ and $B^{\ast }$. It is noted that these estimates of cross sections are highly dependent on the choice of form factor and the value of cutoff, as well as on the values of coupling constants. However, it is observed that the effect of uncertainty in the four point contact coupling $g_{\pi B_{c}B^{\ast }D^{\ast }}$ is marginal on the total cross section due to relatively small value of the cross section of the second process. This is shown in the Fig. 6, in which total absorption cross section for $B_{c}+\pi$ is plotted for three different values of $g_{\pi B_{c}B^{\ast }D^{\ast }}=105, 200, 295$.
![Total $B_{c}$ absorption cross sections by pion for three different values of four-point coupling, $g_{\protect\pi B_{c}B^{\ast }D^{\ast }}=105, 200, 295$ for dotted, solid and dashed curve respectively. Cutoff parameter is taken 1.5 GeV.[]{data-label="fig6"}](fig6.eps){width="70.00000%"}
Concluding Remarks
==================
In this paper, we have calculated $B_{c}$ absorption cross section by $\pi $ mesons using hadronic Lagrangian based on SU(5) flavor symmetry. This approach has already been used for calculating absorption cross sections of $J/\psi $ and $\Upsilon $ mesons by hadrons. In our study, all the coupling constants are preferably determined empirically using vector meson dominance model, heavy quark symmetries or QCD sum rules instead of using SU(5) symmetry. The hadronic Lagrangian based on SU(5) flavor symmetry is developed by imposing the gauge symmetry, but this symmetry is broken when the mass terms are added in the Lagrangian. Thus SU(5) gauge symmetry exists only in limit of zero hadronic masses. Broken SU(5) symmetry does not necessarily implies that the coupling constants of three or four-point vertices should be related through SU(5) universal coupling constant. It is, therefore, justified to empirically fix the couplings. It can also be seen that the empirical values of the couplings also violate SU(5) symmetry relations given in Eqs. 8 and 9. It is also noted that four-point coupling constant $g_{\pi B_{c}B^{\ast }D^{\ast }}$ cannot be fixed empirically. Thus in this case we have no choice except to make a reasonable estimate using SU(5) symmetry as discussed above. Calculated cross sections are found to be in range 2 to 7 mb and 0.2 to 2 mb for the processes $B_{c}^{+}\pi
\rightarrow DB$ and $B_{c}^{+}\pi \rightarrow D^{\ast }B^{\ast }$ respectively, when the form factor is included. These results could be useful in calculating production rate of $B_{c}$ meson in relativistic heavy ion collisions.
[99]{} T. Matsui & H. Satz, Phys. Lett. B **178,** 416 (1986)
M. C. Areu et. Al., NA50 Collaboration, Phys. Lett. B **450**, 456 (1999).
W. Cassing and C. M. Ko, Phys. Lett. B **396**, 39 (1996); W. Cassing and E. L. Bratkovskaya, Nucl. Phys. **A623**, 570 (1997).
N. Armesto and A. Capella, Phys. Lett. B **430**, 23 (1998).
D. E. Kahana and S. H. Kahana, Phys. Rev. C **59**, 1651 (1999).
C. Gale, S. Jeon and J. Kapusta, Phys. Lett. B**459**, 455 (1999).
C. Spieles, R. Vogt, L. Gerland, S. A. Bass, M. Bleicher, H. Stocker and W. Greiner, Phys. Rev. C **60**, 054901 (1999).
Ben-Hao Sa, An Tai, Hui Wang and Geng-He Liu, Phys. Rev. C **59**, 2728 (1999).
D. Kharzeev and H. Satz, Phys. Lett. B** 334**, 155 (1994).
D. Kharzeev, H. Satz, A. Syamtomov, and G. Zinovjev Phys. Lett. B **389**, 595 (1996).
C. Y. Wong, E. S. Swanson and T. Barnes, Phys. Rev. C **62**, 045201 (2000); M. A. Ivanov, J. G. Korner, and P. Santorelli, Phys. Rev. D **70**, 014005 (2004).
Z. Lin and C. M. Ko, Phys. Lett. B **503**, 104 (2001).
K. Haglin, Phys. Rev. C **61**, 031902 (2000).
Z. Lin and C. M. Ko, Phys. Rev. C **62**, 034903 (2000).
W. Liu, C. M. Ko, and Z. W. Lin, Phys. Rev. C **65**, 015203 (2001).
The annual Quark Matter conference 2011, reported in CERN Bulletin Nos **21-22**, (2011).
R. Vogt, Phys. Rept. **310**, 197 (1997).
Martin Schroedter, Robert L. Thews and Johann Rafelski, Phys. Rev. C** 62**, 024905 (2000).
J. Letessier and J. Refelski, *Hadrons and Quark-Gluon Plasma* (CUP, 2002).
M. A. K. Lodhi and R. Marshall, Nucl. Phys. A **790**, 323c-327c (2007).
W. M. Yao, et. al., Journal of Physics G **33,** 1 (2006).
P. Colangelo, F. De Fazio, and G. Nardulli, Phys. Lett. B** 334**, 175 (1994).
V. M. Belyaev, V. M. Braun, A. Khodjamirian and R. Ruckl, Phys. Rev. D** 51**, 6177 (1995).
M. B. Wise, Phys. Rev. D **45**, 2188 (1992); G. Burdman and J. F. Donoghue, Phys. Lett. B** 280**, 287 (1992).
B. Grinstein and P. F. Mende, Phys. Lett. B** 299**, 127 (1993); E. de Rafael and J. Taron, Phys. Rev. D **50**, 373 (1994).
R. Machleid, K. Holinde and C. Elster, Phys. Rev. **149** , 1 (1987); R. Machleid, Adv. Nucl. Phys. **19**, 189 (1989); D. Lohse, J. W. Durso, K. Holinde, and J. Speth, Nucl. Phys. **A516**, 513 (1990).
S. Yasui and K. Sudoh, Phys. Rev. D **80**, 034008 (2009).
B. Holzenkamp, K. Holinde and J. Speth, Nucl. Phys. **A500**, 485 (1989); G. Janssen, J. W. Durso, K. Holinde, B. C. Pearce, and J. Speth, Phys. Rev. Lett. **71**, 1978 (1993).
[^1]: [email protected]
[^2]: [email protected] (corresponding author)
[^3]: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We discuss a possibility to measure the lifetime of charged Wino in supersymmetric model at future $100\ {\rm TeV}$ $pp$ colliders, assuming that (neutral) Wino is the lightest superparticle (LSP). In the Wino LSP scenario, the charged Wino has a lifetime of about $0.2\ {\rm ns}$, and its track may be reconstructed in particular by the inner pixel detectors. We show that the lifetime of charged Wino may be measured by using the information about the distribution of the flight lengths of charged Winos. We propose a procedure for the lifetime determination and show how the accuracy changes as we vary the mass spectrum of superparticle. We also discuss the effects of the detector layouts on the lifetime determination.'
---
UT-19-29\
December, 2019\
.75in
****
Determining Wino Lifetime in Supersymmetric Model\
at Future 100 TeV $pp$ Colliders\
.75in
[So Chigusa${}^{a}$, Yusuke Hosomi${}^{a}$, Takeo Moroi${}^{a}$ and Masahiko Saito${}^{b}$]{}
0.25in
[*${}^{a}$Department of Physics, University of Tokyo, Tokyo 113-0033, Japan\
${}^{b}$International Center for Elementary Particle Physics, University of Tokyo,\
Tokyo 113-0033, Japan*]{}
.5in
Introduction {#sec:intro}
============
Low energy supersymmetry (SUSY), with superparticles at the mass scale much lower than the Planck scale, has been attracted attentions. Even though no direct evidence of superparticles has been experimentally found yet, it is still a well-motivated candidate of physics beyond the standard model. In particular, in models with low energy supersymmetry, gauge coupling unification at the scale of grand unified theory (GUT), i.e., $\sim 10^{16}\ {\rm GeV}$, is possible. In addition, more importantly for our study, the lightest superparticle (LSP) in SUSY model with $R$-parity conservation can be dark matter. The SUSY dark matter is an important target not only of direct detection experiments but also of high energy colliders. Notably, the collider phenomenology of SUSY dark matter depends on the properties of the LSP.
In the present study, we concentrate on SUSY model in which Winos, which are superpartners of $SU(2)_L$ gauge bosons, are lighter than other superparticles and discuss its collider phenomenology. This class of model is motivated by the present constraints on low energy SUSY. First, the neutral Wino can be a viable candidate of dark matter. In addition, the Wino LSP naturally shows up from so-called minimal gravity mediation model [@Ibe:2006de; @Ibe:2011aa; @ArkaniHamed:2012gw] based on anomaly mediation [@Randall:1998uk; @Giudice:1998xp]. In such a model, masses of gauginos are of the order of $(0.1-1)\ {\rm TeV}$, while those of other superparticles are a few orders of magnitude heavier. Notably, such a mass spectrum is well motivated from the Higgs mass point of view because heavier superparticles, in particular heavy stops, can push up the Higgs mass to the observed value of about $125\ {\rm GeV}$ via radiative corrections [@Okada:1990vk; @Haber:1990aw; @Ellis:1990nz; @Ellis:1991zd].
In the SUSY model of our interest, the primary targets of the collider study are Winos as well as other gauginos. The thermal relic abundance of Wino becomes equal to the dark matter density if its mass is about $2.9\ {\rm TeV}$ [@Hisano:2006nn], while lighter Wino can also become dark matter if Winos are non-thermally produced in the early universe [@Giudice:1998xp; @Moroi:1999zb]. Combined with the present collider bounds on superparticles, it may be the case that the Winos (and other superparticles) are out of the kinematic reach of the LHC experiment. Such a possibility motivates us to consider more energetic colliders than the LHC. In particular, $pp$ colliders with the center of mass energy of $\sim 100\ {\rm TeV}$, called future circular collider (or FCC-hh), is now seriously discussed.
Here, we consider the collider phenomenology of supersymmetric model with Wino LSP at FCC-hh. In the previous studies, it has been discussed that the discovery [@Saito:2019rtg] and the mass measurements [@Asai:2019wst] of gauginos are possible at FCC-hh, particularly relying on the existence of disappearing tracks of charged Winos.[We may also use the study of mono-jet events [@Han:2018wus] and precision study of the Drell-yan processes [@Chigusa:2018vxz; @DiLuzio:2018jwd; @Matsumoto:2018ioi; @Abe:2019egv] for the discovery of the signals of electroweakly interacting particles, like Winos, at the FCC-hh.]{} In the case of Wino LSP, it is often the case that the mass difference between charged and neutral Winos is induced dominantly by the radiative correction due to the electroweak gauge bosons; the charged Wino becomes slightly heavier than the neutral one and the mass difference is given by $\sim 160\ {\rm MeV}$. As a result, the charged Wino becomes fairly long-lived; its lifetime is often $\sim
0.2\ {\rm ns}$ and is insensitive to the mass spectrum of superparticles (as far as superparticles other than gauginos are much heavier than Wino). Then, once produced at the colliders, the charged Wino may fly $O(1-10)\ {\rm cm}$ and may be identified as a short high $p_T$ track. In order to understand the properties of Winos, measurement of the lifetime of charged Wino is an important step.
In this letter, we study the possibility of measuring the lifetime of charged Wino at FCC-hh. If the Wino mass is less than $\sim 2.9\ {\rm
TeV}$, which is the upper bound on the Wino mass from the point of view of dark matter, it is expected that the charged Wino is within the reach of FCC-hh [@Saito:2019rtg] and that the Wino mass can be determined [@Asai:2019wst]. Here, using SUSY events with charged Wino production, we show that the lifetime of the charged Wino can be also determined. We discuss the basic procedure to determine the charged Wino lifetime at FCC-hh, and show the expected accuracy of the lifetime measurement. The organization of this letter is as follows. In Section \[sec:formalism\], we explain the method of the lifetime measurements at the FCC-hh. We also summarize important features of the model of our interest. Then, in Section \[sec:mc\], we show our numerical results. Section \[sec:conclusion\] is devoted for conclusions and discussion.
Formalism {#sec:formalism}
=========
Let us explain the setup of our analysis. Here, we concentrate on supersymmetric models in which Winos are lighter than other supersymmetric particles. Detailed mass spectrum of superparticles for our Monte Carlo (MC) analysis will be explained in the next section. We also assume that the mass difference between the charged and neutral Winos dominantly comes from radiative effects due to electroweak gauge bosons. In such models, neutral Wino becomes slightly lighter than charged ones, and hence neutral Wino becomes the LSP while charged Wino becomes long-lived. The mass difference is predicted to be $\sim 160\ {\rm MeV}$ which gives the lifetime of the charged Wino of $\sim 0.2\ {\rm ns}$ [@Cheng:1998hc; @Feng:1999fu; @Gherghetta:1999sw; @Yamada:2009ve; @Ibe:2012sx]. Here, we take the canonical lifetime of charged Wino to be $c\tau=5.75\ {\rm cm}$ [@Ibe:2012sx], with $c$ being the speed of light. In the following, we study how well we can determine the lifetime of charged Wino at the FCC-hh experiment. Although our primary interest is to determine the Wino lifetime, we vary the input value of the lifetime to see how the sensitivity depends on it.
Once produced, charged Wino travels finite distance and decays into neutral Wino (and charged pion). In particular, some of charged Winos travel long enough to go through several layers of inner pixel detectors and to be reconstructed as (disappearing) tracks. Such disappearing tracks can be used not only for the reduction of standard model backgrounds but also for the determination of the lifetime of charged Wino.
Expecting that there exist several layers of pixel detectors, let $\tilde{W}^\pm_i$ ($i=1$ $-$ $n_A$) be charged Winos which arrive $A$-th layer of the pixel detector before decaying. Here, $n_A$ is the number of charged Wino samples available for the lifetime measurement. Then, the expectation value of the number of charged Winos which arrive $B$-th layer ($B>A$) is $$\begin{aligned}
\langle n_B \rangle (\tau) = \sum_{i=1}^{n_A} p_i (\tau),\end{aligned}$$ with $$\begin{aligned}
p_i (\tau) \equiv
e^{-(L_T^{(B)}-L_T^{(A)})/\tau \beta_i \gamma_i \sin\theta_i},
\label{p_i}\end{aligned}$$ where $L_T^{(A)}$ and $L_T^{(B)}$ are transverse distance from the interaction point to the $A$- and $B$-th layers, respectively, $\beta_i$ is the velocity of $i$-th Wino, $\gamma_i\equiv
1/\sqrt{1-\beta_i^2}$, and $\theta_i$ is the angle between the proton beam and the direction of the momentum of $\tilde{W}^\pm_i$. Notice that $\langle n_B \rangle$ is an increasing function of $\tau$ and is sensitive to the lifetime of charged Wino. Thus, with the measurements of the numbers of charged Winos arriving at $A$- and $B$-th layers, we may acquire information about $\tau$, assuming that the velocity and the propagation direction of charged Winos are measurable. In particular, once $n_B$ (i.e., the number of charged Winos reaching to the $B$-th layer) is measured, the best-fit value of the lifetime is given by solving $$\begin{aligned}
\langle n_B \rangle (\tau^{\rm (best)}) = n_B.\end{aligned}$$
The propagations of charged Winos from the $A$-th layer to the $B$-th are multiples of Bernoulli processes with various probabilities. Assuming a test lifetime $\tau^{\rm (test)}$, the probability to realize a specific value of $n_B$ for a given data set is expressed as [@Chen:1997] $$\begin{aligned}
P (n_B;\tau^{\rm (test)}) &=
\left[ \prod_{i=1}^{n_A} (1-p_i^{\rm (test)}) \right]
\sum_{i_1<i_2<\cdots<i_{n_B}}
\left[
\frac{p_{i_1}^{\rm (test)}}{1-p_{i_1}^{\rm (test)}}
\frac{p_{i_2}^{\rm (test)}}{1-p_{i_2}^{\rm (test)}} \cdots
\frac{p_{i_{n_B}}^{\rm (test)}}{1-p_{i_{n_B}}^{\rm (test)}}
\right], \notag \\
&= \frac{1}{n_A+1}
\sum_{l=0}^{n_A}
e^{-2\pi i n_A l/(n_A+1)}
\prod_{k=1}^{n_A}
\left[ 1+(1-e^{2\pi i l/(n_A+1)} p_k^{\rm (test)}) \right].
\label{probdist}\end{aligned}$$ where $p_{i}^{\rm (test)}\equiv p_i(\tau^{\rm (test)})$ and the sum is taken for all the possible sets of $\{i_1,i_2,\cdots,i_{n_B}\}$. The second equality follows from [@Fernandez:2010], which reduces the cost of numerical calculation.
Once charged Winos are observed at future collider experiments with the measurements of their velocities and directions (as well as $n_A$ and $n_B$), we may constrain the lifetime of charged Wino. In our analysis, we define $\alpha\ \%$ “confidence interval,” which we denote $\{n_B\}_\alpha$, as follows. We define integers $I_1$, $\cdots$, $I_{n_A}$ such that $P (I_1; \tau^{\rm
(test)})\geq\cdots\geq P (I_{n_A}; \tau^{\rm (test)})$. Then, the confidence interval, $\{n_B\}_\alpha\equiv\{I_1,\cdots,I_N\}$, is defined as $$\begin{aligned}
\sum_{I\in\{n_B\}_\alpha} P (I; \tau^{\rm (test)})
- P (I_{N}; \tau^{\rm (test)})
<
\alpha\ \%
\leq
\sum_{I\in\{n_B\}_\alpha} P (I; \tau^{\rm (test)}).\end{aligned}$$ A test lifetime $\tau^{\rm (test)}$ is allowed (excluded) if observed $n_B$ is inside (outside) of the confidence interval calculated with $\tau^{\rm (test)}$. In the next section, we discuss how well we can determine the lifetime using MC analysis.
Monte Carlo Analysis {#sec:mc}
====================
In this section, we show that the determination of the lifetime of charged Wino is really possible using MC analysis. For simplicity, motivated by the minimal gravity mediation model based on anomaly mediation, we concentrate on the model in which gauginos (i.e., Bino, Wino, and gluino) are the only superparticles accessible with FCC-hh; other superparticles are assumed to be too heavy to be produced. In addition, Winos are assumed to be lighter than Bino and gluino. We adopt three Sample Points which are based on the minimal gravity mediation model [@Asai:2019wst]. The Sample Points are shown in Table \[table:samplept\]; on the table, the masses of Bino, Wino, and gluino (denoted as $m_{\tilde{B}}$, $m_{\tilde{W}}$, and $m_{\tilde{g}}$, respectively) as well as the gluino pair production cross section and the canonical luminosity in our analysis are given. In our study, we take $\mathrm{Br}(\tilde{g}\rightarrow\tilde{W}\bar{q}q)=\mathrm{Br}(\tilde{g}\rightarrow\tilde{B}\bar{q}q)=0.5$ (with $q$ being quarks), with the assumption of the flavor universality of the final-state quarks. For the cases of the gluino mass of $6$ and $7$ TeV, we assume the integrated luminosity of ${\cal
L}=10\ {\rm ab}^{-1}$, while ${\cal L}=30\ {\rm ab}^{-1}$ is used for the case of $m_{\tilde{g}}=8\ {\rm TeV}$ to compensate the smallness of the gluino production cross section.
Point 1 Point 2 Point 3
---------------------------------------------------- --------- --------- ---------
$m_{\tilde{B}}$ \[GeV\] $3660$ $4060$ $4470$
$m_{\tilde{W}}$ \[GeV\] $2900$ $2900$ $2900$
$m_{\tilde{g}}$ \[GeV\] $6000$ $7000$ $8000$
$\sigma (pp\rightarrow \tilde{g}\tilde{g})$ \[fb\] $7.9$ $2.7$ $1.0$
${\cal L}$ \[1/ab\] $10$ $10$ $30$
: Gaugino masses and the gluino pair production cross section (for the center-of-mass of $100\ {\rm TeV}$) for the Sample Points 1, 2, and 3. We also show the canonical luminosities used for the analysis for these Sample Points.[]{data-label="table:samplept"}
Our method of event generation is mostly the same as that adopted in [@Asai:2019wst]. We use [MadGraph5\_aMC@NLO]{} [@Alwall:2011uj; @Alwall:2014hca] for the generation of $pp\rightarrow\tilde{g}\tilde{g}$ and $pp\rightarrow\tilde{W}^{+}\tilde{W}^{-}+\mathrm{jets}$ events. The results are passed to [PYTHIA8]{} [@Sjostrand:2014zea] for the decay and hadronization processes. Then, [Delphes]{} ([v3.4.1]{}) [@deFavereau:2013fsa] is used for a fast detector simulation; we use the card [FCChh.tcl]{} included in the package. The velocities of charged Winos are smeared by our original code. We expect that, at FCC-hh, the charged Wino track can be reconstructed and that information about the time of flight is available if the charged Wino hits several layers of pixel detector. We assume $6\ \%$ error in the velocity measurement [@Asai:2019wst]; for reconstructed charged Wino tracks, the observed values of the Wino velocity are determined as follows: $$\begin{aligned}
\beta = (1 + 6\ \% \times Z) \beta^{\rm (true)},
\label{betaerror}\end{aligned}$$ where $\beta$ and $\beta^{\rm (true)}$ are observed and true values of the velocity and $Z$ is the $(0,1)$ Gaussian random variable. We neglect the error in the measurement of the directions of Wino tracks.
![Distributions of the true (blue) and observed (orange) velocities of charged Winos for the Sample Point 1. The height of the histogram is the number of charged Winos in the bin for events satisfying the Requirements 1 and 2. The bin width is $\Delta\beta=0.01$. []{data-label="fig:beta"}](betadist.pdf){width="65.00000%"}
In the following analysis, we assume that the charged Winos are required to hit at least inner four layers of the pixel detector for the track reconstruction (and also for background reduction). In order to eliminate standard model backgrounds, we use only the events satisfying the following requirements:
- There exist two “long enough” Wino-like tracks. The transverse length of the tracks should be longer than the transverse distance to the 4th pixel detector $L_{T}^{(4)}$. In addition, the pseudo-rapidities ($\eta$) of the tracks should satisfy $|\eta|<1.5$.
- The missing transverse energy (MET) should be larger than $1\ {\rm TeV}$.
With these requirements, we expect that the standard model backgrounds can become negligible, as discussed in [@Asai:2019wst].
In Fig. \[fig:beta\], we show the distribution of $\beta$ and $\beta^{\rm (true)}$ for the Sample Point 1, using the events satisfying Requirements 1 and 2. (Here, 100 sets of the event samples are used for the figure.) We can see that the peak of the distribution is shifted to the small value of the velocity after the smearing. We will discuss its effects of the lifetime determination later.
For the determination of $\tau$, we use charged Winos in the events satisfying the requirements given above, and hence we take $(A,B)=(4,5)$. In addition, for the lifetime determination, velocity information about charged Winos is necessary. In order for a good velocity measurement, we use only charged Winos whose (observed) velocity is smaller than $0.85$ for the lifetime determination.
![The dots show $P (n_5; \tau^{\rm (test)})$ for $L_{T}^{(4)}=10\ {\rm cm}$ and $L_{T}^{(5)}=15\ {\rm cm}$, using the event samples generated from the Sample Point 1 with $c\tau=5.75\ {\rm cm}$. $c\tau^{\rm (test)}$ is taken to be $3\ {\rm cm}$ (blue), $5.75\ {\rm cm}$ (red), and $10\ {\rm cm}$ (green) from left to right. The solid lines show the Gaussian distribution $N(\hat{n}_5(\tau^{\rm (test)}),\hat{n}_5(\tau^{\rm (test)}))$, where $\hat{n}_5(\tau^{\rm (test)})$ is the value of $n_5$ which maximizes $P (n_5; \tau^{\rm (test)})$.[]{data-label="fig:probdist"}](probdist.pdf){width="65.00000%"}
First, we consider the case where the gluino mass is light enough so that the gluino pair production dominates the SUSY events at FCC-hh (i.e., the case of the Sample Points 1 $-$ 3). In Fig. \[fig:probdist\], taking $L_{T}^{(4)}=10\ {\rm cm}$ and $L_{T}^{(5)}=15\ {\rm cm}$, we plot $P (n_5; \tau^{\rm (test)})$ taking $c\tau^{\rm (test)}=3\ {\rm cm}$ (blue), $c\tau^{\rm
(test)}=5.75\ {\rm cm}$ (red), and $c\tau^{\rm (test)}=10\ {\rm cm}$ (green), using the event sample generated from the Sample Point 1 with $c\tau=5.75\ {\rm cm}$. We can see that the behavior of $P (n_5;
\tau^{\rm (test)})$ is strongly dependent on $\tau^{\rm (test)}$. Thus, with the measurement of the number of charged Winos reaching to the 5th layer, we can obtain information about the lifetime. We also plot the Gaussian distribution $N(\hat{n}_5(\tau^{\rm
(test)}),\hat{n}_5(\tau^{\rm (test)}))$, where $\hat{n}_5(\tau^{\rm
(test)})$ is the value of $n_5$ which maximizes $P (n_5; \tau^{\rm
(test)})$. We can see that the probability distribution is well approximated by the Gaussian distribution when the number of charged Winos reaching to the 5th layer is large enough.
![Same as Fig. \[fig:pp2gogo6TeV\], except for the Sample Point 2.[]{data-label="fig:pp2gogo7TeV"}](pp2gogo6TeV.pdf){width="65.00000%"}
![Same as Fig. \[fig:pp2gogo6TeV\], except for the Sample Point 2.[]{data-label="fig:pp2gogo7TeV"}](pp2gogo7TeV.pdf){width="65.00000%"}
![Same as Fig. \[fig:pp2gogo6TeV\], except for the Sample Point 3.[]{data-label="fig:pp2gogo8TeV"}](pp2gogo8TeV.pdf){width="65.00000%"}
In Figs. \[fig:pp2gogo6TeV\] $-$ \[fig:pp2gogo8TeV\], we plot the best-fit value of the lifetime as well as the expected lower and upper bounds for various choices of $\tau$ for the Sample Points 1 $-$ 3. Here, we use 100 independent data sets for each Sample Point, and determine the lower and upper bounds on the lifetime for each data set using the probability distribution defined in Eq.. The expected lower and upper bounds shown in the figures are obtained by taking the median of those from 100 independent data sets. The regions with $\tau$ giving rise to $\langle n_4\rangle<10$, for which our method of the lifetime measurement becomes difficult, are shaded.
We can see that the best-fit values of the lifetime are systematically overestimated in the present analysis. This is mainly due to the error in the velocity measurement. Here, the Wino velocity is assumed to be measured with the $6\ \%$ accuracy (see Eq. ). As one can see in Fig. \[fig:beta\], the observed Wino velocity is likely to be smaller than the true value, resulting in the overestimation of the lifetime. We checked that the best-fit values become consistent with the input values if the true value of the Wino velocity is used in the analysis. We assume that such a systematic effect originating from the velocity measurement can be well understood in the actual experiment. Thus, we will not include the shift of the best-fit value in estimating the uncertainty of the lifetime determination.
![Accuracies of the lifetime determination for several choices of the detector layout for the Sample Point 1 with $c\tau=5.75\ {\rm cm}$. Here, we take $(L_{T}^{(4)},
L_{T}^{(5)})=(10\ {\rm cm}, 15\ {\rm cm})$, $(11\ {\rm cm},
15\ {\rm cm})$, $(15\ {\rm cm}, 20\ {\rm cm})$, and $(15\ {\rm cm},
27\ {\rm cm})$ from left to right.[]{data-label="fig:pp2gogo6TeVLayout"}](pp2gogo6TeV_Layout.pdf){width="65.00000%"}
Next, we show how the accuracy of the lifetime determination depends on the detector layouts. The accuracy depends on the distances to the pixel layers. In Fig. \[fig:pp2gogo6TeVLayout\], for the Sample Point 1 with taking $c\tau=5.75\ {\rm cm}$, we show the expected lower and upper bounds for several choices of the distances to the layers. As expected, the accuracy becomes worse as the distances to the 4th layer becomes longer; with larger $L_{T}^{(4)}$, the number of charged Winos available for the analysis becomes smaller.
We define the uncertainties in the lifetime determination as $$\begin{aligned}
\delta \tau^{(\pm)} \equiv | \tau^{(\pm)} - \tau^{\rm (best)} |,\end{aligned}$$ where $\tau^{(-)}$ and $\tau^{(+)}$ are lower and upper bounds on the lifetime for a given confidence interval, respectively. For the Sample Points $1$ $-$ $3$ with $c\tau=5.75\ {\rm cm}$, the values of $\tau^{(\pm)}$ for several choices of detector layouts are summarized in Table \[table:error\]. In the same Table, we show the median values of $n_4$ and $n_5$ for each detector layout. For some sample points, $n_4$ for $(L_{T}^{(4)}, L_{T}^{(5)})=(15\ {\rm cm}, 20\ {\rm
cm})$ and that for $(15\ {\rm cm}, 27\ {\rm cm})$ slightly differ; it is due to statistical fluctuations.
[c|cccccc]{}\
$(L_{T}^{(4)},L_{T}^{(5)})$ & $c\delta\tau^{(-)}$ ($68\, \%$) & $c\delta\tau^{(+)}$ ($68\, \%$) & $c\delta\tau^{(-)}$ ($95\, \%$) & $c\delta\tau^{(+)}$ ($95\, \%$) & $n_4$ & $n_5$\
$(10\ {\rm cm}, 15\ {\rm cm})$ & 0.40 & 0.45 & 0.76 & 0.93 & 400 & 180\
$(11\ {\rm cm}, 15\ {\rm cm})$ & 0.51 & 0.57 & 0.94 & 1.2 & 290 & 150\
$(15\ {\rm cm}, 20\ {\rm cm})$ & 0.87 & 1.1 & 1.6 & 2.4 & 85 & 40\
$(15\ {\rm cm}, 27\ {\rm cm})$ & 0.83 & 1.2 & 1.5 & 2.3 & 85 & 16\
[c|cccccc]{}\
\
$(L_{T}^{(4)},L_{T}^{(5)})$ & $c\delta\tau^{(-)}$ ($68\, \%$) & $c\delta\tau^{(+)}$ ($68\, \%$) & $c\delta\tau^{(-)}$ ($95\, \%$) & $c\delta\tau^{(+)}$ ($95\, \%$) & $n_4$ & $n_5$\
$(10\ {\rm cm}, 15\ {\rm cm})$ & 0.65 & 0.74 & 1.2 & 1.5 & 170 & 78\
$(11\ {\rm cm}, 15\ {\rm cm})$ & 0.80 & 0.90 & 1.4 & 2.0 & 120 & 66\
$(15\ {\rm cm}, 20\ {\rm cm})$ & 1.4 & 1.9 & 2.3 & 4.2 & 38 & 17\
$(15\ {\rm cm}, 27\ {\rm cm})$ & 1.2 & 1.7 & 2.1 & 3.5 & 40 & 7\
[c|cccccc]{}\
\
$(L_{T}^{(4)},L_{T}^{(5)})$ & $c\delta\tau^{(-)}$ ($68\, \%$) & $c\delta\tau^{(+)}$ ($68\, \%$) & $c\delta\tau^{(-)}$ ($95\, \%$) & $c\delta\tau^{(+)}$ ($95\, \%$) & $n_4$ & $n_5$\
$(10\ {\rm cm}, 15\ {\rm cm})$ & 0.56 & 0.64 & 1.0 & 1.3 & 230 & 100\
$(11\ {\rm cm}, 15\ {\rm cm})$ & 0.71 & 0.79 & 1.2 & 1.6 & 170 & 90\
$(15\ {\rm cm}, 20\ {\rm cm})$ & 1.1 & 1.4 & 2.0 & 3.2 & 53 & 25\
$(15\ {\rm cm}, 27\ {\rm cm})$ & 1.0 & 1.5 & 1.9 & 3.3 & 53 & 10\
[c|cccccc]{}\
\
$(L_{T}^{(4)},L_{T}^{(5)})$ & $c\delta\tau^{(-)}$ ($68\, \%$) & $c\delta\tau^{(+)}$ ($68\, \%$) & $c\delta\tau^{(-)}$ ($95\, \%$) & $c\delta\tau^{(+)}$ ($95\, \%$) & $n_4$ & $n_5$\
$(10\ {\rm cm}, 15\ {\rm cm})$ & 1.1 & 1.5 & 1.9 & 3.2 & 60 & 28\
$(11\ {\rm cm}, 15\ {\rm cm})$ & 1.3 & 1.8 & 2.2 & 4.0 & 45 & 24\
$(15\ {\rm cm}, 20\ {\rm cm})$ & 2.2 & 3.6 & 3.4 & 9.2 & 15 & 7\
$(15\ {\rm cm}, 27\ {\rm cm})$ & 1.8 & 3.8 & 3.2 & 8.1 & 15 & 3\
When gluino is too heavy to be produced, the gluino pair production process cannot be used for our analysis. Even in such a case, we may use the direct production of Winos for the lifetime determination. In particular, if the Wino mass is $\sim 2.9\ {\rm TeV}$, which is the value of the Wino mass relevant for the thermal Wino dark matter scenario, charged Wino can be within the discovery reach of the disappearing track search at the FCC-hh [@Saito:2019rtg]. This fact indicates that the lifetime determination is also possible. In order to see how well we can determine the lifetime, we consider the process $pp\rightarrow\tilde{W}^{+}\tilde{W}^{-}+\mathrm{jets}$. Here, the extra jets are required for the trigger selection (as well as for the kinematical cut of our choice). For the events, we impose the Requirements 1 and 2, which we mentioned before. Then, using the events satisfying the Requirements, we determine the best-fit value of the lifetime as well as the confidence interval. In Fig.\[fig:DYjLayout\], with adopting several detector layouts, we show the expected accuracy of the determination of the Wino lifetime, taking $m_{\tilde{W}}=2.9\ {\rm TeV}$, $c\tau=5.75\ {\rm cm}$, and ${\cal L}=30\ {\rm ab}^{-1}$. Here, we use independent 500 data sets to calculate the median values of best-fit lifetime as well as lower and upper bounds. However $n_5 = 0$ in 5 data sets of them, so we exclude these sets before taking the median. The uncertainties for our choices of detector layouts are also summarized in Table \[table:error\], taking $c\tau=5.75\ {\rm cm}$. One can see that the uncertainties are larger than the cases of the gluino pair production events. This is mainly due to the smallness of the cross section for the $pp\rightarrow\tilde{W}^{+}\tilde{W}^{-}+\mathrm{jets}$ process. Even so, the Wino lifetime can be determined with a relatively good accuracy, i.e., $\delta\tau^{(\pm)}/\tau\sim O(10)\ \%$, in particular if a compact pixel detector with $L_T^{(5)}\sim 15\ {\rm cm}$ is available. We also comment here that, for $L_{T}^{(4)}=15\ {\rm cm}$, the layout with $L_{T}^{(5)}=27\ {\rm cm}$ gives better accuracy than $L_{T}^{(5)}=20\ {\rm cm}$. This is due to the fact that the accuracy becomes worse when $L_{T}^{(A)}$ and $L_{T}^{(B)}$ take too close values.
![Accuracies of the lifetime determination for several choices of the detector layout for $m_{\tilde{W}}=2.9\ {\rm TeV}$ and $c\tau=5.75\ {\rm cm}$, using the process $pp\rightarrow\tilde{W}^{+}\tilde{W}^{-}+\mathrm{jets}$. Here, we take $(L_{T}^{(4)}, L_{T}^{(5)})=(10\ {\rm cm}, 15\ {\rm
cm})$, $(11\ {\rm cm}, 15\ {\rm cm})$, $(15\ {\rm cm},
20\ {\rm cm})$, and $(15\ {\rm cm}, 27\ {\rm cm})$ from left to right.[]{data-label="fig:DYjLayout"}](pp2xpxm_Layout.pdf){width="65.00000%"}
Conclusions and Discussion {#sec:conclusion}
==========================
In this letter, we have discussed the possibility to determine the lifetime of charged Wino in supersymmetric model, assuming that the neutral Wino is the LSP. In such a case, the lifetime of the charged Wino is given by $c\tau\simeq 5.75\ {\rm cm}$, for which we have seen that a significant number of charged Winos may hit several layers of inner pixel detector and may be used for the lifetime determination. Concentrating on the case with the Wino mass of $2.9\ {\rm TeV}$, which is the relevant value to make thermal relic Wino as dark matter, we have studied the prospect of the Wino lifetime determination at FCC-hh.
If gluino is within the kinematical reach of FCC-hh, we may use the charged Winos produced by the decay of gluino. In such a case, the Wino lifetime may be determined with the accuracy of $14\ \%$ ($30\ \%$) for the $68\ \%$ ($95\ \%$) confidence interval. Even if gluino is out of the kinematical reach, we may use the charged Winos produced by the process $pp\rightarrow\tilde{W}^{+}\tilde{W}^{-}+\mathrm{jets}$. In such a case, the accuracy of the lifetime determination becomes worse, but still it can be $43\ \%$ ($92\ \%$) for the $68\ \%$ ($95\ \%$) confidence interval. These measurements of the Wino lifetime provides an important confirmation that the observed charged particle is really $\tilde{W}^\pm$.
Finally, we comment on the effects of the accidental alignments of the hits on the pixel detector, which has been neglected in our analysis. Potentially, the most serious effect may come from the hits on the 5th layer near the trajectories of the true charged Wino tracks. If there exist such hits for charged Winos which decay before reaching 5th layer, they affect the measurement of $n_5$. According to the study of the fake tracks given in [@Saito:2019rtg], however, the probability to have fake charged Wino tracks decreases by a factor of $O(100)$ with requiring a hit on an extra layer. Thus, we estimate that the error of $n_5$ due to the accidental alignment is less than $O(10^{-2})\times n_4$, which is negligible in the situation of our study.
Acknowledgments {#acknowledgments .unnumbered}
===============
This work was supported by MEXT KAKENHI Grant number JP16K21730 and JSPS KAKENHI Grant (Nos. 17J00813 \[SC\], 16H06490 \[TM\], 18K03608 \[TM\], and 18J11405 \[MS\]).
[99]{}
M. Ibe, T. Moroi and T. T. Yanagida, Phys. Lett. B [**644**]{} (2007) 355 \[hep-ph/0610277\]. M. Ibe and T. T. Yanagida, Phys. Lett. B [**709**]{} (2012) 374 \[arXiv:1112.2462 \[hep-ph\]\]. N. Arkani-Hamed, A. Gupta, D. E. Kaplan, N. Weiner and T. Zorawski, arXiv:1212.6971 \[hep-ph\]. L. Randall and R. Sundrum, Nucl. Phys. B [**557**]{} (1999) 79 \[hep-th/9810155\]. G. F. Giudice, M. A. Luty, H. Murayama and R. Rattazzi, JHEP [**9812**]{} (1998) 027 \[hep-ph/9810442\].
Y. Okada, M. Yamaguchi and T. Yanagida, Prog. Theor. Phys. [**85**]{} (1991) 1. H. E. Haber and R. Hempfling, Phys. Rev. Lett. [**66**]{} (1991) 1815. J. R. Ellis, G. Ridolfi and F. Zwirner, Phys. Lett. B [**257**]{} (1991) 83. J. R. Ellis, G. Ridolfi and F. Zwirner, Phys. Lett. B [**262**]{} (1991) 477. J. Hisano, S. Matsumoto, M. Nagai, O. Saito and M. Senami, Phys. Lett. B [**646**]{} (2007) 34 \[hep-ph/0610249\]. T. Moroi and L. Randall, Nucl. Phys. B [**570**]{} (2000) 455 \[hep-ph/9906527\]. M. Saito, R. Sawada, K. Terashi and S. Asai, Eur. Phys. J. C [**79**]{} (2019) no.6, 469 \[arXiv:1901.02987 \[hep-ph\]\]. S. Asai [*et al.*]{}, JHEP [**1905**]{} (2019) 179 \[arXiv:1901.10389 \[hep-ph\]\].
T. Han, S. Mukhopadhyay and X. Wang, Phys. Rev. D [**98**]{} (2018) no.3, 035026 \[arXiv:1805.00015 \[hep-ph\]\]. S. Chigusa, Y. Ema and T. Moroi, Phys. Lett. B [**789**]{} (2019) 106 \[arXiv:1810.07349 \[hep-ph\]\]. L. Di Luzio, R. Gröber and G. Panico, JHEP [**1901**]{} (2019) 011 \[arXiv:1810.10993 \[hep-ph\]\]. S. Matsumoto, S. Shirai and M. Takeuchi, JHEP [**1903**]{} (2019) 076 \[arXiv:1810.12234 \[hep-ph\]\]. T. Abe, S. Chigusa, Y. Ema and T. Moroi, Phys. Rev. D [**100**]{} (2019) no.5, 055018 \[arXiv:1904.11162 \[hep-ph\]\]. H. C. Cheng, B. A. Dobrescu and K. T. Matchev, Nucl. Phys. B [**543**]{} (1999) 47 \[hep-ph/9811316\]. J. L. Feng, T. Moroi, L. Randall, M. Strassler and S. f. Su, Phys. Rev. Lett. [**83**]{} (1999) 1731 \[hep-ph/9904250\]. T. Gherghetta, G. F. Giudice and J. D. Wells, Nucl. Phys. B [**559**]{} (1999) 27 \[hep-ph/9904378\]. Y. Yamada, Phys. Lett. B [**682**]{} (2010) 435 \[arXiv:0906.5207 \[hep-ph\]\]. M. Ibe, S. Matsumoto and R. Sato, Phys. Lett. B [**721**]{} (2013) 252 \[arXiv:1212.5989 \[hep-ph\]\]. S. X. Chen and J. S. Liu, Statistica Sinica [**7**]{} (1997) 875.
M. Fernandez and S. Williams, IEEE Transactions on Aerospace Electronic Systems [**46**]{} (2010) 803.
J. Alwall, M. Herquet, F. Maltoni, O. Mattelaer and T. Stelzer, JHEP [**1106**]{} (2011) 128 \[arXiv:1106.0522 \[hep-ph\]\]. J. Alwall [*et al.*]{}, JHEP [**1407**]{} (2014) 079 \[arXiv:1405.0301 \[hep-ph\]\]. T. Sjostrand [*et al.*]{}, Comput. Phys. Commun. [**191**]{} (2015) 159 \[arXiv:1410.3012 \[hep-ph\]\]. J. de Favereau [*et al.*]{} \[DELPHES 3 Collaboration\], JHEP [**1402**]{} (2014) 057 \[arXiv:1307.6346 \[hep-ex\]\].
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We study strong-coupling lattice QCD with staggered-Wilson fermions, with emphasis on discrete symmetries and possibility of their spontaneous breaking. We perform hopping parameter expansion and effective potential analyses in the strong-coupling limit. From gap equations we find nonzero pion condensate in some range of a mass parameter, which indicates existence of the parity-broken phase in lattice QCD with staggered-Wilson fermions. We also find massless pions and PCAC relations around second-order phase boundary. These results suggest that we can take a chiral limit by tuning a mass parameter in lattice QCD with staggered-Wilson fermions as with the Wilson fermion.'
author:
- Tatsuhiro Misumi
- 'Takashi Z. Nakano'
- Taro Kimura
- Akira Ohnishi
title: |
Strong-coupling Analysis of Parity Phase Structure\
in Staggered-Wilson Fermions
---
Introduction {#sec:Intro}
============
Since the dawn of lattice field theory [@Wil], the doubling problem of fermions has been a notorious obstacle to lattice simulations. Among several prescriptions for this problem, the Wilson fermion simply bypasses the no-go theorem [@NN] by introducing a species-splitting mass term into the naive lattice fermion. This Wilson term is regarded as one example of “flavored-mass terms" which split original 16 fermion species into plural branches [@CKM1; @CKM12]. It has been recently shown that the flavored-mass terms can also be constructed for staggered fermions [@KS; @Suss; @Sha] in Ref. [@Adams1; @Adams2; @Hoel]. The original purpose of introducing these terms was establishment of the index theorem with staggered fermions [@Adams1]. A bonus here is that staggered fermions with the flavored-mass terms can be applied to lattice QCD simulations as Wilson fermion and an overlap kernel. One possible advantage of these novel formulations, called staggered-Wilson and staggered-overlap, is reduction of matrix sizes in the Dirac operators, which would lead to reduction of numerical costs in lattice simulations. The numerical advantage in the staggered-overlap fermion have been shown in [@PdF]. Now further study is required toward lattice QCD with these lattice fermions.
The purpose of this work is to reveal properties of staggered-Wilson fermions in terms of the parity-phase structure (Aoki phase) [@AokiP; @AokiU1; @CreutzW; @SS; @ACV; @Sharpe]. As is well-known, the existence of the Aoki phase and the second-order phase boundary in Wilson-type fermions enables us to perform lattice QCD simulations by taking a chiral limit since the critical behavior near the phase boundary reproduces massless pions and the PCAC relation. Besides, understanding the phase structure also gives practical information for the application of the overlap [@GW; @Neu] and domain-wall [@Kap; @FuSh] versions, both built on the Wilson-type kernel. Thus, in order to judge applicability of these new lattice fermions, it is essential to investigate the Aoki phase in the staggered-Wilson fermions. The phase structure for the staggered-Wilson fermion was first studied by using the Gross-Neveu model in Ref. [@CKM2; @MCKNO] and the present paper shows further investigation of this topic.
In this paper, we investigate strong-coupling lattice QCD [@KaS] with emphasis on parity-phase structure for two types of staggered-Wilson fermions [@Adams2; @Hoel]. Firstly we discuss discrete symmetries of staggered-Wilson fermions, and show that physical parity and charge conjugation can be defined in both cases while hypercubic symmetry depends on types of staggered-Wilson fermions. Secondly, we perform hopping-parameter expansion and effective potential analysis for meson fields in the strong-coupling limit. For this purpose, we develop a method to derive the effective potential for lattice fermions with multiple-hopping terms. The gap equations show that pion condensate becomes non-zero in some range of a mass parameter, which indicates that parity-broken phase appears in this range. We also study meson masses around the second-order phase boundary, and find that massless pions and PCAC relations are reproduced. Lastly, we discuss parity-flavor symmetry breaking for 2-flavor cases. These results suggest that we can take a chiral limit by tuning a mass parameter in lattice QCD with staggered-Wilson fermions as with the Wilson fermion.
This paper is organized as follows. In Sec. \[sec:SWF\], we review staggered flavored-mass terms and two types of staggered-Wilson fermions. In Sec. \[sec:Sym\], we study discrete symmetries of staggered-Wilson fermions. In Sec. \[sec:HPE\], we study hopping parameter expansion in lattice QCD with these fermions. In Sec. \[sec:EPA\], we investigate Aoki phase structure by effective potential analysis. In Sec. \[sec:Tf\], we discuss parity-flavor symmetry breaking in two-flavor cases. In Sec. \[sec:SD\], we devote ourselves to a summary and discussion.
Staggered-Wilson fermions {#sec:SWF}
=========================
Before looking into staggered-Wilson fermions, we review the Wilson fermion and its relatives. The Wilson term splits 16 species of naive fermions into 5 branches with 1, 4, 6, 4 and 1 fermions. We call this kind of species-splitting terms “flavored-mass terms". As shown in [@CKM1], there are 4 types of flavored-mass terms for naive fermion which satisfy $\gamma_{5}$ hermiticity. ($\gamma_{5}$ in the naive fermion is flavored such as $\gamma_{5}\otimes(\tau_{3}\otimes\tau_{3}\otimes\tau_{3}\otimes\tau_{3})$ in the spin-flavor representation.) All these terms with proper mass shifts lead to a second derivative term as $\sim a\int dx^{4} \bar{\psi}D^{2}_{\mu}\psi$ up to $\mathcal{O}(a^2)$ errors. Thus we can regard them as cousins of Wilson fermion.
There are also non-trivial flavored-mass terms for staggered fermions, which split 4 tastes into branches and satisfy $\gamma_{5}$ hermiticity. Since $\gamma_{5}$ is expressed in spin-taste representation as $\gamma_{5}\otimes\gamma_{5}$ in this case, we only have two flavored-mass terms satisfying $\gamma_{5}$ hermiticity: ${{\bf 1}}\otimes\gamma_{5}$ and ${{\bf 1}}\otimes \sigma_{\mu\nu}$. (For larger discrete symmetry one needs to take a proper sum for $\mu,\nu$ in the latter case.) These spin-flavor representations translate into four- and two-hopping terms in the one-component staggered action up to $\mathcal{O}(a)$ errors. The first type is given by $$M_A= \epsilon\sum_{sym} \eta_{1}\eta_{2}\eta_{3}\eta_{4}
C_{1}C_{2}C_{3}C_{4}
= ({{\bf 1}}\otimes \gamma_{5}) + \mathcal{O}(a)
\ ,
\label{AdamsM}$$ with $$\begin{aligned}
(\epsilon)_{xy}&=(-1)^{x_{1}+...+x_{4}}\delta_{x,y}
\ ,
\\
(\eta_{\mu})_{xy}&=(-1)^{x_{1}+...+x_{\mu-1}}\delta_{x,y}
\ ,
\\
C_{\mu}&=(V_{\mu}+V_{\mu}^{\dag})/2
\ ,
\\
(V_{\mu})_{xy}&=U_{\mu,x}\delta_{y,x+\mu}
\ .\end{aligned}$$ The second type is given by $$\begin{aligned}
M_H&={i\over{\sqrt{3}}}(\eta_{12}C_{12}+\eta_{34}C_{34}+\eta_{13}C_{13}+\eta_{42}C_{42}
+\eta_{14}C_{14}+\eta_{23}C_{23})
\ ,
\nonumber\\
&= [{{\bf 1}}\otimes (\sigma_{12}+\sigma_{34}+\sigma_{13}+\sigma_{42}+\sigma_{14}+\sigma_{23}) ] + \mathcal{O}(a)
\ ,
\label{HoelM}\end{aligned}$$ with $$\begin{aligned}
(\eta_{\mu\nu})_{xy}&=\epsilon_{\mu\nu}\eta_{\mu}\eta_{\nu}\delta_{x,y}
\ ,
\\
(\epsilon_{\mu\nu})_{xy}&=(-1)^{x_{\mu}+x_{\nu}}\delta_{x,y}
\ ,
\\
C_{\mu\nu}&=(C_{\mu}C_{\nu}+C_{\nu}C_{\mu})/2
\ .\end{aligned}$$ We refer to $M_A$ and $M_H$ as the Adams- [@Adams1] and Hoelbling-type [@Hoel], respectively. The former splits the 4 tastes into two branches with positive mass and the other two with negative mass. These two branches correspond to $+1$ and $-1$ eigenvalues of $\gamma_{5}$ in the taste space. The latter splits them into one with positive mass, two with zero mass and the other one with negative mass. We note that $M_{A}$ and $M_{H}$ are also derived from the flavored mass terms for naive fermions through spin-diagonalization as shown in [@CKM1]. Now we introduce a Wilson parameter $r=r\delta_{x,y}$ and shift mass as in Wilson fermions [@Hoel]. Then the Adams-type staggered-Wilson fermion action is given by $$\begin{aligned}
S_{\rm A}\,&=\,\sum_{xy}\bar{\chi}_{x}[\eta_{\mu}D_{\mu}
+r(1+M_A)+M]_{xy}\chi_{y}
\ ,
\label{AdamsS}
\\
D_{\mu}&={1\over{2}}(V_{\mu}-V^\dagger_{\mu})
\ .\end{aligned}$$ Here $M$ stands for the usual taste-singlet mass ($M=M\delta_{x,y}$). The Hoelbling-type fermion action is given by $$\begin{aligned}
S_{\rm H}\,=\,\sum_{xy}\bar{\chi}_{x}[\eta_{\mu}D_{\mu}
+r(2+M_H)+M]_{xy}\chi_{y}
\ .
\label{HoelS}\end{aligned}$$ It is obvious that these lattice fermions have possibility to be two- or one-flavor Wilson fermions. In lattice QCD simulation with these fermions, the mass parameter $M$ will be tuned to take a chiral limit as Wilson fermions. For some negative value of the mass parameter: $-1<M<0$ for Adams-type and $-2<M<0$ for Hoelbling-type, we obtain two-flavor and one-flavor overlap fermions respectively by using the overlap formula in principle.
Discrete Symmetries {#sec:Sym}
===================
In this section, we discuss discrete symmetry of staggered-Wilson fermions. A potential problem with staggered-Wilson fermions in lattice QCD is discrete symmetry breaking. As discussed in [@Adams2; @Hoel], the discrete symmetry possessed by the original staggered fermion is broken to its subgroup both in the Adams-type and Hoelbling-type actions. We below list all the staggered discrete symmetries (shift, axis reversal, rotation and charge conjugation), and look into their status in staggered-Wilson fermions. Shift transformation is given by $$\begin{aligned}
\mathcal{S}_{\mu}:\,\,\chi_{x} \to \zeta_{\mu}(x)\chi_{x+\hat{\mu}}, \,\,\,\,
\bar{\chi}_{x} \to \zeta_{\mu}(x)\bar{\chi}_{x+\hat{\mu}},\,\,\,\,
U_{\nu,x} \to U_{\nu, x+\hat{\mu}}
\ ,
\label{shift1}\end{aligned}$$ with $\zeta_{1}(x)=(-1)^{x_{2}+x_{3}+x_{4}}$, $\zeta_{2}(x)=(-1)^{x_{3}+x_{4}}$, $\zeta_{3}(x)=(-1)^{x_{4}}$ and $\zeta_{4}(x)=1$. It is obvious that this transformation flips the sign of both flavored-mass terms. The Adams type is invariant under the two-shift subgroup as $x \to x+\hat{1}\pm\hat{\mu}$ while the Hoelbling type is invariant under four-shift subgroup as $x\to x+\hat{1}\pm\hat{2}\pm\hat{3}\pm\hat{4}$. Note that these subgroups include the doubled shift $x\to x+2\hat{\mu}$ as their subgroup. The axis reversal transformation is given by, $$\begin{aligned}
\mathcal{I}_{\mu}:\,\,\chi_{x} \to (-1)^{x_{\mu}}\chi_{Ix}, \,\,\,\,
\bar{\chi}_{x} \to (-1)^{x_{\mu}}\bar{\chi}_{Ix},\,\,\,\,
U_{\nu,x} \to U_{\nu, Ix}
\ ,
\label{axis1}\end{aligned}$$ with $I=I^{\mu}$ is the axis reversal $x_{\mu}\to -x_{\mu}$, $x_{\rho}\to x_{\rho}$, $\rho\not= \mu$. It again flips the sign of the flavored-mass terms. The staggered rotational transformation is given by $$\begin{aligned}
\mathcal{R}_{\rho\sigma}:\,\,\chi_{x} \to S_{R}(R^{-1}x)\chi_{R^{-1}x},\,\,\,\,
\bar{\chi_{x}} \to S_{R}(R^{-1}x)\bar{\chi}_{R^{-1}x},\,\,\,\,
U_{\nu, x} \to U_{\nu, Rx}
\ ,
\label{rot1}\end{aligned}$$ where $R_{\rho\sigma}$ is the rotation $x_{\rho}\to x_{\sigma}$, $x_{\sigma}\to -x_{\rho}$, $x_{\tau}\to x_{\tau}$, $\tau \not= \rho, \sigma$ and $S_{R}(x)={1\over{2}}[1\pm\eta_{\rho}(x)\eta_{\sigma}(x)\mp\zeta_{\rho}(x)\zeta_{\sigma}(x)
+\eta_{\rho}(x)\eta_{\sigma}(x)\zeta_{\rho}(x)\zeta_{\sigma}(x)]$ with $\rho$$\sigma$. It is notable that the Adams-type fermion keeps this staggered rotation symmetry while the Hoelbling type loses it. The staggered charge conjugation transformation is given by $$\mathcal{C}:\,\,\chi_{x}\to\epsilon_{x}\bar{\chi}_{x}^{T},\,\,\,\,
\bar{\chi}_{x}\to-\epsilon_{x}\chi_{x}^{T},\,\,\,\,
U_{\nu,x} \to U_{\nu,x}^{*}
\ .
\label{C0}$$ The Adams-type fermion keeps this symmetry while the Hoelbling type loses it.
We next elucidate residual subgroups possessed by staggered-Wilson fermions, and discuss how to define physical discrete symmetries as Parity, Charge conjugation and Hypercubic symmetry. For this purpose we separate spin and flavor rotations in the above transformations. Here we utilize the momentum-space representation in [@GS; @DS]. In this representation we can define two set of clifford generators $\Gamma_{\mu}$ and $\Xi_{\mu}$, which operate on spinor and flavor spaces in the momentum-space field $\phi(p)$, respectively. (Details are shown in Appendix. \[SFS\].) Then the shift transformation translates into $$\mathcal{S}_{\mu}:\,\,\phi(p)\,\,\to\,\, \exp(ip_{\mu})\Xi_{\mu}\,\phi(p)
\ .
\label{shift2}$$ The axis reversal translates into $$\mathcal{I}_{\mu}:\phi(p)\,\,\to\,\,\Gamma_{\mu}\Gamma_{5}\Xi_{\mu}\Xi_{5}\,\phi(Ip)
\ .
\label{axis2}$$ The rotational transformation translates into $$\mathcal{R}_{\rho\sigma}:\,\,\phi(p)\,\,\to\,\,\exp({\pi\over{4}}\Gamma_{\rho}\Gamma_{\sigma})
\exp({\pi\over{4}}\Xi_{\rho}\Xi_{\sigma})\,\phi(R^{-1}p)
\ .
\label{rot2}$$ By using this representation we figure out residual discrete symmetries of staggered-Wilson fermions as follows. We first consider parity. Both staggered-Wilson fermions are invariant under $$\mathcal{I}_{s}\mathcal{S}_{4}\sim
\exp(ip_{4})\Gamma_{1}\Gamma_{2}\Gamma_{3}\Gamma_{5}\,\phi(-{\bf p},p_{4})\sim
\exp(ip_{4})\Gamma_{4}\,\phi(-{\bf p},p_{4})
\ ,
\label{parity}$$ with $\mathcal{I}_{s}\equiv \mathcal{I}_{1}\mathcal{I}_{2}\mathcal{I}_{3}$. This is essentially the continuum parity transformation [@GS]. In the continuum limit the phase factor disappears and it results in the continuum parity transformations $P : \psi(p)\to\gamma_{4}\psi(-{\bf p},p_{4})$ for the Dirac fermion. We thus conclude both staggered-Wilson fermions possess symmetry leading to physical parity symmetry $P$. We note the simple combination of $\mu$-shift and $\mu$-axis reversal (shifted-axis reversal:$\mathcal{S}_{\mu}\mathcal{I}_{\mu}$) is also a symmetry of both fermions.
We next consider physical charge conjugation. In the case of Adams fermion the staggered charge conjugation symmetry $\mathcal{C}$ in Eq. (\[C0\]) remains intact. Thus, physical charge conjugation for the two-flavor branch can be formed in a similar way to usual staggered fermions as shown in [@GS] ($C\sim\mathcal{C}\mathcal{S}_{2}\mathcal{S}_{4}
\mathcal{I}_{2}\mathcal{I}_{4}$). On the other hand, the Hoelbling type breaks $\mathcal{C}$. In this case, however, we can define another charge conjugation by combining $\mathcal{C}$ with rotation transformation as $$\mathcal{C}_{T} : \,\,\mathcal{R}_{21}\mathcal{R}_{13}^{2}\mathcal{C}
\ .
\label{RRRC}$$ The Hoelbling action is invariant under this transformation. By using this symmetry we can define physical charge conjugation $C$ for one-flavor branch as in the Adams type. We thus conclude that both staggered-Wilson fermions have proper charge conjugation symmetry.
$N_f$ $\mathcal{S}$$\&$$\mathcal{I}$-subgroup $\mathcal{R}$-subgroup $P$ $C$ $SW_{4}$
----------- ------- ------------------------------------------------------------------------------------------------------------------ ------------------------------------------------ ------------ ------------ ------------ --
Staggered $4$ $\mathcal{S}_{\mu}$,$\mathcal{I}_{\mu}$ $\mathcal{R}_{\mu\nu}$ $\bigcirc$ $\bigcirc$ $\bigcirc$
Adams $2$ $\mathcal{S}_{\mu}\mathcal{S}_{\nu}$, $\mathcal{S}_{\mu}\mathcal{I}_{\mu}$ $\mathcal{R}_{\mu\nu}$ $\bigcirc$ $\bigcirc$ $\bigcirc$
Hoelbling $1$ $\mathcal{S}_{\mu}\mathcal{S}_{\nu}\mathcal{S}_{\rho}\mathcal{S}_{\sigma}$, $\mathcal{S}_{\mu}\mathcal{I}_{\mu}$ $\mathcal{R}_{\mu\nu}\mathcal{R}_{\rho\sigma}$ $\bigcirc$ $\bigcirc$ $\times$
We lastly consider hypercubic symmetry. In staggered fermions, the rotation Eq. (\[rot1\]) and axis reversal Eq. (\[axis1\]) form hypercubic symmetry [@KilSha], which enhances to Euclidian Lorentz symmetry in the continuum limit. In the case of the Adams-type fermion, the action is invariant under the rotation Eq. (\[rot1\]) and the shifted-axis reversal $\mathcal{S}_{\mu}\mathcal{I}_{\mu}$. These two symmetries can form proper hypercubic symmetry $SW_{4}$ in this case. Thus we conclude that Adams fermion recovers Lorentz symmetry in the continuum. In the Hoelbling-type formulation, the action breaks the rotation symmetry into its subgroup called doubled rotation [@Hoel], which is given by $$\mathcal{R}_{\rho\sigma}\mathcal{R}_{\mu\nu}\sim
\exp[{\pi\over{4}}(\Gamma_{\rho}\Gamma_{\sigma}+\Gamma_{\mu}\Gamma_{\nu})]
\exp[{\pi\over{4}}(\Xi_{\rho}\Xi_{\sigma}+\Xi_{\mu}\Xi_{\nu})]
\phi(R_{\rho\sigma}^{-1}R_{\mu\nu}^{-1}p)
\ ,
\label{DR}$$ where $(\mu,\nu,\sigma,\rho)$ is any permutation of ($1,2,3,4$). It is also invariant under sequential transformations of ($\mu,\nu$ rotation), ($\nu,\mu$ rotation), ($\mu$ shift) and ($\nu$ shift) as $$\mathcal{S}_{\nu}\mathcal{S}_{\mu}\mathcal{R}_{\nu\mu}\mathcal{R}_{\mu\nu}\sim
\exp(ip_{\mu}+ip_{\nu})\Gamma_{\mu}\Gamma_{\nu}\,\phi(\tilde{p})
\ ,
\label{SDR}$$ with $\tilde{p}_{\mu,\nu}=-p_{\mu,\nu}, \tilde{p}_{\tau}=p_{\tau}$, $\tau\not=\mu,\nu$. The loss of rotation symmetry indicates that we cannot define hypercubic symmetry in the Hoelbling fermion. It implies that it could not lead to a correct continuum theory, and we would need to tune parameters to restore Lorentz symmetry. Indeed the recent study on symmetries of staggered-Wilson fermions by Sharpe [@Steve] reports that recovery of Lorentz symmetry requires fine-tuning of coefficients in the gluonic sector in lattice QCD with Hoelbling fermion.
To summarize, Adams fermion possesses physical parity, charge conjugation and hypercubic symmetries while Hoelbling fermion loses hypercubic as shown in Table. \[Table:sym\]. It seems that Hoelbling fermion cannot be straightforwardly applied to lattice QCD while Adams type can. We note that both staggered-Wilson fermions have proper parity symmetry, and we can discuss spontaneous parity symmetry breaking. Moreover, we may find some symptom due to Lorentz symmetry breaking in Hoelbling fermion in the parity-phase structure. This is another motivation to study parity-phase structure in strong-coupling lattice QCD.
In the end of this section, we comment on special symmetries in staggered-Wilson fermions without mass shift. Hoelbling fermion without the mass shift ($\eta_{\mu}D_{\mu}+M_{H}$) possesses special charge conjugation symmetry ($\mathcal{C}_{T}': \chi\to\bar\chi,\,
\bar{\chi}\to\chi$). This topic is beyond the scope of this work, but we note that it can do good to two flavors in the central branch. Use of the central branch is intensively discussed in [@Rev; @PdF].
Hopping Parameter Expansion {#sec:HPE}
===========================
In this section we investigate parity-phase structure in lattice QCD with staggered-Wilson fermions in the framework of hopping parameter expansion (HPE) in the strong-coupling regime [@AokiP]. In the hopping parameter expansion, we treat a mass term as a leading action while we perturbatively treat hopping terms. We thus perform expansion by a hopping parameter which essentially corresponds to inverse of a mass parameter. By using self-consistent equations we derive one- and two-point functions, and calculate meson condensates and meson mass for two types of staggered-Wilson fermions. We for simplicity drop the flavor indices until we discuss the two-flavor case in details in Sec. \[sec:Tf\]. However it is easy to recover the flavor indices for the field $\chi_{f}$, the mass parameter $M_{f}$ and the condensate $\Sigma_{f}$ ($f=1,2,...$).
Hoelbling type {#subsec:HPE-H}
--------------
![Feynman rules in hopping parameter expansion (HPE) with the Hoelbling-type staggered-Wilson fermion. $a$ and $b$ stand for the color indices. $\mathcal{W}_{\alpha\mu\beta\nu,x}^{(2)}$ is given in Table \[Table:U2\].[]{data-label="FR-H"}](HPE-Feynman-rule-Hoelbling.eps){height="6cm"}
$\alpha$ $\beta$ $\mathcal{W}_{\alpha\mu\beta\nu,x}^{(2)}$
---------- --------- ---------------------------------------------------------------------
$+$ $+$ $U_{\mu,x}U_{\nu,x+\hat{\mu}}$
$-$ $-$ $U_{\mu,x-\hat{\mu}}^\dagger U_{\nu,x-\hat{\mu}-\hat{\nu}}^\dagger$
$+$ $-$ $U_{\mu,x}U_{\nu,x+\hat{\mu}-\hat{\nu}}^\dagger$
$-$ $+$ $U_{\mu,x-\hat{\mu}}^\dagger U_{\nu,x-\hat{\mu}}$
![Feynman diagram for mesonic one-point functions in the $\mathcal{O}(K^{3})$ self-consistent equation of HPE with the Hoelbling fermion. Black circles stand for the leading one-point function $\langle \chi_x \bar{\chi}_x \rangle_0$ while white circles stand for $\langle \chi_x \bar{\chi}_x \rangle$ which include next-leading and higher hopping terms. By summing up higher contributions, we obtain the second equality.[]{data-label="One-H"}](HPEwHoelbling.eps){height="6cm"}
We begin with the Hoelbling-type fermion, which contains two-hopping terms in the action. One reason that we start with Hoelbling type regardless of the lower rotation symmetry is that 2-hopping calculation can be a good exercise for 4-hopping case in Adams type. To perform the HPE for the Hoelbling-type fermion, we rewrite the action Eq. (\[HoelS\]) by redefining $\chi \rightarrow \sqrt{2K} \chi$ with the hopping parameter $K=1/[2(M+2r)]$, $$S =
\sum_{x} {{\bar{\chi}}}_x \chi_x + 2K \sum_{x, y} {{\bar{\chi}}}_{x}(\eta_{\mu}
D_\mu )_{xy}\chi_{y}+ 2 K r \sum_{x,y} {{\bar{\chi}}}_x (M_{H})_{xy} \chi_y
\ ,
\label{HPE-Hoel}$$ where $M_{H}$ is given by Eq. (\[HoelM\]). The plaquette action is $1/g^2$ term and we can omit it in the strong-coupling limit. In this section, we derive one- and two-point functions by using a $\mathcal{O}(K^{3})$ self-consistent equation: Solving this equation leads to truncation of diagrams as the ladder approximation having all diagrams to $\mathcal{O}(K^3)$ are taken into account. More precisely, this approximation does not take account of all diagrams, but it successfully includes certain kinds of diagrams to all orders of $K$ thanks to a self-consistent approach. We thus expect that it works to figure out existence of Aoki phase. We note that this approximation especially works well for a small hopping parameter $K\ll1$. In Fig. \[FR-H\], we depict Feynman rules in the HPE for this fermion. The fundamental Feynman rules contain contributions from 0-hopping (mass term), 1-hopping (kinetic term) and 2-hopping (flavored-mass term) terms.
First, by using these Feynman rules, we derive meson condensates from an one-point function of the meson operator $\mathcal{M}_{x}=\bar{\chi}_{x}\chi_{x}$ in the mean-field approximation. The one-point function is defined as $$\begin{aligned}
\langle \chi_x^a \bar{\chi}_x^b \rangle
\equiv& - \delta_{ab} \Sigma_x
= \frac{\int \mathcal{D}[\chi,\bar{\chi},U] \chi_x^a \bar{\chi}_x^b\ e^{S}
}{
\int \mathcal{D}[\chi,\bar{\chi},U] e^{S}
}\ .\end{aligned}$$ Note that we use the partition function $Z=\int \mathcal{D}[\chi,\bar{\chi},U] e^{S}$, not $Z=\int \mathcal{D}[\chi,\bar{\chi},U] e^{-S}$, following the convention for the partition function in the strong-coupling analysis [@KaS]. The leading term in the hopping parameter expansion is given by $$\begin{aligned}
\langle \chi_x^a \bar{\chi}_x^b \rangle_0
= \frac{\int \mathcal{D}[\chi,\bar{\chi},U]\, \chi_x^a \bar{\chi}_x^b\ e^{S_0}}
{\int \mathcal{D}[\chi,\bar{\chi},U]\, e^{S_0}}
= - \delta^{ab}
\ ,\end{aligned}$$ where $S_0=\sum_x \bar{\chi}_x \chi_x$. By using the Feynman rules, we can evaluate the diagrams in Fig. \[One-H\]. $$\begin{aligned}
\langle \chi_x^a \bar{\chi}_x^b \rangle
& \equiv - \delta^{ab} \Sigma_x
{\nonumber}\\
&= \langle \chi_x^a \bar{\chi}_x^b \rangle_0
{\nonumber}\\
&
+ \sum_{\pm \mu}
(-1) (K \eta_{\mu,x})^2
\langle (\chi^a \bar{\chi})_x \rangle_0
U_{\mu,x}
\langle (\chi \bar{\chi})_{x+\hat{\mu}} \rangle_0
U_{\mu,x}^\dagger
\langle (\chi \bar{\chi}^b)_x \rangle_0
{\nonumber}\\
&
+ \sum_{\substack{\pm \mu,\pm \nu \\ (\mu \neq \nu)}}
(-1) \left( 2K r i \eta_{\mu \nu,x} \displaystyle \frac {1}{2^3 \sqrt{3}} \right)^2
\langle (\chi^a \bar{\chi})_x \rangle_0
\mathcal{W}_{\mu\nu,x}^{(2)}
\langle (\chi \bar{\chi})_{x+\hat{\mu}+\hat{\nu}} \rangle_0
\mathcal{W}_{\mu\nu,x}^{(2)\dagger}
\langle (\chi \bar{\chi}^b)_x \rangle_0
{\nonumber}\\
&
+ \sum_{\pm \mu,\pm \nu}
(-1) (K \eta_{\mu,x})^2 (-1) (K \eta_{\nu,x})^2
\langle (\chi^a \bar{\chi})_x \rangle_0
U_{\mu,x}
\langle (\chi \bar{\chi})_{x+\hat{\mu}} \rangle_0
U_{\mu,x}^\dagger
\langle (\chi \bar{\chi})_x \rangle_0 U_{\nu,x}
{\nonumber}\\
& \times
\langle (\chi \bar{\chi})_{x+\hat{\nu}} \rangle_0
U_{\nu,x}^\dagger
\langle (\chi \bar{\chi}^b)_x \rangle_0
{\nonumber}\\
&
+ \sum_{\substack{\pm \mu,\pm \nu ,\pm \rho, \pm \sigma \\ (\mu \neq \nu, \rho \neq \sigma)}}
(-1) \left( 2K r i \eta_{\mu \nu,x} \displaystyle \frac {1}{2^3 \sqrt{3}} \right)^2
(-1) \left( 2K r i \eta_{\rho \sigma,x} \displaystyle \frac {1}{2^3 \sqrt{3}} \right)^2
{\nonumber}\\
& \times
\langle (\chi^a \bar{\chi})_x \rangle_0
\mathcal{W}_{\mu\nu,x}^{(2)}
\langle (\chi \bar{\chi})_{x+\hat{\mu}+\hat{\nu}} \rangle_0
\mathcal{W}_{\mu\nu,x}^{(2)\dagger}
\langle (\chi \bar{\chi})_x \rangle_0
\mathcal{W}_{\rho\sigma,x}^{(2)}
\langle (\chi \bar{\chi})_{x+\hat{\rho}+\hat{\sigma}} \rangle_0
\mathcal{W}_{\rho\sigma,x}^{(2)\dagger}
\langle (\chi \bar{\chi}^b)_x \rangle_0
{\nonumber}\\
&
+ \sum_{\substack{\pm \mu,\pm \nu ,\pm \rho \\ (\mu \neq \nu)}}
(-1) \left( 2K r i \eta_{\mu \nu,x} \displaystyle \frac {1}{2^3 \sqrt{3}} \right)^2
(-1) \left( K \eta_{\rho,x} \right)^2
\langle (\chi^a \bar{\chi})_x \rangle_0
\mathcal{W}_{\mu\nu,x}^{(2)}
{\nonumber}\\
& \times
\langle (\chi \bar{\chi})_{x+\hat{\mu}+\hat{\nu}} \rangle_0
U_{\rho,x+\hat{\mu}+\hat{\nu}}
\langle (\chi \bar{\chi})_{x+\hat{\mu}+\hat{\nu}+\hat{\rho}} \rangle_0
U_{\rho,x+\hat{\mu}+\hat{\nu}}^\dagger
\langle (\chi \bar{\chi})_{x+\hat{\mu}+\hat{\nu}} \rangle_0
\mathcal{W}_{\mu\nu,x}^{(2)\dagger}
\langle (\chi \bar{\chi}^b)_x \rangle_0
{\nonumber}\\
& + \mathcal{O}(K^5)
\ ,
\label{Eq:HPE-Hoelbling1-detail}\end{aligned}$$ where $(\chi \bar{\chi})_x$ stands for $\chi_x \bar{\chi}_x$ and $\mathcal{W}_{\mu\nu,x}^{(2)}=\mathcal{W}_{+\mu+\nu,x}^{(2)}$ in Table. \[Table:U2\]. Note that we consider only connected diagrams in Fig. \[One-H\]. By summing higher hopping terms, the one-point function is obtained as shown in Fig. \[One-H\], which is given by $$- \Sigma_x \equiv - \langle \mathcal{M}_x \rangle =
- \langle \mathcal{M}_x \rangle_0
+ 2K^2 \sum_\mu \Sigma_x \Sigma_{x+\hat{\mu}}
- 2 \cdot \displaystyle \frac {1}{24} (Kr)^2 \sum_{\mu \neq \nu} \Sigma_x
\Sigma_{x+\hat{\mu}+\hat{\nu}}
\ .
\label{Eq:HPE-Hoelbling1}$$ The equation contains terms to $\mathcal{O}(K^{2})$, and $\mathcal{O}(K^{3})$ diagrams are found to vanish due to cancellation between the diagrams. Here we solve it in a self-consistent way for condensate $\Sigma$ within mean-field approximation. We here assume $\Sigma_x=\sigma_x +i \epsilon_x \pi_x$ as the condensate. $\sigma$ and $\pi$ correspond to chiral and pion condensates, respectively. We substitute this form of $\Sigma_{x}$ in Eq. (\[Eq:HPE-Hoelbling1\]) and obtain a self-consistent equation $$- \left( \sigma +i \epsilon_x \pi \right)=-1 + 2K^2 \cdot 4 \left( \sigma^2
+ \pi^2 \right) -2 \cdot \displaystyle \frac {1}{24} (Kr)^2 \cdot 4 \cdot 3
\left( \sigma +i \epsilon_x \pi \right)^2
\ ,
\label{HPE-HoelSelf1}$$ which yields $- \sigma = -1 + 16 K^2 \pi^2$ and $- i \pi = - 8K^2 \cdot 2 i \sigma \pi$. For simplicity, we have set $r=2\sqrt{2}$ to make the equation Eq. (\[HPE-HoelSelf1\]) simpler. Of course we can also discuss for other values of $r$ in general.
Now we have two solutions depending on $\pi=0$ or $\pi\not=0$: For $\pi=0$, we have a trivial solution $\sigma=1$. For $\pi \neq 0$, we have a solution as $$\sigma = \displaystyle \frac{1}{16K^2},\,\,\,\,\,\,\,\,\,\,\,\,\,\,
\pi = \pm \sqrt{ \displaystyle \frac{1}{16K^2} \left( 1- \displaystyle
\frac{1}{16K^2} \right) }
\ .
\label{cond}$$ Nonzero pion condensate implies spontaneous parity breaking for the range $\mid{K}\mid > 1/4$. The sign of the pion condensate in Eq. (\[cond\]) reflects the $Z_2$ parity symmetry of the theory. Thus the parity-broken phase, if it exists, appears in a parameter range $-4\sqrt{2}-2<M<-4\sqrt{2}+2$ in the strong-coupling limit. We note that the critical hopping parameter $|K_c|=1/4$ is small, and we speculate that the $\mathcal{O}(K^{3})$ self-consistent equation is valid around the value.
Next, we discuss the meson mass from a two-point function of the meson operator $\mathcal{S}(0,x)\equiv\langle\mathcal{M}_{0}\mathcal{M}_{x}\rangle$. From Fig. \[Two-H\], we derive the following $\mathcal{O}(K^{3})$ equation for a two-point function.
![Feynman diagram for mesonic two-point functions for $\mathcal{O}(K^{3})$ self-consistent equation with the Hoelbling fermion. []{data-label="Two-H"}](HPEwHoelbling-Mass.eps){height="4cm"}
$$\begin{aligned}
\mathcal{S}(0,x)
= &\langle {{\bar{\chi}}}_0^a\chi_0^a {{\bar{\chi}}}_x^b \chi_x^b \rangle
{\nonumber}\\
= &- \delta_{0x} N_c
{\nonumber}\\
&-K^2 \langle {{\bar{\chi}}}_0^a \chi_0^a {{\bar{\chi}}}_0^c (\eta_{\mu,0})^2
\biggl[
U_{\mu,0}^{cd} \chi_{\hat{\mu}}^d {{\bar{\chi}}}_{\hat{\mu}}^e
(U_{\mu,0}^{\dagger})^{ef}
+ (U_{\mu,-\hat{\mu}}^{\dagger})^{cd}
\chi_{-\hat{\mu}}^d
{{\bar{\chi}}}_{-\hat{\mu}}^e U_{\mu,-\hat{\mu}}^{ef}
\biggr]
\chi_0^f {{\bar{\chi}}}_x^b \chi_x^b \rangle
{\nonumber}\\
& - \left( 2 K r i \displaystyle \frac{1}{2^3 \sqrt{3}} \right)^2
\langle {{\bar{\chi}}}_0^a \chi_0^a {{\bar{\chi}}}_0^c \sum_{\alpha,\beta=\pm} \sum_{\mu \neq \nu} (\eta_{\mu\nu,0})^2
\biggl[
(\mathcal{W}_{\alpha\mu\beta\nu,0}^{(2)})^{cd}
\chi_{\alpha\hat{\mu}\beta\hat{\nu}}^d {{\bar{\chi}}}_{\alpha\hat{\mu}\beta\hat{\nu}}^e
(\mathcal{W}_{\alpha\mu\beta\nu,0}^{(2) \dagger})^{ef}
\biggr]
\chi_0^f {{\bar{\chi}}}_x^b \chi_x^b \rangle
\ ,
\label{HPE-HoelTwo} \end{aligned}$$
where $\mathcal{W}_{\alpha\mu\beta\nu,0}^{(2)}$ is defined in Table. \[Table:U2\]. Note that we consider only connected diagrams in Fig. \[Two-H\]. By integrating out the link variables in the strong-coupling limit, it is simplified as $$\begin{aligned}
\mathcal{S}(0,x) \equiv
\langle {{\bar{\chi}}}_0^a\chi_0^a {{\bar{\chi}}}_x^b \chi_x^b \rangle
= - \delta_{0x} N_c &+ K^2 \sum_{\pm \mu} \langle
\chi_{\hat{\mu}}^a {{\bar{\chi}}}_{\hat{\mu}}^a
{{\bar{\chi}}}_x^b \chi_x^b \rangle
{\nonumber}\\
& + \left( 2 K r i \displaystyle \frac{1}{2^3 \sqrt{3}} \right)^2
\sum_{\substack{\pm \mu,\pm \nu \\ (\mu \neq \nu)}}
\langle \chi_{\hat{\mu}+\hat{\nu}}^a {{\bar{\chi}}}_{\hat{\mu}+\hat{\nu}}^a
{{\bar{\chi}}}_x^b \chi_x^b \rangle
\ .
\label{Eq:HPE-Hoelbling2}\end{aligned}$$ Then the self-consistent equation for $\mathcal{S}$ is given in the momentum space as $$\begin{aligned}
\mathcal{S}(p) &=
- N_c + \biggl[- K^2 \sum_\mu \left( e^{-ip_\mu} + e^{ip_\mu} \right)
{\nonumber}\\
&+ \left( 2 K r \displaystyle \frac{1}{2^3 \sqrt{3}} \right)^2
\sum_{\mu \neq \nu} \left( e^{-i(p_\mu+p_\nu)} + e^{i(p_\mu+p_\nu)}
+ e^{-i(p_\mu-p_\nu)} + e^{i(p_\mu-p_\nu)} \right)\biggr] \mathcal{S}(p)
\ .
\label{HPE-HoelSelf2}\end{aligned}$$ We finally obtain the meson propagator as $$\mathcal{S}(p) = N_c \biggl[ - 2 K^2 \sum_\mu \cos p_\mu
+ 4 \left( 2 K r \displaystyle \frac{1}{2^3 \sqrt{3}} \right)^2
\sum_{\mu \neq \nu} \cos p_\mu \cos p_\nu - 1 \biggr]^{-1}
\ .
\label{MP}$$ Here the pole of $\mathcal{S}(p)$ should give meson mass. Since $\chi$ is an one-component fermion, it may seem to be difficult to find the pion excitation from Eq. (\[MP\]). However, as we discussed, $\gamma_{5}$ in the staggered fermion is given by $\epsilon_{x}=(-1)^{x_{1}+...+x_{4}}$ and the pion operator is given by $\pi_{x}=\bar{\chi}_{x}i\epsilon_{x}\chi_{x}$. We therefore identify momentum of pion by measuring it from a shifted origin $p=(\pi,\pi,\pi,\pi)$. Here we set $p=(i m_\pi a + \pi, \pi,\pi,\pi)$ for $1/\mathcal{S}(p)=0$ in Eq. (\[MP\]). Then we derive the pion mass $m_\pi$ as $$\begin{aligned}
\cosh(m_\pi a) &= 1 + \displaystyle \frac{1-16K^2}{6K^2}
\ ,
\label{HPE-Hoelpi}\end{aligned}$$ where we again set $r=2\sqrt{2}$ for simplicity. In this result, the pion mass becomes zero at $|K|=1/4$, and tachyonic in the range $\mid{K}\mid > 1/4$. It implies that there occurs a phase transition between parity-symmetric and parity-broken phases at $|K|=1/4$, which is consistent with the result from the one-point function in Eq. (\[cond\]). We note that the massless pion at the phase boundary is consistent with the scenario of second-order transition. We can also derive the sigma meson mass by substituting $p=(i m_\pi a, 0, 0, 0)$ for $1/\mathcal{S}(p)=0$ in Eq. (\[MP\]) as $$\begin{aligned}
\cosh(m_\sigma a) &= 1 + \displaystyle \frac{1}{2K^2}
\ .
\label{HPE-Hoelsigma}\end{aligned}$$
Adams type {#subsec:HPE-A}
----------
We investigate the parity-phase structure for the Adams-type staggered-fermion by using $\mathcal{O}(K^{3})$ self-consistent equations in hopping parameter expansion. The approach is basically parallel to that of Hoelbling type. We just need to consider Feynman diagrams for this case. The action Eq. (\[AdamsS\]) is rewritten by redefining $\chi \rightarrow \sqrt{2K} \chi$ with $K=1/[2(M+r)]$ as, $$S =
\sum_{x} {{\bar{\chi}}}_x \chi_x + 2K \sum_{x, y} {{\bar{\chi}}}_{x}(\eta_{\mu}
D_\mu )_{xy}\chi_{y}+ 2 K r \sum_{x,y} {{\bar{\chi}}}_{x}
(M_{A})_{xy} \chi_{y}
\ ,
\label{HPE-Adams}$$ where $M_{A}$ is given in Eq. (\[AdamsM\]). In Fig. \[FR-A\], the Feynman rules in the HPE for this fermion are depicted.
![Feynman rules for the HPE with the Adams fermion. $a$ and $b$ stand for the color indices. We show the concrete forms of $\mathcal{W}_{\alpha\mu\beta\nu\gamma\rho\delta\sigma,x}^{(4)}$ in Table \[Table:U4\] of Appendix \[AdamsEff\].[]{data-label="FR-A"}](HPE-Feynman-rule-Adams.eps){height="6cm"}
First, we derive meson condensates from the one-point function $\mathcal{M}_{x}=\bar{\chi}_{x}\chi_{x}$. The equation for the one-point function is obtained as shown in Fig. \[One-A\], $$\begin{aligned}
- \Sigma_x
& \equiv
- \langle \mathcal{M}_x \rangle
{\nonumber}\\
& =
- \langle \mathcal{M}_x \rangle_0 + 2K^2 \sum_\mu \Sigma_x \Sigma_{x+\hat{\mu}}
- 2 \cdot \displaystyle \frac {1}{(4!)^2 \cdot 2^3} (Kr)^2
\sum_{\mu \neq \nu \neq \rho \neq \sigma}
\Sigma_x \Sigma_{x+\hat{\mu}+\hat{\nu}+\hat{\rho}+\hat{\sigma}}
\ .
\label{Eq:HPE-Adams1}\end{aligned}$$
![Feynman diagram for mesonic one-point functions in the $\mathcal{O}(K^{3})$ self-consistent equations of HPE with the Adams fermion. There is a 4-hopping fundamental diagram, which is peculiar to this fermion.[]{data-label="One-A"}](HPEwAdams.eps){height="3cm"}
We substitute $\Sigma_x=\sigma_x +i \epsilon_x \pi_x$ for $\Sigma_{x}$ in Eq. (\[Eq:HPE-Adams1\]) and obtain the self-consistent equation $$- \left( \sigma +i \epsilon_x \pi \right)=
-1 + 2K^2 \cdot 4 \left( \sigma^2 + \pi^2 \right)
- 2 \cdot \displaystyle \frac {1}{(4!)^2 \cdot 2^3} (Kr)^2 \cdot 4!
\left( \sigma +i \epsilon_x \pi \right)^2
\ .
\label{HPE-AdamsSelf1}$$ From this, we obtain $- \sigma = -1 + 16 K^2 \pi^2$ and $- i \pi = - 8K^2 \cdot 2 i \sigma \pi$. Here we have set $r=16\sqrt{3}$ to make the equation simple. We again have two solutions: For $\pi=0$, we have a trivial solution $\sigma=1$. For $\pi \neq 0$, we have a non-trivial solution as $$\sigma = \displaystyle \frac{1}{16K^2},\,\,\,\,\,\,\,\,\,\,\,\,\,\,
\pi = \pm \sqrt{ \displaystyle \frac{1}{16K^2} \left( 1- \displaystyle
\frac{1}{16K^2} \right) }
\ .
\label{condA}$$ It indicates that parity-broken phase appears in the range of the hopping parameter as $\mid{K}\mid > 1/4$ or equivalently $-16\sqrt{3}-2<M<-16\sqrt{3}+2$.
Next, we discuss the meson mass from a two-point function of the meson operator $\mathcal{S}(0,x)\equiv\langle\mathcal{M}_{0}\mathcal{M}_{x}
\rangle$. From Fig. \[Two-A\] we derive the following equation for two-point functions,
![Feynman diagram for mesonic two-point functions for $ \mathcal{O}(K^{3})$ self-consistent equations of HPE with the Adams fermion.[]{data-label="Two-A"}](HPEwAdams-Mass.eps){height="4cm"}
$$\begin{aligned}
\mathcal{S}(0,x)
= &\langle {{\bar{\chi}}}_0^a\chi_0^a {{\bar{\chi}}}_x^b \chi_x^b \rangle
{\nonumber}\\
=&- \delta_{0x} N_c
{\nonumber}\\
&-K^2 \langle {{\bar{\chi}}}_0^a \chi_0^a {{\bar{\chi}}}_0^c (\eta_{\mu,0})^2
\biggl[
U_{\mu,0}^{cd} \chi_{\hat{\mu}}^d {{\bar{\chi}}}_{\hat{\mu}}^e
(U_{\mu,0}^{\dagger})^{ef}+ (U_{\mu,-\hat{\mu}}^{\dagger})^{cd}
\chi_{-\hat{\mu}}^d {{\bar{\chi}}}_{-\hat{\mu}}^e U_{\mu,-\hat{\mu}}^{ef}\biggr]
\chi_0^f {{\bar{\chi}}}_x^b \chi_x^b \rangle
{\nonumber}\\
& + \left( 2 K r \epsilon \eta_5 \displaystyle \frac{1}{4! \cdot 2^4} \right)^2
\langle {{\bar{\chi}}}_0^a \chi_0^a {{\bar{\chi}}}_0^c
\sum_{\alpha,\beta,\gamma,\delta=\pm} \sum_{\mu \neq \nu \neq \rho \neq \sigma}
\biggl[
\left( \mathcal{W}_{\alpha\mu\beta\nu\gamma\rho\delta\sigma,0}^{(4)} \right)^{cd}
\chi_{\alpha\hat{\mu}\beta\hat{\nu}\gamma\hat{\rho}\delta\hat{\sigma}}^d
{{\bar{\chi}}}_{\alpha\hat{\mu}\beta\hat{\nu}\gamma\hat{\rho}\delta\hat{\sigma}}^e
{\nonumber}\\
& \times
\left( \mathcal{W}_{\alpha\mu\beta\nu\gamma\rho\delta\sigma,0}^{(4) \dagger} \right)^{ef}
\biggr]
\chi_0^f {{\bar{\chi}}}_x^b \chi_x^b \rangle
\ ,
\label{HPE-AdamsTwo} \end{aligned}$$
where $\mathcal{W}_{\alpha\mu\beta\nu\gamma\rho\delta\sigma,x}^{(4)}$ is defined in Table. \[Table:U4\] of Appendix \[AdamsEff\]. By integrating out the link variables in the strong-coupling limit, it is simplified as $$\begin{aligned}
\mathcal{S}(0,x) \equiv
\langle {{\bar{\chi}}}_0^a\chi_0^a {{\bar{\chi}}}_x^b \chi_x^b \rangle
= &- \delta_{0x} N_c + K^2
\sum_{\pm \mu} \langle \chi_{\hat{\mu}}^a {{\bar{\chi}}}_{\hat{\mu}}^a
{{\bar{\chi}}}_x^b \chi_x^b \rangle
{\nonumber}\\
& - \left( 2 K r \displaystyle \frac{1}{4! \cdot 2^4} \right)^2
\sum_{\substack{\pm \mu, \pm \nu, \pm \rho,
\pm \sigma \\ (\mu \neq \nu \neq \rho \neq \sigma)}}
\langle \chi_{\hat{\mu}+\hat{\nu}+\hat{\rho}
+\hat{\sigma}}^a {{\bar{\chi}}}_{\hat{\mu}+\hat{\nu}+\hat{\rho}
+\hat{\sigma}}^a {{\bar{\chi}}}_x^b \chi_x^b \rangle
\ .
\label{Eq:HPE-Adams2}\end{aligned}$$ Then the self-consistent equation for $\mathcal{S}$ is given in the momentum space as $$\begin{aligned}
\mathcal{S}(p) &=
- N_c +
\biggl[- K^2 \sum_\mu \left( e^{-ip_\mu} + e^{ip_\mu} \right)
{\nonumber}\\
&+
\left( 2 K r \displaystyle \frac{1}{4! \cdot 2^4} \right)^2
\sum_{\alpha,\beta,\gamma,\delta=\pm} \sum_{\mu \neq \nu \neq \rho \neq \sigma}
e^{+i(\alpha p_\mu + \beta p_\nu + \gamma p_\rho + \delta p_\sigma)}
\biggr]
\ .
\label{HPE-AdamsSelf2}\end{aligned}$$ We finally obtain the meson propagator as $$S(p) = N_c
\biggl[ - 2 K^2 \sum_\mu \cos p_\mu
+ 16 \left( 2 K r \displaystyle \frac{1}{4! \cdot 2^4} \right)^2
\sum_{\mu \neq \nu \neq \rho \neq \sigma}
\cos p_\mu \cos p_\nu \cos p_\rho \cos p_\sigma- 1 \biggr]^{-1}
\ .
\label{MP2}$$ Here we set $p=(i m_\pi a + \pi, \pi,\pi,\pi)$ for $1/\mathcal{S}(p)=0$ in Eq. (\[MP2\]), which gives the pion mass $m_\pi$ as $$\begin{aligned}
\cosh(m_\pi a) &= 1 + \displaystyle \frac{1-16K^2}{10K^2}
\ ,
\label{HPE-Adamspi}\end{aligned}$$ where we again set $r=16\sqrt{3}$ for simplicity. Here the pion mass becomes zero at $\mid{K}\mid = 1/4$ and becomes tachyonic in the range $\mid{K}\mid > 1/4$. It suggests that there occurs a second-order phase transition between parity-symmetric and broken phases at $|K|=1/4$, which is consistent with Eq. (\[condA\]). We can also derive the sigma meson mass by substituting $p=(i m_\pi a, 0, 0, 0)$ for $1/\mathcal{S}(p)=0$ in Eq. (\[MP2\]) as $$\begin{aligned}
\cosh(m_\sigma a) &= 1 + \displaystyle \frac{1}{6K^2}
\ .
\label{HPE-Asigma}\end{aligned}$$
Effective Potential Analysis {#sec:EPA}
============================
In the previous section, we have investigated the parity-phase structure in hopping parameter expansion. We found a strong sign of parity-broken phase for $\mid{K}\mid >1/4$. In order to judge whether the parity-broken phase is realized as a vacuum, the analysis of the gap solution in the hopping parameter expansion is not enough, and we need to investigate the effective potential for meson fields.
In this section, we consider the effective potential of meson fields for $SU(N)$ lattice gauge theory with staggered-Wilson fermions. In the strong-coupling limit and the large $N$ limit, effective action can be exactly derived by integrating the link variables [@KaS; @AokiP]. Then, by solving a saddle-point equation, we can investigate a vacuum and find meson condensations. In this section we again begin with the Hoelbling case as exercise, and go on to Adams fermion with better discrete symmetry.
Hoelbling type {#hoelbling-type}
--------------
In the strong-coupling limit we can drop plaquette action. Then the partition function for meson fields $\mathcal{M}_x=({{\bar{\chi}}}_x \chi_x)/N$ with the source $J_{x}$ is given by $$\begin{aligned}
Z(J) &= \displaystyle \int {{\cal D}}\left[ \chi, {{\bar{\chi}}}, U \right]
\exp \left[ N \sum_x J_x \mathcal{M}_x + S_F \right]
\ .
\label{ZJ}\end{aligned}$$ where $S_{F}$ stands for the fermion action. Here we have defined $\mathcal{M}_x$ with a $1/N$ factor to ensure it to have a certain large $N$ limit. In this case, $S_F$ is the Hoelbling-type staggered-Wilson action Eq. (\[HoelS\]). $N$ stands for the number of color. In the large $N$ limit, we can perform the link integral. We here consider the effective action up to $\mathcal{O}(\mathcal{M}^{3})$ for meson field $\mathcal{M}$. This order corresponds to the $\mathcal{O}(K^{3})$ self-consistent equation in the hopping parameter expansion.
We develop a method to perform the link-variable integral with multi-hopping fermion action terms. In our method, we perform the link integral by introducing two kinds of link-variable measures. Now we formally rewrite the partition function as, $$\begin{aligned}
Z(J) &= \displaystyle \int {{\cal D}}\left[ \chi, {{\bar{\chi}}}\right]
\exp \left[ \sum_x N \left( J_x + {\hat{M}}\right) \mathcal{M}_x \right]
\exp \left[ \sum_x N W (\Lambda) \right]
\ ,\end{aligned}$$ where we define ${\hat{M}}=M+2r$ and the last term is expressed as, $$\begin{aligned}
\exp \left[ \sum_x N W (\Lambda) \right]
&=
\prod_x Z_x
\ , {\nonumber}\\
Z_x
&= \displaystyle \int \left( \prod_{\mu \neq \nu} {{\cal D}}\left[ U_{\mu,x},
U_{\mu,x+\hat{\nu}} \right] \right)
\exp \left[ - \left( {\mathrm{Tr}}(VE^\dagger) - {\mathrm{Tr}}(V^\dagger E) \right) \right]
\ .
\label{WZ}\end{aligned}$$ \[sec:EPA-H\] $\Lambda$, $E$ and $E^\dagger$ ($V$ and $V^\dagger$) are composites of the fermion field $\chi$ (link variables $U$), which we will explicitly show later. $W(\Lambda)$ is a function of $\Lambda$, which will be an essential part of effective potential of meson fields.
Now we explain how the integral in Eq. (\[WZ\]) can be performed by using two types of the link measure. Let us consider two-dimensional cases in Fig. \[H-2link\] for simplicity. In this case, $U_{\mu,x}$ and $U_{\mu, x+\hat{\nu}} \ (\mu \neq \nu)$ form link variables in a square block. We can classify diagrams to $\mathcal{O}(\mathcal{M}^{3})$ into three types: (1) 1-link ($\mu$) + 1-link ($-\mu$) hoppings, (2) 2-link ($\mu,\nu$) + 2-link ($-\nu,-\mu$) hoppings, (3) 2-link ($\mu,\nu$) + 1-link ($-\nu$) + 1-link ($-\mu$) hoppings. The 1-link hopping comes from the usual staggered kinetic term while the 2-link hopping from the flavored-mass term. (1) and (2) are $\mathcal{O}(\mathcal{M}^{2})$ while (3) is $\mathcal{O}(\mathcal{M}^{3})$. Since one square block contains all the three diagrams, we can derive the effective potential by integrating link variables per each block. We note that $\mathcal{O}(\mathcal{M}^{3})$ diagrams cancel between one another, which is consistent with the HPE calculations. We can also avoid double-counting by adjusting factors for one-link and two-link terms as we will show in Eq. (\[Lambda\]).
![Link variables corresponding to the two kinds of measures in the partition function Eq. (\[WZ\]) in a 2 dimensional case.[]{data-label="H-2link"}](Hoelbling-2link.eps){height="5cm"}
In this method, we need to define sets of link variables and fermion bilinears as $V$ and $E$ in Eq. (\[WZ\]) : $V$ and $E$ are matrices including components corresponding to $1$- and $2$-link terms. We call a space spanned by these matrices “hopping space". Here we define $a,b$ and $\alpha, \beta$ as color and hopping space indices respectively. We also denote ${\mathrm{Tr}}$ as trace for color and the hopping space. Explicit forms of $V$ and $E$ are given by $$\begin{aligned}
V^{ab}_{\alpha \beta}
&= \mathrm{diag} \left(
V_1^{ab}, V_2^{ab}, V_3^{ab}
\right)
\ ,\end{aligned}$$ with $$\begin{aligned}
V_1
&=\mathrm{diag} \left(U_{\mu,x} \right)
\ {\nonumber}\\
&\equiv \mathrm{diag} \underbrace{ \left( U_{1,x}, U_{2,x}, \cdots, U_{4,x} \right) }_4
\ , \\
V_2
&=\mathrm{diag} \left(
U_{\mu,x} U_{\nu,x+\hat{\mu}}
\right)
\ {\nonumber}\\
&\equiv \mathrm{diag} \underbrace{ \left(U_{1,x} U_{2,x+\hat{1}}, U_{1,x} U_{3,x+\hat{1}} , \cdots, U_{4,x} U_{3,x+\hat{4}}\right) }_{12}
\ , \\
V_3
&=\mathrm{diag} \left(
U_{\mu,x+\hat{\nu}} U_{\nu,x+\hat{\mu}}^\dagger
\right)
\ {\nonumber}\\
&\equiv \mathrm{diag} \underbrace{ \left(U_{1,x+\hat{2}} U_{2,x+\hat{1}}^\dagger, U_{1,x+\hat{3}} U_{3,x+\hat{1}}^\dagger , \cdots, U_{4,x+\hat{3}} U_{3,x+\hat{4}}^\dagger \right) }_{12}
\ ,\end{aligned}$$ $$\begin{aligned}
E^{ab}_{\alpha \beta}
&= \mathrm{diag} \left(
E_1^{ab}, E_2^{ab}, E_3^{ab}
\right)
\ ,\end{aligned}$$ and $$\begin{aligned}
E_1
&= \mathrm{diag} \left(
D_{1,\mu}
\right)
\ {\nonumber}\\
&\equiv \mathrm{diag} \underbrace{ \left(
D_{1,1}, D_{1,2}, \cdots, D_{1,4}
\right) }_4
\ , \\
E_i
&= \mathrm{diag} \left(
D_{i,\mu\nu}
\right)
\ {\nonumber}\\
&\equiv
\mathrm{diag} \underbrace{ \left(
D_{i,12}, D_{i,13}, \cdots, D_{i,43}
\right) }_{12}
, \ (i=2,3)
\ ,\end{aligned}$$ where we define the operator $D$ as the fermion bilinears, $$\begin{aligned}
\left( D_{1,\mu}^\dagger \right)^{ab} &=
\displaystyle \frac {1}{2} \eta_{\mu,x} {{\bar{\chi}}}_x^a \chi_{x+\hat{\mu}}^b
\ , \quad
\left( D_{1,\mu} \right)^{ab} =
\displaystyle \frac {1}{2} \eta_{\mu,x} {{\bar{\chi}}}_{x+\hat{\mu}}^a \chi_x^b
\ , \\
\left( D_{2,\mu \nu}^\dagger \right)^{ab} &=
\displaystyle \frac {ir}{2^3 \sqrt{3}} \eta_{\mu \nu,x}
{{\bar{\chi}}}_x^a \chi_{x+\hat{\mu}+\hat{\nu}}^b
\ , \quad
\left( D_{2,\mu \nu} \right)^{ab} =
\displaystyle \frac {ir}{2^3 \sqrt{3}} \eta_{\mu \nu,x}
{{\bar{\chi}}}_{x+\hat{\mu}+\hat{\nu}}^a \chi_x^b
\ , \\
\left( D_{3,\mu \nu}^\dagger \right)^{ab} &=
\displaystyle \frac {ir}{2^3 \sqrt{3}} \eta_{\mu \nu,x+\hat{\nu}}
{{\bar{\chi}}}_{x+\hat{\nu}}^a \chi_{x+\hat{\mu}}^b
\ , \quad
\left( D_{3,\mu \nu} \right)^{ab} =
\displaystyle \frac {ir}{2^3 \sqrt{3}} \eta_{\mu \nu,x+\hat{\nu}}
{{\bar{\chi}}}_{x+\hat{\mu}}^a \chi_{x+\hat{\nu}}^b
\ .\end{aligned}$$ Here $V_{1}$ and $E_{1}$ are $4\times 4$ diagonal matrices while $V_{i}$ and $E_{i}$ $(i=2,3)$ are $12\times 12$ diagonal matrices. Now we have prepared to obtain $W(\Lambda)$. By using the relation $U^{\dagger}U=1$, we obtain the Schwinger-Dyson equation, $$\begin{aligned}
\displaystyle \frac{\partial^2 Z_x}{\partial
E^{ab}_{\alpha \beta} \partial
\left( E^\dagger \right)^{bc}_{\beta \gamma}}
&= - \delta_{ca} \delta^{\alpha \gamma} Z_x
\ .\end{aligned}$$ $W$ should be a function of a gauge-invariant quantities as follows. $$\begin{aligned}
\Lambda^{ab}_{\alpha \beta} &= \displaystyle
\frac {1}{N^2} \left( E^\dagger E \right)^{ab}_{\alpha \beta}
\ .\end{aligned}$$ We can solve the Schwinger-Dyson equation analytically and derive $W$ as a function of $\Lambda$, $$\begin{aligned}
W(\Lambda)
&= {\mathrm{Tr}}\left[ \left( 1-4 \Lambda \right)^{1/2} - 1
- \ln \left[ \displaystyle \frac {1+\left( 1-4 \Lambda \right)^{1/2}}{2}
\right] \right]
\ .\end{aligned}$$ We here perform trace for colors and hopping spaces. $$\begin{aligned}
\sum_x W(\Lambda)
&= - \sum_x \left[ \left( 1-4 \Lambda_x \right)^{1/2}
- 1 - \ln \left[ \displaystyle \frac {1+
\left( 1-4 \Lambda_x \right)^{1/2}}{2} \right] \right]
\ .\end{aligned}$$ Finally we obtain a concrete form of $\Lambda$ as $$\begin{aligned}
\Lambda_x
&=
\displaystyle \frac {1}{8}
\left[
\sum_\mu \mathcal{M}_x \mathcal{M}_{x+\hat{\mu}}
+ \displaystyle \frac {1}{3}
\sum_{\mu \neq \nu} \mathcal{M}_{x+\hat{\mu}} \mathcal{M}_{x+\hat{\mu}+\hat{\nu}}
\right]
-\left( \displaystyle \frac {r}{2^3 \sqrt{3}} \right)^2
\sum_{\mu \neq \nu} \left( \mathcal{M}_x \mathcal{M}_{x+\hat{\mu}+\hat{\nu}}
+ \mathcal{M}_{x+\hat{\nu}} \mathcal{M}_{x+\hat{\mu}}\right)
\ .
\label{Lambda}\end{aligned}$$ The first and second terms correspond to contribution from $D_{1,\mu},D_{1,\mu}^\dagger$, and the third and forth terms correspond to contribution from $D_{i,\mu},D_{i,\mu}^\dagger \ (i=2,3)$. Now we again set $r=2\sqrt{2}$ to compare the result to that of the hopping parameter expansion in Sec. \[sec:HPE\]. We need to change the fermion measure to the meson-field measure as $$\begin{aligned}
\displaystyle \int {{\cal D}}\left[ \chi, {{\bar{\chi}}}\right]
&= \displaystyle \int {{\cal D}}\mathcal{M} \exp \left[ -N \sum_x \ln \mathcal{M}_x \right]
\ .\end{aligned}$$ Then the effective partition function for the meson field is given by $$\begin{aligned}
Z(J)
&= \displaystyle \int {{\cal D}}\mathcal{M} \exp \left[ N \left( \sum_x J_x \mathcal{M}_x
+ S_{\mathrm{eff}}(\mathcal{M}) \right) \right]
\ , \\
S_{\mathrm{eff}}(\mathcal{M})
&= \sum_x \left( {\hat{M}}\mathcal{M}_x - \ln \mathcal{M}_x \right) + \sum_x W(\Lambda)
\ ,
\label{EAc1}\end{aligned}$$ where we denote ${\hat{M}}$ as the shifted mass parameter ${\hat{M}}= M+2r$. The partition function with $J=0$ in the large $N$ limit is reduced to the integrant for the saddle-point values of the meson fields, $$\begin{aligned}
Z(J=0)
&= \displaystyle \int {{\cal D}}\mathcal{M} \exp \left[ N S_{\mathrm{eff}}(\mathcal{M}) \right]
\ {\nonumber}\\
& \sim \exp \left[ N S_{\mathrm{eff}}(\bar{\mathcal{M}}) \right],
\quad \left( N \rightarrow \infty \right)
\ .\end{aligned}$$ Now we consider pion condensate. For now, we consider only scalar $\sigma$ and pseudo-scalar $\pi$ fields as $$\begin{aligned}
\bar{\mathcal{M}}_{x}
&= \sigma + i \epsilon_x \pi
\ ,
\\
&= \Sigma e^{i\epsilon_x \theta}
\ .\end{aligned}$$ By substituting this form of the meson field into the Eq. (\[EAc1\]), we derive the effective action for the $\Sigma$ and $\theta$, $$\begin{aligned}
S_{\mathrm{eff}}(\bar{\mathcal{M}})
&= {\hat{M}}\sum_x \Sigma \cos \theta - \sum_x \ln \Sigma
\ {\nonumber}\\
&- \sum_x \left[ \left( 1- 4 \cdot 2 \Sigma^2 \sin^2 \theta \right)^{1/2}
- \ln \left[ \displaystyle \frac{1+\left( 1- 4 \cdot 2 \Sigma^2 \sin^2
\theta \right)^{1/2}}{2} \right] \right]
\ .\end{aligned}$$ We ignore the irrelevant constant. From the translational invariance, we factorize the 4-dimensional volume $V_4$ from the effective action as $S_{\mathrm{eff}}(\bar{M})=-V_4 V_\mathrm{eff}(\Sigma,\theta)$. Then the effective potential $V_{\mathrm{eff}}$ is given by $$\begin{aligned}
V_{\mathrm{eff}}(\Sigma,\theta)
&= - {\hat{M}}\Sigma \cos \theta + \ln \Sigma
\ {\nonumber}\\
&+ \left[ \left( 1- 4 \cdot 2 \Sigma^2 \sin^2 \theta \right)^{1/2}
- \ln \left[ \displaystyle \frac{1+\left( 1- 4 \cdot 2 \Sigma^2
\sin^2 \theta \right)^{1/2}}{2} \right] \right]
\ .
\label{effS1}\end{aligned}$$ Now let us look into the vacuum structure of this effective potential by solving the saddle-point condition, which are given by $$\begin{aligned}
\displaystyle \frac {\partial V_{\mathrm{eff}}(\Sigma,\theta)}{\partial \Sigma}
&= - {\hat{M}}\cos \theta + \displaystyle \frac {1}{\Sigma}
- \displaystyle \frac {8 \Sigma \sin^2 \theta}{1+
\left( 1- 4 \cdot 2 \Sigma^2 \sin^2 \theta \right)^{1/2}}
\quad
=0
\ , \\
\displaystyle \frac {\partial V_{\mathrm{eff}}(\Sigma,\theta)}{\partial \theta}
&= \Sigma \sin \theta \left[
{\hat{M}}- \displaystyle \frac {8 \Sigma \cos \theta}{1
+ \left( 1- 4 \cdot 2 \Sigma^2 \sin^2 \theta \right)^{1/2}}
\right]
\quad
=0
\ .\end{aligned}$$ Here we find two types of solutions for these equations depending on whether $\theta$ is zero or nonzero: For a trivial solution $\theta=0$, we have $\Sigma=1/{\hat{M}}$. For $\theta \neq 0$, the stationary conditions are written as $$\begin{aligned}
&{\hat{M}}\Sigma - \cos \theta =0
\ , \\
&1 - \displaystyle \frac {8 \Sigma^2}{1+ \left(
1- 4 \cdot 2 \Sigma^2 \sin^2 \theta \right)^{1/2}}
=0
\ .\end{aligned}$$ Then, we find a solution for $\theta\not= 0$ as $$\begin{aligned}
\Sigma &= \bar{\Sigma}= \sqrt{ \displaystyle \frac {1}{8-{\hat{M}}^2}}
\ , \\
\sin^2 \theta &= \sin^2 \bar{\theta}= \displaystyle \frac {2 ( 4-{\hat{M}}^2 ) }{8-{\hat{M}}^2}
\ .\end{aligned}$$ Now we need to figure out which solution is realized as the vacuum of the theory by comparing the potentials for the two solutions. We easily show for ${\hat{M}}^2 < 4$, $$V_{\mathrm{eff}}(1/{\hat{M}},0)-V_{\mathrm{eff}}(\bar{\Sigma},\bar{\theta})>0
\ .$$ while $V_{\mathrm{eff}}(1/{\hat{M}},0)-V_{\mathrm{eff}}(\bar{\Sigma},\bar{\theta})<0$ for ${\hat{M}}^2 > 4$. Thus the vacuum of the strong-coupling QCD with the Hoelbling-type staggered-Wilson fermion is given by the following: For ${\hat{M}}^2 > 4$ or equivalently $M > -4\sqrt{2}+2,\,\, M<-4\sqrt{2}-2$, there is only the chiral condensate as $$\begin{aligned}
\displaystyle \frac {1}{N} \langle {{\bar{\chi}}}\chi \rangle
&= \Sigma \cos \theta \quad = \displaystyle \frac {1}{{\hat{M}}}
\ , \\
\displaystyle \frac {1}{N} \langle {{\bar{\chi}}}i \epsilon_x \chi \rangle
&= \Sigma \sin\theta \quad = 0
\ .\end{aligned}$$ For ${\hat{M}}^2 < 4$ or equivalently $-4\sqrt{2}-2<M<-4\sqrt{2}+2$, there is pion condensate which breaks parity symmetry spontaneously. $$\begin{aligned}
\displaystyle \frac {1}{N} \langle {{\bar{\chi}}}\chi \rangle &=
\bar{\Sigma} \cos \bar{\theta} \quad = \displaystyle \frac {{\hat{M}}}{8-{\hat{M}}^2}
\ , \\
\displaystyle \frac {1}{N} \langle {{\bar{\chi}}}i \epsilon_x \chi \rangle
&= \bar{\Sigma} \sin \bar{\theta} \quad =
\pm \displaystyle \frac {\sqrt{ 2(4-{\hat{M}}^2)}}{8-{\hat{M}}^2}
\ .
\label{picond}\end{aligned}$$ The sign of the pion condensate Eq. (\[picond\]) reflects the $Z_2$ parity symmetry of the theory. The critical mass parameter $M_{c}=-4\sqrt{2}\pm 2$ and the range for the Aoki phase $-4\sqrt{2}-2<M<-4\sqrt{2}+2$ is consistent with those of the hopping parameter expansion shown below Eq. (\[cond\]). These results strongly suggest the existence of the parity-broken phase in the lattice QCD although it is just a strong-coupling limit. Figure \[trans1\] shows that the phase transition is second-order.
![The pion condensate undergoes second-order phase transition.[]{data-label="trans1"}](trans1.eps){height="7cm"}
We can also derive mass spectrum of mesons by expanding the effective action Eq. (\[EAc1\]) up to the quadratic terms of the meson-excitation field $\Pi_x=\mathcal{M}_x-\bar{\mathcal{M}}_x$. Since we are interested in the chiral limit which is taken from the parity-symmetric phase to the critical line, we here concentrate on the pion mass in the parity-symmetric phase. For the parity-symmetric phase (${\hat{M}}^{2}>4$), the quadratic part of the effective action is given by $$\begin{aligned}
S_\mathrm{eff}(\mathcal{M}) - S_\mathrm{eff}(\bar{\mathcal{M}})
&= \sum_{x,y} S_\mathrm{eff}^{(2)} (x,y) \Pi_x \Pi_y
\nonumber\\
&= \int_{-\pi}^\pi \displaystyle \frac{d^4p}{(2\pi)^4}
\Pi(-p) \mathcal{D} \Pi(p)
\ ,\end{aligned}$$ where $\Pi(p)$ is the Fourier component of $\Pi_x$, and $$\begin{aligned}
\mathcal{D}
&= \displaystyle \frac {1}{2\Sigma^2}
+ \left[ \displaystyle \frac{1}{4} \sum_\mu \cos p_\mu
- \displaystyle \frac{1}{24}
\sum_{\mu \neq \nu} \left( \cos p_{\mu + \nu} + \cos p_{\mu - \nu} \right)
\right]
\ ,
\label{HpoD}\end{aligned}$$ with $p_{\mu\pm\nu}\equiv p_{\mu}\pm p_{\nu}$. Then we obtain the pion mass by solving $\mathcal{D}=0$ at $p=(i m_\pi a + \pi, \pi,\pi,\pi)$ as $$\cosh(m_{\pi}a) = 1+{2{\hat{M}}^{2}-8\over{3}}
\ .
\label{HpiM}$$ By using the definition $K=1/2{\hat{M}}$ with ${\hat{M}}=M+2r$ and $r=2\sqrt{2}$, we find $\cosh(m_{\pi}a)=1+(1-16K^{2})/6K^{2}$, which is consistent with the result of the hopping parameter expansion Eq. (\[HPE-Hoelpi\]): The pion mass becomes zero at the critical mass ${\hat{M}}^{2}=4$, which indicates there occurs a second-order phase transition between parity-symmetric and broken phases in the strong-coupling limit. By defining quark mass as $m_{q}a={\hat{M}}-{\hat{M}}_{c}$, we find PCAC relation near the critical mass as $$(m_{\pi}a)^{2} = {16\over{3}}m_{q}a+\mathcal{O}(a^{2})
\ .
\label{HpiM2}$$ We can also study a case for non-zero spacial momenta by considering $p=(iEa+\pi,p_{1}a+\pi,p_{2}a+\pi,p_{3}a+\pi)$ in Eq. (\[HpoD\]). By using the pion mass Eq. (\[HpiM2\]) and re-normalizing the Dirac operator as $-{8\over{3}}\mathcal{D}\to\mathcal{D}$, we show that Eq. (\[HpoD\]) results in the Lorentz-covariant form up to $\mathcal{O}(a)$ discretization errors, $$\mathcal{D}=(E^{2}-{\bf p}^{2}-m_{\pi}^{2})a^{2}+\mathcal{O}(a^{3})
\ ,
\label{HLorentz}$$ with ${\bf p}^{2}=p_{1}^{2}+p_{2}^{2}+p_{3}^{2}$. As we discussed in Sec. \[sec:Sym\], we expected that we may find a sign of Lorentz symmetry breaking of the Hoelbling fermion in this study. However, we eventually cannot find a disease due to the symmetry breaking in the strong-coupling study. We consider that it is because Lorentz symmetry breaking appears mainly in the gluon sector as shown in [@Steve] and it is difficult to find it in the meson sector in this limit. In future work, we may be able to find it by including higher order corrections of $1/g^{2}$ and $1/N$.
We here discuss possibility of other condensation. For this purpose, we consider a general form of the meson field as $$\bar{\mathcal{M}}_{x}=\sigma+i\epsilon_{x}\pi+\sum_{\mu}(-1)^{x_{\mu}}v_{\mu}
+\sum_{\mu}i\epsilon_{x}(-1)^{x_{\mu}}a_{\mu}
+\sum_{\mu>\nu}(-1)^{x_{\mu}+x_{\nu}}t_{\mu\nu}
\ ,
\label{GeM}$$ where we define the vector, axial-vector and tensor meson fields as $v_{\mu}$, $a_{\mu}$ and $t_{\mu}$. We can easily show there is no other condensate by substituting this general form Eq. (\[GeM\]) into the meson action Eq. (\[EAc1\]). Thus we conclude that the vacuum we obtained is a true one.
The results in this section suggest that the chiral limit can be taken in Hoelbling-fermion lattice QCD in a parallel manner to Wilson fermion. However we probably need to tune other parameters to restore Lorentz symmetry in lattice QCD with this type. Therefore, what we can state here is just that, if we succeed to restore Lorentz symmetry by parameter tuning, this fermion could be applied to lattice QCD as Wilson fermion.
Adams type {#sec:EPA-A}
----------
We next investigate the case for the Adams fermion. We again consider the effective potential up to $\mathcal{O}(\mathcal{M}^{3})$. The derivation is almost the same as the Hoelbling case in Subsec. \[sec:EPA-H\]. The main difference between them is the number of the multi-links. The fermion of the Adams type includes the four-hopping terms while the Hoelbling one has the two-hopping terms. In the appendix \[AdamsEff\], we derive the effective potential for the Adams-type fermion. Here we only summarize the results.
In this case, we again set $r=16\sqrt{3}$ to match the result to that of the hopping parameter expansion in Sec. \[sec:HPE\]. We can derive the effective potential for scalar and pseudo-scalar fields by assuming condensation as $\mathcal{M}_{x} = \Sigma e^{i\epsilon_{x}\theta}$. We note that the functional form of the effective potential is the same as Eq. (\[effS1\]). By solving saddle-point equations, we find that the critical mass is given by ${\hat{M}}^{2}_{c}=4$ or equivalently $M_c=-16\sqrt{3}\pm2$ with ${\hat{M}}=M+r$ and $r=16\sqrt{3}$. The vacuum in this case is given by following: For ${\hat{M}}^2 > 4$ or $M > -16\sqrt{3}+2,\,\, M<-16\sqrt{3}-2$, there is only the chiral condensate as $$\begin{aligned}
\displaystyle \frac {1}{N} \langle {{\bar{\chi}}}\chi \rangle
&= \Sigma \cos \theta \quad = \displaystyle \frac {1}{{\hat{M}}}
\ , \\
\displaystyle \frac {1}{N} \langle {{\bar{\chi}}}i \epsilon_x \chi \rangle
&= \Sigma \sin\theta \quad = 0
\ .\end{aligned}$$ For ${\hat{M}}^2 < 4$ or $-16\sqrt{3}-2<M<-16\sqrt{3}+2$, there emerge the pion condensate which breaks the parity symmetry spontaneously. $$\begin{aligned}
\displaystyle \frac {1}{N} \langle {{\bar{\chi}}}\chi \rangle &=
\bar{\Sigma} \cos \bar{\theta} \quad = \displaystyle \frac {{\hat{M}}}{8-{\hat{M}}^2}
\ , \\
\displaystyle \frac {1}{N} \langle {{\bar{\chi}}}i \epsilon_x \chi \rangle
&= \bar{\Sigma} \sin \bar{\theta} \quad =
\pm \displaystyle \frac {\sqrt{ 2(4-{\hat{M}}^2)}}{8-{\hat{M}}^2}
\ .\end{aligned}$$ We note the critical mass and the parameter range of the Aoki phase are consistent with those of the hopping parameter expansion shown below Eq. (\[condA\]). This result supports the existence of the parity-broken phase in the lattice QCD with the Adams fermion again. The behavior of pion condensate in this case is also given by Fig. \[trans1\], which shows the order of phase transition is second-order. This is consistent with the second-order scenario that we can take a chiral limit by a mass parameter tuning.
We derive the pion mass from the quadratic parts of the effective potential in a parallel way to the Hoelbling type. The pion mass for this case is given by $$\cosh(m_{\pi}a) = 1+{2{\hat{M}}^{2}-8\over{5}}
\ ,
\label{ApiM}$$ for the parity-symmetric phase (${\hat{M}}^{2}>4$). By using the definition $K=1/2{\hat{M}}$ with ${\hat{M}}=M+r$ and $r=16\sqrt{3}$, we find $\cosh(m_{\pi}a)=1+(1-16K^{2})/10K^{2}$ which is consistent with the result of the hopping parameter expansion Eq. (\[HPE-Adamspi\]): The pion mass becomes zero at the critical mass ${\hat{M}}^{2}=4$, which indicates there occurs a second-order phase transition between parity-symmetric and broken phases in the strong-coupling limit. The PCAC relation holds near the critical mass also in this case. $$(m_{\pi}a)^{2} = {16\over{5}}m_{q}a+\mathcal{O}(a^{2})
\ .
\label{ApiM2}$$ We can also show that the Lorentz-covariant dispersion relation recovers in the continuum limit in the Adams-type formalism as $\mathcal{D}=(E^{2}-{\bf p}^{2}-m_{\pi}^{2})a^{2}+\mathcal{O}(a^{3})$. It is reasonable that Lorentz-symmetric dispersion recovers since Adams fermion has sufficiently large discrete symmetry for Lorentz symmetry restoration. We can also argue the possibility of other condensations in the same way: We can show there is no other condensates by substituting a general form of the meson fields (\[GeM\]) into the mesonic action for this case.
All these results indicate that, in Adams-type staggered-Wilson, we can take a chiral limit by tuning a mass parameter toward the second-order critical line from the parity-symmetric phase. Since Adams fermion has sufficient discrete symmetry, it can be straightforwardly applied to lattice QCD in a parallel manner to Wilson fermions.
Discussion on two-flavor case {#sec:Tf}
=============================
![Conjectures of pion mass behaviors as a function of a mass parameter in two-flavor Hoelbling fermions and a single Adams fermion. In both cases PCAC relation holds in the parity symmetric phase.[]{data-label="Fig:mpi-HA"}](Aoki-Phase-Mass-Hoelbling-Adams-2flavor.eps){height="5cm"}
In this section, we discuss parity-flavor breaking for two-flavor staggered-Wilson fermions. We first consider two-flavor Hoelbling-type fermion action. In this case, except that the low discrete symmetry would require further parameter tuning, the situation is quite similar to that of the Wilson fermion [@AokiP]. We here assume that mass parameters for two flavors $M_{f}$ $(f=1,2)$ are equal, which means there is exact $SU(2)$ flavor symmetry. The chiral and pion condensates are given by $$\begin{aligned}
{1\over{N}}\langle \bar{\chi}_{f}\chi_{f} \rangle
&= \Sigma_{f} \cos \theta_{f} = {1\over{\hat{M}_{f}}}
\ ,
\label{2fout1}
\\
{1\over{N}}\langle \bar{\chi}_{f}i\epsilon_{x}\chi_{f} \rangle
&= \Sigma_{f} \sin \theta_{f} = 0
\ ,
\label{2fout2}\end{aligned}$$ for $\hat{M}_{f}^{2}\geq 4$ (parity-symmetric phase), while they are given by $$\begin{aligned}
{1\over{N}}\langle \bar{\chi}_{f}\chi_{f} \rangle
&= \bar{\Sigma}_{f} \cos \bar{\theta}_{f} = {\hat{M}_{f}\over{8-\hat{M}_{f}^{2}}}
\ ,
\\
{1\over{N}}\langle \bar{\chi}_{f}i\epsilon_{x}\chi_{f} \rangle
&= \bar{\Sigma}_{f} \sin \bar{\theta}_{f} = \pm {\sqrt{2(4-\hat{M}_{f}^{2})}\over{8-\hat{M}_{f}^{2}}}
\ ,
\label{2fcon}\end{aligned}$$ for $\hat{M_{f}}^{2}< 4$ (Aoki phase). Here we have assumed that only the diagonal condensates in the flavor space ([*i.e.*]{} neutral condensates) can take finite values. Here we remind you of $\hat{M}_{f}=M_{f}+2r$ with $r=2\sqrt{2}$. We first look into the parity-symmetric phase. Although $SU(2)$ chiral symmetry is explicitly broken due to the flavored-mass term, three-massless pions appear on the second-order phase boundary due to divergence of correlation length as shown in Fig. \[Fig:mpi-HA\]. In the parity-broken phase, things depend on whether or not $\bar{\theta}_{1}$ and $\bar{\theta}_{2}$ have the same sign in Eq. (\[2fcon\]). For $\bar{\theta}_{1}=\bar{\theta}_{2}$, $$\begin{aligned}
&\langle \bar{\chi}i\epsilon_{x}\chi \rangle\not= 0
\ ,
\nonumber\\
&\langle \bar{\chi}i\epsilon_{x}\tau_{i}\chi \rangle=0
\ , \,\,\,\,\,(i=1,2,3)
\ , \end{aligned}$$ where $\tau_{i}$ is the Pauli matrix and $\chi$ stands a doublet $\chi=(\chi_{1},\chi_{2})^T$. For $\bar{\theta}_{1}=-\bar{\theta}_{2}$, $$\begin{aligned}
&\langle \bar{\chi}i\epsilon_{x}\chi \rangle=0
\ ,
\nonumber\\
&\langle \bar{\chi}i\epsilon_{x}\tau_{1}\chi \rangle =0
\ ,
\nonumber\\
&\langle \bar{\chi}i\epsilon_{x}\tau_{2}\chi \rangle =0
\ ,
\nonumber\\
&\langle \bar{\chi}i\epsilon_{x}\tau_{3}\chi \rangle \not=0
\ .
\label{2fVW}\end{aligned}$$ From Vafa-Witten’s theorem [@VW], we expect that the latter vacuum ($\bar{\theta}_{1}=-\bar{\theta}_{2}$) realizes [@ACV]. It is also possible to check this by studying next-leading-order calculation of $1/N$ or $1/g^{2}$ expansions. If the latter scenario realizes, the flavored pion condensate Eq. (\[2fVW\]) breaks $SU(2)$ flavor symmetry into its $U(1)$ subgroup as well as parity symmetry [^1]. Thus, in the parity-broken phase, we have two-massless pions as NG bosons associated with spontaneous breaking of the flavor symmetry. We summarize them in Fig. \[Fig:mpi-HA\]. This situation is qualitatively the same as the case of Wilson fermion [@AokiP] except possibility of further parameter tuning to recover Lorentz symmetry.
The Adams-type staggered-Wilson fermion is more fascinating. It has two flavors for each branch in the first place. In this case there is no exact $SU(2)$ flavor symmetry due to taste-mixing in original staggered fermions. It means that in the parity-broken phase there is no massless excitation since there is no continuous symmetry to be broken as Fig. \[Fig:mpi-HA\]. However the number of massless pions in the chiral limit (on the boundary) depends on residual discrete flavor symmetry in the pion sector. If the discrete flavor symmetry is not sufficient to have a degenerate pion triplet, we have only one massless pion in the chiral limit although these three are expected to be degenerate in the continuum limit. If the symmetry is large enough, we have three-massless pions on the phase boundary. The latter is a quite fascinating scenario because we can simulate two-flavor QCD with a single lattice fermion. It is possible to study it by looking into transfer matrix symmetry or chiral Lagrangian potential. Recently Ref. [@Steve] has reported that classification of pion operators from the transfer matrix symmetry indicates three-degenerate pions. Adams fermion can be a new standard of lattice fermion in the near future.
Summary and Discussion {#sec:SD}
======================
In this paper we investigate strong-coupling lattice QCD with staggered-Wilson fermions, with emphasis on the parity-broken phase (Aoki phase) structure. We consider hopping parameter expansion and effective potential analysis in the strong-coupling limit. We have shown that the parity-broken phase and the second-order phase boundary exist for both Adams-type and Hoelbling-type staggered-Wilson fermions, which is consistent with the second-order scenario for a chiral limit.
In Sec. \[sec:Sym\], we discuss and classify discrete symmetries of two types of staggered-Wilson fermions. We show that they are invariant under charge conjugation and parity transformation, the latter of which is defined as 4th-shift followed by spatial axis reversal. We also discuss smaller rotation symmetry in the Hoelbling fermion, which would require further parameter tuning as shown in [@Steve].
In Sec. \[sec:HPE\], we analyze staggered-Wilson fermions by using hopping parameter expansion. From one-point functions of meson fields in the expansion, we find that pion condensate becomes nonzero in some range of the hopping parameter. From two-point functions, we show that square pion mass becomes zero on the boundary and becomes negative in the parameter region with nonzero pion condensate. These results suggest that there is a parity-broken phase and a second-order phase boundary.
In Sec. \[sec:EPA\], we study the effective potential for meson fields in the strong-coupling limit and large $N$ limit to elaborate the phase structure in details. Here we develop a method to derive effective potential for lattice fermion actions with multiple-hopping terms. The gap equations from the effective potential exhibit nonzero pion condensate in the same parameter range as the hopping parameter expansion. From this analysis, we also show that pion becomes massless on the second-order phase boundary, and PCAC relation is reproduced around the boundary. If this property carries over into the weak-coupling regime, we can take a chiral limit by tuning a mass parameter in lattice QCD with staggered-Wilson fermions as with Wilson fermion.
In Sec. \[sec:Tf\], we discuss the two-flavor cases. The situation in two-flavor Hoelbling-type fermions is similar to the original Wilson fermion except less rotational symmetry: Three-massless pions are expected to appear on the second-order critical lines, while two of them remain massless in the Aoki phase due to the flavored pion condensate. However we probably need to care about Lorentz symmetry breaking in this case, thus we cannot straightforwardly apply it to two-flavor lattice QCD. The Adams-type staggered-Wilson fermion contains two flavors in each branch. Although the taste-mixing breaks flavor symmetry at finite lattice spacing, it does not necessarily mean non-degenerate three pions. Moreover SU(2) flavor symmetry should recover in the continuum limit at least, and three-massless pions emerge if we take a chiral and continuum limit. In this case, there is no rotational symmetry breaking, and the hypercubic symmetry will recover in the continuum limit. We can thus perform two-flavor QCD simulations with Adams-type staggered-Wilson fermion more efficiently than usual.
All of these results shows new possibilities of lattice fermion formulations. In particular, the Adams fermion can be straightforwardly applied to 2-flavor lattice QCD since it does not require any other fine-tuning and automatically has two flavors. Taking account of less numerical expenses in the staggered fermion, there is possibility that it would be numerically better than Wilson fermions, especially as an overlap kernel [@PdF]. We finally note that the analysis here does not include contribution from some of higher-hopping terms or higher-meson fields. To confirm our results, we need to perform the same analysis with these higher contributions. In the future work, we can also study detailed mass spectrum of mesons and possibility of small other condensation in the Aoki phase.
TM is thankful to D. Adams, M. Creutz, M. Golterman, T. Izubuchi and S. Sharpe for the fruitful discussions. We are thankful to P. de Forcrand for the fruitful discussions. TK and TN are supported by Grants-in-Aid for the Japan Society for Promotion of Science (JSPS) Research Fellows (Nos. 22-3314, 23-593.). TM is supported by Grant-in-Aid for the Japan Society for Promotion of Science (JSPS) Postdoctoral Fellows for Reseach Abroad (24-8). This work is suppported in part by the Grants-in-Aid for Scientific Research from JSPS (Nos. 09J01226, 10J03314, 11J00593, 23340067, 24340054, and 24540271. ), and by the Grant-in-Aid for the global COE program “The Next Generation of Physics, Spun from Universality and Emergence" from MEXT. This work is based on fruitful discussions in the YIPQS-HPCI workshop “New-Type of Fermions on the Lattice", Feb. 9-24, 2012 in Yukawa Institute for Theoretical Physics. The authors are grateful to the organizers for giving them chances to have interest in the present topics.
spin and flavor separation {#SFS}
==========================
From one-staggered field we define 16 species fields in the momentum space as $\phi(p)_{A}\equiv\chi(p+\pi_{A})$ $(-\pi/2\leq p_{\mu} <\pi/2)$ where $\pi_{A}$ ($A=1,2,...,16$) being 4-dim vectors whose components take $0$ or $\pi$. For convenience, we here consider a 16-multiplet field as $\phi(p)=(\phi(p)_{1},
\phi(p)_{2},\cdots, \phi(p)_{16})^{T}$. As this 16-multiplet field has both the spinor (space-time) and the flavor (taste) indices, we can construct two sets of clifford generators $\Gamma_{\mu}$ and $\Xi_{\mu}$, which operate on spinor and flavor spaces in the momentum field $\phi(p)$. They satisfy the clifford algebra as $$\begin{aligned}
\{ \Gamma_{\mu},\Gamma_{\nu} \}=2\delta_{\mu\nu}
\ ,
\\
\{ \Xi_{\mu},\Xi_{\nu} \}=2\delta_{\mu\nu}
\ ,
\\
\{ \Gamma_{\mu}, \Xi_{\nu} \}=0
\ .\end{aligned}$$ By using these definitions, the Dirac operator for the staggered fermion is given by $D_{st}=i\Gamma_{\mu}\sin p_{\mu}$ for the 16 multiplet $\phi(p)$ [^2].
Strong-coupling analysis for Adams-type {#AdamsEff}
=======================================
$\alpha$ $\beta$ $\gamma$ $\delta$ $\mathcal{W}_{\alpha\mu\beta\nu\gamma\rho\delta\sigma,x}^{(4)}$
---------- --------- ---------- ---------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
$+$ $+$ $+$ $+$ $U_{\mu,x} U_{\nu,x+\hat{\mu}} U_{\rho,x+\hat{\mu}+\hat{\nu}} U_{\sigma,x+\hat{\mu}+\hat{\nu}+\hat{\rho}}$
$-$ $-$ $-$ $-$ $U_{\mu,x-\hat{\mu}}^\dagger U_{\nu,x-\hat{\mu}-\hat{\nu}}^\dagger U_{\rho,x-\hat{\mu}-\hat{\nu}-\hat{\rho}}^\dagger U_{\sigma,x-\hat{\mu}-\hat{\nu}-\hat{\rho}-\hat{\sigma}}^\dagger$
$-$ $+$ $+$ $+$ $U_{\mu,x-\hat{\mu}}^\dagger U_{\nu,x-\hat{\mu}} U_{\rho,x-\hat{\mu}+\hat{\nu}} U_{\sigma,x-\hat{\mu}+\hat{\nu}+\hat{\rho}}$
$+$ $-$ $-$ $-$ $U_{\mu,x} U_{\nu,x+\hat{\mu}-\hat{\nu}}^\dagger U_{\rho,x+\hat{\mu}-\hat{\nu}-\hat{\rho}}^\dagger U_{\sigma,x+\hat{\mu}-\hat{\nu}-\hat{\rho}-\hat{\sigma}}^\dagger$
$+$ $-$ $+$ $+$ $U_{\mu,x} U_{\nu,x+\hat{\mu}-\hat{\nu}}^\dagger U_{\rho,x+\hat{\mu}-\hat{\nu}} U_{\sigma,x+\hat{\mu}-\hat{\nu}+\hat{\rho}}$
$-$ $+$ $-$ $-$ $U_{\mu,x-\hat{\mu}}^\dagger U_{\nu,x-\hat{\mu}} U_{\rho,x-\hat{\mu}+\hat{\nu}-\hat{\rho}}^\dagger U_{\sigma,x-\hat{\mu}+\hat{\nu}-\hat{\rho}-\hat{\sigma}}^\dagger$
$+$ $+$ $-$ $+$ $U_{\mu,x} U_{\nu,x+\hat{\mu}} U_{\rho,x+\hat{\mu}+\hat{\nu}-\hat{\rho}}^\dagger U_{\sigma,x+\hat{\mu}+\hat{\nu}-\hat{\rho}}$
$-$ $-$ $+$ $-$ $U_{\mu,x-\hat{\mu}}^\dagger U_{\nu,x-\hat{\mu}-\hat{\nu}}^\dagger U_{\rho,x-\hat{\mu}-\hat{\nu}} U_{\sigma,x-\hat{\mu}-\hat{\nu}+\hat{\rho}-\hat{\sigma}}^\dagger$
$+$ $+$ $+$ $-$ $U_{\mu,x} U_{\nu,x+\hat{\mu}} U_{\rho,x+\hat{\mu}+\hat{\nu}} U_{\sigma,x+\hat{\mu}+\hat{\nu}+\hat{\rho}-\hat{\sigma}}^\dagger$
$-$ $-$ $-$ $+$ $U_{\mu,x-\hat{\mu}}^\dagger U_{\nu,x-\hat{\mu}-\hat{\nu}}^\dagger U_{\rho,x-\hat{\mu}-\hat{\nu}-\hat{\rho}}^\dagger U_{\sigma,x-\hat{\mu}-\hat{\nu}-\hat{\rho}}$
$+$ $+$ $-$ $-$ $U_{\mu,x} U_{\nu,x+\hat{\mu}} U_{\rho,x+\hat{\mu}+\hat{\nu}-\hat{\rho}}^\dagger U_{\sigma,x+\hat{\mu}+\hat{\nu}-\hat{\rho}-\hat{\sigma}}^\dagger$
$+$ $-$ $+$ $-$ $U_{\mu,x} U_{\nu,x+\hat{\mu}-\hat{\nu}}^\dagger U_{\rho,x+\hat{\mu}-\hat{\nu}} U_{\sigma,x+\hat{\mu}-\hat{\nu}+\hat{\rho}-\hat{\sigma}}^\dagger$
$+$ $-$ $-$ $+$ $U_{\mu,x} U_{\nu,x+\hat{\mu}-\hat{\nu}}^\dagger U_{\rho,x+\hat{\mu}-\hat{\nu}-\hat{\rho}}^\dagger U_{\sigma,x+\hat{\mu}-\hat{\nu}-\hat{\rho}}$
$-$ $+$ $+$ $-$ $U_{\mu,x-\hat{\mu}}^\dagger U_{\nu,x-\hat{\mu}} U_{\rho,x-\hat{\mu}+\hat{\nu}} U_{\sigma,x-\hat{\mu}+\hat{\nu}+\hat{\rho}-\hat{\sigma}}^\dagger$
$-$ $-$ $+$ $+$ $U_{\mu,x-\hat{\mu}}^\dagger U_{\nu,x-\hat{\mu}-\hat{\nu}}^\dagger U_{\rho,x-\hat{\mu}-\hat{\nu}} U_{\sigma,x-\hat{\mu}-\hat{\nu}+\hat{\rho}}$
$-$ $+$ $-$ $+$ $U_{\mu,x-\hat{\mu}}^\dagger U_{\nu,x-\hat{\mu}} U_{\rho,x-\hat{\mu}+\hat{\nu}-\hat{\rho}}^\dagger U_{\sigma,x-\hat{\mu}+\hat{\nu}-\hat{\rho}}$
In this chapter, we show the derivation of the effective potential for the Adams-type fermion in the strong-coupling limit. To derive the effective potential for the Adams type, we replace Eq. (\[WZ\]) by Eq. (\[WZA\]) in the Adams type. $$\begin{aligned}
&\exp \left[ \sum_x N W (\Lambda) \right]
=
\prod_x Z_x ,
{\nonumber}\\
Z_x
&= \displaystyle \int \left( \prod_{\mu \neq \nu \neq \rho \neq \sigma} {{\cal D}}\left[ U_{\mu,x},
U_{\mu,x+\hat{\nu}}, U_{\rho, x+\hat{\mu}+\hat{\nu}}, U_{\sigma, x+\hat{\mu}+\hat{\nu}+\hat{\rho}} \right] \right) \exp \left[ - \left( {\mathrm{Tr}}(VE^\dagger) - {\mathrm{Tr}}(V^\dagger E) \right) \right]
\ .
\label{WZA}\end{aligned}$$ Here we represent the partition function as $4$ link integrals with $U_{\mu,x}$, $U_{\nu, x+\hat{\mu}}$, $U_{\rho, x+\hat{\mu}+\hat{\nu}}$, $U_{\sigma, x+\hat{\mu}+\hat{\nu}+\hat{\rho}}$. $V$ and $E$ in Eq. (\[WZA\]) are the matrices which include components corresponding to $1$-, $2$-, $3$-, and $4$-link terms. The components of $V$ and $E$ consist of link variables and fermion fields respectively. The concrete forms of the $V$ and $E$ for this case are given by $$\begin{aligned}
V^{ab}_{\alpha \beta}
&= \mathrm{diag} \left(
V_1^{ab}, V_2^{ab}, \cdots, V_{11}^{ab}
\right)
\ ,\end{aligned}$$ with$$\begin{aligned}
V_1
&=\mathrm{diag} \left(U_{\mu,x} \right)
\ , \\
V_2
&=\mathrm{diag} \left(
U_{\mu,x} U_{\nu,x+\hat{\mu}} U_{\rho,x+\hat{\mu}+\hat{\nu}} U_{\sigma,x+\hat{\mu}+\hat{\nu}+\hat{\rho}}
\right)
\ , \\
V_3
&=\mathrm{diag} \left(
U_{\mu,x}^\dagger U_{\nu,x} U_{\rho,x+\hat{\nu}} U_{\sigma,x+\hat{\nu}+\hat{\rho}}
\right)
\ , \\
V_4
&=\mathrm{diag} \left(
U_{\mu,x+\hat{\nu}} U_{\nu,x+\hat{\mu}}^\dagger U_{\rho,x+\hat{\mu}} U_{\sigma,x+\hat{\mu}+\hat{\rho}}
\right)
\ , \\
V_5
&=\mathrm{diag} \left(
U_{\mu,x+\hat{\rho}} U_{\nu,x+\hat{\mu}+\hat{\rho}} U_{\rho,x+\hat{\mu}+\hat{\nu}}^\dagger U_{\sigma,x+\hat{\mu}+\hat{\nu}}
\right)
\ , \\
V_6
&=\mathrm{diag} \left(
U_{\mu,x+\hat{\sigma}} U_{\nu,x+\hat{\mu}+\hat{\sigma}} U_{\rho,x+\hat{\mu}+\hat{\nu}+\hat{\sigma}} U_{\sigma,x+\hat{\mu}+\hat{\nu}+\hat{\rho}}^\dagger
\right)
\ , \\
V_7
&=\mathrm{diag} \left(
U_{\mu,x+\hat{\rho}+\hat{\sigma}} U_{\nu,x+\hat{\mu}+\hat{\rho}+\hat{\sigma}} U_{\rho,x+\hat{\mu}+\hat{\nu}+\hat{\sigma}}^\dagger U_{\sigma,x+\hat{\mu}+\hat{\nu}}^\dagger
\right)
\ , \\
V_8
&=\mathrm{diag} \left(
U_{\mu,x+\hat{\nu}+\hat{\sigma}} U_{\nu,x+\hat{\mu}+\hat{\sigma}}^\dagger U_{\rho,x+\hat{\mu}+\hat{\sigma}} U_{\sigma,x+\hat{\mu}+\hat{\rho}}^\dagger
\right)
\ , \\
V_9
&=\mathrm{diag} \left(
U_{\mu,x+\hat{\nu}+\hat{\rho}} U_{\nu,x+\hat{\mu}+\hat{\rho}}^\dagger U_{\rho,x+\hat{\mu}}^\dagger U_{\sigma,x+\hat{\mu}}
\right)
\ , \\
V_{10}
&=\mathrm{diag} \left(
U_{\mu,x+\hat{\nu}}^\dagger U_{\nu,x}^\dagger U_{\rho,x} U_{\sigma,x+\hat{\rho}}
\right)
\ , \\
V_{11}
&=\mathrm{diag} \left(
U_{\mu,x+\hat{\rho}}^\dagger U_{\nu,x+\hat{\rho}} U_{\rho,x+\hat{\nu}}^\dagger U_{\sigma,x+\hat{\nu}}
\right)
\ ,\end{aligned}$$ $$\begin{aligned}
E^{ab}_{\alpha \beta}
&= \mathrm{diag} \left(
E_1^{ab}, E_2^{ab}, \cdots, E_{11}^{ab}
\right)
\ ,\end{aligned}$$ and $$\begin{aligned}
E_1
&= \mathrm{diag} \left(
D_{1,\mu}
\right)
\ , \\
E_i
&= \mathrm{diag} \left(
D_{i,\mu\nu\rho\sigma}
\right)
, \ (i=2,3,\cdots,11)
\ ,\end{aligned}$$ where we define the operator $D$ as the fermion bilinears, $$\begin{aligned}
\left( D_{1,\mu}^\dagger \right)^{ab} &=
\displaystyle \frac {1}{2} \eta_{\mu,x} {{\bar{\chi}}}_x^a \chi_{x+\hat{\mu}}^b
\ , \quad
\left( D_{1,\mu} \right)^{ab} =
\displaystyle \frac {1}{2} \eta_{\mu,x} {{\bar{\chi}}}_{x+\hat{\mu}}^a \chi_x^b
\ , \\
\left( D_{2,\mu \nu \rho \sigma}^\dagger \right)^{ab} &=
- s
{{\bar{\chi}}}_x^a \chi_{x+\hat{\mu}+\hat{\nu}+\hat{\rho}+\hat{\sigma}}^b
\ , \quad
\left( D_{2,\mu \nu} \right)^{ab} =
s
{{\bar{\chi}}}_{x+\hat{\mu}+\hat{\nu}+\hat{\rho}+\hat{\sigma}}^a \chi_x^b
\ , \\
\left( D_{3,\mu \nu \rho \sigma}^\dagger \right)^{ab} &=
- s_\mu
{{\bar{\chi}}}_{x+\hat{\mu}}^a \chi_{x+\hat{\nu}+\hat{\rho}+\hat{\sigma}}^b
\ , \quad
\left( D_{3,\mu \nu} \right)^{ab} =
s_\mu
{{\bar{\chi}}}_{x+\hat{\nu}+\hat{\rho}+\hat{\sigma}}^a \chi_{x+\hat{\mu}}^b
\ , \\
\left( D_{4,\mu \nu \rho \sigma}^\dagger \right)^{ab} &=
- s_\nu
{{\bar{\chi}}}_{x+\hat{\nu}}^a \chi_{x+\hat{\mu}+\hat{\rho}+\hat{\sigma}}^b
\ , \quad
\left( D_{4,\mu \nu} \right)^{ab} =
s_\nu
{{\bar{\chi}}}_{x+\hat{\mu}+\hat{\rho}+\hat{\sigma}}^a \chi_{x+\hat{\nu}}^b
\ , \\
\left( D_{5,\mu \nu \rho \sigma}^\dagger \right)^{ab} &=
- s_\rho
{{\bar{\chi}}}_{x+\hat{\rho}}^a \chi_{x+\hat{\mu}+\hat{\nu}+\hat{\sigma}}^b
\ , \quad
\left( D_{5,\mu \nu} \right)^{ab} =
s_\rho
{{\bar{\chi}}}_{x+\hat{\mu}+\hat{\nu}+\hat{\sigma}}^a \chi_{x+\hat{\rho}}^b
\ , \\
\left( D_{6,\mu \nu \rho \sigma}^\dagger \right)^{ab} &=
- s_\sigma
{{\bar{\chi}}}_{x+\hat{\sigma}}^a \chi_{x+\hat{\mu}+\hat{\nu}+\hat{\rho}}^b
\ , \quad
\left( D_{6,\mu \nu} \right)^{ab} =
s_\sigma
{{\bar{\chi}}}_{x+\hat{\mu}+\hat{\nu}+\hat{\rho}}^a \chi_{x+\hat{\sigma}}^b
\ , \\
\left( D_{7,\mu \nu \rho \sigma}^\dagger \right)^{ab} &=
- s_{\rho+\sigma}
{{\bar{\chi}}}_{x+\hat{\rho}+\hat{\sigma}}^a \chi_{x+\hat{\mu}+\hat{\nu}}^b
\ , \quad
\left( D_{7,\mu \nu} \right)^{ab} =
s_{\rho+\sigma}
{{\bar{\chi}}}_{x+\hat{\mu}+\hat{\nu}}^a \chi_{x+\hat{\rho}+\hat{\sigma}}^b
\ , \\
\left( D_{8,\mu \nu \rho \sigma}^\dagger \right)^{ab} &=
- s_{\hat{\nu}+\hat{\sigma}}
{{\bar{\chi}}}_{x+{\hat{\nu}+\hat{\sigma}}}^a \chi_{x+\hat{\mu}+\hat{\rho}}^b
\ , \quad
\left( D_{8,\mu \nu} \right)^{ab} =
s_{\hat{\nu}+\hat{\sigma}}
{{\bar{\chi}}}_{x+\hat{\mu}+\hat{\rho}}^a \chi_{x+{\hat{\nu}+\hat{\sigma}}}^b
\ , \\
\left( D_{9,\mu \nu \rho \sigma}^\dagger \right)^{ab} &=
- s_{\hat{\nu}+\hat{\rho}}
{{\bar{\chi}}}_{x+\hat{\nu}+\hat{\rho}}^a \chi_{x+\hat{\mu}+\hat{\sigma}}^b
\ , \quad
\left( D_{9,\mu \nu} \right)^{ab} =
s_{\hat{\nu}+\hat{\rho}}
{{\bar{\chi}}}_{x+\hat{\mu}+\hat{\sigma}}^a \chi_{x+\hat{\nu}+\hat{\rho}}^b
\ , \\
\left( D_{10,\mu \nu \rho \sigma}^\dagger \right)^{ab} &=
- s_{\hat{\mu}+\hat{\nu}}
{{\bar{\chi}}}_{x+\hat{\mu}+\hat{\nu}}^a \chi_{x+\hat{\rho}+\hat{\sigma}}^b
\ , \quad
\left( D_{10,\mu \nu} \right)^{ab} =
s_{\hat{\mu}+\hat{\nu}}
{{\bar{\chi}}}_{x+\hat{\rho}+\hat{\sigma}}^a \chi_{x+\hat{\mu}+\hat{\nu}}^b
\ , \\
\left( D_{11,\mu \nu \rho \sigma}^\dagger \right)^{ab} &=
- s_{\hat{\mu}+\hat{\rho}}
{{\bar{\chi}}}_{x+\hat{\mu}+\hat{\rho}}^a \chi_{x+\hat{\nu}+\hat{\sigma}}^b
\ , \quad
\left( D_{11,\mu \nu} \right)^{ab} =
s_{\hat{\mu}+\hat{\rho}}
{{\bar{\chi}}}_{x+\hat{\nu}+\hat{\sigma}}^a \chi_{x+\hat{\mu}+\hat{\rho}}^b
\ .\end{aligned}$$ Note that $s=r \left(\epsilon \eta_5 \right)_x / (4! \cdot 16),
s_\mu=r \left(\epsilon \eta_5 \right)_{x+\hat{\mu}} / (4! \cdot 16),$ and $s_{\mu+\nu}=r \left(\epsilon \eta_5 \right)_{x+\hat{\mu}+\hat{\nu}} / (4! \cdot 16)$. Here $V_{1}$ and $E_{1}$ are $4\times 4$ diagonal matrices while $V_{i}$ and $E_{i}$ $(i=2,3,\cdots,11)$ are $24\times 24$ diagonal matrices. Here we note the situation of the cancellation between the diagrams crossing the different blocks is basically the same as the case for the Hoelbling type although there is difference between the 2-link and 4-link hoppings. We can derive $W$ as a function of $\Lambda$ by using the Schwinger-Dyson equation in a similar way to the Hoelbling type. $\Lambda$ is $$\begin{aligned}
\Lambda_x
&=
\displaystyle \frac {1}{16}
\biggl[
\sum_\mu \mathcal{M}_x \mathcal{M}_{x+\hat{\mu}}
+ \displaystyle \frac {1}{3} \sum_{\mu \neq \nu} \mathcal{M}_{x+\hat{\mu}} \mathcal{M}_{x+\hat{\mu}+\hat{\nu}}
\ {\nonumber}\\
& + \displaystyle \frac {1}{6} \sum_{\mu \neq \nu \neq \rho} \mathcal{M}_{x+\hat{\mu}+\hat{\nu}} \mathcal{M}_{x+\hat{\mu}+\hat{\nu}+\hat{\rho}}
+ \displaystyle \frac {1}{6} \sum_{\mu \neq \nu \neq \rho \neq \sigma} \mathcal{M}_{x+\hat{\mu}+\hat{\nu}+\hat{\rho}} \mathcal{M}_{x+\hat{\mu}+\hat{\nu}+\hat{\rho}+\hat{\sigma}}
\biggr]
\ {\nonumber}\\
&-\left( \displaystyle \frac {r}{4! \cdot 16} \right)^2
\sum_{\mu \neq \nu \neq \rho \neq \sigma}
\left(
2 \mathcal{M}_x \mathcal{M}_{x+\hat{\mu}+\hat{\nu}+\hat{\rho}+\hat{\sigma}}
+4 \mathcal{M}_{x+\hat{\mu}} \mathcal{M}_{x+\hat{\nu}+\hat{\rho}+\hat{\sigma}}
+2 \mathcal{M}_{x+\hat{\mu}+\hat{\nu}} \mathcal{M}_{x+\hat{\rho}+\hat{\sigma}}
\right)
\ .\end{aligned}$$
Order parameter and zero eigenvalue of staggered-Wilson operator {#zeroeigen}
================================================================
We investigate the relation between the order parameter $\langle {{\bar{\chi}}}i \epsilon_x \tau_3 \chi \rangle$ for spontaneous symmetry breaking and zero eigenvalue of staggered-Wilson operator. In QCD, the chiral condensate $\langle \bar{\psi} \psi \rangle$ which is the order parameter for spontaneous breaking of chiral symmetry relates to the zero eigenvalue of Dirac operator. It is called Banks-Casher relation [@BCrel]. In Wilson fermion, the pion condensate $\langle \bar{\psi} i \gamma_5 \tau_3\psi \rangle$ which is the order parameter for spontaneous breaking of parity-flavor symmetry relates to the zero eigenvalue of Wilson operator. In staggered-Wilson fermion (two-flavor Hoelbling type), we derive the relation between $\langle {{\bar{\chi}}}i \epsilon_x \tau_3 \chi \rangle$ and zero eigenvalue as Wilson fermion. Then we add the external field $H$ for order parameter to the Hoelbling-type staggered-Wilson action Eq. (\[HoelS\]), $$\begin{aligned}
S_{H}(H)
&= {{\bar{\chi}}}\left[ D_{SW}(M) + i \epsilon_x \tau_3 H \right] \chi
\ .\end{aligned}$$ The order parameter $\langle {{\bar{\chi}}}i \epsilon_x \tau_3 \chi \rangle$ is represented as, $$\begin{aligned}
\lim_{H \rightarrow 0} \langle {{\bar{\chi}}}i \epsilon_x \tau_3 \chi \rangle
&= -\lim_{H \rightarrow 0} \lim_{V \rightarrow \infty} \displaystyle \frac{1}{V} \mathrm{Tr} \left( i \epsilon_x \tau_3
\displaystyle \frac{1}{D_{SW}+ i\epsilon_x \tau_3 H} \right)
\, {\nonumber}\\
&= -\lim_{H \rightarrow 0} \lim_{V \rightarrow \infty} \displaystyle \frac{1}{V} \mathrm{tr} \left[ i \epsilon_x \left(
\displaystyle \frac{1}{D_{SW}+ i\epsilon_x H} - \displaystyle \frac{1}{D_{SW}- i\epsilon_x H} \right) \right]
\, {\nonumber}\\
&= -\lim_{H \rightarrow 0} \lim_{V \rightarrow \infty} \displaystyle \frac{1}{V} \mathrm{tr} \left[ i \left(
\displaystyle \frac{1}{H_{SW}+ i H} - \displaystyle \frac{1}{H_{SW}- i H} \right) \right]
\, {\nonumber}\\
&= -i \lim_{H \rightarrow 0} \lim_{V \rightarrow \infty} \displaystyle \frac{1}{V} \sum_\lambda \langle \lambda \mid \left[
\displaystyle \frac{1}{\lambda+ i H} - \displaystyle \frac{1}{\lambda- i H} \right] \mid \lambda \rangle
\, {\nonumber}\\
&= -i \lim_{H \rightarrow 0} \lim_{V \rightarrow \infty} \displaystyle \frac{1}{V} \int d\lambda \rho\left( \lambda \right)
\left[
\displaystyle \frac{1}{\lambda+ i H} - \displaystyle \frac{1}{\lambda- i H} \right]
\, {\nonumber}\\
&= - \displaystyle \frac{2 \pi \rho(0)}{V}
\, {\nonumber}\\
&\equiv - \lim_{\epsilon \rightarrow 0} \lim_{H \rightarrow 0} \lim_{V \rightarrow \infty} \displaystyle \frac{2 \pi \rho(\epsilon)}{V}
\ ,\end{aligned}$$ where $H_{SW}=\epsilon_x D_{SW}$ is the Hermitian operator. $\mathrm{Tr}$ means the traces for flavor, color and space while $\mathrm{tr}$ means the traces for color and space. $\lambda$ and $\mid{\lambda}\rangle$ are the eigenvalues and eigenstates and $\rho(\lambda)=\sum_{\lambda^{\prime}} \delta( \lambda - \lambda^\prime )$ is the density of the state. By this analysis, we find the order parameter $\langle {{\bar{\chi}}}i \epsilon_x \tau_3 \chi \rangle$ for spontaneous breaking of parity-flavor symmetry relates to the zero eigenvalue of the staggered-Wilson operator $H_{SW}$. Also, we can derive this relation for Adams fermion in the same way.
[99]{}
K. G. Wilson, Phys. Rev. D [**10**]{}, 2445 (1974).
H. B. Nielsen and M. Ninomiya, Nucl. Phys. B [**185**]{}, 20 (1981); Nucl. Phys. B [**193**]{} 173 (1981); Phys. Lett. B [**105**]{} 219 (1981).
M. Creutz, T. Kimura and T. Misumi, JHEP 1012, 041 (2010) \[[[arXiv:1011.0761]{}]{}\].
T. Kimura, M. Creutz and T. Misumi, PoS Lattice2011 (2011) 106 \[arXiv:1110.2482\].
J. B. Kogut and L. Susskind, Phys. Rev. D [**11**]{}, 395 (1975).
L. Susskind, Phys. Rev. D [**16**]{}, 3031 (1977).
H. S. Sharatchandra, H. J. Thun and P. Weisz, Nucl. Phys. B [**192**]{}, 205 (1981).
D. H. Adams, Phys. Rev. Lett. [**104**]{}, 141602 (2010) \[[[arXiv:0912.2850]{}]{}\].
D. H. Adams, Phys. Lett. B [**699**]{}, 394 (2011) \[[[arXiv:1008.2833]{}]{}\].
C. Hoelbling, Phys. Lett. B [**696**]{}, 422 (2011) \[[[arXiv:1009.5362]{}]{}\].
P. de Forcrand, A. Kurkela and M. Panero, PoS Lattice2010 (2011) 080 \[[[arXiv:1102.1000]{}]{}\]: JHEP [**1204**]{} (2012) 142 \[[[arXiv:1202.1867]{}]{}\].
S. Aoki, Phys. Rev. D [**30**]{}, 2653 (1984).
S. Aoki, Phys. Rev. D [**33**]{}, 2377 (1986); [**34**]{}, 3170 (1986); Phys. Rev. Lett. [**57**]{} 3136 (1986); Nucl. Phys. B [**314**]{}, 79 (1989).
M. Creutz, (1996) \[[[arXiv:hep-lat/9608024]{}]{}\].
S. Sharpe and R. Singleton. Jr, Phys. Rev. D [**58**]{}, 074501 (1998) \[[[arXiv:hep-lat/9804028]{}]{}\].
Another interesting scenario is discussed in V. Azcoiti, G. Di Carlo, A. Vaquero, Phys. Rev. D [**79**]{}, 014509 (2009) \[[[arXiv:0809.2972]{}]{}\] and their other works.
S. R. Sharpe, Phys. Rev. D [**79**]{}, 054503 (2009) \[[[arXiv:0811.0409]{}]{}\].
P. H. Ginsparg and K. G. Wilson, Phys. Rev. D [**25**]{}, 2649 (1982).
N. Neuberger, Phys. Lett. B [**427**]{}, 353 (1998) \[[[arXiv:hep-lat/9801031]{}]{}\].
D. B. Kaplan, Phys. Lett. B [**288**]{}, 342 (1992) \[[[arXiv:hep-lat/9206013]{}]{}\].
V. Furman and Y. Shamir, Nucl. Phys. B [**439**]{}, 54 (1995) \[[[arXiv:hep-lat/9405004]{}]{}\].
M. Creutz, T. Kimura and T. Misumi, Phys. Rev. D [**83**]{}, 094506 (2011) \[[[arXiv:1101.4239]{}]{}\].
T. Misumi, M. Creutz, T. Kimura, T. Z Nakano and A. Ohnishi, PoS Lattice2011 (2011) 108 \[arXiv:1110.1231\].
N. Kawamoto and J. Smit, Nucl. Phys. B[**192**]{}, 100 (1981).
C. van den Doel and J. Smit, Nucl. Phys. B [**228**]{}, 122 (1983).
M. F. L. Golterman and J. Smit, Nucl. Phys. B [**245**]{}, 61 (1984).
G. W. Kilcup and S. Sharpe, Nucl. Phys. B [**283**]{}, 493 (1987).
S. Sharpe, Talk in YIPQS-HPCI workshop “New-Type of Fermions on the Lattice" (2012), http://www2.yukawa.kyoto-u.ac.jp/ws/2011/newtype/Talk-slides/sharpe-kyoto12-1.pdf\
and manuscript in preparation.
T. Kimura, S. Komatsu, T. Misumi, T. Noumi, S. Torii and S. Aoki, JHEP [**01 (2012)**]{} 048 \[[[arXiv:1111.0402]{}]{}\].
C. Vaffa and E. Witten, Phys. Rev. Lett [**53**]{}, 535 (1984).
T. Banks and A. Casher, Nucl. Phys. B [**169**]{}, 103 (1980).
[^1]: In Appendix \[zeroeigen\], we show the relation between an order parameter of the phase transition and zero eigenvalues of staggered-Wilson operator, as Wilson fermion.
[^2]: The origin of the discrepancy between this form and the usual spin-taste representation is clearly elaborated in the reference, G. P. Lepage, \[arXiv:1111.2955\].
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The Langer-Schwartz equations for precipitation are formulated to calculate nucleation, growth and coarsening of second phase precipitates under non-isothermal situations. A field-theoretic steady-state nucleation rate model is used in the analysis. The field-theoretic nucleation rate is compared with the classical nucleation rate relation derived from the Fokker-Planck equation. This integrated model is used to simulate a bcc-to-hcp phase quenching and subsequent annealing in the hcp phase of a dilute zirconium alloy, where the mean precipitate size, number density and the degree of supersaturation are calculated as a function of time. The influence of cooling rate on the aforementioned parameters is evaluated. A lower cooling rate results in larger precipitates with a smaller number density in concordance with observations.'
author:
- 'A. R. Massih$^{1,2}$ and L. O. Jernkvist$^1$'
bibliography:
- 'alma.bib'
title: 'Nucleation and growth of second phase precipitates under non-isothermal conditions [^1]'
---
Introduction {#sec:Intro}
============
Various important properties of engineering alloys, such as their mechanical strength and toughness, creep and corrosion resistance, magnetic and superconducting behaviors, are basically governed by the presence and characteristics of second phase precipitated particles (SPPs) [@Wagner_Kampmann_1991]. In many material processing methods, the SPP characteristic is controlled by quenching the material from a high temperature phase, where the constituents of the SPPs are in solid solution, to a low temperature phase, in which precipitation occurs and a distribution of SPPs is observed. This process embraces such phenomena as nucleation, spinodal decomposition, late stage growth and coarsening, which are broadly described by the framework of the kinetics of first-order transitions [@Gunton_et_al_1983].
For example, when a binary alloy system is quenched from an equilibrium state (phase) to a non-equilibrium state inside a coexistence curve of its phase diagram, the quenched system gradually transforms from the non-equilibrium state to an equilibrium state, consisting of two coexisting phases. This transformation occurs by time-dependent spatial fluctuations of the local concentration of one of the two component species. Commonly, two different kinds of instabilities are considered; in one case the solid solution experiences a shallow quench in the metastable region, in another, it is quenched deeply into the unstable region of the miscibility gap [@Wagner_Kampmann_1991]. In the former case, the instability is due to the formation of stable nuclei via localized composition fluctuations, requiring to surmount an energy barrier that is characterized by an incubation period. This corresponds to a metastable phase and the transformation is by nucleation and growth. In the latter case, the instability is caused by the formation of non-localized, small-amplitude, spatially extended composition fluctuations that grow spontaneously in amplitude as the time after the quench proceeds [@Binder_1991]. For a binary system, this phenomenon is called spinodal decomposition, for which the quenched state lies within the boundary between unstable and metastable states, the so-called spinodal curve.
Langer and Schwartz [@Langer_Schwartz_1980] have developed a detailed theory describing the nucleation, growth and coarsening of droplets in metastable, near-critical fluids. Wendth and Haasen [@Wendt_Haasen_1983] adapted this theory for solid solutions and used it to interpret the precipitation of $\gamma '$-Ni$_3$Al SPPs in Ni-14 at.$\%$Al. Later, Kampmann and Wagner [@Kampmann_Wagner_1984] introduced a non-linearized Gibbs-Thomson relation into the Langer-Schwartz equations in order to account for cases where the surface tension of precipitates and the supersaturation are large. They also included an empirical time-dependent factor on the steady-state nucleation rate to include transient nucleation kinetics that has been observed in many experiments. Kampmann and Wagner used the model to interpret and explain observations on $\beta '$-Cu$_4$Ti particles in Cu-1.9 at.$\%$Ti alloy.
In the present paper, we consider homogeneous nucleation, growth and coarsening of SPPs under non-isothermal conditions. The Langer-Schwartz equations for the precipitation kinetics are formulated in a way that can be used to calculate nucleation and growth of SPPs under time-varying temperature situations such as quenching. A field-theoretic steady-state nucleation rate model is used in our evaluation. Moreover, the basic models for growth and coarsening are presented, followed by a computation of an isothermal experiment on a copper-cobalt alloy phase transformation, and a simulation of a body-centered cubic (bcc) to hexagonal close-packed (hcp) phase quenching and subsequent annealing in the hcp phase of a dilute zirconium alloy.
Models {#sec:model}
======
The models considered here describe the phenomena of homogeneous nucleation, growth and coarsening of second phase precipitates in non-isothermal conditions in an integrated fashion. The precipitates are assumed to be spherical and at each instant are in local equilibrium with their surrounding; therefore the concentration near the SPP/matrix interface is determined by the Gibbs-Thomson boundary condition. A key variable defining the state of the system is the supersaturation; here defined as $x \equiv \ln{(C/C_\infty)}$, where $C$ is the solute concentration in the matrix, which is time-dependent, and $C_\infty$ the solubility limit or the equilibrium solute concentration, which is strongly temperature dependent. The Langer-Schwartz theory of precipitation basically comprises a nucleation model and a growth model. In this section we outline these models and describe our formulation of the theory for non-isothermal applications.
Nucleation {#sec:nuke}
----------
### Classical Theory {#sec:cnt}
Consider that a spherical (or circular) particle with radius $R$ is emerged from a metastable state. The particle free energy consists of the surface and the bulk energy parts, $$\label{eq:spp_energy}
F(R) = S_d\Big(\sigma R^{d-1}-\frac{\mu_e}{d} R^d\Big),$$ where $d$ is the spatial dimensionality, $S_3=4\pi$ and $S_2=2\pi$, $\sigma$ the surface energy, and $\mu_e$ the difference in free energy density (per unit volume) between the metastable and stable phases. The function $F(R)$ goes through a maximum at a critical radius $R=R_c=(d-1)\sigma/\mu_e$. The height of the nucleation energy barrier is given by: $$\label{eq:spp_energy_max}
F_c=F(R_c) = \frac{S_d}{d}\sigma R_c^{d-1}.$$
The probability of formation of a particle of radius $R$ is $P\sim
\exp{[-\beta F(R)]}$, where $\beta=1/k_BT$ with $k_B$ Boltzmann’s constant and $T$ the temperature. Thus the critical nucleus, which maximizes $F$, is the least probable one. Once the particle forms, it will grow to reduce its free energy.
The basic ingredient of the considered model is the nucleation rate of the secondary phase precipitates. In the standard formulation, attributed to Ya. B. Zel’dovich [@Landau_Lifshitz_1981], the nucleation process is described by a Fokker-Planck equation: $$\label{eq:FPE}
\frac{\partial n}{\partial t}=\frac{\partial}{\partial
R}\Big(B\frac{\partial n }{\partial R}\Big) + \beta \frac{\partial}{\partial
R}\Big(B n\frac{ \partial F}{\partial R}\Big),$$ where $n=n(R,t)$ is the number of SPP in the radius interval $[R,
R+dR]$ per unit volume at time $t$, $B=B(R)$ a nucleation size diffusion coefficient and $F$ the SPP free energy. After a transient time $t_0$, a constant nucleation rate of SPPs with radii larger than $R_c$ prevails, $J=\mathrm{const}$. More concretely, if $R>>R_c$, $n(R,t)dR=Jdt$ is the number of growing newly emerging particles in a time interval $dt$ per unit volume and is independent of $R$ [@Onuki_2002]. The steady-state solution of Eq. (\[eq:FPE\]), $n=n_s(R)$ is obtained by integration, viz., $$\label{eq:sss}
n_s(R) = J\, e^{-\beta F(R)}\int_R^\infty dr
B(r)^{-1}e^{\beta F(r)},$$ where the condition $n_s(R)\rightarrow 0$ as $R\rightarrow \infty$ was imposed. Next, noting that the steady-state solution of the Fokker-Planck equation is determined by the Boltzmann distribution, i.e., $n_s=n_\ell\exp{[-\beta F(R)]}$, an expression for $J$ can be obtained, when $R$ is near $R_c$, giving (Appendix \[sec:appenda\]) $$\label{eq:cnt2}
J=(2\pi)^{-1/2}(d-1)^{1/2}D_n(S_d\beta\sigma)^{-1/2}
R_c^{-(d+1)/2} n_l\exp{(-\beta F_c)},$$ where $D_n=S_d\beta\sigma R_c^{d-1}B_c$ is the critical nucleation diffusivity with $B_c=B(R_c)$. Relating the critical radius to the supersaturation $x$ and the capillary length $\ell$ according to: $R_c=\ell/x$, where $\ell=2\beta\sigma v_a$ and $v_a$ the atomic volume of the precipitate, Eq. (\[eq:cnt2\]) is expressed in terms of the supersaturation, i.e., $$\label{eq:cnt3}
J=\sqrt{\frac{d-1}{2\pi d}}D_n n_\ell\Big(\frac{x_0}{\ell}\Big)
\Big(\frac{x}{x_0}\Big)^{\frac{d+1}{2}}\exp{\big[-\big(\frac{x_0}{x}\big)^{d-1}\big]},$$ where $$\label{eq:cap_length}
x_0=\Big(\frac{S_d\beta\sigma}{d}\Big)^{\frac{1}{d-1}}\,\ell.$$ Relation (\[eq:cnt3\]) is the classical expression for the nucleation rate formulated in $d$ spatial dimensions as a function of the supersaturation. The scaling in the critical region suggests that the Boltzmann prefactor is $n_\ell \sim \ell^{-(d+1)}$. Whence, setting $\mathcal{A}\equiv\ \sqrt{(d-1)/2\pi d}$, Eq. (\[eq:cnt3\]) can be written as $$J = \mathcal{A}\frac{D_nx_0}{\ell^{d+2}}\left( \frac{x}{x_o} \right)^{\frac{d+1}{2}}
\exp{\big[{- \left( \frac{x_o}{x}\right)^{d-1}}\big]}.
\label{eq:nr_bdz}$$
### Field Theoretic Approach {#sec:ftn}
The nucleation rate in the field-theoretic formulation [@Langer_1969] begins from the relation that links the steady-state nucleation rate with the imaginary part of the free energy $$\label{eq:nr_gen}
J = \frac{\beta\kappa}{\pi}\,\Im\big[F(\mu_e)\big],$$ where $\kappa$ is a kinetic factor related to the diffusivity via $\kappa=(d-1)D_n\ell/R_c^3$. Relation (\[eq:nr\_gen\]) has been evaluated by various investigators and more carefully by Günther, Nicole and Wallace [@Gunther_Nicole_Wallace_1980], which for $d=3$ takes the form (Appendix \[sec:appendb\]) $$J = \mathcal{B}\frac{D_nx_0^6}{\ell^5}\left( \frac{x}{x_o} \right)^{2/3}
\exp\big[{- \left( \frac{x_o}{x}\right)^2\big]},
\label{eq:nr_gnw}$$ where $\mathcal{B}=6^5/(288\pi\sqrt{3})$. Equation (\[eq:nr\_gnw\]) is strictly valid for low supersaturations, i.e., when $x/x_0 << 1$. Langer and Schwartz [@Langer_Schwartz_1980] have heuristically extended this expression for large values of $x$ according to $$J = \mathcal{B}\frac{D_nx_0^6}{\ell^5}\left( \frac{x}{x_o} \right)^{2/3}
\left( 1+\frac{x}{x_o}\right)^{3.55}
\exp{\big[- \left( \frac{x_o}{x}\right)^2\big]}.
\label{eq:nr_ls}$$
The behavior of the aforementioned models for the nucleation rate, Eqs. (\[eq:nr\_bdz\]), (\[eq:nr\_gnw\]) and (\[eq:nr\_ls\]) are illustrated in Fig. \[fig:nucrate\] for $d=3$, where the scaled nucleation rate $K$ is plotted against the scaled supersaturation $x/x_0$ (Eq. (\[eq:nr\_bdz\]) is scaled by $\mathcal{A}D_nx_0/\ell^{d+2}$, while Eqs. (\[eq:nr\_gnw\]) and (\[eq:nr\_ls\]) are scaled by $ \mathcal{B}D_nx_0^6/\ell^5$).
![The scaled nucleation rate vs. scaled supersaturation for $d=3$, comparing predictions of Eq. (\[eq:nr\_bdz\]), the classical model, Eq. (\[eq:nr\_gnw\]), the GNW model, and Eq. (\[eq:nr\_ls\]), the LS model. The scaling factors used are defined in the text[]{data-label="fig:nucrate"}](fig/NucRate){width="60.00000%"}
Growth and Coarsening
---------------------
### Theory {#sec:gct}
The details of the Langer and Schwartz theory have been clearly described in their original work [@Langer_Schwartz_1980] and reviewed in a number of publications ([@Wagner_Kampmann_1991], [@Gunton_et_al_1983], [@Aaronson_LeGoues_1992], [@Ratke_Voorhees_2002]). We only summarize its basic ingredients in order to fix notation. The basic equations of the Langer-Schwartz theory are those of the Lifshitz and Slyozov [@Lifshitz_Slyozov_1961], supplemented by a source of droplets (spherical precipitates) which describes the nucleation event. The continuity equation for the droplet distribution function $n=n(R,t)$ is $$\label{eq:lasw-1}
\frac{\partial n}{\partial t}+\frac{\partial}{\partial R}
\Big [\frac{dR}{dt}n\Big]=j(R),$$ where $j(R)$ is the distributed nucleation rate of precipitates, $
J=\int_{R_c}^\infty j(R)dR$, and as before $J$ stands for the nucleation rate per unit volume. Langer and Schwartz made a simplifying assumption that only precipitates with $R>R_c$ are to be counted as part of the second phase. The SPP number density $N$ is given by $ N=\int_{R_c}^\infty n(R)dR$ with their mean radius defined as $ \overline{R}=\frac{1}{N}\int_{R_c}^\infty n(R)RdR$.
The growth rate of the spherical nucleus is controlled by the rate the solute atoms supplied to the precipitate/matrix interface via diffusion [@Lifshitz_Slyozov_1961] $$\frac{dR}{dt} = \frac{D}{R} \left( \frac{C(t)-C_R}{C_p-C_R} \right).
\label{eq:Growth_rate}$$ Here, $D$ is the diffusivity of the solute in the matrix, $C$ the matrix average solute concentration, $C_p$ the solute concentration within the precipitate, assumed to be uniform, and $C_R$ the local solute concentration at the precipitate/matrix interface. The latter concentration is related to the solubility limit $C_{\infty}$, a temperature dependent quantity, and the curvature of the precipitate/matrix interface through the Gibbs-Thomson relation $C_R =
C_{\infty} \exp{(\ell/R)}$. Also, the equation for conservation of matter must be satisfied, viz., $$\frac{C-C_0}{C-C_p} = \frac{4\pi}{3} \overline{R}^3N(t).
\label{eq:conserve}$$
For non-isothermal conditions, where temperature is varying with time, all the temperature dependent parameters appearing in the foregoing equations must be considered and treated properly. The resulting Langer-Schwartz differential equations for $N$ and $\overline{R}$, applicable to both isothermal and non-isothermal conditions, can be expressed by $$\frac{dN}{dt} = \frac{g_1-g_1f_2+g_3f_1}{1-f_2-g_2g_3f_3}; \quad
\frac{d\overline{R}}{dt} = \frac{f_1 -f_1g_2+f_3g_1}{1-g_2-f_2f_3g_3},
\label{eq:dNdt_final}$$ where the functions $g_1, g_2, g_3$ and $f_1, f_2, f_3$ are $$\begin{aligned}
g_1 & = & J + \frac{ b \ell N }{ x (\overline{R}-R_c) }
\left( \frac{1}{T} - \frac{1}{x}
\frac{d \ln{C_{\infty}}}{d T} \right)
\frac{dT}{dt},
\label{eq:g_1} \\
g_2 & = & \frac{ -N \overline{R} }{ 3(\overline{R}-R_c) }\mathit{\Gamma},
\qquad g_3 = \frac{ -N^2 }{ (\overline{R}-R_c) } \mathit{\Gamma},
\label{eq:g_2}
$$ $$\begin{aligned}
f_1 & = & \frac{D}{\overline{R}}
\left( \frac{ e^x C_\infty-C_{\overline{R}}}{C_p-C_{\overline{R}}} \right) +
\frac{J}{N} \left( R_c + \delta R_c - \overline{R} \right) + \frac{b\ell}{x} \left( \frac{1}{x}
\frac{d\ln{C_{\infty}}}{d T} - \frac{1}{T} \right) \frac{dT}{dt},
\label{eq:f_1} \\
f_2 & = & N \mathit{\Gamma},
\quad f_3 = \frac{\overline{R}}{3} \mathit{\Gamma}, \quad
\mathit{\Gamma} = \frac{4\pi b \ell \overline{R}^2(C_p-C_0)}
{C_\infty e^x x^2 \big(1-\frac{4\pi\overline{R}^3}{3}N\big)^2}.
\label{eq:f_2}
$$ In the above expressions, $\delta R_c$ is the width of the size distribution and $b=0.317014$ for $\overline{R} \le 3R_c/2$ and $b=0$ otherwise [@Langer_Schwartz_1980]. In computations, we assume $\delta R_c= a \ell$ with $a$, a positive constant $a<1$, taken as a free parameter. It is noted that the particular value of $b$ set by Langer and Schwartz reproduces the familiar Lifshitz-Slyozov coarsening law “$t^{1/3}$” as $t \rightarrow \infty$ [@Langer_Schwartz_1980; @Lifshitz_Slyozov_1961]. The initial conditions for the equations in (\[eq:dNdt\_final\]) are $$\begin{aligned}
N(t=t_o) & = & N_o, \label{eq:N_o} \\
\overline{R}(t=t_o) & = & R_c + \delta R_c = \frac{\ell}{x} + a\ell, \label{eq:R_o} \end{aligned}$$ where the starting time $t_o$ is determined from $N_o = \int_{0}^{t_o}
J(\tau) d\tau$; and $N_o$ defines the lowest particle density of practical interest, here treated as a model parameter.
### Numerical Method
The nonlinear ordinary differential equations (\[eq:dNdt\_final\]) need to be evaluated numerically. In evaluations of the right-hand-side expressions of equations in (\[eq:dNdt\_final\]), we make use of correlations for the solubility limit $C_{\infty}$ and the diffusivity $D$, which are temperature-dependent material properties. We have used the MATLAB programming environment to solve these differential equations; more specifically the MATLAB solver [@Matlab]. This is a variable order solver, which is intended to solve stiff systems of ordinary differential equations. The solver is invoked at each time step of the time-temperature history, and the evolution of $N$ and $\overline{R}$ during the time step is computed with ODE15S for each mesh point, based on the local temperature at beginning and end of the time step. In the solution procedure, the temperature is assumed to vary linearly with time during the time step.
Application {#sec:exp}
===========
Isothermal experiment {#sec:iso}
---------------------
The methods outlined in the foregoing section have been verified [@Jernkvist_2005] against a number of measurements made in isothermal conditions for binary precipitates such $\gamma '$-Ni$_3$Al in Ni-14 at.$\%$Al and Co in Cu-Co alloys [@Wendt_Haasen_1983; @LeGoues_Aaronson_1984; @Wendt_Haasen_1985; @Haasen_Piller_1987]. Here, we only report the results of our computations for cobalt precipitates in Cu-2.7 at.%Co alloy, for which isothermal annealing tests (at 823 K) were carried out by Wendth and Haasen [@Wendt_Haasen_1985]. The considered experiment involved measurements of the SPP number density as a function of annealing time and also the time variation of precipitate mean size was determined. These properties were measured by transmission electron microscopy (TEM) and atom probe field ion microscopy.
In order to simulate this test, we used the solubility limit for cobalt in copper proposed by Servi and Turnbull [@Servi_Turnbull] $C_{\infty} = 712.85 \times 10^{-2875/T}$, where $T$ is the absolute temperature and $C_{\infty}$ is in atomic percent. For the diffusivity of cobalt in copper, we used the correlation by Döhl *et al.* [@Dohl_etal], $D = 4.3 \times 10^{-5} e^{-25738/T}$, where $D$ is in m$^2$s$^{-1}$. The matrix/SPP interface energy $\sigma$ was in our analyses set to 0.22 Jm$^{-2}$, following the evaluation of data by Stowell [@Stowell]. The second phase particles in these alloys are of pure cobalt, which means that $C_p$ is 100 at%. Both the particles and the matrix have a face-centered cubic (fcc) crystal structure. Moreover, the molar volume $v_m$ of cobalt is $6.7 \times
10^{-6}$ m$^{3}$mole$^{-1}$. In analyses, we have set the model parameter $N_o$ in Eq. (\[eq:N\_o\]) to $1 \times 10^{10}$ m$^{-3}$ and $a$ in Eq. (\[eq:R\_o\]) to unity.
![Calculated number density of Co second phase particles in Cu-2.7at%Co at 823 K, in comparison with data from [@Wendt_Haasen_1985]. The calculated nucleation rate, given by Eq. (\[eq:nr\_ls\]), is scaled by a factor 0.05 in order to fit the data.[]{data-label="fig:CuCo_T823_Nnom"}](fig/CuCo_T823_Nnom){width="60.00000%"}
The calculated evolution of SPP number densities are compared with data in Fig. \[fig:CuCo\_T823\_Nnom\]. The results are shown for two cases: For the nominal case, the nucleation rate $J$, given by Eq. (\[eq:nr\_ls\]), has been used. For the scaled case, the coefficient $\mathcal{B}$ in Eq. (\[eq:nr\_ls\]) has been scaled by a factor 0.05, in order to obtain a best fit to the data. For this test at 823 K, shown in Fig. \[fig:CuCo\_T823\_Nnom\], the measured SPP density reaches a peak at about 300 seconds, after which coarsening starts to dominate the picture. At this stage, large particles will grow at the expense of smaller ones, as a consequence of the Gibbs-Thomson relation, see section \[sec:gct\]. The smallest particles will ultimately dissolve, which makes the number density decrease with a rate proportional to $\approx 1/t$. Our model captures this behavior fairly well, but the calculated peak in the SPP density is reached too early, in comparison with data. For Cu-1.0 at%Co alloy, LeGoues and Aaronson [@LeGoues_Aaronson_1984] reported a delay time of about 130 s in their study, whereas Haasen and Piller [@Haasen_Piller_1987] reported that the delay time was no more than 30 s in the study made by Al-Kasab. Since we have utilized a steady-state theory and assumed a homogeneous nucleation rate, we could have overlooked the effect of the delay time, $t_d$, and hence get a discrepancy between the observed temporal data and calculated values; cf. Appendix \[sec:appendc\] for a rough estimation of this effect.
Non-isothermal simulation {#sec:noniso}
-------------------------
In this subsection, we present the results of a simulation of a heat treatment made on a cylindrical specimen. The cylinder is made of Zircaloy-2 (Zr-1.4Sn-0.12Fe-0.09Cr by wt.%) and has a diameter of 25 mm. The heat treatment consists of quenching the sample from a high temperature (1323 K) body-centered cubic $\beta$ phase in water, which is kept at room temperature. After the quenching, the cylinder is subjected to two subsequent annealing steps in Ar gas at 838 K (hexagonal-closed packed $\alpha$ phase) for durations of 1.0 h and 1.5 h, respectively (Fig. \[fig:t-hist\]).
![Temperature history for the annealing steps; the cooling and heating rates for the steps are $\pm 5.4$ Ks$^{-1}$.[]{data-label="fig:t-hist"}](fig/T_hist){width="60.00000%"}
The prominent second phase precipitate in Zircaloys is Zr(Fe,Cr)$_2$ (Laves phase with hexagonal structure) formed at around 1200 K [@Massih_et_al_2003]. Here, we calculate the nucleation and growth of this kind of precipitates under the aforementioned heat treatment (quenching plus annealing) using the models described in the foregoing section, which are implemented in a computer program [@Jernkvist_Massih_2004]. In particular, we have selected the nucleation rate equation (\[eq:nr\_ls\]) in our evaluation. Since the diffusivity of Fe and Cr are similar and solubility data (temperature vs. concentration) are available for (Fe+Cr) in Zircaloy [@Massih_et_al_2003], we have treated the precipitate as a binary compound Zr and (Fe+Cr) with initial solute (Fe+Cr) concentration of $C_0=2010$ wppm (weight parts per million). The SPP molar volume is $v_m=9 \times 10^{-6}$ m$^{3}$mole$^{-1}$ and its composition $C_p =
5.4\times 10^5$ wppm. We have assumed that the diffusivity during nucleation is identical to that during growth. The input model parameters in our evaluations are as follows. We set the effective diffusivity for (Fe+Cr) in Zircaloy as $D(T) = 1.473 \times 10^{-6}
e^{-15930/T}$ , where $T$ is the absolute temperature and $D$ is in m$^2$s$^{-1}$, and the surface tension of the precipitate $\sigma=0.25$ Jm$^{-2}$. The lower cut-off limit for the particle density, $N_o$ in Eq. (\[eq:N\_o\]), is set to $1 \times 10^{10}$ m$^{-3}$, and the model parameter $a$ in Eq. (\[eq:R\_o\]) to 0.25. The motivations for selecting these values are discussed in [@Jernkvist_2005].
The results of our non-isothermal evaluation are presented in a number of diagrams in Fig. \[fig:spp-evol\], which shows (i) the mean precipitate radius, (ii) the precipitate number density, (iii) nucleation rate, (iv) the matrix supersaturation, (v) the matrix solute concentration (Fe+Cr), and (vi) temperature evolution in the first ten seconds of the heat treatment. The computation output corresponds to the center of the cylinder and 0.1 mm from the outer surface. The cooling rates at these locations, during quenching, can be noted from the temperature vs. time diagram in Fig. \[fig:spp-evol\].
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![Evolution of Zr(Fe,Cr)$_2$ SPPs in a Zircaloy-2 cylinder during the heat treatment. Diagrams, left to right, top to bottom, show calculated (i) mean precipitate radius, (ii) precipitate number density, (iii) nucleation rate, (iv) matrix supersaturation, (v) matrix solute concentration (Fe+Cr), and (vi) temperature in the first ten seconds.[]{data-label="fig:spp-evol"}](fig/Radius "fig:"){width="50.00000%"} ![Evolution of Zr(Fe,Cr)$_2$ SPPs in a Zircaloy-2 cylinder during the heat treatment. Diagrams, left to right, top to bottom, show calculated (i) mean precipitate radius, (ii) precipitate number density, (iii) nucleation rate, (iv) matrix supersaturation, (v) matrix solute concentration (Fe+Cr), and (vi) temperature in the first ten seconds.[]{data-label="fig:spp-evol"}](fig/Density "fig:"){width="50.00000%"}
![Evolution of Zr(Fe,Cr)$_2$ SPPs in a Zircaloy-2 cylinder during the heat treatment. Diagrams, left to right, top to bottom, show calculated (i) mean precipitate radius, (ii) precipitate number density, (iii) nucleation rate, (iv) matrix supersaturation, (v) matrix solute concentration (Fe+Cr), and (vi) temperature in the first ten seconds.[]{data-label="fig:spp-evol"}](fig/Nucleation "fig:"){width="50.00000%"} ![Evolution of Zr(Fe,Cr)$_2$ SPPs in a Zircaloy-2 cylinder during the heat treatment. Diagrams, left to right, top to bottom, show calculated (i) mean precipitate radius, (ii) precipitate number density, (iii) nucleation rate, (iv) matrix supersaturation, (v) matrix solute concentration (Fe+Cr), and (vi) temperature in the first ten seconds.[]{data-label="fig:spp-evol"}](fig/Supersat "fig:"){width="50.00000%"}
![Evolution of Zr(Fe,Cr)$_2$ SPPs in a Zircaloy-2 cylinder during the heat treatment. Diagrams, left to right, top to bottom, show calculated (i) mean precipitate radius, (ii) precipitate number density, (iii) nucleation rate, (iv) matrix supersaturation, (v) matrix solute concentration (Fe+Cr), and (vi) temperature in the first ten seconds.[]{data-label="fig:spp-evol"}](fig/Solute "fig:"){width="50.00000%"} ![Evolution of Zr(Fe,Cr)$_2$ SPPs in a Zircaloy-2 cylinder during the heat treatment. Diagrams, left to right, top to bottom, show calculated (i) mean precipitate radius, (ii) precipitate number density, (iii) nucleation rate, (iv) matrix supersaturation, (v) matrix solute concentration (Fe+Cr), and (vi) temperature in the first ten seconds.[]{data-label="fig:spp-evol"}](fig/Temperature "fig:"){width="50.00000%"}
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Discussion {#sec:disc}
----------
From the results displayed in Fig. \[fig:spp-evol\], we note that there is a significant difference in calculated precipitate radius at the different positions in the cylinder, directly upon $\beta$ quenching, but this difference gradually diminishes under subsequent annealing. The initial difference in the radius of the precipitates located in the cylinder central region and those close to the outer surface follows from the difference in the cooling rate, as is seen in the temperature diagram in Fig. \[fig:spp-evol\]. According to our computations, the slow cooling at the cylinder central part results in appreciable precipitate growth already under the quenching phase. This behavior is not predicted at the billet outer surface.
The calculated time variation of the precipitate number density shown in Fig. \[fig:spp-evol\] displays a remarkable difference between the central and periphery of the cylinder directly on quenching. The calculated number density at the outer surface is initially about several orders of magnitude larger than that at the central part, but this large difference gradually disappears under subsequent annealing as a result of coarsening. Figure \[fig:spp-evol\] also shows the calculated nucleation rate as a function of time, evaluated at the considered locations in the cylinder. As can be seen, the nucleation takes place under a very short time span, as the temperature drops below 1118 K and the matrix becomes supersaturated ($C>C_{\infty}$). The nucleation rate pulse width, as can be seen in the diagram, is a couple of ms at the billet outer surface, whereas the corresponding time span is much larger in the central part, i.e., around 50 ms. The calculated matrix supersaturation, $x = \ln{(C/C_\infty)}$, is also shown as a function of time in the figure. The spikes seen in the diagram for $x$ correspond to the temperature drops during the annealing cycles (Fig. \[fig:t-hist\]), which reduce $C_\infty$. Note also the sharp dive in the matrix solute concentration due the temperature dip. The results of the computations presented here, i.e., the precipitate radius and number density, are in qualitative agreement with the observations made on a similar kind of heat treatment of this material [@Massih_etal_CS].
As alluded in section \[sec:iso\], we have supposed a steady-state nucleation model. Such a treatment does not provide information regarding instantaneous SPP size distribution nor on the nucleation rate prior to reaching steady state. Some workers postulate a nucleation relation in the form $J(t)=J(\infty)\exp{(-t_d/t)}$ to fit metallurgical data [@Wagner_Kampmann_1991]. However, this type of approach is simplistic and a more rigorous treatment to solve the time-dependent Fokker-Plank equation (\[eq:FPE\]) [@Kashchiev_1999] or extend the field theoretical approach (section \[sec:iso\]) to the realm of phase transition dynamics can be more expedient. Besides, nucleation does not commonly occur homogeneously in a matrix by means of solely thermal and concentration fluctuations. It is induced most often by heterogeneities and defects in solids, e.g., grain and interphase boundaries, dislocations, stacking faults, free surfaces and vacancies or their clusters. These micro-domains facilitate or enhance the nucleation rate for formation of a new phase during quenching [@Porter_Easterling_1981].
Conclusion {#sec:concl}
==========
The equations for nucleation, growth and coarsening of second phase precipitates are extended to account for the non-isothermal situations. The results of our simulation of the heat treatment of a Zircaloy-2 specimen clearly illustrates the influence of cooling rate during quenching on the properties of SPPs. A lower cooling rate results in larger precipitates with a smaller number density. This calculation is in qualitative agreement with the observations made on a similar kind of heat treatment of this material [@Massih_etal_CS], which also indicate that SPP size and density are quenching rate dependent and they impact macroscopic properties of the alloy.
[^1]: This is an expanded version of the paper presented in the International Conference on Solid-Solid Phase Transformation in Inorganic Materials 2005 (PTM 2005), May 29 - June 3, 2005, Phoenix, Arizona, USA.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We study the weak decays of $\bar{B}_{(s)}$ and $B_c$ into $D$-wave heavy-light mesons, including $J^P=2^-$ ($D_{(s)2},D''_{(s)2},B_{(s)2}, B''_{(s)2}$) and $3^-$ ($D^*_{(s)3}, B^*_{(s)3}$) states. The weak decay hadronic matrix elements are achieved based on the instantaneous Bethe-Salpeter method. The branching ratios for $\bar{B}$ decays are $\mathcal{B}[\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}D_{2}e\bar{\nu}_e] = 1.1^{-0.3}_{+0.3} \times 10^{-3}$, $\mathcal{B}[\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}D''_2e\bar{\nu}_e]=4.1^{-0.8}_{+0.9} \times 10^{-4}$, and $\mathcal{B}[\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}D^*_3e\bar{\nu}_e]=1.0^{-0.2}_{+0.2} \times 10^{-3}$, respectively. For semi-electronic decays of $\bar B_s$ to $D_{s2}$, $D''_{s2}$, and $D^*_{s3}$, the corresponding branching ratios are $1.7^{-0.5}_{+0.5}\times 10^{-3}$, $5.2^{-1.5}_{+1.6}\times 10^{-4}$, and $1.5^{-0.4}_{+0.4}\times 10^{-3}$, respectively. The branching ratios of semi-electronic decays of $B_c$ to $D$-wave $D$ mesons are in the order of $10^{-5}$. We also achieved the forward-backward asymmetry, angular spectra, and lepton momentum spectra. In particular the distribution of decay widths for $2^-$ states $D_2$ and $D''_2$ varying along with mixing angle are presented.'
address: 'Harbin Institute of Technology, Harbin, 150001, P. R. China'
author:
- Qiang Li
- Tianhong Wang
- Yue Jiang
- Han Yuan
- Tian Zhou
- 'Guo-Li Wang'
title: 'Decays of $B$, $B_s$ and $B_c$ to $D$-wave heavy-light mesons'
---
Introduction
============
The $D$-wave $D_{(s)}$ mesons have attracted lots of attention since numerous excited charmed states are discovered by [BaBar]{} [@BaBar-2010], and LHCb [@LHCb1; @LHCb2; @LHCb3; @LHCb4]. In 2010 [BaBar]{} observed four signals $D(2550)^0$, $D^*(2600)^0$, $D(2750)^0$, and $D^*(2760)^0$ for the first time [@BaBar-2010], where the last two are expected to lie in the mass region of four $D$-wave charm mesons [@PRD32-1985]. Later the LHCb reported two natural parity resonances $D^*_J(2650)^0$ and $D^*_J(2760)^0$ in the $D^{*+}\pi^-$ mass spectrum and measured their angular distribution [@LHCb1]. The same final states also show the presence of two unnatural parity states, $D_J(2580)^0$ and $D_J(2740)^0$. Here the natural parity denotes states with $J^P=0^+, ~1^-, ~2^+, ~3^-, \dots$ with $P=(-1)^J$, while the unnatural parity indicates series with $J^P=0^-,~1^+,~2^-,\cdots$.
Then in May 2015, LHCb confirmed that the $D^*_J(2760)^0$ resonance has spin 1 [@LHCb3]. The mass and width are measured as $m[D^*_1(2760)^0]=2781\pm22$ and $\Gamma[D^*_1(2760)^0]=177\pm38$ , where we have combined the statistical and systematic uncertainties in quadrature for simplicity. Later LHCb determined $D^*_J(2760)^-$ to have spin-parity $3^-$ and it is interpreted as $D^*_3(2760)^-$, namely the $^3\!D_3$ $\bar{c}d$ state. The mass and width are measured as $m[D^*_3(2760)]=2798\pm10$ and $\Gamma[D^*_3(2760)]=105\pm30$ [@LHCb4]. For the $D$-wave charm-strange meson, BarBar first observed the $D^*_{sJ}(2860)$ [@BaBar2; @BaBar3]. And then LHCb’s results support that $D^*_{sJ}(2860)$ is an admixture of the spin-1 and spin-3 [@LHCb5; @LHCb6]. The measured mass and width for $D^*_{s3}$ are $2861\pm7$ and $53\pm10$ , respectively. The two $D$-wave charm-strange mesons with $J=2$, namely the $2^-$ states $D_{s2}$ and $D'_{s2}$ are still undiscovered in experiment.
Identification of these new excited charmed mesons can be found in Refs. [@Col-2006; @Col-2012; @PRD82-2010-1; @PRD82-2010-2; @PRD83-2011-1; @PRD83-2011-2; @PRD90-2014; @PRD92-2015; @CPC39-2015; @God-2014; @PRD93-2016]. We will follow Godfrey’s assignments on $D$-wave $D^{(*)}_{(s)J}$ mesons in Ref. [@PRD93-2016], where $D^*_{s3}(2860)$ is identified as $1^3\!D_3$ $c\bar{s}$; $D^*_3(2798)^0$ is identified as $1^3\!D_3(c\bar{q})$ state; $D^*_1(2760)^0$ is interpreted as $1^3\!D_1(c\bar{q})$; and the $D_J(2750)^0$ reported by BaBar and ${D_J(2740)^0}$ reported by LHCb are identified as the same state with $1D_2(c\bar{q})$, where $q$ denotes a light quark $u$ or $d$.
These $D$-wave excited states still need more experimental data to be discovered or confirmed. The identification and spin-parity assignments in above literature are just tentative. As the LHC accumulates more and more data, the study of these $D$-wave charm and charm-strange mesons in the weak decay of $B_{(s)}$ and $B_c$ meson becomes necessary and important. The properties of $D^{(*)}_{(s)J}$ in $B_{(s)}$ and $B_c$ decays would be helpful in identification of these excited $D_{(s)}$ mesons. The semi-leptonic decays of $B_{(s)}$ into $D$-wave charmed mesons have been studied by QCD sum rules [@PLB478-2000; @PRD79-2009; @Gan-2015] and constituent quark models in the framework of heavy quark effective theory (HQET) [@PTP91-1994; @PRD54-1996]. Most of previous work is based on the HQET. The systematic studies on weak decays of $\bar B_{(s)}$ into $D$-wave $D_{(s)2}$, $D'_{(s)2}$ and $D^*_{(s)3}$ are still quite less while all these $D$-wave charmed mesons are hopefully to be detected in the near future experiments. On the other hand, in 2012 the BaBar Collaboration reported the ratio of $\mathcal{B}(\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D^{(*)}\tau^-\bar \nu_\tau)$ relative to $\mathcal{B}(\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D^{(*)}e^-\bar \nu_e)$, which exceed the standard model expectation by $2\sigma~(2.7\sigma)$ [@BaBar-2012] and may hint the new physics. We also noticed that in the very recently Belle measurement[@Belle-2016], the experimental results on this quantity are consistent with the theoretical predictions[@Fajfer-2012; @Tanaka-2013; @Xiao-2015] in the framework of the Standard Model. Anyway, it is still necessary and helpful to investigate these ratios for $\bar B_{(s)}$ and $B_c$ decays into higher excited $D_{(s)}$ mesons.
In this work we will concentrate on the semi-leptonic and non-leptonic decays of $\bar{B}$ ($\bar{B}_s, B_c$) into $D$-wave $D$ ($D_s$) meson, including $2^- ~(D_{(s)2}, D'_{(s)2})$ and $3^-~(D^*_{(s)3})$ states. For completeness, the weak decays of $B_c$ to $D$-wave bottomed mesons are also studied. We use the Instantaneous Bethe-Salpeter equation (IBS) [@PR87-1952] to get the hadronic transition form factors. BS equation [@PR84-1951] is the relativistic two-body bound states formula. Based on our previous studies [@NPP39-2012; @JPG40-2013; @JHEP09-2015; @QLi-2016], the relativistic corrections for transitions of higher excited states are larger and more important than that for the ground states, so the relativistic method is more reliable for the processes involved the high excited states. In the instantaneous approximation of the interaction kernel, we can achieve the Salpeter equation. The Salpeter method has been widely used to deal with heavy mesons’ decay constants calculation [@PLB633-2006; @Guo-2007], annihilation rate [@PLB674-2009; @JHEP03-2013], and hadronic transition [@NPP39-2012; @JPG40-2013; @JHEP09-2015; @QLi-2016].
This paper is organized as follows: first we present the general formalism of semi-leptonic and non-leptonic decay for $\bar B_{(s)}$ meson, including decay width, forward-backward asymmetry, and lepton spectra. In we compute the form factors in hadronic transition by Salpeter method. In we give the numerical results and discussions. Finally we give a short summary of this work.
Formalism of semi-leptonic and non-leptonic decays {#Sec-2}
==================================================
In this section, firstly we will derive the formalism of transition amplitudes for $\bar{B}_{(s)}$ to $D$-wave heavy-light mesons. Then the formalisms of interested observables are presented. We will take the $\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D^{(*)}_{J}$ transition as an example to show the calculation details, while results for transition of $B_s$ and $B_c$ will be given directly.
Semi-leptonic decay amplitude
-----------------------------
The Feynman diagram responsible for $\bar{B}$ semi-leptonic decay is showed in , where we use $P$ and $P_F$ to denote the momenta of $\bar{B}$ and $D^{(*)}_{J}$ respectively.
![Feynman diagram for semi-leptonic decay of $\bar{B}$ to $D^{(*)}_{J}$ $(J=2, 3)$. $m_i(m'_i)$ and $p_i(p_i^\prime)$ are the constituent quark mass and momentum for initial (final) state, respectively.[]{data-label="Fig-semi"}](bd-semi){width="55.00000%"}
The transition amplitude $\mathcal{A}$ for the process $\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D^{(*)}_J\ell\bar\nu$ can be written directly as $$\label{Amp}
\mathcal{A}=\frac{G_{\!F}}{\sqrt{2}}V_{cb} l^\mu \langle D^{(*)}_{J}|J_{\mu}|\bar{B}\rangle.$$ In above equation, $G_{\!F}$ denotes the Fermi weak coupling constant; $V_{cb}$ is the Cabibbo-Kobayashi-Maskawa matrix element; the lepton matrix element $l^\mu$ reads $$\begin{gathered}
l^\mu=\bar{u}(p_\ell)\Gamma^\mu v(p_\nu),\end{gathered}$$ where $\ell$ ($\bar{\nu}_\ell$) denotes the charged lepton (anti-neutrino), and $p_\ell$($p_\nu$) denotes the corresponding momentum, and the definition $\Gamma^\mu=\gamma^\mu(1-\gamma^5)$ is used; $\langle D^{(*)}_{J}|J_{\mu}|\bar{B}\rangle$ is the hadronic transition element, where $J_\mu=\bar{c}\Gamma_\mu b$ is the weak current.
We use $\mathcal{M}^\mu$ to denote hadronic transition element $\langle D^{(*)}_J|J^{\mu}|\bar{B}\rangle$, which can be described with form factors. The general form of the hadronic matrix element depends on the total angular momentum $J$ of the final meson. For $D_2~(D'_2)$ and $D^*_3$ the form factors are defined as $$\label{form}
\mathcal{M}^\mu=
\begin{cases}
e_{\alpha\beta}P^{\alpha}(s_1P^{\beta}P^{\mu}+s_2P^{\beta}P_F^{\mu}+s_3g^{\beta\mu}+\mathrm{i}s_4\epsilon^{\mu\beta PP_F}) & \text{if}~J=2,\\
e_{\alpha\beta\gamma}P^{\alpha}P^{\beta}(h_1P^{\gamma}P^{\mu}+h_2P^{\gamma}P_F^{\mu}+h_3
g^{\gamma\mu}+\mathrm{i}h_4\epsilon^{\mu \gamma PP_F}) & \text{if}~J=3.
\end{cases}$$ In above equation, we used the definition $\epsilon_{\mu \nu P P_F}=\epsilon_{\mu \nu \alpha \beta }P^{\alpha} P^{\beta}_F$ where $\epsilon_{\mu \nu \alpha \beta}$ is the totally antisymmetric Levi-Civita tensor; $g^{\mu\nu}$ is the Minkowski metric tensor; $e_{\alpha\beta}$ and $e_{\alpha\beta\gamma}$ are the polarization tensor for $J=2$ and 3 mesons, respectively, which are completely symmetric; $s_i$ and $h_i~(i=1,2,3,4)$ are the form factors for $J=2$ and 3, respectively. To state it more clearly, we will use $s_i$, $t_i$, and $h_i$ to denote the form factors for transitions $\bar B$ to $D_2$, $D'_2$, and $D^*_3$, respectively. $S_i$, $T_i$, and $H_i$ are used to represent the form factors of $B^-_c$ to $\bar D_2$, $ D'_2$, and $ D^*_3$, respectively. The definition forms are the same with that for transition $\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D^{(*)}_J$, just $s_i$ is replaced by $S_i$, $t_i$ by $T_i$, and $h_i$ by $H_i$. For $\bar B_s$ decays, the corresponding form factor behaviors are very similar to $\bar B$ decays. The detailed calculations of these form factors will be given in next section.
After summing the polarization of all the final states, including the charged lepton, anti-neutrino and the final $D^{(*)}_J$, we obtain $$\label{Amp2}
\overline{|\mathcal{A}|^2} =\frac{G_F^2}{2} |V_{cb}|^2 L^{\mu \nu} H_{\mu\nu} ,$$ where the lepton tensor $L^{\mu \nu}$ has the following form $$\label{T-L}
L^{\mu \nu}= 8(p_\ell^{\mu}p_\nu^{\nu}+p_\nu^{\mu}p_\ell^{\nu}-p_\ell \! \cdot \! p_\nu g^{\mu \nu}-\mathrm{i}\epsilon^{\mu \nu p_\ell p_\nu}).$$ $H_{\mu \nu}$ is the hadronic tensor describing the propagator-meson interaction vertex, which depends on $P$, $P_F$ and the corresponding form factors. It can be written as $$\begin{gathered}
\label{eq-HF}
H_{\mu \nu}=\sum_{s=-J}^{J} \mathcal{M}_\mu^{(s)}\mathcal{M}^{*(s)}_\nu= N_{1}P_{\mu}P_{\nu}+N_2(P_{\mu}{P_F}_\nu+P_{\nu}{P_F}_\mu)+N_4P_{F\mu}P_{F\nu}+
N_5g_{\mu \nu}+\mathrm{i} N_6 \epsilon_{\mu \nu P P_F},\end{gathered}$$ where the summation is over the polarization of final $D^{(*)}_J$ meson; $N_i$ is related to the form factors $s_i$ for $D_2$, $t_i$ for $D'_2$ or $h_i$ for $D^*_3$. The detailed expressions for $N_i$ can be found in appendix \[Ni\].
Non-leptonic decay amplitude
----------------------------
The Feynman diagram for the non-leptonic decay of $\bar B$ to $D^{(*)}_{J}$ and a light meson $X$ is showed in . As a preliminary study for non-leptonic decays of $\bar B$ to $D$-wave $D$ mesons, we will work in the framework of naive factorization approximation [@Fak-1978; @Cab-1978; @Bau-1987; @Ali-1998], which has been widely used in heavy mesons’ weak decays [@cch1; @Col-2000; @PRD73-2006; @PRD74-2006; @Faustov-2013]. The factorization assumption is expected to hold for process that involves a heavy meson and a light meson, provided the light meson is energetic [@Dugan-1991; @Ben-1999; @Keu-2001]. Also we only consider the processes when the light meson $X$ is $\pi$, $\rho$, $K$, or $K^*$.
In the naive factorization approximation, the decay amplitude can be factorized as the product of two parts, the hadronic transition matrix element and an annihilation matrix element. Then we can write the non-leptonic decay amplitude as $$\label{Amp-non}
\mathcal{A}\big [\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D^{(*)}_{J}X \big ] \simeq \frac{G_F}{\sqrt{2}} V_{cb} V_{uq} a_1(\mu) \langle D^{(*)}_{J}|J_\mu|\bar B \rangle \langle X|(\bar qu )_{V-A}|0\rangle,$$ where we have used the definition $(\bar q u)_{V-A}=\bar{q} \Gamma^\mu u$ and $q$ denotes a $d$ or $s$ quark field; $V_{uq}$ denotes the corresponding CKM matrix element; $a_1=c_1+\frac{1}{N_c}c_2$, where $N_c=3$ is the number of colors. For $b$ decays, we take $\mu=m_b$, and $a_1=1.14$, $a_2=-0.2$ [@PRD73-2006] are used in this work. The annihilation matrix element can be expressed by decay constant as $$\langle X|(\bar qu )_{V-A}|0\rangle=
\begin{cases}
\text{i} P^{\mu}_X f_P~~~~~&X{\rm~is~a~pseudoscalar~meson}~(\pi, K),\\
e^{\mu}M_X f_V &{X\rm ~is~a~vector~meson}~(\rho, K^*).
\end{cases}$$ $M_X$, $P_X$ are the mass and momentum of $X$ meson, respectively; the meson polarization vector $e^\mu$ satisfies $e_\mu P^\mu_X=0$ and the completeness relation is given by $\sum_s e^\mu_{(s)} e^\nu_{(s)}=\frac{P_X^\mu P_X^\nu}{M_X^2}-g^{\mu\nu}$, where $s$ denotes the polarization state; $f_P$ and $f_V$ are the corresponding decay constants.
Then the $\overline{|\mathcal{A}|^2}$ can be expressed by hadronic tensor $H_{\mu\nu}$, which is just the same with that in the corresponding semi-leptonic decays, and light meson tensor $X^{\mu\nu}$ as $$\overline{|\mathcal{A}|^2}=\frac{G_F^2}{2}|V_{cb}|^2|V_{uq}|^2a_1^2H_{\mu\nu}X^{\mu\nu},$$ where $X^{\mu\nu}$ has the following expression $$X^{\mu\nu}=\langle X|(\bar q\Gamma^\mu u )|0\rangle\langle X|\bar q\Gamma^\nu u|0\rangle^*=
\begin{cases}
P^{\mu}_X P^{\nu}_X f^2_P &X{\rm~is~a~pseudoscalar~meson},\\
(P^{\mu}_X P^{\nu}_X- M^2_Xg^{\mu\nu}) f^2_V &X{\rm ~is~a~vector~meson}.
\end{cases}$$
Several observables
-------------------
One of the interested quantity in $\bar B$ semi-leptonic decay is the angular distribution of the decay width $\Gamma$, which can be described as $$\frac{\text{d} \Gamma}{ \mathrm{d}\cos\theta} = \int \frac{1}{(2\uppi)^3} \frac{|\bm{p}^*_\ell||\bm{p}_F^*|}{16M^3} \overline{|\mathcal{A}|^2} \text{d} m^2_{\ell\nu},$$ where $M$ is the initial $\bar B$ mass; $m^2_{\ell\nu}=(p_\ell+p_\nu)^2$ is the invariant mass square of $\ell$ and $\bar\nu$; $\bm{p}^*_\ell$ and $\bm{p}_F^*$ are the three momenta of $\ell$ and $D^{(*)}_{J}$ in the $\ell\bar\nu$ rest frame, respectively; $\theta$ is the angle between $\bm{p}^*_\ell$ and $\bm{p}_F^*$; $|\bm{p}^*_{\ell}|=\frac{1}{2m_{\ell\nu}}\lambda^{\frac{1}{2}}(m^2_{\ell\nu}, M_{\ell}^2, M^2_{\nu})$ and $|\bm{p}_F^*|=\frac{1}{2m_{\ell\nu}}\lambda^{\frac{1}{2}}(m^2_{\ell\nu}, M^2, M^2_F)$, where we have used the Källén function $\lambda(a,b,c)=(a^2+b^2+c^2-2ab-2bc-2ac)$, $M_{\ell}$ and $M_{\nu}$ are the lepton mass and anti-neutrino mass, respectively. Another quantity we are interested is the forward-backward asymmetry $A_{FB}$, which is defined as $$A_{FB} = \frac{\Gamma_{\cos\theta>0}-\Gamma_{\cos\theta<0}}{\Gamma_{\cos\theta>0}+\Gamma_{\cos\theta<0}}.$$ The decay width varying along with charged lepton 3-momentum $|\bm{p}_\ell|$ is given by $$\frac{\text{d} \Gamma}{\text{d} |\bm{p}_\ell|} =\int \frac{1}{(2\uppi)^3} \frac{|\bm{p}_\ell|} {16M^2 E_\ell} \overline{|\mathcal{A}|^2} \text{d} m^2_{\ell\nu} ,$$ where $E_\ell$ denotes the charged lepton energy in the rest frame of initial state meson.
The non-leptonic decay width of the $\bar{B}$ meson is given by $$\label{eq-width2}
\Gamma=\frac{|\bm{p}|}{8\uppi M^2} \overline{|\mathcal{A}|^2},$$ where $\bm{p}$ represents the 3-momentum of the final $D^{(*)}_{J}$ in $\bar B$ rest frame, which is expressed as $|\bm{p}|=\frac{1}{2M}\lambda^{\frac{1}{2}}(M^2, M^2_X, M^2_F)$.
Hadronic transition matrix element {#Sec-3}
==================================
The hadronic transition matrix element $\langle D^{(*)}_{J}|J^{\mu}|\bar{B}\rangle$ plays an key role in the calculations of $\bar B$ semi-leptonic and non-leptonic decays. In this section we will give details to calculate the hadronic transition matrix element by Bethe-Salpeter method in the framework of constituent quark model.
Formalism of hadronic transition matrix element with Bethe-Salpeter method
--------------------------------------------------------------------------
According to the Mandelstam formalism [@Man-1955], the hadronic transition amplitude $\mathcal{M}^\mu$ can be written by Beter-Salpeter (BS) wave function as $$\label{M-0}
\mathcal{M}^\mu=-\mathrm{i} \int \frac{\mathrm{d}^4 q \mathrm{d}^4 q'}{(2 \uppi)^4} \mathrm{Tr} \big[ \bar{\Psi}_D(q',P_F) \Gamma^{\mu} \Psi_B(q,P) (m_2+\slashed {p}_2) \delta^{4}(p_2-p'_2)\big],$$ where $\Psi_B(q,P)$ and $\Psi_D(q',P_F)$ are the BS wave functions of the $\bar B$ meson and the final $D^{(*)}_J$, respectively; $\bar\Psi$ is defined as $\gamma^0\Psi^\dagger\gamma^0$; $q$ and $q'$ are respectively the inner relative momenta of $\bar B$ and $D^{(*)}_J$ system, which are related to the quark (anti-quark) momentum $p_1^{(\prime)}$ ($p_2^{(\prime)}$) by $p_i=\alpha_i P+(-1)^{(i+1)}q$ and $p'_i=\alpha^\prime_i P_F+(-1)^{(i+1)}q^\prime$ ($i=1, 2$). And here we defined the symbols $\alpha_i=\frac{m_i}{m_1+m_2}$ and $\alpha^\prime_i=\frac{m^\prime_i}{m'_1+m'_2}$, where $m_i$ and $m'_i$ are masses of the constituent quarks in the initial and final bound states, respectively (see ). Here in $\bar B$ decays we have $m_1=m_b$, $m'_1=m_c$, $m_2=m'_2=m_d$. As there is a delta function in above equation, the relative momenta $q$ and $q^\prime$ are related by $q '=q-(\alpha_2 P-\alpha '_2 P_F)$.
In the instantaneous approximation [@PR87-1952], the inner interaction kernel between quark and anti-quark in bound state is independent of the time component $q_P(=q\cdot P)$ of $q$. By performing the contour integral on $q_P$ and then we can express the hadronic transition amplitude as [@QLi-2016] $$\label{M}
\mathcal{M}^\mu = \int \frac{ \text{d}^3 {q_\perp}}{(2\uppi)^3} \text{Tr} \bigg [ \frac{\slashed{P}}{M} \bar{\psi}_{D}(q'_\perp) \Gamma^\mu \psi_B(q_\perp) \bigg ],$$ where we have used the definitions $q_\perp\equiv q-\frac{P\cdot q}{M^2}P$ and $q'_\perp\equiv q'-\frac{P\cdot q'}{M^2}P$. Here $\psi$ denotes the 3-dimensional positive Salpeter wave function (see appendix \[Salpeter-EQ\]). $\psi_B$ and $\psi_D$ denote the positive Salpeter wave functions for $\bar B$ and $D^{(*)}_J$, respectively, and $\bar \psi_D$ is defined $\gamma^0 \psi_D^\dag \gamma^0$.
The positive Salpeter wave function for ${^1\!S_0}(0^-)$ state can be written as [@PLB584-2004] $$\label{wave-1s0}
\psi_B(^1\!S_0)=\bigg [ A_1+A_2 \frac{\slashed P}{M} +A_3 \frac{\slashed q_{\perp}}{M}+A_4\frac{\slashed P \slashed q_{\perp}}{M^2} \bigg ]\gamma^5,$$ where we have the following constraint conditions, $$\label{par-1s0}
\begin{aligned}
A_1=& \frac{M}{2}\bigg[\frac{\omega_1+\omega_2}{m_1+m_2}k_1+k_2 \bigg],~
&A_3=&-\frac{M(\omega_1-\omega_2)}{m_1\omega_2+m_2\omega_1}A_1, \\
A_2=&\frac{M}{2}\bigg[k_1+\frac{m_1+m_2}{\omega_1+\omega_2}k_2 \bigg],~
&A_4=&-\frac{M(m_1+m_2)}{m_1\omega_2+m_2\omega_1}A_1.
\end{aligned}$$ The definition $\omega_i\equiv \sqrt{m_i^2-q_\perp^2}~(i=1,2)$ is used. The derivation of Eq.(\[wave-1s0\]) and (\[par-1s0\]) can be found in appendix \[Salpeter-EQ\]. So there are only two undetermined wave function $k_1$ and $k_2$ here, which are the functions of $q_\perp$. The positive Salpeter wave function for $3^-(^3\!D_3)$ state with unequal mass of quark and anti-quark has the following forms [@3D23-wave] $$\label{wave-3d3}
\begin{aligned}
\psi_D(^3\!D_3)&=e_{\mu \nu \alpha} q^{\prime\nu}_{\perp} q_{\perp}^{\prime\alpha} \biggl [ q^{\prime\mu}_{\perp}( n_1+n_2\frac{\slashed P_F}{M_F}+n_3 \frac{\slashed q'_{\perp}}{M_F} +n_4\frac{\slashed P_F \slashed q'_{\perp}}{M_F^2})
+ \gamma^\mu ( n_5 M_F + n_6 \slashed P_F) \\
&\quad+n_7 (\gamma^\mu \slashed{q'}-q^{\prime\mu})+ n_8 \frac{(\gamma^\mu \slashed P_F \slashed q'_{\perp} + \slashed P_F q^{\prime\mu}_{\perp})}{M_F}\bigg].
\end{aligned}$$ In above equation $n_i~(i=1,2,\cdots,8)$ can be expressed with 4 wave functions $u_i~(i=3,4,5,6)$ as below, $$\label{par-3d3}
\begin{aligned}
n_1=& \frac{\big[(\omega'_1+\omega'_2)(q_\perp^{\prime2}u_3+M_F^2u_5)+(m'_1+m'_2)(q_\perp^{\prime2}u_4-M_F^2 u_6)\big]}{2M_F(m'_1 \omega'_2+m'_2 \omega'_1)},\\
n_2=& \frac{\big[(m'_1-m'_2)(q_\perp^{\prime2}u_3+M_F^2u_5)+(\omega'_1-\omega'_2)(q_\perp^{\prime2}u_4-M_F^2 u_6)\big]}{2M_F(m'_1 \omega'_2+m'_2 \omega'_1)},\\
n_3=&\frac{1}{2}\biggl[u_3+\frac{m'_2+m'_2}{\omega'_1+\omega'_2}u_4-\frac{2M_F^2}{m'_1\omega'_2+m'_2\omega'_1}u_6 \biggl],\\
n_4=&\frac{1}{2}\biggl[u_4+\frac{\omega'_1+\omega'_2}{m'_1+m'_2}u_3-\frac{2M_F^2}{m'_1\omega'_2+m'_2\omega'_1}u_5 \biggl],\\
n_5=& \frac{1}{2}\biggl[u_5-\frac{\omega'_1+\omega'_2}{m'_1+m'_2}u_6 \biggl],\quad
n_6= \frac{1}{2}\biggl[u_6-\frac{m'_1+m'_2}{\omega'_1+\omega'_2}u_5\biggl],\\
n_7=& \frac{M_F(\omega'_1-\omega'_2)}{(m'_1\omega'_2+m'_2\omega'_1)}n_5,~~~~\quad
n_8= \frac{M_F(\omega'_1+\omega'_2)}{(m'_1\omega'_2+m'_2\omega'_1)}n_6.
\end{aligned}$$ In above Salpeter positive wave functions $\psi_B$ and $\psi_D$, the undetermined wave functions $k_1$, $k_2$ for $0^-$ and $u_i~(i=3,4,5,6)$ for $3^-$ can be achieved by solving the full Salpeter equations numerically (see appendix \[Salpeter-EQ\]). The positive Salpeter wave functions for $^1\!D_2$ [@JHEP03-2013], and $^3\!D_2$ [@3D23-wave] states can be seen in appendix \[sec-wave\]. $e^{\mu\nu\alpha}$ is the symmetric polarization tensor for spin-3 and satisfies the following relations [@PRD43-1991] $$\begin{gathered}
\label{Polarization3}
e^{\mu\nu\alpha} g_{\mu \nu}=0,\quad
e^{\mu\nu\alpha} P_{F\mu}=0,\\
\sum_s e^{abc}_{(s)}e^{\mu\nu\alpha}_{(s)}=\frac{1}{6}\Omega_1^{abc;\mu\nu\alpha}-\frac{1}{15}\Omega_2^{abc;\mu\nu\alpha},\end{gathered}$$ where $$\label{eq-Polar31}
\begin{aligned}
\Omega_1^{abc;\mu\nu\alpha}&=g_\perp^{a\mu}g_\perp^{b\nu}g_\perp^{c\alpha}+
g_\perp^{a\mu}g_\perp^{b\alpha}g_\perp^{c\nu}+
g_\perp^{a\nu}g_\perp^{b\mu}g_\perp^{c\alpha}
+g_\perp^{a\nu}g_\perp^{b\alpha}g_\perp^{c\mu}+
g_\perp^{a\alpha}g_\perp^{b\mu}g_\perp^{c\nu}+g_\perp^{a\alpha}g_\perp^{b\nu}g_\perp^{c\mu},\\
\Omega_2^{abc;\mu\nu\alpha}&=g_\perp^{ab}g_\perp^{c\mu}g_\perp^{\nu\alpha}+
g_\perp^{ab}g_\perp^{c\nu}g_\perp^{\mu\alpha}+
g_\perp^{ab}g_\perp^{c\alpha}g_\perp^{\mu\nu}
+g_\perp^{ac}g_\perp^{b\mu}g_\perp^{\nu\alpha}+
g_\perp^{ac}g_\perp^{b\nu}g_\perp^{\mu\alpha}
+g_\perp^{ac}g_\perp^{b\alpha}g_\perp^{\mu\nu}\\
&+g_\perp^{bc}g_\perp^{a\mu}g_\perp^{\nu\alpha}+
g_\perp^{bc}g_\perp^{a\nu}g_\perp^{\mu\alpha}+g_\perp^{bc}g_\perp^{a\alpha}g_\perp^{\mu\nu}.\\
\end{aligned}$$ and we have used the definition $g_\perp^{\mu\nu}=-g^{\mu\nu}+\frac{P_F^\mu P_F^\nu}{M_F^2}$.
Inserting the initial $\bar B$ wave function $\psi_B(^1\!S_0)$ (Eq. \[wave-1s0\]) and final $D^{*}_3$ wave function $\psi_D(^3\!D_3)$ (Eq. \[wave-3d3\]) into the hadronic transition amplitude Eq. (\[M\]), after calculating the trace and performing the integral in Eq. (\[M\]) we achieve the form factors $h_i$ for $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}D^{*}_3$ transition defined in Eq. (\[form\]). When performing the integral over $\bm{q}$ in the rest frame of the initial meson, the following formulas are used. $$\begin{gathered}
\int \frac{\mathrm{d}^3\bm q}{(2\pi)^3}q^{\mu}_\perp = ~C_1P_{F\perp}^{\mu},\\
\int \frac{\mathrm{d}^3\bm q}{(2\pi)^3}q^{\mu}_\perp q^{\nu}_\perp= ~C_{21}P_{F\perp}^{\mu}P_{F\perp}^{\nu}+C_{22}g_T^{\mu \nu},\\
\int \frac{\mathrm{d}^3\bm q}{(2\pi)^3}q^{\mu}_\perp q^{\nu}_\perp q^{\alpha}_\perp= ~C_{31}P_{F\perp}^{\mu}P_{F\perp}^{\nu}P_{F\perp}^{\alpha}
+ C_{32}(g_T^{\mu \nu}P_{F\perp}^{\alpha}+g_T^{\mu \alpha}P_{F\perp}^{\nu}+g_T^{\alpha \nu}P_{F\perp}^{\mu}),\\
\int \frac{\mathrm{d}^3\bm q}{(2\pi)^3}q^{\mu}_\perp q^{\nu}_\perp q^{\alpha}_\perp q^{\beta}_\perp= ~C_{41}P_{F\perp}^{\mu}P_{F\perp}^{\nu}P_{F\perp}^{\alpha}P_{F\perp}^{\beta}+
C_{42}(g_T^{\mu \nu}P_{F\perp}^{\alpha}P_{F\perp}^{\beta}+g_T^{\mu \alpha}P_{F\perp}^{\nu}P_{F\perp}^{\beta}+
g_T^{\alpha \nu}P_{F\perp}^{\mu}P_{F\perp}^{\beta}+\\ g_T^{\alpha \beta}P_{F\perp}^{\mu}P_{F\perp}^{\nu}+g_T^{\beta \nu}P_{F\perp}^{\mu}P_{F\perp}^{\alpha}+g_T^{\beta \mu}P_{F\perp}^{\nu}P_{F\perp}^{\alpha})+
C_{43}(g_T^{\alpha \beta}g_T^{\mu \nu}+g_T^{\alpha \nu}g_T^{\mu \beta}+g_T^{\alpha \mu}g_T^{\beta \nu}),\end{gathered}$$ where $g_T^{\mu\nu}$ are defined as $(g^{\mu\nu}-\frac{P^\mu P^\nu}{P^2})$ and $P^\mu_{F\perp}=(P_F^\mu-\frac{P_F\cdot P}{M^2}P^\mu)$. From above equations we can easily obtain the following expressions of $C_i$, $$\label{eq-Ci}
\left \{
\begin{aligned}
C_{1~}=&~ |\bm q| \cos \eta, &\qquad C_{21}=&~\frac{1}{2}|\bm q|^2(3\cos^2 \eta-1),\\
C_{22}=&~\frac{1}{2}|\bm q|^2(\cos^2 \eta-1), &\qquad C_{31}=&~\frac{1}{2}|\bm q|^3(5\cos^3 \eta-3 \cos \eta),\\
C_{32}=&~\frac{1}{2}|\bm q|^3(\cos^3 \eta- \cos \eta), &\qquad C_{41}=&~\frac{1}{8}|\bm q|^4(35\cos^4 \eta-30\cos^2 \eta+3),\\
C_{42}=&~\frac{1}{8}|\bm q|^4(5\cos^4 \eta-6\cos^2 \eta+1), &\qquad C_{43}=&~\frac{1}{8}|\bm q|^4(\cos^4 \eta-2\cos^2 \eta+1),
\end{aligned}
\right .$$ where $\eta$ is the angle between $\bm{q}$ and $\bm{P}_F$.
The physical $2^-$ $D$-wave states $D_2$ and $D'_{2}$ are the mixing states of $^3\!D_2$ and ${^1\!D_2}$ states, whose wave functions are what we solve directly from the full Salpeter equations. Here we will follow Ref. [@Matsuki-2010] and Ref. [@Fau-2010], where the mixing form for $D$-wave states is defined with the mixing angle $\alpha$ as $$\label{def-D2}
\begin{aligned}
|D_2\rangle &= +\cos \alpha~ |^1\!D_2\rangle + \sin \alpha ~|^3\!D_2\rangle, \\
|D'_2\rangle &= -\sin \alpha ~|^1\!D_2\rangle + \cos \alpha~ |^3\!D_2\rangle.
\end{aligned}$$ In the heavy quark limit ($m_Q{\hspace{-0.25em}\to\hspace{-0.25em}}\infty$), the $D$ mesons are described in the $|J,j_\ell\rangle$ basis, where $m_Q$ denotes the heavy quark mass and $j_\ell$ denotes the total angular momentum of the light quark. The relations between $|J,j_\ell\rangle$ and $|J,S\rangle$ for $L=2$ are showed by $$\begin{gathered}
\begin{bmatrix} |2,5/2\rangle \\ |2, 3/2\rangle \end{bmatrix}=\frac{1}{\sqrt{5}}\begin{bmatrix}\sqrt{2+1} & \sqrt{2}\\ -\sqrt{2} & \sqrt{2+1}\end{bmatrix} \begin{bmatrix} |^1\!D_2\rangle \\ |^3\!D_2\rangle \end{bmatrix}.\end{gathered}$$ Then the mixing angle for $L=2$ can be expressed as $\alpha=\arctan\sqrt{2/3}=39.23\degree$. So in this definition $D_{2}$ corresponds to the $|J^P,j_\ell\rangle=|2^-,{5/2}\rangle$ state and $D'_{2}$ corresponds to the $|2^-,{3/2}\rangle$ state. In this work the same mixing angle will also be used for $2^-$ states $D^{(\prime)}_{s2}$ and $B^{(\prime)}_{(s)2}$. Here the mixing angle is the ideal case predicted by the HEQT in the limit of $m_Q{\hspace{-0.25em}\to\hspace{-0.25em}}\infty$. The dependence for decay widths varying over the mixing angle can be seen in equations (\[E-mix1\]) and (\[E-mix2\]).
The wave functions of $^1\!D_2$ and $^3\!D_2$ states can be achieved by solving the corresponding Salpeter equations directly. Then the amplitude for physical $2^-$ states can be considered as the mixing of the transition amplitudes for $^1\!D_2$ and $^3\!D_2$ states, namely $$\label{mix-d2}
\begin{aligned}
\mathcal{M}^\mu(D_2) = +\cos\alpha~\mathcal{M}^{\mu}(^1\!D_2) + \sin \alpha~\mathcal{M}^{\mu}(^3\!D_2),\\
\mathcal{M}^\mu(D'_2) = -\sin\alpha~\mathcal{M}^{\mu}(^1\!D_2) + \cos \alpha~\mathcal{M}^{\mu}(^3\!D_2).
\end{aligned}$$
By using Eq. (\[mix-d2\]), replacing the final state’s wave function $\psi_D(^3\!D_3)$ by $\psi_D(^1\!D_2)$ and $\psi_D(^3\!D_2)$, and then repeating the above procedures for $^3\!D_3$ state, we can get the form factors $s_i$ for $D_2$ and $t_i$ for $D'_2$ defined in Eq. (\[form\]).
Form factors
------------
To solve the Salpeter equations, in this work we choose the Cornell potential as the inner interaction kernel as before [@PLB584-2004], which is a linear scalar potential plus a vector interaction potential as below $$\label{Cornell}
\begin{aligned}
V(\bm q)&=(2\uppi)^3V_s(\bm q)+\gamma^0\otimes\gamma_0(2\uppi)^3V_v(\bm q),\\
V_s(\bm q)&=-(\frac{\lambda}{\alpha}+V_0)\delta^3(\bm q)+\frac{\lambda}{\uppi^2(\bm q^2+\alpha^2)^2},\\
V_v(\bm q)&=-\frac{2\alpha_s(\bm q)}{3\uppi^2(\bm q^2+\alpha^2)}, ~
\alpha_s(\bm q)=\frac{12\uppi}{27\ln(a+\frac{\bm q^2}{\Lambda^2_\mathrm{QCD}})}.
\end{aligned}$$ In above equations, the symbol $\otimes$ denotes that the Salpeter wave function is sandwiched between the two $\gamma^0$ matrices. The model parameters we used are the same with before [@JPG40-2013], which read $$\begin{aligned}
a&=e=2.7183, &\alpha&=0.060~\si{GeV}, &\lambda&=0.210~\si{GeV}^2, \\
m_u&=0.305~\si{GeV}, &m_d&=0.311~\si{GeV}, & m_s&=0.500~\si{GeV},\\
m_c&=1.62~\si{GeV}, &m_b&=4.96~\si{GeV}, &\Lambda_\text{QCD}&=0.270~\si{GeV}.\end{aligned}$$ The free parameter $V_0$ is fixed by fitting the mass eigenvalue to experimental value.
With the numerical Salpeter wave function we can obtain the form factors.
\
\
Here we plot the $\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D^{(*)}_J$ form factors $s_i$, $t_i$, and $h_i~(i=1,2,3,4)$ changing with the square of momentum transfer $t^2=(P-P_F)^2$ in $\sim$ , respectively, where $s_3$, $t_3$, and $h_3$ are divided by $M^2_{\bar B}$ in order to keep the dimension consistent. $\sim$ are the distribution of form factors $S_i$, $T_i$, and $H_i$ for $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^{(*)}_J$ transitions. Also we divided $S_3$, $T_3$, and $H_3$ by $M^2_{B_c}$ to keep the dimension consistent. From , we can see that in all the range concerned the form factors are quite smooth along with $t^2$. And for transitions $\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D^{(*)}_J$, the form factors change slowly and almost linearly when $t^2$ varies from $0$ to $(M-M_F)^2$. For transitions $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^{(*)}_J$, the form factors change dramatically over $t^2$, especially in the range with large momentum transfer.
Numerical Results and Discussions {#Sec-4}
=================================
Firstly we specify the meson mass, lifetime, CKM matrix elements and decay constants used in this work. For the mass of $\bar{B}$, $\bar B_s$, and $B_c$ mesons we take the values from PDG [@PDG-2014]. We follow the mass predictions and $J^P$ assignments of Ref. [@PRD93-2016] for $D$-wave charm and charm-strange mesons. For $D$-wave bottom mesons $B_2$, $B'_2$, and $B^*_3$ we use the average values of Ref. [@Dev-2015] and Ref. [@Fau-2010]. Predictions of Ref. [@Fau-2010] and Ref. [@Dev-2012] are averaged to achieve the mass of $D$-wave bottom-strange mesons $B_{s2}$, $B'_{s2}$, and $B^*_{s3}$. These mass values we used can been seen below $$\begin{aligned}
{8}
&M_{B} &=5.280~\si{GeV}, \quad &M_{B_s} &=5.367~\si{GeV}, \quad &M_{B_c} &=6.276~\si{GeV},\\
&M_{D_2} &=2.750~\si{GeV}, \quad &M_{D'_2} &=2.780~\si{GeV}, \quad &M_{D^*_3} &=2.800~\si{GeV},\\
&M_{D_{s2}}&=2.846~\si{GeV}, \quad &M_{D'_{s2}} &=2.872~\si{GeV}, \quad &M_{D^*_{s3}} &=2.860~\si{GeV},\\
&M_{B_{2}} &=6.060~\si{GeV}, \quad &M_{B'_{2}} &=6.100~\si{GeV}, \quad &M_{B^*_{3}} &=6.050~\si{GeV},\\
&M_{B_{s2}}&=6.150~\si{GeV}, \quad &M_{B'_{s2}} &=6.210~\si{GeV}, \quad &M_{B^*_{s3}} &=6.190~\si{GeV}.\end{aligned}$$ The lifetime of initial mesons we used are as below [@PDG-2014] $$\begin{gathered}
\tau_{\bar B}=1.519\times 10^{-12}~\si{s}, \quad \tau_{\bar B_s}=1.512\times 10^{-12}~\si{s}, \quad \tau_{ B_c}=0.452\times 10^{-12}~\si{s}.\end{gathered}$$ The involved CKM matrix element values are [@PDG-2014] $$\begin{gathered}
|V_{ud}|=0.974,~ |V_{us}|=0.225, ~ |V_{ub}|=0.0042, ~ |V_{cd}|=0.23, ~|V_{cs}|=1.006, ~|V_{cb}|=0.041.\end{gathered}$$ In the calculation of non-leptonic decays, the decay constants we used are [@PDG-2014; @PRD73-2006] $$\begin{gathered}
f_\pi=130.4 ~\si{MeV}, ~f_K=156.2 ~\si{MeV},~f_\rho=210 ~\si{MeV},~f_{K^*}=217~ \si{MeV}.\end{gathered}$$
For the theoretical uncertainties, here we will just discuss the dependence of the final results on our model parameters $\lambda,~\Lambda_\text{QCD}$ in the Cornell potential, and the constituent quark mass $m_b,~m_c,~m_s,~m_d$ and $m_u$. The theoretical errors, induced by these model parameters, are determined by varying every parameter by $\pm5\%$, and then scanning the parameters space to find the maximum deviation. Generally, this theoretical uncertainties can amount to $10\%\sim30\%$ for the semi-leptonic decays. The theoretical uncertainties show the robustness of the numerical algorithm.
Lepton spectra and $A_{FB}$
---------------------------
The distribution of $\bar B$ and $B^-_c$ decay width $\Gamma$ varying along with $\cos \theta$ for $e$ and $\tau$ modes can be seen in , from which we can see that, for $\bar B$ decays, the distribution of semi-electronic decay widths are much more symmetric than that for the semi-taunic mode. These asymmetries over $\cos \theta$ can also be reflected by the forward-backward asymmetries $A_{FB}$, which are showed in . We can see that $A_{FB}$ is sensitive to lepton mass $m_\ell$ and is the monotonic function of $m_\ell$. Considering the absolute values of $A_{FB}$, we find that for $\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D^{(*)}_J$ and $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^{(*)}_J$, the $\mu$ decay mode has the smallest $|A_{FB}|$.\
\
Channels $A_{FB}$ Channels $A_{FB}$ Channels $A_{FB}$
-------------------------------------------------------------------- ---------- --------------------------------------------------------------------- ---------- ---------------------------------------------------------------------- ----------
$\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_2e\bar \nu$ $ -0.08$ $\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_2e\bar \nu$ $ -0.08$ $\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_3e\bar \nu$ $-0.10$
$\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_{2}\mu\bar \nu$ $ -0.05$ $\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_{2}\mu\bar \nu$ $ -0.05$ $\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_{3}\mu\bar \nu$ $-0.07$
$\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_2\tau\bar \nu$ $ 0.20$ $\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_2\tau\bar \nu$ $ 0.21$ $\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_3\tau\bar \nu$ $0.12$
$\bar B_s{\hspace{-0.25em}\to\hspace{-0.25em}}D_{s2}e\bar \nu$ $ -0.10$ $\bar B_s{\hspace{-0.25em}\to\hspace{-0.25em}}D'_{s2}e\bar \nu$ $-0.09$ $\bar B_s{\hspace{-0.25em}\to\hspace{-0.25em}}D^*_{s3}e\bar \nu$ $ -0.10$
$\bar B_s{\hspace{-0.25em}\to\hspace{-0.25em}}D_{s2}\mu\bar \nu$ $ -0.07$ $\bar B_s{\hspace{-0.25em}\to\hspace{-0.25em}}D'_{s2}\mu\bar \nu$ $-0.06$ $\bar B_s{\hspace{-0.25em}\to\hspace{-0.25em}}D^*_{s3}\mu\bar \nu$ $ -0.08$
$\bar B_s{\hspace{-0.25em}\to\hspace{-0.25em}}D_{s2}\tau\bar \nu$ $ 0.17$ $\bar B_s{\hspace{-0.25em}\to\hspace{-0.25em}}D'_{s2}\tau\bar \nu$ $ 0.20$ $\bar B_s{\hspace{-0.25em}\to\hspace{-0.25em}}D^*_{s3}\tau\bar \nu$ $ 0.11$
$B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_2e\bar \nu$ $ -0.28$ $B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_2e\bar \nu$ $-0.43$ $B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_3e\bar \nu$ $ -0.24$
$B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_{2}\mu\bar \nu$ $ -0.28$ $B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_{2}\mu\bar \nu$ $-0.42$ $B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_{3}\mu\bar \nu$ $ -0.23$
$B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_2\tau\bar \nu$ $ -0.03$ $B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_2\tau\bar \nu$ $-0.19$ $B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_3\tau\bar \nu$ $ -0.01$
$B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B_2e^+\nu$ $0.04$ $B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B'_2e^+\nu$ $-0.07$ $B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B^*_3e^+\nu$ $0.03$
$B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B_2\mu^+\nu$ $0.23$ $B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B'_2\mu^+\nu$ $0.18$ $B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B^*_3\mu^+\nu$ $0.24$
$B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B_{s2}e^+\nu$ $0.03$ $B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B'_{s2}e^+\nu$ $-0.03$ $B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B^*_{s3}e^+\nu$ $0.01$
: $A_{FB}$ for semi-leptonic decays of $\bar B, \bar B_s$ and $B_c$ to $D$-wave heavy-light mesons.[]{data-label="Tab-AFB"}
The spectra of decay widths for $\bar B$ and $B^-_c$ varying along with $|\bm{p}_\ell|$, the absolute value of the three-momentum for charged leptons, are showed in . This distribution is almost the same for $\bar B$ decays into $D_2$, $D'_2$ or $D^*_3$. For $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^{(*)}_J$, the momentum spectrum of $\bar D'_2$ is sharper than that of $\bar D_2$ and $\bar D^*_3$.
\
\
Branching ratios of semi-leptonic decays
----------------------------------------
The semi-electronic decay widths we got are $\Gamma(\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D_2e\bar\nu_e)=4.9\times10^{-16}$ , $\Gamma(\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D'_2e\bar\nu_e)=1.8\times10^{-16}$ , and $\Gamma(\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D^*_3e\bar\nu_e)=4.5\times10^{-16}$ . The branching ratios of $\bar B$ to $D$-wave charmed mesons are listed in . We have listed others’ results for comparison if available. Our results are about 5 times greater than that in Ref. [@PRD79-2009]. It’s noticeable that our results for decays into $D_2$ and $D'_2$ are in the same order, while in the results of QCD sum rules [@PRD79-2009] $\mathcal{B}(\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D_2)$ is about 25 times larger than $\mathcal{B}(\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D'_2)$. The branching ratios for semi-leptonic decays of $\bar B_s$ into $D_{s2}$, $D'_{s2}$ and $D^*_{s3}$ are listed in . Our results for $\bar B_s$ to $D$-wave charm-strange mesons are also much larger than the results of QCD sum rules in Ref. [@Gan-2015].
The branching ratios for $B_c$ to $D$-wave $\bar D^{(*)}_J$ are listed in . The branching ratios for semi-leptonic decays of $B^-_c$ to $\bar D_2$ and $\bar D^*_3$ are in the order of $10^{-5}$, and for $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}D'_2$ the results are in the order of $10^{-6}$. These results are about 100 times smaller than that for $\bar B_{(s)}$ decays owning to the the different CKM matrix elements.
For completeness of this research, we also give the corresponding results for $B_c$ to the $D$-wave $B^{(*)}_J$ and $B^{(*)}_{sJ}$ in , although their branching ratios are quite small duo to the tiny phase space. For $D$-wave bottom mesons, the semi-taunic mode is not available and for $D$-wave bottom-strange mesons, both the $\mu$ and $\tau$ modes are unavailable since the constraints of phase space. The branching ratios for $B^+_c{\hspace{-0.25em}\to\hspace{-0.25em}}B^{(*)}_J$ are less than $10^{-8}$ and that for $B^+_c{\hspace{-0.25em}\to\hspace{-0.25em}}B^{(*)}_{sJ}$ are less than $10^{-9}$. Based on our results, the possibilities for the $D$-wave bottomed mesons to be detected in $B_c$ decays are quite small by current experiments.
Channels Ours Ref. [@PRD79-2009] Ref. [@PLB478-2000] Ref. [@PTP91-1994] Ref. [@PRD54-1996]
------------------------------------------------------------------ ------------------------------------- ----------------------- --------------------- --------------------- ---------------------
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}D_2 e \bar{\nu} $ $1.1{^{-0.3}_{+0.3}\times 10^{-3}}$ $1.5{\times 10^{-4}}$ $1{\times 10^{-5}}$ - -
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}D_2 \mu \bar{\nu}$ $1.1{^{-0.3}_{+0.3}\times 10^{-3}}$ $1.5{\times 10^{-5}}$ $1{\times 10^{-5}}$ - -
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}D_2 \tau \bar{\nu}$ $8.0{^{-2.0}_{+2.0}\times 10^{-6}}$ - - - -
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}D'_2e\bar{\nu} $ $4.1{^{-0.8}_{+0.9}\times 10^{-4}}$ $6{\times 10^{-6}}$ - $2{\times 10^{-6}}$ $6{\times 10^{-5}}$
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}D'_2 \mu \bar{\nu}$ $4.1{^{-0.8}_{+0.9}\times 10^{-4}}$ $6{\times 10^{-6}}$ - $2{\times 10^{-6}}$ -
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}D'_2\tau \bar{\nu}$ $2.7{^{-0.4}_{+0.5}\times 10^{-6}}$ - - - -
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}D^*_3 e \bar{\nu}$ $1.0{^{-0.2}_{+0.2}\times 10^{-3}}$ $2.1{\times 10^{-4}}$ $1{\times 10^{-5}}$ - -
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}D^*_3\mu\bar{\nu} $ $1.0{^{-0.2}_{+0.2}\times 10^{-3}}$ $2.1{\times 10^{-4}}$ $1{\times 10^{-5}}$ - -
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}D^*_3\tau\bar{\nu}$ $5.4{^{-0.9}_{+1.0}\times 10^{-6}}$ - - - -
: Branching ratios of $\bar{B}$ semi-leptonic decays with $\tau_{\bar B}=1.519\times 10^{-12}$ .[]{data-label="br-semi-bd"}
Channels Ours Ref. [@Gan-2015]
-------------------------------------- ------------------------------------- ------------------------ --
$\bar{B}_s\to D_{s2}e \bar{\nu} $ $1.7{^{-0.5}_{+0.5}\times 10^{-3}}$ $1.02{\times 10^{-4}}$
$\bar{B}_s\to D_{s2} \mu \bar{\nu}$ $1.7{^{-0.5}_{+0.4}\times 10^{-3}}$ $1.02{\times 10^{-4}}$
$\bar{B}_s\to D_{s2} \tau \bar{\nu}$ $1.3{^{-0.4}_{+0.4}\times 10^{-5}}$ -
$\bar{B}_s\to D'_{s2}e\bar{\nu} $ $5.2{^{-1.5}_{+1.6}\times 10^{-4}}$ $3.4{\times 10^{-7}}$
$\bar{B}_s\to D'_{s2} \mu \bar{\nu}$ $5.1{^{-1.5}_{+1.6}\times 10^{-4}}$ $3.4{\times 10^{-7}}$
$\bar{B}_s\to D'_{s2}\tau \bar{\nu}$ $3.4{^{-1.0}_{+1.1}\times 10^{-6}}$ -
$\bar{B}_s\to D^*_{s3} e \bar{\nu}$ $1.5{^{-0.4}_{+0.4}\times 10^{-3}}$ $3.46{\times 10^{-4}}$
$\bar{B}_s\to D^*_{s3}\mu\bar{\nu} $ $1.4{^{-0.4}_{+0.4}\times 10^{-3}}$ $3.46{\times 10^{-4}}$
$\bar{B}_s\to D^*_{s3}\tau\bar{\nu}$ $9.4{^{-2.5}_{+2.8}\times 10^{-6}}$ -
: Branching ratios of $\bar{B}_s$ semi-leptonic decays with $\tau_{\bar B_s}=1.512\times 10^{-12}$ .[]{data-label="br-semi-Bs"}
Channels Br Channels Br Channels Br
------------------------------------------------------------------- -------------------------------------- -------------------------------------------------------------------- -------------------------------------- --------------------------------------------------------------------- --------------------------------------
$B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_2e\bar \nu$ $ 2.2{^{-0.4}_{+0.7}\times 10^{-5}}$ $B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_2e\bar \nu$ $4.0{^{-0.9}_{+1.6}\times 10^{-6}}$ $B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_3e\bar \nu$ $ 1.2{^{-0.2}_{+0.4}\times 10^{-5}}$
$B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_{2}\mu\bar \nu$ $2.2{^{-0.4}_{+0.7}\times 10^{-5}}$ $B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_{2}\mu\bar \nu$ $4.0{^{-0.9}_{+1.6}\times 10^{-6}}$ $B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_{3}\mu\bar \nu$ $1.2{^{-0.2}_{+0.4}\times 10^{-5}}$
$B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_2\tau\bar \nu$ $7.7{^{-1.5}_{+2.6}\times 10^{-6}}$ $B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_2\tau\bar \nu$ $1.2{^{-0.3}_{+0.5}\times 10^{-6}}$ $B_c^-{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_3\tau\bar \nu$ $3.1{^{-0.7}_{+1.1}\times 10^{-6}}$
$B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B_2e^+\nu$ $9.4{^{-1.0}_{+0.7}\times 10^{-9}}$ $B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B'_2e^+\nu$ $1.3{^{-0.3}_{+0.3}\times 10^{-10}}$ $B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B^*_3e^+\nu$ $1.4{^{-0.3}_{+0.3}\times 10^{-10}}$
$B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B_2\mu^+\nu$ $1.7{^{-0.2}_{+0.1}\times 10^{-9}}$ $B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B'_2\mu^+\nu$ $7.6{^{-0.7}_{+0.7}\times 10^{-12}}$ $B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B^*_3\mu^+\nu$ $2.0{^{-0.4}_{+0.4}\times 10^{-11}}$
$B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B_{s2}e^+\nu$ $3.3{^{-0.2}_{+0.2}\times 10^{-9}}$ $B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B'_{s2}e^+\nu$ $3.2{^{-0.4}_{+0.4}\times 10^{-12}}$ $B_c^+{\hspace{-0.25em}\to\hspace{-0.25em}}B^*_{s3}e^+\nu$ $5.6{^{-1.7}_{+1.7}\times 10^{-13}}$
: Semi-leptonic decay branching ratios of $B_c$ to $D$-wave heavy-light mesons with $\tau_{B_c}=0.452\times 10^{-12}$ .[]{data-label="br-semi-Bc"}
The ratio $\mathscr{R}[D^{(*)}_J]$, defined as the ratio of semi-taunic branching fraction over semi-electronic branching fraction for decay $\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D^{(*)}_J$, namely, $\mathscr{R}[D^{(*)}_{J}]=\frac{\mathcal{B}[\bar B\to D^{(*)}_{J}\tau^-\bar \nu_\tau]}{\mathcal{B}[\bar B\to D^{(*)}_{J}e^-\bar \nu_e]}$, may hint the new physics [@BaBar-2012; @Belle-2016]. We present these ratios for decays to $D$-wave charmed mesons in , from which we can see that, $\mathscr{R}[D^{(*)}_{J}]$ for $\bar B$ decays and $\mathscr{R}[D^{(*)}_{sJ}]$ for $\bar B_s$ decays, are almost the same and in the order of $10^{-3}$, while $\mathscr{R}[\bar D^{(*)}_{J}]$ for $B^-_c$ decays are in the order of $10^{-1}$. This big difference is mainly due to the phase space. By simple integral over the phase space, we can find that, the phase space ratio of semi-taunic decay over semi-electronic decay for $B_c^-$ meson is about 30 times larger than that for $\bar B$ or $\bar B_s$ meson.
Modes $D_2$ $D'_2$ $D^*_3$ $D_{s2}$ $D'_{s2}$ $D^*_{s3}$ $\bar D_2$ $\bar D'_2$ $\bar D^*_3$
-------------- ---------- ---------- ---------- ---------- ----------- ------------ ------------ ------------- --------------
$\mathscr R$ $0.0071$ $0.0065$ $0.0052$ $0.0079$ $0.0066$ $0.0064$ $0.35$ $0.29$ $0.25$
: $\mathscr{R}[D^{(*)}_J]=\frac{\mathscr{B}[\bar B\to D^{(*)}_J\tau\bar \nu_\tau]}{\mathscr{B}[\bar B\to D^{(*)}_Je\bar \nu_e]}$, $\mathscr{R}[D^{(*)}_{sJ}]=\frac{\mathscr{B}[\bar B_s\to D^{(*)}_{sJ}\tau\bar \nu_\tau]}{\mathscr{B}[\bar B_s\to D^{(*)}_{sJ}e\bar \nu_e]}$, and $\mathscr{R}[\bar D^{(*)}_J]=\frac{\mathscr{B}[B^-_c\to \bar D^{(*)}_J\tau \bar \nu_\tau]}{\mathscr{B}[B_c^-\to \bar D^{(*)}_Je \bar \nu_e]}$, ratios of semi-taunic branching ratio to semi-electronic branching ratio for $\bar B$, $\bar B_s$ and $B^-_c$ to $D$-wave charmed mesons.[]{data-label="Tab-RD"}
The decay widths for $\bar B_{(s)}$ or $B_c$ to $2^-$ states mesons are dependent on the mixing angle $\alpha$, which can be showed by and . This dependence for $\bar B$ decays can be described by the following equations $$\begin{aligned}
\Gamma(\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D_2e\bar\nu) &=\Gamma_1\big[1+\lambda_1\cos (2\alpha+\Theta_1)\big],\label{E-mix1}\\
\Gamma(\bar B{\hspace{-0.25em}\to\hspace{-0.25em}}D'_2e\bar\nu)&=\Gamma_2\big[1-\lambda_2\cos (2\alpha+\Theta_2)\big].\label{E-mix2}\end{aligned}$$ Our fit results give that the parameters are as $$\begin{gathered}
\Gamma_1=3.46\times10^{-16},\quad \lambda_1=0.709,\quad \Theta_1=-23.7\degree,\\
\Gamma_2=3.06\times10^{-16},\quad \lambda_2=0.711,\quad \Theta_2=-24.1\degree.\end{gathered}$$ The tiny differences in parameters for $D_2$ and $D'_2$ come from the small difference between $m_{D_2}$ and $m_{D'_2}$. In and , we also show the ratios $\frac{\Gamma(\bar B\to{D_2}e\bar\nu)}{\Gamma(\bar B\to{D'_2}e\bar\nu)}$ and $\frac{\Gamma(B_c^-\to\bar D_2e\bar\nu)}{\Gamma(B^-_c\to\bar D'_2e\bar\nu)}$, which are very sensitive to the mixing angle.
\
Non-leptonic decay widths and branching ratios
----------------------------------------------
The non-leptonic decay widths are listed in , where we have kept the Wilson coefficient $a_1$ in order to facilitate comparison with other models. The corresponding branching ratios are listed in , where we have specified the values $a^b_1=1.14$ for $b{\hspace{-0.25em}\to\hspace{-0.25em}}c(u)$ transition and $a^c_1=1.2$ for $c{\hspace{-0.25em}\to\hspace{-0.25em}}d(s)$ transition [@PRD73-2006]. From the non-leptonic decay results we can see that, with the same final $D$ meson, the $\rho$ mode has the largest branching ratio and can reach $10^{-3}$ order in $\bar B_{(s)}$ decays, and $10^{-6}$ order in $B_c$ decay. When the light mesons have the same quark constituents, the width for decay into vector meson $(\rho, K^*)$ mode is about $2\sim3$ times greater than its pseudoscalar meson $(\pi, K)$ mode.
------------------------------------------------------------------ ---------------------------------------- ------------------------------------------------------------------ ---------------------------------------- ------------------------------------------------------------------- ----------------------------------------
Channels Width Channels Width Channels Width
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D_2} \pi^- $ $2.8{^{-0.7}_{+0.7}\times 10^{-16}} $ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_2} \pi^- $ $9.8{^{-2.0}_{+2.2}\times 10^{-17}} $ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_3} \pi^- $ $2.2{^{-0.4}_{+0.4}\times 10^{-16}} $
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D_2} K^- $ $2.0{^{-0.5}_{+0.5}\times 10^{-17}} $ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_2} K^- $ $6.7{^{-1.4}_{+1.5}\times 10^{-18}} $ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_3} K^- $ $1.4{^{-0.3}_{+0.3}\times 10^{-17}} $
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D_2} \rho^-$ $5.5{^{-1.4}_{+1.4}\times 10^{-16}} $ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_2}\rho^- $ $2.0{^{-0.4}_{+0.4}\times 10^{-16}} $ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_3} \rho^-$ $4.7{^{-0.8}_{+0.9}\times 10^{-16}} $
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D_2} K^{*-} $ $2.9{^{-0.7}_{+0.7}\times 10^{-17}} $ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_2}K^{*-} $ $1.0{^{-0.2}_{+0.2}\times 10^{-17}} $ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_3} K^{*-}$ $2.5{^{-0.4}_{+0.5}\times 10^{-17}} $
$\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D_{s2}} \pi^- $ $4.0{^{-1.0}_{+1.1}\times 10^{-16}} $ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_{s2}} \pi^- $ $1.2{^{-0.4}_{+0.4}\times 10^{-16}} $ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_{s3}} \pi^- $ $2.9{^{-0.8}_{+0.7}\times 10^{-16}} $
$\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D_{s2}} K^- $ $2.8{^{-0.8}_{+0.7}\times 10^{-17}} $ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_{s2}} K^- $ $8.2{^{-2.4}_{+2.5}\times 10^{-18}} $ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_{s3}} K^- $ $1.9{^{-0.5}_{+0.5}\times 10^{-17}} $
$\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D_{s2}} \rho^-$ $8.1{^{-2.3}_{+2.1}\times 10^{-16}} $ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_{s2}}\rho^- $ $2.4{^{-0.7}_{+0.8}\times 10^{-16}} $ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_{s3}} \rho^-$ $6.4{^{-1.7}_{+1.7}\times 10^{-16}} $
$\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D_{s2}} K^{*-} $ $4.2{^{-1.2}_{+1.1}\times 10^{-17}} $ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_{s2}}K^{*-} $ $1.3{^{-0.4}_{+0.4}\times 10^{-17}} $ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_{s3}} K^{*-}$ $3.4{^{-0.9}_{+0.9}\times 10^{-17}} $
$B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_2\pi^-$ $1.1{^{-0.4}_{+0.7}\times 10^{-18}} $ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_2\pi^-$ $2.5{^{-0.7}_{+1.3}\times 10^{-19}} $ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_3\pi^-$ $1.1{^{-0.3}_{+0.5}\times 10^{-18}} $
$B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_2K^-$ $9.0{^{-0.3}_{+0.5}\times 10^{-20}} $ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_2K^-$ $2.0{^{-0.6}_{+1.0}\times 10^{-20}} $ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_3K^-$ $8.4{^{-2.2}_{+4.0}\times 10^{-20}} $
$B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_2\rho^-$ $3.5{^{-1.1}_{+1.9}\times 10^{-18}} $ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_2\rho^-$ $7.8{^{-2.2}_{+3.9}\times 10^{-19}} $ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_3\rho^-$ $3.1{^{-0.8}_{+1.4}\times 10^{-18}} $
$B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_2K^{*-}$ $2.1{^{-0.6}_{+1.1}\times 10^{-19}} $ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_2K^{*-}$ $4.7{^{-1.3}_{+2.3}\times 10^{-20}} $ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_3K^{*-}$ $1.9{^{-0.5}_{+0.8}\times 10^{-19}} $
$B^+_c{\hspace{-0.25em}\to\hspace{-0.25em}}B_2 \pi^+ $ $1.4{^{-0.1}_{+0.1}\times 10^{-19}} $ $B^+_c{\hspace{-0.25em}\to\hspace{-0.25em}}B'_2 \pi^+ $ $1.3{^{-0.3}_{+0.2}\times 10^{-21}} $ $B^+_c{\hspace{-0.25em}\to\hspace{-0.25em}}B^*_3 \pi^+ $ $1.9{^{-0.4}_{+0.4}\times 10^{-21}} $
------------------------------------------------------------------ ---------------------------------------- ------------------------------------------------------------------ ---------------------------------------- ------------------------------------------------------------------- ----------------------------------------
: Non-leptonic decay widths of $\bar{B}$, $\bar{B}_s$ and $B_c$ to $D$-wave heavy-light meson with general Wilson coefficient $a_1$.[]{data-label="width-non"}
Channels Br Channels Br Channels Br
----------------------------------------------------------------- ------------------------------------- ------------------------------------------------------------------ ------------------------------------- ------------------------------------------------------------------- --------------------------------------
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D_2} \pi^- $ $8.5{^{-2.2}_{+2.1}\times 10^{-4}}$ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_2} \pi^- $ $2.9{^{-0.6}_{+0.6}\times 10^{-4}}$ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_3} \pi^- $ $6.5{^{-1.2}_{+1.2}\times 10^{-4}}$
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D_2} K^- $ $5.9{^{-1.5}_{+1.4}\times 10^{-5}}$ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_2} K^- $ $2.0{^{-0.4}_{+0.4}\times 10^{-5}}$ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_3} K^- $ $4.3{^{-0.8}_{+0.8}\times 10^{-5}}$
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D_2} \rho^-$ $1.7{^{-0.4}_{+0.4}\times 10^{-3}}$ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_2}\rho^- $ $5.9{^{-1.2}_{+1.3}\times 10^{-4}}$ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_3} \rho^-$ $1.4{^{-0.3}_{+0.3}\times 10^{-3}}$
$\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D_2} K^{*-}$ $8.6{^{-2.2}_{+2.1}\times 10^{-5}}$ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_2}K^{*-}$ $3.1{^{-0.6}_{+0.7}\times 10^{-5}}$ $\bar{B}{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_3} K^{*-}$ $7.5{^{-1.3}_{+1.4}\times 10^{-5}}$
$\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D_{s2}} \pi^- $ $1.2{^{-0.3}_{+0.3}\times 10^{-3}}$ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_{s2}} \pi^- $ $3.6{^{-1.0}_{+1.1}\times 10^{-4}}$ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_{s3}} \pi^- $ $8.5{^{-2.3}_{+2.2}\times 10^{-4}}$
$\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D_{s2}} K^- $ $8.3{^{-2.3}_{+2.1}\times 10^{-5}}$ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_{s2}} K^- $ $2.5{^{-0.7}_{+0.7}\times 10^{-5}}$ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_{s3}} K^- $ $5.7{^{-1.6}_{+1.5}\times 10^{-5}}$
$\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D_{s2}} \rho^-$ $2.4{^{-0.7}_{+0.6}\times 10^{-3}}$ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_{s2}}\rho^- $ $7.3{^{-2.1}_{+2.2}\times 10^{-4}}$ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_{s3}} \rho^-$ $1.9{^{-0.5}_{+0.5}\times 10^{-3}}$
$\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D_{s2}} K^{*-}$ $1.3{^{-0.4}_{+0.3}\times 10^{-4}}$ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D'_{s2}}K^{*-}$ $3.8{^{-1.1}_{+1.2}\times 10^{-5}}$ $\bar{B}_s{\hspace{-0.25em}\to\hspace{-0.25em}}{D^*_{s3}} K^{*-}$ $1.0{^{-0.3}_{+0.3}\times 10^{-4}}$
$B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_2\pi^-$ $1.0{^{-0.3}_{+0.6}\times 10^{-6}}$ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_2\pi^-$ $2.2{^{-0.6}_{+1.2}\times 10^{-7}}$ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_3\pi^-$ $9.6{^{-2.6}_{+4.6}\times 10^{-7}}$
$B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_2K^-$ $8.0{^{-2.5}_{+4.6}\times 10^{-8}}$ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_2K^-$ $1.7{^{-0.5}_{+0.9}\times 10^{-8}}$ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_3K^-$ $7.5{^{-0.2}_{+3.5}\times 10^{-8}}$
$B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_2\rho^-$ $3.1{^{-1.0}_{+1.7}\times 10^{-6}}$ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_2\rho^-$ $6.9{^{-2.0}_{+3.5}\times 10^{-7}}$ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_3\rho^-$ $2.8{^{-0.7}_{+1.3}\times 10^{-6}}$
$B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D_2K^{*-}$ $1.9{^{-0.6}_{+1.0}\times 10^{-7}}$ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D'_2K^{*-}$ $4.2{^{-1.2}_{+2.0}\times 10^{-8}}$ $B^-_c{\hspace{-0.25em}\to\hspace{-0.25em}}\bar D^*_3K^{*-}$ $1.7{^{-0.4}_{+0.7}\times 10^{-7}}$
$B^+_c{\hspace{-0.25em}\to\hspace{-0.25em}}B_2 \pi^+ $ $1.4{^{-0.1}_{+0.1}\times 10^{-7}}$ $B^+_c{\hspace{-0.25em}\to\hspace{-0.25em}}B'_2 \pi^+ $ $1.3{^{-0.2}_{+0.2}\times 10^{-9}}$ $B^+_c{\hspace{-0.25em}\to\hspace{-0.25em}}B^*_3 \pi^+ $ $1.9{^{-0.4}_{+0.4}\times 10^{-9}} $
: Branching ratios of non-leptonic decays for $\bar{B}$, $\bar B_s$ and $B_c$ to $D$-wave heavy-light mesons. $a^b_1=1.14$ for $b$ quark decay and $a^c_1=1.2$ for $c$ quark decay.[]{data-label="br-non"}
Summary {#Sec-5}
=======
In this work we calculated semi-leptonic and non-leptonic decays of $\bar B_{(s)}$ into $D$-wave charmed mesons ($D_{(s)2}$, $D'_{(s)2}$, $D^*_{(s)3}$) and $B_c$ into $D$-wave charmed and bottomed excited mesons. Form factors of hadronic transition are calculated by instantaneous Bethe-Salpeter methods. The semi-electronic branching ratios for $\bar B_{(s)}{\hspace{-0.25em}\to\hspace{-0.25em}}D^{(*)}_{(s)J}$ we got are about $10^{-3}$ order, and for $B_c$ to $D$-wave charmed mesons are about $10^{-5}$ order. The non-leptonic branching ratios for decays to $\rho$ mode can reach $10^{-3}$ order for $\bar B_{(s)}$ decays. So the $D$-wave $D$ and $D_s$ mesons are hopefully to be detected in $\bar B_{(s)}$ decays by current experiments . Our results reveal the branching fractions for $B_c$ to $D$-wave bottomed mesons are less than $10^{-8}$, which makes the $D$-wave bottomed mesons almost impossible to be discovered in $B_c$ decays by current experiments.
We also present the angular distribution and charged lepton spectra for $\bar B$ and $B_c$ decays. The $2^-$ states $D_2$ and $D'_2$ are the mixing states of $^1\!D_2-{^3\!D_2}$, so we present the dependence of the decay width varying along with the mixing angle. Based on our results, the semi-leptontic and non-leptonic branching ratios for $\bar B_{(s)}$ decays to the $D$-wave charm and charm-strange mesons have reached the experimental detection thresholds. These results would be helpful in future detecting and understanding these new $D$-wave excited $D_{(s)}$ mesons.
Acknowledgments {#acknowledgments .unnumbered}
===============
This work was supported in part by the National Natural Science Foundation of China (NSFC) under Grant Nos. 11405037, 11575048 and 11505039, and in part by PIRS of HIT Nos. T201405, A201409, and B201506.
Expressions for $N_i$s in the Hadronic Tensor $H_{\mu\nu}$ {#Ni}
==========================================================
The hadronic tensor $N_i~(i=1,2,4,5,6)$ for $\bar B$ to $D_{2}$ meson are $$\begin{aligned}
N_1 &= \frac{2 M^4 \bm{p}_F^4 s_1^2}{3 M_F^4}-\frac{4 M^2 \bm{p}_F^2 s_1 s_3}{3 M_F^2}-\frac{1}{2} M^2 \bm{p}_F^2 s_4^2+\frac{s_3^2}{6}, \label{N1-2} \\
N_2 &= \frac{2 E_F M^3 \bm{p}_F^2 s_1 s_3}{3 M_F^4}+\frac{E_F M^3 \bm{p}_F^2 s_4^2}{2 M_F^2}-\frac{E_F M s_3^2}{6 M_F^2}+\frac{2 M^4 \bm{p}_F^4 s_1 s_2}{3 M_F^4}-\frac{2 M^2 \bm{p}_F^2 s_2 s_3}{3 M_F^2},\label{N2-2} \\
N_4 &= \frac{4 E_F M^3 \bm{p}_F^2 s_2 s_3}{3 M_F^4}+\frac{2 M^4 \bm{p}_F^4 s_2^2}{3 M_F^4}-\frac{M^4 \bm{p}_F^2 s_4^2}{2 M_F^2}+\frac{M^2 s_3^2 (M_F^2+4 \bm{p}_F^2)}{6 M_F^4}, \label{N4-2} \\
N_5 &= -\frac{M^4 \bm{p}_F^4 s_4^2}{2 M_F^2}-\frac{M^2 \bm{p}_F^2 s_3^2}{2 M_F^2}, \label{N5-2} \\
N_6 &= -\frac{M^2 \bm{p}_F^2 s_3 s_4}{M_F^2}. \label{N6-2}\end{aligned}$$ Here $\bm p_F$ denotes the three-momentum of final $D$ systems and $E_F=\sqrt{M_F^2+\bm p_F^2}$. For $\bar B$ to $D'_{2}$ the relations between $N_i$ and form factors $t_k~(k=1,2,3,4)$ have the same form with that for $D_2$, just $s_k$ are replaced with $t_k$. Both $s_k$ and $t_k$ are functions of $q'^2_\perp$.
The hadronic tensor $N_i$ for $\bar B$ to $D^*_{3}$ are expressed with form factors $h_k~(k=1,2,3,4)$ as $$\begin{aligned}
N_1 &= \frac{2 M^6 \bm{p}_F^6 h_1^2}{5 M_F^6}-\frac{4 M^4 \bm{p}_F^4 h_1 h_3}{5 M_F^4}-\frac{4 M^4 \bm{p}_F^4 h_4^2}{15 M_F^2}+\frac{2 M^2 \bm{p}_F^2 h_3^2}{15 M_F^2}, \label{N1-3}\\
N_2 &= \frac{2 {E_F} M^5 \bm{p}_F^4 h_1 h_3}{5 M_F^6}+\frac{4 {E_F} M^5 \bm{p}_F^4 h_4^2}{15 M_F^4}-\frac{2 {E_F} M^3 \bm{p}_F^2 h_3^2}{15 M_F^4}+\frac{2 M^6 \bm{p}_F^6 h_1 h_2}{5 M_F^6}-\frac{2 M^4 \bm{p}_F^4 h_2 h_3}{5 M_F^4}, \label{N2-3}\\
N_4 &= \frac{4 {E_F} M^5 \bm{p}_F^4 h_2 h_3}{5 M_F^6}+\frac{2 M^6 \bm{p}_F^6 h_2^2}{5 M_F^6}-\frac{4 M^6 \bm{p}_F^4 h_4^2}{15 M_F^4}+\frac{2 M^4 \bm{p}_F^2 h_3^2 (M_F^2+3 \bm{p}_F^2) }{15 M_F^6}, \label{N4-3}\\
N_5 &= -\frac{4 M^6 \bm{p}_F^6 h_4^2}{15 M_F^4}-\frac{4 M^4 \bm{p}_F^4 h_3^2}{15 M_F^4}, \label{N5-3}\\
N_6 &=-\frac{8 M^4 \bm{p}_F^4 h_3 h_4}{15 M_{\!F}^4}. \label{N6-3}\end{aligned}$$.
Full Salpeter equations and the numerical solutions {#Salpeter-EQ}
===================================================
Salpeter equations
------------------
Salpeter wave function $\varphi(q_\perp)$ is related to BS wave function $\Psi(q)$ by the following definition $$\begin{gathered}
\varphi(q_\perp)=\text{i}\int \frac{\mathrm{d}q_P}{2\pi}\Psi(q),\quad \eta(q_\perp)=\int \frac{\mathrm{d}^3k_\perp}{(2\pi)^3}\varphi(k_\perp)V(|q_\perp- k_\perp|),\end{gathered}$$ where the 3-dimensional integration $\eta(q_\perp)$ can be understood as the BS vertex for bound states, and $V(|q_\perp- k_\perp|)$ denotes the instantaneous interaction kernel.
The projection operators $\Lambda^{\pm}_i(q_\perp)$ ($i=1$ for quark and 2 for anti-quark) are defined as $$\begin{gathered}
\Lambda^{\pm}_i=\frac{1}{2\omega_i}\left[ \frac{\slashed P}{M}\omega_i\pm(-1)^{i+1}(m_i+\slashed q_\perp) \right].\end{gathered}$$ Then we define four wave functions $\varphi^{\pm\pm}$ by $\varphi$ and $\Lambda^\pm_i$ as $$\begin{gathered}
\varphi^{\pm\pm}\equiv\Lambda_1^\pm(q_\perp)\frac{\slashed P}{M}\varphi(q_\perp)\frac{\slashed P}{M}\Lambda_2^\pm(q_\perp),\label{Def-np}\end{gathered}$$ where $\varphi^{++}$ and $\varphi^{--}$ are called the positive and negative Salpeter wave function, respectively. And we can easily check that $\varphi=\varphi^{++}+\varphi^{-+}+\varphi^{+-}+\varphi^{--}$.
The full coupled Salpeter equations then can be expressed as [@PR87-1952] $$\begin{gathered}
\varphi^{+-}=\varphi^{-+}=0\label{BS-np},\\
(M-\omega_1-\omega_2)\varphi^{++}=+\Lambda_1^+(q_\perp)\eta(q_\perp)\Lambda^+_2(q_\perp),\label{BS-pp}\\
(M+\omega_1+\omega_2)\varphi^{--}=-\Lambda_1^-(q_\perp)\eta(q_\perp)\Lambda^-_2(q_\perp).\label{BS-nn}\end{gathered}$$ From above equations, we can see that in the weak binding condition $M\sim (\omega_1+\omega_2)$, $\varphi^{--}$ is much smaller compared with $\varphi^{++}$ and can be ignored in the calculations. The normalization condition for Salpeter wave function reads $$\begin{gathered}
\label{Norm-BS}
\int \frac{\text{d}^3q_\perp}{(2\uppi)^3}\left[\overline\varphi^{++}\frac{\slashed P}{M}\varphi^{++}\frac{\slashed P}{M}-\overline\varphi^{--}\frac{\slashed P}{M}\varphi^{--}\frac{\slashed P}{M}\right]=2M.\end{gathered}$$
Numerical solutions of $0^-$ state
----------------------------------
Now we take the $0^{-}~(^1\!S_0)$ state as an example to show the details of achieving Sapeter equations’ numerical results. The Salpeter wave function for $0^-(^1\!S_0)$ state has the following general form [@PLB584-2004] $$\label{BS-wave-1s0}
\varphi(^1\!S_0)= M\bigg[ k_1 \frac{\slashed P}{M}+k_2 +k_3 \frac{\slashed q_{\perp}}{M}+k_4\frac{\slashed P \slashed q_{\perp}}{M^2} \bigg ]\gamma^5.$$ By utilizing the Eq. (\[BS-np\]), we have the following two constraint conditions $$\begin{gathered}
\label{BS-par-1s0}
k_3= \frac{M(\omega_1-\omega_2)}{m_1\omega_2+m_2\omega_1}k_2,\quad
k_4= -\frac{M(\omega_1+\omega_2)}{m_1\omega_2+m_2\omega_1}k_1.\end{gathered}$$ In above wave function, the only undetermined wave functions are $k_1$ and $k_2$, which are the functions of $q^2_\perp$.
By using the definition Eq. (\[Def-np\]), we can easily get the positive Salpeter wave function of ${^1\!S_0}$ state as Eq. (\[wave-1s0\]), and the corresponding constraint conditions Eq. (\[par-1s0\]). Similarly, the Salpeter negative wave function $\varphi^{--}(^1\!S_0)$ is expressed as $$\label{wave-1s0-nn}
\varphi^{--}(^1\!S_0)=\bigg [ Z_1+Z_2 \frac{\slashed P}{M} +Z_3 \frac{\slashed q_{\perp}}{M}+Z_4\frac{\slashed P \slashed q_{\perp}}{M^2} \bigg ]\gamma^5.$$ $Z_i~(i=1,2,3,4)$ has the following forms $$\label{par-1s0-nn}
\begin{aligned}
Z_1=& \frac{M}{2}\biggl[k_2 -\frac{\omega_1+\omega_2}{m_1+m_2}k_1\bigg],&\quad Z_3=&-\frac{M(\omega_1-\omega_2)}{m_1\omega_2+m_2\omega_1}Z_1,\\
Z_2=&\frac{M}{2}\biggl[k_1-\frac{m_1+m_2}{\omega_1+\omega_2}k_2 \bigg],&\quad Z_4=&+\frac{M(m_1+m_2)}{m_1\omega_2+m_2\omega_1}Z_1.
\end{aligned}$$ And now the normalization condition Eq. (\[Norm-BS\]) becomes $$\begin{gathered}
\label{Norm-1S0}
\int \frac{\text{d}^3\bm{q}}{(2\uppi)^3}\frac{8M\omega_1\omega_2k_1k_2}{(m_1\omega_2+m_2\omega_1)}=1.\end{gathered}$$
Inserting the Salpeter positive wave function Eq. (\[wave-1s0\]), and negative wave function Eq. (\[wave-1s0-nn\]) into Salpeter equations Eq. (\[BS-pp\]) and (\[BS-nn\]) respectively, we can obtain two coupled eigen equations on $k_1$ and $k_2$ [@PLB584-2004] as $$\left \{
\begin{aligned}
(M-\omega_1-\omega_2)\left[ck_1(\bm q)+k_2(\bm q)\right]=\frac{1}{2\omega_1\omega_2}\int \text{d}^3\bm{k} \left[\beta_{1}k_1(\bm k)+\beta_{2}k_2(\bm k)\right],\\
(M+\omega_1+\omega_2)\left[k_2(\bm q)-ck_1(\bm q)\right]=\frac{1}{2\omega_1\omega_2}\int \text{d}^3\bm{k} \left[\beta_{1}k_1(\bm k)-\beta_{2}k_2(\bm k)\right],
\end{aligned}
\right .$$ where we have used definition $c=\frac{\omega_1+\omega_2}{m_1+m_2}$ and the shorthand $$\begin{aligned}
\beta_1&=\bm k\cdot \bm q(V_s+V_v)\frac{(\nu_1+\nu_2)(\omega_1+\omega_2)}{m_1\nu_2+m_2\nu_1} -(V_s-V_v)(m_1\omega_2+m_2\omega_1),\\
\beta_2&=\bm k\cdot \bm q(V_s+V_v)\frac{(\nu_1-\nu_2)(m_1-m_2)}{m_1\nu_2+m_2\nu_1} -(V_s-V_v)(m_1m_2+\omega_1\omega_2+\bm q^2).
\end{aligned}$$ In above equations, $V_s$ and $V_v$ are the scalar and vector parts defined in Cornell potential (see Eq. \[Cornell\]) respectively; we have used the definition $\nu_i=\sqrt{m^2_i+\bm k^2}~(i=1,2)$.
Then by solving the two coupled eigen equations numerically, we achieve the mass spectrum and corresponding wave functions $k_1$, $k_2$. Repeating the similar procedures we can obtain the numerical wave functions for $^1\!D_2$, $^3\!D_2$ and $^3\!D_3$ states. Interested reader can see more details on solving the full Salpeter equations in Refs. [@3D23-wave; @JPG40-2013; @PLB584-2004].
Positive Salpeter wave function for $^1\!S_0$, $^1\!D_2$ and $^3\!D_2$ {#sec-wave}
======================================================================
The positive Salpeter wave function and its constraint conditions for $^1\!D_2$ state [@JHEP03-2013] are displayed in \[wave-1d2\] and \[par-1d2\]. And the undetermined wave function are $f_1$ and $f_2$. $$\label{wave-1d2}
\psi_D(^1\!D_2)=e_{\mu \nu} q'^\mu_{\perp} q'^\nu_{\perp} \bigg [ b_1+b_2 \frac{\slashed P_F}{M_F} +b_3 \frac{\slashed q'_\perp}{M_F}+b_4\frac{\slashed P_F \slashed q'_{\perp}}{M_F^2} \bigg]\gamma^5,$$ $$\label{par-1d2}
\begin{aligned}
b_1=& \frac{1}{2}\Big[f_1+\frac{\omega'_1+\omega'_2}{m'_1+m'_2}f_2 \Big],~
&b_3=& -\frac{M_F(\omega'_1-\omega'_2)}{m'_1\omega'_2+m'_2\omega'_1}b_1,\\
b_2=& \frac{1}{2}\Big[f_2+\frac{m'_1+m'_2}{\omega'_1+\omega'_2}f_1 \Big],~
&b_4=& -\frac{M_F(m'_1+m'_2)}{m'_1\omega'_2+m'_2\omega'_1}b_1 .
\end{aligned}$$
The positive Salpeter wave function of ${^3\!D_2}$ state [@3D23-wave] and constraint conditions can be written as $$\label{wave-3d2}
\psi_D(^3\!D_2)=\mathrm{i}\epsilon_{\mu \nu \alpha \beta }\frac{P_F^{\nu}}{M_F}q'^{\alpha}_{\perp} e^{\beta \delta} q'_{\perp \delta} \gamma^{\mu} \bigg[ i_1+i_2 \frac{\slashed P_F}{M_F} +i_3 \frac{\slashed q'_\perp}{M_F}+i_4\frac{\slashed P_F \slashed q'_{\perp}}{M_F^2} \bigg ].$$ $$\label{par-3d2}
\begin{aligned}
i_1=& \frac{1}{2} \biggl[v_1-\frac{\omega'_1+\omega'_2}{m'_1+m'_2}v_2 \bigg],~
&i_3=&+\frac{M_F(\omega'_1-\omega'_2)}{m'_1\omega'_2+m'_2\omega'_1}i_1,\\
i_2=& \frac{1}{2} \biggl[v_2-\frac{m'_1+m'_2}{\omega'_1+\omega'_2}v_1 \bigg],~
&i_4=&-\frac{M_F(m'_1+m'_2)}{m'_1\omega'_2+m'_2\omega'_1}i_1.
\end{aligned}$$ Here we also only have two undetermined wave function $v_1$ and $v_2$.
In above equations \[wave-1d2\] $\sim$ \[par-3d2\] the indeterminate wave functions, such as $f_1$ and $f_2$ in $\psi_D(^1\!D_2)$, $v_1$ and $v_2$ in $\psi_D(^3\!D_2)$, which are functions of $q'^2_\perp$ and can be determined numerically by solving the coupled Salpeter eigen equations \[BS-pp\] and \[BS-nn\].
References {#references .unnumbered}
==========
[99]{}
P. del Amo Sanchez [*et al*]{}. (BABAR Collaboration), [[Phys. Rev. D. [**82**]{}, 111101 (2010)](http://dx.doi.org/10.1103/PhysRevD.82.111101)]{}.
R. Aaij [*et al*]{}. (LHCb Collaboration), [[JHEP. [**09**]{}, 145 (2013)](http://dx.doi.org/10.1007/JHEP09(2013)145)]{}.
R. Aaij [*et al*]{}. (LHCb Collaboration), [[Phys. Rev. D [**90**]{}, 072003 (2014)](http://dx.doi.org/10.1103/PhysRevD.90.072003)]{}.
R. Aaij [*et al*]{}. (LHCb Collaboration), [[Phys. Rev. D [**91**]{}, 092002 (2015)](http://dx.doi.org/10.1103/PhysRevD.91.092002)]{}.
R. Aaij [*et al*]{}. (LHCb Collaboration), [[Phys. Rev. D [**92**]{}, 032002 (2015)](http://dx.doi.org/10.1103/PhysRevD.92.032002)]{}.
S. Godfrey and N. Isgur, [[Phys. Rev. D [**32**]{}, 189 (1985)](http://dx.doi.org/10.1103/PhysRevD.32.189)]{}.
B. Aubert [*et al*]{}. (BABAR Collaboration), [[Phys. Rev. Lett.[**97**]{}, 222001 (2006)](http://dx.doi.org/10.1103/PhysRevLett.97.222001)]{}.
B. Aubert [*et al*]{}. (BABAR Collaboration), [[Phys. Rev. D [**80**]{}, 092003 (2009)](http://dx.doi.org/10.1103/PhysRevD.80.092003)]{}.
R. Aaij [*et al*]{}. (LHCb Collaboration), [[Phys. Rev. Lett. [**113**]{}, 162001 (2014)](http://dx.doi.org/10.1103/PhysRevLett.113.162001)]{}.
R. Aaij [*et al*]{}. (LHCb Collaboration), [[Phys. Rev. D [**90**]{}, 072003 (2014)](http://dx.doi.org/10.1103/PhysRevD.90.072003)]{}.
P. Colangelo, F. De Fazio and S. Nicotri, [[Phys. Lett. B [**642**]{}, 48 (2006)](http://dx.doi.org/10.1016/j.physletb.2006.09.018)]{}.
P. Colangelo, F. De Fazio, F. Giannuzzi and S. Nicotri, [[Phys. Rev. D [**86**]{}, 054024 (2012)](http://dx.doi.org/10.1103/PhysRevD.86.054024)]{}.
Xian-Hui Zhong, [[Phys. Rev. D [**82**]{}, 114014 (2010)](http://dx.doi.org/10.1103/PhysRevD.82.114014)]{}.
Zhi-Feng Sun, Jie-Sheng Yu, Xiang Liu, and T. Matsuki, [[Phys. Rev. D [**82**]{}, 111501 (2010)](http://dx.doi.org/10.1103/PhysRevD.82.111501)]{}.
B. Chen, L. Yuan, and A. Zhang, [[Phys. Rev. D [**83**]{}, 114025 (2011)](http://dx.doi.org/10.1103/PhysRevD.83.114025)]{}.
Zhi-Gang Wang, [[Phys. Rev. D [**83**]{}, 014009 (2011)](http://dx.doi.org/10.1103/PhysRevD.83.014009)]{}.
Qi-Fang L$\ddot{\text{u}}$ and De-Min Li, [[Phys. Rev. D [**90**]{}, 054024 (2014)](http://dx.doi.org/10.1103/PhysRevD.90.054024)]{}.
B. Chen, X. Liu, and A. Zhang, [[Phys. Rev. D [**92**]{}, 034005 (2015)](http://dx.doi.org/10.1103/PhysRevD.92.034005)]{}.
Guo-Liang Yu, Zhi-Gang Wang, and Zhen-Yu Li, [[Chin. Phys. C [**39**]{}, 063101 (2015)](http://dx.doi.org/10.1088/1674-1137/39/6/063101)]{}.
S. Godfrey and K. Moats, [[Phys. Rev. D [**90**]{}, 117501 (2014)](http://dx.doi.org/10.1103/PhysRevD.90.117501)]{}.
S. Godfrey and K. Moats, [[Phys. Rev. D [**93**]{}, 034035 (2016)](http://dx.doi.org/10.1103/PhysRevD.93.034035)]{}.
P. Colangelo, F. De Fazio, and G. Nardulli, [[Phys. Lett. B **478**, 408 (2000)](http://dx.doi.org/10.1016/S0370-2693(00)00291-4)]{}.
Long-Fei Gan and Ming-Qiu Huang, [[Phys. Rev. D **79**, 034025 (2009)](http://dx.doi.org/10.1103/PhysRevD.79.034025)]{}.
Long-Fei Gan [*et al*]{}., [[Eur. Phys. J. C [**75**]{}, 232 (2015)](http://dx.doi.org/10.1140/epjc/s10052-015-3449-y)]{}.
T.B. Suzuki, T. Ito, S. Sawada, and M. Matsuda, [[Prog. Theor. Phys. **91**, 757 (1994)](http://dx.doi.org/10.1143/ptp/91.4.757)]{}.
S. Veseli and M.G. Olsson, [[Phys. Rev. D **54**, 886 (1996)](http://dx.doi.org/10.1103/PhysRevD.54.886)]{}.
J.P. Lees *et al*. (BaBar Collaboration), [[Phys. Rev. Lett. [**109**]{}, 101802 (2012)](http://dx.doi.org/10.1103/PhysRevLett.109.101802)]{}.
S. Hirose *et al*. (Belle Collaboration), [arXiv:1612.00529 \[hep-ex\]](https://arxiv.org/abs/1612.00529v1).
S. Fajfer, J. F. Kamenik and I. Nišandžić, [[Phys. Rev. D [**85**]{}, 094025 (2012)](http://dx.doi.org/10.1103/PhysRevD.85.094025)]{}.
M. Tanaka and R. Watanabe, [[Phys. Rev. D 87, 034028 (2013)](http://dx.doi.org/ 10.1103/PhysRevD.87.034028)]{}.
Ying-Ying Fan, Zhen-Jun Xiao, Ru-Min Wang and Bing-Zhong Li, [[Sci. Bull. [**60**]{}: 2009 (2015)](http://dx.doi.org/10.1007/s11434-015-0959-9)]{}.
E.E. Salpeter, [[Phys. Rev. [**87**]{}, 328 (1952)](http://dx.doi.org/10.1103/PhysRev.87.328)]{}.
E. Salpeter and H. Bethe, [[Phys. Rev. [**84**]{}, 1232 (1951)](http://dx.doi.org/10.1103/PhysRev.84.1232)]{}.
Zhi-hui Wang, Guo-Li Wang, and Chao-Hsi Chang, [[J. Phys. G: Nucl. Part. Phys. [**39**]{}, 015009 (2012)](http://dx.doi.org/10.1088/0954-3899/39/1/015009)]{}.
T. Wang, Guo-Li Wang, Y. Jiang, and Wan-Li Ju, [[J. Phys. G. [**40**]{}, (2013) 035003](http://dx.doi.org/10.1088/0954-3899/40/3/035003)]{}.
Wan-Li Ju, Guo-Li Wang, Hui-Feng Fu, Zhi-Hui Wang, and Ying Li, [[JHEP [**09**]{}, 171 (2015)](http://dx.doi.org/10.1007/JHEP09(2015)171)]{}.
Q. Li, T. Wang, Y. Jiang, H. Yuan, and G.-L. Wang. [arXiv:1603.02013 \[hep-ph\]](http://arxiv.org/abs/1603.02013).
Guo-Li Wang, [[Phys. Lett. B [**633**]{}, 492 (2006)](http://dx.doi.org/10.1016/j.physletb.2005.12.005)]{}.
Guo-Li Wang, [[Phys. Lett. B [**650**]{}, 15 (2007)](http://dx.doi.org/10.1016/j.physletb.2007.05.001)]{}.
Guo-Li Wang, [[Phys. Lett. B [**674**]{}, 172 (2009)](http://dx.doi.org/10.1016/j.physletb.2009.03.030)]{}.
Tianhong Wang, Guo-Li Wang, Wan-Li Ju, and Yue Jiang, [[JHEP [**03**]{}, 110 (2013)](http://dx.doi.org/10.1007/JHEP03(2013)110)]{}.
D. Fakirov and B. Stech, [[Nucl. Phys. B $\bm{133}$, 315 (1978)](http://dx.doi.org/10.1016/0550-3213(78)90306-1)]{}.
N. Cabibbo and L. Maiani, [[Phys. Lett. B $\bm{73}$, 418 (1978)](http://dx.doi.org/10.1016/0370-2693(78)90754-2)]{}.
M. Bauer, B. Stech and M. Wirbel, [[ Z. Phys. C [**34**]{}, 103 (1987)](http://dx.doi.org/10.1007/BF01561122)]{}.
A. Ali, G. Kramer and C. D. Lu, [[ Phys. Rev. D [**58**]{}, 094009 (1998)](http://dx.doi.org/10.1103/PhysRevD.58.094009)]{}.
C.-H. Chang and Y.-Q. Chen, [[Phys. Rev. D [**49**]{}, 3399 (1994).](http://dx.doi.org/10.1103/PhysRevD.49.3399)]{}
P. Colangelo and F. De Fazio, Phys. Rev. D [**61**]{}, 034012 (2000) doi:10.1103/PhysRevD.61.034012 \[hep-ph/9909423\].
M. A. Ivanov, J. G. K$\rm \ddot{o}$rner, and P. Santorelli, [[Phys. Rev. D [**73**]{}, 054024 (2006)](http://dx.doi.org/10.1103/PhysRevD.73.054024)]{}.
E. Hern$\rm \acute{a}$ndez, J. Nieves, and J. M. Verde-Velasco, [[Phys. Rev. D [**74**]{}, 074008 (2006)](http://dx.doi.org/10.1103/PhysRevD.74.074008)]{}.
R.N. Faustov and V.O. Galkin, [[Phys. Rev. D. [**87**]{}, 034033 (2013)](http://dx.doi.org/10.1103/PhysRevD.87.034033)]{}.
M. J. Dugan and B. Grinstein, [[Phys. Lett. B [**255**]{}, 583 (1991)](http://dx.doi.org/10.1016/0370-2693(91)90271-Q)]{}.
M. Beneke, G. Buchalla, M. Neubert and C. T. Sachrajda, Phys. Rev. Lett. [**83**]{}, 1914 (1999) doi:10.1103/PhysRevLett.83.1914 \[hep-ph/9905312\]
Y. Y. Keum and H. n. Li, Phys. Rev. D [**63**]{}, 074006 (2001) doi:10.1103/PhysRevD.63.074006 \[hep-ph/0006001\].
S. Mandelstam, [[Proc. Roy. Soc. A [**233**]{}, 248 (1955)](http://dx.doi.org/10.1098/rspa.1955.0261)]{}.
C.S. Kim and Guo-Li Wang, [[Phys. Lett. B [**584**]{}, 285 (2004)](http://dx.doi.org/10.1016/j.physletb.2004.01.058)]{}.
T. Wang, H.-F. Fu, Y. Jiang, Q. Li, and G.-L. Wang, [arXiv:1601.01047 \[hep-ph\]](http://arxiv.org/abs/1601.01047).
L. Bergstr$\rm {\ddot{o}}$m, H. Grotch, and R.W. Robinett, [[Phys. Rev. D. [**43**]{}, 7 (1991)](http://dx.doi.org/10.1103/PhysRevD.43.2157)]{}.
T. Matsuki, T. Morii, and K. Seo, [[Prog. Theo. Phys. [**124**]{}, 2 (2010)](http://dx.doi.org/10.1143/PTP.124.285)]{}.
D. Ebert, R.N. Faustov, and V.O. Galkin, [[Eur. Phys. J. C [**66**]{}, 197 (2010)](http://dx.doi.org/10.1140/epjc/s10052-010-1233-6)]{}.
K.A. Olive [*et al*]{}. (Particle Data Group), [[Chin. Phys. C [**38**]{}, 090001 (2014)](http://dx.doi.org/10.1088/1674-1137/38/9/090001)]{}.
N. Devlani, V.H. Kher, and A.K. Rai, [[EPJ. Web. Conf. [**95**]{}, 05006 (2015)](http://dx.doi.org/10.1051/epjconf/20159505006)]{}.
N. Devlani and A.K. Rai, [[Eur. Phys. J. A [**48**]{}, 104 (2012)](http://dx.doi.org/10.1140/epja/i2012-12104-8)]{}.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
We present a pulsational stability analysis of hot post-AGB H-deficient pre-white dwarf stars with active He-burning shells. The stellar models employed are state-of-the-art equilibrium structures representative of PG1159 stars derived from the complete evolution of the progenitor stars, through the thermally pulsing AGB phase and born-again episode. On the basis of fully nonadiabatic pulsation computations, we confirmed theoretical evidence for the existence of a separate PG1159 instability strip in the $\log T_{\rm eff} - \log
g$ diagram characterized by short-period $g$-modes excited by the $\epsilon$-mechanism. This instability strip partially overlaps the already known GW Vir instability strip of intermediate/long period $g$-modes destabilized by the classical $\kappa$-mechanism acting on the partial ionization of C and/or O in the envelope of PG1159 stars. We found that PG1159 stars characterized by thick He-rich envelopes and located inside this overlapping region could exhibit both short and intermediate/long periods simultaneously.
As a natural application of our results, we study the particular case of , a pulsating planetary nebula nucleus (PG1159-type) particularly interesting because it has been reported to exhibit a rich and complex pulsation spectrum including a series of unusually short pulsation periods. We found that the long periods exhibited by can be readily explained by the classical $\kappa$-mechanism, while the observed short-period branch below $\approx 300$ s could correspond to modes triggered by the He-burning shell through the $\epsilon$-mechanism, although more observational work is needed to confirm the reality of these short-period modes. Were the existence of short-period $g$-modes in this star convincingly confirmed by future observations, could be the first known pulsating star in which both the $\kappa$-mechanism and the $\epsilon$-mechanism of mode driving are *simultaneously* operating.
author:
- 'A. H. Córsico'
- 'L. G. Althaus'
- 'M. M. Miller Bertolami'
- 'J. M. González Pérez'
- 'S. O. Kepler'
title: 'On the possible existence of short-period $g$-mode instabilities powered by nuclear burning shells in post-AGB H-deficient (PG1159-type) stars'
---
4334
Introduction {#intro}
============
At present, the study of stellar pulsations constitutes one of the most fundamental pillars on which the building of stellar astrophysics rests on. Although the theory of stellar pulsations was initially elaborated to explain the existence of classical variable stars such as Cepheids and RR Lyrae, in the last few decades the study of pulsating stars has been extended to many other different kinds of stars that were regarded as constant stars before (e.g., Unno et al. 1989; Gautschy & Saio 1995). Nowadays, new classes of pulsating stars across the HR diagram are being routinely uncovered from ground-based observations as well as space missions (e.g., CoRoT, MOST; see Aerts et al. 2008). The study of stellar pulsations through the approach of asteroseismology constitutes a powerful tool to probe the internal structure and evolution of stars.
Most of the pulsations exhibited by pulsating stars are self-excited through the classical $\kappa$-mechanism operating in a partial ionization zone near the surface of stars (Cox 1980; Unno et al. 1989). As a matter of fact, this mechanism is responsible for pulsations of the stars in the classical instability strip due to partial ionization of H and HeI and/or HeII. In the driving zone, the opacity perturbation increases outward so that radiative luminosity is blocked in the compression phase of pulsation. The region gains thermal energy in the compression phase and it loses thermal energy in the expansion phase.
A less common —and consequently less explored— pulsation driving mechanism in stars is the $\epsilon$-mechanism. This mechanism is due to vibrational instability induced by thermonuclear reactions. In this case, the driving is due to the strong dependence of nuclear burning on temperature. During maximum compression, the temperature and thus the nuclear energy production rates are higher than at equilibrium. So, in the layers where nuclear reactions take place, thermal energy is gained at compression while the opposite happens during the expansion phase (Unno et al. 1989; Gautschy & Saio 1995). An excellent historical account of studies on vibrational destabilization of stars by the $\epsilon$-mechanism, can be found in Kawaler (1988) —we refer the interested reader to that paper for details.
In this paper, we explore the $\epsilon$-mechanism in connection with pulsating PG1159 stars. These stars, also called GW Vir or DOV, are very hot H-deficient post-Asymptotic Giant Branch (AGB) stars with surface layers rich in He, C and O (Werner & Herwig 2006) that exhibit multiperiodic luminosity variations with periods ranging from 300 to 6000 s, attributable to non-radial $g$-modes driven by the $\kappa$-mechanism acting on the region of partial ionization of C and O (Starrfield et al. 1983, 1984, 1985; Gautschy 1997; Quirion et al. 2004; Gautschy et al. 2005; Córsico et al. 2006; Quirion et al. 2007). Some pulsating PG1159 stars are still embedded in a nebula and are called Planetary Nebula Nuclei Variable (PNNV) stars (see Winget & Kepler 2008 and Fontaine & Brassard 2008 for recent reviews).
Evolutionary models of PG1159 stars with thick He-rich envelopes located at the upper left portion of the HR diagram are characterized by the presence of vigorous He-burning shells. The first attempt to study the effect of the $\epsilon$-mechanism induced by a He-burning shell in H-deficient pre-white dwarf stars was the seminal work by Kawaler et al. (1986). These authors found some $g$-modes excited through this mechanism with periods in the range 70 to 200 s. Observationally, however, no signature of these short pulsation periods was found in the surveys of planetary nebula nuclei conducted at that time (Grauer et al. 1987; Hine & Nather 1987). Later on, stability analysis on simplified PG1159 models by Saio (1996) and Gautschy (1997) also predicted unstable $g$-modes driven by the $\epsilon$-mechanism with periods in the range $110-150$ s.
The interest in the $\epsilon$-mechanism in the context of H-deficient post-AGB stars has recently been renewed by the discovery of luminosity variations in the PNNV star ($T_{\rm eff}=
130\,000 \pm 5000$ K, $\log g= 7 \pm 0.5$, C/He= 1.5 and O/He= 0.4; Werner & Herwig 2006) by González Pérez et al. (2006). The most outstanding feature of is that its period spectrum appears to include a series of unusually short pulsation periods ($\sim 130-300$ s), the shortest periods ever detected in a pulsator of its class. González Pérez et al. (2006) (see also Solheim et al. 2008) speculate that these rapid oscillations could be excited by the $\epsilon$-mechanism.
In this work, we largely extend the pioneering work by Kawaler et al. (1986), Saio (1996) and Gautschy (1997) by performing fully nonadiabatic pulsation computations on realistic PG1159 models extracted from full evolutionary sequences with a wide range of stellar masses and effective temperatures. In particular, we gather strong evidence for the existence of a new short-period $g$-mode instability strip of pulsating PG1159 stars due to the $\epsilon$-mechanism. In addition, we examine the possibility that the short-period $g$-modes of could be excited by this mechanism. The paper is organized as follow: in the next section we briefly describe the input physics of the PG1159 evolutionary sequences analyzed and the nonadiabatic treatment of the pulsations. In Sect. \[results\] we describe the stability analysis. In Sect. \[vv47\] we present the application to the star , and in Sect \[pg1159\] we discuss the case of the prototypical DOV star in the context of our theoretical findings. Finally, in Sect. \[summary\] we summarize our main results and make some concluding remarks.
Evolutionary/pulsational modelling of PG1159 stars {#evolutionary}
==================================================
The PG1159 equilibrium models on which the present investigation rests on were extracted from the evolutionary calculations presented by Althaus et al. (2005), Miller Bertolami & Althaus (2006) and Córsico et al. (2006), who computed the complete evolution of model star sequences with initial masses on the ZAMS in the range $1 - 3.75
M_{\odot}$. The evolutionary tracks for the H-deficient pre-white dwarf remnants are displayed in Fig. \[fig01\]. All of the post-AGB evolutionary sequences, computed with [LPCODE]{} (Althaus et al. 2005), were followed through the very late thermal pulse (VLTP) and the resulting born-again episode that gives rise to the H-deficient, He-, C-, and O-rich composition characteristic of PG1159 stars. For details about the input physics and evolutionary code used, and the numerical simulations performed to obtain the PG1159 evolutionary sequences employed here, we refer the interested reader to the works mentioned above. One distinctive feature, and crucial for this study, that is common to all of our sequences, is that the PG1159 models are characterized by He-rich envelopes thick enough as to sustain active He-burning shell sources during the evolutionary stages of interest. This is at variance with the non-standard PG1159 models employed in Althaus et al. (2008) to explain the $\dot{\Pi}$ values in , which are characterized by thin He-rich envelopes and so they are not able to sustain an active He-burning shell.
![The PG1159 evolutionary tracks of Althaus et al. (2005), Miller Bertolami & Althaus (2006) and Córsico et al. (2006), with stellar masses of (from right to left): $M_*= 0.515, 0.530, 0.542,
0.565, 0.589, 0.609, 0.664, 0.741 M_{\odot}$. Also shown is the location of known PG1159 stars. The error bars for are displayed.[]{data-label="fig01"}](fig01.eps){width="250pt"}
The pulsational stability analysis presented in this work was carried out with the linear, nonradial, nonadiabatic pulsation code described in Córsico et al. (2006). The “frozen-in convection” approximation was assumed because the flux of heat carried by convection is negligible in PG1159 stars. At variance with Córsico et al. (2006), in this work we have fully taken into account the $\epsilon$-mechanism for mode driving operating in the He-shell nuclear-burning region. Because we are interested in PG1159 stars which are H-deficient, we are only concerned with the He-burning reactions. Fortunately, because $g$-mode pulsation timescales are much shorter than the timescales of nucleosynthesis, possible phase delays between the temperature perturbations and the abundance variations are unimportant. Hence, they can be neglected, largely simplifying the pulsational stability analysis (Unno et al. 1989; Kawaler et al. 1986)[^1].
![The dipole ($\ell= 1$) normalized growth rates $\eta$ (black dots connected with continuous lines) in terms of the pulsation periods for a $0.530 M_{\odot}$ PG1159 template model located before the evolutionary knee ($T_{\rm eff}= 138\, 400$ K, $\log (L_*/L_{\odot})= 3.14$). Numbers indicate the radial order $k$ for low order modes. The large numerical range spanned by $\eta$ is appropriately scaled for a better graphical representation. The two ranges of overstable $g$-modes —one due to the $\kappa$-mechanism and the other induced by the $\epsilon$-mechanism— are clearly discernible. Plus symbols connected with dotted lines correspond to the case where the $\epsilon$-mechanism is explicitly suppressed in the stability calculations.[]{data-label="fig02"}](fig02.eps){width="250pt"}
Nonadiabatic results {#results}
====================
We analyzed the stability properties of about 4000 stellar models covering a wide range of effective temperatures ($5.5 \gtrsim \log(T_{\rm eff}) \gtrsim 4.7$) and stellar masses ($0.515 \lesssim M_*/M_{\odot}\lesssim 0.741$). For each model we restricted our study to $\ell= 1$ $g$-modes with periods in the range $50-7\,000$ s.
A single template model
-----------------------
We start our description by focusing on a $0.530 M_{\odot}$ PG1159 template model with $T_{\rm eff}= 138\,400$ K and $\log
(L_*/L_{\odot})= 3.14$ located before the evolutionary knee in Fig. \[fig01\]. The surface chemical composition of the model is $X({^4{\rm He}})= 0.33$, $X({^{12}{\rm C}})= 0.39$, $X({^{13}{\rm C}})= 0.05$, $X({^{14}{\rm N}})= 0.02$, and $X({^{16}{\rm O}})= 0.17$. Fig. \[fig02\] displays the normalized $\ell= 1$ growth rates $\eta= -\Re(\sigma)/\Im(\sigma)$ (where $\Re(\sigma)$ and $\Im(\sigma)$ are the real and the imaginary parts, respectively, of the complex eigenfrequency $\sigma$) in terms of the pulsation periods ($\Pi$) corresponding to our template model. In the interests of a better graphical representation, the huge numerical range spanned by $\eta$ is appropriately scaled (see Gautschy 1997). The sign function allows to discriminate between stable and unstable modes. The presence of two well-defined families of overstable $g$-modes, one at the intermediate- and long-period regime, and the other one at the short period regime, is apparent. The first group of periods ($\approx
750-1800$ s) corresponds to modes driven by the well-known $\kappa$-mechanism operating at the region of the opacity bump due to partial ionization of C and O, centered at $\log T \approx 6.2$ (Gautschy et al. 2005, Córsico et al. 2006). The second group of periods, which are associated to low radial order $g$-modes, are destabilized by the action of the vigorous He-shell burning through the $\epsilon$-mechanism.
The short-period instabilities uncovered here are of the same nature than those found by Kawaler et al. (1986) in the context of H-deficient hot central stars of planetary nebulae. Here, as in that work, the $\epsilon$-mechanism induced by the He-shell burning constitutes the source of driving. In absence of this destabilizing agent, the overstable modes with periods in the range ($\approx
100-180$ s) turn out to be stable, while the remainder modes of the pulsation spectrum remains unchanged. This is vividly displayed in Fig. \[fig02\], that shows with plus symbols the results of additional stability computations in which the nuclear energy production rate, $\epsilon$, and the logarithmic derivatives $\epsilon_T= \left(\frac{\partial \ln \epsilon}{\partial \ln
T}\right)_{\rho}$ and $\epsilon_{\rho}= \left(\frac{\partial \ln \epsilon}{\partial
\ln \rho} \right)_T$, are forced to be zero in the pulsation equations. It is worth emphasizing that in the present effort we are able to obtain destabilization of $g$-modes through both the $\kappa$-mechanism *and* the $\epsilon$-mechanism in the same PG1159 equilibrium model. This is at variance with the study by Kawaler et al. (1986), who reported only $\epsilon$-destabilized modes.
The $\epsilon$-mechanism behaves as an efficient filter of modes that destabilizes only those $g$-modes that have their largest maximum of the temperature perturbation ($\delta T/T$) in the narrow region of the He-burning shell (see Kawaler et al. 1986). In the specific case of our template model, only the $g$-modes with $k= 3,
4, 5$, and $6$ meet such a condition and, as a result, they are $\epsilon$-destabilized. For $k= 1, 2$ the largest maximum of $\delta T/T$ lies at inner layers with respect to the He-burning shell. Thus, these modes are stable. For modes with $k \geq 7$ the opposite is true and these modes also are stable.
Test stability calculations with $\ell= 2$ for our template model indicate that there exist only one quadrupole $\epsilon$-destabilized $g$-mode which corresponds to $k= 5$ with a period $\Pi \sim 95$ s, about $40 \%$ shorter than the corresponding $k= 5$ mode with $\ell=
1$ ($\Pi \sim 157$ s). Hence, in general, a narrower range of shorter periods is expected to be associated with $\epsilon$-destabilized $g$-modes with $\ell= 2$ as compared with the case of $\ell= 1$.
A new PG1159 instability strip
------------------------------
Having described our results for a single template model, we now are in conditions to examine the location and extension of the complete unstable domain associated with the $\epsilon$-mechanism. Our results are depicted in Fig. \[fig03\], which displays the instability strip of $\epsilon$-destabilized modes in the $\log T_{\rm eff}-\log g$ drawn with thick black curves along the PG1159 evolutionary tracks. In addition, the GW Vir instability domain of $\kappa$-destabilized modes (see Córsico et al. 2006) is depicted with red (gray) lines along the tracks. Note that the instability strip for $\epsilon$-destabilized modes partially overlaps the domain of $\kappa$-destabilized modes. So, our results indicate the existence of three well-defined instability regimes in the $\log T_{\rm eff}-\log
g$ plane: a regime —splitted into two regions, one at low gravity and the other at high gravity— in which stellar models harbour intermediate/long period $g$-modes excited by the $\kappa$-mechanism only, another one corresponding to short-period modes destabilized by the $\epsilon$-mechanism only, and finally a region in which models experience pulsational destabilization by the $\kappa$-mechanism and the $\epsilon$-mechanism of driving simultaneously. Notably, the region corresponding to the $\epsilon$-mechanism only is not occupied by any known PG1159 star (see Fig. \[fig03\]).
We stress that in previous works (Kawaler et al. 1986; Saio 1996; Gautschy 1997) only *some* short-period $g$-modes were found to be destabilized by the $\epsilon$-mechanism. Needles to say, due to the very few $\epsilon$-destabilized modes found in those exploratory works, no clear extension and location of the $\epsilon$-mechanism instability domain were obtained, thus hampering those authors from making further consideration of such modes. At variance with those works, in the present study we are able to find a *complete* instability strip of $\epsilon$-destabilized modes.
The degree of driving, and the place that it might occur in the $\log T_{\rm eff}- \log g$ diagram, is sensitive to the stellar mass, previous evolutionary history, and so on. Thus, due to the uncertainties in the stellar evolution modelling (overshooting, nuclear reaction rates, etc), the surface and internal composition of PG1159 stars are not known in detail, and so a clear instability domain for $\epsilon$-destabilized pulsations is difficult to drawn. So, what is shown in the Fig. \[fig03\] is the shape and location of the $\epsilon$-mechanism instability strip obtained by us under the particular assumptions adopted in the construction of the PG1159 evolutionary models of Miller Bertolami & Althaus (2006). The extension and location of this instability domain might change if other assumptions for the evolutionary history of the progenitor stars were adopted.
![Same as Fig. \[fig01\], but including the loci of models having $\ell= 1$ (dipole) $\kappa$-destabilized modes with solid red (gray) curves along the tracks, and models harbouring short-period $\epsilon$-destabilized modes according to the present study. Superposition of both types of curves corresponds to stellar models with both $\epsilon$- and $\kappa$-destabilized modes (shaded area). The location and designation of relevant PG1159 stars is also shown \[Color figure only available in the electronic version of the article\].[]{data-label="fig03"}](fig03.eps){width="250pt"}
All of the overstable $\epsilon$-destabilized modes computed in this work are characterized by very tiny ($10^{-9}-10^{-12}$) linear growth rates $\eta$, by far smaller than those characterizing overstable modes excited by the $\kappa$-mechanism ($10^{-8} \lesssim \eta \lesssim 10^{-4}$). So, the question rises about what would be the chance for a given $\epsilon$-destabilized mode to have plenty of time for developing observable amplitudes. To analyze this question we consider the time interval that the models spend crossing the instability strip of $\epsilon$-destabilized modes, $\Delta t$, and the maximum and minimum $e$-folding times $\tau_e^{\rm max}$ and $\tau_e^{\rm min}$, respectively, of the unstable modes for a given stellar mass. The $e$-folding times are defined as $\tau_e \equiv 1/|\Im(\sigma)|$, such that the time dependence of the amplitude of the pulsations is given by $\xi(t) \propto e^{i \sigma t}$, and $\Im(\sigma) < 0$ for overstable modes.
The values of $\Delta t$, $\tau_e^{\rm min}$, and $\tau_e^{\rm max}$ are provided in Table \[table1\] for each value of the stellar mass. Note that the three timescales monotonically decrease for increasing stellar mass. For all of our PG1159 sequences we found that the most unstable modes —those with the smaller values of $\tau_e$— are found near the low-gravity (high-luminosity) boundary of the instability domain (upper black dashed line in Fig. \[fig03\]), when the models are still evolving to the blue before reaching the evolutionary knee. On the contrary, when models are already evolving toward the white dwarf cooling track, the $\epsilon$-destabilized modes are only marginally unstable, and so they are characterized by large $e$-folding times.
Table \[table1\] shows that $\tau_e^{\rm min} \ll \Delta t$ for all of our sequences. This means that $g$-modes that are destabilized at epochs before the evolutionary knee, characterized by short $e$-folding times, have time enough to reach observable amplitudes before the star leaves the instability strip. On the other hand, it is apparent that $\tau_e^{\rm max} \gtrsim \Delta t$. Thus, the $g$-modes that are destabilized in models close to the high-gravity limit (low-luminosity) of the instability strip (lower black dashed line in Fig. \[fig03\]) have little —or even null— chances to develop observable amplitude before the model abandons the instability domain.
In summary, our computations predict that some $g$-modes (those with short $\tau_e$) could have plenty of time to grow and finally develop observable amplitudes. We caution, however, that this prediction is based on a *linear* stability analysis, and that the last word should came from a detailed non-linear description of nonadiabatic pulsations. Such a nonlinear treatment is not available at the present stage. Also, there are other effects (stellar winds, diffusion, etc) suspected to be present in real stars, that could be affecting the growth of pulsations. The assessment of their effects on the modes predicted to be unstable in the frame of our analysis is beyond the scope of the present study.
----------------- -------------------- -------------------- --------------------
$M_*/M_{\odot}$ $\tau_e^{\rm min}$ $\tau_e^{\rm max}$ $\Delta t$
$0.515$ $3410$ $1.5 \times 10^6$ $1.60 \times 10^5$
$0.530$ $2580$ $1.0 \times 10^6$ $1.01 \times 10^5$
$0.542$ $1610$ $3.8 \times 10^5$ $5.95 \times 10^4$
$0.565$ $1400$ $1.3 \times 10^5$ $2.78 \times 10^4$
$0.589$ $1160$ $1.0 \times 10^5$ $2.47 \times 10^4$
$0.609$ $742 $ $4.7 \times 10^4$ $1.26 \times 10^4$
$0.664$ $361 $ $1.8 \times 10^4$ $4830 $
$0.741$ $180 $ $7000$ $1570 $
----------------- -------------------- -------------------- --------------------
: The minimum and maximum $e$-folding times (in yr), and the time (in yr) that PG1159 models spend within the instability strip of $\epsilon$-destabilized modes.
\[table1\]
The next step in our analysis is to derive the range of periods ($\Pi$) of overstable $\epsilon$-destabilized modes. Fig. \[fig04\] displays the regions of the $\kappa$-mechanism instability domains in the $\log T_{\rm eff}-\Pi$ diagram, depicted with lines of different colours for the various stellar masses. Notably, the figure also shows the presence of a separate, well-defined instability domain for a broad range of effective temperatures ($5.46 \gtrsim \log T_{\rm eff } \gtrsim 4.99$) and pulsation periods in the interval $55 \lesssim \Pi \lesssim 200$ s, associated to $\epsilon$-destabilized $g$-modes with radial orders ranging from $2$ to $5$ for $M_*= 0.515 M_{\odot}$ and from $3$ to $8$ for $M_*= 0.741 M_{\odot}$. The stages corresponding to phases before (after) the evolutionary knee are depicted with small dot (plus) symbols. A close inspection of the figure reveals that for the low-mass models, most of modes are destabilized after the evolutionary knee. For the high-mass models the situation is reversed, that is, most of overstable modes are destabilized before the evolutionary knee. The existence of this new instability domain of short-period $g$-modes in stellar models representative of PG1159 stars is the main result of our study.
In particular, it is worth emphasizing that the $\epsilon$-mechanism should be active in a PG1159 star irrespective of the precise abundances of He, C, and O at the surface, because in this case the mode excitation takes place at deep layers in the star. This is at variance with the $\kappa$-mechanism, which is strongly dependent on the exact O/C/He abundances at the driving regions (see Quirion et al. 2007).
![The dipole ($\ell= 1$) instability domains for overstable $\kappa$-destabilized $g$-modes, shown with thin lines of different colours for the various stellar masses. The $\epsilon$-mechanism instability domain is emphasized with a shaded area. Short-period dipole unstable $\epsilon$-destabilized $g$-modes are depicted with dot (plus) symbols for stages before (after) the evolutionary knee. Numbers indicate the radial order of the modes. \[Color figure only available in the electronic version of the article\].[]{data-label="fig04"}](fig04.eps){width="250pt"}
The case of the PNNV star {#vv47}
--------------------------
An immediate prediction of the present study is that PG1159-type stars populating the overlapping region of $\kappa$- and $\epsilon$-destabilized modes in the $\log T_{\rm eff}-\log g$ diagram (the shaded region in Fig. \[fig03\]) should exhibit both short- and intermediate/long-period luminosity variations simultaneously. Table \[table2\] lists the known PG1159 candidate stars. A glance of this table leads us to a somewhat disappointing conclusion: most of the stars located in the region of interest are not variables at all or have not been scrutinized for variability. Other stars, at most, exhibit intermediate/long-period luminosity variations which are typical of the high/intermediate-order $g$-modes driven by the $\kappa$-mechanism, but not the expected short periods typical of $\epsilon$-destabilized modes. In particular, this is the case for the prototype DOV star, .
There is one object, the PNNV star , which is suspected to pulsate in short- and long-period modes (González Pérez et al. 2006). This star ($T_{\rm eff}= 130\,000 \pm 5000$ K, $\log g= 7 \pm
0.5$, C/He= 1.5 and O/He= 0.4; Werner & Herwig 2006) was first observed as potentially variable by Liebert et al. (1988). Later, it was monitored by Ciardullo & Bond (1996), but no clear variability was found. Finally, González Pérez et al. (2006) were able to confirm the —until then, elusive— intrinsic variability of for the first time. They found evidence that the pulsation spectrum of this star is extremely complex. The most outstanding feature of is the presence of high-frequency peaks (at periods $\sim 130-300$ s) in the power spectrum, which could be serious candidates for low-$k$ radial order $g$-modes triggered by the $\epsilon$-mechanism.
We decided to test the attractive possibility that the short-periods observed in could be due to the $\epsilon$-mechanism. We first estimated the stellar mass of . From the location of in the $\log T_{\rm eff}- \log g$ plane (see Fig. \[fig01\]) it is apparent that the spectroscopic mass of is of $\approx 0.525 M_{\odot}$. In addition, a preliminary adiabatic asteroseismological analysis on this star indicates that the seismological mass of —obtained from the period spacing data ($\Delta \Pi \approx 24$ s)— is of $\approx
0.52-0.53 M_{\odot}$, in excellent agreement with the spectroscopic derivation. So, we shall focus on the case of the evolutionary sequence of $M_*= 0.530 M_{\odot}$. This sequence is characterized by a thick He-rich envelope ($M_{\rm env} \sim 0.045 M_{\odot}$).
We would like to see how well the theoretical ranges of periods of unstable modes corresponding to this sequence fit the observed period spectrum of . Fig. \[fig05\] displays the regions of the $\kappa$-mechanism instability domain (light and dark grey) for the $0.530 M_{\odot}$ sequence. The figure also shows the presence of a well-defined instability domain ($77\lesssim \Pi \lesssim 180$ s) that corresponds to $\epsilon$-destabilized $g$-modes with $k= 2,\cdots,6$ (large and small dots). Also depicted in the plot are the periods reported by González Pérez et al. (2006) for with their corresponding uncertainties in $T_{\rm eff}$. We have emphasized with black small circles the periods associated with modes having the best chances to be real, according to González Pérez et al. (2006). It is apparent that, whereas most of the long periods observed in are qualitatively explained by the $\kappa$-mechanism when the model star is before the evolutionary knee, the short-period branch (below $\sim 300$ s) of the pulsation spectrum of the star is not accounted for at all by the theoretical domains corresponding to this destabilizing agent. We can see, instead, that the short periods of —in particular $\Pi \lesssim 200$ s— are satisfactorily accounted for by the $\epsilon$-destabilized $g$-modes. Note, however, that if only periods detected with sufficiently high significance (black filled circles) are used to compare with our theoretical predictions, then the period at 261.4 s cannot be explained by a low-order $g$-mode excited by the $\epsilon$-mechanism. In fact, this period is considerably longer than the longest period ($\approx 180$ s) of the $g$-modes which can be excited by the $\epsilon$-mechanism as our analysis predicts.
Clearly, more observational work is needed to put the reality of the short periods in on a solid basis. Were the existence of these short periods confirmed by future observations, then they could be attributed to the $\epsilon$-mechanism, and this could be indicating that should have a *thick* He-rich envelope as to support an active He-burning shell.
Star PN Variable Period range \[s\]
---- --------------- ----- ---------- --------------------
1 PG 1159$-$035 no yes $430-840$
2 NGC 650$-$1 yes no $-$
3 VV 47 yes yes (?) $\sim$ 260
$\sim$ 2170-4300
4 PG 1144+005 no no $-$
5 Jn 1 yes yes (?) $454-1860$
6 Abell 21 yes no $-$
7 K 1$-$16 yes yes $1500-1700$
8 Longmore 3 yes no $-$
9 HS 2324+3944 no yes $2005-2570$
10 HS 1517+7403 no no $-$
11 PG 1424+535 no no $-$
12 IW 1 yes no $-$
13 Sh 2$-$68 yes ? $-$
14 HS 0704+6153 no ? $-$
: Known PG1159 stars populating the overlapping instability region of $\epsilon$- and $\kappa$-destabilized modes.
\[table2\]
![The $\ell= 1$ regions of the $\kappa$-mechanism instability domain, shown with light (dark) grey for stages before (after) the evolutionary knee, corresponding to the $0.530 M_{\odot}$ sequence. Arrows indicate the time sense of evolution. Also shown is the evolution of the periods corresponding to the $\epsilon$-destabilized modes $g_2, \cdots, g_6$, with large (small) dots for stages before (after) the evolutionary knee. Finally, the periods reported by González Pérez et al. (2006) for with their corresponding uncertainties in $T_{\rm eff}$ are displayed with small circles. Periods detected with sufficiently high significance are emphasized with black filled circles.[]{data-label="fig05"}](fig05.eps){width="250pt"}
The case of the DOV star {#pg1159}
-------------------------
Another consequence of our investigation concerns the pulsating star PG 1159$-$035, the prototype of the class and the best-studied DOV. Indeed, note from Fig. \[fig03\] that a trend of our results is that this variable star should exhibit short-period $\epsilon$-destabilized modes if the thick He-rich envelopes derived from our evolutionary calculations were representative of the star. These modes are not observed by Costa et al. (2007). This result suggests that the He-burning shell may not be active in . This would indicate that this star has a thinner He-rich envelope than what is traditionally derived from standard evolutionary calculations, in line with the recent finding by Althaus et al. (2008) that a thinner He-rich envelope (at least a factor of two below of the value predicted by the standard evolution theory) for PG 1159$-$035 should be invoked to alleviate the longstanding discrepancy between the observed (Costa & Kepler 2008) and the theoretical (Córsico et al. 2008) rates of period change in that star.
If the short periods observed in were confirmed, then we should face the problem of the coexistence of two PG1159 stars located very close each other in the $\log T_{\rm eff} -\log g$ diagram (see Fig. \[fig03\]) but with substantially different thickness of the He-rich envelopes. This would suggest that these stars could have had a different evolutionary history, a suggestion reinforced by the fact that still retains a planetary nebula while PG 1159$-$035 does not.
Summary and conclusions {#summary}
=======================
In this paper, we have presented a fully nonadiabatic stability analysis on state-of-the-art PG1159 models generated taking into account the complete evolution of progenitor stars, through the thermally pulsing AGB phase and born-again episode. We have explored the possibility that nonradial $g$-mode pulsations could be destabilized by a He-burning shell through the $\epsilon$-mechanism. Our study covers a broad range of stellar masses and effective temperatures. We confirm and extend the pioneering work of Kawaler et al. (1986), Saio (1996) and Gautschy (1997) on this topic. The main results are the following:
- We found strong theoretical evidence for the existence of a separate, well-defined PG1159 instability strip in the $\log T_{\rm eff} - \log
g$ diagram characterized by short-period $g$-modes excited by the $\epsilon$-mechanism due to the presence of active He-burning shells. Notably, this instability strip partially overlaps the already known GW Vir instability strip due to the $\kappa$-mechanism acting on the partial ionization of C and/or O in the envelope of the PG1159 stars. We emphasize that while in previous works only some short-period $g$-modes were found to be destabilized by the $\epsilon$-mechanism, in the present study we found a *complete* instability strip of $\epsilon$-destabilized modes.
- At variance with the classical $\kappa$-mechanism responsible for the intermediate/long-period GW Vir pulsations, the $\epsilon$-mechanism should be efficient even in PG1159 stars with low C and O content in their envelopes.
- The $\epsilon$-driven $g$-modes that are destabilized at epochs before the evolutionary knee are characterized by short $e$-folding times (between $\approx 180$ yr for $M_*= 0.741 M_{\odot}$ and $\approx 3000$ yr for $M_*= 0.515 M_{\odot}$), and so they probably have time enough as to reach observable amplitudes before the star leaves the instability strip. Note, however, that nonlinear effects, or the presence of a variety of phenomena such as stellar winds or diffusion, could affect the growth of pulsations.
- We have closely examined the case of , the only PG1159 star for which observational evidence of the presence of short-period $g$-modes exists (González Pérez et al. 2006). For this star we have derived for the first time a seismological mass of $\approx
0.52-0.53 M_{\odot}$, in excellent agreement with the spectroscopic mass ($\approx 0.525 M_{\odot}$). If we accept that all of the periods reported by González Pérez et al. (2006) are real, our stability analysis provides very strong support to the idea that the physical origin of the short periodicities could be the $\epsilon$-mechanism powered by an active He-burning shell, whereas the long-period branch of the period spectrum of this star should be due to the $\kappa$-mechanism acting on the region of partial ionization of C and O. However, if only periods detected with sufficiently high significance are taken into account, then the period at 261.4 s can not be explained by a low-order $g$-mode excited by the $\epsilon$-mechanism.
- We speculate that the absence of short periods ($\lesssim 300$ s) in the pulsation spectrum of PG 1159$-$035 could be indicating that the He-burning shell may not be active in this star. This would indicate that PG 1159$-$035 has a thinner He-rich envelope than what is traditionally derived from standard evolutionary computations.
In light of our results, if the reality of the short periods of were confirmed by follow-up observations, this star could be the first known pulsating PG1159 star undergoing nonradial $g$-modes destabilized by the $\epsilon$-mechanism. Even more, could be the first known pulsating star in which both the $\kappa$-mechanism and the $\epsilon$-mechanism of mode driving are *simultaneously* operating. Further time-series photometry of will be needed to firmly establish the reality of the short-period pulsations detected in this star.
On the other hand, the apparent absence of short-period pulsations in the remainder variable stars —such as K 1$-$16, HS 2324+3944, and Jn 1— could be an indication that, like , they are characterized by thin He-rich envelopes, as a result of which they should lack of stable He-shell burning. Another possibility is that short-period pulsations could be indeed present in these stars, but with very low amplitudes, below the actual detection limits.
Also, it is quite intriguing the absence of both short- and intermediate/long-period pulsations in the up to now constant stars (NGC 650$-$1, PG 1144+005, Abell 21, Longmore 3, HS 1517+7403, PG 1424+535, IW 1) that populate the overlapping region of the $\epsilon$- and $\kappa$-destabilized modes. In any case, extensive searches for low amplitude intrinsic variability in these stars and also in the stars Sh 2$-$68 and HS 0704+6153, which have not been observed for variability yet, should be worth doing in order to test the existence of the new theoretical instability strip uncovered in this work.
This paper has been benefited from the valuable suggestions and comments of an anonymous referee. Part of this work was supported by AGENCIA through the Programa de Modernización Tecnológica BID 1728/OC-AR, and by the PIP 6521 grant from CONICET. This research has made use of NASA’s Astrophysics Data System. Finally, we thank H. Viturro and R. Martínez for technical support.
Aerts, C., Christensen-Dalsgaard, J., Cunha, M. & Kurtz, D. W. 2008, Solar Phys. 251, 3 Althaus, L. G., Córsico, A. H., Miller Bertolami, M. M., García-Berro, E., & Kepler, S. O. 2008, ApJL, 677, L35 Althaus, L. G., Serenelli, A. M., Panei, J. A. et al. 2005, A&A, 435, 631 Ciardullo, R., & Bond, H. E. 1996, AJ, 111, 217 Córsico, A. H., Althaus, L. G., & Miller Bertolami, M. M. 2006, A&A, 458, 259 Córsico, A. H., Althaus, L. G., Kepler, S. O., Costa, J. E. S., & Miller Bertolami, M. M. 2008, A&A, 478, 869 Costa, J. E. S., Kepler, S. O. 2008, A&A, 489, 1225 Costa, J. E. S., Kepler, S. O., Winget, D. E. et al. 2008, A&A, 477, 627 Cox, J. P., 1980, Theory of Stellar Pulsations (Princeton: Princeton University Press) Fontaine, G. & Brassard, P. 2008, PASP, 120, 1043 Gautschy, A. 1997, A&A, 320, 811 Gautschy, A., & Saio, H. 1995, ARAA, 33, 75 Gautschy, A., Althaus, L. G., & Saio, H. 2005, A&A, 438, 1013 González Pérez, J. M., Solheim, J.-E., & Kamben, R. 2006, A&A, 454, 527 Grauer, A. D., Bond, H. E., Liebert, J., Fleming, T. A., & Green, R. F. 1987, ApJ, 323, 271 Hine, B. P., III, & Nather, R. E. 1987, IAU Colloq. 95: Second Conference on Faint Blue Stars, 619 Kawaler, S. D. 1988, ApJ, 334, 220 Kawaler, S. D., Winget, D. E., Hansen, C., Iben, I., Jr. 1986, ApJL, 306, L41 Liebert, J., Fleming, T. A., Green, R. F, & Grauer, A. D. 1988, PASP, 100, 187 Miller Bertolami, M. M., & Althaus, L. G. 2006, A&A, 454, 845 Quirion, P., Fontaine, G., & Brassard, P., 2007, ApJS, 171, 219 Quirion, P., Fontaine, G., & Brassard, P., 2004, ApJ, 610, 436 Saio, H. 1996, in Hydrogen-Deficient Stars, ed. C. S. Jeffery, & U. Heber, ASP Conf. Ser., 96, 361 Solheim, J.-E., Gonz[á]{}lez P[é]{}rez, J. M., & Vauclair, G. 2008, Hydrogen-Deficient Stars , 391, 195 Starrfield, S., Cox, A. N., Hodson, S. W., & Pesnell, W. D. 1983, ApJL, 268, L27 Starrfield, S., Cox, A. N., Kidman, R. B., & Pesnell, W. D. 1984, ApJ, 281, 800 Starrfield, S., Cox, A. N., Kidman, R. B., & Pesnell, W. D. 1985, ApJL, 293, L23 Unno, W., Osaki, Y., Ando, H., Saio, H., & Shibahashi, H. 1989, Nonradial Oscillations of Stars (University of Tokyo Press), 2nd edition Werner, K., & Herwig, F. 2006, PASP, 118, 183 Winget, D. E., & Kepler, S. O. 2008, ARAA, 46, 157
[^1]: We note that some of our sequences have trace surface abundances of H ($X_{\rm H} \lesssim 10^{-3}$) which give rise to some H burning. However, exhaustive test stability computations demonstrate that H burning is very weak and extends on a extremely narrow layer, as a result of which the H-shell burning is completely irrelevant in destabilizing modes and will not be further considered in this paper.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We present the results of new radio interferometer line observations for the merging galaxy pair NGC 4038/9 (‘The Antennae’), obtained using the Australia Telescope Compact Array. The results improve substantially on those of van der Hulst (1979) and show in detail the two merging galactic disks and the two tidal tails produced by their interaction. The small edge-on spiral dwarf galaxy ESO 572–G045 is also seen near the tip of the southern tail, but distinct from it. It shows no signs of tidal interaction. The northern tidal tail of the Antennae shows no connection to the disks and has an extension towards the west. The southern tidal tail is continuous, with a prominent concentration at its tip, roughly at the location of the tidal dwarf galaxy observed optically by Mirabel, Dottori & Lutz (1992). Clear velocity structure is seen along the tidal tails and in the galactic disks. Radio continuum images at 20 and 13 are also presented, showing the disks in detail.'
author:
- |
Scott Gordon$^1$[^1], Bärbel Koribalski$^2$, Keith Jones$^1$\
$^1$Department of Physics, University of Queensland, St. Lucia, Brisbane, QLD 4072, Australia\
$^2$Australia Telescope National Facility, CSIRO, P.O. Box 76, Epping, NSW 1710, Australia\
date: 'Received date; accepted date'
title: ' Observations of Interacting Galaxy Pair NGC 4038/9'
---
instrument: Australia Telescope Compact Array — galaxies: individual: NGC 4038, NGC 4039, ESO 572–G045 — galaxies: interacting
Introduction {#section:Introduction}
============
The Antennae (NGC 4038/4039 = Arp 244, see Fig. \[figure:AntennaeDSS\]) are a pair of closely-interacting disk galaxies in the advanced stages of merging, named after the appearance of the two spectacular galactic tails produced by gravitational tidal forces during the interaction. There are a number of similar systems known, including ‘The Mice’ (NGC 4676A/B) described by Hibbard (1995) and ‘The Superantennae’ (IRAS 19254–7245) described by Mirabel [et al.]{}(1991). Also visible to the south-west of the southern tail of the Antennae is the edge-on spiral dwarf galaxy ESO 572–G045 (Fig. \[figure:AntennaeDSS\]), which has no measured optical velocity. Some parameters of the Antennae and ESO 572–G045 are given in Table \[table:TheGroup\].
On a large scale, the Antennae appears roughly symmetrical, although the southern tail is more prominent both optically (Fig. \[figure:AntennaeDSS\]) and in (van der Hulst 1979; this paper). The two galaxies retain distinct nuclei as seen from optical and CO observations. However, the galactic disks, while still present, are colliding and have severely disrupted one another. The two nuclei are currently $\sim$113 or $\sim$6.3 kpc apart across the line-of-sight, according to CO measurements by Stanford [et al.]{} (1990). As a result of the collision, vigorous star formation is occurring in the disks and in the overlap region between them. This is evidenced by a variety of data, for example HST optical observations (Whitmore & Schweizer 1995; Whitmore [et al.]{} 1999), and ISO infrared observations ([e.g.]{} Vigroux [et al.]{} 1996) which show a high infrared luminosity in the two nuclei and particularly in the overlap region.
The tidal tails, disturbed structure, and star formation, are all classic features of interacting galaxies. Given this and the closeness of the system ($D = 19.2\,\MpcE$[^2]), the Antennae have attracted a lot of attention in observational and theoretical studies of galaxy interaction.
History of the Antennae {#section:IntroHistory}
-----------------------
The Antennae were first seen by C.O. Lampland of Lowell Observatory in 1917 (Parker 1990), and were later described by Shapley & Paraskevopoulos (1940), and Minkowski (1957), who identified the pair with a radio source and first described them as interacting galaxies. They were also noted as number 244 in the Arp atlas of peculiar galaxies (Arp 1966).
Systems resembling the Antennae were produced in the pioneering computer modelling simulations by Toomre & Toomre (1972), providing the first theoretical proof that features such as galactic tails were produced by gravitational tides between interacting galaxies, leading to much of the present understanding of their evolution. Clutton-Brock (1972) published work similar to the Toomres’, but less extensive. The Antennae are the first or ‘earliest’ of the 11 systems in the ‘Toomre Sequence’ (Toomre 1977), and are still used for modelling ([e.g.]{} Barnes 1988; Mihos, Bothun & Richstone 1993). Collectively, these simulations have disproved, for most interacting systems, earlier explanations for galactic pairs and interaction effects, such as the suggestion by Arp (1969) that when two galaxies of different sizes were interacting, the smaller one was an ejected component of the larger.
Previous Observations {#section:IntroPreviousObs}
---------------------
Observations of the Antennae have been made at many radio, infrared, optical, and X-ray wavelengths, including many emission lines such as (Amram [et al.]{} 1992), (Nikola [et al.]{} 1998), CO J=1–0 (Sanders & Mirabel 1985; Stanford [et al.]{} 1990; Young [et al.]{} 1995; Gao [et al.]{} 1998; Wilson [et al.]{} 2000; Gao [et al.]{} 2001), and (Huchtmeier & Bohnenstengel 1975; van der Hulst 1979), , , , , and (Rubin, Ford & D’Odorico 1970), and also radio continuum (Condon 1983; Condon 1987; Hummel & van der Hulst 1986).
More recently, radio observations of the Antennae have been made using the VLA. These include high-resolution 6 and 3.5 observations by Neff & Ulvestad (2000), and observations by Mahoney, Burke & van der Hulst (1987) and by Hibbard, van der Hulst & Barnes (2001, in prep).
Infrared continuum studies have been made using the ISO satellite by Fischer [et al.]{} (1996), Kunze [et al.]{} (1996), Vigroux [et al.]{}(1996), and Mirabel [et al.]{}(1998). X-ray studies have been made using the Einstein Observatory by Fabbiano & Trinchieri (1983), ROSAT by Read [et al.]{} (1995) and Fabbiano [et al.]{} (1997), ASCA by Sansom [et al.]{} (1996), and most recently using Chandra by Fabbiano, Zezas & Murray (2000).
The Inner Disks {#section:IntroInnerDisks}
---------------
There are numerous regions distributed in a ring in the NGC 4038 (northern) disk and a few also along an arc in the NGC 4039 (southern) disk. The shape of their distribution is roughly that of a letter ‘e’, and is seen clearly as knots, or compact objects, in blue optical images as well as in (Amram [et al.]{} 1992) and in the radio continuum (Hummel & van der Hulst 1986). The Hubble Space Telescope resolved the regions into many groups and clusters of young stars, with over 700 mostly blue compact objects 20–30 wide (Whitmore & Schweizer 1995; Whitmore [et al.]{} 1999).
The location of one particular group of optical and radio continuum knots in the overlap region between the two galactic disks, combined with the youth of the stars, suggests that this star formation activity is caused by the on-going interaction. Further high-resolution mid-infrared studies using the ISO satellite (Mirabel [et al.]{} 1998) show the strongest star formation in the overlap region, accounting for half of the $15\,\umE$ flux, rather than in the more optically spectacular star-forming ring. This is explained as a result of obscuration by the large quantity of dust, which tends to hide star-forming regions at optical wavelengths but also makes them highly visible in the infrared (Kunze [et al.]{} 1996; Mirabel [et al.]{} 1998; Wilson [et al.]{} 2000). Wilson [et al.]{} (2000) produced detailed CO maps and molecular mass estimates for the disk region using the Caltech mm-array. The maps show seven giant molecular cloud complexes, including the two nuclei and five complexes in the disk overlap region which have on-going massive star formation. Gao [et al.]{} (2001) made a similar study combining single-dish and interferometric observations, leading to an estimated molecular gas mass of $\baseexpE{1.38}{10}\,\MsolarE$ (adopting $D=19.2\,\MpcE$), implying a modest star formation efficiency.
Southern-Tail Tidal Dwarf Galaxy {#section:IntroTailDwarf}
--------------------------------
At the tip of the southern tail is an object thought to be a dwarf galaxy formed from material ejected from the galactic disks during the interaction (Mirabel, Dottori & Lutz 1992). It contains many hot, young stars, and coincides with a major concentration of which has been noted in earlier studies (van der Hulst 1979).
This dwarf is of particular interest because it matches the results of computer modelling, confirming that tidally-disturbed material can become concentrated enough to form bound objects (Barnes & Hernquist 1992, 1996). Additionally, observing a dwarf galaxy soon after its formation could shed light on the origin and star-forming properties of such galaxies, about which little is currently known.
------------------------------------------------------------------------
\
Section 2 describes the observations and data reduction. This is followed by a summary of the major features in Section 3. The implications of the data, and its relationship to previous work, is given in Section 4. Finally, Section 5 presents a summary of our new and interesting results.
Observations & Data Reduction {#section:Observations}
=============================
We have observed the Antennae group using the Australia Telescope Compact Array (ATCA[^3]), a linear east-west oriented radio interferometer with six 22-m parabolic dishes. Table \[table:ObsDetails\] gives details of the observations. In total, four successful 12 hour observations were obtained (plus one additional 2–3 hour run) using different antenna arrays chosen to give baselines spread across as wide a range as possible. The dual 20/13 feedhorns on the ATCA, producing two independent IFs, were used to observe in both bands simultaneously. Data reduction was performed, unless otherwise stated, using the MIRIAD analysis package (Sault, Teuben & Wright 1995). All radio images presented have been corrected for primary-beam attenuation. In some figures, pixels below $3\sigma$ have been masked out prior to correction to exclude noise. All velocities are in the heliocentric reference frame and the ‘optical’ definition $v=c\left(\Delta\lambda/\lambda_{\circ}\right)$. The measured velocities have an uncertainty of about 5. Positions are expressed in the J2000 system, including any from other papers.
Line and 20-cm Continuum Observations {#section:Obs20cm}
--------------------------------------
The 20 data (IF1; see Table \[table:ObsDetails\]) covered the red-shifted emission line of neutral atomic hydrogen, plus some continuum channels to each side. The data was Hanning-smoothed to a resolution of $\sim6.6\,\kmpsE$. Channels at each edge of the passband (attenuated by 50% or more) were removed, giving a remaining bandwidth of $\sim6.8\,\MHzE$. This band was used to produce separate line and 20-cm continuum datasets, using a first-order polynomial (or baseline) fit to the line-free channels to each side of the spectrum.
The 20 (1413) flux for PKS 1934–638 is $\sim$14.88 Jy. The derived antenna gains ranged from 0.99 to 1.06, indicating a typical $\sim$3–4% change during calibration.
The line data for the individual observations were then combined and Fourier-transformed to produce a spectral image cube. The visibilities were weighted using ‘robust’ weighting with the robustness parameter equal to 1, chosen to give a good compromise between sensitivity and resolution. This image cube was cleaned by deconvolution using a combined Högbom/Clark/Steer algorithm (Högbom 1974, Clark 1980, Steer [et al.]{} 1984). The maps have a channel width of $10\,\kmpsE$; the pixel size is 5 $\times$ 5. The r.m.s. residual after cleaning is $\sim1~\mJypbE$. The map was restored using a circular restoring beam of 42.
The task was used to produce maps of integrated intensity (moment-0), intensity-weighted mean velocity (moment-1) and velocity-dispersion (moment-2). To smooth the cube spatially and in velocity we used a three pixel ‘boxcar’ function and a five channel ‘Hanning’ function, respectively. A lower flux cutoff of $2.1~\mJypbE$ ($\sim3\sigma$ after smoothing) was applied.
The 20 continuum map (Fig. \[figure:Antennae20\_13CM\_Overlay\]a) was made using ‘uniform’ weighting resulting in a noise after cleaning of 0.23, and a $16{\arcsecE}$ restoring beam was used.
13-cm Continuum Observations {#section:Obs13cm}
----------------------------
The 13 data (IF2; see Table \[table:ObsDetails\]) had 128bandwidth and full polarisation information. After excluding edge channels, the bandwidth was 100. No polarisation was observed above $1\%$.
The 13 continuum map (Fig. \[figure:Antennae20\_13CM\_Overlay\]b) was made using ‘uniform’ weighting, resulting in a noise after cleaning of 0.056 using a $9.5{\arcsecE}$ restoring beam.
The flux of PKS 1934–638 ranges from $\sim$11.5 Jy 2378 to $\sim$11.1 Jy at 2496. The derived antenna gains ranged from 0.99 to 1.05, giving a typical change during calibration of $\sim3\%$.
Results {#section:Results}
=======
Channel Map Summary {#section:ResultsHIChannelMaps}
--------------------
Neutral hydrogen in the Antennae group is observed in the velocity range $\sim1340-1946\,\kmpsE$ (see Fig. \[figure:HIChannelMaps\]). At low velocities the emission consists of a single peak in the disk overlap region, generally brightening with increasing velocity. Above 1560a second peak to the south is seen, moving southward and forming the inner half of the southern tail. A third peak beginning at $\sim1610\,\kmpsE$ forms the outer half of that tail, the two meeting near the tail’s center and disappearing at $\sim1780\,\kmpsE$. The northern tail is also detected ($\sim1540-1620~\kmpsE$) and possesses a previously unknown extension towards the west at a similar velocity.
The dwarf galaxy ESO572–G045 about 30 south-west of the tip of the southern tail is detected in the velocity range $\sim1650-1750\,\kmpsE$.
Statistics {#section:ResultsHIStatistics}
-----------
The disk region and the southern tail have similar masses of $\sim\baseexpE{2.3}{9}\,\MsolarE$ and $\sim\baseexpE{2.5}{9}\,\MsolarE$, respectively (see Table \[table:HIstats\]). The exact division between the two regions is arbitrary since there is no clear boundary between them, although the southern tail only has velocities above $\sim1590\,\kmpsE$. The northern tail has an mass of $\sim\baseexpE{0.2}{9}\,\MsolarE$.
We calculate an mean velocity for the Antennae, of $\sim1664\,\kmpsE$, obtained from the moment maps (Fig. \[figure:HIMomentMaps\] a,b) (the mean for the disks is $\sim1620\,\kmpsE$). This is similar to the $\sim1660\,\kmpsE$ systemic velocity obtained from optical measurements of the nuclear velocities (see Table \[table:TheGroup\]).
The column density in Table \[table:HIstats\] is obtained from the moment-0 map using the following formula: $$\label{eqn:HI_ColumnDensity}
N_{\HI} = \baseexpE{1.11}{24}\,\psqcmE \frac{\int{B(v)dv}}
{\JypbkmpsE} \frac{{\rm arcsec}^2} {\theta_1\theta_2}~~~,$$ where $B(v)$ is surface brightness, $v$ is velocity, and $\theta_1$ and $\theta_2$ are the major and minor beam FWHM. For our maps with a resolution of 42 there is $\baseexpE{6.23}{20}\,\psqcmE$ per of moment-0.
The mass is calculated from the integrated flux, $F(v)$, using the following equation (Giovanelli & Haynes 1988; Roberts 1975):
$$\label{eqn:HI_Mass}
M_{HI} = \baseexpE{2.356}{5}\,\MsolarE {\left(\frac{D}{\MpcE}\right)}^2
\frac{\int{F(v)dv}}{\JykmpsE}~~~.$$
For $D=19.2\,\MpcE$ this gives per of flux.
The Galactic Disks {#section:ResultsGalacticDisks}
------------------
The overall shape of the in the disk region and at the base of the southern tail (Fig. \[figure:HIMomentMaps\]b) follows the shape of the optical emission, except that: [**(1)**]{} the spans the gap between the western ends of the two disks and ends well inside the optical image, particularly in the northern disk, whereas galaxies usually have more extended ; [**(2)**]{} the northern (and smaller) tidal tail is not connected to the disks in , with a $\sim1{\arcminE}$ gap in the distribution; and [**(3)**]{} the intensity peak is midway between the optical nuclei.
The peak in the distribution lies about one third of the distance between the two nuclei, about $32{\arcsecE}$ (3) north of the center of NGC 4039 (Table \[table:HIstats\]).
spectra taken throughout the disks generally show a single peak in the spectrum. However, there are two velocity peaks in a wide region midway between the centres of the two disks (Fig. \[figure:HINuclearSpectra\]), at $\sim1542$ and $\sim1630~\kmpsE$. These peaks are also visible in the overall disk spectrum (Fig. \[figure:HISpectra\]a), but they don’t match the systemic velocities of the disks, which are $\sim1663$ and $\sim1655\,\kmpsE$ (Table \[table:TheGroup\]).
The velocities in the disks (Fig. \[figure:HIMomentMaps\]c,d) range from 1470 at the northern disk edge, to 1920 close to the south-western disk edge.
The disks have a substantial velocity dispersion (see Fig. 2f), with a peak of $\sim150\,\kmpsE$ about $30{\arcsecE}$ south of the NGC 4039 nucleus (Table \[table:TheGroup\]), far from the intensity peak. The high dispersion results from the dual-peaked spectrum in its vicinity.
The dominant feature in the south-east is the base of the southern tail, which attaches smoothly to the disks. Although a clear division is seen in the diagram (Fig. \[figure:HIPV\]a) the two are not clearly divided in moment maps.
The Southern Tail {#section:ResultsSouthernTail}
-----------------
The southern tidal tail describes a gradual but pronounced arc about 90 kpc long which dominates the maps (Fig. \[figure:HIMomentMaps\]). It contains $\sim14$ times more than the northern tail, even though the two appear similar optically. It appears continuous, although some clumping is visible in Fig. \[figure:HIMomentMaps\]a.
The optical width of the tail is quite narrow compared to . Gaussian fits to profiles across the tail give ${\rm FWHM}\approx100{\arcsecE}$.
The flux peaks at $\sim1.59\,\JypbkmpsE$ ($\baseexpE{9.9}{20}\,\psqcmE$) close to the tip of the tail. This is quite a pronounced concentration containing an mass of $\sim\baseexpE{8.2}{8}\,\MsolarE$ or $\sim30\%$ of the total for the tail and was known as a distinct region in the maps by van der Hulst (1979).
The southern tail covers a continuous velocity range of $\sim1590-1784~\kmpsE$, with the velocity changing smoothly along the tail, peaking at $\sim1773\,\kmpsE$. Towards the apparent concentration at the tip of the tail, the velocity changes rapidly towards a minimum of $\sim1634\,\kmpsE$. The iso-velocity contours change orientation suggesting that the tail bends sharply toward the end, in agreement with a sharp bend in the optical image. Hence this concentration might be just the result of projection in a tail with varying orientation.
The Northern Tail and Extension {#section:ResultsNorthernTail}
-------------------------------
Compared to the southern tail, the northern tail contains less ($\sim\baseexpE{1.6}{8}\,\MsolarE$), is shorter and covers a smaller radial velocity range (1550–1620). Its velocity decreases steadily with distance from the disks.
There is a notable gap between the northern tail and the disks (see Fig. \[figure:HIMomentMaps\]b and Fig. \[figure:HIChannelMaps\]). However, the diagrams (Fig. \[figure:HIPV\]) suggest that a faint connection may exist. This gap contrasts with this tail’s optical connection to the disks and the continuous in the southern tail.
Fig. \[figure:HIMomentMaps\]a shows emission slightly west of the tip of the northern tail, not corresponding to any optical feature and not noted in any previously published maps. It has an mass of $\sim\baseexpE{1.8}{7}\,\MsolarE$ and a velocity range of 1550–1580. Its closeness to the tip of the northern tail suggests that this is a westward extension of a curved northern tail.
The dwarf companion ESO 572–G045 {#section:ResultsESO572G045}
--------------------------------
Our maps (see Fig. \[figure:HIMomentMaps\]a,b) show a clear detection of the small edge-on spiral galaxy ESO 572–G045 (Karachentsev [et al.]{} 1993; see also Table \[table:TheGroup\]). Optically it has a size of $\sim64{\arcsecE}\times7{\arcsecE}$. It is $\sim$165 (91 kpc) from the centre of the Antennae, and $\sim$5–6south-west of the end of the southern tail.
In the galaxy is slightly extended north-south, but unresolved east-west, matching quite well the optical appearance. It has a clear and smooth velocity gradient along the north-south major axis, ranging from 1650–1750 (north to south), like an edge-on view of a rotating disk. Its velocity also matches the adjacent portions of the Antennae southern tail. There is no visible bridge of material joining this galaxy to the tidal tail, and also no visible distortion of the tail in the direction of this galaxy.
The mass of ESO 572–G045 is $\sim\baseexpE{1.4}{8}\,\MsolarE$, or $3~\%$ of that in the Antennae. Its systemic velocity is $\sim1700\,\kmpsE$.
Radio Continuum Emission {#section:ResultsRadioContinuum}
------------------------
The ATCA 20 continuum map (Fig. \[figure:Antennae20\_13CM\_Overlay\]a) corresponds to those published by Condon (1983; 1987), and by Hummel & van der Hulst (1986), all using VLA data. However, the 13 map obtained with the ATCA (Fig. \[figure:Antennae20\_13CM\_Overlay\]b) is the first ever published.
The continuum sources seen in and around the Antennae are summarised in Table \[table:ContinuumSources\]. The images show no emission outside of the vicinity of the galactic disks, except for a number of point sources. No continuum emission from the dwarf galaxies was detected.
Fig. \[figure:Antennae20\_13CM\_Overlay\] shows continuum concentrated in the northern starburst ring and in the disk overlap region, with some additional extended emission. The emission is well correlated with the location of star formation activity as seen through optical, and infrared mapping (Whitmore [et al.]{} 1999; Amram [et al.]{} 1992; Mirabel [et al.]{} 1998, respectively). The continuum appears generally more extended than these regions.
The major peak in the radio continuum at both 20 and 13 occurs in the disk overlap, with secondary peaks near the southern nucleus and in the western portion of the starburst ring, as for the mid-infrared (Mirabel [et al.]{} 1998). Sources A, B, and C correspond to the three molecular gas concentrations found in CO by Stanford [et al.]{} (1990).
Source A, in the overlap region, is particularly bright. It corresponds to ‘Region C’ of Whitmore & Schweizer (1995), and to the ‘East Clump’ of Stanford [et al.]{} (1990). Higher-resolution 13 maps (not shown), with a fitted Gaussian restoring beam rather than the larger circular one of Fig. \[figure:Antennae20\_13CM\_Overlay\], show this region resolved into three or four smaller ones, apparently corresponding to the southern-most group of emission regions seen in by Amram [et al.]{} (1992).
Source B is at the catalogued central or nuclear position of NGC 4038 (Table \[table:TheGroup\]), while Source C is $\sim6{\arcsecE}$ west of the nucleus of NGC 4039.
The 20 map has slightly lower resolution than the 13 map (see Fig. \[figure:Antennae20\_13CM\_Overlay\]) and shows more extended emission, seen largely as a number of faint ‘protrusions’ from the Antennae disks, some of which are noise. Because of the higher resolution, the 13 map resolves more emission peaks, particularly in the southern disk, and shows two depressions which are not obvious at 20.
Fig. \[figure:Antennae20\_13CM\_Overlay\] also shows some continuum emission coming from the base of the southern tail, which is rather bright optically and in (Fig. \[figure:HIMomentMaps\]b). There is no trace of radio continuum emission elsewhere in either tail.
We measure the total continuum emission from the disks as $\sim$498 mJy at $1413\,\MHzE$ (20), and $\sim$311 mJy at $2378\,\MHzE$ (13). This includes the emission at the base of the southern tail. The 20 figure compares with the $472\pm23\,\mJyE$ at $1465\,\MHzE$ by Condon (1983), $572\,\mJyE$ at $1490\,\MHzE$ by Condon (1987), and $486\pm20\,\mJyE$ at $1465\,\MHzE$ by Hummel & van der Hulst (1986).
Discussion {#section:Discussion}
==========
Mass Distribution {#section:DiscussionHIMass}
------------------
The total mass of the Antennae, as measured with the ATCA, is $\sim\baseexpE{5}{9}\,\MsolarE$. About 46% of the mass is in the disks, 51% in the southern tail and only 3% in the northern tail (see Table \[table:HIstats\]). Van der Hulst (1979) estimated the corresponding components as 40%, 52% and 8%. The column densities in the disks are much higher than those in the tails. Our measurements agree with the general trends found previously ([e.g.]{} Hibbard & van Gorkom 1996), whereby gas is depleted in galactic disks as their merger progresses, leaving most of it in the tidal tails. For the Antennae, at an early stage of merging, the tails already contain about half of the .
The southern tail is about twice as long optically as the northern tail, but otherwise they are optically similar. In this is quite different, with the southern tail containing $\sim15$ times as much as the northern tail. Additionally, the northern tail is not connected to the disks in .
Wilson [et al.]{} (2000) mapped the molecular gas in the disks and find that about half of the detected gas resides in five super-giant molecular cloud complexes in the disk overlap. The remainder is in the galactic nuclei and the western portion of the northern star-forming ring. However, the authors acknowledge, based on single-dish observations by Gao [et al.]{} (1998), that they probably missed 60% of the gas. More recently, the Gao [et al.]{} (2001) give a molecular mass estimate of $\sim\baseexpE{1.38}{10}\,\MsolarE$ (adopting $D=19.2\,\MpcE$). Combined with our data, this implies a molecular-to- mass ratio of $\sim6:1$ for the disks alone and $\sim3:1$ for the whole Antennae.
Gao [et al.]{} (2001) made the first, tentative detection of CO at the tip of the southern tail, with a molecular mass of $\sim\baseexpE{2}{8}\,\MsolarE$, and hence a molecular-to- ratio of $\sim1:4$.
Van der Hulst (1979) tried to overcome the large angular size of the Antennae, and Westerbork’s low sensitivity to east-west structure, by estimating masses from multiple single-disk measurements by Huchtmeier & Bohnenstengel (1975). The derived total mass is $\baseexpE{3.7}{9}\,\MsolarE$, much more than the $\baseexpE{2.5}{9}\,\MsolarE$ which he measured by synthesis imaging, but less than our measured $\sim\baseexpE{5.1}{9}\,\MsolarE$ (including ESO 572–G045), and the estimate of $\sim\baseexpE{5.3}{9}\,\MsolarE$ by Huchtmeier & Bohnenstengel. The component masses measured by van der Hulst were $\baseexpE{1.0}{9}\,\MsolarE$ for the disks, $\baseexpE{1.3}{9}\,\MsolarE$ for the southern tail, and $\baseexpE{1.8}{8}\,\MsolarE$ for the northern tail (adopting $D=19.2\,\MpcE$).
The Galactic Disks {#section:DiscussionDisks}
------------------
### Disk {#section:DiscussionDiskHI}
Both the distribution and velocity pattern in the central area of the Antennae (Fig. \[figure:HIMomentMaps\] b,d) do not resemble the regular, symmetric pattern seen in an isolated disk or spiral galaxy, and there is also no clear way to model it as the superposition of two disks.
Our data suggests instead a) that disk has been extensively kinematically disturbed by disk collision or the preceding close encounter, b) that it is now largely detached from the individual galaxies, with most of it no longer following closely their original disk structure, and c) that it resides largely in a single structure between and surrounding the disks. Further, this structure is connected smoothly to the -rich southern tidal arm.
Mihos & Hernquist (1994) ran computer simulations of a variety of merging disk galaxy pairs, including both those with and without large bulges. In the simulations, a clear pattern was seen. Galaxies with no bulge experienced nuclear gas inflows early in the merger, producing a steady moderate starburst and depleting the gas prior to the final stages of the merger. However, the presence of a substantial bulge inhibited this inflow, with a low initial star-formation rate and minor gas depletion. The result for these systems was coalescence of the two galaxies’ still abundant gas late in the merger, resulting in a massive burst of star formation and an ultraluminous infrared merger.
This study sheds light on the possible origin of the observed pool in the Antennae, and their future evolution. The star formation consequences of this are described further in Section \[section:DiscussionStarFormation\].
Although there is an apparent pooling of the , the dual-peaked spectra, seen particularly over the central southern disk, indicate that in this region at least there are two clouds overlapping along the line-of-sight, with some of the still attached to the southern disk.
### Disk Radio Continuum {#section:DiscussionDiskContinuum}
Comparison of our ATCA 20 and 13 cm radio continuum data (Fig. \[figure:Antennae20\_13CM\_Overlay\]) with VLA maps at 20 and 6 cm (Hummel & van der Hulst 1986) does not reveal many new features.
The 13 cm continuum map (beam = 95) resembles a smoothed version of the high-resolution 6 and 20 VLA maps (beam = 6). The latter resolve greater detail in the disks, [e.g.]{} more emission peaks in the eastern half. The faint emission at the base of the southern tail is not seen in these maps, but is clearly detected in the low-resolution VLA 20-cm map.
The low-resolution 20 map by Hummel & van der Hulst (1986) has the same sensitivity and resolution as our Fig. \[figure:Antennae20\_13CM\_Overlay\]a, allowing a direct comparison. The measured fluxes agree well within the errors. Our image includes slightly more extended emission at faint levels, but most of it is likely to be noise.
The well-resolved emission peaks in our radio continuum maps, particularly in the 13 image (Fig. \[figure:Antennae20\_13CM\_Overlay\]b), agree with those in the map by Amram [et al.]{} (1992). Our resolution is too low to allow a detailed comparison of the radio continuum and peaks.
With radio continuum closely matching optical regions and maps, the Antennae agree with the correlation between star formation and the , FIR, and radio, noted for example by Sanders & Mirabel (1985).
The Tidal Tails {#section:DiscussionTidalTails}
---------------
### Modelling The Tails {#section:DiscussionTailModelling}
Toomre & Toomre (1972) performed pioneering computer simulations which clearly demonstrated for the first time that galactic tails originate through tidal interaction. However, these simulations entirely neglected the self-gravity of the disk material as well as dispersion effects, which have been considered in more recent simulations. The omission of simulated velocity fields in many theoretical papers on galaxy interactions often renders impossible a direct kinematic comparison between observations and model predictions.
Fortunately, van der Hulst (1979) obtained and published graphically the velocity information not published by Toomre & Toomre (1972), and used it to compare observed diagrams to predicted ones. After rescaling of dimensions to fit the data, the match was quite close.
Computer simulations ([e.g.]{} Appendix VIII of Hibbard 1995; Hibbard & Mihos 1995) predict that material in a tail will be ejected along many different trajectories, with a large-scale filament-like structure due to the similarity of trajectories of neighbouring material, rather than being a stream of material along a common trajectory. In this way, it resembles a density wave, with the wave initially moving outward from the galactic disks, and in most cases later falling back inward under gravity. Consequently, the material velocities tend to be at a large angle to the orientation of the tail, and disk galaxy models do not apply to tidal tails.
More recently, Barnes (1988) carried out tree-code N-body simulations of interacting disk galaxies, incorporating full self-gravity and a bulge/disk/halo model. The first model made was a look-alike of the Antennae, for which a number of figures were published, including a diagram along the orbital plane, approximately north-south (see his Fig. 3d). This is roughly a modelled equivalent of our Fig. \[figure:HIPV\]b, but showing all material originating from the disk rather than just . These two figures bear a very close resemblance, except that the observed southern tail is much more prominent than the northern one, whereas the model diagram shows them fairly comparable in size.
Hibbard & Mihos (1995) performed similar simulations based on earlier VLA data for the merging pair NGC 7252, which closely resembles the Antennae. The observed diagram (their Fig. 1) closely resembles our observed ones for the Antennae (Fig. \[figure:HIPV\]). That work particularly mentioned the final destination of material in tidal tails (their Figs. 5, 7). The bulk of tail material has an elliptical trajectory taking it back towards the disks, much of it quite rapidly. Some, however, may gain escape velocity, with material further from the disks also having higher velocities.
By analogy, it can be inferred that the outer-most material in the southern tail of the Antennae (including the concentration at the tip), may take quite some time to fall back, if it does at all. However, the turn-around in radial velocities in the southern tail indicates that it is substantially bent along the line-of-sight ‘towards’ the disks, suggesting that the tip is falling back. This interpretation, though, is unlikely to be correct since computer simulations show that the energy of tail material increases with distance from the center, and thus outer portions of the tail will take longer to fall back (J. Hibbard, private communication).
### Optical versus Emission in the Tidal Tails {#section:DiscussionTailHIOpticalDifferences}
There is no obvious spatial displacement between the optical and emission in the tails (Fig. \[figure:HIMomentMaps\]) as has been observed in other interacting systems, [e.g.]{} Arp 157, Arp 220, and Arp 299 (see Hibbard, Vacca and Yun 2000). The primary differences are: [**(1)**]{} the widths of the tails, [**(2)**]{} the highly visible pool at the end of the southern tail, [**(3)**]{} the greater length of the southern tail in , and [**(4)**]{} the absence of at the base of the northern tail. The latter might be due to ionisation by emission from the disks as suggested by Hibbard, Vacca and Yun (2000).
The in the tails is clearly more extended laterally than the stellar emission. Short-exposure optical images ([e.g.]{} Fig. 1) show both tidal tails to be fairly narrow with a width of 30–40 (2.8–3.7 kpc), much less than the width of $\sim$100 (9.3 kpc). Deeper optical images of the Antennae (Schweizer 1978), however, show both tails more being extended.
The strong stellar emission from the Antennae tails is somewhat unusual since most tidal tails are optically faint. This general trend may come about because material which is (pre-interaction) further from the galactic centre, is less tightly bound gravitationally, and more affected by intergalactic tidal forces, and thus generally contributes most of the material in tidal tails ([e.g.]{} Toomre & Toomre 1972). This and the fact that in most galaxies the gas is much more extended than the stellar component, leads to tails initially containing a lot of gas and relatively few stars. Furthermore, the rapid decrease in the gas density of tidal arms, due to differential expansion after their creation, should result in a corresponding decrease in star-formation rates, and hence a drop in optical emission (Mihos, Richstone & Bothun 1991). The fact that the Antennae do not show low optical brightness, suggests that the material in the tails derives from a more inner part of the disks than what is often seen in other galaxies.
### No Continuum Emission in the Tidal Tails {#section:DiscussionTailContinuum}
Although the disks show continuum emission, we have found none from either of the tidal tails, except at the base of the southern tail. Since radio continuum emission is generally produced extensively in star-forming regions, [e.g.]{} by hot gas or shocks in supernova remnants, its absence in the tails implies a low level of star formation.
The absence of detectable radio continuum in tidal tails is also quite common for other observed interacting systems, and agrees with predictions by Mihos, Richstone & Bothun (1991) and others, who have undertaken computer simulations of interaction-induced starbursts, modelling galactic close encounters and the resulting collisions of giant molecular clouds and inwards gas flows, believed to be the means by which starbursts are produced in interacting galaxies. Mihos, Richstone & Bothun (1991) studied star-formation rates for pairs of disk galaxies undergoing encounters (but not mergers) at a number of separations and relative orientations. They found that in some cases close encounters will actually reduce star formation, by throwing clouds out of the disk and reducing their density, without producing many collisions. However, prograde encounters (both disks rotating in the same sense as their orbit about one another, [e.g.]{} the Antennae), tend to produce quite strong starbursts and extensive transfer of material between the disks. Such encounters could also produce starbursts along the leading edges of tidal tails. However, such tails expand rapidly and the star formation rate drops considerably after a couple $\times10^8{\rm~years}$ (Mihos [et al.]{} 1991), so that tails will generally have no detectable star formation activity when observed well into a close encounter or merger.
Southern-Tail Tidal Dwarf Galaxy {#section:DiscussionTailDwarf}
--------------------------------
The enhanced column density at the tip of the southern tail coincides with the location of a young ($\sim\baseexpE{3}{8}$ years) star-forming region, suspected to be a dwarf galaxy formed during the interaction (Schweizer 1978; Mirabel, Dottori & Lutz 1992). The enhancement, by a factor of $\sim4.5:1$ compared to the tail’s minimum can at least in part be explained as a projection effect due to the bending of the tail and hence a deeper line of sight. However, such a large ratio requires a substantial bending angle, hence there is likely to be an enhanced local gas density at the location of the star-forming region: a point already made by van der Hulst (1979).
Sizeable concentrations in tidal arms, many of which appear self-gravitating, have been found in studies of a number of interacting galaxies, with typical masses of $\sim\baseexpE{5}{8}-\baseexpE{6}{9}\,\MsolarE$, the resulting overall population of bound dwarf galaxies being perhaps half of all dwarfs (Sanders & Mirabel 1996).
The formation of dwarf galaxies is predicted by computer simulations of galaxy interaction, such as those by Barnes & Hernquist (1992), and the more complete model, including gas dynamics, by Barnes & Hernquist (1996). In models, self-gravitation of some material ejected in tidal arms, may be enough to overcome expansion of the arm over time, forming bound objects. These tend to be gas-rich since their gas and stars have slightly different origins, with the gas being more compact and so surviving tidal disruption. Frequently a number of bound objects are produced, generally in the outer parts of the tail, due to the lower mass needed to avoid disruption there.
Elmegreen [et al.]{} (1993) used computer models of IC 2163/NGC 2207 to show that extended disks of gas can lead to a massive concentration of gas at the end of a tidal arm during interaction. They also showed that if the mass of the second galaxy is similar to, or larger than, the host galaxy, then eventually massive clouds can be ejected from the host galaxy, leading to self-gravitating dwarf galaxies like those observed by the same authors using VLA observations. According to Elmegreen [et al.]{}, the lack of an extended gas distribution in the models of Barnes & Hernquist (1992), prevented the gas pool forming in their simulations. With the Antennae galaxies merging, and of apparently similar masses, they should be sufficient to produce a pool along a tail, or attached or detached dwarf galaxies.
The dwarf companion ESO 572–G045 {#section:DiscussionESO572G045}
--------------------------------
The galaxy ESO 572–G045 is much smaller and less massive than the Antennae and is classified as a spiral dwarf galaxy. It has an radial velocity $\sim1688\,\kmpsE$, similar to that of the southern tidal tail, and a small apparent distance from it.
The association between ESO 572–G045 and the Antennae could be found by searching it for signs of recent interaction ([e.g.]{} starburst activity). If this has not occurred then almost certainly the dwarf is not involved in the Antennae merger. If interaction has occurred, then either the dwarf was a pre-existing galaxy now tidally disturbed, or it is a bound object produced from tidal debris from the merger ([e.g.]{} Elmegreen [et al.]{} 1993; Hernquist 1992, 1996). Given its apparent spiral structure, creation from debris appears a less likely explanation. With an mass alone of $\sim\baseexpE{1.4}{8}\,\MsolarE$, its mass would be at the high end for bound objects (Barnes & Hernquist 1992), but still reasonable. Unfortunately, to date there is no available information which would resolve this question.
The small size of the object relative to the beam allows only an approximate dynamical mass estimate from velocities. It is edge-on, hence its maximum rotational velocity is half of the range ($\sim50\,\kmpsE$). The corresponding radius appears to be $\sim32{\arcsecE}$ (3). The following equation (Giovanelli & Haynes 1988) estimates the total enclosed mass, assuming a spherical distribution:
$$\label{equation:MvsRV_spherical}
M_T(r) = \frac{rV^2(r)} {\kpcE~(\kmpsE)^2} \times \baseexpE{2.33}{5}\,\MsolarE~~~~,$$
from which we derive a dynamical mass of $\sim\baseexpE{1.74}{9}\,\MsolarE$. These estimates imply that makes up about 8% of the dwarf’s mass, roughly in the range observed in larger spiral galaxies, and supporting the suggestion of spiral structure.
The inner-most distance from the center of a globular cluster or orbiting satellite, at which tidal stripping occurs, is the tidal radius ($r_t$), which we take as the Jacobi limit:
$$\label{equation:TidalRadius}
r_t = R {\left(\frac{M}{3M_g}\right)}^{1/3}~~~,$$
where $M$ is the total mass of the satellite, $M_g$ that of the parent galaxy and $R$ their separation (see [e.g.]{} Pisano & Wilcots 2000). Taking the velocity minima and maxima as the extreme velocities in the Antennae disks, and the peak as the center of mass, leads to projected rotation velocity of $\sim228\,\kmpsE$ at a radius of $\sim12.5\,\kpcE$, and a very rough Antennae disk mass estimate of $\sim\baseexpE{1.5}{11}\,\MsolarE$. Treating ESO 572–G045 as a satellite of the Antennae disks ($R \ge91$ kpc) gives a tidal radius of $r_t\ge14$ kpc, much larger than its measured radius. So it is unlikely that ESO 572–G045 is prone to disruption.
Star Formation Patterns {#section:DiscussionStarFormation}
-----------------------
The Antennae starburst is optically spectacular and has a total infrared luminosity of $\sim10^{11}~\LsolarE$ indicating major starburst activity. However, Gao [et al.]{} (2001) concluded that the starburst is modest considering the high molecular gas content of the system ($\baseexpE{1.38}{10}\,\MsolarE$).
Computer models of the Antennae have shown that it may result from the prograde encounter of two disk galaxies. This has been found ([e.g.]{} for the non-merging case, by Mihos [et al.]{} 1991) to generally produce the strongest starburst. In the same study, closer encounters also strengthen the starburst at least for non-mergers. The Antennae encounter and the starburst within it thus agrees with the available models.
Mihos & Hernquist (1994) modelled star formation patterns in merging pairs of disk galaxies, concluding that galaxies with large bulges would delay their major starburst activity until the final stages of merger. At this point a combined central gas pool would produce vigorous starbursts (and a high infrared luminosity). From our studies it appears that such a pool of cool gas is forming in the Antennae.
These observations suggest that the Antennae are an example of the delayed action for bulge galaxies modelled by Mihos & Hernquist, with the current starburst being an early stage, and that the scene is being set for a fairly spectacular final merger of the disks at a later date.
Conclusion {#section:Conclusion}
==========
Our images of the Antennae have higher sensitivity and resolution than those by van der Hulst (1979), revealing more mass and structure as well as some new features not previously observed.
Our radio continuum images include the first published 13 image, and a new 20 image corresponding to previously published ones by Condon (1983, 1987) and Hummel & van der Hulst (1986). No continuum polarisation has been detected. The radio continuum maps closely resemble the maps by Amram [et al.]{} (1992). The distribution approximates the optical star-forming regions, [e.g.]{} seen in HST images. Sources in the overlap region match the locations of giant molecular clouds seen in CO by Stanford [et al.]{} (1990). These similarities suggest that the continuum results largely from star formation. The lack of continuum emission in the tidal tails indicates that recent and on-going star formation in the Antennae is basically confined to the disks.
The presence of a central gas pool along with a modest disk starburst (Gao [et al.]{} 2001) and computer simulations (Mihos & Hernquist 1994) suggest that the current starburst may be a weak precursor to a far larger one occuring much later in the merger.
The southern tidal tail contains abundant mass (), similar to the galactic disks (), but less concentrated. This tail is connected smoothly to the disks, although diagrams show a clear breakaway point at 1590. Definite clumps exist along the tail’s length. It has a substantial width of up to $\sim100\arcsecE$ ($\sim9\,\kpcE$), larger than the optical width. There is no spatial displacement between optical and emission. The tip of the southern tail contains a large concentration containing $\sim\baseexpE{8.2}{8}\,\MsolarE$ or $\sim30\%$ of the tail’s . At this point the tail also bends sharply both in terms of velocity and shape, and the column density is higher by a factor of $\sim4$. This coincides in position with the young tidal dwarf galaxy, observed optically by Mirabel [et al.]{} (1992). Dwarf galaxies have previously been produced as a tidal feature by computer simulations, along with pools of material at the ends of tidal arms (Barnes & Hernquist 1992, 1996; Elmegreen 1993), and they have also been observed in other interacting galaxies ([e.g.]{} Duc & Mirabel 1994, 1998). Although it is not known that any of the pool in the Antennae is associated with the dwarf, it appears likely they are associated, as seen in the simulations.
The smaller northern tail contains only of and has no detectable connection to the disks, which may be partly due to ionisation effects suggested by Hibbard, Vacca & Yun (2000). This tail’s extends much further lengthwise than its starlight, unlike the southern tail, and it curves toward the west along a faint extension newly detected in our data and not visible optically.
The central , while it follows broadly the outline of the optical disks, has a detailed velocity structure and spatial distribution which bear no close resemblance to those normally observed in disks. The appears to have been so kinematically disturbed that it is largely decoupled from the stellar component. The overall velocity range is a high $\sim640\,\kmpsE$ and the is substantially pooled in the region between the disks, peaking midway between them, unlike optical images. While generally there is only one velocity component along the line-of-sight, part of the disk , close to the southern nucleus, contains two components and thus a high dispersion, up to 150.
Comparison of mass measurements to previous CO measurements by Wilson [et al.]{} (2000) and Gao [et al.]{} (2001) gives an estimated molecular-to-atomic gas ratio of $\sim3:1$ for the whole system. The difference between the physical sizes and contents of the tails, suggests a difference in structure or gas content between the two progenitor galaxies, although asymmetry in the encounter geometry may be responsible.
Our images also show the nearby dwarf galaxy ESO 572–G045, which is south-west of the end of the southern tail, and $\sim$91 kpc from the Antennae centre. The matches the optical appearance of the galaxy, an edge-on disk $\sim$1 in width. Using the velocity field we estimate a systemic velocity of $\sim1700\,\kmpsE$ and a rotational velocity of $\sim50\,\kmpsE$, resulting in a dynamic mass of $\sim\baseexpE{1.74}{9}\,\MsolarE$, compared to an mass of $\sim\baseexpE{1.4}{8}\,\MsolarE$. This gives an 8% content which appears reasonable. The mass of ESO 572–G045 is 3% that of the Antennae. The dwarf’s closeness, and quite similar velocity, to the adjoining southern tail of the Antennae, indicates that it is a companion galaxy. It appears likely to have been involved in the interaction between the Antennae galaxies, but no major disturbances are observed. Its tidal radius is $\ge14\,\kpcE$, so it is unlikely to be prone to disruption.
Acknowledgements {#section:Acknowledgements}
================
- We would like to thank both the anonymous referee, and John Hibbard, for their many valuable suggestions, which have been incorporated into this paper.
- Some figures presented use data from the National Aeronautics and Space Administration (NASA)’s [*SkyView*]{} facility () located at NASA Goddard Space Flight Center. In particular, use has been made of Digitised Sky Survey (DSS) images based on photographic data obtained using the UK Schmidt Telescope.
- Various information used in preparing this paper, was taken from the NASA/IPAC Extragalactic Database (NED), operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with NASA.
Amram P., Marcelin M., Boulesteix J., le Coarer E., 1992, A&A, 266, 106
Arp H.C., 1966, ‘Atlas of Peculiar Galaxies’, California Institute of Technology (equivalent to Arp H.C., 1966, ApJS, 14, 1)
Arp H.C., 1969, A&A, 3, 418
Barnes J.E., 1988, ApJ, 331, 669
Barnes J.E., Hernquist L., 1992, Nature, 360, 715
Barnes J.E., Hernquist L., 1996, ApJ, 471, 115
Clark B.G., 1980, A&A, 89, 377
Clutton-Brock M., 1972, Ap&SS, 17, 292
Condon J.J., 1983, ApJS, 53, 459
Condon J.J., 1987, ApJS, 65, 485
de Vaucouleurs G., de Vaucouleurs A., Corwin H.G. Jr. , Buta R.J., Paturel G., Fouqué P., 1991, ‘Third Reference Catalogue of Bright Galaxies’, New York: Springer Verlag \[RC3\]
Douglas J.N., Bash F.N., Bozyan F.A., Torrence G.W., Wolfe C., 1996, AJ, 111, 1945
Duc P.-A., Mirabel I.F., 1994, A&A, 289, 83
Duc P.-A., Mirabel I.F., 1998, A&A, 333, 813
Elmegreen D.M., Kaufman M., Brinks E., Elmegreen B.G., Sundin M., 1995, ApJ, 453, 100
Elmegreen B.G., Kaufman M., Thomasson M., 1993, ApJ, 412, 90
Elmegreen B.G., Sundin M., Kaufman M., Brinks E., Elmegreen B.G., 1995, ApJ, 453, 139
Fabbiano G., Schweizer F., Mackie G., 1997, ApJ, 478, 542
Fabbiano G., Trinchieri G., 1983, ApJ, 266, L5
Fabbiano G., Zezas A., Murray S.S., 2000, BAAS, 32, 1532 \[Meeting 197, abstract 79.01\]
Fischer J., [et al.]{} 1996, A&A, 315, L97
Gao Y., Gruendl R.A., Lo K.Y., Lee S.-W., Hwang C.Y., 1998, BAAS, 30, 923 \[Meeting 192, abstract 69.04\]
Gao Y., Lo K.Y., Lee S.-W., Lee T.-H, 2001, ApJ, 548, 172
Gioia I.M., Maccacaro T., Schild R.E., Wolter A., Stocke J.T., Morris S.L., Henry J.P., 1990, ApJS, 72, 567
Giovanelli R., Haynes M.P., 1988, in ‘Galactic and Extragalactic Radio Astronomy’, Verschuur, G.L. & Kellerman, K.I. (eds), Springer-Verlag, p.522
Hibbard J.E., 1995, PhD Thesis (Columbia University)
Hibbard J.E., Mihos J.C., 1995, AJ, 110, 140
Hibbard J.E., Vacca W.D., Yun, M.S., 2000, AJ, 119, 1130
Hibbard J.E., van der Hulst J.M., Barnes J.E., 2001, in prep.
Hibbard J.E., van Gorkom J.H., 1996, AJ, 111, 655
Högbom J.A., 1974, A&AS, 15, 417
Huchtmeier W.K., Bohnenstengel H.D., 1975, A&A, 41, 477
Hummel E., van der Hulst J.M., 1986, A&A, 155, 151
Karachentsev I.D., Karachentseva V.E., Parnovskij S.L., 1993, Astronomische Nachrichten, 314, 97
Kaufman M., Brinks E., Elmegreen B.G., Elmegreen D.M., Struck C., Thomasson M., Klaric M., 1995, BAAS, 186, \#39.10
Kunze D., [et al.]{}, 1996, A&A, 315, L101
Maccacaro T., della Ceca R., Gioia I.M., Morris S.L., Stocke J.T., Wolter A., 1991, ApJ, 374, 117
Mahoney J.H., Burke B.F., van der Hulst J.M., 1987, IAU Symposium No. 117, 94
Malphrus B.K., Simpson C.E., Gottesman S.T., Hawarden T.G., 1997, ApJ, 114, 1427
Mihos J.C., Bothun G.D., Richstone D.O., 1993, ApJ, 418, 82
Mihos J.C., Hernquist L., 1994, ApJ, 431, L9
Mihos J.C., Richstone D.O., Bothun G.D., 1991, ApJ, 377, 72
Minkowski R., 1957, IAU Symposium No. 4, 107, Cambridge University Press
Mirabel I.F., Dottori H., Lutz D., 1992, A&A, 256, L19
Mirabel I.F., Lutz D., Maza J., 1991, A&A, 243, 367
Mirabel I.F., Vigroux L., [et al.]{} 1998, A&A, 333, L1
Neff S.G., Ulvestad J.S. 2000, AJ, 120, 670
Nikola T., Genzel R., Herrmann F., Madden S.C., Poglitsch A., Geis N., Townes C.H., Stacey G.J., 1998, ApJ, 504, 749
Parker P.R., 1990, ‘Colliding galaxies: the universe in turmoil’, Plenum Press, p.171
Pisano D.J., Wilcots E.M., 2000, MNRAS, 319, 821
Read A.M., Ponman T.J., Wolstencroft R.D., 1995, MNRAS, 277, 397
Roberts M.S., 1975, in ‘Galaxies and the Universe’, Sandage, A., Sandage, M. & Kristian, J. (eds), Chicago: University of Chicago Press, p.309
Rubin V.C., Ford W.K., D’Odorico S., 1970, ApJ, 160, 801
Sanders D.B., Mirabel I.F., 1985, ApJ, 298, L31
Sanders D.B., Mirabel I.F., 1996, ARA&A, 34, 749
Sansom A.E., Dotani T., Okada K., Yamashita A., Fabbiano G., 1996, MNRAS, 281, 48
Sault R.J., Teuben P.J., Wright M.C.H., 1995, in ‘Astronomical Data Analysis Software and Systems IV’, ASP Conf.Series 77, p.433
Schweizer F., 1978, in ‘Structure and Properties of Nearby Galaxies’, Berkhuijsen, E.M. & Wielebinski, R. (eds), IAU Symposium No. 77, p.279
Shapley H., Paraskevopoulos J.S., 1940, Proc. Nat. Acad. Sci., 26, 31
Stanford S.A., Sargent A.I., Sanders D.B., Scoville N.Z., 1990, ApJ, 349, 492
Steer D.G., Dewdney P.E., Ito M.R., 1984, A&A, 137, 159
Stocke J.T., [et al.]{} 1991, ApJS, 76, 813
Toomre A., 1977, in ‘Evolution of Galaxies and Stellar Populations’, Tinsky, B.M. & Larson, R.B. (eds), Yale University Observatory, New Haven, p.401
Toomre A., Toomre J., 1972, ApJ, 178, 623
van der Hulst J.M., 1979, A&A, 71, 131
Vigroux L., Mirabel F., [et al.]{}, 1996, A&A, 315, L93
Whitmore B.C., Schweizer F., 1995, AJ, 109, 960
Whitmore B.C., Zhang Q., Leitherer C., Fall S.M., Schweizer F., Miller B.W., 1999, AJ, 118, 1551
Wilson C.D., Scoville N., Madden S.C., Charmandaris V., 2000, ApJ, 542, 120
Young J.S., [et al.]{}, 1995, ApJS, 98, 219
[lcccccl]{} Name & & Optical Velocity & ${\rm B}^o_T$ & Galaxy Type\
& RA & DEC & (heliocentric)\
NGC 4038 & & & $1663\pm20~\kmpsE~^{(2)}$ & 10.59 & SB(s)m(pec)\
NGC 4039 & & & $1655\pm28~\kmpsE~^{(2)}$ & 10.69 & SA(s)m(pec)\
ESO 572–G045 & & & unknown & & Scd\
\
.\
.\
\
---------------------------------- ------------------------------------------------------------------------------------------
Instrument Australia Telescope Compact Array (ATCA)
Array configurations 6C (1996 July 24); 1.5A (1996 Oct 25); 750A (1996 Nov 12; 13); 375 (1998 Mar 28)
Baseline coverage 30.6–6000 , 75 baselines (56 unique)
Pointing centre RA=, DEC=
Frequency configuration IF1 (20): centre 1413 , bandwidth 8 , 512 channels, 2 polarisations
IF2 (13): centre 2378 (6C, 1.5A, 750A arrays) or 2496 (375 array)
bandwidth 128 MHz, 32 channels, 4 polarisations
Primary beam FWHM $33{\arcminE}$ (IF1 or 20), $21{\arcminE}$ (IF2 or 13)
Observing time per run 11–12 hours overall; $\sim8.5-9.5$ hours on source \[3.8/2.3 hours for second 750A run\]
Total observing time 50 hours overall; 39 hours on source
Flux calibration 10 min per run, using PKS 1934–638 (14.88 in IF1, 11.5 in IF2)
Phase calibration alternately 30 min on-source, 3 min on PKS 1151–348 (6.1/4.1 ) or 1127–145 (5.0/4.65 )
Calibrated Bandwidth IF1 (20): 6.8 (1436 as ) ; IF2 (13): 100
Image Weight; r.m.s. noise; Beam : ‘robust=1’; 1.0 vs 0.95 theoretical; $42{\arcsecE}$ beam
20 : ‘uniform’; 0.23 vs 0.25 theoretical; $16{\arcsecE}$ beam
13 : ‘uniform’; 0.056 vs 0.039 theoretical; 95 beam
---------------------------------- ------------------------------------------------------------------------------------------
[lcccccccc]{} Region & Velocities & Flux & Mass & Fraction & Peak Integrated &\
& () & () & () & of total & Intensity (& RA ($\pm2.5{\arcsecE}$) & DEC ($\pm2.5{\arcsecE}$)\
& & & & & and $\exponE{20}\,\pcmcmE$)\
Galactic disks & 1340–1945 & 26.3 & & 45 % & 5.62 ; 35.0 & &\
Southern tail & 1590–1784 & 28.8 & & 49 % & 1.59 ; 9.9 & &\
– Tip only & 1634–1740 & 10.5 & & 18.0 % & 1.59 ; 9.9 & &\
& 2.05 & & 3.5 % & 0.37 ; 2.3 & &\
– Main section & 1550–1622 & 1.84 & & 3.1 % & 0.37 ; 2.3 & &\
– Extension & 1550–1580 & 0.21 & & 0.25 % & 0.15 ; 0.93 & &\
ESO 572–G045 & 1640–1750 & 1.61 & & 2.7 % & 1.10 ; 6.9 & &\
\
& 30.9 & & 53 % & 1.59 ; 9.9 & &\
& 54.9 & & 94 % & 5.62 ; 35.0 & &\
& 56.9 & & 97 % & 5.62 ; 35.0 & &\
& 58.6 & & 100 % & 5.62 ; 35.0 & &\
[lccccl]{} Source Name & & & Comments\
/ Identifier & RA & DEC & 20 & 13 &\
A & & & 63.6 & 22 & Overlap of disks (4 subregions)\
B & & & 32 & 10.5 & NGC 4038 nucleus\
C & & & 19 & 6.5 & 5 south-east of NGC 4039 nucleus\
D & & & 1.6 & 1.6 & Midway between dwarf galaxies\
E & & & 3.4 & 3.4 & West of the galactic disks\
F & & & 11 & $<2$ & Extended ($\sim100{\arcsecE}\times30{\arcsecE}$) source (N–W)\
TXS 1159-187 & & & 108 & 43 & Catalogued radio source\
TXS 1158-187 & & & 78 & 36 & Catalogued radio source\
MS 1200.2-1829 & & & 5 & & Known X-ray & optical source\
\
\
---------------------------------------------- --------------------------------------------------
[**(a)**]{} moment-0 (integrated intensity) [**(b)**]{} moment-0 (a closeup of [**(a)**]{})
[**(c)**]{} moment-1 (mean velocity) [**(d)**]{} moment-1 (a closeup of [**(c)**]{})
---------------------------------------------- --------------------------------------------------
--------------------------------------------- --------------------------------------------------
[**(e)**]{} moment-2 (velocity dispersion) [**(f)**]{} moment-2 (a closeup of [**(e)**]{})
--------------------------------------------- --------------------------------------------------
[**Figure \[figure:HIMomentMaps\] continued.** ]{}
------------------------------------------------- --------------------------------------------------
[**(a)**]{} Projection onto (RA, velocity) axes [**(b)**]{} Projection onto (DEC, velocity) axes
------------------------------------------------- --------------------------------------------------
---------------------------------------------------- ----------------------------------------------------
[**(a)**]{} Slice along the major axis of NGC 4038 [**(b)**]{} Slice along the major axis of NGC 4039
---------------------------------------------------- ----------------------------------------------------
-- --
-- --
--------------------------------------------- --------------------------------------------
[**(a)**]{} spectrum of the Antennae disks [**(b)**]{} spectrum of ESO 572–G045
[**(c)**]{} spectrum of the northern tail [**(d)**]{} spectrum of the southern tail
--------------------------------------------- --------------------------------------------
[c]{}\
[c]{}\
[**Figure \[figure:HIChannelMaps\]**]{} (continued)
[c]{}\
[**Figure \[figure:HIChannelMaps\]**]{} (continued)
[c]{}\
[**Figure \[figure:HIChannelMaps\]**]{} (continued)
[c]{}\
[**Figure \[figure:HIChannelMaps\]**]{} (continued)
[c]{}\
[**Figure \[figure:HIChannelMaps\]**]{} (continued)
----------------------------------------- ----------------------------------------
[**(a)**]{} 20 radio continuum emission [**(b)**]{} 13 radio ontinuum emission
----------------------------------------- ----------------------------------------
[^1]: E-mail: [email protected]
[^2]: This paper assumes a Hubble constant of and a velocity relative to the Local Group of $1439~\kmpsE$ for the Antennae (Whitmore & Schweizer 1995 from de Vaucouleurs [et al.]{} 1991) giving a Hubble distance of 19.2 and a linear scale of 1 $\approx 5.6$ kpc.
[^3]: The Australia Telescope Compact Array is part of the Australia Telescope which is funded by the Commonwealth of Australia for operation as a National Facility managed by CSIRO.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Is it possible to break the host-vector chain of transmission when there is an influx of infectious hosts into a naïve population and competent vector? To address this question, a class of vector-borne disease models with an arbitrary number of infectious stages that account for immigration of infective individuals is formulated. The proposed model accounts for forward and backward progression, capturing the mitigation and aggravation to and from any stages of the infection, respectively. The model has a rich dynamic, which depends on the patterns of infected immigrant influx into the host population and connectivity of the transfer between infectious classes. We provide conditions under which the answer of the initial question is positive.'
author:
- |
Derdei Bichara\
Department of Mathematics & Center for Computational and Applied\
Mathematics, California State University, Fullerton, CA 92831, USA.
title: 'Effects of migration on vector-borne diseases with forward and backward stage progression'
---
[**Mathematics Subject Classification:**]{} 92D30, 34D23, 34D40, 34A34.
#### **Keywords:**
Vector-borne diseases, Migration, Stage progression, Stage amelioration, Global stability.
Introduction
============
Vector-borne diseases represent a major public health problem around the world, cause over one million deaths, one billion cases each year [@world2014global], and more than half of the world’s population is at risk [@world2004world]. They are typically associated with the tropics and subtropics where these diseases are endemic. However, recently these diseases have expanded their geographical distribution and have been reported in many temperate countries. For instance, Dengue and Chikungunya have been reported in France [@gould2010first; @grandadam2011chikungunya; @paty2014large; @vega2013high], Italy [@poletti2011transmission; @rezza2007infection], and Portugal [@wilder20142012]. The CDC recently [@CDCVectorBorneUSA] reported that mosquito, tick, and flea bite borne diseases tripled in the United States from 2004 through 2016.\
Many drivers are reported to be behind the geographic expansion of vector-borne diseases, including but not limited to trade, travel, climate change, urbanization and other social upheaval phenomena [@jones2008global; @kilpatrick2012drivers; @rogers2006climate]. Particularly, immigration and migration have been pointed to be the leading drivers in the emergence of vector-borne diseases in temperate nations [@barnett2008role]. For instance, a term *Airport malaria* has been coined by [@isaacson1989airport]. Indeed, it describes a malaria infection that has resulted from the bite of an infected tropical anopheline mosquito by persons whose geographic history excludes exposure to this vector in its natural habitat [@isaacson1989airport]. Hereafter, we use the term “immigrants" to represent both endemic area borne individuals migrating into a “naïve" area as well as non-endemic area native individuals who acquired an *airport vector-borne disease* after a stint in an endemic area. Naturally, the latter term follows the definition of *Airport malaria*.\
Moreover, the existence of vector populations that are capable of transmitting several arboviruses in the US and western Europe has been widely documented. For instance, the vector *Aedes albopictus* also known as the Asian tiger mosquito, a vector competent of transmitting many arbovirus including Japanese encephalitis, West Nile, and yellow fever, Dengue, etc., to humans [@mitchell1995geographic; @shroyer1986aedes], is well established in North America [@benedict2007spread; @mitchell1995geographic; @shroyer1986aedes] and Europe [@mitchell1995geographic].\
Together, the increasing interconnectedness of the world brings an influx of viremic (latent or infectious) individuals and their epidemiological life-history into naive geographic areas or populations, and thereby infecting local competent vector populations. This could potentially create a chain reaction that could lead to an autochthonous transmission cycle of arboviruses and sporadic outbreaks of vector-borne diseases in these otherwise naive host populations. For instance, the presumed index case of Italy’s 2007 Chikungunya outbreak was a man from India who developed symptoms while visiting relatives in one of the villages where the outbreak started [@rezza2007infection]. Similarly, an estimated 475 cases of imported chikungunya are reported in mainland France [@paty2014large] from November 2013 to June 2014, and these cases are reportedly traced back to travelers returning from the French Caribbean islands where chikungunya and Dengue are endemic [@paty2014large]. The 2012 Portugal’s Dengue outbreak was reported to be imported by a traveler from Venezuela [@wilder20142012]. It is therefore important the gauge the impacts of infected immigrants of the dynamics of vector-borne diseases.\
Typically, modeling the dynamics of directly transmitted or vector-borne diseases have often been based on the assumption that the recruitment into the considered population is completely susceptible, and thereby sweeping the effects of global movement of individuals across the world at unprecedented levels under the rug. To the best of our knowledge, Brauer and van den Driessche [@BraVdd01] were the to first propose a mathematical model that accounts for immigration that includes infected individuals using an SIS structure in an attempt to study HIV in prisons. Subsequently, McCluskey and van den Driessche [@1056.92052] proposed a model studied with the same features where both immigration of latent and infective are considered, in the context tuberculosis. These two papers [@BraVdd01; @1056.92052] showed that there is no disease-free equilibrium and that the endemic equilibrium (EE) is globally asymptotically stable. Indeed, the model proposed in [@BraVdd01] is a two-dimensional, and the Poincaré-Bendixson theorem has been used to study the global stability of the EE. In [@1056.92052], the authors considered model is an $SEIS$ model and a geometric approach [@Limul95; @MR95k:92022] is used to prove the global stability of the EE. Li et *al.* [@guo2012impacts] generalized the before-mentioned models for staged-progression model – by considering a model with $n$ infectious stages and a proportion $p_i$ of the total influx is incorporated for each infectious class $I_i$. The models in the before-mentioned papers are all for directly transmitted diseases.\
Recently, Tumwiine et *al.* [@tumwiine2010host] investigated the effects of infected immigrants using an $SIR-SI$ Ross-Macdonal’s model in the context of malaria and showed that the disease persists in host and vector populations whenever the proportion of infected immigrants is non zero. However, their model does not account of immigration of latent individuals, a critical category as these pass the precautionary measures of screenings, if these were in place.\
In this paper, we derive and investigate the global behavior of a system that captures the dynamics of a class of vector-borne models accounting for flux of infected individuals at different stages of infection. Of particular interest is the impact of the influx of infected individuals and transfer rates between infectious classes on the overall the dynamics of the model. The paper is organized as follows:\
$\bullet$ We derive a class of vector-borne models with $n$ stages of infection, for which there is a flux of infected and infectious immigrants at all of these infectious stages. The formulated model accounts also for the progression and amelioration during the infectious stages, from an arbitrary stage $i$ to an arbitrary stage $j$. The transfer is considered a progression if $i>j$ and an amelioration if $i<j$ ().\
$\bullet$ We completely study the dynamics of the proposed model (). It turns out that the model has a variety of dynamics, which depends on the patterns of the influx infected host into the population and the transfer rate matrix – that describes the amelioration and deterioration of hosts’ infectivity level. Particularly, we show that, under certain conditions, it is possible to corral the infectious hosts only into the classes in which they are replenished and maintain the vector populations disease-free. A threshold $\mathcal N^2(p,\mathbf{p},p_{n+1})$ plays a critical role for the existence of such steady-state.\
$\bullet$ We provide the global dynamics of the model when there is no influx of infected individuals into the population, which surprisingly has not been done (). In this case, the model exhibits the threshold phenomenon – the basic reproduction number $\mathcal R_0^2$ determines the outcome of the disease both in host and vector population. It happens that $\mathcal R_0^2:=\mathcal N^2(0,\mathbf{0},0)$.\
$\bullet$ Illustrations of the results and numerical simulations are carried out in .
Formulation of the model {#sec:SPHostFlux}
========================
We consider a disease whose evolution is captured by a host-vector interaction for which the host population is composed of susceptible, exposed, recovered and infectious of stage $i$ ($1\leq i\leq n$). These subpopulations are denoted respectively by $S_h$, $E_h$, $R_h$ and $I_i$. The total host population is therefore $N_h=S_h+E_h+\sum_{i=1}^nI_i+R_h$. The vector population $N_v$ is composed of susceptible, exposed, and infectious arthropods; denoted by $S_v$, $E_v$, and $I_v$, respectively.\
The total host population is replenished through a constant recruitment, $\pi_h$, that includes birth and migratory influx of individuals. Of this constant recruitment, a proportion $p$, $p_i$ ($i=1,\dots,n$) and $p_{n+1}$ is latent, infectious at stage $i$ and recovered, respectively. Thus, the total recruitment in the susceptible class is $\pi_h\left( 1-p-\sum_{i=1}^{n+1}p_i \right)$. Naturally, we assume that, $0\leq p\leq1$, $0\leq p_i\leq1$ and $0\leq p+\sum_{i=1}^{n+1}p_i<1$.\
Susceptible hosts are infected at the rate $a\beta_{vh}\frac{I_v}{N_n}$, where $a$ is the biting/landing rate and $\beta_{vh}$ is the Host’s infectiousness by arthropod per biting/landing. Vector’s infection term is captured by $a S_{v}\frac{\sum_{i=1}^n\beta_iI_i}{N_n},$ where $\beta_i$ is the vector’s infectiousness by infected hosts of stage $i$. This accounts for the differential infectivity of vectors with respect to hosts’ infectious stages.\
Motivated by [@guo2012impacts; @guo2012global; @1008.92032], we incorporate incremental and non-incremental amelioration and recrudescence in the infectiosity at each stage of the host’s infection. For instance, for models in [@BicharaIggidrSmith2017; @cruz2012control; @palmer2018dynamics], the transitions between infection stages are incremental, that is, always from stage $i$ to $i+1$. However, with vector-borne diseases, a bite of infected arthropod to an already infected host, say at stage $i$, may increase this host’s parasitemia, thereby catapulting its infectious class from stage $i$ to any stage, say $i+j$, where $1\leq i+j\leq n$. To incorporate this phenomenon, we denote by $\gamma_{ij}$, the *per capita* rate at which the host progresses from stage $i$ to stage $j$. Similarly, the increase of treatments (which decrease the parasiteamia in the blood-stream) of vector-borne diseases could alleviate the host’s infection and therefore, its stage could change form $i$ to $i-k$, where $ i-k\geq 0$. We denote by $\delta_{ij}$ the *per capita* rate at which the host regresses from stage $i$ to stage $j$, where $j\leq i$. These generalizations are illustrated in .\
In concert, the overall dynamics of the Host-Vector infection is given by: $$\label{ModelAD}
\left\{\begin{array}{llll}\displaystyle\dot S_{h}=\pi_h\left( 1-p-\sum_{i=1}^{n+1}p_i \right)-a\,\beta_{vh}\,S_{h}\,\dfrac{I_{v}}{N_{h}}-\mu_h S_h\\
\displaystyle\dot E_{h}=p\pi_h+ \,a\beta_{vh}\,S_{h}\,\dfrac{I_{v}}{N_{h}}-(\mu_h+\nu_h+\eta)E_h\\
\displaystyle\dot I_{1}=p_1\pi_h+\nu_hE_h-(\mu_h+\eta_1)I_1-I_{1}\sum_{j=2}^n\gamma_{1j}+\sum_{j=2}^n\delta_{j1}I_{j}\\
\displaystyle\dot I_{2}=p_2\pi_h-(\mu_h+\eta_2)I_2-I_{2}\left(\sum_{j=1,j<2}^n\delta_{2j}+\sum_{j=1,j>2}^n\gamma_{2j}\right)+\left(\sum_{j=1,j>2}^n\delta_{j2}I_j+\sum_{j=1,j<2}^n\gamma_{j2}I_j\right)\\
\vdots\\
\displaystyle\dot I_{i}=p_i\pi_h-(\mu_h+\eta_i)I_i-I_{i}\left(\sum_{j=1,j<i}^n\delta_{ij}+\sum_{j=1,j>i}^n\gamma_{ij}\right)+\left(\sum_{j=1,j>i}^n\delta_{ji}I_j+\sum_{j=1,j<i}^n\gamma_{ji}I_j\right)\\
\displaystyle\dot I_{n}=p_i\pi_h-(\mu_h+\eta_n)I_n-I_{n}\sum_{j=1}^{n-1}\gamma_{nj}+\sum_{j=1}^{n-1}\delta_{jn}I_{j}\\
\displaystyle \dot R_h=p_{n+1}\pi_h+\sum_{i=1}^n\eta_iI_{i}-\mu_hR\\
\displaystyle\dot S_{v}=\pi_v-a S_{v}\sum_{i=1}^n\dfrac{\beta_i I_{i}}{N_{h}}-(\mu_v+\delta_v)S_{v}\\
\displaystyle\dot E_{v}=a S_{v}\sum_{i=1}^n\dfrac{\beta_i I_{i}}{N_{h}}-(\mu_v+\nu_v+\delta_v)E_{v}\\
\displaystyle\dot I_{v}=\nu_vE_v-(\mu_v+\delta_v)I_{v}
\end{array}\right.$$
To ease the notations, let us denote by $$m_{ij}=\left\{\begin{array}{lll}\delta_{ji}\quad\textrm{if}\quad i<j,\\
0\quad\textrm{if}\quad i=j,\\
\gamma_{ji}\quad\textrm{if}\quad i>j,
\end{array}\right.$$and $\alpha_v=\mu_v+\nu_v+\delta_v$, $\alpha_1=\mu_h+\eta_1 +\sum_{j=2}^n\gamma_{1j}=\mu_h+\eta_1 +\sum_{j=2}^nm_{j1}$ and for $i\geq2$, $$\begin{aligned}
\alpha_i&=&\mu_h+\eta_i +\left(\sum_{j=1,j<i}^n\delta_{ij}+\sum_{j=1,j>i}^n\gamma_{ij}\right)\nonumber\\
&=&\mu_h+\eta_i + \sum_{j=1}^nm_{ji}. \nonumber\end{aligned}$$ The matrix $M=(m_{ij})_{1\leq i,j\leq n}$ is the transfer matrix between Host’s infectious classes and the parameters $\alpha_i$ represent the rates at which infected of stage $i$ leave this stage. The total host population is asymptotically constant. Indeed, its evolution is given by $\dot N_h=\pi_h-\mu_hN_h$ and thus, it is straightforward to show that $\displaystyle\lim_{t\to\infty}N_h=\dfrac{\pi_h}{\mu_h}.$ Moreover, the subsystem describing the dynamics of the host is triangular, and hence we can disregard the dynamics of the recovered host $R_h$. Hence, by abusively denoting $\displaystyle\lim_{t\to\infty}N_h$ again by $N_h$ and using the theory of asymptotically autonomous systems for triangular systems [@CasThi95; @0478.93044], System (\[ModelAD\]) could equivalently be written in a compact form as follows: $$\label{ModelAD2Compact}
\left\{\begin{array}{llll}\displaystyle\dot S_{h}=\pi_h\left( 1-p-\mathds{1}^T\mathbf{p}-p_{n+1} \right)-a\,\beta_{vh}\,S_{h}\,\dfrac{I_{v}}{N_{h}}-\mu_h S_h\\
\displaystyle\dot E_{h}=p\pi_h+ \,a\beta_{vh}\,S_{h}\,\dfrac{I_{v}}{N_{h}}-(\mu_h+\nu_h+\eta)E_h\\
\displaystyle\dot{\mathbf{I}}_h=\pi_h\mathbf{p}+\nu_hE_he_1-(\textrm{diag}(\alpha)-M)\mathbf{I}_h\\
\displaystyle\dot S_{v}=\pi_v
-a\dfrac{ S_{v}}{N_h}\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, \mathbf{I}_h
\bigr\rangle
-(\mu_v+\delta_v)S_{v}\\
\displaystyle\dot E_{v}=a\dfrac{ S_{v}}{N_h}\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, \mathbf{I}_h
\bigr\rangle-(\mu_v+\nu_v+\delta_v)E_{v}\\
\displaystyle\dot I_{v}=\nu_vE_v-(\mu_v+\delta_v)I_{v},
\end{array}\right.$$ where $\mathbf{I}_h=(I_1,I_2,\dots,I_n)^T$, $\Beta=(\beta_1,\beta_2,\dots,\beta_n)^T$, $\mathbf{p}=(p_1,p_2,\dots,p_n)^T$, and $M=G^T+D^T$ with $G=(\gamma_{ij})$ representing the progression matrix, or forward flow transition matrix while $D=(\delta_{ij})$ represents the amelioration matrix, or backward transition flow matrix. More precisely,
$$\label{GD}G=\begin{pmatrix}
0 & \gamma_{12}&\gamma_{13} & \dots & \gamma_{1n}\\
0 &0 &\gamma_{23} & \dots & \gamma_{2n}\\
\vdots& \vdots &\ddots & \ddots & \vdots \\
0 & 0 &0 & \dots & \gamma_{n-1,n}\\
0 & 0 &0 & \dots & 0
\end{pmatrix}
\quad\textrm{and}\quad
D=\begin{pmatrix}
0 & 0& \dots & 0& 0\\
\delta_{21} &0 & \dots & 0& 0\\
\vdots& \ddots &\ddots & \vdots& \vdots \\
\delta_{n-1,1} & \delta_{n-1,2} & \dots &0& 0\\
\delta_{n1} & \delta_{n2} & \dots &\delta_{n,n-1}& 0
\end{pmatrix}.$$
The parameters of System (\[ModelAD\]) are described in Table \[TableHV1\]. The flow chart capturing the infection process is represented in Fig. (\[fig:FlowSEIR1HostAmeliorationDeterioration\]).\
\[TableHV1\]
Parameters Description
--------------- ----------------------------------------------------------------
$\pi_h$ Recruitment of the host
$\pi_v$ Recruitment of vectors
$p$ Proportion of latent immigrants
$p_i$ Proportion of infectious immigrants at stage $i$
$a$ Biting rate
$\mu_h$ Host’s natural death rate
$\beta_{v,h}$ Host’s infectiousness by mosquitoes per biting
$\beta_{i}$ Vector’s infectiousness by host at stage $i$ per biting
$\nu_h$ Host’s rate at which the exposed individuals become infectious
$\eta_i$ Per capita recovery rate of an infected host at stage $i$
$\gamma_{ij}$ Host’s per capita progression rate from stage $i$ to $j$
$\delta_{ij}$ Host’s per capita regression rate from stage $i$ to $j$
$\mu_v$ Vectors’ natural mortality rate
$\delta_v$ Vectors’ control-induced mortality rate
$\nu_v$ Rate at which the exposed vectors become infectious
: Description of the parameters used in System (\[ModelAD\]).[]{data-label="tab:Param"}
Model (\[ModelAD2Compact\]) follows an $SEI^nR-SEI$ structure. That is, of the host and vector populations dynamics follow an and $SEI^nR$ and $SEI$ types of model, respectively. The choices are make to capture some key features in modeling different vector-borne diseases. Indeed, many special cases could be obtained from our general framework to fit a particular arboviral disease. For instance, if $\nu_h\to\infty$, the Host’s dynamics will be an $SI^nR$ model. An $SIR-SI$ model have been considered for malaria [@tumwiine2010host] and Dengue [@Esteva98Dengue] while an $SI^nR-SI$ model was deemed more suited for tick-borne relapsing fever [@johnson2016modeling; @palmer2018dynamics].\
Model (\[ModelAD2Compact\]) generalizes other models proposed in the literature in the following five ways:
- If $D=\mathbf{0}_{n,n}$ and $\gamma_{ij}=0$ for all $i,j$, except when $j=i+1$, Model (\[ModelAD2Compact\]) consists of a class of staged progression vector-borne diseases models with an influx of infected individuals of each class into the considered population. In this case,
- Model (\[ModelAD2Compact\]) extents the existing stage progression vector-borne models to incorporate a differential proportions of the overall recruitment in all infected classes. This allows us to gauge the impact of imported cases on the dynamics of vector-borne infections. When $\mathbf{p}=\mathbf{0}$, $p=0$, in Model (\[ModelAD2Compact\]), we obtain the model proposed and studied in [@BicharaIggidrSmith2017]. Moreover, our model extends also [@palmer2018dynamics], for which $\beta_i=\beta$, for all $i$ and the recruitment constitutes of susceptible individuals only. Our model generalizes also [@cruz2012control], which considers Chagas disease model with two stages, namely acute and chronic phases.
- Model (\[ModelAD2Compact\]) generalizes existing models that investigate staged-progression for directly transmitted infections for which influx of infected individuals are considered [@guo2012impacts; @1056.92052] and [@BraVdd01], where no stages are considered in the latter.
- Model (\[ModelAD2Compact\]) extends also the model in [@tumwiine2010host], where the authors considered a host-vector model $SIR-SI$ with infectious immigrants in investigating the effects of the latter on Malaria dynamics, by incorporating a latent class, $n$ stages of infection in the host’s dynamics and a differential infectivity of vectors with respect to host’s infectious stages.
- If $D$ and $G$ are as defined in (\[GD\]), our model extends [@BicharaIggidrSmith2017; @BraVdd01; @cruz2012control; @guo2012impacts; @1056.92052; @palmer2018dynamics; @tumwiine2010host] by incorporating forward (deterioration) and backward (amelioration) stage progression. Moreover, the progressions or regressions are not necessarily incremental.
- Model (\[ModelAD2Compact\]) extends the models considered in [@guo2012impacts; @guo2012global; @1008.92032] to vector-borne disease models and the incorporation of influx of infected in each of the hosts’ infectious classes.
Overall, Model (\[ModelAD2Compact\]) generalizes in some fashion or aspect models in [@BicharaIggidrSmith2017; @BraVdd01; @cruz2012control; @guo2012impacts; @guo2012global; @1056.92052; @palmer2018dynamics; @tumwiine2010host; @1008.92032].
Also, it worthwhile to notice that our system could be seen as vector-borne model that vertically transmitted at each stage of the infection. That is, when off-springs are infected by mothers during pregnancy or delivery. Zika virus is a natural example of a vector-borne disease that is vertically transmitted [@tabata2016zika].\
The following result shows the solutions of Model (\[ModelAD2Compact\]) are positive and remain bounded at all times, thereby making the model biologically grounded.
The set $$\Omega=\left\{ (S_h,E_h,\mathbf{I}_h,S_v,E_v,I_v)\in\R^{n+5}\;\mid S_h+E_h+\mathds{1}^T\mathbf{I}_h\leq N_h, S_v+E_v+I_v\leq N_v:=\frac{\pi_v}{\mu_v+\delta_v} \right\}$$ is a compact positively invariant for System (\[ModelAD2Compact\]).
In the next section, we investigate the steady states solutions of Model (\[ModelAD2Compact\]) and their asymptotic behavior.
Global Stability Analysis {#sec:GSVBAP}
=========================
The next theorem establishes the existence of endemic equilibria of System (\[ModelAD2Compact\]) and provides conditions under which they may exist. Following Thieme [@MR1993355], we use the nomenclature *strongly* endemic equilibrium if all of its components are positive and *weakly* endemic equilibrium, if at least one of the infected component is positive. Naturally, we start with the assumption that $\mathbf{p}\neq\mathbf{0}_{\mathbb{R}^n}$. The case $\mathbf{p}=\mathbf{0}_{\mathbb{R}^n}$ is dealt in .
\[ExistenceEEAD\]
The equilibria of System (\[ModelAD2Compact\]) are as follows:
1. \[i\] If $\Beta=0$, a unique weakly endemic equilibrium $(\bar{S}_h, \bar{E}_h,\bar{\mathbf{I}}_h,\bar{S}_v,0,0)$ exists.
2. \[ii\] If $\Beta\neq0$ and $p\neq0$, it exists a unique strongly endemic equilibrium $({S}_h^*, {E}_h^*,\mathbf{I}_h^*,S_v^*,E_v^*,I_v^*)$.
3. \[iii\] If $\Beta$ and $\mathbf{p}$ are such that $\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, (\textrm{diag}(\alpha)-M)^{-1} \mathbf{p}
\bigr\rangle\neq0$, a unique strongly endemic equilibrium $({S}_h^\sharp, E_h^\sharp,\mathbf{I}_h^\sharp,{S}_v^\sharp,E_v^\sharp,I_v^\sharp)$ exists.
4. \[iv\] If , and are not satisfied, then a threshold $\mathcal N_0^2(p,\mathbf{p},p_{n+1})$, defined by: $$\mathcal N_0^2(p,\mathbf{p},p_{n+1})=\frac{a^2\beta_{vh}\nu_v\nu_h}{\alpha_v(\mu_v+\delta_v)\alpha_h\mu_h}\frac{N_v}{N_h}\frac{\pi_h\left( 1-p-\mathds{1}^T\mathbf{p}-p_{n+1} \right)}{N_h}\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, (\textrm{diag}(\alpha)-M)^{-1} e_1
\bigr\rangle,$$ exists and for which
- If $\mathcal N_0^2(p,\mathbf{p},p_{n+1})\leq1$, a unique weakly endemic equilibrium $({S}_h^\diamond, 0,\mathbf{I}_h^\diamond,{S}_v^0,0,0)$ where $\mathbf{I}_h^\diamond>0$ exists.
- If $\mathcal N_0^2(p,\mathbf{p},p_{n+1})>1$, a unique strongly endemic equilibrium $(\tilde{S}_h, \tilde{E}_h,\tilde{\mathbf{I}}_h,\tilde{S}_v,\tilde{E}_v,\tilde{I}_v)$ exists.
An equilibrium $(S_h^\ast,E_h^\ast,I_h^\ast,\dots,I_n^\ast,S_v^\ast,E_v^\ast,I_v^\ast)$ for Model (\[ModelAD2Compact\]) satisfies the following relations. $$\label{ModelAD2Endemic1}
\left\{\begin{array}{llll}\displaystyle \Lambda_h=a\,\beta_{vh}\,S_{h}^\ast\,\dfrac{I_{v}^\ast}{N_{h}}+\mu_h S_h^\ast\\
\displaystyle (\mu_h+\nu_h+\eta)E_h^\ast=p\pi_h+ \,\beta_{vh}\,S_{h}^\ast\,\dfrac{I_{v}^\ast}{N_{h}}\\
\displaystyle (\textrm{diag}(\alpha)-M)\mathbf{I}_h^\ast=\pi_h \mathbf{p}+\nu_hE_h^\ast e_1\\
\displaystyle \pi_v=
a\dfrac{ S_{v}^\ast}{N_h}\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, \mathbf{I}_h^\ast
\bigr\rangle
+(\mu_v+\delta_v)S_{v}^\ast\\
\displaystyle (\mu_v+\nu_v+\delta_v)E_{v}^\ast=a\dfrac{ S_{v}^\ast}{N_h}\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, \mathbf{I}_h^\ast
\bigr\rangle\\
\displaystyle (\mu_v+\delta_v)I_{v}^\ast=\nu_vE_v^\ast,
\end{array}\right.$$ where $\displaystyle\Lambda_h=\pi_h\left( 1-p-\mathds{1}^t\mathbf{p} -p_{n+1}\right)$. Using these relationship and $N_v=S_v-E_v-I_v$, one could express $I_v^\ast$ in terms of $\mathbf{I}_h^\ast$, as follows: $$\label{IVstarEE}I_v^\ast\left(\mu_v+\delta_v+\frac{a}{N_h}\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, \mathbf{I}_h^\ast
\bigr\rangle\right)=\frac{a\nu_v}{(\mu_v+\nu_v+\delta_v)}\frac{N_v}{N_h}\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, \mathbf{I}_h^\ast
\bigr\rangle.$$ Moreover, the first equation of (\[ModelAD2Endemic1\]) leads to: $$S_h^\ast=\frac{\Lambda_hN_h}{\mu_hN_h+a\beta_{vh}I_v^\ast}>0.$$ Furthermore, since the matrix $\textrm{diag}(\alpha)-M$ is strictly diagonally dominant and thus invertible, we obtain: $$\mathbf{I}_h^\ast=\pi_h(\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}+\nu_hE_h^\ast(\textrm{diag}(\alpha)-M)^{-1}e_1,$$ and $$\begin{aligned}
E_h^\ast
&=& \frac{1}{(\mu_h+\nu_h+\eta)}\left(p\pi_h+\,\dfrac{\Lambda_ha\,\beta_{vh}I_{v}^\ast}{\mu_hN_h+a\beta_{vh}I_v^\ast}\right).\end{aligned}$$ Therefore, $\mathbf{I}_h^\ast$ could be written as: $$\label{Ihast}
\mathbf{I}_h^\ast=\pi_h(\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}+\frac{\nu_h}{(\mu_h+\nu_h+\eta)}\left(p\pi_h+\,\dfrac{\Lambda_ha\,\beta_{vh}I_{v}^\ast}{\mu_hN_h+a\beta_{vh}I_v^\ast}\right)(\textrm{diag}(\alpha)-M)^{-1}e_1.$$ The relation (\[Ihast\]) is key for the remaining of the proof, as we will use it to compute $\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, \mathbf{I}_h^\ast
\bigr\rangle$ and obtain a quadratic equation in $I_v^\ast$ using . The latter equation leads to:
$$\begin{gathered}
\label{IVstarEEQuadratic}
0=I_v^\ast\left(\mu_v+\delta_v+\frac{a}{N_h}\bigl\langle
\Beta \, \mbox{\Large $\mid$} \, \pi_h(\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}\bigr\rangle+\right.\\
\left.+\frac{a}{N_h}\frac{\nu_h}{(\mu_h+\nu_h+\eta)}\left(p\pi_h+\,\dfrac{\Lambda_ha\,\beta_{vh}I_{v}^\ast}{\mu_hN_h+a\beta_{vh}I_v^\ast}\right)\bigl\langle\Beta \, \mbox{\Large $\mid$} (\textrm{diag}(\alpha)-M)^{-1}e_1
\bigr\rangle\right)-\\
\frac{a\nu_v}{(\mu_v+\nu_v+\delta_v)}\frac{N_v}{N_h}\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, \pi_h(\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}\bigr\rangle-\\
\frac{\nu_h}{(\mu_h+\nu_h+\eta)}\left(p\pi_h+\,\dfrac{\Lambda_ha\,\beta_{vh}I_{v}^\ast}{\mu_hN_h+a\beta_{vh}I_v^\ast}\right)\frac{a\nu_v}{(\mu_v+\nu_v+\delta_v)}\frac{N_v}{N_h}\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \,(\textrm{diag}(\alpha)-M)^{-1}e_1
\bigr\rangle\end{gathered}$$
After some rearrangement, could be written as $$\label{quad}A{I_v^\ast}^2+BI^\ast_v+C=0,$$ where $$\begin{aligned}
A&=a\beta_{vh}\left[ \mu_v+\delta_v+\frac{a}{N_h}\bigl\langle
\Beta \, \mbox{\Large $\mid$} \, \pi_h(\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}\bigr\rangle+\right.\\
& \left.+\frac{a}{N_h}\frac{\nu_h}{(\mu_h+\nu_h+\eta)}\left(p\pi_h+ \Lambda_h \right)\bigl\langle\Beta \, \mbox{\Large $\mid$} (\textrm{diag}(\alpha)-M)^{-1}e_1
\bigr\rangle \right]\\
&>0,\end{aligned}$$ $$\begin{aligned}
B&=(\mu_v+\delta_v)\mu_nN_h +\frac{a}{N_h}\pi_h\bigl\langle
\Beta \, \mbox{\Large $\mid$} \,(\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}\bigr\rangle\mu_hN_h\\
&+\frac{a}{N_h}\frac{\nu_h}{(\mu_h+\nu_h+\eta)}p\pi_h\bigl\langle\Beta \, \mbox{\Large $\mid$} (\textrm{diag}(\alpha)-M)^{-1}e_1
\bigr\rangle \mu_nN_h\\
&-\frac{a^2\beta_{vh}\nu_v}{(\mu_v+\nu_v+\delta_v)}\frac{N_v}{N_h}\pi_h\bigl\langle
\Beta \, \mbox{\Large $\mid$} \,(\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}\bigr\rangle\\
&-\frac{a^2\beta_{vh}\nu_v\nu_h}{(\mu_v+\nu_v+\delta_v)(\mu_h+\nu_h+\eta)}\frac{N_v}{N_h}p\pi_h\bigl\langle
\Beta \, \mbox{\Large $\mid$} \,(\textrm{diag}(\alpha)-M)^{-1}e_1\bigr\rangle\\
&-\frac{a^2\beta_{vh}\nu_v\nu_h}{(\mu_v+\nu_v+\delta_v)(\mu_h+\nu_h+\eta)}\frac{N_v}{N_h}\Lambda_h\bigl\langle
\Beta \, \mbox{\Large $\mid$} \,(\textrm{diag}(\alpha)-M)^{-1}e_1\bigr\rangle,\end{aligned}$$ and $$\begin{aligned}
C&=-\frac{a\nu_v}{(\mu_v+\nu_v+\delta_v)}\frac{N_v}{N_h}\pi_h\mu_hN_h \bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, \pi_h(\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}\bigr\rangle\\
&-
\frac{a\nu_v N_v}{(\mu_v+\nu_v+\delta_v)} \frac{N_v}{N_h} \frac{\nu_h \mu_hN_h}{(\mu_h+\nu_h+\eta)}p\pi_h\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \,(\textrm{diag}(\alpha)-M)^{-1}e_1
\bigr\rangle.\end{aligned}$$ Now, we investigate cases for which has non-negative solutions.\
$\bullet$ If $\Beta=0$, then $C=0$ and $B=(\mu_v+\delta_v)\mu_nN_h>0$. Hence, $I_v^*=0$ is the unique solution of the quadratic equation. Thus, the unique equilibrium for System (\[ModelAD2Compact\]) is $(\bar{S}_h, \bar{E}_h,\bar{\mathbf{I}}_h,\bar{S}_v,0,0)$, where $\bar{S}_h=\frac{\Lambda_h}{\mu_h}$, $ \bar E_h=\frac{p\pi_h}{(\mu_h+\nu_h+\eta)}$, $\bar{\mathbf{I}}_h^\ast=\pi_h(\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}+\frac{\nu_hp\pi_h}{(\mu_h+\nu_h+\eta)} (\textrm{diag}(\alpha)-M)^{-1}e_1,
$ and $\bar{S}_v=\frac{\Lambda_v}{\mu_v+\delta_v}$. This proves .\
$\bullet$ If $\Beta\neq0$ and $p\neq0$, then $C<0$ and therefore has a unique solution such that $I_v^*>0$. Thus, from , System (\[ModelAD2Endemic1\]), and using the fact that $(\textrm{diag}(\alpha)-M)^{-1}e_1\gg0$, we deduce .\
$\bullet$ If $\Beta$ and $p$ are such that $\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, (\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}\bigr\rangle\neq0$, then we also have $C<0$; that is, it exists a unique $I_v^\sharp>0$ of . As in the previous point, this leads to .\
$\bullet$ If the conditions of , and are not satisfied; that is, if $\Beta\neq0$, $p=0$ and $\mathbf{p}$ is such that $\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, (\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}\bigr\rangle=0$. In this case, $C=0$ and $B$ could be written as: $$\begin{aligned}
B &=(\mu_v+\delta_v)\mu_hN_h \\
&-\frac{a^2\beta_{vh}\nu_v\nu_h}{(\mu_v+\nu_v+\delta_v)(\mu_h+\nu_h+\eta)}\frac{N_v}{N_h}\Lambda_h\bigl\langle
\Beta \, \mbox{\Large $\mid$} \,(\textrm{diag}(\alpha)-M)^{-1}e_1\bigr\rangle \\
&=(\mu_v+\delta_v)\mu_hN_h \left[1
-\frac{a^2\beta_{vh}\nu_v\nu_h}{(\mu_v+\nu_v+\delta_v)(\mu_v+\delta_v)(\mu_h+\nu_h+\eta)\mu_h}\frac{N_v}{N_h}\frac{\Lambda_h}{N_h}\bigl\langle
\Beta \, \mbox{\Large $\mid$} \,(\textrm{diag}(\alpha)-M)^{-1}e_1\bigr\rangle \right]\\
&=(\mu_v+\delta_v)\mu_hN_h\left(1-\mathcal N_0^2(p,\mathbf{p},p_{n+1}) \right).\end{aligned}$$ Thus, it follows that if $\mathcal N_0^2(p,\mathbf{p},p_{n+1})\leq1$, then $B\geq0$, leading to $I_v^*=0$ and $\mathbf{I}_h^\ast=\pi_h(\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}>\mathbf{0}_{\R^n}$. If $\mathcal N_0^2(p,\mathbf{p},p_{n+1})>1$, then $B<0$ and therefore $I_v^*>0$, leading to a strongly positive equilibrium.
The following two theorems establish the global stability analysis for the two types of endemic equilibria exhibited in . This gives a complete description of the global asymptotic behavior of System (\[ModelAD2Compact\]) whenever there is an influx of infected or infectious individuals into the population.
\[GASEEAD\]\
Let $({S}_h^*, {E}_h^*,\mathbf{I}_h^*,S_v^*,E_v^*,I_v^*)$ be a strongly endemic equilibrium of Model (\[ModelAD2Compact\]). This equilibrium is globally asymptotically stable whenever it exists.
\
Let consider the following Lyapunov function $ \mathcal V= \mathcal V_h+ \mathcal V_v$, where $$\begin{aligned}
\mathcal V_h&=&c_0\int_{S_h^\ast}^{S_h}\left(1-\frac{S_h^\ast}{x}\right)dx+c_0\int_{E_{h}^\ast}^{E_h}\left(1-\frac{E_h^\ast}{x}\right)dx+\sum_{i=1}^nc_{i}\int_{I_{i}^\ast}^{I_{i}}\left(1-\frac{I_{i}^\ast}{x}\right)dx,
\end{aligned}$$ and, $$\mathcal V_{v}=c_{v}\int_{S_{v}^\ast}^{S_{v}}\left(1-\frac{S_{v}^\ast}{x}\right)dx +c_{v}\int_{E_{v}^\ast}^{E_{v}}\left(1-\frac{E_{v}^\ast}{x}\right)dx+\frac{\mu_v+\nu_v+\delta_v}{\nu_v}c_{v}\int_{I_{v}^\ast}^{I_{v}}\left(1-\frac{I_{v}^\ast}{x}\right)dx.$$ The coefficients $c=(c_1,c_2,\dots,c_n)^T$ are positive to be determined later. The coefficient $c_0$ and $c_{v}$ are related to $c_1$ as follows: $$\label{c0cvc1}c_0a\beta_{vh}\,S_{h}^*\dfrac{I_{v}^*}{N_{h}}=c_1\nu_hE_h^*\quad\textrm{and}\quad c_{v} aS_v^\ast\frac{1}{N_h}=c_1\frac{\nu_hE_h^\ast}{\sum_{i=1}^n \beta_i I_i^\ast}.$$ This function is definite positive. We want to prove that its derivative along the trajectories of System (\[ModelAD2Compact\]) is definite-negative. Throughout the proof, we will be using the component-wise endemic relations (\[ModelAD2Endemic1\]). That is, $$\label{ModelAD2Endemic}
\left\{\begin{array}{llll}
\displaystyle \pi_h\left( 1-p-\sum_{i=1}^{n+1}p_i \right)=a\,\beta_{vh}\,S_{h}^\ast\,\dfrac{I_{v}^\ast}{N_{h}}+\mu_h S_h^\ast\\
\displaystyle (\mu_h+\nu_h+\eta)E_h^\ast=p\pi_h+ \,a\beta_{vh}\,S_{h}^\ast\,\dfrac{I_{v}^\ast}{N_{h}}\\
\displaystyle \alpha_1I_1^\ast=p_1\pi_h+\nu_hE_h^\ast+\sum_{j=2}^nm_{1j}I_{j}^\ast\\
\displaystyle \alpha_2I_{2}^\ast=p_2\pi_h+\sum_{j=1}^nm_{2j}I_j^\ast \\
\vdots\\
\displaystyle \alpha_iI_i^\ast=p_i\pi_h+ \sum_{j=1}^nm_{ij}I_j^\ast\\
\displaystyle \alpha_nI_n^\ast=p_n\pi_h+\sum_{j=1}^{n}m_{nj}I_{j}^\ast\\
\displaystyle \Lambda_v=a S_{v}^\ast\sum_{i=1}^n\dfrac{\beta_i I_{i}^\ast}{N_{h}}+(\mu_v+\delta_v)S_{v}^\ast\\
\displaystyle (\mu_v+\nu_v+\delta_v)E_{v}^\ast=a S_{v}^\ast\sum_{i=1}^n\dfrac{\beta_i I_{i}^\ast}{N_{h}}\\
\displaystyle (\mu_v+\delta_v)I_{v}^\ast=\nu_vE_v^\ast
\end{array}\right.$$
The derivative of $\mathcal{V}_h$ along the trajectories of System (\[ModelAD2Compact\]) is: $$\begin{aligned}
\label{Lyapc1}
\dot {\mathcal V_h} &=&c_0\left(1-\frac{S_h^\ast}{S_h}\right)\dot{S}_h+c_0\left(1-\frac{E_h^\ast}{E_{h}}\right)\dot E_{h}+\sum_{i=1}^nc_i\left(1-\frac{I_i^\ast}{I_i}\right)\dot I_i\nonumber\\
&=&c_0\mu_h S_h^\ast\left(2-\frac{S_h}{S_h^\ast}-\frac{S_h^\ast}{S_h}\right)+c_0a\beta_{vh}S_{h}^*\,\dfrac{I_{v}^*}{N_{h}}\left(2 -\frac{S_h^\ast}{S_h}-\frac{S_h}{S_h^\ast} \frac{I_v}{I_v^\ast} \frac{E_h^\ast}{E_{h}}\right)+c_0p\pi\left(2-\frac{E_h^\ast}{E_{h}}-\frac{E_{h}}{E_h^*} \right)\nonumber\\
&&-c_0\left( \,\beta_{vh}\,S_{h}^*\dfrac{I_{v}^*}{N_{h}} \right)\frac{E_{h}}{E_h^*}+c_0a\,\beta_{vh}S_h^\ast\,\dfrac{I_{v}}{N_{h}}+\sum_{i=1}^nc_i\left(1-\frac{I_i^\ast}{I_i}\right)\dot I_i \end{aligned}$$ Using the endemic relation $\displaystyle\alpha_1I_1^*=p_1\pi_h+\nu_hE_h^*+\sum_{j=1}^nm_{1j}I_{j}^*$, and the relationship between $c_0$ and $c_1$, Equation (\[Lyapc1\]) yields to $$\begin{aligned}
\label{Lyapc111}
\dot{\mathcal V}_h
&=&c_0\mu_h S_h^\ast\left(2-\frac{S_h}{S_h^\ast}-\frac{S_h^\ast}{S_h}\right)+c_0a\beta_{vh}S_{h}^*\,\dfrac{I_{v}^*}{N_{h}}\left(3 -\frac{S_h^\ast}{S_h}-\frac{S_h}{S_h^\ast} \frac{I_v}{I_v^\ast} \frac{E_h^\ast}{E_{h}} -\frac{E_h}{E_h^*}\frac{I_1^\ast}{I_1}\right)+c_0p\pi\left(2-\frac{E_h^\ast}{E_{h}}-\frac{E_{h}}{E_h^*} \right)\nonumber\\
&&+c_0a\,\beta_{vh}S_h^\ast\,\dfrac{I_{v}}{N_{h}}+ c_1p_1\pi_h\left(2-\frac{I_1^\ast}{I_1} -\frac{I_1}{I_1^\ast}\right) -c_1\left(\nu_hE_h^*+\sum_{j=1}^nm_{1j}I_{j}^*\right)\frac{I_1}{I_1^*}+c_1\sum_{j=1}^nm_{1j}I_{j} \nonumber\\
&& +c_1\sum_{j=1}^n m_{1j}I_{j}^*
-c_1\frac{I_1^\ast}{I_1}(\sum_{j=1}^nm_{1j}I_{j}) +\sum_{i=2}^nc_i\left(1-\frac{I_i^\ast}{I_i}\right)\dot I_i
\end{aligned}$$ Noting that, from the endemic relations (\[ModelAD2Endemic\]), we have $\alpha_{i}I_i^\ast=p_i\pi_h+ \sum_{j=1}^nm_{ij}I_j^\ast$, and thus, the last term of Equation (\[Lyapc111\]) leads to $$\begin{aligned}
\label{ciIi2}
c_i\left(1-\frac{I_i^\ast}{I_i}\right)\dot I_i
& =&c_ip_i\pi_h\left(2-\frac{I_i^\ast}{I_i} -\frac{I_i}{I_i^*}\right)+ c_i\sum_{j=1}^nm_{ij}I_j^\ast-c_i\left(\sum_{j=1}^nm_{ij}I_j^\ast\right)\frac{I_i}{I_i^*}\nonumber\\
&&+ c_i\sum_{j=1}^n m_{ij}I_j -c_i\frac{I_i^\ast}{I_i}\left(\sum_{j=1}^n m_{ij}I_j \right)
\end{aligned}$$
Moreover, we can check that the derivative of $\mathcal V_v$ along the trajectories of System (\[ModelAD2Compact\]) is: $$\label{VvDot1}
\dot{\mathcal V_v}=c_{v,1}\left( \mathcal A_v+\sum_{i=1}^naS_v^\ast\frac{\beta_i I_i^\ast}{N_i}\left(3 -\frac{S_v^\ast}{S_v}-\frac{S_v}{S_v^\ast} \frac{I_i}{I_i^\ast} \frac{E_v^\ast}{E_{v}}-\frac{E_vI_v^\ast}{E_v^\ast I_v} \right)+aS_v^\ast\sum_{i=1}^n \frac{\beta_i I_i}{N_h}-aS_v^\ast\sum_{i=1}^n \frac{\beta_i I_i^\ast}{N_h}\frac{I_v}{I_v^\ast}\right),$$ where $\displaystyle\mathcal A_v=(\mu_v+\delta_v)S_v^*\left(2-\dfrac{S_v^*}{S_v}-\dfrac{S_v}{S_v^*}\right).$ Combining equations (\[Lyapc111\]), (\[ciIi2\]), and (\[VvDot1\]), we obtain: $$\begin{aligned}
\label{CombinedLyap}
\dot{\mathcal V}
&=&c_0\mu_h S_h^\ast\left(2-\frac{S_h}{S_h^\ast}-\frac{S_h^\ast}{S_h}\right)+c_0a\beta_{vh}S_{h}^*\,\dfrac{I_{v}^*}{N_{h}}\left(3 -\frac{S_h^\ast}{S_h}-\frac{S_h}{S_h^\ast} \frac{I_v}{I_v^\ast} \frac{E_h^\ast}{E_{h}} -\frac{E_h}{E_h^*}\frac{I_1^\ast}{I_1}\right)+c_0p\pi\left(2-\frac{E_h^\ast}{E_{h}}-\frac{E_{h}}{E_h^*} \right)\nonumber\\
&&+c_0a\,\beta_{vh}S_h^\ast\,\dfrac{I_{v}}{N_{h}}+\sum_{i=1}^nc_ip_i\pi_h\left(2-\frac{I_i^\ast}{I_i} -\frac{I_i}{I_i^*}\right) - c_1\left(\nu_hE_h^*+\sum_{j=1}^nm_{1j}I_{j}^*\right)\frac{I_1}{I_1^*}+c_1\sum_{j=1}^nm_{1j}I_{j} \nonumber\\
&& +c_1\sum_{j=1}^n m_{1j}I_{j}^*
-c_1\frac{I_1^\ast}{I_1}\left(\sum_{j=1}^nm_{1j}I_{j}\right) \nonumber\\
&&+\sum_{i=2}^n\left[
c_i\sum_{j=1}^nm_{ij}I_j^\ast-c_i\left(\sum_{j=1}^nm_{ij}I_j^\ast\right)\frac{I_i}{I_i^*}
+ c_i\sum_{j=1}^n m_{ij}I_j -c_i\frac{I_i^\ast}{I_i}\left(\sum_{j=1}^n m_{ij}I_j \right)
\right]\nonumber\\
&&+c_{v,1}\left( \mathcal A_v+\sum_{i=1}^naS_v^\ast\frac{\beta_i I_i^\ast}{N_i}\left(3 -\frac{S_v^\ast}{S_v}-\frac{S_v}{S_v^\ast} \frac{I_i}{I_i^\ast} \frac{E_v^\ast}{E_{v}}-\frac{E_vI_v^\ast}{E_v^\ast I_v} \right)+aS_v^\ast\sum_{i=1}^n \frac{\beta_i I_i}{N_h}-aS_v^\ast\sum_{i=1}^n \frac{\beta_i I_i^\ast}{N_h}\frac{I_v}{I_v^\ast}\right)
\end{aligned}$$ Given the relationship (\[c0cvc1\]), the linear terms in $I_v$ in Equation (\[CombinedLyap\]) cancel. Furthermore, by substituting $\mathcal A_v$ by its expression and $c_{v,1}$ by their expressions, Equation (\[CombinedLyap\]) leads to $$\begin{aligned}
\label{CombinedLyap3}
\dot{\mathcal V}
&=&c_0\mu_h S_h^\ast\left(2-\frac{S_h}{S_h^\ast}-\frac{S_h^\ast}{S_h}\right)+c_{v,1}(\mu_v+\delta_v)S_v^*\left(2-\dfrac{S_v^*}{S_v}-\dfrac{S_v}{S_v^*}\right)\nonumber\\
&&+c_0p\pi\left(2-\frac{E_h^\ast}{E_{h}}-\frac{E_{h}}{E_h^*} \right)+\sum_{i=1}^nc_ip_i\pi_h\left(2-\frac{I_i^\ast}{I_i} -\frac{I_i}{I_i^*}\right) \nonumber\\
&&+c_{v,1}\sum_{i=1}^naS_v^\ast\frac{\beta_i I_i^\ast}{N_i}\left(6 -\frac{S_h^\ast}{S_h}-\frac{S_h}{S_h^\ast} \frac{I_v}{I_v^\ast} \frac{E_h^\ast}{E_{h}} -\frac{E_h}{E_h^*}\frac{I_1^\ast}{I_1} -\frac{S_v^\ast}{S_v}-\frac{S_v}{S_v^\ast} \frac{I_i}{I_i^\ast} \frac{E_v^\ast}{E_{v}}-\frac{E_vI_v^\ast}{E_v^\ast I_v} \right)\nonumber\\
&& - c_1\left(\nu_hE_h^*+\sum_{j=1}^nm_{1j}I_{j}^*\right)\frac{I_1}{I_1^*}+c_1\sum_{j=1}^nm_{1j}I_{j} +c_1\sum_{j=1}^n m_{1j}I_{j}^*
-c_1\frac{I_1^\ast}{I_1}\left(\sum_{j=1}^nm_{1j}I_{j}\right) \nonumber\\
&&+\sum_{i=2}^n\left[
c_i\sum_{j=1}^nm_{ij}I_j^\ast-c_i\left(\sum_{j=1}^nm_{ij}I_j^\ast\right)\frac{I_i}{I_i^*}
+ c_i\sum_{j=1}^n m_{ij}I_j -c_i\frac{I_i^\ast}{I_i}\left(\sum_{j=1}^n m_{ij}I_j \right)
\right]\nonumber\\
&&+c_1\frac{\nu_hE_h^\ast}{\sum_{l=1}^n \beta_l I_l^\ast}\left(\sum_{i=1}^n \beta_i I_i\right)
\end{aligned}$$ We choose the vector $c=(c_1,c_2,\dots,c_n)^T$ to be the solution of the linear system $Bc=0$, where $$\label{B}
B=\begin{pmatrix}
-\diamond_{11} & m_{21}I_1^\ast & m_{31}I_{1}^\ast & \dots & \dots & m_{n1}I_{1}^\ast\\
\nu_hE_h^*\frac{\beta_2 I_2^\ast}{\sum_{i=1}^n \beta_i I_i^\ast}+ m_{12}I_{2}^\ast & -\diamond_{22} & m_{32}I_2^\ast & \dots &\dots & m_{n2}I_{2}^\ast\\
\nu_hE_h^*\frac{\beta_3 I_3^\ast}{\sum_{i=1}^n \beta_i I_i^\ast} + m_{13}I_{3}^\ast & m_{23}I_{3}^\ast & -\diamond_{33} & \dots& \dots& m_{n3}I_{n}^\ast\\
\vdots& \vdots & \vdots& \ddots & \ddots& \vdots \\
\nu_hE_h^* \frac{\beta_n I_n^\ast}{\sum_{i=1}^n \beta_i I_i^\ast}+m_{1n}I_{n}^\ast & m_{2n}I_{n}^\ast& m_{3n}I_{n}^\ast & \hdots&\hdots &-\diamond_{nn}
\end{pmatrix}$$ where $$\diamond_{11}=\left(\nu_hE_h^*\frac{\sum_{i=2}^n \beta_i I_i^\ast }{\sum_{i=1}^n \beta_i I_i^\ast }+\sum_{j=1}^nm_{1j}I_{j}^*\right),\quad\textrm{
and for } k\geq 2,\quad
\diamond_{kk}=\sum_{j=1}^{n}m_{kj}I_j^\ast.$$ The matrix $B$ is irreducible. Indeed, since $\mathbf{I}_h^*\gg0$, we notice that all elements of the second upper diagonal of $B$ are all non zero, as $m_{i+1,i}=\gamma_{i,i+1}$, and thus represent the incremental transition between infectious classes. This, along with the first column, makes the matrix $B$ irreducible. Hence, it could be shown that $\dim(\ker(B))= 1$; and by the Kirchhoff’s matrix tree theorem[@bollobas2013modern; @moon1970counting], $c_i=-C_{ii}\gg0$ where $C_{ii}$ is the cofactor of the $i^{th}$ diagonal of $B$. Hence, it exists $c=(c_1,c_2,\dots,c_n)^T\gg0$ such that $Bc=0$. Moreover, this implies that, in Equation (\[CombinedLyap3\]), we have: $$\begin{aligned}
0&=& - c_1\left(\nu_hE_h^*+\sum_{j=1}^nm_{1j}I_{j}^*\right)\frac{I_1}{I_1^*}+c_1\sum_{j=1}^nm_{1j}I_{j} +c_1\sum_{j=1}^n m_{1j}I_{j}^*
\nonumber\\
&&+\sum_{i=2}^n\left[
c_i\sum_{j=1}^nm_{ij}I_j^\ast-c_i\left(\sum_{j=1}^nm_{ij}I_j^\ast\right)\frac{I_i}{I_i^*}
+ c_i\sum_{j=1}^n m_{ij}I_j
\right]\nonumber\\
&&+c_1\frac{\nu_hE_h^\ast}{\sum_{l=1}^n \beta_l I_l^\ast}\left(\sum_{i=1}^n \beta_i I_i\right)
\end{aligned}$$ Thus, (\[CombinedLyap3\]) yields to:
$$\begin{aligned}
\label{CombinedLyap4}
\dot{\mathcal V}
&= c_0\mu_h S_h^\ast\left(2-\frac{S_h}{S_h^\ast}-\frac{S_h^\ast}{S_h}\right)+c_{v,1}(\mu_v+\delta_v)S_v^*\left(2-\dfrac{S_v^*}{S_v}-\dfrac{S_v}{S_v^*}\right)\nonumber\\
& +c_0p\pi\left(2-\frac{E_h^\ast}{E_{h}}-\frac{E_{h}}{E_h^*} \right)+\sum_{i=1}^nc_ip_i\pi_h\left(2-\frac{I_i^\ast}{I_i} -\frac{I_i}{I_i^*}\right) \nonumber\\
& +c_{v,1}\sum_{i=1}^naS_v^\ast\frac{\beta_i I_i^\ast}{N_i}\underbrace{\left(6 -\frac{S_h^\ast}{S_h}-\frac{S_h}{S_h^\ast} \frac{I_v}{I_v^\ast} \frac{E_h^\ast}{E_{h}} -\frac{E_h}{E_h^*}\frac{I_1^\ast}{I_1} -\frac{S_v^\ast}{S_v}-\frac{S_v}{S_v^\ast} \frac{I_i}{I_i^\ast} \frac{E_v^\ast}{E_{v}}-\frac{E_vI_v^\ast}{E_v^\ast I_v} \right)}_{\mathcal W_i}\nonumber\\
& +\sum_{i=1}^n\sum_{j=1}^nc_im_{ij}I_j^\ast \left(
1
-\frac{I_i^\ast}{I_i} \frac{I_j^\ast}{I_j}
\right).
\end{aligned}$$
The first three terms of (\[CombinedLyap4\]) are definite-positive. Now, we will break down the last two terms in (\[CombinedLyap4\]) into definite-negative terms. Indeed, following [@guo2012global], we transform each theses expressions as sums of terms in the form of $f(x)=1-x+\ln x$. To this end, we will use the fact that the function $f(x)$ is definite negative around $x^\ast=1$. Indeed, using the properties of natural logarithm function, the expression of $\mathcal W_i$ in (\[CombinedLyap4\]) could be written as: $$\begin{aligned}
\mathcal W_i &= 6 -\frac{S_h^\ast}{S_h}-\frac{S_h}{S_h^\ast} \frac{I_v}{I_v^\ast} \frac{E_h^\ast}{E_{h}} -\frac{E_h}{E_h^*}\frac{I_1^\ast}{I_1} -\frac{S_v^\ast}{S_v}-\frac{S_v}{S_v^\ast} \frac{I_i}{I_i^\ast} \frac{E_v^\ast}{E_{v}}-\frac{E_vI_v^\ast}{E_v^\ast I_v}\nonumber\\
&= \left(1-\frac{S_h^\ast}{S_h}+\ln\frac{S_h^\ast}{S_h}\right)+\left(1-\frac{S_h}{S_h^\ast} \frac{I_v}{I_v^\ast} \frac{E_h^\ast}{E_{h}}+\ln\frac{S_hI_vE_h^\ast}{S_h^\ast I_v^\ast E_{h}} \right)+\left(1-\frac{E_vI_v^\ast}{E_v^\ast I_v} +\ln\frac{E_vI_v^\ast}{E_v^\ast I_v} \right)\nonumber\\
& + \left(1 -\frac{E_h}{E_h^\ast}\frac{I_1^\ast}{I_1} +\ln\frac{E_hI_1^\ast}{E_h^\ast I_1} \right)
+\left(1-\frac{S_vI_i E_v^\ast}{S_v^\ast I_i^\ast E_{v}} +\ln\frac{S_vI_i E_v^\ast}{S_v^\ast I_i^\ast E_{v}} \right)
+\left(1
-\frac{S_v^\ast}{S_v}+\ln\frac{S_v^\ast }{S_v}\right) +\ln\frac{ I_1}{I_1^\ast}\frac{ I_i^\ast }{I_i}
\end{aligned}$$ Noting that $$1- \frac{I_i^\ast}{I_i} \frac{I_j}{I_j^\ast}=1- \frac{I_i^\ast}{I_i} \frac{I_j}{I_j^\ast}+\ln\frac{I_i^\ast I_j}{I_iI_j^\ast} +\ln\frac{I_iI_j^\ast}{I_i^\ast I_j},$$ and substitute the expression of $\mathcal W_i$, becomes $$\begin{aligned}
\label{CombinedLyapGene2}
\dot {\mathcal V}
&= c_0\mu_h S_h^\ast\left(2-\frac{S_h}{S_h^\ast}-\frac{S_h^\ast}{S_h}\right)+c_{v,1}(\mu_v+\delta_v)S_v^*\left(2-\dfrac{S_v^*}{S_v}-\dfrac{S_v}{S_v^*}\right) \nonumber\\
&+c_{v,1}aS_v^\ast\sum_{i=1}^n \frac{\beta_i I_i^\ast}{N_h}\left[
\left(1-\frac{S_h^\ast}{S_h}+\ln\frac{S_h^\ast}{S_h}\right)+\left(1-\frac{S_h}{S_h^\ast} \frac{I_v}{I_v^\ast} \frac{E_h^\ast}{E_{h}}+\ln\frac{S_hI_vE_h^\ast}{S_h^\ast I_v^\ast E_{h}} \right)\right.\nonumber\\
& \left.+\left(1-\frac{E_vI_v^\ast}{E_v^\ast I_v} +\ln\frac{E_vI_v^\ast}{E_v^\ast I_v} \right)
+ \left(1 -\frac{E_h}{E_h^\ast}\frac{I_1^\ast}{I_1} +\ln\frac{E_hI_1^\ast}{E_h^\ast I_1} \right)
+\left(1-\frac{S_vI_i E_v^\ast}{S_v^\ast I_i^\ast E_{v}} +\ln\frac{S_vI_i E_v^\ast}{S_v^\ast I_i^\ast E_{v}} \right)
\right.\nonumber\\
& \left.+\left(1
-\frac{S_v^\ast}{S_v}+\ln\frac{S_v^\ast }{S_v}\right)\right]+c_0p\pi\left(2-\frac{E_h^\ast}{E_{h}}-\frac{E_{h}}{E_h^*} \right)
+\sum_{i=1}^nc_ip_i\pi_h\left(2-\frac{I_i^\ast}{I_i}-\frac{I_i}{I_i^\ast}\right) \nonumber\\
& \sum_{i=1}^nc_i \sum_{j=1}^{n}m_{ij}I_j^\ast\left(1- \frac{I_i^\ast}{I_i} \frac{I_j}{I_j^\ast}+\ln\frac{I_i^\ast I_j}{I_iI_j^\ast}\right)\nonumber\\
&+\underbrace{c_{v,1}aS_v^\ast\sum_{i=1}^n \frac{\beta_i I_i^\ast}{N_h} \ln\frac{ I_1}{I_1^\ast}\frac{ I_i^\ast }{I_i}
+\sum_{i=1}^nc_i \sum_{j=1}^{n}m_{ij}I_j^\ast \ln\frac{I_iI_j^\ast}{I_i^\ast I_j}}_{\mathcal S}
\end{aligned}$$
All but the last two sums in (\[CombinedLyapGene2\]) are definite negative. Let us denote by $\mathcal S$ the sum of these two sums. We focus on proving that $\mathcal S:=0$. Indeed, recall the expression of $c_{v,1}$ in terms of $c_1$, given in (\[c0cvc1\]): $$c_{v,1}\frac{aS_v^\ast}{N_h}=c_1\frac{\nu_hE_h^*}{\sum_{l=1}^n \beta_l I_l^\ast }.$$ By replacing $c_{v,1}$ by its value in $\mathcal S$, we obtain,
$$\begin{aligned}
\label{S}
\mathcal S &= c_{v,1}aS_v^\ast\sum_{i=1}^n \frac{\beta_i I_i^\ast}{N_h} \ln\frac{ I_1}{I_1^\ast}\frac{ I_i^\ast }{I_i}
+\sum_{i=1}^nc_i \sum_{j=1}^{n}m_{ij}I_j^\ast \ln\frac{I_iI_j^\ast}{I_i^\ast I_j}\nonumber\\
& =c_1\frac{\nu_hE_h^*}{\sum_{l=1}^n \beta_l I_l^\ast }\sum_{i=1}^n \beta_i I_i^\ast \ln\frac{ I_1}{I_1^\ast}\frac{ I_i^\ast }{I_i}
+\sum_{i=1}^nc_i \sum_{j=1}^{n}m_{ij}I_j^\ast \ln\frac{I_iI_j^\ast}{I_i^\ast I_j}\nonumber\\
& =c_1\sum_{j=1}^{n}\ln\frac{I_1I_j^\ast}{I_1^\ast I_j}\left[\frac{\nu_hE_h^*}{\sum_{l=1}^n \beta_l I_l^\ast } \beta_j I_j^\ast
+ m_{1j}I_j^\ast \right]
+\sum_{i=2}^nc_i \sum_{j=1}^{n}m_{ij}I_j^\ast \ln\frac{I_iI_j^\ast}{I_i^\ast I_j}
\end{aligned}$$
However, since $c_i$ are the components of the solution of $Bc=0$ where $B$ is given in (\[B\]), it follows that, for any $j\geq2,$
$$c_1\left( \nu_hE_h^*\frac{\beta_j I_j^\ast}{\sum_{l=1}^n \beta_l I_l^\ast} +m_{1j}I_{j}^\ast\right)=
c_j\left(\sum_{k=1}^ n m_{jk}I_k^\ast \right)
-\sum_{i=2}^{n}c_im_{ij}I_j^\ast,$$ Plugging this expression into , and using again the properties of natural logarithms, we obtain:
$$\begin{aligned}
\label{S2}
\mathcal S
& =\sum_{j=1}^{n}\ln\frac{I_1I_j^\ast}{I_1^\ast I_j}\left[c_j\left(\sum_{k=1}^ n m_{jk}I_k^\ast \right)
-\sum_{i=2}^{n}c_im_{ij}I_j^\ast \right]
+\sum_{i=2}^nc_i \sum_{j=1}^{n}m_{ij}I_j^\ast \ln\frac{I_iI_j^\ast}{I_i^\ast I_j}\nonumber\\
&= \sum_{i=1}^{n}c_i\ln\frac{I_1I_i^\ast}{I_1^\ast I_i}\left(\sum_{k=1}^ n m_{ik}I_k^\ast \right)+\sum_{i=2}^nc_i \sum_{j=1}^{n}m_{ij}I_j^\ast \left[ -\ln\frac{ I_1 I_j^\ast}{I_1^\ast I_j}+\ln\frac{I_i I_j^\ast} {I_i^\ast I_j}\right]\nonumber\\
&= \sum_{i=2}^{n}c_i\ln\frac{I_1I_i^\ast}{I_1^\ast I_i}\left(\sum_{j=1}^ n m_{ij}I_j^\ast \right)+\sum_{i=2}^nc_i \sum_{j=1}^{n}m_{ij}I_j^\ast \left[ \ln\frac{I_1^\ast I_i}{ I_1 I_i^\ast}\right]\nonumber\\
&:=0, \end{aligned}$$
since for $i=1$, the coefficient of the sum is $\ln1=0$. Finally using and , the derivative of $\mathcal V$ along the trajectories of is $$\begin{aligned}
\label{CombinedLyapGeneFinal}
\dot {\mathcal V}
&= c_0\mu_h S_h^\ast\left(2-\frac{S_h}{S_h^\ast}-\frac{S_h^\ast}{S_h}\right)+c_{v,1}(\mu_v+\delta_v)S_v^*\left(2-\dfrac{S_v^*}{S_v}-\dfrac{S_v}{S_v^*}\right) \nonumber\\
&+c_{v,1}aS_v^\ast\sum_{i=1}^n \frac{\beta_i I_i^\ast}{N_h}\left[
\left(1-\frac{S_h^\ast}{S_h}+\ln\frac{S_h^\ast}{S_h}\right)+\left(1-\frac{S_h}{S_h^\ast} \frac{I_v}{I_v^\ast} \frac{E_h^\ast}{E_{h}}+\ln\frac{S_hI_vE_h^\ast}{S_h^\ast I_v^\ast E_{h}} \right)\right.\nonumber\\
& \left.+\left(1-\frac{E_vI_v^\ast}{E_v^\ast I_v} +\ln\frac{E_vI_v^\ast}{E_v^\ast I_v} \right)
+ \left(1 -\frac{E_h}{E_h^\ast}\frac{I_1^\ast}{I_1} +\ln\frac{E_hI_1^\ast}{E_h^\ast I_1} \right)
+\left(1-\frac{S_vI_i E_v^\ast}{S_v^\ast I_i^\ast E_{v}} +\ln\frac{S_vI_i E_v^\ast}{S_v^\ast I_i^\ast E_{v}} \right)
\right.\nonumber\\
& \left.+\left(1
-\frac{S_v^\ast}{S_v}+\ln\frac{S_v^\ast }{S_v}\right)\right]+c_0p\pi\left(2-\frac{E_h^\ast}{E_{h}}-\frac{E_{h}}{E_h^*} \right)
+\sum_{i=1}^nc_ip_i\pi_h\left(2-\frac{I_i^\ast}{I_i}-\frac{I_i}{I_i^\ast}\right) \nonumber\\
& \sum_{i=1}^nc_i \sum_{j=1}^{n}m_{ij}I_j^\ast\left(1- \frac{I_i^\ast}{I_i} \frac{I_j}{I_j^\ast}+\ln\frac{I_i^\ast I_j}{I_iI_j^\ast}\right),
\end{aligned}$$ which is definite-negative. Therefore, by Lyapunov’s stability theorem, the unique endemic equilibrium is GAS.
\[WEEGAS\] Let $({S}_h^\diamond, 0,\mathbf{I}_h^\diamond,{S}_v^0,0,0)$ be a weakly endemic equilibrium of Model (\[ModelAD2Compact\]). This equilibrium is GAS.
Let $({S}_h^\diamond, 0,\mathbf{I}_h^\diamond,{S}_v^0,0,0)$ be a weakly endemic equilibrium of Model (\[ModelAD2Compact\]).\
$\bullet$ If $\Beta=0$, we remark from the vector’s equations in Model (\[ModelAD2Compact\]) that $S_v\to S_v^0:=\frac{\pi_v}{\mu_v+\delta_v}$, $E_v\to0$ and $I_v\to0$ as $t\to\infty$. So, by the theory of asymptotically autonomous systems for triangular systems [@CasThi95; @0478.93044], Model (\[ModelAD2Compact\]) is equivalent to $$\left\{\begin{array}{llll}\label{WeakEE}
\displaystyle\dot S_{h}=\pi_h\left( 1-p-\sum_{i=1}^{n+1}p_i \right)-\mu_h S_h\\
\displaystyle\dot E_{h}=p\pi_h-(\mu_h+\nu_h+\eta)E_h\\
\displaystyle\dot{\mathbf{I}}_h=\pi_h\mathbf{p}+\nu_hE_he_1-(\textrm{diag}(\alpha)-M)\mathbf{I}_h
\end{array}\right.$$ System (\[WeakEE\]) is triangular and linear, and its solutions converge toward $(\bar S_h,\bar E_h,\bar{\mathbf{I}}_h)$, where $\bar S_h:=\frac{\Lambda_h}{\mu_h}$, $\bar E_h:=\frac{p\pi_h}{\mu_h+\nu_h+\eta}$ and $\bar{\mathbf{I}}_h:=(\textrm{diag}(\alpha)-M)^{-1}\left( \pi_h\mathbf{p}+\frac{\nu_hp\pi_h}{\mu_h+\nu_h+\eta}e_1 \right)$. Thus, it follows that the weak endemic equilibrium $(\bar S_h,\bar E_h,\bar{\mathbf{I}}_h,S_v^0,0,0)$ of Model (\[ModelAD2Compact\]) is GAS.\
Before we start the proof of the next case, let us define the order relation for the vectors as follows: $u\leq v$ if $u_i\leq v_i$, for all $i$, where $u_i$ and $v_i$ are components of $u$ and $v$ respectively. Similarly, $u< v$ if $u\leq v$ and $u\neq v$. Also $u\gg v$ if $u_i>v_i$, for all $i$.\
$\bullet$ If of is satisfied with $\mathcal N_0^2(p,\mathbf{p},p_{n+1})\leq1$. That is, $\Beta\neq0$, $p=0$ and $\mathbf{p}$ and $M$ are such that $\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, (\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}\bigr\rangle=0$. These imply that, using the endemic relations, $$(\textrm{diag}(\alpha)-M)^{-1} \mathbf{p} >0.$$ Moreover, it follows that $\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, \mathbf{I}_h^\diamond
\bigr\rangle=0$. This implies that it exists a subset $\mathcal J$ of $\{1,2,\dots,n\}$ such that $I^\diamond_i=0$, for all $i\in\mathcal J$, $I_i^\diamond>0$ for $i\in\{1,2,\dots,n\}\setminus\mathcal J$; and $\beta_i^\diamond>0$ for $i\in\{1,2,\dots,n\}\setminus\mathcal J$, and $\beta_i^\diamond=0$ for $i\in\mathcal J$. WLOG, suppose that $\mathcal J=\{1,2,\dots,s-1\}$ with $s\geq2$. Hence, the endemic relation $\mathbf{I}_h^\diamond=\pi_h(\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}$ and the condition $\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, (\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}\bigr\rangle=0$ imply that $M$ has the form $$M=\begin{pmatrix}
M_{11} & \textbf{0}_{s-1,n-s+1}\\
M_{21} & M_{22}
\end{pmatrix},$$ where $M_{11}\in\mathcal M_{s-1,s-1},\;\; M_{21}\in\mathcal M_{n-s+1,s-1},\;\; M_{22}\in\mathcal M_{n-s+1,n-s+1}$. Similarly, $p_i=0$ for all $i\in\mathcal J$ and $\mathbf{I}_h^\diamond=(0,\dots,0, I_s^\diamond,\dots,I_n^\diamond )$ where $I_i^\diamond>0$ for $s\leq i\leq n$.\
Let $\mathbf{c}=(\mathbf{c}_1,\mathbf{c}_2)^T$ where $\mathbf{c}_1=(c_1,\dots,c_{s-1})^T$ and $\mathbf{c}_2=(c_s,\dots,c_{n})^T$. The vector $\mathbf{c}_2$ is the solution of $\tilde B\mathbf{c}_2=0$ where $$\tilde B=\begin{pmatrix}
-\tilde{b}_{s,s} & m_{s+1,s}I_s^\diamond & m_{s+2,s}I_{s}^\diamond & \dots & \dots & m_{n,s}I_{s}^\diamond\\
m_{s,s+1}I_{s+1}^\diamond & -\tilde{b}_{s+1,s+1} & m_{s+2,s+1}I_{s+1}^\diamond & \dots &\dots & m_{n,s+1}I_{s+1}^\diamond\\
m_{s,s+2}I_{s+2}^\diamond & m_{s+1,s+2}I_{s+2}^\diamond & -\tilde{b}_{33} & \dots& \dots& m_{n,s+2}I_{s+2}^\diamond\\
\vdots& \vdots & \vdots& \ddots & \ddots& \vdots \\
m_{s,n}I_{n}^\diamond & m_{s+1,n}I_{n}^\diamond & m_{s+2,n}I_{n}^\diamond & \hdots&\hdots &-\tilde{b}_{nn}
\end{pmatrix},$$ where $\tilde{b}_{kk}=\sum_{j=s}^nm_{kj}I_j^\diamond$ for $s\leq k\leq n$. Since $M_{22}$ is irreducible and $I^\diamond_i>0$ for all $s\leq i\leq n$, the matrix $\tilde B$ is irreducible. Moreover, $\tilde B$ is the Laplacian matrix of the graph interconnecting the stages $I_i$ for $s\leq i\leq n$. Hence, as previously stated, Kirchhoff’s matrix tree theorem affirms that the solution of $\tilde B\textbf{c}_2=0$ is such that $c_i=-C_{ii}\gg0$, where $C_{ii}$ is the cofactor of $i^{th}$ diagonal element of $\tilde B$. Hence $\textbf{c}_2\gg0$.
Let $\mathbf{I}_h^\diamond=(\mathbf{I}_1^\diamond,\mathbf{I}_2^\diamond)^T$ and consider the Lyapunov function candidate $ \mathcal V= \mathcal V_h+ \mathcal V_v$, where $$\mathcal V_h=c_1 \frac{\nu_h}{\alpha_h} E_h+\bigl\langle\textbf{c}_1 \, \mbox{\Large $\mid$} \, \mathbf{I}_1 \bigr\rangle
+\sum_{i=s}^nc_{i}\int_{I_{i}^\diamond}^{I_{i}}\left(1-\frac{I_{i}^\diamond}{x}\right)dx,\quad\textrm{and}\quad \mathcal V_v=c_vE_v+c_v\frac{\alpha_v}{\nu_v}I_v,$$ where $c_v=c_1\frac{\nu_h}{\alpha_h}\frac{a\beta_{vh}\Lambda_h}{\mu_hN_h}\frac{\nu_v}{(\mu_v+\delta_v)\alpha_v}$, $\textbf{c}_1$ is positive vector to be determined later. The derivative of $\mathcal V$ along the trajectories of System (\[ModelAD2Compact\]) is: $$\begin{aligned}
\label{LyapBoundary}
\dot {\mathcal V_h} &=&c_1 \frac{\nu_h}{\alpha_h}\dot{E}_h+\bigl\langle\textbf{c}_1 \, \mbox{\Large $\mid$} \, \dot{\mathbf{I}}_1^\diamond \bigr\rangle
+\sum_{i=s}^nc_i\left(1-\frac{I_i^\diamond}{I_i}\right)\dot I_i+c_v\dot E_v+c_v\frac{\alpha_v}{\nu_v}\dot I_v\nonumber\\
&=&c_1 a\,\beta_{vh}\frac{\nu_h}{\alpha_h}\dfrac{S_{h}I_{v}}{N_{h}}+\bigl\langle\textbf{c}_1 \, \mbox{\Large $\mid$} \, (-\textrm{diag}(\tilde\alpha)+M_{11})\mathbf{I}_1 \bigr\rangle+\sum_{i=s}^nc_i\left(1-\frac{I_i^\diamond}{I_i}\right)\dot I_i, \end{aligned}$$ where $\tilde\alpha=(\alpha_1,\dots,\alpha_{s-1})$. Moreover, as in the proof of , using the fact, for that, for $1\leq i\leq n$, the $c_i$ are the components of the solution of $\tilde B\textbf{c}_2=0$ and $$\sum_{i=s}^nc_i\sum_{j=s}^nm_{ij}I_j^\diamond\left(1 - \frac{I_i^\diamond}{I_i}\frac{I_j}{I_j^\diamond}\right)=
\sum_{i=s}^nc_i\sum_{j=s}^nm_{ij}I_j^\diamond\left(1 - \frac{I_i^\diamond}{I_i}\frac{I_j}{I_j^\diamond}+ \ln\frac{I_i^\diamond}{I_i}\frac{I_j}{I_j^\diamond}\right),$$ it could be shown that Equation (\[LyapBoundary\]) implies that
$$\begin{aligned}
\label{ciM222}\sum_{i=s}^nc_i\left(1-\frac{I_i^\ast}{I_i}\right)\dot I_i
&=&\sum_{i=s}^nc_ip_i\pi_h\left(2-\frac{I_i^\diamond}{I_i} -\frac{I_i}{I_i^\diamond}\right)+\sum_{i=s}^nc_i\sum_{j=s}^nm_{ij}I_j^\diamond\left(1 - \frac{I_i^\diamond}{I_i}\frac{I_j}{I_j^\diamond}+\ln\frac{I_i^\diamond}{I_i}\frac{I_j}{I_j^\diamond}\right)\nonumber\\
& &- \sum_{i=s}^{n}c_i \frac{I_i^\diamond}{I_i}\sum_{i=1}^{s-1}m_{ij}I_j
+\sum_{i=s}^nc_i \sum_{j=1}^{s-1}m_{ij}I_j \nonumber\\
&:=&\sum_{i=s}^nc_ip_i\pi_h\left(2-\frac{I_i^\ast}{I_i} -\frac{I_i}{I_i^*}\right)+\sum_{i=s}^nc_i\sum_{j=s}^nm_{ij}I_j^\diamond\left(1 - \frac{I_i^\diamond}{I_i}\frac{I_j}{I_j^\diamond}+\ln\frac{I_i^\diamond}{I_i}\frac{I_j}{I_j^\diamond}\right)\nonumber\\
& &- \sum_{i=s}^{n}c_i \frac{I_i^\diamond}{I_i}\sum_{i=1}^{s-1}m_{ij}I_j +\textbf{c}_2^TM_{21}\mathbf{I}_1.
\end{aligned}$$
We choose $\mathbf{c}_1$ to be the solution of $(-\textrm{diag}(\tilde\alpha)+M_{11}^T+\bar\Beta \tilde{e}_1^T)\mathbf{c}_1=-M_{21}^T\mathbf{c}_2$, where $\bar\Beta=\frac{a^2\beta_{vh}\Lambda_h}{\mu_hN_h}\frac{\nu_v\nu_h}{(\mu_v+\delta_v)\alpha_v\alpha_h}\frac{N_v}{N_h}\tilde\Beta$, with $\tilde\Beta=(\beta_1,\dots,\beta_{s-1})$ and $\tilde{e}_1$ the fist canonical vector of $\R^{s-1}$. This solution exists and $\mathbf{c}_1\geq0$ since $\mathbf{c}_2\gg0$ and $-(-\textrm{diag}(\tilde\alpha)+M_{11}^T+\bar\Beta \tilde{e}_1^T)^{-1}\geq0$ as $-\textrm{diag}(\tilde\alpha)+M_{11}^T+\bar\Beta \tilde{e}_1^T$ is a Metzler invertible matrix.\
Hence, Equations (\[LyapBoundary\]) and (\[ciM222\]) leads to $$\begin{aligned}
\label{LyapBoundary2}
\dot {\mathcal V_h}
&=& c_1 a\,\beta_{vh}\frac{\nu_h}{\alpha_h}\dfrac{S_{h}I_{v}}{N_{h}}+\bigl\langle\textbf{c}_1 \, \mbox{\Large $\mid$} \, (-\textrm{diag}(\tilde\alpha)+M_{11})\mathbf{I}_1 \bigr\rangle +\sum_{i=s}^nc_ip_i\pi_h\left(2-\frac{I_i^\diamond}{I_i} -\frac{I_i}{I_i^\diamond}\right)\nonumber\\
& &+\sum_{i=s}^nc_i\sum_{j=s}^nm_{ij}I_j^\diamond\left(1 - \frac{I_i^\diamond}{I_i}\frac{I_j}{I_j^\diamond}+\ln\frac{I_i^\diamond}{I_i}\frac{I_j}{I_j^\diamond}\right)
- \sum_{i=s}^{n}c_i \frac{I_i^\diamond}{I_i}\sum_{i=1}^{s-1}m_{ij}I_j +\textbf{c}_2^TM_{21}\mathbf{I}_1
\end{aligned}$$ However, $$\begin{aligned}
\bigl\langle\textbf{c}_1 \, \mbox{\Large $\mid$} \, (-\textrm{diag}(\tilde\alpha)+M_{11})\mathbf{I}_1 \bigr\rangle+\textbf{c}_2^TM_{21}\mathbf{I}_1&=&\bigl\langle(-\textrm{diag}(\tilde\alpha)+M_{11}^T)\textbf{c}_1 +M_{21}^T \textbf{c}_2 \, \mbox{\Large $\mid$} \, \mathbf{I}_1 \bigr\rangle\nonumber\\
&=&-\bigl\langle\bar\Beta \tilde{e}_1^T\textbf{c}_1 \, \mbox{\Large $\mid$} \, \mathbf{I}_1 \bigr\rangle\nonumber\\
&:=&-c_1\frac{a^2\beta_{vh}\Lambda_h}{\mu_hN_h}\frac{\nu_v\nu_h}{(\mu_v+\delta_v)\alpha_v\alpha_h}\frac{N_v}{N_h}\bigl\langle\tilde\Beta \, \mbox{\Large $\mid$} \, \mathbf{I}_1 \bigr\rangle.\nonumber
\end{aligned}$$ Hence, Equation (\[LyapBoundary2\]) leads to $$\begin{aligned}
\label{LyapBoundary3}
\dot {\mathcal V_h}
&=& c_1 a\,\beta_{vh}\frac{\nu_h}{\alpha_h}\dfrac{S_{h}I_{v}}{N_{h}}+\sum_{i=s}^nc_ip_i\pi_h\left(2-\frac{I_i^\diamond}{I_i} -\frac{I_i}{I_i^\diamond}\right) -c_1\frac{a^2\beta_{vh}\Lambda_h}{\mu_hN_h}\frac{\nu_v\nu_h}{(\mu_v+\delta_v)\alpha_v\alpha_h}\frac{N_v}{N_h}\bigl\langle\tilde\Beta \, \mbox{\Large $\mid$} \, \mathbf{I}_1 \bigr\rangle\nonumber\\
& &+\sum_{i=s}^nc_i\sum_{j=s}^nm_{ij}I_j^\diamond\left(1 - \frac{I_i^\diamond}{I_i}\frac{I_j}{I_j^\diamond}+\ln\frac{I_i^\diamond}{I_i}\frac{I_j}{I_j^\diamond}\right)
- \sum_{i=s}^{n}c_i \frac{I_i^\diamond}{I_i}\sum_{i=1}^{s-1}m_{ij}I_j.
\end{aligned}$$ We can check that derivative of $\mathcal V_v$ along the trajectories of (\[ModelAD2Compact\]) $$\begin{aligned}
\label{LyapVectBoundary}
\dot {\mathcal V_v} &=&c_v\dot E_v+c_v\frac{\alpha_v}{\nu_v}\dot I_v\nonumber\\
&=&c_v\left(a\dfrac{ S_{v}}{N_h}\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, \mathbf{I}_h
\bigr\rangle-\alpha_vE_{v}\right)
+c_v\frac{\alpha_v}{\nu_v}\left(\nu_vE_v-(\mu_v+\delta_v)I_{v}\right)\nonumber\\
&=&c_1a^2\frac{\beta_{vh}\Lambda_h}{\mu_hN_h}\frac{\nu_v\nu_h}{(\mu_v+\delta_v)\alpha_v\alpha_h} \dfrac{ S_{v}}{N_h}\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, \mathbf{I}_h
\bigr\rangle
-c_1\frac{\nu_h}{\alpha_h}\frac{a\beta_{vh}\Lambda_h}{\mu_hN_h}I_{v} \nonumber\\
&=&c_1a^2\frac{\beta_{vh}\Lambda_h}{\mu_hN_h}\frac{\nu_v\nu_h}{(\mu_v+\delta_v)\alpha_v\alpha_h} \dfrac{ S_{v}}{N_h}\bigl\langle
\tilde\Beta
\, \mbox{\Large $\mid$} \, \mathbf{I}_1
\bigr\rangle
-c_1\frac{\nu_h}{\alpha_h}\frac{a\beta_{vh}\Lambda_h}{\mu_hN_h}I_{v},
\end{aligned}$$ since $\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, \mathbf{I}_h
\bigr\rangle =\bigl\langle
\tilde\Beta
\, \mbox{\Large $\mid$} \, \mathbf{I}_1
\bigr\rangle$. Finally, the derivative of $\mathcal V=\mathcal V_h+\mathcal V_v$ along the trajectories of (\[ModelAD2Compact\]) is obtained by combining Equation (\[LyapBoundary3\]) and Equation (\[LyapVectBoundary\]) as follows: $$\begin{aligned}
\label{CombinedBoundary}
\dot {\mathcal V}
&=&c_1 a\,\beta_{vh}\frac{\nu_h}{\alpha_h}\dfrac{1}{N_{h}}\left(S_{h}-\frac{\Lambda_h}{\mu_h} \right)I_{v} -c_1\frac{a^2\beta_{vh}\Lambda_h}{\mu_hN_h}\frac{\nu_v\nu_h}{(\mu_v+\delta_v)\alpha_v\alpha_h}\frac{1}{N_h}\left(S_v-N_v \right)\bigl\langle\tilde\Beta \, \mbox{\Large $\mid$} \, \mathbf{I}_1 \bigr\rangle\nonumber\\
& &+\sum_{i=s}^nc_ip_i\pi_h\left(2-\frac{I_i^\diamond}{I_i} -\frac{I_i}{I_i^\diamond}\right)+\sum_{i=s}^nc_i\sum_{j=s}^nm_{ij}I_j^\diamond\left(1 - \frac{I_i^\diamond}{I_i}\frac{I_j}{I_j^\diamond}+\ln\frac{I_i^\diamond}{I_i}\frac{I_j}{I_j^\diamond}\right)
- \sum_{i=s}^{n}c_i \frac{I_i^\diamond}{I_i}\sum_{i=1}^{s-1}m_{ij}I_j.\nonumber\end{aligned}$$ Moreover, using the equation of $\dot S_h$ and $\dot S_v$ in Model (\[ModelAD2Compact\]), it is straightforward that $S_h\leq\frac{\Lambda_h}{\mu_h}$ and $S_v\leq N_v:=\frac{\pi_v}{\mu_v+\delta_v}$, where $\displaystyle\Lambda_h=\pi_h\left( 1-p-\sum_{i=1}^{n+1}p_i \right)$. Hence $\dot{\mathcal V}\leq0$. Therefore, by Lyapunov’s theorem this proves the stability of the weakly endemic equilibrium $({S}_h^\diamond, 0,\mathbf{I}_h^\diamond,{S}_v^0,0,0)$. Furthermore, $\dot{\mathcal V}$ is the sum of five nonpositive terms, of which two are definite-negative. Hence, it is straightforward that the largest invariant on which $\dot{\mathcal V}=0$ is reduced to $({S}_h^\diamond, 0,\mathbf{I}_h^\diamond,{S}_v^0,0,0)$. Thus, by LaSalle’s principle, $({S}_h^\diamond, 0,\mathbf{I}_h^\diamond,{S}_v^0,0,0)$ is asymptotically stable. This completes the proof of the global asymptotic stability of the weakly endemic equilibrium $({S}_h^\diamond, 0,\mathbf{I}_h^\diamond,{S}_v^0,0,0)$.
Per , , a necessary condition to break the host-vector transmission, that is, to maintain the vectors disease-free, is $p=0$ and $\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, (\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}\bigr\rangle=0$. The later quantity has an epidemiological interpretation. Indeed, it means that: a.) there is an influx of infected individuals only to a subset of indices and that the hosts in these stages are unable to infect the vectors and b.) the infectious hosts at these stages do not “ameliorate" their infectiosity to stages in the complement of the subset in which they belong. That is, $\delta_{ij}=0$ for all $i\in A$ and $j\in\{1,2,\dots,n\}\setminus A$, with $p_j>0$ for all $j\in A$ and $p_j=0$ otherwise. In this case, the threshold $\mathcal N_0^2(p,\mathbf{p},p_{n+1})$ determine whether or not the vector populations become disease-free. If $\mathcal N_0^2(p,\mathbf{p},p_{n+1})<1$, the disease dies out in the vector population and it thus, the infectious hosts are contained only into the classes in which they are replenished. This threshold captures the capacity of hosts in stage $A$ to maintain the disease in the vector population. Indeed, we can show that: $$\begin{aligned}
\mathcal N_0^2(p,\mathbf{p},p_{n+1})&=&\frac{a^2\beta_{vh}\nu_v\nu_h}{\alpha_v(\mu_v+\delta_v)\alpha_h\mu_h}\frac{N_v}{N_h}\frac{\pi_h\left( 1-p-\sum_{i=1}^{n+1}p_i \right)}{N_h}\bigl\langle
\Beta
\, \mbox{\Large $\mid$} \, (\textrm{diag}(\alpha)-M)^{-1} e_1
\bigr\rangle\\
&:=&\frac{a^2\beta_{vh}\nu_v\nu_h}{\alpha_v(\mu_v+\delta_v)\alpha_h\mu_h}\frac{N_v}{N_h}\frac{\pi_h\left( 1-p-\sum_{i=1}^{n+1}p_i \right)}{N_h}\bigl\langle
\tilde\Beta
\, \mbox{\Large $\mid$} \, (\textrm{diag}(\tilde\alpha)-M_{11})^{-1} e_1
\bigr\rangle\end{aligned}$$
Sharp threshold property {#SharpThreshold}
------------------------
In this subsection, we investigate the dynamics of Model (\[ModelAD2Compact\]) when $p=p_1=\dots,p_n=0$. In this case, we obtain the model $$\label{ModelAD2CompactSharp}
\left\{\begin{array}{llll}\displaystyle\dot S_{h}=\pi_h-a\,\beta_{vh}\,S_{h}\,\dfrac{I_{v}}{N_{h}}-\mu_h S_h\\
\displaystyle\dot E_{h}= a\,\beta_{vh}\,S_{h}\,\dfrac{I_{v}}{N_{h}}-(\mu_h+\nu_h+\eta)E_h\\
\displaystyle\dot{\mathbf{I}}_h=\nu_hE_he_1-(\textrm{diag}(\alpha)-M)\mathbf{I}_h\\
\displaystyle\dot S_{v}=\Lambda_v
-a\dfrac{ S_{v}}{N_h}\bigl\langle
\Beta^T
\, \mbox{\Large $\mid$} \, \mathbf{I}_h
\bigr\rangle
-(\mu_v+\delta_v)S_{v}\\
\displaystyle\dot E_{v}=a\dfrac{ S_{v}}{N_h}\bigl\langle
\Beta^T
\, \mbox{\Large $\mid$} \, \mathbf{I}_h
\bigr\rangle-(\mu_v+\nu_v+\delta_v)E_{v}\\
\displaystyle\dot I_{v}=\nu_vE_v-(\mu_v+\delta_v)I_{v}
\end{array}\right.$$ For the same reason evoked in , the solutions of System (\[ModelAD2CompactSharp\]) stay positive and bounded. Unlike in Model (\[ModelAD2Compact\]), the Model (\[ModelAD2CompactSharp\]) has a disease free equilibrium (DFE), and is given by $(S_h^0,0,0,S_v^0,0,0)$ with $S_h^0=\frac{\pi_h}{\mu_h}$ and $S_v^0=\frac{\pi_v}{\mu_v+\delta_v}$.
The basic reproduction number $\mathcal R_0^2$ is derived using the next generation method. An explicit expression of it is given by $$\begin{aligned}
\mathcal R_0^2&=&\frac{a^2\beta_{vh}\nu_h\nu_vN_v}{(\mu_h+\nu_h+\eta)(\nu_v+\mu_v+\delta_v)(\nu_v+\delta_v)N_h}\Beta^T(\textrm{diag}(\alpha)-M)^{-1}e_1\\
&:=&\mathcal N_0^2(0,\mathbf{0},0).\end{aligned}$$ Note that since the matrix is $M$ is Metzler (off-diagonal elements are non-negative) and invertible, we have $-M^{-1}\geq0$. Thus, $\mathcal R_0^2\geq0$. The following theorem gives the complete asymptotic behavior of Model (\[ModelAD2CompactSharp\]).
\[theo:SharpThreshold\]
1. If $\mathcal R_0^2\leq1$, the DFE is globally asymptotically stable.
2. If $\mathcal R_0^2>1$, the DFE is unstable and a unique endemic equilibrium exists and is GAS.
The proof of the first part of follows using, for example, a left-eigenvector argument. We omit the details. The second part is particular case of . This result is new in itself.
Illustrations and Simulations {#IllustrationsNSimulations}
=============================
In this section, we provide illustrations to highlight the effects of influx of immigrants and the transfer matrix on the disease dynamics and provide some numerical simulations to showcase the results of . To do so, we consider the case $n=4$. That is, there are four infectious stages in the host’s infectivity. Unless otherwise stated, we consider the following baseline parameters: $$\pi_h=1000, a=0.7, \beta_{vh}=0.3,
\mu_h=\frac{1}{75\times365}\;\textrm{days}^{-1},\;
\nu_h=\frac{1}{15}\;\textrm{days}^{-1},\;
\gamma_{12}=\frac{1}{8}\;\textrm{days}^{-1},\; $$$$ \gamma_{23}=\gamma_{34}=\frac{1}{6}\;\textrm{days}^{-1},\;
\frac{1}{\eta}=0\;\textrm{days}^{-1},\; \eta_1=\frac{1}{50}\;\textrm{days}^{-1},\; \eta_2=\eta_3=\frac{1}{30}\;\textrm{days}^{-1},\;\eta_4=\frac{1}{40}\;\textrm{days}^{-1},\; $$ $$\pi_v=10000, \mu_v=\frac{1}{15}\;\textrm{days}^{-1},\; \nu_v=\frac{1}{4}\;\textrm{days}^{-1},\; \delta_v=\frac{1}{20}\;\textrm{days}^{-1}.$$ It is worthwhile noting that, although reasonable, these values do not necessarily match any particular arbovirus diseases. We have chosen them to encompass results of . The transfer matrix $M$ and the vector proportions of influx of infected $\mathbf{p}$ are given by $$M=\begin{pmatrix}
0 & \delta_{21} &\delta_{31} &\delta_{41}\\
\gamma_{12} &0 &\delta_{32} &\delta_{42}\\
\gamma_{13} & \gamma_{23} &0 &\delta_{43}\\
\gamma_{14} & \gamma_{24} &\gamma_{34} &0
\end{pmatrix},\quad
\mathbf{p}=\begin{pmatrix}
p_1\\
p_2\\
p_3\\
p_4
\end{pmatrix}.
$$ We vary the parameter $p$, the vector $\mathbf{p}$ and the matrix $M$ to investigate their impacts on the disease dynamics.
\
and depict the dynamics of the Model (\[ModelAD2Compact\]) when there is no transmission form hosts to vectors. That is, whenever $\Beta=\mathbf{0}_{\R^n}$. In this case, with $\mathbf{p}\neq\mathbf{0}_{\R^n}$, the infected hosts reach an endemic level ( ) while the disease dies out in the vector population (). This is in accordance is the prediction of , , where the weakly endemic equilibrium is GAS ().
For $\Beta=(0.2,0,0,0.5)^T\neq\mathbf{0}_{\R^4}$ and $p=0.01$, the trajectories converge to a strongly endemic equilibrium ( and ) as the hypotheses of , are satisfied.\
To illustrate , , suppose that $\Beta=(\beta_1,0,0,0)^T$ and $\mathbf{p}=(0,0,p_3,p_4)^T$ where $\beta_1>0$, $p_3>0$ and $p_4>0$. By choosing $\gamma_{14}=\gamma_{24}=\delta_{21}=\delta_{31}=\delta_{42}=0$, we obtain:
$$\label{BetaTM}\Beta^T(\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}=\frac{\beta_1\alpha_2\delta_{41}(p_3\gamma_{34}+p_4\alpha_3)}{\det(\textrm{diag}(\alpha)-M))}>0.$$
Using this setup, of anticipates the existence of an strongly endemic equilibrium. Indeed, represents the dynamics of hosts () and vectors () in Model (\[ModelAD2Compact\]) in this case.
By choosing $\delta_{41}=0$, implies that $\Beta^T(\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}=0$ and thus satisfying the conditions of , . And so, a weakly endemic equilibrium $({S}_h^\diamond, 0,\mathbf{I}_h^\diamond,{S}_v^0,0,0)$ or a strongly endemic equilibrium $(\tilde{S}_h, \tilde{E}_h,\tilde{\mathbf{I}}_h,\tilde{S}_v,\tilde{E}_v,\tilde{I}_v)$ exists depending on whether ${\mathcal N}_0^2(p,\mathbf{p},p_{n+1})$ is below or greater than unity, respectively. shows that the hosts’ infection dies out at stage 1 and 2 while it persists at stage 3 and 4. The disease dies out the vectors’ population (). It is worthwhile noting that the disease is maintained at stages 1 and 2, due to the influx of infectious individuals at these stages, without whom, the interaction between hosts and vectors is not sufficient to sustain the infectious. That is, ${\mathcal N}_0^2(p,\mathbf{p},p_{n+1})\leq1$. Under the same transfer matrix $M$ and the infectious influx $\mathbf{p}$ configurations, but choosing the entomological parameters $a=0.9$ and $\beta_{vh}=0.9$, we obtain ${\mathcal N}_0^2(p,\mathbf{p},p_{n+1})=1.6051>1$. This leads to a strongly endemic equilibrium ( and ).
\
Conclusion
==========
Modeling the dynamics of vector-borne diseases have often been based on the assumption that the recruitment into the population is completely susceptible, and thereby making it difficult to assess the effects of the infected or infectious individuals who enters the population. However, the recent surge of vector-borne diseases such as Chikunguyna and Dengue in areas previously free from the pre-cited diseases, that also coincides with an increase of global travel across the world, makes the study of the effects of new arrivals on vector-borne diseases dynamics a necessity. Indeed, the arrival of new individuals from endemic areas, or the return of local residents after a stint in areas where the vector-borne diseases are endemic, could potentially result in infecting the local vector populations and the cycle of host-vector infection could start or accelerate.\
In this paper, we formulate a general staged-progression and stage-regression vector-borne diseases to capture some key features of their dynamics. Particularly, we investigate the effects of the, often swept under the rug, influx of viremic individuals into the population and vectors’ dynamics. We also explore the impacts of treatment and repeated exposure. Indeed, assuming the infectious individuals in the population are undergoing a treatment program, whereby improving their health status; this could lead an infectious individual to go from stage $i$ to a “lower" stage $i-k$, where $1\leq k\leq i$. Similarly, the repeated exposure of infected hosts to infected vectors could lead to more infectious bite. This could lead to increase in infected hosts’ level of parasitemia and thus worsening its health status. In this case, the infected host progresses from stage $i$ to an “upper" stage $i+k$ where $1\leq i\leq k$. And so, we incorporate of these two phenomena of progression and regression on the hosts’ dynamics, which happens to have an altering effects on the qualitative dynamics of the model.\
We derive an staged-progression vector-borne model with $n$ infectious stages. The host-vector dynamics follows and $SEI^nR-SEI$ framework. We assume that a proportion –of the overall recruitment– $p$ and $p_i$, for $i=1,2,\dots,n$, of latent and infectious of stage $i$, respectively, enter into the population. An infectious host at stage $i$ could improve its status from stage $i$ to $i+j$, with $i\leq j$, at a rate $\delta_{ij}$ or worsen its viremicity form stage $i$ to stage $ i\geq j$, at a rate $\gamma_{ij}$. We derived all steady states of the general system and provided conditions under which they exists (). Its turns out that the model has multiple equilibria, depending on the connectivity configuration between host’s infectious stages and the influx of infectious arrivals. However, all of the equilibria are either strongly endemic (SEE) –for which all of the infected and infectious components are positive–, or weakly endemic (WEE)–for which some of host and vectors’ infected or infectious classes are zero. We show the influx of latent individuals into the population guarantees the existence of an SEE, which is globally asymptotically stable (). This case is particularly important for controlling vector-borne diseases as it pertains to public health policies since it is difficult to detect latent even if screening measures were in place.\
When there is no influx of latent but the host-vector transmission vector $\Beta$, the vector of influx of infectious $\mathbf{p}$ and transfer rates matrix $M$ are such that $\Beta^T(\textrm{diag}(\alpha)-M)^{-1}\mathbf{p}=0$, that is, whenever the infectious stages with non zero influx do not transmit the infection to the susceptible vector population. We show that if there is an influx in all infectious stages –$\mathbf{p}\gg0$–, then $\Beta=0$ and the disease dies out in the vector populations (see and the first part of its proof). If $\Beta\neq0$, then there will not be influx of infectious in all stages. Moreover, if infectious hosts in the stages with influx do not improve their to stages that are capable of transmitting the infection to vectors, that is stages where $\beta_i>0$, then a threshold $\mathcal N_0^2(p,\mathbf{p},p_{n+1})$ arises. The disease will die out in the vector population if $\mathcal N_0^2(p,\mathbf{p},p_{n+1})$ is below unity and persists otherwise.\
Our results show that when there is no influx of infected and infectious individuals, the considered model becomes a vector-borne disease with $n$ infectious stages that accounts for amelioration from and progression to any stages. We show that this model has a sharp threshold phenomenon, for which the dynamics is completely determined by the basic reproduction number $\mathcal R_0^2$ (). It turns out that $\mathcal R_0^2=\mathcal N_0^2(0,\mathbf{0},0)$, and if $\mathcal R_0^2\leq1$, the disease-free equilibrium exists and is globally asymptotically stable. Moreover, if $\mathcal R_0^2>1$, we show that an endemic equilibrium exists and is globally asymptotically stable.
[10]{}
, [*Role of immigrants and migrants in emerging infectious diseases*]{}, Medical Clinics of North America, 92 (2008), pp. 1447–1458.
, [ *Spread of the tiger: global risk of invasion by the mosquito aedes albopictus*]{}, Vector-borne and zoonotic Diseases, 7 (2007), pp. 76–85.
, [*Multi-stage vector-borne zoonoses models: A global analysis*]{}, Bulletin of Mathematical Biology, https://doi.org/10.1007/s11538-018-0435-1 (2018).
, [*Modern graph theory*]{}, Springer Science & Business Media, 2013.
, [*Models for transmission of disease with immigration of infectives*]{}, Math. Biosci., 171 (2001).
, [*Asymptotically autonomous epidemic models*]{}, in Mathematical Population Dynamics: Analysis of Heterogeneity, Volume One: Theory of Epidemics,, O. Arino, A. D.E., and M. Kimmel, eds., Wuerz, 1995.
, [*Illnesses from mosquito, tick, and flea bites increasing in the us*]{}, Centers for Disease Control and Prevention, <https://www.cdc.gov/media/releases/2018/p0501-vs-vector-borne.html>, (2018).
, [*Control measures for chagas disease*]{}, Mathematical biosciences, 237 (2012), pp. 49–60.
, [*Analysis of a dengue disease transmission model*]{}, Math. Biosci., 150 (1998), pp. 131–151.
, [*First cases of autochthonous dengue fever and chikungunya fever in france: from bad dream to reality!*]{}, Clinical microbiology and infection, 16 (2010), pp. 1702–1704.
, [*Chikungunya virus, southeastern france*]{}, Emerging infectious diseases, 17 (2011), p. 910.
, [*Impacts of migration and immigration on disease transmission dynamics in heterogeneous populations*]{}, Discrete Contin. Dyn. Syst. Ser. B, 17 (2012), pp. 2413–2430.
, [*Global dynamics of a general class of multistage models for infectious diseases*]{}, SIAM Journal on applied mathematics, 72 (2012), pp. 261–279.
, [*Airport malaria: a review*]{}, Bulletin of the World Health Organization, 67 (1989), p. 737.
, [*Modeling relapsing disease dynamics in a host-vector community*]{}, PLoS Negl Trop Dis, 10 (2016), p. e0004428.
, [*Global trends in emerging infectious diseases*]{}, Nature, 451 (2008), p. 990.
, [*Drivers, dynamics, and control of emerging vector-borne zoonotic diseases*]{}, The Lancet, 380 (2012), pp. 1946–1955.
, [*On r.a. smith’s automonmous convergence theorem*]{}, Rocky Mountain J. Math., 25 (1995), pp. 365–379.
, [*Global stability for the [SEIR]{} model in epidemiology*]{}, Math. Biosci., 125 (1995), pp. 155–164.
, [*[A model of HIV/AIDS with staged progression and amelioration.]{}*]{}, Math. Biosci., 181 (2003), pp. 1–16.
, [*[Global analysis of two tuberculosis models.]{}*]{}, J. Dyn. Differ. Equations, 16 (2004), pp. 139–166.
, [*Geographic spread of aedes albopictus and potential for involvement in arbovirus cycles in the mediterranean basin*]{}, Journal of Vector ecology, 20 (1985), pp. 44–58.
, [*Counting labeled trees,*]{}. Canadian Math, Monographs, 1970.
, [*The dynamics of vector-borne relapsing diseases*]{}, Math. Biosci., 297 (2018).
, [*Large number of imported chikungunya cases in mainland france, 2014: a challenge for surveillance and response*]{}, Eurosurveillance, 19 (2014), p. 20856.
, [*Transmission potential of chikungunya virus and control measures: the case of italy*]{}, PLoS One, 6 (2011), p. e18860.
, [*Infection with chikungunya virus in italy: an outbreak in a temperate region*]{}, The Lancet, 370 (2007), pp. 1840–1846.
, [*Climate change and vector-borne diseases*]{}, Advances in parasitology, 62 (2006), pp. 345–381.
, [**AEDES ALBOPICTUS* and arboviruses: A concise review of the literaturei*]{}, Journal of the American Mosquito Control Association, (1986).
, [*Zika virus targets different primary human placental cells, suggesting two routes for vertical transmission*]{}, Cell host & microbe, 20 (2016), pp. 155–166.
, [*Mathematics in population biology*]{}, Princeton Series in Theoretical and Computational Biology, Princeton University Press, Princeton, NJ, 2003.
, [*A host-vector model for malaria with infective immigrants*]{}, Journal of Mathematical Analysis and Applications, 361 (2010), pp. 139–149.
, [*High efficiency of temperate aedes albopictus to transmit chikungunya and dengue viruses in the southeast of france*]{}, PLoS One, 8 (2013), p. e59716.
, [*[Decomposition techniques for large-scale systems with nonadditive interactions: Stability and stabilizability.]{}*]{}, IEEE Trans. Autom. Control, 25 (1980), pp. 773–779.
, [*The 2012 dengue outbreak in madeira: exploring the origins*]{}, Euro Surveill., 19 (2014), p. 20718.
, [*The world health report: 2004: changing history*]{}, 2004.
height 2pt depth -1.6pt width 23pt, [*A global brief on vector-borne diseases*]{}, 2014.
|
{
"pile_set_name": "ArXiv"
}
|
@ssage[16]{} @ssage @ssage[if two columns per page, type any character now]{} @ssage[otherwise, or if confused, hit return.]{} @f@ult @col =-1 -1 tonswer =13 nswer@f@ult@colfalse@ssage[Single column.]{} @coltrue@ssage[Two columns per page. Use landscape mode when printing.]{}@stp@geempty ł@stp@geemptyfalse @col
ł@r=L =[ ]{} \#1[ Lł@r 1=1 =\#1 ł@r=R 1=2 ł@r=L]{} =-1.5mm =-15.5mm =1 =8.4truein =0.5truein =0.35truein =6.3truein =10
=cmr7 scaled 4 =cmr10 scaled 2 =cmssbx10 scaled 2
\#1 \#2
[ plus .1-200 plus -.1 ]{}
>
\#1/\#2[.1em .5ex-.1em /-.05em.4ex]{}
\#1\#2
Dilogarithm identities, fusion rules
and structure constants of CFTs
Michael Terhoeven
Physikalisches Institut
der
Rheinischen Friedrich-Wilhelms Universität Bonn
Nussallee 12
D-53115 Bonn
unp044 at ibm.rhrz.uni-bonn.de
**Abstract**
Recently dilogarithm identities have made their appearance in the physics literature. These identities seem to allow to calculate structure constants like, in particular, the effective central charge of certain conformal field theories from their fusion rules. In Nahm, Recknagel, Terhoeven (1992) a proof of identities of this type was given by considering the asymptotics of character functions in the so-called Rogers-Ramanujan sum form and comparing with the asymptotics predicted by modular covariance. Refining the argument, we obtain [*the general connection of quantum dimensions of certain conformal field theories to the arguments of the dilogarithm function*]{} in the identities in question and [*an infinite set of consistency conditions on the parameters of Rogers-Ramanujan type partitions for them to be modular covariant*]{}.
=1
1 Introduction and Overview
Recently, the dilogarithm function has reappeared (Babujan (1983)) in the physics of two-dimensional quantum field theories and lattice models (Bazhanov, Reshetikhin (1989), Klassen, Melzer (1990), Klümper, Pearce (1991), Zamolodchikov (1990)). In Nahm, Recknagel, Terhoeven (1992) a connection of fusion rules to thermodynamic Bethe Ansatz (TBA) type equations was conjectured. In this paper, we make this connection precise in the framework of CFT.
Crucial in the latter approach of the subject is to (re)write the characters of the conformal field theory (CFT) in the Rogers-Ramanujan sum form ((1.1) below). In Terhoeven (1992), Kuniba, Nakanishi, Suzuki (1992) this form was conjectured to exist for general parafermionic theories. Allowing a slight generalization, one obtains formulas of similar type for characters of all unitary Virasoro minimal models (Dashmahapatra, Kedem, Klassen, McCoy, Melzer (1993) and references therein, also for the so-called quasi-particle picture). Thus it seems possible that the class of theories allowing their characters to be written in a suitable form is not too far away from the set of all CFTs. In this paper, we find an infinite set of consistency conditions to be satisfied by all partitions of the form (1.1), which is a first step in the classification of this type of partitions, possibly relevant for a classification of all rational CFTs.
In the following, we consider partition functions of the Rogers-Ramanujan (or Euler) sum form $$\eqalign{
Z(q) &=
\sum_{\matrix{&\scriptstyle m_1,\ldots ,m_n\in\zzz_{\geq 0}\cr
&\scriptstyle c\cdot {\bf m} = \gamma\mod 1\cr}}
{ q^{{\bf m}B{\bf m}^t + {\bf b\cdot m} + \beta }
\over (q)_{\bf m}} \cr
&= \sum_{l\in\zzz_K} {1\over K}
\sum_{ m_1,\ldots ,m_n\in\zzz_{\geq 0}}
{ q^{{\bf m}B{\bf m}^t + {\bf b\cdot m} + \beta }
\over (q)_{\bf m}}\ e^{-2\pi il({\bf c\cdot m}-\gamma)},
\cr }
\eqno(1.1)$$ where $(q)_{\bf m}=(q)_{m_1}\ldots (q)_{m_n}$, $(q)_m = (1-q)(1-q^2)\ldots(1-q^m)$, $B_{ij}$, $b_i$, $c_i$, $\beta$ are rational numbers, $\gamma \in {\zzz_K}/K$ ($K\in\nnn$) and $B$ is symmetric and positive definite. Let $q = e^{2\pi i\tau}=e^{-2\pi t}$ and $\tilde q = e^{-2\pi i/\tau}=e^{-2\pi /t}$, thus $1/\log(q) = \log(\tilde q)/4\pi^2$ and in the limit $t=-i\tau \searrow 0$ we have $q \nearrow 1$.
More accurately, we consider the asymptotic of $(1.1)$ in the above limit. Refining the argument of Nahm, Recknagel, Terhoeven (1992) and requiring (1.1) to be modular covariant, we obtain not only a dilogarithm identity for the lowest energy state in the spectrum. More than that: our main results are [*the connection of fusion rules to TBA-type equations*]{} (meaning a formula relating quantum dimensions and a certain solution of the TBA-type equations, the arguments of the dilogarithm function in the formula for the effective central charge) and [*an infinite set of consistency equations restricting the possible choices of $B$, $\bb$ and $\beta$*]{} (the discussion in section four).
The outline of the paper is as follows:
In the rather technical second section we study the asymptotics of (1.1) for $q \nearrow 1$ using an integral representation and integral transform of the partition function following Meinardus (1954) and a higher order saddle point approximation, preferable diagrammatically (à la Feynman).
In the third section we compare the partition asymptotics calculated explicitly with the one predicted by modular covariance, and thus find the results mentioned above.
Finally, we outline future work and as a short application present a classification of modular covariant partitions of the Rogers-Ramanujan form in the case rank $B =n=1$, which is found from the first few consistency equations in this particular case.
2 Partition asymptotics
In this section, we will extract the first term of the modular transform of $Z(q)$ – a power series in $\tilde q$ – from the asymptotic behavior of (1.1) when $q\nearrow 1$.
Following Meinardus (1954), we write (1.1) in an integral representation using [*Cauchy’s theorem*]{} $$Z(q) =\oint\ \prod_j{dw_j\over 2\pi i\ w_j}\ {q^{\beta}\over K}\
\big( \sum_{\matrix{\scriptstyle \tmu\in\zzz^n\cr
\scriptstyle l\in\zzz_K \cr}}
q^{\tmu B \tmu^t }\ e^{-2\pi i l(\tmu\cdot\cc -\gamma)} \
\prod_j w_j^{-\mu_j} \big)\
\prod_j \Big( \sum_{m_j\geq 0} {q^{ b_j m_j } \over (q)_{m_j}}
w_j^{m_j} \Big) .
\eqno(2.1)$$ On the first bracket inside the multiple integral in (2.1) we apply the well-known [*Jacobi inversion formula*]{} (Gunning (1962)) ($|q|<1$) $$\eqalign{
&\qquad \sum_{\tmu\in\zzz^n,\ l\in\zzz_K}
q^{ \tmu B \tmu}\ e^{-2\pi i l(\tmu\cdot\cc -\gamma)} \ \ww^{\tmu}\cr
&= {1\over \sqrt{\det(2B)\ t^n }}\
\sum_{l\in\zzz_K}\ e^{2\pi i l\gamma}\
\sum_{\tmu\in\zzz^n}
{\tilde q}^{(\tmu+\log(\ww)/2\pi i+l\cc)
B^{-1}(\tmu+\log(\ww)/2\pi i+l\cc)/4}, \cr}
\eqno(2.2)$$ where $\ww^{\tmu}=\prod_j w_j^{\mu_j}$. The terms in the second bracket of (2.1) require some more explicit calculations. First, we use a relation going back to [*Euler*]{} (Andrews (1976)) ($|q|<1, |w|<|q|^{-b}$) $$\sum_{m\geq 0} {q^{ b m } \over (q)_{m}}\ w^m =
\prod_{n\geq 0} (1-w\ q^{n+b})^{-1} .
\eqno(2.3)$$ Second, we apply ${\rm id}=\exp\circ\log$ and the Taylor series expansion of $\log(1-x)$ to obtain $$\exp\big\{ \sum_{r=1}^{\infty} {w^r\over r}
\sum_{n=0}^{\infty} e^{-2\pi t(n+b)r}
\big\} .
\eqno(2.4)$$ Using [*inverse Mellin transformation*]{} (Davies (1978)), the exponent in (2.4) can be written as $${1\over 2\pi i} \int_{3/2-i\infty}^{3/2+i\infty}
ds\ \Gamma(s)\ D(s+1,w)\ \zeta(s,b)\ (2\pi t)^{-s},
\eqno(2.5)$$ where $\Gamma(s)$ is the Gamma function with $\res_{s=-n}\Gamma(s)
= {(-)^n\over n!}$ $(n>0)$, $\zeta(s,b)$ is the generalized Riemann $\zeta$-function $\zeta(s,b)=\sum_{n=0}^{\infty} (n+b)^{-s}$ with $\res_{s=-n}\zeta(s,b) = - {\varphi_{n+1}(b)\over (n+1)}$ $(n>0)$, where $\varphi_n(x)$ are the Bernoulli polynomials ($\varphi_0(x)=1$, $\varphi_1(x)=x-1/2$, $\varphi_2(x)=x^2-x+1/6$, $\ldots$) generated by ${te^{xt}\over (e^t-1)}=\sum_{n=0}^{\infty} {t^n\over n!}\
\varphi_n(x)$ and $D(s,w)=\sum_{r=1}^{\infty} {w^r\over r^s}$ ($|w|<1$).
Third, shifting the contour of integration in (2.5) along the real axis, we pick up one by one the residues of the integrand at $s=1,0,-1,\ldots\ $, the first coming from the pole of $\zeta$, the others from poles of $\Gamma$. This gives the following expression for the exponent of (2.4) $$-{1\over\log(q)} \ \sum_{p=0}^{\infty}\ \log^p(q)\
{\varphi_p(b) \over p!}\
(w\partial_w)^p\
\sum_{r=1}^{\infty} {w^r\over r^2},
\eqno(2.6)$$ which, using the generating function of the Bernoulli polynomials and the series definition of the dilogarithm $\LL(w) =
\sum_{r=1}^{\infty} {w^r\over r^2}$, can be written more compactly as $$- w\partial_w \
{q^{bw\partial_w}\over q^{w\partial_w}-1}\quad
\LL (w).
\eqno(2.7)$$
Taking (2.6) and (2.2), we can write (2.1) in the form $$Z(q) = \oint\
\prod_j{dw_j\over 2\pi i\ w_j}\
{1\over K\sqrt{\det(2B)\ t^n}}\
\sum_{\tmu\in\zzz^n,\ l\in\zzz_K}\ \exp(\emu(q,\ww)),
\eqno(2.8a)$$ where $$\eqalign{
\emu(q,\ww)= &\log(q)\ \beta\ + \ 2\pi il\gamma\ \cr
& + {4\pi^2\over \log(q)}\ {1\over 4} (\tmu+{\log(\ww)\over 2\pi i}+l\cc)
B^{-1}(\tmu+{\log(\ww)\over 2\pi i}+l\cc)^t \cr
& -{1\over\log(q)} \ \sum_{p=0}^{\infty}\ \log^p(q)\
\sum_{a=1}^n\ {\varphi_p(b_a) \over p!}\
(w_a\partial_{w_a})^p\
\sum_{r_a=1}^{\infty} {w_a^{r_a}\over r_a^2}, \cr }
\eqno(2.8b)$$ which in lowest order in $\log(q)$ reduces to $${1\over \log(q)}\Big( 4\pi^2\ { 1\over 4}
(\tmu+{\log(\ww)\over 2\pi i}+l\cc)B^{-1}
(\tmu+{\log(\ww)\over 2\pi i}+l\cc)^t
- \sum_a\ \LL(w_a) \Big).
\eqno(2.9)$$ The sum over $\tmu$ together with the contour integral can be understood as an infinite contour integral over all branches of the analytically continued logarithm $\log(\ww)$. Thus, we forget (the sum over) $\tmu$ in the following and understand the contour integral as the appropriate infinite integral.
We continue with the [*saddle point approximation*]{} of (2.8) (in short: the value of the integral is approximated by the integrand at the saddle point):
A necessary condition for a minimum in a point $\ww$ (in lowest $\log(q)$ order, only then $\ww$ is independent of $q$, a number) is given by $$\eqalign{
0 & = w_a \partial_{w_a}\ \eml(q,\ww) + {\cal O} (\log^0(q)) \cr
& = {1\over \log(q)} \Big(
- 2\pi i B_{aa'}^{-1}({\log(\ww)\over 2\pi i}+l\cc)_{a'}/2
+ \log(1-w_a)
\Big) \cr }
\eqno(2.10)$$ or $${1\over 2} B_{aa'}^{-1}({\log(\ww)\over 2\pi i}+l\cc)_{a'}
= {\log(1-w_a)\over 2\pi i}
\eqno(2.11)$$ or in exponentiated form (ignoring for the moment possible phases) $$w_{a'} = \prod_{a=1}^n\ (1-w_a)^{2B_{aa'}},
\eqno(2.12)$$ which appeared in the physical literature as certain limits of TBA-type (thermodynamic Bethe Ansatz) equations. We will assume in the following that a unique solution of (2.10,11) with $0<w_a<1$ exists and that furthermore $l=0$ for this solution, which is indeed true for all examples known to us.
Substituting (2.11) into (2.8b) and introducing Rogers’ dilogarithm function $$\lll(z) = \LL(z) + {1\over 2} \log(z)\log(1-z),
\eqno(2.13)$$ we obtain the exponent function at the saddle point $\ww=\ww_s (l=0)$ (repeated indices are summed over) $$\eqalign{
\eml(q,\ww=\ww_s) =
&\log(\tilde q)\ \Big\{
-{1\over 24\ \lll(1)}\sum_{a=1}^n \lll (w_a) \Big\} \cr
+ & \log^0(q)\ \Big\{ \varphi_1(b_a)\
\log (1-w_{a}) \Big\}\cr
+ & \log^1(q)\ \Big\{ \beta -
{\varphi_2(b_a)\over 2}\ {w_a\over 1-w_a} \Big\}
+ {\cal O} (\log^2(q)). \cr }
\eqno(2.14)$$
The $\log(\tilde q)$ part of (2.14) yields the first and crudest approximation of the integral. To extract [*higher order corrections*]{}, we change the variable of integration in (2.8) by the substitution $w_a = e^{\sqrt{-2\pi t}\ v_a}=e^{\sqrt{\log(q)}\ v_a}$ ($\partial_{v_a} = \sqrt{\log(q)}\ w_a\partial_{w_a} $) and expand $\eml(q,\vv)$ in a Taylor series around the saddle point $\vv_s$ to obtain $$Z(q) = \int\
{1\over K\sqrt{\det(2B)}}\
\sum_{ l\in\zzz_K}\ \exp(\eml(q,\vv_s +\tD))\
\prod_j{d\tD_j\over\sqrt{2\pi}} ,
\eqno(2.15a)$$ where $$\eqalign{
\eml(q,\vv_s+\tD)=
\eml(q,\vv) \big|_{\vv=\vv_s}
& + \sum_a \tD_a \partial_{\vv_a} \eml(q,\vv)
\big|_{\vv=\vv_s}\cr
& + {1\over 2} \sum_{a,a'} \tD_a \tD_{a'} \ \partial_{v_a}
\partial_{v_{a'}} \eml(q,\vv) \big|_{\vv=\vv_s} \cr
& + {1\over 6} \sum_a \tD_a^3 \ \partial_{v_a}^3
\eml(q,\vv) \big|_{\vv=\vv_s} + \ldots \quad . \cr}
\eqno(2.15b)$$ The first derivative of $\eml(q,\vv)$ is given by $$\partial_{v_a} \eml(q,\vv) \big|_{\vv=\vv_s} =
- \sum_{p=1}^{\infty}\ \log^{p-1/2}(q)\
{\varphi_p(b_a) \over p!}\
\sum_{r_a=1}^{\infty} w_a^{r_a} r_a^{p-1},
\eqno(2.16)$$ the second by $$\eqalign{
\partial_{v_a} \partial_{v_{a'}}\
\eml(q,\vv) \big|_{\vv=\vv_s} =
- & (2 B)^{-1}_{aa'} -\delta_{a,a'} {w_a\over 1-w_a} \cr
& - \delta_{a,a'}\ \sum_{p=1}^{\infty}\ \log^p(q)\
{\varphi_p(b_a) \over p!}\
\sum_{r_a=1}^{\infty} w_a^{r_a} r_a^{p}, \cr }
\eqno(2.17)$$ and the higher ones for $n\geq 3$ by $$\partial_{v_a}^n\ \eml(q,\vv) \big|_{\vv=\vv_s} =
- \sum_{p=0}^{\infty}\ \log^{p-1+n/2}(q)\
{\varphi_p(b_a) \over p!}\
\sum_{r_a=1}^{\infty} w_a^{r_a} r_a^{p+n-2}.
\eqno(2.18)$$
The next $\log(q)$ order of our expression of $Z(q)$ can then be found by evaluating the integral as a gaussian integral. Namely, for any symmetric, real, positive definite matrix $A$, one has $$\int\ e^{-{1\over 2} xAx^t-d\cdot x-c}\ {d^nx\over (2\pi)^{n/2}}
= {1\over\sqrt{\det{A}}}\ e^{{1\over 2} dA^{-1}d^t-c},
\eqno(2.19)$$ leading to $$\eqalign{
Z(q) =
& \exp \bigg( \log({\tilde q}) \Big\{
-{1\over 24\lll(1)}\ \sum_{a=1}^n \lll (w_a) \Big\} \bigg) \cr
& \exp\Big( \sum_{a=1}^n \varphi_1(b_a)\ \log(1-w_a) \Big)
\ \Big/ \ K\sqrt{ \det(2B\cdot A) } \cr
& \Big( 1+{\cal O}\big( \log(q) \big) \Big) , \cr }
\eqno(2.20)$$ where $d_a = \varphi_1(b_a) {w_a\over 1-w_a}\ \sqrt{\log(q)}$ and $$A_{aa'}=(2B)_{aa'}^{-1}+\delta_{a,a'} {w_a\over 1-w_a} .
\eqno(2.21)$$ The next order in $\log(q)$ leads to replacing $
\Big( 1+{\cal O}\big( \log(q) \big) \Big)$ by (sums over all indices $a,a'=1,\ldots n$ implied) $$\eqalign{
\exp \bigg\lbrack \log^1(q)\ \bigg( &
\Big\{ \beta - {\varphi_2(b_a)\over 2}\ {w_a\over 1-w_a}
\Big\} \cr
& + {1\over 2}
\varphi_1(b_a) {w_a\over 1-w_a} A^{-1}_{aa'}
\varphi_1(b_{a'}) {w_{a'}\over 1-w_{a'}} \cr
& - {1\over 2}
\varphi_1(b_a) {w_a\over (1-w_a)^2} A^{-1}_{aa} \cr
& + {1\over 2}
{w_a\over (1-w_a)^2} A^{-1}_{aa} A^{-1}_{aa'}
\varphi_1(b_{a'}) {w_{a'}\over 1-w_{a'}} \cr
& - {1\over 8}
{w_a(1+w_a) \over (1-w_a)^3} (A^{-1}_{aa})^2 \cr
& + {1\over 12}
{w_a\over (1-w_a)^2} (A^{-1}_{aa'})^3
{w_{a'}\over (1-w_{a'})^2} \cr
& + {1\over 8}
{w_a\over (1-w_a)^2} A^{-1}_{aa} A^{-1}_{aa'} A^{-1}_{a'a'}
{w_{a'}\over (1-w_{a'})^2}
\bigg) \bigg\rbrack \cr
& \phantom{mmmmmmmmmmm}
\Big( 1 + {\cal O}\big(\log^2(q)\big) \Big) \cr }
\eqno(2.22)$$ in (2.20). It can be found by taking derivatives of (2.19) with respect to $d$ and then collecting all terms of order $\log(q)$. Once written down, one also recognizes it as the exponential of the generating function of the connected Feynman diagrams of order $\log(q)$ (with corresponding factors of internal symmetry) allowed by the Feynman rules encoded in (2.15).
This insight allows one to write down the next order by hand (with $35$ ’new’ diagrams, meaning diagrams not constructed out of ’dressings’ ($p\mapsto p+1$ in (2.16-18)) of (2.22)). The next to next order already is a bit unwieldly with 367 diagrams before choosing possible dressings. Higher orders can in principle be written down. However, the proliferation of diagrams is enormous and we have looked beyond $\log(q)$ only in the comparably simple case where ${\rm rank}\ B=1$.
3 Modular covariance
In this section we compare the asymptotics found above with the prediction of modular covariance.
In the following, we suppose that the partition $Z(q)$ given in $(1.1)$ equals a character of some CFT with effective central charge $\cef = c-24\ \hm$ (the index $\min$ refers to the primary field of minimal conformal dimension) – more accurately: $Z(q)$ equals the character corresponding to the primary field $\phi_{\lambda}$ of conformal dimension $h_{\lambda}$, thus $$Z^{(\lambda)}(q)=
q^{h_{\lambda}-c/24}\sum_{m\in\zzz_{\geq 0}}d^{(\lambda)}_m q^m .
\eqno(3.1)$$ Then by modular covariance we have $$Z^{(\lambda)}(q)
=\sum_{{\lambda}'} S_{{\lambda},{\lambda}'}\
Z^{(\lambda')}(\tilde q)
=\sum_{{\lambda}'} S_{{\lambda},{\lambda}'}\
{\tilde q}\, {}^{h_{{\lambda}'}-c/24}
\sum_{m\in\zzz_{\geq 0}}d^{(\lambda')}_m\ {\tilde q}\, {}^m,
\eqno(3.2)$$ where the $d_m^{(\lambda)}$ are the multiplicities of states at grade $m$.
Comparing (1.1) with (3.1), we see that $$h_{\lambda}-c/24=
\min_{\bf m\geq 0}({\bf m}B{\bf m}^t + {\bf b\cdot m} + \beta).
\eqno(3.3)$$
Comparing the first term (in powers of ${\tilde q}$) of the right hand side of (3.2) with (2.20,22), we obtain $$\eqalignno{
& \cef = \sum_{a=1}^n \ {\lll (w_a)\over \lll (1)} &(A)\cr
& S_{\lambda ,\min} = {\exp\Big(
\sum_{a=1}^n \varphi_1(b^{(\lambda)}_a)\
\log(1-w_a) \Big) \over
K \sqrt{ \det(2B\cdot A)} } &(B)\cr
& 0 = {\rm exponent\ of\ (2.22)}
\big|_{\beta=\beta^{(\lambda)},\bb=\bb^{(\lambda)}}
&(C)\cr
& 0 = {\rm higher\ orders\ in\ \log(q)\ of\ the\
expansion\ of\ } Z^{(\lambda)}(q), &(D)
}$$ where as before $$A_{aa'}=(2B)_{aa'}^{-1}+\delta_{a,a'}
{w_a\over 1-w_a} .$$ In words:
$(A)$ gives the effective central charge of the theory in terms of a dilogarithmic expression with arguments $w_a$ given by a certain solution of the TBA-type equations (2.12) which only depend on $B$.
$(B)$ gives an expression for certain elements of the modular S-matrix in terms of the $w_a$. From $(B)$ we obtain the following relation between quantum dimensions and solutions of the TBA-type equations $${ S_{\lambda ,\min}\over S_{0 ,\min}}
= \ \prod_{a=1}^n\
(1-w_a)^{\displaystyle
b_a^{(\lambda)}-b_a^{(0)}} ,
\eqno(3.4)$$ where $0$ denotes the vacuum of the theory.
$(C)$ fixes $\beta$ in terms of $w_a$ ($B$) and $\bb$, thus with (3.3) gives an expression for the conformal dimensions of the theory.
Finally, $(D)$ is an innocent looking short-hand for an infinite set of complicated consistency conditions for modular covariance putting constraints on $B$ and $\bb$. However, trying to solve them is highly non trivial and of the type to solve $(C)$ for given $\beta$ in terms of $B$ and $\bb$. For the case ${\rm rank}\ B=1$ we refer the reader to the outlook.
4 Outlook
Assuming the existence of a character with $b=0$, we have classified the partitions of type (1.1) with rank $B=n=1$:
We found numerically the zeroes of the first non-trivial consistency condition (order $\log^2(q)$) in $w\in\rbrack 0,1\lbrack$, namely $\tau,\ 1/2,\ \tau^2$ ($\tau^2+\tau-1=0$), which correspond to the $(3,5)-$, $(3,4)-$ and $(2,5)-$model in the series of Virasoro minimal models, with $B=1/4,1/2,1$ respectively. Given $w$ and thus $B$, we could solve the first consistency conditions for other possible $b\neq 0$ and found the known results.
Certainly one should try to attack the higher rank case by the same method and to extend the argument to the more general partitions mentioned in the introduction.
Also, we would like to understand, in a way similar to the discussion given, other dilogarithm identities leading to the other effective conformal dimensions or possibly (Kuniba, Nakanishi, Suzuki (1992)) to the whole spectrum of CFTs, we would like to be able to explicitly construct the whole modular transform of partitions in the Rogers-Ramanujan form and not only the first term. A ’naive’ analytic continuation of the (di)logarithm in (2.20) leads to a form surprisingly similar to (1.1) again. However, a closer lock reveals certain inconsistencies and substantially reduces the predictive power (only $(3.C)$ seems to hold without any ’tuning’). This problem is solved by W. Nahm (1993).
A -6pt cknowledgements
I thank W. Nahm for encouragement and advice.
I am indebted to W. Eholzer, R. Kellerhals, Andi Recknagel, M. Rösgen, R. Varnhagen and D. Zagier for essential discussions on related subjects.
During the time of writing I was supported by my parents and the Studienstiftung des deutschen Volkes.
R -6pt eferences
=-1truecm =0.2cm =1truecm
G.E. Andrews (1976), [*The theory of partitions*]{}, Encyclopedia of Mathematics and Its Applications, Vol. 2, Addison Wesley.
V. V. Bazahnov, N. Yu. Reshetikhin (1989), [*Int. J. Mod. Phys. A*]{} [**4**]{}, 115.
H. M. Babujan (1983), [*Nucl. Phys.*]{} [**B215**]{}, 317-36.
S. Dasmahapatra, R. Kedem, T. R. Klassen, B. M. McCoy, E. Melzer (1993), [*Quasi-Particles, Conformal Field Theory, and q-Series*]{}, ITP-SB-93-12, RU-93-07, hep-th/9303013, to appear in the Proc. of Yang-Baxter Equations in Paris Conf., Paris, France, July 24-30, (1992?).
B. Davies (1978), [*Integral transforms and their applications*]{}, Applied mathematical sciences, Vol. 25, Springer-Verlag.
R. C. Gunning (1962), [Lectures on Modular Forms]{}, Princeton University Press.
T. R. Klassen, E. Melzer (1990), [*Nucl. Phys.* ]{}[**B338**]{}, 485-528; (1992), [*Nucl. Phys.* ]{}[**B370**]{}, 511.
A. Klümper, P. A. Pearce (1991), [*J. Stat. Phys.*]{} [**64**]{}, 13.
A. Kuniba, T. Nakanishi (1992), [*Mod. Phys. Lett.*]{} [**A7**]{} (1992), 3487-94.
A. Kuniba, T. Nakanishi, J. Suzuki (1992), [Characters in CFTs from Thermodynamic Bethe Ansatz]{}, preprint hep-th/9301018, HUTP-92/A069.
G. Meinardus (1954), [*Math. Zeitschr.* ]{}, Bd. [**1**]{}, 289-302.
W. Nahm, A. Recknagel, M. Terhoeven (1992), hep-th/9211034, preprint BONN-HE-92-35, to appear in [*Mod. Phys. Lett.*]{} [**A**]{}.
W. Nahm (1993), in preparation.
B. Richmond, G. Szekeres (1981), [*J. Austral. Math. Soc.*]{} [*(Series A) 31*]{}, 362-373.
M. Terhoeven (1992), [*Lift of dilogarithm to partition identities*]{}, preprint BONN-HE-92-36, hep-th/9211120.
Al. B. Zamolodchikov (1990), [*Nucl. Phys. B*]{} [**342**]{}, 695.
———————————————————————
michael terhoeven endenicher allee 11-13, avz 104 53115 bonn (53oo bonn 1) deutschland
tel.: o228-733727
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
We compare analytic predictions for real and Fourier space two-point statistics for biased tracers from a variety of Lagrangian Perturbation Theory approaches against those from state of the art N-body simulations in $f(R)$ Hu-Sawicki and the nDGP braneworld modified gravity theories.
We show that the novel physics of gravitational collapse in scalar tensor theories with the chameleon or the Vainshtein screening mechanism can be effectively factored in with bias parameters analytically predicted using the Peak-Background Split formalism when updated to include the environmental sensitivity of modified gravity theories as well as changes to the halo mass function.
We demonstrate that Convolution Lagrangian Perturbation Theory (CLPT) and Standard Perturbation Theory (SPT) approaches provide accurate analytic methods to predict the correlation function and power spectra, respectively, for biased tracers in modified gravity models and are able to characterize both the BAO, power-law and small scale regimes needed for upcoming galaxy surveys such as DESI, Euclid, LSST and WFIRST.
author:
- 'Georgios Valogiannis and Rachel Bean.'
nocite: '[@*]'
title: Convolution Lagrangian perturbation theory for biased tracers beyond general relativity
---
\[firstpage\]
Introduction {#sec:intro}
============
About 20 years after the discovery of cosmic acceleration [@Perlmutter:1998np; @Riess:2004nr], the $\Lambda$CDM model of the universe is rightfully called the standard model of cosmology, after having successfully passed a broad range of tight observational tests [@Eisenstein:2005su; @Percival:2007yw; @Percival:2009xn; @Kazin:2014qga; @Spergel:2013tha; @Ade:2013zuv; @Ade:2015xua]. This model augments the cosmic baryonic and leptonic matter components in the Standard Model, with two additional ingredients that dominate the cosmic energy budget, cold dark matter (CDM) and a cosmological constant, $\Lambda$, with the latter playing the role of vacuum energy that is responsible for the observed late-time acceleration. Central to the above paradigm is also the assumption that gravitational evolution is governed by Einstein’s General Theory of Relativity (GR), which led to the observed inhomogeneous pattern of the Large-Scale Structure (LSS) of the Universe through the process of gravitational instability. The value of the vacuum energy predicted by quantum field theory, however, is orders of magnitude larger than the best-fit one that is necessary to explain cosmic acceleration, so $\Lambda$ needs to be fine-tuned; the so-called cosmological constant problem [@Weinberg:1988cp]. Such an unfortunate mismatch, together with the need to fully explore the space of all theoretical alternatives, has generated growing interest in testing gravity models that self-accelerate though a large-scale modification to GR, instead of dark energy, the so-called Modified Gravity (MG) models [@Koyama:2015vza; @Ishak:2018his].
Modifying the Einstein-Hilbert action, however, introduces in principle an additional degree of freedom that is conformally coupled to matter and can produce significant deviations from the predictions of GR, which have passed a wide array of precise observational tests, especially in the Solar System [@Will:2005va]. Furthermore, the recent simultaneous detection of gravitational waves and EM counterparts by the LIGO/Virgo collaboration [@TheLIGOScientific:2017qsa; @Goldstein:2017mmi; @Savchenko:2017ffs; @Monitor:2017mdv; @GBM:2017lvd], has placed additional constraints [@Sakstein:2017xjx; @Ezquiaga:2017ekz; @Creminelli:2017sry; @Baker:2017hug] into the form of the most general expression of a scalar-tensor theory that produces second order equations of motion, described by the Horndeski action [@Horndeski1974; @PhysRevD.84.064039].
In order to be able to confront such tight constraints successfully, while at the same time provide a stable self-accelerative cosmic mechanism, viable candidates contain a restoring property, called “screening" [@Khoury:2010xi; @Khoury:2013tda], which is a dynamical mechanism that weakens the additional fifth forces in high-density environments through the corresponding scalar field self-interactions. In the Vainshtein mechanism [@VAINSHTEIN1972393; @Babichev:2013usa], GR is recovered thanks to the second derivative terms in the scalar field Lagrangian, that become large in high density environments and effectively weaken the coupling to the matter sources. The Vainshtein mechanism is very efficient in the vicinity of a massive source and contains a rich phenomenology, which makes it particularly attractive. Another popular class of screening consists of the chameleons [@PhysRevD.69.044026; @PhysRevLett.93.171104], where in regions of high potential the scalar fields become massive and cannot propagate, resulting thus in suppression of the fifth forces. Despite the fact that chameleons cannot produce self-acceleration [@Wang:2012kj], their very interesting phenomenology makes them serve as ideal testbeds for gravity and considerable efforts have been put into their study in the past decade [@Burrage:2017qrf]. Other screening classes include the symmetrons [@PhysRevLett.104.231301; @Olive:2007aj], that employ spontaneous symmetry breaking and share qualitative similarities with the chameleons and the K-Mouflage [@Dvali:2010jz; @Babichev:2013usa], in which deviations are suppressed when scalar field gradients exceed a certain value.
The observed inhomogeneous LSS of the Universe, is the outcome of the subsequent nonlinear gravitational evolution of the primordial density fluctuations, partially modulated by the late-time acceleration. As a result, it provides us with an observational window into the fundamental physics that shaped this process, including sensitive tests of the underlying large-scale gravitational law, making it particularly valuable for constraining the various MG models. Indeed, as we are entering the era of “precision cosmology", multiple spectroscopic and photometric surveys of the LSS, both already operating, such as the DES [@Abbott:2005b] and also about to be commissioned in the next decade, like DESI [@Levi:2013gra], Euclid [@Laureijs:2011gra], the LSST [@Abell:2009aa] and WFIRST [@Spergel:2013tha], will provide us with particularly precise maps of the LSS that will shed light on the mysterious nature of the dark sector. Taking full advantage of this wealth of cosmological information poses a great challenge for experiment and theory alike.
From a theoretical standpoint, models of structure formation rely upon accurately tracing the nonlinear evolution of dark matter perturbations. In the linear regime and when gravity is governed by GR, different modes evolve independently, with the time evolution encapsulated in the scale-independent growth factor. At nonlinear scales, however, the dynamics of the self-gravitating dark matter system can only be tracked accurately by full-blown N-body simulations, which are highly computationally expensive . Additional complexity arises in accounting for the fact that the observed galaxies do not perfectly trace the underlying dark matter density field, but are biased tracers of it [@1984ApJ...284L...9K]. While this effect can be easily captured by introducing a multiplicative bias factor in the large scales [@1988MNRAS.235..715E], in the regime of nonlinear dynamics, empirical modeling needs to be combined with sophisticated simulations in order to predict the spatial distribution of galaxies inside gravitationally collapsed dark matter halos [@0004-637X-575-2-587; @0004-637X-609-1-35; @0004-637X-633-2-791]. Furthermore, when MG configurations are considered, the fifth forces introduce an additional layer of complexity, scale-dependent growth is generally present even at the linear level and in the nonlinear scales, one needs to solve the scalar field Klein-Gordon (KG) equation that is highly nonlinear and adds to the computational costs significantly. The intermediate, quasi-linear, scales can fortunately be analytically accessed by higher order Perturbation Theory (PT) [@Bernardeau:2001qr; @2009PhRvD..80d3531C] approaches or hybrid methods [@Tassev:2013pn; @Valogiannis:2016ane].
The Lagrangian Perturbation Theory approach to structure formation is one the oldest and most popular analytical frameworks in the literature, that has been been particularly successful at describing the Baryon Acoustic Oscillation (BAO) peak [@Eisenstein:2005su], observed at a comoving scale of $\sim$110 Mpc/$h$, and the power-law correlation function, on comoving scales $\sim$20-90 Mpc/$h$, in $\Lambda$CDM [@Vlah:2014nta]. Combined with a model for halo bias [@1984ApJ...284L...9K; @Mo:1996cn; @Sheth:1999mn; @2009JCAP...08..020M; @PhysRevD.88.023515; @Matsubara:2008wx], it can be used to predict the 2-point statistics for halos in the real and redshift space [@2013MNRAS.429.1674C; @Matsubara:2008wx; @Vlah:2016bcl], which serves as a crucial step to the theoretical description of galaxy clustering. Additional contributions from small-scale physics can also be included using techniques inspired by effective field theory theory [@Vlah:2015sea; @Vlah:2016bcl]. In the context of MG cosmologies, extensive studies have been performed in the framework of Eulerian Standard Perturbation Theory (SPT) [@PhysRevD.79.123512; @Taruya:2013quf; @PhysRevD.88.023527; @PhysRevD.90.123515; @PhysRevD.92.063522; @Fasiello:2017bot; @Bose:2017dtl; @Bose:2018zpk]. LPT was first found to work very well within the COLA hybrid framework for chameleon and Vainshtein MG cosmologies [@Valogiannis:2016ane], while third order LPT for dark matter was recently developed in the case of scalar-tensor theories in [@Aviles:2017aor].
In this work, we perform a comprehensive study of how LPT can be used to make predictions for biased tracers in modified gravity theories, and how well the predictions compare with full numerical simulations for a variety of modified gravity models. We study chameleon and Vainsthein MG theories, focusing on the $f(R)$ Hu-Sawicki [@Hu:2007nk] and $n$DGP braneworld models [@Dvali:2000hr] as popular, representative examples for each category. The underlying dark matter clustering is described using the formalism in [@Aviles:2017aor]. We then consider the Convolution Lagrangian Perturbation Theory (CLPT) [@2013MNRAS.429.1674C] for biased tracers, as well as the variants, using the particular resummation scheme in [@Vlah:2015sea; @Vlah:2016bcl]. We extend the peak-background split formalism (PBS) [@1984ApJ...284L...9K; @1986ApJ...304...15B; @PhysRevD.88.023515] in which the Lagrangian bias factors are calculated as responses, to account for modifications to the halo mass function in modified gravity theories and the environmental dependence of screening effects. We show how this formalism should be extended in the case of each screening mechanism. Finally, we cross-check and validate our results in terms of the 2-point statistics against state-of-the-art cosmological N-body simulations that allow us to assess the LPT predictions in the nonlinear, quasi non-linear and baryon acoustic oscillation peak regimes. Comparison with simulations is required to ensure that the predictions will be sufficiently robust for upcoming surveys such as DESI, Euclid, LSST and WFIRST.
We note that a recent paper, submitted to the arXiv while this work was being finalized, also studied biased tracers in MG using CLPT [@Aviles:2018saf] but used a different bias scheme and results were not compared against full simulations.
Our paper is structured as follows: in Section \[sec:Models\], we present the MG models we studied and the details of the N-body simulations employed to test our LPT implementation. We summarize the formalism for developing biased tracer statistics for GR in Section \[sec:LPTDM\]. In Section \[sec:Results:LPT\], we discuss the modifications required to the perturbative schemes to predict the real and Fourier space 2-point statistics and the associated bias parameters for tracers of different masses for models beyond GR. The LPT predictions are compared to statistics derived from simulations in Section \[sec:Results:Compsims\]. Finally, we conclude, and discuss implications for future work, in Section \[sec:conclusions\]. More detail on the derivations in the paper are given in the Appendix.
Formalism {#sec:Formalism}
=========
Modified gravity models {#sec:Models}
-----------------------
### The $f(R)$ model
Despite the latest surge in the field of MG, deformations to GR, together with associated experimental tests, have been around for almost as long as GR itself [@Will:2005va]. One of the oldest attempts consisted of adding a nonlinear function $f(R)$ of the Ricci scalar $R$ to the standard Einstein-Hilbert action, the so-called $f(R)$ theories [@DeFelice:2010aj], with a resulting action $S$ of the form: $$\label{actFr}
S=\int d^4x \sqrt{-g} \left[\frac{R+f(R)}{16 \pi G} + \mathcal{L}_m \right],$$ with $\mathcal{L}_m$ denoting the matter sector Lagrangian and G the gravitational constant. Since such an action frees up an additional degree of freedom, the latest interest in this class of theories comes from the idea that a modification of this type is responsible for cosmic acceleration, rather than dark energy [@Carroll:2003wy].
Such a model is the Hu-Sawicki $f(R)$ model [@Hu:2007nk], which we study in this paper and the functional form of which is given by: $$\label{fRHu}
f(R)=-m^2\frac{c_1\left(R/m^2\right)^n}{c_2\left(R/m^2\right)^n+1}.$$ In equation (\[fRHu\]), $\Omega_{m0}$ denotes the matter fractional energy density and $H_0$ the Hubble Constant, both evaluated today, $m=H_0\sqrt{\Omega_{m0}}$, which has dimensions of mass and $n$, $c_1$ and $c_2$ are free parameters.
In order to match the $\Lambda$CDM expansion history and for sufficiently small values of $|f_{R_0}|$, the background value of the Ricci scalar, $\bar{R}$, becomes equal to: $$\begin{aligned}
\bar{R}= 3 \Omega_m H_0^2 \left(1+ 4 \frac{\Omega_{\Lambda0}}{\Omega_{m0}} \right).\end{aligned}$$ where $\Omega_{\Lambda0}$ is the dark energy fractional density evaluated at the present time.
The derivative $f_{R}=\frac{df(R)}{dR}$ becomes a functional of the cosmological parameters when evaluated today, through the relationship $$\bar{f}_{R_0} =-n\frac{c_1}{c_2^2}\left(\frac{\Omega_{m0}}{3(\Omega_{m0}+\Omega_{\Lambda0})}\right)^{n+1}.$$ The above mapping allows us to reduce the number of free parameters and the Hu-Sawicki model is commonly parametrized by quoting the values chosen for $n$ and $|f_{R_0}|$, with the latter being the background value of the fifth potential evaluated today.
The reason the Hu-Sawicki model is so popular is that it can be cast into the form of a scalar-tensor theory that realizes the chameleon screening mechanism [@PhysRevD.69.044026; @PhysRevLett.93.171104], through a conformal transformation [@Brax:2008hh], with the quantity $f_{R}$ identified as the scalar field that is coupled to matter. Through the interplay between matter and the self-interaction potential, the scalar chameleon field becomes massive near high over-densities and the associated fifth forces get exponentially suppressed due to the Yukawa effect. A stronger screening effect is manifested in lower values of the parameter $|f_{R_0}|$, with the limit of $|f_{R_0}|\rightarrow0$ exactly recovering GR. Following the literature and also because of the available simulations (as we will discuss below in \[sim\]), we choose $n=1$ and consider three different $f(R)$ models with $\abs{\bar{f}_{R_0}}=\{10^{-6},10^{-5}, 10^{-4}\}$, which we shall refer to, from now on, as F6, F5 and F4.
When considering perturbations around a homogeneous and isotropic FRW metric in the conformal Newtonian gauge, the resulting system of the Poisson and KG equations becomes [@Hu:2007nk]: $$\begin{aligned}\label{eq:poisson}
\nabla^2 \Phi_N &= 4 \pi G a^2 \delta \rho_m - \frac{1}{2} \nabla^2 f_R, \\
\nabla^2 f_R &= -\frac{a^2}{3} \delta R - \frac{8 \pi G a^2}{3} \delta \rho_m,
\end{aligned}$$ where $\Phi_N$ is the Newtonian potential and $\delta \rho_m$ the matter density perturbation. $\delta R$, the perturbation to the Ricci scalar, can be written as a function of the scalar field $f_R$, which will play a central role in our screening implementation in Section \[sec:LPTDM\].
### The $n$DGP model {#nDGPsec}
The second MG model under consideration comes from the realm of higher-dimensional braneworld cosmology. The simplest example of such a configuration is the so-called Dvali-Gabadadze-Porrati (DGP) model [@Dvali:2000hr], which is described by an action of the following form: $$\label{actDGP}
S=\int d^4x \sqrt{-g} \left[\frac{R}{16 \pi G} + \mathcal{L}_m \right] + \int d^5x \sqrt{-g_5} \left(\frac{R_5}{16 \pi G r_c}\right).$$ In this model, the spacetime consists of 5 dimensions, rather than the usual 4, but the standard model fields are restricted to a 4-dimensional (4D) brane and the free parameter, $r_c$, denotes the length-scale below which gravity becomes 4D. $R_5$ and $g_5$ denote the corresponding 5D versions of the Ricci scalar and metric tensor determinant, respectively. The resulting Friedman equation from (\[actDGP\]) is: $$\label{FridDGP}
\epsilon \frac{H}{r_c}=H^2 - \frac{\rho_m}{24 \pi G},$$ where $\epsilon=\pm 1$. Each of the two values of $\epsilon$ represents a particular branch of the model, with $\epsilon=+1$ producing the self-accelerating solution, which, however, suffers from undesirable “ghost” instabilities [@Koyama:2007zz] and is thus an unphysical model to consider. The value of $\epsilon=-1$ corresponds to the so-called normal branch, hereafter called $n$DGP, which is well behaved, but does not self-accelerate and can match a $\Lambda$CDM expansion history only in the presence of dark energy.
When focusing on the normal branch and in the quasi-static limit for sub-horizon scales, the perturbations in the conformal Newtonian gauge give the modified Poisson system of equations [@PhysRevD.79.123512]: $$\begin{aligned}\label{eq:poisson_dgp}
\nabla^2 \Phi_N &= 4 \pi G a^2 \delta \rho_m + \frac{1}{2} \nabla^2 \varphi, \\
\nabla^2 \varphi &= \frac{8 \pi G a^2}{3 \beta} \delta \rho_m - \frac{r_c^2}{3 \beta a^2}\Big[(\nabla^2 \varphi)^2- (\nabla_i \nabla_j \varphi)^2 \Big],
\end{aligned}$$ with the coupling $\beta$ given by $$\begin{aligned}
\label{eq:dgpbeta}
\beta(a) = 1 + 2 H(a) r_c \left(1 + \frac{\dot{H(a)}}{3 H(a)^2} \right).\end{aligned}$$
The $n$DGP model is a typical example of a scalar-tensor theory that realizes the Vainshtein screening mechanism [@VAINSHTEIN1972393; @Babichev:2013usa], in which the modifications to gravity are suppressed in the existence of large second derivatives of the scalar field. In the second equation of (\[eq:poisson\_dgp\]), in particular, it can be seen how, once the second derivatives of the scalar field become large in high densities, the second term on the right hand side (r.h.s.) becomes significant and effectively weakens the source strength, resulting in strong screening of the fifth forces.
We consider the $n$DGP model for two choices of the free parameter $r_c$, corresponding to $n\equiv H_0 r_c=1$ and $n=5$, which we shall call, from now on, N1 and N5, respectively.
### N-body Simulations {#sim}
Accurate realizations of structure formation in the nonlinear regime can only be achieved by performing N-body simulations. In this paper, we test our results against two groups of state-of-the-art N-body simulations, that serve complementary purposes to each other, as explained below.
The first group of simulations, to which we shall refer as Group I from now on, are the [ELEPHANT]{} simulations, presented in [@Cautun:2017tkc]. These span the parameter space of both MG models we study: the F4, F5 and F6 $f(R)$ models and N1 and N5 $n$DGP models. The $f(R)$ simulations were performed using the [ECOSMOG]{} code [@1475-7516-2012-01-051; @Bose:2016wms], while the $n$DGP ones [@Hellwing:2017pmj] using the [ECOSMOG-V]{} version [@Li:2013nua; @Barreira:2015xvp], both of which were based on the GR code [RAMSES]{} [@Teyssier:2001cp] and where suitably extended to integrate the scalar field KG equation for the corresponding models using adaptive-mesh-refinement techniques. The parameters describing the background $\Lambda$CDM cosmology are the best-fit ones given by the 9-year WMAP release [@0067-0049-208-2-19] and have the following values: $\Omega_b=0.046$, $\Omega_{cdm}=0.235$, $\Omega_m=0.281$, $\Omega_L=0.719$, $h=0.7$, $n_s=0.971$ and $\sigma_8=0.82$. $N_p=1024^3$ equal mass particles were placed in a simulation box with a side $L_{box}=1024$ Mpc/h and the density field was resolved in a $1024^3$ resolution grid. Furthermore, the simulations were initialized at redshift $z_i=49$ using the Zel’dovich approximation [@Zeldovich:1969sb] and evolved through $z_f=0$.
Gravitationally bound dark matter halos were identified using the [ROCKSTAR]{} halo finder [@2013ApJ...762..109B]. Finally, so as to get an estimate of the variance, each model was simulated for 5 random realizations, corresponding to different random phases in the initial density field.
The $1024^3$ $(Mpc/h)^3$ volume simulations’ results in Group I become noisy at scales $r>100$ Mpc/h. To probe the BAO scales, where LPT has been previously found to perform very well for GR [@Vlah:2014nta], we also test our results against the largest volume $f(R)$ simulations performed to date for the modified gravity lightcone simulation project [@Arnold:2018nmv]. In these simulations, which we will call Group II from now on, the box side is $L_{box}=1536$ Mpc/h with $2048^3$ equal mass particles used, for GR and the $\abs{\bar{f}_{R_0}}$=$10^{-5}$ model. The parameters describing the background $\Lambda$CDM cosmology are the best-fit ones given by the Planck collaboration and have the following values: $\Omega_b=0.0486$, $\Omega_m=0.3089$, $\Omega_L=0.6911$, $h=0.6774$, $n_s=0.9667$ and $\sigma_8=0.8159$. The simulations were performed using the MG code [MG-GADGET]{} [@doi:10.1093/mnras/stt1575], which is a MG extension to the code [P-GADGET3]{}, an improved version of the code [GADGET-2]{} [@2005MNRAS.364.1105S], created for GR simulations. Dark matter halo catalogues were produced using the [SUBFIND]{} algorithm [@2001MNRAS.328..726S]. Each model has been simulated for one random realization.
For more detailed discussions on the N-body implementations, we refer interested readers to the corresponding publications.
Convolution Lagrangian Perturbation Theory for biased tracers in MG {#sec:LPTDM}
-------------------------------------------------------------------
### LPT for dark matter {#LPTdarkmatter}
The Lagrangian Perturbation Theory approach to structure formation has been extensively studied in the context of $\Lambda$CDM scenarios. Opposite to the Eulerian picture, in which one monitors the evolution of the desired quantities at a given, fixed, position, in LPT one instead tracks down the evolution of a given fluid element over time. Starting from an initial, Eulerian, comoving position $\bold{q}$ at a desired early time $t_0$, each mass element is mapped to its comoving Lagrangian position $\bold{x}(\bold{q},t)$ at time $t$, through the relationship $$\label{Lagpos}
\bold{x}(\bold{q},t) = \bold{q} + \bold{\Psi}(\bold{q},t).$$ The Lagrangian displacement $\bold{\Psi}(\bold{q},t)$, taken to be 0 at the initial time $t_0$, is the fundamental quantity of interest in LPT. Furthermore, enforcing mass conservation, through the continuity equation, between the initial and final infinitesimal volume elements centered around $\bold{q}$ and $\bold{x}$, respectively, gives $\rho_{m}(\bold{x},t)d^3x=\rho_{m}(\bold{q},t_0)d^3q$. Assuming $t_{0}$ refers to an epoch early enough that the density perturbations around the background density $\bar{\rho}$ are negligible, meaning $\rho_{m}(\bold{q},t_0)=\bar{\rho}_m$, allows us to obtain the dark matter fractional overdensity, $\delta_m$, in the Lagrangian picture: $$\label{delJac}
1 + \delta_m(\bold{x},t) = \int d^3q \delta_{D}\left[\bold{x}-\bold{q}-\bold{\Psi}(\bold{q},t)\right] = \frac{1}{J(\bold{q},t)},$$ with $\delta_{D}$ being the Dirac delta function and $J(\bold{q},t)$ the determinant of the deformation matrix $$\label{Jacobian}
J_{ij} = \frac{\partial x^i}{\partial q^j} = \delta_{ij} + \frac{\partial \Psi^i}{\partial q^j}.$$ For an irrotational flow, which is a good approximation for cold dark matter and assuming that the gravitational evolution is governed by GR, perturbations around a flat FRW metric give the geodesic and Poisson equations, in the quasi-static approximation and for sub-horizon scales, as: $$\begin{aligned}\label{eq:geopoisson}
\ddot{\bold{x}} + 2H\dot{\bold{x}} &= -\frac{1}{a^2}\nabla_{\bold{x}}\psi(\bold{x},t), \\
\frac{1}{a^2}\nabla^2_{\bold{x}}\psi(\bold{x},t) &= 4 \pi G\bar{\rho}_m \delta(\bold{x},t).
\end{aligned}$$ We should point out that in (\[eq:geopoisson\]) $\psi(\bold{x},t)$ denotes the metric perturbation, which should not be confused with the Lagrangian displacement field $\bold{\Psi}(\bold{q},t)$. In the LPT picture, we perturbatively expand $\bold{\Psi}$ as $$\label{eq:psiexp}
\bold{\Psi}(\bold{q},t) = \sum_{n=1}^{\infty}\bold{\Psi}^{(n)}(\bold{q},t) = \bold{\Psi}^{(1)}(\bold{q},t)+\bold{\Psi}^{(2)}(\bold{q},t)+\bold{\Psi}^{(3)}(\bold{q},t)...$$ and equations (\[Lagpos\])-(\[eq:psiexp\]) form a closed system, that is recursively solved for the various orders of $\bold{\Psi}$. The first order solution is the so-called Zel’dovich approximation [@Zeldovich:1969sb].
In MG theories, as also explained in Section \[sec:Models\], an additional degree of freedom is present, that directly couples to matter and causes particles to deviate from the nominal geodesics of GR. Consequently, equations (\[eq:geopoisson\]) are, in principle, modified for a scalar-tensor theory and so is the LPT framework presented above. In [@Valogiannis:2016ane], the LPT approach was expanded for chameleons and symmetrons, including the first order contribution to the Klein-Gordon equation and was shown to perform very well in the context of the COLA hybrid framework. The LPT approach for scalar-tensor theories up to third order was presented in [@Aviles:2017aor], the main results of which we summarize in this section.
In [@PhysRevD.92.063522], which developed an SPT framework for studying MG theories with a screening mechanism in the nonlinear regime, based on the closure theory approximation in [@Taruya:2007xy], the scalar field KG equation for a Brans-Dicke-like (BD) theory with interactions of a scalar field, $\phi$, was written as: $$\label{BransDicke}
(3 + 2 \omega_{BD})\frac{1}{a^2} k_x^2\phi(\bold{k}_x,t) = 8 \pi G \bar{\rho}_m \delta(\bold{k}_x,t) - \mathcal{I}(\phi),$$ where $\mathcal{I}(\phi)$ denotes the perturbative form of the field self-interaction term: $$\begin{aligned}
\label{interaction}
\mathcal{I}(\phi) &=& M_1(\bold{k},t)\phi \nonumber
\\
&& + \frac{1}{2}\int \frac{d^3 k_1 d^3 k_2}{\left(2 \pi\right)^3} \delta_D(\bold{k}-\bold{k}_{12})M_2(\bold{k}_{1},\bold{k}_{2})\phi(\bold{k}_{1})\phi(\bold{k}_{2})\nonumber\\
&& + \frac{1}{6}\int \frac{d^3 k_1 d^3 k_2 d^3 k_3}{\left(2 \pi\right)^6} \delta_D(\bold{k}-\bold{k}_{123})\nonumber \\&& \ \hspace{1.5cm} \times M_3(\bold{k}_{1},\bold{k}_{2},\bold{k}_{3})\phi(\bold{k}_{1})\phi(\bold{k}_{2})\phi(\bold{k}_{3}),\end{aligned}$$ where we adopted the standard notation $\bold{k}_{ijk}=\bold{k}_{i}+\bold{k}_{j}+\bold{k}_{k}$ and $M_1(\bold{k},t)$, $M_2(\bold{k}_{1},\bold{k}_{2})$ and $M_3(\bold{k}_{1},\bold{k}_{2},\bold{k}_{3})$ are mass terms. The higher order piece in the Fourier space representation of the interaction term in equation (\[interaction\]), incorporates the screening effect, up to third order, that is responsible for recovering GR at small scales. The mapping between a given scalar-tensor theory and the BD form above can be easily performed through assigning appropriate values to the mass terms and the BD coupling $\omega_{BD}$ above, as we will later show for our two models of study.
The perturbed modified Einstein equations have the form [@Aviles:2017aor] $$\begin{aligned}\label{eq:modpoisson}
\nabla_{\bold{x}}\mathcal{\hat{T}}\bold{\Psi} &= -\frac{1}{a^2}\nabla^2_{\bold{x}}\psi(\bold{x},t), \\
\frac{1}{a^2}\nabla^2_{\bold{x}}\psi(\bold{x},t) &= 4 \pi G\bar{\rho}_m \delta(\bold{x},t) -\frac{1}{2 a^2} \nabla^2\phi - \frac{1}{2 a^2}\left(\nabla^2_{\bold{x}}\phi -\nabla^2\phi \right),
\end{aligned}$$ where we use (\[Lagpos\]) and introduce the time derivative operator $\mathcal{\hat{T}}=\frac{d^2}{dt^2} + 2H\frac{d}{dt}$, as in [@Matsubara:2007wj]. The last term in the second line of (\[eq:modpoisson\]), called $\it{frame}$-$\it{lagging}$ in [@Aviles:2017aor], is a geometrical term that occurs due to the fact that, in LPT, the KG equation should be expressed in Lagrangian $\left(\nabla^2\right)$, rather than Eulerian coordinates $\left(\nabla^2_{\bold{x}}\right)$. Taking this into account, equations (\[BransDicke\])-(\[eq:modpoisson\]) are combined to give $$\begin{aligned}\label{eq:finpsi}
& \left(J^{-1}\right)_{ij}\mathcal{\hat{T}}\Psi_{i,j}(\bold{k}) = -A(k)\delta(\bold{k}) + \frac{k^2}{a^2 3\Pi(k)}\delta\mathcal{I}(\bold{k}) \\
& + \frac{M_1(k)}{3\Pi(k)}\frac{1}{2 a^2}\left(\nabla^2_{\bold{x}}\phi -\nabla^2\phi \right)(\bold{k}),
\end{aligned}$$ where, following the definitions in [@PhysRevD.92.063522], we have $$\begin{aligned}\label{eq:sources}
A(k) &= 4 \pi G\bar{\rho}_m\left(1 + \frac{k^2}{a^2 3\Pi(k)}\right), \\
\Pi(k) &= \frac{1}{3a^2}\left[\left(3+2\omega_{BD}\right)k^2 + M_1a^2\right]
\end{aligned}$$ and all the quantities are Fourier transforms in the Lagrangian $\bold{q}$-space. The inverse Jacobean in (\[eq:finpsi\]) reflects the derivative transformation to the $\bold{q}$-space, where the Einstein notation is adopted. Furthermore, $\delta(\bold{k})$, $\delta\mathcal{I}(\bold{k})$ and $\left(\nabla^2_{\bold{x}}\phi -\nabla^2\phi \right)(\bold{k})$ are the Lagragian Fourier transformations of the Lagrangian-transformed overdensity (\[delJac\]), the higher order interaction Kernels in (\[interaction\]) and the frame-lagging Kernel, correspondingly. The expression for the latter is given in [@Aviles:2017aor]. Equation (\[eq:finpsi\]) forms a closed system with (\[delJac\]), (\[Jacobian\]), (\[BransDicke\]) and (\[interaction\]) that is solved, perturbatively, to obtain the MG solution up to various orders in $\bold{\Psi}$, as in (\[eq:psiexp\]).
Solving for the first order solution, one gets [@Valogiannis:2016ane]: $$\label{eq:Zeldisp}
\bold{k}\cdot\bold{\Psi}^{(1)} = i D^{(1)}(k,t)\delta^{(1)}(\bold{k},t=0),$$ which can be easily solved for the displacement field, as: $$\label{eq:Zeldispsol}
{\Psi}^{j}(k,t) = \frac{i k^j}{k^2} D^{(1)}(k,t)\delta^{(1)}(\bold{k},t=0).$$ We see that the r.h.s of (\[eq:Zeldispsol\]) can be conveniently decomposed into a product of the first order density mode at very early times, $\delta^{(1)}(\bold{k},t=0)$, early enough to be gaussian and a space-time dependent growth factor $D^{(1)}(k,t)$, given by: $$\label{growth1st}
\mathcal{\hat{T}}D^{(1)}(k,t)=A(k)D^{(1)}(k,t).$$ In the GR limit, $A(k)=A(k=0)=4 \pi G\bar{\rho}_m$, $D^{(1)}$ becomes scale independent and is nothing else than the first order growing mode for GR; the Zel’dovich approximation.
Moving on to the second order piece, we have: $$\label{eq:Zeldisp2}
\bold{k}\cdot\bold{\Psi}^{(2)} = \frac{i}{2} \int \frac{d^3 k_1 d^3 k_2}{\left(2 \pi\right)^3} \delta_D(\bold{k}-\bold{k}_{12}) D^{(2)}(\bold{k}_1,\bold{k}_2)\delta^{(1)}_1\delta^{(1)}_2,$$ where, for compactness, we adopted the notation $\delta^{(1)}_1=(\bold{k}_1,0)$ and the second order growth factor, $D^{(2)}(\bold{k}_1,\bold{k}_2)$, is given by [@Aviles:2017aor]: $$\begin{aligned}
\label{eq:secgrowth}
D^{(2)}(\bold{k}_1,\bold{k}_2) &=& D^{(2)}_a(\bold{k}_1,\bold{k}_2)-D^{(2)}_b(\bold{k}_1,\bold{k}_2)\frac{\bold{k}_1\cdot\bold{k}_2}{k^2_1k^2_1} \nonumber
\\ && +D^{(2)}_{FL}(\bold{k}_1,\bold{k}_2)-D^{(2)}_{\delta \mathcal{I}}(\bold{k}_1,\bold{k}_2).\end{aligned}$$ The four individual components are given by: $$\begin{aligned}\label{eq:D2sources}
\left(\mathcal{\hat{T}}-A(k)\right)D^{(2)}_a(\bold{k}_1,\bold{k}_2) &= A(k)D^{(1)}(k_1)D^{(1)}(k_2), \\
\left(\mathcal{\hat{T}}-A(k)\right)D^{(2)}_b(\bold{k}_1,\bold{k}_2) &= \left(A(k_1)+A(k_2)-A(k)\right)D^{(1)}(k_1)D^{(1)}(k_2), \\
\left(\mathcal{\hat{T}}-A(k)\right)D^{(2)}_{\delta \mathcal{I}}(\bold{k}_1,\bold{k}_2) &= \left(\frac{2A_0}{3}\right)^2\frac{k^2}{a^2}\frac{M_{2}(\bold{k}_1,\bold{k}_2)D^{(1)}(k_1)D^{(1)}(k_2)}{6\Pi(k)\Pi(k_1)\Pi(k_2)}, \\
\left(\mathcal{\hat{T}}-A(k)\right)D^{(2)}_{FL}(\bold{k}_1,\bold{k}_2) &= \left(\frac{M_1}{3\Pi(k)}\right)K^{(2)}_{FL}(\bold{k}_1,\bold{k}_2)D^{(1)}(k_1)D^{(1)}(k_2).
\end{aligned}$$ The two last terms represent the second order contributions to the growth factor, given by the screening and frame-lagging effects, correspondingly, while the expression for $K^{(2)}_{FL}$ is given in [@Aviles:2017aor]. Despite its lengthier expression, when taking the GR limit we get $D^{(2)}_{FL}=D^{(2)}_{\delta \mathcal{I}}=0$ and $D^{(2)}_{a}=D^{(2)}_{b}$, allowing $D^{(2)}$ to become scale-independent, reducing to the known GR result, which can be well approximated by $D^{(2)}(t)=-\frac{3}{7}\left(D^{(1)}(t)\right)^2$ for $\Lambda$CDM cosmologies [@Bouchet:1994xp].
Solving for the third order piece in (\[eq:finpsi\]) results in a lengthy differential equation for the third order MG growth factor, $D^{(3)}(\bold{k}_1,\bold{k}_2,\bold{k}_3)$, that also needs to be symmetrized. The result is given by equation (\[D3MG\]) in the appendix \[polyderiv\]. It should be also noted that equations (\[growth1st\]), (\[eq:D2sources\]) and f(\[D3MG\]) can be either solved by inverting the linear operator $\left(\mathcal{\hat{T}}-A(k)\right)$ using its Green function, as done in [@Matsubara:2015ipa; @Aviles:2017aor], or by numerically solving the corresponding differential equations. Even though both methods give results that agree with each other well, we chose to proceed with the latter because it is computationally faster. The differential equations were solved using a $5^{th}$ order Runge-Kutta scheme, implemented in $\it{Mathematica}$ [@Mathematica].
We finish this section by showing the particular expressions for the mass terms in (\[interaction\]) and the sources in (\[eq:sources\]) for the $f(R)$ and $n$DGP models we study. For the $f(R)$ Hu-Sawicki model, the Brans-Dicke scalar is simply $\omega_{BD}=0$, while the mass terms are given by the expansion [@PhysRevD.79.123512]: $$\label{eq:fRmasses}
M_{n} = \frac{d^{n}\bar{R}(f_{R_0})}{df_{R_0}^{n}},$$ which, using (\[fRHu\]), gives [@Aviles:2017aor]: $$\begin{aligned}\label{eq:fRmasses2}
M_{1}(a) &= \frac{3 H_0^2}{2 |f_{R_0}|}\frac{\left(\Omega_m a^{-3}+4 \Omega_{\Lambda}\right)^3}{\left(\Omega_m+4 \Omega_{\Lambda}\right)^2},\\
M_{2}(a) &= \frac{9 H_0^2}{4 |f_{R_0}|}\frac{\left(\Omega_m a^{-3}+4 \Omega_{\Lambda}\right)^5}{\left(\Omega_m+4 \Omega_{\Lambda}\right)^4},\\
M_{3}(a) &= \frac{45 H_0^2}{8 |f_{R_0}|}\frac{\left(\Omega_m a^{-3}+4 \Omega_{\Lambda}\right)^7}{\left(\Omega_m+4 \Omega_{\Lambda}\right)^6}.
\end{aligned}$$ In the case of the $n$DGP braneworld model, a similar procedure, informed by (\[eq:poisson\_dgp\]), gives the relevant expressions [@PhysRevD.79.123512; @Aviles:2018saf]: $$\begin{aligned}\label{eq:nDGPmasses}
&M_{1}(a) = 0,\\
&M_{2}(\bold{k}_1,\bold{k}_2,a) = 2\frac{n^2}{H_0^2 a^4}\left(k_1^2k_2^2-(\bold{k_1}\cdot\bold{k_2})^2\right),\\
&M_{3}(\bold{k}_1,\bold{k}_2,\bold{k}_3,a) = 18\frac{n^2 \beta(a)}{H_0^2 a^6 4 \pi G\bar{\rho}_m}\Biggl((\bold{k_1}\cdot\bold{k_2})k_1^2k_3^2+2(\bold{k_1}\cdot\bold{k_2})^2k_3^2\\
& -2(\bold{k_1}\cdot\bold{k_2})(\bold{k_1}\cdot\bold{k_3})(\bold{k_2}\cdot\bold{k_3})-(\bold{k_1}\cdot\bold{k_2})(\bold{k_1}\cdot\bold{k_3})^2\Biggr), \\
\end{aligned}$$ with $\beta$ defined in (\[eq:dgpbeta\]). It is interesting to notice that, even though the interaction term in (\[eq:poisson\_dgp\]) contains only second order derivatives, in (\[eq:nDGPmasses\]) a $3^{rd}$ order mass contribution is now present, that arises when transforming the Eulerian derivatives to the Lagrangian space through (\[Jacobian\]).
### 2-point statistics for biased tracers in GR {#sec:twopointGR}
The perturbative theory of galaxy clustering [@Desjacques:2016bnm], which aims to describe the statistics of biased tracers in the quasi-linear regime, consists of a perturbative description for the evolution of the underlying dark matter density field, combined with an analytical description for the bias parameters at each given order. In the case of cold dark matter, the calculation of the 2-point statistics, even in LPT, is a more straightforward process, since one just needs to plug the ${\bf q}$-space Lagrangian overdensity, mapped to the Eulerian frame through (\[delJac\]), into the common expressions for the autocorrelation function: $$\label{xi}
\xi(r) = \langle \delta_m(\bold{x})\delta_m(\bold{x+r})\rangle$$ and its Fourier space counterpart, the matter power spectrum $$\label{Pk}
\left(2\pi\right)^3\delta_D(\bold{k}+\bold{k'})P(k) = \langle \tilde{\delta}_m(\bold{k})\tilde{\delta}_m(\bold{k'})\rangle.$$ When studying biased tracers, like for example dark matter halos, we need an analytical model to describe their statistical prevalance with respect to the underlying density field. Following [@Matsubara:2008wx; @2013MNRAS.429.1674C], we employ a model of a local in matter density Lagrangian bias in which the positions of biased tracers are purely specified by a distribution of the underlying CDM density field $\delta(\bold{q},t=0)\equiv\delta(\bold{q})$, encoded through a function $F\left[\delta_R(\bold{q})\right]$, as $$\label{biasF}
\rho_{X}(\bold{q}) =\bar{\rho}_{X} F\left[\delta_R(\bold{q})\right],$$ where, consistent with the literature, we use the subscript $X$ to indicate biased tracers. $\delta_R(\bold{q})$ denotes the primordial density field smoothed over some scale $R$, while $\bar{\rho}_{X}$ is the mean density of tracers. Density conservation provides the equivalent of equation (\[delJac\]) for tracers, $$\label{delJacX}
\delta_X(\bold{x},t) = \int d^3q F\left[\delta_R(\bold{q})\right]\delta_{D}\left[\bold{x}-\bold{q}-\bold{\Psi}(\bold{q},t)\right] -1.$$ This model of local Lagrangian bias, which corresponds to a non-local bias in the Eulerian space, can be extended to include a biasing scheme that is non-local in the Lagrangian space [@PhysRevD.83.083518]. Combining (\[delJacX\]) and (\[xi\]) and after some transformations one gets the general expression for the 2-point correlation function for biased tracers in LPT, $$\label{xiX}
1+\xi_{X}(r) = \int d^3q\int \frac{d^3k}{\left(2 \pi\right)^3}e^{i\bold{k}\cdot(\bold{q}-\bold{r})} \int \frac{d\lambda_1}{\left(2 \pi\right)}\frac{d\lambda_2}{\left(2 \pi\right)} L(\bold{q},\bold{k},\lambda_1,\lambda_2),$$ with $$\label{LX}
L(\bold{q},\bold{k},\lambda_1,\lambda_2)=\tilde{F}_1 \tilde{F}_2\underbrace{\langle e^{i\left[\lambda_1 \delta_1+\lambda_2 \delta_2+\bold{k}\cdot(\bold{\Delta}))\right]} \rangle}_\text{$K(\bold{q},\bold{k},\lambda_1,\lambda_2)$},$$ where $\tilde{F}_1$, $\tilde{F}_2$ are the Fourier space representations of $F$, with corresponding wavemodes $\lambda_1$ and $\lambda_2$ and $\bold{\Delta}=\bold{\Psi}_2-\bold{\Psi}_1$. The notation $\delta(\bold{q}_1)\equiv \delta_1$ has been adopted for all quantities. The ensemble average K in (\[LX\]) can be cast into an exponent of a power series in cumulants, through the cumulant expansion theorem, $\langle e^{iX} \rangle = \exp\left[\sum_{N=1}^{\infty}\frac{i^N}{N!} \langle X^{N} \rangle_c \right]$, which, combined with a multinomial expansion, gives: $$\label{multinom}
K(\bold{q},\bold{k},\lambda_1,\lambda_2)=\exp \left[\sum \frac{i^{n+m+r}}{m!n!r!}\lambda_1^m\lambda_2^n k_{i_1}..k_{i_r}\langle \delta_1^{m} \delta_1^{n} \Delta_{i_1}..\Delta_{i_r}\rangle_c\right],$$ in terms of a series of correlators $$\begin{aligned}\label{eq:correl}
\sigma_R^2 &= \langle \delta^2 \rangle_c \\
\xi_L(\vec{q}) &= \langle \delta_1 \delta_2 \rangle_c, \\
A_{ij}^{m n}(\vec{q}) &= \langle \delta_i^{m} \delta_j^{n} \Delta_i \Delta_j\rangle_c, \\
W_{ijk}^{m n}(\vec{q}) &= \langle \delta_i^{m} \delta_j^{n} \Delta_i \Delta_j \Delta_k\rangle_c, \\
U_{i}^{m n}(\vec{q}) &= \langle \delta_1^{m} \delta_2^{n} \Delta_i \rangle_c,
\end{aligned}$$ where we adopted the commonly used notation for the Lagrangian cumulants in (\[eq:correl\]).
Keeping all terms in (\[multinom\]) that contain cumulants up to third order, which is the equivalent of the one-loop correction to the linear power spectrum, results in a highly oscillatory integrand that presents challenges when ensuring the integral is fully converged. [@Matsubara:2007wj] proposed expanding all contributions to the exponent but the scale-independent “zero-lag" piece of $A_{ij}^{0 0}$, which results in a non-perturbative resummation scheme that is simpler to handle analytically. Building upon this result, [@2013MNRAS.429.1674C] proposed keeping all the terms of $A_{ij}^{0 0}$ in the exponent, in their Convolution Lagrangian Perturbation Theory (CLPT) scheme. Keeping only the linear component of $A_{ij}^{0 0}$ exponentiated, as done in [@Vlah:2015sea; @Vlah:2016bcl], and performing the $\lambda$ and $k$ integrations in (\[xiX\]) gives a CLPT expression for the 2-point real space correlation function, $$\begin{aligned}
\label{xiXfinal}
1 + \xi_{X}(r) &=& \int d^3q \frac{e^{-\frac{1}{2} (q_i - r_i) (A^{-1}_{L})_{ij}(q_j - r_j)}} {\left(2\pi\right)^{3/2} |A_{L}|^{1/2}} \times \Biggl( 1 - \frac{1}{2}G_{ij} A^{loop}_{ij} \nonumber\\
&& + \frac{1}{6}\Gamma_{ijk} W_{ijk}- b_1 \left(2U_i g_i + A^{10}_{ij}G_{ij}\right) \nonumber \\
&&-b_2\left(U^{(1)}_iU^{(1)}_jG_{ij}+U^{20}_ig_i\right)\nonumber\\
&&+b_1^2\left(\xi_L-U^{(1)}_iU^{(1)}_jG_{ij}-U^{11}_ig_i\right) \nonumber\\
&& +\frac{1}{2}b_2^2\xi_L^2 -2b_1b_2\xi_LU^{(1)}_ig_i \Biggr),\end{aligned}$$ with $$\begin{aligned}
\label{determs}
g_i &\equiv& (A^{-1}_{L})_{ij}(q_j - r_j), \nonumber \\
G_{ij} &\equiv& (A^{-1}_L)_{ij}-g_ig_j, \nonumber\\
\Gamma_{ijk} &\equiv& (A^{-1}_{L})_{ij} g_k +(A^{-1}_{L})_{ki} g_j +(A^{-1}_{L})_{jk} g_i - g_i g_j g_k. \end{aligned}$$ Furthermore, in (\[xiXfinal\]) we define $U^{10}_i\equiv U_i$, $W^{000}_{ijk}\equiv W_{ijk}$ and use superscript numbers in brackets to indicate the various orders of contribution. The $1^{st}$ and $2^{nd}$ order Lagrangian bias factors, $b_1$ and $b_2$, are the expectation values of the $1^{st}$ and $2^{nd}$ order derivatives of the Lagrangian bias function $F$, respectively, through the identity [@Matsubara:2008wx; @PhysRevD.83.083518], $$\begin{aligned}\label{biasfn}
b_n\equiv \int \frac{d \lambda}{2\pi}\tilde{F}(\lambda)e^{-\frac{1}{2}\lambda^2 \sigma^2_R}\left(i \lambda \right)^n = \bigg \langle \frac{d^nF}{d\delta^n} \bigg \rangle.
\end{aligned}$$ In the case of dark matter, we have $F=1$ and $\tilde{F}(\lambda)=2\pi \delta_D(\lambda)$ [@2013MNRAS.429.1674C], and we recover $b_1=b_2=0$ for the unbiased, dark matter distribution.
The Fourier transform gives the CLPT power spectrum for biased tracers [@Vlah:2015sea; @Vlah:2016bcl]: $$\begin{aligned}
\label{PkXfinal}
P_{X}(k) &=& \int d^3q e^{i\bold{k}\cdot\bold{q}}e^{-\frac{1}{2} k_ik_j A^{L}_{ij}} \times \Biggl( 1 - \frac{1}{2}k_ik_j A^{loop}_{ij} - \frac{i}{6}k_ik_jk_k W_{ijk} \nonumber\\
&&+ b_1 \left(2i k_i U_i - k_ik_jA^{10}_{ij}\right) + b_2\left(i k_i U^{20}_i - k_ik_jU^{(1)}_iU^{(1)}_j\right) \nonumber\\
&&+b_1^2\left(\xi_L + ik_iU^{11}_i -k_ik_jU^{(1)}_iU^{(1)}_j\right)\nonumber \\
&& +\frac{1}{2}b_2^2\xi_L^2 + 2b_1b_2\xi_L i k_iU^{(1)}_i\Biggr).\end{aligned}$$ In addition to the one-loop expressions for the two-point statistics, we also calculate the Zel’dovich ($1^{st}$ order LPT) approximation [@White:2014gfa] for biased tracers in the configuration and Fourier space, which can be identified as the subset of terms in (\[xiXfinal\]) and (\[PkXfinal\]) that are linear in $P_{L}(k)$. These are the terms that depend on combinations of $\xi_L$, $U^{(1)}$ and $A^{L}_{ij}$.
While, for GR, CLPT does a very good job at modeling the configuration space $\xi(r)$, it is known to perform less well in reconstructing clustering in the Fourier space [@Vlah:2014nta]. Expanding the resummed exponent in (\[PkXfinal\]) and performing the resulting integrals gives the Eulerian one-loop Standard Perturbation Theory (SPT) power spectrum for biased tracers in GR [@Matsubara:2008wx], $$\begin{aligned}
\label{PkXSPTGR}
P^{SPT}_{X}(k) &=& \left(1-k^2 \sigma_L^2\right)\left(1+b_1\right)^2P_{L}(k) + \frac{9}{98}Q_1(k) + \frac{3}{7}Q_2(k) \nonumber \\
&&+ \frac{1}{2}Q_3(k)+ b_1\left(\frac{6}{7}Q_5(k)+2Q_7(k)\right) + b_2\left(\frac{3}{7}Q_8(k)+Q_9(k)\right)\nonumber\\
&& + b_1^2 \left(Q_9(k)+Q_{11}(k)\right) + 2b_1b_2Q_{12}(k) + \frac{1}{2}b_2^2 Q_{13}(k) \nonumber\\
&&+ \frac{6}{7}\left(1+b_1\right)^2\left[R_1(k)+R_2(k)\right] -\frac{8}{21}\left(1+b_1\right)R_1(k),\end{aligned}$$ where $$\label{sigmalin}
\sigma_L^2=\frac{1}{6 \pi^2}\int_0^{\infty}dkP_L(k),$$ is the 1D variance of the Lagrangian displacement and the functions $Q_n$ and $R_n$ were defined in [@Matsubara:2008wx] for GR. The SPT power spectrum has been shown to follow the power spectrum much better than the CLPT prediction in GR [@Vlah:2014nta].
The correlation function obtained from Fourier transforming (\[PkXSPTGR\]) is, unfortunately, known to be ill-behaved [@Matsubara:2007wj]. However, if one performs an alternative resummation proposed in [@Matsubara:2007wj; @Matsubara:2008wx], known as Lagrangian Resummation Theory (LRT), the resulting power spectrum, $$\begin{aligned}
\label{PkXLRTGR}
P^{LRT}_{X}(k) &=& e^{-k^2 \sigma_L^2} \Biggl[\left(1+b_1\right)^2P_{L}(k) + \frac{9}{98}Q_1(k) + \frac{3}{7}Q_2(k) + \frac{1}{2}Q_3(k)\nonumber \\
&&+ b_1\left(\frac{6}{7}Q_5(k)+2Q_7(k)\right) + b_2\left(\frac{3}{7}Q_8(k)+Q_9(k)\right)\nonumber\\
&& + b_1^2 \left(Q_9(k)+Q_{11}(k)\right) + 2b_1b_2Q_{12}(k)\nonumber \\
&& + \frac{1}{2}b_2^2 Q_{13}(k)+ \frac{6}{7}\left(1+b_1\right)^2\left[R_1(k)+R_2(k)\right] \nonumber \\
&&-\frac{8}{21}\left(1+b_1\right)R_1(k) \Biggr],\end{aligned}$$ which differs from (\[PkXSPTGR\]) only by the exponential prefactor, can be Fourier transformed to the configuration space and is found to characterize the BAO scales well for both dark matter [@Matsubara:2007wj; @Vlah:2014nta] and biased tracers [@Matsubara:2008wx; @2013MNRAS.429.1674C]. It decays sharply for large values of $k$ however.
### Calculation of bias parameters in GR {#sec:biasparametersGR}
In this section, we present an analytical model for the calculation of the bias parameters (\[biasfn\]) in GR, which will be extended to include MG in section \[sec:biasparameters\]. It should be noted though, that even in the complete absence of an analytical model, one or both of the bias parameters in CLPT can be treated as free parameters, to be fitted over simulations, as for example done in [@2013MNRAS.429.1674C; @doi:10.1111/j.1365-2966.2011.19379.x]. With regards to analytical models for bias, arguably the most popular one is the halo approach [@Mo:1995cs; @Mo:1996cn; @Sheth:1998xe; @Sheth:1999mn; @Ma:2000ik; @Peacock:2000qk; @Scoccimarro:2000gm; @Cooray:2002dia], that is based on the extended Press-Schechter (PS) formalism [@1974ApJ...187..425P; @1991ApJ...379..440B], in combination with the Peak-Background Split (PBS) approach [@1984ApJ...284L...9K]. A discussion of the accuracy of such approaches can be found in [@2010MNRAS.402..589M; @Hoffmann:2015mma]. In what follows, we briefly summarize the main ingredients of this prescription in GR.
Let $M_0$ be the mass of a collapsed region at a redshift of interest $z$, that is enclosed in a Lagrangian region of radius $R_0$, which, given the mean matter density $\rho_{m0}$, will be given by $$\label{radiusdef}
R_0 = \left(\frac{3 M_0}{4 \pi \rho_{m0}} \right)^{\frac{1}{3}}.$$ The variance of matter density fluctuations in this region is, $$\label{variance}
\sigma^2(M_0) = \int \frac{dk k^2}{2 \pi^2} W^2\left(k R_0\right) P_L(k,z=0),$$ with $P_L(k,z=0)$ the linear matter power spectrum evaluated today and $W\left(k R_0\right)$ the top-hat smoothing Kernel, $$\label{tophat}
W\left(k R_0\right) = \frac{3\left[\sin(k R_0)-k R_0\cos(k R_0)\right]}{\left(k R_0\right)^3}.$$ For GR, density perturbations are evolved in time, relative to present time using the GR linear growth factor $D(z)$.
Based on the PS theory and its variants [@1974ApJ...187..425P; @1991ApJ...379..440B], the comoving mean number density of halos per logarithmic mass bin $d\ln M$, $\bar{n}_h$, can be analytically modeled as: $$\label{PSfunction}
\bar{n}_h(M) = \frac{\partial^2\bar{N}_h}{\partial V \partial \ln M} = \frac{\bar{\rho}_m}{M}\nu_c(M)f\left[\nu_c(M)\right] \frac{d\ln \nu_c(M)}{dM},$$ where $\bar{N}_h$ is the mean number of halos with mass $M$, in a bin of width $dM$, enclosed in a comoving volume $V$.
The quantity $\nu_c(M)$, the peak significance, is given by $$\label{peak}
\nu_c(M)=\frac{\delta_{cr}}{\sigma(M,z)}=\frac{\delta_{cr}}{D(z)\sigma(M)}.$$ where $D(z)$ is the linear growth factor at the time of collapse $z$, normalized so that $D(z=0)=1$. In (\[peak\]), $\sigma(M,z)=D(z)\sigma(M)$ is the variance at redshift $z$, with $\sigma(M)$ the variance ($\ref{variance}$) evaluated today and $\delta_{cr}$ is the critical overdensity for collapse at redshift $z$. For an Einstein De-Sitter (EDS) cosmology, the latter is always $\delta_{cr}=1.686$, which turns out to be a very good approximation for $\Lambda$CDM cosmologies and will be adopted here.
$f\left[\nu_c(M)\right]$ is the multiplicity function, that in the original PS theory is given by: $$\label{psmult}
\nu_cf\left[\nu_c\right]=\sqrt{\frac{2}{\pi}}\nu_ce^{\frac{-\nu_c^2}{2}}.$$ The prescription (\[PSfunction\]), often referred to as the universal mass function, is exact in an EDS universe with a power law power spectrum. While (\[psmult\]) has been used to describe the halo mass function for a broad range of cosmologies, it lacks the necessary accuracy for precision predictions. For this reason, Sheth and Tormen (ST) [@Sheth:1999mn], introduced an alternative function: $$\label{stmult}
\nu_cf\left[\nu_c\right]=\sqrt{\frac{2}{\pi}}A(p)\left[1+\frac{1}{(q \nu_c^2)^p}\right]\sqrt{q}\nu_ce^{\frac{-q\nu_c^2}{2}},$$ where $A(p)=\left[1+\pi^{-\frac{1}{2}}2^{-p}\Gamma(0.5-p)\right]^{-1}$ and $q,p$ are free parameters that can be fitted over N-body simulations. The best fit pair was initially proposed to be $(q,p)=(0.707,0.3)$ which was later updated to $(q,p)=(0.75,0.3)$. These are considered to be the “standard" ST parameters [@Sheth:1999mn; @Sheth:1999su]. For the PS function, $q=1,p=0$.
Based on the PBS argument [@1984ApJ...284L...9K], a large-wavelength density perturbation $\Delta$ (that is effectively constant on small scales) has the same effect on the formation of biased tracers as a modification to the mean background density by this offset. If by $\bar{n}_h(M, \Delta)$ we denote the halo mass function’s response to such a perturbation, also sometimes called the conditional mass function, then the fractional overdensity of halos will be given by [@Mo:1995cs; @Mo:1996cn; @Sheth:1998xe]: $$\label{halodensity}
1 + \delta_h(M)= \frac{\bar{n}_h(M,\Delta)}{\bar{n}_h(M,0)},$$ where $\bar{n}_h(M,0)=\bar{n}_h(M)$, is the standard, unconditional halo mass function. It also worth noticing that equation (\[halodensity\]) also defines $F\left[\delta_R(\bold{q})\right]$, through (\[biasF\]). The Lagrangian bias of order $n$, is then given by $$\label{biasrig}
b_{n}^L(M)= \frac{1}{\bar{n}_h(M,0)}\frac{d^n\bar{n}_h(M,\Delta)}{d \Delta^n}\Biggr|_{\substack{\Delta=0}},$$ where the time argument in the above is assumed and omitted for simplicity. Equation (\[biasrig\]) is the rigorous definition of the bias parameters, that is exact even in the absence of an analytical description for the halo mass function and can be calculated numerically, for example employing separate universe simulations [@PhysRevD.88.023515]. In the presence of a universal mass function, the conditional mass function is given by the same expression (\[PSfunction\]), but with the modified peak significance $\nu'_c(M)=\frac{\delta_{cr}-\Delta}{D(z)\sigma(M)}$, in which case the the bias factors are easily calculated [@Sheth:1998xe] by (\[halodensity\]), as: $$\label{biasgeneral}
b_{n}^L(M)= \frac{(-1)^n}{\left(D(z)\sigma(M)\right)^n}\frac{1}{\nu_cf\left[\nu_c\right]}\frac{d^n \left(\nu_cf\left[\nu_c\right]\right)}{d\nu_c^n},$$ which are the so-called PBS Lagrangian biases. Applied on the ST function (\[stmult\]), the PBS biases are: $$\begin{aligned}\label{pbsbiasGR}
b_{1}^L (M)&= \frac{-1}{\delta_{cr}}\left[q\nu_c^2-1+\frac{2p}{1+\left(q\nu_c^2\right)^p}\right], \\
b_{2}^L (M)&= \frac{1}{\delta^2_{cr}}\left[q^2\nu_c^4-3q\nu_c^2+\frac{2p\left(2q\nu_c^2+2p-1\right)}{1+\left(q\nu_c^2\right)^p}\right].\\
\end{aligned}$$ These PBS biases are identical to the Lagrangian bias factors defined within the context of CLPT through (\[biasfn\]) [@Matsubara:2008wx; @PhysRevD.83.083518].
Results {#sec:Results}
=======
Lagrangian Perturbation Theory for Biased Tracers in MG {#sec:Results:LPT}
-------------------------------------------------------
### 2-point statistics for Biased Tracers in MG {#sec:twopointMG}
In section \[sec:twopointGR\], we showed the expressions for the calculation of 2-point statistics in CLPT and its variants, under the assumption that gravitational evolution is governed by GR. Here we explain how each of these relationships have to be modified in the case of MG theories. We note that these results are consistent with those recently presented in [@Aviles:2018saf].
The two-point statistics for biased tracers in MG are given, as in GR, by the definitions \[xi\] and \[Pk\], in the configuration and Fourier space, respectively. Considering biased tracers in the Lagrangian space, through (\[biasF\]), the overdensity of biased tracers in LPT is given by $$\label{delJacXMG}
\delta_X(\bold{x},t) = \int d^3q F\left[\delta_R(\bold{q})\right]\delta_{D}\left[\bold{x}-\bold{q}-\bold{\Psi}(\bold{q},t)\right] -1,$$ where we used density conservation. The above equation is similar to (\[delJacX\]) for GR, but differs in that the Lagrangian field $\bold{\Psi}(\bold{q})$ follows the MG evolution presented in Section \[LPTdarkmatter\]. In particular, if we work in terms of the Fourier transform of $\bold{\Psi}(\bold{q})$, labeled as $\tilde{\bold{\Psi}}(\bold{p})$, the $n^{th}$ order LPT solutions in MG will be given by, $$\begin{aligned}
\label{psifourmain}
\tilde{\Psi}^{(n)}_j(\bold{p}) &=& \frac{i}{n!}\int\frac{d^3p_1}{(2\pi)^3}..\frac{d^3p_{n}}{(2\pi)^3} \delta_D^3\left(\sum_{j=1}^{n}p_j-p\right) \nonumber \\
&& \times \ L_j^{(n)}(\bold{p}_1,..,\bold{p}_n)\tilde{\delta}_{L}(\bold{p}_1)..\tilde{\delta}_{L}(\bold{p}_n), \end{aligned}$$ where $\tilde{\delta}_{L}(\bold{p}_n)$ are the linear-density Fourier transformed fields at the time of evaluation and the Kernels $ L_j^{(n)}(\bold{p}_1,..,\bold{p}_n)$ are given by [@Aviles:2017aor]: $$\begin{aligned}\label{LPTKernelsmain}
L_j^{(1)}(\bold{p}) &= \frac{p^j}{p^2}, \\
L_j^{(2)}(\bold{p}_1,\bold{p}_2) &= \frac{p^j}{p^2} \frac{D^{(2)}(\bold{p_1},\bold{p_2})}{D^{(1)}(p_1)D^{(1)}(p_2)}, \\
(L_j^{(3)})_{sym}(\bold{p}_1,\bold{p}_2,\bold{p}_3) &= i\frac{p^j}{p^2} \frac{D^{(3)}_{sym}(\bold{p_1},\bold{p_2},\bold{p}_3)}{D^{(1)}(p_1)D^{(1)}(p_2)D^{(1)}(p_3)}.
\end{aligned}$$ The MG growth factors in (\[LPTKernelsmain\]), are the ones given by (\[growth1st\]), (\[eq:secgrowth\]) and (\[D3MG\]). Plugging the overdensity (\[delJacXMG\]) into (\[xi\]) and (\[Pk\]) and working as in (\[xiX\])-(\[eq:correl\]), we arrive at equations (\[xiXfinal\]) and (\[PkXfinal\]), that give the 2-point statistics for biased tracers in MG and depend on the MG Lagrangian correlators $$\begin{aligned}\label{eq:correlMG}
\sigma_R^2 &= \langle \delta^2 \rangle_c \\
\xi_L(\vec{q}) &= \langle \delta_1 \delta_2 \rangle_c, \\
A_{ij}^{m n}(\vec{q}) &= \langle \delta_i^{m} \delta_j^{n} \Delta_i \Delta_j\rangle_c, \\
W_{ijk}^{m n}(\vec{q}) &= \langle \delta_i^{m} \delta_j^{n} \Delta_i \Delta_j \Delta_k\rangle_c, \\
U_{i}^{m n}(\vec{q}) &= \langle \delta_1^{m} \delta_2^{n} \Delta_i \rangle_c.
\end{aligned}$$ For the MG correlators (\[eq:correlMG\]), we use the same definition and index structure as in GR, but these functions differ from their GR counterparts, because the quantities in the cumulants follow the MG LPT solutions (\[LPTKernelsmain\]). In particular, plugging (\[psifourmain\]) and (\[LPTKernelsmain\]) into (\[eq:correlMG\]), we get the Lagrangian correlators in MG: $$\begin{aligned}
\label{qfuncsmain}
V_1^{(112)}(q)&=&\frac{1}{2\pi^2}\int \frac{dk}{k} \left(-\frac{3}{7}\right)\left[R_1(k)\right]_{MG} j_1(kq), \nonumber\\
V_3^{(112)}(q)&=&\frac{1}{2\pi^2}\int \frac{dk}{k} \left(-\frac{3}{7}\right)\left[Q_1(k)\right]_{MG} j_1(kq), \nonumber\\
S^{(112)}(q)&=&\frac{3}{14\pi^2}\int \frac{dk}{k} \left[2\left[R_1\right]_{MG}+4R_2 +\left[Q_1\right]_{MG} +2Q_2\right]\frac{j_2(kq)}{kq}, \nonumber\\
T^{(112)}(q)&=&\frac{-3}{14\pi^2}\int \frac{dk}{k} \left[2\left[R_1\right]_{MG}+4R_2 +\left[Q_1\right]_{MG} +2Q_2\right]j_3(kq), \nonumber\\
U^{(1)}(q)&=&\frac{1}{2\pi^2}\int dk k \left(-1\right)P_L(k) j_1(kq), \nonumber\\
U^{(3)}(q)&=&\frac{1}{2\pi^2}\int dk k \left(-\frac{5}{21}\right)R_1(k) j_1(kq), \nonumber\\
U_{20}^{(2)}(q)&=&\frac{1}{2\pi^2}\int dk k \left(-\frac{6}{7}\right)Q_8(k) j_1(kq), \nonumber\\
U_{11}^{(2)}(q)&=&\frac{1}{2\pi^2}\int dk k \left(-\frac{6}{7}\right)\left[R_1(k)+R_2(k)\right]_{MG} j_1(kq), \nonumber\\
X_{10}^{(12)}(q)&=&\frac{1}{2\pi^2}\int dk \frac{1}{14}\Biggl(2\left(\left[R_1\right]_{MG}-R_2(k)\right) +3\left[R_1\right]_{MG} j_0(kq) \nonumber\\
-&3&\left[ \left[R_1\right]_{MG} + 2R_2+2\left[R_1(k)+R_2(k)\right]_{MG} +2Q_5 \right]\frac{j_1(kq)}{kq} \Biggr), \nonumber\\
Y_{10}^{(12)}(q)&=&\frac{1}{2\pi^2}\int dk \left(-\frac{3}{14}\right)\Biggl( \left[R_1\right]_{MG} + 2R_2 \nonumber\\
&& +2\left[R_1(k)+R_2(k)\right]_{MG} +2Q_5\Biggr)\times \left[j_0(kq)-3\frac{j_1(kq)}{kq}\right] , \nonumber\\
X(q)&=&\frac{1}{2\pi^2}\int dk\ a(k) \left[\frac{2}{3}-2\frac{j_1(kq)}{kq}\right], \nonumber\\
Y(q)&=&\frac{1}{2\pi^2}\int dk\ a(k) \left[-2j_0(kq)+6\frac{j_1(kq)}{kq}\right], \end{aligned}$$ where we additionally performed the decompositions $$\begin{aligned}
A_{ij}^{mn}(q)&=X_{mn}(q)\delta_{ij}+Y_{mn}\hat{q}_{i}\hat{q}_{j}\\
W_{ijk}(q)&=V_{1}(q)\hat{q}_{i}\delta_{jk}+V_{2}(q)\hat{q}_{j}\delta_{ki}+V_{3}(q)\hat{q}_{k}\delta_{ij}+T(q)\hat{q}_{i}\hat{q}_{j}\hat{q}_{k}
\end{aligned}$$ and defined $a(k)=P_L(k)+\frac{9}{98}Q_1(k)+\frac{10}{21}R_1(k)$. The functions $Q_n(k)$ and $R_n(k)$ that appear in the r.h.s of (\[qfuncsmain\]), are defined, as in GR, to be: $$\begin{aligned}
Q_n(k) &=& \frac{k^3}{4 \pi^2 }\int_0^{\infty} drP_L(kr) \nonumber \\
&& \times \int_{-1}^{1}dx P_L(k\sqrt{1+r^2-2rx})\tilde{Q}_n(r,x) \nonumber
\\
\label{QRformmain}
R_n(k) &= &\frac{k^3}{4 \pi^2 }P_L(k) \int_0^{\infty} drP_L(kr)\int_{-1}^{1}dx \tilde{R}_n(r,x).\end{aligned}$$ The scale and redshift dependency of the growth factors alters the evaluation of these expressions relative to GR: $$\begin{aligned}\label{Qsmain}
&\tilde{Q}_1 = r^2\left(\bar{D}_a^{(2)}-\bar{D}_b^{(2)}\frac{x^2+r^2-2rx}{1+r^2-2rx}+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right)^2 \\
&\tilde{Q}_2 = \frac{rx(1-rx)}{1+r^2-2rx}\left(\bar{D}_a^{(2)} - \bar{D}_b^{(2)}\frac{x^2+r^2-2rx}{1+r^2-2rx}+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right) \\
&\tilde{Q}_3 = \frac{x^2(1-rx)^2}{(1+r^2-2rx)^2} \\
&\tilde{Q}_5 = rx\left(\bar{D}_a^{(2)}-\bar{D}_b^{(2)}\frac{x^2+r^2-2rx}{1+r^2-2rx}+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right) \\
&\tilde{Q}_7 = \frac{x^2(1-rx)}{(1+r^2-2rx)} \\
&\tilde{Q}_8 = r^2\left(\bar{D}_a^{(2)}-\bar{D}_b^{(2)}\frac{x^2+r^2-2rx}{1+r^2-2rx}+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right) \\
&\tilde{Q}_9 = \frac{rx(1-rx)}{1+r^2-2rx} \\
&\tilde{Q}_{11} = x^2 \\
&\tilde{Q}_{12} = rx \\
&\tilde{Q}_{13} = r^2 \\
&\left[\tilde{Q}_1\right]_{MG} = \frac{r^2 (1-x^2)}{1+r^2-2rx}\left(\bar{D}_a^{(2)}-\bar{D}_b^{(2)}\frac{x^2+r^2-2rx}{1+r^2-2rx}+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right).\\
\end{aligned}$$ and $$\begin{aligned}\label{Rsmain}
&\tilde{R}_{1} = \frac{21}{10} r^2 \frac{D^{(3)}_{sym}(\bold{k},-\bold{p},\bold{p})}{D^{(1)}(\bold{k})\left(D^{(1)}(\bold{p})\right)^2} \\
&\tilde{R}_{2} = \frac{rx(1-rx)}{1+r^2-2rx}\left(\bar{D}_a^{(2)}-\bar{D}_b^{(2)}x^2+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right) \\
&\left[\tilde{R_1}(k)+\tilde{R_2}(k)\right]_{MG} = \frac{r^2(1-rx)}{1+r^2-2rx}\left(\bar{D}_a^{(2)}-\bar{D}_b^{(2)}x^2+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right)\\
&\left[\tilde{R}_1\right]_{MG} = \frac{r^2(1-x^2)}{1+r^2-2rx}\left(\bar{D}_a^{(2)}-\bar{D}_b^{(2)}x^2+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right).\\
\end{aligned}$$ The functions $Q_1$-$Q_{13}$ in (\[Qsmain\]) and $R_1$, $R_{2}$ in (\[Rsmain\]) differ from GR as they depend on the MG growth factors (\[growth1st\]), (\[eq:secgrowth\]) and (\[D3MG\]). In addition, the functions $\left[Q_1\right]_{MG}$, $\left[R_1\right]_{MG}$ and $\left[R_1(k)+R_2(k)\right]_{MG}$ are new ones that arise in MG. In the GR limit, that corresponds to $\bar{D}_a^{(2)}=\bar{D}_b^{(2)}=1$ and $\bar{D}_{FL}^{(2)}=\bar{D}_{\delta \mathcal{I}}^{(2)}=0$, $\left[Q_1\right]_{MG}=Q_1$, $\left[R_1\right]_{MG}=R_1$, $\left[R_1(k)+R_2(k)\right]_{MG}=R_1(k)+R_2(k)$ and the functions $Q_n$ and $R_n$ reduce to their GR expressions in [@Matsubara:2007wj]. In that limit, furthermore, the correlators (\[qfuncsmain\]) recover their GR forms presented in [@2013MNRAS.429.1674C]. The derivations of the above, along with a more detailed discussion, are presented in Appendices \[polyderiv\] and \[correlderiv\].
In Figure \[CLPTstuff\], we show the contributions of the different terms in (\[xiXfinal\]) as a function of $r$ for the F4 model (which predicts the largest deviations from GR), evaluated at $z=0.5$.
As in the GR case, proceeding to expand the resummed exponent in (\[PkXfinal\]) and performing the resulting integrals, as shown in Appendix \[spectraderiv\], gives the equivalent of the Eulerian one-loop SPT power spectrum for biased tracers in MG: $$\begin{aligned}\label{PkXSPT}
&P^{SPT}_{X}(k) = \\
& \left(1-k^2 \sigma_L^2\right)\left(1+b_1\right)^2P_{L}(k) + \frac{9}{98}Q_1(k) + \frac{3}{7}Q_2(k) + \frac{1}{2}Q_3(k) \\
&+ b_1\left(\frac{6}{7}Q_5(k)+2Q_7(k)\right) + b_2\left(\frac{3}{7}Q_8(k)+Q_9(k)\right)\\
& + b_1^2 \left(Q_9(k)+Q_{11}(k)\right) + 2b_1b_2Q_{12}(k) \\
& + \frac{1}{2}b_2^2 Q_13(k)+ \frac{6}{7}\left(b_1^2+b_1\right)\left[R_1(k)+R_2(k)\right]_{MG} \\
&+ \frac{6}{7}\left(1+b_1\right)\left[R_1(k)+R_2(k)\right] -\frac{8}{21}\left(1+b_1\right)R_1(k).
\end{aligned}$$ Equation (\[PkXSPT\]), that depends on the functions (\[QRformmain\]), is the MG version of (\[PkXSPTGR\]). Appendix \[polyderiv\] provides a more thorough discussion. While we refer to this as the “SPT" expression, we note that, unlike in GR, where equation (\[PkXSPTGR\]) has been shown to be identical to the SPT expression, in MG, additional terms that appear when transforming the Klein-Gordon equation from Eulerian to Lagrangian coordinates need to be considered to show the equivalence [@Aviles:2018saf].
The LRT power spectrum for MG theories is obtained, just like in GR, by keeping the zero-lag term exponentiated: $$\begin{aligned}
\label{PkXLRT}
P^{LRT}_{X}(k) &=& e^{-k^2 \sigma_L^2} \Biggl[\left(1+b_1\right)^2P_{L}(k) + \frac{9}{98}Q_1(k) + \frac{3}{7}Q_2(k) + \frac{1}{2}Q_3(k) \nonumber \\
&&+ b_1\left(\frac{6}{7}Q_5(k)+2Q_7(k)\right) + b_2\left(\frac{3}{7}Q_8(k)+Q_9(k)\right)\nonumber\\
&& + b_1^2 \left(Q_9(k)+Q_{11}(k)\right) + 2b_1b_2Q_{12}(k)\nonumber \\
&& + \frac{1}{2}b_2^2 Q_13(k)+ \frac{6}{7}\left(b_1^2+b_1\right)\left[R_1(k)+R_2(k)\right]_{MG}\nonumber \\
&&+ \frac{6}{7}\left(1+b_1\right)\left[R_1(k)+R_2(k)\right] -\frac{8}{21}\left(1+b_1\right)R_1(k) \Biggr].\end{aligned}$$ The derivation is discussed in Appendix \[spectraderiv\]. The configuration space counterpart, $\xi^{LRT}_{X}(r)$, is obtained by Fourier transforming (\[PkXLRT\]), $$\label{xiXLRT}
\begin{aligned}
\xi^{LRT}_{X}(r) &=\int \frac{d^3k}{\left(2\pi\right)^3} e^{i\bold{k}\cdot\bold{r}}P^{LRT}_{X}(k) \\
& = \int \frac{dk}{2 \pi^2} k^2P^{LRT}_{X}(k)j_0(kr), \\
\end{aligned}$$ with $j_0(kr)$ the zeroth-order Bessel function.
To evaluate the expressions (\[xiXfinal\]), (\[PkXfinal\]), (\[PkXSPT\]) and (\[PkXLRT\]) we modified a publicly available code released by [@Vlah:2016bcl] in [^1], that efficiently performs the 2D integrals in (\[xiXfinal\]), (\[PkXfinal\]) using Haskel transformations, as well those in (\[qfuncsmain\]) and (\[QRformmain\]). On top of the functions $Q_1$-$Q_{13}$ and $R_1$, $R_{2}$, the code was extended to evaluate the new functions $\left[Q_1\right]_{MG}$, $\left[R_1\right]_{MG}$ and $\left[R_1(k)+R_2(k)\right]_{MG}$, as well as the modified correlators (\[qfuncsmain\]). We make this code publicly available in [^2]. Our modified version accepts the modified gravity model growth factors, $D_{MG}(k,z)$ as input along with the linear power spectrum given by: $$\label{Plin}
P_{MG}^{L}(k,z) = \left(\frac{D^1_{MG}(k,z)}{D^1_{GR}(k,0)}\right)^2P_{GR}^{L}(k,0).$$ The linear power spectrum for the background $\Lambda$CDM cosmology is calculated using the publicly available code CAMB [@Lewis:1999bs].
After calculating the necessary MG growth factors using our $\sc{Mathematica}$ notebook, we feed our modified version of the code with tabulated values of the growth factors for the various values of $k$, $r$ and $x$ needed at a given cosmological redshift $z$. The PYTHON module computes the various $Q_n(k)$ and $R_n(k)$ functions through equations (\[Qsmain\]) and (\[Rsmain\]), which are then used to calculate the various components of the CLPT power spectrum $P_{X}(k)$. The k functions can then be simply combined to give the SPT and LRT power spectra, by equations (\[PkXSPT\]) and (\[PkXLRT\]), respectively. Finally, the modified C$++$ counterpart follows a similar procedure to compute the configuration space two-point correlation function given by CLPT, through (\[xiXfinal\]). The procedure is explained in more detail in the Appendix \[correlderiv\].
We finish this section by noting that, even though we restrict our model to the case of a local Lagrangian bias, our framework can be extended to include a curvature bias and/or corrections from EFT, as in [@Vlah:2016bcl]. For modified gravity theories with scale-dependent growth, a general expansion bias is not possible in principle, though for some theories, such as the $f(R)$, the effects of the fifth force can be perturbatively absorbed in terms of higher-order derivatives [@Desjacques:2016bnm; @Aviles:2018saf].
### Calculation of bias parameters in MG {#sec:biasparameters}
In this section we turn to the final necessary ingredient to describe biased tracers, an analytical model for the calculation of the bias parameters in MG.
Central to the GR derivation in Section \[sec:biasparametersGR\], is the assumption that spherical collapse is independent of the exterior spacetime, which, in the case of GR evolution, is given by Birkhoff’s theorem. In MG theories, however, the additional degree of freedom violates Birkhoff’s theorem, which will have important consequences for the modeling of the halo mass function, as well as on its response to an external density perturbation.
The Press-Schechter formalism (\[PSfunction\]) relies upon the assumption that the linearly evolved critical overdensity, $\delta_{cr}$, is always constant at the time of collapse, for example $\delta_{cr}=1.686$ for an EDS evolution. This is not the case for MG due to the presence of the additional scalar field that generates the fifth forces. Following [@doi:10.1111/j.1365-2966.2011.20404.x; @Lombriser:2013wta], if we define $1+\delta_h = y_h^{-3}$, the evolution of a spherically symmetric halo density perturbation, will be given by: $$\label{collapseh}
y''_h -\left(2-\frac{3}{2}\Omega(a)\right) y'_h +\frac{1}{2}\Omega(a)\frac{G_{eff}}{G}\left(y_h^{-3}-1\right)y_h = 0,$$ where $'$ denotes derivatives with respect to $\ln(a)$ and $G_{eff}$, the modified Newton’s constant, is given by $$G_{eff} = \left(1+E\right)G,
\label{eq:Geff}$$
For the $n$DGP braneworld model,[@2010PhRvD..81f3005S] $$E= \frac{2}{3 \beta(a)}\frac{\sqrt{1+\chi^3}-1}{\chi^{-3}},$$ where $\beta(a)$ was defined in (\[eq:nDGPmasses\]) and $$\label{chinDGP}
\chi^{-3} = \frac{\Omega_m n^2}{1.10894 a^3 \beta^2(a)}\frac{y_h^3-1}{y_h^3},$$ with the $n$DGP parameter, $n=H_0r_c$. Note that the fifth force modification does not depend on either the mass or the environment, a property of the Vainshtein mechanism, which means that the collapse barrier for this model is redshift and scale independent [@2010PhRvD..81f3005S; @Barreira:2013xea]. Thanks to this property, the halo biases for this model can be easily calculated by the corresponding GR expressions (\[pbsbiasGR\]), with a different value for the $\it{constant}$ threshold $\delta_{cr}$. For the background cosmology of the Group I simulations, at $z=0.5$, we integrate equation (\[collapseh\]) and find $\delta_{cr}$ to have values (linearly extrapolated at $z=0.5$) of $\delta_{cr}=1.571$ and $\delta_{cr}=1.657$ for the N1 ($n=1$) and N5 ($n=5$) cases, respectively.
For $f(R)$ models, for a collapsing sphere of mass $M$ and radius $R_{th}$, $E$ is given by $$\label{Fforce}
E = 2\beta^2\left[3\frac{\Delta R}{R_{th}}-3\left(\frac{\Delta R}{R_{th}}\right)^2+\left(\frac{\Delta R}{R_{th}}\right)^3\right],$$ with $\beta=\frac{1}{\sqrt{6}}$. Finally, $\frac{\Delta R}{R_{th}}$ is given by $$\begin{aligned}\label{Shell}
& \frac{\Delta R}{R_{th}} = \frac{|f_{R0}| a^3}{\Omega_m y_h^{-3}H_0^2 R_{th}^2} \\
& \times \left[\left(\frac{1+4\frac{\Omega_{\Lambda}}{\Omega_m}}{\left(y_{env} a\right)^{-3}+4\frac{\Omega_{\Lambda}}{\Omega_m}}\right)^{n+1} - \left(\frac{1+4\frac{\Omega_{\Lambda}}{\Omega_m}}{\left(y_{h} a\right)^{-3}+4\frac{\Omega_{\Lambda}}{\Omega_m}}\right)^{n+1}\right]. \\
\end{aligned}$$ Here the overdensity related to $y_h$ is embedded on a longer wavelength environment with over density, $1+\delta_{env} = y_{env}^{-3}$. In models that realize the chameleon screening mechanism, like the $f(R)$ Hu-Sawicki, only a thin shell of a massive sphere contributes to the fifth force, with a fractional thickness $\frac{\Delta R}{R_{th}}$, that causes an enhancement given by $E$. It is this factor $E$ that causes the environmental dependence of spherical collapse in MG, through (\[collapseh\]). When set equal to zero, we recover the standard GR solution, which is the one that describes the evolution of the environment, since on such a long wavelength perturbations $G_{eff}\approx G$ and we have: $$\label{collapseenv}
y''_{env} -\left(2-\frac{3}{2}\Omega(a)\right) y'_{env} +\frac{1}{2}\Omega(a)\left(y_{env}^{-3}-1\right)y_{env} = 0.$$
In light of the coupling between a collapsing halo and the background on which it evolves, as seen through (\[collapseh\]), it is clear that in MG the collapse barrier, $\delta_{cr}$, that is constant in GR, should now be promoted to a function of both the mass $M$ and the environment overdensity $\delta_{env}$. For each choice of $M$ and $\delta_{env_i}$, equations (\[collapseh\]) and (\[collapseenv\]) form a system of coupled differential equations that we solve simultaneously, as in [@doi:10.1111/j.1365-2966.2011.20404.x; @Lombriser:2013wta]. The critical overdensity is identified as the smallest value of the boundary condition $\delta_{h}(a_i)$ at the initial scale factor $a_i=0.002$, that causes a singularity i.e. signifying the onset of nonlinear collapse, at the scale factor of interest, which is then evolved to that scale factor $a$ through [@doi:10.1111/j.1365-2966.2011.20404.x; @Lombriser:2013wta]: $$\begin{aligned}\label{delhevolv}
\delta_{h}(a) = \frac{D^{(1)}(a)}{D^{(1)}(a_i)}\delta_{h}(a_i),
\end{aligned}$$ with $D^{(1)}$ the linear GR growth factor
In Figure \[collapsevsM\], we show the critical density, $\delta_{cr}(z, M,\delta_{env})$ as a function of mass, $M$, for the three $f(R)$ models at $z=0.5$, as it was obtained for a variety of environmental densities, $\delta_{env}$. Because the gravitational strength, $G_{eff}$, is greater in the modified models this allows haloes to form more easily, translating into a lower value of $\delta_{cr}$ than the value in $\Lambda$CDM. $G_{eff}$ is scale dependent and tends towards the GR value on large scales (which would collapse into large mass haloes). GR is also recovered for highly screened models and high density environments. For this reason the critical threshold tends towards the GR value $\delta_{cr}=1.686$ for smaller values of $f_{R0}$, increasing values of $M$ and more positive values of $\delta_{env}$ (regions with larger screening) as found in [@doi:10.1111/j.1365-2966.2011.20404.x; @Lombriser:2013wta]. The deviations from GR become progressively less pronounced as we move from weaker (F4, top panel) to stronger (F6, lower panel) screening, as one would expect.
Having calculated the function $\delta_{cr}(z, M,\delta_{env})$ for the three $f(R)$ models, the MG halo mass function can be again given by the universal prescription (\[PSfunction\]), but now with a modified peak significance $$\label{peakMG}
\nu_{cMG}(z, M,\delta_{env})=\frac{\delta_{cr}(z, M,\delta_{env})}{D^{(1)}(z)\sigma(M)}.$$ The growth factors in MG are scale-dependent, meaning that in equation (\[peakMG\]) one should in principle use $D^{(1)}_{MG}(k,z)$, however, it has been shown [@doi:10.1111/j.1365-2966.2011.20404.x; @Lombriser:2013wta], that it is sufficient to use the GR growth factor, $D^{(1)}(z)$ to define the $\nu_c$ parameters. Modifications beyond this assumption are accounted for later in the free parameters of the halo mass function.
The dependence on the mass and the environment alters, and makes environment dependent, not only the unconditional halo mass function in MG, but also the conditional one. In the presence of a long-wavelength density perturbation, $\Delta$, the conditional halo mass function will be again described by (\[PSfunction\]), but now with a modified $\nu'_{cMG}$ given by: $$\begin{aligned}\label{peakMGprime}
\nu'_{cMG}(M,\delta_{env})=\frac{\delta_{cr}(z, M,\delta_{env}+\Delta)-\Delta}{D^{(1)}(z)\sigma(M)}.
\end{aligned}$$ Application of the bias definition (\[biasrig\]) in that case, gives the first and second order bias parameters $$\begin{aligned}\label{pbsbiasMG}
& b_{MG}^1(M,\delta_{env}) = \frac{\frac{d\delta_{cr}(M,\delta_{env})}{d \delta_{env}}-1}{\delta_{cr}(M,\delta_{env})}\left[q\nu_{cMG}^2-1+\frac{2p}{1+\left(q\nu_{cMG}^2\right)^p}\right],
\\
& b_{MG}^2(M,\delta_{env}) = \\
& \frac{\left(\frac{d\delta_{cr}(M,\delta_{env})}{d \delta_{env}}-1\right)^2}{\delta^2_{cr}(M,\delta_{env})}\left[q^2\nu_{cMG}^4-3q\nu_{cMG}^2+\frac{2p\left(2q\nu_{cMG}^2+2p-1\right)}{1+\left(q\nu_{cMG}^2\right)^p}\right] \\
& + \frac{d^2\delta_{cr}(M,\delta_{env})}{d \delta_{env}^2}\frac{1}{\delta_{cr}(M,\delta_{env})}\left[q\nu_{cMG}^2-1+\frac{2p}{1+\left(q\nu_{cMG}^2\right)^p}\right].\\
\end{aligned}$$ In the case of a sample of halos in a mass range of width $dM$ around a single value $M$, the conditional and unconditional mass functions need to be first averaged over the mass range, $$\begin{aligned}\label{pbsbiasMGavg}
& b_{MG}^1(M,\delta_{env}) = \\
& \frac{1}{I_{dM}}\int\left[\frac{\left(\frac{d\delta_{cr}(M,\delta_{env})}{d \delta_{env}}-1\right)}{\delta_{cr}(M,\delta_{env})}\frac{\nu_{cMG}}{M}\frac{\partial f[\nu_{cMG}]}{\partial \nu_{cMG}}\frac{d \ln \nu_{cMG}}{dM}\right]dM, \\
& b_{MG}^2(M,\delta_{env}) = \\
& \frac{1}{I_{dM}}\int\left[\frac{\left(\frac{d\delta_{cr}(M,\delta_{env})}{d \delta_{env}}-1\right)^2}{\delta^2_{cr}(M,\delta_{env})}\frac{\nu^2_{cMG}}{M}\frac{\partial^2 f[\nu_{cMG}]}{\partial \nu^2_{cMG}}\frac{d \ln \nu_{cMG}}{dM}\right]dM \\
& + \frac{1}{I_{dM}}\int\left[\frac{\frac{d^2\delta_{cr}(M,\delta_{env})}{d \delta_{env}^2}}{\delta_{cr}(M,\delta_{env})}\frac{\nu_{cMG}}{M}\frac{\partial f[\nu_{cMG}]}{\partial \nu_{cMG}}\frac{d \ln \nu_{cMG}}{dM}\right]dM, \
\end{aligned}$$ with $$I_{dM} = \int\left[\frac{f[\nu_{cMG}]}{M}\frac{d \ln \nu_{cMG}}{dM}\right]dM.$$ The details of the peak-background split derivations, for (\[pbsbiasMG\]), (\[pbsbiasMGavg\]), are shown in more detail in Appendix \[pbsMG\].
We note that another popular method, as an alternative to PBS, for calculating halo biases is the excursion set approach [@1991ApJ...379..440B]. Here the universal halo mass function is associated with the Brownian-walk, first crossing distribution of a collapse threshold. In the GR case, the redshift and scale independence of the collapse barrier leads to an analytical solution that recovers the common PBS biases. Given the potential scale, environment and redshift dependence that $\delta_{cr}$ has in MG models however, there is no analytical solution for the excursion set approach in MG and one would need to perform numerical simulations, rather than analytic prediction available for PBS, to determine the predicted biases [@doi:10.1111/j.1365-2966.2011.20404.x; @doi:10.1111/j.1365-2966.2012.21746.x; @doi:10.1111/j.1365-2966.2012.21592.x].
In order to make predictions to compare against simulations, given that the correlation statistics sample a distribution of environments, rather than a specific value of $\delta_{env}$, we average all environment-dependent quantities over a probability distribution for environments in which halos form and reside, defining these on a fixed scale $\zeta$ which we set to 8 Mpch/$h$ [@Lam:2007qw; @doi:10.1111/j.1365-2966.2012.21592.x; @doi:10.1111/j.1365-2966.2012.21746.x], $$\begin{aligned}\label{probenv}
& p_{\zeta}(\delta_{env})= \frac{\beta^{0.5 \omega}}{\sqrt{2 \pi}}\left[1+(\omega-1)\frac{\delta_{env}}{\delta_{cr}}\right]\left(1-\frac{\delta_{env}}{\delta_{cr}}\right)^{-0.5 \omega-1}, \\
& \times \exp\left[-\frac{\beta^{\omega}}{2}\frac{\delta_{env}}{\left(1-\frac{\delta_{env}}{\delta_{cr}}\right)^{\omega}}\right],
\end{aligned}$$ where $\beta\equiv {\left(\frac{\zeta}{8}\right)^3}/(\delta_{cr}\sigma_8^{(2/\omega)})$ with $\omega\equiv\delta_{cr}\frac{n_s+3}{3}$.
The list of environments over which we average the other dependent quantities for the rest of this work, is $\delta_{env}=\left[-1,-0.72,-0.43,-0.15,0.13,0.42,0.7,0.98,1.27,1.55\right]$.
In Figure \[biasvsM\] we demonstrate the impact of the reduced $\delta_{cr}$ values in modified gravity models, separately from modifications to the halo mass function itself, by plotting the environmentally averaged bias parameters $b_{MG}^1$ and $b_{MG}^2$ for GR and three $f(R)$ models we study calculated from relations (\[pbsbiasGR\]) and (\[pbsbiasMG\]), while assuming [*the same underlying halo mass model*]{}, with $(q,p)=(0.707,0.3)$ for $\it{all}$ models. As described earlier, the increased gravitational strength in modified gravity models, parameterized by $G_{eff}$ in (\[eq:Geff\]), allows haloes of a given mass to collapse more readily, yielding a lower critical threshold values for $\delta_{cr}$ and a lower bias relative to the background (reduced $b_1$ relative to GR). The deviations from GR are most pronounced in the models with least screening, like the $F4$ model and are suppressed in the presence of stronger screening.
To accurately predict the biases in the various modified gravity models, we also need to accurately characterize the halo mass function. The values $(q=0.707,p=0.3)$ used to characterize the ST halo mass function in GR were fixed by fitting to $\Lambda$CDM simulations [@Sheth:1999mn] and would not be expected to predict the mass function for modified gravity theories, given the different physics involved in the growth rate and collapse of nonlinear structures. Since the form of the halo mass function is critical to evaluating the biases for the LPT correlation function and power spectra predictions, we determine the best fit values for $(q,p)$ from the simulated halo mass functions for each model in the mass ranges considered. These then uniquely determine the predictions for the biased tracer correlation and power spectra. This approach minimizes the errors that would be introduced at the outset of the LPT modeling from an inaccurate halo mass function. To do this, we evaluate the environment averaged Sheth-Tormen halo mass function (\[PSfunction\]) for various pairs values of $(q,p)$, each using the MG prescription (\[peakMG\]), over the distribution of environments (\[probenv\]) and identify the pair of values that best fits the corresponding halo mass functions from the simulations, through the simple criterion that minimizes the quantity $$\begin{aligned}\label{criterion}
\sum_i \Biggl\lvert\frac{n^{sim}(M_i)}{n^{ST}(M_i,q,p)}-1\Biggr\rvert,
\end{aligned}$$ previously used to fit the halo mass function in Galileons [@Barreira:2013xea; @Barreira:2014zza]. In (\[criterion\]), $i$ is the number of mass bins over which the sum is performed, which, can be tuned to model a narrow mass range rather than fitting the whole range with a single set of parameters.
In Figure \[massfR\], we plot the halo mass functions for the GR and three $f(R)$ MG models we consider at $z=0.5$, together with the best fit ST halo mass functions obtained through (\[criterion\]). In the case of $f(R)$ gravity, we plot both the mean ST halo mass function, as well as the halo mass function for each value of $\delta_{env}$, to give a sense of the variation among various different environments.
In Table \[tab1\], we show the best fit values of $(q,p)$ for the different models, mass ranges and simulations considered. For Group I simulations, we study predictions at $z=0.5$ for halos in a mass range $(2-3.5)\ \times 10^{12} M_{\odot}/h$, for all models. For the Group II simulations we analyze a $z=1$ snapshot, where we consider halos in three separate mass bins: a lower mass bin of $9\times10^{11}-2\times 10^{12}\ M_{\odot}/h$, an intermediate bin of $5\times10^{12}-1\times 10^{13}\ M_{\odot}/h$ and a higher mass bin, $1.1\times10^{13}-9\times 10^{13}\ M_{\odot}/h$.
[ | p[8.2em]{} | C[3.5em]{} |C[3.5em]{} | C[3.5em]{} |C[3.5em]{} | ]{} & &\
Models & $q$ &$p$ & $b_1$ &$b_2$\
$Group\ I: GR$ & 0.726 & 0.345 & 0.301 & -0.501\
$Group\ I: F4$ & 0.671 & 0.361 & 0.120 & -0.435\
$Group\ I: F5$ & 0.765 & 0.321 & 0.211 & -0.470\
$Group\ I: F6$ & 0.670 & 0.362 & 0.230 & -0.449\
$Group\ I: N1$ & 0.701 & 0.369 & 0.224 & -0.661\
$Group\ I: N5$ & 0.702 & 0.357 & 0.268 & -0.503\
$Group\ II: GR\ Low$ & 0.674 & 0.362 & 0.345 & -0.183\
$Group\ II: GR\ Mid.$ & 0.728 & 0.342 & 0.925 & -0.05\
$Group\ II: GR\ High$ & 0.806 & 0.594 & 1.720 & 1.900\
$Group\ II: F5\ Low$ & 0.733 & 0.314 & 0.295 & -0.170\
$Group\ II: F5\ Mid.$ & 0.788 & 0.282 & 0.909 & -0.033\
$Group\ II: F5\ High$ & 0.746 & 0.305 & 1.491 & 0.416\
In Figure \[xienv\] we present the predicted correlation function, $\xi_{env}$ for different environments for the F6 model, along with the correlation function for the environment averaged bias values. Changing the environment can have a notable effect on the predicted correlation function, with variations of $\sim 10\%$ for the F6 model presented in the figure. A decrease in the background environmental density corresponds to a reduction in the correlation function because lower values of $\delta_{env}$ give rise to a lower $\delta_{cr}$ and consequently a lower value for the linear bias $b_1$.
Comparison with simulations {#sec:Results:Compsims}
---------------------------
In this Section, we compare the performance of the various LPT resummation schemes under consideration, combined with our bias model, against the corresponding results from the Group I and Group II simulations, discussed in Section \[sim\], with respect to the correlation function and the power spectrum. All correlation functions from the simulations have been calculated employing the publicly available code $\sc{CUTE}$ [@2012arXiv1210.1833A], using 30 linearly spaced bins in the range $0-140$ Mpc/$h$. The power spectra, on the other hand, have been extracted using a Cloud-In-Cell (CIC) mass assignment scheme, on a grid with resolution of $N_{grid}=1024^3$ and $N_{grid}=1200^3$ cells, for the Group I and Group II simulations, respectively. The power was binned in 30 linearly spaced points in the $k$ range of $0.008-0.3$ h/Mpc.
As discussed in Section \[sec:biasparameters\], for Group I simulations, we study predictions at $z=0.5$ for halos in the mass range $(2-3.5)\ \times 10^{12} M_{\odot}/h$, for all models. For the Group II simulations we analyze a $z=1$ snapshot, where we consider halos in three separate mass bins: a lower mass bin of $9\times10^{11}-2\times 10^{12}\ M_{\odot}/h$, an intermediate bin of $5\times10^{12}-1\times 10^{13}\ M_{\odot}/h$ and a higher mass bin, $1.1\times10^{13}-9\times 10^{13}\ M_{\odot}/h$.
### Correlation function {#sec:Results:Compsims:xi}
To assess the accuracy and performance of the LPT predictions, we first utilize the Group II simulations as a comparison dataset. Given that we only have 1 realization available for the F5 and GR models, the correlation functions exhibit noise due to the random initial phase, but still facilitate valuable conclusions about the performance of the methods tested given the appropriate combination of large volume and high resolution. These allow us to evaluate the performance of CLPT and its variations simultaneously across a wide range of scales, including both the BAO scales and the region roughly following a power-law scaling relation, down to $r\sim5$ Mpc/$h$.
The LPT predictions use the PBS biases evaluated from the best fit halo mass function, when fitted over the specific mass ranges, for each bin as summarized inTable \[tab1\]). Both the LPT and simulation results are compared to the Zel’dovich prediction for the correlation function.
To benchmark our findings we consider both the GR simulations as well as those for the modified gravity model. Figure \[xiresults\] shows these comparisons for the three mass bin ranges, and Figure \[xicomp\] shows the fractional variations with respect to the Zel’dovich component of CLPT.
For both the GR and modified gravity cases it is important to carefully understand the form of the halo mass function to get accurate LPT predictions. We find that simply adopting the standard ST pair of values, $(0.75,0.3)$, gives a poor approximation to the first order bias $b_1$ (for the various values of halo mass), consistent with the findings of the simulation creators [@Arnold:2018nmv] when they extracted the bias estimate from the simulations and compared it to a standard ST prediction. For the results with the best fit halo mass parameters, we find that the full CLPT results for both the GR and the F5 model, incorporating the bias parameters evaluated using our PBS model (Table \[tab1\]) and the environment averaging where necessary, does a very good job, in describing the power-law correlation function, $20-80$ Mpc/h, for all three mass bins and significantly improves upon linear theory at the BAO peak. For all three mass ranges, shown in Figure \[xicomp\], the simulated correlation function falls below the Zel’dovich approximation and we find that the CLPT predictions are reflecting this better than both the linear and LRT predictions. The three approaches, CLPT, LRT and Zel’dovich, all perform well in characterizing the BAO peak for the low mass and intermediate mass bins, for the largest differences being in the highest mass bin for F5, where the LRT approach performs slightly better. The LRT performs poorly at the smaller scales, under $20Mpc/h$, significantly overshooting the observed correlation function, consistent with the results reported in previous studies performed on $\Lambda$CDM cosmologies [@Vlah:2014nta; @Matsubara:2008wx; @2013MNRAS.429.1674C; @doi:10.1111/j.1365-2966.2011.19379.x]. The Zel’dovich approximation provides the best agreement at scales below $10Mpch/h$.
It is also interesting to notice that, in Figure \[xiresults\], while the correlation functions have similar values for GR and the F5 model in the lowest mass bins, the F5 result is noticeably lower than the GR one for the highest mass bin. The amplitude of the correlation function depends on the interplay between the dark matter component (which has higher values in MG) and mostly the linear bias factor, $b_1$, which is lower for MG. In the lower mass bins, the combination of the above two is such that the difference between the GR and the F5 curves is almost neutralized, while in the highest mass bins the linear bias factor $b_1$ is, relatively, even lower, causing the F5 two-point function to have clearly lower values than in the GR case.
To expand on the results from the large volume simulations, we now look into the comparison with the Group I simulations, that, while spanning a smaller volume, allow us to test our schemes for a wider range of models. For each model, five different random realizations are available and the error bars represent the standard deviations over these realizations. Starting with the $f(R)$ family, which is plotted in Figure \[xifR\], we see that the picture painted for the $F4$ an $F5$ models here is similar with the one for the $F5$ model (at $z=1$), with CLPT performing the best at scales $r>20$ Mpc/h and the Zel’dovich result being superior at capturing the smaller scales, while the trend is more pronounced in the F4 case. For the F6 model however, not only does CLPT perform better at these larger scales, but it seems to trace the simulation results more accurately compared to Zel’dovich down to r$\sim7$ Mpc/h.
To explore this small scale performance sensitivity to screening in more detail note we consider what makes the linear order LPT and its one-loop extensions perform differently. In [@Vlah:2014nta] (and also in [@Tassev:2013rta]), it was argued, in dark matter-only studies, that LPT does a poor job at estimating the higher order corrections to the linear displacement dispersion, given by (\[sigmalin\]) and the one-loop correction piece in LPT given by $$\label{sigmaloop}
\sigma_{1loop}^2=\frac{1}{6 \pi^2}\int_0^{\infty}dk\left(\frac{9}{98}Q_1(k)+\frac{10}{21}R_1(k)\right).$$ Comparison with simulations in [@Tassev:2013rta; @Vlah:2014nta], found CLPT to overestimate the size of this correction at small scales, through (\[sigmaloop\]), with the true value of the total $\sigma_L^2+\sigma_{1loop}^2$ being closer to $\sigma_L^2$, which is what the Zel’dovich result uses. Because one-loop CLPT strongly depends on these corrections, through its zero-lag terms (as can be seen in Appendix \[correlderiv\]), it performs worse at the smaller scales compared to its Zel’dovich counterpart. Calculated from our theory prediction for GR at $z=0.5$, the ratio $\frac{\sigma_{1loop}^2}{\sigma_L^2}=0.126$, close to the value in Figure 5 of [@Vlah:2014nta]. In comparison, for the $f(R)$ models at $z=0.5$, the ratio $\frac{\sigma_{1loop}^2}{\sigma_L^2}$ is $(F4, F5, F6)=(0.212, 0.180, 0.152)$. The higher values for F4 and F5 lead to an overestimation in these cases that is responsible for the worse behavior at smaller $r$. It Is worth noting here that if we do not include screening, the ratio $\frac{\sigma_{1loop}^2}{\sigma_L^2}=0.17$ in the $F6$ model, as opposed to the full value of $0.152$. For the $z=1$ Group II simulations, the ratios are $(GR, F5)=(0.08, 0.103)$, which explains the lower discrepancy and better performance of CLPT for F5. This also is consistent with considering that this is an earlier reference in which clustering differences between the theories will be less pronounced. From a physical standpoint, the overestimation reflects an inability in LPT (including the Zel’dovich result), to trap dark matter particles within halos [@Vlah:2014nta], which seems to be more apparent in the LPT approach for stronger MG chameleons. Fortunately, as we said earlier, these models are the ones that violate the observational constraints and are thus less interesting from an astrophysical standpoint.
Finally, we test our LPT approaches applied on the $n$DGP models, that represent the Vainshtein screening mechanism, and the correlation functions of which are presented in Figure \[xinDGP\], for all LPT schemes and the Group II simulations. Just like in the $f(R)$ models, CLPT does a very good job at describing the correlation function for large scales and beyond that, it even seems to perform at least equally well as the Zel’dovich curve down to scales $r\sim10$ Mpc/h, similar to the F6 and GR cases in Figure \[xifR\] discussed earlier. The measurement of the 1-loop statistic discussed in the previous paragraph is consistent with this; for the $n$DGP models, the ratio$\frac{\sigma_{1loop}^2}{\sigma_L^2}=(0.129, 0.122)$ for (N1, N5) respectively, very consistent with the GR value = 0.126. This is the case even in the weaker screening case, the N5 model, and is a very promising sign, given that the Vainshtein mechanism is highly efficient at screening modifications to gravity at smaller scales and contains viable candidates that self-accelerate (even though this particular model does not). The relative performance among the different resummation schemes is very similar to the one observed in the GR and $f(R)$ cases, with all LPT models improving the accuracy at the BAO peak upon linear theory, with the LRT scheme giving more power that CLPT and then the Zel’dovich result. The characterization of the BAO peak on scales $r>100$ Mpc/h is limited in the simulation box with side 1,024 Mpc/h; larger-volume simulations for the $n$DGP model, comparable to the Group II simulations for F5 or GR, will allow us to more clearly trace the region between $100-140$ Mpc/h and draw stronger conclusions about how our models perform at the BAO scales.
The fact that CLPT performs well for all modified gravity models considered in the power-law and BAO scales is very encouraging. On the smaller scales, its robustness for highly screened models is also a positive result. If an MG model was ever detected, it would be a highly screened case, given the tight constraints placed on GR; models F4 and F5 are actually ruled out by observations [@Burrage:2017qrf], but we include them in our analysis to fully investigate the chameleon phenomenology with LPT.
### Power spectrum
Complementary to the correlation function, we also perform tests on its Fourier space counterpart, the halo power spectrum. The mass bins and bias values used in the power spectra calculations are exactly the same as the ones presented in the correlation function case, but with the additional step that all power spectra are shot noise corrected [@PhysRevLett.103.091303]: $\tilde{P}(k)=P(k)-\frac{1}{n_h}$, where $P(k)$ is the uncorrected power spectrum, and $n_h$ is the number density of halos in each bin. Especially for the higher mass bins that contain less halos, this effect is not negligible, especially at higher $k$. We also identify the scale at which perturbation theory starts to fail, $k_{NL}$, with the vertical dashed-dot blue line, using the definition [@Matsubara:2007wj], $k_{NL} = (2\sigma_L^2)^{-1}$, with $\sigma_L^2$ the linear power spectrum dispersion defined in (\[sigmalin\]).
In Figure \[PklightF5\], we present the F5, Group II snapshot at $z=1$. We find the expanded, SPT power spectrum (\[PkXSPT\]) to perform very well at capturing the small $k$ and to follow the power spectrum until $k\sim0.25h/$Mpc, where it starts to overestimate power compared to the simulations, for all three mass bins. This behavior is consistent with what was found in the GR case in earlier works on [@Vlah:2014nta; @Matsubara:2008wx] and also for dark matter in MG [@Aviles:2017aor]. The linear theory result is only accurate at very large scales and quickly underestimates the power at $k>0.05 h$/Mpc. Unlike in the correlation function comparison, where LPT was found to perform very well at a wide range of scales, here we see that the CLPT power spectrum (\[PkXfinal\]) decays quickly and performs considerably worse than the SPT expansion. This is not unexpected, since the power spectrum in LPT has been found to receive, unlike in the configuration space, significant contributions from large, nonlinear $k$ modes, where LPT performs poorly and fails to trap particles inside dark matter halos [@Vlah:2014nta]. Our results show that this to also the case in our MG models. We find that this effect is even more pronounced in the LRT power spectrum (\[PkXLRT\]), which decays sharply in $k$-space, for this reason we do not include the result in our plots.
In Figure \[PkfR\] summarizing the GR and $f(R)$ cases from the Group I simulations, at $z=0.5$, while the CLPT consistently underestimates the power spectrum for all models, the SPT result tracing the simulation points well for F5 and F6 until $k\sim 0.2 h$/Mpc, at which it starts to overestimate the power spectrum. The performance for the the F4 model is slightly worse on small scales. This earlier deviation is not surprising given that the model has the smallest $k_{NL}$ prediction, resulting from a comparatively higher 1D linear dispersion.
For the two $n$DGP models, also shown in Figure \[PkfR\], we find that the SPT predictions perform well at scales $k<0.15$ Mpc/h but overestimate the power on small scales. The CLPT predictions consistently underestimate the power, and are broadly comparable to the linear prediction.
Conclusions {#sec:conclusions}
============
In this work, we modeled the two-point statistics of biased tracers in modified gravity (MG) up to one-loop order in the linear power spectrum, using the Convolution Lagrangian Perturbation Theory (CLPT) framework and its variants. Following standard methods in the literature, the linear piece of the two-point Lagrangian correlator for dark matter is kept exponentiated in the expression for the two-point correlation function, but everything else is expanded, leading to a series of convolution integrals, the expressions of which we derive for scalar-tensor theories.
The evolution of the underlying dark matter density field is described by the LPT framework for dark matter, suitably extended to study scalar-tensor theories, along with an analytical model for the calculation of the first and second order bias parameters in MG. To perform the bias calculations, we employ the Peak-Background split (PBS) approach, in which biases are modeled rigorously as responses of the universal Sheth-Tormen halo mass function in the presence of a long-wavelength density perturbation. This is extended in MG theories, to account for the dependence of the gravitational collapse on the environment and screening. Our PBS implementation, provides a quantitative prediction for the increased production, and the related lower biases, for haloes of a given mass. We apply this scheme to the $f(R)$ Hu-Sawicki and the $n$DGP braneworld models, that are representatives of the chameleon and Vainshtein screening mechanisms, respectively. We make the code used for the analytic predictions publicly available in [^3] and evaluate their performance against state-of-the-art cosmological N-body simulations, for a variety of MG models at $z=0.5$ and $z=1$, with respect to the correlation function and the power spectrum in a variety of mass regimes and scales.
The CLPT implementation, in combination with the analytical bias model, gives good agreement with the simulations, with the only free parameters necessary being those to best-fit the Sheth-Tormen universal halo mass function at the given mass range. The CLPT predicts the correlation function across scales $20-80$ Mpc/$h$, tracing the simulation results at an accuracy of $2-3\%$ and better. At the BAO scales, that provide a valuable probe of fundamental physics, CLPT was found to improve significantly upon the linear theory and Zel’dovich predictions for the F5 models, just like in GR. The Lagrangian Resummation Theory (LRT) approach improved the accuracy a little further at BAO scales for the highest mass range considered. At scales of $r<20$ Mpc/$h$, the CLPT performed well for the highly screened model $F6$ and for the $n$DGP models, while the Zel’dovich predictions performed better for the weakly screened F5 and F4 models. The reason for this behavior was identified, being an overestimation in these low-screening chameleon models of the one-loop contributions to the zero-lag terms at small scales.
In Fourier space, consistent with findings for GR, the CLPT power spectrum was found to underestimate power quickly, compared to the simulations for all MG models. This is due to the power spectrum receiving significant contributions from large $k$, where LPT performs poorly. The Standard Perturbation Theory (SPT) approach, though, which is the low-$k$ expanded version of this power spectrum, performs very well and remains consistent with the simulation results down to $k\sim0.2 h$/Mpc for the $f(R)$ models and down to $k\sim0.15 h$/Mpc for the two $n$DGP models. Beyond these scales, the SPT curve overestimates the power spectrum, as has been found for GR previously.
While we have focused our analysis on LPT predictions for real space, our model can be expanded to capture the redshift-space distortions required for upcoming LSS surveys. Furthermore, even though we focused on a local in matter density bias scheme in the Lagrangian space, in which the bias is purely a function of the local density, one can extend this to include other factors determining bias into the formalism, such as curvature bias, and model them successfully by this PBS scheme. The same applies for potential extensions to include EFT corrections to our LPT model, as in [@Vlah:2016bcl], which could also be used to calculate the components of the Gaussian Streaming Model for MG theories. Finally, our CLPT MG framework can be used to analytically predict marked statistics in MG and assess their ability to boost the MG signals carried in cosmic density fields, as in [@White:2016yhs; @Valogiannis:2017yxm]. We leave these natural extensions to future work.
In the coming decade, a wide array of cosmological surveys will span a large part of the observable universe, searching for hints of new physics beyond $\Lambda$CDM. In this work we demonstrate that semi-analytical approaches, extensively employed in the context of standard GR, can serve as invaluable tools to predict structure formation in cosmologies with an extra degree of freedom in the gravitational sector. A next step for these approaches are to confront them in comparison to realistic simulations of galaxies and clusters that will be observed with surveys coming online in the coming year or two and assess survey ability to identify and constrain potential deviations from GR.
Acknowledgments {#acknowledgments .unnumbered}
===============
We wish to thank Wojciech Hellwing for kindly making available the $n$DGP simulations, on behalf of [@Hellwing:2017pmj] and Baojiu Li for kindly providing the $\sc{ELEPHANT}$ simulations, on behalf of [@Cautun:2017tkc], for numerous discussions and for making us awaref of the Lightcone simulations. We are also grateful to Christian Arnold, as well as to all the other authors of [@Arnold:2018nmv], for kindly providing their Lightcone simulations and for numerous discussions. We also wish to thank Martin White for useful discussions on available simulations of $\Lambda$CDM cosmologies. The work of Georgios Valogiannis and Rachel Bean is supported by DoE grant DE-SC0011838, NASA ATP grants NNX14AH53G and 80NSSC18K0695, and NASA ROSES grant 12-EUCLID12-0004.
\[sec:app\]
Appendix {#appendix .unnumbered}
========
One loop corrections for biased statistics in MG {#App:AppendixA}
================================================
In section \[sec:twopointMG\] of the main text, it was stated that the two point statistics for biased tracers, up to one loop in CLPT, are given by equations (\[xiXfinal\]) and (\[PkXfinal\]), in the configuration space and the Fourier space, respectively. The expressions are convolutional integrals over a sum of individual terms that depend on the Lagrangian correlators (\[eq:correlMG\]), which are essentially the fundamental blocks of CLPT. These expressions (\[eq:correlMG\]), however, are functions of the Lagrangian coordinates $\bold{q}$, while the LPT solutions for the displacement fields up to various orders are found in the Fourier space (through the growth factors (\[growth1st\]), (\[eq:D2sources\]) and (\[D3MG\])). As was also stated in \[sec:twopointMG\], substituting the LPT solutions (\[psifourmain\]) into (\[eq:correlMG\]) gives the integral expressions for the MG Lagrangian correlators (\[qfuncsmain\]), that depend on the functions (\[QRformmain\]), which are also the building blocks of the SPT and LRT power spectra (\[PkXSPT\]) and (\[PkXLRT\]). We start with the innermost layer of integration, deriving the expressions for the k-dependent functions (\[QRformmain\]) in section \[polyderiv\], before showing how to get to the correlators (\[qfuncsmain\]) in section \[correlderiv\]. Finally, in section \[spectraderiv\], we show how the SPT and LRT expressions for the two-point statistics are derived. The notation and index structure is the one adopted in [@Matsubara:2007wj; @Matsubara:2008wx; @2013MNRAS.429.1674C]. These results are consistent with those recently presented in [@Aviles:2018saf].
Polyspectra and k-functions in MG {#polyderiv}
---------------------------------
In LPT, we solve for the displacement field $\bold{\Psi}(\bold{q})$ across various orders in perturbation theory, as $$\label{eq:psiexpappend}
\bold{\Psi}(\bold{q},t) = \sum_{n=1}^{\infty}\bold{\Psi}^{(n)}(\bold{q},t) = \bold{\Psi}^{(1)}(\bold{q},t)+\bold{\Psi}^{(2)}(\bold{q},t)+\bold{\Psi}^{(3)}(\bold{q},t)...$$ In the Fourier space representation, $\tilde{\bold{\Psi}}(\bold{p})$, the solutions are expanded as $$\begin{aligned}\label{psifour}
&\tilde{\Psi}^{(n)}_j(\bold{p}) = \frac{i}{n!}\int\frac{d^3p_1}{(2\pi)^3}..\frac{d^3p_{n}}{(2\pi)^3} \delta_D^3\left(\sum_{j=1}^{n}p_j-p\right) \\
&\times L_j^{(n)}(\bold{p}_1,..,\bold{p}_n)\tilde{\delta}_{L}(\bold{p}_1)..\tilde{\delta}_{L}(\bold{p}_n),
\end{aligned}$$ where $\tilde{\delta}_{L}(\bold{p}_n)$ are the linear density fields in the Fourier space at the time of evaluation. For gravitational evolution governed by GR, the growth factors are only functions of time, and under the additional assumption of an Einstein-De Sitter evolution, the Kernels $L_j^{(n)}(\bold{p}_1,..,\bold{p}_n)$ admit simple scale-independent expressions and (\[psifour\]) can be simply evolved in time by powers of the linear growth factor [@Matsubara:2007wj]. This assumption gives results accurate at the sub-percent level for $\Lambda$CDM cosmologies [@Bouchet:1994xp].
In MG, however, the simple description presented above does not hold, in principle, because the growth factors depend on both space and time, as we saw in Section \[LPTdarkmatter\]. Following [@Aviles:2017aor], we define $$\begin{aligned}\label{LPTKernels}
&L_j^{(1)}(\bold{p}) = \frac{p^j}{p^2} \\
& L_j^{(2)}(\bold{p}_1,\bold{p}_2) = \frac{p^j}{p^2} \frac{D^{(2)}(\bold{p_1},\bold{p_2})}{D^{(1)}(\bold{p_1})D^{(1)}(\bold{p_2})} \\
& (L_j^{(3)})_{sym}(\bold{p}_1,\bold{p}_2,\bold{p}_3) = i\frac{p^j}{p^2} \frac{D^{(3)}_{sym}(\bold{p_1},\bold{p_2},\bold{p}_3)}{D^{(1)}(\bold{p_1})D^{(1)}(\bold{p_2})D^{(1)}(\bold{p_3})} ,
\end{aligned}$$ where the MG growth factors are calculated through the prescription described in Section \[LPTdarkmatter\] and their time arguments have been dropped for notational simplicity. Furthermore, the subscript in the third order Kernel is meant to emphasize on the fact that the configuration that enters the 2-point statistics should be symmetrized [@Matsubara:2007wj; @Aviles:2017aor]. The symmetrized third order growth factor that enters the two-point statistics is given by [@Aviles:2017aor]: $$D^{(3)}_{sym}(\bold{k},-\bold{p},\bold{p})=D^{(3)}(\bold{k},-\bold{p},\bold{p})+D^{(3)}(\bold{k},\bold{p},-\bold{p}),$$ with $D^{(3)}(\bold{k},-\bold{p},\bold{p})$ given by: $$\begin{aligned}\label{D3MG}
&\left(\mathcal{\hat{T}}-A(k)\right)D^{(3)}(\bold{k},-\bold{p},\bold{p})= D^{(1)}(p)\left(A(p)+\mathcal{\hat{T}}-A(k)\right)D^{(2)}(\bold{p},\bold{k})\times \\
& \left[1-\frac{\left(\bold{p}\cdot\left(\bold{k}+\bold{p}\right)\right)^2}{p^2|\bold{p}+\bold{k}|^2}\right]- D^{(1)}(p)\left(A(p)+A(|\bold{p}+\bold{k}|)-2A(k)\right)D^{(2)}(\bold{p},\bold{k}) \\
&+ \left(2A(k)-A(|\bold{p}+\bold{k}|)-A(p)\right)D^{(1)}(k)\left(D^{(1)}(p)\right)^2\frac{\left(\bold{k}\cdot\bold{p}\right)^2}{k^2p^2}\\
& -\left(A(|\bold{p}+\bold{k}|)-A(k)\right)D^{(1)}(k)\left(D^{(1)}(p)\right)^2 - D^{(1)}(k)\left(D^{(1)}(p)\right)^2 \times \Biggl[ \\
&\frac{M_1(\bold{p}+\bold{k})}{3\Pi(|\bold{p}+\bold{k}|)}K^{(2)}_{FL}(\bold{p},\bold{k})-\left(\frac{2A(0)}{3}\right)^2\frac{M_2(\bold{p},\bold{k})|\bold{p}+\bold{k}|^2}{6 a^2 \Pi(|\bold{p}+\bold{k}|)\Pi(k)\Pi(p)} \Biggr]+\\
&\frac{M_1(k)}{3\Pi(k)}\Biggl[\left(2\frac{\left(\bold{p}\cdot\left(\bold{k}+\bold{p}\right)\right)^2}{p^2|\bold{p}+\bold{k}|^2}-\frac{\bold{p}\cdot\left(\bold{k}+\bold{p}\right)}{p^2}\right)\left(A(p)-A(0)\right)D^{(2)}(\bold{p},\bold{k})D^{(1)}(p) \\
&+\left(2\frac{\left(\bold{p}\cdot\left(\bold{k}+\bold{p}\right)\right)^2}{p^2|\bold{p}+\bold{k}|^2}-\frac{\bold{p}\cdot\left(\bold{k}+\bold{p}\right)}{|\bold{k}+\bold{p}|^2}\right)\left(A(|\bold{k}+\bold{p}|)-A(0)\right)D^{(2)}_{\phi}(\bold{p},\bold{k})D^{(1)}(p) \\
&+3\frac{\left(\bold{k}\cdot\bold{p}\right)^2}{k^2p^2}\left(A(p)+A(k)-2A(0)\right)D^{(1)}(k)\left(D^{(1)}(p)\right)^2 \Biggr] \\
&-\frac{1}{2}\frac{k^2}{6 a^2 \Pi(k)}K^{(3)}_{\delta \mathcal{I} sym}(\bold{k},-\bold{p},\bold{p})D^{(1)}(k)\left(D^{(1)}(p)\right)^2. \\
\end{aligned}$$ In (\[D3MG\]), we additionally defined $$\begin{aligned}\label{D2phi}
&D^{(2)}_{\phi}(\bold{p},\bold{k})= D^{(2)}(\bold{p},\bold{k})+\left(1+\frac{\left(\bold{k}\cdot\bold{p}\right)^2}{k^2p^2}\right)D^{(1)}(k)D^{(1)}(p) \\
&-\frac{2A(0)}{3}\frac{M_2(\bold{p},\bold{k})+2\left(\frac{3}{2A(0)}\right)^2K^{(2)}_{FL}(\bold{p},\bold{k})\Pi(k)\Pi(p)}{3\Pi(k)\Pi(p)}D^{(1)}(k)D^{(1)}(p) \\
\end{aligned}$$ and also used the kernels $K^{(2)}_{FL}$ and $K^{(3)}_{\delta \mathcal{I} sym}$, the forms of which were shown in [@Aviles:2017aor]. The second and third order LPT kernels (\[LPTKernels\]) need to be evaluated numerically, now, for each value of $\bold{p}_1,\bold{p}_2$ and $z$ and so should (\[psifour\]), which is the main point of divergence between the MG implementation and the corresponding one in GR.
Now, following [@Matsubara:2008wx], we define the mixed polyspectra $C_{i_1...i_N}$, as $$\begin{aligned}\label{mixedploy}
& \langle \tilde{\delta}_{L}(\bold{k}_1)...\tilde{\delta}_{L}(\bold{k}_l)\tilde{\Psi}_{i_1}(\bold{p}_1)...\tilde{\Psi}_{i_N}(\bold{p}_N) \rangle_c = \\
& = (2\pi)^3 \delta_D^3(\bold{k}_1+..+\bold{k}_l+..\bold{p}_1+..+\bold{p}_N)(-i)^{N}C_{i_1...i_N}(\bold{p}_1,..,\bold{p}_N).
\end{aligned}$$ It is useful to decompose the various polyspectra into the various constituents, order by order in perturbation theory, as e.g. $$\begin{aligned}\label{polyorders}
& C_{ij}(\vec{p})=C_{ij}^{(11)}(\vec{p})+C_{ij}^{(22)}(\vec{p})+C_{ij}^{(13)}(\vec{p})+C_{ij}^{(31)}(\vec{p})+.... \\
& C_{ijk}(\vec{p}_1,\vec{p}_2,\vec{p}_3)=C_{ijk}^{(112)}(\vec{p}_1,\vec{p}_2,\vec{p}_3)+\\
& C_{ijk}^{(121)}(\vec{p}_1,\vec{p}_2,\vec{p}_3)+C_{ijk}^{(211)}(\vec{p}_1,\vec{p}_2,\vec{p}_3)+...,
\end{aligned}$$ where the additional notation has been adopted $$\begin{aligned}\label{mixedployorder}
& \langle \tilde{\delta}_{L}(\bold{k}_1)...\tilde{\delta}_{L}(\bold{k}_l)\tilde{\Psi}_{i_1}^{(r)}(\bold{p}_1)...\tilde{\Psi}_{i_N}^{(s)}(\bold{p}_N) \rangle_c = \\
& = (2\pi)^3 \delta_D^3(\bold{k}_1+..+\bold{k}_l+..\bold{p}_1+..+\bold{p}_N)(-i)^{N}C_{i_1...i_N}^{(r...s)}(\bold{p}_1,..,\bold{p}_N)
\end{aligned}$$ and as previously, the bracketed numbers in the superscripts indicate the orders of contribution in each $\tilde{\Psi}_i(\bold{p})$. The various polyspectra can be expressed as functions of the Lagrangian kernels (\[LPTKernels\]), by identifying the different contributions across each order in LPT, as in (\[polyorders\]) and plugging the solutions (\[psifour\]) into equation (\[mixedployorder\]). The ones relevant for the two-point statistics of biased tracers are [@Matsubara:2008wx]: $$\begin{aligned}\label{mixedployorderspef}
&C_{ij}^{(11)}(\bold{p}) = L_i^{(1)}(\bold{p})L_j^{(1)}(\bold{p}')P_L(p) \\
&C_{ij}^{(22)}(\bold{p}) =\frac{1}{2}\int \frac{d^3p'}{(2\pi)^3}L_i^{(2)}(\bold{p}',\bold{p}-\bold{p}')L_j^{(2)}(\bold{p}',\bold{p}-\bold{p}')\times \\
&\times P_L(p) P_L(|p-p'|) \\
&C_{ij}^{(13)}(\bold{p}) = C_{ji}^{(31)}(\bold{p}) = -\frac{1}{2}L_i^{(2)}(\bold{p}) P_L(p)\times \\
&\times \int \frac{d^3p'}{(2\pi)^3}(L_j^{(3)})_{sym}(\bold{p},-\bold{p}',\bold{p}')P_L(p') \\
&C_{i}^{(2)}(\bold{p}_1,\bold{p}_2;\bold{p}_3) = -L_i^{(2)}(\bold{p}_1,\bold{p}_2) P_L(p_1) P_L(p_2)\\
&C_{ij}^{(2)}(\bold{p}_1;\bold{p}_2,\bold{p}_3) = C_{ji}^{(21)}(\bold{p}_1;\bold{p}_3,\bold{p}_2) = \\
&= -L_i^{(1)}(\bold{p}_2)L_j^{(2)}(\bold{p}_1,\bold{p}_2) P_L(p_1) P_L(p_2)\\
&C_{ijk}^{(112)}(\bold{p}_1,\bold{p}_2,\bold{p}_3) = C_{kij}^{(211)}(\bold{p}_3,\bold{p}_1,\bold{p}_2) = C_{jki}^{(121)}(\bold{p}_2,\bold{p}_3,\bold{p}_1)=\\
&= L_i^{(1)}(\bold{p}_1)L_j^{(1)}(\bold{p}_2) P_L(p_1) P_L(p_2),\\
\end{aligned}$$ where by $P_L(p)$ we denote the MG linear power spectrum.
The scalar functions $Q_n(k)$ and $R_n(k)$ that contribute to the SPT power spectrum (and as we shall see in the next section, to the Lagrangian correlators (\[eq:correlMG\])), are expressed as functions of the polyspectra (\[mixedployorder\]) in GR [@Matsubara:2008wx]. Fortunately, since in MG the above picture is only modified through the modified Kernels in (\[LPTKernels\]), the relationships that give the various scalar functions are of the same form as the ones presented in [@Matsubara:2008wx] (in particular, equations (A50)-(A67) in Appendix A). However, one should be cautious at this point, because certain symmetries that are present in the GR solutions, are not preserved anymore. In particular, the integral in the l.h.s of eq. (A59) in [@Matsubara:2008wx] will not be equal to $R_1(k)+R_2(k)$ anymore, because of the scale and redshift dependence of the MG growth factors. In a similar manner, the functions resulting from eq. (A57) and (A61), that used to be equal to $R_1(k)$ and $Q_1(k)$, respectively, in GR, will differ for our MG models and should be additionally calculated. We denote these by $\left[R_1(k)+R_2(k)\right]_{MG}$, $\left[R_1(k)\right]_{MG}$ and $\left[Q_1(k)\right]_{MG}$ to emphasize on their GR limit. In order to illustrate how these calculations are performed, we show the derivations for $\left[R_1(k)+R_2(k)\right]_{MG}$ and $\left[Q_1(k)\right]_{MG}$, that are both new in MG and serve as a representative example of each category. For the former, we have: $$\begin{aligned}\label{R12plus}
&\left[R_1(k)+R_2(k)\right]_{MG} = -\frac{7}{3} k_i k_j \int \frac{d^3 p}{8 \pi^3}C_{ji}^{(21)}(-\bold{p};\bold{p}-\bold{k},\bold{k})= \\
& = \frac{7}{3} \int \frac{d^3 p}{8 \pi^3}k_i k_j L_j^{(1)}(\bold{k})L_j^{(2)}(-\bold{p},\bold{k}) P_L(p) P_L(k)=\\
& = P_L(k) \frac{7}{3} \int \frac{dr dx}{4 \pi^2} k^2 k r^2 \frac{k^2 -\bold{k}\cdot\bold{p}}{|k-p|^2} \frac{D^{(2)}(-\bold{p},\bold{k})}{D^{(1)}(p)D^{(1)}(p-k)} P_L(kr) =\\
& = \frac{k^3}{4 \pi^2 }P_L(k) \int_0^{\infty} drP_L(kr)\int_{-1}^{1}dx \frac{r^2 \left(1-rx\right)}{1+r^2-2rx} \bar{D}^{(2)}(-\bold{p},\bold{k}),
\end{aligned}$$ where we defined the quantities $x=\hat{\bold{k}}\cdot\hat{\bold{p}}$, $p=kr$ and $$\begin{aligned}\label{growthdef}
& \bar{D}^{(2)}(-\bold{p},\bold{k}) = \frac{7}{3}\frac{D^{(2)}(-\bold{p},\bold{k})}{D^{(1)}(p)D^{(1)}(k)} = \\
&=\frac{7}{3}\frac{D^{(2)}\left(k\sqrt{1+r^2-2rx},k,kr\right)}{D^{(1)}(kr)D^{(1)}(k)} = \\
& \bar{D}_a-\bar{D}_b x^2+\bar{D}_{FL}-\bar{D}_{\delta \mathcal{I}}, \\
\end{aligned}$$ as was done in [@Aviles:2017aor]. Similarly, for $\left[Q_1(k)\right]_{MG}$ we will have: $$\begin{aligned}\label{Q1mod}
&\left[Q_1(k)\right]_{MG}= \frac{7}{3} (k_i k_j k_l - k^2 k_i \delta_{jl}) \int \frac{d^3 p}{8 \pi^3}C_{ijl}^{(211)}(\bold{k},-\bold{p},\bold{p}-\bold{k})= \\
& = \frac{7}{3} (k_i k_j k_l - k^2 k_i \delta_{jl}) \times \\
& \int \frac{d^3 p}{8 \pi^3}L_l^{(1)}(\bold{p})L_l^{(1)}(\bold{p}-\bold{k})L_i^{(2)}(-\bold{p},\bold{p}-\bold{k}) P_L(p) P_L(|k-p|)=\\
& = \frac{k^3}{4 \pi^2 } \int dr dx \frac{(\bold{k}\cdot\bold{p})(\bold{k}\cdot\bold{p}-k^2)-k^2\bold{p}(\bold{k}-\bold{p})}{p^2 |k-p|^2} \times \\
& \bar{D}^{(2)}(\bold{p},\bold{k}-\bold{p}) P_L(kr)P_L(|k-p|) =\\
& \frac{k^3}{4 \pi^2 }\int_0^{\infty} drP_L(kr)\int_{-1}^{1}dx \frac{r^2 (1-x^2)}{(1+r^2-2rx} \bar{D}^{(2)}(\bold{p},\bold{k}-\bold{p})P_L(|k-p|),
\end{aligned}$$ where $|k-p|=k\sqrt{1+r^2-2rx}$ is used and similarly as before, we defined $$\begin{aligned}\label{growthdef2}
& \bar{D}^{(2)}(\bold{p},\bold{k}-\bold{p}) = \frac{7}{3}\frac{D^{(2)}(\bold{p},\bold{k}-\bold{p})}{D^{(1)}(p)D^{(1)}(|k-p|)} = \\
&=\frac{7}{3}\frac{D^{(2)}\left(k,kr,k\sqrt{1+r^2-2rx}\right)}{D^{(1)}(p)D^{(1)}(k\sqrt{1+r^2-2rx})} = \\
& \bar{D}_a-\bar{D}_b \frac{x^2+r^2-2rx}{1+r^2-2rx}+\bar{D}_{FL}-\bar{D}_{\delta \mathcal{I}}. \\
\end{aligned}$$ The above two Kernels, as well as all of the rest that we need, can be compactly written in the form [@Matsubara:2008wx]: $$\begin{aligned}\label{Qform}
Q_n(k) &= \frac{k^3}{4 \pi^2 }\int_0^{\infty} drP_L(kr) \\
& \times \int_{-1}^{1}dx P_L(k\sqrt{1+r^2-2rx})\tilde{Q}_n(r,x)
\end{aligned}$$ and $$\begin{aligned}\label{Rform}
R_n(k) &= \frac{k^3}{4 \pi^2 }P_L(k) \int_0^{\infty} drP_L(kr)\int_{-1}^{1}dx \tilde{R}_n(r,x).
\end{aligned}$$ Using similar methods as the one presented above we get that the various $Q_n(k)$, and after some algebra, are given by: $$\begin{aligned}\label{Qs}
&\tilde{Q}_1 = r^2\left(\bar{D}_a^{(2)}-\bar{D}_b^{(2)}\frac{x^2+r^2-2rx}{1+r^2-2rx}+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right)^2 \\
&\tilde{Q}_2 = \frac{rx(1-rx)}{1+r^2-2rx}\left(\bar{D}_a^{(2)} - \bar{D}_b^{(2)}\frac{x^2+r^2-2rx}{1+r^2-2rx}+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right) \\
&\tilde{Q}_3 = \frac{x^2(1-rx)^2}{(1+r^2-2rx)^2} \\
&\tilde{Q}_5 = rx\left(\bar{D}_a^{(2)}-\bar{D}_b^{(2)}\frac{x^2+r^2-2rx}{1+r^2-2rx}+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right) \\
&\tilde{Q}_7 = \frac{x^2(1-rx)}{(1+r^2-2rx)} \\
&\tilde{Q}_8 = r^2\left(\bar{D}_a^{(2)}-\bar{D}_b^{(2)}\frac{x^2+r^2-2rx}{1+r^2-2rx}+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right) \\
&\tilde{Q}_9 = \frac{rx(1-rx)}{1+r^2-2rx} \\
&\tilde{Q}_{11} = x^2 \\
&\tilde{Q}_{12} = rx \\
&\tilde{Q}_{13} = r^2 \\
&\left[\tilde{Q}_1\right]_{MG} = \frac{r^2 (1-x^2)}{1+r^2-2rx}\left(\bar{D}_a^{(2)}-\bar{D}_b^{(2)}\frac{x^2+r^2-2rx}{1+r^2-2rx}+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right).\\
\end{aligned}$$ Similarly, in accordance with equation (\[Rform\]), we will have the $R_n(k)$ functions: $$\begin{aligned}\label{Rs}
&\tilde{R}_{1} = \frac{21}{10} r^2 \frac{D^{(3)}_{sym}(\bold{k},-\bold{p},\bold{p})}{D^{(1)}(k)\left(D^{(1)}(p)\right)^2} \\
&\tilde{R}_{2} = \frac{rx(1-rx)}{1+r^2-2rx}\left(\bar{D}_a^{(2)}-\bar{D}_b^{(2)}x^2+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right) \\
&\left[\tilde{R_1}(k)+\tilde{R_2}(k)\right]_{MG} = \frac{r^2(1-rx)}{1+r^2-2rx}\left(\bar{D}_a^{(2)}-\bar{D}_b^{(2)}x^2+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right)\\
&\left[\tilde{R}_1\right]_{MG} = \frac{r^2(1-x^2)}{1+r^2-2rx}\left(\bar{D}_a^{(2)}-\bar{D}_b^{(2)}x^2+\bar{D}_{FL}^{(2)}-\bar{D}_{\delta \mathcal{I}}^{(2)}\right).\\
\end{aligned}$$ The functions $Q_1$-$Q_3$, $R_1$ and $R_2$ are the only ones that are necessary to calculate LPT statistics for pure dark matter considerations in MG, with the rest of them that we present here, being the additional functions needed for statistics of biased tracers in MG (in the configuration space). It should be emphasized at this point, that even the functions that have the same integral structure as in GR (for example, $Q_9$-$Q_{13}$), do differ from their GR values, but this difference is manifested in the MG linear power spectra that appear in the integral relations (\[Qform\]) and (\[Rform\]).
Let us finish this section, by noting that in the GR limit, the above functions can be rather easily shown to recover their standard GR forms given in [@Matsubara:2008wx], if one keeps in mind that $\bar{D}_{\delta \mathcal{I}}^{(2)}=\bar{D}_{FL}^{(2)}=0$ and $\bar{D}_a^{(2)}=\bar{D}_b^{(2)}=1$ (for Einstein-De Sitter) in this limit.
Lagrangian correlators and q-functions in MG {#correlderiv}
--------------------------------------------
Having derived the expressions for the scalar functions $Q_n(k)$ and $R_n(k)$ in MG cosmologies, we will now derive the integral formulas for the Lagrangian correlators (\[eq:correlMG\]), that are the building blocks of the 2-point statistics in CLPT. We will adopt the notation of [@2013MNRAS.429.1674C] in this section and will show how the functions in their Appendix B will change for our MG models. To illustrate how the connection between the functions (\[eq:correlMG\]) and the ones presented in the previous section is drawn and also to show how these calculations are performed, we pick a reprsentative example of one these functions, $U_{11}^{(2)}$ and show the derivation below. Starting with the definition: $$\begin{aligned}\label{Udef}
U_{11}^{(2)}(\bold{q}) = \hat{q}^i\langle \delta_1^{(1)} \delta_2^{(1)} \Delta_i^{(2)} \rangle_c,
\end{aligned}$$ we plug in the Fourier space representation of the field $\Delta_i=\Psi_{i}(\bold{q_2})-\Psi_{i}(\bold{q_1})$, as well of the linear overdensities and get: $$\begin{aligned}\label{Urel}
& U_{11}^{(2)}(\bold{q}) = \hat{q}^i \times\\
&\bigg \langle \int\frac{d^3 p}{\left(2 \pi\right)^3} \frac{d^3 p_1}{\left(2 \pi\right)^3} \frac{d^3 p_2}{\left(2 \pi\right)^3} e^{i \bold{p}_2\cdot\bold{q}_2} e^{i \bold{p}_1\cdot\bold{q}_1}\left(e^{i \bold{p}\cdot\bold{q}_2}-e^{i \bold{p}\cdot\bold{q}_1}\right)\tilde{\delta}(\bold{p}_2)\tilde{\delta}(\bold{p}_1)\tilde{\Psi}^{(2)}(\bold{p})\bigg \rangle_c \\
& = \hat{q}^i \int\frac{d^3 p}{\left(2 \pi\right)^3} \frac{d^3 p_1}{\left(2 \pi\right)^3} \frac{d^3 p_2}{\left(2 \pi\right)^3} e^{i \bold{p}_1\cdot\bold{q}_1}e^{i (\bold{p}+\bold{p}_2)\cdot\bold{q}_2}\bigg \langle \tilde{\delta}(\bold{p}_2)\tilde{\delta}(\bold{p}_1)\tilde{\Psi}^{(2)}(\bold{p})\bigg \rangle_c \\
& - \hat{q}^i \int\frac{d^3 p}{\left(2 \pi\right)^3} \frac{d^3 p_1}{\left(2 \pi\right)^3} \frac{d^3 p_2}{\left(2 \pi\right)^3} e^{i \bold{p}_2\cdot\bold{q}_2}e^{i (\bold{p}+\bold{p}_1)\cdot\bold{q}_1}\bigg \langle \tilde{\delta}(\bold{p}_2)\tilde{\delta}(\bold{p}_1)\tilde{\Psi}^{(2)}(\bold{p})\bigg \rangle_c. \\
\end{aligned}$$ This expression contains two terms, which turn out to be equal. For this reason, we focus on the first one and notice that the cumulant can be expressed as a polyspectrum, through (\[mixedployorder\]). The substitution gives: $$\begin{aligned}\label{Urelpart}
&\hat{q}^i \int\frac{d^3 p}{\left(2 \pi\right)^3} \frac{d^3 p_1}{\left(2 \pi\right)^3} \frac{d^3 p_2}{\left(2 \pi\right)^3} e^{i \bold{p}_1\cdot\bold{q}_1}e^{i (\bold{p}+\bold{p}_2)\cdot\bold{q}_2}\bigg \langle \tilde{\delta}(\bold{p}_2)\tilde{\delta}(\bold{p}_1)\tilde{\Psi}^{(2)}(\bold{p})\bigg \rangle_c = \\
&-i \int\frac{d^3 p_1}{\left(2 \pi\right)^3} \frac{d^3 p}{\left(2 \pi\right)^3} e^{i \bold{p}_1\cdot\bold{q}} \hat{q}^i C_{i}^{(2)}(\bold{p}_1, \bold{p};\bold{p}_1-\bold{p}) = \\
&-i \int\frac{d^3 p_1}{\left(2 \pi\right)^3} \hat{q}^i e^{i \bold{p}_1\cdot\bold{q}} \underbrace{\int \frac{d^3 p}{\left(2 \pi\right)^3} C_{i}^{(2)}(\bold{p}_1, \bold{p};\bold{p}_1-\bold{p})}_\text{$I_{C}$}. \\
\end{aligned}$$ In the last line, we defined the integral $I_{C}$, that can be calculated by using the definitions (\[mixedployorderspef\]): $$\begin{aligned}\label{Urelpart2}
& I_{C} = \int \frac{d^3 p}{\left(2 \pi\right)^3} C_{i}^{(2)}(\bold{p}_1, \bold{p};\bold{p}_1-\bold{p}) = \\
& -\int \frac{d^3 p}{\left(2 \pi\right)^3} L_i^{(2)}(\bold{p}_1-\bold{p})P_L(p_1)P_L(p) =\\
& = -\frac{3}{7}\frac{p_{1i}}{p_1^2}\int \frac{d^3 p}{\left(2 \pi\right)^3} p_{1i} \frac{(p_{1i}-p_i)}{|\bold{p}_{1}-\bold{p}|^2}\bar{D^{(2)}}P_L(p_1)P_L(p) \\
& = -\frac{3}{7}\frac{p_{1i}}{p_1^2}\underbrace{\frac{p_{1}^3}{4 \pi^2 }P_L(p_{1}) \int_0^{\infty} drP_L(p_{1}r)\int_{-1}^{1}dx \frac{r^2 \left(1-rx\right)}{1+r^2-2rx} \bar{D}^{(2)}}_\text{$\left[R_1(p_1)+R_2(p_1)\right]_{MG}$} \\
& = -\frac{3}{7}\frac{p_{1i}}{p_1^2} \left[R_1(p_1)+R_2(p_1)\right]_{MG},\\
\end{aligned}$$ where we made use of the previous result (\[R12plus\]). Plugging the result (\[Urelpart2\]) into (\[Urelpart\]) and relabelling $p_1$ as $k$, we have $$\begin{aligned}\label{Urelpart3}
&\hat{q}^i \int\frac{d^3 p}{\left(2 \pi\right)^3} \frac{d^3 k}{\left(2 \pi\right)^3} \frac{d^3 p_2}{\left(2 \pi\right)^3} e^{i \bold{k}\cdot\bold{q}_1}e^{i (\bold{p}+\bold{p}_2)\cdot\bold{q}_2}\bigg \langle \tilde{\delta}(\bold{p}_2)\tilde{\delta}(\bold{k})\Psi^{(2)}(\bold{p})\bigg \rangle_c = \\
&i \int\frac{dk dx}{4\pi^2} e^{i x(kq)}x k \frac{3}{7}\left[R_1(k)+R_2(k)\right]_{MG} =\\
&\frac{1}{2\pi^2}\int dk k \left(-\frac{3}{7}\right)\left[R_1(k)+R_2(k)\right]_{MG} j_1(kq), \\
\end{aligned}$$ where we made use of the Bessel function identity $\frac{1}{2}\int_{-1}^{1}dxxe^{ixkq}=i j_1(kq)$. In exactly the same way, the second term in (\[Urelpart\]) is equal to the first, which finally gives: $$\begin{aligned}\label{Urelfinal}
&U_{11}^{(2)}(q)=\frac{1}{2\pi^2}\int dk k \left(-\frac{6}{7}\right)\left[R_1(k)+R_2(k)\right]_{MG} j_1(kq). \\
\end{aligned}$$ This is the MG equivalent of equation (B$28$) in Appendix B of [@2013MNRAS.429.1674C], which is obviously recovered in the GR limit. In a similar manner, but after lengthy calculations, we get the expressions for all the correlators: $$\begin{aligned}\label{qfuncs}
&V_1^{(112)}(q)=\frac{1}{2\pi^2}\int \frac{dk}{k} \left(-\frac{3}{7}\right)\left[R_1(k)\right]_{MG} j_1(kq), \\
&V_3^{(112)}(q)=\frac{1}{2\pi^2}\int \frac{dk}{k} \left(-\frac{3}{7}\right)\left[Q_1(k)\right]_{MG} j_1(kq), \\
&S^{(112)}=\frac{3}{14\pi^2}\int \frac{dk}{k} \left[2\left[R_1\right]_{MG}+4R_2 +\left[Q_1\right]_{MG} +2Q_2\right]\frac{j_2(kq)}{kq}, \\
&T^{(112)}=\frac{-3}{14\pi^2}\int \frac{dk}{k} \left[2\left[R_1\right]_{MG}+4R_2 +\left[Q_1\right]_{MG} +2Q_2\right]j_3(kq), \\
&U^{(1)}(q)=\frac{1}{2\pi^2}\int dk k \left(-1\right)P_L(k) j_1(kq), \\
&U^{(3)}(q)=\frac{1}{2\pi^2}\int dk k \left(-\frac{5}{21}\right)R_1(k) j_1(kq), \\
&U_{20}^{(2)}(q)=\frac{1}{2\pi^2}\int dk k \left(-\frac{6}{7}\right)Q_8(k) j_1(kq), \\
&U_{11}^{(2)}(q)=\frac{1}{2\pi^2}\int dk k \left(-\frac{6}{7}\right)\left[R_1(k)+R_2(k)\right]_{MG} j_1(kq), \\
&X_{10}^{(12)}(q)=\frac{1}{2\pi^2}\int dk \frac{1}{14}\Biggl(2\left(\left[R_1\right]_{MG}-R_2(k)\right) +3\left[R_1\right]_{MG} j_0(kq) \\
& -3\left[ \left[R_1\right]_{MG} + 2R_2+2\left[R_1(k)+R_2(k)\right]_{MG} +2Q_5 \right]\frac{j_1(kq)}{kq} \Biggr), \\
&Y_{10}^{(12)}(q)=\frac{1}{2\pi^2}\int dk \left(-\frac{3}{14}\right)\Biggl( \left[R_1\right]_{MG} + 2R_2 \\
& +2\left[R_1(k)+R_2(k)\right]_{MG} +2Q_5\Biggr)\times \left[j_0(kq)-3\frac{j_1(kq)}{kq}\right] , \\
&X(q)=\frac{1}{2\pi^2}\int dk\ a(k) \left[\frac{2}{3}-2\frac{j_1(kq)}{kq}\right], \\
&Y(q)=\frac{1}{2\pi^2}\int dk\ a(k) \left[-2j_0(kq)+6\frac{j_1(kq)}{kq}\right], \\
\end{aligned}$$ where we defined $a(k)=P_L(k)+\frac{9}{98}Q_1(k)+\frac{10}{21}R_1(k)$ and, following [@2013MNRAS.429.1674C], we decomposed the matter terms as $$\begin{aligned}\label{decomp}
&A_{ij}^{mn}(q)=X_{mn}(q)\delta_{ij}+Y_{mn}\hat{q}_{i}\hat{q}_{j}\\
&W_{ijk}(q)=V_{1}(q)\hat{q}_{i}\delta_{jk}+V_{2}(q)\hat{q}_{j}\delta_{ki}+V_{3}(q)\hat{q}_{k}\delta_{ij}+T(q)\hat{q}_{i}\hat{q}_{j}\hat{q}_{k}. \\
\end{aligned}$$
Now that the basic framework has been laid out, let us finish this section by briefly summarizing the steps followed to calculate the two point statistics for a given model: after calculating the necessary MG growth factors, (\[growth1st\]), (\[eq:D2sources\]) and (\[D3MG\]), using our $\sc{Mathematica}$ notebook, we feed our modified version of the code in [^4] with tabulated values of the growth factors for the various values of $k$, $r$ and $x$ needed at a given cosmological redshift $z$, as well as with the MG linear power spectrum given by: $$\label{Plin}
P_{MG}^{L}(k,z) = \left(\frac{D^1_{MG}(k,z)}{D^1_{GR}(k,0)}\right)^2P_{GR}^{L}(k,0).$$ The linear power spectrum for the background $\Lambda$CDM cosmology is calculated using the publicly available code CAMB [@Lewis:1999bs]. The PYTHON module computes the various $Q_n(k)$ and $R_n(k)$ functions through equations (\[Qs\]) and (\[Rs\]), which are then used to calculate the various components of the CLPT power spectrum $P_{X}(k)$, through the integrations (\[qfuncs\]) and (\[PkXfinal\]). To calculate (\[qfuncs\]), the $q-$function module is modified accordingly. The k functions can then be simply combined to give the SPT and LRT power spectra, by equations (\[PkXSPT\]) and (\[PkXLRT\]), respectively. Finally, the modified C$++$ counterpart follows a similar procedure to compute the configuration space two-point correlation function given by CLPT, through (\[xiXfinal\]).
SPT and LRT Power spectra {#spectraderiv}
-------------------------
In the main text, it was stated that the SPT and LRT power spectra, given by (\[PkXSPT\]) and (\[PkXLRT\]), correspondingly, are produced when one expands the resummed terms in the exponent of relation (\[PkXfinal\]). Here we show how this derivation takes place, a process that once again shares a lot of similarities with the corresponding one in GR. When fully expanding the resummed Lagrangian correlator $A^L_{ij}$ in (\[PkXfinal\]), one gets: $$\begin{aligned}\label{PkXinterm}
&P_{X}(k) = \int d^3q e^{i\bold{k}\cdot\bold{q}} \\
& \times \Biggl[ 1 - \frac{1}{2}k_ik_j A_{ij} - \frac{i}{6}k_ik_jk_k W_{ijk} + b_1 \left(2i k_i U_i - k_ik_jA^{10}_{ij}\right) \\
&+ b_2\left(i k_i U^{20}_i - k_ik_jU^{(1)}_iU^{(1)}_j\right)+b_1^2\left(\xi_L + ik_iU^{11}_i -k_ik_jU^{(1)}_iU^{(1)}_j\right) \\
& +\frac{1}{2}b_2^2\xi_L^2 + 2b_1b_2\xi_L i k_iU^{(1)}_i\Biggr].
\end{aligned}$$ As done previously, we pick one of terms that is modified, that is the term $b_1^2 ik_iU^{11}_i$, and perform the integration as an example. Plugging in $U^{11}_i$ from (\[qfuncs\]): $$\begin{aligned}\label{integrSPT}
& i\ b_1^2\int d^3q e^{i\bold{k}\cdot\bold{q}}\hat{q}^ik_iU^{11} = \\
& \frac{-i\ b_1^2}{2 \pi^2}\frac{6}{7}\int d^3q dp e^{i kq x}k x p \left[R_1(p)+R_2(p)\right]_{MG} j_1(pq)=\\
& \frac{-i\ b_1^2}{\pi}\frac{6}{7}\int dq dp k p q^2 \left[R_1(p)+R_2(p)\right]_{MG} j_1(pq) \int^{1}_{-1} dxe^{i kq x}x=\\
& \frac{b_1^2}{\pi}\frac{12}{7}\int dq\ dp\ k p q^2 \left[R_1(p)+R_2(p)\right]_{MG} j_1(pq)j_1(kq) =\\
& \frac{b_1^2}{\pi}\frac{12}{7}\int dp\ k p \left[R_1(p)+R_2(p)\right]_{MG} \int_0^{\infty} dq q^2 j_1(pq)j_1(kq) =\\
& b_1^2\frac{6}{7}\int dp\ \frac{k p}{k\ p} \delta_D(p-k) \left[R_1(p)+R_2(p)\right]_{MG} =\\
& b_1^2 \left[R_1(k)+R_2(k)\right]_{MG}, \\
\end{aligned}$$ where we also used the identity $ \int_0^{\infty} dq q^2 j_1(pq)j_1(kq)=\frac{\pi}{2pk}\delta_D(p-k)$. Similar computations for the rest of the terms give (\[PkXSPT\]). In the LRT case, we expand everything but the $q$-independent, “zero-lag" piece of $A^L_{ij}$, which is equal to $2 \sigma^2_L \delta_{ij}$, with $\sigma_L^2=\frac{1}{6 \pi^2}\int_0^{\infty}dkP_L(k)$. Since this term is scale-independent, it can be pulled out of the q integral and all the other integrations can be performed in the same manner as above, resulting in (\[PkXLRT\]), that differs from (\[PkXSPT\]) only in terms of the resummed exponential factor.
PBS biases in MG {#pbsMG}
================
In this section, we will explain the derivation of the Lagrangian PBS biases (\[pbsbiasMG\]) in MG models with environmentally dependent gravitational collapse. The PBS argument is commonly employed in conjunction with the halo approach [@Mo:1995cs; @Mo:1996cn; @Sheth:1998xe], where one states that the conditional halo mass function $\bar{n}_h(M,\Delta)$, modulated by a long-wavelength density perturbation $\Delta$, is modeled by the universal mass function prescription (\[PSfunction\]), with the collapse threshold shifted as $\delta_{cr}\rightarrow\delta_{cr}-\Delta$. The same result was also derived in [@PhysRevD.88.023515], based on the rigorous definition (\[biasrig\]), following a separate universe approach: with regards to galaxy clustering, a large-scale density perturbation, $\Delta$, can be viewed as a modification of the mean physical density $\bar{\varrho}_m$ by an offset, that is [@Desjacques:2016bnm] $$\label{shift}
\bar{\varrho}'_m = (1+\Delta)\bar{\varrho}_m,$$ where $\bar{\varrho}_m$ should not be confused with the mean comoving density $\bar{\rho}_m$ and in a similar manner, the fractional overdensity at a point $\bold{x}$, $\delta_m(\bold{x})$, is shifted as: $$\label{delshift}
\delta'_m(\bold{x}) = \delta_m(\bold{x})+\Delta.$$ This reasoning can be employed to calculate the conditional halo mass function $\bar{n}_h(\Delta)$, by noticing that (\[PSfunction\]) depends solely on comoving quantities (that will not change), with the only exception of the density threshold $\delta_{cr}$, through the peak significance $$\label{peakapp}
\nu_c= \frac{\delta_{cr}}{D^{(1)}(z)\sigma(M)}=\frac{\varrho_{cr}-\bar{\varrho}_m}{D^{(1)}(z)\sigma(M)\bar{\varrho}_m},$$ that quantifies how rare a fluctuation above the density barrier $\varrho_{cr}=(1+\delta_{cr})\bar{\varrho}_m$ is, given an RMS amplitute at the time of collapse z, $\delta\varrho_{RMS}=D^{(1)}(z)\sigma(M)\bar{\varrho}_m$. Now following Birkhoff’s theorem, the critical density for collapse, $\varrho_{cr}$, will be unaffected in the presence a density perturbation $\Delta$, since a collapsing overdensity will not depend on the external spacetime. Combining this fact with equation (\[shift\]), relation (\[peakapp\]) gives [@PhysRevD.88.023515] $$\begin{aligned}\label{peakGRsep}
\nu'_{c}(M)=\frac{(1+\delta_{cr})\bar{\varrho}_m-(1+\Delta)\bar{\varrho}_m}{D^{(1)}(z)\sigma(M)\bar{\varrho}_m}=\frac{\delta_{cr}-\Delta}{D^{(1)}(z)\sigma(M)}.
\end{aligned}$$ The conditional halo mass function $\bar{n}_h(M,\Delta)$ is now given by the universal prescription (\[PSfunction\]), but with the peak significance $\nu'_{c}(M)$ in (\[peakGRsep\]). Combining this fact with the rigorous definition of the bias (\[biasrig\]), gives the known PBS biases (\[biasgeneral\]) for GR.
Let us now turn to the MG case. Following the discussion in Section \[sec:biasparameters\], it is now clear how the density threshold will not be scale and redshift independent anymore, but will depend on the comoving halo mass $M$, as well as the environmental density $\delta_{env}$, that is, a function $\delta_{cr}(M,\delta_{env})$. As a consequence, the peak significance in MG will now become: $$\label{peakappMG}
\nu_{cMG}= \frac{\delta_{cr}(M,\delta_{env})}{D^{(1)}(z)\sigma(M)}=\frac{\varrho_{cr}(M,\delta_{env})-\bar{\varrho}_m}{D^{(1)}(z)\sigma(M)\bar{\varrho}_m}.$$ In the presence of a long-wavelength density perturbation $\Delta$, $\bar{\varrho}_m$ will again change according to (\[shift\]), but also $\delta_{env}$ will now change as dictated by (\[delshift\]), and so will $\delta_{cr}$ that depends on it, which will become $\delta'_{cr}(M,\delta'_{env})=\delta_{cr}(M,\delta_{env}+\Delta)$. From the equivalent of (\[peakGRsep\]), the MG peak significance will now become: $$\begin{aligned}\label{peakGRsep2}
\nu'_{cMG}(M)=\frac{(1+\delta'_{cr})\bar{\varrho}_m-(1+\Delta)\bar{\varrho}_m}{D^{(1)}(z)\sigma(M)\bar{\varrho}_m}=\frac{\delta_{cr}(M,\delta_{env}+\Delta)-\Delta}{D^{(1)}(z)\sigma(M)}.
\end{aligned}$$ As in the GR case, the conditional halo mass function $\bar{n}_h(M,\Delta)$ for MG is now given by the universal prescription (\[PSfunction\]), but with the peak significance $\nu'_{cMG}(M)$ given by (\[peakGRsep2\]). Calculating, now, the first and second order derivatives that we need in (\[biasrig\]), we will have, starting with the linear order: $$\begin{aligned}\label{deriv1}
& \frac{d\bar{n}_h(M,\Delta)}{d \Delta}\Biggr|_{\substack{\Delta=0}} = \frac{d\nu'_{cMG}}{d \Delta}\Biggr|_{\substack{\Delta=0}} \frac{d\bar{n}_h(M,\Delta)}{d \nu'_{cMG}}\Biggr|_{\substack{\Delta=0}} = \\
&\frac{1}{D^{(1)}(z)\sigma(M)}\frac{d \left[\delta_{cr}(M,\delta_{env}+\Delta)-\Delta\right]}{d \Delta}\Biggr|_{\substack{\Delta=0}} \frac{d\bar{n}_h(M,0)}{d \nu_{cMG}} = \\
&\frac{\left[\frac{d \delta_{cr}(M,\delta_{env})}{d \delta_{env}}-1\right]}{D^{(1)}(z)\sigma(M)}\frac{d\bar{n}_h(M,0)}{d \nu_{cMG}},
\end{aligned}$$ where we used the chain rule a few times and also the fact that $\nu'_{cMG}(M)=\nu_{cMG}(M)$ at $\Delta=0$. The definition (\[biasrig\]), combined with the result (\[deriv1\]) and the universal prescription (\[PSfunction\]), gives: $$\label{bias1st}
b_{MG}^1(M, \delta_{env})= \frac{\left[\frac{d \delta_{cr}(M,\delta_{env})}{d \delta_{env}}-1\right]}{D^{(1)}(z)\sigma(M)}\frac{1}{\nu_{cMG}f\left[\nu_{cMG}\right]}\frac{d \left(\nu_{cMG}f\left[\nu_{cMG}\right]\right)}{d\nu_{cMG}},$$ Similarly, the second order derivative will be: $$\begin{aligned}\label{deriv2}
& \frac{d^2\bar{n}_h(M,\Delta)}{d \Delta^2}\Biggr|_{\substack{\Delta=0}} = \frac{d}{d\Delta}\frac{d\bar{n}_h(M,\Delta)}{d \Delta}\Biggr|_{\substack{\Delta=0}}\\
&=\frac{d}{d\Delta} \left[\frac{\left[\frac{d \delta_{cr}(M,\delta_{env})}{d \delta_{env}}-1\right]}{D^{(1)}(z)\sigma(M)}\frac{d\bar{n}_h(M,0)}{d \nu_{cMG}}\right],\\
& =\frac{d^2 \delta_{cr}(M,\delta_{env})}{d \delta^2_{env}} \frac{1}{D^{(1)}(z)\sigma(M)}\frac{d\bar{n}_h(M,0)}{d \nu_{cMG}} + \\
& \frac{\left[\frac{d \delta_{cr}(M,\delta_{env})}{d \delta_{env}}-1\right]^2}{\left[D^{(1)}(z)\sigma(M)\right]^2}\frac{d\bar{n}^2_h(M,0)}{d \nu^2_{cMG}},\\
\end{aligned}$$ which will give the expression for the second order bias factor: $$\label{bias2st}
\begin{aligned}
&b_{MG}^2(M, \delta_{env})= \frac{\frac{d^2 \delta_{cr}(M,\delta_{env})}{d \delta^2_{env}}}{D^{(1)}(z)\sigma(M)}\frac{1}{\nu_{cMG}f\left[\nu_{cMG}\right]}\frac{d \left(\nu_{cMG}f\left[\nu_{cMG}\right]\right)}{d\nu_{cMG}} + \\
& \frac{\left[\frac{d \delta_{cr}(M,\delta_{env})}{d \delta_{env}}-1\right]^2}{\left[D^{(1)}(z)\sigma(M)\right]^2}\frac{1}{\nu_{cMG}f\left[\nu_{cMG}\right]}\frac{d^2 \left(\nu_{cMG}f\left[\nu_{cMG}\right]\right)}{d\nu^2_{cMG}}.\\
\end{aligned}$$ Equations (\[bias1st\]) and (\[bias2st\]) give the Lagrangian PBS biases of first and second order in MG, for any universal mass function $f\left[\nu_{cMG}\right]$. Applying these to the particular ST form (\[stmult\]), we arrive at the relationships (\[pbsbiasMG\]) and (\[pbsbiasMGavg\]) in the main text. The derivatives of the form $\frac{d \delta_{cr}(M,\delta_{env})}{\delta_{env}}$ can be easily calculated numerically, as soon as the function $\delta_{cr}(M,\delta_{env})$ is known from integrating equation (\[collapseh\]). In MG models that still possess a scale and redshift $\it{independent}$ barrier, like the $n$DGP model (and possibly other models in the Vainshtein family), these derivatives will vanish and we recover the standard GR expressions for the PBS biases (\[biasgeneral\]), but with a different $\delta_{cr}$.
[^1]: <https://github.com/martinjameswhite/CLEFT_GSM>
[^2]: <https://github.com/CornellCosmology/bias_MG_LPT_products>
[^3]: <https://github.com/CornellCosmology/bias_MG_LPT_products>
[^4]: <https://github.com/martinjameswhite/CLEFT_GSM>
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'This paper presents the atmospheric characterisation of three large, gaseous planets: WASP-127b, WASP-79b and WASP-62b. We analysed spectroscopic data obtained with the G141 grism (1.088 - 1.68 $\mu$m) of the Wide Field Camera 3 (WFC3) onboard the Hubble Space Telescope (HST) using the `Iraclis` pipeline and the TauREx3 retrieval code, both of which are publicly available. For WASP-127b, which is the least dense planet discovered so far and is located in the short-period Neptune desert, our retrieval results found strong water absorption corresponding to an abundance of log(H$_2$O) = -2.71$^{+0.78}_{-1.05}$, and absorption compatible with an iron hydride abundance of log(FeH)=$-5.25^{+0.88}_{-1.10}$, with an extended cloudy atmosphere. We also detected water vapour in the atmospheres of WASP-79b and WASP-62b, with best-fit models indicating the presence of iron hydride, too. We used the Atmospheric Detectability Index (ADI) as well as Bayesian log evidence to quantify the strength of the detection and compared our results to the hot Jupiter population study by [@angelos30]. While all the planets studied here are suitable targets for characterisation with upcoming facilities such as the James Webb Space Telescope (JWST) and Ariel, WASP-127b is of particular interest due to its low density, and a thorough atmospheric study would develop our understanding of planet formation and migration.'
author:
- Nour Skaf
- Michelle Fabienne Bieger
- Billy Edwards
- Quentin Changeat
- Mario Morvan
- Flavien Kiefer
- Doriann Blain
- Tiziano Zingales
- Mathilde Poveda
- 'Ahmed Al-Refaie'
- Robin Baeyens
- Amélie Gressier
- Gloria Guilluy
- Adam Yassin Jaziri
- 'Darius Modirrousta-Galian'
- 'Lorenzo V. Mugnai'
- William Pluriel
- Niall Whiteford
- Sam Wright
- Kai Hou Yip
- Benjamin Charnay
- Jérémy Leconte
- Pierre Drossart
- Angelos Tsiaras
- Olivia Venot
- Ingo Waldmann
- 'Jean-Philippe Beaulieu'
bibliography:
- 'main.bib'
title: 'ARES[^1] II: Characterising the Hot Jupiters WASP-127b, WASP-79b and WASP-62b with HST'
---
Introduction {#sec:intro}
============
The currently-known exoplanet population displays a wide range of masses, radii, and orbits. Although many planets have been detected and it is thought that they are common in our Galaxy [@howard; @batalha; @cassan; @dressing; @wright_jup], our current knowledge of their atmospheric characteristics is still very limited. Examining the atmospheres of exoplanets further unveils their planetary properties, with their study made possible by various methods, including transit spectroscopy [e.g. @Tinetti_water; @mark]. Facilities such as the *Hubble Space Telescope* and the *Spitzer Space Telescope*, as well as some ground-based observatories, have provided constraints on these properties for a limited number of targets and, in some cases, have identified the key molecules present in their atmospheres while also detecting the presence of clouds and probing their thermal structure [e.g. @brogi; @majeau; @stevenson; @sing; @fu; @angelos30; @kelt9_iron; @pinhas; @k2_18b; @iron_wasp76b; @ares1].
This paper presents the analysis of data from Hubble’s public archive for the exoplanets WASP-127b, WASP-79b and WASP-62b. They are all inflated, with low eccentricities and short orbital periods around bright stars. Table \[tab:general\_parameters\] presents the stellar and planetary parameters for each of these targets.
WASP-127b is an ideal target for spectroscopic studies, given its unusually low density (with a super-Jupiter radius and a sub-Saturn mass). It is located in the short-period Neptune desert, where it is expected that planets might not survive photo-evaporation [@Owen; @Mazeh]. However, photo-evaporation is strongly case-dependent and this planet receives a relatively low XUV flux [@Chen]. Potential explanations for its inflation include tidal heating, enhanced atmospheric opacity, Ohmic heating, and re-inflation by the host star during inward migration [@jeremy2010; @2010ohm; @2011ohm; @rauscher_menou; @2014wu]. Both WASP-62b and WASP-79b are believed to have an evaporating atmosphere, with mass loss rates estimated at $\approx$ 11 g$\cdot$s$^{-1}$ [@Bourrier2015]. WASP-79b, which has a polar orbit, was originally detected through an aberration in the radial velocity due to the Rossiter McLaughlin effect [@Addison13].
All spectral data presented herein were acquired with the G141 grism (1.088 - 1.68 $\mu$m) of the HST/WFC3 camera and details regarding each observation can be found in Table \[tab:HSTdata\]. In Section \[sec:dataanalysis\], we detail how the data were reduced with the `Iraclis` pipeline [@Iraclis], following the approach described by @angelos30, and summarised here. In Section \[sec:atmochar\], we describe the TauREx retrieval code used to analyse the reduced spectra [@Waldmann2015b; @Waldmann2015a; @al-refaie_taurex3], along with the initial parameters and priors used. Our results can be found in Section \[sec:results\], followed by a discussion on our findings and the implications they hold for future missions, including simulations of data from Ariel and JWST.
[@lccc@]{} Parameter & WASP-127b & WASP-79b & WASP-62b\
\
Spectral type & G5 & F5 & F7\
$T_{\text{eff}}$ \[K\] & 5750$\pm$85 & 6600$\pm$100 & 6230$\pm$80\
$\log g$ (cgs) & 3.9 & 4.06$\pm$0.15 & 4.45$\pm$0.10\
$[$Fe/H$]$ & -0.18$\pm$0.06 & 0.03 & 0.04\
\
$P$ \[days\] & 4.17807015$\pm$2.10$^{-6}$ & 3.662387$\pm$4.10$^{-6}$ & 4.411953$\pm$4.10$^{-6}$\
$T_{mid}$ \[BJD$_{TDB}$ - 2450000\] & 8138.670144 & 7815.89868 & 5855.39195\
$i$ \[$^\circ$\] & 88.2$^{+1.1}_{-0.9}$ & 86.1$\pm$0.2 & 88.5 $^{+0.4}_{-0.7}$\
$M_{\text{P}} [M_J]$ & 0.18$\pm$0.02 & 0.85$\pm$0.8 & 0.58$\pm$0.03\
$R_{\text{P}} [R_J]$ & 1.37$\pm$0.04 & 1.53$\pm$0.04 & 1.34$^{+0.05}_{-0.03} $\
$T_{\text{eq},A=0}$ \[K\] & 1400$\pm$24 & 1716.2$^{+25.8}_{-24.4}$ & 1475.3$^{+25.1}_{-10}$\
$R_{P}$/$R_\star$ & 0.09992$^{+0.0028}_{-0.0029}$ & 0.09609$^{+0.0023}_{-0.0027}$ & 0.1091$^{+0.0038}_{-0.0023}$\
$a$/$R_\star$ & 7.846 & 6.069 & 9.5253\
References & [@Palle] & [@brown_rm] & [@brown_rm]\
----------- -------------- -------------------- ------------------- --------------- -------------
Planet Median epoch Mean exposure time Number of spectra PI name Proposal ID
(MJD) (sec)
WASP-127b 58217.51310 95.782 74 Jessica Spake 14619
WASP-79b 57815.37216 138.381 64 David Sing 14767
WASP-62b 57857.82823 138.381 61 David Sing 14767
----------- -------------- -------------------- ------------------- --------------- -------------
Methodology {#sec:dataanalysis}
===========
HST Observations
----------------
Data reduction and calibration were performed using `Iraclis`, software developed in @Iraclis and available on GitHub[^2]; it has been used to extract HST spectra in multiple studies, including @55cancri [@angelos30; @k2_18b]. We used the Mikulski Archive for Space Telescopes[^3] to assess the spectroscopic observational data of WASP-127b, WASP-79b and WASP-62b and information about the observations can be found in Table \[tab:HSTdata\]. The WASP-127b proposal was led by Jessica Spake while David Sing was the PI for the observations of WASP-79b and WASP-62b. Although more data is available from additional instruments, we have restricted our study to HST/WFC3 data, in order to maintain consistency in comparing the analysis of the planets.
Data Analysis
-------------
The planets in this paper were analysed to be comparable to the thirty planets studied in @angelos30. We followed the same methodology as summarised below; differences between our studies are stated explicitly.
Our analysis began with raw spatially scanned spectroscopic images, with data reduction and correction steps performed in the following order: zero-read subtraction; reference pixel correction; non-linearity correction; dark current subtraction; gain conversion; sky background subtraction; calibration; flat-field correction; bad pixels and cosmic ray correction.
Following the reduction process, the flux was extracted from the spatially scanned spectroscopic images to create the final transit light-curves per wavelength band. We considered one broadband (white) light-curve covering the whole wavelength range in which the G141 grism is sensitive (1.088 - 1.68 $\mu$m) and spectral light-curves with a resolving power of 70 at 1.4 $\mu$m. When extracting the spectral light-curves, `Iraclis` accounts for the geometric distortions induced by the tilt of the detector in the WFC3 infrared channel. The bands of the spectral light-curves are selected such that the SNR is approximately uniform across the planetary spectrum. We extracted our final light-curves from the differential, non-destructive reads. Prior to light curve fitting, we chose to discard the first HST orbit of each visit, as these exhibit much stronger hooks than subsequent orbits.
Our white light-curves were fit using literature values and the only free parameters, other than the coefficients for Hubble systematics, were the planet-to-star radius ratio and the transit mid time. This is motivated by the gaps in the observations, caused by Earth obscuration, which often means the ingress and egress of the transit is missed, limiting our ability to refine the semi-major axis to star radius ratio and the inclination planet’s orbit. The limb-darkening coefficients were selected from the quadratic formula by @Claret, using the stellar parameters in Table \[tab:general\_parameters\]. Figure \[fig:white\] shows the raw white light-curve, the detrended white light-curve, and the fitting residuals for WASP-127b while Figure \[fig:all\] shows the fits of spectral light-curves for each wavelength bin.
![Results of the white light-curve of WASP-127b. Top: raw light-curve, after normalisation. Second: light-curve, divided by the best fit model for the systematics. Third: residuals. Bottom: auto-correlation function of the residuals.[]{data-label="fig:white"}](white_fitting.pdf){width="\columnwidth"}
![Spectral light curves fitted with Iraclis for the transmission spectra where, for clarity, an offset has been applied. Left: the detrended spectral light curves with best-fit model plotted. Right: residuals from the fitting with values for the Chi-squared ($\chi^2$), the standard deviation of the residuals with respect to the photon noise ($\bar{\sigma}$) and the auto-correlation (AC). Similar plots for WASP-79b and WASP-62b are shown in Figures \[fig:wasp79\_lc\] and \[fig:wasp62\_lc\]. The mean $\bar{\sigma}$ for each of the three planets is between 1.02 and 1.25 times the photon noise.[]{data-label="fig:all"}](WASP127b_LC.png){width="\columnwidth"}
Atmospheric characterisation {#sec:atmochar}
----------------------------
The reduced spectra obtained using Iraclis were thereafter fitted using the publicly available[^4] Bayesian atmospheric retrieval framework TauREx 3 [@al-refaie_taurex3]. TauREx uses the nested sampling code Multinest [@multinest] to explore the likelihood space of atmospheric parameters and features highly accurate line lists from the ExoMol project [@ExoMol], along with those from HITRAN and HITEMP [@HITRAN; @HITEMP]. In our retrieval analysis, we used 750 live points and an evidence tolerance of 0.5. Several molecular opacities have been tested to model the spectra of the observations; in this publication, we considered five trace gases: H$_2$O [@polyansky_h2o], CH$_4$ [@CH4], CO[@li_co_2015], CO$_2$ [@HITEMP], NH$_3$ [@ExoMol_NH3] and FeH [@dulick_FeH]. In the wavelength range covered by G141, water vapour is the dominant spectral feature, but these other molecules can present detectable signals, particularly FeH [@tennyson_yurchenko]. Clouds are fitted assuming a grey opacity model.
### General setup
In this study we use the plane-parallel approximation to model the atmospheres, with pressures ranging from $10^{-2}$ to $10^6$ Pa, uniformly sampled in log-space with 100 atmospheric layers. We included the Rayleigh scattering and the collision induced absorption (CIA) of H$_2$–H$_2$ and H$_2$–He [@abel_h2-h2; @fletcher_h2-h2; @abel_h2-he]. A summary of the fitted retrieval parameters is given in Table \[tab:general\_parameters\]. For consistency, the same parameter bounds have been applied for all three planets. Constant molecular abundance profiles were used, and allowed to vary freely between $10^{-12}$ and $10^{-1}$ in volume mixing ratio. The planetary radius was set to vary in our models between 0.5 $R_{ref}$ and 1.5 $R_{ref}$, where $R_{ref}$ the reference radius from the literature for each planet, as shown in Table \[tab:general\_parameters\]. This is assumed to be equivalent to the radius at $10^{6}$ Pa pressure.
The cloud top pressure ranged from $10^{-2}$ to $10^6$ Pa, in log-uniform scale. We consider a cloud top pressure of $10^6$ Pa to be a cloud-free atmosphere; the grey cloud model used for this study corresponds to a fully opaque layer below the cloud top pressure.
An isothermal atmosphere was assumed and the planetary temperature, $T_p$, set to vary from 400 to 2500 K; this is to accommodate the wide range in equilibrium temperatures between our three planets, which are between 1400 K and 1750 K as shown in Table \[tab:general\_parameters\].
Atmospheric Detectability Index - ADI
-------------------------------------
For quantifying the detection significance of an atmosphere, we use the Atmospheric Detection Index (ADI) from @angelos30, positively defined as the Bayes Factor between the nominal atmospheric model and the flat-line model (i.e. a model representing a fully cloudy atmosphere). For the flat line model, the only free parameters are the planet radius and temperature, along with the cloud pressure. The nominal model then includes Rayleigh scattering and the collision induced absorption of H$_2$–H$_2$ and H$_2$–He, as well as molecular opacities. If an atmosphere is detected at 3 $\sigma$ and 5 $\sigma$ level, the corresponding ADI will be above 3 and 11, respectively. An ADI below 3 suggests the atmospheric detection is not significant, indicating the spectral feature amplitudes are insufficient given the uncertainty of the data.
To quantify the detection of particular species, we computed the Bayes factor, which is the ratio of the Bayesian evidences of different models. We follow the formalism by [@kass1995bayes] for model selection significance as well as translate the Bayes factor to the more traditional $\sigma$ significance nomenclature following [@BennekeSeager2012].
Ephemeris Refinement
--------------------
Accurate knowledge of exoplanet transit times is fundamental for atmospheric studies. To ensure the planets studied here can be observed in the future, we used our HST white light curve mid times, along with data from TESS [@ricker], to update the ephemeris of each planet. TESS data is publicly available through the MAST archive and we used the pipeline from [@edwards_orbyts] to download, clean and fit the 2 minute cadence data. WASP-127b had been studied in Sector 9; WASP-79b in Sectors 4 and 5; and WASP-62b in Sectors 1-4 and 6-13. After excluding bad data, we recovered 4, 12 and 60 transits for WASP-127b, WASP-79b and WASP62b respectively. These were fitted individually with the planet-to-star radius ratio $R_p/R_s$, reduced semi-major axis ($a/R_s$), inclination ($i$) and transit mid time ($T_{mid}$) as free parameters. Finally, we fitted a linear period (P) to these mid times and selected the updated transit mid time (T$_0$) such that the co-variance between T$_0$ and P was minimised.
Results {#sec:results}
=======
Each planet’s retrieval produced results consistent with the significant presence of water vapour, with opaque clouds in two of the three planets. While we did attempt to retrieve the carbon-based molecules, CO, CO$_2$, and CH$_4$, only their upper value could be constrained as they lack strong absorption features in the G141 wavelength range. In each case our best-fit solution also indicates the presence of FeH, with abundances of log(FeH) between -3.04 and -5.25.
The relatively high water abundances retrieved (10$^{-2}$-10$^{-3}$) for these three planets can be suggestive of metallicities in the super-solar regime [@solarmadhu; @Pinhas_2018; @Charnay_2018]. However, there are known degeneracies between the the cloud pressure, $10^6$Pa radius and water abundance retrieved from WFC3 data [@griffith; @heng_10bar]. Additionally, due to the restrictions of the WFC3 wavelength ranges, these observations are not generally sensitive to the main carbon bearing species and arguments of high metallicities are usually based on retrieved water abundances alone, assuming that half of the oxygen is in H$_2$O as expected for a solar C/O ratio at high temperatures [@solarmadhu]. Hence, observations covering longer wavelength ranges are needed to further constrain the C/O ratios of these planets, to fully understand their metallicity.
Our findings are shown in Table \[tab:retrieval\], with Figure \[fig:fit\_spectra\] showcasing all three retrieved spectra with the corresponding contributions for each opacity source. For WASP-127b, the posteriors are shown in Figure \[posteriors127\], with equivalent results for the other planets available in Figure \[fig:posteriors79\] and \[fig:posteriors62\].\
[lllll]{} Retrieved Parameters & Bounds & **WASP-127b** & **WASP-79b** & **WASP-62b**\
$\log(H_2O)$ & 1e-12 - 1e-1 & $-2.71^{+0.78}_{-1.05}$ & $-2.43^{+0.57}_{-0.76}$ & $-2.03^{+0.52}_{-1.27}$\
$\log(FeH)$ & 1e-12 - 1e-1 & $-5.25^{+0.88}_{-1.10}$ & $-4.42^{+0.91}_{-1.18}$ & $-3.04^{+2.18}_{-2.27}$\
$\log(CH_4)$ & 1e-12 - 1e-1 & $< -5$ & $< -5$ & $< -5$\
$\log(CO)$ & 1e-12 - 1e-1 & $< -3$ & $< -3$ & $< -3$\
$\log(CO_2)$ & 1e-12 - 1e-1 & $< -3$ & $< -3$ & $< -3$\
$\log(NH3)$ & 1e-12 - 1e-1 & $< -5$ & $< -5$ & $< -5$\
$T_p$ \[K\] & 400-2500 & $ 1304^{+185}_{-175}$ & $996^{+249}_{-228}$ & $891^{+211}_{-164}$\
$R_p\, [R_{J}]$ & $\pm $ 50% & $1.15^{+0.04}_{-0.04}$ & $1.55^{+0.02}_{-0.02}$ & $1.35^{+0.01}_{-0.02}$\
$\log(P_{clouds})$ & 1e-2 - 1e6 & $1.7^{+0.93}_{-0.66}$ & $> 4$ & $3.63^{+1.46}_{-1.29}$\
$\mu$ (derived) & & $ 2.34^{+0.20}_{-0.03}$ & $2.38^{+0.33}_{-0.07}$ & $2.39^{+0.51}_{-0.08}$\
ADI & - & 167.9 & 17.1 & 16.2\
$\sigma$-level & - & $>5\sigma$ & $>5\sigma$ & $>5\sigma$\
\
P \[days\] & - & 4.1780619$\pm$1.3x10$^{-6}$ & 3.66239344$\pm$3.5x10$^{-7}$ & 4.41194014$\pm$7.4x10$^{-7}$\
T$_0$ \[BJD$_{TDB}$-2450000\] & - & 8238.943367$\pm$5.5x10$^{-5}$ & 8160.186968$\pm$3.9x10$^{-5}$ & 8476.084602$\pm$4.0x10$^{-5}$\
\[tab:retrieval\]
WASP-127b
---------
As expected given the low density, we retrieved a statistically significant atmosphere around WASP-127b with a strong detection of water and opaque clouds. The retrieved radius is 1.16$^{+0.04}_{-0.04}$ R$_J$ at a pressure of 10 bar, which is smaller than stated in previous studies (1.37 R$_J$, @Chen). However, our analysis is best fit with high altitude opaque clouds (log(P$_{clouds}$) = 1.85$^{+0.97}_{-0.66}$ Pa), which corresponds to approximately 1.37 R$_J$, thus explaining this difference between the retrieved radius and the radius in the literature.
In terms of chemistry, our best fit solution indicates significant amounts of water at log(H$_2$O)= -2.71$^{+0.78}_{-1.05}$, and constraints on FeH. FeH produces the flat absorption between 1.2 and 1.3 $\mu$m, whilst deepening the slope in the longer wavelengths (around 1.5 - 1.6 $\mu$m). We also note a correlation between the amount of these two molecules, the radius and the cloud pressure. For less H$_2$O and FeH, the model requires deeper clouds, but a higher base planet radius. In particular, the abundance of FeH can vary from $10^{-4}$ to $10^{-7}$, depending the complementary contribution of clouds.
The posteriors for FeH are, however, always distinct; clouds cannot be used to completely replace the additional visible absorption provided by FeH. A lower metallicity, and larger radius, could be consistent with current data, but is not the best-fit solution. Given the posterior distribution, we don’t find a clear correlation between the radius and the water abundance.
WASP-79b
--------
For WASP-79b, we get very similar results to those of WASP-127b, with the exception of the cloud deck. Following our baseline approach, we find a large abundance of water at log(H$_2$O)= -2.43$^{+0.57}_{-0.76}$ and well defined constraints on the abundance of FeH with log(FeH)= -4.42$^{+0.91}_{-1.18}$. The clouds, however, do not impact the model, and we only retrieve a lower limit on their top pressure (P$_{clouds}$ >$10^3$ Pa). This means that either the planet possesses a clear atmosphere, or that the clouds are located below the visible pressure, at which the atmosphere is opaque due to molecular or collision induced absorption. We do not detect signatures of CH$_4$, CO, or CO$_2$. The retrieved temperature of $\sim$1000 K is lower than the calculated equilibrium temperature for this type of planet; this was also found in [@sotzen_w79] and is discussed further in Section \[discussion\].
WASP-62b
--------
The recovered spectrum of WASP-62b was flatter than the two other planets. However, we found that the data was best explained by the presence of H$_2$O and FeH and, for this retrieval, the recovered abundances are log(H$_2$O)=-2.03$^{+0.52}_{-1.27}$ and log(FeH)=-3.04$^{+2.18}_{-2.27}$. These results stem from detections in the lower-wavelength spectrum, below 1.5 $\mu$m, which guides the retrieval towards non-fully opaque sources, such as clouds and high-radius solutions. Again, the retrieved temperature is lower than the expected 1475 K equilibrium temperature, which is indicative of a large day-night temperature contrast and/or efficient cooling mechanisms. Our analysis indicates that clouds are likely to be present, but the quality of our data means that we cannot completely rule out a clear atmosphere.
The retrieved abundances are very high, but we note that the posteriors allow for a wide range of abundances and present interesting correlations; such as, the lower the abundance of H$_2$O and FeH are, the higher in the atmosphere the clouds are located. There is also a negative correlation between the molecular abundances and the temperature and, from the posterior distributions, we see that the data is consistent with abundances of order of $10^{-4}$ in H$_2$O and FeH.
Finally, we note that given the low spectral variations in this spectrum, the retrieval may lack a scale height constraint, which would provide a relevant baseline in predicting the molecular abundances and temperature more accurately.
\[bayes\_evid\]
{width="\textwidth"}
Ephemeris Refinement
--------------------
We found that the observed HST and TESS transits were consistent with literature ephemeris within 1$\sigma$. Nevertheless, we refined the period and reference mid transit time for each planet. The updated ephemeris is given in Table \[tab:retrieval\] while the fitting for the TESS data can be seen in Figures \[fig:wasp127\_tess\], \[fig:wasp79\_tess\] and \[fig:wasp62\_tess\]. The observed minus calculated plots are given in Figure \[fig:ephm\_refine\] and all transit mid times used for the fitting are listed in Table \[tab:mid\_times\]. These have been uploaded to ExoClock[^5], an initiative to ensure transiting planets are regularly followed-up, keeping their ephemeris up-to-date for the ESA Ariel mission [@tinetti_ariel].
Discussion
==========
Initially our baseline model did not include FeH, but these models struggled to fit the data, forcing solutions to lower temperatures and nonphysical values in order to account for the opacity sources at shorter wavelengths with a grey cloud deck. FeH has strong absorption features in the visible and near-infrared, and can be expected at the temperatures of these planets [@tennyson_yurchenko; @madhu]; hence we propose it as the possible absorber to suit our spectral features and explore our justifications for FeH over molecules with similar spectral signatures, such as TiO or VO, in this discussion. FeH was not included in the analysis of @angelos30 and thus, for the hotter planets in that study, retrievals with FeH may alter the retrieved atmospheric characteristics.
Theoretical equilibrium chemistry models predict FeH [@sharp; @woitke] to be stable in the gas phase at the temperatures and pressures consistent with the planetary atmospheres considered here. FeH has previously been observed in L and M brown dwarfs at 1800 K [@visscher_feh]. In cooler T dwarfs, it has been shown to appear where brown dwarfs have temperatures below 1350 K [@coolBD], with some additional studies [@browndwarf] confirming FeH detection in dwarfs with temperatures of 1000 K. The latter of these detections is at temperatures comparable to the retrieved temperatures the planets here.
A recent study from [@iron_wasp76b] found atomic iron (Fe) in the day-side of the planet WASP-76b, and not in the terminator, concluding that Fe is condensing on the night-side, then falling into deeper layers of the atmosphere. Furthermore, the results in [@Pluriel] and [@Caldas2019] have investigated how the 3D structure of the atmosphere biases the abundances retrieved with typical retrieval codes, since there is a chemical dichotomy between the day- and night-side that is not considered in a 1D treatment of transit geometry.
We therefore identify three possible scenarios for the detection of FeH in these planets:\
- FeH is orginating from the day-side where the temperature is much higher, and leaks in the night-side before it is able to condense due to circulation processes [@HengCirculation].
- Atmospheric retrieval studies involve temperature bias due to 3D effects, and we retrieve indeed a cooler temperature than expected; we discuss this in Section \[temperature\].
- A 3D effect is in play and we retrieve the FeH in the day-side inflated region of the limb [@Caldas2019; @Pluriel].\
Table \[bayes\_evid\] contains the log evidence of several retrievals for each planet. In all cases, the addition of FeH increases the goodness of fit, while also raising the retrieved temperature. We note that all models included clouds. By comparing the log evidence of the models with only H$_2$O and the models with H$_2$O and FeH, we confirm for all planets that clouds are not a suitable opacity substitute for FeH.
The difference in log evidence for these models $(\Delta\text{log}(E))$ is 8.33, 3.39, and 5.37 for WASP-127b, WASP-79b and WASP-62b respectively (4.48$\sigma$, 3.09$\sigma$ and 3.72$\sigma$ detection of FeH, respectively). This indicates strong to decisive evidence in favour of models containing FeH [@kass1995bayes; @BennekeSeager2012; @le2layer].
While we postulate that our evidence holds for FeH, it is possible that we detect another, yet unidentified opacity source with absorption characteristics similar to those of FeH over the WFC3 passband. For instance, similar absorption features can be produced with metal oxides such as TiO, VO and YO. However, we do not expect the presence of these molecules in these planets due to the planets’ low equilibrium temperatures. TiO and VO have condensation temperatures of over $\sim$2000K [@Lodders2002; @HubenyTiOVO; @FortneyTiOVO]; the highest equilibrium temperature featured of these three planets is WASP-79b’s 1716K, as referenced in Table \[tab:general\_parameters\], thus rendering it less likely that the spectral features are due to TiO and VO compared to FeH. Ultimately, this further exemplifies the need for longer wavelength coverage with JWST or Ariel to confirm the nature of observed absorption in the future.
For each planet, we have calculated the ADI and found significant evidence of atmospheric features for all three. Given the water detection on all three planets, our results support the conclusions drawn by @angelos30; inflated, hot Jupiter-like planets do not necessarily destroy water in their upper atmospheres.
Retrieved temperature {#temperature}
---------------------
For the three planets considered, the temperature retrieved is notably lower than the equilibrium temperature. In Figure \[fig:temperature\], we present a plot analysing the temperatures retrieved for other planets, particularly giving attention to the population paper we based our study on, [@angelos30]. Indeed, retrieved temperatures are typically lower than the equilibrium ones, and we derived a best fit of this.
One of the key assumptions leading to this effect is that the equilibrium temperature is usually calculated for the planet day-side and considering a planetary albedo of zero. Considering an albedo greater than zero necessarily implies a loss of energy, and therefore a lower equilibrium temperature.
Furthermore, the region probed during transit eclipse observations is the terminator region: a mix between the day- and the night-side. The temperature difference observed may indicate a bias in the retrievals, which consider exclusively a 1D geometry of the atmospheres. This bias has been pointed out by several studies, especially [@Caldas2019; @cold] and [@Pluriel].
WASP-127b
---------
We used ExoREM (Exoplanet Radiative-convective Equilibrium Model; [@Baudino2017; @Charnay_2018]), a self-consistent simulation software for brown dwarfs and giant exoplanets, to calculate the mean temperature profile and the expected abundances of WASP-127b assuming a solar composition.
The model suggests significant abundances of H$_2$O, CO and Na; though, as stated previously, the WFC3 coverage means our data set is only sensitive to H$_2$O.
From Figure \[fig:contribution\_function\], we can see that our retrieval is sensitive at pressures between $\approx 10^{4}$ and $10^{2}$ Pa. Figure \[fig:127b\_abd\] indicates the retrieved abundances of WASP-127b are compatible with a solar composition in this pressure range. The mean retrieved abundance of FeH is higher than expected values; however, the error spans three orders of magnitude, allowing for more physical solutions, as discussed at the beginning of this section.
As displayed in Figure \[fig:127b\_TP\], the retrieved temperature of WASP-127b is compatible with the calculated mean temperature profile within our pressure sensitivity range. We can also see that the calculated temperature profile crosses the condensation curves of MnS and Cr between $10^{4}$ and $10^{3}$ Pa. We could therefore expect clouds composed of these species to form at these pressures.
![\[fig:contribution\_function\] Typical contribution function of our retrievals. We can see the effect of the H$_2$O lines. The raise in sensitivity at $\approx$ 10$^4$ Pa is due to the opaque cloud.](Contribution.png){width="50.00000%"}
![Expected abundances of WASP-127b generated by Exo-REM, assuming chemical non-equilibrium for C, N and O bearing species, an eddy diffusion coefficient of 10$^8$ cm$^2$.s$^{-1}$, and a metallicity of 3 times solar [@Lodders2019]. We have included TiO and VO to the Exo-REM run but found their predicted abundances to be negligible ($<$ 10$^{-14}$).[]{data-label="fig:127b_abd"}](fig8.pdf){width="50.00000%"}
![\[fig:127b\_TP\] Mean temperature profile of WASP-127b, assuming radiative transfer equilibrium, generated by Exo-REM. The condensation profiles of various species are represented as dotted lines.](fig9.pdf){width="\columnwidth"}
Low and high resolution spectra of WASP-127b have been collected with ground-based instruments. @Palle obtained low resolution spectroscopy with the Andalucia Faint Object Spectrograph and Camera (ALFOSC) spectrograph mounted on the Nordic Optical Telescope (NOT), covering the spectral range 0.45 - 0.85 $\mu m$. A slope was detected in the optical wavelength spectrum, interpreted as Rayleigh scattering and potentially Na. They also attribute the trend to TiO/VO with low significance. @Chen observed with the OSIRIS spectrograph, mounted on the Gran Telescopio Canarias (GTC) telescope, covering the range 0.4 - 1.0 $\mu m$. They presented detection of alkali metals (Na, K and Li), and hints of clouds and water with a retrieved abundance of log(H$_2$O)=-2.60$^{+0.94}_{-4.56}$. A recent study with Hubble STIS and WFC3, combined with Spitzer data from IRAC 1 and 2, also concluded that water was present in the atmosphere. [@spake2019] used an MCMC model to fit the data, resulting in a best-fit solution detecting H$_2$O, CO$_2$, Na and K; their water abundance of log(H$_2$O)=-2.87$^{+0.58}_{-0.61}$ is similar to the value retrieved here.
Hence, our independent data reduction and analysis of the water content in the atmosphere of WASP-127b is consistent with these studies. We do not attempt a joint retrieval with this data due to the potential incompatibility between the data sets from different instruments as highlighted in @yip. Given that the planet lies within the short-period Neptunian desert and has large atmospheric features, it will be an intriguing target for further characterisation.
WASP-79b
--------
@sotzen_w79 utilised the same WFC3 data set, along with observations from ground-based facilities, TESS and Spitzer, to study the atmosphere of WASP-79b. Their retrieval results indicates the presence of H$_2$O, Na and FeH. Our retrieved water abundance is consistent with that from @sotzen_w79; (–2.20 <log(H$_2$O) <–1.55).
In our retrievals without FeH as an opacity source, our solution is driven to low temperatures; [@sotzen_w79] encountered a similar predicament when attempting to fit a chemical equilibrium model to the data. Here, by adding FeH as a retrieval parameter, our recovered temperature increases to 996 K$^{+249}_{-228}$, which more readily agrees with what is expected for the terminator region.
While the temperature is still cooler than expected, we note the degeneracy with the 10 bar radius. Our analysis of purely the HST/WFC3 data also favours the presence of H$_2$O and FeH. Na does not have features within the WFC3 spectral range, and we do not attempt the addition of other data for the aforementioned reasons.
WASP-62b
--------
WASP-62b has demonstrably similar bulk characteristics to HD209458b; both planets have roughly the same radius and effective temperature, although HD209458b is $\approx$ 20% more massive than WASP-62b [@bonomo2017]. Given their similarities, we may expect them to exhibit a similar atmospheric chemistry and structure. HD 209458b has been extensively analysed in the literature, with 3D simulations [@Showman2009] and cloud analysis [@Sing2016], making it ideal to interpret the results on WASP-62b. We observe a cloud deck located at $\approx 2.5 \times 10^3$ Pa. This cloud deck could be explained by the condensation of MgSiO$_3$ in the atmosphere, as was the case with HD209458b [@Sing2016].
Using the models of @Showman2009 and @Caldas2019, we may expect the temperature at the terminator to be close to $\approx$ 1350 K. This is somewhat hotter than the $891^{+211}_{-164}$ K retrieved in our standard setup (2.2$\sigma$). However, as seen in the posterior plot in Figure \[fig:posteriors62\], there is a strong correlation between the temperature, the planet radius and the cloud pressure; so the data remains consistent with the expected temperature.
Wavelength WASP-127b WASP-79b WASP-62b
------------ ---------------- ---------------- ----------------
1.12625 9800 $\pm$ 53 11420 $\pm$ 74 12012 $\pm$ 56
1.14775 10030 $\pm$ 48 11477 $\pm$ 59 11993 $\pm$ 51
1.16860 10026 $\pm$ 72 11371 $\pm$ 71 11890 $\pm$ 61
1.18880 10070 $\pm$ 50 11362 $\pm$ 71 11983 $\pm$ 65
1.20835 9967 $\pm$ 53 11372 $\pm$ 61 11909 $\pm$ 61
1.22750 9936 $\pm$ 61 11350 $\pm$ 70 11909 $\pm$ 58
1.24645 9990 $\pm$ 54 11398 $\pm$ 72 12036 $\pm$ 64
1.26550 9938 $\pm$ 53 11248 $\pm$ 64 11913 $\pm$ 64
1.28475 10050 $\pm$ 49 11325 $\pm$ 76 12024 $\pm$ 65
1.30380 10068 $\pm$ 52 11434 $\pm$ 73 11999 $\pm$ 56
1.32260 10123 $\pm$ 46 11292 $\pm$ 58 11820 $\pm$ 54
1.34145 10318 $\pm$ 53 11600 $\pm$ 72 12151 $\pm$ 61
1.36050 10500 $\pm$ 70 11417 $\pm$ 74 11964 $\pm$ 52
1.38005 10513 $\pm$ 65 11448 $\pm$ 59 12120 $\pm$ 53
1.40000 10417 $\pm$ 53 11548 $\pm$ 72 12065 $\pm$ 62
1.42015 10415 $\pm$ 50 11664 $\pm$ 71 12082 $\pm$ 54
1.44060 10485 $\pm$ 59 11457 $\pm$ 56 12019 $\pm$ 49
1.46150 10408 $\pm$ 54 11384 $\pm$ 73 12102 $\pm$ 71
1.48310 10305 $\pm$ 51 11375 $\pm$ 75 11901 $\pm$ 62
1.50530 10246 $\pm$ 61 11476 $\pm$ 78 11901 $\pm$ 59
1.52800 9989 $\pm$ 60 11383 $\pm$ 89 11907 $\pm$ 61
1.55155 9990 $\pm$ 55 11283 $\pm$ 58 11837 $\pm$ 68
1.57625 9816 $\pm$ 56 11245 $\pm$ 61 11907 $\pm$ 65
1.60210 9781 $\pm$ 58 11245 $\pm$ 87 11796 $\pm$ 63
1.62945 9665 $\pm$ 65 11053 $\pm$ 80 11734 $\pm$ 58
: WFC3 transit depths and errors (in ppm) for for WASP-127b, WASP-79b and WASP-62b.
Future Characterisation
-----------------------
Upcoming ground and space-based telescopes such as the European Extremely Large Telescope [E-ELT, @brandl], the Thirty Meter Telescope [TMT, @skidmore], the Giant Magellan Telescope [GMT, @fanson], the James Webb Space Telescope [JWST, @greene]), Twinkle [@twinkle] and Ariel [@tinetti_ariel] will characterise the atmospheres of a large population of exoplanets via transit and eclipse spectroscopy at visible and infrared wavelengths. These missions will move the exoplanet field from an era of detection into one of characterisation, allowing for the identification of the molecular species present and their chemical profile, insights into the atmospheric temperature profile and the detection and characterisation of clouds [e.g. @rocchetto; @rodler; @le2layer].
Ariel has been selected as ESA’s M4 mission adoption candidate for launch in 2028 and is designed for the characterisation of a large and diverse population of exoplanetary atmospheres to provide insights into planetary formation and evolution within our Galaxy. Ariel will provide simultaneous photometry and spectroscopy over 0.5 - 7.8 $\mu$m. Each of the planets studied here is an excellent target for atmospheric studies with Ariel [@GUILLY] and we use ArielRad [@mugnai] to simulate observations of this forthcoming mission. For each of the planets we take the best-fit solution from the Hubble WFC3 analysis to model Ariel observations at the native resolution of its instruments. Figure \[fig:ariel\_plot\] highlights the increased wavelength coverage and data quality that will be achieved with Ariel, allowing for a deeper understanding of each of these worlds. WASP-79b is part of the JWST Early release Science (ERS) program and will be observed by JWST with several different instruments [@bean]. Here we simulate JWST observations for these planets, assuming NIRISS GR700XD and NIRSpec G395H are used. Again, the increase in data quality is easily discernible and, although it is not a dedicated exoplanet mission, JWST promises to provide exquisite data for atmospheric characterisation.
Conclusion
==========
We have presented the analysis of data from Hubble’s WFC3 G141 grism for three planets. By using the `Iraclis` pipeline and fitting the resultant spectra with TauREx, we have characterised the atmospheres of WASP-127b, WASP-79b and WASP-62b, recovering best fit models which favour the presence of H$_2$O and FeH in each case. This was performed during the ARES Summer School, using software and data publicly available to the community in order to allow for reproducible results.
The properties of WASP-127b, particularly its extended atmosphere with clouds and large spectral features; the resultant high atmospheric detectability; and its unusually low density; make it an ideal target for further characterisation with the next generation of facilities. Large spectral features were also detected in WASP-79b and WASP-62b, with clouds in the atmosphere of the latter.
None of the three planets studied have strong features in their spectra that can be linked to NH$_3$, CH$_4$, CO, or CO$_2$. This is expected, given their spectroscopic lines do not have major bands in this wavelength range compared to the H$_2$O and FeH lines and higher quality data, with a broader spectral coverage, is required to improve constraints on the atmospheric chemistry. Nevertheless, studying the atmospheric composition of these planets has extended the catalogue of hot Jupiters studied with WFC3 from those by [@angelos30]. The ADI introduced therein has been utilised effectively in this paper to estimate the significance of these atmospheric observations. This was done in order to unify the statistical results between our study and that of further populations studies, which remain fundamental tools in understanding the nature and evolutionary history of planets.\
Acknowledgements:
We want to thank the anonymous reviewer for the insightful and constructive comments which helped improve the quality of the manuscript.
This work was realised as part of ARES, the Ariel Retrieval Exoplanet School, in Biarritz in 2019. The school was organised by Jean-Philippe Beaulieu with the financial support of CNES. JPB acknowledges the support of the University of Tasmania through the UTAS Foundation and the endowed Warren Chair in Astronomy, Rodolphe Cledassou, Pascale Danto and Michel Viso (CNES). BE, QC, MM, AT and IW rfunding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme grant ExoAI (GA No. 758892) and the STFC grants ST/P000282/1, ST/P002153/1, ST/S002634/1 and ST/T001836/1. NS acknowledges the support of the IRIS-OCAV, PSL. MP acknowledges support by the European Research Council under Grant Agreement ATMO 757858 and by the CNES. RB is a Ph.D. fellow of the Research Foundation–Flanders (FWO). WP, TZ, and AYJ have received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement n$^\circ$ 679030/WHIPLASH). OV thank the CNRS/INSU Programme National de Planétologie (PNP) and CNES for funding support. GG acknowledges the financial support of the 2017 PhD fellowship programme of INAF.
This paper includes data collected by the TESS mission which is funded by the NASA Explorer Program. TESS data is publicly available via the Mikulski Archive for Space Telescopes (MAST).
**Software:** Iraclis [@Iraclis], TauREx3 [@al-refaie_taurex3], pylightcurve [@tsiaras_plc], ExoTETHyS [@morello_exotethys], ArielRad [@mugnai], ExoWebb [@edwards_exowebb], Astropy [@astropy], h5py [@hdf5_collette], emcee [@emcee], Matplotlib [@Hunter_matplotlib], Multinest [@multinest], Pandas [@mckinney_pandas], Numpy [@oliphant_numpy], SciPy [@scipy].
{width="\textwidth"}
{width="\textwidth"}
![Top figure: Results of the white light-curve of WASP-79b. Top: raw light-curve, after normalisation. Second: light-curve, divided by the best fit model for the systematics. Third: residuals. Bottom: auto-correlation function of the residuals. Bottom figure: Spectral light curves fitted with Iraclis for the transmission spectra where, for clarity, an offset has been applied. Left: the detrended spectral light curves with best-fit model plotted. Right: residuals from the fitting with values for the Chi-squared ($\chi^2$), the standard deviation of the residuals with respect to the photon noise ($\bar{\sigma}$) and the auto-correlation (AC).[]{data-label="fig:wasp79_lc"}](WASP79_white.pdf "fig:"){width="\columnwidth"} ![Top figure: Results of the white light-curve of WASP-79b. Top: raw light-curve, after normalisation. Second: light-curve, divided by the best fit model for the systematics. Third: residuals. Bottom: auto-correlation function of the residuals. Bottom figure: Spectral light curves fitted with Iraclis for the transmission spectra where, for clarity, an offset has been applied. Left: the detrended spectral light curves with best-fit model plotted. Right: residuals from the fitting with values for the Chi-squared ($\chi^2$), the standard deviation of the residuals with respect to the photon noise ($\bar{\sigma}$) and the auto-correlation (AC).[]{data-label="fig:wasp79_lc"}](WASP79_LC.png "fig:"){width="0.95\columnwidth"}
![Top figure: Results of the white light-curve of WASP-62b. Top: raw light-curve, after normalisation. Second: light-curve, divided by the best fit model for the systematics. Third: residuals. Bottom: auto-correlation function of the residuals. Bottom figure: Spectral light curves fitted with Iraclis for the transmission spectra where, for clarity, an offset has been applied. Left: the detrended spectral light curves with best-fit model plotted. Right: residuals from the fitting with values for the Chi-squared ($\chi^2$), the standard deviation of the residuals with respect to the photon noise ($\bar{\sigma}$) and the auto-correlation (AC).[]{data-label="fig:wasp62_lc"}](WASP62_white.pdf "fig:"){width="0.8\columnwidth"} ![Top figure: Results of the white light-curve of WASP-62b. Top: raw light-curve, after normalisation. Second: light-curve, divided by the best fit model for the systematics. Third: residuals. Bottom: auto-correlation function of the residuals. Bottom figure: Spectral light curves fitted with Iraclis for the transmission spectra where, for clarity, an offset has been applied. Left: the detrended spectral light curves with best-fit model plotted. Right: residuals from the fitting with values for the Chi-squared ($\chi^2$), the standard deviation of the residuals with respect to the photon noise ($\bar{\sigma}$) and the auto-correlation (AC).[]{data-label="fig:wasp62_lc"}](WASP62_LC.png "fig:"){width="0.95\columnwidth"}
\[tab:mid\_times\]
\[tab:mid\_times2\]
![TESS observations of WASP-127b presented in this work. Left: detrended data and best-fit model. Right: residuals from fitting.[]{data-label="fig:wasp127_tess"}](WASP-127b_All.png){width="\columnwidth"}
![TESS observations of WASP-79b presented in this work. Left: detrended data and best-fit model. Right: residuals from fitting.[]{data-label="fig:wasp79_tess"}](WASP-79b_All.png){width="\columnwidth"}
![TESS observations of WASP-62b presented in this work. Left: detrended data and best-fit model. Right: residuals from fitting.[]{data-label="fig:wasp62_tess"}](WASP-62b_All.png){width="\columnwidth"}
![Observed minus calculated (O-C) transit mid times for WASP-127b (top), WASP-79b (middle) and WASP-62b (bottom). Transit mid time measurements from this work are shown in gold (HST) and blue (TESS), while literature T$_0$ values are in red. The black line denotes the new ephemeris of this work with the dashed lines showing the associated 1$\sigma$ uncertainties and the black data point indicating the updated T$_0$. For comparison, the previous literature ephemeris and their 1$\sigma$ uncertainties are given in red. In all cases, our results are compatible with the literature but provide a further refinement of the ephemeris.[]{data-label="fig:ephm_refine"}](WASP-127b_Deviations.png "fig:"){width="\columnwidth"} ![Observed minus calculated (O-C) transit mid times for WASP-127b (top), WASP-79b (middle) and WASP-62b (bottom). Transit mid time measurements from this work are shown in gold (HST) and blue (TESS), while literature T$_0$ values are in red. The black line denotes the new ephemeris of this work with the dashed lines showing the associated 1$\sigma$ uncertainties and the black data point indicating the updated T$_0$. For comparison, the previous literature ephemeris and their 1$\sigma$ uncertainties are given in red. In all cases, our results are compatible with the literature but provide a further refinement of the ephemeris.[]{data-label="fig:ephm_refine"}](WASP-79b_Deviations.png "fig:"){width="\columnwidth"} ![Observed minus calculated (O-C) transit mid times for WASP-127b (top), WASP-79b (middle) and WASP-62b (bottom). Transit mid time measurements from this work are shown in gold (HST) and blue (TESS), while literature T$_0$ values are in red. The black line denotes the new ephemeris of this work with the dashed lines showing the associated 1$\sigma$ uncertainties and the black data point indicating the updated T$_0$. For comparison, the previous literature ephemeris and their 1$\sigma$ uncertainties are given in red. In all cases, our results are compatible with the literature but provide a further refinement of the ephemeris.[]{data-label="fig:ephm_refine"}](WASP-62b_Deviations.png "fig:"){width="\columnwidth"}
[^1]: ARES: Ariel Retrieval of Exoplanets School
[^2]: <https://github.com/ucl-exoplanets/Iraclis>
[^3]: <https://archive.stsci.edu/hst/search>
[^4]: <https://github.com/ucl-exoplanets/TauREx3_public>
[^5]: <https://www.exoclock.space>
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We prove that every finite connected simplicial complex is homotopy equivalent to the quotient of a contractible manifold by proper actions of a virtually torsion-free group. As a corollary, we obtain that every finite connected simplicial complex is homotopy equivalent to the classifying space for proper bundles of some virtual Poincaré duality group.'
author:
- Raeyong Kim
bibliography:
- 'referencesforproject2.bib'
title: 'Every finite complex is the classifying space for proper bundles of a virtual Poincaré duality group.[^1]'
---
Introduction
============
Let $G$ be a discrete group. A $G$-CW-complex is, by definition, a CW-complex on which $G$ acts by permuting the cells and cell stabilizers act trivially on cells. A $G$-CW-complex $Y$ is said to be a model for ${\underline{{\mathrm E}}}G$ if every cell stabilizer is finite and, for every finite subgroup $H \leq G$, the fixed set $Y^{H}$ is contractible. The existence of such a model can be established by Milnor’s or Segal’s argument for the construction of the universal space for $G$. (See [@LeaNuc] for the general construction). Applying an equivariant obstruction theory proves that any two models for ${\underline{{\mathrm E}}}G$ are $G$-homotopy equivalent. We call ${\underline{{\mathrm E}}}G$ the classifying space for proper $G$-actions. We write the quotient ${\underline{{\mathrm E}}}G/G$ by ${\underline{{\mathrm B}}}G$ and call it the classifying space for proper $G$-bundles. Our main theorem can be stated as follows.
\[mainthm\] For any finite connected simplicial complex $X$, there exists a virtually torsion-free group $G$ with ${\underline{{\mathrm E}}}G$ a cocompact manifold such that ${\underline{{\mathrm B}}}G$ is homotopy equivalent to $X$.
A group $\Gamma$ is called *a Poincaré duality group of dimension $n$* if $H^{i}(\Gamma;A) \cong H_{n-i}(\Gamma;A)$ for any $\mathbb{Z}\Gamma$-module $A$. Many interesting examples of Poincaré duality groups are manifold groups. More specifically, the fundamental group of a closed, aspherical $n$-dimensional manifold is a Poincaré duality group of dimension $n$. (The converse is false.) See [@Bro], [@DavPOINCARE] for details about Poincaré duality groups. Finally, recall that a group *virtually* has some property if a finite index subgroup has the property.
Let $T$ be a torsion-free finite index subgroup of $G$ in Theorem \[mainthm\]. Then one can take ${\underline{{\mathrm E}}}G$ as a model for ${\underline{{\mathrm E}}}T = ET$, where $ET$ is the universal space for $T$. Since $ET/T$ is a closed aspherical manifold, $T$ is a Poincaré duality group.
\[maincor\] For any finite connected simplicial complex $X$, there is a virtual Poincaré duality group $G$ such that ${\underline{{\mathrm B}}}G$ is homotopy equivalent to $X$.
The statement of Corollary \[maincor\] is related to the theorem of Kan-Thurston, which says that every connected complex has the same homology as the classifying space for some group (See [@KanThu]). This theorem has been extended and generalized by a number of authors. For example, see [@BauDyeHel],[@Mau],[@Hau],[@LeaMKT],[@McD],[@LeaNuc],[@RKim]. Among many extensions and generalizations, Leary and Nucinkis proved in [@LeaNuc] that every connected CW-complex has the same homotopy type as the classifying space for proper bundles of some group. Corollary \[maincor\] says that the group can be taken as a virtual Poincaré duality group if the simplicial complex is finite.
The proof of Theorem \[mainthm\] consists of three steps. In Section \[equiembedding\], we outline the embedding trick, due to Floyd [@Flo], for equivariantly embedding a simplicial complex with an involution into some Euclidean space. In Section \[equireflection\], we review the equivariant reflection group trick, due to Davis [@DavLea]. Finally, in Section \[resultproof\], we use the two tricks to construct a contractible manifold, whose quotient by some group $G$ is homotopy equivalent to a given finite simplicial complex $X$. We also prove that the contractible manifold is the classifying space for proper $G$-actions and introduce the torsion-free subgroup of finite index to complete the proof of Theorem \[mainthm\].
The paper is part of author’s Ph.D. thesis. The author thanks the thesis advisor Ian Leary for his guidance throughout this research project. The author also thanks Jean Lafont for his careful reading of an earlier version of this paper.
The Equivariant Embedding Trick {#equiembedding}
===============================
Let $Z$ be a finite simplicial complex with a simplicial map of period $p$. In [@Flo], Floyd introduced the embedding trick, namely, $Z$ can be embedded in some Euclidean space such that the restriction of specific coordinate changing map on $Z$ is the given simplicial map. We outline his construction in the case that $p=2$. See [@Flo Section 2] for the full construction.
Let $L$ be a finite connected simplicial complex with a simplicial involution $T$. Embed $L$ into ${{\mathbb R}}^{n}$ for some $n$ and suppose that ${{\mathbb R}}^{n}$ is triangulated so that $L$ is a subcomplex. Consider the following map. $$\phi : L \to {{\mathbb R}}^{n} \times {{\mathbb R}}^{n}(={{\mathbb R}}^{2n}) ,\qquad x \mapsto (x,T(x)).$$
Note that a cell in the cellular decomposition of ${{\mathbb R}}^{2n}$ has the type of $s_1 \times s_2$, where each $s_i$ is a simplex in ${{\mathbb R}}^{n}$. We use the first barycentric subdivision of this cellular decomposition for the subdivision of ${{\mathbb R}}^{2n}$.
By passing to the barycentric subdivision ${\mathrm{Sd}}(L)$ of $L$, we obtain that $\phi$ is a simplicial homeomorphism of ${\mathrm{Sd}}(L)$ onto a subcomplex of ${{\mathbb R}}^{2n}$. Furthermore, the map $S : {{\mathbb R}}^{n} \times {{\mathbb R}}^{n} \to {{\mathbb R}}^{n} \times {{\mathbb R}}^{n}$ defined by $S(x,y)=(y,x)$ is simplicial and satisfies $S\circ\phi=\phi\circ T$, hence $\phi$ is an equivalence between $({\mathrm{Sd}}(L),T)$ and $(\phi({\mathrm{Sd}}(L)),S)$. Hereafter, we suppose that $L$ is a subcomplex of ${{\mathbb R}}^{2n}$ and $S=T$ on $L$.
We may as well assume that $L$ is a full subcomplex of ${{\mathbb R}}^{2n}$. For if not, ${\mathrm{Sd}}(L)$ is a full subcomplex of ${\mathrm{Sd}}({{\mathbb R}}^{2n})$. Let $U$ be the first regular neighborhood of $L$, i.e. the union of all open stars of vertices of ${\mathrm{Sd}}(L)$ relative to ${\mathrm{Sd}}({{\mathbb R}}^{2n})$. Then $K=\overline{U}$ is a manifold with boundary of dimension $2n$. Denote the boundary of $K$ by $\partial K$.
Let $v_0, \cdots, v_{k}, v_{k+1}, \cdots$ be vertices of ${{\mathbb R}}^{2n}$, where $v_0,\cdots, v_{k}$ are vertices of $L$. Every point $x$ in ${{\mathbb R}}^{2n}$ has a unique barycentric representation $\displaystyle{\sum t_{i}v_{i}}$. Furthermore, $K$ consists of points $x$ for which $\mathrm{max}(t_0, \cdots , t_k) \geq \mathrm{max}(t_{k+1}, \cdots )$ and $\partial K$ consists of points $x$ for which $\mathrm{max}(t_0, \cdots , t_k) = \mathrm{max}(t_{k+1}, \cdots )$. Consider $f : K \to L$ defined by $\displaystyle{f(x) = f(\sum t_{i}v_{i}}) = \frac{\sum_{i=0}^{k} t_{i}v_{i}}{\sum_{i=0}^{k} t_{i}}$. Then $\Phi : K \times I \to K$ defined by $\Phi(x,t) = (1-t)x+ tf(x)$ is a deformation retract of $K$ onto $L$.
By the unique barycentric representation of points in ${{\mathbb R}}^{2n}$, $\Phi(S(x),t) = S(\Phi(x),t)$ for any $x \in K$ and $t \in I$ so that $\Phi$ is $S$-equivariant. Therefore, $K$ equivariantly deformation retracts onto $L$. In particular, the fixed set $K^S$ is homotopy equivalent to the fixed set $L^T$ and $K/\langle S\rangle$ is homotopy equivalent to $L/\langle T \rangle$.
The Equivariant Reflection Group Trick {#equireflection}
======================================
Suppose that we are given a space $M$ and a subspace $N \subset M$ such that $N$ is triangulated as a finite dimensional flag complex. Recall that a simplicial complex is a *flag complex* if any finite set of vertices, which are pairwise connected by edges, spans a simplex. Let a discrete group $G$ act on $M$ so that $G$ stabilizes the subspace $N$ and $G$ acts on $N$ by simplicial automorphisms. Following [@DavLea], we will associate a right-angled Coxeter group $W$ and construct a $(W \rtimes G)$-action on a space ${{\mathcal U}}(M,N,G)$.
Let $I$ be a vertex set of $N$. Define a right-angled Coxeter group $W$ as follows. There is one generator $s_{i}$ for each $i \in I$. Relations are given by $s_{i}^{2} = 1$ and $(s_{i}s_{j})^{2} = 1$ if $\{i,j\}$ spans an edge in $N$. For $x \in N$, let $\sigma(x) = \{i \in I | x \in N_i\}$, where $N_i$ is the closed star of the vertex $i$ in the barycentric subdivision of $N$ and $W_{x}$ be the subgroup generated by $\{s_i | i \in \sigma(x)\}$.
Note that $G$ acts on $N$ by permuting vertices, so we can form $W \rtimes G$.
Define the space ${{\mathcal U}}(M,N,G)$ by $${{\mathcal U}}(M,N,G) = W \times M / \sim,$$ where $(w,x) \sim (w',x')$ if $x=x'$ and $w^{-1}w' \in W_{x}$. For $[w,x] \in {{\mathcal U}}(M,N,G)$ and $(v,g) \in W \rtimes G$, the action of $W\rtimes G$ on ${{\mathcal U}}(M,N,G)$ is defined by $$(v,g).[w,x] = [vw^{g},g .x].$$
This construction enjoys the following properties. For details or proofs, see [@DavBOOK Section 11.7], [@Dav].
1. If $M$ is contractible, then so is ${{\mathcal U}}(M,N,G)$.
2. If $M$ is an $n$-dimensional manifold with boundary and $N= \partial M$, then ${{\mathcal U}}(M,N,G)$ is an $n$-dimensional manifold.
3. Let $C(N)$ be the cone on $N$. Then ${{\mathcal U}}(C(N),N,G)$ has a natural $\CAT(0)$ cubical structure so that the link of each vertex is isomorphic to $N$, and so that $W \rtimes G$ acts by a group of isometries. In particular, for any finite subgroup $F$ of $W \rtimes G$, the fixed point set ${{\mathcal U}}(C(N),N,G)^{F}$ is contractible.
A group action on a simplicial complex is said to be *admissible* if, for any simplex, setwise stabilizers are equal to pointwise stabilizers. If the $G$-action on $N$ is admissible, we have the following.
\[finitesubgroup\] Let $H$ be a finite subgroup in $G$. Then $${{\mathcal U}}(M,N,G)^{H} = {{\mathcal U}}(M^{H},N^{H},V_{H}),$$ where $M^{H}$, and $N^{H}$ respectively, is the $H$-fixed set in $M$, and $N$ respectively, and $V_{H} = N_{G}(H)/H$.
The above lemma is stated in [@DavBOOK Proposition 11.7.1] without a proof. We provide the proof below. Note also that $N^{H}$ is a flag complex.
It is obvious that ${{\mathcal U}}(M^{H},N^{H},V_{H})$ is a subspace of ${{\mathcal U}}(M,N,G)$.
Let $[w,m] \in {{\mathcal U}}(M,N,G)^{H}$ be given. In order to prove that ${{\mathcal U}}(M,N,G)^{H}$ is contained in ${{\mathcal U}}(M^{H},N^{H},V_{H})$, it suffices to show that $m \in M^{H}$ and $w \in W_{H}$, where $W_{H}$ is the subgroup of $W$ generated by $\{s_{i}| i \,\,\textrm{is a vertex in}\,\, N^{H}\}$. For any $h \in H$, $$\begin{array}{cc} &(1,h).[w,m] = [w^{h},h.m] = [w,m]\\\Rightarrow&h.m = m,\qquad w^{-1}w^{h} \in W_{m}.\end{array}$$
Since $h.m=m$ for all $h \in H$, it follows that $m \in M^{H}$. We prove that $w \in W_{H}$ by induction on the length $l(w)$ of $w$. To begin with, we point out that since the $G$-action on $M$ is admissible, every generator in $W_{m}$ is fixed by $H$. In particular, $W_{m}$ is a subgroup in $W_{H}$. Also note that $W_{m}$ is finite. (Consider the simplex containing $m$ of minimal dimension.)
Suppose that $l(w)=1$, i.e. $w=w^{-1}$ . Since $W_{m}$ is finite, $ww^{h}$ has finite order. But this happens only if two vertices corresponding to $w$ and $w^{h}$ are connected. Admissibility implies that $w$ is fixed by $h$, and hence, $w \in W_{H}$.
Suppose that $w = s_1 \cdots s_n$ is a reduced word (of length $n$). Let $t_i = s_{i}^{h}$.
Suppose $s_{1} \neq t_{1}$. Again, $w^{-1}w^{h}$ has finite order. $$\begin{array}{ccc}& (w^{-1}w^{h})^{n} = 1\,\, \mathrm{for\,\,some}\,\, n\\\Rightarrow&(s_n \cdots s_1\cdot t_1 \cdots t_n) (s_n \cdots s_1\cdot t_1 \cdots t_n)\cdots(s_n \cdots s_1\cdot t_1 \cdots t_n)=1\end{array}$$
In order for the left hand side to be reduced to the identity, in particular, there exists $t_{i}$ for $2 \leq i \leq n$ such that $t_{i}$ commutes with $t_{1}$ and cancels with $s_{1}$, i.e. $t_{i}=s_{1}$. But this is impossible, because $s_{1}$ and $t_{1}$ do not commute. Therefore, $s_{1}=t_{1}$.
$$w^{-1}w^{h} = s_n \cdots s_1\cdot t_1 \cdots t_n = s_n \cdots s_2\cdot t_2 \cdots t_n \in W_{m}$$ By the induction hypothesis, $s_{1} w \in W_{H}$, so $w \in W_{H}$.
The proof of Theorem \[mainthm\] {#resultproof}
================================
The proof of Theorem \[mainthm\] consists of three steps. First, we use the equivariant embedding trick to embed a given finite simplicial complex $X$ into the manifold $M$ with an involution $\tau$ such that $M/\langle \tau \rangle$ is homotopy equivalent to $X$. Then we apply the equivariant reflection group trick on $M$ with boundary to obtain a contractible manifold on which some group $G$ acts. The quotient of the contractible manifold by $G$ will be homotopy equivalent to $X$. Finally, we show that the contractible manifold is the classifying space for proper $G$-actions. Additionally, we introduce a finite index torsion-free subgroup of $G$, which proves that $G$ is a virtual Poincaré duality group.\
Let $X$ be a finite connected simplicial complex. Note that the equivariant embedding trick requires a simplicial complex with a periodic simplicial map. In this paper, we use the construction appearing in [@LeaMKT]. Applying [@LeaMKT Theorem A], we obtain a finite connected locally $\CAT(0)$ cubical complex $Y$ with a cubical involution $\tau$ such that $Y/\langle\tau\rangle$ is homotopy equivalent to $X$. By passing to the barycentric subdivision, we may assume that $Y$ is a finite connected simplicial complex and $\tau$ is a simplicial involution on $Y$. Now we apply the equivariant embedding trick introduced in Section \[equiembedding\] to obtain a manifold $M$ with a boundary $N$ and a simplicial involution $\omega$ on $M$ such that $M$ equivariantly deformation retracts onto $Y$. By passing to the barycentric subdivision, we can assume that $N$ is a flag complex and a cyclic group of order two, $C_{2}=\langle \omega \rangle$, acts on $N$ admissibly. Note that $Y$ is locally $\CAT(0)$. Therefore, $M$ is aspherical and $M^{\omega}$ is homotopy equivalent to $Y^{\tau}$.
Next we apply the equivariant reflection group trick from Section \[equireflection\] to obtain a space ${{\mathcal U}}(M,N,C_{2})$ on which $W \rtimes C_{2}$ acts, where $W$ is a right-angled Coxeter group associated to $N$. Let ${{\widetilde M}}$ be the universal cover of $M$, ${{\widetilde N}}$ be the inverse image of $N$ in ${{\widetilde M}}$ and ${{\widetilde W}}$ be the associated right-angled Coxeter group to ${{\widetilde N}}$. Repeat the equivariant reflection group trick to obtain a space ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)$ on which ${{\widetilde W}}\rtimes \Gamma$ acts, where $\Gamma$ is the group of liftings of the $C_{2}$-action on $M$ to ${{\widetilde M}}$. Note that every torsion element in $\Gamma$ has order at most two and every finite subgroup of $\Gamma$ is cyclic of order two.
\[equi\] Let ${{\mathcal U}}(M,N,C_2)$ and ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)$ be the spaces constructed above. Then
1. \[p1\] ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)$ and ${{\mathcal U}}(M,N,C_2)$ are manifolds.
2. \[p2\] ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)$ is the universal cover of ${{\mathcal U}}(M,N,C_2)$.
3. \[p3\] ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)/({{\widetilde W}}\rtimes \Gamma)$ is homotopy equivalent to $X$.
The first statement follows from the fact that $M$ and ${{\widetilde M}}$ are manifolds with boundary. It is clear that ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)$ is a cover of ${{\mathcal U}}(M,N,C_2)$. Since $M$ is aspherical, ${{\widetilde M}}$ is contractible and so is ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)$. This proves the second statement. By construction,
$${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)/({{\widetilde W}}\rtimes \Gamma) \simeq {{\mathcal U}}(M,N,C_2)/ (W\rtimes C_{2}) \simeq M/C_{2} \simeq X,$$ where $\simeq$ is a homotopy equivalence.
We compare the $({{\widetilde W}}\rtimes \Gamma)$-action on the space ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)$ with the same action on ${{\mathcal U}}(C({{\widetilde N}}),{{\widetilde N}},\Gamma)$, and prove ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma) = {\underline{{\mathrm E}}}({{\widetilde W}}\rtimes \Gamma)$. Denote the image of $\{{{\widetilde w}}\} \times {{\widetilde M}}$ in ${{\mathcal U}}({{\widetilde M}}, {{\widetilde N}}, \Gamma)$ by ${{\widetilde w}}{{\widetilde M}}$ and the image of $\{{{\widetilde w}}\} \times ({{\widetilde M}}\setminus {{\widetilde N}})$ by $int({{\widetilde w}}{{\widetilde M}})$. First, we prove that all stabilizers are finite.
Let $H$ be a subgroup of ${{\widetilde W}}\rtimes \Gamma$ fixing some point in ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)$. Then $H$ is finite.
Suppose that $H$ fixes some point in $int({{\widetilde w}}{{\widetilde M}})$ for some ${{\widetilde w}}$.
Then $H' = ({{\widetilde w}},1)^{-1}H({{\widetilde w}},1)$ fixes some point in $int({{\widetilde M}})$. Denote this point by $[1,x]$. For any $({{\widetilde v}},\gamma) \in H'$, $$({{\widetilde v}},\gamma).[1,x] = [{{\widetilde v}},\gamma.x] =[1,x] \Rightarrow \gamma.x=x,\,\, {{\widetilde v}}\in {{\widetilde W}}_{x}$$
Since $x \in {{\widetilde M}}\setminus {{\widetilde N}}$(recall that ${{\widetilde N}}=\partial {{\widetilde M}}$, and $x \in int({{\widetilde M}})$.), ${{\widetilde W}}_{x}$ is trivial. It follows that $H'$ is a subgroup of $\Gamma$. Recall that $\Gamma$ is the group of liftings of the $C_{2}$-action on $M$ to ${{\widetilde M}}$. Therefore, if a nontrivial element $\gamma$ fixes some point $x$, $\gamma$ is the only nontrivial element in $\Gamma$ fixing $x$. This tells us that $H'$ must be finite of order $2$.
Suppose that the fixed point is not contained in $int({{\widetilde w}}{{\widetilde M}})$ for any ${{\widetilde w}}$. As in the previous case, choose some ${{\widetilde w}}'$ so that $H''=({{\widetilde w}}',1)^{-1}H({{\widetilde w}}',1)$ fixes some point in the image of ${{\widetilde N}}$ in ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)$. Denote such a point by $[1,y]$. For any $({{\widetilde v}}',\gamma') \in H''$, $$({{\widetilde v}}',\gamma').[1,y] = [{{\widetilde v}}',\gamma'.y] =[1,y] \Rightarrow \gamma'.y=y,\,\, {{\widetilde v}}' \in {{\widetilde W}}_{y}$$
As before, we have at most two possibilities for $\gamma'$. Furthermore, ${{\widetilde W}}_{y}$ is finite. (Consider the simplex containing $y$ of minimal dimension.) Therefore, $H''$ is finite, and hence, $H$ is finite.
\[clas\] ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma) = {\underline{{\mathrm E}}}({{\widetilde W}}\rtimes \Gamma)$.
It suffices to prove that the fixed point set by a finite subgroup is contractible. As mentioned before, we consider the $({{\widetilde W}}\rtimes \Gamma)$-action on ${{\mathcal U}}(C({{\widetilde N}}),{{\widetilde N}},\Gamma)$.
Let $F$ be a finite subgroup of ${{\widetilde W}}\rtimes \Gamma$. Recall that ${{\mathcal U}}(C({{\widetilde N}}),{{\widetilde N}},\Gamma)^{F}$ is contractible. (See Remark 4.) In particular, it is nonempty.
Suppose that $F$ does not fix any cone point in ${{\mathcal U}}(C({{\widetilde N}}),{{\widetilde N}},\Gamma)$. Then $F$ fixes no point in $int({{\widetilde w}}{{\widetilde M}})$ for any ${{\widetilde w}}$. In other words,
$${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)^{F} \subset \displaystyle{{{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma) - \bigcup_{{{\widetilde w}}\in {{\widetilde W}}} int({{\widetilde w}}{{\widetilde M}})}.$$ Therefore,
$${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)^{F} = {{\mathcal U}}(C({{\widetilde N}}),{{\widetilde N}},\Gamma)^{F}$$ and ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)^{F}$ is contractible.
Suppose that $F$ fixes some cone point in ${{\mathcal U}}(C({{\widetilde N}}),{{\widetilde N}},\Gamma)$. Choose some ${{\widetilde w}}''$ so that $F' = ({{\widetilde w}}'',1)^{-1}F({{\widetilde w}}'',1)$ fixes the cone point of $int(C{{\widetilde N}})$. Denote the cone point by $c$. For any $({{\widetilde v}}'',\gamma'') \in F'$, $$({{\widetilde v}}'',\gamma''),[1,c] = [{{\widetilde v}}'',\gamma''.c] = [1,c] \Rightarrow \gamma''.c=c,\,\,{{\widetilde v}}'' \in {{\widetilde W}}_{c}.$$
Since $c$ is a cone point, ${{\widetilde v}}''$ is trivial. It follows that $F'$ is a finite subgroup of $\Gamma$, and hence, cyclic of order $2$. By Lemma \[finitesubgroup\], it follows that ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)^{F'}$ is ${{\mathcal U}}({{\widetilde M}}^{F'},{{\widetilde N}}^{F'},V_{F'})$. Recall $M$ is equivariantly homotopy equivalent to a locally $\mathrm{CAT}(0)$ space $Y$. Therefore, ${{\widetilde M}}^{F'}$ is homotopy equivalent to the fixed set of $CAT(0)$ space by a cyclic group of order $2$. It follows that ${{\widetilde M}}^{F'}$ is contractible, hence so is ${{\mathcal U}}({{\widetilde M}}^{F'},{{\widetilde N}}^{F'},V_{F'})$. Finally, ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)^{F} = ({{\widetilde w}}'',1).{{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma)^{F'}$ is contractible.
Consider the commutator subgroup $T$ of $W$ and its inverse image ${{\widetilde T}}$ in ${{\widetilde W}}$. Since ${{\widetilde T}}$ is torsion-free and of finite index in ${{\widetilde W}}$, ${{\widetilde T}}\rtimes \pi_{1}(M)$ is a torsion-free finite index subgroup of ${{\widetilde W}}\rtimes \Gamma$. Since ${{\mathcal U}}({{\widetilde M}},{{\widetilde N}},\Gamma) / ({{\widetilde T}}\rtimes \pi_{1}(M))$ is an aspherical closed manifold, ${{\widetilde T}}\rtimes \pi_{1}(M)$ is a Poincaré duality group. This verifies that ${{\widetilde W}}\rtimes \Gamma$ is a virtual Poincaré duality group.
Raeyong Kim:
Department of Mathematics, The Ohio State University,
231 West 18th Avenue, Columbus, Ohio 43210-1174, United States.
[[email protected]]{}
[^1]: 2010 *AMS Subject Classification* 55R35 (57P10, 20J05, 57S99)
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'We provide a representation of the homomorphisms $U\longrightarrow \R$, where $U$ is the lattice of all uniformly continuous on the line. The resulting picture is sharp enough to describe the fine topological structure of the space of such homomorphisms.'
address: 'Departamento de Matemáticas, UEx and IMUEx, 06071-Badajoz, Spain'
author:
- Félix Cabello Sánchez
title: Fine structure of the homomorphisms of the lattice of uniformly continuous functions on the line
---
=2000
\
Introduction
============
The purpose of this short note is to describe, as accurately as possible, the real-valued homomorphisms of the lattice of all uniformly continuous functions on the (half) line.
We denote by $U(X)$ the lattice of all real-valued uniformly continuous functions on $X$, which is invariably assumed to be a metric space. The sublattice of bounded functions is denoted by $U^*(X)$. When $X$ is the half-line $\H=[1,\infty)$ with the distance given by the absolute value we just write $U$ and $U^*$.
By a homomorphism of vector lattices we mean a linear map preserving joins and meets (equivalently, absolute values). Given a vector lattice $\mathscr L$, we denote by $H(\mathscr L)$ the set of all homomorphisms $\phi:\mathscr L\longrightarrow\mathbb R$.
We are interested in $H(U)$ for two good reasons, apart from sheer curiosity. The first and most obvious one is that, given an object of a category, the study of the homomorphisms against the “simplest object” in the category (if there is one) is interesting in its own right and often enlights the initial object. Quite clearly, $\R$ can be considered as the simplest vector lattice.
The study of the lattices of uniformly continuous functions and their homomorphisms has spurred a sustained, though moderate, interest for some time now; see the papers [@shi; @ga-ja; @javi; @hus; @hus-pul] and their references.
The second motivation springs from the circle of ideas around the Samuel-Smirnov compactification, a classical construction in topology; see [@s] and [@wi Chapter 9, §41]. The space $H(\mathscr L)$ can be given the relative product topology it inherits from $\R^\mathscr L$. The Samuel-Smirnov compactification of $X$ is then the subspace of those homomorphisms $\phi: U^*(X)\longrightarrow\R$ that are unital in the sense that they send the function [**1**]{} to the number $1$. This construction has attracted a considerable attention, even for very simple choices of the base space, such as $X=\R^n$; see [@w; @japos]
Very recently Garrido and Meroño [@g-m] have used the unital homomorphisms on $U(X)$ to construct a realcompactification of $X$ which plays the same role for general uniformly continuous functions than the Samuel-Smirnov compactification for those which are bounded.
Our modest contribution to this line of research is a description of the topological space $H(U)$ in the spirit of Woods’ [@w Section 4]. As we shall see, the most interesting homomorphisms are not unital and actually they vanish on every bounded function. So, somehow, there is a better life beyond 1.
We are aware of the fact that this is just one example and that most of the arguments presented here depend heavily on the peculiarities of the line.
This is compensated in part by the chief role played by the line amongst metric spaces as well as by the neat description of $H(U)$ that is achieved.
Moreover, as a byproduct, we compute the spaces of homomorphisms of other important lattices such as $U(\R)$ and $\Lip(\H)$ or $\Lip(\N)$.
Elementary stuff
================
This part contains a rather pedestrian description of $H(U)$, based on ultrafilters over the positive integers. If $\mathscr L$ is a vector lattice, then $H(\mathscr L)$ is a subset of $\R^\L$. This can be used to transfer to $H(\mathscr L)$ the product topology of $\R^\L$: a typical neighbourhood of $\phi$ has the form $$\{\psi\in H(\mathscr L): |\phi(f_i)-\psi(f_i)|<\varepsilon \text{ for $1\leq i\leq n$}\},$$ where $f_1,\dots, f_n\in \mathscr L$ and $\varepsilon >0$. This is the only topology that we will consider on $H(\mathscr L)$.
Let us begin with the observation that every $t\in\H$ gives rise to a unital homomorphism $\delta_t:U\longrightarrow\R$ by evaluation: $\delta_t(f)=f(t)$. It is *really* easy to see that all unital homomorphisms arise as evaluation at some $t\in\H$. Indeed, put $$H_{\bf 1}=\{\phi\in H(U): \phi({\bf 1})=1\}.$$ Let us first check that these evaluations are dense in $H_{\bf 1}$. Pick $\phi\in H_{\bf 1}$, finitely many functions $f_i\in U$ and $\e>0$. We have to find a point $t\in \H$ such that $$|\phi(f_i)-f_i(t)|<\e\quad\quad(1\leq i\leq n).$$ If we assume that no such $t$ exists, then, letting $c_i=\phi(f_i){\bf 1}$, we have $$\bigvee_{1\leq i\leq n}|c_i {\bf 1}- f_i|\geq \e{\bf 1}.$$ But $$\phi\left(\bigvee_{1\leq i\leq n}|c_i {\bf 1}- f_i|\right)
=
\bigvee_{1\leq i\leq n}\phi(|c_i {\bf 1}- f_i|)
=
\bigvee_{1\leq i\leq n}|c_i \phi({\bf 1})- \phi(f_i)| =0,$$ while $\phi(\e{\bf 1})=\e$, a contradiction.
Thus, given $\phi\in H_{\bf 1}$ we can find a net $(t_\alpha)$ such that $(\delta_{t_\alpha})$ converges to $\phi$ in $H(U)$. In particular we have $$\phi(f)=\lim_\alpha f(t_\alpha)$$ for every $f\in U$. Taking $f={\frak t}$ as the identity of $\H$ and $t^*=\phi({\frak t})\geq 1$, we have $t^*=\lim_\alpha t_\alpha
$ and so $\phi=\delta_{t^*}$. We have thus proved:
$\bigstar$ *A homomorphism $\phi: U\longrightarrow \R$ has the form $\phi=c\delta_t$ for some $t\in\H$ and $0<c<\infty$ if and only if $\phi({\bf 1})>0.$ Otherwise $\phi$ vanishes on every bounded function.*
The “otherwise” part is due to the fact that if $\phi({\bf 1})=0$, then $\phi(f)=0$ for every bounded $f\geq 0$ since $f\leq n{\bf 1}$ for some $n\in\N$ and so $\phi(g)=0$ for every $g\in U^*$ since such a $g$ is the difference of two nonnegative functions in $U^*$.
It is clear that the preceding proof relies on Heine-Borel theorem as it depends on the local compactness of the line. Let us remark, however, that $H(U^*)$ contains many unital homomorphisms which are not evaluations at points of $\H$. These form the Samuel-Smirnov compactification of $\H$; see [@s] and specially Section 4 in Woods’ classical paper [@w].
To see how these “outer” homomorphisms arise, take any uniformly separated sequence $(t_n)$, that is, such that $|t_n-t_k|>\e$ for some positive $\e$ and every $n\neq k$, and let $\mathscr U$ be a free ultrafilter on the integers. Then set $$\phi(f)=\lim_{\mathscr U(n)}f(t_n)\quad\quad(f\in U^*).$$
The space $H(U)$ contains outer homomorphisms as well. These have to vanish on $U^*$ and, as we shall see, also at each function $f$ such that $t^{-1}f(t)\to 0$ as $t\to\infty$.
The main property of the half-line required here is that every $f\in U$ is Lipschitz for large distances: for every $\e>0$ there is a constant $L$, depending on $\e$ and $f$, such that $$|f(s)-f(t)|\leq L|s-t|\quad\text{provided}\quad |s-t|\geq\e;$$ see [@b-l Proposition 1.11] or [@turkiyos Lemma 2.2] for the easy proof. In particular the limit $L(f)=\limsup_{t\to\infty}t^{-1}|f(t)|$ is finite for every $f\in U$, which implies that for every $f\in U$ there is $c>0$ such that $c|f|\leq {\frak t}$ and so each homomorphism vanishing at $\frak t$ has to be zero. Thus, one can use $\phi({\frak t})$ to measure size in $H(U)$.
Note that there are (many) unbounded functions in $U$ such that $L(f)=0$, for instance $f(t)=t^\alpha$ for $0<\alpha<1$ or $f(t)=\log t$.
Going back to $H(U)$, let $\U$ be a free ultrafilter on $\mathbb N$ and put $$\label{phiU}
\phi_\U(f)=\lim_{\mathscr U(n)}\frac{f(n)}{n}\quad\quad(f\in U).$$ Clearly, $\phi_\U$ is correctly defined, belongs to $H(U)$, vanishes on $U^*$ and $\phi_\U(\frak t)=1$. Note that only the values of $f$ at the integers are used in the definition of $\phi_\U$. Now, consider the following subsets of $H(U)$: $$\begin{aligned}
H_\frak t&=\{\phi\in H(U): \phi(\frak t)=1\},\\
H_\frak t^0&=\{\phi\in H(U): \phi(\frak t)=1 \text{ and } \phi({\bf 1})=0\}.\end{aligned}$$ Every nonzero $\phi\in H(U)$ falls into $H_\frak t$ after renormalization: just take $\phi(\frak t)^{-1}\phi$.
{#our}
Our immediate aim is to show that every $\phi\in H_\frak t^0$ comes from a free ultrafilter on $\N$, as in (\[phiU\]).
Let us first check that the closure of the set $\{n^{-1}\delta_n:n\in\N\}$ in $H(U)$ contains $H_\frak t^0$.
This amounts to verifying that, given $\phi\in H_\frak t^0, f_i,\dots, f_k\in U$ and $\e>0$ there is $n\in\N$ such that $$\left|\phi(f_i)-\frac{f_i(n)}{n}\right|<\e\quad\quad(1\leq i\leq k).$$ Assuming the contrary we have $$\bigvee_{1\leq i\leq k}\left|n\phi(f_i)-f_i(n)\right|\geq \e n\quad\quad(n\in\N).$$ Letting $c_i=\phi(f_i)$ and taking into account that a uniformly continous function on $\H$ is bounded if and only if it bounded on $\N$, we see that the function $$g= 0 \wedge \left(\bigvee_{1\leq i\leq k}\left|c_i{\frak t}-f_i\right|-\e {\frak t}\right)$$ belongs to $U^*$, since it vanishes on $\mathbb N$, and satisfies $$\bigvee_{1\leq i\leq k}\left|c_i{\frak t}-f_i\right|\geq \e {\frak t}+g,$$ which cannot be since $$\phi\left( \bigvee_{1\leq i\leq k}\left|c_i{\frak t}-f_i\right|\right)=
\bigvee_{1\leq i\leq k}\left|c_i\phi({\frak t})-\phi(f_i)\right|=0,$$ while $\phi(\e {\frak t}+g)= \e\phi( {\frak t})+\phi(g)=\e$.
Now, let us fix $\phi\in H_\frak t^0$. It is clear that there is a filter on the integers, say $\mathscr F$, containing every set of the form $
\{n\in \N: n^{-1}\delta_n\in V\},
$ where $V$ runs over the neighbourhoods of $\phi$ in $H(U)$. Now, if $\mathscr U$ is any ultrafilter refining $\mathscr F$, then $\phi=\phi_\U$ since for every $f\in U$ one has $$\phi(f)=\lim_{\mathscr F(n)}\frac{f(n)}{n}= \lim_{\mathscr U(n)}\frac{f(n)}{n}= \phi_\U(f).$$ We therefore have:
$\bigstar$ *Let $\phi\in H_\frak t$. If $\phi({\bf 1})>0$, then $\phi=t^{-1}\delta_t$ for some $t\in\H$. Otherwise there is a free ultrafilter $\U$ on the positive integers such that $\phi=\phi_\U$, as in $(\ref{phiU})$.*
Hence, if $\phi$ vanishes at ${\bf 1}$ it also vanishes at every function with $L(f)=0$.
Of course we are proud of this statement. However in its present form it cannot be used to detect when and why two ultrafilters induce the same homomorphism. This question leads to very interesting maths, as we will see in the next Section.
More advanced stuff
===================
In all what follows we denote by $\beta\N$ the Stone-Čech compactification of the positive integers and $\N^*=\beta\N\backslash\N$ will be the remainder. We understand each element of $\N^*$ as a free ultrafilter on $\N$ and each point of $\N$ as a fixed ultrafilter. Let, as usual, $\ell_\infty$ denote the algebra of all bounded functions on $\mathbb N$, with the pointwise operations and order.
As it is well-known, the ultrafilters on $\N$ are in exact correspondence with the algebra homomorphisms $\ell_\infty\longrightarrow \R$ through the formula $$f\longmapsto \lim_{\U(n)}f(n).$$
{#declare}
Let us declare the ultrafilters $\U$ and $\mathscr V$ equivalent (and write $\U\approx\mathscr V$ for short) if they induce the same homomorphism on $U$, that is, when $\phi_\U=\phi_\mathscr V$. While two ultrafilters inducing the same homomorphism on $\ell_\infty$ actually agree, this is not the case for the notion of equivalence we have just introduced.
To see this, take $\U\in\N^*$ and put $\mathscr V=1+\U$, that is, the sets of $\mathscr V$ are obtained by translating those of $\U$ by a unit. Then for $f\in U$ we have $$\phi_\mathscr V(f)=\lim_{\mathscr V(n)}\frac{f(n)}{n}
=\lim_{\mathscr U(n)}\frac{f(n+1)}{n+1}=\lim_{\mathscr U(n)}\frac{f(n)}{n}= \phi_\mathscr U(f)$$ since $f(n+1)-f(n)$ is bounded. Needless to say $\U$ and $\mathscr V$ are different as exactly one of them contains the set of even numbers.
Let us explain the notion of the image of an ultrafilter, which is implicit in the construction of the “translate” $1+\U$. Let $g:X\longrightarrow Y$ be a mapping, where $X$ and $Y$ are sets with no additional structure. If $\U$ is an ultrafilter on $X$, then the image of $\U$ under $g$ is the ultrafilter $$\mathscr V= g[\U]=\{B\subset Y: g^\leftarrow[B]\in\U\}.$$ Quite clearly, if $K$ is a compact Hausdorff space and $f:Y\longrightarrow K$ is any mapping, then one has $$\lim_{\mathscr V(y)}f(y)= \lim_{\mathscr U(x)}f(g(x)).$$ In this way $1+\U$ is just the image of $\U$ under the translation $1+\bullet:\N\longrightarrow\N$ given by $(1+\bullet)(n)=1+n$.
Now, the idea is that if $\U\neq \mathscr V$ and $g:\N\longrightarrow\N$ increases fast enough, then $g[\U]\not\approx g[\V]$.
Indeed, consider the function $2^\bullet:\N\longrightarrow\N$ defined by $2^\bullet(n)=2^n$. Let $\mathscr U$ and $\mathscr V$ be two different ultrafilters on $\N$. We will prove that $\phi_{2^{\bullet}[\U]}\neq \phi_{2^{\bullet}[\V]}$, that is, that there is $f\in U$ such that $$\label{distinto}
\lim_{\mathscr U(n)}\frac{f(2^n)}{2^n}\neq \lim_{\mathscr V(n)}\frac{f(2^n)}{2^n}$$ Let $A$ be a witness set, so that $A$ belongs to $\U$ but not to $\V$. We define a Lipschitz $f:\H\longrightarrow\R$ as follows. First, we put $$f_0(2^n)=\begin{cases}
2^n& \text{if $n\in A$}\\
0 & \text{if $n\notin A$}
\end{cases}$$ and $f_0(1)=1$ which corresponds to $n=0$. Then we extend $f_0$ to a piecewise linear function on $\H$ thus: write $t\in [2^n,2^{n+1}]$ as $t=(1-s)2^n+s2^{n+1}$ with $0\leq s\leq 1$ and put $$f(t)=(1-s)f_0(2^n)+sf_0(2^{n+1}).$$ The resulting function is Lipschitz (hence uniformly continuous) with Lipschitz constant at most $$\sup_{n\geq 0}\frac{|f(2^{n+1})-f(2^n)|}{|2^{n+1}-2^n|}\leq \sup_n\frac{2^{n+1}}{2^n}=2.$$ Needless to say, for this $f$ the limit in the left-hand side of (\[distinto\]) equals 1, while that on the right-hand side is 0 since $A^c$ belongs to $\mathscr V$. It’s nice, isn’t it?
\[fig:all\] {width="154mm"}
{#thepreceding}
The preceding observation is the key of the ensuing argument which allows us to give a neat description of the “fiber” $H_\frak t$ and so of $H(U)$. The fact that the exponential functions have exactly the growth-rate that is needed to separate ultrafilters is certainly a stroke of luck.
Let us denote by $\N_0$ the set of all nonnegative integers. Every point $t\in[1,\infty)$ can be written as $t=c\cdot 2^n$ for some $n\in\N_0$ and $c\in[1,2]$. Let $m:[1,2]\times\N_0\longrightarrow[1,\infty)$ be the map sending $(c,n)$ to $c\cdot 2^n$ and notice that $m(c,n)=m(d,k)$ if and only if $c=2, d=1$ and $k=n+1$ or vice-versa. Composing $m$ with the mapping $[1,\infty)\longrightarrow H(U)$ sending $t$ to $t^{-1}\delta_t$ we obtain a map $\mu: [1,2]\times\N_0\longrightarrow H(U)$ namely $\mu(c,n)=m(c,n)^{-1}\delta_{m(c,n)}$, so $$\mu(c,n)(f)=\frac{f(c\cdot 2^n)}{c\cdot 2^n}\quad\quad(f\in U).$$ Clearly, $\mu$ takes values in $
H_\frak t
$ which is a compact subset of $H(U)$ since it is homeomorphic to a closed subset of the product space $
\prod_{f\leq\frak t}[0,1]
$. The argument appearing in [**\[our\]**]{} shows that the range of $\mu$ is dense in $
H_\frak t
$. (Intermission: $
H_\frak t$ is a compactification of the half-line and also of the positive integers, with remainder $H^0_\frak t$ in both cases.)
Now we put Stone and Čech to work to obtain an extension $\tilde\mu:
[1,2]\times\beta \N_0\longrightarrow H_\frak t$ so that, for every $c\in[1,2]$ and every $\U\in\N_0^*$ one has $
\tilde{\mu}(c,\U)=\lim_{\U(n)}\mu(c,n)
$, that is, $$\tilde{\mu}(c,\U)(f)= \lim_{\U(n)} \frac{f(c\cdot 2^n)}{c\cdot 2^n}\quad\quad(f\in U).$$ Observe that the definition of $\tilde{\mu}$ guarantees continuity in the second variable, but not joint continuity that we now check “by hand”.
{#joint}
Recall that the topology of $\beta\N_0$ comes from $\ell_\infty(\N_0)$ in the sense that, given $\U\in\beta\N_0$, the sets of the form $$\{\mathscr V: |f(\mathscr V)-f(\U)|\leq\e\}\quad\quad(f\in\ell_\infty,\e>0),$$ where $g(\mathscr W)$ is interpreted as the limit of $g(n)$ along $\mathscr W$, form a subbase of the topology at $\U$.
Thus, in order to stablish the continuity of $\tilde{\mu}$ at $(c,\U)$, it suffices to prove that, given $f\in U$ and $\e>0$, there exist $g\in\ell_\infty(\N_0)$ and $\delta>0$ so that $|c-d|<\delta$ and $|g(\mathscr V)-g(\U)|<\delta$ imply $$\left|\lim_{\U(n)} \frac{f(c\cdot 2^n)}{c\cdot 2^n}-
\lim_{\mathscr V(n)} \frac{f(d\cdot 2^n)}{d \cdot 2^n}
\right|<\e.$$ Let us assume that $\U$ is free. The other case is easier. As the reader may guess we take $g(n)=f(c2^n)/(c2^n)$. Also, let $L$ be such that $|f(s)-f(t)|\leq L|s-t|$ provided $|s-t|\geq 1$ and suppose $|d-c|<\e$ and $|g(\mathscr V)-g(\U)|<\e$. Then $$\left|\lim_{\U(n)} \frac{f(c2^n)}{c 2^n}-
\lim_{\mathscr V(n)} \frac{f(d 2^n)}{d 2^n}
\right|\leq
\underbrace{\left|\lim_{\U(n)} \frac{f(c2^n)}{c 2^n}-
\lim_{\mathscr V(n)} \frac{f(c 2^n)}{d 2^n}
\right|}_{(\dagger)}
+
\underbrace{\left|\lim_{\mathscr V(n)} \frac{f(c2^n)}{d 2^n}-
\lim_{\mathscr V(n)} \frac{f(d 2^n)}{d 2^n}
\right|}_{(\ddagger)}\!.$$ Now, $$(\dagger)= \left|g(\U)-\frac{c}{d}g(\mathscr V)\right|\leq
\left|g(\U)-g(\mathscr V)\right|+ \left|g(\mathscr V)-\frac{c}{d}g(\mathscr V)\right|
\leq \e+ \e g(\mathscr V)\leq \e(1+L),$$
$$(\ddagger)\leq \lim_{\mathscr V(n)} \left| \frac{f(c2^n)-f(d2^n)}{d 2^n} \right|
\leq \lim_{\mathscr V(n)} \frac{L|c-d|2^n}{d2^n}\leq L\e,$$ and so $\tilde{\mu}$ is a continuous mapping *onto* $H_\frak t$.
{#identify}
We have just seen that $\tilde\mu:
[1,2]\times\beta \N_0\longrightarrow H_\frak t$ is a continuous surjection. The next task is to find out when two points of $[1,2]\times\beta \N_0$ have the same image in $H_\frak t$. Most of the work has been already done in [**\[declare\]**]{}, though in an implicit way.
Pick $(c,\U)$ and $(d,\mathscr V)$ in $[1,2]\times\beta \N_0$.
We claim that $\tilde{\mu}(c,\U)=\tilde{\mu}(d,\mathscr V)$ if and only if $c=2, d=1$ and $\mathscr V=1+\U$ or vice-versa. The meaning of this equality was explained in [**\[declare\]**]{}.
Let us first show that $\tilde{\mu}(2,\U)=\tilde{\mu}(1,1+\mathscr U)$. Take $f\in U$. Then $$\tilde{\mu}(2,\U)(f)=
\lim_{\U(n)} \frac{f(2\cdot 2^n)}{2\cdot 2^n}=
\lim_{\U(n)} \frac{f(2^{1+n})}{2^{1+n}}=
\lim_{(1+\U)(n)} \frac{f(1\cdot 2^{n})}{1\cdot 2^{n}}=
\tilde{\mu}(1,1+\mathscr U)(f),$$ as required. To check the converse we may assume $1\leq c\leq d<2$ since otherwise we could replace $(2,\U)$ by $(1,1+\U)$ and/or $(2,\V)$ by $(1,1+\V)$. Let us consider the case where $\U\neq \V$. Then there is $A\in \U$ which does not belong to $\V$ and so $A^c\in \V$. Here $A^c=\mathbb N_0\backslash A$ is the complement of $A$. Write $$\mathbb H=\bigsqcup_{n\geq 0}[2^n, 2^{n+1}).$$ Each interval $[2^n, 2^{n+1})$ contains exactly one point of the form $c2^n$ with $n\in\N_0$ and another one of the form $d2^n$. We define an increasing sequence $(p_n)_{n\geq 0}$ taking $$p_n=\begin{cases}
c2^n &\text{if $n\in A$}\\
d2^n &\text{if $n\notin A$}
\end{cases}$$ Note that $$|p_{n+1}-p_n|\geq \min(c,d)2^{n+1}- \max(c,d)2^{n} =
c2^{n+1}- d2^{n} = (2c-d)2^n.$$ Since $2c-d>0$ there exists a Lipschitz (hence uniformly continuous) $f:\mathbb H\longrightarrow\R$ such that $$f(t)=\begin{cases}
c2^n & \text{if $t=c2^n$ and $n\in A$}\\
0 & \text{if $t=d2^n$ and $n\in A^c$}
\end{cases}$$ namely the function whose graph is the polygonal joining $(p_0,q_0), (p_1,q_1); (p_2,q_2)...$ with $q_n=p_n$ if $n\in A$ and $q_n=0$ otherwise. For this $f$ one clearly has $\tilde{\mu}(c,\U)(f)=1$, while $\tilde{\mu}(d,\mathscr V)(f)=0$, so $\tilde{\mu}(c,\U)\neq\tilde{\mu}(d,\mathscr V)$.
Finally, if $\U=\V$, but $c\neq d$, then one easily finds a Lipschitz $f$ such that $$f(t)=\begin{cases}
c2^n & \text{if $t=c2^n$}\\
0 & \text{if $t=d2^n$}
\end{cases}$$ for every $n$ from where it follows that $\tilde{\mu}(c,\U)\neq\tilde{\mu}(d,\mathscr U)$. Since the case where $\U$ or $\V$ are fixed is trivial this leads to the following description of $H_{\frak t}$. Note that every continuous surjection between Hausdorff compacta is automatically a quotient map (cf. Willard [@w Chapter 3, §9]).
$\bigstar$ *The fiber $H_{\frak t}$ is homeomorphic to the quotient obtained from $[1,2]\times\beta\N_0$ after identifying each point of the form $(2,\U)$ with $(1,1+\U)$. The map sending the class of $(c,\U)$ to the homomorphism defined by the formula $$\phi(f)=\lim_{\U(n)} \frac{f(c\cdot 2^n)}{c\cdot 2^n}\quad\quad(f\in U)$$ is a homeomorphism.*
To complete our picture of $H(U)$, note that $H(U)\backslash\{0\}$ is homeomorphic to $H_\frak t\times(0,\infty)$: the map $(\phi,\lambda)\longmapsto \lambda\phi$ is continuous, with continuous inverse given by $\varphi\longmapsto(\varphi(\frak t)^{-1}\varphi, \varphi(\frak t))$.
On the other hand since for each $f\in U$ there exist $c>0$ such that $c|f|\leq\frak t$ we see that the sets $
\{\phi\in H(U): \phi(\frak t)<\e\}
$ form a base of neighbourhoods of $0$. Hence:
$\bigstar$ *The space $H(U)$ is homeomorphic to the quotient of $[1,2]\times\beta\N_0\times(0,\infty)$ with one point ${\bf 0}$ added, where we identify points of the form $(2,\U,\lambda)$ and $(1,1+\U,\lambda)$ and the neighbourhoods of the point ${\bf 0}$ are those sets containing a subset of the form $\{(c,\U,\lambda): \lambda<\e \}$ for some $\e>0$ together with the point ${\bf 0}$.*
{#other}
It is clear that everything what has been said about $U$ applies verbatim to $\Lip(\mathbb H)$, the lattice of Lipschitz functions on the half-line. Hence the spaces of homomorphisms of $\Lip(\mathbb H)$ and $U$ agree, in the sense that each homomorphism $\Lip(\mathbb H)\longrightarrow\R$ is the restriction of a unique $\phi\in H(U)$. We refer the reader to Chapter 5 of Weaver booklet [@weaver] for basic information about Lipschitz lattices.
Also, since the line $\R$ can be obtained by “gluing” two half-lines, $H(U(\R))$ can be easily computed using two copies of $H(U)$ and the same applies to $\Lip(\R)$. We will not give the details.
Finally, let us describe the homomorphisms $\Lip(\N)\longrightarrow\R$, where $\N$ carries the metric inherited from $\R$. Let $E:\Lip(\N)\longrightarrow U$ be the linear map sending $f$ into the piecewise linear function on $\H$ that interpolates $f$ on $\N$. If we consider $\ell_\infty$ as the set of bounded functions in $\Lip\N$, then $E$ maps $\ell_\infty$ to $U^*$ and we cannot help to display the following commutative diagram of linear maps $$\xymatrix{
0\ar[r] & U^* \ar[r] & U \ar[r] & U/U^* \ar[r] & 0\\
0\ar[r] & \ell_\infty \ar[r] \ar[u]^E & \Lip(\N) \ar[r] \ar[u]^E & \Lip(\N)/\ell_\infty \ar[r] \ar@{=}[u] & 0\\
}$$ Here, the rows are exact and the equal sign on the right reflects the fact that every $f\in U$ agrees with one of the form $E(g)$ for some $g\in \Lip\N$ modulo a bounded function: actually one can take $g=E(f|_\N)$. Now, let $L=H(\Lip \N)$ and
- $L_{\frak t}=\{\phi\in L: \phi({\frak t})=1\}$,
- $L_{\frak t}^0=\{\phi\in L: \phi({\frak t})=1 \text{ and } \phi({\bf 1})=0\}$.
If $\phi\in L_{\frak t}$, then either $\phi({\bf 1})>0$, in which case $\phi=n^{-1}\delta_n$ for some integer $n$, or $\phi$ vanishes on every bounded function and so it factors throught the quotient $\Lip(\N)/\ell_\infty=U/U^*$. If so, there is $\U\in \N^*$ and $c\in[1,2]$ such that $$\phi(f)=\lim_{\U(n)}\frac{Ef(c2^n)}{c2^n}= \lim_{\U(n)}\frac{Ef([c2^n])}{[c2^n]},$$ where $[\cdot]$ is the integer part function. Hence $L_{\frak t}^0$ is homeomorphic to $H_{\frak t}^0$.
Coda
====
This note lived a hard life until the referee it was looking for came along. In the meantime, we have explored in [@quiz] the consequences of the research reported here, encountered nice descriptions of the Samuel-Smirnov compactification of the line in [@flows Lemma 2.1] and [@turkiyos Theorem 2.1], and registered some curious connections between the homomorphisms on $\Lip(\N)$ and the so-called density measures (cf. [@density Theorem 3.1]).
[00]{}
Yuji Akaike, Naotsugu Chinen, Kazuo Tomoyasu, The Smirnov remainders of uniformly locally connected proper metric spaces. Top. Appl. 158 (2011), no. 1, 69–83.
İlker Akça, Mahmut Koçak, Non-homogeneity of the remainder $s\R\backslash\R$ of the Samuel compactification of $\R$, Top. Appl. 149 (2005) 239–242.
Yoav Benyamini, Joram Lindenstrauss, Geometric nonlinear functional analysis. Amer. Math. Soc. Colloquium Publications, 48. Providence, RI, 2000.
Félix Cabello Sánchez, Javier Cabello Sánchez, Lattices of uniformly continuous functions, Top. Appl. 160 (2013) 50–55.
Félix Cabello Sánchez, Javier Cabello Sánchez, Quiz your maths: Do the uniformly continuous functions on the line form a ring? To appear in Proc. Amer. Math. Soc.
M. Isabel Garrido, Jesús Á. Jaramillo, A Banach-Stone theorem for uniformly continuous functions, Monatsh. Math. 131 (2000) 189–192.
M. Isabel Garrido, Ana S. Meroño, The Samuel realcompactification of a metric space. J. Math. Anal. Appl. 456 (2017), no. 2, 1013–1039.
Miroslav Hušek, Lattices of uniformly continuous functions determine sublattices of bounded functions, Top. Appl. 182 (2015) 71–76.
Miroslav Hušek, Antonio Pulgarín, Banach-Stone-like theorems for lattices of uniformly continuous functions, Quaest. Math. 35 (2012) 417–430.
Ryoichi Kunisada, Density measures and additive property, J. Number Theory 176 (2017) 184–203
Pierre Samuel, Ultrafilters and compactifications of uniform spaces. Trans. Amer. Math. Soc. 64 (1948) 100–132.
Taira Shirota, A generalization of a theorem of I. Kaplansky, Osaka Math. J. 4 (1952) 121–132.
Jun-Ichi Tanaka, Flows in fibers, Trans. Amer. Math. Soc. 343 (1994) 77–804.
Nik Weaver, Lipschitz Algebras, World Scientific, Singapore, 1999.
Stephen Willard, General Topology, Addison-Wesley, Reading, 1970.
R. Grant Woods, The minimum uniform compactification of a metric space, Fund. Math. 147 (1995) 39–59.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The possibility to observe and manipulate Majorana fermions as end states of one-dimensional topological superconductors has been actively discussed recently. In a quantum wire with strong spin-orbit coupling placed in proximity to a bulk superconductor, a topological superconductor has been expected to be realized when the band energy is split by the application of a magnetic field. When a periodic lattice modulation is applied multiple topological superconductor phases appear in the phase diagram. Some of them occur for higher filling factors compared to the case without the modulation. We study the effects of phase jumps and argue that the topologically nontrivial state of the whole system is retained even if they are present. We also study the effect of the spatial modulation in the hopping parameter.'
author:
- Masaki
- Norio
title: Reentrant topological transitions with Majorana end states in 1D superconductors by lattice modulation
---
Introduction
============
Much work has been devoted to realize topologically nontrivial states of matter with topologically protected surface states. Majorana surface states are expected to be formed at the boundaries and vortex cores of topological superconductors (TSs).[@Read2000; @Kitaev2001; @Ivanov2001; @Fu2008; @Fujimoto2008; @Sato2009; @Linder2010] Such states have been expected in a one-dimensional (1D) quantum wire with spin-orbit interaction (SOI) placed under a Zeeman field and in proximity to a bulk superconductor,[@Sau2010; @Alicea2010; @Lutchyn2010; @Oreg2010] in cold atoms in optical lattices with effective gauge fields generated by spatially varying laser fields,[@Sato2009] bilayer electron gases in semiconductor heterostructure with interlayer Coulomb coupling,[@Nakosai2012] among others. Several groups of experimentalists have recently reported that they have observed the signatures of Majorana fermions appearing at the ends of nanowires attached to superconductors.[@Mourik2012; @Rodrigo2012; @Deng2012; @Rokhinson2012] For a review we refer to Ref. .
The effect of spatial inhomogeneity on a superconducting quantum wire has been a nontrivial problem.[@Motrunich2001; @Gruzberg2005; @Brouwer2011; @Lutchyn2012; @Lobos2012; @Bagrets2012; @Takei2012; @Adagideli2013; @Sau2013] The energy distributions of the end states have been obtained for the Dirac equation with random mass and a 1D spinless superconductor.[@Brouwer2011] The interplay of disorder and correlation in 1D TSs has also been investigated.[@Lobos2012] Among the possible realizations of spatial inhomogeneity, quasiperiodic (Harper) potential modulation forms a special class in that in 1D system all the single particle eigenstates become localized at the same modulation strength. Quasiperiodic potentials have been experimentally studied in optically trapped cold atom systems [@QuasiperiodicColdAtoms] as well as in solid state, misfit compound [@1301.6888] systems. Signatures of a Hofstadter butterfly-like band structure have been observed in a van der Waals system of monolayer graphene on top of a hexagonal boron nitrade surface.[@Hunt1303.6942]
The 1D effectively spinless superconductor, expected in 1D superconductor with strong spin-momentum coupling under magnetic field, is of the D symmetry class.[@PeriodicTable] Therefore its topology is classified by a $Z_2$ topological number. This suggests that the boundary between two topologically non-trivial 1D superconductors would not have a localized mode. Boundary phenomena between topologically equivalent or distinct phases, with Harper or Fibonacci potentials, have been experimentally studied using photonic quasicrystals.[@VerbinPRL2013]
In the scenario of Refs. the chemical potential needs to lie close to the band edge so that the band degeneracy is removed by the external magnetic field. However, the present authors have observed that, by a quasiperiodic lattice modulation with a fixed wavenumber, effectively single-band superconductor with end Majorana fermions is realized even when the chemical potential is closer to the center of the original cosine band, because energy separations are introduced within each of the Zeeman-split bands.[@Tezuka2012] We have also demonstrated that this physics is stable even in the presence of a Hubbard-like on-site interaction and/or a harmonic trapping potential. More recently, the effect of incommensurate potentials on 1D p-wave superconductors have been studied.[@DeGottardi1208.0015; @Cai1208.2532; @Satija1210.5159] *Commensurate* diagonal or off-diagonal Harper model has also attracted theoretical attention.[@Lang1207.6192; @Ganeshan1301.5639]
Here we are interested in characterizing the new TS regions further, focusing on when they emerge, and what happens when the quasiperiodically modulated quantum wire is connected with other wires with different modulation phases or an unmodulated one. Particularly, we would like to (i) clarify the correspondence between the energy spectrum of the single particle states and emergence of the TS states, in the presence of either a quasiperiodic modulation with a general wavenumber or the external magnetic field to a general direction, (ii) understand the effect of phase jumps of the quasiperiodic modulation and that of a quasiperiodic modulation applied to only a limited part of the one-dimensional system, and (iii) study the effect of a quasiperiodic modulation of the *hopping* parameter.
In section II we define our model and introduce our Bogoliubov-de Gennes (BdG) equation approach to study the system. We find that with a lattice-site energy modulation with a generic wavenumber, quasiperiodic or periodic, new topological superconducting regions with end Majorana fermions emerge. We also study the effect of the direction of the external magnetic field. Then in section III we observe that those regions are stable against phase jumps in one-dimensional systems, while when the modulated wire is connected with an unmodulated wire, the locations of the localized modes are determined by which of the wires becomes TS. We also study the case with quasiperiodic modulation in the intersite hopping parameter, and find that multiple topological transitions into and out of TS states occur. Finally, in section IV, we summarize our findings.
Site level modulation by a single (quasi)periodic lattice potential
===================================================================
We consider a one-dimensional quantum wire parallel to the $\hat{{\mbox{\boldmath $x$}}}$ direction, coupled to a bulk superfluid whose surface is perpendicular to $\hat{{\mbox{\boldmath $z$}}}$. We study a tight-binding one-dimensional model of spin-$1/2$ fermions with the Rashba-type spin-orbit coupling, the mean-field coupling to the bulk superconductor, and the Zeeman energy due to the external magnetic field ${\mbox{\boldmath $B$}}$.
The Hamiltonian we have adopted is $$\begin{aligned}
\mathcal{H}
&=& -\frac{t}{2}\sum_{l=0}^{L-2}\sum_{\sigma=\uparrow,\downarrow}
(\hat c_{\sigma,l}^\dag \hat c_{\sigma,l+1} + \mathrm{h.c.})\nonumber\\
&+& \frac{\alpha}{2} \sum_{l=0}^{L-2}
\left(
(\hat c_{\downarrow,l}^\dag \hat c_{\uparrow,l+1}
-\hat c_{\uparrow,l}^\dag \hat c_{\downarrow,l+1}) + \mathrm{h.c.}\right)\nonumber\\
&+& \sum_{l=0}^{L-1} \left(\Delta(\hat c_{\uparrow,l} \hat c_{\downarrow,l} + \mathrm{h.c.})
+\frac{2\Gamma}{\hbar}\hat{{\mbox{\boldmath $B$}}}\cdot{\mbox{\boldmath $S$}}_l\right)\nonumber\\
&+& \sum_{l=0}^{L-1} \sum_{\sigma=\uparrow,\downarrow}(t-\mu+\epsilon_{\sigma,l}) \hat n_{\sigma,l}
,
\label{eqn:Hamiltonian_gen}\end{aligned}$$ with $\Gamma \equiv \frac{g\mu_B}{2}B$. We set $t=1$ as the unit of energy and set $\hbar = 1$ in the following. In the case of $\hat{{\mbox{\boldmath $B$}}} = \hat{{\mbox{\boldmath $z$}}}$ the third line becomes $$\begin{aligned}
\sum_{l=0}^{L-1} \left(\Delta(\hat c_{\uparrow,l} \hat c_{\downarrow,l} + \mathrm{h.c.})
+\Gamma(\hat n_{\uparrow,l}-\hat n_{\downarrow,l})\right).\nonumber\end{aligned}$$ Here, $\hat c_{\sigma,l}$ annihilates a fermion with spin $\sigma(=\uparrow, \downarrow)$ at site $l (=0, 1, \ldots, L-1)$, $\hat n_{\sigma,l} \equiv \hat c^\dag_{\sigma,l} c_{\sigma,l}$, $t$ determines the nearest-neighbor hopping, $\alpha$ is the Rashba-type SOI, $\Delta$ is the coupling to the bulk superconductor, $\Gamma$ is the Zeeman energy, $\mu$ is the chemical potential, and $\epsilon_{\sigma,l}$ is the site energy for spin $\sigma$ on site $l$.
In this work we limit our discussion to the case with $\epsilon_{\sigma,l} = \epsilon_l$. In the following we introduce a quasiperiodic modulation to the site energy, and study the energy distribution of the single particle state and its correspondence with the realization of Majorana end modes.
![(Color online) The distribution of single particle state energy of eq. (\[eqn:Hamiltonian\_gen\]) for $L = 200$, $(\Gamma, \alpha)=(0.3, 0.3)$, $V_\mathrm{Q}=0, 0.2, 0.6, 1.25$, $\hat{{\mbox{\boldmath $B$}}} = \hat{{\mbox{\boldmath $z$}}}$. The single particle Hamiltonian, obtained by setting $\Delta = 0$ and $\mu=0$ in eq. (\[eqn:Hamiltonian\_gen\]), is diagonalized for $\kappa = 2\pi j/400$ ($j=0,1,.\ldots,400$) and each eigenstate is plotted in a color corresponding to the expectation value of the $z$ component of the spin for that eigenstate. The plot for $V_\mathrm{Q} = 1.25$ is similar to that for $V_\mathrm{Q} = 0.8$ (not shown) except that the former is scaled horizontally, because of the duality of the lattice between $V_\mathrm{Q} = x$ and $1/x$. $V_\mathrm{Q} = 1$ is the self-similar point, at which the modulation potential equals the unmodulated band width for $\Gamma = \alpha = 0$. []{data-label="fig:DoubleHofstadter"}](Full_Fig1_200){width="8.66cm"}
Double Hofstadter butterfly
---------------------------
Let us first consider the case of $\hat{{\mbox{\boldmath $B$}}} = \hat{{\mbox{\boldmath $z$}}}$. For an infinitely long system having $\Delta=\epsilon_{l}=0$ we easily obtain the single particle energy as a function of the quasimomentum $k$, $$E^{\pm}(k) = t(1-\cos(k)) \pm \sqrt{\alpha^2\sin^2(k) + \Gamma^2}.
\label{eqn:RZbands}$$ We call them the upper and lower Rashba–Zeeman (RZ) bands. [@Tezuka2012] The mapping of the Hamiltonian to that of a spinless system is possible if $\mu$ lies in only one of the RZ bands. [@Lutchyn2010; @Oreg2010; @Stoudenmire2011] In such a case, by introducing the pairing $\Delta$ such that $|\Delta| \lesssim \Gamma$, we obtain the topological superconductor phase.
We consider a site potential which is given by $$\epsilon_{l} = V_\mathrm{Q} \cos( \kappa (l-l_\textrm{c}) + \phi_0),
\label{eqn:sitePotential}$$ in which $V_\mathrm{Q}\geq 0$, $l_\textrm{c} \equiv (L-1)/2$, $\phi_0$ is the phase of the potential at the center of the system and $\kappa = 2 \pi g$, in which $g$ is a real number such that $0<g<1$. The potential is periodic for a rational $g$, while it is quasiperiodic for an irrational $g$. In the following we choose $\phi_0 = 0$, except when we study the effect of $\phi_0$ and when we study the effect of phase jumps in the system.
In a finite length system with $L$ lattice sites, we numerically obtain the set of single particle level energies. For $\alpha = \Gamma = 0$, each single body wavefunction is extended for $V_\mathrm{Q}<t$ and localized for $V_\mathrm{Q}>t$ in the $L\rightarrow\infty$ limit. [@Kohmoto1983] The self-similar structure of the two-dimensional spectrum plotted against $\kappa$ for $V_\mathrm{Q} = t$ is called the Hofstadter butterfly. [@Hofstadter1976; @Kohmoto1983]
When the $2L$ energy levels obtained are plotted against various values of $\kappa$, for $V_\mathrm{Q} \sim t$, the spectrum shows a self-similar structure resembling two Hofstadter butterflies shifted in energy and braided together, as shown in Fig. \[fig:DoubleHofstadter\]. We call this structure the double Hofstadter butterfly. [@Tezuka2012]
The spin-orbit coupling $\alpha$ mixes the spin-up states and spin-down states differently at each value of the quasimomentum $k$ of the resulting RZ bands. Therefore the spin-independent site potential, which has components with $|k|=\pm \kappa$ and $|k| = \pm(2\pi-\kappa)$, further mixes the upper and lower RZ bands. Most of the states in the resulting double Hofstadter butterfly do not have a completely polarized spin. We may, however, obtain the expectation value of the $z$ component of the spin, $\langle S_z\rangle$, for each of the $2L$ eigenstate. The energy levels plotted in Fig. \[fig:DoubleHofstadter\] have been color-coded according to the value of $\langle S_z\rangle$.
We find that, for a fixed value of $\kappa$, the sets of states from two Hofstadter-butterfly-like structures with separated values of spin polarizations overlap within some energy ranges. In some regions in energy there are no single particle states, even inside the range of $|\epsilon - t|\leq t + \Gamma$, which was occupied by states of eq. (\[eqn:RZbands\]) before the introduction of the site potential modulation eq. (\[eqn:sitePotential\]). Other regions are occupied by states in only one of the Hofstadter-butterfly-like structure. We study the consequences of the site potential modulation on the realization of TS states for the many-body states with a finite chemical potential in the following.
Bogoliubov-de Gennes equation: zero modes and Majorana fermions
---------------------------------------------------------------
The Hamiltonian (\[eqn:Hamiltonian\_gen\]) is bilinear in operators $\hat{c}$ and $\hat{c}^\dag$. It can be exactly diagonalized in the Nambu spinor space $\{ ({\mbox{\boldmath $u$}}_\uparrow, {\mbox{\boldmath $u$}}_\downarrow, {\mbox{\boldmath $v$}}_\uparrow, {\mbox{\boldmath $v$}}_\downarrow)^\mathrm{T} \}$, with the basis obtained as the set of the eigenvectors of the Bogoliubov-de Gennes (BdG) equation.
Alternatively, fermion pairing via a short-range attractive interaction can also be simulated by introducing a pairing constant $g$ and solving the Bogoliubov-de Gennes equation self-consistently. In a lattice system the energy cutoff which renormalizes $g$ can be in principle determined from the lattice constant. However, here we fix the value of $\Delta$, a homogeneous proximity pairing, by hand. This is because we do not expect that the BdG approximation, which simulates the pair formation *within* the 1D wire, directly corresponds to our Hamiltonian (\[eqn:Hamiltonian\_gen\]).
We solve [@Iskin2012; @Liu2012] $$\left(\begin{array}{cccc}
\hat{H_{\uparrow\uparrow}} & \hat{H_{\uparrow\downarrow}} & 0 & \Delta\\
\hat{H_{\downarrow\uparrow}} & \hat{H_{\downarrow\downarrow}} & -\Delta & 0\\
0 & -\Delta & -\hat{H_{\uparrow\uparrow}} & -\hat{H_{\uparrow\downarrow}}\\
\Delta & 0 & -\hat{H_{\downarrow\uparrow}} & -\hat{H_{\downarrow\downarrow}}\\
\end{array}\right)
\Psi = \epsilon \Psi,
\label{eqn:BdG}$$ in which $\Psi = ({\mbox{\boldmath $u$}}_\uparrow, {\mbox{\boldmath $u$}}_\downarrow, {\mbox{\boldmath $v$}}_\uparrow, {\mbox{\boldmath $v$}}_\downarrow)^\mathrm{T}$ is a $4L$-dimensional vector and $\hat{H_{\sigma\sigma'}}$ are the single-particle components of the Hamiltonian, $$\begin{aligned}
\hat{H_{\sigma\sigma}} = -\frac{t}{2}\sum_{l=0}^{L-2}\sum_{\sigma=\uparrow,\downarrow}
(\hat c_{\sigma,l}^\dag \hat c_{\sigma,l+1} + \mathrm{h.c.})\nonumber\\
+\sum_{l=0}^{L-1} \sum_{\sigma=\uparrow,\downarrow}(t-\mu+\epsilon_{l}
+(-)^\sigma (\hat{{\mbox{\boldmath $B$}}}\cdot\hat{{\mbox{\boldmath $z$}}})\Gamma
) \hat n_{\sigma,l},\end{aligned}$$ $$\begin{aligned}
\hat{H_{\downarrow\uparrow}} &=& \hat{H_{\uparrow\downarrow}}^\dag =
\frac{\alpha}{2} \sum_{l=0}^{L-2}
(\hat c_{\downarrow,l}^\dag \hat c_{\uparrow,l+1}
-\hat c_{\downarrow,l+1}^\dag \hat c_{\uparrow,l})\\
&-&((\hat{{\mbox{\boldmath $B$}}}\cdot\hat{{\mbox{\boldmath $x$}}}) + i(\hat{{\mbox{\boldmath $B$}}}\cdot\hat{{\mbox{\boldmath $y$}}}))\Gamma
\sum_{l=0}^{L-1}\hat c_{\downarrow,l}^\dag \hat c_{\uparrow,l},\end{aligned}$$ in which $(-)^\sigma = \delta_{\sigma\downarrow}-\delta_{\sigma\uparrow}$. In the following we call the $4L$-dimensional matrix in the left hand side of eq. (\[eqn:BdG\]) $M_\mathrm{BdG}$.
We work in the limit of low temperature $T\rightarrow 0$. We obtain the particle distribution by $$\begin{aligned}
n_{\sigma,l} &=& \sum_q\left[f(\epsilon_q) |u_{\sigma,l}|^2 + f(-\epsilon_q) |v_{\sigma,l}|^2\right],
\label{eqn:density}\end{aligned}$$ in which $f(\epsilon)\equiv (1+\mathrm{e}^{\epsilon/(k_\mathrm{B}T)})
\rightarrow \Theta(-\epsilon)$ ($T\rightarrow 0$) is the Fermi distribution function, with $\Theta(x)$ being the step function. The total number of fermions with spin $\sigma$ is obtained as $N_\sigma = \sum_l n_{\sigma,l}$.
The sum in eq. (\[eqn:density\]) is taken over all $q$. The matrix $M_\mathrm{BdG}$ is Hermitian, and has pairs of positive and negative eigenvalues with equal absolute values, because if $M_\mathrm{BdG}({\mbox{\boldmath $u$}}_\uparrow, {\mbox{\boldmath $u$}}_\downarrow, {\mbox{\boldmath $v$}}_\uparrow, {\mbox{\boldmath $v$}}_\downarrow)^\mathrm{T}
= \epsilon({\mbox{\boldmath $u$}}_\uparrow, {\mbox{\boldmath $u$}}_\downarrow, {\mbox{\boldmath $v$}}_\uparrow, {\mbox{\boldmath $v$}}_\downarrow)^\mathrm{T}$, we have $M_\mathrm{BdG}({\mbox{\boldmath $v$}}_\uparrow, {\mbox{\boldmath $v$}}_\downarrow, {\mbox{\boldmath $u$}}_\uparrow, {\mbox{\boldmath $u$}}_\downarrow)^\mathrm{T}
= -\epsilon({\mbox{\boldmath $v$}}_\uparrow, {\mbox{\boldmath $v$}}_\downarrow, {\mbox{\boldmath $u$}}_\uparrow, {\mbox{\boldmath $u$}}_\downarrow)^\mathrm{T}.$ For a positive (negative) eigenvalue, only $|v_{\sigma,l}|^2$ ($|u_{\sigma,l}|^2$) contributes to the particle distribution in the $T\rightarrow0$ limit.
![(a–b) Largest negative ($-$) and smallest positive ($+$) eigenvalues of the BdG equation (\[eqn:BdG\]) plotted against the chemical potential $\mu$ for $g = \sqrt{5}-2$, $(\Gamma, \alpha, \Delta)=(0.2, 0.3, 0.1)$ and $V_\mathrm{Q} = $(a) $0$, (b) $0.5$. (c–e) Amplitudes of the particle- and hole-like parts of the eigenvectors $\sum_\sigma|u_\sigma|^2$, $\sum_\sigma|v_\sigma|^2$ plotted against the lattice site for $(\Gamma, \alpha, \Delta)=(0.2, 0.3, 0.1)$ and $\mu = $(c) $-0.04$, (d) $-0.02$, (e) $-0.01$.[]{data-label="fig:localization"}](Full_Fig2_200){width="8.66cm"}
### Distribution of eigenvalues
For $L$ sites in the system, because of the spin and particle–hole degrees of freedom, we have $4L$ eigenstates of the BdG equation (\[eqn:BdG\]). The introduction of $\Delta$ opens a gap in the eigenvalue spectrum of eq. (\[eqn:BdG\]) in the absence of the spin-orbit coupling $\alpha$ or the Zeeman field $\Gamma$. With the spin-orbit coupling and the Zeeman field, when the chemical potential $\mu$ satisfies $$\mathrm{(lower\ band\ bottom)} < \mu < \mathrm{(upper\ band\ bottom)}$$ so that the Kitaev model [@Kitaev2001] is effectively realized in the spinful case, the system is a topological superconductor for $0 < \Delta \lesssim \Gamma$ when ${\mbox{\boldmath $B$}}$ is in the $z$–$x$ plane.[@Sau2010; @Alicea2010; @Lutchyn2010; @Oreg2010; @Stanescu1302.5433] In this case the BdG equation has two eigenstates with $\epsilon\sim 0$.
Let us consider the $(2L)$-th and $(2L+1)$-th smallest eigenvalues, $\epsilon^-$ and $\epsilon^+$, which satisfy $-\epsilon^- = \epsilon^+$ because eigenvalues appear in pairs with same absolute value and opposite signs as mentioned above. We can only have an even number of vanishing eigenvalues, and if we have them $\epsilon^-$ and $\epsilon^+$ should be included in them; otherwise $\epsilon^- < 0 < \epsilon^+$.
For $\mu\ll\{-t,\min(\epsilon_{l})\}$, the number of fermions in the system is negligible. As $\mu$ is increased, $N_\sigma$ increases, with $N_\downarrow > N_\uparrow$ for $\Gamma > 0$. $\epsilon^+$ initially decreases linearly in $\mu$, reflecting the linear decrease of the required energy to add a single particle in the system. When it is closer to zero, however, the value of $\epsilon_\pm$ approaches more slowly to zero, especially for a smaller system. When we fit the decrease by a function of the shape $\exp{(-c\vert\mu-\mu_0\vert)}$, the exponent $c$ is roughly in proportion to the system size $L$. This also suggests that the modes corresponding to $\epsilon_\pm$ are spatially localized.
### Detection of end Majorana fermions
From the eigenvector of $M_\mathrm{BdG}$ corresponding to the eigenvalue $\epsilon^+$, $({\mbox{\boldmath $u$}}_\uparrow, {\mbox{\boldmath $u$}}_\downarrow, {\mbox{\boldmath $v$}}_\uparrow, {\mbox{\boldmath $v$}}_\downarrow)^\mathrm{T}$, we define the averaged separation from the system center $$\sqrt{\langle x^2 \rangle} \equiv
\sqrt{
\left(\sum_{\sigma,l} (l-l_c)^2 \vert{\mbox{\boldmath $v$}}_{\sigma,l}\vert^2\right)
/
\left(\sum_{\sigma,l} \vert{\mbox{\boldmath $v$}}_{\sigma,l}\vert^2\right)
}.$$ If the modes localize to the system ends, $\sqrt{\langle x^2 \rangle}$ are close to $l_c$, the maximum value it can take. Note that, when $\epsilon^-$ and $\epsilon^+$ are numerically degenerate ($\epsilon^+ \lesssim 10^{-12}$ in our work), any linear combination of the two eigenvectors that correspond to these eigenvalues would be obtained as the eigenvector corresponding to $\epsilon^+$, so within our BdG calculation we do not directly observe eigenmodes that are localized at only one of the ends of the system.
In the density-matrix renormalization group (DMRG) simulations of the same model, [@Stoudenmire2011; @Tezuka2012] however, a pair of Majorana modes localized at each end of the system have been obtained. We believe that the DMRG calculation, with limitation on the entanglement entropy between the two ends of the system, automatically chooses less entangled degenerate ground states, which are connected to each other by operating either of the two Majorana operators, for the subspaces with even and odd numbers of fermions. We find that the localization of ${\mbox{\boldmath $u$}}_{\sigma,l}$ and ${\mbox{\boldmath $v$}}_{\sigma,l}$, in terms of $\sqrt{\langle x^2\rangle}$, precisely corresponds to the localization of the Majorana operators observed by DMRG [@Tezuka2012].
We have plotted the values of $\epsilon^\pm$ as a function of $\mu$ along with the single particle state energy in Fig. \[fig:localization\] (a–b). In Fig. \[fig:localization\] (a) with $V_\mathrm{Q} = 0$, the overlapping RZ bands are clearly observed, and $\epsilon^\pm$ vanish only when the chemical potential crosses only one of the RZ bands.
In Fig. \[fig:localization\] (b) with $V_\mathrm{Q} = 0.5$ several regions with vanishing $\epsilon^\pm$ are found, each corresponding to the energy region with states from just one of the two Hofstadter-butterfly-like structure. The components of the eigenvector corresponding to the eigenvalue $\epsilon^+$ are shown in Fig. \[fig:localization\] (c–e) for values of $\mu$ approaching to one of the regions with vanishing $\epsilon^\pm$. The localization of the mode is clearly observed with an increase of $\sqrt{\langle x^2 \rangle}$. In all regions with vanishing $\epsilon^\pm$, we observe a clear localization of the corresponding eigenvectors of $M_\mathrm{BdG}$.
![(Color online) Top figure: Single particle eigenstates of eq. (\[eqn:Hamiltonian\_gen\]) color-coded according to the value of $S_z$ for $\hat{{\mbox{\boldmath $B$}}}=\hat{{\mbox{\boldmath $z$}}}$, $L=400$, $(\Gamma, \alpha, V_\mathrm{Q}) = (0.3, 0.3, 0.2)$ and $-0.6\leq\mu\leq1$, $0 < \kappa < \pi$. Lower figures: Grayscale plots of $\epsilon^+$ for the same parameters with $\Delta = 0.05, 0.1, 0.15, 0.2$.[]{data-label="fig:L400lambda0.2"}](Full_Fig3_200){width="8.66cm"}
![(Color online) Top figure: Single particle eigenstates of eq. (\[eqn:Hamiltonian\_gen\]) color-coded according to the value of $S_z$ for $\hat{{\mbox{\boldmath $B$}}}=\hat{{\mbox{\boldmath $z$}}}$, $L=400$, $(\Gamma, \alpha, V_\mathrm{Q}) = (0.3, 0.3, 0.5)$ and $-0.6\leq\mu\leq1$, $0 < \kappa < \pi$. Lower figures: Grayscale plots of $\epsilon^+$ for the same parameters with $\Delta = 0.05, 0.1, 0.15, 0.2$.[]{data-label="fig:L400lambda0.5"}](Full_Fig4_200){width="8.66cm"}
Dependence on the lattice modulation wavenumber
-----------------------------------------------
In Ref. [@Tezuka2012] we fixed the wave vector $\kappa$ of the quasiperiodic lattice modulation. However, as we have observed in Fig. \[fig:DoubleHofstadter\], we have effectively single-band regions of the chemical potential for a wide range of the value of $\kappa$. Therefore it is interesting to study the dependence of the appearance of topologically non-trivial superfluid with end Majorana fermions on the value of $\kappa$. Especially it is intriguing whether a $\kappa$ such that $2\pi/\kappa$ is an integer has any difference.
In Figs. \[fig:L400lambda0.2\] and \[fig:L400lambda0.5\] we plot the value of $\epsilon^+$ in a color code, along with the single particle state energies obtained from eq. (\[eqn:Hamiltonian\_gen\]). For smaller values of $\Delta$, we notice that vanishing eigenvalue corresponds to regions covered by a single band of single particle eigenstates, regardless of the value of $\kappa$. Particularly, even for values of $\kappa$ such as $\pi/3$ and $\pi/2$, $\epsilon^+$ vanishes when the chemical potential lies in the region where single particle eigenstates are effectively spinless.
In these regions, because of the effectively single band structure of the non-interacting Hamiltonian, the Kitaev model is realized, with Majorana fermion modes localizing at system ends. [@Tezuka2012] The value of $\epsilon^+$ becomes closer to $0$ as $L$ is increased, reflecting that the end modes are more separated. For smaller $L$, the separation between the single particle eigenstates is comparable to $\Delta$ when $\Delta\ll\{\Gamma, t\}$. In this case, strength of the induced superconductivity depends on the relative location of a level and the chemical potential. When the separation is larger (the density of states is lower) the value of $\epsilon^+$ usually stays larger, but changes rapidly. $\epsilon^+$ becomes more homogeneous and generally reduced as $\Delta$ is increased inside each sub-band which does not overlap in energy with another.
We note that, while $N_\downarrow$ is always larger than $N_\uparrow$ for $\Gamma > 0$, there are sub-bands having fermions almost polarized in the $+\hat{{\mbox{\boldmath $z$}}}$ direction ($\langle S_z \rangle \sim 1/2$) even when $\mu < 1$ with less than half filling factor ($N_\downarrow + N_\uparrow < L$). Remarkably, even when the chemical potential lies in one of such bands, a TS state with end Majorana fermions can be formed.
However, as $\Delta$ is increased, smaller features in the single particle state distribution, which remained in the structure of the value of $\epsilon^+$ for smaller $\Delta$, gradually disappear and only the widest single-band regions remain visible. Similar simplification of the phase diagram is also observed in Fig. 2 of Ref. for a quasiperiodic system of spinless fermions. Finally, for $\Delta \gtrsim \Gamma$, the in-gap state disappears and the eigenvalue spectrum of $M_\mathrm{BdG}$ has a gap of the order of $\Delta$, regardless of the value of $V_\mathrm{Q}$, so that the system is topologically trivial.
In summary, the new topologically non-trivial regions with end Majorana fermions appear for general site energy modulations of the type of eq. (\[eqn:sitePotential\]), and their existence is not limited to some special irrational values of $g=\kappa/(2\pi)$. The commensurate case can be considered as a kind of multi-band wire. The possibility of TS states with end Majorana modes has also been studied in multi-band systems.[@Potter2010; @Lutchyn2011a; @Rieder2012] We have observed that while the range of the chemical potential reflects the single particle eigenstate spectrum strongly, the value of $\Delta$ also plays an important role. If $\Delta$ is too small $\epsilon^\pm$ vanishes only for regions with higher density of states. If $\Delta$ is too large, smaller features in the single particle spectrum become smeared. This occurs before $\Delta$ exceeds $\Gamma$ so that the system becomes topologically trivial regardless of the values of $\mu$.
Dependence on the direction of the external magnetic field
----------------------------------------------------------
![(Color online) Grayscale plot of the value of $\epsilon^+$ for $\hat{{\mbox{\boldmath $B$}}}=\hat{{\mbox{\boldmath $x$}}}\cos\theta + \hat{{\mbox{\boldmath $y$}}}\sin\theta$, $L=200$, $(\Gamma, \alpha, \Delta, V_\mathrm{Q}) = (0.3, 0.3, 0.1, 0.5)$, $g = (\sqrt5-1)/2$, and $-0.6\leq\mu\leq1$, $0 \leq \theta \leq \pi/2$. []{data-label="fig:rotateTheta"}](Full_Fig5){width="8.66cm"}
The existence of the Majorana end fermions depends on the direction of the applied magnetic field. [@Sau2010; @Alicea2010; @Lutchyn2010; @Oreg2010] Namely, the effective magnetic field introduced by the Rashba spin-momentum coupling needs to have perpendicular component to the external magnetic field. In our model the Rashba spin-momentum coupling is in the $y$ direction, so the perpendicular directions lie in the $z$–$x$ plane, Here we ask: can the spin-insensitive quasiperiodic modulation change this situation?
To answer this question we now consider $\hat{{\mbox{\boldmath $B$}}} = \hat{{\mbox{\boldmath $x$}}}\cos\theta + \hat{{\mbox{\boldmath $y$}}}\sin\theta$, with $0\leq \theta \leq \pi/2$. We plot the value of $\epsilon^+$ in a color-coded plot in Fig. \[fig:rotateTheta\]. While the region with vanishing $\epsilon^+$ persists up to $\theta \lesssim \pi/4$, which corresponds to $\Gamma_x\equiv \Gamma\cos\theta
\gtrsim 0.2$, for larger $\theta$ with $\Gamma_x \lesssim 0.2$ we no longer have a vanishing eigenvalue of the BdG equation and the system is topologically nontrivial.
Therefore, the introduction of the site level modulation still does not lift the limitation in the direction of ${\mbox{\boldmath $B$}}$ for TS states to be observed. Because $\Delta = 0.1$ still does not exceed $\Gamma_x$, the result above indicates that the $y$ component of the applied magnetic field is rather detrimental for the realization of a TS in our model.
Dependence on the phase of the modulation potential
---------------------------------------------------
![(Color online) Top figure: Single particle eigenstates of eq. (\[eqn:Hamiltonian\_gen\]) color-coded according to the value of $S_z$ for $g=\sqrt5-2$, $\hat{{\mbox{\boldmath $B$}}}=\hat{{\mbox{\boldmath $z$}}}$, $L=200$, $(\Gamma, \alpha, V_\mathrm{Q}) = (0.3, 0.3, 0.5)$ and $-0.6\leq\mu\leq1$, $0 \leq \phi_0 \leq \pi$. Lower figures: Grayscale plots of $\epsilon^+$ for the same set of parameters except that $\Delta = 0.1$ is introduced and $L=50, 100, 200$.[]{data-label="fig:f-L200lambda0.5"}](Full_Fig6_200){width="8.66cm"}
It has been known (see *e.g.* [@Mei2012; @Ganeshan1301.5639]) for systems with $\Gamma = \alpha = 0$ that when we fix the value of $\kappa$ one or more in-gap states emerge within the energy gaps due to the potential $\epsilon_{l}$. Such in-gap states also exist in our model with non-zero $\Gamma$ and $\alpha$, and can cross each other or another sub-band because the RZ bands are shifted in energy. As we change the value of the phase of the site energy modulation, $\phi_0$, the energies of these in-gap states change rapidly, while other states in ‘bulk’ sub-bands do not change significantly.
In Fig. \[fig:f-L200lambda0.5\] we change the value of $\phi_0$ to study the effect of such in-gap states. We have plotted the single particle state energies color-coded by the value of $\langle S_z \rangle$ as well as the value of $\epsilon^+$ for different system sizes. The plot at the bottom with $L = 200$ looks similar to that of the single particle state energies at the top, except that the regions with two overlapping sub-bands do not have a vanishing $\epsilon^+$.
We observe that the dependence of the eigenvalues on the choice of the phase $\phi_0$ becomes weaker as $L$ increases. Bulk sub-bands are not shifted, and crossing with a single in-gap state does not usually break the effectively single band situation. Therefore the topological equivalence between systems with different modulation phases is clear. In the next section we study what happens if the modulation phase abruptly changes inside the quantum wire, or if the modulation disappears from a part of the wire.
Effects of different types of lattice modulation
================================================
Effect of phase jumps
---------------------
In applications to the quantum information field, namely quantum computation utilizing the pair annihilation or creation of multiple Majorana fermions via gates, [@Wu1302.3947] end Majorana fermions should be stable against minor changes of the condition of the internals of the one-dimensional system, which would occur when two or more quantum wires are joined via gates. Our model of modulated lattices is characterized by the pair of the wavenumber $\kappa$ and the phase $\phi_0$. Here, it is of much interest what happens if we have phase jumps of the lattice modulation, which would correspond to joints between quantum wires, in our system.
Let us consider a system with $N_\mathrm{J}$ phase jumps, $$\epsilon_{l} = V_\mathrm{Q} \cos( \kappa (l-l_\textrm{c}) + \phi_0 + \lfloor l/W \rfloor \phi_\mathrm{J}),
\label{eqn:sitePotentialJump}$$ in which $\lfloor x \rfloor$ denotes the largest integer not exceeding $x$ and $W = L / (N_\mathrm{J} + 1)$ is the distance between phase jumps of $\phi_\mathrm{J}$. If the phase jumps do not affect the Majorana modes, the regions of $\mu$ with vanishing $\epsilon^\pm$ should not change, and when they vanish, the corresponding eigenmodes should occupy the ends of the system in spite of the internal phase jumps.
![(a) Grayscale plots of $\epsilon^+$ for $g=\sqrt5-2$, $\hat{{\mbox{\boldmath $B$}}}=\hat{{\mbox{\boldmath $z$}}}$, $L=200$, $(\Gamma, \alpha, V_\mathrm{Q}) = (0.3, 0.3, 0.5)$, $\Delta = 0.1$, $-0.6\leq\mu\leq1$, $N_\mathrm{J}$ and $0 \leq \phi_\mathrm{J} \leq \pi$. (b) Value of $\epsilon^+$ plotted against $\mu$ for $N_\mathrm{J} = 1, 3, 7$ and $\phi_\mathrm{J} = \pi$. Eigenstates for $\mu = 0.04$ are plotted for (c) $N_\mathrm{J} = 1$, (d) $N_\mathrm{J} = 3$, and (e) $N_\mathrm{J} = 7$. []{data-label="fig:phiJ"}](Full_Fig7_200){width="8.66cm"}
In Fig. \[fig:phiJ\] we plot the value of $\epsilon^+$ for (a) a single jump with different sizes of phase jump $\phi_\mathrm{J}$, and (b) different numbers of phase jumps with $\phi_\mathrm{J} = \pi$. In Fig. \[fig:phiJ\] (a), $\phi_\mathrm{J} = 0$ corresponds to a system without a phase jump. Introduction of a single phase jump almost does not change the locations of the regions with vanishing $\epsilon^\pm$, though we observe a few curves of kinks in $\epsilon^+$ corresponding to a single particle state running between RZ bands as $\phi_\mathrm{J}$ is changed. Increasing the number of phase jumps does not change the picture significantly, as long as the localization of the end modes is almost within a single section between phase jumps. This is observed in Fig. \[fig:phiJ\] (b); while the values of $\epsilon^+$ sometimes differ between systems with different numbers of phase jumps of $\pi$, the regions with $\epsilon^+\ll 1$ is not shifted or removed. Also we find in Fig. \[fig:phiJ\] (c–e) that the eigenvalues of $M_\mathrm{BdG}$ other than $\epsilon^\pm$ do not vanish in this case. There are only one pair of Majorana fermions appearing at the both ends of the system, rather than more than one pairs of them appearing also at phase jumps.
The results above reflect the fact that a phase jump does not change the topological character of the system. If both sides of the introduced phase jump are in the state characterized by the same topological quantum number, boundary states do not form at the phase jump.
Site level modulation limited to a part of the system
-----------------------------------------------------
![ (a) Value of $\epsilon^+$ plotted against $\mu$ for $n_\mathrm{B} = 100, 140, 180$ and $g=\sqrt5-2$, $\hat{{\mbox{\boldmath $B$}}}=\hat{{\mbox{\boldmath $z$}}}$, $L=200$, $(\Gamma, \alpha, V_\mathrm{Q}) = (0.2, 0.3, 0.5)$, $\Delta = 0.1$. Eigenstates for $n_\mathrm{B} = 100$ are plotted for (b) $\mu = -0.28$ (only the right side of the system is single band), (c) $\mu = 0$ (both sides are single band), and (d) $\mu = 0.1$ (only the left side is single band). []{data-label="fig:partialModulation"}](Full_Fig8_200){width="8.66cm"}
The view above is further confirmed when we remove the lattice modulation from some part of the system, by having $$\epsilon_l = \left\{
\begin{array}{cc}
0, & (0 \leq l < n_\mathrm{B})\\
V_\mathrm{Q} \cos(\kappa(l - l_c)). & (n_\mathrm{B} \leq l < L)
\end{array}
\right.$$ In Fig. \[fig:partialModulation\] (a) we plot the value of $\epsilon^+$ for different values of $n_\mathrm{B}$ for the case in which the part of the system to the left of site $n_\mathrm{B}$ has the same parameter as in Fig. \[fig:localization\] (a), whereas the right part has the same parameter as in Fig. \[fig:localization\] (b). For $n_\mathrm{B} = L/2 = 100$ we observe that $\epsilon^+$ vanishes when it vanishes either in Fig. \[fig:localization\] (a) with $V_\mathrm{Q} = 0$ or (b) with $V_\mathrm{Q} = 0.5$. In Fig. \[fig:partialModulation\] (b–d) we have plotted the spatial distribution of eigenvectors of (\[eqn:BdG\]) corresponding to $\epsilon^+$. In Fig. \[fig:partialModulation\] (b) with $\mu = -0.28$, only the right side of the system with $V_\mathrm{Q} = 0.5$ is an effectively single band system at the chemical potential and becomes a topological superconductor. The eigenvector has most of its amplitude localized equally to the two ends of this region, and the value of $\sqrt{\langle x^2 \rangle}$ is close to $L/\sqrt{2}$, as expected in such a case. Also, in Fig. \[fig:partialModulation\] (d) with $\mu = 0.1$, only the left side of the system without the lattice modulation becomes a topological superconductor, and the eigenvector is this time localized to the two ends of the left part, with the value of $\sqrt{\langle x^2 \rangle}$ similar to that in (b). On the other hand, in Fig. \[fig:partialModulation\] (c) both sides are topologically nontrivial, and the eigenvector has its amplitudes localized at the two ends of the whole system.
As we enlarge the part without lattice modulation by increasing the value of $n_\mathrm{B}$ in Fig. \[fig:partialModulation\] (a), the plot of $\epsilon^+$ approaches that in Fig. \[fig:localization\] (a). This is because the part with the lattice modulation cannot form a well-defined topological superconductor if it is too short.
In summary, when two TS regions are joined, the resulting system become a TS with end Majorana states at the ends. On the other hand, if a TS is joined with a topologically trivial chain, the resulting system becomes a TS whose Majorana states appear close to where they had been in the original TS. This does not depend on which of the chains has spatial modulation. If the chemical potential can be controlled in the system one can control which region has boundary Majorana modes in the setting above.
Case of a hopping modulation
----------------------------
![(Color online) Effect of quasiperiodic modulation of the hopping parameter — Top: Single particle state energy for $\hat{{\mbox{\boldmath $B$}}}=\hat{{\mbox{\boldmath $z$}}}$, $L=200$, $(\Gamma, \alpha, V_\mathrm{J}) = (0.3, 0.3, 0.5)$ and $-0.6\leq\mu\leq1$, $0 \leq \kappa \leq \pi$. Bottom: $\epsilon^+$ calculated for the same parameters and $\Delta = 0.1$. []{data-label="fig:hopMod05"}](Full_Fig9_200){width="8.66cm"}
The correspondence between the ‘diagonal’ modulation in the lattice site energy and the ‘off-diagonal’ one in the hopping amplitude in quasiperiodic systems has attracted a renewed attention. [@Kraus2012; @Thiem1212.6337] Here we study the effect of such a modulation in the hopping amplitude in our system, in which case the Hamiltonian is given by $$\begin{aligned}
\mathcal{H}
&=& -\sum_{l=0}^{L-2}\frac{t_l}{2}\sum_{\sigma=\uparrow,\downarrow}
(\hat c_{\sigma,l}^\dag \hat c_{\sigma,l+1} + \mathrm{h.c.})\nonumber\\
&+& \frac{\alpha}{2} \sum_{l=0}^{L-2}
\left(
(\hat c_{\downarrow,l}^\dag \hat c_{\uparrow,l+1}
-\hat c_{\uparrow,l}^\dag \hat c_{\downarrow,l+1}) + \mathrm{h.c.}\right)\nonumber\\
&+& \sum_{l=0}^{L-1} \left(\Delta(\hat c_{\uparrow,l} \hat c_{\downarrow,l} + \mathrm{h.c.})
+\frac{2\Gamma}{\hbar}\hat{{\mbox{\boldmath $B$}}}\cdot{\mbox{\boldmath $S$}}_l\right)\nonumber\\
&+& \sum_{l=0}^{L-1} \sum_{\sigma=\uparrow,\downarrow}(t-\mu+\epsilon_{l}) \hat n_{\sigma,l}
,
\label{eqn:Hamiltonian_offDiagMod}\end{aligned}$$ with $$t_l = t\left[1 + V_\mathrm{J} \cos\left(\kappa (l + 1/2 - l_c) + \phi_0\right)\right]$$ and $\epsilon_{l} = 0$.
In Fig. \[fig:hopMod05\] we plot the single particle state energy as well as the value of $\epsilon^+$ obtained by solving the BdG equation similar to (\[eqn:BdG\]) but the diagonal (spin-preserving) blocks of the kinetic term substituted by the one with the modulated hopping parameter. We observe that, while the details of the single particle spectrum is changed, the correspondence between the effectively single-band region in the spectrum and the zeros of the $\epsilon^+$ is also observed here.
The symmetry of the Hamiltonian is not changed by going from the site level modulation to the hopping parameter modulation. Therefore similar response to phase jumps or partial modulations inside the system is expected for the latter case as in the former case, which has been studied in the above.
Conclusion
==========
In summary, we have studied the effect of different types of spatial modulations on the realization of a topological superconductor in a 1D conductor with proximity-induced superfluidity and spin-orbit coupling, extending our previous work, Ref. .
The combination of a quasiperiodic site energy modulation with the external Zeeman field and the spin-orbit coupling results in the single particle state energy distribution having a fractal pattern, *double Hofstadter butterfly*. Within the mean-field, Bogoliubov-de Gennes approximation, our model Hamiltonian can be diagonalized. We have demonstrated that the smallest positive eigenvalue of the Hamiltonian is strongly governed by the single particle energy spectrum for relatively weak induced superfluidity. Localized end modes, which are Majorana fermions, exist when two eigenvalues are degenerate at zero energy. As we change the chemical potential or the modulation wavenumber, we observe multiple reentrant transitions into and out of topologically nontrivial states. However, for stronger superfluidity, small patterns of the double Hofstadter butterfly are smeared from the eigenvalue plot showing the topologically nontrivial parameter ranges. The resulting topological superconductor is sensitive to the direction of the magnetic field, while the phase of the modulation does not affect the system.
We have also studied the effects of the phase jump of the quasiperiodic potential and what happens when the potential is applied to only a part of the quantum wire. The results reflect that the system is characterized by a $Z_2$ quantum number, that is, all the topologically nontrivial states are indistinguishable, and if two regions with such states are joined, the Majorana end modes appear only at the ends of the resulting system. If the chemical potential can be changed, the locations of the end modes can be manipulated. A quasiperiodic hopping modulation also exhibits a similar phase diagram with reentrant topological transitions.
Recently a scheme for topological superconductivity without proximity effect has been proposed [@Farrell2012]. Our study of the correlation between the band structure and realization of topological superconductivity could also be relevant in such cases.
M. T. acknowledges the hospitality of Cavendish Laboratory, University of Cambridge, where this work was completed. This work was partially supported by the Grant-in-Aid for the Global COE Program “The Next Generation of Physics, Spun from Universality and Emergence” from MEXT of Japan. N. K. is supported by KAKENHI (Grants No. 22103005 and No. 25400366) and JSPS through its FIRST Program. Part of the computation in this work has been performed using the facilities of the Supercomputer Center, Institute for Solid State Physics, University of Tokyo and Yukawa Institute for Theoretical Physics, Kyoto University.
[99]{} N. Read and D. Green, Phys. Rev. B **61**, 10267 (2000). A. Yu. Kitaev, Phys.-Usp. **44**, 131, 268 (2001). D. A. Ivanov, Phys. Rev. Lett. **86**, 268 (2001). L. Fu and C. L. Kane, Phys. Rev. Lett. **100**, 096407 (2008). S. Fujimoto, Phys. Rev. B **77**, 220501 (2008); M. Sato and S. Fujimoto, Phys. Rev. B **79**, 094504 (2009). M. Sato, Y. Takahashi, and S. Fujimoto, Phys. Rev. Lett. **103**, 020401 (2009). Y. Tanaka, T. Yokoyama, and N. Nagaosa, Phys. Rev. Lett. **103**, 107002 (2009); J. Linder, Y. Tanaka, T. Yokoyama, A. Sudbø, and N. Nagaosa, Phys. Rev. Lett. **104**, 067001 (2010). J. D. Sau, R. M. Lutchyn, S. Tewari, and S. Das Sarma, Phys. Rev. Lett. **104**, 040502 (2010). J. Alicea, Phys. Rev. B **81**, 125318 (2010). R. M. Lutchyn, J. D. Sau, and S. Das Sarma, Phys. Rev. Lett. **105**, 077001 (2010). Y. Oreg, G. Refael, and F. von Oppen, Phys. Rev. Lett. **105**, 177002 (2010). S. Nakosai, Y. Tanaka, N. Nagaosa, Phys. Rev. Lett. **108**, 147003 (2012). V. Mourik, K. Zuo, S. M. Frolov, S. R. Plissard, E. P. A. M. Bakkers, and L. P. Kouwenhoven, Science **336**, 1003 (2012). J. G. Rodrigo, V. Crespo, H. Suderow, S. Vieira and F. Guinea, **109**, 237003 (2012). M. T. Deng, C. L. Yu, G. Y. Huang, M. Larsson, P. Caroff, and H. Q. Xu, Nano Lett. **12**, 6414 (2012). L. P. Rokhinson, X. Liu, and J. K. Furdyna, Nature Phys. **8**, 795 (2012). T. D. Stanescu and S. Tewari, J. Phys.: Condens. Matter **25**, 233201 (2013). O. Motrunich, K. Damle, and D. A. Huse, Phys. Rev. B **63**, 224204 (2001). I. A. Gruzberg, N. Read, and S. Vishveshwara, Phys. Rev. B **71**, 245124 (2005). P. W. Brouwer, M. Duckheim, A. Romito, and F. von Oppen, Phys. Rev. Lett. **107**, 196804 (2011); **84**, 144526 (2011). R. M. Lutchyn, T. D. Stanescu, and S. Das Sarma, **85**, 140513(R) (2012). A. M. Lobos, R. M. Lutchyn, and S. Das Sarma, **109**, 146403 (2012). D. Bagrets and A. Altland, **109**, 227005 (2012); P. Neven, D. Bagrets, and A. Altland, New J. Phys. **15**, 055019 (2013). S. Takei, B. M. Fregoso, H. Y. Hui, A. M. Lobos, and S. Das Sarma, **110**, 186803 (2013). İ. Adagideli, M. Wimmer, and A. Teker, arXiv:1302.2612. J. D. Sau, and S. Das Sarma, arXiv:1305.0554. G. Roati *et al.*, Nature **453**, 895 (2008); E. Lucioni *et al.*, Phys. Rev. Lett. **106**, 230403 (2011). V. Brouet *et al.*, **87**, 041106 (R). B. Hunt *et al.*, Science **340**, 1427 (2013). A. P. Schnyder, S. Ryu, A. Furusaki, and A. W. W. Ludwig, **78**, 195125 (2008); A. Kitaev, arXiv:0901.2686. M. Verbin, O. Zilberberg, Y. E. Kraus, Y. Lahini, and Y. Silberberg, **110**, 076403 (2013). M. Tezuka and N. Kawakami, Phys. Rev. B **85**, 140508(R). X. Cai, L.-J. Lang, S. Chen, and Y. Wang, **110**, 176403 (2013). W. DeGottardi, D. Sen, and S. Vishveshwara, **110**, 146404 (2013). I. I. Satija and G. G. Naumis, arXiv:1210.5159. L.-J. Lang and S. Chen, **86**, 205135 (2012). S. Ganeshan, K. Sun, and S. Das Sarma, **110**, 180403 (2013). E. M. Stoudenmire, J. Alicea, O. A. Starykh, and M. P. A. Fisher, Phys. Rev. B **84**, 014503 (2011). M. Kohmoto, Phys. Rev. Lett. **51**, 1198 (1983); C. Tang and M. Kohmoto, Phys. Rev. B **34**, 2041 (1986). D. R. Hofstadter, Phys. Rev. B **14**, 2239 (1976). M. Iskin, Phys. Rev. A. **85**, 013622 (2012). X.-J. Liu, L. Jiang, H. Pu, and H. Hu, Phys. Rev. A **85**, 021603(R) (2012). A. C. Potter and P. A. Lee, Phys. Rev. Lett. **105**, 227003 (2010); Phys. Rev. B **83**, 094525 (2011). R. M. Lutchyn, T. D. Stanescu, and S. Das Sarma, Phys. Rev. Lett. **106**, 127001 (2011); T. D. Stanescu, R. M. Lutchyn, and S. Das Sarma, Phys. Rev. B **84**, 144522 (2011). M.-T. Rieder, G. Kells, M. Duckheim, D. Meidan, and P. W. Brouwer, **86**, 125423 (2012). F. Mei, S. -L. Zhu, Z. -M. Zhang, C. H. Oh, and N. Goldman, Phys. Rev. A **85**, 013638 (2012). L.-H. Wu, Q.-F. Liang, Z. Wang, and X. Hu, arXiv:1302.3947. Y. E. Kraus and O. Zilberberg, Phys. Rev. Lett. **109**, 116404 (2012). S. Thiem and M. Schreiber, arXiv:1212.6337. A. Farrell and T. Pereg-Barnea, **87**, 214517 (2013).
|
{
"pile_set_name": "ArXiv"
}
|
---
author:
- 'B. Berche[^1]'
- 'C. von Ferber[^2]'
- 'T. Holovatch[^3]'
- 'Yu. Holovatch[^4]'
date: 'Coventry, May 11, 2009'
title: Resilience of public transport networks against attacks
---
[leer.eps]{} gsave 72 31 moveto 72 342 lineto 601 342 lineto 601 31 lineto 72 31 lineto showpage grestore
Introduction {#I}
============
The question of resilience or vulnerability of a complex network [@networks] against failure of its parts has, beside purely academic interest a whole range of important practical implications. In what follows below any such failure will be called an [*attack*]{}. In practice, the origin of the attack and its scenario may differ to large extent, ranging from random failure, when a node or a link in a network is removed at random to a targeted destruction, when the most influential network constituents are removed according to their operating characteristics. The notion of attack vulnerability of complex networks originates from studies of computer networks and was coined to denote the decrease of network performance as caused by the removal of either nodes or links. The behavior of a complex network under attack has been observed to drastically differ from that of regular lattices. Early evidence of this fact was found in particular for real world networks that show scale-free behavior: the world wide web and the internet [@Albert00; @Tu00], as well as metabolic [@Jeong00], food web [@Sole01], and protein [@Jeong01] networks. It appeared that these networks display a high degree of robustness against random failure. However, if the scenario is changed towards targeted attacks, the same networks may appear to be especially vulnerable [@Cohen00; @Callaway00].
Essential progress towards a theoretical description of the attack vulnerability of complex networks is due to the application of the tools and concepts of percolation phenomena [@percolation]. On a lattice percolation occurs e.g. when at a given concentration of bonds a spanning cluster appears. This concentration $c_{\rm perc}$ which is determined by an appropriate ensemble average in the thermodynamic limit is the so-called percolation threshold which is in general lattice dependent. On a general network the corresponding phenomenon is the emergence of a giant connected component (GCC) i.e. a connected subnetwork which in the limit of an infinite network contains a finite fraction of the network. For a random graph where given vertices are linked at random this threshold has been shown to be reached at one bond per vertex [@Erdos]. However the distribution $p(k)$ of the degrees $k$ of vertices in a random graph is Poissonian. A more general criterion applicable to networks with given degree distribution $p(k)$ but otherwise random linking between vertices has been proposed by Molloy and Reed [@Molloy; @Cohen00; @Callaway00]. For such equilibrium networks a GCC can be shown to be present if $$\label{1b}
\langle k(k-2) \rangle \geq 0$$ with the appropriate ensemble average $\langle \dots \rangle$ over networks with given degree distribution. Defining the Molloy-Reed parameter as the ratio of the moments of the degree distribution $$\label{1}
\kappa^{(k)} \equiv \langle k^2 \rangle / \langle k \rangle$$ the percolation threshold can then be determined by $$\label{1a}
\kappa^{(k)}=2
\hspace{1em}\mbox{ at }\hspace{1em} c_{\rm perc}.$$ Taken that for scale-free networks the degree distribution obeys power law scaling $$\label{2}
p(k) \sim k^{-\gamma}$$ one finds that the second moment $\langle k^2 \rangle$ diverges for $\gamma<3$. Thus, the value $\gamma=3$ separates two different regimes for the percolation on equilibrium scale free networks [@Cohen00]. Indeed, for infinite equilibrium scale-free networks $\kappa^{(k)}$ (\[1\]) remains finite for $\gamma>3$, however for $\gamma<3$ a GCC is found to exist at any concentration of removed sites: the network appears to be extremely robust to random removal of nodes. Therefore, observed transitions for real-world systems [@Albert00; @Tu00; @Jeong00; @Sole01; @Jeong01] from the theoretical standpoint may be seen as finite-size effects or resulting from essential degree-degree correlations. The tolerance of scale-free networks to intentional attacks (when the highest degree nodes are removed) was studied in Ref. [@Cohen01]. It was shown that even networks with $\gamma<3$ may be sensitive to intentional attacks.
Obviously, the above theoretical results apply to ideal complex networks and for ensemble averages and may be confirmed within certain accuracy when applied to different individual real-world networks. Not only finite-size effects are the origin of this discrepancy [@Kalisky06]. Furthermore, even networks of similar type (e.g. of similar node degree distribution and size) may be characterized by a large variety of other characteristics. While some of them may have no impact on the percolation properties [@Xulvi-Brunet03], others do modify their behavior under attack, as empirically revealed in Ref. [@Holme02] for two different real-world scale-free networks (computer and collaboration networks). Therefore, an empirical analysis of the behavior of different real-world networks under attack appears timely and will allow not only to elaborate scenarios for possible defence mechanisms of operating networks but also to create strategies of network constructions, that are robust to attacks of various types.
In this paper, we present results of the analysis of the behavior of networks of public transport in large cities (public transport networks, PTNs) and consider attacks by various scenarios. To our knowledge the resilience of PTNs under attack has so far not been treated in terms of complex network concepts. Furthermore, in parallel we analyze a number of complex networks of the same type. Previous analysis usually focussed on a single instance of a network of given type [@note2]. Our study intends to show that even within a sample of several networks that were created for the same purpose, namely PTNs, one may observe essential diversity with respect to the behavior under attacks of various scenarios.
As we have mentioned above, the attack resilience of a network may be tested within a variety of different attack scenarios. In a given one, a list of nodes ordered by decreasing degree may be prepared for the unperturbed network and the attack successively removes vertices according to this original list [@Barabasi99; @Broder00]. In a slightly different scenario the vertex degrees are recalculated and the list is reordered after each removal step [@Albert00]. In initial studies only little difference between these two scenarios was observed [@Callaway00], however further analysis showed [@Holme02; @Girvan02] that attacks according to recalculated lists often turn out to be more harmful than the attack strategies based on the initial list, suggesting that the network structure changes as important vertices or edges are removed. Other scenarios consider attacks following an order imposed by other measures of the centrality of a node, e.g. the so-called betweenness centrality [@Holme02]. In particular for the world-wide airport network, it has been shown recently [@Guimera04; @Guimera05] that nodes with higher betweenness play a more important role in keeping the network connected than those with high degree. In our study, we will make use of the scenarios proposed so far as well as develop further algorithms to perform network attacks.
The paper is organized as follows, in the next Section we describe the database, define observables in terms of which we are going to follow the changes in the network properties under attacks, and describe the different attack strategies we will use. We display our principal results in sections \[III\], \[IV\]. There, we formulate criteria that allow to estimate the resilience of networks against attacks and discuss behavior of the PTNs during attacks following different strategies, outlining the most effective ones. Conclusions and an outlook are given in Section \[V\].
Databases, observables, and attack strategies {#II}
=============================================
This study continues our analysis of the properties of PTNs initiated in Refs. [@Ferber; @Ferber09a; @Ferber09b]. As in these works, we rely on the publicly available information about PTNs of a set of fourteen major cities of the world [@database]. Our choice for the selection of these cities was motivated by the idea to collect network samples from cities of different geographical, cultural, and economical background. In Table \[tab1\] we give some information summarizing the empirical analysis of some of the properties of the PTNs under consideration.
--------------------- --------------------- ---------------------
  
[**a.**]{} [**b.**]{} [**c.**]{}
--------------------- --------------------- ---------------------
1.2mm
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
City Type $N$ $R$ $\langle k_{\mathds{L}} $\ell_{\mathds{L}}^{\rm max}$ $\langle $c_{\mathds{L}}$ $\kappa^{(z)}_{\mathds{L}}$ $\kappa^{(k)}_{\mathds{L}}$ $\gamma_{\mathds{L}}$ $\langle k_{\mathds{P}} \rangle$ $\ell_{\mathds{P}}^{\rm max}$ $\langle \ell_{\mathds{P}} \rangle$ $c_{\mathds{P}}$ $\kappa^{(z)}_{\mathds{P}}$ $\kappa^{(k)}_{\mathds{P}}$ $\gamma_{\mathds{P}}$
\rangle$ \ell_{\mathds{L}} \rangle$
------------- ------- ------- ------ ------------------------- ------------------------------- ---------------------------- ------------------ ----------------------------- ----------------------------- ----------------------- ---------------------------------- ------------------------------- ------------------------------------- ------------------ ----------------------------- ----------------------------- -----------------------
Berlin BSTU 2992 211 2.58 68 18.5 52.8 1.96 3.16 (4.30) 56.61 5 2.9 41.9 11.47 84.51 (5.85)
Dallas B 5366 117 2.18 156 52.0 55.0 1.28 2.35 5.49 100.58 8 3.2 48.6 11.23 145.65 (4.67)
Düsseldorf BST 1494 124 2.57 48 12.5 24.4 1.96 3.16 3.76 59.01 5 2.6 19.7 10.56 91.17 (4.62)
Hamburg BFSTU 8084 708 2.65 156 39.7 254.7 1.85 3.26 (4.74) 50.38 11 4.7 132.2 7.96 79.43 4.38
Hong Kong B 2024 321 3.59 60 11.0 60.3 3.24 5.34 (2.99) 125.67 4 2.2 11.7 10.20 232.73 (4.40)
Istanbul BST 4043 414 2.30 131 29.7 41.0 1.54 2.69 4.04 76.88 6 3.1 41.5 10.59 140.13 (2.70)
London BST 10937 922 2.60 107 26.5 320.6 1.87 3.22 4.48 90.60 6 3.3 90.0 16.94 166.95 3.89
Los Angeles B 44629 1881 2.37 210 37.1 645.3 1.59 2.73 4.85 97.99 11 4.4 399.6 17.21 159.86 3.92
Moscow BEST 3569 679 3.32 27 7.0 127.4 6.25 7.91 (3.22) 65.47 5 2.5 38.0 26.48 130.65 (2.91)
Paris BS 3728 251 3.73 28 6.4 78.5 5.32 6.93 2.62 50.92 5 2.7 59.6 24.06 88.89 3.70
Rome BT 3961 681 2.95 87 26.4 163.4 2.02 3.67 (3.95) 69.05 6 3.1 41.4 11.34 108.08 (5.02)
Saõ Paolo B 7215 997 3.21 33 10.3 268.0 4.17 5.95 2.72 137.46 5 2.7 38.2 19.61 333.73 (4.06)
Sydney B 1978 596 3.33 34 12.3 82.9 2.54 4.37 (4.03) 42.88 7 3.0 33.6 7.79 74.63 (5.66)
Taipei B 5311 389 3.12 74 20.9 186.2 2.42 4.02 (3.74) 236.65 6 2.4 15.4 12.96 415.46 (5.16)
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
There are various ways to represent a PTN in terms of a graph [@ptns]. These different representations allow for a comprehensive analysis of various PTN properties reflecting their operating functions. It is natural to perform the analysis of PTN attack resilience in terms of these representations. These are briefly summarized in Fig. \[fig1\]. For the purpose of the present analysis, we will make use of the so-called $\mathds{L}$ and $\mathds{P}$-space graphs. In $\mathds{L}$-space representation [@ptns] the PTN is represented by a graph with nodes that correspond to the stations, whereas links correspond to connections between stations within one stop distance (Fig. \[fig1\]b). In the $\mathds{P}$-space [@Sen03] all station-nodes that belong to the same route form of a complete subgraph of the network (Fig. \[fig1\]c).
{width="80mm"} {width="80mm"}
[**a.**]{}
Let us take the $\mathds{L}$-space representation to introduce the observables we will use to quantify the PTN behavior under attack. Keep in mind however, that in our analysis presented in Section \[III\] we will deal also with the $\mathds{P}$-space. There are two intrinsically connected questions that naturally arise when one wants to describe quantitatively how a certain network changes when its nodes are removed [@note3]. The first is how to choose the ’order-parameter’ variable that signals the quantitative change in the network behavior (i.e. the break down of the network), the second is how to locate the value of concentration of removed nodes at which this change occurs. As we have mentioned in the introduction, in a theoretical description a useful quantity is the GCC: its disappearance can be associated with a network breakdown. Strictly speaking, the GCC is well-defined only in the $N\to \infty$ limit, therefore in practice dealing with a network of a finite size $N$ it is substituted by the size of the largest connected component. We will use in the following its normalized value defined by: $$\label{3}
S=N_1/N,$$ with $N$ and $N_1$ being number of nodes of the network and of its largest component correspondingly. By definition (\[3\]), a largest component is always present in a network of non-zero size. A useful quantity to measure network connectivity is the average shortest path: $$\label{4}
\langle \ell \rangle = \frac{2}{N(N-1)}\sum_{i>j}\ell (i,j),$$ where $\ell (i,j)$ is the length of a shortest path from node $i$ to $j$ and the sum spans all pairs $i,j$ of sites of the network. However, $\langle \ell \rangle$ is ill-defined for a disconnected network. Alternatively, one can suitably define the mean inverse shortest path length [@Holme02] by: $$\label{5}
\langle \ell^{-1} \rangle = \frac{2}{N(N-1)}\sum_{i>j}\ell^{-1}
(i,j),$$ with $\ell^{-1} (i,j)=0$ if nodes $i,j$ are disconnected. As one can see, Eq. (\[5\]) is well-defined even for a disconnected network and as such can be used to trace changes of network behavior under attack. To give an example, we show in Fig. \[fig2\] how the largest component fraction $S$, Eq. (\[3\]) and the mean inverse shortest path length $\langle \ell^{-1} \rangle$, Eq. (\[5\]), change upon random removal of nodes in each of fourteen PTNs selected for our study. More precisely, we measure these quantities as functions of the fraction of removed nodes $c$ starting from the unperturbed network ($c=0$) and eliminating at random step-by-step 1 % of the nodes up to $c=1$. In what follows below we will call this scenario a [*random scenario*]{}.
Already this first attack attempt brings about interesting (and in part unexpected) PTN features. Namely:
\(i) different PTNs react on random removal of their nodes in different ways, that range from rapid abrupt breakdown (Dallas) to a slow almost linear decrease (Paris);
\(ii) although qualitatively similar, the observed impact of the attack differs depending on which variable is used as indicator, either $S$ or $\langle \ell^{-1} \rangle$. Ordering the PTNs by their vulnerability, this order may thus differ depending on the applied indicator;
\(iii) up to $c=1$, there is no general ’percolation threshold’ concentration of removed nodes $c$ at which $S$ (or $\langle \ell^{-1} \rangle$) vanishes that would hold for all PTNs. Rather for some individual PTNs one observes various values of $c$ at which these PTNs show abrupt changes of their properties.
Figs. \[fig2\] [**a**]{},[**b**]{} display how the different PTNs react on a [*random*]{} removal of their nodes. Obviously, the question immediately arises how this behavior changes if one removes the nodes not at random, but following a given order or scheme (we call this the scenario of the attack). As we have mentioned in the introduction, a number of different attack scenarios have been proposed . These are generally based on the intuitive assumption that the largest impact on a network is caused by the removal of its most ’important’ nodes. A number of indicators have been developed in particular in applications of graph theory for social science to measure the importance of a node. Besides the node degree $k_j$, which is equivalent to the number of nearest neighbors $z_1(j)$ of a given node $j$, different centralities have been introduced for this purpose. In particular, the closeness $C_C(j)$, graph $C_G(j)$, stress $C_S(j)$, and betweenness centralities $C_B(j)$ of a node $j$ are defined as follows (see e.g. [@Brandes01]): $$\begin{aligned}
\label{6}
C_C(j) &=& \frac{1}{\sum_{t\in \cal{N}} \ell(j,t)}, \\
\label{7}
C_G(j) &=& \frac{1}{{\rm
max}_{t\in \cal{N}} \ell(j,t)}, \\
\label{8}
C_S(j) &=& \sum_{s\neq j \neq t\in \cal{N}} \sigma_{st}(j), \\
\label{9}
C_B(j) &=& \sum_{s\neq j \neq t\in \cal{N}}
\frac{\sigma_{st}(j)}{\sigma_{st}}.
\end{aligned}$$ In Eqs. (\[6\])–(\[8\]), $\ell(j,t)$ is the length of a shortest path between the nodes $j,t$ that belong to the network $\cal{N}$, $\sigma_{st}$ is the number of shortest paths between the two nodes $s,t\in \cal{N}$, and $\sigma_{st}(j)$ is the number of shortest paths between nodes $s$ and $t$ that go through the node $j$. Alternatively, one may measure the importance of a given node $j$ by the number of its second nearest neighbors $z_2(j)$ or its clustering coefficient $C(j)$. The latter is the ratio of the number of links $E_j$ between the $k_j$ nearest neighbors of $j$ and the maximal possible number of mutual links between them: $$\label{10}
C(j) =\frac{2E_j}{k_j(k_j-1)}.$$
![(color online). Largest component size of the PTN of Paris as function of the fraction of removed nodes for different attack scenarios. Each curve corresponds to a different scenario as indicated in the legend. Lists of removed nodes were prepared according to their degree $k$, closeness $C_C$, graph $C_G$, stress $C_S$, and betweenness $C_B$ centralities, clustering coefficient $C$, and next nearest neighbors number $z_2$. A superscript $i$ refers to lists prepared for the initial PTN before the attack. RV and RN denote the removal of a random vertex (RV) or of its randomly chosen neighbor (RN), respectively.[]{data-label="fig3"}](S_Paris.eps){width="80mm"}
Removing important nodes according to lists prepared in the order of decreasing node degrees $k$, centralities (\[6\])–(\[9\]), number of their second nearest neighbors $z_2$, and increasing clustering coefficient $C$ defines seven different attack scenarios. As we have already mentioned in the introduction, the scenarios can be either implemented according to lists prepared for the [*initial*]{} PTN before the attacks (we will indicate the corresponding scenario by a superscript $i$, e.g. $C_B^{i}$) or by lists rebuilt by recalculating the order of the remaining nodes after each step. Together, this leads to fourteen different attack scenarios. In addition, we will keep the above described random scenario (denoted further as RV) and add one scenario more, removing a randomly chosen neighbor of a randomly chosen node (RN). The latter scenario appears to be effective for immunization problems [@Cohen03] and it is based on the fact, that in this way nodes with a high number of neighbors will be selected with higher probability. Note that in this scenario only a neighbor node is removed and not the initially chosen one.
All together, this defines sixteen different scenarios to attack a network and we apply these to all fourteen PTNs that form our database. A typical result for a single PTN is displayed in Fig. \[fig3\]. Here, we show how the largest connected component size $S$ of the Paris PTN changes under the influence of the above described attack scenarios. Already from this plot one may discriminate between the most effective scenarios that result in a fast decrease of the largest component size (those governed by betweenness and stress centralities, node degree, and next nearest neighbors number – see the Figure) and the less harmful ones. In the following, instead of displaying the results of all attacks for all different PTNs we will focus on the results of the most effective scenarios comparing them with those of random failure as introduced by the random scenario. As outlined in the introduction, we make use of different PTN representations (different ’spaces’ of Fig. \[fig1\]). In the following section, we present the analysis of PTN resilience in the $\mathds{L}$-space representation.
{width="80mm"} {width="80mm"}
[**a.**]{}
Results in $\mathds{L}$-space {#III}
=============================
The $\mathds{L}$-space representation of a PTN is a graph that represents each station by a node, a link between nodes indicates that there is at least one route that services the two corresponding stations consecutively. No multiple links are allowed (see Fig. [\[fig1\]**b**]{}). Therefore, attacks in the $\mathds{L}$-space correspond to situations, in which given public transport stations cease to operate for all means of traffic that go through them. Note however, that in this representation, the removal of a station node does not otherwise interfere with the operation of a route that includes this station. It rather splits this route into two (operating) pieces. An alternative situation will be considered in the forthcoming section.
In order to answer some of the questions raised in Section \[II\], let us return to Fig. \[fig3\], where the impact on the largest component size $S$ of the PTN of Paris is shown for sixteen different attack scenarios as function of the fraction of removed nodes. As we have already remarked, for this PTN the most influential are the scenarios where nodes are removed according to lists ordered by $C_B$, $k$, $C_S$, $k^{i}$, $C_B^{i}$, $C_S^{i}$ (we list the characteristics in a decreasing order of effectiveness of the corresponding scenario). For a small value of $c$ ($c<0.07$) these scenarios cause practically indistinguishable impact on $S$ with a linear behavior $S\sim (1-c)$. As $c$ increases, deviations from the linear behavior arise and the impact of different scenarios start to vary. In particular, there appear differences between the role played by the nodes with highest value of $k$ and highest betweenness centrality $C_B$. Whereas the first quantity is a local one, i.e. it is calculated from properties of the immediate environment of each node, the second one is global. Moreover, the $k$-based strategy aims to remove a maximal number of edges whereas the $C_B$-based strategy aims to cut as many shortest paths as possible. In addition, there arise differences between the ’initial’ and ’recalculated’ scenarios, suggesting that the network structure changes as important nodes are removed. Similar behavior of $S(c)$ is observed for all PTNs included in this study, with certain peculiarities in the order of effectiveness of different attack scenarios. Note however, that the difference between ’initial’ and ’recalculated’ scenarios is less evident for strategies based on local characteristics, as e.g. the node degree or the number of second nearest neighbors (c.f. curves for $k$, $k^i$ and $z_2$, $z_2^i$, respectively). This difference between initial and recalculated characteristics is even more pronounced for the centrality-based scenarios.
Now let us return to some of the observations of Section \[II\]. Namely, we noted that the observed impact of an attack may differ depending on which observable is used as the ’order-parameter’ variable (c.f. Fig. \[fig2\] where this is shown for the RV attack scenario taking either $S$ or $\langle \ell^{-1} \rangle$ as ’order-parameter’). Similar differences we observe also in the case of the other scenarios. For the sake of uniqueness in the following we will use the value of $S$ to measure the effectiveness of a given attack. This choice is motivated by several reasons: (i) in an infinite network limit $S$ defines an order parameter of the classical percolation problem [@percolation]; (ii) differences between network resilience as judged e.g. by the behavior of $S$ or by that of $\langle \ell^{-1} \rangle$ are not significant enough to be a subject of special analysis (at least not for the PTNs we consider); (iii) considering $S$ naturally leads to other useful characteristics that allow to estimate the PTN operating ability and its segmentation. Let us stop to elaborate the latter point in more detail.
As we have already emphasized, there is no well defined ’percolation threshold’ concentration of removed nodes $c_{\rm perc}$ at which $S$ (or $\langle \ell^{-1} \rangle$) vanishes (see Figs. \[fig2\], \[fig3\]) which could serve as evidence of a break down of the largest PTN component and hence of the loss of operating ability [@note4]. In Ref. [@Ferber09b] it has been proposed to use the behavior of maximal shortest path length $\ell_{\rm max}$ as a possible indicator of the network break down. This was based on the observation, that as the concentration of removed nodes $c$ increases, the value of $\ell_{\rm max}$ for different PTNs displays similar typical behavior: initial growth and then an abrupt decrease when a certain threshold is reached (see e.g. Fig. \[fig4\] [**a**]{} where this value is shown for the recalculated highest degree attack scenario of the PTNs of Paris and London). Obviously, removing the nodes initially increases the path lengths as deviations from the original shortest paths need to be taken into account. Further removing nodes then at some point leads to the breakup of the network into smaller components on which the paths are naturally limited by the size of these components which explains the sudden decrease of their lengths. For comparison, in Fig. \[fig4\] [**b**]{} we show how the value of $S$ changes under the recalculated highest degree scenario for the above PTNs.
Being certainly useful for many instances of the PTNs analyzed, the above $\ell_{\rm max}$-based criterion cannot serve as an universal tool to determine the region of $c$, where the network stops to operate. One of the reasons is that for certain PTNs (as well as for certain attack scenarios) we have found that $\ell_{\rm max}$ does not show a pronounced maximum, but rather shows several maxima at different values of $c$. Therefore, to devise a criterion which may be equally well used for any of the networks we decided to define characteristic concentration of removed nodes $c_{\rm s}$ at which the size of the largest component $S$ decreases to one half of its initial value. This characteristic concentration allows us to compare the effective robustness of different PTNs or of the same PTN when different attack scenarios are applied. In what follows below, we will call this concentration the [*segmentation*]{} concentration $c_{\rm s}$, with the obvious condition: $$\label{11}
S(c_{\rm s}) = \frac{1}{2}S(c=0).$$
![(color online). $\mathds{L}$-space. Random scenario. Size of the largest cluster $S$ normalized by its value at $c=0$ as function of a fraction of removed nodes. From this figure it is easy to define the fraction of nodes $c_{\rm s}$ which satisfies Eq. (\[11\]).[]{data-label="fig5"}](SrandomL05_norm.eps){width="80mm"}
In Fig. \[fig5\] we plot the size of the largest connected component $S$ for different PTNs as function of the fraction of removed nodes $c$ for the random vertex scenario (RV) in $\mathds{L}$-space. The choice of the lowest $S$ value $S=1/2$ in this figure enables one to find the value $c_{\rm s}$ as the crossing point of $S(c)$ with the horizontal axis. The values of $c_{\rm s}$ obtained for this scenario are given in the last column of Table \[tab2\]. Note that the PTNs under consideration react on random attack in many different ways: some of them slowly decrease without any abrupt changes in $S$ (like PTNs of Paris, Moscow, Sydney) while others are characterized by rather fast decay of $S$ (Dallas, Los Angeles, Istanbul).
2.5mm
City $c_{\rm s}$ $c_{\rm s}$ $c_{\rm s}$ $c_{\rm s}$ $c_{\rm s}$ $c_{\rm s}$
------------- ------------- --------- ------------- --------- ------------- ----------- ------------- --------- ------------- ----------- ------------- ----
Berlin .060 $C_B$ .065 $k^{i}$ .065 $C_S$ .070 $k$ .075 $z_2$ .220 RV
Dallas .025 $k^{i}$ .030 $k$ .030 $C_B$ .045 $z_2$ .055 $z_2^{i}$ .090 RV
Düsseldorf .075 $C_B$ .080 $k$ .080 $k^{i}$ .095 $C_S$ .105 $z_2$ .240 RV
Hamburg .040 $C_B$ .040 $C_C$ .045 $C_S$ .045 $k^{i}$ .060 $z_2$ .150 RV
Hong Kong .030 $C_B$ .040 $C_C$ .050 $z_2^{i}$ .060 $C_S$ .090 $k^{i}$ .300 RV
Istanbul .025 $C_S$ .030 $C_C$ .030 $C_B$ .035 $k^{i}$ .035 $k$ .140 RV
London .055 $k$ .060 $k^{i}$ .065 $C_B$ .075 $C_C$ .085 $z_2$ .175 RV
Los Angeles .040 $k$ .060 $k^{i}$ .065 $z_2$ .075 $C_B$ .100 $z_2^{i}$ .130 RV
Moscow .070 $C_B$ .085 $C_S$ .085 $k$ .085 $k^{i}$ .100 $C_C$ .350 RV
Paris .105 $C_B$ .120 $k$ .125 $C_S$ .130 $k^{i}$ .140 $C_B^{i}$ .375 RV
Rome .050 $C_B$ .060 $C_C$ .065 $k$ .065 $k^{i}$ .085 $C_S$ .215 RV
Saõ Paolo .040 $k$ .040 $k^{i}$ .045 $C_B$ .060 $C_S$ .060 $C_S^{i}$ .320 RV
Sydney .040 $C_B$ .040 $C_C$ .065 $C_S$ .075 $k^{i}$ .085 $C_G,k$ .350 RV
Taipei .105 $C_B$ .105 $C_G$ .115 $k$ .120 $k^{i}$ .120 $C_C$ .240 RV
Now, applying these attacks according to the sixteen scenarios described above we are in the position to discriminate them by their degree of destruction and to single out those with the highest impact on each of the PTNs considered. To this end, for each PTN we give in Table \[tab2\] the segmentation concentration $c_{\rm s}$ for the five most harmful attack scenarios. The obtained values of $c_{\rm s}$ are given in increasing order. Near each value we denote the scenario that was implemented. Our analysis reveals the most harmful scenarios as those targeted at nodes with the highest values of either the node degree $k$, the betweenness centrality $C_B$, the next nearest neighbor number $z_2$, or the stress centrality $C_S$ recalculated after each step of the attack.
{width="80mm"} {width="80mm"}
It is instructive to observe correlations between the characteristics of unperturbed PTNs (see Table \[tab1\]) and their robustness to attacks. Such correlations may allow for an a priory estimate of the resilience of a network with respect to attacks. As discussed in the introduction, percolation theory for uncorrelated networks predicts that the value of the Molloy-Reed parameter $\kappa^{(k)}$, Eq. (\[1\]), can be used to measure the distance to the percolation point $\kappa^{(k)}=2$. We may therefore expect that networks with a higher value of $\kappa^{(k)}$ show higher resilience. To this end let us first compare the values of $c_{\rm
s}$ for certain scenarios with the value of $\kappa^{(k)}$ for the unperturbed PTN. Before doing this let us note that for an uncorrelated network the value of $\kappa^{(k)}$ can be equally represented by the ratio between the mean next neighbors number of a node $z_1$ (which is by definition equal to the mean node degree $\langle k \rangle$) and the mean second nearest neighbors number $z_2$: $$\label{12}
\kappa^{(z)}=z_2/z_1.$$ Indeed, given that for such a network (see e.g. [@networks]) $$\label{13}
z_2= \langle k^2 \rangle - \langle k \rangle ,$$ one can rewrite (\[1a\]) as: $$\label{14}
\kappa^{(z)}=1 \hspace{1em}\mbox{ at }\hspace{1em} c_{\rm perc}.$$ The relation $\kappa^{(k)}=\kappa^{(z)}+1$ holds only approximately for the real-world networks we consider in our study, as one can see, e.g., from the Table \[tab1\]. In Fig. \[fig6\][**a**]{} we compare both quantities $\kappa^{(k)}$, $\kappa^{(z)}$ for unperturbed PTNs with the corresponding segmentation concentration $c_{\rm s}$ for the random attack scenario. Within the expected scatter of data one can definitely observe a general tendency of $c_{\rm s}$ to increase with both $\kappa^{(k)}$ and $\kappa^{(z)}$: the higher the value of $\kappa$ for an unperturbed network, the more robust it is to random removal of its vertices. This conclusion, however with a more pronounced scatter of data even holds if one repeats the same analysis for the case of the scenario based on recalculated node degrees, as shown in Fig. \[fig6\][**b**]{}. Again, one observes $c_{\rm s}$ to increase with increasing $\kappa$. For the betweenness-based attack scenarios the data is even more scattered and a prediction based on the a priori calculated ratios is unreliable.
{width="80mm"} {width="80mm"}
[**a.**]{}
Another useful observation concerns the correlation between the PTN attack resilience and the node-degree distribution exponent $\gamma$ (\[2\]). As we have observed in the previous studies [@Ferber; @Ferber09a] some of the PTNs under consideration are scale-free: their node-degree distributions have been fitted to a power-law decay (\[2\]) with the exponents shown in Table \[tab1\]. Others are characterized rather by an exponential decay, but up to a certain accuracy they can also be approximated by a power-law behavior (then, the corresponding exponent is shown in Table \[tab1\] in brackets). In Fig. \[fig7\][**a**]{} we show the correlation between the fitted node-degree distribution exponent $\gamma$ and $c_{\rm s}$ for the random attack scenario. Filled circles correspond to scale-free PTNs, open circles correspond to the PTNs where the scale-free behavior is less pronounced. It is interesting to observe, that even if we include the PTNs which are better described by the exponential decay of the node-degree distributions, there is a notable tendency to find PTNs with smaller values of $\gamma$ to be more resilient as indicated by larger values of $c_{\rm s}$. This tendency is again confirmed if one considers the recalculated node degree attack scenario, as shown in Fig. \[fig7\][**b**]{}.
{width="80mm"} {width="80mm"}
[**a.**]{}
{width="80mm"} {width="80mm"}
[**c.**]{}
The above observed correlation between the exponent $\gamma$ that characterizes the unperturbed network (i.e. a PTN at $c=0$) and the segmentation concentration $c_{\rm s}$ at which however the PTN is to a large part unperturbed indicates that some global properties of the node-degree distribution may remain essentially unchanged when the nodes are removed (i.e. a scale-free distribution remains scale-free as $c$ increases, $0<c<c_{\rm s}$). To check that assumption for the RV scenario, we analyzed the averaged cumulative node degree distributions for each of the PTNs with 3,5, and 10 % of removed nodes. The cumulative distribution $P(k)$ is defined in terms of the node-degree distribution $p(q)$ (\[2\]) as: $$\label{15}
P(k)=\sum_{q=k}^{k^{\rm max}} p(q),$$ with $k^{\rm max}$ the maximal node degree in the given PTN. Typical results of this analysis are shown in Fig. \[fig8\], for the PTN of Paris. We compare the cumulative node degree distribution $P(k)$ of the unperturbed PTN with that of the PTN where a given fraction $c$ part of the nodes ($c=0.03$, 0.05, and 0.1, correspondingly) was removed according to the random attack scenario (RV). For each of the concentrations of the removed nodes, $P(k)$ was averaged over 2000 repeated attacks.
In the first plot, Fig. \[fig8\][**a**]{}, we compare the three resulting average distributions (for $c=0.03$, 0.05, and 0.1) with the original one ($c=0$). One clearly sees that there is no qualitative or even quantitative (change of exponent) change of the distributions for any of the three cases. Indeed, if one has a large set of nodes with a given node-degree distribution any sufficiently large random subset of these nodes should have the same distribution; in particular this holds if one averages these subset distributions over many instances. The above argument seems to ignore the change of degrees in the subset due to cutting off those vertices not remaining in the set. However, due to the random choice of the removed nodes the share of lost degree will on the average be proportional to the degree of each vertex: the higher its degree the more probable it is that one of its neighbors is chosen to be removed and this probability is proportional to its degree. Thus, the sum of degrees in the remaining subset is lower; but the degree distribution $P(k)$ is effectively transformed to $P'(ck) = n P(k)$ where $c$ is the probability of any node being removed and $P'(k)$ is the distribution in the remaining subset of nodes, $n$ a normalization. For an exponential distribution this transformation shifts the scale. However, a scale free distribution keeps its exponent under such a transformation.
In the other three plots, Figs. \[fig8\][**b**]{}-[**d**]{} we show for each amount of removed nodes the average cumulative distribution together with statistical errors calculated as the standard deviation within the ensemble of the 2000 instances generated in the sample. Even on the logarithmic scale these are very small for all but the very high degrees where fluctuations of small numbers of often less than one node for a given degree occur.
Results in $\mathds{P}$-space {#IV}
=============================
Let us complement the $\mathds{L}$-space analysis performed above by observing the reaction of PTN graphs under attack when one observes them in another representation. In particular, we will investigate $\mathds{P}$-space graphs.
{width="80mm"} {width="80mm"}
[**a.**]{}
First let us recall that in this representation each node corresponds to a PTN station, i.e. it has the same interpretation as in the $\mathds{L}$-space. However, the interpretation of a link differs from that in the $\mathds{L}$-space: now all station-nodes that belong to the same route are connected and thus each route enters the $\mathds{P}$-space network as a complete subgraph. This results in the main peculiarity of the interpretation of the behavior under attacks of these graphs. Consider as an example the $\mathds{P}$-space graph of Fig. \[fig1\][**c**]{} and compare it to the original PTN map, Fig. \[fig1\][**a**]{}. Whereas the removal of station node C in the map (Fig. \[fig1\][**a**]{}) disconnects the nodes B and D, the removal of the same node in the $\mathds{P}$-space (Fig. \[fig1\][**c**]{}) keeps nodes B and D connected, as far as they still belong to the same route. Therefore, the removal of nodes in $\mathds{P}$-space, performed either in a random way or according to certain lists, has a different interpretation in comparison to that occurring in the $\mathds{L}$-space. An interpretation of the removal of nodes in $\mathds{P}$-space is the following: if a node is removed, the corresponding stop of the route is canceled while the route otherwise keeps operating. If in the above example the station-node C is removed, route No 2 still keeps operating and station-node B can be reached from D, only without stopping at C (e.g. the bus takes a shortcut). In this way, as we will see below, the removal of nodes in $\mathds{P}$-space allows us to gain additional insight into the PTN structure.
{width="80mm"} {width="80mm"}
[**a.**]{}
2.5mm
City $c_{\rm s}$ $c_{\rm s}$ $c_{\rm s}$ $c_{\rm s}$ $c_{\rm s}$ $c_{\rm s}$
------------- ------------- ------------- ------------- ------- ------------- ----------- ------------- ----------- ------------- ------------------- ------------- ----
Berlin .155 $C_B$ .175 $C_C$ .215 $C_S$ .285 $C^{i}$ .290 $C^{i}_B$ .490 RV
Dallas .065 $C_B$ .075 $C_C$ .095 $C_S$ .115 $C$ .130 $C^{i}$ .490 RV
Düsseldorf .160 $C_B$ .185 $C_S$ .255 $C_C$ .295 $C^{i}$ .300 $k^{i}$ .495 RV
Hamburg .050 $C_C$ .065 $C_B$ .145 $C_G$ .170 $C$ .175 $C^{i}_C$ .490 RV
Hong Kong .285 $C_B$ .295 $C_S$ .335 $C_C$ .365 $C$ .380 $C^{i}$ .505 RV
Istanbul .060 $C_C$ .060 $C_B$ .060 $C^{i}_B$ .115 $C^{i}_C$ .175 $C$ .500 RV
London .155 $C_B$ .205 $C_C$ .305 $C_G$ .330 $C$ .350 $C^{i}$ .495 RV
Los Angeles .065 $C_B$ .095 $C_C$ .145 $C_S$ .145 $C^{i}_B$ .150 $C$ .480 RV
Moscow .175 $C_B$ .255 $C_C$ .285 $C_S$ .345 $C$ .395 $C^{i}$,$C^{i}_S$ .495 RV
Paris .115 $C_B$ .165 $C_S$ .215 $C_C$ .235 $C^{i}_B$ .240 $C$,$C^{i}$ .500 RV
Rome .135 $C_C$ .160 $C_B$ .225 $C_G$ .285 $C_S$ .305 $C$ .495 RV
Saõ Paolo .205 $C_B$,$C_C$ .240 $C_S$ .355 $C_G$ .365 $C$ .390 $C^{i}$ .500 RV
Sydney .075 $C_C$ .085 $C_B$ .105 $C_S$ .225 $C$ .240 $C^{i}$ .510 RV
Taipei .290 $C_B$ .320 $C_S$ .370 $C_C$ .430 $C_G$ .440 $k,C_S^{i}$ .495 RV
As in the case of the $\mathds{L}$-space representation, we study the resilience of the $\mathds{P}$-space PTN graphs to attacks performed following the sixteen different scenarios defined in Section \[III\]. In Fig. \[fig9\] we show the change of the size of the largest cluster $S$ ([**a**]{}) and the average inverse mean shortest path length $\langle\ell^{-1}\rangle$ ([**b**]{}) under random attacks (RV). If one compares this behavior with that observed for the RV scenario in $\mathds{L}$-space (see Fig. \[fig2\]) one sees, that all PTNs under consideration react in a much more homogeneous way. In $\mathds{L}$-space random attacks lead to changes of the largest connected component $S$ that range from an abrupt breakdown (Dallas) to a slow smooth decrease (Paris). In $\mathds{P}$-space one observes for the same scenario only a decrease of $S$ which corresponds to the number of removed nodes. No break-down of this cluster occurs in this scenario. The value of $S(c_{\rm s})$ defined by the condition (\[11\]) is given in the last column of Table \[tab3\]. It is worth to note, that the behavior of the mean inverse shortest path length $\langle
\ell^{-1} \rangle$ as function of the fraction $c$ of disabled nodes is also qualitatively different between the two RV scenarios in $\mathds{L}$- (Fig. \[fig2\][**b**]{}) and $\mathds{P}$- (Fig. \[fig9\][**b**]{}) spaces. In $\mathds{L}$-space $\langle \ell^{-1}
\rangle$ decreases in general faster than linearly indicating an increase of the path length between the nodes as well as partitioning of the network. In $\mathds{P}$-space $\langle
\ell^{-1} \rangle$ remains for a large part unperturbed as the nodes of the complete subgraph remain essentially connected and the shortest path lengths remain almost unchanged until only a small fraction of the network remains.
To further detail the situation, similar as in Section \[III\], we summarize in Table \[tab3\] the outcome of the five most harmful attack scenarios and compare those with the random attack scenario. As it follows from the Table and as is further supported by Fig. \[fig10\], the betweenness-targeted scenarios appear to be the most harmful. Following this observation let us investigate the role of the highest betweenness nodes: above all these are the nodes (and not the highest-$k$ hubs) that control the PTN behavior under attack. The $\mathds{P}$-space degrees of these high-betweenness nodes do not essentially differ from those of the hubs, therefore they cannot be easily distinguished from the other nodes during attacks according to highest-$k$ scenario. To support this assumption, let us recall that in the $\mathds{P}$-space representation each route enters the overall network as a complete subgraph, with all nodes interconnected. Removing nodes from a complete graph does not lead to any segmentation. The decrease of the normalized size of this graph will be given by the exact formula $S=1-c$ (which is - almost - reproduced by the RV scenario, c.f. Fig. \[fig9\][**a**]{}). Under such circumstances a special role is played by those nodes that join different complete graphs (different routes). The removal of such nodes will separate different complete routes and as a result may lead to network segmentation. Naturally, being between different complete subgraphs such nodes are characterized by high centrality indices, as observed above. Moreover, as far as their direct neighbors belong to different complete graphs, these neighbors are not connected between each other resulting in a lower value of the clustering coefficient $C$. From Table \[tab3\] one sees that attacks based on choosing nodes with low-$C$ values are very effective in $\mathds{P}$-space.
To conclude this section, we ask the question if a simple criterion can be found that allows to predict a priori the $\mathds{P}$-space PTN vulnerability. Namely, given the general PTN characteristics (see Table \[tab1\]) can one forecast resilience against attacks in $\mathds{P}$-space? The answer is given by the observation that the networks with low mean shortest path length $\langle
\ell_{\mathds{P}} \rangle$ are the best connected in $\mathds{P}$-space and hence may be expected to be less vulnerable. Indeed, on the one hand, for the above example of a complete graph (a single PTN route) $\langle \ell_{\mathds{P}} \rangle=1$ and it is extremely robust to $\mathds{P}$-space attacks. On the other hand, a high value of $\langle \ell_{\mathds{P}} \rangle$ indicates numerous intermediate nodes between different routes. As we have checked above, the targeted removal of such nodes leads to rapid network segmentation. In support of the above reasoning, in Fig. \[fig11\] we plot $c_{\rm s}$ as function of $\langle \ell_{\mathds{P}}
\rangle$ for attacks based on the highest betweenness centrality scenario. There, within the expected scatter of data one observes a clear evidence of the decrease of $c_{s}$ with $\langle
\ell_{\mathds{P}} \rangle$, i.e. networks with higher mean path length break down at smaller values of $c$ and are thus more vulnerable.
It is worth to note here, that in $\mathds{P}$-space it is only the RV attack that has very similar impact on all PTNs (see Fig. \[fig9\]). As we have just observed, similar to the $\mathds{L}$-space also in $\mathds{P}$-space the PTNs manifest different level of robustness against attacks targeted on the most important nodes. However, the order of vulnerability changes if one compares the outcome of the $\mathds{L}$-space and $\mathds{P}$-space attacks. This means that PTNs that were vulnerable in the $\mathds{L}$-space may appear to be robust against attacks in $\mathds{P}$-space. From Table \[tab3\] we see that the PTNs that are most stable against highest $C_B$-targeted attacks in $\mathds{P}$-space are the PTNs of Hong Kong, Saõ Paolo, and Moscow, with $c_{\rm s}= 0.285$, $0.205$, and $0.175$, correspondingly. When attacked in $\mathds{L}$-space, the PTN of Moscow keeps its robustness: $c_{\rm s}= 0.07$ during $C_B$-targeted attack, which is one of highest $c_{\rm s}$ values for the $\mathds{L}$-space, see Table \[tab2\]. This is however not the case for the PTNs of Hong Kong and Saõ Paolo. In $\mathds{L}$-space, these belong to the most vulnerable PTNs.
![$\mathds{P}$-space. Correlations between the mean shortest path length $\langle \ell_{\mathds{P}} \rangle$ and segmentation concentration $c_{s}$ in the highest betweenness centrality scenario. The line serves as a guide to observe the tendency of $c_{s}$ to decrease with increasing $\langle \ell_{\mathds{P}} \rangle$. \[fig11\]](cs_lp_mod.eps){width="80mm"}
Conclusions and outlook {#V}
=======================
In this paper, we have studied the behavior of city public transportation networks (PTNs) under attacks. In our analysis we have examined PTNs of fourteen major cities of the world. The principal motivation behind this study was to observe the behavior under attack of a sample of networks that were constructed for the same purpose, to compare these with available analytical results for percolation of complex networks, and possibly to derive some conclusions about correlations between PTN characteristics calculated a priory and the resilience to attacks. Furthermore, the resilience behavior of a network against different attack scenarios gives additional insight into the network architecture, discovering structures on different scales. This approach has been termed the ’tomography’ of a network [@Xulvi-Brunet03].
In our study we have also attempted to compare our results with the predictions of percolation theory on networks. Due to the sizes of these systems which are far from the thermodynamic limit and the rather small sample of networks no quantitative comparison appeared possible. However, qualitative predictions about the location of segmentation thresholds and thus the vulnerability could be verified. Although our study was not primarily motivated by applications, some of the results and methods developed within this study may be useful for planning and risk assessment of PTNs. Our analysis has identified PTN structures which are especially vulnerable and others, which are particulary resilient against attacks. Further investigation of other relevant network properties may reveal mechanisms behind this structural resilience [@Holovatch09]. Furthermore we note that the methods developed here also allow to identify minimal strategies to obstruct the operation of the PTN of a city e.g. for the purposes of industrial action and possibly achieve a successful end of a social conflict.
To analyze PTN resilience we have applied different attack scenarios, that range from a random failure to a targeted destruction, when the most influential network nodes were removed according to their operating characteristics. To choose the most influential nodes, we have used different graph theoretical indicators and determined in such a way the most effective attack scenarios. By our paper we show that even within a sample of networks that were created for the same purpose one observes essential diversity with respect to their behavior under attacks of various scenarios. Results of our analysis show that PTNs demonstrate rich variety of behavior under attacks, that range from smooth decay to abrupt change.
As shown by our study, the impact of attacks may be measured by different quantities. As a criterion that is well defined and easily reproducible we choose to define the segmentation concentration $c_{\rm s}$ to correspond to the situation where the largest remaining cluster contains one half of the original nodes of the network. Let us note as well, that definitely not all of the PTNs analyzed demonstrated scale-free behavior in $\mathds{P}$-space (and even less in $\mathds{L}$-space). Nevertheless, in spite of the diversity of behavior we clearly see common tendencies in their reaction to attacks. In particular, this enabled us to propose criteria that allow an a priori estimate of PTN robustness. In $\mathds{L}$-space resilience is indicated by a high value of the Molloy-Reed parameter $\kappa$, Eqs. (\[1\]), (\[12\]) or by a small value of the exponent $\gamma$, if a power law is observed for the PTN node degree distribution, in $\mathds{P}$-space high resilience is indicated by a small mean shortest path length $\langle \ell_{\mathds{P}} \rangle$.
One of possible continuations of our study will be the analysis of PTN resilience in other graph representations, than those that were described above.
Acknowledgements {#acknowledgements .unnumbered}
================
CvF wishes to thank Reinhard Folk for his hospitality at the Institute of Theoretical Physics, JKU Linz. This work was supported by the Austrian Fonds zur Förderung der wissenschaftlichen Forschung under Project No. P19583-N20 and by the cooperation programme ’Dnipro’ between the Ministry of Foreign Affairs of France and the Ministry of Education and Science of Ukraine.
[50]{}
R. Albert and A.-L. Barabási, Rev. Mod. Phys. [**74**]{}, 47 (2002); S. N. Dorogovtsev and J. F. F. Mendes, Adv. Phys. [**51**]{}, 1079 (2002); M. E. J. Newman, SIAM Review [**45**]{}, 167 (2003); S. N. Dorogovtsev and S. N. Mendes, [*Evolution of Networks*]{} (Oxford University Press, Oxford, 2003).
R. Albert, H. Jeong, and A.-L. Barabási, Nature (London) [**406**]{}, 378 (2000).
Y. Tu, Nature (London) [**406**]{}, 353 (2000).
H. Jeong, B. Tombor, R. Albert, Z. N. Oltvai, and A.-L. Barabási, Nature (London) [**407**]{}, 651 (2000).
R. V. Solé and J. M. Montoya: Proc. R. Soc. Lond. B [**268**]{}, 2039 (2001).
H. Jeong, S. P. Mason, A.-L. Barabási, and Z. N. Oltvai, Nature (London) [**411**]{}, 41 (2001).
R. Cohen, K. Erez, D. ben-Avraham, and S. Havlin, Phys. Rev. Lett. [**85**]{}, 4626 (2000).
D. S. Callaway, M. E. J. Newman, S. H. Strogatz, and D. J. Watts, Phys. Rev. Lett. [**85**]{}, 5468 (2000).
J. W. Essam, Rep. Prog. Phys. [**43**]{}, 833 (1980); D. Stauffer and A. Aharony, [*Introduction to Percolation Theory*]{}, (Taylor & Francis, London, 1991).
P. Erdös and A. Rényi, Publ. Math. (Debrecen) [**6**]{}, 290 (1959); Publ. Math. Inst. Hung. Acad. Sci. [**5**]{}, 17 (1960); Bull. Inst. Int. Stat. Tokyo [**38**]{}, 343 (1961). B. Bollobás, [*Random Graphs*]{} (Academic Press, London, 1985).
M. Molloy and B. A. Reed, Random Struct. Algorithms [**6(2/3)**]{}, 161 (1995); M. Molloy and B. Reed, Combinatorics, Probability and Computing [**7**]{}, 295 (1998).
R. Cohen, K. Erez, D. ben-Avraham, and S. Havlin, Phys. Rev. Lett. [**86**]{}, 3682 (2001).
T. Kalisky and R. Cohen, Phys. Rev. E [**73**]{}, 035101(R) (2006).
R. Xulvi-Brunet, W. Pietsch, and I. M. Sokolov, Phys. Rev. E [**68**]{}, 036119 (2003).
P. Holme, B. J. Kim, C. N. Yoon, and S. K. Han, Phys. Rev. E [**65**]{}, 056109 (2002).
Note however that an early paper on metabolism networks [@Jeong01] checked 43 different organisms.
A.-L. Barabási and R. Albert, Science [**286**]{}, 509 (1999); A.-L. Barabási, R. Albert, and H. Jeong, Physica A [**272**]{}, 173 (1999).
A. Broder, R. Kumar, F. Maghoul, P. Raghavan, S. Rajagopalan, R. Stata, A. Tomkins, and J. Wiener, Comput. Netw. [**33**]{}, 309 (2000).
M. Girvan and M. E. J. Newman, Proc. Natl. Acad. Sci. USA [**99**]{}, 7821 (2002).
R. Guimera and L. A. N. Amaral, Eur. Phys. J. B [**38**]{}, 381 (2004).
R. Guimera, S. Mossa, A. Turtschi and L.A.N. Amaral, Proc. Nat. Acad. Sci. USA [**102**]{}, 7794 (2005).
C. von Ferber, Yu. Holovatch, and V. Palchykov, Condens. Matter Phys. [**8**]{}, 225 (2005); C. von Ferber, T. Holovatch, Yu. Holovatch, and V. Palchykov, Physica A [**380**]{}, 585 (2007).
C. von Ferber, T. Holovatch, Yu. Holovatch, and V. Palchykov, Eur. Phys. Jour. B [**68**]{}, 261 (2009).
C. von Ferber, T. Holovatch, and Yu. Holovatch, In: C. Appert-Rolland, F. Chevoir, P. Gondret, S. Lassarre, J.-P. Lebacque, M. Schreckenberg (Eds.) [*Traffic and Granular Flow ’07*]{}. Springer (2009); arXiv:0709.3206.
For links see [*http://www.apta.com*]{}. The database is described in details in Ref. [@Ferber09a].
M. Marchiori and V. Latora, Physica A [**285**]{}, 539 (2000); V. Latora and M. Marchiori, Phys. Rev. Lett. [**87**]{}, 198701 (2001); V. Latora and M. Marchiori, Physica A [**314**]{}, 109 (2002); K. A. Seaton and L. M. Hackett, Physica A [**339**]{}, 635 (2004); J. Sienkiewicz and J. A. Holyst, Phys. Rev. E [**72**]{}, 046127 (2005); J. Sienkiewicz and J. A. Holyst, Acta Phys. Polonica B [**36**]{}, 1771 (2005); P. Angeloudis and D. Fisk, Physica A [**367**]{}, 553 (2006); P.-P. Zhang, K. Chen, Y. He, T. Zhou, B.-B. Su, Y. Jin, H. Chang, Y.-P. Zhou, L.-C. Sun, B.-H. Wang, and D.-R. He, Physica A [**360**]{}, 599 (2006); X. Xu, J. Hu, F. Liu, and L. Liu, Physica A [**374**]{}, 441 (2007); H. Chang, B.-B. Su, Y.-P. Zhou, and D.-R. He, Physica A [**383**]{}, 687 (2007).
P. Sen, S. Dasgupta, A. Chatterjee, P. A. Sreeram, G. Mukherjee, and S. S. Manna, Phys. Rev. E [**67**]{}, 036106 (2003).
In the following we will only consider the removal of nodes. This is along the lines of the site percolation phenomenon.
U. Brandes, J. Math. Sociology, [**25**]{}, 163 (2001).
R. Cohen, S. Havlin, and D. ben-Avraham, Phys. Rev. Lett. [**91**]{}, 247901 (2003).
Another obvious dfference between the relative size of the largest cluster $S$ of a network considered here and a spanning cluster size on a regular lattice is that the former can be zero only in the infinite network limit, whereas the latter can be zero even for a finite size system.
T. Holovatch, B. Berche, C. von Ferber, unpublished.
[^1]: [email protected]
[^2]: [email protected]
[^3]: [email protected]
[^4]: [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'There is an ever growing need to ensure the quality of food and assess specific quality parameters in all the links of the food chain, ranging from processing, distribution and retail to preparing food. Various imaging and sensing technologies, including X-ray imaging, ultrasound, and near infrared reflectance spectroscopy have been applied to the problem. Cost and other constraints restrict the application of some of these technologies. In this study we test a novel Multiplexing Electric Field Sensor (MEFS), an approach that allows for a completely non-invasive and non-destructive testing approach. Our experiments demonstrate the reliable detection of certain foreign objects and provide evidence that this sensor technology has the capability of measuring fat content in minced meat. Given the fact that this technology can already be deployed at very low cost, low maintenance and in various different form factors, we conclude that this type of MEFS is an extremely promising technology for addressing specific food quality issues.'
author:
- 'Anne E. Rittscher'
- 'Aslam Sulaimalebbe$^*$'
- 'Yves Capdeboscq [^1]'
- 'Jens Rittscher$^*$'
bibliography:
- 'food-analysis-lit.bib'
title: Foreign Object Detection and Quantification of Fat Content Using A Novel Multiplexing Electric Field Sensor
---
Acknowledgements {#acknowledgements .unnumbered}
================
The sensor was provided by Zedsen Ltd. All measurements were generated by AER and JR outside the Zedsen laboratories. The study was sponsored by Zedsen.
[^1]: Corresponding Authors: [email protected], [email protected], [email protected]
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In this paper we study the problem of estimating the alpha-, beta- and phi-mixing coefficients between two random variables, that can either assume values in a finite set or the set of real numbers. In either case, explicit closed-form formulas for the beta-mixing coefficient are already known. Therefore for random variables assuming values in a finite set, our contributions are two-fold: (i) In the case of the alpha-mixing coefficient, we show that determining whether or not it exceeds a prespecified threshold is NP-complete, and provide efficiently computable upper and lower bounds. (ii) We derive an exact closed-form formula for the phi-mixing coefficient. Next, we prove analogs of the data-processing inequality from information theory for each of the three kinds of mixing coefficients. Then we move on to real-valued random variables, and show that by using percentile binning and allowing the number of bins to increase more slowly than the number of samples, we can generate empirical estimates that are consistent, i.e., converge to the true values as the number of samples approaches infinity.'
author:
- 'Mehmet Eren Ahsen and M. Vidyasagar [^1]'
title: |
Mixing Coefficients Between\
Discrete and Real Random Variables:\
Computation and Properties
---
Introduction {#sec:1}
============
The notion of independence of random variables is central to probability theory. In [@Kolmogorov33 p. 8], Kolmogorov says:
> “Indeed, as we have already seen, the theory of probability can be regarded from the mathematical point of view as a special application of the general theory of additive set functions.
and
> “Historically, the independence of experiments and random variables represents the very mathematical concept that has given the theory of probability its peculiar stamp.”
In effect, Kolmogorov is saying that, if the notion of independence is removed, then probability theory reduces to just measure theory.
Independence is a binary concept: Either two random variables are independent, or they are not. It is therefore worthwhile to replace the concept of independence with a more nuanced measure that [*quantifies*]{} the extent to which given random variables are dependent. In the case of stationary stochastic processes, there are various notions of ‘mixing’, corresponding to long term asymptotic independence. These notions can be readily adapted to define various mixing coefficients between two random variables. In this setting, the mixing rate of a stochastic process can be interpreted as the mixing coefficient between the semi-infinite ‘past’ and ‘future’ variables. Several such definitions are presented in [@Doukhan94 p. 3], out of which three are of interest to us, namely the $\al$-, $\beta$- and $\phi$-mixing coefficients. While the definitions themselves are well-known, there is very little work on actually [*computing*]{} (or at least estimating) these mixing coefficients in a given situation. The $\beta$-mixing coefficient is easy to compute but this is not the case for the $\al$- and the $\phi$-mixing coefficients.
Against this background, the present paper makes the following specific contributions: For random variables that assume values in a finite set:
In the case of the $\al$-mixing coefficient, it is shown that determining whether or not it exceeds a prespecified threshold is NP-complete, and efficiently computable upper and lower bounds are derived.
An efficiently computable exact formula is derived for the $\phi$-mixing coefficient.
We study the case of three random variables $X,Y,Z$, where $X,Z$ are conditionally independent given $Y$, or equivalently, $X \ap Y \ap Z$ is a short Markov chain. In this case a well-known inequality from information theory [@CT06 p. 34], usually referred to as the ‘data processing inequality (DPI)’, states that \[eq:11\] I(X,Z) { I(X,Y) , I(Y,Z) } , where $I(\cdot,\cdot)$ denotes the mutual information. We state and prove analogs of the DPI for each of the $\al$-, $\beta$- and $\phi$-mixing coefficients. Next we turn to real-valued random variables.
Suppose $X,Y$ are real-valued random variables whose joint distribution has a density with respect to the Lebesgue measure, and that $\{ (x_1 , y_1 ) , \ldots, ( x_l , y_l ) \}$ are independent samples of $(X,Y)$. If we compute the empirical joint distribution of $(X,Y)$ from these samples, then the Glivenko-Cantelli Lemma states that the empirical joint distribution converges with probability one to the true joint distribution; in other words, the empirical distribution gives a [*consistent*]{} estimate. However, it is shown here that if the empirical distribution is used to estimate the mixing coefficients, then with probability one both the estimated $\beta$-mixing coefficient and the estimated $\phi$-mixing coefficient approach one as $\lai$, irrespective of what the true value might be. Thus a quantity derived from a consistent estimator need not itself be consistent.
On the other hand, if we bin the $l$ samples into $k_l$ bins and choose $k_l$ in such a way that $k_l \ap \infty$ and $k_l/l \ap 0$ as $\lai$, and a few technical conditions are satisfied, then the empirically estimated $\al$-, $\beta$- and $\phi$-mixing coefficients converge to their true values as $\lai$, with probability one.
The notion of a mixing process and various definitions of mixing coefficients originated in an attempt to establish the law of large numbers for stationary stochastic processes that are not i.i.d. The problem of determining (or at least bounding) the mixing coefficients of random variables and stochastic processes arises in various contexts, including system identification and statistical learning. Traditional theories of system identification are based on the assumption that the input sequence to the unknown system is i.i.d. However, it became clear over time that much of the theory continues to hold even if the input sequence is not i.i.d., but is mixing in an appropriate sense. See [@Weyer00; @VK07] as just two examples of such an approach. Similarly, the standard formulation of PAC (probably approximately correct) learning in statistical learning theory is based on the assumption that the inputs are i.i.d. See [@MV97; @Vapnik98] for example. However, subsequently PAC learning theory has been extended to the case where the learning inputs are not i.i.d., but are mixing instead; see for example the book [@MV03] and the references therein, as well as [@Meir00]. In adapting results in system identification or statistical learning theory from the i.i.d. case to the case of mixing processes, it becomes necessary to obtain at least upper bounds for the mixing coefficients, if not exact values. The results presented here have some relevance to this problem, as do other recent results such as [@MSS11]. We shall return to this topic in the concluding remarks.
Proving that various mixing coefficients satisfy analogs of the data processing inequality (DPI) is not just of academic interest. Recent work on reverse-engineering genome-wide interaction networks from gene expression data is based on first constructing a complete graph where each node corresponds to a gene, and then using the DPI to “prune” the graph. Among the first algorithms to use this approach is ARACNE [@ARACNE], which is based on using mutual information as a measure of interaction between genes. However, because mutual information is a symmetric quantity, the resulting graphs are [*undirected*]{}, which is quite contrary to biological realism, because in reality the interactions between genes are not symmetric. This led the authors to explore whether the $\phi$-mixing coefficient, which is asymmetric, can be used as a measure of the interaction between two genes. Once it is established that the $\phi$-mixing coefficient satisfies an analog of the DPI (which is one of the principal results of this paper), it is possible to develop a method for constructing [*directed graphs*]{} that represent whole genome regulatory networks. However, this by itself is not sufficient. If there are $n$ genes in the study, this approach requires the computation of $n^2$ $\phi$-mixing coefficients. So for a typical genome-wide study involving $20,000$ genes, it becomes necessary to compute $400$ million $\phi$-mixing coefficients. Hence it is mandatory to have a method for the efficient computation of the $\phi$-mixing coefficient. Such a method is also provided in the present paper. Please see [@MV-EJC11; @GIN; @MV-Brief] for a discussion of how the methods presented here can be applied to reverse engineering gene regulatory networks.
Definitions of Mixing Coefficients {#sec:2}
==================================
Definitions of the $\al$-, $\beta$- and $\phi$-mixing coefficients of a stationary stochastic process can be found, among other places, in [@MV03 pp. 34-35]. The $\al$-mixing coefficient was introduced by Rosenblatt [@Rosenblatt56]. According to Doukhan [@Doukhan94 p. 5], Kolmogorov introduced the $\beta$-mixing coefficient, but it appeared in print for the first time in a paper published by some other authors. The $\phi$-mixing coefficient was introduced by Ibragimov [@Ibragimov62].
Essentially, all notions of mixing processes try to quantify the idea that, in a stationary stochastic process of the form $\{ X_t \}_{t = -\infty}^\infty$, the random variables $X_t$ and $X_\t$ become more and more independent as $| t - \t |$ approaches infinity, in other words, there is an asymptotic long-term near-independence. However, these very general notions can be simplified and readily adapted to define mixing coefficients between a pair of random variables $X$ and $Y$. This is how they are defined in [@Doukhan94]. Note that, strictly speaking, mixing is a property not of the random variables $X$ and $Y$, but rather of the $\s$-algebras generated by $X$ and $Y$. Note also that, if $\{ X_t \}_{t = -\infty}^\infty$ is a stationary stochastic process, then the $k$-th ($\al$, $\beta$ or $\phi$) mixing coefficient of the stochastic process is just the corresponding mixing coefficient as defined in [@Doukhan94] between the semi-infinite past $X_{-\infty}^0 := (X_t, t \leq 0)$ and the semi-infinite future $X_k^\infty := (X_t , t \geq k)$. Though mixing coefficients can be defined for arbitrary random variables, in the interests of avoiding a lot of technicalities we restrict our attention in this paper to just two practically important cases: real-valued and finite-valued random variables. We first define mixing coefficients between real-valued random variables, and then between finite-valued random variables.
\[def:mix-real\] Suppose $X$ and $Y$ are real-valued random variables. Let $\B$ denote the Borel $\s$-algebra of subsets of $\R$. Then we define (X,Y) & := & \_[S , T ]{} | { X S & Y T }\
& - & { X S } { Y T } | .\[eq:21\] (X|Y) & := & \_[S , T ]{} | { X S | Y T } - { X S } |\
& = & \_[S , T ]{} | - { X S } | . \[eq:22\]
In applying the above definition, in case $\Pr \{ Y \in T \} = 0$, we use the standard convention that { X S | Y T } = { X S } . Note that the $\al$-mixing coefficient is symmetric: $\al(X,Y) = \al(Y,X)$. However, in general $\phi(X|Y) \neq \phi(Y|X)$.
The third coefficient, called the $\beta$-mixing coefficient, has a somewhat more elaborate definition, at least in the general case. Let $\th$ denote the probability measure of the joint random variable $(X,Y)$, and let $\mu,\nu$ denote the marginal measures of $X$ and $Y$ respectively. Note that $\th$ is a measure on $\R^2$ while $\mu,\nu$ are measures on $\R$. If $X$ and $Y$ were independent, then $\th$ would equal $\mu \times
\nu$, the product measure. With this in mind, we define \[eq:23\] (X,Y) = (, ) , where $\r$ denotes the total variation distance between two measures. That is, if $\th,\eta$ are probability measures on a common measure space $(\OM,\SI)$, then (,) := \_[S ]{} | (S) - (S) | . The $\beta$-mixing coefficient is also symmetric.
Next we deal with finite-valued random variables, and for this purpose we introduce some notation that is used throughout the remainder of the paper. The most important notational change is that, since probability distributions on finite sets can be represented by vectors, we use bold-face Greek letters to denote them, whereas we use normal Greek letters to denote measures on $\R$ or $\R^2$. For each integer $n$, let $\Sm_n$ denote the $n$-dimensional simplex. Thus \_n := { v\^n : v\_i 0 i , \_[i=1]{}\^n v\_i = 1 } . If $\Abb = \{ a_1 , \ldots , a_n \}$ and $\bmu \in \Sm_n$, then $\bmu$ defines a measure $P_\bmu$ on the set $\Abb$ according to P\_(S) = \_[i=1]{}\^n \_i I\_S(a\_i) , where $I_S(\cdot)$ denotes the indicator function of $S$. Suppose $\bmu,\bnu \in \Sm_n$ are probability distributions on a set $\Abb$ of cardinality $n$. Then the [**total variation distance**]{} between $\bmu$ and $\bnu$ is defined as \[eq:24\] (,) := \_[S ]{} | P\_(S) - P\_(S) | . It is easy to give several equivalent closed-form formulas for the total variation distance. (,) = 0.5 - \_1 = \_[i=1]{}\^n ( \_i - \_i )\_+ = - \_[i=1]{}\^n ( \_i - \_i )\_- , where as usual $(\cdot)_+$ and $(\cdot)_-$ denote the nonnegative and the nonpositive parts of a number: (x)\_+ = { x,0 } , (x)\_- = { x, 0 } .
Now suppose $\Abb,\Bbb$ denote sets of cardinality $n,m$ respectively, and that $\bmu \in \Sm_n , \bnu \in \Sm_m$. Then the distribution $\bpsi \in \Sm_{nm}$ defined by $\psi_{ij} = \mu_i \nu_j$ is called the [**product distribution**]{} on $\Abb \times \Bbb$, and is denoted by $\bmu \times \bnu$. In the other direction, if $\bth \in \Sm_{nm}$ is a distribution on $\Abb \times \Bbb$, then $\bth_\Abb \in \Sm_n , \bth_\Bbb \in \Sm_m$ defined respectively by (\_)\_i := \_[j=1]{}\^m \_[ij]{} , (\_)\_j := \_[i=1]{}\^n \_[ij]{} are called the [**marginal distributions**]{} of $\bth$ on $\Abb$ and $\Bbb$ respectively.
The earlier definitions of mixing coefficients become quite explicit in the case where $X,Y$ are random variables assuming values in the finite sets $\Abb,\Bbb$ of cardinalities $n,m$ respectively. In this case it does not matter whether the ranges of $X,Y$ are finite subsets of $\R$ or some abstract finite sets. Definition \[def:mix-real\] can now be restated in this context. Note that, since $\Abb,\Bbb$ are finite sets, the associated $\s$-algebras are just the power sets, that is, the collection of all subsets.
\[def:mix-finite\] With the above notation, we define \[eq:3a\] (X,Y) := \_[S , T ]{} | P\_( S T ) - P\_(S) P\_(T) | , \[eq:3b\] (X,Y) := ( , ) , \[eq:3c\] (X|Y) := \_[S , T ]{} | - P\_(S) | .
Whether $X,Y$ are real-valued or finite-valued random variables, the mixing coefficients satisfy the following inequalities; see [@Doukhan94 p. 4]: \[eq:25\] (X,Y) , (X,Y) , (X,Y) , 0 & & 2 (X,Y) (X,Y) { (X|Y) , (Y|X) }\
& & { (X|Y) , (Y|X) } 1 . Also, the following statements are equivalent:
$X$ and $Y$ are independent random variables.
$\al(X,Y) = 0$.
$\beta(X,Y) = 0$.
$\phi(X|Y) = 0$.
$\phi(Y|X) = 0$.
Computation of Mixing Coefficients for Finite-Valued Random Variables {#sec:3}
=====================================================================
From the definitions, it is clear that $\beta(X,Y)$ can be readily computed in closed form. As before, let us define $\bpsi = \bmu \times \bnu$ to be the product distribution of the two marginals, and define \_[ij]{} := \_[ij]{} - \_[ij]{} , := \[ \_[ij]{} \] \^[n m]{} . Then it readily follows from (\[eq:21\]) that (X,Y) := ( , ) & = & 0.5 \_[i=1]{}\^n \_[j=1]{}\^m | \_[ij]{} |\
& = & \_[i=1]{}\^n \_[j=1]{}\^m ( \_[ij]{} )\_+\
& = & - \_[i=1]{}\^n \_[j=1]{}\^m ( \_[ij]{} )\_- . In addition, there is a very useful upper bound on the $\beta$-mixing coefficient in terms of the so-called “Pinsker’s inequality”, though it may be appropriate to credit this inequality to Csiszár; see [@CK81] or [@CT06]. This inequality states that, for any two probability distributions $\bth$ and $\bphi$ on a common set, ( , ) , where $D(\cdot,\cdot)$ is the Kullback-Leibler divergence. Now apply this inequality with $\bphi = \bmu \times \bnu$. This leads to ( , ) , However, $\r( \bth , \bmu \times \bnu) = \beta(X,Y)$ whereas $ D( \bth \nm \bmu \times \bnu ) = I(X,Y)$, the mutual information between $X$ and $Y$. Therefore (X,Y) . On the other hand, computing $\al(X,Y)$ or $\phi(X|Y)$ directly from Definition \[def:mix-finite\] would require $2^{n+m}$ computations, since $S,T$ must be allowed to vary over all subsets of $\Abb,\Bbb$ respectively. Therefore the question arises as to whether this is an artefact of the definition, or an inherent barrier to efficient computation. In the present section, the following results are established:
As stated in (\[eq:25\]), the quantity $\al(X,Y)$ always lies in the interval $[0,0.25]$. It is shown that the problem of determining whether $\al(X,Y) = 0.25$ for a given pair of random variables $X,Y$ is NP-complete. More generally, given any number $\e \in (0,0.25]$, determining whether $\al(X,Y) \geq \e$ is NP-complete.
Some efficiently computable upper and lower bounds are derived for $\al(X,Y)$. These bounds become germane in view of the above complexity result.
An exact and efficiently computable formula is derived for $\phi(X,Y)$.
In proceeding further, the first step is to get rid of the absolute value signs in the definitions of the $\al$- and $\phi$-mixing coefficients.
\[thm:33\] It is the case that \[eq:34\] (X,Y) = \_[S , T ]{} , \[eq:35\] (X|Y) = \_[S , T ]{} .
[**Proof:**]{} Define \_:= { P\_( S T ) - P\_(S) P\_(T) , S , T } . Then ${\cal R}_\al$ is a finite subset of the real line consisting of at most $2^{n+m}$ elements. Now it is claimed that the set ${\cal R}_\al$ is symmetric; that is, $x \in {\cal R}_\al$ implies that $-x \in {\cal R}_\al$. If this claim can be established, then (\[eq:34\]) follows readily. So suppose $x \in {\cal R}_\al$, and choose $S \seq \Abb , T \seq \Bbb$ such that P\_( S T ) - P\_(S) P\_(T) = x . Let $S^c$ denote the complement of $S$ in $\Abb$. Then, using the facts that P\_(S\^c) = 1 - P\_(S) , P\_( S\^c T) & = & P\_( T) - P\_( S T )\
& = & P\_(T) - P\_( S T ) , it is easy to verify that P\_(S\^c T) - P\_( S\^c) P\_(T) = -x . So ${\cal R}_\al$ is symmetric and (\[eq:34\]) follows. By analogous reasoning, the set \_:= { - P\_(S) : S , T } is also symmetric, which establishes (\[eq:35\]). $\halmos$
NP-Completeness of Estimating the Alpha-Mixing Coefficient {#ssec:31}
----------------------------------------------------------
We begin by revisiting the definition of $\al(X,Y)$, and determine the conditions under which it can attain its theoretical maximum value of $0.25$.
\[thm:3a\] Suppose $X,Y$ are random variables assuming values in finite sets $\Abb,\Bbb$ respectively, with marginal distributions $\bmu,\bnu$ respectively, and joint distribution $\bth$. Then $\al(X,Y) \leq 0.25$. Moreover, $\al(X,Y) = 0.25$ if and only if there exist subsets $S \seq \Abb,T \seq \Bbb$ such that $P_\bmu(S) = 0.5$, $P_\bnu(T) = 0.5$, and $P_\bth(S \times T) = 0$.
[**Proof:**]{} It is easy to see that the following relationship, which is the mirror image of (\[eq:34\]), is true: \[eq:3d\] (X,Y) = - \_[S , T ]{} . Indeed, as shown in the proof of Theorem \[thm:33\], if $S,T$ achieve the maximum in (\[eq:34\]), then $S^c,T$ achieve the minimum in (\[eq:3d\]), and vice versa. Given sets $S \seq \Abb, T \seq \Bbb$, define a := P\_(S T) , b := P\_(S T\^c) , c := P\_(S\^c T) , d := P\_(S\^c T\^c) . Then it is evident that $a + b + c + d = 1$. Moreover, P\_(S) = a + b , P\_(T) = a + c . Therefore P\_( S T ) - P\_(S) P\_(T) & = & a - (a+b) (a+c)\
& = & -a\^2 + a ( 1 - b - c ) - bc\
& =: & f(a) . Let us think of the above quantity as a function of $a$ with $b,c$ fixed. This amounts to fixing the measures of the sets $S,T$ while adjusting the joint distribution to change $P_\bth(S \times T)$. Then $f(0) = - bc$, and $f'(0) = 1 - b - c \geq 0$. So $f(a)$ is nondecreasing at $a = 0$. The maximum permissible value of $a$ is $1 - b - c$ (amounting to setting $d = 0$), and $f(1 - b - c)$ again equals $-bc$. Simple high school algebra shows that $f(a)$ achieves a maximum at $a^* = (1 - b - c)/2$, and then begins to decrease. Therefore it follows that $f(a) \geq - bc$. Now $b,c$ satisfy $b+c \leq 1$, whence it is immediate that (X,Y) & & - \_[b,c]{} - bc b + c 1\
& = & \_[b,c]{} bc b + c 1\
& = & 0.25 . Moreover, $\al(X,Y) = 0.25$ if only if the choice $b = c = 0.5$ (which in turn implies that $a = d = 0$) is compatible with the given joint distribution. Recalling what these symbols represent shows that (i) $\al(X,Y) \leq 0.25$ always, and (ii) $\al(X,Y) = 0.25$ if and only if there exist subsets $S \seq \Abb,T \seq \Bbb$ such that $P_\bth(S \times T) = 0$, $P_\bmu(S) = P_\bnu(T) = 0.5$. $\halmos$
The next step is to map a problem that is known to be NP-complete into the problem of checking whether or not $\al(X,Y) = 0.25$. Our choice is the so-called “normalized partition” problem, which is a variant of the “partition problem”, which can be found in [@Garey-Johnson79 p. 47], among other places. We begin by stating the partition problem.
[**Problem Partition:**]{}\
[**Instance:**]{} A positive integer $m$, and a set of positive integers $a_1 , \ldots , a_m$.\
[**Question:**]{} Does there exist a subset $I \seq \M := \{ 1 , \ldots , m \}$ such that \_[i I]{} a\_i = \_[j I]{} a\_j ? This problem is known to be NP-Complete; see [@Garey-Johnson79 p. 47]. For our purposes we modify the problem as follows:
[**Problem Normalized Partition:**]{}\
[**Instance:**]{} A positive integer $m$, and a set of positive rational numbers $a_1 , \ldots , a_m$ such that $\sum_{i=1}^m a_i = 1$.\
[**Question:**]{} Does there exist a subset $I \seq \M := \{ 1 , \ldots , m \}$ such that \_[i I]{} a\_i = \_[j I]{} a\_j ? It is clear that this problem is equivalent to the partition problem, and is therefore NP-complete.
\[thm:3b\] The following problem is NP-complete:
[**Problem:**]{}\
[**Instance:**]{} Positive integers $n,m$ and a set of nonnegative rational numbers $\th_{ij}, i = 1 , \ldots , n , j = 1 , \ldots , m$ such that $\sum_{i,j} \th_{ij} = 1$.\
[**Question:**]{} Let $(X,Y)$ be random variables assuming values in $\N := \{ 1 , \ldots , n \}$, $\M := \{ 1 , \ldots , m \}$ respectively with the joint distribution $\Pr \{ X = i \& Y = j \} =\th_{ij}$. Is $\al(X,Y) = 0.25$?
[**Proof:**]{} By Theorem \[thm:3a\], we know that $\al(X,Y) = 0.25$ if and only if there exist subsets $S \seq \N, T \seq \M$, such that $P_\bmu(S) = 0.5$, $P_\bnu(T) = 0.5$ and $P_\bth(S \times T) = 0$, where $P_\bmu,P_\bnu$ denote the marginals of $P_\bth$. Hence, given a candidate solution in terms of sets $S,T$, all one has to do is to verify the above three relationships, which can be done in polynomial time. So the problem is in NP.
To show that it is NP-complete, we map the normalized partition problem into it. Given positive rational numbers $a_1 , \ldots , a_m$ such that $\sum_{i=1}^m a_i = 1$, define $n = m$ and $\th_{ij} = a_i \d_{ij}$, where $\d_{ij}$ is the Kronecker delta. Thus, under this joint distribution, $n = m$, both $X$ and $Y$ have the vector $\a = [ a_1 , \ldots , a_m]$ as their marginal distributions, and $\Pr \{ X = Y \} = 1$. Given subsets $S , T \seq \M$, it is easy to verify that $P_\bth (S \times T) = P_\a(S \cap T)$. (Note that $P_\bth$ is a measure on $\M \times \M$ while $P_\a$ is a measure on $\M$.) Therefore $\al(X,Y) = 0.25$ if and only if there exist subsets $S,T \seq
\M$ such that $P_\a(S) = 0.5$, $P_\a(T) = 0.5$, and $P_\a(S \cap T) = 0$. These conditions imply that $S,T$ form a partition of $\M$, and that $I = S, \M \setminus I = T$ is a solution of the normalized partition problem. Hence this problem is NP-complete. $\halmos$
\[corr:3a\] The following problem is NP-complete:
[**Problem:**]{}\
[**Instance:**]{} Positive integers $n,m$, a set of nonnegative rational integers $\th_{ij}, i = 1 , \ldots , n , j = 1 , \ldots , m$ such that $\sum_{i,j} \th_{ij} = 1$, and a rational number $\e \in (0,0.25]$.\
[**Question:**]{} Let $(X,Y)$ be random variables assuming values in $\N := \{ 1 , \ldots , n \}$, $\M := \{ 1 , \ldots , m \}$ respectively with the joint distribution $\Pr \{ X = i \& Y = j \} =\th_{ij}$. Is $\al(X,Y) \geq \e$?
[**Proof:**]{} If we choose $\e = 0.25$, this problem reduces to that studied in Theorem \[thm:3b\], which is NP-complete. Therefore the present problem is NP-hard. On the other hand, given a candidate solution in terms of subsets $S \seq \N,T \seq \M$, it is possible to verify in polynomial time that $| P_\bth(S \times T) - P_\bmu(S) P_\bnu(T) | \geq \e$. Therefore the problem is NP-complete. $\halmos$
Upper and Lower Bounds for the Alpha-Mixing Coefficient {#ssec:32}
-------------------------------------------------------
Since computing the $\al$-mixing coefficient is NP-hard (because merely testing whether it exceeds a prespecified threshold is NP-complete), it is worthwhile to have efficiently computable upper and lower bounds for this mixing coefficient. The aim of this subsection is to present such bounds.
To contrast with later results on the $\phi$-mixing coefficient, we introduce a bit of notation. Suppose $\GA \in \R^{n \times m}$, and that $p,q \in [1,\infty]$. Then the induced norm $\nm \GA \nm_{p,q}$ is defined as \_[p,q]{} := \_[v\_p 1]{} v\_q . Explicit closed-form formulas are available for $\nm \GA \nm_{p,q}$ when $ (p,q) = (1,1), (2,2), (\infty,\infty), (2,\infty)$; see for example [@MV93]. However, not much is known about other combinations.
\[thm:3c\] Suppose $X,Y$ are random variables over finite sets $\Abb,\Bbb$ with joint distribution $\bth$ and marginals $\bmu,\bnu$ respectively. Define \[eq:3e\] = - \^t \^[n m]{}, where $\Th = [ \th_{ij} ]$. Then \[eq:3f\] (X,Y) = 0.25 \_[,1]{} .
[**Proof:**]{} Let $n$ denote $| \Abb |$, and define a map $h : 2^\Abb \ap \bi^n$ as follows: For a subset $S \seq \Abb$ h\_i(S) = { 1, & & a\_i S ,\
0, & & a\_i S . . \[perm\] Note that by definition we have: (S)+(S\^c) = \_n, where $\eb$ denotes a column vector whose components all equal one, and the subscript denotes its dimension. A similar map can be defined for $\Bbb$ as well. With this notation, for any subsets $S \seq \Abb, T \seq \Bbb$, we have that P\_(S) = \^t (S) = \[(S)\]\^t , P\_(T) = \^t (T) = \[(T)\]\^t . Moreover, with the joint distribution $\bth$, we have that P\_(S T) = \^t . Since the function $h$ is a bijection, it follows from (\[eq:34\]) that (X,Y) & = & \_[ {0,1}\^n, {0,1}\^m ]{} \^t - \^t \^t\
\[eq:3i\] & = & \_[ {0,1}\^n, {0,1}\^m ]{} \^t .
Let $\b \in \R^m$ be any fixed vector; then $\a^t \GA \b$ is maximized with respect to $\a \in \bi^n$ by choosing $a_i = 1$ if $(\GA \b)i \geq 0$, and $a_i = 0$ if $(\GA \b)_i < 0$. In other words, \_[\^n]{} \^t = \_i ( ( )\_i )\_+ . However, since the product distribution $\bmu \bnu^t$ and the joint distribution $\Th$ have the same marginals, it follows that \_m = \_n , \_n\^t = \_m . This implies that, for any vector $\a \in \bi^n$ and any $\b \in \R^m$, we have \^t = - ( \_n - )\^t , and also that \_i ( ( )\_i )\_+ = - \_i ( ( )\_i )\_- = 0.5 \_1 . Therefore \_[\^n]{} \^t = - \_[\^n]{} \^t , whence \_[\^n]{} | \^t | & = & \_[\^n]{} \^t\
& = & \_i ( ( )\_i )\_+\
& = & 0.5 \_1 . As a consequence, it now follows from (\[eq:3i\]) that (X,Y) & = & \_[\^m]{} \_i ( ( )\_i )\_+\
& = & 0.5 \_[\^m]{} \_1 . \[eq:3g\]
The proof is completed by showing that the quantity on the right side of (\[eq:3g\]) equals $\nm \GA \nm_{\infty,1}$. For an arbitrary $\b \in \R^m$, define the associated vector $\z \in \R^m$ by $\z = 2 \b - \eb_m$, and observe that, as $\b$ varies over $\bi^m$, $\z$ varies over $\bp^m$. Also, = 0.5 ( - \_m ) = 0.5 , because $\GA \eb_m = \bz_m$. Therefore (X,Y) = 0.5 \_[\^m]{} \_1 = 0.25 \_[\^m]{} \_1 . Now consider the optimization problem \_\_1 . Since the objective function is convex and the feasible region is convex and polyhedral, the optimum occurs at an extremum point. In other words, \_[\_1]{} \_1 = \_[\^m]{} \_1 . However, by definition the left side equals $\nm \GA \nm_{\infty,1}$. This shows that (X,Y) = 0.25 \_[,1]{} , which is the desired conclusion. $\halmos$
By combining Theorems \[thm:3b\] and \[thm:3c\], we can conclude that computing the induced norm $\nm \cdot \nm_{\infty,1}$ of an arbitrary matrix is NP-hard. However, this result is already shown in [@Nesterov98], which also gives an efficiently computable upper bound for this induced norm, with a guaranteed suboptimality. By adapting that result, we can derive efficiently computable upper and lower bounds for the $\al$-mixing coefficient.
\[thm:3d\] [@Nesterov98] Given a matrix $\GA \in \R^{n \times m}$, define $c(\GA)$ to be the value of the following optimization problem: \[eq:3j\] c() := 0.5 \_[\^n , \^m]{} \_+\_1 subject to 0 , where $M \geq 0$ denotes that $M$ is positive semidefinite. Then \[eq:3k\] 0.1086 c() (X,Y) 0.25 c(),
[**Proof:**]{} It is shown in [@Nesterov98] that \_[,1]{} c() 2.3 \_[,1]{}. Therefore (1/2.3) c() \_[,1]{} c() . Combining this with (\[eq:3f\]) leads to the desired conclusion. $\halmos$
Note that the computation of $c(\GA)$ requires the solution of a semidefinite optimization program. Efficient algorithms to solve semidefinite programs can be found in [@VB96].
An Exact Formula for the Phi-Mixing Coefficient {#ssec:33}
-----------------------------------------------
We have seen in Section \[ssec:31\] that estimating the $\al$-mixing coefficient is NP-complete. Though the definition of the $\phi$-mixing coefficient resembles that of the $\al$-mixing coefficient in terms of (apparently) requiring an enumeration of all subsets, it turns out that there is an efficiently computable exact formula for the $\phi$-mixing coefficient.
\[thm:3e\] Suppose $X,Y$ are random variables over finite sets $\Abb,\Bbb$ with joint distribution $\bth$ and marginals $\bmu,\bnu$ respectively. Then (X,Y) & = & \_j \_i ( \_[ij]{} )\_+\
& = & 0.5 \_j \_i | \_[ij]{} |\
& = & 0.5 \^[-1]{} \_[1,1]{} \[eq:3h\]
[**Proof:**]{} We already know from Theorem \[thm:33\] that (X|Y) := \_[S , T ]{} { X S | Y T } - { X S } . Now define g(S,T) := { X S | Y T } - { X S } Then (X|Y) = \_[T ]{} \_[S ]{} g(s,T) .
Next, using obvious notation, let us rewrite $g(S,T)$ as g(S,T) & = & P(S|T) - P(T)\
& = & . Now, suppose $T_1 , T_2$ are [*disjoint*]{} subsets of $\Bbb$. Then P(T \_1 T\_2) = P(T\_1) + P(T\_2) , P(S (T\_1 T\_2) ) = P(S T\_1) + P(S T\_2) because the events $S \times T_1$ and $S \times T_2$ are also disjoint. Therefore g(S,T\_1 T\_2) & = &\
& = &\
& + &\
& = &\
& + &\
& = & ł\_1 g(S,T\_1) + ł\_2 g(S,T\_2) , where ł\_1 = , ł\_2 = . Therefore $g(S,T_1 \cup T_2)$ is a convex combination of $g(S,T_1)$ and $g(S,T_2)$. There is nothing special about writing $T$ as a disjoint union of [*two*]{} subsets. In general, if $T = \{ j_1 , \ldots , j_k \}$, then the above reasoning can be repeated to show that g(S,T) = \_[l=1]{}\^k g(S, { j\_l }) , that is, $g(S,T)$ is a convex combination of $g(S, \{ j_l \})$. This shows that, if $T = \{ j_1 , \ldots , j_k \}$, then g(S,T) \_[1 l k]{} g(S, { j\_l }) . Hence, for a given subset $S \seq \Abb$, we have \_[T ]{} g(S,T) = \_[j ]{} g(S,{ j } ) . The importance of the above equation lies in enabling us to replace a maximum over all [*subsets*]{} of $\Bbb$ with a maximum over all [*elements*]{} of $\Bbb$. This is how we break through the barrier of enumerating an exponential number of subsets. As a consequence we have (X|Y) = \_[j ]{} \_[S ]{} g(S,{ j } ) .
Now for a fixed subset $S \seq \Abb$ and fixed element $j \in \Bbb$, we have g(S,{ j } ) & = & { X S | Y = j } - { X S }\
& = & \_[i S ]{}\
& = & \_[i S ]{} \[ \_[ij]{} - \_i \_j \]\
& = & \_[i S ]{} \_[ij]{} . Hence, for a fixed $j \in \Bbb$, the summation is maximized with respect to $S$ by choosing $i \in S$ if $\g_{ij} \geq 0$ and $i \not \in S$ if $\g_{ij} < 0$. The resulting maximum value (for a fixed $j \in \Bbb$) is \_[S ]{} g(S,{ j } ) = \_i (\_[ij]{})\_+ . So finally (X|Y) & = & \_[j ]{} \_[S ]{} g(S,{ j } )\
& = & \_j \_i (\_[ij]{})\_+ , which is the first equation in (\[eq:3h\]). The second equation in (\[eq:3h\]) follows from the the fact that $\eb_n^t \GA = \bz_m$, which implies in turn that, for each fixed $
j \in \Bbb$, we have that \_i (\_[ij]{})\_+ = - \_i (\_[ij]{})\_- = 0.5 \_i |\_[ij]{}| . Lastly, the fact that \_j \_i | \_[ij]{} | = \^[-1]{} \_[1,1]{} is standard and can be found in many places, e.g. [@MV93]. $\halmos$
We conclude this section by observing that the $\al$-mixing coefficient is proportional to the $(\infty,1)$-induced norm of the matrix $\GA$, whereas the $\phi$-mixing coefficient is proportional to the $(1,1)$-induced norm of the matrix $\GA [ {\rm Diag} ( \bnu ) ]^{-1}$. Therefore the reason for the NP-hardness of computing the $\al$-mixing coeffient and the efficient computability of the $\phi$-mixing coefficient lies in the nature of the induced norms that need to be computed.
Data Processing-Type Inequalities for Mixing Coefficients {#sec:6}
=========================================================
In this section we study the case where two finite-valued random variables are conditionally independent given a third finite-valued random variable, and prove inequalities of the data processing-type for the associated mixing coefficients. The nomenclature ‘data processing-type’ is motivated by the well-known data processing inequality in information theory.
\[def:cond\] Suppose $X,Y,Z$ are random variables assuming values in finite sets $\Abb, \Bbb, \Cbb$ respectively. Then $X,Z$ are said to be [**conditionally independent**]{} given $Y$ if, $\fa i \in \Abb , j \in \Bbb , k \in \Cbb$, it is true that { X = i & Z = k | Y = j } & = & { X = i | Y = j }\
& & { Z = k | Y = j } . \[eq:61\]
If $X,Z$ are conditionally independent given $Y$, we denote this by $(X \perp Z) | Y$. Some authors also write this as ‘$X \ap Y \ap Z$ is a short Markov chain’, ignoring the fact that the three random variables can belong to quite distinct sets. In this case, it makes no difference whether we write $X \ap Y \ap Z$ or $Z \ap Y \ap X$, because it is obvious from (\[eq:61\]) that conditional independence is a symmetric relationship. Thus (X Z) | Y (Z X) | Y . Also, from the definition, it follows readily that if $(X \perp Z) | Y$, then $\fa S \seq \Abb , j \in \Bbb , U \seq \Cbb$ we have that { X S & Z U | Y = j } & = & { X S | Y = j }\
& & { Z U | Y = j } .\[eq:62\] However, in general, it is [*not true*]{} that, $\fa S \seq \Abb ,T \seq \Bbb , U \seq \Cbb$, { X S & Z U | Y T } & = & { X S | Y T }\
& & { Z U | Y T } . In fact, by setting $T = \Bbb$, it would follow from the above relationship that $X$ and $Z$ are independent, which is a stronger requirement than conditional independence.
Given two random variables $X,Y$ with joint distribution $\bth$ and marginal distributions $\bmu,\bnu$ of $X,Y$ respectively, the quantity H( ) := - \_[i = 1]{}\^n \_i \_i is called the [**entropy**]{} of $\bmu$, with analogous definitions for $H( \bnu)$ and $H( \bth )$; and the quantity I(X,Y) = H( ) + H( ) - H( ) is called the [**mutual information**]{} between $X$ and $Y$. It is clear that $I(X,Y) = I(Y,X)$. The following well-known inequality, referred to as the [**data-processing inequality**]{}, is the motivation for the contents of this section; see [@CT06 p. 34]. Suppose $(X \perp Z) | Y$. Then \[eq:63\] I(X,Z) { I(X,Y) , I(Y,Z) } .
\[thm:61\] Suppose $(X \perp Z) | Y$. Then \[eq:64\] (X,Z) { (X,Y) , (Y,Z) } .
\[thm:62\] Suppose $(X \perp Z) | Y$. Then \[eq:65\] (X,Z) { (X,Y) , (Y,Z) } .
\[thm:63\] Suppose $(X \perp Z) | Y$. Then \[eq:66\] (X|Z) { (X|Y) , (Y|Z) } , \[eq:67\] (Z|X) { (Z|Y) , (Y|X) } .
[**Proof of Theorem \[thm:61\]:**]{} Let $S \seq \Abb , U \seq \Cbb$ be arbitrary, and define r\_(S,U) := { X S & Z U } - { X S } { Z U } . Then (X,Y) = \_[ S , U ]{} r\_(S,U) . Recall from (\[eq:3g\]) that (X,Y) = \_[\^m]{} \_i ( ( )\_i )\_+ . Using the definition of the matrix $\GA$ and the one-to-one relationship between vectors in $\bi^m$ and subsets of $\Bbb$, we can rewrite the above equation equivalently as (X,Y) = \_[ T ]{} \_[i=1]{}\^n & \[ & { X = i & Y T }\
& - & { X = i } { Y T } \]\_+ .\[eq:312\]
Now we manipulate the quantity $r_\al(S,U)$ for arbitrary subsets $S \seq \Abb,
U \seq \Cbb$ to prove the desired conclusion.[^2] r\_(S,U) & = & \_[j=1]{}\^m \[ { X S & Y = j & Z U }\
& - & { X S & Y = j } { Z U } \]\
& = & \_[j=1]{}\^m \[ { X S | Y = j } { Z U | Y = j }\
& & { Y = j }\
& - & { X S | Y = j } { Y = j } { Z U } \]\
& = & \_[j=1]{}\^m { X S | Y = j } \[ { Z U & Y = j }\
& - & { Y = j } { Z U } \]\
& & \_[j=1]{}\^m { X S | Y = j } \[ { Z U & Y = j }\
& - & { Y = j } { Z U } \]\_+\
& & \_[j=1]{}\^m \[ { Z U & Y = j }\
& - & { Y = j } { Z U } \]\_+\
& & \_[U ]{} \_[j=1]{}\^m \[ { Z U & Y = j }\
& - & { Y = j } { Z U } \]\_+\
& = & (Y,Z) .
Since $S$ and $U$ are arbitrary, this implies that $\al(X,Z) \leq \al(Y,Z)$ whenever $X \ap Y \ap Z$ is a short Markov chain. Since $X \ap Y \ap Z$ is the same as $Z \ap Y \ap X$, it also follows that $\al(Z,X) \leq \al(Y,X)$. Finally, since $\al$ is symmetric, the desired conclusion (\[eq:64\]) follows. $\halmos$
[**Proof of Theorem \[thm:62\]:**]{} Suppose that $\Abb,\Bbb,\Cbb$ have cardinalities $n,m,l$ respectively. (The symbols $n,m$ have been introduced earlier and now $l$ is introduced.) Let $\bdelta$ denote the joint distribution of $(X,Y,Z)$, $\bzeta$ the joint distribution of $(X,Z)$, $\boldeta$ the joint distribution of $(Y,Z)$, and as before, $\bth$ the joint distribution of $(X,Y)$. Let $\bxi$ the marginal distribution of $Z$, and as before, let $\bmu, \bnu$ denote the marginal distributions of $X$ and $Y$. Finally, define c\_[jk]{} = = { Z = k | Y = j } . As can be easily verified, the fact that $(X \perp Z ) | Y$ (or (\[eq:61\])) is equivalent to \_[ijk]{} = = \_[ij]{} c\_[jk]{} , i , j , k . Also note the following identities: \_[i=1]{}\^n \_[ij]{} = \_j , \_[j=1]{}\^m \_[ij]{} = \_i , \_[j=1]{}\^m \_[ijk]{} = \_[ik]{} , i, j, k . Now it follows from the various definitions that (X,Z) & = & \_[i=1]{}\^n \_[k=1]{}\^l ( \_[ik]{} - \_i \_k ) \_+\
& = & \_[i=1]{}\^n \_[k=1]{}\^l ( \_[j=1]{}\^m ( \_[ijk]{} - \_[ij]{} \_k ) )\_+\
& & \_[i=1]{}\^n \_[k=1]{}\^l \_[j=1]{}\^m ( \_[ijk]{} - \_[ij]{} \_k )\_+\
& = & \_[i=1]{}\^n \_[k=1]{}\^l \_[j=1]{}\^m ( \_[ij]{} c\_[jk]{} - \_[ij]{} \_k )\_+\
& = & \_[k=1]{}\^l \_[j=1]{}\^m ( c\_[jk]{} - \_k )\_+\
& = & \_[k=1]{}\^l \_[j=1]{}\^m ( \_j c\_[jk]{} - \_j \_k )\_+\
& = & \_[k=1]{}\^l \_[j=1]{}\^m ( \_[jk]{} - \_j \_k )\_+\
& = & (Y,Z) . Now the symmetry of $\beta(\cdot,\cdot)$ serves to show that $\beta(X,Z) \leq \beta(X,Y)$. Putting both inequalities together leads to the desired conclusion. $\halmos$
[**Proof of Theorem \[thm:63\]:**]{} Suppose $(X \perp Z) | Y$. Since the $\phi$-mixing coefficient is not symmetric, it is necessary to prove two distinct inequalities, namely: (i) $\phi(X|Z) \leq \phi(X|Y)$, and (ii) $\phi(X|Z) \leq \phi(Y|Z)$.
[**Proof that $\phi(X|Z) \leq \phi(X|Y)$:**]{} For $S \seq \Abb$, define r\_(S) := \_[T ]{} { X S | Y T } , and observe that (X|Y) = \_[S ]{} \[ r\_(S) - P\_(S) \] . Suppose $S \seq \Abb, U \seq \Cbb$ are arbitrary. Then { X S & Z U } & = & \_[j=1]{}\^m { X S & Y = j & Z U }\
& = & \_[j=1]{}\^m { X S | Y = j }\
& & { Z U | Y = j } { Y = j }\
& = & \_[j=1]{}\^m { X S | Y = j }\
& & { Z U & Y = j }\
& & r\_(S) \_[j=1]{}\^m { Z U & Y = j }\
& = & r\_(S) { Z U } . Dividing both sides by $\Pr \{ Z \in U \}$ leads to { X S | Z U } r\_(S) , { X S | Z U } - P\_(S) r\_(S) - P\_(S) Taking the maximum of both sides with respect to $S \seq \Abb$, $U \seq \Cbb$ shows that (X|Z) (X|Y) .
[**Proof that $\phi(X|Z) \leq \phi(Y|Z)$:**]{} We begin by rewriting the expression for $\phi(X|Y)$. In order to make the equations fit, for $S \seq \Abb, T \seq \Bbb$ we will use $P(S|T)$ as a shorthand for $\Pr \{ X \in S | Y \in T \}$, and so on. With this convention, for $S \seq \Abb, T \seq \Bbb$, we have that P(S|T) - P(S) & = & \_[i S]{} \[ P(i|T) - P(i) \]\
& & \_[i S]{} \[ P(i|T) - P(i) \]\_+\
& & \_[i ]{} \[ P(i|T) - P(i) \]\_+ . Therefore (X|Y) & = & \_[S,T]{} \[ P(S|T) - P(S) \]\
& & \_T \_[i ]{} \[ P(i|T) - P(i) \]\_+ . \[eq:3l\] Actually this can be shown to be an equality, and not an inequality, but we will not expend space on that.
Let us define c(S,U) := { X S | Z U } - P\_(S) , and reason as follows: c(S,U) & = & { X S | Z U } - { X S }\
& = & \_[j=1]{}\^m \[ { X S & Y = j | Z U }\
& - & { X S & Y = j } \]\
& = & \_[j=1]{}\^m \[ { X S | Y = j & Z U }\
& & { Y = j | Z U }\
& - & { X S | Y = j } { Y = j } \]\
& = & \_[j=1]{}\^m { X S | Y = j } \[ { Y = j | Z U }\
& - & { Y = j } \]\
& & \_[j=1]{}\^m { X S | Y = j } \[ { Y = j | Z U }\
& - & { Y = j } \]\_+\
& & \_[j=1]{}\^m \[ { Y = j | Z U } - { Y = j } \]\_+\
& & \_[U ]{} \_[j=1]{}\^m \[ { Y = j | Z U } - { Y = j } \]\_+\
& & (Y|Z) , where the last step follows from (\[eq:3l\]). Since the right side is independent of both $S$ and $U$, the desired conclusion follows. $\halmos$.
Inconsistency of an Estimator for Mixing Coefficients {#sec:4}
=====================================================
Suppose $X,Y$ are real-valued random variables with some unknown joint distribution, and suppose we are given an infinite sequence of independent samples $\{ (x_i, y_i) , i = 1 , 2 , \ldots \}$. The question studied in this section and the next is whether it is possible to construct empirical estimates of the various mixing coefficients that converge to the true values as the number of samples approaches infinity. Let \_[X,Y]{}(a,b) = { X a & Y b } denote the true but unknown joint distribution function of $X$ and $Y$, and let $\Phi_X(\cdot),\Phi_Y(\cdot)$ denote the true but unknown marginal distribution functions of $X,Y$ respectively. Using the samples$\{ (x_i, y_i) , i = 1 , 2 , \ldots \}$, we can construct three ‘stair-case functions’ that are empirical estimates of $\Phi_X,\Phi_Y$ and $\Phi_{X,Y}$ based on the first $l$ samples, as follows: \[eq:41\] \_X(a;l) := \_[i=1]{}\^l I\_[ { x\_i a } ]{} , \[eq:42\] \_Y(b;l) := \_[i=1]{}\^l I\_[ { y\_i b } ]{} , \[eq:43\] \_[X,Y]{}(a,b;l) := \_[i=1]{}\^l I\_[ { x\_i a & y\_i b } ]{} , where as usual $I$ denotes the indicator function. Thus $\Phih_X(a;l)$ counts the fraction of the first $l$ samples that are less than or equal to $a$, and so on. With this construction, the well-known Glivenko-Cantelli lemma (see [@Glivenko33; @Cantelli33] or [@Loeve77 p. 20]) states that the empirical estimates converge uniformly and almost surely to their true functions as the number of samples $\lai$. Thus $\Phih_{X,Y}$ is a consistent estimator of the true joint distribution. Thus one might be tempted to think that an empirical estimate of any (or all) of the three mixing coefficients based on $\Phih_{X,Y}$ will also converge to the true value as $\lai$. The objective of this brief section is to show that this is not so. Hence estimates of mixing coefficients derived from a consistent estimator of the joint distribution need not themselves be consistent.
\[thm:41\] Suppose $\Phih_{X,Y}$ is defined by (\[eq:43\]), and that $x_i \neq x_j$ and $y_i \neq y_j$ whenever $i \neq j$. Let $\betah_l$ denote the $\beta$-mixing coefficient associated with the joint distribution $\Phih_{X,Y}(\cdot,\cdot;l)$. Then $\betah_l = (l-1)/l$.
[**Proof:**]{} Fix the integer $l$ in what follows. Note that the empirical distribution $\Phih_{X,Y}(\cdot,\cdot;l)$ depends only the totality of the $l$ samples, and not the order in which they are generated. Without loss of generality, we can replace the samples $(x_1 , \ldots , x_n)$ by their ‘order statistics’, that is, the same samples arranged in increasing order, and do the same for the $y_i$. Thus the assumption is that $x_1 < x_2 < \ldots < x_l$ and similarly $y_1 < y_2 < \ldots < y_l$. With this convention, the empirical samples will be of the form $\{ (x_1 , y_{\pi(1)}) , \ldots , (x_l , y_{\pi(l)}) \}$ for some permutation $\pi$ of $\{ 1, \ldots , l \}$. Therefore the probability measure associated with the empirical distribution $\Phih$ is purely atomic, with jumps of magnitude $1/l$ at the points $\{ (x_1 , y_{\pi(1)}) , \ldots , (x_l , y_{\pi(l)}) \}$. So we can simplify matters by replacing the real line on the $X$-axis by the finite set $\{ x_1 , \ldots , x_l \}$, and the real line on the $Y$-axis by the finite set $\{ y_1 , \ldots , y_l \}$. With this redefinition, the joint distribution $\bth$ assigns a weight of $1/l$ to each of the points $(x_i, y_{\pi(i)})$ and a weight of zero to all other points $(x_i , y_j)$ whenever $j \neq \pi(i)$, while the marginal measures $\bmu,\bnu$ of $X$ and $Y$ will be uniform on the respective finite sets. Thus the product measure $\bmu \times \bnu$ assigns a weight of $1/l^2$ to each of the $l^2$ grid points $(x_i,y_j)$. From this, it is easy to see that \_l = (,) = (l-1)/l . This is the desired conclusion. $\halmos$
\[corr:41\] Suppose the true but unknown distribution $\Phi_{X,Y}$ has a density with respect to the Lebesgue measure. Then $\betah_l \ap 1, \phih_l \ap 1$ almost surely as $\lai$.
[**Proof:**]{} If the true distribution has a density, then it is nonatomic, which means that with probability one, samples will be pairwise distinct. It now follows from Theorem \[thm:41\] that \_l \_l = 1 . This is the desired conclusion. $\halmos$
Consistent Estimators for Mixing Coefficients {#sec:5}
=============================================
The objective of the present section is to show that a simple modification of the ‘naive’ algorithm proposed in Section \[sec:4\] does indeed lead to consistent estimates, provided appropriate technical conditions are satisfied.
The basic idea behind the estimators is quite simple. Suppose that one is given samples $\{ (x_i , y_i) , i \geq 1 \}$ generated independently and at random from an unknown joint probability measure $\th \in \M(\R^2)$. Given $l$ samples, choose an integer $k_l$ of bins. Divide the real line into $k_l$ intervals such that each bin contains $\lfloor l/k_l \rfloor$ or $\lfloor l/k_l \rfloor + 1$ samples for both $X$ and $Y$. In other words, carry out percentile binning of both random variables. One way to do this (but the proof is not dependent on how precisely this is done) is as follows: Define $m_l = \lfloor l/k_l \rfloor , r = l - k_l m_l$, and place $m_l +1$ samples in the first $r$ bins and $m_l$ samples in the next $m_l - r$ bins. This gives a way of discretizing the real line for both $X$ and $Y$ such that the discretized random variables have nearly uniform marginals. With this binning, compute the corresponding joint distribution, and the associated empirical estimates of the mixing coefficients. The various theorems below show that, subject to some regularity conditions, the empirical estimates produced by this scheme do indeed converge to their right values with probability one as $\lai$, [*provided that*]{} $m_l \ap \infty$, or equivalently, $k_l / l \ap 0$, as $\lai$. In other words, in order for this theorem to apply, the number of bins must increase more slowly than the number of samples, so that the number of samples per bin must approach infinity. In contrast, in Theorem \[thm:41\], we have effectively chosen $k_l = l$ so that each bin contains precisely one sample, which explains why that approximation scheme does not work.
To state the various theorems, we introduce a little bit of notation, and refer the reader to [@Berberian70] for all concepts from measure theory that are not explicitly defined here. Let $\M(\R), \M(\R^2)$ denote the set of all measures on $\R$ or $\R^2$ equipped with the Borel $\s$-algebra. Recall that if $\th , \eta \in \M(\R)$ or $\M(\R^2)$, then $\th$ is said to be [**absolutely continuous**]{} with respect to $\eta$, denoted by $\th \ll \eta$, if for every measurable set $E$, $\eta(E) = 0 \imp \th(E) = 0$.
Next, let $\th$ denote the joint probability measure of $(X,Y)$, and let $\mu,\nu$ denote the marginal measures. Thus, for every measurable[^3] subset $S \seq \R$, the measure $\mu(S)$ is defined as $\th(S \times \R)$and similarly for all $T \seq \R$, the measure $\nu(T)$ is defined as $\th(\R \times T)$. Now the key assumption made here is that [*the joint measure $\th$ is absolutely continuous with respect to the product measure $\mu \times \nu$*]{}. In the case of finite-valued random variables, this assumption is automatically satisfied. Suppose that for some pair of indices $i,j$, it is the case that $\mu_i \cdot \nu_j = 0$. Then either $\mu_i = 0$ or $\nu_j = 0$. If $\mu_i = 0$, then it follows from the identity $\sum_{j'} \th_{i j'} = \mu_i$ that $\th_{ij'} = 0$ for all $j'$, and in particular $\th_{ij} = 0$. Similarly if $\nu_j = 0$, then it follows from the identity $\sum_{i'} \th_{i'j} = \nu_j$ that $\th_{i'j} = 0$ for all $i'$, and in particular $\th_{ij} = 0$. In either case it follows that $\th_{ij} = 0$, so that $\th \ll \mu \times \nu$. However, in the case of real random variables, this need not be so. For example, replace $\R \times \R$ by the unit square, and let $\th$ be the diagonal measure. Then both marginals $\mu,\nu$ are the uniform measures on the unit interval, and the product $\mu \times \nu$ is the uniform measure on the unit square – and $\th$ is singular with respect to the uniform measure.
Next we introduce symbols for the various densities. Since $\th \ll \mu \times \nu$, it follows that $\th$ has a Radon-Nikodym derivative with respect to $\mu \times \nu$, which is denoted by $f(\cdot,\cdot)$. So for any sets $S, T \seq \R$, it follows that (S T) & = & \_S \_T f(x,y) d (y) d (x)\
& = & \_T \_S f(x,y) d (x) d (y) . For any $T \seq \R$ with $\nu(T) > 0$, the conditional probability $\Pr \{ X \in S | Y \in T \}$ is given by { X S | Y T } & = & =\
& = & \_S d (x) .
\[thm:51\] Suppose $\th \ll \mu \times \nu$, and that $k_l \ap \infty$, $k_l/l \ap 0$ as $\lai$. Then the empirically estimated $\beta$-mixing coefficient $\betah_l$ converges almost surely to the true value $\beta$ as $\lai$.
\[thm:52\] Suppose $\th \ll \mu \times \nu$, and in addition that the density $f(\cdot,\cdot)$ belongs to $L_\infty(\R^2)$. Suppose that $k_l \ap \infty$, $k_l/l \ap 0$ as $\lai$. Then the empirically estimated $\al$-mixing coefficient $\alh_l$ converges almost surely to the true value $\al$ as $\lai$, and the empirically estimated $\phi$-mixing coefficient $\phi_l$ converges almost surely to the true value $\phi$ as $\lai$.
Note that the absolute continuity assumption $\th \ll \mu \times \nu$ guarantees that the density $f(\cdot,\cdot) \in L_1(\R^2, \mu \times \nu)$. So no additional technical assumptions are needed to ensure that the sequence of empirical estimates $\betah_l$ converges to its true value. However, in order to establish that the sequences of empirical estimates $\alh_l$ and $\phih_l$ converge to their true values, we have added an assumption that the density $f$ is bounded almost everywhere. This condition is intended to ensure that conditional densities do not ‘blow up’. In the case of finite-valued variables, we have already seen that the condition $\th \ll \mu \times \nu$ holds automatically, which means that the ‘density’ $f_{ij} \th_{ij}/(\mu_i \nu_j)$ is always well-defined. Since there are only finitely many values of $i$ and $j$, this ratio is also bounded. However, in the case of real-valued random variables, this condition needs to be imposed explicitly.
The proofs of these two theorems are based on arguments in [@LN96; @WKV05]. In the proof of Theorem \[thm:51\], we can use those arguments as they are, whereas in the proof of Theorem \[thm:52\], we need to adapt them. To facilitate the discussion, we first reprise the relevant results from [@LN96; @WKV05].
\[def:51\] Let $( \OM, \F )$ be a measurable space, and let $Q$ be a probability measure on $( \OM, \F )$. Suppose $\{ I_1 , \ldots , I_L \}$ is a finite partition of $\OM$, and that $\{ I_1^{(m)} , \ldots , I_L^{(m)} \}$ is a sequence of partitions of $\OM$. Then $\{ I_1^{(m)} , \ldots , I_L^{(m)} \}$ is said to [**converge to**]{} $\{ I_1 , \ldots , I_L \}$ with respect to $Q$ if, for every probability measure $P$ on $( \OM, \F )$ such that $P \ll Q$, it is the case that P(I\_i\^[(m)]{}) P(I\_i) .
See [@WKV05 Definition 1].
\[thm:54\] Suppose $Q$ is a probability measure on $(\R,\B)$ that is absolutely continuous with respect to the Lebesgue measure, $L$ is a fixed integer, and that $\{ I_1 , \ldots , I_L \}$ is an equiprobable partitioning of $\R$. In other words, choose numbers - = a\_0 < a\_1 < …< a\_[L-1]{} < a\_L = + such that the semi-open intervals $I_i = (a_{i-1},a_i]$ satisfy Q(I\_i) = 1/L , i = 1 , …, L . Suppose $\{ y_1 , \ldots , y_m \}$ are i.i.d. samples generated in accordance with $Q$, and that $m = l_m T$ with $l_m \in \Nbb$, an integer. Let $\{ I_1^{(m)} , \ldots , I_L^{(m)} \}$ denote the empirical equiprobable partitioning associated with the samples $\{ y_1 , \ldots , y_m \}$. Then $\{ I_1^{(m)} , \ldots , I_L^{(m)} \}$ converges to $\{ I_1 , \ldots , I_L \}$ with respect to $Q$ as $\mai$.
[**Proof:**]{} See [@WKV05 Lemma 1].
\[thm:55\] Let $( \OM, \F )$ be a measurable space, and let $Q$ be a probability measure on $( \OM, \F )$. Suppose $\{ I_1^{(m)} , \ldots , I_L^{(m)} \}$ is a sequence of partitions of $\OM$ that converges with respect to $Q$ to another partition $\{ I_1 , \ldots , I_L \}$ as $\mai$. Suppose $\{ x_1 , \ldots , x_n \}$ are i.i.d. samples generated in accordance with a probability measure $P \ll Q$, and let $P_n$ the empirical measure generated by these samples. Then \_ \_ P\_n(I\_i\^[(m)]{}) = P(I\_i) , i .
[**Proof:**]{} See [@WKV05 Lemma 2].
Before proceeding to the proofs of the two theorems, we express the three mixing coefficients in terms of the densities. As stated in (\[eq:23\]), we have that \[eq:43a\] (X,Y) = 0.5 \_\_| f(x,y) - 1 | d (x) d (y) . Here we take advantage of the fact that the ‘density’ of $\mu$ with respect to itself is one, and similarly for $\nu$. Next, as in Theorem \[thm:33\], we can drop the absolute value signs in the definitions of $\al(X,Y)$ and of $\phi(X|Y)$. Therefore the various mixing coefficients can be expressed as follows: \[eq:44\] (X,Y) = \_T \_S \_S \_T \[ f(x,y) - 1 \] d (y) d (x) , \[eq:45\] (X,Y) = \_T \_S \_S d (x) . Now, for each fixed set $T$, let us define signed measures $\kappa_T$ and $\d_T$ as follows: \_T(x) = \_T \[ f(x,y) - 1 \] d (y) , \_T(x) = \_T d (y) - 1 , and associated support sets A\_+(T) = { x : \_T(x) 0 } , B\_+(T) = { x : \_T(x) 0 } . Then it is easy to see that, for each fixed set $T$, the supremum in (\[eq:44\]) is achieved by the choice $S = A_+(T)$ while the supremum in (\[eq:45\]) is achieved by the choice $S = B_+(T)$. Therefore (X,Y) & = & \_T \_[A\_+(T)]{} \_T(x) d (x)\
& = & \_T \_\_+ d (x) , \[eq:46\] (X|Y) & = & \_T \_[B\_+(T)]{} \_T(x) d (x)\
& = & \_T \_\_+ d (x) . \[eq:47\] These formulas are the continuous analogs of (\[eq:312\]) and (\[eq:3l\]) respectively.
[**Proof of Theorem \[thm:51\]:**]{} For a fixed integer $L \geq 2$, choose real numbers - = a\_0 < a\_1 < …< a\_[L-1]{} < a\_L = + , - = b\_0 < b\_1 < …< b\_[L-1]{} < b\_L = + such that the semi-open intervals $I_i = (a_{i-1},a_i], J_i = (b_{i-1},b_i]$ satisfy (I\_i) = 1/L , (J\_i) = 1/L , i = 1 , …, L . Now define the equiprobable partition of $\R^2$ consisting of the $L \times L$ grid $\{ I_i \times J_j , i , j = 1 , \ldots , L \}$. Next, based on the $l$-length empirical sample $\{ (x_1 , y_1) , \ldots ,
(x_l,y_l) \}$, construct empirical marginal distributions $\hat{\mu}$ for $X$ and $\hat{\nu}$ for $Y$. Based on these empirical marginals, divide both the $X$-axis $\R$ and $Y$-axis $\R$ into $L$ bins each having nearly equal fractions of the $l$ samples in each bin. This gives an empirical $L \times L$ partitioning of $\R^2$, which is denoted by $\{ I_i^{(L)} \times J_j^{(L)} , i , j = 1 , \ldots , L \}$. Using this grid, compute the associated empirical joint distribution $\hat{\th}_l$ on $\R^2$. Then the proof of [@WKV05 Lemma 1] can be adapted to show that the empirical partition $\{ I_i^{(L)} \times J_j^{(L)} , i , j = 1 , \ldots , L \}$ converges to the true partition $\{ I_i \times J_j , i , j = 1 , \ldots , L \}$ as $\lai$, with respect to the product measure $\mu \times \nu$. The only detail that differs from [@WKV05] is the computation of the so-called ‘growth function’. Given a set $A \seq \R^2$ of cardinality $m$, the number of different ways in which this set can be partitioned by a rectangular grid of dimension $L \times L$ is called the growth function, denoted by $\D_m$. It is shown in [@WKV05 Eq. (15)] that when the partition consists of $L$ intervals and the set being partitioned is $\R$, then $\D_m$ is given by the combinatorial parameter \_m = ( m+L\
L ) = . It is also shown in [@WKV05 Eq. (21)] that 2m h( 1/L) , where $h(\cdot)$ is defined by h(x) = - x x - (1-x) (1-x) , x (0,1) . When $\R$ is replaced by $\R^2$ and a set of $L$ intervals is replaced by a grid of $L^2$ rectangles, it is easy to see that the growth function is [*no larger than*]{} \_m \^2 . Therefore 4m h( 1/L) . In any case, since $L$, the number of grid elements, approaches $\infty$ as $\lai$, it follows that the growth condition proposed in [@LN96] is satisfied. Therefore the empirical partition converges to the true partition as $\lai$.
Next, let $\{ I_i \times J_j , i , j = 1 , \ldots , L \}$ denote, as before, the true equiprobable $L \times L$ gridding of $\R^2$. Suppose that, after $l$ samples $(x_r,y_r) , r = 1 , \ldots , l$ have been drawn, the data is put into $k_l$ bins. Then the expression (\[eq:43a\]) defining the true $\beta$-mixing coefficient can be rewritten as (X,Y) = 0.5 \_[i=1]{}\^[k\_l]{} \_[j=1]{}\^[k\_l]{} \_[I\_i]{} \_[J\_j]{} | f(x,y) - 1 | d (x) d (y) . Now suppose $l$ is an exact multiple of $k_l$. Then the empirical estimate based on the $k_l \times k_l$ empirical grid can be written as \_l = 0.5 \_[i=1]{}\^[k\_l]{} \_[j=1]{}\^[k\_l]{} | C\_[ij]{} - 1 | k\_l\^[-2]{} , where $C_{ij}$ denotes the number of samples $(x_r,y_r)$ in the $ij$-th cell of the [*empirical*]{} (not true) equiprobable grid. If $l$ is not an exact multiple of $k_l$, then some bins will have $\lfloor l/k_l \rfloor$ elements while other bins will have $\lfloor l/k_l \rfloor + 1$ elements. As a result, the term $k_l^{-2}$ gets replaced by $(s_i t_j)/l^2$ where $s_i$ is the number of samples in $I_i^{(l)}$ and $t_j$ is the number of samples in $J_j^{(l)}$. Now, just as in [@WKV05 Eq. (36) [*et seq.*]{}], the error $| \betah_l - \beta (X,Y) |$ can be bounded by the sum of two errors, the first of which is caused by the fact that the empirical equiprobable grid is not the same as the true equiprobable grid (the term $e_1$ of [@WKV05]), and the second is the error caused by approximating an integral by a finite sum over the true equiprobable grid (the term $e_2$ of [@WKV05]). Out of these, the first error term goes to zero as $\lai$ because, if $k_l/l \ap 0$ so that each bin contains increasingly many samples, the empirical equiprobable grid converges to the true equiprobable grid. The second error terms goes to zero because the integrand in (\[eq:43a\]) belongs to $L_1(\R^2,\mu \times \nu)$, as shown in [@WKV05 Eq. (37)]. $\halmos$
[**Proof of Theorem \[thm:52\]:**]{} The main source of difficulty here is that, whereas the expression for $\beta(X,Y)$ involves just a single integral, the expressions for $\al(X,Y)$ and for $\phi(X,Y)$ involve the supremum over all sets $T \seq \R$. Thus, in order to show that the empirical estimates converge to the true values, we must show not only that empirical estimates of integrals of the form $\int_\R [\kappa_T]_+ d \mu(x)$ and $\int_\R [\d_T]_+ d \mu(x)$ converge to their correct values for each fixed set $T$, but also that the convergence is in some sense uniform with respect to $T$. This is where we use the boundedness of the density $f(\cdot,\cdot)$. The details are fairly routine modifications of arguments in [@WKV05]. Specifically, (switching notation to that of [@WKV05]), suppose that in their Equation (27), we have not just one measure $\mu$, but rather a family of measures $\mu_T$, indexed by $T$, and suppose there exists a finite constant $c$ such that for every set $S$ we have $\mu_T(S) \leq c Q(S)$. Then it follows from Equation (27) [*et seq.*]{} of [@WKV05] that \_T ( ( a\_i a\_l\^m , a\_i a\_i\^m \] ) c Q( ( a\_i a\_l\^m , a\_i a\_i\^m \] ), T . Therefore \_ \_T \_T ( ( a\_i a\_l\^m , a\_i a\_i\^m \] ) = 0 . With this modification, the rest of the proof in [@WKV05] can be mimicked to show the following: In the interests of brevity, define r\_T = \_\_+ d (x) and let $\hat{r}_{T,l}$ denote its empirical approximation. Then, using the above modification of the argument in [@WKV05], it follows that \_ \_T | r\_T - \_T | = 0 . As a consequence, \_ \_T \_T = \_T r\_T = (X,Y) . The proof for the $\phi$-mixing coefficient is entirely similar. $\halmos$
Concluding Remarks {#sec:7}
==================
In this paper we have studied the problems of computing and estimating the mixing coefficients between two random variables in two important cases, namely: finite-valued and real-valued random variables. Three different mixing coefficients were studied, namely $\al$-mixing, $\beta$-mixing and $\phi$-mixing coefficients. In the case of finite-valued random variables, it has been shown that determining whether the $\al$-mixing coefficient exceeds a prespecified threshold is an NP-complete problem. Efficiently computable upper and lower bounds for the $\al$-mixing coefficients have been derived. In contrast, an explicit and efficiently computable formula has been derived for the $\phi$-mixing coefficient. Analogs of the data-processing inequality from information theory have been established for each of the three kinds of mixing coefficients. In the case of real-valued random variables, by using percentile binning and allowing the number of bins to increase more slowly than the number of samples, we can generate empirical estimates that converge to the true values for all the three kinds of mixing coefficients.
Several interesting questions are thrown up by the contents of this paper. As mentioned in the introduction, mixing coefficients were originally introduced as a way of extending the law of large numbers to stochastic processes that are not i.i.d. The problem studied in Section \[sec:5\] is to estimate the mixing coefficient between two real-valued random variables $X$ and $Y$, based on i.i.d. samples of the pair. A counterpoint to this problem is that studied in [@MSS11], where the objective is to estimate the $\beta$-mixing rate of a stationary stochastic process $\{ X_t \}$ from a single sample path. For a fixed integer $k$, the rate $\beta(k)$ can be interpreted as the $\beta$-mixing coefficient between the semi-infinite past $X_{-\infty}^0$ and the semi-infinite future $X_k^\infty$. However, the techniques presented here do not work for that problem, whereas [@MSS11] presents a comprehensive solution in terms of “blocking” and “histogramming”, that is, estimating the joint distribution of $d_l$ consecutive random variables, when $l$ samples are available in all. It is interesting to note that the convergence results in [@MSS11] also depend on letting $d_l$ grow more slowly than $l$. Specifically, as shown in [@MSS11 Theorem 2.3], the estimator converges to the true value provided $d_l = O(\exp[W(\log l)])$, where $W$ is the Lambert $W$ function. More details can be found in [@MSS11]. It would be worthwhile to explore whether similar estimators can be constructed for the $\al$-mixing rate of a stochastic process.
Another direction is to explore whether analogs of the data processing inequality, namely (\[eq:64\]) through (\[eq:67\]), hold for real-valued random variables. The proof techniques in Section \[sec:6\] make heavy use of the finiteness of the underlying sets where the various random variables assume their values. On the other hand, there are analogous formulas for real-valued random variables, namely (\[eq:46\]) and (\[eq:47\]). It might therefore be possible to extend the proofs in Section \[sec:6\] making use of these formulas. However, the technicalities may prove to be formidable.
In the consistency theorems of Section \[sec:5\], the requirement that the bins should consist of empirically equiprobable (or percentile) samples is really not necessary. A close examination of the proof techniques used in [@WKV05] shows that, so long as the minimum number of samples in each bin approaches infinity as $\lai$, the results would still hold. We leave it to the reader to state and prove such results. The later parts of the paper [@WKV05] contain some proposals on how to speed up the convergence of the empirical estimates of the Kullback-Leibler divergence between two unknown measures. It would be worthwhile to explore whether similar speed-ups can be found for the algorithms proposed here for estimating mixing coefficients from empirical data.
[99]{}
S. Boyd and L. Vandenberghe, *Convex optimization*.1em plus 0.5em minus 0.4emCambridge university press, 2004.
Sterling K. Berberian, [*Measure and Integration*]{}, Chelsea, New York, 1970.
F. P. Cantelli, “Sulla determinazione empirica delle legge di probabilità”, [*Giornali dell’Istituto Italia degli Attuari*]{}, 4, 421-424, 1933.
Thomas M. Cover and Joy A. Thomas, [*Elements of Information Theory*]{}, (Second Edition) John Wiley, New York, 2006.
I. Csiszár and J. Körner, [*Information Theory: Coding Theorems for Discrete Memoryless Systems*]{}, Academic Press, New York, 1981.
Paul Doukhan, [*Mixing: Properties and Examples*]{}, Springer-Verlag, Heidelberg, 1994.
M. R. Garey and D. Johnson, *Computers and intractability*.1em plus 0.5em minus 0.4emFreeman San Francisco, CA, 1979, vol. 174.
V. I. Glivenko, “Sulla determinazione empirica delle legge di probabilità”, [*Giornali dell’Istituto Italia degli Attuari*]{}, 4, 92-99, 1933.
I. A. Ibragimov, “Some limit theorems for stationary processes”, [*Theory of Probability and its Applications*]{}, 7, 349-382, 1962.
A. N. Kolmogorov, [*Foundations of Probability*]{}, (Second English Edition), Chelsea, New York, 1956.
Michel Loève, [*Probability Theory I*]{}, (4th Edition), Springer-Verlag, Heidelberg, 1977.
Gabor Lugosi and Andrew Nobel, “Consistency of data-driven histogram methods for density estimation and classification”, [*The Annals of Statistics*]{}, 24(2), 687-706, 1996.
A. A. Margolin *et al.*, “ARACNE: An algorithm for the reconstruction of gene regulatory networks in a cellular context,” *BMC Bioinformatics*, vol. 7(Supplement 1):S7, 20 March 2008.
Daniel J. McDonald, Cosma Rohilla Shalizi and Mark Schervish, “Estimating beta-mixing coefficients via histograms”, [*arXiv*]{}, 1109.5998, September 2011.
Ron Meir, “Nonparametric time series prediction through adaptive model selection”, [*Machine Learning*]{}, 39(1), 5-34, 2000.
Y. Nesterov, “Semidefinite relaxation and nonconvex quadratic optimization,” *Optimization methods and software*, vol. 9, no. 1-3, pp. 141–160, 1998.
M. Rosenblatt, “A central limit theorem and a strong mixing condition”, [*Proc. Nat’l. Acad. Sci.*]{}, 42(1), 43-47. January 1956.
Nitin Kumar Singh et al., “Reverse engineering gene interaction networks using the phi mixing coefficient”, under preparation; preliminary version [*arXiv*]{}, 1288.4066.
L. Vandenberghe and S. Boyd, “Semidefinite programming,” *SIAM review*, vol. 38, no. 1, pp. 49–95, 1996.
V. N. Vapnik, [*Statistical Learning Theory*]{}, John Wiley, New York, 1998.
M. Vidyasagar, [*Nonlinear Systems Analysis*]{}, (Second Edition), Prentice-Hall, Englewood Cliffs, NJ, 1993.
M. Vidyasagar, [*A Theory of Learning and Generalization*]{}, Springer-Verlag, London, 1997.
M. Vidyasagar, [*Learning and Generalization: With Applications to Neural Networks and Control Systems*]{}, Springer-Verlag, London, 2003.
M. Vidyasagar, “Probabilistic methods in cancer biology”, [*European Journal of Control*]{}, 17(5-6), 483-511, September-December 2011.
M. Vidyasagar, [*Computational Cancer Biology: An Interaction Network Approach*]{}, Springer-Verlag, London, 2013.
M. Vidyasagar and R. L. Karandikar, “System identification – A learning theory approach”, [*Journal of Process Control*]{}, 18, 421-430, 2007.
Qing Wang, Sanjeev R. Kulkarni and Sergio Verdú, “Divergence estimation of continuous distributions based on data-dependent partitions”, [*IEEE Transactions on Information Theory*]{}, 51(9), 3064-3074, September 2005.
Erik Weyer, “Finite sample properties of system identification of ARX models under mixing conditions”, [*Automatica*]{}, 36(9), 1291-1299, 2000.
[^1]: Department of Bioengineering, University of Texas at Dallas, 800 W. Campbell Road, Richardson, TX 75080; Emails: {Ahsen, M.Vidyasagar}@utdallas.edu. This work is supported by National Science Foundation Award \#1001643.
[^2]: Due to the width limitations of the two-column format, the long equations that follow have been split across two and sometimes three lines.
[^3]: Hereafter we drop this adjective; it is assumed that all sets that are encountered are measurable.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'A description of photoassociation by CW laser is formulated in the framework of grid methods. The Hamiltonian describing one or several bound states coupled to a multiple of continuum manifolds via a radiative field is written in the energy representation and diagonalized. The generality of the treatment allows to compute accurately and efficiently physical properties such as non-linear high-intensity energy shifts, line shapes, and photoassociation rates both for isolated and non-isolated resonances. Application is given to sodium photoassociation in the experimental conditions of Mc Kenzie [*et al*]{} \[Phys. Rev. Lett. [**88**]{}, 090403 (2002)\]. Inverted region for the dependency of the rate vs. the intensity and non-symmetric lineshapes were predicted to occur above the saturation limit. Comparison with the model of Bohn and Julienne \[Phys. Rev. A [**60**]{}, 414 (1999)\] is discussed.'
author:
- 'S. Kallush$^{1,2}$, R. Kosloff$^{2}$ and F. Masnou-Seeuws$^{1}$'
title: ' Grid methods for cold molecules : determination of photoassociation lineshapes and rate constants'
---
Introduction
============
Many experimental groups are presently working on the formation of ultracold molecules and molecular quantum gases. A very efficient scheme is the photoassociation reaction, where two colliding atoms absorb a photon to make a molecule [@thorsheim87]; moreover ultracold photoassociation spectroscopy [@jones2006] is capable of providing deep insight into the properties of long range molecules, cold atomic collisions, or pair correlations in a condensate. Theoretical two-body scattering models have been developed to compute the photoassociation rates [@pillet97; @napolitano97; @cote98; @bohn99; @taylor2004], in the framework of a perturbative treatment, adapted to situations where low-intensity CW lasers are employed and many-body effects are negligible. A linear variation of the photoassociation rate as a function of intensity is then predicted. In the work of Bohn and Julienne [@bohn99], which develops a semi-analytic theory yielding laser-induced energy shifts and l ine broadening, the departure from this linear behavior is described, the saturation limit at larger laser intensities being evaluated. In a condensate, the two-body model should fail and Javanainen and Mackie [@javanainen2002] have developed a many-body theory which predicts a saturation limit occurring at lower intensities and attributed to rogue photodissociation. The search for saturation effects has stimulated several experiments. Measurement, by Mc Kenzie [*[et al ]{}*]{} [@mckenzie2002], of the photoassociation rate in a sodium condensate as a function of laser intensity has shown a linear behavior in the intensity range considered, in agreement with the two-body Bohn and Julienne estimation, and in contrast with the prediction of Javanainen and Mackie. For a quantum degenerate lithium atomic gas, measurement [@prodan2003] of the intensity dependence of the photoassociation rate has verified both the large value of this rate (at a temperature $T$=600 nK) and the two-body saturation effect at intensities $I \ge$ 30 W/cm$^2$. The agreement with theory is considered as reasonable within the present experimental accuracy: the higher value of the measured saturation limit, and the oscillations in the experimental rate in the 30-80 W/cm$^2$ intensity range should be confirmed by measurements with a smaller error bar. For a non-degenerate gas of cesium atoms, at a temperature of 40 $\mu$K, within an optical dipole trap, saturation limit is reached for intensities above 100 W/cm$^2$[@kraft2005], and the experimental values then lay around the theoretical curve : it is not clear whether such oscillations are due to uncertainties or should be attributed to a physical effect.\
Recently Naidon and Masnou-Seeuws [@naidon2006] have developed a time-dependent many-body theory of photoassociation in an atomic Bose-Einstein condensate, showing that several physical effects depend on pair correlations. They show that the two existing models for the description of such correlations [@kohler2003b; @naidon2003], differ in their predictions of the photoassociation line shapes at large intensities. Precise measurements of the variation of the line shapes as a function of cw laser intensity should therefore give insight into the optimal description of many-body effects.\
The aim of the present paper is to anticipate such measurements by revisiting the theoretical treatment of photoassociation line shapes, taking advantage of the efficiency of grid methods [@kgrid96; @fatal; @slava99; @willner2004] for the numerical description of cold collisions and long range molecules, particularly in cases where several channels are involved. Actually, the later methods have been widely used for the computation of photoassociation rates [@dion01; @kerman04; @azzizi2004; @sage2005], since they can determine very precise energies and wave functions, both for bound and for continuum levels. It has been demonstrated that due to their global character, mapped grid methods are ideally suited to precise computation of the perturbations occurring in molecular spectra, such as Rb$_2$[@slava00a; @slava00b; @bergeman2006] or [K$_2$]{}[@lisdat01] singlet-triplet mixing. In such cases, the perturbation of one level of a given series should be attributed to a large number of bound or continuum levels of the other series, and local perturbation treatments are poorly adapted. However, up to know, attempts to compute the photoassociation line shapes via grid methods have not been very efficient [@pellegrini2003; @naidon2004] and the interpretation of precise measurements is routinely performed with Numerov type approaches [@portier2006]. The present paper is proposing an alternative way of implementing grid methods for the calculation of photoassociation line shapes, energy shifts, and rates.\
The paper is organized as follows: The description of the present model is reported in Section II-IV. The link with previous models, such as the Bohn and Julienne model is discussed in Section V, where the various assumptions and approximations are underlined. Application to the cases of Sodium photoassociation in the experimental conditions of Ref. [@mckenzie2002] is presented in Section VI. Section VII will summaries and discuss the outlook of the method.
The present model - preliminary remarks
=======================================
The physical process of photoassociation (PA) can be divided into three steps. [*Initially*]{} (step I), a thermal ensemble of cold atoms is in the ground electronic state; the probability to find a colliding pair with relative motion at energy $E$ and angular momentum $l$ is given by: $$p_{B}(E)= \frac{(2l+1)\exp(- E/(k_BT))}{Z_{eq}},
\label{boltz}$$ where $T$ is the temperature, and $k_B$ is the Boltzmann constant. The partition function $Z_{eq}$ was shown in [@koch06] to be well approximated by the classical expression ${Z_{eq}}=Q_{T}V = V(2 \pi \mu k_BT)^{3/2}/h^3$ for a gas of non-interacting pairs in a volume $V$. Further discussion regarding $Z_{eq}$ could be found in [@koch06].\
The radial wave function which corresponds to the pair is represented with a continuum eigenfunction $\left| {\psi^g_{E,l}(R) } \right\rangle$ of the single-channel Hamiltonian ${{{\bf {\hat {H}}}}}_{sc}$ $${{{\bf {\hat {H}}}}}_{sc} = -\frac{{\hbar ^2 }}{{2\mu }}\frac{{d^2 }}{{dR^2 }} + \frac{{l\left( {l + 1} \right)}}{{R^2 }} + V \left( R \right).
\label{eq:01}$$ Here, $\mu $ is the reduced mass of the pair, and $V \left( R \right)$ is the Born-Oppenheimer potential curve for the relevant, here the ground, electronic state.\
At the second step (step II), the intensity of a CW laser with frequency $\omega$ is ramped within a time scale that can be varied from hundreds of ns to several $\mu$s. Along this [*preparation*]{} period, the continuum states are coupled to one or more bound vibrational levels $v,J$ that belong to one or more excited electronic states $e$. Each of the levels in the excited state, denoted by $\left| {\phi^e_{v,J}(R)} \right\rangle$, may decay spontaneously with a natural free width $ \gamma^{0e}_{v,J}$, which is typically at the order of ns. The field that is applied on this stage varies with time, and so are the states of the total system.\
Finally, during the stable period of the experiment (step III), the laser reaches its peak constant intensity $I$. For the trap-loss or multiphoton detection scheme which will be described here, the duration of this period is on the order of ms.
Each of the energy levels of the system at this step can now be written as: $$\left| {\Psi _j(E)} \right\rangle = \sum\limits_{e,v,J} {a_{e,v,J}^j \left| {\phi^e_{v,J}} \right\rangle } + {\sum\limits_{l}{\int {b_j \left( {E} \right)\left| {\psi^g_{E,l} } \right\rangle dE} }} + \sum\limits_{
v',l} {c_{j} \left|{\psi^g_{v',l}}\right\rangle }$$ where we include the possibility of populating bound levels $\left|{\psi^g_{v',l}}\right\rangle $ of the ground state. In the present work, we shall consider only this third step, which of course is influenced by the preparation step. Since we shall not treat explicitly the preparation step, and possible transitory effects, a rapid switching on of the photoassociation laser will be assumed, justifying sudden approximation. The density of probability for a transition from initial state $\left| {\psi^g_{E,l}} \right\rangle$ to a final state $\left| {\Psi_j(E) } \right\rangle$ is given by: $$P_{(E,l) \to j} = \left| {\left\langle {\Psi _j(E) }
\mathrel{\left | {\vphantom {{\Psi _j(E) } {\psi^g_{E,l} }}}
\right. \kern-\nulldelimiterspace}
{\psi^g_{E,l} } \right\rangle } \right|^2$$\
At section IV it will be shown that a knowledge of $P_{(E,l) \to j}$ is sufficient to extract the experimental parameters of the PA process. An efficient method to compute the initial and the final states, $\left| \psi^g_{E,l} \right\rangle$ and $\left| {\Psi _j(E)} \right\rangle$, is the subject of the next section.
Numerical description of the field-molecule system
==================================================
The initial state is a thermal ensemble composed of atom pairs, with several angular momenta $l$. Each pair is coupled to several $J$ partial waves components on the excited state. For simplicity, the equations will be written hereby for one value $l$ and$J$ of the angular momentum. To obtain $P_{(E) \to j}$ one has to compute the initial and final eigenstates of the Hamiltonian without and with the light interaction. In the global grid method used here, this is achieved by representation of the Hamiltonian in a certain basis, and then a diagonalization. The diagonalization step is the most expensive computationally, and it scales as $N^3$ where $N$ is the number of basis function. A selection of an appropriate basis prior to the diagonalization is hence crucial for the efficiency of the method.\
The initial states: $R$-representation of the unperturbed molecule in a box
---------------------------------------------------------------------------
Using numerical grid methods, a numerically complete set of of eigenvalues and eigenvectors of the time-independent Schrödinger Hamiltonian ${{{\bf {\hat {H}}}}}^R_{sc}$ is obtained for the isolated molecule, for the ground and excited potentials. This is achieved by using the mapped Fourier grid method (see for example the references [@kgrid96; @fatal; @slava99; @nest]). Within the procedure, a unitary transformation ${{{\bf {\hat {U}}}}}_{mapp}$ is performed on the Hamiltonian eq. (\[eq:01\]), which is written in equally space grid $R$. The resulted Hamiltonian ${{{\bf {\hat {H}}}}}^x_{sc} = {{{\bf {\hat {U}}}}}^\dag_{mapp}{{{\bf {\hat {H}}}}}^R_{sc}{{{\bf {\hat {U}}}}}_{mapp}$, is represented on a grid with non-uniform distribution of points in the position space $x$, to take advantage of the phase-space occupancy of the problem. The potential energy operator is written simply as ${{{\bf {\hat {V}}}}}(x)$, and the kinetic energy operator reads: $${{{\bf {\hat {T}}}}} = - \frac{\pi ^2 }{2\mu L_x^2}{{{\bf {\hat {J}}}}}^{ - 1/2}{{{\bf {\hat {D}}}}} {{{\bf {\hat {J}}}}}^{ - 1} {{{\bf {\hat {D}}}}}{{{\bf {\hat {J}}}}}^{ - 1/2}$$ $L_x$ is the length of the grid, ${{{\bf {\hat {D}}}}}$ is the first derivative operator matrix, given by (assuming even number of grid points): $${{{\bf {\hat {D}}}}}_{i,j} = \left( \frac{d}{dx} \right)_{i,j} = \left\{ {\begin{array}{*{20}c}
{0,i = j} \\
{\frac{{\left( { - 1} \right)^{i - j} }}{{\sin \left[ {\left( {i - j} \right)\pi /N} \right]}},i \ne j} \\
\end{array}} \right.$$ and ${{{\bf {\hat {J}}}}}\left( {x} \right) ={\frac{{dx}}{{dR}}}$ is the (diagonal) Jacobian operator for the transformation.
A sufficient representation for ${{{\bf {\hat {H}}}}}^R_{sc}$ by an equally spaced grid demands typically $N \propto 10^4$ basis functions for the example considered below in section \[sec:sodium\]. As was shown in [@slava99], the use of a mapped grid leads to a reduction of about an order of magnitude in the basis size, to several thousands.\
A diagonalization of ${{{\bf {\hat {H}}}}}^x_{sc}$ gives the initial, interaction-free, wave functions and their energies for both electronic states $\left| {\psi^g_{E}(R) } \right\rangle$ and $\left| {\phi^e_{v,J}(R)} \right\rangle$. The eigenvalues with $E<0$, correspond to bound, discrete states and the ones with $E>0$ refer to box-normalized quasi-continuum states. The continuum under this picture is therefore tied to the box size from which is was deduced.\
In order to take account of the spontaneous emission in the excited channel $e$, an imaginary potential is introduced: [@cohen73] $${{{\bf {\hat {V}}}}}_d(R) = i \left( {\frac{2}{{3c}}} \right)^3 \frac{{\left| {\mu_{eg} \left( R \right)} \right|^2 {\omega\left( R \right)} ^3 }}{\hbar }$$ where ${ \mu_{eg} \left( R \right)}$ is the transition dipole moment, and ${\omega\left( R \right)} 2\pi \left( {V_e \left( R \right) - V_g \left( R \right)} \right)/\hbar $.
A diagonalization of ${{{\bf {\hat {H}}}}}^x_{sc}+ {{{\bf {\hat {V}}}}}_d $ results a complex eigen-energies. The imaginary part of the eigenvalues, $\gamma_{v,J}^{0e} = {\mathop{\rm Im}\nolimits} \left( {E_{v,J}^e } \right)$ is interpreted as the lifetime of the level $\left| {\phi _{v,J}^e \left( R \right)} \right\rangle $.
The final states
----------------
### The field-dressed molecule
A CW laser with constant intensity $I$, detuned by $\Delta_{a}$ relative to the atomic resonance line, introduces a radiative coupling between the ground state and the excited state. After making the rotating wave approximation, the full dressed Hamiltonian under the interaction can be written as: $${ {{{\bf {\hat {H}}}}}_R} = \left[ {\begin{array}{*{20}c}
{{{{\bf {\hat {H}}}}}^g_{sc}} & {{{{\bf {\hat {\Omega}}}}}(R)} \\
{{{{\bf {\hat {\Omega}}}}}(R)} & {{{{\bf {\hat {H}}}}}^e_{sc}} + \Delta_{a} \\
\end{array}} \right]$$ The coupling between the ground and the excited state is introduced by the Rabi coupling operator through the transition dipole moment: ${{{\bf {\hat {\Omega}}}}}(R) = -A{{{\bf {\hat {\mu}}}}}_{eg} (R)\sqrt {\frac{I}{{2c\varepsilon _0 }}}$, where ${{{\bf {\hat {\mu}}}}}_{eg}$ is the spatially dependent transition dipole moment between the two electronic states, and $A$ is an angular factor depending upon the polarization of the laser and the orientation of the molecular axis. In the $R$ representation $\Omega(R)$ is a local, i.e. diagonal, operator. A diagonalization of ${ {{{\bf {\hat {H}}}}}_R}$ gives the final states $\left| {\Psi _j(E) } \right\rangle$.\
### Energy eigenstates representation
For the final states, one has to diagonalize the Hamiltonian describing the interaction of the molecule with the radiation field routinely for many experimental parameters. A direct diagonalization of ${{{{\bf {\hat {H}}}}}_R}$ even for the mapped grid representation, is inefficient, mainly due to mainly due to the redundancy in the representation of the basis for both potentials. An improved representation for the coupled system is hence crucial. This can be obtained by writing ${{{\bf {\hat {H}}}}}_R$ in the energy eigenstates of the single channels Hamiltonians ${{{{\bf {\hat {H}}}}}_{sc}}$. The unitary transformation ${{{\bf {\hat {U}}}}}_{E-rep}$ is performed on ${{{{\bf {\hat {H}}}}}_R}$ to give $${{{\bf {\hat {H}}}}}_{E-rep}={{{\bf {\hat {U}}}}}^{\dag}_{E-rep} {{{\bf {\hat {H}}}}}_R {{{\bf {\hat {U}}}}}_{E-rep},$$ the Hamiltonian in the energy representation. The eigenvectors that were obtained as the solutions for the single channel problem serve as the transformation matrices between the two representations, so that: $${ {{{\bf {\hat {U}}}}}_{E-rep}} = \left[ {\begin{array}{*{20}c}
{{{\bf {\hat {U}}}}}^g_{E-rep} & 0 \\
0 & {{{\bf {\hat {U}}}}}^e_{E-rep} \\
\end{array}} \right]$$ where ${{{\bf {\hat {U}}}}}^{g/e}_{E-rep}$ are the matrices which contain the complete set of eigenvectors for the ground and excited states, $\left| {\psi^g_{E}(R) } \right\rangle$ and $\left| {\phi^e_{v,J}(R)} \right\rangle$, respectively.\
The coupled channel Hamiltonian in the energy states picture becomes:\
- [for the diagonal part in the excited channel subspace $$\left\langle { \phi_{v}^e \left( R \right)} \right|{{{\bf {\hat {H}}}}}\left| {\phi_{v}^e \left( R \right)} \right\rangle = E^{0e}_{v} + \Delta_{a} -i\gamma^{0e}_{v}, E^{0e}_{v} < 0 ,
\label{eq:hdiag_e}$$ where $\Delta_a$ is the detuning from the atomic resonance, and $\gamma^{0e}_{v}$ the natural width for the vibrational level $e,v$.]{}
- [for the diagonal matrix elements in the ground channel subspace: $$\left\langle { \psi_{n}^g \left( R \right)} \right|{{{\bf {\hat {H}}}}}\left| {\psi_{n}^g \left( R \right)} \right\rangle = E^{0g}_{n} .
\label{eq:hdiag_g}$$ For clarity, explicit $l$ dependency is ommited. We remark here again that $E^{0g}_{n}$ include both the bound and the quasi-continuum parts of the spectrum. $E^{0g}_{n}>0$ is therefore discrete, for a given box normalization.]{}\
- [ for the non-diagonal part of ${{{\bf {\hat {H}}}}}_{E-rep}$: $$\begin{aligned}
\hbar \Omega _{g,n}^{e,v} = \langle {\phi^{e}_{v} \left( R \right)} |{{{\bf {\hat {H}}}}}| {\psi _{n}^g \left( R \right)} \rangle
\label{eq:hcoupling}
\end{aligned}$$ We can employ the R-centroid approximation for ${{{\bf {\hat {\mu}}}}}_{eg}(R)$ and write: $$\hbar \Omega _{g,n}^{e,v} = {\bar \Omega} \left\langle {{\phi^{e,v} }}
\mathrel{\left | {\vphantom {{\phi^{e,v} } {\psi_{n}^g }}}
\right. \kern-\nulldelimiterspace}
{{\psi_{n}^g }} \right\rangle \equiv {\bar \Omega} {\cal \tilde F}_{g,n}^{e,v}
\label{eq:hcoupling2}$$ where now ${\cal \tilde F}_{g,n}^{e,v}$ denotes the Frank-Condon factor between vibrational levels from the two different electronic states.]{}
The advantage of the energy representation is the possibility to pick the energy levels that participate in the process and thus reduce the required basis set. We shall consider formally the case of an isolated resonance where the ground state continuum $\left| {\psi_{n}^g \left( R \right)} \right\rangle$ is coupled only to single bound level in the excited state $\left| {\phi_{v}^e \left( R \right)} \right\rangle$. The detuning from the atomic resonance $\Delta_{a}$ could be replaced by $\Delta$, the detuning from the excited bound level. A generalization to several bound excited states, or even a whole continuum, is straightforward. From this point on, we reduce all the indices for the excited state.\
By moving to the energy representation we reduce the representation for the coupled channels problem by about one half, to the order of $N \approx 1000$ in the example discussed below in section \[sec:sodium\].
### Getting out of the box - from $E$ to $E'$ representation.
Note that so far, the representation of the continuum is done using a box unity-normalized wave functions. This box-dependent basis for the continuum is arbitrary, and therefore it could be reduced.\
In order to do so, one had to defined a box-independent Frank-Condon factor. Box-independent quantities for the continuum are obtained by moving from the box-normalized to the energy-normalized wave functions. The relation between the two is given by: $$\left| {\tilde\psi_{n}^g \left( R \right)} \right\rangle = \left( {\frac{{\partial E}}{{\partial n}}} \right)^{ - 1/2} \left| {{\psi}_{n}^g \left( R \right)} \right\rangle
\label{eq:en_norm}$$ where the inverse of the level density, $\frac{{\partial E}}{{\partial n}}$, is given simply, for a large enough box, by the energy difference between two adjacent quasi-continuum levels, given under the box-normalization constraint. An energy normalized Frank-Condon profile: $${\cal F}^{e}(E) =\sqrt{{\frac{{\partial E}}{{\partial n}}}}{\cal \tilde F}_{g}^{e,n}
\label{eq:FC_profile}$$ could now be obtained. ${\cal F}^{e}(E)$ is a box-independent, numerically continuous function of the scattering energy. In fact, ${\cal F}^{e}(E)$ is a physical measure for the energy dependence of the interaction and it thus the function that one would like to sample for obtaining the minimal energy representation. The alternative basis has to capture the functional energy dependence of the Frank-Condon profile, using the smallest basis. This could be done in principle with any sampling procedure [@numerical_recipies]. The technical details regarding the numerical procedure we employed in this work is described at the appendix.\
Assuming such a minimal representation basis was found, we will denote it by $\left| g,n' \right\rangle$ for the ground state continuum, for $n'>n_B$, $n_B$ being the number of bound states. The expression for the quasi-continuum diagonal part of the full Hamiltonian (eq.\[eq:hdiag\_g\]) reads now: $$\left\langle { g,n'} \right|{{{\bf {\hat {H}}}}}\left| {g,n'} \right\rangle = E^{0g}_{n'},n'>n_B
\label{eq:althdiag_g}$$ and all the other diagonal parts remained unchanged. The nondiagonal part in the alternative basis is defined through the energy normalized Frank-Condon profile as: $$\hbar \Omega _{g,n'}^{e} = {\bar \Omega} {\cal \tilde F}_{g}^{e,n'} = {\bar \Omega} {\sqrt{{\frac{{\partial E}}{{\partial n'}}}}}^{-1}{\cal F}^{e}(E).
\label{eq:alt_nond}$$\
A diagonalization of the Hamiltonian (eqs.\[\[eq:althdiag\_g\],\[eq:alt\_nond\]\]) gives the set of eigenstates for the coupled Hamiltonian, $\left| {\Psi _j(E) } \right\rangle$, which include components on both channels, with possible contributions from all the bound or free levels. The energy eigenvalues which are obtained are again complex and $\gamma^{j} = \mathop{\rm Im} (E_j)$ is the width of the level $\left| {\Psi _j(E) } \right\rangle$. The transition probability from any initially populated state $\left| g,n' \right\rangle$ to any final state $\left| {\Psi _j(E) } \right\rangle$, is simply: $$P_{(g,n') \to j} = \left| {\left\langle {\Psi _j(E) }
\mathrel{\left | {\vphantom {{\Psi _j(E) } {g,n' }}}
\right. \kern-\nulldelimiterspace}
{g,n' } \right\rangle } \right|^2$$\
It is important to notice here that because $\left| {\Psi _j } \right\rangle$ is an eigenstate of the coupled Hamiltonian, a pair that reaches this state at the beginning of the final step of the experiment remains in the same state without any significant dynamics beside an exponential decay with the lifetime $\gamma^{j}$.\
Table 1 summarizes our procedure so far with the different representations and their typical necessary basis sizes. We started from the Hamiltonian on the equally spaced grid representation. A minimal representation of the problem with such a grid demands a basis of about $N_{R}=10^4$ function. We then used the phase space occupation of the problem to move to a mapped grid, and reduce the basis set in about an order on magnitude to $N_{x}=10^3$, for each of the electronic states. The transformation to the energy representation reduces the excited state manifold. The reduced $E'$ representation minimizes the basis for the ground state, so that finally we could solve the whole coupled channels problem with a basis at the order of $N_{E'}\sim 200$, or even smaller. At this sizes of basis the method become efficient, even compared to the semianalytic perturbative methods.\
Representation $N_g$ $N_e$
------------------------------------- ------------- --------
R - Equally spaced position grid $10^4$ $10^4$
x - Mapped position grid $10^3$ $10^3$
E - Energy eigenstates - boxed $10^3$ $<10$
E’ - Energy eigenstates - non boxed $\sim 10^2$ $<10$
: Summary of the different representations used in this paper. $N_g$ and $N_e$ are the typical basis set sizes for the ground and the excited electronic state, respectively, adapted to the example of section \[sec:sodium\].[]{data-label="param"}
In the next section the deduction of the various PA experimental parameters from $P_{(g,n') \to j}$ (and $\gamma^{j}$) is presented.
Extraction of the energy shifts, trap-loss rates, and lineshapes.
=================================================================
With the formal definitions of the previous subsection we can now move to define the various physical quantities that are being measured on the system. The probability to find an atom pair at the beginning of the stable period with energy $E_j$ (which we will denote as zero time) is given by: $$p(E_j,0) = N_{pair}(0)\sum\limits_{n'} { P_{(g,n') \to j}p_{B}(E_{n'})}=\frac{N^2_{atom}(0)}{2}\sum\limits_{n'} { P_{(g,n') \to j}p_{B}(E_{n'})}$$ where $N_{atom}(0)$ and $N_{pair}(0)$ are the number of atoms and atom pairs at time $0$, and $p_B$ is defined above (see eq. (\[boltz\])). As was explained at the last paragraph the only dynamics at the stable period are the exponential decay of the $j$-th level. At later times thus: $$p(E_j,t) = \exp(-\gamma_j t/\hbar) p(E_j,0)$$ so that the total probability density for the whole ensemble is: $$P_{tot}(t)= \sum\limits_{j} p(E_j,t) = \frac{N^2_{atom}(0)}{2} \sum\limits_{n',j} \exp(-\gamma_j t/\hbar) { P_{(g,n') \to j}p_{B}(E_{n'})}$$ The observed reaction rate $v(t)$ at any given time is the change of $P_{tot}(t)$ versus time: $$v(t) = -\frac{dP_{tot}(t)}{dt} = \frac{N^2_{atom}(0)}{2\hbar} \sum\limits_{n',j}\gamma_j \exp(-\gamma_j t/\hbar) { P_{(g,n') \to j}p_{B}(E_{n'})} \equiv \frac{N^2_{atom}(0)}{2\hbar} \left\langle {\gamma(t)} \right\rangle$$ The rate could be thus interpreted as the thermally averaged lifetime. On the other hand, the experimental definition of the diatomic reaction rate for a sample containing $N_{atoms}$ is: $$v(t)= - K \frac{N_{atoms}^2(t)}{V}.$$ Now, a comparison between the two expressions yields the relation between the calculated and experimental rate constants, which holds for any given time. Specifically for t=0 one gets: $$K(I,\Delta) = \frac{V \left\langle {\gamma(0)} \right\rangle}{2Z_{eq}} = \frac{\left\langle {\gamma(0)} \right\rangle}{2Q_T}
\label{our_K}$$ In a fashion similar to the experimental, a calculation of $K$ as a function of the detuning yields the lineshape. The peak value of $K$ as a function of $\Delta$, and the value of the detuning at the peak are termed as the maximum reaction rate $K_{max}$ and the energy shift $E_1$ for a given intensity and temperature, respectively.\
Before moving forward, we remark that the model is non perturbative with respect to the laser intensity. It is also well oriented to dynamical, time dependent-type calculations. Furthermore, a generalization of the method to multiple coupled continuum manifolds, several bound states, or to include hyperfine structure, is strightforward. Its drawback is mainly being a little less efficient computationally, mainly due to the expansive diagonalization step. Nevertheless, this could be overcomed, especially for calculating lineshapes and saturation effects, as we will show at the application of section VI. But, before doing so, we will present briefly in the next subsection the necessary expressions of the method of Bohn and Julienne[@bohn99] (B $\&$ J) to which our expression (\[our\_K\]) will be compared to.
Adjustment of Bohn and Julienne model to the grid methods framework
===================================================================
In this subsection we present the method of B $\&$ J [@bohn99], and adjust it to take advantage of the available grid methods. We will limit the treatment to the simple case of a single isolated resonance embedded in a continuum. The scattering-theory-based method defines a scattering amplitude $S_{E}$ for a pair of atoms in the center of mass framework. The square of $S_{E}$ is referred to as the probability to initiate in a unbound state with kinetic energy $E$ and be lost at infinite time, due to a light-induced coupling to a bound molecular state.
According to Fano[@bohn99; @fano61], $|S_{E}|^2$ is given by: $$|S_E|^2 = \frac{{\gamma \Omega_E }}{{\left( {E - \Delta - E_1 } \right) - \left(
{\frac{{\gamma + \Omega _E }}{2}} \right)2 }}.$$ here $\Delta$ is the detuning of the light frequency of the resonance bound state’s energy , and $\gamma$ is the natural linewidth of the bound level. $E_1$ and $\Omega_E$ are the energy shift and the stimulated absorption/emission rate, respectively. With the help of second order perturbation theory it was shown in [@fano61] that: $$E_1 = {\cal P}.\int {dE'\frac{{\left| {\Omega _{E'} } \right|^2 }}{{\Delta -
E'}}}
\label{fanoshift}$$ where ${\cal P}.$ stands for ‘the principle part of’, and $\Omega_E$ is the dipole moment coupling that was defined above (see eq.(\[eq:hcoupling\])).
The rate of loss from the system is given by performing a velocity thermal averaging of $|S_E|^2$, for all the initial scattering energies, and with respect to all the relevant partial waves, to obtain the observed rate loss: $$K = \sum\limits_l {\frac{{\pi v}}{k}\int {p_B \left( {E^l } \right)\left| {S_E^l } \right|^2 dE} }
\label{KBohn}$$ where the integral over the energy indicates for a thermal averaging over the Boltzmannian distribution $p_B$, defined above (see eq.\[boltz\]). The superscript for $S$ denotes the rotational quantum number for the partial wave $l$, and $v$ and $k=\sqrt(2\mu E)/\hbar$ are the relative velocity and the wavenumber of the atom-pair.
It is important to comment here, that in order to calculate the energy dependent $\Omega_E$ for eqs. (26-28) the energy dependency of the continuum has to be calculated \[see eq. (\[boltz\])\]. The common use of local solvers like the Numerov propagator [@numerical_recipies] for that purpose made Fano’s formalism impractical computationally. An energy independent $\Omega_E$ was assumed for demonstrative purposes in Ref. [@fano61]. Obviously, as one can see from figure 3 this is far from being adequate to describe molecular systems in any other circumstances.
Using the ideas of Du et al. [@du89] Bohn and Juliene combined the couple of regular and irregular solutions of eq.(\[eq:01\]) to write the energy shift $E_1$ as: $$E_1 = -2\pi \int\limits_0^\infty {{^r\Omega _E} \left( R \right)}dR
\int\limits_0^R {{^i\Omega _E} \left( {R'} \right)}dR'$$ where the superscripts denote the regular and irregular part of the continuum wavefunction for $E\to 0$, and $^{r/i}\Omega_E(R)$ is understood to be the integrand of the integral of eq.(\[eq:hcoupling\]). As noted in [@du89; @portier2006], this alternative expression could be viewed as a way to write the Green kernel ${{{\bf {\hat {G}}}}} = \left({{{\bf {\hat {H}}}}}-E\right)^{-1}$ in the position representation instead of the energy space used in Fano’s formalism.\
To reduce even further the effort in the computation of $\Omega _E$, the assumption of a slowly varying continuum with respect to $E$ is employed, i.e., $\lim_{E\to 0}\frac{d\Omega _E}{dE} \ll p_B(E\to 0)$. Under these conditions, the rate is assumed to be influenced by the on-resonance light interaction.
The continuum is represented now by only a single continuum state and the thermal averaging in eq.(\[KBohn\]) can be avoided. As shown in[@Bohn99], this leads to: $$K = \frac{h}{2\mu k}\frac{4\gamma\Omega(E_R)}{(\gamma + \Omega(E_R))2}
\label{Kapp}$$ where now the coupling $\Omega_E$ (defined above in eq.(\[eq:hcoupling\])) is taken only on the resonance energy $E_R$.
An analytic alternative way for calculating the energy dependence of $\Omega$ was suggested in Ref. [@crubellier2006]. In the latter work, strong energy dependence, was found in some cases, for instance $^{85}$Rb.
In this paper, we adopt the advantages of the grid methods in the E-representation, and calculate the explicit dependency of $\Omega(E)$ directly from the continuous energy normalized Frank-Condon profile with any desired accuracy for any energy range. This allows us to calculate $E_1$ directly using Fano’s formalism, without a need to have the irregular solution of the single channel Schrodinger equation. As noted in [@portier2006] the two expressions give the same results.\
At the next section a comparison between the various expressions will be demonstrated by a concrete application.
Application {#sec:sodium}
===========
The example chosen for numerical application will be the on-resonance photoassociation of the $v$=135 , $J$=1 level of Na$_{2}$ A$^{1}\Sigma_{u}^{+}$, in the experimental conditions of the NIST experiment [@mckenzie2002]. The initial state is a pair of colliding atoms with $l=0$ or $2$ (see figure \[fig:scheme\]).
![A schematic view of the CW laser PA experiment. Atoms pairs colliding in $S\, (l=0)$ and $D\, (l=2)$ partial waves on the ground electronic state, are coupled by a laser with constant frequency and intensity to the $v=135, P\, (J=1)$ bound level in the excited electronic state. The nearby bound levels on the excited state, as well as few bound levels within the ground electronic state could be coupled as well. The natural lifetime of the bound state $\gamma$ is $2\pi\times$ 18.36 MHz.[]{data-label="fig:scheme"}](fig1.eps){width="0.95\linewidth"}
The calculations have been performed using the same potentials as in Ref.[@naidon2006]. The position of the repulsive wall in the ground state X$^{1}\Sigma_{g}^{+}$ has been slightly modified, so that the binding energy of the last levels is $E(v=65)=0.013cm^{-1}$ for $l$=0 and $E(v=64)$=0.25cm$^{-1}$ for $l$=2. The experimental values for these levels are $E(v=65)=0.013$cm$^{-1}$ and $E(v=64)=0.25$cm$^{-1}$, respectively. In the excited curve, the level $v$=135 is bound by 49.23 cm$^{-1}$, the two neighboring levels being $v$=134 at -52.95 cm$^{-1}$ and $v$=136 at -45.76 cm$^{-1}$. Note that in experiment the binding energy of the resonant level is 43 cm$^{-1}$. The natural width is set to $\gamma$ = 2 $\pi \times$ 18.36 MHz to match the experiment.\
Figure \[fig:coupling\_bound\] presents the coupling profile, i.e., the Frank Condon factors multiplied by the Rabi frequency (see eq. \[eq:hcoupling2\]), for various bound levels $n$ of the ground state coupled to one level (v=135) of the excited state at intensity of 1 W/cm$^2$. In the main frame $n$ varies from 0 to 64 and oscillations in the coupling term are visible as the Condon point is moving to large internuclear distances . The FC factors are identical for both partial waves when $n \le 64$ . The $n = 65$ level is bound only for the S wave.\
At the upper and the lower panels of figure \[fig:coupling\_cont\] the coupling profile for the continuum part of the spectrum is shown, for both $S$ and $D$ partial wave, respectively. The left and the right panels show the same profile in logarithmic and linear scale. The last bound level $n = 65$ for the $S$ wave, is shifted for the $D$ wave and appears as a shape resonance with a well defined width in the $D$ continuum. For the continuum part of the spectrum, the Condon point is energy independent. The structure of the coupling profile of the continuum is a signature of the energy dependence of the continuum wave functions[@koch06]. The rotational barrier on the $D$ potential surface makes the $D$ profile to vanish more strongly near the threshold than does the $S$ profile.\
In the next subsections the results of grid method will be presented in three steps: First, the light induced energy shift will be calculated and discussed. Then, the results for the rest of the PA observables will be presented in the low intensity regime, where the perturbative method should be adequate. Finally, the results for in the high intensity regime, where the appearance of saturation effect are explored.\
Light Induced Energy Shifts
---------------------------
Using Fano’s perturbative expression for the energy shift (eq.\[fanoshift\]) and the right panels of figure \[fig:coupling\_cont\] it is understood that the energy scale which is needed for the calculation of the energy shift is larger in many order of magnitude then the thermal energy. (see the appendix for a numerical example). On the other hand, as will be shown below, the lineshape is influenced only by the coupling profile in the vicinity of the thermally populated energy. This separation of scales can help us furthermore in reducing the basis set, depending upon which physical quantity one would like to calculate.\
The light induced energy shift for intensities up to 100W/cm$^2$, are presented in figure \[fig:eshiftS\]. The energy shift was calculated using the $S$ continuum part of the ground state manifold. The green line is the energy shift that was calculated using the modified Bohn and Julienne perturbative method and is temperature independent by definition. In red, blue and black, we present the energy shift using our method for $T = 20 \mu K,2 \mu K$ and $200nK$, respectively. All the curves are linear as expected. A difference exists between the energy shifts, but it is too small to be measurable. The difference in the slopes between the present calculations and the perturbative method is also within the experimental accuracy. An important result we observed in our calculation is that the perturbative linearity of the energy shift with respect to the intensity was conserved for any feasible intensity that was checked (up to $100kW/cm^2$ and higher). Another feature of the perturbative limit is the additivity of the contributions from different partial waves with respect to the bound and the continuum. Figure \[fig:eshiftall\] presents the energy shift for the full interaction, i.e., bound and continuum part of the spectrum for both $S$ and $D$ partial waves. No deviations from the perturbative treatment were found with respect to the additivity.\
Low Intensity regime: the perturbative limit
--------------------------------------------
Saturation effects for the case of photoassociation in a sodium condensate were not observed up to 1.2kW/cm$^2$. A typical lineshape for an intensity of 100W/cm$^2$, well below the saturation limit, is presented in Figure \[fig:line\_low\]. The lines were all centered to zero detuning. The red line presents the rate versus the detuning for our method for T=500 nK. In black line the correspondent result for the Bohn and Julienne method with the same temperature is shown. The two lines show the same width but differ at the peak height by less then $10\%$, small comparable to the experimental accuacy. Moreover, as can be seen from the blue line in the figure, a small change of the temperature for the perturbative method (to $400nK$)reproduces exactly the same lineshape as in our method. Figure \[fig:K\_vs\_Ilow\] presents with the same colors of figure \[fig:line\_low\] the results for the maximum rate lineshape $K_{max}$ as a function of the intensity. All the curves are linear, as expected for the perturbative regime. Again, the agreement between the two methods is within the experimental accuracy.\
We have to remark here that, to our best knowledge, an extensive study of the temperature dependency of photoassociation lineshapes has never been performed. Figure \[fig:T\_dep\] presents the temperature dependency of $K_{max}$ for intensity of 100W/cm$^2$. The red and blue lines present the values for the thermal averaged rates of expressions (\[our\_K\],\[KBohn\]). The difference between the methods at the most common temperature for PA experiments in condensates, i.e., a few hundreds of nano Kelvins to micro Kelvins, is at the order the experimental uncertainty. The values deviate significantly from each other at relatively high ($T>10\mu K$) and low ($T<100nK$)temperatures. The black line in figure \[fig:T\_dep\] represents the values obtained from the perturbative method without energy averaging (see eq. \[Kapp\]). A single state approximation for the rate seems to be inadequate, at least for the description of most of the temperature dependence of th e trap loss. In addition, we checked the results obtained applying the perturbative model (\[KBohn\]) with thermal energy averaging but assuming a constant coupling $\Omega(E)$. The cyan and green lines show the values for $\Omega(E)$ taken at $k_bT$ and $0.5k_bT$, respectively. The constant coupling approximation fits perfectly to the fully thermally averaged model for the low temperature regime with $E=0.5k_bT$. Two factors play a role in this situation: the coupling profile (see fig. \[fig:coupling\_cont\]) and the Boltzmann thermal distribution (see eq. \[boltz\]). The first increases with $E$ while the last decreases. From the results it seems that in our case, the compensation of the two factors takes place at $E$=0.5$K_BT$. Above this temperature, the thermal width starts to play a role, and the constant coupling approximation is no longer valid.\
Non Perturbative regime - saturation effects
--------------------------------------------
We now move to explore the regime of high intensity. In figure \[fig:K\_vs\_Ihigh\_nK\] we present the intensity dependence of $K_{max}$ for $I$ up to 300 kW/cm$^2$ . The blue, black and red lines are the results for our method, the thermally averaged, and the non averaged perturbative method, respectively. The temperature is $200 nK$, typical for a condensate. All the lines show saturation behavior, but the saturation limit is not the same. The results for the single energy perturbative approximation are significantly lower then the other two methods. The intensities for reaching saturation are way too high to be observed experimentally, and, indeed, were never observed. The validity of the two-body model does not hold for such intensities, and therefore the significance of the results is mainly in predicting a saturation effect for other species where the saturation is reached at lower intensities, e.g., cesium[@kraft2005], lithium[@prodan2003] or strontium [@nagel2005].\
Figure \[fig:K\_vs\_Ihigh\_muK\] presents the same calculations as figure \[fig:K\_vs\_Ihigh\_nK\], but for $T = 20\mu K$, a typical temperature for a MOT. At a MOT temperature saturation is reached at lower intensities for all the models. The non-perturbative method presents an inverted region in which the peak rate constant decrease with the light intensity. This prediction has never been observed, and it might be feasible experimentally for species other then Sodium. The saturation for the case of cesium in a MOT may give clue to this inverted behavior[@kraft2005]\
A better understanding for the origin of the inverted behavior could be acquired by an inspection of the lineshapes near the threshold. Figure \[fig:line\_high\_us\] exhibits the lineshapes computed with the non-perturbative method around the saturation regime. The four lines corresponds to the intensities of 1,2,3 and 6 kW/cm$^2$ in blue, red, black and green, respectively. All the lines are centered to zero detuning for comparison. A shoulder at the positive tail of the Lorentzian lineshape starts to develop as saturation is approached. Beyond saturation, the shoulder takes over the Lorentzian lineshape at the positive tail of the line to create an unsymmetrical lineshape. The power broadening at the negative detuning tail differs from the positive one, and the unsymmetrical form of the lineshape is more pronounced with the increase of the intensity. Figure \[fig:line\_high\_BJ\] presents the lineshapes close to and beyond the saturation limit of the perturbative method, for $T=20 \mu K$. The lines are centered to zero detuning. Beside the obvious power broadening, the lineshapes remain symmetrical even for high intensities.\
As a first order perturbative method, Bohn and Julienne model takes into account only single path processes. That is, the rate constant measures the population flux from the unbound continuum levels to the bound level in the excited state. Any flux that is transfered to the bound level is lost at a rate proportional to the natural linewidth of the bound level. For low temperatures, the thermal energy width is narrow. The loss rate for the off resonance laser is determined only by the distance from the threshold. Positive and negative detuning are identical in that sense and the lineshape is therefore symmetric. For higher temperatures, the population spread is larger and the inner structure of the continuum start to play a role. This is the origin for the deviation from a Lorentzian lineshapes which presented in figure 5 of [@bohn99], for temperatures of $1mK$. This non symmetric form is expected to be washed out at low temperatures and high intensities, where the thermal wi dth is much smaller then the radiative width.\
In the non-perturbative method, all paths are being taken into account. At large intensity, multi-paths processes can take place. A flux could bounce from the bound level back to the continuum and vice verse. As can be seen in figure \[fig:coupling\_cont\], the coupling profile indicates for a maximum of the interaction around $10^{-2}$cm$^{-1}$. The shoulder of the lineshape appears at a detuning equal to the maximum. Another support for this interpretation could be found in the energy dependence of the transition probability, $P_{(g,n') \to j}$. At low intensity single path processes take place and the transition probability is significant only close to the threshold. At large intensities other parts of the profile play a role, $P_{(g,n') \to j}$ deviates from a symmetric Lorentzian and peaks also at the energy of the maximum of the interaction profile. A further increase of the intensity enhances even further the weight of the non-linear effects on the line shape, and so the lineshape at positive detuning is rolled by non linear effects. A clue to support this result could be seen maybe at the apparent unsymmetrical lineshapes that were measured for the saturation limit of cesium in a MOT(see [@kraft2005]).
Discussion and Summary
======================
In the present paper we have explored the advantage of grid methods for performing a non-perturbative calculation of photoassociation lineshapes, energy shifts and rates. Several physical properties determine the energy scales that are involved in the PA experiment and characterize the outcomes of the various observables. The energy shift, which is the value of the detuning at which a photoassociation lineshape peaks, is the parameter which is the less sensitive to the structure of the interaction profile and the thermal population. In order to calculate it accurately one has to consider an interaction energy which goes up to several hundreds of Kelvins and to include all the manifolds which participate in the interaction with the bound level. The interaction energy is independent of the thermal occupation of these levels. A full non-perturbative treatment for the energy shift is therefore extremely difficult, but fortunately, it is also unnecessary. The advantage of the grid methods in calculating energy shifts is minor and they could be obtained efficiently and accurately by perturbative methods.\
For low laser intensities, the natural linewidth of the bound level, which is usually at the order of hundred of MHz, or tens of milliKelvin, is the next energy scale in the hierarchy. The width of the lineshape and the scaling of the peak height are determined by this parameter at low intensities. As the intensity of the laser increases, the radiative coupling, or the Rabi frequency of the given scattering energy $\Omega(E)$ begin to dominates, and lead to power broadening and saturation.\
Notation Parameter Order of magnitude(Kelvins)
------------ ---------------------------------- -----------------------------
$E_{I}$ Energy shifts interaction energy $ 10^1 - 10^3$
$\gamma$ Natural linewidth $10^{-1}-10^{-3}$
$\Omega_E$ Stimulated emission rate Depends on $\Delta$ and $I$
$E_{T}$ Thermal Energy $10^{-7}-10^{-4}$
: Parameters involve in PA and their typical energy scales in Kelvins.
The thermal energy is commonly the smallest energy scale. It ranges from about a hundred of nanokelvins for the coldest condensates, to a few hundreds microkelvins, for ‘hot’ MOTs. For low temperatures, the narrow width of the populated energy band makes the coupling profile to be almost insignificant. Saturation could then be described quite well with the perturbative methods. As the temperature increases, the inner structure of the continuum is encrypted into deviations of the lineshape from a symmetric Lorentzian, and an inverted region of $K_{max}$ as the intensity increases. Some of these deviations, especially the ones which are visible at the low intensity limit, for relatively high temperatures, could be well described by the single-path perturbative models. Other features which result only at high intensities, could be calculated only with the help of the multi-path model.\
The non perturbative model that we presented in this paper can serve as a useful tool to model easily other cases where non trivial lineshapes appear. These include:
- [Photoassociation close to threshold.]{}
- [Accidentally degenerate or almost degenerate levels that originate from different electronic states. In both cases, interaction between several levels in the excited state need to be put in the framework of a higher order model. The case of resonant coupling photoassociation [@slava00a; @dion01; @kerman04; @bergeman2006] is a good example.]{}
- [Active interaction between multiple continuum manifolds, where more than a single continuum is populated or depopulated during the process. One possibility is when more then one partial wave participates in the scattering, e.g., the $D$ resonance in $^85$Rb (see ref. [@crubellier2006]). In cases of lower barriers, higher temperatures, or in the cases of processes that take place by explicitly populating the continuum, highly structured lineshapes might be resolved by our method.]{}
Finally, the dynamical approach which is the basis for the present study is an important building block for a future comprehensive time dependent description of the many body interaction in the description of PA.
Acknowledgments {#acknowledgments .unnumbered}
---------------
This work was supported by the Israel Science Foundation and by the European Commission in the frame of the Cold Molecule Research Network under Contract No. HPRN-CT-2002-00290. The Fritz Haber Center is supporter by the Minerva Gesellschaft fur die Forschung GmbH Munchen, Germany.
[10]{}
H. R. Thorsheim, J. Weiner, and P. S. Julienne. “Laser-induced photoassociation of ultracold sodium atoms.” Phys. Rev. Lett. [**58**]{}, 2420 (1987).
K. M. Jones, E. Tiesinga, P. D. Lett, and P. S. Julienne. “Ultracold photoassociation spectroscopy: long range molecules and atomic scattering.” Rev. Mod. Phys. [**78**]{}, 483 (2006).
P. Pillet, A. Crubellier, A. Bleton, O. Dulieu, P. Nosbaum, I. Mourachko, and F. Masnou-Seeuws. “Photoassociation in a gas of cold alkali atoms. [I: P]{}erturbative quantum approach.” J. Phys. B [**30**]{}, 2801 (1997).
R. Napolitano. “A two-state model for controlling scattering lengths and photoassociation spectral line shapes alkali-metal atoms by resonant light in the regime of finite ultracold temperature.” Brazilian Journal of Physics [**27**]{}, 162 (1997).
R. Côté and A. Dalgarno. “Photoassociation intensities and radiative trap loss in lithium.” Phys. Rev. A [**58**]{}, 498 (1998).
J. L. Bohn and P. S. Julienne. “Semianalytic theory of laser-assisted resonant cold collisions.” Phys. Rev. A [**60**]{}, 414 (1999).
E. Taylor-Juarros, R. Côté, and K. Kirby. “Formation of ultracold polar molecules via raman excitation.” Eur. Phys. J. D [**31**]{}, 213 (2004).
J. Javanainen and M. Mackie. “Rate limit for photoassociation of a [Bose-Einstein]{} condensate.” Phys. Rev. Lett. [**88**]{}, 090403 (2002).
C. McKenzie, J. H. Denschlag, H. H[ä]{}ffner, A. Browaeys, L. E. de Araujo, F. K. Fatemi, K. M. Jones, J. E. Simsarian, D. Cho, A. Simoni, E. Tiesinga, P. S. Julienne, K. Helmerson, P. D. Lett, S. L. Rolston, and W. D. Phillips. “Photoassociation of sodium in a [Bose-Einstein]{} condensate.” Phys. Rev. Lett. [**88**]{}, 120403 (march 2002).
I. D. Prodan, M. Pichler, M. Junker, R. J. Hulet, and J. L. Bohn. “Intensity dependence of photoassociation in a quantum degenerate atomic gas.” Phys. Rev. Lett. [**91**]{}, 080402 (august 2003).
S. D. Kraft, M. Mudrich, M. U. Staudt, J. Lange, O. Dulieu, R. Wester, and M. Weidemuller “Saturation of [Cs$_2$]{} photoassociation in an optical dipole trap.” Phys. Rev. A [**71**]{}, 013417 (2005).
P. Naidon and F. Masnou-Seeuws. “Photoassociation and optical Feshbach resonances in an atomic [Bose-Einstein]{} condensate: treatment of correlation effects.” Phys. Rev. A [**73**]{}, 043611 (april 2006).
T. Köhler, T. Gasenzer, and K. Burnett. “Microscopic theory of atom-molecule oscillations in a [Bose-Einstein]{} condensate.” Phys. Rev. A [**67**]{}, 013601 (2003).
P. Naidon and F. Masnou-Seeuws. “Pair dynamics in the formation of molecules in a [B]{}ose-[E]{}instein condensate.” Phys. Rev. A [**68**]{}, 033612.
R. Kosloff. “[Quantum Molecular Dynamics on Grids]{}.” In R. E. Wyatt and J. Z. Zhang, editors, Dynamics of Molecules and Chemical Reactions, pages 185-230, Marcel Dekker, New York (1996).
E. Fattal, R. Baer and R. Kosloff, “[Phase Space Approach for Optimizing Grid Representation: The Mapped Fourier Method]{}.” Phys. Rev. E [**53**]{}, 1217 (1996).
V. Kokoouline, O. Dulieu, R. Kosloff, and F. Masnou-Seeuws. “[Mapped Fourier methods for long range molecules: Application to perturbations in the [Rb$_2(0_u^+$)]{} spectrum]{}.” J. Chem. Phys. [**110**]{}, 9865 (1999).
K. Willner, O. Dulieu, and F. Masnou-Seeuws. “A mapped sine grid method for long range molecules and cold collisions.” J. Chem. Phys. [**120**]{}, 548 (2004).
C. M. Dion, C. Drag, O. Dulieu, [Laburthe Tolra, B.]{}, F. Masnou-Seeuws, and P. Pillet. “Resonant coupling in the formation of ultracold ground state molecules via photoassociation.” Phys. Rev. Lett. [**86**]{}, 2253 (2001).
A. J. Kerman, J. M. Sage, S. Sainis, T. Bergeman, and D. DeMille. “Photoassociation and state-selective detection of ultracold $RbCs$ molecules.” Phys. Rev. Lett. [**92**]{}, 153001 (2004).
S. Azizi, M. Aymar, and O. Dulieu. “Prospects for the formation of ultracold ground state polar molecules from mixed alkali atom pairs.” Eur. Phys. J. D [**31**]{}, 195 (december 2004).
J. M. Sage, S. Sainis, T. Bergeman, and D. DeMille. “Optical production of ultracold polar molecules.” Phys. Rev. Lett. [**94**]{}, 203001 (2005).
V. Kokoouline, O. Dulieu, and F. Masnou-Seeuws. “Theoretical treatment of channel mixing in excited [Rb$_2$]{} and [Cs$_2$]{} ultra-cold molecules. Perturbations in $0_u^+$ photoassociation and fluorescence spectra.” Phys. Rev. A [**62**]{}, 022504 (2000).
V. Kokoouline, O. Dulieu, R. Kosloff, and F. Masnou-Seeuws. “Theoretical treatment of channel mixing in excited [Rb$_2$]{} and [Cs$_2$]{} ultra-cold molecules : determination of predissociation lifetimes with coordinate mapping.” Phys. Rev. A [**62**]{}, 032716 (2000).
T. Bergeman, J. Qi, D. Wang, Y. Huang, H. K. Pechkis, E. E. Eyler, P. L. Gould, W. C. Stwalley, R. A. Cline, J. D. Miller, and D. J. Heinzen. “Photoassociation of $^{85}$[Rb]{} atoms into 0$_u^+$ states near the 5[S]{} +5[P]{} atomic limit.” J. Phys. B: At. Mol. Opt. Phys. [**39**]{}, S813 (2006).
C. Lisdat, O. Dulieu, H.Knockel, and E. Tiemann. “Inversion analysis of [K$_2$]{} coupled electronic states with the [F]{}ourier grid method.” Eur. Phys. J. D [**17**]{}, 319 (2001).
P. Pellegrini. . Thése de doctorat (Ph.D thesis), Université Paris-Sud, Centre d’Orsay (2003).
P. Naidon. . Thése de doctorat (ph.d thesis), Université Pierre et Marie Curie Paris VI (2004).
M. Portier, S. Moal, J. Kim, M. Leduc, C. Cohen-Tannoudji, and O. Dulieu. “Analysis of light-induced frequency shifts in the photoassociation of ultracold metastable helium atoms.” J. Phys. B: At. Mol. Opt. Phys. [**39**]{} S881 (2006).
C. P. Koch, R. Kosloff, E. Luc-Koenig, F. Masnou-Seeuws, and A. Crubellier “Photoassociation with chirped laser pulses : Calculation of the absolute number of molecules per pulse .” J. Phys. B: At. Mol. Opt. Phys. , [**39**]{} S1017 (2006).
M. Nest, H.-D. Meyer. Chem. Phys. Lett. [**352**]{} 486 (2002).
F. Laloe. Claude Cohen-Tannoudji, Bernard Diu. . (Hermann, Paris, 1973).
W.H. Press, S.A. Teukolsky, W.T. Vetterling, and B.P. Flannery. . (Second edition, Cambridge University Press, 1995).
U. Fano. “Effect of configuration interaction on intensities and phaseshifts.” Phys. Rev. [**124**]{}, 1866 (1961).
M. L. Du, A. Dalgarno, and M. J. Jamieson. “Level shifts of discrete states embedded in a continuum.” J. Chem. Phys. [**5**]{}, 2980 (1991).
J. L. Bohn and P. S. Julliene Phys. Rev. A [**60**]{}, 414 (1999).
A. Crubellier and E. Luc-Koenig. “Threshold effects in the photoassociation of cold atoms : [$R^{-6}$]{} model in the milne formalism.” J. Phys. B [**39**]{}, 1417 (2006).
S. B. Nagel, P. G. Mickelson, A. D. Saenz, Y. N. Martinez, Y. C. Chen, T. C. Killian, P. Pellegrini, and R. Côté “Photoassociative Spectroscopy at Long Range in Ultracold Strontium.” Phys. Rev. Lett. [**94**]{}, 083004 (2005).
APPENDIX {#appendix .unnumbered}
========
Two purposes are the objects of this appendix: the numerical details of the sampling of the continuum profile ${\cal F}^e (E)$ (see eq. (\[eq:FC\_profile\])), and the demonstration of the energy scale that is needed for the calculation of the energy shift. The sampling of the continuum profile ${\cal F}^e (E)$ could be done in principle by a global or local numerical sampling procedure (see for example [@numerical_recipies]).The common procedures take advantage of an a-priori knowledge about the given function to be sampled. In this work we limited ourselves to a naive approach and determined the sampling of the profile a-posteriori according to the results that were obtained with a given basis. The criterion we employed was the numerical convergence of the integral over a given moment of the interaction profile ${\cal F}^e (E)$: $${\cal{I}} = \int\limits_{0}^{E_{cut}} { {[{\cal F}^e (E)]^n dE}} \approx \sum_j \Delta(E_j)[{\cal F}^e (E_j)]^n$$ where $n$ is the order of the moment, $E_j$ is the discrete sampling energy points, and $\Delta(E_j)$ is the energy difference between two adjacent points. The upper bound for the profile $E_{cut}$ is discussed below. The procedure initiates with a small set of points, and a new point is added or rejected between any two old points according to its influence on the value of ${\cal{I}}$. The quality of the procedure is determined by the tolerance parameter $\tau = \left| {\cal{I}}^{new}/{\cal{I}}^{old}-1 \right|$. The number of basis functions for the sampling the $S$ continuum, as a function of the tolerance parameter is presented in figure A1. The various lines present different moments.\
As can be understood, the convergence of the sampling becomes more efficient for higher moments, which emphasize more the features in the profile. Too high moments, however, smear completely the functional structure and so an optimum moment have to be found. We have to comment here that in order to increase the resolution of the basis set with respect to the thermally populated energy near the threshold (for $E < Nk_{B}T$, N is usually 10), we inserted a small energy grid (typically 50-100 basis functions) with constant energy step at close-to-threshold energies.\
In figure A2 the resulted energy shifts are plotted as a function of the tolerance parameter for the various moment orders. The numerically exact value which was calculated using the perturbative method is $27.8 MHz$. The values for moment order above the $5^{th}$, convergence to this values within less then 5% for tolerance value of $5\times 10^{-5}$. The basis set that was taken in this paper uses $\tau = 5\times 10^{-7}$ and $n = 9$. Larger moments than the chosen exhibit slower convergence and demand much smaller values of $\tau$ to produce an adequate basis set.\
We must remark also, that the only physical parameter which demands such a care in the sampling is the energy shift. The values for the lifetime are much less influenced by the size of the basis, and converged even at $\tau$ as small as $1\times10^{-3}$. To demonstrate the importance of very large energy scale that are needed to calculate the energy shift correctly we present in figure A3 the calculated energy shift as a function of $E_{cut}$ the upper bound of the energy grid. Convergence of the energy shift is achieved only when reaching hundreds of Kelvins. Note, again, that only the energy shift exhibit a sensitivity for such high energies. For low laser intensities the lifetime is found to converge $E_{cut}$ at the order of the thermal energy grid. Convergence near the saturation limit demands extension of the grid up to the first minimum of the coupling profile, i.e.,$10^{-2}cm^{-1}$, or tens of milikelvins. A basis set for such energies contained less then 150 function and is adequate for calculating most of the lineshapes features, except for energy shifts.\
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Published dielectric and shear data of six molecular glass formers and one polymer are evaluated in terms of a spectrum of thermally activated processes, with the same barrier density for the retardation spectrum of shear and dielectrics. The viscosity, an independent parameter of the fit, seems to be related to the high-barrier cutoff time of the dielectric signal, in accordance with the idea of a renewal of the relaxing entities after this critical time. In the five cases where one can fit accurately, the temperature dependence of the high-barrier cutoff follows the shoving model. The Johari-Goldstein peaks, seen in four of our seven cases, are describable in terms of gaussians in the barrier density, superimposed on the high-frequency tail of the $\alpha$-process. Dielectric and shear measurements of the same substance find the same peak positions and widths of these gaussians, but in general a different weight.'
author:
- 'U. Buchenau'
date: 'April 2, 2007'
title: Mechanical and dielectric relaxation spectra in seven highly viscous glass formers
---
Introduction
============
The publications of Kia Ngai deal with more substances and more measurement techniques than the work of any other scientist in the field of undercooled liquids. Within the past three decades, whenever a new development appeared, he was the quickest to appreciate it, analyze it and bring it to the general attention, thus speeding up the progress substantially.
Many scientists in the field share his conviction that the flow process in highly viscous liquids can only be understood by combining all possible techniques for its study [@birge; @ngai; @chang; @donth; @bow; @bzow]. The present paper evaluates recently published [@niss; @jakobsen] broadband shear and dielectric relaxation data on seven glass formers. The mechanical shear data were obtained with a new technique [@christensen] which allows to cover a large dynamical range. Samples for dielectric and shear measurements were taken from the same charge, and the temperature sensors of both measurements were calibrated to each other.
The data show a striking similarity of $G''(\omega)$ and $\epsilon''(\omega)$ on the right hand side of the $\alpha$-peak, a similarity which is sometimes perturbed by the secondary Johari-Goldstein peak [@johari1; @johari2]. The similarity suggests a common origin of the $\alpha$-peak in dielectrics and shear. The fact that the shear peak appears at a higher frequency than the dielectric peak is explainable in terms of the viscosity, which in a compliance treatment [@ferry] is a free parameter.
In order to identify the elementary processes with thermally activated jumps over an energy barrier $V$, one can use a recent translation [@fv] of the textbook [@ferry] retardation spectrum $L(\ln\tau)$ into a barrier density function $l(V)$. We will argue that the compliance barrier density of the shear equals the electric dipole moment barrier density $l_\epsilon(V)$ of the dielectric data.
The next section (section II) explains and motivates this approach in more detail. The results of the data treatment are described in section III. They are discussed and compared to other approaches in section IV. Section V summarizes and concludes the paper.
The barrier density functions for shear and dielectrics
=======================================================
The choice of a retardation spectrum for the shear is motivated by a surprising coincidence, which is more or less visible in the data of all seven substances. We show the two examples where it is most clearly seen in Fig. 1 and Fig. 2.
Fig. 1 compares $G''(\omega)$ and $-\epsilon''(\omega)$ at the same temperature for the type-A glass former PPE. PPE, 1,3-Bis(3-phenoxyphenoxy)benzene, is a diffusion pump oil with the commercial name Santovac-5P, a molecule consisting of five phenyl rings connected by four oxygens to form a short chain. In terms of the classification proposed by the Bayreuth group [@kudlik], it is a type-A glass former, a glass former which shows no or at least no pronounced secondary Johari-Goldstein peak. The (negative) $\epsilon''(\omega)$-data have been scaled to coincide with the $G''(\omega)$-data on the right hand side of the peak. One finds good agreement between $G''(\omega)$ and $-\epsilon''(\omega)$ as soon as the frequency is two decades higher than the one of the peak in $G''(\omega)$.
In addition, Fig. 1 shows a change of slope of the $\alpha$-tail, from an $\omega^{-1/2}$- to an $\omega^{-1/3}$-behavior. The tendency is seen most clearly in the dielectric data, but it is also in the shear data; their fit improves markedly if one allows for a $\omega^{-1/3}$-component. This might be the influence of a hidden Johari-Goldstein peak, an explanation which has been favored for glycerol on the basis of pressure, aging and chemical series measurements (for a review, see Ngai and Paluch [@paluch]). But there is also impressive experimental evidence for a limiting $\omega^{-1/3}$-behavior in shear compliance data of type-A molecular glass formers [@plazek], which show the so-called Andrade [@andrade] creep, $J(t)\sim t^{1/3}$, in the short-time limit. Therefore we will fit our data in terms of a sum of an $\omega^{-\beta}$-term (with $\beta$ as fit parameter) and an $\omega^{-1/3}$-term, dominating at high frequency.
Toward lower frequency, the common slope terminates for the shear data already at a higher frequency (a shorter time) than for the dielectric data. The natural explanation for this is that the parallel between dielectrics and shear is in fact between shear compliance and dielectric susceptibility. In a comparison of these two quantities, the shear compliance starts to deviate from the dielectric susceptibility as soon as the viscous flow sets in. This suggests a treatment of the shear data in terms of a retardation spectrum, with the viscosity as an independent parameter [@ferry].
The good agreement at the right-hand side of the $\alpha$-peak is found to be a general feature of all seven measured substances, as long as there is no disturbing influence from the secondary Johari-Goldstein peak [@johari1; @johari2]. This is seen in Fig. 2, which shows that the good agreement between $G''$ and $-\epsilon''$ disappears as the peaks merge. The substance, tripropylene glycol (TPG), $C_6H_{20}O_4$, is still a molecule and not yet a polymer (the whole series from the small molecule propylene glycol to long-chain polypropylene glycol is well-investigated by dielectrics [@leon; @lars1; @lars2]). TPG itself has been studied under aging [@dyretpg] and under pressure [@pawlus]. It consists of three connected propylene groups, with a large dielectric moment and a pronounced Johari-Goldstein peak (which is absent or at least much less pronounced in propylene glycol [@leon]).
Fig. 2 shows another tendency which will be evaluated quantitatively in this paper, namely a much stronger increase of the imaginary quantities in the $\alpha$-peak region with temperature than in the Johari-Goldstein peak region.
If one wants to decompose a measured relaxation into a spectrum of exponential decays in time, one can choose between two equivalent possibilities [@ferry], the relaxation spectrum in which the elementary exponential relaxators add to decrease the modulus or the retardation spectrum in which they add to increase a susceptibility. In principle, the choice is not crucial, because the two spectra can be calculated from each other. Here, we choose the retardation spectrum, with the viscosity $\eta$ as an independent variable.
For this choice, one has the textbook expressions [@ferry] for the real and imaginary parts of the complex frequency-dependent shear compliance $$\label{jpferry}
J'(\omega)=J_g+\int_{-\infty}^\infty L(\ln\tau)\frac{1}{1+\omega^2\tau^2} d(\ln\tau)$$ and $$\label{jppferry}
J''(\omega)=\int_{-\infty}^\infty L(\ln\tau)\frac{\omega\tau}{1+\omega^2\tau^2} d(\ln\tau)+\frac{1}{\omega\eta},$$ where $\tau$ is the relaxation time and $L(\ln\tau)$ is the weight of this relaxation time in the retardation spectrum. $J_g$, the glass compliance, is the inverse of the infinite frequency modulus $G_\infty$.
In an energy landscape picture [@goldstein], one reckons with thermally activated jumps over the energy barrier between two neighboring minima. In fact, one very often finds a broad secondary relaxation peak (the Johari-Goldstein peak [@johari1; @johari2]) below the $\alpha$-peak of the flow process. This peak follows the Arrhenius relation in the glass phase, indicating that it stems from local thermally activated jumps. For a jump over an energy barrier of height $V$, the Arrhenius relation for the relaxation time $\tau_V$ reads $$\tau_V=\tau_0{\rm e}^{V/k_BT},$$ where $\tau_0=10^{-13}$ s and $T$ is the temperature.
For a spectrum of thermally activated jumps, one defines [@fv] the barrier density function $l_s(V)$ $$l_s(V)=\frac{G_\infty}{k_BT} L(V/k_BT+\ln\tau_0).$$ The index $s$ stands for the shear. With this definition, the complex shear compliance equations (\[jpferry\]) and (\[jppferry\]) transform into $$\label{jp}
J'(\omega)=J_g+J_g\int_0^\infty l_s(V)\frac{1}{1+\omega^2\tau_V^2} dV$$ and $$\label{jpp}
J''(\omega)=J_g\int_0^\infty l_s(V)\frac{\omega\tau_V}{1+\omega^2\tau_V^2} dV+\frac{1}{\omega\eta}.$$
The dielectric susceptibility can also be described [@pollak; @gilroy] in terms of a dielectric barrier density function $l_\epsilon(V)$ $$\label{ep}
\frac{\epsilon'(\omega)-\epsilon_\infty}{\epsilon(0)-\epsilon_\infty}=\int_0^\infty l_\epsilon(V)\frac{1}{1+\omega^2\tau_V^2} dV$$ and $$\label{epp}
\frac{\epsilon''(\omega)}{\epsilon(0)-\epsilon_\infty}=\int_0^\infty l_\epsilon(V)\frac{\omega\tau_V}{1+\omega^2\tau_V^2} dV.$$ Here $\epsilon(0)$ is the static dielectric susceptibility, $\epsilon_\infty$ is the real part of $\epsilon(\omega)$ in the GHz range (larger than $n^2$, the square of the refractive index, because of vibrational contributions [@bzow]).
The above definitions of eqs. (\[jp\]-\[epp\]) imply a normalization of both $l_s(V)$ and $l_\epsilon(V)$ with $$\label{snorm}
\int_0^\infty l_s(V)dV=\frac{J_e^0-J_g}{J_g},$$ where $J_e^0$ is the recoverable compliance of the steady-state flow [@ferry], and $$\label{enorm}
\int_0^\infty l_\epsilon(V)dV=1.$$
The dielectric $\alpha$-peak occurs always at a lower frequency than the shear one and seems to coincide with the heat capacity and the structural relaxation peaks [@birge; @ngai; @chang; @donth; @bow; @bzow]. Below, we will adopt the view that the left side of the dielectric peak marks the disappearance and renewal of the relaxing entities.
In order to describe this decay, one needs to multiply the barrier density of the energy landscape with an appropriate cutoff function at a cutoff barrier $V_c$. Here, we will assume that the relaxing entities decay exponentially in time with the critical relaxation time $\tau_c$. With the Arrhenius relation $\tau_c=\tau_0\exp(V_c/k_BT)$, this translates into a double-exponential cutoff $$c(V)=\exp(-\exp((V-V_c)/k_BT)).$$
Equations (\[jpp\]) and (\[epp\]) show that a Johari-Goldstein peak in $G''(\omega)$ or $\epsilon''(\omega)$ at the peak frequency $\omega_1$ corresponds to a peak in $l(V)$ at a peak barrier $V_1=k_BT\ln(1/\omega_1\tau_0)$. We will see that the Johari-Goldstein peaks are reasonably well described by gaussians in $l(V)$.
To describe both the $\alpha$-peak and the Johari-Goldstein-peak in terms of a barrier density, $l_s(V)$ and $l_\epsilon(V)$ will be fitted by the form $$\label{lvform}
l(V)=(a_\beta {\rm e}^{\beta V/k_BT}+a_{1/3}{\rm e}^{V/3k_BT}+a_1{\rm e}^{-\gamma_1(V-V_1)^2})c(V).$$ The first two terms describe the high-frequency tail of the $\alpha$-process, the third term the Johari-Goldstein peak (if there is one; in three of our seven examples, it is not needed). The dimensionless parameter $\beta$ determines the slope $\omega^{-\beta}$ at the beginning of the $\alpha$-tail in the double-log plot of Fig. 1.
Instead of using the three prefactors $a_\beta$, $a_{1/3}$ and $a_1$ as fit parameters, it is better to use the corresponding weights $w_\beta$, $w_{1/3}$ and $w_1$ in the integral over the barriers, equs. (\[snorm\]) and (\[enorm\]). A type-A glass former without Johari-Goldstein peak with $w_1=0$ is characterized by the two dimensionless parameters $\beta$ and $b_2=w_{1/3}/(w_\beta+w_{1/3})$, at least as far as the form of its spectrum is concerned. $\beta$ and $b_2$ have the advantage to be reasonably temperature-independent.
With this prescription, one can fit the $\epsilon''(\omega)$ of a type-A glass former with two temperature-independent parameters, $\beta$ and $b_2$, and two temperature-dependent parameters, $\Delta\epsilon=\epsilon(0)-\epsilon_\infty$ and $V_c$. Their temperature dependence is a decrease with increasing temperature, which is well fitted by an appropriate power law $$\label{tscal}
\Delta\epsilon(T)=\Delta\epsilon(T_g)\left(\frac{T_g}{T}\right)^{\gamma_\epsilon},$$ where $T_g$ is the glass temperature. Similarly, one describes the decrease of $V_c$ with the exponent $\gamma_V$ and the one of $G_\infty$ with $\gamma_G$.
The strategy of our evaluation is to fit $l(V)$ to the dielectric data, and then use the same spectral form to describe the shear. The fit of the shear data requires three additional temperature-dependent parameters, $J_g$, $J_e^0$ and $\eta$. Again, it is worthwhile to look for combinations which might turn out to be temperature-independent. One of them is the ratio $$\label{f0}
f_0=\frac{J_e^0-J_g}{J_g},$$ which appears in the normalization of the shear spectrum, eq. (\[snorm\]). A second interesting possibility is not to fit the viscosity $\eta$, but the ratio $$\label{fjc}
f_{jc}=\frac{\tau_c}{f_0J_g\eta},$$ where $\tau_c$ is the Arrhenius relaxation time of the terminal barrier $V_c$. As we will show in the discussion, one can argue that the ratio $f_{jc}$ should be 2 for a renewal of the relaxing entities within the critical time $\tau_c$.
In the case of a type-B glass former, Fig. 2 shows that one needs another dimensionless parameter, because the weight of the Johari-Goldstein peak is different in the two quantities. In Fig. 2, the Johari-Goldstein peak is more prominent in the shear signal, but this varies from substance to substance.
Data evaluation
===============
The three type-A glass formers
------------------------------
Three of our seven substances, TPE, DC704 and PPE, happen to have no or at least only a rather weak Johari-Goldstein peak. Let us begin with TPE. TPE stands for triphenylethylene, $C_{20}H_{16}$, a rather flexible molecule with three phenyl rings attached to a central $C=C$ double bond. Fig. 3 (a) shows data and fit for $\epsilon''(\omega)$ in a double-log plot, Fig. 3 (b) the ones for $G(\omega)$. The dielectric data in Fig. 3 (a) are well fitted with only the first two terms of eq. (\[lvform\]), without any Johari-Goldstein peak. $\beta$ and $b_2$ turn out to be temperature-independent within experimental accuracy.
One gets a good fit for the shear data in Fig. 3 (b), taking over $\beta$, $b_2$ and the cutoff barrier $V_c$ from the fit of the dielectric data at the given temperature and fitting $G_\infty$, $f_0$ and $f_{jc}$. $G_\infty$ is temperature-dependent, but $f_0$ and $f_{jc}$ are again temperature-independent within the experimental accuracy, thus justifying our choice of variables. The parameters and their temperature dependence are listed in Table I.
The temperature exponents $\gamma_V$ and $\gamma_G$ of the critical barrier $V_c$ and the infinite frequency shear modulus $G_\infty$ are the same within their error bars (about 5 % for $\gamma_V$ and about 10 % for $\gamma_G$). This shows the validity of the shoving model [@shoving], according to which the energy barrier of the $\alpha$-process should be proportional to the infinite frequency shear modulus $G_\infty$. The shoving model postulates that the $\alpha$-process happens when the local energy concentration exceeds the product $G_\infty\Delta_v$, where $\Delta_v$ is a volume expansion.
The same results, maybe even a bit clearer because of the stronger dielectric signals, are obtained for the two other type-A glass formers PPE and DC704. Again, the fit parameters are listed in Table I. In particular, the $\omega^{-1/3}$-contribution is much better seen, as illustrated in Fig. 1 for PPE. In DC704, again a diffusion pump oil (1,3,3,5-tetramethyl-1,1,5,5-tetraphenyl-trisiloxane, a rather large molecule) we have the additional advantage of a large temperature range of the measurement, from 209 to 239 K. As in TPE, we find temperature-independent parameters $\beta$, $b_2$, $f_0$ and $f_{jc}$. Again, we find the shoving model [@shoving] confirmed in both glass formers. In DC704, one even sees the curvature of both curves (see Fig. 4), which justifies our temperature exponent Ansatz, eq. (\[tscal\]).
Table I comprises the fit parameters for these three type-A glass formers. Note that our formalism allows to describe both shear and dielectric data over the whole temperature range with eleven temperature-independent parameters.
glass former TPE DC704 PPE
----------------------- -------- ------- -------
$T_g$ (K) 249 211 244
$\Delta\epsilon$ 0.0491 0.257 2.011
$\gamma_\epsilon$ 1.85 2.26 1.90
$\beta$ 0.77 0.85 1.04
$b_2$ 0.18 0.27 0.215
$V_c(T_g)$ (eV) 0.767 0.639 0.755
$\gamma_V$ 4.3 4.6 4.6
$G_\infty(T_g)$ (GPa) 1.38 1.80 1.27
$\gamma_G$ 4.7 4.2 4.7
$f_0$ 1.65 2.38 2.22
$f_{jc}$ 2.5 2.45 2.05
: Parameters of the three type-A glass formers. Upper part $\epsilon(\omega)$, lower part $G(\omega)$.[]{data-label="tab:tab1"}
The four type-B glass formers
-----------------------------
In the type-B glass formers DHIQ, PB20, Squalane and TPG, one needs to fit a Johari-Goldstein peak on top of the high-frequency tail of the $\alpha$-process. This is illustrated in Fig. 5 for our first type-B example, squalane.
Squalane, $C_{30}H_{62}$, is a short chain molecule with 24 carbon atoms in the backbone and 6 attached $CH_3$-groups, rather polymerlike. It has a strong and well-separated Johari-Goldstein peak (see Fig. 5), much better visible in the shear data than in the dielectric data. The dielectric dipole moment is very weak. Nevertheless, it is possible to fit both sets of data with the same retardation spectrum, attaching a substantially higher weight to the Johari-Goldstein peak in the shear (see Table II).
In this substance, it is not possible to fit the shear data with a temperature-independent parameter $f_0$; one has to postulate a rather strong increase of $f_0$ with increasing temperature $$\label{f0t}
f_0(T)=f_0(T_g)+f_0'(T-T_g),$$ but one can keep the parameter $f_{jc}$ constant (see Table II).
PB20 is a true polymer, relatively short (5000 g/mol), composed of 80 % 1,4-polybutadiene monomers and 20 % 1,2-polybutadiene monomers. The results look very similar to those of squalane, and the resulting fit parameters in Table II are in fact close to those of squalane. Even more than squalane, it has the polymer feature of a relatively slow decrease of $\epsilon''(\omega)$ at low frequency, explainable in terms of chain modes with long relaxation times [@ferry]. This is illustrated in Fig. 6, which shows the deviation between fit and data at low frequency. As a consequence, the resulting parameters have a larger error bar in squalane and polybutadiene than in the two molecular substances TPG and DHIQ. In particular, the deviations between $\gamma_V$ and $\gamma_G$ do not demonstrate a failure of the shoving model.
TPG is a much more favorable case, with a very large dipole moment and no problems at the cutoff barrier. As Fig. 7 (a) shows, our spectrum of eq. (\[lvform\]) provides beautiful fits over a large temperature range. One needs to take the temperature dependence of the Johari-Goldstein peak position $V_1$ into account. Our fit found $$\label{tpgshift}
V_1=0.295\frac{T}{T_g},$$ a bit smaller shift than the one found in aging experiments [@dyretpg].
Fig. 7 (b) shows that the shear data are well described in terms of the dielectric retardation spectrum. There is a small temperature dependence of $f_0$, but $f_{jc}$ is again a temperature-independent constant. The shoving model is found to be well fulfilled (see Fig. 8).
Finally, DHIQ, decahydroisoquinoline, $C_9H_{17}N$, is best described as two cyclohexanol rings sharing one $C-C$-bond, one of the two rings having an $NH$ replacing a $CH_2$-group. In this case, the Johari-Goldstein peak is very prominent in the dielectric data [@ranko; @paluch2], comparable to the one in $G(\omega)$. The dipole moment is large; both $V_c$ and $G_\infty$ can be determined with high accuracy. Again, their temperature exponents $\gamma_V$ and $\gamma_G$ agree within the error bars (see Table II), in agreement with the shoving model [@shoving]. Since both are exceptionally large (DHIQ is very fragile, m=158 in Angell’s scheme [@bohmer]), their good agreement provides a strong argument for the validity of the model.
In Table II, the Johari-Goldstein peak is characterized by the weight of the peak $$w(T)=a_1\sqrt(\pi/\gamma_1)=a_1FWHM\sqrt(\pi/4\ln 2)$$ which shows a Boltzmann factor behavior $$\label{boltz}
w(T)=w(T_g)\exp(-E_a(1/k_BT-1/k_BT_g)),$$ with a formation energy $E_a$ which is on the average 2/3 of the peak position $V_1$.
glass former Squalane PB20 TPG DHIQ
----------------------------------------- ---------- ------- -------- -------
$T_g$ (K) 167 176 184 175
$\Delta\epsilon(T_g)$ 0.0155 0.132 23.3 1.707
$\gamma_\epsilon$ 2.1 0.0 1.51 0.0
$\beta$ 0.6 0.44 0.85 0.4
$b_2$ 0.2 0.2 0.22 0.2
$V_c(T_g)$ (eV) 0.517 0.54 0.63 0.635
$\gamma_V$ 3.2 3.8 3.0 6.4
$G_\infty(T_g)$ (GPa) 1.33 1.63 2.69 3.1
$\gamma_G$ 2.5 2.7 2.8 6.3
$f_0(T_g)$ 2.4 3.25 6.7 1.56
$f_0'$ (1/K) 0.4 0.41 -0.04 0.25
$f_{jc}$ 2.7 2.4 2.5 2.0
$V_1$ (eV) 0.27 0.28 0.32\* 0.32
$FWHM$(eV) 0.135 0.170 0.154 0.16
$w_s(T_g)$ 0.56 0.55 0.15 0.50
$w_\epsilon(T_g)$ 0.03 0.23 0.02 0.48
$E_a$ (eV) 0.25 0.12 0.19 0.25
\*average value, see eq. (\[tpgshift\])
: Parameters of the four type-B glass formers. Upper part $G(\omega)$, middle part $\epsilon(\omega)$, lower part Johari-Goldstein peak parameters for both.[]{data-label="tab:tab2"}
Discussion
==========
The preceding section presented a quantitative description of the $\alpha$- and the $\beta$-process in dielectrics and shear for seven different glass formers, a description which is based on the concept of isolated and independent thermally activated jumps in the energy landscape. The description allows for a reasonable fit of the temperature dependence in terms of temperature-independent parameters. The number of parameters is not small; one needs eleven or twelve parameters for a type-A glass former (depending on whether $f_0$ is temperature-independent or not, see Table I and II) and five additional parameters for the description of the $\beta$- or Johari-Goldstein peak (see Table II).
Nevertheless, the exercise is not completely meaningless. One does indeed get meaningful quantitative information, which is impossible to obtain otherwise. The first and rather important one is the probable equality of the retardation spectra of shear and dielectrics (but with a different weight of the Johari-Goldstein peak), an information which one can guess from the raw data (see Figs. 1 and 2), but which requires a full fit for its quantitative check.
The second and equally important quantitative information concerns the dimensionless ratio $f_{jc}$ between the terminal dielectric relaxation time $\tau_c$ and the product of the viscosity with the total retardation compliance, eq. (\[fjc\]). The seven fitted values lie between 2 and 2.7 (average value 2.37). This indicates a general relation between the dielectric terminal time and the viscosity. Since the dielectric terminal time seems to coincide with the structural lifetime [@bow], it is probably also the lifetime of the double-well potentials which are responsible for the retardation spectrum.
Question: What do we expect for the ratio $f_{jc}$ if this is indeed the case? To answer this question, consider a constant applied shear stress. After the time $\tau_c$, all the double-well potentials of the spectrum would have reached thermal equilibrium, giving their full contribution to the compliance. From this consideration, if we renew them at the time $\tau_c$, we would naively expect them to be able to give their contribution again after this time, yielding $f_{jc}=1$.
But this answer is not correct. To get the correct answer, one must consider the difference between energy and free energy in these double-well potentials. To keep the argument simple, let us restrict ourselves to the special case of a symmetric double-well; it applies as well to the asymmetric case.
If the double-well is initially symmetric and if it couples to the shear stress $\sigma$ with a coupling constant $v$ (the coupling constant has the dimension of a volume), then the asymmetry $\Delta$ under the stress is $\sigma v$. One well has the energy $-\sigma v/2$, the other has the energy $+\sigma v/2$.
In thermal equilibrium, the population of the two wells is given by their Boltzmann factors. It is easy to calculate the energy $U$ of the equilibrated system in the limit of a small stress $$U=-\frac{\sigma^2v^2}{4k_BT}.$$ This is the energy transported to the heat bath in the equilibration of the relaxing entity after switching on the stress.
The free energy $F$ is $$F=-\frac{\sigma^2v^2}{8k_BT},$$ only half of the energy itself. If one thinks about it, the reason is clear: spending the energy, one has spanned an entropic spring by the population difference in the two wells. If one removes the stress slowly, one gets half the energy back. But if the double-well potential decays, one gets nothing back.
The contribution of the relaxing entity to the compliance is given by the second derivative of the free energy with respect to the stress. But if we now deal with the effect of a renewal of the double-well potential on the viscosity, we have to count the energy. This means we spend twice as much energy under a constant stress as the one calculated above in our first oversimplified picture. And this means the viscosity must be a factor of 2 smaller, which implies $f_{jc}=2$. This is reasonably close to the fitted values in Table I and II.
A third quantitative conclusion of the present study is a surprising agreement with the conclusions of Plazek et al [@plazek] from their recoverable shear compliance experiments. If one takes the parameters of Table I to calculate the recoverable compliance, one gets curves which closely resemble those reported by them. Obviously, it is experimentally much easier to detect the Andrade creep [@andrade] in creep experiments than in dynamical ones. If one calculates $f_0$ from their data, one finds values between 1.5 and 2.3, similar to those in Table I.
Here, however, a word of caution is in place. Our data, taken as they are, do not imply a limited recoverable shear compliance. In fact, they are well fitted by the BEL model [@bel], which has a divergent recoverable compliance. The values in the two tables stem from the assumption that the two retardation spectra of dielectrics and shear (at least as far as the $\alpha$-peak is concerned) are the same.
The same is true for the fourth conclusion, the validity of the shoving model [@shoving]. The fitted $G_\infty$-values were obtained under the same assumption.
Finally, the Johari-Goldstein peak increases its height with increasing temperature. The increase follows a Boltzmann factor, with a formation energy of about two thirds of the barrier height at the center of the peak.
Summary and conclusions
=======================
Dielectric and shear relaxation data in seven highly viscous liquids, most of them molecular liquids, were evaluated in terms of a barrier density of independent thermally activated relaxation centers. Three of the substances are type-A glass formers without or with only a very small Johari-Goldstein peak, four of them show a pronounced Johari-Goldstein peak.
The most important conclusion is the probable equality of the dielectric and shear retardation spectra, guessed from the raw data and confirmed by a quantitative fit. The difference in the peak positions is due to the influence of the viscosity. The Johari-Goldstein peak has different weight in dielectrics and shear.
The second important conclusion concerns the viscosity. It seems probable that the viscosity results from the constant renewal of the double-well potentials in the sample within the terminal dielectric relaxation time.
Our data support earlier recovery compliance results by Plazek et al [@plazek], according to which one has an Andrade [@andrade] creep $J\sim t^{1/3}$ at short times in type-A glass formers (glass formers without Johari-Goldstein peak). They further support the shoving model [@shoving], which postulates a proportionality between the infinite frequency shear modulus and the Arrhenius barrier of the terminal relaxation time.
Acknowledgement: The author is deeply thankful to Kristine Niss and Bo Jakobsen for communicating their beautiful data to him, to Niels Boye Olsen and Tage Christensen for enlightening discussions and to Jeppe Dyre for constant encouragement and a lot of helpful advice.
[99]{} N. O. Birge and S. R. Nagel, Phys. Rev. Lett. [**54**]{}, 2674 (1985); N. O. Birge, Phys. Rev. B [**34**]{}, 1631 (1986) K. L. Ngai and R. W. Rendell, Phys. Rev. B [**41**]{}, 754 (1990) I. Chang and H. Sillescu, J. Chem. Phys. [**101**]{}, 8794 (1997) and further references therein K. Schröter and E. Donth, J. Non-Cryst. Solids [**307-310**]{}, 270 (2002) U. Buchenau, M. Ohl and A. Wischnewski, J. Chem. Phys. [**124**]{}, 094505 (2006) U. Buchenau, R. Zorn, M. Ohl and A. Wischnewski, cond-mat/0607056 and Phil. Mag. 2006 (in press) K. Niss, B. Jakobsen and N. B. Olsen, J. Chem. Phys. [**123**]{}, 234510 (2005) B. Jakobsen, K. Niss and N. B. Olsen, J. Chem. Phys. [**123**]{}, 234511 (2005) T. Christensen and N. B. Olsen, Rev. Sci. Instrum. [**66**]{}, 5019 (1995) G. P. Johari and M. Goldstein, J. Chem. Phys. [**53**]{}, 2372 (1970) G. P. Johari and M. Goldstein, J. Chem. Phys. [**55**]{}, 4245 (1971) D. J. Ferry, “Viscoelastic properties of polymers”, 3rd ed., John Wiley, New York 1980 U. Buchenau, Phys. Rev. B [**63**]{}, 104203 (2001) A. Kudlik, Ch. Tschirwitz, S. Benkhof, T. Blochowicz and E. Rössler, Europhys. Lett. [**40**]{}, 649 (1997) K. L. Ngai and M. Paluch, J. Chem. Phys. [**120**]{}, 857 (2004) D. J. Plazek, C. A. Bero and I.-C. Chay, J. Non-Cryst. Solids [**172-174**]{}, 181 (1994) E. N. da C. Andrade, Proc. Roy. Soc. A [**84**]{}, 1 (1910) C. Leon, K. L. Ngai and C. M. Roland, J. Chem. Phys. [**110**]{}, 11585 (1999) J. Mattsson, R. Bergman, P. Jacobsson and L. Börjesson, Phys. Rev. Lett. [**90**]{}, 075702 (2003) J. Mattsson, R. Bergman, P. Jacobsson and L. Börjesson, Phys. Rev. Lett. [**94**]{}, 165701 (2005) J. C. Dyre and N. B. Olsen, Phys. Rev. Lett. [**91**]{}, 155703 (2003) S. Pawlus, S. Hensel-Bielowska, K. Grzybowska, J. Ziolo and M. Paluch, Phys. Rev. B [**71**]{}, 174107 (2005) M. Goldstein, J. Chem. Phys. [**51**]{}, 3728 (1968) M. Pollak and G. E. Pike, Phys. Rev. Lett. [**28**]{}, 1449 (1972) K. S. Gilroy and W. A. Phillips, Phil. Mag. B [**43**]{}, 735 (1981) J. C. Dyre, N. B. Olsen and T. Christensen, Phys. Rev. B [**53**]{}, 2171 (1996) R. Richert, K. Duvvuri and L.-T. Duong, J. Chem. Phys. [**118**]{}, 1828 (2003) M. Paluch, S. Pawlus, S. Hensel-Bielowska, E. Kaminska, D. Prevosto, S. Capaccioli, P. A. Rolla and K. L. Ngai, J. Chem. Phys. [**122**]{}, 234506 (2005) R. Böhmer, K. L. Ngai, C. A. Angell and D. J. Plazek, J. Chem. Phys. [**99**]{}, 4201 (1993) A. J. Barlow, A. Erginsav and J. Lamb, Proc. Roy. Soc. A [**309**]{}, 473 (1969)
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
We study the autocorrelation function of a conserved spin system following a quench at the critical temperature. Defining the correlation length $L(t)\sim t^{1/z}$, we find that for times $t'$ and $t$ satisfying $L(t')\ll L(t)\ll L(t')^\phi$ well inside the scaling regime, the spin autocorrelation function behaves like $\langle s(t)s(t')\rangle\sim
L(t')^{-(d-2+\eta)}\left[{L(t')}/{L(t)}\right]^{\lambda^\prime_c}$. For the $O(n)$ model in the $n\to\infty$ limit, we show that $\lambda^\prime_c=d+2$ and $\phi=z/2$. We give a heuristic argument suggesting that this result is in fact valid for any dimension $d$ and spin vector dimension $n$. We present numerical simulations for the conserved Ising model in $d=1$ and $d=2$, which are fully consistent with the present theory.
author:
- Clément Sire
title: Autocorrelation exponent of conserved spin systems in the scaling regime following a critical quench
---
The quench of a ferromagnetic spin system [@bray], from high temperature ($T_0>T_c$) to low temperature (typically $T=0$ or $T=T_c$) is characterized by the growth of a correlation length scale (or domain length scale when domains can be identified), $L(t)\sim t^{1/z}$. In the non conserved case, $z$ depends on the final temperature of the quench ($z=2$ for $T<T_c$, while $z$ is the dynamical critical exponent for $T=T_c$ [@hh]). If the order parameter $s({\bf x},t)$ (possibly a vector) is locally conserved, $z=3$ (scalar) or $z=4$ (vector) for a quench below $T_c$ [@bray], while $z=4-\eta$ [@hh] for a quench at $T_c$. Another interesting and fundamental quantity is the spin autocorrelation $A(t,0)=\langle s({\bf x},t)s({\bf
x},0)\rangle\sim L(t)^{-\lambda}$ [@lambdac; @lambdabibth; @lambdabibexp]. For non conserved dynamics, whatever the temperature of the quench, $\lambda$ is non trivial (except in $d=1$ [@bray]) and only approximate theories are available for $T=0$ [@bray; @lambdabibth], while for $T=T_c$ [@lambdac], the $\varepsilon$-expansion of $\lambda_c$ can be calculated. In the case of conserved dynamics, it is now well established that $\lambda_c=\lambda=d$ for quenches at and below $T_c$ [@satya1; @satya2; @huse]. Hence, for fixed $t'$ and $t\to+\infty$, $A(t,t')\sim L(t)^{-d}$. However, for $t'$ and $t>t'$ both in the scaling regime (in a sense to be defined later), several authors have observed numerically [@lambdapbib1; @lambdapbib2; @godreche] and experimentally [@explamb] a faster power law decay of the autocorrelation. More precisely, in the case of a quench of an Ising system at $T_c$ (critical quench), the authors of [@godreche] obtained numerically $$A(t,t')\sim
L(t')^{-(d-2+\eta)}\left[\frac{L(t')}{L(t)}\right]^{\lambda'_c},
\label{lambdap}$$ in $d=1$ (where formally $\eta=1$ and $T_c=0$) and $d=2$. They respectively found $\lambda'_c\approx 2.5$ in $d=1$ and $\lambda'_c\approx 3.5$ in $d=2$. They also suggested a general scaling relation $$A(t,t')\sim L(t)^{-d}C\left[\frac{L(t)}{L(t')^\phi}\right],
\label{scalam}$$ where $C(x)$ goes to a non zero constant for $x\to +\infty$, $C(x)\sim x^{-(\lambda'_c-d)}$ for $x\to 0$, and $$\phi=1+\frac{2-\eta}{\lambda'_c-d}.\label{phigen}$$ As noticed in [@godreche], this scaling implies the existence of a new relevant length scale $L(t')^\phi$, which is the crossover length between the two observed regimes. Its physical meaning has yet to be elucidated.
In the present Letter, we address the problem of the actual analytical derivation of $\lambda'_c$ in the case of the $O(n)$ model in the limit of infinite $n$. Within this model, the diffusive nature of this new length scale can be understood, and one finds $\lambda'_c=d+2$ and $\phi=2$. By generalizing the interpretation of this diffusive crossover length scale to any $O(n)$ spin system, we conjecture that the result $\lambda'_c=d+2$ holds and that $\phi=2-\eta/2=z/2$.
We first examine the exactly solvable $O(n)$ model in the limit $n\to\infty$ and for dimensions $d>2$. This model is known to be pathological for a quench at zero temperature, displaying multiscaling [@coniglio], whereas normal scaling should be restored at finite $n$ [@bray; @bray1]. However, after a quench at $T_c$, the structure factor obeys standard scaling even for $n\to\infty$ [@satya2]. In the standard Cahn-Hilliard equation describing the evolution of the magnetization field ${\bf s}({\bf x},t)$, ${\bf s}^2({\bf
x},t)/n$ can be replaced by its average in the limit $n\to\infty$. Thus, any spin component satisfies $$\frac{\partial s}{\partial t}=-\Delta\left[\Delta s+k_0^2s-\langle
s^2\rangle s\right]+\eta,\label{ch}$$ where $k_0^2$ is a constant, $\eta({\bf x},t)$ is a conserved delta-correlated noise satisfying $\langle\eta({\bf k},t)\eta({\bf
k'},t')\rangle=2T_ck^2\delta({\bf k}+{\bf k'})\delta(t-t')$, and $\langle s^2\rangle$ has to be computed self-consistently. Although the derivation of the structure factor has already appeared in the literature [@satya2], we briefly repeat it as it furnishes a useful basis for our final derivation.
Eq. (\[ch\]) can be solved in Fourier space, leading to $$s({\bf k},t)=\left[s({\bf k},0)+\int_0^t{\rm
e}^{q(k,\tau)}\eta({\bf k},\tau)\,d\tau\right]{\rm e}^{-q(k,t)},
\label{s}$$ where $$q(k,t)=k^4t-k^2\int_0^t[k_0^2-\langle s^2({\bf
x},\tau)\rangle]\,d\tau.$$ Assuming an uncorrelated initial condition such that $\langle
s(-{\bf k},0)s({\bf k},0)\rangle=s_0^2$, we then find the structure factor $S(k,t)=\langle s(-{\bf k},t)s({\bf k},t)\rangle$ $$S(k,t)=\left[s_0^2+2T_ck^2\int_0^t{\rm
e}^{2q(k,\tau)}\,d\tau\right]{\rm e}^{-2q(k,t)}.$$ We now express the self-consistent condition $\langle s^2({\bf
x},t)\rangle=\int^\Lambda S(k,t)\,\frac{d^d{\bf k}}{(2\pi)^d}$, where $\Lambda$ is the inverse of a lattice cut-off. $T_c$ is such that $S(k,t\to\infty)\sim k^{-2+\eta}$, where $\eta$ is the usual critical exponent controlling the decay of the static correlation function ($\eta=0$ for $n\to\infty$). This leads to $$T_c\int^\Lambda k^{-2}\,\frac{d^d{\bf
k}}{(2\pi)^d}=k_0^2.\label{tc}$$ Finally, if the above condition is satisfied, we find that $q(k,t)$ obeys a scaling relation for large $t$ $$q(k,t)=q(kt^{1/4})=k^4t-c_dk^2t^{1/2},$$ where $c_d$ is a universal constant determined by a simple integral relation ($c_d=0$ for $d>4$) [@satya2], and $q(u)=u^4-c_du^2$. We thus find $L(t)=t^{1/z}$, with $z=4$, in agreement with the general result $z=4-\eta$ [@hh]. We hence reproduce the general form of the structure factor $$S(k,t)=s_0^2{\rm e}^{-2q(kt^{1/z})}+t^{(2-\eta)/z}F(kt^{1/z}).
\label{scas}$$ For the $O(n=\infty)$ model, we have $z=4$, $\eta=0$, and $$F(u)=2T_cu^2\int_0^1{\rm e}^{-2u^4(1-v)+2c_du^2(1-v^{1/2})}\,dv.$$ Note the following asymptotics for $F(u)$ $$\begin{aligned}
F(u)&\sim &2T_c u^{2}, \quad u\to 0,\label{as0}\\
F(u)&\sim &T_c u^{-2}, \quad u\to +\infty.\label{asinf}\end{aligned}$$ In the scaling limit, the first term of the right hand side (RHS) of Eq. (\[scas\]) is negligible compared to the second term. In real space, Eq. (\[scas\]) illustrates the fact that conventional (critical) scaling is obeyed $$\langle{\bf s}({\bf x},t){\bf s}({\bf
0},t)\rangle=L(t)^{-(d-2+\eta)}f[x/L(t)],\label{f}$$ where $f$ is simply the inverse Fourier transform of $F$.
We now move to the calculation of the two-time correlation function, focusing on the case where both considered times $t'$ and $t>t'$ are in the scaling regime, a notion which will be made more precise hereafter. Using Eq. (\[s\]), and working along the line of the derivation of $S(k,t)$, we find the following expression for $C(k,t)=\langle s(-{\bf k},t')s({\bf k},t)\rangle$ $$\begin{aligned}
%{\rm e}^{q[kt'^{1/4}]-q[kt^{1/4}]}S(k,t')
C(k,t,t')&=& A_1(k,t,t')+A_2(k,t,t'),\\
A_1(k,t,t')&=& s_0^2{\rm e}^{-q[kL(t')]-q[kL(t)]},\label{c1}\\
A_2(k,t,t')&=& L(t')^2{\rm e}^{q[kL(t')]-q[kL(t)]}F[k
L(t')].\label{c2}\end{aligned}$$ For a fixed $t'$ and $t\to\infty$, the contribution of $A_2$ becomes negligible, as for large $t$ and hence $L(t)$, only the contribution of small wave vector $k\sim L(t)^{-1}$ matters. Using the result of Eq. (\[as0\]), we indeed find that that this term is of order $k^2\sim L(t)^{-2}$, whereas the main contribution $A_1$ in Eq. (\[c1\]) is of order $s_0^2$ which is a constant. Contrary to what occurs for $S(k,t)$, it is now the term depending on the initial conditions via $s_0^2$ which dominates. Hence in this limit of fixed $t'$ and $t\to\infty$, we find $$C(k,t,t')\approx C(k,t,0)= s_0^2{\rm e}^{-q[kL(t)]}=G[kL(t)],$$ and in real space $$\langle{\bf s}({\bf x},t){\bf s}({\bf 0},t')\rangle=
L(t)^{-d}g[x/L(t)],$$ where $g$ is the inverse Fourier transform of $G$. One recovers, in the limit $t\gg t'$ to be made more precise later, that the large time autocorrelation exponent is $\lambda_c=d$, which is observed in all conserved models including thermal fluctuations [@satya1; @satya2]. In this limit, conventional scaling holds. However, we will now show that the contribution of Eq. (\[c2\]) which has not so far been considered dominates in a well defined time regime, and will prompt us to introduce another autocorrelation exponent $\lambda'_c$.
For general $t'$ and $t>t'$, we now proceed to calculate the autocorrelation for a spin on a given lattice site. Defining $A(t,t')=\langle s({\bf x},t')s({\bf x},t)\rangle$, we finally find $A(t,t')=A_1(t,t')+A_2(t,t')$, where $$A_1(t,t')=s_0^2\int^\Lambda {\rm
e}^{-q[kL(t')]-q[kL(t)]}\,\frac{d^d{\bf k}}{(2\pi)^d}.$$ After a change of variable and noting that the region of $k\gg
L(t)^{-1}$ barely contributes to the integral, we find $$A_1(t,t')=L(t)^{-d}a_1[L(t')/L(t)].$$ Thus, $A_1(t,t')$ obeys conventional scaling for any $t'$ and $t>t'$. We explicitly find $$a_1(u)=s_0^2\int^\infty {\rm
e}^{-k^4(1+u^4)+c_dk^2(1+u^2)}\,\frac{d^d{\bf k}}{(2\pi)^d},$$ where this integral is now over the entire space. $a_1(u)$ remains bounded and of order $s_0^2$ for any value of $u=L(t')/L(t)\leq 1$. Keeping the notation $u=L(t')/L(t)\leq 1$, the expression for $A_2(t,t')$ can be written in the rescaled form $$\begin{aligned}
&&A_2(t,t')=L(t')^{-(d-2)}u^d
{\times}\label{a2}\\
&&\int^{L(t)\Lambda}{\rm
e}^{-k^4(1-u^4)+c_dk^2(1-u^2)}F(ku)\,\frac{d^d{\bf
k}}{(2\pi)^d}.\nonumber\end{aligned}$$
Let us analyze the different asymptotics for $A_2(t,t')$. First of all, for large $t=t'$ ($u=1$), the integral is dominated by the region of large $k$’s. Using Eq. (\[tc\]), we find the expected result $A_2(t,t)\approx k_0^2$, which is the equilibrium value of $\langle s^2\rangle$. Note that if $t-t'\ll 1$, we obtain $A_2(t,t')=A_2(t,t)-K_d(t-t')+...$, where $K_d$ is a computable constant. We now assume that $1\ll L(t)-L(t')\ll L(t')$, which ensure that $u$ is very close to 1. In this regime, we find that $$\begin{aligned}
A_2(t,t')&\sim &J_d
L(t)^{-(d-2)}\left[1-\frac{L(t')}{L(t)}\right]^{-(d-2)/4}, \\
&\sim & J'_d\,(t-t')^{-(d-2)/4},\end{aligned}$$ where $J_d$ and $J'_d$ can be written exactly as simple integrals. Finally, and this constitutes the central result of this Letter, we consider the regime $1\ll L(t')\ll L(t)$. In this case, $u\ll
1$, and the integral of Eq. (\[a2\]) is dominated by the region of $k$ of order unity, so that the small argument asymptotics can be taken for $F(ku)$ in Eq. (\[a2\]). We find $$\begin{aligned}
A_2(t,t')&\sim &\kappa_d\,
L(t')^{-(d-2)}\left[\frac{L(t')}{L(t)}\right]^{d+2},\label{ln} \\
\kappa_d&= &2T_c\int^\infty k^2{\rm
e}^{-k^4+c_dk^2}\,\frac{d^d{\bf k}}{(2\pi)^d}.\end{aligned}$$ Eq. (\[ln\]) takes exactly the expected form of Eq. (\[lambdap\]), with $$\lambda'_c=d+2.$$ Hence, we find that $A_2(t,t')$ [*prevails over*]{} $A_1(t,t')$ for $L(t')\leq L(t)\ll L_0(t')$, with $L_0(t')\sim L(t')^\phi$ and $\phi=2$. For $L(t')\ll L(t)\ll L_0(t')$, the autocorrelation $A(t,t')\approx A_2(t,t')$ is then given by Eq. (\[ln\]). Moreover, Eq. (\[a2\]) shows that instead of Eq. (\[scalam\]), the correct scaling is rather $$A(t,t')= A(t,0)+L(t')^{-(d-2+\eta)}D[L(t)/L(t')],\label{scageneral}$$ with $D(1/u)\sim u^{\lambda'_c}$ for $u\ll 1$. Both scaling are equivalent only for $u\ll 1$. We now present an heuristic argument based on dimensional analysis which suggests that the result $\lambda'_c=d+2$ may be of general validity for conserved spin systems. Indeed, the occurrence of a new length scale bigger than $L(t)$ could have been inferred from the small $k$ behavior of $S(k,t)$. In the $n\to\infty$ limit and for $k\to 0$, Eq. (\[scas\]) leads to $$S(k,t)\approx s_0^2+2T_ck^2 L(t)^4+...\label{smallk}$$ A natural momentum scale $k_0(t)\sim L_0(t)^{-1}$ arises by matching the two terms of the RHS of Eq. (\[smallk\]), which leads to $\phi=2$ and hence $\lambda'_c=d+2$.
In the general case, for short-range correlated initial conditions, we expect the following general form to hold $$S(k,t)=F_1[k L(t)]+L(t)^{2-\eta} F_2[k L(t)],$$ with $F_1(0)=s_0^2$ being a non zero constant (equal to the variance of the initial total magnetization normalized by the volume), while the scaling contribution should vanish for $k=0$, implying $F_2(0)=0$. Imposing $F_2(p)\sim p^\gamma$, $\gamma$ is necessarily an even integer. If $\gamma$ were not integer, the correlation function scaling function $f$ defined in Eq. (\[f\]) would have a power law decay for large distance, which is unphysical as such correlations cannot develop in a finite time starting from short-range ones. $\gamma$ cannot be an odd integer as space isotropy guarantees that $f$ should be an even function. Contrary to the case of a quench at $T=0$, for which convincing theoretical arguments for $d\geq 2$ [@k4] and experiments [@expk4] show that $F_2(p)\sim p^4$, there is no reason to expect the same for critical quenches. Generically, we expect $F_2(p)\sim p^2$ as found for the $d=1$ conserved Ising model [@satya1; @satya2], and in the present Letter for the $O(n)$ model for $n\to\infty$. Finally, the small $k$ behavior of the structure factor should be of the form $$S(k,t)\approx s_0^2+C_0k^2 L(t)^{4-\eta}+...,\label{smallkgen}$$ where $C_0>0$ is a constant. Assuming that the length scale obtained by matching both terms of the RHS of Eq. (\[smallkgen\]) is the same as the crossover length between the two observed regimes for the autocorrelation, and using the general result of Eq. (\[phigen\]), we obtain $$\phi=2-\eta/2=1+\frac{2-\eta}{\lambda'_c-d},$$ which implies $\lambda'_c=d+2$. This result also extends to $d=1$ after formally taking $\eta=1$, leading to $\lambda'_c=3$ and $\phi=3/2$. Note that the crossover scale can also be written $L_0(t)\sim t^{\phi/z}\sim t^{1/2}$, which is the diffusion scale associated to thermal noise. At least in $d=1$, this scale can be related to the equilibrium diffusion of tagged spins observed in [@luck].
We now present simulations of the Ising model Kawasaki dynamics in $d=1$ and $d=2$ after a quench at $T_c$. In the $d=1$ case, we use the accelerated algorithm introduced in [@satya2], which is faster than that used in [@godreche] (but does not permit to compute simply the response function as was needed in [@godreche]). By fitting $A(t,t')$ in the scaling regime, the authors of [@godreche] found $\lambda'_c\approx 2.5$ lower than our prediction $\lambda'_c=3$. However, for the moderately large numerically accessible times, the contribution of $A_1(t,t')\approx A(t,0)$ is significant. When plotting $A(t,t')-A(t,0)$ as a function of $L(t)$, one actually finds $\lambda'_c\approx 3$ instead of $\lambda'_c\approx 2.5$. Result of simulations for the $d=2$ Ising model evolving with Kawasaki dynamics at $T_c$ are shown on Fig. 2. Considering the very slow growth of $L(t)\sim t^{4/15}$, it is difficult to obtain data spanning more than one decade in $L(t)$. Hence, the regime of interest $1\ll L(t')\ll L(t)$ cannot be reached and the separation of scales properly achieved. Still, subtracting $A(t,0)$ from $A(t,t')$ leads to $\lambda'_c\approx 4$, significantly greater than the value $\lambda'_c\approx 3.5$ found in [@godreche].
In conclusion, in view of the exact result for the $O(n=\infty)$ model, a general argument for any $n$ and $d$, and convincing simulations in $d=1$ (and consistent in $d=2$), we have strongly suggested that $\lambda'_c=d+2$ and $\phi=z/2$ generally holds. We also find that the scaling form of Eq. (\[scageneral\]) is more appropriate than Eq. (\[scalam\]). The compelling generalization of our heuristic argument to a quench at $T<T_c$ (in $d\geq 2$, and admitting $F_2(p)\sim p^4$) leads to $A(t,t')\sim [L(t')/L(t)]^{\lambda'}$ for $L(t')\ll L(t)\ll
L(t')^\phi$, with $\lambda'=d+4$ and $\phi=1+d/4$. In $d=2$, the prediction $\lambda'=6$ is significantly larger than the numerical result $\lambda'\approx 4$ [@lambdapbib1]. However, the fit in [@lambdapbib1] was performed in the short scaling regime over less than a decade in $L(t)$, and subtracting $A(t,0)$ before performing the fit could lead to a significantly higher value for $\lambda'$, as noted in the two examples treated in this Letter.
[10]{} A.J. Bray, Adv. Phys. [**43**]{}, 357 (1994). P. Hohenberg and B. Halperin, Rev. Mod. Phys. [**49**]{}, 435 (1977). H.K. Janssen, B. Schaub, and B. Schmittmann, Z. Phys. B [**73**]{}, 539 (1989); D.A. Huse, Phys. Rev. B [**40**]{}, 304 (1989). D.S. Fisher and D.A. Huse, Phys. Rev. B [**38**]{}, 373 (1988); G. F. Mazenko, Phys. Rev. B [**42**]{}, 4487 (1990); C. Sire and S. N. Majumdar, Phys. Rev. Lett. [**74**]{}, 4321 (1995). N. Mason, A.N. Pargellis, and B. Yurke, Phys. Rev. Lett. [**70**]{}, 190 (1993). S.N. Majumdar, D.A. Huse, and B.D. Lubachevsky, Phys. Rev. Lett. [**73**]{}, 182 (1994). S.N. Majumdar and D.A. Huse, Phys. Rev. E [**52**]{}, 270 (1995). F.J. Alexander, D.A. Huse, and S.A. Janowsky, Phys. Rev. B [**50**]{}, 663 (1994). C. Yeung, M. Rao, and R.C. Desai, Phys. Rev. E [**53**]{}, 3073 (1996). J.F. Marko and G.T. Barkema, Phys. Rev. E [**52**]{}, 2522 (1995). C. Godrèche, F. Krzakala, and F. Ricci-Tersenghi, JSTAT P04007 (2004). T. Nagaya and J.-M. Gilli, Phys. Rev. Lett. [**92**]{}, 145504 (2004). A. Coniglio and M. Zannetti, Europhys. Lett. [**10**]{}, 575 (1989); A. Coniglio, P. Ruggiero, and M. Zannetti, Phys. Rev. E [**50**]{}, 1046 (1994). A.D. Rutenberg and A.J. Bray, Phys. Rev. E [**51**]{}, 5499 (1995); S. Puri, A.J. Bray, and F. Rojas, Phys. Rev. E [**52**]{}, 4699 (1995); F. Rojas and A.J. Bray, Phys. Rev. E [**51**]{}, 188 (1995). C. Yeung, Phys. Rev. Lett. [**61**]{}, 1135 (1988); H. Furukawa, Phys. Rev. Lett. [**62**]{}, 2567 (1989); H. Furukawa, J. Phys. Soc. Jpn. [**58**]{}, 216 (1989). P. Wiltzius, F.S. Bates, and W.R. Heffner, Phys. Rev. Lett. [**60**]{}, 1538 (1988). C. Godrèche and J.M. Luck, J. Phys. A [**36**]{}, 9973 (2003).
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'In this review, we discuss the role of the various experimental programs taking part in the broader effort to identify the particle nature of dark matter. In particular, we focus on electroweak scale dark matter particles and discuss a wide range of search strategies being carried out and developed to detect them. These efforts include direct detection experiments, which attempt to observe the elastic scattering of dark matter particles with nuclei, indirect detection experiments, which search for photons, antimatter and neutrinos produced as a result of dark matter annihilations, and collider searches for new TeV-scale physics. Each of these techniques could potentially provide a different and complementary set of information related to the mass, interactions and distribution of dark matter. Ultimately, it is hoped that these many different tools will be used together to conclusively identify the particle or particles that constitute the dark matter of our universe.'
author:
- 'Dan Hooper Edward A. Baltz'
title: |
Strategies for Determining the Nature of Dark Matter\
------------------------------------------------------------------------
---
psfig.sty
dark matter
INTRODUCTION
============
There exists a wide array of evidence in support of the conclusion that most of the matter in our universe is non-luminous. This includes observations of the rotational speeds of galaxies [@rotationcurves], the orbital velocities of galaxies within clusters [@clusters], gravitational lensing [@lensing], the cosmic microwave background [@wmap], the light element abundances [@bbn] and large scale structure [@lss]. Despite these many observational indications of dark matter, it is clear that it does not consist of baryonic material or other known forms of matter. For the time being, we remain ignorant of the particle identity of this substance.
In this review, we summarize some of the most promising strategies and techniques being pursued to elucidate the nature of dark matter. These efforts include direct detection experiments designed to observe the elastic scattering of dark matter particles with nuclei, indirect detection experiments which hope to detect the annihilation products of dark matter such gamma rays, neutrinos, positrons, antiprotons, antideuterons, synchrotron radiation and X-rays, and collider searches for dark matter and associated particles.
The material presented here is not intended to be an exhaustive summary of the field of dark matter physics. Such reviews can be found elsewhere [@Bertone:2004pz]. In this article, we limit our discussion to the case of candidate dark matter particles with electroweak scale masses and couplings. Furthermore, for the sake of length, we do not discuss every experimental approach being pursued, but instead focus on several of the most promising direct, indirect and collider efforts, and on the interplay and complementarity between these various programs.
We would like to emphasize that the detection of dark matter particles in any one of the experimental channels discussed here will not alone be sufficient to conclusively identify the nature of dark matter. The direct or indirect detection of the dark matter particles making up our galaxy’s halo is highly unlikely to provide enough information to reveal the underlying physics (supersymmetry, etc.) behind these particles. In contrast, collider experiments may identify a long-lived, weakly interacting particle, but will not be able to test its cosmological stability or abundance. Only by combining the information provided by many different experimental approaches is the mystery of dark matter’s particle nature likely to be solved. Although the detection of dark matter in any one search channel would constitute a discovery of the utmost importance, it would almost certainly leave many important questions unanswered.
THE WIMP HYPOTHESIS
===================
In this review, we limit our discussion to dark matter candidates which are heavy, electrically neutral and weakly-interacting. This class of particles, collectively known as WIMPs, are particularly well motivated, especially when their mass and couplings are tied to the physics of the electroweak scale. Before we discuss the experimental techniques for detecting dark matter particles, we will briefly discuss some of the most compelling motivations for electroweak-scale dark matter.
The challenge of stabilizing the mass of the Higgs boson ([*ie.*]{} the hierarchy problem) leads us to expect new forms of matter to appear at or near the electroweak scale. The nature of any physics beyond the Standard Model which might appear at the TeV scale, however, is tightly constrained by the precision electroweak measurements made at LEP. In particular, new discrete symmetries are required of most phenomenologically viable models of TeV scale physics [@hier]. Such symmetries naturally lead to a stable particle or particles, which may potentially constitute the dark matter of our universe.
A number of extensions of the Standard Model have been proposed which introduce new particle content at or near the electroweak scale, and which include a discrete symmetry of the form required to stabilize a potential dark matter candidate. The most well studied example is the lightest neutralino in supersymmetric models. Others examples include Kaluza-Klein hypercharge gauge bosons in models with universal extra dimensions [@kkdm], and the lightest T-parity odd particle in little Higgs theories [@lh].
Each of these candidates have similar masses and couplings, and thus will undergo similar thermal histories in the early universe. At high temperatures, WIMPs are abundant, being freely created and annihilated in pairs. As the universe expands and the temperature drops below the WIMPs’ production threshold, however, the number density of these particles becomes rapidly suppressed. Ultimately, the WIMPs will “freeze out” and remain as a thermal relic of the universe’s hot youth. The resulting density of WIMPs is given by: \_h\^2 = [s\_0\_c/h\^2]{} ( [45g\_\*]{})\^[1/2]{} [ x\_f m\_]{}[1]{}, where $s_0$ is the current entropy density of the universe, $\rho_c$ is the critical density, $h$ is the (scaled) Hubble constant, $g_*$ is the effective number of relativistic degrees of freedom at the time that the dark matter particle goes out of thermal equilibrium, $m_\Pl$ is the Planck mass, $x_f = m/T_f \approx 25$ is the inverse freeze-out temperature in units of the WIMP mass, and $\VEV{\sigma v}$ is the thermal average of the dark matter pair annihilation cross section times the relative velocity.
In order for this process to yield a thermal abundance of dark matter within the range measured by WMAP ($0.095 < \Omega h^2 <
0.129$) [@wmap], the thermally averaged annihilation cross section is required to be $\sigv \approx 3 \times 10^{-26}$ cm$^3$/s (or alternatively, $\sigv \approx
0.9$ pb). Remarkably, this is quite similar to the value obtained for a generic electroweak mass particle annihilating through the exchange of the electroweak gauge or Higgs bosons. In particular, we notice that $\VEV{\sigma v} = \pi
\alpha^2/8m^2$ leads us to a WIMP mass on the order of $m \sim 100$ GeV.
We conclude that if a stable, weakly interacting, electroweak-scale particle exists, then it is likely to be present in the universe today with an abundance similar to the measured dark matter density. With this in mind, we focus our dark matter search strategy on this particularly well motivated scenario in which the dark matter particle has electroweak interactions and a mass near the electroweak scale.
DIRECT DETECTION {#sec:section5}
================
Experiments such as XENON [@xenon], CDMS [@cdmssi; @cdmssd], ZEPLIN [@zeplin], Edelweiss [@edelweiss], CRESST [@cresst], WARP [@warp] and COUPP [@coupp] are designed to detect dark matter particles through their elastic scattering with nuclei. This class of techniques is collectively known as direct detection, in contrast to indirect detection efforts which attempt to observe the annihilation products of dark matter particles.
The role played by direct detection is important for a number of reasons. Firstly, although collider experiments may be capable of detecting dark matter particles, they will not be able to distinguish a cosmologically stable WIMP from a long-lived but unstable particle. More generally speaking, colliders will not inform us as to the cosmological abundance of a WIMP they might observe. Furthermore, while the mass of the dark matter particle could potentially be measured by a collider experiment such as the LHC, its couplings are much more difficult to access in this way. Direct detection experiments, in contrast, provide a valuable probe of the dark matter’s couplings to the Standard Model. Finally, the uncertainties involved in direct detection are likely to be significantly smaller than in most indirect detection channels. Whereas indirect detection rates rely critically on the distribution of dark matter, especially in high density regions, and on other astrophysical properties such as the galactic magnetic and radiation fields, direct detection experiments rely only on the local dark matter density and velocity distribution.
The density of dark matter in the local neighborhood is inferred by fitting observations to models of the galactic halo. These observations including the rotational speed of stars at the solar circle and other locations, the total projected mass density (estimated by considering the motion of stars perpendicular to the galactic disk), peak-to-trough variations in the rotation curve ([*ie.*]{} the ‘flatness constraint’), and microlensing. Taken together, these constraints can be used to estimate the local halo density to lie between 4 $\times$ 10$^{-25}$ g/cm$^{-3}$ and 13 $\times$ 10$^{-25}$ g/cm$^3$ ($0.22-0.73\,$GeV/cm$^3$) [@GGTurner]. Limits on the density of MACHO microlensing objects imply that at least 80% of this is cold dark matter. The velocity of the WIMPs is expected be close to the galactic rotation velocity, $230 \pm 20$ km/sec [@Drukier].
These observations, however, only constrain the dark matter density as averaged over scales larger than a kiloparsec or so. In contrast, the solar system moves a distance of $\sim$$10^{-3}$ parsecs relative to the dark matter halo each year. If dark matter is distributed in an inhomogeneous way over milliparsec scales ([*ie.*]{} as a collection of dense clumps and voids), then the density along the path of the Earth, as seen by direct detection experiments, could be much larger or smaller than is inferred by the rotational dynamics of our galaxy.
Throughout most of our galaxy’s halo, however, inhomogeneities in the small scale dark matter distribution are not anticipated to be large. The vast majority of the dark matter in the inner regions of our galaxy has been in place for $\sim$$10^{10}$ years; ample time for the destruction of clumps through tidal interactions. Using high-resolution simulations, Helmi, White and Springel find that the dark matter in the solar neighborhood is likely to consist of a superposition of hundreds of thousands of dark matter streams, collectively representing a very smooth and homogeneous distribution [@white]. That being said, if we happen to find our Solar System residing in a overdense clump or stream of dark matter, high direct detection rates could lead us to mistakenly infer an artificially large WIMP-nucleon elastic scattering cross section.
The nuclear physics involved in WIMP-nuclei elastic scattering also introduces uncertainties which may ultimately limit the accuracy to which the dark matter’s couplings to the Standard Model can be measured. In many models, including many supersymmetric models, the WIMP-nucleon scattering cross section is dominated by the $t$-channel exchange of a Higgs boson. The coupling of the Higgs boson to the proton receives its dominant contributions from two sources, the coupling of the Higgs to gluons through a heavy quark loop and the direct coupling of the Higgs to strange quarks [@jungman]. That means that this coupling depends on the parameter $$f_{Ts} = {\bra{p} m_s \bar s s \ket{p} \over \bra{p} H_{QCD}
\ket{p}} \ ,
\label{fTsdefin}$$ that is, the fraction of the mass of the proton that arises from the mass of the non-valence strange quarks in the proton wavefunction. It has been known for some time that there is significant uncertainty in this quantity [@KaplanNelson], and several recent papers have pointed out the uncertainty this introduces to calculations of the WIMP-nucleon elastic scattering cross section [@Bottino; @EllisUpdate]. In particular, in the case of WIMPs which couple dominantly to the strange content of the nucleon, this can lead to an uncertainty in the direct detection cross section of a factor of 4 or even larger [@Bottino]. It is possible that this uncertainty could be reduced in the future through the use of lattice gauge theory [@UKQCD; @Weise].
The processes of WIMP-nuclei elastic scattering can be naturally divided into spin-dependent and spin-independent contributions. The spin-independent, or coherent scattering, term is enhanced in WIMP-nucleus cross sections by factors of $A^2$, making it advantageous to use targets consisting of heavy nuclei. This enhancement is due to the fact that the WIMP wavelength is of order the size of the nucleus, thus the scattering amplitudes on individual nucleons add coherently. The spin-dependent contribution, in contrast, couples to the spin of the target nuclei and scales with $J (J+1)$. Naively, this could be considered a coherent subtraction of amplitudes of opposite signs of pairs of nucleons. As the current spin-dependent scattering constraints are not strong enough to test many dark matter models, we devote our attention primarily to the process of spin-independent scattering.
The spin-independent WIMP-nucleus elastic scattering cross section is given by: $$\label{sig}
\sigma \approx \frac{4 m^2_{X} m^2_{T}}{\pi (m_{X}+m_T)^2} [Z f_p + (A-Z) f_n]^2,$$ where $m_T$ is the mass of the target nucleus, $m_X$ is the WIMP’s mass and $Z$ and $A$ are the atomic number and atomic mass of the nucleus. $f_p$ and $f_n$ are the WIMP’s couplings to protons and neutrons, given by: $$f_{p,n}=\sum_{q=u,d,s} f^{(p,n)}_{T_q} a_q \frac{m_{p,n}}{m_q} + \frac{2}{27} f^{(p,n)}_{TG} \sum_{q=c,b,t} a_q \frac{m_{p,n}}{m_q},
\label{feqn}$$ where $a_q$ are the WIMP-quark couplings and $f^{(p,n)}_{T_q}$ denote the quark content of the nucleon.
The first term in Eq. \[feqn\] corresponds to interactions with the quarks in the target nuclei. In the case of neutralino dark matter, this can occur through either $t$-channel CP-even Higgs exchange, or $s$-channel squark exchange:
(222,254)(3,4.3) (5.,20.)(20.,10.)(0.,)[/Straight]{}[0]{} (0.,18.0)\[b\][$\chi^0$]{} (20.,-5.0)(20.,10.0)(0.,)[/ScalarDash]{}[0]{} ( 5.,-15.0)(20.,-5.0)(0.,)[/Straight]{}[+1]{} (0.,-18.0)\[b\][$q$]{} (25.,0.0)\[b\][$H, h$]{} (20.,10.)(35.,20.)(0.,)[/Straight]{}[0]{} (20.,-5.)(35.,-15.0)(0.,)[/Straight]{}[+1]{} (40.,18.0)\[b\][$\chi^0$]{} (40.,-18.)\[b\][$q$]{} (70.,14.)(85.,3.)(0.,)[/Straight]{}[0]{} (65.,12.0)\[b\][$\chi^0$]{}
(70.,-7.0)(85.,3.0)(0.,)[/Straight]{}[+1]{} (65.,-9.0)\[b\][$q$]{}
(85.,3.)(105.,3.0)(0.,)[/ScalarDash]{}[0]{} (95.,9.)\[t\][$\tilde{q}$]{}
(120.,14.)(105.,3.0)(0.,)[/Straight]{}[0]{} (125.,12.)\[b\][$\chi^0$]{}
(120.,-7.)(105.,3.0)(0.,)[/Straight]{}[-1]{} (125.,-9.)\[b\][$q$]{}
The second term corresponds to interactions with the gluons in the target through a loop diagram (a quark/squark loop in the case of supersymmetry). $f^{(p)}_{TG}$ is given by $1 -f^{(p)}_{T_u}-f^{(p)}_{T_d}-f^{(p)}_{T_s}
\approx 0.84$, and analogously, $f^{(n)}_{TG} \approx 0.83$.
Besides its mass, the only thing we need to know about the WIMP itself to calculate this cross section are its couplings to quarks, $a_q$. In the case of neutralino dark matter, the value of this coupling depends on many features of the supersymmetric spectrum. The contribution resulting from Higgs exchange depends on the neutralino composition, as well as the Higgs masses and couplings.
In the case of heavy squarks, small wino component and little mixing between the CP-even Higgs bosons ($\cos \alpha \approx 1$), neutralino-nuclei elastic scattering is dominated by $H$ exchange with strange and bottom quarks, leading to a neutralino-nucleon cross section approximately given by: $$\sigma_{\chi N} \sim \frac{g^2_1 g^2_2 f_{\tilde{B}} f_{\tilde{H}} \,m^4_N}{4\pi m^2_W \cos^2 \beta \, m^4_H} \bigg(f_{T_s}+\frac{2}{27}f_{TG}\bigg)^2, \,\,\,\, (m_{\tilde{q}}\, \rm{large}, \cos \alpha \approx 1),
\label{case1}$$ where $f_{\tilde{B}}$ and $f_{\tilde{H}}$ denote the bino and higgsino fractions of the lightest neutralino and $\tan \beta$ is the ratio of the vevs of the two Higgs doublets in the MSSM. Note that the coupling involves the product of $f_{\tilde{B}}$ and $f_{\tilde{H}}$. Neutralinos that are purely gaugino-like or purely higgsino-like have zero cross section with nuclei. The fundamental reason for this is that the relevant vertex is gaugino-higgsino-Higgs.
If the heavier of the two CP-even Higgs bosons is very heavy and/or $\tan \beta$ is small, scattering with up-type quarks through light Higgs exchange can dominate: $$\sigma_{\chi N} \sim \frac{g^2_1 g^2_2 f_{\tilde{B}} f_{\tilde{H}} \, m^4_N}{4\pi m^2_W \, m^4_h} \bigg(f_{T_u}+\frac{4}{27}f_{TG}\bigg)^2, \,\,\,\, (m_{\tilde{q}}, m_H\, \rm{large}, \cos \alpha \approx 1).
\label{case2}$$ If $\tan \beta$ and $m_H$ are large and the squarks somewhat light, elastic scattering can instead be dominated by squark exchange: $$\sigma_{\chi N} \sim \frac{g^2_1 g^2_2 f_{\tilde{B}} f_{\tilde{H}} \, m^4_N}{4\pi m^2_W \cos^2 \beta \, m^4_{\tilde{q}}} \bigg(f_{T_s}+\frac{2}{27}f_{TG}\bigg)^2, \,\,\,\, (\tilde{q}\,\, \rm{dominated}, \tan \beta \gg 1).
\label{case3}$$ From these expressions [@scatteraq], it is clear that the direct detection of dark matter alone will not be very capable of revealing much about supersymmetry or the other underlying physics. There are a large number of degeneracies which can lead to a given value of the WIMP-nucleon cross section. Only by combining this information with collider and/or indirect detection data can one hope to infer the nature of the dark matter particle.
Currently, the strongest limits on the spin-independent WIMP-nucleon cross section come from the XENON [@xenon] and CDMS [@cdmssi] experiments, which have obtained an upper bound on the cross section of a $\sim$100 GeV WIMP at the $\sim$$10^{-7}$ pb level. These constraints, along with those of other experiments, are shown in Fig. \[directcurrent\].
There is currently a great deal of progress being made in the experimental field of direct detection. Within the next several months (early 2008), the CDMS collaboration is expected to release a new limit which will likely be the most stringent (assuming no detection is made). In the meantime, the XENON collaboration is preparing for a run with a larger detector, with results expected within a year or so of this time. Beyond the next year or two, it is difficult to foresee which experiment(s) will be leading this search. It is still not clear whether detectors using liquid noble elements or cryogenic technologies will advance most rapidly. For the time being, there are clear advantages to proceeding with multiple technologies.
Despite our inability to predict how this field will develop, it is reasonable to expect that by 2010 or so direct detection experiments will reach the $\sim 10^{-9}$ pb level of sensitivity. Roughly speaking, such cross sections are sufficient to test many, if not most, supersymmetric models, as well as many WIMPs candidates in other particle physics frameworks.
Given the rate at which direct dark matter experiments are developing, it is interesting to recognize that such experiments are likely to see their first evidence for WIMPs within the same time frame that the Large Hadron Collider is expected to reveal the presence of the associated physics. In such a scenario, it will be essential to compare the mass of the WIMP observed in each experimental program.
Direct detection experiments can determine the mass of the WIMP by measuring the distribution of the recoil energy, $E_R$ [@Green:2007rb]. This varies with the mass of the WIMP, with a resonance where the WIMP mass equals the target mass. Roughly, one expects , where $m_T$ is the target mass and $v$ is the WIMP velocity, with corrections depending on the precise target material and the properties of the detector [@LewinSmith]. Assuming the standard velocity distribution in smooth halo models, with approximately 10% uncertainty, an experiment with a Xenon or Germanium target that detects 100 signal events for a WIMP of mass $m_X = 100$ GeV can expect to measure the mass of this particle at the 20% level, thus potentially confirming the cosmological stability (and abundance) of a WIMP detected at the LHC.
If the WIMP mass inferred in a direct detection experiment was not consistent with that measured at the LHC, this could imply that different particle species are being observed, or could be the result of a nonstandard dark matter velocity distribution. In the future, directional dark matter detectors may help to clear up such a scenario.
INDIRECT DETECTION
==================
In parallel to direct detection experiments, a wide range of indirect detection programs have been developed to search for the annihilation products of dark matter particles. In particular, searches are underway to detect neutrinos from dark matter annihilations in the core of the Sun, antimatter particles from dark matter annihilations in the galactic halo, and photons from dark matter annihilations in the halo of the Milky Way, galactic substructure and the dark matter distribution integrated over cosmological volumes. In this section, we briefly describe the role of these experimental programs in the overall strategy to reveal dark matter’s identity.
Gamma Rays
----------
Searches for prompt photons generated in dark matter annihilations have a key advantage over other indirect detection channels in that they travel essentially unimpeded from their production site. In particular, gamma rays are not deflected by magnetic fields, and thus can potentially provide valuable angular information. For example, point sources of dark matter annihilation radiation might appear from high density regions such as the Galactic Center or dwarf spheroidal galaxies. Furthermore, over galactic distance scales, gamma rays are not attenuated, and thus retain their spectral information. In other words, the spectrum observed at Earth is the same spectrum that was generated in the dark matter annihilations.
The spectrum of photons produced in dark matter annihilations depends on the details of the WIMP being considered. Supersymmetric neutralinos, for example, typically annihilate to final states consisting of heavy fermions and gauge or Higgs bosons [@jungman]. Generally speaking, each of these annihilation modes typically result in a very similar spectrum of gamma rays (see, however, Ref. [@bringmann]). The gamma ray spectrum from a WIMP which annihilates to light leptons can be quite different, however. This can be particularly important in the case of Kaluza-Klein dark matter in models with one universal extra dimension, for example, in which dark matter particles annihilate significantly to $e^+ e^-$ and $\mu^+ \mu^-$ [@kkdm].
The Galactic Center has long been considered to be one of the most promising regions of the sky in which to search for gamma rays from dark matter annihilations [@gchist]. The prospects for this depend, however, on a number of factors including the nature of the WIMP, the distribution of dark matter in the region around the Galactic Center, and our understanding of the astrophysical backgrounds.
The gamma ray spectrum produced through dark matter annihilations is given by $$\Phi_{\gamma}(E_{\gamma},\psi) = \frac{1}{2}\sigv \frac{dN_{\gamma}}{dE_{\gamma}} \frac{1}{4\pi m^2_X} \int_{\rm{los}} \rho^2(r) dl(\psi) d\psi.
\label{flux1}$$ Here, $\sigv$ is the thermally averaged WIMP annihilation cross section, $m_X$ is the mass of the WIMP, $\psi$ is the angle observed relative to the direction of the Galactic Center, $\rho(r)$ is the dark matter density as a function of distance to the Galactic Center, and the integral is performed over the line-of-sight. $dN_{\gamma}/dE_{\gamma}$ is the gamma ray spectrum generated per WIMP annihilation.
Averaging over a solid angle centered around a direction, $\psi$, we arrive at $$\Phi_{\gamma}(E_{\gamma}) \approx 2.8 \times 10^{-12} \, \rm{cm}^{-2} \, \rm{s}^{-1} \, \frac{dN_{\gamma}}{dE_{\gamma}} \bigg(\frac{\sigv}{3 \times 10^{-26} \,\rm{cm}^3/\rm{s}}\bigg) \bigg(\frac{1 \, \rm{TeV}}{m_{\rm{X}}}\bigg)^2 J(\Delta \Omega, \psi) \Delta \Omega,
\label{flux2}$$ where $\Delta \Omega$ is the solid angle observed. The quantity $J(\Delta \Omega, \psi)$ depends only on the dark matter distribution, and is the average over the solid angle of the quantity: $$J(\psi) = \frac{1}{8.5 \, \rm{kpc}} \bigg(\frac{1}{0.3 \, \rm{GeV}/\rm{cm}^3}\bigg)^2 \, \int_{\rm{los}} \rho^2(r(l,\psi)) dl.$$ $J(\psi)$ is normalized such that a completely flat halo profile, with a density equal to the value at the solar circle, integrated along the line-of-sight to the Galactic Center would yield a value of one. In dark matter distributions favored by N-body simulations, however, this value can be much larger. The Narvarro-Frenk-White profile [@nfw], which is a commonly used benchmark halo model, leads to values of $J(\Delta \Omega=10^{-5} \, {\rm sr}, \psi=0) \sim 10^5$. The effects adiabatic contraction due to the cooling of baryons is further expected to increase this quantity [@ac].
The recent discovery of a bright, very high-energy gamma ray source in the galactic center region by the Atmospheric Cerenkov Telescopes HESS [@hess], MAGIC [@magic], WHIPPLE [@whipple] and CANGAROO-II [@cangaroo] has made efforts to identify gamma rays from dark matter annihilations more difficult. This source appears to be coincident with the dynamical center of the Milky Way (Sgr A$^*$) and has no detectable angular extension (less than 1.2 arcminutes). Its spectrum is well described by a power-law, $dN_{\gamma}/dE_{\gamma} \propto E_{\gamma}^{-\alpha}$, where $\alpha=2.25 \pm
0.04 (\rm{stat}) \pm 0.10 (\rm{syst})$ over the range of 160 GeV to 20 TeV. Although speculations were initially made that this source could be the product of annihilations of very heavy ($\gsim 10$ TeV) dark matter particles [@actdark], the spectral shape appears inconsistent with a dark matter interpretation. The source of these gamma rays is more likely an astrophysical accelerator associated with our Galaxy’s central supermassive black hole [@hessastro]. Although this gamma ray source represents a formidable background for GLAST and other experiments searching for dark matter annihilation radiation [@gabi], it may be possible to reduce the impact of this and other backgrounds by studying the angular distribution of gamma rays from this region of the sky [@Dodelson:2007gd].
The prospects for identifying dark matter annihilation radiation from the Galactic Center depends critically on the unknown dark matter density within the inner parsecs of the Milky Way and on the properties of the astrophysical backgrounds present. If these characteristics are favorable, then the Galactic Center is very likely to be the most promising region of the sky to study. If not, other regions with high dark matter densities may be more advantageous.
Dwarf spheroidal galaxies within and near the Milky Way provide an opportunity to search for dark matter annihilation radiation with considerably less contamination from astrophysical backgrounds. The flux of gamma rays from dark matter annihilations in such objects, however, is also expected to be lower than from a cusp in the center of the Milky Way [@Evans:2003sc; @Bergstrom:2005qk; @Strigari:2007at]. As a result, planned experiments are likely to observe dark matter annihilation radiation from dwarf galaxies only in the most favorable range of particle physics models.
The integrated gamma ray signal from dark matter annihilations throughout the cosmological distribution of dark matter may also provide an opportunity to identify the products of dark matter annihilations. The ability of future gamma ray telescopes to identify a dark matter component of the diffuse flux depends strongly on the fraction of the extragalactic gamma ray background observed by the EGRET experiment which will be resolved as individual sources, such as blazars. If a large fraction of this background is resolved, the remaining extragalactic signal could potentially contain identifiable signatures of dark matter annihilations [@egdiffuse].
The telescopes potentially capable of detecting gamma rays from dark matter annihilations in the near future include the satellite-based experiment GLAST [@glast], and a number ground based Atmospheric Cerenkov Telescopes, including HESS, MAGIC and VERITAS. The roles played by each of these two classes of experiments in the search for dark matter are quite different. GLAST will continuously observe a large fraction of the sky, but with an effective area far smaller than possessed by ground based telescopes. Ground based telescopes, in contrast, study the emission from a small angular field, but with far greater exposure. Furthermore, while ground based telescopes can only study gamma rays with energy greater than $\sim$100 GeV, GLAST will be able to directly study gamma rays with energies over the range of 100 MeV to 300 GeV.
As a result of the different energy ranges accessible by these experiments, searches for dark matter particles lighter than a few hundred GeV are most promising with GLAST, while ground based telescopes are better suited for heavier WIMPs. The large field-of-view of GLAST also makes it well suited for measurements of the diffuse gamma ray background. GLAST is also expected to detect a number of unidentified sources, some of which could potentially be signals of dark matter substructures. Follow up observations with ground based gamma ray telescopes would be very useful for clarifying the nature of such sources.
Antimatter
----------
WIMP annihilations in the galactic halo generate charged anti-matter particles: positrons, anti-protons and anti-deuterons. Unlike gamma rays, which travel along straight lines, charged particles move under the influence of the Galactic Magnetic Field, diffusing and losing energy, resulting in a diffuse spectrum at Earth. By studying the cosmic anti-matter spectra, satellite-based experiments such as PAMELA [@pamela] and AMS-02 [@ams02] may be able to identify signatures of dark matter. PAMELA began its three-year satellite mission in June of 2006. AMS-02 is planned for later deployment onboard the International Space Station.
As compared to antiprotons and antideuterons, cosmic positrons are attractive probes of dark matter for several reasons. In particular, positrons lose the majority of their energy over typical length scales of a few kiloparsecs or less [@baltzpos]. The cosmic positron spectrum, therefore, samples only the local dark matter distribution and is thus subject to considerably less uncertainty than the other anti-matter species. Additionally, data from the HEAT [@heat] and AMS-01 experiments [@ams01] contain features which could plausibly be the consequence of dark matter annihilations in the local halo.
The spectral shape of the cosmic positron spectrum generated in dark matter annihilation depends on the leading annihilation modes of the WIMP in the low velocity limit. Bino-like neutralinos, for example, typically annihilate to heavy fermion pairs: $b\bar{b}$ with a small $\tau^+ \tau^-$ admixture, along with a fraction to $t\bar{t}$ if $m_{\chi} \gsim m_t$. Wino or higgsino-like neutralinos annihilate most efficiently to combinations of Higgs and gauge bosons. In other particle dark matter candidates, such as Kaluza-Klein dark matter in models with universal extra dimensions, annihilation to light charged leptons can lead to a much harder positron spectrum than is expected from neutralinos [@kkpos].
Once positrons are injected into the local halo through dark matter annihilations, they propagate under the influence of galactic magnetic fields, gradually losing energy through synchrotron emission and inverse Compton scattering with radiation fields, such as starlight and the cosmic microwave background. The spectrum observed at Earth is found by solving the diffusion-loss equation [@diffusion]: $$\begin{aligned}
\frac{\partial}{\partial t}\frac{dn_{e^{+}}}{dE_{e^{+}}} = \vec{\bigtriangledown} \cdot \bigg[K(E_{e^{+}},\vec{x}) \vec{\bigtriangledown} \frac{dn_{e^{+}}}{dE_{e^{+}}} \bigg] + \frac{\partial}{\partial E_{e^{+}}} \bigg[b(E_{e^{+}},\vec{x})\frac{dn_{e^{+}}}{dE_{e^{+}}} \bigg] + Q(E_{e^{+}},\vec{x}),
\label{dif}\end{aligned}$$ where $dn_{e^{+}}/dE_{e^{+}}$ is the number density of positrons per unit energy, $K(E_{e^{+}},\vec{x})$ is the diffusion constant, $b(E_{e^{+}},\vec{x})$ is the rate of energy loss and $Q(E_{e^{+}},\vec{x})$ is the source term, which contains all of the information about the dark matter annihilation modes, cross section, and distribution. To solve the diffusion-loss equation, a set of boundary conditions must be adopted. In this application, the boundary condition is described as the distance from the galactic plane at which the positrons can freely escape, $L$. These diffusion parameters can be constrained by studying the spectra of various species of cosmic ray nuclei, most importantly the boron-to-carbon ratio [@btoc].
In Fig. \[heat\], the ratio of positrons to positrons plus electrons in the cosmic ray spectrum is shown as a function of energy, including a possible contribution from dark matter annihilations. Also shown are the measurements from the HEAT experiment [@heat], which may possibly contain an excess in comparison to standard astrophysical expectations at energies above 7 GeV or so. While positrons from dark matter annihilations are indeed able to generate this possible excess, it requires a somewhat larger annihilation rate than is typically expected. In particular, if a smooth dark matter halo and an annihilation cross section of $\sigma v \approx 3\times 10^{-26}$ cm$^{3}$/s (as required to thermally produce the observed dark matter abundance via S-wave processes) are assumed, the annihilation rate will be a factor of 50 or more too low to generate the spectrum measured by HEAT. Fluctuations in the local dark matter density, however, could lead to enhancements in the local annihilation rate, known as the “boost factor”. It is typically expected that this quantity could be as large as 5 to 10. Although boost factors of 50 or more are not impossible, such large values would be somewhat surprising.
If the positron flux observed by HEAT is in fact the result of annihilating dark matter, then the corresponding spectrum will be precisely measured by PAMELA [@pamela] and AMS-02 [@ams02]. If not, then the detection of positrons from dark matter annihilations will be more difficult, but perhaps still possible [@silkpos].
Unlike gamma ray measurements of the Galactic Center or dwarf galaxies, observations of the cosmic positron spectrum (as well as the antiproton and antideuteron spectra) could potentially provide a measurement of the dark matter annihilation rate over large volumes of space. Such a measurement, therefore, could be used to determine the product of the WIMP’s annihilation cross section and its density squared, averaged over the sampled volume (roughly a few cubic kiloparsec region, corresponding to the distance a typical positron travels from its point of origin before losing the majority of its energy). As a result of this limited range, only the dark matter distribution in the local halo is relevant to the observed cosmic positron flux. Assuming there are no very large and unknown clumps of dark matter in the surrounding kiloparsecs (which, although not impossible, is very unlikely [@hoopertaylorsilk]), a measurement of the cosmic positron spectrum could be used to infer the dark matter particle’s annihilation cross section (in the low velocity limit) with a comparatively modest degree of uncertainty coming from the unknown distribution of dark matter.
Neutrino Telescopes
-------------------
Although dark matter annihilations in the galactic halo produce too few neutrinos to be detected [@neutrinoshalo], annihilations which occur in the center of the Sun could potentially generate an observable flux of high energy neutrinos [@neutrinosun].
Dark matter particles scatter elastically with and become captured in the Sun at a rate given by [@capture] $$C^{\odot} \approx 3.35 \times 10^{19} \, \mathrm{s}^{-1}
\left( \frac{\sigma_{\mathrm{H, SD}} +\, \sigma_{\mathrm{H, SI}}
+ 0.07 \, \sigma_{\mathrm{He, SI}} } {10^{-7}\, \mathrm{pb}} \right)
\left( \frac{100 \, \mathrm{GeV}}{m_{X}} \right)^2 ,
\label{capture}$$ where $m_{X}$ is the dark matter particle’s mass. $\sigma_{\mathrm{H, SD}}$, $\sigma_{\mathrm{H, SI}}$ and $\sigma_{\mathrm{He, SI}}$ are the spin dependent (SD) and spin independent (SI) elastic scattering cross sections of the WIMP with hydrogen and helium nuclei, respectively. The factor of $0.07$ reflects the solar abundance of helium relative to hydrogen and well as dynamical factors and form factor suppression.
Notice that the capture rate is suppressed by two factors of the WIMP mass. One of these is simply the result of the depleted number density of WIMPs in the local halo ($n \propto 1/m$) while the second factor is the result of kinematic suppression for the capture of a WIMP much heavier than the target nuclei, in this case hydrogen or helium. If the WIMP’s mass were comparable to the masses of hydrogen or helium nuclei, these expressions would no longer be valid. For WIMPs heavy enough to generate neutrinos detectable in the high-energy neutrino telescopes, Eq. \[capture\] should be applicable.
If the capture rate and annihilation cross sections are sufficiently large, equilibrium will be reached between these processes. For a number of WIMPs in the Sun, $N$, the rate of change of this quantity is given by $$\dot{N} = C^{\odot} - A^{\odot} N^2 ,$$ where $C^{\odot}$ is the capture rate and $A^{\odot}$ is the annihilation cross section times the relative WIMP velocity per volume. $A^{\odot}$ can be approximated by $$A^{\odot} \approx \frac{\sigv}{V_{\mathrm{eff}}},$$ where $V_{\mathrm{eff}}$ is the effective volume of the core of the Sun determined roughly by matching the core temperature with the gravitational potential energy of a single WIMP at the core radius. This was found in Refs. [@equ1; @equ2] to be $$V_{\rm eff} \approx 5.7 \times 10^{27} \, \mathrm{cm}^3
\left( \frac{100 \, \mathrm{GeV}}{m_{X}} \right)^{3/2} \;.$$ The present WIMP annihilation rate in the Sun is given by $$\Gamma = \frac{1}{2} A^{\odot} N^2 = \frac{1}{2} \, C^{\odot} \,
\tanh^2 \left( \sqrt{C^{\odot} A^{\odot}} \, t_{\odot} \right) \;,$$ where $t_{\odot} \approx 4.5$ billion years is the age of the solar system. The annihilation rate is maximized when it reaches equilibrium with the capture rate. This occurs when $$\sqrt{C^{\odot} A^{\odot}} t_{\odot} \gg 1 \; .$$ If this condition is met, the final annihilation rate (and corresponding neutrino flux and event rate) has no further dependence on the dark matter particle’s annihilation cross section.
WIMPs can generate neutrinos through a wide range of annihilation channels. Annihilations to heavy quarks, tau leptons, gauge bosons and Higgs bosons can all generate neutrinos in the subsequent decay. In some models, WIMPs can also annihilate directly to neutrino pairs.
Once produced, neutrinos can travel to the Earth where they can be detected. The muon neutrino spectrum at the Earth from WIMP annihilations in the Sun is given by: $$\frac{dN_{\nu_{\mu}}}{dE_{\nu_{\mu}}} = \frac{ C_{\odot} F_{\rm{Eq}}}{4 \pi D_{\rm{ES}}^2} \bigg(\frac{dN_{\nu}}{dE_{\nu}}\bigg)^{\rm{Inj}},
\label{wimpflux}$$ where $C_{\odot}$ is the WIMP capture rate in the Sun, $F_{\rm{Eq}}$ is the non-equilibrium suppression factor ($\approx 1$ for capture-annihilation equilibrium), $D_{\rm{ES}}$ is the Earth-Sun distance and $(\frac{dN_{\nu}}{dE_{\nu}})^{\rm{Inj}}$ is the neutrino spectrum from the Sun per WIMP annihilating. Due to $\nu_{\mu}-\nu_{\tau}$ vacuum oscillations, the muon neutrino flux from WIMP annihilations in the Sun observed at Earth is the average of the $\nu_{\mu}$ and $\nu_{\tau}$ components.
Muon neutrinos produce muons in charged current interactions with ice or water nuclei inside or near the detector volume of a high energy neutrino telescope. The rate of neutrino-induced muons observed in a high-energy neutrino telescope is estimated by: $$N_{\rm{events}} \simeq \int \int \frac{dN_{\nu_{\mu}}}{dE_{\nu_{\mu}}}\, \frac{d\sigma_{\nu}}{dy}(E_{\nu_{\mu}},y) \,R_{\mu}((1-y)\,E_{\nu})\, A_{\rm{eff}} \, dE_{\nu_{\mu}} \, dy,$$ where $\sigma_{\nu}(E_{\nu_{\mu}})$ is the neutrino-nucleon charged current interaction cross section, $(1-y)$ is the fraction of neutrino energy which goes into the muon, $A_{\rm{eff}}$ is the effective area of the detector, $R_{\mu}((1-y)\,E_{\nu})$ is the distance a muon of energy, $(1-y)\,E_{\nu}$, travels before falling below the muon energy threshold of the experiment (ranging from $\sim$1 to 100 GeV), called the muon range.
The spectrum and flux of neutrinos generated in WIMP annihilations depends on the annihilation modes which dominate, and thus is model dependent. For most annihilation modes, however, the variation from model-to-model is not dramatic. In Fig. \[ratecompare\], the event rate in a kilometer-scale neutrino telescope (with a 50 GeV muon energy threshold) is shown as a function of the WIMP’s effective elastic scattering cross section for a variety of annihilation modes [@Halzen:2005ar]. The effective elastic scattering cross section is defined as $\sigma_{\rm{eff}} = \sigma_{\mathrm{H, SD}} +\, \sigma_{\mathrm{H,
SI}} + 0.07 \, \sigma_{\mathrm{He, SI}}$, following Eq. \[capture\]. These rates are indicative of that expected for experiments such as IceCube at the South Pole [@icecube], or a future kilometer-scale neutrino telescope built in the Mediterranean Sea [@km3]. To detect neutrinos from WIMP annihilations in the Sun over the background of atmospheric neutrinos, a rate in the range of 10-100 events per square-kilometer, per year is required.
![The event rate in a kilometer-scale neutrino telescope such as IceCube as a function of the WIMP’s effective elastic scattering cross section in the Sun for a variety of annihilation modes. The effective elastic scattering cross section is defined as $\sigma_{\rm{eff}} = \sigma_{\mathrm{H, SD}} +\, \sigma_{\mathrm{H, SI}}
+ 0.07 \, \sigma_{\mathrm{He, SI}}$, following Eq. \[capture\]. The dashes, solid and dotted lines correspond to WIMPs of mass 100, 300 and 1000 GeV, respectively.[]{data-label="ratecompare"}](indirectratebb.ps "fig:"){width="2.2in"} ![The event rate in a kilometer-scale neutrino telescope such as IceCube as a function of the WIMP’s effective elastic scattering cross section in the Sun for a variety of annihilation modes. The effective elastic scattering cross section is defined as $\sigma_{\rm{eff}} = \sigma_{\mathrm{H, SD}} +\, \sigma_{\mathrm{H, SI}}
+ 0.07 \, \sigma_{\mathrm{He, SI}}$, following Eq. \[capture\]. The dashes, solid and dotted lines correspond to WIMPs of mass 100, 300 and 1000 GeV, respectively.[]{data-label="ratecompare"}](indirectrateWW.ps "fig:"){width="2.2in"}\
![The event rate in a kilometer-scale neutrino telescope such as IceCube as a function of the WIMP’s effective elastic scattering cross section in the Sun for a variety of annihilation modes. The effective elastic scattering cross section is defined as $\sigma_{\rm{eff}} = \sigma_{\mathrm{H, SD}} +\, \sigma_{\mathrm{H, SI}}
+ 0.07 \, \sigma_{\mathrm{He, SI}}$, following Eq. \[capture\]. The dashes, solid and dotted lines correspond to WIMPs of mass 100, 300 and 1000 GeV, respectively.[]{data-label="ratecompare"}](indirectratett.ps "fig:"){width="2.2in"} ![The event rate in a kilometer-scale neutrino telescope such as IceCube as a function of the WIMP’s effective elastic scattering cross section in the Sun for a variety of annihilation modes. The effective elastic scattering cross section is defined as $\sigma_{\rm{eff}} = \sigma_{\mathrm{H, SD}} +\, \sigma_{\mathrm{H, SI}}
+ 0.07 \, \sigma_{\mathrm{He, SI}}$, following Eq. \[capture\]. The dashes, solid and dotted lines correspond to WIMPs of mass 100, 300 and 1000 GeV, respectively.[]{data-label="ratecompare"}](indirectratetautau.ps "fig:"){width="2.2in"}
Currently, the Super-Kamiokande experiment has placed the strongest bounds on high-energy neutrinos from the direction of the Sun [@superk]. In this application, Super-K has two primary advantages over other neutrino detectors. Firstly, they have analyzed data over a longer period than most of their competitors, a total of nearly 1700 live days. Secondly, Super-K was designed to be sensitive to low energy ($\sim$GeV) neutrinos, which gives them an advantage in searching for lighter WIMPs. Super-K’s upper limit on neutrino-induced muons above 1 GeV from WIMP annihilations in the Sun is approximately 1000 to 2000 per square kilometer per year for WIMPs heavier than 100 GeV, and approximately 2000 to 5000 per square kilometer per year for WIMPs in the 20 to 100 GeV range. The precise value of these limits depends on the WIMP annihilation mode considered.
The Amanda-II [@amanda], Baksan [@baksan] and Macro [@macro] experiments have each placed limits on the flux of neutrino-induced muons from the Sun that are only slightly weaker than Super-Kamiokande’s. The limit placed by the Amanda experiment resulted from only 144 live days of data. Having operated the detector for seven years, Amanda is expected to produce significantly improved bounds in the future.
In addition to these experiments, the next generation neutrino telescopes IceCube and Antares are currently under construction at the South Pole and in the Mediterranean, respectively. IceCube, with a full cubic kilometer of instrumented volume, will be considerably more sensitive to WIMP annihilations in the Sun than other planned or existing experiments [@icecube]. Antares, with less than one tenth of the effective area of IceCube, will have the advantage of a lower energy threshold, and may thus be more sensitive to low mass WIMPs [@antares]. Beyond Antares, there are also plans to build a kilometer-scale detector in the Mediterranean Sea [@km3].
From Fig. \[ratecompare\], we see that a WIMP-proton elastic scattering cross section on the order of $10^{-6}$ pb or greater is needed if kilometer-scale neutrino telescopes are to detect a signal from dark matter annihilations. Elastic scattering cross sections of this size are constrained by the absence of a positive signal in direct detection experiments, however. Currently, the strongest constraints on the WIMP-nucleon, spin-independent elastic scattering cross section have been made by the XENON [@xenon] and CDMS experiments [@cdmssi], who each place limits below $10^{-6}$ pb. Therefore, if current or planned neutrino telescopes are to detect neutrinos from dark matter annihilations in the Sun, they must scatter elastically with nuclei in the Sun via spin-dependent interactions, which are far less strongly constrained by direct detection experiments. The strongest bounds on the WIMP-proton spin-dependent cross section have been made by the NAIAD experiment [@naiad]. This result limits the spin-dependent cross section with protons to be less than approximately 0.3 pb for a WIMP in the mass range of 50-100 GeV and less than 0.8 pb ($m_{X}$/500 GeV) for a heavier WIMP. The PICASSO [@picasso] and CDMS [@cdmssd] experiments have placed limits on the spin-dependent WIMP-proton cross section roughly one order of magnitude weaker than the NAIAD result.
A WIMP with a largely spin-dependent scattering cross section with protons may thus be capable of generating large event rates in high energy neutrino telescopes. Considering, for example, a 300 GeV WIMP with an elastic scattering cross section near the experimental limit, Fig. \[ratecompare\] suggests that rates as high as $\sim 10^6$ per year could be generated if purely spin-dependent scattering contributes to the capture rate of WIMPs in the Sun.
The relative size of the spin-independent and spin-dependent elastic scattering cross sections depend on the nature of the WIMP in question. For a neutralino, these cross sections depend on its composition and on the mass spectrum of the exchanged Higgs bosons and squarks. Spin dependent, axial-vector, scattering of neutralinos with quarks within a nucleon is made possible through the t-channel exchange of a $Z$, or the s-channel exchange of a squark. Spin independent scattering occurs at the tree level through s-channel squark exchange and t-channel Higgs exchange, and at the one-loop level through diagrams involving a loop of quarks and/or squarks.
For higgsino-like or mixed higgsino-bino neutralinos, the spin dependent cross section can be somewhat larger than the spin independent, which is potentially well suited for the prospects for indirect detection. In particular, spin-dependent cross sections as large as $\sim 10^{-3}$pb are possible even in models with very small spin-independent scattering rates. Such neutralinos would go easily undetected in all planned direct detection experiments, while still generating on the order of $\sim 1000$ events per year at IceCube.
Synchrotron Emission
--------------------
As described in Sec. \[antimatter\], electrons and positrons produced in dark matter annihilations travel under the influence of the Galactic Magnetic Field, losing energy through Compton scattering off of starlight, cosmic microwave background photons and far infrared emission from dust, and through synchrotron emission from interactions with the Galactic magnetic field. The relative importance of these processes depends on the energy densities of radiation and magnetic fields.
The processes of synchrotron emission and inverse Compton scattering each lead to potentially observable byproducts [@syndm]. For dark matter particles with electroweak scale masses, the resulting synchrotron photons typically fall in the microwave frequency band, and thus are well suited for study with cosmic microwave background experiments [@haze]. The inverse Compton scattering of highly relativistic electrons and positrons with starlight photons, on the other hand, can generate photons with MeV-GeV energies.
THE ROLE OF COLLIDERS
=====================
Among other new states, particles with TeV scale masses and QCD color are generic features of models of electroweak symmetry breaking. These particles appear as counterparts to the quarks to provide new physics associated with the generation of the large top quark mass. In many scenarios, including supersymmetry, electroweak symmetry breaking arises as a result of radiative corrections due to these particles, enhanced by the large coupling of the Higgs boson to the top quark.
Any particle with these properties will be pair-produced at the Large Hadron Collider (LHC) with a cross section of tens of picobarns [@sqgl]. That particle (or particles) will then subsequently decay to particles including quark or gluon jets and the lightest particle in the new sector ([*ie.*]{} the dark matter candidate) which proceeds to exit the detector unseen. For any such model, the LHC experiments are, therefore, expected to observe large numbers of events with many hadronic jets and an imbalance of measured momentum. These ‘missing energy’ events are signatures of a wide range of models that contains an electroweak scale candidate for dark matter.
If TeV-scale supersymmetry exists in nature, it will very likely be within the discovery reach of the Large Hadron Collider (LHC). The rate of missing energy events depends strongly on the mass of the colored particles that are produced and only weakly on other properties of the model. In Fig. \[fig:ATLAS\], the estimates of the ATLAS collaboration are shown for the discovery of missing energy events [@Tovey]. If squarks or gluinos have masses below 1 TeV, the missing energy events can be discovered with an integrated luminosity of 100 pb$^{-1}$, about 1% of the LHC first-year design luminosity. Thus, we will know very early in the LHC program that a WIMP candidate is being produced.
By studying the decays of squarks and/or gluinos it will also be possible to discover other superpartners at the LHC. For example, in many models, decays of the variety, $\tilde{q} \rightarrow \chi^0_2 q \rightarrow \tilde{l}^{\pm} l^{\mp} q \rightarrow \chi^0_1 l^+ l^- q$, provide a clean signal of supersymmetry in the form of $l^+l^- +\, \rm{jets}\, + \,\,\rm{missing}\,\, E_T$. By studying the kinematics of these decays, the quantities $m_{\tilde{q}}$, $m_{\chi^0_2}$, $m_{\tilde{l}}$ and $m_{\chi^0_1}$ could each be potentially reconstructed [@recon; @drees; @fitter]. More generally speaking, the LHC is, in most models, likely to measure the mass of the lightest neutralino to roughly 10% accuracy, and may also be able to determine the masses of one or more of the other neutralinos, and any light sleptons [@lhc]. Charginos are more difficult to study at the LHC.
The heavy, neutral Higgs bosons of the MSSM ($A$, $H$), can also be potentially produced and studied at the LHC. In particular, in models with large $\tan \beta$, heavy Higgs bosons have enhanced couplings to down-type fermions, thus leading to potentially observable di-tau final states. If enough of these events are observed, the masses of the heavy Higgs bosons could be potentially reconstructed, and $\tan \beta$ measured [@ditau; @higgsmeasure].
Prospects for the discovery of supersymmetry at the Tevatron, although not nearly as strong as at the LHC, are also exciting. The most likely discovery channel at the Tevatron is probably through clean tri-lepton plus missing energy events originating from the production of a chargino and a heavy neutralino, followed by a decay of the form, $\chi^{\pm} \chi^0_2 \rightarrow \tilde{\nu} l^{\pm} l^{+} \tilde{l^{-}} \rightarrow \nu \chi^0_1 l^{\pm} l^+ l^- \chi^0_1$ [@fermilab]. Only models with rather light gauginos (neutralinos and charginos) and sleptons can be discovered in this way, however. For some of the recent results from supersymmetry searches at the Tevatron, see Ref. [@recenttevatron].
Measurements of particle masses and other properties at the LHC can provide an essential cross check for direct and indirect detection channels. In particular, neither direct nor indirect detection experiments provide information capable of identifying the overall cosmological abundance of a WIMP, but instead infer only combinations of density and interaction cross section, leaving open the possibility that an observation may be generated by a sub-dominant component of the cosmological dark matter with a somewhat larger elastic scattering or annihilation cross section. Collider measurements can help to clarify this situation.
In the left frame of Fig. \[LCC2\], we show the ability of the LHC to infer the thermal neutralino relic abundance from measurements of sparticle masses and other properties. The results shown are for a specific benchmark supersymmetric model (see Ref. [@Baltz:2006fm]), but are not atypical. In this case, the LHC can infer $\Omega_{\chi} h^2$ to lie roughly within 0.05 to 0.2 (assuming properties such as R-parity conservation), which, along with a detection in either a direct or indirect channel, would provide a strong confirmation that the observed neutralino does in fact constitute the bulk of the cosmological dark matter.
In the right frame of Fig. \[LCC2\], the same supersymmetric model is considered, but instead showing the LHC’s ability to determine the neutralino-nucleon elastic scattering cross section. This information would be very useful in combination with a direct detection signal. In particular, it would enable uncertainties in the local dark matter density to be reduced with confidence.
Also shown in each frame of Fig. \[LCC2\] are the results which could be obtained from a future 500 GeV or 1 TeV (center-of-mass) $e^+ e^-$ linear collider, such as the International Linear Collider (ILC) [@ilc]. Such an experiment would have considerable advantages over hadron colliders such as the Tevatron or the LHC. Although hadron colliders can reach very high center-of-mass energies, and thus play an essential role as discovery machines, lepton colliders are best suited for lower energy, precision measurements. In particular, at an electron-positron collider, the process $\ee \to
X\bar X$ can provide an exquisite diagnostic of the quantum numbers of the massive particle, $X$. As long as only the diagrams with annihilation through $\gamma$ and $Z$ are relevant, the angular distribution and threshold shape of the reaction are characteristic for each spin, and the normalization of the cross section directly determines the $SU(2)\times U(1)$ quantum numbers. These tests can be applied to any particles with electric or weak charge whose pair-production thresholds lie in the range of the collider. Such a measurement could be used to pin down the spin and quantum numbers of a given particle and bring us a long way toward the qualitative identification of the underlying model.
We cannot emphasize enough the importance of the complementary roles played by each of the LHC and ILC programs. Whereas the LHC can more easily reach high energies and offers very large cross sections for specific states of a model of new physics, the ILC will likely reach fewer states in the new particle spectrum, but will provide extremely incisive measurements of the properties of the particles that are available to it. Furthermore, the particles within the ILC reach are typically the ones on which the dark matter density depends most strongly. Although both the LHC and ILC can make precision measurements, the measurements at the ILC typically have a more direct interpretation in terms of particle masses and couplings.
SUMMARY
=======
In this review, we have attempted to summarize the diverse and complementary roles played by the various direct, indirect and collider searches for particle dark matter. As of 2008, there has not yet been a clear or conclusive detection of dark matter’s non-gravitational interactions. There is reason to be optitmistic, however, that such a detection will be made within the next few years, moving the field beyond the discovery phase and into the measurement phase of the quest to reveal dark matter’s nature and particle identity. As next generation direct detection experiments such as Super-CDMS, XENON-plus, LUX and others come online, most TeV-scale models containing a viable WIMP candidate will become within reach of these programs. Indirect detection experiments, including GLAST, VERITAS, HESS, MAGIC, PAMELA, AMS-02, IceCube and others are also rapidly advancing, and may see the first signals of dark matter annihilations. As the Large Hadron Collider begins its operation later this year, a new window into high-energy phenonoma will be opened. If dark matter is associated with physics of the electroweak scale, it is very likely to be within the discovery reach of this experiment.
The various experimental programs described in this review are each potentially capable of bringing very different measurments to the table. Although any one of these programs may be the first to discover particle dark matter, no single experiment or observation will answer all of our questions concerning this substance. Only by combining several of these detection methods together will it be possible to conclusively identify the dark matter of our universe.
Acknowledgments {#acknowledgments .unnumbered}
===============
DH is supported by the United States Department of Energy and NASA grant NAG5-10842. Fermilab is operated by the Fermi Research Alliance, LLC under Contract No. DE-AC02-07CH11359 with the United States Department of Energy.
[99]{}
A. Borriello and P. Salucci, Mon. Not. Roy. Astron. Soc. [**323**]{}, 285 (2001) \[arXiv:astro-ph/0001082\]. F. Zwicky, Helv. Phys. Acta 6 (1933) 110.
J. A. Tyson, G. P. Kochanski and I. P. Dell’Antonio, Astrophys. J. [**498**]{}, L107 (1998) \[arXiv:astro-ph/9801193\]. H. Dahle, arXiv:astro-ph/0701598. D. Clowe, M. Bradac, A. H. Gonzalez, M. Markevitch, S. W. Randall, C. Jones and D. Zaritsky, arXiv:astro-ph/0608407. D. N. Spergel [*et al.*]{} \[WMAP Collaboration\], Astrophys. J. Suppl. [**170**]{}, 377 (2007) \[arXiv:astro-ph/0603449\]. K. A. Olive, G. Steigman and T. P. Walker, Phys. Rept. [**333**]{}, 389 (2000) \[arXiv:astro-ph/9905320\]. M. Tegmark [*et al.*]{} \[SDSS Collaboration\], Astrophys. J. [**606**]{}, 702 (2004) \[arXiv:astro-ph/0310725\].
G. Bertone, D. Hooper and J. Silk, Phys. Rept. [**405**]{}, 279 (2005) \[arXiv:hep-ph/0404175\]. H. C. Cheng and I. Low, JHEP [**0309**]{}, 051 (2003) \[arXiv:hep-ph/0308199\]; J. Wudka, arXiv:hep-ph/0307339. G. Servant and T. M. P. Tait, Nucl. Phys. B [**650**]{}, 391 (2003) \[arXiv:hep-ph/0206071\]; H. C. Cheng, J. L. Feng and K. T. Matchev, Phys. Rev. Lett. [**89**]{}, 211301 (2002) \[arXiv:hep-ph/0207125\]; D. Hooper and S. Profumo, Phys. Rept., in press, arXiv:hep-ph/0701197. H. C. Cheng and I. Low, JHEP [**0309**]{}, 051 (2003) \[arXiv:hep-ph/0308199\]; H. C. Cheng and I. Low, JHEP [**0408**]{}, 061 (2004) \[arXiv:hep-ph/0405243\].
J. Angle [*et al.*]{} \[XENON Collaboration\], arXiv:0706.0039 \[astro-ph\]. D. S. Akerib [*et al.*]{} \[CDMS Collaboration\], arXiv:astro-ph/0509259. D. S. Akerib [*et al.*]{} \[CDMS Collaboration\], arXiv:astro-ph/0509269.
G. J. Alner [*et al.*]{}, Astropart. Phys. [**28**]{}, 287 (2007) \[arXiv:astro-ph/0701858\]; G. J. Alner [*et al.*]{} \[UK Dark Matter Collaboration\], Astropart. Phys. [**23**]{}, 444 (2005). V. Sanglard [*et al.*]{} \[The EDELWEISS Collaboration\], Phys. Rev. D [**71**]{}, 122002 (2005) \[arXiv:astro-ph/0503265\].
G. Angloher [*et al.*]{}, Astropart. Phys. [**23**]{}, 325 (2005) \[arXiv:astro-ph/0408006\].
P. Benetti [*et al.*]{}, arXiv:astro-ph/0701286; R. Brunetti [*et al.*]{}, New Astron. Rev. [**49**]{}, 265 (2005) \[arXiv:astro-ph/0405342\]. W. J. Bolte [*et al.*]{}, J. Phys. Conf. Ser. [**39**]{}, 126 (2006). E. I. Gates, G. Gyuk and M. S. Turner, Phys. Rev. D [**53**]{} (1996) 4138 \[arXiv:astro-ph/9508071\], E. Gates, G. Gyuk and M. S. Turner, arXiv:astro-ph/9704253. A. K. Drukier, K. Freese and D. N. Spergel, Phys. Rev. D [**33**]{}, 3495 (1986).
A. Helmi, S. D. M. White and V. Springel, Phys. Rev. D [**66**]{}, 063502 (2002) \[arXiv:astro-ph/0201289\].
G. Jungman, M. Kamionkowski and K. Griest, Phys. Rept. [**267**]{}, 195 (1996).
A. E. Nelson and D. B. Kaplan, Phys. Lett. B [**192**]{}, 193 (1987); D. B. Kaplan and A. Manohar, Nucl. Phys. B [**310**]{}, 527 (1988). A. Bottino, F. Donato, N. Fornengo and S. Scopel, “Size of the neutralino nucleon cross-section in the light of a new Astropart. Phys. [**18**]{}, 205 (2002) \[arXiv:hep-ph/0111229\]. J. R. Ellis, K. A. Olive, Y. Santoso and V. C. Spanos, Phys. Rev. D [**71**]{}, 095007 (2005) \[arXiv:hep-ph/0502001\].
C. Michael, C. McNeile and D. Hepburn \[UKQCD Collaboration\], Nucl. Phys. Proc. Suppl. [**106**]{}, 293 (2002) \[arXiv:hep-lat/0109028\]. M. Procura, T. R. Hemmert and W. Weise, Phys. Rev. D [**69**]{}, 034505 (2004) \[arXiv:hep-lat/0309020\].
G. B. Gelmini, P. Gondolo and E. Roulet, Nucl. Phys. B [**351**]{}, 623 (1991); M. Srednicki and R. Watkins, Phys. Lett. B [**225**]{}, 140 (1989); M. Drees and M. Nojiri, Phys. Rev. D [**48**]{}, 3483 (1993) \[arXiv:hep-ph/9307208\]; M. Drees and M. M. Nojiri, Phys. Rev. D [**47**]{}, 4226 (1993) \[arXiv:hep-ph/9210272\]; J. R. Ellis, A. Ferstl and K. A. Olive, Phys. Lett. B 481, (2000) 304, \[arXiv:hep-ph/0001005\].
A. M. Green, JCAP [**0708**]{}, 022 (2007) \[arXiv:hep-ph/0703217\]. J. D. Lewin and P. F. Smith, Astropart. Phys. [**6**]{}, 87 (1996).
T. Bringmann, L. Bergstrom and J. Edsjo, JHEP [**0801**]{}, 049 (2008) \[arXiv:0710.3169 \[hep-ph\]\]. L. Bergstrom, P. Ullio and J. H. Buckley, Astropart. Phys. [**9**]{}, 137 (1998) \[arXiv:astro-ph/9712318\]; L. Bergstrom, J. Edsjo and P. Ullio, Phys. Rev. Lett. [**87**]{}, 251301 (2001) \[arXiv:astro-ph/0105048\]; V. Berezinsky, A. Bottino and G. Mignola, Phys. Lett. B [**325**]{}, 136 (1994) \[arXiv:hep-ph/9402215\]; A. Cesarini, F. Fucito, A. Lionetto, A. Morselli and P. Ullio, Astropart. Phys. [**21**]{}, 267 (2004) \[arXiv:astro-ph/0305075\]; P. Ullio, L. Bergstrom, J. Edsjo and C. G. Lacey, Phys. Rev. D [**66**]{}, 123502 (2002) \[arXiv:astro-ph/0207125\].
J. F. Navarro, C. S. Frenk and S. D. M. White, Astrophys. J. [**462**]{}, 563 (1996) \[arXiv:astro-ph/9508025\]; J. F. Navarro, C. S. Frenk and S. D. M. White, Astrophys. J. [**490**]{}, 493 (1997).
F. Prada, A. Klypin, J. Flix, M. Martinez and E. Simonneau, arXiv:astro-ph/0401512; G. Bertone and D. Merritt, Mod. Phys. Lett. A [**20**]{}, 1021 (2005) \[arXiv:astro-ph/0504422\]; G. Bertone and D. Merritt, Phys. Rev. D [**72**]{}, 103502 (2005) \[arXiv:astro-ph/0501555\]. F. Aharonian [*et al.*]{} \[The HESS Collaboration\], arXiv:astro-ph/0408145; For example, see F. Ahronian, Talk at TeV Particle Astrophysics Workshop, Batavia, USA, July 2005.
J. Albert [*et al.*]{} \[MAGIC Collaboration\], Astrophys. J. [**638**]{}, L101 (2006) \[arXiv:astro-ph/0512469\]. K. Kosack [*et al.*]{} \[The VERITAS Collaboration\], Astrophys. J. [**608**]{}, L97 (2004) \[arXiv:astro-ph/0403422\]. K. Tsuchiya [*et al.*]{} \[CANGAROO-II Collaboration\], Astrophys. J. [**606**]{}, L115 (2004) \[arXiv:astro-ph/0403592\].
D. Hooper and J. March-Russell, Phys. Lett. B [**608**]{}, 17 (2005) \[arXiv:hep-ph/0412048\]; D. Hooper, I. de la Calle Perez, J. Silk, F. Ferrer and S. Sarkar, JCAP [**0409**]{}, 002 (2004) \[arXiv:astro-ph/0404205\]; S. Profumo, Phys. Rev. D [**72**]{}, 103521 (2005) \[arXiv:astro-ph/0508628\]; L. Bergstrom, T. Bringmann, M. Eriksson and M. Gustafsson, Phys. Rev. Lett. [**94**]{}, 131301 (2005) \[arXiv:astro-ph/0410359\].
F. Aharonian and A. Neronov, Astrophys. J. [**619**]{}, 306 (2005) \[arXiv:astro-ph/0408303\]; arXiv:astro-ph/0503354; AIP Conf. Proc. [**745**]{}, 409 (2005); A. Atoyan and C. D. Dermer, Astrophys. J. [**617**]{}, L123 (2004) \[arXiv:astro-ph/0410243\].
G. Zaharijas and D. Hooper, Phys. Rev. D [**73**]{}, 103501 (2006). \[arXiv:astro-ph/0603540\].
S. Dodelson, D. Hooper and P. D. Serpico, arXiv:0711.4621 \[astro-ph\].
N. W. Evans, F. Ferrer and S. Sarkar, Phys. Rev. D [**69**]{}, 123501 (2004) \[arXiv:astro-ph/0311145\]. L. Bergstrom and D. Hooper, Phys. Rev. D [**73**]{}, 063510 (2006) \[arXiv:hep-ph/0512317\]. L. E. Strigari, S. M. Koushiappas, J. S. Bullock, M. Kaplinghat, J. D. Simon, M. Geha and B. Willman, arXiv:0709.1510 \[astro-ph\].
L. Bergstrom, J. Edsjo and P. Ullio, Phys. Rev. Lett. [**87**]{}, 251301 (2001) \[arXiv:astro-ph/0105048\]; P. Ullio, L. Bergstrom, J. Edsjo and C. G. Lacey, Phys. Rev. D [**66**]{}, 123502 (2002) \[arXiv:astro-ph/0207125\]; D. Elsaesser and K. Mannheim, Astropart. Phys. [**22**]{}, 65 (2004) \[arXiv:astro-ph/0405347\]; D. Hooper and P. D. Serpico, JCAP [**0706**]{}, 013 (2007) \[arXiv:astro-ph/0702328\].
N. Gehrels and P. Michelson, Astropart. Phys. [**11**]{}, 277 (1999); S. Peirani, R. Mohayaee and J. A. de Freitas Pacheco, Phys. Rev. D [**70**]{}, 043503 (2004) \[arXiv:astro-ph/0401378\]; A. Cesarini, F. Fucito, A. Lionetto, A. Morselli and P. Ullio, Astropart. Phys. [**21**]{}, 267 (2004) \[arXiv:astro-ph/0305075\].
A. Morselli and P. Picozza, [*Prepared for 4th International Workshop on the Identification of Dark Matter (IDM 2002), York, England, 2-6 Sep 2002*]{}.
M. Sapinski \[AMS Collaboration\], Acta Phys. Polon. B [**37**]{}, 1991 (2006); C. Goy \[AMS Collaboration\], J. Phys. Conf. Ser. [**39**]{}, 185 (2006).
E. A. Baltz and J. Edsjo, Phys. Rev. D [**59**]{} (1999) 023511 \[arXiv:astro-ph/9808243\].
S. W. Barwick [*et al.*]{} \[HEAT Collaboration\], Astrophys. J. [**482**]{}, L191 (1997) \[arXiv:astro-ph/9703192\]; S. Coutu [*et al.*]{} \[HEAT-pbar Collaboration\], in Proceedings of 27th ICRC (2001).
Olzem, Jan \[AMS Collaboration\], Talk given at the 7th UCLA Symposium on Sources and Detection of Dark Matter and Dark Energy in the Universe, Marina del Ray, CA, Feb 22-24, 2006.
D. Hooper and G. D. Kribs, Phys. Rev. D [**70**]{}, 115004 (2004) \[arXiv:hep-ph/0406026\]. W. R. Webber, M. A. Lee and M. Gupta, Astrophys. J.[**390**]{} (1992) 96; I. V. Moskalenko, A. W. Strong, S. G. Mashnik and J. F. Ormes, Astrophys. J. [**586**]{}, 1050 (2003) \[arXiv:astro-ph/0210480\]; I. V. Moskalenko and A. W. Strong, Phys. Rev. D [**60**]{}, 063003 (1999) \[arXiv:astro-ph/9905283\].
D. Maurin, F. Donato, R. Taillet and P. Salati, Astrophys. J. [**555**]{}, 585 (2001) \[arXiv:astro-ph/0101231\]; D. Maurin, R. Taillet and F. Donato, Astron. Astrophys. [**394**]{}, 1039 (2002) \[arXiv:astro-ph/0206286\].
D. Hooper and J. Silk, Phys. Rev. D [**71**]{}, 083503 (2005) \[arXiv:hep-ph/0409104\]. I. V. Moskalenko and A. W. Strong, Astrophys. J. [**493**]{}, 694 (1998) \[arXiv:astro-ph/9710124\].
D. Hooper, J. E. Taylor and J. Silk, Phys. Rev. D [**69**]{}, 103509 (2004) \[arXiv:hep-ph/0312076\].
G. Bertone, E. Nezri, J. Orloff and J. Silk, Phys. Rev. D [**70**]{}, 063503 (2004) \[arXiv:astro-ph/0403322\].
L. Bergstrom, J. Edsjo and P. Gondolo, Phys. Rev. D [**55**]{}, 1765 (1997) \[arXiv:hep-ph/9607237\]; Phys. Rev. D [**58**]{}, 103519 (1998) \[arXiv:hep-ph/9806293\]; V. D. Barger, F. Halzen, D. Hooper and C. Kao, Phys. Rev. D [**65**]{}, 075022 (2002) \[arXiv:hep-ph/0105182\]; V. Barger, W. Y. Keung, G. Shaughnessy and A. Tregre, Phys. Rev. D [**76**]{}, 095008 (2007) \[arXiv:0708.1325 \[hep-ph\]\]; V. D. Barger, W. Y. Keung and G. Shaughnessy, arXiv:0709.3301 \[astro-ph\]. A. Gould, Astrophys. J. [**388**]{}, 338 (1991).
K. Griest and D. Seckel, Nucl. Phys. B [**283**]{}, 681 (1987) \[Erratum-ibid. B [**296**]{}, 1034 (1988)\]. A. Gould, Astrophys. J. [**321**]{}, 571 (1987). G. Jungman and M. Kamionkowski, Phys. Rev. D [**51**]{} (1995) 328 \[arXiv:hep-ph/9407351\]; For a more recent calculation see: M. Cirelli, N. Fornengo, T. Montaruli, I. Sokalski, A. Strumia and F. Vissani, arXiv:hep-ph/0506298.
F. Halzen and D. Hooper, Phys. Rev. D [**73**]{}, 123507 (2006) \[arXiv:hep-ph/0510048\].
T. DeYoung \[IceCube Collaboration\], Int. J. Mod. Phys. A [**20**]{}, 3160 (2005); J. Ahrens [*et al.*]{} \[The IceCube Collaboration\], Nucl. Phys. Proc. Suppl. [**118**]{}, 388 (2003) \[arXiv:astro-ph/0209556\]. P. Sapienza, Nucl. Phys. Proc. Suppl. [**145**]{}, 331 (2005).
S. Desai [*et al.*]{} \[Super-Kamiokande Collaboration\], Phys. Rev. D [**70**]{}, 083523 (2004) \[Erratum-ibid. D [**70**]{}, 109901 (2004)\] \[arXiv:hep-ex/0404025\].
M. Ackermann [*et al.*]{} \[AMANDA Collaboration\], arXiv:astro-ph/0508518. M. M. Boliev [*et al.*]{}, Proc. of the Intl. Workshop on Aspects of Dark Matter in Astrophysics and Particle Physics, Heidelberg, Germany, 1996. Edited by H. V. Klapdor-Kleingrothaus, Y. Ramachers. Singapore, World Scientific, 1997.
M. Ambrosio [*et al.*]{} \[MACRO Collaboration\], Phys. Rev. D [**60**]{}, 082002 (1999) \[arXiv:hep-ex/9812020\].
J. Hossl \[ANTARES Collaboration\], Proc. of the Identification of Dark Matter (IDM), Edinburgh, Scotland, 2004; J. Brunner \[ANTARES Collaboration\], Nucl. Phys. Proc. Suppl. [**145**]{}, 323 (2005).
G. J. Alner [*et al.*]{} \[UK Dark Matter Collaboration\], Phys. Lett. B [**616**]{}, 17 (2005) \[arXiv:hep-ex/0504031\].
M. Barnabe-Heider [*et al.*]{} \[PICASSO Collaboration\], arXiv:hep-ex/0502028.
E. A. Baltz and L. Wai, Phys. Rev. D [**70**]{}, 023512 (2004) \[arXiv:astro-ph/0403528\]; S. Colafrancesco, S. Profumo and P. Ullio, Astron. Astrophys. [**455**]{}, 21 (2006) \[arXiv:astro-ph/0507575\]; L. Bergstrom, M. Fairbairn and L. Pieri, Phys. Rev. D [**74**]{}, 123515 (2006) \[arXiv:astro-ph/0607327\].
D. P. Finkbeiner, arXiv:astro-ph/0409027; D. Hooper, D. P. Finkbeiner and G. Dobler, Phys. Rev. D [**76**]{}, 083012 (2007), arXiv:0705.3655 \[astro-ph\].
S. Dawson, E. Eichten and C. Quigg, Phys. Rev. D [**31**]{}, 1581 (1985).
D. R. Tovey, Eur. Phys. J. direct C [**4**]{}, N4 (2002).
H. Bachacou, I. Hinchliffe and F. E. Paige, Phys. Rev. D [**62**]{}, 015009 (2000) \[arXiv:hep-ph/9907518\].
M. Drees, Y. G. Kim, M. M. Nojiri, D. Toya, K. Hasuko and T. Kobayashi, Phys. Rev. D [**63**]{}, 035008 (2001) \[arXiv:hep-ph/0007202\]. R. Lafaye, T. Plehn and D. Zerwas, arXiv:hep-ph/0404282; P. Bechtle, K. Desch and P. Wienemann, Comput. Phys. Commun. [**174**]{}, 47 (2006) \[arXiv:hep-ph/0412012\].
H. Bachacou, I. Hinchliffe and F. E. Paige, Phys. Rev. D [**62**]{}, 015009 (2000) \[arXiv:hep-ph/9907518\]; B. C. Allanach, C. G. Lester, M. A. Parker and B. R. Webber, JHEP [**0009**]{}, 004 (2000) \[arXiv:hep-ph/0007009\]; H. Baer, C. h. Chen, F. Paige and X. Tata, Phys. Rev. D [**52**]{}, 2746 (1995) \[arXiv:hep-ph/9503271\]; Phys. Rev. D [**53**]{}, 6241 (1996) \[arXiv:hep-ph/9512383\]; S. Abdullin and F. Charles, Nucl. Phys. B [**547**]{}, 60 (1999) \[arXiv:hep-ph/9811402\].
S. Abdullin [*et al.*]{}, Eur. Phys. J. C [**39S2**]{}, 41 (2005); A. Datta, A. Djouadi, M. Guchait and F. Moortgat, Nucl. Phys. B [**681**]{}, 31 (2004) \[arXiv:hep-ph/0303095\]. R. Kinnunen, S. Lehti, F. Moortgat, A. Nikitenko and M. Spira, Eur. Phys. J. C [**40N5**]{}, 23 (2005) \[arXiv:hep-ph/0503075\].
S. Abel [*et al.*]{} \[SUGRA Working Group Collaboration\], arXiv:hep-ph/0003154.
D. Bortoletto \[CDF and D0 Collaborations\], PoS [**HEP2005**]{}, 347 (2006); A. Canepa \[CDF Collaboration\], arXiv:hep-ex/0603032; A. Anastassov \[CDF and D0 Collaborations\], PoS [**HEP2005**]{}, 326 (2006); V. Abazov \[D0 Collaboration\], arXiv:hep-ex/0604029.
E. A. Baltz, M. Battaglia, M. E. Peskin and T. Wizansky, Phys. Rev. D [**74**]{}, 103521 (2006) \[arXiv:hep-ph/0602187\].
G. Weiglein [*et al.*]{} \[LHC/LC Study Group\], arXiv:hep-ph/0410364.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Standard sirens are the gravitational wave (GW) analog of the astronomical standard candles, and can provide powerful information about the dynamics of the Universe. In this work, we simulate a catalog with 1000 standard siren events from binary neutron star mergers, within the sensitivity predicted for the third generation of the ground GW detector called Einstein telescope. After correctly modifying the propagation of GWs as input to generate the catalog, we apply our mock data set on scalar-tensor theories where the speed of GW propagation is equal to the speed of light. As a first application, we find new observational bounds on the running of the Planck mass, when considering appropriate values within the stability condition of the theory, and we discuss some consequences on the amplitude of the running of the Planck mass. In the second part, we combine our simulated standard sirens catalog with other geometric cosmological tests (Supernovae Ia and cosmic chronometers measurements) to constrain the Hu-Sawicki $f(R)$ gravity model. We thus find new and non-null deviations from the standard $\Lambda$CDM model, showing that in the future the $f(R)$ gravity can be tested up to 95% confidence level. The results obtained here show that the statistical accuracy achievable by future ground based GW observations, mainly with the ET detector (and planed detectors with a similar sensitivity), can provide strong observational bounds on modified gravity theories.'
author:
- 'Rocco D’Agostino'
- 'Rafael C. Nunes'
title: 'Probing observational bounds on scalar-tensor theories from standard sirens'
---
Introduction
============
After 20 years of research, the nature of the physical mechanism responsible for accelerating the Universe at late times is still an open question, and a large variety of cosmological models have been and are continually proposed in the literature to explain such observations (see [@DE_review; @MG_review01; @MG_review03] for review). This is essentially due to the difficulty of discriminating among different scenarios that respond to the observations in the same way, leading to a theoretical degeneracy.
The observation of new astrophysical sources, through a direct manifestation of gravitational effects, can provide rich physical information about the nature of gravity, which should play a key role to probe new (or rule out) additional gravitational degree(s) of freedom, or exotic forms of energy such as dark energy. The gravitational waves (GWs) issued by binary systems, such as binary black hole (BBH) and/or binary neutron star (BNS) detected by LIGO/VIRGO, certainly open a new window to investigate fundamental physics in this direction. At present, catalogues of GWs from 10 BBH mergers and 1 BNS merger are available [@LIGO01]. The latter, the GW170817 event [@GW170817], observed at $z \simeq 0.009$, has imposed strong constraints on modified gravity/dark energy models [@GW_MG01; @GW_MG02; @GW_MG03; @GW_MG04; @GW_MG041; @GW_MG05]. Also, GW170817 was the first standard siren (the gravitational-wave analog of an astronomical standard candle) event to be catalogued, once its electromagnetic counterpart (GRB170817) was measured. These observations were also used to measure $H_0$ at the 12% accuracy, assuming a fiducial $\Lambda$CDM cosmology [@SS01]. We refer the reader to [@H0_GW1; @H0_GW2; @H0_GW3] for proposals to use standard siren to measure $H_0$ with more accuracy.
Given the central importance of GW astronomy, beyond the present performance of the LIGO and Virgo interferometers, plans for construction of several GW observatory interferometers (on earth and in space) are currently in preparation, such like LIGO Voyager [@CE], Cosmic Explore [@CE], Einstein Telescope (ET) [@ET_design01; @ET_design02], LISA [@LISA], DECIGO [@DECIGO], TianQin [@TianQin], among others, to observe GWs in the most diverse frequencies bands and different types of GW sources. In this paper, we are particularly interested to use the sensitivity predicted for the ET [@ET_design01; @ET_design02], which is a third-generation ground-based detector and it is envisaged to be several times more sensitive in amplitude than the advanced ground-based detectors in operation, covering the frequency band range from 1 Hz to $10^4$ Hz. Also, the ET is expected to have signal-to-noise ratio (SNR) for BBH and BNS mergers several times larger than the current measures, as well as to observe hundreds, thousands of these events throughout the whole operational time. Several works have been done using the ET sensitivity to simulate GWs standard siren in order to investigate diverse aspects in cosmology [@ET01; @ET02; @ET03; @ET04; @ET05; @ET06; @ET07; @ET08; @ET09; @ET10; @ET11; @ET12; @ET13; @ET14; @ET15; @ET16; @ET17; @ET18; @ET19; @ET20; @ET21; @ET22; @ET23].
In this work, we generate a simulated catalog with 1000 standard siren events from BNS mergers, from the ET power spectral density noise, in order to evaluate forecasting observational constraints on scalar-tensor theories where the speed of GW propagation is equal to speed of light. First, assuming a well-known parametric model for the running of the Planck mass, and assuming appropriate stability conditions on the theory, we find new observational bounds on the amplitude of the running of the Planck mass and we discuss its possible implications. In the second part, we apply our simulated standard siren data on $f(R)$ gravity given by the Hu-Sawicki model in order to find new observational limits on such model. In both analysis, we find that the parameters that characterize deviations from General Relativity (GR) may be non-null, within some statistical borders.
The manuscript is organized as follows. In Section \[sec-model\], we set our theoretical framework to show how the GW propagation is modified from scalar-tensor theories. In Section \[Methodology\], we describe our methodology to generate standard siren mock catalogs. In Section \[alphaM0\] and \[f(R)\], we present our main results. Finally, in Section \[Conclusions\], we outline our final considerations and future perspectives.
Throughout the text, we use units such that $c=\hbar=1$, and $M_P=1/\sqrt{8\pi G}$ is the Planck mass. Moreover, we adopt the flat Friedmann-Lemaître-Robertson-Walker (FLRW) metric, $ds^2=-dt^2+a(t)^2\delta_{ij}dx^idx^j$, where $a$ is the scale factor, normalized to unity today. As usual notation, we denote by a subscript ‘0’ physical quantities evaluated at the present time, and by the prime and dot symbols the derivatives with respect to the conformal time ($\tau$) and cosmic time ($t$), respectively, related by $dt=a d\tau$. Lastly, we express the Hubble constant $(H_0)$ results in units of km/s/Mpc.
Modified Gravitational Wave Propagation in Scalar-Tensor Gravity {#sec-model}
================================================================
The Horndeski theories of gravity [@Horndeski; @Deffayet] (see [@Horndeski_revisao01; @Horndeski_revisao02] for a review) are the most general Lorentz invariant scalar-tensor theories with second-order equations of motion and where all matter is universally coupled to gravity. They include, as a sub set, the archetypal modifications of gravity such as metric and Palatini $f(R)$ gravity, Brans-Dicke theories, galileons, among others. The Horndeski action reads $$\label{acao_geral}
S = \int d^4 x \sqrt{-g} \left[ \sum_{i=2}^{5} M_P^2 \mathcal{L}_i + \mathcal{L}_m \right],$$ where $g$ is the determinant of the metric tensor, and $$\begin{aligned}
\mathcal{L}_2 = &\ G_2(\phi, X), \\
\mathcal{L}_3 = &-G_3(\phi, X) \Box \phi, \\
\mathcal{L}_4 =& -G_4(\phi, X)R + G_{4,X} [( \Box \phi)^2 - \phi_{;\mu \nu} \phi^{;\mu \nu}], \\
\mathcal{L}_5 = &- G_5(\phi, X)G_{\mu \nu}\phi^{;\mu \nu} -
\dfrac{1}{6}G_{5,X}[(\Box \phi)^3 \\
&+ 2 \phi_{;\mu \nu} \phi^{;\mu \sigma} \phi^{;\nu}_{;\sigma}
- 3 \phi_{;\mu \nu} \phi^{;\mu \nu} \Box \phi].\end{aligned}$$ Here, $G_i$ ($i$ runs over 2, 3, 4, 5) are functions of a scalar field $\phi$ and the kinetic term $X \equiv -1/2 \nabla^\nu \phi \nabla_\nu \phi $, and $G_{i,X} \equiv \partial G_i/\partial X$. For $G_2 = \Lambda$, $G_4 = M^2_P/2$ and $G_3 = G_5 = 0$, we recover GR with a cosmological constant. For a general discussion on the model varieties for different $G_i$ choices after GW170817, see [@Horndeski_revisao02].
Recently, the GW170817 event together with the electromagnetic counterpart showed that the speed of GW, $c_T$, is very close to the speed of light for $z < 0.01$, i.e. $ |c_T/c - 1| < 10^{-15}$ [@GW170817]. In the context of Horndeski gravity, in order to explain these constraints, the only option is to consider $G_{4,X} \approx 0$ and $G_5 \approx constant$ in the action above. It is important to note that this restriction applies only to the local Universe ($\lesssim$ 40 Mpc). Thus, in principle, nothing prevents from considering the presence of these terms at redshifts larger than $z = 0.01$. In fact, only future measurements at high $z$ can confirm whether $c_T = c$. Here, we assume $c_T = c$, without loss of generality in the analysis we are going to develop. Under this condition, the GW propagation obeys the equation of motion [@Saltas04] $$\label{h}
h''_{ij} + (2 + \alpha_M) \mathcal{H} h'_{ij} + k^2 h_{ij} = 0,$$ where $h_{ij}$ is the metric tensor perturbation and $\mathcal{H}\equiv a'/a$ is the Hubble rate in conformal time. Moreover, $\alpha_M$ is the running of the Planck mass, which enters as a friction term responsible for modifying the amplitude of GWs acting as a damping term: $$\label{alphaM}
\alpha_M = \dfrac{1}{H M^2_{*}} \dfrac{dM^2_{*}}{dt},$$ where $M_{*}$ is the effective Planck mass: $$M^2_{*} = 2(G_4 - 2XG_{4X} + XG_{5 \phi} - \dot{\phi} H X G_{5X}),$$ and $H\equiv \dot{a}/a$ is the Hubble parameter. Following the methodology presented in [@Atsushi01] (see also [@Atsushi02; @Atsushi03]), we can write a generalized GW amplitude propagation for scalar-tensor theories as $$\label{hDT}
h = e^{-\mathcal{D}} h_{GR},$$ where $$\label{DD}
\mathcal{D} = \dfrac{1}{2} \int^{\tau} \alpha_M \mathcal{H} d\tau'.$$ Note that due the condition $c_T = c$, that is, $G_{4,X} \approx 0$ and $G_5 \approx constant$, we do not have phase corrections in Eq. (\[hDT\]). As the GW amplitude is inversely proportional to the distance, one can interpret the amplitude modification in Eq. (\[hDT\]) as a correction to the luminosity distance, defining an effective luminosity distance, or equivalently, an effective amplitude correction as [@Atsushi02; @MG_amplitude01; @MG_amplitude02]: $$\label{dL_Gw}
d_L^{GW}(z) = d_L^{EM}(z) \exp \left[\dfrac{1}{2} \int_0^{z} \dfrac{dz'}{1 + z'} \alpha_M (z') \right],$$ where $d_L^{EM}$ is the standard electromagnetic luminosity distance as a function of the redshift[^1]: $$\label{dL_em}
d_L^{EM}(z) = (1 + z) \int_0^z \dfrac{dz'}{H(z')}.$$ This generalization has been recently investigated in some contexts of modified gravity (see, e.g., [@Atsushi01; @Atsushi02; @Atsushi03; @MG_amplitude01; @MG_amplitude02; @MG_dL01; @MG_dL02; @MG_dL03; @MG_dL04; @MG_dL05; @MG_dL06; @MG_dL07; @MG_dL08]).
{width="3.2in"} {width="3.2in"}
It is usual to choose phenomenologically motivated functional forms for the functions $\alpha_i$ (see, e.g., [@Bellini; @alphai_01; @alphai_02; @alphai_03]). Typically, their evolution are tied to the scale factor $a(t)$ or to the dark energy density $\Omega_{de}(a)$ raised to some power $n$. On the other hand, an important point within Horndeski gravity are the stability conditions of the theory. Appropriate values of the free parameters functions must be considered in order to have a stable theory throughout the evolution of the Universe (see [@alphai_02] and reference therein). Following [@alphai_02], we adopt the parametrization $\alpha_M =\alpha_{M0} a^n$, so that the stability conditions can be summarized as follows:
1. $n > \dfrac{5}{2}$: stable for $\alpha_{M0} < 0$;
2. $0 < n < 1 + \dfrac{3\Omega_{m0}}{2}$: stable for $\alpha_{M0} > 0$.
Here, $\Omega_{m0}$ is the present dimensionless matter density. Under these considerations, we can note from Eq. (\[dL\_Gw\]) that the changes in the GW amplitude propagation will be sensitive to the sign of $\alpha_{M0}$. Possible corrections with $\alpha_{M0} > 0$ or $\alpha_{M0}< 0$ will induce $d_L^{GW} > d_L^{EM}$ and $d_L^{GW} < d_L^{EM}$, respectively. We quantify these effects in Figure \[fig:dL\]. We note that variations on $\alpha_{M0} > 0 \, (< 0)$ can produce changes up to 10% (5%), respectively, on the effective GW amplitude, for reasonable values of the running of the Planck mass today.
Methodology and GW standard sirens data set {#Methodology}
===========================================
In order to move on, we need to define the GW signal $h_{GR}$. In modeling the gravitational waves form, given a GW strain signal $h(t) = A(t) \cos [\Phi(t)]$, we can obtain its Fourier transform $\tilde{h}(f)$ using the stationary phase approximation for the orbital phase of inspiraling binary system. For a coalescing binary system with component masses $m_1$ and $m_2$, we have $$\label{waveform}
\tilde{h}(f) = Q \mathcal{A} f^{-7/6} e^{i\Phi(f)},$$ where $\mathcal{A}$ is the GW inspiral amplitude computed perturbatively within the so-called post-Newtonian (PN) formalism up until 3 PN corrections, $$\label{A}
\mathcal{A} = \sqrt{\dfrac{5}{96}} \dfrac{\mathcal{M}^{5/6}_c}{\pi^{2/3} d_L^{GW}} \left( \sum_{i=0}^6 A_i(\pi f)^{i/3} \right),$$ where $d_L^{GW}$ is the modified luminosity distance as in Eq. (\[dL\_Gw\]), and the coefficients $A_i$ are given in Appendix A. The function $Q$ is expressed by $$\label{Q}
Q^2 = F^2_{+}(1+cos^2(\iota))^2 + 2F^2_{\times}cos^2(\iota),$$ where $\iota$ is the inclination angle of the binary orbital angular momentum with respect to the line of sight, and $F^2_{+}$, $F^2_{\times}$ are the two antenna pattern functions. In Eq. (\[waveform\]), the function $\Phi(f)$ is the inspiral phase of the binary system: $$\label{phi}
\Phi(f) = 2 \pi f t_c - \phi_c - \dfrac{\pi}{4} + \dfrac{3}{128 \eta v^5} \left[ 1 + \sum_{i=2}^7 \alpha_i v^i \right],$$ where the coefficients $\alpha_i$ are the corrections up to the 3.5 PN corrections. In Appendix A, we also provide the expressions for these coefficients. In the above equation, we have defined $v \equiv (\pi M f)^{1/3}$, $M \equiv m_1 + m_2$, $\eta \equiv m_1 m_2/(m_1 + m_2)^2$, and $\mathcal{M}_c \equiv (1+z) M \eta^{3/5}$ to be the inspiral reduced frequency, total mass, symmetric mass ratio, and the redshifted chirp mass, respectively. The quantities $t_c$ and $\phi_c$ are the time and phase of coalescence, respectively.\
After having defined the modified GW signal for compact binaries, in what follows we summarize the already know methodology used to estimate $d_L(z)$ measures from GW standard sirens. We refer to [@Schutz; @Holz] for pioneer works in this regard.
For a high enough signal-to-noise ratio (SNR) and a given waveform model, $h(f, \theta_i )$, with free parameters $\theta_i$, we can use the Fisher matrix analysis to provide upper bounds for the free parameters of the models by means of the Cramer-Rao bound [@Fisher01; @Fisher02]. We refer the reader to [@Fisher03; @Fisher04; @Fisher05; @Fisher06; @Fisher07; @Fisher08] for a discussion on the Fisher analysis to estimate parameters in binary systems for a given GW signal. Once the waveform model is defined, the root-mean-squared error on any parameter is determined by $$\begin{aligned}
\label{}
\Delta \theta^i = \sqrt{\Sigma^{ii}},\end{aligned}$$ where $\Sigma^{ij}$ is the covariance matrix, i.e, the inverse of the Fisher matrix, $\Sigma^{ij} = \Gamma_{ij}^{-1}$. The Fisher matrix is given by $$\label{}
\Gamma_{ij} = \left( \dfrac{\partial \tilde{h}}{\partial \theta^i} \mid \dfrac{\partial \tilde{h}}{\partial \theta^j} \right).$$ The inner product between two waveform models is defined as $$\label{}
(\tilde{h}_1 \mid \tilde{h}_2) \equiv 2 \int_{f_{low}}^{f_{upper}} \dfrac{\tilde{h}_1 \tilde{h}_2^{*} + \tilde{h}_1^{*} \tilde{h}_2}{S_n(f)} df,$$ where the ‘star’ stands for complex conjugation, and $S_n(f)$ is the detector spectral noise density. With this definition of the inner product, the SNR is defined as $${\rm SNR}^2 \equiv 4 Re \int_{f_{low}}^{f_{upper}} \, \dfrac{\vert h(f)\vert ^2}{S_n} df .
\label{eq:snr}$$
In what follows, we consider the Einstein telescope (ET) detector power spectral density noise. The ET is a third-generation ground-based detector of GWs and it is envisaged to be ten times more sensitive in amplitude than the advanced ground-based detectors in operation nowadays, covering the frequency range $1-10^4$ Hz. Unlike the current detectors, from the ET conceptual design study, the expected rates of BNS detections per year are of the order of $10^3-10^7$ [@ET02]. However, we can expect only a small fraction ($\sim 10^{-3}$) of them accompanied with the observation of a short $\gamma$-ray burst. If we assume that the detection rate is in the middle range around $\mathcal{O}(10^5)$, we can expect to see $\mathcal{O}(10^2)$ events with short $\gamma$-ray bursts per year.
Thus, let us consider in our simulations a mock GW standard sirens data set composed by 1000 BNS merger events. Assuming that the errors on $d_L$ are uncorrelated with errors on the remaining GW parameters, we have $$\sigma_{d_L}^2 = \left( \dfrac{\partial \tilde{h}(f)}{\partial d_L}, \dfrac{\partial \tilde{h}(f)}{\partial d_L} \right)^{-1}.$$ Since $\tilde{h}(f) \propto {(d_L^{GW}})^{-1}$, then $\sigma_{d_L} = d_L/ {\rm SNR}$. However, when we estimate the practical uncertainty of the measurements of $d_L$, we should take the orbital inclination into account. The maximal effect of the inclination on the SNR is a factor of 2 (between $\iota =0^{\circ}$ and $\iota = 90^{\circ}$). Therefore, we add this factor to the instrumental error for a conservative estimation. Thus, the estimate of the instrumental error is given by $\sigma_{d_L} = 2 d_L/ {\rm SNR}$. On the other hand, GWs are lensed in the same way as the electromagnetic waves, resulting into a weak lensing effect error, which we model as $\sigma^{\rm lens}_{d_L}= 0.05\ z\ d_L(z)$ [@ET02; @ET_lens]. In our study, we do not consider possible errors induced from the peculiar velocity due to the clustering of galaxies. Since we are interested in simulating events at high $z$ mainly, we can neglect such contributions, which are significant only for $z \ll 1$. In fact, at high $z$, the dominant source of uncertainty is the one due to weak lensing. Therefore, the total uncertainty $\sigma_{d_L}$ on the luminosity distance measurements associated to each event is obtained by combining the instrumental and weak lensing uncertainties as $$\begin{aligned}
\sigma_{d_L} &= \sqrt{ \left ( \sigma^{\rm ins}_{d_L} \right )^2 + \left ( \sigma^{\rm lens}_{d_L} \right )^2} \nonumber \\
&= \sqrt{ \left ( \dfrac{2d_L(z)}{\rm SNR} \right )^2 + (0.05 z d_L(z))^2}\,.
\label{eq:sigmatot}\end{aligned}$$ The redshift distribution of the BNS sources is taken to be of the form $$P(z) \propto \dfrac{4 \pi d^2_C(z) r(z)}{H(z)(1+z)},$$ where $d_C(z)$ is the comoving distance, and $r(z)$ describes the time-evolution of the burst rate: $$r(z)=\left\{
\begin{aligned}
& 1 + 2z, \hspace{0.2cm} z \leq 1 \\
& (15 - 3z)/4, \hspace{0.2cm} 1 < z < 5 \\
& 0,\hspace{0.2cm} z \geq 5
\end{aligned}
\right.$$ The distribution of the neutron star masses is chosen to be randomly sampled from uniform distributions within $[1 - 2]~{\rm M}_{\odot}$, also under the condition $m_1 \gtrsim m_2$ and $\eta < 0.25$. In this case, in the mock data generation we take $\chi_1 = \chi_2 = 0$, where $\chi_1, \chi_2$ are the associated spin magnitudes on each mass component. Then, we simulate BNS mergers up to $z =2$, which represents the maximum distance at which these events can be observed from the power spectral density noise from the ET [@ET02]. Also, we checked that, beyond $z = 2$, the SNR presents low values, which also can limit the use of the Fisher information for mock data. Now, in order to realistically generate a mock catalog using modified gravity, we shall consider non-zero values for the parameters $\alpha_{M0}$ and $n$, which are compatible with the current cosmological observation as well as with the stability criteria of the theory. Lastly, when generating our mock GW standard sirens data set, we only consider BNS mergers with ${\rm SNR} > 8$.
In order to estimate the observational constraints on the free parameters of the models, we apply Markov Chain Monte Carlo (MCMC) method through the Metropolis-Hastings algorithm [@Metropolis-Hastings], where the likelihood function for the GW standard sirens mock data set is built in the form $$\mathcal{L}_\text{GW} \propto \exp \left[ -\dfrac{1}{2} \sum_{i=1}^{1000} \left( \dfrac{d_L^{obs}(z_i) - d_L^{th}(z_i)}{\sigma_{d_L,i}}\right) \right].$$ Here, $d_L^{obs}(z_i)$ are the 1000 simulated BNS merger events with their associated uncertainties $\sigma_{d_L,i}$, while $d_L^{th}(z_i)$ is the theoretical prediction on each $i$th event.
Constraints on the running of the Planck mass {#alphaM0}
=============================================
Stability Conditions $H_0$ $\Omega_{m0}$ $\alpha_{M0}$
---------------------- -------------------------------------------- ------------------------------------------ -------------------------------------------- --
$\alpha_{M0}<0$ $67.466^{+0.036(0.143)}_{-0.067(0.179)} $ $0.328^{+0.015(0.028)}_{-0.014(0.028)} $ $-0.100^{+0.051(0.092)}_{-0.043(0.085)} $
$\alpha_{M0}>0$ $67.390_{-0.050(0.095)}^{+0.047(0.098)}$ $0.297^{+0.029(0.083)}_{-0.044(0.072)} $ $0.199^{+0.069(0.178)}_{-0.097(0.167)} $
{width="3.2in"} {width="3.2in"}
In this section, we present and discuss our results regarding the future observational constraints that GW standard sirens can impose on a possible time variation of the Planck mass within of the ET sensitivity. The running of the Planck mass is an important physical quantity, which essentially is present in any and all modified gravity models. To generate a simulated $d_L(z_i)$ catalog using modified gravity, we assume realistic values for the pair ($\alpha_{M0}$, $n$), on each triplet ($z_i, d_L(z_i), \sigma_{d_L(z_i)}$) evaluated at each point $i$, as follows.
We first note that the parameter $n$ is statistically degenerate. This fact is already well known and expected to happen. In the literature, it is usual to assume $n = 1$, but here we follow the stability conditions discussed in Section \[sec-model\] and weakly generate random values for $n$, within the range of stability of the theory: $i$) for the case $\alpha_{M0} > 0$, we randomly sampled from uniform distributions: $\alpha_{M0} \in [0, 0.5]$ and $n \in [0, 1.40]$; $ii$) for the case $\alpha_{M0} < 0$, we randomly sampled $\alpha_{M0} \in [-0.5, 0]$ and $n \in [2.5, 3.5]$. We found that different prior ranges on $n$ change the simulated catalogs very weakly. Only very different prior ranges on $\alpha_{M0}$ can significantly change the pair ($d_L(z_i), \sigma_{d_L(z_i)}$). The range assumed on $\alpha_{M0}$ is fully compatible with current constraints [@Horndeski_constraints_01; @Horndeski_constraints_02; @Horndeski_constraints_03; @Horndeski_constraints_04; @Horndeski_constraints_05; @Horndeski_constraints_06; @Horndeski_constraints_07; @Horndeski_constraints_08; @Horndeski_constraints_09; @Horndeski_constraints_10; @Horndeski_constraints_11]. We used as input values $H_0 = 67.4$ km/s/Mpc and $\Omega_{m0} = 0.31$ for the Hubble constant and matter density parameter, respectively, in agreement with the most recent Planck CMB data [@Planck2018]. Hence, these values are reasonable for our purpose to generate GW standard sirens mock data.
In the realization of the MCMC analysis, the sampling has been done assuming the following uniform priors for the cosmological parameters: $H_0 \in [55, 90]$, $\Omega_{m0} \in [0, 1]$, and $\alpha_{M0} \in [-1, 0]$, $\alpha_{M0} \in [0, 1]$ for each case. Due to the large statistical degeneracy on $n$, as commented above, we fixed $n= 3$ and $n=1$ for the cases of $\alpha_{M0} > 0$ and $\alpha_{M0}< 0$, respectively. Table \[tab:results\] summarizes the constraints at the 68% and 95% confidence levels (CL). In Figure \[fig:contours\_alphaM0\], we show the parametric space and the one-dimensional marginalized distribution for the parameters $\Omega_{m0}$, $H_0$ and $\alpha_{M0}$, in both $\alpha_{M0} > 0$ and $\alpha_{M0}< 0$ cases. We note from both analyses that the parameter $\alpha_{M0}$ is non-null at the 68% CL. Assuming the stability condition where $\alpha_{M0}$ is negative, we find the new lower limit $\alpha_{M0} > -0.2$ at the 95% CL. On the other hand, under the condition that the running of the Planck mass is positive defined, we find that $\alpha_{M0}$ can be non-null up to 95% CL, more specifically $0.03 \lesssim \alpha_{M0} \lesssim 0.38$. For a qualitative comparison, we show also the constraints from the $\Lambda$CDM model in both cases. It is important to note that the $\Lambda$CDM model is a particular case of our simulated data, where both mock catalogs are mainly controlled by possible corrections on the amplitude of the running of the Planck mass as explained above. Therefore, extended estimates on the pair ($d_L(z_i), \sigma_{d_L(z_i)}$) beyond the $\Lambda$CDM prediction will be dependent on the $\alpha_{M0}$ correction, where we have considered two different scenarios $\alpha_{M0} > 0$ and $\alpha_{M0} < 0$. This should induce a minimal bias when analyzing the parameters within the $\Lambda$CDM scenario due to the $\alpha_{M0}$ prior range in the mock data generation. In the case $\alpha_{M0} < 0$, our minimal baseline, i.e. $\Omega_{m0}$ and $H_0$, is completely compatible for the two scenarios. On the other hand, when analyzing $\alpha_{M0} > 0$, we can note a minimal bias manifested on $\Omega_{m0}$, resulting in a minimum tension at 1$\sigma$ CL on this parameter, but still making the model compatible beyond that statistical significance.
It is interesting to compare our results with others already obtained in the literature. For instance, in [@Horndeski_constraints_11], using measurements of the growth rate of structures from DESI survey, it is observed that the amplitude of the running of the Planck mass (quantity physically analogous to our $\alpha_{M0} > 0$) can be detected up to 99% CL. In [@Horndeski_constraints_08], a 95% CL upper limit of 0.015 is found from CMB data. In [@MG_dL01], analyzing the standard siren GW17081 event, the authors found the amplitude of the running of the Planck mass to be $\in$ \[-80, 28\] at the 95% CL. In [@Atsushi02], the amplitude damping $\alpha_{M0} < 0$ appear to be preferentially at low $z$ from GWs observations. Other analyzes can be found in [@Horndeski_constraints_01; @Horndeski_constraints_02; @Horndeski_constraints_03; @Horndeski_constraints_04; @Horndeski_constraints_05; @Horndeski_constraints_06; @Horndeski_constraints_07; @Horndeski_constraints_08; @Horndeski_constraints_09; @Horndeski_constraints_10; @Horndeski_constraints_11]. We note that the new borders on the amplitude of the running of the Planck mass derived in this work may have also an impact on the modified propagation primordial gravitational waves spectrum [@Horndeski_constraints_12].\
Now, we shall briefly discuss the consequences of our results. Based on the arguments developed in Section \[sec-model\], we can write the running of the Planck mass as $$\label{running planck mass}
\alpha_M = \dfrac{\dot{G_4}}{H G_4}.$$ One of the surviving classes of models under the condition $c_T = c$ are the non-minimal theories in which the scalar field $\phi$ is coupled with the curvature scalar $R$ in the form $G_4(\phi) R$. This class includes the metric $f(R)$ gravity and the Brans-Dicke theory [@Brans61]. The original Brans-Dicke theory, for instance, is obtained by setting $G_4 = \phi$. By substituting this in Eq. (\[running planck mass\]), it is possible to obtain $\phi$ as a function of the cosmic time. In Figure \[fig:phi\], we show a reconstruction for the evolution of the field $\phi/\phi_0$ in Planck mass units, where $\phi_0$ is $\phi(z = 0)$. Also, it is important to remember that in such a theory the gravitational constant is not presumed to be constant, but $G_N(\phi) \varpropto 1/\phi$. This fact is physically encoded in $\alpha_M(a)$, which measures the gravity strength. We can note that at late times, the gravity strength, $G_N$, is greater (smaller) than predicted by GR when $\alpha_{M0} > 0 \, (< 0)$, up to 20% (30%) at $z = 0$. On the other hand, when $z \gg 0$, GR is recovered and we do not expect to have significant variations at early times.
{width="3.2in"} {width="3.2in"}
Constraints on parametric $f(R)$ gravity {#f(R)}
========================================
In this section, we briefly review $f(R)$ cosmology and show new observational constraints on the Hu-Sawicki (HS) model from our standard siren mock data set. We refer to [@fR1; @fR2; @Capozziello11; @Nojiri17] for reviews on $f(R)$ gravity.
The $f(R)$ gravitational theories consist of extending the Einstein-Hilbert action in the form $$S = \int d^4 x \sqrt{-g}\,\, \dfrac{M_P^2}{2}f(R) + S_m ,
\label{action0}$$ where $f(R)$ is a function of the Ricci scalar, and $S_m$ is the action for matter fields. For $f(R) = R$, the GR case is recovered.
Let us consider a spatially flat FLRW Universe dominated by pressureless matter (baryonic plus dark matter) and radiation with energy densities $\rho_m$, $\rho_r$ and pressures $P_m$, $P_r$, respectively. The modified Friedmann equations in the metric formalism are given by [@fR1; @fR2] $$\begin{aligned}
3FH^2=8\pi G \left(\rho_m+\rho_r\right) +\dfrac{1}{2} \left( FR - f \right)-3H\dot{F}\,, \label{FR1a} \\
-2F\dot{H} = 8\pi G \left( \rho_m + P_m +\rho_r + P_r \right)+\ddot{F}-H\dot{F}\
\label{FR2a}\end{aligned}$$ Moreover, one obtains the following useful relation: $$R=6\left(2H^2+\dot{H}\right).$$ It can be shown that, through the transformation of the scalar degree of freedom, $\phi = M_{P} df(R)/dR$, the metric $f(R)$ gravity is equivalent to the Brans-Dicke theory (with $w_{BD} = 0$). Thus, when considering a parametric function $f(R)$, given our constraints on $\phi$, some bounds can also be found on $f(R)$ gravity. For example, adopting the formalism presented in [@Basilakos; @Rafael] (and reference therein), one can write $$\label{fR_models}
f(R) = R - 2 \Lambda y(R, b),$$ where the function $y(R, b)$ quantifies the deviation from Einstein gravity, i.e. the effect of the $f(R)$ modification, through the distortion parameter $b$. Then, the scalar field can be expressed as $$\phi \simeq M_{P} \left[1 - 2 \Lambda \dfrac{\partial y(R, b)}{\partial R} \right].
$$ Interpreting $\phi_0 = M_{P}$ and given a function $y(R, b)$, we can use our limits on $\phi/\phi_0$ and place observational bounds on $b$. We note that the stability conditions assumed in Section \[sec-model\] and used in the development of this section are completely in agreement with $f(R)$ gravity [@alphai_02]. Clearly, more direct observational boundaries on $f(R)$ gravity can be obtained by modifying appropriately Eq. (\[dL\_Gw\]) to include a function $f(R)$. We are motivated to present a more detailed study in this sense in a future communication.
Without loss of generality, we can use our simulated standard siren catalog from BNS mergers, within a parametric limit and certain stability conditions, to model the $f(R)$ gravity dynamic. Thus, in what follows, we consider a parametric $f(R)$ gravity scenario and investigate the observational bounds that our standard siren mock data set can impose on the free parameters of the theory.
Let us consider viable models that have up to two parameters, where $f(R)$ function is given by Eq. (\[fR\_models\]). This methodology has been used earlier to investigate the observational constraints on $f(R)$ gravity in [@Basilakos; @Rafael]. In this respect, one of the most well-known models in the modified gravity theory literature is the Hu-Sawicki (HS) model [@Hu:2007nk], which satisfies all the dynamics conditions required for a given $f(R)$ function. The function $y(R, b)$ for the HS model is given by $$\label{y_HS}
y(R, b) = 1- \dfrac{1}{1+ \Bigl(\dfrac{R}{\Lambda b} \Bigr)^n},$$ where $n$ is an intrinsic parameter of the model. In what follows, we assume $n = 1$ and refer to [@Basilakos; @Rafael] for details.
As a direct application of the standard siren events from BNS mergers, within the sensitivity predicted for the ET, we proceed to constrain the HS model using our mock data set generated by the condition $\alpha_{M0} < 0$. Together with the standard siren data, in the present analysis we also employed the type Ia Supernova (SN) Pantheon data [@Scolnic18] and the cosmic chronometers (CC) measurements [@Moresco12] in order to obtain tighter constraints on the free parameters of the HS scenario (see Appendix B for the details on the SNe and CC data.) In the MCMC analysis, we assumed uniform priors for the cosmological parameters: $H_0 \in [55, 90]$, $\Omega_{m0} \in [0, 1]$, and $b \in [0, 1]$. In Figure \[fig:contours\_Hu-Sawicki\], we show the parametric space and the one-dimensional posterior distributions for the HS baseline model from SN + CC + GW and SN + CC analysis. Thus, we can quantify how much the addition of the GWs standard siren data can improve the constraints and break the degeneracy on the parametric space of the model, in particular on $b$. For comparison, we show also the constraints for the $\Lambda$CDM model. In particular, we found the following mean values with the relative 68% CL and 95% CL errors from the joint analysis (SNe + CC + GWs): $$\begin{aligned}
&H_0 = 69.37^{+0.67(1.45)}_{-0.80(1.41)}\ , \\
&\Omega_{m0} = 0.303^{+0.019(0.038)}_{-0.019(0.037)} \ , \\
&b = 0.383^{+0.134(0.229)}_{-0.116(0.257)}\ .\end{aligned}$$ We note that $b$ tends to be non-null up to the 95% CL with the addition of GWs data, therefore breaking the degeneracy enough to have a non-null value on $b$ up to 2$\sigma$. This demonstrates the potential of the future standard sirens catalogs in joint analysis with another geometric probes. As a comparison, in [@Rafael] it was found $b < 0.50$ and $b<0.13$ at the 95% CL from CC + $H_0$ and JLA + BAO + CC + $H_0$, respectively, while in [@Basilakos] it was found $b < 0.25 $ at the 68% CL. The data used in these works are from different physical nature and accuracy and, thus, a direct comparison seems not to be appropriate. However, all these constraints are compatible with each other within the 95% CL. Also, we can note how much the addition of GWs data can improve the constraints on $\Omega_{m0}$ and $H_0$ parameters. A detailed analysis from GW standard sirens on other viable $f(R)$ models as well as full discussion will be presented in a future communication.
Finally, it is interesting to analyze the consequences of our results on the effective dark energy equation of state parameter. This can be expressed as [@Basilakos] $$w_{DE}(a)=\dfrac{-1-\frac{2}{3}a \frac{d\ln{E}}{da}}{1-\Omega_m(a)},$$ where $E(a)= H(a)/H_0$ and $$\Omega_m(a)=\dfrac{\Omega_{m0}a^{-3}}{E^2(a)}.$$ In the case of the standard $\Lambda$CDM model, $w_{DE}=-1$ throughout the entire cosmological evolution. In Figure \[fig:wDE\], we show the $1\sigma$ and $2\sigma$ reconstructions of $w_{DE}(z)$ for the Hu-Sawicki model compared to the prediction of the standard $\Lambda$CDM model. The non-vanishing value of $b$ makes the effective dark energy term behave as quintessence at late times for $z < 0.4\ (0.1)$ at the 68% (95%) CL. Beyond this range, the effective equation of state is compatible $w_{DE} = -1$ at the 95% CL. The green line represents the evolution of $w_{DE}$ from the mean values of the MCMC analysis on SNe + CC + GWs.
![Parametric space at 68% and 95% CL and the one-dimensional marginalized distribution for $\Omega_{m0}$, $H_0$ and $b$, for the Hu-Sawicki model from the joint analysis SNe + CC + GW and SNe + CC. The predictions of the $\Lambda$CDM model are shown for comparison.[]{data-label="fig:contours_Hu-Sawicki"}](triangle_LCDM_HuS_SN_OHD_GW_alphaM0_negative.pdf){width="3.2in"}
![Reconstruction of the effective dark energy equation of state parameter at the 68% and 95% CL from the SN + CC + GW analysis for the Hu-Sawicki model. The prediction of the $\Lambda$CDM model (red solid line) is shown for comparison.[]{data-label="fig:wDE"}](w_z1.pdf){width="3.3in"}
Final remarks {#Conclusions}
=============
Assuming an effective luminosity distance for GWs, which can be physically interpreted as possible corrections on the GW amplitude propagation between source and detector, we obtained new observational constraints on the running of the Planck mass under appropriate stability conditions (see summary in Table \[tab:results\]), from 1000 standard siren events from binary neutron star mergers within the ET power spectral density noise. We found that the GW amplitude damping correction can be preferentially non-zero at the 68% CL and 95% CL, for $\alpha_{M0} < 0$, $\alpha_{M0} > 0$, respectively.
Furthermore, we combined our simulated GW data with the latest available SNe and CC measurements to constrain the parametric Hu-Sawicki gravity model. In doing so, we found that the deviation from GR may be evidenced as non-null at the 95% CL. These results demonstrate the statistical accuracy that can be achieved by future ground based GW observatory as ET detector.
The underlying assumption of our analysis is $c_T = c$ at all redshifts. This hypothesis is strongly motivated from the GW170817 constraints to hold locally. On the other hand, the evidence for $c_T = c$ might not be falsified due to the lack the gravitational wave observations at high $z$, even in the near future, since within the LIGO/VIRGO sensitivity, we expect to measure standard siren events from BNS only at very low $z$. Thus, as significant deviations from GR are expected only at moderate to high $z$ and/or large scale, it may be interesting to relax the condition $c_T/c = 1$ and perform forecast analyses on the ratio $c_T/c$ from binary systems at high $z$ within the sensitivity of the future detectors such as ET and DECIGO, and check for possible consequences on modified gravity phenomenology. It would be also interesting to apply a similar methodology as developed here to investigate other well-motivated classes of modified gravity models, as well as to perform joint analysis between standard siren mock events and CMB data (present and future missions). In this way, we will be able to determine the level of deviations from GR as suggested by the combination of future astronomical observations in the next decade.\
The authors are grateful to Salvatore Capozziello and Jose C. N. de Araujo for useful comments and discussions. R.C.N. would like to thank FAPESP for financial support under the project \# 2018/18036-5. The authors also wish to thank the referee for his/her constructive comments.
Appendix A: PN coefficients {#appendix-a-pn-coefficients .unnumbered}
===========================
For the convenience of the reader, we list below the PN coefficients [@TF2] we used for the waveform model Eq. (\[A\]) and Eq. (\[phi\]). The individual masses and spin parameters, $m_i$ and $\chi_i$ $(i = 1, 2)$, are encoded in the following parameter combinations: $$\begin{aligned}
&\delta = \ \dfrac{m_1 - m_2}{M}, \\
&\chi_s = \ \dfrac{\chi_1 + \chi_2}{2}, \\
&\chi_a = \ \dfrac{\chi_1 - \chi_2}{2}.\end{aligned}$$\
The PN amplitude expansion coefficients are:
$$\begin{aligned}
A_0 = &\ 1, \\
A_1 = &\ 0, \\
A_2 = &\ \dfrac{451}{168} - \dfrac{323}{224}, \\
A_3 = &\ \dfrac{27 \delta \chi_a}{8} + \left(\dfrac{27}{8} - \dfrac{11 \eta}{6} \right) \chi_s, \\
A_4 = &\ \dfrac{27312085}{8128512} - \dfrac{1975055 \eta}{338688} + \dfrac{105271 \eta^2}{24192}+ \left( 8\eta + \dfrac{81}{32} \right) \chi^2_a - \dfrac{81}{16} \delta \chi_a \chi_s + \left(\dfrac{81}{32} + \dfrac{17 \eta}{8} \right) \chi^2_s, \\
A_5 = & - \dfrac{85 \pi}{64} + \dfrac{85 \pi \eta}{16} + \delta \left( \dfrac{285197}{16128} - \dfrac{1579 \eta}{4032} \right) \chi_a + \left( \dfrac{285197}{16128} - \dfrac{15317 \eta}{672} - \dfrac{2227 \eta^2}{1008} \right) \chi_s, \\
A_6 = & - \dfrac{177520268561}{8583708672} + \Big( \dfrac{545384828789}{5007163392} - \dfrac{205 \pi^2}{48}\Big) \eta - \dfrac{3248849057 \eta^2}{178827264}
+ \dfrac{34473079 \eta^3}{6386688} \\
&+ \Big( \dfrac{1614569}{64512} - \dfrac{1873643 \eta}{16128} + \dfrac{2167 \eta^2}{42} \Big) \chi^2_a +
\Big( \dfrac{31 \pi}{12} - \dfrac{7 \pi \eta}{3} \Big) \chi_s + \Big( \dfrac{1614569}{64512} - \dfrac{61391 \eta}{1344} + \dfrac{57451 \eta^2}{4032} \Big) \chi^2_s \\
&+
\delta \chi_a \Big( \dfrac{31 \pi}{12} + \Big( \dfrac{1614569}{32256} - \dfrac{165961 \eta}{2688} \Big) \chi_s \Big) \end{aligned}$$
The phase $\Phi(f)$ expansion coefficients are:
$$\begin{aligned}
\alpha_2 = &\ \dfrac{3715}{756} + \dfrac{55 \eta}{9}, \\
\alpha_3 = & -16\pi + \dfrac{113 \delta \chi_a}{3} + \left( \dfrac{113}{3} - \dfrac{76 \eta}{3} \right) \chi_s, \\
\alpha_4 = &\ \dfrac{15293365}{508032} + \dfrac{27145 \eta}{504} + \dfrac{3085 \eta^2}{72} + \left(200 \eta - \dfrac{405}{8} \right) \chi^2_a - \dfrac{405}{4} \delta \chi_a \chi_s + \left(\dfrac{5 \eta}{2} - \dfrac{405}{8} \right) \chi^2_s, \\
\alpha_5 = &\ [ 1 + \log (\pi M f) ] \left[ \dfrac{38645\pi}{756} - \dfrac{65 \pi \eta}{9} + \delta \left( -\dfrac{140 \eta}{9} - \dfrac{732985}{2268} \right) \chi_a
+ \left( - \dfrac{732985}{2268} + \dfrac{24260 \eta}{81} + \dfrac{340 \eta^2}{9} \right) \chi_s \right], \\
\alpha_6 = &\ \dfrac{11583231236531}{4694215680} - \dfrac{6848 \gamma_E}{21} - \dfrac{640 \pi^2}{3} + \left(\dfrac{15737765635}{3048192} + \dfrac{2255 \pi^2}{12} \right) \eta + \dfrac{76055 \eta^2}{1728} - \dfrac{127825 \eta^3}{1296} \\
&- \dfrac{6848}{63} \log(64 \pi M f) + \dfrac{2270}{3} \pi \delta \chi_a \left( \dfrac{2270 \pi}{3} - 520 \pi \eta \right) \chi_s, \\
\alpha_7 = &\ \dfrac{77096675 \pi}{254016} + \dfrac{378515 \pi \eta}{1512} - \dfrac{74045 \pi \eta^2}{756} + \delta \left(- \dfrac{25150083775}{3048192} + \dfrac{26804935 \eta}{6048} - \dfrac{1985 \eta^2}{48} \right)\chi_a \\
&+ \left( -\dfrac{25150083775}{3048192} + \dfrac{10566655595 \eta}{762048} - \dfrac{1042165 \eta^2}{3024} + \dfrac{5345 \eta^3}{36} \right) \chi_s.\end{aligned}$$
Appendix B: SNe and CC data set {#appendix-b-sne-and-cc-data-set .unnumbered}
===============================
We provide here some details of the low-redshift cosmological observables that we used to complement the GW mock data in the statistical analysis on the HS model.\
The first data set is the Pantheon sample [@Scolnic18], composed of 1048 Supernovae (SNe) Ia in the redshift range $0.01<z<2.3$. In this compilation, all the SNe are standardized through the SALT2 light-curve fitter, in which the distance modulus is modelled as follows [@SALT2]: $$\mu=m_B-M+\alpha x_1-\beta C + \Delta_M +\Delta_B ,
\label{mu_SALT2}$$ where $m_b$ is the $B$-band apparent magnitude of each SN and $M$ is its absolute magnitude, while $\Delta_M$ and $\Delta_B$ account for the host-mass galaxy and the distance bias corrections, respectively. Moreover, $x_1$ and $C$ are the stretch and color parameters of each SN light-curve, respectively, with their relative coefficients $\alpha$ and $\beta$. On the other hand, the distance modulus predicted by a cosmological model is given as $$\mu(z)=5\log_{10}\left[\dfrac{d_L(z)}{1\text{ Mpc}}\right]+25.$$ As shown in [@Riess18], under the assumption of a flat universe one can compress the full SN sample into a set of 6 cosmological model-independent measurements of $E(z)^{-1}$, where $E(z)\equiv H(z)/H_0$ is the dimensionless Hubble parameter. This approach allows us to properly marginalize over the SN nuisance parameters in the fit. Thus, taking into account the correlations among the $E^{-1}(z)$ measurements, we can write the likelihood function associated to the SN data as [@DAgostino19] $$\mathcal{L}_\text{SN}\propto \exp\left[-\dfrac{1}{2}\mathbf{A}^\text{T} \mathbf{C}_\text{SN}^{-1} \mathbf{A}\right],$$ where $\mathbf{A}\equiv E^{-1}_{obs,i}-E^{-1}_{th}(z_i)$ quantifies the difference between the measured values and the values predicted by a cosmological model, and $\mathbf{C}_\text{SN}$ is the covariance matrix resulting from the correlation matrix given in [@Riess18].\
The second data set is built upon the differential age approach [@Jimenez02], which represents a model-independent method to characterize the expansion of the universe up to $z<2$. In this technique, passively evolving red galaxies are used as cosmic chronometers (CC) to measure the age difference $(dt)$ of the universe at two close redshifts $(dz)$. Thus, one can estimate the Hubble parameter as $$H(z)=-\dfrac{1}{(1+z)}\dfrac{dz}{dt} .$$ In our analysis, we used the compilation of 31 $H(z)$ uncorrelated measurements collected in [@Capozziello18] (see references therein). We can then write the likelihood function relative to the CC data as $$\mathcal{L}_\text{CC}\propto\exp\left[-\dfrac{1}{2}\displaystyle{\sum_{i=1}^{31}}\left(\dfrac{H_{obs,i}-H_{th}(z_i)}{\sigma_{H,i}}\right)^2\right] ,$$ where $H_{obs,i}$ are the observed measurements with their relative uncertainties $\sigma_{H,i}$, while $H_{th}(z_i)$ are the theoretical values of the Hubble parameter estimated from using a specific cosmological model.
[99]{}
D. Huterer and D. L. Shafer, Rep. Prog. Phys. [**81**]{}, 016901 (2018), arXiv:1709.01091 \[astro-ph.CO\].
S. Capozziello, R. D’Agostino and O. Luongo, doi:10.1142/S0218271819300167, arXiv:1904.01427 \[gr-qc\] (2019).
M. Ishak, Living Rev. Rel. [**1**]{}, 22 (2019), arXiv:1806.10122 \[astro-ph.CO\].
B. P. Abbott et al., arXiv:1811.12907 \[astro-ph.HE\].
B. P. Abbott et al., Phys. Rev. Lett. **119**, 161101 (2017), arXiv:1710.05832 \[gr-qc\].
T. Baker et al., Phys. Rev. Lett. **119**, 251301 (2017), arXiv:1710.06394 \[astro-ph.CO\].
P. Creminelli and F. Vernizzi, Phys. Rev. Lett. **119**, 251302 (2017), arXiv:1710.05877 \[astro-ph.CO\].
J. Sakstein and J. Jain, Phys. Rev. Lett. **119**, 251303 (2017), arXiv:1710.05893 \[astro-ph.CO\].
J. M. Ezquiaga and M. Zumalacárregui, Phys. Rev. Lett. [**119**]{}, 251304 (2017), arXiv:1710.05901 \[astro-ph.CO\].
S. Boran, S. Desai, E. Kahya, R. Woodard, Phys. Rev. D **97**, 041501 (2018), arXiv:1710.06168 \[astro-ph.HE\].
J. M. Ezquiaga and M. Zumalacárregui, Front. Astron. Space Sci. [**5**]{}, 44 (2018), arXiv:1807.09241 \[astro-ph.CO\].
B. P. Abbott et al., Nature **551** 7678 (2017), arXiv:1710.05835 \[astro-ph.CO\].
H. Y. Chen, M. Fishbach and D. E. Holz, Nature **562** 7728 (2018), arXiv:1712.06531 \[astro-ph.CO\].
S. M. Feeney et al., Phys. Rev. Lett. **122**, 061105 (2019), arXiv:1802.03404 \[astro-ph.CO\].
A. Nishizawa, Phys. Rev. D **96**, 101303 (2017), arXiv:1612.06060 \[astro-ph.CO\].
L. S. Collaboration, “Ligo-t1400316-v4: Instrument science white paper".
Einstein Telescope Science Team, ET Document: ET-0106C-10, 2011. http://www.et-gw.eu/.
M. Punturo, et al., Class. Quant. Grav. [**27**]{}, 173001 (2010), arXiv:1003.2480 \[astro-ph.HE\].
H. Audleyet et al. (2017), arXiv:1702.00786 \[astro-ph.IM\].
N. Seto, S. Kawamura and T. Nakamura, Phys. Rev. Lett. [**87**]{}, 221103 (2001), arXiv: 0108011 \[astro-ph\].
J. Luo et al., Class. Quant. Grav. **33** 3 035010 (2016), arXiv:1512.02076 \[astro-ph.IM\].
B. S. Sathyaprakash, B. F. Schutz and C. Van Den Broeck, Class. Quant. Grav. [**27**]{}, 215006 (2010), arXiv:0906.4151 \[astro-ph.CO\].
W. Zhao, C. Van Den Broeck, D. Baskaran and T. G. F. Li, Phys. Rev. D [**83**]{}, 023005 (2011), arXiv:1009.0206 \[astro-ph.CO\].
R. G. Cai and T. Yang, Phys. Rev. D [**95**]{}, 044024 (2017), arXiv:1608.08008 \[astro-ph.CO\].
M. Du et al., arXiv:1812.01440 \[astro-ph.CO\].
W. Yang et al., arXiv:1904.11980 \[astro-ph.CO\].
J. Wei, Astrophys. J. 868, [**1**]{} 29 (2018), arXiv:1806.09781 \[astro-ph.CO\].
X. N. Zhang et al., arXiv:1804.08379 \[astro-ph.CO\].
L. F. Wang, X. N. Zhang, J. F. Zhang and X. Zhang, Physics Letters B [**782**]{}, 87 (2018), arXiv:1802.04720 \[astro-ph.CO\].
R. G. Cai et al., Phys. Rev. D [**97**]{}, 103005 (2018), arXiv:1712.00952 \[astro-ph.CO\].
T. Yang, R. F. L. Holanda and B. Hu, Astropart. Phys. [**108**]{} (2019), arXiv:1710.10929 \[astro-ph.CO\].
X. Zhang, J. Yu, T. Liu, W. Zhao, and A. Wang, Phys. Rev. D [**95**]{}, 124008 (2017), arXiv:1703.09853 \[gr-qc\].
J. F. Zhang, H. Y. Dong, J. Z. Qi and X. Zhang, arXiv:1906.07504 \[astro-ph.CO\].
J. Mendonca and R. Sturani, arXiv:1905.03848 \[gr-qc\].
J. Z. Qi et al., Phys. Rev. D [**99**]{}, 063507 (2019), arXiv:1902.01988 \[astro-ph.CO\].
J. Z. Qi, S. Cao, Y. Pan and J. Li, arXiv:1902.01702 \[astro-ph.CO\].
W. Yang et al., arXiv:1905.08286 \[astro-ph.CO\].
X. Fu, L. Zhou and J. Chen, Phys. Rev. D [**99**]{}, 083523 (2019), arXiv:1903.09913 \[gr-qc\].
X. N. Zhang, L. F. Wang, J. F. Zhang and X. Zhang, Phys. Rev. D [**99**]{}, 063510 (2019), arXiv:1804.08379 \[astro-ph.CO\].
S. Vitale and M. Evans, Phys. Rev. D [**95**]{}, 064052 (2017), arXiv:1610.06917 \[gr-qc\].
S. R. Taylor and J. R. Gair, Phys. Rev. D [**86**]{}, 023502 (2012), arXiv:1204.6739 \[astro-ph.CO\].
R. R. A. Bachega, E. Abdalla, and K. S. F. Fornazier, arXiv:1906.08909 \[astro-ph.CO\].
K. Liao, Phys. Rev. D [**99**]{}, 083514 (2019), arXiv:1904.01744 \[astro-ph.CO\].
J. F. Zhang, M. Zhang, S. J. Jin, J. Z. Qi and X. Zhang, arXiv:1907.03238 \[astro-ph.CO\].
C. Deffayet, X. Gao, D. A. Steer and G. Zahariade, Phys. Rev. D **84**, 064039 (2011), arXiv:1103.3260 \[hep-th\].
T. Kobayashi, M. Yamaguchi and J. Yokoyama, Prog. Theor. Phys. **126**, 511 (2011), arXiv:1105.5723 \[hep-th\].
G. W. Horndeski, Int. J. Theor. Phys. **10**, 363 (1974).
T. Kobayashi, arXiv:1901.07183 \[gr-qc\].
R. Kase and S. Tsujikawa, Int. J. Mod. Phys. D **28** 05, 1942005 (2019), arXiv:1809.08735 \[gr-qc\].
I. D. Saltas, I. Sawicki, L. Amendola and M. Kunz (2014), arXiv:1406.7139 \[astro-ph.CO\].
A. Nishizawa, Phys. Rev. D [**97**]{}, 104037 (2018), arXiv:1710.04825 \[gr-qc\].
A. Nishizawa and S. Arai, Phys. Rev. D [**99**]{}, 104038 (2019), arXiv:1901.08249 \[gr-qc\].
S. Arai and A. Nishizawa, Phys. Rev. D [**97**]{}, 104038 (2018), arXiv:1711.03776 \[gr-qc\]
E. Belgacem, Y. Dirian, S. Foffa, and M. Maggiore, Phys. Rev. D [**97**]{}, 104066 (2018), arXiv:1712.08108 \[astroph.CO\].
E. Belgacem, Y. Dirian, S. Foffa, and M. Maggiore, Phys. Rev. D [**98**]{}, 023510 (2018), arXiv:1805.08731 \[gr-qc\].
M. Lagos, M. Fishbach, P. Landry and D. E. Holz, arXiv:1901.03321 \[astro-ph.CO\].
R. C. Nunes, M. E. S. Alves and J. C. N. de Araujo, arXiv:1905.03237 \[gr-qc\].
E. Belgacem, Y. Dirian, A. Finke, S. Foffa and M. Maggiore, arXiv:1907.02047 \[astro-ph.CO\].
E. Belgacem et al., arXiv:1906.01593 \[astro-ph.CO\].
X. Gao and X. Y. Hong, arXiv:1906.07131 \[gr-qc\].
C. Dalang and L. Lombriser, arXiv:1906.12333 \[astro-ph.CO\].
E. Belgacem et al., arXiv:1907.01487 \[astro-ph.CO\].
G. Calcagni et al., arXiv:1907.02489 \[gr-qc\].
E. Bellini and I. Sawicki, J. Cosm. Astrop. Phys. **1407**, 050 (2014), arXiv:1404.3713 \[astro-ph.CO\].
J. Kennedy, L. Lombriser and Andy Taylor, Phys. Rev. D **98**, 044051 (2018), arXiv:1804.04582 \[astro-ph.CO\].
M. Denissenya and E. V. Linder, J. Cosm. Astrop. Phys., [**1811**]{}, 010 (2018), arXiv:1808.00013 \[astro-ph.CO\].
M. Zumalacarregui, E. Bellini, I. Sawicki, J. Lesgourgues, P. G. Ferreira, J. Cosm. Astrop. Phys. **08** 019 (2017), arXiv:1605.06102 \[astro-ph.CO\].
B. F. Schutz, Nature, [**323**]{}, 310 (1986).
D. E. Holz and S. A. Hughes, Astrophys. J., [**629**]{}, 15 (2005), arXiv:astro-ph/0504616.
H. Cramer, Mathematical Methods of Statistics (Princeton University Press, Princeton, 1946).
C. R. Rao, in Bulletin of the Calcutta Mathematical Society (1945).
M. Vallisneri, Phys. Rev. Lett. [**107**]{}, 191104 (2011), arXiv:1108.1158 \[gr-qc\].
C. Cutler and E. Flanagan, Phys. Rev. D [**49**]{} 2658-2697 (1994), arXiv:gr-qc/9402014.
K. G. Arun, B. R. Iyer, B. S. Sathyaprakash, P. A. Sundararajan, Phys. Rev. D [**71**]{} 084008 (2005), arXiv:gr-qc/0411146.
H. S. Cho and C. H. Lee, arXiv:1310.3050 \[gr-qc\].
C. L. Rodriguez, B. Farr, W. M. Farr, I. Mandel, Phys. Rev. D [**88**]{}, 084013 (2013), arXiv:1308.1397 \[astro-ph.IM\].
E. Berti, A. Buonanno, and C. M. Will, Phys. Rev. D [**71**]{} 084025 (2005).
S. Nissanke, D. E. Holz, S. A. Hughes, N. Dalal and J. L. Sievers, Astrophys. J. [**725**]{}, 496 (2010), arXiv:0904.1017 \[astro-ph.CO\].
N. Metropolis et al., J. Chem. Phys. 21, 1087 (1953); W. K. Hastings, Biometrika, [**57**]{}, 97 (1970).
J. Espejo, S. Peirone, M. Raveri, K. Koyama, L. Pogosian, and A. Silvestri, arXiv:1809.01121 \[astro-ph.CO\].
E. Bellini, A. J. Cuesta, R. Jimenez, and L. Verde, J. Cosm. Astrop. Phys. **1602** 02, 053 (2016), arXiv:1509.07816 \[astro-ph.CO\].
D. Alonso, E. Bellini, P. G. Ferreira, and M. Zumalacarregui, Phys. Rev. D **95**, 063502 (2017), arXiv:1610.09290 \[astro-ph.CO\].
N. Frusciante, S. Peirone, S. Casas, and N. A. Lima, Phys. Rev. D [**99**]{}, 063538 (2019), arXiv:1810.10521 \[astro-ph.CO\].
C. D. Kreisch and E. Komatsu, J. Cosm. Astrop. Phys. [**12**]{}, 030 (2018), arXiv:1712.02710 \[astro-ph.CO\].
R. A. Battye, F. Pace, and D. Trinh, Phys. Rev. D **98**, 023504 (2018), arXiv:1802.09447 \[astro-ph.CO\].
J. Kennedy, L. Lombriser and A. Taylor, Phys. Rev. D **98**, 044051 (2018), arXiv:1804.04582 \[astro-ph.CO\].
M. Brush, E. V. Linder, M. Zumalacarregui, J. Cosm. Astrop. Phys. [**1901**]{}, 029 (2019), arXiv:1810.12337 \[astro-ph.CO\].
J. Noller and A. Nicola, arXiv:1811.03082 \[astro-ph.CO\].
J. Gleyzes, Phys. Rev. D **96**, 063516 (2017), arXiv:1705.04714 \[astro-ph.CO\].
E. V. Linder, J. Cosm. Astrop. Phys. [**1803**]{}, 005 (2018), arXiv:1801.01503 \[astro-ph.CO\].
\[Planck Collaboration\] N. Aghanim et al., arXiv:1807.06209 \[astro-ph.CO\].
R. C. Nunes, M. E. S. Alves and J. C. N. de Araujo, Phys. Rev. D [**99**]{}, 084022 (2019), arXiv:1811.12760 \[gr-qc\].
C. Brans, R. H. Dicke, Phys. Rev. [**124**]{}, 925 (1961).
S. Basilakos, S. Nesseris and L. Perivolaropoulos, Phys. Rev. D [**87**]{}, 12, 123529 (2013). R. C. Nunes, S. Pan, E. N. Saridakis and E. M. C. Abreu, J. Cosm. Astrop. Phys. [**1701**]{} 01, 005 (2017), arXiv:1610.07518 \[astro-ph.CO\].
A. De Felice and S. Tsujikawa, Living Rev. Rel. [**13**]{}, 3 (2010), arXiv:1002.4928 \[gr-qc\].
T. P. Sotiriou and V. Faraoni, Rev. Mod. Phys. [**82**]{}, 451-497 (2010), arXiv:0805.1726 \[gr-qc\].
S. Capozziello and M. De Laurentis, Phys. Rept. [**509**]{}, 167-321 (2011), arXiv:1108.6266 \[gr-qc\].
S. Nojiri, S. D. Odintsov and V. K. Oikonomou, Phys. Rept. [**692**]{}, 1-104 (2017), arXiv:1705.11098 \[gr-qc\].
W. Hu and I. Sawicki, Phys. Rev. D [**76**]{}, 064004 (2007), arXiv:0705.1158 \[astro-ph\].
D. M. Scolnic, Astrophys. J. [**859**]{}, 101 (2018), arXiv:1710.00845 \[astro-ph.CO\].
M. Moresco et al., JCAP [**08**]{}, 006 (2012), arXiv:1201.3609 \[astro-ph.CO\].
S. Khan et al., Phys. Rev. D [**93**]{}, 044007 (2016), arXiv:1508.07253 \[gr-qc\].
J. Guy et al., Astron. Astrophys. [**466**]{}, 11 (2007), arXiv:astro-ph/0701828.
A. G. Riess et al., Astrophys. J. [**853**]{}, 126 (2018), arXiv:1710.00844 \[astro-ph.CO\].
R. Jimenez and A. Loeb, Astrophys. J. [**573**]{}, 37 (2002), arXiv:astro-ph/0106145.
R. D’Agostino, Phys. Rev. D [**99**]{}, 103524 (2019), arXiv:1903.03836 \[gr-qc\].
S. Capozziello, R. D’Agostino and O. Luongo, Mon. Not. Roy. Astron. Soc. [**476**]{}, 3924 (2018), arXiv:1712.04380 \[astro-ph.CO\].
[^1]: The redshift is defined as $z=a^{-1}-1$.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Atom-atom interactions within a small volume were investigated through the excitation of ultracold Rb atoms. The application of microwaves enhances these interactions, causing the suppression of Rydberg state excitation. The suppression of Rydberg atom excitation was both qualitatively observed and quantitatively analyzed using a universal scaling law, giving a measure of the atom-atom interaction strength in agreement with theoretical prediction.'
author:
- 'E. Brekke'
- 'J. O. Day'
- 'T. G. Walker'
bibliography:
- 'MyCollection.bib'
title: 'Excitation Suppression Due to Interactions Between Microwave-Dressed Rydberg Atoms'
---
Introduction
============
Strong dipole-dipole interactions between atoms in Rydberg states allow conditional quantum manipulations of atom pairs at micron-scale interatomic separations [@Saffman2010a]. The blockade effect [@Jaksch2000] uses these interactions to generate entanglement, as recently demonstrated experimentally [@Wilk2010; @Isenhower2010a; @Zhang2010]. A related fundamental goal is quantum manipulation of small ensembles of atoms using the blockade effect [@Lukin2001]. Collective Rabi flopping was recently experimentally observed [@Dudin2012]. In extended samples, [*i.e.*]{} atom clouds whose volume substantially exceeds the range of the Rydberg-Rydberg interactions, a number of experiments (reviewed in Ref. [@Saffman2010a]) have observed signatures of collective blockade.
In such extended samples, excitation of more than one atom per blockade volume is suppressed. The excitation of a single atom to a Rydberg state is sufficient to shift the corresponding energy levels of its neighbors out of resonance with the exciting laser photons. If there are $N$ atoms within the blockade volume, the effective Rabi frequency of the excitation of the blockaded “superatom" is collectively enhanced by $\sqrt{N}$. The Stuttgart group [@Heidemann2007; @Heidemann2008; @Low2009] has derived and experimentally confirmed the existence of non-trivial scaling relations that show how the excitation fraction depends on experimental parameters such as density and intensity.
In the main, experiments on blockade in extended samples have been done in the absence of applied fields, so that the Rydberg-Rydberg interactions are dominated by $r^{-6}$ van der Waals interactions[@Walker2008], where $r$ is the interatomic separation. These can be tuned with DC electric fields to give Förster resonances, with potentially dramatic increases in interaction strength. A variety of experiments have utilized this capability, as reviewed in [@Comparat2010].
Rydberg atoms [also]{} interact strongly with microwave electric fields. In the context of laser cooled Rydberg atoms, microwave spectroscopy has been used to probe energy level shifts and atomic interactions [@Younge2010; @Anderson2011; @Park2011]. Microwaves were utilized to transfer population to nearby states [@Maeda2011; @Afrousheh2004], and to investigate ionization of Rydberg states [@Park2011b].
Microwaves can also be used to increase the interaction strength between Rydberg atoms. The strong coupling of Rydberg atoms to resonant microwave fields makes it easy to dress Rydberg atoms in superpositions of opposite parity states, so that the atoms possess oscillating dipoles and thereby interact via a dipole-dipole interaction of comparable strength ($\sim n^4(ea_0)^2/r^3$) to the Förster interaction. The first evidence for enhanced dipole-dipole interactions from resonant microwaves was recently presented in the context of EIT in cold Rydberg gases [@Tanasittikosol2011]. In that work, Autler-Townes dressing led to enhanced suppression of EIT (as compared to van der Waals interactions) due to the stronger, longer range dipole-dipole interaction.
In this paper we use all-optical means to study the enhanced interactions between microwave-dressed Rydberg atoms. Using stimulated emission detection, we observe the suppression of Rydberg excitation in a magneto-optical trap. By monitoring the number of Rydberg atoms created by two-photon excitation while varying the density and laser intensity, a deviation from linear dependence is observed when microwave coupling is applied. The resulting suppression in the excitation of Rydberg atoms is not only qualitatively observed, but additionally quantitatively compared to theoretical predictions using scaling law arguments and the superatom concept. The results confirm the enhancement of Rydberg-Rydberg interactions through microwave coupling.
Our experiment, described in Sec. \[sec:expt\], consists of exciting 47s Rydberg states using two-photon excitation pulses. We observe the Rydberg production by laser deexcitation to the 6p state, detecting the subsequent 6p-6s photons. Analysis of the transients allows measurement of the excitation rates. In Sec. \[sec:scaling\], we show how the superatom concept leads to a simple model for the dependence of Rydberg production on density and intensity, for the case of a 1/$r^3$ interaction between the atoms. We then present a calculation of the expected interaction strengths for microwave-dressed 47s Rydberg atoms, including the effects of angular momentum structure. We then present in Sec. \[sec:suppression\] our results on suppression of Rydberg excitation, comparing microwave dressed and undressed atoms, and compare to the scaling theory.
Experimental Setup {#sec:expt}
==================
The basic geometry and energy-level scheme of the experiment is shown in Fig. \[fig:elevel\]. The cold atom sample is a magneto-optical trap of $^{87}$Rb atoms in the F=2 state. In order to produce a fairly small volume of excited Rydberg atoms, we cross a 780 nm laser beam with a 50 $\mu$m waist with another 480 nm laser focused to about 13 $\mu$m. The two lasers are tuned to produce 47s Rydberg atoms via the pathway $5s_{1/2}$(F=2)$\rightarrow5p_{3/2}\rightarrow47s$, detuned $\Delta=2\pi\times500$ MHz above the intermediate $5p_{3/2}$ F$'=3$ state. The Rydberg atoms are de-excited by stimulated emission from a uniform intensity 1016 nm laser tuned to the $47s\rightarrow6p_{3/2}($F$'=2)$ transition, with detection of the process by observation of the subsequent $6p_{3/2}\rightarrow5s_{1/2}$ fluorescence [@Day2008; @Brekke2008].
![(Color online) a) Energy levels for Rydberg atom production and detection. Two photon excitation produces Rydberg atoms, and a stimulated emission probe moves them to the 6p state where decay photons are detected. b) Experimental setup for the production of Rydberg atoms in a small spatial sample. Not to scale.[]{data-label="fig:elevel"}](elevel3.eps){width="3.5"}
The Rabi frequency of the 780 nm laser is $\Omega=2\pi \times 370$ MHz at its focus, giving rise to a spatially dependent repulsive AC Stark shift of about ${\Omega_{780}^2}/{4 \Delta}\sim2\pi\times$ 75 MHz for the 5$s$ atoms. This inhomogeneously broadens the two-photon excitation process, as illustrated in Fig. \[fig:acshift\]. There we show the 6$p$ fluorescence as a function of the frequency of the 480 nm beam, measured relative to the free-space resonance frequency. By tuning the 480 nm laser to the “Deep" end of the spectrum, we selectively excite atoms at the center of the 780 beam. We estimate that this reduces the effective waist of the 780 nm beam to about 20 $\mu$m, so that the Rydberg atoms are produced from an excitation volume of about 2$\times$20 $\mu$m$\times\pi\times$(13$\mu$m)$^2=2\times10^4$ $\mu$m$^3$. [All results in this paper were obtained by tuning the 480 nm laser to the “Deep" portion of the spectrum at -68 MHz. ]{}
![a) Rydberg atoms are created over a large range of 480 nm frequencies due to the spatial variation of the AC Stark shift from the 780 nm beam. This could give extra spatial selection by choosing the largest shift, and hence the most intense part of the 780 nm beam, here labeled as “Deep". b) With resonant microwaves coupling the ${47s}$ state to the ${47p_{1/2}}$ state, the resulting Autler-Townes splitting can be observed. []{data-label="fig:acshift"}](acshift.eps)
Left on continuously, the repulsion of the AC Stark shift would deplete the local atom density. Thus the 780 nm laser is pulsed on for only 20 $\mu$s at a time, once every 500 $\mu$s, reducing the time-averaged force by a factor of 25. The 780 nm laser also produces significant depumping of the F=2 atoms; we estimate that about 1/4 of the atoms will be in the F=1 state at the center of the excitation region.
The 5$p_{3/2}\rightarrow47s$ Rabi frequency at the center of the crossed 480 nm beam can be calculated from the laser intensities and matrix elements, resulting in $\Omega_{480}$=2$\pi\times$11.2 MHz. Combining this with the 780 Rabi frequency, we estimate the effective two-photon Rabi frequency to be $\Omega_{2}$=${\Omega_{780}\Omega_{480}/2\Delta}$=2$\pi\times$3.8 MHz.
As discussed in our previous work [@Brekke2008; @Day2008], we find that the linewidth for excitation of Rydberg states by our lasers is about 2$\pi\times$6 MHz. (The lasers are all external cavity diode lasers, locked to $\sim$ 10 MHz linewidth transfer cavities that are themselves locked to Rb saturated absorption lines [@Brekke2009].) The linewidth being larger than the two-photon Rabi frequency, we conclude that the excitation process is best modeled by assuming incoherent excitation. Thus we will analyze the population dynamics for this paper in terms of populations and excitation rates, instead of using optical Bloch equations. For excitation of a transition with a linewidth $\gamma$, the effective excitation rate is $R = 2 \pi {\Omega^2/\gamma}$ which gives an expected peak two-photon excitation rate of $R_2=1.5 \cdot 10^7 /$s at the center of the crossed lasers. As a result of the distribution of excitation rates within the atomic sample, the mean excitation rate is expected to be $R_2$=$3.7 \cdot 10^6 $/s.
[We detect Rydberg excitation by adding a third spatially uniform laser]{} that is tuned to be resonant on the 47$s_{1/2}\longrightarrow6p_{3/2}$ transition at 1016 nm. [Atoms in the 6$p_{3/2}$ state decay rapidly]{} by spontaneous emission of 420 nm photons. The intensity of this laser was chosen to give a stimulated emission rate of typically $R_3=3\times 10^5$/s. This value was chosen as a compromise between small signal size (when $R_3$ is significantly smaller), or insensitivity to suppression effects (when $R_3$ is larger).
The 420 nm photons, detected by a Hamamatsu H7360-01 photon counting module, are binned in 100 ns intervals. The photon counter is triggered with the start of each pulse of the 780 nm beam, and the counts collected from many of these pulses averaged together. Fig. \[fig:Spulse\] shows the result of averaging 300,000 such time sequences. Since the decay rate from the intermediate 6$p_{3/2}$ state is significantly faster than the excitation rates used in our sample, the 420 nm photon count rate is proportional to the number of Rydberg atoms. From the count rate, we determined the number of Rydberg atoms using the known solid angle and branching ratio ($2.4 \times 10^{-4} $), collection efficiency (12.7%), and the measured stimulated emission rate.
![The photon count rate is recorded vs the time after the start of the excitation pulse, and averaged for many cycles. The number of Rydberg atoms is proportional to the equilibrium count rate, and the rise/fall times of the signal are experimental measures of the excitation and stimulated emission rates. []{data-label="fig:Spulse"}](Spulse2.eps){width="3.5"}
The rise and fall times of the transients in Fig. \[fig:Spulse\] gives an experimental measure of the excitation and stimulated emission rates through an analysis of atomic populations. The resulting excitation is incoherent, and the optical Bloch equations can be simplified to rate equations for the three level system:
$$\begin{aligned}
{dN_g\over dt} &=& R_2 (N_r-N_g) + \Gamma_{6p} N_{6p} \nonumber\\
{dN_{r}\over dt} &=& R_2 (N_g - N_r) - R_{3} (N_{r}-N_{6p}) \nonumber \\
{dN_{6p}\over dt} &=& R_{3} (N_{r}-N_{6p}) - \Gamma_{6p} N_{6p},
\label{eqn:risemodel}\end{aligned}$$
Since $\Gamma_{6p} >> R_3$, $N_{6p} \propto N_{r}$, allowing $N_{6p}$ to be eliminated. Solving the resulting two level system gives: $$N_r= \frac{R_2}{2R_2+R_{3}}(1-e^{-(2 R_2 + R_{3})t}).
\label{eqn:nsat}$$ The observed rise time is then equal to ${1}/(2 R_2+R_{3})$ and fall time (in the absence of excitation light) is equal to ${1}/{R_{3}}$. The stimulated emission rate $R_3$ determined from the fit to the data agrees within a factor of four of our calculated value. The excitation rate $R_2$ determined from the fits is 0.6$\pm$0.2$\times 10^6$/s, smaller than the expected value by a factor of six. It should be noted from Fig. \[fig:Spulse\] that there is an initial very rapid increase in population for the first microsecond after turning on the Rydberg excitation, then a slower population increase after that. For this data we are assuming little or no blockade, but if there is some blockade, one might explain this behavior by noting that as excited atoms move it might be possible for previously blockaded atoms to be excited. This effect might explain why the excitation transient in Fig. \[fig:Spulse\] is slower than expected. For the following analysis, we have used the deduced values of $R_2$ and $R_3$ from data such as Fig. \[fig:Spulse\] at low densities where suppression effects should be small.
Finally, in order to assure an accurate correlation of photon counts to number of Rydberg atoms created, there are two concerns which must be addressed: Superradiance and collision induced ionization. We have previously seen that superradiance has played a key role in Rydberg dynamics [@Day2008]. However, in this experiment the small excitation size greatly reduces the number of Rydberg atoms, and we calculate the expected superradiant rates to be negligible. In this environment collision induced ionization and subsequent electron trapping present a more serious concern due to the high Rydberg densities. Collision induced ionization and subsequent electron trapping have been shown to greatly affect the Rydberg population [@Li04] [@Walz-Flannigan2004]. The effect of ionization is often difficult to differentiate from a blockade signal [@Vogt2007], and so it is important to prevent this process in our sample. Indeed, when we switched to excitation of d-states, we found complex population dynamics on a few microsecond time scales [@Brekke2009]. S-state excitation, on the other hand, showed no sign of such effects. This is consistent with previous observations that $s$-states ionize on much slower time scales than $d$-states do [@Li2005].
In order to study how Rydberg excitation is modified when resonant microwave fields are applied, we use an HP 83640A synthesized sweeper with a microwave horn to beam the microwaves into the vacuum chamber through a glass window. Since Rydberg atoms have very strong coupling between nearest-neighbor transitions, the states can be easily brought into the strongly coupled regime with only $\sim$0.1 mW reaching the atoms. The resulting Autler-Townes splitting can be seen in Fig. \[fig:acshift\]b, and is well described elsewhere [@Autler1955]. We have examined the excitation and stimulated emission rates with microwave coupling, and seen that the microwave coupling reduces the effective rates by $\sim$2, consistent with our expectations.
The basic experiment, then, consists of studying the Rydberg population dynamics as a function of density and laser intensity, with and without microwave dressing of the Rydberg states. To interpret our results, we use scaling arguments presented in the next section.
Interaction Theory and Scaling {#sec:scaling}
==============================
In order to develop expectations for the scaling of Rydberg atom production in a strongly interacting gas, we use the idea of collective excitation [@Heidemann2007], where the effective Rabi frequency is enhanced by the number of atoms in the blockaded region according to $$\Omega_{2b} = \sqrt{N_b} \Omega_2 = \sqrt{\eta \frac{4 \pi}{3} r_b^3 } \Omega_2
\label{eqn:effexc}$$ where $N_b$ is the number of atoms in the blockaded region, which is equal to the blockade volume $4 \pi r_b^3/3 $ times the ground state density $\eta$. The resulting scaling in a regime where the enhanced Rabi frequency is much greater than the linewidth of the transition $\gamma$ was recently developed [@Low2009]. However, it is expected that in our sample the effective enhanced Rabi frequency will remain smaller than the transition linewidth. In this case, to be blockaded the atom-atom interaction energy must be comparable to or greater than the transition linewidth. For microwave coupled interactions this would give: $$\frac{C_3 \beta}{r_b^3} = \gamma.
\label{eqn:uscalingw}$$ The factor $C_3\beta$ accounts for the strength and angular dependence of the interaction, as will be discussed further below. In the limit of strong blockade, the number of excitations contained in the extended sample volume $V$ will then be $$N_s = \frac{V}{\frac{4 \pi}{3} r_b^3} = \frac{3V \gamma}{{4 \pi}C_3 \beta},
\label{eqn:sscalingw}$$ This is independent of the density or excitation rate.
In the absence of blockade, there are potentially $N_s=\eta V$ Rydberg atoms that can be excited. To generalize the scaling laws for the transition between the realm where there is no blockade ($N_s \propto \eta$) and full blockade ($N_s$ constant), we use the following interpolation: $$N_s=\frac{\eta V}{\sqrt{1+( \frac{3V \gamma}{{4 \pi}C_3 \beta})^2}}.
\label{eqn:sscalingg}$$ Of course, other interpolation functions could be chosen. This particular function was chosen for its simplicity. Several other functions with the appropriate limiting dependence were tried, and resulted in less than a 20% change in the extracted parameters.
The relationship between the actual number of Rydberg atoms created in steady-state and the number of available excitation sites depends upon the relative rates of excitation and stimulated emission. The steady-state number of Rydberg atoms is given by: $$N_r= \frac{R_{2b}}{2 R_{2b} +R_{3}} N_s.
\label{eqn:nscalingg}$$ In the limit of strong blockade, the effective excitation rate $R_{2b}$ is collectively enhanced by the number of atoms in a blockade volume, and so can similarly be generalized as $$R_{2b} = \sqrt{1+\left(\frac{\eta {4 \pi} C_3 \beta}{{3}\gamma}\right)^2} R_2,
\label{eqn:effrateg}$$ so Eqn. \[eqn:nscalingg\] can be written explicitly as $$N_r= \frac{R_{2} V \eta}{2 R_{2} \sqrt{1+\left(\frac{\eta {4 \pi} C_3 \beta}{{3}\gamma}\right)^2} +R_{3}}.
\label{eqn:nscalinggf}$$
With this scaling relation, the number of Rydberg atoms expected can be determined for a particular set of [$\eta$, $R_2$, $R_3$, V, and $C_3 \beta$]{}, giving a means to quantitatively examine the number of Rydberg atoms created as a function of several experimental parameters. It is also worthwhile to consider the scaling of the change in Rydberg atom number, ${dN_r}/{dt}$, at small times when the Rydberg state has minimal population. Following the line of thought described above, $$\frac{dN_r}{dt} = N_s R_{2b},$$ which in the most general case can be written $$\frac{dN_r}{dt} = \frac{\eta V}{\sqrt{1+\left(\frac{\eta{4 \pi} C_3 \beta}{3\gamma}\right)^2}} \frac{2\pi\Omega_{2b}^2}{\sqrt{\gamma^2 +2 \Omega_{2b}^2}},
\label{eqn:dNdtscalingg}$$ As already discussed, for the densities available in our experiment $\gamma \gg \Omega_{2b}$. This results in the simplification of Eqn. \[eqn:dNdtscalingg\] to $$\frac{dN_r}{dt} = \frac{2\pi \eta V \Omega_2^2}{\gamma},
\label{eqn:dNdtscalingw}$$ which would scale linearly in both the ground state density and the 480 nm power. This is a different scaling than the regime where the enhanced Rabi frequency is larger than the transition linewidth, where a non-linear scaling is seen[@Low2009].
We now calculate the atom-atom interaction strength for use in analyzing our results. When the microwaves are off, the Rydberg atoms interact via van der Waals forces, which have been considered elsewhere [@Walker2008]. The interaction energy is given by $$U_6(R)_{vdW}=\frac{C_6}{r^6},$$ with relevant numbers given in Table \[table:strengths\].
For microwave dressed Rydberg states, the atoms acquire oscillating dipole moments. The time-averaged force between the dipoles is non-zero and gives an effective interaction. Neglecting spin, dressed $s$-state atoms would experience a dipole-dipole interaction of the classic form $$U_3={C_3\over r^3}P_2(\cos\theta)$$ where $\theta$ is the angle between the interatomic axis and the polarization of the microwave field, and $P_2$ is a Legendre polynomial. Including the degeneracies associated with spin modifies this simple picture. In the Appendix we consider the case of dressing on an $s_{1/2}\rightarrow p_{1/2}$ transition, as we have done in this experiment. The results are summarized in Fig. \[fig:channels\]. There are now four different interaction potentials, with corresponding eigenstates that are superpositions of the four possible $\ket{m_1m_2}$ combinations of angular momentum states of the pair of atoms. Two of the potentials have strong angular dependences, but the other two are isotropic, and one of them has zero dipole-dipole interaction. In general, we express the angular averaged strength of the interaction via the parameter $\beta$ introduced in Eq. \[eqn:uscalingw\].
We have used the calculated interaction strengths to estimate the blockade radii and the number of atoms within a blockade sphere for various states (Table \[table:strengths\]). Simulations suggest that a reasonable accounting for the blockade effects of the angular dependence of the microwave-induced interaction can made by taking $\beta=1/9$ for the three interaction curves of Fig. \[fig:channels\], and assuming a van der Waals interaction for the $\beta=0$ curve. For our densities, we do not expect to see suppression at the $30s$ even with microwaves, while the $60s$ state would show strong suppression with either van der Waals or microwave coupled interactions. The $47s$ level presents a compromise where the difference between the two cases is more pronounced, and hence the $47s$ state is used in the experiments described here.
------- ------------------- ----------- ----------- ------------------- ----------- -----------
State $C_6$ $r_b$ $N_b$ $C_3\beta$ $r_b$ $N_b$
$30s$ 30 1.3 0.10 76 2.3 0.43
$47s$ $7.1 \times 10^3$ 3.3 1.6 $530$ 4.1 3.1
$60s$ $1.3 \times 10^5$ 5.3 6.8 $1.4 \times 10^3$ 5.7 8.5
------- ------------------- ----------- ----------- ------------------- ----------- -----------
: The interaction strengths $C_6$(MHz $\mu m^6$) and $C_3\beta$(MHz $\mu m^3$), blockade radius $r_b$ ($\mu$m), and blockaded number $N_b$ for various Rydberg levels accessible to our experiment. These values are shown for both van der Waals interactions and microwave coupled interactions. The transition linewidth is assumed to be 6 MHz. []{data-label="table:strengths"}
![The microwave-induced portion of the interaction energy as a function of the angle between the two dipoles for the possible angular momentum states for microwave coupling on ${s_{1/2}} \rightarrow {p_{1/2}}$. A state exists with zero microwave-induced interaction energy, which therefore experiences only weaker van der Waals interactions. The ${s_{1/2}} \rightarrow {p_{3/2}}$ coupling does not have this zero.[]{data-label="fig:channels"}](channels.eps){width="3.2"}
\[t\] 
Studies of Suppression {#sec:suppression}
======================
We now turn to our experimental studies of suppression in the presence of resonant microwaves.
Using transients such as in Fig. \[fig:Spulse\], we study Rydberg atom production as the density of atoms is changed. We reduce the density from its maximum value by placing an iris in the MOT trapping beams, thus reducing the MOT loading rate [@Brekke2009]. This technique enables variation of the number of atoms without significantly altering the volume of the atom cloud. If no microwaves are applied, van der Waals interactions dominate. From the considerations of the previous section (Table \[table:strengths\]), less than 2 atoms are present in each blockaded volume at maximum density, and so minimal suppression is expected.
Figure \[fig:allNdep\]a shows the number of Rydberg atoms versus the ground state density, with no applied microwaves. During this series of scans, the excitation and stimulated emission rates were maintained at their maximum value: $R_2=0.6\times10^6$/s and $R_3=0.3\times10^6$/s. A linear dependence is observed, supporting the estimate that atom-atom interactions do not significantly effect the excitation in this regime.
The situation changes as microwaves are applied, increasing the Rydberg-Rydberg interactions. Figure \[fig:allNdep\]b shows the number of Rydberg atoms versus the ground state density, while the excitation and stimulated emission rates were again maintained at their maximum value, this time with microwaves on resonance to the $47p_{1/2}$ state. At high ground state densities, the number of Rydberg atoms produced ceases to grow linearly. This deviation from a linear dependence suggests that the increased interactions between atoms have started to suppress further Rydberg excitation as the density is increased. The slope at low density gives the effective excitation volume, while the density at which saturation begins determines the interaction strength. We fit the data using Eqn. \[eqn:nscalinggf\], and obtain an excitation volume of $2.1(4)\times 10^4$ $\mu$m$^3$ and an effective interaction strength $C_3 \beta$ of 530(90) MHz $\mu$m$^3$. Both the deduced interaction strength and the excitation volume are consistent with expectations from Sect. \[sec:expt\] and Sect. \[sec:scaling\].
A prediction of the scaling model in Section \[sec:scaling\] is that under our conditions the slope of the Rydberg excitation curves ${dN_r}/{dt}$ should be proportional to the atom density for both the van der Waals and microwave interaction cases. Indeed, for both the van der Waals and microwave coupled cases a linear dependence on the MOT density is observed, consistent with expectations for the regime where the transition linewidth is significantly larger than the effective excitation rate. If higher densities, and hence higher effective excitation rates, are achieved in the future, it will be interesting to further examine this dependence.
As a second means of studying suppression of Rydberg atom excitation, we vary the 480 nm laser power. The data, both with and without microwave excitation, are shown in Fig. \[fig:microidepscans\]. Even in the absence of interactions, the power dependence is nonlinear due to the competition between excitation and de-excitation (see Eq. \[eqn:nscalingg\]). The effect of blockade is to reduce the laser power at which the competition becomes apparent. From fitting the data to Eq. \[eqn:nscalinggf\] we obtain an effective interaction strength $C_3 \beta$ of 370(100) MHz $\mu$m$^3$ and a volume of $1.5(4)\times 10^4$ $\mu$m$^3$. The deduced interaction strength is somewhat smaller (1.5 standard deviations) from our expectations and the value deduced from the density dependence data.
![An examination of the Rydberg atom population with microwaves coupling the ${47s}$ and ${47p_{1/2}}$ states. Rydberg atom number is plotted versus 480 nm laser power, both with and without microwaves to allow observation of the difference which increased interactions provide. The microwave coupled data is fit using Equation \[eqn:nscalinggf\], giving $C_3 \beta$ = 370(100) MHz $\mu$m$^3$.[]{data-label="fig:microidepscans"}](NrvsPblueboth.eps){width="3.5"}
Conclusions
===========
We have shown that the interactions between Rydberg atoms are increased by dressing them with resonant microwave fields, essentially giving them oscillating electric dipole moments. In our experiment, the enhanced interactions were studied by measuring the suppression of Rydberg excitation due to the blockade effect. We derived scaling relations valid for our experimental parameters that accounted well for the observed excitation suppression and its dependence on density and laser intensity. The inferred interaction strengths were consistent with expectations.
Theoretically, we have shown that the Zeeman degeneracy of the dressed states alters the angular dependence of the Rydberg-Rydberg interactions. In particular, we predict that for the $s_{1/2}\rightarrow p_{1/2}$ dressing used in our experiment, there is a particular superposition of Rydberg states that is non-interacting for all angles. This state, which still experiences van der Waals interactions, might be more apparent in Rydberg EIT experiments, or at smaller principle quantum numbers where the van der Waals interactions are smaller relative to the microwave-dressed dipole-dipole interactions.
Interesting future experiments would be to repeat the current study using $s_{1/2}\rightarrow p_{3/2}$ dressing, where we predict that there is no non-interacting state. In addition, our relatively broad excitation line widths of 6 MHz resulted in clearly observable but modest differences in the excitation under conditions of microwave dressing as opposed to van der Waals interactions. With narrower excitation lines, and going to smaller principle quantum numbers, the effects should be greatly enhanced. This suggests that it might be possible to implement asymmetric blockade schemes [@Saffman2009b] using time-dependent microwave fields to alternately produce strong or weak blockade.
Finally, we note that this work, as with our previous experiments and much recent work on quantum manipulation experiments using Rydberg atoms, uses photons for Rydberg detection. It is interesting to contrast this trend with decades of experiments using ionization detection [@Gallagher1994].
This research was supported by the National Science Foundation.
Dipole-dipole interactions in dressed microwave fields
======================================================
We consider the interactions between Rydberg atoms dressed in resonant microwave fields. For concreteness, we consider $\hat{z}$-polarized microwaves dressing an $s_{1/2}$ level (with states denoted $\ket{sm_s}$) and a $p_{1/2}$ level ($\ket{pm_p}$), but it is not difficult to extend the results to other situations. We shall see that the Zeeman degeneracy of the levels leads to unusual angular dependences and coupling strengths.
We assume that the microwave Rabi frequency is large compared to the dipole-dipole coupling, optical Rabi frequencies, and to Zeeman shifts. Then, to first order, the atoms are in dressed Autler-Townes eigenstates that are linear combinations of the bare states $\ket{sm}$ and $\ket{pm}$: $$\ket{m}^\pm=\left(\ket{sm}\pm(-1)^{m-1/2}\ket{pm}e^{-i\omega t}\right)/\sqrt{2}$$ with the $\pm$ depending on which of the two dressed states are being excited by the light. For each atom, there are two pairs of dressed states, for $m=\pm1/2$. Since the microwave coupling is assumed to be the strongest in the system, dipole-dipole interactions will to first order only cause transitions between the Zeeman-degenerate states within a given Autler-Townes manifold $\ket{m}^+$ or $\ket{m}^-$.
We assume that two such dressed atoms are a distance $R$ apart, with their interatomic axis oriented at an angle $\theta$ with respect to the microwave polarization. The dipole-dipole interaction is $$V={e^2\over r^3}(x'_Ax'_B+y'_Ay'_B-2z'_Az'_B)$$ where, for example, $x'_A$ is the $x$-coordinate of the electron position operator on atom A, measured with respect to its nucleus, with the prime denoting that the operator is written in a coordinate system whose $z'$-axis is oriented along the interatomic axis.
Considering the matrix elements of $V$ within a single dressed-state manifold, we get matrix elements like $$\begin{aligned}
\bra{m_1 m_2}&&x'_Ax'_B\ket{m_3 m_4}=\\\nonumber
&&{1\over 4}\left(\bra{sm_1}x'_A\ket{pm_3}\bra{pm_2}x'_B\ket{sm_4}+\right.\\\nonumber
&&\left.\bra{pm_1}x'_A\ket{sm_3}\bra{sm_2}x'_B\ket{pm_4}\right)+\ldots\end{aligned}$$ The ellipses denote terms such as $$\bra{sm_1sm_2}x'_Ax'_B\ket{pm_3pm_4}e^{-2i\omega t}$$ that rapidly oscillate in sign and time-average to zero.
For the $s_{1/2}\rightarrow p_{1/2}$ case considered, the explicit matrix for $V$, with the basis states ordered as $\ket{1/2,1/2},\ket{1/2,-1/2},\ket{-1/2,1/2},\ket{-1/2,-1/2}$, is $$V={C_3\over 9 r^3}
\left(
\begin{array}{cccc}
-P_2 & 0 & 0 & 1-P_2 \\
0 & -P_2 & P_2 & 0 \\
0 & P_2 & -P_2 & 0 \\
1-P_2 & 0 & 0 & -P_2
\end{array}
\right)$$ where $C_3=\left(e R_{n_ss}^{n_pp}\right)^2$ in the notation of Ref. [@Walker2008], and the Legendre polynomial is $P_{2}=(1+3\cos(2\theta))/4$. The matrix divides into two non-interacting subspaces with $|M|=|m_A+m_B|=0$ or 1.
The $M=0$ subspace, with all four elements proportional to $P_2$, has a zero eigenvalue; there is no energy shift from the dipole-dipole interaction for this state. The other three eigenvalues of $V$ are non-zero and are shown in Fig. \[fig:channels\]. The explicit energies are $$\begin{aligned}
V={C_3\over 9 r^3}\left\{0,-1,-2 P_2,1-2 P_2\right\}\end{aligned}$$ Note that we have neglected van der Waals interactions in this calculation, so the zero-eigenvalue state will generally still have those interactions.
We find that the case of an $s_{1/2}$ level dressed with a $p_{3/2}$ level has no zero eigenvalue case. The energies are $$\begin{aligned}
V&=&{C_3\over 9 r^3}\left\{\frac{3}{8} \left(2-7 P_2\right),-\frac{3}{8} \left(7 P_2-1\right)\right.\nonumber \\&&\hspace{0.25 in}\left.-\frac{3}{8} \left(P_2+1\right),-\frac{3}{8} \left(P_2+2\right)\right\}.\end{aligned}$$
|
{
"pile_set_name": "ArXiv"
}
|
---
address: |
Physics Department, Columbia University\
538 West 120th Street, New York, NY 10027, U.S.A.\
E-mail: [email protected]
author:
- Dénes Molnár
title: The gluon plasma at RHIC
---
Introduction {#Section:intro}
============
At present we have a very limited understanding of the properties of the partonic environment created in heavy-ion collisions at RHIC. Predictions for the density of the gluons produced vary by a factor of five depending on the model considered. Elliptic flow, $v_2(p_\perp)=\langle \cos(2\phi)\rangle_{p_\perp}$, the differential second moment of the azimuthal momentum distribution, and the high-$p_\perp$ suppression of the particle spectra have been the subject of increasing interest[@hydro; @Zhang:1999rs; @molnar_v2; @gvw; @v2_cascade] because they provide important constraints on the density and effective energy loss of partons.
The origin of the remarkable saturation of elliptic flow $v_2(p_\perp) \to 0.2$ above $p_\perp \sim 2$ GeV reported by STAR at Quark Matter 2001[@STARv2] is an open question for theory. Calculations based on inelastic parton energy loss[@gvw] do predict saturation or decreasing $v_2$ at high $p_\perp$. These calculations are valid for high $p_\perp$, where collective transverse flow from lower-$p_\perp$ partons can be neglected and Eikonal dynamics is applicable. However, a constant spatial anisotropy was assumed throughout the evolution, while in reality, it decreases and probably even changes sign. This is likely to reduce the generated elliptic flow much below[@gvw] the preliminary data.
Ideal hydrodynamics[@hydro], the simplest theoretical framework to study elliptic flow, agrees remarkably well with the measured elliptic flow data[@STARv2] up to transverse momenta $\sim 1.5$ GeV$/c$. However, it fails to saturate at high $p_\perp>2$ GeV as does the preliminary data.
A theoretical problem with ideal hydrodynamics is that it assumes local equilibrium throughout the whole evolution. This idealization is marginal for conditions encountered in heavy ion collisions[@nonequil]. Covariant Boltzmann transport theory provides a convenient framework for nonequilibrium dynamics that depends on the local mean free path $\lambda(x) \equiv 1/\sigma n(x)$.
Parton cascade simulations[@Zhang:1999rs; @molnar_v2] show on the other hand, that the initial parton density based on HIJING[@Gyulassy:1994ew] is too low to produce the observed elliptic flow unless the pQCD cross sections are artificially enhanced by a factor $\sim 2-3$. However, gluon saturation models[@Eskola:2000fc] predict up to five times higher initial densities, and these may be dense enough to generate the observed collective flow even with pQCD elastic cross sections.
In this study, we explore the dependence of elliptic flow and the high-$p_\perp$ suppression of the particle spectra on the initial density and the elastic $gg$ cross section. Though parton cascades lack at present covariant inelastic energy loss, elastic energy loss alone may account for the observed high-$p_\perp$ azimuthal flow pattern as long as the number of elastic collisions is large enough[@molnar_v2].
Covariant parton transport theory {#Section:transport_theory}
=================================
We consider here, as in [@molnar_v2; @nonequil; @Yang; @Zhang:1998ej], the simplest nonlinear form of Lorentz-covariant Boltzmann transport theory in which the on-shell phase space density, evolves with an elastic $2\to 2$ rate. We solve the transport equation via the MPC algorithm[@nonequil], which maintains Lorentz covariance using the parton subdivision technique[@Yang; @Zhang:1998tj]. See Ref. [@molnar_v2] and references therein for details.
For a given nuclear geometry and formation time, the solutions of the nonlinear transport equation has been shown[@molnar_v2] to depend mainly on the [*transport opacity*]{} $\chi\equiv\int dz \sigma_t\rho_g
=N\langle\sin^2 \theta_{cm}\rangle$ and the impact parameter $b$. Here $\sigma_t$ is the elastic [*transport*]{} cross section, $N$ is the average number of collisions per parton during the whole evolution, while $\theta_{cm}$ is the collision deflection angle in the c.m. frame. To good accuracy the transport opacity factorizes[@molnar_v2] as $\chi = C(b) \sigma_t(T_0) dN_g/d\eta$.
We label our results by the transport opacity $\chi$ and impact parameter $b$. Furthermore, we quote our transport opacities [*relative*]{} to that for the pQCD minijet gluons predicted by HIJING \[$dN/d\eta = 210$, $\sigma_{gg\to gg} \approx 3$ mb $\Rightarrow \sigma_t \approx 1$ mb, $\chi_{b=0} \approx 0.3$\].
\[Figure:v2\]
\[Figure:pt\]
Numerical results {#Section:glue_results}
=================
The initial condition was a longitudinally boost invariant thermal Bjorken tube at proper time $\tau_0=0.1$ fm/$c$ with uniform pseudo-rapidity distribution between $|\eta| < 5$ and transverse density distribution proportional to the binary collision distribution for the two gold nuclei. Based on HIJING, the pQCD jet cross section was normalized to yield $dN_g/d\eta=210$ in central collisions and the initial temperature was chosen to be $T_0=700$ MeV.
Two different hadronization schemes were applied. One is based on local parton-hadron duality, where each gluon is assumed to convert to a pion[@Eskola:2000fc]. The other hadronization prescription is independent fragmentation, where we considered only the $g\to \pi^{\pm}$ channel. See Ref. [@molnar_v2] for details.
Fig. \[Figure:v2\] shows the impact-parameter-averaged elliptic flow as a function of $p_\perp$. With increasing $p_\perp$, elliptic flow increases until $p_\perp\sim 1.5-2$ GeV, where it saturates. To reproduce the preliminary STAR data, $\sim 80$ times more opaque initial gluon plasma is needed than the pQCD prediction from HIJING. Surprisingly, the results show no sensitivity to the applied hadronization prescription.
Fig. 2 shows that such large transport opacities are also consistent with the preliminary charged hadron spectra measured by STAR[@Dunlop:2001vh]. Hadronization via parton-hadron duality yields too little suppression at high $p_\perp$ because it only incorporates quenching due to elastic energy loss. However, with the additional quenching due to independent fragmentation, the parton cascade results approach the preliminary STAR data.
Acknowledgements {#acknowledgements .unnumbered}
================
We acknowledge the Parallel Distributed Systems Facility at the National Energy Research Scientific Computing Center for providing computing resources.
This work was supported by the Director, Office of Energy Research, Division of Nuclear Physics of the Office of High Energy and Nuclear Physics of the U.S. Department of Energy under contract No. DE-FG-02-93ER-40764.
[99]{}
P. F. Kolb, P. Huovinen, U. Heinz and H. Heiselberg, Phys. Lett. B [**500**]{}, 232 (2001) \[hep-ph/0012137\]; P. Huovinen, P. F. Kolb, U. Heinz, P. V. Ruuskanen and S. A. Voloshin, Phys. Lett. B [**503**]{}, 58 (2001) \[hep-ph/0101136\].
B. Zhang, M. Gyulassy and C. M. Ko, Phys. Lett. [**B455**]{}, 45 (1999) \[nucl-th/9902016\].
D. Molnar and M. Gyulassy, Nucl. Phys. [**A**]{} in press \[nucl-th/0104073\]; Nucl. Phys. [**A661**]{}, 236c (1999) \[nucl-th/9907090\]; and nucl-th/0102031.
X.-N. Wang, nucl-th/0009019; M. Gyulassy, I. Vitev and X.-N. Wang, Phys. Rev. Lett. [**86**]{}, 2537 (2001) \[nucl-th/0012092\]; M. Gyulassy, I. Vitev, X.-N. Wang, and P. Huovinen, nucl-th/0109063.
H. Sorge, Nucl. Phys. [**A661**]{}, 577 (1999) \[nucl-th/9906051\]; M. Bleicher and H. Stocker, hep-ph/0006147.
K. H. Ackermann [*et al.*]{} \[STAR Collaboration\], Phys. Rev. Lett. [**86**]{}, 402 (2001) \[nucl-ex/0009011\]; R. J. Snellings \[STAR Collaboration\], nucl-ex/0104006.
D. Molnar and M. Gyulassy, Phys. Rev. [**C 62**]{}, 054907 (2000) \[nucl-th/0005051\]. Parton cascade code MPC 1.0.6 used in the present study can be downloaded from http://www-cunuke.phys.columbia.edu/OSCAR
M. Gyulassy and X. Wang, Comput. Phys. Commun. [**83**]{}, (1994) 307 \[nucl-th/9502021\].
K. J. Eskola, K. Kajantie, P. V. Ruuskanen, and K. Tuominen, Nucl. Phys. [**B570**]{}, 379 (2000) \[hep-ph/9909456\].
Y. Pang, RHIC 96 Summer Study, CU-TP-815 preprint (unpublished); Generic Cascade Program (GCP) documentation available at WWW site http://www-cunuke.phys.columbia.edu/OSCAR
B. Zhang, Comput. Phys. Commun. [**109**]{}, 193 (1998) \[nucl-th/9709009\].
B. Zhang, M. Gyulassy, and Y. Pang, Phys. Rev. C [**58**]{}, (1998) 1175 \[nucl-th/9801037\].
J. C. Dunlop \[STAR Collaboration\], CERN-ALICE-PUB-2001-011
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'Conventionally, model-based reinforcement learning (MBRL) aims to learn a global model for the dynamics of the environment. A good model can potentially enable planning algorithms to generate a large variety of behaviors and solve diverse tasks. However, learning an accurate model for complex dynamical systems is difficult, and even then, the model might not generalize well outside the distribution of states on which it was trained. In this work, we combine model-based learning with model-free learning of primitives that make model-based planning easy. To that end, we aim to answer the question: how can we discover skills whose outcomes are easy to predict? We propose an unsupervised learning algorithm, Dynamics-Aware Discovery of Skills (DADS), which simultaneously discovers *predictable* behaviors and learns their dynamics. Our method can leverage continuous skill spaces, theoretically, allowing us to learn infinitely many behaviors even for high-dimensional state-spaces. We demonstrate that *zero-shot planning* in the learned latent space significantly outperforms standard MBRL and model-free goal-conditioned RL, can handle sparse-reward tasks, and substantially improves over prior hierarchical RL methods for unsupervised skill discovery. We have open-sourced our implementation at: <https://github.com/google-research/dads>'
author:
- |
Archit Sharma[^1], Shixiang Gu, Sergey Levine, Vikash Kumar, Karol Hausman\
Google Brain\
`{architsh,shanegu,slevine,vikashplus,karolhausman}@google.com`\
bibliography:
- 'iclr2020\_conference.bib'
title: 'Dynamics-Aware Unsupervised Discovery of Skills'
---
{width="\textwidth"}
Introduction
============
Deep reinforcement learning (RL) enables autonomous learning of diverse and complex tasks with rich sensory inputs, temporally extended goals, and challenging dynamics, such as discrete game-playing domains [@mnih2013playing; @silver2016mastering], and continuous control domains including locomotion [@schulman2015trust; @heess2017emergence] and manipulation [@rajeswaran2017learning; @kalashnikov2018qt; @gu2017deep]. Most of the deep RL approaches learn a Q-function or a policy that are directly optimized for the training task, which limits their generalization to new scenarios. In contrast, MBRL methods [@li2004iterative; @deisenroth2011pilco; @watter2015embed] can acquire dynamics models that may be utilized to perform unseen tasks at test time. While this capability has been demonstrated in some of the recent works [@levine2016end; @nagabandi2018neural; @chua2018deep; @kurutach2018model; @ha2018recurrent], learning an accurate global model that works for all state-action pairs can be exceedingly challenging, especially for high-dimensional system with complex and discontinuous dynamics. The problem is further exacerbated as the learned global model has limited generalization outside of the state distribution it was trained on and exploring the whole state space is generally infeasible. Can we retain the flexibility of model-based RL, while using model-free RL to acquire proficient low-level behaviors under complex dynamics?
While learning a global dynamics model that captures all the different behaviors for the entire state-space can be extremely challenging, learning a model for a specific behavior that acts only in a small part of the state-space can be much easier. For example, consider learning a model for dynamics of all gaits of a quadruped versus a model which only works for a specific gait. If we can learn many such behaviors and their corresponding dynamics, we can leverage model-predictive control to plan in the *behavior space*, as opposed to planning in the action space. The question then becomes: how do we acquire such behaviors, considering that behaviors could be random and unpredictable? To this end, we propose *Dynamics-Aware Discovery of Skills* (DADS), an unsupervised RL framework for learning low-level skills using model-free RL with the explicit aim of making model-based control easy. Skills obtained using DADS are directly optimized for *predictability*, providing a better representation on top of which predictive models can be learned. Crucially, the skills do not require any supervision to learn, and are acquired entirely through autonomous exploration. This means that the repertoire of skills and their predictive model are learned before the agent has been tasked with any goal or reward function. When a task is provided at test-time, the agent utilizes the previously learned skills and model to immediately perform the task without any further training.
The key contribution of our work is an unsupervised reinforcement learning algorithm, DADS, grounded in mutual-information-based exploration. We demonstrate that our objective can embed learned primitives in continuous spaces, which allows us to learn a large, diverse set of skills. Crucially, our algorithm also learns to model the dynamics of the skills, which enables the use of model-based planning algorithms for downstream tasks. We adapt the conventional model predictive control algorithms to plan in the space of primitives, and demonstrate that we can compose the learned primitives to solve downstream tasks without any additional training.
Preliminaries
=============
Mutual information can been used as an objective to encourage exploration in reinforcement learning [@DBLP:journals/corr/HouthooftCDSTA16; @mohamed2015variational]. According to its definition, $\mathcal{I}(X;Y) = \mathcal{H}(X) - \mathcal{H}(X \mid Y)$, maximizing mutual information $\mathcal{I}$ with respect to $Y$ amounts to maximizing the entropy $\mathcal{H}$ of $X$ while minimizing the conditional entropy $\mathcal{H}(X \mid Y)$. In the context of RL, $X$ is usually a function of the state and $Y$ a function of actions. Maximizing this objective encourages the state entropy to be high, making the underlying policy to be exploratory. Recently, multiple works [@eysenbach2018diversity; @gregor2016variational; @achiam2018variational] apply this idea to learn diverse skills which maximally cover the state space.
To leverage planning-based control, MBRL estimates the true dynamics of the environment by learning a model $\hat{p}(s' \mid s, a)$. This allows it to predict a trajectory of states $\hat{\tau}_H = (s_t ,\hat{s}_{t+1}, \ldots \hat{s}_{t+H})$ resulting from a sequence of actions without any additional interaction with the environment. While model-based RL methods have been demonstrated to be sample efficient compared to their model-free counterparts, learning an effective model for the whole state-space is challenging. An open-problem in model-based RL is to incorporate temporal abstraction in model-based control, to enable high-level planning and move-away from planning at the granular level of actions. These seemingly unrelated ideas can be combined into a single optimization scheme, where we first discover skills (and their models) without any extrinsic reward and then compose these skills to optimize for the task defined at test time using model-based planning. At train time, we assume a Markov Decision Process (MDP) $\mathcal{M}_1 \equiv (\mathcal{S}, \mathcal{A}, p)$. The state space $\mathcal{S}$ and action space $\mathcal{A}$ are assumed to be continuous, and the $\mathcal{A}$ bounded. We assume the transition dynamics $p$ to be stochastic, such that $p: \mathcal{S} \times \mathcal{A} \times \mathcal{S} \mapsto [0, \infty)$. We learn a skill-conditioned policy $\pi(a \mid s, z)$, where the skills $z$ belongs to the space $\mathcal{Z}$, detailed in Section \[sec:DADS\]. We assume that the skills are sampled from a prior $p(z)$ over $\mathcal{Z}$. We simultaneously learn a skill-conditioned transition function $q(s' \mid s, z)$, coined as *skill-dynamics*, which predicts the transition to the next state $s'$ from the current state $s$ for the skill $z$ under the given dynamics $p$. At test time, we assume an MDP $\mathcal{M}_2 \equiv (\mathcal{S}, \mathcal{A}, p, r)$, where $\mathcal{S}, \mathcal{A}, p$ match those defined in $\mathcal{M}_1$, and the reward function $r: \mathcal{S} \times \mathcal{A} \mapsto (-\infty, \infty)$. We plan in $\mathcal{Z}$ using $q(s' \mid s, z)$ to compose the learned skills $z$ for optimizing $r$ in $\mathcal{M}_2$, which we detail in Section \[sec:planning\].
Dynamics-Aware Discovery of Skills (DADS) {#sec:DADS}
=========================================
![The agent $\pi$ interacts with the environment to produce a transition $s \rightarrow s'$. Intrinsic reward is computed by computing the transition probability under $q$ for the current skill $z$, compared to random samples from the prior $p(z)$. The agent maximizes the intrinsic reward computed for a batch of episodes, while $q$ maximizes the log-probability of the actual transitions of $(s, z) \rightarrow s'$.[]{data-label="Fig:skill_discovery_summary"}](images/skill_learning.png){width=".9\textwidth"}
\[algorithm:usl\] Initialize $\pi, q_\phi$
We use the information theoretic paradigm of mutual information to obtain our unsupervised skill discovery algorithm. In particular, we propose to maximize the mutual information between the next state $s'$ and current skill $z$ conditioned on the current state $s$. $$\begin{aligned}
\label{eq:mi_objective}
\mathcal{I}(s';z \mid s) &= \mathcal{H}(z\mid s) - \mathcal{H}(z \mid s', s)\\
\label{eq:mi_mbrl_decomposition}
&= \mathcal{H}(s' \mid s) - \mathcal{H}(s' \mid s, z)\end{aligned}$$ Mutual information in Equation \[eq:mi\_objective\] quantifies how much can be known about $s'$ given $z$ and $s$, or symmetrically, $z$ given the transition from $s \rightarrow s'$. From Equation \[eq:mi\_mbrl\_decomposition\], maximizing this objective corresponds to maximizing the diversity of transitions produced in the environment, that is denoted by the entropy $\mathcal{H}(s' \mid s)$, while making $z$ informative about the next state $s'$ by minimizing the entropy $\mathcal{H}(s' \mid s, z)$. Intuitively, skills $z$ can be interpreted as abstracted action sequences which are identifiable by the transitions generated in the environment (and not just by the current state). Thus, optimizing this mutual information can be understood as encoding a diverse set of skills in the latent space $\mathcal{Z}$, while making the transitions for a given $z \in \mathcal{Z}$ predictable. We use the entropy-decomposition in Equation \[eq:mi\_mbrl\_decomposition\] to connect this objective with model-based control.
We want to optimize the our skill-conditioned controller $\pi(a \mid s, z)$ such that the latent space ${z \sim p(z)}$ is maximally informative about the transitions $s \rightarrow s'$. Using the definition of conditional mutual information, we can rewrite Equation \[eq:mi\_mbrl\_decomposition\] as: $$\begin{aligned}
\mathcal{I}(s'; z \mid s) = \int p(z, s, s') \log\frac{p(s' \mid s, z)}{p(s' \mid s)} ds' ds dz\end{aligned}$$ We assume the following generative model: $p(z, s, s') = p(z) p(s \mid z) p(s' \mid s, z)$, where $p(z)$ is user specified prior over $\mathcal{Z}$, $p(s | z)$ denotes the stationary state-distribution induced by $\pi(a \mid s, z)$ for a skill $z$ and $p(s' \mid s, z)$ denotes the transition distribution under skill $z$. Note, $p(s' \mid s, z) = \int p(s' \mid s, a) \pi(a \mid s, z) da$ is intractable to compute because the underlying dynamics are unknown. However, we can variationally lower bound the objective as follows: $$\begin{aligned}
\mathcal{I}(s';z \mid s) &= \mathbb{E}_{z, s, s' \sim p} \Big[\log \frac{p(s' \mid s, z)}{p(s' \mid s)}\Big]\nonumber\\
&= \mathbb{E}_{z, s, s' \sim p}\Big[\log\frac{q_\phi(s' \mid s, z)}{p(s' \mid s)}\Big] + \mathbb{E}_{s, z \sim p}\Big[\mathcal{D}_{KL}(p(s' \mid s, z) \mid\mid q_\phi(s' \mid s, z))\Big]\label{eq:alternate_opt}\nonumber\\
&\geq \mathbb{E}_{z, s, s' \sim p}\Big[\log\frac{q_\phi(s' \mid s, z)}{p(s' \mid s)}\Big]\end{aligned}$$ where we have used the non-negativity of KL-divergence, that is $\mathcal{D}_{KL} \geq 0$. Note, skill-dynamics $q_\phi$ represents the variational approximation for the transition function $p(s' \mid s, z)$, which enables model-based control as described in Section \[sec:planning\]. Equation \[eq:alternate\_opt\] suggests an alternating optimization between $q_\phi$ and $\pi$, summarized in Algorithm \[algorithm:usl\]. In every iteration:\
(*Tighten variational lower bound*) We minimize $D_{KL}(p(s' \mid s, z) \mid\mid q_\phi(s' \mid s, z))$ with respect to the parameters $\phi$ on $z, s \sim p$ to tighten the lower bound. For general function approximators like neural networks, we can write the gradient for $\phi$ as follows: $$\begin{aligned}
\nabla_\phi \mathbb{E}_{s, z} [\mathcal{D}_{KL}(p(s' \mid s, z )\mid\mid q_\phi(s' \mid s, z))] &= \nabla_\phi \mathbb{E}_{z, s, s'} \Big[\log \frac{p(s' \mid s, z)}{q_\phi(s' \mid s, z)}\Big] \nonumber\\
&= -\mathbb{E}_{z, s, s'} \Big[ \nabla_\phi \log q_\phi(s' \mid s, z)\Big]\end{aligned}$$ which corresponds to maximizing the likelihood of the samples from $p$ under $q_\phi$.
(*Maximize approximate lower bound*) After fitting $q_\phi$, we can optimize $\pi$ to maximize $\mathbb{E}_{z, s, s'} [ \log q_\phi(s' \mid s, z) - \log p(s' \mid s)]$. Note, this is a reinforcement-learning style optimization with a reward function $\log q_\phi(s' \mid s, z) - \log p(s' \mid s)$. However, $\log p(s' \mid s)$ is intractable to compute, so we approximate the reward function for $\pi$: $$\begin{aligned}
r_z(s, a, s') = \log{\frac{q_\phi(s' \mid s, z)}{\sum_{i=1}^L q_\phi(s' \mid s, z_i)}} + \log{L}, \quad z_i \sim p(z). \label{eq:final_reward}\end{aligned}$$ The approximation is motivated as follows: $p(s' \mid s) = \int p(s' \mid s, z) p(z | s) dz \approx \int q_\phi(s' \mid s, z) p(z) dz \approx \frac{1}{L}\sum_{i=1}^L q_\phi(s' \mid s, z_i)$ for $z_i \sim p(z)$, where $L$ denotes the number of samples from the prior. We are using the marginal of variational approximation $q_\phi$ over the prior $p(z)$ to approximate the marginal distribution of transitions. We discuss this approximation in Appendix \[appendix:reward\_approximation\]. Note, the final reward function $r_z$ encourages the policy $\pi$ to produce transitions that are (a) predictable under $q_\phi$ (*predictability*) and (b) different from the transitions produced under $z_i \sim p(z)$ (*diversity*).
To generate samples from $p(z, s, s')$, we use the rollouts from the current policy $\pi$ for multiple samples $z \sim p(z)$ in an episodic setting for a fixed horizon $T$. We also introduce entropy regularization for $\pi(a \mid s, z)$, which encourages the policy to discover action-sequences with similar state-transitions and to be clustered under the same skill $z$, making the policy robust besides encouraging exploration [@haarnoja2018soft]. The use of entropy regularization can be justified from an information bottleneck perspective as discussed for Information Maximization algorithm in [@mohamed2015variational]. This is even more extensively discussed from the graphical model perspective in Appendix \[appendix:infobot\], which connects unsupervised skill discovery and information bottleneck literature, while also revealing the temporal nature of skills $z$. Details corresponding to implementation and hyperparameters are discussed in Appendix \[appendix:implementation\].
Planning using Skill Dynamics {#sec:planning}
=============================
![At test time, the planner executes simulates the transitions in environment using skill-dynamics $q$, and updates the distribution of plans according to the computed reward on the simulated trajectories. After a few updates to the plan, the first primitive is executed in the environment using the learned agent $\pi$.[]{data-label="Fig:planning_overview"}](images/model_predictive_control.png){width="\textwidth"}
\[algorithm:planning\] $s\gets s_0$ Initialize parameters $\mu_1, \ldots \mu_{H_P}$
Given the learned skills $\pi(a \mid s, z)$ and their respective skill-transition dynamics $q_\phi(s' \mid s, z)$, we can perform model-based planning in the latent space $\mathcal{Z}$ to optimize for a reward $r$ that is given to the agent at test time. Note, that this essentially allows us to perform zero-shot planning given the unsupervised pre-training procedure described in Section \[sec:DADS\].
In order to perform planning, we employ the model-predictive-control (MPC) paradigm [@garcia1989model], which in a standard setting generates a set of action plans $P_k = (a_{k,1}, \ldots a_{k,H}) \sim P$ for a planning horizon $H$. The MPC plans can be generated due to the fact that the planner is able to simulate the trajectory $\hat{\tau}_k = (s_{k, 1}, a_{k, 1} \ldots s_{k, H+1})$ assuming access to the transition dynamics $\hat{p}(s' \mid s, a)$. In addition, each plan computes the reward $r(\hat{\tau}_k)$ for its trajectory according to the reward function $r$ that is provided for the test-time task. Following the MPC principle, the planner selects the best plan according to the reward function $r$ and executes its first action $a_1$. The planning algorithm repeats this procedure for the next state iteratively until it achieves its goal.
We use a similar strategy to design an MPC planner to exploit previously-learned skill-transition dynamics $q_\phi(s' \mid s,z)$. Note that unlike conventional model-based RL, we generate a plan $P_k = (z_{k,1}, \ldots z_{k, H_P})$ in the latent space $\mathcal{Z}$ as opposed to the action space $\mathcal{A}$ that would be used by a standard planner. Since the primitives are temporally meaningful, it is beneficial to hold a primitive for a horizon $H_Z > 1$, unlike actions which are usually held for a single step. Thus, effectively, the planning horizon for our latent space planner is $H = H_P \times H_Z$, enabling longer-horizon planning using fewer primitives. Similar to the standard MPC setting, the latent space planner simulates the trajectory $\hat{\tau}_k = (s_{k, 1}, z_{k, 1}, a_{k, 1}, s_{k, 2}, z_{k, 2}, a_{k, 2}, \ldots s_{k, H+1})$ and computes the reward $r(\hat{\tau}_k)$. After a small number of trajectory samples, the planner selects the first latent action $z_1$ of the best plan, executes it for $H_Z$ steps in the environment, and the repeats the process until goal completion.
The latent planner $P$ maintains a distribution of latent plans, each of length $H_P$. Each element in the sequence represents the distribution of the primitive to be executed at that time step. For continuous spaces, each element of the sequence can be modelled using a normal distribution, $\mathcal{N}(\mu_1, \Sigma), \ldots \mathcal{N}(\mu_{H_P}, \Sigma)$. We refine the planning distributions for $R$ steps, using $K$ samples of latent plans $P_k$, and compute the $r_k$ for the simulated trajectory $\hat{\tau}_k$. The update for the parameters follows that in Model Predictive Path Integral (MPPI) controller [@williams2016aggressive]: $$\begin{aligned}
\mu_i = \sum_{k=1}^K\frac{\exp(\gamma r_k)}{\sum_{p=1}^K \exp(\gamma r_p)} z_{k, i} \quad \forall i = 1, \ldots H_P\end{aligned}$$ While we keep the covariance matrix of the distributions fixed, it is possible to update that as well as shown in [@williams2016aggressive]. We show an overview of the planning algorithm in Figure \[Fig:planning\_overview\], and provide more implementation details in Appendix \[appendix:implementation\].
Related Work
============
Central to our method is the concept of skill discovery via mutual information maximization. This principle, proposed in prior work that utilized purely model-free unsupervised RL methods [@daniel2012hierarchical; @florensa2017stochastic; @eysenbach2018diversity; @gregor2016variational; @warde2018unsupervised; @thomas2018disentangling], aims to learn diverse skills via a discriminability objective: a good set of skills is one where it is easy to distinguish the skills from each other, which means they perform distinct tasks and cover the space of possible behaviors. Building on this prior work, we distinguish our skills based on how they modify the original uncontrolled dynamics of the system. This simultaneously encourages the skills to be both *diverse* and *predictable*. We also demonstrate that constraining the skills to be predictable makes them more amenable for hierarchical composition and thus, more useful on downstream tasks. Another line of work that is conceptually close to our method copes with intrinsic motivation [@oudeyer2009intrinsic; @oudeyer2007intrinsic; @schmidhuber2010formal] which is used to drive the agent’s exploration. Examples of such works include empowerment [@klyubin2005empowerment; @mohamed2015variational], count-based exploration [@bellemare2016unifying; @oh2015action; @tang2017exploration; @NIPS2017_6851], information gain about agent’s dynamics [@StadieLA15] and forward-inverse dynamics models [@pathakICMl17curiosity]. While our method uses an information-theoretic objective that is similar to these approaches, it is used to learn a variety of skills that can be directly used for model-based planning, which is in contrast to learning a better exploration policy for a single skill. The skills discovered using our approach can also provide extended actions and temporal abstraction, which enable more efficient exploration for the agent to solve various tasks, reminiscent of hierarchical RL (HRL) approaches. This ranges from the classic option-critic architecture [@sutton1999between; @stolle2002learning; @perkins1999using] to some of the more recent work [@bacon2017option; @vezhnevets2017feudal; @nachum2018data; @hausman2018learning]. However, in contrast to end-to-end HRL approaches [@heess2016learning; @peng2017deeploco], we can leverage a stable, two-phase learning setup. The primitives learned through our method provide action and temporal abstraction, while planning with skill-dynamics enables hierarchical composition of these primitives, bypassing many problems of end-to-end HRL. In the second phase of our approach, we use the learned skill-transition dynamics models to perform model-based planning - an idea that has been explored numerous times in the literature. Model-based reinforcement learning has been traditionally approached with methods that are well-suited for low-data regimes such as Gaussian Processes [@rasmussen2003gaussian] showing significant data-efficiency gains over model-free approaches [@deisenroth2013gaussian; @kamthe2017data; @kocijan2004gaussian; @ko2007gaussian]. More recently, due to the challenges of applying these methods to high-dimensional state spaces, MBRL approaches employs Bayesian deep neural networks [@nagabandi2018neural; @chua2018deep; @gal2016improving; @fu2016one; @lenz2015deepmpc] to learn dynamics models. In our approach, we take advantage of the deep dynamics models that are conditioned on the skill being executed, simplifying the modelling problem. In addition, the skills themselves are being learned with the objective of being predictable, further assists with the learning of the dynamics model. There also have been multiple approaches addressing the planning component of MBRL including linear controllers for local models [@levine2016end; @kumar2016optimal; @chebotar2017combining], uncertainty-aware [@chua2018deep; @gal2016improving] or deterministic planners [@nagabandi2018neural] and stochastic optimization methods [@williams2016aggressive]. The main contribution of our work lies in discovering model-based skill primitives that can be further combined by a standard model-based planner, therefore we take advantage of an existing planning approach - Model Predictive Path Integral [@williams2016aggressive] that can leverage our pre-trained setting.
Experiments
===========
Through our experiments, we aim to demonstrate that: (a) DADS as a general purpose skill discovery algorithm can scale to high-dimensional problems; (b) discovered skills are amenable to hierarchical composition and; (c) not only is planning in the learned latent space feasible, but it is competitive to strong baselines. In Section \[sec:qualitative\], we provide visualizations and qualitative analysis of the skills learned using DADS. We demonstrate in Section \[sec:variance\_primitives\] and Section \[sec:hrl\_on\_skills\] that optimizing the primitives for predictability renders skills more amenable to temporal composition that can be used for Hierarchical RL.We benchmark against state-of-the-art model-based RL baseline in Section \[sec:mbrl\], and against goal-conditioned RL in Section \[sec:gcrl\].
Qualitative Analysis {#sec:qualitative}
--------------------
![Skills learned on different MuJoCo environments in the OpenAI gym. DADS can discover diverse skills without any extrinsic rewards, even for problems with high-dimensional state and action spaces.[]{data-label="Fig:skills"}](images/humanoid_fused.png){width="\textwidth"}
![Skills learned on different MuJoCo environments in the OpenAI gym. DADS can discover diverse skills without any extrinsic rewards, even for problems with high-dimensional state and action spaces.[]{data-label="Fig:skills"}](images/ant_skills.png){width="\textwidth"}
![Skills learned on different MuJoCo environments in the OpenAI gym. DADS can discover diverse skills without any extrinsic rewards, even for problems with high-dimensional state and action spaces.[]{data-label="Fig:skills"}](images/walker_fused.png){width="\textwidth"}
In this section, we provide a qualitative discussion of the unsupervised skills learned using DADS. We use the MuJoCo environments [@todorov2012mujoco] from the OpenAI gym as our test-bed [@DBLP:journals/corr/BrockmanCPSSTZ16]. We find that our proposed algorithm can learn diverse skills without any reward, even in problems with high-dimensional state and actuation, as illustrated in Figure \[Fig:skills\]. DADS can discover primitives for Half-Cheetah to run forward and backward with multiple different gaits, for Ant to navigate the environment using diverse locomotion primitives and for Humanoid to walk using stable locomotion primitives with diverse gaits and direction. The videos of the discovered primitives are available at: <https://sites.google.com/view/dads-skill>
Qualitatively, we find the skills discovered by DADS to be predictable and stable, in line with implicit constraints of the proposed objective. While the Half-Cheetah will learn to run in both backward and forward directions, DADS will disincentivize skills which make Half-Cheetah flip owing to the reduced predictability on landing. Similarly, skills discovered for Ant rarely flip over, and tend to provide stable navigation primitives in the environment. This also incentivizes the Humanoid, which is characteristically prone to collapsing and extremely unstable by design, to discover gaits which are stable for sustainable locomotion.
One of the significant advantages of the proposed objective is that it is compatible with continuous skill spaces, which has not been shown in prior work on skill discovery [@eysenbach2018diversity]. Not only does this allow us to embed a large and diverse set of skills into a compact latent space, but also the smoothness of the learned space allows us to interpolate between behaviors generated in the environment. We demonstrate this on the Ant environment (Figure \[Fig:latent\_space\_visualization\]), where we learn two-dimensional continuous skill space with a uniform prior over $(-1, 1)$ in each dimension, and compare it to a discrete skill space with a uniform prior over $20$ skills. Similar to [@eysenbach2018diversity], we restrict the observation space of the skill-dynamics $q$ to the cartesian coordinates $(x, y)$. We hereby call this the *x-y prior*, and discuss its role in Section \[sec:variance\_primitives\].
Trajectories in Discrete Skill Space
![(Left, Centre) X-Y traces of Ant skills and (Right) Heatmap to visualize the learned continuous skill space. Traces demonstrate that the continuous space offers far greater diversity of skills, while the heatmap demonstrates that the learned space is smooth, as the orientation of the X-Y trace varies smoothly as a function of the skill.[]{data-label="Fig:latent_space_visualization"}](images/traj_plot_discrete_fix1.png){width="\textwidth"}
Trajectories in Continuous Skill Space
![(Left, Centre) X-Y traces of Ant skills and (Right) Heatmap to visualize the learned continuous skill space. Traces demonstrate that the continuous space offers far greater diversity of skills, while the heatmap demonstrates that the learned space is smooth, as the orientation of the X-Y trace varies smoothly as a function of the skill.[]{data-label="Fig:latent_space_visualization"}](images/traj_plot_continuous_fix1.png){width="\textwidth"}
Orientation of Ant Trajectory
![(Left, Centre) X-Y traces of Ant skills and (Right) Heatmap to visualize the learned continuous skill space. Traces demonstrate that the continuous space offers far greater diversity of skills, while the heatmap demonstrates that the learned space is smooth, as the orientation of the X-Y trace varies smoothly as a function of the skill.[]{data-label="Fig:latent_space_visualization"}](images/heatmap_fix4.png){width="\textwidth"}
In Figure \[Fig:latent\_space\_visualization\], we project the trajectories of the learned Ant skills from both discrete and continuous spaces onto the Cartesian plane. From the traces of the skills, it is clear that the continuous latent space can generate more diverse trajectories. We demonstrate in Section \[sec:mbrl\], that continuous primitives are more amenable to hierarchical composition and generally perform better on downstream tasks. More importantly, we observe that the learned skill space is semantically meaningful. The heatmap in Figure \[Fig:latent\_space\_visualization\] shows the orientation of the trajectory (with respect to the $x$-axis) as a function of the skill $z \in \mathcal{Z}$, which varies smoothly as $z$ is varied, with explicit interpolations shown in Appendix \[Appendix:interpolation\].
Skill Variance Analysis {#sec:variance_primitives}
-----------------------
Standard Deviation of Trajectories
![(Top-Left) Standard deviation of Ant’s position as a function of steps in the environment, averaged over multiple skills and normalized by the norm of the position. (Top-Right to Bottom-Left Clockwise) X-Y traces of skills learned using DIAYN with $x$-$y$ prior, DADS with $x$-$y$ prior and DADS without x-y prior, where the same color represents trajectories resulting from the execution of the same skill $z$ in the environment. High variance skills from DIAYN offer limited utility for hierarchical control.[]{data-label="Fig:variance_analysis"}](images/primitive_variance_y_fix3.png "fig:"){width="98.00000%"} DADS without x-y prior
![(Top-Left) Standard deviation of Ant’s position as a function of steps in the environment, averaged over multiple skills and normalized by the norm of the position. (Top-Right to Bottom-Left Clockwise) X-Y traces of skills learned using DIAYN with $x$-$y$ prior, DADS with $x$-$y$ prior and DADS without x-y prior, where the same color represents trajectories resulting from the execution of the same skill $z$ in the environment. High variance skills from DIAYN offer limited utility for hierarchical control.[]{data-label="Fig:variance_analysis"}](images/multi_executions_full_obs_fix1.png){width="90.00000%"}
DIAYN with x-y prior
![(Top-Left) Standard deviation of Ant’s position as a function of steps in the environment, averaged over multiple skills and normalized by the norm of the position. (Top-Right to Bottom-Left Clockwise) X-Y traces of skills learned using DIAYN with $x$-$y$ prior, DADS with $x$-$y$ prior and DADS without x-y prior, where the same color represents trajectories resulting from the execution of the same skill $z$ in the environment. High variance skills from DIAYN offer limited utility for hierarchical control.[]{data-label="Fig:variance_analysis"}](images/multi_executions_diayn_xy_fix1.png "fig:"){width="90.00000%"} DADS with x-y prior
![(Top-Left) Standard deviation of Ant’s position as a function of steps in the environment, averaged over multiple skills and normalized by the norm of the position. (Top-Right to Bottom-Left Clockwise) X-Y traces of skills learned using DIAYN with $x$-$y$ prior, DADS with $x$-$y$ prior and DADS without x-y prior, where the same color represents trajectories resulting from the execution of the same skill $z$ in the environment. High variance skills from DIAYN offer limited utility for hierarchical control.[]{data-label="Fig:variance_analysis"}](images/multi_executions_dads_xy_fix1.png){width="90.00000%"}
In an unsupervised skill learning setup, it is important to optimize the primitives to be diverse. However, we argue that diversity is not sufficient for the learned primitives to be useful for downstream tasks. Primitives must exhibit low-variance behavior, which enables long-horizon composition of the learned skills in a hierarchical setup. We analyze the variance of the $x$-$y$ trajectories in the environment, where we also benchmark the variance of the primitives learned by DIAYN [@eysenbach2018diversity]. For DIAYN, we use the $x$-$y$ prior for the skill-discriminator, which biases the discovered skills to diversify in the $x$-$y$ space. This step was necessary for that baseline to obtain a competitive set of navigation skills. Figure \[Fig:variance\_analysis\] (Top-Left) demonstrates that DADS, which optimizes the primitives for predictability and diversity, yields significantly lower-variance primitives when compared to DIAYN, which only optimizes for diversity. This is starkly demonstrated in the plots of X-Y traces of skills learned in different setups. Skills learned by DADS show significant control over the trajectories generated in the environment, while skills from DIAYN exhibit high variance in the environment, which limits their utility for hierarchical control. This is further demonstrated quantitatively in Section \[sec:hrl\_on\_skills\].
While optimizing for predictability already significantly reduces the variance of the trajectories generated by a primitive, we find that using the $x$-$y$ prior with DADS brings down the skill variance even further. For quantitative benchmarks in the next sections, we assume that the Ant skills are learned using an $x$-$y$ prior on the observation space, for both DADS and DIAYN.
Model-Based Reinforcement Learning
----------------------------------
\[sec:mbrl\] The key utility of learning a parametric model $q_\phi(s' | s, z)$ is to take advantage of planning algorithms for downstream tasks, which can be extremely sample-efficient. In our setup, we can solve test-time tasks in zero-shot, that is *without any learning on the downstream task*. We compare with the state-of-the-art model-based RL method [@DBLP:journals/corr/abs-1805-12114], which learns a dynamics model parameterized as $p(s' | s, a)$, on the task of the Ant navigating to a specified goal with a dense reward. Given a goal $g$, reward at any position $u$ is given by $r(u) = -\lVert g - u\rVert_2$. We benchmark our method against the following variants:
- Random-MBRL (*rMBRL*): We train the model $p(s' | s, a)$ on randomly collected trajectories, and test the zero-shot generalization of the model on a distribution of goals.
- Weak-oracle MBRL (*WO-MBRL*): We train the model $p(s' | s, a)$ on trajectories generated by the planner to navigate to a goal, randomly sampled in every episode. The distribution of goals during training matches the distribution at test time.
- Strong-oracle MBRL (*SO-MBRL*): We train the model $p(s' | s, a)$ on a trajectories generated by the planner to navigate to a specific goal, which is fixed for both training and test time.
Amongst the variants, only the rMBRL matches our assumptions of having an unsupervised task-agnostic training. Both WO-MBRL and SO-MBRL benefit from goal-directed exploration during training, a significant advantage over DADS, which only uses mutual-information-based exploration.
We use $\Delta = \sum_{t=1}^H\frac{-r(u)}{H\lVert g\rVert_2}$ as the metric, which represents the distance to the goal $g$ averaged over the episode (with the same fixed horizon $H$ for all models and experiments), normalized by the initial distance to the goal $g$. Therefore, lower $\Delta$ indicates better performance and $0 < \Delta \leq 1$ (assuming the agent goes closer to the goal). The test set of goals is fixed for all the methods, sampled from $[-15, 15]^2$.
Figure \[Fig:mbrl\_benchmark\] demonstrates that the zero-shot planning significantly outperforms all model-based RL baselines, despite the advantage of the baselines being trained on the test goal(s). For the experiment depicted in Figure \[Fig:mbrl\_benchmark\] (Right), DADS has an unsupervised pre-training phase, unlike SO-MBRL which is training directly for the task. A comparison with Random-MBRL shows the significance of mutual-information-based exploration, especially with the right parameterization and priors. This experiment also demonstrates the advantage of learning a continuous space of primitives, which outperforms planning on discrete primitives.
![(Left) The results of the MPPI controller on skills learned using DADS-c (continuous primitives) and DADS-d (discrete primitives) significantly outperforms state-of-the-art model-based RL. (Right) Planning for a new task does not require any additional training and outperforms model-based RL being trained for the specific task. []{data-label="Fig:mbrl_benchmark"}](images/mbrl_bar_fix4.png){width="\textwidth"}
![(Left) The results of the MPPI controller on skills learned using DADS-c (continuous primitives) and DADS-d (discrete primitives) significantly outperforms state-of-the-art model-based RL. (Right) Planning for a new task does not require any additional training and outperforms model-based RL being trained for the specific task. []{data-label="Fig:mbrl_benchmark"}](images/so_mbrl_dads_fix4.png){width="\textwidth"}
Hierarchical Control with Unsupervised Primitives {#sec:hrl_on_skills}
-------------------------------------------------
![(Left) A RL-trained meta-controller is unable to compose primitive learned by DIAYN to navigate Ant to a goal, while it succeeds to do so using the primitives learned by DADS. (Right) Goal-Conditioned RL (GCRL-dense/sparse) does not generalize outside its training distribution, while MPPI controller on learned skills (DADS-dense/sparse) experiences significantly smaller degrade in performance.[]{data-label="Fig:mfrl_benchmark"}](images/hierarchical_control_usl_fix4.png){width="\textwidth"}
![(Left) A RL-trained meta-controller is unable to compose primitive learned by DIAYN to navigate Ant to a goal, while it succeeds to do so using the primitives learned by DADS. (Right) Goal-Conditioned RL (GCRL-dense/sparse) does not generalize outside its training distribution, while MPPI controller on learned skills (DADS-dense/sparse) experiences significantly smaller degrade in performance.[]{data-label="Fig:mfrl_benchmark"}](images/gcrl_benchmark_30_fix4.png){width="\textwidth"}
We benchmark hierarchical control for primitives learned without supervision, against our proposed scheme using an MPPI based planner on top of DADS-learned skills. We persist with the task of Ant-navigation as described in \[sec:mbrl\]. We benchmark against Hierarchical DIAYN [@eysenbach2018diversity], which learns the skills using the DIAYN objective, freezes the low-level policy and learns a meta-controller that outputs the skill to be executed for the next $H_Z$ steps. We provide the $x$-$y$ prior to the DIAYN’s disciminator while learning the skills for the Ant agent. The performance of the meta-controller is constrained by the low-level policy, however, this hierarchical scheme is agnostic to the algorithm used to learn the low-level policy. To contrast the quality of primitives learned by the DADS and DIAYN, we also benchmark against Hierarchical DADS, which learns a meta-controller the same way as Hierarchical DIAYN, but learns the skills using DADS.
From Figure \[Fig:mfrl\_benchmark\] (Left) We find that the meta-controller is unable to compose the skills learned by DIAYN, while the same meta-controller can learn to compose skills by DADS to navigate the Ant to different goals. This result seems to confirm our intuition described in Section \[sec:variance\_primitives\], that the high variance of the DIAYN skills limits their temporal compositionality. Interestingly, learning a RL meta-controller reaches similar performance to the MPPI controller, taking an additional $200,000$ samples per goal.
Goal-conditioned RL {#sec:gcrl}
-------------------
To demonstrate the benefits of our approach over model-free RL, we benchmark against goal-conditioned RL on two versions of Ant-navigation: (a) with a dense reward $r(u)$ and (b) with a sparse reward $r(u) = 1$ if $\lVert u-g \rVert_2 \leq \epsilon$, else 0. We train the goal-conditioned RL agent using soft actor-critic, where the state variable of the agent is augmented with $u-g$, the position delta to the goal. The agent gets a randomly sampled goal from $[-10, 10]^2$ at the beginning of the episode. In Figure \[Fig:mfrl\_benchmark\] (Right), we measure the average performance of the all the methods as a function of the initial distance of the goal, ranging from 5 to 30 metres. For dense reward navigation, we observe that while model-based planning on DADS-learned skills degrades smoothly as the initial distance to goal to increases, goal-conditioned RL experiences a sudden deterioration outside the goal distribution it was trained on. Even within the goal distribution observed during training of goal-conditioned RL model, skill-space planning performs competitively to it. With sparse reward navigation, goal-conditioned RL is unable to navigate, while MPPI demonstrates comparable performance to the dense reward up to about 20 metres. This highlights the utility of learning task-agnostic skills, which makes them more general while showing that latent space planning can be leveraged for tasks requiring long-horizon planning.
Conclusion
==========
We have proposed a novel unsupervised skill learning algorithm that is amenable to model-based planning for hierarchical control on downstream tasks. We show that our skill learning method can scale to high-dimensional state-spaces, while discovering a diverse set of low-variance skills. In addition, we demonstrated that, without any training on the specified task, we can compose the learned skills to outperform competitive model-based baselines that were trained with the knowledge of the test tasks. We plan to extend the algorithm to work with off-policy data, potentially using relabelling tricks [@DBLP:journals/corr/AndrychowiczWRS17; @nachum2018data] and explore more nuanced planning algorithms. We plan to apply the hereby-introduced method to different domains, such as manipulation and enable skill/model discovery directly from images.
Acknowledgements
================
We would like to thank Evan Liu, Ben Eysenbach, Anusha Nagabandi for their help in reproducing the baselines for this work. We are thankful to Ben Eysenbach for their comments and discussion on the initial drafts. We would also like to acknowledge Ofir Nachum, Alex Alemi, Daniel Freeman, Yiding Jiang, Allan Zhou and other colleagues at Google Brain for their helpful feedback and discussions at various stages of this work. We are also thankful to Michael Ahn and others in Adept team for their support, especially with the infrastructure setup and scaling up the experiments.
Implementation Details {#appendix:implementation}
======================
All of our models are written in the open source Tensorflow-Agents [@TFAgents], based on Tensorflow [@tensorflow2015-whitepaper].
Skill Spaces
------------
When using discrete spaces, we parameterize $\mathcal{Z}$ as one-hot vectors. These one-hot vectors are randomly sampled from the uniform prior $p(z) = \frac{1}{D}$, where $D$ is the number of skills. We experiment with $D \leq 128$. For discrete skills learnt for MuJoCo Ant in Section \[sec:mbrl\], we use $D=20$. For continuous spaces, we sample $z \sim \textrm{Uniform}(-1, 1)^D$. We experiment with $D=2$ for Ant learnt with x-y prior, $D=3$ for Ant learnt without x-y prior (that is full observation space), to $D=5$ for Humanoid on full observation spaces. The skills are sampled once in the beginning of the episode and fixed for the rest of the episode. However, it is possible to resample the skill from the prior within the episode, which allows for every skill to experience a different distribution than the initialization distribution. This also encourages discovery of skills which can be composed temporally. However, this increases the hardness of problem, especially if the skills are re-sampled from the prior frequently.
Agent
-----
We use SAC as the optimizer for our agent $\pi(a \mid s, z)$, in particular, EC-SAC [@DBLP:journals/corr/abs-1812-05905]. The $s$ input to the policy generally excludes global co-ordinates $(x, y)$ of the centre-of-mass, available for a lot of enviroments in OpenAI gym, which helps produce skills agnostic to the location of the agent. We restrict to two hidden layers for our policy and critic networks. However, to improve the expressivity of skills, it is beneficial to increase the capacity of the networks. The hidden layer sizes can vary from $(128, 128)$ for Half-Cheetah to $(512, 512)$ for Ant and $(1024, 1024)$ for Humanoid. The critic $Q(s, a, z)$ is similarly parameterized. The target function for critic $Q$ is updated every iteration using a soft updates with co-efficient of $0.005$. We use Adam [@kingma2014adam] optimizer with a fixed learning rate of $3e-4$ , and a fixed initial entropy co-efficient $\beta=0.1$. While the policy is parameterized as a normal distribution $\mathcal{N}(\mu(s, z), \Sigma(s, z))$ where $\Sigma$ is a diagonal covariance matrix, it undergoes through tanh transformation, to transform the output to the range $(-1, 1)$ and constrain to the action bounds.
Skill-Dynamics
--------------
Skill-dynamics, denoted by $q(s' \mid s, z)$, is parameterized by a deep neural network. A common trick in model-based RL is to predict the $\Delta s = s'-s$, rather than the full state $s'$. Hence, the prediction network is $q(\Delta s \mid s, z)$. Note, both parameterizations can represent the same set of functions. However, the latter will be easy to learn as $\Delta s$ will be centred around $0$. We exclude the global co-ordinates from from the state input to $q$. However, we can (and we still do) predict $\Delta_x, \Delta_y$, because reward functions for goal-based navigation generally rely on the position prediction from the model. This represents another benefit of predicting state-deltas, as we can still predict changes in position without explicitly knowing the global position.
The output distribution is modelled as a Mixture-of-Experts [@jacobs1991adaptive]. We fix the number of experts to be 4. We model each expert as a Gaussian distribution. The input $(s, z)$ goes through two hidden layers (the same capacity as that of policy and critic networks, for example $(512, 512)$ for Ant). The output of the two hidden layers is used as an input to the mixture-of-experts, which is linearly transformed to output the parameters of the Gaussian distribution, and a discrete distribution over the experts using a softmax distribution. In practice, we fix the covariance matrix of the Gaussian experts to be an identity matrix, so we only need to output the means for the experts. We use batch-normalization for both input and the hidden layers. We normalize the output targets using their batch-average and batch-standard deviation, similar to batch-normalization.
Other Hyperparameters
---------------------
The episode horizon is generally kept shorter for stable agents like Ant ($200$), while longer for unstable agents like Humanoid ($1000$). For Ant, longer episodes do not add value, but Humanoid can benefit from longer episodes as it helps it filter skills which are unstable. The optimization scheme is on-policy, and we collect $2000$ steps for Ant and $4000$ steps for Humanoid in one iteration. The intuition is to experience trajectories generated by multiple skills (approximately $10$) in a batch. Re-sampling skills can enable experiencing larger number of skills. Once a batch of episodes is collected, the skill-dynamics is updated using Adam optimizer with a fixed learning rate of $3e-4$. The batch size is $128$, and we carry out $32$ steps of gradient descent. To compute the intrinsic reward, we need to resample the prior for computing the denominator. For continuous spaces, we set $L=500$. For discrete spaces, we can marginalize over all skills. After the intrinsic reward is computed, the policy and critic networks are updated for $128$ steps with a batch size of $128$. The intuition is to ensure that every sample in the batch is seen for policy and critic updates about $3-4$ times in expectation.
Planning and Evaluation Setups
------------------------------
For evaluation, we fix the episode horizon to $200$ for all models in all evaluation setups. Depending upon the size of the latent space and planning horizon, the number of samples from the planning distribution $P$ is varied between $10-200$. For $H_P=1, H_Z=10$ and a $2D$ latent space, we use $50$ samples from the planning distribution $P$. The co-efficient $\gamma$ for MPPI is fixed to $10$. We use a setting of $H_P=1$ and $H_Z=10$ for dense-reward navigation, in which case we set the number of refine steps $R=10$. However, for sparse reward navigation it is important to have a longer horizon planning, in which case we set $H_P=4, H_Z=25$ with a higher number of samples from the planning distribution ($200$ from $P$). Also, when using longer planning horizons, we found that smoothing the sampled plans help. Thus, if the sampled plan is $z_1, z_2, z_3, z_4 \ldots$, we smooth the plan to make $z_2 = \beta z_1 + (1-\beta)z_2$ and so on, with $\beta=0.9$.
For hierarchical controllers being learnt on top of low-level unsupervised primitives, we use PPO [@schulman2017proximal] for discrete action skills, while we use SAC for continuous skills. We keep the number of steps after which the meta-action is decided as 10 (that is $H_Z=10$). The hidden layer sizes of the meta-controller are $(128, 128)$. We use a learning rate of $1e-4$ for PPO and $3e-4$ for SAC.
For our model-based RL baseline PETS, we use an ensemble size of $3$, with a fixed planning horizon of $20$. For the model, we use a neural network with two hidden layers of size $400$. In our experiments, we found that MPPI outperforms CEM, so we report the results using the MPPI as our controller.
Graphical models, Information Bottleneck and Unsupervised Skill Learning {#appendix:infobot}
========================================================================
We now present a novel perspective on unsupervised skill learning, motivated from the literature on information bottleneck. This section takes inspiration from [@alemi2018therml], which helps us provide a rigorous justification for our objective proposed earlier. To obtain our unsupervised RL objective, we setup a graphical model $P$ as shown in Figure \[Fig:real\_dynamics\], which represents the distribution of trajectories generated by a given policy $\pi$. The joint distribution is given by: $$\begin{aligned}
p(s_1, a_1 \ldots a_{T-1}, s_T, z) = p(z)p(s_1) \prod_{t=1}^{T-1} \pi(a_t | s_t, z) p(s_{t+1} | s_{t}, a_t).\end{aligned}$$
We setup another graphical model $N$, which represents the desired model of the world. In particular, we are interested in approximating $p(s' | s, z)$, which represents the transition function for a particular primitive. This abstraction helps us get away from knowing the exact actions, enabling model-based planning in behavior space (as discussed in the main paper). The joint distribution for $N$ shown in Figure \[Fig:approx\_dynamics\] is given by: $$\begin{aligned}
\eta(s_1, a_1, \ldots s_T, a_T, z) = \eta(z)\eta(s_1) \prod_{t=1}^{T-1} \eta(a_t)\eta(s_{t+1} | s_{t}, z).\end{aligned}$$
The goal of our approach is to optimize the distribution $\pi(a | s, z)$ in the graphical model $P$ to minimize the distance between the two distributions, when transforming to the representation of the graphical model $Z$. In particular, we are interested in minimizing the KL divergence between $p$ and $\eta$, that is $\mathcal{D}_{KL}(p || \eta)$. Note, if $N$ had the same structure as $P$, the information lost in projection would be $0$ for any valid $P$. Interestingly, we can exploit the following result from in [@friedman2001multivariate] to setup the objective for $\pi$, without explicitly knowing $\eta$: $$\begin{aligned}
\min_\eta \mathcal{D}_{KL}(p || \eta) = \mathcal{I}_P - \mathcal{I}_N,\end{aligned}$$
where $\mathcal{I}_P$ and $\mathcal{I}_N$ represents the multi-information for distribution $P$ on the respective graphical models. Note, $\min_{\eta \in N} \mathcal{D}_{KL}(p || \eta)$, which is the reverse information projection [@csiszar2003information]. The multi-information [@slonim2005estimating] for a graphical model $G$ with nodes $g_i$ is defined as: $$\begin{aligned}
\mathcal{I}_G = \sum_i I(g_i;Pa(g_i)),\end{aligned}$$ where $Pa(g_i)$ denotes the nodes upon which $g_i$ has direct conditional dependence in $G$. Using this definition, we can compute the multi-information terms: $$\begin{aligned}
\mathcal{I}_P &= \sum_{t=1}^T I(a_t; \{s_t, z\}) + \sum_{t=2}^T I(s_t ; \{s_{t-1}, a_{t-1}\}) \quad \textrm{and} \quad
\mathcal{I}_N = \sum_{t=2}^T I(s_t ; \{s_{t-1}, z\}).\end{aligned}$$ Following the Optimal Frontier argument in [@alemi2018therml], we introduce Lagrange multipliers $\beta_t \geq 0, \delta_t \geq 0$ for the information terms in $\mathcal{I}_P$ to setup an objective $R(\pi)$ to be maximized with respect to $\pi$: $$\begin{aligned}
R(\pi) &= \sum_{t=1}^{T-1} I(s_{t+1} ; \{s_{t}, z\}) - \beta_t I(a_t ; \{s_t, z\}) -\delta_t \mathcal{I}(s_{t+1}; \{s_{t}, a_{t}\})\\\end{aligned}$$ As the underlying dynamics are fixed and unknown, we simplify the optimization by setting $\delta_t=0$ which intuitively corresponds to us neglecting the unchangeable information of the underlying dynamics. This gives us $$\begin{aligned}
R(\pi) &= \sum_{t=1}^{T-1} I(s_{t+1} ; \{s_{t}, z\}) - \beta_t I(a_t ; \{s_t, z\}) \\
& \geq \sum_{t=1}^{T-1} I(s_{t+1};z \mid s_t) - \beta_t I(a_t ; \{s_t, z\}) \label{eq:bottleneck_obj}\end{aligned}$$ Here, we have used the chain rule of mutual information: $\mathcal{I}(s_{t+1} ; \{s_t, z\}) = \mathcal{I}(s_{t+1} ; s_t) + \mathcal{I}(s_{t+1};z \mid s_t) \geq \mathcal{I}(s_{t+1};z \mid s_t)$, resulting from the non-negativity of mutual information. This yield us an information bottleneck style objective where we maximize the mutual information motivated in Section \[sec:DADS\], while minimizing $\mathcal{I}(a_t; \{s_t, z\})$. We can show that the minimization of the latter mutual information corresponds to entropy regularization of $\pi(a_t \mid s_t, z)$, as follows: $$\begin{aligned}
\mathcal{I}(a_t; \{s_t, z\}) &= \mathbb{E}_{a_t \sim \pi(a_t \mid s_t, z), s_t, z \sim p}\Big[\log \frac{\pi(a_t \mid s_t, z)}{\pi(a_t)} \Big] \\
&= \mathbb{E}_{a_t \sim \pi(a_t \mid s_t, z), s_t, z \sim p}\Big[\log \frac{\pi(a_t \mid s_t, z)}{p(a_t)}\Big] - \mathcal{D}_{KL}(\pi(a_t) \mid\mid p(a_t))\\
&\leq \mathbb{E}_{a_t \sim \pi(a_t \mid s_t, z), s_t, z \sim p}\Big[\log \frac{\pi(a_t \mid s_t, z)}{p(a_t)}\Big] \label{eq:entropy_inequality}\end{aligned}$$ for some arbitrary distribution $\log p(a_t)$ (for example uniform). Again, we have used the non-negativity of $\mathcal{D}_{KL}$ to get the inequality. We use Equation \[eq:entropy\_inequality\] in Equation \[eq:bottleneck\_obj\] to get: $$\begin{aligned}
R(\pi) \geq \sum_{t=1}^{T-1} \mathcal{I}(s_{t+1};z \mid s_t) -\beta_t\mathbb{E}_{a_t \sim \pi(a_t \mid s_t, z), s_t, z \sim p}\Big[\log \pi(a_t \mid s_t, z)\Big]\end{aligned}$$ where we have ignored $p(a_t)$ as it is a constant with respect to optimization for $\pi$. This motivates the use of entropy regularization. We can follow the arguments in Section \[sec:DADS\] to obtain an approximate lower bound for $\mathcal{I}(s_{t+1};z \mid s_t)$. The above discussion shows how DADS can be motivated from a graphical modelling perspective, while justifying the use of entropy regularization from an information bottleneck perspective. This objective also explicates the temporally extended nature of $z$, and how it corresponds to a sequence of actions producing a predictable sequence of transitions in the environment.
We can carry out the exercise for the reward function in [@eysenbach2018diversity] (DIAYN) to provide a graphical model interpretation of the objective used in the paper. To conform with objective in the paper, we assume to be sampling to be state-action pairs from skill-conditioned stationary distributions in the world $P$, rather than trajectories. The objective to be maximized is given by: $$\begin{aligned}
R(\pi) &= -\mathcal{I}_P + \mathcal{I}_Q\\
&= -I(a;\{s, z\}) + I(z;s)\\
&=\mathbb{E}_\pi[ \log{\frac{p(z|s)}{p(z)}} - \log{\frac{\pi(a | s, z)}{\pi(a)}}]\\
&\geq \mathbb{E}_\pi[\log{q_\phi(z | s)} - \log{p(z)} - \log{\pi(a | s, z)}] = R(\pi, q_\phi)\end{aligned}$$ where we have used the variational inequalities to replace $p(z|s)$ with $q_\phi(z|s)$ and $\pi(a)$ with a uniform prior over bounded actions $p(a)$ (which is ignored as a constant).
Approximating the Reward Function {#appendix:reward_approximation}
=================================
We revisit Equation \[eq:alternate\_opt\] and the resulting approximate reward function constructed in Equation \[eq:final\_reward\]. The maximization objective for policy was: $$\begin{aligned}
R(\pi \mid q_\phi) = \mathbb{E}_{z, s, s'}\big[\log q_\phi(s' \mid s, z) - \log p(s' \mid s) \big]\end{aligned}$$
The computational problem arises from the intractability of $p(s' \mid s) = \int p(s' \mid s, z) p(z \mid s) dz$, where both $p(s' \mid s, z)$ and $p(z \mid s) \propto p(s \mid z) p(z)$ are intractable. Unfortunately, any variational approximation results in an improper lower bound for the objective. To see that: $$\begin{aligned}
R(\pi \mid q_\phi) &= \mathbb{E}_{z, s, s'} \big[\log q_\phi(s' \mid s, z) - \log q(s' \mid s)\big] - \mathcal{D}_{KL}(p(s' \mid s) \mid\mid q(s' \mid s))\\
&\leq \mathbb{E}_{z, s, s'} \big[\log q_\phi(s' \mid s, z) - \log q(s' \mid s)\big]\end{aligned}$$ where the inequality goes the wrong way for any variational approximation $q(s' \mid s)$. Our approximation can be seen as a special instantiation of $q(s'\mid s) = \int q_\phi(s' \mid s, z) p(z) dz$. This approximation is simple to compute as generating samples from the prior $p(z)$ is inexpensive and effectively requires only a forward pass through $q_\phi$. Reusing $q_\phi$ to approximate $p(s' \mid s)$ makes intuitive sense because we want $q_\phi$ to reasonably approximate $p(s' \mid s, z)$ (which is why we collect large batches of data and take multiple steps of gradient descent for fitting $q_\phi$). While sampling from the prior $p(z)$ is crude, sampling $p(z \mid s)$ can be computationally prohibitive. For a certain class of problems, especially locomotion, sampling from $p(z)$ is a reasonable approximation as well. We want our primitives/skills to be usable from any state, which is especially the case with locomotion. Empirically, we have found our current approximation provides satisfactory results. We also discuss some other potential solutions (and their limitations):
\(a) One could potentially use another network $q_\beta(z \mid s)$ to approximate $p(z\mid s)$ by minimizing $\mathbb{E}_{s, z \sim p}\big[D_{KL}(p(z\mid s) \mid\mid q_\beta(z \mid s))\big]$. Note, the resulting approximation would still be an improper lower bound for $R(\pi \mid q_\phi)$. However, sampling from this $q_\beta$ might result in a better approximation than sampling from the prior $p(z)$ for some problems.
\(b) We can bypass the computational intractability of $p(s' \mid s)$ by exploiting the variational lower bounds from [@agakov2004algorithm]. We use the following inequality, used in [@hausman2018learning]: $$\begin{aligned}
\mathcal{H}(x) \geq \int p(x,z) \log \frac{q(z|x)}{p(x,z)} dx dz\end{aligned}$$ where $q$ is a variational approximation to the posterior $p(z|x)$. $$\begin{aligned}
I(s';z | s) &= -\mathcal{H}(s' | s, z) + \mathcal{H}(s'|s)\\
&\geq \mathbb{E}_{z, s, s' \sim p}\big[\log q_\phi(s' | s, z)] + \mathbb{E}_{z, s, s' \sim p}\big[\log q_\alpha(z | s', s)\big] + \mathcal{H}(s', z | s)\\
&= \mathbb{E}_{z, s, s' \sim p}\big[\log{q_\phi(s' | s, z)} + \log{q_\alpha}(z | s', s)] + \mathcal{H}(s', z | s)\end{aligned}$$ where we have used the inequality for $\mathcal{H}(s' | s)$ to introduce the variational posterior for skill inference $q_\alpha (z \mid s', s)$ besides the conventional variational lower bound to introduce $q(s' \mid s, z)$. Further decomposing the leftover entropy: $$\begin{aligned}
\mathcal{H}(s', z | s) = \mathcal{H}(z|s) + \mathcal{H}(s' | s, z)\end{aligned}$$ Reusing the variational lower bound for marginal entropy from [@agakov2004algorithm], we get: $$\begin{aligned}
\mathcal{H}(s'|s,z) &\geq \mathbb{E}_{s, z}\Big[\int p(s', a | s, z) \log \frac{q(a | s', s, z)}{p(s', a | s, z)} ds' da\Big]\\
&= -\log c + \mathcal{H}(s', a | s, z)\\
&= -\log c + \mathcal{H}(s' | s, a, z) + \mathcal{H}(a | s, z)\end{aligned}$$ Since, the choice of posterior is upon us, we can choose $q(a | s', s, z) = 1/c$ to induce a uniform distribution for the bounded action space. For $\mathcal{H}(s' | s, a, z)$, notice that the underlying dynamics $p(s' | s, a)$ are independent of $z$, but the actions do depend upon $z$. Therefore, this corresponds to entropy-regularized RL when the dynamics of the system are deterministic. Even for stochastic dynamics, the analogy might be a good approximation , assuming the underlying dynamics are not very entropic. The final objective (making this low-entropy dynamics assumption) can be written as: $$\begin{aligned}
I(s';z | s) \geq \mathbb{E}_s\mathbb{E}_{p(s',z | s)}[\log{q_\phi(s' | s, z)} + \log{q_\alpha}(z | s', s) - \log p(z|s)] + \mathcal{H}(a | s, z)\end{aligned}$$ While this does bypass the intractability of $p(s' \mid s)$, it runs into the intractable $p(z \mid s)$, despite deploying significant mathematical machinery and additional assumptions. Any variational approximation for $p(z \mid s)$ would again result in an improper lower bound for $\mathcal{I}(s' ; z \mid s)$.
\(c) One way to a make our approximation $q(s' \mid s)$ to more closely resemble $p(s' \mid s)$ is to change our generative model $p(z, s, s')$. In particular, if we resample $z \sim p(z)$ for every timestep of the rollout from $\pi$, we can indeed write $p(z \mid s) = p(z)$. Note, $p(s' \mid s)$ is still intractable to compute, but marginalizing $q_\phi(s' \mid s, z)$ over $p(z)$ becomes a better approximation of $p(s' \mid s)$. However, this severely dampens the interpretation of our latent space $\mathcal{Z}$ as temporally extended actions (or skills). It becomes better to interpret the latent space $\mathcal{Z}$ as dimensional reduction of action space. Empirically, we found that this significantly throttles the learning, not yielding useful or interpretable skills.
Interpolation in Continuous Latent Space {#Appendix:interpolation}
========================================
![Interpolation in the continuous primitive space learned using DADS on the Ant environment corresponds to interpolation in the trajectory space. (Left) Interpolation from $z=[1.0, 1.0]$ (solid blue) to $z=[-1.0, 1.0]$ (dotted cyan); (Middle) Interpolation from $z = [1.0, 1.0]$ (solid blue) to $z = [-1.0, -1.0]$ (dotted cyan); (Right) Interpolation from $z=[1.0, 1.0]$ (solid blue) to $z=[1.0, -1.0]$ (dotted cyan).](images/image18.png "fig:"){width="30.00000%"} ![Interpolation in the continuous primitive space learned using DADS on the Ant environment corresponds to interpolation in the trajectory space. (Left) Interpolation from $z=[1.0, 1.0]$ (solid blue) to $z=[-1.0, 1.0]$ (dotted cyan); (Middle) Interpolation from $z = [1.0, 1.0]$ (solid blue) to $z = [-1.0, -1.0]$ (dotted cyan); (Right) Interpolation from $z=[1.0, 1.0]$ (solid blue) to $z=[1.0, -1.0]$ (dotted cyan).](images/image19.png "fig:"){width="30.00000%"} ![Interpolation in the continuous primitive space learned using DADS on the Ant environment corresponds to interpolation in the trajectory space. (Left) Interpolation from $z=[1.0, 1.0]$ (solid blue) to $z=[-1.0, 1.0]$ (dotted cyan); (Middle) Interpolation from $z = [1.0, 1.0]$ (solid blue) to $z = [-1.0, -1.0]$ (dotted cyan); (Right) Interpolation from $z=[1.0, 1.0]$ (solid blue) to $z=[1.0, -1.0]$ (dotted cyan).](images/image26.png "fig:"){width="30.00000%"}
Model Prediction {#Appendix:model_prediction}
================
![(Left) Prediction error in the Ant’s co-ordinates (normalized by the norm of the actual position) for skill-dynamics. (Right) X-Y traces of actual trajectories (colored) compared to trajectories predicted by skill-dynamics (dotted-black) for different skills.[]{data-label="fig:prediction_error"}](images/prediction_error_xy.png){width="\textwidth"}
![(Left) Prediction error in the Ant’s co-ordinates (normalized by the norm of the actual position) for skill-dynamics. (Right) X-Y traces of actual trajectories (colored) compared to trajectories predicted by skill-dynamics (dotted-black) for different skills.[]{data-label="fig:prediction_error"}](images/pred_vs_actual_traj_plot.png){width="\textwidth"}
From Figure \[fig:prediction\_error\], we observe that skill-dynamics can provide robust state-predictions over long planning horizons. When learning skill-dynamics with $x-y$ prior, we observe that the error in prediction rises slower with horizon as compared to the norm of the actual position. This provides strong evidence of cooperation between the primitives and skill-dynamics learned using DADS with $x-y$ prior. As the error-growth for skill-dynamics learned on full-observation space is sub-exponential, similar argument can be made for DADS without $x-y$ prior as well (albeit to a weaker extent).
[^1]: Work done a part of the Google AI Residency program.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: |
The use of computational fluid dynamics (CFD) in industry is severely limited by the inability to accurately and reliably simulate complex turbulent flows. With the increase in computing power, high-order spectral/$hp$ methods for large-eddy simulation (LES) are emerging as a promising approach to predict this type of flows. However, the lack of robustness of these methods for under-resolved simulations, such as in LES, still inhibits their in industrial applications. In order to help overcome this issue, we introduce a non-modal analysis technique that characterizes the short-term dissipation of the scheme for linear convection-diffusion systems. While strictly speaking only valid for linear problems, the analysis is devised such that it can provide critical insights on the following questions: Why do high-order spectral/$hp$ methods suffer from stability issues in under-resolved computations? Why do they successfully predict under-resolved turbulent flows even without a subgrid-scale model? And, most importantly, how can we devise more robust and accurate schemes for these problems?
For illustration purposes, this analysis technique is applied to the hybridized discontinuous Galerkin methods as representatives of high-order spectral/$hp$ methods. The effect of the polynomial order, the upwinding parameter and the Péclet number on the short-term diffusion of the scheme are investigated. From non-modal analysis, polynomial orders between $2$ and $4$ with standard upwinding would be the best suited for under-resolved turbulence simulations. For larger polynomial orders, as well as for under- and over-upwinded numerical fluxes, robustness and accuracy issues can be expected. The non-modal analysis results are then tested against under-resolved turbulence simulations of the Burgers, Euler and Navier-Stokes equations. While devised in the linear setting, non-modal analysis succeeds to predict the behavior of the scheme in the nonlinear problems considered.
author:
- 'P. Fernandez[^1], R. Moura[^2], G. Mengaldo[^3], N.C. Nguyen[^4], J. Peraire[^5]'
title: 'Numerical dissipation of spectral/$hp$ methods via non-modal analysis and its application to under-resolved turbulence simulations'
---
\[s:introduction\]Introduction
==============================
High-order spectral/$hp$ methods for large-eddy simulation (LES), including continuous Galerkin (CG) methods [@Karamanos:2000; @Kirby:2003], standard discontinuous Galerkin (DG) methods [@Beck:14; @Frere:15; @Gassner:13; @Murman:16; @Renac:15; @Uranga:11; @Wiart:15], hybridized DG methods [@Fernandez:16a; @Fernandez:17a], spectral difference (SD) methods [@Liang:2009; @Parsani:2010] and flux reconstruction (FR) methods [@Park:2017; @Vermeire:2013; @Vermeire:2014], are emerging as a promising approach to solve complex turbulent flows. First, they allow for high-order discretizations on complex geometries and unstructured meshes. This is critical to accurately propagate small-scale, small-magnitude features, such as in transitional and turbulent flows, over the complex three-dimensional geometries commonly encountered in industrial applications. Second, spectral/$hp$ methods are highly tailored to emerging computing architectures, including graphics processing units (GPUs) and manycore architectures, due to their high flop-to-communication ratio. The use of these methods for LES is being further encouraged by successful numerical predictions (see references before).
The diffusion characteristics of the discretization scheme, which are still not well understood for spectral/$hp$ methods, play a critical role in robustness and accuracy in large-eddy simulation. In order to enhance the current predictive capabilities of computational fluid dynamics, the following questions must be addressed: How much numerical dissipation is introduced by the scheme? At what scales is it applied? And, how does this numerical diffusion relate to the level of under-resolution in the simulation? As customary, we use the term [*under-resolved*]{} to refer to simulations in which the exact solution contains scales that are smaller than the Nyquist wavenumber of the grid (the so-called subgrid scales) and thus cannot be captured with the grid resolution. For numerical schemes with more than one degree of freedom (DOF) per computational cell, such as in high-order spectral/$hp$ methods, several ways of investigating the diffusion characteristics of the scheme are possible. The most widely used technique is the eigensolution analysis; which has been succesfully applied to CG [@Moura:16b], standard DG [@Hu:1999; @Hu:2002; @Mengaldo:ComputersFluids:2017; @Moura:15a], hybridized DG [@Moura:18a] and FR [@Mengaldo:eigenFR:2018a] methods. Eigenanalysis concerns about the diffusion and dispersion characteristics of the discretization of the linear convection-diffusion equation, and relies on Fourier modes being eigenmodes of the discretization. This is a good approximation for well-resolved wavenumbers and becomes less accurate as the Fourier mode approaches the Nyquist wavenumber. We note that the behavior of the scheme for the latter wavenumbers is critical in the accuracy and robustness of under-resolved turbulence simulations. As an alternative and complementary approach to eigenanalysis, we are interested in the short-term dynamics of the semi-discrete system arising from the spatial discretization of the linear convection-diffusion equation. This is motivated by the idea that nonlinear systems behave similarly to its linearized version during a short period of time, and it is some form of the short-term behavior of the linearized system that is likely to be most informative about the nonlinear dynamics. To this end, we introduce a new analysis framework and refer to it as non-modal analysis as it resembles non-modal stability theory [@Schmid:2007; @Trefethen:1993; @Trefethen:1997]. Non-modal stability theory studies the transient growth of non-modal disturbances in linear dynamical systems (non-modal in the sense that they are not eigenmodes) and was a major breakthrough to characterize nonlinear instabilities by analyzing the short-term behavior of the linearized dynamics. Our non-modal analysis does not require Fourier modes to be eigenmodes of the discretization and reconciles with eigenanalysis whenever they actually are. The remainder of the paper is organized as follows. In Section \[s:nonModal\_HDG\], we introduce the non-modal analysis framework and apply it to the hybridized discontinuous Galerkin methods. The short-term diffusion characteristics of hybridized DG methods are investigated in that section. In Section \[s:applicationNonLinear\], we assess how non-modal analysis results extend to the nonlinear setting. To that end, we compare non-modal analysis with numerical results for the Burgers, Euler and Navier-Stokes equations. A discussion on how to devise superior schemes using insights from non-modal analysis is presented in Section \[s:discussion\]. We conclude the paper with some remarks in Section \[s:conclusions\].
\[s:nonModal\_HDG\]Non-modal analysis for hybridized DG
=======================================================
We illustrate the non-modal analysis framework by applying it to the hybridized discontinuous Galerkin methods [@Nguyen:15; @Fernandez:17a], a particular instance of high-order spectral/$hp$ method. More precisely, the hybridized DG methods are a class of discontinuous Galerkin methods that generalizes the HDG [@Cockburn:09a; @Nguyen:12], EDG [@Cockburn:09a; @Cockburn:09b] and IEDG [@Fernandez:16a] methods. This family of schemes is becoming increasingly popular for fluid mechanics [@Dahm:2014; @Fernandez:17a; @Giorgiani:2014; @Lehrenfeld:2016; @Rhebergen:2012; @Schutz:2013; @Ueckermann:2016; @Woopen:2014], solid mechanics [@Sheldon:2016; @Terrana:2018] and electromagnetism [@Chen:2015; @Christophe:2018; @Feng:2016; @Li:2017; @VidalCodina:2018; @Yoo:2016] as it allows (for moderately high accuracy orders) for more computationally efficient implementations than standard DG methods. The analysis herein can be readily extended to other high-order spectral/$hp$ methods, including continuous Galerkin, standard discontinuous Galerkin, spectral difference, and flux reconstruction methods. We note that hybridized and standard DG methods are equivalent for the linear convection equation (see Appendix \[s:connectionStHyDG\]), and therefore the hybridized DG non-modal analysis for pure convection also applies to standard DG. Similarly, the results for pure convection carry over to certain types of flux reconstruction schemes (see [@DeGrazia:2014; @Mengaldo:2016; @Vincent:2011] for the connections between DG and FR methods).
Spatial discretization of the linear convection-diffusion equation
------------------------------------------------------------------
We consider the linear convection-diffusion equation with constant coefficients in a one-dimensional domain $\Omega = (-\infty , \infty)$, given by
\[e:lcd\] $$\begin{aligned}
{2}
& \frac{\partial u}{\partial t} + a \frac{\partial u}{\partial x} = \nu \frac{\partial ^2 u}{\partial x^2} , \qquad & t > 0 , \\
& u = u_0 , \qquad & t = 0 , \end{aligned}$$
where $a$ is the convection velocity, $\nu \geq 0$ is the diffusion coefficient, and $u_0 \in \mathcal{C}^2(\mathbb{R} ; \mathbb{C})$ is a twice continuously differentiable (possibly complex-valued) initial condition. In order to discretize Eq. in space by hybridized DG methods, we first rewrite it in the following conservation form
\[e:lcd2\] $$\begin{aligned}
{2}
\label{e:lcd2_1}
& \frac{\partial u}{\partial t} + \frac{\partial f}{\partial x} = 0 , \qquad & t > 0 \mbox{ ,} \\
\label{e:lcd2_2}
& q = \frac{\partial u}{\partial x} , \qquad & t \geq 0 \mbox{ ,} \\
\label{e:lcd2_3}
& u = u_0 , \qquad & t = 0 \mbox{ ,} \end{aligned}$$
where $q$ is the auxiliary gradient variable and $f(u,q) = a \, u - \nu \, q$ is the flux function. After $\Omega$ is partitioned into uniform non-overlapping elements $\Omega_e$ of size $h$, the numerical solution and its gradient in a given element $\Omega_e$ are approximated by polynomial expansions of the form $$\label{eq:exp}
u_h|_{\Omega_e} = \sum_{j=0}^{p}{\tilde{u}_{h,j}(t) \, \phi_j(\xi(x))} \mbox{ ,} \qquad \qquad q_h|_{\Omega_e} = \sum_{j=0}^{p}{\tilde{q}_{h,j}(t) \, \phi_j(\xi(x))} \mbox{ .}$$ where $\phi_j$ are polynomial basis functions of degree up to $p$, defined in the reference domain $\Omega_{\textnormal{ref}} = [-1,1]$. A linear mapping relation is assumed between the physical coordinate $x$ in element $\Omega_e$ and the coordinate $\xi = \xi(x) \in \Omega_{\textnormal{ref}}$. Multiplying Equations $-$ by $\phi_i$, integrating over $\Omega_e$, and applying integration by parts leads to
\[eq:disc\] $$\begin{aligned}
{2}
\label{eq:discU}
& \frac{h}{2} \int_{\Omega_{\textnormal{ref}}}{\frac{\partial u_h}{\partial t} \phi_i} + \left( \widehat{f}_h \phi_i \right)^{\oplus}_{\ominus} = \int_{\Omega_{\textnormal{ref}}}{f \frac{\partial \phi_i}{\partial \xi}} \mbox{ ,} \\
\label{eq:discG}
& \frac{h}{2} \int_{\Omega_{\textnormal{ref}}}{q_h \phi_i} + \int_{\Omega_{\textnormal{ref}}}{u_h \frac{\partial \phi_i}{\partial \xi}} = \big( \widehat{u}_h \phi_i \big)^\oplus_\ominus \mbox{ ,} \end{aligned}$$
where the symbols $\ominus$ and $\oplus$ denote the left and right boundaries of $\Omega_e$, respectively. As customary in the context of DG methods, expressions are inserted into ; which are then required to hold for $i = 0, \dots, p$. Note that we have introduced the interface quantities $\widehat{f}_h$ and $\widehat{u}_h$. The former is the so-called interelement flux, interface flux or numerical flux, and appears in standard DG methods as well. The latter is particular of hybridized DG methods and is an approximation for the solution $u$ on the element faces that takes the same value on the two elements neighboring the considered interface.
In order to complete the definition of the hybridized DG scheme, it remains to define the numerical flux $\widehat{f}_h$ and enforce its continuity from the left ($L$) to the right ($R$) elements sharing the interface. For convection-diffusion systems, the numerical fluxes are usually defined as [@Nguyen:09; @Peraire:10]
$$\begin{aligned}
{2}
\label{eq:fluxmi}
\widehat{f}_{h,\ominus} = f(\widehat{u}_{h,\ominus}, q_{h,\ominus}) - \sigma (u_{h,\ominus} - \widehat{u}_{h,\ominus}) \mbox{ ,} \\
\label{eq:fluxpl}
\widehat{f}_{h,\oplus} = f(\widehat{u}_{h,\oplus}, q_{h,\oplus}) + \sigma (u_{h,\oplus} - \widehat{u}_{h,\oplus}) \mbox{ ,}\end{aligned}$$
where $\sigma = \beta |a|$ is the stabilization constant and $\beta \geq 0$ the upwinding parameter. Note that no explicit stabilization is used for the diffusive term. Since hybridized DG methods have some form of built-in stabilization for second-order operators [@Fernandez:PhD:2018], this choice of $\sigma$ is customary for high Reynolds number flows [@Fernandez:17a; @Nguyen:2011; @Peraire:10] and has been adopted here for consistency with the literature. The flux continuity condition is then given by $$\label{eq:fluxcont}
\widehat{f}_{h,\oplus}^L = \widehat{f}_{h,\ominus}^R \mbox{ .}$$ We note that Eq. ensures local conservation regardless of the chosen numerical flux formula. Also, for pure convection and our choice of numerical fluxes, it follows that $\widehat{u}_h = (u_{h,\oplus}^L + u_{h,\ominus}^R) / 2$ and, furthermore, hybridized and standard DG methods lead to the same numerical solution (see Appendix \[s:connectionStHyDG\]). This does not hold, however, when diffusion is taken into account. In this case $\widehat{u}_h$ is only given implicitly from the flux continuity at interfaces, namely, $$\label{eq:fluxcont2}
a \widehat{u}_h - \nu q_{h,\oplus}^L + \sigma ( u_{h,\oplus}^L - \widehat{u}_{h} ) = a \widehat{u}_h - \nu q_{h,\ominus}^R - \sigma ( u_{h,\ominus}^R - \widehat{u}_h ) \mbox{ ,}$$ where $q_{h,\oplus}^L$ and $q_{h,\ominus}^R$ in turn depend on the values of $\widehat{u}_h$ at two other interfaces via .
To simplify the analysis, we rewrite the hybridized DG discretization in matrix notation. To that end, we note that
$$\begin{aligned}
{2}
\label{eq:minusUG}
u_{h,\ominus} = \sum_{j=0}^{p}{\tilde{u}_{h,j} \, \phi_j(-1)} \mbox{ ,} \quad q_{h,\ominus} = \sum_{j=0}^{p}{\tilde{q}_{h,j} \, \phi_j(-1)} \mbox{ ,} \\
\label{eq:plusUG}
u_{h,\oplus} = \sum_{j=0}^{p}{\tilde{u}_{h,j} \, \phi_j(+1)} \mbox{ ,} \quad q_{h,\oplus} = \sum_{j=0}^{p}{\tilde{q}_{h,j} \, \phi_j(+1)} \mbox{ ,}\end{aligned}$$
and introduce the vectors $\tilde{u}_h = \{ \tilde{u}_0, \dots, \tilde{u}_p \}^T$, $\tilde{q} = \{ \tilde{q}_0, \dots, \tilde{q}_p \}^T$, $\tilde{\phi}_\oplus = \{ \tilde{\phi}_0(+1), \dots, \tilde{\phi}_p(+1) \}^T$ and $\tilde{\phi}_\ominus = \{ \tilde{\phi}_0(-1), \dots, \tilde{\phi}_p(-1) \}^T$. The flux continuity condition then can be expressed as $$\label{eq:III}
\widehat{u}_h = \frac{1}{2} \left( \tilde{\phi}^T_{\oplus} \tilde{u}_h^L + \tilde{\phi}^T_\ominus \tilde{u}_h^R \right) + \frac{\nu}{2 \sigma} \left( \tilde{\phi}^T_\ominus \tilde{q}_h^R - \tilde{\phi}^T_\oplus \tilde{q}_h^L \right) \mbox{ .}$$ Likewise, the auxiliary equation can be written as $$\label{eq:II}
\frac{h}{2} M \tilde{q}_h + D \tilde{u}_h = \tilde{\phi}_\oplus \widehat{u}_{h,\oplus} - \tilde{\phi}_\ominus \widehat{u}_{h,\ominus} \mbox{ ,}$$ where $M$ and $D$ are the mass and convection matrices defined as $$\label{eq:matsMD}
M_{ij} = \int_{\Omega_{\textnormal{ref}}} \phi_i \, \phi_j \mbox{ ,} \qquad \qquad D_{ij} = \int_{\Omega_{\textnormal{ref}}} \frac{\partial \phi_i}{\partial \xi} \, \phi_j \mbox{ .}$$ Finally, Eq. becomes $$\label{eq:I}
\frac{h}{2} M \frac{d \tilde{u}_h}{d t} + \tilde{\phi}_\oplus \widehat{f}_{h,\oplus} - \tilde{\phi}_\ominus \widehat{f}_{h,\ominus} = a D \tilde{u}_h - \nu D \tilde{q}_h \mbox{ ,}$$ with
\[eq:fluxI\] $$\label{eq:fluxImi}
\widehat{f}_{h,\ominus} = a \widehat{u}_{h,\ominus} - \nu \tilde{\phi}^T_\ominus \tilde{q}_h - \sigma ( \tilde{\phi}^T_\ominus \tilde{u}_h - \widehat{u}_{h,\ominus} ) \mbox{ .}$$ $$\label{eq:fluxIpl}
\widehat{f}_{h,\oplus} = a \widehat{u}_{h,\oplus} - \nu \tilde{\phi}^T_\oplus \tilde{q}_h + \sigma ( \tilde{\phi}^T_\oplus \tilde{u}_h - \widehat{u}_{h,\oplus} ) \mbox{ ,}$$
Note that is a scalar equation written from the point of view of a given interface, whereas and are vector equations written from the viewpoint of an arbitrary element $\Omega_e$ of size $h$.
At this point, it is convenient to eliminate $\tilde{q}_h$ from the formulation and work with the variables $\tilde{u}_h$ and $\widehat{u}_h$ only. To this end, we use to obtain $\tilde{q}_h$ as a function of $\tilde{u}_h$ and $\widehat{u}_h$, and then substitute the resulting expression into and . The former substitution leads, after some algebra, to $$\label{eq:star}
\left( \bar{\sigma} + \frac{m_\ominus^\ominus}{\mbox{Pe}} + \frac{m_\oplus^\oplus}{\mbox{Pe}} \right) \widehat{u}_h - \frac{m_\oplus^\ominus}{\mbox{Pe}} \widehat{u}_{h,\ominus}^L - \frac{m_\ominus^\oplus}{\mbox{Pe}} \widehat{u}_{h,\oplus}^R = \tilde{\phi}^T_\oplus B_\oplus \tilde{u}_h^L + \tilde{\phi}^T_\ominus B_\ominus \tilde{u}_h^R \mbox{ ,}$$ where $\bar{\sigma} = \sigma / |a|$ is a non-dimensional stabilization parameter and $\mbox{Pe}$ denotes the cell Péclet number $\mbox{Pe} = |a| h / \nu$. Moreover, the following scalar constants ‘$m$’ $$\label{eq:emes}
m_{\ominus}^{\ominus} = \tilde{\phi}_{\ominus}^T M^{-1} \tilde{\phi}_{\ominus} \mbox{ ,} \quad
m_{\oplus}^{\ominus} = \tilde{\phi}_{\oplus}^T M^{-1} \tilde{\phi}_{\ominus} \mbox{ ,} \quad
m_{\ominus}^{\oplus} = \tilde{\phi}_{\ominus}^T M^{-1} \tilde{\phi}_{\oplus} \mbox{ ,} \quad
m_{\oplus}^{\oplus} = \tilde{\phi}_{\oplus}^T M^{-1} \tilde{\phi}_{\oplus} \mbox{ .}$$ and matrices $$B_{\ominus} = \bigg( \frac{\bar{\sigma}}{2} \, I - \frac{M^{-1} D}{\textnormal{Pe}} \bigg) \mbox{ ,} \qquad \qquad B_{\oplus} = \bigg( \frac{\bar{\sigma}}{2} \, I + \frac{M^{-1} D}{\textnormal{Pe}} \bigg) \mbox{ ,}$$ have been introduced in . Note that Eq. links the solution vectors $\tilde{u}_h$ on two adjacent elements ($\Omega_L$ and $\Omega_R$) with the three interface variables $\widehat{u}_h$ corresponding to the boundaries of these elements.
The second substitution, namely inserting $\tilde{q}_h$ from into , and using also Equation for the numerical fluxes, yields $$\label{eq:aster}
\frac{h}{2 a} M \frac{d \tilde{u}_h}{d t} + A \tilde{u}_h = A_\ominus \tilde{\phi}_\ominus \widehat{u}_{h,\ominus} + A_\oplus \tilde{\phi}_\oplus \widehat{u}_{h,\oplus} \mbox{ ,}$$ where
$$\label{eq:matA1}
A = \bar{\sigma} \, \big( \Phi_{\ominus}^{\ominus} + \Phi_{\oplus}^{\oplus} \big) + \bigg( \frac{2N}{\textnormal{Pe}} - I \bigg) \, D \mbox{ ,}$$
$$\label{eq:matA2}
A_{\ominus} = (\bar{\sigma} + 1) \, I - \frac{2 N}{\textnormal{Pe}} \mbox{ ,} \qquad \qquad A_{\oplus} = (\bar{\sigma} - 1) \, I + \frac{2 N}{\textnormal{Pe}} \mbox{ ,}$$
and
$$\label{eq:matsPhi}
\Phi_{\ominus}^{\ominus} = \tilde{\phi}_{\ominus} \tilde{\phi}_{\ominus}^T \mbox{ ,} \qquad \qquad \Phi_{\oplus}^{\oplus} = \tilde{\phi}_{\oplus} \tilde{\phi}_{\oplus}^T \mbox{ ,}$$
$$\label{eq:matN}
N = \big( \Phi_{\oplus}^{\oplus} - \Phi_{\ominus}^{\ominus} - D \big) \, M^{-1} \mbox{ .}$$
Note that Eq. links the solution vector $\tilde{u}_h$ and its time derivative to the two interface variables $\widehat{u}_h$ at the boundaries of the considered element.
The hybridized DG discretization of the linear convection-diffusion equation in matrix notation is given by Equations , and ; which are required to hold in all elements and all faces. Equations and are an equivalent formulation in terms of $\tilde{u}_h$ and $\widehat{u}_h$ only. Both formulations need to be further equipped with the initial condition $\tilde{u}_h(t=0) = \tilde{u}_{h,0}$, the discretized version of Eq. , where the right-hand side is the vector of coefficients of the Galerkin projection of $u_0$ and is given by $$\tilde{u}_{h,0} = M^{-1} d \mbox{ ,}$$ where we have introduced the vector $$d_j = \int_{\Omega_{\textnormal{ref}}} u_0 \, \phi_j \mbox{ .}$$ We note that all the methods within the hybridized DG family, including HDG, EDG and IEDG, reduce to the same scheme in one-dimensional problems, and therefore no difference between them has been made here.
Non-modal analysis formulation
------------------------------
We are concerned about the short-term diffusion properties, in wavenumber space, of the hybridized DG discretization of . That is, if the initial condition is a single Fourier mode $u_0 \propto \exp(i \kappa x)$, where $\kappa \in \mathbb{R}$ denotes the wavenumber, how does the magnitude of the numerical solution evolve over time, and in particular right after $t = 0$? To this end, we define the short-term diffusion as $$\label{e:lambdaDef}
\varpi^* := \frac{d \log {\left|\left|u_h\right|\right|}}{d \tau^*} \bigg |_{\tau^*=0} \mbox{ ,}$$ where $\tau^* = \tau \, (p+1) = t \, a \, (p+1) / h$ is a non-dimensional time based on the convection time between degrees of freedom. Note that we define the distance between degrees of freedom as $h^* = h / (p+1)$ and that $\tau^* = 1$ represents the time it takes for a signal convected with speed $a$ to cross a single DOF. The $*$ superscript, such as in $\varpi^*$, $\tau^*$ and $h^*$, is used to indicate that a $(p+1)$ factor has been applied to account for the $p+1$ degrees of freedom per element. Also, we note that Eq. can be rewritten as $$\label{e:lambdaDef_v2}
\varpi^* = \lim_{\tau^* \downarrow 0} \, \frac{1}{\tau^*} \log \Bigg( \frac{{\left|\left|u_h\right|\right|}}{{\left|\left|u_{h,0}\right|\right|}} \Bigg) \mbox{ ,}$$ which some readers may find easier to interpret.
As we shall see, $\varpi^*$ depends on the wavenumber $\kappa$, the modified Péclet number $\textnormal{Pe}^* = |a| \, h^* / \nu = \textnormal{Pe} / (p+1)$ and the details of the hybridized DG scheme, such as the polynomial order and the upwinding parameter. Intuitively, $\varpi^*$ informs of the relative magnitude growth in the numerical solution ([*relative*]{} with respect to its initial magnitude, thereby the $\log$ in ) per [*unit convection time between degrees of freedom*]{} (thereby the use of $\tau^*$ in the derivative) at early times (thereby the $|_{\tau^* = 0}$) , starting from the initial condition $\exp(\kappa x)$. In particular, ${\left|\left|u_h(\tau^*)\right|\right|} \approx {\left|\left|u_{h,0}\right|\right|} \exp(\varpi^* \tau^*)$ at early times, and thus $\exp(\varpi^*)$ can be considered as a [*damping factor per DOF crossed*]{}. Next, we derive an explicit expression for $\varpi^*$. First, it can be shown (see footnote [^6]) that if $u_0 \propto \exp(i \kappa x)$, then the relations
\[e:waveLike\] $$\label{e:waveLike_u}
\tilde{u}_h^L = \tilde{u}_h \exp(-i \kappa h) \mbox{ ,} \qquad \qquad \tilde{u}_h^R = \tilde{u}_h \exp(+i \kappa h) \mbox{ ,}$$ $$\label{e:waveLike_uHat}
\widehat{u}_{h,\ominus} = \widehat{u}_{h,\oplus} \exp(-i \kappa h) \mbox{ ,}$$
hold for all elements and all times. Similarly to the notation adopted before for the elements neighboring an interface, we use the superscripts $L$ and $R$ in to denote the left and right neighboring elements of a given element.
The wave-like behavior of the numerical solution allows to reduce the dimensionality of the problem from countably many (infinite) degrees of freedom to $p+1$ degrees of freedom, and this in turn makes our non-modal analysis possible. In particular, it now follows from Equations and that $$\label{eq:starb}
\widehat{u}_h = \left( \tilde{\phi}^T_\oplus B_\oplus \tilde{u}_h^L + \tilde{\phi}^T_\ominus B_\ominus \tilde{u}_h^R \right) b^{-1} \mbox{ ,}$$ where $b = b (\kappa h; \mbox{Pe}, p, \bar{\sigma})$ is a scalar defined as $$\label{eq:scalarb}
b = \bar{\sigma} + \left( m_{\ominus}^{\ominus} - m_{\oplus}^{\ominus} \exp(- i \kappa h) - m_{\ominus}^{\oplus} \exp(+i \kappa h) + m_{\oplus}^{\oplus} \right) \, \textnormal{Pe}^{-1} \mbox{ .}$$ Inserting and into , one finally obtains $$\label{e:discrete_cd_Fourier_1}
\frac{h}{a} \, \frac{d \tilde{u}_h}{dt} = \tilde{Z}_h \tilde{u}_h ,$$ where $\tilde{Z}_h = \tilde{Z}_h (\kappa h; \mbox{Pe}^*, p, \bar{\sigma})$ is a square matrix given by $$\label{e:Zh_tilde}
\tilde{Z}_h = 2b^{-1} M^{-1} \big( A_{\ominus} \Phi_{\ominus}^{\ominus} B_{\ominus} + A_{\ominus} \Phi_{\oplus}^{\ominus} B_{\oplus} \exp(- i \kappa h) + A_{\oplus} \Phi_{\ominus}^{\oplus} B_{\ominus} \exp(+i \kappa h) + A_{\oplus} \Phi_{\oplus}^{\oplus} B_{\oplus} - A b \big) \mbox{ ,}$$ with $\Phi_\ominus^\ominus$ and $\Phi_\oplus^\oplus$ are given by , and $$\label{eq:matsPhiNew}
\Phi_{\oplus}^{\ominus} = \tilde{\phi}_{\ominus} \tilde{\phi}_{\oplus}^T \mbox{ ,} \qquad \qquad \Phi_{\ominus}^{\oplus} = \tilde{\phi}_{\oplus} \tilde{\phi}_{\ominus}^T \mbox{ .}$$ Since $\varpi^*$ is independent of the choice of basis, we assume that $\phi_j$ is the orthonormal Legendre polynomial of degree $j$ in $\Omega_{\textnormal{ref}} = [-1,1]$; in which case we can obtain a closed-form expression for $\varpi^*$. Combining Equations and , using inner product properties, orthonormality of Legendre polynomials and the wave-like behavior of the numerical solution, it follows that $$\label{e:lambda_expr}
\begin{split}
\varpi^* = & \, \frac{d \log {\left|\left|u_h\right|\right|}}{d \tau^*} \bigg |_{\tau^*=0} = \frac{1}{{\left|\left|u_h\right|\right|}} \, \frac{d {\left|\left|u_h\right|\right|}}{d \tau^*} \bigg |_{\tau^*=0} = \frac{h^*}{a} \, \frac{1}{(\tilde{u}_{h}^{\dagger} \tilde{u}_{h})^{1/2}} \, \frac{d \, (\tilde{u}_{h}^{\dagger} \tilde{u}_{h})^{1/2}}{d t} \bigg |_{t=0} \\
= & \, \frac{h^*}{2 a} \, \frac{1}{\tilde{u}_{h}^{\dagger} \tilde{u}_{h}} \, \bigg( \frac{d \tilde{u}_h^{\dagger}}{d t} \tilde{u}_h + \tilde{u}_h^{\dagger} \frac{d \tilde{u}_h}{d t} \bigg) \bigg |_{t=0} = \frac{1}{2} \, \frac{1}{p+1} \, \frac{u_{h,0}^{\dagger} \, \tilde{Z}_h^{\dagger} \, u_{h,0} + u_{h,0}^{\dagger} \, \tilde{Z}_h \, u_{h,0}}{\tilde{u}_{h,0}^{\dagger} \, \tilde{u}_{h,0}} \\
= & \, \frac{1}{p+1} \, \mathbb{R}e \Bigg[ \frac{\tilde{u}_{h,0}^{\dagger} \, \tilde{Z}_h \, \tilde{u}_{h,0}}{\tilde{u}_{h,0}^{\dagger} \, \tilde{u}_{h,0}} \Bigg] \mbox{ .}
\end{split}$$ where the $\dagger$ superscript denotes conjugate transpose and $\mathbb{R}e$ the real part of a complex number. Note that the value of $\varpi^*$ is independent of the amplitude of the Fourier mode. Taking $u_0 = \exp(i \kappa x)$, it follows that $\tilde{u}_{h,0} = \alpha = \alpha(\kappa h)$ with
\[e:alpha\] $$\begin{aligned}
{2}
\label{e:alpha_1}
\alpha_0 = \sqrt{2} \, \frac{\sin z}{z} \mbox{ ,} \qquad \qquad \alpha_1 = \frac{i \sqrt{6}}{z} \bigg( \frac{\sin z}{z} - \cos z \bigg) \mbox{ ,} \\
\intertext{and, for $j \geq 1$,}
\label{e:alpha_2}
\alpha_{j+1} = \frac{\sqrt{4 j+6}}{z} \, \bigg( M_j \sin z + i \bigg( \frac{\sin z}{z} - \cos z \bigg) M_{j+1} + i \sum_{k=1}^j \sqrt{k+1/2} M_{j+k+1} \, \alpha_k \bigg) \mbox{ ,}\end{aligned}$$
where $z = \kappa h / 2$ and $M_j = \textnormal{mod}(j,2)$ is the modulus of $j$ after division by two. Equations , and provide a closed-form expression for $\varpi^* = \varpi^* (\kappa h; \mbox{Pe}^*, p, \bar{\sigma})$.
We note that $\tilde{Z}_h$ is the same matrix (up to complex sign) as that in eigenanalysis [@Moura:18a]. The difference between eigenanalysis and non-modal analysis is that the former concerns about the eigenvalues and eigenvectors of $\tilde{Z}_h$, whereas non-modal analysis concerns about the non-modal behavior of $\tilde{Z}_h$, and in particular about its short-term dynamics. Also, if the Fourier mode is an eigenmode of $\tilde{Z}_h$, as it is the case for $p=0$, then $(p+1)\, \varpi^*$ coincides with the real part of the corresponding eigenvalue, and non-modal analysis reconciles with eigenanalysis.
[**Remark:**]{} Strictly speaking, Equations $-$ should read as $$\varpi^* := \lim_{n \to \infty} \frac{d \log {\left|\left|\chi_{[-n,n]} \, u_h\right|\right|}}{d \tau^*} \Bigg |_{\tau^*=0} \mbox{ ,} \qquad \qquad \varpi^* = \lim_{n \to \infty} \lim_{\tau^* \downarrow 0} \frac{1}{\tau^*} \log \Bigg( \frac{{\left|\left|\chi_{[-n,n]} \, u_h\right|\right|}}{{\left|\left|\chi_{[-n,n]} \, u_{h,0}\right|\right|}} \Bigg) \mbox{ ,}$$ respectively, where $\chi_{[-n,n]}$ denotes the indicator function in $[-n,n]$. For any $u_0 \propto \exp(i \kappa x)$ such as in our analysis, the limits are well-defined and equal to the right-hand sides of Equations $-$. A limiting process is also required, and intentional abuse of notation is used, whenever ${\left|\left|u_h\right|\right|}$ is written outside of a logarithm.
Non-modal analysis results
--------------------------
We present the non-modal analysis results through the so-called short-term diffusion curves. For given $\textnormal{Pe}^*$, $p$ and $\beta$, the short-term diffusion curves show $\varpi^*$ ($y$ axis) as a function of the non-dimensional wavenumber $\bar{\kappa}^* = \kappa h^* = \kappa h / (p+1)$ ($x$ axis). The right limit of the $x$ axis corresponds to the Nyquist wavenumber of the grid, defined as $\kappa_N = \pi / h^*$, and thus $\bar{\kappa}_N^* = \pi$. The exact dissipation curves will be indicated with dashed lines.
Before presenting the short-term diffusion curves, we briefly discuss on how these curves [*should look like*]{} from the perspectives of robustness and accuracy. For robustness purposes, monotonic ($d \varpi^* / d \kappa \leq 0$) and slowly-varying curves are preferred as they lead to more stable discretizations, particularly for nonlinear systems due to the nonlinear interactions between wavenumbers. Under-resolved simulations of systems featuring a kinetic energy cascade, such as turbulent flows, additionally require the dissipation near the Nyquist wavenumber to be [*sufficiently large*]{} to avoid energy accumulation at large wavenumbers. Regarding accuracy, the short-term dissipation should agree with the exact dissipation as closely as possible. Like for robustness purposes, however, some additional dissipation near the Nyquist wavenumber is desired for under-resolved simulations of systems with an energy cascade in order to account for the dissipation in the subgrid scales. These considerations are based on [*a priori*]{} knowledge and [*a posteriori*]{} insights from the numerical results in Section \[s:applicationNonLinear\].
### Effect of the polynomial order
Figure \[f:shortTimeDiffusionCurves\_pStudy\] shows the short-term diffusion curves as a function of the polynomial order for standard upwinding $\beta = 1$ in convection-dominated $\textnormal{iPe}^* = 0.001$ (left) and diffusion-dominated $\textnormal{iPe}^* = 10$ (right) regimes, where $\textnormal{iPe}^* = (\textnormal{Pe}^*)^{-1}$ denotes the inverse of the Péclet number. The exact solution of the convection-diffusion equation is shown in dashed black line. For convection-dominated flows, high polynomial orders lead to non-monotonic short-term diffusion characteristics in wavenumber space. In particular, very small dissipation is introduced at some specific wavenumbers. As discussed before and shown by the numerical results in Sections \[s:burgers\] and \[s:TGV\], this may potentially lead to nonlinear instability. Also, we recall that some amount of numerical dissipation near $\kappa_N$ (preferably monotonic in wavenumber space) is desired (both for accuracy and robustness) in under-resolved turbulence simulations in order to replicate the dissipation that takes place in the subgrid scales. The short-term diffusion properties that are better suited, both in terms of accuracy per DOF and robustness, for convection-dominated under-resolved turbulence simulations, seem to be those for polynomial orders $p=2$, $3$ and $4$. We note that convection-dominated, from the cell Péclet number perspective, is the regime featured by most under-resolved turbulent flows in practical applications. For diffusion-dominated problems, however, higher $p$ improves both accuracy per degree of freedom and robustness.
![\[f:shortTimeDiffusionCurves\_pStudy\] Short-term diffusion curves as a function of the polynomial order for $\beta = 1$ in convection-dominated $\textnormal{iPe}^* = 0.001$ (left) and diffusion-dominated $\textnormal{iPe}^* = 10$ (right) regimes. The exact solution of the convection-diffusion equation is shown in dashed black line. Note a different scale in the $y$ axis is used for each figure. ](shortTimeDiffusionCurves_pStudy_iPe0.eps "fig:"){width="45.00000%"} ![\[f:shortTimeDiffusionCurves\_pStudy\] Short-term diffusion curves as a function of the polynomial order for $\beta = 1$ in convection-dominated $\textnormal{iPe}^* = 0.001$ (left) and diffusion-dominated $\textnormal{iPe}^* = 10$ (right) regimes. The exact solution of the convection-diffusion equation is shown in dashed black line. Note a different scale in the $y$ axis is used for each figure. ](shortTimeDiffusionCurves_pStudy_iPe1e1.eps "fig:"){width="45.00000%"}
### Effect of the Péclet number
Figure \[f:shortTimeDiffusionCurves\_iPeStudy\] shows the short-term diffusion curves as a function of the Péclet number for standard upwinding $\beta = 1$ with polynomial orders $p=1$ (left) and $p=6$ (right). The exact solution of the convection-diffusion equation is shown in dashed lines. As noted in the polynomial order study, high $p$ schemes are better suited to diffusion-dominated problems, both in terms of accuracy and robustness. As for low polynomial orders, moderately high wavenumbers are poorly resolved regardless of the Péclet number. Robustness of low $p$ schemes seems to improve in the convection-dominated regime.
![\[f:shortTimeDiffusionCurves\_iPeStudy\] Short-term diffusion curves as a function of the Péclet number for standard upwinding $\beta = 1$ and polynomial orders $p=1$ (left) and $p=6$ (right). The exact solution of the convection-diffusion equation is shown in dashed lines. ](shortTimeDiffusionCurves_iPeStudy_p1.eps "fig:"){width="45.00000%"} ![\[f:shortTimeDiffusionCurves\_iPeStudy\] Short-term diffusion curves as a function of the Péclet number for standard upwinding $\beta = 1$ and polynomial orders $p=1$ (left) and $p=6$ (right). The exact solution of the convection-diffusion equation is shown in dashed lines. ](shortTimeDiffusionCurves_iPeStudy_p6.eps "fig:"){width="45.00000%"}
### Effect of the upwinding parameter
Figure \[f:shortTimeDiffusionCurves\_betaStudy\] shows the short-term diffusion curves as a function of the upwinding parameter for $p=3$ in convection-dominated $\textnormal{iPe}^* = 0.001$ (left) and diffusion-dominated $\textnormal{iPe}^* = 10$ (right) regimes. The exact solution of the convection-diffusion equation is shown in dashed black line. Both under- and over-upwinding may lead to accuracy and robustness issues in the convection-dominated regime. For diffusion-dominated problems, however, the scheme benefits form over-upwinding; which is no completely surprising since no explicit stabilization is used for the diffusion operator.
![\[f:shortTimeDiffusionCurves\_betaStudy\] Short-term diffusion curves as a function of the upwinding parameter for $p = 3$ in convection-dominated $\textnormal{iPe}^* = 0.001$ (left) and diffusion-dominated $\textnormal{iPe}^* = 10$ (right) regimes. The exact solution of the convection-diffusion equation is shown in dashed black line. Note a different scale in the $y$ axis is used for each figure.](shortTimeDiffusionCurves_betaStudy_iPe1eMinus3.eps "fig:"){width="45.00000%"} ![\[f:shortTimeDiffusionCurves\_betaStudy\] Short-term diffusion curves as a function of the upwinding parameter for $p = 3$ in convection-dominated $\textnormal{iPe}^* = 0.001$ (left) and diffusion-dominated $\textnormal{iPe}^* = 10$ (right) regimes. The exact solution of the convection-diffusion equation is shown in dashed black line. Note a different scale in the $y$ axis is used for each figure.](shortTimeDiffusionCurves_betaStudy_iPe1e1.eps "fig:"){width="45.00000%"}
\[s:applicationNonLinear\]Application to nonlinear systems
==========================================================
In order to assess how the non-modal analysis results extend to the nonlinear setting, we apply it to the Burgers, Euler and Navier-Stokes equations.
\[s:burgers\]Application to the Burgers equation
------------------------------------------------
### Problem description
We consider the following forced Burgers turbulence problem [@Chekhlov:1995]
\[e:burgersTurb\] $$\begin{aligned}
{2}
\label{e:burgersTurb1}
& \frac{\partial u}{\partial t} + \frac{1}{2} \frac{\partial u^2}{\partial x} = \frac{A_F}{\sqrt{\Delta t}} \sum_{N \in \mathbb{N}_F} \frac{\sigma_N(t)}{\sqrt{|N|}} \exp \bigg( i \frac{2 \pi N}{L} x \bigg) , \qquad & t > 0 , \\
& u|_{x=-L/2} = u|_{x=L/2} , \qquad & t \geq 0 , \\
& u = u_0 , \qquad & t = 0 , \end{aligned}$$
where $u_0 > 0$ denotes the initial velocity (constant in the domain), $L$ is the length of the computational domain $\Omega$, $\Delta t$ is the time-step size used in the simulation, $A_F$ is an amplitude constant for the forcing term, $\mathbb{N}_F = \{ \pm 1 , \dots , \pm N_c \}$ is a collection of integers, and $\sigma_{N}$ is a standard Gaussian random variable (zero mean and unit variance) that is independent for each wavenumber and each time step. We set $N_c=80$ and $A_F = \sqrt{8} \cdot 10^{-1} \, u_0^{3/2} \, L^{-1/2}$ for the numerical experiments in this section. This completes the non-dimensional description of the problem. The choice of forcing in yields, for wavenumbers below the cut-off wavenumber $\kappa_c = 2 \pi N_c / L$, a $-5/3$ slope for the inertial range of the energy spectrum [@Adams:2004; @Chekhlov:1995; @Moura:15a; @Zikanov:1997] and thus resembles Navier-Stokes turbulence within the Burgers setting. As customary in the literature, we use the term Burgers [*turbulence*]{} to refer to the chaotic and turbulent-like behavior featured by the solution of the Burgers equation.
### Details of the numerical discretization
We use the hybridized DG method with various polynomial orders to discretize Eq. in space. We recall that HDG, EDG, IEDG and all other schemes within the hybridized DG family reduce to the same scheme in one-dimensional problems, and there is only one type of hybridized DG method for this problem. We refer the interested reader to [@Nguyen:09:nonlinear] for the details of the hybridized DG discretization of the one-dimensional Burgers equation. The upwinding parameter is $\beta = 1$ (and thus $\sigma = |u|$), and the total number of degrees of freedom is $N_{DOF} = (p+1) \cdot \lceil 1024 / (p+1) \rceil \approx 1024$, where $\lceil \ \ \rceil$ denotes the rounding of a positive real number to the closest larger (or equal) integer. Note this is required in order to obtain an integer number of elements in $\Omega$. Exact integration is used both for the Burgers flux and the forcing term. For the former, Gauss-Legendre quadrature with the required number of points to ensure exact integration of polynomials of degree $3 \, p$ (and thus of the Burgers flux term in the hybridized DG discretization) is used. The Galerkin projection of the forcing term is integrated exactly using the analytical expressions in [@Moura:15a]. The third-order, three-stage $L$-stable diagonally implicit Runge-Kutta DIRK(3,3) scheme [@Alexander:77] is used for the temporal discretization with CFL number based on the initial velocity of $\textnormal{CFL} = u_0 \, \Delta t / h^* = 0.05$. The solution is computed from $t = 0$ to $t = 8 \, L / u_0$.
### Numerical results
The time-averaged kinetic energy spectra from $t = 2 \, L / u_0$ to $t = 8 \, L / u_0$ for $p = 1 , \dots , 7$ are shown on the top of Figure \[f:spectraBurgers\]. Polynomial orders $p \geq 8$ failed to converge for this problem. The spectra are shifted up by a multiplicative factor of $4^{p-1}$ to allow for easier visualization. We note that the mean of $u(t)$ in the computational domain remains constant, and in particular equal to $u_0$, throughout the simulation since neither the forcing nor the nonlinear dynamics of the Burgers equation affect the average value (component $\kappa = 0$ of the spectrum) of the solution. All the spectra feature an inertial range of turbulence with slope $-5/3$ up to $\log_{10} (\kappa_c \, L) = \log_{10} (2 \pi N_c) \approx 2.7$, as expected from the forcing strategy adopted. After the cut-off wavenumber, a slope of $-2$, typical of unforced Burgers turbulence [@Bec:2007], takes place whenever numerical dissipation is still small enough over these wavenumbers. In all the simulations, numerical dissipation eventually becomes significant and affects the shape of the energy spectra near the Nyquist wavenumber of the grid $\kappa_N \, L = \pi N_{DOF} \approx 3217$; which corresponds to the right limit of the $x$ axis. The short-term diffusion curves from non-modal analysis are shown in the bottom of Figure \[f:spectraBurgers\], where the $x$ axis has been mapped from $\kappa \, h / (p+1)$ (as in Figures \[f:shortTimeDiffusionCurves\_pStudy\]$-$\[f:shortTimeDiffusionCurves\_betaStudy\]) to $\kappa \, L$ to facilitate the comparison with the energy spectra. Note that $\textnormal{iPe}^* = 0$ in this problem due to the lack of physical viscosity. The trends observed in the energy spectra are consistent with non-modal analysis results. First, a numerically induced dissipation range near $\kappa_N$ is observed in the spectrum of the $p = \{ 1 , 2 \}$ and, to a lessen extent, the $p=3$ discretizations; which is consistent with the large short-term dissipation of these schemes right before the Nyquist wavenumber. Second, bottlenecks in the turbulence cascade (in the sense of energy accumulation at some specific wavenumbers) are observed for the high $p$ discretizations; which is consistent with the non-monotonicity in the short-term diffusion curves.
![\[f:spectraBurgers\] Results for the Burgers turbulence problem. Top: Time-averaged energy spectra from $t = 2 \, L / u_0$ to $t = 8 \, L / u_0$ for $p = 1 , \dots , 7$. The spectra are shifted up by a multiplicative factor of $4^{p-1}$ to allow for easier visualization. Bottom: Short-term diffusion curves from non-modal analysis. The right limit of the $x$ axis corresponds to the Nyquist wavenumber of the grid. ](burgersSpectra.eps "fig:"){width="100.00000%"} ![\[f:spectraBurgers\] Results for the Burgers turbulence problem. Top: Time-averaged energy spectra from $t = 2 \, L / u_0$ to $t = 8 \, L / u_0$ for $p = 1 , \dots , 7$. The spectra are shifted up by a multiplicative factor of $4^{p-1}$ to allow for easier visualization. Bottom: Short-term diffusion curves from non-modal analysis. The right limit of the $x$ axis corresponds to the Nyquist wavenumber of the grid. ](shortTimeDiffusionBurgers_v2.eps "fig:"){width="100.00000%"}
\[s:TGV\]Application to the Euler and Navier-Stokes equations. The Taylor-Green vortex
--------------------------------------------------------------------------------------
### Problem description
The Taylor-Green vortex (TGV) problem [@Taylor:37] describes the evolution of the fluid flow in a cubic domain $\Omega = [-L \pi, L \pi)^3$ with triple periodic boundaries, starting from the smooth initial condition $$\label{initialCondTGV}
\begin{split}
\rho & = \rho_0 , \\
u_1 & = U_0 \sin \Big( \frac{x}{L} \Big) \cos \Big( \frac{y}{L} \Big) \cos \Big( \frac{z}{L} \Big) , \\
u_2 & = - U_0 \cos \Big( \frac{x}{L} \Big) \sin \Big( \frac{y}{L} \Big) \cos \Big( \frac{z}{L} \Big) , \\
u_3 & = 0 , \\
p & = P_0 + \frac{\rho_0 \, U_0^2}{16} \, \bigg( \cos \Big( \frac{2x}{L} \Big) + \cos \Big( \frac{2y}{L} \Big) \bigg) \, \bigg( \cos \Big( \frac{2z}{L} \Big) + 2 \bigg) ,
\end{split}$$ where $\rho$, $p$ and $(u_1, u_2, u_3)$ denote density, pressure and the velocity vector, respectively. Governed by the Navier-Stokes equations (Euler equations in the inviscid case), the large-scale eddy in the initial condition leads to smaller and smaller structures through vortex stretching. For Reynolds numbers $\textnormal{Re} = U_0 \, L / \nu$ below about $1000$, the flow remains laminar at all times [@Brachet:91]. Above this threshold, the vortical structures eventually break down and the flow transitions to turbulence[^7]. After transition, the turbulent motion dissipates all the kinetic energy, and the flow eventually comes to rest through a decay phase similar to that in decaying homogeneous isotropic turbulence, yet not isotropic here. In the high Reynolds number limit (inviscid TGV), there is no decay phase due to the lack of viscosity and the smallest turbulent scales thus become arbitrarily small as time evolves.
In order to investigate different Péclet numbers and flow regimes, we consider the Reynolds numbers $100$, $400$, $1600$ and $\infty$. The reference Mach number is set to $\textnormal{Ma} = U_0 / c_0 = 0.1$ in all cases in order to render the flow nearly incompressible, where $c_0$ denotes the speed of sound at temperature $T_0 = P_0 / (\gamma - 1) \, c_v \, \rho_0$.
### Details of the numerical discretization
The computational domain is partitioned into a uniform $64 \times 64 \times 64$ Cartesian grid and the Embedded DG (EDG) scheme with $p=2$ is used for the spatial discretization. The details of the EDG discretization of the Euler and Navier-Stokes equations are presented in [@Peraire:11]. We consider stabilization matrices of the form $$\bm{\sigma} = \beta \, |\bm{A}_n (\widehat{\bm{u}}_h)| ,$$ with $\beta = 0.25$ (under-upwinding) and $\beta = 1.00$ (standard upwinding), and where $\bm{A}_n = \partial (\bm{F} \cdot \bm{n}) / \partial \bm{u}$ denotes the Jacobian matrix of the inviscid flux normal to the element face. We note that the stabilization matrix implicitly defines the Riemann solver in hybridized DG methods, and in particular a Roe-type solver is recovered in the case $\beta = 1.00$. The interested reader is referred to [@Fernandez:AIAA:17a; @Fernandez:PhD:2018] for additional details on the relationship between the stabilization matrix and the resulting Riemann solver. The third-order, three-stage $L$-stable diagonally implicit Runge-Kutta DIRK(3,3) scheme [@Alexander:77] is used for the temporal discretization with CFL number $U_0 \, \Delta t / h^* = 0.1$. The solution is computed from $t = 0$ to $t = 15 \, L / U_0$.
### Numerical results
Figures \[f:kinEnSpTGV\_Re100\] to \[f:kinEnSpTGV\_ReInf\] show the time evolution of one-dimensional kinetic energy spectrum at the Reynolds numbers considered. The left and right images in these figures correspond to $\beta = 0.25$ and $1.00$, respectively, and the inverse of $\textnormal{Pe}^*$ at each time is indicated in the legends. The modified Péclet number in this problem is defined as $\textnormal{Pe}^* = h^* \, u_{rms} / \nu$, where $u_{rms}$ is the root mean square velocity and $\nu$ the kinematic viscosity. We note that, for a given $\textnormal{Re}$, the Péclet number slightly changes over time due to differences in $u_{rms}$. The short-term diffusion curves from non-modal analysis at the relevant Péclet numbers are shown in Figure \[f:shortTimeDiffusionTGV\], where the $x$ axis has been mapped from $\kappa \, h / (p+1)$ (as in Figures \[f:shortTimeDiffusionCurves\_pStudy\]$-$\[f:shortTimeDiffusionCurves\_betaStudy\]) to $\kappa \, L$ to facilitate the comparison with the energy spectra. Like in Section \[s:burgers\], non-modal analysis results show good agreement with the turbulent energy spectrum in the simulations. First, energy pileups are observed in the spectrum whenever the short-term diffusion curves are non-monotonic. These bottlenecks are induced by the lower short-term dissipation at specific wavenumbers. Particularly informative is the Reynolds number $400$. From non-modal analysis, the short-term dissipation curves at the corresponding $\textnormal{Pe}^* \approx 2.0 - 3.0$ are non-monotonic and monotonic near the Nyquist wavenumber with $\beta = 0.25$ and $1.00$, respectively. As a consequence, an energy bump appears in the spectrum near the Nyquist wavenumber with $\beta = 0.25$; which is not present with standard upwinding.
Second, non-modal analysis predicts a small dissipation at high wavenumbers with under-upwinding in convection-dominated regimes, and this directly translates to the TGV results. In particular, when the physical viscosity is small (i.e. in the high Reynolds number cases), the dissipation at high wavenumbers with $\beta = 0.25$ does not suffice to dissipate all the energy that is being transferred from the larger scales through the turbulence cascade. As a consequence, energy starts to accumulate near the Nyquist wavenumber from the beginning of the simulation. As time evolves, this bottleneck extends to larger scales due to the insufficient dissipation of energy at high wavenumbers, and eventually leads to nonlinear instability and the simulation breakdown at times $t \approx 4.42 \, L / U_0$ and $4.01 \, L / U_0$ for $\textnormal{Re} = 1600$ and $\infty$, respectively. In addition to non-monotonic dissipation characteristics, insufficient dissipation (specially at large wavenumbers) is [*per se*]{} another mechanism for nonlinear instability in under-resolved turbulence simulations.
![\[f:kinEnSpTGV\_Re100\] Time evolution of one-dimensional kinetic energy spectra for the Taylor-Green vortex at $\textnormal{Re} = 100$ with $\beta = 0.25$ (left) and $\beta = 1.00$ (right). The inverse of $\textnormal{Pe}^*$ at each time is indicated in the legend. ](kinEnSpTGV_Re100beta025.eps "fig:"){width="45.00000%"} ![\[f:kinEnSpTGV\_Re100\] Time evolution of one-dimensional kinetic energy spectra for the Taylor-Green vortex at $\textnormal{Re} = 100$ with $\beta = 0.25$ (left) and $\beta = 1.00$ (right). The inverse of $\textnormal{Pe}^*$ at each time is indicated in the legend. ](kinEnSpTGV_Re100beta1.eps "fig:"){width="45.00000%"}
![\[f:kinEnSpTGV\_Re400\] Time evolution of one-dimensional kinetic energy spectra for the Taylor-Green vortex at $\textnormal{Re} = 400$ with $\beta = 0.25$ (left) and $\beta = 1.00$ (right). The inverse of $\textnormal{Pe}^*$ at each time is indicated in the legend. ](kinEnSpTGV_Re400beta025.eps "fig:"){width="45.00000%"} ![\[f:kinEnSpTGV\_Re400\] Time evolution of one-dimensional kinetic energy spectra for the Taylor-Green vortex at $\textnormal{Re} = 400$ with $\beta = 0.25$ (left) and $\beta = 1.00$ (right). The inverse of $\textnormal{Pe}^*$ at each time is indicated in the legend. ](kinEnSpTGV_Re400beta1.eps "fig:"){width="45.00000%"}
![\[f:kinEnSpTGV\_Re1600\] Time evolution of one-dimensional kinetic energy spectra for the Taylor-Green vortex at $\textnormal{Re} = 1600$ with $\beta = 0.25$ (left) and $\beta = 1.00$ (right). The inverse of $\textnormal{Pe}^*$ at each time is indicated in the legend. ](kinEnSpTGV_Re1600beta025.eps "fig:"){width="45.00000%"} ![\[f:kinEnSpTGV\_Re1600\] Time evolution of one-dimensional kinetic energy spectra for the Taylor-Green vortex at $\textnormal{Re} = 1600$ with $\beta = 0.25$ (left) and $\beta = 1.00$ (right). The inverse of $\textnormal{Pe}^*$ at each time is indicated in the legend. ](kinEnSpTGV_Re1600beta1.eps "fig:"){width="45.00000%"}
![\[f:kinEnSpTGV\_ReInf\] Time evolution of one-dimensional kinetic energy spectra for the inviscid Taylor-Green vortex with $\beta = 0.25$ (left) and $\beta = 1.00$ (right). The inverse of $\textnormal{Pe}^*$ at each time is indicated in the legend. ](kinEnSpTGV_ReInfbeta025.eps "fig:"){width="45.00000%"} ![\[f:kinEnSpTGV\_ReInf\] Time evolution of one-dimensional kinetic energy spectra for the inviscid Taylor-Green vortex with $\beta = 0.25$ (left) and $\beta = 1.00$ (right). The inverse of $\textnormal{Pe}^*$ at each time is indicated in the legend. ](kinEnSpTGV_ReInfbeta1.eps "fig:"){width="45.00000%"}
![\[f:shortTimeDiffusionTGV\] Short-term diffusion curves with $\beta = 0.25$ (left) and $\beta = 1.00$ (right) from non-modal analysis. ](shortTimeDiffusionTGV_beta025.eps "fig:"){width="45.00000%"} ![\[f:shortTimeDiffusionTGV\] Short-term diffusion curves with $\beta = 0.25$ (left) and $\beta = 1.00$ (right) from non-modal analysis. ](shortTimeDiffusionTGV_beta1.eps "fig:"){width="45.00000%"}
\[s:discussion\]Discussion
==========================
We present a discussion on how to construct superior schemes using insights from non-modal analysis and numercal experiments. First, the two following conditions are critical for robust simulations of nonlinear systems, particularly for under-resolved simulations of systems featuring a kinetic energy cascade, such as turbulent flows. ([*i*]{}) Monotonic and slowly varying short-term diffusion characteristics to avoid energy bottlenecks at some specific wavenumbers. ([*ii*]{}) Sufficient dissipation near the Nyquist wavenumber of the grid to avoid energy accumulation at large wavenumbers. Otherwise, ([*i*]{}) and ([*ii*]{}) may lead to nonlinear instability and the simulation breakdown.
Non-modal analysis also provides insights to understand the built-in “subgrid-scale model” for under-resolved turbulence simulations. In particular, non-modal analysis indicates that, for moderately high accuracy orders, hybridized DG methods introduce numerical dissipation in under-resolved computations of convection-dominated flows, and this dissipation is localized near the Nyquist wavenumber. The implicit subgrid-scale model resembles variational multiscale [@Collis:2002; @Hughes:1998; @Hughes:2000; @Murman:2014] and spectral vanishing viscosity [@Karamanos:2000; @Kirby:2002; @Tadmor:1989] approaches in the sense that dissipation is applied to the smallest resolved scales and the amount of such dissipation depends mostly on the energy in those scales. By choosing the element size $h$ and the polynomial order $p$ inside of each element, the diffusion properties of the scheme can be tuned to obtain an equivalent filter width that can be readily used within an implicit LES context. The information regarding diffusion properties can also be used to improve accuracy in classical (explicit) LES, by better decoupling the wavenumber of the LES filter from the diffusion introduced by the numerics.
Overall, the polynomial orders $p=2$, $3$ and $4$ with standard upwinding seem to be the most adequate for under-resolved simulations of turbulent flows, at least in the implicit LES context. Under- and over-upwinding, as well as higher polynomial orders, may lead to stability issues. We note that Riemann solvers that are based on the maximum-magnitude eigenvalue of the Jacobian matrix of the Euler fluxes, such as the Lax-Friedrichs and the HLL solvers [@Toro:1999], produce over-upwinding at low Mach numbers.
The non-modal analysis framework can be generalized in several ways. First, one may study the finite-time behavior of the system (for some fixed $t > 0$), as opposed to its short-term behavior (limit $t \downarrow 0$). Second, the analysis can be extended to arbitrary initial conditions $u_0$, instead of Fourier modes only. We note, however, that Fourier modes are arguably the best choice to provide insights on the robustness and accuracy of the scheme. Third, more complex discretizations, including non-uniform meshes, non-constant coefficients, multi-dimensional problems and systems of conservation laws, could be considered. All these generalizations, however, would add more parameters and partially defeat the purpose of the analysis; which is to provide with a tool that, with a few inputs, approximately describes the behavior of the scheme for nonlinear problems.
\[s:conclusions\]Conclusions
============================
We introduced a non-modal analysis framework to investigate the short-term dynamics, and in particular the short-term dissipation in wavenumber space, of the semi-discrete system arising from the spatial discretization of the linear convection-diffusion equation. While applicable to high-order spectral/$hp$ methods in general, the non-modal analysis methodology was illustrated for the particular instance of the hybridized discontinuous Galerkin methods. The effect of the polynomial order, the Péclet number and the upwinding parameter on the short-term diffusion of the hybridized DG scheme were investigated. From these studies, the diffusion characteristics that are better suited, in terms of accuracy per DOF and robustness, for under-resolved simulations of nonlinear systems, seem to be those for polynomial orders $p=2$, $3$ and, to a lessen extent, $4$. Beyond these polynomial orders, the dissipation curves become strongly non-monotonic; which may lead to nonlinear instability due to bottlenecks in the energy spectrum. Under- and over-upwinded numerical fluxes, such as with Lax-Friedrichs type Riemann solvers at low Mach numbers, may similarly lead to stability and accuracy issues. While devised in the linear setting, non-modal analysis succeeded to predict the trends observed in the nonlinear problems considered. In particular, non-modal analysis results showed excellent agreement with numerical results for the Burgers, Euler and Navier-Stokes equations. From a practical perspective, non-modal analysis provides insights on why high-order spectral/$hp$ methods may suffer from stability issues in under-resolved simulations, as well as on how to devise better schemes for these problems. Furthermore, it provides insights to understand and improve the built-in subgrid-scale model in the scheme for under-resolved turbulence simulations. A discussion on these topics was presented.
Acknowledgments {#acknowledgments .unnumbered}
===============
The authors acknowledge the Air Force Office of Scientific Research (FA9550-16-1-0214), the National Aeronautics and Space Administration (NASA NNX16AP15A) and Pratt & Whitney for supporting this effort. The first author also acknowledges the financial support from the Zakhartchenko and “la Caixa” Fellowships.
\[s:connectionStHyDG\]Connection between standard DG and hybridized DG for linear convection
============================================================================================
Standard DG {#standard-dg .unnumbered}
-----------
We consider standard DG numerical fluxes of the form
\[e:fluxDG\] $$\begin{aligned}
{2}
\widehat{f}_{h,\ominus} = a \, \frac{u_{h,\ominus} + u^L_{h,\oplus}}{2} - \beta \, |a| \, \frac{u_{h,\ominus} - u^L_{h,\oplus}}{2} \mbox{ ,} \\
\widehat{f}_{h,\oplus} = a \, \frac{u_{h,\oplus} + u^R_{h,\ominus}}{2} + \beta \, |a| \, \frac{u_{h,\oplus} - u^R_{h,\ominus}}{2} \mbox{ ,}
\end{aligned}$$
where $\beta \geq 0$ is the upwinding parameter. The cases $\beta = 0$ and $\beta = 1$ correspond to the central flux and the standard upwinding, respectively.
Hybridized DG {#hybridized-dg .unnumbered}
-------------
We consider hybridized DG numerical fluxes of the form
\[e:fluxhDG\] $$\begin{aligned}
{2}
\tag{\ref{eq:fluxmi}}
\widehat{f}_{h,\ominus} = a \, \widehat{u}_{h,\ominus} - \beta \, |a| \, (u_{h,\ominus} - \widehat{u}_{h,\ominus}) \mbox{ ,} \\
\tag{\ref{eq:fluxpl}}
\widehat{f}_{h,\oplus} = a \, \widehat{u}_{h,\oplus} + \beta \, |a| \, (u_{h,\oplus} - \widehat{u}_{h,\oplus}) \mbox{ .}\end{aligned}$$
For $\beta > 0$, the numerical trace is uniquely defined and given by $\widehat{u}_h = (u^L_{h,\oplus} + u^R_{h,\ominus}) / 2$. The standard DG numerical fluxes are therefore recovered, and both the standard and hybridized DG schemes lead to the same numerical solution $u_h$. We note that the hybridized DG discretization is singular for $\beta = 0$, and in particular $\delta \widehat{u}_h$ is in the nullspace.
The 1% rule
===========
In analogy with eigenanalysis, we introduce the “1% rule” that indicates the largest wavenumber for which wave propagation is accurately resolved. We consider a wavenumber is well-resolved if the numerical damping per degree of freedom crossed is smaller than 1%. Since $\log (0.99) \approx -0.01$, we then define $\bar{\kappa}_{1\%}$ as the smallest wavenumber such that $\varpi^*_{\bar{\kappa}_{1\%}} = -0.01$. The “1% rule” thus differentiates between regions of negligible ($\bar{\kappa} \leq \bar{\kappa}_{1\%}$) and significant ($\bar{\kappa} > \bar{\kappa}_{1\%}$) numerical dissipation. We note that both diffusion and dispersion errors must be small for acurate propagation of a given wavenumber. However, diffusion errors dominate dispersion errors in this regime [@Ainsworth:2004; @Hu:1999] and thus the focus here on the diffusion properties only.
Figure \[TBD\] shows the short-term diffusion curves for pure convection and varying $p$. The plateau region in the diffusion curves become larger for increasing orders. For given $p$, it is possible to determine the largest value of $\bar{\kappa}$ within the plateau region through the 1% rule. Then, given $h$, one can obtain the corresponding value of $\kappa_{1\%}$. A summary of the short-term diffusion quantities for standard upwinding is shown in Table \[t:1PercentRule\].
The second and third columns in Table \[t:1PercentRule\] show the values of $\bar{\kappa}_{1\%}$ and $\kappa h|_{1\%}$, respectively, that can be achieved within an $hp$ setting of standard upwinded hybridized DG. Through these values, one can also estimate the coarsest mesh spacing that can be used in order to resolve scales up to a given wavenumber. The fourth column contains the associated number of degrees of freedom per wavelength, $\textnormal{DOFpW} =2 \pi / \bar{\kappa}_{1\%}$. This column shows that the hybridized DG discretization becomes more efficient [*per degree of freedom*]{} as the polynomial order is increased. In fact, spectral-like resolution [@Lele:1992] is approached for increasing $p$ on a [*per degree of freedom*]{} basis, as shown in Figure \[TBD\].
Finally, the last two columns in Table \[t:1PercentRule\] show the values of $\varpi^*_{\kappa}$ at $\bar{\kappa} = \pi$ and the associated damping factors per DOF crossed, $\exp(\varpi^*_{\pi})$, respectively. These quantities represent the amount of dissipation affecting the smallest captured scales and can be seen to become stronger for increasing P. For $p=2$, where $\exp(\varpi^*_{\pi}) \approx 2 \cdot 10^{-2}$, the smallest scales would cross more than two DOFs to suffer the same damping yielded for a single DOF by the $p=7$ discretization, where $\exp(\varpi^*_{\pi}) \approx 5 \cdot 10^{-4}$. This is probably an extra advantage of employing higher order discretizations, since as previously recognized a faster damping of the poorly-resolved scales further precludes them from polluting the numerical solution.
$p$ $\bar{\kappa}_{1\%}$ $\kappa h|_{1\%}$ $\textnormal{DOFpW}$ $\varpi^*_{\pi}$ $\exp(\varpi^*_{\pi})$
----- ---------------------- ------------------- ---------------------- ------------------ ------------------------
$1$ $0.7502$ $1.500$ $8.37$ $-3.00$ $0.0498$
$2$ $1.4275$ $4.283$ $4.40$ $-3.88$ $0.0207$
$3$ $1.0511$ $4.204$ $5.98$ $-3.94$ $0.0194$
$4$ $1.2004$ $6.002$ $5.23$ $-2.81$ $0.0601$
$5$ $1.3937$ $8.362$ $4.51$ $-1.36$ $0.2578$
$6$ $1.5875$ $11.113$ $3.96$ $-0.42$ $0.6603$
$7$ $1.7623$ $14.098$ $3.57$ $-0.04$ $0.9621$
$8$ $1.4605$ $13.145$ $4.30$ $-0.03$ $0.9698$
: \[t:1PercentRule\] Relevant short-term diffusion quantities of standard upwinded $\beta = 1$ hybridized DG for pure convection as a function of the polynomial order $p$.
[10]{}
N. Adams, S. Hickel, S. Franz, Implicit subgrid-scale modeling by adaptive deconvolution, J. Comput. Phys. 200 (2) (2004) 412–431. M. Ainsworth, Dispersive and dissipative behaviour of high order discontinuous Galerkin finite element methods, J. Comput. Phys. 198 (1) (2004) 106–130. R. Alexander, Diagonally implicit Runge-Kutta methods for stiff ODEs, SIAM J. Numer. Anal. 14 (6) (1977) 1006–1021. J. Bec, K. Khanin, Burgers turbulence, Phys. Rep. 447 (1) (2007) 1–66. A.D. Beck, T. Bolemann, D. Flad, H. Frank, G.J. Gassner, F. Hindenlang, C.-D. Munz, High-order discontinuous Galerkin spectral element methods for transitional and turbulent flow simulations, Int. J. Numer. Meth. Fl. 76 (8) (2014) 522–548. M.E. Brachet, Direct simulation of three-dimensional turbulence in the Taylor-Green vortex, Fluid Dyn. Res. 8 (1–4) (1991) 1–8. A. Chekhlov, V. Yakhot, Kolmogorov turbulence in a random-force-driven Burgers equation, Phys. Rev. E 51 (4) (1995) R2739. X. Chen, N.C. Nguyen, S.-H. Oh, J. Peraire, Nanogap-enhanced Terahertz sensing of 1-nm-thick dielectric films, ACS Photonics 2 (3) (2015) 417–424. A. Christophe, S. Descombes, S. Lanteri, An implicit hybridized discontinuous Galerkin method for the 3D time-domain Maxwell equations, Appl. Math. Comput. 319 (2018) 395–408. B. Cockburn, J. Gopalakrishnan, R. Lazarov, Unified hybridization of discontinuous Galerkin, mixed and continuous Galerkin methods for second order elliptic problems, SIAM J. Numer. Anal. 47 (2) (2009) 1319–1365. B. Cockburn, J. Guzman, S.C. Soon, H.K. Stolarski, An Analysis of the Embedded Discontinuous Galerkin Method for Second-Order Elliptic Problems, SIAM J. Numer. Anal. 47 (4) (2009) 2686–2707. S.S. Collis, The DG/VMS method for unified turbulence simulation, In: 32nd AIAA Fluid Dynamics Conference and Exhibit, St. Louis, USA, 2002. J.P.S. Dahm, K.J. Fidkowski, Error estimation and adaptation in hybridized discontinous Galerkin methods, In: 52nd Aerospace Sciences Meeting, National Harbor, USA, 2014. D. De Grazia, G. Mengaldo, D. Moxey, P.E. Vincent, S.J. Sherwin, Connections between the discontinuous Galerkin method and high-order flux reconstruction schemes, Int. J. Numer. Methods Fluids 75 (12) (2014) 860–877. X. Feng, P. Lu, X. Xu, A Hybridizable Discontinuous Galerkin Method for the Time-Harmonic Maxwell Equations with High Wave Number, Comput. Methods Appl. Math. 16 (3) (2016) 429–445. P. Fernandez, N.C. Nguyen, X. Roca, J. Peraire, Implicit large-eddy simulation of compressible flows using the Interior Embedded Discontinuous Galerkin method, In: 54th AIAA Aerospace Sciences Meeting, San Diego, USA, 2016. P. Fernandez, N.C. Nguyen, J. Peraire, The hybridized Discontinuous Galerkin method for Implicit Large-Eddy Simulation of transitional turbulent flows, J. Comput. Phys. 336 (1) (2017) 308–329. P. Fernandez, N.C. Nguyen, J. Peraire, Subgrid-scale modeling and implicit numerical dissipation in DG-based Large-Eddy Simulation, In: 23rd AIAA Computational Fluid Dynamics Conference, Denver, USA, 2017. P. Fernandez, The hybridized discontinuous Galerkin methods for large-eddy simulation of transitional and turbulent flows, PhD Thesis, Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, 2018. A. Frere, K. Hillewaert, H. Sarlak, R.F. Mikkelsen, Cross-Validation of Numerical and Experimental Studies of Transitional Airfoil Performance, In: 33rd ASME Wind Energy Symposium, Kissimmee, USA, 2015. G. Gassner, D. Kopriva, A comparison of the dispersion and dissipation errors of Gauss and Gauss-Lobatto discontinuous Galerkin spectral element methods, SIAM J. Sci. Comput. 33 (5) (2011) 2560–2579. G.J. Gassner, A.D. Beck, On the accuracy of high-order discretizations for underresolved turbulence simulations, Theor. Comput. Fluid Dyn. 27 (3) (2013) 221–237. G. Giorgiani, S. Fernandez-Mendez, A. Huerta, Hybridizable Discontinuous Galerkin with degree adaptivity for the incompressible Navier-Stokes equations, Comput. Fluids 98 (2014) 196–208. F. Hu, M. Hussaini, P. Rasetarinera, An analysis of the discontinuous Galerkin method for wave propagation problems, J. Comput. Phys. 151 (2) (1999) 921–946. F. Hu, H. Atkins, Eigensolution analysis of the discontinuous Galerkin method with nonuniform grids: I. One space dimension, J. Comput. Phys. 182 (2) (2002) 516–545. T.J.R. Hughes, G.R. Feijoo, L. Mazzei, J.B. Quincy, The variational multiscale method - a paradigm for computational mechanics, Comput. Methods Appl. Math. 166 (1) (1998) 3–24. T.J.R. Hughes, L. Mazzei, K.E. Jansen, Large Eddy Simulation and the variational multiscale method, Comput. Visual Sci. 3 (1) (2000) 47–59. H.T. Huynh, A flux reconstruction approach to high-order schemes including discontinuous Galerkin methods, In: 18th AIAA Computational Fluid Dynamics Conference, Miami, USA, 2007. G.S. Karamanos, G.E. Karniadakis, A spectral vanishing viscosity method for large-eddy simulations, J. Comput. Phys. 163 (1) (2000) 22–50. R.M. Kirby, G.E. Karniadakis, Coarse resolution turbulence simulations with spectral vanishing viscosity large-eddy simulations (SVV-LES), J. Fluids Eng. 124 (4) (2002) 886–891. R.M. Kirby, G.E. Karniadakis, De-aliasing on non-uniform grids: algorithms and applications, J. Comput. Phys. 191 (1) (2003) 249–264. C. Lehrenfeld, J. Schoberl, High order exactly divergence-free Hybrid Discontinuous Galerkin Methods for unsteady incompressible flows, Comput. Methods Appl. Mech. Eng. 307 (2016) 339–361. L. Li, S. Lanteri, N.A. Mortensen, M. Wubs, A hybridizable discontinuous Galerkin method for solving nonlocal optical response models, Comput. Phys. Commun. 219 (2017) 99–107. C. Liang, S. Premasuthan, A. Jameson, Z. Wang, Large eddy simulation of compressible turbulent channel flow with spectral difference method, In: 47th AIAA Aerospace Sciences Meeting, Orlando, USA, 2009. G. Mengaldo, D. De Grazia, P.E. Vincent, S.J. Sherwin, On the connections between discontinuous Galerkin and flux reconstruction schemes: extension to curvilinear meshes, J. Sci. Comput. 67 (3) (2016) 1272–1292. G. Mengaldo, R.C. Moura, B. Giralda, J. Peir[ó]{}, S.J. Sherwin, Spatial eigensolution analysis of discontinuous Galerkin schemes with practical insights for under-resolved computations and implicit LES, Comput. Fluids (2017), In Press. G. Mengaldo, D. De Grazia, R.C. Moura, S.J. Sherwin, Spatial eigensolution analysis of energy-stable flux reconstruction schemes and influence of the numerical flux on accuracy and robustness, J. Comp. Phys. 358 (1) (2018) 1–20. R.C. Moura, S.J. Sherwin, J. Peir[ó]{}, Linear dispersion-diffusion analysis and its application to under-resolved turbulence simulations using discontinuous Galerkin spectral/hp methods, J. Comp. Phys. 298 (2015) 695–710. R.C. Moura, S.J.Sherwin, J.Peir[ó]{}, Eigensolution analysis of spectral/hp continuous Galerkin approximations to advection-diffusion problems: Insights into spectral vanishing viscosity, J. Comp. Phys. 307 (2016) 401–422. R.C. Moura, P. Fernandez, G. Mengaldo, J. Peraire, S. Sherwin, Temporal eigenanalysis of hybridized discontinuous Galerkin methods for advection-diffusion problems, J. Comp. Phys., In preparation for submission. S.M. Murman, L.T. Diosady, A. Garai, Development of dynamic sub-grid models for variational multiscale methods, In: Proceedings of the 2014 Summer Program, Center for Turbulence Research, Stanford University, 2014. S.M. Murman, L.T. Diosady, A. Garai, M. Ceze, A Space-Time Discontinuous-Galerkin Approach for Separated Flows, In: 54th AIAA Aerospace Sciences Meeting, San Diego, USA, 2016. N.C. Nguyen, J. Peraire, B. Cockburn, An implicit high-order hybridizable discontinuous Galerkin method for linear convection-diffusion equations, J. Comp. Phys. 228 (9) (2009) 3232–3254. N.C. Nguyen, J. Peraire, B. Cockburn, An implicit high-order hybridizable discontinuous Galerkin method for nonlinear convection-diffusion equations, J. Comp. Phys. 228 (23) (2009) 8841–8855. N.C. Nguyen, J. Peraire, An Adaptive Shock-Capturing HDG Method for Compressible Flows, In: 20th AIAA Computational Fluid Dynamics Conference, Honolulu, USA, 2011. N.C. Nguyen, J. Peraire, Hybridizable discontinuous Galerkin methods for partial differential equations in continuum mechanics, J. Comput. Phys. 231 (18) (2012) 5955–5988. N.C. Nguyen, J. Peraire, B. Cockburn, A class of embedded discontinuous Galerkin methods for computational fluid dynamics, J. Comput. Phys. 302 (1) (2015) 674–692. J.S. Park, F.D. Witherden, P.E. Vincent, High-Order Accurate Implicit Large Eddy Simulations of Flow over a NACA0021 Aerofoil in Deep Stall, AIAA J. 55 (7) (2017) 2186–2197. M. Parsani, G. Ghorbaniasl, C. Lacor, E. Turkel, An implicit high-order spectral difference approach for large eddy simulation, J. Comput. Phys. 229 (14) (2010) 5373–5393. J. Peraire, N.C. Nguyen, B. Cockburn, A hybridizable discontinuous Galerkin method for the compressible Euler and Navier-Stokes equations, In: 48th AIAA Aerospace Sciences Meeting, Orlando, USA, 2010. J. Peraire, N.C. Nguyen, B. Cockburn, An Embedded Discontinuous Galerkin Method for the Compressible Euler and Navier-Stokes Equations, In: 20th AIAA Computational Fluid Dynamics Conference, Honolulu, USA, 2011. F. Renac, M. de la Llave Plata, E. Martin, J.-B. Chapelier, V. Couaillier, Aghora: A High-Order DG Solver for Turbulent Flow Simulations, In: IDIHOM: Industrialization of High-Order Methods - A Top-Down Approach, Notes on Numerical Fluid Mechanics and Multidisciplinary Design 128 (2015) 315–335. S. Rhebergen, B. Cockburn, A space-time hybridizable discontinuous Galerkin method for incompressible flows on deforming domains, J. Comput. Phys. 231 (11) (2012) 4185–4204. P.J. Schmid, Nonmodal Stability Theory, Annu. Rev. Fluid Mech. 39 (2007) 129–162. J. Sch[" u]{}tz, G. May, A hybrid mixed method for the compressible Navier-Stokes equations, J. Comput. Phys. 240 (2013) 58–75. J.P. Sheldon, S.T. Miller, J.S. Pitt, A hybridizable discontinuous Galerkin method for modeling fluid-structure interaction, J. Comput. Phys. 326 (2016) 91–114. E. Tadmor, Convergence of spectral methods for nonlinear conservation laws, SIAM J. Numer. Anal. 26 (1) (1989) 30. G.I. Taylor, A.E. Green, Mechanism of the production of small eddies from large ones, P. R. Soc. Lond. A. 158 (1937) 499–521. S. Terrana, J.-P. Vilotte, L. Guillot, A spectral hybridizable discontinuous Galerkin method for elastic-acoustic wave propagation, Geophys. J. Int. 213 (1) (2018) 574–602. E.F. Toro, Riemann Solvers and Numerical Methods for Fluid Dynamics, Springer, Berlin, 1999. L.N. Trefethen, A.E. Trefethen, S.C. Reddy, T.A. Driscoll, Hydrodynamic Stability Without Eigenvalues, Science 261 (5121) (1993) 578–584. L.N. Trefethen, Pseudospectra of Linear Operators, SIAM Rev. 39 (3) (1997) 383-406. M.P. Ueckermann, P.F.J. Lermusiaux, Hybridizable discontinuous Galerkin projection methods for Navier-Stokes and Boussinesq equations, J. Comput. Phys. 306 (2016) 390–421. A. Uranga, P.-O. Persson, M. Drela, J. Peraire, Implicit Large Eddy Simulation of transition to turbulence at low Reynolds numbers using a Discontinuous Galerkin method, Int. J. Numer. Meth. Eng. 87 (2011) 232–261. B. Vermeire, J. Cagnone, S. Nadarajah, ILES using the correction procedure via reconstruction scheme, In: 51st AIAA Aerospace Sciences Meeting, Grapevine, USA, 2013. B. Vermeire, S. Nadarajah, P. Tucker, Canonical test cases for high-order unstructured implicit large eddy simulation, In: 52nd AIAA Aerospace Sciences Meeting, National Harbour, USA, 2014. F. Vidal-Codina, N.C. Nguyen, S.-H. Oh, J. Peraire, A hybridizable discontinuous Galerkin method for computing nonlocal electromagnetic effects in three-dimensional metallic nanostructures, J. Comput. Phys. 355 (2018) 548–565. P.E. Vincent, P. Castonguay, A. Jameson, Insights from von Neumann analysis of high-order flux reconstruction schemes, J. Comput. Phys. 230 (22) (2011) 8134–8154. C.C. de Wiart, K. Hillewaert, Development and Validation of a Massively Parallel High-Order Solver for DNS and LES of Industrial Flows, In: IDIHOM: Industrialization of High-Order Methods - A Top-Down Approach, Notes on Numerical Fluid Mechanics and Multidisciplinary Design 128 (2015) 251–292. M. Woopen, A. Balan, G. May, A hybridized discontinuous Galerkin method for three-dimensional compressible flow problems, In: 52nd AIAA Aerospace Sciences Meeting, National Harbor, USA, 2014. D. Yoo, N.C. Nguyen, L. Martin-Moreno, D.A. Mohr, S. Carretero-Palacios, J. Shaver, J. Peraire, T.W. Ebbesen, S.H. Oh, High-throughput fabrication of resonant metamaterials with ultrasmall coaxial apertures via atomic layer lithography, Nano Lett. 16 (3) (2016) 2040–2046. O. Zikanov, A. Thess, R. Grauer, Statistics of turbulence in a generalized random-force-driven burgers equation, Phys. Fluids 9 (5) (1997) 1362–1367.
[^1]: Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, USA.
[^2]: Instituto Tecnol[' o]{}gico de Aeron[' a]{}utica (ITA), S[\~ a]{}o Jos[' e]{} dos Campos, SP 12228-900, Brazil.
[^3]: Division of Engineering and Applied Science, California Institute of Technology, 1200 E. California Boulevard, Pasadena, CA 91125, USA.
[^4]: Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, USA.
[^5]: Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, USA.
[^6]: The Galerkin projection of $\exp(i \kappa x)$ trivially features this wave-like behavior, and thus holds at $t=0$. Equation at $t = 0$ then follows from . Since Equations are satisfied at $t = 0$, it follows from (which holds at any given time under the previous assumptions) that they are also satisfied at all subsequent times $t \geq 0$.
[^7]: While no temporal chaos exists in the Taylor-Green vortex, we use the term [*turbulence*]{} to refer to the phase of spatial chaos (spatial decoherence) that takes place after $t \approx 7 - 9 \, L / U_0$ for Reynolds numbers above about $1000$.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'A mechanism of disappearance and formation of the Efimov levels of the helium $^4$He$_3$ trimer is studied when the force of interatomic interaction is changed. The resonances including virtual levels are calculated by the method based on the solving the boundary value problem, at complex energies, for the Faddeev differential equations describing the $(2+1\to 2+1;\,1+1+1)$ scattering processes.'
address:
- 'Laboratory of Computing Techniques and Automation, Joint Institute for Nuclear Research, Dubna, 141980, Russia'
- 'Physikalishes Institut, Universität Bonn, Endenicher Allee 11–13, D-53115 Bonn, Germany'
author:
- 'E. A. Kolganova'
- 'A. K. Motovilov[^1]'
title: 'Scattering and Resonances in the $^4$He Three-Atomic System[^2] '
---
Introduction
============
The system of three $^4$He atoms is of considerable interest for various fields of physical chemistry and molecular physics. The present paper is a sequel of studies of the $^4$He$_3$ system undertaken in [@KMS-JPB], where the repulsive component of the He–He interaction at short distances between atoms is approximated by a hard core. This allows one to investigate the $^4$He$_3$ system within a mathematically rigorous method of solving a three-body problem in the Boundary-Condition Model developed in [@MerMot; @MMYa]. In [@KMS-JPB], such an approach has been successfully applied for calculating not only scattering but also the ground- and excited-state energies of the helium trimer. Investigation made in [@KMS-JPB] has shown that the method proposed in [@MerMot; @MMYa] is well suited for performing three–body molecular computations in the case where repulsive components of interatomic interactions are of a hard core nature.
There is a series of works [@EsryLinGreene; @Gloeckle; @KMS-JPB] showing that the excited state of the $^4$He trimer is initiated indeed by the Efimov effect [@VEfimov]. In these works the various versions of the Aziz $^4$He–$^4$He potential were employed However, the basic result of Refs. [@EsryLinGreene; @Gloeckle; @KMS-JPB] on the excited state of the helium trimer is the same: this state disappears when the interatomic potential is multiplied by the “amplification factor" $\lambda$ of order 1.2. More precisely, if this potential is multiplied by the increasing factor $\lambda>1$ then the following effect is observed. First, the difference $\epsilon_d(\lambda)-E_t^{(1)}(\lambda)$ between the dimer energy $\epsilon_d(\lambda)$ and the energy of the trimer excited state $E_t^{(1)}(\lambda)$ increases. Then the behavior of this difference radically changes and with further increase of $\lambda$ it monotonously decreases. At $\lambda\approx 1.2$ the level $E_t^{(1)}$ disappears. It is just such a nonstandard behavior of the energy $E_t^{(1)}(\lambda)$ as the coupling between helium atoms becomes more and more strengthening, points to the Efimov nature of the trimer excited state. And vice versa, when $\lambda$ slightly decreases (no more than 2%), the second excited state $E_t^{(2)}$ appears in the trimer [@EsryLinGreene; @Gloeckle].
Here we present the results of our numerical study of a mechanism of disappearance and formation of the Efimov levels of the helium $^4$He$_3$ trimer using the method of search for resonances in a three–body system on the basis of the Faddeev differential equations. The idea of the method formulated and proved in [@MotMathNachr] consists in calculating the analytic continuation of the component ${\rm S}_0(z)$ of the scattering matrix corresponding to the ($2+1\to 2+1$) process in the physical sheet. For the potentials we use, the three–body resonances (including virtual levels) lying in the unphysical sheet of energy $z$ plane adjoining the physical sheet along the interval $(\epsilon_d,0)$ are the roots of the function ${\rm S}_0(z)$ in the physical sheet. We have earlier employed this method for computing resonances as roots of ${\rm
S}_0(z)$ in the three–nucleon problem [@YaFKM].
Method
======
In this work we consider the three-atomic $^4$He system with the total angular momentum $L=0$. The angular partial analysis reduces the initial Faddeev equation for three identical bosons to a system of coupled two-dimensional integro-differential equations (see Ref. [@KMS-JPB] and references therein) $$\begin{aligned}
\label{FadPartCor}
\lefteqn{ \left[-\frac{\partial^2}{\partial x^2}
-\frac{\partial^2}{\partial y^2}
+l(l+1)\left(\frac{1}{x^2}
+\frac{1}{y^2}\right)
-E\right]F_l(x,y) }
\qquad\qquad\qquad\qquad\qquad\qquad \\
\nonumber
&&=\left\{\begin{array}{cl} -V(x)\Psi_l(x,y), & x>c \\
0, & x<c\,.
\end{array}\right.\end{aligned}$$ Here, $x,y$ stand for the standard Jacobi variables and $c$, for the core range. At $L=0$ the partial angular momentum $l$ corresponds both to the dimer subsystem and a complementary atom. The energy $z$ can get both real and complex values. The He–He potential $V(x)$ acting outside the core domain is assumed to be central. The partial wave function $\Psi_l(x,y)$ is related to the Faddeev components $F_l(x,y)$ by $
\Psi_l(x,y)=F_l(x,y) + \sum_{l'}\int_{-1}^{+1}
d\eta\,h_{l l'}(x,y,\eta)\,F_{l'}(x',y')
$ where $
x'=(\frac{1}{4}\,x^2+
\frac{3}{4}\,y^2-\frac{\sqrt{3}}{2}\,xy\eta)^{1/2}\,,
y'=(\frac{3}{4}\,x^2+
\frac{1}{4}\,y^2+\frac{\sqrt{3}}{2}\,xy\eta)^{1/2}\,
$ and $1 \leq{\eta}\leq 1$. The explicit form of the function $h_{ll'}$ can be found in Refs. [@MF; @MGL]. The functions $F_{l}(x,y)$ satisfy the boundary conditions $$\label{BCStandardCor}
F_{l}(x,y)\left.\right|_{x=0}
=F_{l}(x,y)\left.\right|_{y=0}=0\, \qquad {\rm and} \qquad
\Psi_{l}(c,y)=0\,.$$
Here we only deal with a finite number of equations (\[FadPartCor\]), assuming that $l\leq l_{\rm max}$ where $l_{\rm max}$ is a certain fixed even number. The condition $0\leq l\leq l_{\rm max}$ is equivalent to the supposition that the potential $V(x)$ only acts in the two-body states with $l=0,2,\ldots,l_{\rm max}$. We assume that the potential $V(x)$ is finite, i.e., for $x>r_0$, $r_0>0$. The asymptotic conditions as $\rho\rightarrow\infty$ and/or $y\rightarrow\infty$ for the partial Faddeev components of the $(2+1\rightarrow 2+1\,;\,1+1+1)$ scattering wave functions for $z=E+{\rm i}0$, $E>0$, read (see, e.g., Ref. [@MF]) $$\begin{aligned}
\nonumber
F_l(x,y;z) & = &
\delta_{l0}\psi_d(x)\left\{\sin(\sqrt{z-\epsilon_d}\,y)
+ \exp({\rm i}\sqrt{z-\epsilon_d}\,y)
\left[{\rm a}_0(z)+o\left(1\right)\right]\right\} \\
\label{AsBCPartS}
&& +
\displaystyle\frac{\exp({\rm i}\sqrt{z}\rho)}{\sqrt{\rho}}
\left[A_l(z,\theta)+o\left(1\right)\right]\,.\end{aligned}$$ We assume that the $^4$He dimer has an only bound state with an energy $\epsilon_d$, and wave function $\psi_d(x)$, $\psi_d(x)=0$ for $0\leq x\leq c$. The notations $\rho$, $\rho=\sqrt{x^2+y^2}$, and $\theta$, $\theta=\mathop{\rm
arctan}(y/x)$, are used for the hyperradius and hyperangle. The coefficient ${\rm a}_0(z)$, $z=E+{\rm i}0$, for $E>\epsilon_d$ is the elastic scattering amplitude. The functions $A_l(E+{\rm i}0,\theta)$ provide us, at $E>0$, the corresponding partial Faddeev breakup amplitudes. For real $z=E+{\rm i}0$, $E>\epsilon_d$, the component of the $s$-wave partial scattering matrix for a system of three helium atoms is given by the expression $${\rm S}_0(z)=1+2{\rm i}{\rm a}_0(z)\,.
$$ Our goal is to study the analytic continuation of the function ${\rm S}_0(z)$ into the physical sheet. As it follows from the results of Refs. [@MotMathNachr], the ${\rm S}_0(z)$ is just that truncation of the total scattering matrix whose roots in the physical sheet correspond to location of resonances in the unphysical sheet adjoining the physical one along the spectral interval $(\epsilon_d,0)$.
Results of computations {#results}
=======================
In the present work we make use of the Faddeev equations (\[FadPartCor\]) considered together with the boundary conditions (\[BCStandardCor\]), (\[AsBCPartS\]) to calculate the values of the $^4$He$_3$ scattering matrix ${\rm S}_0(z)$ in the physical sheet. We search for the resonances including the virtual levels as roots of ${\rm S}_0(z)$ and for the bound-state energies as positions of poles of ${\rm S}_0(z)$. All the results presented below are obtained for the case $l_{\rm max}=0$.
In all our calculations, As the interatomic He–He-interaction we employed the widely used semiempirical potential HFD-B [@Aziz87]. The value of the core range $c$ is chosen to be equal $1.3$[Å]{} providing at least six reliable figures of the dimer binding energy $\epsilon_d$ and three figures of the trimer ground state energy $E_t^{(0)}$. A detailed description of the numerical method we use is presented in Ref. [@KMS-JPB]. In contrast to [@KMS-JPB], in the present work we solve the block-three-diagonal algebraic system on the basis of the matrix sweep method. This allows us to dispense with writing the system matrix on the hard drive and to carry out all the operations related to its inversion immediately in RAM. Besides, the matrix sweep method reduces almost by one order the computer time required for computations on the grids of the same dimensions as in [@KMS-JPB].
We searched for the resonances (roots of the function ${\rm
S}_0(z)$ on the physical sheet) and bound-state energies (roots of the function ${\rm S}_0^{-1}(z)$ for real $z<\epsilon_d$) of the helium trimer by using the complex version of the secant method. We found positions of the four “resonances”, the roots of ${\rm S}_0(z)$, in case of the grid parameters and [Å]{}. Complex roots of the function ${\rm S}_0(z)$ are located at points mK, mK, mK and mK. These “resonances” are situated beyond the domain of scattering matrix holomorphy $\Pi^{(S)}$ where the applicability of our method is proved [@KolMotYaF]. So we do not consider the roots of function ${\rm S}_0(z)$ as genuine resonances for the $^4$He$_3$ system. However it is remarkable that the “true" (i.e., getting inside $\Pi^{(S)}$) virtual levels and then the energies of the excited (Efimov) states appear just due to these (quasi)resonances when the potential $V(x)$ is weakened.
Following [@EsryLinGreene; @Gloeckle; @KMS-JPB], instead of the initial potential $V(x)=V_{\rm HFD-B}(x)$, we consider the potentials $
V(x)=\lambda\cdot V_{\rm HFD-B}(x).
$ To establish the mechanism of formation of new excited states in the $^4$He trimer, we have first calculated the scattering matrix ${\rm S}_0(z)$ for $\lambda<1$. We have found that for a value of $\lambda$ slightly smaller than $0.9885$, the (quasi)resonance closest to the real axis gets on it and transforms into a virtual level of the second order. This virtual level is preceded by the (quasi)resonances mK for $\lambda=0.989$ and mK for $\lambda=0.9885$. The originating virtual level is of the second order since simultaneously with the root of the function ${\rm S}_0(z)$, also the conjugate root of this function gets on the real axis. With a subsequent decrease of $\lambda$ the virtual level of the second order splits into a pair of the virtual levels of the first order which move in opposite directions. One of the virtual level moves towards the threshold $\epsilon_d$ and “collides” with it at $\lambda<0.98$. For $\lambda=0.975$ the function ${\rm S}_0(z)$ instead of the root acquires a new pole corresponding to the second excited state of the trimer with the energy $E_t^{(2)}$.
To determine the mechanism of disappearance of the excited state of the helium trimer we have calculated the scattering matrix ${\rm S}_0(z)$ when the two-body interactions become stronger owing to the increasing coupling constant $\lambda>1$. We found that this disappearance proceeds just according to the scheme of the formation of new excited states; only the order of occurring events is inverse. In the interval between $\lambda=1.17$ and $\lambda=1.18$ there occurs a “jump" of the level $E_t^{(1)}$ on the unphysical sheet and it transforms from the pole of the function ${\rm
S}_0(z)$ into its root corresponding to the trimer virtual level.
Acknowledgements {#acknowledgements .unnumbered}
================
The authors are grateful to Prof. V.B.Belyaev and Prof. H.Toki for help and assistance in calculations at the supercomputer of the Research Center for Nuclear Physics of the Osaka University, Japan. One of the authors (A.K.M.) is much indebted to Prof. W.Sandhas for his hospitality at the Universität Bonn. The support of this work by the Deutsche Forschungsgemeinschaft and Russian Foundation for Basic Research is gratefully acknowledged.
[99]{} E. A. Kolganova, A. K. Motovilov, S. A. Sofianos: J. Phys. B. [**31**]{} (1998) 1279 (LANL E-print [physics/9612012]{}). S. P. Merkuriev, A.K. Motovilov: Lett. Math. Phys. [**7**]{} (1983) 497. S. P. Merkuriev, A. K. Motovilov, S. L. Yakovlev: Theor. Math. Phys. [**94**]{} (1993) 306 (also see LANL E-print [nucl-th/9606022]{}). B. D. Esry, C. D. Lin, C. H. Greene: Phys. Rev. A. [**54**]{} (1996) 394. T. Cornelius, W. Glöckle: J. Chem. Phys. [**85**]{} (1986) 3906. V. Efimov: Nucl. Phys. A. [**210**]{} (1973) 157. A. K. Motovilov: Math. Nachr. [**187**]{} (1997) 147 (LANL E-print [funct-an/9509003]{}). E. A. Kolganova, A. K. Motovilov: Phys. Atom. Nucl. [**60**]{} (1997) 177 (LANL E-print [nucl-th/9602001]{}); also see LANL E-print [nucl-th/9702037]{}. L. D. Faddeev, S. P. Merkuriev: [*Quantum Scattering Theory for Several Particle Systems*]{}. Doderecht: Kluwer Academic Publishers 1993. S. P. Merkuriev, C. Gignoux, A. Laverne: Ann. Phys. (N.Y.) [**99**]{} (1976) 30. E. A. Kolganova, A. K. Motovilov: Preprint JINR E4-98-243 (LANL E-print [physics/9808027]{}). R. A. Aziz, F. R. W. McCourt, C. C. K. Wong: Mol. Phys. [**61**]{} (1987) 1487.
[^1]: On leave of absence from the Laboratory of Theoretical Physics, Joint Institute for Nuclear Research, Dubna, 141980, Russia
[^2]: Contribution to Proceedings of the First International Conference on Modern Trends in Computational Physics, June 15–20, 1998, Dubna (Russia). LANL E-print [physics/9810005]{}.
|
{
"pile_set_name": "ArXiv"
}
|
---
abstract: 'The purpose of this material is to review the Adler Kostant Symes scheme as a theory which can be developped succesfully in different contexts. It was useful to describe some mechanical systems, the so called generalized Toda, and now it was proved to be a tool for the study of the linear approach to the motion of n uncoupled harmonic oscillators. The complete integrability of these systems has an algebraic description. In the original theory this is related to ad-invariant functions, but new examples show that new conditions should be investigated.'
address: 'G. Ovando: CONICET y ECEN-FCEIA, Universidad Nacional de Rosario, Pellegrini 250, 2000 Rosario, Santa Fe, Argentina'
author:
- 'Gabriela P. Ovando'
title: The Adler Kostant Symes scheme in physics
---
[^1]
[^2]
Introduction
============
In this work we are interested in the use of Lie theory to understand some Hamiltonian systems. For the study of completely integrable systems one needs to identify the following: i) the symplectic structure, which gives the system its Hamiltonian character, ii) first integrals or constants of motion, iii) action angle variables, and the computation of their evolution. Indeed this is a very difficult approach but it is possible for systems related to certain Lie groups.
To this end there are several methods with a common idea: the realization of canonical equations on Lie algebras, or on orbits of a certain action or on symmetric spaces. These ideas appeared in the 70’s and were developped, under other by several authors as Adler, Fomenko, Kostant, Mischenko, Olshanetsy, Perelomov, Trofimov, Symes, etc. (see for instance [@Ad1] [@F-M1] [@F-M2] [@F-T] [@Ko1] [@O-P] [@P] and [@Sy] and their references).
In any method all of the above steps i) ii) iii) are reflected by algebraic circumstances. One needs a way for imbedding a certain Hamiltonian system into a Lie algebra, effective methods for constructing sets of involution and the proof of the full integrability of a wide family of functions in involution.
In this chapter we are concerned with so called Adler-Kostant-Symes scheme, which brings together a mathematical framework with Lie theory but also consequences in the dynamics of the Hamiltonian system. This method was successful when studying some mechanical systems such as the rigid body or the generalized Toda lattice [@Ad2] [@Ko2] [@Sy] [@R2]. In this setting the phase space of the Hamiltonian systems become coadjoint orbits represented on a Lie algebra and the functions in involutions are presented as ad-invariant functions. On the one hand for this kind of functions, the corresponding Hamiltonian systems become a Lax equation and on the other hand they are in involution on the orbits. Whenever studying Poisson commuting conditions the ad-invariance property can be replaced by a weaker one as in [@R1]. In the framework of this theory what we need is a Lie algebra with an ad-invariant metric, a splitting of this Lie algebra into a direct sum as vector subspaces of two subalgebras and a given function. These algebraic tools were used with semisimple Lie algebras, where the Killing form is the natural candidate for the ad-invariant metric.
However there are more Lie algebras admitting an ad-invariant metric. We shall examplify here how can be applied the theory for semisimple Lie algebras, and also for other ones, such as the solvable ones. For the general case one should see that any Lie algebra with an ad-invariant metric can be constructed by a double extension procedure, whose more simple application follows from $\RR^m$. In this way one gets a solvable Lie algebra $\ggo$, that results a semidirect extension of the 2n+1-dimensional Heisenberg Lie algebra $\hh_n$ and that can be endowed with an ad-invariant metric which is an extension of a non degenerate bilinear form on $\RR^{2n}$. But for other cases the resulting Lie algebras could be no semisimple and no solvable.
In any case, the Lie algebra $\ggo$ splits naturally as a direct sum of vector spaces of two subalgebras. Looking at the coadjoint orbits of one of the Lie subalgebras, one gets Hamiltonian systems on these orbits and one can identify the original Hamiltonian system with one of these. In particular for the restriction of the quadratic corresponding to the ad-invariant metric we obtain a Hamiltonian system that becomes a Lax equation, whose solution can be computed with the Adjoint representation.
As example we work out the Toda lattice and the linear equation of motion of n-uncoupled harmonic oscillators. The first one corresponds to a semisimple Lie algebra, and the second one is associated to a solvable one. Furthermore it is proved that the Hamiltonian for the last one is completely integrable on all maximal orbits. We notice that the functions in involution we are making use, are not ad-invariant and they do not satisfy the involution conditions of [@R1].
The setting for the second example applies for quadratic hamiltonians. The Poisson commutativity conditions we get for some polynomials can be read off in the Lie algebra $sp(n)$ of derivations of the Heisenberg Lie algebra of dimension 2n+1 $\hh_n$. In particular for the case of the motion of n-uncoupled harmonic oscillators we need a abelian subalgebra in the Lie algebra of isometries of the Heisenberg Lie group $\HH_n$, endowed with its canonical inner product.
This is not surprising if we consider that symplectic automorphisms of the Heisenberg Lie group produce symplectic symmetries of p-mechanical, quantum and classical dynamics for more general systems than the linear ones (see [@Ki2]).
The appearence of the Heisenberg Lie algebra related to the motion of n-uncoupled harmonic oscillators is not so surprising. In fact, it is known that in quantum mechanics a good approach to the simple harmonic oscillator is through the Heisenberg Lie algebra. In dimension three this is the Lie algebra generated by the position operator $Q$ = [*multiplication by x*]{}, the momentum operator $P = -i \frac
{d}{dx}$ and $1$ with the only non trivial commutation relation $$[Q,P]=1$$ These operators evolve according to the Heisenberg equations $$\frac{dP}{dt}=-Q \qquad \qquad \frac{dQ}{dt}= P$$
An attempt to relate the classical mechanical system of the linear approximation of the motion of n-uncoupled harmonic oscillators was presented by the theory of p-mechanics, which makes use of the representation theory of the Heisenberg Lie group to show that both quantum and classical mechanics can be derived from the same source (see for instance [@Ki1] [@Ki2]). This theory contructs a more general setting that unifies both quantum and classical mechanics. The starting point for p-mechanics is the method of orbit of Kirillov [@K1] [@K2], which says that the orbits of the coadjoint representation of the Heisenberg Lie group parametrise all unitary irreducible representations [@F]. Thus non commutative representations are known to be connected with quantum mechanics. In the contrast commutative representations are related to classical mechanics in the observation that the union of one dimensional representations naturally acts as the classical phase space in p-mechanics. In this theory the time evolution of both quantum and classical mechanics observables can be derived from the time evolution of p-observables, choosen as particular functions or distributions on the Heisenberg Lie group.
These considerations allow to suppose that new applications of the Adler Kostant Symes scheme are possible and maybe it comes a new time to understand old mechanical systems with new tools, which should be developped for these purposes. As an introduction to the topic one can find exceptional ideas in the books of Arnold, Abraham and Marsden, Ratiu and Marsden, etc. all of them classics in the literature concerning classical mechanics.
The chapter is organised as follows: in the first part we present basic ideas concerning symplectic geometry. The second part is devoted to the Adler-Kostant-Symes scheme and the third part to the examples: on the one hand the Toda lattice with generalization in ([@Ko2] and [@Sy]), and on the other hand the systems corresponding to quadratic Hamiltonians on $\RR^{2n}$.
Basic notions on symplectic manifolds
=====================================
In this section we present the basic elements to work with symplectic geometry. Some texts concerning this topic are [@L-M] [@CdS].
Let $M$ denote a differentiable manifold.
A 2-form on $M$, $\omega$ is called a symplectic form if $d\omega =0$ and $\omega_p$ is non degenerate for every $p\in M$.
The pair $(M, \omega)$ is a symplectic manifold.
It follows that the dimension of $M$ must be even.
\[exa1\] Let $\RR^{2n}$ be the usual euclidean space equipped with global coordinates $x_1, \hdots, x_n$, $y_1, \hdots, y_n$. The 2-form given by $$\omega = \sum_{i=1}^n dy_i \wedge dx_i$$ defines a symplectic form on $\RR^{2n}$.
Note that if $(\,,\,)$ denotes the canonical inner product on $\RR^{2n}$ and $J$ the canonical complex structure $$J= \left( \begin{matrix} 0 & -I \\ I & 0\end{matrix} \right)$$ where $I$ is the identity $n \times n$ matrix, then $$\omega(X,Y) =(X,JY)$$
If $(M_1, \omega_1)$ and $(M_2, \omega_2)$ are symplectic manifolds, the direct product $M_1 \times M_2$ is a symplectic manifold.
\[coadac\] Coadjoint orbits. Let $G$ denote a Lie group with Lie algebra $\ggo$ and let $\ggo^{\ast}$ be the dual space of $\ggo$. The coadjoint action of $G$ on $\ggo^{\ast}$ is defined as: $$g \cdot \varphi = \varphi \circ \Ad(g^{-1}) \qquad \qquad g\in G, \varphi\in \ggo^{\ast}.$$ Notice that the orbit throught $\varphi$ is the set $G\cdot \varphi =\{ g \cdot \varphi : g \in G\}$ and the isotropy subgroup at $\varphi$ is $G_{\varphi}=\{ g \in G \, : \, \varphi \circ Ad(g^{-1})=\varphi\}$; thus as usual one has $G \cdot \varphi= G / G_{\varphi}$.
The action of $G$ on $\ggo^{\ast}$ induces an action of $\ggo$ on $\ggo^{\ast}$ as $$X \cdot \varphi= - \varphi \circ \ad(X) \qquad \qquad X\in \ggo, \varphi \in \ggo^{\ast}$$ that cames from the derivative $$\frac{d}{dt}_{|_{t=0}} \exp(tX) \cdot \varphi = -\varphi \circ \ad(X),$$ in other words $$\tilde{X}(\varphi) = - \varphi \circ \ad(X)$$ is the infinitesimal generator induced by $X\in \ggo$ at $\varphi\in \ggo^*$.
Any coadjoint orbit $G\cdot \varphi$ is a symplectic manifold with the 2-form $$\omega_{\varphi}(\tilde{X},\tilde{Y})=-\varphi ([X,Y]), \qquad \varphi \in \ggo^{\ast}, \, X,Y\in \ggo.$$ called [*the Kirillov-Kostant-Souriau*]{} symplectic structure.
Locally any symplectic manifold looks like the example (\[exa1\]) above. This is a classical result of Darboux.
[**Darboux.**]{} Let $(M, \omega)$ be a symplectic manifold. For every $p\in M$ there exists coordinate system $(U, (x^1, \hdots, x^n, y^1, \hdots, y^n))$ such that $p\in U$ and $\omega_{|_U}= \sum_{i=1}^n dy_i \wedge dx_i$.
The symplectic 2-form $\omega$ bilds a isomorphism on $T_xM$ for $x\in M$. In fact, since $\omega_x:T_x M \times T_xM$ is non degenerate, there is a linear isomorphism $K_x: T_x M \to T_x^{\ast}M$ defined by $$K_x(v) (u) = \omega_x(u,v)=(-i_v\omega)(u).$$
Let $H:M \to \RR$ be a differentiable function, since its differential belongs to $T^{\ast}M$, via the isomorphism above we get a vector field $X_H$ given by $$\label{hvf} X_H(x)=K_x( dH_x),$$ that is, $X_H$ is the vector field on $M$ satisfying $$v(H) = dH(v) = \omega(v, X_H)$$ and this is called the [*Hamiltonian vector field*]{} associated to the Hamiltonian function $H$.
For the standard symplectic structure on $\RR^{2n}$ the isomorphism $K_x$ is given by $K_x v= - J v$, where $J$ denotes the canonical complex structure on $\RR^{2n}$.
Let $H\in C^{\infty}(\RR^{2n})$, its associated Hamiltonian vector field is $$X_H(m)=J(\nabla H) = \sum_i (\frac{\partial H}{\partial y_i} \frac{\partial}{\partial x_i}-\frac{\partial H}{\partial x_i} \frac{\partial}{\partial y_i}),$$ where $\nabla H$ is the gradient of $H$, with respect to the canonical inner product.
The Hamiltonian system for a Hamiltonian $H\in C^{\infty}(M, \omega)$ is $$\label{hamsis}
x'(t)=X_H(x(t)).$$
Let $H$ be a smooth function on $\RR^{2n}$, the Hamiltonian equation is the classical one $$\begin{array}{rcl}
x_i' & = & \frac{\partial f}{\partial y_i}\\
y_i' & = & - \frac{\partial f}{\partial x_i}
\end{array}$$ where $x_i$ is actually $x_i(t)$, that is it depends on $t$, for all $i$ (and also for any $y_i$).
\[exa2\] On $\RR^{2n}$ a [*quadratic Hamiltonian*]{} is a smooth function as $$H(x)=\frac12 (Ax,x)\qquad \mbox{ for } \qquad A \mbox{ symmetric linear map,}$$ which yields the Hamiltonian system $$\label{ham1}
x' = JA x$$ In classical mechanic this system describes “small oscillations”, that is, it approximates the motion of a particle on $\RR^n$ or equivalently the motion of $n$ uncoupled particles on $\RR$, near an equilibrium position.
For instance the motion of $n$-uncoupled harmonic oscillators near an equilibrium position can be approximated with $H$ a quadratic Hamiltonian as above by taking $A=I$; therefore (\[ham1\]) becomes $$\label{ho}
\begin{array}{rcl}
x_i'(t) & = & y_i(t)\\
y_i'(t) & = & -x_i(t)
\end{array}$$
where $x(t)=(x_1(t), \hdots, x_n(t), y_1(t), \hdots, y_n(t))$.
In classical mechanics it is usual to name the coordinates as $x_i$ the position coordinates and $y_i$ as the velocity coordinates for every $i=1, \hdots, n$.
A diffeomorphism $\phi$ on a symplectic manifold $(M, \omega)$ is symplectic if $\phi^{\ast} \omega = \omega$.
Recall that the Lie derivative on a smooth manifold $M$ given as $$L_X T = \frac{d}{dt}_{|_{t=0}} \psi_t^{\ast}(T)$$ where $X$ is a vector field on $M$ with one parameter group $\psi_t$ and $T$ a tensor, satisfies the following identities $$\begin{array}{rcl}
L_X & = & i_X d + d i_X\\
L_X i_Y & = & i_{L_X Y} + i_Y L_X =i_{[X,Y]} + i_Y L_X
\end{array}$$ For a proof see for instance [@Wa].
A vector field $X$ on a symplectic manifold $(M, \omega)$ is symplectic if $L_X \omega=0$.
A vector field $X\in \chi(M)$ is symplectic if and only if the one parameter subgroup $\psi_t$ generated by $X$ is symplectic.
If $\psi_t$ is symplectic, using the definition of $L_X$ it is easy to see that $L_X \omega=0$. Conversely assume $L_X \omega =0$, then $$\frac{d}{dt}_{|_{t=s}} \psi_t^{\ast} \omega = \frac{d}{dt}_{|_{t=0}} \psi_t^{\ast} \psi_s^{\ast} \omega = L_X \psi_s^{\ast} \omega = \psi_s^{\ast} L_X \omega=0$$ hence $\psi_t^{\ast} \omega$ is constant. But $\psi_0=Id$ and so $\psi_t^{\ast}\omega = \omega$.
i\) If $\omega$ is a symplectic form then $L_X \omega= di_X \omega$.
ii\) A vector field on $(M, \omega)$ is symplectic if and only if $i_X \omega$ is closed.
A vector field $X$ on a symplectic manifold $(M,\omega)$ is Hamiltonian if and only if $-i_X\omega$ is exact.
The vector field associated to a Hamiltonian function $H$ defined in (\[hvf\]) is Hamiltonian.
Notice that the fact of being $X$ Hamiltonian says that there is $H \in C^{\infty}(M)$ such that $dH =-i_X \omega$, therefore $X$ is symplectic. On the other hand for any $p \in M$ there always exists local solutions to $dH=-i_X \omega$ for any $X\in \chi(M)$. For global solutions we must ask extra conditions as below.
Let $(M, \omega)$ be a symplectic manifold such that $H^1(M, \RR)=0$. Every symplectic vector field on $M$ is Hamiltonian.
A symplectic 2-form $\omega$ on a symplectic manifold $M$ induces a [*Poisson*]{} bracket $\{\,,\,\}$ on $C^{\infty}M$ by: $$\{f, g\}(p)=\omega_p(X_f, X_g)=X_f(g) =-X_g(f)\qquad
\qquad \mbox{ for any } f,g\in C^{\infty}M.$$
Let $C^{\infty}(M)$ is a Lie algebra under the Poisson bracket defined above and $f \to X_f$ is a Lie algebra anti-homomorphism of $C^{\infty}(M)$ into $\chi(M)$.
Since $K_x:T_xM \to T_x^{\ast}M$ is a linear isomorphism, the map $f\to X_f$ is linear. Now we should prove that $[X_f, X_g]=X_{\{f,g\}}$. Using the properties of $L_X$ one gets $$L_{X_f} i_{X_g} \omega = i_{[X_f, X_g]} \omega + i_{X_g} L_{X_f}\omega.$$ Since $L_{X_f}\omega=0$, one gets $$L_{X_f} i_{X_g} \omega = i_{[X_f, X_g]} \omega.$$ The Lie derivative on 1-forms follows $$L_X \theta = i_X d \theta + di_X \theta.$$ Taking $i_{X_g}\omega=d g$ and applying above it holds $$L_{X_f} i_{X_g} \omega = L_{X_f} dg = i_{X_f} d^2 g + di_{X_f} dg = d(X_f(g))= d\{f,g\}.$$ Therefore $$i_{[X_f, X_g]}\omega = d\{f,g\}$$ Thus the left side of the equality above $i_{[X_f, X_g]}\omega_x$ coincides with $-K_x(X_{\{f,g\}}$, and since $K_x$ is an isomorphism $[X_f, X_g]=-X_{\{f,g\}}$.
Recall that a [*Poisson structure*]{} is a bracket $\{\,,\,\}$ on a associative algebra $A$, such that
$\bullet$ $\{\,,\,\}$ is a Lie bracket on $A$ and
$\bullet$ $f\{g,h\} = \{fg, h\} + \{g, fh\}\qquad \mbox{ for all } f,g, h \in A.$
the last one is called the Leibnitz rule. In [@Sy] a such structure is called [*Hamiltonian*]{}.
The space of smooth functions on a differentiable manifold is a associative Lie algebra, hence a natural space to be endowed with a Poisson structure. The Proposition we already proved says that whenever $(M, \omega)$ is a symplectic manifold, $C^{\infty}(M)$ has a Poisson structure induced by $\omega$: the Poisson bracket $\{\, , \}$ is a Lie bracket and the Leibnitz rule holds, since any vector field is a derivation on $C^{\infty}(M)$.
On $\RR^{2n}$, the Poisson structure associated to the standard symplectic form is given by $$\label{pb}
\{f, g \} = (\nabla f, J \nabla g)= \sum_i \frac{\partial f}{\partial x_i} \frac{\partial
g}{\partial y_i}-
\frac{\partial f}{\partial y_i} \frac{\partial g}{\partial x_i}.$$
Let $\ggo$ be a Lie algebra and $\ggo^{\ast}$ its dual. As usual one identifies $\ggo^{\ast}$ with its tangent space. Given a function $F:\ggo^{\ast} \to \RR$, we define the gradient of $F$ at $\alpha\in \ggo^{\ast}$, denoted by $\nabla F(\alpha)$, as an element $\nabla F(\alpha)\in \ggo$ such that $\la \beta, \nabla F(\alpha) \ra = d F_{\alpha} (\beta)$ for any $\beta \in \ggo^{\ast}$, where $\la \, , \, \ra$ denotes the evaluation map.
The Kirillov’s Poisson bracket on $\ggo^{\ast}$ is given by $$\{f,h\}(\alpha)=\la alpha, [\nabla f(\alpha), \nabla h(\alpha)]\ra.$$
If $\{f,g\}=0$ then $g$ is constant on the integral curves of $X_f$.
Assume $x'(t)=X_f(x(t))$ then $$\frac{d}{dt} g(x(t)) = dg(x'(t)) = X_f(g)(x(t))=\{f,g\}(x(t))=0.$$
Thus $g$ is called a [*constant of motion*]{} of the flow defined by $X_f$. Since $\{\, , \, \}$ is skew symmetric, $g$ is a constant of motion of $X_f$ if and only if $f$ is a constant of motion of $X_g$. Constant of motion always exist, in fact $f$ is a constant of motion of $X_f$.
A function $f$ on a 2n-dimensional Poisson manifold $(M, \{ \,, \, \})$ is [*completely integrable*]{} if there exist $n$ functions $f_1, \hdots, f_n\in C^{\infty}M$ such that:
i\) $\{f, f_i\}= 0$, $\{f_i, f_j\}=0$ for all $1\leq i, j \leq n$,
ii\) The differentials $df_1, \hdots, df_n$ are linearly independent on a open set invariant under the flow of $X_f$.
Two functions $f,g: M \to \RR$ such that $\{f,g\}=0$ are said to be in involution or Poisson commute.
A subset $N \subset M$ is [*invariant*]{} under the flow of $X_f$ if the solution $x$ for the Hamiltonian system (\[hamsis\]) corresponding to the Hamiltonian $f$ lies on $N$ if $x(0)\in N$.
On $\RR^{2n}$ for $H(x)=\frac12 (x,x)$ the polynomials $$f_i(x)=\frac12 (p_i^2+q_i^2)\quad i=1, \hdots,n$$ shows that $H$ is completely integrable. In fact it is easy to check that $\{H, f_i\}=0=\{f_i, f_j\}$ for all $i=1, \hdots, n$.
Let $F=(f_1, \hdots , f_n)$, then $F^{-1}(c)$ is a torus which is invariant under the flow generated by $X_H$. Let $(\theta_1, \hdots, \theta_n)$ denote the angle variable on the torus $F^{-1}(c)$. Then $(f_1, \hdots , f_n, \theta_1, \hdots, \theta_n)$ is a local coordinate on $\RR^{2n}$. With these coordinates, the Hamiltonian equation becames $$\begin{array}{rcl}
f_i' & = & 0\\
\theta_i' & = & -1
\end{array}$$ and the coordinate functions satisfy $$\{f_i, f_j\}=\{\theta_i, \theta_j\} = 0, \qquad \{f_i, \theta_j\}=\delta_{ij},$$ therefore the flow $X_h$ is linear on $F^{-1}(c)$ for $c\in \RR^n$.
Moreover since the level sets $\{x\in \RR^{2n}: H(x)=c\}$ are compact we have action angle coordinates (see Liouville Theorem below).
Generally $m$ Poisson commuting functions $f_1, f_2, \hdots, f_m$ on a symplectic manifold $(M, \omega)$ give rise to an action of $\RR^m$ on $M$. Let $(\psi_i)_t$ be the one parameter subgroup generated by $X_{f_i}$. Then $$(t_1, \hdots, t_m)\cdot p=(\psi_1)_{t_1} (\psi_2)_{t_2} \hdots (\psi_m)_{t_m}(p).$$ defines a $\RR^M$ action on $M$. Since $\{f_i, f_j\}=0$ for all $i,j$, the set $N=\{ x\in M,\,:f_i(x)=c_i\}$ is invariant under the $\RR^m$-action, for constants $c_1, \hdots, c_m$. If $N$ is compact, the $\RR^m$-action descends to a torus action on $N$. When $m=1/2 \dim M$, one gets the Liouville theorem.
\[Liouville\] Let $f$ be a completely integrable function on $M$, with $\dim M=2n$, and assume $f_1:=f, f_2, \hdots, f_n$ are commuting Hamiltonians which are linearly independent and let $F=(f_1, \hdots, f_n):M \to \RR^n$ be proper. Then $F^{-1}(c)$ is invariant under the $\RR^n$ action and it descends to a torus $T^n$-action. Let $\theta_1, \hdots, \theta_n$ denote the angle coordinates on the invariant tori. Then $\{f_i, f_j\}=\{\theta_i, \theta_j\}=0$ and $\{f_i, \theta_j\}=c_{ij}(F)$ for some functions $c_{ij}:\RR^n\to \RR$. In particular, the flow of $X_f$ in coordinates $(f_1, \hdots, f_n, \theta_1, \hdots, \theta_n)$ is linear.
Coordinates as above, are called [*action-angle variables*]{} for the Hamiltonian system of $f$.
Symplectic actions: the AKS-Scheme
==================================
Let $M$ denote a differentiable manifold and let $G$ be a Lie group. An differentiable action of $G$ on $M$ is a differentiable map $\eta: G\times M \to M$, $\eta: (g,m) \to \eta(g,m):= g\cdot m$ such that $$\begin{array}{lrlll}
\mbox{ i)} & e \cdot m & = & m & \mbox{ for all } m\in M \mbox{ and }\\
\mbox{ii)} & (g h)\cdot m & = & g\cdot(h \cdot m) &\mbox{ for all } m\in M, g,h \in G.
\end{array}$$ Notice that if $\eta$ is an action, the applications $\eta_g:M \to M$ given by $\eta_g(m)=g \cdot m$ are diffeomorphisms of $M$. In fact, $\eta_g$ are differentiable for any $g$ and they are diffeomorphisms since the inverse of any $\eta_g$ is $\eta_{g^{-1}}$ (see ii) above). Therefore an action of a Lie group on $M$ induces a representation of $G$ on $Diff(M)$ the diffeomorphisms of $M$, given by $g \to \eta_g$.
Let $GL(n, \RR)$ denote the Lie group of non singular transformations of $\RR^n$. This acts on $\RR^n$ as evaluation: $A \cdot v= v$ for $A\in GL(n, \RR)$ and $v\in \RR^n$. It is easy to verify that this is in fact an action.
Let $H$ be a Lie subgroup of a Lie group $G$, then $H$ acts on $G$ by conjugation, $H\times G \to G$, $(h,x) = h^{-1} x h$, for any $h\in H$, $x\in G$. If $H$ is a normal subgroup, one can consider the action of $G$ on $H$ by conjugation.
Let $G$ be a Lie group with Lie algebra $\ggo$, then $G$ acts on $\ggo$ by the Adjoint action, $G\times \ggo \to \ggo$, $(g,X) = \Ad(g) X$, for any $g\in G$, $X\in \ggo$. Recall that $\Ad(g)= dI(g)_e$ where $I_g$ denotes the conjugation by $g$ (see the previous example). It is easy to see that $I_{gh} = I_g \circ I_h$ for all $g,h\in G$, hence the map $G \to GL(\ggo)$ is a representation of $G$, called the Adjoint representation. This has a correlative at the Lie algebra level, the adjoint representation: $\ggo \times \ggo \to \ggo$ given by $X \cdot Y = [X,Y]$ for all $X,Y\in \ggo$.
Recall that in (\[coadac\]) we defined the [*coadjoint action*]{} of a Lie group $G$ on the space $\ggo^{\ast}$, the dual of the Lie algebra $$g \cdot \varphi = \varphi \circ \Ad(g^{-1}) \qquad g\in G, \varphi \in \ggo^{\ast}$$ and also we gave the corresponding action of $\ggo$ on $\ggo^{\ast}$ by $$X \cdot \varphi = -\varphi \circ \ad(X) \qquad X\in \ggo, \varphi \in \ggo^{\ast}.$$
The orbit of an action of a Lie group $G$ on a set $M$ is $$G \cdot m=\{ g \cdot m : g\in G\}$$ and the [*isotropy*]{} or [*stabilizer*]{} group of the action at the point $m$ is the closed subgroup of $G$ given by $$G_m=\{ g\in G \mbox{ such that } g \cdot m=m\}.$$
It is known that the orbit at $m$ is diffeomorphic to the quotient space of $G$ and the isotropy group, $G \cdot m \simeq G/G_m$ (see [@Wa] for instance). Thus any curve at the orbit $G \cdot m$ through $m$ is $\gamma (t)=\exp tX \cdot m$ and this generates the infinitesimal vector $\tilde{X}$ at $T_m(G\cdot m)$ by $$\tilde{X}(m)= \frac{\rm d}{\rm{dt}}_{|_{t=0}} \exp tX \cdot m.$$ Hence the tangent space of a $G$-orbit at $m$ is $$T_m(G \cdot m)=\{ \tilde{X}, \quad X \in \ggo\}$$ being $\ggo$ the Lie algebra of $G$.
Assume $M$ and $N$ are two differentiable maps on which a given Lie group $G$ acts. A map $F: M \to N$ is called [*equivariant*]{} if $F( g \cdot m) = g \cdot F(m)$ for all $m\in M$, $g\in G$. The condition is also expressed as $F$ [*intertwines*]{} the two $G$-actions.
Let $(M, \omega)$ be a symplectic manifold. An action $\eta$ of a Lie group $G$ on $M$ is called symplectic if the diffeomorphisms $\eta_g$ are symplectic maps for any $g\in G$, that is $\eta_g^{\ast} \omega=\omega$.
The coadjoint orbits are examples of symplectic manifolds. Recall that they are endowed with the 2-form given by: $$\omega_{\beta}(\tilde{X},\tilde{Y})=-\beta ([X,Y]), \qquad \beta \in G\cdot \mu$$
which is symplectic. In fact, it is closed since for $X_1, X_2, X_3\in \ggo$ one has $$\omega([\tilde{X_1}, \tilde{X_2}], \tilde{X_3})=-\varphi([[X_1, X_2], X_3]]),$$ hence $$d\omega(\tilde{X_1}, \tilde{X_2}, \tilde{X_3})=-\varphi([[X_1, X_2], X_3]])-\varphi([[X_2, X_3], X_1]])-\varphi([[X_3, X_1], X_2]])=0$$ where the last equality holds after Jacobi for $[\cdot, \cdot]$.
The 2-form $\omega$ is non degenerate on a orbit: let $\varphi\in \ggo^{\ast}$ and let $X\in \ggo$ such that $\omega(\tilde{X}, \tilde{Y})=0$ for all $Y\in \ggo$.
Then $-\varphi([X,Y])=0$ for all $Y\in \ggo$, says that $X \cdot \varphi=0$ implying that $X\in L(G_{\varphi})$. In fact $\exp tX\in G_{\varphi}$ if and only if $\exp tX \cdot \varphi = \varphi$ for $t$ near $0$. Thus taking derivative at $t=0$ we have $X \cdot \varphi=0$, and this is the set corresponding to the Lie algebra of $G_{\varphi}$. Since the tangent space of the orbit at $\varphi$ is $T_{\varphi} (G \cdot \varphi)=\ggo/L(G_{\varphi})$, one gets $\tilde{X}=0$.
An ad-invariant metric on $\ggo$ is a bilinear map $\la \,,\,\ra: \ggo\times \ggo \to \RR$, which is a non-degenerate symmetric and such that $\ad(X)$ is skew symmetric for any $X\in \ggo$, that is $$\la [X,Y], Z\ra +\la Y, [X,Z]\ra=0\qquad \qquad \mbox{ for all } X,Y, Z\in \ggo.$$
This ad-invariant metric gives rise to a bi-invariant pseudo Riemannian metric on a connected Lie group $G$ with Lie algebra $\ggo$; bi-invariant means that the maps $\Ad(g)$ are isometries for all $g\in G$, that is $$\la Ad(g) Y, Ad(g) Z\ra=\la Y, Z\ra \qquad \qquad \mbox{ for all } Y, Z\in \ggo, g\in G,$$ and conversely any bi-invariant pseudo Riemannian metric on $G$ induces an ad-invariant metric on its Lie algebra, just by taking derivative of the last equality at $t=0$ with $g=\exp t X$.
Examples of Lie algebras with ad-invariant metrics are:
a\) semisimple Lie algebras with the Killing form;
b\) semidirect products $\ggo \ltimes_{coad} \ggo^*$ with the canonical neutral metric $$\la (x_1, \varphi_1),(x_2, \varphi_2)\ra=\varphi_1(x_2)+\varphi_2(x_1)$$
An ad-invariant metric $\la \,,\, \ra$ induces a diffeomorphism between the adjoint orbit $G\cdot X$ and the coadjoint orbit $G\cdot \ell_X$ where $\ell_X (Y)=\la X, Y\ra$. In fact $$g \cdot \ell_X (Y)=\la X, Ad(g^{-1}) Y\ra= \la Ad(g) X, Y\ra \qquad \mbox{ for all } X,Y\in \ggo, g\in G,$$ implying that the map $\ell: X \to \ell_X$ is equivariant. Thus the adjoint orbits become symplectic manifolds with the 2-form: $$\omega_X(\tilde{Y}, \tilde{Z})=\la X, [Y,Z]\ra\qquad \mbox{ for } X,Y, Z\in \ggo.$$
We shall consider these ideas to construct Hamiltonian systems on orbits that are included on Lie algebras.
Recall that given a metric $\la \, ,\, \ra$ on $\ggo$ the gradient of a function $f:\ggo \to \RR$ at the vector $X\in \ggo$ is defined by $$\label{gra}
\la \nabla f(X), Y\ra = df_X(Y)\qquad \qquad Y\in \ggo.$$
Suppose $\ggo_+$, $\ggo_-$ are Lie subalgebras of the Lie algebra $\ggo$ such that $$\ggo = \ggo_+ \oplus \ggo_-$$ as a direct sum of linear subspaces, that is $(\ggo, \ggo_+, \ggo_-)$ is a product structure on $\ggo$. The Lie algebra $\ggo$ also splits as $$\ggo=\ggo_+^{\perp} \oplus \ggo_-^{\perp},$$ and $$\ggo_{\pm}^{\perp}\quad \mbox{ is isomorphic as
vector spaces to }\quad \ggo_{\mp}^{\ast}.$$
This follows from the isomorphism $\ell:\ggo \to \ggo^{\ast}$. In fact, let $X\in \ggo_+^{\perp}$ maps to $\ell_X$. Since $\ell_X(Y)=0$ for all $Y\in \ggo_+$, the image of $\ell(\ggo_+^{\perp}$ belongs to $\ggo_-^{\ast}$, and the isomorphism follows from dimensions.
Let $G_-$ denote a subgroup of $G$ with Lie algebra $\ggo_-$. Then the coadjoint action of $G_-$ on $\ggo_-^{\ast}$ induces an action of $G_-$ on $\ggo_+^{\perp}$: for $g_-\in G_-$, $X\in \ggo_+^{\perp}$, $Y\in \ggo_-$ one has: $$g_- \cdot \ell_X (Y)= \la X, \Ad(g^{-1}) Y\ra = \la \Ad(g) X, Y\ra= \pi_{\ggo_+^{\perp}}(\Ad(g_-)X ),Y\ra,$$ where $\pi_{\ggo_+^{\perp}}$ denotes the projection of $\ggo$ on $\ggo_+^{\perp}$; therefore the action is given as $$g_- \cdot X = \pi_{\ggo_+^{\perp}}(\Ad(g_-)X ),$$ and $\ell:\ggo_+^{\perp} \to \ggo_-^{\ast}$ is equivariant.
The infinitesimal generator corresponding to $Y_-\in \ggo_-$ is $$\tilde{Y}_-(X)= \frac{\rm d}{\rm dt}_{|_{t=0}} \exp tY_- \cdot X= \pi_{\ggo_+^{\perp}}([Y_-, X]) \qquad X\in \ggo_+^{\perp}.$$ The orbit $G_-\cdot Y$ becomes a symplectic manifold with the symplectic structure given by $$\omega_X(\tilde{U_-}, \tilde{V_-})=\la X, [U_-,V_-]\ra \qquad
\mbox{ for } U_-, V_- \in \ggo_-, X\in G_- \cdot Y$$ which is induced from the Kostant-Kirillov-Souriau symplectic form on the coadjoint orbits in $\ggo_-^{\ast}$.
Consider a smooth function $f:\ggo \to \RR$ and restrict it to an orbit $G_-\cdot X
:=\mathcal M\subset \ggo_+^{\perp}$. Then the Hamiltonian vector field of the restriction $H=f_{|_{\mathcal M}}$ is the infinitesimal generator corresponding to $-\nabla f_-$ , that is $$\label{e3}
X_H(Y)=-\pi_{\ggo_+^{\perp}}([\nabla f_-(Y),Y])$$ where $Z_{\pm}$ denotes the projection of $Z \in \ggo$ with respect to the decomposition $\ggo=\ggo_+\oplus \ggo_-$. In fact for $Y\in
\ggo_+^{\perp}$, $V_-\in \ggo_-$ we have $$\begin{array}{rcl}
\omega_Y(\tilde{V_-}, X_H) & = & dH_Y(\tilde{V_-}) = \la \nabla f(Y),
\pi_{\ggo_+^{\perp}}([V_-,Y])\ra= \la \nabla f_-(Y),[V_-,Y]\ra \\
& = & \la Y,[\nabla f_-(Y),V_-]\ra = \omega_Y(\tilde{\nabla
f_-(Y)},\tilde{V_-}).
\end{array}$$ Since $\omega$ is non degenerate, one gets (\[e3\]).
Therefore the Hamiltonian equation for $x:\RR \to \ggo$ follows $$\label{e4}
x'(t)=-\pi_{\ggo_+^{\perp}}([\nabla f_-(x),x]).$$ In particular if $f$ is ad-invariant then $0 = [\nabla f(Y),Y]=
[\nabla f_-(Y),Y]+
[\nabla f_+(Y),Y]$. Since the metric is ad-invariant $[\ggo_+, \ggo_+^{\perp}]\subset \ggo_+^{\perp}$, in fact $$\la [\ggo_+, \ggo_+^{\perp}], \ggo_+ \ra= \la \ggo_+^{\perp}, [\ggo_+, \ggo_+]\ra =0.$$
Hence the equation (\[e4\]) takes the form $$\label{e5}
x'(t)=[\nabla f_+(x),x]=[x,\nabla f_-(x)],$$ that is, (\[e4\]) becomes a [*Lax equation*]{}, that is, it can be written as $x'=[P(x),x]$.
If we assume now that the multiplication map $G_+\times G_- \to G$, $(g_+, g_-) \to g_+g_-$, is a diffeomorphism, then the initial value problem $$\label{e6}
\left\{ \begin{array}{rcl}
\frac{dx}{dt} & = & [\nabla f_+(x),x] \\
x(0) & = & x_0
\end{array}
\right.$$ can be solved by factorization. In fact if $\exp t\nabla
f(x_0)=g_+(t)g_-(t)$, then $x(t)=\Ad(g_+(t))x_0$ is the solution of (\[e6\]).
[Remark.]{} If the multiplication map $G_+ \times G_- \to G$ is a bijection onto an open subset of $G$, then equation (\[e4\]) has a local solution in an interval $(-\varepsilon, \varepsilon)$ for some $\varepsilon >0$.
The theory we already exposed shows the application of Lie theory to the study of ODE’s as in equation (\[e5\]). Even when it is possible to give the solution, one need more information. This can be obtained from involution conditions. They help in some sense to control the solutions.
A first step in the construction of action angle variables is to search for functions which Poisson commute. The Adler-Kostant-Symes Theorem shows a way to get functions in involution on the orbits $\mathcal M$. We shall formulate it in its classical Lie algebra setting.
\[AKS1\] Let $\ggo$ be a Lie algebra with an ad-invariant metric $\la \,, \,\ra$. Assume $\ggo_-, \ggo_+$ are Lie subalgebras such that $\ggo=\ggo_-\oplus \ggo_+$ as direct sum of vector subspaces. Then any pair of ad-invariant functions on $\ggo$ Poisson commute on $\ggo_+^{\perp}$ (resp. on $\ggo_-^{\perp}$).
Sometimes the ad-invariant condition is too strong, so the following version of the previous Theorem given by Ratiu [@R1] asks for a weaker condition.
\[AKS2\] Let $\ggo$ be a Lie algebra carrying an ad-invariant metric $\la \,, \,\ra$. Assume it admits a splitting into a direct sum as vector spaces $\ggo=\ggo_+\oplus \ggo_-$, where $\ggo_+$ is an ideal and $\ggo_-$ is a Lie subalgebra. If $f,h$ are smooth Poisson commuting functions on $\ggo$, then the restrictions of $f$ and $h$ to $\ggo_+^{\perp}$ are in involution in the Poisson structure of $\ggo_+^{\perp}$.
This theorem was used in [@R2] to prove the involution of the Manakov integrals for the free n-dimensional rigid body motion.
Applications of the Adler-Kostant-Symes-scheme to classical mechanics
=====================================================================
In this section we show the explicit use of the theory above in some Lie groups and Lie algebras. The first example is done with semisimple Lie algebras, and it is known as the Toda Lattice.
The Toda lattice
----------------
The Toda lattice is the mechanical system which describes the motion of n particles on a line with an exponential restoring force, that is the Hamiltonian function on $\RR^{2n}$ is $$H(x,y)=\frac12 \sum_{i=1}^n y_i^2+\sum_{i=1}^{n-1} e^{x_i-x_{i-1}}.$$ The phase space is $\RR^{2n}$ which is a symplectic manifold with its canonical symplectic structure. It follows that the Hamiltonian equation is
$$\label{toda1}
\begin{array}{rcl}
x_k' & = & y_k \\
y_k' & = & e^{x_{k-1} - x_k} - e^{x_k - x_{k+1}}
\end{array}$$
and with $e^{x_{0} - x_1} =0= e^{x_n - x_{n+1}}$. Flaschka considered a change of coordinates (called Falschka transform) as follows $$\phi:\RR^{2n} \to \RR^{2n}, \qquad \phi(x,y)=(a,b)$$ where $$\begin{array}{rcll}
a_k & = &-\frac12 y_k & 1 \leq k \leq n,\\
b_k & = & \frac12 e^{\frac12{x_k - x_{k+1}}} & 1 \leq k \leq n-1 \\
b_n & = & \frac12 e^{\frac{x_n}{2}}
\end{array}$$ Therefore the equation (\[toda1\]) yields
$$\begin{array}{rcll}
a_k' & = & 2(b_k^2 - b_{k-1}^2) & 1 \leq k \leq n \\
b_k' & = & b_k(a_{k+1} - a_{k}) & 1 \leq k \leq n
\end{array}$$
with $a_{n+1}=0=b_0$. Notice that $\sum_i y_i'=0$. Assume $\sum x_i=\sum y_i=0$ and let $V=\{(x,y) / \sum_i x_i = 0=\sum_i y_i\}$, then the system above becomes $$\label{toda2}
\begin{array}{rcll}
a_k' & = & 2(b_k^2 - b_{k-1}^2) & 1 \leq k \leq n-1 \\
b_k' & = & b_k(a_{k+1} - a_{k}) & 1 \leq k \leq n-1,
\end{array}$$
Consider $\ggo$ the semisimple Lie algebra of traceless real matrices $sl(n, \RR$ equipped with the ad-invariant metric $\la x, y\ra= tr(x,y)$ for all $x,y \in sl(n,\RR)$.
Let $\ggo_+=so(n)$ the Lie subalgebra of skew symmetric real matrices and $\ggo_-$ the Lie algebra of upper triangular matrices of trace zero.
Then $\ggo_+^{\perp}$ is the space of real symmetric matrices in $sl(n,\RR)$ and $\ggo_-^{\perp})$ is the space of strictly upper triangular matrices in $sl(n, \RR)$.
The coadjoint orbit $\mathcal M= G_-\cdot x_0$ for $x_0=\sum_{i=1}^{n-1} e_{i, i+1} + e_{i+1,i}$ is the set of tri-diagonal real symmetric matrices $$\sum_{i=1}^n a_{i} E_{i,i}+ \sum_{i=1}^{n-1} b_i (E_{i, i+1} + E_{i+1, i})\qquad \sum_i a_i=0,\quad b_i >0 \quad \forall i.$$ where $E_{i,j}$ denotes the matrix with a 1 at the place $i,j$ and 0 in the others components.
Let $f: sl(n,R) \to \RR$ be the function given by $f(X)=\frac12 \la X, X\ra= \frac12 tr(X X)$. It is easy to see that the gradient of $f$ at $X$ is $X$, and hence applying the theory of the previous section we get (\[e5\]) $$x'=[x_+ , x]$$ for $x_+ \subset \ggo_+$ a curve in $\ggo_+$. Writing the last system in terms of coordinates $(a,b)$ we get the system (\[toda2\]).
A generalization of this system can be read in [@Sy], where also aplications of the theory to other differential equations are explained.
The motion of n uncoupled Harmonic oscillators
----------------------------------------------
Recall that the motion of $n$-uncoupled harmonic oscillators near an equilibrium position can be approximated with $H$ the quadratic Hamiltonian as $\frac12(x,x)$ where $(\, , \, )$ is the canonical inner product in $\RR^{2n}$. Let $\omega$ the canonical symplectic structure, the corresponding Hamitonian system follows $$\label{ho}
\begin{array}{rcl}
x_i'(t) & = & y_i(t)\\
y_i'(t) & = & -x_i(t)
\end{array}$$ where $x(t)=(x_1(t), \hdots, x_n(t), y_1(t), \hdots, y_n(t))$.
The associated Poisson structure on $\RR^{2n}$ is given as follows $$\label{pb}
\{f, g \} = (\nabla f, J \nabla g)= \sum_i \frac{\partial f}{\partial x_i} \frac{\partial
g}{\partial y_i}-
\frac{\partial f}{\partial y_i} \frac{\partial g}{\partial x_i}.$$ for smooth functions $f, g $ on $\RR^{2n}$. Thus with respect to the Lie bracket $\{ \,, \,\}$ the subspace over $\RR$ generated by the functions $H=\frac12\sum_i(x_i^2+y_i^2)$, the coordinates $x_i$, $y_i$, and $1$ form a solvable Lie algebra of dimension 2n+2, which is a semidirect extension of the Heisenberg Lie algebra spanned by the functions $x_i,y_i,1$ i=1, $\hdots$,n. In fact they obey the following non trivial rules $$\{x_i, y_j \} =\delta_{ij} \qquad \{H, x_i \} =-y_i \qquad \{H, y_i\} =x_i.$$
In order to simplify notations let us rename these elements identifying $X_{n+1}$ with $H$, $X_i$ with $x_i$, $Y_i$ with $y_i$ and $X_0$ with the constant function 1 $$\begin{array}{rcl}
1 & \leftrightarrow & X_0\\
x_i & \leftrightarrow & X_i \\
y_i & \leftrightarrow & Y_i \\
H & \leftrightarrow & X_{n+1}
\end{array}$$ and set $\ggo$ denotes the Lie algebra generated by these vectors with the Lie bracket $[\cdot , \cdot]$ derived from the Poisson structure. This Lie algebra is known as a [*oscillator*]{} Lie algebra.
The Lie algebra $\ggo$ splits into a vector space direct sum $\ggo = \ggo_+ \oplus \ggo_-$, where $\ggo_{\pm}$ denote the Lie subalgebras $$\label{deco} \ggo _- = span\{X_0, X_i, Y_j\}_{i,j=1, \hdots n},\qquad
\qquad \ggo_+ = \RR{X_{n+1}}.$$
Notice that $\ggo_-$ is isomorphic to the 2n+1-dimensional Heisenberg Lie algebra we denote $\hh_n$.
The quadratic form on $\ggo$ which for $X =x_0(X) X_0 + \sum_i (x_i(X) X_i + y_i(X) Y_i) + x_{n+1}(X)X_{n+1}$ is given by $$f(X)= \frac12 \sum_i(x_i^2+ y_i^2)+ x_0x_{n+1}$$ induces an ad-invariant metric on $\ggo$ denoted by $\la \,, \,\ra$. It is easy to show that the gradient of $f$ at a point $X$ is $$\nabla f(X) = X.$$
The restriction of the quadratic form to $\vv:=span\{X_i, Y_j\}$ i, j=1, $\hdots $, n, coincides with the canonical one $(\, , \, )$ on $\RR^{2n}\simeq \vv$.
The metric induces a decomposition of the Lie algebra $\ggo$ into a vector subspace direct sum of $\ggo_+^{\perp}$ and $\ggo_-^{\perp}$ where $$\ggo_-^{\perp} =
span\{X_0\} \qquad \qquad \ggo_+^{\perp} = \RR X_{n+1} \oplus \,span\{X_i,
Y_j\}_{i,j=1, \hdots, n},$$ and it also induces linear isomorphisms $\ggo_{\pm}^{\ast}\simeq\ggo^{\perp}_{\mp}$. Let $G$ denote a Lie group with Lie algebra $\ggo$ and $G_{\pm}\subset G$ is a Lie subgroup whose Lie algebra is $\ggo_{\pm}$. Hence the Lie subgroup $G_-$ acts on $\ggo_+^{\perp}$ by the “coadjoint” representation; which in terms of $U_-\in \ggo_-$ and $V\in \ggo_+^{\perp}$ is given by $$\label{m2}
\begin{array}{rcl}
\ad^{\ast}_{U_-} V & = & x_{n+1}(V) \sum_i(y_i(U)
X_i - x_i(U) Y_i)
\end{array}$$ It is not difficult to see that the orbits are 2n-dimensional if $x_{n+1}(V) \ne 0$ and furthermore $V$ and $W$ belong to the same orbit if and only if $x_{n+1}(V)=x_{n+1}(W)$, hence the orbits are parametrized by the $x_{n+1}$-coordinate; so we denote them by $\mathcal M_{x_{n+1}}$. They are topologically like $\RR^{2n}$. In fact $\mathcal M_{x_{n+1}}= G_- \cdot V\simeq
\HH_n/Z(\HH_n)$, where $\HH_n$ denotes the Heisenberg Lie group with center $Z(\HH_n)$.
Equipp these coadjoint orbits with the canonical symplectic structure, that is for $U_-, V_- \in \ggo_-$ take $$\omega_Y(\tilde{U}_-, \tilde{V}_-)=\la Y, [U_-, V_-]\ra = x_{n+1}(Y) \sum_{i=1}^n (x_i(U_-) y_i(V_-)-x_i(V_-)y_i(U_-)).$$ Indeed on the orbit $\mathcal M_1$ the coordinates $x_i, y_j$, $i,j =1, \hdots n$, are the canonical symplectic coordinates and one can identify this orbit with $\RR^{2n}$ in a natural way. This says that the identification is a symplectomorphism between $\RR^{2n}$ with the canonical symplectic structure and the orbit with the Kirillov-Kostant-Souriau symplectic form.
Consider $H$, the restriction to a orbit $\mathcal M_{x_{n+1}}$ of the function $f$. Since $f$ is ad-invariant the Hamiltonian system of $H=f_{|_{\mathcal M_{x_{n+1}}}}$ reduces to $$\label{osc}
\begin{array}{rcl}
\frac{\rm dx}{\rm dt}& = & [x_{n+1}X_{n+1}, x_{\vv}+ x_{n+1} X_{n+1}]\\
x(0) & = & x^0
\end{array}$$ where $x^0= x_{\vv}^0 +x_{n+1}^0 X_{0}$ and $x_{\vv}^0=\sum_i (x_i^0 X_i + y_i^0
Y_i)$.
For $x_{n+1}\equiv x_{n+1}^0\equiv 1$ this system is that one we get on $\RR^{2n}$.
The trajectories $x(t)$ with coordinates $x_i(t)$, $y_j(t)$, $x_{n+1}^0$ are parametrized circles of angular velocity $x_{n+1}^0$, for all i,j, that is $$\begin{array}{rcl}
x_i(t) & = & x_i^0 \cos(x_{n+1}^0 t) + y_i^0 \sin( x_{n+1}^0 t)\\
y_j(t) & = & - x_j^0 \sin(x_{n+1}^0 t) + y_j^0 \cos(x_{n+1}^0 t)\\
x_{n+1}(t) & = & x_{n+1}^0
\end{array}$$ This solution coincides with that computed in the previous section, when we considered systems on coadjoint orbits. In fact it can be written as $$x(t)= \Ad(exp\,\, t x_{n+1}^0 X_{n+1}) x^0,$$ and one verifies that the flow at the point $X^0 \in \ggo_+^{\perp}$ is $$\label{flow}
\begin{array}{rcl}
\Delta^t(X^0) & = &
\sum_i [(x_i^0 \cos(x_{n+1}^0 t) + y_i^0 \sin( x_{n+1}^0 t))X_i +
(-x_i^0 \sin(x_{n+1}^0 t) + \\ \\
& & y_i^0 \cos(x_{n+1}^0 t))Y_i] + x_{n+1}^0 X_{n+1}
\end{array}$$
By taking $L$ and $M$ the following matrices: $$M = \left(
\begin{matrix}
0 & x_{n+1} & 0& 0 &&& &0 & 0\\ -x_{n+1} & 0 & 0 & 0
&&&&0 &
0 \\ 0 & 0 & 0 & x_{n+1} &&&&0& 0\\ 0& 0 & -x_{n+1}&
0&&&&0&0\\ & & & & \ddots & & &\vdots & \vdots\\ & & & & & 0 &
x_{n+1} &0 & 0\\ & & & & 0& -x_{n+1}& 0 &0& 0\\ 0& 0&
\hdots
& & & & & 0 & 0\\ 0 & 0 & \hdots & & & & & 0 & 0
\end{matrix}
\right)$$ $$L = \left(
\begin{matrix}
0 & x_{n+1} & 0& 0 & & & & &x_1\\ -x_{n+1} & 0 & 0 & 0
& &
& & & y_1\\ 0 & 0 & 0 & x_{n+1} & & & & & x_2\\ 0& 0 &
-x_{n+1} & 0 & & & & & y_2\\ & & & & \ddots & & & \vdots &
\vdots\\ & & & & & & x_{n+1} & 0& x_n\\ & & & & & -x_{n+1}
& 0 &0 & y_n\\ -\frac12 y_1& \frac12 x_1&-\frac12 y_2 &\frac12
x_2 &\hdots & -\frac12 y_n& \frac12 x_n & 0 & 0\\ 0& 0 & 0 & 0 &
\hdots & 0& 0& 0 & 0
\end{matrix}
\right)$$ we get $L^{\prime} = [M, L]= ML -
LM$, the Lax pair equation.
Quadratic Hamiltonians and coadjoint orbits
===========================================
In this section we shall prove that Hamiltonian systems corresponding to quadratic Hamiltonians in $\RR^{2n}$ of the form $H(x) = \frac12(Ax,x)$ where $A$ is a non singular symmetric map, can be described using the scheme of Adler-Kostant-Symes on a solvable Lie algebra.
Let us consider the linear system of one degree of freedom on $\RR^{2n}$ with Hamiltonian given by: $$H(x)= \frac12 (Ax, x)$$ where $x=(q_1, \hdots, q_n, p_1, \hdots, p_n)$ is a vector in $\RR^{2n}$ written in a symplectic basis and $A$ is a non singular symmetric linear operator with respect to the canonical inner product $(\, ,\, )$. This yields the following Hamiltonian equation
$$(\ref{ham1}) \qquad \qquad \qquad \qquad \qquad
x'=JA x , \qquad\qquad \mbox{ with } J = \left( \begin{matrix} 0 & -Id \\ Id & 0
\end{matrix} \right)$$
and being $Id$ the identity. The phase space for this system is $\RR^{2n}$. We shall construct a solvable Lie algebra that admits an ad-invariant metric on which the system (\[ham1\]) can be realized as a Hamiltonian system on coadjoint orbits. Moreover it can be written as a Lax pair equation.
Let $b$ denote the non degenerate bilinear form on $\RR^{2n}=span \{X_i, Y_j\}_{i,j=1}^n$ given by $b(X,Y)=(AX,Y)$. In our terms, $b$ defines a metric on $\RR^{2n}$ but it is not necessary definite. Note that the linear $JA$ is non singular and skew symmetric with respect to $b$, where $J$ is the canonical complex structure on $\vv\simeq \RR^{2n}$ as above: $$b(JAX,Y)=(AJAX,Y)=(JAX,AY)=-(AX,JA Y)=-b(X,JA).$$ Let $\ggo$ denote the Lie algebra $\ggo$ which as vector space is the diract sum $$\ggo=\RR X_0 \oplus \vv \oplus \RR X_{n+1}$$ where $\vv= \RR^{2n}$ and with the Lie bracket given by the non trivial relations $$\label{lb}
[U, V] = b(JA U, V) X_0 \qquad \ [X_{n+1}, U ] = JA U \quad\mbox{ for all } U\in \vv.$$
Thus in this way one defines a structure of a solvable Lie algebra on $\ggo$. Note that $A=Id$ is the particular case we considered in the previous subsection.
This Lie algebra $\ggo$ can be equipped with the ad-invariant metric defined by $$\label{metric}
\la x_0^1X_0 + U^1 + x_{n+1}^1 X_{n+1}, x_0^2X_0 + U^2 + x_{n+1}^2 X_{n+1}\ra = b(U^1, U^2) + ( x_0^1 x_{n+1}^2
+ x_0^2 x_{n+1}^1).$$
Thus if $\la \,, \,\ra_{\vv}$ denotes the restriction of the metric of $\ggo$ to $\vv=span\{X_i, Y_j\}_{i,j=1, \hdots , n}$, then clearly $\la \, , \,\ra$ is a generalization of the non degenerate symmetric bilinear map $b$ of $\RR^{2n}$. Moreover $\ggo$ admits a orthogonal splitting $$\ggo= span\{X_0,X_{n+1}\}\oplus \vv.$$
Denote by $\ggo_{\pm}$ the Lie subalgebras $$\ggo_+=\RR X_{n+1}, \qquad \ggo_-=\RR X_0 \oplus span\{X_i,, Y_i\}.$$ They induce the splitting of $\ggo$ into a vector space direct sum $\ggo = \ggo_+ \oplus \ggo_-$, which by the ad-invariant metric gives the following linear decomposition $\ggo= \ggo_+^{\perp} \oplus \ggo_-^{\perp}$, direct sum as vector spaces, for $$\ggo_-^{\perp} = \RR X_0 \qquad \qquad \ggo_+^{\perp} = span\{X_i, Y_i\}_{i=1, \hdots, n}
\oplus \RR X_{n+1}.$$ Note that $\ggo_-$ is an ideal of $\ggo$ isomorphic to the 2n+1-dimensional Heisenberg Lie algebra $\hh_n$.
Let $G$ denote a Lie group with Lie algebra $\ggo$, set $G_-\subset G$ the Lie subgroup with Lie subalgebra $\ggo_-$. As we already explained $G_-$ acts on $\ggo_+^{\perp}$ by the coadjoint action $$g_- \cdot X= \pi_{\ggo_+^{\perp}}(\Ad(g_-)X ) \quad g_-\in G_-, \quad X \in
\ggo_+^{\perp},$$ where $\pi_{\ggo_+^{\perp}}$ is the projection of $\ggo$ on $\ggo_+^{\perp}$, which in infinitesimal terms gives the following action of $\ggo_-$ on $\ggo_+^{\perp}$ $$\label{m22}
\begin{array}{rcl}
\ad^{\ast}_{U} V : = U \cdot V & = & x_{n+1}(V) JA X_{\vv}(U)
\end{array} \qquad \mbox{ for } U\in \ggo_-, \, V\in \ggo_+^{\perp}.$$ being $X_{\vv}(U)$ the projection of $U$ onto $\vv$ with respect to the orthogonal splitting $\ggo = span\{X_0,\, X_{n+1}\} \oplus \vv$.
The orbits are 2n-dimensional if $x_{n+1}(V) \ne 0$ and furthermore $V$ and $W$ belong to the same orbit if and only if $x_{n+1}(V)=x_{n+1}(W)$, and therefore one parametrizes the orbits by the $x_{n+1}$-coordinate and one enotes them by $\mathcal M_{x_{n+1}}$. The orbits are topologically like $\RR^{2n}$ since they are diffeomorphic to the quotient $\HH_n/Z(\HH_n)$, if $Z(\HH_n) = \RR X_0$ is the center of the Heisenberg subgroup.
Endow the orbits with the canonical symplectic structure of the coadjoint orbits, that is for $X\in \ggo_+^{\perp}$, $U_-, V_-\in \ggo_-$ set $$\omega_X(\tilde{U_-}, \tilde{V_-})=\la X, [U_-, V_-]\ra= x_{n+1}(X) b(JA U_{\vv}, V_{\vv}).$$
Consider $f:\ggo \to \RR$ the ad-invariant function given by $$f(X)= \frac12 \la X, X\ra.$$ The gradient of the function $f$ at a point $X$ is the so called position vector $$\nabla f(X) = X.$$ Since $f$ is ad-invariant the Hamiltonian system of $H=f_{|_{\mathcal M_{x_{n+1}}}}$, the restriction of $f$ to the orbit $\mathcal M_{x_{n+1}}$, given by (\[e5\]) becomes $$\label{ipe}
\begin{array}{rcl}
\frac{dx}{dt}& = & [\nabla f_+(x),x] = [x_{n+1}X_{n+1}, x_{\vv} + x_{n+1} X_{n+1}]= x_{n+1} JA x_{\vv}\\
x(0) & = & X^0
\end{array}$$ where $X^0 \in \ggo_+^{\perp}$.
Thus this Hamiltonian system written as a Lax pair equation is equivalent to (\[ham1\]) for $x_{n+1} = x_{n+1}^0 = 1$. The solution $X(t)$ for the initial condition $X^0 \in \ggo_+^{\perp}$ can be computed via the Adjoint map on $G$, that is, $$X(t)= \Ad(exp\,\, t x_{n+1}^0 X_{n+1}) X^0.$$ The previous explanations prove the following result.
Let $H(X)= \frac12(AX,X)$ be a quadratic Hamiltonian on $\RR^{2n}$ with corresponding Hamiltonian system (\[ham1\]). Then $H$ can be extended to a quadratic function $f$ on a solvable Lie algebra $\ggo$ containing the Heisenberg Lie algebra as a proper ideal. The function $f$ induces a Hamiltonian system on coadjoint orbits of the Heisenberg Lie group, that can be written as a Lax pair equation and which is equivalent to (\[ham1\]). Moreover the trajectories on $\RR^{2n}$ for the initial condition $V^0$ can be computed with help of the Adjoint map on $\ggo$. Explicitely they are the curves $x(t) = \exp^{t J A} V^0$, where $\exp$ denotes the usual exponential map of matrices.
If we take $L, M \in M(2n+2, \RR)$ as $$M = \left(
\begin{matrix}
x_{n+1}JA & 0 & z \\
0 & 0 & 0 \\ 0 & 0 & 0
\end{matrix}
\right) \quad
L = \left(
\begin{matrix}
x_{n+1} JA & 0 & z \\
i\frac12z^T & 0 & 0 \\ 0 & 0 & 0 \end{matrix}
\right)$$ where $z^T=(x_1, x_2, \cdots, x_n, y_1, y_2, \hdots, y_n)$ then the Hamiltonian equation can be written in the following way $$L^{\prime} = [M, L].$$
As example of the previous construction consider the linear approximation of the motion of n uncoupled inverse pendula. This corresponds to the Hamiltonian $
H(x)= \frac12 (Ax, x)$ with $$A=\left( \begin{matrix} Id & 0 \\ 0 & -Id
\end{matrix} \right).$$ This yields the Hamiltonian system $x' = JAx$, which in coordinates takes the form $$\label{33}
\begin{array}{rclcl}
\frac{dx_i}{dt} & = & y_i\\
\frac{dy_i}{dt} & = & {x_i}
\end{array}$$
As we said the phase space is $\RR^{2n}$. In the setting of the AKS scheme we can construct coadjoint orbits $\mathcal M$ of the Heisenberg Lie group, that are included in a solvable Lie algebra $\ggo$ with Lie bracket (\[lb\]) and ad-invariant metric (\[metric\]). The Hamiltonian system for the restriction to the orbits of the ad-invariant function on $\ggo$, $f(X)=\frac12\la X, X \ra$, can be written as $$\label{ipel}
\begin{array}{rcl}
\frac{dx}{dt}& = & [x_{n+1}X_{n+1}, x_{\vv} + x_{n+1} X_{n+1}]\\
x(0) & = & X^0
\end{array}$$ where $X^0= \sum_i (x_i^0 X_i + y_i^0 Y_i)+x_{n+1}^0 X_{n+1}$. The Hamiltonian system above on the coadjoint orbit $\mathcal M_1$ written in coordinates is clearly equivalent to (\[33\]).
The trajectories on $\ggo_+^{\perp}$, $x= \sum_i (x_i(t)X_i+ y_i(t) Y_i) + x_{n+1} X_{n+1}$ are parametrized by $$\begin{array}{rcl}
x_i(t) & = & x_i^0 \cosh(x_{n+1}^0 t) + y_i^0 \sinh( x_{n+1}^0 t)\\
y_i(t) & = & x_i^0 \sinh(x_{n+1}^0 t) + y^0_i \cosh(x_{n+1}^0 t)\\
x_{n+1}(t) & = & x_{n+1}^0
\end{array}$$
The flow at the point $X^0 \in \ggo_+^{\perp}$ is $$\label{iflow}
\begin{array}{rcl}
\Delta^t(X^0) & = & \sum_i[( x_i^0 \cosh(x_{n+1}^0 t) - y_i^0 \sinh( x_{n+1}^0 t)X_i
+ \\
& & + (x_i^0 \sinh(x_{n+1}^0 t) + y_i^0 \cosh(x_{n+1}^0 t)Y_i] +
x_{n+1}^0 X_{n+1}
\end{array}$$
.2cm
The system (\[ipel\]) is a Lax pair equation $L^{\prime} = [M, L]= ML -
LM$, and has a matricial representation by choosing $L$ and $M$ the following matrices in $M(2n+2,\RR)$: $$M = \left(
\begin{matrix}
0 & x_{n+1}& 0& 0 &&& &0 & 0\\ x_{n+1}& 0 & 0 & 0
&&&&0 &
0 \\ 0 & 0 & 0 & x_{n+1} &&&&0& 0\\ 0& 0 & x_{n+1}&
0&&&&0&0\\ & & & & \ddots & & &\vdots & \vdots\\ & & & & & 0 &
x_{n+1}&0 & 0\\ & & & & 0& x_{n+1}& 0 &0& 0\\ 0& 0&
\hdots
& & & & & 0 & 0\\ 0 & 0 & \hdots & & & & & 0 & 0
\end{matrix}
\right)$$ $$L = \left(
\begin{matrix}
0 & x_{n+1} & 0& 0 & & & & &x_1\\ x_{n+1} & 0 & 0 & 0
& &
& & & y_1\\ 0 & 0 & 0 & x_{n+1} & & & & & x_2\\ 0& 0 &
x_{n+1} & 0 & & & & & y_2\\ & & & & \ddots & & & \vdots &
\vdots\\ & & & & & & x_{n+1} & 0& x_n\\ & & & & & x_{n+1}
& 0 &0 & y_n\\ -\frac12 y_1& \frac12 x_1&-\frac12 y_2 & \frac12
x_2 &\hdots & -\frac12 y_n& \frac12 x_n & 0 & 0\\ 0& 0 & 0 & 0 &
\hdots & 0& 0& 0 & 0
\end{matrix}
\right)$$
Now we shall investigate involution conditions on the coadjoint orbits of the Heisenberg Lie group for the restrictions of the quadratic functions $f(X)=\frac12 \la X,X\ra$, where $\la \,,\, \ra$ denotes the ad-invariant metric on the solvable Lie algebra $\ggo$.
Let $g_i, g_j$ be two quadratics on $\RR^{2n}$ that are realted to the symmetric maps $A_i, A_j:\vv \to \vv$ respectively, that is $$g_i(X) = \frac12 (A_i X, X) \qquad \qquad g_j(X)=\frac12 (A_jX,X).$$ Consider quadratic functions on the solvable Lie algebra $\ggo$, which are extensions of $g_i, g_j$ to $\RR X_0 \oplus \RR X_{n+1}$, for instance as $$g_i(X) = \frac12 (A_i X_{\vv}, X_{\vv}) + x_0 x_{n+1}\qquad \qquad g_j(X)=\frac12 (A_jX_{\vv},X_{\vv}) +x_0 x_{n+1}.$$ For the following results these extensions are not unique. For instance extending them trivially we get the same conclusions.
Let $H_i, H_j$ denote the restrictions of $g_i, g_j$ to the orbits $\mathcal M_{x_{n+1}}$ and let $X\in \mathcal M_{x_{n+1}} \subset \ggo_+^{\perp}$. The symplectic structure on the orbits induces a Poisson bracket which for the functions $H_i, H_j$ follows: $$\{H_i, H_j\}(X) = \la X, [\nabla {g_i}_-(X), \nabla {g_j}_-(X)]\ra$$ By computing one can see that the gradients of $g_i$ and $g_j$ are $$\nabla g_i(X) = A^{-1} A_i X_{\vv}+x_0X_0 + x_{n+1} X_{n+1} \qquad \nabla g_j(X) = A^{-1} A_j X_{\vv}+x_0 X_0 + x_{n+1} X_{n+1}.$$
Thus we are ready to prove the following result.
\[[@O2]\] \[c11\] The functions $H_i, H_j$ are in involution on the orbits $\mathcal
M_{x_{n+1}}$ if and only if $$\label{c1}
[JA_i, JA_j]=0$$ where $J$ is the canonical complex structure on $\RR^{2n}$.
Let $X\in \mathcal M_{x_{n+1}} \subset \ggo_+^{\perp}$. For the functions $H_i, H_j$ the Poisson bracket on the orbit $\mathcal M_{x_{n+1}}$ follows: $$\begin{array}{rcl}
\{H_i, H_j\}(X) & = & \la X, [A_i X_{\vv}, A_jX_{\vv}]\ra = \la x_{n+1} [X_{n+1},
A^{-1}A_i X_{\vv}], A^{-1}A_j X_{\vv}\ra \\
& = & x_{n+1} \la J A_i X_{\vv}, A^{-1} A_jX_{\vv}\ra = x_{n+1} (J A_i X_{\vv}, A_jX_{\vv})
\end{array}$$ Therefore $\{H_i, H_j \}(X)=0$ if and only if $( A_j J A_i X_{\vv}, X_{\vv}\ra=0$ which is equivalent to $A_j J A_i= A_i J A_j$, if and only if $JA_j J A_i= JA_i J A_j$, that is $[JA_i, JA_j]=0$.
The natural question is what is the meanning of (\[c1\])?
Fix $\la \, , \,\ra'$ the inner product on $\hh_n$ defined so that the vectors $X_i, Y_j, X_0$ are orthonormal for all i,j=1,$\hdots$, n. The metric is an extension of the canonical one on $\RR^{2n}$. The Lie bracket on $\hh_n = \RR X_0 \oplus \vv$ where $\RR^{2n}\simeq \vv = span\{ X_i, Y_j\}_{i,j=1, \hdots, n}$ is expressed as $$\la [X, Y], x_0 X_0 \ra' = x_0 \la J X, Y \ra' \quad \mbox{ with } J \mbox{ as in } (\ref{ham1})$$ and note that $\la \,, \,\ra_{|_{\vv \times\vv}} = (\, ,\,)$. A derivation $D$ of $\hh_n$ acting trivially on the center must satisfy $[DU, V]= - [U, DV]$ for all $U, V \in \vv$. Equivalently in terms of $\la \,,\, \ra'$, we have that a map $D$ in $\hh_n$ is a derivation acting trivially on the center of $\hh_n$ if and only if the restriction of $D$ to $\vv$ (denoted also $D$) satisfies $$( J DU, V ) = - (JU, DV) \qquad \mbox{ for all } U, V \in \vv,$$ where we replaced $\la \,, \,\ra'_{\vv}$ by $(\, ,\, )$ since they coincide on $\vv \simeq \RR^{2n}$. Denote by $\dd$ the set of derivations on $\hh_n$ acting trivially on the center of $\hh_n$.
\[bij\] There is a bijection between the set of derivations of $\hh_n$ acting trivially on the center and the set $\mathfrak{so}(n)$ of symmetric linear maps on $\RR^{2n}$. This correspondence is given by $D \in \dd \to JD \in \mathfrak{so}(n)$, where $J$ is the complex structure as in (\[ham1\]).
If there exists an n-dimensional abelian subalgebra on $z(JA)_{\dd}$, where $$z(JA)_{\dd}=\{ D \in \dd \mbox{ such that } [D, JA]=0\}$$ then the Hamiltonian function $H$ restriction of the function $f(X)= \frac12(AX,X)$ is completely integrable on the orbits $\mathcal M_{x_{n+1}}$ for $x_{n+1} \neq 0$.
The previous theorem says that the restrictions to the orbit $\mathcal M_{x_{n+1}}$ of the functions $g_i, g_j$ are in involution if their corresponding derivations commute in $\dd$. In particular for $g_i$ and $f$, we have that $H$ and $H_i$ Poisson commute on the orbit if and only if $JA_i$ belongs to the centralizer of $JA$ in $\dd$, $z(JA)_{\dd}$. Since the complete integrability requires of n linearly independent functions, this can be done with a basis of an $n$-dimensional abelian subalgebra of $z(JA)_{\dd}$, finishing the proof.
A linear map $t$ is a derivation of $\hh_n$ acting trivially on the center $\zz(\hh_n)$ if and only if $J t + t^* J=0$, if and only if $t\in \mathfrak{sp}(n)$. The derivations of nilpotent Lie algebras of H-type were computed in ([@Sa]).
In the case of the motion of n-uncoupled harmonic oscillators, we can see that the corresponding derivation is an element of a Cartan subalgebra of $\mathfrak{sp}(n)$.
[GGGG]{}
, [*Foundations of Mechanics*]{}, Second edition. The Benjamin Cummings publishing company, (1985).
, [ A new integrable system and a conjecture by Calogero]{}, , [****]{}, , (1975).
, [ On a trace for formal pseudodifferential operators and the symplectic structure for the KdV type equations]{}, Invent. Math., [**50**]{}, 219-248, (1979).
, [*Mathematical methods of classical mechanics*]{}, Springer Verlag, (1980).
, [*Lectures on symplectic geometry*]{}, Lecture notes in Math. [**1764**]{}, Springer Verlag, (2001).
, [ *Symmetric, invariant, non-degenerate bilinear form on a Lie algebra*]{}, J. Algebra [**105**]{}, 451–464 (1987).
, [*Harmonic analysis in phase space*]{}, Annals of Math. Studies, [**122**]{}, Princeton University Press, Princeton, NJ, (1989).
, [*Euler equations on finite-dimensional Lie groups*]{}, Izv. Akad.Nauk SSSR Ser Mat. [**42**]{}, (1978), 396-415 \[rus\]; english: Math. USSR Izv.[**12**]{}, (1978), 371 –389.
, [ Generalized Liouville method of integration of Hamiltonian systems]{}, Funct. Anal. and its Applic., [**12**]{}, 113 -121, (1978).
, [ Integrable systems in Lie algebras and symmetric spaces]{}, Gordon and Breach Sc. Publ., (1988).
, [*Harmonic Maps, Loop Groups and Integrable Systems*]{}. (London Math. Soc. Student Texts; 38). New York: Cambridge University Press (1997).
, [*Symplectic techniques in physics*]{}. Cambridge New York Port Chester Melbourne Sydney: Cambridge University Press (1991).
, [*Infinite-dimensional Lie algebras*]{}, Cambridge University Press, Cambridge, 1985.
, [*Elements of the theory of representations*]{}, Springer-Verlag, (1976).
, [*Merits and demerits of the orbit method*]{}, bULL. AMS (N.S.), [**36**]{} 4, 433-488, (1999).
, [*Plain mechanics: classical and quantum*]{}, J. Natur. Geom., [**9**]{} 1, 1-14, (1996).
, [*p-mechanics as a physical theory: an introduction*]{}, J. Physics, [**37**]{} 1, 183-204, (2004) (arXiv:quant-ph/0212101).
, [ Quantization and Representation Theory, in: [*Representation Theory of Lie groups, Proc. SRC/LMS Res. Symp., Oxford 1977. London Math. Soc. Lecture Notes Series*]{}]{}, [**34**]{}, 287-316, (1979).
, [ The solution to a generalized Toda lattice and representation theory]{}, Advances in Math., [**39**]{}, 195 - 338, (1979).
, [*Symplectic Geometry and Analytical Mechanics*]{}, D. Reidel Publishing Company, 1987.
, [ Algèbres de Lie et produit scalaire invariant]{}, Ann. scient. Éc. Norm. Sup., 4$^e$ série, [ **t. 18**]{}, 391 - 404,(1985).
, [*Completely integrable hamiltonain systems connected with semisimple Lie algebras*]{}, Inventiones math. [**37**]{}, 93–108 (1976).
, [Estructuras complejas y sistemas hamiltonianos en grupos de Lie solubles]{}, Tesis Doctoral, Fa.M.A.F. Univ. Nac. de Córdoba,( Marzo 2002).
, [*Small oscillations and the Heisenberg Lie algebra*]{}, J. Phys. A: Math. Theor. [**40**]{}, 2407–2424 (2007).
, [*Integrable Systems of Classical Mechanics and Lie Algebras*]{}, vol. I, Birkhäuser Verlag, Basel - Boston - Berlin, (1990).
, [Discrete subgroups of Lie groups]{}, Springer, New York,(1972).
, [*Involution theorems*]{}, Geometric methods in Math. Phys., Lect. Notes in Math., [**775**]{}, Procedings, Lowell, Massachusetts 1979, Springer Verlag, (1980).
, [ The motion of the free n-dimensional rigid body]{}, Indiana Univ. Math. Journal, [**29**]{}, 609 - 629, (1980).
Rend. Sem. Mat. Univ. Pol. Torino, [**54**]{} 2, (1996).
, [Systems of Toda type, inverse spectral problems and representation theory]{}, Invent. Math., [**59**]{}, 13 - 53, (1978).
, [*Lie groups, Lie algebras and their representations*]{}, Springer, (1984).
, [*Fundations of differentiable manifolds and Lie groups*]{}, Springer Verlag, New York (1983).
, [*Geometric quantization*]{}, Oxford Math. Monographs, The Clarendon Press Oxford Univ. Press, New York 1992, Oxford Science Publication.
[^1]: The author GO was partially supported by CONICET, ANPCyT and SECyT-UNC
[^2]: [*(2000) Mathematics Subject Classification*]{}: 53C15, 53C55, 53D05, 22E25, 17B56
|
{
"pile_set_name": "ArXiv"
}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.