1: \documentclass[12pt]{article}

2: \usepackage{amsmath}

3: \usepackage{amssymb}

4: \usepackage{amscd}

5: \usepackage{epsfig}

6: \usepackage{amsthm}

7: 

8: 

9: \def\mini{\scriptsize}

10: \def\t{\tilde}

11: \newcommand{\diam}{\operatorname{diam}}

12: \newcommand{\Isom}{\operatorname{Isom}}

13: \newcommand{\rank}{\operatorname{rank}}

14: \newcommand{\length}{\operatorname{\ul{length}_{\Delta}}}

15: \newcommand{\Length}{\operatorname{length_{\Delta}}}

16: 

17: 

18: 

19: 

20: 

21: 

22: \def\ora{\overrightarrow}

23: \def\half{\frac{1}{2}}

24: \def\no{\noindent}

25: \setlength{\parindent}{.25in} \setlength{\textwidth}{6in}

26: \setlength{\oddsidemargin}{.25in} \setlength{\evensidemargin}{.25in}

27: \setlength{\textheight}{9in} \setlength{\headheight}{0.2in}

28: \setlength{\topmargin}{-.25in}

29: \setlength{\parskip}{\smallskipamount}

30: 

31: 

32: 

33: 

34: \newtheorem{dfn}{Definition}[section]

35: \newtheorem{rem}[dfn]{Remark}

36: \newtheorem{thm}[dfn]{Theorem}

37: \newtheorem{defn}[dfn]{Definition}

38: \newtheorem{lem}[dfn]{Lemma}

39: \newtheorem{lemma}[dfn]{Lemma}

40: \newtheorem{sublem}[dfn]{Sublemma}

41: \newtheorem{prob}[dfn]{Problem}

42: \newtheorem{prop}[dfn]{Proposition}

43: \newtheorem{ass}[dfn]{Assumption}

44: \newtheorem{add}[dfn]{Addendum}

45: \newtheorem{claim}[dfn]{Claim}

46: \newtheorem{cor}[dfn]{Corollary}

47: \newtheorem{conj}[dfn]{Conjecture}

48: \newtheorem{conjecture}[dfn]{Conjecture}

49: \newtheorem{conc}[dfn]{Conclusion}

50: \newtheorem{cons}[dfn]{Consequence}

51: \newtheorem{condition}[dfn]{Condition}

52: \newtheorem{conv}[dfn]{Convention}

53: \newtheorem{ex}[dfn]{Example}

54: \newtheorem{example}[dfn]{Example}

55: \newtheorem{fact}[dfn]{Fact}

56: \newtheorem{question}[dfn]{Question}

57: \newtheorem{sit}[dfn]{Situation}

58: \newtheorem{spec}[dfn]{Special case}

59: \newtheorem{crit}[dfn]{Criterion}

60: \newtheorem{lis}[dfn]{List}

61: 

62: 

63: \def\proof{\par\medskip\noindent{\it Proof: }}

64: \def\sketch{\par\medskip\noindent{\it Sketch of proof: }}

65: \def\lra{\longrightarrow}

66: \def\Lra{\Longrightarrow}

67: \def\ra{\rightarrow}

68: \def\Ra{\Rightarrow}

69: \def\CR{\curvearrowright}

70: \def\acts{\CR}

71: \def\lh{\hookleftarrow}

72: \def\embed{\hookrightarrow}

73: \def\ti{\tilde}

74: 

75: 

76: \def\back{\backslash}

77: 

78: 

79: \def\P{{\mathbb P}}

80: \def\C{{\mathbb C}}

81: \def\R{{\mathbb R}}

82: \def\H{{\mathbb H}}

83: \def\Z{{\mathbb Z}}

84: \def\K{{\mathbb K}}

85: \def\M{{\mathcal M}}

86: \def\e{{\mathcal E}}

87: 

88: 

89: \def\O{{\mathcal O}}

90: 

91: 

92: 

93: 

94: 

95: 

96: \def\Q{{\mathbb Q}}

97: 

98: \def\B{{\mathbb B}}

99: \def\N{{\mathbb N}}

100: \def\l{{\mathbb l}}

101: \def\F{{\mathbb F}}

102: \def\g{{\mathfrak g}}

103: \def\k{{\mathfrak k}}

104: \def\p{{\mathfrak p}}

105: \def\a{{\mathfrak a}}

106: \def\eps{\epsilon}

107: \def\al{\alpha}

108: \def\be{\beta}

109: \def\ga{\gamma}

110: 

111: 

112: \def\Ga{\Gamma}

113: \def\de{\delta}

114: \def\De{\Delta}

115: 

116: 

117: \def\Del{\Delta}

118: \def\Si{\Sigma}

119: \def\si{\sigma}

120: \def\th{\theta}

121: \def\L{{\cal L}}

122: \def\la{\lambda}

123: \def\La{\Lambda}

124: \def\Om{\Omega}

125: \def\om{\omega}

126: 

127: 

128: \def\>{\rangle}

129: \def\<{\langle}

130: \def\del{\delta}

131: \def\D{\partial}

132: \def\3{\ss}

133: \def\8{\infty}

134: 

135: 

136: \def\ol{\overline}

137: \def\ul{\underline}

138: \def\ov{\overrightarrow}

139: \newcommand{\restr}{\mbox{\Large \(|\)\normalsize}}

140: \def\nbd{neighborhood~}

141: \def\mf{\mathfrak }

142: 

143: 

144: \overfullrule=0pt

145: 

146: 

147: 

148: 

149: \begin{document}

150: 

151: 

152: \title{A path model for geodesics in Euclidean buildings and its

153: applications to  representation theory}

154: \author{Misha Kapovich and John J. Millson}

155: \date{April 12, 2008}

156: \maketitle

157: 

158: 

159: \begin{abstract}

160: In this paper we give a combinatorial characterization of projections of geodesics in Euclidean buildings to

161: Weyl chambers. We apply these results to the representation theory of

162: complex reductive Lie groups and to spherical Hecke rings associated with split nonarchimedean reductive

163: Lie groups. Our main application is a generalization of the

164: saturation theorem of Knutson and Tao for $SL_n$ to other complex semisimple

165: Lie groups.

166: \end{abstract}

167: 

168: 

169: \section{Introduction}

170: 

171: Let $\ul{G}$  be a $\Q$-split reductive algebraic group defined over

172: $\Z$ and $\ul{G}^\vee$ be its Langlands' dual. In this paper we

173: continue our study (which we began in \cite{KLM3}) of the interaction

174: between the representation theory of the group

175: $G^\vee:=\ul{G}^\vee(\C)$ and geometry of the Bruhat--Tits building

176: associated with the nonarchimedean group

177: $G=\ul{G}(\K)$, where $\K$ is a complete field with discrete

178: valuation. We restrict ourselves to the case when $\K$ is a local field, in which case, algebraically speaking,

179: we will be studying the relation between

180: the representation ring of the group

181: $G^\vee$ and the {\em spherical Hecke algebra}

182: $\mathcal{H}_G$ associated with  $G$.

183: 

184: In his papers \cite{Littelmann1}, \cite{Littelmann2},

185: P.~Littelmann introduced a {\em path model} for the

186: representations  of complex reductive Lie groups $G^\vee$. The

187: Littelmann path model gives a method to compute the structure

188: constants of the representation ring of $G^\vee$ by counting

189: certain piecewise linear paths, called LS paths.

190: 

191: 

192: In this paper we define a class of piecewise-linear paths in $\De$, a Weyl chamber of the Weyl group $W$ of $G^\vee$.

193: These paths will be called {\em Hecke paths}  (see Definition \ref{Hecke}) because of their

194: connection with $\mathcal{H}_G$. We will prove that a path $p$ in $\De$ is a Hecke path if and only if $p$

195: is the projection into $\De$ of a geodesic segment in the

196: Euclidean (Bruhat-Tits) building $X$ associated with $G$.  Thus, unlike LS paths which had to be {\em invented}, the

197: Hecke paths appear very naturally as projections of geodesic

198: segments. Hecke paths are defined by eliminating one of the axioms for

199: LS paths, therefore {\em each LS path for $G^\vee$ is a Hecke path

200: for $G$}.

201: 

202: The converse relation is more subtle and is discussed later in the

203: introduction. To state our main results we need a definition of

204: $k_R$, the {\em saturation factor} of the root system $R$ of the

205: group $\ul{G}$. Let $\alpha_1,...,\alpha_l\in R$ be the simple

206: roots (corresponding to $\De$). Let $\theta$ be the highest root

207: and define positive integers $m_1,...,m_l$ by

208: $$

209: \theta = \sum_{i=1}^l m_i \alpha_i.$$

210: Then $k_R$ is the least common multiple of the numbers $m_i$, $i=1,...,l$.

211: We refer to section \ref{kfactors} for the computation of $k_R$.

212: 

213: \medskip

214: Below, $L$ is the character lattice of a maximal torus in $G^\vee$ (so that $\De\subset L\otimes \R$),

215: $Q(R^\vee)$ is the root lattice of $R^\vee$. Our main result is the following

216: theorem (see section \ref{saturationsection}), which, in a weaker

217: form, has been conjectured by S.~Kumar:

218: 

219: 

220: \begin{thm}

221: [Saturation theorem]\label{main}  Let $G^{\vee}$ and $L$ be as above. Suppose that

222: $\al,\be,\ga\in L$ are dominant characters such that $\al+\be+\ga\in

223: Q(R^\vee)$ and that there exists $N\in \N$ so that

224: $$

225: (V_{N\al}\otimes V_{N\be} \otimes V_{N\ga})^{G^\vee}\ne 0.

226: $$

227: Then for $k=k_R^2$ we have:

228: $$

229: (V_{k\al}\otimes V_{k\be} \otimes V_{k\ga})^{G^\vee}\ne 0.

230: $$

231: \end{thm}

232: 

233: Here and in what follows $V_\la$ is the irreducible representation

234: of $G^\vee$ associated with the dominant weight $\la$ of $G^\vee$.

235: Also it will be convenient to introduce integers $n_{\al,\be}(\ga^*)$,

236: the structure constants of the representation ring of the group $G^\vee$. (Here and in what follows, $\ga^*$

237: is the dominant weight contragredient to $\ga$.) Hence

238: $n_{\al,\be}(\ga^*)$ are defined by the equation

239: $$

240: V_{\alpha} \otimes V_{\be} = \bigoplus_{\ga} n_{\al,\be}(\ga^*) V_{\ga^*}.$$

241: Here the right-hand side is the decomposition of the tensor product

242: of the irreducible representations $V_{\alpha}$ and $V_{\be}$

243: into a direct sum of irreducible representations.

244: We can then formulate the above theorem as

245: $$n_{N\alpha, N\beta}(N \gamma^*)\ne 0 \Rightarrow

246: n_{k\alpha, k\beta}(k \gamma^*)\ne 0.

247: %\ \text{for all dominant weights} \ \alpha,\beta,\gamma.

248: $$

249: 

250: 

251: We will freely move back and forth between the symmetric formulation of

252: saturation in Theorem \ref{main} and the asymmetric formulation immediately

253: above.

254: 

255: 

256: 

257: As an immediate corollary of Theorem \ref{main} we obtain a new

258: proof of the saturation theorem of A.~Knutson and T.~Tao

259: \cite{KnutsonTao}:

260: 

261: \begin{cor}

262: Suppose that $R=A_l$, i.e. the semisimple part of $G^\vee$ is

263: locally isomorphic to $SL_{l+1}$. Suppose that $\al,\be,\ga$ are

264: dominant characters such that $\al+\be+\ga\in Q(R^\vee)$ and that

265: there exists $N\in \N$ so that

266: $$

267: (V_{N\al}\otimes V_{N\be} \otimes V_{N\ga})^{G^\vee}\ne 0.

268: $$

269: Then

270: $$

271: (V_{\al}\otimes V_{\be} \otimes V_{\ga})^{G^\vee}\ne 0.

272: $$

273: \end{cor}

274: 

275: Another proof of this theorem was given by H.\ Derksen and J.\ Weyman in \cite{DW}. However

276: the proofs of \cite{KnutsonTao} and \cite{DW} do not work for root systems different from $A_l$.

277: 

278: 

279: \begin{question}

280: \label{simplyconjecture}

281: Is it true that  if $G$ is a simple simply-laced group, then in Theorem \ref{main}

282: one can always take $k=1$ and in the case of non-simply laced groups the

283: smallest $k$ which suffices is $k=2$?

284: \end{question}

285: 

286: 

287: The affirmative answer to this question is supported by the odd orthogonal groups, symplectic groups and $G_2$,

288: when one gets the saturation constant $k=2$ rather than $2^2$ and $6^2$ \cite{KLM3, BK},

289: the group $Spin(8)$, when the saturation constant equals $1$

290: \cite{KKM}, as well as by a number of computer

291: experiments with the exceptional root systems and the root systems $D_l$.

292: 

293: 

294: Using the results of \cite{KLM2, KLM3} one can reformulate  Theorem

295: \ref{main} as  follows:

296: 

297: 

298: \begin{thm}

299: \label{reform} There exists a convex homogeneous cone $D_3\subset \De^3$,

300: defined by the {\em generalized triangle inequalities},

301: which depends only on the Weyl group $W$, so that the following hold:

302: 

303: 

304: 1. If a triple $(\al,\be,\ga)\in (\De\cap L)^3$ satisfies

305: $$

306: (V_{\al}\otimes V_{\be} \otimes V_{\ga})^{G^\vee}\ne 0,

307: $$

308: then $(\al,\be,\ga)\in D_3$ and

309: $$

310: \al+\be+\ga\in Q(R^\vee).

311: $$

312: 

313: 

314: 2. ``Conversely'', if $(\al,\be,\ga)\in k^2_R \cdot L^3 \cap D_3$ and

315: $\al+\be+\ga\in k^2_R \cdot Q(R^\vee)$, then

316: $$

317: (V_{\al}\otimes V_{\be} \otimes V_{\ga})^{G^\vee}\ne 0.

318: $$

319: \end{thm}

320: 

321: 

322: 

323: We now outline the steps required to prove the Theorem \ref{main}. We will need several facts about Hecke rings

324: ${\mathcal H}$. Let $\mathcal{O}$ denote the valuation ring of $\K$, then

325: $K:=\ul{G}(\mathcal{O})$ is a maximal compact subgroup in $G$.

326: The lattice $L$ defined above is the cocharacter lattice of a maximal torus $T\subset G$.

327: The Hecke ring ${\mathcal H}$, as a $\Z$-module, is freely generated by the characteristic functions

328: $\{c_\la: \la \in L \cap \Delta\}$. The

329: multiplication on ${\mathcal H}$ is defined via the convolution

330: product $\star$. Then the  structure constants $m_{\al,\be}(\ga)$ of ${\mathcal H}$

331: are defined by

332: $$

333: c_\al \star c_\be= \sum_\ga m_{\al,\be}(\ga) c_\ga.

334: $$

335: We refer the reader to  \cite{Gross},

336: \cite{KLM3} and to section \ref{heck} of this paper for more details.

337: 

338: 

339: Let $o\in X$ be the special vertex of the building $X$ which is

340: fixed by $K$. In section \ref{distances} we define the notion

341: $d_\De(x,y)$ of the $\De$-valued distance between points $x, y\in

342: X$. Given a piecewise-geodesic path $p$ in $X$ we define its

343: $\De$-length as the sum of $\De$-distances between the consecutive

344: vertices.

345: 

346: The structure constants for ${\mathcal H}$ are related to the geometry of $X$ via the following

347: 

348: 

349: \begin{thm}

350: \cite[Theorem 8.12]{KLM3}.

351: \label{Hecke=triangles}

352: The number $m_{\al,\be}(\ga)$ is equal to the product

353: of a certain positive constant  %(depending only on $G$)

354: by the  number of geodesic triangles $T\subset X$ whose vertices are {\em special vertices}

355: of $X$ with the first vertex equal to $o$ and whose $\De$-side

356: lengths are $\al, \be, \ga^*$.

357: \end{thm}

358: 

359: Theorem \ref{main} is essentially Statement 3, which follows from

360: Statements 1 and 2, of the following

361: 

362: 

363: \begin{thm}

364: \label{3->4}

365: Set $\ell:=k_R$.

366: 

367: 1. Suppose that

368: $(\al,\be,\ga)\in D_3\cap L^3$ and $\al+\be+\ga\in Q(R^\vee)$.

369: Then the structure constants $m_{\cdot,\cdot}(\cdot)$ of the Hecke ring of the group $G$ satisfy

370: $$

371: m_{\ell\al,\ell\be}(\ell\ga)\ne 0.$$

372: 

373: 2. Suppose that $\al, \be, \ga$ are dominant coweights of $\ul{G}$, such that

374: $m_{\al,\be}(\ga)\ne 0$. Then

375: 

376: $$ n_{\ell\al,\ell\be}(\ell\ga)\ne 0.$$

377: 

378: 

379: 3. As a consequence of 1 and 2 we have: Suppose that

380: $(\al,\be,\ga)\in D_3\cap L^3$ and $\al+\be+\ga\in Q(R^\vee)$.

381: Then

382: $$

383: n_{\ell^2\al,\ell^2\be}(\ell^2\ga)\ne 0.$$

384: \end{thm}

385: 

386: \begin{rem}

387: a. Part 1 of the above theorem was proven in \cite{KLM3}. Thus the point

388: of this paper is to prove Part 2 of the above theorem.

389: 

390: b. Examples in \cite{KLM3} show that both implications

391: $$

392: (\al,\be,\ga)\in D_3\cap L^3, \al+\be+\ga\in Q(R^\vee) \Rightarrow

393: m_{\al,\be}(\ga)\ne 0

394: $$

395: and

396: $$

397: m_{\al,\be}(\ga)\ne 0 \Rightarrow n_{\al,\be}(\ga)\ne 0

398: $$

399: are {\bf false} for the groups $G_2$ and $SO(5)$. Therefore the dilation by $k_R$ in both cases is necessary at least

400: for these groups.

401: \end{rem}

402: 

403: More generally, we prove (Theorem \ref{mini3->4})

404: 

405: 

406: 

407: \begin{thm}

408: \label{minus}

409: Suppose that $\al, \be, \ga\in L$ are dominant weights so that one of them

410: is the sum of minuscule weights. Then

411: $$

412: m_{\alpha, \beta}(\gamma)\ne 0 \Rightarrow

413: n_{\alpha, \beta}(\gamma)\ne 0.

414: $$

415: \end{thm}

416: 

417: The proof of Part 2 of Theorem \ref{3->4} proceeds as follows.

418: In section \ref{characterizationsection} we prove a characterization

419: theorem for folded triangles which implies:

420: 

421: 

422: \begin{thm}

423: \label{char}

424:  There exists a geodesic triangle $T\subset X$ whose

425: vertices are {\em special vertices} of $X$ and whose $\De$-side

426: lengths are $\al, \be, \ga^*$ if and only if there exists a Hecke

427: path $p: [0,1]\to \De$ of $\De$-length $\be$ so that

428: $$

429: p(0)=\al, p(1)=\ga.

430: $$

431: \end{thm}

432: 

433: 

434: 

435: The directed segments $\pi_\alpha$, the Hecke path $p$ and the (reversed) directed

436: segment $\pi_\gamma$ fit together to form a ``broken triangle'',

437: see Figure \ref{triangle.fig}. Here and in what follows $\pi_\la$ is the geodesic path

438: parameterizing the directed segment $\ov{ox}=\la$.

439: 

440: \begin{figure}[tbh]

441: \centerline{\epsfxsize=4.5in \epsfbox{triangle.eps}}

442: \caption{\sl A broken triangle.}

443: \label{triangle.fig}

444: \end{figure}

445: 

446: 

447: Then, by combining theorems \ref{Hecke=triangles} and \ref{char}, we

448: obtain

449: 

450: 

451: \begin{thm}

452: \label{comp}

453: $m_{\al,\be}(\ga)\ne 0$ if and only if there exists a Hecke path

454: $p: [0,1]\to \De$ of $\De$-length $\be$  so that

455: $$

456: p(0)=\al, p(1)=\ga.

457: $$

458: \end{thm}

459: 

460: This statement is an analogue of Littelmann's theorem which relates structure constants $n_{\al,\be}(\ga)$

461: of the representation ring with LS paths. The problem however is that not every Hecke path is an LS

462: path (even for the group $SL(3)$).

463: 

464: In order to prove Theorem \ref{main} our ``path model'' for the nonvanishing of

465: the Hecke structure constants must be generalized to a model

466: where Hecke paths are replaced by {\em generalized Hecke

467: paths} (we also have to replace the LS-paths by {\em generalized LS paths}). More precisely,

468: begin with a geodesic triangle $[x,y,z]\subset X$ whose vertices are special vertices of $X$ and whose $\De$-side lengths

469: are $\be, \ga^*, \al$. Now replace the geodesic segment $\ol{xy}$

470: with a certain piecewise-geodesic path $\t{p}\subset X$

471: connecting $x$ and $y$ in $X$, and

472: which is contained in the 1-skeleton of a single apartment in $X$. We then show that the projection $p$ of $\t{p}$

473: to $\De$ is a {\em generalized Hecke path}. The $\De$-length of the path $p$ still equals $d_\De(x,y)=\be$.

474: 

475: The generalized Hecke paths have the advantage over Hecke

476: paths that  {\em their break points occur only at vertices of the building}.

477: By using this observation we obtain

478: 

479: \begin{thm}\label{dilate}

480: If $p$ is a generalized Hecke path with $p(0)=0$, then $k_R \cdot p$, the image

481: of $p$ under dilation by $k_R$, is a generalized LS path of $\De$-length equal to $k_R \Length(p)$.

482: \end{thm}

483: 

484: Lastly, for {\em generalized LS paths} we prove, by modifying slightly Littelmann's arguments, the following:

485: 

486: \begin{thm}

487: \label{Li}

488: $n_{\al,\be}(\ga)\ne 0$ if and only if there exists a generalized LS path $q$ with $\De$-length $\be$ so that

489: $$

490: \al+q(1)=\ga

491: $$

492: and the concatenation $\pi_\al*q$ is contained in $\De$.

493: \end{thm}

494: 

495: 

496: Theorem \ref{main} is now obtained in 2 steps, the first of which is contained in \cite{KLM3} and the second

497: is at the heart of the present paper:

498: 

499: Step 1. It was shown in \cite[Theorem 9.17]{KLM3} that

500: $$

501: n_{N\al,N\be}(N\ga)\ne 0 \Rightarrow m_{N\al,N\be}(N\ga)\ne 0.

502: $$

503: Therefore, the vector $(N\al, N\be, N\ga^*)$ belongs to the homogeneous cone $D_3$.

504: Hence, by Part 1 of Theorem \ref{3->4}

505: we conclude that $m_{\ell\al,\ell\be}(\ell\ga)\ne 0$.

506: 

507: Step 2. $m_{\ell\al,\ell\be}(\ell\ga)\ne 0$ implies existence of a geodesic triangle

508: $[z, x, y]\subset X$ with the special vertices and $\De$-side lengths $\ell\al, \ell\be, \ell\ga^*$.

509: Then by projecting the corresponding path $\t{p}$ to $\De$ and dilating it by $k_R$ we obtain

510: a path $k\al+q(t)$ in $\De$ connecting $k\al$ to $k\ga^*$, where $q(t)$ is a generalized LS path of

511: $\De$-length $k\be$. (Recall that $k=\ell^2$.) Therefore, by appealing to Theorem \ref{Li}, we see that

512: $$

513: n_{k\al,k\be}(k\ga)\ne 0

514: $$

515: which concludes the proof.

516: 

517: 

518: 

519: 

520: 

521: \medskip

522: This paper is organized as follows. Preliminary material is

523: discussed in section \ref{prelim}, where we review the concepts of

524: Coxeter complexes, buildings and piecewise-linear paths in buildings, as well as

525: the {\em generalized distances} in buildings. In section

526: \ref{sectionchains} we define the notion of {\em chains}, which is

527: essentially due to Littelmann. This concept allows one to define

528: both LS paths and Hecke paths, as well as to relate Hecke paths and

529: foldings of geodesic paths, which is discussed in the next section.

530: 

531: 

532: The main technical tool of this paper is the concept of {\em

533: folding} of geodesics in a building into an apartment, which is done

534: via retraction of the entire building to an apartment or chamber.

535: \footnote{A similar idea is used by S. Gaussent and P. Littelmann in

536: \cite{GL}, where they fold galleries in a building to galleries in

537: an apartment.} Properties of foldings are discussed in section

538: \ref{sectionfolding}. In section \ref{euclideanfolding} we prove

539: that the image of each geodesic segment in $X$ under {\em folding}

540: $f: X\to \De$ of $X$ to a Weyl chamber, is a Hecke path, Theorem \ref{chaincondition}.

541: We then prove a partial converse to this result, i.e. that each

542: Hecke path which satisfies the {\em simple chain condition} can be

543: unfolded in $X$. We also find a necessary and sufficient condition

544: for  unfolding of a path $p$ which is {\em local}, i.e. it depends

545: only on germs of the path $p$ at its break-points.

546: 

547: 

548: 

549: 

550: In section \ref{LSpolygons} we review Littelmann's

551: path model for the representation theory of complex semisimple Lie

552: groups, in particular we discuss LS paths and generalized LS paths

553: as well as raising and lowering operators.

554: 

555: 

556: We use {\em approximation of LS paths by paths satisfying the simple

557: chain condition} to unfold LS paths in $X$, Theorem \ref{LSfolded}

558: in section \ref{unfoldingLS}. Although there are Hecke paths which

559: are not LS paths, since the unfolding condition is local, by restricting

560: the root system we reduce the general unfolding problem to the case

561: of the LS paths. We thus establish that a path in $\De$ is

562: unfoldable in $X$ it and only if it is a Hecke path, this is done in

563: section \ref{characterizationsection}, Theorem

564: \ref{chain-unfolding}. The reader interested only in the proof of

565: the saturation theorem can omit this section.

566: 

567: 

568: 

569: We prove the saturation theorem in section \ref{saturationsection},

570: Corollary \ref{corQ3->Q4}. As we stated above, the idea of the proof is to replace Hecke

571: paths with piecewise-linear paths contained in the 1-skeleton of the Euclidean

572: Coxeter complex. We show that dilation by $k_R$ of such a path

573: results in a {\em generalized LS path} which in turn suffices for

574: finding nonzero invariant vectors in triple tensor products.

575: 

576: \paragraph{Hecke paths and folded galleries.} It is interesting to ask what the relationship is between our paper

577: using Hecke paths and the results of S. Gaussent and P.

578: Littelmann in \cite{GL} and others using positively folded galleries:

579: 

580: (a) {\em  Hecke paths} correspond to the {\em positively folded galleries} defined in \cite{GL}.

581: However this correspondence can be established only {\bf after} the folding--unfolding results of the present

582: paper are proven.

583: Therefore it appears that

584: one cannot prove our results characterizing Hecke paths as projections of geodesic segments

585: in building using the results of \cite{GL} and vice versa.

586: 

587: 

588: (b) In \cite{Schwer}, C. Schwer has used the gallery approach to

589: compute the Hecke structure constants $m_{\alpha,\beta}(\gamma)$.

590: In an earlier version of this paper we applied our theory to

591: compute some Hecke structure coefficients. Thus  both Hecke path

592: and gallery models can be used  to compute the structure constants

593: of the spherical Hecke algebra.

594: 

595: (c) One can prove that

596: $$n_{\al,\be}(\ga)\ne 0\Rightarrow m_{\al,\be}(\ga)\ne 0$$

597: \cite[Theorem 9.17]{KLM3} using both Hecke paths (as it is done in section \ref{characterizationsection})

598: and positively folded galleries.

599: 

600: (d) There does not  seem to be a direct way to carry over our

601: proof of the main Theorem \ref{main} or part 2 of Theorem

602: \ref{3->4} to a proof using positively folded galleries. It is

603: clear from the above, that one of the critical steps in our

604: argument is provided by Theorem \ref{dilate} stating that dilation

605: by $k_R$ converts generalized Hecke paths to generalized LS paths.

606: But there does not appear to be a way to ``stretch'' a folded

607: gallery by the factor $k_R$ --- or, more generally, to produce an

608: LS gallery from a positively folded gallery. This is an advantage

609: of the Hecke paths, which are {\em geometric} objects, over

610: positively folded galleries, which are {\em combinatorial}

611: objects. Even when $k_R=1$, there is no obvious (at least to us)

612: reason why existence of a positively folded gallery would imply

613: existence of an LS gallery. We note that if $k_R \neq 1$ then the

614: existence of a positively folded gallery does not imply the

615: existence of an LS gallery, there are counterexamples for $SO(5)$

616: and $G_2$.

617: 

618: \medskip

619: {\bf Acknowledgments.} This paper grew out of our joint work

620: \cite{KLM1, KLM2, KLM3} with Bernhard Leeb. We are grateful to him

621: for all his ideas that he contributed to those papers, especially

622: the idea of folding that came out of his contribution to

623: \cite{KLM3}. We are also grateful to Tom Haines for valuable discussions. We thank the referee for his/her remarks.

624: Part of the work on this paper was done when the first author was

625: visiting the Max Plank Institute (Bonn), during this work he was also supported by

626: the NSF grant DMS-02-03045 and DMS-04-05180. The second author was

627: supported by the NSF grant DMS-01-04006 and DMS-04-05606. Together, the authors were suported by the NSF grant

628: DMS-05-54349. The authors gratefully acknowledge support of these institutions.

629: 

630: 

631: \tableofcontents

632: 

633: \section{Definition and notation}

634: \label{prelim}

635: 

636: 

637: 

638: 

639: 

640: 

641: 

642: 

643: \subsection{Root systems and Coxeter complexes}

644: 

645: 

646: A (discrete, nonnegatively curved) {\em Coxeter complex} is a pair

647: $(A,W)$, where $A$ is either a Euclidean space or the unit sphere

648: and $W$ is a discrete reflection group acting on $A$. The {\em rank} of

649: the Coxeter complex is the dimension of $A$. The group $W$ is called {\em Weyl group} of the Coxeter complex.

650: It is called an {\em affine Weyl group} if $A$ is a Euclidean space.

651: 

652: 

653: %\begin{rem}

654: %There are also Euclidean Coxeter complexes which are nondiscrete (i.e. the

655: %group $W$ is nondiscrete, but has discrete linear part), as well as hyperbolic Coxeter complexes,

656: %where one uses reflection groups of isometries of the hyperbolic

657: %space. However we do not need them in the present paper.

658: %\end{rem}

659: 

660: An isomorphism of Coxeter complexes $(A,W), (A',W')$ is an isometry $\iota: A\to A'$ so that

661: $$

662: \iota W \iota^{-1}=W'.

663: $$

664: 

665: {\em Walls} in

666: $(A,W)$ are fixed point sets of reflections $\tau\in W$. A point

667: $x\in A$ is called {\em regular} if it does not belong to any wall

668: and {\em singular} otherwise. The closure of each connected

669: component of the set of regular points is called an {\em alcove} in

670: the Euclidean case and a {\em chamber} in the spherical case.

671: 

672: In the case when $W$ acts cocompactly on $A$, each alcove (resp. chamber) is a product (resp. join) of simplices

673: and $(A,W)$ determines structure of a {\em polysimplicial complex} on $A$. In general

674: there exists a totally-geodesic subspace

675: $A'\subset A$ which is $W$--invariant and such that $A'/W$ is compact.

676: Therefore each alcove in $A$ is a product

677: of simplices and a Euclidean subspace in $A$. Thus much of the discussion

678:  of Euclidean Coxeter complexes can be reduced to the case when $A/W$ is compact.

679: 

680: \begin{rem}

681: \label{tri}

682: A  triangulation of a fundamental alcove (chamber) in $A$ determines a

683: $W$--invariant simplicial complex on $A$. Thus

684: we can always think of $A$ as a simplicial complex.

685: \end{rem}

686: 

687: A {\em half-apartment} in $A$ is the closure of a

688: connected component of $A\setminus H$, where $H$ is a wall in $A$.

689: 

690: 

691: ``Most'' Coxeter complexes are associated with root systems as we

692: describe below. Suppose that $R$ is a root system on a vector space

693: $V$ (i.e. each element $\al\in R$ is a linear functional on $V$).

694: The {\em rank} of $R$ is the number of simple roots in $R$, i.e. is the rank of the free abelian

695: subgroup in $V^*$ generated by $R$.

696: Let $A$ denote the Euclidean affine space corresponding to $V$. This

697: data defines a finite Coxeter group $W_{sph}$, which is a reflection

698: group generated by reflections in the hyperplanes $H_{\al}=\{ x:

699: \al(x)=0\}$. {\em Weyl chambers} of $W_{sph}$ are closures of the

700: connected components of the complement to

701: $$

702: \cup_{\al\in R} H_\al.

703: $$

704: In what follows we fix a {\em positive Weyl chamber} $\De$, it

705: determines the subset of positive roots  $R^+\subset R$ and of {\em

706: simple roots} $\Phi\subset R^+$. We also have the group of coweights

707: $P(R^\vee)$ associated with $R$:

708: $$

709: \la\in P(R^\vee) \iff \forall \al\in R, \quad \al(\la)\in \Z.

710: $$

711: Let $W_{aff}$ denote the {\em affine Coxeter group} determined by

712: the above data, this group is generated by reflections in the

713: hyperplanes ({\em affine walls})

714: $$

715: H_{\al,t}=\{\al(x)=  t\}, t\in \Z.$$

716: 

717: Given a vector $\ga\in \De$, we define the {\em contragredient vector} $\ga^*$

718: as $w_0(-\ga)$, where $w_0$ is the longest element of $W$.

719: In other words, $\ga^*$ is the intersection of the $W$-orbit

720: $W\cdot (-\ga)$ with the chamber $\De$.

721: 

722: 

723: The translation subgroup in $W_{aff}$ is the {\em coroot lattice}

724: $Q(R^\vee)$,  it is generated by the coroots $\al^\vee, \al\in R$.

725: The group  $P(R^\vee)$ is the normalizer of $W_{aff}$ in the group

726: of translations $V$. Note that $V/P(R^\vee)$ is compact. The

727: dimension of this quotient is the same as the dimension of $V$

728: provided that $rank(R)$ equals the dimension of $V$.

729: 

730: 

731:  The {\em special vertices} of a Euclidean Coxeter complex are the

732: points whose stabilizer in $W_{aff}$ is isomorphic to $W_{sph}$.

733: Equivalently, they are the points in the $P(R^\vee)$-orbit of the

734: origin.

735: 

736: 

737: \begin{rem}

738: If $A/W_{aff}$ is compact, the  vertices of $(A,W_{aff})$ are the

739: vertices of the polysimplicial complex determined by $W_{aff}$.

740: \end{rem}

741: 

742: Let $A^{(0)}$ denote the {\em vertex

743: set} of $(A, W_{aff})$, which consists of points of maximal

744: intersection of walls in $A$. If $R$ spans $V^*$, the set $A^{(0)}$

745: equals the vertex set of the polysimplicial complex in $A$ defined

746: by tessellation of $A$ via alcoves of $W_{aff}$.

747: 

748: Given a Coxeter complex $(A,W)$ and a point $x\in A$ we define a new

749: Coxeter complex $(S_x,W_x)$ where $S_x$ is the unit tangent sphere

750: at $x$ and $W_x$ is the stabilizer of $x$ in $W$.

751: 

752: 

753: For a nonzero vector $\nu\in V$ we let $\bar{\nu}:= \nu/|\nu|$

754: denote the {\em normalization} of $\nu$. We define {\em rational}

755: elements of the unit sphere $S$ to be the unit vectors of the form

756: $$

757: \eta=\bar\nu, \quad \nu\in P(R^\vee).

758: $$

759: The next lemma follows immediately from compactness of

760: $V/P(R^\vee)$:

761: 

762: 

763: \begin{lem}

764: \label{density} Rational points are dense in $S$.

765: \end{lem}

766: 

767: 

768: Suppose that $(A, W)$ is a Euclidean Coxeter complex. A {\em

769: dilation} of $(A, W)$ is a dilation $h$ (i.e. a composition of

770: translation and similarity $v\mapsto \la v, \la>0$) in the affine

771: space $A$ so that

772: $$

773: h W h^{-1} \subset W.

774: $$

775: We let $Dil(A, W)$ denote the semigroup of dilations of the complex

776: $(A, W)$. We will refer to the number $\la$ as the {\em conformal

777: factor} of the dilation $h$.

778: 

779: 

780: Given a point $x\in A$ and a dilation $h\in Dil(A,W)$, we can define

781: a new spherical Coxeter complex $(S_x, W'_x)$ on the unit tangent

782: sphere $S_x$ at $x$ via pull-back

783: $$

784: W'_x:= h^*(W_{t(x)}),

785: $$

786: where $W_{h(x)}$ is the stabilizer of $h(x)$ in $W$.

787: 

788: 

789: 

790: 

791: \begin{defn}

792: \label{relation} Suppose that $W$ is a finite Coxeter group acting

793: on a vector space $V$. Define a (nontransitive) relation $\sim_W$ on

794: $V\setminus \{0\}$ by

795: $$

796: \mu \sim_W \nu \iff

797: $$

798: \centerline{$\mu, \nu$ belong to the same Weyl chamber of $W$.}

799: 

800: 

801: \noindent We will frequently omit the subscript $W$ in this

802: notation.

803: \end{defn}

804: 

805: 

806: 

807: 

808: \begin{defn}

809: \cite[page 514]{Littelmann2}

810: \label{porder} We say that

811: nonzero vectors $\nu, \mu\in V$ satisfy $\nu \rhd_W \mu$ (for short,

812: $\nu \rhd \mu$) if for each positive root $\al$,

813: $$

814: \al(\nu)< 0 \Rightarrow \al(\mu)\leq 0.

815: $$

816: \end{defn}

817: 

818: \begin{lem}

819: \label{difference}

820: Suppose that $\nu, \mu\in P(R^\vee)$, $w\in W=W_{aff}$ is such that $w(\nu)=\mu$. Then

821: $$

822: \mu-\nu\in Q(R^\vee).

823: $$

824: \end{lem}

825: \proof The mapping $w$ is a composition of reflections $\tau_i\in W$.

826: Therefore it suffices to prove the assertion in case when $w$

827: is a reflection $\tau$. This reflection is a composition of

828: a translation $t$  and a reflection $\si\in W_o$.

829: The translation $t$ belongs to the translation subgroup $Q(R^\vee)$ of $W$,

830: therefore it suffices to consider the case when $\tau=\si\in W_o$.

831: Then $\tau=\tau_\be$, where $\be$ is a root and we have

832: $$

833: \mu-\nu= -\be(\nu)\be^\vee.

834: $$

835: Since $\be(\nu)\in \Z$ and $\be^\vee\in Q(R^\vee)$, the assertion follows. \qed

836: 

837: 

838: 

839: 

840: 

841: \subsection{Paths}

842: \label{paths}

843: 

844: Suppose that $A, V, W_{aff}$, etc., are as in the previous section.

845: 

846: 

847: Let $\tilde{\mathcal P}$ denote the set of all piecewise-linear paths $p: [a,b]\to

848: V$.  We will be identifying paths that differ by

849: orientation-preserving re-parameterizations $[a,b]\to [a',b']$.

850: Accordingly, we will always (re)parameterize a piecewise-linear path with

851: constant speed. We let $p'_-(t), p'_+(t)$ denote the derivatives of the function $p$

852: from the left and from the right. The space $\t{\mathcal P}$ will be

853: given the topology of uniform convergence.

854: 

855: 

856: If $p, q: [0,1]\to A$ are piecewise-linear paths in a simplicial complex such that

857: $p(1)=q(0)$, we define their {\em composition} $r=p\cup q$ by

858: $$

859: r(t)=\left\{

860: \begin{array}{c}

861: p(t), ~~t\in [0,1],\\ q(t-1), ~~t\in [1, 2].

862:  \end{array}\right.

863: $$

864: Let ${\mathcal P}\subset \tilde{\mathcal P}$ denote the set of paths

865: $p: [0,1]\to V$ such that $p(0)=0$. Given a path $p\in {\mathcal P}$

866: we let $p^*\in {\mathcal P}$ denote the {\em reverse} path

867: $$

868: p^*(t)= p(1-t)-p(1).

869: $$

870: For a vector $\la\in V$ define a

871: geodesic path $\pi_\la\in {\mathcal P}$ by

872: $$

873: \pi_\la(t)=t\la, \quad t\in [0,1].

874: $$

875: Given two paths $p_1, p_2\in {\mathcal P}$ define their {\em

876: concatenation} $p=p_1* p_2$ by

877: $$

878: p(t)= \left\{

879: \begin{array}{c}

880: p_1(2t), \quad t\in [0, 1/2],\\ p_1(1)+ p_2(2t-1), \quad t\in [1/2,

881: 1].

882:  \end{array}\right.

883: $$

884: 

885: \medskip

886: Suppose that $p\in {\mathcal P}$ and $J=[a,b]$ is nondegenerate

887: subinterval in $I=[0,1]$. We will use the notation  $p|J\in

888: \tilde{\mathcal P}$ to denote the function-theoretic restriction of

889: $p$ to $[a,b]$. We will use the notation $p|_J$ to denote the path

890: in ${\mathcal P}$ obtained from $p|J$ by pre-composing $p|J$ with an

891: increasing linear bijection $\ell: I\to J$ and post-composing it

892: with the translation by the vector $-p(a)$.

893: 

894: \medskip

895: Fix a positive Weyl chamber $\De\subset V$; this determines the set of positive roots $R^+\subset R$,

896: the set of simple roots $\Phi\subset R^+$. We define the

897: subset ${\mathcal P}^+\subset {\mathcal P}$ consisting of the paths

898: whose image is contained in $\De$.

899: 

900: For a path $p\in {\mathcal P}$ and a positive root $\al\in R^+$ define the

901: {\em height} function

902: $$

903: h_\al(t)= \al(p(t))

904: $$

905: on $[0,1]$. Let $m_\al=m_\al(p)\in \R$ denote the minimum of

906: $h_\al$. Clearly $m _\al(p)\le 0$ for all $p\in {\mathcal P}$.

907:  We define the set of ``integral paths''

908: $$

909: {\mathcal P}_\Z:= \{ p\in {\mathcal P}: \forall \al\in \Phi,

910: m_\al(p)\in \Z\}.

911: $$

912: More restrictively, we define the set ${\mathcal P}_{\Z,loc}$ of paths $p\in {\mathcal

913: P}$  which satisfy the following {\em local integrality condition}:

914: 

915: 

916: For each simple root $\al\in \Phi$ the function $h_\al$  takes

917: integer values at the points of local minima.

918: 

919: 

920: 

921: 

922: 

923: 

924: 

925: \subsection{The saturation factors associated to a root system}

926: \label{kfactors}

927: 

928: 

929: In this section we define and compute {\em saturation factors}

930: associated with root systems.  Let $o\in A$ be a

931: special vertex, which we will identify with

932: $0\in V$.

933: 

934: 

935: \begin{dfn}

936: We define the {\em saturation factor} $k_R$ for the root system $R$

937: to be the least natural number $k$ such that $k\cdot A^{(0)}\subset

938: P(R^\vee)\cdot o$. The numbers $k_R$ for the irreducible root

939: systems are listed in the table (\ref{ta}).

940: \end{dfn}

941: 

942: 

943: Note that the condition that $k\cdot A^{(0)}\subset P(R^\vee)\cdot

944: o$ is equivalent to that each point of $k\cdot A^{(0)}$ is a special

945: vertex.

946: 

947: 

948: 

949: 

950: Below we explain how to compute the saturation factors $k_R$

951: following \cite{KLM3}. First of all, it is clear that if the root

952: system $R$ is reducible and $R_1,...,R_s$ are its irreducible

953: components, then $k_R= LCM(k_{R_1},...,k_{R_s})$, where  $LCM$

954: stands for the {\em

955:   least common multiple}. Henceforth we can assume that the system

956: $R$ is reduced, irreducible and has rank $n=\dim(V)$. Then the

957: affine Coxeter group $W_{aff}$ acts cocompactly on $A$ and its

958: fundamental domain (a {\em Weyl alcove}) is a simplex.

959: 

960: 

961: Let $\{\al_1,...,\al_{\ell}\}$ be the collection of simple roots in $R$

962: (corresponding to the positive Weyl chamber $\De$) and $\theta$ be

963: the highest root. Then

964: 

965: 

966: \begin{equation}

967: \label{highestroot} \theta= \sum_{i=1}^{\ell} m_i \al_i.

968: \end{equation}

969: 

970: 

971: We have

972: 

973: 

974: \begin{lem}

975: \cite[Section 2]{KLM3}

976: \label{kexists} $k_R=

977: LCM(m_1,...,m_n)$.

978: \end{lem}

979: 

980: 

981: Below is the list of saturation factors:

982: 

983: 

984: \begin{equation}

985: \label{ta}

986: \begin{array}{|c|c|c|}

987: \hline ~ & ~ & ~  \\ \hbox{Root system} & \theta &  k_R \\ \hline

988: A_\ell & \al_1+...+\al_\ell & 1\\ \hline B_\ell & \al_1+

989: 2\al_2+...+2\al_\ell & 2\\ \hline C_\ell & 2\al_1+

990: 2\al_2+...+2\al_{\ell-1}+\al_\ell & 2\\ \hline D_\ell & \al_1+

991: \al_2+\al_3+ 2\al_4+...+2\al_\ell & 2 \\ \hline G_2 & 3\al_1+2\al_2&

992: 6 \\ \hline F_4 & 2\al_1+ 3\al_2+4\al_3 +2\al_4 &  12\\ \hline E_6 &

993: \al_1+\al_2+ 2\al_3+2\al_4+2\al_5+3\al_6 &  6 \\ \hline E_7 & \al_1+

994: 2\al_2+2\al_3+2\al_4+3\al_5+&  12 \\ ~ & +3\al_6+ 4\al_7 & ~ \\

995: \hline E_8 & 2\al_1+2\al_2+3\al_3+3\al_4+ 4\al_5+ &  60\\ ~&

996: +4\al_6+ 5\al_7 +6\al_8 & ~  \\ \hline

997: \end{array}

998: \end{equation}

999: 

1000: 

1001: \subsection{Buildings}

1002: 

1003: 

1004: Our discussion of buildings follows \cite{KleinerLeeb}. We refer the reader

1005: to \cite{Brown}, \cite{Ronan}, \cite{Rousseau} for the more combinatorial discussion.

1006: 

1007: 

1008: Fix a spherical or Euclidean (discrete) Coxeter complex $(A,W)$, where $A$ is a

1009: Euclidean space $E$ or a unit sphere $S$ and $W=W_{aff}$ or $W=W_{sph}$

1010: is a discrete Euclidean or a spherical Coxeter group acting on $A$.

1011: 

1012: A metric space $Z$ is called {\em geodesic} if any pair of points $x, y$ in $Z$ can be connected by a

1013: geodesic segment $\ol{xy}$.

1014: 

1015: 

1016: Let $Z$ be a metric space. A {\em geometric structure} on $Z$ {\em

1017: modeled on} $(A,W)$ consists of an atlas of isometric embeddings

1018: $\varphi:A\embed Z$ satisfying the following compatibility

1019: condition: For any two charts $\varphi_1$ and $\varphi_2$, the

1020: transition map $\varphi_2^{-1}\circ\varphi_1$ is the restriction of

1021: an isometry in $W$. The charts and their images,

1022: $\varphi(A)=a\subset Z$, are called {\em apartments}. We will

1023: sometimes refer to $A$ as the {\em model apartment}. We will require

1024: that there are {\em plenty of apartments} in the sense that any two

1025: points in $Z$ lie in a common apartment. All $W$-invariant notions

1026: introduced for the Coxeter complex $(A, W)$, such as rank, walls, singular

1027: subspaces, chambers etc., carry over to geometries modeled on $(A,

1028: W)$. If $a, a'\subset X$ are alcoves (in the Euclidean case) or chambers

1029: (in the spherical case) then there exists an apartment $A'\subset X$ containing $a\cup a'$:

1030: Just take regular points $x\in a, x'\in a'$ and an apartment $A'$ passing through $x$ and $x'$.

1031: 

1032: 

1033: 

1034: A geodesic metric space $Z$ is said to be a $CAT(\kappa)$-space if

1035: {\em geodesic triangles in $Z$ are ``thinner'' than geodesic

1036: triangles in a simply-connected complete surface of the constant

1037: curvature $\kappa$}. We refer the reader to \cite{Ballmann}  for the

1038: precise definition. Suppose that $Z$ is a (non-geodesic) metric space with the discrete metric:

1039: $$

1040: d(x,y)=\pi \iff x\ne y.

1041: $$

1042: We will regard such a space as a $CAT(1)$ space as well.

1043: 

1044: \begin{dfn}

1045: A {\em spherical building} is a $CAT(1)$-space modeled on a spherical

1046: Coxeter complex.

1047: \end{dfn}

1048: 

1049: 

1050: Spherical buildings have a natural structure of polysimplicial

1051: piecewise spherical complexes. We prefer the geometric to the

1052: combinatorial view point because it appears to be more appropriate

1053: in the context of this paper.

1054: 

1055: 

1056: 

1057: 

1058: \begin{defn}

1059: A Euclidean building is a $CAT(0)$-space modeled on a (discrete)

1060: Euclidean Coxeter complex.

1061: \end{defn}

1062: 

1063: 

1064: 

1065: 

1066: A building is called {\em thick} if every wall is an intersection of

1067: apartments. A non-thick building can always be equipped with a

1068: natural structure of a thick building by reducing the Coxeter group  \cite{KleinerLeeb}.

1069: 

1070: \medskip

1071: Let $\K$ be a local field with a (discrete) valuation $\nu$ and valuation ring $\O$.

1072: Throughout this paper, we will be working with $\Q$-split reductive algebraic groups (i.e. Chevalley groups) $\ul{G}$ over $\Z$.

1073: We refer the reader to \cite{Borel, Demazure} for a detailed discussion of such groups. A reader

1074: can think of $\K=\Q_p$ (the $p$-adic numbers) and a {\em classical} Chevalley group $\ul{G}$,

1075: e.g. $GL(n)$ or $Sp(n)$, in which case  Bruhat-Tits buildings

1076: below can be described rather explicitely, see e.g. \cite{Garrett}.

1077: 

1078: Given a group $\ul{G}$ as above, we get a {\em nonarchimedean Lie

1079: group} $G=\ul{G}(\K)$, to which we can associate a Euclidean building (a

1080: Bruhat-Tits building) $X=X_G$. We refer the reader to \cite{BT},

1081: \cite{KLM3} and \cite{Rousseau} for more detailed discussion of the

1082: properties of $X$. Here we only recall that:

1083: 

1084: 

1085: 1. $X$ is thick and locally compact.

1086: 

1087: 

1088: 2. $X$ is modeled on a Euclidean Coxeter complex $(A,W_{aff})$ whose

1089: dimension equals the rank of $\ul{G}$, and the root system is

1090: isomorphic to the root system of $\ul{G}$.

1091: 

1092: 3. $X$ contains a special vertex $o$ whose stabilizer in $G$ is $\ul{G}(\O)$.

1093: 

1094: 

1095: \begin{ex}

1096: Let $X$ be a (discrete) Euclidean building, consider the  {\em

1097: spaces of directions} $\Si_x X$. We will think of this space as the

1098: space of germs of non-constant geodesic segments $\ol{xy}\subset X$.

1099: As a polysimplicial complex $\Si_xX$ is just the link of the point

1100: $x\in X$. The space of directions has the structure of a spherical

1101: building modeled on $(S,W_{sph})$, which is thick if and only if $x$

1102: is a special vertex of $X$  \cite{KleinerLeeb}. The same applies

1103: in the case when $X$ is a spherical building.

1104: \end{ex}

1105: 

1106: If $X$ is a Euclidean building modeled on $(A,W)$, for each point

1107: $x\in X$ the space of directions $\Si_x(X)$ has {\em two structures}

1108: of a spherical building:

1109: 

1110: 

1111: 1. The {\em restricted building structure} which is modeled on the

1112: Coxeter complex $(S, W_{x})$, where $S=S_x(A)$ is the unit tangent

1113: sphere at $x$ and $W_{x}$ is the stabilizer of $x$ in the Coxeter

1114: group $W$. This building structure is thick.

1115: 

1116: 

1117: 2. The {\em unrestricted building structure} which is modeled on the

1118: Coxeter complex $(S, W_{sph})$, where $S=S_x(A)$ is the unit tangent

1119: sphere at $x$ and $W_{sph}$ is the linear part of the affine Coxeter

1120: group $W_{aff}$. This building structure is not thick, unless $x$ is

1121: a special vertex.

1122: 

1123: 

1124: Let $B$ be a spherical building modeled on a spherical Coxeter

1125: complex $(S, W_{sph})$. We say that two points $x, y\in B$ are

1126: antipodal, if $d(x,y)=\pi$; equivalently, they are antipodal points

1127: in an apartment $S'\subset B$ containing both $x$ and $y$. The

1128: quotient map $S\ra S/W_{sph}\cong\De_{sph}$ induces a canonical

1129: projection $\theta:B\ra \De_{sph}$ folding the building onto its

1130: model Weyl chamber. The $\theta$-image of a point in $B$ is called

1131: its {\em type}.

1132: 

1133: 

1134: \begin{rem}

1135: To define $\theta(x)$ pick an apartment $S'$ containing $x$ and a

1136: chart $\phi: S\to S'$. Then $\theta(x)$ is the projection of

1137: $\phi^{-1}(x)$ to  $S/W_{sph}\cong\De_{sph}$. We note that this is

1138: clearly independent of $S'$ and $\phi$.

1139: \end{rem}

1140: 

1141: 

1142: \begin{lem}

1143: \label{thetaproperties}

1144:  1. If $h: A\to A'$ is an isomorphism of apartments in $B$ (i.e.

1145: ${\phi'}^{-1}\circ h\circ \phi\in W$) then $\theta\circ h=\theta$.

1146: 

1147: 

1148: 2. If $x, x'\in B$ which belong to apartments $A, A'$ respectively

1149: and $-x\in A, -x'\in A'$ are antipodal to $x, x'$, then

1150: $\theta(x)=\theta(x')$ implies $\theta(-x)=\theta(-x')$.

1151: \end{lem}

1152: \proof (1) is obvious, so we prove (2). Pick an isomorphism $h: A\to

1153: A'$. Then (since $\theta(x)=\theta(x')$) there exists $w\in W\acts

1154: A'$ such that $w(h(x))=x'$. Hence $w\circ h(-x)=-w\circ(x)=-x'$. The

1155: claim now follows from 1. \qed

1156: 

1157: 

1158: 

1159: 

1160: 

1161: 

1162: \medskip

1163: We will regard {\em $n$-gons} $P$ in a building $X$ as maps $\nu:

1164: \{1,...,n\}\to X$, $\nu(i)=x_i$, where $x_i$ will be the vertices of

1165: $P$.  If $rank(X)\ge 1$ we can connect the consecutive vertices

1166: $x_i, x_{i+1}$ by shortest geodesic segments $\ol{x_i

1167: x_{i+1}}\subset X$ thus creating a {\em geodesic polygon} $[x_1,

1168: x_2,\ldots, x_n]$ in $X$ with the edges $\ol{x_i x_{i+1}}$. Observe

1169: that in case $x_i, x_{i+1}$ are antipodal, the edge $\ol{x_i

1170: x_{i+1}}$ is not unique.

1171: 

1172: 

1173: 

1174: 

1175: We say that two subsets $F, F'$ in a building $X$ are {\em

1176: congruent} if there exist apartments $A, A'$ in $X$ containing $F,

1177: F'$ resp., and an isomorphism $A\to A'$ of Coxeter complexes which

1178: carries $F$ to $F'$.

1179: 

1180: 

1181: \begin{conv}

1182: Suppose that $X$ is a spherical building. We will be considering

1183: only those geodesic triangles $T$ in $X$ for which the length of

1184: each geodesic side of $T$ is $\le \pi$.

1185: \end{conv}

1186: 

1187: 

1188: 

1189: 

1190: 

1191: \medskip

1192: Let $X, Y$ be buildings and $f: X\to Y$ a continuous map satisfying

1193: the following: For each alcove (in Euclidean case) or spherical

1194: chamber (in the spherical case) $a\subset X$, the image $f(a)$ is

1195: contained in an apartment of $Y$ and the restriction $f|a$ is

1196: an isometry or similarity. Then we call $f$ {\em differentiable} and

1197: define the {\em derivative} $df$ of $f$ as follows. Given a point

1198: $x\in X$ and $y=f(x)$, the derivative $df_x$ is a map $\Si_x(X)\to

1199: \Si_y(Y)$. For each $\xi\in \Si_x(X)$ let $\ol{xz}\subset X$ be a

1200: geodesic segment whose interior consists of regular points only and

1201: so that $\xi$ is the unit tangent vector to  $\ol{xz}$. Then $f$

1202: sends $\ol{xz}$ to a nondegenerate geodesic segment $\ol{y f(z)}$

1203: contained in an apartment $A\subset Y$. Then we let $df_x(\xi)\in

1204: \Si_y(Y)$ be the unit tangent vector to $\ol{y f(z)}$.

1205: 

1206: We will be also using the above definition in the setting when a building $Y$ is a Euclidean

1207: Coxeter complex $(A,W)$ and $h\in Dil(A,W)$. Then after letting $X:= h^*(A,W)$ we get an isometry $h: X\to Y$.

1208: 

1209: 

1210: \begin{conv}

1211: Throughout the paper we will be mostly using roman letters $x, y,

1212: z$, etc., to denote points in Euclidean buildings and Greek

1213: letters $\xi, \eta, \zeta$, etc., to denote points in spherical

1214: buildings. Sometimes however (e.g., in section \ref{converting})

1215: we will be working simultaneously with a spherical building $X$

1216: and its links $\Si_x(X)$, which are also spherical buildings. In

1217: this case we will use roman letters for points in $X$ and Greek

1218: letters for points  in $\Si_x(X)$.

1219: \end{conv}

1220: 

1221: \subsection{Generalized distances and lengths in buildings}

1222: \label{distances}

1223: 

1224: Let $(A,W)$ be a spherical or Euclidean Coxeter complex. The

1225: complete invariant of a pair of points $(x,y)\in A^2$ with respect

1226: to the action $W\acts A$, is its image $d_{ref}(x,y)$ under the

1227: canonical projection to $A\times A/W$. Following \cite{KLM2} we call $d_{ref}(x,y)$ the

1228: {\em refined distance from $x$ to $y$}. This notion carries over to

1229: buildings modeled on the Coxeter complex $(A,W)$: For a pair of

1230: points $(x,y)$ pick an apartment $A'$ containing $x,y$ and, after

1231: identifying $A'$ with the model apartment $A$, let $d_{ref}(x,y)$ be

1232: the projection of this pair to $A\times A/W$.

1233: 

1234: 

1235: 

1236: 

1237: If points $\xi, \eta$ in a spherical building are antipodal we will

1238: use $\pi$ for the refined distance $d_{ref}(\xi,\eta)$: This does

1239: not create much ambiguity since given apartment contains unique

1240: point antipodal to $\xi$.

1241: 

1242: 

1243: 

1244: 

1245: 

1246: 

1247: In the case of Euclidean Coxeter complexes there is an extra

1248: structure associated with the concept of refined length. Given a

1249: Euclidean Coxeter complex $(A, W_{aff})$, pick a special vertex

1250: $o\in A$. Then we can regard $A$ as a vector space $V$, with the

1251: origin $0=o$. Let $\De\subset A$ denote a Weyl chamber

1252:  of $W_{sph}$, the tip of $\De$ is at $o$.

1253: 

1254: 

1255: Then following \cite{KLM2}, we define the {\em $\De$-distance} between points of $(A,

1256: W_{aff})$ by composing $d_{ref}$ with the natural forgetful map

1257: \[ A\times A/W_{aff}\ra A/W_{sph}\cong\De .\]

1258: To compute the $\De$-distance $d_\De (x,y)$ we regard the oriented

1259: geodesic segment $\ol{xy}$ as a vector in $V$ and project it to

1260: $\De$. Again, the concept of $\De$-distance, carries over to the

1261: buildings modeled on $(A,W_{aff})$.

1262: 

1263: 

1264: \begin{defn}

1265: \label{maindef} Let $X$ be a thick Euclidean building. Define the

1266: set $D_n(X)\subset \De^n$ of $\De$-side lengths which occur for

1267: geodesic $n$-gons in $X$.

1268: \end{defn}

1269: 

1270: 

1271: It is one of the results of \cite{KLM2} that $D_n:=D_n(X)$ is a convex

1272: homogeneous polyhedral cone in $\De^n$, which depends only on $(A,W_{sph})$.

1273: The polyhedron $D_3$ in Theorem \ref{reform} is the polyhedron $D_3(X)$.

1274: The set of {\em stability} inequalities defining $D_n$

1275: is determined in \cite{KLM1} and

1276: \cite{BS}.

1277: 

1278: 

1279: 

1280: 

1281: \begin{thm}

1282: \label{klm2}\cite[Corollary 8.4]{KLM3}.

1283:  Let $X$ be a thick

1284: Euclidean building modeled on $(A,W_{aff})$. Suppose that $\al, \be,

1285: \ga\in P(R^\vee)$, $\al+\be+\ga\in Q(R^\vee)$ and $(\al,\be,\ga)\in

1286: D_3(X)$. Then there exists a geodesic triangle $T\subset X$ whose

1287: vertices are vertices of $X$ and the $\De$-side lengths are $\al,

1288: \be, \ga$.

1289: \end{thm}

1290: 

1291: 

1292: 

1293: \medskip

1294: Suppose that $p$ is a piecewise-linear path in a Euclidean building $X$. We say

1295: that $p$ is a {\em billiard path} if for each $t, s\in [0,1]$ the

1296: tangent vectors $p'(t), p'(s)$ have the same projection to the

1297: chamber $\De$ in the model apartment. If $p$ is a path which is the

1298: composition

1299: $$

1300: \ol{x_0 x_1}\cup ...\cup \ol{x_{m-1} x_{m}}

1301: $$

1302: of geodesic paths, then the $\De$-length of $p$ is defined as

1303: $$

1304: \Length(p):=\sum_{i=1}^m d_\De (x_{i-1}, x_i)

1305: $$

1306: where $d_\De (x,y)$ is the $\De$-distance from $x$ to $y$

1307: 

1308: 

1309: Each piecewise-linear path $p$ admits a unique representation

1310: $$

1311: p=p_1\cup...\cup p_n

1312: $$

1313: as a composition of maximal billiard subpaths so that

1314: $$

1315: \la_i= \Length (p_i).

1316: $$

1317: We define

1318: $$

1319: \length(p):= \ul{\la}= (\la_1,...,\la_n).

1320: $$

1321: Clearly, $\Length(p)$ is the sum of the vector components of

1322: $\ul{\la}$.

1323: 

1324: 

1325: 

1326: \subsection{The Hecke ring}

1327: \label{heck}

1328: 

1329: In this section we review briefly the definition of spherical

1330: Hecke rings and their relation to the geometry of Euclidean buildings; see

1331: \cite{Gross, KLM3} for more details.

1332: 

1333: Let $\K$ denote a locally compact field with discrete valuation

1334: $v$, and (necessarily) finite residue field of the order $q$. Let  $\mathcal{O}$

1335: be the subring of $\K$ consisting of elements with nonnegative

1336: valuation. Choose a uniformizer $\pi\in \mathcal{O}$.

1337: 

1338: Consider a connected reductive algebraic group $\ul{G}$ over $\K$.

1339: We fix a maximal split torus $\ul{T}\subset \ul{G}$ defined over

1340: $\mathcal{O}$. We put $G := \ul{G}(\K), K:= \ul{G}(\mathcal{O})$ and

1341: $T :=\ul{T}(\K)$. We let $\ul{B} \subset \underline{G}$ be a Borel subgroup

1342: normalized by $\underline{T}$ and set $B:= \underline{B}(\K)$.

1343: 

1344: Let $X$ denote the Bruhat-Tits building

1345: associated with the group $\ul{G}$; $o\in X$ is a distinguished special vertex

1346: stabilized by the compact subgroup $K$.

1347: 

1348: We also have  free abelian cocharacter group

1349: $X_*(\underline{T})$  of $\ul{T}$ whose rank equals $\dim(\underline{T})$.

1350: This group contains the set of coroots $R^{\vee}$ of the group $\ul{G}$.

1351: The roots are the characters of $\underline{T}$

1352: that occur in the adjoint representation on the Lie algebra of $\underline{G}$.

1353: The subset $R^+$ of the roots that occur in representation on the Lie algebra

1354: of $\underline{B}$ forms a positive system and the indecomposable elements

1355: of that positive system form a system of simple roots $\Phi$. We let $W$ denote

1356: the

1357: corresponding (finite) Weyl group.

1358: 

1359: The set of positive roots $\Phi$ determines a positive Weyl chamber $\Pi^+$ in

1360: $X_*(\underline{T})$, by

1361: $$

1362: \Pi^+ = \{\lambda \in X_*(\underline{T}): \ \<\lambda,\alpha\> \geq 0, \alpha

1363: \in \Phi\}.

1364: $$

1365: This chamber is a fundamental domain for the action of $W$ on

1366: $X_*(\underline{T})$.

1367: 

1368: We define a partial ordering on $\Pi^+$ by $\la \succ \mu$ iff the difference

1369: $\la-\mu$

1370: is a sum of positive coroots.

1371: 

1372: \begin{dfn}

1373: The (spherical) Hecke ring $\mathcal{H} = \mathcal{H}_{G}$ is

1374: the ring of all locally constant, compactly supported functions $f: G\lra \Z$

1375: which are $K$--biinvariant. The multiplication in $\mathcal{H}$ is by the

1376: convolution

1377: $$

1378: f \star g (z) = \int_G f(x) \cdot g(x^{-1}z)dx$$

1379: where $dx$ is the Haar measure on $G$ giving $K$ volume $1$.

1380: \end{dfn}

1381: 

1382: The ring $\mathcal{H}$ is commutative and associative (see e.g. \cite{Gross}).

1383: For $\lambda \in X_*(\ul{T})$ let

1384: $c_{\lambda}$ be the characteristic function of the corresponding

1385: $K$-double coset $\la(\pi)\in K\backslash G/K$.

1386: Then the functions $c_\la, \la\in \De$ freely generate ${\mathcal H}$ as an

1387: (additive) abelian group. Deep result of Satake \cite{Satake} relates the Hecke ring of $G$ and

1388: the representation ring of $G^\vee$.

1389: 

1390: 

1391: 

1392: The structure constants  for ${\mathcal H}$ are defined by the formula

1393: 

1394: \begin{equation}

1395: \label{heckesum}

1396: c_{\lambda} \star c_{\mu} = \sum_\nu

1397: m_{\lambda,\mu}(\nu) c_{\nu} =

1398: c_{\lambda + \mu} + \sum

1399: m_{\lambda,\mu}(\nu) c_{\nu},

1400: \end{equation}

1401: where the last sum is taken over all $\nu\in \Pi^+$ such that $\lambda + \mu

1402: \succ \nu$ and therefore is

1403: finite.

1404: 

1405: It turns out that the structure constants $m_{\lambda,\mu}(\nu)$

1406: are nonnegative integers which are polynomials in $q$ with integer

1407: coefficients. These constants are determined by the geometry of

1408: the building as follows. Given $\al, \be, \ga$ let

1409: $\mathcal{T}=\mathcal{T}_{\al,\be}(\ga)$ denote the (finite) set

1410: of geodesic triangles $[o, x, y]$ in the building $X$ which have

1411: the $\De$-side lengths $\al, \be, \ga^*$, so that $y$ is the

1412: projection of the point

1413: $$

1414: \gamma(\pi)\in T\subset G

1415: $$

1416: into $X$ under the map $g\mapsto g\cdot o$.

1417: Recall that $\ga$ as a cocharacter and therefore it defines a homomorphism

1418: ${\mathbb K}^*\to T$.

1419: 

1420: 

1421: \begin{thm}

1422: \cite[Theorem 9.11]{KLM3}.

1423: \label{trcount}

1424: $m_{\alpha,\beta}(\ga)$ equals the cardinality of $\mathcal{T}$.

1425: \end{thm}

1426: 

1427: \begin{rem}

1428: Instead of relating geometry of locally compact buildings to representation theory of $G^\vee$

1429: as it is done in this paper, one can use the non-locally compact building

1430: associated with the group $\ul{G}(\C((t)))$,

1431: as it is done, for instance, in \cite{GL}. It was shown in \cite{KLM2} that the choice of

1432: a field with discrete valuation (or, even, the entire Euclidean building)

1433: is irrelevant as far as the existence of triangles with the

1434: given $\De$-side lengths is concerned: What is important is the affine Weyl group.

1435: For our purposes, moreover, it is more convenient to work with local fields and locally-compact buildings.

1436: In particular, it allows us to compare structure constants of Hecke and representation rings.

1437: \end{rem}

1438: 

1439: 

1440: \section{Chains}

1441: \label{sectionchains}

1442: 

1443: 

1444: \subsection{Absolute chains}

1445: \label{absolute}

1446: 

1447: Let $R$ be a root system on a Euclidean vector space $V$,

1448: $W=W_{sph}$ be the finite Coxeter group associated with $R$, let

1449: $W_{aff}$ denote the affine Coxeter group associated to $R$. Our

1450: root system $R$ is actually the {\em coroot system} for the one

1451: considered by Littelmann in \cite{Littelmann2}. Accordingly, we will

1452: switch weights to coweights, etc. We pick a Weyl chamber $\Del$ for

1453: $W$, this determines the positive roots and the simple roots in $R$.

1454: Let $-\Del$ denote the negative chamber.

1455: 

1456: 

1457: We get the Euclidean Coxeter complex $(A,W_{aff})$, where $A=V$ and

1458: the spherical Coxeter complex $(S,W)$ where $S$ is the

1459: unit sphere in $V$. By abusing notation we will also use the

1460: notation $\De, -\De$ for the positive and negative chambers in

1461: $(S,W)$. We will use the notation $(A,W,-\De)$ for a

1462: Euclidean/spherical Coxeter complex with chosen negative chamber.

1463: More generally, we will use the notation $(A,W_{aff},a)$ for a Euclidean

1464: Coxeter complex with chosen alcove $a$.

1465: 

1466: 

1467: \begin{figure}[tbh]

1468: \centerline{\epsfxsize=4.5in \epsfbox{Fc.eps}} \caption{\sl A

1469: chain.} \label{Fc}

1470: \end{figure}

1471: 

1472: 

1473: \begin{defn}\label{D0}

1474:  A $W$--{\em chain} in

1475: $(A,W,-\De)$ is a finite sequence $(\eta_0,...,\eta_m)$ of nonzero

1476: vectors in $V$ so that for each $i=1,...,m$ there exists a positive

1477: root $\be_i\in R^+$ so that the corresponding reflection

1478: $\tau_i:=\tau_{\be_i}\in W$ satisfies

1479: 

1480: 

1481: 1. $\tau_i(\eta_{i-1})= \eta_{i}$.

1482: 

1483: 

1484: 2. $\be_i(\eta_{i-1}) <0$.

1485: 

1486: 

1487: \noindent Sometimes we will refer to a chain as a $(V,W,-\De)$-chain to

1488: emphasize the choice of $V$, $W$ and $\De$. When the choice of $W$

1489: is clear we will frequently refer to $W$-chains as {\em chains}.

1490: \end{defn}

1491: 

1492: \begin{rem}

1493: One could call these chains $(V,W,\De)$-chains instead,

1494: but this definition would not generalize to affine chains.

1495: \end{rem}

1496: 

1497: 

1498: 

1499: 

1500: Recall that $\tau_i$ is the reflection in the wall

1501: $H_i=\{\be_i=0\}$. More geometrically one can interpret the

1502: condition (2) by saying that the wall $H_i$ separates the negative

1503: chamber $-\De$ from the vector $\eta_i$. In other words, the

1504: reflection $\tau_i$ moves the vector $\eta_{i-1}$ ``closer'' to the

1505: positive chamber.

1506: 

1507: \bigskip

1508: The concept of a chain generalizes naturally to Euclidean Coxeter complexes $(A,W_{aff})$. Pick an

1509: alcove $a$ in $A$.

1510: 

1511: \begin{defn}

1512: \label{DA}

1513: An affine {\em chain} in $(A,W_{aff},a)$ is a finite sequence $(\eta_0,...,\eta_m)$ of

1514: elements in $A$ so that for each $i=1,...,m$ there exists a reflection $\tau_i\in W_{aff}$ such that

1515: 

1516: 1. $\tau_i(\eta_{i-1})= \eta_{i}$.

1517: 

1518: 

1519: 2. The hyperplane $H_i\subset A$ fixed by $\tau_i$ separates $a$ from $\eta_i$.

1520: \end{defn}

1521: 

1522: 

1523: 

1524: We now return to the chains as in definition \ref{D0}.

1525: By restricting vectors $\eta_i$ to have unit length we define chains

1526: in the spherical Coxeter complex, see Figure \ref{Fc}.

1527: 

1528: 

1529: \begin{defn}

1530: The points $\eta_i$ as in Definition \ref{D0} will be called {\em vertices} of the

1531: chain. We will say that the chain begins at $\eta_0$ and {\em ends}

1532: at $\eta_m$, or that this chain is {\em from} $\eta_0$ {\em to}

1533: $\eta_m$. We refer to a subsequence $(\eta_i,

1534: \eta_{i+1},...,\eta_m)$ as a {\em tail} of the chain.

1535: \end{defn}

1536: 

1537: 

1538: We will refer to the number $m$ as the {\em length} of the chain. A

1539: chain $(\eta_i)$ is called {\em simple} if it has length 1. Set

1540: $$

1541: dist_W(\nu,\mu)=dist(\nu, \mu)$$ to be the maximal length $m$ of a

1542: $W$-chain which begins at $\nu$ and ends at $\mu$.

1543: 

1544: 

1545: 

1546: 

1547: 

1548: Given a chain $(\eta_0,...,\eta_m)$ we define a {\em subdivision} of

1549: this chain to be a new chain in $(A,W)$ which is still a chain

1550: from $\eta_0$ to $\eta_m$ and which contains all the vertices of

1551: the original chain.

1552: 

1553: 

1554: The concept of chain determines a partial order on the $W$-orbits in

1555: $V$:

1556: 

1557: 

1558: \begin{defn}

1559: \cite[page 509]{Littelmann2}

1560: \label{order}

1561:  For a pair of nonzero vectors $\nu, \mu\in V$ which

1562: belong to the same $W$-orbit, write $\nu\ge_{W} \mu$ (or simply

1563: $\nu\ge \mu$) if there exists a $W$-chain from $\nu$ to $\mu$.

1564: Accordingly, $\nu >\la$ if $\nu\ge \la$ and $\nu \ne \la$.

1565: \end{defn}

1566: 

1567: 

1568: 

1569: 

1570: \begin{lem}

1571: \label{easy} Suppose that $\nu \ge \mu$ and $\al$ is a positive root

1572: such that $\al(\mu)\le 0$. Then

1573: $$

1574: \nu\ge \tau_\al(\mu).

1575: $$

1576: \end{lem}

1577: \proof If $\al(\mu)=0$ then $\tau_\al(\mu)=\mu$ and there is nothing

1578: to prove. Thus we assume that $\al(\mu)<0$. Consider a chain

1579: $(\nu=\nu_0,...,\nu_s=\mu)$. Then, since $\al(\mu)<0$, we also have

1580: $\al(\tau_\al(\mu))>0$ and thus we get a longer chain

1581: $$

1582: (\nu=\nu_0,...,\nu_s=\mu, \tau(\mu)). \qed

1583: $$

1584: 

1585: 

1586: The word metric $d_W$ on the finite Coxeter group $W$ defines the

1587: {\em length function}

1588: $$

1589: \ell: W\cdot \la \to \N

1590: $$

1591: by

1592: $$

1593: \ell(\mu):= \min\{ d_W(w,1):  w\in W, w^{-1}(\mu)\in \De\}.

1594: $$

1595: 

1596: 

1597: \begin{prop}

1598: \label{wordlength} If $\nu > \mu$ then $\ell(\nu)> \ell(\mu)$.

1599: \end{prop}

1600: \proof It suffices to prove the assertion in the case when

1601: $$

1602: \mu=\tau(\nu), \tau=\tau_\be,

1603: $$

1604: where $\be$ is a positive root, $\be(\nu)<0, \be(\mu)>0$. If $W\cong

1605: \Z/2$ the assertion is clear, so we suppose that it is not the case.

1606: Then we can embed the Cayley graph $\Ga$ of $W$ as a dual graph to

1607: the tessellation of $V$ by the Weyl chambers of $W$. Suppose that

1608: $\nu \in w^{-1}(\De)$, then the wall $H_{\be}=\{\be=0\}$ separates

1609: $w^{-1}(\De)$ from $\De$, where $w\in W$ is the shortest element

1610:  such that $w(\nu)\in \De$. Let $p: [0,1]\to \Ga$

1611: denote the shortest geodesic from $1$ to $w$ in $\Ga$. The path $p$

1612: crosses the wall $H$ at a point $x=p(T)$. We construct a new path

1613: $q$  by

1614: $$

1615: q|[0,T]=p|[0,T], \quad q|[T, 1]= w\circ p|[T,1].

1616: $$

1617: The path $q$ connects $1\in \De$ to the Weyl chamber $\tau w(\De)$

1618: containing $\mu$. This path has a break-point at $x$, which is not a

1619: vertex of the Cayley graph. Therefore, by eliminating the

1620: backtracking of $q$ at $x$, we obtain a new path which connects $1$

1621: to $\tau w(\De)$ and whose length is one less than the length of

1622: $p$. \qed

1623: 

1624: 

1625: \begin{cor}

1626: \label{maxi} The length of a chain in $V$ does not exceed the

1627: diameter of the Cayley graph of $W$.

1628: \end{cor}

1629: 

1630: 

1631: \begin{cor}

1632: \label{end} Suppose that $\nu\in \De$ and $\nu\ge \mu$. Then

1633: $\mu=\nu$.

1634: \end{cor}

1635: \proof Since $\nu\in \De$, $\ell(\nu)=0$. Hence by Proposition \ref{wordlength},

1636: $\ell(\mu)= 0$, which implies  that $\mu=\nu$. \qed

1637: 

1638: 

1639: 

1640: \begin{lem}

1641: \label{2orders}

1642: Suppose that $\nu=w(\mu)\ne 0$ for some $w\in W$ and $\mu\rhd \nu$. Then

1643: $\nu \ge \mu$.

1644: \end{lem}

1645: \proof Let $\De$ be the positive Weyl chamber.

1646: Suppose that $\De_0$ is a chamber containing $\nu$ and $\De'$

1647: is a chamber containing $\mu$. These chambers are non-unique, but

1648: we can choose them in such a way that for all $\xi\in \De_0, \xi'\in \De'$

1649: $$

1650: \xi'\rhd \xi.

1651: $$

1652: In other words, if a wall separates $\De'$ from $\De$ then it also separates $\De_0$ from $\De$.

1653: Let

1654: $$

1655: (\De_0, \De_1,...,\De_m=\De')

1656: $$

1657: be a gallery of Weyl chambers, i.e. for each $i$, $\De_i\cap

1658: \De_{i+1}$ is a codimension 1 face $F_i$ of $\De_i, \De_{i+1}$. We choose this gallery to have

1659: the shortest length, i.e. so that the number $m$ is minimal.

1660: Let $H_i$ be the wall containing $F_i$ and $\tau_i$ be the reflection in $H_i$.

1661: We claim that the sequence

1662: $$

1663: \nu=\eta_0, \eta_1=\tau_1(\eta_0),..., \mu=\eta_m=

1664: \tau_m(\eta_{m-1}),

1665: $$

1666: after deletion of equal members, is a chain.

1667: 

1668: Our proof is by induction on $m$. If $m=0$ and $\nu=\mu$, there is

1669: nothing to prove. Suppose the assertion holds for $m-1$, let us

1670: prove it for $m$. We claim that for all points $\xi_1\in \De_1,

1671: \xi_0\in \De_0$, $\xi_1 \rhd \xi_0$. Indeed, otherwise the wall

1672: $H_1$ does not separate $\De_0$ from $\De$, but separates $\De_1$

1673: from $\De$. Then $H_1$ does not separate $\De_m$ from $\De$

1674: either. Thus, as in the proof of Proposition \ref{wordlength}, we can

1675: replace the gallery $(\De_0, \De_1,...,\De_m)$ with a shorter

1676: gallery connecting $\De_0$ to $\De'$, contradicting minimality of

1677: $m$. Now, clearly,

1678: $$

1679: \eta_0\ge \eta_1, \eta_0\rhd \eta_1.

1680: $$

1681: Therefore, by the induction

1682: $$

1683: \eta_0\ge \eta_1 \ge \eta_m \Rightarrow \nu=\eta_0\ge \eta_m=\mu.  \qed

1684: $$

1685: 

1686: \begin{rem}

1687: The converse to the above lemma is false for instance for

1688: the root system $A_2$. See Figure \ref{Fchain}, where $\eta_0\ge \eta_1$ but  $\eta_0\ntriangleright \eta_1$.

1689: \end{rem}

1690: 

1691: 

1692: 

1693: 

1694: \begin{figure}[tbh]

1695: \centerline{\epsfxsize=4.5in \epsfbox{ch.eps}} \caption{\sl A

1696: chain.} \label{Fchain}

1697: \end{figure}

1698: 

1699: 

1700: As a corollary of Lemma \ref{2orders} we obtain:

1701: 

1702: \begin{lem}

1703: \label{constructingachain}

1704: Let $\nu\in V\setminus \{0\}$ and let $\mu$ be the unique vector in

1705: $W\cdot \nu$ which belongs to $\De$. Then $\nu\ge \mu$.

1706: \end{lem}

1707: \proof Clearly, $\mu \rhd \nu$. Then the assertion follows from

1708: Lemma \ref{2orders}.\qed

1709: 

1710: 

1711: 

1712: 

1713: 

1714: 

1715: 

1716: \begin{defn}

1717: [Maximality condition] \label{maximality}

1718:   We say that a chain

1719: $(\eta_0, \eta_1,..., \eta_m)$ is  {\em maximal} if it cannot be

1720: {\em subdivided} into a longer $W$-chain. Equivalently, \newline

1721: $dist(\eta_i,\eta_{i+1})=1$ for each $i=0,...,m-1$.

1722: \end{defn}

1723: 

1724: 

1725: 

1726: 

1727: 

1728: 

1729: 

1730: 

1731: 

1732: 

1733: \begin{lem}

1734: \label{simplemax}

1735: Suppose that $\nu\ge \mu$, and there exists a simple root $\be$ such that

1736: $\tau_\be(\nu)=\mu$ and  $\be(\nu)<0$. Then $dist(\nu,\mu)=1$.

1737: \end{lem}

1738: \proof Consider a chain from $\nu$ to $\mu$, i.e. a sequence of vectors $\nu=\nu_0, \nu_1,...,

1739: \nu_s=\mu$ and positive roots $\be_1,...,\be_s$ so that

1740: $$

1741: \nu_i= \tau_{\be_i}(\nu_{i-1}) \quad \hbox{and}\quad

1742: \be_i(\nu_{i-1}) <0, i=1,...,s.

1743: $$

1744: Then

1745: $$

1746: \mu-\nu= -\be(\nu) \be^\vee,

1747: $$

1748: and

1749: $$

1750: \mu-\nu= \sum_{i=1}^s -\be_{i-1} (\nu_i)\be^\vee_i.

1751: $$

1752: Thus

1753: $$

1754: \be= \sum_{i=1}^s \frac{ \< \nu_i, \be_i^\vee\>}{\<\nu, \be^\vee\>}

1755: \be_{i-1}

1756: $$

1757: i.e. the simple root $\be$ is a positive linear combination of

1758: positive roots. It follows that $s=1$ and $\be_1=\be$. \qed

1759: 

1760: 

1761: 

1762: 

1763: 

1764: 

1765: 

1766: 

1767: 

1768: 

1769: 

1770: \subsection{Relative chains}

1771: 

1772: 

1773: {\bf 1. Chains relative to a root subsystem.}

1774: 

1775: 

1776: \medskip

1777: Let $R$ be a root system on $V$ with the set of simple roots $\Phi$,

1778: $W$ be the corresponding Weyl group. Let $\Phi'\subset \Phi$ be a

1779: subset, $W'\subset W$ the corresponding reflection subgroup and

1780: $\De'$ the positive chamber for $W'$, defined by the property that

1781: all simple roots $\al\in \Phi'$ are nonnegative on $\De'$. Thus we

1782: will be (frequently) considering $(V,W',-\De')$-chains rather than $(V,W,-\De)$-chains.

1783: In this paper we will be using subgroups $W'$ which are stabilizers

1784: of points $x$ in $(A,W_{aff})$ (where $W=W_{sph}$). Then we will

1785: think of a relative $(V, W')$-chain as a chain in the tangent space

1786: $T_x A$ of the point $x$.

1787: 

1788: 

1789: \begin{lem}

1790: \label{simplesubsystem} Given a nonzero vector $\eta\in V$ there

1791: exists a $W'$-chain \newline $(\eta=\eta_0,...,\eta_m)$ with

1792: $\eta_m\in \De'$, so that the chain $(\eta_i)$ is maximal with

1793: respect to the original root system $R$.

1794: \end{lem}

1795: \proof The proof is by induction on the number $r=r(\eta)$ of simple

1796: roots in $\Phi'$ which are negative on the vector $\eta$. If

1797: $r(\eta)=0$ then $\eta\in \De'$ and there is nothing to prove.

1798: Suppose the assertion holds for all $\eta$ with $r(\eta)\le k$. Let

1799: $\eta$ be such that $r(\eta)=k+1$. Pick a root $\be\in \Phi'$ such

1800: that $\be(\eta)<0$. Then for the vector

1801: $$

1802: \zeta:= \tau_{\be}(\eta)

1803: $$

1804: we have:

1805: $$

1806: \zeta= \eta- \be(\eta)\be^\vee.

1807: $$

1808: Clearly, $\beta(\zeta) >0$. Thus the pair $(\eta,\zeta)$ is a

1809: $W'$-chain; this chain is maximal as a $W$-chain by Lemma

1810: \ref{simplemax}, since it is defined using a simple reflection in

1811: $W$.

1812: 

1813: 

1814: For each simple root $\al\in \Phi'\setminus \{\be\}$ which is

1815: nonnegative on $\eta$

1816: $$

1817: \al(\zeta) = \al(\eta)- \be(\eta) \al(\be^\vee)\ge \al(\eta) \ge 0.

1818: $$

1819: Therefore $r(\zeta)< r(\eta)$ and we are done by the induction. \qed

1820: 

1821: 

1822: 

1823: 

1824: \begin{lem}

1825: \label{ifff} Suppose that $W'\subset W$ is a reflection subgroup as

1826: above. For any two vectors $\al, \del$ the following are equivalent:

1827: 

1828: 

1829: 1. There exists $\be, \ga$ so that $\al \ge_{W'} \be \sim_{W} \ga

1830: \ge_{W'}  \del$.

1831: 

1832: 

1833: 2. There exists $\be$ so that $\al \ge_{W'}  \be \sim_{W} \del$.

1834: 

1835: 

1836: 3. There exists $\ga$ so that $\al \sim_{W} \ga \ge_{W'}  \del$.

1837: 

1838: 

1839: \noindent Here $\sim$ is the relation from Definition \ref{relation}.

1840: \end{lem}

1841: \proof It is clear that 2 $\Rightarrow$ 1 and 3 $\Rightarrow$ 1. We

1842: will prove that 1 $\Rightarrow$ 2, since the remaining implication

1843: is similar. We have chains

1844: $$

1845: (\al=\eta_0,...,\eta_m=\be), \quad \eta_i=\tau_i(\eta_{i-1}), \ i=1,...,m,

1846: $$

1847: $$

1848: (\ga=\eta_0',...,\eta_s'=\del), \quad \eta_i'=\tau_i'(\eta_{i-1}'), \ i=1,...,s.

1849: $$

1850: Then we can extend the first chain to

1851: $$

1852: (\al=\eta_0,...,\eta_m=\be, \tau_1'(\be)=\eta_{m+1},...,

1853: \tau_s'(\eta_{m+s-1})=\eta_{m+s}=:\eps).

1854: $$

1855: After discarding equal members of this sequence we obtain a chain

1856: from $\al$ to $\eps$. Since $\be, \ga$ belong to the same chamber,

1857: the vectors $\eps$ and $\del=\tau_s'\circ ... \circ \tau_1'(\ga)$

1858: also belong to the same chamber. Therefore we obtain

1859: $$

1860: \al\ge \eps \sim \del. \qed

1861: $$

1862: 

1863: 

1864: \begin{defn}

1865: We will write $\al\gtrsim_{W'}  \del$ if one of the equivalent

1866: conditions in the above lemma holds. We will frequently omit the

1867: subscript $W'$ in this notation when the choice of the subgroup $W'$

1868: is clear or irrelevant.

1869: \end{defn}

1870: 

1871: 

1872: The reason for using the relation $\sim$ with respect to $W$ rather

1873: than its subgroup $W'$ (as may seem more natural) is that we will be

1874: taking limits under which the subgroup $W'$ is increasing (but the

1875: limit is still contained in $W$). Such limits clearly preserve the

1876: relation $\gtrsim_{W'}$, but not the relation defined using

1877: $\sim_{W'}$.

1878: 

1879: 

1880: 

1881: 

1882: The next corollary immediately follows from Lemma \ref{ifff}:

1883: 

1884: 

1885: \begin{cor}

1886: \label{iff}

1887:  $\al \gtrsim \del \iff -\del \gtrsim -\al$.

1888: \end{cor}

1889: 

1890: 

1891: \begin{lem}

1892: \label{hard} Suppose that $\nu \gtrsim \mu$ and $\al\in \Phi$ is

1893: such that $\al(\nu)<0$, $\al(\mu)\ge 0$. Then

1894: $$

1895: \tau_\al(\nu)\gtrsim \mu.

1896: $$

1897: \end{lem}

1898: \proof Let $\la$ be such that

1899: $$

1900: \nu \ge \la \sim \mu.

1901: $$

1902: Then $\al(\la)\ge 0$ and it follows that $\la\ne \nu$, i.e. $\nu >

1903: \la$. By applying \cite[Lemma 4.3]{Littelmann2} we get

1904: $$

1905: \tau_\al(\nu) \ge \la \sim \mu. \qed

1906: $$

1907: 

1908: 

1909: \medskip

1910: {\bf 2. Chains relative to positive real numbers ($a$-chains in the

1911: sense of Littelmann).}

1912: 

1913: Let $a$ be a positive real number

1914: and let $\nu, \mu\in V$ be nonzero vectors in the same $W$-orbit.

1915: 

1916: 

1917: \begin{defn}

1918: [P.~Littelmann, \cite{Littelmann2}] \label{achain}  {\em An

1919: $a$-chain} for $(\nu,\mu)$ is a chain $(\la_0,\ldots,\la_s)$ which

1920: starts at $\nu$, ends at $\mu$ and satisfies

1921: 

1922: 

1923: (i) For each $i>0$ we have

1924: $$

1925: t_i:=\be_i(a\la_{i-1})\in \Z,

1926: $$

1927: where $\la_i=\tau_{\be_i}(\la_{i-1})$ as in the Definition

1928: \ref{D0}.

1929: 

1930: 

1931: (ii) For each $i$, $dist(\la_{i-1}, \la_i)=1$.

1932: \end{defn}

1933: 

1934: 

1935: \begin{rem}

1936: Our root system $R$ is the coroot system for the one considered by

1937: Littelmann.

1938: \end{rem}

1939: 

1940: 

1941: Our goal is to give this definition a somewhat more geometric

1942: interpretation. In particular, we will see that the concept of an

1943: $a$-chain is a special case of the concept of a chain relative to a

1944: root subsystem.

1945: 

1946: 

1947: 

1948: 

1949: The root system $R$ defines an affine Coxeter complex $(A, W_{aff})$

1950: on $A$. Let $x\in P(R^\vee)$ be a special vertex; set $x_i:=

1951: x+a\la_i$, $\tau_i:=\tau_{\be_i}$, $i=0,...,s$. Thus $t_i=

1952: \be_i(x_{i-1})$. Note that $t_i\in \Z$ iff $\be_i(x_{i-1})\in \Z$.

1953: 

1954: 

1955: \begin{prop}

1956: $\be_i( x_0)\in \Z$ for each $i=1, 2,...,s$.

1957: \end{prop}

1958: \proof It suffices to consider the case $x=0$. We have:

1959: $$

1960: x_{i-1}= \tau_{i-1}(x_{i-2})= x_{i-2} - \be_{i-1}( x_{i-2})

1961: \be_{i-1}^\vee= x_{i-2} - t_{i-1} \be_{i-1}^\vee= \ldots = x_0-

1962: \sum_{j=1}^{i-1}  t_j \be_{j}^\vee.

1963: $$

1964: Hence

1965: $$

1966: t_i= \be_i( x_{i-1})= \be_i( x_0) -  \sum_{j=1}^{i-1}  t_j

1967: \be_i(\be_{j}^\vee).

1968: $$

1969: Since $t_i\in \Z$ and $\be_i(\be_{j}^\vee)\in \Z$ for all $j$, it

1970: follows that $\be_i( x_0)\in \Z$. \qed

1971: 

1972: 

1973: 

1974: 

1975: \medskip

1976: We define the integers $k_i:= \be_i(x_0)$ and the affine walls

1977: $$

1978: H_i:=H_{\be_{i},k_{i}}= \{ v\in V: \be_{i}(v)=k_i\}.

1979: $$

1980: The reflection $\si_i$ in  the wall $H_i$ belongs to the group

1981: $W_{aff}$, its linear part is the reflection $\tau_i\in W_{sph}$,

1982: $i=1,...,s$.

1983: 

1984: 

1985: 

1986: 

1987: The argument in the above proof can be easily reversed

1988: and hence we get

1989: 

1990: 

1991: \begin{cor}

1992: \label{C1} The integrality condition (i) is equivalent to the

1993: assumption that the point $x_0$ lies on the intersection of walls

1994: $H_i$ of the Euclidean Coxeter complex $(A,W_{aff})$, where each

1995: $H_i$ is parallel to the reflection hyperplane of $\tau_i$.

1996: Equivalently, the $W$-chain $(\la_0,...,\la_s)$ is actually a

1997: $W'$-chain, where $W'=W_{x_0}$ is the stabilizer of $x_0$ in

1998: $W_{aff}$.

1999: \end{cor}

2000: 

2001: 

2002: Therefore, identify the vectors $\la_i$ with vectors in the tangent

2003: space $V':= T_{x_0}(A)$, let $\De'\subset V'$ denote the Weyl

2004: chamber of $W'$ which contains the (parallel transport of the)

2005: positive chamber $\De$. We obtain

2006: 

2007: 

2008: \begin{prop}

2009: \label{eq}  Littelmann's definition of an $a$-chain is equivalent to

2010: the conjunction of

2011: 

2012: 

2013: 1. $(\la_0,...,\la_s)$ is a chain in $(V',W',-\De')$.

2014: 

2015: 

2016: 2. This chain is maximal as a $W$-chain.

2017: \end{prop}

2018: 

2019: 

2020: Thus the choice of the real number $a$ amounts to choosing a

2021:  Coxeter subcomplex $(V', W')$ in $(V,W_{sph})$. The reader will also note

2022: the discrepancy between (1) and (2): The chain condition refers to

2023: the restricted Coxeter complex $(V', W')$, while the maximality

2024: condition refers to the unrestricted one, $(V,W_{sph})$. This is the

2025: key difference between {\em LS paths} and {\em Hecke paths}.

2026: 

2027: 

2028: \begin{rem}

2029: \label{R1} Note that both conditions (1) and (2) are vacuous if

2030: $x_0$ is a special vertex in the Euclidean Coxeter complex

2031: (equivalently, if $a\nu$ is a coweight): If

2032: $$

2033: (\la_0, \la_1,...,\la_s)

2034: $$

2035: is a $W'$-chain, since $W'\cong W_{sph}$,  we can subdivide this

2036: chain to get a longer chain

2037: $$

2038: (\la_0=\la'_0, \la_1',...,\la_{m-1}', \la_m'=\la_s)

2039: $$

2040: between $\la_0$ and $\la_s$ which satisfies the unit distance

2041: condition  $dist(\la_i', \la_{i+1}')=1$ for all $i$.

2042: \end{rem}

2043: 

2044: 

2045: 

2046: 

2047: 

2048: 

2049: 

2050: 

2051: 

2052: 

2053: \subsection{Hecke paths}

2054: 

2055: 

2056: The goal of this section is to introduce a class of piecewise-linear paths which

2057: satisfy a condition similar to Littelmann's definition of LS (and

2058: generalized LS) paths. These paths ({\em Hecke paths}) play the role

2059: in the problem of computing structure constants for spherical Hecke

2060: algebras which is analogous to the role that LS paths play in the

2061: representation theory of complex semisimple Lie groups.

2062: 

2063: 

2064: \medskip

2065: Let $(A, W_{aff}, -\De)$ be a Euclidean Coxeter complex

2066: corresponding to a root system $R$, with fixed negative chamber

2067: $-\De$.  Let $p\in \tilde{\mathcal P}$ be a path equal to the

2068: composition

2069: $$

2070: \ol{x_{1} x_{2}}\cup ... \cup \ol{x_{n-1} x_{n}}.

2071: $$

2072: For each vertex $x=x_i, i=2,...,n-1$ we define the unit tangent

2073: vectors $\xi, \eta \in S_x$ to the segments $\ol{x_ix_{i-1}},

2074: \ol{x_ix_{i+1}}$.

2075: 

2076: 

2077: 

2078: 

2079: \begin{defn}

2080: \label{D3} We say that the path $p$ {\em satisfies the chain

2081: condition} if for each $x=x_i, i=2.,...,n-1$ there exists a unit

2082: vector $\mu$ so that

2083: 

2084: 

2085: 1.

2086: $$

2087: -\xi \ge_{W_x} \mu, \hbox{in the spherical Coxeter complex~}

2088: (S_x,W_x,\De).

2089: $$

2090: 2. $\mu \sim \eta$ in the (unrestricted) spherical Coxeter complex

2091: $(S, W_{sph})$, i.e. for each root $\al\in R$ we have

2092: $$

2093: \al(\mu) \ge 0 \iff \al(\eta) \ge 0.

2094: $$

2095: In other words, for each $t\in [0, 1]$ we have

2096: $$

2097: p'_-(t)\gtrsim_{W_{p(t)}} p'_+(t).

2098: $$

2099: \end{defn}

2100: 

2101: 

2102: \noindent Intuitively, at each break-point $p(t)$ the path $p$

2103: ``turns towards the positive chamber''. In what follows we will use

2104: the notation ${\mathcal P}_{chain}$ for the set of all paths $p\in

2105: {\mathcal P}$ satisfying the chain condition.

2106: 

2107: 

2108: 

2109: 

2110: \begin{defn}

2111: \label{Hecke} A path $p\in \t{\mathcal P}$ is called a {\em Hecke

2112: path} if it is a billiard path which satisfies the chain condition, i.e.

2113: for each $t$

2114: $$

2115: p'_-(t)\ge_{W_{p(t)}} p'_+(t).

2116: $$

2117: \end{defn}

2118: 

2119: 

2120: 

2121: 

2122: 

2123: 

2124: Below is example of a class of Hecke paths. Suppose that $p\in

2125: {\mathcal P}$ and for each $t\in [0, 1]$ either $p$ is smooth at $t$

2126: or there exists a reflection $\tau\in W_{p(t)}$ so that the

2127: derivative of $\tau$ equals $\tau_\be$, $\be \in R^+$ and

2128: 

2129: 

2130: 1. $d\tau_{p(t)} (p'_-(t))=p'_+(t)$.

2131: 

2132: 

2133: 

2134: 

2135: 2. $\be(p'_-(t))<0$, $\be(p'_+(t))>0$.

2136: 

2137: 

2138: Then $p$ is a Hecke path with the length of each chain in

2139: $(S_{p(t)}, W_{p(t)})$ equal $0$ or $1$. See Figure \ref{Fa}.

2140: 

2141: 

2142: \begin{defn}

2143: We say that $p$ satisfies the {\em simple chain} condition if at

2144: each break-point $x=p(t)$ the chain can be chosen to be simple, i.e.

2145: of length $1$.

2146: \end{defn}

2147: 

2148: 

2149: 

2150: 

2151: 

2152: 

2153: \begin{figure}[tbh]

2154: \centerline{\epsfxsize=3.5in \epsfbox{Fa.eps}} \caption{\sl A

2155: billiard path satisfying the simple chain condition.} \label{Fa}

2156: \end{figure}

2157: 

2158: 

2159: In what follows we will also need

2160: 

2161: \begin{defn}

2162: \label{gH}

2163: Suppose that $p$ is a path satisfying the chain condition. We call $p$ a {\em generalized Hecke path} if

2164: each geodesic segment in $p$ (regarded as a vector)

2165: is an integer multiple of some $w\varpi$ where $w\in W$ and $\varpi=\varpi_i$

2166: is one of the fundamental coweights.

2167: \end{defn}

2168: 

2169: 

2170: 

2171: 

2172: 

2173: \subsection{A compactness theorem}

2174: 

2175: 

2176: Pick $\eps>0$. We define the subset ${\mathcal P}_{m,\eps}\subset

2177: {\mathcal P}$ consisting of paths $p$ with

2178: $$

2179: \length(p)=\ul{\la}=(\la_1,...,\la_m)$$

2180:  so that for each $i$,

2181: \begin{equation}

2182: \label{bdd}

2183: \eps\le |\la_i|\le \eps^{-1}.

2184: \end{equation}

2185: 

2186: 

2187: 

2188: 

2189: \begin{thm}

2190: \label{compactness} For each $\eps>0$ the set ${\mathcal

2191: P}_{chain,m,\eps}:= {\mathcal P}_{chain}\cap {\mathcal P}_{m,\eps}$

2192: is compact in ${\mathcal P}_{m,\eps}$.

2193: \end{thm}

2194: \proof Suppose that $p\in {\mathcal P}_{chain}$ is a concatenation

2195: of $m$ billiard paths $p_i$. Then the number of breaks in the broken

2196: geodesic $p_i$ is bounded from above by a constant $c$ equal to the

2197: length of a maximal chain in the Bruhat order of the finite Weyl

2198: group $W_{sph}$, see Corollary \ref{maxi}. This immediately implies

2199: that the subset ${\mathcal P}_{chain,m,\eps}$ is precompact in

2200: ${\mathcal P}$. What has to be proven is that this subset is closed.

2201: 

2202: 

2203: \bigskip

2204: Let $\la_{ij}\in V$ be vectors of nonzero length so that

2205: $$

2206: p_i= \pi_{\la_{i1}} * \pi_{\la_{i2}}*...*\pi_{\la_{in}}

2207: $$

2208: are Hecke paths, $i=1, 2,...$. We suppose that

2209: $$

2210: \lim_{i\to\infty} \la_{ij}=0, j=2,...,n-1

2211: $$

2212: and

2213: $$

2214: \lim_{i\to\infty} \la_{i1}=\la_{\infty,1}, \lim_{i\to\infty} \la_{in}=\la_{\infty, n}

2215: $$

2216: are nonzero vectors. It is clear that

2217: $$

2218: \lim_i p_i= p_{\infty}:= \pi_{\la_{\infty,1}} * \pi_{\la_{\infty,n}}.

2219: $$

2220: 

2221: 

2222: 

2223: \begin{lem}

2224: \label{1}

2225: Under the above conditions the path $p_{\infty}$ is again a Hecke path.

2226: \end{lem}

2227: \proof Let $x:= \la_{\infty,1}$. We need to check that the unit vectors $\bar\la_{\infty,1}, \bar\la_{\infty,n}$ satisfy

2228: $$

2229: \bar\la_{\infty,1} \ge_{W_x} \bar\la_{\infty,n}.

2230: $$

2231: Here and below, $W_x$ is the stabilizer of $x\in V$ in the Coxeter group $W_{aff}$.

2232: 

2233: 

2234: Let $x_{ij}$ denote the break-point of $p_i$ which is the concatenation point between $\la_{i,j}$ and $\la_{i,j+1}$. Then

2235: $$

2236: \bar\la_{i,j} \ge_{W_{x_{ij}}} \bar\la_{i,j+1},

2237: $$

2238: $\lim_i x_{ij}=x$. If $\si_{ij}\in W_{aff}$ is a reflection fixing $x_{ij}$,

2239: then, up to a subsequence,

2240: $$

2241: \si_{\infty,j}= \lim_i \si_{ij}\in W_{aff}

2242: $$

2243: fixes the point $x$. Therefore it follows from the definition of a chain that

2244: $$

2245: \bar\la_{\infty,j} \ge_{W_x} \bar\la_{\infty,j+1}

2246: $$

2247: for each $j$. By putting these inequalities together we obtain

2248: $$

2249: \bar\la_{\infty,1} \ge_{W_x} \bar\la_{\infty,n}. \qed

2250: $$

2251: 

2252: 

2253: 

2254: \medskip

2255: Suppose now that $p_i, q_i$ are Hecke paths,

2256: $$

2257: p_i= \pi_{\la_{i1}} * \pi_{\la_{i2}}*...*\pi_{\la_{in}}, q_i= \pi_{\mu_{i1}} * \pi_{\mu_{i2}}*...*\pi_{\mu_{im}}

2258: $$

2259: and the concatenation $r_i:= p_i*q_i$ satisfies the chain condition. We suppose that $r_{\infty}=p_\infty * q_\infty$

2260: is the limit of the sequence of paths $r_i$ and the paths $p_\infty, q_\infty$ are not constant.

2261: 

2262: 

2263: \begin{lem}

2264: \label{2}

2265: Under the above assumptions, the path $r_\infty$ also satisfies the chain condition.

2266: \end{lem}

2267: \proof Lemma \ref{1} implies that the paths $p_\infty, q_\infty$ are Hecke. Therefore it suffices to verify the chain condition

2268: at the concatenation point $x=p_\infty(1)$. Up to passing to a subsequence, we have:

2269: $$

2270: \lim_{i\to\infty} \la_{ij}=\la_{\infty,j}, \lim_{i\to\infty} \mu_{ij}=\mu_{\infty,j},

2271: $$

2272: $$

2273: \lim_{i\to\infty} \bar\la_{ij}=\bar\la_{\infty,j}, \lim_{i\to\infty} \bar\mu_{ij}=\bar\mu_{\infty,j}.

2274: $$

2275: 

2276: Suppose that

2277: $$

2278: \lim_i \la_{ij}=0, j=k+1,...,n, \quad \lim_i \mu_{ij}=0, j=1,...,l,

2279: $$

2280: and

2281: $$

2282: \la_{\infty,k}\ne 0, \quad \mu_{\infty,l+1}\ne 0.

2283: $$

2284: 

2285: 

2286: By Lemma \ref{1},

2287: $$

2288: \bar\la_{\infty,k} \ge_{W_x} \bar\la_{\infty,n}, \quad \bar\mu_{\infty,1} \ge_{W_x} \bar\mu_{\infty,l+1}.

2289: $$

2290: On the other hand, it is clear that

2291: $$

2292: \bar\la_{\infty,n} \gtrsim_{W_x} \bar\mu_{\infty,1}.

2293: $$

2294: Therefore, by Lemma \ref{ifff}, we get

2295: $$

2296: \bar\la_{\infty,k} \gtrsim_{W_x} \bar\mu_{\infty,l+1}. \qed

2297: $$

2298: 

2299: 

2300: We now can finish the proof Theorem \ref{compactness}. Suppose that $p_i=p_{i1}*...*p_{im}$ is a sequence of paths

2301: in ${\mathcal P}_{chain,m,\eps}$, where each $p_{ij}$ is a billiard path, and $p_{\infty}$ is the limit of this sequence.

2302: Each sequence of billiard paths $(p_{ij})_{i\in \N}$ converges to a billiard path $p_{\infty,j}$ which is a Hecke path according to Lemma \ref{1}.

2303: Consider now a concatenation point $x$ of the subpaths $p_{\infty}:=p_{\infty,j}, q_{\infty}:= p_{\infty,j+1}$. The paths

2304: $p_\infty, q_\infty$ are non-constant by the inequality (\ref{bdd}). Therefore we

2305: can apply Lemma \ref{2} to conclude that the path $p_\infty$ satisfies chain condition at the point $x$. \qed

2306: 

2307: 

2308: 

2309: 

2310: 

2311: 

2312: \begin{rem}

2313: It is easy to see that the assumption that the length of each billiard subpath $p_{ij}$ is bounded away

2314: from zero, is necessary in Theorem \ref{compactness}.

2315: \end{rem}

2316: 

2317: 

2318: 

2319: 

2320: 

2321: 

2322: \section{Folding}

2323: \label{sectionfolding}

2324: 

2325: 

2326: The key tool for proving the main results of this paper is {\em

2327: folding} of polygons in a building $X$ into apartments and Weyl

2328: chambers. The folding construction replaces a geodesic segment $\t{p}$ in $X$

2329: with a piecewise-linear path $p$ in an apartment.

2330: This construction was used in \cite{KLM3} to construct various

2331: counter-examples.  The reader will note that the folding

2332: construction used in the present paper is somewhat different from

2333: the one in \cite{KLM3}.

2334: 

2335: 

2336: 

2337: 

2338: 

2339: 

2340: 

2341: 

2342: \subsection{Folding via retraction}

2343: 

2344: 

2345: Suppose that $X$ is a Euclidean or spherical building modeled on the

2346: Coxeter complex $(A,W)$, we identify the model apartment $A$ with an

2347: apartment $A\subset X$; let $a\subset A$ be an alcove (or a chamber

2348: in the spherical case). Recall that the {\em retraction, or folding,

2349: to an apartment} $f=Fold_{a,A}: X\to A$ is defined as follows (see

2350: for instance \cite{Rousseau}):

2351: 

2352: 

2353: Given a point $x\in X$ choose an apartment $A_x$ containing $x$ and

2354: $a$. Then there exists a (unique) isomorphism $\phi: A_x\to A$

2355: fixing $A\cap A_x$ pointwise and therefore fixing $a$ as well. We

2356: let $f(x):= \phi(x)$. It is easy to see that $f(x)$ does not depend

2357: on the choice of $A_x$. Observe that $f$ is an isometry on each

2358: geodesic $\ol{xy}$, where $y\in a$.

2359: 

2360: 

2361: \medskip

2362: The retraction can be generalized as follows.

2363: 

2364: 

2365: Suppose that $X$ is a Euclidean building, $a$ is an alcove with a

2366: vertex $v$ (not necessarily special). Let $\De\subset A$ denote a

2367: Weyl chamber with  tip $o$. Choose a dilation $h\in Dil(A,W)$ which

2368: sends $v$ to $o$. Let ${\mathbb P}: A\to \De$ denote the natural

2369: projection which sends points $x\in A$ to $W_{sph}\cdot x \cap

2370: \Del$, where $W_{sph}$ is the stabilizer of $o$ in $W_{aff}$.

2371: 

2372: 

2373: We define a folding $g=Fold_{v,h,\De}: X\to \De$ as the composition

2374: $$

2375: \P\circ h\circ Fold_{a,A}.

2376: $$

2377: The mapping $g$ will be called a {\em folding} into a Weyl chamber.

2378: Observe that $g$ (unlike $Fold_{a,A}$) does not depend upon the

2379: choice of the alcove $a$, therefore it will be denoted in what

2380: follows $g=Fold_{v,h,\De}$. Note that

2381: $$

2382: Fold_{v,k\circ h,\De}= k\circ Fold_{v,h,\De}.

2383: $$

2384: 

2385: 

2386: In case when $h=Id$ we will abbreviate $Fold_{v,h,\De}$ to

2387: $Fold_{\De}$.

2388: 

2389: 

2390: 

2391: 

2392: \begin{rem}

2393: The folding maps are Lipschitz and differentiable. The restriction

2394: of the retraction $Fold_{a,A}$ to each chamber (alcove) is a

2395: congruence of two chambers (alcoves).

2396: \end{rem}

2397: 

2398: 

2399: 

2400: 

2401: 

2402: 

2403: %\begin{rem}

2404: %Observe that instead of retracting to an apartment one can retract

2405: %$X$ to the parallel set of a flat in $X$. We will explore this in \cite{HKM}.

2406: %\end{rem}

2407: 

2408: 

2409: If $h=Id$ (and thus $v=o$) one can describe $f=Fold_{\De}$ as

2410: follows. Given a point $x\in X$ find an apartment $A_x$ through $v,

2411: x$ and a Weyl chamber $\De_x\subset A_x$ with  tip $v$. Let

2412: $\phi: \De_x\to \De$ be the unique isometry extending to an

2413: isomorphism of Coxeter complexes $A_x\to A$. Then $f(x)=\phi(x)$.

2414: 

2415: 

2416: \medskip

2417: Suppose now that $X$ is a Euclidean building, $x\in X$; we give the

2418: link $\Si_x(X)$ structure of an unrestricted spherical building $Y$.

2419: Let $R$ denote the corresponding root system. Let $\del$ be a

2420: chamber in $Y$ and $\xi, \mu\in \del$. Let $f: X\to \Del$ be a

2421: folding of $X$ to a Weyl chamber. Let $x'$, $\xi', \mu'$ denote the

2422: images of $x$, $\xi, \mu$ under $f$ and $df_x$. Then

2423: 

2424: 

2425: \begin{lem}

2426: \label{positivity} 1. $d_{ref}(\xi, \mu)= d_{ref}(\xi', \mu')$.

2427: 

2428: 

2429: 2. For each $\al\in R$,

2430: $$

2431: \al(\mu')\ge 0 \iff \al(\xi')\ge 0.

2432: $$

2433: \end{lem}

2434: \proof The restriction $df|_\del$ is an isometry which is the

2435: restriction of an isomorphism of spherical apartments. This proves

2436: (1). To prove (2) observe that $df$ sends $\del$ to a spherical Weyl

2437: chamber $\del'$ in $\Si_{x'}X$. \qed

2438: 

2439: 

2440: \medskip

2441: Let $f$ be a folding of $X$ into an apartment or a chamber.

2442: 

2443: 

2444: \begin{lem}

2445: \label{broken} For each geodesic segment $\ol{xy}\subset X$ its

2446: image $f(\ol{xy})$ is a broken geodesic, i.e. it is a concatenation

2447: of geodesic segments.

2448: \end{lem}

2449: \proof We give a proof in the case of a folding into an apartment

2450: and will leave the other case to the reader. Let $A'\subset X$

2451: denote an apartment containing the geodesic segment $\ol{xy}$. Let

2452: $a_1,...,a_m$ denote the alcoves (or chambers in the spherical case)

2453: in $A'$ covering $\ol{xy}$, set $\ol{x_i x_{i+1}}:= \ol{xy}\cap

2454: a_i$. For each $a_i$ there exists an apartment $A_i$ containing the

2455: alcoves $a$ and $a_i$. The restriction of the retraction $f$ to

2456: $A_i$ is an isometry. It is now clear that the path

2457: $f(\ol{xy})$ is a composition of the geodesic paths $f(\ol{x_i

2458: x_{i+1}})$.

2459:  \qed

2460: 

2461: 

2462: We let $x_i'=f(x_i)$ denote the break points of $f(\ol{xy})$. For

2463: each $x_i'$ let $\xi_i', \eta_i'$ denote the unit tangent vectors in

2464: $T_{x_i'} A$ which are tangent to the segments $\ol{x_i' x_{i-1}'},

2465: \ol{x_i' x_{i+1}'}$ respectively.

2466: 

2467: 

2468: \begin{lem}

2469: \label{billiard}

2470:  The broken geodesic $f(\ol{xy})$ is a {\em billiard

2471: path}, i.e. for each break point $x_i'$ the vectors  $\xi':=\xi_i',

2472: \eta':=\eta_i'$ satisfy

2473: $$

2474: \exists w\in W_{x_i'} : w(\xi')=-\eta'.

2475: $$

2476: \end{lem}

2477: \proof We again present a proof only in the case of a folding into

2478: an apartment. Let $Y=\Si_{x_i} X$ denote the spherical building

2479: which is the space of directions of $X$  at $x_i$. Let $\De_Y$

2480: denote the Weyl chamber of this building and $\theta: Y\to \De_Y$

2481: the canonical projection. The directions $\xi=\xi_i$ and

2482: $\eta=\eta_i$ of the segments $\ol{x_i x_{i-1}}, \ol{x_i x_{i+1}}$

2483: are antipodal in the building $Y$. Since the folding $f$ is an

2484: isomorphism of the apartments $A_i\to A, A_{i+1}\to A$, and

2485: $$

2486: df(\eta)=\eta', \quad df(\xi)=\xi',

2487: $$

2488: we see that

2489: $$

2490: \theta(\xi)=\theta(\xi'), \quad \theta(\eta)=\theta(\eta').

2491: $$

2492: The assertion now follows from Lemma  \ref{thetaproperties}, part

2493: (2). \qed

2494: 

2495: 

2496: \begin{lem}

2497: \label{invarianceoflength} Suppose that $X$ is a Euclidean building,

2498: $f=Fold_{a,A}: X\to A$ and $g=Fold_{z,h,\De}$ are foldings to an

2499: apartment and a chamber respectively. Then for each piecewise-linear path $p$ in

2500: $X$ we have:

2501: 

2502: 

2503: 1. $\length(f(p))=\length(p)$.

2504: 

2505: 

2506: 2. $\length(g(p))=k\cdot \length(p)$, where $k>0$ is the conformal

2507: factor of the dilation $h$.

2508: \end{lem}

2509: \proof We will prove the first assertion since the second assertion

2510: is similar. It suffices to give a proof in the case when $p$ is a

2511: billiard path. Then, analogously to the proof of Lemma

2512: \ref{billiard}, there exists a representation of $p$ as a composition

2513: of geodesic subpaths

2514: $$

2515: p=p_1\cup ...\cup p_m

2516: $$

2517: so that the restriction of $f$ to each $p_i$ is a congruence.

2518: Therefore

2519: $$

2520: \Length(p_i)= \Length(f(p_i))

2521: $$

2522: and hence

2523: $$

2524: \Length (p)= \sum_i \Length(p_i)= \Length (f(p)). \qed

2525: $$

2526: 

2527: 

2528: 

2529: 

2530: 

2531: 

2532: \medskip

2533: {\bf Derivative of the retraction.} We assume that $rank(X)\ge 1$.

2534: We identify the model apartment $A$ with an apartment in $X$. Pick

2535: $a\subset A$ which is an alcove (in the Euclidean case) or a chamber

2536: (in the spherical case). Given a point $x'\in X$ choose an apartment

2537: $(A',W')$ through $a$ and $x'$ and let $\phi: A\to A'$ denote the

2538: inverse to the retraction $f=Fold_{a,A}: A'\to A$. Set $x=f(x')$ and

2539: let $W_x'$ denote the stabilizer of $x'$ in $W'$. Then the link

2540: $Y=\Si_{x'}(X)$  has a natural structure of a thick spherical

2541: building modeled on $(S,W_x')$. It is easy to see that $(S, W_x')$

2542: is independent of the choice of $A'$. Observe that if $x'$ is

2543: antipodal to a regular point $y\in a$ then $x'$ is regular itself

2544: and therefore $W_x'=\{1\}$. We next define a chamber $s\subset S$:

2545: 

2546: 

2547: Given a regular point $y\in a\setminus \{x'\}$ and a geodesic

2548: segment $\ol{x'y}$, let $\zeta=\zeta(y)$ denote the unit tangent

2549: vector to $\ol{x'y}$ at $x'$. Then the set

2550: $$

2551: \{\zeta(y): y \hbox{~is a regular point in~} a\},

2552: $$

2553: is contained in a unique spherical chamber $s\subset S$. (If $x'$ is

2554: antipodal to some $y\in int(a)$ then $s=S$.)

2555: 

2556: 

2557: Set $f':= \phi\circ f= Fold_{a,A'}$.

2558: 

2559: 

2560: 

2561: 

2562: \begin{lemma}

2563: \label{consistent}

2564:  The derivative $d_{x'}f': Y\to S$ equals $Fold_{s,S}$.

2565: \end{lemma}

2566: \proof Given $\eta\in Y$, find an alcove (or a spherical chamber)

2567: $c$ so that $\eta\in \Si_{x'} c$. Then there exists an apartment

2568: $A_\eta\subset X$ containing both $a$ and $c$. Let $S_\eta$ denote

2569: the unit tangent sphere of $A_\eta$ at $x'$. Then $\eta\in S_\eta$

2570: and $s\subset S_\eta$. Now it is clear from the definition that

2571: $$

2572: df'(\eta)=Fold_{s,S}(\eta)$$

2573:  since both maps send $S_\eta$ to $S$ and fix $s$ pointwise. \qed

2574: 

2575: 

2576: 

2577: 

2578: \medskip

2579: {\bf Folding of polygons.} Suppose now that $X$ is a building and

2580: $\t{P}=[\t{z}, \t{x}_1, \ldots, \t{x}_n]$ is a geodesic polygon in $X$. Pick an

2581: apartment $A\subset X$ which contains $\ol{\t{z} \t{x}_1}$ and an alcove

2582: $a\subset A$ which contains $\t{z}$. Let $\De\subset A$ denote a Weyl

2583: chamber (in case $X$ is Euclidean) with  tip $o$. Let $f$ be a

2584: folding of $X$ of the form

2585: $$

2586: f=Fold_{a,A}$$

2587:  or

2588: $$

2589: f=Fold_{\t{z},h,\De},$$

2590:  where $h$ is a dilation sending $\t{z}$ to $o$. We will

2591: then apply $f$ to $\t{P}$ to obtain a {\em folded polygon} $P:= f(\t{P})$

2592: in $A$ or $\De$ respectively.

2593: 

2594: 

2595: Observe that the restriction of $f$ to the edges $\ol{\t{z} \t{x}_1}$ and

2596: $\ol{\t{x}_n \t{z}}$ of $P$ is an isometry or a similarity. The restriction

2597: of $f$ to the path

2598: $$

2599: \t{p}=\ol{\t{x}_1 \t{x}_2}\cup ...\cup \ol{\t{x}_{n-1}\t{x}_n}

2600: $$

2601: preserves the {\em type} of the unit tangent vectors, cf. Lemma

2602: \ref{invarianceoflength}. We will be using foldings into apartments

2603: and chambers to transform geodesic polygons in $X$ into {\em folded

2604: polygons}.

2605: 

2606: 

2607: In the special case when $\t{P}=T$ is a triangle (and thus $n=2$), the

2608: folded triangle $P=f(T)$ has two geodesic sides $\ol{z w_1}:=f(\ol{\t{z} \t{x}_1}),

2609: \ol{x_2 z}:= f(\ol{\t{x}_2 \t{z} })$ and one {\em broken side} $p:=f(\ol{\t{x}_1 \t{x}_2})$, so we

2610: will think of $f(T)$ as a {\em broken triangle}.

2611: 

2612: 

2613: 

2614: 

2615: The next proposition relates folding into a Weyl chamber with the

2616: concept of folding of polygons used in \cite{KLM3}. Let $P=[o,

2617: x_1, x_2, \ldots , x_n]$ be a polygon in $\De$. Triangulate $P$

2618: from the vertex $o$ into geodesic triangles $T_i=[x_i,

2619: x_{i+1},o]$. Suppose that $\t{P}\subset X$ is a geodesic polygon

2620: $$

2621: \t{P}=[o, \t{x}_1, \t{x}_2, \ldots, \t{x}_n], \t{x}_1=x_1,

2622: $$

2623: triangulated into geodesic triangles $\t{T}_i=[\t{x}_i, \t{x}_{i+1},o]$, where

2624: each $\t{T}_i$ is contained in an apartment $A_i$. Assume that for each

2625: $i$ there exists a congruence

2626: $$

2627: \phi_i: \t{T}_i\to T_i

2628: $$

2629: i.e. an isometry sending $\t{x}_j$ to $x_j$ ($j=i, i+1$) which extends

2630: to an isomorphism of Coxeter complexes $\phi_i: A_i\to A$.

2631: 

2632: 

2633: \begin{prop}

2634: \label{uniqueness}

2635:  Under the above assumptions, for each $i$,

2636: $Fold_{\De}|\t{T}_i=\phi_i|\t{T}_i$.

2637: \end{prop}

2638: \proof Let $\De_i\subset A_i$ denote the preimage of $\De$ under

2639: $\phi_i$. Then each $\De_i$ is a Weyl chamber, hence

2640: $\phi_i|\De_i=Fold_{\De}|\De_i$, by the alternative description of

2641: $Fold_{\De}$ given earlier in this section. \qed

2642: 

2643: 

2644: \medskip

2645: The following lemma shows that unfolding of polygons is a {\em local

2646: problem}. Suppose that $T=[o, x_1,...,x_n]\subset A$ is a

2647: geodesic polygon so that $x_i\ne o$ for each $i$. For each $i=2,...,

2648: n-1$ we define the unit vectors

2649: $$

2650: \xi_i, \eta_i, \zeta_i\in \Si_{x_i} A

2651: $$

2652: which are tangent to the segments $\ol{x_i x_{i-1}},

2653: \ol{x_{i}x_{i+1}}$, $\ol{x_i o}$. Define thick spherical

2654: buildings $Y_i:= \Si_{x_i}(X)$. By combining the above proposition

2655: with \cite[Condition 7.5]{KLM3} we obtain

2656: 

2657: 

2658: \begin{lem}

2659: \label{locality} The polygon $T$ can be unfolded in $X$ to a

2660: geodesic triangle $\t{T}$ whose vertices project to $o, x_1, x_n$ if

2661: and only if for each $i=2,...,n-1$, there exists a triangle

2662: $[\t\xi_i, \t\zeta_i, \t\eta_i]\subset Y_i$ so that

2663: $$

2664: d_{ref}(\t\xi_i, \t\zeta_i) = d_{ref}(\xi_i, \zeta_i),

2665: $$

2666: $$

2667: d_{ref}(\t\eta_i, \t\zeta_i) = d_{ref}(\eta_i, \zeta_i),

2668: $$

2669: $$

2670: d(\t\xi_i, \t\eta_i) = \pi.

2671: $$

2672: \end{lem}

2673: 

2674: 

2675: 

2676: 

2677: \medskip

2678: We will eventually obtain a characterization of the broken triangles

2679: in $\De$ which are foldings of geodesic triangles in $X$ as {\em

2680: billiard triangles satisfying the chain condition}, see section

2681: \ref{last}. The goal of the next section is to give a {\em

2682: necessary} condition for a broken triangle to be unfolded; we also

2683: give a partial converse to this result.

2684: 

2685: 

2686: 

2687: 

2688: 

2689: 

2690: \subsection{Converting folded triangles in spherical buildings into chains}

2691: \label{converting}

2692: 

2693: Suppose that $X$ is a (thick) spherical or Euclidean building modeled on $(A,W)$. Consider a

2694: triangle $\tilde{T}=[x, y, z]\subset X$ with

2695: $$

2696: \be:=d_{ref}(x, y), \quad \ga:=d_{ref}(y, z).

2697: $$

2698:  Assume that $A$ is embedded in $X$ so that it contains $x$ and $z$.

2699: Let $a\subset A$ be a spherical chamber or a Euclidean alcove containing $z$.

2700: In the spherical case we regard $a$ as the {\em negative chamber} in $A$,

2701: let $\De$ denote the positive chamber $-a$. We have the retraction $f:=

2702: Fold_{a,A}:X\to A$.

2703: 

2704: 

2705: 

2706: 

2707: 

2708: 

2709: \begin{figure}[tbh]

2710: \centerline{\epsfxsize=5in \epsfbox{chain.eps}} \caption{\sl

2711: Converting geodesic triangle to a chain.} \label{chain.fig}

2712: \end{figure}

2713: 

2714: 

2715: 

2716: 

2717: \begin{thm}\label{T2}

2718: There exists a $(A,W,a)$--chain $(y_0,..., y_m)$  such that

2719: $y_m=f(y)$, $d_{ref}(x, y_0)=\be$, $d_{ref}(y_m, z)=\ga$. (In the case when $X$ is a Euclidean building

2720: the above chain is an affine chain.)

2721: See Figure \ref{chain.fig}.

2722: \end{thm}

2723: \proof

2724: We prove the assertion for the spherical buildings as the Euclidean case is completely analogous.

2725: (This is also the only case when this theorem is used in the present paper.)

2726: 

2727: Our proof is by induction on the rank of the building.

2728: Consider first the case when $rank(X)=0$ (i.e. $A=S^0$ is the

2729: 2-point set). If $y$ and $x$ are both distinct from $z$, then

2730: $f(y)\ne z$. This implies that $f(y)=x$ and we take

2731: $$

2732: y_0:= z, y_1:= y, m=1.

2733: $$

2734: In the remaining cases we will use the chain $y_0=f(y)=y_m$.

2735: 

2736: 

2737: Suppose now that $rank(X)=r\ge 1$ and the assertion holds for all

2738: (spherical) buildings of rank $r-1$, let's prove it for buildings of

2739: rank $r$.

2740: 

2741: 

2742: We let $\tilde{p}: [0, c]\to \ol{xy}$ denote the unit speed parametrization

2743: of $\ol{xy}$ and set $p:= f(\t{p})$. {\em We assume for now that $z\notin \ol{xy}$}.

2744: 

2745: 

2746: As in the proof of Lemma \ref{broken}, we ``triangulate'' the geodesic

2747: triangle $\t{T}$ into geodesic triangles $\t{T}_i:= [z, \t{x}_i,

2748: \t{x}_{i+1}]$, where the points $\t{x}_i=\t{p}(t_i)$, $i=1,...,n$, are chosen

2749: so that each triangle $\t{T}_i$ is contained in an apartment

2750: $A_i\subset X$ and the map $f$ restricts to an isometry $f:

2751: \t{T}_i\to f(\t{T}_i)\subset A$. Here $\t{x}_0:=x, \t{x}_{n+1}=y$.

2752: Observe that each side of $\t{T}_i$ has positive length.

2753: 

2754: 

2755: 

2756: 

2757: Let $\t{S}_i:=\Si_{\t{x}_i}(A_i)$ denote the unit tangent sphere at

2758: $\t{x}_i$. Define $\t{s}_i$ to be the (unique) chamber in $\t{S}_i$

2759: containing all the directions of the geodesic segments from

2760: $\t{x}_i$ to the interior of $a$. This determines the {\em positive

2761: chamber} $\t{\De}_i=-\t{s}_i\subset S_i$.

2762: 

2763: 

2764: Set

2765: $$

2766: \t\eta_i:=\t{p}'(t_i), \t\xi_i:= -\eta_i,

2767: $$

2768: and let $\t\zeta_i\in \t{S}_i$ denote the unit tangent vector to

2769: $\ol{\t{x}_i z}$.

2770: 

2771: 

2772: Now, applying the retraction $f$ to all this data, we obtain:

2773: 

2774: 1. The folded triangle $T=f(\t{T})$ which has two geodesic sides

2775: $\ol{zx}, \ol{zy_m}$ (where $y_m=f(y)$), and the broken side

2776: represented by the path $p=f(\t{p})$. In particular, $d_{ref}(y_m, z)=d_{ref}(y, z)$ (as required by the theorem).

2777: 

2778: 

2779: 2. The vertices $x_i=p(t_i)= f(\t{x}_i)$ of the broken geodesic $p$.

2780: 

2781: 3. Unit tangent vectors $\xi_i=df(\t{\xi}_i), \eta_i=df(\t{\eta}_i),

2782: \zeta_i=df(\t{\zeta_i})$ in $\Si_x A$. These vectors are tangent to

2783: the segments $\ol{x_i x_{i-1}}, \ol{x_i x_{i+1}}, \ol{x_i z}$

2784: respectively.

2785: 

2786: 

2787: 4. The positive chamber $\De_i=df(\t{\De}_i)$ and the negative

2788: chamber $s_i=df(\t{s}_i)$ in the spherical Coxeter complex

2789: $(S_i=\Si_{x_i}(A), W_i=W_{x_i})$. The negative chamber contains the

2790: directions tangent to the geodesic segments from $x_i$ to the

2791: chamber $a\subset A$.

2792: 

2793: Our goal is to convert the broken side $p$ of $T$ into a chain in

2794: $A$ by ``unbending'' the broken geodesic $p$ to a geodesic segment

2795: in $A$. See Figure \ref{con.fig}.

2796: 

2797: 

2798: 

2799: \begin{figure}[tbh]

2800: \centerline{\epsfxsize=5in \epsfbox{cone.eps}} \caption{\sl

2801: Forming a chain by unbending.} \label{con.fig}

2802: \end{figure}

2803: 

2804: 

2805: \begin{lem}

2806: \label{trivial} The path $p$ satisfies the following:

2807: 

2808: 

2809: 1. The metric lengths of $p$ and $\tilde{p}=\ol{xy}$ are the same.

2810: 

2811: 

2812: 2. $p'(0)=\tilde{p}'(0)$.

2813: 

2814: 

2815: 3. At each break-point $x_i$ there exists an $(S_i, W_{i}, s_i)$-chain from $-\xi_i$ to $\eta_i$.

2816: \end{lem}

2817: \proof The first two assertions are clear from the construction.

2818: Let's prove the last statement. For each $i$ and the point

2819: $\t{v}=\t{x}_i$ we have the spherical building $Y:=\Si_{\t{v}}(X)$

2820: which has rank $r-1$. This building contains the antipodal points

2821: $$

2822: \t\xi_i, \t\eta_i

2823: $$

2824: and the point $\t\zeta_i$. We form the geodesic triangle

2825: $\tau=[\t\xi_i, \t\eta_i, \t\zeta_i]\subset Y$, where we use an

2826: arbitrary shortest geodesic in $Y$ to connect $\xi_i$ to $\eta_i$.

2827: Therefore $\xi_i, \eta_i, \zeta_i$ are vertices of the broken

2828: geodesic triangle $df(\tau)\subset S_i$.

2829: 

2830: 

2831: As in Lemma \ref{consistent}, we use the isomorphism $S_i\to

2832: \t{S}_i$ (sending $s_i$ to $\t{s}_i$) to identify these apartments.

2833: Under this identification, $df: Y\to S_i$ is the retraction

2834: $Fold_{s_i, S_i}$ of $Y$ to the apartment $S_i$. Thus $df(\tau)$ is

2835: a folded triangle in $S_i$.

2836: 

2837: 

2838: Hence, by the (rank) induction hypothesis, for each $i$ there exists

2839: a chain

2840: $$

2841: (-\xi_i,..., \eta_i)

2842: $$

2843: in the spherical Coxeter complex $(S_i, W_{i}, s_i)$. \qed

2844: 

2845: 

2846: 

2847: 

2848: \begin{lem}

2849: For each path $p: [0, c]\to A$ satisfying the conclusion of Lemma

2850: \ref{trivial}, there exists a point $y'\in A$ such that

2851: $$

2852: d_{ref}(x, y')= \be %= \Length(p),

2853: $$

2854: and

2855: $$

2856: y' \ge p(c)

2857: $$

2858: in $(A,W,-\De)$.

2859: \end{lem}

2860: \proof We use the second induction, on the number $n$ of vertices in

2861: the broken geodesic $p$. Set $u:= p(c)$.

2862: 

2863: 

2864: The metric length of the path $p$ equals the metric length of the

2865: path $\t{p}=\ol{x y}$, the tangent directions of these paths at $x$

2866: are the same. Therefore, if $n=0$ (and hence the path $p$ is

2867: geodesic) there is nothing to prove, one can simply take $y'=u$.

2868: 

2869: 

2870: Assume that the assertion holds for all $n\le N-1$, let's prove it

2871: for $N$. We treat the path $p$ as the composition

2872: $$

2873: p|[0, t_N]\cup \ol{x_N u}

2874: $$

2875: Our goal is to replace the geodesic subpath $\ol{x_N u}$ with a

2876: geodesic path $w( \ol{x_N u})$, where $w\in W$ is fixing $x_N$, so

2877: that:

2878: 

2879: 

2880: 1. $\ol{x_{N-1} x_N}\cup w( \ol{x_N u})$ is a geodesic segment.

2881: 

2882: 

2883: 2. There exists an $(A, W, -\De)$-chain between $w(u)$ and $u$. Then

2884: we would be done by the induction on $n$. Indeed, the new path

2885: $$

2886: p|[0, t_N]\cup w(\ol{x_N u})

2887: $$

2888: has one less break-point and still satisfies the conclusion of Lemma

2889: \ref{trivial}. Thus, by the induction hypothesis, there exists

2890: $y'\in A$ so that

2891: $$

2892: y' \ge w(u) \ge u \Rightarrow y' \ge u,

2893: $$

2894: $$

2895: d_{ref}(x, y')= d_{ref}(x, y).

2896: $$

2897: 

2898: 

2899: {\bf Construction of $w$.} Recall that there exists an $(S_N, W_{N},

2900: -\De_N)$-chain

2901: $$

2902: (-\xi_N=\nu_0,..., \nu_k=\eta_N),

2903: $$

2904: hence we have a sequence of reflections  $r_1,...,r_k\in W_{N}$

2905: (fixing walls $H_i\subset S_N$, $i=1,...,k$) so that:

2906: $$

2907: r_i(\nu_{i-1})=\nu_i, i=1,...,k,

2908: $$

2909: and each wall $H_i$ separates $\nu_{i}$ from the negative chamber

2910: $s_N$. We extend each reflection $r_i$ from $S_N$ to a reflection

2911: $r_i$ in $A$, and each $H_i$ to a wall $H_i$ in $A$.

2912: 

2913: 

2914: We therefore define the following points in $A$:

2915: $$

2916: y_k:=u, y_{k-1}:= r_{k}(y_k), y_{k-2}:= r_{k-1}(y_{k-1}),\ldots,

2917: y_{0}:= r_{1}(y_1).

2918: $$

2919: Note that the directions $\nu_i$ are tangent to the segments

2920: $\ol{x_N y_i}$. Thus for each $i$, the wall $H_i$ separates the

2921: point $y_{i}$ from the negative chamber $a\subset A$ and the

2922: sequence

2923: $$

2924: (y_0,...,y_k=u)

2925: $$

2926: forms a chain. We set $w=r_1\circ ... \circ r_k$. The vector $\nu_0$

2927: is antipodal to $-\xi_N$, hence the path

2928: $$

2929: p|[0, T_N] \cup w(\ol{x_N u})

2930: $$

2931: is geodesic at the point $x_N=p(T_N)$. \qed

2932: 

2933: 

2934: This concludes the proof of Theorem in the case when $z\notin\ol{x

2935: y}$.

2936: 

2937: 

2938: We now consider the special case when the above proof has to be

2939: modified: The triangle $T$ is degenerate, i.e. $z\in \ol{xy}$, but

2940: the alcove $a$ is such that $y\notin A$.  Thus the folding

2941: $Fold_{a,A}$ is not an isometry on $T$. Then the tangent direction

2942: $\t\zeta_i$ is not defined when $x_i=z$. Note that $x_i=z$ then is

2943: the only break-point in the broken side of $T'$.

2944: 

2945: 

2946: In this case we replace the vertex $z$ with an arbitrary point $z'$

2947: in the interior of $a$ and repeat the above arguments. The chains

2948: constructed in the process will be independent of the choice of $z'$

2949: and thus, after taking the limit $z'\to z$, we obtain a chain as

2950: required by the assertion of Theorem. \qed

2951: 

2952: 

2953: 

2954: 

2955: \begin{cor}

2956: Cf. \cite[Theorem 8.2, Part 4]{KLM3}.  Suppose that $X$ is a Euclidean building.

2957: Assume that $\al:=d_\De(z,x), \be:=d_\De(x,y), \ga:=d_\De(y,z)$ are in $P(R^\vee)$

2958: and $x, y, z$ are special vertices of $X$. Then

2959: $$

2960: \al+\be+\ga \in Q(R^\vee).

2961: $$

2962: \end{cor}

2963: \proof

2964: Let $(y_0,...,y_m)$ be an affine chain given by Theorem \ref{T2}.

2965: We regard the point $x$ as the origin $o$ in $A$; thus we will regard $z$, $y_0, y_m$ as vectors in $V$.

2966: Then, according to Lemma \ref{difference},

2967: $$

2968: y_m-y_0\in Q(R^\vee).

2969: $$

2970: Consider the vectors $\be':= y_0-x, \ga':= z-y_m, \al':= x-z$ in $P(R^\vee)$. By the definition of $\De$-length,

2971: $$

2972: \al'\in W_{sph} \al, \quad \be'\in W_{sph} \be, \quad \ga'\in W_{sph} \ga.$$

2973:  Therefore, by applying  Lemma \ref{difference} again we see that

2974: the differences

2975: $$

2976: \al-\al',  \quad \be-\be',  \quad \ga-\ga'

2977: $$

2978: all belong to $Q(R^\vee)$. Since

2979: $$

2980: \al'+\be'+\ga'=y_0-y_m\in Q(R^\vee),

2981: $$

2982: the assertion of lemma follows. \qed

2983: 

2984: 

2985: 

2986: 

2987: 

2988: \medskip

2989: The following simple proposition establishes a partial converse to

2990: Theorem \ref{T2}:

2991: 

2992: 

2993: \begin{prop}

2994: \label{simplechain} Suppose that $X$ is a thick spherical building and, as before, the point $z$ belongs to a

2995: negative chamber $a=-\De$. Then, for each simple chain $(y_0, y_1)$

2996: such that $d_{ref}(x, y_0)=\pi, d_{ref}(z, y_1)=\be$, there exists a

2997: point $y\in X$ so that

2998: $$

2999: d_{ref}(y, z)= \ga \quad \hbox{and}\quad  d_{ref}(x , y)= \pi.$$

3000: \end{prop}

3001: 

3002: \begin{rem}

3003: Recall that $d_{ref}(x , y)= \pi$ means that the points $x$ and $y$ are antipodal.

3004: \end{rem}

3005: 

3006: \proof Let $\tau(y_0)=y_1$ where $\tau$ is a reflection in a wall

3007: $H\subset A$ as in the definition of a chain. Let $A=A^-\cup A^+$ be

3008: the union of half-apartments, where $A^-$ is bounded by $H$ and contains $a$.

3009: By the definition of a chain, $y_1\in A^-, y_0\in A^+$

3010: and hence the antipodal point $x=-y_0$ belongs to $A^+$.

3011: 

3012: 

3013: Since $X$ is thick, there exists a half-apartment $B^-\subset X$

3014: which intersects $A$ along $H$. Define the apartment $B:= A^-\cup

3015: B^-$; then there exists  an isomorphism of Coxeter complexes

3016: $$

3017: \phi: A\to B, \phi|A^-=id.

3018: $$

3019: We set $y:= \phi(y_0)$.

3020: 

3021: 

3022: Since $\phi$ is an isomorphism of Coxeter complexes which fixes $z$,

3023: it preserves the refined distance to the point $z$ and hence

3024: $$

3025: d_{ref}(y, z)= d_{ref}(y_1, z)=\ga.$$

3026: 

3027: 

3028: The union $C:= A^+\cup B^-$ is also an apartment in $X$. Then there

3029: exists an isomorphism  $\psi: B^-\to A^-$ so that

3030: $$

3031: \psi\circ\phi|A^+= \tau|A^+.

3032: $$

3033: The isomorphism $\psi$ extends to an isomorphism $\rho: C\to A$

3034: fixing $A^+$ pointwise and hence fixing the point $x$. Therefore

3035: $$

3036: d_{ref}(x , y)= d_{ref}(x, y_0)= \pi.  \qed

3037: $$

3038: 

3039: 

3040: 

3041: 

3042: 

3043: \subsection{Folding polygons in Euclidean buildings}

3044: \label{euclideanfolding}

3045: 

3046: 

3047: Our next goal is to show that each folding transforms certain piecewise-linear

3048: paths in Euclidean buildings to paths satisfying the chain condition.

3049: 

3050: 

3051: Suppose that $X$ is a Euclidean building with model apartment

3052: $(A,W_{aff})$, $\De\subset A$ is the positive Weyl chamber with the

3053: tip $o$. Consider a piecewise-linear path $\tilde{p}: [0,c]\to \t{A}$, which is

3054: parameterized  with the unit speed, where $\tilde{A}\subset X$ is an

3055: apartment. We assume that for each $t\in [0,c]$

3056: $$

3057: \tilde{p}'_-(t) \sim_{W_{sph}} \tilde{p}'_+(t),

3058: $$

3059: for instance, $\tilde{p}$ could be a geodesic path.

3060: 

3061: 

3062: Thus the path $\tilde{p}$ {\em trivially satisfies the chain

3063: condition}. Let $g: X\to \De$ be a folding into $\De$,

3064: $g=Fold_{z,h,\De}$ for a certain $z\in A$ and $h$. Recall that the

3065: folding $g$ is the composition of three maps:

3066: $$

3067: g= \P_{\Del} \circ h \circ f, \quad f=Fold_{a,A},

3068: $$

3069: where $a$ is an alcove in $A$ containing $z$, $h\in Dil(A,W_{aff})$ is a

3070: dilation sending $z$ to the point $o$. Consider

3071: the structure of a Coxeter complex on $A$ given by the pull-back

3072: $$

3073: h^*(A,W_{aff}).

3074: $$

3075: We thus get a new (typically non-thick) building structure for $X$,

3076: the one modeled on $h^*(A,W_{aff})$.

3077: 

3078: \begin{defn}

3079: We say that a path $\tilde{p}$ is {\em generic} if it is disjoint

3080: from $z$ and from the codimension 2 skeleton of $X$ and the

3081: break-points of $\tilde{p}$ are disjoint from the codimension 1

3082: skeleton of $X$, where $X$ is regarded as a building modeled on

3083: $h^*(A,W_{aff})$.

3084: \end{defn}

3085: 

3086: 

3087: 

3088: The main result of this section is

3089: 

3090: 

3091: \begin{thm}

3092: \label{chaincondition} The folded path $p= g(\tilde{p})$ satisfies

3093: the chain condition.

3094: \end{thm}

3095: \proof The proof of this theorem is mostly similar (except for the

3096: projection $\P$ which causes extra complications) to the proof of

3097: Theorem \ref{T2} in the previous section. We will prove Theorem

3098: \ref{chaincondition} in two steps: We first establish it for the

3099: paths $\tilde{p}$ which are {\em generic}. Then we use the {\em

3100: compactness theorem} to prove it in general.

3101: 

3102: 

3103: \begin{prop}

3104: \label{chaincondition0} The conclusion of Theorem

3105: \ref{chaincondition} holds for {\em generic} paths $\tilde{p}$.

3106: \end{prop}

3107: \proof If a point $\tilde{x}=\tilde{p}(t)$ is a regular point of

3108: $X$, then

3109: $$

3110: dg_{\tilde{x}}: \Si_{\tilde{x}}(X)\to \Si_x(A), \quad x=g(\tilde{x})

3111: $$

3112: is an isometry. Thus the path $p$ trivially satisfies the chain

3113: condition at the point $x$.

3114: 

3115: 

3116: Therefore we assume that $\tilde{x}$ is a singular point of $X$. Since

3117: $\tilde{p}$ is assumed to be generic, this point lies on exactly one

3118: wall of $X$; moreover, $\tilde{p}$ is geodesic near $\tilde{x}$.

3119: 

3120: 

3121: 

3122: 

3123: We first analyze what happens to the germ of $\tilde{p}$ at

3124: $\tilde{x}$ under the retraction $f$. We suppose that the

3125: restriction of $f$ to the germ $(\t{p}, \t{x})$ is not an isometry

3126: (otherwise there is nothing to discuss).  Let $\tilde\zeta\in

3127: \Si_{\tilde{x}}(X)$ denote the tangent to the geodesic

3128: segment $\ol{\tilde{x}z}$. Let $\eta\in \Si_{\tilde{x}}(X\cap

3129: \tilde{A})$ be the tangent vector $p'(t)$, $\tilde\xi:= -\tilde\eta$

3130: (this vector is also tangent to the path $p$). Set $\be:=

3131: d_{ref}(\tilde\zeta, \tilde\eta)$.

3132: 

3133: 

3134: We obtain the  triangle $\tau=[\tilde\xi, \tilde\eta,

3135: \tilde\zeta]$ in $\Si_{\tilde{x}}(X)$. The derivative of the

3136: retraction $f$ at $\t{x}$ is a retraction of the spherical building

3137: $\Si_{\t{x}}(X)$ into its apartment $S$, after identification of $S$

3138: with the sphere $S_{x'}(A)$, $x':=f(\t{x})$ (see Lemma

3139: \ref{consistent}). Define the following elements of $S$:

3140: $$

3141: \eta':=df_{\tilde{x}}(\tilde\eta), \quad \xi':=

3142: df_{\tilde{x}}(\tilde\xi), \quad

3143: \zeta':=df_{\tilde{x}}(\tilde\zeta).

3144: $$

3145: 

3146: 

3147: Therefore, according to Theorem \ref{T2}, the folded triangle

3148: $\tau'=df_x(\tau)\subset S$ yields an $(S,W_{x'},

3149: -\De_{x'})$-chain\footnote{Which is necessarily simple since

3150: $\tilde{p}$ is assumed generic.}

3151: $$

3152: (\mu'=-\xi', \eta')

3153: $$

3154: such that $d_{ref}(\zeta, \eta')= d_{ref}(\t\zeta, \t\eta)=\be$.

3155: 

3156: 

3157: Here $-\De_{x'}$ is a chamber in $(S,W_{x'})$ which contains

3158: the unit tangent vector to the segment $\ol{x' z'}$ where $z'\in a$ is a regular point.

3159: 

3160: 

3161: \begin{rem}

3162: Note that our assumptions on $\tilde{p}$ imply that $(S,W_{x'})$

3163: has a unique wall. If the corresponding wall in $A$ does not

3164: pass through $z$, then the negative chamber $-\De_{x'}$  in $(S,W_{x'})$ is

3165: uniquely determined by the condition that it contains the direction

3166: tangent to

3167:  $\ol{ x' z}$.

3168: \end{rem}

3169: 

3170: 

3171: Consider now the effect of the rest of the folding $g$ on the path

3172: $\tilde{p}$ at $\tilde{x}$. Let $x:= g(\tilde{x})$. We identify $S$

3173: with the unit tangent sphere at the point $x$.

3174: 

3175: 

3176: The dilation $h$ clearly preserves the chain condition at $x'$

3177: (since it acts trivially on the unit tangent sphere). The

3178: restriction of the projection $\P=\P_{\Del}$ to the germ of

3179: $hf(\tilde{p})$ at $hf(\tilde{x})$ is necessarily an isometry (since

3180: $\tilde{p}$ is generic), hence it is given by an element $w\in

3181: W=W_{sph}$. This element transforms the above chain to another

3182: $(S,W_x, -\De_x)$--chain, where

3183: $$

3184: \De_x:= d(w\circ h) (\De_{x'}).

3185: $$

3186: What is left to verify is that the positive chamber $\De_x$ in this

3187: complex contains a translate of the positive chamber $\De$. In case

3188: when $x$ belongs to the interior of $\De$, the segment $\ol{ox}$ is

3189: not contained in any wall and thus the negative chamber $-\De_x$ has

3190: to contain the initial direction of the segment $\ol{xz}$ (see the

3191: remark above). However this initial direction belongs to $-\Del$ and

3192: thus $\De_x$ contains $\De$.

3193: 

3194: 

3195: Consider the exceptional case when $x$ is on the boundary of $\De$.

3196: It then belongs to a unique wall $H$ in the Coxeter complex

3197: $(A,W_{aff})$ and this wall passes through the origin $o$. Rather

3198: than trying to use Theorem \ref{T2} to verify the chain condition at

3199: $x$, we give a direct argument. Let $\eta, \xi$ be the unit vectors

3200: which are the images of $\eta', \xi'$ under

3201: $$

3202: d(w\circ h): \Si_{x'}(A)\to \Si_x(A).

3203: $$

3204: Since the path $p$ is entirely contained in $\De$, the vector

3205: $p'_-(t)$ points {\em outside} of $\De$ and the vector $p'_+(t)$

3206: points {\em inside}. The reflection $\si$ in the wall $H$ sends the

3207: vector $-\xi=p'_-(t)$ the vector $\eta=p'_+(t)$. It is then clear

3208: that the (simple) chain condition is satisfied at the point $x$.

3209: 

3210: 

3211: \medskip

3212: Lastly, we consider the points $\tilde{x}=\tilde{p}(t)$ for which

3213: $f$ is an isometry on the germ of $\tilde{p}$ at $\tilde{x}$.

3214: The point $x=g(\tilde{x})$ belongs to

3215: a face of $\De$ contained in a wall $H$, and this is the only wall

3216: of $(A,W_{aff})$ which passes through $x$. Then,

3217: necessarily, the germ of the path $hf(\tilde{p})$ at

3218: $hf(\tilde{x})$ is a geodesic.  We now simply repeat the

3219: arguments of the exceptional case in the above proof (see also the

3220: proof of Proposition \ref{chaincondition1}) to see that $p$

3221: satisfies the chain condition at $x$. \qed

3222: 

3223: 

3224: \medskip

3225: We are now ready to prove Theorem \ref{chaincondition} for arbitrary

3226: paths $\tilde{p}$. We will do so by approximating the path

3227: $\tilde{p}$ via {\em generic paths}. Let $\la$ be an arbitrary

3228: vector in $\t{A}$. We let $\tilde{q}_\la:=\tilde{p}+\la$ denote the

3229: translation of the path $\tilde{p}$ by the vector $\la$. It is

3230: clear, from the dimension count, that for an open and dense set of

3231: vectors $\la$, the path $\tilde{q}_\la$ is {\em generic}.

3232: 

3233: 

3234: Since the folding $g$ is continuous,

3235: $$

3236: p=g(\tilde{p})= \lim_{\la\to 0} g(\tilde{q}_\la).

3237: $$

3238: By the Proposition \ref{chaincondition0}, each $g(\tilde{q}_\la)$

3239: satisfies the chain condition. Observe that the $\De$-lengths of the

3240: paths $\tilde{p}+\la$ are independent of $\la$. Since $f$ and $\P$

3241: preserve $\De$-lengths of piecewise-linear paths and the dilation $h$ changes them

3242: by a fixed amount, we can apply the compactness theorem (Theorem

3243: \ref{compactness}) to conclude that the limiting path $p$ satisfies

3244: the chain condition as well. \qed

3245: 

3246: 

3247: \medskip

3248: We now verify that, at certain points, the folded path $p$ satisfies

3249: the {\em maximal chain condition}.

3250: 

3251: 

3252: 

3253: 

3254: \begin{prop}

3255: \label{chaincondition1} Under the assumptions of Theorem

3256: \ref{chaincondition} let $\tilde{x}=\tilde{p}(t)$ be such that the

3257: folding $f$ restricts to an isometry on the germ $(\tilde{p},

3258: \tilde{x})$. Then the path $p=g(\tilde{p})$ satisfies the {\em

3259: maximal chain condition} at $x=g(\tilde{x})$.

3260: \end{prop}

3261: \proof Our proof follows Littelmann's arguments in his proof of the

3262: PRV Conjecture  \cite{Littelmann1}. We fold the path

3263: $q:=hf(\tilde{p})$ into $\De$ inductively.

3264: 

3265: 

3266: We subdivide the interval $[0,c]$ as

3267: $$

3268: 0=t_0 < t_1<...< t_k=c

3269: $$

3270: such that $[t_i, t_{i+1}]$ are maximal subintervals so that $q|[t_i,

3271: t_{i+1}]$ is contained in a Weyl chamber of $W_{sph}$.

3272: 

3273: 

3274: We first apply to $q$ an element $w_0\in W_{sph}$ which sends $q([0,

3275: t_1])$ into $\De$, so we can assume that this subpath belongs to

3276: $\De$. Assume that the restriction of $f$ to the germ $(\t{p}, \t{p}(t_1))$ is an isometry.

3277: Let $\mu', \eta'$ be the vectors $q'_-(t_1), q'_+(t_1)$. Then

3278: $$

3279: \mu'\sim \eta',

3280: $$

3281: see Lemma \ref{positivity}. Set $x:= q(t_1)$. The image $\eta$ of

3282: the vector $\eta'$ under $\P$ is obtained as

3283: $$

3284: dw_1(\eta'),

3285: $$

3286: where $w_1\in W_{sph}$ fixes the point $x$ and $\eta$ is the unique

3287: vector in the $W_x$-orbit of $\eta'\in S_{x}$ which points inside

3288: $\De$. Below we describe $w_1$ as a composition of reflections.

3289: 

3290: 

3291: Let $R'$ denote the root subsystem in $R$ generated by the set of

3292: simple roots $\Phi'$ which vanish at the point $x$. Let $\De'$

3293: denote the positive chamber for $W_x$ defined via $\Phi'$. Then the

3294: vector $\eta$ can be described as the unique vector in the

3295: $W_x$-orbit of $\eta'$ (now, regarded as a vector in $V=T_o(A)$)

3296: which belongs to the interior of $\De'$. According to Lemma

3297: \ref{simplesubsystem},

3298: $$

3299: w_1= \tau_m\circ .. \circ \tau_1, \hbox{~where for each~} i, \quad

3300: \tau_i=\tau_{\be_i}, \be_i\in \Phi',

3301: $$

3302: so that the sequence of vectors

3303: $$

3304: (\eta_0=\eta', \eta_1:=

3305: \tau_1(\eta_{0}),...,\eta_m=\tau_m(\eta_{m-1})=\eta),$$ is a chain

3306: in $(S, W_x,-\De)$ which is maximal as a chain in $(S,W_{sph},-\De)$.

3307: 

3308: 

3309: We therefore apply the identity transformation to the path $q|[0,

3310: t_1]$ and the element $w_1$ to the path $q|[t_1, c]$ to transform

3311: the path $q$ to the new path

3312: $$

3313: q_1= q|[0, t_1] \cup w_1\circ q|[t_1, c].

3314: $$

3315: Clearly, $\P(q)= \P(q_1)$ and $\eta$ is the unit vector tangent to

3316: $q_1|[t_1,c]$ at $x$. The above arguments therefore show that $q_1$

3317: satisfies the {\em maximal chain condition} at the point $x$.

3318: 

3319: 

3320: We then proceed to the next point $t_2$, $q_1(t_2)$ belongs to the

3321: boundary of $\De$ and we transform $q_1$ to $q_2$ by

3322: $$

3323: q_2|[0,t_2]=q_1|[0,t_2], \quad q_2|[t_2, c]= w_2\circ q_1|[t_2,c],

3324: $$

3325: where $w_2$ is a certain element of $W_{sph}$ fixing

3326: $q_1(t_2)=q_2(t_2)$. Therefore $\P(q_2)=\P(q_1)=\P(q)$ and we repeat the

3327: above argument. \qed

3328: 

3329: 

3330: \begin{defn}

3331: Suppose that $P=\ol{zx}\cup p \cup \ol{yz}$ is a polygon in $A$,

3332: where $p:[0,1]\to A$ is a piecewise-linear path such that $p(0)=x, p(1)=y$. We say

3333: that $P$ {\em satisfies the chain condition} (resp. simple chain

3334: condition, resp. maximal chain condition) if its subpath $p$

3335: satisfies the chain condition (resp. simple chain condition, resp.

3336: maximal chain condition).

3337: \end{defn}

3338: 

3339: 

3340: Therefore, as an application of Theorem \ref{chaincondition} we

3341: obtain

3342: 

3343: 

3344: 

3345: 

3346: \begin{cor}

3347: \label{folded-chain}

3348:  Suppose that $T=[\t{z}, \t{x}, \t{y}]\subset X$ is a

3349: geodesic triangle, $\t{z}$ is a special vertex which belongs to an

3350: alcove $a\subset A$. Let $\De\subset A$ be a Weyl chamber with the

3351: tip $\t{z}=o$ and $P=Fold_{\De}(T)$ be the folding of $T$ into $\De$. Then

3352: the folded triangle $P$ satisfies the chain condition.

3353: \end{cor}

3354: 

3355: 

3356: 

3357: 

3358: A converse to this corollary will be proven in Theorem

3359: \ref{characterization}; the following is a {\em partial} converse to

3360: Corollary \ref{folded-chain} (which is essentially contained in

3361: \cite[Lemma 7.7]{KLM3}):

3362: 

3363: 

3364: \begin{cor}

3365: \label{chain-folded0} Let $\De\subset A$ be a Weyl chamber with  tip

3366: $o$ in $X$. Suppose that a polygon $P= [o, x_1,...,x_{n}]\subset

3367: \De$, satisfies the {\em simple chain} condition (at each vertex

3368: $x_i, 0<i<n$) and

3369: $$

3370: p=\ol{x_1 x_2}\cup ... \cup \ol{x_{n-1} x_n}

3371: $$

3372: is a {\em billiard} path. Then $P$ unfolds to a geodesic triangle

3373: $T\subset X$, i.e. $Fold_{\De}(T)=P$.

3374: \end{cor}

3375: \proof Let $f:=Fold_{\De}$.  We run the argument from the proof of

3376: Theorem \ref{T2} in the reverse; the reader will observe that our

3377: argument is essentially the same as in the proof of the Transfer

3378: Theorem in \cite{KLM2}. Triangulating the polygon $P$ from the

3379: vertex $o$ we obtain geodesic triangles $P_i=[o, x_i, x_{i+1}]$, $i=1, 2,...,n-1$.

3380: Let $\xi_i, \zeta_i, \eta_i \in \Si_{x_i}(A)$ denote the unit tangent vectors to the

3381: segments $\ol{x_i x_{i-1}}, \ol{x_i o}, \ol{x_i x_{i+1}}$

3382: respectively.

3383: 

3384: We unfold $P$ inductively. Set $T_1:= P_1$; let $A_1:= A$,

3385: this apartment contains the triangle $T_1$. Set $\t{x}_1:= x_1, \t{x}_2:= x_2$,

3386: 

3387: Suppose that we have constructed apartments $A_i\subset X$ and flat triangles

3388: $T_i=[o, \t{x}_i, \t{x}_{i+1}]\subset A_i$, $i=1,...,m-1$,

3389: so that $T_i$ is congruent to $P_i$ ($i=1,...,m-1$) and

3390: $$

3391: \angle( \t{\xi}_i, \t{\eta}_i)=\pi, i=1,...,m-1.

3392: $$

3393: Here $\t{\xi}_i, \t{\eta}_i, \t{\zeta}_i$

3394: are directions in $\Si_{\t{x}_i}(X)$ which correspond to the directions $\xi_i, \eta_i, \zeta_i$ under the

3395: congruences $T_i\to P_i$. Our goal is to produce a flat triangle $T_m\subset A_m\subset X$ so that the above

3396: properties still hold.

3397: 

3398: Since we have a simple chain $(-\xi_m, \eta_m)$ in

3399: $(S_{x_m}, W_{x_m})$, it follows from Proposition \ref{simplechain}

3400: that there exists a point $\tilde{\eta}_m\in \Si_{\t{x}_m}(X)$ so that

3401: $$

3402: d(\tilde{\eta}_m, \t{\xi}_m)=\pi, \quad d(\tilde{\eta}_m,

3403: \zeta_m)=d_{ref}(\eta_m, \zeta_m).

3404: $$

3405: Let $A_m$ denote an apartment in $X$ which contains $\ol{o \t{x}_m}$ and

3406: such that $\tilde{\eta}_m$ is tangent to $A_m$. Construct a geodesic segment

3407: $\ol{\t{x}_m \t{x}_{m+1}}\subset A_m\subset X$ whose metric length equals the one of $\ol{x_m x_{m+1}}$

3408: and whose initial direction is $\tilde{\eta}_m$. This defines a flat triangle

3409: $$

3410: T_m=[o, \t{x}_m, \t{x}_{m+1}]\subset A_m.

3411: $$

3412: It is clear from the construction that the triangle $T_m$ is congruent to $P_m$, in particular,

3413: $d_{ref}(o, x_3)=d_{ref}(o, y_3)$. Observe also that

3414: $$

3415: \ol{\t{x}_{m-1} \t{x}_m}\cup \ol{\t{x}_m \t{x}_{m+1}}$$

3416: is a geodesic segment (because $\angle(\tilde{\eta}_m, \t{\xi}_m)=\pi$). See Figure \ref{unfold.fig}.

3417: 

3418: Therefore, by induction we obtain a geodesic triangle $T=[o, x_1, \t{x}_n]\subset X$, which

3419: is triangulated (from $o$) into flat geodesic triangles $T_i$ which are congruent  to $P_i$'s.

3420: We claim that $f(T)=P$. For each $i$ the folding $f$ sends the

3421: triangle $T_i$ to $P_i$, according to Proposition \ref{uniqueness}.

3422: Therefore $f(T)=P$.

3423: 

3424: 

3425: 

3426: \begin{figure}[tbh]

3427: \centerline{\epsfxsize=5in \epsfbox{unfold.eps}} \caption{\sl

3428: Unfolding a broken triangle.} \label{unfold.fig}

3429: \end{figure}

3430: 

3431: 

3432: \medskip

3433: As in the proof of Theorem \ref{T2}, the argument has to be modified

3434: in case when $x_i=o$ for some $i$, $x_i=p(t_i)$. Then the vector

3435: $\mu=p'_-(t_i)$ belongs to the negative chamber $-\De$ and the

3436: vector $\la=p'_+(t_i)$ belongs to the positive chamber $\De$. Since

3437: $p$ is a billiard path, there exists $w\in W_{sph}$ which sends

3438: $\mu$ to $\la$. Now the chain and billiard conditions imply that

3439: $x_i$ is the  only break-point in $p$. Thus we can take

3440: $$

3441: T:= \ol{ox_1} \cup p([0,t_i]) \cup w^{-1} p([t_i, 1]) \cup

3442: w^{-1}(\ol{ox_n}).

3443: $$

3444: This degenerate geodesic triangle (it is contained in the geodesic

3445: through the points $x_1, w(x_n)$) folds to $P$ under the projection

3446: $\P_\De: A\to \De$. \qed

3447: 

3448: 

3449: 

3450: 

3451: \medskip

3452: The same argument as above proves the following generalization of

3453: Corollary \ref{chain-folded0}

3454: 

3455: 

3456: \begin{cor}

3457: \label{chain-folded} Suppose that $P$ is a  polygon in $\De$ which

3458: is the composition

3459: $$

3460: \ol{ox} \cup   p \cup  q \cup  \ol{yo}.

3461: $$

3462: Assume that paths $p, q$ satisfy the simple chain condition. Then

3463: there exists a polygon $\tilde{P}\subset X$ of the form

3464: $$

3465: \ol{ox} \cup \tilde{p} \cup  \tilde{q} \cup \ol{\t{y}o}

3466: $$

3467: so that $f(\tilde{P})=P$, $f(\tilde{p})=p, f(\tilde{q})=q$,

3468: $f(\t{y})=y$, and $\tilde{p}, \tilde{q}$ are geodesic paths.

3469: \end{cor}

3470: 

3471: 

3472: 

3473: 

3474: 

3475: 

3476:  We now use our analysis of the folded triangles

3477: (polygons) to relate them to the Littelmann triangles (polygons).

3478: 

3479: 

3480: \section{Littelmann polygons}

3481: \label{LSpolygons}

3482: 

3483: 

3484: \subsection{LS paths}

3485: \label{LSpaths}

3486: 

3487: 

3488:  Let $R$ be a root system on a Euclidean vector space

3489: $V$, $W=W_{sph}$ be the finite Coxeter group associated with $R$,

3490: let $W_{aff}$ denote the affine Coxeter group associated to $R$.

3491: This root system $R$ is actually the {\em coroot system} for the one

3492: considered by Littelmann in \cite{Littelmann2}. Accordingly, we will

3493: switch weights to coweights, etc. We pick a Weyl chamber $\Del$ for $W$, this

3494: determines the positive

3495: roots and the simple roots in $R$. We get the Euclidean Coxeter

3496: complex $(A,W_{aff})$, where $A$ is the affine space corresponding

3497: to $V$. Given $x\in A$ let $W_x$ denote the stabilizer of $x$ in $W_{aff}$.

3498: 

3499: 

3500: \medskip

3501: Suppose we are given a  vector $\la\in  \De\subset V$, a

3502: sequence of real numbers

3503: $$

3504: \ul{a}=( a_0=0< a_1 < ... < a_r=1),

3505: $$

3506: and a sequence of vectors in $W\la$

3507: $$

3508: \ul\nu= (\nu_1, ... , \nu_r), \hbox{~~so that~~} \nu_1 > ... > \nu_r

3509: $$

3510: with respect to the order in Definition \ref{order}.

3511: 

3512: 

3513: \begin{defn}

3514: The pair $(\ul\nu, \ul{a})$ is called a {\em real (billiard) path}

3515: of the $\De$-length $\la$.

3516: \end{defn}

3517: 

3518: 

3519: \begin{defn}

3520: (P. Littelmann \cite{Littelmann2}.) A real path of $\De$-length

3521: $\la$ is called {\em rational} if  $\la$ is a coweight and all

3522: numbers $a_i$ a rational.

3523: \end{defn}

3524: 

3525: 

3526: \begin{rem}

3527: Littelmann uses the notion {\em path of type $\la$} rather than of

3528: the $\De$-length $\la$.

3529: \end{rem}

3530: 

3531: 

3532: Set $a_i':= a_i -a_{i-1}$, $i=1, 2,...,r$. The data $(\ul\nu,

3533: \ul{a})$ determines a piecewise-linear path $p\in {\mathcal P}$ whose restriction

3534: to each interval $[a_{i-1}, a_i]$ is given by

3535: \begin{equation}

3536: \label{path}

3537:  p(t)= \sum_{k=1}^{i-1} a_k' \nu_k + (t-a_{i-1})\nu_i,

3538: t\in [a_{i-1}, a_i].

3539: \end{equation}

3540: 

3541: 

3542: 

3543: 

3544: Our interpretation of real and rational paths is the one of a broken

3545: (oriented) geodesic $L$ in $V$. Each oriented geodesic subsegment of

3546: $L$ is parallel to a positive multiple of an element of $W\la$, thus

3547: $L$ is a billiard path. The break points of the above path are the

3548: points

3549: $$

3550: x_1=a_1\nu_1, \ldots, x_i=x_{i-1}+ a_i' \nu_{i}, \ldots

3551: $$

3552: Since $\sum_i a_i'=1$, is clear that

3553: $$

3554: \Length (L)=\la,

3555: $$

3556: in the sense of the definition in section \ref{distances}. This

3557: justifies our usage of the name $\De$-length $\la$ in the above

3558: definitions, rather than Littelmann's notion of {\em type}.

3559: 

3560: 

3561: 

3562: 

3563: Observe that given a piecewise-linear path $p(t)\in {\mathcal P}$ (parameterized

3564: with the constant speed) one can recover the nonzero vectors

3565: $\nu_i\in V$ and the numbers $a_i$ and $a_i'$.

3566: 

3567: 

3568: 

3569: 

3570: 

3571: 

3572: 

3573: 

3574: 

3575: 

3576: 

3577: 

3578: \begin{defn}

3579: \label{LSdef} (P. Littelmann \cite{Littelmann2}.)

3580: A rational path

3581: $p(t)$ is called an {\em LS path}\footnote{a Lakshmibai-Seshadri

3582: path} if it satisfies further {\em integrality condition}:

3583: 

3584: 

3585: For each $i=1,...,s-1$ there exists an $a_i$-chain for the pair

3586: $(\nu_i, \nu_{i+1})$ (in the sense of Definition \ref{achain}).

3587: \end{defn}

3588: 

3589: 

3590: Observe that, since

3591: $$

3592: \eta \ge \tau \iff -\tau \ge -\eta,

3593: $$

3594: it follows that $p$ is an LS path if and only if $p^*$ is.

3595: 

3596: 

3597: 

3598: 

3599: \begin{thm}

3600: \label{locint}

3601:  (P.\ Littelmann \cite[Lemma 4.5]{Littelmann2}) Each LS path belongs to

3602: ${\mathcal P}_{\Z,loc}$.

3603: \end{thm}

3604: 

3605: 

3606: Our next goal is to give a more geometric interpretation of LS

3607: paths. Suppose that $p\in {\mathcal P}$ is a billiard path given by

3608: the equation (\ref{path}), with the vertices

3609: $$

3610: 0=x_0, x_1,...,x_r.

3611: $$

3612: At each vertex point $x_i, 0<i<r$,  we have unit tangent vectors

3613: $\xi_i, \mu_i$ which are tangent to the segments $\ol{x_i x_{i-1}},

3614: \ol{x_i  x_{i+1}}$. Note that at each vertex $x_i, 0<i<r$ we have

3615: the restricted and unrestricted spherical Coxeter complexes; the

3616: positive chamber $\De$ in $V$ determines positive chambers $\De_i$

3617: in the restricted spherical complexes $(S_{x_i}, W_{x_i})$.

3618: 

3619: 

3620: 

3621: 

3622: \begin{thm}

3623: \label{T3} A billiard path $p(t)$ of $\De$--length $\la\in P(R^\vee)$

3624: is an LS path if and only if it is a Hecke path which satisfies the

3625: {\em maximal chain condition} (cf. Definition \ref{D3}): At each

3626: vertex $x_i, 0<i<r$ there exists a $(S_{x_i}, W_{x_i},-\De_i)$-chain

3627: between $-\xi_i$ and $\mu_i$ and this chain is maximal as a

3628: $(S_{x_i}, W)$-chain.

3629: \end{thm}

3630: \proof Recall that given a nonzero vector $v\in V$, $\bar{v}$

3631: denotes its normalization $v/|v|$.

3632: 

3633: 

3634: 

3635: 

3636: 

3637: 

3638: \begin{figure}[tbh]

3639: \centerline{\epsfxsize=4.5in \epsfbox{LS.eps}} \caption{\sl

3640: Unbending a path.} \label{LS.fig}

3641: \end{figure}

3642: 

3643: 

3644: 

3645: 

3646: It is easy to see (and left to the reader) that if $p(t)$ is a

3647: satisfies the above chain condition and $\la=\Length(p)$ is a

3648: coweight, then all numbers $a_i$ are rational.

3649: 

3650: 

3651: Consider the first break point $x_1=x_0+a_1\nu_1$ of the broken

3652: geodesic path $p(t)$. Observe that

3653: $$

3654: \bar\nu_1=-\xi_1, \bar\nu_2=\mu_1\in S_{x_1}.

3655: $$

3656: According to Proposition \ref{eq}, existence of an $a_1$-chain for

3657: the pair $(\nu_1, \nu_{2})$ is equivalent to existence of an

3658: $(S_{x_1}, W_{x_1}, \De_1)$-chain, which is maximal in the

3659: unrestricted Coxeter complex,

3660: $$

3661: (\bar\nu_1=\eta_{1,0}, \eta_{1,1},...,\eta_{1,s_1}=\bar\nu_2).

3662: $$

3663: Thus the path $p$ satisfies the maximal chain  condition at the

3664: first break point $x_1$ if and only if it satisfies the {\em

3665: integrality condition} as in Definition \ref{LSdef}, at the point

3666: $x_1$.

3667: 

3668: 

3669: We now proceed to the next break point $x_2=x_1+ a_2' \nu_2$. We

3670: identify normalized vectors $\bar\nu_1, \bar\nu_2$ with unit vectors

3671: in $S_{x_1}$. Note that if $p(t)$ is an LS path of the $\De$-length

3672: $\la$, then there exists an element $w_1\in W_{x_1}$ which sends

3673: $\bar\nu_1$ to $\bar\nu_2$. The same is true if $p$ is a Hecke path.

3674: 

3675: 

3676: 

3677: 

3678: Set $x_2':= w_1^{-1}(x_2)$. Observe that, in both cases of an LS

3679: path and a Hecke path,

3680: $$

3681: \ol{x_0 x_1}\cup \ol{x_1 x_2'}

3682: $$

3683: is a geodesic segment $\ol{x_0 x_2'}$; the corresponding directed

3684: segment represents the vector $a_2 \nu_1$. Let $w_1'\in W_{sph}$

3685: denote the linear part of $w_1$. Set $x_2'':=w_1'(x_2')$. We

3686: translate the vectors $\bar\nu_2, \bar\nu_3$ to the unit tangent

3687: sphere $S_{x_2''}$.  The directed segment $\ora{x_0 x_2''}$

3688: represents the vector $a_2 \nu_2$. See Figure \ref{LS.fig}.

3689: 

3690: 

3691: We are now again in position to apply Proposition \ref{eq} with

3692: $a=a_2$: There exists a maximal chain

3693: $$

3694: (\bar\nu_2=\eta_{2,0}, \eta_{2,1},...,\eta_{2,s_2}=\bar\nu_3)

3695: $$

3696: if and only if there exists an $a_2$-chain for the pair

3697: $(\nu_2,\nu_3)$. The product $w_1\circ (w_1')^{-1}$ is a translation

3698: in $W_{aff}$ which carries $x_2''$ back to $x_2$. Therefore it

3699: induces an isomorphism of the restricted Coxeter spherical complexes

3700: $$

3701: (S_{x_2''}, W_{x_2''})\to (S_{x_2}, W_{x_2})

3702: $$

3703: which carries positive chamber to positive chamber. Hence this

3704: translation sends the chain $(\eta_{2,i})$ to a maximal chain in

3705: $(S_{x_2}, W_{x_2})$.

3706: 

3707: 

3708: We continue in this fashion: On the $i$-th step we ``unbend'' the

3709: broken geodesic

3710: $$

3711: \ol{x_0 x_1}\cup ... \cup\ol{x_{i-1} x_i}

3712: $$

3713: to a directed geodesic segment $\ora{x_0 x_i'}$ representing the

3714: vector $a_i\nu_1$, then apply an appropriate element $w_{i-1}'\in

3715: W_{sph}$ to transform segment $\ol{x_0 x_i'}$  to $\ol{x_0 x_i''}$;

3716: finally, appeal to Proposition \ref{eq} to establish equivalence

3717: between the maximal chain condition and the LS path axioms. \qed

3718: 

3719: 

3720: As a corollary of Theorem \ref{T3} we obtain

3721: 

3722: 

3723: \begin{cor}

3724: \label{C3} Let $T=[z,x,y]\subset X$ be a geodesic triangle and

3725: $f=Fold_{z,h,\De}$ be a folding into the Weyl chamber. Set

3726: $\be:=d_\De (x,y)$. Assume that $T'=f(T)$ is such that $f(x), f(y)$

3727: and all break-points of the broken geodesic $f(\ol{xy})$ are special

3728: vertices. Then $f(\ol{xy})$ is an LS path of the $\De$-length $k\be$.

3729: Here $k$ is the conformal factor of the dilation $h$.

3730: \end{cor}

3731: 

3732: 

3733: 

3734: 

3735: 

3736: 

3737: \subsection{Root operators}

3738: \label{operators}

3739: 

3740: 

3741: With each simple root $\al\in \Phi$, Littelmann

3742: \cite{Littelmann2} associates {\em raising and lowering} root

3743: operators $e_\al$ and $f_\al$ acting ${\mathcal P}$ as follows.

3744: 

3745: 

3746: Recall that given a path $p(t)$ and a root $\al$ we have the height function

3747: $h_\al(t):=\al(p(t))$. The number $m_\al$ is the minimal value of

3748: $h_\al$ on $[0,1]$.

3749: 

3750: 

3751: If $m_\al>-1$ then $e_\al$ is not defined on $p$. Otherwise let

3752: $t_1$ be the minimal $t$ for which $h_\al(t)=m_\al$ and let $t_0\in

3753: [0, t_1]$ be maximal such that $h_\al(t)\ge m_\al+1$ for all $t\in

3754: [0, t_0]$.

3755: 

3756: 

3757: The operator $e_\al$ will not change the path $p$ for $t\in [0,

3758: t_0]$ and, as far as $[t_1, 1]$ is concerned, the path $p|[t_1, 1]$

3759: will change only by a translation in $W_{aff}$ along the line

3760: $L_\al$ parallel to the vector $\al^\vee$. Thus it remains to

3761: describe the path $q=e_\al(p)$ on $[t_0, t_1]$. If $h_\al$ were not

3762: to have any local minima on $[t_0, t_1]$ then $q|_{[t_0, t_1]}$

3763: would be obtained by the reflection

3764: $$

3765: q|_{[t_0, t_1]}:= \tau_\al\circ p|_{[t_0, t_1]}

3766: $$

3767: and we would set

3768: $$

3769: q:= p|_{[0,t_0]} *\tau_\al\circ p|_{[t_0, t_1]} * p|_{[t_1,1]}.

3770: $$

3771: (Here we treat the paths resulting from the restriction of $p$ to

3772: subintervals of $[0,1]$ as elements of ${\mathcal P}$, according to

3773: the convention in section \ref{prelim}.)

3774: 

3775: 

3776: This is the definition of $e_\al$ of \cite{Littelmann1}, however the definition of

3777: $e_\al$ which we will need in this paper is the more refined

3778: one of \cite{Littelmann2}. Call a  subinterval $[s,u]\subset

3779: [t_0, t_1]$ a {\em spike} if it is a maximal interval satisfying

3780: $$

3781: h_\al(s)=h_\al(u) =\min( h_\al \restr [s,u]).

3782: $$

3783:  Thus $h_\al\restr [t_0,t_1]$ is decreasing on the complement to the

3784: union of {\em spikes}. The restriction of $q$ to each {\em spike} is

3785: obtained from $p$ by a translation along $L_\al$. The restriction to

3786: each subinterval disjoint from a {\em spike} is obtained by a

3787: reflection. To be more precise, subdivide the interval $[t_0,t_1]$

3788: into

3789: $$

3790: [t_0, s_1]\cup [s_1,s_2]\cup ... \cup [s_k, t_1],

3791: $$

3792: where the {\em spike} and {\em non-spike} intervals alternate. Observe

3793: that $[t_0, s_1], [s_k, t_1]$ are not {\em spikes}. Then

3794: $$

3795: q:= p|_{[0,t_0]} *\tau_\al(p|_{[t_0, s_1]}) * p|_{[s_1,s_2]}*...

3796: *\tau_\al( p|_{[s_k, t_1]}) * p|_{[t_1,1]}.

3797: $$

3798: 

3799: 

3800: 

3801: 

3802: Note that the operator $e_\al$ changes the geometry of the path $p$

3803: by an isometry near every point $p(t)$ which is neither a point of

3804: local minimum for $h_\al$ nor is a point where $h_\al(t)=m_\al-1$.

3805: Otherwise the local change is done by a ``bending''  with respect to

3806: a hyperplane parallel to $H_\al$. These hyperplanes are not

3807: necessarily walls of $W_{aff}$. However, if all local minimal values

3808: of $h_\al$ belong to $\Z$, these hyperplanes are indeed walls and we

3809: obtain:

3810: 

3811: 

3812: For each path $p\in {\mathcal P}_{\Z,loc}$, for each simple root

3813: $\al$, the path $q=e_\al(p)$ satisfies the following: The interval

3814: $[0,1]$ can be subdivided into subintervals $[s_i, s_{i+1}]$ such

3815: that the restriction $q|[s_i, s_{i+1}]$ is obtained from the

3816: restriction of $p$ by post-composition with an element of $W_{aff}$.

3817: 

3818: 

3819: \medskip

3820: The lowering operators $f_\al$ are defined analogously to the

3821: raising operators, we refer the reader to \cite{Littelmann2} for the

3822: precise definition. (See however Property 1 below.) At this stage we

3823: note only that $f_\al$ is undefined on $p$ iff $m_\al> h_\al(1)-1$.

3824: Let ${\mathcal E}$ be the semigroup generated by $e_\al$'s,

3825: ${\mathcal F}$ be the semigroup generated by $f_\al$'s and

3826: ${\mathcal A}$ be the semigroup generated by all root operators. The

3827: semigroups contain the identity operator by default. For each

3828: $\phi\in {\mathcal A}$ let $Dom(\phi)$ denote the domain of $\phi$.

3829: 

3830: 

3831: 

3832: 

3833: 

3834: 

3835: \begin{rem}

3836: In fact, Littelmann extends the operators $f_\al, e_\al$ to the

3837: entire ${\mathcal P}$ by declaring $f_\al(p)={\mathbf 0}$ for all

3838: $p$ for which $f_\al$ is undefined. However we will not need this

3839: extension in the present paper.

3840: \end{rem}

3841: 

3842: 

3843: Below we list certain properties of the root operators. Most of them

3844: are either clear from the definition or are proven in

3845: \cite{Littelmann2}. Most proofs that we present are slight

3846: modifications of the arguments in \cite{Littelmann2}.

3847: 

3848: 

3849: \medskip

3850: {\bf Property 1.} (P. Littelmann, \cite[Lemma 2.1

3851:  (b, e)]{Littelmann2}.)

3852: $$

3853: e_\al\circ f_\al(p)= p, \hbox{~~if~~} p\in Dom(f_\al),

3854: $$

3855: $$

3856: f_\al\circ e_\al(p)= p, \hbox{~~if~~} p\in Dom(e_\al),

3857: $$

3858: $$

3859: e_\al(p^*)= (f_\al(p))^*, \quad (e_\al(p))^*= f_\al(p^*),

3860: $$

3861: the latter could be taken as the definition of $f_\al$.

3862: 

3863: 

3864: 

3865: 

3866: \medskip

3867: {\bf Property 2.} \cite[Lemma 2.1]{Littelmann2}. For each

3868: $p\in Dom(e_\al)\cap {\mathcal P}$,

3869: $$

3870: m_\al(e_\al(p))= m_\al(p)+1,

3871: $$

3872: $$

3873: p\in Dom(e^N_\al) \iff N < |m_\al|.

3874: $$

3875: 

3876: 

3877: 

3878: 

3879: 

3880: 

3881: \medskip

3882: {\bf Property 3.} Suppose that $p$ is a path in ${\mathcal P}_\Z$

3883: which does not belong to the domain of any $e_\al, \al\in \Phi$.

3884: Then $p$ is contained in $\Del$. Indeed, for each simple root $\al$

3885: we have to have $m_\al(p)>-1$. Since $p\in {\mathcal P}_\Z$,

3886: $m_\al(p)=0$. Thus $p\in {\mathcal P}^+$.

3887: 

3888: 

3889: \medskip

3890: {\bf Property 4.}  \cite[Proposition 3.1 (a, b)]{Littelmann2}.

3891: For each $\al \in \Phi$, $Dom(f_\al)\cap {\mathcal P}_\Z$ is open

3892: and $f_\al|{\mathcal P}_\Z$ is continuous.

3893: 

3894: 

3895: \medskip

3896: {\bf Property 5.} \cite[\S 7, Corollary 1 (a)]{Littelmann2}.

3897: Let $p\in {\mathcal P}^+$ and $\phi$ be a composition of lowering

3898: operators defined on $p$. Then $\phi(p)\in {\mathcal P}_\Z$ .

3899: 

3900: 

3901: \medskip

3902: {\bf Property 6.} Combining Properties 4 and 5 we conclude that for

3903: each ${\mathbf f}\in {\mathcal F}$, $Dom({\mathbf f})\cap {\mathcal

3904: P}^+$ is open and ${\mathbf f}|{\mathcal P}^+$ is continuous.

3905: 

3906: 

3907: \medskip

3908: {\bf Property 7.} \cite[Corollary 3, Page 512]{Littelmann2}.

3909: $p$ is an LS path of the $\De$-length $\la$ if and only if there

3910: exists ${\mathbf f}\in {\mathcal F}$ such that

3911: $$

3912: p={\mathbf f}(\pi_\la)

3913: $$

3914: 

3915: 

3916: \medskip

3917: {\bf Property 8.} \cite[Corollary 2(a), page

3918: 512]{Littelmann2}. The set of LS paths of the given $\De$-length is

3919: stable under ${\mathcal A}$.

3920: 

3921: 

3922: \medskip

3923: {\bf Property 9.} Suppose that $p\in {\mathcal P}$, $t\in [0,1]$,

3924: $\al\in \Phi$ and $x:=p(t)$ satisfy

3925: $$

3926: \al(x)\in \Z,  \quad p'_-(t)\gtrsim_{W_x} p'_+(t).

3927: $$

3928: Then the path $q=e_\al(p)$ also satisfies

3929: $$

3930: q'_-(t)\gtrsim_{W_y} q'_+(t),

3931: $$

3932: for $y=q(t)$. \proof If $h_\al(t)\ne m_\al, m_\al-1$, the germs of

3933: the paths $p$ and $q$ at $t$ differ by a translation. Thus the

3934: conclusion trivially holds in this case. The same argument applies

3935: if $h_\al(t)=m_\al-1$ and

3936: $$

3937: \al(p_-'(t)) \ge 0, \al(p_+'(t))\le 0.

3938: $$

3939: The nontrivial cases are:

3940: 

3941: 

3942: 1. $h_\al(t)=m_\al-1$, $\al(p'_-(t))\le 0$, $\al(p'_+(t))\le 0$. In

3943: this case the assertion follows from Lemma \ref{easy} with

3944: $\nu=p'_-(t), \mu=p'_+(t)$.

3945: 

3946: 

3947: 2. $h_\al(t)=m_\al$, $\al(p'_-(t))\le 0$, $\al(p'_+(t))\ge 0$. In

3948: this case the assertion follows from Lemma \ref{hard} with

3949: $\nu=p'_-(t), \mu=p'_+(t)$. \qed

3950: 

3951: 

3952: 

3953: 

3954: \medskip

3955: {\bf Property 10.}  Suppose that $p=p_1*p_2$ where $p_1\in {\mathcal

3956: P}^+$. Then for each $e\in {\mathcal E}$ defined on $p$ we have

3957: $$

3958: e(p)= p_1* e(p_2).

3959: $$

3960: \proof It suffices to prove this for $e=e_\al$, $\al\in \Phi$. In

3961: the latter case it follows directly from the definition of the

3962: operator $e_\al$. \qed

3963: 

3964: 

3965: 

3966: 

3967: The next property is again clear from the definition:

3968: 

3969: 

3970: \medskip

3971: {\bf Property 11.} Suppose that

3972: $$

3973: {\mathbf e}= e_{\be_m}\circ ... \circ e_{\be_1}

3974: $$

3975: where $\be_i\in \Phi$, $p\in Dom({\mathbf e})$. Set

3976: $$

3977: p_i:= e_{\be_i}\circ ...\circ e_{\be_1}(p), i=1,...,m.

3978: $$

3979: Then for each $T\in [0, 1]$, the sequence of vectors

3980: $$

3981: (p'_+(T), (p_1)_+'(T),..., (p_m)_+'(T)),

3982: $$

3983: after deleting equal members, forms a chain.

3984: 

3985: 

3986: \begin{lem}

3987: \label{finiteness} Given a path $p$ there are only finitely many

3988: operators ${\mathbf e}\in {\mathcal E}$ which are defined on $p$.

3989: \end{lem}

3990: \proof Break the path $p$ as the concatenation

3991: $$

3992: p_1*...*p_s.

3993: $$

3994: of geodesic paths each of which is contained in a single alcove and

3995: let $T_i\in [0, 1]$ be such that $p(T_i)=p_i(1/2)$; set $T_0:=0$.

3996: Then for each $e_\al\in {\mathcal E}$ defined on $p$ there exists

3997: $i$ such that the derivatives of ${e}_\al(p)$ and $p$ at $T_i$ are

3998: not the same. Moreover,

3999: $$

4000: q={e}_\al(p)= q_1*...*q_s,$$ where each $q_i$ is a geodesic path

4001: contained in an alcove. Consider the vector

4002: $$

4003: L(p):= (\ell( p'(0)), \ell(p'(T_1)),..., \ell(p'(T_s))) \in (\N\cup

4004: \{0\}) ^{s+1},

4005: $$

4006: where $\N^{s+1}$ is given the lexicographic order and $\ell$ is the length

4007: function on $W_{sph}$-orbits induced from the word metric on

4008: $W_{sph}$ as in Proposition \ref{wordlength}. Then, by combining Proposition

4009: \ref{wordlength} and Property 11 above, for each $\al\in \Phi$,

4010: $$

4011: L(e_\al(p))< L(p).

4012: $$

4013: Lemma follows. \qed

4014: 

4015: 

4016: 

4017: 

4018: 

4019: 

4020: 

4021: 

4022: 

4023: 

4024: \subsection{Generalized LS paths}

4025: \label{GLS}

4026: 

4027: 

4028: 

4029: 

4030:  In this paper we will need two generalizations of the

4031: concept of an LS path; the first one will be needed for the proof of

4032: the saturation theorem (section \ref{saturationsection}), the second

4033: will be used in section \ref{unfoldingLS} for the proof of the

4034: unfolding theorem. Although we will use the name {\em generalized LS

4035: path} for both generalizations, it will be clear from the context

4036: which generalization is being referred to.

4037: 

4038: 

4039: \medskip

4040: {\bf The first generalization, ${\mathcal L}{\mathcal S}_1$}.

4041: \footnote{This generalization of LS paths will be used in the proof of the saturation theorem.}

4042: 

4043: \medskip

4044: Suppose we are given a collection of LS paths $p_i$ of the

4045: $\De$-length $\la_i\in \De\cap P(R^\vee)$, $i=0,...,m$. We will use

4046: the notation

4047: $$

4048: \ul{\la}=(\la_0,...,\la_m)

4049: $$

4050: and

4051: $$

4052: \la:= \sum_{i=0}^m \la_i.

4053: $$

4054: 

4055: 

4056: \begin{rem}

4057: Actually, for our main application it will suffice to consider

4058: $\la_i$'s which are multiples of the fundamental coweights

4059: $\varpi_i$. Therefore such paths are automatically {\em

4060: generalized Hecke paths} as defined in Definition \ref{gH}.

4061: \end{rem}

4062: 

4063: 

4064: 

4065: 

4066: \begin{defn}

4067: \label{LS1} The concatenation

4068: $$

4069: p= p_0*p_1*...*p_m

4070: $$

4071: will be called a {\em generalized LS path} with $\length

4072: (p)=\ul{\la}$, if for each $i=0,...,m-1$

4073: $$

4074: p'_i(1) \gtrsim  p'_{i+1}(0).

4075: $$

4076: The set of such generalized LS paths will be denoted ${\mathcal

4077: L}{\mathcal S}_1$.

4078: \end{defn}

4079: 

4080: 

4081: This definition is a very special case of the one used by Littelmann

4082: in \cite{Littelmann3} under the name of a {\em locally integral

4083: concatenation}.

4084: 

4085: 

4086: Recall that according to the definition of $\De$-length,

4087: $$

4088: \la= \Length(p).

4089: $$

4090: Observe that each LS path $p$ satisfies the above definition, since

4091: for each $i$

4092: $$

4093: p'_i(1) \ge p'_{i+1}(0).

4094: $$

4095: 

4096: 

4097: \begin{ex}

4098: Suppose that $u, v$ are dominant coweights. Then $p=\pi_u*\pi_v$ is

4099: a generalized LS path.

4100: \end{ex}

4101: 

4102: 

4103: 

4104: 

4105: 

4106: \medskip

4107: {\bf The second generalization ${\mathcal L}{\mathcal S}_2$}.

4108: \footnote{This notion of generalized LS path will be used only to unfold Hecke paths.}

4109: 

4110: 

4111: \medskip

4112: Suppose that $p_1, p_2\in {\mathcal P}$ appear as

4113: $$

4114: p_1= \tilde{p}_1|_{[0,a]}, \quad 0<a<1,

4115: $$

4116: $$

4117: p_2= \tilde{p}_2|_{[b,1]}, \quad 0<b<1,

4118: $$

4119: where $\tilde{p}_1, \tilde{p}_2$ are LS paths, $a, b\in \Q$. (See

4120: section \ref{paths} for the definition of $\tilde{p}_1|_{[0,a]}$

4121: and $\tilde{p}_2|_{[b,1]}$.) Define the path $p:= p_1*p_2$. Assume

4122: that

4123: $$

4124: p_1'(1)\rhd p_2'(0);

4125: $$

4126: in other words, if $t$ is such that $p(t)= p_1(1)$ then

4127: $$

4128: p'_-(t) \rhd p_+(t).

4129: $$

4130: 

4131: \begin{defn}

4132: The concatenation $p$ will be called a {\em generalized LS path}

4133: if the {\em concatenation point} $p_1(1)$ is a regular

4134: point\footnote{I.e. it does not belong to any wall.} of

4135: $(A,W_{aff})$ and $p(1)\in P(R^\vee)$.

4136: 

4137: 

4138: The set of such generalized LS paths will be denoted ${\mathcal

4139: L}{\mathcal S}_2$.

4140: \end{defn}

4141: 

4142: 

4143: \begin{ex}

4144: \label{e2} Suppose that $u\in\De, v\in V$ are such that $u, v \in

4145: P(R^\vee)\otimes \Q$, $u+v\in P(R^\vee)\cap \De$ and the head of the

4146: vector $u$ is a regular point in $(A,W_{aff})$. Then $p=\pi_u*

4147: \pi_v\in {\mathcal L}{\mathcal S}_2$.

4148: \end{ex}

4149: 

4150: 

4151: This definition is again a very special case of the one given by

4152: Littelmann in \cite[5.3]{Littelmann2}. Littelmann does not assume

4153: that $p_1(1)$ is regular, but instead imposes certain chain

4154: conditions at this point.

4155: 

4156: 

4157: \medskip

4158: {\bf Properties of generalized LS paths:}

4159: 

4160: 

4161: \medskip

4162: {\bf Property 0.} If $p\in {\mathcal L}{\mathcal S}_1$ then $p^*$ is

4163: also in ${\mathcal L}{\mathcal S}_1$.

4164: 

4165: 

4166: \proof Represent $p$ as a concatenation $p_1*...*p_m$ of LS paths as

4167: in the Definition \ref{LS1}. Then

4168: $$

4169: p^*=(p_m^*)*...*(p_1^*),

4170: $$

4171: where each path $p_i^*$ is again an LS path. Now the assertion

4172: follows from Lemma \ref{iff}. \qed

4173: 

4174: 

4175: \medskip

4176: {\bf Property 1.} ${\mathcal L}{\mathcal S}_2$ is stable under the

4177: root operators  \cite[Lemma 5.6, 2-nd assertion]{Littelmann2}.

4178: In particular, suppose that $p$ is as in Example \ref{e2}. Then for

4179: each ${\mathbf f}\in {\mathcal F}$  (defined on $p$), ${\mathbf

4180: f}(p)$ is a generalized LS path of the $\De$-length $\la=u+v$.

4181: 

4182: 

4183: {\bf Property 2.}

4184:  ${\mathcal L}{\mathcal S}_1$ is stable

4185:  under the root operators.

4186: 

4187: 

4188: \proof Suppose that $p=p_1*...*p_m$ is a concatenation of LS paths

4189: as above and $e_\al$ is a raising operator. In particular, for each

4190: $i$ we have a vector $u_i$ so that

4191: $$

4192: p'_i(1) \ge u_i \sim p'_{i+1}(0).

4193: $$

4194: For each $i$, $e_\al(p_i)$ is again a Littelmann path. Therefore

4195: $e_\al(p)$ is a concatenation of LS paths $q_1*...*q_m$. We have to

4196: verify that for each $i$ there is a vector $v_i\in V$ so that

4197: $$

4198: q'_i(1) \ge v_i \sim q'_{i+1}(0).

4199: $$

4200: This however follows from the Property 9 in the previous section. To

4201: check that ${\mathcal L}{\mathcal S}_1$ is preserved by $f_\al$ we

4202: use that $q\in {\mathcal L}{\mathcal S}_1\iff q^* \in {\mathcal

4203: L}{\mathcal S}_1$ and

4204: $$

4205: f_\al(p)= (e_\al(p^*))^*.  \qed

4206: $$

4207: 

4208: 

4209: 

4210: 

4211: \medskip

4212: {\bf Property 3.} ${\mathcal L}{\mathcal S}_1$ and ${\mathcal

4213: L}{\mathcal S}_2$ are contained in ${\mathcal P}_{\Z,loc}$. For

4214: ${\mathcal L}{\mathcal S}_1$ it is immediate since, by Theorem

4215: \ref{locint}, the set of LS

4216: paths is contained in ${\mathcal P}_{\Z,loc}$.

4217: For ${\mathcal L}{\mathcal S}_2$ it is a special case

4218: of \cite[Lemma 5.5]{Littelmann2}.

4219: 

4220: 

4221: \medskip

4222: {\bf Property 4.} Suppose that $p\in {\mathcal L}{\mathcal S}_1$.

4223: Then there exists an element ${\mathbf e}\in {\mathcal E}$

4224: defined on $p$ such that $q={\mathbf e}(p)\in {\mathcal P}^+$.

4225: 

4226: 

4227: \proof If $m_{\al_1}(p)\le -1$ then we apply a power

4228: $e_{\al_1}^{k_1}$ to $p$ so that $q_1:=e_{\al_1}^{k_1}(p)$ satisfies

4229: $m_{\al_1}(q_1)>-1$. However, since $e_{\al_1}^{k_1}(p)\in {\mathcal

4230: L}{\mathcal S}_1\subset {\mathcal P}_{\Z,loc}$, it follows that

4231: $q_1\in {\mathcal P}_{\Z,loc}$ and so $m_{\al_1}(q_1)=0$. We then

4232: apply a power of $e_{\al_2}$ to $q_1$, etc. According to Lemma

4233: \ref{finiteness}, this process must terminate. Therefore, in the end

4234: we obtain a path

4235: $$

4236: {\mathbf e}(p)=q

4237: $$

4238: which does not belong to the domain of any raising operator. Since

4239: $q\in {\mathcal P}_{\Z,loc}$ it follows that $q$ is entirely

4240: contained in $\De$. \qed

4241: 

4242: 

4243: Recall  \cite[Chapter VI, section 10]{Bourbaki} that if a root system

4244: $R$ spans $V$ then each dominant coweight $\la\in \De$ is a

4245: positive integral combination

4246: $$

4247: \la= \sum_{i=1}^l n_i \varpi_i,

4248: $$

4249: where $\varpi_i$ are fundamental coweights. This assertion (as it stands) is false without the above assumption

4250: on $V$. In the general case we  have

4251: $$

4252: \la= \la'+ \sum_{i=1}^l n_i \varpi_i,

4253: $$

4254: where $\la'\in V'$, $n_i\in \N\cup \{0\}$.

4255: As an alternative the reader can restrict the discussion to semisimple groups only,

4256: when $V'=0$.

4257: 

4258: 

4259: \begin{conv}

4260: \label{con} From now on we will be assuming that in Definition

4261: \ref{LS1}

4262: $$

4263: \la_j=k_j \varpi_j, \quad k_j\in \N,

4264: $$

4265: for each $j=1,...,m$, where $\varpi_j$ is the $j$-th fundamental coweight, and

4266: $$

4267: \la_0\in V'.

4268: $$

4269: Then the subpath $p_0$ is necessarily geodesic.

4270: \end{conv}

4271: 

4272: 

4273: \begin{lem}

4274: \label{positiv} Suppose that $p\in {\mathcal L}{\mathcal S}_1 \cap

4275: {\mathcal P}^+$ is a generalized LS path with $\length(p)=\ul{\la}$.

4276: Then

4277: $$

4278: p=\pi_{\la_0}*...*\pi_{\la_m}.

4279: $$

4280: \end{lem}

4281: \proof Represent $p$ as the concatenation of maximal LS subpaths,

4282: $p=p_0*p_1...*p_m$. The geodesic subpath $p_0$ clearly equals $\pi_{\la_0}$.

4283: Since $p_1$ is an LS path and $p_1'(0)\in \De$, we

4284: see that $p_1$ is a geodesic path (see Corollary \ref{end}) which

4285: therefore equals $\pi_{\la_1}$. Moreover, because

4286: $$

4287: p_1'(1)\ge u_1 \sim p'_2(0),

4288: $$

4289: it follows that $u_1=p_1'(1)$ and thus $p_1'(1)=\la_1\sim p_2'(0)$.

4290: Let $x_1:= p_1(1)$; this point lies on the boundary face of $\De$

4291: which does not contain $\la_2$. Note that the vector $p_2'(0)$,

4292: regarded as an element of $T_{x_1}(A)$, points inside the Weyl

4293: chamber $\De$ (for otherwise $p$ is not contained in $\De$). On the

4294: other hand, since $\la_1\sim p_2'(0)$, the vector $p_2'(0)$ belongs

4295: either to $\De$ or to the Weyl chamber

4296: $$

4297: \tau_{\be_2}(\De)

4298: $$

4299: adjacent to $\De$. Since $p\in {\mathcal P}^+$, it is clear that

4300: $p_2'(0)\in \De$. Thus $p_2$ is the geodesic path $\pi_{\la_2}$.

4301: Continuing in this fashion we conclude that

4302: $$

4303: p=\pi_{\la_0}*\pi_{\la_1}*...*\pi_{\la_m}. \qed

4304: $$

4305: 

4306: 

4307: \begin{thm}

4308: \label{pathlow}

4309:  Suppose that $p$ is a generalized LS path in the sense of ${\mathcal L}{\mathcal S}_1$

4310:  with $\length(p)=\ul{\la}$

4311: (satisfying convention \ref{con}). Then there exists ${\mathbf f}\in

4312: {\mathcal F}$ such that

4313: $$

4314: p={\mathbf f}(\pi_{\la_0}*...*\pi_{\la_m})

4315: $$

4316: \end{thm}

4317: \proof If $p\in {\mathcal P}^+$ then we are done. Otherwise, by

4318: combining Lemma \ref{positiv} with the Property 4, we find an

4319: ${\mathbf e}\in {\mathcal E}$,

4320: $$

4321: {\mathbf e}= e_{\al_1}^{k_1}\circ ... e_{\al_n}^{k_n}

4322: $$

4323:  such that $p\in Dom({\mathbf e})$ and ${\mathbf e}(p)=q\in

4324: {\mathcal P}^+$. Therefore

4325: $$

4326: q=\pi_{\la_0}*\pi_{\la_1}*...*\pi_{\la_m}

4327: $$

4328: and thus the composition

4329: $$

4330: {\mathbf f}= f_{\al_n}^{k_n}\circ ... f_{\al_1}^{k_1}

4331: $$

4332: satisfies $p={\mathbf f} {\mathbf e}(p)={\mathbf

4333: f}(\pi_{\la_0}*\pi_{\la_1}*...\pi_{\la_m})$. \qed

4334: 

4335: 

4336: 

4337: 

4338: \subsection{Path model for the representation theory of Lie groups}

4339: \label{LSrep}

4340: 

4341: 

4342: Suppose that

4343: $$p(t)\in {\mathcal L}{\mathcal S}_1 \cup {\mathcal L}{\mathcal S}_2$$

4344: is a generalized LS path with $\Length(p)=\be$ and

4345: $$\length(p)=\ul{\la}.$$

4346: Suppose that $\al\in P(R^\vee)$ is such that $\al+p(t)$ is contained

4347: in $\De$. Then $\al$ and $p$ define a polygon

4348: $$

4349: P:=\ol{o y_0} \cup (p+\al)\cup \ol{y_n o} \subset \De$$

4350:  where $\al=\ov{o y_0}$, $y_n=\al+p(1)$. Let $\ga$ denote the vector

4351: $\ov{o y_{n}}$; then $\ga$ is also a dominant coweight. Recall that

4352: the contragredient dominant coweight $\ga^*\in \De$ is obtained by

4353: projecting the vector $-\ga$ to the Weyl chamber $\De$ by the

4354: projection $\P: V\to \De$.

4355: 

4356: 

4357: \begin{defn}

4358: 1. A polygon $P$ above will be called a {\em (broken) Littelmann

4359: polygon} with the $\De$-side lengths $\al, \be, \ga^*$.

4360: 

4361: 

4362: 2. If $p(t)$ is an LS path then $P$ will be called a {\em (broken)

4363: Littelmann triangle} with the $\De$-side lengths $\al, \be, \ga^*$.

4364: \end{defn}

4365: 

4366: 

4367: Pick a lattice $L$ such that

4368: $$

4369: Q(R^\vee)\subset L\subset P(R^\vee).

4370: $$

4371: Then there exists a unique connected semisimple complex Lie group

4372: $G^\vee$  with the root system $R^\vee$ and the character lattice

4373: $L$ of the maximal torus $T^\vee\subset G^\vee$.

4374:  Recall that irreducible representations $V$ of $G^\vee$ are

4375: parameterized by their dominant weights, $V=V_\la$, $\la\in \De\cap

4376: L$.

4377: 

4378: 

4379: Pick a path $q\in {\mathcal P}^+$ such that $q(1)=\be$. Then,

4380: according to \cite[Decomposition formula, Page 500]{Littelmann2} we

4381: have

4382: 

4383: 

4384: \begin{thm}

4385: \label{Lit2} The tensor product $V_\al\otimes

4386: V_\be$  contains $V_\ga$ as a subrepresentation if and only if there

4387: exists a path $p\in {\mathcal F}(q)$ such that $\pi_{\al}*p\in

4388: {\mathcal P}^+$ and $\pi_{\al}*p(1)=\ga$.

4389: \end{thm}

4390: 

4391: 

4392: \begin{rem}

4393: Littelmann works with simply-connected group $G^\vee$ and weights

4394: $\al, \be,\ga$ in $P(R^\vee)$. The statement for

4395: non-simply-connected groups trivially follows from the

4396: simply-connected case.

4397: \end{rem}

4398: 

4399: 

4400: 

4401: 

4402: In particular, since $p$ is an LS paths of the $\De$-length $\be$ if

4403: and only if $p\in {\mathcal F}(\pi_\be)$, it follows that

4404: 

4405: 

4406: \begin{thm}

4407: \cite{Littelmann1, Littelmann2}. \label{Lit} The tensor

4408: product $V_\al\otimes V_\be$  contains $V_\ga$ as a

4409: subrepresentation if and only if there exists a (broken) Littelmann

4410: triangle in $\Del\subset V$, with the $\De$-side-lengths $\al, \be,

4411: \ga^*$.

4412: 

4413: 

4414: In other words, $V_\ga\subset V_\al\otimes V_\be$ if and only if

4415: there exists an LS path $p$ of $\De$-length $\be$ such that

4416: 

4417: 

4418: 1. $\pi_\al *p\in {\mathcal P}^+$.

4419: 

4420: 

4421: 2. $p(1)+\al=\ga$.

4422: \end{thm}

4423: 

4424: 

4425: 

4426: 

4427: 

4428: 

4429: We will apply Theorem \ref{Lit2} as follows.

4430: Represent the vector $\be$ as the integer linear combination of

4431: fundamental coweights

4432: $$

4433: \be =\sum_{i=1}^n k_i \varpi_i.

4434: $$

4435: We reorder the fundamental coweights so that $k_i>0$ for all

4436: $i=1,...,m$ and $k_i=0, i\ge m+1$. Set $\la_i:= k_i \varpi_i, 1\le

4437: i\le m$ and let $\ul{\la}= (\la_1,...,\la_m)$. Therefore the path

4438: $$

4439: \pi_{\ul\la}:= \pi_{\la_1}*... \pi_{\la_m}

4440: $$

4441: belongs to ${\mathcal L}{\mathcal S}_1$ and $\pi_{\ul\la}(1)=\be$.

4442: Then

4443: 

4444: 

4445: 

4446: 

4447: \begin{cor}

4448: The tensor product $V_\al\otimes V_\be$ contains $V_\ga$ as a

4449: subrepresentation

4450:  if and only if there exists a generalized LS path $p$ so that

4451: 

4452: 

4453: 1. $\length(p)=\ul{\la}$.

4454: 

4455: 

4456: 2. $\pi_{\al}*p\in {\mathcal P}^+$.

4457: 

4458: 

4459: 3. $\pi_{\al}*p(1)=\ga$.

4460: \end{cor}

4461: \proof Set

4462: $$

4463: q=\pi_{\ul{\la}}.$$

4464:  According to Theorems \ref{pathlow} and Property 2 of

4465: generalized LS paths (section \ref{GLS}), $p\in {\mathcal P}$ is a

4466: generalized LS path with $\length(p)=\ul\la$ if and only if $p\in

4467: {\mathcal F}(q)$. Now the assertion follows from Theorem \ref{Lit2}.

4468: \qed

4469: 

4470: 

4471: Combining Corollary \ref{folded-chain}, Theorem \ref{T3} and Theorem

4472: \ref{Lit} we obtain

4473: 

4474: 

4475: \begin{cor}

4476: Suppose that $X$ is a thick Euclidean building modeled on the

4477: Coxeter complex $(A,W_{aff})$. Let $\al, \be, \ga^*\in L$ be

4478: dominant coweights. Suppose that $a\subset A$ is an alcove

4479: containing a special vertex $o$, $T=[o, x, y]\subset X$ is a

4480: geodesic triangle with the special vertices  and the $\De$-side

4481: lengths $\al, \be, \ga^*$. Assume also that the broken side

4482: $Fold_{a,A}(\ol{x y})$ of the folded triangle

4483: $$

4484: Fold_{a,A}(T)

4485: $$

4486: has breaks only at the special vertices of $A$. Then

4487: 

4488: 

4489: 1. The folded triangle $T'=Fold_{o,id,\De}(T)\subset \De$ is a

4490: Littelmann triangle.

4491: 

4492: 

4493: 2. $V_\ga\subset V_\al\otimes V_\be$.

4494: \end{cor}

4495: \proof Indeed, according to Corollary \ref{folded-chain}, the folded

4496: triangle $T'$ satisfies the chain condition. Each break point $x_i$

4497: on the broken side of $T'$ is

4498: 

4499: 

4500: 1. Either a special vertex, in which case it satisfies maximal chain

4501: condition by Remark \ref{R1}, or

4502: 

4503: 

4504: 2. $Fold_{a,A}(\ol{xy})$ is geodesic at the point corresponding to

4505: $x_i$, so the chain at $x_i$ can be chosen to be maximal by

4506: Proposition \ref{chaincondition1}.

4507: 

4508: 

4509: Hence Theorem \ref{T3} implies that $T'$ is a Littelmann triangle.

4510: The second assertion now follows from Theorem \ref{Lit}. \qed

4511: 

4512: 

4513: \medskip

4514: Of course, the assumption that the break points occur only at the

4515: special vertices is very restrictive. In section

4516: \ref{saturationsection} we will get rid of this assumption at the

4517: expense of dilation of the side-lengths.

4518: 

4519: 

4520: \section{Unfolding}

4521: \label{last}

4522: 

4523: 

4524: 

4525: 

4526: The goal of this section is to establish an intrinsic

4527: characterization of folded triangles as the broken billiard

4528: triangles satisfying the chain condition. We first prove this

4529: characterization for Littelmann triangles and then, using this, give

4530: a general proof.

4531: 

4532: 

4533: 

4534: 

4535: Throughout this section we assume that $X$ is a thick locally

4536: compact Euclidean building modeled on the Coxeter complex $(A,W)$,

4537: $\Del\subset A$ is a Weyl chamber with tip $o$. Let $g: X\to

4538: \De$ denote the folding $Fold_{\De}$.

4539: 

4540: 

4541: 

4542: Let $T\subset \Del$ be a  billiard triangle which is the

4543: composition

4544: $$

4545: T=\ov{ox}\cup r \cup \ol{y o},

4546: $$

4547: where $r(t)=p(t)+\al$, $\al=\ov{ox}$ and $p\in {\mathcal P}$ is a Hecke path. Thus $T$

4548: has the geodesic sides $\ol{ox}, \ol{oy}$ and the broken side

4549: $r$. We set $\ga:= \ov{o y}$ and let $\be\in \De$ denote

4550: $\De$-length of the path $p$.

4551: 

4552: 

4553: 

4554: 

4555: \subsection{Unfolding Littelmann triangles}

4556: \label{unfoldingLS}

4557: 

4558: 

4559: \begin{thm}

4560: \label{LSfolded}

4561:  Suppose that, in addition, $T$ is a {\em Littelmann triangle},

4562: i.e. $\al, \ga\in L\subset P(R^\vee)$ and $p$ is an LS path. Then

4563: $T$ can be {\em unfolded} in $X$, i.e. there exists a geodesic

4564: triangle $\t{T}\subset X$ such that $g(\t{T})=T$.

4565: \end{thm}

4566: \proof Here is the idea of the proof: We know that billiard

4567: triangles in $\De$ satisfying the {\em simple chain condition} can be

4568: unfolded to geodesic triangles in $X$, see Corollary

4569: \ref{chain-folded}. Littelmann triangle $T$ is billiard, satisfies

4570: the chain condition, but not necessarily the {\em simple chain condition}.

4571: Our goal is to approximate $T$ by Littelmann polygons $P_\eps$,

4572: $\lim_{\eps\to 0} P_\eps=T$, which satisfy the simple chain condition.

4573: We then unfold each $P_\eps$ to a geodesic quadrilateral

4574: $\t{T}_\eps\subset X$. Since $X$ is locally compact, there is a

4575: convergent sequence $\t{T}_{\eps_j}$ whose limit is a geodesic triangle

4576: $\t{T}$ which folds to $T$. Below is the detailed argument.

4577: 

4578: 

4579: 

4580: 

4581: 

4582: 

4583: \begin{figure}[tbh]

4584: \centerline{\epsfxsize=3.5in \epsfbox{unfolding.eps}} \caption{\sl

4585: Approximation.} \label{unfolding.fig}

4586: \end{figure}

4587: 

4588: 

4589: 

4590: 

4591: \medskip

4592: Consider the geodesic path $\pi_\be=\ol{ob}\in {\mathcal P}^+$.

4593: Since $p$ is an LS path with the $\De$-length $\be$, according to

4594: Property 7 in section \ref{operators}, there exists a composition

4595: $\phi\in {\mathcal F}$ of lowering operators so that

4596: $$

4597: \phi(\pi_\be)= p.

4598: $$

4599: 

4600: 

4601: Let $c\subset \De$ denote an alcove which contains the germ of the

4602: segment $\ol{ob}$ at $b$. Pick a point $u$ in the interior of

4603: $\ol{ob}\cap c$. Then for each $\eps>0$ there exists a

4604: point $u_\eps\in int(c)\cap P(R^\vee)\otimes \Q$ such that

4605: 

4606: 

4607: 1. $|u-u_\eps|<\eps$.

4608: 

4609: 

4610: 2. The segments $\ol{ou_\eps}, \ol{u_\eps b}$ do not pass

4611:    through any point of intersection of two or more walls

4612:    (except for the end-points of these segments).

4613: 

4614: 

4615: Observe that $u_\eps$ is a regular point in $(A,W_{aff})$, i.e. it

4616: does not belong to any wall.

4617: 

4618: 

4619: In other words, the path

4620: $$

4621: \hat{p}_\eps:= \ol{ou_\eps}\cup \ol{u_\eps b} \in {\mathcal P}

4622: $$

4623: is {\em generic}. Parameterize $\hat{p}_\eps$ with the constant speed

4624: so that $\hat{p}_\eps(t_\eps)=u_\eps$. See Figure \ref{unfolding.fig}.

4625: 

4626: 

4627: Then the path $\hat{p}_{\eps}$ belongs to ${\mathcal P}^+$; clearly it

4628: is also a generalized LS path: $\hat{p}_\eps\in {\mathcal L}{\mathcal

4629: S}_2$. Moreover,

4630: $$

4631: \lim_{\eps\to 0} \hat{p}_{\eps}= \pi_\be.

4632: $$

4633: Therefore, according to Property 6 of the root operators (see

4634: section \ref{operators}), the operator $\phi$ is defined on all

4635: $\hat{p}_{\eps}$ for $\eps$ sufficiently small and

4636: $$

4637: \lim_{\eps\to 0} \phi(\hat{p}_{\eps})= \phi(\pi_\be)=p.

4638: $$

4639: Set $p_\eps:= \phi(\hat{p}_\eps)$. Since $\hat{p}_\eps$ was generic, the

4640: path $p_\eps$ is generic as well. By construction, for each

4641: sufficiently small $\eps$,

4642: $$

4643: p_\eps(1)=p(1).

4644: $$

4645: Observe also that the germ of the path $p_{\eps}$ at the point

4646: $p_\eps(t_\eps)$ is isomorphic (via an element of

4647: $W_{aff})$ to the germ of $\hat{p}_{\eps}$ at $u_\eps$ (since $u_\eps$

4648: is regular). Similarly, $p_\eps$ is the composition of the path

4649: $p_\eps|[0, t_\eps]$ with the path that belongs to the

4650: $W_{aff}$-orbit of $\ol{u_\eps b}$.

4651: 

4652: 

4653: For each $\eps$ we form a new polygon $P_\eps$ by replacing

4654: the broken side $r(t)=\al+p(t)$ (in $T'$) with the path $\al+p_\eps(t)$.

4655: Clearly,

4656: $$

4657: \lim_{\eps\to 0} P_\eps=T.

4658: $$

4659: 

4660: 

4661: To simplify the notation we now fix $\eps>0$ and let $q:= p_\eps$.

4662: 

4663: 

4664: \begin{lem}

4665: For all sufficiently small $\eps$, the polygon $P_\eps$ is

4666: contained in $\De$.

4667: \end{lem}

4668: \proof Suppose that $\la$ is a simple root which is negative at some

4669: point of the path $\al+q(t)$.

4670: 

4671: 

4672: Since $\la$ is nonnegative on the limiting path $\al+p$, the minimum

4673: of the function $J_\la(t):= \la(q(t)), t\in [0,1]$, converges to zero as $\eps\to 0$.

4674: However, as a generalized LS path, $q$ belongs to ${\mathcal P}_\Z$

4675: (see Property 5 in section \ref{operators}). Since $\al\in

4676: P(R^\vee)$, it follows that the minimum of  $J_\la(t)$ is an

4677: integer. Hence it has to be equal to zero for all sufficiently small

4678: values of $\eps$. Contradiction. \qed

4679: 

4680: 

4681: 

4682: 

4683: Since each $q$ is a generalized LS path and $P_\eps\subset \De$,

4684: the polygon $P_\eps$ is a Littelmann polygon. Moreover, since $q$

4685: is {\em generic}, the polygon $P_\eps$ satisfies the {\em simple

4686: chain condition}. Thus

4687: 

4688: 

4689: 1. For each $t\in [0,t_{\eps})$ either $q$ is smooth at $t$ or

4690: $$

4691: (q'_-(t), q'_+(t))

4692: $$

4693: is a chain of length 1: At $m=q(t)$ the above tangent vectors are

4694: related by a single reflection in $W_{m}$. The fixed-point set of

4695: this reflection is the unique wall passing through $m$.

4696: 

4697: 

4698: 2. The subpath $q([t_\eps,1])$ in $q$ is a geodesic segment and

4699: $$

4700: \del(\eps):= \angle(\ov{u_\eps o}, \ov{u_\eps b})= \pi- \angle(

4701: q'_-(t(\eps)), q'_+(t(\eps))).

4702: $$

4703: 

4704: 

4705: 

4706: 

4707: 

4708: 

4709: Now we are in position to apply Corollary \ref{chain-folded} and

4710: unfold $P_\eps$ in $X$: For each $\eps$ there exists a geodesic

4711: quadrilateral $\t{T}_\eps$ (with one vertex at $o$) in $X$ such that

4712: $$

4713: g(\t{T}_\eps)= P_\eps.

4714: $$

4715: Let $\t{z}=\t{z}_\eps$ denote the point of $\t{T}_\eps$ which maps to $z=q(t_\eps)$

4716: under the folding map $f$. Since $z$ is a regular point, the point

4717: $\t{z}$ is regular as well and the angle between the sides of $\t{T}_\eps$

4718: at $\t{z}$ is the same as the angle between the sides of $P_\eps$ at

4719: $z$, i.e. equals $\del(\eps)$.

4720: 

4721: 

4722: Since the building $X$ is locally compact, the sequence of

4723: quadrilaterals $\t{T}_\eps$ subconverges to a geodesic quadrilateral

4724: $T\subset X$ which is a geodesic triangle since

4725: $$

4726: \lim_{\eps\to 0}\del(\eps)=\pi.

4727: $$

4728: By continuity of the folding $g: X\to\De$,

4729: $$

4730: g(T)= \lim_{\eps\to 0} P_{\eps} =T. \qed

4731: $$

4732: 

4733: 

4734: 

4735: 

4736: In the above proof we assumed that the polygon $T$ is entirely

4737: contained in $\De$. This assumption can be weakened. Let $f: X\to A$

4738: denote the folding $Fold_{a,A}$ into the apartment $A$, where $a$ is

4739: an alcove containing $o$. Suppose that $T\subset A$ is as above, so

4740: that $\al, \ga\in P(R^\vee)$, $p$ is a billiard path, $r=p+\al$. Define two subsets $J, J'\subset I=[0,1]$:

4741: $$

4742: J:= cl( r^{-1}(int( \De))), \quad J':= cl( r^{-1}(int(

4743: V\setminus \De))).

4744: $$

4745: Clearly, $I=J\cup J'$ and the set $J\cap J'$ is finite.

4746: 

4747: 

4748: We assume that for each $t\in J$ the germ of $p$ at $t$ satisfies

4749: the maximal chain condition, and for each $t\in J'$ the germ of $p$

4750: at $t$ is geodesic.

4751: 

4752: 

4753: 

4754: 

4755: \begin{thm}

4756: \label{addendum} Under the above assumptions the polygon $T$ can be

4757: unfolded to a geodesic triangle in $X$ via the retraction $f$.

4758: \end{thm}

4759: \proof Recall that unfolding of $T$ is a local problem of behavior of

4760: the path $r$ at the break-points (Lemma \ref{locality}), which

4761: in our case all occur inside $\De$.

4762: 

4763: 

4764: We first replace $T$ with the polygon $P=\P(T)$, where

4765: $\P=\P_{\De}$ is the projection of $A$ to the Weyl chamber $\De$.

4766: Then, analogously to the proof of Proposition \ref{chaincondition1},

4767: the new polygon $P$ still satisfies the maximal chain condition.

4768: Therefore, according to the previous theorem, the polygon $P$

4769: unfolds in $X$ via the folding map $g=Fold_{\De}: X\to \De$. However

4770: this means that the unfolding condition (stated in Lemma

4771: \ref{locality}) is satisfied at each break-point of the polygon $T$

4772: (since the germs of $r$ and of $\P(r)$ are the same). Hence

4773: the original polygon $T$ unfolds to a geodesic triangle in $X$ via $f: X\to A$. \qed

4774: 

4775: 

4776: 

4777: 

4778: 

4779: 

4780: 

4781: 

4782: 

4783: 

4784: \medskip

4785: Let $\ul{G}$ be a connected split semisimple algebraic group with

4786: the root system $R$ and the cocharacter

4787: lattice $L$ of the maximal torus $\ul{T}\subset \ul{G}$. We let

4788: $\ul{G}^\vee$ denote its Langlands' dual and set

4789: $$

4790: G^\vee:=\ul{G}^\vee(\C).

4791: $$

4792: We assume that $\al,\be,\ga^*\in L$ are dominant weights of $G^\vee$

4793: such that

4794: $$

4795: (V_\al\otimes V_\be \otimes V_{\ga^*})^{G^\vee}\ne 0,$$

4796: equivalently,

4797: $$

4798: V_{\ga}\subset V_\al\otimes V_\be.

4799: $$

4800: 

4801: 

4802: As a corollary of Theorem \ref{LSfolded} we get a new proof of

4803: 

4804: 

4805: \begin{thm}

4806: [Theorem 9.17 in \cite{KLM3}, also proven in \cite{Haines}] \label{klm3}

4807:   Under the above assumptions,

4808: in the thick Euclidean building $X$ there exists a geodesic triangle

4809: with special vertices and the $\De$-side lengths $\al, \be, \ga^*$. In other words,

4810: $$

4811: n_{\al,\be}(\ga)\ne 0 \Rightarrow m_{\al,\be}(\ga)\ne 0.

4812: $$

4813: \end{thm}

4814: \proof Since

4815: $$

4816: V_{\ga}\subset V_\al\otimes V_\be,

4817: $$

4818: according to Littelmann's Theorem \ref{Lit}, there exists a

4819: Littelmann triangle $T'\subset \De$, as in Theorem \ref{LSfolded}.

4820: Let $p\in {\mathcal P}$ denote the LS path (of the $\De$-length

4821: $\be$) representing the broken side of $T'$; $p=\phi(\pi_\be)$,

4822: where $\phi\in {\mathcal F}$ is a composition of lowering operators.

4823: Thus, by Theorem \ref{LSfolded}, there exists a triangle $T=[o, x,

4824: y]\subset X$ such that $Fold_\De(T)=T'$. Therefore, by the

4825: definition of folding,

4826: $$

4827: d_{ref}(o, x)= d_{ref}(o,x')=\al, \quad d_{ref}(o, y)=

4828: d_{ref}(o,y')=\ga.

4829: $$

4830: Since we assumed that $\al, \be\in L\subset P(R^\vee)$ then $x, y$

4831: are special vertices of $x$. Since folding preserves the

4832: $\De$-length,

4833: $$

4834: d_\De(x,y)= \Length(p)=\be. \qed

4835: $$

4836: 

4837: 

4838: 

4839: 

4840: 

4841: 

4842: \subsection{Characterization of folded triangles}

4843: \label{characterizationsection}

4844: 

4845: 

4846: The goal of this section is to extend the results of the previous

4847: one from the case of Littelmann triangles to general broken

4848: triangles satisfying the chain condition.

4849: 

4850: 

4851: 

4852: 

4853: \begin{thm}

4854: \label{chain-unfolding} Suppose that $p\in {\mathcal P}$ is a Hecke

4855: path, $\al=\ov{o u}\in \De$ is such that the path $q:= \al+p$ is

4856: contained in $\De$. Define the billiard triangle $T':=\ol{o u}\cup q

4857: \cup \ol{ q(1) o}$.

4858:  Then $T'$ can be unfolded in $X$.

4859: \end{thm}

4860: \proof The idea of the proof is that the set of unfoldable billiard

4861: paths is closed, thus it suffices to approximate $p$ by unfoldable

4862: paths. We first prove the theorem in the case when $o$ does not

4863: belong to the image of the path $q$.

4864: 

4865: 

4866: 

4867: 

4868: According to Lemma \ref{locality}, unfolding of a path is a purely

4869: local matter. Therefore the problem reduces to the case when $q$ has

4870: only one break-point, $x=q(t_1)$. If $p$ were an LS path, we would

4871: be done. In general it is not, for instance, because it might fail

4872: the {\em maximal chain condition}. We resolve this difficulty by

4873: {\em passing to a smaller Coxeter complex} and a smaller building.

4874: 

4875: 

4876: Let $R_x$ denote the root subsystem in $R$ which is generated by the

4877: roots corresponding to the walls passing through $x$. This root

4878: system determines a Euclidean Coxeter complex where the stabilizer

4879: of the origin is a finite Coxeter group $W'_{sph}$ which is

4880: conjugate to the group $W_x$ via the translation by the vector

4881: $\ov{ox}$. Let $\De_x$ denote the positive Weyl chamber of $(V,

4882: W'_{sph})$ (the unique chamber which contains $\De$). Let $\xi$,

4883: $\eta$ and $\zeta$ denote the normalizations of the vectors

4884: $-p'_-(t_1), p'_+(t_1), \ov{xo}$.

4885: 

4886: 

4887: Then, since $p$ satisfies the chain condition, there exists an

4888: $(S,W_x,\De_x)$-chain

4889: $$

4890: (\nu_0,...,\nu_m), \quad \nu_0=-\xi, \quad \nu_m=\eta, \quad

4891: \nu_i=\tau_i(\nu_{i-1}), \quad 1\le i\le m.

4892: $$

4893: Our first observation is that although this chain may fail to be a

4894: maximal chain with respect to the unrestricted Coxeter complex

4895: $(S,W_{sph})$, we can assume that it is maximal with respect to the

4896: restricted Coxeter complex $(S,W_x)$.

4897: 

4898: 

4899: Next, the initial and final points of $q$ may not belong to

4900: $P(R^\vee_x)$. Recall however that {\em rational points} are dense

4901: in $S$, see Lemma \ref{density}; therefore, there exists a sequence

4902: of rational points (with respect to $R_x$) $\xi_j\in S$ which

4903: converges to $\xi$. Thus, using the same reflections $\tau_i$ as

4904: before, we obtain a sequence of rational chains $(\nu_i^j),

4905: i=0,...,m$, where $\nu_0^j=-\xi_j, \nu_m^j=\eta_j$. We set

4906: $\zeta_j:= \zeta$.

4907: 

4908: 

4909: Hence for each $j$ there exists a number $c=c_j\in \R_+$ so that the

4910: points

4911: $$

4912: x_j=x+ c\xi_j, \quad y_j:= x+ c\eta_j

4913: $$

4914: belong to $P(R^\vee)$.  We define a sequence of paths

4915: $$

4916: q_j:=\ol{x_j x}\cup \ol{x y_j} \in \t{\mathcal P}.

4917: $$

4918: Our next goal is to choose the sequence $\xi_j$ so that the germ of

4919: each $q_j$ at $x$ is contained in $\De$. If $x$ belongs to the

4920: interior of $\De$ then we do not need any restrictions on the

4921: sequence $\xi_j$. Assume therefore that $x$ belongs to the boundary

4922: of $\De$. Let $F$ denote the smallest face of the Coxeter complex

4923: $(V, W_{sph})$ which contains the point $x$ and let $H$ denote the

4924: intersection of all walls through the origin which  contain $x$. It

4925: is clear that $F$ is a convex homogeneous polyhedral cone contained

4926: in $H$ and $x$ belongs to the interior of $F$ in $H$. If $w\in W_x$

4927: is such that $w(-\xi)= \eta$ then $w$ fixes $H$ (and $F$) pointwise.

4928: 

4929: 

4930: By Lemma \ref{density}, applied to the root system $R_x$, there

4931: exists a sequence of unit rational vectors $\xi_j$ and positive

4932: numbers $\eps_j$ converging to zero so that points $x\pm \eps_j

4933: \xi_j$ belong to $int_H(F)$; therefore the sequence $w(x+ \eps_j

4934: \xi_j)$ is also contained in $int_H(F)$.

4935: 

4936: 

4937: Using this sequence $\xi_j$ we define the paths $q_j$; clearly the

4938: germ of $q_j$ at $x$ is contained in $int_H(F)\subset \De\subset

4939: \De_x$.

4940: 

4941: 

4942: \begin{rem}

4943: Note that, typically, the sequence $(c_j)$ is unbounded and the

4944: paths $q_j$ are not contained in $\De_x$.

4945: \end{rem}

4946: 

4947: 

4948: We let $p_j\in {\mathcal P}$ denote the path $q_j- q_j(0)$. Then

4949: each $p_j$ is an LS path with respect to the root system $R_x$:

4950: Integrality and the maximal chain condition now hold. Set

4951: $$

4952: \la_j:= \hbox{length}_{\De_x}(p_j).

4953: $$

4954: 

4955: 

4956: \begin{rem}

4957: Observe that,

4958: $$

4959: \lim_j \bar\la_j = \bar\la\in \De,

4960: $$

4961: where $\la$ is the $\De_x$-length of $p$.

4962: \end{rem}

4963: 

4964: 

4965: Therefore, according to Theorem \ref{addendum} for each $j$ the path

4966: $q_j$ is unfoldable in a thick Euclidean building $X_x$ modeled on

4967: the Coxeter complex

4968: $$

4969: (A, W'_{aff}), \hbox{~~where~~} W'_{aff}= V\ltimes W_x.

4970: $$

4971: This means that there exists a geodesic path $\tilde{q}_j$ in $X_x$,

4972: whose $\De_x$-length is $\la_j$, and which projects to $q_j$ under

4973: the folding $X_x\to A$.

4974: 

4975: 

4976: Let $z_j\in \t{q}_j$ be the points which correspond to the point $x$

4977: under the folding map

4978: $$

4979: \t{q}_j \to q_j.

4980: $$

4981: Thus the ``broken triangle'' $[\xi_j, \zeta, \eta_j]$ in $S_x$

4982: unfolds in $\Si_{z_j}(X_x)$ into a triangle $[\t\xi_j,

4983: \t\zeta_j,\t\eta_j]$ such that

4984: $$

4985: d_{ref}(\t\xi_j, \t\zeta_j) = d_{ref}(\xi_j, \zeta_j),

4986: $$

4987: $$

4988: d_{ref}(\t\eta_j, \t\zeta_j) = d_{ref}(\eta_j, \zeta_j),

4989: $$

4990: $$

4991: d(\t\eta_j,\t\zeta_j)=\pi.

4992: $$

4993: 

4994: 

4995: Observe that the metric distance from $o$ to $z_j$ is uniformly

4996: bounded. Since $X_x$ is locally compact, the sequence of buildings

4997: $\Si_{z_j}(X_x)$ subconverges to the link of a vertex $u\in

4998: X_x\subset X$.

4999: 

5000: 

5001: \begin{rem}

5002: The spherical buildings $\Si_{z_j}(X_x)$, $\Si_{u}(X_x)$ have to be

5003: modeled on the same spherical Coxeter complex $(S,W_x)$, since the

5004: structure group can only increase in the limit and the structure

5005: group at $z_j$ was already maximal possible, i.e. $W_x$.

5006: \end{rem}

5007: 

5008: 

5009: Accordingly, the triangles $[\t\xi_j, \t\zeta_j,\t\eta_j]$

5010: subconverge to a triangle $[\t\xi, \t\zeta,\t\eta]$ whose refined

5011: side-lengths are

5012: $$

5013: d_{ref}(\xi, \zeta),   d_{ref}(\zeta, \eta),\pi.

5014: $$

5015: This shows that the triangle $[\xi,\zeta,\eta]$ can be unfolded in a

5016: building which is modeled on $(S,W_x)$. We now apply the {\em

5017: locality} lemma \ref{locality} to conclude that the path $q$ can be

5018: unfolded in $X$ to a geodesic path. Thus the broken triangle $T'$

5019: unfolds to a geodesic triangle as well.

5020: 

5021: 

5022: \medskip

5023: If $o$ belongs to the image of $q$ we argue as follows. The path

5024: $q$, as before, has only one break point, which in this case occurs

5025: at the origin:

5026: $$

5027: q= \ol{zo}\cup \ol{oy}.

5028: $$

5029: There exists an element $w\in W_{sph}$ which sends the vector

5030: $\eta=-\xi$ to $\xi$, where $\eta$ is the normalization of the

5031: vector $\ov{oy}$. Then consider the geodesic path

5032: $$

5033: \tilde{q}:= \ol{zo}\cup w(\ol{oy}).

5034: $$

5035: It is clear that $g(\tilde{q}) = \P(\tilde{q}) =q$. \qed

5036: 

5037: 

5038: By combining Theorem \ref{chain-unfolding} and Corollary

5039: \ref{folded-chain} we obtain the following

5040: 

5041: 

5042: \begin{thm}

5043: [Characterization of folded triangles]\label{characterization} A

5044: polygon $P\subset \De$ of the form

5045: $$

5046: \ol{ox} \cup (p+\al) \cup \ol{yo}, \hbox{~where~}\al=\ov{ox},

5047: \ga=\ov{oy}\in \De,

5048: $$

5049: can be unfolded to a geodesic triangle in $X$ if and only if $p$ is

5050: a Hecke path, i.e. a billiard path which satisfies the chain

5051: condition.

5052: \end{thm}

5053: 

5054: 

5055: 

5056: 

5057: 

5058: \section{Proof of the saturation theorem}

5059: \label{saturationsection}

5060: 

5061: 

5062: We first prove Theorem \ref{3->4} formulated in the Introduction. Part 1 of Theorem was proven

5063: in \cite{KLM3}, so we prove Part 2.

5064: Let $X$ be a (thick) Euclidean building of rank $r$ modeled on a

5065: discrete Coxeter complex $(A,W)$; the building $X$ is the Bruhat-Tits building associated with

5066: the group $G=\ul{G}(\K)$. Then the assumption  that $m_{\al,\be}(\ga)\ne 0$ is equivalent to

5067: the assumption that there exists a geodesic triangle $T=[\t{x}, \t{y}, \t{z}]\subset X$,

5068: where $\t{x}, \t{y}, \t{z}$ are special vertices of

5069: $X$ and whose $\De$-side-lengths are $\al, \be, \ga^* \in \De\cap P(R^\vee)$.

5070: 

5071: 

5072: Recall that there exists an apartment $\tilde{A}\subset X$ which

5073: contains the segment $\ol{\t{x} \t{y}}$. We let $\t{W}$ denote the

5074: affine Weyl group operating on $\t{A}$. Our first step is to replace

5075: the geodesic triangle $T$ with a geodesic polygon

5076: $$

5077: \tilde{P} := [\tilde z, \t{x}=\tilde{x}_1,..., \tilde{x}_n,

5078: \tilde{x}_{n+1}=\tilde{y}]

5079: $$

5080: as follows. We now treat the point $\t{x}$ as the origin in the

5081: affine space $\t{A}$. Let $\tilde\De\subset \tilde{A}$ be a Weyl chamber in $(\t{A},\t{W})$,

5082: so that $\t\De$ has its tip at $\tilde{x}$ and  $\ol{\t{x}\t{y}}\subset \t\De$.

5083: 

5084: 

5085: Consider the vectors $\varpi_1,...,\varpi_r\in \t\De$ which are the

5086: fundamental coweights of our root system. Then the vector

5087: $\ov{\t{x}\t{y}}$ is the integer linear combination

5088: $$

5089: \ov{\t{x}\t{y}}= \sum_{i=1}^r n_i \varpi_i, n_i\in \N\cup \{0\}.

5090: $$

5091: Accordingly, we define a path $\t{p}$ in $\t\De$ with the initial

5092: vertex $\t{x}$ and the final vertex $\t{y}$ as the concatenation

5093: $$

5094: \t{p}=\pi_{\ul{\la}}= \pi_{\la_1}...*\pi_{\la_r}= \t{p}_1\cup..\cup

5095: \t{p}_r,

5096: $$

5097: where $\la_i:=n_i \varpi_i$, $\ul{\la}=(\la_1,...,\la_r)$. Observe

5098: that the path  $\t{p}$ satisfies the assumptions of Theorem

5099: \ref{chaincondition}. Moreover, $\t{p}$ is a generalized Hecke

5100: path.

5101: 

5102: 

5103: \medskip

5104: Next, let $A\subset X$ be an apartment containing $\ol{\t{z}\t{x}}$,

5105: $a\subset A$ be an alcove containing $\t{z}$; consider the

5106: retraction $f=Fold_{a,A}: X\to A$. This retraction transforms

5107: $\t{P}$ to a polygon $\hat{P}:=f(\t{P})\subset A$ which has geodesic

5108: sides $f(\ol{\t{z}\t{x}})$, $f(\ol{\t{y}\t{z}})$. Note that the

5109: break-points in

5110: $$

5111: \hat{p}:=f(\t{p})= \hat{p}_1\cup...\cup \hat{p}_r,

5112: $$

5113: are the images of the vertices $\t{x}_i$ of $\t{P}$ which are

5114: break-points $\t{p}$, but in addition we possibly have break-points

5115: within the segments $f(\t{p}_i)$. The latter can occur only at the

5116: values of $t$ for which the geodesic segments of $\t{p}_i$ intersect

5117: transversally the walls of $(\t{A},\t{W})$. Since $\t{x}$ is a special vertex and the edges of

5118: each $\t{p}_i$ are parallel to multiples of $\varpi_i$, it follows that each segment $\t{p}_i$ is

5119: contained in the 1-skeleton of $X$. Thus the break-points of $f(\t{p}_i)$ are

5120: automatically vertices of $\t{A}$. We subdivide the path $\t{p}$ so

5121: that all break-points of $\hat{p}$ are the images of the vertices of

5122: $\t{P}$.

5123: 

5124: 

5125: Let $k=k_R$ be the saturation constant of the root system $R$. Then,

5126: according to Lemma \ref{kexists}, for each vertex $v\in A$, the

5127: point $kv\in A$ is a special vertex of $A$. Therefore, applying a

5128: dilation $h\in Dil(A,W)$ with the conformal factor $k$ to the polygon

5129: $\hat{P}$, we obtain a new polygon $k\cdot \hat{P}=h(\hat{P})$,

5130: whose vertices are all special vertices of $A$. Thus we can identify

5131: the Weyl chamber $\De$ with a chamber in $A$ whose tip $o$ is at the

5132: vertex $h(\tilde{z})$ and which contains the geodesic segment

5133: $h(\ol{\tilde{z}\tilde{x}})$.

5134: 

5135: 

5136: Lastly, let $P=\P(k \hat{P})=g(\t{P})$ denote the projection of the

5137: polygon $k\hat{P}$ to the Weyl chamber $\De$, where

5138: $g=Fold_{z,h,\De}$. We set $x:=g(\t{x}), y:= g(\t{y}), x_i:=

5139: g(\t{x}_i)$, $p:= g(\t{p})$, etc.

5140: 

5141: 

5142: \begin{prop}

5143: $P=\ol{ox} \cup p \cup \ol{yo}$ is a Littelmann polygon such that

5144: $$

5145: \length(p)= k\cdot \length(\t{p}).

5146: $$

5147: \end{prop}

5148: \proof We have to show that the path $p$ satisfies the chain

5149: condition with maximal chains at each vertex.

5150: 

5151: 

5152: The chain condition at each vertex of the path $p$ follows

5153: immediately from Theorem \ref{chaincondition}.

5154: 

5155: 

5156: The maximality condition is immediate for the break-points which

5157: occur at the special vertices of $A$, in particular, for all

5158: break-points which are images of the break-points of $\hat{p}$. The

5159: remaining break-points are the ones which occur at the points $\P

5160: (\hat{x}_i)$, where $\hat{x}_i$ are smooth points of $k\hat{p}$ at

5161: which this path transversally intersects the  walls of $A$ passing

5162: through $o$. However at these points the maximality condition

5163: follows from Proposition \ref{chaincondition1}.

5164: 

5165: 

5166: The second assertion of the proposition was proven in Lemma

5167: \ref{invarianceoflength}.  \qed

5168: 

5169: 

5170: \begin{rem}

5171: \label{miniscule}

5172:  Observe that in the case when $\be$ is the sum of

5173: {\em minuscule fundamental coweights}, the multiplication by $k$ in the above

5174: proof is unnecessary since all the vertices of the polygon $\hat{P}$

5175: (and hence $P=\P(\hat{P})$) are already special. Thus, in this case,

5176: the polygon

5177: $$P=Fold_{\De}(\t{P})$$

5178: is a Littelmann polygon.

5179: \end{rem}

5180: 

5181: 

5182: \medskip

5183: The above proposition shows that the polygon $P$ is a Littelmann polygon in $\De$,

5184: which has two geodesic sides having the $\De$-lengths $k\al, k\ga^*$ and

5185: the concatenation of the remaining sides equal to a generalized LS path of the $\De$-length $k\be$.

5186: Therefore, according to Littelmann's theorem (see Theorem

5187: \ref{Lit}),

5188: $$

5189:  V_{k\ga}\subset V_{k\al}\otimes V_{k\be}.

5190: \qed

5191: $$

5192: This concludes the proof of Theorem \ref{3->4}. \qed

5193: 

5194: 

5195: \begin{cor}

5196: \label{corQ3->Q4} Suppose that $\al,\be,\ga\in L\subset P(R^\vee)$

5197: are dominant weights for the complex semisimple Lie group

5198: $\ul{G}^\vee(\C)$, such that $\al+\be+\ga\in Q(R^\vee)$ and that

5199: there exists $N\in \N$ so that

5200: $$

5201:  V_{N\ga}\subset V_{N\al}\otimes V_{N\be}.

5202: $$

5203: Then for the saturation constant $k=k^2_R$ we have

5204: $$

5205: V_{k\ga}\subset V_{k\al}\otimes V_{k\be} .

5206: $$

5207: \end{cor}

5208: \proof Let $X$ be the Euclidean (Bruhat-Tits) building associated to

5209: the group $\ul{G}(\K)$. Then the assumption that

5210: $$

5211:  V_{N\ga}\subset V_{N\al}\otimes V_{N\be}

5212: $$

5213: implies that $(N\al,N\be,N\ga^*)$ belongs to $D_3(X)$ (see

5214: \cite[Theorem 9.17]{KLM3}, or \cite[Theorem 10.3]{KLM3},  or Theorem

5215: \ref{klm3}). Since $D_3(X)$ is a homogeneous cone and $N> 0$, $(\al,\be,\ga^*)\in

5216: D_3(X)$ as well. Moreover, according to Theorem \ref{klm2}, since

5217: $\al, \be, \ga\in P(R^\vee)$ and $\al+\be+\ga\in Q(R^\vee)$, there

5218: exists a triangle $T\subset X$ with the $\De$-side-lengths

5219: $\al,\be,\ga^*$, whose vertices are also vertices of $X$. Now the

5220: assertion follows from Part 3 of theorem \ref{3->4}. \qed

5221: 

5222: 

5223: Using Remark \ref{miniscule} we also obtain:

5224: 

5225: 

5226: \begin{thm}

5227: \label{mini3->4}

5228: Let $X$ be a building as above. Suppose that $T=[x,y,z]$ is a

5229: geodesic triangle in $X$ with the $\De$-side-lengths

5230: $(\al,\be,\ga^*)$, which are dominant weights of $G^\vee$ and so that

5231: one (equivalently, all) vertices of $T$ are special and at least one

5232: of the weights $\al,\be,\ga$ is the sum of minuscule weights. Then

5233: $$

5234: V_{\ga} \subset V_{\al}\otimes V_{\be}.

5235: $$

5236: \end{thm}

5237: 

5238: 

5239: This theorem was originally proven by Tom Haines in the case when

5240: all the weights $\al, \be, \ga$ are sums of minuscules.

5241: 

5242: 

5243: \begin{conjecture}

5244: (T. Haines) Suppose that $\al, \be, \ga$ are sums of minuscule

5245: weights. Then, in the above theorem, the assumption that one vertex

5246: of $T$ is special can be replaced by $\al+\be+\ga\in Q(R^\vee)$.

5247: \end{conjecture}

5248: 

5249: 

5250: Note that (among irreducible root systems) the root systems $G_2, F_4, E_8$ have no minuscule weights,

5251: $B_n, C_n, E_7$ have exactly one minuscule weight and the root systems $A_n, D_n, E_6$ have more than 1 minuscule weights.

5252: For the root system $A_n$ Haines conjecture follows from the saturation theorem. For $D_n$ and $E_6$ it would

5253: follow from the affirmative answer to Question \ref{simplyconjecture}.

5254: 

5255: 

5256: 

5257: \begin{figure}[tbh]

5258: \centerline{\epsfxsize=5in \epsfbox{hecke.eps}} \caption{\sl A Hecke

5259: path which does not satisfy the integrality condition.}

5260: \label{hecke}

5261: \end{figure}

5262: 

5263: \begin{prop}

5264: Suppose that the root system $R$ has exactly one minuscule coweight

5265: $\la$. Then the above conjecture holds for $R$.

5266: \end{prop}

5267: \proof Let $(A,W)$ denote the Euclidean Coxeter complex

5268: corresponding to the root system $R$ and let $X$ be a thick Euclidean building modeled on $(A,W)$.

5269: Given $(\al,\be,\ga^*)\in D_3(X)$ such that $\al, \be, \ga\in P(R^\vee), \al+\be+\ga\in Q(R^\vee)$

5270: we have to construct a geodesic triangle $T=[o,x,y]\subset X$ with special vertices and

5271: the $\De$-side lengths $(\al,\be,\ga^*)$. Clearly, it suffices to treat the case when the root system $R$ is

5272: irreducible and spans $V$. Therefore $R$ has type $B_n, C_n$, or $E_7$. In particular, the index of connection $i$ of $R$ equals $2$

5273: and $-1\in W_{sph}$. In particular, $\ga=\ga^*$.

5274: Let $\la$ denote the unique minuscule coweight of $R$ and let $\La$ denote the span in $\la$ in $V$.

5275: 

5276: Observe that $\la$ does not belong to the coroot lattice $Q(R^\vee)$ and thus, since $i=2$,

5277: $$

5278: \N\cdot \la\cap Q(R^\vee)= 2\N \cdot \la.

5279: $$

5280: Suppose now that $\al=a\la, \be=b\la, \ga=b\la$, where $a, b, c\in \N\cup \{0\}$ and

5281: $$

5282: (\al,\be,\ga)\in D_3(X), \quad \al+\be+\ga\in Q(R^\vee).

5283: $$

5284: Thus $a+b+c$ is an even number and the triple $(a,b,c)$ satisfies

5285: the ordinary metric triangle inequalities.

5286: 

5287: Let $(A',W')=(\R, 2\Z\ltimes \Z/2)$ denote the rank 1 Coxeter complex;  its

5288: vertex set equals $\Z$. The positive Weyl chamber in $(A',W')$ is $\R_+$ and

5289: we can identify $\De'$-distances with the usual metric distances. Let $X'$ denote a thick

5290: building  which is modeled on $(A',W')$ (i.e. a simplicial tree with edges of unit length and thickness $\ge 3$).

5291: Then the above properties of $a, b, c$ imply that $X'$ contains a triangle $T'=[o', x', y']$ with the metric

5292: side-lengths $a, b, c$. If this triangle is contained in a single apartment $A'\subset X'$, we send $T'$ to

5293: a geodesic triangle $T\subset X$ via the isometry $A'\to \La\to A\to X$. If not, we

5294: obtain a folded (Hecke) triangle $P=f'(T')=[o', x', u', f'(y')]\subset \De'$. Note that the unit tangent directions

5295: $\xi', \eta'\in S_{u'}(A')$ to the segments $\ol{u'x'}, \ol{u' f(y')}$ are antipodal. Now embed

5296: the apartment $A'$ into $A\subset X$ via the isometry $\iota$ that sends $A'$ to $\La$, $o'$ to $o$, $1$ to $\la$ (the latter

5297: is a special vertex). Then the point $u:=\iota(u')$ is also a special vertex in $A$.

5298: We claim that the resulting  broken triangle $[o, x, u, y]\subset A$ is a Hecke triangle in $A$.

5299: Indeed, the directions $\xi, \eta$ at $\Si_u$ which are images of $\xi', \eta'$ under $\iota$ are

5300: antipodal and $\eta\in \De$. Therefore, since $u$ is a special vertex and  $-1\in W_{sph}$, according to Lemma

5301: \ref{constructingachain} $\xi\ge \eta$. Thus $\ol{xu}\cup \ol{uy}$ is a Hecke path. It follows that

5302: the broken triangle $[o, x, u, y]$ is a Hecke triangle and hence it unfolds to a geodesic triangle $T$ in $X$.

5303: The triangle $T$  has special vertices and $\De$-side lengths $(\al,\be,\ga)$.

5304: 

5305: Below is an alternative to the above argument. Let $\Si\subset \De^3$ denote the collection of

5306: triples of dominant weights $\tau, \eta, \mu$ such that

5307: $$

5308: (V_\tau \otimes V_\eta \otimes V_\mu)^{G^\vee}\ne 0,

5309: $$

5310: where $G^\vee$ is assumed to be simply-connected.

5311: This set is an additive semigroup, see for instance  \cite[Appendix]{KLM3}.

5312: It suffices to prove that $(\al,\be,\ga)\in \Si$.

5313: 

5314: Set $t:= \frac{1}{2}(a+b-c)$. (The number $t$ is the metric length of the ``leg'' of the geodesic triangle

5315: $T'\subset  X'$ in the above argument, the leg which contains the vertex $x'$.) Since $a+b+c$ is even, the number $t$

5316: is an integer. Set

5317: $$

5318: a_1:= a-t, b_1:= b-t, c_1:= c, \quad \al_1:= a_1\la, \be_1:= b_1\be, \ga_1:= c_1\ga.

5319: $$

5320: Then $c_1=a_1+b_1$ and the metric triangle inequalities for $a, b, c$ imply that $t\ge 0, a_1\ge 0, b_1\ge 0$.

5321: Thus $\al_1, \be_1, \ga_1$ are still dominant weights of $G^\vee$ and they satisfy

5322: $$

5323: \ga_1= \al_1+\be_1.

5324: $$

5325: Then, since $-1\in W_{sph}$ and $\ga_1=\ga_1^*$, we have: $(\al_1,\be_1,\ga_1)\in \Si$. Moreover,

5326: $(t\la, t\la, 0)$ also clearly belongs to $\Si$ and we have

5327: $$

5328: (\al,\be,\ga)= (\al_1,\be_1,\ga_1) + (t\la, t\la, 0)= (\al,\be,\ga). \qed

5329: $$

5330: 

5331: 

5332: 

5333: 

5334: \begin{ex}

5335: There exists a Hecke path $p\in {\mathcal P}$ such that $p(1)\in

5336: P(R^\vee)$, however for the saturation constant $k=k_R$, the path

5337: $k\cdot p$ is not an LS path.

5338: \end{ex}

5339: \proof Our example is for the root system $A_2$, in which case

5340: $k=1$. We will give an example of a Hecke path $p\in {\mathcal P}$

5341: such that $p(1)\in P(R^\vee)$ but $p$ does not belong to ${\mathcal

5342: P}_\Z$. Since, according to Theorem \ref{locint}, each LS path

5343: belong to ${\mathcal P}_\Z$, it proves that $p$ is not an LS path.

5344: 

5345: 

5346: The Hecke path $p$ in question has $\De$-length $\varpi_1+\varpi_2$,

5347: where $\varpi_1, \varpi_2$ are the fundamental coweights; the

5348: break-point of $p$ occurs at the point $-(\varpi_1+\varpi_2)/2$

5349: where the path $p$ backtracks back to the origin.

5350: Thus, for the simple roots $\al$ and $\be$, the minimum of the

5351: functions $\al(p(t)), \be(p(t))$ equals $-1/2$. See Figure

5352: \ref{hecke}.  \qed

5353: 

5354: 

5355: 

5356: 

5357: 

5358: \newpage

5359: \begin{thebibliography}{BaBE}

5360: \addcontentsline{toc}{section}{Bibliography}

5361: 

5362: 

5363: 

5364: 

5365: \bibitem[Ba]{Ballmann}

5366: W.\ Ballmann, ``Lectures on spaces of nonpositive curvature. With an

5367: appendix by Misha Brin.'' DMV Seminar, vol. 25. Birkhauser Verlag,

5368: Basel, 1995.

5369: 

5370: \bibitem[BK]{BK}

5371: P. Belkale, S. Kumar, {\em Eigencone, saturation and Horn problems for symplectic and odd orthogonal groups},

5372:  Preprint arXiv:0708.0398, 2007.

5373: 

5374: \bibitem[BS]{BS}

5375: A.\ Berenstein and R.\ Sjamaar, {\em Coadjoint orbits, moment

5376: polytopes, and the Hilbert-Mumford criterion},  Journ. Amer. Math.

5377: Soc., vol. {\bf 13} (2000), no. 2, p. 433--466.

5378: 

5379: \bibitem[B]{Borel}

5380: A. Borel, ``Linear Algebraic Groups'', Springer Verlag,

5381: Graduate Texts in Mathematics , Vol. 126.

5382: 2nd printing, 1991.

5383: 

5384: 

5385: \bibitem[Bo]{Bourbaki}

5386: N. Bourbaki, ``Lie groups and Lie algebras", Chap. 4, 5, 6. Springer Verlag, 2002.

5387: 

5388: 

5389: \bibitem[Br]{Brown}

5390: K. Brown,``Buildings'', Springer-Verlag, New York, 1989.

5391: 

5392: 

5393: \bibitem[BT]{BT}

5394: F.\ Bruhat\ and\ J. Tits, {\em Groupes reductifs sur un corps

5395: local,} Publ. Math. IHES, No. 41, (1972), p.

5396: 5--251.

5397: 

5398: \bibitem[D]{Demazure}

5399: M. Demazure, {\em Sch\'emas en groupes r\'eductifs},

5400: Bull. Soc. Math. France,  {\bf 93}  (1965) p. 369--413.

5401: 

5402: \bibitem[DW]{DW}

5403: H.\ Derksen and J.\ Weyman, {\em Semi-invariants of quivers and saturation for Littlewood-Richardson coefficients}

5404: J. Amer. Math. Soc.  {\bf 13}  (2000),  no. 3, p. 467--479.

5405: 

5406: 

5407: \bibitem[G]{Garrett}

5408: P. Garrett, ``Buildings and classical groups,'' Chapman \& Hall, London, 1997.

5409: 

5410: 

5411: \bibitem[GL]{GL}

5412: S. Gaussent and P. Littelmann, {\em LS-Galleries, the path model and

5413: MV-cycles},  Duke Math. J.  127  (2005),  no. 1, p. 35--88.

5414: 

5415: 

5416: \bibitem[Gro]{Gross}

5417: B.\ Gross, {\em On the Satake isomorphism}, In: ``Galois

5418: representations in arithmetic algebraic geometry

5419:               (Durham, 1996)'',

5420: London Math. Soc. Lecture Notes, vol. {\bf 254}, (1998) p. 223--237.

5421: 

5422: 

5423: \bibitem[Ha]{Haines}

5424: T.\ J.\ Haines, {\em Structure constants for Hecke and

5425: representations rings}, IMRN vol. {\bf 39} (2003), p. 2103--2119.

5426: 

5427: 

5428: \bibitem[KKM]{KKM}

5429: M. Kapovich, S. Kumar and J. J. Millson, {\em Saturation and irredundancy for $Spin(8)$},

5430: To appear in  Pure and Applied Mathematics Quarterly.

5431: 

5432: \bibitem[KLM1]{KLM1}

5433: M. Kapovich, B.\ Leeb and J.\ J.\ Millson, {\em Convex functions on

5434: symmetric spaces, side lengths of polygons and the stability

5435: inequalities for weighted configurations at infinity}, Preprint,

5436: June 2004.

5437: 

5438: 

5439: \bibitem[KLM2]{KLM2}

5440: M. Kapovich, B.\ Leeb and J.\ J.\ Millson,  {\em Polygons in buildings and

5441: their side-lengths}, Preprint, 2004.

5442: 

5443: 

5444: \bibitem[KLM3]{KLM3}

5445: M. Kapovich, B.\ Leeb and J.\ J.\ Millson,  {\em Polygons in symmetric

5446: spaces and buildings with applications to

5447:   algebra},  Memoirs of AMS, to appear.

5448: 

5449: 

5450: \bibitem[KL]{KleinerLeeb}

5451: B.\ Kleiner and B.\ Leeb, {\em Rigidity of quasi-isometries for

5452: symmetric spaces and Euclidean buildings}, Publ. Math. IHES, vol.

5453: {\bf 86} (1997), p. 115--197.

5454: 

5455: 

5456: 

5457: 

5458: \bibitem[KT]{KnutsonTao}

5459: A. Knutson\ and\ T. Tao, {\em The honeycomb model of ${\rm

5460: GL}_n(\C)$ tensor products. I. Proof of the saturation conjecture,}

5461: J. Amer. Math. Soc., vol. {\bf 12} (1999), no.~4, p. 1055--1090.

5462: 

5463: 

5464: 

5465: 

5466: \bibitem[L1]{Littelmann1}

5467: P.\ Littelmann, {\em  A Littlewood-Richardson rule for symmetrizable

5468: Kac-Moody  algebras},

5469: Invent. Math. 116 (1994), no. 1-3, p. 329--346.

5470: 

5471: 

5472: \bibitem[L2]{Littelmann2}

5473: P.\ Littelmann, {\em Paths and root operators in representation

5474: theory}, Annals of Math. (2)  142 (1995) no. 3, p. 499--525.

5475: 

5476: 

5477: \bibitem[L3]{Littelmann3}

5478: P.\ Littelmann, {\em Characters of representations and paths in

5479: ${\mathfrak H}^*_{\R}$}, In: ``Representation theory and automorphic

5480: forms'' (Edinburgh, 1996), p. 29--49, Proc. Sympos. Pure Math., 61,

5481: Amer. Math. Soc., Providence, RI, 1997.

5482: 

5483: 

5484: \bibitem[Ron]{Ronan}

5485: M. Ronan,``Lectures on buildings'', Perspectives in Mathematics, 7.

5486: Academic Press, Inc., 1989.

5487: 

5488: 

5489: \bibitem[Rou]{Rousseau}

5490: G. Rousseau, {\em Euclidean buildings}, Lectures at \'Ecole d'\'et\'e de

5491: math\'ematiques ``Nonpositively curved geometries, discrete groups and

5492: rigidities'', Institut Fourier, Grenoble, 2004.

5493: 

5494: \bibitem[Sat]{Satake}

5495: I.\ Satake, {\em Theory of spherical functions on reductive algebraic groups

5496: over $p$-adic fields}, Publ. Math. IHES, vol. {\bf 18} (1963), p. 1--69.

5497: 

5498: \bibitem[Sc]{Schwer}

5499: C. Schwer, {\em Galleries, Littlewood polynomials and

5500: structure constants of the spherical Hecke algebra},

5501: Int. Math. Res. Not.  2006, Art. ID 75395, 31 pp.

5502: 

5503: 

5504: \end{thebibliography}

5505: 

5506: 

5507: 

5508: 

5509: 

5510: 

5511: 

5512: 

5513: \bigskip

5514: \noindent Michael Kapovich: \newline Department of Mathematics,

5515: \newline University of California, \newline Davis, CA 95616, USA

5516: \newline

5517:  kapovich$@$math.ucdavis.edu

5518: 

5519: 

5520: \smallskip

5521: \noindent John J. Millson: \newline Department of Mathematics,

5522: \newline University of Maryland, \newline College Park, MD 20742,

5523: USA

5524: \newline

5525:  jjm$@$math.umd.edu

5526: 

5527: 

5528: 

5529: 

5530: \end{document}\end

5531: