1: % look up papers by J.J. Zhang on dim(W+WS) ~ dim(W)
2:
3: % relate subgroup growth to amenability and algebraic growth
4: % (Aljadeff)
5:
6: % try to do the same for property (T)
7:
8: \documentclass{amsart}
9: \usepackage{graphicx}
10: \usepackage{amsrefs,mathrsfs,math}
11: \newcommand\K{{\mathbb K}}
12: \numberwithin{equation}{section}
13:
14: \begin{document}
15: \title{On Amenability of Group Algebras, II: graded algebras}
16: \author{Laurent Bartholdi}
17: \date{typeset \today; last timestamp 20070902}
18: \address{\'Ecole Polytechnique F\'ed\'erale de Lausanne (EPFL),
19: Institut de Math\'ematiques B (IMB), 1015 Lausanne, Switzerland}
20: \email{laurent.bartholdi@gmail.com}
21: \begin{abstract}
22: We show that, in a finitely generated amenable group $G$ with lower
23: central series $(\gamma_n(G))$, the function
24: $n\mapsto\rank(\gamma_n(G)/\gamma_{n+1}(G))$ grows subexponentially.
25:
26: This paper continues~\cite{bartholdi:aa1}'s study of amenability of
27: affine algebras (based on the notion of almost-invariant
28: finite-dimensional subspace), and applies it to graded algebras
29: associated with finitely generated groups.
30:
31: We consider the graded deformation associated with the filtration of
32: $\Bbbk G$ by powers of its augmentation ideal, and show that it has
33: subexponential growth if $G$ is amenable. This yields the statement
34: in the first paragraph, and answers a question by
35: Vershik~\cite{vershik:amenability}, and another one by de la
36: Harpe~\cite{harpe:uniform}.
37:
38: We also consider the graded deformation of a group ring $\Bbbk G$
39: associated with a metric on $G$, and note that this deformation is
40: amenable whenever $G$ is --- but also if $G$ has ``dead ends'' in
41: its Cayley graph.
42: \end{abstract}
43: \maketitle
44:
45: Warning! This paper contains a mistake in Lemma 4.3, and its main
46: claims should not be considered as proven!
47:
48: \section{Introduction}
49: Throughout this paper, let $\Bbbk$ denote a commutative ring, and for
50: a $\Bbbk$-module $M$ let $\rank(M)$ denote its minimal number of
51: generators. Recall from~\cite{bartholdi:aa1} the notion of amenable
52: algebra, which first appeared (in a slightly different form)
53: in~\cites{elek:amenaa,gromov:topinv1} and in~\cite{vaillant:folner} in
54: the context of $C^*$-algebras:
55:
56: \begin{defn}\label{def:aa}
57: Let $R$ be a $\Bbbk$-algebra and let $M$ be a right $R$-module. It
58: is \emph{amenable} if, for every $\epsilon>0$ and every finite-rank
59: subspace $S$ of $R$, there exists a finite-rank subspace $F$ of $M$
60: such that
61: \[\frac{\rank(F+FS)-\rank(F)}{\rank(F)}<\epsilon;\]
62: any such $F$ is called \emph{$(S,\epsilon)$-invariant}.
63:
64: $M$ is \emph{exhaustively amenable} if furthermore in the definition
65: above the space $F$ may be required to contain any specified
66: finite-rank subspace.
67: \end{defn}
68:
69: This is a counterpart to amenability of $G$-sets for a group $G$,
70: where according to F\o lner's definition~\cite{folner:banach} the
71: $G$-set $X$ is \emph{amenable} if, for every $\epsilon>0$ and every
72: finite subset $S\subset G$, there exists a finite subset $F\subset X$
73: such that $(\#(F\cup FS)-\#F)/\#F<\epsilon$; such $F$ being also
74: called $(S,\epsilon)$-invariant. The main result
75: of~\cite{bartholdi:aa1} was:
76: \begin{thm}\label{thm:main}
77: Let $\K$ be a field, and let $X$ be a $G$-set. Then $X$ is amenable
78: if and only if its linear envelope $\K X$ is amenable.
79: \end{thm}
80:
81: %In this text, all groups shall be finitely generated, and all algebras
82: %shall be \emph{affine}, i.e.\ finitely generated associative over
83: %$\Bbbk$. If $R$ is affine, then any given finite-dimension subspace
84: %$S$ generating $R$ may be fixed in Definition~\ref{def:aa}.
85:
86: \subsection{Growth of modules}\label{ss:growth}
87: Let $R$ be an \emph{affine} algebra, i.e.\ a finitely generated
88: associative algebra over a commutative ring $\Bbbk$, generated by the
89: finite-rank subspace $S$. Let $M$ be a finitely-generated $R$-module,
90: generated (as an $R$-module) by a finite-rank subspace $T$. Let us
91: write here and below
92: \[TS^{\le n}=T+TS+TS^2+\dots+TS^n.\] The module $M$ is then filtered
93: by the exhausting sequence of subspaces $T\le TS^{\le1}\le
94: TS^{\le2}\le\cdots\le TS^{\le n}\le\cdots$, and the \emph{growth} of
95: $M$ is the sequence $(r_n)_{n\ge1}$ defined by $r_n=\rank(TS^{\le n})$. We
96: say that $M$ has \emph{subexponential growth} if
97: $\lim\sqrt[n]{r_n}=1$. This property does not depend on the choices of
98: $S$ and $T$.
99: \begin{prop}\label{prop:subam}
100: If $M$ has subexponential growth, then $M$ is exhaustively amenable.
101: \end{prop}
102: \begin{proof}
103: Given $U\le M$ of finite rank and $\epsilon>0$, let $d\in\N$ be such
104: that $U\le TS^{\le d}$. Since $(r_n)$ grows subexponentially,
105: $\liminf r_{n+1}/r_n=1$, so there exists $n>d$ such that
106: $r_{n+1}/r_n<1+\epsilon$. Set $F=TS^{\le n}$. Then $F+FS\le TS^{\le
107: n+1}$, so $(\rank(FS)-\rank(F))/\rank(F)<\epsilon$ and $F$
108: contains $U$.
109: \end{proof}
110:
111: We consider in~\S\ref{ss:crystal} the graded module $M_0=\bigoplus_n
112: TS^{\le n}/TS^{\le n-1}$ associated with the ascending filtration of
113: $M$ by its finite-rank subspaces $TS^n$. If we consider $R=\Bbbk G$ a
114: group ring, $S$ the $\Bbbk$-span of a finite generating set of $G$,
115: and $M=R$ with $T=\{1\}$, then $M_0=R_0$ is an algebra with basis in
116: bijection with $G$, and with product derived from multiplication in
117: $G$ except that $g\cdot h=0$ if the length of $gh$ is strictly less
118: than the sum of the lengths of $g$ and $h$. We prove:
119: \begin{thm}\label{thm:dead}
120: Let $G=\langle S\rangle$ be a finitely generated group. If $G$ is
121: amenable, or has dead ends, then $(\Bbbk G)_0$ is amenable. If $G$
122: is amenable, or has infinitely many dead ends, then $(\Bbbk G)_0$ is
123: exhaustively amenable.
124: \end{thm}
125:
126: The proof of Theorem~\ref{thm:dead} relies on fact that $(\Bbbk
127: G)_0$ has a monomial basis, namely a basis $B$ with $B\cdot B\subseteq
128: B\cup\{0\}$, this prompts the
129: \begin{question}
130: If $M$ is an amenable finitely-generated $R$-module, is then $M_0$
131: also amenable?
132: \end{question}
133:
134: \subsection{Augmented algebras}\label{ss:augmented}
135: Assume now that $R$ has an augmentation\footnote{Namely, a morphism of
136: unital algebras to the scalars $\Bbbk$.}$\varepsilon:R\to\Bbbk$,
137: with kernel $\varpi$; as a typical example $R=\Bbbk G$ and
138: $\varpi=\langle g-1:\,g\in G\rangle$. Let $M$ be an $R$-module. Then
139: $M$ admits a descending filtration $M\ge M\varpi\ge
140: M\varpi^2\ge\cdots\ge M\varpi^n\ge\cdots$; and an associated graded
141: module \[\overline M=\bigoplus_{n\ge 0}M\varpi^n/M\varpi^{n+1}.\] In
142: particular, $R$ admits an associated graded algebra $\overline
143: R=\bigoplus_{n\ge0}\varpi^n/\varpi^{n+1}$. If $M$ is generated (as an
144: $R$-module) by a subspace $T$, then $\overline M$ is a graded
145: $\overline R$-module, generated by the image $\overline T$ of $T$ in
146: $M/\varpi M$.
147:
148: If $R$ is affine, say generated by a finite-rank subspace $S$, then
149: $\overline R$ is also affine: the projection $\overline
150: S=\{s-\varepsilon(s)+\varpi^2:\,s\in S\}$ of $S$ in $\varpi/\varpi^2$
151: actually equals $\varpi/\varpi^2$ irrespective of $S$, and generates
152: $\overline R$. The \emph{growth} of $\overline M$ is then its growth
153: in the sense of~\S\ref{ss:growth}, with respect to the generating
154: subspace $\overline S$ of $\overline R$ and $\overline T$ of
155: $\overline M$.
156:
157: The main result of this paper is the following theorem, proven
158: in~\S\ref{ss:vershik}:
159: \begin{thm}\label{thm:vershik}
160: If $G$ is an amenable, finitely generated group, then
161: $\overline{\Bbbk G}$ has subexponential growth (and therefore is
162: exhaustively amenable by Proposition~\ref{prop:subam}).
163: \end{thm}
164:
165: Vershik conjectured in~\cite{vershik:amenability}*{page 326} that if
166: $G$ is an amenable, finitely generated group and if $\varpi$ denote
167: the augmentation ideal in $\Z G$, then $\rank(\varpi^n/\varpi^{n+1})$
168: grows subexponentially. This follows from Theorem~\ref{thm:vershik}
169: with $\Bbbk=\Z$.
170:
171: Theorem~\ref{thm:vershik} actually follows from the following
172: statement: if $R$ is an amenable augmented affine algebra with a basis
173: consisting of invertible elements, then $\overline R$ has
174: subexponential growth. This prompts the
175: \begin{question}
176: Let $R$ be an amenable augmented affine algebra. Does $\overline R$
177: necessarily have subexponential growth?
178:
179: Let more generally $M$ be an amenable $R$-module, where $R$ is any
180: augmented affine algebra. Does $\overline M$ necessarily have
181: subexponential growth?
182: \end{question}
183:
184: %Note that Theorem~\ref{thm:vershik} does not extend directly to
185: %modules; indeed consider $G=F_2=\langle x,y\rangle$ a free group, and
186: %$H=\langle y^{x^n}:n<0\rangle$. Then $X=H\backslash G$ is an amenable
187: %$G$-set (with F\o lner sets $\{x^n:\,0\le n<N\}$); however,
188: %$\Lambda=\overline{\Bbbk G}$ is the free unital algebra generated by
189: %$X=x-1$ and $Y=y-1$, and $\overline{\Bbbk
190: % X}=\Lambda/\{X^nY:\,n>0\}\Lambda$ has basis
191: %$\{X^n:\,n\ge0\}\cup\{Yw:\,w\in\{X,Y\}^*\}$, so in particular has
192: %exponential growth. Note that nevertheless it is an amenable
193: %$\Lambda$-module.
194:
195: \subsection{Golod-Shafarevich groups}
196: Let $\K$ be a field of characteristic $p>0$, and assume that $G$ is a
197: residually-$p$ group. This means that the series $(G_{n,p})_{n\ge1}$
198: of \emph{dimension} subgroups defined by $G_{1,p}=G$ and
199: $G_{n+1,p}=[G_{n,p},G](G_{\lceil n/p\rceil,p})^p$ for $n\ge1$,
200: satisfies $\bigcap G_{n,p}=\{1\}$. The \emph{degree} of $g\in G$,
201: written $\deg_p(g)$, is the maximal $n\in\N\cup\{\infty\}$ such that
202: $g\in G_{n,p}$.
203:
204: It is well known~\cites{jennings:gpring,lazard:uea} that $\mathcal
205: L:=\bigoplus_{n\ge1}(G_{n,p}/G_{n+1,p})\otimes_{\F}\K$ has the structure of
206: a restricted Lie algebra over $\K$; and that $\overline{\K G}$ is the
207: universal restricted enveloping algebra of $\mathcal L$. We will use
208: this fact to relate Theorem~\ref{thm:vershik} to purely
209: group-theoretical statements, e.g.\ Corollary~\ref{cor:lcs}.
210:
211: Golod constructed in~\cite{golod:nil} for every prime $p$ a finitely
212: generated infinite torsion $p$-group. His method is quite flexible,
213: and was generalized as follows:
214: \begin{defn}
215: A group $G$ is a \emph{Golod-Shafarevich group} for the prime $p$ if
216: it admits a presentation $G=F/\langle \mathscr R\rangle^F$ in which
217: $F$ is a free group of rank $d$ and $\mathscr R\subset F$ is a set
218: of relators, such that for some $t\in(0,1)$ we have
219: \[1-dt+\sum_{r\in\mathscr R}t^{\deg_pr}<0,\]
220: where $\deg_p$ denotes degree with respect to the filtration
221: $(F_{n,p})_{n\ge1}$ of $F$.
222: \end{defn}
223: \begin{prop}[\cite{herstein:rings}; \cite{koch:galois};
224: \cite{huppert-b:fg2}*{\S VIII.12}; \cite{bartholdi-g:lie}*{}]
225: All Golod-Shafarevich groups are infinite. If $\varpi$ denote the
226: augmentation ideal in $\F G$, then $\dim_{\F}(\varpi^n/\varpi^{n+1})$
227: grows exponentially (at rate at least $1/t$).
228: \end{prop}
229: On the other hand, there are torsion groups that are Golod-Shafarevich
230: (this solved Burnside's problem~\cite{burnside:question}).
231:
232: The second part of the following result
233: answers~\cite{harpe:uniform}*{Open Problem 5.2}
234: and~\cite{harpe:mfap}*{Question 7}:
235: \begin{cor}
236: Golod-Shafarevich groups are not amenable. In particular, there
237: exist non-amenable residually-$p$ torsion groups.
238: \end{cor}
239: Ershov~\cite{ershov:goshat} has constructed Golod-Shafarevich groups
240: that have property (T) --- and no infinite (T) group can be amenable.
241: Any Golod-Shafarevich group admits a Golod-Shafarevich torsion
242: quotient, which will still have property (T). This
243: answers~\cite{harpe:uniform}*{Open Problem 5.2} by a different method.
244:
245: \subsection{Lower central series}
246: Theorem~\ref{thm:vershik} has the following purely group-theoretic
247: consequence, also proven in~\S\ref{ss:vershik}:
248: \begin{cor}\label{cor:lcs}
249: Let $G$ be an amenable, finitely-generated group, and let
250: $(\gamma_n(G))_{n\ge1}$ denote its lower central series. Then the
251: function $n\mapsto\rank(\gamma_n(G)/\gamma_{n+1}(G))$ grows
252: subexponentially.
253: \end{cor}
254: Note that this function may grow arbitrarily close to an exponential
255: function. Indeed Petrogradsky showed
256: in~\cite{petrogradsky:polynilpotent} that if $G$ be the free
257: $k$-generated soluble group of solubility class $q\ge3$, then
258: \[\rank(\gamma_n(G)/\gamma_{n+1}(G))\cong\exp\left(\bigg(\frac{(k-1)\zeta(k)}{\log\log\cdots n}\bigg)^{1/k}n\right),\]
259: with $q-3$ iterated logarithms in the expression above.
260:
261: The converse of Theorem~\ref{thm:vershik} does not hold (see
262: Remark~\ref{rem:vershik}). However, the next statement trivially
263: follows from Theorem~\ref{thm:vershik}, and raises the question after
264: it:
265: \begin{cor}
266: Let $G$ be an amenable group. Then for every finitely-generated
267: subgroup $H$ of $G$ the function
268: $n\mapsto\rank(\gamma_n(H)/\gamma_{n+1}(H))$ grows subexponentially.
269: \end{cor}
270: \begin{question}
271: Does there exist a non-amenable
272: residually-nilpotent\footnote{Namely, such that
273: $\bigcap\gamma_n(G)=1$.} group $G$ such that for every
274: finitely-generated subgroup $H$ of $G$ the function
275: $n\mapsto\rank(\gamma_n(H)/\gamma_{n+1}(H))$ grows subexponentially?
276: \end{question}
277:
278: \subsection{Acknowledgments}
279: The author is grateful to Mikhail Ershov, Fran\c cois Gu\'eritaud,
280: Pierre de la Harpe and Fabrice Krieger for generous feedback and/or
281: entertaining and stimulating discussions.
282:
283: \section{Hecke and crystal algebras}\label{ss:crystal}
284: We prove Theorem~\ref{thm:dead} in this section, phrasing it in a
285: slightly more general manner. Let $G$ be a group with fixed
286: generating set $S$. Denote by $\ell(g)$ the length of $g\in G$ in the
287: word metric:
288: \[\ell(g)=\min\{n:\, g=s_1\dots s_n,\,s_i\in S\}.\]
289: Choose $\lambda\in\Bbbk$, and define the ``Hecke
290: algebra''\footnote{The terminology comes from the classical Hecke
291: algebra associated with the symmetric group.} $(\Bbbk G)_\lambda$
292: as follows: it is isomorphic to $\Bbbk G$ as a $\Bbbk$-module; it has
293: a basis $\{\delta_g\}_{g\in G}$; and multiplication is given by
294: \[\delta_g\delta_h=\lambda^{\ell(g)+\ell(h)-\ell(gh)}\delta_{gh}.\]
295: Note that, although the notation does not make it explicit, $(\Bbbk
296: G)_\lambda$ depends on the choice of $S$.
297:
298: If $\lambda$ is invertible, then $(\Bbbk G)_\lambda$ is isomorphic to
299: $\Bbbk G$, the isomorphism being given by $\delta_g\mapsto
300: \lambda^{\ell(g)}g$. The universal cases $\Bbbk=\Z[\lambda]$ and
301: $\Q[[\lambda]]$ should deserve particular consideration.
302:
303: Quite on the contrary, $(\Bbbk G)_0$ is a graded algebra (with degree
304: function $\ell$), which we call the \emph{crystal}\footnote{The
305: terminology comes from statistical mechanics, where the parameter
306: $\lambda$ of the deformation of $\Bbbk G$ is interpreted as
307: temperature.} of $\Bbbk G$. Note that $(\Bbbk G)_0$ is the
308: associated graded algebra $\bigoplus\Bbbk S^{\le n}/\Bbbk S^{\le n-1}$
309: of the filtered algebra considered in \S\ref{ss:growth}. The first
310: part of Theorem~\ref{thm:dead} generalizes as:
311: \begin{prop}\label{prop:crystal}
312: Let $G$ be an amenable group with fixed generating set. Then $(\Bbbk
313: G)_\lambda$ is amenable for all $\lambda\in\Bbbk$.
314: \end{prop}
315: \begin{proof}
316: Let $\epsilon>0$ be given, and let\footnote{In this proof, we use
317: $S$ for a subset of a group, and $S'$ for a subspace of an
318: algebra.} $S'$ be a finite-rank subspace of $(\Bbbk G)_\lambda$.
319: Let $S$ denote the support of $S'$, i.e.\ the set of those $g\in G$
320: such that $\delta_g$ has a non-zero co\"efficient in some element of
321: $S'$; it is a finite subset of $G$. Since $G$ is amenable, there
322: exists a finite subset $F$ of $G$ with
323: \begin{equation}\label{eq:crystal:1}
324: (\#(F\cup FS)-\#F)/\#F<\epsilon.
325: \end{equation}
326: Set $F'=\bigoplus_{g\in F}\Bbbk\delta_g$, a finite-rank subspace of
327: $(\Bbbk G)_\lambda$. We have $\rank F'=\#F$ and $F'S'\le\Bbbk(FS)$,
328: so $\rank(F'S')\le\#FS$, and $\rank(F'+F'S')\le\#(F\cup FS)$, whence
329: \begin{equation}\label{eq:crystal:2}
330: \frac{\rank(F'+F'S')-\rank(F')}{\rank(F')}<\epsilon,
331: \end{equation}
332: so $(\Bbbk G)_\lambda$ is amenable.
333: \end{proof}
334:
335: It is however possible for $(\Bbbk G)_0$ to be amenable, yet for $G$
336: not to be amenable. The example in Proposition~\ref{prop:bogo} appears
337: in~\cite{bogopolski:bilipschitz} (with a small typographical mistake).
338:
339: \subsection{Dead ends} Say $g\in G$ is a \emph{dead end} if
340: $\ell(gs)\le\ell(g)$ for all $s\in S$. Note that, although the
341: notation does not make it apparent, this property strongly depends on
342: the choice of $S$. Zoran \v Suni\'k has informed me that every group
343: admits a generating set for which the group contains a dead end. The
344: second part of Theorem~\ref{thm:dead} reads:
345:
346: \begin{prop}
347: If $G$ has a dead end (with respect to $S$), then $(\Bbbk G)_0$ is
348: amenable.
349:
350: If $G$ has infinitely many dead ends, then $(\Bbbk G)_0$ is
351: exhaustively amenable.
352: \end{prop}
353: \begin{proof}
354: Let first $g\in G$ be a dead end. Set $F=\Bbbk\delta_g$. Then
355: $F+Fs=F$ for any $s\in(\Bbbk G)_0$, so $(\Bbbk G)_0$ is amenable.
356:
357: Let now $g_1,g_2,\dots$ be an infinite set of dead ends in $G$.
358: Given $\epsilon>0$ and $E\le(\Bbbk G)_0$ of rank $n$, consider
359: $F=E+\Bbbk\delta_{g_1}+\dots+\Bbbk\delta_{g_{\lceil
360: n/\epsilon\rceil}}$. Then $F$ contains $E$ and
361: $\rank(F+Fs)\le\rank F+\rank E$ and $\rank F\ge\rank E/\epsilon$.
362: \end{proof}
363:
364: \begin{prop}[Bogopolski]\label{prop:bogo}
365: For all $k\ge 3$, the ``triangle'' group
366: \[G_k=T_{3,3,k}=\langle x,y\mid x^3,y^3,(xy)^k\rangle\] contains
367: infinitely many dead ends.
368: \end{prop}
369: \begin{proof}
370: This group is hyperbolic; it acts by isometries on hyperbolic space
371: and preserves the semiregular tiling of $\mathbb H^2$ by triangles
372: (with edges labeled $x$ and $y$) and $2k$-gons (with edges labeled
373: periodically $x,y$). The $1$-skeleton of this tiling is the Cayley
374: graph of $G_k$.
375:
376: Assume first that $k$ is even. Consider for all
377: $n\in\Z\setminus\{0\}$ the element $d_n\in G_k$ defined by
378: \[d_{2n}=((xy)^{k/2}(yx)^{k/2})^n,\qquad
379: d_{2n+1}=((xy)^{k/2}(yx)^{k/2})^n(xy)^{k/2}.\]
380:
381: Consider furthermore the automorphism $\phi$ of $G_k$ defined by
382: $x\mapsto y^{-1}$ and $y\mapsto x^{-1}$. It is easy to check that
383: $d_n$ is fixed by $\phi$.
384:
385: Geometrically, $d_n$ is the hyperbolic isometry whose axis cuts
386: through a doubly-infinite sequence of vertex-abutting $2k$-gons in
387: their middle, and translates by $n$ of them; $\phi$ is the
388: reflection through this axis.
389:
390: Let $s_1\dots s_m$ be a word of minimal length representing $d_n$,
391: with $s_i\in S=\{x,y,x^{-1},y^{-1}\}$; so $\ell(d_n)=m$. Then
392: $\ell(d_ns)\le m$ for all $s\in S$:
393: \begin{itemize}
394: \item if $s=s_m^{-1}$ this is clear;
395: \item if $s=s_m$ then $d_ns=s_1\dots s_{m-1}s_m^{-1}$ whence
396: $\ell(d_ns)\le m$;
397: \item otherwise, $d_n^\phi=d_n=s_1^\phi\dots s_m^\phi$ also of
398: minimal length, and $s_m^\phi\in\{s,s^{-1}\}$, so the previous
399: cases apply.
400: \end{itemize}
401:
402: If $k$ is odd, consider for all $n\ne0$ the element
403: $d_n=((xy)^{(k-1)/2}x)^n$, with a similar geometric interpretation
404: as above. The same arguments apply.
405: \end{proof}
406: It is in fact not hard to see that these are the only dead ends in
407: $G_k$. Since for $k\ge4$ the $G_k$ are non-elementary hyperbolic
408: groups, they are certainly not amenable.
409:
410: There unfortunately does not seem to be any natural condition to
411: impose on $(\Bbbk G)_0$ to ensure that $G$ be amenable.
412:
413: \section{Tileable amenable groups}\label{ss:tag}
414: We prove in this section a result by Weiss~\cite{weiss:monotileable},
415: based on earlier work by Ornstein and
416: Weiss~\cite{ornstein-weiss:entropyiso}*{\S I.2}. We follow the sketch
417: of a proof by Gromov~\cite{gromov:topinv1}*{pages~336--337}, adapting
418: it so as to prepare the ground for a generalization to modules
419: in~\S\ref{ss:tae}. We will not use the results in this \S, but present
420: them as a warm-up for \S\ref{ss:tae}.
421:
422: Let $X$ denote a $G$-set. We defined amenability with respect to the
423: \emph{outer envelope} $AK$ of a subset $A\subset X$ with respect to
424: $K\subset G$. It will be useful in this section to consider, again for
425: $A\subset X$ and $K\subset G$, the \emph{inverse envelope}
426: \[AK^* := \{x\in X:\,xK\cap A\neq\emptyset\}.\]
427: The easy properties
428: \[A(K\cup L)^*=AK^*\cup AL^*,\qquad (A\cup B)K^*=AK^*\cup BK^*\]
429: follow immediately from $AK^*=A\{k^{-1}:\,k\in K\}$.
430:
431: \begin{thm}[Weiss]\label{thm:weiss}
432: Let $G$ be an amenable group, let $K\subseteq G$ be a finite subset,
433: let $\epsilon>0$ be given, and let $N_0\ge N_1\ge\dots$ be a nested
434: sequence of finite-index normal subgroups of $G$ such that
435: $\bigcap_{n\in\N} N_n=\{1\}$.
436:
437: Then for all $n\gg0$ there exists a $(K,\epsilon)$-invariant subset
438: $T_n\subseteq G$ that is a transversal for $N_n$ in $G$.
439: \end{thm}
440:
441: Let $\zeta\ge1$ and $\delta\in(0,1)$ be constants to be fixed later.
442: In this section and the next we shall use, for $\mu\ge\delta$, the
443: transformation $\Theta_\mu:\R_+^2\to\R_+^2$ given by
444: \[\Theta_\mu(\nu,\alpha)=\Big(\nu+\mu(1-\alpha),\alpha+\frac{\mu(1-\alpha)}{1-\delta}\zeta\Big).\]
445: The proof of Theorem~\ref{thm:weiss} relies on the following
446: \begin{lem}\label{lem:weiss}
447: Let $\Omega$ be a finite $G$-set; let $B$ be a subset of $\Omega$
448: and let $K,L$ be finite subsets of $G$; assume that for all
449: $x\in\Omega$ the orbit map
450: \begin{equation}\label{eq:weiss:i}
451: K\to\Omega,\; k\mapsto xk\qquad\text{is injective}.
452: \end{equation}
453: Let $\alpha,\nu\in[0,1)$ and $\zeta\ge1$ be such that
454: \begin{xalignat*}{2}
455: \#(KL^*)&\le\zeta\#K, & \#(BK^*)&\le\alpha\#\Omega,\\
456: \#B&=\nu\#\Omega, & \#(BL^*)&\le\alpha\#\Omega;
457: \end{xalignat*}
458: Let furthermore $\delta\in(0,1)$ be given.
459:
460: Then there exist $s\ge 1$, elements $x_1,\dots,x_s\in\Omega$, and
461: $\mu\ge\delta$ such that, setting $B_0=B$, $B_i=B_{i-1}\cup x_iK$
462: for $1\le i\le s$ and $\Theta_\mu(\nu,\alpha)=(\nu',\alpha')$, we
463: have
464: \begin{gather}
465: \label{eq:weiss:1}
466: \#(x_iK\cap B_{i-1}) \le\delta\#K\text{ for all }i\in\{1,\dots,s\};\\
467: \label{eq:weiss:2} \#B_s =\nu'\#\Omega;\\
468: \label{eq:weiss:3} \#(B_sL^*) \le\alpha'\#\Omega.
469: \end{gather}
470: \end{lem}
471: Informally, the lemma says that, given a set $K$ with small
472: $L$-boundary, and a set $B\subset\Omega$ with small $K$- and
473: $L$-boundary, one can construct a quantifiably larger set $B_s$ with
474: small $L$-boundary.
475:
476: \begin{proof}
477: Let $x_1,\dots,x_s\in\Omega$ be a maximal-length sequence of
478: elements such that~\eqref{eq:weiss:1} holds. Then, by maximality of
479: $(x_1,\dots,x_s)$, for all $x\in\Omega$ we have
480: \begin{equation}\label{eq:weiss:p1}
481: \#(B_s\cap xK)=\#\{(b,k):b\in B_s,k\in K,xk=b\}>\delta\#K.
482: \end{equation}
483: We deduce
484: \begin{align*}
485: \#B_s\#K&=\#\{(b,k,x):\,b\in B_s,k\in K,x\in\Omega,xk=b\}\\
486: &=\sum_{x\in\Omega}\#\{(b,k):b\in B_s,k\in K,xk=b\}
487: =\sum_{x\in\Omega}\#(B_s\cap xK)\\
488: &=\sum_{x\in BK^*}\#\{(b,k):b\in B_s,k\in K,xk=b\}+\sum_{x\in
489: \Omega\setminus BK^*}\#(B_s\cap xK)\\
490: &=\sum_{b\in B_s}\sum_{k\in K}\#\{BK^*\cap
491: \{bk^{-1}\}\}+\sum_{x\in\Omega\setminus BK^*}\#(B_s\cap xK)\\
492: &\ge\sum_{b\in B}\sum_{k\in K}\#\{BK^*\cap
493: \{bk^{-1}\}\}+\sum_{x\in\Omega\setminus BK^*}\delta\#K\text{ by }\eqref{eq:weiss:p1}\\
494: &\ge\#B\#K+(1-\alpha)\#\Omega\delta\#K.
495: \end{align*}
496: If we divide by $\#K\#\Omega$ and set
497: $\mu=(\#B_s/\#\Omega-\nu)/(1-\alpha)$, then we get $\mu\ge\delta$,
498: and therefore $s\ge1$. Note that~\eqref{eq:weiss:2} holds by the
499: choice of $\mu$. Next,
500: \[\#B_s\ge\#B+\sum_{i=1}^s\#(x_iK\setminus
501: B_{i-1})\ge\#B+s(1-\delta)\#K\text{ by }(\ref{eq:weiss:i},\ref{eq:weiss:1})
502: \]
503: so $s\#K\le(\nu'-\nu)\#\Omega/(1-\delta)$; and
504: \[\#(B_sL^*)\le\#(BL^*)+\sum_{i=1}^s\#(x_iKL^*)\le\alpha\#\Omega+s\zeta\#K,
505: \]
506: from which~\eqref{eq:weiss:3} follows.
507: \end{proof}
508:
509: \begin{rem}\label{rem:tiling}
510: We shall apply Lemma~\ref{lem:weiss} using the following strategy:
511: we construct a finite but very long sequence $K_1,\dots,K_t$ of F\o
512: lner sets, each with a very small boundary with respect to its
513: predecessors and to $K$. We then find a finite quotient $\Omega$ of
514: $G$ in which these F\o lner sets embed. Then, starting from $K_t$
515: down to $K_1$, we apply $t$ times Lemma~\ref{lem:weiss} to cover
516: most of $\Omega$ by images of translates of the $K_i$, taking each
517: time as many copies as possible subject to them having extremely
518: small overlaps (see condition~\eqref{eq:weiss:1}). There remains a
519: small part of $\Omega$ that is not covered. We then lift these
520: images back to $G$, and lift the small remainder arbitrarily. We
521: have obtained a transversal, consisting mostly of pieces carved out
522: of F\o lner sets by other F\o lner sets. See Figure~\ref{fig:1} for
523: an illustration.
524: \begin{figure}
525: \[\includegraphics[height=15cm]{weiss.pdf}\]
526: \caption{The proof of Theorem~\ref{thm:weiss}, illustrated in a picture}
527: \label{fig:1}
528: \end{figure}
529:
530: Note that both co\"ordinates of $\Theta_\mu$ increase monotonically
531: in $\mu$. This property helps us formalize the intuitive notion that
532: ``it cannot hurt us, in constructing the covering of $\Omega$, if
533: the biggest F\o lner sets leave unexpectedly small holes between
534: them, even though small holes are harder to cover efficiently with
535: smaller F\o lner sets''.\hfill $\triangle$
536: \end{rem}
537:
538: \begin{proof}[Proof of Theorem~\ref{thm:weiss}]
539: We are given $K$ and $\epsilon$. Without loss of generality, we
540: suppose $1\in K$. Fix in order the following data:
541: \begin{enumerate}
542: \item Choose $\delta>0$, the ``allowed overlap'' as
543: in~\eqref{eq:weiss:1}, such that $\delta\#K<\frac\epsilon2$ and
544: $(1+\frac\epsilon2)(1-\delta)>1$;
545: \item Choose $\zeta>1$ such that
546: $(1-\delta)/\zeta>1-\epsilon/(2\#K)$, the ``relative F\o lner
547: constant''.
548: \end{enumerate}
549: For all $t\in\N$ set
550: $(\overline{\nu_t},\overline{\alpha_t})=\Theta_\delta^t(0,0)$. Then
551: it is clear that
552: $\overline{\nu_t}/\overline{\alpha_t}=(1-\delta)/\zeta$ for all
553: $t>0$, and that $\limsup\overline{\alpha_t}\ge1$, so
554: $\limsup\overline{\nu_t}\ge(1-\delta)/\zeta$.
555: \begin{enumerate}\setcounter{enumi}{2}
556: \item Let $t\in\N$, the ``tower height'', be such that
557: $\overline{\nu_t}>1-\epsilon/(2\#K)$.
558: \end{enumerate}
559: Then for any $\mu_1,\dots,\mu_t\ge\delta$ we also have
560: \[\Theta_{\mu_t}(\cdots\Theta_{\mu_1}(0,0)\cdots)\in(1-\epsilon/(2\#K),1]\times[0,1].\]
561: \begin{enumerate}\setcounter{enumi}{3}
562: \item Using amenability of $G$, construct finite subsets
563: $K_0=K,K_1,\dots,K_t$ of $G$, the ``Rokhlin tower'', such that for
564: all $j<i$ in $\{1,\dots,t\}$ one has $\#(K_iK_j^*)<\zeta\#K_i$,
565: and such that
566: \begin{equation}\label{eq:weiss:6}
567: \#(K_iK)\le\big(1+\frac\epsilon2\big)(1-\delta)\#K_i \text{ for all
568: }i\in\{1,\dots,t\}.
569: \end{equation}
570: \item Choose $n\in\N$, the ``quotient index'', large enough so that
571: $K_iK_i^*\cap N_n=\{1\}$ for all $i\in\{1,\dots,t\}$.
572: \end{enumerate}
573: Set $\Omega=G/N_n$, let $\pi:G\to\Omega$ denote the natural quotient
574: map, and remark, by the choice of $n$, that $K_i\to\Omega$,
575: $k\mapsto xk$ is injective for all $i\in\{1,\dots,t\}$ and all
576: $x\in\Omega$.
577:
578: Now start with $A_{t,0}=\emptyset$ and
579: $(\nu_t,\alpha_t)=(0,0)=(\overline{\nu_0},\overline{\alpha_0})$, and
580: apply $t$ times Lemma~\ref{lem:weiss}: at step $i=t,t-1,\dots,1$,
581: % \\ \begin{tabular}{|p{\textwidth}}
582: apply it with $B:=A_{i,0}$, $K:=\pi(K_i)$, $L:=\pi(K_{i-1})$, and
583: $(\nu,\alpha):=(\nu_i,\alpha_i)$. Lemma~\ref{lem:weiss} produces for
584: us
585: \begin{itemize}
586: \item constants $s\ge1$ and $\mu\ge\delta$, which we rechristen
587: $s(i)$ and $\mu_i$;
588: \item a sequence $x_1,\dots,x_{s(i)}$ in $\Omega$, which we rechristen
589: $\overline{x_{i,1}},\dots,\overline{x_{i,s(i)}}$;
590: \item a sequence $B_0\subseteq\dots\subseteq B_{s(i)}$ of subsets of
591: $\Omega$, which we rechristen
592: $A_{i,0},\dots,A_{i,s(i)}=:A_{i-1,0}$.
593: \end{itemize}
594: Set $(\nu_{i-1},\alpha_{i-1})=\Theta_{\mu_i}(\nu_i,\alpha_i)$, and
595: note that $\nu_{i-1}\ge\overline{\nu_{t-(i-1)}}$ by induction on $i$
596: and because $\mu_i\ge\delta$ and $\Theta_\mu$ is monotonically
597: increasing in $\mu$. Finally define $K_{i,j}\subset G$ by
598: $\overline{x_{i,j}}K_{i,j}=\overline{x_{i,j}}K_i\setminus
599: A_{i,j-1}$, and note by~\eqref{eq:weiss:1} that
600: \begin{equation}\label{eq:weiss:9}
601: \#(\overline{x_{i,j}}K_{i,j})\ge(1-\delta)\#K_i.
602: \end{equation}
603:
604: If $\alpha_{i-1}\ge1$, then for all $j<i-1$ set
605: $A_{j,0}=A_{i-1,0}$, $s(j)=0$ and $\nu_j=\nu_{i-1}$, and stop;
606: otherwise, decrease $i$ and continue.
607: % \end{tabular}
608:
609: After all these steps, we have obtained a decomposition
610: \begin{equation}\label{eq:weiss:4}
611: \Omega=\overline Q\sqcup\bigsqcup_{i=1}^t\bigsqcup_{j=1}^{s(i)}\overline{x_{i,j}}K_{i,j},
612: \end{equation}
613: with $\overline Q=\Omega\setminus A_{0,0}$. If the iteration was
614: stopped because $\alpha_{i-1}\ge1$, then
615: $\nu_0=\alpha_{i-1}(1-\delta)/\zeta>1-\epsilon/(2\#K)$; otherwise,
616: $\nu_0\ge\overline{\nu_t}>1-\epsilon/(2\#K)$. In all cases, we have
617: \begin{equation}\label{eq:weiss:7}
618: \#\overline Q\le(1-\nu_0)\#\Omega<\frac\epsilon{2\#K}\#\Omega.
619: \end{equation}
620: Lift $\overline Q$ and $\overline{x_{i,j}}$ to $Q\subseteq G$ and
621: $x_{i,j}\in G$ respectively. We have obtained a finite subset
622: \begin{equation}\label{eq:weiss:5}
623: T_n=Q\sqcup\bigsqcup_{i=1}^t\bigsqcup_{j=1}^{s(i)}x_{i,j}K_{i,j}
624: \end{equation}
625: of $G$. Furthermore, the natural restriction
626: $\pi\downharpoonright_{T_n}:T_n\to\Omega$ is a bijection, i.e.\
627: $T_n$ is a transversal to $N_n$. We compute
628: \begin{align*}
629: \#(T_n\cup T_nK)&=\#(T_nK)\text{ since }1\in K\\
630: &\le\#(QK)+\sum_{i,j}\#(x_{i,j}K_iK)\text{ by }\eqref{eq:weiss:5}\\
631: &\le\#Q\#K+\sum_{i,j}(1+{\textstyle\frac\epsilon2})(1-\delta)\#(x_{i,j}K_i)\text{ by }\eqref{eq:weiss:6}\\
632: &\le\#Q\#K+\sum_{i,j}(1+{\textstyle\frac\epsilon2})\#(x_{i,j}K_{i,j})\text{ by }\eqref{eq:weiss:9}\\
633: &\le\frac{\epsilon\#\Omega}{2\#K}\#K+(1+{\textstyle\frac\epsilon2})\#\Omega\text{ by }(\ref{eq:weiss:7}+\ref{eq:weiss:4})\\
634: &\le(1+\epsilon)\#T_n.\qedhere
635: \end{align*}
636: \end{proof}
637:
638: \section{Tileable amenable algebras}\label{ss:tae}
639: We prove in this section an analogue of Theorem~\ref{thm:weiss} for
640: algebras. We follow as much as possible the notation of the previous
641: section; this mainly amounts to replacing `$\#$' by `$\dim$',
642: `$\sqcup$' by `$\oplus$' and so on. The equation numbers match between
643: these two sections; Section~\ref{ss:tag} also contains a few informal
644: remarks that may help the reader along the proof(s).
645:
646: Throughout this section we consider an associative algebra $R$ over a
647: field $\K$, we use $\oplus$ and $\otimes$ to denote direct sum and
648: tensor products as $\K$-vector spaces, and we denote by `$\dim$' the
649: dimension as a $\K$-vector space. An \emph{i-subspace} of $R$ is a
650: subspace of $R$ admitting a basis consisting of invertible elements.
651: An \emph{i-algebra} is an algebra $R$ which is an i-subspace of
652: itself, and $R$ is \emph{i-amenable} if for every finite-dimensional
653: subspace $K$ and every $\epsilon>0$ there exists a
654: $(K,\epsilon)$-invariant i-subspace of $R$, in the sense of an
655: i-subspace $F$ such that
656: \[\frac{\rank(F+FK)-\rank(F)}{\rank(F)}<\epsilon.\]
657:
658: \begin{lem}
659: The group ring of an amenable group is an i-amenable i-algebra.
660: \end{lem}
661: \begin{proof}
662: The basis $G$ of $\K G$ consists of invertible elements. Following
663: the proof of Proposition~\ref{prop:crystal}, there exist
664: $(K,\epsilon)$-invariant subspaces of the form $\K F$ with
665: $F\subseteq G$, which have a basis $F$ consisting of invertible
666: elements.
667: \end{proof}
668:
669: Given a finite-dimensional subspace $A$ and a finite-dimensional
670: i-subspace $K$ of $R$, with fixed basis $\{k_1,\dots,k_d\}\subseteq
671: R^\times$, define
672: \[AK^*:=A\,\K\{k_1^{-1},\dots,k_d^{-1}\}.\]
673: Note that, although the notation does not make it apparent, $AK^*$ may
674: depend on the choice of basis of $K$.
675:
676: \begin{thm}\label{thm:tiling}
677: Let $R$ be an amenable i-algebra, let $K\le R$ be a
678: finite-dimensional i-subspace, let $\epsilon>0$ be given, and let
679: $I_0,I_1,\dots$ be a sequence of finite-codimension ideals in $R$
680: such that $\bigcap_{n\in\N} I_n=\{0\}$.
681:
682: Then for all $n\gg0$ there exists a $(K,\epsilon)$-invariant subspace
683: $T_n\le R$ that is a vector-space complement for $I_n$ in $R$.
684: \end{thm}
685:
686: Let $\zeta\ge1$ and $\delta\in(0,1)$ be constants to be fixed later.
687: In this section we shall use, for $\mu\ge\delta$, the transformation
688: $\Theta_\mu:\R_+^2\to\R_+^2$ given by
689: \[\Theta_\mu(\nu,\alpha)=\Big(\nu+\mu(1-\alpha),\alpha+\frac{\mu(1-\alpha)}{1-\delta}\zeta\Big).\]
690: The proof of Theorem~\ref{thm:tiling} relies on the following
691: \begin{lem}\label{lem:tiling}
692: Let $\Omega$ be a finite-dimensional $R$-module over an i-algebra
693: $R$, with spanning subset $\mathscr C$; let $B$ be a subspace of
694: $\Omega$ and let $K,L$ be finite-dimensional i-subspaces of $R$;
695: assume that for all $x\in\mathscr C$ the orbit map
696: \begin{equation}\label{eq:tiling:i}
697: K\to\Omega,\; k\mapsto xk\qquad\text{is injective}.
698: \end{equation}
699: Let
700: $\alpha,\nu\in[0,1)$ and $\zeta\ge1$ be such that
701: \begin{xalignat*}{2}
702: \dim(KL^*)&\le\zeta\dim K, & \dim(BK^*)&\le\alpha\dim\Omega,\\
703: \dim B&=\nu\dim\Omega, & \dim(BL^*)&\le\alpha\dim\Omega;
704: \end{xalignat*}
705: Let furthermore $\delta\in(0,1)$ be given.
706:
707: Then there exist $s\ge1$, elements $x_1,\dots,x_s\in\mathscr C$, and
708: $\mu\ge\delta$ such that, setting $B_0=B$ and $B_i=B_{i-1}+x_iK$ for
709: $1\le i\le s$ and $\Theta_\mu(\nu,\alpha)=(\nu',\alpha')$, we have
710: \begin{gather}
711: \label{eq:tiling:1}
712: \dim(x_iK\cap B_{i-1}) \le\delta\dim K\text{ for all }i\in\{1,\dots,s\};\\
713: \label{eq:tiling:2} \dim B_s=\nu'\dim\Omega;\\
714: \label{eq:tiling:3} \dim(B_sL^*)\le\alpha'\dim\Omega.
715: \end{gather}
716: \end{lem}
717: \begin{proof}
718: Let $x_1,\dots,x_s\in\mathscr C$ be a maximal-length sequence of
719: elements such that~\eqref{eq:tiling:1} holds. Let
720: $\{v_1,\dots,v_d\}$ be a maximal subset of $\mathscr C$ whose image
721: is independent in $\Omega/BK^*$; then $V:=\Bbbk\{v_1,\dots,v_d\}$ is
722: a vector space complement to $BK^*$ in $\Omega$. By maximality of
723: $(x_1,\dots,x_s)$, for all $x\in\mathscr C$ we have
724: \begin{equation}\label{eq:tiling:p1}
725: \dim(B_s\cap xK)=\dim\K\big\{x\otimes k\in\Omega\otimes K:\,xk\in B_s\big\}>\delta\dim K.
726: \end{equation}
727:
728: Let $K$ have i-basis $\{k_i\}\subseteq R^\times$; then for any
729: $A\le\Omega$ the vector space $\{\sum\omega_i\otimes k_i\in\Omega\otimes
730: K:\,\omega_ik_i\in A\text{ for all }i\}$ is isomorphic to $A\otimes K$
731: via the map
732: \begin{equation}\label{eq:tensoriso}\tag{\ref{ss:tae}.\hbox{$\dagger$}}
733: \sum \omega_i\otimes k_i\mapsto\sum \omega_ik_i\otimes k_i
734: \end{equation}
735: with inverse $\sum\xi_i\otimes k_i\mapsto\sum\xi_ik_i^{-1}\otimes
736: k_i$. We compute
737: \begin{align*}
738: \dim B_s\dim K&=\dim(B_s\otimes K)\\
739: &=\dim\Big\{\sum \omega_i\otimes k_i\in\Omega\otimes
740: K:\,\omega_i\in B_s\;\forall i\Big\}\\
741: &=\dim\Big\{\sum\omega_i\otimes k_i\in\Omega\otimes K:\,\omega_ik_i\in
742: B_s\;\forall i\Big\}\text{ by }\eqref{eq:tensoriso}\\
743: &\ge\dim\Big\{\sum\omega_i\otimes k_i\in BK^*\otimes K:\,\omega_ik_i\in
744: B_s\;\forall i\Big\}\\
745: \intertext{\hfill $\displaystyle+\dim\Big\{\sum \omega_i\otimes
746: k_i\in V\otimes K:\,\omega_ik_i\in B_s\;\forall i\Big\}$}
747: &\ge\dim\Big\{\sum\omega_ik_i^{-1}\otimes k_i\in BK^*\otimes
748: K:\,\omega_i\in B_s\;\forall i\Big\}+\dim(VK\cap B_s)\\
749: &\ge\dim(B\otimes K)+\sum_{j=1}^d\dim(v_jK\cap B_s)\text{ by }\eqref{eq:tiling:p1}\\
750: &\ge\dim B\dim K+d\delta\dim K;
751: \end{align*}
752: and $d\ge(1-\alpha)\dim\Omega$. If we divide by $\dim K\dim\Omega$
753: and set, $\mu=(\dim B_s/\dim\Omega-\nu)/(1-\alpha)$, then we get
754: $\mu\ge\delta$, and therefore $s\ge1$. Note that~\eqref{eq:tiling:2}
755: holds by the choice of $\mu$. Next,
756: \begin{align*}
757: \dim B_s&\ge\dim B+\sum_{i=1}^s\dim(x_iK/(x_iK\cap B_{i-1}))\\
758: &\ge\dim B+s(1-\delta)\dim K\text{ by }(\ref{eq:tiling:i},\ref{eq:tiling:1}),
759: \end{align*}
760: so $s\dim K\le(\nu'-\nu)\dim\Omega/(1-\delta)$; and
761: \[\dim(B_sL^*)\le\dim(BL^*)+\sum_{i=1}^s\dim(x_iKL^*)
762: \le\alpha\dim\Omega+s\zeta\dim K,
763: \]
764: from which~\eqref{eq:tiling:3} follows.
765: \end{proof}
766:
767: For a sketch of the proof of Therom~\ref{thm:tiling}, which applies as
768: well to algebra setting as to the group setting, see
769: Remark~\ref{rem:tiling} on page~\pageref{rem:tiling}.
770:
771: \begin{proof}[Proof of Theorem~\ref{thm:tiling}]
772: Without loss of generality (possibly adjoining a unit to $R$ first),
773: we suppose $1\in K$. Fix in order the following data:
774: \begin{enumerate}
775: \item Choose $\delta>0$, the ``allowed overlap'' as
776: in~\eqref{eq:tiling:1}, such that $\delta\dim K<\frac\epsilon2$ and
777: $(1+\frac\epsilon2)(1-\delta)>1$;
778: \item Choose $\zeta>1$ such that $(1-\delta)/\zeta>1-\epsilon/(2\dim
779: K)$, the ``relative F\o lner constant''.
780: \end{enumerate}
781: For all $t\in\N$ set
782: $(\overline{\nu_t},\overline{\alpha_t})=\Theta_\delta^t(0,0)$. Then
783: it is clear that
784: $\overline{\nu_t}/\overline{\alpha_t}=(1-\delta)/\zeta$ for all
785: $t>0$, and that $\limsup\overline{\alpha_t}\ge$, so
786: $\limsup\overline{\nu_t}\ge(1-\delta)/\zeta$.
787: \begin{enumerate}\setcounter{enumi}{2}
788: \item Let $t\in\N$, the ``tower height'', be such that
789: $\overline{\nu_t}>1-\epsilon/(2\dim K)$.
790: \end{enumerate}
791: Then for any $\mu_1,\dots,\mu_t\ge\delta$ we also have
792: \[\Theta_{\mu_t}(\cdots\Theta_{\mu_1}(0,0)\cdots)\in(1-\epsilon/(2\dim K),1]\times[0,1].\]
793: \begin{enumerate}\setcounter{enumi}{3}
794: \item Using i-amenability of $R$, construct finite-dimensional
795: i-subspaces $K_0=K$, $K_1,\dots,K_t$ of $R$, the ``Rokhlin tower'',
796: such that for all $j<i$ in $\{1,\dots,t\}$ one has
797: $\dim(K_iK_j^*)<\zeta\dim K_i$, and such that
798: \begin{equation}\label{eq:tiling:6}
799: \dim(K_iK)\le\big(1+\frac\epsilon2\big)(1-\delta)\dim K_i \text{ for all
800: }i\in\{1,\dots,t\}.
801: \end{equation}
802: \item Choose $n\in\N$, the ``quotient index'', large enough so that
803: $K_iK_i^*\cap I_n=(0)$ for all $i\in\{1,\dots,t\}$.
804: \end{enumerate}
805: Set $\Omega=G/I_n$, let $\pi:R\to\Omega$ denote the natural quotient
806: map, and let $\mathscr C$ denote the image in $\Omega$ of the
807: spanning set $R^\times$ of $R$. Remark, by the choice of $n$, that
808: $K_i\to\Omega$, $k\mapsto xk$ is injective for all
809: $i\in\{1,\dots,t\}$ and all $x\in\mathscr C$.
810:
811: Now start with $A_{t,0}=\{0\}\le\Omega$ and
812: $(\nu_t,\alpha_t)=(0,0)=(\overline{\nu_0},\overline{\alpha_0})$, and
813: apply $t$ times Lemma~\ref{lem:tiling}: at step $i=t,t-1,\dots,1$,
814: % \\ \begin{tabular}{|p{\textwidth}}
815: apply it with $B:=A_{i,0}$, $K:=\pi(K_i)$, $L:=\pi(K_{i-1})$, and
816: $(\nu,\alpha):=(\nu_i,\alpha_i)$. Lemma~\ref{lem:tiling} produces
817: for us
818: \begin{itemize}
819: \item constants $s\ge1$ and $\mu\ge\delta$, which we rechristen
820: $s(i)$ and $\mu_i$;
821: \item a sequence $x_1,\dots,x_{s(i)}$ in $\mathscr C$, which we
822: rechristen $\overline{x_{i,1}},\dots,\overline{x_{i,s(i)}}$;
823: \item a sequence $B_0\subseteq\dots\subseteq B_{s(i)}$ of subspaces
824: of $\Omega$, which we rechristen
825: $A_{i,0},\dots,A_{i,s(i)}=:A_{i-1,0}$.
826: \end{itemize}
827: Set $(\nu_{i-1},\alpha_{i-1})=\Theta_{\mu_i}(\nu_i,\alpha_i)$, and
828: note that $\nu_{i-1}\ge\overline{\nu_{t-(i-1)}}$ by induction on $i$
829: and because $\mu_i\ge\delta$ and $\Theta_\mu$ is monotonically
830: increasing in $\mu$. Finally let $K_{i,j}\le K_i$ be any subspace
831: such that $\overline{x_{i,j}}K_{i,j}\oplus(\overline{x_{i,j}}K_i\cap
832: A_{i,j-1})=\overline{x_{i,j}}K_i$, and note by~\eqref{eq:tiling:1}
833: that
834: \begin{equation}\label{eq:tiling:9}
835: \dim(\overline{x_{i,j}}K_{i,j})\ge(1-\delta)\dim K_i,
836: \end{equation}
837: because the $\overline{x_{i,j}}$ are invertible.
838:
839: If $\alpha_{i-1}\ge1$, then for all $j<i-1$ set $A_{j,0}=A_{i-1,0}$,
840: $s(j)=0$ and $\nu_j=\nu_{i-1}$, and stop; otherwise, decrease $i$
841: and continue.
842: % \end{tabular}
843:
844: After all these steps, we have obtained a decomposition
845: \begin{equation}\label{eq:tiling:4}
846: \Omega=\overline Q\oplus\bigoplus_{i=1}^t\bigoplus_{j=1}^{s(i)}
847: \overline{x_{i,j}}K_{i,j},
848: \end{equation}
849: where $\overline Q$ is any vector space complement to $A_{0,0}$ in
850: $\Omega$. If the iteration was stopped because $\alpha_{i-1}\ge1$,
851: then $\nu_0=\alpha_{i-1}(1-\delta)/\zeta>1-\epsilon/(2\dim K)$;
852: otherwise, $\nu_0\ge\overline{\nu_t}>1-\epsilon/(2\dim K)$. In all
853: cases, we have
854: \begin{equation}\label{eq:tiling:7}
855: \dim\overline Q\le(1-\nu_0)\dim\Omega
856: <\frac\epsilon{2\dim K}\dim\Omega.
857: \end{equation}
858: Lift $\overline Q$ and $\overline{x_{i,j}}$ to $Q\le R$ and
859: $x_{i,j}\in R^\times$ respectively. We have obtained a
860: finite-dimensional subspace
861: \begin{equation}\label{eq:tiling:5}
862: T_n=Q\oplus\bigoplus_{i=1}^t\bigoplus_{j=1}^{s(i)} x_{i,j}K_{i,j}
863: \end{equation}
864: of $R$. Furthermore, the natural restriction
865: $\pi\downharpoonright_{T_n}:T_n\to\Omega$ is a bijection, i.e.\
866: $T_n$ is a vector space complement to $I_n$. We compute
867: \begin{align*}
868: \dim(T_n+T_nK)&=\dim(TK)\text{ because }1\in K\\
869: &\le\dim(QK)+\sum_{i,j}\dim(x_{i,j}K_iK)\text{ by }\eqref{eq:tiling:5}\\
870: &\le\dim Q\dim
871: K+\sum_{i,j}(1+\textstyle{\frac\epsilon2})(1-\delta)\dim(x_{i,j}K_i)\text{
872: by }\eqref{eq:tiling:6}\\
873: &\le\dim Q\dim K+\sum_{i,j}(1+\textstyle{\frac\epsilon2})\dim(x_{i,j}K_i)\text{ by }\eqref{eq:tiling:9}\\
874: &\le\frac{\epsilon\dim\Omega}{2\dim K}\dim
875: K+(1+\textstyle{\frac\epsilon2})\dim\Omega\text{ by }(\ref{eq:tiling:7}+\ref{eq:tiling:4})\\
876: &\le(1+\epsilon)\dim T_n.\qedhere
877: \end{align*}
878: \end{proof}
879:
880: \section{Proof of Theorem~\ref{thm:vershik} and Corollary~\ref{cor:lcs}}\label{ss:vershik}
881: We start by a ``Reidemeister-Schreier'' result for algebras and ideals:
882: \begin{lem}\label{lem:rs}
883: Let $R$ be a unital $\Bbbk$-algebra generated by a subspace $S$; let
884: $I\triangleleft R$ be a right ideal in $R$; and let $F\le R$ be a
885: complement of $I$, so we have $R=I\oplus_\Bbbk F$. Let
886: $x\mapsto\overline x$ be the projection $R\to F$; assume that $1\in
887: F$ and $\overline 1=1$. Then
888: \[I=\{fs-\overline{fs}:\, f\in F,\,s\in S\}R.\]
889: \end{lem}
890: \begin{proof}
891: Write $J=\langle fs-\overline{fs}\rangle$; then obviously
892: $J\subseteq I$. Conversely, consider first $s_1\dots s_n\in R$, and write
893: $\equiv_J$ for congruence modulo $J$. Then
894: \begin{align*}
895: s_1\dots s_n &= (\overline1s_1-\overline{s_1})s_2\dots s_n+\overline{s_1}s_2\dots s_n\\
896: &\equiv_J (\overline{s_1}s_2-\overline{\overline{s_1}s_2})s_3\dots s_n+\overline{\overline{s_1}s_2}s_3\dots s_n\\
897: &\equiv_J\dots\equiv_J\overline{\overline{\overline{s_1}\cdots} s_n}.
898: \end{align*}
899: Consider now any $x=\sum_i s_{i,1}\dots s_{i,n_i}\in R$. Then
900: $x\equiv_J\sum_i\overline{\overline{\overline{s_{i,1}}\cdots}s_{i,n_i}}\in
901: F$, so $R=J+F$ and therefore $I=J$.
902: \end{proof}
903:
904: \begin{rem}
905: Consider the right ideal $I=(H-1)\Bbbk G\triangleleft\Bbbk G$ for
906: some subgroup $H$ of $G=\langle S\rangle$. Let $T$ be a right
907: transversal of $H$ in $G$; then $\Bbbk T$ is a complement of $I$ in
908: $\Bbbk G$, so $I$ is generated by $\{fs-\overline{fs}:\,f\in T,s\in
909: S\}$ by Lemma~\ref{lem:rs}, and therefore also by
910: $\{fs\overline{fs}^{-1}-1\}$; so $H$ is generated by
911: $\{fs\overline{fs}^{-1}:\,s\in S,f\in T\}$, which is the
912: Reidemeister-Schreier generating set of $H$.\hfill $\triangle$
913: \end{rem}
914:
915: \begin{cor}\label{cor:rs}
916: Let $R=\langle S\rangle$ be an augmented algebra, and consider
917: $I\triangleleft R$ with $R=I\oplus_\Bbbk F$. Then $I/I\varpi$ is
918: spanned by the image of $(F+FS)\cap I$ in $I/I\varpi$.
919: \end{cor}
920:
921: \noindent We are now ready to prove Theorem~\ref{thm:vershik}. We
922: start by a special case:
923: \begin{proof}[Proof of Theorem~\ref{thm:vershik} for $\Bbbk=\K$ a
924: field of positive characteristic]
925: Let a finite subset $S$ of $G$ and $\eta>1$ be given; we will show
926: that $\lim\sqrt[n]{\dim(\K G/\varpi^n)}\le\eta$.
927:
928: Since $\K$ has positive characteristic and $G$ is finitely
929: generated, the dimension subgroups $G_n=(1+\varpi^n)\cap G$ have
930: finite index in $G$ for all $n$. Since any quotient of an amenable
931: group is amenable, we may replace $G$ by $G/\bigcap_{n\ge1} G_n$ and
932: assume from now on that $\bigcap G_n=1$. The ideals $\varpi^n$ have
933: finite codimension in $\K G$, and have trivial intersection.
934:
935: We apply Theorem~\ref{thm:tiling} to $I_n=\varpi^n$: let $n_0\in\N$
936: be such that for all $n\ge n_0$ there is a subspace $F_n$ of $\K G$
937: with $\K G=I_n\oplus_\K F_n$ and $\dim(F_n+F_nS)<\eta\dim F_n$. Then
938: \begin{align*}
939: \dim(\K G/\varpi^{n+1})&=\dim(\K G/\varpi^n)+\dim(\varpi^n/\varpi^{n+1})\\
940: &=\dim F_n+\dim(I_n/I_n\varpi)\\
941: &\le\dim F_n+\dim((F_n+F_nS)\cap I_n)\text{ by Corollary~\ref{cor:rs}}\\
942: &\le\dim((F_n+F_nS)\cap F_n)+\dim((F_n+F_nS)\cap I_n)\\
943: &\le\dim((F_n+F_nS)\cap (F_n+I_n))=\dim(F_n+F_nS)\\
944: &\le\eta\dim F_n=\eta\dim(\K G/\varpi^n).
945: \end{align*}
946: Set $C=\dim(\K G/\varpi^{n_0})/\eta^{n_0}$. We therefore have
947: $\dim(\K G/\varpi^n)<C\eta^n$ for all $n\ge n_0$, so
948: $\lim\sqrt[n]{\dim(\K G/\varpi^n)}\le\eta$ for all $\eta>1$.
949: \end{proof}
950:
951: \begin{proof}[Proof of Theorem~\ref{thm:vershik} for $\Bbbk=\Z$]
952: For $n\ge0$ set $r_n=\rank(\varpi^n/\varpi^{n+1})$, where $\varpi$
953: denotes the augmentation ideal of $\Z G$; for $n\ge0$ and $p$ prime
954: set $s_{n,p}=\rank(\varpi_p^n/\varpi_p^{n+1})$, where $\varpi_p$
955: denotes the augmentation ideal of $\F G$; let
956: $r_{n,p}=\rank(\varpi^n/\varpi^{n+1}\otimes_\Z\Z[\frac1p])$ denote
957: the rank of the part of $\varpi^n/\varpi^{n+1}$ that is coprime to
958: $p$, and let $r_{n,0}=\rank(\varpi^n/\varpi^{n+1}\otimes_\Z\Q)$
959: denote the free rank of $\varpi^n/\varpi^{n+1}$.
960:
961: Since $G$ is finitely generated (say by $d$ elements),
962: $\varpi^n/\varpi^{n+1}$ is a finite-rank abelian group (of rank at
963: most $d^n$), so $\cong\Z^{r_{n,0}}\oplus\text{torsion}$. We thus
964: have $r_{n,p}\ge r_{n,0}$, and for fixed $n$ we have $r_{n,p}=r_{n,0}$
965: for almost all $p$.
966:
967: By Theorem~\ref{thm:vershik} for $\Bbbk=\F$, the sequence $s_{n,p}$
968: grows subexponentially for any fixed $p$. Every $\F$ factor in
969: $\varpi_p^n/\varpi_p^{n+1}$ lifts to a $\Z$-factor in $\Z G$, which
970: then gives either a $\Z$-factor in $\varpi^n/\varpi^{n+1}$, or gives
971: a torsion factor in $\varpi^m/\varpi^{m+1}$ for all $m\ge n$.
972: Therefore $r_{m,p}\le\sum_{n\le m}s_{m,p}$, so $r_{n,p}$ grows
973: subexponentially for any fixed $p$.
974:
975: The multiplication maps
976: $\varpi^m/\varpi^{m+1}\otimes\varpi^n/\varpi^{n+1}\to\varpi^{m+n}/\varpi^{m+n+1}$
977: are onto, so the sequence $r_n$ is submultiplicative ($r_mr_n\ge
978: r_{m+n}$); the same holds for the sequences $r_{n,p}$ for fixed $p$,
979: and for the sequence $r_{n,0}$.
980:
981: Let $\eta>1$ be given. Then for some $n\in\N$ we have
982: $r_{n,0}<\eta^n$; and for some $p_0$ we have $r_{n,p}=r_{n,0}$ if
983: $p>p_0$. By submultiplicativity, $r_{kn,p}\le r_{n,p}^k<\eta^{kn}$
984: for all $k$.
985:
986: For all $p\le p_0$ there exists $k_p\in\N$ such that
987: $r_{k_pn,p}<\eta^{k_pn}$, because $r_{n,p}$ grows
988: subexponentially. Set
989: $m=n\cdot\operatorname{lcm}(k_1,\dots,k_{p_0})$. Then
990: $r_{m,p}\le\eta^m$ for all $p$.
991:
992: We have $r_n=\max_pr_{n,p}$ for all $n\in\N$, so
993: $r_{km}\le\eta^{km}$ for all $k\in\N$. Since $r_n$ is
994: submultiplicative, we have
995: \[\limsup\sqrt[n]{r_n}=\lim(r_{kn})^{1/kn}\le\eta\]
996: \setbox9=\hbox{\cite{fekete:verteilung}*{page~233}}%
997: by Fekete's Lemma~\cite{polya-s:analysis}*{volume~1, part~I,
998: problem~98; originally~\box9}.
999: Since $\eta>1$ was arbitrary, the sequence $r_n$ grows
1000: subexponentially.
1001: \end{proof}
1002:
1003: \begin{proof}[Proof of Theorem~\ref{thm:vershik} for general $\Bbbk$]
1004: Let $\varpi$ denote the augmentation ideal of $\Z G$, and let
1005: $\overline\varpi$ denote the augmentation ideal of $\Bbbk G$. Since
1006: the natural map
1007: $\varpi^n/\varpi^{n+1}\otimes\Bbbk\to\overline\varpi^n/\overline\varpi^{n+1}$
1008: is onto for all $n\in\N$, we have $\rank_\Bbbk(\Bbbk
1009: G/\overline\varpi^n)\le\rank_\Z(\Z G/\varpi^n)$, so the claim
1010: follows from Theorem~\ref{thm:vershik} for $\Bbbk=\Z$.
1011: \end{proof}
1012:
1013: \begin{rem}\label{rem:vershik}
1014: The converse of Theorem~\ref{thm:vershik} does not hold: the group
1015: $SL(d,\Z)$ for $d\ge3$ is certainly not amenable (it contains free
1016: subgroups), and neither is its congruence subgroup
1017: $K=\ker(SL(d,\Z)\to SL(d,\Z/p\Z))$. This subgroup is residually-$p$
1018: if $p\ge3$, since\footnote{As usual in these situations, one should
1019: treat sometimes $2$, sometimes $4$ as the even prime to extend
1020: this result to characteristic $2$.} the subgroups
1021: $K_n=\ker(SL(d,\Z)\to SL(d,\Z/p^n\Z))$ have trivial intersection and
1022: index $p^{(n-1)(d^2-1)}$ in $K$. Then, because $K$ has the
1023: congruence property~\cite{bass-l-s:congruence}, the congruence
1024: subgroup $K_n$ coincides with the dimension subgroup as defined
1025: in~\S\ref{ss:augmented}, so $\dim(\varpi^n/\varpi^{n+1})$ grows
1026: subexponentially (approximately at rate
1027: $e^{(d^2-1)\pi\sqrt{2n/3}}$). I wish to thank M.\ Ershov for
1028: pointing out this example to me.\hfill $\triangle$
1029: \end{rem}
1030:
1031: Corollary~\ref{cor:lcs} could follow along the same lines as the proof
1032: for $\Bbbk=\Z$ of Theorem~\ref{thm:vershik}, by reducing from
1033: $\gamma_n(G)/\gamma_{n+1}(G)$ to quotients of $p$-dimension subgroups
1034: $G_{n,p}/G_{n+1,p}$ at all primes $p$, and using the fact that
1035: $\bigoplus_{n\ge1}G_{n,p}/G_{n+1,p}$ is the primitive part of the Hopf
1036: algebra $\overline{\F G}$ and therefore has subexponential growth. We
1037: will however opt for a shortcut:
1038: \begin{proof}[Proof of Corollary~\ref{cor:lcs}]
1039: The classical \emph{dimension subgroups} of $G$ are the subgroups
1040: $\delta_n(G)=G\cap(1+\varpi^n)$, where $\varpi$ denotes the
1041: augmentation ideal in $\Z G$. By a result of
1042: Gupta~\cite{gupta:odddimension}, the quotient
1043: $\delta_n(G)/\gamma_n(G)$ is a finite $2$-group. Now
1044: $\rank(\delta_n(G)/\delta_{n+1}(G))$ grows subexponentially by
1045: Theorem~\ref{thm:vershik} for $\Bbbk=\Z$, since
1046: $\delta_n(G)/\delta_{n+1}(G)$ is a submodule of
1047: $\varpi^n/\varpi^{n+1}$; and $\dim_{\F[2]}G_{n,2}/G_{n+1,2}$ grows
1048: subexponentially since $\overline{\F[2]G}$ has subexponential growth
1049: by Theorem for $\Bbbk=\F[2]$. We conclude that
1050: \begin{align*}
1051: \rank(\gamma_n(G)/\gamma_{n+1}(G))&\le\rank(\gamma_n(G)/\gamma_{n+1}(G)\otimes\Z[{\textstyle\frac12}])+\rank(\gamma_n(G)/\gamma_{n+1}(G)\otimes\F[2])\\
1052: &\le\rank(\delta_n(G)/\delta_{n+1}(G))+\rank(G_{n,2}/G_{n+1,2})
1053: \end{align*}
1054: grows subexponentially.
1055: \end{proof}
1056:
1057: % \section{Isoperimetric Profile}
1058: % There is a quantitative estimate of amenability, called the
1059: % \emph{isoperimetric profile} (see~\cite{varopoulos-s-c:analysis}*{\S
1060: % VI.1} or~\cite{gromov:asympt}*{\S5.E}\footnote{The definition is
1061: % also hinted at in~\cite{vershik:amenability}*{page 325}.}): for
1062: % groups, this is the function
1063: % \[I_G(n,S)=\max_{m\le n}\min_{\substack{F\subseteq G\\ \#F= m}}\#(F\cup
1064: % FS)-\#F.\] Amenability of
1065: % $G$ is equivalent to $\lim_{n\to\infty}I_G(n,S)/n=0$ for all finite
1066: % $S\subseteq G$.
1067: %
1068: % If $G$ is a residually-$p$ group, it is tempting to conjecture that
1069: % optimal F\o lner sets are transverse to the $\varpi^n$, in the
1070: % following sense:
1071: % \begin{conj}
1072: % For any generating set $S$ of the residually-$p$ group $G$, its
1073: % isoperimetric profile satisfies
1074: % \[I_G(n,S)\ge\max_{\rank(\F G/\varpi^t)\le
1075: % n}\rank(\varpi^t/\varpi^{t+1}).\]
1076: % \end{conj}
1077:
1078: \begin{bibsection}
1079: \begin{biblist}
1080: \bibselect{bartholdi,math}
1081: \end{biblist}
1082: \end{bibsection}
1083: \end{document}
1084: