1: %&biglatex

2: \documentclass[color]{pjm}

3: \usepackage{color,bbold}

4: \usepackage{amsrefs}

5: %\usepackage[colorlinks,citecolor=red,pagebackref,hypertexnames=false]{hyperref}

6: \usepackage[nohams]{math}

7: \usepackage[arrow,curve,matrix]{xy}

8: \setlength\textheight{22cm}

9: \setlength\textwidth{14.1cm}

10: \setlength\parskip{1ex}

11: \def\pair<#1>{{\ll}#1{\gg}}

12: 

13: \newcommand\gr{{\operatorname{\mathsf{growth}}}}

14: \newcommand\alt[1]{{\mathfrak A_{#1}}}

15: \newcommand\squish[1]{{\makebox[3em]{$\langle#1\rangle^\Gg$}}}

16: \newcommand\ssquish[1]{{\makebox[3em]{$\scriptstyle\langle#1\rangle^\Gg$}}}

17: \newcommand\sssquish[1]{{\makebox[3em]{$\scriptscriptstyle\langle#1\rangle^\Gg$}}}

18: \def\lar{\ar@{-}}

19: \def\tar{\ar@{->}} % BLUE

20: \def\aar{\ar@{.>}} % RED

21: \newcommand\rmI{{\mathrm{I}}}

22: \newcommand\rmII{{\mathrm{I\!I}}}

23: \newcommand\rmIII{{\mathrm{I\!I\!I}}}

24: \newcommand\6{\genfrac{}{}{0pt}1{\mathbb0}{\mathbb1}}

25: \newcommand\7{{\mathbb0}}

26: \newcommand\8{{\mathbb1}}

27: \newcommand\9{{\mathbb2}}

28: \newcommand\usp{{\mathbb p'}}

29: 

30: \CompileMatrices

31: %\UseCrayolaColors

32: \begin{document}

33: \title{Lie Algebras and Growth in Branch Groups}

34: \author{Laurent Bartholdi}

35: \date{April 18, 2002; Revised December 5, 2002 and \today}

36: \email{laurent@math.berkeley.edu}

37: \urladdr{\texttt{http://www.math.berkeley.edu/\char126laurent}}

38: \address{Dept of Mathematics, Evans Hall 970, U. C. Berkeley, CA

39:   94720-3840 U.S.A.}

40: \thanks{The author acknowledges support from the ``Swiss National Fund

41:   for Scientific Research'', and the Hebrew University of Jerusalem.}

42: \keywords{Lie algebra; Growth of groups; Lower Central Series}

43: \subjclass{\parbox[t]{0.55\textwidth}{

44:     \textbf{20F14} (Derived series, central series, and generalizations),\\

45:     \textbf{20F40} (Associated Lie structures),\\

46:     \textbf{17B70} (Graded Lie (super)algebras),\\

47:     \textbf{16P90} (Growth rate),\\

48:     \textbf{20E08} (Groups acting on trees)}}

49: \begin{abstract}

50:   We compute the structure of the Lie algebras associated to two

51:   examples of branch groups, and show that one has finite width while

52:   the other, the ``Gupta-Sidki group'', has unbounded width

53:   (Corollary~\ref{cor:gamma:rk}). This answers a question by Sidki.

54:   More precisely (Corollary~\ref{cor:gsgk}) the Lie algebra of the

55:   Gupta-Sidki group has Gelfand-Kirillov dimension

56:   $\log3/\log(1+\sqrt2)$.

57:   

58:   We then draw a general result relating the growth of a branch group,

59:   of its Lie algebra, of its graded group ring, and of a natural

60:   homogeneous space we call \emph{parabolic space}, namely the

61:   quotient of the group by the stabilizer of an infinite ray. The

62:   growth of the group is bounded from below by the growth of its

63:   graded group ring, which connects to the growth of the Lie algebra

64:   by a product-sum formula, and the growth of the parabolic space is

65:   bounded from below by the growth of the Lie algebra

66:   (see Theorem~\ref{thm:growth}).

67:   

68:   Finally we use this information to explicitly describe the normal

69:   subgroups of $\Gg$, the ``Grigorchuk group''.  All normal subgroups

70:   are characteristic, and the number $b_n$ of normal subgroups of

71:   $\Gg$ of index $2^n$ is odd and satisfies

72:   $\{\limsup,\liminf\}b_n/n^{\log_2(3)}=\{5^{\log_2(3)},\frac29\}$

73:   (see Corollary~\ref{cor:asympns}).

74: \end{abstract}

75: \maketitle

76: 

77: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

78: \section{Introduction}

79: The first purpose of this paper is to describe explicitly the Lie

80: algebra associated to the Gupta-Sidki group

81: $\GS$~\cite{gupta-s:burnside}, and show in this way that this group

82: is not of finite width (Corollary~\ref{cor:gamma:rk}). We shall

83: describe in Theorem~\ref{thm:gamma:struct} the Lie algebra as a graph,

84: somewhat similar to a Cayley graph, in a formalism close to that

85: introduced in~\cite{bartholdi-g:lie}.

86: 

87: We shall then consider another group, $\FG$, and show in

88: Corollary~\ref{cor:delta:rk} that although many similarities exist

89: between $\GS$ and $\FG$, the Lie algebra of $\FG$ does have

90: finite width.

91: 

92: These results follow from a description of group elements as ``branch

93: portraits'', exhibiting the relation between the group and its Lie

94: algebra. They lead to the notion of infinitely iterated ``wreath

95: algebras'', similar to wreath products of

96: groups~\cite{bartholdi:wreathalg}, to appear in a subsequent paper.

97: 

98: We shall show in Theorem~\ref{thm:growth} that, in the class of branch

99: groups, the growth of the homogeneous space $G/P$ (where $P$ is a

100: parabolic subgroup) is larger than the growth of the Lie algebra

101: $\Lie(G)$. This result parallels a lower bound on the growth of $G$ by

102: that of its graded group ring $\overline{\Bbbk G}$

103: (Proposition~\ref{prop:gpgr}).

104: 

105: Finally, we shall describe all the normal subgroups of the first

106: Grigorchuk group, using the same formalism as that used to describe

107: the lower central series. We confirm the description by Ceccherini et

108: al.\ of the low-index normal subgroups of

109: $\Gg$~\cite{ceccherini-s-t:grigns}.  It turns out that all non-trivial

110: normal subgroups are characteristic, and have finite index a power of

111: $2$.  Call $b_n$ the number of normal subgroups of index $2^n$

112: (Finite-index, non-necessarily-normal subgroups always have index a

113: power of $2$; this follows from $G$ being a $2$-torsion group.) Then

114: there are $3^k+2$ subgroups of index $2^{5\cdot2^k+1}$ and

115: $\frac293^k+1$ subgroups of index $2^{2^k+2}$; these two values are

116: extreme, in the sense that $b_n/n^{\log_2(3)}$ has lower limit

117: $5^{-\log_2(3)}$ and upper limit $\frac29$.  Also, $b_n$ is odd for

118: all $n$ (see Corollaries~\ref{cor:asympns} and~\ref{cor:allodd}).

119: 

120: \subsection{Philosophy}

121: One can hardly exaggerate the importance of Lie algebras in the study

122: of Lie groups. Lie subgroups correspond to subalgebras, normal

123: subgroups correspond to ideals; simplicity, nilpotence etc.\ match

124: perfectly. This is due to the existence of mutually-inverse functions

125: $\exp$ and $\log$ between a group and its algebra, and the

126: Campbell-Hausdorff formula expressing the group operation in terms of

127: the Lie bracket.

128: 

129: In the context of (discrete) $p$-groups and Lie algebras of

130: characteristic $p$, the correspondence is not so perfect. First, in

131: general, there is no exponential, and the best one can consider is the

132: degree-$1$ truncations

133: \[\exp(x) = 1+x+\mathcal O(x^2),\quad\log(1+x)=x+\mathcal O(x^2);\]

134: more terms would introduce denominators that in general are not

135: invertible; and no reasonable definition of convergence can be imposed

136: on $\F$. As a consequence, the group has to be subjected to a

137: filtration to yield a Lie algebra. Then there is no perfect bijection

138: between group and Lie-algebra objects.

139: 

140: However, the numerous results obtained in the area show that much can

141: be gained from consideration of these imperfect algebras. To name a

142: few, the theory of groups of finite width is closely related to the

143: classification of finite $p$-groups

144: (see~\cites{leedham-green:prop,shalev-z:finite-coclass}) and the

145: theory of pro-$p$-groups is intimately Lie-algebraic;

146: see~\cite{shalev:pro-p}, \cite{shalev:finite-p}*{\S8}

147: and~\cite{klass-lg-p:fw} with its bibliography. The solution to

148: Burnside's problems by Efim Zelmanov relies also on Lie algebras. The

149: results by Lev Kaloujnine on the $p$-Sylow subgroups of $\sym{p^n}$,

150: even if in principle independent, can be restated in terms of Lie

151: algebras in a very natural way~(see Theorem~\ref{thm:psylow}).

152: 

153: In this paper, I wish to argue that questions of growth, geometry and

154: normal subgroup structure are illuminated by Lie-algebraic

155: considerations.

156: 

157: \subsection{Notation}

158: We shall always write commutators as $[g,h]=g^{-1}h^{-1}gh$,

159: conjugates as $g^h=h^{-1}gh$, and the adjoint operators $\Ad(g)=[g,-]$

160: and $\ad(x)=[x,-]$ on the group and Lie algebra respectively. $\sym n$

161: is the symmetric group on $n$ letters, and $\alt n$ is the alternate

162: subgroup of $\sym n$.  Polynomials and power series are all written

163: over the formal variable $\hbar$, as is customary in the theory of

164: quantum algebras.  The Galois field with $p$ elements is written $\F$.

165: The cyclic group of order $n$ is written $C_n$.

166: 

167: The lower central series of $G$ is $\{\gamma_n(G)\}$, the lower

168: $p$-central series is $\{P_n(G)\}$, the dimension series is $\{G_n\}$,

169: the Lie dimension series is $\{L_n(G)\}$, and the derived series is

170: $G^{(n)}$, and in particular $G'=[G,G]$ --- the definitions shall be

171: given below.

172: 

173: As is common practice, $H<G$ means that $H$ is a

174: not-necessarily-proper subgroup of $G$.  For $H<G$, the subgroup of

175: $H$ generated by $n$-th powers of elements in $H$ is written

176: $\mho_n(H)$, and $H^{\times n}$ denotes the direct product of $n$

177: copies of $H$, avoiding the ambiguous ``$H^n$''. The normal closure of

178: $H$ in $G$ is $H^G$.

179: 

180: Finally, ``$*$'' stands for ``anything'' --- something a speaker would

181: abbreviate as ``blah, blah, blah'' in a talk. It is used to mean

182: either that the value is irrelevant to the rest of the computation, or

183: that it is the only unknown in an equation and therefore does not

184: warrant a special name.

185: 

186: \subsection{$N$-series}

187: We first recall a classical construction of Magnus~\cite{magnus:lie},

188: described for instance in~\cite{lazard:nilp}

189: and~\cite{huppert-b:fg2}*{Chapter~VIII}.

190: \begin{defn}

191:   Let $G$ be a group. An \emdef{$N$-series} is series $\{H_n\}$ of

192:   normal subgroups with $H_1=G$, $H_{n+1}\le H_n$ and $[H_m,H_n]\leq

193:   H_{m+n}$ for all $m,n\ge1$.

194:   The associated Lie ring is

195:   \[\Lie(G) = \bigoplus_{n=1}^\infty \Lie_n,\]

196:   with $\Lie_n=H_n/H_{n+1}$ and the bracket operation

197:   $\Lie_n\otimes\Lie_m\to\Lie_{m+n}$ induced by commutation in $G$.

198: 

199:   For $p$ a prime, an \emdef{$N_p$-series} is an $N$-series

200:   $\{H_n\}$ such that $\mho_p(H_n)\le H_{pn}$, and the associated Lie ring

201:   is a restricted Lie algebra over $\F$.

202:   \[\Lie_{\F}(G) = \bigoplus_{n=1}^\infty \Lie_n,\]

203:   with the $p$-mapping $\Lie_n\to\Lie_{pn}$ induced by raising to the

204:   power $p$ in $H_n$.

205: \end{defn}

206: 

207: We recall that $\Lie$ is a \emph{restricted} Lie algebra

208: (see~\cite{jacobson:restr} or~\cite{strade-f:mla}*{Section~2.1}) if it

209: is over a field $\Bbbk$ of characterstic $p$, and there exists a

210: mapping $x\mapsto x^{[p]}$ such that $\ad x^{[p]}=\ad(x)^p$, $(\alpha

211: x)^{[p]}=\alpha^px^{[p]}$ and

212: $(x+y)^{[p]}=x^{[p]}+y^{[p]}+\sum_{i=1}^{p-1}s_i(x,y)$, where the

213: $s_i$ are obtained by expanding

214: $\ad(x\otimes\hbar+y\otimes1)^{p-1}(a\otimes1)=\sum_{i=1}^{p-1}s_i(x,y)\otimes

215: i\hbar^{i-1}$ in $\Lie\otimes\Bbbk[\hbar]$.  Equivalently,

216: \begin{prop}[Jacobson]\label{prop:jacobson}

217:   Let $(e_i)$ be a basis of $\Lie$ such that, for some $y_i\in\Lie$,

218:   we have $\ad(e_i)^p=\ad(y_i)$. Then $\Lie$ is restricted; more

219:   precisely, there exists a unique $p$-mapping such that

220:   $e_i^{[p]}=y_i$.

221: \end{prop}

222: 

223: The standard example of $N$-series is the \emph{lower central series},

224: $\{\gamma_n(G)\}_{n=1}^\infty$, given by $\gamma_1(G)=G$ and

225: $\gamma_n(G)=[G,\gamma_{n-1}(G)]$, or the \emph{lower exponent-$p$

226:   central series} or \emph{Frattini series} given by $P_1(G)=G$ and

227: $P_n(G)=[G,P_{n-1}(G)]\mho_p(P_{n-1}(G))$. It differs from the lower

228: central series in that its successive quotients are all elementary

229: $p$-groups.

230: 

231: The standard example of $N_p$-series is the \emph{dimension series},

232: also known as the $p$-lower central,

233: Zassenhaus~\cite{zassenhaus:ordnen}, Jennings~\cite{jennings:gpring},

234: Lazard~\cite{lazard:nilp} or Brauer series, given by $G_1=G$ and

235: $G_n=[G,G_{n-1}]\mho_p(G_{\lceil n/p \rceil})$, where $\lceil n/p

236: \rceil$ is the least integer greater than or equal to $n/p$. It can

237: alternatively be described, by a result of Lazard~\cite{lazard:nilp},

238: as

239: \begin{equation}\label{eq:lazard}

240:   G_n = \prod_{i\cdot p^j\ge n}\mho_{p^j}(\gamma_i(G)),

241: \end{equation}

242: or as

243: \[G_n = \{g\in G|\,g-1\in\varpi^n\},\]

244: where $\varpi$ is the augmentation (or fundamental) ideal of the group

245: algebra $\F G$. Note that this last definition extends to

246: characteristic $0$, giving a graded Lie algebra $\Lie_\Q(G)$ over

247: $\Q$. In that case, the subgroup $G_n$ is the isolator of

248: $\gamma_n(G)$:

249: \[G_n = \sqrt{\gamma_n(G)} = \{g\in G|\,\langle

250: g\rangle\cap\gamma_n(G)\neq\{1\}\}.\]

251: A good reference for these results is~\cite{passi:gr}*{Chapter~VIII}.

252: 

253: We mention finally for completeness another $N_p$-series, the

254: \emdef{Lie dimension series} $L_n(G)$ defined by

255: \[L_n(G) = \{g\in G|\,g-1\in\varpi^{(n)}\},\]

256: where $\varpi^{(n)}$ is the $n$-th Lie power of $\varpi<\Bbbk G$,

257: given by $\varpi^{(1)}=\varpi$ and

258: $\varpi^{(n+1)}=[\varpi^{(n)},\varpi]=\{xy-yx|\,x\in\varpi^{(n)},y\in\varpi\}$.

259: It is then known~\cite{passi-s:liedim} that

260: \[L_n(G) = \prod_{(i-1)\cdot p^j\ge n}\mho_{p^j}(\gamma_i(G))\]

261: if $\Bbbk$ is of characteristic $p$, and

262: \[L_n(G) = \sqrt{\gamma_n(G)}\cap [G,G]\]

263: if $\Bbbk$ is of characteristic $0$.

264: 

265: In the sequel we will only consider the $N$-series $\{\gamma_n(G)\}$

266: and $\{P_n(G)\}$ and the $N_p$-series $\{G_n\}$ of dimension

267: subgroups. We reserve the symbols $\Lie$ and $\Lie_{\F}$ for their

268: respective Lie algebras.

269: 

270: \begin{defn}

271:   Let $\{H_n\}$ be an $N$-series for $G$. The \emph{degree} of $g\in

272:   G$ is the maximal $n\in\N\cup\{\infty\}$ such that $g$ belongs to

273:   $H_n$.

274: \end{defn}

275: 

276: A series $\{H_n\}$ has \emdef{finite width} if there is a constant $W$

277: such that $\ell_n:=\rank[H_n:H_{n+1}]\le W$ holds for all $n$ (Here

278: $\rank A$ is the minimal number of generators of the abelian group

279: $A$).  A group has \emdef{finite width} if its lower central series

280: has finite width --- this definition comes from~\cite{klass-lg-p:fw}.

281: 

282: \begin{defn}

283:   Let $a=\{a_n\}$ and $b=\{b_n\}$ be two sequences of real numbers. We

284:   write $a\precsim b$ if there is an integer $C>0$ such that

285:   $a_n<Cb_{Cn+C}+C$ for all $n\in\N$, and write $a\sim b$ if

286:   $a\precsim b$ and $b\precsim a$.

287: \end{defn}

288: In the sense of this definition, a group has finite width if and only

289: if $\{\ell_n\}\sim\{1\}$.

290: 

291: I do not know the answer to the following natural

292: \begin{question}

293:   If $\rank(\gamma_n(G)/\gamma_{n+1}(G))$ is bounded, does that imply

294:   that $\rank(G_n/G_{n+1})$, $\rank(P_n(G)/P_{n+1}(G))$ or

295:   $\rank(L_n(G)/L_{n+1}(G))$ is bounded?  and conversely?

296:   

297:   More generally, say an $N$-series $\{H_n\}$ has \emdef{finite width}

298:   if $\rank(H_n/H_{n+1})$ is bounded over $n\in\N$. If $G$ has a

299:   finite-width $N$-series intersecting to $\{1\}$, are all $N$-series

300:   of $G$ of finite width?

301: \end{question}

302: 

303: The following result is well-known, and shows that sometimes the

304: Lie ring $\Lie(G)$ is actually a Lie algebra over $\F$.

305: \begin{lem}\label{lem:liealg}

306:   Let $G$ be a group generated by a set $S$. Let $\Lie(G)$ be the Lie

307:   ring associated to the lower central series.

308:   \begin{enumerate}

309:   \item If $S$ is finite, then $\Lie_n$ is a finite-rank $\Z$-module

310:     for all $n$.

311:   \item If there is a prime $p$ such that all generators $s\in S$ have

312:     order $p$, then $\Lie_n$ is a vector space over $\F$ for all $n$.

313:     It then follows that the Frattini series (for that prime $p$) and

314:     the lower central series coincide.

315:   \end{enumerate}

316: \end{lem}

317: \begin{proof}

318:   First, $\Lie_1$ is generated by $\overline S$, the image of $S$ in

319:   $G/G'$. Since $\Lie$ is generated by $\Lie_1$, in particular

320:   $\Lie_n$ is generated by the finitely many $(n-1)$-fold products of

321:   elements of $\overline S$; this proves the first point.

322:   

323:   Actually, far fewer generators are required for $\Lie_n$; in the

324:   extremal case when $G$ is a free group, a basis of $\Lie_n$ is given

325:   in terms of ``standard monomials'' of degree $n$; see

326:   Subsection~\ref{subs:free} or~\cite{hall:liebasis}.

327:   

328:   For the second claim, assume more generally that $s^p\in

329:   G'$ for all $s\in S$, so that  $G/G'$ is an $\F$-vector space. We

330:   use the identity $[x,y]^p\equiv [x,y^p]\mod\gamma_3\langle

331:   x,y\rangle$, due to Philip Hall. Let $g=[x,y]$ be a generator of

332:   $\gamma_n(G)$, with $x\in G$ and $y\in\gamma_{n-1}(G)$. Then

333:   $y^p\in\gamma_n(G)$ by induction, so $g^p\in\gamma_{n+1}(G)$ and

334:   $\Lie_n$ is an $\F$-vector space.

335: \end{proof}

336: 

337: Anticipating, we note that the groups $\GS$ and $\FG$ we shall

338: consider satisfy these hypotheses for $p=3$, and $\Gg$ satisfies them

339: for $p=2$.

340: 

341: \subsection{Growth of groups and vector spaces}

342: Let $G$ be a group generated by a finite set $S$. The \emdef{length}

343: $|g|$ of an element $g\in G$ is the minimal number $n$ such that $g$

344: can be written as $s_1\dots s_n$ with $s_i\in S$. The \emdef{growth

345:   series} of $G$ is the formal power series

346: \[\gr(G)=\sum_{g\in G}\hbar^{|g|}=\sum_{n\ge0}f_n\hbar^n,\]

347: where $f_n=\#\{g\in G|\,|g|=n\}$.  The \emdef{growth function} of $G$ is

348: the $\sim$-equivalence class of the sequence $\{f_n\}$. Note that

349: although $\gr(G)$ depends on $S$, this equivalence class is

350: independent of the choice of $S$.

351: 

352: Let $X$ be a transitive $G$-set and $x_0\in X$ be a fixed base point.

353: The \emdef{length} $|x|$ of an element $x\in X$ is the minimal length

354: of a $g\in G$ moving $x_0$ to $x$. The \emdef{growth series} of $X$ is

355: the formal power series

356: \[\gr(X,x_0)=\sum_{x\in X}\hbar^{|x|}=\sum_{n\ge0}f_n\hbar^n,\]

357: where $f_n=\{x\in X|\,\min_{gx_0=x}|g|=n\}$.  The \emdef{growth

358:   function} of $X$ is the the $\sim$-equivalence class of the sequence

359: $\{f_n\}$. It is again independent of the choice of $x_0$ and of

360: generators of $G$.

361: 

362: Let $V=\bigoplus_{n\ge0}V_n$ be a graded vector space. The

363: \emdef{Hilbert-Poincar\'e} series of $V$ is the formal power series

364: \[\gr(V)=\sum_{n\ge0}v_n\hbar^n=\sum_{n\ge0}\dim V_n\hbar^n.\]

365: 

366: We return to the dimension series of $G$. Consider the graded algebra

367: \[\overline{\F G}=\bigoplus_{n=0}^\infty\varpi^n/\varpi^{n+1}.\]

368: A fundamental result connecting $\Lie_{\F}(G)$ and $\overline{\F G}$ is

369: the

370: \begin{thm}[Quillen~\cite{quillen:ab}]\label{thm:quillen}

371:   $\overline{\F G}$ is the restricted enveloping algebra of the Lie

372:   algebra $\Lie_{\F}(G)$ associated to the dimension series.

373: \end{thm}

374: 

375: The Poincar\'e-Birkhoff-Witt Theorem then gives a basis of

376: $\overline{\F G}$ consisting of monomials over a basis of

377: $\Lie_{\F}(G)$, with exponents at most $p-1$. As a consequence, we

378: have the

379: \begin{prop}[Jennings~\cite{jennings:gpring}]\label{prop:sumprod}

380:   Let $G$ be a group, and let $\sum_{n\ge1}\ell_n\hbar^n$ be the

381:   Hilbert-Poincar\'e series of $\Lie_{\F}(G)$. Then

382:   \[\gr(\overline{\F G})

383:   =\prod_{n=1}^\infty\left(\frac{1-\hbar^{pn}}{1-\hbar^n}\right)^{\ell_n}.\]

384: \end{prop}

385: 

386: Approximations from analytical number theory~\cite{li:nt} and complex

387: analysis give then the

388: \begin{prop}[\cite{bartholdi-g:lie}, Proposition~2.2

389:   and~\cite{petrogradsky:polynilpotent}, Theorem~2.1]\label{prop:grlgr}

390:   Let $G$ be a group and expand the power series

391:   $\gr(\Lie_{\F}(G))=\sum_{n\ge1}\ell_n\hbar^n$ and

392:   $\gr(\overline{\F G})=\sum_{n\ge0}f_n\hbar^n$. Then

393:   \begin{enumerate}

394:   \item $\{f_n\}$ grows exponentially if and only if $\{\ell_n\}$

395:     does, and we have

396:     \[\limsup_{n\to\infty}\frac{\ln \ell_n}n

397:     =\limsup_{n\to\infty}\frac{\ln f_n}n.\]

398:   \item If $\ell_n\sim n^d$, then $f_n\sim e^{n^{(d+1)/(d+2)}}$.

399:   \end{enumerate}

400: \end{prop}

401: 

402: The Lie algebras we consider have polynomial growth, i.e.\ finite

403: Gelfand-Kirillov dimension. This notion is more commonly studied for

404: associative rings~\cite{gelfand-k:dimension}:

405: \begin{defn}\label{def:gkdim}

406:   Let $\Lie=\oplus\Lie_n$ be a graded Lie algebra. Its

407:   \emph{Gelfand-Kirillov dimension} is

408:   \[\dim_{GK}(\Lie)=\limsup_{n\to\infty}\frac{\log\left(\dim\Lie_1+\dots+\dim{\Lie_n}\right)}{\log n}.\]

409: \end{defn}

410: Note that if $\ell_n\sim n^d$, then $\Lie$ has Gelfand-Kirillov

411: dimension $d+1$. However, the converse is not true, since the sequence

412: $\log(\ell_1+\dots+\ell_n)/\log n$ need not converge. If the group $G$ has

413: finite width, then its algebra $\Lie(G)$ has Gelfand-Kirillov

414: dimension $1$.

415: 

416: Note also that if $A$ is any algebra generated in degree $1$, then

417: $\dim_{GK}(A)=0$ or $\dim_{GK}(A)\ge1$. Furthermore, George Bergman

418: showed in~\cite{bergman:growth} that if $A$ is associative, then

419: $\dim_{GK}(A)=1$ or $\dim_{GK}(A)\ge2$. Victor Petrogradsky showed

420: in~\cite{petrogradsky:anydim} that there exist Lie algebras of any

421: Gelfand-Kirillov dimension $\ge1$.

422: 

423: %%%%% FALSE!

424: %Let $G$ be a group generated by a finite set $S$, and let $(X,x_0)$ be

425: %a $G$-space. The representation space $\Bbbk X$ naturally has the

426: %structure of a $\Bbbk G$-module via $\rho:\Bbbk G\to\End(\Bbbk X)$,

427: %and the algebra $T=\Bbbk G/\ker\rho$, as a quotient of $\Bbbk G$, is

428: %naturally filtered by powers of $\varpi$.  Denote by $\overline T$ the

429: %associated graded algebra. The algebra $T$ was first considered by

430: %Sidki in~\cite{sidki:primitive} in the special case $X=G/\stab(e)$,

431: %with $e$ an infinite ray in the tree.

432: %

433: %We use $\sum f_n\hbar^n\ge\sum g_n\hbar^n$ to mean $f_n\ge g_n$ for

434: %all $n\in\N$.

435: %\begin{prop}\label{prop:hsgr}

436: %  Let $G$ be a group generated by a finite set $S$ and act on

437: %  $(X,x_0)$. Then

438: %  \[\frac{\gr(X,x_0)}{1-\hbar}\ge\gr(\overline T).\]

439: %\end{prop}

440: %\begin{proof}

441: %  By restricting to the orbit of $x_0$, we may at no cost assume that

442: %  $X$ is a transitive $G$-space, i.e.\ $X=G/H$ with $x_0=H$; then

443: %  $T=\Bbbk G/\langle h-1:\,h\in H\rangle$.  Denote again by $\varpi$ the

444: %  image of the augmentation ideal of $\Bbbk G$ in $T$.

445: %  

446: %  First, $\varpi^n$ is generated by all expressions of the form

447: %  $g_0(1-f_1)g_1\cdots(1-f_n)g_n+\langle H-1\rangle\equiv

448: %  g_0(1-f_1)g_1\cdots(1-f_n)g_nH$ with $f_i,g_i\in G$. Adding terms

449: %  from $\varpi^{n+1}$, we see that $\varpi^n/\varpi^{n+1}$ is

450: %  generated by the images of $(1-f_1)\cdots(1-f_n)H$. Then note the

451: %  identities

452: %  \[1-xy \equiv (1-x)+(1-y)\mod\varpi^2,\quad (1-x^{-1})\equiv-(1-x)\mod\varpi^2;\]

453: %  using them, we may pick a basis for $\varpi^n/\varpi^{n+1}$ among

454: %  all $(1-s_1)\cdots(1-s_n)H$ with $s_i\in S$. Now all these products

455: %  lie in the vector subspace $\Bbbk\{gH\in X|\,|g|\le n\}$ of $\Bbbk

456: %  X$, and therefore $\dim(\varpi^n/\varpi^{n+1})\le f_0+\dots+f_n$,

457: %  where $\gr(X,x_0)=\sum f_i\hbar^i$.

458: %\end{proof}

459: Finally, we recall a connection between the growth of $G$ and that of

460: $\overline{\F G}$. We use the notation $\sum f_n\hbar^n\ge\sum

461: g_n\hbar^n$ to mean $f_n\ge g_n$ for all $n\in\N$.

462: 

463: \begin{prop}[\cite{grigorchuk:hp}, Lemma~8]\label{prop:gpgr}

464:   Let $G$ be a group generated by a finite set $S$. Then

465:   \[\frac{\gr(G)}{1-\hbar}\ge\gr(\overline{\Bbbk G}).\]

466: \end{prop}

467: 

468: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

469: \section{Branch groups}

470: Branch groups were introduced by Rostislav Grigorchuk

471: in~\cite{grigorchuk:jibg}, where he develops a general theory of

472: groups acting on rooted trees. We shall content ourselves with a

473: restricted definition; recall that $G\wr\sym d$ is the \emdef{wreath

474:   product} $G^{\times d}\rtimes\sym d$, the action of $\sym d$ on the

475: direct product induced by the permutation action of $\sym d$ on

476: $\Sigma=\{\mathsf1,\dots,\mathsf d\}$.

477: \begin{defn}

478:   A group $G$ is \emdef{regular branch} if for some $d\in\N$ there is

479:   \begin{enumerate}

480:   \item an embedding $\psi:G\hookrightarrow G\wr\sym d$ such that the

481:     image of $\psi(G)$ in $\sym d$ acts transitively on $\Sigma$.

482:     Define for $n\in\N$ the subgroups $\stab_G(n)$ of $G$ by

483:     $\stab_G(0)=G$, and inductively

484:     \[\stab_G(n)=\psi^{-1}(\stab_G(n-1)^{\times d})\]

485:     where $\stab_G(n-1)^{\times d}$ is seen as a subgroup of $G\wr\sym

486:     d$. One requires then that $\bigcap_{n\in\N}\stab_G(n)=\{1\}$;

487:   \item a subgroup $K<G$ of finite index with $\psi(K)<K^{\times d}$.

488:   \end{enumerate}

489: \end{defn}

490: To avoid ambiguous bracket notations, we write the decomposition map

491: \[\psi(g)=\pair<g_1,\dots,g_d>\pi,\]

492: with $\pi$ expressed as a permutation in disjoint cycle notation.

493: 

494: We shall abbreviate ``regular branch'' to ``branch'', since all the

495: branch groups that appear in this paper are actually regular branch.

496: We shall usually omit $d$ from the description, and say that ``$G$

497: branches over $K$''.

498: 

499: \begin{lem}\label{lem:branchnormal}

500:   If $G$ is a branch group, then $G$ branches over a subgroup $K$ of

501:   $G$ such that $K$ is normal in $G$, and $K^{\times d}$ is normal in

502:   $\psi(K)$.

503: \end{lem}

504: \begin{proof}

505:   Let $G$ be branch over $L$ of finite index, and set $K=\bigcup_{g\in

506:     G}L^g$, the \emph{core} of $L$. Then obviously $L\triangleleft G$;

507:   and since $(L^{\times d})^{\psi g}<\psi(K^g)$ for all $g\in G$, we

508:   have, writing $\psi(g)=\pair<g_1,\dots,g_d>\pi$,

509:   \[K^{\times d}\le\bigcap_{g\in G}(L^{g_{1^\pi}}\times\dots\times

510:   L^{g_{d^\pi}})=\bigcap_{g\in G}(L^{\times d})^{\psi g}<K,

511:   \]

512:   and $(K^{\times d})^{\psi(g)}=K^{g_{1^\pi}}\times\dots\times

513:   K^{g_{d^\pi}}=K^{\times d}$, so $K^{\times d}\triangleleft\psi(G)$.

514: \end{proof}

515: 

516: Let $G$ be a branch group, with $d$, $\Sigma$ and $K$ as in the

517: definition. The \emdef{rooted tree} on $\Sigma$ is the free monoid

518: $\Sigma^*$, with root the empty sequence $\emptyset$; it is a metric

519: space for the distance

520: \[\dist(\sigma,\tau)=|\sigma|+|\tau|-2\max\{n\in\N|\,\sigma_n=\tau_n\}.\]

521: The \emdef{natural action} of $G$ is an action on $\Sigma^*$, defined

522: inductively by

523: \begin{equation}

524:   (\sigma_1\sigma_2\dots\sigma_n)^g

525:   =(\sigma_1)^\pi(\sigma_2\dots\sigma_n)^{g_{\sigma_1}}\text{ for

526:   }\sigma_1,\dots,\sigma_n\in\Sigma,\label{eq:action}

527: \end{equation}

528: where $\psi(g)=\pair<g_1,\dots,g_d>\pi$. By the condition

529: $\bigcap\stab_G(n)=\{1\}$, this action is faithful and $G$ is

530: residually finite. Note that $\stab_G(n)$ is the fixator of $\Sigma^n$

531: in this action.

532: 

533: Note that the action~\eqref{eq:action} gives geometrical meaning to

534: the branch structure of $G$ that closely parallels the structure of

535: the tree $\Sigma^*$. Indeed one may consider $G$ as a group acting on

536: the tree $\Sigma^*$; then the choice of a vertex $\sigma$ of

537: $\Sigma^*$ and of a subgroup $J$ of $K$ determines a subgroup

538: $L_\sigma$ of $K$, namely the group of tree-automorphisms of

539: $\Sigma^*$ that fix $\Sigma^*\setminus\sigma\Sigma^*$ and whose action

540: on $\sigma\Sigma^*$ is that of an element of $J$ on $\Sigma^*$. The

541: choice of a subgroup $J_\sigma$ for all $\sigma\in\Sigma^*$ determines

542: a subgroup $M$ of $K$, namely the closure of the $L_\sigma$ associated

543: to $\sigma$ and $J_\sigma$ when $\sigma$ ranges over $\Sigma^*$.

544: 

545: This geometrical vision can also give pictorial descriptions of 

546: the group elements:

547: \begin{defn}\label{defn:bp}

548:   Suppose $G$ branches over $K$; let $T$ be a transversal of $K$ in

549:   $G$, and let $U$ be a transversal of $\psi^{-1}(K^{\times d})$ in

550:   $K$. The \emdef{branch portrait} of an element $g\in G$ is a

551:   labeling of $\Sigma^*$, as follows: the root vertex $\emptyset$ is

552:   labeled by an element of $TU$, and all other vertices are labeled

553:   by an element of $U$.

554:   

555:   Given $g\in G$: write first $g=kt$ with $k\in K$ and $t\in T$; then

556:   write $k=\psi^{-1}(k_{\mathsf1},\dots,k_{\mathsf d})u_\emptyset$,

557:   and inductively

558:   $k_\sigma=\psi^{-1}(k_{\sigma\mathsf1},\dots,k_{\sigma\mathsf

559:     d})u_\sigma$ for all $\sigma\in\Sigma^*$. Label the root vertex by

560:   $tu_\emptyset$ and the label the vertex $\sigma\neq\emptyset$ by

561:   $u_\sigma$.

562: \end{defn}

563: 

564: There are uncountably many branch portraits, even for a countable

565: branch group. We therefore introduce the following notion:

566: \begin{defn}

567:   Let $G$ be a branch group. Its \emdef{completion} $\overline G$ is

568:   the inverse limit

569:   \[\projlim_{n\to\infty}G/\stab_G(n).\]

570:   This is also the closure in $\aut\Sigma^*$ of $G$ seen through its

571:   natural action~\eqref{eq:action}.

572: \end{defn}

573: Note that since $\overline G$ is closed in $\aut\Sigma^*$ it is a

574: profinite group, and thus is compact, and totally disconnected. If $G$

575: has the ``congruence subgroup property''~\cite{grigorchuk:jibg},

576: meaning that all finite-index subgroups of $G$ contain $\stab_G(n)$

577: for some $n$, then $\overline G$ is also the profinite completion of

578: $G$.

579: 

580: \begin{lem}\label{lem:portraits}

581:   Let $G$ be a branch group and $\overline G$ its completion. Then

582:   Definition~\ref{defn:bp} yields a bijection between the set of

583:   branch portraits and $\overline G$.

584: \end{lem}

585: 

586: We shall often simplify notation by omitting $\psi$ from subgroup

587: descriptions, as for instance in statements like

588: ``$\stab_G(n)<\stab_G(n-1)^{\times d}$.''

589: 

590: %----------------------------------------------------------------

591: \subsection{The group $\Gg$}

592: We shall consider more carefully three examples of branch groups in

593: the sequel. The first example of branch group was considered by

594: Rostislav Grigorchuk in 1980, and appeared innumerably often in

595: recent mathematics --- the entire chapter~VIII of~\cite{harpe:ggt} is

596: devoted to it. It is defined as follows: it is a $4$-generated group

597: $\Gg$ (with generators $a,b,c,d$), its map $\psi$ is given by

598: \[\psi:\begin{cases}

599:   \Gg&\hookrightarrow (\Gg\times\Gg)\rtimes\sym 2\\

600:   a&\mapsto\makebox[7em][l]{$\pair<1,1>(\mathsf1,\mathsf2),$}

601:   b\mapsto\pair<a,c>,\\

602:   c&\mapsto\makebox[7em][l]{$\pair<a,d>,$}d\mapsto\pair<1,b>

603: \end{cases}\]

604: and its subgroup $K$ is the normal closure of $[a,b]$, of index $16$.

605: Rostislav Grigorchuk proved

606: in~\cites{grigorchuk:burnside,grigorchuk:growth} that $\Gg$ is an

607: intermediate-growth, infinite torsion group. Its lower central series

608: was computed in~\cite{bartholdi-g:lie}, along with a description of

609: its Lie algebra. We shall reproduce that result using a more general

610: method.

611: 

612: %---------------------------------------------------------------

613: \subsection{The group $\GS$}

614: This $2$-generated group was introduced by Narain Gupta and Said Sidki

615: in~\cite{gupta-s:burnside}, where they proved it to be an infinite

616: torsion group. Later Said Sidki obtained a complete description of its

617: automorphism group~\cite{sidki:subgroups}, along with information on

618: its subgroups. It is a branch group with generators $a,t$, its

619: map $\psi$ is given by

620: \[\psi:\begin{cases}

621:   \GS&\hookrightarrow(\GS\times\GS\times\GS)\rtimes\alt3\\

622:   a&\mapsto\pair<1,1,1>(\mathsf1,\mathsf2,\mathsf3)\\

623:   t&\mapsto\pair<a,a^{-1},t>,

624: \end{cases}\]

625: and its subgroup $K$ is $\GS'$, of index $9$.

626: 

627: The author proved recently~\cite{bartholdi:phd} that $\GS$ has

628: intermediate growth, which increases its analogy with the Grigorchuk

629: group mentioned above. An outstanding question was whether $\GS$

630: has finite width. Ana Cristina Vieira computed

631: in~\cites{vieira:lcs,vieira:subgroups} the first $9$ terms of the lower

632: central series and showed that there are all of rank at most $2$. We

633: shall shortly see, however, that $\GS$ has unbounded width.

634: 

635: The following lemma is straightforward:

636: \begin{lem}

637:   $\GS'/(\GS'\times\GS'\times\GS')$ is isomorphic to

638:  $C_3\times C_3$, generated by $c=[a,t]$ and $u=[a,c]$.

639: \end{lem}

640: 

641: Note finally that the notations in~\cite{sidki:subgroups} are slightly

642: different: his $x$ is our $a$, and his $y$ is our $t$.

643: In~\cite{vieira:lcs} her $y^{[1]}$ is our $u$, and more generally her

644: $g_1$ is our $\7(g)$ and her $g^{[1]}$ is our $\9(g)$.

645: In~\cite{bartholdi-g:parabolic}, where a great deal of information on

646: $\GS$ is gathered, the group is called $\doverline\Gamma$.

647: 

648: %---------------------------------------------------------------

649: \subsection{The group $\FG$}

650: This other group is at first sight close to $\GS$: it is also

651: branch, and generated by two elements $a,t$. Its map $\psi$ is given

652: by

653: \[\psi:\begin{cases}

654:   \FG&\hookrightarrow(\FG\times\FG\times\FG)\rtimes\alt3\\

655:   a&\mapsto\pair<1,1,1>(\mathsf1,\mathsf2,\mathsf3)\\

656:   t&\mapsto\pair<a,1,t>,

657: \end{cases}\]

658: and its subgroup $K$ is $\FG'$, of index $9$.

659: 

660: This group was first considered by Jacek Fabrykowski and Narain

661: Gupta~\cite{fabrykowski-g:growth2}, who studied its growth.

662: In~\cite{bartholdi-g:parabolic}, Rostislav Grigorchuk and the author

663: proved that it is a branch group, and that its subgroup $L=\langle

664: at,ta\rangle$ has index $3$ and is torsion-free.

665: In~\cite{bartholdi:phd} another proof of $\FG$'s subexponential growth

666: is given.

667: 

668: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

669: \section{Lie algebras}

670: We shall now describe the Lie algebras associated to the groups $\Gg$,

671: $\GS$ and $\FG$ defined in the previous section. We start by

672: considering a group $G$, and make the following hypotheses on $G$,

673: which will be satisfied by $\Gg$, $\GS$ and $\FG$:

674: \begin{enumerate}

675: \item $G$ is finitely generated by a set $S$;

676: \item there is a prime $p$ such that all $s\in S$ have order $p$.

677: \end{enumerate}

678: Under these conditions, it follows from Lemma~\ref{lem:liealg} that

679: $\gamma_n(G)/\gamma_{n+1}(G)$ is a finite-dimensional vector space

680: over $\F$, and therefore that $\Lie(G)$ is a Lie algebra over $\F$

681: that is finite at each dimension. Clearly the same property holds for

682: the restricted algebra $\Lie_{\F}(G)$.

683: 

684: We propose the following notation for such algebras:

685: \begin{defn}\label{defn:liegf}

686:   Let $\Lie=\bigoplus_{n\ge1}\Lie_n$ be a graded Lie algebra over

687:   $\F$, and choose a basis $B_n$ of $\Lie_n$ for all $n\ge1$. For

688:   $x\in\Lie_n$ and $b\in B_n$ denote by $\langle x|b\rangle$ the

689:   $b$-coefficient of $x$ in base $B_n$.

690:   

691:   The \emdef{Lie graph} associated to these choices is an abstract

692:   graph. Its vertex set is $\bigcup_{n\ge1}B_n$, and each vertex $x\in

693:   B_n$ has a degree, $n=\deg x$. Its edges are labeled as $\alpha x$,

694:   with $x\in B_1$ and $\alpha\in\F$, and may only connect a vertex

695:   of degree $n$ to a vertex of degree $n+1$. For all $x\in B_1$, $y\in

696:   B_n$ and $z\in B_{n+1}$, there is an edge labeled $\langle

697:   [x,y]|z\rangle x$ from $y$ to $z$.

698:   

699:   If $\Lie$ is a restricted algebra of $\F$, there are additional

700:   edges, labeled $\alpha\cdot p$ with $\alpha\in\F$, from vertices

701:   of degree $n$ to vertices of degree $pn$. For all $x\in B_n$ and

702:   $y\in B_{pn}$, there is an edge labeled $\langle x^p|y\rangle\cdot

703:   p$ from $x$ to $y$.

704: 

705:   Edges labeled $0x$ are naturally omitted, and edges labeled $1x$

706:   are simply written $x$.

707: \end{defn}

708: 

709: There is some analogy between this definition and that of a Cayley

710: graph --- this topic will be developed in Section~\ref{sec:parabolic}.

711: The generators (in the Cayley sense) are simply chosen to be the

712: $\ad(x)$ with $x$ running through $B_1$, a basis of $G/[G,G]$.

713: 

714: A presentation for the $\Lie$ can also be read off its Lie graph. For

715: every $n$, consider the set $\mathcal W$ of all words of length $n$

716: over $B_1$. For a path $\pi$ in the Lie graph, define its weight as

717: the product of the labels on its edges. Each $w\in\mathcal W$ defines

718: an element of $\Lie_n$, by summing the weights of all paths labeled

719: $w$ in the Lie graph. Let $\mathcal R_n$ be the set of all linear

720: dependence relations among these words. Then $\Lie$ admits a

721: presentation by generators and relations as

722: \[\Lie = \langle B_1|\,\mathcal R_1,\mathcal R_2,\dots\rangle.\]

723: 

724: Let us give a few examples of Lie graphs. First, if $G$ is abelian,

725: then its Lie graph has $\rank(G)$ vertices of weight $1$ and no other

726: vertices. If $G$ is the quaternion group $Q_8=\{\pm1,\pm i,\pm j,\pm

727: k\}$, then its Lie ring is an algebra over $\F[2]$, and the Lie graph

728: of $\Lie(Q_8)=\Lie_{\F[2]}(Q_8)$ is

729: \[\xymatrix{{i}\ar[dr]^{j}\\ & {-1}\\ {j}\ar[ur]^{i}}\]

730: 

731: \subsection{The infinite dihedral group} As another example, let $G$

732: be the infinite dihedral group $D_\infty=\langle

733: a,b|\,a^2,b^2\rangle$. Then $\gamma_n(G)=\langle

734: (ab)^{2^{n-1}}\rangle$ for all $n\ge2$, and its Lie ring is again a

735: Lie algebra over $\F[2]$, with Lie graph

736: \[\xymatrix{{a}\ar[dr]^{b}\\

737:   & {(ab)^2}\ar[r]^{a,b} & {(ab)^4}\ar[r]^{a,b} &

738:   {(ab)^8}\ar@{.>}[r]^{a,b} & {}\\

739:   {b}\ar[ur]_{a}}\]

740: 

741: Note that the lower $2$-central series of $G$ is different: we have

742: $G_{2^n}=G_{2^n+1}=\dots=G_{2^{n+1}-1}=\gamma_{n+1}(G)$, so the

743: Lie graph of $\Lie_{\F[2]}(G)$ is

744: \[\xymatrix{{a}\ar[dr]^{b}\\

745:   & {(ab)^2}\ar[rr]^{\cdot2} & & {(ab)^4}\ar[rrrr]^{\cdot2} & & & &

746:   {(ab)^8}\ar@{.>}[rr]^{\cdot2} & & {}\\

747:   {b}\ar[ur]_{a}}\]

748: 

749: \subsection{The free group}\label{subs:free}

750: Consider, as an example producing exponential growth, the free group

751: $F_r$ and its Lie algebra $\Lie$; this is a free Lie algebra of rank

752: $r$. Using Theorem~\ref{thm:quillen} and M\"obius inversion, we get

753: \[\dim_\Q(\gamma_n(F_r)/\gamma_{n+1}(F_r)\otimes\Q)=\#\{u\in\mathcal

754: M|\,\deg u=n\} =\frac1n\sum_{d|n}\mu_{n/d}r^d\precsim r^n,

755: \]

756: where $\mu$ is the M\"obius function; therefore $\gr(\overline{\Q

757:   F_r})\le\frac1{1-r\hbar}$.  Recall that

758: $\gr(F_r)=\frac{1+\hbar}{1-(2r-1)\hbar}$, so the group growth rate

759: can be strictly larger than the algebra growth rate in

760: Proposition~\ref{prop:gpgr}.

761: 

762: It is an altogether different story to find explicitly a basis of

763: $\Lie$. Pick a basis $X$ of $F_r$; its image in $\Lie_1\cong\Z^r$ is a

764: generating set of $\Lie$, still written $X$. A \emph{Hall set} is a

765: linearly ordered set of non-associative words $\mathcal M$ with

766: $X\subset\mathcal M$ and

767: \[[u,v]\in\mathcal M\text{ if and only if }u<v\in\mathcal M\text{ and

768: }(u\in X\text{ or }u=[p,q],q\ge v);\]

769: furthermore one requires $[u,v]<v$. Note that an order on the

770: non-associative words uniquely defines a corresponding Hall set.

771: 

772: There are many examples of Hall sets, and for each Hall set $\mathcal

773: M$ the set $\{u\in\mathcal M|\,|u|=r\}$ is a basis of the abelian

774: group $\gamma_n(F_r)/\gamma_{n+1}(F_r)$. For example, the \emph{Hall

775:   basis}~\cite{hall:liebasis} is the linearly ordered set $\mathcal M$

776: having as maximal elements $X$ in an arbitrary order, and such that

777: $u<v$ in $\mathcal M$ whenever $\deg(u)>\deg(v)$. It contains then all

778: $[x,y]$ with $x,y\in X$ and $x>y$; then all $[[u,v],w]$ whenever

779: $[u,v]<w\le v$ and $u,v,w\in\mathcal M$.

780: 

781: Another basis, more computationally efficient (it is a Lie algebra

782: equivalent of ``Gr\"obner bases''), is the ``Lyndon-Shirshov

783: basis''~\cites{shirshov:liebases,lothaire:mots,reutenauer:fla}. It is

784: defined as follows: order $X$ arbitrarily; on the free monoid $X^*$

785: put the lexicographical ordering: $u\le uv$, and $uxv<uyw$ for all

786: $u,v,w\in X^*$ and $x<y\in X$. A non-empty word $w\in X^*$ is a

787: \emph{Lyndon-Shirshov word} if for any non-trivial factorization

788: $w=uv$ we have $w<v$. If furthermore we insist that $v$ be

789: $<$-minimal, then $u$ and $v$ are again Lyndon-Shirshov words. For a

790: Lyndon-Shirshov word $w$, define its \emph{bracketing} $B(w)$

791: inductively as follows: if $w\in X$ then $B(w)=w$. If $w=uv$ with $v$

792: minimal then $B(w)=[B(u),B(v)]$.  Then $\{B(w)\}$ is a basis of

793: $\Lie$.

794: 

795: From our perspective, an optimal basis $B$ would consist only of

796: left-ordered commutators, and be prefix-closed, i.e.\ be such that

797: $[u,x]\in B$ implies $u\in B$; then indeed the Lie algebra structure

798: of an arbitrary Lie algebra would be determined $\ad(u)$ for all $u\in

799: B$, and therefore would be a tree in the case of a free Lie algebra.

800: Kukin announced in~\cite{kukin:liebases} a construction of such bases,

801: but his proof does not appear to be altogether

802: complete~\cite{blessenohl-l:liebases}, and the problem of construction

803: of a left-ordered basis seems to be considered open.

804: 

805: \subsection{The lamplighter group} As another example, consider the

806: ``lamplighter group'' $G=C_2\wr\Z$, with $a$ generating $C_2$ and

807: $t$ generating $\Z$.  Define the elements

808: \[a_n=\prod_{i=0}^{n-1}a^{(-1)^i\binom{n-1}it^i}

809: =at^{-1}a^{-(n-1)}t^{-1}\dots a^{(-1)^{n-1}}t^{n-1}\]

810: of $G$. Then its Lie algebra $\Lie_{\F[2]}(G)$ is as follows:

811: \[\xymatrix{{a}\ar[r]^{t} & {a_2}\ar[r]^{t} & {a_3}\ar[r]^{t} &

812:   {a_4}\ar[r]^{t} & {a_5}\ar[r]^{t} & {a_6}\ar[r]^{t} &

813:   {a_7}\ar[r]^{t} & {a_8}\ar[r]^{t} & {a_9}\ar@{.>}[r]^{t} & {}\\

814:   {t}\ar[ur]_{a}\ar[r]_{\cdot2} &

815:   {t^2}\ar[ur]_{a}\ar[rr]_{\cdot2} & &

816:   {t^4}\ar[ur]_{a}\ar[rrrr]_{\cdot2} & & & &

817:   {t^8}\ar[ur]_{a}\ar@{.>}[rr]_{\cdot2} & & {}}\]

818: Note that $\Lie_{\F[2]}(G)$ has bounded width, while $G$ has

819: exponential growth! This shows that in Proposition~\ref{prop:gpgr} the

820: group growth rate can be exponential while the algebra growth rate is

821: polynomial.

822: 

823: \subsection{The Nottingham group} As a final example, we give the Lie

824: graph of the Nottingham group's Lie

825: algebra~\cites{jennings:sgfps,camina:ng}. Recall that for odd prime $p$

826: the Nottingham group $J(p)$ is the group of all formal power series

827: \[\hbar+\sum_{i>1}a_i\hbar^i\in{\F}[[\hbar]],\]

828: with composition (i.e.\ substitution) as binary operation.

829: The lower central series is given by

830: \[J_n=\{\hbar+\sum_{i>\left\lceil\frac{np-1}{p-1}\right\rceil}a_i\hbar^i\},\]

831: and a basis of $\Lie$ is $\{f_i=\hbar(1+\hbar^i)\}_{i\ge1}$, where

832: $f_i$ has degree $\left\lfloor\frac{(p-1)i+1}{p}\right\rfloor$. As

833: basis of $J_1/J_2$, we take

834: $B_1=\{x=\hbar+\hbar^2+\hbar^3,y=\hbar+\hbar^3\}$. The commutations

835: are given by

836: \[[f_i,x]=(i-1)f_{i+1},\qquad

837: [f_i,y]=\begin{cases}-2f_{i+2}&\text{ if }i\equiv0\mod p\\

838:   -f_{i+2}&\text{ if }i\equiv1\mod p\\

839:   0&\text{ otherwise,}

840: \end{cases}.\]

841: This gives the Lie graph with ``diamond'' structure~\cite{caranti:nottingham}

842: {\tiny\[\xymatrix{

843:     {1} & {2} & {3} & {\cdots} & {p-2} & {p-1} & {p} & {p+1} & {p+2}\\

844:     {x}\ar[dr]^{-y} & & & & & & {f_{p+1}}\ar[dr]^{-y}\\

845:     & {f_3}\ar[r]^{2x} & {f_4}\ar@{.>}[rr] & & {f_{p-1}}\ar[r]^{-2x} &

846:     {f_p}\ar[ur]^{-x}\ar[dr]_{-2y} & &

847:     {f_{p+3}}\ar[r]^{2x} & {f_{p+4}}\ar@{.>}[r] & {}\\

848:     {y}\ar[ur]_{x} & & & & & & {f_{p+2}}\ar[ur]_{x}}\]}

849: 

850: %---------------------------------------------------------------

851: \subsection{The tree automorphism group's pro-$p$-Sylow

852:   $\aut_p(\Sigma^*)$}\label{subs:psylow} We start by considering a

853: typical example of branch group. Let $p$ be prime; write $p'=p-1$ for

854: notational simplicity. Let $\Sigma$ be the $p$-letter alphabet

855: $\{\mathsf1,\dots,\mathsf p\}$, and let $x_n$, for $n\in\N$, be the

856: $p$-cycle permuting the first $p$ branches at level $n+1$ in the tree

857: $\Sigma^*$. Therefore $x_0$ acts just below the root vertex, and

858: $x_{n+1}=\pair<x_n,1,\dots,1>$ for all $n$.

859: 

860: For all $n\in\N$ we define $G_n=\aut_p(\Sigma^*)$ as the group

861: generated by $\{x_0,\dots,x_{n-1}\}$, and $G=\langle

862: x_0,x_1,\dots\rangle$. Clearly $G=\injlim G_n$, while its closure is

863: $\overline G=\projlim G_n$.  Note that $G_n$ is a $p$-Sylow of

864: $\sym{p^n}$, and $\overline G$ is a pro-$p$-Sylow of

865: $\aut(\Sigma^*)$.

866: 

867: \begin{lem}

868:   $G=G\wr C_p$; therefore $G$ is a regular branch group over

869:   itself.

870: \end{lem}

871: \begin{proof}

872:   The subgroup $\langle x_1,x_2,\dots\rangle$ of $G$ is isomorphic to

873:   $G$ through $x_i\mapsto x_{i-1}$, and its $p$ conjugates under

874:   powers of $x_0$ commute, since they act on disjoint subtrees.

875: \end{proof}

876: 

877: Lev Kaloujnine described in~\cite{kaloujnine:struct} the lower central

878: series of $G_n$, using his notion of \emph{tableau}. Our purpose here

879: shall be to describe the Lie algebra of $G_n$ (and therefore $G$ and

880: $\overline G$) using our more geometric approach. Let us just mention

881: that in Kaloujnine's theory of tableaux his polynomials

882: $x_1^{e_1}\dots x_n^{e_n}$ correspond to our $\mathbb e_1\dots\mathbb

883: e_n(x_0)$.

884: 

885: \begin{lem}\label{lem:poisson}

886:   For $u,v\in G$ and $X,Y\in\{\7,\dots,\usp\}^n$ we have

887:   \[[X(u),Y(v)]\equiv(X_1+Y_1-\usp)\dots(X_n+Y_n-\usp)

888:   ([u,v])^{\prod_{i=1}^n(-1)^{p'-Y_i}\binom{X_i}{p'-Y_i}},

889:   \]

890:   modulo terms in $[[X(u),Y(v)],G]$.

891: \end{lem}

892: \begin{proof}

893:   The proof follows by induction, and we may suppose $n=1$ without

894:   loss of generality. Multiplying by terms in $[[X(u),Y(v)],G]$, we

895:   may assume $Y(v)$ by some element acting only on the last $Y_1$

896:   subtrees below the root vertex. Then

897:   \begin{align*}

898:     [X(u),Y(v)] &\equiv [\pair<u,\dots,u^{(-1)^{X_1}},1,\dots,1>,

899:     \pair<1,\dots,1,v,\dots,v^{(-1)^{Y_1}}>]\\

900:     &= \pair<[u,1],\dots,[u^{(-1)^{p'-Y_1}\binom{p}{p'-Y_1}},v],

901:     \dots,[u^{(-1)^{X_1}},v^{(-1)^{X_1}\binom{p}{X_1}}],\dots,[1,v]>\\

902:     &\equiv (X+Y-\usp)([u,v])^{(-1)^{p'-Y_1}\binom{p}{p'-Y_1}}.

903:   \end{align*}

904: \end{proof}

905: 

906: Note that in the Kaloujnine terminology there is a beautiful

907: description of $[X(u),Y(v)]$ in terms of Poisson brackets, due to

908: Vitaly Sushchansky, and due to appear in a forthcoming paper of his.

909: 

910: \begin{thm}\label{thm:psylow}

911:   Consider the following Lie graph: its vertices are the symbols $X$

912:   for all words $X\in\{\7,\dots,\usp\}^*$, including the empty

913:   word $\lambda$.  Their degrees are given by

914:   \[\deg X_1\dots X_n = 1 + \sum_{i=1}^n X_ip^{i-1}.\]

915:   For all $m>n\ge0$ and all choices of $X_i$, there is an arrow

916:   labeled $\7^n$ from $\usp^nX_{n+1}\dots X_m$ to

917:   $\7^n(X_{n+1}+\8)X_{n+2}\dots X_m$, and an arrow labeled $\7^m$ from

918:   $\usp^n$ to $\7^n\8\7^{m-n-1}$.

919:   

920:   Then the resulting graph is the Lie graph of $\Lie(G)$ and of

921:   $\Lie_{\F}(G)$.

922:   

923:   The subgraph spanned by all words of length up to $n-1$ is the Lie

924:   graph of $\Lie(G_n)$ and of $\Lie_{\F}(G_n)$.

925: \end{thm}

926: \begin{proof}

927:   We interpret $X$ in the Lie graph as $X(x_0)$ in $G$. The generator

928:   $x_n$ is then $\7^n(x_0)$. By Lemma~\ref{lem:poisson}, the adjoint

929:   operators $\ad(x_n)$ correspond to the arrows labeled $\7^n$. The

930:   arrows connect elements whose degree differ by $1$, so the degree of

931:   the element $X(x_0)$ is $\deg(X)$ as claimed.

932: 

933:   The power maps $g\mapsto g^p$ are all trivial on the elements

934:   $X(x_0)$, so the Lie algebra and restricted Lie algebra coincide.

935: 

936:   The elements $X(x_0)$ for $|X|\ge n$ belong to $\stab_G(n)$, and

937:   hence are trivial in $G_n$.

938: \end{proof}

939: 

940: \begin{figure}\label{fig:lie:psylow}

941:   \tiny\[\xymatrix@-2ex{

942:     {}\ar@{--}[d] &

943:     {}\ar@{--}[d] &

944:     {}\ar@{--}[d] &

945:     {}\ar@{--}[d] &

946:     {}\ar@{--}[d] &

947:     {}\ar@{--}[d] &

948:     {}\ar@{--}[d] &

949:     {}\ar@{--}[d] \\

950: %

951:     {\7^n}\ar@{--}[d]\ar[r]^{\lambda} &

952:     {\8\7^{n-1}}\ar@{--}[d]\ar@{.>}[r]^{\lambda} &

953:     {\usp\7^{n-1}}\ar@{--}[d]\ar[r]^{\7} &

954:     {\7\8\7^{n-2}}\ar@{--}[d]\ar[r]^{\lambda} &

955:     {\8\8\7^{n-2}}\ar@{--}[d]\ar@{.>}[r]^{\lambda} &

956:     {\usp\8\7^{n-2}}\ar@{--}[d]\ar@{.>}[r]^{\lambda} &

957:     {\usp^2\7^{n-2}}\ar@{--}[d]\ar[r]^{\7^2} &

958:     {\7^2\8\7^{n-3}}\ar@{--}[d]\ar@{.>}[r] & {} \\

959: %

960:     {\7^3}\ar[r]^{\lambda} & {\8\7\7}\ar@{.>}[r]^{\lambda} &

961:     {\usp\7\7}\ar[r]^{\7} & {\7\8\7}\ar[r]^{\lambda} &

962:     {\8\8\7}\ar@{.>}[r]^{\lambda} & {\usp\8\7}\ar@{.>}[r]^{\lambda} &

963:     {\usp\usp\7}\ar[r]^(0.4){\7^2} & {\7\7\8}\ar@{.>}[r] & {} \\

964: %

965:     {\7^2}\ar[r]^(0.3){\lambda} & {\8\7}\ar@{.>}[r]^{\lambda} &

966:     {\usp\7}\ar[r]^(0.25){\7} & {\7\8}\ar[r]^{\lambda} &

967:     {\8\8}\ar@{.>}[r]^{\lambda} & {\usp\8}\ar@{.>}[r]^{\lambda} &

968:     {\usp\usp}\ar@{.>}[uur]_(0.7){\7^n}\ar[ur]_{\7^3}\\

969: %

970:     {\7}\ar[r]^(0.45){\lambda} & {\8}\ar@{.>}[r]^{\lambda} &

971:     {\usp}\ar@{.>}[uuur]^{\7^n}\ar[uur]_(0.7){\7^3}\ar[ur]_{\7^2}\\

972: %

973:     {\lambda}\ar@{.>}[uuuur]^(0.6){\7^n}\ar[uuur]_(0.8){\7^3}

974:     \ar[uur]_(0.7){\7^2}\ar[ur]_{\7}}\]

975:   \caption{The beginning of the Lie graph of $\Lie(G)$ for $G$ the

976:     $p$-Sylow of $\aut(\Sigma^*)$}

977: \end{figure}

978: 

979: %---------------------------------------------------------------

980: \subsection{The group $\Gg$}\label{subs:Gg}

981: We give an explicit description of the Lie algebra of $\Gg$, and

982: compute its Hilbert-Poincar\'e series. These results were obtained

983: in~\cite{bartholdi-g:lie}, and partly before in~\cite{rozhkov:lcs}.

984: 

985: Set $x=[a,b]$. Then $\Gg$ is branch over $K=\langle x\rangle^\Gg$, and

986: $K/(K\times K)$ is cyclic of order $4$, generated by $x$.

987: 

988: \newcommand\sm[2]{{\{\begin{smallmatrix}#1\\#2\end{smallmatrix}\}}}

989: 

990: Extend the generating set of $\Gg$ to a formal alphabet

991: $S=\big\{a,b,c,d,\sm bc,\sm cd,\sm db\big\}$.  Define the

992: transformation $\sigma$ on words in $S^*$ by

993: \[\sigma(a)=a\sm bca,\quad\sigma(b)=d,\quad\sigma(c)=b,\quad\sigma(d)=c,\]

994: extended to subsets by $\sigma\sm xy=\sm{\sigma x}{\sigma y}$. Note

995: that for any fixed $g\in G$, all elements $h\in\stab_\Gg(1)$ such that

996: $\psi(h)=\pair<g,*>$ are obtained by picking a letter from each set in

997: $\sigma(g)$. This motivates the definition of $S$.

998: 

999: \begin{thm}\label{thm:G:struct}

1000:   Consider the following Lie graph: its vertices are the symbols

1001:   $X(x)$ and $X(x^2)$, for words $X\in\{\7,\8\}^*$. Their degrees are

1002:   given by

1003:   \begin{align*}

1004:     \deg X_1\dots X_n(x) = 1 + \sum_{i=1}^n X_i2^{i-1} + 2^n,\\

1005:     \deg X_1\dots X_n(x^2) = 1 + \sum_{i=1}^n X_i2^{i-1} + 2^{n+1}.\\

1006:   \end{align*}

1007:   There are four additional vertices: $a,b,d$ of degree $1$, and

1008:   $[a,d]$ of degree $2$.

1009:   

1010:   Define the arrows as follows: an arrow labeled $\sm xy$ or

1011:   ``$x,y$'' stands for two arrows, labeled $x$ and $y$, and the

1012:   arrows labeled $c$ are there to expose the symmetry of the graph

1013:   (indeed $c=bd$ is not in our chosen basis of $G/[G,G]$):

1014:   \[\xymatrix@R-3ex{

1015:     {a}\ar[r]^-{b,c} & {x} &         {a}\ar[r]^-{c,d} & {[a,d]}\\

1016:     {b}\ar[r]^{a} & {x} &            {d}\ar[r]^-{a} & {[a,d]}\\

1017:     {x}\ar[r]^-{a,b,c} & {x^2} &     {x}\ar[r]^-{c,d} & {\7(x)}\\

1018:     {[a,d]}\ar[r]^-{b,c} & {\7(x)} & {\7*}\ar[r]^-{a} & {\8*}\\

1019: %

1020:     {\8^n(x)}\ar[r]^{\sigma^n\sm cd} & {\7^{n+1}(x)} &

1021:     {\8^n(x)}\ar[r]^{\sigma^n\sm bd} & {\7^n(x^2)}\\

1022:     {\8^n\7*}\ar[r]^{\sigma^n\sm cd} & {{\7^n\8*}\makebox[0mm][l]{ if

1023:         $n\ge1$.}}}\]

1024:   

1025:   Then the resulting graph is the Lie graph of $\Lie(\Gg)$. A slight

1026:   modification gives the Lie graph of $\Lie_{\F[2]}(\Gg)$: the degree

1027:   of $X_1\dots X_n(x^2)$ is then $2\deg X_1\dots X_n(x)$; and the

1028:   $2$-mappings are given by

1029:   \begin{align*}

1030:     X(x)&\overset{\cdot2}\longrightarrow X(x^2),\\

1031:     \8^n(x^2)&\overset{\cdot2}\longrightarrow \8^{n+1}(x^2).\\

1032:   \end{align*}

1033: 

1034:   The subgraph spanned by $a,t$, the $X_1\dots X_i(x)$ for $i\le n-2$

1035:   and the $X_1\dots X_i(x^2)$ for $i\le n-4$ is the Lie graph

1036:   associated to the finite quotient $\Gg/\stab_\Gg(n)$.

1037: \end{thm}

1038: 

1039: Figure~\ref{fig:lie:Gg} describes as Lie graphs the top of the Lie

1040: algebras associated to $\Gg$. Note the infinite path, labeled by

1041: \[\sm cda\sigma(\sm cda)\sigma^2(\sm cda)\dots=\sm cda\sm bca\sm

1042: bca\sm bda\sm bca\sm bda\sm bca\sm cda\sm bca\dots;\]

1043: it is the same as the labeling of the parabolic space of $\Gg$ ---

1044: see Section~\ref{sec:parabolic} and~\cite{bartholdi-g:parabolic}.

1045: 

1046: \begin{figure}\label{fig:lie:Gg}

1047:   \tiny\[\xymatrix@d@-7ex@C+2em{

1048:     {b}\ar[dr]^{a} & & {x^2}\\

1049:     & {x}\ar[ur]^{a,b,c}\ar[dr]^{c,d}

1050:     & & & {\7(x^2)}\ar[r]^{a} & {\8(x^2)} & & & {\7\7(x^2)}\ar[r]^{a} &

1051:     {\8\7(x^2)}\ar[r]^{b,c} & {\7\8(x^2)}\ar[r]^{a} & {\8\8(x^2)} & &

1052:     & & & {\7\7\7(x^2)}\ar@{--}[r] &\\

1053:     {a}\ar[ur]^{b,c}\ar[dr]_{c,d}  & & {\7(x)}\ar[r]_{a} &

1054:     {\8(x)}\ar[dr]_{b,c}\ar[ur]^{c,d} & & & {\7\8(x)}\ar[r]_{a} &

1055:     {\8\8(x)}\ar[dr]_{b,d}\ar[ur]^{b,c} & & & & &

1056:     {\7\7\8(x)}\ar[r]_{a} & {\8\7\8(x)}\ar[r]_{b,c} &

1057:     {\7\8\8(x)}\ar[r]_{a} & {\8\8\8(x)}\ar[dr]_{c,d}\ar[ur]^{b,d}\\

1058:     & {[a,d]}\ar[ur]_{b,c} & & & {\7\7(x)}\ar[r]_{a} &

1059:     {\8\7(x)}\ar[ur]_{b,c} & & & {\7\7\7(x)}\ar[r]_{a} &

1060:     {\8\7\7(x)}\ar[r]_{b,c} & {\7\8\7(x)}\ar[r]_{a} &

1061:     {\8\8\7(x)}\ar[ur]_{b,d} & & & & & {\7\7\7\7(x)}\ar@{--}[r] &\\

1062:     {d}\ar[ur]_{a}\\

1063: %

1064:     {1}\ar@{.}[u]\ar@{.}[d] & {2} & {3} &

1065:     {4}\ar@{.}[uuu]\ar@{.}[dd] &  {5} & {6} & {7} &

1066:     {8}\ar@{.}[uuu]\ar@{.}[dd] & {9} & {10} & {11} &

1067:     {12}\ar@{.}[uu]\ar@{.}[dd] & {13} & {14} & {15} &

1068:     {16}\ar@{.}[uuu]\ar@{.}[dd] & {17}\\

1069: %

1070: %!    {b}\ar[dr]^{a}\\

1071: %!    & {x}\ar[dr]^{c,d}\ar[rr]^{\cdot2} & & {x^2}\ar[dr]^(0.4){c,d}

1072: %!    \ar@/^1pc/[rrrr]^/-1cm/{\cdot2}

1073: %!    & & {\7(x^2)} & & {\8(x^2)}\ar[dr]^(0.56){b,c}

1074: %!    \ar@/^1.5pc/[rrrrrrrr]^/-1cm/{\cdot2}

1075: %!    & & {\7\7(x^2)} & & {\8\7(x^2)} & & {\7\8(x^2)} & &

1076: %!    {\8\8(x^2)}\ar[dr]^{b,d}\ar@{--}[rr] & &\\

1077: %!    {a}\ar[ur]^{b,c}\ar[dr]_{c,d} & &

1078: %!    {\7(x)}\ar[r]_{a}\ar[urrr]^(0.4){\cdot2} &

1079: %!    {\8(x)}\ar[r]_{b,c}\ar[urrrr]^{\cdot2} &

1080: %!    {\7\7(x)}\ar[r]_{a}\ar[urrrrr]^{\cdot2} &

1081: %!    {\8\7(x)}\ar[r]_{b,c}\ar[urrrrrr]^{\cdot2} &

1082: %!    {\7\8(x)}\ar[r]_{a}\ar[urrrrrrr]^{\cdot2} &

1083: %!    {\8\8(x)}\ar[r]_{b,d}\ar[urrrrrrrr]^{\cdot2} &

1084: %!    {\7\7\7(x)}\ar[r]_{a} & {\8\7\7(x)}\ar[r]_{b,c} &

1085: %!    {\7\8\7(x)}\ar[r]_{a} & {\8\8\7(x)}\ar[r]_{b,d} &

1086: %!    {\7\7\8(x)}\ar[r]_{a} & {\8\7\8(x)}\ar[r]_{b,c} &

1087: %!    {\7\8\8(x)}\ar[r]_{a} & {\8\8\8(x)}\ar[r]_{c,d} &

1088: %!    {\7\7\7\7(x)}\ar@{--}[r] &\\

1089: %!    & {[a,d]}\ar[ur]_{b,c}\\

1090: %!    {d}\ar[ur]_{a}}\]

1091:     {b}\ar[dr]^{a}\\

1092:     & {x}\ar[dr]^{c,d}\ar[rr]^{\cdot2}

1093:     & & {x^2}\ar[dr]^{c,d}\ar@/^0.2pc/[rrrr]^{\cdot2}

1094:     & & & & {\8(x^2)}\ar[dr]^{b,c}\ar@/^0.3pc/[rrrrrrrr]^(0.4){\cdot2}

1095:     & & & & & & & & {\8\8(x^2)}\ar[dr]^{b,d}\ar@{--}[rr] & &\\

1096:     {a}\ar[ur]^{b,c}\ar[dr]_{c,d} & &

1097:     {\7(x)}\ar[r]^{a}\ar[drrr]_{\cdot2} &

1098:     {\8(x)}\ar[r]_{b,c}\ar@/^0.7pc/[urrrr]^{\cdot2} &

1099:     {\7\7(x)}\ar[r]^{a}\ar[ddrrrrr]_{\cdot2} &

1100:     {\8\7(x)}\ar[r]^{b,c}\ar[ddrrrrrr]_{\cdot2} &

1101:     {\7\8(x)}\ar[r]^{a}\ar[ddrrrrrrr]_{\cdot2} &

1102:     {\8\8(x)}\ar[r]_{b,d}\ar@/^1.2pc/[urrrrrrrr]^(0.4){\cdot2} &

1103:     {\7\7\7(x)}\ar[r]^{a}\ar@{--}[drrr] &

1104:     {\8\7\7(x)}\ar[r]^{b,c}\ar@{--}[drrr] &

1105:     {\7\8\7(x)}\ar[r]^{a}\ar@{--}[drrr] &

1106:     {\8\8\7(x)}\ar[r]^{b,d}\ar@{--}[drrr] &

1107:     {\7\7\8(x)}\ar[r]^{a}\ar@{--}[drrr] &

1108:     {\8\7\8(x)}\ar[r]^{b,c}\ar@{--}[drrr] &

1109:     {\7\8\8(x)}\ar[r]^{a}\ar@{--}[drrr] &

1110:     {\8\8\8(x)}\ar[r]_{c,d}\ar@{--}[urr] &

1111:     {\7\7\7\7(x)}\ar@{--}[r] &\\

1112:     & {[a,d]}\ar[ur]_{b,c} & & & & {\7(x^2)} & & & & & &

1113:     \ar@{--}[rrrrrr] & & & & & & \\

1114:     {d}\ar[ur]_{a} & & & & & & & & &

1115:     {\7\7(x^2)} & & {\8\7(x^2)} & & {\7\8(x^2)}}\]

1116:   \caption{The beginning of the Lie graphs of $\Lie_{\F[2]}(\Gg)$

1117:     (left) and $\Lie(\Gg)$ (right).}

1118: \end{figure}

1119: 

1120: For proof requires computation, given a term $N$ of a central series

1121: and a generator $s\in\{a,b,c,d\}$, of $[N,x]$ modulo $[N,G,G]$. We

1122: do slightly better in the following lemma --- this will be useful in

1123: Section~\ref{sec:nsGg} where we describe all normal subgroups of $G$.

1124: For that purpose we introduce a symbol $\6(x)=\7(x)\8(x)^{-1}$. We

1125: then have

1126: \[\7(x)=\pair<x,1>,\quad \8(x)=\pair<x,x^{-1}>,\quad \6(x)=\pair<1,x>.\]

1127: \begin{lem}\label{lem:commGg}

1128:   Assume $N$ is a normal subgroup containing the left-hand operand of

1129:   the commutators below. Then modulo $[N,G]'$ we have

1130:   \begin{xalignat*}{2}

1131:     [\7X,a]&=\8X     & [\8X,a]&=\8X^2\\

1132:     [\7X,b]&=\7[X,a] & [\8X,b]&=\7[X,a]+\6[X,c]\\

1133:     [\7X,c]&=\7[X,a] & [\8X,c]&=\7[X,a]+\6[X,d]\\

1134:     [\7X,d]&=1       & [\8X,d]&=\6[X,b]\\

1135:     [x,a]&=x^2       & [x^2,a]&=x^4=\8(x^2+\8x)\\

1136:     [x,b]&=x^2       & [x^2,b]&=\8(x^2+\8x)\\

1137:     [x,c]&=\7(x)+x^2 & [x^2,c]&=\7(x^2+\7x)+\8(x^2+\8x)\\

1138:     [x,d]&=\7(x)     & [x^2,d]&=\7(x^2+\7x)\\

1139:   \end{xalignat*}

1140: \end{lem}

1141: \begin{proof}

1142:   Direct computation, using the decompositions

1143:   $\psi(b)=(a,c)=\7(a)\cdot\6(c)$ etc. and linearizing.

1144: \end{proof}

1145: 

1146: \begin{proof}[Proof of Theorem~\ref{thm:G:struct}]

1147:   The proof proceeds by induction on length of words, or, what amounts

1148:   to the same, on depth in the lower central series.

1149: 

1150:   First, the assertion is checked ``manually'' up to degree $3$. The

1151:   details of the computations are the same as

1152:   in~\cite{bartholdi-g:lie}.

1153: 

1154:   We claim that for all words $X,Y$ with $\deg Y(x)>\deg X(x)$ we have

1155:   $Y(x)\in\langle X(x)\rangle^\Gg$, and similarly $Y(x^2)\in\langle

1156:   X(x^2)\rangle^\Gg$. The claim is verified by induction on $\deg X$.

1157:   

1158:   We then claim that for any non-empty word $X$, either

1159:   $\ad(a)X(*)=0$ (if $X$ starts by ``$\8$'') or

1160:   $\ad(v)X(*)=0$ for $v\in\{b,c,d\}$ (if $X$ starts by ``$\7$'').

1161:   Again this holds by induction.

1162:   

1163:   We then prove that the arrows are as described above; this follow

1164:   from Lemma~\ref{lem:commGg}. For instance,

1165:   \begin{align*}

1166:     \ad(\sigma^n\sm cd)\8^n\7* &= \begin{cases}

1167:       \big(\ad(\sigma^n\sm db)\8^{n-1}\7*,\ad(\sm a1)\8^{n-1}{\7*}\big)\\

1168:       \hfill=\7\ad(\sigma^{n-1}\sm cd)\8^{n-1}\7*=\7^n\8* & \text{ if }n\ge2,\\

1169:       (\ad(\sm bc)\7*,\ad(a)\7*) = \7\8* & \text{ if }n=1.

1170:     \end{cases}

1171:   \end{align*}

1172:   

1173:   Finally we check that the degrees of all basis elements are as

1174:   claimed. For that purpose, we first check that the degree of an

1175:   arrow's destination is always one more than the degree of its

1176:   source.

1177:   Then fix a word $X(*)$, and consider the largest $n$ such that

1178:   $X(*)$ belongs to $\gamma_n(\Gg)$. There is then an expression of

1179:   $X(*)$ as a product of $n$-place commutators on elements of

1180:   $\Gg\setminus[\Gg,\Gg]$, and therefore in the Lie graph there is a

1181:   family of paths starting at some element of $B_1$ and following

1182:   $n-1$ arrows to reach $X(*)$. This implies that the degree of $X(*)$

1183:   is $n$, as required.

1184: 

1185:   The modification giving the Lie graph of $\Lie_{\F[2]}(\Gg)$ is

1186:   justified by the fact that in $\Lie(\Gg)$ we always have $\deg

1187:   X(x^2)\le2\deg X(x)$, so the element $X(x^2)$ appears always last as

1188:   the image of $X(x)$ through the square map. The degrees are modified

1189:   accordingly. Now $X(x^2)=X\8(x^2)$, and $2\deg X\8(x)\ge 4\deg X(x)$,

1190:   with equality only when $X=\8^n$. This gives an additional square map

1191:   from $\8^n(x^2)$ to $\8^{n+1}(x^2)$, and requires no adjustment of

1192:   the degrees.

1193: \end{proof}

1194: 

1195: \begin{cor}\label{cor:g:rk}

1196:   Define the polynomials

1197:   \begin{align*}

1198:     Q_2&=-1-\hbar,\\

1199:     Q_3&=\hbar+\hbar^2+\hbar^3,\\

1200:     Q_n(\hbar)&=(1+\hbar)Q_{n-1}(\hbar^2)+\hbar+\hbar^2\text{ for }n\ge4.

1201:   \end{align*}

1202:   Then $Q_n$ is a polynomial of degree $2^{n-1}-1$, and the first

1203:   $2^{n-3}-1$ coefficients of $Q_n$ and $Q_{n+1}$ coincide.  The

1204:   term-wise limit $Q_\infty=\lim_{n\to\infty}Q_n$ therefore exists.

1205:   

1206:   The Hilbert-Poincar\'e series of $\Lie(\Gg/\stab_\Gg(n))$ is

1207:   $3\hbar+\hbar^2+\hbar Q_n$, and the Hilbert-Poincar\'e series of

1208:   $\Lie(\Gg)$ is $3\hbar+\hbar^2+\hbar Q_\infty$.

1209: 

1210:   The Hilbert-Poincar\'e series of $\Lie_{\F[2]}(\Gg)$ is

1211:   $3+\frac{2\hbar+\hbar^2}{1-\hbar^2}$.

1212: 

1213:   As a consequence, $\Gg/\stab_\Gg(n)$ is nilpotent of class

1214:   $2^{n-1}$, and $\Gg$ has finite width.

1215: \end{cor}

1216: 

1217: \begin{proof}

1218:   Consider the sequence of coefficients of $Q_n$. They are, in

1219:   condensed form,

1220:   \[1,2^{2^0},1^{2^0},2^{2^1},1^{2^1},\dots, 2^{n-4},1^{n-4},1^{n-2}.\]

1221:   The $i$th coefficient is $2$ if there are $X(x)$ and $Y(x^2)$ of

1222:   degree $i$ in $\Gg/\stab_\Gg(n)$, and is $1$ if there is only

1223:   $X(x)$. All conclusions follow from this remark.

1224: \end{proof}

1225: 

1226: %---------------------------------------------------------------

1227: \subsection{The group $\GS$}

1228: We now give an explicit description of the Lie algebra of $\GS$,

1229: and compute its Hilbert-Poincar\'e series.

1230: 

1231: Introduce the following sequence of integers:

1232: \[\alpha_1=1,\quad\alpha_2=2,

1233: \quad\alpha_n=2\alpha_{n-1}+\alpha_{n-2}\text{ for }n\ge3,\]

1234: and $\beta_n=\sum_{i=1}^n\alpha_i$. One has

1235: \begin{align*}

1236:   \alpha_n &= \frac1{2\sqrt2}\left((1+\sqrt2)^n-(1-\sqrt2)^n\right),\\

1237:   \beta_n &= \frac14\left((1+\sqrt2)^{n+1}+(1-\sqrt2)^{n+1}-2\right).

1238: \end{align*}

1239: The first few values are

1240: \[\begin{array}{c|cccccccc}

1241:   n & 1 & 2 & 3 & 4 & 5 & 6 & 7 & 8\\ \hline

1242:   \alpha_n & 1 & 2 & 5 & 12 & 29 & 70 & 169 & 398\\

1243:   \beta_n & 1 & 3 & 8 & 20 & 49 & 119 & 288 & 686

1244: \end{array}\]

1245: 

1246: \begin{thm}\label{thm:gamma:struct}

1247:   In $\GS$ write $c=[a,t]$ and $u=[a,c]=\9(t)$.  Consider the following

1248:   Lie graph: its vertices are the symbols $X_1\dots X_n(x)$ with

1249:   $X_i\in\{\7,\8,\9\}$ and $x\in\{c,u\}$.  Their degrees are given by

1250:   \begin{align*}

1251:     \deg X_1\dots X_n(c)&=1+\sum_{i=1}^nX_i\alpha_i+\alpha_{n+1},\\

1252:     \deg X_1\dots X_n(u)&=1+\sum_{i=1}^nX_i\alpha_i+2\alpha_{n+1}.

1253:   \end{align*}

1254:   There are two additional vertices, labeled $a$ and $t$, of degree

1255:   $1$.

1256: 

1257:   Define the arrows as follows:

1258:   \[\xymatrix@R-3ex{

1259:     {a}\tar[r]^{-t} & {c} & {c}\tar[r]^-{t} & {\7(c)}\\

1260:     {t}\aar[r]^{a} & {c} &  {c}\aar[r]^{a} & {u}\\

1261:     & {u}\tar[r]^{t} & {\8(c)}\\

1262: %

1263:     {\7*}\aar[r]^{a} & {\8*} & {\8*}\aar[r]^{a} & {\9*}\\

1264:     & {\9*}\tar[r]^{t} & {\7\#\makebox[0mm][l]{ whenever

1265:         $*\blue{\overset t\longrightarrow}\#$}}\\

1266:     {\9(c)}\tar[r]^{t} & {\8(u)} &  {\8(c)}\tar[r]^{-t} & {\7(u)}\\

1267:     {\8\7*}\tar[r]^{-t} & {\7\8*} & {\8\8*}\tar[r]^{-t} & {\7\9*}\\

1268:     {\9\7*}\tar[r]^{t} & {\8\8*} &  {\9\8*}\tar[r]^{t} & {\8\9*}}\]

1269:   (Note that these last $3$ lines can be replaced by the rules

1270:   ${\9*}\blue{\overset t\longrightarrow}\8\#$ and

1271:   ${\8*}\blue{\overset{-t}\longrightarrow}\7\#$ for all arrows

1272:   $*\red{\overset a\longrightarrow}\#$.)

1273: 

1274:   Then the resulting graph is the Lie graph of $\Lie(\GS)$. It is

1275:   also the Lie graph of $\Lie_{\F[3]}(\GS)$, with the only

1276:   non-trivial cube maps given by

1277:   \[\9^n(c)\overset{\cdot3}\longrightarrow\9^n\7\7(c),\qquad

1278:   \9^n(c)\overset{\cdot3}\longrightarrow\9^n\8(u).\]

1279:   

1280:   The subgraph spanned by $a,t$, the $X_1\dots X_i(c)$ for $i\le n-2$

1281:   and the $X_1\dots X_i(u)$ for $i\le n-3$ is the Lie graph associated

1282:   to the finite quotient $\GS/\stab_\GS(n)$.

1283: \end{thm}

1284: 

1285: \begin{figure}

1286:   \tiny\[\xymatrix@!@-2ex@C-3.4em{

1287:     {1} & {2} & {3} & {4} & {5}\ar@{--}[ddddd] & {6} & {7} & {8} & {9}

1288:     & {10}\ar@{--}[dddd] & {11} & {12} & {13} & {14} & {15}\ar@{--}[d] &

1289:     {16} & {17} & {18} & {19} & {20}\ar@{--}[dd] & {21}\\

1290: %

1291:     & & & & & & & & & & & & & & {\9\7\7(c)}\tar[dr] & &

1292:     {\7\9\7(c)}\aar[dr] & & {\9\9\7(c)}\tar[dr]\\

1293: %

1294:     & & & & & & & & & & & & & {\8\7\7(c)}\tar[dr]\aar[ur] & &

1295:     {\8\8\7(c)}\aar[dr]\tar[ur] & & {\8\9\7(c)}\aar[ur] & &

1296:     {\7\8\8(c)}\aar[dr] & & {}\\

1297: %

1298:     & & & & & & & & & & & & {\7\7\7(c)}\aar[ur] & &

1299:     {\7\8\7(c)}\aar[ur] & & {\9\8\7(c)}\tar[dr]\tar[ur] & &

1300:     {\8\7\8(c)}\tar[ur]\aar[dr] & & {\8\8\8(c)}\aar[ur]\tar[dr]\\

1301: %

1302:     & & & & & {\7\7(c)}\aar[dr] & & {\9\7(c)}\tar[dr]

1303:     & & {\7\9(c)}\aar[dr] & & {\9\9(c)}\tar[ur]\tar[dr]

1304:     & & & & & & {\7\7\8(c)}\aar[ur] & & {\9\7\8(c)}\tar[ur] & & {}\\

1305: %

1306:     {a}\tar[dr] & & {\7(c)}\aar[dr] & & {\9(c)}\tar[ur]\tar[dr] & &

1307:     {\8\7(c)}\aar[ur]\tar[dr] & & {\8\8(c)}\tar[ur]\aar[dr] & &

1308:     {\8\9(c)}\aar[ur] & & {\7\8(u)}\aar[dr] & & {\9\8(u)}\tar[dr] & &

1309:     {\9\9(u)}\tar[ur]\\

1310: %

1311:     & {c}\tar[ur]\aar[dr] & & {\8(c)}\aar[ur]\tar[dr] & &

1312:     {\8(u)}\aar[dr] & & {\7\8(c)}\aar[ur] & &

1313:     {\9\8(c)}\tar[ur]\tar[dr] & & {\8\7(u)}\tar[ur]\aar[dr] & &

1314:     {\8\8(u)}\tar[dr]\aar[ur] & & {\8\9(u)}\aar[ur]\\

1315: %

1316:     {t}\aar[ur] & & {u}\tar[ur] & & {\7(u)}\aar[ur] & &

1317:     {\9(u)}\tar[ur] & & & & {\7\7(u)}\aar[ur] & & {\9\7(u)}\tar[ur] &

1318:     & {\7\9(u)}\aar[ur]\\

1319: %

1320:     {2} & {1} & {2} & {1}\ar@{--}[uu] & {2} & {2} & {2} & {2} &

1321:     {1}\ar@{--}[uuu] & {2} & {2} & {2} & {3} & {2} & {4} & {2} & {3} &

1322:     {2} & {2} & {2} & {1}\ar@{--}[uuuuu]}\]

1323:   \caption{The beginning of the Lie graph of $\Lie(\GS)$. The

1324:     generator $\ad(t)$ is shown by plain/blue arrows, and the generator

1325:     $\ad(a)$ is shown by dotted/red arrows.}

1326: \end{figure}

1327: 

1328: \begin{proof}

1329:   We perform the computations in the completion of $\GS$, still

1330:   written $\GS$. With Lemma~\ref{lem:portraits} in mind, $\GS'$

1331:   is the subgroup generated by all $X(c)$ and $X(u)$, for

1332:   $X\in\{\7,\8,\9\}^*$.

1333:   

1334:   We claim inductively that if $X_i\ge Y_i$ at all positions $i$,

1335:   then $X(c)\in\langle Y(c)\rangle^\GS$, and similarly for $u$.

1336:   Therefore some terms may be neglected in the computations of

1337:   brackets.

1338:   

1339:   Now we compute $\ad(x)y$ for $x,y\in\{a,t,c,u\}$. Here $\equiv$

1340:   means some terms of greater degree have been neglected:

1341:   \begin{align*}

1342:     [a,\7*] &= \8*,\quad [a,\8*] = \9*,

1343:     \quad [a,\9*] = 1\text{ by definition,}\\

1344:     [t,\7*] &= [\pair<a,a^{-1},t>,\pair<*,1,1>]

1345:     =\pair<[a,*],1,1>=\7[a,*]\\

1346:     &\equiv[\pair<a^{-1},t,a>,\pair<*,1,1>]=-\7[a,*],

1347:     \text{ so }[t,\7*]=1,\\

1348:     [t,\8*] &= [\pair<a,a^{-1},t>,\pair<*,*^{-1},1>]\equiv-\7[a,*],\\

1349:     [t,\9*] &= [\pair<a,a^{-1},t>,\pair<*,*,*>]\equiv\8[a,*]+\7[t,*].\\

1350:   \end{align*}

1351:   All asserted arrows follow from these equations.

1352: 

1353:   Finally, we prove that the degrees of $X(c)$ and $X(u)$ are as

1354:   claimed, by remarking that $\deg c=3$ and $\deg u=4$, that $\deg

1355:   \ad(s)*\ge\deg(*)$ for $s=a,t$ and all words $*$ (so the claimed

1356:   degrees smaller of equal to their actual value), and that each word

1357:   of claimed degree $n$ appears only as $\ad(s)*$ for words $*$ of

1358:   degree at most $n-1$ (so the claimed degrees are greater or equal to

1359:   their actual value).

1360:   

1361:   The last point to check concerns the cube map; we skip the details.

1362: \end{proof}

1363: 

1364: \begin{cor}\label{cor:gamma:rk}

1365:   Define the following polynomials:

1366:   \begin{align*}

1367:     Q_1&=0,\\

1368:     Q_2&=\hbar+\hbar^2,\\

1369:     Q_3&=\hbar+\hbar^2+2\hbar^3+\hbar^4+\hbar^5,\\

1370:     Q_n&=(1+\hbar^{\alpha_n-\alpha_{n-1}})Q_{n-1}

1371:     +\hbar^{\alpha_{n-1}}(\hbar^{-\alpha_{n-2}}+1+\hbar^{\alpha_{n-2}})Q_{n-2}

1372:     \text{ for }n\ge3.

1373:   \end{align*}

1374:   Then $Q_n$ is a polynomial of degree $\alpha_n$, and the polynomials

1375:   $Q_n$ and $Q_{n+1}$ coincide on their first $2\alpha_{n-1}$ terms.

1376:   The coefficient-wise limit $Q_\infty=\lim_{n\to\infty}Q_n$ therefore

1377:   exists.

1378: 

1379:   The largest coefficient in $Q_{2n+1}$ is $2^n$, at position

1380:   $\frac12(\alpha_{2n+1}+1)$, so the coefficients of $Q_\infty$ are

1381:   unbounded. The integers $k$ such that $\hbar^k$ has coefficient $1$ in

1382:   $Q_\infty$ are precisely the $\beta_n+1$.

1383:   

1384:   The Hilbert-Poincar\'e series of $\Lie(\GS/\stab_\GS(n))$ is

1385:   $\hbar+Q_n$, and the Hilbert-Poincar\'e series of $\Lie(\GS)$ is

1386:   $\hbar+Q_\infty$. The same holds for the Lie algebra

1387:   $\Lie_{\F[3]}(\GS/\stab_\GS(n))$ and $\Lie_{\F[3]}(\GS)$.

1388: 

1389:   As a consequence, $\GS/\stab_\GS(n)$ is nilpotent of class

1390:   $\alpha_n$, and $\GS$ does not have finite width.

1391: \end{cor}

1392: \begin{proof}

1393:   Define polynomials

1394:   \[R_n=\sum_{w\in\{\7,\8,\9\}^n}\hbar^{\deg

1395:     w(c)}+\sum_{w\in\{\7,\8,\9\}^{n-1}}\hbar^{\deg w(u)}+\hbar.\]

1396:   Then one checks directly that the polynomials $R_n$ satisfy the same

1397:   initial values and recurrence relation as $Q_n$, hence are

1398:   equal. All convergence properties also follow from the definition of

1399:   $R_n$.

1400:   

1401:   The words of degree $\frac12(\alpha_{2n+1}+1)$ are

1402:   $(\7\8)^{n-1}\7(c)$, $(\7\8)^{n-2}\7\9(u)$, and all the words that

1403:   can be obtained from these by iterating the substitutions

1404:   $\7\7\8\mapsto\8\9\7$, $\8\7\8\mapsto\9\9\7$, $\7\7\9\mapsto\8\9\8$,

1405:   $\8\7\9\mapsto\9\9\8$ along with $\7\8\mapsto\9\7$ and

1406:   $\7\9\mapsto\9\8$ at the beginning of the word. This gives $2^n$

1407:   words in total, half of the form $X(c)$ and half $X(u)$.

1408: 

1409:   There is a unique word of degree $\beta_n+1$, and that is $\8^n(c)$.

1410:   

1411:   Note that these last two claims have a simple interpretation: there

1412:   are $2^{n-1}$ ways of writing

1413:   $\frac12(\alpha_{2n+1})-1-\alpha_{n+1}$ in base $\alpha$ using only

1414:   the digits $0,1,2$; there is a unique way of writing $\beta_n$ in

1415:   base $\alpha$ using these digits.

1416: \end{proof}

1417: 

1418: We note as an immediate consequence that

1419: \[[\GS:\gamma_{\beta_n+1}(\GS)]=3^{\frac12(3^n+1)},\]

1420: so that the asymptotic growth of

1421: $\ell_n=\dim(\gamma_n(\GS)/\gamma_{n+1}(\GS))$ is polynomial of degree

1422: $d=\log3/\log(1+\sqrt2)-1$:

1423: \begin{cor}\label{cor:gsgk}

1424:   The Gelfand-Kirillov dimension of $\Lie(\GS)$ is

1425:   $\log3/\log(1+\sqrt2)-1$.

1426: \end{cor}

1427: 

1428: We then deduce:

1429: \begin{cor}

1430:   The growth of $\GS$ is at least

1431:   $e^{n^{\frac{\log3}{\log(1+\sqrt2)+\log3}}}\cong e^{n^{0.554}}$.

1432: \end{cor}

1433: \begin{proof}

1434:   Apply Proposition~\ref{prop:gpgr} to the series $\sum n^d \hbar^n$,

1435:   which is comparable to the Hilbert-Poincar\'e series of $\Lie(\GS)$.

1436: \end{proof}

1437: 

1438: Turning to the derived series, we may also improve the general result

1439: $\GS^{(k)}<\gamma_{2^k}(\GS)$ to the following

1440: \begin{thm} For all $k\in\N$ we have

1441:   \[\GS^{(k)}<\gamma_{\alpha_{k+1}}(\GS).\]

1442: \end{thm}

1443: \begin{proof}

1444:   Clearly true for $k=0,1$; then a direct consequence of

1445:   $\GS^{(k)}=\gamma_5(\GS)^{\times3^{k-2}}$ (obtained by Ana

1446:   Vieira in~\cite{vieira:lcs}) and

1447:   $\gamma_{\alpha_j}(\GS)^{\times3}<\gamma_{\alpha_{j+1}}(\GS)$

1448:   for $j=3,\dots,k$.

1449: \end{proof}

1450: 

1451: %---------------------------------------------------------------

1452: \subsection{The group $\FG$}

1453: We now give an explicit description of the Lie algebra of $\FG$,

1454: and compute its Hilbert-Poincar\'e series.

1455: 

1456: \begin{thm}\label{thm:delta:struct}

1457:   In $\FG$ write $c=[a,t]$ and $u=[a,c]\equiv\9(at)$. For words

1458:   $X=X_1\dots X_n$ with $X_i\in\{\7,\8,\9\}$ define symbols

1459:   $\overline{X_1\dots X_n}(c)$ (representing elements of

1460:   $\FG$) by

1461:   \begin{align*}

1462:     \overline{\mathbb i\7}(c) &= \mathbb i\7(c)/\mathbb i(u),\\

1463:     \overline{\mathbb i\9^{m+1}\8^n}(c) & =\mathbb i\big(

1464:     \overline{\9^{m+1}\8^n}(c)\cdot\7\8^m\7^n(u)^{(-1)^n}\big),\\

1465:     \overline{\mathbb iX}(c) &= \mathbb i\overline X(c)\text{ for all

1466:       other }X.\\

1467:   \end{align*}

1468:   

1469:   Consider the following Lie graph: its vertices are the symbols

1470:   $\overline{X}(c)$ and $X(u)$. Their degrees are given by

1471:   \begin{align*}

1472:     \deg\overline{X_1\dots X_n}(c)&=1+\sum_{i=1}^nX_i3^{i-1}+\frac12(3^n+1),\\

1473:     \deg X_1\dots X_n(u)&=1+\sum_{i=1}^nX_i3^{i-1}+(3^n+1).

1474:   \end{align*}

1475:   There are two additional vertices, labeled $a$ and $t$, of degree

1476:   $1$.

1477: 

1478:   Define the arrows as follows, for all $n\ge1$:

1479:   \[\xymatrix@R-3ex{

1480:     {a}\tar[r]^{-t} & {c} &      {t}\aar[r]^{a} & {c}\\

1481:     {c}\tar[r]^-{-t} & {\7(c)} & {c}\tar[r]^{a} & {u}\\

1482:     {u}\tar[r]^-{-t} & {\8(c)} & {\overline{\9^n}(c)}\ar[r]^-{-t} &

1483:     {\overline{\7^{n+1}}(c)}\\

1484: %

1485:     {\7*}\aar[r]^{a} & {\8*} &         {\8*}\aar[r]^{a} & {\9*}\\

1486:     {\9^n\7*}\tar[r]^{t} & {\7^n\8*} & {\9^n\8*}\tar[r]^{t} & {\7^n\9*}\\

1487:     & {\makebox[5mm][r]{$\overline{X_1\dots X_n}(c)$}}

1488:     \tar[r]^{-(-1)^{\sum X_i}t} &

1489:     {\makebox[5mm][l]{$(X_1-1)\dots(X_n-1)(u)$}}}\]

1490:   

1491:   Then the resulting graph is the Lie graph of $\Lie(\FG)$.

1492: 

1493:   The subgraph spanned by $a,t$, the $\overline{X_1\dots X_i}(c)$ for

1494:   $i\le n-2$ and the $X_1\dots X_i(u)$ for $i\le n-3$ is the Lie graph

1495:   associated to the finite quotient $\FG/\stab_\FG(n)$.

1496: \end{thm}  

1497: 

1498: \begin{figure}

1499:   \tiny\[\xymatrix@!@-2ex@C-3.4em{

1500:     {1} & {2} & {3} & {4} & {5}\ar@{--}[dd] &

1501:     {6} & {7} & {8} & {9} & {10}\ar@{--}[d] &

1502:     {11} & {12} & {13} & {14} & {15}\ar@{--}[dd] &

1503:     {16} & {17} & {18} & {19} & {20}\ar@{--}[d] & {21}\\

1504: %

1505:     & & & & & {\overline{\7\7}(c)}\aar[dr] & &

1506:     {\overline{\9\7}(c)}\tar[dr] & &

1507:     {\overline{\8\8}(c)}\aar[dr]\tar[dddr] & &

1508:     {\overline{\7\9}(c)}\aar[dr] & &

1509:     {\overline{\9\9}(c)}\tar[dr]\tar[dddr] & &

1510:     {\overline{\8\7\7}(c)}\aar[dr] & & {\overline{\7\8\7}(c)}\aar[dr]

1511:     & & {\overline{\9\8\7}(c)}\tar[dr]\\

1512: %

1513:     {a}\tar[dr] & & {\overline{\7}(c)}\aar[dr] & &

1514:     {\overline{\9}(c)}\tar[dr]\tar[ur] & &

1515:     {\overline{\8\7}(c)}\aar[ur] & & {\overline{\7\8}(c)}\aar[ur] & &

1516:     {\overline{\9\8}(c)}\tar[dr]\tar[ur] & &

1517:     {\overline{\8\9}(c)}\aar[ur]\tar[dr] & &

1518:     {\overline{\7\7\7}(c)}\aar[ur] & & {\overline{\9\7\7}(c)}\tar[ur]

1519:     & & {\overline{\8\8\7}(c)}\aar[ur] & & {}\\

1520: %

1521:     & {c}\tar[ur]\aar[dr] & & {\overline{\8}(c)}\aar[ur]\tar[dr] & &

1522:     {\8(u)}\aar[dr] & & & & & & {\8\7(u)}\aar[dr] & &

1523:     {\7\8(u)}\aar[dr] & & {\9\8(u)}\tar[dr] & & {\8\9(u)}\aar[dr]\\

1524: %

1525:     {t}\aar[ur] & & {u}\tar[ur] & & {\7(u)}\aar[ur] & & {\9(u)} & & &

1526:     & {\7\7(u)}\aar[ur] & & {\9\7(u)}\tar[ur] & & {\8\8(u)}\aar[ur] &

1527:     & {\7\9(u)}\aar[ur] & & {\9\9(u)}\\

1528: %

1529:     {2} & {1} & {2} & {1}\ar@{--}[uu] & {2} & {2} & {2} & {1} & {1} &

1530:     {1}\ar@{--}[uuuu] & {2} & {2} & {2} & {2} & {2} & {2} & {2} & {2} &

1531:     {2} & {1} & {1}}\]

1532:   \caption{The beginning of the Lie graph of $\Lie(\FG)$. The

1533:     generator $\ad(t)$ is shown by plain/blue arrows, and the generator

1534:     $\ad(a)$ is shown by dotted/red arrows.}

1535: \end{figure}

1536: 

1537: \begin{proof}

1538:   The proof is similar to that of Theorems~\ref{thm:G:struct}

1539:   and~\ref{thm:gamma:struct}, but a bit more tricky. Again we perform

1540:   the computations in the completion of $\FG$, still written

1541:   $\FG$. Again $\FG'$ is the subgroup generated by all

1542:   $\overline X(c)$ and $X(u)$, for $X\in\{\7,\8,\9\}^*$.

1543:   

1544:   We claim inductively that if $X_i\ge Y_i$ at all positions $i$,

1545:   then $X(c)\in\langle Y(c)\rangle^\FG$, and similarly for $u$.

1546:   Therefore some terms may be neglected in the computations of

1547:   brackets.

1548:   

1549:   Now we compute $\ad(x)y$ for $x,y\in\{a,t,c,u\}$.  Here $\equiv$

1550:   means some terms of greater degree have been neglected:

1551:   \begin{align*}

1552:     [a,\7*] &= \8*,\quad [a,\8*] = \9*,\quad [a,\9*] = 1\text{ by

1553:       definition,}\\

1554:     [t,\7*] &\equiv [\pair<1,t,a>,\pair<*,1,1>]=1,\\

1555:     [t,\8*] &= [\pair<a,1,t>,\pair<*,*^{-1},1>]=\7[a,*]\\

1556:     &\equiv [\pair<1,t,a>,\pair<*,*^{-1},1>]\equiv -\7[t,*]\\

1557:     [t,\9*] &= [\pair<a,1,t>,\pair<*,*,*>]\equiv \7[a,*]+\7[t,*]+\8[t,*].\\

1558:   \end{align*}

1559:   Note that in the last line the ``negligible'' term $\8[t,*]$ has been

1560:   kept; this is necessary since sometimes the $\7[t,*]$ term cancels

1561:   out.

1562: 

1563:   Now we check each of the asserted arrows against the

1564:   relations described above. First the ``$a$'' arrows are clearly as

1565:   described, and so are the ``$t$'' arrows on $X(u)$; for instance,

1566:   \begin{align*}

1567:     \ad(t){\9^n\8*}(u) &= \7\ad(a){\9^{n-1}\8*}(u) + \7\ad(t){2^{n-1}\8*}(u)

1568:     + \8\ad(t){\9^{n-1}\8*}(u)\\

1569:     &\equiv \7^n\big(\ad(a){\8*}(u) + \ad(t){\8*}(u)\big) \equiv {\7^n\9*}(u),

1570:   \end{align*}

1571:   which holds by induction on the length of $*$. Next, the ``$t$''

1572:   arrows on $\overline X(c)$ agree; for instance,

1573:   \begin{align*}

1574:     \ad(t)\overline{\9\8^n}(c) &= \7\ad(a)\8^n(c) + \7\ad(t)\8^n(c)

1575:     + \8\ad(t)\8^n(c)\\

1576:     &= \7\9\8^{n-1}(c) + (-1)^n\cdot\7^{n+1}(u) + (-1)^n\cdot\8\7^n(u)\big)\\

1577:     &= \overline{\7\9\8^{n-1}}(c) + (-1)^n\cdot\8\7^n(u)

1578:     \text{ by induction on }n,\\\

1579:     \ad(t)\overline{\9^n}(c) &=

1580:     \ad(t)\9\big(\overline{\9^{n-1}}(c)\cdot\7\8^{n-2}(u)\big)\\

1581:     &\equiv \7\8^{n-1}(u) + \7\big(-\overline{\7^n}(c)-\8^{n-1}(u)\big) +

1582:     \8\big(-\overline{\7^n}(c)-\8^{n-1}(u)\big)\\

1583:     &\equiv-\overline{\7^{n+1}}(c) - \8^n(u).

1584:   \end{align*}

1585:   All other cases are similar. Note how the calculation for

1586:   $\overline{\9\8^n}(c)$ explains the definition of $\overline X(c)$:

1587:   both $\7\9\8^{n-1}(c)$ and $\7^{n+1}(u)$ have degree smaller than

1588:   $d=\deg\overline{\9\8^n}(c)$ in $\Lie(\FG)$, but they are linearly

1589:   dependent in $\gamma_{d-1}(\FG)/\gamma_d(\FG)$.

1590: 

1591:   Finally, we prove that the degrees of $X(c)$ and $X(u)$ are as

1592:   claimed, by remarking that $\deg c=3$ and $\deg u=4$, that $\deg

1593:   \ad(s)*\ge\deg(*)$ for $s=a,t$ and all words $*$ (so the claimed

1594:   degrees smaller of equal to their actual value), and that each word

1595:   of claimed degree $n$ appears only as $\ad(s)*$ for words $*$ of

1596:   degree at most $n-1$ (so the claimed degrees are greater or equal to

1597:   their actual value).

1598: \end{proof}

1599: 

1600: \begin{cor}\label{cor:delta:rk}

1601:   Define the integers $\alpha_n=\frac12(5\cdot3^{n-2}+1)$, and the

1602:   polynomials

1603:   \begin{align*}

1604:     Q_2&=1,\\

1605:     Q_3&=1+2\hbar+\hbar^2+\hbar^3+\hbar^4+\hbar^5+\hbar^6,\\

1606:     Q_n(\hbar)&=(1+\hbar+\hbar^2)Q_{n-1}(\hbar^3)+\hbar+

1607:     \hbar^{\alpha_n-2}\text{ for }n\ge4.

1608:   \end{align*}

1609:   Then $Q_n$ is a polynomial of degree $\alpha_n-2$, and the first

1610:   $3^{n-2}+1$ coefficients of $Q_n$ and $Q_{n+1}$ coincide.  The

1611:   term-wise limit $Q_\infty=\lim_{n\to\infty}Q_n$ therefore exists.

1612:   

1613:   The Hilbert-Poincar\'e series of $\Lie(\FG/\stab_\FG(n))$ is

1614:   $2\hbar+\hbar^2Q_n$, and the Hilbert-Poincar\'e series of $\Lie(\FG)$ is

1615:   $2\hbar+\hbar^2Q_\infty$.

1616: 

1617:   As a consequence, $\FG/\stab_\FG(n)$ is nilpotent of class

1618:   $\alpha_n$, and $\FG$ has finite width.

1619: \end{cor}

1620: 

1621: \begin{proof}

1622:   Consider the sequence of coefficients of $2\hbar+\hbar^2Q_n$. They

1623:   are, in condensed form,

1624:   \[2,1,2^{3^0},1^{3^0},2^{3^1},1^{3^1},\dots,

1625:   2^{3^{n-3}},1^{3^{n-3}},1^{\frac12(3^{n-1}+1)}.\] The $i$th

1626:   coefficient is $2$ if there are $\overline X(c)$ and $Y(u)$ of

1627:   degree $i$ in $\FG/\stab_\FG(n)$, and is $1$ if there is only

1628:   $\overline X(c)$. All conclusions follow from this remark.

1629: \end{proof}

1630: 

1631: In quite the same way as for $\GS$, we may improve the general

1632: result $\FG^{(k)}<\gamma_{2^k}(\FG)$:

1633: \begin{thm}

1634:   The derived series of $\FG$ satisfies $\FG'=\gamma_2(\FG)$

1635:   and $\FG^{(k)}=\gamma_5(\FG)^{\times3^{k-2}}$ for $k\ge2$. We

1636:   have for all $k\in\N$

1637:   \[\FG^{(k)}<\gamma_{2+3^{k-1}}(\FG).\]

1638: \end{thm}

1639: \begin{proof}

1640:   It is a general fact for a $2$-generated group $\FG$ that

1641:   $\FG''<\gamma_5(\FG)$. Since $[c,\7(c)]\equiv \7(u)^{-1}$ and

1642:   $[c,u]\equiv \9(c)^{-1}$ (modulo $\gamma_6(\FG)$), we have

1643:   $[\gamma_2(\FG),\gamma_3(\FG)]=\gamma_5(\FG)$ and therefore

1644:   $\FG''=\gamma_5(\FG)$.

1645:   

1646:   Next, $\gamma_5(\FG)=\gamma_3(\FG)^{\times3}\cdot\9(c)$, so

1647:   $\FG^{(3)}=[\gamma_3(\FG),c]^{\times3}=\gamma_5^{\times3}$,

1648:   and the claimed formula holds for all $\FG^{(k)}$ by induction.

1649:   Finally

1650:   $\gamma_{2+3^{j-2}}(\FG)^{\times3}<\gamma_{2+3^{j-1}}(\FG)$

1651:   for all $j=3,\dots,k$.

1652: \end{proof}

1653: 

1654: We omit altogether the proof of the following two results, since it is

1655: completely analogous to that of Theorem~\ref{thm:delta:struct}.

1656: \begin{thm}

1657:   Keep the notations of Theorem~\ref{thm:delta:struct}. Define now

1658:   furthermore symbols $\overline{X_1\dots X_n}(u)$ (representing

1659:   elements of $\FG$) by

1660:   \begin{align*}

1661:     \overline{\9^n}(u)&=\9^n(u)\cdot \9^{n-1}\7(c)\cdot \9^{n-2}\7\8(c)\cdots

1662:     \9\7\8^{n-2}(c),\\

1663:     \text{and }\overline X(u) &= X(u)\text{ for all other }X.\\

1664:   \end{align*}

1665:   

1666:   Consider the following Lie graph: its vertices are the symbols

1667:   $\overline{X}(c)$ and $\overline X(u)$. Their degrees are given by

1668:   \begin{align*}

1669:     \deg\overline{X_1\dots X_n}(c)&=1+\sum_{i=1}^nX_i3^{i-1}+\frac12(3^n+1),\\

1670:     \deg \9^n(u)&=3^{n+1},\\

1671:     \deg X_1\dots X_n(u)&=\max\{1+\sum_{i=1}^nX_i3^{i-1}+(3^n+1),

1672:     \frac12(9-3^n)+3\sum_{i=1}^nX_i3^{i-1}\}.

1673:   \end{align*}

1674:   There are two additional vertices, labeled $a$ and $t$, of degree

1675:   $1$.

1676: 

1677:   Define the arrows as follows, for all $n\ge1$:

1678:   \[\xymatrix@R-3ex{

1679:     {a}\tar[r]^{-t} & {c} &      {t}\aar[r]^{a} & {c} & \\

1680:     {c}\tar[r]^-{-t} & {\7(c)} & {c}\tar[r]^{a} & {u} & \\

1681:     {u}\tar[r]^-{-t} & {\8(c)} & {\overline{\9^n}(c)}\ar[r]^{-t} &

1682:     {\overline{\7^{n+1}}(c)}\\

1683: %

1684:     {\7*}\aar[r]^{a} & {\8*} &         {\8*}\aar[r]^{a} & {\9*}\\

1685:     {\9^n\7*}\tar[r]^{t} & {\7^n\8*} & {\9^n\8*}\tar[r]^{t} & {\7^n\9*}\\

1686:     & {\makebox[5mm][r]{$\overline{X_1\dots X_n}(c)$}}

1687:     \tar[r]^{-(-1)^{\sum X_i}t} &

1688:     {\makebox[5mm][l]{$(X_1-1)\dots(X_n-1)(u)$}}\\

1689:     {c}\ar[r]^{\cdot3} & {\overline{\7\7}(c)} &

1690:     {\overline{2^n}(u)}\ar[r]^{\cdot3} & {\overline{2^{n+1}}(u)}\\

1691:     {*\7(c)}\ar[r]^{\cdot3} & {*\9(u)\makebox[0mm][l]{ if

1692:         $3\deg*\7(c)=\deg*\9(u)$}}}\]

1693:   

1694:   Then the resulting graph is the Lie graph of $\Lie_{\F[3]}(\FG)$.

1695:   

1696:   The subgraph spanned by $a,t$, the $\overline{X_1\dots X_i}(c)$ for

1697:   $i\le n-2$ and the $X_1\dots X_i(u)$ for $i\le n-3$ is the Lie graph

1698:   of the Lie algebra $\Lie_{\F[3]}(\FG/\stab_\FG(n))$.

1699: 

1700:   As a consequence, the dimension series of $\FG/\stab_\FG(n)$

1701:   has length $3^{n-1}$ (the degree of $\overline{2^n}(u)$), and

1702:   $\FG$ has finite width.

1703: \end{thm}

1704: 

1705: We have then from Proposition~\ref{prop:grlgr}

1706: \begin{cor}

1707:   The growth of $\FG$ is at least $e^{n^{\frac12}}$.

1708: \end{cor}

1709: 

1710: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1711: \section{Parabolic Space}\label{sec:parabolic}

1712: In the natural action of a branch group $G$ on the tree $\Sigma^*$,

1713: consider a ``parabolic subgroup'' $P$, the stabilizer of an infinite

1714: ray in $\Sigma^*$. (The terminology comes from geometry, where a

1715: parabolic subgroup is the stabilizer of a point on the boundary of an

1716: appropriate $G$-space). Such a parabolic subgroup may be defined

1717: directly as follows: let

1718: $\omega=\omega_1\omega_2\dots\in\Sigma^\infty$ be an infinite

1719: sequence. Set $P_0^\omega=G$ and inductively

1720: \[P_n^\omega = \psi^{-1}(G\times\dots\times P_{n-1}^\omega\dots\times G),\]

1721: with the `$P_{n-1}^\omega$' in position $\omega_n$. Set

1722: $P^\omega=\bigcap_{n\ge0}P_n^\omega$.

1723: 

1724: In the natural tree action~\eqref{eq:action} of $G$ on $\Sigma^*$ or

1725: on $\Sigma^\infty$ its boundary, $P_n^\omega$ is the stabilizer of the

1726: point $\omega_1\dots\omega_n$, and $P^\omega$ is the stabilizer of the

1727: infinite sequence $\omega$.

1728: 

1729: The following facts easily follow from the definitions:

1730: \begin{lem}

1731:   $\bigcap_{\omega\in\Sigma^\infty}P^\omega = 1$. The index of

1732:   $P_n^\omega$ in $G$ is $d^n$, and that of $P^\omega$ is infinite.

1733: \end{lem}

1734: 

1735: \begin{defn}

1736:   Let $G$ be a branch group. A \emdef{parabolic space} for $G$ is a

1737:   homogeneous space $G/P$, where $P$ is a parabolic subgroup.

1738: \end{defn}

1739: 

1740: Suppose now that $G$ is finitely generated by a set $S$.

1741: \begin{prop}[\cite{bartholdi-g:spectrum}]\label{prop:polygrowth}

1742:   Suppose that the length $|\cdot|$ on the branch group $G$ satisfies

1743:   the following condition: there are constants $\lambda,\mu$ such that

1744:   for all $g\in\stab_G(1)$, writing $\psi(g)=(g_1,\dots,g_d)$, one has

1745:   $|g_i|<\lambda|g|+\mu$.

1746: 

1747:   Then all parabolic spaces of $G$ have polynomial growth of degree at

1748:   most $\log_{1/\lambda}(d)$.

1749: \end{prop}

1750: 

1751: \begin{thm}\label{thm:growth}

1752:   Let $G$ be a finitely generated branch group. Then there exists a

1753:   constant $C$ such that for any $x_0\in G$ we have

1754:   \[\frac{C\gr(G/P,x_0P)}{1-\hbar}\ge \frac{\gr\Lie(G)}{1-\hbar}.\]

1755: \end{thm}

1756: \begin{proof}

1757:   Assume $G$ acts on a $d$-regular tree, and write as before $d'=d-1$.

1758:   The proof relies on an identification of the Lie action on group

1759:   elements and the natural action on tree levels. We first claim that

1760:   for any $u\in K$ and $W\in\{\6,\dots,\mathbb d'\}^*$

1761:   \[\deg W(u)\ge \deg(\7^{|W|}(u)) + d_{G/P}(\7^{|W|},W),\]

1762:   where $d(W,X)$ is the length of a minimal word moving $W$ to $X$ in

1763:   the tree $\Sigma^*$.

1764:   

1765:   Therefore the growth of $\Lie(G)$ and $G/P$ may be compared just by

1766:   considering the degrees of elements of the form $\7^n(u)$ for some

1767:   fixed $u\in K$; indeed the other $W(u)$ will contribute a smaller

1768:   growth to the Lie growth series than the corresponding vertices to

1769:   the parabolic growth series, and the $N$ finitely many values $u$

1770:   may take in a branch portrait description will be taken care of by

1771:   the constant $C$.

1772: 

1773:   Now there is a constant $\ell\in\N$ such that $\7^{\ell+m}(u)$ has

1774:   greater degree than $(\mathbb d')^m(u)$ for all $m\in\N$. Indeed

1775:   there exists $k\in K$ and $\ell\in\N$ such that

1776:   $[k,u]=\7^\ell(u)$, and then

1777:   $[\7^mk,{\mathbb d'}^m(u)]=\7^{\ell+m}(u)$, proving the claim.

1778: 

1779:   We may now take $C=\ell N$. The Lie growth series is the sum over

1780:   all $n\in\N$ and coset representatives $u\in T$ of the

1781:   power series counting the growth of $W(u)$ over words $W$ of length

1782:   $n$. There are $N$ choices for $u$, and for given $u$ at most $\ell$

1783:   of these power series overlap.

1784: %

1785: %  More precisely, we shall show that for a generic $P$, if one sets

1786: %  $b_n=\deg\gr(G/P_n)$, then

1787: %  \[\frac1{1-\hbar}\Big(C\sum_{n\ge0}\hbar^{b_0+\dots+b_{n-1}}\gr(G/P_n) -

1788: %  \gr\Lie(G)\Big)\]

1789: %  is a positive series. Indeed for each $\gamma_n(G)$ consider all the

1790: %  branch portraits of its elements.  These portraits will be arbitrary

1791: %  from some level on, but subject to constraints in their top levels.

1792: %  It is therefore possible to express $\gamma_n(G)$ as $K^{\times

1793: %    d^m}\cdot T$ for some finite set $T$.  Suppose first $n_0$ chosen

1794: %  so that $T$ is maximal. The bracket operations with Lie elements of

1795: %  weight $1$ replace $T$ by a subset $T'$, etc., up to some point when

1796: %  $\gamma_{n_1}$ does not contain $K^{\times d^m}$. This, however,

1797: %  happens when $T$ becomes very small.

1798: %  

1799: %  Now $T$ is a set of branch portraits above level $m$. It injects in

1800: %  bijection with $\Bbbk^C\Sigma^m$, where $C$ is the index of

1801: %  $K^{\times d}$ in $K$. This last space is isomorphic to the

1802: %  $G$-space $\Bbbk^C[G/P_m]$, so the growth series of $\Lie(G)$

1803: %  between $n_0$ and $n_1$ is dominated by the growth series of

1804: %  $\Bbbk^G[G/P_m]$, which is precisely $C\gr(G/P_m)$.

1805: \end{proof}

1806: 

1807: Note that this result is valid even if the action on the rooted tree

1808: is not cyclic, i.e.\ even if in the decomposition map $G\to G\wr A$

1809: the finite group $A$ is not cyclic. If $A$ is not nilpotent, then the

1810: Lie algebra $\Lie$ is no longer isomorphic to $G$, so the best we can

1811: hope for is an inequality bounding the growth of $\Lie$ by that of $G/P$.

1812: 

1813: There are examples of groups of exponential growth, whose Lie algebra

1814: has subexponential growth --- see for example~\cite{bartholdi:zp2-1}.

1815: 

1816: %Let us suppose now that $A$ is nilpotent; this implies that

1817: %$\bigcup_{n\ge0}\gamma_n(G)=\{1\}$.

1818: %\begin{conj}

1819: %  Consider the square

1820: %  \[\xymatrix{{G}\ar@{-}[r]^{\ge}\ar@{-}[d]_{?} &

1821: %    {\overline{\F G}}\ar@{-}[d]^{\sum=\prod}\\

1822: %    {G/P}\ar@{-}[r]_{\ge} & {\Lie_{\F}(G)}}\]

1823: %  where the relations between the growth series are indicated: the top

1824: %  inequality is Proposition~\ref{prop:gpgr}, the bottom one is

1825: %  Theorem~\ref{thm:growth} and the right relation is

1826: %  Proposition~\ref{prop:sumprod}. Is there a general relation on the right

1827: %  side?

1828: %

1829: %  Such a relation, conjecturally, is

1830: %  \[\gr(G)\cong\prod_{n=1}^\infty\left(\frac1{1-\hbar^n}\right)^{p_nd},\]

1831: %  where $d$ is the Hausdorff dimension of $G$

1832: %  (see~\cite{barnea-s:hausdorff}) and $\sum p_n\hbar^n=\gr(G/P,*)$.

1833: %\end{conj}

1834: % 

1835: %One of the outstanding questions in the subject is the existence of

1836: %contracting branch groups of exponential growth. Since (under mild

1837: %conditions, like those of Proposition~\ref{prop:polygrowth}) the

1838: %parabolic space of a branch group has polynomial growth, its Lie

1839: %algebra also has polynomial growth, so its graded group ring has

1840: %subexponential growth.  A branch group of exponential growth with a

1841: %subexponentially-growing graded group ring would be a peculiar

1842: %phenomenon.

1843: %

1844: %Note that Said Sidki and John Wilson constructed

1845: %in~\cite{sidki-w:freebranch} branch groups containing free subgroups

1846: %--- and therefore of exponential growth. These groups are not

1847: %contracting, however.

1848: 

1849: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1850: \section{Normal Subgroups}\label{sec:nsGg}

1851: Using the notion of branch portrait, it is not too difficult to

1852: determine the exact structure of normal subgroups in a branch group.

1853: Consider a $p$-group $G$ and its $p$-Lie algebra $\Lie$ over $\F$.

1854: Normal subgroups of $G$ correspond to ideals of $\Lie$, just as

1855: subgroups of $G$ correspond to subalgebras of $\Lie$; and the index of

1856: $H<G$ is $p^{\dim \Lie/\mathcal M}$, where the subgroup $H$

1857: corresponds to the subalgebra $\mathcal M$. This correspondence is not

1858: exact, and we shall neither use it nor make it explicit; however it

1859: serves as a motivation for relating subgroup growth and the study of

1860: Lie algebras. In all cases, sufficient knowledge of $\Lie$, as well as

1861: its finiteness of width, allow an explicit description of the normal

1862: subgroup lattice of $G$.

1863: 

1864: We focus on the first and most important example, $\Gg$, for which we

1865: obtain an explicit answer. The computations presented here clearly

1866: extend, \emph{mutatis mutandis}, to any regular branch group.

1867: 

1868: Set $\mathcal W=\{\7,\8\}^*$, and order words $X\in\mathcal W$ by

1869: ``reverse shortlex'': the rank of $X_1\dots X_n$ is

1870: \[\#X_1\dots X_n = 1 + \sum_{i=1}^n X_i2^{i-1} + 2^n.\]

1871: (Note that $\#X=\deg X(x)$ according to the definition in

1872: Subsection~\ref{subs:Gg}.) We write $<$ the order induced by rank.

1873: 

1874: \begin{thm}\label{thm:normal:Gg}

1875:   The non-trivial normal subgroups of $\Gg$ are as follows:

1876:   \begin{itemize}

1877:   \item there are respectively $1,7,7,1$ subgroups of index $1,2,4,8$

1878:     corresponding to the lifts to $\Gg$ of subgroups of

1879:     $\Gg/[\Gg,\Gg]=C_2^{\times3}$;

1880:   \item there are $12$ other subgroups of $\Gg$ not contained in $K$:

1881:     six of index $8$, namely $\langle[a,c],d^ab\rangle^\Gg$, $\langle

1882:     c\rangle^\Gg$, $\langle x,c^ad\rangle^\Gg$, $\langle

1883:     b\rangle^\Gg$, $\langle[a,d],b^ac\rangle^\Gg$, and

1884:     $\langle d,x^2\rangle^\Gg$; four of index $16$, namely

1885:     $\langle[a,c]\rangle^\Gg$, $\langle[a,d],x^2\rangle^\Gg$,

1886:     $\langle d\rangle^\Gg$, and $\langle[a,d],x^2d\rangle^\Gg$;

1887:     and two of index $32$, namely $\langle[a,d]x^2\rangle^\Gg$ and

1888:     $\langle[a,d]\rangle^\Gg$;

1889:   \item all normal subgroups $N\triangleleft\Gg$ contained in $K$ are

1890:     of the form

1891:     \begin{equation}\label{eq:defW}

1892:       W(A;B_1,\dots,B_m;C):=

1893:       \langle A(x)B_1(x^2)\dots B_m(x^2),C(x^2)\rangle^\Gg,

1894:     \end{equation}

1895:     for words $A,B_i,C\in\mathcal W$. Assume the functions

1896:     $M(A,\{B_i\},C)$ and $S(A,\{B_i\},C)$, with values in $\mathcal

1897:     W$, defined in the proof. Then there is a unique description of

1898:     $N$ in the form~\ref{eq:defW} satisfying $B_1<B_2<\dots<B_m\le

1899:     S(A,\{B_i\},C)<C\le M(A,\{B_i\})$.

1900:     

1901:     The index of $N$ is $2^{\#A+\#S(A,\{B_i\},C)}$.

1902: 

1903:     The groups can furthermore be subdivided in three types:

1904:     \begin{description}

1905:     \item[$\rmI$] $C\le\7^{|A|}$ and $A\le\7^{|C|}\8\7$. Then all $B_i$

1906:       are optional, i.e.\ there are $2^m$ groups with these $A$ and

1907:       $C$, obtained by choosing any subset of the $B_i$'s;

1908:     \item[$\rmII$] $C>\7^{|A|}$ and $C\le\7^{|A|+1}$. Then $A=B_1\8$ and

1909:       all other $B_i$'s are optional;

1910:     \item[$\rmIII$] $A=\7^n$ and some $B_i=\7^{n-1}$. Then in fact an

1911:       alternate description exists, obtained by suppressing $A$ and

1912:       $B_i$ from the description.

1913:     \end{description}

1914:   \end{itemize}

1915: \end{thm}

1916: Note that we have only described finite-index subgroups of $\Gg$.

1917: Since $\Gg$ is just-infinite, all its non-trivial normal subgroups

1918: have finite index.

1919: 

1920: We represent the top of the lattice in Figure~\ref{fig:latticeG},

1921: containing all normal subgroups of index at most $2^{13}$ (there are

1922: never more than $7$ subgroups of given lesser index).

1923: 

1924: The first few subgroups of $K$ are described in Table~\ref{table:ns},

1925: sorted by their index in $\Gg$, and identified by their type in

1926: $\{(\rmI),(\rmII),(\rmIII)\}$. We write $\lambda$ for the empty

1927: sequence. An argument $[B_i]$ means that term is optional, and

1928: therefore stands for two groups, one with that term and one without.

1929: 

1930: Among the remarkable subgroups are: the subgroup $K^{\times

1931:   2^n}=\langle \7^n(x)\rangle^\Gg$, written $K_n$

1932: in~\cite{bartholdi-g:parabolic}; the subgroup $K^{\times

1933:   2^n}\mho_2(K)^{\times 2^{n-1}}=\langle \7^n(x),\7^{n-1}(x^2)\rangle$,

1934: written $N_n$ in~\cite{bartholdi-g:parabolic}; and $\stab_G(n)=\langle

1935: \7^{n-3}(\8(x)x^2),\7^{n-2}(x^2)\rangle$.

1936: 

1937: The lattice of normal subgroups of $\Gg$ is described in

1938: Figure~\ref{fig:latticeG}. Even though I do not understand completely

1939: the lattice's structure, some remarks can be made: the lattice has a

1940: fractal appearance; all its nodes have $1$ or $3$ descendants, and $1$

1941: or $3$ ascendants. Large portions of it have a grid-like structure.

1942: This can be explained by the construction $N\rightsquigarrow N\times

1943: N$ of normal subgroups, lending the lattice some self-similarity.

1944: 

1945: \begin{table}

1946:   \begin{center}

1947:     \begin{tabular}{r|c|lll}

1948:     Index  & Count & Description\\ \hline

1949:     $2^4$&    $1$&      $W(\lambda;;\lambda)_\rmI=K$\\

1950:     $2^5$&    $1$&      $W(\7;;\lambda)_\rmI$\\

1951:     $2^6$&    $3$&      $W(\8;;\lambda)_\rmI$& $W(\7;[\lambda];\7)_\rmI$\\

1952:     $2^7$&    $3$&      $W(\7\7;;\lambda)_\rmI$& $W(\8;[\lambda];\7)_\rmI$\\

1953:     $2^8$&   $5$&      $W(\8\7;;\lambda)_\rmI$& $W(\7\7;;\7)_\rmI$\\

1954:     &&&$W(\8;\lambda,[\7];\8)_\rmII$& $W(\infty;\lambda,\7;\8)_\rmIII$\\

1955:     $2^9$&   $5$&      $W(\8\7;;\7)_\rmI$& $W(\7\7;[\7];\8)_\rmI$&

1956:     $W(\8;\lambda,[\8];\7\7)_\rmII$\\

1957:     $2^{10}$&   $7$&      $W(\7\8;;\7)_\rmI$& $W(\8\7;[\7];\8)_\rmI$&

1958:     $W(\7\7;[\7],[\8];\7\7)_\rmI$\\

1959:     $2^{11}$&  $5$&      $W(\8\8;;\7)_\rmI$& $W(\7\8;[\7];\8)_\rmI$&

1960:     $W(\8\7;[\8];\7\7)_\rmI$\\

1961:     $2^{12}$&  $7$&      $W(\7\7\7;;\7)_\rmI$& $W(\8\8;[\7];\8)_\rmI$&

1962:     $W(\7\8;[\7],[\8];\7\7)_\rmI$\\

1963:     $2^{13}$&  $7$&      $W(\8\7\7;;\7)_\rmI$& $W(\7\7\7;[\7];\8)_\rmI$&

1964:     $W(\8\8;[\8];\7\7)_\rmI$\\

1965:     &&&&$W(\7\8;\7,[\7\7];\8\7)_\rmII$\\

1966:     $2^{14}$&$13$&    $W(\7\8\7;;\7)_\rmI$& $W(\8\7\7;[\7];\8)_\rmI$&

1967:     $W(\7\7\7;;\7\7)_\rmI$\\

1968:     &&&$W(\8\8;\8,[\7\7];\8\7)_\rmII$& $W(\7\8;\7,[\7\7],[\8\7];\7\8)_\rmII$\\

1969:     &&&$W(\infty;\8,0\7;\8\7)_\rmIII$& $W(\infty;\7,[\8],0\7;\7\8)_\rmIII$\\

1970:     $2^{15}$&$9$&     $W(\7\8\7;;\8)_\rmI$& $W(\8\7\7;;\7\7)_\rmI$&

1971:     $W(\7\7\7;[\7\7];\8\7)_\rmI$\\

1972:     &&$W(\8\8;\8,[\8\7];\7\8)_\rmII$& $W(\7\8;\7,[\7\8];\8\8)_\rmII$&

1973:     $W(\infty;\8,1\7;\7\8)_\rmIII$\\

1974:     $2^{16}$&$13$&    $W(\7\8\7;;\7\7)_\rmI$& $W(\8\7\7;[\7\7];\8\7)_\rmI$&

1975:     $W(\7\7\7;[\7\7],[\8\7];\7\8)_\rmI$\\

1976:     &&&$W(\8\8;\8,[\7\8];\8\8)_\rmII$& $W(\7\8;\7,[\7\8],[\8\8];\7\7\7)_\rmII$\\

1977:     $2^{17}$&$11$&    $W(\8\8\7;;\7\7)_\rmI$& $W(\7\8\7;[\7\7];\8\7)_\rmI$&

1978:     $W(\8\7\7;[\8\7];\7\8)_\rmI$\\

1979:     && $W(\7\7\7;[\7\7],[\7\8];\8\8)_\rmI$&& $W(\8\8;\8,[\8\8];\7\7\7)_\rmII$\\

1980:     $2^{18}$&$19$&    $W(\7\7\8;;\7\7)_\rmI$& $W(\8\8\7;[\7\7];\8\7)_\rmI$&

1981:     $W(\7\8\7;[\7\7],[\8\7];\7\8)_\rmI$\\

1982:     && $W(\8\7\7;[\8\7],[\7\8];\8\8)_\rmI$&

1983:     \multicolumn{2}{l}{$W(\7\7\7;[\7\7],[\7\8],[\8\8];\7\7\7)_\rmI$}\\

1984:     \hline

1985:     \end{tabular}

1986:   \end{center}

1987:   \caption{Normal subgroups of index up to $2^{18}$ in $\Gg$,

1988:     contained in $K$}

1989:   \label{table:ns}

1990: \end{table}

1991: 

1992: \begin{proof}[Proof of Theorem~\ref{thm:normal:Gg}]

1993:   The first two assertions are checked directly as follows. Let

1994:   $\mathcal F$ be the set of finite-index subgroups of $\Gg$ not in

1995:   $K$. Consider the finite quotient $Q=\Gg/\stab_6(\Gg)$, and the

1996:   preimage $P$ of $\Gg$ defined as

1997:   \[P = \big\langle a,b,c,d\big|\,a^2,b^2,c^2,d^2,bcd,

1998:   \sigma^i(ad)^4,\sigma^i(adacac)^4\quad(i=0\dots5)\big\rangle.\]

1999:   Clearly the image of $\mathcal F$ in $Q$ is at most as large as

2000:   $\mathcal F$, and the preimage of $\mathcal F$ in $P$ is at least as

2001:   large as $\mathcal F$. Now we use the algorithms in

2002:   \textsc{Gap}~\cite{gap4:manual} computing the top of the lattice of

2003:   normal subgroups for finite groups ($Q$) and finitely presented

2004:   groups ($P$). The number of subgroups not contained in $K$ agree in

2005:   $P$ and $Q$, so give the structure of the lattice not below $K$ in

2006:   $\Gg$.

2007:   

2008:   Let now $N$ be a normal subgroup of $\Gg$, contained in $K$. If $N$

2009:   is non-trivial, then it has finite

2010:   index~\cite{grigorchuk:jibg}*{Corollary to Proposition~9}. It is

2011:   easy to see that $N$ contains $C(x^2)$ and $D(x)$ for some words

2012:   $C,D$, using for instance the congruence

2013:   property~\cite{grigorchuk:jibg}*{Proposition~10}; therefore the

2014:   generators of $N$ may be chosen as

2015:   \[\{A_1(x)\cdots A_n(x)B_1(x^2)\cdots B_m(x^2),

2016:   A'_1(x)\cdots A'_{n'}(x)B'_1(x^2)\cdots B'_{m'}(x^2),\dots,C(x^2),D(x)\},\]

2017:   with $A_i^{(j)}<D$ and $B_i^{(j)}<C$ for all $i,j$.

2018:   

2019:   Taking the commutators of these generators with the appropriate

2020:   generator among $\{a,b,c,d\}$, we shift the ranks of the $A$-terms

2021:   up by $1$, and multiplying a generator by another we may get rid of

2022:   all generators except $C(x^2)$ and the one with $A_1$ of smallest

2023:   rank.

2024:   

2025:   We therefore consider all subgroups $W(A;B_1,\dots,B_m;C)$, and seek

2026:   conditions on $A$, $\{B_i\}$ and $C$ so that to each normal subgroup

2027:   in $K$ there corresponds a unique expression of the form

2028:   $W(A;B_1,\dots,B_m;C)$.

2029: 

2030:   Let first $C$ be minimal such that $C(x^2)\in N$; then take $A$

2031:   minimal such that for some $B_1<\dots<B_m<C$ we have

2032:   $A(x)B_1(x^2)\cdots B_m(x^2)\in N$ . Take also $B'_1$ minimal such

2033:   that $B'_1(x^2)\cdots B'_{m'}(x^2)\in N$ for some $B'_i$.

2034:   

2035:   Define the functions $M,S:\mathcal W\times2^{\mathcal

2036:     W}\times\mathcal W\to\mathcal W$ as follows ($M$ stands for

2037:   ``monomial'' and $S$ stands for ``squares''): Consider

2038:   $A(x)B_1(x^2)\dots B_m(x^2)$ as an element of $\Lie_{\F[2]}(\Gg)$,

2039:   truncated at degree $C$.  Successive commutations with generators

2040:   $s\in\{a,b,c,d\}$, \textbf{according the the rules of

2041:     Lemma~\ref{lem:commGg}}, give rise to other elements of

2042:   $\Lie_{\F[2]}(\Gg)$. We stress that we use the complete computations

2043:   of commutators, and not just those in the filtered Lie algebra.

2044:   Define $M(A,\{B_i\})$ as the minimal word $D$ such that $D(x^2)$

2045:   that arises in this process; if no such word occurs,

2046:   $M(A,\{B_i\},C)=C$. Define $S(A,\{B_i\})$ as the minimal $B'_{m'}$

2047:   such that $B'_1(x^2)\cdots B'_{m'}(x^2)$ occurs in this process; if

2048:   no such product occurs, $S(A,\{B_i\},C)=C-1$.

2049:   

2050:   Now, since $M(A,\{B_i\},C)(x^2)\in N$, we necessarily have $C\le

2051:   M(A,\{B_i\})$. Also, all $B_i$ of degree at least $B'_{m'}$ can be

2052:   replaced by terms of lower degree $B'_1,\dots,B'_{m-1}$. This proves

2053:   the claimed inequalities. Conversely, if there existed another

2054:   description $A(x)\tilde B_1(x^2)\dots\tilde B_m(x^2)\in N$ for

2055:   another choice of $\tilde B$'s, then by dividing we would obtain a

2056:   product of $B_i(x^2)$ in $N$, contradicting $B_m<S(A,\{B_i\},C)$.

2057:   The data $(A;B_1,\dots,B_m;C)$ subjected to the Theorem's

2058:   constraints therefore bijectively correspond to $N$'s.

2059: 

2060:   The index of $N$ can be computed in $\Lie_{\F[2]}(\Gg)$. Seeing

2061:   elements of $N$ as inside $\Lie$, a vector-space complement of $N$

2062:   is spanned by all $\tilde A(x)$ of rank less than $A$, and all

2063:   $\tilde B(x^2)$ of rank less than $S(A,\{B_i\},C)$.

2064: 

2065:   We consider finally three cases: first assume $C\le\7^{|A|}$ and

2066:   $|B_1|\ge|A|-1$. Then $C(x^2)$ gives $\7^{|C|+1}(x^2)\7^{|C|+2}(x)$ by

2067:   commutation with $\sigma^{|A|}(d)$, which itself gives

2068:   $\7^{|C|}\8\7(x)$ by commutation with $a$, so we may suppose

2069:   $A\le\7^{|C|}1\7$. Various $B_i$'s can be added, giving the

2070:   description $(\rmI)$.

2071:   

2072:   Now assume $C>\7^{|A|}$. Then since $A(x)$ would produce

2073:   $\7^{|A|}(x^2)$ by commutation with an appropriate conjugate of

2074:   $\sigma^{|A|}(b)$, we must have $A=B_1\8$ so that the same

2075:   commutation vanishes, giving the description $(\rmII)$.

2076: 

2077:   Finally assume we have $C\le\7^{|A|}$ and $|B_1|<|A|-1$. Then

2078:   necessarily $A=\7^n$; and taking appropriate commutations we see that

2079:   the normal subgroup under consideration contains

2080:   $\7^n(x)\7^{n-1}(x^2)$. We may then replace the generator

2081:   $A(x)B_1(x^2)\dots B_m(x^2)$ by $\7^{n-1}(x^2)B_1(x^2)\dots

2082:   B_m(x^2)$, and obtain the description $(\rmIII)$.

2083: \end{proof}

2084: 

2085: \begin{cor}

2086:   Let $N$ be a normal subgroup of $\Gg$. Then $N/[N,\Gg]$ is an

2087:   elementary $2$-group of rank $1$ or $2$, unless it is $N=\Gg$ (of

2088:   rank $3$).

2089: \end{cor}

2090: 

2091: \begin{cor}

2092:   Every normal subgroup of $G$ is characteristic.

2093: \end{cor}

2094: \begin{proof}

2095:   The automorphism group of $\Gg$ is determined

2096:   in~\cite{bartholdi-s:at}: it also acts on the binary tree, and is

2097:   \[\aut\Gg = \langle G, \8^j\7[a,d]\text{ for all }j\in\N\rangle.\]

2098:   It then follows that $[K,\aut\Gg]=\langle\7(x),x^2\rangle^\Gg$ is a

2099:   strict subgroup of $K$; and hence $[N,\aut\Gg]<N$ for any normal

2100:   subgroup that is generated by expressions in $W(x)$ and $W(x^2)$ for

2101:   words $W\in\{\7,\8\}^*$. The theorem asserts that all normal

2102:   subgroups of $\Gg$ below $K$ have this form; it then suffices to

2103:   check, for instance using the algorithms in

2104:   \textsc{Gap}~\cite{gap4:manual}, that the finitely many normal

2105:   subgroups of $\Gg$ not in $K$ are characteristic.

2106: \end{proof}

2107: 

2108: \begin{cor}\label{cor:asympns}

2109:   The number $b_n$ of normal subgroups of $\Gg$ of index $2^n$ starts

2110:   as follows, and is asymptotically $n^{\log_2(3)}$. More precisely,

2111:   we have $\liminf b_n/n^{\log_2(3)}=5^{-\log_2(3)}\approx0.078$ and

2112:   $\limsup b_n/n^{\log_2(3)}=\frac29\approx0.222$.

2113:   \[\begin{array}{c|cccccccccccc}

2114:     \text{index }2^n & 2^0 & 2^1 & 2^2 & 2^3 & 2^4 & 2^5 & 2^6 & 2^7 &

2115:     2^8 & 2^9 & 2^{10} & 2^{11}\\

2116:     |\{N\triangleleft\Gg\}| &

2117:     1 & 7 & 7 & 7 & 5 & 3 & 3 & 3 & 5 & 5 & 7 & 5\\ \hline

2118:     & \rule{0em}{3ex} 2^{12} & 2^{13} & 2^{14} & 2^{15} & 2^{16} &

2119:     2^{17} & 2^{18} & 2^{19} & 2^{20} & 2^{21} & 2^{22} & 2^{23}\\

2120:     & 7 & 7 & 13 & 9 & 13 & 11 & 19 & 11 & 13 & 11 & 19 & 15\\ \hline

2121:     & \rule{0em}{3ex} 2^{24} & 2^{25} & 2^{26} & 2^{27} & 2^{28} &

2122:     2^{29} & 2^{30} & 2^{31} & 2^{32} & 2^{33} & 2^{34}\\

2123:     & 25 & 21 & 37 & 23 & 31 & 23 & 37 & 25 & 37 & 31 & 55

2124:   \end{array}\]

2125: \end{cor}

2126: \begin{proof}

2127:   The number of subgroups of index $2^n$ behaves in a somewhat erratic

2128:   way, but is greater when $n$ is of the form $2^k+2$, so that there

2129:   is a maximal number of choices for $A$ and $C$, and is smaller when

2130:   $n$ is of the form $5\cdot2^k+1$. We compute the numbers $F_k$ and

2131:   $f_k$ of normal subgroups of $\Gg$ contained in $K$ of index $2^n$,

2132:   with respectively $n=2^k+2$ and $n=5\cdot2^k+1$, yielding the upper

2133:   and lower bounds. The computations are simplified by the fact that

2134:   for these two values of $n$ there are only subgroups of type $\rmI$.

2135:   

2136:   Let us start by the upper bound, when $n=2^k+2$. First, for $k=2$,

2137:   the subgroups of index $2^n$ are $W(\7;;\7)$, $W(\7;\lambda;\7)$ and

2138:   $W(\8;;\lambda)$, giving $F_2=3$.  Then, for $k>2$, the subgroups can

2139:   of index $2^n$ can be described as follows:

2140:   \begin{enumerate}

2141:   \item $W(A\8\7;\mathcal B\7;C\7)$ for all $W(A\7;\mathcal B;C)$ 

2142:     counted in $F_{k-1}$, except when $C=\7^{k-3}$, when no subgroup

2143:     appears in $F_k$, and when $C=\7^{k-2}$, when $C\7$ should be

2144:     replaced by $\7^{k-3}\8$;

2145:   \item $W(A\7;\mathcal B\8;C\8)$ for all $W(A;\mathcal B;C)$ counted

2146:     in $F_{k-1}$, except when $C=\7^{k-3}$, when no subgroup

2147:     appears in $F_k$, and when $C=\7^{k-2}$, when $C\8$ should be

2148:     replaced by $\7^{k-1}$;

2149:   \item $W(A\7;\{A\}\cup\mathcal B\8;C\8)$, with the same

2150:     qualifications as above;

2151:   \item $W(\7^{k-2}\8;;\7^{k-2})$.

2152:   \end{enumerate}

2153:   It then follows that $F_k=3(F_{k-1}-1)+1$, so $F_k=\frac293^k+1$ for

2154:   all $k\ge2$.

2155:   

2156:   For the lower bound, we have $f_0=F_2=3$; and for $k>0$, when

2157:   $n=5\cdot2^k+1$, the subgroups can of index $2^n$ can be described

2158:   as follows:

2159:   \begin{enumerate}

2160:   \item $W(A\8\8;\mathcal B\7;C\7)$ for all $W(A\8;\mathcal B;C)$ 

2161:     counted in $f_{k-1}$;

2162:   \item $W(A\7\8;\mathcal B\8;C\8)$ for all $W(A\8;\mathcal B;C)$

2163:     counted in $f_{k-1}$;

2164:   \item $W(A\7\8;\{A\7\}\cup\mathcal B\8;C\8)$, with the same

2165:     qualifications as above;

2166:   \item $W(\8^k\7;;\7^{k+1})$ and $W(\8^k\7;\8^k;\7^{k+1})$.

2167:   \end{enumerate}

2168:   It then follows that $f_k=3(f_{k-1}-2)+2$, so $F_k=3^k+2$ for all

2169:   $k\ge0$.

2170:   

2171:   In summary, the number of normal subgroups of index $2^n$ oscillates

2172:   between $3^{\log_2(\frac{n-1}{5})}+2$ and $\frac293^{\log_2(n-2)}+1$

2173:   for $n\ge6$ (when all normal subgroups of $\Gg$ are contained in

2174:   $K$). These bounds give respectively

2175:   $5^{-\log_2(3)}(n-1)^{\log_2(3)}$ and $\frac29(n-2)^{\log_2(3)}$.

2176: \end{proof}

2177: 

2178: Note also the following curiosity:

2179: \begin{cor}\label{cor:allodd}

2180:   The number of normal subgroups of index $r$ of $G$ is odd for all

2181:   $r$'s a power of $2$, and even (in fact, $0$) for all other $r$.

2182: \end{cor}

2183: \noindent(The same congruence phenomenon holds for the group

2184: $C_2*C_3$, as observed by Thomas M\"uller~\cite{muller:subgroups})

2185: \begin{proof}

2186:   The proof follows from the description of

2187:   Theorem~\ref{thm:normal:Gg}.  Assume $r=2^k$. To determine the

2188:   parity of the number of subgroups of index $r$, it suffices to

2189:   consider which $W(A;\mathcal B;C)$ expressions have no choices for

2190:   $\mathcal B$. These are precisely the $W(A;;\7^n)_\rmI$ with

2191:   $2^{n+1}<\#A\le5\cdot2^n$, the $W(\7^n\8\7;;C)_\rmI$ with

2192:   $2^n<\#C\le2^{n+1}$ and the $W(\infty;\8^n,C-1;C)_\rmIII$ with

2193:   $2^{n+1}+1<\#C\le3\cdot2^n+1$.

2194: 

2195:   Now these last two families yield a subgroup for precisely the same

2196:   values of $k$, namely those satisfying $6\cdot2^j+2\le

2197:   k\le7\cdot2^j+1$, and therefore contribute nothing modulo $2$. The

2198:   first family contributes a subgroup for all $k$.

2199: \end{proof}

2200: 

2201: %\begin{figure}\label{fig:topG}

2202: %  \[\xymatrix{

2203: %    & & & {\Gg}\lar[d]\\

2204: %    {\raisebox{2ex}{\tiny(7 groups)}} & & &

2205: %    {\makebox[0mm]{$\stab_\Gg(1)$}}\lar[d] & & & {}

2206: %    \save"2,1"."2,7"*[F-]\frm{}\restore\\

2207: %    {\raisebox{2ex}{\tiny(7 groups)}} &

2208: %    {\langle c,\Gg'\rangle}\lar[dl]\lar[d]\lar[drr] & & 

2209: %    {\langle b,\Gg'\rangle}\lar[dl]\lar[d]\lar[dr] & &

2210: %    {\langle d,\Gg'\rangle}\lar[dll]\lar[d]\lar[dr] & {}

2211: %    \save"3,1"."3,7"*[F-]\frm{}\restore\\

2212: %    \squish{[a,c],d^ab}\lar[dr] & \squish{c}\lar[d] &

2213: %    \squish{[a,b],c^ad}\lar[d] & {\Gg'}\lar[dll]\lar[dl]\lar[dr] &

2214: %    \squish{b}\lar[dll] & \squish{[a,d],b^ac}\lar[dl] &

2215: %    \squish{d,x^2}\lar[dll]\lar[dl]\lar[d]\\

2216: %    & \squish{[a,c]} & {K}\lar[ddll]\lar[ddrr] & &

2217: %    \squish{[a,d],x^2}\lar[d]\lar[dr] & \squish{d}\lar[d] &

2218: %    \squish{dx^2}\lar[dl]\\

2219: %    & & & & \squish{[a,d]x^2} & \squish{[a,d]}\\

2220: %    {} & & & & {}}\]

2221: %  \caption{The top of the lattice of normal subgroups of $\Gg$ not below

2222: %    $K$. The boxes abbreviate the lattice between $\Gg$ and

2223: %    $\Gg'$}\label{fig:latticeG}

2224: %\end{figure}

2225: 

2226: \begin{figure}

2227:   \footnotesize\[\kern-2cm\xymatrix@!@R-3ex@C+1em{

2228:     {2^0} &

2229:     & & & {\Gg}\lar[dlll]\lar[dll]\lar[dl]\lar[d]\lar[dr]\lar[drr]\lar[drrr]\\

2230: %

2231:     {2^1} &

2232:     \squish{a,c}\lar[d]\lar[dr]\lar[drr] &

2233:     \squish{a,b}\lar[dl]\lar[drr]\lar[drrrr] &

2234:     \squish{a,d}\lar[dll]\lar[drr]\lar[drrrr] &

2235:     \squish{b,c,d}\lar[dl]\lar[d]\lar[dr] &

2236:     \squish{ab,c}\lar[dll]\lar[dr]\lar[drr] &

2237:     \squish{ad,b}\lar[dllll]\lar[dll]\lar[dr] &

2238:     \squish{ac,d}\lar[dlllll]\lar[dll]\lar[dl]\\

2239: %

2240:     {2^2} &

2241:     \squish{a}\lar[drrr] &

2242:     \squish{ac}\lar[drr] &

2243:     {\langle c,\Gg'\rangle}\lar[dll]\lar[dl]\lar[dr] &

2244:     {\langle b,\Gg'\rangle}\lar[dl]\lar[d]\lar[dr] &

2245:     {\langle d,\Gg'\rangle}\lar[dl]\lar[dr]\lar[drr] &

2246:     \squish{ab}\lar[dll] &

2247:     \squish{ad}\lar[dlll]\\

2248: %

2249:     {2^3} &

2250:     \squish{c}\lar[drr] &

2251:     \squish{[a,c],cx}\lar[dr] &

2252:     \squish{b}\lar[dr] &

2253:     {\Gg'}\lar[dl]\lar[d]\lar[dr] &

2254:     \squish{x,[a,c]b}\lar[dl] & 

2255:     \squish{d,x^2}\lar[dl]\lar[d]\lar[dr] &

2256:     \squish{\8(b),xd}\lar[dll]\\

2257: %

2258:     {2^4} &

2259:     & & \squish{[a,c]}\lar[dr]

2260:     & {K}\lar[d]

2261:     & \squish{\8(b),x^2}\lar[dl]\lar[d]\lar[dr]

2262:     & \squish{d}\lar[d]

2263:     & \squish{dx^2}\lar[dl]\\

2264: %

2265:     {2^5} &

2266:     & & & \squish{\7(x),x^2}\lar[dl]\lar[d]\lar[dr]

2267:     & \squish{\8(b)x^2}\lar[d]

2268:     & \squish{\8(b)}\lar[dl]\\

2269: %

2270:     {2^6} &

2271:     & & \squish{\7(x)x^2}\lar[dr] &

2272:     \squish{\8(x),x^2}\lar[dl]\lar[d]\lar[dr] &

2273:     \squish{\7(x)}\lar[dl]\\

2274: %

2275:     {2^7} &

2276:     & & \squish{\8(x)x^2,\7(x^2)}\lar[dl]\lar[d]\lar[dr] &

2277:     \squish{\8(x)}\lar[d] &

2278:     \squish{\7\7(x),x^2}\lar[dl]\lar[d]\lar[dr]\\

2279: %

2280:     {2^8} &

2281:     & \squish{\8(x)x^2,\8(x^2)}\lar[dl]\lar[d]\lar[dr] &

2282:     \squish{\8(x)x^2\7(x^2)}\lar[d] &

2283:     \squish{\7\7(x),\7(x^2)}\lar[dl]\lar[d]\lar[dr] &

2284:     \squish{\7\7(x)x^2}\lar[d] &

2285:     \squish{x^2}\lar[dl]\\

2286: %

2287:     {2^9} &

2288:     \squish{\8(x)x^2}\lar[d] &

2289:     \squish{\8(x)x^2\8(x^2)}\lar[dl] &

2290:     \squish{\7\7(x),\8(x^2)}\lar[dll]\lar[dl]\lar[dr] &

2291:     \squish{\8\7(x),\7(x^2)}\lar[dl]\lar[d]\lar[dr] &

2292:     \ssquish{\7\7(x)\7(x^2),\8(x^2)}\lar[dl]\lar[dr]\lar[drr]\\

2293: %

2294:     {2^{10}} &

2295:     \squish{\7\7(x)}\lar[dr] &

2296:     \squish{\7\7(x)\8(x^2)}\lar[d] &

2297:     \squish{\8\7(x)\7(x^2)}\lar[dr] &

2298:     \squish{\8\7(x),\8(x^2)}\lar[dll]\lar[d]\lar[dr] &

2299:     \squish{\7\8(x),\7(x^2)}\lar[dll]\lar[dl]\lar[dr] &

2300:     \squish{\7\7(x)\7(x^2)}\lar[dl] &

2301:     \ssquish{\7\7(x)\7(x^2)\8(x^2)}\lar[dll]\\

2302: %

2303:     {2^{11}} &

2304:     & \squish{\8\7(x)}\lar[drr] &

2305:     \squish{\8\8(x),\7(x^2)}\lar[dll]\lar[dl]\lar[drr] &

2306:     \squish{\7\8(x),\8(x^2)}\lar[dl]\lar[d]\lar[dr] &

2307:     \squish{\8\7(x)\8(x^2)}\lar[dl] &

2308:     \ssquish{\7\8(x)\7(x^2),\8(x^2)}\lar[dl]\lar[d]\lar[dr]\\

2309: %

2310:     {2^{12}} &

2311:     \squish{\7\7\7(x),\7(x^2)}\lar[d]\lar[dr]\lar[drr] &

2312:     \squish{\8\8(x)\7(x^2)}\lar[dr] &

2313:     \squish{\7\8(x)\8(x^2)}\lar[dr] &

2314:     \squish{\7\8(x)}\lar[d] &

2315:     \squish{\8\8(x),\8(x^2)}\lar[dll]\lar[dl]\lar[dr] &

2316:     \ssquish{\7\8(x)\7(x^2),\7\7(x^2)}\lar[dl]\lar[d]\lar[dr] &

2317:     \ssquish{\7\8(x)\7(x^2)\8(x^2)}\lar[dl]\\

2318: %

2319:     {2^{13}} &

2320:     \squish{\8\7\7(x),\7(x^2)} &

2321:     \ssquish{\7\7\7(x)\7(x^2),\8(x^2)} &

2322:     \squish{\7\7\7(x),\8(x^2)} &

2323:     \squish{\8\8(x)} &

2324:     \ssquish{\7\8(x)\7(x^2),\8\7(x^2)} &

2325:     \ssquish{\8\8(x)\8(x^2),\7\7(x^2)} &

2326:     \ssquish{\7\8(x)\7(x^2)\7\7(x^2),\8\7(x^2)}}\]

2327:   \caption{The top of the lattice of normal subgroups of $\Gg$, of

2328:     index at most $2^{13}$}\label{fig:latticeG}

2329: \end{figure}

2330: 

2331: \subsection{Normal subgroups in $\GS$} The normal subgroup growth

2332: of $\GS$ is much larger. As a crude lower bound, consider the

2333: quotient $A=\gamma_k(\GS)/\gamma_{k+1}(\GS)$ for

2334: $k=\frac12(\alpha_{2n+1}+1)$. It is abelian of rank $2^n$, and the

2335: index of $\gamma_k(\GS)$, respectively $\gamma_{k+1}(\GS)$, is

2336: $3^{3^{2n-1}\pm2^{n-1}+1}$.

2337: 

2338: In the vector space $\F[3]^j$, there are roughly $3^{\binom j2}$

2339: subspaces; so $A$ has about $3^{4^n}$ subgroups

2340: $S=N/\gamma_{k+1}(\GS)$, each of them giving rise to a subgroup $N$ of

2341: index roughly $3^{9^n}$.

2342: 

2343: It then follows that the number of normal subgroups of $\GS$ of

2344: index $3^n$ is at least $3^{n^{\log_3(2)}}$, a function intermediate

2345: between polynomial and exponential growth. More precise estimations of

2346: the normal subgroup growth of $\GS$ will be the topic of a future

2347: paper.

2348: 

2349: \subsection*{Acknowledgments}

2350: I wish to express my immense gratitude to the referee who helped me

2351: clarify many parts of the present and forthcoming paper.

2352: 

2353: \begin{bibsection}

2354: \begin{biblist}

2355: \bibselect{math}

2356: \end{biblist}

2357: \end{bibsection}

2358: 

2359: %\bibliography{mrabbrev,people,math,bartholdi}

2360: %\bibselect{grigorchuk}

2361: %\bibliographystyle{amsxport}

2362: \end{document}

2363: