1: 

2: \documentclass[dvips]{amsart}

3: \usepackage[dvips]{graphics}

4: \usepackage{amscd,amssymb,amsxtra,latexsym,pictex,epsfig,psfrag}

5: %\usepackage[notcite,notref]{showkeys}

6: %\input mac

7: \newcommand{\vanish}[1]{}

8: \newcommand{\F}{\mathcal{F}}

9: \newcommand{\G}{\mathcal{G}}

10: \newcommand{\A}{\mathcal{A}}

11: \newcommand{\B}{\mathcal{B}}

12: \renewcommand{\L}{\mathcal{L}}

13: \newcommand{\M}{\mathcal{M}}

14: \newcommand{\C}{\mathcal{C}}

15: %\newcommand{\D}{\mathcal{D}}

16: \newcommand{\T}{\mathcal{S}}

17: \newcommand{\D}{\Delta}

18: \newcommand{\R}{\mathbb{R}}

19: \newcommand{\Q}{\mathbb{Q}}

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

21: \newcommand{\Z}{\mathbb{Z}}

22: \newcommand{\fq}{\mathbb{F}_q}

23: \newcommand{\abs}[1]{\lvert#1\rvert}

24: \newcommand{\norm}[1]{\lVert#1\rVert}

25: \newcommand{\zerohat}{\hat{0}}

26: \newcommand{\onehat}{\hat{1}}

27: \newcommand{\id}{\operatorname{id}}

28: \newcommand{\supp}{\operatorname{supp}}

29: \newcommand{\rank}{\operatorname{rank}}

30: \newcommand{\Span}{\operatorname{span}}

31: \newcommand{\pointsat}{\mathrel{\to\mkern-10mu-}}

32: \newcommand{\gl}{\operatorname{GL}}

33: \newcommand{\des}{\operatorname{des}}

34: \newcommand{\degg}{\operatorname{deg}}

35: \newcommand{\sign}{\operatorname{sign}}

36: \newcommand{\Aut}{\operatorname{Aut}}

37: \newcommand{\Der}{\operatorname{Der}}

38: \newcommand{\Des}{\operatorname{Des}}

39: \newcommand{\type}{\operatorname{type}}

40: \newcommand{\End}{\operatorname{End}}

41: \newcommand{\inv}{\operatorname{inv}}

42: \newcommand{\Stab}{\operatorname{Stab}}

43: \newcommand{\onto}{\twoheadrightarrow}

44: \newcommand{\into}{\hookrightarrow}

45: \newcommand{\iso}{\cong}

46: \newcommand{\yright}{\stackrel{y}{\longrightarrow}}

47: \newcommand{\yleft}{\stackrel{y}{\longleftarrow}}

48: \newcommand{\isoto}{\stackrel{\cong}{\longrightarrow}}

49: \newcommand{\app}{\stackrel{(R2)}{\rightsquigarrow}}

50: \newcommand{\fsp}{{{\mathfrak{sp}}(2n)}}

51: \newcommand{\J}{\mathcal{J}}

52: \newcommand{\qbinom}[2]{\genfrac{[}{]}{0pt}{}{#1}{#2}}

53: %\newcommand{\lhor}{\rule[.7ex]{.2in}{.3pt}}%horiz. line for lattice

54: \newcommand{\lhor}{\rule[.7ex]{.2in}{.2pt}}%horiz. line for lattice

55: %\newsavebox{\lhorbox}

56: %\savebox{\lhorbox}{$\relbar\joinrel\relbar\joinrel\relbar$}

57: %\newcommand{\lhor}{\usebox{\lhorbox}}

58: 

59: %\newcommand{\kid}{\text{\textcircled{x}}}

60: %\newcommand{\nokid}{\text{\textcircled{}}}

61: 

62: \theoremstyle{plain}

63: \newtheorem*{corollary}{Corollary}

64: \newtheorem{lemma}{Lemma}

65: \newtheorem{theorem}{Theorem}

66: \newtheorem*{theorem*}{Theorem}

67: \newtheorem*{main}{Main theorem}

68: \newtheorem*{reduction}{Reduction lemma}

69: \newtheorem{proposition}{Proposition}

70: \theoremstyle{definition}

71: \newtheorem{definition}{Definition}

72: \newtheorem*{intuition}{Intuition}

73: 

74: \theoremstyle{remark}

75: \newtheorem*{remark}{Remark}

76: \newtheorem{example}{Example}

77: 

78: %%% Abbreviations for references%%%%%%%%%%%

79: \newcommand{\adams}{adams78:_infin}

80: \newcommand{\bm}{bar-natan01:_graph}

81: \newcommand{\bl}{bocklandt:_neckl_lie}

82: \newcommand{\bll}{bergeron98:_combin}

83: \newcommand{\boothby}{boothby86:_rieman}

84: \newcommand{\kbrown}{brown94:_cohom}

85: \newcommand{\cv}{conant:_infin_operat_graph_graph_homol}

86: \newcommand{\drinfeld}{drinfelprimed90:_hopf_gal_q_q}

87: \newcommand{\frt}{faddeev88:_quant_lie_lie}

88: \newcommand{\fh}{fulton91:_repres}

89: \newcommand{\fuks}{fuks86:_cohom_lie}

90: \newcommand{\gelfand}{gelprimefand70:_cohom_lie}

91: \newcommand{\ferenc}{gerlits02:_calcul}

92: \newcommand{\getzler}{getzler94:_batal_vilkov}

93: \newcommand{\getzkap}{getzler98:_modul}

94: \newcommand{\getzlerkap}{getzler95:_cyclic}

95: \newcommand{\ginz}{ginzburg01:_non}

96: \newcommand{\gk}{ginzburg94:_koszul}

97: \newcommand{\jacobson}{jacobson79:_lie}

98: \newcommand{\joyal}{joyal81:_une}

99: \newcommand{\kapranov}{kapranov}

100: \newcommand{\karoubi}{karoubi95:_formes}

101: \newcommand{\kassel}{kassel95:_quant}

102: \newcommand{\kont}{kontsevich94:_feynm}

103: \newcommand{\god}{kontsevich93:_formal}

104: \newcommand{\lazard}{lazard95:_lois}

105: \newcommand{\lsv}{loday97:_operad}

106: \newcommand{\loday}{loday98:_cyclic}

107: \newcommand{\lodayo}{loday97:_overv_leibn}

108: \newcommand{\lodayetal}{loday01:_dialg}

109: \newcommand{\dear}{mahajan01:_sympl}

110: \newcommand{\me}{mahajan01:_quant}

111: \newcommand{\manin}{manin92:_notes_rham}

112: \newcommand{\markl}{markl99:_cyclic}

113: \newcommand{\may}{may72}

114: \newcommand{\mccleary}{mccleary01}

115: \newcommand{\mcduff}{mcduff98:_introd}

116: \newcommand{\mm}{milnor65:_hopf}

117: \newcommand{\conj}{morita99:_struc}

118: \newcommand{\morita}{morita93:_abelian}

119: \newcommand{\spivak}{spivak79}

120: \newcommand{\stasheff}{stasheff63:_homot_h}

121: \newcommand{\sweedler}{sweedler69:_hopf}

122: \newcommand{\thurston}{thurston:_integ_expres_vassil_knot_invar}

123: \newcommand{\voronov}{voronov:_notes}

124: \newcommand{\weibel}{weibel94}

125: \newcommand{\wz}{wess90:_covar}

126: 

127: \hyphenation{mul-ti-pli-ci-ties}

128: 

129: \begin{document}

130: \title{Symplectic operad geometry and graph homology} 

131: \author{Swapneel Mahajan}

132: 

133: 

134: \address{Department of Mathematics\\

135: Cornell University\\

136: Ithaca, NY 14853}

137: \email{swapneel@math.cornell.edu}

138: %\date{\today}

139: %\date{September 20, 1999}

140: 

141: 

142: %\thanks{Preliminary version; comments and suggestions are welcome.}

143: \footnote{

144: 2000 \emph{Mathematics Subject Classification.}

145: Primary 18D50, 17B65, 05C15;

146: Secondary 53D55.

147:  

148: \emph{Key words and phrases.}

149: species; operads; symplectic geometry; differential forms;

150: graph (co)homology; deformation theory.}

151: 

152: \begin{abstract}

153: A theorem of Kontsevich relates the homology 

154: of certain infinite dimensional Lie algebras

155: to graph homology.

156: We formulate this theorem using the language 

157: of reversible operads and mated species.

158: All ideas are explained using a pictorial calculus

159: of cuttings and matings.

160: The Lie algebras are constructed as 

161: Hamiltonian functions on a symplectic operad manifold.

162: And graph complexes are defined for any mated species.

163: The general formulation gives us many examples

164: including a graph homology for groups.

165: We also speculate on the role of deformation theory

166: for operads in this setting.

167: \end{abstract}

168: 

169: \maketitle

170: 

171: %\input o

172: \section{Introduction}  

173: 

174: This paper is my humble tribute to the genius of Maxim Kontsevich.

175: Needless to say, the credit for any new ideas that occur here

176: goes to him, and not me.

177: For how I got involved in this wonderful project,

178: see the historical note (\ref{ss:hn}).

179: 

180: In the papers~\cite{\god,\kont},

181: Kontsevich defined three Lie algebras and

182: related their homology with classical invariants,

183: including the homology of 

184: the group of outer automorphisms of a free group and

185: mapping class groups.

186: He showed this by reducing the homology computation

187: in each case to three graph complexes.

188: His main theorem can be informally stated as

189: 

190: \begin{theorem*}

191: $H_*(\text{Lie algebra}) = H_*(\text{Graph complex}) = 

192: H_*(\text{Group}).$

193: \end{theorem*}

194: 

195: In this paper, we consider only the first part of this theorem.

196: The definitions of the Lie algebras are motivated by

197: classical symplectic geometry.

198: Kontsevich considered three worlds - commutative, associative 

199: and Lie.

200: He developed formal analogues of classical symplectic geometry

201: (which is the commutative case)

202: for the associative and Lie worlds.

203: Each of the three Lie algebras was then defined as

204: Hamiltonian functions on a ``symplectic non-commutative manifold''

205: with the bracket being the analogue of the usual Poisson bracket.

206: The symplectic Lie algebra $\fsp$ was a Lie subalgebra

207: of all three Lie algebras.

208: 

209: In the commutative case, Kontsevich defined a chain complex using graphs.

210: The homology of this chain complex is known as

211: graph homology.

212: He also gave an analogous definition in the associative case

213: using ribbon graphs

214: and in the Lie case using more complicated graphs,

215: which one may call Lie graphs.

216: 

217: The connection between an algebraic or geometric object

218: like a Lie algebra on one side and 

219: a combinatorial object like a graph on the other side

220: is the content of Kontsevich's theorem.

221: The main tool in proving this connection is to use

222: classical invariant theory of $\fsp$.

223: 

224: \subsection{The goal of this paper}

225: From the unified way in which Kontsevich treated the three cases,

226: one expected them to be a part of a more general theory.

227: In fact, in \cite{\god}, Kontsevich made the following statement.

228: \begin{quotation}

229: Our formalism could be extended to the case of

230: ``Koszul dual pairs of quadratic operads''

231: (see \cite{\gk}) including Poisson algebras and,

232: probably, operator algebras etc.

233: \end{quotation}

234: This paper arose in an attempt to understand this statement.

235: In mathematics, just as we have the notion of a 

236: theorem, proposition, corollary, etc, 

237: so should we have the notion of a ``Kontsevich sentence'',

238: the above being a classic example.

239: 

240: Earlier we used the term ``world'' to stand for

241: one of the three words - commutative, associative 

242: and Lie.

243: A possible mathematical substitute for this term is

244: the word ``operad''.

245: It is not possible to develop an analogue of symplectic geometry

246: for every operad.

247: However, one can do so if the operad is ``reversible''.

248: The commutative, associative and Lie operads are three examples.

249: The notion of a reversible operad 

250: was developed independently by Getzler and Kapranov

251: under the name of cyclic operad~\cite{\getzlerkap}.

252: For some information on the history,

253: see the discussion in (\ref{ss:hn}).

254: 

255: The goal of this paper is to formulate 

256: the first part of Kontsevich's theorem

257: in what appears to be its most natural setting -

258: the world of a reversible operad.

259: To such an operad $P$,

260: we associate a Lie algebra $QA_{\infty}$ and a graph complex

261: $(Q\mathcal G,\partial_E)$

262: and then show that they have the same homology.

263: A precise statement is given in Theorem~\ref{t:mt}

264: in Section~\ref{s:mt}.

265: 

266: 

267: 

268: \subsection{Organisation of the paper} \label{ss:op}

269: 

270: We now explain the structure of the paper 

271: and give a guide to how readers 

272: with different interests and backgrounds can read it.

273: 

274: The two words that dominate the paper are 

275: species and operads.

276: These notions are explained in Section~\ref{s:ro}.

277: In the same section we define reversible operads.

278: We give plenty of simple examples 

279: with stress laid on drawing pictures.

280: More examples are discussed in Sections~\ref{s:me}

281: and \ref{s:gp}.

282: We hope that the amount of detail given

283: along with the examples will be sufficient 

284: for the reader unfamiliar with these concepts.

285: 

286: In Section~\ref{s:mf}, we construct a 

287: $$

288: \text{Mating Functor} \ : \ \{\text{Reversible operads}\}

289: \longrightarrow \{\text{Species}\}.

290: $$

291: Let $P$ be a reversible operad and

292: $Q$ its image under this functor.

293: Remember that our first goal is to associate a Lie algebra 

294: and a graph complex to $P$.

295: The fact is that the mating functor is the main step

296: in constructing these two objects,

297: and the species $Q$ plays a key role.

298: So we call these two objects $QA_{\infty}$ and

299: $(Q\mathcal G,\partial_E)$ respectively,

300: stressing the close connection to the species $Q$.

301: The letters $A$ and $\mathcal G$ stand for 

302: algebra and graph respectively.

303: The subscript $\infty$ refers to the fact that $QA_{\infty}$

304: will be defined as a direct limit 

305: of a family of Lie algebras $QA_n$.

306: 

307: Sections~\ref{s:ro} and \ref{s:mf},

308: on which we have elaborated so far,

309: form the basic structure

310: on which the rest of the paper is built.

311: It is a good idea to avoid technical details

312: on a first reading and just keep a few concrete examples in mind.

313: The rest of the paper can be split into three parts.

314: Part A is about the left hand side of Theorem~\ref{t:mt}

315: (Lie algebras),

316: Part B is about the right hand side (graph homology)

317: and Part C is about stating and proving the theorem.

318: 

319: \subsubsection*{Part A : Sections~\ref{s:sgo}-\ref{s:me}}

320: This deals with symplectic operad geometry.

321: We start with an overview in Section~\ref{s:sgo}.

322: The goal of Section~\ref{s:cm}

323: is to define the Lie algebra $QA_{\infty}$.

324: This is done via a simple pictorial calculus 

325: of cuttings and matings.

326: The Lie algebra homology $H_*(QA_{\infty})$

327: has the structure of a Hopf algebra,

328: which we then explain.

329: This completes the definition of the left hand side of

330: Theorem~\ref{t:mt}.

331: 

332: In Section~\ref{s:st},

333: we explain Kontsevich's symplectic mini-theory.

334: This is logically not essential to the rest of the paper.

335: However, it is conceptually the most important section

336: in Part A.

337: It shows how the Lie structure on $QA_{\infty}$

338: emerges naturally from a symplectic form.

339: This allows us to think of $QA_{\infty}$

340: as Hamiltonian functions on a ``symplectic operad manifold''.

341: 

342: 

343: \subsubsection*{Part B : Sections~\ref{s:gh}-\ref{s:mgh}}

344: 

345: This deals with graph homology.

346: The reader, who is mainly interested in graph homology,

347: can directly start with this part even skipping

348: Sections~\ref{s:ro} and \ref{s:mf}.

349: We first define a general graph complex $(\mathcal G,\partial_E)$

350: and then look at other relevant (and smaller) graph complexes,

351: $(Q\mathcal G,\partial_E)$ being one of them.

352: This completes the definition of the right hand side of

353: Theorem~\ref{t:mt}.

354: 

355: 

356: Section~\ref{s:gp} is optional.

357: It contains an interesting example of the theory 

358: which is based on groups.

359: In Section~\ref{s:mgh},

360: we continue with graph homology and develop some tools 

361: that will be used in the proof of the main theorem.

362: This involves defining a cochain complex $(\mathcal G,\delta_E)$,

363: i.e. graph cohomology.

364: The homology and cohomology are related 

365: by an interesting and highly non-trivial pairing on graphs

366: (\ref{ss:pg}).

367: 

368: \subsubsection*{Part C : Sections~\ref{s:mt}-\ref{s:con}}

369: 

370: This deals with the connection between symplectic geometry

371: and graph homology.

372: Section~\ref{s:mt} contains a precise formulation

373: of the main theorem.

374: To digest the statement completely,

375: the material on Lie algebras and graph homology

376: in Sections~\ref{s:cm} and \ref{s:gh} respectively

377: is a prerequisite.

378: The next three sections deal with the proof of the theorem.

379: 

380: The main ideas of the proof are already present 

381: in the commutative case.

382: Since the proof is quite involved,

383: we suggest that the reader specialise to this case

384: on a first reading.

385: The Lie algebra $QA_{\infty}$

386: in this case is easy to define directly.

387: So Sections~\ref{s:ro}-\ref{s:me} are not necessary.

388: Thus the reader, who mainly wants to see the ideas in the proof,

389: may start with Part C directly,

390: skipping Part A and referring back to Part B as necessary.

391: 

392: The proof of Theorem~\ref{t:mt} is given in three parts

393: (Sections~\ref{s:fin},\ref{s:sta} and~\ref{s:con}).

394: In each of these sections, we state and prove

395: a theorem of the form

396: \[

397: H_*(\text{Lie algebra}) = H_*(\text{Graph complex}),

398: \]

399: gradually getting closer to our goal.

400: As mentioned earlier,

401: the main tool in the proof

402: is classical invariant theory of $\fsp$.

403: This part is done in Section~\ref{s:fin}.

404: The second part of the theorem given in

405: Section~\ref{s:sta} deals with the stability issue

406: $(n \to \infty)$

407: and makes strong use of the ideas of 

408: Section~\ref{s:mgh}.

409: In Section~\ref{s:con}, we understand the space

410: of primitive elements of the Hopf algebra $H_*(QA_{\infty})$.

411: This is largely a matter of unwinding definitions

412: and there are no new ideas here.

413: Since Hopf algebras play a minor role in this paper,

414: the reader unfamiliar with them may simply omit this section

415: without losing any of the main ideas.

416: 

417: \subsubsection*{Appendix}

418: We have included two appendices,

419: which explain how deformation theory

420: relates to the ideas of this paper.

421: They are not logically essential

422: to the understanding of the main theorem.

423: Appendix~\ref{ss:dq} speculates 

424: on the deformation quantisation problem for operads

425: in this setting.

426: In Appendix~\ref{ss:dg},

427: we show that a certain Lie bracket 

428: on graph homology defined in~\cite{\cv} is zero. 

429: 

430: 

431: \subsection{A historical note} \label{ss:hn}

432: 

433: This work grew out of a seminar

434: organised by Karen Vogtmann,

435: devoted to understanding Kontsevich's work

436: (Fall 2000).

437: Other participants included 

438: D. Ciubotaru, F. Gerlits, D. Brown, J. Conant, 

439: M. Horak, F. Schwartz, M. Cohen and J. West.

440: My understanding of Kontsevich's ideas,

441: particularly the proof of the commutative case,

442: is due to them.

443: The material in (\ref {ss:og}-\ref {ss:gc})

444: and (\ref {ss:lah}-\ref {ss:inv}) is based 

445: on the seminar notes.

446: Conant and Vogtmann are writing an exposition 

447: that has some overlap with this.

448: 

449: The notion of a reversible operad and

450: its role in symplectic operad geometry

451: was done in July 2001 (Sections~\ref{s:ro}-\ref{s:me}).

452: The rest of the paper, namely, 

453: defining graph homology for a mated species and Theorem~\ref{t:mt},

454: was then relatively easy. 

455: The result in Appendix~\ref{ss:dg} was proved later

456: in November 2001.

457: These ideas, in some form or the other,

458: had already appeared in the works of 

459: Getzler and Kapranov~\cite{\getzlerkap,\getzkap},

460: Ginzburg~\cite{\ginz} and Markl~\cite{\markl}. 

461: Being ignorant of the operad literature at that time,

462: this was a gradual discovery for me.

463: In this paper, I give my own viewpoint of the subject,

464: which was formed by reading Kontsevich.

465: 

466: In \cite{\getzlerkap},

467: Getzler and Kapranov introduced the notion of cyclic operads

468: in order to extend the formalism of cyclic homology

469: for associative algebras~\cite{\loday} to operad algebras.

470: This notion coincides with what we call a reversible operad.

471: For such an operad $P$, they introduced a functor

472: $$

473: \lambda(P,_{\ -}) \ : \ \{P\text{-algebras}\}

474: \longrightarrow \{\text{Vector spaces}\}.

475: $$

476: In our notation (see Sections~\ref{s:sgo} and \ref{s:cm}), 

477: $\lambda(P,PA) = QA$,

478: and equation~\eqref{e:r} in (\ref {ss:fa})

479: can be taken as the definition 

480: of $\lambda$.

481: The mating functor is similar in spirit to $\lambda$, except that

482: we deal with operads and species rather than

483: algebras and vector spaces.

484: 

485: In \cite{\ginz}, Ginzburg explained Kontsevich's

486: symplectic mini-theory for a cyclic Koszul operad.

487: He denoted the functor $\lambda$ by the letter $R$.

488: Since a Koszul operad is quadratic,

489: the relations in equation~\eqref{e:r} take the simpler form

490: \begin{equation*}

491: a \otimes b = b \otimes a  \quad \text{and} \quad

492: \mu(a,b) \otimes c = a \otimes \mu(b,c) \quad \text{for} \quad \mu \in P[2].

493: \end{equation*}

494: These are precisely the relations that Kontsevich 

495: wrote in~\cite{\god}.

496: Elements of associative and Lie calculus 

497: had appeared earlier in the works of Karoubi~\cite{\karoubi}

498: and Drinfeld~\cite{\drinfeld}.

499: 

500: However, as we show, to do symplectic geometry,

501: it is sufficient to assume that $P$ is reversible,

502: not necessarily Koszul or quadratic.

503: In fact, all the examples in Section~\ref{s:me}

504: and some earlier ones too are non-quadratic.

505: We discovered reversibility by requiring that

506: the notion of a partial derivative make sense

507: (Proposition~\ref{p:pd}).

508: So in this sense, it is also a necessary condition

509: to do symplectic geometry.

510: 

511: In~\cite{\markl}, Markl considered graph complexes

512: associated to cyclic operads.

513: And for this, he referred back to the Feynman transform

514: construction of Getzler and Kapranov~\cite{\getzkap}.

515: 

516: We conclude the introduction with two remarks. 

517: Koszul operads will play an important role in the second part

518: of Kontsevich's theorem,

519: which we do not consider here.

520: If we switch from Lie to Liebniz algebras~\cite{\lodayo}

521: then we will end up with another avatar of graph homology.

522: Now vertices of graphs will have honest labels, 

523: unlike in usual graph homology,

524: where vertices are labelled only upto orientation.

525: 

526: \newpage

527: \section{Species, operads and reversible operads}  \label{s:ro}

528: 

529: In this section, we give a brief introduction 

530: to species and operads and then define reversible operads.

531: An excellent exposition on species can be found in the book

532: by Bergeron, Labelle and Leroux~\cite{\bll}.

533: The notes by Voronov~\cite{\voronov}

534: contain a good review on operads.

535: A more comprehensive reference is the seminal work

536: of Ginzburg and Kapranov~\cite{\gk}.

537: For some recent developments on operads, 

538: see~\cite{\getzler,\getzlerkap,\getzkap,\lsv,\lodayetal}.

539: For earlier literature,

540: see~\cite{\joyal,\may,\stasheff,\lazard},

541: where the concepts of species and operads first originated.

542: 

543: \subsection{Species}  \label{ss:s}

544: 

545: A species $Q$ is a functor from the category of 

546: (Finite sets, bijections) to the category of 

547: (Sets, maps).

548: We denote the image of a set $I$ by $Q[I]$

549: and say that $Q[I]$ is the set of $Q$-structures on the set $I$.

550: The set $Q[\phi]$ will always be empty.

551: 

552: \medskip

553: \begin{center}

554: \begin{tabular}{|c|c|} \hline 

555: Species $Q$ & Picture of an element of $Q[I]$ 

556: for $I=\{a,b,c,d\}$ \\ \hline \hline

557: $c$ & 

558: \begin{minipage}{.4 in}

559: \vspace{.1in}

560: \begin{center}

561: \psfrag{a}{\Huge $a$}

562: \psfrag{b}{\Huge $b$}

563: \psfrag{c}{\Huge $c$}

564: \psfrag{d}{\Huge $d$}

565: \resizebox{.4 in}{.4 in}{\includegraphics{ccd.eps}}

566: \end{center}

567: \vspace{.1in}

568: \end{minipage} 

569: \ \ = \ \

570: \begin{minipage}{.4 in}

571: \vspace{.1in}

572: \begin{center}

573: \psfrag{a}{\Huge $a$}

574: \psfrag{b}{\Huge $b$}

575: \psfrag{c}{\Huge $c$}

576: \psfrag{d}{\Huge $d$}

577: \resizebox{.4 in}{.4 in}{\includegraphics{cc.eps}}

578: \end{center}

579: \vspace{.1in}

580: \end{minipage} 

581: \ \ = \ \

582: \begin{minipage}{.4 in}

583: \vspace{.1in}

584: \begin{center}

585: \psfrag{a}{\Huge $a$}

586: \psfrag{b}{\Huge $b$}

587: \psfrag{c}{\Huge $c$}

588: \psfrag{d}{\Huge $d$}

589: \resizebox{.4 in}{.4 in}{\includegraphics{csubt.eps}}

590: \end{center}

591: \vspace{.1in}

592: \end{minipage} \\ \hline

593: $a$ &  

594: \begin{minipage}{.65 in}

595: \vspace{.1in}

596: \begin{center}

597: \psfrag{a}{\Huge $a$}

598: \psfrag{b}{\Huge $b$}

599: \psfrag{c}{\Huge $c$}

600: \psfrag{d}{\Huge $d$}

601: \resizebox{.65 in}{.2 in}{\includegraphics{ad.eps}}

602: \end{center}

603: \vspace{.1in}

604: \end{minipage}

605: \ \ = \ \

606: \begin{minipage}{.55 in}

607: \vspace{.1in}

608: \begin{center}

609: \psfrag{a}{\Huge $a$}

610: \psfrag{b}{\Huge $b$}

611: \psfrag{c}{\Huge $c$}

612: \psfrag{d}{\Huge $d$}

613: \resizebox{.55 in}{.5 in}{\includegraphics{asubt.eps}}

614: \end{center}

615: \vspace{.1in}

616: \end{minipage} \\ \hline

617: $aa$ &  

618: \begin{minipage}{.5 in}

619: \vspace{.1in}

620: \begin{center}

621: \psfrag{a}{\Huge $a$}

622: \psfrag{b}{\Huge $b$}

623: \psfrag{c}{\Huge $c$}

624: \psfrag{d}{\Huge $d$}

625: \resizebox{.5 in}{.5 in}{\includegraphics{aa.eps}}

626: \end{center}

627: \vspace{.1in}

628: \end{minipage} \\ \hline

629: $t$ &

630: \begin{minipage}{.8 in}

631: \vspace{.1in}

632: \begin{center}

633: \psfrag{a}{\Huge $a$}

634: \psfrag{b}{\Huge $b$}

635: \psfrag{c}{\Huge $c$}

636: \psfrag{d}{\Huge $d$}

637: \resizebox{.75 in}{.4 in}{\includegraphics{tsubt.eps}}

638: \end{center}

639: \vspace{.1in}

640: \end{minipage} \\ \hline

641: $tt$ & 

642: \begin{minipage}{.8 in}

643: \vspace{.1in}

644: \begin{center}

645: \psfrag{a}{\Huge $a$}

646: \psfrag{b}{\Huge $b$}

647: \psfrag{c}{\Huge $c$}

648: \psfrag{d}{\Huge $d$}

649: \resizebox{.7 in}{.45 in}{\includegraphics{tt.eps}}

650: \end{center}

651: \vspace{.1in}

652: \end{minipage} \\ \hline

653: \end{tabular}

654: \end{center}

655: 

656: \vspace{0.3 cm}

657: \noindent

658: We now give some examples of species.

659: The pictures that go along with the examples are shown in the table.

660: \begin{itemize}

661: 

662: \item

663: $c[I]=\{I\}$,

664: that is, there is exactly one $c$-structure on the set $I$.

665: There are a variety of ways to show this via pictures.

666: 

667: \item

668: $a[I]=$ the set of linear orders on the set $I$.

669: We have shown two pictures for it.

670: 

671: \item

672: $aa[I]=$ the set of cyclic orders on the set $I$

673: if $|I| \geq 2$ and empty otherwise.

674: 

675: \item

676: $t[I]=$ the set of rooted trees with leaves labelled 

677: by elements of the set $I$.

678: 

679: \item

680: $tt[I]=$ the set of trees with leaves labelled 

681: by elements of the set $I$

682: if $|I| \geq 2$ and empty otherwise.

683: 

684: \end{itemize}

685: %

686: An element of $Q[I]$ can be schematically drawn as

687: 

688: 

689: \[

690: \begin{minipage}{.8 in}

691: \begin{center}

692: \psfrag{a}{\Huge $a$}

693: \psfrag{b}{\Huge $b$}

694: \psfrag{c}{\Huge $c$}

695: \psfrag{d}{\Huge $d$}

696: \resizebox{.6 in}{.6 in}{\includegraphics{q.eps}}

697: \end{center}

698: %\label{f:q}

699: \end{minipage}

700: \in Q[I] \ \ \text{for} \ \ I=\{a,b,c,d\}.

701: \]

702: Note that the picture for the species ``$aa$''

703: fits in with this representation perfectly.

704: And the remaining examples can also be

705: made to fit in without difficulty.

706: For example,

707: the species ``$c$'' can be drawn as

708: \[

709: \begin{minipage}{.8 in}

710: \begin{center}

711: \psfrag{a}{\Huge $a$}

712: \psfrag{b}{\Huge $b$}

713: \psfrag{c}{\Huge $c$}

714: \psfrag{d}{\Huge $d$}

715: \resizebox{.5 in}{.5 in}{\includegraphics{ccs.eps}}

716: \end{center}

717: \end{minipage},

718: \]

719: and so on for the others too.

720: 

721: A species can be equivalently defined as a

722: sequence $Q[0],Q[1],Q[2], \ldots$,

723: where $Q[n]$ is a $\Sigma_n$ set,

724: where $\Sigma_n$ is the symmetric group

725: on $n$ letters.

726: Here we have abbreviated our notation so that

727: $Q[n]$ stands for $Q[\{1,2,\ldots,n\}]$.

728: In the pictures, 

729: the group $\Sigma_n$ acts by permuting the $n$ letters.

730: 

731: It is also useful to use a more general definition of species 

732: where the target category is replaced

733: by the category of vector spaces.

734: In this case,

735: a species is a sequence as above with

736: $Q[n]$ being a linear representation of $\Sigma_n$.

737: It is clear that using vector spaces as the target category

738: is more general because

739: one can go from sets to vector spaces 

740: by linearising the representation.

741: 

742: \begin{remark}

743: Throughout this paper,

744: we use only sets and vector spaces as the target categories.

745: However, this restriction is mainly for simplicity.

746: For example,

747: one may consider a species in the category of topological spaces.

748: It would be a sequence $X[0],X[1],X[2], \ldots$

749: of topological spaces with an action of

750: $\Sigma_n$ on $X[n]$.

751: \end{remark}

752: 

753: \subsection{Operads}  \label{ss:o}

754: 

755: Note that for each of the examples $c,a,t$,

756: we suggested a picture with an arrow ``$\rightarrow$''

757: drawn in it.

758: This is possible because these species

759: have the additional structure of an operad.

760: 

761: We begin with an informal discussion on operads.

762: An operad $P$ is a species in which there is a substitution rule.

763: It can be shown schematically as

764: \[

765: \begin{minipage}{.8 in}

766: \begin{center}

767: \psfrag{a}{\Huge $a$}

768: \psfrag{b}{\Huge $b$}

769: \psfrag{c}{\Huge $c$}

770: \psfrag{d}{\Huge $d$}

771: \resizebox{.6 in}{.6 in}{\includegraphics{p.eps}}

772: \end{center}

773: %\label{f:p}

774: \end{minipage}

775: \in P[I] \ \ \text{for} \ \ I=\{a,b,c,d\}.

776: \]

777: One thinks of $a,b,c,d$ as four inputs and the arrow as an output.

778: The substitution rule allows us to feed 

779: the output of one object $p_2$ into 

780: the input of another object $p_1$.

781: We write this as $p_2 \rightarrow p_1$ or

782: $p_1 \leftarrow p_2$.

783: This can be shown as

784: \[

785: \begin{minipage}{.6 in}

786: \begin{center}

787: \psfrag{a}{\Huge $a$}

788: \psfrag{b}{\Huge $b$}

789: \psfrag{c}{\Huge $c$}

790: \psfrag{d}{\Huge $d$}

791: \psfrag{p2}{\Huge $p_2$}

792: \resizebox{.6 in}{.6 in}{\includegraphics{psubt.eps}}

793: \end{center}

794: %\label{f:psubt}

795: \end{minipage}

796: \hfill

797: \begin{minipage}{.8 in}

798: \begin{center}

799: \psfrag{x}{\Huge $x$}

800: \psfrag{y}{\Huge $y$}

801: \psfrag{z}{\Huge $z$}

802: \psfrag{p1}{\Huge $p_1$}

803: \resizebox{.6 in}{.6 in}{\includegraphics{psubi.eps}}

804: \end{center}

805: %\label{f:psubi}

806: \end{minipage}

807: =

808: \begin{minipage}{.8 in}

809: \begin{center}

810: \psfrag{a}{\Huge $a$}

811: \psfrag{b}{\Huge $b$}

812: \psfrag{c}{\Huge $c$}

813: \psfrag{d}{\Huge $d$}

814: \psfrag{x}{\Huge $x$}

815: \psfrag{z}{\Huge $z$}

816: \resizebox{.6 in}{.7 in}{\includegraphics{psubo.eps}}

817: \end{center}

818: %\label{f:psubo}

819: \end{minipage}.

820: \]

821: If we wish to be more specific, we will write

822: $p_2 \yright p_1$.

823: This means that we feed $p_2$ to that input of $p_1$

824: whose label is $y$.

825: 

826: Now we show how the substitution rule

827: works in each of the examples $c,a,t$.

828: \[

829: \begin{minipage}{.6 in}

830: \begin{center}

831: \psfrag{a}{\Huge $a$}

832: \psfrag{b}{\Huge $b$}

833: \psfrag{c}{\Huge $c$}

834: \psfrag{d}{\Huge $d$}

835: \resizebox{.5 in}{.5 in}{\includegraphics{csubt.eps}}

836: \end{center}

837: %\label{f:csubt}

838: \end{minipage}

839: \hfill

840: \begin{minipage}{.8 in}

841: \begin{center}

842: \psfrag{x}{\Huge $x$}

843: \psfrag{y}{\Huge $y$}

844: \psfrag{z}{\Huge $z$}

845: \resizebox{.6 in}{.6 in}{\includegraphics{csubi.eps}}

846: \end{center}

847: %\label{f:csubi}

848: \end{minipage}

849: =

850: \begin{minipage}{.8 in}

851: \begin{center}

852: \psfrag{a}{\Huge $a$}

853: \psfrag{b}{\Huge $b$}

854: \psfrag{c}{\Huge $c$}

855: \psfrag{d}{\Huge $d$}

856: \psfrag{x}{\Huge $x$}

857: \psfrag{z}{\Huge $z$}

858: \resizebox{.6 in}{.6 in}{\includegraphics{csubo.eps}}

859: \end{center}

860: %\label{f:csubo}

861: \end{minipage}.

862: \]

863: 

864: \[

865: \begin{minipage}{.6 in}

866: \begin{center}

867: \psfrag{a}{\Huge $a$}

868: \psfrag{b}{\Huge $b$}

869: \psfrag{c}{\Huge $c$}

870: \psfrag{d}{\Huge $d$}

871: \resizebox{.6 in}{.6 in}{\includegraphics{asubt.eps}}

872: \end{center}

873: %\label{f:asubt}

874: \end{minipage}

875: \hfill

876: \begin{minipage}{.8 in}

877: \begin{center}

878: \psfrag{x}{\Huge $x$}

879: \psfrag{y}{\Huge $y$}

880: \psfrag{z}{\Huge $z$}

881: \resizebox{.6 in}{.6 in}{\includegraphics{asubi.eps}}

882: \end{center}

883: %\label{f:asubi}

884: \end{minipage}

885: =

886: \begin{minipage}{.8 in}

887: \begin{center}

888: \psfrag{a}{\Huge $a$}

889: \psfrag{b}{\Huge $b$}

890: \psfrag{c}{\Huge $c$}

891: \psfrag{d}{\Huge $d$}

892: \psfrag{x}{\Huge $x$}

893: \psfrag{z}{\Huge $z$}

894: \resizebox{.6 in}{.7 in}{\includegraphics{asubo.eps}}

895: \end{center}

896: %\label{f:asubo}

897: \end{minipage}.

898: \]

899: For the species $t$, we graft the root of the first tree 

900: into the specified leaf of the second tree.

901: \[

902: \begin{minipage}{.7 in}

903: \begin{center}

904: \psfrag{a}{\Huge $a$}

905: \psfrag{b}{\Huge $b$}

906: \psfrag{c}{\Huge $c$}

907: \psfrag{d}{\Huge $d$}

908: \resizebox{.7 in}{.6 in}{\includegraphics{tsubt.eps}}

909: \end{center}

910: %\label{f:tsubt}

911: \end{minipage}

912: \hfill

913: \begin{minipage}{.8 in}

914: \begin{center}

915: \psfrag{x}{\Huge $x$}

916: \psfrag{y}{\Huge $y$}

917: \psfrag{z}{\Huge $z$}

918: \resizebox{.6 in}{.5 in}{\includegraphics{tsubi.eps}}

919: \end{center}

920: %\label{f:tsubi}

921: \end{minipage}

922: =

923: \begin{minipage}{1 in}

924: \begin{center}

925: \psfrag{a}{\Huge $a$}

926: \psfrag{b}{\Huge $b$}

927: \psfrag{c}{\Huge $c$}

928: \psfrag{d}{\Huge $d$}

929: \psfrag{x}{\Huge $x$}

930: \psfrag{z}{\Huge $z$}

931: \resizebox{.9 in}{.6 in}{\includegraphics{tsubo.eps}}

932: \end{center}

933: %\label{f:tsubo}

934: \end{minipage}.

935: \]

936: Note that there is another way possible.

937: One could contract the edge at which the grafting took place.

938: This is what happened for the species ``$c$''.

939: There are numerous other variations one can do on this example;

940: for instance, one could use rooted binary trees.

941: 

942: To give a formal definition,

943: an operad $P$ is a species with a substitution rule

944: which is

945: \begin{itemize}

946: \item

947: associative, and

948: 

949: \item

950: compatible with the morphisms in the source category.

951: 

952: \end{itemize}

953: The first condition says that 

954: if we perform two substitutions,

955: one after the other,

956: then the order in which we do them does not matter.

957: The second condition says that substitution

958: commutes with relabelling of the inputs.

959: For more detail on these conditions,

960: see the references cited earlier.

961: 

962: For convenience of bookkeeping, it is useful to label

963: the output as well.

964: This idea can be formalised as follows.

965: For any operad $P$, define a bi-functor

966: \[

967: P : (\text{Sets,\ bijections}) \times

968: (\text{Singleton sets,\ bijections}) \to

969: (\text{Vector spaces,\ maps})

970: \]

971: by setting $P[I,U]=P[I]$,

972: where $U$ is a singleton set.

973: One thinks of an element of $P[I,U]$

974: as an element of $P[I]$

975: whose output is labelled by the element of $U$.

976: For the substitution rule in this situation, 

977: we demand that the output label of $p_2$

978: match the input label of $p_1$.

979: So the notation $p_2 \yright p_1$

980: now means that 

981: the label of the output of $p_2$ is $y$

982: and it is fed to an input of $p_1$

983: whose label is again $y$.

984: 

985: Just as for species,

986: we can define an operad as a sequence $P[n]$

987: of $\Sigma_n$ modules

988: equipped with a substitution rule

989: that is associative and $\Sigma_n$ invariant. 

990: A unit element of $P$ is an element of $P[1]$ 

991: which when substituted into an input of any

992: $p \in P[n]$ gives back $p$.

993: We assume that our operads have units

994: and call the unit element $u$.

995: In the second notation,

996: we would write $P[n,1]$ instead of $P[n]$

997: and the unit element would lie in $P[1,1]$.

998: 

999: 

1000: 

1001: \subsection{Examples} \label{ss:eg}

1002: 

1003: We now organise together a basic set of examples.

1004: These include the ones shown in the table (\ref{ss:s}).

1005: The motivation for organising them in this manner will become clear

1006: when we discuss the mating functor (\ref{ss:mf}).

1007: We make two quick remarks.

1008: All examples here are based on sets.

1009: Furthermore, the ones that are operads have unit elements

1010: and satisfy $P[1]=\Q$,

1011: i.e. $P[1]$ is just the span of the unit element $u$ of $P$.

1012: 

1013: 

1014: \subsubsection{The unit operad $u$ and species $uu$} \label{sss:u}

1015: Define 

1016: $u[I]=\{I\}$ if $I$ is a singleton and empty otherwise.

1017: And similarly, put

1018: $uu[I]=\{I\}$ if $|I|=2$ and empty otherwise.

1019: These two examples play a fundamental role in this paper

1020: because of their connection to the symplectic Lie algebra

1021: $\fsp$, see (\ref{ss:sl}).

1022: \[

1023: \begin{minipage}{1 in}

1024: \begin{center}

1025: \psfrag{a}{\huge $a$}

1026: \psfrag{b}{\huge $b$}

1027: \psfrag{u}{\Huge $u$}

1028: \resizebox{.7 in}{.25 in}{\includegraphics{u.eps}}

1029: \end{center}

1030: \end{minipage}

1031: \hfill

1032: \begin{minipage}{2 in}

1033: \begin{center}

1034: \psfrag{1}{}

1035: \psfrag{2}{}

1036: \psfrag{p1}{\huge $a$}

1037: \psfrag{q2}{\huge $b$}

1038: \psfrag{uu}{\huge $uu$}

1039: \resizebox{1 in}{.2 in}{\includegraphics{sf.eps}}

1040: \end{center}

1041: \end{minipage}

1042: \]

1043: 

1044: \subsubsection{The commutative operad $c$ and species $cc$}

1045: \label{sss:c}

1046: 

1047: Define 

1048: $c[I]=\{I\}$,

1049: that is, there is exactly one $c$-structure on the set $I$.

1050: Also define $cc[I]=\{I\}$ if $|I| \geq 2$ and empty otherwise.

1051: Though this is also an operad (without a unit),

1052: we call it the commutative species.

1053: As far as pictures go,

1054: we will use the one with arrow ``$\rightarrow$'' for ``$c$''

1055: and the one without it for ``$cc$''.

1056: 

1057: \subsubsection{The associative operad $a$ and species $aa$}  \label{sss:a}

1058: Let $a[I]=$ the set of linear orders on the set $I$.

1059: Also let $aa[I]=$ the set of cyclic orders on the set $I$

1060: if $|I| \geq 2$ and empty otherwise.

1061: 

1062: \subsubsection{The tree operad $t$ and species $tt$}  \label{sss:t}

1063: 

1064: Let $t[I]=$ the set of rooted trees with leaves labelled 

1065: by elements of the set $I$.

1066: For the tree species $tt$, we drop the word ``rooted''. 

1067: In other words,

1068: $tt[I]=$ the set of trees with leaves labelled 

1069: by elements of the set $I$

1070: if $|I| \geq 2$ and empty otherwise.

1071: Since there is no root, one cannot define a substitution rule.

1072: 

1073: \subsubsection{The chord operad $k$ and species $kk$}  \label{sss:k}

1074: The chord species $kk[I]$

1075: is the set of chord diagrams on the set $I$.

1076: That is, an element of $kk[I]$

1077: specifies a way to pair off the elements of $I$.

1078: Clearly, this is non-empty only when the cardinality of $I$ is even.

1079: Define the chord operad $k[I]$

1080: in the same way, except that

1081: one of the chords is left hanging at one end.

1082: Hence this is non-empty only when the cardinality of $I$ is odd.

1083: The substitution rule should be clear.

1084: 

1085: \medskip

1086: \noindent

1087: The species $aa,tt,kk$ have no natural substitution rule;

1088: so they are not operads.

1089: We will give some more examples later (Sections~\ref{s:me}

1090: and \ref{s:gp}).

1091: 

1092: 

1093: \subsection{Reversible operads} \label{ss:ro}

1094: 

1095: An operad $P$ is reversible if there is a rule that allows

1096: us to switch the output with any given input.

1097: One represents this pictorially as

1098: \[

1099: r_{x,y}

1100: \left(

1101: \begin{minipage}{.8 in}

1102: \begin{center}

1103: \psfrag{a}{\Huge $a$}

1104: \psfrag{b}{\Huge $b$}

1105: \psfrag{x}{\Huge $x$}

1106: \psfrag{y}{\Huge $y$}

1107: \resizebox{.7 in}{.6 in}{\includegraphics{prevt.eps}}

1108: \end{center}

1109: %\label{f:prevt}

1110: \end{minipage}

1111: \right)

1112: =

1113: \begin{minipage}{.8 in}

1114: \begin{center}

1115: \psfrag{a}{\Huge $a$}

1116: \psfrag{b}{\Huge $b$}

1117: \psfrag{x}{\Huge $x$}

1118: \psfrag{y}{\Huge $y$}

1119: \resizebox{.7 in}{.6 in}{\includegraphics{prevo.eps}}

1120: \end{center}

1121: %\label{f:prevo}

1122: \end{minipage}

1123: .

1124: \]

1125: To give a formal definition, for each set $I$ and $x \in I$, 

1126: there exist reversal maps

1127: \[

1128: r_{x,y} : P[I,\{y\}] \to P[I \setminus \{x\} \cup \{y\},\{x\}],

1129: \]

1130: which are subject to two conditions.

1131: 

1132: 1. We require

1133: $r_{y,x} \circ r_{x,y} = \ \text{identity}$ and

1134: $r_{x,y} \circ r_{y,z} = r_{x,z}.$ 

1135: %

1136: It follows from these two relations that the composite

1137: of a sequence of reversals is either the identity map

1138: or a single reversal. 

1139: 

1140: 2. Reversal must be compatible with substitution.

1141: That is,

1142: $$

1143:   r_{x,z} (p_2 \yright p_1) \ = \left\{ \begin{array}{c l}

1144:     p_2 \yright r_{x,z} (p_1) &

1145: \mbox{ if $x$ is an input of $p_1$,} \\

1146:     r_{x,y} (p_2) \yleft r_{y,z} (p_1) &

1147: \mbox{ if $x$ is an input of $p_2$.} 

1148:                     \end{array} \right. $$

1149: %

1150: Here $y$ labels one of the inputs of $p_1$

1151: as well as the output of $p_2$ 

1152: and $z$ labels the output of $p_1$.

1153: 

1154: The examples of operads $(P=u,c,a,t,k)$

1155: that we gave in (\ref {ss:eg}) are all reversible.

1156: For each of these examples,

1157: there is a picture of the operad

1158: that fits in with the schematic one.

1159: The reversal map in all cases then works exactly

1160: as shown above in the schematic picture.

1161: 

1162: We give some intuitive meaning to the conditions

1163: on the reversal maps.

1164: Condition 1 says that reversal is an external operation.

1165: The internal data of the operad remains unchanged.

1166: Condition 2 says that operad substitution

1167: commutes with operad reversal.

1168: This will become clearer from the examples

1169: of non-reversible operads given in (\ref{ss:ne}).

1170: 

1171: \begin{remark}

1172: The reversal rules may not be unique.

1173: For example, in (\ref{ss:g}), we give two distinct ways

1174: to reverse the graph operad.

1175: \end{remark}

1176: 

1177: 

1178: \subsection{Non-examples}  \label{ss:ne}

1179: We now give two examples of operads that are not reversible.

1180: So they play no role in the rest of the paper.

1181: Let Perm$[I] = I$ and

1182: Dias$[I] =$ the set of linear orders with 

1183: a distinguished element on the set $I$.

1184: \[

1185: \begin{minipage}{.8 in}

1186: \begin{center}

1187: \psfrag{a}{\Huge $a$}

1188: \psfrag{b}{\Huge $b$}

1189: \psfrag{c}{\Huge $c$}

1190: \psfrag{d}{\Huge $d$}

1191: \resizebox{.5 in}{.5 in}{\includegraphics{permt.eps}}

1192: \end{center}

1193: \end{minipage}

1194: \in \text{Perm}[I], \qquad \qquad \qquad

1195: \begin{minipage}{.8 in}

1196: \begin{center}

1197: \psfrag{a}{\Huge $a$}

1198: \psfrag{b}{\Huge $b$}

1199: \psfrag{c}{\Huge $c$}

1200: \psfrag{d}{\Huge $d$}

1201: \resizebox{.7 in}{.6 in}{\includegraphics{diast.eps}}

1202: \end{center}

1203: \end{minipage}

1204: \in \text{Dias}[I].

1205: \]

1206: The pictures show examples for $I=\{a,b,c,d\}$.

1207: 

1208: We explain the substitution rule for the operad Dias.

1209: There are two cases depending on whether

1210: the input into which the substitution occurs

1211: is circled or not.

1212: \[

1213: \begin{minipage}{.6 in}

1214: \begin{center}

1215: \psfrag{a}{\Huge $a$}

1216: \psfrag{b}{\Huge $b$}

1217: \psfrag{c}{\Huge $c$}

1218: \psfrag{d}{\Huge $d$}

1219: \resizebox{.6 in}{.6 in}{\includegraphics{diast.eps}}

1220: \end{center}

1221: %\label{f:diast}

1222: \end{minipage}

1223: \hfill

1224: \begin{minipage}{.8 in}

1225: \begin{center}

1226: \psfrag{x}{\Huge $x$}

1227: \psfrag{y}{\Huge $y$}

1228: \psfrag{z}{\Huge $z$}

1229: \resizebox{.6 in}{.6 in}{\includegraphics{diasi.eps}}

1230: \end{center}

1231: %\label{f:diasi}

1232: \end{minipage}

1233: =

1234: \begin{minipage}{.8 in}

1235: \begin{center}

1236: \psfrag{a}{\Huge $a$}

1237: \psfrag{b}{\Huge $b$}

1238: \psfrag{c}{\Huge $c$}

1239: \psfrag{d}{\Huge $d$}

1240: \psfrag{x}{\Huge $x$}

1241: \psfrag{z}{\Huge $z$}

1242: \resizebox{.6 in}{.7 in}{\includegraphics{diaso.eps}}

1243: \end{center}

1244: %\label{f:diaso}

1245: \end{minipage}.

1246: \]

1247: The picture illustrates the second case.

1248: The first case works as follows.

1249: If the input $y$ were circled instead of $x$

1250: then the result would be a circle on the input $b$.

1251: The substitution rule for Perm is similar.

1252: 

1253: Observe that Perm and Dias are closely related

1254: to the commutative (\ref{sss:c}) and associative (\ref{sss:a})

1255: operads respectively.

1256: They are binary quadratic and were introduced

1257: by Loday and his collaborators 

1258: with motivation from algebraic K-theory~\cite{\lodayetal}.

1259: For other examples of non-reversible operads,

1260: see~\cite{\getzlerkap}.

1261: 

1262: \section{The mating functor}  \label{s:mf}

1263: 

1264: In this section, we introduce the mating functor,

1265: which allows us to construct a mated species 

1266: from a reversible operad.

1267: We then explain the concept of a partial derivative

1268: for a mated species (Proposition~\ref{p:pd}).

1269: This is the most crucial part of the theory and 

1270: shows exactly how the conditions 

1271: on the reversal map given in (\ref{ss:ro}) arise.

1272: 

1273: 

1274: 

1275: \subsection{The mating functor}  \label{ss:mf}

1276: 

1277: Define a species $Q$ starting with a reversible operad $P$

1278: as follows. 

1279: For any finite set $K$, let

1280: $$Q[K] = \bigoplus_{I \sqcup J = K} P[I,U] \otimes P[J,U],$$

1281: subject to the two relations:

1282: 

1283: (R1) $p_2 \otimes p_1 = p_1 \otimes p_2,$ and 

1284: 

1285: (R2) $p_3 \rightarrow p_2 \otimes p_1 = p_3 \otimes r(p_2) \leftarrow p_1.$

1286: 

1287: \medskip

1288: \noindent

1289: The set $U$ is any singleton,

1290: all choices being considered equivalent.

1291: We interpret the tensor sign as a mating, and say that

1292: $p_2 \otimes p_1$ is a mating of $p_2$ and $p_1$.

1293: We show it as 

1294: \begin{minipage}{1.2 in}

1295: \begin{center}

1296: \psfrag{p1}{\Huge $p_1$}

1297: \psfrag{p2}{\Huge $p_2$}

1298: \resizebox{1 in}{.3 in}{\includegraphics{pm.eps}}

1299: \end{center}

1300: \end{minipage}.

1301: For simplicity of notation,

1302: we are suppressing the labels.

1303: The symmetry of relation $(R1)$ is thus built into the picture.

1304: If either of $p_2$ or $p_1$ is the unit element $u$

1305: then we say that the mating 

1306: $p_2 \otimes p_1$ is trivial.

1307: We show it as 

1308: \begin{minipage}{1 in}

1309: \begin{center}

1310: \psfrag{u}{\Huge $u$}

1311: \psfrag{p2}{\Huge $p_2$}

1312: \resizebox{.8 in}{.3 in}{\includegraphics{ptm.eps}}

1313: \end{center}

1314: \end{minipage}.

1315: In a nontrivial mating,

1316: we will call the segment joining $p_2$ and $p_1$ an \emph{ideal edge}.

1317: It has two opposing arrowheads in the centre.

1318: 

1319: \begin{remark}

1320: Ideal edges will play an important role later

1321: in the definition of graph cohomology (\ref{ss:stacb}).

1322: \end{remark}

1323: %

1324: 

1325: The second relation $(R2)$ is written as under

1326: \[

1327: \begin{minipage}{1.9 in}

1328: \begin{center}

1329: \psfrag{p1}{\Huge $p_3$}

1330: \psfrag{p2}{\Huge $p_2$}

1331: \psfrag{p3}{\Huge $p_1$}

1332: \resizebox{1.7 in}{.4 in}{\includegraphics{pmr.eps}}

1333: \end{center}

1334: \end{minipage}

1335: =

1336: \begin{minipage}{1.9 in}

1337: \begin{center}

1338: \psfrag{p1}{\Huge $p_3$}

1339: \psfrag{pr}{\Huge $\overline{p_2}$}

1340: \psfrag{p3}{\Huge $p_1$}

1341: \resizebox{1.7 in}{.4 in}{\includegraphics{pmrr.eps}}

1342: \end{center}

1343: \end{minipage}

1344: .

1345: \]

1346: %

1347: The notation $\overline{p_2}$ stands for $r(p_2)$.

1348: 

1349: Since we assume that the operad $P$ has a unit,

1350: we can use relation $(R2)$ to express any mating

1351: $p_2 \otimes p_1$ as a trivial mating as follows.

1352: Write $p_1 = p_1 \leftarrow u$,

1353: where $u$ is the unit element of $P$. Then

1354: \begin{equation}  \label{e:tm}

1355: p_2 \otimes p_1 =

1356: p_2 \otimes p_1 \leftarrow u =

1357: p_2 \rightarrow r(p_1) \otimes u.

1358: \end{equation}

1359: This allows us to think of a mating $p_2 \otimes p_1$

1360: roughly

1361: as a reversal of $p_1$ followed by a substitution of

1362: $p_2$ into $p_1$ or vice-versa.

1363: 

1364: We will refer to a species obtained from a reversible operad

1365: by the above procedure as a \emph{mated species}.

1366: For obvious reasons,

1367: $Q[K]$ is non-empty only if $|K| \geq 2$.

1368: 

1369: \subsection{How elements of a reversible operad mate}  \label{ss:am}

1370: In examples, substitutions and reversals 

1371: usually have a pictorial description.

1372: And hence so do matings. 

1373: We now show the mating in the associative case.

1374: \[

1375: \begin{minipage}{1.3 in}

1376: \begin{center}

1377: \psfrag{a}{\Huge $a$}

1378: \psfrag{b}{\Huge $b$}

1379: \psfrag{c}{\Huge $c$}

1380: \psfrag{d}{\Huge $d$}

1381: \psfrag{x}{\Huge $x$}

1382: \psfrag{y}{\Huge $y$}

1383: \psfrag{z}{\Huge $z$}

1384: \resizebox{1.2 in}{.6 in}{\includegraphics{amb.eps}}

1385: \end{center}

1386: \end{minipage}

1387: =

1388: \begin{minipage}{1.1 in}

1389: \begin{center}

1390: \psfrag{a}{\Huge $a$}

1391: \psfrag{b}{\Huge $b$}

1392: \psfrag{c}{\Huge $c$}

1393: \psfrag{d}{\Huge $d$}

1394: \psfrag{x}{\Huge $x$}

1395: \psfrag{y}{\Huge $y$}

1396: \psfrag{z}{\Huge $z$}

1397: \resizebox{1 in}{.6 in}{\includegraphics{amt.eps}}

1398: \end{center}

1399: \end{minipage}

1400: =

1401: \begin{minipage}{1.1 in}

1402: \begin{center}

1403: \psfrag{a}{\Huge $a$}

1404: \psfrag{b}{\Huge $b$}

1405: \psfrag{c}{\Huge $c$}

1406: \psfrag{d}{\Huge $d$}

1407: \psfrag{x}{\Huge $x$}

1408: \psfrag{y}{\Huge $y$}

1409: \psfrag{z}{\Huge $z$}

1410: \resizebox{1 in}{.6 in}{\includegraphics{amm.eps}}

1411: \end{center}

1412: \end{minipage}

1413: =

1414: \begin{minipage}{.9 in}

1415: \begin{center}

1416: \psfrag{a}{\Huge $a$}

1417: \psfrag{b}{\Huge $b$}

1418: \psfrag{c}{\Huge $c$}

1419: \psfrag{d}{\Huge $d$}

1420: \psfrag{x}{\Huge $x$}

1421: \psfrag{y}{\Huge $y$}

1422: \psfrag{z}{\Huge $z$}

1423: \resizebox{.8 in}{.6 in}{\includegraphics{amd.eps}}

1424: \end{center}

1425: \end{minipage}.

1426: \]

1427: The  picture shows that the mating functor maps

1428: the associative operad $a$ to the associative species $aa$.

1429: 

1430: For the commutative case,

1431: mating occurs as follows.

1432: \[

1433: \begin{minipage}{1.2 in}

1434: \begin{center}

1435: \psfrag{a}{\Huge $a$}

1436: \psfrag{b}{\Huge $b$}

1437: \psfrag{c}{\Huge $c$}

1438: \psfrag{d}{\Huge $d$}

1439: \psfrag{x}{\Huge $x$}

1440: \psfrag{z}{\Huge $z$}

1441: \resizebox{1 in}{.5 in}{\includegraphics{cmb.eps}}

1442: \end{center}

1443: \end{minipage}

1444: =

1445: \begin{minipage}{.8 in}

1446: \begin{center}

1447: \psfrag{a}{\Huge $a$}

1448: \psfrag{b}{\Huge $b$}

1449: \psfrag{c}{\Huge $c$}

1450: \psfrag{d}{\Huge $d$}

1451: \psfrag{x}{\Huge $x$}

1452: \psfrag{z}{\Huge $z$}

1453: \resizebox{.5 in}{.5 in}{\includegraphics{cmd.eps}}

1454: \end{center}

1455: \end{minipage}.

1456: \]

1457: Thus the commutative operad $c$ 

1458: maps to the commutative species $cc$.

1459: 

1460: Similarly, one sees that an operad $P$ 

1461: maps to the species $Q=PP$ when $P=u,t,k$,

1462: see (\ref{ss:eg}).

1463: The letter repetition is used to indicate that

1464: the species is obtained as a mating.

1465: A more correct notation would be to write

1466: $Q=P\reflectbox{$P$}$,

1467: but we will not bother with that.

1468: 

1469: \subsection{The special role of $u$ and $uu$}  \label{ss:uu}

1470: 

1471: In all examples so far (\ref{ss:eg}),

1472: we have $P[1]=\Q$.

1473: In other words, 

1474: it is just the span of the unit element $u$ of $P$.

1475: Hence $Q[2]=\Q$ is the span of the element

1476: obtained by mating the unit element of $P$

1477: with itself, which is $uu$.

1478: And so $cc[2]=aa[2]=tt[2]=kk[2]$.

1479: For example,

1480: there is only one cyclic order or

1481: unrooted tree or chord diagram on $2$ letters.

1482: Thus if we ignore the pieces of degree $>$ 1 in $P$

1483: and $>$ 2 in $Q$

1484: then we are left precisely with the unit operad $u$

1485: and species $uu$.

1486: 

1487: In general, we always assume that $P$ has a unit.

1488: Hence $u$ and $uu$ are always a part of $P$ and $Q$ respectively.

1489: For convenience, we use the letter $u$

1490: for both the unit operad and 

1491: the unit element in an operad.

1492: Similarly for $uu$.

1493: 

1494: \subsection{The partial derivative}  \label{ss:pd}

1495: 

1496: Now we introduce the notion of a partial derivative 

1497: for a mated species.

1498: 

1499: \begin{proposition}  \label{p:pd}

1500: Let $P$ be a reversible operad with unit $u$ and

1501: $Q$ be its mated species.

1502: For $a \in K$ and $q \in Q[K]$,

1503: there is a unique element

1504: $p = \frac{\partial q}{\partial a} \in P[K \setminus a]$

1505: such that $q = \frac{\partial q}{\partial a} \otimes u$. 

1506: In terms of pictures, 

1507: \[

1508: \begin{minipage}{1 in}

1509: \begin{center}

1510: \psfrag{a}{\Huge $a$}

1511: \psfrag{b}{\Huge $b$}

1512: \psfrag{c}{\Huge $c$}

1513: \psfrag{d}{\Huge $d$}

1514: \psfrag{q}{\Huge $q$}

1515: \resizebox{.8 in}{.6 in}{\includegraphics{s.eps}}

1516: \end{center}

1517: \end{minipage}

1518: = 

1519: \begin{minipage}{1.3 in}

1520: \begin{center}

1521: \psfrag{a}{\Huge $a$}

1522: \psfrag{b}{\Huge $b$}

1523: \psfrag{c}{\Huge $c$}

1524: \psfrag{d}{\Huge $d$}

1525: \psfrag{p}{\Huge $p$}

1526: \psfrag{u}{\Huge $u$}

1527: \resizebox{1.1 in}{.6 in}{\includegraphics{der.eps}}

1528: \end{center}

1529: \end{minipage}.

1530: \]

1531: \end{proposition}

1532: \begin{proof}

1533: There are two parts to the proposition.

1534: The first one is the existence of

1535: $\frac{\partial q}{\partial a}$.

1536: This is same as saying that $q$ 

1537: can be written as a trivial mating

1538: (at a specified input).

1539: We have already derived this in

1540: equation~\eqref{e:tm}

1541: as a simple consequence of relation $(R2)$,

1542: see (\ref{ss:mf}).

1543:  

1544: The second part is to show the uniqueness of 

1545: $\frac{\partial q}{\partial a}$.

1546: Suppose that $q= p_1 \otimes u = p_2 \otimes u$.

1547: This means that one can obtain

1548: $p_2 \otimes u$ from 

1549: $p_1 \otimes u$ by successive applications of relation $(R2)$.

1550: Now the reduction lemma below says that

1551: this can be done in one step.

1552: This implies that $p_1 = p_2$.

1553: \end{proof}

1554: 

1555: To complete the proof of the above proposition, 

1556: we prove a reduction lemma

1557: that analyses the relation $(R2)$

1558: in the definition of a mated species (\ref{ss:mf}).

1559: The proof of the lemma will use the two conditions

1560: imposed on the reversal maps

1561: in the definition of a reversible operad (\ref{ss:ro}).

1562: 

1563: \subsection*{Local notation}

1564: 

1565: We will follow the convention that

1566: subscripts increase from right to left.

1567: If $p = p_2 \rightarrow p_1$  

1568: then we say that $p_2 \rightarrow p_1$ is a splitting of $p$.

1569: Similarly, $p = p_3 \rightarrow (p_2 \rightarrow p_1) = 

1570: (p_3 \rightarrow p_2) \rightarrow p_1$ 

1571: is a splitting of $p$ into three parts.

1572: Since operad substitution is associative,

1573: we may also write

1574: $p = p_3 \rightarrow p_2 \rightarrow p_1$.

1575: For simplicity of notation, we are suppressing labels.

1576: \vanish{

1577: Now say $p = p_2 \rightarrow p_1$.

1578: Hence

1579: $p \otimes p_0 = (p_2 \rightarrow p_1) \otimes p_0$.

1580: If we apply relation $(R2)$, we get

1581: $p_2 \otimes r(p_1) \leftarrow p_0$.

1582: We say that

1583: $p_2 \otimes r(p_1) \leftarrow p_0$

1584: is obtained from $p \otimes p_0$ by

1585: (splitting $p$ and) applying relation $(R2)$.

1586: }

1587: \begin{reduction} 

1588: Let $P$ be a reversible operad 

1589: with reversal rule $r$ and unit $u$.

1590: The result obtained by two successive applications

1591: of relation $(R2)$ to $p \otimes u$ can, in fact, 

1592: be obtained by a single application of $(R2)$.

1593: \end{reduction}

1594: \begin{proof}

1595: We may assume that the first application of relation $(R2)$ to

1596: $p \otimes u$ involves splitting $p$.

1597: If we split $u$ then nothing changes

1598: and the lemma is proved directly.

1599: Hence say

1600: $p = p^{\prime}_2 \rightarrow p_1$

1601: and applying $(R2)$, we obtain

1602: $$p \otimes u \ = \ p^{\prime}_2 \rightarrow p_1 \otimes u

1603: \ \ \app \ \ p^{\prime}_2 \otimes r(p_1) \leftarrow u \ = \

1604: p^{\prime}_2 \otimes p^{\prime}_1,$$

1605: where $p^{\prime}_1 = r(p_1) \leftarrow u$.

1606: 

1607: For the second application, we apply relation $(R2)$ to 

1608: $p^{\prime}_2 \otimes p^{\prime}_1$.

1609: This involves two cases.

1610: 

1611: 1. We split $p^{\prime}_2$ as say $p^{\prime}_2 = p_3 \rightarrow p_2$.

1612: 

1613: 2. We split $p^{\prime}_1$.\\

1614: \noindent

1615: Let us look at the first case. We have

1616: $$p^{\prime}_2 \otimes p^{\prime}_1 \ = \ p_3 \rightarrow p_2 \otimes p^{\prime}_1

1617: \ \ \app \ \ p_3 \otimes r(p_2) \leftarrow p^{\prime}_1.$$

1618: Substituting $p^{\prime}_1 = r(p_1) \leftarrow u$

1619: and using the associativity of substitution and

1620: the compatibility of reversal with substitution,

1621: we get

1622: $$p_3 \otimes r(p_2) \leftarrow p^{\prime}_1 \ = \

1623: p_3 \otimes r(p_2) \leftarrow (r(p_1) \leftarrow u) \ = \

1624: p_3 \otimes r(p_2 \rightarrow p_1) \leftarrow u.$$

1625: Hence the net effect of two applications of relation $(R2)$ has been

1626: $$p \otimes u \ \ \rightsquigarrow \ \

1627: p_3 \otimes r(p_2 \rightarrow p_1) \leftarrow u,$$

1628: where

1629: $p = p^{\prime}_2 \rightarrow p_1 =

1630: (p_3 \rightarrow p_2) \rightarrow p_1 =

1631: p_3 \rightarrow (p_2 \rightarrow p_1)$.

1632: It is clear that the same effect is achieved by

1633: $$p \otimes u \ = \ p_3 \rightarrow (p_2 \rightarrow p_1) \otimes u

1634: \ \ \app \ \ p_3 \otimes r(p_2 \rightarrow p_1) \leftarrow u,$$

1635: which is just one application of relation $(R2)$

1636: to $p \otimes u$.

1637: This completes the analysis for the first case.

1638: 

1639: For the second case, we just point out

1640: a simple subcase and leave the rest out.

1641: If we split $p^{\prime}_1$ as $p^{\prime}_1 = r(p_1) \leftarrow u$

1642: and apply relation $(R2)$ then

1643: we just reverse the first step and get back $p \otimes u$.

1644: And here we used the relation

1645: $r_{y,x} \circ r_{x,y} = \ \text{identity}$.

1646: 

1647: \end{proof}

1648: 

1649: 

1650: 

1651: \section{Overview of symplectic operad geometry} \label{s:sgo}

1652: 

1653: In this section, we explain

1654: some of the philosophy of symplectic operad geometry.

1655: The reader can just glance through it on a first reading.

1656: Complete details are given in 

1657: Sections~\ref{s:cm}-\ref{s:me}.

1658: The reader more interested in graph homology

1659: may directly go to Sections~\ref{s:gh}-\ref{s:mgh}.

1660: 

1661: \subsection{What is an operad manifold?} \label{ss:om}

1662: 

1663: An ordinary manifold $X$ is a commutative object 

1664: in the following sense.

1665: Functions on $X$ form a commutative algebra,

1666: the product being pointwise multiplication.

1667: To go a little further,

1668: differential forms on $X$ form 

1669: a graded or supercommutative algebra

1670: with the usual wedge product.

1671: Many geometric notions associated to $X$

1672: can be captured by these commutative algebras.

1673: For example, vector fields on $X$

1674: are derivations of the algebra of functions on $X$.

1675: Among all manifolds, 

1676: $\R^n$ plays a special role.

1677: And the space of polynomial functions on $\R^n$

1678: is the free commutative algebra on $n$ generators.

1679: 

1680: Now for a general operad $P$,

1681: there is a notion of a $P$-algebra, a $P$-superalgebra, 

1682: a free $P$-algebra, etc.

1683: For example,

1684: for the associative operad, 

1685: one has associative algebras;

1686: for the Lie operad, one has Lie algebras and so on.

1687: We would like to think of a $P$-algebra 

1688: as the algebra of functions on a ``$P$-manifold'',

1689: and of the free $P$-algebra on $n$ generators

1690: as the algebra of functions on 

1691: the $n$ dimensional ``$P$-manifold''

1692: which is the analogue of $\R^n$.

1693: We emphasise that a ``$P$-manifold'' as a geometric object 

1694: does not make any sense.

1695: However, thinking of a $P$-algebra in this manner

1696: is useful because it allows us to make analogies

1697: with the commutative case.

1698:  

1699: In this paper, we will only deal with free $P$-algebras.

1700:  

1701: \subsection{What is symplectic operad geometry?} \label{s:sog}

1702: 

1703: Let us start with the commutative case.

1704: For an introduction to classical symplectic topology, 

1705: see the book by McDuff and Salamon~\cite{\mcduff}.

1706: A symplectic manifold $(X,\omega)$ is a manifold $X$

1707: with a closed non-degenerate 2 form $\omega$.

1708: The existence of such a 2 form implies that $X$

1709: is even dimensional.

1710: Further, the algebra of smooth functions

1711: $C^{\infty}(X)$ has the structure of a Lie algebra.

1712: The standard example of a symplectic manifold is

1713: $(\R^{2n}, \omega_0)$.

1714: Here $\omega_0$ is the standard symplectic form

1715: $\sum dp_i \wedge dq_i$,

1716: where $p_1,\ldots,p_n,q_1,\ldots,q_n$

1717: are coordinates on $\R^{2n}$.

1718: The Lie algebra structure on $C^{\infty}(\R^{2n})$

1719: is given by the usual Poisson bracket

1720: \begin{equation} \label{e:pb}

1721: \{F,H\}=\sum_{i=1}^n \frac{\partial F}{\partial p_i}\frac{\partial H}{\partial q_i} - \frac{\partial F}{\partial q_i}\frac{\partial H}{\partial

1722: p_i} \ \ \text{for} \ \ F,H \in C^{\infty}(\R^{2n}).

1723: \end{equation}

1724: %

1725: Instead of smooth functions, 

1726: it is easier to deal with polynomial functions

1727: and henceforth we will always do so.

1728: 

1729: \subsubsection{The general case}

1730: We want to do something similar for a general operad $P$.

1731: Let $X$ be a ``$P$-manifold''.

1732: As explained in (\ref{ss:om}),

1733: this simply means that we have a $P$-algebra.

1734: And since we are only going to deal with the free case,

1735: we further assume that we have a free $P$-algebra.

1736: Call it $PA$.

1737: Since one thinks of $PA$ as polynomial functions on $X$,

1738: one would expect a Lie algebra structure on $PA$,

1739: in analogy with the commutative case.

1740: However, this is not true.

1741: 

1742: It turns out that one needs to consider 

1743: another algebraic object $QA$

1744: (Section~\ref{s:cm}).

1745: And this object can be constructed only if $P$ is reversible.

1746: Then if $PA$ is free on an even number of generators,

1747: it is true that $QA$ is a Lie algebra.

1748: The two important operations are 

1749: $$\frac{\partial }{\partial x_i} : QA \longrightarrow PA 

1750: \ \ \ \text{and} \ \ \ 

1751: \{\ ,\ \} : QA \otimes QA \longrightarrow QA.$$

1752: %

1753: The Poisson bracket $\{\ ,\ \}$ equips $QA$ 

1754: with the structure of a Lie algebra.

1755: It is given by a formula that is almost identical 

1756: to equation~\eqref{e:pb}, see 

1757: equation~\eqref{e:ptb} in (\ref{ss:mate}).

1758: 

1759: Just as $PA$ is related to the operad $P$,

1760: the space $QA$ is related to an object $Q$.

1761: The object $Q$ is not an operad but a simpler object called a species.

1762: To give an analogy, operads are like algebras and

1763: species are like vector spaces.

1764: The species $Q$ is the image of $P$ under the mating functor

1765: constructed in Section~\ref{s:mf}.

1766: The space $QA$ can be directly defined 

1767: in terms of $Q$.

1768: 

1769: So far, we have not talked about any symplectic form

1770: in the general case.

1771: The point is that the Lie structure on $QA$ 

1772: emerges naturally from a symplectic form.

1773: This is explained in Section~\ref{s:st}.

1774: A ``symplectic $P$-manifold'' must be a ``$P$-manifold''

1775: which has a symplectic form.

1776: The only example we give is that of a free $P$-algebra

1777: on an even number of generators, for $P$ reversible.

1778: When specialised to the commutative case,

1779: this says that we are only considering the symplectic manifolds

1780: $(\R^{2n}, \omega_0)$.

1781: 

1782: \subsubsection{Back to the classical case}  \label{ss:tcc}

1783: 

1784: We now explain how the general discussion above

1785: specialises to the commutative situation.

1786: That is, $P$ is the commutative operad $c$, see (\ref{sss:c}).

1787: For $P=c$, we have the free (non-unital) $P$-algebra

1788: $PA=$ polynomial functions with no constant terms

1789: on the symplectic manifold $(\R^{2n}, \omega_0)$.

1790: The mated species $Q$ in this case,

1791: which is $cc$, is almost the same.

1792: Hence the corresponding algebraic object $QA$ is also similar.

1793: We have $QA=$ polynomial functions in $2n$ variables 

1794: with no constant or linear terms. 

1795: In this case, the partial derivative has the usual meaning 

1796: and the Poisson bracket on $QA$ is given by the formula

1797: in equation~\eqref{e:pb}.

1798: Thus the content of the general discussion in quite simple

1799: in this case.

1800: 

1801: We now give a pictorial description of this calculus. 

1802: If $F \in QA$ is a monomial, say

1803: $F = x_1^2 x_2 x_3$,

1804: then we represent it as 

1805: $F = 

1806: \mbox{

1807: \begin{minipage}{.4 in}

1808: \begin{center}

1809: \psfrag{a}{\Huge $x_2$}

1810: \psfrag{b}{\Huge $x_1$}

1811: \psfrag{c}{\Huge $x_3$}

1812: \psfrag{d}{\Huge $x_1$}

1813: \resizebox{.4 in}{.4 in}{\includegraphics{cc.eps}}

1814: \end{center}

1815: \end{minipage} 

1816: }

1817: $.

1818: And if $F$ is a polynomial rather than a monomial

1819: then we represent it as a formal sum of pictures.

1820: Though $PA$ is almost the same as $QA$,

1821: we will represent its elements slightly differently.

1822: Namely, if $x_1^2 x_2 x_3 \in PA$

1823: then we write it as

1824: $F = 

1825: \mbox{

1826: \begin{minipage}{.4 in}

1827: \begin{center}

1828: \psfrag{a}{\Huge $x_2$}

1829: \psfrag{b}{\Huge $x_1$}

1830: \psfrag{c}{\Huge $x_3$}

1831: \psfrag{d}{\Huge $x_1$}

1832: \resizebox{.4 in}{.4 in}{\includegraphics{csubt.eps}}

1833: \end{center}

1834: \end{minipage} 

1835: }

1836: $.

1837: 

1838: We explain how the partial derivative

1839: $\frac{\partial }{\partial x_i} : QA \longrightarrow PA$ 

1840: works by an example.

1841: \[

1842: \frac{\partial}{\partial x_1}

1843: \left(

1844: \begin{minipage}{.6 in}

1845: \begin{center}

1846: \psfrag{a}{\Huge $x_2$}

1847: \psfrag{b}{\Huge $x_1$}

1848: \psfrag{c}{\Huge $x_3$}

1849: \psfrag{d}{\Huge $x_1$}

1850: \resizebox{.4 in}{.4 in}{\includegraphics{cc.eps}}

1851: \end{center}

1852: \end{minipage} 

1853: \right) = 

1854: \begin{minipage}{.6 in}

1855: \begin{center}

1856: \psfrag{a}{\Huge $x_2$}

1857: \psfrag{b}{\Huge $$}

1858: \psfrag{c}{\Huge $x_3$}

1859: \psfrag{d}{\Huge $x_1$}

1860: \resizebox{.4 in}{.4 in}{\includegraphics{car.eps}}

1861: \end{center}

1862: \end{minipage} 

1863: \ \ + \ \

1864: \begin{minipage}{.6 in}

1865: \begin{center}

1866: \psfrag{a}{\Huge $x_2$}

1867: \psfrag{b}{\Huge $x_1$}

1868: \psfrag{c}{\Huge $x_3$}

1869: \psfrag{d}{\Huge $$}

1870: \resizebox{.4 in}{.4 in}{\includegraphics{cas.eps}}

1871: \end{center}

1872: \end{minipage} 

1873: .

1874: \]

1875: In other words,

1876: cut all inputs with label $x_1$, one at a time.

1877: The above pictorial equation just says that

1878: $\frac{\partial}{\partial x_1}

1879: (x_1^2 x_2 x_3) = 2 x_1 x_2 x_3$.

1880: 

1881: Now we illustrate the Poisson bracket on $QA$

1882: by an example.

1883: \[

1884: \left\{

1885: \begin{minipage}{.6 in}

1886: \begin{center}

1887: \psfrag{x}{\Huge $p_1$}

1888: \psfrag{y}{\Huge $q_1$}

1889: \psfrag{z}{\Huge $p_2$}

1890: \resizebox{.5 in}{.4 in}{\includegraphics{cca.eps}}

1891: \end{center}

1892: \end{minipage}

1893: \ \ , \ \

1894: \begin{minipage}{.8 in}

1895: \begin{center}

1896: \psfrag{p1}{\Huge $q_2$}

1897: \psfrag{p2}{\Huge $p_2$}

1898: \psfrag{q1}{\Huge $q_1$}

1899: \psfrag{q2}{\Huge $p_2$}

1900: \psfrag{1}{}

1901: \psfrag{2}{}

1902: \psfrag{uu}{}

1903: \resizebox{.7 in}{.15 in}{\includegraphics{sf.eps}}

1904: \end{center}

1905: \end{minipage}

1906: \right\}

1907: \ \ = \ \

1908: \begin{minipage}{1 in}

1909: \begin{center}

1910: \psfrag{x}{\Huge $p_1$}

1911: \psfrag{y}{\Huge $q_1$}

1912: \psfrag{z}{\Huge $p_2$}

1913: \psfrag{p2}{\Huge $p_2$}

1914: \resizebox{1 in}{.4 in}{\includegraphics{ccmb.eps}}

1915: \end{center}

1916: \end{minipage}

1917: \ \ = \ \

1918: \begin{minipage}{.6 in}

1919: \begin{center}

1920: \psfrag{x}{\Huge $p_1$}

1921: \psfrag{y}{\Huge $q_1$}

1922: \psfrag{z}{\Huge $p_2$}

1923: \resizebox{.5 in}{.4 in}{\includegraphics{cca.eps}}

1924: \end{center}

1925: \end{minipage}

1926: .

1927: \]

1928: In usual terms, this says that

1929: $\{p_1 q_1 p_2, p_2 q_2 \} = p_1 q_1 p_2$.

1930: So to compute $\{F,H\}$,

1931: cut a $p_i$ from $F$, and

1932: a $q_i$ from $H$ (or vice-versa),

1933: do a mating and then sum over all possibilities.

1934: Compare the picture above

1935: with the mating suggested for the commutative operad 

1936: in (\ref{ss:am}).

1937: 

1938: The pictorial calculus may look silly but

1939: the point is that it generalises very nicely

1940: to reversible operads (Section~\ref{s:cm}).

1941: The pictorial way of thinking can also be extended 

1942: to the computation of the homology of the Lie algebra $QA$.

1943: This involves the building of a chain complex of graphs,

1944: which leads to graph homology.

1945: This connection is made in the first part of the proof

1946: of the main theorem (Section~\ref{s:fin}).

1947: 

1948: We now record these ideas in a nutshell as follows.

1949: \medskip

1950: 

1951: \begin{center}

1952: \begin{tabular}{|c|c|} \hline 

1953: Pictorial & Formal \\ \hline \hline

1954: Cutting of a vertex & Partial derivative \\ \hline

1955: Mating two cut vertices & Poisson bracket \\ \hline

1956: Building a graph & Homology \\ \hline 

1957: \end{tabular}

1958: \end{center}

1959: 

1960: 

1961: \section{Cuttings and Matings}  \label{s:cm}

1962: 

1963: The goal of this section

1964: is to define the Lie algebra $QA_{\infty}$

1965: that occurs in the left hand side of the main theorem.

1966: Let $V$ be the vector space with basis

1967: $x_1, \ldots,x_n$.

1968: Later to do symplectic theory,

1969: we will use the basis

1970: $p_1, \ldots,p_n,q_1, \ldots,q_n$.

1971: If we want to show the dependence of $n$ explicitly,

1972: we will write $V_n$ for $V$.

1973: 

1974: Let $P$ be a reversible operad and $Q$ its mated species

1975: obtained by applying the mating functor (Section~\ref{s:mf}).

1976: We will first define the algebraic objects $QA$ and $PA$.

1977: Using a calculus of cuttings and matings,

1978: we will give $QA$ a Lie algebra structure.

1979: The example of the commutative operad $(P=c)$

1980: was discussed in (\ref{ss:tcc}).

1981: A trivial looking but important example

1982: is the unit operad $(P=u)$.

1983: The Lie algebra in this case is 

1984: the usual symplectic Lie algebra $\fsp$.

1985: This is explained in (\ref{ss:sl}).

1986: 

1987: As in the case of the underlying vector space $V$, 

1988: we will sometimes write $QA_n$ and $PA_n$,

1989: to show the dependence of $n$ explicitly.

1990: The Lie algebra $QA_{\infty}$ will be defined

1991: as a direct limit $\underset{n \to \infty}{\lim} QA_n$.

1992: After this, we will briefly explain 

1993: the Hopf algebra structure on $H_*(QA_{\infty})$.

1994: 

1995: 

1996: \subsection{The free algebras} \label{ss:fa}

1997: 

1998: We now define the free algebra $QA$ of a species $Q$

1999: on generators $x_1, \ldots,x_n$.

2000: This also includes the case of $PA$,

2001: since every operad is a species.

2002: 

2003: \medskip

2004: \begin{center}

2005: \begin{tabular}{|c|c|c|c|} \hline 

2006: Operad $P$ & An element of $PA$ & 

2007: Species $Q$ & An element of $QA$ \\ \hline \hline

2008: $c$ & 

2009: \begin{minipage}{.4 in}

2010: \vspace{.1in}

2011: \begin{center}

2012: \psfrag{a}{\Huge $x_2$}

2013: \psfrag{b}{\Huge $x_1$}

2014: \psfrag{c}{\Huge $x_3$}

2015: \psfrag{d}{\Huge $x_1$}

2016: \resizebox{.4 in}{.4 in}{\includegraphics{csubt.eps}}

2017: \end{center}

2018: \vspace{.1in}

2019: \end{minipage} &

2020: $cc$ &

2021: \begin{minipage}{.4 in}

2022: \vspace{.1in}

2023: \begin{center}

2024: \psfrag{a}{\Huge $x_2$}

2025: \psfrag{b}{\Huge $x_1$}

2026: \psfrag{c}{\Huge $x_3$}

2027: \psfrag{d}{\Huge $x_1$}

2028: \resizebox{.4 in}{.4 in}{\includegraphics{cc.eps}}

2029: \end{center}

2030: \vspace{.1in}

2031: \end{minipage} \\ \hline

2032: $a$ &  

2033: \begin{minipage}{.4 in}

2034: \vspace{.1in}

2035: \begin{center}

2036: \psfrag{a}{\Huge $x_2$}

2037: \psfrag{b}{\Huge $x_1$}

2038: \psfrag{c}{\Huge $x_3$}

2039: \psfrag{d}{\Huge $x_1$}

2040: \resizebox{.4 in}{.4 in}{\includegraphics{asubt.eps}}

2041: \end{center}

2042: \vspace{.1in}

2043: \end{minipage} &

2044: $aa$ &

2045: \begin{minipage}{.4 in}

2046: \vspace{.1in}

2047: \begin{center}

2048: \psfrag{a}{\Huge $x_2$}

2049: \psfrag{b}{\Huge $x_1$}

2050: \psfrag{c}{\Huge $x_3$}

2051: \psfrag{d}{\Huge $x_1$}

2052: \resizebox{.4 in}{.4 in}{\includegraphics{aa.eps}}

2053: \end{center}

2054: \vspace{.1in}

2055: \end{minipage} \\ \hline

2056: $t$ &

2057: \begin{minipage}{.6 in}

2058: \vspace{.1in}

2059: \begin{center}

2060: \psfrag{a}{\Huge $x_2$}

2061: \psfrag{b}{\Huge $x_1$}

2062: \psfrag{c}{\Huge $x_3$}

2063: \psfrag{d}{\Huge $x_1$}

2064: \resizebox{.6 in}{.4 in}{\includegraphics{tsubt.eps}}

2065: \end{center}

2066: \vspace{.1in}

2067: \end{minipage} &

2068: $tt$ &

2069: \begin{minipage}{.6 in}

2070: \vspace{.1in}

2071: \begin{center}

2072: \psfrag{a}{\Huge $x_2$}

2073: \psfrag{b}{\Huge $x_1$}

2074: \psfrag{c}{\Huge $x_3$}

2075: \psfrag{d}{\Huge $x_1$}

2076: \resizebox{.6 in}{.4 in}{\includegraphics{tt.eps}}

2077: \end{center}

2078: \vspace{.1in}

2079: \end{minipage} \\ \hline

2080: \end{tabular}

2081: \end{center}

2082: \vspace{0.3 cm}

2083: \noindent

2084: To get an element of $QA$,

2085: take any element of $Q[I]$,

2086: for some finite non-empty set $I$

2087: and replace each element of $I$ by one of the generators

2088: $x_1, \ldots,x_n$.

2089: Call such an element a monomial.

2090: To get a general element of $QA$,

2091: take linear combinations of monomials.

2092: \[

2093: \begin{minipage}{.8 in}

2094: \begin{center}

2095: \psfrag{a}{\Huge $a$}

2096: \psfrag{b}{\Huge $b$}

2097: \psfrag{c}{\Huge $c$}

2098: \psfrag{d}{\Huge $d$}

2099: \resizebox{.6 in}{.6 in}{\includegraphics{q.eps}}

2100: \end{center}

2101: \end{minipage}

2102: \in Q[I] \qquad \rightsquigarrow \qquad

2103: \begin{minipage}{.8 in}

2104: \begin{center}

2105: \psfrag{x1}{\Huge $x_1$}

2106: \psfrag{x2}{\Huge $x_2$}

2107: \psfrag{x4}{\Huge $x_4$}

2108: \resizebox{.8 in}{.6 in}{\includegraphics{qa.eps}}

2109: \end{center}

2110: \end{minipage}

2111: \in QA^4.

2112: \]

2113: We may also write this as

2114: $QA = \bigoplus_{j\geq 2} (Q[j] \otimes V^{\otimes j})_{\Sigma_j}$,

2115: where $\Sigma_j$ acts on $Q[j]$

2116: by permuting letters and

2117: on $V^{\otimes j}$ by permuting the tensor factors.

2118: Strictly speaking, the grading starts in degree $1$.

2119: But since we are mainly interested in mated species,

2120: we started the grading in degree $2$.

2121: We will sometimes write $QA^j$ for the degree $j$ piece.

2122: Though we call $QA$ the free algebra of $Q$,

2123: it only has the structure of a graded vector space.

2124: 

2125: 

2126: If the species $Q$ has the structure of an operad $P$

2127: then we emphasise it by drawing instead

2128: \[

2129: \begin{minipage}{.8 in}

2130: \begin{center}

2131: \psfrag{a}{\Huge $a$}

2132: \psfrag{b}{\Huge $b$}

2133: \psfrag{c}{\Huge $c$}

2134: \psfrag{d}{\Huge $d$}

2135: \resizebox{.7 in}{.6 in}{\includegraphics{p.eps}}

2136: \end{center}

2137: \end{minipage}

2138: \in P[I] \qquad \rightsquigarrow \qquad

2139: \begin{minipage}{.8 in}

2140: \begin{center}

2141: \psfrag{x1}{\Huge $x_1$}

2142: \psfrag{x2}{\Huge $x_2$}

2143: \psfrag{x4}{\Huge $x_4$}

2144: \resizebox{.7 in}{.6 in}{\includegraphics{pa.eps}}

2145: \end{center}

2146: \end{minipage}

2147: \in PA^4.

2148: \]

2149: Just as for a species, we get

2150: $PA = \bigoplus_{j\geq 1} (P[j] \otimes V^{\otimes j})_{\Sigma_j}.$

2151: However, the situation for operads is more interesting.

2152: The substitution rule for operads allows us

2153: to define binary (and higher) products on $PA$.

2154: Hence $PA$ looks more like the algebras 

2155: that we are used to.

2156: For example,

2157: $cA=$ free non-unital commutative algebra and

2158: $aA=$ free non-unital associative algebra

2159: on the generators $x_1, \ldots,x_n$.

2160: 

2161: As this discussion suggests,

2162: it is possible to give a more abstract definition of $PA$.\

2163: To every operad $P$, one can associate a category of $P$-algebras

2164: (\ref{ss:om});

2165: the object $PA$ is then the free object,

2166: with respect to $V$, in this category.

2167: 

2168: \begin{remark}

2169: Recall that, roughly speaking, 

2170: the mated species

2171: $Q$ was defined as a quotient of $P \otimes P$,

2172: see (\ref{ss:mf}).

2173: In the same way, one can define

2174: $QA$ as $PA \otimes PA$ subject to the two relations:

2175: \begin{equation}  \label{e:r}

2176: a \otimes b = b \otimes a  \quad \text{and} \quad

2177: a \rightarrow b \otimes c = a \otimes r(b) \leftarrow c.

2178: \end{equation}

2179: \end{remark}

2180: 

2181: \subsubsection*{Examples}  

2182: 

2183: Refer to the table in (\ref{ss:fa}).

2184: In the commutative case, the pictures for $c$ and $cc$

2185: contain the same amount of information,

2186: they represent a monomial in the commutative variables

2187: $x_1,x_2,\ldots,x_n$.

2188: In the associative case, the pictures for $a$ and $aa$

2189: are different.

2190: They represent linear and cyclic orders respectively

2191: in non-commutative variables

2192: $x_1,x_2,\ldots,x_n$.

2193: In the tree case, the objects $PA$ and $QA$ are 

2194: best described using pictures rather than words.

2195: 

2196: Following the analogy with the commutative case

2197: discussed in (\ref{ss:tcc}),

2198: we now discuss the calculus of cuttings and matings 

2199: in the general case.

2200: The reader may keep the above examples in mind.

2201: 

2202: \subsection{Cutting}  \label{ss:cut}

2203: We define

2204: $\frac{\partial }{\partial x_i} : QA \longrightarrow PA$

2205: by showing how it works on a schematic example.

2206: \[

2207: \begin{minipage}{.9 in}

2208: \begin{center}

2209: \psfrag{x1}{\Huge $x_1$}

2210: \psfrag{x2}{\Huge $x_2$}

2211: \psfrag{x4}{\Huge $x_4$}

2212: \resizebox{.8 in}{.6 in}{\includegraphics{qa.eps}}

2213: \end{center}

2214: \end{minipage}

2215: \qquad \buildrel \frac{\partial}{\partial x_1}\over \longmapsto \qquad

2216: \begin{minipage}{.9 in}

2217: \begin{center}

2218: \psfrag{x1}{\Huge $x_1$}

2219: \psfrag{x2}{\Huge $x_2$}

2220: \psfrag{x4}{\Huge $x_4$}

2221: \resizebox{.8 in}{.6 in}{\includegraphics{par.eps}}

2222: \end{center}

2223: \end{minipage}

2224: \ \ + \ \

2225: \begin{minipage}{.9 in}

2226: \begin{center}

2227: \psfrag{x1}{\Huge $x_1$}

2228: \psfrag{x2}{\Huge $x_2$}

2229: \psfrag{x4}{\Huge $x_4$}

2230: \resizebox{.8 in}{.6 in}{\includegraphics{pas.eps}}

2231: \end{center}

2232: \end{minipage}

2233: .

2234: \]

2235: Namely, to define

2236: $\frac{\partial}{\partial x_1}$,

2237: cut all inputs with label $x_i$,

2238: one at a time.

2239: The justification for such a definition 

2240: is given by Proposition~\ref{p:pd}

2241: and the above remark.

2242: 

2243: As an example, in the associative case,

2244: the derivative with respect to $x_1$ of

2245: the cyclic word $x_1 x_4 x_1 x_2$ is the sum of

2246: two linear words, $x_4 x_1 x_2$ and $x_2 x_1 x_4$.

2247: 

2248: \subsection{Mating}  \label{ss:mate}

2249: In order to perform matings, 

2250: one requires an even number of characters.

2251: So assume that

2252: $V$ is even dimensional and has a basis

2253: $p_1, \ldots,p_n,q_1, \ldots,q_n$.

2254: Define the Poisson bracket

2255: $\{\ ,\ \} : QA \otimes QA \longrightarrow QA$

2256: using the formula

2257: \begin{equation} \label{e:ptb}

2258: \{F,H\}=\sum_{i=1}^n \frac{\partial F}{\partial p_i} \otimes \frac{\partial H}{\partial q_i} - 

2259: \frac{\partial F}{\partial q_i} \otimes \frac{\partial H}{\partial p_i}.

2260: \end{equation}

2261: In (\ref{ss:am}), we explained how the tensor sign 

2262: can be interpreted as a mating.

2263: The only difference now is that the objects being mated,

2264: instead of being labelled by elements of a set $I$,

2265: are labelled by elements of a vector space $V$.

2266: So one can continue with the same interpretation.

2267: \[

2268: \left\{

2269: \begin{minipage}{.8 in}

2270: \begin{center}

2271: \psfrag{a}{\Huge $p_2$}

2272: \psfrag{c}{\Huge $p_3$}

2273: \psfrag{d}{\Huge $q_3$}

2274: \psfrag{b}{\Huge $q_1$}

2275: \resizebox{.7 in}{.6 in}{\includegraphics{qaf.eps}}

2276: \end{center}

2277: \end{minipage}

2278: \ \ , \ \

2279: \begin{minipage}{.9 in}

2280: \begin{center}

2281: \psfrag{p1}{\Huge $p_1$}

2282: \psfrag{p2}{\Huge $p_2$}

2283: \psfrag{q1}{\Huge $q_1$}

2284: \psfrag{q2}{\Huge $q_2$}

2285: \resizebox{.8 in}{.6 in}{\includegraphics{qas.eps}}

2286: \end{center}

2287: \end{minipage}

2288: \right\}

2289: \ \ = \ \

2290: -

2291: \begin{minipage}{1.4 in}

2292: \begin{center}

2293: \psfrag{a}{\Huge $p_2$}

2294: \psfrag{c}{\Huge $p_3$}

2295: \psfrag{d}{\Huge $q_3$}

2296: \psfrag{x}{\Huge $p_2$}

2297: \psfrag{z}{\Huge $q_1$}

2298: \resizebox{1.2 in}{.6 in}{\includegraphics{qast.eps}}

2299: \end{center}

2300: \end{minipage}

2301: .

2302: \]

2303: In the above example,

2304: there is only one mating possible.

2305: So there is only one term on the right hand side.

2306: The minus sign reflects the fact that 

2307: the ``$q$'' was cut from the first term and

2308: the ``$p$'' from the second term.

2309: 

2310: \begin{proposition}  \label{p:pb}

2311: The bracket $\{\ ,\ \}$ equips $QA$ 

2312: with the structure of a Lie algebra.

2313: \end{proposition}

2314: 

2315: \noindent

2316: This can be verified directly.

2317: It will also follow from the symplectic operad theory

2318: that we will present in Section~\ref{s:st}.

2319: So we assume this result for the moment.

2320: Note that the Lie algebra $QA$ is graded

2321: and the bracket $\{\ ,\ \}$ is a map of degree $-2$.

2322: 

2323: \begin{remark}

2324: If the reversible operad $P$ is based on sets

2325: then $QA$ also has a Lie coalgebra structure 

2326: defined using breakups (rather than matings).

2327: This parallels the definition of the coboundary operator

2328: on graphs (\ref{ss:stacb}).

2329: \end{remark}

2330: 

2331: \subsection{Connection to the symplectic Lie algebra $\fsp$}  \label{ss:sl}

2332: 

2333: We now look at an important example

2334: that is already present in the classical theory.

2335: This is the simplest case, i.e.,

2336: $P=u$ and $Q=uu$,

2337: see (\ref{ss:uu}).

2338: It is easy to see that $uA_n=V$, 

2339: the underlying vector space and

2340: $uuA_n=$ homogeneous polynomials of degree 2

2341: in the variables $p_1, \ldots,p_n,q_1, \ldots,q_n$.

2342: In this case, $\{\ ,\ \}$ is the usual Poisson bracket

2343: with the tensor sign in equation~\eqref{e:ptb}

2344: being ordinary multiplication.

2345: 

2346: Going to the general case,

2347: note that $uuA$ is always a subalgebra of $QA$ 

2348: that sits inside $QA^2$, the piece of degree 2.

2349: This is because $uu$ is a subspecies of $Q$,

2350: see (\ref{ss:uu}).

2351: Moreover, since the bracket $\{\ ,\ \}$ has degree $-2$,

2352: it restricts to a map

2353: $QA^j \otimes uuA \longrightarrow QA^j$.

2354: Hence each graded piece $QA^j$ is a right $uuA$ module.

2355: As an example,

2356: \[

2357: \left\{

2358: \begin{minipage}{.9 in}

2359: \begin{center}

2360: \psfrag{p1}{\Huge $p_1$}

2361: \psfrag{p2}{\Huge $p_2$}

2362: \psfrag{q1}{\Huge $q_1$}

2363: \psfrag{q2}{\Huge $q_2$}

2364: \resizebox{.8 in}{.6 in}{\includegraphics{qas.eps}}

2365: \end{center}

2366: \end{minipage}

2367: \ \ , \ \

2368: \begin{minipage}{.8 in}

2369: \begin{center}

2370: \psfrag{p1}{\Huge $p_1$}

2371: \psfrag{p2}{\Huge $p_2$}

2372: \psfrag{q1}{\Huge $q_1$}

2373: \psfrag{q2}{\Huge $q_2$}

2374: \psfrag{1}{}

2375: \psfrag{2}{}

2376: \psfrag{uu}{}

2377: \resizebox{.7 in}{.15 in}{\includegraphics{sf.eps}}

2378: \end{center}

2379: \end{minipage}

2380: \right\}

2381: \ \ = \ \

2382: \begin{minipage}{.9 in}

2383: \begin{center}

2384: \psfrag{p1}{\Huge $p_1$}

2385: \psfrag{p2}{\Huge $p_2$}

2386: \psfrag{q1}{\Huge $q_1$}

2387: \psfrag{q2}{\Huge $q_2$}

2388: \resizebox{.8 in}{.6 in}{\includegraphics{qasr.eps}}

2389: \end{center}

2390: \end{minipage}

2391: \ \ - \ \

2392: \begin{minipage}{.9 in}

2393: \begin{center}

2394: \psfrag{p1}{\Huge $p_1$}

2395: \psfrag{p2}{\Huge $p_2$}

2396: \psfrag{q1}{\Huge $q_1$}

2397: \psfrag{q2}{\Huge $q_2$}

2398: \resizebox{.8 in}{.6 in}{\includegraphics{qass.eps}}

2399: \end{center}

2400: \end{minipage}

2401: .

2402: \]

2403: Observe that the action only involves changing labels.

2404: It does not change the internal structure 

2405: of the vertex.

2406: This is because the matings are all trivial.

2407: 

2408: The symplectic Lie algebra $\fsp$ is defined

2409: as the space of linear maps on $V$

2410: that kill the symplectic form

2411: $\sum dp_i \wedge dq_i$

2412: (the action is as a derivation);

2413: the bracket being the usual commutator.

2414: Note that

2415: $QA^j = (Q[j] \otimes V^{\otimes j})_{\Sigma_j}$

2416: is a left $\fsp$ module with the action 

2417: induced by the left action on $V$ and

2418: the trivial action on $Q[j]$.

2419: 

2420: We now explain the relation between the Lie algebras

2421: $uuA_n$ and $\fsp$.

2422: 

2423: \begin{proposition}  \label{p:degtwo}

2424: There is a Lie algebra anti-isomorphism

2425: $uuA_n \to \fsp$ given by $H \mapsto \xi_H$,

2426: where

2427: $\xi_H(p_i)=\frac{\partial H}{\partial q_i}$ and

2428: $\xi_H(q_i)=-\frac{\partial H}{\partial p_i}$.

2429: \end{proposition}

2430: 

2431: \noindent

2432: Again this can be checked directly

2433: but will be a consequence of the general theory

2434: (Section~\ref{s:st}).

2435: The definition of the module structures on $QA$

2436: together with the above proposition gives us the following.

2437: 

2438: \begin{corollary}

2439: The right $uuA_n$ and the left $\fsp$ 

2440: module structures on $QA^j$ are compatible

2441: via the above anti-isomorphism $uuA_n \to \fsp$.

2442: In other words,

2443: $\{F,H\} = \xi_H \cdot F$ for $F \in QA^j$

2444: and $H \in uuA_n$.

2445: \end{corollary}

2446: 

2447: 

2448: 

2449: \subsection{The Lie algebra $QA_{\infty}$}

2450: \label{ss:ha}

2451: 

2452: We have the underlying vector space 

2453: $V_{n} \subset V_{n+1}$, with

2454: the basis of $V_{n+1}$ extending the basis on $V_n$ to

2455: $p_1, \ldots,p_{n+1},q_1, \ldots,q_{n+1}$.

2456: This gives us a sequence of Lie algebra inclusions

2457: $$QA_1 \subset \ldots \subset QA_n \subset QA_{n+1} \subset \ldots$$

2458: We denote the direct limit by $QA_{\infty}$.

2459: 

2460: A Hopf algebra is a vector space 

2461: which has compatible algebra and coalgebra structures

2462: and an inverse which is called the antipode.

2463: For basic information on Hopf algebras,

2464: see~\cite{\sweedler,\kassel}.

2465: The graded vector space $H_*(QA_{\infty})$

2466: has the structure of a Hopf algebra,

2467: which we now explain.

2468: 

2469: To describe the product on $H_*(QA_{\infty})$,

2470: we start with a morphism of Lie algebras

2471: $QA_n \oplus QA_m \into QA_{n+m}$, defined by

2472: $(F,H) \mapsto F+H$,

2473: where we shift up the indices of $H$ by $n$.

2474: In other words, we think of

2475: $QA_n$ and $QA_m$ as disjoint Lie subalgebras of $QA_{n+m}$.

2476: This defines a map

2477: $H_*(QA_{n}) \otimes H_*(QA_{m}) \to H_*(QA_{n+m})$.

2478: Taking direct limits, we get a product on $H_*(QA_{\infty})$.

2479: We point out that the diagram

2480: \begin{equation*}

2481: \begin{CD}

2482: QA_{n} \oplus QA_{m} @>>> QA_{n+m} \\

2483: @VVV @VVV \\

2484: QA_{n+1} \oplus QA_{m} @>>> QA_{n+m+1} \\

2485: \end{CD}

2486: \end{equation*}

2487: commutes only upto index shifting.

2488: However, this cannot be detected on the homology level.

2489: 

2490: The coproduct is relatively straightforward to define.

2491: We start with the diagonal map

2492: $QA_{\infty} \to QA_{\infty} \oplus QA_{\infty}$,

2493: which is a morphism of Lie algebras.

2494: This induces a cocommutative coproduct on

2495: $H_*(QA_{\infty})$.

2496: 

2497: 

2498: 

2499: 

2500: 

2501: 

2502: \section{Symplectic operad theory}  \label{s:st}

2503: 

2504: In this section, we give a self-contained treatment of 

2505: Kontsevich's symplectic mini-theory.

2506: But in order to appreciate it,

2507: it is important to be familiar with

2508: classical symplectic theory.

2509: This involves the principles of Hamiltonian mechanics

2510: \cite{\mcduff} and

2511: basic notions of differential topology such as

2512: vector fields, differential forms,

2513: Lie derivatives and contraction operators

2514: \cite{\boothby,\spivak}.

2515: 

2516: For some recent ideas and applications 

2517: in the associative case, see the papers of

2518: Ginzburg~\cite{\ginz} and Bocklandt and Le Bruyn~\cite{\bl}. 

2519: For a detailed exposition,

2520: see Chapter $7$ of Le Bruyn's book ``Non-commutative geometry'',

2521: which is available on his homepage.

2522: 

2523: The basic objects of interest are summarised in the table.

2524: We have already seen $PA$ and $QA$ before.

2525: The remaining ones will be defined in this section.

2526: \medskip

2527: \begin{center}

2528: \begin{tabular}{|c|c|c|} \hline 

2529: Geometric objects & \multicolumn2{c|}{Algebraic objects} \\ \hline \hline

2530: $X$ & $PA$ & $QA$ \\ 

2531: $\mathfrak X(X)$ & $\Der(PA)$ & $\Der(QA)$ \\ 

2532: $\Pi TX$ & $\Omega(PA)$ & $\Omega(QA)$ \\ 

2533: $\mathfrak X(\Pi TX)$ & $\Der(\Omega(PA))$ & $\Der(\Omega(QA))$ \\ \hline

2534: \end{tabular}

2535: \end{center}

2536: 

2537: \medskip

2538: \begin{intuition}   

2539: The geometric objects in the first column 

2540: do not exist except in the commutative case $(P=c)$

2541: in which case $X=\R^n$.

2542: One may say that they are defined 

2543: via their algebraic counterparts.

2544: It is useful to have the following dictionary:

2545: 

2546: $X=$ operad manifold,

2547: $\mathfrak X(X)=$ vector fields on $X$,

2548: $\Pi TX=$ total space of the odd tangent bundle to $X$,

2549: $\mathfrak X(\Pi TX)=$ vector fields on $\Pi TX$.

2550: \end{intuition}

2551: 

2552: Let $X$ be an operad manifold with

2553: the free $P$-algebra $PA$ as its algebra of functions

2554: (\ref{ss:om}).

2555: Starting with this purely algebraic data,

2556: we want to construct an algebra of

2557: ``differential forms on X'', 

2558: which is defined as some differential $P$-superalgebra

2559: $\Omega(PA)$.

2560: One would guess that the symplectic form $\omega$ should be 

2561: a suitable element of $\Omega^2(PA)$.

2562: However, this is not true.

2563: 

2564: It turns out that one needs to consider 

2565: $\Omega(QA)$, which is the corresponding object for $QA$.

2566: And the symplectic form $\omega$ will lie in $\Omega^2(QA)$.

2567: So in this sense,

2568: it is $QA$ and $\Omega(QA)$, which should be regarded

2569: as functions and forms on $X$.

2570: 

2571: \subsection{Algebraic objects for an operad}  \label{ss:ao}

2572: 

2573: We first look more generally at operad manifolds

2574: (\ref{ss:om}).

2575: To every operad $P$, one can associate a category of $P$-algebras.

2576: Similarly, one can consider the graded version,

2577: which is the category of $P$-superalgebras.

2578: And one can talk of derivations of $P$-algebras

2579: or superderivations of $P$-superalgebras~\cite{\gk}.

2580: 

2581: In what follows, we will freely use this language.

2582: For example,

2583: $PA$ is the free object in the category

2584: of $P$-algebras.

2585: However, just as we did for $PA$,

2586: we will also give explicit descriptions

2587: of all the objects we deal with.

2588: This should help the reader 

2589: who is unfamiliar with the above language.

2590: We also recommend that the reader keep

2591: a concrete case, say commutative or associative,

2592: in mind.

2593: 

2594: \begin{definition}   

2595: Let $\Der(PA)$ be the space of derivations of $PA$.

2596: It is a Lie algebra with the bracket being the commutator.

2597: \[

2598: \begin{minipage}{.8 in}

2599: \begin{center}

2600: \psfrag{x1}{\Huge $x_1$}

2601: \psfrag{x2}{\Huge $x_2$}

2602: \resizebox{.7 in}{.7 in}{\includegraphics{pad.eps}}

2603: \end{center}

2604: \end{minipage}

2605: \qquad \buildrel {\xi}\over \longmapsto \qquad

2606: \begin{minipage}{.8 in}

2607: \begin{center}

2608: \psfrag{x1}{\Huge $\xi(x_1)$}

2609: \psfrag{x2}{\Huge $x_2$}

2610: \resizebox{.7 in}{.7 in}{\includegraphics{pad.eps}}

2611: \end{center}

2612: \end{minipage}

2613: \ \ + \ \

2614: \begin{minipage}{.8 in}

2615: \begin{center}

2616: \psfrag{x1}{\Huge $x_1$}

2617: \psfrag{x2}{\Huge $\xi(x_2)$}

2618: \resizebox{.7 in}{.7 in}{\includegraphics{pad.eps}}

2619: \end{center}

2620: \end{minipage}

2621: .

2622: \]

2623: \end{definition}

2624: %

2625: \noindent

2626: A derivation ${\xi} \in \Der(PA)$ is

2627: an operator on $PA$.

2628: It is uniquely determined by its values 

2629: ${\xi}(x_1),\ldots,{\xi}(x_n) \in PA$ on the generators

2630: $x_1,\ldots,x_n$.

2631: An example of how this works is shown above;

2632: for each input, we replace

2633: the value $x_i$ on it by $\xi(x_i)$.

2634: In the first term on the right,

2635: $\xi(x_1)$ is written on one input.

2636: This is to be understood as a substitution of $\xi(x_1)$

2637: into that input.

2638: 

2639: \begin{definition}   

2640: The algebra of forms $\Omega(PA)$ 

2641: is the free differential superalgebra 

2642: on the generators $x_1, \ldots,x_n$.

2643: The term ``differential'' means that

2644: there is a superderivation $d:\Omega(PA) \to \Omega(PA)$ 

2645: of odd degree such that $d^2=0$.

2646: \end{definition}

2647: %\noindent

2648: The algebra of forms $\Omega(PA)$ is freely generated 

2649: as a $P$-superalgebra by the symbols 

2650: $x_1, \ldots,x_n, dx_1, \ldots,dx_n$.

2651: Hence a superderivation on $\Omega(PA)$ is uniquely determined

2652: by its values on these generators.

2653: Note that the differential $d$ is specified by

2654: $d(x_i)=dx_i$ and $d(dx_i)=0$.

2655: 

2656: Just as for $PA$,

2657: one can give an explicit description of $\Omega(PA)$.

2658: To get a monomial in $\Omega(PA)$,

2659: take any element of $P[I]$,

2660: for some finite non-empty set $I$

2661: and replace each element of $I$ by one of the generators

2662: $x_1, \ldots,x_n,dx_1, \ldots,dx_n$.

2663: Since we are now dealing with a super-object,

2664: we need to be more careful.

2665: We order the inputs with differential symbols 

2666: in the sense of orientation as follows.

2667: \begin{equation}  \label{e:dpa}

2668: \begin{minipage}{.9 in}

2669: \begin{center}

2670: \psfrag{x1}{\Huge $x_1$}

2671: \psfrag{x2}{\Huge $x_2$}

2672: \psfrag{x3}{\Huge $x_3$}

2673: \psfrag{dx2}{\Huge $dx_2$}

2674: \psfrag{dx1}{\Huge $dx_1$}

2675: \psfrag{1}{\Huge $1$}

2676: \psfrag{2}{}

2677: \psfrag{3}{\Huge $2$}

2678: \psfrag{4}{}

2679: \resizebox{.9 in}{.7 in}{\includegraphics{dpa.eps}}

2680: \end{center}

2681: \end{minipage}

2682: \qquad = \qquad - \ \ \ \ 

2683: \begin{minipage}{.9 in}

2684: \begin{center}

2685: \psfrag{x1}{\Huge $x_1$}

2686: \psfrag{x2}{\Huge $x_2$}

2687: \psfrag{x3}{\Huge $x_3$}

2688: \psfrag{dx2}{\Huge $dx_2$}

2689: \psfrag{dx1}{\Huge $dx_1$}

2690: \psfrag{1}{\Huge $2$}

2691: \psfrag{2}{}

2692: \psfrag{3}{\Huge $1$}

2693: \psfrag{4}{}

2694: \resizebox{.9 in}{.7 in}{\includegraphics{dpa.eps}}

2695: \end{center}

2696: \end{minipage}

2697: .

2698: \end{equation}

2699: In other words, 

2700: an even permutation of the order leaves an element unchanged

2701: while an odd permutation gives its negative.

2702: We now explain how the differential $d$ works on this example.

2703: \[

2704: \begin{minipage}{.9 in}

2705: \begin{center}

2706: \psfrag{x1}{\Huge $x_1$}

2707: \psfrag{x2}{\Huge $x_2$}

2708: \psfrag{x3}{\Huge $x_3$}

2709: \psfrag{dx2}{\Huge $dx_2$}

2710: \psfrag{dx1}{\Huge $dx_1$}

2711: \psfrag{1}{\Huge $1$}

2712: \psfrag{2}{}

2713: \psfrag{3}{\Huge $2$}

2714: \psfrag{4}{}

2715: \resizebox{.9 in}{.7 in}{\includegraphics{dpa.eps}}

2716: \end{center}

2717: \end{minipage}

2718: \qquad \buildrel {d}\over \longmapsto \qquad

2719: \begin{minipage}{.9 in}

2720: \begin{center}

2721: \psfrag{x1}{\Huge $x_1$}

2722: \psfrag{x2}{\Huge $dx_2$}

2723: \psfrag{x3}{\Huge $x_3$}

2724: \psfrag{dx2}{\Huge $dx_2$}

2725: \psfrag{dx1}{\Huge $dx_1$}

2726: \psfrag{1}{\Huge $2$}

2727: \psfrag{2}{\Huge $1$}

2728: \psfrag{3}{\Huge $3$}

2729: \psfrag{4}{}

2730: \resizebox{.9 in}{.7 in}{\includegraphics{dpa.eps}}

2731: \end{center}

2732: \end{minipage}

2733: \ \ + \ \

2734: \begin{minipage}{.9 in}

2735: \begin{center}

2736: \psfrag{x1}{\Huge $x_1$}

2737: \psfrag{x2}{\Huge $x_2$}

2738: \psfrag{x3}{\Huge $dx_3$}

2739: \psfrag{dx2}{\Huge $dx_2$}

2740: \psfrag{dx1}{\Huge $dx_1$}

2741: \psfrag{1}{\Huge $2$}

2742: \psfrag{2}{}

2743: \psfrag{3}{\Huge $3$}

2744: \psfrag{4}{\Huge $1$}

2745: \resizebox{.9 in}{.7 in}{\includegraphics{dpa.eps}}

2746: \end{center}

2747: \end{minipage}

2748: .

2749: \]

2750: In general, we replace each $x_i$ by a $dx_i$,

2751: one at a time.

2752: And the input at which the replacement occurs

2753: is given label $1$ and the remaining labels

2754: are shifted up by $1$.

2755: 

2756: It is clear that $\Omega(PA)$ is in fact $\Z$ graded

2757: with the grading given by the number of differential symbols

2758: and $\Omega^0(PA)=PA$.

2759: 

2760: 

2761: \begin{definition}   

2762: Let $\Der(\Omega(PA))$ be the space 

2763: of superderivations of $\Omega(PA)$.

2764: It is a Lie superalgebra with the bracket being the supercommutator.

2765: \end{definition}

2766: 

2767: To act by a superderivation $L$

2768: on an element of $\Omega(PA)$,

2769: one replaces the value $v$ at each input by $L(v)$,

2770: one at a time.

2771: If $v=x_i$ then we shift the labels by $\degg(L)$.

2772: If $v=dx_i$ then we first reorder such that the $dx_i$ 

2773: being substituted into has label $1$ and then 

2774: shift the remaining labels by $\degg(L)$.

2775: The special case when $L$ is the differential $d$ was shown above.

2776: 

2777: \begin{definition}   

2778: For any derivation ${\xi} \in \Der(PA)$,

2779: we define the Lie derivative $L_{\xi} \in \Der(\Omega(PA))$ 

2780: and the contraction operator $i_{\xi} \in \Der(\Omega(PA))$ 

2781: by specifying them on the generators as under

2782: $$L_{\xi}(x_i)={\xi}(x_i), \ L_{\xi}(dx_i)=d{\xi}(x_i) \ \ \text{and} \ \

2783: i_{\xi}(x_i)=0, \ i_{\xi}(dx_i)={\xi}(x_i).$$

2784: They have degrees $0$ and $-1$ respectively.

2785: \end{definition} 

2786: 

2787: \begin{lemma} \label{l:rel}

2788: For derivations ${\xi}, {\eta} \in \Der(PA)$,

2789: the superderivations $L_{\xi}, i_{\xi}, d$ satisfy

2790: the following relations.

2791: \begin{itemize}

2792: \item[(1)]

2793: $[i_{\xi},d] = L_{\xi}$ \ \ \ (Cartan's formula). 

2794: 

2795: \item[(2)]

2796: $[i_{\xi},i_{\eta}] = 0$.

2797: 

2798: \item[(3)]

2799: $[L_{\xi},i_{\eta}] = i_{[{\xi},{\eta}]}$.

2800: 

2801: \item[(4)]

2802: $[L_{\xi},L_{\eta}] =  L_{[{\xi},{\eta}]}$.

2803: \end{itemize}

2804: 

2805: \end{lemma}

2806: \begin{proof}

2807: As mentioned earlier, superderivations are preserved

2808: by the supercommutator.

2809: Hence in all cases, both sides are superderivations,

2810: so we need to check $(1)-(4)$ only on the generators

2811: $x_1, \ldots,x_n, dx_1, \ldots,dx_n$.

2812: This is a straightforward check that is

2813: independent of the operad $P$.

2814: \end{proof}

2815: 

2816: \begin{lemma} \label{l:exact}

2817: The algebra of forms $\Omega(PA)$ is exact.

2818: \end{lemma}

2819: \begin{proof}

2820: Consider the Euler vector field $e \in \Der(PA)$,

2821: defined by $e(x_i)=x_i$ for all $i$.

2822: Then $L_e(x_i)=x_i$ and $L_e(dx_i)=dx_i$

2823: and hence $L_e$ maps every monomial

2824: to a non-zero multiple of itself.

2825: Therefore it is invertible on $\Omega(PA)$.

2826: So it induces an isomorphism on cohomology.

2827: On the other hand, Cartan's formula

2828: (item $(1)$ in Lemma~\ref{l:rel}) shows that

2829: it induces the zero map on cohomology.

2830: So we conclude that $\Omega(PA)$ is exact.

2831: \end{proof}

2832: 

2833: For the operad $P=c$,

2834: the lemma just says that the algebra of polynomial forms on $\R^n$,

2835: that vanish at the origin, is exact.

2836: 

2837: \subsection{Algebraic objects for a species}  \label{ss:as}

2838: 

2839: Now assume that the operad $P$ is reversible;

2840: so we can talk of its mated species $Q$.

2841: Analogues for $Q$ of the objects in (\ref{ss:ao})

2842: can be defined without difficulty.

2843: As far as pictures are concerned,

2844: we draw the same ones as for operads

2845: except that the output arrow is now omitted.

2846: 

2847: We have seen that $QA$ can be written

2848: as a quotient of $PA \otimes PA$ (equation~\eqref{e:r}).

2849: Similarly, one can define $\Omega(QA)$

2850: as a quotient of $\Omega(PA) \otimes \Omega(PA)$ 

2851: subject to the super-relations:

2852: \begin{equation}  \label{e:sr}

2853: a \otimes b = (-1)^{|a||b|} b \otimes a

2854: \quad \text{and} \quad

2855: a \rightarrow b \otimes c = a \otimes r(b) \leftarrow c.

2856: \end{equation}

2857: %

2858: Just like $\Omega(PA)$,

2859: we see that $\Omega(QA)$ is $\Z$ graded

2860: and $\Omega^0(QA)=QA$.

2861: We also define $\Der(QA)$, $\Der(\Omega(QA))$ 

2862: and for $\xi \in \Der(PA)$,

2863: the operators

2864: $L_{\xi}, i_{\xi} \in \Der(\Omega(QA))$.

2865: The relations in Lemma~\ref{l:rel} continue to hold.

2866: Hence it follows that $\Omega(QA)$ 

2867: is also exact.

2868: 

2869: 

2870: Note that any element of $\Omega^1(QA)$

2871: can be uniquely written as

2872: $\sum_{i=1}^{n} f_i \otimes dx_i$, with

2873: $f_i \in PA$.

2874: And for $H \in QA$,

2875: we get $dH = \sum_{i=1}^{n} \frac{\partial H}{\partial x_i} \otimes dx_i$.

2876: These facts are again a consequence of Proposition~\ref{p:pd}

2877: and will be crucial in what follows.

2878: 

2879: 

2880: \subsection{The symplectic form $\omega$}  

2881: 

2882: Since $P$ has a unit element $u$,

2883: its mated species $Q$ has the special element $uu$

2884: in degree 2, see (\ref{ss:uu}).

2885: It allows us to define the symplectic form

2886: $\omega = \sum dp_i \otimes dq_i \in \Omega^2(QA)$.

2887: For a picture, we may draw 

2888: $\sum

2889: \begin{minipage}{.8 in}

2890: \begin{center}

2891: \psfrag{p1}{\Huge $dp_i$}

2892: \psfrag{q2}{\Huge $dq_i$}

2893: \psfrag{1}{\Huge $1$}

2894: \psfrag{2}{\Huge $2$}

2895: \psfrag{uu}{}

2896: \resizebox{.7 in}{.15 in}{\includegraphics{sf.eps}}

2897: \end{center}

2898: \end{minipage}

2899: $.

2900: Note that if we switch the order of the factors,

2901: we pick a minus sign (equation~\eqref{e:sr}).

2902: This can also be seen from the picture and

2903: the analogue of equation~\eqref{e:dpa} for species.

2904: 

2905: Also $dw=0$ and hence the form is closed.

2906: This gives us a symplectic operad manifold $X$.

2907: We think of $\Der(PA)$ as vector fields on $X$

2908: that vanish at a point and 

2909: $\Omega(QA)$ as differential forms on $X$ 

2910: with no constant or linear terms,

2911: with the $0$-forms $QA$ being Hamiltonian functions on $X$.

2912: 

2913: \begin{lemma} \label{l:iso}

2914: The map $\Der(PA) \to \Omega^1(QA)$ which sends

2915: ${\xi}$ to $i_{\xi} w$ is an isomorphism.

2916: \end{lemma}

2917: \begin{proof}

2918: We have the chain of equalities

2919: $$i_{\xi} w = \sum_{i=1}^{n} i_{\xi} (dp_i \otimes dq_i) =

2920: \sum_{i=1}^{n} {\xi}(p_i) \otimes dq_i - {\xi}(q_i) \otimes dp_i.$$

2921: This shows that $i_{\xi} w$ is determined by 

2922: ${\xi}(p_i)$ and ${\xi}(q_i)$, which also determine ${\xi}$.

2923: Since one can represent any element of $\Omega^1(QA)$ 

2924: uniquely by a sum of the form

2925: $\sum_{i=1}^{n} ( \ ) \otimes dp_i + ( \ ) \otimes dq_i$,

2926: the map in the lemma is an isomorphism.

2927: \end{proof}

2928: 

2929: \begin{definition}   

2930: Let $\Der(PA, w) = \{ {\xi} \in \Der(PA) \ | \ L_{\xi} w = 0 \}$.

2931: In analogy with the classical case, 

2932: we call these the symplectic vector fields.

2933: \end{definition}

2934: 

2935: It follows from Cartan's formula (Lemma~\ref{l:rel})

2936: that 

2937: $${\xi} \in \Der(PA, w) \Longleftrightarrow d(i_{\xi} w) = 0.$$

2938: Hence under the isomorphism of Lemma~\ref{l:iso},

2939: one sees that symplectic vector fields correspond 

2940: to closed 1 forms.

2941: Define Hamiltonian vector fields to be the ones that correspond 

2942: to exact 1 forms.

2943: However, since $\Omega(QA)$ is exact, 

2944: the Hamiltonian and symplectic vector fields coincide

2945: in this case.

2946: And there is an isomorphism

2947: $QA \buildrel d \over \to \Omega^1(QA)_{\text{closed}}$.

2948: Putting all facts together, 

2949: there is an isomorphism 

2950: $\xi : QA \isoto \Der(PA, w)$, 

2951: where $H \mapsto {\xi}_H$

2952: is defined by the equation 

2953: $dH = i_{{\xi_H}}(w)$.

2954: Writing both sides in the unique form

2955: $\sum_{i=1}^{n} ( \ ) \otimes dp_i + ( \ ) \otimes dq_i$,

2956: we derive the Hamiltonian equations

2957: \begin{equation}  \label{e:ham}

2958: {\xi_H}(p_i) = \frac{\partial H}{\partial q_i} \ \ \text{and} \ \ 

2959: {\xi_H}(q_i) = - \frac{\partial H}{\partial p_i}.

2960: \end{equation}

2961: We now derive the isomorphism between

2962: Hamiltonian functions and Hamiltonian vector fields.

2963: By Kontsevich kimaya,

2964: the proof is same as in the classical case;

2965: we give it here for completeness.

2966: 

2967: \begin{proposition}  

2968: The map $QA \to \Der(PA, w)$ defined above

2969: is an anti-isomorphism of Lie algebras,

2970: with the bracket on $QA$ being the Poisson bracket

2971: and on $\Der(PA, w)$ 

2972: being the usual commutator.

2973: \end{proposition}

2974: \begin{proof}

2975: 

2976: First note that $\Der(PA, w)$ is closed under

2977: taking commutators

2978: (item $(4)$ in Lemma~\ref{l:rel}).

2979: This gives it a Lie algebra structure.

2980: So to prove the proposition,

2981: we trace through the isomorphism 

2982: $QA \isoto \Der(PA, w)$,

2983: compute the induced Lie structure on $QA$

2984: and see that upto a minus sign 

2985: it coincides with the Poisson bracket

2986: (\ref{ss:mate}).

2987: We do the computation in two steps.

2988: 

2989: \subsection*{Claim $1$}

2990: $i_{[{\xi_F},{\xi_H}]}w = d(i_{\xi_F}(dH))$.

2991: $$

2992: \begin{array}{r c l l}

2993: i_{[{\xi_F},{\xi_H}]}w & = & [L_{\xi_F},i_{\xi_H}]w & \mbox{(item (3) in Lemma~\ref{l:rel})} \\

2994:              & = & L_{\xi_F}(i_{\xi_H} w) & (L_{\xi_F} w=0 \ \mbox{since} \ \xi_F \ \mbox{is symplectic}) \\

2995:              & = & L_{\xi_F}(dH) & \\

2996:              & = & [i_{\xi_F},d](dH) & \mbox{(item (1) in Lemma~\ref{l:rel})} \\

2997:              & = & d(i_{\xi_H} (dH)). &  \\

2998: \end{array}

2999: $$

3000: 

3001: \subsection*{Claim $2$}

3002: $i_{\xi_F}(dH) = - \{F,H\}$.\\

3003: $$

3004: \begin{array}{r c l l}

3005: i_{\xi_F}(dH) & = & i_{\xi_F}(\sum_{i=1}^{n} \frac{\partial H}{\partial p_i} \otimes dp_i+ \frac{\partial H}{\partial q_i} \otimes dq_i) & \\

3006:              & = & \sum_{i=1}^{n} \frac{\partial H}{\partial p_i} \otimes {\xi_F}(p_i) + \frac{\partial H}{\partial q_i} \otimes {\xi_F}(q_i) & \\

3007:              & = & \sum_{i=1}^n \frac{\partial H}{\partial p_i} \otimes \frac{\partial F}{\partial q_i} - \frac{\partial H}{\partial q_i} \otimes \frac{\partial F}{\partial p_i} & \mbox{(equation~\eqref{e:ham})} \\

3008:              & = & - \{F,H\}. & \mbox{(equation~\eqref{e:ptb})} \\

3009:                     

3010: \end{array}

3011: $$

3012: %

3013: The proposition now follows from the two claims.

3014: \end{proof}

3015: 

3016: As a corollary, we obtain Proposition~\ref{p:pb}.

3017: Also, if we specialise to $P=u$ and $Q=uu$,

3018: we obtain

3019: Proposition~\ref{p:degtwo}.

3020: 

3021: 

3022: \section{Examples motivated by PROPS}  \label{s:me}

3023: 

3024: We have, so far, only defined the notion of a derivative.

3025: If one wants to define higher order derivatives

3026: then one should consider a more general structure,

3027: namely, a PROP.

3028: For some motivation for PROPs

3029: from the viewpoint of homotopy theory, see~\cite{\adams}. 

3030: 

3031: \subsection{PROP}  \label{ss:prop}

3032: 

3033: We denote a PROP by $Pr$.

3034: It consists roughly of objects 

3035: with many inputs and many outputs.

3036: For a precise definition, 

3037: see the notes by Voronov~\cite{\voronov}. 

3038: We write $Pr[I,J]$ for the ``set of $Pr$-structures 

3039: on the set $I \sqcup J$ that have $I$ inputs and $J$ outputs''.

3040: There is also a composition law

3041: which allows to feed the outputs of one object

3042: to the inputs of another.

3043: \[

3044: \begin{minipage}{2 in}

3045: \begin{center}

3046: \resizebox{1.8 in}{.6 in}{\includegraphics{prsubt.eps}}

3047: \end{center}

3048: \end{minipage}

3049: =

3050: \begin{minipage}{1 in}

3051: \begin{center}

3052: \resizebox{.85 in}{.6 in}{\includegraphics{prsubo.eps}}

3053: \end{center}

3054: \end{minipage}.

3055: \]

3056: 

3057: It is useful to think of species and operads

3058: as parts of a PROP corresponding to 

3059: no output ($J$ empty) and

3060: a single output ($J$ singleton) respectively.

3061: We saw that a partial derivative (Proposition~\ref{p:pd})

3062: sends a species element to an operad element.

3063: Stated differently, it just converts 

3064: an input to an output.

3065: If this is to be the meaning of a derivative

3066: then to define higher derivatives

3067: one is forced to consider objects with many outputs.

3068: Thus one sees that PROPs provide a natural framework

3069: to talk of higher derivatives.

3070: To make the theory of higher matings work,

3071: one would need a ``reversible PROP''.

3072: We do not make these ideas precise.

3073: Instead, we look at some examples

3074: that fit this pattern.

3075: They may be of independent interest.

3076: 

3077: \subsection{The graph operad $g$ and species $gg$} \label{ss:g}

3078: 

3079: Let $g[I]=$ set of graphs with a specified source and sink vertex

3080: and directed edges labelled by elements of the set $I$.

3081: The figure shows an element of $g[\{a,b,c,d\}]$.

3082: \[

3083: \begin{minipage}{1.6 in}

3084: \begin{center}

3085: \psfrag{a}{\Huge $a$}

3086: \psfrag{b}{\Huge $b$}

3087: \psfrag{c}{\Huge $c$}

3088: \psfrag{d}{\Huge $d$}

3089: \psfrag{so}{\Huge $\text{source}$}

3090: \psfrag{si}{\Huge $\text{sink}$}

3091: \resizebox{1 in}{.6 in}{\includegraphics{gsubt.eps}}

3092: \end{center}

3093: \end{minipage}

3094: \]

3095: We indicate the source (resp. sink) vertex 

3096: by a half-edge with an arrow going in (resp. out).

3097: We regard the two half-edges together as 

3098: constituting the output edge of the operad.

3099: Substitution works as under.

3100: \[

3101: \begin{minipage}[b]{.9 in}

3102: \begin{center}

3103: \psfrag{a}{\Huge $a$}

3104: \psfrag{b}{\Huge $b$}

3105: \psfrag{c}{\Huge $c$}

3106: \psfrag{d}{\Huge $d$}

3107: \psfrag{so}{\Huge $$}

3108: \psfrag{si}{\Huge $$}

3109: \resizebox{.9 in}{.7 in}{\includegraphics{gsubt.eps}}\\

3110: $p_2$

3111: \end{center}

3112: %\label{f:gsubt}

3113: \end{minipage}

3114: \hfill

3115: \begin{minipage}[b]{1.2 in}

3116: \begin{center}

3117: \psfrag{x}{\Huge $x$}

3118: \psfrag{y}{\Huge $y$}

3119: \psfrag{z}{\Huge $z$}

3120: \psfrag{w}{\Huge $$}

3121: \psfrag{so}{\Huge $\text{source}$}

3122: \psfrag{si}{\Huge $\text{sink}$}

3123: \resizebox{.9 in}{.5 in}{\includegraphics{gsubi.eps}}\\

3124: \vspace{.1 in}

3125: $p_1$

3126: \end{center}

3127: %\label{f:gsubi}

3128: \end{minipage}

3129: \hspace{1 in}

3130: \begin{minipage}[b]{1.4 in}

3131: \begin{center}

3132: \psfrag{a}{\Huge $a$}

3133: \psfrag{b}{\Huge $b$}

3134: \psfrag{c}{\Huge $c$}

3135: \psfrag{d}{\Huge $d$}

3136: \psfrag{x}{\Huge $x$}

3137: \psfrag{z}{\Huge $z$}

3138: \psfrag{w}{\Huge $$}

3139: \psfrag{so}{\Huge $\text{source}$}

3140: \psfrag{si}{\Huge $\text{sink}$}

3141: \resizebox{1.3 in}{.7 in}{\includegraphics{gsubo.eps}}\\

3142: $p_2 \yright p_1$

3143: \end{center}

3144: %\label{f:gsubo}

3145: \end{minipage}.

3146: \]

3147: In other words,

3148: to obtain

3149: $p_2 \yright p_1$,

3150: we identify the source (resp. sink) vertex of $p_2$

3151: with the initial (resp. terminal) vertex of the edge $y$ in $p_1$.

3152: 

3153: 

3154: \subsubsection{Sub-examples}

3155: The commutative, associative, chord 

3156: (\ref {ss:eg}) and many other operads

3157: can be obtained as special cases

3158: by simply restricting the type of graphs allowed.

3159: There is seemingly no end to the number of interesting examples

3160: one can obtain in this way.

3161: We illustrate the commutative and associative cases.

3162: \[

3163: \begin{minipage}[b]{1.8 in}

3164: \begin{center}

3165: \vspace{.1 in} 

3166: \psfrag{a}{\Huge $a$}

3167: \psfrag{b}{\Huge $b$}

3168: \psfrag{c}{\Huge $c$}

3169: \psfrag{d}{\Huge $d$}

3170: \psfrag{x}{\Huge $x$}

3171: \psfrag{z}{\Huge $z$}

3172: \psfrag{w}{\Huge $$}

3173: \psfrag{so}{\Huge $\text{source}$}

3174: \psfrag{si}{\Huge $\text{sink}$}

3175: \resizebox{1.8 in}{.4 in}{\includegraphics{gc.eps}}\\

3176: commutative

3177: \end{center}

3178: \end{minipage}

3179: \hspace{1 in}

3180: \begin{minipage}[b]{1.6 in}

3181: \begin{center}

3182: \vspace{.1 in} 

3183: \psfrag{a}{\Huge $a$}

3184: \psfrag{b}{\Huge $b$}

3185: \psfrag{c}{\Huge $c$}

3186: \psfrag{d}{\Huge $d$}

3187: \psfrag{x}{\Huge $x$}

3188: \psfrag{z}{\Huge $z$}

3189: \psfrag{w}{\Huge $$}

3190: \psfrag{so}{\Huge $\text{source}$}

3191: \psfrag{si}{\Huge $\text{sink}$}

3192: \resizebox{1.6 in}{.4 in}{\includegraphics{ga.eps}}\\

3193: %\vspace{.1 in} 

3194: associative

3195: \end{center}

3196: \end{minipage}

3197: \]

3198: Thus to get the commutative operad,

3199: we only allow graphs with two vertices (source and sink)

3200: and all edges are directed from the source to the sink.

3201: For the associative operad,

3202: we allow graphs whose edges form a directed path 

3203: from the source to the sink.

3204: 

3205: There is a variation on the associative case

3206: (and the commutative),

3207: where one allows the arrows on the edges to point

3208: in either direction.

3209: This is a binary quadratic operad.

3210: Axioms for an algebra over this operad can be written explicitly.

3211: We do not know whether they have been considered before.

3212: 

3213: \subsubsection{The reversal maps}

3214: The graph operad is reversible 

3215: with the reversal map as shown.

3216: \[

3217: r_{x,y}

3218: \left(

3219: \begin{minipage}{1.4 in}

3220: \begin{center}

3221: \psfrag{x}{\Huge $a$}

3222: \psfrag{y}{\Huge $x$}

3223: \psfrag{z}{\Huge $b$}

3224: \psfrag{w}{\Huge $y$}

3225: \psfrag{so}{\Huge $\text{source}$}

3226: \psfrag{si}{\Huge $\text{sink}$}

3227: \resizebox{1 in}{.5 in}{\includegraphics{gsubi.eps}}

3228: \end{center}

3229: \end{minipage}

3230: \right)

3231: \ \ = \ \

3232: \begin{minipage}{1.4 in}

3233: \begin{center}

3234: \psfrag{x}{\Huge $a$}

3235: \psfrag{y}{\Huge $x$}

3236: \psfrag{z}{\Huge $b$}

3237: \psfrag{w}{\Huge $y$}

3238: \psfrag{so}{\Huge $\text{source}$}

3239: \psfrag{si}{\Huge $\text{sink}$}

3240: \resizebox{1.2 in}{.5 in}{\includegraphics{gsubir.eps}}

3241: \end{center}

3242: \end{minipage}

3243: .

3244: \]

3245: Namely,

3246: we open the input edge $x$ and close the output edge $y$.

3247: 

3248: There is another way to reverse the graph operad,

3249: where in addition to the above,

3250: we also switch the directions on $x$ and $y$.

3251: The two methods of reversal are compatible

3252: with the associative and commutative cases respectively.

3253: Since the two ways are analogous, we will work with the first definition.

3254: 

3255: 

3256: \subsubsection{How graph objects mate}

3257: So far, we have only defined the graph operad.

3258: But the definition of the graph PROP should be clear;

3259: we simply allow more output edges.

3260: In other words, graphs can now have many pairs

3261: of source and sink vertices.

3262: And for the composition law in the PROP,

3263: we perform simultaneous substitutions.

3264: The graph species $gg$ is the part of the graph PROP

3265: with no outputs.

3266: So these are graphs with no source and sink vertex.

3267: We now show how graph objects mate.

3268: \[

3269: \begin{minipage}{1.5 in}

3270: \begin{center}

3271: \psfrag{a}{\Huge $a$}

3272: \psfrag{b}{\Huge $b$}

3273: \psfrag{c}{\Huge $c$}

3274: \psfrag{d}{\Huge $d$}

3275: \psfrag{x}{\Huge $x$}

3276: \psfrag{y}{\Huge $y$}

3277: \psfrag{z}{\Huge $z$}

3278: \resizebox{1.4 in}{.7 in}{\includegraphics{gmb.eps}}

3279: \end{center}

3280: \end{minipage}

3281: =

3282: \begin{minipage}{1.5 in}

3283: \begin{center}

3284: \psfrag{a}{\Huge $a$}

3285: \psfrag{b}{\Huge $b$}

3286: \psfrag{c}{\Huge $c$}

3287: \psfrag{d}{\Huge $d$}

3288: \psfrag{x}{\Huge $x$}

3289: \psfrag{y}{\Huge $y$}

3290: \psfrag{z}{\Huge $z$}

3291: \resizebox{1.3 in}{.7 in}{\includegraphics{gmt.eps}}

3292: \end{center}

3293: \end{minipage}

3294: =

3295: \begin{minipage}{1.1 in}

3296: \begin{center}

3297: \psfrag{a}{\Huge $a$}

3298: \psfrag{b}{\Huge $b$}

3299: \psfrag{c}{\Huge $c$}

3300: \psfrag{d}{\Huge $d$}

3301: \psfrag{x}{\Huge $x$}

3302: \psfrag{y}{\Huge $y$}

3303: \psfrag{z}{\Huge $z$}

3304: \resizebox{.9 in}{.7 in}{\includegraphics{gmd.eps}}

3305: \end{center}

3306: \end{minipage}.

3307: \]

3308: Namely, the sink of one merges (mates) with the source of the other

3309: and vice-versa.

3310: The result is a graph with no source and sink vertex.

3311: Thus the mating functor maps the graph operad $g$

3312: to the graph species $gg$.

3313: 

3314: \subsubsection{The algebraic objects}

3315: The free algebras $gA$ and $ggA$ consist of graphs

3316: with directed edges labelled by one of the generators

3317: $x_1, \ldots,x_n$.

3318: Of course, for elements of $gA$,

3319: graphs also have a specified source and sink vertex.

3320: The partial derivative 

3321: $\frac{\partial }{\partial x_i} : ggA \longrightarrow gA$

3322: works very nicely.

3323: \[

3324: \begin{minipage}{1 in}

3325: \begin{center}

3326: \psfrag{c}{\Huge $x_1$}

3327: \psfrag{z}{\Huge $x_1$}

3328: \psfrag{b}{\Huge $x_2$}

3329: \psfrag{a}{\Huge $x_2$}

3330: \psfrag{d}{\Huge $x_4$}

3331: \psfrag{y}{\Huge $x_4$}

3332: \psfrag{x}{\Huge $x_3$}

3333: \resizebox{.8 in}{.7 in}{\includegraphics{gmd.eps}}

3334: \end{center}

3335: \end{minipage}

3336: \qquad \buildrel \frac{\partial}{\partial x_1}\over \longmapsto \qquad

3337: \begin{minipage}{1 in}

3338: \begin{center}

3339: \psfrag{c}{\Huge $x_1$}

3340: \psfrag{z}{\Huge $x_1$}

3341: \psfrag{b}{\Huge $x_2$}

3342: \psfrag{a}{\Huge $x_2$}

3343: \psfrag{d}{\Huge $x_4$}

3344: \psfrag{y}{\Huge $x_4$}

3345: \psfrag{x}{\Huge $x_3$}

3346: \resizebox{.9 in}{.7 in}{\includegraphics{gmdr.eps}}

3347: \end{center}

3348: \end{minipage}

3349: \ \ + \ \

3350: \begin{minipage}{1 in}

3351: \begin{center}

3352: \psfrag{c}{\Huge $x_1$}

3353: \psfrag{z}{\Huge $x_1$}

3354: \psfrag{b}{\Huge $x_2$}

3355: \psfrag{a}{\Huge $x_2$}

3356: \psfrag{d}{\Huge $x_4$}

3357: \psfrag{y}{\Huge $x_4$}

3358: \psfrag{x}{\Huge $x_3$}

3359: \resizebox{.8 in}{.7 in}{\includegraphics{gmds.eps}}

3360: \end{center}

3361: \end{minipage}

3362: .

3363: \]

3364: In words,

3365: $\frac{\partial F}{\partial x_i}$ 

3366: is the sum of graphs obtained by cutting edges of $F$

3367: that are labelled $x_i$, one at a time.

3368: The half-edges that result from a cut get induced directions

3369: and provide the source and sink vertex

3370: of the graph created.

3371: We leave it to the reader to explicitly describe

3372: the algebra of forms $\Omega(gA)$, $\Omega(ggA)$

3373: and other notions described in Section~\ref{s:st}.

3374: 

3375: For higher order derivatives,

3376: we consider the free algebra of the graph PROP.

3377: The order of the derivative determines 

3378: the number of edges that are cut.

3379: And for higher order matings,

3380: we take two graphs with the same number of cuts (outputs)

3381: and the result is a graph with no cuts.

3382: This generalises the first order mating 

3383: given by the mating functor.

3384: 

3385: \subsection{A generalisation of the tree example

3386: (\ref {sss:t})} \label{ss:gt}

3387: 

3388: We now present a second example based on graphs.

3389: It complements the earlier example in the sense

3390: that (univalent) vertices rather than edges

3391: play the role of inputs.

3392: Let $hh[I]=$ set of graphs with univalent vertices

3393: labelled by elements of the set $I$.

3394: The operad $h$ is defined as the species $hh$

3395: except that we use one of the univalent vertices

3396: as the output or root.

3397: And for the PROP,

3398: we do not place any restriction

3399: on the number of outputs or roots.

3400: Operad substitution and reversal and matings work

3401: exactly as in the tree case.

3402: \[

3403: \begin{minipage}[b]{1.8 in}

3404: \begin{center}

3405: \psfrag{a}{\Huge $a$}

3406: \psfrag{b}{\Huge $b$}

3407: \psfrag{y}{\Huge $y$}

3408: \psfrag{z}{\Huge $z$}

3409: \resizebox{1.7 in}{.5 in}{\includegraphics{hsubi.eps}}

3410: \end{center}

3411: \end{minipage}

3412: \hspace{1 in}

3413: \begin{minipage}[b]{1.6 in}

3414: \begin{center}

3415: \psfrag{a}{\Huge $a$}

3416: \psfrag{b}{\Huge $b$}

3417: \psfrag{y}{\Huge $y$}

3418: \psfrag{z}{\Huge $z$}

3419: \resizebox{1.6 in}{.5 in}{\includegraphics{hsubo.eps}}

3420: \end{center}

3421: \end{minipage}.

3422: \]

3423: Note that for the above composition law

3424: of the PROP to make sense,

3425: we necessarily have to work with graphs and not just trees.

3426: We mention that the graph complexes that arise from this species

3427: have been considered in~\cite{\bm}.

3428: 

3429: 

3430: \subsection{The surface operad $s$ and species $ss$} \label{ss:ss}

3431: 

3432: Let $ss[I]=$ set of compact orientable surfaces

3433: (not necessarily connected) with boundary circles

3434: labelled by elements of the set $I$.

3435: For the operad $s[I]$, 

3436: we use one of the boundary circles as the output.

3437: Operad substitution works by gluing.

3438: For the PROP, we allow more outputs as usual.

3439: The composition law in the PROP is simultaneous gluing.

3440: We leave the details to the reader.

3441: Some pictures may be found in~\ref{ss:ae}.

3442: 

3443: \section{Graph homology} \label{s:gh}

3444: 

3445: In this section, we define a generalisation of 

3446: Kontsevich's graph homology to mated species.

3447: This coincides with the graph complexes

3448: considered by Markl~\cite{\markl}.

3449: When one specialises to the commutative species

3450: (\ref{sss:c}),

3451: one recovers the usual definition.

3452: The case of the associative species (\ref{sss:a}) is 

3453: also explained in~\cite{\voronov}.

3454: By a graph, we mean a finite 1-dimensional CW complex.

3455: 

3456: In (\ref {ss:graph}-\ref {ss:gc}),

3457: we define a chain complex $(\mathcal G,\partial_E)$

3458: based on graphs.

3459: Here $\mathcal G$ refers to the chain groups

3460: and $\partial_E$ to the boundary operator.

3461: The boundary operator $\partial_E$

3462: can be regarded as a limit of a family of boundary operators

3463: $\partial_n$.

3464: This family is discussed in 

3465: (\ref {ss:finb}).

3466: In (\ref {ss:cc}),

3467: we list the subcomplexes of $(\mathcal G,\partial_E)$

3468: that are relevant to us.

3469: The graph complex $(Q\mathcal G,\partial_E)$,

3470: which appears in the main theorem, is one of them.

3471: 

3472: 

3473: \subsection{Graphs}  \label{ss:graph}

3474: 

3475: Let $Q$ be any species.

3476: Define a $Q$-graph to be a graph whose vertices are fattened 

3477: by elements of the species $Q$.

3478: To be more precise,

3479: it is a graph $\Gamma$ such that

3480: for every vertex $v$ of $\Gamma$,

3481: a $Q$-structure is specified 

3482: on the set of half-edges incident to $v$.

3483: \[

3484: \begin{minipage}{1.8 in}

3485: \begin{center}

3486: \psfrag{1}{\Huge $1$}

3487: \psfrag{2}{\Huge $2$}

3488: \psfrag{3}{\Huge $3$}

3489: \psfrag{4}{\Huge $4$}

3490: \resizebox{.9 in}{.9 in}{\includegraphics{man.eps}}\\

3491: \vspace{.1 in} Greetings from a $Q$-graph.

3492: \end{center}

3493: \end{minipage}

3494: \]

3495: 

3496: \noindent

3497: We now give some examples;

3498: see the table of pictures

3499: and examples of species in (\ref{ss:s}) and (\ref{ss:eg}).

3500: \begin{itemize}

3501: \item

3502: A commutative graph $(Q=cc)$  is simply a graph.

3503: 

3504: \item

3505: An associative graph $(Q=aa)$ is a ribbon graph;

3506: that is a cyclic ordering of

3507: the half-edges is fixed at each vertex.

3508: %And in the Lie graph $(Q=ll)$, a bracket expression of the half-edges is given.

3509: 

3510: \item

3511: The figure below shows a typical graph for a species based on trees

3512: $(Q=tt)$.

3513: \[

3514: \begin{minipage}{1.8 in}

3515: \begin{center}

3516: \psfrag{1}{\Huge $1$}

3517: \psfrag{2}{\Huge $2$}

3518: \psfrag{3}{\Huge $3$}

3519: \psfrag{4}{\Huge $4$}

3520: \resizebox{1.3 in}{1 in}{\includegraphics{tree.eps}}

3521: \end{center}

3522: \end{minipage}

3523: \]

3524: 

3525: \item

3526: For the chord species $(Q=kk)$,

3527: the vertices instead of being trees are chord diagrams

3528: (\ref{sss:k}).

3529: In our terminology, these are chord graphs.

3530: 

3531: \item

3532: For an example with a somewhat different flavour,

3533: consider the surface species $Q=ss$, see (\ref{ss:ss}).

3534: A surface graph is a compact orientable surface

3535: with a collection of disjoint loops.

3536: In this case,

3537: the loops play the role of edges and

3538: the pieces obtained 

3539: by cutting along the loops play the role of vertices.

3540: This example was first mentioned to me by J. Conant.

3541: \end{itemize}

3542: The interested reader can also work out the graphs

3543: for the species $hh$ and the graph species $gg$

3544: in Section~\ref{s:me}.

3545: Note that all the species above have been denoted

3546: by a letter repetition.

3547: This is to indicate the fact that they are

3548: mated species (\ref{ss:mf}-\ref{ss:am}).

3549: 

3550: The reader, who skipped Sections~\ref{s:ro} and \ref{s:mf}

3551: on species and operads, 

3552: can read this section and Section~\ref{s:mgh}

3553: by keeping the above concrete examples in mind.

3554: For simplicity of notation, from now on,

3555: we will just write ``graph'' instead of ``$Q$-graph''.

3556: 

3557: \subsection{Oriented graphs}  \label{ss:og}

3558: 

3559: For a graph $\Gamma$, we shall denote the set of vertices by

3560: $V(\Gamma)$, the set of edges by $E(\Gamma)$

3561: and the set of ends of

3562: an edge $e$ by $V(e)$.

3563: %, and the set of edges emanating from a

3564: %vertex $v$ by $E(v)$.  

3565: Finally, $\mathbb{R}X$ will

3566: denote the real vector space which has the elements of $X$ as a

3567: basis.

3568: 

3569: \begin{definition}   

3570: An orientation $\sigma$ of a graph $\Gamma$ is an

3571: orientation of the one-dimensional real vector space $\det

3572: \mathbb RV(\Gamma) \otimes\bigotimes_{e\in E(\Gamma)}\det \mathbb

3573: RV(e)$. 

3574: For a vector space $W$ of dimension $n$,

3575: we are using the notation $\det W = \Lambda^n W$. 

3576: We say that $(\Gamma, \sigma)$ is an oriented graph.

3577: There is another notion of orientation of a graph

3578: equivalent to the above; see \cite{\thurston} for details.

3579: \end{definition}

3580: 

3581: A way to represent an orientation is to order the

3582: vertices and orient each edge of the graph.  An odd permutation

3583: of the labels on the vertices reverses the orientation, and a

3584: single change of the orientation of one edge reverses it as

3585: well.  An even number of these transformations produces an

3586: orientation equivalent to the original one.

3587: So a graph $\Gamma$ has two orientations,

3588: which we may call $\sigma$ and $- \sigma$.

3589: \[

3590: \begin{minipage}{\linewidth}

3591: \begin{center}

3592: \psfrag{1}{\Huge $1$}

3593: \psfrag{2}{\Huge $2$}

3594: \psfrag{3}{\Huge $3$}

3595: \psfrag{4}{\Huge $4$}

3596: \resizebox{1.3 in}{1 in}{\includegraphics{orient.eps}}\\

3597: \vspace{.1 in} A representative of an orientation on a $Q$-graph.

3598: \end{center}

3599: \end{minipage}

3600: \]

3601: 

3602: 

3603: We say that $(\Gamma, \sigma )$ and $(\Gamma^\prime,

3604: \sigma^\prime )$ are isomorphic if there is a graph isomorphism

3605: $h:\Gamma \to \Gamma^\prime$ such that $h$ carries the

3606: orientation $\sigma$ to $\sigma^\prime$.

3607: 

3608: Later for simplicity,

3609: we will sometimes suppress the orientation

3610: from our notation and picture.

3611: It is understood that from now on,

3612: all graphs are oriented.

3613: 

3614: 

3615: \subsection{The graph complex}  \label{ss:gc}

3616: 

3617: From now on, we restrict to those species $Q$ that arise 

3618: from the symplectic theory.

3619: Namely, we assume that $Q$ is a mated species (\ref{ss:mf}-\ref{ss:am}).

3620: The examples in (\ref{ss:graph}) were all of this type.

3621: Hence the vertices necessarily have valence at least 2.

3622: We now define a chain complex $(\mathcal G,\partial_E)$ for such species.

3623: 

3624:  

3625: \begin{definition}  

3626: The $k$th chain group of $\mathcal G$, 

3627: which we denote $\mathcal G_k$, 

3628: is the vector space over $\mathbb Q$ generated by all oriented

3629: graphs $(\Gamma,\sigma)$ with $k$ vertices

3630: subject to the relations:

3631: \begin{equation} \label{e:grel}

3632: (\Gamma,\sigma) = -(\Gamma,-\sigma),\ \ \text{and} \ \

3633: (\Gamma,\sigma)=(\Gamma^\prime,\sigma^\prime) \ \text{if} \

3634: (\Gamma,\sigma)\cong(\Gamma^\prime,\sigma^\prime).

3635: \end{equation}

3636: \end{definition}

3637: 

3638: \noindent

3639: A graph can have an orientation reversing automorphism, that is

3640: $(\Gamma,\sigma)\cong (\Gamma,-\sigma)$, where $\sigma$ and $-\sigma$ are

3641: the two orientations of $\Gamma$. 

3642: Note that the first relation then makes the class of such a graph 

3643: equal to zero.

3644: 

3645: \begin{definition}  

3646: The boundary map $\partial_E : \mathcal G_k\to \mathcal G_{k-1}$

3647: is defined using edge contractions, i.e. matings

3648: (\ref{ss:am}).

3649: This is where one uses that $Q$ is a mated species.

3650: We do not contract loops.

3651: More precisely, we have

3652: 

3653: $$\partial_E (\Gamma,\sigma) = \sum_{e \in E(\Gamma)} (\Gamma /e,

3654: \sigma /e),$$

3655: %

3656: where $\Gamma /e$ is the graph $\Gamma$ with the edge $e$ contracted

3657: (see figure), and $\sigma /e$ is obtained the following way: 

3658: choose a

3659: representative of $\sigma$ where $e$ points from vertex 1 to vertex 2,

3660: give the new vertex arising from the contraction of $e$ the label 1, and

3661: subtract 1 from the label of each of the other vertices; 

3662: finally, keep the

3663: orientations on the edges (other than $e$) unchanged.

3664: \[

3665: \begin{minipage}{\linewidth}

3666: \begin{center}

3667: \begin{minipage}{1 in}

3668: \begin{center}

3669: \psfrag{1}{\Huge $1$}

3670: \psfrag{2}{\Huge $2$}

3671: \psfrag{3}{\Huge $3$}

3672: \psfrag{4}{\Huge $4$}

3673: \resizebox{.8 in}{.8 in}{\includegraphics{bi.eps}}

3674: \end{center}

3675: \end{minipage}

3676: $\qquad \buildrel {\text{contract} \ e}\over \longmapsto \qquad$

3677: \begin{minipage}{1 in}

3678: \begin{center}

3679: \psfrag{1}{\Huge $1$}

3680: \psfrag{2}{\Huge $2$}

3681: \psfrag{3}{\Huge $3$}

3682: \psfrag{4}{\Huge $4$}

3683: \psfrag{5}{\Huge $5$}

3684: \resizebox{.8 in}{.8 in}{\includegraphics{bo.eps}}

3685: \end{center}

3686: \end{minipage}.\\

3687: \vspace{.1 in}

3688: A mating of two vertices along an edge $e$.

3689: \end{center}

3690: \end{minipage}

3691: \]

3692: \end{definition}

3693: 

3694: \noindent

3695: An equivalent way to describe $\sigma /e$ is the following: if the labels

3696: on the endpoints of $e$ are $i < j$, collapse $e$, label the resulting

3697: vertex $i$, decrease the labels greater than $j$ by one, and multiply this

3698: orientation by $(-1)^j$ if $e$ points from $i$ to $j$, and by

3699: $(-1)^{j+1}$ if it points from $j$ to $i$.

3700: 

3701: Note that $\sigma /e$ is well-defined and that $\partial_E$ respects the

3702: relations in equation~\eqref{e:grel}. 

3703: 

3704: \subsubsection*{Edge contractions in examples}

3705: 

3706: We explain how edge contractions work in the examples

3707: in (\ref{ss:graph}).

3708: A commutative graph is an ordinary graph.

3709: And edge contraction has the usual meaning.

3710: For associative or ribbon graphs,

3711: a (local) picture for an edge contraction

3712: is given in (\ref{ss:am}).

3713: For tree (resp. chord) graphs,

3714: we merge the trees (resp. chords)

3715: at the two vertices along the edge being contracted.

3716: For a surface graph,

3717: an edge is a loop on the surface.

3718: To contract an edge, simply delete the loop.

3719: 

3720: \begin{lemma} 

3721: $\partial_E^2 =0.$

3722: \end{lemma}

3723: 

3724: \begin{proof}

3725: If we collapse first the edge $e = (i, j)$ and then the edge $e' =

3726: (i',j')$, we get the same graph as by collapsing $e'$ first and

3727: then $e$, but with the opposite orientation.  This is easiest to

3728: see using the second definition of $\sigma /e$ above; we may

3729: assume that $j < j'$ and $i < j$, $i' < j'$. We get the same

3730: orientation either way, but in the first case with a coefficient

3731: $(-1)^{j}(-1)^{j'-1}$, in the second case with

3732: $(-1)^{j'}(-1)^{j}$.

3733: \end{proof}

3734: 

3735: Thus $(\mathcal G,\partial_E)$ is a chain complex.

3736: We will denote its homology by $H_*(\mathcal G,\partial_E)$.

3737: 

3738: \subsection{A family of boundary operators}  \label{ss:finb}

3739: 

3740: The reader, who has never seen graph homology before,

3741: may omit this section on a first reading.

3742: We now define a second boundary operator $\partial_H$

3743: that plays an important part in this theory.

3744: 

3745: Let $E_q(\Gamma)$ be the set of all \emph{quasi-edges} of $\Gamma$.

3746: These are unordered pairs of distinct half-edges of $\Gamma$.

3747: A quasi-edge is specified by two vertices, say $v_1$ and $v_2$, 

3748: and two edges $e_1$ and $e_2$ incident to $v_1$ and $v_2$ respectively.

3749: We say that

3750: $e$ is a \emph{quasi-loop} if $v_1 = v_2$.

3751: Observe that $E(\Gamma) \subset E_q(\Gamma)$. 

3752: 

3753: \begin{definition}  

3754: The boundary map $\partial_H : \mathcal G_k\to \mathcal G_{k-1}$

3755: is defined by contracting quasi-edges, which are not edges.

3756: We do not contract quasi-loops. We have

3757: $$\partial_{H} (\Gamma,\sigma) = \sum_{e \in E_q(\Gamma) \setminus E(\Gamma)} 

3758: (\Gamma /e,\sigma /e),$$

3759: where 

3760: $\Gamma /e$ and $\sigma /e$ 

3761: are defined as follows.

3762: 

3763: 1. Choose a representative for $\sigma$ where

3764: $v_1$ and $v_2$ have labels $1$ and $2$ respectively

3765: and the arrow on edge $e_1$

3766: (resp. $e_2$) points out of $v_1$ (resp. into $v_2$).

3767: 

3768: 2. Cut the edges $e_1$ and $e_2$ and join them 

3769: so that the quasi-edge $e$ and its partner $f$ turn into edges. 

3770: In effect,

3771: $e_1$ and $e_2$ have been replaced by two new edges 

3772: (which get an induced orientation).

3773: \[

3774: \begin{minipage}{\linewidth}

3775: \begin{center}

3776: \begin{minipage}{1.6 in}

3777: \begin{center}

3778: \psfrag{v1}{\Huge $v_1$}

3779: \psfrag{v2}{\Huge $v_2$}

3780: \psfrag{e1}{\Huge $e_1$}

3781: \psfrag{e2}{\Huge $e_2$}

3782: \psfrag{e}{\Huge $e$}

3783: \psfrag{f}{\Huge $f$}

3784: \psfrag{pi}{\Huge $$}

3785: \psfrag{qi}{\Huge $$}

3786: \resizebox{1.4 in}{.9 in}{\includegraphics{bqi.eps}}

3787: \end{center}

3788: \end{minipage}

3789: $\qquad \buildrel {\text{contract} \ e}\over \longmapsto \qquad$

3790: \begin{minipage}{1.6 in}

3791: \begin{center}

3792: \psfrag{v1}{\Huge $v_1$}

3793: \psfrag{v2}{\Huge $v_2$}

3794: \psfrag{e}{\Huge $e$}

3795: \psfrag{f}{\Huge $f$}

3796: \resizebox{1.4 in}{.9 in}{\includegraphics{bqo.eps}}

3797: \end{center}

3798: \end{minipage}.\\

3799: \vspace{.1 in} A mating along a quasi-edge $e$.

3800: \end{center}

3801: \end{minipage}

3802: \]

3803: 

3804: 3. Collapse the edge joining $v_1$ and $v_2$. 

3805: Thus $v_1$ and $v_2$ become a single vertex, say $v$. 

3806: The resulting graph is $\Gamma /e$. 

3807: Note that the edges of $\Gamma /e$ are already oriented. 

3808: To get an order on the vertices, 

3809: label the new vertex $v$ by $1$ and

3810: subtract $1$ from the labels of the other vertices. 

3811: The resulting orientation is $\sigma /e$. 

3812: \end{definition}

3813: 

3814: 

3815: %

3816: \begin{definition}  

3817: We now define a family of boundary maps by the formula

3818: $$\partial_n = 2n \partial_E + \partial_H.$$

3819: To put in words, $\partial_n(\Gamma)$

3820: is obtained by contracting quasi-edges of $\Gamma$.

3821: And if the quasi-edge is an edge 

3822: then we multiply by a factor of $2n$.

3823: \end{definition}

3824: %

3825: 

3826: The fact

3827: that $\partial_n = 2n \partial_E + \partial_H$ 

3828: is a boundary operator for all $n$

3829: will follow from the proof 

3830: of Theorem~\ref{t:fin} (Section~\ref{s:fin}).

3831: This automatically implies that $\partial_E^2=\partial_H^2=0$

3832: and $\partial_E \partial_H = - \partial_H \partial_E$.

3833: In other words, $\partial_E$ and $\partial_H$ together

3834: span an abelian Lie superalgebra

3835: of operators on $\mathcal G$.

3836: We will call $\partial_n$

3837: the finite boundary operator

3838: and write the chain complex as $(\mathcal G,\partial_n)$.

3839: 

3840: \begin{remark}

3841: If the boundary operator is not stated explicitly,

3842: then we take it to be the stable boundary operator $\partial_E$.

3843: \end{remark}

3844: 

3845: 

3846: \subsection{An important simplification}  \label{ss:is}

3847: 

3848: Note that the boundary maps $\partial_E$ and $\partial_H$ preserve

3849: the Euler characteristic $\chi$ of any graph $\Gamma$.

3850: This is because they decrease the number of edges and vertices

3851: both by 1.

3852: So the graph complex,

3853: irrespective of the boundary map used,

3854: splits as a direct sum of chain complexes

3855: $\mathcal G^{(r)}$:

3856: \begin{gather*}

3857: \cdots \to \mathcal G^{(1)}_3 \to \mathcal G_2^{(1)} \to \mathcal G^{(1)}_1 \to 0\\

3858: \cdots \to \mathcal G^{(2)}_3 \to \mathcal G_2^{(2)} \to \mathcal G^{(2)}_1 \to 0\\

3859: \cdots \to \mathcal G^{(3)}_3 \to \mathcal G_2^{(3)} \to \mathcal G^{(3)}_1 \to 0\\

3860: \hspace{1em}\vdots \hspace{3em} \vdots \hspace{3em}\vdots

3861: \end{gather*}

3862: The useful point is that the chain groups 

3863: $\mathcal G_k^{(r)}$ are finite dimensional.

3864: Instead of $\chi$,

3865: we have used the superscript $r = 1- \chi$ for convenience.

3866: If the graph $\Gamma$ is connected

3867: then $r=\dim_{\Q}H_1(\Gamma, \Q)$ 

3868: is just the rank of $\Gamma$.

3869: 

3870: 

3871: 

3872: \subsection{Other relevant graph complexes}  \label{ss:cc}

3873: 

3874: From now on, assume that the species $Q$

3875: is based on sets rather than vector spaces.

3876: All the examples

3877: in this paper are of this type.

3878: We first give a preliminary definition.

3879: 

3880: \begin{definition}  

3881: We say a vertex in a $Q$-graph is fake 

3882: if it corresponds to the element $uu$ in $Q[2]$,

3883: the degree 2 piece of $Q$, see (\ref{ss:uu}).

3884: Else we say that the vertex is real.

3885: \[

3886: \begin{minipage}{4 in}

3887: \begin{center}

3888: \resizebox{1 in}{.6 in}{\includegraphics{f.eps}}\\

3889: \vspace{.1 in} The dots stand for the fake vertices.

3890: \end{center}

3891: \end{minipage}

3892: \]

3893: \end{definition}

3894: Note that the fake vertices are always bivalent

3895: and they all look identical.

3896: Furthermore, they behave as unit elements 

3897: in contractions (matings).

3898: So in this sense, they are extremely inert.

3899: 

3900: For all the examples in (\ref{ss:graph}),

3901: except the surface species $ss$,

3902: $Q[2]$ is a singleton.

3903: For example,

3904: there is only one cyclic order or chord diagram or unrooted tree

3905: on 2 letters.

3906: Hence, in these cases, 

3907: the fake vertices are precisely the bivalent vertices.

3908: And real vertices are exactly those with valence greater than 2.

3909: For the surface species, 

3910: $ss[2]$ consists of compact orientable surfaces 

3911: whose boundary is two disjoint circles.

3912: And there are clearly many such.

3913: Among them, cylinders are precisely the fake vertices,

3914: while the rest are real.

3915: 

3916: It can even happen that

3917: contracting an edge joining two real vertices 

3918: produces a fake vertex.

3919: An example is given in Section~\ref{s:gp}.

3920: In this case, the two real vertices must be necessarily bivalent.

3921: In any case, the boundary map can reduce the number

3922: of real vertices of a graph by at most 2.

3923: 

3924: We now look at the various subcomplexes of $(\mathcal G,\partial_E)$,

3925: that appear in the main theorem and its proof.

3926: They are all defined by putting some restriction

3927: on the type of graphs that are allowed.

3928: \begin{itemize}

3929: 

3930: \item

3931: $\mathcal T =$ graphs all of whose vertices are bivalent.

3932: 

3933: \item

3934: $\mathcal F =$ graphs all of whose vertices are fake.

3935: 

3936: \item

3937: $\mathcal C =$ connected graphs.

3938: 

3939: \item

3940: $Q\mathcal G =$ connected graphs all of whose vertices are real

3941: and with at least one vertex of degree greater than 2.

3942: 

3943: \item

3944: $\mathcal B =$ connected graphs all of whose vertices are bivalent.

3945: 

3946: \item

3947: $\mathcal P =$ connected graphs all of whose vertices are fake.

3948: \end{itemize}

3949: 

3950: It is clear that $\partial_E$ preserves the above chain groups

3951: except $Q\mathcal G$.

3952: The trouble with $Q\mathcal G$ is that an edge contraction

3953: can result in a fake vertex.

3954: Hence for $Q\mathcal G$, modify the definition of $\partial_E$

3955: such that edge contractions that create fake vertices are ignored.  

3956: So, strictly speaking, it is not a subcomplex of $(\mathcal G,\partial_E)$.

3957: In keeping with Kontsevich's original definition,

3958: we will call $H_*(Q\mathcal G,\partial_E)$

3959: the graph homology of the mated species $Q$.

3960: 

3961: Note that $\mathcal F$ and $\mathcal P$

3962: do not depend on the species $Q$.

3963: The complex $\mathcal P$ just consists of polygons and

3964: $\mathcal F$ consists of disjoint unions of polygons.

3965: Further, since fake vertices are always bivalent,

3966: the complex $\mathcal F$ (resp. $\mathcal P$)

3967: is a subcomplex of $\mathcal T$ (resp. $\mathcal B$).

3968: The containment is strict if $Q[2]$ has more than 1 element,

3969: as for the surface or graph species.

3970: We discuss another example of this kind in Section~\ref{s:gp}. 

3971: 

3972: \subsection{A reduction step}  

3973: 

3974: The chain complex that appears in the statement

3975: of the main theorem is $(Q\mathcal G,\partial_E)$.

3976: However, the one that appears naturally in the proof is

3977: $(\mathcal C,\partial_E)$.

3978: The reduction from the homology 

3979: of this larger complex to 

3980: graph homology $H_*(Q\mathcal G,\partial_E)$

3981: is the content of Proposition~\ref{p:ss}.

3982: It will be the last step in the proof

3983: of the main theorem.

3984: Though the statement of the proposition 

3985: is quite intuitive,

3986: its proof is a little technical and may be skipped

3987: on a first reading.

3988: 

3989: \begin{proposition}  \label{p:ss}

3990: $H_*(\mathcal C) = H_*(Q\mathcal G) \bigoplus H_*(\mathcal B)$.

3991: \end{proposition}

3992: \begin{proof}

3993: The bivalent graph complex $\mathcal B$ is a direct summand 

3994: of the graph complex $\mathcal C$.

3995: The complement is the subcomplex consisting of connected graphs

3996: containing at least one vertex of degree $\geq 3$.

3997: Denote this complex by $\mathcal D$, graded as usual by the number of

3998: vertices. 

3999: The difference between $\mathcal D$ and $Q\mathcal G$ is that

4000: graphs in $\mathcal D$ are allowed to contain fake vertices.

4001: To finish the proof, we need to show that 

4002: $H_*(\mathcal D) = H_*(Q\mathcal G)$.

4003: This can be done by a spectral sequence argument.

4004: For a treatment of spectral sequences, see \cite{\mccleary}.

4005: 

4006: We have already defined real and fake vertices.

4007: In addition, call an edge fake if it is incident to a fake vertex.

4008: Next, define a filtration on $\mathcal D$ by

4009: $F_p \mathcal D_m =$ graphs in $\mathcal D_m$ with upto $p$ real vertices.

4010: Recall that for the spectral sequence associated to a filtration,

4011: $E_{p q}^0 = F_p \mathcal D_{p+q} \big/ F_{p-1} \mathcal D_{p+q}$.

4012: In our case, $E_{p q}^0$ can be described as the span of oriented

4013: graphs with $p$ 

4014: real and $q$ fake vertices.

4015: The vertical maps on the $E^0$ page, namely,

4016: $d_0 : E_{p q}^0 \mapsto E_{p q-1}^0$

4017: are defined exactly as 

4018: the boundary map for an oriented graph, 

4019: except that now we are

4020: allowed to collapse only the fake edges.

4021: Note that the positive $X$-axis of the $E^0$ page consists exactly of

4022: graphs with no fake vertices. In other words, they are 

4023: the chain groups of the graph complex $Q\mathcal G$.

4024: 

4025: \medskip

4026: \noindent{\bf Claim}. The homology on the $E^0$ page is trivial except

4027: at $q = 0$, that is, the $X$-axis. Equivalently, 

4028: $E_{p q}^1 = 0$ for $q > 0$. 

4029: In addition, we claim that $E_{p 0}^1 = Q\mathcal G_p$.

4030: 

4031: Assuming the claim,

4032: it is clear that the induced boundary map on the $E^1$ page 

4033: $d_1 : E_{p 0}^1 \mapsto E_{p-1 0}^1$ coincides with the boundary map

4034: defined on $Q\mathcal G$. 

4035: Namely, edge contractions that create fake vertices are ignored.

4036: The rest of the $E^1$ page is zero.

4037: Thus the spectral sequence associated to our filtration of $\mathcal D$

4038: collapses at the first term to the graph complex $Q\mathcal G$.

4039: This shows that $H_*(\mathcal D) = H_*(Q\mathcal G)$.

4040: 

4041: \medskip

4042: \noindent{\bf Proof of the claim}. 

4043: To every connected graph containing at least one real vertex, 

4044: one can

4045: associate a new graph all of whose vertices are real. 

4046: The new graph is just the old graph with all the fake vertices removed. 

4047: This gives us

4048: an equivalence relation on the set of all connected graphs containing

4049: a real vertex. 

4050: The equivalence classes are indexed by graphs all of

4051: whose vertices are real.

4052: Now consider the complex 

4053: $$\ldots\to E_{p q}^0\buildrel d_0\over \to E_{p q-1}^0\buildrel 

4054: d_0\over \to \ldots \buildrel d_0\over \to E_{p 0}^0 = Q\mathcal G_p 

4055: \buildrel d_0\over \to 0$$ 

4056: whose homology we want to compute.

4057: This splits as a direct sum of subcomplexes one for each graph

4058: $\Gamma$ in $E_{p 0}^0 = Q\mathcal G_p$. 

4059: Call the subcomplex corresponding

4060: to $\Gamma$ the standard complex of $\Gamma$. 

4061: Note that the graphs

4062: that occur in the standard complex of $\Gamma$ are precisely those

4063: that lie in the same equivalence class as $\Gamma$. 

4064: We need to show that the

4065: only non-trivial homology occurs in degree $0$.

4066: 

4067: To understand this complex, 

4068: first look at the standard complex of a single edge. 

4069: It has dimension $1$ in each degree $k \geq 0$, 

4070: because there is a unique (upto isomorphism) way to put $k$ points in

4071: the interior of an edge. 

4072: The differential in this standard complex

4073: is zero for $k=0$ and all odd $k$ and 

4074: an isomorphism for positive even $k$. 

4075: Hence it has non-trivial homology only in degree $0$.

4076: 

4077: Now the standard complex of $\Gamma$ is 

4078: the tensor product of the

4079: standard  complexes of all its edges

4080: modulo the action of the finite group $\Aut(\Gamma$). 

4081: Hence by the Kunneth formula 

4082: and the fact that finite groups have trivial rational homology

4083: \cite{\kbrown},

4084: we are done.

4085: \end{proof}

4086: 

4087: \begin{corollary}

4088: Let $Q$ be a mated species, 

4089: where all bivalent vertices in any graph are fake. 

4090: Or equivalently,

4091: let $Q$ be obtained from a reversible operad

4092: satisfying $P[1]=\Q$.

4093: Then

4094: $H_*(\mathcal C) = H_*(Q\mathcal G) \bigoplus H_*(\mathcal P)$.

4095: \end{corollary}

4096: 

4097: 

4098: \section{Graph homology for groups}  \label{s:gp}

4099: 

4100: We now digress to give an example of the theory

4101: that is based on groups.

4102: It can be defined more generally for algebras with involution;

4103: we then recover dihedral homology \cite{\loday}.

4104: We hope that this example will further clarify the concepts

4105: discussed in previous sections.

4106: An attraction of this example is that

4107: it is tractable to computations.

4108: Some known results on dihedral homology 

4109: can be obtained this way.

4110: 

4111: \subsection{Groups as reversible operads} 

4112: Let $K$ be any finite group.

4113: Let $P$ be the operad whose elements 

4114: are given by 

4115: \begin{minipage}{.6 in}

4116: \begin{center}

4117: \psfrag{u}{\Huge $g$}

4118: \psfrag{a}{\Huge $a$}

4119: \psfrag{b}{\Huge $b$}

4120: \resizebox{.5 in}{.2 in}{\includegraphics{ur.eps}}

4121: \end{center}

4122: \end{minipage}

4123: for $g \in K$,

4124: with substitution being group multiplication.

4125: In other words, $P[1]=\Q K$,

4126: the group algebra of $K$ over $\Q$, and

4127: $P[2]=P[3]=\ldots=0$.

4128: If $K$ is the trivial group

4129: then $P$ is just the unit operad $u$,

4130: see (\ref {sss:u}).

4131: More generally, one can take $P[1]$ 

4132: to be any algebra with a unit.

4133: 

4134: For this example,

4135: $P$ is reversible (\ref{ss:ro}),

4136: if there exists a map 

4137: $^{*}: K \to K$ which satisfies

4138: $1^*=1$ and $(g h)^* = h^* g^*$

4139: for any $g,h \in K$.

4140: Natural candidates for the $^{*}$ map are 

4141: the identity

4142: (if $K$ is abelian)

4143: and the inverse map.

4144: Thus, for $K$ abelian,

4145: there are two distinct ways of reversal.

4146: The reversal map is given by 

4147: $$

4148: r_{a,b} \left(

4149: \begin{minipage}{.6 in}

4150: \begin{center}

4151: \psfrag{u}{\Huge $g$}

4152: \psfrag{a}{\Huge $a$}

4153: \psfrag{b}{\Huge $b$}

4154: \resizebox{.5 in}{.2 in}{\includegraphics{ur.eps}}

4155: \end{center}

4156: \end{minipage}

4157: \right)

4158: =

4159: \begin{minipage}{.6 in}

4160: \begin{center}

4161: \psfrag{a}{\Huge $a$}

4162: \psfrag{b}{\Huge $b$}

4163: \psfrag{u}{\Huge $g^*$}

4164: \resizebox{.5 in}{.2 in}{\includegraphics{u.eps}}

4165: \end{center}

4166: \end{minipage}

4167: .$$

4168: 

4169: Applying the mating functor to $P$ gives us a mated species $Q$.

4170: This species $Q$ lives entirely in degree 2.

4171: We leave it to the reader to check that an element of 

4172: $Q[\{a,b\}]$ can be specified by the picture

4173: \begin{equation}  \label{e:gr}

4174: \begin{minipage}{.7 in}

4175: \begin{center}

4176: \psfrag{g}{\Huge $g$}

4177: \psfrag{a}{\Huge $a$}

4178: \psfrag{b}{\Huge $b$}

4179: \resizebox{.7 in}{.3 in}{\includegraphics{g.eps}}

4180: \end{center}

4181: \end{minipage}

4182: \ \ \ = \ \ \

4183: \begin{minipage}{.7 in}

4184: \begin{center}

4185: \psfrag{g}{\Huge $g^*$}

4186: \psfrag{a}{\Huge $a$}

4187: \psfrag{b}{\Huge $b$}

4188: \resizebox{.7 in}{.3 in}{\includegraphics{gr.eps}}

4189: \end{center}

4190: \end{minipage}

4191: .

4192: \end{equation}

4193: %

4194: 

4195: 

4196: \subsection{Graph homology for groups}  \label{ss:gp}

4197: 

4198: For the above species $Q$, the corresponding graphs

4199: are necessarily bivalent (polygons),

4200: with each vertex labelled (more or less) by a group element.

4201: The vertices labelled by the unit element 1 are fake

4202: while the rest are real.

4203: 

4204: We now explain how an edge contraction (mating) works.

4205: \[

4206: \begin{minipage}{1.2 in}

4207: \begin{center}

4208: \psfrag{g}{\Huge $g$}

4209: \psfrag{h}{\Huge $h$}

4210: \psfrag{e}{\Huge $e$}

4211: \resizebox{1.2 in}{.6 in}{\includegraphics{gmi.eps}}

4212: \end{center}

4213: \end{minipage}

4214: \qquad \buildrel {\text{contract} \ e}\over \longmapsto \qquad

4215: \begin{minipage}{1.1 in}

4216: \begin{center}

4217: \psfrag{gh}{\Huge $gh$}

4218: \resizebox{1 in}{.6 in}{\includegraphics{gmo.eps}}

4219: \end{center}

4220: \end{minipage}

4221: .

4222: \]

4223: Apply relation~\eqref{e:gr} if necessary,

4224: so that both vertices incident to the edge $e$ point in the same direction.

4225: Contract it and make the new vertex also point in the same direction.

4226: And label it with the product of the two labels

4227: (in the order specified by the direction).

4228: 

4229: We will call the homology of this bivalent complex

4230: as the graph homology of the group $K$

4231: and denote it $H_*(K,^{*})$.

4232: In the notation of Section~\ref{s:gh},

4233: it coincides with both $H_*(\mathcal C)$ and $H_*(\mathcal B)$ .

4234: 

4235: 

4236: \subsection{Graph homology computations}  \label{ss:cal}

4237: 

4238: The problem of computing graph homology 

4239: seems to be difficult in general.

4240: The only instance where computations 

4241: have been made is the commutative case (\ref{sss:c});

4242: see the thesis of Ferenc Gerlits~\cite{\ferenc}.

4243: This is the case where we are dealing with usual graphs.

4244: 

4245: We give two instances where graph homology $H_*(\mathcal C)$

4246: can be completely computed.

4247: They are the first two cases of the example

4248: under discussion.

4249: It is easy to generalise them but we do not do it here.

4250: 

4251: \subsubsection{The trivial group} \label{sss:ctg}

4252: 

4253: If $K$ is the trivial group then $P=u$ and $Q=uu$,

4254: see (\ref{ss:uu}).

4255: The computation for this case,

4256: which coincides with $H_*(\mathcal P)$, see~(\ref{ss:cc}),

4257: is discussed in~\cite{\god}.

4258: The graphs are polygons with only fake vertices.

4259: These graphs have two kinds of automorphisms: rotations and reflections.

4260: It is easy to see that a simple rotation is orientation reversing if and

4261: only if $k$, the number of vertices, is even, and a reflection reverses

4262: the orientation iff $k \equiv 1$ or $2$ modulo $4$.  So there is one

4263: non-zero chain in each degree of the form $4i+3$, and hence all the

4264: boundary maps are zero, therefore these are the homology groups also.

4265: Hence 

4266: $$

4267:   H_k(\{1\},\text{identity}) \ = \left\{ \begin{array}{c l}

4268:     \Q & \mbox{ if $k \equiv 3 \ (\text{mod} \ 4)$}, \\

4269:     0  & \mbox{ otherwise.} 

4270:                     \end{array} \right. $$

4271: %

4272: By Corollary to Theorem~\ref{t:con} (Section~\ref{s:con}),

4273: this is same as the primitive homology

4274: of the Lie algebra ${{{\mathfrak{sp}}(2 \infty)}}$.

4275: 

4276: 

4277: \subsubsection{The group $\Z_2$}  

4278: 

4279: For this case, the graphs are polygons with two types of vertices

4280: corresponding to the two group elements.

4281: Computing the homology directly from the definition is not easy.

4282: So we rerun the spectral sequence argument

4283: (Proposition~\ref{p:ss})

4284: with some modification.

4285: The stable page is $E^2$ with non-zero terms

4286: only on the positive X and Y axis.

4287: The conclusion is

4288: $$

4289:   H_k(\Z_2,\text{identity}) \ = \left\{ \begin{array}{c l}

4290:     \Q \oplus \Q & \mbox{ $k \equiv 3 \ (\text{mod} \ 4)$}, \\

4291:     0           & \mbox{ otherwise.} 

4292:                     \end{array} \right. $$

4293: This example, we hope, gives an idea 

4294: of the complexity of computing graph homology.

4295: 

4296: 

4297: 

4298: 

4299: \section{Graph cohomology}  \label{s:mgh}

4300: 

4301: The ideas in this section are based on a letter of Kontsevich.

4302: It leads to the birth of many interesting operations

4303: on graphs.

4304: I thank Jim Conant for helping me understand its contents

4305: and also for providing his notes related to this.

4306: 

4307: We begin by defining graph cohomology.

4308: The homology and cohomology are related 

4309: by an interesting and highly non-trivial pairing on graphs.

4310: This is explained in (\ref{ss:pg}-\ref{ss:ad}).

4311: Later in (\ref{ss:dmg}),

4312: we use it to define a deformation map on graphs.

4313: These ideas will be used in Section~\ref{s:sta} 

4314: in the proof of Theorem~\ref{t:sta}

4315: that deals with stability.

4316: Throughout this section,

4317: we assume that $Q$ is a mated species based on sets. 

4318: 

4319: \subsection{The blowup coboundary operator $\delta_E$}  \label{ss:stacb}

4320: 

4321: Let $\Gamma$ be an oriented $Q$-graph and

4322: $I(\Gamma)$ 

4323: be the set of its ideal edges.

4324: These are the edges that are ``present''

4325: in the internal structure of the vertices of $\Gamma$,

4326: see (\ref{ss:mf}).

4327: 

4328: \begin{definition}  

4329: The coboundary map $\delta_E : \mathcal G_k\to \mathcal G_{k+1}$

4330: is defined using ideal edge expansions, i.e. breakups.

4331: It is given by the formula

4332: 

4333: $$\delta_E (\Gamma,\sigma) = \sum_{e \in I(\Gamma)} (\Gamma \backslash e,

4334: \sigma \backslash e),$$

4335: %

4336: where $\Gamma \backslash e$ is the graph $\Gamma$ 

4337: with the ideal edge $e$ expanded

4338: (see figure), and 

4339: $\sigma \backslash e$ is obtained the following way: 

4340: choose a representative of $\sigma$ 

4341: where the vertex of the ideal edge $e$ has label 1,

4342: give the two new vertices arising from the breakup

4343: the labels 1 and 2,

4344: add 1 to the label of each of the other vertices, and

4345: direct the newly formed edge $e$ from vertex 1 to 2;

4346: finally, keep the

4347: orientations on the other edges unchanged.

4348: \[

4349: \begin{minipage}{\linewidth}

4350: \begin{center}

4351: \begin{minipage}{1.2 in}

4352: \begin{center}

4353: \psfrag{1}{\Huge $1$}

4354: \psfrag{2}{\Huge $2$}

4355: \psfrag{e}{\Huge $e$}

4356: \resizebox{1 in}{.6 in}{\includegraphics{cbi.eps}}

4357: \end{center}

4358: \end{minipage}

4359: $\qquad \buildrel {\text{expand} \ e}\over \longmapsto \qquad$

4360: \begin{minipage}{1.2 in}

4361: \begin{center}

4362: \psfrag{1}{\Huge $1$}

4363: \psfrag{2}{\Huge $2$}

4364: \psfrag{e}{\Huge $e$}

4365: \resizebox{1 in}{.6 in}{\includegraphics{cbo.eps}}

4366: \end{center}

4367: \end{minipage}\\

4368: \vspace{.1in}

4369: A local picture of a breakup along an ideal edge $e$.

4370: \end{center}

4371: \end{minipage}

4372: \]

4373: \end{definition}

4374: %

4375: Similarly, one can define a coboundary operator

4376: $\delta_H$ as the co-analogue of $\partial_H$,

4377: see (\ref{ss:finb}).

4378: We will not deal with $\delta_H$ in this paper.

4379: 

4380: \subsection{A family of pairings $M(n)$}  \label{ss:pg}

4381: 

4382: For every $n$,

4383: we will define a pairing

4384: $M(n) : \mathcal G \otimes \mathcal G \to \Q$.

4385: The definition, though explicit, will be somewhat complicated.

4386: A way to understand it is given in

4387: (\ref{ss:pl}-\ref{ss:pr}), where we derive it 

4388: as a restriction of a simpler pairing

4389: defined on a larger space.

4390: The reader, who is more interested in Lie algebras

4391: or the main theorem rather than just graphs,

4392: may read that part first,

4393: referring back as necessary.

4394: 

4395: Let $(\Gamma_1,\sigma_1),(\Gamma_2,\sigma_2) \in \mathcal G$

4396: be two oriented graphs.

4397: For simplicity,

4398: we suppress orientations from the notation.

4399: 

4400: \begin{definition}  \label{d:gm}

4401: A matching $m:\Gamma_1 \to \Gamma_2$ is a bijection

4402: between the vertex sets

4403: $V(\Gamma_1)$ and $V(\Gamma_2)$

4404: that preserves the internal structure of the vertices.

4405: We imagine this as an overlaying of the vertex sets of

4406: $\Gamma_1$ and $\Gamma_2$.

4407: \[

4408: \begin{minipage}{1.6 in}

4409: \begin{center}

4410: \begin{minipage}{1.2 in}

4411: \begin{center}

4412: \resizebox{1.2 in}{1.2 in}{\includegraphics{mr.eps}}

4413: \end{center}

4414: \end{minipage}\\

4415: \vspace{.1in}

4416: $\Gamma_1$

4417: \end{center}

4418: \end{minipage}

4419: \begin{minipage}{1.6 in}

4420: \begin{center}

4421: \begin{minipage}{1.2 in}

4422: \begin{center}

4423: \resizebox{.7 in}{1.2 in}{\includegraphics{ms.eps}}

4424: \end{center}

4425: \end{minipage}\\

4426: \vspace{.1in}

4427: $\Gamma_2$

4428: \end{center}

4429: \end{minipage}

4430: \begin{minipage}{1.6 in}

4431: \begin{center}

4432: \begin{minipage}{1.2 in}

4433: \begin{center}

4434: \psfrag{d}{\Huge $d$}

4435: \psfrag{c}{\Huge $c$}

4436: \psfrag{b}{\Huge $b$}

4437: \psfrag{a}{\Huge $a$}

4438: \resizebox{1.2 in}{1.2 in}{\includegraphics{m.eps}}

4439: \end{center}

4440: \end{minipage}\\

4441: \vspace{.1in}

4442: Matching of $\Gamma_1$ and $\Gamma_2$.

4443: \end{center}

4444: \end{minipage}

4445: \]

4446: An example of a matching is shown above.

4447: For clarity, orientations 

4448: and the internal structure of the vertices

4449: have been omitted.

4450: It is clear that for a matching to exist,

4451: $\Gamma_1$ and $\Gamma_2$

4452: must have the same number of vertices and edges.

4453: \end{definition}

4454: 

4455: \begin{definition} \label{d:csm}

4456: We now define the number of components 

4457: $c(m)$ and $\sign(m)$

4458: of a matching

4459: $m:\Gamma_1 \to \Gamma_2$.

4460: Delete the vertices from the overlaying of

4461: $\Gamma_1$ and $\Gamma_2$ specified by $m$.

4462: We are left with a disjoint union of even sided polygons

4463: whose edges alternate between those of

4464: $\Gamma_1$ and $\Gamma_2$.

4465: The quantity $c(m)$ counts the number of these polygons.

4466: In the example above, we get a hexagon and a square;

4467: so $c(m)=2$.

4468: \[

4469: \begin{minipage}{.7 in}

4470: \begin{center}

4471: \psfrag{a}{\Huge $d$}

4472: \psfrag{b}{\Huge $a$}

4473: \psfrag{c}{\Huge $c$}

4474: \psfrag{d}{\Huge $d$}

4475: \psfrag{e}{\Huge $c$}

4476: \psfrag{f}{\Huge $b$}

4477: \resizebox{.7 in}{.7 in}{\includegraphics{ch.eps}}

4478: \end{center}

4479: \end{minipage}

4480: \hspace{1 in}

4481: \begin{minipage}{.6 in}

4482: \begin{center}

4483: \psfrag{a}{\Huge $b$}

4484: \psfrag{b}{\Huge $b$}

4485: \psfrag{c}{\Huge $a$}

4486: \psfrag{d}{\Huge $d$}

4487: \resizebox{.5 in}{.5 in}{\includegraphics{cs.eps}}

4488: \end{center}

4489: \end{minipage}

4490: \]

4491: The purpose of $\sign(m)$ is to take 

4492: orientations into account.

4493: Choose representatives for $\sigma_1$ and $\sigma_2$.

4494: The matching $m$ gives a bijection of the vertex sets.

4495: First note the sign of this permutation.

4496: Then for each polygon, fix say the clockwise direction

4497: for $\Gamma_1$ and anticlockwise direction for $\Gamma_2$. 

4498: And write a minus sign for all edges that are out of order.

4499: The product of all signs gives $\sign(m)$.

4500: \end{definition}

4501: 

4502: 

4503: \begin{definition}  

4504: Define the pairing 

4505: $M(n) : \mathcal G \otimes \mathcal G \to \Q$ by the formula 

4506: \begin{equation}  \label{e:pg}

4507: M(n)(\Gamma_1,\Gamma_2) = \underset{m:\Gamma_1 \to \Gamma_2}

4508: {\sum} \sign(m) (2n)^{c(m)},

4509: \end{equation}

4510: where the sum is over all matchings

4511: $m:\Gamma_1 \to \Gamma_2$

4512: and $\sign(m)$ and $c(m)$ 

4513: are the sign and the number of components of the matching $m$.

4514: \end{definition}

4515: 

4516: \begin{remark}

4517: For $\Gamma_1$ and $\Gamma_2$ fixed,

4518: the maximum value of $c(m)$ is

4519: $e=$ number of edges of $\Gamma_1$ (and $\Gamma_2$).

4520: This happens iff every polygon in the matching $m$

4521: has exactly two sides, i.e.

4522: iff $m:\Gamma_1 \to \Gamma_2$ is an isomorphism.

4523: We will call this a perfect matching.

4524: And in this case, 

4525: the coefficient of $(2n)^e$, upto sign, is

4526: $|\Aut(\Gamma_1)| = |\Aut(\Gamma_2)|$.

4527: %$|\text{Aut}(\Gamma_1)| = |\text{Aut}(\Gamma_2)|$.

4528: \end{remark}

4529: 

4530: \subsection{The adjoint property} \label{ss:ad}

4531: 

4532: We now relate the finite boundary operator $\partial_n$

4533: in (\ref{ss:finb})

4534: to the (stable) coboundary operator $\delta_E$

4535: in (\ref{ss:stacb}).

4536: There is a simpler relation between $\partial_E$

4537: and $\delta_E$ which can be derived from this one,

4538: by looking at the leading coefficient.

4539: This is explained in (\ref{ss:dmg}).

4540: Recall that $\partial_n$ is defined by contracting quasi-edges

4541: and $\delta_E$ is defined by expanding ideal edges.

4542: 

4543: \begin{proposition}  \label{p:ad}

4544: The maps $\partial_n$ and $\delta_E$ are adjoints

4545: with respect to the pairing $M(n)$. In other words,

4546: $M(n)(\partial_n \Gamma_1,\Gamma_2)=M(n)(\Gamma_1,\delta_E \Gamma_2).$

4547: \end{proposition}

4548: \begin{proof}

4549: To prove the above identity, 

4550: we express both sides as weighted state sums,

4551: and then give a bijection of the state space in the LHS

4552: with that in the RHS such that it respects weights.

4553: Note that for either side to be nonzero,

4554: $\Gamma_1$ must have one vertex and edge more than $\Gamma_2$.

4555: 

4556: Define a state $S_L$ in the LHS to be a pair $(e,m)$

4557: of a quasi-edge $e \in E_q(\Gamma_1)$

4558: and a matching $m:\Gamma_1 /e \to \Gamma_2$.

4559: Then by the definition of the boundary map $\partial_n$ 

4560: and the pairing $M(n)$, the

4561: LHS = $M(n)(\partial_n \Gamma_1,\Gamma_2)=\sum_{S_L} w(S_L), \ 

4562: \text{where}$

4563: $$

4564:   w(S_L) \ = \left\{ \begin{array}{c l}

4565:     \sign(m) (2n)^{c(m)+1} &

4566: \mbox{ if $e \in E(\Gamma_1)$}, \\

4567:     \sign(m) (2n)^{c(m)} &

4568: \mbox{ if $e \in E_q(\Gamma_1) \setminus E(\Gamma_1).$}

4569:                     \end{array} \right. $$

4570: An example of a state $S_L$ for

4571: $e \in E_q(\Gamma) \setminus E(\Gamma)$

4572: is shown in the figure.

4573: The quasi-edge $f$ is the partner of $e$.

4574: The graph $\Gamma_2$ is not shown separately,

4575: since it is visible from the overlaying.

4576: Also, in the graph $\Gamma_1 /e$,

4577: the edge $e$ has been shown as an ideal edge.

4578: \[

4579: \begin{minipage}[b]{2.4 in}

4580: \begin{center}

4581: \psfrag{e}{\Huge $e$}

4582: \psfrag{f}{\Huge $f$}

4583: \resizebox{1 in}{1 in}{\includegraphics{gone.eps}}\\

4584: $\Gamma_1$ with a quasi-edge $e$.

4585: \end{center}

4586: \end{minipage}

4587: \begin{minipage}[b]{2.4 in}

4588: \begin{center}

4589: \psfrag{e}{\Huge $e$}

4590: \psfrag{f}{\Huge $f$}

4591: \resizebox{1.4 in}{.8 in}{\includegraphics{gml.eps}}\\

4592: $S_L$ = A matching of $\Gamma_1 /e$ and $\Gamma_2$.

4593: \end{center}

4594: \end{minipage}

4595: \]

4596: Similarly,

4597: a state $S_R$ in the RHS is defined to be a pair $(e,m)$

4598: of an ideal edge $e \in I(\Gamma_2)$

4599: and a matching $m:\Gamma_1 \to \Gamma_2 \backslash e$.

4600: Then by definition the

4601: RHS = $M(n)(\Gamma_1,\delta_E(\Gamma_2))=\sum_{S_R} w(S_R), \ 

4602: \text{where}$

4603: $w(S_R) = \sign(m) (2n)^{c(m)}$.

4604: 

4605: We now indicate the bijection between the two state spaces

4606: by continuing our example.

4607: We draw the state $S_R$ 

4608: that corresponds to the state $S_L$ shown in the figure above.

4609: \[

4610: \begin{minipage}[b]{2.4 in}

4611: \begin{center}

4612: \psfrag{e}{\Huge $e$}

4613: \resizebox{1 in}{.8 in}{\includegraphics{gtwo.eps}}\\

4614: $\Gamma_2$ with an ideal edge $e$.

4615: \end{center}

4616: \end{minipage}

4617: \begin{minipage}[b]{2.4 in}

4618: \begin{center}

4619: \psfrag{e}{\Huge $e$}

4620: \resizebox{1 in}{1 in}{\includegraphics{gmr.eps}}\\

4621: $S_R$ = A matching of $\Gamma_1$ and $\Gamma_2 \backslash e$.

4622: \end{center}

4623: \end{minipage}

4624: \]

4625: The two pictures that represent 

4626: the states $S_L$ and $S_R$

4627: are identical except for the local behaviour at the edge $e$.

4628: 

4629: In general, to see that $w(S_L) = w(S_R)$,

4630: note that a state $S_R$ is of two kinds 

4631: depending on whether the ideal edge $e \in I(\Gamma_2)$

4632: is overlayed on a quasi-edge

4633: $e \in E(\Gamma_1)$ or $e \in E_q(\Gamma_1) \setminus E(\Gamma_1).$

4634: In the first case, $S_R$ has one more component than $S_L$

4635: while in the second they are the same.

4636: We illustrated the second case in our pictures.

4637: This fits in with the two cases for $w(S_L)$.

4638: It can also be checked that the signs work out correctly.

4639: \end{proof}

4640: 

4641: 

4642: \subsection{Non-degeneracy of the pairing}  \label{ss:pnd}

4643: 

4644: Let $\mathcal G_k^e$ be the span of all oriented graphs

4645: with $k$ vertices and $e$ edges.

4646: Observe that the subspaces $\mathcal G_k^e$ of $\mathcal G$,

4647: as $k$ and $e$ vary,

4648: are mutually orthogonal with respect to the pairing $M(n)$.

4649: By our earlier notation (\ref{ss:is}),

4650: we have

4651: $\mathcal G_k^e=\mathcal G_k^{(r)}$,

4652: with $r=1- \chi=e-k+1$.

4653: The notation $\mathcal G_k^e$ is local to this section

4654: and is introduced to improve clarity.

4655: 

4656: For any $k$ and $e$, 

4657: the pairing $M(n)$ defines a map $\mathcal G_k^e \to (\mathcal G_k^e)^*$,

4658: which we write $M_k^e(n)$.

4659: Regard $M_k^e(n)$

4660: as a matrix whose rows and columns are indexed by 

4661: (isomorphism classes of) oriented graphs $\Gamma$

4662: with $k$ vertices and $e$ edges and

4663: where the $(\Gamma_i,\Gamma_j)$ entry is given by

4664: $<M_k^e(n)(\Gamma_i),\Gamma_j>$.

4665: By the definition of the pairing $M(n)$, 

4666: the entries of this matrix 

4667: are polynomials in $2n$ of degree $\leq e$

4668: and moreover the degree $e$ entries

4669: are precisely the diagonal entries.

4670: Furthermore, the coefficient of $(2n)^e$ of

4671: $<M_k^e(n)(\Gamma),\Gamma>$ is just $|\Aut(\Gamma)|$.

4672: This is clear from the remark at the end of

4673: (\ref{ss:pg}).

4674: 

4675: With $e$ and $k$ fixed and $n$ large enough,

4676: the diagonal entries dominate,

4677: so the matrix $M_k^e(n)$ is invertible, and

4678: $\mathcal G_k^e \isoto (\mathcal G_k^e)^*$.

4679: In other words,

4680: the pairing $M(n)$ is non-degenerate

4681: for $e$ and $k$ fixed and $n$ large enough.

4682: 

4683: \subsection{The deformation map $D(n)$} \label{ss:dmg}

4684: 

4685: The adjoint property of $\partial_n$ and $\delta_E$

4686: established in Proposition~\ref{p:ad} says that

4687: \begin{equation}  \label{e:ad}

4688: M_{k-1}^{e-1}(n)

4689: \circ \partial_n = 

4690: \delta_E^* \circ M_k^e(n).

4691: \end{equation}

4692: Let $A_k^e$ denote the diagonal matrix with $(\Gamma,\Gamma)$

4693: entry equal to $|\Aut(\Gamma)|$.

4694: Then the degree $e$ part of the above equation is just

4695: \begin{equation}  \label{e:sad}

4696: A_{k-1}^{e-1}

4697: \circ \partial_E = 

4698: \delta_E^* \circ A_k^e.

4699: \end{equation}

4700: Equation~\eqref{e:ad} (resp.~\eqref{e:sad}) shows the precise sense in which

4701: $\partial_n$ (resp $\partial_E$) 

4702: is dual to $\delta_E$.

4703: The results for $n$ large enough can be summarised 

4704: in the following commutative diagram.

4705: \begin{equation*}

4706: \begin{CD}

4707: \mathcal G_k^e @>M_k^e(n)>\iso> (\mathcal G_k^e)^* @<A_k^e<\iso< \mathcal G_k^e \\ 

4708: @V{2n \partial_E + \partial_H}VV @V{\delta_E^*}VV @V{\partial_E}VV\\

4709: \mathcal G_{k-1}^{e-1} @>\iso>M_{k-1}^{e-1}(n)> (\mathcal G_{k-1}^{e-1})^* @<\iso<A_{k-1}^{e-1}< \mathcal G_{k-1}^{e-1}.\\ 

4710: \end{CD}

4711: \end{equation*}

4712: %

4713: Now switching back to our earlier notation, 

4714: the above commutative diagram can be rewritten as

4715: \begin{equation}  \label{e:cdg}

4716: \begin{CD}

4717: \mathcal G_k^{(r)} @>M(n)>\iso> (\mathcal G_k^{(r)})^* @<A<\iso< \mathcal G_k^{(r)} \\ 

4718: @V{2n \partial_E + \partial_H}VV @V{\delta_E^*}VV @V{\partial_E}VV\\

4719: \mathcal G_{k-1}^{(r)} @>\iso>M(n)> (\mathcal G_{k-1}^{(r)})^* @<\iso<A< \mathcal G_{k-1}^{(r)}.\\ 

4720: \end{CD}

4721: \end{equation}

4722: We point out that $M(n)$ is used to denote 

4723: both the pairing and the map it induces.

4724: 

4725: \begin{corollary}

4726: Let $k$ and $r$ be fixed. Then for $n$ large enough

4727: $$H_k(\mathcal G^{(r)},\partial_n) = H_k(\mathcal G^{(r)},\partial_E).$$

4728: \end{corollary}

4729: %

4730: \noindent

4731: We will call the map $G^{(r)} \to G^{(r)}$,

4732: which induces the above isomorphism,

4733: the deformation map $D(n)$.

4734: It is the map $M(n)$ followed by $A^{-1}$.

4735: We note that for a graph $\Gamma$,

4736: its image $D(n) \Gamma$

4737: is a polynomial in $2n$ of degree $e$

4738: with coefficients in $\mathcal G$.

4739: Furthermore, the coefficient of $(2n)^e$ is exactly $\Gamma$.

4740: Here $e=|E(\Gamma)|$, the number of edges in $\Gamma$.

4741: The deformation map is a key step in the proof of 

4742: the main theorem.

4743: We will also study it briefly in Appendix~\ref{ss:dg}.

4744: 

4745: 

4746: 

4747: \section{The main theorem} \label{s:mt} 

4748: 

4749: In this section,

4750: we give a precise statement of the main theorem.

4751: We will also discuss the classical case briefly.

4752: First recall some facts from Section~\ref{s:cm}.

4753: 

4754: Let $P$ be a reversible operad and 

4755: $Q$ be its mated species.

4756: There are two algebraic objects $PA$ and $QA$

4757: associated to $P$ and $Q$ respectively.

4758: One thinks of $QA$ as 

4759: ``Hamiltonian functions on the symplectic $P$-manifold''.

4760: It has the structure of a Lie algebra.

4761: The Lie algebra $QA$

4762: depends on the dimension of the symplectic $P$-manifold.

4763: So we write $QA_n$ when the dimension is $2n$.

4764: The dimension is necessarily even.

4765: We then have a sequence of Lie algebra inclusions

4766: $$QA_1 \subset \ldots \subset QA_n \subset QA_{n+1} \subset \ldots$$

4767: We denote the direct limit by $QA_{\infty}$.

4768: And $QA_n$ always contains the symplectic Lie algebra

4769: $\fsp$ as an anti-subalgebra.

4770: We write

4771: ${{{\mathfrak{sp}}(2 \infty)}}$

4772: for the corresponding direct limit.

4773: The rational homology of $QA_{\infty}$, 

4774: which we denote $H_*(QA_{\infty})$,

4775: has the structure of a Hopf algebra.

4776: We write $PH_*(QA_{\infty})$

4777: for the subspace of primitive elements.

4778: 

4779: \subsection{Statement of the main theorem}

4780: Kontsevich's result can now be stated as

4781: 

4782: \begin{theorem}  \label{t:mt}

4783: $PH_*(QA_{\infty}) = 

4784: H_*(Q\mathcal G,\partial_E) \oplus

4785: PH_*({{{\mathfrak{sp}}(2 \infty)}}).$

4786: \end{theorem}

4787: 

4788: \noindent

4789: The term $H_*(Q\mathcal G,\partial_E)$

4790: is the graph homology of the mated species $Q$.

4791: It is the homology of the chain complex

4792: $(Q\mathcal G,\partial_E)$ of graphs

4793: that was defined in Section~\ref{s:gh}.

4794: 

4795: There are two conditions that we require in the theorem.

4796: We assume that the operad $P$ is based on sets

4797: rather than vector spaces.

4798: This is because the proof involves a pairing on graphs

4799: (\ref{ss:pg}),

4800: which we know how to define only in the former case.

4801: We hope that eventually this restriction

4802: would not be necessary.

4803: Hence, as of now,

4804: the theorem cannot be applied to the Lie operad,

4805: which is one of the cases claimed in~\cite{\god}.

4806: 

4807: Secondly, we assume that $P[1]$ 

4808: is a singleton consisting of the unit element $u$,

4809: see (\ref{ss:o}).

4810: This is done mainly for simplicity.

4811: If we drop this assumption 

4812: then the summand $PH_*({{{\mathfrak{sp}}(2 \infty)}})$

4813: has to be replaced by the homology of the bivalent graph complex

4814: $(\mathcal B,\partial_E)$ defined in Section~\ref{s:gh}.

4815: 

4816: \begin{remark}

4817: The primitive homology of ${{{\mathfrak{sp}}(2 \infty)}}$

4818: is known (\ref{sss:ctg}).

4819: It will be computed in the course of proving the theorem.

4820: \end{remark}

4821: 

4822: \subsection{The classical case} \label{ss:ecc}

4823: 

4824: This is the commutative case, $P=c$ and $Q=cc$,

4825: see (\ref{ss:tcc}).

4826: In this case, the ``symplectic operad manifold''

4827: actually exists and is simply $(\R^{2n},\omega_0)$.

4828: The Lie algebra $QA_n$,

4829: which we write as $ccA_n$, consists of 

4830: polynomial functions in $2n$ variables 

4831: with no constant or linear terms. 

4832: The Lie structure is given by

4833: the usual Poisson bracket 

4834: (equation~\eqref{e:pb} in Section~\ref{s:sog}).

4835: There is an evident subalgebra $uuA_n$

4836: consisting of all homogeneous polynomials

4837: of degree 2.

4838: The reason for this notation is that 

4839: $uuA_n$ can also be seen as an example of the theory

4840: with $P=u$ and $Q=uu$, see (\ref{ss:sl}).

4841: The Lie algebra $uuA_n$ is anti-isomorphic to 

4842: the symplectic Lie algebra $\fsp$.

4843: 

4844: It is also clear that we have a sequence of Lie algebra inclusions

4845: $$ccA_1 \subset \ldots \subset ccA_n \subset ccA_{n+1} \subset \ldots$$

4846: The direct limit $ccA_{\infty}$ consists of finite polynomials

4847: in infinitely many variables 

4848: $p_1,p_2,\ldots,q_1,q_2,\ldots$.

4849: 

4850: \section{Proof of the main theorem-Part I}  \label{s:fin}

4851: 

4852: We now begin the proof of the main theorem.

4853: It will be done in three steps.

4854: In this section, we take the first step of relating 

4855: the homology of a Lie algebra to graph homology.

4856: We will prove the following theorem.

4857: 

4858: \begin{theorem}  \label{t:fin}

4859: $H_*(QA_n) = H_*(\mathcal G,\partial_n).$

4860: \end{theorem}

4861: 

4862: \noindent

4863: For definitions of the above terms, see

4864: (\ref{ss:fa}-\ref{ss:mate}) and (\ref{ss:gc}-\ref{ss:finb}).

4865: For the commutative case, 

4866: the Lie algebra $QA_n$ is easy to define, see (\ref{ss:ecc}).

4867: The main ideas of the proof are already present 

4868: in the commutative case.

4869: Hence the reader may specialise to this case

4870: on a first reading.

4871: 

4872: \begin{corollary}

4873: $H_*(\fsp) = H_*(\mathcal F,\partial_n).$

4874: \end{corollary}

4875: \begin{proof}

4876: We apply the theorem to the unit species, i.e. $Q=uu$.

4877: In this case, $QA_n=uuA_n$, which is anti-isomorphic to $\fsp$,

4878: see (\ref{ss:sl}).

4879: Due to the trivial nature of the species,

4880: the graphs have only fake (bivalent) vertices.

4881: Hence the chain complex $\mathcal G$,

4882: in this case, 

4883: is simply the fake bivalent complex $\mathcal F$,

4884: see (\ref{ss:cc}). 

4885: \end{proof}

4886: 

4887: We now start the proof of Theorem~\ref{t:fin}.

4888: It is best summarised in the following Kontsevich sentence.

4889: \begin{quotation}

4890: The spirit of the (quite simple) computations

4891: is somewhere between Gelfand-Fuks computations

4892: (see~\cite{\gelfand} and~\cite{\fuks}) and cyclic homology.

4893: \end{quotation}

4894: 

4895: 

4896: \subsection{Lie algebra homology}  \label{ss:lah}

4897: 

4898: A good introduction to Lie algebra homology

4899: can be found in Weibel's book~\cite[Chapter 7]{\weibel}.

4900: Recall that the homology of the Lie algebra $QA_n$ 

4901: can be computed using

4902: the Chevalley-Eilenberg or standard complex

4903: $$\ldots\longrightarrow \mathcal C_{k+1}\buildrel \partial_{k+1}\over \longrightarrow 

4904: \mathcal C_{k}\buildrel 

4905: \partial_{k}\over \longrightarrow \mathcal C_{k-1}\buildrel \partial_{k-1}\over

4906: \longrightarrow \ldots $$ with

4907: $\mathcal C_k=\Lambda^k(QA_n)$ 

4908: and 

4909: \begin{equation}  \label{e:lb}

4910: \partial_k(F_1 \wedge \ldots \wedge F_k)

4911: =\sum_{1\leq s<t\leq k}(-1)^{s+t-1}\{F_s,F_t\}\wedge F_1\ldots \hat 

4912: F_s\ldots \hat

4913: F_t\ldots

4914: \wedge F_k.

4915: \end {equation}

4916: The boundary operator commutes with the action

4917: of the Lie algebra $QA_n$ 

4918: on the exterior powers $\Lambda^k(QA_n)$. 

4919: 

4920: We have learnt to think of the Lie bracket $\{\ ,\ \}$ on $QA_n$

4921: as a mating.

4922: For the commutative case, this was discussed in

4923: (\ref{ss:tcc}).

4924: In the formula~\eqref{e:lb},

4925: we take $k$ elements of $QA_n$,

4926: say $F_1,\dots,F_k$, and to apply the boundary map

4927: we do pairwise matings.

4928: Now let us think for a moment about graphs.

4929: The boundary operators on a graph are all defined

4930: using quasi-edge contractions,

4931: which are again matings.

4932: Hence to relate the two notions,

4933: all one needs to do is imagine the elements

4934: $F_1,\dots,F_k$ as being the vertices of a graph.

4935: We now work towards making this precise.

4936: 

4937: \subsection{Passing to the subcomplex of $\fsp$ invariants}  \label{ss:in}

4938: 

4939: Recall from (\ref{ss:fa}) that

4940: \begin{equation}  \label{e:fsa}

4941: QA_n = \bigoplus_{j\geq 2} (Q[j] \otimes V^{\otimes j})_{\Sigma_j}

4942: = \bigoplus_{j\geq 2} QA_n^j.

4943: \end{equation}

4944: In the commutative case, $QA_n^j = S^j(V)$

4945: is the $j$th symmetric tensor power of $V$.

4946: This is just the space of commuting polynomials of degree $j$

4947: in a basis of $V$.

4948:  

4949: The space $QA_n^j$ is a left $\fsp$ module with the action 

4950: induced by the usual action on $V$ and

4951: the trivial action on $Q[j]$, see (\ref{ss:sl}).

4952: Thus, the Lie algebra $QA_n$

4953: is a direct sum of finite dimensional $\fsp$ modules.

4954: Since the Lie algebra $\fsp$ is simple,

4955: it follows that $QA_n$ and $\Lambda^k(QA_n) = \mathcal C_k$ 

4956: are semisimple $\fsp$ modules.

4957: This allows us to write

4958: $\mathcal C_{k} = (\mathcal C_{k})^{\fsp} \oplus \fsp \cdot \mathcal C_k$.

4959: A standard argument now shows that

4960: the subcomplex $\fsp \cdot \mathcal C$ is exact.

4961: Hence

4962: the homology of the standard complex is the same as the 

4963: homology of the subcomplex of $\fsp$-invariants:

4964: $$\ldots\longrightarrow (\mathcal C_{k+1})^{\fsp}\buildrel \partial_{k+1}\over \longrightarrow 

4965: (\mathcal C_{k})^{\fsp}{ \buildrel \partial_{k}\over \longrightarrow} (\mathcal C_{k-1})^{\fsp}\buildrel

4966: \partial_{k-1}\over

4967: \longrightarrow \ldots\qquad $$

4968: 

4969: \begin{remark}

4970: The standard argument is as under.

4971: Let $\mathfrak g$ be a Lie algebra and

4972: $\mathcal C$ be its standard complex.

4973: Then for $\xi \in \mathfrak g$,

4974: define Lie derivative $L_{\xi}$ (action of $\xi$)

4975: and contraction operator $i_{\xi}$ (wedge with $\xi$)

4976: as operators on $\mathcal C$.

4977: The relations of Lemma~\ref{l:rel} (Section~\ref{s:st})

4978: hold in this case also.

4979: Cartan's formula implies that the Lie derivative

4980: is zero on $H_*(\mathcal C)$.

4981: Now let $\mathfrak h$ be a simple subalgebra of $\mathfrak g$

4982: such that $\mathcal C$ is a semisimple $\mathfrak h$ module.

4983: Then the subcomplex

4984: $\mathfrak h \cdot \mathcal C$ is exact.

4985: 

4986: 

4987: For the semisimplicity of $\Lambda^k(QA_n)$,

4988: we used two facts.

4989: Any finite dimensional module of a simple Lie algebra is semisimple.

4990: The tensor product of two finite dimensional semisimple modules 

4991: over any Lie algebra

4992: is again semisimple~\cite[pg 83]{\jacobson}.

4993: \end{remark}

4994: 

4995: \subsection{Passing from invariants to oriented graphs}  \label{ss:inv}

4996: 

4997: We want to relate the above subcomplex of $\fsp$-invariants

4998: to the graph complex $(\mathcal G,\partial_n)$.

4999: To this end, we first try to get a better understanding

5000: of the chain groups $(\wedge^k QA_n)^{\fsp}$.

5001: 

5002: The description of $QA_n$ given by equation~\eqref{e:fsa}

5003: gives us the following formula:

5004: \[

5005: \wedge^k QA_n = \bigoplus_{\tiny \begin{array}{c}k_1+k_2+\ldots+k_r=k\\

5006:                              2\leq j_1<j_2<\ldots <j_r\end{array}}

5007: \left(\wedge^{k_1}QA^{j_1} \otimes \cdots \otimes

5008: \wedge^{k_r}QA^{j_r}\right).

5009: \]

5010: %

5011: We try to understand the $\fsp$ invariants in each summand.

5012: This will be done in three stages.

5013: As a concrete example, we will consider the summand

5014: $\wedge^{1}QA^{2} \otimes \wedge^{2}QA^{3} \otimes 

5015: \wedge^{1}QA^{4}$

5016: and see how the analysis works on it at every stage.

5017: We will illustrate it with the tree species $(Q=tt)$.

5018: 

5019: The term $\wedge^{k_1}QA^{j_1}$ is a quotient of the tensor power

5020: $\left(Q[j_1] \otimes V^{\otimes j_1}\right)^{\otimes k_1}$.

5021: Hence

5022: each summand of $\wedge^kQA_n$ is a quotient of the tensor power

5023: $\otimes_{t=1}^r (Q[j_t] \otimes V^{\otimes j_t})^{\otimes k_t}$. 

5024: In the first two stages,

5025: we will figure out the $\fsp$-invariants in this tensor power.

5026: In the third stage,

5027: we will mod out by appropriate actions of the symmetric groups.

5028: 

5029: \subsubsection{The first stage}

5030: We find the invariants in

5031: $V^{\otimes \sum_{t=1}^r j_tk_t}$. 

5032: By the invariant theory of $\fsp$, we know that

5033: a base for the invariants in $V^{\otimes \sum_{t=1}^r k_tj_t}$ 

5034: is given by

5035: oriented chord diagrams on $\sum_{t=1}^r k_tj_t$ vertices,

5036: if $n$ is sufficiently large;

5037: see~\cite[Appendix F]{\fh}. 

5038: 

5039: For the definition of a chord diagram, see~(\ref{sss:k}).

5040: By an oriented chord diagram,

5041: we mean that each chord is oriented 

5042: and reversing the orientation of a single chord

5043: incurs a minus sign.

5044: \[

5045: \begin{minipage}{\linewidth}

5046: \begin{center}

5047: \psfrag{1}{\Huge $1$}

5048: \psfrag{2}{\Huge $2$}

5049: \psfrag{3}{\Huge $3$}

5050: \psfrag{4}{\Huge $4$}

5051: \psfrag{5}{\Huge $5$}

5052: \psfrag{6}{\Huge $6$}

5053: \psfrag{7}{\Huge $7$}

5054: \psfrag{8}{\Huge $8$}

5055: \psfrag{9}{\Huge $9$}

5056: \psfrag{10}{\Huge $10$}

5057: \psfrag{11}{\Huge $11$}

5058: \psfrag{12}{\Huge $12$}

5059: \resizebox{1.5 in}{1.3 in}{\includegraphics{fs.eps}}\\

5060: An oriented chord diagram on $12$ vertices.

5061: It gives a $\fsp$ invariant in

5062: $V^{\otimes 12} =

5063: (V^{\otimes 2})^{\otimes 1} \otimes

5064: (V^{\otimes 3})^{\otimes 2} \otimes

5065: (V^{\otimes 4})^{\otimes 1}$.

5066: \end{center}

5067: \end{minipage}

5068: \]

5069: %

5070: \subsubsection{On how an oriented chord diagram gives an invariant}

5071: \label{sss:ci}

5072: Each vertex of the diagram represents a tensor factor,

5073: in the order given by the vertex labelling.

5074: For each edge, we put a $p_i$ at the tail of the arrow

5075: and a $q_i$ at the head or 

5076: we put a $q_i$ at the tail of the arrow

5077: and a $p_i$ at the head,

5078: incurring a minus sign as a result.

5079: We then sum over all possible choices to get the invariant.

5080: 

5081: The smallest $\fsp$ invariant lies in $V \otimes V$.

5082: It is given by 

5083: $\sum_{i=1}^n p_i \otimes q_i - q_i \otimes p_i$,

5084: which we represent by the chord diagram

5085: $1 \pointsat 2$.

5086: 

5087: \subsubsection{The second stage}

5088: To get the invariants in 

5089: $\otimes_{t=1}^r (Q[j_t] \otimes V^{\otimes j_t})^{\otimes k_t}$, 

5090: we tensor the space of invariants obtained above by

5091: $\otimes_{t=1}^r Q[j_t]^{\otimes k_t}$. 

5092: This is alright because the $\fsp$ action is trivial 

5093: on the $Q[j]$'s. 

5094: This has the following effect on our picture.

5095: \[

5096: \begin{minipage}{1.8 in}

5097: \begin{center}

5098: \psfrag{1}{\Huge $1$}

5099: \psfrag{2}{\Huge $2$}

5100: \psfrag{3}{\Huge $3$}

5101: \psfrag{4}{\Huge $4$}

5102: \psfrag{5}{\Huge $5$}

5103: \psfrag{6}{\Huge $6$}

5104: \psfrag{7}{\Huge $7$}

5105: \psfrag{8}{\Huge $8$}

5106: \psfrag{9}{\Huge $9$}

5107: \psfrag{10}{\Huge $10$}

5108: \psfrag{11}{\Huge $11$}

5109: \psfrag{12}{\Huge $12$}

5110: \resizebox{1.5 in}{1.3 in}{\includegraphics{ss.eps}}

5111: \end{center}

5112: \end{minipage}

5113: \]

5114: 

5115: \begin{remark}

5116: We have drawn trees inside the circles because

5117: we are illustrating with the tree species.

5118: In the commutative case,

5119: we will instead get ordinary graphs

5120: with directed edges and an order on the set of half-edges.

5121: And in the associative case, we get ribbon graphs

5122: with similar data.

5123: \end{remark}

5124: 

5125: \subsubsection{The third stage}  

5126: 

5127: Moding out the actions of the symmetric groups

5128: has the following effect on our picture.

5129: Again, we are illustrating with the tree case.

5130: \[

5131: \begin{minipage}{1.8 in}

5132: \begin{center}

5133: \psfrag{1}{\Huge $1$}

5134: \psfrag{2}{\Huge $2$}

5135: \psfrag{3}{\Huge $3$}

5136: \psfrag{4}{\Huge $4$}

5137: \resizebox{1.5 in}{1.3 in}{\includegraphics{ts.eps}}

5138: \end{center}

5139: \end{minipage}

5140: \]

5141: \noindent

5142: Firstly, we have removed all the labels on the half-edges

5143: and instead given an ordering to the vertices.

5144: This is because we are moding out the action of the $\Sigma_j$'s.

5145: Secondly, we must now interpret the order on the vertices

5146: of the graph in the sense of orientation.

5147: That is, if we interchange the order of two consecutive vertices

5148: then we pick a minus sign.

5149: This is due to the presence of the wedges. 

5150: What we are left with is precisely an oriented $Q$-graph,

5151: see (\ref{ss:og}).

5152: Note that the valence of the vertices is at least 2,

5153: since the grading on $QA$ begins at 2.

5154: Thus we get an isomorphism of chain groups

5155: $(\mathcal C_k)^{\fsp} \cong \mathcal G_k$,

5156: see (\ref{ss:gc}).

5157: 

5158: \subsection{Comparison of the boundary maps of the two complexes}  

5159: \label{ss:cb}

5160: 

5161: To complete the proof of Theorem~\ref{t:fin},

5162: we need to show that the following diagram commutes.

5163: \begin{equation}  \label{e:comp}

5164: \begin{CD}

5165: \mathcal G_k @>\iso>> (\mathcal C_k)^{\fsp}\\

5166: @V{2n \partial_E + \partial_H}VV @V{\partial}VV\\

5167: \mathcal G_{k-1} @>\iso>> (\mathcal C_{k-1})^{\fsp}.\\

5168: \end{CD}

5169: \end{equation}

5170: %

5171: Define 

5172: $\partial^{\prime} : \mathcal G_k \rightarrow \mathcal G_{k-1}$

5173: as the composite of three maps in the above diagram

5174: (the middle map is $\partial$).

5175: We want to show that

5176: $\partial^{\prime} = 2n \partial_E + \partial_H$.

5177: For this, we first understand the map

5178: $\mathcal G_k \rightarrow (\mathcal C_k)^{\fsp}$ better.

5179: 

5180: \subsubsection{On how an oriented graph gives an invariant}  \label{sss:gi}

5181: 

5182: Starting with an oriented graph with $k$ vertices,

5183: we want to construct a $\fsp$ invariant in $\mathcal C_k=\Lambda^k(QA_n)$.

5184: The description that we give follows

5185: directly from the one that we gave 

5186: for an oriented chord diagram (\ref{sss:ci}).

5187: 

5188: Let $(\Gamma,\sigma)$ be an oriented graph.

5189: Choose a representative for $\sigma$.

5190: This means that the vertices of $\Gamma$ are ordered 

5191: and the edges are oriented.

5192: Each vertex of the graph represents a tensor factor,

5193: in the order given by the vertex labelling.

5194: For each edge, we put a $p_i$ at the tail of the arrow

5195: and a $q_i$ at the head or 

5196: we put a $q_i$ at the tail of the arrow

5197: and a $p_i$ at the head,

5198: but picking a minus sign.

5199: This is called a state of the edge.

5200: And a \emph{state} of the graph is a choice of a state for every edge.

5201: Summing over all states of $\Gamma$ and

5202: passing to the wedge product

5203: gives an $\fsp$ invariant in $\Lambda^*(QA_n)$.

5204: \[

5205: \begin{minipage}{4 in}

5206: \begin{center}

5207: \psfrag{p1}{\Huge $p_1$}

5208: \psfrag{p2}{\Huge $p_2$}

5209: \psfrag{p3}{\Huge $p_3$}

5210: \psfrag{q1}{\Huge $q_1$}

5211: \psfrag{q2}{\Huge $q_2$}

5212: \psfrag{q3}{\Huge $q_3$}

5213: \resizebox{1.1 in}{.7 in}{\includegraphics{state.eps}}

5214: \end{center}

5215: \end{minipage}

5216: \]

5217: 

5218: \noindent

5219: An example of a state is shown above.

5220: It gives us the term

5221: \[

5222: \begin{minipage}{\linewidth}

5223: \begin{center}

5224: \begin{minipage}{1 in}

5225: \begin{center}

5226: \psfrag{p1}{\Huge $q_1$}

5227: \psfrag{p2}{\Huge $p_2$}

5228: \psfrag{q1}{\Huge $q_1$}

5229: \psfrag{q2}{\Huge $p_2$}

5230: \psfrag{1}{}

5231: \psfrag{2}{}

5232: \psfrag{uu}{}

5233: \resizebox{.7 in}{.15 in}{\includegraphics{sf.eps}}

5234: \end{center}

5235: \end{minipage}

5236: $\bigwedge$

5237: \begin{minipage}{1 in}

5238: \begin{center}

5239: \psfrag{p1}{\Huge $p_1$}

5240: \psfrag{p2}{\Huge $p_2$}

5241: \psfrag{q1}{\Huge $q_1$}

5242: \psfrag{q2}{\Huge $q_3$}

5243: \psfrag{1}{}

5244: \psfrag{2}{}

5245: \psfrag{uu}{}

5246: \resizebox{.7 in}{.15 in}{\includegraphics{sf.eps}}

5247: \end{center}

5248: \end{minipage}

5249: $\bigwedge$

5250: \begin{minipage}{1 in}

5251: \begin{center}

5252: \psfrag{p1}{\Huge $p_3$}

5253: \psfrag{p2}{\Huge $q_2$}

5254: \psfrag{q1}{\Huge $q_2$}

5255: \resizebox{.8 in}{.6 in}{\includegraphics{qas.eps}}

5256: \end{center}

5257: \end{minipage}

5258: $\bigwedge$

5259: \begin{minipage}{1 in}

5260: \begin{center}

5261: \psfrag{p1}{\Huge $p_1$}

5262: \psfrag{p2}{\Huge $p_2$}

5263: \psfrag{q1}{\Huge $q_1$}

5264: \resizebox{.8 in}{.6 in}{\includegraphics{qas.eps}}

5265: \end{center}

5266: \end{minipage}\\

5267: \end{center}

5268: \end{minipage}

5269: \]

5270: in $\Lambda^4(QA_n)$.

5271: Since there are an even number of negative signs,

5272: the net sign is positive.

5273: To get the invariant, we sum over all states.

5274: 

5275: \subsubsection{Comparing the matings}  

5276: 

5277: Now we show that

5278: $\partial^{\prime} = 2n \partial_E + \partial_H$.

5279: Let $(\Gamma,\sigma) \in \mathcal G_k$. 

5280: And let $I$ be the invariant in $(\mathcal C_k)^{\fsp}$

5281: corresponding to $\Gamma$

5282: obtained by the procedure described above.

5283: 

5284: In order to compute $\partial(I)$,

5285: we apply the formula for $\partial$ given by equation \eqref{e:lb}

5286: to each state of $I$.

5287: So for each state,

5288: we must compute the Lie bracket of the data

5289: at each pair of vertices.

5290: However, recall from (\ref{ss:mate})

5291: that the Lie bracket is simply a sum of matings.

5292: Hence $\partial(I)$ can be computed as a sum

5293: over all states of $I$,

5294: where for each state,

5295: we mate a $p_i$ on a half-edge at a vertex

5296: with a $q_i$ on another half-edge at a different vertex

5297: in all possible ways.

5298: Another way to say this is that

5299: a mating occurs along a quasi-edge

5300: (which is a pair of distinct half-edges)

5301: that is not a quasi-loop (\ref{ss:finb}).

5302: 

5303: Now we rewrite $\partial(I)$ as a sum indexed by quasi-edges $e$

5304: of $\Gamma$.

5305: The summand for a quasi-edge $e$ 

5306: is the sum of those terms for which

5307: the mating occurs along $e$.

5308: It is obtained in two steps.

5309: 

5310: 1. Fix a state for $e$. Then sum over all possible states 

5311: of the other edges.

5312: 

5313: 2. Do this for each of the $2n$ states of $e$.

5314: 

5315: \noindent

5316: We explain a little of how the orientation will work out.

5317: We fix a representative $\sigma$ for $\Gamma$.

5318: Let $v_1$ and $v_2$ with labels $s < t$ 

5319: be the vertices of the quasi-edge $e$ 

5320: that we want to contract.

5321: In order to get a representative for $\sigma /e$,

5322: we first reorder the vertices so that

5323: $v_1$ and $v_2$ have labels $1$ and $2$ respectively.

5324: In doing so, we pick up a factor of $(-1)^{s+t-1}$. 

5325: This is precisely the factor that appears in formula \eqref{e:lb}

5326: for $\partial(I)$.

5327: 

5328: Now for a quasi-edge $e$,

5329: we perform the two steps above.

5330: The analysis splits into two cases.

5331: In the first case, assume that $e$ is an actual edge,

5332: i.e. $e \in E(\Gamma)$.

5333: Then the first step itself gives an invariant that comes from 

5334: $(\Gamma /e,\sigma /e).$

5335: Hence, after the second step, 

5336: the net contribution is $2n (\Gamma /e,\sigma /e).$

5337: 

5338: In the second case, assume that $e$ is a quasi-edge

5339: that is not an edge,

5340: i.e. $e \in E_q(\Gamma) \setminus E(\Gamma)$. 

5341: So the first step does not give an invariant.

5342: This is because, if the state of $e$ is fixed

5343: then the state of its partner, the quasi-edge $f$,

5344: is also fixed. 

5345: This is shown in the figure below.

5346: \[

5347: \begin{minipage}{1.6 in}

5348: \begin{center}

5349: \psfrag{v1}{\Huge $v_1$}

5350: \psfrag{v2}{\Huge $v_2$}

5351: \psfrag{e1}{\Huge $$}

5352: \psfrag{e2}{\Huge $$}

5353: \psfrag{e}{\Huge $e$}

5354: \psfrag{f}{\Huge $f$}

5355: \psfrag{pi}{\Huge $p_i$}

5356: \psfrag{qi}{\Huge $q_i$}

5357: \resizebox{1.4 in}{.9 in}{\includegraphics{bqi.eps}}

5358: \end{center}

5359: \end{minipage}

5360: \]

5361: However, after the second step, 

5362: we do get the invariant $(\Gamma /e,\sigma /e).$

5363: 

5364: Hence putting the two cases together, we obtain

5365: $$\partial^{\prime}_{} (\Gamma,\sigma) = 2n \sum_{e \in E(\Gamma)} (\Gamma /e,

5366: \sigma /e) + \sum_{e \in E_q(\Gamma) \setminus E(\Gamma)} 

5367: (\Gamma /e,\sigma /e).$$ 

5368: The right hand side is precisely

5369: $2n \partial_E + \partial_H$.

5370: This shows the commutativity of diagram~\eqref{e:comp} and

5371: completes the proof of Theorem~\ref{t:fin}.

5372: 

5373: 

5374: \section{Proof of the main theorem-Part II}  \label{s:sta}

5375: 

5376: In the previous section, the Lie algebra homology

5377: $H_*(QA_n)$ was related to graph homology.

5378: In this section, we show that the homology 

5379: of the Lie algebra $QA_n$, as $n$ varies, is ``stable''.

5380: Recall that 

5381: $QA_\infty = \underset{n \to \infty}{\lim} QA_n$,

5382: see (\ref{ss:ha}).

5383: We will prove the following theorem.

5384: 

5385: \begin{theorem}  \label{t:sta}

5386: $H_*(QA_{\infty}) = H_*(\mathcal G,\partial_E).$

5387: \end{theorem}

5388: %

5389: \noindent

5390: As in the previous section,

5391: we apply the theorem to the unit species $uu$

5392: and obtain the following corollary.

5393: 

5394: \begin{corollary}

5395: $H_*({{{\mathfrak{sp}}(2 \infty)}}) = H_*(\mathcal F,\partial_E).$

5396: \end{corollary}

5397: 

5398: \noindent

5399: Let us try to prove this theorem. We have

5400: $$

5401: \begin{array}{r c l}

5402: H_k(QA_{\infty})

5403:  & = & \underset{n \to \infty}{\lim} H_k(QA_n). \\

5404:  & = & \underset{n \to \infty}{\lim} H_k(\mathcal G,\partial_n). \\

5405:  & = & \underset{r}{\bigoplus} \underset{n \to \infty}{\lim} H_k(\mathcal G^{(r)},\partial_n). 

5406: \end{array}

5407: $$

5408: %

5409: The first equality says that homology commutes with direct limits.

5410: And the second equality is the content of Theorem~\ref{t:fin}.

5411: The bonding maps

5412: $H_k(\mathcal G,\partial_n) \to H_k(\mathcal G,\partial_{n+1})$

5413: for the direct limit are defined using the isomorphism

5414: $H_k(QA_n) \iso H_k(\mathcal G,\partial_n)$.

5415: For the third equality, 

5416: see the discussion in (\ref{ss:is}).

5417: There is a subtle point here.

5418: We need to know that the bonding maps restrict to

5419: $H_k(\mathcal G^{(r)},\partial_n) \to H_k(\mathcal G^{(r)},\partial_{n+1})$.

5420: Even assuming this, we are stuck.

5421: To complete the argument, one needs the following proposition.

5422: 

5423: \begin{proposition}  \label{p:sta}

5424: $\underset{n \to \infty}{\lim} H_k(\mathcal G^{(r)},\partial_n) = 

5425: H_k(\mathcal G^{(r)},\partial_E)$.

5426: \end{proposition}

5427: \noindent

5428: It will be proved using the ideas of Section \ref{s:mgh}. 

5429: The proof will be completed by the end of this section.

5430: The subtle point about the definition of the LHS

5431: raised above will be dealt in the course of the proof.

5432: 

5433: 

5434: \subsection{A pairing on $\Lambda^*(QA_n)$}  \label{ss:pl}

5435: 

5436: We have fixed a basis $p_1, \ldots,p_n,q_1, \ldots,q_n$

5437: for $V$.

5438: This gives us a basis for the Lie algebra $QA_n$

5439: and its exterior algebra $\Lambda^*(QA_n)$.

5440: To be explicit, a basis for $QA_n$ is given by distinct monomials.

5441: And for $\Lambda^*(QA_n)$ is given by taking wedge products

5442: of such monomials.

5443: \[

5444: \begin{minipage}{\linewidth}

5445: \begin{center}

5446: \begin{minipage}{1.2 in}

5447: \begin{center}

5448: \psfrag{x1}{\Huge $p_1$}

5449: \psfrag{x2}{\Huge $p_2$}

5450: \psfrag{x4}{\Huge $p_4$}

5451: \resizebox{.8 in}{.6 in}{\includegraphics{qa.eps}}

5452: \end{center}

5453: \end{minipage}

5454: $\bigwedge$

5455: \begin{minipage}{1.2 in}

5456: \begin{center}

5457: \psfrag{p1}{\Huge $p_1$}

5458: \psfrag{p2}{\Huge $p_2$}

5459: \psfrag{q1}{\Huge $q_1$}

5460: \psfrag{q2}{\Huge $q_2$}

5461: \psfrag{1}{}

5462: \psfrag{2}{}

5463: \psfrag{uu}{}

5464: \resizebox{.7 in}{.15 in}{\includegraphics{sf.eps}}

5465: \end{center}

5466: \end{minipage}

5467: $\bigwedge$

5468: \begin{minipage}{1.2 in}

5469: \begin{center}

5470: \psfrag{p1}{\Huge $p_1$}

5471: \psfrag{p2}{\Huge $p_2$}

5472: \psfrag{q1}{\Huge $q_1$}

5473: \psfrag{q2}{\Huge $q_2$}

5474: \resizebox{.8 in}{.6 in}{\includegraphics{qas.eps}}

5475: \end{center}

5476: \end{minipage}\\

5477: \vspace{.1 in} A basis element of $\Lambda^3(QA_4)$.

5478: \vspace{.1 in}

5479: \end{center}

5480: \end{minipage}

5481: \]

5482: Define a pairing

5483: $M^{\prime}(n) : \Lambda^*(QA_n)\otimes \Lambda^*(QA_n)\to \Q$ by

5484: $M^{\prime}(n)(\psi_i,\psi_j) = \delta_{ij}$,

5485: where $\{\psi_i\}$ is the basis of the exterior algebra

5486: that was chosen above.

5487: The subspaces $\Lambda^k(QA_n)$, as $k$ varies,

5488: are clearly orthogonal with respect to 

5489: $M^{\prime}(n)$.

5490: 

5491: To be more explicit, let

5492: $\varphi_1 = F_1 \wedge \ldots \wedge F_k$ and

5493: $\varphi_2 = H_1 \wedge \ldots \wedge H_k$ be wedges of monomials.

5494: Then

5495: $M^{\prime}(n)(\varphi_1,\varphi_2)$ is nonzero 

5496: only if $\varphi_1=\varphi_2$ or $\varphi_1=-\varphi_2$.

5497: This happens if

5498: $F_i = H_{\pi(i)}$ for a permutation $\pi$,

5499: i.e. there is a matching of the $F_i$'s with the $H_j$'s.

5500: This way of saying it brings us closer to the pairing $M(n)$,

5501: that was defined on graphs (\ref{ss:pg}). We say this more precisely.

5502: 

5503: The isomorphism

5504: $\mathcal G_k \isoto (\Lambda^k(QA_n))^{\fsp}$,

5505: explained in (\ref{sss:gi}),

5506: gives us an inclusion map 

5507: $\mathcal G \into \Lambda^*(QA_n)$.

5508: Under this inclusion,

5509: the pairing $M^{\prime}(n)$ has two important restrictions, namely,

5510: $M^{\prime}(n) : \Lambda^*(QA_n)\otimes \mathcal G \to \Q$ and 

5511: $M^{\prime}(n) : \mathcal G \otimes \mathcal G \to \Q$.

5512: We will see in Proposition \ref{p:pr}

5513: that the second pairing is just the pairing

5514: $M(n)$,

5515: that was defined in (\ref{ss:pg}).

5516: 

5517: \subsection{Connection between the pairings $M^{\prime}(n)$ and $M(n)$} 

5518: \label{ss:pr} 

5519: 

5520: 

5521: We will now see how formula \eqref{e:pg} for the pairing $M(n)$

5522: emerges naturally by analysing $M^{\prime}(n)$.

5523: 

5524: \begin{proposition}  \label{p:pr}

5525: The pairing $M^{\prime}(n) : \Lambda^*(QA_n)\otimes \Lambda^*(QA_n)\to \Q$ restricts to the pairing

5526: $M(n) : \mathcal G \otimes \mathcal G \to \Q$ 

5527: under the inclusion map $\mathcal G \into \Lambda^*(QA_n)$.

5528: \end{proposition}

5529: \begin{proof}

5530: Let $(\Gamma_1,\sigma_1),(\Gamma_2,\sigma_2) \in \mathcal G$.

5531: We will suppress the inclusion map and

5532: orientations from the notation.

5533: We want to show

5534: $M^{\prime}(n)(\Gamma_1,\Gamma_2) = M(n)(\Gamma_1,\Gamma_2)$.

5535: We compute the LHS by summing over all pairs of states of

5536: $\Gamma_1$ and $\Gamma_2$.

5537: Recall that a state of a graph is a choice

5538: of $p_i \lhor q_i$ for every edge (\ref{sss:gi}).

5539: We write, 

5540: \[

5541: M^{\prime}(n)(\Gamma_1,\Gamma_2) = \sum_{S_1,S_2} M^{\prime}(n)(S_1,S_2),

5542: \]

5543: where $S_i$ is a state of $\Gamma_i$.

5544: A term in this sum is nonzero only if

5545: one can match $S_1$ and $S_2$ in the following sense.

5546: 

5547: There is a matching $m:\Gamma_1 \to \Gamma_2$ 

5548: (Definition \ref{d:gm})

5549: with the following additional piece of data.

5550: For each component in the matching

5551: (Definition \ref{d:csm}), 

5552: there is an index $i$ such that

5553: the vertices of the polygon are alternately

5554: labelled $p_i$ and $q_i$.

5555: \[

5556: \begin{minipage}{.7 in}

5557: \begin{center}

5558: \psfrag{a}{\Huge $q_i$}

5559: \psfrag{b}{\Huge $p_i$}

5560: \psfrag{c}{\Huge $q_i$}

5561: \psfrag{d}{\Huge $p_i$}

5562: \psfrag{e}{\Huge $q_i$}

5563: \psfrag{f}{\Huge $p_i$}

5564: \resizebox{.7 in}{.7 in}{\includegraphics{ch.eps}}

5565: \end{center}

5566: \end{minipage}

5567: \hspace{1 in}

5568: \begin{minipage}{.6 in}

5569: \begin{center}

5570: \psfrag{a}{\Huge $q_j$}

5571: \psfrag{b}{\Huge $p_j$}

5572: \psfrag{c}{\Huge $q_j$}

5573: \psfrag{d}{\Huge $p_j$}

5574: \resizebox{.5 in}{.5 in}{\includegraphics{cs.eps}}

5575: \end{center}

5576: \end{minipage}

5577: \]

5578: Hence we can group terms and sum over all

5579: matchings $m:\Gamma_1 \to \Gamma_2$.

5580: For each matching $m$, there are $c(m)$ number of polygons.

5581: And for each polygon, there are $2n$ choices of the index $i$.

5582: This gives a factor of $(2n)^{c(m)}$ 

5583: with a sign which can be checked to be 

5584: $\sign(m)$. 

5585: Hence

5586: $M^{\prime}(n)(\Gamma_1,\Gamma_2) = \underset{m:\Gamma_1 \to \Gamma_2}{\sum}

5587: \sign(m) (2n)^{c(m)}$,

5588: which by definition is

5589: $M(n)(\Gamma_1,\Gamma_2)$.

5590: This completes the proof.

5591: 

5592: \end{proof}

5593: 

5594: \subsection{A commutative diagram}  \label{ss:cd}

5595: 

5596: Consider the restriction of the pairing

5597: $M^{\prime}(n) : \Lambda^k(QA_n) \otimes \mathcal G_k^{(r)} \to \Q$.

5598: This then defines a map

5599: $\Lambda^k(QA_n) \to (\mathcal G_k^{(r)})^*$,

5600: which we again denote by $M^{\prime}(n)$.

5601: From Proposition \ref{p:pr},

5602: it is clear that this map is an extension of the map

5603: $\mathcal G_k^{(r)} \to (\mathcal G_k^{(r)})^*$

5604: defined using the pairing $M(n)$.

5605: We now claim that the following diagram commutes.

5606: \begin{equation*}

5607: \begin{CD}

5608: \mathcal G_k^{(r)} @>>> \Lambda^k(QA_n)@>M^{\prime}(n)>> (\mathcal G_k^{(r)})^* @<A<\iso< \mathcal G_k^{(r)} \\ 

5609: @V{2n \partial_E + \partial_H}VV @V{\partial}VV @V{\delta_E^*}VV @V{\partial_E}VV\\

5610: \mathcal G_{k-1}^{(r)} @>>> \Lambda^{k-1}(QA_n)@>>M^{\prime}(n)> (\mathcal G_{k-1}^{(r)})^* @<\iso<A< \mathcal G_{k-1}^{(r)}.\\ 

5611: \end{CD}

5612: \end{equation*}

5613: %

5614: This is a fattening of the diagram \eqref{e:cdg} 

5615: at the end of Section \ref{s:mgh},

5616: with the Chevalley-Eilenberg complex added in.

5617: 

5618: The commutativity of the first square follows from 

5619: diagram \eqref{e:comp} in (\ref{ss:cb}).

5620: Saying that the second square commutes 

5621: is equivalent to saying that

5622: the boundary map $\partial:\Lambda^*(QA_n)\to \Lambda^*(QA_n)$

5623: and the coboundary map $\delta_E:\mathcal G \to \mathcal G$ are adjoints

5624: with respect to the pairing $M^{\prime}(n)$.

5625: This follows by generalising the proof of Proposition \ref{p:ad}.

5626: This proves the claim.

5627: 

5628: \subsection{Invariance of the pairing}  

5629: 

5630: The pairing 

5631: $M^{\prime}(n) : \Lambda^*(QA_n)\otimes \Lambda^*(QA_n)\to \Q$

5632: is $\fsp$ invariant 

5633: in a precise sense.

5634: It is more convenient to express it using 

5635: the Lie algebra $uuA_n$, see (\ref{ss:sl}), 

5636: which is anti-isomorphic to $\fsp$.

5637: 

5638: \begin{proposition}  \label{p:in}

5639: We have

5640: $M^{\prime}(n)(\varphi_1,\varphi_2 \cdot H) + 

5641: M^{\prime}(n)(\varphi_1 \cdot JH, \varphi_2) = 0$,

5642: where $H \in uuA_n$ and $J$ 

5643: is the element of the symplectic group defined by

5644: $J(p_i)=q_i$ and $J(q_i)= - p_i$.

5645: \end{proposition}

5646: 

5647: \begin{proof}

5648: We may assume that $\varphi_1,\varphi_2$ 

5649: are wedges of monomials, 

5650: i.e. they are two basis elements (upto sign) of

5651: $\Lambda^k(QA_n)$.

5652: We may also assume that $H$ is a monomial in $uuA_n$.

5653: Say, for definiteness, that $H=q_1 q_2$ and so

5654: $JH = p_1 p_2$.

5655: Note that $\varphi_2 \cdot H$ involves replacing an occurrence of

5656: $p_2$ (resp. $p_1$) in $\varphi_2$ by $q_1$ (resp. $q_2$),

5657: see (\ref{ss:sl}).

5658: And $\varphi_1 \cdot JH$ has exactly the opposite effect.

5659: It involves replacing an occurrence of

5660: $q_1$ (resp. $q_2$) in $\varphi_1$ by

5661: $p_2$ (resp. $p_1$) and picking a minus sign.

5662: It is now fairly clear how terms 

5663: from the two pairing expressions would cancel.

5664: \end{proof}

5665: 

5666: \begin{corollary}

5667: The subspaces $(\Lambda^k(QA_n))^{\fsp}$ and

5668: $\fsp \cdot (\Lambda^k(QA_n))$ are orthogonal 

5669: with respect to the pairing $M^{\prime}(n)$.

5670: \end{corollary}

5671: 

5672: \subsection{Stability of the pairings}  

5673: 

5674: Consider the Lie algebra $QA_{n+1}$.

5675: The underlying vector space $V_{n+1}$ has dimension $2(n+1)$.

5676: The basis of $V_{n+1}$ is an extension of the one on $V$ to

5677: $p_1, \ldots,p_{n+1},q_1, \ldots,q_{n+1}$.

5678: Hence the basis for $QA_n$ and $\Lambda^*(QA_n)$

5679: includes into the basis for $QA_{n+1}$ and $\Lambda^*(QA_{n+1})$

5680: respectively.

5681: 

5682: We now show that the restricted pairings 

5683: $M^{\prime}(n) : \Lambda^*(QA_n)\otimes \mathcal G \to \Q$ 

5684: are stable, that is,

5685: for $\varphi \in \Lambda^*(QA_n)$ and $\Gamma \in \mathcal G$,

5686: \begin{equation}  \label{e:ps}

5687: M^{\prime}(n)(\varphi,\Gamma) = M^{\prime}(n+1)(\varphi,\Gamma). 

5688: \end{equation}

5689: The $\varphi$ in the RHS is obtained by using the inclusion of

5690: $\Lambda^*(QA_n)$ in $\Lambda^*(QA_{n+1})$.

5691: 

5692: Express both sides as sums over the states of $\Gamma$,

5693: see (\ref{sss:gi}).

5694: Though the graph $\Gamma$ is the same on both sides,

5695: it has more states for ``$n+1$'' than ``$n$''

5696: because there are more variables available.

5697: The extra states of $\Gamma$ in the RHS are the ones that

5698: involve at least one $p_{n+1}$ or $q_{n+1}$.

5699: However, they do not contribute anything to the pairing

5700: $M^{\prime}(n+1)$, since by assumption, 

5701: $\varphi$ does not involve either

5702: $p_{n+1}$ or $q_{n+1}$. 

5703: This proves that equation \eqref{e:ps} holds.

5704: 

5705: 

5706: \subsection{The stability commutative diagram}  

5707: 

5708: The chain group 

5709: $\mathcal C_{k} = \Lambda^k(QA_n)$ can be written as

5710: $\mathcal C_{k} = (\mathcal C_{k})^{\fsp} \oplus \fsp \cdot \mathcal C_k$,

5711: see (\ref{ss:in}).

5712: It depends on $n$ 

5713: but we will suppress that in our notation.

5714: The stability of the pairings $M^{\prime}(n)$ (equation \eqref{e:ps})

5715: says that the following diagram commutes.

5716: \begin{equation*}

5717: \begin{CD}

5718: (\mathcal C_{k})^{\fsp} \oplus \fsp \cdot \mathcal C_k @>M^{\prime}(n)>> (\mathcal G_k^{(r)})^* @<A<\iso< \mathcal G_k^{(r)} \\ 

5719: @VVV @V{\id}VV @V{\id}VV\\

5720: (\mathcal C_{k})^{{\mathfrak{sp}}(2n+2)} \oplus {{\mathfrak{sp}}(2n+2)} \cdot \mathcal C_k 

5721: @>>M^{\prime}(n+1)> (\mathcal G_{k}^{(r)})^* @<\iso<A< \mathcal G_{k}^{(r)}.\\ 

5722: \end{CD}

5723: \end{equation*}

5724: The first vertical map does not restrict to a map

5725: on the invariants.

5726: However, by projecting on the first factor,

5727: we obtain maps

5728: $(\mathcal C_{k})^{\fsp} \to (\mathcal C_{k})^{{{\mathfrak{sp}}(2n+2)}}$,

5729: one for each $n$.

5730: And these are the maps that one can use 

5731: for computation of stable homology

5732: $\underset{n \to \infty}{\lim} H_k(QA_n)$.

5733: This gives us the diagram

5734: \begin{equation*}

5735: \begin{CD}

5736: \mathcal G_k \iso (\mathcal C_{k})^{\fsp} @>M^{\prime}(n)>> (\mathcal G_k^{(r)})^* @<A<\iso< \mathcal G_k^{(r)} \\ 

5737: @VVV @V{\id}VV @V{\id}VV\\

5738: \mathcal G_k \iso (\mathcal C_{k})^{{\mathfrak{sp}}(2n+2)}

5739: @>>M^{\prime}(n+1)> (\mathcal G_{k}^{(r)})^* @<\iso<A< \mathcal G_{k}^{(r)}.\\ 

5740: \end{CD}

5741: \end{equation*}

5742: By the Corollary to Proposition \ref{p:in},

5743: the map

5744: $\fsp \cdot \mathcal C_k \to (\mathcal G_k)^*$ is zero.

5745: Hence the above diagram still commutes.

5746: Furthermore, if one thinks of the map

5747: $(\mathcal C_{k})^{\fsp} \to (\mathcal C_{k})^{{{\mathfrak{sp}}(2n+2)}}$

5748: in terms of graphs,

5749: then it is clear that it preserves the Euler characteristic.

5750: This gives a map

5751: $\mathcal G_k^{(r)} \to \mathcal G_k^{(r)}$ for each $n$.

5752: The direct limit

5753: $\underset{n \to \infty}{\lim} H_k(\mathcal G^{(r)},\partial_n)$

5754: is taken with respect to these maps.

5755: Consider the stability commutative diagram

5756: \begin{equation*}

5757: \begin{CD}

5758: \mathcal G_k^{(r)} @>>> (\mathcal C_{k})^{\fsp} @>M^{\prime}(n)>> (\mathcal G_k^{(r)})^* @<A<\iso< \mathcal G_k^{(r)} \\ 

5759: @VVV @VVV @V{\id}VV @V{\id}VV\\

5760: \mathcal G_{k}^{(r)} @>>> (\mathcal C_{k})^{\fsp+2} @>>M^{\prime}(n)> (\mathcal G_{k}^{(r)})^* @<\iso<A< \mathcal G_{k}^{(r)}.\\ 

5761: \end{CD}

5762: \end{equation*}

5763: %

5764: We know that the map

5765: $M(n): \mathcal G_k^{(r)} \to (\mathcal G_k^{(r)})^*$

5766: is an isomorphism for all $n$ large enough (\ref{ss:pnd}).

5767: In the above diagram,

5768: it is the composite of the first two horizontal maps.

5769: Proposition \ref{p:sta} now follows

5770: by the isomorphism between the leftmost and rightmost columns.

5771: 

5772: \begin{remark}

5773: In \cite{\god}, Kontsevich conjectured that

5774: the stable homology groups

5775: $H_k(QA_\infty)$ are finite dimensional.

5776: In terms of graphs, this says that

5777: $H_k(\mathcal G^{(r)},\partial_E) = 0$ 

5778: for fixed $k$ and sufficiently large $r$.

5779: \end{remark}

5780: 

5781: \section{Proof of the main theorem-Part III}  \label{s:con}

5782: 

5783: The limit Lie algebra,

5784: $QA_\infty = \underset{n \to \infty}{\lim} QA_n$,

5785: has the structure of a Hopf algebra on its homology

5786: (\ref{ss:ha}).

5787: Let

5788: $PH_*(QA_{\infty})$

5789: denote the subspace of primitive elements of

5790: $H_*(QA_{\infty})$.

5791: Also let $(\mathcal C,\partial_E)$ be the subcomplex of

5792: $(\mathcal G,\partial_E)$ spanned by oriented connected graphs.

5793: We will prove the following theorem.

5794: 

5795: \begin{theorem}  \label{t:con}

5796: $PH_*(QA_{\infty}) = H_*(\mathcal C,\partial_E).$

5797: \end{theorem}

5798: \begin{proof}

5799: The Hopf algebra $H_*(QA_{\infty})$ is commutative

5800: and cocommutative in the graded sense.

5801: Hence by the structure theorem of Milnor and Moore

5802: \cite{\mm},

5803: $H_*(QA_{\infty})$ is a graded polynomial algebra

5804: generated by the primitive elements.

5805: By Theorem \ref{t:sta},

5806: we know $H_*(QA_{\infty}) = H_*(\mathcal G,\partial_E).$

5807: This along with the identification given by Proposition \ref{p:du}

5808: (which we are going to prove)

5809: shows that the subspace of primitive elements 

5810: is precisely $H_*(\mathcal C,\partial_E).$

5811: \end{proof}

5812: %

5813: \noindent

5814: As always,

5815: we apply the above theorem to the unit species $uu$

5816: and obtain the following corollary.

5817: 

5818: \begin{corollary}

5819: $PH_*({{{\mathfrak{sp}}(2 \infty)}}) = H_*(\mathcal P,\partial_E).$

5820: \end{corollary}

5821: 

5822: \begin{proposition}  \label{p:du}

5823: The product in the stable homology

5824: $H_*(QA_{\infty})$ can be defined on the chain complex

5825: $(\mathcal G,\partial_E)$ as disjoint union of graphs.

5826: \end{proposition}

5827: \begin{proof}

5828: We make use of the deformation map

5829: $D(n) : \Lambda^k(QA_n) \to \mathcal G_k^{(r)}$.

5830: This is a composition of the maps

5831: $\Lambda^k(QA_n) \to (\mathcal G_k^{(r)})^* \to \mathcal G_k^{(r)}$

5832: that we had in the previous section (\ref{ss:cd}).

5833: It is an extension of the deformation map

5834: $D(n) : \mathcal G_k^{(r)} \to \mathcal G_k^{(r)}$ 

5835: defined at the end of (\ref{ss:dmg}).

5836: For simplicity,

5837: we will suppress the dependence on $n$

5838: and use the notation

5839: $D(\varphi) = \sum_{\Gamma} D(\varphi,\Gamma) \ \Gamma$

5840: for $\varphi \in \Lambda^k(QA_{n})$ and

5841: $\Gamma \in \mathcal G_k^{(r)}$.

5842: The usage of $D$ for $D(n)$ 

5843: is local to this proposition.

5844: Later in Appendix~\ref{ss:dg},

5845: the letter $D$ will be used for

5846: something slightly different.

5847: 

5848: The map $QA_n \oplus QA_m \into QA_{n+m}$ induces a map

5849: $\Lambda^i(QA_{n}) \otimes \Lambda^{k-i}(QA_{m}) \to

5850: \Lambda^k(QA_{n+m})$.

5851: More explicitly,

5852: $\varphi_1 \otimes \varphi_2 \mapsto \varphi_1 \wedge \varphi_2$, 

5853: with the indices of $\varphi_2$ shifted up by $n$.

5854: In order to complete the proof,

5855: it is enough to show that the following diagram commutes.

5856: 

5857: \begin{equation*}

5858: \begin{CD}

5859: \Lambda^i(QA_{n}) \otimes \Lambda^{k-i}(QA_{m}) @>D \otimes D>> \mathcal G_i \otimes \mathcal G_{k-i} \\

5860:  @VVV @V \mu VV \\

5861: \Lambda^k(QA_{n+m}) @>D>> \mathcal G_k. \\

5862: \end{CD}

5863: \end{equation*}

5864: Here the map $\mu$ stands for disjoint union of graphs.

5865: The commutativity is equivalent to showing the identity

5866: $$D(\varphi_1 \wedge \varphi_2,\Gamma) = \underset{\Gamma_1 \sqcup \Gamma_2 = \Gamma}{\sum}

5867: D(\varphi_1,\Gamma_1) D(\varphi_2,\Gamma_2).$$

5868: To calculate the LHS,

5869: we match the $i$ pieces of $\varphi_1$ 

5870: and $k-i$ pieces of $\varphi_2$ 

5871: with the $k$ vertices of $\Gamma$ and then contract indices.

5872: And we sum over all matchings.

5873: 

5874: The key observation is that $\varphi_1$ and $\varphi_2$

5875: must land on disconnected parts of $\Gamma$,

5876: say $\Gamma_1$ and $\Gamma_2$,

5877: in order to get a non-zero contraction.

5878: This is because the indices that occur in $\varphi_1$

5879: are disjoint from those in $\varphi_2$.

5880: The rest of the argument is now fairly clear and we omit it.

5881: \end{proof}

5882: 

5883: \begin{remark}

5884: In fact, the chain complex $(\mathcal G,\partial_E)$ 

5885: is a differential graded Hopf algebra

5886: with product $\mu$ given by disjoint union and

5887: the coproduct $\Delta$ defined for a connected graph $\Gamma$

5888: by $\Delta (\Gamma) = 1 \otimes \Gamma + \Gamma \otimes 1$

5889: and extended to $\mathcal G$ as a morphism of algebras.

5890: Here 1 stands for the unit in $\mathcal G$ and

5891: may be thought of as the empty graph.

5892: The boundary map $\partial_E$ is a

5893: derivation with respect to the product $\mu$

5894: and a coderivation with respect to the coproduct $\Delta$.

5895: And this induces the Hopf algebra structure on

5896: $H_*(\mathcal G,\partial_E)$. 

5897: \end{remark}

5898: 

5899: \subsection*{Proof of the main theorem concluded}  

5900: 

5901: Our main theorem (Theorem \ref{t:mt}) now follows by putting together

5902: Theorem \ref{t:con} and its corollary

5903: and the corollary to Proposition \ref{p:ss}.

5904: 

5905: \appendix

5906: 

5907: \section{Deformation quantisation} \label{ss:dq}

5908: 

5909: For a good review of the problem of

5910: deformation quantisation, 

5911: see the notes by Voronov~\cite{\voronov}.

5912: The main object of interest is a Poisson algebra.

5913: The algebra of functions on a symplectic,

5914: or more generally, a Poisson manifold 

5915: form a Poisson algebra.

5916: This is the classical case.

5917: In (\ref {ss:tp}-\ref {ss:se}),

5918: we provide some background on the classical case.

5919: This would also be useful for some of the ideas 

5920: in Appendix~\ref{ss:dg}.

5921: The material in~\ref{ss:tp} is taken from~\cite{\voronov}.

5922: 

5923: In (\ref {ss:qoa}),

5924: we speculate on the form of this problem for operads.

5925: The main question is:

5926: What is a ``Poisson operad manifold''?

5927: Unlike the classical case,

5928: a symplectic operad manifold is not automatically

5929: a Poisson operad manifold.

5930: The relation between our viewpoint and 

5931: the standard deformation theory of operads

5932: considered by Balavoine in~\cite{\lsv}

5933: is not clear.

5934: 

5935: 

5936: \subsection{The problem}  \label{ss:tp}

5937: 

5938: Let $A$ be a commutative algebra over a field $k$

5939: of characteristic zero.

5940: A formal deformation of $A$ is a $k[[t]]$ bilinear product

5941: (which we denote $\star$) on the space $A[[t]]$

5942: of formal power series in a variable $t$ satisfying:

5943: $$F \star H = F \cdot H + \mu_1(F,H) t + \mu_2(F,H) t^2 +

5944: \ldots \ \ \text{for} \ \ F,H \in A,$$

5945: where $F \cdot H$ is the original product on $A$

5946: and the star product $\star$ is associative.

5947: The deformation is called \emph{trivial} if

5948: all the higher products $\mu_1,\mu_2,\ldots$ are zero.

5949: 

5950: Suppose that $\star$ is a deformation of $\cdot$,

5951: the original product.

5952: Then define $\{F,H\} = \frac{1}{2} (\mu_1(F,H) - \mu_1(H,F))$.

5953: One can check that $\{\ ,\ \}$ 

5954: is a Lie bracket on $A$ 

5955: and further that the triple $(A,\cdot,\{\ ,\ \})$ is a Poisson algebra.

5956: Recall that a Poisson algebra is a space

5957: with compatible commutative and Lie structures.

5958: The compatibility relation is 

5959: $\{F \cdot G,H\} = F \cdot \{G,H\} + G \cdot \{F,H\}$.

5960: 

5961: In physical terms, one regards

5962: the Poisson algebra $A$ as the quasi-classical limit 

5963: of the associative algebra $A[[t]]$, 

5964: and the algebra $A[[t]]$ as a deformation quantisation

5965: of the Poisson algebra $A$.

5966: The deformation quantisation problem is the inverse problem:

5967: given a Poisson algebra $(A,\cdot,\{\ ,\ \})$,

5968: find a formal deformation $\star$ returning the original

5969: Poisson algebra structure on $A$ in the quasi-classical limit.

5970:  

5971: \subsection{Gauge equivalence}  \label{ss:ge}

5972: 

5973: There is a natural gauge group acting on star products.

5974: This group consists of automorphisms $D$ of $A[[t]]$

5975: which are $k[[t]]$ linear.

5976: They have the form

5977: $D = D_0 + t D_1 + t^2 D_2 + \ldots$,

5978: where $D_i : A \to A$ are operators 

5979: with $D_0$ being the identity.

5980: This means that

5981: $$F \mapsto F + t D_1(F) + t^2 D_2(F) + \ldots \ \ \text{for}

5982: \qquad F \in A,$$

5983: and for a general element in $A[[t]]$,

5984: one uses $k[[t]]$ linearity.

5985: 

5986: Two star products $\star$ and $\star^{\prime}$

5987: are gauge equivalent if there is an automorphism $D$

5988: as above so that the following diagram commutes.

5989: \begin{equation*}

5990: \begin{CD}

5991: A[[t]] \otimes A[[t]] @>D \otimes D>> A[[t]] \otimes A[[t]] \\

5992:  @V \star VV @V \star^{\prime} VV \\

5993: A[[t]] @>D>> A[[t]]. \\

5994: \end{CD}

5995: \end{equation*}

5996:  

5997: 

5998: \subsection{The simplest example}  \label{ss:se}

5999: 

6000: Let $A$ be the space of all polynomial functions on $\R^{2n}$,

6001: i.e. polynomials in the $2n$ variables 

6002: $p_1, \ldots,p_{n},q_1, \ldots,q_{n}$.

6003: Then the triple

6004: $(A,\cdot,\{\ ,\ \})$ is a Poisson algebra,

6005: where $\cdot$ is the usual product in $A$

6006: and $\{\ ,\ \}$ is the Poisson bracket

6007: defined by equation~\eqref{e:pb} in (\ref{s:sog}).

6008: 

6009: In this case, the solution to the

6010: deformation quantisation problem is 

6011: given by the Moyal $\star$ product.

6012: In fact, we will have

6013: $F \star H = F \cdot H + \{F,H\} t + \ \text{higher terms}$;

6014: that is, $\mu_1(F,H) = \{F,H\}$.

6015: 

6016: \subsubsection{The Moyal product}  \label{ss:mp}

6017: 

6018: Let $B$ be any operator on $A \otimes A$.

6019: For $F,H \in A$, define

6020: $B(F,H)$ to be the element of $A$ obtained

6021: by applying the product in $A$ to 

6022: $B(F \otimes H)$.

6023: Namely, if

6024: $B = \sum B_{(1)} \otimes B_{(2)}$ then

6025: $B(F,H) = \sum B_{(1)} F \cdot B_{(2)} H$.

6026: 

6027: Now let 

6028: $B = \sum_{i=1}^n \frac{\partial}{\partial p_i} \otimes \frac{\partial}{\partial q_i} - 

6029: \frac{\partial}{\partial q_i} \otimes \frac{\partial}{\partial p_i}$

6030: be a bi-differential operator on $A \otimes A$.

6031: By our notation,

6032: $$B(F,H)=\sum_{i=1}^n \frac{\partial F}{\partial p_i} \frac{\partial H}{\partial q_i} - 

6033: \frac{\partial F}{\partial q_i} \frac{\partial H}{\partial p_i}.$$  

6034: In other words, $B(F,H) = \{F,H\}$ is just the Poisson bracket.

6035: Now define the Moyal $\star$ product by

6036: \begin{equation}  \label{e:mp}

6037: F \star H = e^{tB}(F,H) = \sum_{n \geq 0} t^n

6038: \frac{B^n}{n!}(F,H).

6039: \end{equation}

6040: Observe that

6041: $F \star H = F \cdot H + \{F,H\} t + \ \text{higher terms}$,

6042: as claimed earlier.

6043: To show that the Moyal product 

6044: solves the deformation quantisation problem,

6045: one must prove that it is associative.

6046: We will do this by interpreting equation~\eqref{e:mp} using pictures.

6047:   

6048: \subsubsection{Associativity of the Moyal product}  \label{ss:amp}

6049: 

6050: If $F \in A$ is a monomial, say

6051: $F = p_1^2 p_2 q_2$,

6052: then we represent it as 

6053: $F = 

6054: \fbox{

6055: \begin{minipage}{.4 in}

6056: \begin{center}

6057: \psfrag{a}{\Huge $p_1$}

6058: \psfrag{b}{\Huge $p_1$}

6059: \psfrag{c}{\Huge $p_2$}

6060: \psfrag{d}{\Huge $q_2$}

6061: \resizebox{.4 in}{.4 in}{\includegraphics{ccd.eps}}

6062: \end{center}

6063: \end{minipage} 

6064: }

6065: $.

6066: And if $F$ is a polynomial rather than a monomial

6067: then we represent it as a formal sum of pictures.

6068: It is clear that the product of two monomials in $A$ is defined

6069: as the disjoint union of pictures.

6070: 

6071: Our goal is to understand the operator $\frac{B^n}{n!}$

6072: via pictures.

6073: As a start, we understand the term

6074: $\frac{\partial F}{\partial p_i} \frac{\partial H}{\partial q_i}$.

6075: This is a sum of pictures of the form

6076: \begin{minipage}{1 in}

6077: \begin{center}

6078: \psfrag{pi}{\Huge $p_i$}

6079: \psfrag{qi}{\Huge $q_i$}

6080: \psfrag{F}{\Huge $F$}

6081: \psfrag{H}{\Huge $H$}

6082: \resizebox{1 in}{.4 in}{\includegraphics{cm.eps}}

6083: \end{center}

6084: \end{minipage} 

6085: ,

6086: where the edge goes from a point labelled $p_i$

6087: to a point labelled $q_i$.

6088: The other points in $F$ and $H$ are not shown in the picture.

6089: We will have the following interpretation.

6090: A picture of the form

6091: \begin{minipage}{1 in}

6092: \begin{center}

6093: \psfrag{pi}{\Huge $$}

6094: \psfrag{qi}{\Huge $$}

6095: \psfrag{F}{\Huge $$}

6096: \psfrag{H}{\Huge $$}

6097: \resizebox{1 in}{.4 in}{\includegraphics{cm.eps}}

6098: \end{center}

6099: \end{minipage} 

6100: means that we delete the edge and its endpoints

6101: and then take disjoint union of the two parts.

6102: Earlier in the paper,

6103: such an operation was called a mating.

6104: 

6105: Now we claim that

6106: \begin{equation}  \label{e:mpp}

6107: \frac{B^n}{n!}(F,H) = 

6108: \sum_P (-1)^{\sign(P)}

6109: \left(

6110: \begin{minipage}{1.2 in}

6111: \begin{center}

6112: \psfrag{F}{\Huge $F$}

6113: \psfrag{H}{\Huge $H$}

6114: \psfrag{n}{\Huge $n$}

6115: \resizebox{1 in}{.4 in}{\includegraphics{cmm.eps}}

6116: \end{center}

6117: \end{minipage} 

6118: \right)

6119: ,

6120: \end{equation}

6121: where $P$ is a picture as shown.

6122: There are $n$ edges between $F$ and $H$ 

6123: each connecting a $p_i$ in $F$ with a $q_i$ in $H$ or vice-versa.

6124: And $\sign(P)$ is the number of edges 

6125: that are out of order;

6126: i.e. they connect

6127: a $q_i$ in $F$ to a $p_i$ in $H$.  

6128: The interpretation for $P$ is that 

6129: we delete the $n$ edges and their endpoints

6130: and take disjoint union of what remains.

6131: 

6132: Note that the factor $n!$ is no longer necessary

6133: in the pictorial description.

6134: Each of the $n$ $B$'s in $B^n$ contributes

6135: to one of the $n$ edges in the picture.

6136: So there are $n!$ ways to obtain the same picture.

6137: After dividing by $n!$,

6138: we get every picture exactly once,

6139: giving us equation~\eqref{e:mpp}.

6140: 

6141: The associativity of the Moyal product can be seen from

6142: the formula

6143: \begin{equation}  \label{e:amp}

6144: F \star G \star H = 

6145: \sum_{\substack {n \geq 0 \\ n=k+l+m}} t^n

6146: \sum_P (-1)^{\sign(P)}

6147: \left(

6148: \begin{minipage}{1.2 in}

6149: \begin{center}

6150: \psfrag{F}{\Huge $F$}

6151: \psfrag{G}{\Huge $G$}

6152: \psfrag{H}{\Huge $H$}

6153: \psfrag{k}{\Huge $k$}

6154: \psfrag{l}{\Huge $l$}

6155: \psfrag{m}{\Huge $m$}

6156: \psfrag{n}{\Huge $n$}

6157: \resizebox{1 in}{.8 in}{\includegraphics{cmam.eps}}

6158: \end{center}

6159: \end{minipage} 

6160: \right)

6161: ,

6162: \end{equation}

6163: where the picture $P$ means that

6164: delete all edges and their endpoints and

6165: then take disjoint union of what remains.

6166: And  $\sign(P)$ is the number of edges that connect

6167: a $q_i$ to a $p_i$. 

6168: (The edges go from $F$ to $G$, $G$ to $H$ 

6169: and $F$ to $H$.)

6170: The letters $k$, $l$ and $m$ refer to the number of connecting edges.

6171: 

6172: \subsection{An example based on the surface species (\ref{ss:ss})}  \label{ss:ae}

6173: 

6174: Before going to the general case,

6175: let us do another example.

6176: Consider a picture of the form

6177: $ 

6178: \fbox{

6179: \begin{minipage}{.6 in}

6180: \begin{center}

6181: \psfrag{a}{\Huge $p_1$}

6182: \psfrag{b}{\Huge $p_2$}

6183: \resizebox{.6 in}{.4 in}{\includegraphics{ssa.eps}}

6184: \end{center}

6185: \end{minipage} 

6186: }

6187: $,

6188: namely, we have a compact orientable surface

6189: (not necessarily connected)

6190: whose boundary circles are labelled by

6191: $p_1,\ldots,p_n,q_1,\ldots,q_n$.

6192: Let $A$ be the vector space spanned by such pictures,

6193: each picture being a monomial.

6194: The space $A$ is a Poisson algebra 

6195: with the commutative product $\cdot$ given by disjoint union of pictures

6196: and the Lie bracket $\{\ ,\ \}$ given by

6197: \begin{equation}  \label{e:mps}

6198: \{F,H\} = 

6199: \sum_P (-1)^{\sign(P)}

6200: \left(

6201: \begin{minipage}{1 in}

6202: \begin{center}

6203: \psfrag{pi}{\Huge $$}

6204: \psfrag{qi}{\Huge $$}

6205: \psfrag{F}{\Huge $F$}

6206: \psfrag{H}{\Huge $H$}

6207: \resizebox{1 in}{.4 in}{\includegraphics{cm.eps}}

6208: \end{center}

6209: \end{minipage} 

6210: \right)

6211: ,

6212: \end{equation}

6213: where the edge connects a boundary circle 

6214: labelled $p_i$ in $F$ to a boundary circle 

6215: labelled $q_i$ in $H$ or vice-versa.

6216: The picture means that the two boundary circles

6217: joined by the edge are glued together

6218: to form a single surface.

6219: As an example,

6220: \[

6221: \left\{

6222: \begin{minipage}{.8 in}

6223: \begin{center}

6224: \psfrag{a}{\Huge $p_1$}

6225: \resizebox{.6 in}{.2 in}{\includegraphics{ssaf.eps}}

6226: \end{center}

6227: \end{minipage}

6228: \ \ , \ \

6229: \begin{minipage}{.8 in}

6230: \begin{center}

6231: \psfrag{a}{\Huge $q_1$}

6232: \psfrag{b}{\Huge $p_2$}

6233: \resizebox{.7 in}{.17 in}{\includegraphics{ssah.eps}}

6234: \end{center}

6235: \end{minipage}

6236: \right\}

6237: \ \ = \ \

6238: \begin{minipage}{.8 in}

6239: \begin{center}

6240: \psfrag{b}{\Huge $p_2$}

6241: \resizebox{.7 in}{.2 in}{\includegraphics{ssar.eps}}

6242: \end{center}

6243: \end{minipage}

6244: .

6245: \]

6246: The deformation quantisation for the Poisson algebra $A$

6247: is again given by equations~\eqref{e:mp} and~\eqref{e:mpp}.

6248: The picture in equation~\eqref{e:mpp} means that

6249: we glue together the boundary circles that are

6250: connected by edges.

6251: There will be a total of $n$ gluings.

6252: The associativity of the star product follows

6253: from equation~\eqref{e:amp}.

6254: 

6255: 

6256: \subsection{Quantization of operad algebras}  \label{ss:qoa}

6257: 

6258: We now initiate a general discussion of deformation quantization

6259: for operads.

6260: Precise definitions are not given;

6261: so this section should be read mainly for the philosophy.

6262: 

6263: As a candidate for the Poisson algebra $A$,

6264: take the free algebra over a mated species $Q$,

6265: which we have previously denoted $QA$, see (\ref{ss:fa}).

6266: On the geometric side,

6267: we are now dealing with

6268: a symplectic operad manifold.

6269: The examples in (\ref{ss:se}) and (\ref{ss:ae})

6270: correspond to the commutative and surface species

6271: respectively.

6272: 

6273: \begin{remark}

6274: By our construction, the vector space $QA$ is graded

6275: beginning with degree $2$.

6276: For example,

6277: $ccA$ are polynomials with no constant or linear terms.

6278: Also the monomials in $ssA$ have at least two boundary circles.

6279: However, for the deformation quantization problem,

6280: one needs to enlarge $QA$ suitably 

6281: by adding stuff in degrees 0 and 1,

6282: as was done for the previous two examples.

6283: \end{remark}

6284: 

6285: We first want to give a Poisson structure to $QA$

6286: and then deform it by the Moyal product.

6287: We know that $QA$ has a Lie structure $\{\ ,\ \}$.

6288: This was discussed in (\ref{ss:mate}).

6289: In terms of pictures, it is given by equation~\eqref{e:mps}.

6290: The interpretation of the picture is that

6291: a mating occurs along the edge.

6292: But what is the commutative structure on $QA$ ?

6293: We have emphasised that an element of $QA$ 

6294: can be represented by a picture.

6295: So take $F \cdot H$ to be the

6296: ``disjoint union of pictures''

6297: that represent $F$ and $H$.

6298: The implicit assumption is that

6299: the species $Q$ can be written as the exponential

6300: of some other species.

6301: This is our proposal for the Poisson structure on $QA$.

6302: From the viewpoint of pictures,

6303: the compatibility of the commutative and Lie products is clear.

6304: 

6305: Next we want to define the Moyal product

6306: and show that it solves the deformation quantisation problem.

6307: In other words, we want to understand the meaning of

6308: $\frac{B^n}{n!}(F,H)$ for $n \geq 2$,

6309: where $B$ is the bi-differential operator

6310: defined in (\ref{ss:mp}).

6311: For $n=0,1$, we know that 

6312: $B^0(F,H)=F \cdot H$ and $B(F,H) = \{F,H\}$.

6313: Writing

6314: $\frac{B^n}{n!} = \sum B_{(1)} \otimes B_{(2)}$,

6315: where $B_{(1)}$ and $B_{(2)}$ are $n$th order differential operators,

6316: we obtain

6317: $\frac{B^n}{n!}(F,H) = \sum B_{(1)} F \otimes B_{(2)} H$.

6318: One needs to make sense of this.

6319: We first explain how a higher order derivative works

6320: by showing it on a schematic example.

6321: \begin{equation}  \label{e:hd}

6322: \begin{minipage}{.8 in}

6323: \begin{center}

6324: \psfrag{a}{\Huge $x_2$}

6325: \psfrag{b}{\Huge $x_1$}

6326: \psfrag{c}{\Huge $x_1$}

6327: \psfrag{d}{\Huge $x_4$}

6328: \resizebox{.6 in}{.6 in}{\includegraphics{q.eps}}

6329: \end{center}

6330: \end{minipage}

6331: \qquad \buildrel \frac{\partial^2}{\partial x_1^2}\over \longmapsto \qquad

6332: 2 \left(

6333: \begin{minipage}{.8 in}

6334: \begin{center}

6335: \psfrag{x1}{\Huge $x_1$}

6336: \psfrag{x2}{\Huge $x_2$}

6337: \psfrag{x4}{\Huge $x_4$}

6338: \resizebox{.7 in}{.6 in}{\includegraphics{pra.eps}}

6339: \end{center}

6340: \end{minipage}

6341: \right).

6342: \end{equation}

6343: Thus a $n$th order differential operator acting on $F$

6344: produces an algebraic object with $n$ outputs

6345: (\ref{ss:prop}).

6346: We now interpret the tensor sign as a $n$th order mating.

6347: We write

6348: $$

6349: \frac{B^n}{n!}(F,H) = 

6350: \sum_P (-1)^{\sign(P)}

6351: \left(

6352: \begin{minipage}{1.2 in}

6353: \begin{center}

6354: \psfrag{F}{\Huge $F$}

6355: \psfrag{H}{\Huge $H$}

6356: \psfrag{n}{\Huge $n$}

6357: \resizebox{1 in}{.4 in}{\includegraphics{pmm.eps}}

6358: \end{center}

6359: \end{minipage} 

6360: \right)

6361: ,$$

6362: where the picture $P$ shows a higher order mating.

6363: There are $n$ ideal edges between $F$ and $H$ 

6364: each connecting a $p_i$ with a $q_i$ or vice-versa.

6365: And $\sign(P)$ is the number of edges that connect

6366: a $q_i$ on $F$ to a $p_i$ on $H$.

6367: For the definition of an ideal edge, see (\ref{ss:mf}).

6368: 

6369: The examples discussed in Section~\ref{s:me}

6370: have associated PROPs.

6371: In those cases, 

6372: we understand the meaning of higher order derivatives,

6373: i.e. the right hand side of equation~\eqref{e:hd}

6374: makes sense.

6375: Similarly, higher order matings 

6376: have a natural meaning in those examples.

6377: The example of the surface species was illustrated in

6378: (\ref{ss:ae}).

6379: The deformation quantisation for the Poisson algebra $A$

6380: is again given by the formula in equation~\eqref{e:mp}.

6381: The Moyal $\star$ product is associative

6382: in these cases for the same reason 

6383: as in the commutative case $(Q=cc)$.

6384: The associativity can be seen from the formula

6385: $$

6386: F \star G \star H = 

6387: \sum_{\substack {n \geq 0 \\ n=k+l+m}} t^n

6388: \sum_P (-1)^{\sign(P)}

6389: \left(

6390: \begin{minipage}{1.2 in}

6391: \begin{center}

6392: \psfrag{F}{\Huge $F$}

6393: \psfrag{G}{\Huge $G$}

6394: \psfrag{H}{\Huge $H$}

6395: \psfrag{k}{\Huge $k$}

6396: \psfrag{l}{\Huge $l$}

6397: \psfrag{m}{\Huge $m$}

6398: \psfrag{n}{\Huge $n$}

6399: \resizebox{1 in}{.8 in}{\includegraphics{pmam.eps}}

6400: \end{center}

6401: \end{minipage} 

6402: \right)

6403: ,$$

6404: \noindent

6405: where $\sign(P)$ is the number of ideal edges that connect

6406: a $q_i$ to a $p_i$. 

6407: (The edges go from $F$ to $G$, $G$ to $H$ 

6408: and $F$ to $H$.)

6409: The letters $k,l,m$ stand for the number of edges.

6410: 

6411: The interested reader can work out the details 

6412: for the examples

6413: based on graphs (\ref{ss:g}-\ref{ss:gt}).

6414: Another nice example is obtained 

6415: by exponentiating the associative species.

6416: 

6417: \section{The deformation map on graphs}  \label{ss:dg}

6418: 

6419: Recall that the chain complex $(\mathcal G,\partial_E)$ 

6420: is a differential graded Hopf algebra

6421: with product $\mu$ given by disjoint union.

6422: For a definition of $(\mathcal G,\partial_E)$,

6423:  see Section~\ref{s:gh}

6424: and for the Hopf algebra structure,

6425: see the remark at the end of Section~\ref{s:con}.

6426: In addition, recall that we defined 

6427: another boundary operator $\partial_H$ on $\mathcal G$

6428: that satisfied

6429: $\partial_E \partial_H = - \partial_H \partial_E$,

6430: see (\ref{ss:finb}).

6431: So $\partial_H$ induces a map on homology

6432: $H_*(\mathcal G,\partial_E) \to H_*(\mathcal G,\partial_E)$. 

6433: 

6434: The boundary map $\partial_E$ is a

6435: derivation with respect to the product $\mu$.

6436: This can be written as $\partial_E \mu = \mu \partial_E$.

6437: However $\partial_H$ is not a derivation with respect to $\mu$.

6438: Following~\cite{\cv},

6439: define a bracket

6440: $[\ ,\ ] : \mathcal G \otimes \mathcal G \to \mathcal G$ by

6441: \[

6442: [\ ,\ ] = \partial_H \mu - \mu \partial_H.

6443: \]

6444: In other words,

6445: $[\ ,\ ]$ measures the failure of $\partial_H$

6446: to be a derivation with respect to $\mu$.

6447: As a formal consequence of the previous three identities,

6448: it follows that

6449: $\partial_E [\ ,\ ] = - [\ ,\ ] \partial_E$.

6450: This gives us an induced map on homology

6451: $[\ ,\ ] : H_*(\mathcal G,\partial_E) \otimes H_*(\mathcal G,\partial_E)

6452: \to H_*(\mathcal G,\partial_E)$.

6453: 

6454: In this section, we show the following.

6455: 

6456: \begin{proposition}  \label{p:triv}

6457: The maps $\partial_H$ and $[\ ,\ ]$ are trivial on homology.

6458: \end{proposition}

6459: 

6460: This answers some of the questions raised in~\cite{\cv}.

6461: The proof is largely a matter of putting together

6462: things we already know.

6463: The main step is to use 

6464: the pairing $M(n)$ on $\mathcal G$ 

6465: and its adjoint property 

6466: proved in Proposition~\ref{p:ad}. 

6467: 

6468: The coproduct on $\mathcal G$ is irrelevant to Proposition~\ref{p:triv}.

6469: So we write the differential algebra $\mathcal G$ as a triple

6470: $(\mathcal G,\partial_E,\mu)$.

6471: Extending $\partial_E$ and $\mu$ by $\Q[[t]]$ linearity,

6472: we get a differential algebra

6473: $(\mathcal G[[t]],\partial_E,\mu)$. 

6474: In other words, this is just the

6475: trivial deformation of the algebra $\mathcal G$,

6476: see (\ref{ss:tp}).

6477: We now consider another deformation of $\mathcal G$ that is

6478: gauge equivalent to this one (\ref{ss:ge}).

6479: 

6480: \subsection{The deformation map}

6481: 

6482: Let $D(n) : \mathcal G \to \mathcal G$ be the deformation map

6483: defined at the end of (\ref{ss:pnd}).

6484: It was defined by $D(n) = A^{-1} \circ M(n)$,

6485: where the map $M(n)$ was induced by a pairing on $\mathcal G$

6486: (equation \eqref{e:pg}),

6487: while the map $A$ scaled a graph $\Gamma$ by 

6488: the factor $|\Aut(\Gamma)|$.

6489: 

6490: Recall that $D(n) \Gamma$

6491: is a polynomial in $2n$ of degree $e$

6492: with coefficients in $\mathcal G$.

6493: Furthermore, the coefficient of $(2n)^e$ is exactly $\Gamma$.

6494: Here $e=|E(\Gamma)|$, the number of edges in $\Gamma$.

6495: Write

6496: $D(n) \Gamma = (2n)^e (D_0 + \frac{1}{2n} D_1 + \ldots)(\Gamma)$.

6497: This defines maps $D_i : \mathcal G \to \mathcal G$ 

6498: which do not depend on $n$. Also $D_0$ is the identity.

6499: 

6500: Consider the $\Q[[t]]$ linear automorphism of $\mathcal G[[t]]$

6501: given by $D = D_0 + t D_1 + t^2 D_2 + \ldots$,

6502: where $D_i : \mathcal G \to \mathcal G$ are as above.

6503: The map $D$ defines an element of the gauge group

6504: (\ref{ss:ge}).

6505: 

6506: \subsection{Comparing the two gauge equivalent situations}

6507: 

6508: Let $(\mathcal G[[t]],\partial_t,\mu_t)$

6509: be the differential algebra got by applying $D$

6510: to the trivial deformation $(\mathcal G[[t]],\partial_E,\mu)$. 

6511: Stated differently, one has two commutative diagrams. 

6512: \begin{equation}  \label{e:dcdiag}

6513: \begin{CD}

6514: \mathcal G[[t]] @>D>> \mathcal G[[t]]\\

6515:  @V \partial_E VV @V \partial_t VV \\

6516: \mathcal G[[t]] @>D>> \mathcal G[[t]]. \\

6517: \end{CD}

6518: \hspace{.6 in}

6519: \begin{CD}

6520: \mathcal G[[t]] \otimes \mathcal G[[t]] @>D \otimes D>> \mathcal G[[t]] \otimes \mathcal G[[t]] \\

6521:  @V \mu VV @V \mu_t VV \\

6522: \mathcal G[[t]] @>D>> \mathcal G[[t]]. \\

6523: \end{CD}

6524: \end{equation}

6525: %

6526: Since $\partial_E$ is a

6527: derivation with respect to the product $\mu$,

6528: we have $\partial_E \mu = \mu \partial_E$.

6529: Hence

6530: we know $\partial_t \mu_t = \mu_t \partial_t$.

6531: Write

6532: $\partial_t = \partial_0 + t \partial_1 + t^2 \partial_2 + \ldots$

6533: and

6534: $\mu_t = \mu_0 + t \mu_1 + t^2 \mu_2 + \ldots$.

6535: Then the commutativity of the diagrams in~\eqref{e:dcdiag}

6536: implies

6537: $\partial_0 = \partial_E$ and $\mu_0 = \mu$.

6538: 

6539: \begin{lemma} \label{l:drel}

6540: The following relations hold.

6541: \begin{itemize}

6542: \item[(1)]

6543: $- \partial_1 = - D_1 \partial_0 + \partial_0 D_1$.

6544: 

6545: \item[(2)]

6546: $\mu_1 = D_1 \mu_0 - \mu_0 D_1$.

6547: 

6548: \item[(3)]

6549: $- \partial_1 \mu_0 + \mu_0 \partial_1 = \partial_0 \mu_1 - \mu_1 \partial_0$.

6550: \end{itemize}

6551: \end{lemma}

6552: \begin{proof}

6553: The first two items follow by looking at the coefficient of $t$

6554: in the diagrams in~\eqref{e:dcdiag}

6555: and the third item follows by looking at the coefficient of $t$

6556: in $\partial_t \mu_t = \mu_t \partial_t$.

6557: \end{proof}

6558: 

6559: \begin{lemma}

6560: $\partial_1 = - \partial_H$.

6561: \end{lemma}

6562: \begin{proof}

6563: The commutative diagram~\eqref{e:cdg} in Section~\ref{s:mgh}

6564: can be redrawn as under

6565: \begin{equation*} 

6566: \begin{CD}

6567: \mathcal G[[t]] @>D>> \mathcal G[[t]]\\

6568:  @V \partial_E + t \partial_H VV @V \partial_E VV \\

6569: \mathcal G[[t]] @>D>> \mathcal G[[t]]. \\

6570: \end{CD}

6571: \end{equation*}

6572: Comparing the coefficient of $t$ gives

6573: $\partial_H = - D_1 \partial_0 + \partial_0 D_1$.

6574: This together with Lemma~\ref{l:drel} (item (1))

6575: shows that $\partial_1 = - \partial_H$.

6576: \end{proof}

6577: 

6578: \subsection*{Proof of Proposition~\ref{p:triv}}

6579: Items $(1)$ and $(3)$ in Lemma~\ref{l:drel}

6580: can now be rewritten as 

6581: $\partial_H = - D_1 \partial_E + \partial_E D_1$ and

6582: $[\ ,\ ] = \partial_E \mu_1 - \mu_1 \partial_E$.

6583: Thus $\partial_H$ and $[\ ,\ ]$ induce the zero map on homology,

6584: with $D_1$ and $\mu_1$ providing the respective chain homotopies.

6585: This completes the proof of Proposition~\ref{p:triv}.

6586: 

6587: \begin{remark}

6588: It is possible to just give an explicit definition of $\mu_1$

6589: and check the relation $[\ ,\ ] = \partial_E \mu_1 - \mu_1 \partial_E$

6590: directly.

6591: This will give us that $[\ ,\ ]$ is zero on homology.

6592: And similarly for $D_1$.

6593: But we prefer the more conceptual approach 

6594: via deformation theory.

6595: It would be interesting to also describe explicitly

6596: the higher products

6597: $\mu_2, \mu_3, \ldots$ and

6598: $\partial_2, \partial_3, \ldots$, etc.

6599: 

6600: Recall that $\mathcal G$ also has a coproduct $\Delta$ and

6601: $\partial_E$ is a coderivation with respect to the coproduct $\Delta$.

6602: However, $\partial_H$ is not and

6603: this failure can be measured by a cobracket

6604: $\theta : \mathcal G \to \mathcal G \otimes \mathcal G$ given by

6605: $\theta = \partial_H \Delta - \Delta \partial_H$.

6606: Similar arguments show that 

6607: $\theta$ also induces the zero map on homology.

6608: \end{remark}

6609: 

6610: \subsection*{Acknowledgements}

6611: I thank Ken Brown for being a constant source of

6612: knowledge and encouragement.

6613: Among the participants of the Bernstein seminar

6614: at Cornell University (Fall 2000),

6615: Dan Ciubotaru, Ferenc Gerlits and Jim Conant

6616: deserve a special thank you. 

6617: I also thank M. Aguiar, Y. Berest and M. Yakimov

6618: for helpful comments.

6619: Finally, I would like to thank Kontsevich

6620: without whom this paper would never have been written.

6621: I hope that the reader will go back 

6622: and read his original papers,

6623: for which there can be no substitute.

6624: It is fitting to conclude with a Kontsevich sentence.

6625: We choose the opening sentence of~\cite{\kont}.

6626: \begin{quotation}

6627: We shall describe a program here relating Feynman diagrams,

6628: topology of manifolds, homotopical algebra,

6629: non-commutative geometry and

6630: several kinds of ``topological physics.''

6631: \end{quotation}

6632: Enjoy!

6633: 

6634: %\input e

6635: 

6636: %\input app

6637: 

6638: %\bibliographystyle{hamsplain}

6639: %\bibliography{quantum}

6640: 

6641: 

6642: %\input a.bbl

6643: \def\cprime{$'$} \def\cprime{$'$} \def\cprime{$'$}

6644: \providecommand{\bysame}{\leavevmode\hbox to3em{\hrulefill}\thinspace}

6645: \begin{thebibliography}{10}

6646: 

6647: \bibitem{adams78:_infin}

6648: John~Frank Adams, \emph{Infinite loop spaces}, Princeton University Press,

6649:   Princeton, N.J., 1978.

6650: 

6651: \bibitem{bar-natan01:_graph}

6652: Dror Bar-Natan and Brendan~D. Mckay, \emph{Graph cohomology - an overview and

6653:   some computations}, in preparation, 2001.

6654: 

6655: \bibitem{bergeron98:_combin}

6656: F.~Bergeron, G.~Labelle, and P.~Leroux, \emph{Combinatorial species and

6657:   tree-like structures}, Cambridge University Press, Cambridge, 1998,

6658:   Translated from the 1994 French original by Margaret Readdy, With a foreword

6659:   by Gian-Carlo Rota.

6660: 

6661: \bibitem{bocklandt:_neckl_lie}

6662: Raf Bocklandt and Lieven~Le Bruyn, \emph{{Necklace Lie algebras and

6663:   noncommutative symplectic geometry}}, UIA-00-12,

6664:   \mbox{arXiv:math.AG/0010030}.

6665: 

6666: \bibitem{boothby86:_rieman}

6667: William~M. Boothby, \emph{An introduction to differentiable manifolds and

6668:   {R}iemannian geometry}, second ed., Academic Press Inc., Orlando, FL, 1986.

6669: 

6670: \bibitem{brown94:_cohom}

6671: Kenneth~S. Brown, \emph{Cohomology of groups}, Springer-Verlag, New York, 1994,

6672:   Corrected reprint of the 1982 original.

6673: 

6674: \bibitem{conant:_infin_operat_graph_graph_homol}

6675: Jim Conant and Karen Vogtmann, \emph{{Infinitesimal Operations on Graphs and

6676:   Graph Homology}}, \mbox{arXiv:math.QA/0111198}.

6677: 

6678: \bibitem{drinfelprimed90:_hopf_gal_q_q}

6679: V.~G. Drinfel{\cprime}d, \emph{On quasitriangular quasi-{H}opf algebras and on

6680:   a group that is closely connected with {${\rm Gal}(\overline{\bf Q}/{\bf

6681:   Q})$}}, Algebra i Analiz \textbf{2} (1990), no.~4, 149--181.

6682: 

6683: \bibitem{fuks86:_cohom_lie}

6684: D.~B. Fuks, \emph{Cohomology of infinite-dimensional {L}ie algebras},

6685:   Contemporary Soviet Mathematics, Consultants Bureau, New York, 1986,

6686:   Translated from the Russian by A. B. Sosinski\u\i.

6687: 

6688: \bibitem{fulton91:_repres}

6689: William Fulton and Joe Harris, \emph{Representation theory}, Springer-Verlag,

6690:   New York, 1991, A first course, Readings in Mathematics.

6691: 

6692: \bibitem{gelprimefand70:_cohom_lie}

6693: I.~M. Gel{\cprime}fand and D.~B. Fuks, \emph{Cohomologies of the {L}ie algebra

6694:   of formal vector fields}, Izv. Akad. Nauk SSSR Ser. Mat. \textbf{34} (1970),

6695:   322--337.

6696: 

6697: \bibitem{gerlits02:_calcul}

6698: Ferenc Gerlits, \emph{Calculations in graph homology}, Ph.D. thesis, Cornell

6699:   University, 2002.

6700: 

6701: \bibitem{getzler94:_batal_vilkov}

6702: E.~Getzler, \emph{Batalin-{V}ilkovisky algebras and two-dimensional topological

6703:   field theories}, Comm. Math. Phys. \textbf{159} (1994), no.~2, 265--285.

6704: 

6705: \bibitem{getzler95:_cyclic}

6706: E.~Getzler and M.~M. Kapranov, \emph{Cyclic operads and cyclic homology},

6707:   Geometry, topology, \& physics, Internat. Press, Cambridge, MA, 1995,

6708:   pp.~167--201.

6709: 

6710: \bibitem{getzler98:_modul}

6711: \bysame, \emph{Modular operads}, Compositio Math. \textbf{110} (1998), no.~1,

6712:   65--126.

6713: 

6714: \bibitem{ginzburg01:_non}

6715: Victor Ginzburg, \emph{Non-commutative symplectic geometry, quiver varieties,

6716:   and operads}, Math. Res. Lett. \textbf{8} (2001), no.~3, 377--400.

6717: 

6718: \bibitem{ginzburg94:_koszul}

6719: Victor Ginzburg and Mikhail Kapranov, \emph{Koszul duality for operads}, Duke

6720:   Math. J. \textbf{76} (1994), no.~1, 203--272.

6721: 

6722: \bibitem{jacobson79:_lie}

6723: Nathan Jacobson, \emph{Lie algebras}, Dover Publications Inc., New York, 1979,

6724:   Republication of the 1962 original.

6725: 

6726: \bibitem{joyal81:_une}

6727: Andr{\'e} Joyal, \emph{Une th\'eorie combinatoire des s\'eries formelles}, Adv.

6728:   in Math. \textbf{42} (1981), no.~1, 1--82.

6729: 

6730: \bibitem{karoubi95:_formes}

6731: Max Karoubi, \emph{Formes diff\'erentielles non commutatives et cohomologie \`a

6732:   coefficients arbitraires}, Trans. Amer. Math. Soc. \textbf{347} (1995),

6733:   no.~11, 4277--4299.

6734: 

6735: \bibitem{kassel95:_quant}

6736: Christian Kassel, \emph{Quantum groups}, Springer-Verlag, New York, 1995.

6737: 

6738: \bibitem{kontsevich93:_formal}

6739: Maxim Kontsevich, \emph{Formal (non)commutative symplectic geometry}, The

6740:   Gel\cprime fand Mathematical Seminars, 1990--1992, Birkh\"auser Boston,

6741:   Boston, MA, 1993, pp.~173--187.

6742: 

6743: \bibitem{kontsevich94:_feynm}

6744: \bysame, \emph{Feynman diagrams and low-dimensional topology}, First European

6745:   Congress of Mathematics, Vol.\ II (Paris, 1992), Birkh\"auser, Basel, 1994,

6746:   pp.~97--121.

6747: 

6748: \bibitem{lazard95:_lois}

6749: Michel Lazard, \emph{Lois de groupes et analyseurs}, S\'eminaire Bourbaki,

6750:   Vol.\ 3, Soc. Math. France, Paris, 1995, pp.~Exp.\ No.\ 109, 77--91.

6751: 

6752: \bibitem{loday01:_dialg}

6753: J.-L. Loday, A.~Frabetti, F.~Chapoton, and F.~Goichot, \emph{Dialgebras and

6754:   related operads}, Springer-Verlag, Berlin, 2001.

6755: 

6756: \bibitem{loday97:_overv_leibn}

6757: Jean-Louis Loday, \emph{Overview on {L}eibniz algebras, dialgebras and their

6758:   homology}, Cyclic cohomology and noncommutative geometry (Waterloo, ON,

6759:   1995), Fields Inst. Commun., vol.~17, Amer. Math. Soc., Providence, RI, 1997,

6760:   pp.~91--102.

6761: 

6762: \bibitem{loday98:_cyclic}

6763: \bysame, \emph{Cyclic homology}, second ed., Springer-Verlag, Berlin, 1998,

6764:   Appendix E by Mar\'\i a O. Ronco, Chapter 13 by the author in collaboration

6765:   with Teimuraz Pirashvili.

6766: 

6767: \bibitem{loday97:_operad}

6768: Jean-Louis Loday, James~D. Stasheff, and Alexander~A. Voronov (eds.),

6769:   \emph{Operads: {P}roceedings of {R}enaissance {C}onferences}, American

6770:   Mathematical Society, Providence, RI, 1997, Papers from the Special Session

6771:   on Moduli Spaces, Operads and Representation Theory held at the AMS Meeting

6772:   in Hartford, CT, March 4--5, 1995, and from the Conference on Operads and

6773:   Homotopy Algebra held in Luminy, May 29--June 2, 1995.

6774: 

6775: \bibitem{markl99:_cyclic}

6776: Martin Markl, \emph{Cyclic operads and homology of graph complexes}, Rend.

6777:   Circ. Mat. Palermo (2) Suppl. (1999), no.~59, 161--170, The 18th Winter

6778:   School ``Geometry and Physics'' (Srn\'\i, 1998).

6779: 

6780: \bibitem{may72}

6781: J.~P. May, \emph{The geometry of iterated loop spaces}, Springer-Verlag,

6782:   Berlin, 1972, Lectures Notes in Mathematics, Vol. 271.

6783: 

6784: \bibitem{mccleary01}

6785: John McCleary, \emph{A user's guide to spectral sequences}, second ed.,

6786:   Cambridge University Press, Cambridge, 2001.

6787: 

6788: \bibitem{mcduff98:_introd}

6789: Dusa McDuff and Dietmar Salamon, \emph{Introduction to symplectic topology},

6790:   second ed., The Clarendon Press Oxford University Press, New York, 1998.

6791: 

6792: \bibitem{milnor65:_hopf}

6793: John~W. Milnor and John~C. Moore, \emph{On the structure of {H}opf algebras},

6794:   Ann. of Math. (2) \textbf{81} (1965), 211--264.

6795: 

6796: \bibitem{spivak79}

6797: Michael Spivak, \emph{A comprehensive introduction to differential geometry.

6798:   {V}ol. {I}}, second ed., Publish or Perish Inc., Wilmington, Del., 1979.

6799: 

6800: \bibitem{stasheff63:_homot_h}

6801: James~Dillon Stasheff, \emph{Homotopy associativity of ${H}$-spaces. {I},

6802:   {I}{I}}, Trans. Amer. Math. Soc. 108 (1963), 275-292; ibid. \textbf{108}

6803:   (1963), 293--312.

6804: 

6805: \bibitem{sweedler69:_hopf}

6806: Moss~E. Sweedler, \emph{Hopf algebras}, W. A. Benjamin, Inc., New York, 1969.

6807: 

6808: \bibitem{thurston:_integ_expres_vassil_knot_invar}

6809: Dylan~P. Thurston, \emph{{Integral Expressions for the Vassiliev Knot

6810:   Invariants}}, \mbox{arXiv:math.QA/9901110}.

6811: 

6812: \bibitem{voronov:_notes}

6813: Alexander~A. Voronov, \emph{{Notes on universal algebra}},

6814:   \mbox{arXiv:math.QA/0111009}.

6815: 

6816: \bibitem{weibel94}

6817: Charles~A. Weibel, \emph{An introduction to homological algebra}, Cambridge

6818:   University Press, Cambridge, 1994.

6819: 

6820: \end{thebibliography}

6821: 

6822: 

6823: \end{document}

6824: 

6825: 

6826: 

6827: 

6828: 

6829: 

6830: