1: \documentclass[a4paper, 12pt, draft, reqno]{amsart}

2: \pagestyle{headings}

3: \usepackage[final]{graphicx}

4: \usepackage{amsmath}

5: \usepackage{amssymb}

6: \usepackage{latexsym}

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

8: \UseComputerModernTips

9: \CompileMatrices

10: 

11: \setlength{\textheight}{8.9in}

12: \setlength{\textwidth}{6.5in}

13: \setlength{\topmargin}{0in}

14: \setlength{\oddsidemargin}{0in}

15: \setlength{\evensidemargin}{0in}

16: 

17: \DeclareMathOperator{\sgn}{sgn}

18: \newcommand{\ZZ}{{\mathbb Z}}

19: \newcommand{\RR}{{\mathbb R}}

20: \newcommand{\NN}{{\mathbb N}}

21: \newcommand{\CC}{{\mathbb C}}

22: \newcommand{\DD}{{\mathcal D}}

23: \newcommand{\dd}{{\mathcal d}}

24: \DeclareMathOperator{\Hom}{Hom}

25: \DeclareMathOperator{\Ext}{Ext}

26: \DeclareMathOperator{\End}{End}

27: \DeclareMathOperator{\soc}{soc}

28: \DeclareMathOperator{\rad}{rad}

29: \DeclareMathOperator{\im}{im}

30: \DeclareMathOperator{\hd}{hd}

31: \DeclareMathOperator{\ch}{ch}

32: \DeclareMathOperator{\mn}{min}

33: \DeclareMathOperator{\res}{res}

34: \DeclareMathOperator{\add}{add}

35: \DeclareMathOperator{\remo}{rem}

36: \newcommand{\pow}{{\mathcal P}}

37: \newcommand{\lat}{{\mathcal L}}

38: \newcommand{\dual}{^{\ast}}

39: \newcommand{\frob}{^{\mbox{\rm\tiny F}}}

40: \newcommand{\tlam}{\tilde{\lambda}}

41: \newcommand{\bmu}{\bar{\mu}}

42: \newcommand{\otherwise}{\mbox{\rm otherwise}}

43: \newcommand{\wif}{\mbox{\rm if }}

44: \newcommand{\wand}{\mbox{\rm and }}

45: \DeclareMathOperator{\good}{{\mathcal F}(\nabla)}

46: \newcommand{\Mod}{\mathrm{mod}}

47: \newcommand{\MMod}{\mathrm{Mod}}

48: \newcommand{\even}{\ \mbox{\rm even}}

49: \newcommand{\odd}{\ \mbox{\rm odd}}

50: \newcommand{\bg}{\bar{G}}

51: \newcommand{\gi}{G_1}

52: \newcommand{\GL}{\mathrm{GL}}

53: \newcommand{\SL}{\mathrm{SL}}

54: \newcommand{\St}{\mathrm{St}}

55: \newcommand{\notequiv}{\not\equiv}

56: \newcommand{\wfor}{\mbox{\rm for }}

57: \DeclareMathOperator{\ind}{ind}

58: \DeclareMathOperator{\wt}{wt}

59: \DeclareMathOperator{\pr}{pr}

60: \DeclareMathOperator{\Stab}{Stab}

61: \DeclareMathOperator{\Grot}{Grot}

62: \newcommand{\fit}{\phantom{\Big(}}

63: \newcommand{\alphac}{\alpha\check{\ }\,}

64: \newcommand{\betac}{\beta\check{\ }\,}

65: \newcommand{\embeds}{\hookrightarrow}

66: \newcommand{\ses}{short exact sequence }

67: \newcommand{\Tt}{\mathtt{T}}

68: \newcommand{\Tu}{\mathtt{U}}

69: \newcommand{\Ts}{\mathtt{S}}

70: \newcommand{\rr}{\mathfrak{r}}

71: \newcommand{\Sym}{\mathfrak{S}}

72: \DeclareMathOperator{\partn}{\Lambda^+}

73: \DeclareMathOperator{\cpartn}{\Lambda^+_{\mbox{\scriptsize{\rm{col}}}}}

74: \DeclareMathOperator{\rpartn}{\Lambda^+_{\mbox{\scriptsize{\rm{row}}}}}

75: \newcommand{\half}{\frac{1}{2}}

76: \newcommand{\bnabla}{\overline{\nabla}}

77: \newcommand{\bi}{\bar{\imath}}

78: \newcommand{\bj}{\bar{\jmath}}

79: \newcommand{\too}{\longrightarrow}

80: 

81: 

82: \begin{document}

83: \theoremstyle{plain}

84: \newtheorem{thm}{Theorem}[section]

85: \newtheorem{prop}[thm]{Proposition}

86: \newtheorem{lem}[thm]{Lemma}

87: \newtheorem{cor}[thm]{Corollary}

88: \newtheorem{conj}[thm]{Conjecture}

89: \newtheorem{claim}[thm]{Claim}

90: \theoremstyle{definition}

91: \newtheorem{rem}[thm]{Remark}

92: \newtheorem{ass}[thm]{Assumption}

93: \newtheorem{defn}[thm]{Definition}

94: \newtheorem{example}[thm]{Example}

95: 

96: 

97: \setlength{\parskip}{1ex}

98: 

99: \title[The blocks of the 

100: Brauer algebra]{The blocks of the Brauer algebra\\ in characteristic zero}

101: \author{Anton Cox}

102: \email{A.G.Cox@city.ac.uk, M.Devisscher@city.ac.uk, P.P.Martin@city.ac.uk}

103: \author{Maud De Visscher} 

104: \author{Paul Martin} 

105: \address{Centre for Mathematical Science\\

106: City University\\

107: Northampton Square\\ 

108: London\\ 

109: EC1V 0HB\\

110: England.} 

111: \subjclass[2000]{Primary 20G05}

112: 

113: \begin{abstract}We determine the blocks of the Brauer algebra

114: in characteristic zero. We also give information on

115:   the submodule structure of standard modules for this algebra.

116: \end{abstract}

117: 

118: \maketitle

119: \medskip

120: 

121: \section{Introduction}

122: 

123: The Brauer algebra $B_n(\delta)$ was introduced in \cite{brauer}

124: in the study of the representation theory of orthogonal and sympletic

125: groups. Over $\CC$, and for integral values of $\delta$, its action on

126: tensor space $T = (\CC^{|\delta|})^{\otimes n}$ can be identified with

127: the centraliser algebra for the corresponding group action. This

128: generalises the Schur-Weyl duality between symmetric and general

129: linear groups \cite{weyl}. 

130: 

131: 

132: If $n$ is fixed, then for all $\delta \geq n$ the centraliser algebra

133: $\End_{O(\delta)}(T)$ has multimatrix structure {\em independent} of

134: $\delta$, and Brauer's algebra $B_n(\delta)$ unifies these algebras,

135: having a basis independent of $\delta$, and a composition which makes

136: sense over any field $k$ and for any $\delta \in k$. The Brauer

137: algebra is well defined in particular for positive integral $\delta <

138: n$, but the action on $T$ is faithful for positive integral $\delta$

139: if and only if $\delta\geq n$.

140: 

141: In classical invariant theory one is interested in the Brauer algebra

142: {\it per se} only in so far as it coincides with the centraliser of

143: the classical group action on $T$; i.e., in the case of $\delta$

144: integral with $|\delta|$ large compared to $n$.  Here we take another

145: view, and consider the stable properties for fixed $\delta$ and

146: arbitrarily large $n$.  In such cases $B_n(\delta)$ is not semisimple

147: for $\delta$ integral.  However it belongs to a remarkable family of

148: algebras arising both in invariant theory and in statistical mechanics

149: for which this view is very natural. (For example when considered from

150: the point of view of transfer matrix algebras in statistical mechanics

151: \cite{marbook}.)  Indeed much of the structure of $B_n(\delta)$ can be

152: recovered from a suitable global limit of $n$ by localisation (and in

153: this sense its structure does not depend on $n$).

154: 

155: This family of algebras can be introduced as follows.

156: Consider the diagram of commuting actions on $T$, with $|\delta|=N$:

157: \[

158: \xymatrix@C=20pt@R=10pt{ 

159: GL(N)   \ar[ddr]  &   & \ar[ddl] \CC\Sigma_n \\

160: \cup              &   &          \cap     \\

161: O(N)    \ar[r]    & T & \ar[l]   B_n(N)   \\

162: \cup              &   &          \cap     \\

163: \Sigma_N \ar[uur] &   & \ar[uul] P_n(N)

164: }

165: \]

166: where the actions of the algebra on the right centralise the action of

167: the group on the left in the same row, and vice versa. The bottom row

168: consists of the diagonal action of $\Sigma_N$ permuting the standard

169: ordered basis of $\CC^N$ on the left, and the partition algebra

170: $P_n(N)$ on the right.  The partition algebra $P_n(\delta)$ (for any

171: $\delta$) has a basis of partitions of two rows of $n$ vertices.  The

172: Brauer algebra is the subalgebra with basis the subset of pair

173: partitions, and $\CC\Sigma_n$ is the subalgebra with basis the pair

174: partitions such that each pair contains a vertex from each row.  The

175: Brauer algebra also has a subalgebra with basis the set of pair

176: partitions which can be represented by noncrossing lines drawn

177: vertex-to-vertex in an interval of the plane with the rows of vertices

178: on its boundary. This is the Temperley-Lieb algebra $T_n(\delta)$.

179: 

180: All of these algebras are rather well understood over $\CC$, with the

181: exception of $B_n$. All their decomposition matrices are known, and

182: all of their blocks can be described by an appropriate geometric

183: linkage principle. 

184: For $\Sigma_n$ both data are trivial, since it is semisimple.

185: For $T_n$ each standard module has either one or two composition

186: factors and its alcove geometry is affine $A_1$ 

187: (affine reflections on the real line). 

188: For $P_n$ each standard module has either one or two composition

189: factors and its alcove geometry is affine $A_{\infty}$ (although

190: locally the block structure looks like affine $A_1$).

191: 

192:  Over $\CC$, the Brauer algebra is semisimple for $\delta$

193:  sufficiently large, and is generically semisimple

194:  \cite{brownbrauer}. Hanlon and Wales studied these algebras in a

195:  series of papers \cite{hw1,hw2,hw3,hw4} and conjectured that

196:  $B_n(\delta)$ is semi-simple for all non-integral choices of

197:  $\delta$. This was proved by Wenzl \cite{wenzlbrauer}. 

198: 

199: In this paper we determine the blocks of $B_n$ for $\delta$ integral.

200:  The simple modules of $B_n$ may be indexed by partitions of those

201:  natural numbers congruent to $n$ modulo 2 and not exceeding $n$, and

202:  hence by Young diagrams (if $\delta=0$ then the empty partition is

203:  omitted). We will call these indexing objects {\it weights}. Given

204:  $\delta \in R$ a ring we can associate a {\em charge}

205:  $\ch(\epsilon)\in R$ to each box $\epsilon$ in a Young diagram, as

206:  shown in Figure \ref{charge}.  We will also refer later to the usual

207:  {\em content} of boxes which, for the box $\epsilon$ in $i$-th row

208:  and $j$-th column is $c(\epsilon)=j-i$. It is easy to see that

209:  $ch(\epsilon)=\delta-1+2 c(\epsilon)$.  For each pair of diagrams

210:  $\lambda$ and $\mu$ we will also need to consider the

211: skew partitions $\lambda/(\lambda\cap\mu)$ and $\mu/(\lambda\cap\mu)$ 

212:  consisting of those boxes occuring in $\lambda$ but not $\mu$ and in

213:  $\mu$ but not $\lambda$.

214: 

215: \begin{figure}[ht]

216: $$\begin{array}{|c|c|c|cc}

217: \hline \delta -1 & \delta +1 & \delta +3 &

218: \raisebox{0pt}[15pt][10pt]{$\cdots$} \\ \hline \delta -3 & \delta -1

219: & \delta +1 & \raisebox{0pt}[15pt][10pt]{$\cdots$} \\ \hline \delta

220: -5 & \delta -3 & \delta -1 & \raisebox{0pt}[15pt][10pt]{$\cdots$} \\

221: \hline \raisebox{0pt}[10pt][10pt]{$\vdots$} &

222: \raisebox{0pt}[15pt][10pt]{$\vdots$} &

223: \end{array}$$

224: \caption{\label{charge}The charges associated to boxes in a Young diagram}

225: \end{figure}

226: 

227: %\]

228: \newcommand{\Yta}[1]{\begin{array}{|r|} \hline #1 \\ 

229:     \hline \end{array}}

230: \newcommand{\Ytab}[2]{\begin{array}{|r|} \hline #1 \\ \hline #2 \\

231:     \hline \end{array}}

232: %

233: 

234: With these notations we can now state the two main results of the

235: paper (which are valid without restriction on $\delta$).

236: 

237: \newtheorem*{thm1}{Corollary \ref{blocks}}

238: \newtheorem*{thm2}{Theorem \ref{lattice}}

239: \begin{thm1}\it

240: The simple modules  $L(\lambda)$ and $L(\mu)$ are in the same block if

241: and only if

242: 

243: (i) The boxes in $\lambda/(\lambda\cap \mu)$ (respectively

244: $\mu/(\lambda\cap\mu)$) can be put into pairs whose charges sum to

245: zero;

246: 

247: (ii) if $\lambda/(\lambda\cap \mu)$ (respectively

248: $\mu/(\lambda\cap\mu)$ contains $\Ytab{1}{{\!\! -\! 1 }}$ with no

249: $\Yta{1}$ to the right of these boxes then it contains an even number

250: of 1/-1 pairs.

251: \end{thm1}

252: 

253: Examples illustrating this result are given in Example \ref{exbet}.

254: 

255: \begin{thm2}[Summary]\it 

256: For any integral $\delta$ and natural number $l$ a standard module can be 

257: constructed (for some $B_n(\delta)$) whose socle series length is

258:   greater than $l$.  This module also has a socle layer containing at

259:   least $l$ simples.

260: \end{thm2}

261: 

262: The second result shows that the structure of standard modules can

263: become arbitrarily complicated. This is in marked contrast to the

264: partition and Temperley-Lieb algebra, and symmetric group, cases.

265: 

266: 

267: To prove these results we use the theory of towers of recollement 

268: developed in \cite{cmpx}.  This approach is already closely modelled,

269: for $B_n$, in work of Doran, Wales, and Hanlon \cite{dhw} (since both

270: papers use the methods developed in \cite{mar1}). This key paper of Doran,

271: Wales and Hanlon will be the starting point for our work, and we will

272: generalise and refine several of their results.

273: 

274: The `diagram' algebras $P_n \supset B_n \supset T_n$ are amenable to

275: many powerful representation theory techniques, and yet the

276: representation theory of the Brauer algebra is highly non-trivial in

277: comparison to the others. We shall see that, in terms of degree of

278: difficulty, the study of Brauer representation theory in

279: characteristic zero is an intermediate between the study of

280: `classical' objects in characteristic zero and the {\em grand theme}

281: of the representation theory of finite dimensional algebras, the study

282: of $\Sigma_n$ in characteristic $p$.

283: 

284: Another such intermediate class of objects are the Hecke algebras of

285: type $A$ at roots of unity, which are Ringel dual to the generalised

286: Lie objects known as quantum groups. The Brauer algebra $B_n$ in

287: characteristic zero has, through its global limit, more

288: Lie-theory-like structure than $\Sigma_n$ in characteristic $p$ (for

289: which not even a good organisational scheme within which to address

290: the problem is known, for small primes $p$). This is reminiscent of

291: the virtual algebraic Lie theory discussed for the (generalised) blob

292: algebras in \cite{mryom, mwgen}. However in the Brauer algebra case,

293: any candidate for an alcove geometry formulation will be considerably

294: more complicated \cite{nazbrauer, orram}. For these reasons we

295: consider the further study of the Brauer algebra in characteristic

296: zero to be an important problem in representation theory.

297: 

298: 

299: 

300: The paper begins with a section defining the various objects of

301: interest, and a review of their basic properties in the spirit of

302: \cite{cmpx}. This is followed by a brief section describing some basic

303: results about Littlewood-Richardson coefficients which will be needed

304: in what follows.  In Section \ref{partial} we begin the analysis of

305: blocks by giving a necessary condition for two weights to be in the

306: same block. This is based on an analysis of the action of certain

307: central elements in the algebra on standard modules, and inductive

308: arguments using Frobenius reciprocity. Section \ref{detour} constructs

309: homomorphisms between standard modules in certain special cases,

310: generalising a result in \cite{dhw}. Although not necessary

311: for the main block result, this is of independent interest.

312: 

313: The classification of blocks is completed in Section

314: \ref{blocksec}. The main idea is to show that every block contains a

315: unique minimal weight, and that there is a homomorphism from any

316: standard labelled by a non-minimal weight to one labelled by a

317: smaller weight. We also describe precisely which weights are minimal

318: in their blocks.

319: 

320: In Section \ref{bigmod} we consider certain explicit choice of

321: weights, and show inductively, via Frobenius reciprocity arguments,

322: that the corresponding standards can have arbitrarily complicated

323: submodule structures. We conclude by outlining the modifications to

324: our arguments required in the case $\delta=0$.

325: 

326: The structure of the Brauer algebra becomes mouch more complicated

327: when considered over an arbitrary field $k$. For general $k$ and

328: $\delta$ integral it is expected that this algebra still acts as a

329: centraliser algebra, and this has been shown in a recent series of

330: papers for the symplectic case \cite{dotpoly1,oe1,ddh}.  A necessary

331:  and sufficient condition for semisimplicity (which holds over

332:  arbitrary fields) was given recently by Rui \cite{ruibrauer}. The study of

333: Young and permutation modules for these algebras has been started in

334: \cite{hpbrauer}.

335:  

336: 

337: \section{Preliminaries}

338: 

339: In this section we will consider the Brauer algebra defined over a

340: general field $k$ of characteristic $p\geq 0$, although we will later

341: restrict attention to the case $k=\CC$. After reviewing the definition

342: of the Brauer algebra, we will show that families of such algebras

343: form towers of recollement in the sense of \cite{cmpx} (which we will

344: see follows from various results of Doran et.~al.~\cite{dhw}). This

345: will be the framework in which we base our analysis of these algebras.

346: 

347: Given $n\in \NN$ and $\delta \in k$, the {\it Brauer algebra}

348: $B_n(\delta)$ is a finite dimensional associative $k$-algebra

349: generated by certain Brauer diagrams.  A general {\it $(n,t)$-(Brauer)

350: diagram} consists of a rectangular box (or {\it frame}) with $n$

351: distinguished points on the northern boundary and $t$ distinguished

352: points on the southern boundary, which we call {\it nodes}. Each node

353: is joined to precisely one other by a line, and there may also be one

354: or more closed loops inside the frame. Those diagrams without closed

355: loops are called {\it reduced}. We will

356: label the northern nodes from left to right by $1,2,\ldots, n$ and the

357: southern nodes from left to right by

358: $\bar{1},\bar{2},\ldots,\bar{t}$.  We identify diagrams if they

359: connect the same

360: pairs of labelled nodes, and have the same number of closed loops.

361: Lines which connect two nodes on the

362: northern (respectively southern) boundary will be called {\it

363: northern} (respectively {\it southern}) {\it arcs}; those connecting a

364: northern node to a southern node will be called {\it propagating

365: lines}.

366: 

367: \begin{figure}[ht]

368: \includegraphics{prodex.eps}

369: \caption{Multiplication of two diagrams in $B_6(\delta)$}

370: \label{prodex}

371: \end{figure} 

372: 

373: Given an $(n,t)$-diagram $A$ and a $(t,u)$-diagram $B$, we define the

374: product $AB$ to be the $(n,u)$-diagram obtained by concatenation of

375: $A$ above $B$ (where we identify the southern nodes of $A$ with the

376: northern nodes of $B$ and then ignore the section of the frame common

377: to both diagrams). As a set, the Brauer algebra $B_n(\delta)$ consists

378: of linear combinations of $(n,n)$-diagrams. This has an obvious

379: additive structure, and multiplication is induced by concatenation. We

380: also impose the relation that any non-reduced diagram containing $m$

381: closed loops equals $\delta^m$ times the same diagram with all closed

382: loops removed. A basis is then given by the set of reduced

383: diagrams. An example of a product of two diagrams in given in Figure

384: \ref{prodex}. For convenience, we set $B_0(\delta)=k$. When

385: no confusion is likely to arise, we denote the algebra $B_n(\delta)$

386: simply by $B_n$.

387: 

388: We will now apply as much as possible from the general setup of

389: \cite{cmpx} to the Brauer algebra. The labels (A1), (A2), etc., refer

390: to the axioms in that paper. Henceforth, we assume that $\delta\neq

391: 0$; for the case $\delta=0$ see Section \ref{last}.

392: 

393: For $n\geq 2$ consider the idempotent $e_n$ in $B_n$ defined by $1/\delta$

394: times the Brauer diagram where $i$ is joined to $\bar{i}$ for

395: $i=1,\ldots n-2$, and $n-1$ is joined to $n$ and $\overline{n-1}$ is

396: joined to $\bar{n}$. This is illustrated in Figure

397: \ref{idem}.

398: 

399: \begin{figure}[ht]

400: \includegraphics{idem.eps}

401: \caption{The idempotent $e_8$}

402: \label{idem}

403: \end{figure} 

404: 

405: \begin{lem}[A1]\label{A1} For each $n\geq 2$, we have an algebra isomorphism

406: $$\Phi_n \, : \, B_{n-2}\longrightarrow e_n B_n e_n$$ which takes a

407: diagram in $B_{n-2}$ to the diagram in $B_n$ obtained by adding an

408: extra northern and southern arc to the righthand end.

409: \end{lem}

410: 

411: This allows us to define, following Green \cite{green}, an exact

412: localization functor

413: \begin{eqnarray*}

414: F_n \, :\, B_n\mbox{\rm -mod} &\longrightarrow& B_{n-2}\mbox{\rm -mod}\\

415:  M &\longmapsto& e_n M

416:  \end{eqnarray*}

417: and  a right exact globalization functor

418:  \begin{eqnarray*}

419:  G_n\, : \,  B_n\mbox{\rm -mod} &\longrightarrow& B_{n+2}\mbox{\rm -mod}\\

420: M &\longmapsto& B_{n+2}e_{n+2}\otimes_{B_n}M.

421: \end{eqnarray*}

422:  Note that $F_{n+2}G_n(M)\cong M$ for all $M\in B_n\mbox{\rm -mod}$, and

423:  hence $G_n$ is a full embedding.

424: 

425: From this we can quickly deduce an indexing set for the isomorphism

426: classes of simple $B_n$-modules. It is easy to see that

427: \begin{equation}\label{symiso}

428: B_n/B_ne_nB_n\cong k\Sigma_n

429: \end{equation} 

430: the group algebra of the symmetric

431: group on $n$ symbols. If the simple $k\Sigma_n$-modules are indexed by

432: the set $\Lambda^n$ then by \cite{green} and Lemma \ref{A1}, the

433: simple $B_n$-modules are indexed by the set

434: $$\Lambda_n = \Lambda^n \sqcup \Lambda_{n-2}= \Lambda^n \sqcup

435: \Lambda^{n-2} \sqcup \cdots \sqcup \Lambda^{\mn}$$ where $\min=0$ or

436: $1$ depending on the parity of $n$. If $p=0$ or $p>n$ then the set $\Lambda^n$

437: corresponds to the set of partitions of $n$; we write $\lambda \vdash n$ if

438: $\lambda$ is such a partition.

439: 

440: For $m-n$ even we write $\Lambda_n^m$ for $\Lambda^m$ regarded as a

441: subset of $\Lambda_n$. (If $m>n$ then $\Lambda_n^m=\emptyset$.) We

442: also write $\Lambda$ for the disjoint union of all the $\Lambda^n$,

443: and call this  the set of {\it weights} for the Brauer algebra. We

444: will henceforth abuse terminology and refer to weights as being in the

445: same block of $B_n$ if the corresponding simple modules are in the same block.

446: 

447: For $n\geq 2$ and $0\leq t\leq n/2$, define the idempotent $e_{n,t}$

448: to be $1$ if $k=0$ or $1/\delta^t$ times the Brauer diagram with edges

449: between $i$ and $\bar{i}$ for all $1\leq i\leq n-2t$ and between $j$

450: and $j+1$, and $\bar{j}$ and $\overline{j+1}$ for $n-2t+1 \leq j\leq

451: n-1$. (This is the image of $e_t$ via the isomorphisms arising in

452: Lemma \ref{A1}.)  Set $B_{n,t}=B_n / B_n e_{n,t} B_n$.

453: 

454: \begin{lem}[A2]\label{A2} The natural multiplication map

455: $$B_{n,t}e_{n,t} \otimes_{e_{n,t}B_{n,t}e_{n,t}} e_{n,t}B_{n,t}

456: \longrightarrow B_{n,t}e_{n,t}B_{n,t}$$ is bijective. If $\delta\neq

457: 0$ and either $p=0$ or

458: $p>n$ then $B_n/B_ne_nB_n$ is semisimple.

459: \end{lem}

460: \begin{proof} The second part follows from (\ref{symiso}) and standard

461:   symmetric group results.  For the first part, the map is clearly

462: surjective so we only need to show that it is also injective. It is

463: easy to verify that: \\ (i) $B_{n,t}$ has a basis given by all reduced

464: diagrams having at least $n-2t$ propagating lines,\\ (ii)

465: $B_{n,t}e_{n,t}B_{n,t}$ has a basis given by all reduced diagrams

466: having exactly $n-2t$ vertical edges, and\\ (iii)

467: $e_{n,t}B_{n,t}e_{n,t} \cong k\Sigma_{n-2t}$.

468: 

469: Now suppose that $X$ and $X'$ are diagrams in $B_{n,t}e_{n,t}$. Any

470: such diagram has a southern edge where the leftmost $n-2t$ nodes lie

471: on propagating lines, with the remaining southern nodes paired

472: consecutively. The northern edge has exactly $t$ northern arcs. We

473: will label such a diagram by $X_{v,1,\sigma}$, where $v$ represents

474: the configuration of northern arcs, $1$ represents the fixed southern

475: boundary, and $\sigma\in\Sigma_{n-2t}$ is the permutation obtained by

476: setting $\sigma(i)=j$ if the $i$th propagating northern node from the

477: left is connected to $\bar{j}$. (For later use we will denote the set

478: of elements $v$ arising thus by $V_{n,t}$, and call such elements {\it

479: partial one-row diagrams}.) Similarly a diagram $Y$ in

480: $e_{n,t}B_{n,t}$ will be labelled by $Y_{1,v,\sigma}$.

481: 

482: It will be enough to show that the multiplication map is injective on

483: the set of tensor products of diagram elements.  Given

484: $X=X_{v,1,\sigma}$ and $X'=X_{v',1,\sigma'}$ in $B_{n,t}e_{n,t}$ and

485: $Y=Y_{1,w,\tau}$, $Y'=Y_{1,w',\tau'}$ in $e_{n,t}B_{n,t}$, assume that

486: $XY=X'Y'$. Then we must have $v=v'$, $w=w'$ and $\sigma \circ \tau =

487: \sigma' \circ \tau'$. It now follows from the identifcation in (iii)

488: that $X\otimes Y = X'\otimes Y'$ in $B_{n,t}e_{n,t}

489: \otimes_{e_{n,t}B_{n,t}e_{n,t}} e_{n,t}B_{n,t}$.

490: \end{proof}

491: 

492: We immediately obtain

493: 

494: \begin{cor}[A2$'$]\label{A2'} If $\delta\neq 0$ and either 

495: $p=0$ or $p>n$ then $B_n$ is a

496:   quasi-hereditary algebra, with heredity chain given by

497: $$0\subset \cdots \subset B_n e_{n,k} B_n \subset \cdots \subset B_n e_{n,0}

498: B_n.$$ The partial ordering is given as follows: for $\lambda, \mu\in

499: \Lambda_n$ we have $\lambda \leq \mu$ if and only if either $\lambda =

500: \mu$ or $\lambda \in \Lambda_n^s$ and $\mu \in \Lambda_n^t$ with

501: $s>t$.

502: \end{cor}

503: 

504: Henceforth we assume that $p$ satisfies the conditions in Corollary

505: \ref{A2'}.  It follows from the quasi-hereditary structure that for

506: each $\lambda\in \Lambda_n$ we have a standard module

507: $\Delta_n(\lambda)$ having simple head $L_n(\lambda)$ and all other

508: composition factor $L_n(\mu)$ satisfying $\mu < \lambda$. Note that if

509: $\lambda \in \Lambda_n^n$ then

510: $$\Delta_n(\lambda)=L_n(\lambda) \cong S^\lambda$$ the lift to $B_n$

511: of the Specht module for $B_n/B_ne_n B_n \cong k\Sigma_n$.

512: 

513: 

514: Note also that by \cite[A1]{don2} and arguments as in

515: \cite[Proposition 3]{mryom}, the quasi-hereditary structure is

516: compatible with the globalization and localization functors. That is,

517: for all $\lambda\in \Lambda_n$ we have

518: \begin{eqnarray}\label{makestds}

519: &&G_n(\Delta_n(\lambda))=\Delta_{n+2}(\lambda)\\

520: &&F_n(\Delta_n(\lambda))=\left\{ \begin{array}{ll}

521:     \Delta_{n-2}(\lambda) & \mbox{if $\lambda \in \Lambda_{n-2}$}\\ 0

522:     & \mbox{otherwise} \end{array} \right. 

523: \end{eqnarray}

524: As $F_n$ is exact we also have that

525: \begin{eqnarray}\label{Lres}

526: F_n(L_n(\lambda))=\left\{ \begin{array}{ll}

527:     L_{n-2}(\lambda) & \mbox{if $\lambda \in \Lambda_{n-2}$}\\ 0

528:     & \mbox{otherwise} \end{array} \right.

529: \end{eqnarray}

530: 

531: For every partition $\mu$ of some $m=n-2t$ we can give an explicit

532: construction of the modules $\Delta_n(\mu)$.  Let $e=e_{n,t}\in B_n$

533: be as above, so that $eB_ne \cong B_m$. If we denote by $S^{\mu}$ the

534: lift of the Specht module labelled by $\mu$ for $k\Sigma_m$ to $B_m$,

535: then by (\ref{makestds}) we have that

536: \begin{equation}\label{altdef}

537: \Delta_n(\mu)\cong B_n e\otimes_{eB_ne}S^{\mu}.

538: \end{equation} 

539:  Using this fact, it is

540: easy to give a basis for this module in terms of some basis ${\mathcal

541: B}(\mu)$ of $S^{\mu}$, using the notation introduced during the proof

542: of Lemma \ref{A2}.

543: 

544: \begin{lem} \label{basisstandard} If $\mu$ is a partition of $n-2t$ then

545: the module $\Delta_n(\mu)$ has a basis given by

546: $$\{X_{v,1,id}\otimes x\,\, |\,\, v\in V_{n,t} ,\, x\in {\mathcal

547: B}(\mu)\}.$$ 

548: \end{lem}

549: 

550: Via this Lemma we may identify our standard modules

551: $\Delta_n(\lambda)$ with the modules ${\mathcal S}_{\lambda}(n)$ in

552: \cite{dhw} (which in turn come from \cite{brownbrauer}).

553: Note that if we define $\Delta_n(\mu)$ as the tensor product in

554: (\ref{altdef}) then we have a definition that makes sense for all

555: values of $p$. In the non-quasi-hereditary cases these modules still

556: play an important role, as the algebras are cellular \cite{gl} with

557: the $\Delta_n(\mu)$ as cell modules. 

558: 

559: 

560: We will frequently need a second way to relate different Brauer algebras.

561: 

562: \begin{lem}[A3]\label{A3} For each $n\geq 1$, the algebra

563:   $B_n$ can be identified as a subalgebra of

564:   $B_{n+1}$ via the homomorphism which takes a Brauer diagram

565:   $X$ in $B_n$ to the Brauer diagram in $B_{n+1}$

566:   obtained by adding two vertices $n+1$ and

567:   $\overline{n+1}$ with a line between them. 

568: \end{lem}

569: 

570: Lemma \ref{A3} implies that we can consider the usual restriction and

571: induction functors

572: \begin{eqnarray*}

573: \res_n \, :\, B_n\mbox{\rm -mod} &\longrightarrow& B_{n-1}\mbox{\rm -mod}\\

574:  M &\longmapsto& M|_{B_{n-1}}

575:  \end{eqnarray*}

576: and  

577:  \begin{eqnarray*}

578: \ind_n\, : \,  B_n\mbox{\rm -mod} &\longrightarrow& B_{n+1}\mbox{\rm -mod}\\

579: M &\longmapsto& B_{n+1}\otimes_{B_n}M.

580: \end{eqnarray*}

581: 

582: We can relate these functors to globalisation and localisation via

583: 

584: \begin{lem}[A4]\label{A4} (i) For all $n\geq 2$ we have that 

585: $$B_ne_n \cong B_{n-1}$$ as a left $B_{n-1}$, right

586:   $B_{n-2}$-bimodule.\\

587: (ii) For all $B_n$-modules

588: $M$ we have $$\res_{n+2}(G_n(M))\cong \ind_n(M).$$

589: \end{lem}

590: \begin{proof}

591:  (i) Every Brauer diagram in $B_ne_n$ has an edge between

592: $\overline{n-1}$ and $\bar{n}$. Define a map from $B_ne_n$ to

593: $B_{n-1}$ by sending a diagram $X$ to the diagram with $2(n-1)$

594: vertices obtained from $X$ by removing the line connecting

595: $\overline{n-1}$ and $\bar{n}$ and and the line from $n$, and pairing

596: the vertex $\overline{n-1}$ to the vertex originally paired with $n$

597: in $X$. It is easy to check that this gives an isomorphism.\\ 

598: (ii) Using (i) we have

599: \begin{eqnarray*} 

600: \res_{n+2}(G_n(M)) &= (B_{n+2}e_{n+2}\otimes_{B_n}M)|_{B_{n+1}}\\

601:  &\cong  B_{n+1} \otimes_{B_n} M\cong \ind M.

602:  \end{eqnarray*}

603: \end{proof}

604: 

605: 

606: 

607: Let $\lambda$ be a partition of $n$ and $\mu$ be a partition of $n-1$.

608: We write $\lambda \rhd \mu$ and $\mu \lhd \lambda$ if $\mu$ is

609: obtained from $\lambda$ by removing a box from its Young diagram

610: (equivalently if $\lambda$ is obtained from $\mu$ by adding a box to

611: its Young diagram). The following result does {\it not} require any

612: restriction on the characteristic of our field.

613: 

614: \begin{prop}[A5 and 6]\label{indres} For 

615: $\lambda \in \Lambda_n$ we have short exact sequences 

616: $$0\rightarrow \bigoplus_{\mu \lhd \lambda} \Delta_{n+1}(\mu)

617: \rightarrow \ind_n\, \Delta_n(\lambda) \rightarrow \bigoplus_{\mu

618: \rhd \lambda} \Delta_{n+1}(\mu)\rightarrow 0$$

619: and

620: $$0\rightarrow \bigoplus_{\mu \lhd \lambda} \Delta_{n-1}(\mu)

621: \rightarrow \res_n\, \Delta_n(\lambda) \rightarrow \bigoplus_{\mu

622: \rhd \lambda} \Delta_{n-1}(\mu)\rightarrow 0.$$

623: \end{prop}

624: \begin{proof} This was proved for $k=\CC$ in \cite[Theorem

625:     4.1 and Corollary 6.4]{dhw} (as the condition $\lambda\vdash n$ in

626:       \cite[Corollary 6.4]{dhw} is not needed). However, their proof

627:       is valid over any field.

628: \end{proof}

629: 

630: Wenzl \cite{wenzlbrauer} has shown that $B_n$ is semisimple when

631: $k=\CC$ and $\delta\notin\ZZ$. (Over an arbitrary field, a necessary

632: and sufficient condition for semisimplicity has been given by Rui

633: \cite{ruibrauer}.) For this reason we do not consider the case of

634: non-integral $\delta$. As we will regularly need to appeal to the

635: representation theory of the symmetric group, which is not well

636: understood in positive characteristic, we will also only consider the

637: characteristic zero case. In summary:

638: 

639: {\it Henceforth we will assume that $k=\CC$ and $\delta\in\ZZ\backslash\{0\}$,

640: unless otherwise stated.}

641: 

642: \section{Some Littlewood-Richardson coefficients}

643: 

644: One of the key results used by \cite{dhw} in their analysis of the

645: Brauer algebra is \cite[Theorem 4.1]{hw3} which decomposes standard

646: modules $\Delta_n(\lambda)$ with $\lambda\vdash n$ as symmetric group

647: modules. Recall that a partition is {\it even} if every part of the

648: partition is even, and that $c_{\mu\eta}^{\lambda}$ denotes a

649: Littlewood-Richardson coefficient.  If $\lambda\vdash n$ and

650: $\mu\vdash m$ then \cite[Theorem 4.1]{hw3} states that  either 

651: $[\res_{\CC\Sigma_n}\Delta_n(\mu):S^{\lambda}]=0$ or

652: $m=n-2t$ for

653: some $t\geq 0$ and

654: \begin{equation}\label{ressymis}

655: [\res_{\CC\Sigma_n}\Delta_n(\mu):S^{\lambda}]=

656: \sum_{\begin{array}{c}\eta

657: \vdash 2t \\ \eta \,\,\mbox{even}\end{array}}c_{\mu \eta}^{\lambda}

658: \end{equation}

659: As this result is stated in terms of Littlewood-Richardson

660: coefficients, we will find it useful to calculate these in certain

661: special cases.

662: 

663: \begin{lem}\label{LRstuff} If

664: $\mu \subset \lambda$ are partitions such that $\nu = \lambda / \mu$ is

665: also a partition

666: then

667: $$c_{\mu \eta}^\lambda = \left\{ \begin{array}{ll} 1 & \mbox{if $\eta

668: = \nu$}\\ 0 & \mbox{otherwise.} \end{array}\right.$$

669: \end{lem}

670: \begin{proof}

671: This follows immediately from the definition of

672: Littlewood-Richardson coefficients in terms of rectification of skew

673: tableaux (see \cite[Section 5.1, Corollary 2]{fultab})

674: \end{proof}

675: 

676: For our second calculation we will need an alternative definition of

677: Littlewood-Richardson coefficients (which can be found in \cite[2.8.14

678: Corollary]{jk}). When considering a configuration of boxes labeled by

679: elements $b_{ij}$ we say that the configuration is {\it valid} if:\\

680: \phantom{ i} \quad\quad\quad (i) For all $i$, if $y<j$ then $b_{iy}$

681: is in a later column than $b_{ij}$.\\ \phantom{ } \quad\quad\quad(ii)

682: For all $j$, if $x<i$ then $b_{xj}$ is in an earlier row than

683: $b_{ij}$.\\ For each box $(i,j)$ of $\eta$ consider a symbol

684: $b_{ij}$. Then the Littlewood-Richardson coefficient $c_{\mu

685: \eta}^\lambda$ is the number of ways one can form $\lambda$ from $\mu$

686: by adding the boxes of $\eta$ to $\mu$ in the following manner. First

687: add $b_{11},b_{12},\ldots, b_{1\eta_1}$ to $\eta$ to form a new

688: partition $\eta^1$. Continue inductively by adding

689: $b_{i1},b_{i2},\ldots, b_{i\eta_i}$ to $\eta^{i-1}$ to form a new

690: partition $\eta^{i}$. We require that the final configuration of the

691: elements $b_{ij}$ is valid.

692: 

693: 

694: \begin{lem}\label{LRrec} If

695: $\mu \subset \lambda$ are partitions with $\lambda=(a^b)$ for some $a$

696: and $b$ then there is a unique partition $\eta=(\eta_1,\ldots,\eta_r)$

697: such that $c_{\mu\eta}^{\lambda}\neq 0$, and for this partition we

698: have $c_{\mu\eta}^{\lambda}=1$. Further, 

699: $(\lambda/\mu)_i=\eta_{r-i}$.

700: \end{lem}

701: \begin{proof} 

702: Consider valid extensions of $\mu$ by any $\eta$ to form $\lambda$.

703: As $\lambda$ is a rectangle, the final row of $\eta$ can only be

704: placed as illustrated in Figure \ref{etaplace}(a). Then the

705: penultimate row of $\eta$ must be placed as illustrated in Figure

706: \ref{etaplace}(b). Continuing in this way we see that the choice of

707: $\eta$ is unique, and the number of boxes in the final row of

708: $\lambda/\mu$ must equal $\eta_1$, in the penultimate row must equal

709: $\eta_2$, and so on.

710: \end{proof}

711: 

712: \begin{figure}[ht]

713: $$

714: \begin{tabular}{c|c|c|c|c|c|c|c|c|c|}

715: \multicolumn{7}{l}{ }&

716: $\phantom{\raisebox{0pt}[15pt][10pt]{$\cdots$}}$\\ 

717: \cline{5-8}

718: \multicolumn{3}{l}{ } &

719: & $b_{(r-1)\eta_r}$ & $\raisebox{0pt}[15pt][10pt]{$\cdots$}$ & $b_{(r-1)2}$ &

720: $b_{(r-1)1}$\\ 

721: \cline{2-8}

722: $\phantom{\raisebox{0pt}[15pt][10pt]{$\cdots$}}$& $b_{(r-1)\eta_{r-1}}$&

723: $\raisebox{0pt}[15pt][10pt]{$\cdots$}$ & $b_{(r-1)(\eta_r+1)}$ &

724:  $b_{r\eta_r}$ & $\raisebox{0pt}[15pt][10pt]{$\cdots$}$ & $b_{r2}$ &

725: $b_{r1}$ \\ \hline

726: \end{tabular}

727: $$\caption{\label{etaplace}The final two rows of $\eta$ in $\lambda$} 

728: \end{figure} 

729: 

730: 

731: \section{\label{partial}A partial block result}

732: 

733: Doran, Wales, and Hanlon \cite{dhw} have given a necessary condition

734: for the existence of a non-zero homomorphism of $B_n$-modules from

735: $\Delta_n(\lambda)$ to $\Delta_n(\mu)$. We will first elevate this

736: condition to a partial block result, and then give a

737: stronger necessary condition that must also hold for two weights to be

738: in the same block.  In section \ref{blocksec} we will see that this

739: stronger condition is also sufficient for two weights to be in the

740: same block.

741: 

742: Let $\lambda$ be a partition. For a box $d$ in the corresponding Young

743: diagram $[\lambda]$, we denote by $c(d)$ the {\it content} of $d$.

744: Recall that if $d=(x,y)$ is in the $x$-th row (counting from top to

745: bottom) and in the $y$-th column (counting from left to right) of

746: $[\lambda]$, then $c(d)= y-x$. We denote by ${\bf c}(\lambda)$ the

747: {\it multiset} $\{ c(d)\,\, : \,\, d\in [\lambda]\}$. If $\mu$ is a

748: partition with $[\mu]\subseteq[\lambda]$ we write

749: $\mu\subseteq\lambda$, and denote the skew partition obtained by

750: removing $\mu$ from $\lambda$ by $\lambda/\mu$. We then denote by

751: ${\bf c}(\lambda/\mu)$ the multiset ${\bf c}(\lambda)\backslash{\bf

752: c}(\mu)$.

753: 

754: Write $X_{i,j}$ for the Brauer diagram in $B_n$ with edges between $t$

755: and $\bar{t}$ for all $t\neq i,j$ and with edges between $i$ and $j$

756: and between $\bar{i}$ and $\bar{j}$. Note that $B_n$ is generated by

757: the elements $X_{i,j}$ together with the symmetric group $\Sigma_n$

758: (identified with the set of diagrams with $n$ propagating lines).  We

759: denote by $T_n$ the element $\sum_{1\leq i <j \leq n} X_{i,j}$ in

760: $B_n$. Recall also the definition of partial one-row diagrams in the proof

761: of Lemma \ref{A2}.

762: 

763: \begin{lem}

764: \label{Taction} 

765: Let $\mu$ be a partition of $m$ with $m=n-2t$. For all $w\in V_{n,t}$ and

766: $x\in S^\mu$ we have that

767: $$ T_n(X_{w,1,id}\otimes x)=\Big( t(\delta -1) - \sum_{d\in [\mu]}

768: c(d) + \sum_ {1\leq i<j\leq n}(i,j)\Big) (X_{w,1,id}\otimes x)$$

769: where $(i,j)$ denotes the element of $\Sigma_n$ which transposes $i$ and $j$. 

770: Hence for all $y\in \Delta_n(\mu)$ we have

771: $$T_ny=\Big( t(\delta -1) - \sum_{d\in [\mu]} c(d) + \sum_{1\leq

772: i<j\leq n}(i,j)\Big)y.$$

773: \end{lem}

774: \begin{proof} This is essentially \cite[Lemma 3.2]{dhw}, together with

775:   observations in the proof of \cite[Theorem 3.3]{dhw}.

776: \end{proof}

777: 

778: The next result is a slight strengthening of \cite[Theorem 3.3]{dhw}

779: (which in turn generalises \cite[formula before (2.13)]{nazbrauer},

780: which considers the case $\delta\in\NN$). The original results provide

781: a necessary condition for the existence of a homomorphism between two

782: standard modules, but can be refined to prove

783: 

784: \begin{prop}\label{constar} Suppose that $[\Delta_n(\mu):L_n(\lambda)]\neq 0$.

785: Then either $\lambda=\mu$ or $\lambda\in \Lambda^r_n$ and $\mu \in

786: \Lambda_n^s$ for some $r-s=2t>0$. Further, we must have

787:  $$\mu \subseteq \lambda\quad

788: \mbox{and}\quad t(\delta -1) + \sum_{d\in [\lambda/ \mu]} c(d)

789: =0.$$

790: \end{prop}

791: \begin{proof}

792: The first part of the proposition is clear from the quasi-hereditary

793: structure of $B_n$. For the second part, note that by using the

794: exactness of the localization functor we have

795: $$[\Delta_n(\mu):L_n(\lambda)]=[\Delta_r(\mu):L_r(\lambda)]$$ and

796: hence we may assume that $\lambda$ is a partition of $n$.  In this

797: case, $L_n(\lambda)=\Delta_n(\lambda)=S^\lambda$, the lift of the

798: Specht module for $\CC\Sigma_n$ to $B_n(\delta)$, and so any Brauer

799: diagram having fewer than $n$ propagating lines must act as zero on

800: $L_n(\lambda)$. In particular, all the $X_{i,j}$'s act as zero and

801: hence so does $T_n$.

802: 

803: The condition that $\mu \subseteq \lambda$ now follows by

804: regarding $\Delta_n(\mu)$ as a $\CC\Sigma_n$-module by restriction and

805: using (\ref{ressymis}) which describes the multiplicities of

806: composition factors of such a module.

807: 

808: For the final condition, we know by assumption that there must exist a 

809: $B_n$-submodule $M$ of $\Delta_n(\mu)$ and a $B_n$-homomorphism

810: $$\phi \, : \, L_n(\lambda) \longrightarrow \Delta_n(\mu) / M.$$

811: Let $N$ be the $B_n$-submodule of $\Delta_n(\mu)$ containing $M$ such that 

812: $$\phi(L_n(\lambda))=N/M.$$ As $N|_{\CC\Sigma_n}$ is semisimple, we can

813: find a $\CC\Sigma_n$-submodule $W$ of $N$ such that $N=W\oplus M$ and

814: $W\cong S^{\lambda}$.  By Lemma \ref{Taction} we have for all $y\in W$

815: that

816: $$T_ny=\Big( t(\delta -1) - \sum_{d\in [\mu]} c(d) + \sum_{1\leq

817: i<j\leq n}(i,j)\Big)y.$$

818: 

819: But $W\cong S^\lambda$ is a simple $\CC\Sigma_n$-module and $\sum_{1\leq

820: i<j\leq n}(i ,j)$ is in the centre of $\CC\Sigma_n$, so it must act

821: as a scalar on $W$. It is well known \cite[Chapter 1]{Diaconis} that

822: this scalar is given by $\sum_{d\in [\lambda]}c(d)$. Hence we have

823: \begin{eqnarray*}

824: T_ny &=& \Big( t(\delta -1) - \sum_{d\in [\mu]} c(d) + \sum_{d\in

825: [\lambda]}c(d)\Big)y\\ &=&\Big( t(\delta -1) + \sum_{d\in

826: [\lambda / \mu]} c(d)\Big)y.

827: \end{eqnarray*}

828: But $T_n$ must act as zero on $N$ 

829: and hence $t(\delta-1) + \sum_{d\in

830: [\lambda/ \mu]}c(d)=0$.

831: \end{proof}

832: 

833: By standard quasi-heredity arguments \cite[Appendix]{don2} 

834: we deduce

835: 

836: \begin{cor}\label{weak} 

837: Suppose that $\lambda\in \Lambda_n^r$ and $\mu\in \Lambda_n^{s}$

838:  with $s<r$. If $\lambda$ and $\mu$ are in the same block then

839:  $s=r-2t$ for some $t\in\NN$ and

840: \begin{equation}\label{bal}

841: t(\delta -1) + \sum_{d\in [\lambda]}c (d) - \sum_{d\in [\mu]}c(d) =0.

842: \end{equation}

843: \end{cor}

844: 

845: When $t=2$ \cite{dhw} gave a necessary and sufficient

846: condition for the existence of a standard module homomorphism. From

847: their results we obtain

848: 

849: \begin{thm}\label{dhwhom}

850: Suppose that $\mu\subset\lambda$ with

851: $|\lambda/\mu|=2$. Then

852: $$\dim\Hom(\Delta_n(\lambda),\Delta_n(\mu))\leq 1$$

853: and is non-zero if and only if $\lambda$ and $\mu$ satisfy (\ref{bal}) with

854: $\lambda/\mu\neq(1^2)$. Indeed, if $\lambda/\mu=(1^2)$ then 

855: $$[\Delta_n(\mu):L_n(\lambda)]=0.$$

856: \end{thm}

857: \begin{proof} It is enough to consider the case when 

858: $\lambda\vdash n$, as the general case follows by globalisation.  If

859: $\lambda$ and $\mu$ do not satisfy the required conditions then there

860: is no composition factor $L_n(\lambda)$ in $\Delta_n(\mu)$

861: (and hence no homomorphism) by Corollary \ref{weak} and the remarks after

862: \cite[Theorem 3.1]{dhw}. In the remaining cases the existence of such

863: a homomorphism was shown in \cite[Theorem 3.4]{dhw}. By the remarks

864: after \cite[Theorem 3.1]{dhw} the multiplicity of the simple module

865: $\Delta_n(\lambda)$ in $\Delta_n(\mu)$ is $1$, and the dimension

866: result is now immediate.

867: \end{proof}

868: 

869: The next result is a strengthening of Proposition \ref{constar}.

870: 

871: \begin{prop}\label{strongstar} Suppose that

872: $[\Delta_n(\mu):L_n(\lambda)]\neq 0$. Then 

873: there is a pairing of the boxes in $\lambda / \mu$ such that the sum

874: of the content of the boxes in each pair is equal to $1-\delta$.

875: \end{prop}

876: \begin{proof}

877: We use induction on $n$; the case $n=2$ is covered by Proposition

878: \ref{constar}. Thus we assume that the result holds for $n-1$ and will

879: show that it holds for $n$. 

880: 

881:  If $[\Delta_n(\mu):L_n(\lambda)]\neq 0$ then by Proposition

882: \ref{constar} we know that $\mu \subseteq \lambda$ and

883: \begin{equation}\label{tocompare}

884: t(\delta -

885: 1) + \sum_{d\in [\lambda / \mu]} c(d) =0

886: \end{equation}

887:  where

888: $2t=|\lambda|-|\mu|$. Now suppose, for a contradiction, that there is

889: no pairing of the boxes of $[\lambda / \mu]$ satisfying the

890: condition of the proposition.  By localising we may assume that

891: $\lambda$ is a partition of $n$, so that

892: $L_n(\lambda)=\Delta_n(\lambda)$. Thus $\Delta_n(\mu)$ has a submodule

893: $M$ such that $\Delta_n(\lambda)\hookrightarrow \Delta_n(\mu)/M $. 

894: 

895: The partition $\lambda$ has a removable box $\epsilon_i$ of

896: content $s$ say and by Proposition \ref{indres} we have a surjection 

897: $\ind_{n-1}\Delta_{n-1}(\lambda - \epsilon_i) \rightarrow

898: \Delta_n(\lambda)$. Hence we have

899: $$\Hom(\ind_{n-1}\Delta_{n-1}(\lambda - \epsilon_i),

900: \Delta_n(\mu)/M)\neq 0$$ 

901: and so by Frobenius reciprocity we have

902: $${\rm Hom}(\Delta_{n-1}(\lambda - \epsilon_i), 

903: {\res_n}(\Delta_n(\mu)/M))\neq 0.$$ This implies that

904:   $\Delta_{n-1}(\lambda - \epsilon_i)=L_{n-1}(\lambda - \epsilon _i)$

905:   is a composition factor of $\res_n(\Delta_n(\mu))$. Now using

906:   Proposition \ref{indres} we see that either\\ 

907: (i) the weight $\mu$ has a removable box $\epsilon_j $ such that

908:   $[\Delta_{ n-1}(\mu - \epsilon_j):L_{n-1}(\lambda - \epsilon_i)]\neq

909:   0$, or\\ 

910: (ii) the weight $\mu$ has an addable box $\epsilon_j$ such that

911:   $[\Delta_{ n-1}(\mu + \epsilon_j):L_{n-1}(\lambda - \epsilon_i)]\neq

912:   0$.\\

913: We consider each case in turn.

914: 

915: In case (i), Proposition \ref{constar} implies that $[\mu -

916: \epsilon_j]\subseteq [\lambda - \epsilon_i]$ and

917: $$t(\delta - 1) + \sum_{d\in [\lambda / \mu]}c(d) -

918: c(\epsilon_i) + c(\epsilon_j ) = 0.$$ Hence from (\ref{tocompare})

919: we must have

920: $$c(\epsilon_j)= c(\epsilon_i)=s$$ and by induction we can find a

921: pairing of the boxes in $(\lambda - \epsilon_i)/(\mu -\epsilon_j)$

922: such that the sum of the content of the boxes in each pair is equal

923: to $1-\delta$. But as multisets

924: $${\bf c}((\lambda - \epsilon_i) / (\mu - \epsilon_j)) = {\bf

925: c}(\lambda/ \mu) - c(\epsilon _i) + c(\epsilon_j) = {\bf

926: c}(\lambda / \mu)$$ and hence there is such a pairing for the

927: boxes of $\lambda/ \mu$. This gives the desired contradiction.

928: 

929: Now consider case (ii). Here $\mu$ has an addable box $\epsilon_j$ such

930: that $[\mu + \epsilon_j] \subseteq [\lambda - \epsilon_i]$ and

931: $$(t-1)(\delta - 1) + \sum_{d\in [\lambda / \mu]}c(d) - c(\epsilon_i)

932: - c(\epsilon_j) = 0.$$ Comparing with (\ref{tocompare}) we deduce that

933: $$c(\epsilon_j) + c(\epsilon_i) = 1-\delta.$$ By induction there

934: is a pairing of the boxes of $(\lambda - \epsilon_i) / (\mu +

935: \epsilon_j)$ satisfying the condition of the Proposition. But as multisets

936: $${\bf c}((\lambda - \epsilon_i) / (\mu + \epsilon_j)) = {\bf

937: c}(\lambda / \mu) \, - c(\epsilon_i) - c(\epsilon_j)$$ and as

938: observed above the $c(\epsilon_i)$ and $c(\epsilon_j)$ can be paired

939: in the right way. Hence the boxes of $\lambda / \mu$ can be

940: paired appropriately, which again gives the desired contradiction.

941: \end{proof}

942: 

943: When $\delta$ is even we will need a further refinement of Proposition

944: \ref{constar}. Given $\mu\subset\lambda$, consider the boxes with

945: content $-\frac{\delta}{2}$ and $\frac{2-\delta}{2}$ in

946: $\lambda/\mu$. If $[\Delta_n(\mu):L_n(\lambda)]\neq 0$ then these must

947: be paired by Proposition \ref{strongstar}, and so must be in one of

948: the two chain configurations illustrated in Figure \ref{diag} (for

949: some length of chain).

950: 

951: \begin{figure}[ht]

952: \includegraphics{diag.eps}

953: \caption{The two possible configurations of paired boxes of 

954: contents $-\frac{\delta}{2}$ and $\frac{2-\delta}{2}$} 

955: \label{diag}

956: \end{figure} 

957: 

958: \begin{prop}\label{beststar} Suppose that

959: $[\Delta_n(\mu):L_n(\lambda)]\neq 0$ and $\delta$ is even. If the

960: boxes of content $-\frac{\delta}{2}$ and $\frac{2-\delta}{2}$ are

961: configured as in Figure \ref{diag}(b) then the number of columns in

962: this configuration must be even.

963: \end{prop}

964: \begin{proof}

965: We will show by induction on $n$ that in case

966: (b) the number of columns must be even. The case $n=2$ is covered by 

967: Theorem \ref{dhwhom}.

968: 

969: By repeated applications of $F$ we may assume that $\lambda\vdash

970: n$. Let $\epsilon_i$ be a removable box of $\lambda$. As in the proof

971: of Proposition \ref{strongstar} we have that if 

972: $$[\Delta_n(\mu):L_n(\lambda)]\neq 0$$

973: then either 

974: $$[\Delta_{n-1}(\mu-\epsilon_j):L_{n-1}(\lambda-\epsilon_i)]\neq 0$$ for

975: some removable box $\epsilon_j$ of $\mu$ with

976: $c(\epsilon_i)=c(\epsilon_j)$ and $\mu-\epsilon_j\subset\lambda-\epsilon_i$, or

977: $$[\Delta_{n-1}(\mu+\epsilon_j):L_{n-1}(\lambda-\epsilon_i)]\neq 0$$ for

978: some addable box $\epsilon_j$ of $\mu$ with

979: $c(\epsilon_i)+c(\epsilon_j)=1-\delta$ and 

980: $\mu+\epsilon_j\subseteq\lambda-\epsilon_i$.

981: 

982: If $c(\epsilon_i)$ is not equal to either $-\frac{\delta}{2}$ or

983: $\frac{2-\delta}{2}$ then the boxes of

984: $(\lambda-\epsilon_i)/(\mu-\epsilon_j)$ (respectively of

985: $(\lambda-\epsilon_i)/(\mu+\epsilon_j)$) of content

986: $-\frac{\delta}{2}$ and $\frac{2-\delta}{2}$ are the same as those

987: boxes in $\lambda/\mu$, and so the result follows by induction. Also,

988: by our assumption on the configuration of such boxes the partition

989: $\lambda$ does not have a removable box of content

990: $\frac{2-\delta}{2}$.

991: Thus we may assume that $\lambda$ has only one removable box

992: $\epsilon_i$ of content $-\frac{\delta}{2}$ (and hence that $\lambda$

993: is a rectangle).

994: 

995: \begin{figure}[ht]

996: \includegraphics{rectlm.eps}

997: \caption{The partitions $\mu\subset\lambda$, with the

998:  configuration as in Figure \ref{diag}(b) shaded}

999: \label{rectlm}

1000: \end{figure} 

1001: 

1002: We have that $\lambda$ and $\mu$ are of the form shown in Figure

1003: \ref{rectlm}, with $[\Delta_n(\mu):L_n(\lambda)]\neq 0$. So in particular

1004: $$[\res_{\CC\Sigma_n}\Delta_n(\mu):S^{\lambda}]\neq 0.$$ 

1005: By (\ref{ressymis}) we have 

1006: $$[\res_{\CC\Sigma_n}\Delta_n(\mu):S^{\lambda}]= \sum_{\eta\ \mbox{\rm

1007: \tiny even}}c_{\mu\eta}^{\lambda}$$ and hence we must have

1008: $c_{\mu\eta}^{\lambda}\neq 0$ for some even partition

1009: $\eta=(\eta_1,\ldots,\eta_r)$. As $\lambda$ is a rectangle Lemma \ref{LRrec}

1010: implies there is only one possible $\eta$, and that each row of

1011: $\lambda/\mu$ has length $\eta_i$ for some $1\leq i\leq r$. But $\eta$

1012: was an even partition and hence these lengths are all even, which implies

1013: that the number of columns occupied by shaded boxes in Figure

1014: \ref{rectlm} is also even as required.

1015: \end{proof}

1016: 

1017: 

1018: \begin{defn} We say that $\lambda$ and $\mu$ are {\em $\delta$-balanced} 

1019: (or just {\em balanced} when the context is clear) if:

1020: (i) there

1021: exists a pairing of the boxes in $\lambda/(\lambda\cap \mu)$

1022: (respectively in $\mu/(\lambda\cap\mu)$) such that

1023: the contents of each pair sum to $1-\delta$, and  (ii) if $\delta$

1024: is even and the boxes with content $-\frac{\delta}{2}$ and

1025: $\frac{2-\delta}{2}$ in $\lambda/(\lambda\cap \mu)$

1026: (respectively in $\mu/(\lambda\cap\mu)$) are configured as

1027: in Figure \ref{diag}(b), then the number of columns in this

1028: configuration is even.

1029: \end{defn}

1030: 

1031: Just as for Corollary \ref{weak} we can

1032: immediately deduce from Propositions \ref{strongstar} and

1033: \ref{beststar} the following block result.

1034: 

1035: \begin{cor}\label{better}

1036:  If $\lambda$ and $\mu$ are in the same block then they are balanced.

1037: \end{cor}

1038: 

1039: 

1040: \begin{figure}[ht]

1041: \includegraphics{whybetter.eps}

1042: \caption{The diagrams $[\lambda]$, $[\mu]$ and

1043: $[\tau]$ in Example \ref{exbet}(i)}

1044: \label{whybetter}

1045: \end{figure} 

1046: 

1047: \begin{example}\label{exbet} (i)

1048: Let $\lambda=(6,4^2,2,1)$, $\mu=(5,2^2)$, $\tau=\lambda/(\lambda\cap\mu)$,

1049: and $\delta=1$. The diagrams $[\lambda]$, $[\mu]$, and $[\tau]$ are

1050: illustrated (with their contents) in Figure \ref{whybetter}. Clearly

1051: $$\sum_{d\in [\lambda]}c (d) - \sum_{d\in [\mu]}c(d) =0$$ and hence

1052: $\lambda$ and $\mu$ satisfy the conditions in Corollary

1053: \ref{weak}. However, there is no pairing of the boxes in $[\tau]$ such

1054: that the content of each pair sums to zero, and hence $\lambda$ and

1055: $\mu$ cannot lie in the same block.\\ (ii) Let $\alpha=(5,4^4)$,

1056: $\beta=(5,1^4)$, $\gamma=\alpha/(\alpha\cap\beta)$, and $\delta=2$. The

1057: diagrams $[\alpha]$, $[\beta]$, and $[\gamma]$ are illustrated (with

1058: their contents) in Figure \ref{whybetter}. In this case the boxes in

1059: $[\gamma]$ can be put into pairs such that each pair sums to

1060: $1-\delta=-1$, but the boxes with contents $0$ and $-1$ are in

1061: configuration (b) from Figure \ref{diag}, and occupy an odd number of columns. Hence $\alpha$ and $\beta$ cannot lie in the same block.

1062: \end{example}

1063: 

1064: \begin{figure}[ht]

1065: \includegraphics{why2better.eps}

1066: \caption{The diagrams $[\alpha]$, $[\beta]$ and

1067: $[\gamma]$ in Example \ref{exbet}(ii)}

1068: \label{why2better}

1069: \end{figure} 

1070: 

1071: 

1072:  By Corollary \ref{better} weights which are not balanced

1073: will lie in different blocks. Hence for a $B_n$-module $X$ we will

1074: denote by $\pr_{\lambda}X$ the direct summand of $X$ with composition

1075: factors $L_n(\mu)$ such that $\mu$ and $\lambda$ are balanced.

1076: 

1077: \begin{lem} \label{nonsp}

1078: Suppose that $\lambda\vdash n$ and $\epsilon_i\in\remo(\lambda)$.\\

1079: (i) There exists a $B_n$-module $X$ and a short exact sequence

1080: $$0\too

1081: X\too\pr_{\lambda}\ind_{n-1}\Delta_{n-1}(\lambda-\epsilon_i)\too

1082: \Delta_n(\lambda)\too 0.$$ Here

1083: $X\cong\Delta_n(\lambda-\epsilon_i-\epsilon_j)$ if

1084: $(\lambda-\epsilon_i-\epsilon_j,\lambda)$ is a balanced pair

1085: or $X=0$ if no such $\epsilon_j$ exists. In the former case the

1086: sequence is non-split. 

1087: 

1088: \noindent (ii) If 

1089: $$\Hom(\pr_{\lambda}\ind_{n-1}\Delta_{n-1}(\lambda-\epsilon_i),

1090: \Delta_n(\mu))\neq

1091: 0$$ then $[\Delta_n(\mu):L_n(\lambda)]\neq 0$.

1092: \end{lem}

1093: \begin{proof} (i)

1094: The existence of such a sequence, and the form of $X$, follows from

1095: Proposition \ref{indres} and Corollary \ref{better}.

1096: To see that the sequence

1097: is non split, we proceed by induction on $|\lambda|$, the case where

1098: $\lambda=\emptyset$ being clear. By Frobenius reciprocity we have

1099: \begin{equation}\label{lhrh}

1100: \Hom(\Delta_{n-1}(\lambda-\epsilon_i),

1101: \res\Delta_n(\lambda-\epsilon_i-\epsilon_j))\cong

1102: \Hom(\ind\Delta_{n-1}(\lambda-\epsilon_i),

1103: \Delta_n(\lambda-\epsilon_i-\epsilon_j))

1104: \end{equation}

1105: By (\ref{makestds})  and Lemma \ref{A4}(ii) the left-hand side equals

1106: $$\Hom(\Delta_{n-1}(\lambda-\epsilon_i),

1107: \ind\Delta_{n-2}(\lambda-\epsilon_i-\epsilon_j)).$$ As

1108: $\Delta_{n-1}(\lambda-\epsilon_i)$ is simple, we have by the induction

1109: hypothesis and Theorem \ref{dhwhom} that this Hom-space is one

1110: dimensional. Hence the right-hand side of (\ref{lhrh}) is also one

1111: dimensional, which by another application of Theorem \ref{dhwhom}

1112: implies that the desired sequence is non-split as required.

1113: 

1114: Part (ii) is an immediate consequence of (i).

1115: \end{proof}

1116: 

1117: 

1118: \section{Computing some composition multiplicities\label{detour}} 

1119: 

1120: So far we have concentrated on conditions which imply that weights lie

1121: in different blocks of the algebra. In this section we will find

1122: certain pairs of weights which do lie in the same block, which we will

1123: demonstrate by determining certain composition factors of standard

1124: modules, and homomorphisms between such modules.

1125: 

1126: We first consider the special case where the skew partition

1127: $\lambda/\mu$ is itself a partition. For such pairs we will be able to

1128: show precisely when $L_n(\lambda)$ is a composition factor of

1129: $\Delta_n(\mu)$.  We first give a necessary condition, in Proposition

1130: \ref{rectanglenecess}, which is a generalisation of \cite[Corollary

1131: 9.1]{dhw} (the latter only considers the case $\mu=\emptyset$ and

1132: homomorphisms rather than composition factors). 

1133: 

1134: \begin{prop}\label{rectanglenecess}

1135: Let $\mu \subset \lambda$ are partitions such that $\nu = \lambda /

1136: \mu$ is also a partition. If

1137: $$[\Delta_n(\mu):L_n(\lambda)]\neq 0$$ then $\nu = (a^b)$ where $a$ is

1138: even and $b=\delta + a - 1 + 2c$, where $c$ is the content of the

1139: top lefthand box of $\nu$. Moreover, in this case we have

1140: $$[\Delta_n(\mu): L_n(\lambda)]=1.$$

1141: \end{prop}

1142: \begin{proof}

1143: As usual, by localisation we can assume that $\lambda$ is a partition of

1144: $n$. First suppose that $[\Delta_n(\mu) : L_n(\lambda)]\neq 0$. As

1145: $L_n(\lambda)$ is simply the lift of $S^{\lambda}$ for $\CC\Sigma_n$, we

1146:   have that

1147: $$[\res_{\CC\Sigma_n}\Delta_n(\mu) \, : \, S^\lambda ]\neq 0.$$

1148: By (\ref{ressymis}) we have

1149: $$ [\res_{\CC\Sigma_n}\Delta_n(\mu):S^{\lambda}]=

1150: \sum_{\begin{array}{c}\eta

1151: \vdash 2k \\ \eta \,\,\mbox{even}\end{array}}c_{\mu \eta}^{\lambda}.

1152: $$ Hence we see that $\nu$ must be an even partition, and by

1153: Lemma \ref{LRstuff} that $[\Delta_n(\mu) : L_n(\lambda)]=1$.

1154: 

1155:   On the other hand, using Proposition

1156: \ref{strongstar} we know that there is a pairing of the boxes of

1157: $\nu$ such that the sum of the content of the boxes in each pair is

1158: equal to $1-\delta$.  Clearly we have a submodule $M$ of

1159: $\Delta_n(\mu)$ and an embedding 

1160: $$\Delta_n(\lambda) \hookrightarrow\Delta_n(\mu)/M.$$ 

1161: If $\epsilon_i$ is any removable box of

1162: $\lambda$ then we have a surjective homomorphism

1163: $${\ind_{n-1}}\, \Delta_{n-1}(\lambda - \epsilon_i) \rightarrow

1164: \Delta_n(\lambda).$$ 

1165: Composing these maps we see that

1166: $${\Hom}\, ({\ind_{n-1}}\, \Delta_{n-1}(\lambda - \epsilon_i),

1167: \Delta_n(\mu) / M ) \neq 0$$ 

1168: and so by Frobenius reciprocity we have

1169: $${\Hom}\, (\Delta_{n-1}(\lambda - \epsilon_i), {\res_n}\,

1170: (\Delta_n(\mu)/M) )\neq 0.$$ 

1171: Thus 

1172: $$[{\res_n}\, \Delta_n(\mu): L_{n-1}(\lambda - \epsilon_i)]\neq 0$$

1173: and

1174: hence either $\mu$ must have a removable box $\epsilon_j$ such that 

1175: $$[\Delta_{n-1}(\mu - \epsilon_j) : L_{n-1}(\lambda - \epsilon_i)]\neq 0$$

1176: or $\mu$ must have an addable box $\epsilon_j$ such that 

1177: $$[\Delta_{n-1}(\mu + \epsilon_j) : L_{n-1}(\lambda - \epsilon_i)]\neq

1178: 0.$$ 

1179: 

1180: In the first case we have $\mu - \epsilon_j \subset \lambda -

1181: \epsilon_i$ and hence $c (\epsilon_j) = c(\epsilon_i)$. However, as

1182: $\lambda/\mu$ is a partition this is impossible, as no removable box

1183: in $\mu$ can have the same content as some box in $\lambda/\mu$. Hence

1184: we must be in the second case with $\mu + \epsilon_j \subset \lambda -

1185: \epsilon_ i$, so in fact $\epsilon_j$ must be a box in $\nu = \lambda

1186: / \mu$. As $\nu$ is a partition, there is only one such addable box

1187: and its content is given by $c$.  Thus we must have

1188: $$c(\epsilon_i) = 1-\delta - c.$$

1189: 

1190:  Now, if $\nu = \lambda / \mu$

1191: had another removable box then it would have to have the same

1192: content. But different removable boxes have different contents. Hence

1193: $\nu$ can only have one removable box, i.~e. it is a rectangle $\nu =

1194: (a^b)$, where $a$ is even as $\nu$ must be an even partition.  The

1195: content of the only removable box of $\nu$ inside of $\lambda$ is

1196: given by $c + a-1 - (b-1) = c+a-b$ and this must be equal to $1-\delta

1197: -c$. Hence  we get

1198: $$b=\delta - 1 +a +2c$$

1199: as required.

1200: \end{proof}

1201: 

1202: We will show that the condition in Proposition

1203: \ref{rectanglenecess} is also sufficient. This generalises

1204: \cite[Theorem 9.2]{dhw}, which again only considers homomorphisms and

1205: the case $\mu=\emptyset$. Before doing this we will review some

1206: standard symmetric groups results which we will require. Details can be found in \cite[Chapter 7]{fultab}

1207: 

1208: We will need to consider a set of idempotents $\{e_{\lambda}\, :\,

1209: \lambda\vdash n\}$ in $\CC\Sigma_n$, such that

1210: $\CC\Sigma_ne_{\lambda}\cong S^{\lambda}$. We will choose

1211: \begin{equation}\label{edef}

1212: e_{\lambda}=\frac{f^{\lambda}}{n!}\sum_{\sigma\in

1213: C_{\lambda}}\sum_{\tau\in R_{\lambda}}\sgn(\sigma)\sigma\tau

1214: \end{equation} where

1215: $f^{\lambda}=\dim S^{\lambda}$, $C_{\lambda}$ is the column stabiliser

1216: of $[\lambda]$ and $R_{\lambda}$ is the row stabiliser of

1217: $[\lambda]$. For example $e_{(2)}$ and $e_{(1,1)}$ (regarded as

1218: elements of $B_2$) are illustrated in Figure \ref{twoids}.

1219: 

1220: \begin{figure}[ht]

1221: \includegraphics{twoids.eps}

1222: \caption{The elements  $e_{(2)}$ and  $e_{(1,1)}$}

1223: \label{twoids}

1224: \end{figure} 

1225: 

1226:  We will also need the fact that

1227: $$\ind_{\CC(\Sigma_a\times\Sigma_b)}^{\CC\Sigma_{a+b}}\left(S^{\mu}\otimes

1228: S^{\nu}\right)\cong \bigoplus_{\lambda\vdash

1229: (n+m)}c_{\mu\nu}^{\lambda}S^\lambda.$$ As all these group algebras

1230: are semisimple, this implies by Frobenius reciprocity that

1231: \begin{equation}\label{resdec}

1232: \res^{\CC\Sigma_n}_{\CC(\Sigma_{a}\times \Sigma_b)}S^{\lambda} \cong 

1233: \bigoplus_{\mu \vdash

1234: a,\ \nu\vdash b} c_{\mu\nu}^{\lambda}(S^\mu \otimes S^{\nu}).

1235: \end{equation}

1236: Particular values of $c_{\mu\nu}^{\lambda}$ which we will need are

1237: those where $\nu=(2)$, respectively $\nu=(1,1)$.  In these cases

1238: $c_{\mu\nu}^{\lambda}$ is at most $1$, and is non-zero precisely when

1239: $\lambda/\mu$ consists of two boxes in different columns, respectively

1240: different rows.

1241: 

1242: \begin{thm}

1243: \label{rectanglesuff} Supppose that 

1244: $\mu \subset \lambda$ and $\lambda /\mu = \nu =(a^b)$. If $a$ is even

1245:  and $b=\delta - 1 + a +2c$ where $c$ is the content of the top left

1246:  box of $\nu$ then $$[\Delta_n(\mu)\, :\, L_n(\lambda)]=1.$$ Moreover,

1247:  if $\lambda \vdash n$ then

1248: $${\rm Hom}_{B_n}(L_n(\lambda), \Delta_n(\mu))={\mathbb C}.$$

1249: \end{thm}

1250: \begin{proof}

1251: We can assume without loss of generality that $\lambda \vdash n$. We

1252: have seen in the proof of Proposition \ref{rectanglenecess} that

1253: $[\res_{\CC\Sigma_n}\Delta_n(\mu)\, :\, S^\lambda]=1$. Let $W=e_\lambda

1254: \Delta_n(\mu)$, which is isomorphic to $S^\lambda$ as a

1255: $\Sigma_n$-module. To show this is in fact a $B_n$-submodule of

1256: $\Delta_n(\mu)$, it will be enough to show that $X_{i,j}W=0$ for all

1257: $1\leq i < j \leq n$. Indeed, it is enough to show that this holds for a

1258: single choice of $i$ and $j$, as

1259: $$\sigma X_{i,j}\sigma^{-1}=X_{\sigma(i),\sigma(j)}$$

1260: for all $\sigma\in\Sigma_n$.

1261: 

1262: So let us fix $i$ and $j$ with $1\leq i < j \leq n$ and use the embedding 

1263: $$\Sigma_{n-2} \times \Sigma_{2} \subset \Sigma_{n}$$ where $\Sigma_2$

1264: is the symmetric group on $\{i,j\}$ and $\Sigma_{n-2}$ the symmetric

1265: group on $\{1,\ldots , n\} \setminus \{i,j\}$.  By (\ref{resdec}) and

1266: the remarks following we have

1267: $$\res_{\CC(\Sigma_{n-2}\times \Sigma_2)}W \cong \bigoplus_{\alpha \vdash n-2}

1268: (S^\alpha \otimes S^{(1,1)}) \bigoplus_{\beta \vdash n-2} (S^\beta

1269: \otimes S^{(2)})$$ where we sum over all $\alpha$'s obtained from

1270: $\lambda$ by removing 2 boxes in different rows and over all $\beta$'s

1271: obtained from $\lambda$ by removing two boxes in different columns.

1272: 

1273: The map $X_{i,j}\, : \, \Delta_n(\mu) \longrightarrow \Delta_n(\mu)$

1274: is a $\CC\Sigma_{n -2}\times \CC\Sigma_2$-homomorphism.  Note that

1275: we have $X_{i,j}(\Delta_n(\mu)) \subset U$ where $U$ is the span of all

1276: elements of the form $X_{w,1,id} \otimes x$ where $w$ has an arc

1277: between $i$ and $j $ and $x\in S^\mu$. Regarding $U$ as a

1278: $B_{n-2}$-module acting on the strings excluding $i$ and $j$ it is

1279: easy to see that $U$ is isomorphic to $\Delta_{n-2}(\mu)$, and the

1280: restriction of this action to $\CC\Sigma_{n-2}$ is the same as

1281: restriction to the action of the first component of $\CC\Sigma_{n

1282: -2}\times \CC\Sigma_2$ regarded as a subalgebra of $B_n$. Also, it is

1283: clear that $X_{ij}$ kills the element $e_{(1,1)}$ in Figure

1284: \ref{twoids}, and hence kills the simple module $S^{(1,1)}$. Combining

1285: these observations with (\ref{ressymis}) 

1286: we deduce that, as a $\CC\Sigma_{n-2}\times

1287: \CC\Sigma_2$-module, $U$ decomposes as

1288: $$U=\bigoplus_\tau c_\tau (S^\tau \otimes S^{(2)})$$ 

1289: where

1290: $$c_\tau = \sum_{\begin{array}{c} \tau\vdash n-2 \\ \eta \,

1291: \mbox{even}\end{array}} c_{\mu \eta}^\tau.$$

1292: 

1293: Consider the restriction $X_{i,j}\, : \,

1294: W\longrightarrow U$. We want to show that $X_{i,j}W=0$. Look at the

1295: simple summands of $W$. Every summand of the form $S^\alpha \otimes

1296: S^{(1,1)}$ is sent to zero as it does not appear in $U$. Moreover, if

1297: $\mu$ is not contained in $\beta$ then $S^\beta \otimes S^{(2)}$ is

1298: sent to zero as $U$ only contains simple modules $S^\eta \otimes

1299: S^{(2)}$ with $\mu \subset \eta$. So we only need to show that

1300: $$X_{i,j} (S^\beta \otimes S^{(2)})= 0$$ for any $\beta \vdash n-2$

1301: with $\mu \subset \beta$ and $\beta$ obtained from $\lambda$ by

1302: removing two boxes in different columns. But there is only one such $

1303: \beta$, namely the partition obtained from $\lambda$ by removing two

1304: boxes from the last row of $\nu$, i.e $\beta / \mu = (a^{b-1}, a-2)$,

1305: and by Lemma \ref{LRstuff} the coefficient of $S^{\beta}\otimes

1306: S^{(2)}$ in $U$ equals $1$.

1307: 

1308: Write $W=V\oplus Y$ where $V=S^\beta \otimes S^{(2)}$. As $V$ is

1309: simple, either $X_{i,j}$ embeds $V$ into $U$ or $X_{i,j}V=0$.  Label

1310: the boxes of the partition $\lambda$ with the numbers $1,2, \ldots ,

1311: n$ starting with the first row from left to right, then the second row

1312: from left to right, etc., until the last row. Say that the last box of

1313: the partition $\nu = (a^b)$ inside of $\lambda$ is labelled by $l$. Up

1314: until now $X_{i,j}$ was arbitrary; we now fix $i=l-1$ and $j= l$ and

1315: we want to show that $X_{l-1, l}V=0$.  

1316: 

1317: \begin{figure}[ht]

1318: \includegraphics{labels.eps}

1319: \caption{The labelling of $\lambda$, with $\nu$ shaded

1320: and $\mu$ unshaded}

1321: \label{labels}

1322: \end{figure} 

1323: 

1324: Fix a partial one-row diagram $w_0$ with $t$ arcs defined as follows:

1325: suppose the $u$-th row of $\nu$ inside of $\lambda$ is labelled by

1326: $x_u, x_u+1, \ldots, x_u + a -1$ for $1\leq u \leq b$, as illustrated

1327: in Figure \ref{labels}. Then $w_0$ is defined to have arcs $\{x_u, x_u

1328: +1\}, \{x_u+2, x_u +3\}, \ldots \{x_u+a-2, x_u+a-1\}$ for $1\leq u

1329: \leq b$. (Note that $x_b +a-1 = l$.) We will represent elements of

1330: $V_{n,t}$ by adding bars to the Young tableau joining each pair of

1331: nodes connected by an arc. Thus the element $w_0$ will be represented

1332: by the diagram in Figure \ref{bars}. Usually we will only represent

1333: the boxes of $\nu$ in such a diagram.

1334: 

1335: \begin{figure}[ht]

1336: \includegraphics{bars.eps}

1337: \caption{A diagrammatic representation of the element

1338: $w_0$ }

1339: \label{bars}

1340: \end{figure} 

1341: 

1342: Now consider the element of

1343: $\Delta_n(\mu)$ given by $X_{w_0,1,id}\otimes x $ for some $x\in

1344: S^\mu$. Then $e_\lambda (X_{w_0,1,id}\otimes x)\in W$, so it decomposes

1345: as

1346: $$e_\lambda (X_{w_0,1,id}\otimes x) = v + y$$ where $v\in V$ and $y\in

1347: Y$. Note that this decomposition is independent of $\delta$.  As

1348: observed above, we have $X_{l-1,l}e_\lambda(X_{w_0,1,id}\otimes

1349: x)=X_{l-1 ,l}v$. Consider the coefficient of $X_{w_0,1,id}\otimes x$

1350: in $X_{l-1,l}v$. We will show that it is a non-zero multiple of

1351: $$\delta -1 + a - b +2c.$$ Hence, as $v$ is independent of $\delta$ we

1352: see that $v\neq 0$, but when $\delta -1 + a - b +2c=0$, we have

1353: $X_{l-1,l}v=0$. Thus $X_{l-1,l}$ cannot embed $V$ into $U$ and so it

1354: must map $V$ to zero.

1355: 

1356: Using the labelling of the boxes of $\lambda$ defined above, we will

1357: identify the row and column stabilisers $R_\lambda$ and $C_\lambda$ as

1358: subgroups of $\Sigma_n$, the symmetric group on

1359: $\{1,\ldots,n\}$. From (\ref{edef}) we have

1360: $$e_\lambda (X_{w_0,1,id}\otimes x) = \frac{f^\lambda}{n!}

1361: \sum_{\sigma\in C_\lambda} \sum_{\tau\in R_\lambda}{\rm

1362: sgn}(\sigma)\sigma \tau (X_{w_0,1,id}\otimes x),$$ and so

1363: $$X_{l-1,l}e_\lambda (X_{w_0,1,id}\otimes x) = \frac{f^\lambda}{n!}

1364: \sum_{\sigma \in C_\lambda} \sum_{\tau\in R_\lambda}{\rm

1365: sgn}(\sigma)X_{l-1,l}\sigma \tau (X_ {w_0,1,id}\otimes x).$$ We want

1366: to find the coefficient of $X_{w_0,1,id}\otimes x$ in this  sum. We

1367: consider several cases.

1368: 

1369: 

1370: \noindent\textbf{Case 1:} Suppose that $\sigma\tau X_{w_0,v_k,id}$ has

1371: an arc $\{l-1,l\}$.

1372: 

1373: In this case  $X_{l-1,l}\sigma \tau (X_{w_0,1,id}\otimes

1374: x)= \delta \sigma \tau (X_{w_0,1 ,id}\otimes x)$. If we want

1375: $\sigma\tau (X_{w_0,1,id}\otimes x)$ to be in ${\rm span}\{X

1376: _{w_0,1,id}\otimes S^\mu\}$ then we must have

1377: \begin{eqnarray*}

1378: & \tau = \tau_1 \tau_2 \qquad &\mbox{with} \,\, \tau_1 \in

1379: R_{\mu\subset\lambda},

1380:  \, \tau _2 \in 

1381: R_{\lambda}^{0}\\ & \sigma = \sigma_1  \sigma_2 \qquad

1382: &\mbox{with} \,\, \sigma_1 \in C_{\mu\subset \lambda}, \, \sigma_2 

1383: \in C_{\lambda}^{0}

1384: \end{eqnarray*}

1385: where $R_{\mu\subset\lambda}$ denotes the subgroup of $R_\lambda$

1386: (isomorphic to $R_{\mu}$) which preserves the rows of $\mu$ and fixes

1387: everything in $\nu$ and $R_{\lambda}^{0}$ denotes the subgroup of

1388: $R_\lambda$ which fixes $X_{w_0,1,id}$ as a diagram (i.e. fixes all

1389: but the $t$ northern arcs, which may be permuted amongst themselves

1390: and be reversed). In a similar way we define $C_{\mu\subset\lambda}$

1391: and $C_{\lambda}^{0}$.

1392: 

1393: Set $r=|R_{\lambda}^0|$. As the $a$ columns of $\nu$ are paired by the

1394: bars in $w_0$, and each pair of such columns may be permuted freely by

1395: $C_{\lambda}^0$ we have $|C_{\lambda}^0|=(b!)^{a/2}$. Moreover ${\rm

1396: sgn}(\sigma_2)=1$ as $\sigma_2$ is an even permutation (as it is made

1397: up of pairs of identical permutations, corresponding to the paired

1398: ends of a bar) and so ${\rm sgn}(\sigma) = {\rm sgn}(\sigma _1)$.

1399: Hence in this case we get the contribution

1400: \begin{eqnarray*}

1401: && \frac{f^\lambda}{n!} \sum_{\sigma_2\in C_{\lambda}^0}

1402: \sum_{\sigma_1\in C_{\mu\subset\lambda}} \sum_{\tau _2\in

1403: R_{\lambda}^0} \sum_{\tau_1\in R_{\mu\subset\lambda}} {\rm

1404: sgn}(\sigma_1 \sigma_2)\sigma_1 \sigma_2 \tau_1 \tau_2

1405: (X_{w_0,1,id}\otimes x)\\ && = \frac{f^\lambda}{n!} \sum_{\sigma_2\in

1406: C_{\lambda}^0}\sum_{\tau_2\in R_{\lambda}^0} \sigma_2 \tau_2

1407: (X_{w_0,1,id}\otimes \sum_{\sigma_1\in C_{\mu\subset\lambda}}

1408: \sum_{\tau_1\in R_{\mu\subset\lambda}}{\rm sgn}(\sigma_1)\sigma_1

1409: \tau_1 (x))\\ && = \frac{f^\lambda}{n!}  \frac{|\mu

1410: |}{f^\mu}\sum_{\sigma_2\in C_{\lambda}^0} \sum_{\tau_2\in

1411: R_{\lambda}^0} \sigma_2 \tau_2 (X_{w_0,1,id}\otimes e_\mu (x))\\ && =

1412: \frac{f^\lambda}{n!} \frac{|\mu |}{f^\mu} r

1413: (b!)^{a/2}(X_{w_0,1,id}\otimes x)

1414: \end{eqnarray*}

1415: using for the second equality the isomorphisms

1416: $C_{\mu\subset\lambda}\cong C_\mu$ and $R_{\mu\subset\lambda}\cong R_\mu$,

1417: and for the final equality the fact that $e_\mu (x)=x$ for all $x\in

1418: S^\mu$.

1419: 

1420: 

1421: \noindent

1422: \textbf{Case 2:} Suppose that neither $l-1$ nor $l$ is part of an

1423: arc in $\sigma \tau X_{w_0, 1,id}$.  

1424: 

1425:  In this case $X_{l-1,l}\sigma\tau

1426: X_{w_0,1,id}$ has $t+1$ arcs in the top row and so $X

1427: _{l-1,l}(X_{w_0,1,id}\otimes x)=0$.

1428: 

1429: 

1430: \noindent

1431: \textbf{Case 3:} Suppose that in $\sigma \tau X_{w_0,v_k,id}$ there are

1432: arcs $\{l-1, i\}$ and $\{l,j\}$.

1433: 

1434:  In this case, $X_{l-1,l}\sigma\tau

1435: X_{w_0,1,id}$ is obtained from $\sigma \tau X _{w_0,1,id}$ by

1436: replacing the arcs $\{l-1, i\}$ and $\{ l,j\}$ by the arcs $\{

1437: i,j\}$ and $\{l-1, l\}$. Hence if we want to have $X_{l-1,l}\sigma\tau

1438: (X_{w_0,1,id}\otimes x)$ lying in ${\rm span}\, \{X_{w_0,1,id}\otimes

1439: S^\mu\}$ then $\{i,j\}$ must be an arc of $w$ and $i=j\pm 1$. Here

1440: we consider two subcases.

1441: 

1442: \noindent

1443: \textbf{Subcase 3(a):} First assume that the pair $\{i, j\}$ is not in

1444: the last double column. Then $ \tau = \tau_2 \tau_1$ with $\tau_1\in

1445: R_{\mu\subset\lambda}$ and $\tau_2\in \tilde{\tau} R_{\lambda}^0$,

1446: where $\tilde{\tau}= (u-1,v)$ or $(u,v)$ such that $v$ is a box of

1447: $\nu$ in the same column as $l$ (possibly $l$ itself) and $u$ is the

1448: box of $\nu$ in the same row as $v$ and in the same column as

1449: $\max(i,j)$. An example of such a situation is illustrated in Figure

1450: \ref{3acase}.

1451: 

1452: \begin{figure}[ht]

1453: \includegraphics{3acase.eps}

1454: \caption{An example of subcase 3(a)}

1455: \label{3acase}

1456: \end{figure} 

1457: 

1458: 

1459: Thus we have $b$ choices for $v$ and $(\frac{a}{2}-1)$ choices for the

1460: position of $\{i,j\}$ (and hence of $u$), and so there are $2 b

1461: (\frac{a}{2}-1)$ choices for $\tilde{\tau}$. Hence there are $2 r b

1462: (\frac{a}{2}-1)$ choices for $\tau_2$. Now $\sigma = \sigma_2 \sigma

1463: _1$ where $\sigma_1 \in C_{\mu\subset\lambda}$, and $\sigma_2$

1464: permutes the pairs in all double columns, except the last and the

1465: double column containing $\{j-1,j\}$, arbitrarily. In the last double

1466: column it must send $v-1$ to $l-1$ and $v$ to $l$, and in the double

1467: column containing $\{j-1, j\}$, it can permute the pairs in any way

1468: (as $\{j-1,j\}$ can be any pair in this double column). So we get

1469: $(b!)^{\frac{a}{2}-2} (b-1)! \, b!$ possibilities for $\sigma_2$. Note

1470: also that $\sigma_ 2$ is always an even permutation and so ${\rm

1471: sgn}(\sigma) = {\rm sgn}(\sigma_1) $.  Thus in this subcase, we get a

1472: contribution of

1473: \begin{eqnarray*}

1474: &&\frac{f^\lambda}{n!}\,

1475: 2\,r\,b\,(\frac{a}{2}-1)\,(b!)^{(\frac{a}{2}-2)}\,(b-1) !\, b!\,

1476: X_{w_0,1,id}\otimes \sum_{\sigma_1\in C_{\mu\subset\lambda}}

1477:  \sum_{\tau_1\in R_{\mu\subset\lambda}}

1478: {\rm sgn}(\sigma_1) \sigma_1 \tau_1 (x)\\ && = \frac{f^\lambda}{n!}

1479: \, \frac{|\mu |!}{f^\mu}\, r \,(a-2)\, (b!)^{\frac{a}{ 2}}

1480: \,X_{w_0,1,id}\otimes x

1481: \end{eqnarray*} 

1482: where the equality follows as in Subcase 1.

1483: 

1484: 

1485: \noindent{\bf Subcase 3(b):} Next assume that the pair $\{i,j\}$ is in

1486: the last column. We must have $\tau = \tau_2 \tau_1$ where $\tau_1 \in

1487: R_{\mu\subset\lambda}$ and $\tau_2 \in R_{\lambda}^0$. Also $\sigma =

1488: \sigma _2 \sigma_1$ where $\sigma_1 \in C_{\mu\subset\lambda}$ and

1489: $\sigma_2 \in (j,l)C_{\lambda}^0$. We have $b-1$ choices for $j$ being

1490: a box of $\nu$ in the same column as $l$. Note that in this case ${\rm

1491: sgn}(\sigma_2)=-1$ and so ${\rm sgn}(\sigma)=-{\rm sgn}(\sigma_1)$.

1492: Hence arguing as in Subcases 1 and 3(a) we get a

1493: contribution of

1494: $$-\, \frac{f^\lambda}{n!}\, \frac{|\mu |!}{f^\mu} \, r\, (b-1)\,

1495: (b!)^{\frac{a} {2}}\, X_{w_0,1,id}\otimes x.$$

1496: 

1497: \noindent

1498: \textbf{Case 4:} Suppose that in $\sigma \tau X_{w_0,1,id}$ there is a

1499:  link from $l-1$ to $i$, say, and $l$ is not part of an

1500:  arc (or vice versa).

1501: 

1502: In this case $X_{l-1,l} \sigma \tau X_{w_0,1,id}$ is obtained from

1503:  $\sigma \tau X_{w_0,1 ,id}$ by replacing the arc $\{i,l-1\}$ (or

1504:  $\{i,l\}$) with the arc $\{l-1, l\}$ and $i$ is not part of an arc

1505:  any more. So, if we want to have $X_{l-1,l} \sigma \tau X_{w_0,1,id}$

1506:  in ${\rm span}\{X_{w_0,1,id}\otimes S^\mu\}$ then $i$ cannot be one

1507:  of the boxes of $\nu$. There are various potential subcases that can

1508:  arise. After action by an element of $ R_{\mu\subset\lambda}$ the

1509:  element $i$ may be in any box in the same row of $\mu$. There are

1510:  three cases: (a) $i$ is now to the left of the first column of $\nu$;

1511:  (b) $i$ is above $\nu$ but not above $l-1$ or $l$; (c) $i$ is above

1512:  $l-1$ or $l$.

1513: 

1514: \noindent{\bf Subcase 4(a):} First, assume that the box $i$ is in a

1515: column to the left of $\nu$ in $\lambda$.  In this case, $\tau =

1516: \tau_2 \tau_1$ where $\tau_1 \in R_{\mu\subset\lambda}$ (as we have

1517: already acted by such an element to put $i$ in this case above) and

1518: $\tau_2 \in ( v-1, u)R_{\lambda}^0$, or $\tau_2 \in(v,u)R_{\lambda}^0$

1519: where $v$ is any box in $\nu$ in the same column as $l$ and $u$ is the

1520: box of $\mu$ in the same row as $v$ and in the same column as $i$.  An

1521: example of such a situation is illustrated in Figure \ref{4acase}.

1522: 

1523: \begin{figure}[ht]

1524: \includegraphics{4acase.eps}

1525: \caption{An example of subcase 4(a)}

1526: \label{4acase}

1527: \end{figure} 

1528: 

1529: Let $c_1$ be the number of columns of $\lambda$ to the left of $\nu$.

1530:  Then there are $ 2 r \, b\, c_1$ possible choices of $\tau_2$. Now

1531:  $\sigma=\sigma_2 \sigma_1$ where $\sigma_1 \in C_{\mu\subset\lambda}$

1532:  (as $i$ is an arbitrary element in its column of $\mu$) and $\sigma_2$

1533:  permutes the pairs in each of the first $(\frac{a}{2}-1)$ double

1534:  columns of $\nu$ arbitrarily, and in the last double column sends

1535:  $v-1$ to $l-1$ and $v$ to $l$ and then permutes the other pairs

1536:  arbitrarily. Note that ${\rm sgn}(\sigma_1)={\rm sgn}(\sigma)$. Hence

1537:  (arguing as in earlier cases) we get a contribution of

1538: \begin{eqnarray*}

1539: && \frac{f^\lambda}{n!}\, \frac{|\mu |!}{f^\mu} \, 2 \, r\, b\, c_1 \,

1540: (b!)^{(\frac{a}{2}-1)} \, (b-1)!\, X_{w_0,1,id}\otimes x\\ && =

1541: \frac{f^\lambda}{n!}\, \frac{|\mu |!}{f^\mu} \, r\, (b!)^{\frac{a}{2}}

1542: \, 2 c_1 \, X_{w_0,1,id}\otimes x.

1543: \end{eqnarray*}

1544: 

1545: 

1546: \noindent{\bf Subcase 4(b):} Suppose that $i$ is a box of $\mu$ which

1547: is above some column of $\nu$ but to the left of $l-1$. Then the only

1548: way to use row and column permutations not involving

1549: $R_{\mu\subset\lambda}$ (which we have already used to position $i$)

1550: to connect $i$ and $l$ (or $l-1$) is by some pair $\tau$ and $\sigma$

1551: similar to that shown in Figure \ref{4bcase}. But (as illustrated) any

1552: such pair does not preserve the remaining edges in $\nu$. Hence this

1553: subcase cannot arise.

1554: 

1555: \begin{figure}[ht]

1556: \includegraphics{4bcase.eps}

1557: \caption{An example of the impossibility of subcase

1558: 4(b)}

1559: \label{4bcase}

1560: \end{figure} 

1561: 

1562: 

1563: \noindent{\bf Subcase 4(c):} Finally we are left with the subcase

1564: where after action by $R_{\mu\subset\lambda}$ the element $i$ is in a

1565: box of $\mu$ which is either in the same column as $l-1$ or in the

1566: same column as $l$. In this case $\tau = \tau_2 \tau_1$ where $\tau_1

1567: \in R_{\mu\subset\lambda}$ and $\tau_2 \in R_{\lambda}^0$. Also,

1568: $\sigma = \sigma_2 \sigma_1$ where $\sigma_1 \in

1569: C_{\mu\subset\lambda}$ (as $i$ is an arbitrary element in its column

1570: of $\mu$) and either $\sigma_2 \in (i,l)C_{\lambda}^0$ or $\sigma_ 2

1571: \in (i, l-1)C_{\lambda}^0$. If $c_2$ is the number of columns above

1572: $\nu$ in $\lambda$ then there are $ 2 c_2$ choices for the position of

1573: $i$. Note that here ${\rm sgn}(\sigma_2) = -1 $ and so ${\rm

1574: sgn}(\sigma)= - {\rm sgn}(\sigma_1)$. Hence, in this case we get a

1575: contribution of

1576: $$- \frac{f^\lambda}{n!}\, \frac{|\mu |!}{f^\mu} \, r\,

1577: (b!)^{\frac{a}{2}} 2 c_2 \, \, X_{w_0,1,id}\otimes x.$$

1578: 

1579: Note that the final sets of permutations obtained in Subcases 4(a) and

1580: 4(c) are disjoint, so there is no double counting in these

1581: contributions. Now on adding up all contributions from Cases 1--4 we

1582: see that the coefficient of $X_{w_0,1,id }\otimes x$ inside of

1583: $X_{l-1,l}e_\lambda (X_{w_0,1,id}\otimes x)$ is given by

1584: $$\frac{f^\lambda}{n!}\, \frac{|\mu |!}{f^\mu} \, r\,

1585: (b!)^{\frac{a}{2}}\, (\delta -1 +a -b +2 (c_1 - c_2)).$$

1586: 

1587:  The content of the top left box of the partition $\nu$ inside the

1588: partition $\lambda$ is given by $c=(c_1 +1) - (c_2 +1)=c_1 -

1589: c_2$. Thus we have proved that this coefficient is a non-zero

1590: multiple of $(\delta - 1) +a -b +2c$ as required. 

1591: \end{proof}

1592: 

1593: \section{\label{blocksec}The blocks of the Brauer algebra}

1594: 

1595: In section \ref{partial} we saw that a necessary condition for two

1596: weights $\lambda$ and $\mu$ to be in the same block was that the pair

1597: was balanced. We will now show that this condition is also

1598: sufficient. The key idea will be to construct from any partition

1599: $\lambda$ in a balanced pair with some $\mu\subset\lambda$ a partition

1600: $\nu\subset \lambda$ and a homomorphism connecting $\Delta_n(\lambda)$

1601: and $\Delta_n(\nu)$. This will allow us to proceed by induction.

1602: 

1603: Given a partition $\lambda$ we denote by $\add(\lambda)$ the set of

1604: {\it addable} boxes of $\lambda$ (i.e. the set of boxes which may be

1605: added to $\lambda$ such that the new shape is still a

1606: partition). Similarly we denote by $\remo(\lambda)$ the set of {\it

1607: removable} boxes of $\lambda$. If $\mu\subset\lambda$ then we denote

1608: the set of boxes in $\remo(\lambda)$ which are also boxes of

1609: $\lambda/\mu$ by $\remo(\lambda/\mu)$.  Distinct boxes in

1610: $\add(\lambda)$ (respectively in $\remo(\lambda)$) have distinct

1611: contents, and we will identify such boxes by their contents. We will

1612: order the boxes in $\lambda$ with a given content by saying that box

1613: $\epsilon$ is {\it smaller} than box $\epsilon'$ if $\epsilon$ appears on an

1614: earlier row than $\epsilon'$.

1615: 

1616: \begin{defn}\label{maxisub}

1617: Suppose that $\mu\subset\lambda$ is a balanced pair. For each

1618: $\epsilon_i\in\remo(\lambda/\mu)$ we wish to consider $\mu^i$, the

1619: {\em $i$-maximal balanced subpartition between $\mu$ and

1620: $\lambda$}. This is the maximal partition $\mu^i\subset\lambda$ such

1621: that $\mu^i$ does not contain $\epsilon_i$ and $\lambda$ and $\mu^i$

1622: form a balanced pair. We will construct $\mu^i$ by recursively

1623: defining a series of skew partitions $(\lambda/\mu^i)_j$ which will

1624: eventually equal the skew partition $\lambda/\mu^i$. There is by the

1625: pairing condition a maximal box (i.e. all others smaller) with

1626: content $c(\epsilon_i')$ such that

1627: $c(\epsilon_i)+c(\epsilon_i')=1-\delta$. Let

1628: $(\lambda/\mu^i)_0=\{\epsilon_i,\epsilon_i'\}$. Given

1629: $(\lambda/\mu^i)_m$, we set

1630: $$(\lambda/\mu^i)_{m+1}=(\lambda/\mu^i)_m\cup A_{m+1}\cup A_{m+1}'$$

1631: where $A_{m+1}$ is the set of boxes $\epsilon$ in $\lambda$ such that

1632: $\epsilon$ is to the right of or below a box in $(\lambda/\mu^i)_m$,

1633: and $A'_{m+1}$ is the set of boxes $\epsilon'$ in $(\lambda/\mu)$ such that

1634: $c(\epsilon)+c(\epsilon')=1-\delta$ for some $\epsilon\in A_{m+1}$ and

1635: $\epsilon'$ is maximal with such content among the boxes of

1636: $\lambda/\mu$ not already in $(\lambda/\mu^i)_m$.

1637: 

1638: This iterative process eventually stabilises, and we obtain

1639: $(\lambda/\mu^i)_t$ which is a (possibly disconnected) subset of the

1640: edge of $\lambda/\mu$, having width one. (In particular it does not

1641: contain two boxes with the same content.) If $\delta$ is even and

1642: $(\lambda/\mu^i)_t$ does not contain a vertical pair of boxes with

1643: content $\frac{2-\delta}{2}$ and $-\frac{\delta}{2}$, or $\delta$ is

1644: odd and $(\lambda/\mu^i)_t$ does not contain a box of content

1645: $\frac{1-\delta}{2}$ then we set

1646: $\lambda/\mu^i=(\lambda/\mu^i)_t$. Otherwise if $\delta$ is even we

1647: set

1648: \begin{equation}\label{twocase}

1649: (\lambda/\mu^i)_{t+1}=(\lambda/\mu^i)_t\cup\{x,y\}

1650: \end{equation}

1651:  where $x,y$ are

1652: the maximal boxes in $\lambda$ of content $\frac{2-\delta}{2}$ and

1653: $-\frac{\delta}{2}$ not in $(\lambda/\mu^i)_t$, and if $\delta$ is odd we set 

1654: \begin{equation}\label{onecase}

1655: (\lambda/\mu^i)_{t+1}=(\lambda/\mu^i)_t\cup\{z\}

1656: \end{equation}

1657:  where $z$ is the

1658: maximal box in $\lambda$ of content $\frac{1-\delta}{2}$ not in

1659: $(\lambda/\mu^i)_t$. This new skew partition is not necessarily stable

1660: under the addition of boxes $A$ and $A'$ as above, and we repeat that

1661: process again until the skew partition eventually stabilises at some

1662: step $s$. We then set $\lambda/\mu^i=(\lambda/\mu^i)_s$. Thus

1663: $\lambda/\mu^i$ is a removable subset of $\lambda/\mu$ having width at

1664: most two (so at most two boxes with any given content).

1665: \end{defn}

1666: 

1667: \begin{figure}[ht]

1668: \includegraphics{3ex.eps}

1669: \caption{Two examples of the $\lambda/\mu^i$ construction}

1670: \label{3ex}

1671: \end{figure} 

1672: 

1673: \begin{example}

1674: We will now consider several examples of this construction. First let

1675: $\lambda=(6,5,5,2,1)$ and $\mu=(6,4,1)$; this is a balanced pair for

1676: $\delta=2$. If $\epsilon_i$ is any of the removable boxes in

1677: Figure \ref{3ex}(a), then $\lambda/\mu^i$ is the shaded region

1678: shown.  For an example where the resulting skew

1679: partition is connected, consider $\lambda=(7,6,5,5,2,2)$ and

1680: $\mu=(7,4,4,1,1)$. This is a balanced pair for $\delta=2$. If $\epsilon_i$

1681: is  any of the removable boxes in $\lambda/\mu$ then the skew

1682: partition $\lambda/\mu^i$ is the shaded region shown in Figure

1683: \ref{3ex}(b). In this case there is a pair of boxes in the skew

1684: partition with contents $\frac{2-\delta}{2}$ and $-\frac{\delta}{2}$

1685: (i.e. $0$ and $-1$), but we do not get a strip of width $2$ because

1686: these boxes are not vertically aligned.

1687: 

1688: For an example of the full iterative process consider

1689: $\lambda=(7,6,4^4,1^2)$ and $\mu=(4,3^4)$. This is a balanced pair for

1690: $\delta=2$, and after the first part of the iterative process the skew

1691: partition stabilises into the lightly shaded region shown in Figure

1692: \ref{4ex}(a). However, we now have a vertical pair in the skew partition

1693: with contents $\frac{2-\delta}{2}$ and $-\frac{\delta}{2}$ (i.e. $0$

1694: and $-1$). Thus we have to apply (\ref{twocase}), and

1695: add the darkly shaded boxes with content $0$ and $-1$ to this skew

1696: partition. The complement of this is no longer a partition, so we

1697: remove the remaining darkly shaded region by one further application of

1698: the iterative procedure.

1699: \end{example}

1700: 

1701: \begin{figure}[ht]

1702: \includegraphics{4ex.eps}

1703: \caption{More examples of the $\lambda/\mu^i$

1704: construction}

1705: \label{4ex}

1706: \end{figure} 

1707: 

1708: \begin{defn} \label{maxsub}

1709: We now wish to define a {\em maximal balanced subpartition 

1710: between $\mu$ and $\lambda$}, which we will denote by $\lambda/\mu'$.

1711: Having constructed a skew partition $\lambda/\mu^i$ for each removable

1712: box $\epsilon_i$ of $\lambda$, we partially order this collection by

1713: inclusion. We then take $\lambda/\mu'$ to be some minimal element of

1714: this set. 

1715: \end{defn}

1716: 

1717: \begin{example}To see a non-trivial example of this choice, consider

1718: $\lambda=(7,6^2,5,4^2,2)$ and $\mu=(5,3,2^3,1)$. This is a balanced

1719: pair for $\delta=1$, but has several different associated skew

1720: partitions. If we take $\epsilon_i$ to be one of the removable boxes

1721: labelled by $6$ or $-5$ then $\lambda/\mu^i$ equals the entire shaded

1722: region in Figure \ref{4ex}(b). However, if we take $\epsilon_j$ to be

1723: any of the other removable boxes then $\lambda/\mu^j$ consists of the

1724: six darkly shaded boxes. As $\lambda/\mu^j\subset\lambda/\mu^i$, we

1725: take $\lambda/\mu'$ to equal $\lambda/\mu^j$ in this case, and hence

1726: $\mu'=(7,6,4^2,3,2^2)$. (Note that if this example had one additional

1727: box of content $0$ between the two darkly shaded regions, then we

1728: would have to apply (\ref{onecase}) and this box would have associated

1729: skew partition all of the darkly shaded region together with itself

1730: and the diagonally adjacent box with content $0$.)

1731: \end{example}

1732: 

1733: The importance of this construction is given by

1734: 

1735: \begin{thm}\label{balhom}

1736: If $\mu\subset\lambda$ is a balanced pair, then for any maximal

1737: balanced subpartition $\mu'$ between $\mu$ and $\lambda$  we have

1738: $$\Hom(\Delta_n(\lambda),\Delta_n(\mu'))\neq 0.$$

1739: \end{thm}

1740: 

1741: \begin{proof} As usual, we may assume that $\lambda$ is a partition of $n$.

1742: Pick $\epsilon\in\remo(\lambda/\mu')$ with

1743: $|c(\epsilon)-\frac{1-\delta}{2}|$ maximal. (Note that there are at

1744: most two such boxes.) If $\delta$ is even and

1745: $c(\epsilon)=\frac{-\delta}{2}$ or $c(\epsilon)=\frac{2-\delta}{2}$

1746: then $\lambda/\mu'$ is one of the two cases in Figure \ref{triv}(a) or

1747: (b), while if $\delta$ is odd and $c(\epsilon)=\frac{1-\delta}{2}$

1748: then $\lambda/\mu'$ is as in Figure \ref{triv}(c). In each of these

1749: cases there is a non-zero homomorphism from $\Delta_n(\lambda)$ to

1750: $\Delta_n(\mu')$ by Theorem \ref{rectanglesuff} (or more directly by

1751: repeated applications of Frobenius reciprocity). Thus we henceforth

1752: assume we are not in any of these cases.

1753: 

1754: \begin{figure}[ht]

1755: \includegraphics{triv.eps}

1756: \caption{Some small $\epsilon$ cases, with matched box denoted

1757: by $\epsilon'$}

1758: \label{triv} 

1759: \end{figure} 

1760: 

1761: Suppose that $\epsilon$ is paired with a maximal $\epsilon'$ of

1762: content $1-\delta-c(\epsilon)$. We will assume that $\epsilon$ is

1763: above, or to the right of, $\epsilon'$, and leave the (obvious)

1764: modifications required for the other case to the reader.

1765: 

1766: We will be able to proceed by induction using the following claim.

1767: 

1768: \begin{claim}\label{claim}

1769: (i) There is no box of content $c(\epsilon)$ in $\remo(\mu')$.\\

1770: (ii)There is a unique box $\epsilon'$ of content

1771: $1-\delta-c(\epsilon)$ in $\add(\mu')$.\\ (iii) If $|\lambda/\mu'|>2$

1772: then the pair $\lambda-\epsilon$ and $\mu'+\epsilon'$ is balanced, and

1773: the associated skew partition is minimal in the set of those of the

1774: form $(\lambda-\epsilon)/(\mu+\epsilon')^k$, with $\epsilon_k$ in

1775: $\remo((\lambda-\epsilon)/(\mu'+\epsilon'))$. Equivalently, for every

1776: $\epsilon_k$ in $\remo((\lambda-\epsilon)/(\mu'+\epsilon'))$ we have

1777: $$(\lambda-\epsilon)/(\mu+\epsilon')^k=(\lambda-\epsilon)/(\mu+\epsilon').$$

1778: \end{claim}

1779: 

1780: Before proving this claim, we show how it can be used to complete the

1781: proof of Theorem \ref{balhom}.  Note that if $\lambda-\epsilon$ has

1782: a removable box $\tau$ with content $1-\delta-c(\epsilon)$ then by

1783: minimality $\lambda/\mu'=\{\epsilon,\tau\}$, and we are done

1784: by Theorem \ref{dhwhom} and our assumptions on $\lambda$ . Thus we

1785: assume that there is no such removable box.  By Frobenius reciprocity,

1786: Corollary \ref{better}, and Lemma \ref{nonsp}, we have

1787: $$\begin{array}{ll}\Hom(\Delta_n(\lambda),\Delta_n(\mu'))

1788: &\cong\,\Hom(\pr_{\lambda}\ind_{n-1}\Delta_{n-1}(\lambda-\epsilon),

1789: \Delta_n(\mu'))\\

1790: &\cong\,\Hom(\Delta_{n-1}(\lambda-\epsilon),

1791: \pr_{\lambda-\epsilon}\res_{n}\Delta_n(\mu')).

1792: \end{array}$$

1793: By the first two parts of Claim \ref{claim} this latter Homspace is

1794: isomorphic to 

1795: $$\Hom(\Delta_{n-1}(\lambda-\epsilon),\Delta_{n-1}(\mu'+\epsilon'))$$

1796: and by the final part of Claim \ref{claim} (and induction) this is

1797: non-zero as required.

1798: 

1799: Thus it only remains to prove Claim \ref{claim}.

1800: 

1801: \noindent{\bf Proof of Claim \ref{claim}:}

1802: (i) First suppose that there is only one box in $\lambda/\mu'$ with

1803: content $c(\epsilon)$. By construction, if there are any boxes above

1804: $\epsilon$ in $\lambda/\mu'$ then the one with largest content, or its

1805: matched pair, is removable. But this contradicts the choice of

1806: $\epsilon$.  The other possibility is that there is a second box

1807: $\tau$ in

1808: $\lambda/\mu'$ with content $c(\epsilon)$, occuping the opposite corner

1809: of a two by two square. Arguing as in the previous case, if there are

1810: any boxes in $\lambda/\mu'$ above this square then this contradicts

1811: the choice of $\epsilon$. These two cases are illustrated in Figure

1812: \ref{corner}(a) and (b). In both these cases we deduce that $\mu'$

1813: cannot have a removable box of content $c(\epsilon)$, as there must be

1814: boxes to the right of any such box in $\mu'$.

1815: 

1816: \begin{figure}[ht]

1817: \includegraphics{corner.eps}

1818: \caption{Two corner cases}

1819: \label{corner}

1820: \end{figure} 

1821: 

1822: (ii) Note that if $\lambda/\mu'$ consists of two boxes then the result

1823: is obvious, so we assume this is not the case. It is also clear that

1824: any addable box of a given content must be unique. Let $\epsilon'$ be

1825: the maximal box in $\lambda/\mu'$ with content

1826: $1-\delta-c(\epsilon)$. 

1827: 

1828: First suppose that $\lambda/\mu'$ has only one

1829: box with content $c(\epsilon)$, so that we are in the case shown in

1830: Figure \ref{corner}(a). The box $*'$ paired with $*$ in Figure

1831: \ref{corner}(a) must be to the right or above $\epsilon'$, and hence

1832: we are in one of the two configurations shown in Figure

1833: \ref{cor1}.

1834: 

1835: \begin{figure}[ht]

1836: \includegraphics{cor1.eps}

1837: \caption{The first corner case}

1838: \label{cor1}

1839: \end{figure} 

1840: 

1841: The case in Figure \ref{cor1}(a) is impossible by our assumption on

1842: the size of $\lambda/\mu'$ (and minimality), as both $\epsilon$ and

1843: $\epsilon'$ are removable boxes. In the remaining case it is clear

1844: that $\mu'$ has addable box $\epsilon'$, as required.

1845: 

1846: Next suppose that $\lambda/\mu'$ has two boxes with content

1847: $c(\epsilon)$, so that we are in the case shown in Figure

1848: \ref{corner}(b). As in the previous case, the box $\alpha'$ paired

1849: with $\alpha$ must be to the right or above $\epsilon'$. If it is

1850: above then we have a configuration similar to that in Figure

1851: \ref{cor1}(a), and hence $\epsilon'$ is a removable box. But this is

1852: impossible exactly as for the case in Figure \ref{cor1}(a). Hence

1853: $\alpha'$ must be to the right of $\epsilon'$, and we must have a

1854: configuration as in Figure \ref{cor2}. But this configuration clearly

1855: has an addable box, $\tau'$, of content $c(\epsilon')$.

1856: 

1857: \begin{figure}[ht]

1858: \includegraphics{cor2.eps}

1859: \caption{The second corner case}

1860: \label{cor2}

1861: \end{figure} 

1862: 

1863: (iii) The two partitions $\lambda-\epsilon$ and $\mu'+\epsilon'$ are

1864: clearly balanced. For minimality we consider the various cases that

1865: can arise. If we are in the case shown in Figure \ref{corner}(a), then

1866: paired boxes are as shown in Figure \ref{cor1}(b). Suppose for a

1867: contradiction that $(\lambda-\epsilon)/(\mu'+\epsilon')$ is not

1868: minimal, and hence contains a smaller skew partition

1869: $\eta$. If $\eta$ does not involve $*$ and $*'$ then it is

1870: also contained in $\lambda/\mu'$, which contradicts the minimality of

1871: this original pair. If $\eta$ does involve $*$ and

1872: $*'$ then this contradicts $\lambda/\mu'$ being minimal, as

1873: $\lambda/\mu'$ contains $\eta\cup\{*,*'\}$, which is a smaller

1874: sub-skew partition of $\lambda/\mu'$.

1875: 

1876: Now consider the case shown in Figure \ref{corner}(b), where the

1877: paired boxes are as in Figure \ref{cor2}. As before, suppose for a

1878: contradiction that $(\lambda-\epsilon)/(\mu'+\tau')$ is not minimal,

1879: and hence contains a smaller skew partition $\eta$. If $\eta$ does not

1880: involve $\alpha$ and $\alpha'$ then it is also contained in

1881: $\lambda/\mu'$. If $\eta$ does involve $\alpha$ and $\alpha'$ but not

1882: $\tau$ and $\epsilon'$, then $\eta\cup\{\epsilon,\epsilon'\}$ is a

1883: removable skew inside $\lambda/\mu'$. Finally, if $\eta$ involves all

1884: of $\alpha$, $\alpha'$, $\tau$, and $\epsilon'$, then $\eta$ must also

1885: involve $\beta$ and $\beta'$. Now the skew obtained from $\eta$ by

1886: replacing $\tau$ by $\epsilon$ can be removed from $\lambda/\mu'$. In

1887: each of these three cases we have found a proper removable skew inside

1888: $\lambda/\mu'$, which contradicts the minimality of

1889: $\lambda/\mu'$. Thus $(\lambda-\epsilon)/(\mu'+\tau')$ must be

1890: minimal, which completes the proof of Claim \ref{claim}, and hence

1891: also of Theorem \ref{balhom}.

1892: \end{proof}

1893: 

1894: \begin{cor}\label{blocks}

1895: Two weights $\lambda$ and $\mu$ are in the same block of $B_n$ if and

1896: only if they are balanced. Each block contains a unique minimal weight.

1897: \end{cor}

1898: \begin{proof} In Corollary \ref{better} we proved that two weights in

1899: the same block must be balanced. For the reverse implication, we will

1900: proceed by induction. By Theorem \ref{balhom}, if $\lambda$ contains a

1901: smaller partition $\mu$ with which it is balanced, then there exists

1902: some $\mu'\subset\lambda$ with a non-zero homomorphism from

1903: $\Delta_n(\lambda)$ to $\Delta_n(\mu')$. In particular, $\lambda$ and

1904: $\mu'$ will lie in the same block of $B_n$. Thus it is enough to show

1905: that there is a unique minimal partition in the set of partitions

1906: which are balanced with $\lambda$.

1907: 

1908: But if there are two such minimal partitions $\mu$ and $\nu$, then set

1909: $\eta=\mu\cap\nu$. Clearly $\eta$ is a partition, and it forms a balanced

1910: pair with both $\mu$ and $\nu$ (and hence with $\lambda$). This

1911: contradicts our assumption  of minimality

1912: \end{proof}

1913: 

1914: We conclude this section with a description of the minimal partitions

1915: in each block (and hence give a parametrisation of the blocks). We

1916: begin by constructing inductively a skew partition $\hat{\lambda}$

1917: related to $\lambda$. Let $\lambda(0)=\lambda$. Given $\lambda(i)$,

1918: consider $\epsilon\in\remo(\lambda(i))$ such that

1919: $|c(\epsilon)-\frac{1-\delta}{2}|$ is maximal. Suppose that

1920: there does not exist $\epsilon'\in[\lambda]$ with

1921: $c(\epsilon)+c(\epsilon')=1-\delta$ and $\epsilon'\neq

1922: \epsilon$. Hence either the set of rows $\lambda^t$ above and

1923: including the row containing $\epsilon$ (if

1924: $c(\epsilon)-\frac{1-\delta}{2}>0$) or the set of columns $\lambda^l$

1925: to the the left of and including the column containing $\epsilon$ (if

1926: $c(\epsilon)-\frac{1-\delta}{2}<0$) cannot be removed. In this case

1927: set $\lambda(i+1)=\lambda(i)/\lambda^t$, respectively

1928: $\lambda(i+1)=\lambda(i)/\lambda^l$.  If there exists

1929: $\epsilon'\in[\lambda]$ with $c(\epsilon)+c(\epsilon')=1-\delta$ and

1930: $\epsilon'\neq \epsilon$ then $\hat{\lambda}=\lambda(i)$.  This

1931: procedure will eventually terminate in the construction of

1932: $\hat{\lambda}$.

1933: 

1934: \begin{figure}[ht]

1935: \includegraphics{hat.eps}

1936: \caption{An example of the construction of $\hat{\lambda}$}

1937: \label{hat}

1938: \end{figure} 

1939: 

1940: \begin{example} 

1941: As an example of this construction, consider $\delta=1$ and

1942: $\lambda=(7^2,6,5,4,2,1^2)$, as illustrated in Figure \ref{hat}. At

1943: the first stage, we take $\epsilon$ to be the box labelled $-7$, and

1944: hence remove the first column. Next we take the box labelled $5$, and

1945: remove the first two rows. This is followed by the removal of the

1946: second column, then the third row, leaving the skew partition

1947: illustrated in the figure. As the two remaining removable nodes both

1948: have a paired partner (in this case each other) no more rows or

1949: columns need be removed, and we have constructed $\hat{\lambda}$.

1950: \end{example}

1951: 

1952: 

1953: \begin{prop}

1954: The minimal partitions in each block are precisely those for which

1955: either $\hat{\lambda}=\emptyset$ or a single row or column, or

1956: $\delta$ is even and $\hat{\lambda}$ consists of two rows, the second

1957: of which has final box of content $-\frac{\delta}{2}$.

1958: \end{prop}

1959: \begin{proof}

1960: Clearly if $\hat{\lambda}=\emptyset$ then $\lambda$ is minimal in its

1961: block. In the remaining cases, removal of any part of $\lambda$ can

1962: only involve boxes in $\hat{\lambda}$, and hence to be balanced must

1963: involve either a single unpaired box of content $\frac{1-\delta}{2}$

1964: or a single vertical pair in the configuration shown in Figure

1965: \ref{diag}(b). But this is impossible. Hence we assume that

1966: $\hat{\lambda}$ is not of the form given in the proposition, and will show

1967: that $\lambda$ is not minimal.

1968: 

1969: First suppose that $\delta$ is odd. If $\hat{\lambda}$ contains two

1970: boxes of content $\frac{1-\delta}{2}$ then we can construct a maximal

1971: balanced subpartition of $\lambda$, mimicking the process in

1972: Definitions \ref{maxisub} and \ref{maxsub} by starting with

1973: $\epsilon$. Hence by Theorem \ref{balhom} $\lambda$ is non

1974: minimal. If $\hat{\lambda}$ only contains one box $\omega$ with

1975: content $\frac{1-\delta}{2}$ then, again by considering Definitions

1976: \ref{maxisub} and \ref{maxsub} and Theorem \ref{balhom}, any removable

1977: balanced skew-partition must involve $\omega$. The assumption also implies that

1978: $\epsilon$ is in the first row or column of $\hat{\lambda}$.

1979: 

1980: Suppose that $\epsilon$ is in the first row of $\hat{\lambda}$ and

1981: there is more than one row (the case where $\epsilon$ is in the first

1982: column is similar). If $\lambda$ is minimal, then no final segment of

1983: this row has a removable paired segment in $\hat{\lambda}$; this can

1984: only arise if $\hat{\lambda}$ is of the form show in Figure

1985: \ref{oddcase} (where shaded areas indicate boxes definitely not in

1986: $\hat{\lambda}$), where $\tau$ is not paired with any box to the right

1987: of $\omega$. But this means that $\tau$ has content

1988: $\frac{1-\delta}{2}$ which is impossible, and hence $\lambda$ is not

1989: minimal.

1990: 

1991: \begin{figure}[ht]

1992: \includegraphics{oddcase.eps}

1993: \caption{Possible configuration of $\hat{\lambda}$ when

1994: $\delta$ is odd}

1995: \label{oddcase}

1996: \end{figure} 

1997: 

1998: Now suppose that $\delta$ is even. If $\hat{\lambda}$ contains either

1999: of the configurations shown in Figure \ref{nogo}(a) and (b) then we

2000: can again construct a maximal balanced subpartition, and by Theorem

2001: \ref{balhom} $\lambda$ is not minimal.

2002: 

2003: \begin{figure}[ht]

2004: \includegraphics{nogo.eps}

2005: \caption{Possible configurations in $\hat{\lambda}$ when

2006: $\delta$ is even}

2007: \label{nogo}

2008: \end{figure} 

2009: 

2010: If $\hat{\lambda}$ contains only one box with content either

2011: $-\frac{\delta}{2}$ or $1-\frac{\delta}{2}$ then this box is either at

2012: the end of the first row or bottom of the first column, which

2013: contradicts the definition of $\hat{\lambda}$. Thus we must have one of

2014: the configurations in Figure \ref{nogo}(c) or (d).

2015: 

2016: In case (c) $\epsilon$ must lie at the end of the first column, and in

2017: case (d) at the end of the first row. Arguing as in the $\delta$ odd

2018: case, we see in case (c) that if $\lambda$ is minimal then

2019: $\hat{\lambda}$ must consist of a single column. However, in case (d),

2020: if $\lambda$ is minimal then we either have a single row or we are in

2021: a similar situation to that in Figure \ref{oddcase} and $\tau$ must

2022: have content $-\frac{\delta}{2}$. But this implies that

2023: $\hat{\lambda}$ consists of two rows with the final box having of the

2024: second having content $-\frac{\delta}{2}$, which contradicts our

2025: assumptions on $\lambda$.

2026: 

2027: Thus the only cases where $\lambda$ is a minimal partition are those

2028: described in the theorem, and so we are done.

2029: \end{proof}

2030: 

2031: \begin{example} To illustrate the last result, consider $\delta=1$

2032:   with $\lambda=(7,6^2,5,2^2)$ as shown in Figure \ref{hat2}. The

2033:   associated $\hat{\lambda}$ is also shown, and has only one row,

2034:   and it is easy to see that $\lambda$ is indeed minimal inside its block.

2035: \end{example}

2036: 

2037: 

2038: \begin{figure}[ht]

2039: \includegraphics{hat2.eps}

2040: \caption{A minimal weight $\lambda$ and the associated

2041: $\hat{\lambda}$}

2042: \label{hat2}

2043: \end{figure} 

2044: 

2045: 

2046: \section{On the submodule structure of certain standard modules\label{bigmod}}

2047:  

2048: In this section we will show that the structure of standard modules

2049: can become arbitrarily complicated (as measured by their Loewy length

2050: and number of simples in each Loewy layer). For this it will be

2051: sufficient to consider certain special partitions which can be more

2052: easily analysed.

2053: 

2054: 

2055: 

2056: 

2057: \begin{lem}\label{spotl}

2058: If $\epsilon_i\in\remo(\lambda)$ then 

2059: $$[\res_nL_n(\lambda):L_{n-1}(\lambda-\epsilon_i)]\neq 0.$$

2060: \end{lem}

2061: \begin{proof} By (\ref{makestds}) and (\ref{Lres}) we may assume

2062: that $\lambda\vdash n$; the result then follows from Proposition

2063: \ref{indres}.

2064: \end{proof}

2065: 

2066:  When considering a multi-skew-partition of

2067: differences these skew partitions will be listed in the  order

2068: from top right to bottom left. We will extend the power

2069: notation for partitions to multi{\it partitions}, so $((2)^2,(21^3))$ will

2070: denote the triple of partitions $(2)$, $(2)$, and $(21^3)$.

2071: 

2072: \begin{example}

2073: To illustrate these definitions we return to the partitions $\lambda$

2074: and $\mu$ considered in Figure \ref{whybetter}. In this case we have

2075: $\add(\lambda)=\{-5,-3,-1,3,6\}$ and

2076: $\remo(\lambda)=\{-4,-2,1,5\}$. Similarly $\add(\mu)=\{-3,1,5\}$ and

2077: $\remo(\mu)=\{-1,4\}$. The pair $(\lambda,\mu)$ is not

2078: $\delta$-balanced for any $\delta$, and $\lambda/(\lambda\cap\mu)$ has

2079: shape $((1),(2^2),(2,1))$.

2080: \end{example}

2081: 

2082: We will be interested in $\delta$-balanced pairs $\mu\subset\lambda$

2083: such that the associated skew partition consists entirely of isolated

2084: boxes.  If $\mu\subset\lambda$

2085: are balanced with $\lambda/\mu=((1)^{2m})$, denote the matched pairs

2086: of boxes in $\lambda/\mu$ by $\epsilon_1, \epsilon_1', \ldots,

2087: \epsilon_m, \epsilon_m'$ with respective contents $a_1, a_1',

2088: \ldots, a_m, a_m'$. Let ${\pow}(m)$ denote the power set of

2089: $\{1,2,\ldots,m\}$, and for $x\in{\pow}(m)$ set

2090: $$\lambda-x=\lambda-\sum_{i\in x}(\epsilon_i+\epsilon_i').$$

2091: For example, $\lambda-\{1,\ldots,m\}=\mu$.

2092: 

2093: \begin{thm}\label{lattice}

2094: Let $\lambda\vdash n$ and $\mu\subset\lambda$ be a balanced pair with

2095: $\lambda/\mu=((1)^{2m})$. Then 

2096: $$\dim\Hom(\Delta_n(\lambda),\Delta_n(\mu))=1$$ and

2097: $$[\Delta_n(\mu):L_n(\lambda-x)]=1$$ for all $x\in{\pow}(m)$.

2098: 

2099: Further, denote by $\lat(\mu,\lambda)$ the induced lattice in the full

2100: submodule lattice of $\Delta_n(\mu)$ with vertices those simple

2101: modules of the form $L_n(\lambda-x)$ for some $x\in\pow(m)$. Then

2102: $\lat(\mu,\lambda)$ is isomorphic to the superset lattice on

2103: $\pow(m)$; i.e. every submodule of $\Delta_n(\mu)$ which contains

2104: $L_n(\lambda-x)$ contains $L_n(\lambda-y)$ for all $y\subset x$. 

2105: 

2106: In particular the length of the socle series of $\Delta_n(\mu)$ is at

2107: least $m+1$ and there is a socle series layer containing at least $m$ simple

2108: modules.

2109: \end{thm}

2110: 

2111: \begin{rem}

2112: (i) Note that for the  induced lattice we are only considering factors of the

2113: form $L_n(\lambda-x)$. In general the module $\Delta_n(\mu)$ will

2114: have many other composition factors. Thus an arrow $A\rightarrow B$ in

2115: our induced lattice structure is to be understood as representing some

2116: non-trivial extension in $\Delta_n(\mu)$ with $A$ in the head and $B$ in 

2117: the socle.\\

2118: (ii) Clearly the final part of

2119: the theorem can be strengthened, but is already enough to show

2120: that standard modules can have arbitrarily large socle series lengths

2121: (and layers of arbitrary width).

2122: \end{rem}

2123: 

2124: \begin{example} If $\lambda$ and $\mu$ are balanced with 

2125: $\lambda/\mu=((1)^{6})=

2126: \{\epsilon_1,\epsilon_1',\epsilon_2,\epsilon_2',\epsilon_3,\epsilon_3'\}$

2127: then the lattice $\lat(\mu,\lambda)$ is illustrated in Figure \ref{latis}.

2128: \end{example}

2129: 

2130: \begin{figure}

2131: $$\xymatrix{ & {L_n(\mu)} \ar[dl] \ar[d] \ar[dr]

2132: \\ {L_n(\mu+\epsilon_1+\epsilon_1')} 

2133: \ar[d] \ar[dr] & {L_n(\mu+\epsilon_2+\epsilon_2')} \ar[dl]

2134: \ar[dr] & {L_n(\mu+\epsilon_3+\epsilon_3')} \ar[dl] \ar[d] \\

2135: {L_n(\lambda-\epsilon_3-\epsilon_3')} 

2136: \ar[dr] & {L_n(\lambda-\epsilon_2-\epsilon_2')} \ar[d] &

2137: {L_n(\lambda-\epsilon_1-\epsilon_1')} \ar[dl] \\ & L_n(\lambda) }$$

2138: \caption{An example of $\lat(\mu,\lambda)$}

2139: \label{latis}

2140: \end{figure}

2141: 

2142: \begin{proof}

2143: We proceed by induction on $m$, the result being obvious for $m=0$. By

2144: Frobenius reciprocity we have

2145: \begin{equation}\label{indstep}

2146: \Hom(\ind_{n-1}\Delta_{n-1}(\lambda-\epsilon_i),\Delta_n(\mu)) \cong

2147: \Hom(\Delta_{n-1}(\lambda-\epsilon_i),\res_{n}\Delta_n(\mu)).

2148: \end{equation}

2149: By Proposition \ref{indres} and Corollary \ref{better}, the only

2150: submodule of $\res_n\Delta_n(\mu)$ which can lie in the same block as

2151: $\Delta_{n-1}(\lambda-\epsilon_i)$ is isomorphic to

2152: $\Delta_{n-1}(\mu+\epsilon_i')$, and hence by the inductive hypothesis

2153: the right-hand side of (\ref{indstep}) is one dimensional. Lemma

2154: \ref{nonsp} now implies that $L_n(\lambda)$ is a composition factor of

2155: $\Delta_n(\mu)$. To show that

2156: $\dim\Hom(\Delta_n(\lambda),\Delta_n(\mu))=1$ it will be enough to

2157: show that there is precisely one copy of this composition factor in

2158: $\Delta_n(\mu)$ (which will necessarily lie in the socle).

2159: 

2160:  By assumption the pair

2161: $(\lambda,\mu)$ is balanced. We will define the {\it bias} of a pair

2162: $(\lambda,\tau)$ with $|\lambda\vartriangle\tau|=2t$ to be 

2163: $$b(\lambda,\tau)= \left(\sum_{d\in

2164: \lambda\vartriangle\tau}c(d)\right)-t(1-\delta).$$ Thus a balanced

2165: pair has zero bias. Consider the restriction $\res_n\Delta_n(\mu)$. By

2166: Proposition \ref{indres} we have a short exact sequence

2167: \begin{equation}\label{resbias}

2168: 0\rightarrow \bigoplus_{\tau \lhd \mu} \Delta_{n-1}(\tau)

2169: \rightarrow \res_n\, \Delta_n(\mu) \rightarrow \bigoplus_{\tau \rhd

2170: \mu} \Delta_{n-1}(\tau)\rightarrow 0.

2171: \end{equation}

2172: Note that $\mu$ has no removable boxes with content $\pm a_i$ for

2173: $1\leq i\leq m$, as this would contradict the existence of an addable

2174: node with such a content. Thus the only modules $\Delta_{n-1}(\tau)$

2175: in the sequence (\ref{resbias}) with bias $\pm a_i$ are

2176: $\Delta_{n-1}(\mu+\epsilon_i)$ and $\Delta_{n-1}(\mu+\epsilon_i')$

2177: 

2178: 

2179: By Lemma \ref{spotl} we have that 

2180: $$[\res_{n}L_n(\lambda-x):L_{n-1}(\lambda-x-\epsilon_i)]=1$$ provided

2181: that $i\notin x$. But (by the observations on bias above)

2182: $L_{n-1}(\lambda-x-\epsilon_i)$ can only occur in

2183: $\Delta_{n-1}(\mu+\epsilon_i')$, and by the inductive hypothesis it

2184: occurs there precisely once. By varying $i$ we deduce that there is at

2185: most one copy of each $L_n(\lambda-x)$ in $\Delta_n(\mu)$. But by

2186: induction we know that there is a homomorphism from

2187: $\Delta_{n'}(\lambda-x)$ to $\Delta_{n'}(\mu)$ where $n'=|\lambda-x|$, and

2188: hence by repeated applications of $G$ that there is a homomorphism

2189: from $\Delta_{n}(\lambda-x)$ to $\Delta_{n}(\mu)$. Hence we see that

2190: $L_n(\lambda-x)$ occurs exactly once in $\Delta_n(\mu)$.

2191: 

2192: Now consider the summand $\Delta_{n-1}(\mu+\epsilon_i')$ in

2193: $\res_n\Delta_n(\mu)$. This is the only summand of the restriction in

2194: which $L_{n-1}(\lambda-x-\epsilon_i)$ (with $i\notin x$) can arise,

2195: and this simple appears in an extension below

2196: $L_{n-1}(\lambda-y-\epsilon_i)$ for all $y\supset x$ (with $i\notin

2197: y$), by the inductive hypothesis. In particular the copy of

2198: $L_{n-1}(\lambda-x-\epsilon_i)$ appearing in $\res_nL_n(\lambda-x)$

2199: appears below $L_{n-1}(\lambda-x-\epsilon_i-\epsilon_j-\epsilon_j')$

2200: in an extension, and this latter simple must come from

2201: $\res_nL_n(\lambda-x-\epsilon_j-\epsilon_j')$. It follows that

2202: $L_n(\lambda-x)$ must occur in some extension beneath

2203: $L_n(\lambda-x-\epsilon_j-\epsilon_j')$. This argument works for all

2204: $j$ and $x$, and hence verifies the claimed submodule structure except

2205: for the top two layers. However, these are forced by the structure of

2206: standard modules.

2207: \end{proof}

2208: 

2209: \section{\label{last}The case $\delta=0$}

2210: 

2211: In this section we will sketch the modifications to the preceding

2212: arguments which are required when $\delta=0$. The most obvious change

2213: is that the idempotents $e_n$ considered thus far no longer exist. This

2214: is easily remedied --- however a more serious complication is the

2215: failure of the algebras to be quasihereditary when $n$ is even.

2216: 

2217: \begin{figure}[ht]

2218: \includegraphics{alte.eps}

2219: \caption{The element $\bar{e}_n$ in $B_n$}

2220: \label{alte}

2221: \end{figure} 

2222: 

2223: For $n\geq 3$ let $\bar{e}_n$ be the element illustrated in Figure

2224: \ref{alte}. This is an idempotent for every value of $\delta$, and

2225: satisfies (A1), i.e.

2226: $$\bar{e}_nB_n\bar{e}_n\cong B_{n-2}.$$ Unfortunately we can no longer

2227: prove an analogue of (A2) in general, as the algebras are not

2228: quasihereditary. If $n$ is odd then there are no problems, and the

2229: arguments in the $\delta\neq0$ case for (A1-6) go through unchanged.

2230: The results in Sections 4-7 also generalise, as the various results

2231: needed from \cite{dhw} include the case $\delta=0$, and we thus

2232: deduce the block result in this case.

2233: 

2234: For $n$ even, we can no longer appeal directly to the general machinery in

2235: \cite{cmpx}. However, the algebras in this case are cellular, and the

2236: modules considered by \cite{dhw} are precisely the cell modules for

2237: these algebras. The necessary results coming from the general theory

2238: in \cite{cmpx} now have to be verified on an {\it ad hoc} basis, but

2239: this has been carried out in \cite{dhw}.  Thus, again, the results in

2240: Sections 4-7 go through unchanged (noting that it is enough to analyse

2241: cell modules when determining blocks by \cite[(3.9.8)]{gl} (see

2242: \cite[2.22 Corollary]{mathas})).

2243: 

2244: 

2245: \providecommand{\bysame}{\leavevmode\hbox to3em{\hrulefill}\thinspace}

2246: \providecommand{\MR}{\relax\ifhmode\unskip\space\fi MR }

2247: % \MRhref is called by the amsart/book/proc definition of \MR.

2248: \providecommand{\MRhref}[2]{%

2249:   \href{http://www.ams.org/mathscinet-getitem?mr=#1}{#2}

2250: }

2251: \providecommand{\href}[2]{#2}

2252: \begin{thebibliography}{DWH99}

2253: 

2254: \bibitem[Bra37]{brauer}

2255: R.~Brauer, \emph{On algebras which are connected with the semisimple continuous

2256:   groups}, Ann. of Math. \textbf{38} (1937), 857--872.

2257: 

2258: \bibitem[Bro55]{brownbrauer}

2259: W.~Brown, \emph{An algebra related to the orthogonal group}, Michigan Math. J.

2260:   \textbf{3} (1955), 1--22.

2261: 

2262: \bibitem[CMPX]{cmpx}

2263: A.~G. Cox, P.~P. Martin, A.~E. Parker, and C.~Xi, \emph{Representation theory

2264:   of towers of recollement: theory, notes, and examples}, J. Algebra, to

2265:   appear.

2266: 

2267: \bibitem[DDH]{ddh}

2268: R.~Dipper, S.~Doty, and J.~Hu, \emph{Brauer's centralizer algebras, symplectic

2269:   {S}chur algebras and {Schur-Weyl} duality}, preprint.

2270: 

2271: \bibitem[Dia]{Diaconis}

2272: P.~Diaconis, \emph{Group representations in probability and statistics},

2273:   Institute of Mathematical Statistics Lecture Notes, Monograph Series 11.

2274: 

2275: \bibitem[Don98]{don2}

2276: S.~Donkin, \emph{{The $q$-Schur} algebra}, LMS Lecture Notes Series, vol. 253,

2277:   Cambridge University Press, 1998.

2278: 

2279: \bibitem[Dot98]{dotpoly1}

2280: S.~Doty, \emph{Polynomial representations, algebraic monoids, and {S}chur

2281:   algebras of classical type}, J. Pure Appl. Algebra \textbf{123} (1998),

2282:   165--199.

2283: 

2284: \bibitem[DWH99]{dhw}

2285: W.~F. Doran, D.~B. Wales, and P.~J. Hanlon, \emph{On the semisimplicity of the

2286:   {B}rauer centralizer algebras}, J. Algebra \textbf{211} (1999), 647--685.

2287: 

2288: \bibitem[Ful97]{fultab}

2289: W.~Fulton, \emph{Young tableaux}, LMS Student Texts, vol.~35, Cambridge, 1997.

2290: 

2291: \bibitem[GL96]{gl}

2292: J.~J. Graham and G.~I. Lehrer, \emph{Cellular algebras}, Invent. Math.

2293:   \textbf{123} (1996), 1--34.

2294: 

2295: \bibitem[Gre80]{green}

2296: J.~A. Green, \emph{Polynomial representations of {GL$_n$}}, Lecture Notes in

2297:   Mathematics 830, Springer, 1980.

2298: 

2299: \bibitem[HP]{hpbrauer}

2300: R.~Hartmann and R.~Paget, \emph{{Y}oung modules and filtration multiplicities

2301:   for {B}rauer algebras}, preprint.

2302: 

2303: \bibitem[HW89a]{hw2}

2304: P.~J. Hanlon and D.~B. Wales, \emph{Eigenvalues connected with {B}rauer's

2305:   centralizer algebras}, J. Algebra \textbf{121} (1989), 446--476.

2306: 

2307: \bibitem[HW89b]{hw1}

2308: \bysame, \emph{On the decomposition of {B}rauer's centralizer algebras}, J.

2309:   Algebra \textbf{121} (1989), 409--445.

2310: 

2311: \bibitem[HW90]{hw3}

2312: \bysame, \emph{Computing the discriminants of {B}rauer's centralizer algebras},

2313:   Math. Comp. \textbf{54} (1990), 771--796.

2314: 

2315: \bibitem[HW94]{hw4}

2316: \bysame, \emph{A tower construction for the radical in {B}rauer's centralizer

2317:   algebras}, J. Algebra \textbf{164} (1994), 773--830.

2318: 

2319: \bibitem[JK81]{jk}

2320: G.~D. James and A.~Kerber, \emph{The representation theory of the {S}ymmetric

2321:   group}, Encyclopedia of Mathematics and its Applications, vol.~16,

2322:   Addison-Wesley, 1981.

2323: 

2324: \bibitem[Mar91]{marbook}

2325: P.~P. Martin, \emph{Potts models and related problems in statistical

2326:   mechanics}, World Scientific, Singapore, 1991.

2327: 

2328: \bibitem[Mar96]{mar1}

2329: \bysame, \emph{The structure of the {P}artition algebras}, J. Algebra

2330:   \textbf{183} (1996), 319--358.

2331: 

2332: \bibitem[Mat99]{mathas}

2333: A.~Mathas, \emph{Iwahori-{Hecke algebras and Schur algebras of the symmetric

2334:   groups}}, University lecture series, vol.~15, American Mathematical Society,

2335:   1999.

2336: 

2337: \bibitem[MRH04]{mryom}

2338: P.~P. Martin and S.~Ryom-Hansen, \emph{Virtual algebraic {L}ie theory: {Tilting

2339:   modules and Ri}ngel duals for blob algebras}, Proc. LMS \textbf{89} (2004),

2340:   655--675.

2341: 

2342: \bibitem[MW03]{mwgen}

2343: P.~P. Martin and D.~Woodcock, \emph{Generalized blob algebras and alcove

2344:   geometry}, LMS J. of Comp. and Math. \textbf{6} (2003), 249--296.

2345: 

2346: \bibitem[Naz96]{nazbrauer}

2347: M.~Nazarov, \emph{{Young's orthogonal form for Brauer's centralizer algebra}},

2348:   J. Algebra \textbf{182} (1996), 664--693.

2349: 

2350: \bibitem[Oeh01]{oe1}

2351: S.~Oehms, \emph{Centralizer coalgebras, {FRT}-construction and symplectic

2352:   monoids}, J. Algebra \textbf{244} (2001), 19--44.

2353: 

2354: \bibitem[OR01]{orram}

2355: R.~Orellana and A.~Ram, \emph{Affine braids, markov traces and the category

2356:   {$\mathcal O$}}, preprint, 2001.

2357: 

2358: \bibitem[Rui05]{ruibrauer}

2359: H.~Rui, \emph{A criterion on the semisimple {B}rauer algebras}, J. Comb. Theory

2360:   Ser. A \textbf{111} (2005), 78--88.

2361: 

2362: \bibitem[Wen88]{wenzlbrauer}

2363: H.~Wenzl, \emph{On the structure of {B}rauer's centralizer algebra}, Ann. Math.

2364:   \textbf{128} (1988), 173--193.

2365: 

2366: \bibitem[Wey46]{weyl}

2367: H.~Weyl, \emph{The classical groups, their invariants and representations},

2368:   Priceton University Press, 1946.

2369: 

2370: \end{thebibliography}

2371: 

2372: \end{document}

2373: 

2374: