math0608302/aa1.tex
1: \documentclass{amsart}
2: \usepackage{math,amsrefs}
3: \newcommand\K{{\mathbb K}}
4: \newcommand\dimK{\dim_\K}
5: \newcommand\support{\operatorname{support}}
6: \newcommand\sphere{{\mathbb S^{n-1}}}
7: 
8: \begin{document}
9: \title{On Amenability of Group Algebras, I}
10: \author{Laurent Bartholdi}
11: \date{typeset \today; last timestamp 20061002}
12: \subjclass[2000]{43A07; 20C07}
13: \begin{abstract}
14:   We study amenability of algebras and modules (based on the notion of
15:   almost-invariant finite-dimensional subspace), and apply it to
16:   algebras associated with finitely generated groups.
17: 
18:   We show that a group $G$ is amenable if and only if its group ring
19:   $\K G$ is amenable for some (and therefore for any) field $\K$.
20: 
21:   Similarly, a $G$-set $X$ is amenable if and only if its span $\K X$
22:   is amenable as a $\K G$-module for some (and therefore for any)
23:   field $\K$.
24: \end{abstract}
25: \maketitle
26: 
27: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
28: \section{Introduction}
29: Amenability of groups was introduced in 1929 by Von
30: Neumann~\cite{vneumann:masses}:
31: \begin{defn}
32:   A (discrete) group $G$ is \emph{amenable} if it admits a measure
33:   $\mu:2^G\to[0,1]$ such that $\mu(G)=1$ and $\mu(A\sqcup
34:   B)=\mu(A)+\mu(B)$ and $\mu(Ag)=\mu(A)$ for all disjoint
35:   $A,B\subseteq G$ and $g\in G$.
36: \end{defn}
37: This notion may serve as a witness to the ``structure'' of groups:
38: either a group is \emph{amenable}, in which case it admits a
39: right-translation invariant finitely additive measure, or it is
40: \emph{non-amenable}, in which case it admits a ``paradoxical''
41: decomposition in finitely many pieces, which can be reassembled by
42: left-translation in two copies of the original group;
43: see~\cite{wagon:banachtarski}.
44: % Although the definition of amenability extends \emph{verbatim} to
45: % semigroups, no such dichotomy is known in that case.\footnote{For an
46: % example of the difference between groups and semigroups, note that
47: % not all finite semigroups are left
48: % amenable~\cite{paterson:amenability}*{Corollary~1.19}!} 
49: More generally:
50: \begin{defn}
51:   Let $G$ be a group acting on the right on a set $X$. This action is
52:   \emph{amenable} if there exists a measure $\mu:2^X\to[0,1]$ such
53:   that $\mu(X)=1$ and $\mu(A\sqcup B)=\mu(A)+\mu(B)$ and
54:   $\mu(Ag)=\mu(A)$ for all disjoint $A,B\subseteq X$ and $g\in G$.
55: \end{defn}
56: Under this definition, a group $G$ is amenable if its action on itself
57: by right-multiplication is amenable. This definition will be
58: reformulated in terms of F\o lner sets (see Lemma~\ref{lem:ag}).
59: 
60: \subsection{Amenable algebras} The present note explores the notion of
61: amenability for associative algebras, which appeared
62: in~\cites{elek:amenaa,gromov:topinv1}. Throughout this note, $\K$
63: denotes an arbitrary field --- although the results easily extend to
64: integral domains. We shall actually phrase it in the more natural
65: language of modules:
66: 
67: \begin{defn}\label{def:aa}
68:   Let $R$ be an associative algebra, and let $M$ be a right
69:   $R$-module. It is \emph{amenable} if, for every $\epsilon>0$ and
70:   every finite-dimensional subspace $S$ of $R$, there exists a
71:   finite-dimensional subspace $F$ of $M$ such that
72:   \[\frac{\dimK((F+Fs)/F)}{\dimK(F)}<\epsilon\text{ for all }s\in S.\]
73: 
74:   The same definition holds, \emph{mutatis mutandis}, for left
75:   modules.
76: \end{defn}
77: 
78: \noindent The main result of this note is the following, proved
79: in~\S\ref{ss:main}:
80: \begin{thm}\label{thm:main}
81:   Let $\K$ be any field, and let $X$ be a right $G$-set. Then $X$ is
82:   amenable if and only if its linear span $\K X$ is amenable.
83: \end{thm}
84: 
85: \noindent Letting $G$ act on itself by right-multiplication, we obtain:
86: \begin{cor}\label{cor:main}
87:   Let $\K$ be any field, and let $G$ be a group. Then $G$ is amenable
88:   if and only if its group algebra $\K G$ is amenable.
89: \end{cor}
90: 
91: The ``only if'' part of the corollary is claimed
92: in~\cite{elek:amenaa}, where the ``if'' part is proven in case
93: $\K=\C$. M.\ Gromov pointed out to me that the ``if'' part admits a
94: simple proof if $\K$ has characteristic $0$.
95: 
96: \subsection{Acknowledgments}
97: The author is grateful to Yves de Cornulier, G\'abor Elek, Anna
98: Erschler, Misha Gromov, Tracy Hall, Fabrice Krieger, Nicolas Monod,
99: and Christophe Weibel for generous feedback and/or entertaining and
100: stimulating discussions.
101: 
102: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
103: \section{Convex sets}
104: We recall the notion of \emph{Steiner point} of a convex
105: polytope~\cite{grunbaum:cp}*{\S14.3}. Let $P$ be a convex polytope in
106: $\R^n$. For $x\in P$ set
107: \[C(x,P)=\{v\in\sphere:\,\langle x'-x|v\rangle\ge0\text{ for
108:   all }x'\in P\};
109: \]
110: this is the set of outer normal vectors of half-spaces containing $P$
111: and with $x$ on their boundary. Let $\measuredangle(x,P)$ denote the
112: normalized content of $C(x,P)$:
113: \[\measuredangle(x,P)=\frac{\lambda(C(x,P))}{\lambda(\sphere)},
114: \quad\text{where $\lambda$ denotes Lebesgue measure.}
115: \]
116: For obvious geometric reasons the number $\measuredangle(x,P)$ is
117: called the \emph{exterior angle} of $P$ at $x$.
118: 
119: Recall that the \emph{Minkowski sum} of two polytopes $P,Q$ is the
120: polytope $P\dot+Q=\{x+y:\,x\in P,y\in Q\}$.
121: \begin{lem}\label{lem:C+}
122:   $C(x+y,P\dot+Q)=C(x,P)\cap C(y,Q)$.
123: \end{lem}
124: \begin{proof}
125:   \begin{align*}
126:     C(x+y,P\dot+Q)&=\{v\in\sphere:\,\langle
127:     x'+y'-(x+y)|v\rangle\ge0\text{ for all }x'\in P,y'\in Q\}\\
128:     &=\{v\in\sphere:\,\langle x'-x|v\rangle\ge0\text{ for all }x'\in P\\
129:     &\hspace{45mm}\text{and }\langle y'-y|v\rangle\ge0\text{ for all }y'\in Q\}\\
130:     &=C(x,P)\cap C(y,Q).\qedhere
131:   \end{align*}
132: \end{proof}
133: 
134: Let $V$ denote the set of extremal points of $P$; then
135: $\measuredangle(x,P)$ is non-zero if and only if $x\in V$.  The
136: \emph{Steiner point} of $P$ is
137: \begin{equation}\label{eq:steiner}
138:   m(P)=\sum_{x\in V}\measuredangle(x,P)x.
139: \end{equation}
140: Up to measure-zero sets, $\{C(x,P):\,x\in V\}$ is a partition of
141: $\sphere$, so $\sum_{x\in P}\measuredangle(x,P)=1$ and thus
142: $m(V)\in V$.
143: 
144: \begin{prop}[\cite{schneider:steiner}]\label{prop:steiner}
145:   The function $m$ is the only continuous $\R^n$-valued function on
146:   convex polytopes in $\R^n$ that satisfies $m(\alpha
147:   A\dot+(1-\alpha)B)=\alpha m(A)+(1-\alpha)m(B)$ for any convex
148:   polytopes $A,B$ and $\alpha\in[0,1]$ and $m(gA)=gm(A)$ for any
149:   similarity $g:\R^n\to\R^n$.
150: \end{prop}
151: 
152: Let $F$ be a subspace of the vector space $\K^n$. For any
153: $S\subseteq\{1,\dots,n\}$, let $\pi_S:\K^n\to\K^S$ denote the
154: projection $(v_1,\dots,v_n)\mapsto(v_i)_{i\in S}$. Define
155: \begin{equation}
156:   X_F = \big\{S:\,\pi_S\text{ restricts to an isomorphism }F\to\K^S\big\}.
157: \end{equation}
158: Let $e_i$ be the $i$th basis vector in $\R^n$, and set
159: \begin{equation}\label{eq:def:PVm}
160:   V_F=\Big\{\sum_{i\in S}e_i:\,S\in X_F\Big\},\quad P_F=\text{the
161:     convex hull of }V_F,\quad m_F=m(P_F).
162: \end{equation}
163: \begin{lem}\label{lem:ccc}
164:   All the $v\in V_F$ are $\{0,1\}$-vectors. The sets $X_F$ and $V_F$
165:   are non-empty, and $P_F$ is a non-empty, closed, convex polytope in
166:   $[0,1]^n$.
167: \end{lem}
168: \begin{proof}
169:   The only non-trivial statements are that $X_F$, and therefore $V_F$
170:   and $P_F$, are non-empty. Let $S$ be maximal such that $\pi_S$
171:   restricts to a surjection $F\to\K^S$. If $\pi_S|F$ were not
172:   injective, there would be $v\neq0$ in $\ker(\pi_S|F)$; let
173:   $k\in\{1,\dots,n\}$ be a non-zero co\"efficient of $v$; then
174:   $k\not\in S$ and $\pi_{S\cup\{k\}}$ is surjective from $F$ onto
175:   $\K^{S\cup\{k\}}$, since its image contains $0\times\K^{\{k\}}$ and
176:   projects onto $\K^S$. This contradicts the maximality of $S$. We
177:   therefore have $S\in X_F$.
178: \end{proof}
179: 
180: \noindent The proof of Theorem~\ref{thm:main} hinges on the following
181: \begin{prop}\label{prop:mono}
182:   Let $E\le F\le\K^n$ be subspaces. Then $m_E\le m_F$
183:   co\"ordinate-wise.
184: \end{prop}
185: 
186: \begin{lem}\label{lem:mono}
187:   Let $E\le F\le\K^n$ be subspaces. Then
188:   \begin{enumerate}
189:   \item\label{lem:mono:1} for every $S\in X_E$ there exists $T\in X_F$
190:     with $S\subseteq T$;
191:   \item\label{lem:mono:2} for every $T\in X_F$ there exists $S\in X_E$
192:     with $S\subseteq T$;
193:   \item\label{lem:mono:3} for every $S\in X_E,T\in X_F$ and $k\in S$
194:     there exists $\ell\in T$ with $S\setminus\{k\}\cup\{\ell\}\in X_E$
195:     and $T\setminus\{\ell\}\cup\{k\}\in X_F$.
196:   \end{enumerate}
197: \end{lem}
198: \begin{proof}
199:   \eqref{lem:mono:1} Consider $D=\ker(\pi_S)\cap F$. By
200:   Lemma~\ref{lem:ccc}, there exists $U\subset\{1,\dots,n\}$ such that
201:   $\pi_U:D\to\K^U$ is an isomorphism. Clearly $U\cap S=\emptyset$, so
202:   $T=S\sqcup U\in X_F$.
203: 
204:   \eqref{lem:mono:2} Apply Lemma~\ref{lem:ccc} to the inclusion
205:   $\pi_T(E)\le\K^T$.
206: 
207:   \eqref{lem:mono:3} Let $(e_i)_{i\in\{1,\dots,n\}}$ be the standard
208:   basis of $\K^n$. Choose a basis $(\epsilon_i)_{i\in S}$ of $E$ such
209:   that $\langle\epsilon_i|e_j\rangle=\delta_{ij}$ for all $i,j\in S$,
210:   and choose a basis $(\phi_i)_{i\in S}$ of $F$ such that
211:   $\langle\phi_i|e_j\rangle=\delta_{ij}$ for all $i,j\in T$.
212: 
213:   Since $E\le F$, we may write $\epsilon_k=\sum_{\ell\in
214:     T}\alpha_\ell\phi_\ell$; and for all $\ell\in T$ we have
215:   $\langle\epsilon_k|e_\ell\rangle=\sum_{\ell'\in
216:     T}\alpha_{\ell'}\langle\phi_{\ell'}|e_\ell\rangle=\alpha_\ell$.
217:   Therefore
218:   \[1=\langle\epsilon_k|e_k\rangle=\sum_{\ell\in
219:     T}\alpha_\ell\langle\phi_\ell|e_k\rangle=\sum_{\ell\in
220:     T}\langle\epsilon_k|e_\ell\rangle\langle\phi_\ell|e_k\rangle;\] so
221:   $\langle\epsilon_k|e_\ell\rangle\langle\phi_\ell|e_k\rangle\neq0$
222:   for some $\ell\in T$. This implies that
223:   $\langle\epsilon_k|e_\ell\rangle\neq0$, so
224:   $\pi_{S\setminus\{k\}\cup\{\ell\}}:E\to\K^{S\setminus\{k\}\cup\{\ell\}}$
225:   is an isomorphism: its image surjects onto $\K^{S\setminus\{k\}}$,
226:   and contains
227:   $0\times\K^{\{\ell\}}=\pi_{S\setminus\{k\}\cup\{\ell\}}(\K\epsilon_k)$.
228:   Since $\pi_{S\setminus\{k\}\cup\{\ell\}}$ maps onto a space of
229:   dimension $\#S$, it is an isomorphism. We also have
230:   $\langle\phi_\ell|e_k\rangle\neq0$, which by the same argument
231:   implies that
232:   $\pi_{T\setminus\{\ell\}\cup\{k\}}:F\to\K^{T\setminus\{\ell\}\cup\{k\}}$
233:   is an isomorphism.
234: \end{proof}
235: 
236: \begin{proof}[Proof of Proposition~\ref{prop:mono}]
237:   For $\varepsilon\in[0,1]$, let
238:   $P_\varepsilon=(1-\varepsilon)P_E\dot+\varepsilon P_F$ be the Minkowski
239:   linear combination of $P_E$ and $P_F$. It is the convex envelope of
240:   $(1-\varepsilon)V_E+\varepsilon V_F$. Set
241:   \[V_\varepsilon=\{(1-\varepsilon)x+\varepsilon y:\,x\in V_E,y\in
242:   V_F,\text{ and }x\le y\text{ co\"ordinatewise}\}.
243:   \]
244:   \begin{lem}\label{lem:mono:aux} $P_\varepsilon$ is the convex envelope
245:     of $V_\varepsilon$.
246:   \end{lem}
247:   \begin{proof}
248:     If $\varepsilon\in\{0,1\}$ this follows from
249:     Lemma~\ref{lem:mono}\eqref{lem:mono:1},\eqref{lem:mono:2}.
250:     Consider then $\varepsilon\in(0,1)$ and $x\in V_E,y\in V_F$ with
251:     $x\not\le y$. By Lemma~\ref{lem:mono}\eqref{lem:mono:3} there
252:     exist $k,\ell$ such that $x':=x-e_k+e_\ell\in V_E$ and
253:     $y':=y-e_\ell+e_k\in V_F$.  Furthermore $k\neq\ell$ because
254:     $x\not\le y$. Now
255:     \[(1-\varepsilon)x+\varepsilon y=
256:     (1-\varepsilon)\big((1-\varepsilon)x+\varepsilon x'\big)+
257:     \varepsilon\big(\varepsilon y+(1-\varepsilon)y'\big)\] is a convex
258:     combination of non-extremal points of $P_E$ and $P_F$, so is not
259:     an extremal point of $P_\varepsilon$.
260:   \end{proof}
261:   Suppose now $\varepsilon\in(0,1)$. Let
262:   $\alpha_\varepsilon:V_\varepsilon\to V_E$ be the map
263:   $(1-\varepsilon)x+\varepsilon y\mapsto x$; it truncates non-$1$
264:   co\"ordinates down to $0$, so $\alpha(z)\le z$ co\"ordinatewise for
265:   all $z\in V_\varepsilon$. By Lemma~\ref{lem:mono}\eqref{lem:mono:1}
266:   this map is onto. By Lemma~\ref{lem:mono:aux} we have
267:   $\lambda(C((1-\varepsilon)x+\varepsilon y,P_\varepsilon)=0$ if
268:   $y\not\ge x$. By Lemma~\ref{lem:C+} we compute
269:   \begin{align*}
270:     \sum_{z\in\alpha_\varepsilon^{-1}(x)}\measuredangle(z,P_\varepsilon)
271:     &=\sum_{z\in\alpha_\varepsilon^{-1}(x)}\frac{\lambda(C(z,P_\varepsilon))}{\lambda(\sphere)}
272:     =\sum_{y\in V_F}\frac{\lambda(C(x,P_E)\cap C(y,P_F))}{\lambda(\sphere)}\\
273:     &=\frac{\lambda(C(x,P_E))}{\lambda(\sphere)}=\measuredangle(x,P_E).
274:   \end{align*}
275:   We conclude
276:   \begin{align*}
277:     m(P_\varepsilon)&=\sum_{z\in V_\varepsilon}\measuredangle(z,P_\varepsilon)z
278:     =\sum_{x\in V_E}\sum_{z\in\alpha_\varepsilon^{-1}(x)}\measuredangle(z,P_\varepsilon)(x+(z-x))\\
279:     &=\sum_{x\in V_E}\measuredangle(x,P_E)x+\sum_{z\in V_\varepsilon}\measuredangle(z,P_\varepsilon)(z-x)\\
280:     &=m(P_E)+\text{something non-negative}\ge m_E\text{ co\"ordinatewise}.
281:   \end{align*}
282:   The conclusion holds for $m(P_1)=m_F$ by continuity of $m$, see
283:   Proposition~\ref{prop:steiner}.
284: \end{proof}
285: 
286: Note that $\#S=\dimK F$ for all $S\in X_F$, and $\|x\|_1=\dimK F$ for
287: all $x\in V_F$, so $\|m_F\|_1=\dimK F$. 
288: \begin{cor}\label{cor:mono}
289:   Let $E\le F\le\K^n$ be subspaces. Then $\|m_F-m_E\|_1=\dimK(F/E)$.
290: \end{cor}
291: \begin{proof}
292:   By Proposition~\ref{prop:mono},
293:   \[\|m_F-m_E\|_1=\|m_F\|_1-\|m_E\|_1=\dimK(F)-\dimK(E)=\dimK(F/E).\qedhere\]
294: \end{proof}
295: 
296: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
297: \section{Proof of Theorem~\ref{thm:main}}\label{ss:main}
298: We recall that there are sundry equivalent definitions of amenability
299: for $G$-sets:
300: \begin{lem}\label{lem:ag}
301:   Let $G$ be a group and let $X$ be a right $G$-set. The following are
302:   equivalent:
303:   \begin{enumerate}
304:   \item\label{lem:ag:0} $X$ is amenable;
305:   \item\label{lem:ag:1} for every $\epsilon>0$ and every finite subset
306:     $S$ of $G$, there exists a finite subset $F$ of $X$ such that
307:     \[\frac{\#(F\cup FS)-\#F}{\#F}<\epsilon;\]
308:   \item\label{lem:ag:1'} for every $\epsilon>0$ and every finite subset
309:     $S$ of $G$, there exists a finite subset $F$ of $X$ such that
310:     \[\frac{\#(F\cup Fs)-\#F}{\#F}<\epsilon\text{ for all }s\in S;\]
311:   \item\label{lem:ag:2} for every $\epsilon>0$ and every finite subset
312:     $S$ of $G$, there exists $f:X\to\R_+$, with finite support, such
313:     that
314:     \[\frac{\|f-fs\|_1}{\|f\|_1}<\epsilon\text{ for all }s\in S.\]
315:   \end{enumerate}
316: \end{lem}
317: The equivalence between~\eqref{lem:ag:1}, \eqref{lem:ag:1'}
318: and~\eqref{lem:ag:2} is classical, see
319: e.g.~\cite{paterson:amenability}*{Theorems~4.4, 4.10, 4.13}. The
320: equivalence of these with~\eqref{lem:ag:0} is proven there in the case
321: $X=G$; see also~\cite{rosenblatt:folner}.
322: 
323: Similarly, there are various equivalent definitions of amenability for
324: modules:
325: \begin{lem}\label{lem:aa}
326:   Let $R$ be an affine algebra and let $M$ be a right module. The
327:   following are equivalent:
328:   \begin{enumerate}
329:   \item\label{lem:aa:2} $M$ is amenable;
330:   \item\label{lem:aa:1} for every $\epsilon>0$ and every
331:     finite-dimensional subspace $S$ of $R$, there exists a
332:     finite-dimensional subspace $F$ of $M$, such that
333:     \[\frac{\dimK((F+FS)/F)}{\dimK(F)}<\epsilon.\]
334:   \end{enumerate}
335: \end{lem}
336: \begin{proof}
337:   Assume first that $M$ is amenable.  Let there be given $\epsilon>0$
338:   and a finite-dimensional subspace $S\le R$. Let $F$ be a
339:   finite-dimensional subspace of $M$ such that
340:   $\dimK(F+Fs)<(1+\epsilon/\dimK S)\dimK F$ for all $s\in S$. Then
341:   $\dimK(F+FS)<(1+\epsilon)\dimK F$, so~\eqref{lem:aa:1} holds. The
342:   converse implication is trivial.
343: \end{proof}
344: 
345: \begin{proof}[Proof of Theorem~\ref{thm:main}]
346:   Suppose first that $X$ is amenable. Let $\epsilon>0$ be given, and
347:   let $S$ be a finite-dimensional subspace of $\K G$. Let $S'$ be the
348:   support of $S$, i.e.\ the union of the supports of all elements of
349:   $S$; it is a finite subset of $G$. By
350:   Lemma~\ref{lem:ag}\eqref{lem:ag:1} there exists a finite subset $F'$
351:   of $X$ with $(\#(F'\cup F'S')-\#F')/\#F'<\epsilon$. Set $F=\K F'$, a
352:   finite-dimensional subspace of $\K X$. We have $\dimK F=\#F'$ and
353:   $\dimK(FS)\le\#F'S'$, so $\dimK(F+FS)\le\#(F'\cup F'S')$, whence
354:   \[\frac{\dimK((F+FS)/F)}{\dimK(F)}<\epsilon,\]
355:   so $\K X$ is amenable by Lemma~\ref{lem:aa}\eqref{lem:aa:1}.
356: 
357:   Suppose now that the $\K G$-module $\K X$ is amenable.  Let
358:   $\epsilon>0$ be given, and let $S$ be a finite subset of $G$.  Set
359:   $S'=\K S$ and, using Lemma~\ref{lem:aa}\eqref{lem:aa:2}, let $F$ be
360:   a finite-dimensional subspace of $\K X$ such that
361:   $\dimK((F+Fs)/F)/\dimK(F)<\frac\epsilon2$ for all $s\in S$.  Set
362:   $f=m_F$ as defined in~\eqref{eq:def:PVm}, page~\pageref{eq:def:PVm}.
363:   We have $\dimK((F+Fs)/F)<\frac\epsilon2\dimK(F)$, so
364:   $\|m_{F+Fs}-m_F\|_1<\frac\epsilon2\dimK(F)$ by
365:   Corollary~\ref{cor:mono}; and similarly
366:   $\|m_{F+Fs}-m_{Fs}\|_1<\frac\epsilon2\dimK(Fs)$. Now $\dimK
367:   F=\dimK(Fs)=\|f\|_1$, so we get
368:   \begin{align*}
369:     \|f-fs\|_1 &= \|m_F-m_{Fs}\|_1 \le \|m_{F+Fs}-m_F\|_1+\|m_{F+Fs}-m_{Fs}\|_1\\
370:     & <\frac\epsilon2\|f\|_1+\frac\epsilon2\|f\|_1=\epsilon\|f\|_1,
371:   \end{align*}
372:   and therefore $X$ is amenable by Lemma~\ref{lem:ag}\eqref{lem:ag:2}.
373: \end{proof}
374: 
375: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
376: \section{Exhaustively amenable sets and modules}
377: The original definition of amenability for algebras was formulated
378: slightly differently~\cite{gromov:topinv1}*{\S1.11}. We show here that
379: it is equivalent to Definition~\ref{def:aa} for group algebras.
380: 
381: \begin{defn}
382:   A right $G$-set $X$ is \emph{exhaustively amenable} if there exists
383:   an increasing net $(F_\lambda)_{\lambda\in\Lambda}$ of finite
384:   subsets of $X$ such that $\bigcup_{\lambda\in\Lambda}F_\lambda=X$
385:   and for all $g\in G$:
386:   \[\lim_{\lambda\in\Lambda}\frac{\#(F_\lambda\cup F_\lambda
387:     g)}{\#F_\lambda}=1.\]
388: \end{defn}
389: \begin{lem}\label{lem:eag}
390:   Let $G$ be a group and let $X$ be a right $G$-set.
391:   \begin{enumerate}
392:   \item\label{lem:eag:1} $X$ is exhaustively amenable if and only if
393:     for every $\epsilon>0$ and all finite sets $S\subseteq G$ and
394:     $U\subseteq X$ there exists a finite subset $F\subseteq X$ such that
395:     \[\frac{\#(F\cup FS)}{\#F}<1+\epsilon.\]
396:   \item\label{lem:eag:2} If $X$ is exhaustively amenable, then it is
397:     amenable.
398:   \item\label{lem:eag:3} If $X$ is amenable and has no finite orbit,
399:     then it is exhaustively amenable.
400:   \end{enumerate}
401: \end{lem}
402: \begin{proof}
403:   \eqref{lem:eag:1} Assume that $X$ is exhaustively amenable,
404:   exhausted by a net $(F_\lambda)_{\lambda\in\Lambda}$.  Let there be
405:   given $\epsilon>0$ and finite subsets $S\subseteq G$, $U\subseteq
406:   M$.  Let $\lambda$ be large enough so that $F_\lambda$ contains $U$
407:   and $\#(F_\lambda\cup F_\lambda s)<(1+\frac\epsilon{\#S})\#F_\lambda$ for
408:   all $s\in S$.  Then $\#(F_\lambda\cup F_\lambda
409:   S)<(1+\epsilon)\#F_\lambda$.
410: 
411:   Assume then the converse. Let $(g_\lambda)_{\lambda\in\Lambda'}$ be
412:   a well-ordering of $G$. Let
413:   $(x_\lambda)_{\lambda\in\Lambda''}$ be a well-ordering of $X$.
414:   Set $\Lambda=\Lambda'\times\Lambda''$ with the product order; set
415:   $g_{(\lambda',\lambda'')}=g_{\lambda'}$ and
416:   $x_{(\lambda',\lambda'')}=x_{\lambda''}$. Let
417:   $\epsilon:\Lambda\to\R_+$ be a decreasing function with
418:   $\lim_{\lambda\in\Lambda}\epsilon_\lambda=0$.  For every
419:   $\lambda\in\Lambda$, let $F_\lambda$ be a finite subset of $X$,
420:   containing $F_\mu$ and $x_\mu$ for all $\mu<\lambda$, and such that
421:   $\#(F_\lambda\cup F_\lambda g_\mu)<(1+\epsilon_\lambda)\#F_\lambda$ for
422:   all $\mu<\lambda$.  This is an exhausting sequence of asymptotically
423:   invariant subspaces, showing that $X$ is exhaustively amenable.
424: 
425:   \eqref{lem:eag:2} follows clearly from \eqref{lem:eag:1}.
426: 
427:   \eqref{lem:eag:3} Let $(g_\lambda)_{\lambda\in\Lambda}$ be a
428:   well-ordering of $G$. Let $\epsilon:\Lambda\to\R_+$ be a decreasing
429:   function with $\lim_{\lambda\in\Lambda}\epsilon_\lambda=0$.  For
430:   every $\lambda\in\Lambda$, let $F_\lambda$ be a finite subset of
431:   $X$, such that $\#(F_\lambda\cup F_\lambda
432:   g_\mu)<\epsilon_\lambda\#F_\lambda$ for all $\mu<\lambda$.
433: 
434:   If $\#F_\lambda$ is unbounded, let $S\subseteq G$ and $U\subseteq X$
435:   be finite subsets, and let $\epsilon>0$ be given. Let $\lambda$ be
436:   large enough so that $\epsilon_\lambda\le\frac\epsilon{2\#S}$ and
437:   $\max\{\mu:\,g_\mu\in S\}<\lambda$ and
438:   $\#F_\lambda\ge\frac2\epsilon\#(U\cup US)$. Set $F=F_\lambda\cup U$;
439:   then
440:   \[\frac{\#(F\cup FS)}{\#F}\le\frac{\#(F_\lambda\cup\F_\lambda
441:     S)+\#(U\cup US)}{\#F_\lambda}<\frac\epsilon2+\frac\epsilon2=\epsilon,\]
442:   so $X$ is exhaustively amenable by~\eqref{lem:eaa:1}.
443: 
444:   Assume therefore that $\#F_\lambda\le m$ for all
445:   $\lambda\in\Lambda$. Then $F_\lambda g_\mu=F_\lambda$ for all
446:   $\lambda>\mu$, as soon as $\epsilon_\mu<\frac1m$. Set
447:   $F_\infty=\bigcup_{\lambda\in\Lambda}F_\lambda$.
448: 
449:   If $F_\infty$ is infinite, let $N:\Lambda\to\N$ be an increasing
450:   function with $\lim_{\lambda\in\Lambda}N_\lambda=\infty$. For all
451:   $\lambda\in\Lambda$, the set $\bigcup_{\mu>\lambda}F_\mu$ is
452:   infinite. Let $\widetilde F_\lambda$ be a union of finitely many
453:   $F_\mu$ with $\mu\ge\lambda$, such that $\#\widetilde F_\lambda\ge
454:   N_\lambda$. Then we still have $\widetilde F_\lambda
455:   g_\mu=\widetilde F_\lambda$ for all $\lambda>\mu$, as soon as
456:   $\epsilon_\mu<\frac1m$. We are back in the case ``$\#F_\lambda$
457:   unbounded''.
458: 
459:   Finally, if $F_\infty$ is finite, then there exists $F\subseteq
460:   F_\infty$ such that for every $\lambda\in\Lambda$, there exists
461:   $\mu\ge\lambda$ with $F_\mu=F$. This $F$ is a finite $G$-orbit.
462: \end{proof}
463: 
464: The following definition generalizes~\cite{gromov:topinv1}*{\S1.11} to
465: uncountable-dimensional algebras and to modules:
466: \begin{defn}
467:   Let $R$ be an algebra, and let $M$ be a right $R$-module.  It is
468:   \emph{exhaustively amenable} if there exists an increasing net
469:   $(F_\lambda)_{\lambda\in\Lambda}$ of finite-dimensional subspaces of
470:   $M$ such that $\bigcup_{\lambda\in\Lambda}F_\lambda=M$ and for all
471:   $r\in R$:
472:   \[\lim_{\lambda\in\Lambda}\frac{\dimK(F_\lambda+F_\lambda r)}{\dimK F_\lambda}=1.\]
473: \end{defn}
474: \begin{lem}\label{lem:eaa}
475:   \begin{enumerate}
476:   \item\label{lem:eaa:1} $M$ is exhaustively amenable if and only if
477:     for every $\epsilon>0$ and all finite-dimensional subspaces $S\le
478:     R$ and $U\le M$ there exists a finite-dimensional subspace $F\le
479:     M$ such that
480:     \[\frac{\dimK(F+FS)}{\dimK F}<1+\epsilon.\]
481:   \item\label{lem:eaa:2} If $M$ is exhaustively amenable, then it is
482:     amenable.
483:   \item\label{lem:eaa:3} If the $\K G$-module $\K X$ is amenable and
484:     $X$ has no finite orbits, then $\K X$ is exhaustively amenable.
485:   \end{enumerate}
486: \end{lem}
487: \begin{proof}
488:   \eqref{lem:eaa:1} Assume that $M$ is exhaustively amenable,
489:   exhausted by a net $(F_\lambda)_{\lambda\in\Lambda}$.  Let there be
490:   given $\epsilon>0$ and finite-dimensional subspaces $S\le R$, $U\le
491:   M$.  Choose a basis $(b_1,\dots,b_d)$ of $S$, and let $\lambda$ be
492:   large enough so that $F_\lambda$ contains $U$ and
493:   $\dimK(F_\lambda+F_\lambda b_i)<(1+\frac\epsilon d)\dimK F_\lambda$ for
494:   all $i\in\{1,\dots,d\}$.  Then $\dimK(F_\lambda+F_\lambda
495:   S)<(1+\epsilon)\dimK F_\lambda$.
496: 
497:   Assume then the converse. Let $(r_\lambda)_{\lambda\in\Lambda'}$ be
498:   a well-ordered basis of $R$. Let $(m_\lambda)_{\lambda\in\Lambda''}$
499:   be a well-ordered basis of $M$. Set
500:   $\Lambda=\Lambda'\times\Lambda''$ with the product order; set
501:   $r_{(\lambda',\lambda'')}=r_{\lambda'}$ and
502:   $m_{(\lambda',\lambda'')}=m_{\lambda''}$. Let
503:   $\epsilon:\Lambda\to\R_+$ be a decreasing function with
504:   $\lim_{\lambda\in\Lambda}\epsilon_\lambda=0$.  For every
505:   $\lambda\in\Lambda$, let $F_\lambda$ be a finite-dimensional
506:   subspace of $M$, containing $F_\mu$ and $m_\mu$ for all
507:   $\mu<\lambda$, and such that $\dimK(F_\lambda+F_\lambda
508:   r_\mu)<(1+\epsilon_\lambda)\dimK(F_\lambda)$ for all $\mu<\lambda$.
509:   This is an exhausting sequence of asymptotically invariant
510:   subspaces, showing that $M$ is exhaustively amenable.
511: 
512:   \eqref{lem:eaa:2} follows clearly from \eqref{lem:eaa:1}.
513: 
514:   \eqref{lem:eaa:3} If $\K X$ is amenable, then $X$ is amenable by
515:   Theorem~\ref{thm:main}; since it has no finite orbits, it is
516:   exhaustively amenable by Lemma~\ref{lem:eag}\eqref{lem:eag:3}. The
517:   first part of the proof of Theorem~\ref{thm:main} extends easily to
518:   show that $\K X$ is exhaustively amenable.
519: \end{proof}
520: 
521: \begin{cor}
522:   The following are equivalent:
523:   \begin{enumerate}
524:   \item $\K G$ is amenable;
525:   \item $\K G$ is exhaustively amenable;
526:   \item $G$ is amenable.
527:   \end{enumerate}
528: \end{cor}
529: \begin{proof}
530:   (1) and (3) are equivalent by Theorem~\ref{thm:main}, and (1)
531:   follows from (2) by Lemma~\ref{lem:eaa}\eqref{lem:eaa:2}. If $G$ is
532:   finite then there is nothing to prove; otherwise $G$, as a right
533:   $G$-set, has a single orbit, which is infinite, so
534:   Lemma~\ref{lem:eaa}\eqref{lem:eaa:3} applies.
535: \end{proof}
536: 
537: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 
538: \section{Isoperimetric profile}
539: There is a quantitative estimate of amenability, called the
540: \emph{isoperimetric profile} (see~\cite{varopoulos-s-c:analysis}*{\S
541:   VI.1}, \cite{gromov:asympt}*{\S5.E}
542: and~\cite{vershik:amenability}*{page 325} for its first appearances):
543: for $G$-sets $X$, this is the function
544: \[I_X(v,S)=\min_{\substack{F\subseteq X\\ \#F\le v}}\frac{\#(F\cup FS)-\#F}{\#F}.
545: \]
546: Then by Lemma~\ref{lem:ag}\eqref{lem:ag:1} amenability of $G$ is
547: equivalent to $\lim_{v\to\infty}I_G(v,S)=0$ for all finite
548: $S\subseteq G$.  Note that, following the equivalence
549: between~\eqref{lem:ag:1} and~\eqref{lem:ag:2} in Lemma~\ref{lem:ag},
550: we have
551: \begin{equation}\label{eq:isog}
552:   I_X(v,S)\sim\inf_{\substack{f\in\ell^1(X)\\
553:     \#\support(f)\le v}}\max_{s\in
554:   S}\frac{\|f-fs\|_1}{\|f\|_1},
555: \end{equation}
556: where `$I(n,S)\sim J(n,S)$' means that, for any $S\subseteq G$, the
557: quotient $I(n,S)/J(n,S)$ is bounded over all $n\in\N$.
558: 
559: If $X$ is amenable, a better normalization of its isoperimetric
560: profile (see~\cite{grigorchuk-z:lamplighter}
561: and~\cite{erschler:isoperimetric}) is
562: \[\Phi_X(n,S)=\min\{v\in\N:\,I_X(v,S)\le 1/n\}.\]
563: For two functions $\Phi,\Psi:\N\to\N$ we write `$\Phi(n)\sim\Psi(n)$'
564: to mean that there exists $K\in\N$ with $\Phi(n)\le\Psi(Kn)$ and
565: $\Psi(n)\le\Phi(Kn)$ for all $n\in\N$. If $X=G$ is a
566: finitely-generated group, then the equivalence class of $\Phi_G(n,S)$
567: is independent of the choice of generating set $S$ of $G$, and is
568: denoted $\Phi_G(n)$.  For example, $\Phi_\Z(n)\sim n$.
569: 
570: The function $\Phi_X(n,S)$ is well-defined if and only if $X$ is
571: amenable. A general result is that $\Phi_X$ is at least as large as
572: the growth function of $X$, see~\cite{varopoulos-s-c:analysis}*{\S
573:   VI.1}.
574: 
575: Similarly, for a right $R$-module $M$ we define
576: \begin{equation}\label{eq:isoa}
577:   I_M(v,S)=\min_{\substack{F\subseteq M\\\dimK(F)\le
578:       v}}\frac{\dimK((F+FS)/F)}{\dimK F}.
579: \end{equation}
580: Then by Lemma~\ref{lem:aa}\eqref{lem:aa:1} amenability of $M$ is
581: equivalent to $\lim_{v\to\infty}I_M(v,S)=0$ for all finite-dimensional
582: $S<R$. We also set
583: \[\Phi_M(n,S)=\min\{v\in\N:\,I_M(v,S)\le 1/n\}.\]
584: 
585: We then remark that the proof of Theorem~\ref{thm:main} shows that
586: $I_{\K X}(n,\K S)\le I_X(n,S)$ and $\Phi_{\K X}(n,\K
587: S)\le\Phi_X(n,S)$.
588: 
589: On the other hand, let $G=(\Z/2\Z)\wr\Z$ be the `lamplighter group',
590: generated for definiteness by $\pm1\in\Z$ and $\delta_0:\Z\to\Z/2$ the
591: Dirac mass at $0$.  Then $\Phi_G(n)\sim 2^nn$: examples of subsets
592: $F\subseteq G$ that achieve the minimum in~\eqref{eq:isog} are of the
593: form
594: \[F=\big\{(f,t)\in G:\,1\le t\le n\text{ and
595: }\support(f)\subseteq\{1,\dots,n\}\big\},
596: \]
597: with $v=2^nn$ elements and $\#(F\cup FS)=\frac{n+2}{n}v$.
598: Nevertheless, $\Phi_{\K G}(n)\sim n$: examples of subspaces $F\le\K G$
599: that achieve the minimum in~\eqref{eq:isoa} are of the form
600: \[F=\bigg\langle\sum_{\substack{f:\Z\to\Z/2\Z\\\support(f)\subseteq\{1,\dots,n\}}}
601: (f,t):\,t\in\{1,\dots,n\}\bigg\rangle,
602: \]
603: of dimension $v=n$ and with $\dimK(F+FS)=n+2$.
604: 
605: \begin{bibsection}
606: \begin{biblist}
607: \bibselect{bartholdi,math}
608: \end{biblist}
609: \end{bibsection}
610: \end{document}
611: