1: 

2: \documentclass[paper, 12pt, letterpaper, epsf]{JHEP} 

3: \input{epsf.tex}

4: \usepackage{graphics} 

5: \usepackage{epsfig} 

6: %\usepackage{amsmath}

7: \usepackage{amssymb}

8: %\renewcommand{\theequation}{\thesubsection.\arabic{equation}}

9: 

10: %\NeedsTeXFormat{LaTeX2e}[1995/12/01]

11: %\documentclass[titlepage,12pt]{utarticle}   

12: 

13: \usepackage{graphics}

14: \usepackage{latexsym,amsmath,amssymb,amscd,cite}

15: \usepackage{pdfsync}

16: \usepackage{epsf}

17: %\usepackage{showkeys}

18: %\usepackage{hyperref}

19: %\hypersetup{colorlinks=true, linkcolor=green,  anchorcolor=magenta,

20: %citecolor=blue, filecolor=blue, menucolor=blue, pagecolor=blue, urlcolor=red}

21: \usepackage[all]{xy}

22: %\usepackage{amsmath}

23: \usepackage{amsthm}

24: 

25: \newtheorem{Theorem}{Theorem}[section]

26: \newtheorem{Lemma}[Theorem]{Lemma}

27: \newtheorem{Proposition}[Theorem]{Proposition}

28: \theoremstyle{definition}

29: \newtheorem{Definition}[Theorem]{Definition}

30: \theoremstyle{remark}

31: \newtheorem*{Remark}{Remark}

32: 

33: \newcommand{\Hom}{{\rm Hom}}

34: 

35: \newcommand{\A}{{\Bbb{A}}}

36: \newcommand{\Spec}{{\rm Spec}}

37: 

38: \newcommand{\bp}{\begin{Proposition}}

39: \newcommand{\ep}{\end{Proposition}}

40: \newcommand{\bl}{\begin{Lemma}}

41: \newcommand{\el}{\end{Lemma}}

42: \newcommand{\bt}{\begin{Theorem}}

43: \newcommand{\et}{\end{Theorem}}

44: \newcommand{\bd}{\begin{Definition}}

45: \newcommand{\ed}{\end{Definition}}

46: \newcommand{\End}{\rm{End}}

47: \newcommand{\Aut}{\rm{Aut}}

48: %\newcommand{\id}{\rm{id}}

49: \newcommand{\td}{\tilde{}}

50: \newcommand{\Mod}{\rm{Mod}}

51: \newcommand{\Mom}{\rm{Mom}}

52: \newcommand{\Edual}{E[1]^{\rm{v}}}

53: \newcommand{\iEnd}{{\underline \End}}

54: \newcommand{\iHom}{{\underline \Hom}}

55: \newcommand{\Rep}{{\rm Rep}}

56: \newcommand{\g}{{\bf g}}

57: \newcommand{\Mat}{{\rm Mat}}

58: \newcommand{\ev}{{\rm ev}}

59: 

60: \DeclareFontFamily{U}{rsf}{}

61: \DeclareFontShape{U}{rsf}{m}{n}{<5> <6> rsfs5 <7> <8> <9> rsfs7 <10-> rsfs10}{}

62: \DeclareMathAlphabet\Scr{U}{rsf}{m}{n}

63: 

64: \def\N{{\Bbb N}}

65: \def\Z{{\Bbb Z}}

66: \def\C{{\Bbb C}}

67: \def\R{{\Bbb R}}

68: \def\rk{{\rm rk}}

69: \def\deg{{\rm deg}}

70: \def\Der{{\rm Der}}

71: \def\Sym{{\rm Sym}}

72: 

73: \def\lp{\overrightarrow{\partial}}

74: \def\rp{\overleftarrow{\partial}}

75: \def\ld{\overrightarrow{\delta}}

76: \def\rd{\overleftarrow{\delta}}

77: \def\li{\overrightarrow{i}}

78: \def\ri{\overleftarrow{i}}

79: \def\lL{\overrightarrow{L}}

80: \def\rL{\overleftarrow{L}}

81: \def\ltheta{\overrightarrow{\theta}}

82: \def\rtheta{\overleftarrow{\theta}}

83: \def\tf{{\tilde f}}

84: \def\tom{{\tilde \omega}}

85: \def\th{{\tilde \theta}}

86: \def\ta{{\tilde a}}

87: \def\tb{{\tilde b}}

88: \def\lD{\overrightarrow{D}}

89: \def\rD{\overleftarrow{D}}

90: \def\l{\partial^l}

91: 

92: %:wlmacros

93: \def\frc#1#2{\frac1{#2}{#1}}

94: %\def\bib#1#2{\bib{#1}{#2}}%  for conventional use

95: \def\bib#1#2{\newcommand{#1}{#2}}%  for use with refsgen.sh

96: 

97: \newcommand{\be}{\begin{equation}}

98: \newcommand{\ee}{\end{equation}}

99: \newcommand{\bea}{\end{eqnarray}}

100: \newcommand{\eea}{\end{eqnarray}}

101: \newcommand{\nn}{\nonumber}

102: 

103: \newcommand{\beql}[1]{\begin{eqnarray}\label{eq:#1}}

104: 

105: \newcommand{\eeql}{\end{eqnarray}}

106: 

107: \long\def\eqn#1#2{\begin{eqnarray}\label{eq:#1}#2\end{eqnarray}}

108: \long\def\begdel#1\enddel{}

109: %\long\def\new#1\endnew{{\bf #1}}

110: \def\rf#1{(\ref{eq:#1})}

111: 

112: 

113: 

114: \newcommand{\Res}{{\rm Res}}

115: \newcommand{\id}{{\rm id}}

116: \newcommand{\Tr}{{\rm Tr}}

117: \newcommand{\tr}{{\rm tr}}

118: \newcommand{\str}{{\rm str}}

119: \newcommand{\diag}{{\rm diag}}

120: 

121: \def\lto{\longrightarrow}

122: 

123: \def\lab#1{^{(#1)}}

124: \def\Lab#1{^{[#1]}}

125: 

126: \def\half{{1\over2}}

127: 

128: \def\non{\nonumber}

129: 

130: \def\del{\partial}

131: \def\cR{{\mathcal R}}

132: \def\cC{{\mathcal C}}

133: 

134: \def\cB{\Scr B} 

135: \def\BC{{\mathbb C}}

136: 

137: \def\BZ{{\mathbb Z}}

138: \def\BM{{\mathbb M}}

139: \def\BS{{\mathbb S}}

140: \def\BE{{\mathbb E}}

141: \def\BX{{\mathbb X}}

142: 

143: \def\BJ{{\mathbb J}}

144: \def\BA{{\mathbb A}}

145: \def\BB{{\mathbb B}}

146: \def\cZ{{\cal Z}}

147: \def\al{\alpha}

148: \def\bfone{{\bf 1}}

149: \def \unity{\bf 1}

150: 

151: 

152: %paper specific macros

153: \def\bc{boundary condition changing\ }

154: 

155: %\def\bp{boundary preserving\ }

156: 

157: \def\weff{{\mathcal W}_{\rm eff}}

158: 

159: %\def\wlg{{W}_{\rm LG}}

160: \def\wlg{W}

161: 

162: \def\hll{\half(\ell_1+\ell_2)}

163: %\def\vp{\varphi}

164: \def\vp{x}

165: \def\del{\partial}

166: \def\JJ{\BJ}

167: \def\EE{\BE}

168: 

169: 

170: 

171: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% conventions

172: 

173: 

174: \title{On the non-commutative geometry of topological D-branes}

175: \author{~~~Calin Iuliu Lazaroiu\\Department of Mathematics\\Trinity College Dublin\\Dublin 2, Ireland\\calin@maths.tcd.ie\\}

176: 

177: \abstract{This is a noncommutative-geometric study of the

178: semiclassical dynamics of finite topological D-brane systems. Starting

179: from the formulation in terms of $A_\infty$ categories, I show that

180: such systems can be described by the noncommutative symplectic

181: supergeometry of $\Z_2$-graded quivers, and give a synthetic formulation of the 

182: boundary part of the generalized WDVV equations. In particular, a {\em faithful} generating

183: function for integrated correlators on the disk can be constructed as a

184: linear combination of quiver necklaces, i.e.  a function

185: on the noncommutative symplectic superspace defined by the quiver's

186: path algebra. This point of view allows one to construct extended

187: moduli spaces of topological D-brane systems as non-commutative

188: algebraic `superschemes'. They arise by imposing further relations on

189: a $\Z_2$-graded version of the quiver's preprojective algebra, and

190: passing to the subalgebra preserved by a natural group of symmetries. }

191: 

192: \preprint{}

193: 

194: \begin{document}

195: 

196: 

197: \tableofcontents

198: 

199: \pagebreak

200: 

201: \vskip .6in

202: 

203: 

204: \section{Introduction}

205: 

206: The extended moduli spaces \cite{Witten_mirror} of closed topological

207: strings are Frobenius supermanifolds\cite{Dubrovin}, certain types of

208: flat Riemannian supermanifolds whose metric is induced by the

209: generating functional of tree-level closed string amplitudes. This

210: description follows from the consistency constraints on closed string

211: amplitudes on the sphere, known as  the WDVV equations

212: \cite{WDVV}. The theory of Frobenius supermanifolds encodes many

213: interesting properties such as the generic lack of

214: obstructions\footnote{A manifestation of this is the

215: Bogomolov-Tian-Todorov theorem \cite{Tian, Todorov} on

216: unobstructedness of deformations of complex structure for Calabi-Yau

217: manifolds, and its generalization to extended deformations due to

218: Barannikov and Kontsevich \cite{BarKon}.  The extended moduli space of

219: complex structures for such manifolds coincides with the moduli space

220: of deformations of the associated topological B-type string. }  of

221: (not necessarily conformal) topological bulk deformations, and has

222: well-known applications to closed string mirror symmetry.

223: 

224: 

225: It is natural to ask if a similar description can be given for

226: topological deformations of open strings.  A boundary topological

227: string theory admits two types of deformations, which are induced by

228: bulk and boundary observables. As in the closed string case, the

229: open-closed tree-level amplitudes obey consistency conditions (the

230: so-called {\em generalized WDVV equations}), which were derived in

231: \cite{HLL} by worldsheet arguments. While bulk deformations have the

232: same character as in the boundary-free case, boundary deformations

233: behave quite differently.  As shown in \cite{HLL}, they are

234: constrained by a {\em homotopy} version of the associativity

235: conditions, leading to an intricate structure known as a cyclic and

236: unital weak $A_\infty$ category (see \cite{CIL5, Stasheff_Kajiura, Kajiura, Gaberdiel, Fukaya_mirror, KS, 

237: CIL4, CIL8, CIL9, CIL10, Costello1, Costello2} for related

238: work).  The complication arises due to the non-commutativity of

239: boundary insertions on the disk, and leads to difficulties

240: when analyzing the boundary moduli problem.  Among these is the observation that the 

241: the homotopy associativity constraints for the integrated boundary disk amplitudes

242: $W_{a_1\ldots a_n}$ cannot be `integrated' faithfully to an ordinary

243: generating function of the boundary deformation parameters.

244: 

245: In the present paper, we investigate a resolution to this problem, by

246: arguing quite generally that the semiclassical dynamics of

247: open strings in finite topological D-brane systems can be described in

248: the framework of supersymplectic noncommutative geometry. This

249: approach, which is already implicit in the $A_\infty$ constraints of

250: \cite{HLL}, leads us to consider the boundary deformation potential as

251: a function on a {\em noncommutative} space, and allows for a synthetic

252: formulation of the boundary WDVV equations. Moreover, it leads

253: naturally to a construction of boundary moduli spaces of topological

254: D-branes as noncommutative superspaces.

255: 

256: Let us explain this for the simple example of a single topological

257: D-brane.  In this situation, one could try the following naive proposal

258: for the generating function: \be \nn

259: W_{naive}=\sum_{n}{\frac{1}{n}W_{a_1\ldots a_n}\sigma^{a_1}\ldots

260: \sigma^{a_n}}~~, \ee where one views the boundary deformation

261: parameters as (super)commuting variables $\sigma^a$.  However,

262: supercommutativity of the parameters implies that $W_{naive}$ reduces

263: to: \be\nn \sum_{n}{\frac{1}{n}W_{(a_1\ldots a_n)}\sigma^{a_1}\ldots

264: \sigma^{a_n}} \ee where $W_{(a_1\ldots a_n)}$ is the (super)

265: symmetrization of the amplitudes.  Thus a generating function based on

266: (super) commuting deformation parameters does not faithfully encode

267: the topological tree-level data of the worldsheet theory, and cannot

268: generally be used to reconstruct the latter.

269: 

270: It was suggested in \cite{HLL} (see also \cite{Kajiura}) that this

271: problem might be overcome by viewing the boundary deformation

272: parameters as {\em non-commuting}.  While this might seem unusual at

273: first sight, it is in fact quite natural if one recalls that any

274: boundary theory admits Chan-Paton extensions, whose effect is to

275: promote the deformation parameters to (super)matrices $X^a$. As a

276: result, some of the information lost by $W_{naive}$ is preserved by

277: the matrix potential: \be \nn {\hat

278: W}=\sum_{n}{\frac{1}{n}W_{a_1\ldots a_n}{\rm str}(X^{a_1}\ldots

279: X^{a_n})}~~.  \ee However, supertraces of matrix monomials of finite

280: dimensions generally obey polynomial constraints. As a consequence,

281: the matrix potential of a fixed Chan-Paton extension can be reduced by

282: such relations, and again fails to faithfully encode the data of the

283: theory.  To completely resolve the issue, one has to remove all

284: constraints on $X^a$, which amounts to replacing them by free

285: (and in particular non-commuting) supervariables $s^a$. Hence one is lead to the {\em

286: non-commutative generating function} of \cite{HLL}:

287: \be

288: \nn

289: W=\sum_{n} \frac{1}{n}W_{a_1\ldots a_n}(s^{a_1}\ldots s^{a_n})_c~~,

290: \ee

291: where $(.)_c$ denotes the graded-cyclization operation, which gives an

292: abstract analogue of the supertrace. Notice that 

293: $W$ allows one to recover any Chan-Paton extension upon

294: replacing $s^a$ with supermatrices $X^a$, which amounts to considering

295: a finite-dimensional representation of the free associative

296: superalgebra $A=\C\langle\{s^a\} \rangle$.  In this way, one can study

297: at once {\em all} Chan Paton extensions of the theory, as well as more general representations 

298: obtained by taking morphisms from $A$ to an arbitrary associative superalgebra.

299: 

300: 

301: Remarkably, the procedure outlined above agrees with a key

302: principle of noncommutative algebraic geometry espoused in

303: \cite{Konts_formal} and developed further in

304: \cite{LB_lectures} (see \cite{Ginzburg_lectures} for an introduction). 

305: According to this ideology, `good' notions in

306: affine non-commutative algebraic geometry should induce the

307: corresponding classical notions on the moduli spaces of

308: finite-dimensional representations of the noncommutative coordinate

309: ring.  In the example above, the free superalgebra $A$ is the

310: coordinate ring of a noncommutative affine superspace $\A$, while

311: finite-dimensional representations of this algebra

312: (i.e. supermatrix-valued points of the `noncommutative scheme' $\A$) 

313: correspond to Chan-Paton extensions of the theory. By insisting

314: that the generating function should faithfully encode the information

315: of integrated amplitudes on the disk, we are lead to consider $W$ as

316: an element of the cyclic subspace $A_c$ of $A$

317: (namely the subspace of $A$ spanned by all graded-cyclic

318: monomials in the generators $s^a$). This matches the interpretation 

319: \cite{Konts_formal} of $A_c$ as the space of regular 

320: functions on $\A$.  Thus $W$ is a function on a 

321: non-commutative affine space, and we find that physics

322: reasoning agrees with the approach to non-commutative algebraic

323: geometry advocated in \cite{Konts_formal, LB_lectures}. Moreover, one

324: can show that the boundary topological metric induces an (even or odd)

325: noncommutative symplectic form on $\A$, which makes this affine superspace into 

326: a noncommutative symplectic supermanifold. As

327: in the supercommutative case, the symplectic structure

328: determines a bracket $\{.,.\}$ on $A_c$, and one finds that the homotopy associativity

329: constraints of \cite{HLL} are equivalent with the equation:

330: \be

331: \label{BV}

332: \{W,W\}=0~~.

333: \ee

334: Moreover, the unitality condition \cite{HLL, CIL5} on the underlying

335: $A_\infty$ algebra can also be formulated as

336: a constraint on $W$. This gives a

337: non-commutative geometric interpretation of the boundary part of the 

338: generalized WDVV equations.  As explained in \cite{HLL}, the

339: boundary topological metric of a general worldsheet theory can be even

340: or odd; as a consequence, $\{.,.\}$ is an even or odd Lie bracket. In

341: the latter case, the constraint (\ref{BV}) is a {\em non-commutative}

342: analogue of the classical master equation.

343: 

344: Non-commutativity of boundary observable insertions is also 

345: responsible for the fact that deformations of topological D-branes

346: are generally obstructed, which is reflected in

347: the typically singular nature of the boundary moduli space. An algebro-geometric approach to

348: boundary deformations was developed in relation to homological mirror

349: symmetry \cite{HMS} in \cite{Fukaya_mirror} and related to the deformation theory of open strings 

350: in \cite{CIL4} (see \cite{Polishchuk} for related work); these proposals

351: rely on constructing the moduli space as a commutative algebraic

352: or complex-analytic variety.  

353: 

354: 

355: The observations made above suggest that the deformation theory of

356: topological D-branes can be considered as a problem in {\em

357: noncommutative} geometry.  In particular, the possibility of

358: Chan-Paton extensions implies that the boundary moduli space can be

359: viewed as a non-commutative algebraic variety.  This is obtained by

360: `extremizing' the noncommutative function $W$ and modding out via

361: appropriate symmetries.  More precisely, one can impose the algebraic

362: relations:

363: \be

364: \label{crit}

365: \ld_a W=0~~,

366: \ee 

367: where $\ld_a$ is a $\Z_2$-graded version of the so-called {\em cyclic

368: derivatives} of \cite{RSS, Voiculescu} (see also \cite{Konts_formal}).

369: If $J$ is the two-sided ideal generated by $\ld_a W$, then the

370: quotient algebra $\C[{\cal Z}]:=\C[\{s^a\}]/J$ can be viewed as the

371: coordinate ring of a `noncommutative affine scheme' ${\cal Z}$ sitting

372: inside the affine superspace $\A$. Moreover, one can show that the symmetries of 

373: the system

374: form a subgroup ${\cal G}$ of the group of noncommutative symplectomorphisms

375: of $\A$.  These symmetries also preserve $J$, and thus descend to

376: automorphisms of $\C[{\cal Z}]$.  They can be viewed as gauge

377: transformations acting along the `noncommutative vacuum space' ${\cal Z}$.

378: One can thus define a {\em non-commutative extended moduli space}

379: ${\cal M}$ as the affine `noncommutative scheme' whose coordinate ring

380: $\C[{\cal M}]=\C[{\cal Z}]^{\cal G}$ is the ${\cal G}$-invariant part

381: of $\C[{\cal Z}]$.  The existence of a unit observable in the boundary

382: sector implies that one of the conditions (\ref{crit}) is a

383: non-commutative moment map constraint in the sense of \cite{BEV,

384: Bergh}. Therefore, the (invariant theory) quotient leading to ${\cal M}$ amounts to

385: modding out a zero-level 'symplectic reduction' of $A$ through the

386: ideal defined by the remaining relations.

387: 

388: It turns out that the construction outlined above can be carried out in much greater generality. 

389: In fact, as pointed out in \cite{CIL5, HLL}, the

390: homotopy-associativity constraints on disk boundary amplitudes 

391: generalize to {\em systems} of D-branes. In

392: this situation, boundary observables are either boundary-preserving or

393: boundary condition-changing, a decomposition which defines the

394: boundary sectors discussed in \cite{CIL1}.  The homotopy associativity constraints of \cite{HLL}

395: admit an obvious extension to this case, which can be formulated by

396: saying \cite{CIL5} that the D-brane system defines a (generally weak) cyclic and

397: unital $A_\infty$ category. The objects $u$ of this category are the

398: D-branes themselves, while the morphism spaces $\Hom(u,v)=E_{uv}$ are

399: the spaces of topological observables of strings stretching from $u$

400: to $v$.  In this case, the boundary deformation parameters $s^a$ are

401: replaced by $s^i_{uv}$, where $u,v$ run over the

402: topological D-branes, while

403: $i$ indexes a basis $\psi^i_{uv}$ of $E_{uv}$ (in fact, $\{s\}$ can

404: be viewed as a parity-changed dual basis to $\{\psi\}$). Treating

405: $s^i_{uv}$ as non-commuting supervariables leads one to replace the

406: free superalgebra $\C\langle \{s^a\}\rangle$ with the associative

407: superalgebra generated\footnote{Over the subalgebra spanned by the trivial paths.} 

408: by $s^i_{uv}$ with the relations:

409: \be

410: s^i_{uv}s^j_{v'w}=0~~{\rm unless~}v=v'~~.

411: \ee

412: This is the the path algebra $A_{\cal Q}$ of a quiver ${\cal Q}$

413: whose vertices are the D-branes $u$, and whose arrows from $u$ to $v$

414: are the index triples $(u,v,i)$ associated to $s^i_{uv}$. This quiver is $\Z_2$-graded since $s^i_{uv}$ can be

415: even or odd; as a consequence, $A_{\cal Q}$ is an

416: associative superalgebra.  The case of a single D-brane corresponds to

417: a quiver with a single vertex $u$, whose path algebra coincides with

418: the free superalgebra generated by the $\Z_2$-graded loops at that vertex.

419: 

420: It is known that path algebras of quivers are formally smooth in the

421: sense of \cite{CQ}, and in fact they provide good non-commutative

422: generalizations of the coordinate rings of smooth affine varieties

423: \cite{LB_lectures}.  One can formulate a non-commutative symplectic

424: geometry for such spaces \cite{Ginzburg, LB_quivers} by extending the

425: construction of \cite{Konts_formal}.  For a general topological D-brane system,

426: the space $\A_{\cal Q}$ with coordinate ring $A_{\cal Q}$ 

427: is a noncommutative {\em super}space, whose symplectic

428: form is induced by the parity change of the boundary topological

429: metrics.  As for affine superspaces, one finds an even or odd Lie

430: superbracket $\{.,.\}$ which acts on the space $C^0_R(A_{\cal Q})$ of

431: regular functions on $\A_{\cal Q}$.  The latter can be viewed as the

432: vector superspace spanned by necklaces of the quiver (i.e. cycles of

433: the quiver whose marking by the start=endpoint is forgotten). The

434: generating function $W$ of integrated disk boundary amplitudes is an

435: element of this space, i.e. a linear combination of such necklaces;

436: its parity is opposite that of the boundary topological metrics. The

437: categorical $A_\infty$ constraints for the entire collection of

438: boundary operators amount to equation (\ref{BV}). The existence of

439: unit observables in the boundary-preserving sectors can be expressed

440: as a differential constraint on $W$. Once again, a noncommutative

441: moduli space can be constructed as the invariant theory quotient of the

442: noncommutative critical variety ${\cal Z}$ through the symmetries of

443: the system.  The existence of boundary unit observables implies that

444: this quotient can be be viewed as a noncommutative 'symplectic

445: reduction', followed by imposing further constraints.

446: 

447: 

448: The paper is organized as follows. In Section 2, we describe the

449: algebraic structure of finite D-brane systems, starting from the

450: formulation in terms of $A_\infty$ categories found in \cite{HLL,

451: CIL5} (which is summarized in Appendix \ref{data}). Upon introducing a

452: finite-dimensional commutative semisimple algebra $R$, we show that

453: one can express the boundary sector decomposition of \cite{CIL1} as an

454: $R$-superbimodule structure on the total space

455: $E=\oplus_{u,v}{\Hom(u,v)}$ of boundary observables. As a consequence,

456: the cyclic and unital weak $A_\infty$ category determined by

457: integrated string correlators can be described as a cyclic and unital

458: weak $A_\infty$ algebra on the superbimodule $E$.  We also summarize

459: the strategy which will be followed in later sections in order to

460: encode this data into a noncommutative generating function. The main

461: step is passing from the superbimodule $E$ to the tensor algebra

462: $A=T_R (E[1]^{\rm v})$ of its parity-changed dual, which is a

463: superalgebra over $R$. The boundary topological metrics of the theory

464: induce an (even or odd) noncommutative symplectic form on $A$. This

465: leads us to study the noncommutative symplectic geometry of

466: $R$-superalgebras, a subject which is addressed in Section 3.  In

467: Section 4, we apply this to the tensor algebra $A$, discussing the

468: realization of the abstract objects introduced previously. Upon

469: picking appropriate bases in the space of boundary observables, we show that $A$

470: can be identified with the path algebra of a $\Z_2$-graded quiver, and

471: give coefficient expressions for the various quantities of

472: interest. We also discuss the quiver version of cyclic derivatives and

473: the so-called loop partial derivatives, two types of operators acting

474: on the space of noncommutative functions. In Section 5, we apply the

475: machinery developed in Sections 3 and 4 to finite D-brane

476: systems. Using the results of Section 2, we show that the cyclic and

477: unital weak $A_\infty$ structure on $E$ is encoded faithfully by a

478: noncommutative generating function $W$, which can be expressed as a

479: linear combination of quiver necklaces. After discussing the

480: constraints obeyed by $W$, we say a few words about deformations of

481: the underlying string theory (as opposed to deformations of the

482: boundary data). In Section 6, we discuss symmetries of the topological

483: D-brane system and give the algebraic construction of the

484: noncommutative moduli space. Section 7 gives the realization of our

485: construction in the case of a single D-brane. In Section 8,

486: we discuss the simplest examples with an even and odd boundary topological

487: metric, and a rather general class of examples with odd boundary metrics, some

488: particular cases of which appeared recently in \cite{Katz}. Section 9

489: contains our conclusions.

490: 

491: 

492: \paragraph{Conventions}

493: Unless specified otherwise, an algebra means an associative and unital algebra over the complex numbers. 

494: All morphisms of algebras are assumed to be unital. We will often encounter $\Z\times \Z_2$-graded algebras. To fix the

495: sign convention for such algebras, one must chose a $\Z_2$-valued pairing on the

496: Abelian group $\Gamma:=\Z\times \Z_2$, i.e.  a biadditive and

497: symmetric map $\cdot:\Gamma\times \Gamma\rightarrow \Z_2$ specifying

498: how the Koszul rule is applied to bigraded objects. Namely,  one agrees

499: that commuting two objects of bidegrees $(n,\alpha)$ and $(m,\beta)$

500: always produces the sign $(n,\alpha)\cdot (m,\beta)$.  In the present

501: paper, we work with the following choice of pairing:

502: \be

503: \label{pairing}

504: (n,\alpha)\cdot (m,\beta)=[mn]+\alpha\beta~~,

505: \ee

506: where here and in the rest of the paper the notation $[k]$ for $k\in

507: \Z$ stands for the ${\rm mod}~2$ reduction of $k$. If $\alpha$ is an

508: element of $\Z_2$, we let $(-1)^{\alpha}=+1$ if $\alpha=0$ and $-1$ if

509: $\alpha=1$. We let $[1]$ be the parity change functor on the category

510: of super-vector spaces and $\Sigma_U$ the suspension operator of

511: the supervector space $U$ (i.e. $\Sigma:U\rightarrow U[1]$ is the

512: identity operator of $U$, viewed as an odd map from $U$ to itself).

513: We have $\Sigma_U^2=\id_U$ and $[1]^2\approx {\rm Id}$, where ${\rm Id}$ in the

514: last relation is the identity functor. We will often write $\Sigma$

515: without indicating the vector space on which it acts, since the latter

516: is usually clear from the context. Given a ring $R$, we let ${\rm

517: Mod-R}$, ${\rm R-Mod}$ and ${\rm R-Mod-R}$ and  ${\rm

518: Smod-R}$, ${\rm R-Smod}$ and ${\rm R-Smod-R}$ 

519: be the categories of left-, right- and bi- modules, respectively supermodules over $R$.  

520: For a pair $U,V$ of

521: objects of any of these categories, we let $\Hom(U,V)$ be the space of

522: morphisms from $U$ to $V$ in that category. For supermodules, this consists of

523: degree zero $R$-linear maps and is an ordinary (i.e. ungraded) module.  For two supermodules,

524: we also let $\iHom(U,V)$ be the space of morphisms in the corresponding

525: category ${\rm Mod-R}$, ${\rm R-Mod}$ or ${\rm R-Mod-R}$ of {\em

526: ungraded} objects, obtained by forgetting the $\Z_2$-grading of

527: $U,V$. The latter consists of all linear maps, without degree

528: conditions; it is known as the {\em inner} morphism space and is

529: $\Z_2$-graded. In fact, we have $\iHom(U,V)=\iHom^0(U,V)\oplus

530: \iHom^1(U,V)$, where the degree zero component coincides with the

531: space of degree zero maps, $\iHom^0(U,V)=\Hom(U,V)$. We will use the same notational 

532: convention for the endomorphism and automorphism spaces, for example $\End(U)$ 

533: is the endomorphism space of $U$ as a supermodule and $\iEnd(U)$ its inner endomorphism 

534: space etc. Given an $R$-superbimodule $U$, 

535: we set $T_R U:=\oplus_{n\geq 0}{U^{\otimes_R n}}$ (with $U^{\otimes 0}:=R$), the tensor 

536: algebra of $U$, viewed as an $\N\times \Z_2$ graded algebra.

537: 

538: {\bf Note} Throughout this paper, an $A_\infty$ algebra or category will mean 

539: an $A_\infty$ algebra/category whose sequence of defining products $(r_n)$ terminates 

540: (i.e. $r_n=0$ for sufficiently large $n$).  This condition is purely technical, 

541: being needed if one wishes to pass 

542: to the dual of a coalgebra in naive manner. The condition can be 

543: removed in standard fashion, by considering profinite modules and formal algebras and 

544: replacing the relevant maps by continuous maps; then the non-commutative geometric 

545: objects mentioned above can  be understood in the formal sense. Because most of our 

546: considerations generalize straightforwardly to formal case, we adopted the convention of 

547: sometimes writing finite sums without indicating the upper bound; in this paper, it is 

548: understood that all such sums terminate. When we write a finite sum this manner, it is 

549: implied that the sum can be extended to a series in the formal theory.

550: 

551: 

552: \section{Algebraic description of finite topological D-brane systems} 

553: \label{algebraic}

554: 

555: In this section, we consider finite D-brane systems in an open

556: topological string theory, i.e.  finite collections of D-branes,

557: together with the spaces of boundary observables of all topological

558: strings stretched between them. We will assume that the total space of

559: zero-form boundary observables is finite-dimensional, which is the usual case

560: for topological strings. Using a description derived in  \cite{CIL5, HLL}, we 

561: will encode the information of all tree-level boundary string amplitudes into 

562: an algebraic structure on a certain superbimodule built out of the total space 

563: of zero-form boundary observables. 

564: 

565: 

566: The results of  \cite{CIL5, HLL} imply that D-brane systems in topological string theory are

567: described by $A_\infty$ categories.  The precise statement is 

568: as follows (see Appendix \ref{data} for mathematical background):

569: 

570: \

571: 

572: {\em A topological D-brane system is described by a weak, cyclic and

573: unital $A_\infty$ category ${\cal A}$.}

574: 

575: \

576: 

577: \noindent 

578: 

579: In general, the $A_\infty$ category is only $\Z_2$-graded. This

580: grading can be lifted to a $\Z$-grading provided that the relevant

581: $U(1)$ symmetry of the worldsheet theory is non-anomalous.  The

582: objects of ${\cal A}$ are the D-branes themselves, while the morphism

583: space $ \Hom_{\cal A}(u,v)$ for two objects $u,v$ is the complex

584: supervector space of boundary zero-form observables for the

585: topological string stretching from $u$ to $v$. The degree of such

586: observables is given by worldsheet Grassmann parity, and will be

587: denoted by $|.|$. With respect to this grading, the $A_\infty$

588: products have degrees $[n]$, where the square bracket indicates mod 2

589: reduction.  As in Appendix \ref{data}, it is convenient to work with

590: the parity-changed spaces $\Hom_{\cal A}(u,v)[1]$, the degree of whose

591: elements we denote by a tilde. Hence ${\tilde x}=[|x|+1]$ for all

592: homogeneous morphisms $x$.  The unitality condition involves even

593: elements $1_u$ of the endomorphism spaces $\Hom_{\cal A}(u,u)$

594: (equivalently, their odd suspensions $\lambda_u:=\Sigma 1_u\in \Hom_{\cal

595: A}(u,u)[1]$), while the cyclicity constraint involves non-degenerate

596: $\Z_2$-homogeneous pairings (the boundary topological metrics)

597: $\rho_{uv}:\Hom_{\cal A}(u,v)\times \Hom_{\cal A}(v,u)\rightarrow \C$

598: of common degree ${\tilde \omega}\in \Z_2$, obeying the

599: graded-symmetry condition $\rho_{uv}(x,y)=(-1)^{|x||y|}\rho_{vu}(y,x)$

600: for homogeneous elements $x\in \Hom_{\cal A}(u,v)$ and $y\in

601: \Hom_{\cal A}(v,u)$. The latter can also be expressed in

602: terms of the graded-antisymmetric forms $\omega_{uv}:=\rho_{uv}\circ

603: \Sigma^{\otimes 2}:\Hom_{\cal A}(u,v)[1]\times \Hom_{\cal

604: A}(v,u)[1]\rightarrow \C$ obtained from the topological metrics via

605: suspension. The detailed formulation of this structure (which arises

606: by introducing boundary sectors in the derivation of \cite{HLL}) is

607: given in Appendix \ref{data}.

608: 

609: 

610: 

611: \paragraph{Observation}

612: 

613: It was argued in \cite{CIL5} (see \cite{CIL2, CIL3} for the dG case) that the $A_\infty$ category

614: ${\cal A}_{full}$ obtained by considering `all' D-branes of a

615: topological string theory must be endowed with a parity-change functor

616: and be equivalent with its category of twisted complexes. This encodes

617: the physical requirement that the collection of all topological

618: D-branes of a given theory is closed under formation of topological

619: D-brane composites.  This `quasiunitarity constraint' implies that the

620: cohomology category $H^0({\cal A}_{full})$ is an enhanced triangulated

621: category (with an $A_\infty$ enhancement).  In the present paper, we

622: work with a fixed $A_\infty$ sub-category ${\cal A}$ of the full

623: D-brane category ${\cal A}_{full}$, so ${\cal A}$ need not obey this

624: supplementary condition (which is impossible to satisfy with a finite collection of objects).

625: 

626: \

627: 

628: For the remainder of this paper, we focus on finite D-brane systems,

629: which means that $Ob{\cal A}$ is a finite set and all morphism spaces

630: $\Hom(u,v)$ are finite-dimensional. In this case, we say that the $A_\infty$ category ${\cal A}$ is {\em finite}. 

631: 

632: 

633: \subsection{Encoding the boundary sector decomposition}

634: 

635: One can encode the categorical data of ${\cal A}$ 

636: in an equivalent, but more amenable form. Setting ${\cal Q}_0:=Ob{\cal A}$, 

637: consider the $\Z_2$-graded vector space 

638: $E:=\oplus_{u,v\in {\cal Q}_0}{\Hom(u,v)}$, which is known as 

639: the total boundary space of the D-brane system \cite{CIL1}.

640: 

641: 

642: \

643: 

644: \noindent {\bf Definition}

645: A {\em binary homogeneous decomposition} over the set ${\cal Q}_0$ is

646: a pair $(U, (U_{uv})_{u,v\in {\cal Q}_0})$ such that $U, U_{uv}$ are

647: finite-dimensional complex supervector spaces and $U=\oplus_{(u,v)\in {\cal Q}_0\times {\cal

648: Q}_0}{U_{uv}}$ is a homogeneous decomposition of $U$ indexed by the

649: Cartesian product ${\cal Q}_0\times {\cal Q}_0$. The {\em opposite} of the binary decomposition 

650: $(U, (U_{uv})_{(u,v)\in {\cal Q}_0\times {\cal Q}_0})$ is the binary homogeneous decomposition 

651: $(U, (U^{opp}_{uv})_{(u,v)\in {\cal Q}_0\times{\cal Q}_0})$, where $U^{opp}_{uv}=U_{vu}$.

652: 

653: \

654: 

655: A morphism of binary homogeneous decompositions over ${\cal Q}_0$ from

656: $(U, (U_{uv})_{(u,v)\in {\cal Q}_0\times {\cal Q}_0})$ to 

657: $(U',(U'_{uv})_{(u,v)\in {\cal Q}_0\times {\cal Q}_0})$

658: is a degree zero linear map $\phi:U\rightarrow U'$ such that

659: $\phi(U_{uv})\subset U'_{uv}$ for all $(u,v)\in {\cal Q}_0\times {\cal Q}_0$. 

660: With this definition, binary

661: homogeneous decompositions over ${\cal Q}_0$ form a category. A morphism

662: $\phi$ in this category is an isomorphism iff it is a bijective

663: map. In this case, we have $\phi(U_{uv})=U'_{uv}$ for all $u,v\in

664: {\cal Q}_0$.

665: 

666: 

667: A topological D-brane system determines two opposite decompositions of its total boundary space, 

668: namely $E=\oplus_{(u,v)\in {\cal Q}_0\times {\cal Q}_0}{\Hom_{\cal A}(u,v)}$ and 

669: $E=\oplus_{(u,v)\in {\cal Q}_0\times {\cal Q}_0}{\Hom_{\cal A}(v,u)}$. We will view 

670: the first of these  as fundamental, so we set $E_{uv}:=\Hom_{\cal A}(u,v)$; then 

671: $E_{uv}^{opp}:=\Hom_{\cal A}(v,u)$. This is consistent with the convention that morphisms 

672: compose forward in the definition of an $A_\infty$ category given in Appendix \ref{data}.  

673: The binary homogeneous decomposition $(E, (E_{uv})_{(u,v)\in {\cal Q}_0\times {\cal Q}_0})$

674: is  the so-called {\em boundary sector decomposition} of the topological D-brane system \cite{CIL1}. 

675: 

676: Let us consider the finite-dimensional semisimple commutative algebra $R=\oplus_{u\in {\cal Q}_0}{\C}$,

677: where the right hand side is a direct sum of copies of $\C$, viewed as

678: an algebra over itself.  We let $\epsilon_u$ $(u\in {\cal Q}_0$) be

679: the commuting idempotents of $R$ corresponding to the canonical basis

680: elements of the vector space $\C^{{\cal Q}_0}$, so $R=\oplus_{u\in

681: {\cal Q}_0}{\C \epsilon_u}$ with

682: $\epsilon_u\epsilon_v=\delta_{uv}\epsilon_u$.  The unit of $R$ is

683: $1_R=\sum_{u\in {\cal Q}_0}{\epsilon_u}$. It is not hard to

684: see that a binary

685: homogeneous decomposition $U=\oplus_{(u,v)\in {\cal Q}_0\times {\cal Q}_0 }{U_{uv}}$

686: amounts to giving an $R$-superbimodule structure on $U$. 

687: Indeed, such a decomposition amounts to giving degree

688: zero commuting idempotents $\epsilon_u^l, \epsilon_u^r\in \End_\C(U)$

689: for all $u\in {\cal Q}_0$, namely the projectors on the subspaces

690: $\oplus_{v}{U_{uv}}$ and $\oplus_{v}{U_{vu}}$ respectively (then

691: $U_{uv}=\epsilon_u^l\epsilon_v^r(U)=\epsilon_v^r\epsilon_u^l(U)$). Notice

692: that $U_{uv}$ can be recovered as $U_{uv}=\epsilon_uU\epsilon_v$ from

693: knowledge of the $R$-superbimodule structure.  

694: Moreover, a

695: morphism of binary homogeneous decompositions over ${\cal Q}_0$

696: amounts to a morphism of $R$-superbimodules.  Hence the category of

697: binary homogeneous decompositions over ${\cal Q}_0$ is equivalent with

698: the category of superbimodules over $R$.

699: 

700: Since $R$ is commutative, any $R$-superbimodule $U$ defines another $R$-superbimodule $U^{opp}$ 

701: whose underlying supervector space coincides with that of $U$ but whose external multiplications 

702: are given by:

703: \be

704: \nn

705: \alpha*x*\beta=\beta x\alpha~~~~~~\forall \alpha,\beta\in R~~\forall x\in U~~.

706: \ee

707: The relations $\epsilon_u*E*\epsilon_v=\epsilon_v E \epsilon_u$ show that the binary decomposition 

708: determined by $U^{opp}$ is the opposite of the binary decomposition determined by $U$:

709: \be

710: \nn

711: (U^{opp})_{uv}=U_{uv}^{opp}=U_{vu}~~.

712: \ee

713: 

714: 

715: Applying this to a topological D-brane system, we find that the boundary 

716: sector decomposition $E=\oplus_{(u,v)\in {\cal Q}_0\times {\cal Q}_0}{E_{uv}}=

717: \oplus_{(u,v)\in {\cal Q}_0\times {\cal Q}_0}{\Hom_{\cal A}{(u,v)}}$ 

718: is encoded by an $R$-superbimodule structure on $E$. Moreover, the opposite decomposition 

719: $E=\oplus_{(u,v)\in {\cal Q}_0\times {\cal Q}_0}{\Hom_{\cal A}{(v,u)}}$ is encoded by the opposite 

720: superbimodule $E^{opp}$. These observations allow one to encode the category-theoretic

721: structure of $A_\infty$ products into compatibility with the

722: $R$-superbimodule structure of $E$. Before stating the relevant

723: result, we need a few more preparations.

724: 

725: 

726: Given an $R$-superbimodule $U$, its dual $U^{\rm v}:=\Hom_{\rm R-mod} (U,R)=\iHom_{\rm R-Smod}(U,R)$ 

727: as a left $R$-module becomes an $R$-superbimodule 

728: with respect to the external multiplications defined through:

729: \be

730: \label{dual_multiplications}

731: (\alpha f \beta)(x):=f(\alpha x \beta)=\alpha f(x\beta) ~~.

732: \ee

733: We warn the reader that the usual definition of an $R$-superbimodule structure on $U^{\rm v}$ corresponds 

734: to the opposite of that given in (\ref{dual_multiplications}). We adopted the convention 

735: (\ref{dual_multiplications}) in order to avoid notational morass later on. With this definition, some of 

736: the usual isomorphisms involve taking the opposite of certain superbimodules, as we explain in Appendix 

737: \ref{isomorphisms}; in return, the formulas of Section 4,5 and 6 simplify considerably.  

738: 

739: 

740: It is not hard to see that the binary homogeneous decomposition

741: $U^{\rm v}=\oplus_{(u,v)\in {\cal Q}_0\times {\cal Q}_0}{(U^{\rm v})_{uv}}$ determined by

742: the $R$-superbimodule structure (\ref{dual_multiplications})  has components: 

743: \be

744: \label{dual_spaces}

745: (U^{\rm v})_{uv}=(U_{uv})^*~~,

746: \ee

747: where $(U_{uv})^*:=\iHom_\C(U_{uv},\C)$ is the linear dual of $U_{uv}$ viewed as a supervector space.

748: Notice that there is no reversal of the positions of $u$ and $v$ in relation (\ref{dual_spaces}).

749: 

750: 

751: With the definition (\ref{dual_multiplications}), a homogeneous

752: $R$-bilinear form\footnote{Recall that a multilinear map $f$ of $R$-superbimodules is required to satisfy

753: $f(\alpha x_1, x_2\ldots x_{n-1}, x_n\beta)=\alpha f(x_1\ldots

754: x_n)\beta$ as well as the balance condition $f(x_1\ldots

755: x_{j-1}\alpha, x_j\ldots x_n)=f(x_1\ldots x_{j-1}, \alpha x_j\ldots

756: x_n)$ for all $j$, where $\alpha,\beta\in R$. 

757: In particular, a bilinear map $\sigma:U\times U\rightarrow R$ 

758: satisfies $\sigma (x\alpha,y)=\sigma (x,\alpha y)$ and $\sigma (\alpha x, y\beta)=\alpha\sigma(x,y)\beta$.}

759:  $\sigma:U\times U\rightarrow R$ of degree ${\tilde \sigma}$ induces

760: an $R$-superbimodule map $U^{opp}\stackrel{j_\sigma}{\rightarrow}U[{\tilde \sigma}]^{\rm v}$ given

761: by $x\rightarrow f_x(.):=\sigma(\cdot, x)$.  

762: The form is

763: called {\em graded-symmetric} if

764: $\sigma(x,y)=(-1)^{\deg x \deg y}\sigma(y,x)$ for all homogeneous $x,y\in U$

765: and {\em graded-antisymmetric} if

766: $\sigma(x,y)=(-1)^{1+\deg x \deg y}\sigma(y,x)$.  In any of these case, it is called {\em

767: nondegenerate} if the map $x\rightarrow f_x$ is an isomorphism of vector spaces. 

768: 

769: 

770: A graded-symmetric form $\rho$ on

771: $U$ induces a graded-antisymmetric form $\omega=\rho\circ

772: \Sigma^{\otimes 2}$ on $U[1]$, given explicitly by:

773: \be

774: \nn

775: \omega(\Sigma x,\Sigma y)=(-1)^{\tilde x}\rho(x,y)~~~~~~\forall x,y \in U~~,

776: \ee

777: where the sign prefactor is due to the Koszul rule (the

778: suspension operator $\Sigma$ is odd). If $\rho$ is homogeneous, then

779: $\omega$ is homogeneous of the same $\Z_2$ degree. Hence giving a

780: graded-symmetric form on $U$ is equivalent to giving a graded-antisymmetric 

781: form on $U[1]$. Moreover, $\rho$ is non-degenerate iff

782: $\omega$ is.  A non-degenerate, homogeneous and graded-symmetric

783: $R$-bilinear form will be called a {\em metric}, while a

784: non-degenerate, homogeneous and graded-antisymmetric R-bilinear form

785: will be called a {\em symplectic form}.  A $R$-superbimodule is called

786: a {\em metric superbimodule} if it is endowed with a metric, and a {\em

787: symplectic superbimodule} if it is endowed with a symplectic form.  Metric

788: and symplectic $R$-superbimodules form categories if one defines

789: morphisms in the obvious fashion.  Notice that parity change induces

790: an idempotent equivalence between these categories. This reflects the

791: general principle that metric and symplectic superdata are related

792: through parity change.  In particular, $(U[1],\omega)$ is a symplectic

793: R-superbimodule iff $(U,\rho)$ (with $\omega=\rho\circ \Sigma^{\otimes

794: 2}$) is a metric R-superbimodule.

795: 

796: Giving an $R$-bilinear form $\sigma$ on $U$ amounts to giving $\C$-bilinear forms 

797: $\sigma_{uv}:U_{uv}\times U_{vu}\rightarrow \C$ for all $u,v$. Indeed, $R$-bilinearity 

798: of $\sigma$ implies:

799: \be

800: \nn

801: \sigma(x\epsilon_v, y)=\sigma(x, \epsilon_vy)~~{\rm and}~~\sigma(\epsilon_u x, y\epsilon_w)=

802: \sigma(x,y)\epsilon_u\epsilon_w~~.

803: \ee

804: Writing $x=\sum_{u,v}{x_{uv}}$ and $y=\sum_{u,v}{y_{uv}}$ with $x_{uv}:=\epsilon_u x \epsilon_v\in U_{uv}$ and 

805: $y_{uv}:=\epsilon_u y\epsilon_v\in U_{uv}$, this shows that $\sigma(x_{uv}, y_{u'v'})$ vanishes unless $v=u'$ and 

806: $u=v'$. Hence $\sigma$ is completely determined by its restrictions $\sigma'_{uv}$ to the subspaces 

807: $U_{uv}\times U_{vu}$. Explicitly, we have $\sigma(x,y)=\sum_{u,v}\sigma'_{uv}(x_{uv},y_{vu})$. 

808: Each restriction takes $U_{uv}\times U_{vu}$ into the one-dimensional subspace $\C \epsilon_u$ of $R$, so 

809: we can write $\sigma'_{uv}=\sigma_{uv}\epsilon_u$ for some complex-linear maps 

810: $\sigma_{uv}:U_{uv}\times U_{vu}\rightarrow \C$. Then: 

811: \be

812: \label{sigma_dec}

813: \sigma(x,y)=\sum_{u,v}\sigma_{uv}(x_{uv},y_{vu})\epsilon_u

814: \ee

815: and $\sigma$ is completely determined by $\sigma_{uv}$. Conversely,

816: any family of complex-bilinear maps $\sigma_{uv}:U_{uv}\times

817: U_{vu}\rightarrow \C$ determines an $R$-bilinear map $\sigma$ through

818: relation (\ref{sigma_dec}).  The map $U\stackrel{j_\sigma}{\rightarrow} U^{\rm v}$ induced by $\sigma$ 

819: has the property $j_\sigma(U^{\rm opp}_{uv})\subset (U^{\rm v})_{uv}$ i.e. 

820: $j_\sigma(U_{vu})\subset (U_{uv})^*$, and its restrictions to the subspaces $U_{vu}$ can be identified 

821: with the maps $U_{vu}\rightarrow (U_{uv})^*$ induced by $\sigma_{uv}$. 

822: Thus $\sigma$ is non-degenerate iff $\sigma_{uv}$ are (the later

823: means that all maps $U_{vu}\rightarrow (U_{uv})^*$ determined by

824: $\sigma_{uv}$ are linear isomorphisms). It is also clear that $\sigma$ is

825: graded-symmetric iff $\sigma_{uv}(x,y)=(-1)^{{\tilde x}{\tilde

826: y}}\sigma_{uv}(y,x)$ for all homogeneous $x\in U_{uv}$ and $y\in

827: U_{vu}$; a similar statement holds for graded-antisymmetric forms.

828: 

829: 

830: For a topological D-brane system, we have $E_{uv}=\Hom_{\cal

831: A}(u,v)$ and the topological metrics $\rho_{uv}:\Hom_{\cal

832: A}(u,v)\times \Hom_{\cal A}(v,u)\rightarrow \C$ are homogeneous of common

833: degree ${\tilde \omega}$, so they determine an $R$-bilinear form

834: $\rho$ on $E$ of the same degree.  Moreover, the graded-symmetry and

835: non-degeneracy conditions for $\rho_{uv}$ amount to the condition that

836: $\rho$ is a superbimodule metric on $E$.  Similarly, the symplectic

837: forms $\omega_{uv}=\rho_{uv}\circ \Sigma^{\otimes 2}$ determine a

838: superbimodule symplectic form on $E[1]$ of degree ${\tilde \omega}$.

839: 

840: 

841: A {\em weak $A_\infty$ structure} on a $R$-superbimodule $U$ is a

842: countable family of odd $R$-linear maps $r_n:U[1]^{\otimes

843: n}\rightarrow U[1]$ (equivalently, odd $R$-multilinear maps

844: $r_n:U[1]^n\rightarrow U[1]$, which we denote by the same letters)

845: with $n\geq 0$, which satisfy the conditions\footnote{There is

846: some ambiguity in the sign conventions for various objects

847: related to $A_\infty$ algebras and categories. In this paper, we view the homological

848: derivation $Q$ discussed in Section \ref{geometrization} as the

849: fundamental object, and have defined $r_n$ such that most signs

850: simplify. We refer the reader to \cite{Penkava} for a discussion of

851: other conventions.}:

852: \be

853: \label{ainf}

854: \sum_{0\leq i+j\leq n} (-1)^{{\tilde x}_1+\ldots +{\tilde x}_i} r_{n-j+1}(x_1\ldots x_i, r_j(x_{i+1}

855: \ldots x_{i+j}), x_{i+j+1}\ldots x_n)=0~~

856: \ee

857: for all $n\geq 0$. In these relations, it is understood that $x_1,

858: \ldots, x_n$ are arbitrary homogeneous elements of $U[1]$.  The

859: structure is called {\em strong}, if $r_0=0$ and {\em minimal} if

860: $r_0=r_1=0$.

861: 

862: We say that a weak $A_\infty$ structure on $U$ is {\em

863: unital} if there exists an even element $1\in U^R$ such

864: that its odd suspension $\lambda =\Sigma 1\in U[1]^R$ satisfies the

865: following conditions:

866: \begin{eqnarray}

867: \label{unitality}

868: r_n(x_1\ldots x_{j-1},\lambda,x_{j+1}\ldots x_n)&=&0~~{\rm for~all}~~~~ n\neq 2 ~{\rm~and~all}~ j~~\nn\\

869: -r_2(\lambda,x)=(-1)^{{\tilde x}} r_2(x,\lambda)&=&x~~,

870: \end{eqnarray}

871: for all homogeneous elements $x,x_j$ of $U$.  In this case, $1$ is

872: called the {\em even unit} of the $A_\infty$ structure, while

873: $\lambda$ will be called the {\em odd unit}. It is not hard to

874: see that the unit of a unital $A_\infty$

875: structure is unique. Indeed, given another unit $1'$, set $\lambda':=\Sigma

876: 1'$. Then $r_2(\lambda,\lambda')=-\lambda=-\lambda'$, where we used

877: the second row in (\ref{unitality}) by viewing either $\lambda$ or

878: $\lambda'$ as the unit. This implies $1=1'$. 

879: 

880: Recall that the {\em center} $U^R$ of an $R$-superbimodule $U$ 

881: (a.k.a the centralizer of $R$ in $U$) is the 

882: homogeneous sub-bimodule consisting of all central elements $x\in U$, i.e. those

883: elements which satisfy $\alpha x=x\alpha$ for all $\alpha\in R$.  In

884: terms of the homogeneous binary decomposition, we have

885: $U^R=\oplus_{u\in {\cal Q}_0}{U_{uu}}$.  Thus a central element has

886: the form $x=\oplus_{u\in {\cal Q}_0}{x_u}$, with

887: $x_u=\epsilon_ux\epsilon_u=\epsilon_u x=x\epsilon_u\in U_{uu}$. Since

888: $R$ is semisimple, we have a direct sum decomposition of vector spaces:

889: \be

890: \label{CQ_decomp}

891: U=U^R\oplus [R,U]~~,

892: \ee

893: where $[R,U]$ is the complex supervector space generated by the

894: commutators $[\alpha,x]=\alpha x-x\alpha$ with $\alpha\in R$ and $x\in

895: U$. This follows as in \cite{CQ} due to the fact that $\oplus_{u\in

896: Q_0}{\epsilon_u\otimes \epsilon_u}$ is a so-called separability

897: element. It can also be seen directly by using the identities

898: $x-\sum_{u}{\epsilon_u

899: x\epsilon_u}=\frac{1}{2}\sum_{u}{[\epsilon_u,[\epsilon_u, x]]}$ and

900: $\epsilon_u[\epsilon_v, x]\epsilon_u=0$, which

901: hold for any element $x\in U$.

902: 

903: Semisimplicity of $R$ implies that the unit of a unital $A_\infty$ structure 

904: on an $R$-superbimodule $U$ must be central. To see this, let $\lambda$ be the odd 

905: $A_\infty$ unit and consider the central element $\lambda'=\sum_{u}{\epsilon_u \lambda \epsilon_u}$.

906: Then the last row in (\ref{unitality}) implies: 

907: \be

908: \nn

909: r_2(x,\lambda')=\sum_{u} r_2(x,\epsilon_u \lambda \epsilon_u)=

910: \sum_{u} r_2(x\epsilon_u, \lambda )\epsilon_u=(-1)^{\tilde x} \sum_u x\epsilon_u =(-1)^{\tilde x}x~~

911: \ee

912: and: 

913: \be

914: \nn

915: r_2(\lambda',x)=\sum_{u} r_2(\epsilon_u \lambda \epsilon_u, x)=

916: \sum_{u} \epsilon_u r_2(\lambda, \epsilon_u x)=- \sum_u \epsilon_u x=-x~~,

917: \ee

918: where we used $R$-bilinearity of $r_2$ and the equation $\sum_u \epsilon_u=1$. These two equations 

919: imply $\lambda=\lambda'$ by the argument used above to show unicity of the $A_\infty$ unit. 

920: This shows that $\lambda$ must be central. 

921: 

922: 

923: 

924: A weak $A_\infty$ structure $(r_n)$ on $U$ is

925: called {\em cyclic} if $U[1]$ is endowed with a homogeneous symplectic form

926: $\omega$, such that the following relations are satisfied:

927: \be

928: \label{rrcyc}

929: \omega(x_0,r_n(x_1\ldots x_n))=(-1)^{{{\tilde x}_0+{\tilde x}_1+\tilde x}_0

930: ({\tilde x}_1+\ldots +{\tilde x}_n)}\omega(x_1,r_n(x_2\ldots x_n, x_0))~~.

931: \ee

932: In terms of the metric $\rho$ determined by $\omega=\rho\circ \Sigma^{\otimes 2}$,

933: these relations take the following form, which might be more

934: familiar to some readers: \be

935: \label{rcyc}

936: \rho(x_0,r_n(x_1\ldots x_n))=(-1)^{{\tilde x}_0({\tilde x}_1+\ldots

937: +{\tilde x}_n)}\rho(x_1,r_n(x_2\ldots x_n, x_0))~~.  \ee

938: 

939: \

940: 

941: \noindent We can now state a basic equivalence: 

942: \paragraph{\bf Proposition}

943: Giving a finite weak cyclic and unital $A_\infty$ category with object

944: set ${\cal Q}_0$ amounts to giving a weak, cyclic and unital

945: $A_\infty$ structure on a $\Z_2$-graded $R$-superbimodule $E$ of finite

946: complex dimension, over the finite-dimensional semisimple commutative

947: algebra $R=\oplus_{u\in {\cal Q}_0}\C \epsilon_u$.

948: 

949: 

950: \

951: 

952: \noindent 

953: 

954: In view of this proposition, one can {\em define} finite tree-level

955: topological D-brane systems to be cyclic and unital weak $A_\infty$

956: structures on some $R$-superbimodule of finite dimension over $\C$, where $R$ is a 

957: finite-dimensional semisimple commutative $\C$-algebra. The decomposition of the

958: system into constituent D-branes and the decomposition of the total

959: boundary space $E$ into boundary sectors

960: $E_{uv}$ are both encoded by the $R$-superbimodule structure. 

961: 

962: 

963: \paragraph{Sketch of proof}

964: As explained above, the superbimodule structure on $E$ amounts to a

965: homogeneous decomposition $E=\oplus_{(u,v)\in {\cal Q}_0\times {\cal

966: Q}_0}{E_{uv}}$, where $E_{uv}:=\Hom_{\cal A}(u,v)$.  The rest of the

967: proof is a lengthy but straightforward check of conditions, showing

968: that compatibility of various maps with the $R$-superbimodule

969: structure of $E$ allows one to translate superbimodule $A_\infty$ data

970: into the categorical $A_\infty$ data listed in Appendix \ref{data}.

971: We already showed above that giving the categorical bilinear forms $\rho_{uv}$

972: amounts to giving a metric $\rho$ on this superbimodule.  Similarly,

973: giving an $R$-multilinear map $r_n:E^n\rightarrow E$ amounts to giving

974: $\C$-multilinear maps $r_{u_1\ldots u_{n+1}}: E_{u_1 u_2}\times E_{u_2

975: u_3}\times \ldots \times E_{u_n u_{n+1}}\rightarrow E_{u_1u_{n+1}}$,

976: and it is clear that the weak $A_\infty$ constraints for $r_n$ amount

977: to the categorical weak $A_\infty$ constraints for these maps.

978: Moreover, the cyclicity conditions for $r_n$ with respect to

979: $\omega=\rho\circ\Sigma^{\otimes 2}$ amount to the categorical

980: cyclicity constraints with respect to $\omega_{uv}:=\rho_{uv}\circ

981: \Sigma^{\otimes 2}$.  Finally, the $A_\infty$ units $1_u$ of ${\cal

982: A}$ give an even central element $1=\oplus_{u}{1_u}$ in the

983: superbimodule $E$, which is a unit for the superbimodule 

984: $A_\infty$ structure. Conversely, giving such a unit amounts

985: to giving elements $1_u$ in each `diagonal' subspace $E_{uu}$, since 

986: the unit of an $R$- superbimodule $A_\infty$ structure must be central. 

987: The unitality constraints for $r_n$ amount to the categorical unitality constraints

988: for $r_{u_1\ldots u_{n+1}}$.

989: 

990: \paragraph{Observation} When the $A_\infty$ structure $(r_n)$ is minimal, the first 

991: non-trivial $A_\infty$ constraint implies that the product

992: $\cdot=\Sigma\circ r_2\circ \Sigma^{\otimes 2}$ is associative.

993: Moreover, cyclicity and unitality imply that the triple $(E,\cdot

994: ,\rho)$ is a (non-commutative) Frobenius algebra, whose multiplication and 

995: bilinear form are $R$-bilinear. On the other hand, forgetting all higher products of ${\cal A}$ gives a usual

996: (i.e. associative and unital) category endowed with non-degenerate and

997: graded symmetric bilinear pairings $\rho_{uv}$ between opposite spaces

998: of morphisms. As explained in \cite{CIL1}, such a category describes

999: the boundary part of a two-dimensional topological field theory

1000: defined on bordered Riemann surfaces.  This gives the following:

1001: 

1002: 

1003: 

1004: \paragraph{\bf Corollary} The boundary sector of a topological field theory in

1005: two-dimensions, in the presence of a finite system of topological D-branes,  

1006: is described by a noncommutative Frobenius structure on

1007: an $R$-superbimodule $E$, i.e. a unital noncommutative

1008: Frobenius algebra on the vector superspace $E$, whose associative

1009: product and pairing are $R$-bilinear. 

1010: 

1011: \

1012: 

1013: \subsection{The geometrization strategy}

1014: 

1015: The algebraic formulation of the previous subsection 

1016: allows one to avoid the notational morass of the category-theoretic description. One

1017: is still left with the rather complicated data of a cyclic and unital

1018: weak $A_\infty$ structure on $E$.  To express this synthetically, we will 

1019:  use a $\Z_2$-graded version of the non-commutative symplectic

1020: geometry of quivers developed in \cite{Ginzburg, LB_quivers}. Much of the content of the next two sections is a

1021: relatively straightforward, though tedious, superextension of the construction 

1022: of \cite{Ginzburg, LB_quivers}, so it might be useful to

1023: summarize the main points. Start with a cyclic and unital $A_\infty$

1024: structure on the $R$-superbimodule $E$. To encode this geometrically,

1025: we will proceed as follows:

1026: 

1027: (1) Giving a weak

1028: $A_\infty$ structure on the superbimodule $E$ amounts to giving an odd

1029: derivation $Q$ of the tensor algebra $A=T_R E[1]^{\rm v}$, satisfying

1030: the condition $Q^2=0$. Hence $(A,Q)$ can be viewed as a noncommutative version of 

1031: the $Q$-manifolds considered in \cite{Konts_Schwarz}.

1032: 

1033: (2) The symplectic form on $E[1]$ induces a non-commutative symplectic

1034: form on $A$, and one can develop the non-commutative symplectic

1035: supergeometry of this algebra by extending the approach of

1036: \cite{Ginzburg, LB_quivers}. This gives notions of symplectic and

1037: Hamiltonian superderivations having the classical properties. The

1038: Karoubi complex $C_R(A)$ is acyclic in positive degrees so all symplectic

1039: superderivations are Hamiltonian. There is a $\Z_2$-graded version

1040: $\{.,.\}$ of the Kontsevich bracket, a super-Lie bracket on

1041: $C^0_R(A)[{\tilde \omega}]$, where ${\tilde \omega}$ is the

1042: $\Z_2$-degree of the symplectic form.

1043: 

1044: (3) Cyclicity of $(r_n)$ amounts to the condition that $Q$ be a

1045: symplectic derivation, i.e. $L_Q\omega=0$, where $L_Q$ is the Lie

1046: superderivative along $Q$. Thus $(A,Q,\omega)$ is a noncommutative 

1047: version of the QP-manifolds of \cite{Konts_Schwarz}. 

1048: 

1049: 

1050: (4) The non-commutative generating function $W$ of the D-brane system

1051: is the Hamiltonian of the homological derivation $Q$; to determine this uniquely, we 

1052: require that it `vanishes at zero' in an appropriate sense.   Thus $W$ is an

1053: element of degree ${\tilde \omega}+1$ of the supervector space

1054: $C^0_R(A)=A/[A,A]$. The weak $A_\infty$ constraint $Q^2=0\Leftrightarrow

1055: [Q,Q]=0$ is equivalent with the condition $\{W,W\}=0$. The superderivation $Q$ can be

1056: reconstructed as the Hamiltonian derivation determined by $W$, so the $A_\infty$

1057: structure defined by $W$ is automatically cyclic.

1058: 

1059: 

1060: 

1061: (6) Unitality of the weak $A_\infty$ structure is equivalent to the

1062: condition $\frac{1}{2}\ld_\lambda W=\mu$, where $\mu$ is a superized version of

1063: the moment map of \cite{BEV} and $\ld_\lambda$ is the cyclic

1064: derivative of $W$ with respect to the odd $A_\infty$ unit

1065: $\lambda$. The noncommutative generating function determines the

1066: symplectic form through this relation.

1067: 

1068: 

1069: (7) Since $R=\oplus_{u\in {\cal Q}_0}{\C \epsilon_u}$, the tensor algebra $A$ can be

1070: viewed as the path algebra of a superquiver ${\cal Q}$, a quiver on the

1071: vertex set ${\cal Q}_0$ endowed with a $\Z_2$-valued map on its set of

1072: arrows. The quiver presentation arises by choosing homogeneous bases

1073: of the vector space $E$ which are adapted to its binary homogeneous

1074: decomposition. Using quiver language amounts to working in `special'

1075: coordinates on the non-commutative space determined by $A$, where

1076: `special' means that the coefficients of the non-commutative

1077: symplectic form in such coordinates are complex numbers.

1078: 

1079: (8) All constructions have natural quiver interpretations. For

1080: example, $C^0_R(A)$ can be described in terms of necklaces, which in our

1081: case are $\Z_2$-graded. Cyclic derivatives with respect to elements of

1082: the adapted basis translate into cyclic derivatives with respect to

1083: the quiver's arrows $a$.  Relative to an appropriate adapted basis,

1084: the unitality constraint amounts to the requirement that $W$ has a certain

1085: dependence on an odd element $\sigma$ of $E[1]^{\rm v}$ determined by

1086: the odd $A_\infty$ unit $\lambda$.

1087: 

1088: (9) One can view $A$ as the coordinate ring of a non-commutative

1089: supermanifold $\A_{\cal Q}$. Imposing the relations $\ld_a W=0$ gives the

1090: non-commutative extended vacuum space ${\cal Z}$, a `noncommutative subscheme'

1091: of $\A_{\cal Q}$. The non-commutative extended moduli space ${\cal M}$ is obtained by modding 

1092: ${\cal Z}$ (in the GIT sense) through those

1093: symplectomorphisms which correspond to unital and cyclic $A_\infty$ automorphisms 

1094: of the underlying D-brane category.  

1095: 

1096: We now proceed with the detailed discussion of these points.

1097: 

1098: \section{Noncommutative symplectic geometry of $R$-superalgebras}

1099: \label{sg}

1100: 

1101: In this section, we extend the construction of

1102: \cite{Ginzburg,LB_quivers} to the case of superalgebras.  The proofs

1103: of most statements are straightforward adaptations of those given in

1104: \cite{Ginzburg,LB_quivers}, so I will only indicate the points where

1105: our conventions are important or something interesting happens.

1106: 

1107: Let $R$ be a unital and commutative algebra over $\C$.  An {\em

1108: $R$-superalgebra} is a unital superalgebra $A$ containing $R$ as a

1109: subalgebra in even degrees (notice that this is stronger than 

1110: requiring that $A$ be a superalgebra over $R$, since 

1111: we require that $R$ sits inside the degree zero subalgebra of $A$).  

1112: A morphism of $R$-superalgebras is a morphism of

1113: superalgebras whose restriction to $R$ equals the identity map.

1114: 

1115: 

1116: \subsection{Noncommutative differential superforms}

1117: 

1118: Given an $R$-superalgebra $A$, consider the $R$-superbimodule

1119: $A_R:=A/R$, where $A/R$ stands for the vector space quotient. 

1120: We define the {\em space of relative noncommutative

1121: forms of $A$ over $R$} by $\Omega_R A:=\oplus_{n\geq 0}\Omega^n_R A$, where:

1122: \be

1123: \nn

1124: \Omega^n_R A= A\otimes_R T_R^n (A_R)=A\otimes_R A_R^{\otimes_R n}~~.

1125: \ee

1126: We write the elements of this space as $w=a_0da_1\ldots da_n$, where $da$ is the

1127: image of $a\in A$ under the projection $A\stackrel{d}{\rightarrow}

1128: A_R=A/R$ and juxtaposition stands for the tensor product over $R$.

1129: The space $\Omega_R A$ is given a differential algebra structure with product: 

1130: \begin{eqnarray}

1131: &&(a_0 da_1\ldots  da_n) (b_0 db_1 \ldots db_m):=a_0da_1\ldots da_n d(a_nb_0)db_1\ldots db_m\nn\\

1132: &+&\sum_{i=1}^{n-1}{(-1)^{i}a_0  

1133: da_1\ldots  d(a_{n-i}a_{n-i+1}) \ldots 

1134: da_n db_0}\ldots  db_m + (-1)^n a_0 a_1 da_2\ldots da_n db_0\ldots db_m\nn

1135: \end{eqnarray}

1136: and differential $d (a_0da_1\ldots d a_n)=da_0da_1\ldots da_n$. 

1137: In this paper, we view $\Omega_R A$ as an $\N\times \Z_2$ graded algebra,

1138: whose $\N$-grading (with components $\Omega^n_R A$) is given by the `rank' of forms, and whose 

1139: $\Z_2$ grading is induced from $A$. We denote the bidegree of homogeneous elements by $\deg w=({\bar

1140: w},{\tilde w})\in \N\times \Z_2$.  As explained in the introduction, we

1141: always work with the pairing (\ref{pairing}), and will require that $d$

1142: has bidegree $(1,0)\in \Z\times \Z_2$. This means that $d$ has the derivation property:

1143: \be

1144: \nn

1145: d(w_1w_2)=(dw_1)w_2+(-1)^{{\bar w_1}}w_1\cdot d w_2~~{\rm for~all~}~~w_1,w_2\in \Omega_R A~~.

1146: \ee

1147: We stress that in our conventions $d$ has degree {\em zero} with

1148: respect to the $\Z_2$-grading.  The detailed construction of $\Omega_R

1149: A$ is given in Appendix \ref{envelope}.  

1150: The space $\Omega^1_R A$ has an $A$-superbimodule structure with multiplications: 

1151: \be

1152: \nn

1153: \alpha(adb)\beta=(\alpha a)d(b\beta) -(\alpha a b)d\beta~~~~~~~~\forall a,b\in A,~~\alpha,\beta\in R~~.

1154: \ee

1155: As in \cite{CQ}, one has an isomorphism of bigraded 

1156: algebras $\Omega_R A\approx T_A (\Omega^1_R A)$, which induces an $A$-superbimodule 

1157: structure on $\Omega_R A$.  In particular, $\Omega_R A$ is

1158: an $R$-superalgebra (since $R\subset A=\Omega^0_R A \subset \Omega_R A$).  

1159: 

1160: 

1161: As usual, the pair $(\Omega_R A, d)$ has a universality property.  To

1162: formulate it, we define an $R$-{\em differential superagebra} to be

1163: an $\N\times \Z_2$-graded unital differential algebra $(\Omega,d)$

1164: such that $\deg(d)=(1,0)$, $R\subset \Omega^{0,0}$ and $d(R)=0$, where

1165: $\Omega^{0,0}$ is the subspace of elements of vanishing bidegree. If

1166: $(\Omega_j,d_j)$ are two such algebras, a map

1167: $\phi:\Omega_1\rightarrow \Omega_2$ is called a morphism of

1168: $R$-differential superalgebras if:

1169: 

1170: \noindent (1) $\phi$ is a morphism of unital $\N\times \Z_2$-graded

1171: algebras (in particular, $\phi$ has vanishing bidegree)

1172: 

1173: \noindent (2) $\phi|_R=\id_R $

1174: 

1175: \noindent (3) $d_2\circ \phi=\phi\circ d_1$~~.

1176: 

1177: \noindent With this definition, $R$-differential superalgebras form a

1178: category. The universality property of $(\Omega_R A, d)$ is as follows.  

1179: Given any $R$-differential superalgebra $(\Omega, d)$ and a morphism of $R$-superalgebras

1180: $\rho:A\rightarrow \Omega$ such that $\rho(A)\subset \Omega^{0}$, there exists a unique morphism

1181: $u:\Omega_R A\rightarrow \Omega$ of $R$-differential superalgebras such

1182: that $\rho=uj$, where $j:A\rightarrow \Omega_R A$ is the

1183: inclusion. Hence $(\Omega_R A,j)$ is an initial object among the pairs

1184: $(\Omega,\rho)$.  In fact, the correspondence $\Omega\rightarrow

1185: \{$ $R$-superalgebra morphisms $\rho:A\rightarrow

1186: \Omega$ with $\rho (A)\subset \Omega^{0}~ \}$ defines a

1187: functor from the category of $R$-differential superalgebras to

1188: the category of sets. The universality property means that $(\Omega_R

1189: A, d)$ represents this functor, so it is the

1190: superdifferential envelope of $A$. In particular, any morphism

1191: $\phi:A_1\rightarrow A_2$ of $R$-superalgebras extends uniquely to a

1192: morphism $\phi^*:\Omega_R A_1\rightarrow \Omega_R A_2$ of $R$-differential

1193: superalgebras.

1194: 

1195: 

1196: \subsection{Super Lie derivatives and contractions}

1197: 

1198: Recall that a left (right) derivation of $A$ is a derivation of $A$

1199: viewed as a left (right) supermodule over $A\otimes A^{op}$. Thus a homogeneous 

1200: left derivation $D$ satisfies $D(ab)=(Da)b+(-1)^{{\tilde a}{\tilde D}}a Db$, while 

1201: a homogeneous right derivation satisfies $(ab)D=a(bD)+(-1)^{{\tilde b}{\tilde D}}(aD) b$, where ${\tilde D}$ is the 

1202: degree of $D$ and we use the convention that left derivations are written to the left, and right derivations 

1203: are written to the right. We will sometimes also indicate this by writing arrows above $D$.

1204: 

1205: 

1206: A relative derivation of $A$ is a derivation which is $R$-linear,

1207: i.e. vanishes on the subalgebra $R$.  We let $\Der_l(A) $ and

1208: $\Der_r(A)$ be the complex supervector spaces of {\em relative} left

1209: and right derivations of $A$, viewed as Lie superalgebras with respect to the

1210: supercommutator $[D_1,D_2]=D_1\circ D_2-(-1)^{{\tilde D}_1 {\tilde

1211: D_2}}D_2\circ D_1$, which satisfies $[D_1,D_2]=(-1)^{1+{\tilde D}_1{\tilde D}_2}[D_2,D_1]$.  

1212: We let $\Der_{l,r}^\alpha (A)$ be the subspaces

1213: consisting of left and right relative derivations of degree $\alpha$.

1214: 

1215: Similarly, let $\Der_{l,r}(\Omega_R A)$ be the $\Z\times \Z_2$-graded

1216: complex vector spaces of {\em relative} left/right derivations of $\Omega_R A$,

1217: i.e. those left/right derivations of $\Omega_R A$ which vanish on $R$. In this

1218: definition, we view $\Omega_R A$ as a $\Z\times \Z_2$ graded algebra

1219: with the degree pairing (\ref{pairing}); thus a bihomogeneous left derivation $D$ of

1220: $\Omega_R A$ satisfies:

1221: \be

1222: \nn

1223: D(w_1w_1)=Dw_1 w_2+(-1)^{\deg D \cdot \deg w_1}w_1 Dw_2~~,

1224: \ee

1225: while a bihomogeneous right derivation obeys:

1226: \be

1227: \nn

1228: (w_1w_2)D=w_1 (w_2D)+(-1)^{\deg D\cdot \deg w_2}(w_1D)w_2~~.

1229: \ee

1230: The spaces $\Der_{l,r}(\Omega_R A)$ are bigraded Lie superalgebras

1231: with respect to the bigraded supercommutator $[D_1,D_2]=D_1\circ

1232: D_2-(-1)^{\deg D_1 \cdot \deg D_2} D_2 \circ D_1$, which satisfies

1233: $[D_1,D_2]=(-1)^{1+\deg D_1\cdot \deg D_2} [D_2,D_1]$.  We let

1234: $\Der_{l,r}^\alpha (\Omega_R A)$ be the subspaces consisting of left

1235: relative derivations of bidegree $\alpha$.

1236: 

1237: Let $\theta\in \Der_l(A)$ be a homogeneous relative left derivation of

1238: degree ${\tilde \theta}$.  The {\em contraction by $\theta$} is the

1239: unique relative left derivation $i_\theta \in \Der_{l}^{-1, {\tilde

1240: \theta}}(\Omega_R A)$ which satisfies $i_\theta a =0$ and $i_\theta (da)=\theta(a)$

1241: for all $a\in A$. The {\em Lie derivative along $\theta$} is

1242: the unique left derivation $L_\theta \in \Der_l^{0, {\tilde

1243: \theta}}(\Omega_R A)$ which satisfies $L_\theta a=\theta(a)$ and $L_\theta(da)=d\theta(a)$ 

1244: for all $a$. There are obvious versions of these definitions for right derivations. 

1245: It is easy to check that $i_\theta$ and $L_\theta$ are well-defined.

1246: Let $\Aut_R(A)$ be the space of automorphisms of $A$ as an

1247: $R$-superalgebra (in particular, all maps $\phi\in \Aut_R(A)$ are even and restrict to the identity on the subalgebra $R$). 

1248: For any $\theta,\gamma \in \Der_l(A)$ and $\phi\in \Aut_R(A)$, we have the identities:

1249: \begin{eqnarray}

1250: \label{Cartan}

1251: L_\theta&=&\left[i_\theta, d\right]\nn\\

1252: \left[L_\theta, i_\gamma\right]&=&i_{\left[\theta, \gamma\right]}\nn\\

1253: \left[L_\theta, L_\gamma\right]&=&L_{\left[\theta, \gamma\right]}\nn\\

1254: \left[i_\theta, i_\gamma\right]&=&0\\

1255: \left[L_\theta, d\right]&=&0 \nn\\

1256: L_{\phi \theta \phi^{-1}}&=&\phi^*L_\theta \phi^{* -1}\nn\\

1257: i_{\phi \theta \phi^{-1}}&=&\phi^*i_\theta \phi^{* -1}\nn~~.

1258: \end{eqnarray}

1259: As usual, these follow by noticing that all left and right hand sides

1260: are derivations of the same bi-degree on $\Omega_R A$, and checking agreement on the 

1261: generators $a$ and $da$ ($a\in A$). Given $w=a_0da_1\ldots da_n$ with $a_i\in A$, we have: 

1262: \begin{eqnarray}

1263: \label{ids}

1264: i_\theta w &=& \sum_{i=1}^n{

1265: (-1)^{i-1+{\tilde \theta}({\tilde

1266:     a}_0+\ldots +{\tilde a}_{i-1})}

1267: a_0da_1\ldots da_{i-1} \theta(a_i)da_{i+1}\ldots da_n}\nn\\

1268: L_\theta w&=& \theta(a_0) da_1\ldots da_n+

1269: \sum_{i=1}^n{(-1)^{{\tilde \theta}({\tilde

1270:     a}_0+\ldots +{\tilde a}_{i-1})}

1271: a_0da_1\ldots da_{i-1} d\theta(a_i)da_{i+1}\ldots da_n}~~.\nn

1272: \end{eqnarray}

1273: 

1274: \subsection{The bigraded Karoubi complex}

1275: 

1276: Consider the $\N\times \Z_2$-graded vector space: 

1277: \be

1278: C_R(A):=\Omega_R A/[\Omega_R A,\Omega_R A]~~,

1279: \ee

1280: where $[\Omega_R A,\Omega_R A]\subset \Omega_R A$ is the image of the bigraded 

1281: commutator map $[.,.]:\Omega_R A\times \Omega_R A\rightarrow \Omega_R A$:

1282: \be

1283: \nn

1284: [w_1,w_2]=w_1w_2-(-1)^{\deg w_1\cdot \deg w_2}w_2 w_1~~.

1285: \ee

1286: Notice that $[\Omega_R A,\Omega_R A]$ is a homogeneous subspace of $\Omega_R A$ but not a

1287: subalgebra. We let $\pi:\Omega_R A\rightarrow C_R(A)$ be the

1288: projection, and use the notation:

1289: \be

1290: \nn

1291: \pi(w):=(w)_c~~{\rm for}~~w\in \Omega_R A~~.

1292: \ee

1293: We also let $C_R^n(A)$ be the $\N$-homogeneous components of $C_R(A)$. 

1294: 

1295: Any relative derivation of $\Omega_R A$ preserves the subspace $[\Omega_R A, \Omega_R A]$,

1296: so it descends to a well-defined linear operator in $C_R(A)$.  In

1297: particular, $d$ induces a differential ${\bar d}$ on $C_R(A)$. The bigraded 

1298: vector space $(C_R(A),{\bar d})$ is the {\em relative Karoubi (or

1299: non-commutative de Rham) complex} of $A$ over $R$. This differential space is

1300: $\N\times \Z_2$-graded, and we have $\deg {\bar d}=(1,0)$.  The

1301: supervector spaces $H^n_R(A):=H_{\bar d}^n(C_R(A))$ are called the

1302: {\em relative de Rham cohomology groups of $A$}.

1303: 

1304: Given a morphism of $R$-superalgebras, the induced map 

1305: $\phi^*:(\Omega_R A_1,d_1)\rightarrow (\Omega_R A_2, d_2)$ of $R$-differential superalgebras 

1306: satisfies $\phi^*([\Omega_R A_1,\Omega_R A_1])\subset [\Omega_R A_2,\Omega_R

1307: A_2]$, so it descends to a morphism of bigraded complexes

1308: ${\bar \phi}^*:(C_R(A_1),{\bar d}_1)\rightarrow (C_R(A_2),{\bar

1309: d}_2)$.  In particular, any endomorphism $\phi$ of $A$ induces an

1310: endomorphism ${\bar \phi}^*$ of $(C_R(A),{\bar d})$, which is an automorphism 

1311: if $\phi$ is. We let ${\bar

1312: \phi}$ be the restriction of ${\bar \phi}^*$ to $C^0_R(A)=A/[A,A]$ (here $[A,A]\subset A$ is 

1313: the image of the supercommutator $[.,.]:A\times A\rightarrow A$).

1314: 

1315: The contraction and Lie operators $i_\theta$, $L_\theta$ also descend

1316: to well-defined $\C$-linear maps on $C_R(A)$, which we denote by ${\bar

1317: i}_\theta$, ${\bar L}_\theta$. It is clear that the induced operators

1318: satisfy all properties listed in eqs. (\ref{Cartan}):

1319: \begin{eqnarray}

1320: \label{Cartan_c}

1321: {\bar L}_\theta &=&\left[{\bar i}_\theta, {\bar d}\right]\nn\\

1322: \left[{\bar L}_\theta, {\bar i}_\gamma\right]&=&{\bar i}_{\left[\theta, \gamma\right]}\nn\\

1323: \left[{\bar L}_\theta, {\bar L}_\gamma\right]&=&{\bar L}_{\left[\theta, \gamma\right]}\nn\\

1324: \left[{\bar i}_\theta, {\bar i}_\gamma\right]&=&0\\

1325: \left[{\bar L}_\theta, {\bar d}\right]&=&0 \nn\\

1326: {\bar L}_{\phi \theta \phi^{-1}}&=&{\bar \phi}^*{\bar L}_\theta {\bar \phi}^{* -1}\nn\\

1327: {\bar i}_{\phi \theta \phi^{-1}}&=&{\bar \phi}^*{\bar i}_\theta {\bar \phi}^{* -1}\nn~~,

1328: \end{eqnarray}

1329: where $\phi\in \Aut_R(A)$~~.

1330: 

1331: \subsection{Noncommutative supersymplectic forms}

1332: 

1333: An element $\omega\in C^2_R(A)$ is called non-degenerate if the following complex-linear map is bijective: 

1334: \be

1335: \nn

1336: \theta\in \Der_l(A)\rightarrow {\bar i}_\theta \omega\in C^1_R(A)~~.

1337: \ee

1338: A {\em relative non-commutative symplectic form} on $A$ is 

1339: a $\Z_2$-homogeneous element $\omega\in C^2_R(A)$ which is closed (${\bar d}\omega=0$) and non-degenerate. 

1340: 

1341: 

1342: Given a symplectic form $\omega$ of $\Z_2$-degree ${\tilde \omega}$, a relative derivation $\theta\in \Der_l(A)$

1343: is called {\em symplectic} if ${\bar L}_\theta\omega=0$.  Let

1344: $\Der_l^\omega(A)\subset \Der_l(A)$ be the subspace of all symplectic

1345: derivations. By the third property in (\ref{Cartan_c}), this is a (super) Lie subalgebra of $\Der_l(A)$. 

1346: As in the even case, it is easy to see that the following map is an isomorphism of supervector spaces:

1347: \be

1348: \label{isom_0}

1349: \theta\in \Der_l^\omega(A)\rightarrow {\bar i}_\theta\omega\in C^1_R(A)_{\rm closed}[{\tilde \omega}]~~.

1350: \ee 

1351: Here $C^1_R(A)_{\rm closed}=\{\eta\in C^1_R(A)|{\bar d} \eta=0\}$, a homogeneous subspace of $C^1_R(A)$.

1352: 

1353: This implies that any $f\in C^0_R(A)$ defines a unique

1354: element $\theta_f\in \Der_l^\omega(A)$, determined by the equation

1355: ${\bar i}_{\theta_f}\omega={\bar d}f$. Let

1356: $\psi_\omega:C^0_R(A)\rightarrow \Der_l^\omega(A)$ be the complex-linear map

1357: given by $\psi_\omega(f):=\theta_f$. The relation ${\bar

1358: i}_{{\theta}}\omega={\bar d}f$ implies ${\tilde \theta}_f={\tilde

1359: \omega}+{\tilde f}$ for any homogeneous $f$, so the map $\psi_\omega$

1360: is homogeneous of degree ${\tilde \omega}$.  It is clear that the

1361: following sequence of supervector spaces is exact:

1362: \be

1363: \label{ex_0}

1364: 0\rightarrow

1365: H^0_R(A)\hookrightarrow C^0_R(A)\stackrel{\psi_\omega}{\rightarrow}\Der_l^\omega(A)[\tom]~~, 

1366: \ee

1367: where the map in the middle is the inclusion. Elements $\theta\in {\rm

1368: im}\psi_\omega$ are called {\em Hamiltonian derivations}.  Given a

1369: Hamiltonian derivation $\theta$, an element $f\in C^0_R(A)$ such that

1370: $\psi_\omega(f)=\theta\Leftrightarrow \theta=\theta_f$ is called a

1371: {\em Hamiltonian} associated with $\theta$. The sequence (\ref{ex_0})

1372: shows that the Hamiltonian of a Hamiltonian derivation is determined

1373: up to addition of elements of $H^0_R(A)$.

1374: 

1375: \subsection{$\Z_2$-graded version of the Kontsevich bracket}

1376: \label{sec:bracket}

1377: 

1378: We have the following generalization of an operation introduced in \cite{Konts_formal}. 

1379: 

1380: \paragraph{\bf Definition}

1381: The {\em  Kontsevich bracket} induced by $\omega$ is the $\Z_2$-homogeneous 

1382: complex-linear map $\{.,.\}:C^0_R(A)\otimes_\C C^0_R(A)\rightarrow C^0_R(A)$ of degree $\tom$ defined through: 

1383: \be

1384: \label{bdef}

1385: \{f,g\}:={\bar i}_{\theta_f}{\bar i}_{\theta_g}\omega~~~~~\forall f,g \in C^0_R(A)~~,\nn

1386: \ee

1387: where $\theta_f=\psi_\omega(f)$ and $\theta_g=\psi_\omega(g)$. 

1388: 

1389: \

1390: 

1391: \noindent Notice the relation:

1392: \be

1393: \label{bL}

1394: \{f,g\}={\bar L}_{\theta_f}g~~.

1395: \ee

1396: The following result gives the basic properties of the bracket in the $\Z_2$-graded case.

1397: \paragraph{\bf Proposition}

1398: The Kontsevich bracket satisfies the identities: 

1399: \be

1400: \label{gasym}

1401: \{g,f\}=(-1)^{1+({\tilde f}+{\tilde \omega})({\tilde g}+{\tilde \omega})}\{f,g\}~~.

1402: \ee

1403: and:

1404: \be

1405: \label{gjacobi}

1406: (-1)^{(\tf_1+\tom)(\tf_3+\tom)}

1407: \{f_1,\{f_2,f_3\}\}+(-1)^{(\tf_2+\tom)(\tf_1+\tom)}\{f_2,\{f_3,f_1\}\}+

1408: (-1)^{(\tf_3+\tom)(\tf_2+\tom)}\{f_3,\{f_1,f_2\}\}=0~~.

1409: \ee

1410: Hence $(C^0_R(A)[\tom],\{.,.\})$ is a Lie superalgebra. 

1411: 

1412: \begin{proof}

1413: 

1414: The first property follows immediately from $[{\bar i}_{\theta_f},{\bar i}_{\theta_g}]=0$. 

1415: For the second property, let $f_1,f_2,f_3\in C^0_R(A)$ and

1416: $\theta_i:=\phi_\omega(f_i)$. Identities (\ref{Cartan_c}) give:

1417: \begin{eqnarray}

1418: 0={\bar i}_{\theta_2}{\bar i}_{\theta_1}

1419: {\bar i}_{\theta_0}d\omega&=&(-1)^{{\tilde

1420:     \theta}_0({\tilde \theta_1}+{\tilde

1421:     \theta}_2)}{\bar L}_{\theta_0}{\bar i}_{\theta_2}{\bar i}_{\theta_1}\omega+

1422: (-1)^{1+{\tilde \theta}_1 {\tilde

1423:     \theta}_2}{\bar L}_{\theta_1}{\bar i}_{\theta_2}{\bar i}_{\theta_0}\omega+

1424: {\bar L}_{\theta_2}{\bar i}_{\theta_1}{\bar i}_{\theta_0}

1425: \omega\nn\\

1426: &+&

1427:     (-1)^{1+\th_0\th_1}{\bar i}_{\theta_2}{\bar i}_{[\theta_0,\theta_1]}\omega+(-1)^{\th_2(\th_0+\th_1)}

1428: {\bar i}_{\theta_1}

1429:     {\bar i}_{[\theta_0,\theta_2]}\omega+(-1)^{1+\th_0\th_1+\th_2(\th_0+\th_1)}{\bar i}_{\theta_0}

1430: {\bar i}_{[\theta_1,\theta_2]}\omega~~.\nn

1431: \end{eqnarray}

1432: Using the relations

1433: $[\theta_i,\theta_j]=(-1)^{1+\th_i\th_j}[\theta_j,\theta_i]$ and the

1434: properties of ${\bar L}_{\theta_i}$, the

1435: right hand side can be brought to the form:

1436: \be

1437: 2\left[(-1)^{1+\th_0\th_1+\th_1\th_2+\th_2\th_0}\{f_0,\{f_1,f_2\}\}+(-1)^{1+\th_1\th_2}

1438: \{f_1,\{f_2,f_0\}\}+(-1)^{1+\th_0\th_1}\{f_2,\{f_0,f_1\}\}\right]\nn

1439: \ee

1440: Multiplying with $(-1)^{1+\th_0\th_1+\th_1\th_2}$ leads to the

1441: identity: 

1442: \be

1443: (-1)^{\th_0\th_2}\{f_0,\{f_1,f_2\}\}+(-1)^{\th_1\th_0}\{f_1,\{f_2,f_0\}\}+

1444: (-1)^{\th_2\th_1}\{f_2,\{f_0,f_1\}\}=0~~.\nn

1445: \ee

1446: This implies equation (\ref{gjacobi}) upon changing the indices

1447: $0,1,2$ into $1,2,3$ and using $\th_i={\tilde

1448:   f}_i+{\tilde \omega}$. 

1449: 

1450: \end{proof}

1451: 

1452: 

1453: \noindent The map $\psi_\omega$ has another property which parallels classical behavior. 

1454: 

1455: \paragraph{\bf Proposition}

1456: {\em We have: 

1457: \be

1458: \label{Lie_morphism}

1459: \theta_{\{f,g\}}=[\theta_f, \theta_g]~~~~~~\forall f,g\in C^0_R(A)~~.

1460: \ee

1461: Thus $\psi_\omega:(C^0_R(A)[\tom],\{.,.\}) \rightarrow (\Der_l^\omega(A),[.,.])$ is a morphism of Lie

1462: superalgebras over $\C$. }

1463: 

1464: \begin{proof}

1465: Compute: 

1466: \be

1467: {\bar d}\{f,g\}={\bar d}{\bar i}_{\theta_f}{\bar i}_{\theta

1468: g}\omega={\bar L}_{\theta_f}{\bar i}_{\theta_g}\omega+{\bar i}_{\theta_f}

1469: {\bar d}({\bar i}_{\theta_g}\omega)={\bar L}_{\theta_f}{\bar i}_{\theta_g}\omega=[{\bar L}_{\theta_f},

1470: {\bar i}_{\theta_g}]\omega={\bar i}_{[\theta_f,\theta_g]}\omega~~.\nn

1471: \ee

1472: In the third equality, we used ${\bar i}_{\theta_g}\omega={\bar d}g$ and ${\bar d}^2=0$,

1473: while in the fourth we used ${\bar L}_{\theta_f}\omega=0$.

1474: \end{proof}

1475: 

1476: 

1477: An $R$-superalgebra automorphism $\phi\in \Aut_R(A)$ is called a {\em relative

1478: symplectomorphism} if ${\bar \phi}^*(\omega)=\omega$.  We let

1479: $\Aut^\omega_R(A)\subset \Aut_R (A)$ be the subgroup of relative

1480: symplectomorphisms of $A$.  By the sixth property in (\ref{Cartan_c}),

1481: the obvious action of $\Aut^\omega_{alg}(A)$ on $\Der_l(A)$ preserves

1482: the Lie subalgebra $\Der_l^\omega(A)$. Given a symplectomorphism

1483: $\phi$, the last property in (\ref{Cartan_c}) implies $\phi\circ

1484: \theta_f\circ \phi^{-1}=\theta_{{\bar \phi}(f)}$ for $f\in C^0_R(A)$,

1485: i.e.  $\psi_\omega\circ {\bar \phi} ={\rm Ad}_{\phi}\circ \psi_\omega$

1486: for all $\phi\in \Aut^\omega_R(A)$. In turn, this gives $\{{\bar

1487: \phi}(f),{\bar \phi}(g)\}={\bar \phi}(\{f,g\})$.  Hence

1488: $\Aut^\omega_R(A)$ acts on $(A[{\tilde \omega}],\{.,.\})$ by Lie

1489: algebra automorphisms.

1490: 

1491: 

1492: 

1493: \section{Noncommutative calculus for finite D-brane systems}

1494: 

1495: Consider a finite D-brane system with object set ${\cal Q}_0$. As in

1496: Section \ref{algebraic}, we set $R=\oplus_{u\in {\cal Q}_0}{\C \epsilon_u}$ and let $E$ be the

1497: $R$-superbimodule of boundary sectors.  Recall that $E[1]$ carries a

1498: symplectic form $\omega=\rho\circ \Sigma^{\otimes 2}$, the parity change of the topological metric $\rho$.

1499: Setting $V=E[1]^{\rm v}$, we consider the $\N\times \Z_2$ graded tensor

1500: algebra $A=T_R (V)$ whose $\N$-homogeneous subspaces we denote by $A_n=T_R^n (V)$.  

1501: With respect to its $\Z_2$-grading, $A$ is an $R$-superalgebra with  $R=A_0$. 

1502: 

1503: \subsection{Generalities}

1504: \label{tensor}

1505: 

1506: The symplectic form $\omega$ on $E[1]$ induces a

1507: relative noncommutative symplectic form $\omega_{form}$ on $A$ as

1508: follows. Recall from Section 2 that the symplectic form $\omega$ defines 

1509: a map $j_\omega: E[1]^{opp}\rightarrow E[1]^{\rm v}$ which is a morphism of $R$-bimodules, i.e. 

1510: an element of $\iHom(E[1]^{opp},E[1]^{\rm v})$. 

1511: As explained in Appendix \ref{isomorphisms}, there exists 

1512: an isomorphism of $R$-superbimodules between 

1513: $\iHom(E[1]^{opp},E[1]^{\rm v})$ and the center of the superbimodule 

1514: $E[1]^{\rm v}\otimes_R E[1]^{\rm v}$,

1515: which allows us to view $\omega$ as an element ${\hat \omega}$ of degree ${\tilde \omega}$

1516: of the space $(V\otimes_R V)^R$. If ${\hat \omega}=\sum_{i}{f_i\otimes_R g_i}$ with $f_i,g_i\in V$, 

1517: then it is shown in Appendix \ref{isomorphisms} that $\omega$ can be recovered as: 

1518: \be

1519: \label{omega_rec}

1520: \omega(x,y)=\sum_i f_i(xg_i(y))~~.

1521: \ee

1522: Using the element ${\hat \omega}$, we define a noncommutative two-form on $A=T_R V$ through the relation:

1523: \be

1524: \label{omega_form_gen}

1525: \omega_{form}=-\frac{1}{2}\sum_{i}{(df_i dg_i)_c}\in C^2_R(A)~~,

1526: \ee

1527: where the minus sign is introduced for later convenience. 

1528: It is easy to see that $\omega_{form}$ is well-defined

1529: and of the same $\Z_2$-degree as $\omega$.  Moreover, it is not hard to check that this two-form is 

1530: symplectic.

1531: 

1532: 

1533: Since $A=T_R V$, the algebra of noncommutative forms $\Omega_RA$ has a

1534: second $\N$-grading, which is induced from the $\N$-grading of $A$

1535: (with respect to this grading, we have $\deg a =\deg da=1$ for all

1536: $a\in V$, while $R$ sits in degree zero).  This induces a similar

1537: $\N$-grading on $C_R(A)$.  We let $(\Omega_RA)_n$ and $C_R(A)_n$ be

1538: the homogeneous subspaces determined by this grading. A {\em

1539: constant} noncommutative two-form \footnote{The notion of constant

1540: noncommutative symplectic form depends on the specific realization of

1541: $A$ as a tensor algebra $T_R V$. In particular, this concept is not

1542: invariant under the $R$-superalgebra automorphism group $\Aut_R(A)$

1543: (because a superalgebra automorphism need not preserve the

1544: $\N$-grading of $A$).} on $A$ is an element of $C^2_R(A)_2$. Thus

1545: (\ref{omega_form_gen}) is a constant symplectic form, and any constant

1546: symplectic form on $A$ has such an expansion.  We let $CNS^{\tilde \omega}(V)$ be the

1547: vector space of all constant noncommutative symplectic forms on $T_R V$ of degree ${\tilde \omega}$. In

1548: what follows, we often denote $\omega_{form}$ by $\omega$; which of

1549: the two is meant should be clear from the context.

1550: 

1551: 

1552: Notice that $[A,A]\cap A_0=[A_0,A_0]=[R,R]=0$ since $A_0=R$ and $R$ is commutative.  In particular, 

1553: we have $C^0_R(A)_0=R$. The following result follows as in \cite{Ginzburg}, by

1554: considering the Euler derivation associated with the $\N$-grading induced from $A$.

1555: 

1556: \paragraph{\bf Proposition} {\em We have $H^0_R(A)=C^0_R(A)_0=R$ and $H^n_R(A)=0$ for all $n\geq 1$.}

1557: 

1558: \

1559: 

1560: \noindent Using this in (\ref{ex_0}) gives a short exact sequence: 

1561: \be

1562: \label{ex}

1563: 0\rightarrow

1564: R\rightarrow C^0_R(A)\stackrel{\psi_\omega}{\rightarrow}\Der_l^\omega(A)[\tom]\rightarrow 0~~,

1565: \ee

1566: where surjectivity of $\psi_\omega$ follows by using $H^1_R(A)=0$

1567: in (\ref{isom_0}). In particular, {\em any relative symplectic derivation of $A$

1568: is Hamiltonian}. The sequence (\ref{ex}) also shows that the

1569: Hamiltonian of a symplectic derivation is determined up to addition of

1570: elements of $R$, which can be viewed as the subspace $C^0_R(A)_0$ of $C^0_R(A)$.

1571: 

1572: Any element $f$ of $C^0_R(A)$ has a decomposition $f=\sum_{n\geq

1573: 0}{f_n}$, with $f_n\in C^0_R(A)_n$ (in particular, $f_0\in C^0_R(A)_0=R$).  

1574: We say that $f$ {\em has order $k$

1575: at zero} if $f_0= f_1= \ldots =f_{k-1}=0$ and $f_k\neq 0$. We say that

1576: $f$ {\em vanishes at zero} if $f_0=0$, i.e. $f$ has order at least one

1577: at zero. We define the {\em canonical Hamiltonian} of a symplectic

1578: derivation $\theta$ to be that Hamiltonian of $\theta$ which vanishes

1579: at zero. The sequence (\ref{ex}) shows that the canonical

1580: Hamiltonian exists and is unique.

1581: 

1582: 

1583: 

1584: 

1585: 

1586: \subsection{Adapted bases and superquivers} 

1587: \label{coord_description}

1588: 

1589: Using the binary decomposition $E=\oplus_{(u,v)\in {\cal Q}_0\times {\cal Q}_0}{E_{uv}}$, 

1590: consider a homogeneous basis $(\psi_a)$ of the supervector space $E$ having the following properties:

1591: 

1592: \

1593: 

1594: (1)   $a=(u,v, j)$ is a multi-index with $u,v \in {\cal Q}_0$ and $j=1\ldots \dim_\C E_{uv}$

1595: 

1596: \

1597: 

1598: (2) $\psi_{uvj}$ for $j=1\ldots \dim_\C E_{uv}$ is a 

1599: homogeneous basis of $E_{uv}$ for all $u,v\in {\cal Q}_0$.

1600: 

1601: \

1602: 

1603: \noindent We say that such a homogeneous basis is {\em adapted} to the binary decomposition of $E$. 

1604: Setting $e_a:=\Sigma \psi_a$, we let $(s^a)$ be the basis of the super-vector space 

1605: $V=E[1]^{\rm v}$ dual to $(e_a)$: 

1606: \be

1607: \nn

1608: s^a(e_b)=\delta^a_b~~.

1609: \ee

1610: Relation (\ref{dual_spaces}) shows that 

1611: $s^{uvj}$ are bases of $V_{uv}=(E[1]^{\rm v})_{uv}=(E_{uv}[1])^*$, odd dual to the 

1612: bases  $\psi_{uvj}$  of $E_{uv}$. We set ${\tilde

1613: a}:=\deg s^a=\deg e_a=[|a|+1]$, where $|a|:=|\psi_a|$.  

1614: 

1615:  

1616: It is convenient to keep track of indices by considering a quiver

1617: ${\cal Q}$ determined by the multi-indices $a$. Specifically, the {\em index quiver} ${\cal Q}$ is the quiver 

1618: on the vertex set ${\cal Q}_0$ obtained by drawing an arrow from $u$ to $v$ for

1619: each $j=1\ldots \dim_\C V_{uv}$. With this construction, we can identify each

1620: multi-index $a$ with the corresponding arrow of ${\cal Q}$. We let ${\cal Q}_1$ be the set of all

1621: arrows and ${\cal Q}_1(u,v)$ the subset of arrows going from $u$ to $v$.  We

1622: also let $h,t:{\cal Q}_1\rightarrow {\cal Q}_0$ be the head and tail maps of ${\cal Q}$.

1623: 

1624: The index quiver is in fact a {\em superquiver},

1625: being endowed with a map $\deg: {\cal Q}_1\rightarrow \Z_2$ 

1626: given by $\deg (a)={\tilde a}$.  An arrow $a$

1627: is called even if ${\tilde a}=0$ and odd if ${\tilde a}=1$.  The path algebra

1628: $\C {\cal Q}$ becomes a superalgebra by declaring a path $p$ to be even

1629: or odd if it contains an even or odd number of odd arrows. That is, we

1630: define the degree of $p$ by the formula:

1631: \be

1632: \nn

1633: {\tilde p}:=\sum_{j=1}^k{\tilde a_j}~~,

1634: \ee 

1635: where $p=a_1\ldots a_k$ is the arrow decomposition. The trivial paths

1636: are taken to be even.  The path algebra is in fact $\N\times

1637: \Z_2$-graded, where the $\N$-grading is induced by the length of paths. 

1638: We let $(\C {\cal Q})_n$ be the components of degree $n$ with respect to the length grading.

1639: 

1640: As usual, the subspace spanned by the trivial paths forms a finite-dimensional semisimple commutative algebra. 

1641: We identify this with the boundary sector algebra $R$ by sending the trivial path at $u$ 

1642: into the idempotent $\epsilon_u$.  On the other hand, the subspace spanned by

1643: the arrows is isomorphic with the supervector space $V$:

1644: \be

1645: \nn

1646: (\C {\cal Q})_1\approx \oplus_{u,v\in {\cal Q}_0}\oplus_{a\in {\cal Q}_1(u,v)}\C[{\tilde a}]\approx V~~

1647: \ee

1648: via the identification $a\equiv s^a$. It is also clear that the

1649: $R$-superbimodule structure of $V$ coincides with the

1650: $(\C{\cal Q})_0$-superbimodule structure induced on $(\C{\cal Q})_1$ by multiplication in

1651: the path algebra. In fact,  the entire path algebra is isomorphic

1652: with the tensor algebra $T_R (\C {\cal Q})_1$ as an $\N\times

1653: \Z_2$-graded algebra. Combining these observations, we find an isomorphism of bigraded algebras: 

1654: \be

1655: \nn

1656: A\approx \C{\cal Q}

1657: \ee

1658: which extends the isomorphism $V \approx (\C {\cal Q})_1$. Hence:

1659: 

1660: \

1661: 

1662: {\em Choosing an adapted basis of $E$ identifies the tensor algebra

1663: $A=T_R V$ with the path algebra of the index superquiver ${\cal Q}$.}

1664: 

1665: \

1666: 

1667: \noindent In the next subsections, we explore the consequences of this identification. 

1668: 

1669: \subsection{Quiver description of the symplectic structure}

1670: \label{symplectic_structure}

1671: 

1672: Recall that the symplectic form $\omega$ on $E[1]$ corresponds to bilinear forms $\omega_{uv}:E[1]_{uv}\times

1673: E[1]_{vu}\rightarrow \C$ of common degree ${\tilde \omega}$, such that $\omega_{uv}(x,y)=(-1)^{{\tilde

1674: x}{\tilde y}+1}\omega_{vu}(y,x)$.  We define coefficients $(\omega_{ab})_{a,b\in {\cal Q}_1}$ 

1675: through:

1676: \be

1677: \nn

1678: \omega_{ab}=-\omega_{t(a)h(a)}(e_a,e_b)\in \C~~{\rm~if~}~~t(a)=h(b)~~{\rm~and~}~~h(a)=t(b)

1679: \ee

1680: and zero otherwise. The minus sign in this expression is introduced in 

1681: order to simplify certain formulas which will appear in the next sections. 

1682: In terms of the superbimodule symplectic form, we have 

1683: $\omega(e_a,e_b)=-\omega_{ab}\epsilon_u$ where $u:=t(a)=h(b)$. 

1684: The coefficients have the graded-antisymmetry properties:

1685: \be

1686: \label{omega_symm}

1687: \omega_{ab}=(-1)^{{\tilde a}{\tilde b}+1}\omega_{ba}~~

1688: \ee

1689: and satisfy the selection rule:

1690: \be

1691: \label{omega_sel}

1692: \omega_{ab}=0~~{\rm~unless~}~~{\tilde a}+{\tilde b}={\tilde \omega}~~.

1693: \ee

1694: It is not hard to see that the inverse matrix $(\omega^{ab})_{a,b\in {\cal Q}_1}$ also satisfies the selection

1695: rule:

1696: \be

1697: \nn

1698: \omega^{ab}=0~~{\rm unless}~~{\tilde a}+{\tilde b}={\tilde \omega}

1699: \ee 

1700: and graded antisymmetry property:

1701: \be

1702: \nn

1703: \omega^{ab}=(-1)^{{\tilde a}{\tilde b}+1}\omega^{ba}~~.

1704: \ee

1705: Moreover, $\omega^{ab}$ vanishes unless $h(a)=t(b)$ and $t(a)=h(b)$. 

1706: 

1707: Let ${\hat \omega}$ be the central element of $V\otimes_R V$ determined by $\omega$. Since $s^a$ is a vector 

1708: space basis of $V$, we can expand ${\hat \omega}=\sum_{a,b}{{\hat \omega}_{ab}s^a\otimes_R s^b}=

1709: \sum_{a}{s^a\otimes_R (s^a)'}$, where we set $(s^a)':=\sum_{b}{\hat \omega}_{ab}s^b$. 

1710: Then equation (\ref{omega_rec}) gives $\omega(e_a,e_b)=\sum_{c}s^c(e_a (s^c)'(e_b))={\hat \omega}_{ab}$. 

1711: Thus ${\hat \omega}_{ab}=-\omega_{ab}\epsilon_{t(a)}$ and we find that our definition of coefficients 

1712: corresponds to the expansion: 

1713: \be

1714: \nn

1715: {\hat \omega}=-\sum_{a,b\in {\cal Q}_1}\omega_{ab}s^a\otimes_R s^b\in V\otimes_R V~~,

1716: \ee

1717: where we used the relation $\epsilon_{t(a)} s^a=s^a$, which holds because $s^a\in V_{t(a)h(a)}$. 

1718: Using equation (\ref{omega_form_gen}), 

1719: it follows that the non-commutative symplectic form induced on $A$  is given by:

1720: \be

1721: \label{omega_form}

1722: \omega_{form}=\frac{1}{2}\sum_{\tiny 

1723: \begin{array}{c}a,b\in {\cal Q}_1 \\h(a)=t(b),  h(b)=t(a)\end{array}}\omega_{ab}(ds^a ds^b)_c\equiv

1724: \frac{1}{2}\sum_{a,b\in Q_1}\omega_{ab}(da db)_c~~.

1725: \ee

1726: In the last equality, notice that $(dadb)_c$ vanishes unless

1727: $h(a)=t(b)$ and $h(b)=t(a)$, which can be seen immediately by

1728: inserting idempotents $\epsilon_u$ in the appropriate places.  As in

1729: \cite{Ginzburg, LB_quivers}, it is easy to check that 

1730: non-degeneracy of the constant two-form (\ref{omega_form}) 

1731: amounts to non-degeneracy of the matrix $(\omega_{ab})_{a,b\in {\cal Q}_1}$.  A {\em symplectic

1732: superquiver} is a superquiver whose path algebra is endowed with a

1733: constant symplectic form of type (\ref{omega_form}). Hence picking an

1734: adapted homogeneous basis allows us to encode the information of the

1735: symplectic superbimodule $(E[1],\omega)$ into a symplectic

1736: superquiver. Of course, the inverse correspondence also holds.

1737: 

1738: 

1739: 

1740: Recall that the topological metric $\rho$ on $E$ is given by

1741: $\omega=\rho\circ \Sigma^{\otimes 2}$. We define its coefficients

1742: trough:

1743: \be

1744: \nn

1745: \rho_{ab}=\rho(\psi_a,\psi_b)~~,

1746: \ee

1747: {\em without} a minus sign insertion. Equation $\omega(e_a,e_b)=(-1)^{\tilde a}\rho(\psi_a,\psi_b)$ gives: 

1748: \be

1749: \label{omega_rho}

1750: \omega_{ab}=(-1)^{{\tilde a}+1}\rho_{ab}~~.

1751: \ee

1752: The coefficients of $\rho$ have the properties:

1753: \be

1754: \nn

1755: \rho_{ab}=(-1)^{{\tilde a}{\tilde b}+{\tilde \omega}+1}\rho_{ba}=(-1)^{|a||b|}\rho_{ba}

1756: \ee 

1757: and: 

1758: \be

1759: \nn

1760: \rho_{ab}=0{\rm~unless~}{\tilde a}+{\tilde b}={\tilde \omega}\Longleftrightarrow |a|+|b|={\tilde \omega}~~.\nn

1761: \ee

1762: Relation (\ref{omega_rho}) shows that the inverse of the matrix $(\rho_{ab})$ takes the form:

1763: \be

1764: \rho^{ab}=(-1)^{{\tilde b}+1}\omega ^{ab}~~.\nn

1765: \ee

1766: It is clear that the inverse matrix satisfies the relations:

1767: \be

1768: \label{oinv_symm}

1769: \rho^{ab}=(-1)^{{\tilde a}{\tilde b}+{\tilde \omega}+1}\rho^{ba}=(-1)^{|a||b|}\rho^{ba}~~.

1770: \ee

1771: and

1772: \be

1773: \label{oinv_sel}

1774: \rho^{ab}=0{\rm~unless~}{\tilde a}+{\tilde b}={\tilde \omega}\Longleftrightarrow |a|+|b|={\tilde \omega}~~.

1775: \ee

1776: 

1777: The structure theorem for graded antisymmetric matrices implies the following: 

1778: 

1779: \noindent (1) If ${\tilde \omega}=0$, then we can find an adapted basis and an ordering  

1780: $a_1,\ldots, a_{2m},a_{2m+1},\ldots, a_N$ of the arrows ($N={\rm Card} {\cal Q}_1$) 

1781: such that $a_1,\ldots, a_{2m}$ are even, $a_{2m+1},\ldots, a_N$ are odd and:

1782: \be

1783: \nn

1784: \omega=\sum_{i=1}^m{(da_id a_{i+m})_c}+\frac{1}{2}\sum_{j=2m+1}^N{(da_j da_j)_c}~~.

1785: \ee

1786: Setting $p_i:=a_i$ and $q_i=a_{i+m}$ for $i=1\ldots m$ and $\xi_\alpha:=a_{2m+\alpha}$ for $\alpha=1\ldots N-2m$, 

1787: we can write this in the form: 

1788: \be

1789: \label{e_can}

1790: \omega=(dp_idq_i)_c+\frac{1}{2}(d\xi_\alpha d\xi_\alpha)_c

1791: \ee

1792: with even $p_i,q_i$ and odd $\xi_\alpha$. Hence $\omega_{p_iq_j}=-\omega_{q_jp_i}=\delta_{ij}$ 

1793: and $\omega_{\xi_\alpha\xi_\beta}=\delta_{\alpha\beta}$. 

1794: 

1795: \noindent (2) If ${\tilde \omega}=1$, then ${\rm Card} {\cal Q}_1=2m$ for some integer $m$ and 

1796: we can find an adapted basis and an ordering 

1797: $a_1,\ldots,a_{2m}$ of the arrows such that $a_1,\ldots, a_{m}$ are odd, $a_{m+1},\ldots, a_{2m}$ are even and:

1798: \be

1799: \nn

1800: \omega=\sum_{i=1}^m{(da_id a_{i+m})_c}~~.

1801: \ee

1802: Setting $p_i:=a_i$ and $q_i=a_{i+m}$ for $i=1\ldots m$, this becomes: 

1803: \be

1804: \label{o_can}

1805: \omega=(dp_idq_i)_c

1806: \ee

1807: with even $p_i$ and odd $q_i$. Hence $\omega_{p_iq_j}=-\omega_{q_jp_i}=\delta_{ij}$.

1808: 

1809: In general, one can set $a^*=\sum_{b\in {\cal Q}_1}\omega_{ab}b$, 

1810: which brings $\omega$ to the form $\omega=\frac{1}{2}(dada^*)_c$. 

1811: For an even $\omega$ in the canonical basis (\ref{e_can}), we have $p_i^*=q_i$, $q_i^*=-p_i$ and 

1812: $\xi_\alpha^*=\xi_\alpha$. In this case, $*$ squares to minus the identity on the subspace spanned by $q_i, p_i$ 

1813: but to plus the identity on the subspace spanned by $\xi_\alpha$. 

1814: For odd $\omega$ in the basis (\ref{o_can}), we have $p_i^*=q_i$ and $q_i^*=-p_i$, so 

1815: $*$ squares to minus the identity on the entire subspace $A_1=V$. 

1816: 

1817: It is clear from the above that a D-brane system has different behavior 

1818: depending on the parity of $\omega$. We say that the system 

1819: is {\em even} or {\em odd} if ${\tilde \omega}=0$, respectively $1$. 

1820: 

1821: 

1822: \subsection{Quiver description of $C^0_R(A)$}

1823: \label{quivc0}

1824: 

1825: Any element $f\in A$ has an expansion: 

1826: \be

1827: \label{f_path}

1828: f=\sum_{p=path}{f_p p}

1829: \ee

1830: where the sum is over all paths $p$ of ${\cal Q}$ (including the

1831: trivial paths) and where $f_p\in \C$. In this and subsequent

1832: relations, we agree that only a finite number of coefficients 

1833: are nonzero, so that all sums are finite. We can also write (\ref{f_path}) as:

1834: \be

1835: \label{f}

1836: f=\sum_{n\geq 0}{f_{a_1\ldots a_n}a_1\ldots a_n}~~,

1837: \ee

1838: where we use implicit summation over the arrows $a_j$ and we agree

1839: that $f_{a_1\ldots a_0}a_1\ldots a_0$ stands for the sum

1840: $c(f):=\sum_{u\in {\cal Q}_0}{f_u\epsilon_u}$ (with $f_u\in \C$).  

1841: The product $a_1\ldots a_n$ vanishes unless it is a path. This is seen by

1842: inserting idempotents:

1843: \be

1844: \nn

1845: a_1\ldots a_ia_{i+1}\ldots a_n=a_1\ldots a_i\epsilon_{h(a_i)}\epsilon_{t(a_{i+1})}a_{i+1}\ldots a_n~~,

1846: \ee

1847: and noticing that the right hand side vanishes unless

1848: $h(a_i)=t(a_{i+1})$. Thus only the coefficients $f_{a_1\ldots a_n}$

1849: which correspond to paths $a_1\ldots a_n$ are defined; for convenience, we 

1850: define $f_{a_1\ldots a_n}$ to vanish if the word $a_1\ldots a_n$ is not a path. 

1851: 

1852: The obvious relations

1853: \be

1854: \label{lindep}

1855: ({a_1}\ldots {a_n})_c=(-1)^{({\tilde a}_1+\ldots +

1856:   {\tilde a}_i)({\tilde a}_{i+1}+\ldots +{\tilde a}_n)}({a_{i+1}}\ldots

1857:   {a_n}{a_1}\ldots {a_i})_c~~.

1858: \ee 

1859: show that $({a_1}\ldots {a_n})_c$ vanishes unless $a_1\ldots

1860: a_n$ is a cycle of ${\cal Q}$ (as we will see below, it can still

1861: vanish even for a cycle). For a general element (\ref{f}), relations (\ref{lindep}) give:

1862: \be

1863: \nn

1864: f_c:=\pi(f)=\sum_{n\geq 0}{f_{a_1\ldots a_n}({a_1}\ldots

1865:   {a_n}})_c=\sum_{n\geq 0}{f_{(a_1\ldots a_n)}({a_1}\ldots

1866:   {a_n})_c}~~,

1867: \ee

1868: where we introduced the `cyclicized coefficients': 

1869: \be

1870: \nn

1871: f_{(a_1\ldots a_n)}:=\frac{1}{n}\sum_{i=0}^{n-1}{(-1)^{({\tilde a}_1+\ldots +{\tilde

1872:       a}_i)({\tilde a}_{i+1}+\ldots +{\tilde a}_n)}f_{a_{i+1}\ldots

1873:     a_n, a_1\ldots a_i}}~~,

1874: \ee

1875: which satisfy $f_{(a_1\ldots a_n)}=(-1)^{({\tilde a}_1+\ldots +

1876:   {\tilde a}_i)({\tilde a}_{i+1}+\ldots +{\tilde a}_n)} f_{(a_{i+1}\ldots

1877:       a_n, a_1\ldots a_i)}$.

1878: For $n=0$, we set $f_{(u)}=f_{u}$ for all $u\in {\cal Q}_0$.

1879: 

1880: The observations made above show that any element $f\in C^0_R(A)$ can be expanded as: 

1881: \be

1882: \label{fc_expansion}

1883: f=\sum_{n\geq 0}{f_{a_1\ldots a_n}(a_1\ldots a_n)_c}

1884: \ee

1885: where the coefficients are taken to be graded cyclic: 

1886: \be

1887: \nn

1888: f_{a_1\ldots a_n}=(-1)^{({\tilde a}_1+\ldots +

1889:   {\tilde a}_i)({\tilde a}_{i+1}+\ldots +{\tilde a}_n)} f_{a_{i+1}\ldots

1890:       a_n, a_1\ldots a_i}~~

1891: \ee

1892: and the term $n=0$ in the sum stands for $\sum_{u\in {\cal Q}_0}{f_u\epsilon_u}\in R$. 

1893: We also define the {\em strict coefficients} of $f\in C^0_R(A)$ by: 

1894: \be

1895: \nn

1896: {\bar f}_{a_1\ldots a_n}:=nf_{a_1\ldots a_n}~~{\rm if}~~ n\neq 0~~

1897: \ee

1898: and ${\bar f}_{u}:=f_u$ for $n=0$. Then the expansion of $f_c$ becomes:

1899: \be

1900: \nn

1901: f_c=c(f)+\sum_{n\geq 1}{\frac{{\bar f}_{a_1\ldots a_n}}{n}({a_1}\ldots {a_n})_c}~~.

1902: \ee 

1903: 

1904: Consider the set ${\cal C}({\cal Q})$ of cycles of ${\cal Q}$. We say

1905: that two cycles $\gamma_1$ and $\gamma_2$ are {\em equivalent}, and

1906: write $\gamma_1\sim \gamma_2$, if they have the same length and differ

1907: by a cyclic permutation of their arrows (i.e. they differ only in the

1908: choice of the initial=terminal point of the cycle).  This is an

1909: equivalence relation on ${\cal C}({\cal Q})$, whose equivalence classes are

1910: known as {\em necklaces}. We let $N({\cal Q}):={\cal C}({\cal Q})/\sim$ denote the

1911: set of necklaces, and write $[\gamma]$ for the equivalence class of a

1912: cycle $\gamma$. We define the length $l([\gamma])$ to be the length of

1913: any representative cycle. The $\Z_2$ degree of a necklace is the

1914: degree of any of its representatives.  This gives a well-defined map

1915: from $N({\cal Q})$ to $\Z_2$.

1916: 

1917: A cycle $\gamma$ is called primitive  if it cannot be

1918: written in the form $\gamma=u^k$ with $u$ a non-trivial cycle and

1919: $k\geq 2$ (with this definition, the trivial paths $\epsilon_u$ are

1920: primitive).  Any non-trivial cycle $\gamma$ can be written uniquely in

1921: the form $\gamma=r(\gamma)^{p(\gamma)}$ where $p(\gamma)\in \N^*$ and

1922: $r(\gamma)$ is a primitive cycle. This representation is called the

1923: primitive decomposition of $\gamma$.  The integer $p(\gamma)$ is

1924: called the period, while the path $r(\gamma)$ is called the primitive

1925: root of $\gamma$.  Given a necklace $\nu$ and representatives

1926: $\gamma_1,\gamma_2\in \nu$, we have $p(\gamma_1)=p(\gamma_2)$ and

1927: $r(\gamma_1)\sim r(\gamma_2)$. This allows us to define the period and

1928: primitive root of necklaces through $p([\gamma])=p(\gamma)$ and

1929: $r([\gamma])=[r(\gamma)]$.

1930: 

1931: \paragraph{\bf Definition}

1932: A {\em null necklace} is a necklace $\nu$ such that $p(\nu)$ is

1933: even and $\deg~r(\nu)=1\in \Z_2$. 

1934: 

1935: \

1936: 

1937: \paragraph{\bf Proposition}

1938: Let $\gamma$ be a cycle of the quiver ${\cal Q}$. The vector

1939: $\gamma_c=\pi(\gamma)\in C^0_R(A)$ vanishes if and only if the necklace

1940: $\nu:=[\gamma]$ is null.

1941: 

1942: \

1943: 

1944: \noindent We let $N_\bullet({\cal Q}):=\{ \nu\in N({\cal Q}) |\nu~{\rm~is~{\rm

1945: not}~null} \}$ be the set of non-null necklaces. Notice that the

1946: trivial paths $\epsilon_u$ are not null, and thus belong to $N_\bullet({\cal Q})$.

1947: \begin{proof}

1948: 

1949: Let $\gamma=r^p$ be the primitive decomposition of $\gamma$. Commuting

1950: one copy of $r$ to the right gives:

1951: \be 

1952: \gamma_c=\pi(r^p)=(-1)^{(p-1){\tilde r}}\gamma_c,\nn 

1953: \ee 

1954: where we noticed that $\deg (r)\deg(r^{p-1})={\tilde r}(p-1)$ in

1955: $\Z_2$.  Thus $\gamma_c=0$ if $p$ is even and ${\tilde r}$ is odd. The

1956: converse follows from the description of $\pi$ given in Section

1957: \ref{tensor} and the fact that $r$ is the period of $\gamma$.

1958: \end{proof}

1959: 

1960: \noindent

1961: 

1962: Consider a necklace $\nu=[a_1\ldots

1963:   a_n]$. Relations (\ref{lindep}) show that the subspace $V_\nu:=\C({a_1}\ldots

1964:   {a_n})_c\subset C^0_R(A)$ depends only on $\nu$, while the proposition implies that $V_\nu=0$ if $\nu$ is null

1965:   and $V_\nu\approx \C$ otherwise.  This gives a direct sum

1966:   decomposition: \be C^0_R(A)=\oplus_{\nu \in N_\bullet({\cal Q})}{V_\nu}~~,\nn

1967:   \ee where $V_{u}:=\C\epsilon_u$ for the null paths $\epsilon_u$.

1968:   Accordingly, any element $f\in C^0_R(A)$ decomposes as: \be

1969:   f=\sum_{\nu \in N_\bullet({\cal Q})}{f(\nu)}~~,\nn \ee where all but a finite number

1970:   of the vectors $f(\nu)\in V_\nu$ vanish.  If $l(\nu)=n>0$, then: \be

1971:   f(\nu)=\sum_{a_1a_2\dots a_n\in \nu}f_{a_1a_2\ldots

1972:   a_n}({a_1}\ldots {a_n})_c~~,\nn \ee where the coefficients are

1973:   defined as in (\ref{fc_expansion}). Since $f_{a_1\ldots a_n}$ are

1974:   cyclic, all vectors $f_{a_1\ldots a_n}({a_1}\ldots {a_n})_c\in

1975:   C^0_R(A)$ for which the cycle $a_1\ldots a_n$ belongs to a given

1976:   necklace $\nu$ are equal. Thus: \be\nn f(\nu)={\bar f}_{a_1\ldots

1977:   a_n}({a_1}\ldots {a_n})_c~~~{\rm for~any~fixed~cycle~}~~a_1\dots a_n

1978:   \in \nu~~.  \ee (in this relation, no summation over $a_j$ is

1979:   implied).

1980: 

1981: In view of these observations, the coefficients in the expansion

1982: (\ref{fc_expansion}) associated to null necklaces are not defined.

1983: For simplicity, we will define all such coefficients to be zero. Thus

1984: all coefficients associated with words $a_1 \ldots a_n$ which fail to

1985: correspond to a cycle or correspond to a cycle in a null necklace are

1986: set to zero by definition. We will use this convention repeatedly in

1987: what follows.

1988: 

1989: 

1990: \paragraph{Observation}

1991: $C^0_R(A)$ can be identified with the vector space $\C^{N_{\cal Q}}=\oplus_{\nu

1992: \in N_{\cal Q}}\C\nu $ generated by the set of non-null necklaces $N_{\cal Q}$.  For

1993: this, pick an enumeration of ${\cal Q}_1$, let $c_\nu:=\min \nu$ be the

1994: minimal representative of $\nu$ with respect to the induced

1995: lexicographic order on the set of paths of ${\cal Q}$, and identify $V_\nu$ with $\C$

1996: by sending $\pi(c_\nu)$ into the complex unit.  With this

1997: identification, we have $f(\nu)\equiv {\bar

1998: f}_{c_\nu}$ and $f\equiv \sum_{\nu \in N_\bullet({\cal Q})} {\bar f}_{c_\nu}

1999: \nu$.  Notice that such an identification requires that we pick an enumeration of

2000: ${\cal Q}_1$.

2001: 

2002: 

2003: \subsection{Cyclic derivatives and loop partial derivatives}

2004: \label{cycloop}

2005: 

2006: The isomorphism $\Omega^1_RA\approx A\otimes V$ of Section \ref{tensor} 

2007: shows that any one-form $w\in \Omega_R^1 A$ has well-defined coefficients 

2008: $w_a\in A$ determined by:

2009: \be

2010: \nn

2011: w=\sum_{a\in {\cal Q}_1}{(w_a d a)_c}=\sum_{a\in {\cal Q}_1}{ (-1)^{{\tilde a}({\tilde w}+1)}((d a) w_a)_c}~~,

2012: \ee

2013: where we noticed that ${\tilde w}_a={\tilde w}+{\tilde a}$. Inserting

2014: idempotents shows that each $w_a\in A$ must be a linear combination of

2015: paths starting at $h(a)$ and ending at $t(a)$.

2016: 

2017: We define the {\em left and right quiver cyclic derivatives} $\ld_a f, f\rd_a\in A$ of 

2018: an element $f\in C^0_R(A)$ through:

2019: \be

2020: \label{df}

2021: {\bar d} f=\sum_{a} ((f\rd_a)d a)_c =\sum_{a} (d a (\ld_a f))_c~~.

2022: \ee

2023: This gives linear maps $\ld_a, \rd_a: C^0_R(A)\rightarrow A$, satisfying $f\rd_a=

2024: (-1)^{{\tilde a}({\tilde f}+1)}\ld_a f$. 

2025: Equation (\ref{df}) gives: 

2026: \begin{eqnarray}

2027: \ld_a f&=&\sum_{n} {\bar f}_{a a_1\ldots a_n}a_1\ldots a_n ~~\nn\\

2028: f\rd_a &=&\sum_{n} {\bar f}_{a_1\ldots a_n a }a_1\ldots a_n~~.\nn

2029: \end{eqnarray}

2030: Notice that our convention (\ref{pairing}) is crucial for these simple formulas. 

2031: In particular, for any cycle $\gamma=a_1\ldots a_n$ of ${\cal Q}$: 

2032: \begin{eqnarray}

2033: \label{cd_props_general}

2034: \ld_a({a_1}\ldots {a_n})_c &=&\sum_{i=1}^n{(-1)^{({\tilde a}_1+\ldots

2035:     +{\tilde a}_{i-1})({\tilde a}+{\tilde

2036:       a}_{i+1}+\ldots +{\tilde a}_{n})}\delta_{a,a_i}{a_{i+1}}\ldots

2037:   {a_n} {a_1}\ldots {a_{i-1}}}\\

2038: ({a_1}\ldots {a_n})_c\rd_a &=&\sum_{i=1}^n{(-1)^{({\tilde a}+{\tilde a}_1+\ldots +{\tilde a}_{i-1})({\tilde

2039:       a}_{i+1}+\ldots +{\tilde a}_{n})}\delta_{a,a_i}{a_{i+1}}\ldots

2040:   {a_n} {a_1}\ldots {a_{i-1}}}~~.\nn

2041: \end{eqnarray}

2042: These equations imply:

2043: \begin{eqnarray}

2044: \label{cd_proj}

2045: \pi(\ld_a({a_1}\ldots {a_n})_c) &=&\sum_{i=1}^n(-1)^{{\tilde a}({\tilde a}_1+\ldots

2046:     +{\tilde a}_{i-1})}\delta_{a,a_i}(a_1\ldots a_{i-1}a_{i+1}\ldots a_n)_c\\

2047: \pi(({a_1}\ldots {a_n})_c \rd_a) &=&\sum_{i=1}^n (-1)^{{\tilde a}({\tilde a}_{i+1}+\ldots

2048:     +{\tilde a}_n)}\delta_{a,a_i}(a_1\ldots a_{i-1}a_{i+1}\ldots a_n)_c~~.\nn

2049: \end{eqnarray}

2050: It is clear that all terms in right hand side vanish 

2051: unless $a_i$ is a loop.  Hence the

2052: linear operators $\pi\circ \ld_a$ and $\pi\circ \rd_a$ induced

2053: on $C^0_R(A)$ are non-trivial only when  $a$ is a loop.

2054: 

2055: 

2056: We also define {\em loop partial derivatives} $\lp_a,\rp_a$ if $a$ is

2057: a (non-trivial) {\em loop} of the quiver.  Given such a loop of

2058: ${\cal Q}$, the loop derivatives $\lp_a\in \Der_l(A)$ and $\rp_a\in

2059: \Der_r(A)$ are the unique $R$-linear left and right derivations of

2060: $A$ of degree ${\tilde a}$ such that:

2061: \be

2062: \label{lp_def}

2063: \lp_a b:=b\rp_a=\delta^b_a \epsilon_u ~~{\rm~for~all}~b\in {\cal Q}_1~~,

2064: \ee

2065: where $u=h(a)=t(a)$. 

2066: It is not hard to see that $\lp_a$ and $\rp_a$ are well-defined and $\lp_a f=(-1)^{{\tilde a}({\tilde f}+1)}f\rp_a$.

2067: It is clear from  (\ref{lp_def}) that loop partial derivatives supercommute:

2068: \be

2069: \nn

2070: [\lp_a,\lp_b]=[\rp_a,\rp_b]=0~~{\rm for~all~loops}~~a,b\in {\cal Q}_1.

2071: \ee

2072: 

2073: Let ${\cal Q}_1(u)$ be the subset of loops at the vertex $u$ and 

2074: let $a^*=\sum_{b\in {\cal Q}_1}{\omega_{ab}b}=\sum_{b\in {\cal Q}_1(u)}{\omega_{ab}b} \in A$  be the 

2075: conjugate element of the loop $a$ introduced in Subsection \ref{symplectic_structure}. 

2076: A simple computation gives the relation: 

2077: \be

2078: \nn

2079: {\bar i}_{\lp_a}\omega={\bar d}\pi(a^*)~~,

2080: \ee

2081: which shows that $\lp_a$ is the symplectic relative left derivation

2082: with Hamiltonian $\pi(a^*)\in C^0_R(A)$. Similarly, one can

2083: view $\rp_a$ as the symplectic relative right derivation with Hamiltonian

2084: $\pi(a^*)$. Notice that the necklace of a loop $a$ has only one

2085: representative, so one can identify loops with their projections

2086: through $\pi$; we will sometimes do so in what follows.

2087: 

2088: Let $\partial^r_a$ and $\partial_a^l$ be the complex-linear maps

2089: induced by $\lp_a$ and $\rp_a$ on $C_R^0(A)=A/[A,A]$.  For any

2090: cycle $\gamma=a_1\ldots a_n$, we have:

2091: \begin{eqnarray}

2092: \partial^l_a\gamma_c&=&\pi(\lp_a \gamma)=\sum_{i=1}^n(-1)^{{\tilde a}({\tilde a}_1+\ldots

2093:     +{\tilde a}_{i-1})}\delta_{a,a_i}(a_1\ldots a_{i-1}a_{i+1}\ldots a_n)_c~~\nn\\

2094: \gamma_c\partial^r_a &=&\pi(\gamma\rp_a)=\sum_{i=1}^n(-1)^{{\tilde a}({\tilde

2095:       a}_{i+1}+\ldots +{\tilde a}_{n})}\delta_{a,a_i}(a_1\ldots a_{i-1}a_{i+1}\ldots a_n)_c~~.\nn

2096: \end{eqnarray}

2097: Comparing with (\ref{cd_proj}) gives: 

2098: \be

2099: \nn

2100: \partial^l_a=\pi\circ \ld_a~~,~~\partial^r_a=\pi\circ \rd_a~~.

2101: \ee

2102: Hence the cyclic and loop partial derivatives induce the same complex-linear operators on $C^0_R(A)$. 

2103: 

2104: Given an element $x\in E[1]$, we expand $x=\sum_{a\in {\cal Q}_1}{x^a e_a}$ with

2105: $x^a\in \C$ and define the (relative) left cyclic derivative along $x$

2106: via:

2107: \be

2108: \nn

2109: \ld_x f:=\sum_{a\in {\cal Q}_1}{x^a\ld_a f}~~~~\forall f\in C^0_R(A)~~.

2110: \ee

2111: If $x$ is central in $E$, then $x^a$ vanish unless $a$ is a loop. In

2112: this case, we define the (relative) loop left partial derivative along

2113: $x$ via:

2114: \be

2115: \nn

2116: \lp_x f=\sum_{a=loop}{x^a\lp_a f}~~.

2117: \ee

2118: For a central element, these two notions induce the same map on $C^0_R(A)$:

2119: \be

2120: \nn

2121: \partial^l_x f=\sum_{a=loop}{x^a\partial^l_a f}~~.

2122: \ee

2123: These definitions are well-behaved with respect to changes of adapted bases. 

2124: It is clear that $\partial_{e_a}^l=\partial_a^l$ etc. 

2125: 

2126: \paragraph{Observations}

2127: 

2128: (1) Loop partial derivatives and quiver cyclic derivatives are related to certain double 

2129: derivations introduced in \cite{Bergh}. Consider 

2130: the $A$-superbimodule $A\otimes_\C A$, where we use the so-called outer superbimodule structure: 

2131: \be

2132: \nn

2133: \alpha (a\otimes b)\beta:=(\alpha a)\otimes (b\beta)~~~~\forall \alpha,\beta, a, b \in A~~. 

2134: \ee

2135: Since $R$ is a subalgebra of $A$, this is also an $R$-superbimodule. A {\rm relative double derivation} of $A$ 

2136: is an $R$-linear derivation of the $A$-superbimodule $A\otimes A$. 

2137: As in \cite{Bergh}, consider the double left derivations determined by: 

2138: \be

2139: \nn

2140: {\bf D}_a(b)=\epsilon_{t(a)}\otimes \epsilon_{h(a)}\delta^b_a~~~~~\forall a,b\in {\cal Q}_1~~.

2141: \ee

2142: Then the loop partial derivatives can be recovered as:

2143: \be

2144: \nn

2145: \lp_a={\bf m}\circ {\bf D}_a~~{\rm for~~}a{\rm ~~a~loop}

2146: \ee

2147: where ${\bf m}:A\otimes A\rightarrow A$ is the $A$-superbimodule morphism: 

2148: \be

2149: \nn

2150: {\bf m}(a\otimes b)=ab~~.

2151: \ee

2152: Similarly, relations (\ref{cd_props_general}) show that the cyclic

2153: derivatives are induced by the maps $m\circ {\bf D}_a$, where

2154: $m:A\otimes A\rightarrow A$ is the linear map ({\em not} a bimodule

2155: morphism !): \be \nn m(a\otimes b)=(-1)^{{\tilde a}{\tilde b}}ba~~.

2156: \ee Indeed, it is not hard to see that $m\circ {\bf D}_a$ vanishes on

2157: $[A,A]$, so it induces a map from $C^0_R(A)$ to $A$ which coincides

2158: with the cyclic derivative $\ld_a$.

2159: 

2160: (2) Let ${\cal Q}$ be a quiver with a single vertex $u$. Then all

2161: arrows are loops and $A$ is the free superalgebra $\C\langle

2162: \{a\}\rangle$ generated by these loops.  In this case, $\lp_a$ and

2163: $\rp_a$ reduce to the standard left and right partial derivatives of

2164: the free algebra $A$. Moreover, the observations above show that

2165: $\ld_a$ are a superized version of the objects considered in

2166: \cite{RSS, Voiculescu}. This justifies our terminology.

2167: 

2168: 

2169: \subsection{Description of one-forms and closed two-forms}

2170: \label{12forms}

2171: 

2172: Consider the reduced tensor algebra $A_{\geq 1}:=V\otimes_R

2173: A=T_R^{\geq 1} V=\oplus_{n\geq 1}{V^{\otimes_R n}}$, which coincides

2174: with the subspace of $A$ spanned by all paths of length at least one. As in \cite{Ginzburg,LB_quivers} 

2175: (see also \cite{Lazarev}), it is easy to see that the super-vector space $C^1_R(A)$ is isomorphic

2176: with the center $A_{\geq 1}^R=(V\otimes_R A)^R$, the space spanned by the non-trivial cycles of the quiver. 

2177: The isomorphism $\Xi:A_{\geq 1}^R\rightarrow C^1_R(A)$ has the form: 

2178: \be

2179: \label{Xi}

2180: \sum_{i}{x_i\otimes_R f_i}\stackrel{\Xi}{\rightarrow} \sum_{i}{(dx_i f_i)_c}\in C^1_R(A)~~.

2181: \ee

2182: This follows from the observation that any element of $C^1_R(A)$ can be

2183: written uniquely as: 

2184: \be

2185: \nn

2186: w=\sum_{n\geq 0}f_{a a_1\ldots a_n}(da a_1\ldots a_n )_c=\sum_{a}(da f_a)_c=\Xi(\sum_a af_a)~~,

2187: \ee

2188: where $f_a:=\sum_{n\geq 0}f_{a a_1\ldots a_n }a_1\ldots a_n$ and 

2189: the complex coefficients $f_{a_1\ldots a_n}$ are taken to vanish unless $a_1\ldots a_n$ is 

2190: a cycle. Then $\Xi^{-1}(w)=\sum_{n\geq 0} f_{a_1\ldots a_n}a_1\ldots a_n:=f=\sum_a af_a$. 

2191: For $g=\sum_{n\geq 0}g_{a_1\ldots a_n} a_1\ldots a_n \in A^R_{\geq 1}$, we have 

2192: ${\bar d}\pi(g)={\bar d}g_c=(da\ld_a g_c)_c=

2193: \sum_{n\geq 0} {\bar g}_{(a a_1\ldots a_n)}(da a_1\ldots a_n)_c$, where 

2194: ${\bar g}_{(a_1\ldots a_n)}:=n g_{(a_1\ldots a_n)}$ are the strict coefficients of $g_c$. 

2195: Thus $w={\bar d}(g)_c$ iff 

2196: $f_{a_1\ldots a_n}={\bar g}_{(a_1\ldots a_n)}$. It is clear from these observations that 

2197: $w$ is exact iff its coefficients $f_{a_1\ldots a_n}$ are graded-cyclic; in this case, we 

2198: can take $g=\frac{1}{n} f_{a_1\ldots a_n}a_1\ldots a_n$ and we have $f_a=\ld_a g$. 

2199: 

2200: Thus $\Xi(f)\in C^1_R(A)_{\rm closed}= C^1_R(A)_{\rm exact}$. We let $A_{\geq 1}^c$ be the {\em

2201: cyclic subspace} of $A_{\geq 1}^R$, i.e. the subspace 

2202: consisting of elements with graded-cyclic coefficients. The space $A_{\geq 1}^c$

2203: consists of linear combinations of non-trivial cycles, such

2204: that cycles belonging to the same necklace of length $n$ appear with

2205: coefficients related by the action of $\Z_n$. This subspace provides

2206: an embedding of $C^0_R(A)/R$ into $A_{\geq 1}$. In fact, the projection

2207: $\pi:A\rightarrow C^0_R(A)$ induces an isomorphism $A_{\geq

2208: 1}^c\approx C^0_R(A)/R=A_{\geq 1}/[A,A]$ (recall that

2209: $[A,A]_0=[R,R]=0$ so $[A,A]\subset A_{\geq 1}$). Thus we have a vector

2210: space decomposition $A_{\geq 1}=[A,A]\oplus A_{\geq 1}^c$, as well as 

2211: $A_{\geq 1}^R=[A,A]^R\oplus A_{\geq 1}^c$. In particular, the subspace $A_{\geq 1}^c$ gives a 

2212: natural complement of $[A,A]^R$ inside $A_{\geq 1}^R$.

2213: 

2214: 

2215: Since the Karoubi complex is acyclic in positive degrees, we have $C^2_R (A)_{\rm closed}=

2216: {\bar d}(C^1_R (A))$ and the isomorphism $\Xi$ shows that any closed 

2217: two-form can be written as: 

2218: \be

2219: \label{dXi}

2220: u=-\sum_{n\geq 0}{(da df_a)_c}={\bar d}\Xi(f)~~.

2221: \ee

2222: where $f=af_a\in A_{\geq 1}^R$ is a combination of cycles of 

2223: length at least one. Since $\Xi(A_{\geq 1}^c)=C^1_R(A)_{\rm closed}$, the two-form (\ref{dXi}) 

2224: vanishes precisely when $f$ belongs to the cyclic subspace $A_{\geq 1}^c$. Thus $\ker(d\Xi)=A_{\geq 1}^c$, 

2225: and the map $d\Xi$ induces an isomorphism:

2226: \be

2227: \label{kappa_gen}

2228: \kappa:[A,A]^R\stackrel{\approx}{\rightarrow} C^2_R (A)_{\rm closed}

2229: \ee

2230: between the complement $[A,A]^R$ of this subspace in $A_{\geq 1}^R$ and the space 

2231: of closed two forms\footnote{This result is also discussed in \cite{VG}. I thank V. Ginzburg for pointing this 

2232: out.}.

2233: 

2234: 

2235: 

2236: 

2237: \subsection{Quiver description of the Kontsevich bracket}

2238: 

2239: Consider the non-commutative symplectic form (\ref{omega_form}) on $A$.

2240: For $\theta\in \Der_l(A)$, we set $\theta(a):=\theta^a\in A$. $R$-linearity of $\theta$ implies 

2241: that $\theta^a$ is a linear combination of paths which start at $t(a)$ and end at $h(a)$.

2242: Equation (\ref{df}) gives:  

2243: \be

2244: \nn

2245: {\bar L}_\theta f={\bar i}_\theta {\bar d} f=

2246: \pi(i_\theta\sum_{a\in {\cal Q}_1} (da) \ld_a f)=\sum_{a\in {\cal Q}_1}{(\theta(a )\ld_a f)_c}~~.

2247: \ee

2248: Thus: 

2249: \be

2250: \label{dftheta}

2251: {\bar L}_\theta f=\sum_{a\in {\cal Q}_1}(\theta^a \ld_a f)_c~~~~\forall \theta\in \Der_l(A)~~.

2252: \ee

2253: If $\theta$ is homogeneous of degree ${\tilde \theta}$, we set: 

2254: \be

2255: \theta_a:=\theta^b\omega_{ba}~~.\nn

2256: \ee

2257: Notice that $\tilde{\theta_a}={\tilde a}+{\tilde \omega}+{\tilde \theta}$. Expanding 

2258: $\theta^a=\sum_{n\geq 0}\theta_{a_1\ldots a_n}\,\,^a {a_1}\ldots {a_n}$ with $\theta_{a_1\ldots a_n}\,\,^a\in \C$, 

2259: we find  $\theta_a=\sum_{n\geq 0}\theta_{a_1\ldots a_n a}{a_1}\ldots {a_n}$, where: 

2260: \be

2261: \nn

2262: \theta_{a_1\ldots a_na}=\theta_{a_1\ldots a_n}\,\,^b\omega_{ba}~~.

2263: \ee

2264: As usual,  $\theta_{a_1\ldots a_n}\,\,^a$ are taken to vanish  unless $a_1\ldots a_n$ is a path. Also notice 

2265: that  $\theta_{a_1\ldots a_n}\,\,^a$ vanishes automatically unless this path starts at $t(a)$ and ends at $h(a)$.

2266: Similarly, $\theta_{a_1\ldots a_n}$ vanishes unless $a_1\ldots a_n$ is a cycle of ${\cal Q}$. 

2267: 

2268: An easy computation gives:

2269: \be

2270: \label{hash}

2271: {\bar i}_\theta \omega=(\theta_a da)_c=\sum_{n\geq 0}\theta_{a_1\ldots a_n a}({a_1}\ldots {a_n}da)_c~~.

2272: \ee

2273: Given $f\in C^0_R(A)$, we have ${\bar d}f=(f\rd_a

2274: da)_c$. Comparing with (\ref{hash}) gives $(\theta_f)_a=f\rd_a$, where

2275: $\theta_f$ is the Hamiltonian vector field of $f$.  Hence the map

2276: $\psi_\omega:C^0_R(A)\rightarrow \Der_l(A)$ of (\ref{ex}) is

2277: given by:

2278: \be

2279: \label{phiomega}

2280: \theta_f(a):=\theta_f^a=f\rd_b\omega^{ba}=\sum_{n\geq 0}{\bar f}_{a_1\ldots a_n b}\omega^{ba} s^{a_1}\ldots s^{a_n}~~,

2281: \ee

2282: where ${\bar f}_{a_1\ldots a_n}$ are the strict 

2283: coefficients of $f$. This allows us to write the Kontsevich bracket in more familiar form. 

2284: 

2285: \paragraph{\bf Proposition} 

2286: \label{coord_bracket}

2287: We have $\{f,g\}= (f\rd_a\omega^{ab}\ld_b g)_c$ for all $f,g\in

2288: C^0_R(A)$. 

2289: 

2290: \

2291: 

2292: \begin{proof}

2293: 

2294: Using (\ref{bL}), (\ref{dftheta}) and (\ref{phiomega}), we compute

2295: $\{f,g\}={\bar L}_{\theta_f}(g)=(\theta_f^a\ld_a

2296: g)_c=(f\rd_b\omega^{ba}\ld_a g)_c$.

2297: 

2298: \end{proof}

2299: 

2300: 

2301: \subsection{Some canonical forms and coefficient expressions}

2302: \label{canforms}

2303: 

2304: In this subsection, we  give some expressions which are useful in applications. 

2305: Let $W\in C^0_R(A)$ be an element of degree ${\tilde \omega}+1$. As we will see in 

2306: the next section, the boundary generating function of a topological D-brane system 

2307: is such an element.

2308: 

2309: For ${\tilde \omega}=0$, let us choose an adapted basis as in (\ref{e_can}). 

2310: Then $W$ is odd, and one finds:

2311: \be

2312: \nn

2313: \{W, W\}=(W\rd_{p_i}\ld_{q_i}W-W\rd_{q_i}\ld_{p_i}W)_c+(W\rd_{\xi_\alpha}\ld_{\xi_\alpha}W)_c=

2314: 2(W\rd_{p_i}\ld_{q_i}W)_c+(W\rd_{\xi_\alpha}\ld_{\xi_\alpha}W)_c

2315: \ee

2316: since $(W\rd_{p_i}\ld_{q_i}W)_c=-(W\rd_{q_i}\ld_{p_i}W)_c$. Also notice that $W\rd_{\xi_\alpha}=\ld_{\xi_\alpha} W$. 

2317: 

2318: Now let ${\tilde \omega}=1$ and choose an adapted basis as in (\ref{o_can}). 

2319: Then $W$ is even and we have:

2320: \be

2321: \nn

2322: \{W, W\}=(W\rd_{p_i}\ld_{q_i}W-W\rd_{q_i}\ld_{p_i}W)_c=2(W\rd_{p_i}\ld_{q_i}W)_c~~,

2323: \ee

2324: since again $(W\rd_{p_i}\ld_{q_i}W)_c=-(W\rd_{q_i}\ld_{p_i}W)_c$. 

2325: 

2326: One can also extract the coefficient expression of the cyclic bracket by

2327: direct computation. The case relevant for us is as follows.  For $W$ as above, notice that $\rho^{ab}$ can be used to

2328: raise and lower indices `from the left':

2329: \begin{eqnarray}

2330: \label{lifts}

2331: {\bar W}_{a_1\ldots a_{i-1}} {\,\, }^a_{\,\,a_{i+1}\ldots

2332: a_{n}}&:=&\rho^{ab}{\bar W}_{a_1\ldots a_{i-1}b a_{i+1}\ldots

2333: a_n}~~.\nn\\ {\bar W}_{a_1\ldots a_{i-1}b a_{i+1}\ldots

2334: a_n}&:=&\rho_{ab}{\bar W}_{a_1\ldots a_{i-1}} {\,\, }^b_{\,\,

2335: a_{i+1}\ldots a_{n}}\nn~~.

2336: \end{eqnarray}

2337: Then it is shown in Appendix \ref{coeffs} that that bracket of $W$ with itself takes the form: 

2338: \be

2339: \frac{1}{2}\{W,W\}=\frac{1}{2}{\bar W}_a{\bar W}^a+

2340: \sum_{n\geq 1}\frac{1}{n}\left(\sum_{0\leq i+j\leq n}(-1)^{{\tilde a_1}+\ldots +{\tilde a}_i}

2341: {\bar W}_{a_1\ldots a_i a a_{i+j+1}\ldots a_n} {\bar W}^a_{\,\,a_{i+1}\ldots

2342:       a_{i+j}}\right)({a_1}\ldots {a_n})_c~~,\nn

2343: \ee

2344: which is valid irrespective of the degree of $\omega$. 

2345: 

2346: 

2347: \section{Geometry of finite D-brane systems}

2348: \label{geometrization}

2349: 

2350: Consider a finite topological D-brane system with total boundary space

2351: $E$ and boundary algebra $R$.  As before, we let $V=E[1]^{\rm v}$ and

2352: consider the tensor algebra $A=T_R V$. As explained in Section

2353: \ref{algebraic}, the data of all integrated boundary correlators on

2354: the disk is encoded by an $R$-superbimodule structure on $E$, together with

2355: a cyclic and unital weak $A_\infty$ structure on this

2356: superbimodule. We will use the machinery developed in the previous two

2357: sections to encode this into a `noncommutative

2358: generating function' $W\in C^0_R(A)$ subject to simple constraints. To

2359: this end, we pick an adapted basis of $E$ and let ${\cal Q}$ be its

2360: index superquiver.

2361: 

2362: 

2363: 

2364: 

2365: \subsection{Geometric description of cyclic weak $A_\infty$ structures}

2366: 

2367: It turns out that a weak $A_\infty$

2368: structure on the $R$-superbimodule $E$ is the same as an {\em odd}

2369: relative derivation $Q$ of the tensor algebra $A$.

2370: With our conventions, the relation is as follows.  Picking adapted

2371: coordinates, we define the coefficients of $Q$ through:

2372: \be

2373: \label{Q_exp}

2374: Q(a)=\sum_{n\geq 0} Q_{a_1\ldots a_n}{\,}^a a_1\ldots a_n

2375: \ee

2376: and construct odd linear maps $r_n:E[1]^{\otimes n}\rightarrow E[1]$ via: 

2377: \be

2378: \label{rQ}

2379: r(e_{a_1}\ldots e_{a_n})=Q_{a_1\ldots a_n}{\,}^a e_a~~. 

2380: \ee

2381: Thus $Q(s^a)=\sum_{n\geq 0} s^a(r(e_{a_1}\ldots e_{a_n})) s^{a_1}\otimes_R \ldots \otimes_R s^{a_n}$.

2382: Thinking in terms of arrows, it is clear that $r_n$ are $R$-multilinear.

2383: Since $Q$ is odd, we have $[Q,Q]= 2 Q^2$, which implies that $Q^2$ is

2384: a derivation of $A$. Since $a$ generate the algebra, this means that the condition $Q^2=0$ is equivalent with

2385: $Q^2(a)=0$ for all $a$. Using expansion (\ref{Q_exp}), one finds

2386: that this amounts to the relations:

2387: \be

2388: \nn

2389: \sum_{0\leq i+j\leq n}(-1)^{{\tilde a_1}+\ldots +{\tilde a}_i}

2390: Q_{a_1\ldots a_i b a_{i+j+1}\ldots a_n}{\,}^a Q_{\,\,a_{i+1}\ldots

2391:       a_{i+j}}{\,}^b=0~~{\rm for~all~}~~ n\geq 0~~,

2392: \ee

2393: which are the $A_\infty$ constraints (\ref{ainf}).

2394: 

2395: It is also not hard to check that the nilpotent derivation $Q$ is

2396: symplectic iff the associated $A_\infty$ structure is cyclic.  An easy

2397: way to see this is as follows. By the exact sequence (\ref{ex}), we

2398: have that $Q$ is symplectic iff it is Hamiltonian, which via equation

2399: (\ref{phiomega}) amounts to the existence of a $W\in C^0_R(A)$ such

2400: that:

2401: \be

2402: Q_{a_1\ldots a_n}{\,}^a={\bar W}_{a_1\ldots a_nb}\omega^{ba}~~

2403: \ee

2404: or, equivalently:

2405: \be

2406: \label{QW}

2407: Q_{a_1\ldots a_n}={\bar W}_{a_1\ldots a_n}~~.

2408: \ee

2409: As usual, we have set $Q_{a_1\ldots a_n a}:=Q_{a_1\ldots a_n}{\,}^b\omega_{ba}$. 

2410: It is clear that a $W$ exists if and only if 

2411: the coefficients $Q_{a_1\ldots a_n}$ are cyclic. We have:

2412: \begin{eqnarray}

2413: \label{cyc_arg}

2414: \rho(e_{a_0},r_n(e_{a_1}\ldots e_{a_n}))&=&(-1)^{{\tilde a}_0}\omega(e_{a_0},

2415: r_n(e_{a_1}\ldots e_{a_n}))=(-1)^{1+{\tilde a}_0{\tilde \omega}}

2416: \omega(r_n(e_{a_1}\ldots e_{a_n}), e_{a_0})\nn\\

2417: &=&(-1)^{{\tilde a}_0 {\tilde \omega}}{\bar W}_{a_1\ldots a_n a_0}={\bar W}_{a_0\ldots a_n}~~,

2418: \end{eqnarray}

2419: where we used the superselection rules for $\omega$ and $W$.

2420: Thus:

2421: \be

2422: \label{QW1}

2423: \rho(e_{a_0},r_n(e_{a_1}\ldots e_{a_n}))={\bar W}_{a_0a_1\ldots a_n}~~,

2424: \ee

2425: and we see that $L_Q \omega=0$ implies that the left hand side is

2426: cyclic, which is the cyclicity constraint (\ref{rrcyc}).  Conversely,

2427: if the LHS is cyclic then we define $W$ through equation

2428: (\ref{QW1}). Then relations (\ref{cyc_arg}) show that $Q_{a_1\ldots

2429: a_n a_0}={\bar W}_{a_1\ldots a_n a_0}$, i.e. $Q$ is symplectic with

2430: Hamiltonian $W$.  Combining everything and  noticing that ${\tilde W}={\tilde \omega}+1$ (because

2431: $Q$ is odd), we have:

2432: 

2433: \

2434: 

2435: {\em Giving a cyclic weak $A_\infty$ structure on $E$ amounts to giving an element $W\in C^0_R(A)$, 

2436: of degree ${\tilde \omega}+1$,  such that $\{W,W\}=0$.}

2437: 

2438: \

2439: 

2440: 

2441: \paragraph{Observation} The triplet $(A,Q,\omega_{form})$ can be viewed as a noncommutative generalization 

2442: of the so-called $QP$-manifolds of \cite{Konts_Schwarz}, while the doublet $(A,Q)$ generalizes the concept of 

2443: $Q$-manifold discussed in the same paper (notice, though, that we consider both even and odd symplectic forms, 

2444: so we generalize the work of \cite{Konts_Schwarz} in two directions). It was shown in  \cite{Konts_Schwarz} that 

2445: $QP$-manifolds give the general geometric setting of the classical BV-formalism. 

2446: Accordingly, for odd symplectic forms, the 

2447: triplet $(A,Q,\omega_{form})$ defines a noncommutative version of that formalism. 

2448: 

2449: 

2450: 

2451: 

2452: \subsection{The unitality constraint}

2453: \label{sec:unitality}

2454: 

2455: We saw that a cyclic weak $A_\infty$ structure on $E$ is the same as

2456: an $R$-linear symplectic derivation $Q\in \Der_l^\omega(A)$ such that

2457: $Q^2=0$.  We let $W$ be the {\em canonical} Hamiltonian of $Q$, i.e. that Hamiltonian which vanishes at zero 

2458: (see Section \ref{tensor}). Explicitly, equations (\ref{QW}) and (\ref{QW1}) give: 

2459: \be

2460: \nn

2461: W=\sum_{n\geq 0}{\frac{1}{n+1}\rho(e_{a_0}, r_n(e_{a_1}\ldots e_{a_n}))(s^{a_0}\ldots s^{a_n}})_c~~,

2462: \ee

2463: which allows us to reconstruct $r_n$ from $W$ provided that we know $\omega=\rho\circ \Sigma^{\otimes 2}$.

2464: The homological derivation $Q$ can be recovered as the Hamiltonian derivation defined by  $W$, 

2465: which amounts to relations (\ref{QW}). 

2466: 

2467: For a topological D-brane system, the underlying weak $A_\infty$ structure should be unital.

2468: To formulate this condition in terms of $W$, we write the unitality constraints (\ref{unitality}) as:

2469: \begin{eqnarray}

2470: \label{ueqs}

2471: r_n(e_{a_1}\ldots e_{a_{j-1}},\lambda,e_{a_{j+1}}\ldots e_{a_n})&=&0~~{\rm for~all}~~~~ n\neq 2 ~{\rm~and~all}~

2472: j=1\ldots n~~\nn\\

2473: -r_2(\lambda,e_{a})=(-1)^{{\tilde a}} r_2(e_{a},\lambda)&=&e_a~~,

2474: \end{eqnarray}

2475: where $\lambda$ is an odd central element of $E$. 

2476: Given $\lambda=\oplus_{u\in {\cal Q}_0}{\lambda_u}\in E[1]^R$ with $\lambda_u\in E_{uu}[1]$, we

2477: can choose adapted coordinates such that each $\lambda_u$ is one of the odd basis elements $\{e_a\}$. 

2478: We then let $\sigma_u\in E[1]^{\rm v}$ be the corresponding elements of the dual basis $\{s^a\}$ of $V$ (those 

2479: dual basis elements which satisfy $s^{\sigma_u}(e_a)=\delta_{e_a,\lambda_u}$). 

2480: It is clear that each $\sigma_u$ is an odd loop of the quiver starting and ending at the vertex $u$. With such 

2481: a choice of adapted basis, we have $\lambda_u=e_{\sigma_u}$ and the first row in (\ref{ueqs}) is equivalent with:

2482: \be

2483: \label{un1}

2484: W_{a_1\ldots a_n}=0 {\rm ~if~}n\neq 3{\rm ~and~any~of~the~arrows~}

2485: a_j{\rm ~coincides~with~any~of~the~loops}~\sigma_u~~

2486: \ee

2487: while the second row amounts to: 

2488: \be

2489: \label{un2}

2490: {\bar W}_{\sigma_u ab}=-\omega_{ab}\Longleftrightarrow   W_{\sigma_u ab}=-\frac{1}{3}\omega_{ab}

2491: ~~~{\rm~for~}t(a)=h(b)=u~~{\rm~and~~}h(a)=t(b)~~. 

2492: \ee

2493: Hence unitality of $(r_n)$ boils down to the requirement that

2494: the adapted basis $\{\psi_a\}$ can be chosen such that the vertices 

2495: the index quiver carry distinguished odd loops $\sigma_u$ satisfying

2496: (\ref{un1}) and (\ref{un2}).

2497: 

2498: The two conditions above say that $W$ takes the form:

2499: \be

2500: \label{W_decomp}

2501: W=W_g +W_d

2502: \ee

2503: where the `generic' contribution is given by:

2504: \be

2505: \label{Wg}

2506: W_g:=-\omega_{ab}(\sigma a b)_c=

2507: -\sum_{\tiny \begin{array}{c}a,b\in {\cal Q}_1, u\in {\cal Q}_0\\t(a)=h(b)=u\\h(a)=t(b)

2508: \end{array}}{\omega_{ab}(\sigma_u ab)_c}~~

2509: \ee

2510: while the `deformation part'  

2511: $W_d$ vanishes at zero and is independent of all $\sigma_u$. In the first form of the last expression, 

2512: we used Einstein summation over $a$ and $b$ and have set $\sigma=\sum_{u\in {\cal Q}_0}{\sigma_u}\in A_1$. 

2513: Notice the lack of a $1/3$ prefactor in (\ref{Wg}); 

2514: this is because we brought all terms to a form in which a $\sigma_u$ insertion appears in the first position.

2515: 

2516: To describe this more elegantly, notice\footnote{For an 

2517: element $f=\sum_{n\geq 0} f_{a_1\ldots a_n}(a_1\ldots a_n)_c\in C^0_R(A)$, the condition $\ld_a f=0$ amounts 

2518: to vanishing of all cyclic coefficients $f_{a_1\ldots a_n}$ for which one of the $a_j$ coincides with $a$. 

2519: Further, the cyclic derivative $\ld_{\lambda }W$ determines $\ld_{\sigma_u}W=\epsilon_u \ld_{\lambda} W$. 

2520: Thus (\ref{W_decomp}) and (\ref{Wg}) amount to $\ld_\lambda\left(W+\omega_{ab}(\sigma a b)_c\right)=0$, 

2521: which gives the desired statement.} 

2522: that (\ref{W_decomp}) together with (\ref{Wg}) amounts to the condition: 

2523: \be

2524: \nn

2525: \ld_\lambda W=-\sum_{a,b\in {\cal Q}_1}\omega_{ab} ab~~

2526: \ee

2527: where $\lambda:=\sum_{u\in {\cal Q}_0}{\lambda_u}$ and 

2528: $\ld_\lambda:=\sum_{u\in {\cal Q}_0} \ld_{\sigma_u} $ as in Subsection \ref{cycloop}. 

2529: Using the graded antisymmetry of $\omega_{ab}$, the last relation takes the form:

2530: \be

2531: \label{ldcond1}

2532: \ld_\lambda W=-\frac{1}{2}\sum_{a\in {\cal Q}_1}[a,a^*]\in [A,A]_2^R~~,

2533: \ee

2534: where we introduced the conjugate variables

2535: \be

2536: \label{adual_expansion}

2537: a^*:=\sum_{b\in {\cal Q}_1}\omega_{ab}b=

2538: \sum_{\tiny \begin{array}{c}b\in {\cal Q}_1 \\h(a)=t(b),  h(b)=t(a)\end{array}}{\omega_{ab}b}~~

2539: \ee 

2540: as in Subsection \ref{symplectic_structure}.  Here $[A,A]_2$ is the subspace of $[A,A]$ 

2541: consisting of elements of degree two with respect to the

2542: $\N$-grading, while $[A,A]_2^R\subset [A,A]^R$ is the centralizer of $[A,A]_2$ in $R$.  

2543: 

2544: Relation (\ref{ldcond1}) allows one reconstruct $\omega$ from $W$. To

2545: formulate this invariantly, remember from Subsection \ref{12forms}

2546: that the space of closed noncommutative two-forms $C^2_R

2547: (A)_{\rm closed}$ is isomorphic with $[A,A]^R$.  Restricting the map

2548: (\ref{kappa_gen}) to the subspace $C^2_R (A)_2 \subset C^2_R(A)_{\rm

2549: closed}$ of constant two-forms gives an isomorphism

2550: $[A,A]_2^R\stackrel{\approx}{\rightarrow} C^2_R (A)_2$ whose

2551: explicit form is given by relation (\ref{dXi}):

2552: \be

2553: \label{kappa_0}

2554: \sum_{a,b\in {\cal Q}_1}{f_{ab}ab}\stackrel{\kappa}{\rightarrow} -\sum_{a, b\in {\cal Q}_1}f_{ab}(dadb)_c~~.

2555: \ee

2556: Here $f=\sum_{a,b}f_{ab}ab$ is the general element of $[A,A]_2^R$, with graded-antisymmetric complex coefficients 

2557: $f_{ab}=(-1)^{1+{\tilde a}{\tilde b}}f_{ba}$, so we can also write $f=\frac{1}{2}\sum_{a,b}f_{ab}[a,b]$. 

2558: Applying this to the noncommutative symplectic form, we find:

2559: \be

2560: \kappa^{-1}(\omega)=-\frac{1}{2}\omega_{ab}ab=-\frac{1}{4}\sum_{a\in {\cal Q}_1}[a,a^*]~~.

2561: \ee

2562: Thus relation (\ref{ldcond1}) can be written as either of the following equivalent conditions: 

2563: \be

2564: \label{diff_unitality}

2565: \ld_\lambda W=2\kappa^{-1}(\omega)\Leftrightarrow \omega=\frac{1}{2}\kappa(\ld_\lambda W)~~.

2566: \ee

2567: These observations allow us to write the unitality constraint

2568: (\ref{ldcond1}) as condition (\ref{diff_unitality}). In

2569: particular, the element $\ld_\lambda W\in [A,A]_2^R $ must belong to the 

2570: subspace spanned by the $\kappa$-preimages of quiver symplectic forms.  To

2571: describe this space, notice that any element $\mu\in [A,A]_2^R$ can

2572: be expanded uniquely as:

2573: \be

2574: \mu=-\frac{1}{4}\sum_{a\in {\cal Q}_1}{[a,a^*]}~~,

2575: \ee

2576: where each $a^*$ is a linear combination of arrows going from $h(a)$

2577: to $t(a)$.  We say that $\mu$ is {\em non-degenerate} if the elements

2578: $(a^*)_{a\in {\cal Q}_1}$ form a basis of $V$; in this

2579: case, we can expand $a^*$ as in (\ref{adual_expansion}), with

2580: coefficients $\omega_{ab}$ forming the entries of a

2581: graded-antisymmetric non-degenerate matrix. Moreover, it is clear that $\mu$ has 

2582: $\Z_2$-degree ${\tilde \omega}$ iff this matrix satisfies the selection 

2583: rules (\ref{omega_sel}). We let  $\Mom_V\subset [A,A]_2^R$ 

2584: be the $\Z_2$-homogeneous subspace of non-degenerate elements in $[A,A]_2^R$ and 

2585: let $\Mom^0_V$ and $\Mom^1_V$ be its homogeneous components.

2586: The observations made above show that $\kappa$ induces an isomorphism between 

2587: the space $CNS^{\tilde \omega}(V)$ of constant 

2588: non commutative symplectic forms on $A$ having degree ${\tilde \omega}$ and the space 

2589: $\Mom^{\tilde \omega}_V$. It follows from Proposition 8.1.1 of \cite{BEV} that

2590: ${\rm Mom}_V$ is the space of noncommutative moment maps

2591: associated to quiver symplectic forms on the path algebra $A$.

2592: 

2593: \

2594: 

2595: \noindent We can now formulate the unitality criterion as follows: 

2596: 

2597: \paragraph{\bf Proposition}

2598: Let $E[1]$ be a symplectic $R$-superbimodule of finite complex

2599: dimension whose symplectic form has degree ${\tilde \omega}$, let 

2600: $A:=T_R E[1]^{\rm v}$ and let $W$ be an element of $C^0_R(A)$

2601: which has degree ${\tilde \omega}+1$ and vanishes at zero. Let $\omega$ be the noncommutative symplectic

2602: form induced on $A$ and assume that $\{W,W\}=0$, where $\{.,.\}$ is the Kontsevich bracket defined by

2603: $\omega$.  Then the following statements are equivalent:

2604: 

2605: (1) The cyclic weak $A_\infty$ structure determined by $W$ on $E[1]$

2606:     is unital

2607: 

2608: (2) There exists an odd central element $\lambda\in E^R$ such that

2609:     $\frac{1}{2}\ld_\lambda W=\kappa^{-1}(\omega)$.

2610: 

2611: \noindent In this case:

2612: 

2613: (a) $\Sigma \lambda$ is the unit of the $A_\infty$ structure.

2614: 

2615: (b) The element $\mu:=\frac{1}{2}\ld_\lambda W\in A$ belongs to the

2616: subspace $\Mom_V$ of $A$

2617: 

2618: (c) The non-commutative symplectic form can be recovered via the

2619: relation $\omega=\kappa (\mu)$.

2620: 

2621: (d) Let $\lambda=\sum_{u\in {\cal Q}_0}{\lambda_u}$ ($\lambda_u\in

2622: E_{uu}$) be the decomposition of $\lambda$, and choose an adapted

2623: basis $e_a$ of $E[1]$ containing $\lambda_u$ among the basis

2624: elements. Let $s^a\equiv a$ be the dual basis, and let $\sigma_u$ its

2625: elements associated with $\lambda_u$.  Then $\sigma_u$ correspond to odd

2626: loops of the associated superquiver and $W$ takes the form given in

2627: eqs. (\ref{W_decomp}) and (\ref{Wg}), where $W_d$ vanishes at zero and

2628: is independent of all $\sigma_u$.

2629: 

2630: \

2631: 

2632: \subsection{Noncommutative geometry of D-brane systems}

2633: 

2634: Combining the discussion of the previous subsections, we have the following non-commutative geometric description 

2635: of finite topological D-brane systems:

2636: 

2637: \

2638: 

2639: {\em Let $R$ be a finite-dimensional semisimple commutative algebra over $\C$ 

2640: and $E$ an $R$-superbimodule which is finite-dimensional over $\C$. 

2641: Giving a finite topological D-brane system with boundary

2642: decomposition described by ($R$, $E$) and topological metrics of  $\Z_2$-degree ${\tilde \omega}$ amounts to giving

2643: a `noncommutative function' $W\in C^0_R(A)$ on the tensor algebra $A=T_R E[1]^{\rm v}$ 

2644: and an odd central element $\lambda \in E^R$ such that

2645: the following conditions are satisfied:

2646: 

2647: \

2648: 

2649: \noindent (1) $W$ vanishes at zero and has $\Z_2$-degree ${\tilde \omega}+1$.

2650: 

2651: \noindent (2) The element $\mu:=\frac{1}{2}\ld_\lambda W$ belongs to $\Mom_V^{\tilde \omega}$ 

2652: 

2653: \noindent (3) We have $\{W,W\}=0$, where $\{.,.\}$ is the Kontsevich bracket determined on $C^0_R(A)$ by 

2654: the constant noncommutative symplectic form  $\omega:=\kappa (\mu)$. 

2655: }

2656: 

2657: \paragraph{Observations} 

2658: 

2659: (1) The coefficient expressions given in Section

2660:     \ref{canforms} show that equation $\{W,W\}=0$ is

2661:     equivalent with:

2662: \be

2663: \label{W_inf}

2664: \sum_{0\leq i+j\leq n}(-1)^{{\tilde a_1}+\ldots +{\tilde a}_i}

2665: {\bar W}_{a_1\ldots a_i a a_{i+j+1}\ldots a_n} {\bar W}^a_{\,\,a_{i+1}\ldots

2666:       a_{i+j}}=0~~{\rm for~all~}~~ n\geq 0~~,

2667: \ee

2668: where lifting of indices is done from the left  with $\rho^{ab}=(-1)^{\tilde

2669: b+1}\omega^{ab}$, and $\omega^{ab}$ is the inverse of the matrix $\omega_{ab}=-{\bar W}_{\sigma ab}$.  

2670: The first equation (for $n=0$) is ${\bar W}_a{\bar

2671: W}^a=0$.  Notice that $\omega$ is determined by $W$, so the equations are not quadratic. 

2672: This countable system of nonlinear algebraic conditions is

2673: a non-commutative analogue of the WDVV equations \cite{WDVV}.

2674: 

2675: (2) It was shown in \cite{HLL} that the background satisfies the string equations of motion iff 

2676: the underlying $A_\infty$ algebra is minimal. It is clear that 

2677: this amounts to the requirement that $W$ has order at least $3$ at the origin.

2678: 

2679: (3) The structure given above can be viewed as an 'off-shell extension' of the 'boundary part' of the data described in 

2680: \cite{CIL1, Moore_Segal, Moore}. The latter arises in the particular case when $W$  has degree three at the origin

2681: (i.e. the underlying $A_\infty$ structure is minimal), and can be recovered 

2682: by forgetting all terms of $W$ of order higher than $3$. 

2683: In physics language, the structure of \cite{CIL1, Moore_Segal, Moore} corresponds to keeping only the 

2684: boundary three-point functions on the disk, thereby forgetting all {\em integrated} amplitudes. As explained in the 

2685: introduction to \cite{CIL1}, this reflects the difference between two-dimensional topological field theory and 

2686: topological string theory, namely in the topological field theory one does not consider integration of amplitudes 

2687: over the moduli space of the underlying Riemann surface (since by definition 

2688: the worldsheet metric is not a dynamical variable).  

2689: 

2690: \subsection{Deformations of the underlying string theory}

2691: \label{theory_deformations}

2692: 

2693: Each solution of the constraints described in the previous subsection represents the tree-level boundary data of 

2694: an open topological string theory. We would like to make some basic observations about the space of such 

2695: theories.

2696: 

2697: It is instructive to consider the trivial approximation $W_d=0$ i.e. $W=W_g$, which --- as we shall see in 

2698: a moment --- is appropriate under certain assumptions. Starting from $W_g=-\omega_{ab}(\sigma ab)_c$, we 

2699: compute: 

2700: \be

2701: \label{rdWg}

2702: -W_g\rd_a=\omega_{\alpha a}\sigma \alpha+(-1)^{{\tilde \beta}{\tilde \omega}} \omega_{a\beta}\beta\sigma+

2703: (-1)^{\tilde \omega}\delta^\sigma_ a\omega_{\alpha\beta}\alpha\beta~~

2704: \ee

2705: and: 

2706: \be

2707: \label{ldWg}

2708: -\omega^{ab}\ld_b W_g=(-1)^{\tilde \omega}a\sigma +(-1)^{{\tilde a}+{\tilde \omega}}\sigma a+\omega^{a\sigma}

2709: \omega_{\gamma\delta}\gamma\delta~~.

2710: \ee

2711: Combining these equations and using appropriate cyclic permutations gives: 

2712: \be

2713: \nn

2714: \{W_g,W_g\}_c=(W_g\rd_a\omega^{ab}\ld_bW_g)_c=(-1)^{\tilde \omega}\omega^{\sigma\sigma}\omega_{\alpha\beta}

2715: \omega_{\gamma\delta}(\alpha\beta\gamma\delta)_c~~.

2716: \ee

2717: Since $\sigma$ is odd, $\omega^{\sigma\sigma}$ vanishes for degree reasons unless ${\tilde \omega}=0$. When 

2718: $\omega$ is even, the term in the right hand side need not vanish.  

2719: 

2720: Let us assume that ${\tilde \omega}=1$ or that ${\tilde \omega}=0$ but 

2721: $\omega^{\sigma\sigma}\omega_{\alpha\beta}\omega_{\gamma\delta}(\alpha\beta\gamma\delta)_c$ vanishes. 

2722: In this case, we have $\{W_g,W_g\}=0$ and $W_g$ gives a marked point in the space of open 

2723: string theories with underlying supermodule $E$, unit $\lambda$ and symplectic form $\omega$. The cyclic unital 

2724: $A_\infty$ algebra corresponding to this solution has a single product $r_2^g$, which is given by:

2725: \be

2726: \nn

2727: r_2^g(e_a,e_b)=(-1)^{{\tilde \omega}+1}\omega_{ab}\omega^{\lambda c}e_c~~{\rm for}~~a,b\neq \sigma

2728: \ee

2729: and by the unitality constraint $-r_2^g(\lambda,e_a)=(-1)^{{\tilde a}} r_2^g(e_a,\lambda)=e_a$ for the 

2730: remaining combinations of basis elements. The $A_\infty$ constraints (\ref{ainf}) reduce to: 

2731: \be

2732: \nn

2733: r_2^g(r_2^g(x,y),z)+(-1)^{{\tilde x}}r_2^g(x,r_2^g(y,z))=0~~,

2734: \ee

2735: which means that $m_g:=\Sigma\circ r_2\circ \Sigma^{\otimes 2}:E^{\otimes 2}\rightarrow E$ satisfies the associativity condition: 

2736: \be

2737: \nn

2738: m_g(m_g(x,y),z)=m_g(x,m_g(y,z))~~.

2739: \ee

2740: Moreover, the unitality constraint for $r_2^g$ amounts to $m_g(\lambda, x)=m_g(x,\lambda)=x$.  Hence the distinguished

2741: solution $W=W_g$ corresponds to an associative superalgebra

2742: structure on $E$, and the underlying $A_\infty$ category reduces to

2743: an ordinary (i.e. associative) $\Z_2$-graded category.  In the

2744: topological string theory, all integrated boundary correlators on the

2745: disk vanish and the entire information is contained in the boundary

2746: three-point functions. Fixing $\lambda$ and $\omega$, other string

2747: theories with the same units and topological metrics are given by

2748: solutions of the equations $\{W_g+W_d, W_g+W_d\}=0$ and $\ld_\lambda W_d=0$, the first of which reduces to:

2749: \be

2750: \nn

2751: \{W_g, W_d\}+\frac{1}{2}\{W_d,W_d\}=0~~,

2752: \ee

2753: where we used the graded antisymmetry property of the Kontsevich bracket. Letting $Q_g=\theta_{W_g}$ 

2754: be the (odd) Hamiltonian vector field defined by $W_g$ (i.e. $dW_g=i_{Q_g} \omega$), this equation takes the form:

2755: \be

2756: \label{MC}

2757: L_{Q_g}W_d+\frac{1}{2}\{W_d,W_d\}=0~~.

2758: \ee

2759: Notice that $L_{Q_g}$ squares to zero (since $L_{Q_g}^2=\frac{1}{2}[L_{Q_g}, L_{Q_g}]=\frac{1}{2}L_{[Q_g,Q_g]}=

2760: \frac{1}{2}L_{\theta_{\{W_g,W_g\}}}=0$) and that it acts as an odd derivation of the Lie superalgebra 

2761: $(C^0_R(A)[{\tilde \omega}],\{.,\})$ (due to the Jacobi identity for the Kontsevich bracket):

2762: \be

2763: \nn

2764: L_{Q_g}\{f,g\}=\{L_{Q_g}f, g\}+(-1)^{{\tilde f}+{\tilde \omega}}\{f, L_{Q_g} g\}~~.

2765: \ee 

2766: Hence $(C^0_R(A)[{\tilde \omega}], L_Q, \{.,.\})$ is a differential Lie superalgebra, and (\ref{MC}) is its 

2767: Maurer-Cartan equation. This means that one can study the moduli space of boundary string {\em theories}

2768: with fixed units and topological metrics by using the deformation theory of Lie superalgebras.

2769: 

2770: 

2771: \section{The noncommutative moduli space}

2772: 

2773: In this section, we use the formalism developed above to construct a

2774: noncommutative version of the extended moduli space of finite D-brane

2775: systems (the boundary part of the extended moduli space of topological

2776: strings).

2777: 

2778: \subsection{Symmetries}

2779: \label{automf_group}

2780: 

2781: Let us fix a D-brane system described by the noncommutative

2782: generating function $W$, with Hamiltonian derivation $Q$ and

2783: symplectic form $\omega$. We assume given adapted coordinates

2784: including odd loops $\sigma_u$ associated with the units of the

2785: underlying $A_\infty$ structure. It is clear from the categorical

2786: formulation of Appendix \ref{data} that a symmetry of the D-brane

2787: system amounts to an automorphism of the underlying cyclic and unital

2788: weak $A_\infty$ category, called a cyclic and unital

2789: $A_\infty$ automorphism (an automorphism is a strict autoequivalence, as 

2790: appropriate for a finite category). In this subsection, we describe such symmetries 

2791: as symplectomorphisms of $A$ which obey certain supplementary properties. 

2792: 

2793: Given a relative automorphism $\phi$ of $A$, we set

2794: $a':=\phi(a)$ for all $a\in {\cal Q}_1$, and let $V'\subset A$ be the $R$-sub-bimodule

2795: spanned by the elements $a'$.  Then $A$ is isomorphic {\em as a superalgebra} with the tensor

2796: algebra $T_R V'$, and $\phi$ can be viewed as a change of 

2797: coordinates from $a$ to $a'$.  More precisely, the restriction of $\phi$ to $A_1=V$ 

2798: gives an isomorphism of $R$-superbimodules $\phi_1:V\rightarrow V'$ and $\phi$ can be identified 

2799: with the isomorphism of bigraded $R$-superalgebras  

2800: $T_R(\phi)=\oplus_{n\geq 0}{\phi^{\otimes_R n}}:T_RV\rightarrow T_R V'$ induced by $\phi_1$. 

2801: Of course, we have $a'=\phi_1(a)=\phi(a)$ and $a'$ is an adapted basis for the $R$-superbimodule $V'$.

2802: 

2803: $R$-linearity of $\phi$ implies that

2804: each $a'$ is a linear combination of paths starting at $t(a)$ and

2805: ending at $h(a)$. However, $\phi$ need not be homogeneous with respect

2806: to the $\N$-grading of $A$, so generally each $\phi(a)$ can be a

2807: linear combination of paths of different length. Defining $A'_n$ to be

2808: the subspace spanned by $n$-factor monomials in $a'$, we have

2809: $A'_n\approx T_R^n V'$ and a new decomposition:

2810: \be

2811: \nn

2812: A=\oplus_{n\geq 0}{A'_n}~~

2813: \ee

2814: with $A'_0=R$. In particular, a generic $R$-superalgebra automorphism induces a change of $\N$-grading.

2815: 

2816: Equation (\ref{df}) implies:

2817: \be

2818: \nn

2819: {\bar d} {\bar \phi}(f)={\bar\phi}^*({\bar d}f)=\sum_{a} (\phi(f\rd_a)d a')_c~~,

2820: \ee

2821: which shows that the cyclic derivatives with respect to the new coordinates are given by:

2822: \be

2823: \label{ld_tf}

2824: {\bar \phi}(f)\rd_{a'}=\phi(f\rd_a)\Leftrightarrow \ld_{a'} {\bar \phi}(f)=\phi(\ld_a f)~~.

2825: \ee

2826: 

2827: Relative superalgebra endomorphisms $\phi$ of $A$ having the property $\phi\circ Q \circ \phi=Q$ correspond to 

2828: endomorphisms of the underlying weak $A_\infty$ category. The correspondence is obtained by expanding:

2829: \be

2830: \label{phi_expansion}

2831: \phi(a)=\sum_{n\geq 0}\phi_{a_1\ldots a_n}^a a_1\ldots a_n~~,

2832: \ee

2833: where the complex-valued coefficients $\phi_{a_1\ldots a_n}^a$ vanish unless $a_1\ldots a_n$ is a path starting 

2834: at $t(a)$ and ending at $h(a)$. The evenness condition on $\phi$ gives the selection rules:

2835: \be

2836: \nn

2837: \phi^a_{a_1\ldots a_n}=0~~{\rm unless}~{\tilde a}={\tilde a}_1+\ldots +{\tilde a}_n~~.

2838: \ee

2839: The $n=0$ part of (\ref{phi_expansion}) stands for the sum over loops 

2840: $\sum_{a\in Q_1(u,u)}\phi^{a}\epsilon_u$, i.e. we use the convention $\phi_{a_1\ldots a_0}^a:=

2841: \phi^a \delta^u_{h(a)}\delta^u_{t(a)}\epsilon_u$. Then the $A_\infty$ morphism associated with $\phi$ is 

2842: given by the even $R$-multilinear maps $\phi_n:E[1]^n\rightarrow E[1]$ defined through:

2843: \be

2844: \label{phi_maps}

2845: \phi_n(e_{a_1}\ldots e_{a_n})=\phi^a_{a_1\ldots a_n}e_a~~.

2846: \ee

2847: The conditions $\phi(ab)=\phi(a)\phi(b)$ amount to the complicated relations giving the traditional 

2848: definition.  The maps (\ref{phi_maps}) define an endomorphism of the

2849: weak $A_\infty$ structure on the superbimodule $E$; as usual,

2850: $R$-multilinearity allows one to decompose them into

2851: complex-multilinear maps describing an endomorphism of the underlying

2852: weak $A_\infty$ category.  In particular, the map $\phi_0:R\rightarrow

2853: E[1]$ gives even linear maps $\phi_u:\C\rightarrow E_{uu}[1]$ via the

2854: decomposition $\phi_0(\sum_{u}{\alpha_u

2855: \epsilon_u})=\sum_{u}{\epsilon_u \phi_u(\alpha_u)\epsilon_u}$ for

2856: complex $\alpha_u$; these can also be viewed as the odd elements

2857: $\phi_u(1_\C)=\sum_{h(a)=t(a)=u}\phi^a e_a \in E_{uu}$.

2858: 

2859: 

2860: An $A_\infty$ endomorphism of $(E,(r_n))$ is called {\em unital} if $\phi_1(\lambda)=\lambda$

2861: and $\phi_n(e_{a_1}\ldots e_{a_n})$ for $n\neq 1$ vanishes when any of 

2862: the elements $e_{a_1}\ldots e_{a_n}$ coincides with the odd $A_\infty$ unit $\lambda$. In terms of the 

2863: coefficients of $\phi$, this means $\phi^a_\sigma=\delta^a_\sigma$ and 

2864: $\phi^{a}_{a_1\ldots a_n}=0$ for all $n\neq 1$, if $\sigma\in \{a_1\ldots a_n\}$. Plugging this into  

2865: expansion (\ref{phi_expansion}), we see that the $A_\infty$ endomorphism is unital iff:

2866: \be

2867: \label{phi_unitality}

2868: \phi(\sigma)=\sigma~~~{\rm and}~~~\phi(a)={\rm ~independent~of~}\sigma {\rm~for~all}~~a\neq \sigma~~.

2869: \ee

2870: 

2871: The $A_\infty$ endomorphism determined by $\phi$ is {\em cyclic} if $\phi^*(\omega)=\omega$; writing

2872: this condition explicitly gives a series of complicated relations used in the traditional definition.

2873: In particular, a cyclic $A_\infty$ automorphism of $(E,\rho, (r_n)))$ amounts to a 

2874: relative symplectomorphism of $A$ preserving  the homological derivation $Q$. 

2875: 

2876: Let $\phi\in \Aut_R^\omega(A)$ be a relative symplectomorphism. 

2877: Remember from the end of Subsection \ref{sec:bracket} that 

2878: the map $\psi_\omega:C^0_R(A)[{\tilde \omega}]\rightarrow \Der_l^\omega(A)$  

2879: is equivariant with respect to the action of $\Aut_R^\omega(A)$. Moreover, the exact sequence (\ref{ex}) shows that 

2880: $\psi_\omega$ induces an isomorphism of vector spaces $C^0_R(A)[{\tilde \omega}]/R\approx \Der_l^\omega(A)$.

2881: Thus $Q$ is $\phi$-invariant iff its (canonical) Hamiltonian $W$ is invariant under the action of $\phi$ up to

2882: addition of elements of $R$: 

2883: \be

2884: \label{W_variation}

2885: {\bar \phi}(W)=W+\alpha~~{\rm~for~some~}\alpha\in R~~.

2886: \ee

2887: Moreover, the associated $A_\infty$ endomorphism is unital iff $\phi$ satisfies relations 

2888: (\ref{phi_unitality}).

2889: 

2890: Combining these observations, we see that a cyclic and unital

2891: endomorphism of the underlying $A_\infty$ structure amounts to a

2892: symplectomorphism of $A$ which satisfies (\ref{phi_unitality}) and

2893: (\ref{W_variation}).  The {\em symmetry group}  ${\cal G}\subset

2894: \Aut^\omega_R(A)$ of the system is the group of all such

2895: symplectomorphisms of $A$.  If $\phi$ belongs to ${\cal G}$, equations

2896: (\ref{ld_tf}) and (\ref{phi_unitality}), (\ref{W_variation})  imply: \be \nn \mu=\frac{1}{2}\ld_\sigma

2897: W=\frac{1}{2}\ld_{\sigma'} {\bar \phi}(W)=

2898: \frac{1}{2}\phi(\ld_\sigma W)=\phi(\mu)~~, \ee so $\phi$ preserves

2899: the moment element $\mu$. 

2900: 

2901: 

2902: \subsection{Algebraic construction of the noncommutative moduli space}

2903: 

2904: \label{nc_moduli}

2905: 

2906: Consider the two-sided ideal $J$ of $A$ generated by the elements: 

2907: \be

2908: \nn

2909: \ld_a W\in A ~~(a\in {\cal Q}_1)~~,

2910: \ee

2911: which we shall call the {\em critical ideal} of the noncommutative generating function. Notice that $J$

2912: is also generated by $W\rd_a$, due to the relations

2913: $\ld_a W=(-1)^{{\tilde a}{\tilde \omega}}W\rd_a$. We let

2914: $\C[{\cal Z}]:=A/J$. Since $J$ is $\Z_2$-homogeneous (being generated by

2915: $\Z_2$-homogeneous relations), the associative algebra $\C[{\cal Z}]$ is $\Z_2$

2916: graded.  Passage to $\C[{\cal Z}]$ implements the conditions:

2917: \be

2918: \nn

2919: \sum_{n\geq 0} {\bar W}_{aa_1\ldots a_n}a_1\ldots a_n=0~~\forall a\in {\cal Q}_1 \Leftrightarrow 

2920: \sum_{n\geq 0} {\bar W}_{a_1\ldots a_n a}a_1\ldots a_n=0~~\forall a \in {\cal Q}_1~~,

2921: \ee

2922: which (in view of the isomorphism $V\otimes_R A\approx C^1_R(A)$) can also be written as:

2923: \be

2924: \nn

2925: {\bar d}W=0~~.

2926: \ee

2927: In particular, the distinguished central element $\lambda=\sum_{u\in

2928: {\cal Q}_0}{\lambda_u}$ gives the relation:

2929: \be

2930: \nn

2931: \ld_\lambda W=0\Leftrightarrow \mu=0\Leftrightarrow \omega_{ab}ab=\frac{1}{2}\omega_{ab}[a,b]=0\Leftrightarrow 

2932: \sum_{a\in {\cal Q}_1}[a,a^*]=0~~,

2933: \ee

2934: where, as usual,  $[a,b]=ab-(-1)^{{\tilde a}{\tilde b}}ba$ is the supercommutator

2935: in $A$. Hence `extremizing' $W$ automatically imposes the zero-level 

2936: constraint for the noncommutative moment  map of \cite{BEV, Bergh}.

2937: 

2938: For $\phi\in \Aut_R(A)$, relations (\ref{df}) imply $\phi^*({\bar d}W)=(\phi(W\rd_a) da')_c$

2939: and $\phi^*({\bar d}W)={\bar d}({\bar \phi}(W))=({\bar \phi}(W)\rd_a da)_c=({\bar \phi}(W)\rd_a d\phi^{-1}(a'))_c$, 

2940: where $a':=\phi(a)$. Expanding $d\phi^{-1}(a')$ in the second expression and comparing with the first, 

2941: we find that

2942: $\phi(W\rd_a)$ belongs to the ideal generated by ${\bar

2943: \phi}(W)\rd_a$. This shows that relative automorphisms which preserve

2944: $W$ (i.e. ${\bar \phi}(W)=W$) also preserve the ideal $J$, so they

2945: descend to $R$-linear automorphisms of $\C[{\cal Z}]$ (the $R$-superbimodule

2946: structure on $\C[{\cal Z}]$ is induced from its obvious $A$-superbimodule

2947: structure). In particular, the group ${\cal G}$ preserves $J$, and we obtain 

2948: a group morphism $\gamma:{\cal G}\rightarrow \Aut_R(\C[{\cal Z}])$, i.e.  

2949: an action of ${\cal G}$ by $R$-linear automorphisms of the superalgebra $\C[{\cal Z}]$.

2950: The canonical epimorphism $\zeta:A\rightarrow \C[{\cal Z}]=A/J$ is ${\cal G}$-equivariant:

2951: \be

2952: \nn

2953: \zeta\circ \phi=\gamma(\phi)\circ \zeta~~~~~~\forall \phi\in {\cal G}~~.

2954: \ee

2955: We set $\C{\cal M}=\C[{\cal Z}]^{\cal G}$, the homogeneous 

2956: subalgebra of elements invariant under the action of ${\cal G}$. We will view these algebras 

2957: as noncommutative coordinate rings of `noncommutative schemes' ${\cal Z}$, ${\cal M}$, which we 

2958: call the {\em noncommutative extended vacuum space} and {\em noncommutative extended moduli space} 

2959: respectively.

2960: 

2961: 

2962: \section{The case of a single D-brane}

2963: \label{single_brane}

2964: 

2965: Let us illustrate the general discussion with the simple case of a single D-brane. 

2966: Then ${\cal Q}$ consists of $m$ loops at a single vertex, and we 

2967: let $m_\pm$ be the numbers of even and odd loops. 

2968: The space of boundary observables is 

2969: $E=\C^{m_-|m_+}$, with parity-changed dual 

2970: $V=E[1]^*=\C^{m_+|m_-}$. The boundary algebra $R$ coincides with $\C$ while 

2971: the path algebra is the free superalgebra $A=\C\langle\{a\}\rangle $ generated by all loops. The underlying 

2972: $A_\infty$ category has a single object, so it reduces to a weak,

2973: cyclic and unital $A_\infty$ algebra on the supervector space $E$. 

2974: This is the structure found in \cite{HLL}.  

2975: 

2976: 

2977: It is easy to see that $C^0(A):=C^0_\C(A)$ can be identified with the cyclic subspace $A_{\rm cyclic}$ of $A$, 

2978: defined as the image of the idempotent operator:

2979: \be

2980: \nn

2981: P=\id_{\C}\oplus \oplus_{n\geq 1}\left[\frac{1}{n}\sum_{i=0}^{n-1}{(\gamma_n)^i}\right]\in \End_\C(A)~~.

2982: \ee 

2983: Here $\gamma_n\in \End_\C (A_n)$ are the generators of the obvious $\Z_n$ action on $A_n$:

2984: \be

2985: \nn

2986: \gamma_n(x_1\otimes \ldots \otimes x_n)=(-1)^{{\tilde x}_1({\tilde x}_2+\ldots +{\tilde x}_n)}x_2\otimes \ldots 

2987: \otimes x_n \otimes x_1~~.

2988: \ee

2989: Thus $A_{\rm cyclic}$ consists of all polynomials $f=\sum_{n \geq 0}{f_{a_1\ldots a_n}a_1\ldots a_n}\in A$

2990: whose complex coefficients satisfy the conditions $f_{a_1\ldots a_n}=(-1)^{{\tilde a}_1({\tilde a}_2+\ldots +

2991: {\tilde a}_n)}f_{a_2\ldots a_n a_1}$. Writing $f=\sum_{n\geq 1}{f_n}$ with 

2992: $f_n\in A_n =V^{\otimes n}$, such a polynomial belongs to $A_{\rm cyclic}$ iff $\gamma_n(f_n)=f_n$ for 

2993: all $n$. 

2994: 

2995: Let $\sigma$ be the dual basis element corresponding to the parity changed $A_\infty$ unit $\lambda$. 

2996: We assume given a basis of $E$ such that $\sigma$ is one of the odd loops. 

2997: The generating function is a constant-free polynomial:

2998: \be

2999: \nn

3000: W= \sum_{n\geq 1}W_{a_1\ldots a_n}a_1\ldots a_n\in A_{\rm cyclic}

3001: \ee

3002: in the non-commuting variables $a$ such that

3003: $W_{a_1\ldots a_n}$ are graded-cyclic and satisfy the conditions

3004: $W_{a_1\ldots a_n}=0$ unless ${\tilde a}_1+\ldots +{\tilde

3005: a}_n={\tilde \omega}+1$, as well as the $A_\infty$ constraints

3006: (\ref{W_inf}). Moreover, we must have $W=W_g +W_d$ with

3007: $W_g=-\frac{1}{3}\omega_{ab}[\sigma ab+(-1)^{{\tilde a}+{\tilde b}}ab\sigma +

3008: (-1)^{{\tilde b}({\tilde a}+1)}b\sigma a]$ 

3009: and where $W_d$ vanishes at zero and is independent of $\sigma$. The

3010: matrix $(\omega_{ab})$ satisfies properties (\ref{omega_symm}) and

3011: (\ref{omega_sel}) of Section \ref{symplectic_structure} and no further

3012: constraints.

3013: 

3014: Any endomorphism $\phi$ of $A$ is determined by its values on the generators:

3015: \be

3016: \nn

3017: \phi(a):=\phi_{a}=\sum_{n\geq 0} \phi^{a}_{a_1\ldots a_n}a_1\ldots a_n

3018: \ee

3019: and can be viewed as an $m$-tuple $(\phi_{a_1}\ldots \phi_{a_m})$ of polynomials 

3020: in the non-commuting variables $a$. The degree zero condition on $\phi$ gives the constraints

3021: $\deg\phi(a)={\tilde a}$, so  the complex coefficient $\phi^{a}_{a_1\ldots a_n}$ must vanish 

3022: unless ${\tilde a}_1+\ldots +{\tilde a}_n={\tilde a}$. The relative automorphism group is the usual 

3023: group $\Aut(A)$ of superalgebra automorphisms. An automorphism $\phi$ belongs to the symmetry group ${\cal G}$ 

3024: if $\phi(\sigma)=\sigma$, $\phi(a)$ are independent of $\sigma$ for $a\neq \sigma$  

3025: and $W(\{\phi(a)\})$ equals $W(\{a\})$ as a polynomial in the non-commuting variables $\{a\}$. This imposes 

3026: nonlinear algebraic conditions on the coefficients $\phi^a_{a_1\ldots a_n}$. 

3027: 

3028: \paragraph{Observation} The automorphism group $\Aut(A)$ is a rather exotic object. 

3029: In the even case $m_-=0$, it is known\cite{Umirbaev} (see also \cite{DY}) that

3030: $\Aut(A)$ contains wild automorphisms\footnote{ An automorphism is

3031: called wild if it is not a composition of the so-called elementary

3032: automorphisms $(a_1\ldots a_m)\rightarrow (a_1\ldots a_i, \alpha

3033: a_i+f(a_1\ldots a_{i-1}, a_{i+1} \ldots a_m), a_{i+1}\ldots a_m)$ with

3034: $\alpha\in \C^*$ and $f$ a polynomial independent of $x_j$.} as soon

3035: as $m\geq 3$. Even in the commutative case, there are well-known

3036: open problems about automorphisms of polynomial algebras such as the

3037: Jacobian conjecture. The $\Z_2$-graded, noncommutative case does not

3038: seem to have been studied systematically.

3039: 

3040: 

3041: 

3042: \

3043: 

3044: 

3045: 

3046: \section{Examples} 

3047: \label{examples}

3048: 

3049: \subsection{Even system with a single boundary degree of freedom}

3050: 

3051: This is the simplest example relevant for topological sigma models

3052: with target spaces of even complex dimension. In this case, we

3053: have a single D-brane ($R=\C$) with $E=\C$ (concentrated in even degree), $V=\C[1]$ ( a purely odd 

3054: supervector space) and ${\tilde

3055: \omega}=0$.  The canonical forms of Subsection \ref{symplectic_structure} 

3056: show that this is the only possibility when the boundary superspace has dimension one.

3057: 

3058: Up to rescaling, the boundary sector contains a single boundary observable, 

3059: namely the identity operator $1$, which is the even $A_\infty$ unit.  

3060: We set $\lambda=\Sigma 1$ and let $\sigma$ be dual odd element in $V=\C[1]$ (of course, $\sigma$ can be 

3061: identified with $\lambda$ since we identify $\C^*$ with $\C$ using the canonical basis of $\C$ given by the unit).

3062: The superquiver consists of the single odd loop

3063: $\sigma$, with path superalgebra $A=\C\langle \sigma\rangle$. 

3064: By a change of normalization of the $A_\infty$ products, we can take $\omega_{\sigma \sigma}=1$; 

3065: then $\omega_{form}=\frac{1}{2}(d\sigma^2)_c$, and $\sigma$ can also be viewed as the canonical odd coordinate 

3066: $\xi$ in equation (\ref{e_can}). 

3067:   

3068: The non-commutative generating function is an odd

3069: polynomial $W=\sum_{n=odd}{W_n(\sigma^n)_c}\in C^0(A)=A_{\rm cyclic}=A^{\rm odd}$. The

3070: unitality constraint requires the splitting $W=W_d+W_g$ with $W_d$

3071: a constant.  Since $W$ must vanish at zero, this gives

3072: $W_d=0$.  Thus we must have: \be\nn

3073: W=W_g=-\frac{1}{3} \sigma^3 \ee and the only

3074: non-trivial strict coefficient (see Subsection \ref{quivc0}) is ${\bar W}_{\sigma\sigma\sigma}=-1$. The $A_\infty$

3075: constraint (\ref{W_inf}) is trivially satisfied. Since $W$ is cubic, the $A_\infty$ algebra 

3076: contains only the product $r_2$, which is completely determined by the unitality constraint

3077: $r_2(\lambda, \lambda)=-\lambda$. 

3078: The associative product $\cdot=\Sigma \circ r_2 \circ \Sigma^{\otimes 2}$ on $E$ is given 

3079: by $1\cdot 1=1$, which of course is the unique associative product on $\C$ with unit $1$.

3080: 

3081: Since elements

3082: $\phi\in {\cal G}$ must preserve $\sigma$, we have ${\cal

3083: G}=\{\id_A\}$. The critical ideal is generated by $\ld_\sigma W=-\sigma^2$, so

3084: $J=A\sigma^2A$ consists of all polynomials of order at least two at the

3085: origin. Thus $\C[{\cal Z}]=A/(\sigma^2)$ is the Grassmann algebra $\C[\sigma]$

3086: on the odd generator $\sigma$. The noncommutative moduli space coincides

3087: with the noncommutative vacuum space, having coordinate ring: \be

3088: \C[{\cal M}]=\C[{\cal Z}]=\C[\sigma]~~.\nn \ee Hence ${\cal M}={\cal

3089: Z}=\C^{0|1}$, the odd point of usual supergeometry.

3090: 

3091: In this extremely simple example, passage to the noncommutative moduli

3092: space gives nothing new, since supercommutativity is imposed as a

3093: consequence of the equations of motion. 

3094: 

3095: 

3096: \subsection{Odd system with two boundary degrees of freedom}

3097: This is the simplest example with an odd topological boundary metric, obtained for 

3098: $E=\C^{1|1}$. Choosing canonical coordinates as in (\ref{o_can}), we have:

3099: \be

3100: \nn

3101: \omega_{form}=(dp_0dq_0)_c

3102: \ee  

3103: with odd $q_0$ and even $p_0$, and take $\sigma=q_0$. Since

3104: $\sigma$ must be a loop, the only possibility for the boundary algebra

3105: is $R=\C$, i.e. the index superquiver consists of an even and an

3106: odd loop at a single vertex (indeed, $\omega_{form}$ vanishes for any

3107: other boundary structure). Thus $A=\C\langle p_0,q_0\rangle$ is a free

3108: associative superalgebra. In this case, we have $W_g=2(q_0^2p_0)_c$

3109: and $W_d=W_d(p_0)$ must be a constant-free univariate polynomial in

3110: $p_0$.  It is not hard to see that the Maurer-Cartan equation

3111: (\ref{MC}) for $W_d$ is trivially satisfied. Indeed, it is clear that

3112: $\{W_d,W_d\}=0$, while direct computation gives:

3113: \be

3114: \nn

3115: \{W_g,W_d\}=-(W_g\rd_{q_0}\ld_{p_0}W_d)_c=

3116: 2([q_0,p_0]\ld_{p_0}W_d)_c=2(q_0[p_0, \ld_{p_0}W_d])_c=0

3117: \ee

3118: where we noticed that the commutator $[p_0, \ld_{p_0}W_d]$ vanishes

3119: because $W_d$ is a polynomial in $p_0$.  Hence the general system of

3120: this type is described by $W=W_g+W_d$, where $W_d$ is an arbitrary

3121: constant-free polynomial in $p_0$. The defining equations $\ld_a W=0$

3122: for the noncommutative vacuum space take the form:

3123: \be

3124: \label{def_eqs}

3125: [q_0,p_0]=0~~,~~q_0^2=-\frac{1}{2}\ld_{p_0}W_d(p_0)~~.

3126: \ee

3127: Thus $\C[{\cal Z}]=\C\langle q_0,p_0\rangle/([q_0,p_0],

3128: q_0^2+\frac{1}{2}\ld_{p_0}W_d(p_0))$ and ${\cal Z}$ is a bona-fide

3129: noncommutative superspace. It can be viewed as a `fibration' over the

3130: noncommutative affine line $\A^1$ with coordinate $p_0$, where the

3131: (pure fuzz) fiber is a point-dependent deformation of the usual odd point

3132: $\C^{0|1}$.

3133: 

3134: It is known \cite{Cz, ML} that all automorphisms of a free associative

3135: algebra on two generators are tame, i.e. given by iterated composition

3136: of elementary automorphisms of the form $(q_0,p_0)\rightarrow (q_0,

3137: \alpha p_0+f(q_0))$ and $(q_0,p_0)\rightarrow (\beta q_0+g(p_0),p_0)$

3138: with $\alpha,\beta\in \C^*$ and $f,g$ arbitrary univariate

3139: polynomials. Moreover (see \cite{DY} for a more general result), 

3140: all algebra automorphisms fixing one variable are triangular. In particular, 

3141: automorphisms fixing $q_0$ have the form $(q_0,p_0)\stackrel{\phi}{\rightarrow}(q_0,\alpha p_0+f(q_0))$, 

3142: with {\em super}algebra automorphisms obtained by restricting to even polynomials. 

3143: Obviously $\phi(p_0)$ is independent of $\sigma=q_0$ iff $f$ is the constant polynomial. In this 

3144: case, $\phi$ is a symplectomorphism iff $\alpha=1$.  Hence ${\cal G}$ is 

3145: a subgroup of the one-dimensional translation group:

3146: \be

3147: \nn

3148: {\cal T}:~~q_0\rightarrow q_0~~,~~p_0\rightarrow p_0+t~~~~(t\in \C)~~.

3149: \ee

3150: Such an automorphism preserves $W_g$ up to addition of constants, and takes $W_d(p_0)$ into $W_d(p_0+t)$. 

3151: It follows that $W$ is preserved up to constants iff $W_d$ is linear in $p_0$. 

3152: Thus ${\cal G}=\{\id_A\}$ unless $W_d=w(p_0)_c$ for some $w\in \C$, in which case ${\cal G}={\cal T}$.

3153: 

3154: Hence the generic case of a nonlinear $W_d$ gives ${\cal M}={\cal Z}$. 

3155: When $W_d=w(p_0)_c$, equations (\ref{def_eqs})

3156: reduce to $[q_0,p_0]=0$ and $q_0^2=-\frac{w}{2}$

3157:  and we find $\C[{\cal M}]=\C[{\cal Z}]^{\cal T}=\C\langle

3158: q_0\rangle/(q_0^2+\frac{w}{2})$, i.e ${\cal M}$ is the quantum

3159: deformation of the odd point $\C^{0|1}$ given by $q_0^2=-\frac{w}{2}$.

3160: 

3161:  

3162: 

3163: \subsection{A family of odd examples}

3164: 

3165: Consider a theory with ${\tilde \omega}=1$, where we choose adapted

3166: coordinates $p_0\ldots p_m, q_0\ldots q_m$ (with $2(m+1)={\rm Card}

3167: {\cal Q}_0$, odd $q_i$ and even $p_i$) such that that $\omega$ has

3168: canonical form (\ref{o_can}).  Remember from Subsection

3169: \ref{symplectic_structure} that the coordinates $a^*=\omega^{ab}b$ are

3170: given by:

3171: \be

3172: \nn

3173: p_i^*=q_i~~,~~q_i^*=-p_i~~,

3174: \ee

3175: where $*$ can be viewed as an involution on $A_1=V$. Also remember

3176: that $p_i$ are even and $q_i$ are odd.  We assume that $q_0=\sigma$

3177: corresponds to the unit.

3178: 

3179: 

3180: Since ${\tilde \omega}=1$, we have $\{W_g,W_g\}=0$ and the discussion

3181: of Subsection \ref{theory_deformations} applies. In particular, a

3182: general theory with the given unit and symplectic form is specified by

3183: a solution $W_d$ of the Maurer-Cartan equation (\ref{MC}).  For

3184: ${\tilde \omega}=1$, relation (\ref{rdWg}) gives:

3185: \be

3186: \nn

3187: W_g\rd_a=[\sigma,a^*]+\delta_a^\sigma\sum_{\alpha\in Q_1}{\alpha\alpha^*}

3188: \ee

3189: and the conditions $W\rd_a=0$ take the form: 

3190: \be

3191: \nn

3192: [\sigma,a^*]+\delta_a^\sigma\sum_{\alpha\in Q_1}{\alpha\alpha^*}=-W_d\rd_a~~,

3193: \ee

3194: where, as usual, $[.,.]$ stands for the graded commutator.  In

3195: canonical coordinates, we have $[\sigma,

3196: \sigma^*]=[q_0,q_0^*]=-[q_0,p_0]=[p_0,q_0]$ and $\sum_{\alpha\in

3197: Q_1}{\alpha\alpha^*}=\sum_{i=0}^m{[p_i,q_i]}$. The noncommutative

3198: criticality constraints become:

3199: \be

3200: \label{E1}

3201: 2[p_0,q_0]+\sum_{i=1}^m{[p_i,q_i]}=-W_d\rd_{q_0}~~,

3202: \ee

3203: and:

3204: \be

3205: \label{E2}

3206: [q_0,p_i]=W_d\rd_{q_i}~~\forall i=1\ldots m,~~[q_0, q_i]=-W_d\rd_{p_i}~~~~~\forall i=0\ldots m~~.

3207: \ee

3208: 

3209: Let us assume that we are given a particular solution $W=W_g+W_d$ for

3210: which $W_d=W_d(p_1\ldots p_m)$ depends only on $p_1\ldots

3211: p_m$. Special solutions of this type were found in \cite{Katz} for

3212: systems which obey $\Z$-valued selection rules (for the examples of

3213: \cite{Katz}, equation (\ref{MC}) is trivially satisfied by the ansatz

3214: $W_d=W_d(p_1\ldots p_m)$ due to the integer-valued degree condition

3215: obeyed by $W$ in the case of Calabi-Yau compactifications).  In this

3216: case, eqs. (\ref{E1}) and (\ref{E2}) reduce to the following defining

3217: relations for the noncommutative vacuum space ${\cal Z}$:

3218: \begin{eqnarray}

3219: \label{Z_eqs}

3220: [q_0,q_i]&=&-\ld_{p_i}W_d~~~~~\forall i=1\ldots m\nn\\

3221: \left[q_0,p_i\right]&=&0~~~~~~~~~~~~~~~~\forall i=1\ldots m\\

3222: \left[q_0,p_0\right] &=&\frac{1}{2}\sum_{j=1}^m{[p_j,q_j]}~~\nn\\

3223: q_0^2&=&0~~\nn~~.

3224: \end{eqnarray}

3225: To arrive at this form, we noticed that $\ld_{p_i} W=W\rd_{p_i}$ since

3226: $p_i$ are even. Note that $[q_0,q_i]=q_0q_i+q_iq_0$ since $[.,.]$ is

3227: the graded commutator.

3228: 

3229: For simplicity, let us consider the case when the underlying quiver

3230: has a single vertex.  Then we can view ${\cal Z}$ as a `fibration'

3231: over the noncommutative affine plane $\A^{m+1}$ with coordinates

3232: $p_0\ldots p_m$, whose `fiber' is a subspace of the noncommutative

3233: affine space $\A^{m+1}$ (with coordinates $q_0\ldots q_m$) determined

3234: by (\ref{Z_eqs}).  The Abelian locus ${\cal Z}_{Ab}$ in ${\cal Z}$ is

3235: obtained by requiring that all variables supercommute. In this case,

3236: eqs. (\ref{Z_eqs}) reduce to the conditions

3237: $\partial_{p_i}W_d^{Ab}(p_1\ldots p_m)=0$ and we find that ${\cal

3238: Z}_{Ab}$ coincides with the critical locus ${\rm

3239: Crit}(W_d^{Ab})\subset \C^{m}$ of $W_d^{Ab}$, which is the usual

3240: vacuum space expected in the supercommutative formulation. The

3241: Abelianization epimorphism $\C\langle q_i, p_i\rangle \rightarrow

3242: \C[q_i,p_i]$ induces an embedding of ${\cal Z}_{Ab}$ into the much

3243: larger noncommutative space ${\cal Z}$. In the noncommutative vacuum

3244: space, one can move away from the critical locus of $W_{Ab}$ at the

3245: price of allowing for a non-vanishing commutator of $q_0$ with $q_i$;

3246: notice that this is possible even along the locus in ${\cal Z}$ where

3247: $p_i$ are required to commute.

3248: Determining the symmetry group ${\cal G}$ and noncommutative moduli

3249: space ${\cal M}$ in this class of examples is rather formidable in

3250: general and will not be attempted here. We only note that the trivial case $m=0$ corresponds 

3251: to the limit $W_d=0$ of the example discussed in the previous subsection. 

3252: 

3253: 

3254: 

3255: 

3256: 

3257: 

3258: \section{Conclusions}

3259: 

3260: We showed that the totality of boundary tree-level data determined by

3261: a topological string theory in a finite D-brane background can be

3262: encoded {\em faithfully} by using the non-commutative algebraic

3263: geometry of a superquiver determined by the boundary decomposition of

3264: the D-brane system. In particular, cyclicity of integrated boundary

3265: amplitudes on the disk and the weak $A_\infty$ constraints on such

3266: amplitudes amount to the condition $\{W,W\}=0$, where the boundary

3267: potential $W$ is a function defined on a {\em noncommutative}

3268: superspace $\A_{\cal Q}$ determined by the quiver. We also found a

3269: differential constraint on $W$ which expresses the presence of unit

3270: boundary observables in the boundary-preserving sector.

3271: 

3272: Fixing the bulk worldsheet data, but varying the D-brane background,

3273: gives rise to the (extended) boundary moduli space of such a

3274: system. We argued that this moduli space can be viewed as a

3275: noncommutative superspace ${\cal M}$ constructed as an (invariant

3276: theory) quotient of the `noncommutative critical locus' ${\cal Z}$ of

3277: $W$ by a certain group of symplectomorphisms acting on ${\cal Z}$.

3278: 

3279: One upshot of this analysis is that the complicated structure

3280: determined by {\em all} integrated boundary correlators on the disk is

3281: encoded faithfully by a form of noncommutative geometry. According to

3282: this point of view, the theory of topological D-brane deformations is

3283: intrinsically noncommutative.  This gives a stringy realization of

3284: non-commutativity at the level of such moduli spaces. It should be

3285: compared with the realization of \cite{SW}, which arises by

3286: translating the effective action of open strings into an action

3287: governing objects (such as connections) defined over a noncommutative

3288: space determined by the {\em closed} strings.  In both cases,

3289: non-commutativity originates \cite{Schomerus} from the fact that

3290: insertions of boundary observables on the disk do not commute. Hence

3291: these ostensibly different realizations are related, and it would be

3292: interesting to understand precisely how.

3293: 

3294: 

3295: \

3296: 

3297: \acknowledgments{  I am grateful to V. Ginzburg for pointing out

3298: the relevance of quiver algebras, and to M. Kontsevich for sharing his

3299: ideas. This work originated from discussions during the Third Workshop

3300: on Noncommutative Algebraic Geometry held at the Mittag Leffler

3301: Institute.}

3302: 

3303: \appendix

3304: 

3305: \section{Topological D-brane systems as cyclic and unital weak $A_\infty$ categories}

3306: \label{data}

3307: 

3308: 

3309: \subsection{Mathematical background}

3310: 

3311: A weak ($\Z_2$-graded) $A_\infty$ category ${\cal A}$ consists of a

3312: collection of objects $Ob{\cal A}$ and complex supervector spaces

3313: $\Hom_{\cal A}(u,v):=\Hom(u,v)$ for $u,v\in Ob {\cal A}$, together with

3314: odd multilinear maps\footnote{Notice that we take morphisms to compose {\em forward} under the \

3315: $A_\infty$ products.}  $r_{u_1\ldots

3316: u_{n+1}}:\Hom(u_1, u_2)[1]\times \Hom(u_2, u_3)[1]\times\ldots \times \Hom(u_n, u_{n+1})[1]

3317: \rightarrow \Hom(u_1,u_{n+1})[1]$ for all $n\geq 0$.  The

3318: case $n=0$ corresponds to odd maps $r_u:\C\rightarrow \Hom(u,u)[1]$, which

3319: amounts to giving even elements $\theta_u=r_u(1)\in \Hom(u,u)$.  The maps $r$ are

3320: required to satisfy certain conditions called the weak $A_\infty$

3321: constraints.  To formulate them, let $t(x), h(x)\in Ob {\cal A}$

3322: denote the tail and head of a morphism $x$, the unique objects of

3323: ${\cal A}$ such that $x\in \Hom(t(x), h(x))$. We say that an ordered

3324: collection of morphisms $(x_1,\ldots, x_n)$ is {\em composable} if

3325: $h(x_j)=t(x_{j+1})$ for all $j=1\ldots n-1$. In this

3326: case, we let:

3327: \be

3328: [x_1\ldots x_n]:=t(x_1)t(x_2)\ldots t(x_n)h(x_n)~~,

3329: \ee

3330: viewed as a word on the set $Ob {\cal A}$. 

3331: In particular, we set $[x]=t(x) h(x)$ for all morphisms $x$. 

3332: 

3333: We use $|.|$ to denote the degree of homogeneous elements of

3334: $\Hom(u,v)$ and ${\tilde .}$ for the degree of homogeneous elements of

3335: $\Hom(u,v)[1]$.  Then the maps $r_n$ are required to satisfy:

3336: \be

3337: \label{cat_ainf}

3338: \sum_{0\leq i+j\leq n} (-1)^{{\tilde x}_1+\ldots +{\tilde x}_i} 

3339: r_{[x_1\ldots x_i][x_{i+j+1}\ldots  x_{n}]}

3340: (x_1\ldots x_i, r_{[x_{i+1}\ldots x_{i+j}]}(x_{i+1}\ldots x_{i+j}), x_{i+j+1}\ldots x_n)=0~~

3341: \ee

3342: for any system of $\Z_2$-homogeneous and composable morphisms $(x_1, \ldots, x_n)$.

3343: 

3344: A weak $A_\infty$ category is called {\em strong} if $r_u=0\Leftrightarrow \theta_u=0$ for all

3345: $u\in Ob {\cal A}$ and {\em minimal} if it is strong and $r_{uv}=0$

3346: for all $u,v\in Ob{\cal A}$. It is called {\em unital} if one is given

3347: even elements $1_u\in \Hom(u,u)$ for each object $u$, such that the following conditions are

3348: satisfied:

3349: \begin{eqnarray}

3350: \label{cat_unitality}

3351: r_{[x_1\ldots x_{j-1}] [x_{j+1}\ldots x_n]}(x_1\ldots x_{j-1},e_{u_j},x_{j+1}\ldots x_n)&=&0~~

3352: {\rm for~all}~~~~ n\neq 2 ~{\rm~and~all}~ j~~\nn\\

3353: r_{[\lambda_u,x]}(\lambda_u,x)=-x~~,~~r_{[y,\lambda_u]}(y,\lambda_u)&=&(-1)^{{\tilde y}} y~~,

3354: \end{eqnarray}

3355: where $\lambda_u:=\Sigma 1_u\in \Hom(u,u)[1]$ and $u_j:=h(x_{j-1})=t(x_{j+1})$. In these relations, it is understood 

3356: that $(x_1,\ldots, x_{j-1},x_{j+1},\ldots, x_n)$ is any composable system consisting of $\Z_2$-homogeneous elements. 

3357: The last conditions in (\ref{cat_unitality}) are 

3358: imposed for any homogeneous $x\in \Hom(u,v)$ and $y\in \Hom(v,u)$, with arbitrary $u,v$.  

3359: The elements $1_u, \lambda_u$ are called {\em units} and {\em odd units} respectively. 

3360: 

3361: The $A_\infty$ category is called {\em cyclic} if one is given

3362: non-degenerate\footnote{This means that all linear maps $\Hom(v,u)\rightarrow \Hom(u,v)^*$ determined by 

3363: $\rho_{uv}$ are bijective.} bilinear forms $\rho_{uv}:\Hom(u,v)\times

3364: \Hom(v,u)\rightarrow \C$, homogeneous of the same $\Z_2$-degree ${\tilde

3365: \omega}$, such that $\rho_{uv}(x,y)=(-1)^{|x||y|}\rho_{vu}(y,x)$ and such that the

3366: following identities are satisfied:

3367: \be

3368: \label{cat_rcyc}

3369: \rho_{t(x_0) h(x_0)}(x_0,r_{[x_1\ldots x_n]}(x_1\ldots x_n))=

3370: (-1)^{{\tilde x}_0({\tilde x}_1+\ldots +{\tilde x}_n)}

3371: \rho_{t(x_1) h(x_1)}(x_1,r_{[x_2\ldots x_0]}(x_2\ldots x_n, x_0))~~,

3372: \ee

3373: whenever $(x_0,x_1 ,\ldots, x_n)$ is a homogeneous composable system 

3374: and $[x_0\ldots x_n]$ is a cyclic word, i.e. $h(x_n)=t(x_0)$. Equivalently,

3375: \be

3376: \label{cat_rrcyc}

3377: \omega_{t(x_0) h(x_0)}(x_0,r_{[x_1\ldots x_n]}(x_1\ldots x_n))=

3378: (-1)^{{{\tilde x}_0+{\tilde x}_1+\tilde x}_0({\tilde x}_1+\ldots +{\tilde x}_n)}

3379: \omega_{t(x_1) h(x_1)}(x_1,r_{[x_2\ldots x_0]}(x_2\ldots x_n, x_0))~~.

3380: \ee

3381: where $\omega_{uv}=\rho_{uv}\circ \Sigma^{\otimes 2}$

3382: (i.e. $\omega_{uv}(x,y)=(-1)^{\tilde x}\rho_{uv}(x,y)$) are the

3383: suspended bilinear forms $\omega_{uv}:\Hom(u,v)[1]\times

3384: \Hom(v,u)[1]\rightarrow \C$, which satisfy

3385: $\omega_{uv}(x,y)=(-1)^{{\tilde x}{\tilde y}+1}\omega_{vu}(y,x)$.

3386: 

3387: 

3388: The concept of $A_\infty$ category was introduced in \cite{Fukaya_inf}

3389: as a generalization of the 

3390: notion of $A_\infty$ algebra \cite{Stasheff}. These objects are studied mathematically 

3391: in \cite{Keller, Fukaya_mirror, KS, Hasegawa, Lyubashenko1,Lyubashenko2,Lyubashenko3, Lyubashenko4}. 

3392: It is by now well-understood that they play an important role in homological algebra, in particular 

3393: in giving a natural formulation of derived categories \cite{KS_book}. They also play a crucial 

3394: role in the homological mirror symmetry program \cite{HMS}. 

3395: 

3396: \subsection{Category-theoretic description of topological D-brane systems}

3397: 

3398: It was pointed out in \cite{CIL5} (and derived in \cite{HLL} from the

3399: worldsheet perspective, see also \cite{Costello1, Costello2}) that topological D-branes in string theory

3400: form the structure of a weak, cyclic and unital $A_\infty$

3401: category. In this realization, the D-branes of the theory are the

3402: objects of $A$, while each morphism space $\Hom_{\cal A}(u,v)$ is the

3403: space of zero-form topological observables for a string stretching from $u$ to

3404: $v$. The bilinear forms $\rho_{uv}$ are the topological metrics, while

3405: the $A_\infty$ units $1_u$ are the identity observables in the

3406: boundary sectors $\Hom_{\cal A}(u,u)$.  The $A_\infty$ products arise

3407: by dualizing the integrated correlators on the disk. These can be

3408: recovered from the former with the aid of the topological metrics:

3409: \be

3410: \nn

3411: \langle \langle x_0\ldots x_n\rangle \rangle =\rho_{h(x_0) t(x_0)}(x_0,r_{[x_1\ldots x_n]}(x_1\ldots x_n))~~,

3412: \ee

3413: for composable $(x_0\ldots x_n)$ such that $[x_0\ldots x_n]$ is a cyclic word. 

3414: We refer the reader to \cite{HLL, CIL5} for further details.

3415: 

3416: As explained in \cite{HLL, CIL5}, non-vanishing maps $r_u$ are present when

3417: the background of the topological string theory does not satisfy the string equations of motion. 

3418: In this case, the elements $\theta_u=r_u(1)\in \Hom_{\cal A}(u,u)$ correspond to

3419: tadpoles. This can usually be corrected by shifting the string vacuum

3420: until all tadpoles are eliminated \cite{CIL5, HLL}. One sometimes has an obstruction to reaching a 

3421: solution, a phenomenon which was originally noticed in \cite{Fukaya_book}. 

3422: 

3423: The structure described above is a homotopy-theoretic generalization

3424: of the boundary data of a topological field theory defined on open

3425: Riemann surfaces, which was studied in \cite{CIL1, Moore_Segal,

3426: Moore}. The latter data arises by forgetting all integrated boundary

3427: correlators, thereby keeping only the information contained in the

3428: boundary units, topological metrics and three-point functions.

3429: 

3430: The fact that weak cyclic and unital $A_\infty$ categories describe

3431: topological D-brane systems is implicit in the homological mirror

3432: symmetry conjecture \cite{HMS} and in the work of Fukaya and

3433: collaborators \cite{Fukaya_inf, Fukaya_book, Fukaya_mirror}.  The

3434: physical interpretation of this was discussed in \cite{CIL5}, the dG

3435: case having been considered previously in \cite{CIL2, CIL3, CIL6}. A

3436: connection with the D-brane superpotential and Chern-Simons theory was

3437: made in \cite{CIL4, CIL5}, realizing explicitly an observation

3438: originally due to \cite{Witten_CS}.  See \cite{CIL7,CIL8,CIL9,CIL10}

3439: for further physics discussion. $A_\infty$ algebras as descriptions of

3440: open string vertices appeared originally in \cite{Gaberdiel} in the

3441: context of bosonic string field theory, which was further studied in

3442: \cite{Kajiura}. As discussed in \cite{CIL4, CIL5, Kajiura}, the

3443: $A_\infty$ structure describing open topological strings can also be

3444: obtained from the results of \cite{Gaberdiel}, by using the well-known

3445: formal analogy of bosonic and topological string theories.

3446: 

3447: \section{Some basic isomorphisms}

3448: \label{isomorphisms}

3449: 

3450: Convention (\ref{dual_multiplications}) for the superbimodule structure of the dual 

3451: affects some standard isomorphisms. Given two $R$-superbimodules $U,V$, the 

3452: supervector space $\Hom_{\rm Mod-R}(U^{opp}, V)$ becomes an $R$- superbimodule

3453: with respect to the external multiplications given by $(\alpha\phi\beta)(x)=\alpha\phi(x\beta)$. 

3454: There exists an isomorphism of $R$-superbimodules 

3455: $\Hom_{\rm Mod-R}(U^{opp}, V) \approx V\otimes_R U^{\rm v}$, whose inverse takes 

3456: $y\otimes_R f\in V\otimes_R U^{\rm v}$

3457: into the right $R$-supermodule morphism $\phi:U^{opp}\rightarrow V$ given by 

3458: $\phi(x):=yf(x)$. This isomorphism maps the $R$-sub-bimodule $\Hom_{\rm R-Mod-R}(U^{opp},V)$

3459: of $\Hom_{\rm Mod-R}(U^{opp}, V)$ into the center of $V\otimes_R U^{\rm v}$:

3460: \be

3461: \label{basic_isomorphism}

3462: \iHom(U^{opp},V)=\Hom_{\rm R-Mod-R}(U^{opp},V)\approx [V\otimes_R U^{\rm v}]^R~~.

3463: \ee

3464: In particular, we have $\iHom(U^{opp},U)\approx [U^{\rm v}\otimes_R U^{\rm v}]^R$, so 

3465: given a bilinear form $\sigma$ on $U$, the map $j_\sigma:U^{opp}\rightarrow U^{\rm v}$ defined in 

3466: Section 2 can be identified with a central element ${\hat \sigma}\in [U^{\rm v}\otimes_R U^{\rm v}]^R$. 

3467: Tracing through the identifications, one finds that $\sigma$ can be recovered from ${\hat \sigma}$ 

3468: as follows. If ${\hat \sigma}=\sum_{i} f_i\otimes_R g_i $ with $f_i,g_i\in U^{\rm v}$, then

3469: $\sigma(x,y)=\sum_{i}f_i(xg_i(y))$. 

3470: 

3471: 

3472: 

3473: \section{Explicit construction of the differential envelope}

3474: \label{envelope} 

3475: The construction below is a slight adaptation of that given \cite{CK}. 

3476: Let $A$ be an $R$-superalgebra.

3477: \bd

3478: Consider the $R$-superbimodule $A_R:=A/R$ and let $\nu:A\rightarrow

3479: A_R$ be the natural projection.  We let $\nu(a):={\bar a}$ for all

3480: $a\in A$ and let $K_R =A\oplus A_R $, viewed as an $R$-superbimodule.

3481: Define:

3482: \be

3483: \Omega_R A=T_R K_R /J

3484: \ee  

3485: where $T_R K$ (endowed with its obvious $\Z\times \Z_2$ grading) is

3486: the tensor algebra on $K$ and $J\subset T_R K_R $ is the $\Z\times

3487: \Z_2$-homogeneous two-sided ideal generated by all elements of the

3488: form ${\overline a}\otimes b+a\otimes {\overline b}-\overline{ab}$ and

3489: $a\otimes b-ab$ with $a,b\in A$.  We let $d$ be the unique $R$-linear

3490: derivation of $\Omega_R A$ of bidegree $(1,0)\in \Z\times \Z_2$ (with

3491: respect to the pairing (\ref{pairing})) which satisfies the relations:

3492: \be

3493: da={\bar a}~~,~~d{\bar a}=0~~{\rm for~all~}a\in A~~

3494: \ee

3495: (this derivation is well-defined). The $\Z\times \Z_2$-grading of $\Omega_R A$ is induced by the obvious 

3496: $\Z\times \Z_2$-grading of $T_R K_R$. 

3497: \ed

3498: 

3499: \noindent If $\cdot$ denotes multiplication in $\Omega_R A$, we find

3500: that $\Omega_R A$ is spanned by all finite products of elements of the

3501: form $a$ and $db={\bar b}$ with $a,b\in A$, with the relations

3502: $d(ab)=(da)\cdot b+a\cdot (db)$ and $a\cdot b=ab$

3503: Notice that the unit element of $\Omega_R A$ coincides with $1_A$ due to

3504: the relation $1_A\cdot a=1_A a=a=a1_A=a\cdot 1_A$ for $a\in A$, and

3505: that $d$ squares to zero due to the relations $d{\bar

3506: a}=0\Leftrightarrow d^2a=0$ for $a\in A$. Moreover, any element of

3507: $\Omega_ A$ can be written as a finite sum of elements of the form

3508: $a_0\cdot da_1\cdot \ldots \cdot da_n$ with $a_i\in A$, by applying

3509: the identity:

3510: \begin{eqnarray}

3511: a_0 \cdot{\bar a}_1\cdot\ldots \cdot{\bar a}_n\cdot b_0 \cdot{\bar

3512: b}_1\cdot \ldots \cdot{\bar b}_m&:=&a_0 \cdot{\bar a}_1 \cdot \ldots

3513: {\bar a}_{n-1}\cdot\overline{a_n b_0}\cdot~{\bar b}_1\cdot\ldots

3514: \cdot{\bar b}_m\nn\\ &&+\sum_{i=1}^{n-1}{(-1)^{n-i}a_0 \cdot{\bar

3515: a}_1\cdot\ldots \cdot\overline{a_{n-i}a_{n-i+1}} \cdot\ldots \cdot

3516: {\bar a}_n\cdot {\bar b}_0}\cdot\ldots \cdot{\bar b}_m \nn\\ &&+(-1)^n

3517: (a_0a_1) \cdot{\bar a}_2 \cdot\ldots \cdot{\bar a}_n \cdot{\bar

3518: b}_0\cdot \ldots \cdot{\bar b}_m~~\nn\\ ~~{\rm for~all}~~ a_0\ldots

3519: a_n, b_0\ldots b_m\in A~~, \nn

3520: \end{eqnarray}

3521: which follows from the relations valid in $\Omega_R A$. Furthermore, we

3522: can denote the product in $\Omega_R A$ by juxtaposition.

3523: 

3524: \section{Coefficient expression for the cyclic bracket}

3525: \label{coeffs}

3526: 

3527: \bp

3528: \label{cycbracket}

3529: Let $f, g\in C^0_R(A)$ have the forms: 

3530: \begin{eqnarray}

3531: f&=&\sum_{n\geq 0}{f_{a_1\ldots a_n}({a_1}\ldots {a_n})_c}=c(f)+

3532: \sum_{n\geq 1}{\frac{{\bar f}_{a_1\ldots a_n}}{n}({a_1}\ldots {a_n})_c}

3533: ~~\nn\\

3534: g&=&\sum_{n\geq 0}{g_{a_1\ldots a_n}({a_1}\ldots {a_n})_c}= 

3535: c(g)+

3536: \sum_{n\geq 1}{\frac{{\bar g}_{a_1\ldots a_n}}{n}({a_1}\ldots {a_n})_c}~~\nn

3537: \end{eqnarray}

3538: with strict coefficients ${\bar f}_{a_1\ldots a_n}$, ${\bar g}_{a_1\ldots a_n}$

3539: and assume that both $f$ and $g$ are homogeneous of degree

3540: ${\tilde \omega}+1\in \Z_2$.  Then the cyclic bracket of $f$ and $g$: 

3541: \be

3542: \{f,g\}=c(\{f,g\})+\sum_{n\geq 1}{\frac{{\bar \phi}_{a_1\ldots a_n}}{n} ({a_1}\ldots {a_n})_c}~~,\nn

3543: \ee

3544: has strict coefficients:

3545: \begin{eqnarray}

3546: \label{strict_bracket}

3547: {\bar \phi}_{a_1\ldots a_n}=\sum_{j=0}^n \sum_{i=0}^{n-j} (-1)^{{\tilde a}_1+\ldots + {\tilde a}_i}

3548: &[&{\bar f}_{a_1\ldots a_i a a_{i+j+1}\ldots a_n} {\bar g}^a_{\,\,a_{i+1}\ldots

3549:       a_{i+j}}\nn\\

3550: &+&{\bar g}_{a_1\ldots a_i a a_{i+j+1}\ldots a_n}{\bar f}^a_{\,\,

3551:       a_{i+1}\ldots a_{i+j}}]~~.

3552: \end{eqnarray}

3553: and: 

3554: \be

3555: \label{c}

3556: c(\{f,g\}):={\bar f}_a{\bar g}^a={\bar f}^a{\bar g}_a~~,

3557: \ee

3558: where we lift coefficients with $\rho^{ab}$ as in (\ref{lifts}).

3559: \ep

3560: 

3561: \noindent With our conventions, expression (\ref{c}) can be obtained by formally setting 

3562: $n=0$ in equation (\ref{strict_bracket})) and dividing by two. The second equality in (\ref{c}) follows upon using the 

3563: symmetry property of $\rho^{ab}$ and the selection rule ${\tilde a}={\tilde \omega}+1$. 

3564: 

3565: 

3566: \begin{proof}

3567: By Proposition \ref{coord_bracket}, we have:

3568: \be

3569: \{f,g\}=\sum_{n\geq 0}\sum_{j=0}^n F^{(j)}_{(a_1\ldots a_n)}({a_1}\ldots {a_n})_c\nn

3570: \ee

3571: where: 

3572: \be

3573: F^{(j)}_{a_1\ldots a_n}={\bar f}_{a_1\ldots a_jb}\omega^{ba}{\bar g}_{a a_{j+1}\ldots a_n}~~.\nn

3574: \ee

3575: The case $n=0$ can be checked directly, so we discuss only the case $n\geq 1$. 

3576: 

3577: \noindent If $i+j\leq n$, we compute: 

3578: \be

3579: F^{(j)}_{a_{i+1}\ldots a_n, a_1\ldots a_i}={\bar f}_{a_{i+1}\ldots a_{i+j}b}\omega^{ba}

3580: {\bar g}_{aa_{i+j+1}\ldots a_na_1\ldots a_i}=

3581: (-1)^{\sigma_1} {\bar g}_{a_1\ldots a_i a a_{i+j+1}\ldots a_n}{\bar f}^a_{\,\, a_{i+1}\ldots a_{i+j}}~~,\nn

3582: \ee

3583: where we used cyclicity of $f$ and $g$ to permute indices. The sign factor is easily determined 

3584: keeping in mind the symmetry property (\ref{oinv_symm}) and the selection rule (\ref{oinv_sel}):

3585: \be

3586: \sigma_1={\tilde \omega}({\tilde a}+1)+({\tilde a}_1+\ldots +{\tilde a}_i)({\tilde a}+

3587: {\tilde a}_{i+j+1}+\ldots +{\tilde a}_n)+({\tilde a}+{\tilde \omega})({\tilde a}_{i+1}+\ldots 

3588: +{\tilde a}_{i+j})~~({\rm mod}~2)~~.\nn

3589: \ee

3590: For $i+j>n$, we find:

3591: \be

3592: F^{(j)}_{a_{i+1}\ldots a_n, a_1\ldots a_i}=

3593: (-1)^{{\tilde a}+{\tilde \omega}+1}{\bar f}_{a_{i+1}\ldots a_na_1\ldots a_{j-n+i}a}{\bar g}^a_{\,\,a_{j-n+i+1}\ldots a_i}=

3594: (-1)^{\sigma_2}{\bar f}_{a_1\ldots a_{j-n+i}aa_{i+1}\ldots a_n}{\bar g}^a_{\,\,a_{j-n+i+1}\ldots a_i}~~,\nn

3595: \ee

3596: with:

3597: \be

3598: \sigma_2={\tilde a}+{\tilde \omega}+1+

3599: ({\tilde a}_{i+1}+\ldots +{\tilde a}_n)({\tilde a}+{\tilde a}_1+\ldots +{\tilde a}_{j-n+i})

3600: ~~({\rm mod}~2)~~.\nn

3601: \ee

3602: 

3603: \noindent This allows us to write: 

3604: \be

3605: \label{two_sums}

3606: \sum_{j=0}^n F^{(j)}_{(a_1\ldots a_n)}=\frac{1}{n}\left[\sum_{j=0}^n\sum_{i=1}^{n-j}

3607: (-1)^{\epsilon_1} F^{(j)}_{a_{i+1}\ldots a_n, a_1\ldots a_i}+ \sum_{j=0}^n\sum_{i=n-j+1}^{n}

3608: (-1)^{\epsilon_2} F^{(j)}_{a_{i+1}\ldots a_n, a_1\ldots a_i}\right]\nn

3609: \ee

3610: where:

3611: \begin{eqnarray}

3612: \epsilon_1&=&{\tilde a}+{\tilde \omega}+

3613: ({\tilde a}+{\tilde \omega}+{\tilde a}_1+\ldots +{\tilde a}_i)({\tilde a}+{\tilde a}_{i+1}+\ldots 

3614: +{\tilde a}_{i+j})\nn\\

3615: \epsilon_2&=& 1+{\tilde a}+{\tilde \omega}+

3616: ({\tilde a}_{i+1}+\ldots +{\tilde a}_n)({\tilde a}+{\tilde a}_{j-n+i+1}+\ldots +{\tilde a}_i)~~.\nn

3617: \end{eqnarray} 

3618: We now perform the replacement $i\rightarrow i'=i+j-n, j\rightarrow j'=n-j$ in the second 

3619: double sum. Denoting the new summation indices $i',j'$ by $i,j$, this gives: 

3620: \be

3621: \sum_{j=0}^n F^{(j)}_{(a_1\ldots a_n)}=\frac{1}{n}\sum_{j=0}^n\sum_{i=1}^{n-j}\left[(-1)^{\epsilon_1}

3622: {\bar g}_{a_1\ldots a_i a a_{i+j+1}\ldots a_n}{\bar f}^a_{\,\, a_{i+1}\ldots a_{i+j}}+(-1)^{\tilde \epsilon_2}

3623: {\bar f}_{a_1\ldots a_i a a_{i+j+1}\ldots a_n}{\bar g}^a_{\,\, a_{i+1}\ldots a_{i+j}}\right]~~,\nn

3624: \ee

3625: where:

3626: \be

3627: {\tilde \epsilon}_2=1+{\tilde a}+{\tilde \omega}+({\tilde a}+{\tilde a}_{i+1}+\ldots +{\tilde a}_{i+j})({\tilde a}_{i+j+1}+

3628: \ldots +{\tilde a}_n)~~.\nn

3629: \ee

3630: The next step is to notice that the first term in square brackets vanishes unless: 

3631: \be

3632: {\tilde a}+{\tilde a}_1+\ldots +{\tilde a}_i={\tilde g}+{\tilde

3633:   a}_{i+j+1}+\ldots +{\tilde a}_n ~~({\rm mod}~2)\nn

3634: \ee

3635: and: 

3636: \be

3637: {\tilde a}+{\tilde a}_{i+1}+\ldots +{\tilde a}_{i+j}={\tilde f}+{\tilde \omega}~~ 

3638: ~~({\rm mod}~2)~~,\nn

3639: \ee

3640: which allows us to replace $\epsilon_1$ by:

3641: \be

3642: \epsilon_1'={\tilde a}+{\tilde \omega}+

3643: ({\tilde f}+{\tilde \omega})({\tilde g}+{\tilde \omega}+{\tilde a}_{i+j+1}+\ldots +

3644: {\tilde a}_n)~~.\nn

3645: \ee

3646: Similarly, the second term in the square brackets vanishes unless: 

3647: \be

3648: {\tilde a}+{\tilde a}_{i+1}+\ldots +{\tilde a}_{i+j}={\tilde g}+{\tilde \omega}~~,\nn

3649: \ee

3650: which allows us to replace ${\tilde \epsilon}_2$ by: 

3651: \be

3652: \epsilon_2'={\tilde a}+ {\tilde \omega}+1+({\tilde g}+{\tilde \omega})({\tilde a}_{i+j+1}+\ldots +

3653: {\tilde a}_n)~~.\nn

3654: \ee

3655: Finally, using the assumption ${\tilde f}={\tilde g}={\tilde \omega}+1$, we find: 

3656: \be

3657: \epsilon'_1=\epsilon_2'=1+{\tilde \omega}+{\tilde a}+{\tilde a}_{i+j+1}+\ldots +{\tilde a}_n={\tilde a}_1+\ldots +

3658: {\tilde a}_i~~({\rm mod}~2)~~,\nn

3659: \ee

3660: where we used the selection rule ${\tilde a}_1+\ldots +{\tilde a}_i+{\tilde a}+

3661: {\tilde a}_{i+j+1}+\ldots 

3662: +{\tilde a}_n={\tilde f}={\tilde g}={\tilde \omega}+1$. 

3663: 

3664: \end{proof}

3665: 

3666: \paragraph{Corollary}

3667: \label{maincor}

3668: Let $f=c(f)+\sum_{n\geq 0}{\frac{{\bar f}_{a_1\ldots a_n}}{n}({a_1}\ldots {a_n}})_c$ be a 

3669: cyclic element of $A$ of degree 

3670: ${\tilde \omega}+1$. Then: 

3671: \be

3672: \frac{1}{2}\{f,f\}=\frac{1}{2}{\bar f}_a{\bar f}^a+

3673: \sum_{n\geq 1}\frac{1}{n}\left(\sum_{0\leq i+j\leq n}(-1)^{{\tilde a_1}+\ldots +{\tilde a}_i}

3674: {\bar f}_{a_1\ldots a_i a a_{i+j+1}\ldots a_n} {\bar f}^a_{\,\,a_{i+1}\ldots

3675:       a_{i+j}}\right)({a_1}\ldots {a_n})_c~~.\nn

3676: \ee

3677: 

3678: 

3679: 

3680: 

3681: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

3682: \begin{thebibliography}{100}

3683: 

3684: %%%%%%%%%%%%%%%%%%%%%%%%% for bibtex

3685: 

3686: \bibitem{Witten_mirror}{E.~Witten,

3687: ``Mirror manifolds and topological field theory,'', in {\em Essays on mirror

3688: manifolds},  pp 120--158, Internat. Press, Hong Kong, 1992 [arXiv:hep-th/9112056].}

3689: %%CITATION = HEP-TH 9112056;%%

3690: \bibitem{Dubrovin}{ B.~Dubrovin,``Geometry of 2-D topological field

3691: theories,'' LNM {\bf 1620}, Springer, 1996 [arXiv:hep-th/9407018].}

3692: %%CITATION = HEP-TH 9407018;%%

3693: \bibitem{WDVV}{R.~Dijkgraaf, H.~Verlinde and E.~Verlinde,

3694: ``Topological Strings In D $<$ 1,'' Nucl.\ Phys.\ B {\bf

3695: 352} (1991) 59.}

3696: %%CITATION = NUPHA,B352,59;%%

3697: \bibitem{BarKon}{ S.~Barannikov, M.~Kontsevich, ``Frobenius

3698: Manifolds and Formality of Lie Algebras of Polyvector Fields'', Internat. Math. Res. Notices {\bf 4} 

3699: (1998) 201-215 [alg-geom/9710032]}

3700: \bibitem{Tian}{G. Tian , ``Smoothness of the Universal Deformation Space of

3701: Compact Calabi-Yau Manifolds and its Petersson-Weil Metric,'' in {\it Mathematical

3702: Aspects of String Theory,} 629-646, ed. S.-T. Yau, World Scientific, Singapore,

3703: 1987.}

3704: \bibitem{Todorov}{ A. Todorov, ``The Weil-Petersson Geometry of the Moduli Space of

3705: $SU(n\ge 3)$ (Calabi-Yau) Manifolds I,'' {\it Commun. Math. Phys} {\bf 126} (1989) 325-346.}

3706: \bibitem{HLL}{M.~Herbst, C.~I.~Lazaroiu and W.~Lerche,

3707: ``Superpotentials, A(infinity) relations and WDVV equations for open

3708: topological strings,'' JHEP {\bf 0502} (2005) 071 [arXiv:hep-th/0402110].}

3709: %%CITATION = HEP-TH 0402110;%%

3710: \bibitem{CIL5}{C.~I.~Lazaroiu,

3711: ``D-brane categories,'' Int.\ J.\ Mod.\ Phys.\ A {\bf 18} (2003) 5299

3712: [arXiv:hep-th/0305095].}

3713: \bibitem{Costello1}{K. Costello, ``The A-infinity operad and the moduli space of curves'', 

3714: math.AG/0402015.}

3715: \bibitem{Costello2}{K. Costello, ``Topological conformal field theories and Calabi-Yau categories'', 

3716: math.QA/0412149.} 

3717: %%CITATION = HEP-TH 0305095;%%

3718: \bibitem{Stasheff_Kajiura}{H.~Kajiura, J.~Stasheff, 

3719: ``Homotopy algebras inspired by classical open-closed string field theory'', 

3720: math.QA/0410291.}

3721: \bibitem{Kajiura}{H. Kajiura, ``Noncommutative homotopy algebras

3722:       associated with open strings'', Ph.D. thesis, math.QA/0306332.}

3723: \bibitem{Gaberdiel}{M.~R.~Gaberdiel and B.~Zwiebach, 

3724: ``Tensor constructions of open string theories I: Foundations,''

3725:   Nucl.\ Phys.\ B {\bf 505} (1997) 569  [arXiv:hep-th/9705038].}

3726: %%CITATION = HEP-TH 9705038;%%

3727: \bibitem{Konts_formal}{M. Kontsevich, ``Formal (non)commutative

3728: symplectic geometry.'' , {\em The Gelfand Mathematical Seminars, 1990--1992},

3729: 173--187, Birkh\"auser Boston, Boston, MA, 1993.}

3730: \bibitem{LB_lectures}{L. LeBruyn, ``Noncommutative~geometry $@n$'',

3731: unpublished book, available at $www.neverendingbooks.org$; see also

3732: ``Three~talks on noncommutative geometry$@n$'', math.RA/0312221.}

3733: \bibitem{Ginzburg_lectures}{V. Ginzburg, ``Lectures on Noncommutative Geometry'', 

3734: math.AG/0506603.}

3735: \bibitem{HMS}{M.~Kontsevich, ``Homological algebra of mirror symmetry'', Proc.

3736:   Internat. Congr. Math., Z\"urich, Switzerland 1994 (Basel), vol.~1,

3737:   Birkh\"auser Verlag, 1995, 120--139.}

3738: \bibitem{Polishchuk}{A. Polishchuk, ``Homological mirror symmetry

3739: with higher products'', {\em Winter School on Mirror Symmetry, Vector Bundles and

3740: Lagrangian Submanifolds} (Cambridge, Massachusetts, 1999) in 

3741: AMS/IP Stud. Adv. Math. {\bf 23} pp.~247--259 [math.AG/9901025].}

3742: \bibitem{RSS}{G.-C.~Rota, B.~Sagan, P.~R. Stein, ``A cyclic

3743:       derivative in noncommutative algebra'', J. of Algebra {\bf 64}

3744:       (1980) 54-75.}

3745: \bibitem{Voiculescu}{D.~Voiculescu, ``A note on cyclic gradients'',

3746:     Indiana Univ. Math. J {\bf 49} (2000) 3, 837-841.}

3747: \bibitem{BEV}{ W. Crawley-Boevey, P. Etingof, V. Ginzburg, 

3748: ``Noncommutative Geometry and Quiver algebras'', math.AG/0502301}

3749: \bibitem{Bergh}{M. Van den Bergh, ``Double Poisson algebras'', math.QA/0410528.}

3750: \bibitem{CIL1}{C.~I.~Lazaroiu,

3751:  ``On the structure of open-closed topological field theory in two dimensions,''

3752: Nucl.\ Phys.\ B {\bf 603}  (2001) 497 [arXiv:hep-th/0010269].}

3753: %%CITATION = HEP-TH 0010269;%%

3754: \bibitem{CQ}{J. Cuntz, D. Quillen, ``Algebra extensions and

3755: nonsingularity'', Journal AMS {\bf 8} (1995) 251-289.}

3756: \bibitem{Ginzburg}{ V. Ginzburg, ``Noncommutative symplectic

3757: geometry, quiver varieties and operads'', Math. Res. Lett {\bf 8} (2001) 3, pp

3758: 377-400.}

3759: \bibitem{LB_quivers}{ R. Bocklandt, L. Le Bruyn, ``Necklace Lie

3760: algebras and noncommutative symplectic geometry'', Math. Z. {\bf 240} (2002) 141-167 [math.AG/0010030]}

3761: \bibitem{Katz}{P.~S.~Aspinwall and S.~Katz, ``Computation of superpotentials for D-Branes,''

3762:   arXiv:hep-th/0412209.}

3763: %%CITATION = HEP-TH 0412209;%%

3764: \bibitem{CIL2}{C.~I.~Lazaroiu,

3765: ``Generalized complexes and string field theory,''

3766: JHEP {\bf 0106} (2001) 052 [arXiv:hep-th/0102122].}

3767: %%CITATION = HEP-TH 0102122;%%

3768: \bibitem{CIL3}{C.~I.~Lazaroiu,

3769: ``Unitarity, D-brane dynamics and D-brane categories,''

3770: JHEP {\bf 0112} (2001) 031 [arXiv:hep-th/0102183].}

3771: %%CITATION = HEP-TH 0102183;%%

3772: \bibitem{Konts_Schwarz}{ M. Alexandrov, M. Kontsevich, A. Schwarz, O. Zaboronsky,

3773: ``The Geometry of the Master Equation and Topological Quantum Field

3774: Theory'', Int. J. Mod.Phys. {\bf A12} (1997) 1405-1430 [hep-th/9502010].}

3775: \bibitem{Penkava}{M. Penkava, 

3776: ``Infinity Algebras, Cohomology and Cyclic Cohomology, and Infinitesimal Deformations'',

3777:  math.QA/0111088.}

3778: \bibitem{Lazarev}{ A. Hamilton, A. Lazarev, ``Homotopy algebras and

3779: noncommutative geometry'', math.QA/0410621.}

3780: \bibitem{Umirbaev}{U. U. Umirbaev,

3781: ``Tame and wild automorphisms of polynomial algebras and free associative

3782: algebras'', Max-Planck Institute Bonn, Preprint MPIM2004-108.}

3783: \bibitem{DY}{ V. Drensky, J.-T. Yu, `` The strong Anick conjecture is true'', 

3784: math.RA/0507170.} 

3785: \bibitem{Cz}{A.J. Czerniakiewicz, ``Automorphisms of a

3786: free associative algebra of rank 2. I, II'',

3787: Trans. Amer. Math. Soc. {\bf 160} (1971) 393-401;

3788: {\bf 171} (1972) 309-315.}

3789: \bibitem{ML}{L.G. Makar-Limanov,

3790: ``On automorphisms of free algebra with two generators'', Functional Anal. Appl. {\bf 4} (1970), 262-263.}

3791: \bibitem{SW}{N.~Seiberg, E.~Witten, ``String theory and

3792:   noncommutative geometry,'' JHEP {\bf 9909} (1999) 032 [arXiv:hep-th/9908142].}

3793: %%CITATION = HEP-TH 9908142;%%

3794: \bibitem{Schomerus}{V.~Schomerus,``D-branes and deformation

3795:   quantization,'' JHEP {\bf 9906} (1999) 030 [arXiv:hep-th/9903205].}

3796: %%CITATION = HEP-TH 9903205;%%

3797: \bibitem{Moore_Segal}{G. Moore and G. Segal, unpublished; see http://online.kitp.ucsb.edu/online/mp01/}

3798: \bibitem{Moore}{G.~W.~Moore,

3799: ``Some comments on branes, G-flux, and K-theory,''

3800: Int.\ J.\ Mod.\ Phys.\ A {\bf 16} (2001) 936 [arXiv:hep-th/0012007].}

3801: %%CITATION = HEP-TH 0012007;%%

3802: \bibitem{Fukaya_mirror}{

3803: K.~Fukaya, ``Floer homology and mirror symmetry. II'', in {\em Minimal

3804:   surfaces, geometric analysis and symplectic geometry} (Baltimore, MD,

3805:   1999), Adv. Stud. Pure Math. {\bf 34}, Math. Soc. Japan, Tokyo,

3806:   2002, pp.~31--127.}

3807: \bibitem{KS}{M.~Kontsevich, Y.~Soibelman, 

3808: {\em Homological mirror symmetry and torus fibrations}, math.SG/0011041.}

3809: \bibitem{CIL4}{C.~I.~Lazaroiu, ``String field theory and brane superpotentials,'' JHEP {\bf 0110} (2001) 018

3810: [arXiv:hep-th/0107162].}

3811: %%CITATION = HEP-TH 0107162;%%

3812: \bibitem{CIL8}{C.~I.~Lazaroiu and R.~Roiban,

3813: ``Holomorphic potentials for graded D-branes,''

3814: JHEP {\bf 0202} (2002) 038 [arXiv:hep-th/0110288].}

3815: %%CITATION = HEP-TH 0110288;%%

3816: \bibitem{CIL9}{C.~I.~Lazaroiu,

3817: ``An analytic torsion for graded D-branes,''

3818: JHEP {\bf 0209} (2002) 023 [arXiv:hep-th/0111239].}

3819: %%CITATION = HEP-TH 0111239;%%

3820: \bibitem{CIL10}{C.~I.~Lazaroiu and R.~Roiban,

3821: ``Gauge-fixing, semiclassical approximation and potentials for graded

3822: Chern-Simons theories,'' JHEP {\bf 0203} (2002) 022 [arXiv:hep-th/0112029].}

3823: %%CITATION = HEP-TH 0112029;%%

3824: \bibitem{CK}{R. Coquereaux, D. Kastler, ``Remarks on the

3825: differential envelopes of associative algebras'', Pacific. J. Math {\bf 137} (1989) 2.}

3826: \bibitem{Fukaya_inf}{

3827: K.~Fukaya,  ``Morse homotopy, $A_\infty$-category, and Floer

3828:   homologies'', Proc. of GARC Workshop on Geometry and Topology '93

3829:   (H.~J. Kim, ed.), Lecture Notes {\bf 18}, Seoul Nat. Univ., Seoul,

3830:   1993, pp.~1--102.}

3831: \bibitem{Keller}{B.~Keller, ``Introduction to {A}-infinity algebras and modules'',

3832:   Homology, Homotopy and Applications {\bf 3} (2001)1, pp 1--35 [math.RA/9910179]}

3833: \bibitem{Hasegawa}{ K. Lefèvre-Hasegawa ``Sur les A-infini cat\'egories'', Ph.D. Thesis, 

3834: math.CT/0310337.}

3835: \bibitem{Lyubashenko1}{V.~V. Lyubashenko, ``Category of $A_\infty$-categories'',

3836:   Homology, Homotopy and Applications \textbf{5} (2003) 1,pp. 1--48 [math.CT/0210047] }

3837: \bibitem{Lyubashenko2}{ V. Lyubashenko, S. Ovsienko, ``A construction of an A-infinity category'',

3838: math.CT/0211037.}

3839: \bibitem{Lyubashenko3}{V. Lyubashenko, O. Manzyuk, ``Quotients of unital ${A}_\infty$-categories'',

3840: math.CT/0306018.}

3841: \bibitem{Lyubashenko4}{ V. Lyubashenko, O. Manzyuk, ``Free $A_\infty$-categories'',

3842: math.CT/0312339.}

3843: \bibitem{Stasheff}{J.~D. Stasheff, ``Homotopy associativity of H-spaces, I $\&$ II'',

3844:   Trans. Amer. Math. Soc. {\bf 108} (1963), pp. 275--292 and 293--312.}

3845: \bibitem{KS_book}{M. Kontsevich, Y. Soibelman, book in preparation}

3846: \bibitem{Fukaya_book}{K. Fukaya, Y.G. Oh, H. Ohta, K. Ono, 

3847: ``Lagrangin intersection Floer theory - anomaly and obstructon'', preprint. }

3848: \bibitem{CIL6}{C.~I.~Lazaroiu,

3849: ``Graded Lagrangians, exotic topological D-branes and enhanced  triangulated

3850: categories,''JHEP {\bf 0106} (2001)  064 [arXiv:hep-th/0105063].}

3851: %%CITATION = HEP-TH 0105063;%%

3852: \bibitem{CIL7}{C.~I.~Lazaroiu, R.~Roiban and D.~Vaman,

3853: ``Graded Chern-Simons field theory and graded topological D-branes,''

3854: JHEP {\bf 0204}  (2002)023 [arXiv:hep-th/0107063].}

3855: %%CITATION = HEP-TH 0107063;%% 

3856: \bibitem{Witten_CS}{

3857: E.~Witten, ``Chern-Simons gauge theory as a string theory'',

3858: The Floer memorial volume, 637--678, Progr. Math. {\bf 133}, Birkhauser, Basel,

3859: 1995 [hep-th/9207094].}

3860: \bibitem{VG}{V. Ginzburg, ``Double derivations and Cyclic homology'', math.KT/0505236.}

3861: \end{thebibliography}

3862: \end{document}

3863: 

3864: