1: \documentclass[11pt]{amsart}

2: 

3: %Packages:

4: \usepackage{latexsym}

5: \usepackage{amsthm,amssymb,amsmath,amsopn}

6: %\usepackage[metapost]{mfpic} 

7: %\opengraphsfile{myfigs}

8: \usepackage[all]{xypic}

9: \usepackage{pstricks}

10: 

11: %Theorems

12: \newtheorem{theorem}{Theorem}[section]

13: \newtheorem{lemma}[theorem]{Lemma}

14: \newtheorem{corollary}[theorem]{Corollary}

15: \newtheorem{proposition}[theorem]{Proposition}

16: \newtheorem{defn}[theorem]{Definition}

17: \newtheorem{remark}[theorem]{Remark}

18: \newtheorem{definition}[theorem]{Definition}

19: \newtheorem{example}[theorem]{Example}

20: 

21: %Operators

22: \DeclareMathOperator{\sgn}{sgn}

23: \DeclareMathOperator{\Def}{def}

24: \DeclareMathOperator{\cyc}{cyc}

25: \renewcommand{\neg}{\mathrm{neg}} %always write \neg()

26: \DeclareMathOperator{\desc}{desc}

27: 

28: %Euler's number

29: \newcommand{\e}{\mathrm{e}}

30: 

31: %Greek letters

32: \renewcommand{\epsilon}{\varepsilon}

33: \renewcommand{\theta}{\vartheta}

34: 

35: %Shuffle product

36: %\newcommand{\sshuf}{\sqcup}

37: %\newcommand{\shuf}{\sqcup}

38: %Tried wedging two sqcups together, but spacing was inconsistent

39: %Also the symbol ends up looking rather big.

40: %\newcommand{\sshuf}{\hbox{$\sqcup\!\sqcup$}}

41: %\newcommand{\shuf}{\hbox{$\sqcup\hskip -4.25pt plus 0pt minus 0pt\sqcup$}}

42: %Spacing

43: \newcommand{\vthinspace}{\hspace{0.5pt}}

44: 

45: \def\l{\lambda}

46: \def\L{\Lambda}

47: \def\g{\gamma}

48: \def\a{\alpha}

49: \def\b{\beta}

50: \def\d{\delta}

51: \def\e{\varepsilon}

52: \def\s{\sigma}

53: \def\Bbb{\mathbb}

54: 

55: \def\R{\mathbf{R}}

56: \def\E{\mathbb{E}}

57: \def\N{\mathbf{N}}

58: \def\e{\varepsilon}

59: \def\g{\gamma}

60: \def\O{\Omega}

61: \def\AA{{\mathcal A}}

62: 

63: %\renewenvironment{proof}[1][Proof]{\textbf{#1:} }{\ \rule{0.5em}{0.5em}}

64: %\input{tcilatex}

65: %\usepackage[dvips]{graphicx}

66: %\usepackage{pst-text,pstricks}

67: 

68: \newcommand\shuf{

69: \setlength{\unitlength}{.4pt}

70: \begin{picture}(40,20)

71: \put(10,2){\line(1,0){20}}

72: \put(10,2){\line(0,1){10}}

73: \put(20,2){\line(0,1){10}}

74: \put(30,2){\line(0,1){10}}

75: \end{picture}

76: }

77: 

78: \newcommand\sshuf{\!

79: \setlength{\unitlength}{.3pt}

80: \begin{picture}(40,20)

81: \put(10,2){\line(1,0){20}}

82: \put(10,2){\line(0,1){10}}

83: \put(20,2){\line(0,1){10}}

84: \put(30,2){\line(0,1){10}}

85: \end{picture}\!

86: }

87: 

88: \begin{document}

89: 

90: \title[The moments of  L\'evy area]{A combinatorial method for calculating

91: the moments of  L\'evy area}

92: %I think this shorter title is clearer

93: 

94: \author{Daniel Levin}

95: \address{Mathematical Institute, University of Oxford, 24-29 St Giles', Oxford

96: OX1 3LB, United Kingdom}

97: \email{levin@maths.ox.ac.uk}

98: \thanks{The first author is supported by the EPSRC Fellowship ``Partial

99: differential equations --- A rough

100: path approach'' GR/S18526/01}

101: \author{Mark Wildon}

102: \address{Department of Mathematics, University of Wales, Swansea, Singleton 

103: Park, Swansea SA2 8PP, United Kingdom}

104: \email{m.j.wildon@swansea.ac.uk}

105: \thanks{The second author is supported by EPSRC Grant EP/D054664/1}

106: \keywords{L\'evy area, shuffle product, signature of a path}

107: \date{\today \newline

108:      \indent 2000 \emph{Mathematics Subject Classification} 60J65

109:      (primary), 05A15 (secondary).} 

110: 

111: \begin{abstract}

112: We present a new way to compute the moments of the L\'evy 

113: area of a two-dimensional Brownian motion. Our approach uses

114: iterated integrals 

115: and combinatorial arguments involving the shuffle product. 

116: \end{abstract}

117: 

118: \maketitle

119: 

120: %\setcounter{tocdepth}{1}

121: %\tableofcontents

122: \section {Introduction}

123: 

124: In this paper we present a new approach to the problem of finding 

125: the moments of the signed area swept out by a two-dimensional

126: Brownian motion. This is a classical problem of great importance, 

127: originally solved by L{\'e}vy (see~\cite{Levy}).

128: 

129: We begin by explaining how these moments may

130: be defined. 

131: Given a piecewise smooth path $\gamma_t : [0,T] \rightarrow \R^2$ 

132: we may complete it to a loop  $\bar{\gamma}$

133: by closing it with the chord from $\gamma_T$ to $\gamma_0$. 

134: We may then define its \emph{signed area} to be

135: \[ \iint_{\R^2} n(\bar{\gamma},x) dx \]

136: where $n(\bar{\gamma},x)$ is the winding number of $\bar{\gamma}$

137: about the point $x \in \R^2$. 

138: % as illustrated in the figure below.

139: 

140: \psset{xunit=0.8cm, yunit=0.8cm}

141: \begin{figure}[h]

142: \begin{pspicture}(8,6)    

143: 	\psline{<-}(4,3)(8,6)

144:     \psline{-}(0,0)(4,3)

145:     {\pscurve{-}(0,0)(2.5,1)(4,5)(6,2)(8,6)} 

146:     \psline{->}(6,2)(6.05,2.0)

147:     \rput(-0.4,0){$\gamma_0$}

148:     \rput(8.35,6.0){$\gamma_T$}

149:     \rput(6.3,4.0){$+$}

150:     \rput(3.85,3.5){$-$}

151:     \rput(1.7,0.9){$+$}

152:     \rput(3.85,2.0){$0$}

153:     \rput(0.05,0){\circle*{0.1}}

154:     \rput(8.05,6){\circle*{0.1}}

155: \end{pspicture}

156: \caption{Contributions to the signed area of 

157: $\gamma : [0,T] \rightarrow \R^2$.}

158: \end{figure}

159: 

160: Now suppose that $B_t : [0,T] \rightarrow \R^2$ is

161: a two-dimensional Brownian motion. If  we complete~$B$ 

162: to a loop~$\bar{B}$ and attempt to define the 

163: signed area of~$B_t$ as before, then we immediately encounter the problem

164: that almost surely,~$n(\bar{B}_t, \cdot)$ is

165: not integrable on~$\R^2$ (see \cite[Theorem 55.I]{Levy3}). 

166: %[\textbf{I'd rather not say this is an ``exercise''. Is there a reference?}]

167: One solution is to replace~$B_t$ with

168: a sequence of piecewise linear dyadic approximations. 

169: In~\cite[Theorem 10]{Levy2} and~\cite[Chapter 55]{Levy3}, L{\'e}vy

170: proved that almost surely, the winding number integral

171: is defined for each approximation, and that the sequence

172: of areas converges. This gives one possible definition

173: of the L{\'e}vy area of the process~$B_t$. 

174: 

175: We may also define the signed

176: area corresponding to a smooth path  

177: $\gamma : [0,T] \rightarrow \R^2$ with $\gamma_t = (x_t,y_t)$

178: by

179: \[  \frac{1}{2} 

180: \int_0^T \bigl( (x_s-x_0) y'_s  - (y_s-x_0) x'_s \bigr) ds  

181: .\]

182: This observation motivates the following alternative definition of L{\'e}vy

183: area.

184: 

185: \begin{defn}\label{defn:levyarea}

186: Let $B_t = (X_t,Y_t)$ for $t\in [0,T]$ be a two-dimensional

187: Brownian motion starting at $0$. The \emph{L{\'e}vy area}

188: of $(B_t)_{0 \le t \le T}$ is given by the stochastic integral

189: \[ \AA_{T} = \frac{1}{2} \int_0^T (X_s dY_s - Y_s dX_s). \]

190: \end{defn}

191: 

192: \begin{remark}

193: In the sequel, we use $\AA$ as a shorter notation for $\AA_1$.

194: \end{remark}

195: 

196: 

197: %\begin{remark}

198: %In \cite{Lyons} Lyons showed

199: %that we may interpret the definition of~$\AA_{T}$

200: %using either an

201: %{It\^o} or a Stratonovich integral and almost

202: %surely get the same value; here it will

203: %be convenient to

204: %always use the latter interpretation.

205: %\end{remark}

206: 

207: 

208: 

209: L{\'e}vy showed in \cite{Levy} that almost surely the

210: definitions of signed area by dyadic approximation

211: and  by stochastic integration agree.

212: In his paper he also gave two different ways

213: to find the characteristic function, and hence

214: the moments, of $\AA_{T}$ when~$T = 2\pi$.

215: 

216: \begin{theorem}[L{\'e}vy]\label{thm:levy} If $T = 2\pi$ then

217: \begin{equation*}

218: %\label{char_fn}

219: \E \left( \exp \left(

220: i z \AA_{2\pi} \right) 

221: \right) =\left( \cosh \pi z 

222: \right)^{-1}.

223: %\E \left( \mathrm{e}^{i z \AA_{2\pi}} \right) =\left(\cosh \pi z \right)^{-1}.

224: \end{equation*}

225: \end{theorem}

226: 

227: L{\'e}vy's first proof uses the  definition of $\AA_{T}$ by

228: dyadic approximation.

229: His second starts from Definition~\ref{defn:levyarea}, 

230: but depends on earlier work by Kac, Siegert, Cameron and Martin (see

231: references cited in \cite[\S 1.6]{Levy}). 

232: 

233: In this paper we present

234: a direct and largely self-contained 

235: proof of Levy's Theorem, using Definition 1.1 to define L{\'e}vy area.

236: Our approach, which we outline in~\S 2 below,

237: is based on the fact that moments of L{\'e}vy area can be 

238: expressed as iterated integrals and hence calculated explicitly 

239: by exploiting the combinatorics of shuffle products. 

240: 

241: It seems likely that our methods can be applied

242: more broadly. We mention here that by 

243: using the multiplication

244: \[ (x,y,a)(x',y',a') = \Bigl( x+x',y+y',a+a' + \frac{1}{2}(xy'-yx') \Bigr) \]

245: we may identify points $(x,y,a) \in \R^3$ with elements

246: of the Heisenberg group. It is known (see \cite[Section 3.2.2]{LyLeCar}) 

247: that the process

248: $(X_t,Y_t,\AA_t)$ is a Brownian motion on this group.

249: Computing aspects of the joint distribution of $(X_t,Y_t,\AA_t)$

250: is a hard problem, involving Harish-Chandra formulae

251: (see \cite{Connell}); it is in effect aimed at understanding

252: the heat kernel on the Heisenberg group (see \cite{Gaveau}).

253: It seems likely that our approach may shed some light 

254: on these questions. Moreover, it should be possible to use

255: our methods to study the signed areas obtained when we replace

256: Brownian motion by measures related to higher order PDEs.

257: 

258: 

259: 

260: 

261: 

262: \section{Outline}

263: The outline of our proof is as follows. In \S 3 we 

264: use a simple scaling argument to show

265: that Theorem~\ref{thm:levy} is implied by the

266: following assertion about the moments of $\AA$.

267: 

268: \begin{theorem}\label{thm:moments}

269: $\E \AA^n = 2^{-n} E_n$.

270: \end{theorem}

271: 

272: \noindent Here $E_n$ is the $n$th Euler number, as 

273: defined by the generating

274: function

275: \[ \sum_{n=0}^\infty E_n \frac{z^n}{n!} = (\cos z)^{-1}.\]

276: The first few non-zero Euler numbers are $E_0 = 1$, $E_2 = 1$,

277: $E_4=5$, $E_6=61$. Of course all odd-numbered

278: Euler numbers are zero. (Correspondingly, one

279: can easily see that $\E \AA^n = 0$ if $n$ is odd.)

280: We then show that the moments of $\AA$ can be

281: expressed using iterated integrals.

282: 

283: In \S 4 we introduce the shuffle product on the

284: tensor algebra of a vector space, and use it to give 

285: an expression for $\E \AA^n$

286: as a certain coefficient in the

287: expansion of a shuffle product. In \S 5 we use a combinatorial

288: argument to determine this coefficient, thereby proving 

289: Theorem~\ref{thm:moments} and hence L{\'e}vy's

290: Theorem.

291: 

292: %This method might be also applied for signed measures when we replace the 

293: %Brownian motion by measures related to high-order PDEs. 

294: 

295: %[\textbf{I am not sure if this is a good comment. DL}]

296: 

297: %[\textbf{Missed out comment on heat kernel / joint distribution.

298: %Could maybe expand this? Say something about how method might generalise to

299: %other problems?}]

300: 

301: \section{The moments of L{\'e}vy's area for  Brownian motion}

302: 

303: We first show that L{\'e}vy's theorem (Theorem~\ref{thm:levy}) 

304: follows from

305: Theorem~\ref{thm:moments}. 

306: If we scale the Brownian path $B_t=(X_t, Y_t)$ defined for 

307: $0\le t \le 1$ by setting $\tilde{B}_s=\sqrt{T} B_{s/T}$

308: then we obtain a new Brownian path defined for

309: $0 \le t \le T$. As before, ~$\AA$ is the L\'evy area of~$B_t$ at time~$1$

310: and ~$\AA_{T}$ is the L\'evy area of $\tilde{B}_s$ at time $T$.

311: It follows easily from Definition~\ref{defn:levyarea} that~$\AA_{T}=T \AA_{1}$.

312: %and 

313: %\[

314: %\E \left( \exp\left({iz \AA_T}\right)\right)= \E \left( \exp\left({i T z \AA}

315: %\right)\right).

316: %\]

317: %

318: %In particular, by for the Brownian motion on $[0,1]$

319: %\[

320: %\E \left( \exp\left({iz \AA}\right)\right)= 

321: %\left( \cosh \frac{\pi z}{T}\right)^{-1}=\left( \cosh \frac{z}{2} 

322: %\right)^{-1}.

323: %\]

324: %

325: %Since $\left( \cosh \frac{z}{2} \right)^{-1}$ is analytic at $z=0$, $\E \left(

326: %\AA^n\right)<+\infty$ for any $n \in \N$, and moreover, the series 

327: %\[

328: %\E \left( \exp\left({iz \AA}\right)\right)=\sum_{n=0}^{\infty} 

329: %\frac{(iz)^n}{n!} \E\left(\AA^n \right)

330: %\]

331: %is absolutely convergent for any $|z|<\pi$.  

332: Hence, assuming that Theorem~\ref{thm:moments} holds, the moments of

333: L{\'e}vy area at time~$T$ are given by

334: \[ \E \AA_{T}^n = T^n n!\vthinspace E_n. \]

335: In particular, by setting $T = 2\pi$ we find that the

336: characteristic function of L{\'e}vy area at time $2\pi$ is 

337: \[

338: \E \left( \exp \left(i z \AA_{2\pi} \right) \right)=\sum_{n=0}^\infty \pi^n E_n

339:  \frac{(iz)^n}{n!} = (\cos \pi iz)^{-1}=(\cosh \pi z)^{-1},

340: \]

341: where we have absolute convergence of the series for $|z|<1/2$.

342: 

343: Therefore, to prove Theorem~\ref{thm:levy},

344: we may concentrate on finding the moments of $\AA$.

345: For this we shall need the following algebraic definition.

346: 

347: %In particular, setting $T=2\pi$ we see that

348: %\begin{equation}

349: %\label{char_fn01}

350: %\E \left( e^{i z S} \right)=\left( \cosh \frac{\pi z}{T} \right)^{-1}=

351: %\left( \cosh \frac{z}{2} \right)^{-1}.

352: %\end{equation}

353: %Since $\left( \cosh \frac{z}{2} \right)^{-1}$ is analytic at $z=0$, 

354: %$\E\left( S^n \right) < +\infty$ for any $n \in \N$, and moreover, 

355: %the series

356: %\begin{equation*}

357: %\E \left( e^{i z S} \right) =\sum_{n=0}^{\infty} \frac{\left(iz\right)^n}{n!} 

358: %\E\left(S^n \right)

359: %\end{equation*} 

360: %is absolutely convergent for any $|z|<R$ for some $R>0$. 

361: 

362: \begin{definition}

363: Let $V$ be a real vector space. Let

364: \[ 

365: T((V))=\prod_{k=0}^{\infty} V^{\otimes k}.

366: \]

367: Thus a typical element of $T((V))$ is a formal infinite sequence

368: $(a_0,a_1,a_2,\ldots)$ where $a_k \in V^{\otimes k}$. (By

369: convention $V^{\otimes 0} = \R$.) Clearly $T((V))$ is a real

370: vector space. It is easy to check that $T((V))$ becomes an

371: algebra with unit if we define the product of

372: $\mathbf{a}=(a_0, a_1, a_2, \dots),

373: \mathbf{b}=(b_0, b_1, b_2, \dots) \in T((V))$ by

374: \[ \mathbf{a} \otimes \mathbf{b} =\left(\dots, \sum_{j=0}^k a_j \otimes

375: b_{k-j}, \dots \right). \]

376: We also define the \emph{exponential} of a tensor $\mathbf{a}$

377: to be the formal sum

378: \[

379: \exp\left( {\bf a} \right)=\sum_{n=0}^{\infty} \frac{{\bf a}^{\otimes n}}{n!}.

380: \]

381: \end{definition}

382: 

383: %\bigskip

384: 

385: %{\bf [I suppressed again the truncated tensor definition DL]}

386: 

387: %\begin{definition}

388: %The truncated tensor algebra $T^{(n)}$ is defined to be

389: %\begin{equation*}

390: %T^{(n)}(V)=\bigoplus_{k=0}^{n} V^{\otimes k},

391: %\end{equation*}

392: %equipped with the product for ${\bf a}=(a_0, a_1, \dots a_n)$,

393: % ${\bf b}=(b_0, b_1, \dots, b_n)$,

394: %\begin{equation*}

395: %{\bf a} \otimes {\bf b}=\left( a_0 b_0, \dots,  \sum_{j=0}^n a_j \otimes

396: %b_{n-j} \right).

397: %\end{equation*}

398: %\end{definition}

399: %We use $\pi_n$ to denote the natural projection of

400: %$T(V)$ into $T^{(n)}(V)$; $\pi_n$ is an algebra

401: %homomorphism.

402: 

403: We can now define  the signature of a Brownian

404: motion. As motivation, we first recall this definition for a path. 

405: If $\gamma : [0,T] \rightarrow \R^2$ 

406: is a path of finite length (see \cite{LyHa}) then its \emph{signature} 

407: is the formal infinite sum  

408: \[ \textbf{X}_{s,t}(\gamma) = \sum_{k=0}^{\infty} \int_{s<t_1<\dots<t_k<t} 

409: d\gamma_{t_1} \otimes \dots \otimes d\gamma_{t_k} \in T((\R^2)) \]

410: defined for $0 \le s < t \le T$.

411: 

412: \begin{example}

413: For example, if $e_0, e \in \R^2$ and $\gamma_t = e_0 + t e$, 

414: then

415: \[ \textbf{X}_{s,t}(\gamma) = \exp  (t-s) e . \]

416: \end{example}

417: 

418: One of the most important properties of the signature is that it is 

419: multiplicative (for the proof of the following theorem see \cite{Chen}). 

420: 

421: \begin{theorem}

422: For any $0\le r<s<t\le T$ 

423: \begin{equation*}

424: %\label{ChenID}

425: {\bf X}_{r,t}(\g)={\bf X}_{r,s}(\g) \otimes {\bf X}_{s,t}(\g).

426: \end{equation*}

427: \end{theorem}

428: 

429: \begin{definition}

430: Let $B : [0,T] \rightarrow \R^2$ be a two-dimensional 

431: Brownian motion. For $0 \le s <t \le T$ we define 

432: the \emph{signature} of $B$ to be the formal infinite sum

433: \begin{equation*}

434: %\label{iter}

435:     \textbf{X}_{s,t}(B) = \sum_{k=0}^{\infty} \int_{s<t_1<\ldots<t_k<t} 

436:     dB_{t_1}\otimes \dots \otimes dB_{t_k}.

437: \end{equation*}

438: %Here the integral may be interpreted in either the Stratonovich or 

439: %the It\^o stochastic sense.

440: \end{definition}

441: 

442: \begin{remark}

443: Up till now, the signature of a Brownian motion and other auxiliary objects 

444: could be interpreted either in the Stratonovich or It\^o sense. However, the 

445: next theorem holds only for the Stratonovich integral. For its

446: proof see either Fawcett \cite{Fawcett} or Lyons--Victoir \cite{LyVic}.

447: %[\textbf{I suggest we delete this remark, and maybe add

448: %something to Remark 1.2.}]

449: \end{remark}

450: 

451: %By a result of Fawcett \cite{Fawcett} or Lyons--Victoir \cite{LyVic}

452: %we have the following theorem.

453: \begin{theorem}

454: \label{LV}

455: If $B$ is a Brownian motion in $\R^2$ then

456: %and ${\bf X}_{0,1}(B)$ is defined using 

457: %\eqref{iter}, then

458: \begin{equation} 

459: \label{LV_result}

460: \E \bigl(\mathbf{X}_{0,1}  (B) \bigr)=\exp \left( \frac{1}{2} \left( 

461: e_1 \otimes e_1 + e_2 \otimes e_2 \right) \right)

462: \end{equation}

463: where $e_1$ and $e_2$ are any two orthogonal vectors in $\R^2$. 

464: \end{theorem}

465: 

466: %\bigskip

467: 

468: %{\bf [I suppressed again the truncated iterated integrals definition DL]}

469: 

470: %We denote by ${\bf X}^{(n)}_{s,t}\left(\g\right)$ the truncated series of

471: %degree $n$ of iterated integrals $\pi_n \left( {\bf X}_{s,t}\left( \g\right)

472: %\right)$. More precisely,

473: %\begin{equation*}

474: %{\bf X}_{s,t}^{(n)}\left( \g\right)=\left( 1, X_{s,t}, \dots,

475: %X_{s,t}^n \right),

476: %\end{equation*}

477: %where for each $n\ge1$,

478: %\begin{equation*}

479: %X_{s,t}^{n}=\int_{s<t_1<\dots<t_n<t} d\g_{t_1} \otimes \dots \otimes 

480: %d\g_{t_n}.

481: %\end{equation*}

482: 

483: In particular,~\eqref{LV_result} implies that 

484: \begin{equation}

485: \label{trunc_ID} 

486: \E \left(X^{2n}_{0,1} \left( B \right) \right)=\frac{1}{2^n n!} 

487: \left( e_1 \otimes e_1 + e_2 \otimes e_2 \right)^{\otimes n}

488: \end{equation}

489: where $X^{2n}_{0,1}$ is the component of $\mathbf{X}_{0,1}$

490: lying in $(\R^2)^{\otimes 2n}$.

491: 

492: %\begin{remark}

493: %We denote by ${\bf X}^{(n)}_{s,t}\left(\g\right)$ the truncated series of 

494: %degree $n$ of iterated integrals $\pi_n \left( {\bf X}_{s,t}\left( \g\right)

495: %\right)$. More precisely,    

496: %\begin{equation*}

497: %{\bf X}_{s,t}^{(n)}\left( \g\right)=\left( 1, X_{s,t}, \dots, 

498: %X_{s,t}^n \right), 

499: %\end{equation*}

500: %where for each $n\ge1$, 

501: %\begin{equation*}

502: %X_{s,t}^{n}=\int_{s<t_1<\dots<t_n<t} d\g(t_1) \otimes \dots \otimes d\g(t_n).

503: %\end{equation*}

504: %In particular, \eqref{LV_result} implies 

505: %\begin{equation}

506: %\label{trunc_ID} 

507: %\E \left({X}^{2n}_{0,1} \left( B \right) \right)=\frac{1}{2^n n!} 

508: %\left( e_1 \otimes e_1 + e_2 \otimes e_2 \right)^{\otimes n}.

509: %\end{equation}

510: %\end{remark}

511: 

512: \section {Shuffle products and other combinatorial objects}

513: 

514: We now introduce an important combinatorial object which will be used in the 

515: sequel.

516: \begin{definition}

517: \label{shuffle}

518: We define the set $S_{m,n}$ of $(m,n)$ shuffles to be the subset of 

519: permutations in the symmetric group $S_{m+n}$ defined by

520: \begin{equation*}

521: S_{m,n}=\{ \s \in S_{m+n}: \s(1) < \dots < \s(m), \; \s(m+1)<\dots<\s(m+n)\}.

522: \end{equation*}

523: %The subset $S_{m,n} \subset S_{m+n}$ is called $(m,n)$-shuffles. 

524: \end{definition}

525: 

526: \begin{remark}

527: The term ``shuffle'' is used because such permutations arise when

528: one riffle 

529: shuffles a deck of $m+n$ cards cut into one pile of $m$ cards and a second

530: pile of $n$ cards.

531: \end{remark}

532: 

533: Let $V=\R^2$ with the orthogonal basis $e_1$ and $e_2$.  

534: Let $V^{*}$ be the space dual to $V$ and let $e^1$ and $e^2$ 

535: be its dual basis. 

536: Let $n \in \mathbf{N}$. The elements $e_{i_1} \otimes \dots \otimes 

537: e_{i_n}$, where each $i_k \in \{1,2\}$, for

538: $k=1, \dots, n$, form a basis of $V^{\otimes n}$. The

539: corresponding dual basis of ${\left( V^{*} \right)}^{\otimes n}$ is 

540: given by the elements $e^{i_1} \otimes 

541: \dots \otimes e^{i_n}$.

542:   

543: There is a natural duality $\left<\;, \; \right> : V^{\otimes n} \times

544: (V^{\star})^{\otimes n} \rightarrow \R$ defined by:

545: \begin{equation*}

546: \left<e^{i_1} \otimes \dots \otimes e^{i_n}, e_{j_1} \otimes \dots \otimes

547: e_{j_n}\right>=\d_{i_1 j_1} \dots \d_{i_n j_n}.

548: \end{equation*}

549: 

550: \begin{definition}

551: Set $(k_1, \dots k_{m+n})=(i_1, \dots, i_m, j_1, \dots, j_n)$. The shuffle 

552: product of two tensors

553: $e^I=e^{i_1}\otimes \dots \otimes e^{i_m}$ and 

554: $e^J=e^{j_1}\otimes \dots \otimes e^{j_n}$, is the tensor 

555: $e^I \shuf e^J$ defined by

556: \begin{equation*}

557: e^I \shuf e^J=\sum_{\s \in S_{m,n}} e^{k_{\s^{-1}(1)}} \otimes \dots \otimes 

558: e^{k_{\s^{-1}(m+n)}}. 

559: \end{equation*}

560: %where $k_1, \ldots, k_{m+n}$ are given by

561: %\[ k_l = \begin{cases} i_l & \text{if $1 \le l \le m$} \\

562: %           j_{l-m} & \text{if $m+1 \le l \le m+n$.} \end{cases} \]

563: Let

564: $T(V^\star) = \bigoplus_{k=0}^\infty (V^\star)^{\otimes k}$ be the ordinary tensor algebra on $V^\star$.

565: The shuffle product~$\shuf$ extends to a bilinear map 

566: \[ T(V^\star) \times T(V^\star) \rightarrow T(V^\star).\]

567: %\[ {\left( V^{*} \right)}^{\otimes m} \times 

568: %{\left( V^{*} \right)}^{\otimes n} \rightarrow 

569: %{\left( V^{*} \right)}^{\otimes (m+n)}. \]

570: \end{definition}

571: 

572: For example, the reader may check that

573: \[  (e \otimes f) \shuf g = e \otimes f \otimes g + e \otimes g \otimes f

574: + g \otimes e \otimes f \]

575: for any $e,f,g \in (\R^2)^\star$. For an alternative description

576: of the shuffle product, see Definition~5.14 in \cite{LyHa}.

577: 

578: 

579: \begin{remark}

580: It follows easily from the definition that the

581: shuffle product is commutative and 

582: associative. We shall use the following notation for the shuffle product 

583: applied $N$ times:

584: \begin{equation*}

585: {\bf a}^{\sshuf N}=\underbrace{{\bf a} \shuf \dots \shuf {\bf a}}_N.

586: \end{equation*}

587: Later in \S 5 we shall also use the analogous version of the shuffle

588: product defined on the tensor powers of $V$.

589: \end{remark}

590: 

591: %For each path $\gamma_s$, $s\in [0,T]$ of

592: %finite length and each tensor ${\bf e} = e^{i_1} \otimes \ldots

593: %\otimes e^{i_n}$, we may now define 

594: %a real number denoted $\varphi_{\bf e}(\gamma)$ by

595: %Change in Final6rev5

596: For each path $\g_s$, $s \in [0,T]$ of finite

597: length we now introduce a real-valued function

598: %We can now introduce the following real valued functions  

599: \[ \varphi(\g): T(V^{\star}) \rightarrow \R \] 

600: defined on the tensor 

601: ${\bf e}=e^{i_1} \otimes \dots \otimes e^{i_n}$ by  

602: \begin{align}

603: \varphi_{\bf e} (\g)

604: %(e^{i_1} \otimes \dots \otimes e^{i_n})

605: %={\bf e} \left( {\bf X}_{0,T}\left( \g \right) \right) 

606: %Commented out, as you never use this alternative form later

607: \notag 

608: &= \left<e^{i_1} \otimes \dots \otimes e^{i_n}, 

609: \int_{0<t_1<\dots<t_n<T} d\g_{t_1} \otimes \dots \otimes d\g_{t_n}\right>

610: \notag \\

611: &=\int_{0<t_1<\dots<t_n<T} d\g^{i_1}_{t_1} \dots d\g^{i_n}_{t_n}.

612: \label{coord}

613: \end{align}

614: %Here $T(V^\star)$ is the ordinary tensor algebra of $V^\star$.

615: The fundamental property of 

616: $\varphi_{\bf e}(\g)$ (see \cite[Theorem 2.15]{LyLeCar}) is that   

617: \begin{equation}

618: \label{shuffle_prod}

619: \varphi_{\bf e}(\g) \varphi_{\bf f}(\g)=\varphi_{{\bf e} \sshuf {\bf f}}(\g).

620: \end{equation}

621: From now on we change slightly notations for $e_1, e_2$ and $e^1, e^2$. Let 

622: $x$, $y$ be a basis for $\R^2$ and let

623:  $x^{*}$, $y^{*}$ be the dual basis of $(\R^2)^\star$. 

624: We use these techniques together to prove the following theorem.

625: \begin{theorem}

626: \label{Main}

627: The $n$th moment of L\'evy area at time $1$ for a two-dimensional

628: Brownian motion~$B$

629: starting at zero is the signature 

630: ${\bf X}_{0,1}(B)$ contracted with the shuffle powers of the dual 

631: tensor $\frac{1}{2}\left( x^{*} \otimes y^{*} - y^{*} \otimes x^{*} \right)$.

632: That is,

633: \begin{equation*}

634: \E \left( \AA^n \right)=

635: 2^{-2n} \left<(x^{*} \otimes y^{*} - y^{*} \otimes x^{*})^{\sshuf n}, 

636: \frac{(x \otimes x + y \otimes y)^{\otimes n}}{n!}\right>.

637: \end{equation*}

638: \end{theorem}

639: 

640: \begin{proof}

641: Let $B=(X_s,Y_s)$, $0\le s \le 1$ be a Brownian path in $\R^2$ starting at 

642: zero. Then 

643: \begin{equation*}

644: \AA=\frac{1}{2} \int_0^1 \left( X_s \, d Y_s - Y_s \, d X_s \right)=

645: \frac{1}{2} \int_{0<t<s<1} \left( d X_{t} \, d Y_{s} - d Y_t \, d X_s \right).

646: \end{equation*} 

647: In a more canonical notation as in \eqref{coord} we may write this down as 

648: \begin{equation*}

649: \AA=\frac{1}{2} \int_{0<t_1<t_2<1} \left( d X_{t_1} \, d Y_{t_2} - d Y_{t_1} 

650: \, d X_{t_2} \right).

651: \end{equation*}

652: So $\AA=\varphi_{\frac{1}{2}\left( x^{*} \otimes y^{*} - y^{*} \otimes x^{*} 

653: \right)} 

654: \left( B \right)=\frac{1}{2} \left( \varphi_{x^{*} \otimes y^{*}} \left(B 

655: \right) - \varphi_{y^{*} \otimes x^{*}} \left( B \right) \right)$. 

656: Further, using~(\ref{shuffle_prod}) we have 

657: \begin{align*}

658: \AA^n&=\frac{1}{2^n} \Bigl<\left( x^{*}  \otimes y^{*} - y^{*} \otimes x^{*} 

659: \right), \int_{0<t_1<t_2<1} dB_{t_1} \otimes dB_{t_2}\Bigr>^n

660: \\

661: &=\frac{1}{2^n} \Bigl<\left( x^{*}  \otimes y^{*} - y^{*} \otimes x^{*} 

662: \right)^{\sshuf n}, \;

663: X_{0,1}^{2n}\left( B \right)\Bigr>.

664: \end{align*} 

665: 

666: Taking the expectation of $\AA^n$ we have by \eqref{trunc_ID},

667: \begin{align*}

668: \E\left( \AA^n \right) &= 2^{-n} \left<\left( x^{*} \otimes y^{*} - y^{*} 

669: \otimes x^{*} \right)^{\sshuf n}, \; \E \left( X_{0,1}^{2n}\left(B \right)

670: \right)\right>

671: \\ 

672: &=2^{-2n} \left<\left( x^{*} \otimes y^{*} - y^{*} \otimes x^{*} 

673: \right)^{\sshuf n}, \; 

674: \frac{\left( x \otimes x + y \otimes y \right)^{\otimes n}}{n!}\right>. 

675: \end{align*}

676: 

677: \vspace{-22pt}

678: 

679: \end{proof} 

680: 

681: Hence to prove Theorem~\ref{thm:moments} it is sufficient to prove

682: the following theorem.

683: 

684: \begin{theorem}\label{thm:MW1} %(Combinatorial identities)

685: \[ \left<\left( x^{*} \otimes y^{*} - y^{*} \otimes x^{*} \right)^{\sshuf n}, 

686: \; 

687: \left( x \otimes x + y \otimes y \right)^{\otimes n}\right>

688: = 2^n n!\vthinspace E_n.

689: \]

690: \end{theorem}

691: 

692: \section{Proof of Theorem~\ref{thm:MW1}}  

693: \label{Proof_Theor}

694: 

695: \subsection{}

696: From now on it will often be convenient to use

697: a shorter notation for elements in the standard basis

698: of the tensor

699: algebra~$T(V)$, in which we write 

700: $xy$ rather than~\hbox{$x\otimes y$},~$x^2$ rather than~\hbox{$x

701: \otimes x$},

702: and so on. Using this notation 

703: the standard basis elements of~$T(V)$ are simply the words

704: in the letters~$x$ and~$y$. 

705: 

706: We shall say that a 

707: word in the letters $x$ and $y$ is \emph{even} if (i) it is of the

708: form %$z_1^2 \ldots z_r^2$ 

709: $z_1^2 \ldots z_n^2$ where each $z_i \in \{x,y\}$, and (ii)

710: there are equal numbers of $x$'s and $y$'s. 

711: When we expand

712: \[ (x \otimes x + y \otimes y)^{\otimes n} \]

713: we obtain the sum of all words in $x$ and $y$ of length $2n$

714: satisfying condition~(i). If such a word is not killed

715: by $(x^\star \otimes y^\star - y^\star \otimes x^\star)^{\sshuf n}$ then

716: clearly it must also satisfy~(ii). Hence Theorem~\ref{thm:MW1} is equivalent

717: to the following assertion.

718: 

719: \begin{theorem}\label{thm:eqvform}

720: Let 

721: \[ (xy-yx)^{\sshuf n} = \sum_v \lambda_v v \]

722: where the sum is over all words $v$ of length $2n$. Let

723: $u_n = \sum \lambda_w$ 

724: where the sum is over all even words $w$ of length $2n$. Then

725: \begin{equation*} u_n = 2^n n!\vthinspace E_n 

726: %\begin{cases} 

727:  %                      2^n n!\vthinspace E_n & \text{if $n$ is even} \\

728:  %                      0 & \text{if $n$ is odd}.

729:  %        \end{cases}

730: \end{equation*}

731: \end{theorem}

732: 

733: %If we use your notation for Euler numbers, need $(-1)^n$.

734: 

735: \subsection{}

736: We begin the proof of 

737: Theorem~\ref{thm:eqvform} by noting that if $n$ is odd then there are

738: no even words of length $2n$, and so $u_n = 0$, as required.

739: We may therefore assume that~$n = 2m$ is even, so the

740: even words that appear in $(xy-yx)^{\sshuf n}$ are of length $4m$,

741: with the pairs $xx$ and $yy$ each appearing exactly $m$ times. 

742: 

743: Our proof

744: depends on counting the combinatorial objects

745: introduced in the next definition.

746: 

747: \begin{defn}An \emph{$xy$-matching} is a pair $(w,\sigma)$

748: where $w$ is an even word, of length $2m$ say,

749: and $\sigma$ is a fixed-point-free involution in

750: the symmetric group $S_{2m}$

751: such that $w_i = x$ if and only if $w_{\sigma(i)} = y$.

752: 

753: Given an $xy$-matching $\delta = (w,\sigma)$

754: we define the \emph{negativity} of $\delta$ by

755: \[ \neg(\delta) = \# \{i \in \{1 \ldots 2m\} : 

756: \text{$w_i = x$ and $\sigma(i) < i$}\}.

757: \]

758: We define the \emph{sign} of $\delta$ by $\sgn(\delta) = (-1)^{\neg(\delta)}$.

759: Let $N_t(w)$ be the number of $xy$-matchings with underlying word

760: $w$ and negativity $t$.

761: \end{defn}

762: 

763: It will be very useful to represent $xy$-matchings by diagrams

764: such as the one below.

765: 

766: \bigskip

767: \bigskip

768: \hfil

769: \xymatrix@C=6pt{x \ar@{-}@/^12pt/[rr]&

770: x \ar@{-}@/^24pt/@<2pt>[rrrrrr]&

771: y &

772: y \ar@{-}@/_12pt/[rr]&

773: x \ar@{-}@/^12pt/[rr]&

774: x &

775: y &

776: y &

777: }

778: \hfill

779: 

780: \bigskip

781: \begin{center}

782: {\sc Figure 2.} The $xy$-matching $(xxyyxxyy, (13)(28)(46)(57))$.

783: \end{center}

784: 

785: \bigskip

786: 

787: \noindent The arcs contributing to $\neg(\delta)$ are those drawn

788: below the word, thus here $\neg (\delta) = 1$ and

789: $\sgn(\delta) = -1$.

790: As an exercise, the reader may check that

791: there are in total $16$ $xy$-matchings

792: with underlying word $xxyyxxyy$ and negativity~$1$, and so

793: $N_1(xxyyxxyy) = 16$.

794: 

795: %We now show how $xy$-matchings are relevant to our problem.

796: 

797: \begin{proposition}\label{prop:nasty}

798: Let $w$ be an even word of length $2m$ and

799: let $s + t = 2m$. The coefficient

800: of $w$ in $(xy)^{\sshuf s} \shuf (yx)^{\sshuf t}$ is

801: $s!t!\vthinspace N_t(w)$.

802: \end{proposition}

803: 

804: The proof of this proposition

805: is postponed to the appendix at the end of this 

806: paper. The idea is to 

807: associate to each $xy$-matching

808: with underlying word~$w$ and negativity~$t$ exactly

809: $s!t!$ ways to obtain $w$ by expanding the shuffle product

810: \hbox{$(xy)^{\sshuf s} \shuf (yx)^{\sshuf t}$}. With the help

811: of a formal definition of an expansion of a shuffle product

812: of this form,

813: we are able to

814: show that these possibilities are exhaustive.

815: Here we shall illustrate

816: the correspondence when

817: $m = 4$, $w = xxyyxxyy$,~$s=3$ and~$t=1$.

818: 

819: The figure below shows one

820: way to obtain $w$ by expanding the shuffle prouduct~$xy \shuf xy

821: \shuf xy \shuf yx$.

822: 

823: 

824: \medskip

825: %To get round spacing problems with the tail of the y 

826: \newcommand{\y}{\raisebox{3pt}[10pt]{$y$}}

827: \vbox{

828: \hfil

829: \xymatrix@C=3pt@R=42pt{

830: x\ar@{.>}[d] & \y\ar@{.>}[drr] 

831: & \shuf & x\ar@{.>}[dll] & \y\ar@{.>}[drrrrrr]  & \shuf 

832: & x\ar@{.>}[d] & \y\ar@{.>}[drr]  &  \shuf & \y\ar@{-->}[dlllll] & x\ar@{-->}[dlll]  \\

833: x & x & 

834: & y & y& & x  & x & & y & y}

835: \hfill

836: 

837: \smallskip

838: \begin{center}

839: {\sc Figure 3.} An expansion of $xy \shuf xy \shuf xy \shuf yx$.

840: \end{center}}

841: 

842: \medskip

843: \noindent

844: 

845: 

846: We obtain the corresponding $xy$-matching by connecting the letters

847: coming from the same $xy$ or $yx$ term in $xy \shuf xy \shuf xy \shuf yx$.

848: 

849: \bigskip\bigskip

850: \hfil

851: \newcommand{\cupspace}{\hspace{8pt}}

852: \xymatrix@C=3pt@R=36pt{

853: x\ar@{-}@/^12pt/[rrr] & x \ar@{-}@/^24pt/[rrrrrrrrr] & \cupspace

854: & y & y\ar@{-}@/_12pt/[rrr] & \cupspace& x\ar@{-}@/^12pt/[rrr] & x 

855: & \cupspace& y & y}

856: \hfill

857: 

858: \bigskip

859: \begin{center}

860: {\sc Figure 4.} The $xy$-matching corresponding to the expansion in

861: Figure 3.

862: \end{center}

863: 

864: \medskip

865: \noindent 

866: Note that this matching, $(xxyyxxyy, (13)(28)(46)(57))$,

867: has negativity~$1$, corresponding to the single $yx$ term.

868: 

869: 

870: There are in total $3! 1!$

871: ways to obtain~$w$ by expanding the shuffle product 

872: $xy \shuf xy

873: \shuf xy \shuf yx$

874: which correspond to this matching. The

875: remaining five are obtained by permuting identical words

876: in the top line of Figure~3.  For example,

877: we could just as well get the loop joining the first

878: $x$ to the first $y$ 

879: by expanding the shuffle product as shown below.

880: 

881: \medskip

882: \hfil

883: \xymatrix@C=3pt@R=42pt{

884: x\ar@{.>}[dr] & \y\ar@{.>}[drrrrrrrrr] 

885: & \shuf & x\ar@{.>}[dlll] & \y\ar@{.>}[dl]  & \shuf 

886: & x\ar@{.>}[d] & \y\ar@{.>}[drr]  

887: &  \shuf & \y\ar@{-->}[dlllll] & x\ar@{-->}[dlll]  \\

888: x\ar@{-}@/^12pt/@<-2pt>[rrr] & x \ar@{-}@/^18pt/[rrrrrrrrr] & \cupspace

889: & y & y\ar@{-}@/_12pt/[rrr] & \cupspace& x\ar@{-}@/^12pt/@<-1pt>[rrr] & x 

890: & \cupspace & y & y}

891: %x & x & 

892: %& y & y& & x  & x & & y & y}

893: \hfill

894: 

895: \medskip

896: \begin{center}

897: {\sc Figure 5.} Another of the $3!1!$ expansions of $xy \shuf xy \shuf xy \shuf yx$

898: giving the $xy$-matching shown in Figure~4.

899: \end{center}

900: 

901: %\bigskip

902: %Conversely, given a way to obtain $w$ by

903: %expanding $xy \shuf xy \shuf xy \shuf xy$, we may keep track of where each letter goes, and draw the corresponding diagram. The bottom row will be 

904: %the diagram of an $xy$-matching with negativity $1$ and

905: %underlying word $w$. This shows that the six ways

906: %to obtain $w$ described above exhaust the possibilities.

907: 

908: \begin{corollary}\label{cor:nasty} Let $w$ be an even word of length $2m$.

909: The coefficient of $w$ in $(xy-yx)^{\sshuf 2m}$ is

910: \[ (2m)! \sum_{\delta} \sgn(\delta) \]

911: where the sum is over all $xy$-matchings with underlying word $w$.

912: \end{corollary}

913: 

914: \begin{proof}Let $s + t = 2m$. As the shuffle

915: product is commutative,

916: when we expand $(xy-yx)^{\sshuf 2m}$ we obtain

917: $(-1)^t (xy)^{\sshuf s} \shuf (yx)^{\sshuf t}$ exactly $\binom{2m}{s}$

918: times.

919: Hence,

920: by Proposition~\ref{prop:nasty}, the coefficient we seek is

921: \[ \sum_{s+t = 2m} \binom{2m}{s} (-1)^t s!t! N_t(w) =

922: (2m)! \sum_{t=0}^{2m} (-1)^t N_t(w) 

923: = (2m)! \sum_{\delta} \sgn(\delta). \]

924: where the final equality holds because

925: the middle sum counts each $xy$-matching exactly

926: once, with the appropriate sign.

927: \end{proof}

928: 

929: \subsection{}

930: We now introduce our second and final

931: combinatorial object.

932: 

933: \begin{defn}An \emph{$XY$-matching} is a pair $(W,\sigma)$

934: where $W$ is a word in letters $X$ and $Y$ with equal

935: numbers of $X\!$'s and $Y\!$'s, say $m$ of each, and 

936: $\sigma \in S_{m}$ is a permutation such that

937: $W_i = X$ if and only if $W_{\sigma(i)} = Y$. 

938: \end{defn}

939: 

940: Once again, it is very useful to represent

941: $XY$-matchings  by diagrams. The figure below

942: shows a typical example.

943: 

944: \vbox{

945: \bigskip

946: \bigskip

947: \bigskip

948: 

949: \hfil

950: \xymatrix{A \ar@/^12pt/[r]& 

951: B \ar@/_12pt/[r]& 

952: A \ar@/^12pt/[r]& 

953: B \ar@/_15pt/@<-9pt>[lll]& 

954: B \ar@/_9pt/[r]& 

955: A \ar@/^15pt/[l]

956: }

957: \hfill

958: 

959: \bigskip

960: \begin{center}

961: {\sc Figure 5.} The $XY$-matching $(XYXYYX, (1234)(56))$.

962: \end{center}}

963: \medskip

964: 

965: As before, we need certain quantities associated with

966: an $XY$-matching.

967: 

968: \begin{defn}

969: Let $\Delta = (W,\sigma)$ be an $XY$-matching. The \emph{length} of $\Delta$

970: is the length of the word $W$. 

971: We define the \emph{negativity} of $\Delta$ by

972: \[ \neg (\Delta) = \# \{ i : \text{$W_i = X$ and $\sigma(i) < i$ \emph{or}

973: $W_i = Y$ and $\sigma(i) > i$} \}. \]

974: We define the \emph{sign} of $\Delta$ by

975: \[ \sgn(\Delta) = (-1)^{\neg(\Delta)}.\] 

976: We define the \emph{cycle count} of $\Delta$ by

977: \[ \cyc(\Delta) = \text{$\#$disjoint cycles in the permutation $\sigma$}.\]

978: Finally we let $e(\Delta)$

979: be the even word obtained from $\Delta$ by replacing each $X$

980: with $xx$ and each $Y$ with $yy$.

981: \end{defn}

982: 

983: For example, if $\Delta$ is the $XY$-matching shown in Figure 5 above, 

984: then $\Delta$ has length~$6$,

985: $\neg(\Delta) = 3$ (the $3$ arcs drawn below the word contributing),

986: $\sgn(\Delta) = -1$,~$\cyc(\Delta) = 2$ and $e(\Delta) =

987: xxyyxxyyyyxx$.

988: 

989: 

990: \begin{proposition} Let $w$ be an even word of length $2m$ and

991: let $s + t = 2m$. The coefficient

992: of $w$ in $(xy)^{\sshuf s} \shuf (yx)^{\sshuf t}$ is

993: \[ 2^{2m} s!t!  \sum_{\Delta}

994: 2^{-c(\Delta)} \]

995: where the sum is over all the $XY$-matchings $\Delta$ such that $e(\Delta) = w$

996: and \hbox{$\neg(\Delta) = t$}.

997: \end{proposition}

998: 

999: \begin{proof}

1000: By Proposition~2.2, it is sufficient to prove that

1001: \[ N_t(w) = 2^{2m} \sum_{\Delta} %

1002: 2^{-\cyc(\Delta)} \]

1003: where the sum is over all the $XY$-matchings $\Delta$ such that $e(\Delta) = w$

1004: and \hbox{$\neg(\Delta) = t$}. To do this, we shall  associate to 

1005: each $XY$-matching $\Delta = (W,\sigma)$ 

1006: exactly $2^{2m-\cyc(\Delta)}$ $xy$-matchings

1007: with underlying word $e(\Delta)$. 

1008: 

1009: The canonical such $xy$-matching is $\delta = (e(W),\tau)$

1010: where $\tau$ is the involution defined by 

1011: \[ \tau(2i-1) = 2\sigma(i) \qquad \text{if  $1 \le i \le m$} \]

1012: %\begin{align*} \tau(2i-1) &= 2\sigma(i) \qquad\hspace{7.9pt} \text{if  $1 \le i \le m$} \\

1013: %    \tau(2i) &= 2\sigma(i) - 1 \quad \text{if $1 \le i \le m$}.

1014: %    %Hack to get spacing right

1015: %    \end{align*}

1016: For example, if $W = XYXYYX$

1017: and $\sigma = (1234)(56)$ then 

1018: \[ \tau =

1019: (14)(27)(36)(58)(9\:12)(10\:11), \] as shown below.

1020: 

1021: \vbox{

1022: \bigskip

1023: \medskip

1024: \bigskip

1025: 

1026: \hfil

1027: \xymatrix@C=6pt@R=42pt{ 

1028: X \ar@/^12pt/[rr]\ar@{-->}[d]\ar@{-->}[dr]& &

1029: Y \ar@/_12pt/[rr]\ar@{-->}[d]\ar@{-->}[dr]& &

1030: X \ar@/^12pt/[rr]\ar@{-->}[d]\ar@{-->}[dr]& &

1031: Y \ar@/_15pt/@<-9pt>[llllll]\ar@{-->}[d]\ar@{-->}[dr]& & 

1032: Y \ar@/_9pt/[rr]\ar@{-->}[d]\ar@{-->}[dr]& &

1033: X \ar@/^15pt/[ll]\ar@{-->}[d]\ar@{-->}[dr]& \\

1034: x\ar@{-}@/^12pt/[rrr] & 

1035: x & 

1036: y \ar@{-}@/_12pt/[rrr] & 

1037: y  & 

1038: x \ar@{-}@/^12pt/[rrr]& 

1039: x  & 

1040: y \ar@{-}@/_18pt/[lllll]& 

1041: y & 

1042: y \ar@{-}@/_15pt/[rrr] & 

1043: y \ar@{-}@/_9pt/[r] & 

1044: x & 

1045: x

1046: }

1047: \hfill

1048: 

1049: \bigskip

1050: \begin{center}

1051: {\sc Figure 6.} The canonical $xy$-matching associated to $(W,\sigma)$.

1052: \end{center}}

1053: \smallskip

1054: 

1055: We obtain the remaining $xy$-matchings  by conjugating $\tau$ by

1056: the \hbox{$2^{2m}$} % = 2^6$} 

1057: elements of the group $\left<(12), \ldots, (2m-1\:2m)

1058: \right>$. There is, however, some double counting, which accounts for the

1059: factor of $2^{-\cyc(\Delta)}$. There are two cases we must consider.

1060: 

1061: Firstly, suppose $\sigma$ has a $2$-cycle, say $(k\:l)$. Then

1062: $\tau$ involves

1063: \[ (2k-1 \:\: 2l)(2k \:\: 2l-1), \] 

1064: which is stabilised

1065: by conjugation by $(2k-1 \:\: 2k)(2l-1 \:\: 2l)$. This gives us a

1066: factor of $1/2$ for each $2$-cycle. 

1067: 

1068: Secondly, if $\hat{\sigma}$

1069: is a cycle of length $4$ or more in $\sigma$, then we

1070: can replace $\hat{\sigma}$ with $\hat{\sigma}^{-1}$

1071: without changing the $xy$-matchings obtained. So again we

1072: must compensate by a factor of $1/2$ to avoid overcounting.

1073: For example, the two $XY$-matchings $(XYXY, (1234))$ and

1074: $(XYXY, (1432))$ both give the same set

1075: of $16$ $xy$-matchings.

1076: %

1077: %\bigskip

1078: %\medskip

1079: %\hfil

1080: %\xymatrix@C=9pt@R=36pt{X \ar@/^10pt/@<-2pt>[r]& 

1081: %Y \ar@/_8pt/[r]& 

1082: %X \ar@/^10pt/@<-2pt>[r]& 

1083: %Y, \ar@/_18pt/@<-4pt>[lll]& 

1084: % & & 

1085: %X \ar@/^18pt/@<4pt>[rrr]& 

1086: %Y \ar@/_10pt/@<2pt>[l]& 

1087: %X \ar@/^10pt/[l]& 

1088: %Y \ar@/_10pt/@<2pt>[l]

1089: %} 

1090: %\hfill

1091: 

1092: %\bigskip

1093: Combining these observations gives the required result.

1094: \end{proof}

1095: 

1096: \bigskip

1097: 

1098: By the same argument used to deduce Corollary~\ref{cor:nasty} 

1099: from Proposition~\ref{prop:nasty}

1100: we obtain the following corollary.

1101: 

1102: \begin{corollary} Let $w$ be an even word of length $2m$.

1103: The coefficient of $w$ in $(xy-yx)^{\sshuf 2m}$ is

1104: \[ (2m)! \vthinspace 2^{2m} \sum_{\Delta} 2^{-\cyc(\Delta)} \sgn(\Delta) \]

1105: where the sum is over all $XY$-matchings $\Delta$ such that $e(\Delta) = w$.

1106: Hence

1107: \begin{equation}\label{eq:XY} 

1108: u_{2m} = (2m)! \vthinspace 2^{2m} \sum_{\Delta} 2^{-\cyc(\Delta)} \sgn(\Delta) 

1109: \end{equation}

1110: where the sum is over all $XY$-matchings $\Delta$ of length $m$.

1111: \hfill$\qed$

1112: \end{corollary}

1113: 

1114: 

1115: 

1116: \subsection{} Let

1117: \[ c_{2r} = \sum_{\Delta} \sgn(\Delta) \]

1118: where the sum is over all $XY$-matchings $\Delta$ of length $2r$ with

1119: \emph{just one} cycle. 

1120: We  use an argument from the theory of exponential

1121: structures (see~\cite[\S5.4]{StanleyII} for the general setting) to obtain

1122: an expression for $u_{2m}$ in terms of the~$c_{2r}$.

1123: This reduces our problem to finding the $c_{2r}$.

1124: 

1125: \begin{lemma}\label{lemma:easy}

1126: \[ u_{2m} = (2m)!2^{2m} \sum_{\genfrac{}{}{0pt}{}{a_1,\ldots,a_m \ge 0}

1127: {a_1 + 2a_2 + \ldots +ma_m  =m}}

1128: \frac{(2m)!}{(2!)^{a_1} \ldots (2m)!^{a_m}} \frac{c_2^{a_1}}{2^{a_1}a_1!}

1129: \ldots \frac{c_{2m}^{a_m}}{2^{a_m}a_m!} \]

1130: %where the sum is over all  $(a_1,\ldots,a_m) \in \mathbf{N}_0^m$ such that

1131: %$a_1 + 2a_2 + \ldots + ma_m = m$.

1132: \end{lemma}

1133: 

1134: \begin{proof}

1135: Consider the contribution to the sum in \eqref{eq:XY} coming from 

1136: those $XY$-matchings

1137: whose underlying permutation has cycle type $(1^{a_1},\ldots,m^{a_m})$. 

1138: To construct such an $XY$-matching, we must first partition $\{1 \ldots 2m\}$

1139: into~$a_1$ subsets of size~$2$,~$a_2$ subsets of size~$4$, and so on,

1140: up to~$a_m$ subsets of size~$2m$. This can be done

1141: in 

1142: \[ \frac{(2m)!}{(2!)^{a_1} \ldots (2m)!^{a_m}} \]

1143: ways. Then we must choose for each subset an $XY$-matching with

1144: just one cycle

1145: on that subset. For the $a_r$ subsets chosen of size $2r$, this

1146: can be done in~$c_{2r}^{a_r}$ ways. However we only care

1147: about the matchings chosen, not the order we choose them in, so we must divide by~$a_r!$, giving

1148: \[ \frac{c_2^{a_1}}{2^{a_1}a_1!}

1149: \ldots \frac{c_{2m}^{a_m}}{2^{a_m}a_m!} \]

1150: choices for the $XY$-matchings on the subsets. The sign of the resulting 

1151: $XY$-matching is the product of the signs of the $XY$-matchings on the

1152: subset, and the total number of cycles is $a_1 + \ldots + a_m$. The result

1153: now follows from~$\eqref{eq:XY}$.

1154: \end{proof}

1155: 

1156: %Lemma~\ref{lemma:exp1} is most useful to us when expressed

1157: %in terms of generating functions.

1158: 

1159: \begin{lemma}

1160: Let 

1161: \[ f(z) = \sum_{m=0}^\infty \frac{u_{2m}}{2^{2m}(2m)!} \frac{z^{2m}}{(2m)!}

1162: \]  be the exponential

1163: generating function of $\frac{u_{2m}}{2^{2m}(2m)!}$. Then

1164: \begin{equation}\label{eq:fgenfunc} 

1165: f(z) = \exp \left( \sum_{r=1}^\infty \frac{c_{2r}}{2} \frac{z^{2r}}{(2r)!}\right). 

1166: \end{equation}

1167: \end{lemma}

1168: 

1169: \begin{proof}

1170: This is an immediate consequence of Lemma~\ref{lemma:easy}. 

1171: \end{proof}

1172: 

1173: \subsection{}

1174: To finish the proof we need to know the $c_{2r}$.

1175: Since the sign of an $XY$-matching is not affected by swapping the letters $X$ and $Y$, we have

1176: \[ c_{2r} = 2\sum_{\Delta} \sgn(\Delta) \]

1177: where the sum is over all $XY$-matchings $\Delta$ of length $2r$ such that 

1178: $\cyc(\Delta) = 1$ and the first letter of $W$ is $X$. 

1179: 

1180: To evaluate this sum we need the following lemma.

1181: Recall that a permutation 

1182: $\tau$ is said to have a \emph{descent} at $i$

1183: if $\tau(i) > \tau(i+1)$.

1184: 

1185: 

1186: \begin{lemma}

1187: Let $\Delta = (W,\sigma)$ be an $XY$-matching of length $2r$ such that $\cyc(\Delta) = 1$ and $W_1 = X$.

1188: Suppose that $\sigma = (1 \: b_1 \ldots b_{2r-1})$.

1189: Let $\tau \in S_{\{2 \ldots 2m\}}$ be the permutation 

1190: defined by $\tau(i) = b_i$. Then

1191: \[ \sgn(\Delta) = (-1)^{r-1} (-1)^{\desc(\tau)} \]

1192: where $\desc(\tau)$ is the number of descents in $\tau$.

1193: \end{lemma}

1194: 

1195: \begin{proof}

1196: If  $i$ is even then $W_{b_i} = A$, and if $i$ is odd then $W_{b_i} = B$. Hence

1197: \[ \neg(\Delta) = \#\{\text{$i$ : $i$ is even and $b_{i+1} < b_i$}\}

1198: +  \#\{\text{$i$ : $i$ is odd and $b_{i+1} > b_i$}\}. \]

1199: But 

1200: $\#\{\text{$i$ : $i$ is odd and $b_{i+1} > b_i$}\} +

1201: \#\{\text{$i$ : $i$ is odd and $b_{i+1} < b_i$}\}  = r-1$

1202: as there are $r-1$ odd numbers $i$ such that $1 \le i < 2r-1$.

1203: Hence

1204: \[ \begin{split} \#\{\text{$i$ : $i$ is odd and $b_{i+1} > b_i$} \} \equiv

1205: \#\{\text{$i$ : $i$ is odd and $b_{i+1} < b_i$}\} \\ + (r-1) \bmod 2. \end{split} \]

1206: The result now follows.

1207: \end{proof}

1208: 

1209: Let $\left<\genfrac{}{}{0pt}{}{t}{d}\right>$ denote the number of

1210: permutations in $S_t$ with exactly $d$ descents.  (These

1211: are known

1212: as the \emph{Eulerian numbers}; our notation for them is taken from

1213: \cite[\S5.1.3]{KnuthTAOCPIII}.) By the previous lemma,

1214: \[ c_{2r} = 2 (-1)^{r-1} \sum_{d=0}^{2r-1} (-1)^d \left<\genfrac{}{}{0pt}{}{2r-1}{d}\right>\]

1215: for each $r \in \N$.

1216: By a well-known property of Eulerian numbers --- see for instance

1217: \cite[Exercise 5.1.3(3)]{KnuthTAOCPIII} --- we have

1218: \[  \sum_{d=0}^{2r-1} (-1)^d \left<\genfrac{}{}{0pt}{}{2r-1}{d}\right>

1219: = (-1)^{r-1}T_r\]

1220: where the $T_r$ are the \emph{tangent numbers}, defined by

1221: \[ \sum_{r=1}^{\infty} T_r \frac{z^{2r-1}}{(2r-1)!} = \tan z.\]

1222: We have therefore shown that

1223: \begin{equation}\label{eq:g} c_{2r} = 2T_r \quad\text{for all $r\in \N$}. \end{equation}

1224: 

1225: \subsection{}

1226: By \eqref{eq:fgenfunc} and \eqref{eq:g},

1227: \[ \begin{split}

1228: f(z) = \exp \left(\sum_{r=1}^\infty T_r \frac{z^{2r}}{(2r)!} \right) 

1229: = \exp \left( \int \sum_{r=1}^\infty T_r \frac{z^{2r-1}}{(2r-1)!} \right) \\

1230: = \exp \left( \int \tan z\right) 

1231:  = \exp \left( \log \sec z \right) 

1232:  = \sec z. \end{split}

1233: \]

1234: The right hand side is the exponential generating function for the 

1235: Euler numbers, so

1236: comparing coefficients, we see that $\frac{u_{2m}}{2^{2m}(2m)!} = E_{2m}$.

1237: This completes the proof. % of Theorem~\ref{thm:eqvform}. 

1238: 

1239: 

1240: \section{Appendix: Proof of Proposition~\ref{prop:nasty}}

1241: We repeat the statement of this proposition below.

1242: 

1243: \setcounter{section}{5}

1244: \setcounter{theorem}{2}

1245: \begin{proposition}

1246: Let $w$ be an even word of length $2m$ and

1247: let $s + t = 2m$. The coefficient

1248: of $w$ in $(xy)^{\sshuf s} \shuf (yx)^{\sshuf t}$ is

1249: $s!t! \vthinspace N_t(w)$.

1250: \end{proposition}

1251: 

1252: We urge the reader to read the discussion

1253: following the original statement of this proposition in \S 5.2

1254: before proceeding. What

1255: follows is a formalised version of the argument we indicated there.

1256: 

1257: \begin{proof} By

1258: an \emph{expansion} of $(xy)^{\sshuf s} \shuf (yx)^{\sshuf t}$

1259: we mean a word of length $2m$ using

1260: each of the letters 

1261: \[ x^1,y^1, \ldots, x^s,y^s, y_1,x_1, \ldots 

1262: y_t,x_t\] exactly once, and such that for each $i$, 

1263: $x^i$ appears before $y^i$, and $x_i$ appears after $y_i$.

1264: It should be clear that the coefficient

1265: of $w$ in $(xy)^{\sshuf s} \shuf (yx)^{\sshuf t}$ is equal

1266: to the number of expansions of $(xy)^{\sshuf s} \shuf (yx)^{\sshuf t}$

1267: which become $w$ when the numbers attached to the letters are erased.

1268: 

1269: The orbits of $S_s \times S_t$ on 

1270: expansions of $(xy)^{\sshuf s} \shuf (yx)^{\sshuf t}$

1271: are all of size $s!t!$. Given such an orbit we shall a corresponding

1272: $xy$-matching with underlying

1273: word~$w$, and permutation $\sigma$. Choose any representative of

1274: the orbit, $w^\star$ say. For $k \in \{1 \ldots 2m\}$ set $\sigma(k)= l$

1275: where $l$ is defined by:

1276: \[ \text{if} \:\begin{cases}

1277: \!\text{ $w^\star_k = x_i$ then $w^\star_l = y_i$} \\

1278: \!\text{ $w^\star_k = x^i$ then $w^\star_l = y^i$} \\

1279: \!\text{ $w^\star_k = y_i$ then $w^\star_l = x_i$} \\

1280: \!\text{ $w^\star_k = y^i$ then $w^\star_l = x^i$} \\

1281: \end{cases}.

1282: \]

1283: It is easy to check that $\sigma$ is a fixed-point-free involution,

1284: that $\sigma$ does not depend on the choice of $w^\star$,

1285: and that $\neg(w,\sigma) = t$. 

1286: 

1287: Conversely, suppose we are

1288: given an $xy$-matching with underlying word $w$ and permutation $\sigma$

1289: and negativity $t$. Set $s = m-t$.

1290: There is a unique

1291: way to write $\sigma$ in the form

1292: \[ (i^1 \: \sigma(i^1)) \ldots (i^s \: \sigma(i^s)) 

1293: (\sigma(i_1) \: i_1) \ldots (\sigma(i_t) \: i_t) \]

1294: such that the following conditions hold:

1295: \begin{align*}

1296: (1) \:\: & i^1 < i^2 < \ldots < i^s, \quad w_{i^j} = x, \quad

1297: \sigma(i^j) > i^j 

1298: \text{ \hskip 2pt if $1 \le j \le s$}; \\

1299: (2) \:\: & i_1 < i_2 < \ldots < i_t\hskip 1pt, \quad w_{i_k} = x, \quad

1300:  \sigma(i_k) < i_k 

1301: \text{ if $1 \le k \le t$}.

1302: \end{align*}

1303: 

1304: We shall define an associated expansion $w^\star$ of

1305: $(xy)^{\sshuf s} \shuf (yx)^{\sshuf t}$. The underlying

1306: word of $w^\star$ is, of course, $w$. 

1307: For $j \in \{1 \ldots s\}$ we set $w^\star_{i^j} = x^j$ and

1308: $w^\star_{\sigma(i^j)} = y^j$. For $k \in  \{1 \ldots t\}$

1309: we set $w^\star_{i_j} = x_j$ and $w^\star_{\sigma(i_j)} = y_j$. 

1310: By virtue of

1311: our expression for $\sigma$, 

1312: the expansion $w^\star$ we have defined is \emph{canonical}, in the sense

1313: that the subscripts and superscripts on its letters appear in

1314: increasing order. 

1315: 

1316: Clearly each $S_s \times S_t$-orbit contains a unique

1317: canonical expansion.

1318: Hence to prove the proposition

1319: it is sufficient to prove that the two maps we have defined give a bijection

1320: between $xy$-matchings and canonical expansions. This is merely

1321: a matter of definition chasing. We illustrate it by an example.

1322: 

1323: Suppose $\delta = (xxyyxxyy, (13)(28)(46)(57))$. 

1324: As $\neg(\delta) = 1$, we associate to $\delta$

1325: an expansion of $xy \shuf xy \shuf xy \shuf yx$. Following

1326: the given algorithm, we take $i^1 = 1, i^2 = 2, i^3 = 5$ and $i_1 = 6$

1327: and assign the labels $(x^1 x^2 y^1 y_1 x^3 x_1 y^3 x^2)$.

1328: Conversely, given this expansion, we get back the permutation

1329: $(13)(28)(46)(57)$.

1330: \end{proof}

1331: 

1332: \section*{Acknowledgements}

1333: 

1334: The first-named author wishes to thank Thierry L\'evy and Terry Lyons for

1335: valuable discussions.

1336: 

1337: %Let me know if you'd like the BibTeX file I used.

1338: 

1339: \begin{thebibliography}{11}

1340: 

1341: %\bibitem{AbrSt} 

1342: %{\sc Abramowitz M. and Stegun, I.} 

1343: %\newblock {\em Handbook of mathematical functions with formulas, graphs, and

1344: % mathematical tables.}

1345: %\newblock Edited by Milton Abramowitz and Irene A. Stegun. Reprint

1346: %of the 1972 edition. Dover Publications, Inc., New York, 1992.

1347: 

1348: %\bibitem{Breiman} 

1349: %{\sc Breiman L.} 

1350: %\newblock {\em Probability,}

1351: %\newblock Addison-Wesley Publishing Company, Reading, Mass.-London-Don Mills,

1352: %Ont.  1968.

1353: 

1354: \bibitem{Chen}

1355: {\sc Chen K.-T.} 

1356: \newblock {\em Integration of paths, geometric invariants and a generalized 

1357: Baker-Hausdorff formula,} 

1358: \newblock Ann. of Math. (2) {\bf 65} (1957), 163--178

1359: 

1360: \bibitem{Connell}

1361: {\sc O'Connell, N.} 

1362: \newblock {\em Conditioned random walks and the RSK correspondence. 

1363: Random matrix theory,}  

1364: \newblock J. Phys. A  {\bf 36}  (2003),  no. 12, 3049--3066. 

1365: 

1366: %\bibitem{Comtet} 

1367: %{\sc Comtet C. L.,} 

1368: %\newblock {\em Advanced combinatorics,} 

1369: %\newblock Revised and enlarged edition, Reidel, Dordrecht, 1974.

1370: 

1371: %\bibitem{Durrett}

1372: %Durrett, Richard, {\it Probability. Theory and examples}, 

1373: %The Wadsworth \& Brooks/Cole Statistics/Probability Series. 

1374: %Wadsworth \& Brooks/Cole Advanced Books \& Software, Pacific Grove, CA, 1991. 

1375: 

1376: \bibitem{Fawcett}

1377: {\sc Fawcett T.} 

1378: \newblock {\em Problems in stochastic analysis. Connections between

1379: rough paths and non-commutative harmonic analysis.} 

1380: \newblock D.~Phil thesis, Oxford University, 2003.

1381: 

1382: \bibitem{Gaveau}

1383: {\sc Gaveau, B.} 

1384: \newblock {\em Principe de moindre action, propagation de la chaleur et 

1385: estim\'ees sous elliptiques sur certains groupes nilpotents,}  

1386: \newblock Acta Math.  {\bf 139}  (1977), no. 1-2, 95--153. 

1387: 

1388: %\bibitem{Grimmett}

1389: %Grimmett, Geoffrey R. and Stirzaker, David R.  

1390: %Probability and random processes. Third edition.

1391: %Oxford University Press, New York,  2001. 

1392: 

1393: \bibitem{LyHa}

1394: {\sc Hambly B. and Lyons T.}

1395: \newblock {\em Uniqueness for the signature of a path of

1396: bounded variation and continuous analogues of the free group,}

1397: \newblock arXiv:math.CA/0507536 (submitted).

1398: 

1399: \bibitem{KnuthTAOCPIII}

1400: {\sc Knuth, D.~E.}

1401: \newblock {\em The art of computer programming. {V}olume 3}.

1402: \newblock Addison-Wesley Publishing Co., Reading, Mass.-London-Don Mills, Ont.,

1403: 1973.

1404: \newblock Sorting and searching, Addison-Wesley Series in Computer Science

1405: and Information Processing.

1406: 

1407: \bibitem{Levy2} 

1408: {\sc L\'evy P.} 

1409: \newblock {\em Le mouvement Brownien plan.} (French)  

1410: \newblock Amer. J. Math.  

1411: \newblock {\bf 62} (1940), 487--550.

1412: 

1413: \bibitem{Levy3}

1414: {\sc L\'evy P.} 

1415: \newblock {\em Processus stochastiques et mouvement Brownien,} 

1416: \newblock Suivi d'une note de M. Lo\`eve. (French) 

1417: Gauthier-Villars, Paris, 1948. 

1418: 

1419: \bibitem{Levy} 

1420: {\sc L\'evy P.} 

1421: \newblock {\em Wiener's random function, and other Laplacian random

1422: functions,} 

1423: \newblock in {\em Proceedings of the Second Berkeley Symposium on 

1424: Mathematical Statistics and Probability, 1950}, 171--187, Univ. California 

1425: Press, Berkeley and Los Angeles, 1951.

1426: 

1427: 

1428: 

1429: \bibitem{LyVic}

1430: {\sc Lyons T. and Victoir N.} 

1431: \newblock {\em Cubature on Wiener space,} 

1432: \newblock Proc. R. Soc. Lond. Ser. A Math. Phys. Eng. Sci. 

1433: \newblock {\bf 460} (2004), no.~2041, 169--198.

1434: 

1435: \bibitem{Lyons}

1436: {\sc Lyons T.} 

1437: \newblock {\em Differential equations driven by rough signals,} 

1438: \newblock Rev. Mat. Iberoamericana {\bf 14} (1998), no.~2, 215--310.

1439: 

1440: \bibitem{LyLeCar} 

1441: {\sc Lyons T. J., L\'evy T., Caruana M.}

1442: \newblock {\em Differential equations driven by rough paths,}  

1443: \newblock {\em Ecole d'Et\'e de Probabilit\'es de Saint-Flour XXXIV-2004}, 

1444: Springer, 2006 (to appear).

1445: 

1446: %\bibitem{Ree}

1447: %{\sc Ree R.} 

1448: %\newblock{\em Lie elements and an algebra associated with shuffles,} 

1449: %\newblock Ann. of Math. (2) {\bf 68} (1958), 210--220. 

1450: 

1451: %\bibitem{Reut}

1452: %{\sc Reutenauer C.} 

1453: %\newblock {\em Free Lie algebras,} 

1454: %\newblock Oxford Univ. Press, New York, 1993.

1455: 

1456: %\bibitem{Shir}

1457: %{\sc Shiryayev, A. N.}  

1458: %\newblock {\em Probability,} 

1459: %\newblock Translated from the Russian by R. P. Boas. Graduate Texts in 

1460: %Mathematics, 95. Springer-Verlag, New York,  1984.

1461:  

1462: \bibitem{StanleyII}

1463: {\sc Stanley, R.~P.}

1464: \newblock {\em Enumerative combinatorics. {V}ol. 2}, vol.~62 of {\em Cambridge

1465: Studies in Advanced Mathematics}. \newblock Cambridge University Press, 

1466: Cambridge, 1999.

1467: \newblock With a foreword by Gian-Carlo Rota and an appendix by Sergey Fomin.

1468: 

1469: \end{thebibliography}

1470: 

1471: \end{document}

1472: