1: \documentclass[12pt]{article}

2: \usepackage{latexsym}

3: \usepackage{amssymb}

4: %\usepackage{showlabels}

5: \usepackage{graphicx}

6: 

7: %%%%%%%%%%%%%%%%%%% DEFINITIONS %%%%%%%%%%%%%%%%%%%%%%%%

8: \def\be{\begin{equation}}

9: \def\ee{\end{equation}}

10: \def\bear{\begin{eqnarray}}

11: \def\eear{\end{eqnarray}}

12: \def\nn{\nonumber}

13: 

14: 

15: \newcommand\bra[1]{{\langle {#1}|}}

16: \newcommand\ket[1]{{|{#1}\rangle}}

17: 

18: \def\tr{{\mbox{tr}}}

19: \def\Atr{{\mbox{Tr}}}

20: 

21: \def\a{\alpha}

22: \def\b{\beta}

23: \def\g{\gamma}

24: \def\d{\delta}

25: \def\lam{\lambda}

26: \def\u{\mu}

27: \def\v{\nu}

28: \def\r{\rho}

29: \def\t{\tau}

30: \def\z{\zeta}

31: \def\s{\sigma}

32: \def\th{\theta}

33: 

34: \def\Qh{\hat{Q}}

35: \def\baret{\overline{\eta}}

36: \def\homega{{\hat{\omega}}}

37: \def\bpsi{{\overline{\psi}}}

38: \def\wtau{{\widetilde{\tau}}}

39: \def\bth{{\overline{\theta}}}

40: \def\blam{{\overline{\lambda}}}

41: 

42: \def\da{{\dot{\a}}}

43: \def\db{{\dot{\b}}}

44: \def\dg{{\dot{\g}}}

45: 

46: \def\bj{{\overline{j}}}

47: \def\bk{{\overline{k}}}

48: \def\bz{{\overline{z}}}

49: \def\sa{{\hat{a}}}

50: \def\sb{{\hat{b}}}

51: \def\sc{{\hat{c}}}

52: \def\wa{{\tilde{a}}}

53: \def\wb{{\tilde{b}}}

54: \def\wc{{\tilde{c}}}

55: \def\oa{{\overline{a}}}

56: \def\ob{{\overline{b}}}

57: \def\oc{{\overline{c}}}

58: \def\od{{\overline{d}}}

59: 

60: 

61:  \def\CA{{\cal A}}

62:  \def\CC{{\cal C}}

63:  \def\CF{{\cal F}}

64:  \def\CI{{\cal I}}

65:  \def\cJ{{\cal J}}

66:  \def\tJ{{\tilde J}}

67:  \def\tcJ{{\tilde\cal J}} 

68:  \def\CO{{\cal O}}

69:  \def\o{{\rm ord}}

70:  \def\Ph{{\Phi }}

71:  \def\L{{\Lambda}}

72:  \def\CN{{\cal N}}

73:  \def\p{\partial}

74:  \def\pslash{\p \llap{/}}

75:  \def\Dslash{D \llap{/}}

76: % \def\mp{m_{{\rm P}}}

77:  \def\apm{{\a^{\prime}}}

78:  \def\r{\rightarrow}

79: \def\ts{\tilde s}

80: \def\tu{\tilde u}

81:  \def\BR{\IR}

82:  \def\BZ{\IZ}

83:  \def\BC{\IC}

84:  \def\BM{\QM}

85:  \def\BP{\IP}

86:  \def\BH{\QH}

87:  \def\BX{\QX}

88:  \def\sym#1{{{\rm SYM}} _{#1 +1}}

89:  \def\imp{$\Rightarrow$}

90:  \def\IZ{\relax\ifmmode\mathchoice

91:  {\hbox{\cmss Z\kern-.4em Z}}{\hbox{\cmss Z\kern-.4em Z}}

92:  {\lower.9pt\hbox{\cmsss Z\kern-.4em Z}}

93:  {\lower1.2pt\hbox{\cmsss Z\kern-.4em Z}}\else{\cmss Z\kern-.4em Z}\fi}

94:  \def\IB{\relax{\rm I\kern-.18em B}}

95:  \def\IC{{\relax\hbox{$\inbar\kern-.3em{\rm C}$}}}

96:  \def\Ic{{\relax\hbox{$\inbar\kern-.22em{\rm c}$}}}

97:  \def\ID{\relax{\rm I\kern-.18em D}}

98:  \def\IE{\relax{\rm I\kern-.18em E}}

99:  \def\IF{\relax{\rm I\kern-.18em F}}

100:  \def\IG{\relax\hbox{$\inbar\kern-.3em{\rm G}$}}

101:  \def\IGa{\relax\hbox{${\rm I}\kern-.18em\Gamma$}}

102:  \def\IH{\relax{\rm I\kern-.18em H}}

103:  \def\II{\relax{\rm I\kern-.18em I}}

104:  \def\IK{\relax{\rm I\kern-.18em K}}

105:  \def\IP{\relax{\rm I\kern-.18em P}}

106: 

107: 

108: 

109: \def\Tr{{\rm Tr}}

110:  \font\cmss=cmss10 \font\cmsss=cmss10 at 7pt

111:  \def\IR{\relax{\rm I\kern-.18em R}}

112: 

113: \def\wdg{{\wedge}}

114: 

115: \newcommand\ev[1]{{\langle {#1}\rangle}}

116: \newcommand\SUSY[1]{{{\cal N} = {#1}}}

117: \newcommand\diag[1]{{\mbox{diag}({#1})}}

118: \newcommand\com[2]{{\left\lbrack {#1}, {#2}\right\rbrack}}

119: 

120: \newcommand\px[1]{{\partial_{#1}}}

121: \newcommand\qx[1]{{\partial^{#1}}}

122: 

123: \newcommand\rep[1]{{\bf {#1}}}

124: 

125: 

126: \def\gam{{\widetilde{\gamma}}} % Dirac matrices

127: \def\sig{{\sigma}} % Pauli matrices of SO(3,1)

128: \def\hsig{{\hat{\sigma}}} % Pauli matrices of Euclidean SO(4)

129: 

130: \def\eps{{\epsilon}} % epsilon symbol

131: 

132: \def\bZ{{\overline{Z}}}

133: \def\BR{{\mathbb R}}

134: \def\Lag{{\cal L}}

135: \def\cO{{\cal O}}

136: \def\wcO{{\widetilde{\cal O}}}

137: \def\vL{{\vec{L}}}

138: \def\vx{{\vec{x}}}

139: \def\vy{{\vec{y}}}

140: 

141: 

142: \def\vLf{{\vec{\lambda}}}

143: 

144: \newcommand\cvL[1]{{L^{#1}}} % component of $L$

145: 

146: \def\npsi{{\psi^{(inv)}}}

147: \def\bnpsi{{\bpsi^{(inv)}}}

148: \def\wpsi{{\widetilde{\psi}}}

149: \def\bwpsi{{\overline{\widetilde{\psi}}}}

150: \def\hpsi{{\hat{\psi}}}

151: \def\bhpsi{{\overline{\hat{\psi}}}}

152: 

153: \newcommand\nDF[1]{{{D^F}_{#1}}}

154: \def\wA{{\widetilde{A}}}

155: \def\wF{{\widetilde{F}}}

156: \def\wJ{{\widetilde{J}}}

157: \def\hA{{\hat{A}}}

158: \def\hF{{\hat{F}}}

159: \def\hJ{{\hat{J}}}

160: \def\MapL{{\Upsilon}}

161: \def\hR{{\hat{R}}}

162: 

163: \def\cpl{{\lambda}}  % Coupling constant

164: \def\mcr{{\mathcal{R}}}

165: \def\xv{{\vec{x}}}

166: \def\xvt{{\vec{x}^{\top}}}

167: \def\nv{{\hat{n}}}

168: \def\nvt{{\hat{n}^{\top}}}

169: \def\hM{{M}}

170: \def\hgM{{\widetilde{M}}}

171: \def\utr{{\mbox{tr}}}

172: 

173: %%%%%%%%%%%%%%%%%%%%%%%%%%%

174: \newcommand{\fig}[1]{fig.\ (\ref{#1})}

175: \def\dd{\mbox{d}}

176: \def\ddd{\mbox{\sm d}}

177: \def\O{\Omega}

178: \def\o{\omega}

179: \def\bra{\langle}

180: \def\ket{\rangle}

181: \def\a{\alpha}

182: \def\b{\beta}

183: \def\d{\delta}

184: \def\D{\Delta}

185: \def\LL{\triangle}

186: \def\g{\gamma}

187: \def\G{\Gamma}

188: \def\e{\epsilon}

189: \def\ve{\varepsilon}

190: \def\et{\eta}

191: \def\f{\phi}

192: \def\F{\Phi}

193: \def\vf{\varphi}

194: \def\k{\kappa}

195: \def\l{\lambda}

196: \def\L{\Lambda}

197: \def\m{\mu}

198: \def\n{\nu}

199: \def\s{\sigma}

200: \def\S{\Sigma}

201: \def\o{\omega}

202: \def\p{\pi}

203: \def\r{\rho}

204: \def\t{\tau}

205: \def\th{\theta}

206: \def\vt{\vartheta}

207: \def\ra{\rightarrow}

208: \def\la{\leftarrow}

209: \def\pa{\partial}

210: \def\ov{\overline}

211: \def\Pl{\ell_{\sm{Pl}}}

212: \def\tr{\tilde R}

213: \def\td{\tilde d}

214: \def\gmn{g_{\mu \nu}}

215: \newcommand{\ti}[1]{\tilde{#1}}

216: \renewcommand{\^}[1]{\hat{#1}}

217: \newcommand{\sm}[1]{\mbox{\scriptsize #1}}

218: \newcommand{\tn}[1]{\mbox{\tiny #1}}

219: \renewcommand{\@}[1]{\sqrt{#1}}

220: \renewcommand{\le}[1]{\label{#1}\end{eqnarray}}

221: \newcommand{\bea}{\begin{eqnarray}}

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

223: \newcommand{\ep}{e^{2 \phi}}

224: \newcommand{\eq}[1]{(\ref{#1})}

225: \def\nn{\nonumber\\}

226: \def\nm{\nonumber}

227: \newcommand{\rf}[1]{\cite{ref:#1}}

228: \newcommand{\rr}[1]{\bibitem{ref:#1}}

229: \def\qu{\ {\buildrel{\displaystyle ?} \over =}\ }

230: \def\smqu{\ {\buildrel ?\over =}\ }

231: \def\ffract#1#2{\raise .35 em\hbox{$\scriptstyle#1$}\kern-.25em/

232: \kern-.2em\lower .22 em \hbox{$\scriptstyle#2$}}

233: \def\GN{G_{\mbox{\tn N}}}

234: \def\lPl{\ell_{\mbox{\tn{Pl}}}}

235: \def\fn{f_{(0)}}

236: \def\fe{f_{(1)}}

237: \def\ft{f_{(2)}}

238: \def\fd{f_{(3)}}

239: \def\Ric{{\mbox{Ric}}}

240: \def\Rie{{\mbox{Rie}}}

241: \def\Ein{{\mbox{Ein}}}

242: \def\nl{\newline}

243: \def\cl{\textcolor}

244: \def\clb{\colorbox}

245: \def\na{\nabla}

246: \def\half{{1\over2}\,}

247: \setlength{\oddsidemargin}{0in}\setlength{\textwidth}{6.3in}

248: \setlength{\topmargin}{0in}\setlength{\textheight}{9in}

249: \def\nonu{\nonumber \\{}}

250: \def\da{\dot{a}}

251: \def\db{\dot{b}}

252: \def\dc{\dot{c}}

253: \def\tg{\tilde{\g}}

254: \def\bF{\bar{F}}

255: 

256: 

257: %%%%%%%%%%%%%%%%%%%%%%%%%%%

258: 

259: 

260: 

261: %%%

262: \def\vwsX{{\bf X}} % string worldsheet coordinate

263: \def\dvwsX{{\dot{\bf X}}} % string worldsheet coordinate

264: \newcommand\vxmod[1]{{{\bf X}_{#1}}} % string worldsheet mode

265: \newcommand\dvxmod[1]{{\dot{{\bf X}}_{#1}}} % time derivative

266: 

267: \def\vwsP{{\bf \Pi}} % string worldsheet conjugate momentum

268: 

269: 

270: \def\wsX{{X}} % string worldsheet coordinate

271: \def\wsZ{{Z}} % string worldsheet coordinate

272: \def\bwsZ{{\overline{Z}}} % string worldsheet coordinate

273: \def\wsPsi{{\Psi}} % string worldsheet coordinate

274: \def\bwsPsi{{\overline{\Psi}}} % string worldsheet coordinate

275: 

276: \def\twsZ{{\widetilde{Z}}} % string worldsheet coordinate

277: \def\twS{{\widetilde{S}}} % string worldsheet coordinate

278: \def\btwsZ{{\overline{\widetilde{Z}}}} % string worldsheet coordinate

279: \def\twsPsi{{\widetilde{\Psi}}} % string worldsheet coordinate

280: \def\btwsPsi{{\overline{\widetilde{\Psi}}}} % string worldsheet coordinate

281: \def\tV{{\widetilde{V}}}

282: \def\talp{{{\tilde{\a}}'}} % S-dual Regge slope

283: 

284: \def\hLf{{\hat{\lambda}}}

285: \def\hL{{\hat{L}}}

286: 

287: % distance from one line to the next

288: %\setlength{\baselineskip}{20pt}

289: 

290: 

291: %%                              TABLEAUX.TEX

292: %%      This  macro file is for producing a ``Young Tableau'' which is

293: %%      an array of little squares sometimes used in mathematical physics.

294: %%      For instance, the command $\tableau{6 3 2}$ will produce a tableau

295: %%      with 6 squares in the top row, 3 in the next, and 2 in the last.

296: %%                                  OOOOOO

297: %%      This tableau will look like OOO    but made of squares instead of O's.

298: %%                                  OO

299: %%      Any number of rows may be present, each having a nonzero number of

300: %%      squares.

301: %%

302: %%      A tableau is math mode material, so use $ or $$ to enclose it.

303: %%

304: %%      The size and line-thickness of the little boxes are controlled by the

305: %%      dimension parameters --

306: %%              \tableauside=1.0ex              %(size)

307: %%              \tableaurule=0.4pt              %(line-thickness)

308: %%      Change them if you want.

309: %%

310: %%                                                      -- Doug Eardley 9/19/8%%

311: %%

312: \newdimen\tableauside\tableauside=1.0ex

313: \newdimen\tableaurule\tableaurule=0.4pt

314: \newdimen\tableaustep

315: \def\phantomhrule#1{\hbox{\vbox to0pt{\hrule height\tableaurule width#1\vss}}}

316: \def\phantomvrule#1{\vbox{\hbox to0pt{\vrule width\tableaurule height#1\hss}}}

317: \def\sqr{\vbox{%

318:   \phantomhrule\tableaustep

319:   \hbox{\phantomvrule\tableaustep\kern\tableaustep\phantomvrule\tableaustep}%

320:   \hbox{\vbox{\phantomhrule\tableauside}\kern-\tableaurule}}}

321: \def\squares#1{\hbox{\count0=#1\noindent\loop\sqr

322:   \advance\count0 by-1 \ifnum\count0>0\repeat}}

323: \def\tableau#1{\vcenter{\offinterlineskip

324:   \tableaustep=\tableauside\advance\tableaustep by-\tableaurule

325:   \kern\normallineskip\hbox

326:     {\kern\normallineskip\vbox

327:       {\gettableau#1 0 }%

328:      \kern\normallineskip\kern\tableaurule}%

329:   \kern\normallineskip\kern\tableaurule}}

330: \def\gettableau#1 {\ifnum#1=0\let\next=\null\else

331:   \squares{#1}\let\next=\gettableau\fi\next}

332: 

333: \tableauside=1.0ex

334: \tableaurule=0.4pt

335: 

336: \newcommand{\Section}[1]{\section{#1} \setcounter{equation}{0}}

337: \makeatletter

338: \@addtoreset{equation}{section}

339: \makeatother

340: \renewcommand{\theequation}{\thesection.\arabic{equation}}

341: 

342: 

343: 

344: \begin{document}

345: 

346: 

347: \begin{flushright}

348: %\rightline{\today}

349: KCL-MTH-07-02\\

350: UPR-1163-T\\

351: \end{flushright}

352: \vskip0.1truecm

353: 

354: 

355: 

356: \begin{center}

357: \vskip 1truecm {\Large \textbf{Electric-magnetic Duality \\

358: \vskip .3truecm

359: and Deformations of Three-Dimensional CFT's}}

360: \vskip 2truecm

361: %\vfill

362: 

363: {\large Sebastian de Haro${}^\star$ and Peng Gao${}^\dagger$}\\

364: 

365: 

366: \vskip .4truecm

367: \vskip 4truemm ${}^\star${\it Department of Mathematics\\

368: King's College, London WC2R 2LS, UK}\\

369: \tt{sebastian.deharo@kcl.ac.uk}\\

370: \vskip .3truecm

371: $^\dagger${\it Department of Physics and Astronomy\\

372: University of Pennsylvania\\

373: Philadelphia, PA 19104-6396, USA}\\

374: {\tt gaopeng@physics.upenn.edu}

375: 

376: \end{center}

377: \vskip 1truecm

378: 

379: \begin{center}

380: 

381: \textbf{\large \bf Abstract}

382: 

383: \end{center}

384: 

385: $SL(2,{\mathbb{Z}})$ duality transformations in asymptotically AdS$_4\times S^7$ act non-trivially on the

386: three-dimensional SCFT of coincident M2-branes on the boundary. We show how $S$-duality acts away from the

387: IR fixed point. We develop a systematic method to holographically obtain the deformations 

388: of the boundary CFT and show how electric-magnetic duality relates different deformations.

389: We analyze in detail marginal deformations and deformations by dimension 4 operators. In 

390: the case of massive deformations, the RG flow relates $S$-dual CFT's. Correlation functions in the CFT are 

391: computed by varying magnetic bulk sources,

392: whereas correlation functions in the dual CFT are computed by electric bulk sources. Under 

393: massive deformations, the boundary 

394: effective action is generically minimized by massive self-dual configurations of the 

395: $U(1)$ gauge field. We show that a self-dual choice of boundary conditions exists, and it 

396: corresponds to the self-dual topologically massive gauge theory in 2+1 dimensions. Thus, self-duality in 

397: three dimensions can be understood as a consequence of electric-magnetic invariance in the 

398: bulk of AdS$_4$.

399: 

400: 

401: 

402: \newpage

403: 

404: \tableofcontents

405: 

406: %\newpage

407: 

408: 

409: 

410: \section{Introduction and summary of the results}

411: 

412: Electric-magnetic duality has played an important role in understanding non-perturbative 

413: aspects of supersymmetric gauge theories and string theory. It was instrumental in the description of 

414: confinement by monopole condensation in ${\cal N}=2$ SYM theory by Seiberg and Witten \cite{SW}. In the 

415: case ${\cal N}=4$ SYM \cite{VW}, it is related via AdS/CFT to the $SL(2,{\mathbb{Z}})$ 

416: invariance of type IIB string theory. The latter puts very strong constraints on the form of

417: the coefficients or the effective action, as it relates the string tree-level and one-loop 

418: terms to instanton contributions \cite{green}. The topologically twisted version of 

419: ${\cal N}=4$ has recently been shown to realize the basic geometric Langlands 

420: correspondence \cite{KW}.

421: 

422: The question whether four-dimensional gauge theories coupled to gravity exhibit  

423: duality properties has a long history \cite{CJ}. Gravity itself in four dimensions is known 

424: to be invariant under rotations of the linearized Riemann tensor and its dual \cite{HT}. 

425: For toroidal compactifications of M-theory, generalized electric-magnetic dualities are 

426: well-known \cite{CJ,hull}. 

427: In this paper we will consider compactifications to asymptotically AdS$_4\times S^7$. 

428: Finding dualities in this case might mean significant progress towards understanding the 

429: physical properties of the theory describing interacting M2-branes at strong coupling.

430: 

431: An important source of motivation comes from \cite{Wittensl2z}, where it was argued that 

432: the $SL(2,{\mathbb{Z}})$ duality of abelian gauge fields in the bulk of AdS$_4$ relates 

433: seemingly different CFT's to each other. The CFT's in question turn out to be relevant for 

434: the quantum Hall effect, where the $S$-duality of \cite{Wittensl2z} had independently been 

435: found \cite{BD1} and  verified experimentally. In retrospect, 

436: this might be seen as an experimental prediction of AdS/CFT. These modular properties have 

437: recently been used to propose new predictions for the quantum Hall effect in graphene 

438: \cite{BD2}. The $S$-duality of \cite{Wittensl2z} is also related to 

439: the IR limit of three-dimensional mirror symmetry \cite{KS,BKW}.

440: %which exchanges masses and 

441: %FI-terms, twisted vector multiplets and hypermultiplets, and Coulomb and Higgs 

442: %branches. 

443: It implies that electrically charged particles in one theory correspond to vortices 

444: in the dual.

445: 

446: In this paper we develop the holographic map between $SL(2,{\mathbb{Z}})$ transformations

447: in the bulk and dualities between three-dimensional effective CFT's away from the 

448: conformal fixed point. The massive deformations are induced by generalized boundary conditions in

449: the bulk of AdS. We show that under RG flow towards the IR, the IR theory is described by a dimension

450: 2 current which is $S$-dual to the UV current. The UV current-current correlator is given in terms of the dual 

451: gauge field. This is in agreement with the field 

452: theory predictions of \cite{tassos}.

453: 

454: Before we describe our results in more detail, let us give our third piece of motivation. It

455: comes from the fact that conformal theories with instanton solutions in AdS have rather 

456: special properties. One can often write an effective boundary action for them, 

457: whose classical solutions reproduce the bulk instantons. The exact quantum effective action can be computed from the 

458: bulk at leading order in $N$. The fact that one is dealing with instantons implies that the stress-energy tensor is zero and there is no back reaction. In \cite{sebastassos} it was shown in the context of a toy model that 

459: for a conformally coupled scalar field with a quartic interaction potential the classical 

460: bulk solutions can be matched exactly with classical solutions of a boundary effective 

461: field theory. The boundary field theory had appeared earlier in \cite{HH}. The conformally

462: coupled scalar field was embedded in M-theory in \cite{dHPP}, where it was also shown that

463: the boundary effective action can be exactly computed using AdS/CFT.

464: 

465: In this paper we continue this program of conformal holography \cite{sebastassos}  for the

466: case of abelian gauge fields. Again, the boundary effective action can be exactly computed 

467: for the class of solutions under study. The large $N$ effective boundary theory which 

468: reproduces the bulk instantons is the self-dual topologically massive three-dimensional theory of Pilch, Townsend and van Nieuwenhuizen \cite{PTvN,DJ}. Conformal holography is expected to play a role for higher spins also.

469: 

470: We now summarize our main results. At the level of classical bulk solutions, we will give

471: the most general solution for a $U(1)$ gauge field in pure AdS$_4$, and compute the 

472: leading back-reaction effects in the Fefferman-Graham expansion. The on-set of back-reaction is one order up 

473: from the Fefferman-Graham ambiguity, and for the purpose of computing two-point functions the 

474: effective action can be computed by the method of holographic renormalization \cite{dHSS} without using 

475: back-reaction. 

476: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

477: %This is due to conformal invariance as well as the absence of conformal anomalies in three 

478: %dimensions, and constitutes an important simplification. 

479: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%5

480: We also obtain the regularity condition for Euclidean gauge 

481: fields, which does not seem to have appeared in the literature. This regularity condition relates

482: the boundary value of the electric field to the gauge invariant part of the gauge field.

483: In momentum space it takes the form

484: \be\label{1}

485: f_i(p)=-|p|\,A_i^{\sm{T}}(p)~,

486: \ee

487: where $f_i(p)$ is the boundary value of the electric field (see notation in the next subsection).

488: We explain the physical meaning of this boundary condition in Lorentzian signature: it is the

489: condition that waves travel from the boundary into the bulk. In terms of the original M2-geometry,

490: this is the condition that all matter falls into the M2-brane. We will show that self-dual

491: bulk solutions are in one-to-one correspondence with a choice of what we call ``self-dual

492: boundary conditions'':

493: \be\label{2}

494: f_i(x)=\pm \e_{ijk}\pa_jA_k(x)

495: \ee

496: for (anti-)SD solutions, respectively. Combining \eq{1} (or a massive version of it) and

497: \eq{2} gives the self-dual topogically massive theory of \cite{PTvN,DJ}.

498: We show that, 

499: although in flat space there are no abelian instantons, this is not necessarily true in

500: AdS: configurations such as \eq{1}-\eq{2} can lead to finite Euclidean action. The theory of

501: \cite{PTvN,DJ} has a single massive degree of freedom, and the on-shell value of the action

502: is explicitly written in terms of it.

503: 

504: The bulk theory is electric-magnetic invariant only up to boundary terms. We will compute

505: these explicitly for various choices of boundary conditions. Boundary conditions are

506: imposed by the addition of boundary terms in the action, corresponding to deformations of

507: the CFT. We will consider marginal operators and operators of classical dimension 4, and

508: show how electric-magnetic duality maps different operators into each other. We identify

509: configurations which are self-dual under duality, and these include the self-dual configurations

510: of \cite{PTvN,DJ}.

511: 

512: Scalars in the particular range of masses  $-d^2/4<m^2<-d^2/4+1$ in AdS$_{d+1}$ are well-known to admit two kinds of 

513: normalizable solutions, which give rise to two possible quantization schemes and hence two different

514: boundary interpretations, one in terms of an operator of dimension $\D_+$ and the other in

515: terms of an operator of dimension $\D_-$. In the conformal case we will be interested in, and in $d=3$,

516: the dimensions are $\D_+=2$ and $\D_-=1$. Four-dimensional gauge fields have the same property: both modes are 

517: normalizable, and two quantization schemes are available, corresponding to Dirichlet or Neumann boundary conditions. 

518: For Dirichlet boundary conditions where the boundary value of the gauge field (or, equivalently, the magnetic

519: field) is held fixed, the holographic interpretation is the usual one: $A$ is a source for a global

520: symmetry current in the CFT. In the Neumann quantization scheme, the electric field is held fixed instead.

521: The dual operator has dimension $\Delta_-=1$ and is identified with a gauge field that corresponds to the 

522: boundary value of the gauge field $A$. This does not present a problem because one can always construct from it

523: a conserved current of dimension 2 which saturates the unitarity bound $\D\geq2$. Observables of the theory are then correlation functions of the

524: new current. It is dual to the boundary value of the bulk magnetic field. Witten showed 

525: \cite{Wittensl2z} that the bulk $S$-duality operation induces an $S$-duality operation on the CFT that consists of 

526: first gauging the global $U(1)$ symmetry, promoting $A$ to a dynamical field, and coupling it to an external field 

527: $v$ via a Chern-Simons coupling. This corresponds precisely to the above discussion of boundary conditions. A third choice of boundary condition is possible in this case, a mixed one, which 

528: corresponds to a multi-trace deformation of the CFT. In this paper we will examine $S$-duality in the context of 

529: massive deformations of the CFT which induce a renormalization group flow. We will find an interesting 

530: interpretation of these dualities in 

531: terms of bulk sources. If we use the usual bulk magnetic source (one that couples to the electric field), the 

532: correlators that we find are the usual ones. On the other hand, if we use a bulk electric source (one 

533: that couples to the gauge field), we find the $S$-dual correlators.  As we mentioned, RG flow

534: interchanges these two. We also give a bulk proof of Witten's CFT argument that $S$-duality acts

535: as a Legendre transform \cite{Wittensl2z} (this was first shown in \cite{HUY}, where marginal deformations were also discussed). In fact, there are two types of Legendre transforms,

536: one that changes the dimension of the operator and one that doesn't. We interpret these dualities as the

537: particle-vortex duality of \cite{BD1}. The generic IR effective action that we find is the

538: one of QED in three dimensions at large $N_f$. Although this basically follows from conformal invariance,

539: the fact that the qualitative behavior under massive deformations is similar indicates that

540: QED may indeed play a role in the IR for the $U(1)$ gauge fields studied here. The supergravity

541: configuration that we study is ${\cal N}=2$ supergravity in four dimensions, which is 

542: obtained as an M-theory compactification on an $S^7$ where all scalar fields have been 

543: projected out. The still unknown dual SCFT is the strong-coupling limit of 2+1 dimensional SYM theory

544: in the large $N$ limit. We compute its effective action, and a relation with the 

545: large $N_f$ limit of QED is suggested. Three-dimensional $S$-duality of $U(1)$ gauge theories including Chern-Simons terms has been discussed earlier in a somewhat different context in \cite{KOO}.

546: 

547: The plan of the paper is as follows. In section 2 we review the discussion of scalar fields

548: in the special range of masses mentioned above. We will use many of those properties

549: in the gauge field case as well. In that section we also give the basic classical solutions

550: that we will use and apply the method of holographic renormalization to compute the effective

551: action. In section 3 we discuss electric-magnetic duality in AdS$_4$. We give a novel proof of

552: $S$-duality based on the first order formalism and show how the boundary terms change under

553: duality. Section 4 is devoted to the deformations of the boundary conditions and the deformations

554: of the CFT that they induce. In section 5 we show how electric-magnetic transformations

555: act on the general boundary conditions. In section 6 we discuss the interplay of $S$-duality,

556: Legendre transforms, and RG-flow. The results are summarized in figure \ref{diagram} on

557: page 37. In that section we compare to the known CFT results and comment on the non-abelian

558: generalization. Some conclusions and future lines of investigation are presented in section

559: 7. In five appendices we present details concerning $S$-duality, the various solutions, holographic renormalization,

560: and supersymmetry of the solutions.

561: 

562: 

563: \subsection*{Notation and conventions}\label{notation}

564: 

565: $A_i(r,x)$ will denote the bulk gauge field and $A_i(x)=A_{(0)i}(x)$ its restriction to the boundary at $r=0$. The indices $i=0,1,2$ run over the boundary. The electric field will be denoted by $E_i(r,x)=F_{ri}(r,x)$ and its boundary component is $f_i(x)=E_i(0,x)$. We will use the following field redefinitions by a $\th$-term:

566: \bea

567: {\sf{E}}&=&E-\th\,F\nn

568: {\sf{f}}&=&f-\th\,F\nn

569: {\sf{v}}&=&v-\th\,A~.

570: \eea

571: We will use form notation for three-dimensional quantities, and will use a short-hand notation for contractions: $Af=A_if^i$ etc. Boundary indices of $r$-dependent quantities are raised and lowered with $g_{ij}(r,x)$, indices of quantities evaluated at $r=0$ are raised and lowered with $g_{(0)ij}$. We will use a projector onto the part of the gauge field transverse to the momentum:

572: \be\label{id2}

573: \Pi_{ij}=g_{ij(0)}-{\pa_i\pa_j\over\Box}~,~A^{\sm{T}}_i=\Pi_{ij}A_j=(\Pi A)_i~.

574: \ee

575: We will denote $M=\Box^{1/2}=\sqrt{|\Box|}$ which both in Lorentzian and Euclidean signature is a positive definite quantity ($M=|p|$ in momentum space). The quantity $t$ will be +1 in Euclidean signature, and -1 in Lorentzian signature.

576: 

577: The boundary magnetic field will be denoted by a vector $F$,

578: \be

579: F_i=\half\e_{ijk}F_{jk}=\e_{ijk}\pa_j A_k~.

580: \ee

581: 

582: 

583: Twiddles denote Legendre transforms, tildes denote $S$-duality. So $\ti S$ is the Legendre transformed action, $S'$ denotes its $S$-dual.

584: 

585: Boundary operators will be denoted ${\cal O}_{\D i}(x)$, with $\D=1$ for gauge fields and $\D=2$ for currents.

586: Boundary sources are denoted by $J_i(x)$.

587: 

588: \section{The bulk solution}

589: 

590: \subsection{Duality for scalar fields}\label{scalardualities}

591: 

592: 

593: 

594: In this section we review some facts known for scalar fields in AdS$_4$ which will set the stage of the discussion of gauge fields in the rest of the paper. It will also serve to fix notation.

595: 

596: Scalar fields in AdS$_{d+1}$ with mass within the range

597: \be\label{massrange}

598: -{d^2\over4}<m^2<-{d^2\over4}+1

599: \ee

600: are well-known to admit two types of possible boundary conditions at the boundary preserving the asymptotic symmetry group and leading to finite energy \cite{isham,bf,Breitenlohner:1982jf}. For definiteness, consider the Fefferman-Graham form of the metric

601: \be\label{FG}

602: \dd s^2={\ell^2\over r^2}\left(\dd r^2+g_{ij}(r,x)\,\dd x^i\dd x^j\right)

603: \ee

604: where $g_{ij}(r,x)$ has an expansion in the distance $r$ to the boundary, such that at the boundary $r=0$ the metric is the Poincare metric of AdS. A scalar field of mass $m$ will now have the asymptotic expansion

605: \be

606: \f=r^{\D_-}\f_{(0)}(x)+\ldots+r^{\D_+}\f_{(2\D-d)}(x)+\ldots

607: \ee

608: where $\D_\pm={d\over2}\pm\sqrt{{d^2\over4}+m^2}$. $\f_{(0)}$ and $\f_{(2\D-d)}$ are not determined by the field equations. They are boundary conditions and have a holographic CFT interpretation as the source and 1-point function of an operator of dimension $\D_+$, respectively, to which the source is coupled. For the range of masses \eq{massrange}, it was argued \cite{klebanovwitten} that the two quantization schemes,

609: \bea\label{scDN}

610: \d\f_{(0)}(x)&=&0~~~~~~{\mbox{Dirichlet}}\nn

611: \d\f_{(2\D-d)}(x)&=&0~~~~~~{\mbox{Neumann}}~,

612: \eea

613: correspond to two possible boundary CFT's. The usual CFT is the one discussed above, where a source $\f_{(0)}$ couples to an operator of dimension $\D=\D_+$. The usual AdS/CFT recipe can then be summarized as (up to terms that contribute contact terms in the two-point functions and which vanish identically in our case):

614: \bea\label{recipe}

615: \f_{(0)}&=&J(x)\nn

616: S_{\sm{on-shell}}[\f_{(0)}]&=&-W[J]\nn

617: \bra {\cal O}_{\D}(x)\ket_J&=&-{\d W\over\d J}={\d S_{\sm{on-shell}}\over\d\f_{(0)}(x)}=(2\D-d)\,\f_{(2\D-d)}\nn

618: \bra {\cal O}_{\D}(x){\cal O}_{\D}(x')\ket_J&=&-{\d^2 S_{\sm{on-shell}}\over\d\f_{(0)}(x)\d\f_{(0)}(x')}+\bra {\cal O}_{\D}(x)\ket\bra {\cal O}_{\D}(x')\ket~,

619: \eea

620: and likewise for the higher-point functions. The signature here is Euclidean. Our conventions are such that in the Lorentzian we have $S_{\sm{on-shell}}=W$ and hence $\bra {\cal O}_{\D}\ket=-(2\D-d)\f_{(2\D-d)}$.

621: 

622: Thus, the renormalized bulk on-shell action is interpreted as minus the (renormalized) generating functional of connected correlation functions in the CFT, $W$, at leading order in $N$. We can also define the effective action of the boundary CFT, which is its Legendre transform:

623: \be

624: \Gamma[\f_+]=-\int\dd^dx\,\phi_-(x)\phi_+(x) +W[\f_-]~.

625: \ee

626: 

627: The second CFT is what we call the ``dual CFT''  \cite{klebanovwitten}, and it contains an operator of dimension $\D_-$, which couples to a source of dimension $\D_+$. This theory is defined as the Legendre transform of the usual CFT,

628: \be

629: W[\f_-]=\ti W[\f_+]+\int\dd^dx\,\f_-\f_+~.

630: \ee

631: Its effective action $\ti\Gamma$ is given by

632: \be

633: \ti W[\f_+]=-\ti\G[\f']+\int\dd^dx\,\f_+\f'~.

634: \ee

635: Comparing both equations, we see that the effective action of the dual CFT is identified with the bulk action, for $\f'=-\f_-$, and the generating functional of the dual CFT is identified with the effective action of the usual CFT:

636: \be

637: \ti\G[-\f_-]=-W[\f_-]=S[\f_-]~,~\ti W[\f_+]=\G[\f_+]~.

638: \ee

639: where $\f_-$ is interpreted as the expectation value of an operator of dimension $\D_-$ and $\ti J=\f_+$ is the dual source. The operator\footnote{For the conformal equivalence between $\f_+$ and $\f_-$, see \cite{Dobrev}.} is given by:

640: \be

641: \bra{\cal O}_{\D_-}\ket_{\ti J}={\d\ti W[\f_+]\over\d\f_+}~.

642: \ee

643: The fact that we are in the mass range \eq{massrange} ensures that $\D_-$ is above the unitarity bound $\D_->{d-2\over2}$. 

644: 

645: For spin-1 fields we have $\D_\pm={d\over2}\pm\sqrt{{(d-2)^2\over4}+m^2}$, hence for massless spin-1 fields in AdS$_4$ we get $\D_-=1$, $\D_+=2$ as in the case of conformally coupled scalar fields. As we will see, this fact implies that massless gauge fields and conformally coupled scalar fields share many important properties. For scalars in the range of masses \eq{massrange}, a generalized choice of boundary conditions is possible where the total energy is finite and conserved \cite{HM,designer}:

646: \be\label{genbc}

647: \f_+=\f_+(\f_-)={\cal F}'(\f_-)

648: \ee

649: for some function ${\cal F}'$. Since $\f_-$ is itself a gauge invariant operator in the CFT, \eq{genbc} deforms the boundary theory by a multiple-trace operator \cite{wittendt}. For example, in the case at hand, in order to obtain a generalized boundary condition that preserves all of the AdS isometries,

650: \be\label{trtr}

651: \f_+=\a\,\f_-^2

652: \ee

653: we need to deform the CFT by a marginal triple-trace operator $\int\dd^3x\,\f_-^3(x)$.

654: 

655: In order to allow for generalized boundary conditions \eq{genbc}, the variational principle needs to be modified. Indeed, the usual variational principle implies either Dirichlet or Neumann boundary 

656: conditions, which correspond to the $\a=\infty$ and $\a=0$ limits of \eq{trtr}, respectively. In  \cite{MSS}, and 

657: independently in \cite{designer}, a recipe was given where the variational principle is well-defined and 

658: automatically leads to boundary condition \eq{genbc}. This amounts to adding a boundary term to the action such that 

659: the boundary condition is enforced. Let us see this in some detail. Under usual Dirichlet/Neumann boundary conditions 

660: \eq{scDN}, the bulk path integral computes the CFT correlator with a {\it fixed} external source,

661: \be

662: Z[J]=\int_{\f_{\mp}=J}{\cal D}\f\,e^{-S[\f]}=\bra e^{\int\sm{d}^dx\,J(x){\cal O}_{\D_{\pm}}(x)}\ket_{\sm{CFT}}~,

663: \ee

664: and integration over all the bulk fields should be understood here. On the other hand, the boundary condition \eq{genbc} is achieved if we integrate over both the bulk and the boundary values of the scalar field:

665: \be\label{pathi}

666: Z=\int{\cal D}\f_-\int_{\f|_{r=0}=\f_-}{\cal D}\f\,e^{-S[\f]+S_{\mbox{\tiny{bdy}}}[\f_-]}=\int{\cal D}J\,e^{S_{\sm{bdy}}[J]}\bra 

667: e^{\int\sm{d}^dx\,J(x){\cal O}_{\D_+}(x)}\ket_{\sm{CFT}}~.

668: \ee

669: Thus, the path integral is evaluated without any boundary conditions \cite{MSS}. In general, the saddle point approximation for 

670: \eq{pathi} gives two terms. The first one is the bulk equations of motion. The second one is proportional to

671: \be

672: -\int\dd^3x\,\f_+\d\f_-+\d S_{\sm{bdy}}[\f_-]

673: \ee

674: in other words, excluding the trivial case of Dirichlet boundary conditions, the general boundary condition is

675: \be

676: \f_+={\d S_{\sm{bdy}}\over\d \f_-}[\f_-]~,

677: \ee

678: again for the relevant case of the conformally coupled scalar field in AdS$_4$. In \cite{MSS} it was shown that, for 

679: quadratic $S_{\sm{bdy}}[\f_-]$, the modified two-point functions extracted from this recipe agree with those of the 

680: field theory where an explicit double-trace deformation is introduced in the action. In this paper we will use a 

681: similar procedure for gauge fields.

682: 

683: \subsubsection*{Lorentzian signature}

684: 

685: In Lorentzian signature there are a number of subtleties in the above picture. These have 

686: to do with the fact that propagating, normalizable modes now exist, and hence 

687: the solution is not uniquely determined by the boundary condition at the boundary plus regularity of the solution 

688: at infinity. In particular, the functional relation between $\f_+$ and $\f_-$ is no longer fixed. The propagating modes

689: correspond to a non-trivial choice of state in the boundary theory, and their wave-functions have to be included in the 

690: path integral \cite{don}. The solution that is analytically continued from the Euclidean corresponds to the vacuum state -- in that

691: case, the saddle point approximation to the bulk partition function gives the Euclidean CFT partition function, which when

692: analytically continued gives the vacuum amplitude. In the Lorentzian, choosing any other solution than the regular one gives

693: an amplitude between non-trivial states.

694: 

695: %In fact, as explained in \cite{don}, the states {\it are} determined by boundary conditions: the boundary condition

696: %on the spacelike surface $\S_{\pm}$ that bounds the bulk and intersects the boundary. In summary, in the Lorentzian case

697: %we have one boundary condition at $r=0$ and one at $t=t_\pm$.

698: 

699: In this paper we will largely use the regular boundary condition. In section \ref{regularitysec} we will derive this condition for gauge fields and give a purely Lorentzian interpretation of it: instead of as a boundary condition at $\S_{\pm}$, it can be regarded as a condition for the flux to be purely incoming at the boundary. This is the natural boundary condition in the original M2-brane geometry. This implies that we can in fact impose this as an independent boundary condition

700: at the boundary. 

701: 

702: \subsection{Gauge fields: solutions and dual interpretation}

703: 

704: We will now solve the equations of motion for the gauge field. For later reference, we give the Lorentzian form of the action:

705: \be\label{actionwtheta}

706: S[A]=\int\dd^4x\,\sqrt{-g}\left(-{1\over4g^2}F_{\m\n}^2+{\th\over32\pi^2}\,\e^{\m\n\l\s}F_{\m\n}F_{\l\s}\right)~.

707: \ee

708: The $\th$-term is the appropriate one for a spin manifold. It gives a Chern-Simons term at level 1/2. If the manifold is not spin, the coefficient would have an additional factor of 2. The equations of motion that follow from the above matter part of the action are:

709: \be

710: \nabla^\m F_{\m\n}=0~,

711: \ee

712: In the Fefferman-Graham coordinate system \eq{FG}, we can do the following asymptotic expansion for a massless field of spin 1:

713: \bea

714: A_r(r,x)&=&A_r^{(0)}(x)+r\,A_r^{(1)}(x)+r^2\,A_r^{(2)}(x)+\ldots\nn

715: A_i(r,x)&=&A_i^{(0)}(x)+r\,A_i^{(1)}(x)+r^2\,A_i^{(2)}(x)+\ldots

716: \eea

717: Solving Maxwell's equations then determines all the higher coefficients $A_r^{(n)}, A_i^{(n)}$, $n\geq2$, in terms of 

718: the boundary conditions $A_i^{(0)}(x),A_i^{(1)},A_r^{(0)},A_r^{(1)}$, as well as the metric. The radial components 

719: $A_r^{(0),(1)}$ are of 

720: course not real degrees of freedom as they can always be gauged away. Notice that $A_i^{(1)}$ is not left totally undetermined by the equations of motion. There is a constraint on its gauge-invariant part, which for future convenience we will denote $f_i(x)$:

721: \be

722: \nabla_i^{(0)}f^i(x)=0~,

723: \ee

724: where the covariant derivative is defined with respect to the boundary metric $g_{(0)ij}$,

725: boundary indices are raised and lowered with respect to this metric, and where we defined

726: \bea

727: f_i(x)&=&A_i^{(1)}(x)-\pa_iA_r^{(0)}(x)\nn

728: F_{ri}(r,x)&=&f_i(x)+{\cal O}(r)~,

729: \eea

730: with a conventional minus sign in Lorentzian signature.

731: Of course, $f_i(x)$ is nothing but the boundary value of the electric field $E_i=F_{ri}$. The fact that the bulk equations only fix $f_i(x)$ up to an arbitrary conserved vector is called the Fefferman-Graham ambiguity.

732: 

733: Since $f_i(x)$ is conserved, in three dimensions and if $H^1(M)=0$ where $M$ is the boundary, we can introduce a vector $v_i$ such that

734: \be

735: f^i=\e^{ijk}_{(0)}\pa_jv_k~.

736: \ee

737: Again, the epsilon tensor is defined with respect to $g_{(0)ij}$.

738: From now on, we will write $A_i(x)$ instead of $A_i^{(0)}(x)$ for the boundary value of the 

739: bulk field $A_i(r,x)$.

740: 

741: We can take advantage of conformal invariance to completely solve the Maxwell equations 

742: throughout the bulk. In Appendix \ref{regularityapp} we do this for pure AdS. The general solution 

743: in that case is

744: \be\label{Aiexpansion}

745: A_i^{\tiny{\mbox{T}}}(r,p)=A_i^{\tiny{\mbox{T}}}(p)\cosh(|p|r)+{1\over|p|}\,f_i(p)\sinh(|p|r)~,

746: \ee

747: in momentum space. Here $A_i^{\tiny{\mbox{T}}}$ is the transverse, gauge invariant part of 

748: the gauge field, $A^{\tiny{\mbox{T}}}_i(p)=\Pi_{ij} A_j(p)$ with the projector defined in 

749: \eq{id2}. By construction, $f_i(x)$ is also gauge invariant.

750: 

751: In this paper we will take advantage of the holographic interpretation in the dual CFT that is relevant to the Neumann and 

752: mixed boundary problem, as discussed in sections 1 and 2.1. In the standard CFT, the boundary values of the gauge field 

753: $A$ (or alternately the magnetic field) are fixed in terms of a boundary source for a global $U(1)$ current. The 

754: boundary values of the electric field correspond to the conserved current. In the dual CFT, the global symmetry has 

755: been gauged and $A$ has been promoted to a dynamical field that is integrated over in the path integral. Its one-point 

756: function corresponds holographically to the boundary value of the (transverse part of the) gauge field. It gives 

757: rise to a new conserved current $\ti f=*\dd A$ which corresponds to the magnetic field and which is the new

758: gauge invariant observable of the theory saturating the unitarity bound $\D\geq 2$. Finally, a new background 

759: field $v$ has been introduced which couples to this conserved current. This new background field corresponds to the 

760: bulk electric field via $f=*\dd v$. 

761: 

762: \subsection{Holographic renormalization}\label{backreaction}

763: 

764: We now summarize the analysis of the coupled gravity-Maxwell system and the holographic renormalization of the action, which is done in Appendix \ref{holren}.

765: 

766: The total Euclidean action is written in \eq{Euclaction}. In the first order 

767: formalism, we simply replace the matter part of the action by the corresponding first order matter

768: action \eq{1storderform}:

769: \be

770: S=-{1\over16\pi G_N}\int\dd^4x\sqrt{g}\,\left(R-2\L\right)-{1\over8\pi G_N}\int\dd^3x\sqrt{\g}\,K+S_{\sm{bulk}}[A,E]+S_{\sm{bdy}}[A,E]~.

771: \ee

772: The components of the stress-energy tensor are

773: \bea\label{stressenergy}

774: T_{\m\n}&=&{r^2\over\ell^2}\,\t_{\m\n}~;\nn

775: \t_{rr}&=&\half(E^2-F^2)~,\nn

776: \t_{ri}&=&E^jF_{ij}=\e_{ijk}E_jF_k~,\nn

777: \t_{ij}&=&E_iE_j-F_iF_j-\half\,g_{ij}(E^2-F^2)~.

778: \eea

779: The stress-energy tensor is manifestly invariant under  electric-magnetic transformations (which are hyperbolic in Euclidean signature), as it should. In this formulation, it is obvious that the stress-energy tensor is zero if and only if

780: \be

781: E_i=\ve\, F_i~,

782: \ee

783: where $\ve^2=1$, i.e. the solution is (anti-)self-dual.

784: 

785: In Appendix \ref{holren}, Einstein's equations are solved in the coordinate system \eq{FG}:

786: \bea\label{grx}

787: g(r,x)&=&g_{(0)}+r^3g_{(3)}+r^4g_{(4)}+\cdots\nn

788: \nabla^jg_{(3)ij}&=&-{16\pi G_N\over3\ell^2}\,f^jF_{ij}\nn

789: g_{(4)ij}&=&-{4\pi G_N\over\ell^2}\left(f_if_j-F_iF_j-{1\over4}\,g_{ij}(f^2-F^2)\right)~,

790: \eea

791: where we have chosen a Ricci-flat boundary metric, hence $g_{(2)}=0$. $g_{(3)}$ is otherwise undetermined by the 

792: field equations and has the interpretation as the stress-energy tensor of the boundary theory. This is the usual 

793: Fefferman-Graham ambiguity for the metric and it is fixed once one imposes regularity of the bulk solution. As 

794: explained in the Appendix, the middle equation above is a Ward identity that follows from the diffeomorphism 

795: invariance of the on-shell action. In fact, it is of the form 

796: \be

797: \nabla^j\bra T_{ij}\ket=T_{ir}=-F_{(0)ij}f^j~.

798: \ee

799: $\bra T_{ij}\ket$ is also the quasi-local stress-energy tensor of Brown and York \cite{BY}. The above identity measures the matter flow through the boundary. From \eq{stressenergy} we see that it vanishes if $f_i$ is proportional to $F_i$, which are indeed AdS-preserving boundary conditions. Using the fact that the boundary current is

800: \be

801: \bra J_i\ket=-f_i~,

802: \ee

803: the above is the expected field theory Ward identity \cite{dHSS,BFS}.

804: 

805: In Appendix \ref{holren} we show that, after subtracting the divergent part of the effective action by adding the 

806: usual counterterm, there is no near-boundary contribution to the on-shell action coming from the gravity part of the 

807: action. 

808: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%5

809: %This follows from the fact that there is no trace anomaly in three dimensions and amounts to neglecting 

810: %back-reaction, which sets in only at order $r^4$ in the metric. 

811: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

812: Taking into account back-reaction would result in 

813: triple and higher order terms in the boundary operators, which we may neglect since we are only interested in

814: two-point functions. Therefore, in this approximation the only contribution is from the matter part of the action, 

815: which we analyze in the next subsection.

816: 

817: \subsection{Regularity and incoming boundary conditions}\label{regularitysec}

818: 

819: 

820: We now impose the usual normalizability conditions in AdS \cite{isham,bf,vijayper}. The situation for gauge fields is similar to the case of conformally coupled scalar fields. Let us denote the boundary momentum by $p=(\o,k)$. In Euclidean signature, as well as for Lorentzian tachyonic modes $\o^2<k^2$, the general solution of the wave equation is

821: \be

822: A_i^{\tiny{\mbox{T}}}(r,p)=\half\left(A_i^{\tiny{\mbox{T}}}(p)+{1\over|p|}\,f_i(p)\right)e^{|p|r}+\half\left(A_i^{\tiny{\mbox{T}}}(p)-{1\over|p|}\,f_i(p)\right)e^{-|p|r}~.

823: \ee

824: The first solution blows up at $r=\infty$. It is clear that if we demand 

825: \be\label{regularity}

826: A_i^{\tiny{\mbox{T}}}(p)+{1\over|p|}\,f_i(p)=0~,

827: \ee

828: then regularity is ensured. One can check \cite{marolf} that the solutions are normalizable. In position space the above is a non-local relation:

829: \be

830: A_i^{\tiny{\mbox{T}}}(x)=-{i\over \Box^{1/2}}\,f_i(x)=-\int\dd^3y\,{f_i(y)\over|x-y|^4}~.

831: \ee

832: Notice that this is simply the linear boundary condition \eq{dtbc} where $m=-i\Box^{1/2}$ is an operator. 

833: For later convenience we re-write the gauge invariant regularity condition as 

834: \be\label{gaugeinvreg}

835: f=-\sqrt{-\Box}\Pi A~.

836: \ee 

837: 

838: For regular bulk solutions, the gauge field and field strength now take the following form:

839: \bea\label{regulargauge}

840: A_i(r,x)&=&A_i(x)\,e^{-|p|r}\nn

841: F_{ri}(r,p)&=&-|p|\Pi_{ij}A_j(p)\,e^{-|p|r}\nn

842: F_{ij}(r,p)&=&i(p_iA_j(p)-p_jA_i(p))\,e^{-|p|r}\nn

843: F_i(r,p)&=&i\e_{ijk}p_jA_k(p)\,e^{-|p|r}~.

844: \eea

845: 

846: Using the equations of motion, it is also easy to see that if we impose \eq{regularity} at $r=0$ then we also have

847: \be\label{regularityr}

848: A_i^{\sm{T}}(r,p)+{1\over|p|}\,E_i(r,p)=0

849: \ee

850: for all $r$. This will ensure that the contribution from $r=\infty$ to the on-shell action vanishes. The transverse 

851: part of the gauge field is defined in \eq{transverse}, hence the above equation is gauge invariant with respect to 

852: boundary gauge transformations of the dual CFT (see end of section 2.2).

853: 

854: In Lorentzian signature, all the modes satisfying $\o^2>k^2$ are normalizable and so there is no regularity condition for these modes. There is, however, a natural choice of boundary condition \cite{herzogson} which corresponds to imposing that all waves are ingoing at the boundary -- that is, all waves travel from the boundary into the bulk. Notice that this is true both at $r=0$ and $r=\infty$; since we are dealing with pure AdS, waves that are ingoing at the boundary remain ingoing near $r=\infty$. From the field theory point of view, this choice is natural because it corresponds to defining the vacuum in the Lorentzian field theory as the analytically continued Euclidean vacuum; with this choice of boundary conditions, a Lorentzian transition function in the vacuum corresponds to the Euclidean partition function. This is the choice that leads to {\it absorption} cross section by the M2-brane. Let us explain this point. When solving the wave equation in the original M2-brane geometry \cite{maldacena}, 

855: \bea\label{M2}

856: \dd s^2&=&f^{-2/3}\,(-\dd t^2+\dd\vec x^2)+f^{1/3}(\dd R^2+R^2\dd\Omega_7^2)\nn

857: f&=&1+{2^5\pi^2N\ell_{\sm{Pl}}^6\over R^6}~,

858: \eea

859: with a non-trivial four-form flux, 

860: a choice of {\it incoming} waves must be made at the horizon -- in other words, only the modes describing infalling matter are kept \cite{unruh,GK} (see \cite{GKT} for the higher-spin extension). Taking the near-horizon limit of \eq{M2} we get

861: \be

862: \dd s^2={\ell^2\over r^2}(\dd r^2-\dd t^2+\dd\vec x^2)+4\ell^2\dd\Omega_7^2

863: \ee

864: with $2\ell=\ell_{\sm{Pl}}/(2^5\pi^2N)^{1/6}$ where $R\sim 1/r$. Hence, the modes absorbed by the M2-brane will be modes travelling from the boundary towards the bulk. Such modes have the form $e^{-i\o t+i|p|r}$: indeed, for such modes, as time elapses the wave front increases $r$ and so they propagate to the bulk. Since we are doing classical field theory near the M2-brane, it is natural to take the waves to be ingoing at the horizon. Using the change of coordinates in \cite{herzogson}, it is easy to check that this remains true in the non-extremal case. We will now see this for the gauge field.

865: 

866: We write the general form of the solution of Maxwell's equations \eq{Aiexpansion} as

867: \bea\label{Lor}

868: A_i^{\tiny{\mbox{T}}}(r,p)&=&A_i^{\tiny{\mbox{T}}}(p)\,\cos(|p|r)+{1\over|p|}\,A_{i(1)}(p)\,\sin(|p|r)\nn

869: &=&\half\left(A_i^{\tiny{\mbox{T}}}(p)+{1\over i|p|}\,A_{i(1)}(p)\right)e^{i|p|r}+ \half\left(A_i^{\tiny{\mbox{T}}}(p)-{1\over i|p|}\,A_{i(1)}(p)\right)e^{-i|p|r}

870: \eea

871: where now $|p|=\sqrt{\o^2-k^2}$. Analytic continuation from the Euclidean is done by $|p_{\sm{E}}|\rightarrow -i|p_{\sm{L}}|$, therefore the regularity condition \eq{regularity} turns into

872: \be\label{Lreg}

873: A_{(0)}^{\sm{T}}(p)+{1\over i|p|}\,A_{(1)}(p)=0

874: \ee

875: whenever $\o>0$. This leaves only the first factor in \eq{Lor}. Taking into account the fact that $A_{(0)}$ contains a factor of $e^{-i\o t}$, we get a wave $e^{-i\o t+i|p|r}$. This agrees with the behavior of the natural boundary condition in \cite{herzogson}. 

876: 

877: One can check that, by construction, the bulk fields are purely negative frequency in the far past and purely positive frequency in the far future, as one would expect from the analytic continuation of the Euclidean 2-point function which is the Feynmann Green's function.

878: 

879: \subsubsection*{In-going vacuum and vanishing action}

880: 

881: Incoming boundary conditions of the type just found in the Lorentzian describe a special vacuum that we will now analyze in more detail. As we will see, they have exactly zero action. Expand the bulk solutions again in plane waves:

882: \bea\label{Arx}

883: A(r,x)&=&\int\dd^3p\,a(p)\,e^{i|p|r+ip\cdot x}+\mbox{c.c.}\nn

884: &=&\int_0^\infty\dd\o\int\dd^2k\left(a(p)\,e^{i(|p|r-\o t)}+a^*(-p)e^{-i(|p|r+\o t)}\right)e^{ikx}+\mbox{c.c.}

885: \eea

886: where $k$ is a 2-vector that labels spatial momenta along the boundary.

887: As before, the in-going vacuum requires vanishing of the second factor, therefore

888: \be\label{ing}

889: a(p)=a(\o,k)=0~~~\mbox{if}~\o<0~.

890: \ee

891: Notice that only the sign of $\o$ matters here, with respect to which positive frequencies are defined. We now look at the boundary expansion of \eq{Arx}:

892: \be

893: A(r,x)=\int_0^\infty\dd\o\int\dd^2k\left[A_{(0)}(p)\cos(|p|r)+iB_{(1)}(p)\sin(|p|r)\right]e^{-i\o t+ikx}+\mbox{c.c.}

894: \ee

895: where $A_{(0)}(p)=a(p)+a^*(-p)$, $B_{(1)}(p)=a(p)-a^*(-p)$, and $A_{(0)}(p)$ defines the natural field theory vacuum 

896: $A_{(0)}(p)|0\ket=0$, since at $r=0$ $B_{(1)}$ disappears. The in-going vacuum \eq{ing} gives

897: $A_{(0)}(p)=\mbox{sgn}(\o)\,B_{(1)}(p)$. Defining the vacuum in terms of $A_{(0)}$ makes sense in the dual CFT, where 

898: $A_{(0)}$ is interpreted as an operator.

899: 

900: A similar analysis for the electric field gives:

901: \be

902: E(r,x)=\int_0^\infty\dd\o\int\dd^2k\left[G_{(1)}(p)\sin(|p|r)+f(p)\cos(|p|r)\right]e^{-i\o t+ikx}+\mbox{c.c.}

903: \ee

904: where $f(p)=-i|p|B_{(1)}(p)$ and $G_{(1)}(p)=|p|A_{(0)}(p)$.

905: The in-going boundary condition now gives

906: \be\label{nreg}

907: f(p)=-i|p|\,\mbox{sgn}(\o)\,A_{(0)}(p)

908: \ee

909: which is the generalization of \eq{Lreg} to arbitrary frequencies.

910: 

911: It is now straightforward to check that for such states the bulk action is zero. As we will 

912: prove in the next section, the on-shell action is proportional to $\int\dd^3x\,fA$.

913: Expanding this in Fourier modes, and using \eq{nreg}, this is readily seen to be identically zero. On the other hand, the Hamiltonian is non-zero.

914: 

915: \subsection{Self-duality and topologically massive theories}\label{sdtmt}

916: 

917: We will now analyze self-dual solutions of the Euclidean field equations. The Euclidean action is:

918: \be

919: S={1\over4g^2}\int\dd^4x\sqrt{g}\,F_{\m\n}F^{\m\n}+{\th\over32\pi^2}\int\dd^4x\,\sqrt{g}\,\e^{\m\n\l\s}F_{\m\n} F_{\l\s}~.

920: \ee

921: We will prove three main results concerning self-dual solutions. First, we will show that there is a one-to-one correspondence between self-dual solutions in the bulk and a choice of boundary conditions that we will call ``self-dual boundary conditions''. Second, we will show that bulk self-dual solutions together with a boundary condition that breaks conformal invariance leads to the self-dual topological theory of \cite{PTvN}. Finally, we will analyze regularity of the solutions and show that massive solutions are regular for negative value of the deformation parameter. 

922: 

923: \subsubsection*{Self-duality and topologically massive theories}

924: 

925: We start with (anti-)self-dual solutions in the bulk. These satisfy:

926: \be\label{sd}

927: F_{\m\n}=-{\ve\over2}\,\e_{\m\n\a\b}F^{\a\b}~,

928: \ee

929: This equation is conformally invariant, therefore in the Poincare patch this is an equation on ${\mathbb R}_+^4$. $\ve$ can take the values $\pm1,0$. We get:

930: \be\label{selfduality}

931: F_{ri}=-{\ve\over2}\,\e_{ijk}F_{jk}~~~,~~~F_{ij}=-\ve\,\e_{ijk}F_{rk}~.

932: \ee

933: The condition \eq{sd} is a first order equation. Together with the Bianchi it implies the field equation. Therefore it determines one of the two boundary conditions. From \eq{selfduality}:

934: \be\label{sdbc}

935: f_i=-\ve\,\e_{ijk}\pa_jA_k~.

936: \ee

937: This condition is what we will call the ``self-dual boundary condition''.

938: 

939: Next we will show that, given the self-dual boundary condition \eq{sdbc}, any solution of the bulk equations is self-dual 

940: and hence satisfies \eq{selfduality} throughout the whole of AdS. This is easy to see. We use the solution of the equations

941: of motion \eq{Aiexpansion} to work out the field strength,

942: \bea

943: F_{ri}(r,p)&=&|p|\,A_i^{\tiny{\mbox{T}}}(p)\,\sinh(|p|r)+f_i(p)\,\cosh(|p|r)\nn

944: F_{ij}(r,p)&=&i(p_iA_j-p_iA_j)\,\cosh(|p|r)+{i\over|p|}\,(p_if_j-p_jf_i)\,\sinh(|p|r)~.

945: \eea

946: Filling the boundary condition \eq{sdbc}, it is easy to see that the self-duality condition \eq{selfduality} is satisfied

947: 

948: %Filling this into \eq{selfduality}, we find that the self-duality condition is satisfied if the following two equations hold:

949: %\bea

950: %f_i&=&-i\ve\,\e_{ijk}p_jA_k\nn

951: %A_i^{\tiny{\mbox{T}}}&=&-{i\ve\over|p|^2}\,\e_{ijk}p_jf_k~.

952: %\eea

953: %But these are nothing than the self-dual boundary condition \eq{sdbc} provided $\ve^2=1$. Hence, given the self-dual boundary condition, self-duality is satisfied everywhere throughout the bulk. Notice that in doing this we have not used regularity of the solution: self-dual bulk solutions are in one-to-one correspondence with self-dual boundary conditions.

954: 

955: This also implies that the self-dual boundary condition \eq{sdbc} is enough to ensure everywhere vanishing of the 

956: stress-energy tensor (see section \ref{backreaction}).

957: 

958: Now we will consider massive deformations of the boundary theory. This is analogous to the scalar field case, where a 

959: choice of massive boundary condition corresponds to adding relevant terms to the boundary theory. In this case, the relevant massive

960: boundary condition is:

961: \be\label{dtbc}

962: A_i^{\tiny{\mbox{T}}}={1\over m}\,f_i

963: \ee

964: where $m$ is a dimensionful parameter. We will call this the ``massive boundary condition''.

965: 

966: Combining \eq{dtbc} with \eq{sdbc}, we see that the boundary value of the gauge field satisfies the following equation:

967: \be\label{SD}

968: A_i^{\tiny{\mbox{T}}}=-{\ve\over m}\,\e_{ijk}\pa_jA_k^{\tiny{\mbox{T}}}.

969: \ee

970: Surprisingly, this is the self-dual abelian theory in three dimensions of \cite{PTvN}! Deser and Jackiw showed \cite{DJ} that this theory is equivalent to topologically massive QED in three dimensions. We will come back to its interpretation later.

971: 

972: \subsubsection*{Regularity and massive solutions}

973: 

974: Finally, we analyze regularity of the different solutions discussed here. Regularity relates the two boundary conditions of the solution in a non-local way, in this case $A_i(x)$ and $f_i(x)$, as we showed explicitly in \eq{regularity}. 

975: %Therefore, one would think that this leaves room for only one other choice of boundary condition. However, this is not necessarily the case, as a given set of boundary conditions might automatically lead to regular solutions. In other words, one needs to check whether the choice of boundary conditions leads to a regular solution or not.

976: We will first construct a regular self-dual solution, that is we combine equation \eq{regularity} with \eq{sdbc}. We get the following non-local equation for the boundary gauge field:

977: \be\label{rmas}

978: A_i^{\tiny{\mbox{T}}}=-{i\ve\over\Box^{1/2}}\,\e_{ijk}\pa_jA_k^{\tiny{\mbox{T}}}~,

979: \ee

980: again a gauge invariant equation. This gives the most general regular self-dual solution in the bulk. We solve this equation in Appendix \ref{solutionSD}. In fact, this is the most general solution with vanishing stress-energy tensor.

981: 

982: Next we construct regular massive solutions. Thus we combine \eq{dtbc} and \eq{regularity}. This gives the following condition in momentum space:

983: \be\label{sign(m)}

984: |p|=-m~.

985: \ee

986: Notice that in this case regularity does not act as a boundary condition on the boundary values of the fields. Instead, it puts the momentum of the boundary theory on the mass shell. Obviously, this is only possible if $m<0$. This means that, for regular solutions, the deformation parameter of the boundary theory has negative sign. Thus, in this case we can still impose a second boundary condition. It is indeed possible to combine the self-dual boundary condition \eq{sdbc} with the massive one \eq{dtbc}, subject to the regularity condition \eq{sign(m)}. 

987: 

988: This fact is actually generic: the deformation parameter in the linear boundary condition, which corresponds to the deformation parameter of the double trace deformation, is required to be negative in order for two-point functions of dual operators to be positive definite. This is reminiscent of the regularity analysis in \cite{dHPP} for triple-trace deformations and the constraints on the deformation parameters found there. 

989: 

990: \subsection{Abelian instantons in AdS$_4$}\label{iads4}

991: 

992: We have obtained regular self-dual solutions of the Euclidean equations of motion. We can now ask whether these solutions actually have finite action and correspond to instantons. In flat space there are of course no abelian instantons because one cannot form a topological number. However, things may change in AdS due to the boundary terms.

993: 

994: The on-shell value of the action is computed in the next section:

995: \be

996: S_{\sm{on-shell}}=-\int\dd^3p\,A(p)f(-p)~,

997: \ee

998: where we have also included the contribution of explicit boundary terms (worked out in \eq{fF}).

999: We will now ask whether there are solutions such that this number is finite. We consider the regular instanton solutions of section \ref{sdtmt}. The explicit solution of \eq{rmas} is given in \eq{Asolution} of the Appendix. First of all we remark that the action simply reduces to the Chern-Simons term $\int A\wedge\dd A$, which for these solutions can be written as:

1000: \be

1001: S_{\sm{on-shell}}={1\over g^2}\int\dd^3p\,|p|A_i^{\tiny{\mbox{T}}}(p)A_i^{\tiny{\mbox{T}}}(-p)~.

1002: \ee

1003: We should remember the constraints \eq{B3}. In the notation in the Appendix,

1004: \be

1005: A_i^{\tiny{\mbox{T}}}(p)A_i^{\tiny{\mbox{T}}}(-p)=-\left(1+{4p_1^2p_2^2\over(p_1^2+p_2^2)^2}\right)f(p)f(-p)~.

1006: \ee

1007: Redefine now

1008: \be

1009: F(p)=\sqrt{1+{4p_1^2p_2^2\over(p_1^2+p_2^2)^2}}\,f(p)~.

1010: \ee

1011: The action becomes

1012: \be

1013: S_{\sm{on-shell}}={2\sqrt{2}\over g^2}\int\dd p_1\int\dd p_2\,\sqrt{p_1^2+p_2^2}\,|F(p)|^2

1014: \ee

1015: and the integral is over the remaining ${\mathbb{R}}^2$. The following choice will obviously make the action finite:

1016: \be\label{Fp}

1017: F(p)={2^{1/4}\sqrt{\pi a\l}\over a^2+p_1^2+p_2^2}~.

1018: \ee

1019: $\l$ is an arbitrary dimensionless constant. $a$ is an arbitrary scale and as usual for instantons the value of the action is independent of it. We get

1020: \be

1021: S_{\sm{on-shell}}={8\pi^2\over g^2}\,|\l|~.

1022: \ee

1023: The value of $|\l|$ can be fixed by fixing, for example, the holonomy of the gauge field. Since $\l$ must be independent of the coupling, it is remarkable that the action is in fact proportional to $1/g^2$. 

1024: 

1025: \section{$S$-duality in AdS$_4$}

1026: 

1027: We will now analyze the action of $S$-duality on the bulk action. Since electric-magnetic rotations are not canonical transformations, they act non-trivially on the canonically conjugated variables $A$ and $E$. It is well-known that, whereas electric-magnetic transformations are a manifest symmetry of the equations of motion, they are not a manifest symmetry of the action. They have to be realized at the level of the gauge potential, and this is non-trivial. The solution of this problem is well known (see for example \cite{DT,dght,ss,Deser:2004xt}). Here we use the first order formulation in \cite{deser,Deser82}.

1028: 

1029: In AdS, electric-magnetic invariance is broken due to the presence of the boundary. Therefore, an electric-magnetic transformation changes the boundary terms in the action. We will first explicitly show the $S$-duality of the bulk part of the action and compute the boundary terms. We will then generalize this to the full $SL(2,{\mathbb{R}})$ ($SL(1,1)$ in the Euclidean case). Witten's proof of $S$-duality for abelian theories 

1030: involves introducing an additional two-form field in the action, together with an enlarged 

1031: gauge symmetry \cite{wittenabelian}. $S$-duality is then showed after integrating out two different fields. We 

1032: will instead give two elementary proofs that rely only on the first order formalism of the 

1033: action. The advantage of our first proof is that it does not involve integrating out fields 

1034: but only a simple field redefinition. The price we pay is that Lorentz invariance is not 

1035: manifest. In this proof, we write the action in terms of first order fields $\sf E$ and $A$, 

1036: where $\sf E$ is the redefined conjugate momentum. In terms of these variables, the action 

1037: is manifestly invariant under $S$-duality including the $\th$-angle. The second proof, given in Appendix \ref{2nds},

1038: involves integrating out either $E$ or $A$, to get two actions with the same form which are 

1039: dual to each other. It presents some similarities to the methods in \cite{wittenabelian,yolanda}.

1040: 

1041: Then, using holographic renormalization, we will show how $S$-duality acts on the boundary effective potential, and on the boundary 1- and 2-point functions, and how the Legendre transform arises.

1042: 

1043: $T$-transformations simply act by $\th\rightarrow\th+2\pi$ in \eq{actionwtheta}. These do not leave the action invariant but they do leave the partition function 

1044: invariant.

1045: 

1046: \subsection{The first order formalism}\label{1of}

1047: 

1048: We first work out the first order form of the action \eq{actionwtheta} in Lorentzian signature\footnote{For AdS/CFT in the Hamiltonian formalism for gravity and scalars and the role of the canonical momenta in correlation functions, see \cite{YK}.}. We define canonical 

1049: momenta:

1050: \be

1051: \Pi^i={1\over\sqrt{-G}}{\d S\over\pa_rA_i}=-{1\over g^2}\,g^{rr}g^{ij}(\pa_rA_i-\pa_iA_r)~.

1052: \ee

1053: It is easy to see that, in the Fefferman-Graham coordinate system \ref{FG}, one gets the 

1054: same Hamiltonian and action if one defines all quantities with respect to the conformally

1055: rescaled metric $\ti G_{\m\n}={r^2\over\ell^2}\,G_{\m\n}$:

1056: \be

1057: E^i={1\over\sqrt{-g}}{\d S\over\d\pa_rA_i}=-{1\over g^2}\,g^{ij}(\pa_rA_j-\pa_jA_r)~.

1058: \ee

1059: This amounts to removing the factors of $g^{rr}$ from the formulas and raising and lowering

1060: indices with the boundary metric $g_{ij}(r,x)$. This is the procedure we will adopt in what

1061: follows, which of course is possible due to conformal invariance of the matter part 

1062: of the action. From the analysis in Appendix \ref{holren}, equation \eq{FGexp}, it follows that

1063: for a Ricci-flat boundary metric the first correction to the boundary metric appears at

1064: order $r^3$ (this is the case, for instance, of the AdS black hole), therefore most boundary 

1065: quantities we will be interested in are be unaffected by that.

1066: 

1067: The $\th$-term is a boundary term and does not contribute to the canonical momentum. The 

1068: Hamiltonian now takes the following form:

1069: \be

1070: H=\int\dd^4x\left(-{g^2\over2}\,E^2-{1\over2g^2}\,F^2+E^i\pa_iA_r\right)-{\th\over8\pi^2}\int\dd^3x\,AF~.

1071: \ee

1072: Notice the unusual sign, which is due to the fact that we are doing radial quantization. The equations of motion give:

1073: \bea

1074: E_i&=&-{1\over g^2}\,(\pa_rA_i-\pa_iA_r)\nn

1075: \pa_rE&=&{1\over g^2}\,*\dd F\nn

1076: \na^iE_i&=&0~.

1077: \eea

1078: The action, written in first order form, now gives

1079: \bea\label{1storderform}

1080: S&=&\int\dd^4x\left(E^i(\pa_rA_i-\pa_iA_r)+{g^2\over2}\,E^2+{1\over2g^2}\,F^2-{\th\over4\pi^2}\,(\pa_rA)F\right)\nn

1081: &=&\int\dd^4x\left(E^i(\pa_rA_i-\pa_iA_r)+{g^2\over2}\,E^2+{1\over2g^2}\,F^2\right)+{\th\over8\pi^2}\int\dd^3x\,AF~.

1082: \eea

1083: The bulk part of the action has the obvious symmetry $E'=F$, $F'=-E$, $g'=1/g$, $\th'=-g^4\th$. 

1084: In the Euclidean signature action, the quadratic term in $E_i$ has a $-$ sign instead.

1085: 

1086: $A_r$ is a Lagrange multiplier. Its equation of motion gives the Gauss law

1087: \be

1088: \na^iE_i=0~.

1089: \ee

1090: We can solve the latter by

1091: \be\label{top}

1092: E^i(r,x)=\e^{ijk}\pa_jv_k(r,x)~

1093: \ee

1094: where $\e^{ijk}$ is the epsilon-tensor in the boundary metric $g_{ij}(r,x)$.

1095: Notice that $E_i$ and $v_i$ are $r$-dependent. We will denote

1096: \bea

1097: v_i&=&v_i(x)=v_i(0,x)\nn

1098: f_i&=&f_i(x)=E_i(0,x)~.

1099: \eea

1100: One can fix a gauge $A_r=0$, although this is not necessary.

1101: As we mentioned earlier, there is a residual gauge invariance that leaves this gauge fixed, namely 

1102: $A_\m\rightarrow A_\m+\pa_\m \vf$ where $\vf=\vf(x)$ has no $r$-dependence. These residual gauge transformations will act 

1103: on the boundary value of the gauge field $A_i(x)$ and account for the gauge invariance of the dual CFT.

1104: 

1105: 

1106: 

1107: We fill in the equations of motion and get for the on-shell action:

1108: \be\label{onshellwth}

1109: S_{\sm{on-shell}}=\int\dd^3x\left(-\half\,EA+{\th\over8\pi^2}\,AF\right)~.

1110: \ee

1111: 

1112: \subsection{Proof of $S$-duality}

1113: 

1114: We will give a proof of $S$-duality which, to our knowledge, is new. It relies on recasting the first order 

1115: form of the action in a way that is manifestly $S$-duality invariant. We first remark that, since $E$ is a Lagrange multiplier, we can make the following redefinition:

1116: \be\label{eredef}

1117: {\sf E}=E-{\th\over4\pi^2}\,F~.

1118: \ee

1119: Since $F$ is conserved, this is compatible with a conservation constraint on $\sf E$. In fact, written as a function of $\sf E$ the action 

1120: takes the form:

1121: \be\label{1orderwcoupling}

1122: S=\int\dd^4x\left({\sf E}^i(\pa_r A_i-\pa_iA_r)+\half\,g^2{\sf E}^2+{1\over2g^2}\left(1+{g^4\th^2\over(4\pi^2)^2}\right)\,F^2+{\th g^2\over4\pi^2}\,{\sf E}F\right)~.

1123: \ee

1124: With this rewriting, we have eliminated the boundary term. Indeed, the $\th$-angle in this form is not a boundary term.

1125: 

1126: As is easy to check, this form of the action has manifest $S$-duality invariance:

1127: \bea\label{Sduality}

1128: {\sf E}'&=&F\nn

1129: F'&=&-{\sf E}\nn

1130: \tau'&=&-{1\over\tau}~,

1131: \eea

1132: where

1133: \bea\label{Scouplings}

1134: \t&=&{\th\over4\pi^2}+{i\over g^2}=\tau_1+i\t_2\nn

1135: {\th'\over4\pi^2}&=&-{{\th\over4\pi^2}\over {\th^2\over(4\pi^2)^2}+{1\over g^4}}=\t_1'=-{\t_2\over\t_1^2+\t_2^2}\nn

1136: (g')^2&=&{1\over g^2}+{g^2\th^2\over(4\pi^2)^2}~~,~~\t_2'={\t_2\over\t_1^2+\t_2^2}~.

1137: \eea

1138: More precisely, we get

1139: \be\label{trafowtheta}

1140: S[A',{\sf E}']=S[A,{\sf E}]+\int\dd^3x\,{\sf E}A~.

1141: \ee

1142: In other words, in terms of $\sf E$ the boundary term that we get from the variation is the {\it usual} one that we get in the absence of 

1143: the $\th$-angle.

1144: 

1145: 

1146: 

1147: 

1148: One can see $S$-duality directly in the original first form order of the action \eq{1storderform}. We rewrite the $S$-duality transformations \eq{Sduality} in terms of the original variables 

1149: \bea\label{dualEF}

1150: E'&=&\left(1+{\th\th'\over(4\pi^2)^2}\right) F-{\th'\over4\pi^2}\,E\nn

1151: F'&=&-E +{\th\over4\pi^2}\,F

1152: \eea

1153: where $\th'$ is the transformed coupling as in \eq{Scouplings}. Of course, this form of $S$-duality still satisfies $S^2=-1$, as it should. It is now easy to check that this leaves the action unchanged, up to boundary terms:

1154: \be\label{SAE}

1155: S[A',E']=S[A,E]+\int\dd^3x\left(E-{\th\over4\pi^2}\, F\right)A.

1156: \ee

1157: Of course, the extra terms are just the ones in \eq{trafowtheta}, as it should. This gives a direct proof of $S$-duality that doesn't involve any field redefinitions or integrating out fields.

1158: 

1159: It is interesting to note that the above can be simply rewritten as:

1160: \be\label{SSA}

1161: S[A',E']=S[A,E]-\int A'\wedge\dd A~.

1162: \ee

1163: %If we integrate out $E$ and $E'$ now, so that

1164: %\be

1165: %S[A']=S[A]-\int\dd^3x\,F'A~.

1166: %\ee

1167: This is in agreement with the definition of $S$-duality in \cite{Wittensl2z} from the CFT point of view. In fact, the above is like a Legendre transform between $A$ and a field of the same dimension $A'$.

1168: Using the fact that $S$ squares to $-1$, we get

1169: \be

1170: S[A'']=S[A]~,

1171: \ee

1172: so we see that, whereas $S$ is not an invariance of the boundary theory, $S^2$ is. The above agrees with Witten's definition of $S$-duality in CFT \cite{Wittensl2z}.

1173: 

1174: 

1175: 

1176: \subsection{$SL(2,{\mathbb{R}})$ duality}\label{sl2rd}

1177: 

1178: We can generalize $SL(2,{\mathbb{Z}})$ invariance of the action to the full  $SL(2,{\mathbb{R}})$ that is relevant classically and plays a role in M-theory compactifications. Thus, we will generalize $S$-duality to include a full $SO(2)$ transformation. Let us define $g^*=g^*(g,\th)$ and $\th^*=\th^*(\th,g)$ as the couplings that we get after the $SL(2,\mathbb{R})$ transformation. We will take $E$ and $F$ to transform as

1179: \bea

1180: E^*&=&a\,E+b\,F\nn

1181: F^*&=&c\,E+d\,F~.

1182: \eea

1183: Requiring invariance of the potential terms in \eq{1storderform} gives

1184: \bea\label{SL2R}

1185: g^*E^*&=&\eta\,g\,E+\d\,\sqrt{1-\eta^2}\,F/g\nn

1186: F^*/g^*&=&-\d\,\sqrt{1-\eta^2}\,g\,E+\eta\,F/g

1187: \eea

1188: where $\d^2=1$ is an arbitrary sign which simply corresponds to the choice of sign in the square root. There is another sign that we have fixed here. The sign in front of $F$ on the r.h.s. of the above two equations is not fixed by invariance of the potential terms. It is fixed by invariance of the kinetic term (up to boundary terms), and the above choice is the one that renders the full bulk action invariant. $\eta$ is arbitrary and parametrizes the rotation part of the transformation. The transformed action is

1189: \bea\label{actiontransform}

1190: S[A^*,E^*]&=&S[A,E]+\int\dd^3x\left[\left({\d\over2}g^2\eta\sqrt{1-\eta^2}+g^2g^*{}^2{\th^*\over8\pi^2}(1-\eta^2)\right)fv\right.\nn

1191: &+&\left.\left((1-\eta^2)-g^*{}^2{\th^*\over4\pi^2}\d\eta\sqrt{1-\eta^2}\right)fA+\left(-{\d\eta\over2g^2}\sqrt{1-\eta^2}+{g^*{}^2\over g^2}{\th^*\over8\pi^2}\eta^2-{\th\over8\pi^2}\right)FA\right]~.\nn

1192: \eea

1193: 

1194: 

1195: It is easy to check that the general transformation \eq{SL2R} automatically preserves the stress-energy tensor, as it should. 

1196: It is also easy to see that the transformation \eq{SL2R} has unit determinant. 

1197: In fact, it is most convenient to parametrize the above $SO(2)$ transformation as

1198: \bea

1199: \left(

1200: \begin{array}{c}

1201: g^*\,E^*  \\

1202: F^*/g^* \end{array}

1203: \right)&=&\left(

1204: \begin{array}{cc}

1205: \cos\a & \,\,\,\sin\a \\

1206: -\sin\a & \,\,\,\cos\a

1207: \end{array}

1208: \right)

1209: \left(

1210: \begin{array}{c}

1211: g\,E \\

1212: F/g \end{array}

1213: \right)~,

1214: \eea

1215: where without loss of generality we absorbed $\d$ in the sign of $\a$ and $\eta=\cos\a$. This transformation has an 

1216: analytical continuation to Euclidean signature:

1217: \bea\label{SL2RE}

1218: \left(

1219: \begin{array}{c}

1220: g^*\,E^*  \\

1221: F^*/g^* \end{array}

1222: \right)&=&\left(

1223: \begin{array}{cc}

1224: \cosh\s & \,\,\,\sinh\s \\

1225: \sinh\s & \,\,\,\cosh\s

1226: \end{array}

1227: \right)

1228: \left(

1229: \begin{array}{c}

1230: g\,E \\

1231: F/g \end{array}

1232: \right)~,

1233: \eea

1234: which is a symmetry of the Euclidean bulk action up to boundary terms. The Euclidean transformation again preserves the stress tensor and $\dd E\wedge \dd F$. 

1235: 

1236: We also notice the regularity condition is invariant under the duality transformations.  

1237: This can be seen as \eq{gaugeinvreg} implies

1238: \be\label{regv}

1239: F=-t\sqrt{-\Box}\Pi\,v~.

1240: \ee

1241: Under the transformation 

1242: \be\label{regflow}

1243: E^*=cE+sF=-c\sqrt{-\Box}\Pi\,A-t\,s{\sqrt{-\Box}}\Pi\,v=-\sqrt{-\Box}\Pi\,A^*

1244: \ee

1245: where implicit factors of $g,g^*$ are understood.

1246: 

1247: 

1248: %While it's easily seen the general $SL(2,\mathbb{R})$ transformation $S^*$ is uni-modular, this of course does not satisfy 

1249: %in general $S^*{}^2=-1$. For example 

1250: If we take $g^*=g'$ and $\th^*=\th'$ as in \eq{Scouplings}, we get back the $S$-duality transformation \eq{dualEF} if we set in addition

1251: \bea\label{EFtrafo}

1252: \cos\a={{g^2\th\over(4\pi)^2}\over\sqrt{1+{g^4\th^2\over(4\pi^2)^2}}}~,\quad \sin\a={1\over \sqrt{1+{g^4\th^2\over(4\pi^2)^2}}}\ .

1253: \eea

1254: Fill this in \eq{actiontransform} we recover \eq{SAE} as we should.

1255: If we set instead $g^*=1/g$ and $\th'=-g^4\th$ then 

1256: \be\label{flowend}

1257: \sin\a=1

1258: \ee

1259: in order to obtain the S-dual transformation of \eq{1storderform}. Notice in both cases $g^*{}^2\th^*=-g^2\th$, which implements $S^2=-1$.

1260: 

1261: We note here that invariance of the bulk part of the action under the transformation \eq{SL2R} does not require the couplings to transform as \eq{Scouplings}. As we see, $g$ can actually be rescaled away, and $\th$ does not even enter \eq{SL2R}. To find out the transformation of $g$, the rest of the fields in the supermultiplet must be taken into account. In our case, as we will see in section \ref{ssusy} the relevant theory is the ${\cal N}=2$ supergravity in four dimensions, where $g\sim G_{\sm{N}}/\ell^2\sim 1/N^{3/2}$. Therefore, electric-magnetic duality will only be a symmetry of the action (up to boundary terms) if $g^*=g$. On the other hand, \eq{Scouplings} relates theories in asymptotically AdS$_4\times S^7$ of different size. In such a situation, the duality of \cite{BD1,BD2,Wittensl2z} which we prove here for any $g^*(g,\th),\th^*(g,\th)$ might actually be used to explore small AdS spaces.

1262: 

1263: %The identity matrix $S^*=\bf{1}$ is realized by taking $g^*=g$, $\th^*=\th$ and $\cos\a=1$.

1264: Since $g^*(g,\th)$ is in general an arbitrary function, we can in fact get the $T$-generator from \eq{SL2R} by choosing $g^*=g$, $\cos\a=1$ and $\th^*=\th+1$. Using \eq{actiontransform}, this indeed simply shifts the $\theta$-term in the action. In this sense, our transformations are not $SO(2)$ but we in fact generate the full $SL(2,{\mathbb{R}})$.

1265: 

1266: %{\it Show that the S, T transformations realized above indeed satisfy the expected algebraic relations.}

1267: 

1268: 

1269: 

1270: 

1271: 

1272: 

1273: \section{Generalized boundary conditions}\label{gbc}

1274: 

1275: Our goal is to study massive deformations of three-dimensional CFT's and how to describe them from the bulk. We will now develop the formalism to impose general boundary conditions for gauge fields and compute their on-shell action.

1276: 

1277: 

1278: \subsection{Variation and consistent boundary conditions}

1279: 

1280: In the first order formalism we vary the action with respect to $E$ and $A$. The usual variational principle imposes on us a choice of Dirichlet or Neumann boundary conditions:

1281: \be\label{bdyterm}

1282: \d_AS_{\sm{bulk}}[A,E]=-\int\dd^3x\, E^i\,\d A_i|_{r=0} +({\mbox{eom}})~.

1283: \ee

1284: so that we get for the two independent fields:

1285: \bea\label{DN}

1286: \d A_i|_{r=0}&=&0~~~~~~{\mbox{Dirichlet}}\nn

1287: E^i|_{r=0}&=&0~~~~~~{\mbox{Neumann}}~.

1288: \eea

1289: Since we are assuming that the boundary topology is trivial and we integrate out $A_r$, $E_i$ is expressed in terms of $v_i$ as in \eq{top} and there is no difference whether we vary with respect to $E$ or $v$.

1290: 

1291: The variation of the action \eq{bdyterm} also contains a contribution from the same boundary term evaluated at $r=\infty$. For regular solutions \eq{regulargauge}, using \eq{regularityr} this contribution is seen to be zero.

1292: 

1293: As explained in section \ref{scalardualities} for the scalar field case, in order to enforce general boundary conditions

1294: we can add a boundary term that will modify the stationarity conditions \eq{DN}:

1295: \be\label{newaction}

1296: S[E,A]=S_{\sm{bulk}}[E,A]+S_{\sm{bdy}}[E,A]~.

1297: \ee

1298: The Dirichlet condition $\d A_i=0$ is obviously always a possible boundary condition. We will consider the generalization of the Neumann boundary condition. Varying \eq{newaction} and demanding that it be stationary, we find

1299: \bea

1300: \d_ES_{\sm{bdy}}[E,A]&=&0\nn

1301: \d_AS_{\sm{bdy}}[E,A]&=&E^i(0,x)~.

1302: \eea

1303: In other words, in order for the action \eq{newaction} to be stationary under variations we need to require

1304: \bea\label{bc}

1305: {\d S_{\sm{bdy}}\over\d f_i}[f,A]&=&0\nn

1306: {\d S_{\sm{bdy}}\over\d A_i}[f,A]&=&f^i~.

1307: \eea

1308: These two equations define our set of boundary conditions.

1309: 

1310: We can now obtain the on-shell effective action by adding the boundary term to \eq{onshellwth}:

1311: \be\label{onshell}

1312: S_{\sm{on-shell}}[f,A]=-\half\int\dd^3x\,A_if^i+S_{\sm{bdy}}[f,A]

1313: %=-\half\int\dd^3x\,\e^{ijk}A_i\pa_jv_k+S_{\sm{bdy}}[f,A]

1314: ~,

1315: \ee

1316: and $f$ and $A$ satisfy \eq{bc}.

1317: As usual, we are assuming that there are no boundary contributions from $|x|=\infty$, while we keep boundary terms at $r=0$. As explained in section \ref{scalardualities} (see also section \ref{symf}), this can always be arranged by a choice of boundary terms or boundary conditions there \cite{don}. Unless the regularity condition at $r=\infty$ is imposed, the above only describes the contribution to the effective action coming from $r=0$. Notice that thanks to the classical conformal invariance of the bulk action, the on-shell action is completely finite at $r=0$ without the need to include any counterterms for the matter fields.

1318: 

1319: 

1320: \subsection{General boundary action}\label{gba}

1321: 

1322: %\subsubsection*{Two topological massive degrees of freedom}

1323: 

1324: %In general, for solutions that are non-extremal, the boundary value of the gauge field $A_i$ describes two independent 

1325: %massive degrees of freedom. We can illustrate this with the following example:

1326: %\be

1327: %S_{\sm{bdy}}={1\over m_1-m_2}\int\dd^3x\left(m_1\,A_if_i-f_iF_i\right)~.

1328: %\ee

1329: %The variation gives

1330: %\bea

1331: %A_i&=&{1\over m_1}\,\e_{ijk}\pa_jA_k\nn

1332: %f_i&=&{1\over m_2}\,\e_{ijk}\pa_jf_k~,

1333: %\eea

1334: %which are indeed two independent massive excitations with masses $m_1$ and $m_2$. Thus, the boundary gauge fields $A$ and $v$ are now completely decoupled and they describe independent degrees of freedom. Notice that this is possible in the Lorentzian since the regularity (or in-going) boundary condition is not necessary. Only the CFT's are not in their ground state -- they are excited.

1335: 

1336: %This example seems to correspond to two coupled boundary CFT's. They are not two different descriptions arising at a different regime of energies, but actually two different theories. It would be interesting to repeat the above computation including sources. Coupled CFT's have been considered in \cite{aharony}.

1337: 

1338: \subsubsection*{General boundary couplings}

1339: 

1340: We now construct the most general action with marginal operators and operators of dimension 4 that is gauge invariant, 

1341: local, and covariant in $v$ and $A$\footnote{A priori there is no reason to require that the action be local in $v$ and 

1342: $A$. In fact, the effective action will be non-local, as we discuss in section \ref{RGf}. The action 

1343: described here corresponds to the simplest choice of local boundary conditions for the gauge field.}:

1344: \be\label{generalaction}

1345: S_{\sm{bdy}}[v,A]=\int\dd^3x\left(\half\a\, AF+\half\b\, vf+\g\,Af+\half\d\, F^2+\half\e\,f^2+\m\,fF -A\ti{\cal J}-fJ\right)

1346: \ee

1347: (for notation, see section \ref{notation}), where we leave the six coefficients arbitrary. $\ti{\cal J}$ is necessarily conserved by gauge invariance. This means there is a 

1348: $\ti J$ such that $\ti{\cal J}=*\dd\ti J$. We will also define ${\cal J}=*\dd J$. 

1349: The boundary conditions \eq{bc} derived from the above action take the form:

1350: \bea\label{generaleom}

1351: \b\,\Pi v+\g\,\Pi A+\e\,f+\m\,F&=&J\nn

1352: \a\,F+(\g-1)\,f+\d\,*\dd F+\m\,*\dd f&=&\ti{\cal J}~,

1353: \eea

1354: both in Euclidean and in Lorentzian signature.

1355: 

1356: Taking the derivative of the first equation, we get 

1357: \bea

1358: \b\,f+\g\,F+\e\,*\dd f+\m\,*\dd F&=& {\cal J}\nn

1359: \a\,F+(\g-1)\,f+\d\,*\dd F+\m\,*\dd f&=&\ti{\cal J}~.

1360: \eea

1361: The equations have an obvious symmetry under simultaneous exchange $f\leftrightarrow F$, $\a\leftrightarrow\b$, 

1362: $\ve\leftrightarrow\d$ and $\gamma\leftrightarrow\g-1$. Notice that $\g$ plays a special role since it couples both 

1363: equations. The two limiting cases $\g=0$ and $\g=1$ are in fact each other's $S$-duals. 

1364: 

1365: The above system can be reduced to

1366: \bea\label{generaleom1}

1367: a_1\,f+a_2\,F+a_3\,*\dd f&=&0\nn

1368: b_1\,f+b_2\,F-a_3\,*\dd F&=&0~,

1369: \eea

1370: where the constants $a_i,b_i$ are given in \eq{a123} and we have set the sources to zero. It is possible to solve this system in general. One gets the following equation for $A$:

1371: \be\label{generalmassiveeq}

1372: (1+a\Box)\,A=b\,F~.

1373: \ee

1374: The values of $a,b$ are given in \eq{ab}. There is a similar equation for $f$, which for generic values of the constants $\a,\b,\g,\d,\e,\m$ is independent of the above. Again, for non-zero $b$ this is a (higher-derivative) massive equation for $A_i$. In fact, it is again the self-dual equation for the gauge field $A$, where the mass $m$ is now an operator, namely $m=(1+a\Box)/b\ve$.

1375: 

1376: In Appendix \ref{nearextremal} we give one further example, corresponding to near-extremal solutions.

1377: 

1378: 

1379: 

1380: 

1381: 

1382: \subsection{Symplectic flux through the boundary}\label{symf}

1383: 

1384: In section \ref{regularitysec} we derived the normalizability condition for Euclidean solutions. For non-tachyonic Lorentzian solutions, all solutions are normalizable, and the Euclidean boundary condition turns into an in-going boundary condition near the brane. In order to have a well-defined quantization problem, one also needs to ensure that the symplectic structure is finite and conserved \cite{marolf}. Whereas this is automatically ensured for Neumann or Dirichlet boundary conditions, it needs to be checked for generalized boundary conditions.

1385: 

1386: Given either of the boundary conditions

1387: \bea

1388: f_i&=&J_i^f\nn

1389: A_i&=&J_i^A~,

1390: \eea

1391: vanishing of the symplectic flux in four dimensions requires that the matrices

1392: \be\label{sym}

1393: {\d J_i^f(x)\over\d A_j(x')}\quad\mbox{and}\quad{\d J_i^A(x)\over\d f_j(x')}

1394: \ee

1395: be symmetric. For our choice of boundary terms, the explicit expressions for $J_i^f$ and $J_i^A$ follow from \eq{generaleom}. It is then straightforward to check that the symmetry condition on \eq{sym} is met. Evaluating \eq{sym}, we encounter two types of structures: $\d_{ij}\d(x-x')$ and $(*\dd)_{ij}\d(x-x')$ (or the inverse of the latter). Both terms are symmetric. Symmetry of the latter is shown using partial integration.

1396: 

1397: In the scalar field case, there is an ultralocality requirement on the sources \eq{sym} \cite{marolf} in order to avoid partial derivatives contributing to the symplectic flux. For gauge fields, this requirement would seem to be spoiled by the presence of for example Chern-Simons terms. A modified symplectic structure can be defined \cite{Wittensl2z,marolf} such that there is no net contribution to the symplectic flux through the boundary.

1398: 

1399: 

1400: \subsection{Supersymmetric boundary conditions}\label{ssusy}

1401: 

1402: Another interesting aspect of the generalized boundary conditions is which ones give solutions preserving certain amount of supersymmetry asymptotically. This then tells us which deformations of the boundary CFT are supersymmetric deformations. The conclusion however usually does depend on how we embed the truncated bosonic field content in a bigger supergravity theory. 

1403: 

1404: In our case the bosonic fields consist of only the graviton and abelian gauge field in asymptotically ${\rm{AdS}}_4$ space, in particular we have no scalars. The minimal theory to embed these as a consistent truncation is the the four dimensional gauged $\CN=2$ supergravity \cite{Freedman:1976aw} where our fields furnish the gravity multiplet. This is an extended supergravity theory with gauge symmetry $SO(2)=U(1)$ which can be further viewed as the consistent $SO(5)^V$ truncation of the gauged maximal $SO(8)$ supergravity \cite{Romans:1983qi}. This special $SO(5)$ invariance projects out all scalar fields and our theory is now indeed a sector of M-theory dimensionally reduced on ${\rm{S}}^7$. 

1405: 

1406: In order to find the supersymmetric boundary conditions, we follow the procedure of \cite{Hollands:2006zu}. First, since fermionic fields contribute to the symplectic flux through ${\rm AdS}$ boundary, we impose conditions on the gravitini fields so that the latter vanishes. Second, we find boundary conditions for the gauge fields consistent with supersymmetry transformations. We relegate the detailed analysis to Appendix \ref{susyapp} and simply present the conclusion here. We find that the only supersymmetric boundary condition for the gauge field is the Dirichlet boundary condition:

1407: \be

1408: A_i=0~,

1409: \ee

1410: which corresponds to the boundary action $S_{\sm{bdy}}=\int\dd^3x fA$, in Lorentzian signature. 

1411: 

1412: 

1413: \section{The flow of the boundary CFT}\label{fl}

1414: 

1415: %In the previous section we saw how the $S$- and $T$-generators of $SL(2,{\mathbb{Z}})$ act on the Neumann boundary condition problem, thereby reproducing Witten's results \cite{Wittensl2z}. 

1416: In this section we will 

1417: %apply this method to 

1418: analyze the electric-magnetic flow of a general CFT with an AdS$_4$ dual with ${\cal N}=2$ supersymmetry. This is done by studying how generalized boundary conditions transform under electric-magnetic transformations. For convenience we will work in the $SL(2,{\mathbb{R}})$ formulation ($SL(1,1)$ in the Euclidean) relevant at the level of supergravity. Of course, quantum mechanically only a discrete subgroup of this is a symmetry. The latter is obtained by evaluating our formulas at discrete points.

1419: 

1420: \subsection{Flow of Dirichlet-Neumann boundary conditions}\label{clem}

1421: 

1422: We now investigate the properties of the effective action under $SL(2,\mathbb{R})$ transformations. Since $T$-transformations simply act as translations of $\th$, in this section we will work with the compact part, i.~e.~electric-magnetic rotations. For this purpose, we may take $\th^*=\th=0$. In the Lorentzian case, electric-magnetic duality acts as $SO(2)$ transformations. As seen above, the relevant pair is not $(E,A)$, but $(v,A)$ and their corresponding dual field strengths $(E,F)$. In the Euclidean case, we get $SO(1,1)$ instead. 

1423: 

1424: The relevant transformation is

1425: \bea\label{lorenflow}

1426: g^*v_i^*&=&c\, gv_i+s\,{A_i\over g}\nn

1427: {A_i^*\over g^*}&=&ts\, gv_i+c\,{A_i\over g}~.

1428: \eea

1429: Here we recall $t=-1$ for Lorentzian signature and $1$ for Euclidean. Also $c,s$ stands for $\cos\a,\sin\a$ in the former and $\cosh\s,\sinh\s$ in the latter. Taking the curl of this rotates electric and magnetic fields into each other. In the Euclidean case these are boosts, so we will loosely refer to them as the $SO(1,1)$ ``flow'' from now on. The first order bulk action is invariant under this up to a boundary term \eq{actiontransform}, which present below:

1430: \be\label{actionemrot}

1431: S_{\sm{bulk}}[v^*,A^*]-S_{\sm{bulk}}[v,A]=-t\int\left(s^2\,A\wedge\dd v+\half sc(g^2\,v\wedge\dd v+{t\over g^2}A\wedge\dd A)\right)~.

1432: \ee

1433: Of course, the boundary term $S_{\sm{bdy}}[v,A]$ of the effective action will transform as well, and its transformation properties need to be analyzed case by case. We will do this after introducing the general boundary conditions.

1434: 

1435: By minimizing the $SO(2)$-transformed action, it is not hard to check that the boundary conditions required for stationarity satisfy:

1436: \be

1437: A_i+{c\over s}\, v_i=0~,

1438: \ee

1439: This boundary condition is still linear, yet in contrast to \eq{dtbc} it preserves conformal invariance as it does not introduce a new scale. Thus, introducing the $SO(2)$ flow we probe a one-parameter family of conformally invariant boundary conditions. The $SO(2)$ flow corresponds to a 1-parameter deformation of boundary theories.

1440: 

1441: It is clear that for $c/s=\infty$ we get the Neumann boundary condition, and for $c/s=0$ the Dirichlet one.

1442: We started with a situation where $A$ was integrated over but $v$ was fixed such that $F_{ij}[v]$ is a flat connection. After applying the $S$-dual transformation, we get a new boundary condition where $A$ is fixed whereas $v$ becomes dynamical. This precisely parallels the CFT discussion in \cite{Wittensl2z}.

1443: 

1444: There is a unique marginal purely three dimensional action that is invariant under the electric-magnetic duality \eq{lorenflow}. It is the difference (Euclidean) or sum (Lorentzian) of two Chern-Simons terms:

1445: \be\label{invariant}

1446: S_{\sm{invariant}}[v,A]=\int\dd^3 x\left(v\wedge\dd v-tA\wedge\dd A\right)~.

1447: \ee

1448: 

1449: 

1450: \subsection{$SL(2,\mathbb{R})$ transformed boundary conditions}

1451: 

1452: We will now include the $\th$-term and analyze the transformation of general boundary conditions.

1453: The bulk action changes by (see \eq{actiontransform})

1454: \bea\label{actionflow}

1455: S[A^*,E^*]&=&S[A,E]+\int\dd^3x\left[\left(-{t\over2}g^2\,cs+g^2g^*{}^2{\th^*\over8\pi^2}\,s^2\right)fv\right.\nn

1456: &-&\left.t\left(s^2-{g^*{}^2\th^*\over4\pi^2}\,cs\right)fA+\left(-{cs\over2g^2}+{g^*{}^2\over g^2}{\th^*\over8\pi^2}\,c^2-{\th\over8\pi^2}\right)FA\right]~.\nn

1457: \eea

1458: The corresponding equations of variation are

1459: \bea\label{slflow}

1460: {1\over g}\,{\d S^*_{\sm{bdy}}\over \d f_i}&=&ts\left(g^*v^*-{g^*{}^2\th^*\over4\pi^2}{A^*\over g^*}\right)\,\nn

1461: g\,{\d S^*_{\sm{bdy}}\over\d A_i}&=& c\left(g^*f^*-{g^*{}^2\th^*\over4\pi^2}{F^*\over g^*}\right)~.

1462: \eea

1463: This can be re-organized in both signatures to

1464: \bea\label{slcomu}

1465: {\d S^*_{\sm{bdy}}\over \d f_i^*}&=&0\nn

1466: {\d S^*_{\sm{bdy}}\over\d A^*_i}&=&f^*-{\th^*\over4\pi^2}F^*=\sf f^*~.

1467: \eea

1468: Thanks to the fact that we are working off-shell, \eq{slcomu} says that extremization commutes with duality transformations. 

1469: 

1470: We can further simplify our discussion by viewing the $\th$-term as a term in the boundary action. We may now keep factors of $g$ implicit and use \eq{actionemrot} to find  

1471: \bea\label{floweqa}

1472: {\d S^*_{\sm{bdy}}\over \d f_i}&=&ts\left(c\,v+s\,{A}\right)\,\nn

1473: {\d S^*_{\sm{bdy}}\over\d A_i}&=& c\left(c\,f+s\,{F}\right)~.

1474: \eea

1475: Since these transformations interpolate between different boundary effective actions, we will refer to this as the flow of the boundary theory (which is not to be confused with RG flow, although in the next section we will explain how they are connected). 

1476:  The flow starts at $\s=0$, $s=0$, $c=1$, where it of course reduces to \eq{bc}. The Lorentzian flow swaps the value of $c,s$ at the end. In contrast, the Euclidean flow ends at $\sigma=-\infty$ where $s=-c$ blows up. 

1477:  

1478: \subsection{Massive deformation}

1479: 

1480: We now come to the case of massive deformations, given by  the linear boundary condition \eq{dtbc} which explicitly breaks conformal invariance. In fact, it adds a relevant operator to the action. It corresponds to the following choice of boundary term:

1481: \be\label{relevantdef}

1482: S_{\sm{bdy}}[f,A]=\int\dd^3x\left(A_i f_i-{1\over2m}\, f_i^2\right)~.

1483: \ee

1484: Without the deformation term, this action simply imposes the Dirichlet boundary condition $A=0$. It is easy to check that after the flow we get again a single independent equation

1485: \be

1486: c\,A_i+ts\,v={1\over m}\left(c\,f_i+s\,F_i\right)~,

1487: \ee

1488: which is of course nothing else than the flow of the linear boundary condition \eq{dtbc}.

1489: In the Euclidean case, it is interesting to note that in the limit $\s\rightarrow\infty$ the above reduces to

1490: \be

1491: \hat A_i=-{1\over m}\,\e_{ijk}\pa_j\hat A_k

1492: \ee

1493: where $\hat A_i=A_i-v_i$. Hence, in the limit $\s\rightarrow\infty$ we get the self-dual massive theory of \cite{PTvN}!

1494: 

1495: In Lorentzian signature, we can continue the flow to $c=0,s=1$ where the boundary condition becomes 

1496: \be\label{srelevant}

1497: v=-{1\over m}\,F~,

1498: \ee

1499: which is the linear boundary condition on the $S$-dual fields. 

1500: 

1501: 

1502: 

1503: 

1504: \subsection{Self-dual boundary conditions}

1505: 

1506: We will now study the self-dual case, which corresponds to the equations of motion of the topologically massive theory \cite{PTvN}. We look for a choice of boundary terms that combines the self-dual and massive boundary conditions leading to \eq{SD}. This is achieved by the choice:

1507: \be\label{fF}

1508: S_{\sm{bdy}}[A,f]=-\int\dd^3x\left({g^2\over2m}\,f^2+{\ve\over m}\,fF\right)~,

1509: \ee

1510: where we have reinstated factors of $g$ left out in \eq{sdbc}, and for simplicity we first consider $\th=0$. After some algebra, we find that the modified equations of motion combined together lead to

1511: \bea\label{sdtransform}

1512: f&=&-{\ve\over g^2}\,F\nn

1513: A&=&{\ve\over m}\,F~,

1514: \eea

1515: that is, precisely the equations of motion of the ``master action'' for the self-dual theory! \cite{DJ}. It is remarkable that all $\s$-dependence in \eq{sdtransform} has disappeared. Another way to see this is to simply let the flow act directly on \eq{sdtransform}. It is easy to see these equations are invariant under it. So we find that the self-dual theory remains invariant during the whole flow! This is exactly what we expect: if we start with the self-dual boundary condition \eq{sdbc} which is a solution of the bulk self-duality equation \eq{SD}, the solution has vanishing bulk stress-energy tensor:

1516: \be\label{tmn=0}

1517: T_{\m\n}=0~.

1518: \ee

1519: Now since electric-magnetic duality preserves the form of the stress-energy tensor, \eq{tmn=0} should be true throughout the flow. This is exactly what we find: \eq{sdtransform} are valid throughout, and this ensures vanishing stress tensor. 

1520: Of course, the above action again contains relevant deformations since the theory is massive. 

1521: 

1522: We should notice here that, although the equations of motion \eq{sdtransform} are invariant under $SO(1,1)$, the action \eq{fF} (or, for that matter, the full action including the bulk term) is not invariant under it. In Appendix \ref{sdactions} we give two other choices of boundary terms that give the same boundary conditions. These choices should be continuously related to the one in \eq{fF} by the flow.

1523: %In other words, $SO(1,1)$ generates a one-parameter family of actions, all of which correspond to the same invariant choice of boundary conditions. A

1524: 

1525: This discussion should shed more light on the ``self-duality'' property of this three-dimensional theory: this is the boundary theory that corresponds to a choice of boundary conditions that leaves the action invariant under four-dimensional electric-magnetic transformations. Notice that the bulk part of the action is essential in checking this invariance. It is also easy to check that none of the purely three-dimensional actions \cite{DJ} are invariant under \eq{lorenflow}. They are only invariant under the transformation in \cite{hagen}, which is different from electric-magnetic duality in the bulk. In fact, the unique boundary invariant at this order was written in equation \eq{invariant}, which is simply two non-interacting copies of abelian Chern-Simons theory. The point is that the effective action receives a boundary contribution coming from the variation of the bulk term $\int\dd^4x\,E_i\pa_r A_i$. 

1526: 

1527: 

1528: 

1529: 

1530: \subsection{General case}\label{generalflow}

1531: 

1532: We are now ready to study the more general boundary action \eq{generalaction}. For simplicity we will set $g=1$. The transformed equations \eq{generaleom} read

1533: \bea\label{generalflow}

1534: \b\,v'+\g\,A'+\e\,f'+\m\,F'&=&0\nn

1535: \a\,F'+(\g-1)\,f'+\d\,*\dd F'+\m\,*\dd f'&=&0~.

1536: \eea

1537: Performing linear combinations as in \eq{generaleom1}

1538: %\bea\label{generalflow1}

1539: %a_1\,f'+a_2\,F'+a_3\,*\dd f'&=&0\nn

1540: %b_1\,f'+b_2\,F'-a_3\,*\dd F'&=&0~.

1541: %\eea

1542: %In terms of $f,F$, this 

1543: leads to 

1544: \bea\label{generalflow2}

1545: a_1'\,f+a_2'\,F+a_3'\,*\dd f&=&0\nn

1546: b_1'\,f+b_2'\,F-a_3'\,*\dd F&=&0~,

1547: \eea

1548: where we have the transformed coefficients

1549: \bea\label{flowcoeff}

1550: &&a_1'=a_1c^2+t(a_2+tb_1)cs+tb_2s^2\nn

1551: &&a_2'=a_2c^2+(a_1+b_2)cs+b_1s^2\nn

1552: &&b_1'=b_1c^2+t(a_1+b_2)cs+a_2s^2\nn

1553: &&b_2'=b_2c^2+t(a_2+tb_1)cs+ta_1s^2\nn

1554: &&a_3'=a_3~.

1555: \eea

1556: 

1557: 

1558: Remarkably it can be shown that the following are invariants under the flow \eq{flowcoeff}

1559: \bea\label{invariants}

1560: && a_1'-b_2'=a_1-b_2,\quad a_2'-t\,b_1'=a_2-t\,b_1,\quad a_3'=a_3,\nn

1561: && a_1'b_2'-a_2'b_1'=a_1b_2-a_2b_1,\quad c_1'd_2'-c_2'd_1'=0~.

1562: \eea

1563: Using the definition of $a,b$  in \eq{ab} we see that these are invariants as well. This means under the flow, the equation for $A$ remains unchanged

1564: \be

1565: (1+a\Box)A=bF~.

1566: \ee

1567: From this, we conclude that for generic values of coupling constants, the equations \eq{generaleom} and hence the boundary conditions are invariant under the flow! We should stress though that the effective action \eq{generalaction} itself is in general not invariant and hence the invariance of the equation is not a consequence of

1568: the invariance of the action.This is reminiscent of the realization of EM

1569: duality in the 4d theory!

1570: 

1571: This would seem to contradict what we found above where there were nontrivial flows of 

1572: boundary condition. However recall $a,b$ are well-defined only if $a_1b_2-a_2b_1\ne0$. In 

1573: fact, it's easy to check that actions given in \eq{relevantdef} \eq{fF} 

1574: \eq{moresdac}, as well as the trivial case $S_{\sm{bdy}}=0$, all have $a_1b_2-a_2b_1=0$. 

1575: 

1576: %Let us comment that in the flows analyzed in this section we set $g=1$. This is fine as long as $g^*=g$ (see comments at the end of section \ref{sl2rd}), however if the dual coupling is for example $g^*=1/g$ then the boundary theory is only invariant at the self-dual point $g=1$. Similar comments apply to other expressions for $g^*(g,\th)$.

1577: 

1578: \section{RG-flows, $S$-duality, and particle-vortex duality}\label{RGf}

1579: 

1580: Now we come to the interplay between $S$-duality and RG-flows. We will first compute the one- and two-point functions using the method of holographic renormalization, and then explain how $S$-duality acts on them. We will then discuss the physical interpretation, which requires considering both electric and magnetic sources in the AdS/CFT dictionary. Finally we will discuss massive deformations leading to RG-flows from one theory to its $S$-dual, and compute the effective action.

1581: 

1582: \subsection{$S$-duality of the two-point function and Legendre transforms}

1583: 

1584: \subsubsection*{Legendre transformed one-point function}

1585: 

1586: We can now easily compute the 1- and 2-point functions. We use the method of holographic renormalization \cite{dHSS} (see also \cite{Skenderis}). After solving the bulk equations of motion, one regularizes the action and adds counterterms. To compute the 1-point function, one needs to take the derivative of the action with respect to the source $A_{i(0)}(x)$. Doing this in general is non-trivial, as the dependence of $f_i(x)$ on $A_{i(0)}$ depends on the particular solution. The way to do this in general is to write the boundary action in terms of the regulated field $A_i(\e,x)$. We find for the dimension-2 current:

1587: \be\label{dim2}

1588: \d S=\int\dd^3x\left[{1\over g^2}\,(\pa_rA_i(r,x)-\pa_iA_r(r,x))+{\th\over4\pi^2}\,F_i(r,x)\right]\d A_i(r,x)|_{r=\e}~,

1589: \ee

1590: where we have used the equations of motion. The above variation is completely finite, as it should. It is shown in section \ref{backreaction} how to renormalize the action, including the coupling to gravity\footnote{In usual holographic renormalization, one should also use the induced boundary metric $\g_{ij}(\e,x)=1/\e^2g_{ij}(\e,x)$. Thanks to conformal invariance of the matter part of the action the powers of $\e$ drop out. See section \ref{1of}}. We are left with:

1591: \bea\label{1pt}

1592: \bra {\cal O}_{2i}(x)\ket_A&=&\lim_{\e\rightarrow0}{\d S[A]\over\d A_i(\e,x)}=\left(-E_i(r,x)+{\th\over4\pi^2}\,F_i(r,x)\right)|_{r=0}=-f_i(x)+{\th\over4\pi^2}\,F_i(x)\nn

1593: &=&-{\sf f_i}(x)~.

1594: \eea

1595: The first term is the usual one, see \eq{recipe}. The second term is the contribution from the $\th$-term. The $\th$-term does not modify the equations of motion in the bulk. However, it does modify the on-shell action, and therefore it contributes to the expectation value of the conserved current.

1596: 

1597: %We emphasise that, when we consider particular solutions, this result is valid for {\it any} choice of $f_i(x)$. Being a generic result, it will be enough to do the computations of 1- and 2-point functions for particular solutions and write them back in terms of $f$ and $F$. For simplicity, the condition we will use is the regularity condition. We get

1598: %\be

1599: %\bra J\ket_A=-f_i-{g^2\th\over\sqrt{-\Box}}\,*\dd f~,

1600: %\ee

1601: %which when written back in local form agrees with the general result \eq{1pt}.

1602: 

1603: 

1604: In the theory where we have the operator of dimension 2, we have shown that the generating functional of connected correlation functions is: $W=-S_{\sm{on-shell}}=\half\int\dd^3x\,{\sf f} A$.

1605: One can check that the effective action is minus this $\G=S_{\sm{on-shell}}$.

1606: 

1607: Now as we discussed in section \ref{scalardualities}, we can get to the dual CFT if we do a Legendre transform. The roles of the effective action and the generating functional are then interchanged. In particular, for a linear choice of boundary conditions we have

1608: \bea

1609: \ti\G&=&W\nn

1610: \ti W&=&\G~,

1611: \eea

1612: where the right-hand side should be viewed as a function of the source $A$, and the left-hand side as a function of the dual source $\ti f$.

1613: %In four dimensions two quantization schemes, so we can do Legendre transform:

1614: %\be

1615: %\ti S[\ti f]=S[A]-\int\dd^3x\,A\,{\d S\over\d A}~.

1616: %\ee

1617: %The on-shell acion is $S_{\sm{on-shell}}=-\half\int\,\ti fA$, hence

1618: %\be

1619: %\ti S_{\sm{on-shell}}[\ti f]= \half\int\dd^3x\,\ti fA=-S_{\sm{on-shell}}[A]

1620: %\ee

1621: %In this simple case, the dual action is just minus the action, written in the dual variables. The tilde on $\ti f$ is appropriate, as in the presence of a $\th$-term the conjugate to $A$ is $\ti f$ not $f$ as we see from \eq{1pt}.

1622: It is easy to check that the one-point function is independent of the linear choice of boundary condition. That 

1623: is basically what the generic holographic result \eq{dim2} tells us. For any such choice we get:

1624: %Using the boundary condition $f=\a |p|A$ for some $\a$, we get

1625: %\be

1626: %\ti S_{\sm{on-shell}}[\ti f]=\half\int\dd^3x\,{1\over\a|p|\left(1+{\th^2\over\a^2}\right)}\,\ti f\left(\Pi+{\th\over\a|p|}\,*\dd\right)\ti f

1627: %\ee

1628: %and the 1-point function of the dual theory is indeed

1629: \be\label{1ptt}

1630: \bra \ti {\cal O}_1\ket_{\sf f}=-{\d\ti W[\sf f]\over\d\sf f}=-\Pi\,A

1631: \ee

1632: as we would expect. This is a gauge field of dimension 1, projected onto its transverse part. From this we now get

1633: the conserved current of the dual theory, 

1634: \be

1635: \bra\ti{\cal O}_2\ket_{\sf f}=-F~.

1636: \ee

1637: 

1638: \subsubsection*{$S$-duality of the one- and two-point functions}

1639: 

1640: We saw that the action $S[A,E]$ and its $S$-dual $S[A',E']$ are each other's Legendre transforms, \eq{SSA}. This explains the minus sign in comparing $S[A]$ \eq{Aaction} with its $S$-dual \eq{Sd}. The generating functional of connected correlation functions of the $S$-dual theory thus coincides with the Legendre transformed functional, with the $S$-dual couplings:

1641: \be\label{Wp}

1642: W'[{\sf f}']=\ti W[{\sf f}';g',\th']~.

1643: \ee

1644: Let us spell this out in more detail in this case. Recall the definition of the generator of connected correlation functions: $-W[A]=S_{\sm{on-shell}}[A]=-\half\int{\sf f}A$, where ${\sf f}={\sf f}[A]$. Define now the Legendre transform:

1645: \be\label{lW}

1646: \ti W[\ti A]=W[A]+\int\ti A F~,

1647: \ee

1648: therefore

1649: \be\label{tiW}

1650: {\d\ti W\over\d\ti A}=F~~,~~{\d W\over\d A}=-\ti F~.

1651: \ee

1652: Using \eq{tiW} and filling the value of $W[A]$ in \eq{lW}, we find in this case $\ti W[\ti A]=-\half\int{\sf f}A=S[A]$. From \eq{tiW} we then precisely find the definition of $S$-duality \eq{Sduality}, as it should. This justifies \eq{Wp} and agrees with what we found in \eq{SSA}. We rewrite the dual potential as:

1653: \be

1654: \ti W[\ti A]=\half\int\ti{\sf f}\ti A=-\ti S[\ti A]~.

1655: \ee

1656: This is just what we expect -- the dual CFT is related to the dual bulk action in the standard way.

1657: 

1658: 

1659: The one-point function of the $S$-dual conserved current is

1660: \be\label{1pfd}

1661: \bra {\cal O}_{2i}'\ket_{A'}={\d S'\over\d A'}=-F=-{\sf f}'~.

1662: \ee

1663: 

1664: We now compute the two-point function and its $S$-dual. The two-point function is defined as

1665: 

1666: \bea

1667: \bra {\cal O}_{2i}(x){\cal O}_{2j}(x')\ket_{A=0}&=&{\d\over\d A_i(x)}{\d\over\d A_j(x')}\,S_{\sm{on-shell}}|_{A=0}\nn

1668: &=&{\sqrt{-\Box}\over g^2}\,\Pi_{ij}(x-x')+{\th\over4\pi^2}\,(*\dd)_{ij}\d(x-x')~.

1669: \eea

1670: In momentum space,

1671: \be\label{2ptem}

1672: \bra {\cal O}_{2i}(p){\cal O}_{2i}(-p)\ket_{A=0}={1\over g^2}|p|\Pi_{ij}-{\th\over(4\pi^2)^2}i\e_{ijk}p_k~.

1673: \ee

1674: Since the right-hand side only depends on the couplings, in order to compute the $S$-dual of this we might just apply \eq{Scouplings} and 

1675: get the $S$-dual result in \cite{Wittensl2z,tassos}. However, our formalism allows us to perform a check from first principles, which is what

1676: we will now do. This will be useful both in order to understand the structure of $S$-duality, and apply it in the general case.

1677: 

1678: The definition of the two-point function in the $S$-dual theory is simply:

1679: \be

1680: \bra{\cal O}_{2i}'{\cal O}_{2j}'\ket={\d\over\d A_i'}{\d\over\d A'_j}\,S'_{\sm{on-shell}}~.

1681: \ee

1682: We use the transformations \eq{dualEF} to compute the $S$-dual of the on-shell action

1683: \bea

1684: S'_{\sm{on-shell}}&=&-\half\int\dd^3x\,{\sf E}' A'=\half\int\dd^3x\,EA-{\th\over4\pi^2}\,AF\nn

1685: &=&\half\int\dd^3x\left(-{1\over g^2}\,AM\Pi A-{\th\over4\pi^2}\,A*\dd A\right)~,

1686: \eea

1687: where in the last line we have used the in-going boundary condition. This is, as announced, minus the on-shell action.

1688: 

1689: We now solve \eq{dualEF} and write the functional derivative as follows:

1690: \be

1691: {\d\over\d A_i'}=-\left({\th'\over4\pi^2}+g^2\left(1+{\th\th'\over(4\pi^2)^2}\right)M^{-1}*\dd\right)_{ij}{\d\over\d A_j}~,

1692: \ee

1693: where we have used the first of the identities \eq{id4}.

1694: 

1695: Since $\Pi$ is non-invertible, these identities should be understood as follows. One adds a gauge-fixing term to the action and inverting the

1696: resulting modified propagator. The final result does not contain any inverse propagators, and one can safely take the limit that the

1697: gauge-fixing terms go to zero. We get:

1698: \bea\label{dual}

1699: \bra {\cal O}_{2i}'{\cal O}_{2j}'\ket &=&{\d\over\d A_i'}{\d\over\d A_j'}\,S'\nn

1700: &=&\left(-{\th'\over4\pi^2}+g^2\left(1+{\th\th'\over(4\pi^2)^2}\right)M^{-1}*\dd\right)_{ik}\left(-{\th'\over4\pi^2}

1701: +g^2\left(1+{\th\th'\over(4\pi^2)^2}\right)M^{-1}*\dd\right)_{jl}\nn

1702: &\times&{\d\over\d A_k}{\d\over\d A_l}S

1703: \nn

1704: %&=&\left({\th'\over4\pi^2}-g^2\left(1+{\th\th'\over(4\pi^2)^2}\right)M^{-1}*\dd\right)_{ik}\left({1\over g^2}M\Pi-{\th*\over4\pi^2}\dd\right)_{kl}\left({\th'\over4\pi^2}+g^2\left(1+{\th\th'\over(4\pi^2)^2}\right)M^{-1}*\dd\right)_{lj}\nn

1705: &=&\left({\th'^2\over(4\pi^2)^2 g^2}+g^2\left(1+{\th\th'\over(4\pi^2)^2}\right)^2\right)M\Pi-\left({\th\th'^2\over(4\pi^2)^3}+{\th\over4\pi^2} g^4(1+{\th\th'\over(4\pi^2)^2})^2\right)*\dd\nn

1706: &=&{g^2\over1+g^2{\th^2\over(4\pi^2)^2}}\,M\Pi-{{g^4\th\over4\pi^2}\over1+{g^4\th^2\over(4\pi^2)^2}}\,*\dd

1707: \eea

1708: which is indeed the $S$-dual transform in the usual sense.

1709: 

1710: 

1711: \subsection{Physical interpretation of S-duality}\label{phys}

1712: 

1713: 

1714: 

1715: We have seen how $S$-duality acts in the bulk and on the boundary, and we now come to the physical interpretation. 

1716: 

1717: In this section we will deal with the pure Dirichlet/Neumann problems, therefore it will not be necessary to add any additional boundary terms. More general cases will be worked out in the next section.

1718: 

1719: We saw in section \ref{gba} that the parameter $\g$ interpolates between mutually $S$-dual solutions. This is easy to see from \eq{generaleom}: the case $\g=1$ corresponds to the Dirichlet boundary condition, whereas $\g=0$ is the Neumann boundary condition. Inclusion of the sources $J$ and $\ti{\cal J}$ will clarify what this $S$-duality actually means for the boundary CFT. Indeed, \eq{generaleom} is not quite symmetric with respect to $J$ and $\ti{\cal J}$: $J$ has dimension 1, whereas $\ti {\cal J}$ has dimension 2. Therefore, there exists a $\ti J$ such that $\ti {\cal J}=*\dd\ti J$. We can then rewrite the source terms in the following symmetric form:

1720: \be\label{ems}

1721: S_{\sm{source}}=-\int\dd^3x\left(F\ti J+fJ\right)=-\int\dd^3x\left(A\ti{\cal J}+fJ\right)~.

1722: \ee

1723: Although $\ti J$ is not necessarily conserved, this expression is gauge invariant and hence well-defined. In 

1724: a theory with gauge fields, this expression has a natural interpretation. Let us promote $J$ and $\ti J$ to bulk

1725: sources $J(r,x)$, $\ti J(r,x)$ localized at the boundary, with a delta-function at $r=0$. ${\cal J}$ is 

1726: then the magnetic source that couples to the bulk electric field $E(r,x)$. $\ti J(r,x)$, on the other hand,

1727: acts as a source for the magnetic field in the directions transverse to it. This is easy to see: if the sources would contain any bulk contributions, they would modify the equations of motion as

1728: \bea

1729: E&=&-{1\over g^2}\,\pa_rA+{1\over g^2}\,J\nn

1730: \pa_rE&=&{1\over g^2}\,*\dd F-\ti J~.

1731: \eea

1732: Clearly then, $\ti J$ will contribute to the curl of the magnetic field, and $*\dd J$ will contribute to the curl of the electric field. We write $(J_e,J_m)=(\ti{\cal J},{\cal J})$. In this paper we will not consider any bulk contributions coming from the sources. When restricted to the boundary, both sources have dimension 2. However, since they are

1733: conserved we can write $(\ti{\cal J},{\cal J})=*\dd(\ti J,J)$ where $(\ti J,J)$ have dimension 1 in the Legendre transformed theory.

1734: Bulk electric-magnetic duality interchanges them: $(\ti J',J')=(J,-\ti J)$.

1735: 

1736: In AdS/CFT with the usual boundary conditions, $J$ is interpreted as the source in the boundary CFT.

1737: However, the above makes clear that, in the presence of both electric and magnetic charges, the theory

1738: can also be deformed by $\ti J$, hence it ought to have a boundary interpretation. Since it has dimension

1739: 1, this suggests that it is dual to an operator of dimension 2, and in fact this operator should be 

1740: $S$-dual to ${\cal O}_2$. Furthermore, from the existence of an alternative quantization scheme where

1741: the boundary operator has dimension 1, one would guess that the bulk sources $\ti{\cal J}$ and ${\cal J}$

1742: themselves (restricted to the boundary) have an interpretation in terms of dual operators of dimension 1.

1743: This is in fact very natural, as ${\cal J}$ and $\ti{\cal J}$ are conserved so they might give rise to

1744: a boundary operator which is the gauge field itself. We will now verify that this is indeed the case. 

1745: 

1746: We start with the on-shell action $S_{\sm{on-shell}}=-\half\int\dd^3x\,{\sf f}A$. Now instead of fixing

1747: the gauge field $A$, we will fix the electric field by a source of dimension 1:

1748: \be

1749: \ti J=-J'=-A'={\sf v}~.

1750: \ee

1751: The prime suggests that this is the $S$-dual to $J$, however at this stage we are not making use of 

1752: $S$-duality and we regard the above purely as a boundary problem for ${\sf{v}}$. Clearly, the dual operator will have dimension 2. We compute:

1753: \be

1754: \bra{\cal O}_2'\ket=-{\d W'\over\d{\sf v}}=-F~,

1755: \ee

1756: where we have used $W'=S'_{\sm{on-shell}}=\half\int\dd^3x\,{\sf f}A$. This result agrees with \eq{1pfd}. We now compute the two-point function. To do that, we use regularity to write $F=M\Pi v$ and take a further derivative with respect to ${\sf v}$. In order to do that, we invert ${\sf v}=v-\th A$. We get:

1757: \be

1758: v={1\over1+{\th^2\over(4\pi^2)^2}}\left({\sf v}-{\th\over4\pi^2 M}\,*\dd{\sf v}\right)~.

1759: \ee

1760: So we get

1761: \be

1762: \bra{\cal O}_2'{\cal O}_2'\ket={M\over1+{\th^2\over(4\pi^2)^2}}\,\Pi-{{\th\over4\pi^2}\over1+{\th^2\over(4\pi^2)^2}}\,*\dd~.

1763: \ee

1764: This is exactly the $S$-dual \eq{dual} of the two-point function of a conserved current! (we have set 

1765: $g=1$). Notice that we have not assumed any knowledge of $S$-duality here: it simply follows from the fact that we identify $\ti J$ as the source. This describes the lower-left corner of figure \ref{diagram}.

1766: 

1767: The upper right corner of figure \ref{diagram} is the (equivalent) description in terms of the dual gauge field:

1768: this is obtained by taking the dual source ${\cal J}'=-\ti{\cal J}$ and fixing:

1769: \be

1770: \ti{\cal J}={\sf f}.

1771: \ee

1772: We find:

1773: \bea

1774: \bra{\cal O}_1'\ket_{\cal J}&=&\Pi A\nn

1775: \bra{\cal O}_1'{\cal O}'_1\ket&=&{1\over M\left(1+{\th^2\over(4\pi^2)^2}\right)}\,\Pi-{{\th\over4\pi^2}\over M^2\left(1+{\th^2\over(4\pi^2)^2}\right)}\,*\dd~.

1776: \eea

1777: \begin{figure}[t]

1778: \begin{center}

1779: \includegraphics[height=4in,width=5in]{diagram}

1780: \caption{\small Diagram of the various S-duality transformations.}

1781: \label{diagram}

1782: \end{center}

1783: \end{figure}

1784: This is indeed the $S$-dual of the two-point function of a gauge field and agrees with \eq{1ptt}.  Of course, taking the derivative of the above we 

1785: recover the correlation functions of the dual current. This is another

1786: check that the method of dual sources gives correct results.

1787: 

1788: We now complete the Dirichlet case on the main diagonal of \ref{diagram}. The upper-left corner is the usual Dirichlet boundary condition $A=J$. The lower-right case is obtained identifying $F={\cal J}$. The one- and two-point functions are then:

1789: \bea

1790: \bra{\cal O}_1\ket&=&-\Pi{\sf v}\nn

1791: \bra{\cal O}_1{\cal O}_1\ket&=&{1\over M}\,\Pi-{\th\over4\pi^2 M^2}\,*\dd~.

1792: \eea

1793: There is no $S$-duality transform here, as it should. Once more, taking the derivative we get the usual current.

1794: 

1795: We can now rephrase physically the situation in Figure \ref{diagram}. First of all, notice that the Legendre 

1796: transform and $S$-duality transformations that translate us vertically or horizontally are the same. In both cases, 

1797: the term that we add to the action is $\int{\sf f}A$. Only in one case (the horizontal one) this is regarded as a 

1798: coupling between the source and the current, whereas in the vertical case it is a Chern-Simons coupling between the 

1799: background field and the new dynamical gauge field. Obviously, the diagonal action simply corresponds to the action 

1800: of $*\dd$ and ensures that the theory is unitary. The difference between $S$-dual theories lies in what we 

1801: interpret as a source and as an operator. In the upper-left corner, we fix as a source the bulk gauge field $A$. The 

1802: corresponding CFT operator that couples to it is then an induced current of dimension 2. In the upper-right, on the 

1803: other hand, $A$ has become a dynamical field with corresponding conserved current $F$, and the background field $v$ 

1804: is fixed via its curl $f$. This is the particle-vortex duality. In the next subsection we will see how this duality 

1805: is achieved by an RG-flow. From the bulk point of view, particle-vortex duality corresponds to interchanging $J$ and 

1806: ${\cal J}'$ as we have shown. The latter fact may have applications to other situations, such as the non-abelian case 

1807: (see section 6.5).

1808: 

1809: \subsection{RG-flow of the two-point function and $S$-duality}

1810: 

1811: 

1812: %\subsubsection*{IR/UV flows and $S$-duality}

1813: 

1814: 

1815: 

1816: It is known from field theory that massive deformations of three-dimensional CFT's by double-trace operators lead to RG flows from the UV towards the IR fixed point. In \cite{tassos} it was shown using field theory arguments that if the IR dimension of a conserved current is 2, in the UV it corresponds to an operator that is $S$-dual to a gauge field of dimension 1. In this section we will show how this result comes about from the bulk point of view, and how it fits in the general picture of Figure \ref{diagram}.

1817: 

1818: We will modify the massive boundary condition to include a $\th$-angle:

1819: \be

1820: A+{1\over m}\,{\sf f}=J~,

1821: \ee

1822: in other words

1823: \be\label{massive}

1824: A-{\th\over m}\,F+{1\over m}\,f=J~.

1825: \ee

1826: As our second boundary condition, we take

1827: \be

1828: f=cM\Pi A~.

1829: \ee

1830: We will choose $c=-1$, which corresponds (for the tachyonic case) to the regular configurations, as explained in section \ref{regularitysec}. The value of $c$ only results in an overall rescaling of the two-point function, however $c$ has to be negative in order for the two-point function to be positive definite, as explained in section \ref{sdtmt}. We obtain the above boundary conditions if we choose $\g=1$, $\e=1/m$, $\m=-\th/m$, $\a=-M\th/m$ in \eq{generaleom}, and the other coefficients zero (in the rest of this section we rescale $\th$ by a factor of $4\pi^2$ compared to our previous formulas). The on-shell action is:

1831: \be\label{osA}

1832: S_{\sm{on-shell}}=\half\int\left(-{M\over m}(m-M)A\Pi A+{M\th\over m}\,A\wedge\dd A+2MJA\right)~.

1833: \ee

1834: Now we solve for $A$ in \eq{massive} in terms of the source. We get:

1835: \bea

1836: A&=&{m\over\G}\left((m-M)\,J+\th*\dd J\right)\nn

1837: \G&=&(m-M)^2-t\th^2M^2~.

1838: \eea

1839: Plugging this back in the action, we get

1840: \be\label{mons}

1841: S_{\sm{on-shell}}=\int{mM\over2\G}\left(J(m-M)\Pi J+\th J\wedge\dd J\right)~.

1842: \ee

1843: The one-point function is easily computed and found to be $\bra{\cal O}_2\ket=MA=-f$. It is clear that the IR limit $|p|\ll m$ of the two-point function is the usual one for a conserved current of dimension 2, \eq{2ptem}. On the other hand, it is quite remarkable that in the UV limit we get

1844: \be\label{2ptUV}

1845: \bra{\cal O}_{2i}{\cal O}_{2j}\ket={m\over1+\th^2}\left(\Pi_{ij}-\th {i\e_{ijk}p_k\over|p|}\right)+{m^2\over|p|(1+\th^2)^2} \left((1-\th^2)\,\Pi_{ij}-2\th {i\e_{ijk}p_k\over|p|}\right)~.

1846: \ee

1847: The second term in \eq{2ptUV} was also found in \cite{tassos} and corresponds to the ($S$-dual) two-point 

1848: function of an operator of dimension 1 -- a gauge field. This is what we expect. 

1849: Notice that, once again, there is no assumption about $S$-duality made; the massive deformation automatically leads us to 

1850: the upper-right part of the diagram \ref{diagram}.

1851: 

1852: The first term in \eq{2ptUV} is a contact term \cite{tassos}. Naively it would correspond to an operator with UV dimension 3/2. However, a spin one 

1853: operator cannot have dimension 3/2 in a conformal theory, therefore it cannot be a chiral primary and it must be 

1854: a descent operator. There is indeed such an operator in QED in 3 dimensions \cite{BKW,KS}. In that case, the 

1855: current gets dimension 3/2 and it can be written as $f_i=\pa_i\f$ where $\f$ is a scalar field. However, the 

1856: presence of such a term without its primary field in the OPE breaks conformal invariance in the UV. 

1857: 

1858: 

1859: The result can be given a different interpretation that agrees with what we said earlier if we take 

1860: ${\cal J}=*\dd J$ as the source. In that case, we find the two-point function of $\Pi{\sf{v}}$ with itself, and 

1861: we are in the 

1862: lower-right part of \ref{diagram}! In particular, the $S$-duality transform is absent here and will instead 

1863: appear in the IR.

1864: 

1865: 

1866: It is straightforward to check that the method to get the $S$-dual two-point functions outlined in section \ref{phys} works in this case as well. If we fix the boundary value of the gauge field, $J=A$, we easily find the two-point function from \eq{osA}. If instead we couple the theory to an electric source $\ti{\cal J}$  as in \eq{ems} and take derivatives of \eq{osA} with respect to $\ti{\cal J}$, we get the $S$-dual two-point function. This explains the $S$-duality in the flow when we turn on the mixed boundary condition \eq{massive}: in the IR limit we are fixing $A$ and $J$ is the usual magnetic source; in the UV, the second term dominates and $J$ becomes an electric source! Thus, \eq{massive} interpolates between the two.

1867: 

1868: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1869: %\subsubsection*{Boundary degrees of freedom}

1870: 

1871: %The types of boundary conditions we have chosen do not fully determine the gauge field. For example, the two boundary conditions leading to \eq{SD} still leave us with one massive degree of freedom, namely that described by the self-dual topologically massive theory \cite{PTvN,DJ}. We will now explain what this means from the CFT point of view. Consider the case where \eq{regeom} reduces to

1872: %\be\label{Asource}

1873: %A+{\ve\over m}\,F=J~,

1874: %\ee

1875: %for {\it some} $m$, not necessarily related to $|p|$. Of course, this is obtained from the previous example \eq{massive} if we send $m\rightarrow\infty$ while keeping $\th/m$ fixed. In \eq{regeom} we set $\a=\g=1$, $\m=0$, $\d=\ve/m$ in the first equation. For the second equation we have several choices. We will choose $\e=1/M$ and everything else zero. We solve \eq{Asource} as follows:

1876: %\be\label{Asolution0}

1877: %A_i=\left(1+{t\over m^2}\Box\Pi\right)^{-1}\left(J_i-{\ve\over m}\,\e_{ijk}\pa_jJ_k\right)+\hat A_i\equiv j+\hat A_i~,

1878: %\ee

1879: %and $f$ is determined by regularity. $\hat A$ is now the fluctuating piece the topologically massive equation. This will give a non-vanishing one-point function of the dual operator when we take the source to zero.

1880: %

1881: %We now compute the on-shell action as a function of the source. We get:

1882: %\be

1883: %S_{\sm{bdy}}=\int\dd^3x\left(\half\, AF+{\ve\over2m}\,F^2-\ti{\cal J}A\right)~,

1884: %\ee

1885: %where the non-local term miraculously cancels out. Adding the bulk contribution, we get:

1886: %\be\label{Sos}

1887: %S_{\sm{on-shell}}=\int\dd^3x\left(\half\, AM\Pi A+\half\,AF+{\ve\over2m}\,F^2-\ti{\cal J} A\right)

1888: %\ee

1889: %which filling in the solution \eq{Asolution0} gives

1890: %\be

1891: %S_{\sm{on-shell}}=\int\dd^3x\left(\half\,\hat A\left(M-{t\ve\over m}\,\Box\right)\Pi\hat A+\half\,\hat A\hat F

1892: %+jM\Pi\hat A+\half\,j\left(M+{t\ve\over m}\,\Box\right)\Pi j-\half\,j\wedge\dd j\right)~.

1893: %\ee

1894: %This gives

1895: %\bea

1896: %\bra{\cal O}_i\ket_{j=0}&=&-M\Pi_{ij}\hat A_j\nn

1897: %\bra{\cal O}_i{\cal O}_j\ket_{j=0}&=&\half\,M\Pi_{ij}-{\ve\over2m}\,(*\dd*\dd)_{ij}-\half\,(*\dd)_{ij}\nn

1898: %&=&\half\left(M-{\ve\over m}\,\Box\right)\Pi_{ij}+\half\,\dd_{ij}~.

1899: %\eea

1900: %By regularity,  $\bra {\cal O}\ket$ is the expected $f_i[\hat A]$. As we claimed, the one-point function is 

1901: %non-zero after we take the source to zero. This is a rather special property of boundary conditions like 

1902: %\eq{Asource} and \eq{massive}, which allow a fluctuating piece. The renormalization of the two-point function 

1903: %seems to have the a familiar form. As before, the IR and UV limits of the two-point function give rise to 

1904: %different operators.

1905: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1906: 

1907: \subsection{The effective action}

1908: 

1909: We now present an application that motivated this work, namely finding the 

1910: effective action on a stack of M2-branes using AdS/CFT for the particular bulk configurations that we have considered. This 

1911: effective action gives the response of the boundary theory to the action of an external source. Our abelian configuration corresponds

1912: either to a Higgs or a Coulomb phase, depending on the operator that gets a vev. As we have discussed, there are two possible 

1913: meanings to this effective action; the usual theory and the dual one where the on-shell bulk action is identified 

1914: with the boundary effective action. Since the main focus of this paper has been the dual CFT, we will consider 

1915: the latter.

1916: 

1917: In the case of instantons there is no back-reaction, hence we are ensured that the asymptotic expansion of the 

1918: on-shell action gives the full quantum effective action at large $N$. For a particular solution with the sources set 

1919: to zero this was shown in section 2.6 to be a pure number, $8\pi^2|\l|/g^2$. In the case of non-zero back-reaction, 

1920: since the back-reaction sets in only at order $r^4$, gravity and the matter fields decouple near the boundary and we 

1921: can solve the asymptotic equations in a fixed background. At order $r^4$ the back-reaction sets in as in \eq{grx}, 

1922: but this does not affect the two-point functions. The effective action up to this order 

1923: reduces in the IR to:

1924: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1925: %is given by \eq{Sos}, where $\hat A$ just contributes the constant term. In the IR, this reduces to:

1926: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1927: \be

1928: S_{\sm{eff}}=\half\int \left(F_{ij}[A]\Box^{-1/2}F_{ij}[A]+\th \e^{ijk}A_i\pa_jA_k\right)~.

1929: \ee

1930: $A$ is the one-point function of the gauge field as a response to the source $J$. The above is in fact the 

1931: on-shell action for the Dirichlet boundary problem, and it is the generic behavior that we found 

1932: under massive deformations: the theory flows towards the Dirichlet problem in the IR. We get a similar result if we start with the massive case \eq{mons}:

1933: \be

1934: S_{\sm{eff}}=\half\int \left(F_{ij}[A]\Box^{-1/2}F_{ij}[A]+\th m\e^{ijk}A_i\pa_j\Box^{-1/2}A_k\right)~.

1935: \ee

1936: We remark here that for zero theta angle, the above is the effective action of three-dimensional QED at large $N_f$. As we 

1937: noted in the previous section, there is another point where our results seem to agree with QED. The naive UV dimension of the 

1938: current is 3/2 rather than 2; the two-point function of such an operator in momentum space would be a contact term; this is 

1939: exactly what we find in \eq{2ptUV}. As discussed, if such a term is to be interpreted as an operator then it must be a non-primary operator. The present computation also predicts that the UV and IR couplings are 

1940: each other's S-duals\footnote{The existence of the two-point function with the operator of dimension 1/2 suggests that in 

1941: the UV the gauge field is dualized to a scalar: $F_i=\pa_i\f$, in agreement with \cite{BKW}.}. The subleading correction is 

1942: the two-point function of the dual gauge field.

1943: 

1944: 

1945: 

1946: \subsection{Non-abelian case}

1947: 

1948: Due to interactions, classical electric-magnetic duality does not straightforwardly generalize to non-abelian gauge theories

1949: \cite{DT} (but see \cite{DS} where electric-magnetic duality is achieved at the cubic level). The key point is that, as 

1950: explained earlier, a formal replacement $E\leftrightarrow\pm B$ in the Lagrangian does not suffice. Duality has to be

1951: realized at the level of the gauge potential. Duality in non-abelian theories involves quantum effects in an essential way,

1952: which goes beyond the scope of this work. We will suffice with a few comments and suggestions for further work.

1953: 

1954: Instanton configurations that minimize the energy will satisfy:

1955: \be\label{SDnA}

1956: E_i^a=F_{ri}^a=-\half\,\ve\,\e_{ijk}F^{a\,jk}~.

1957: \ee

1958: Restricted to the boundary, this gives again the same boundary condition as before. 

1959: %Now again in the non-abelian case, from Gauss constraint $D^iE_i^a=0$ (define dual $E_i^a=\e_{ijk} E_{jk}^a$) Bianchi 

1960: %follows: $D_{[i}E_{jk]}^a=0$, which is solved as $E_{ij}^a=F_{ij}^a[v]$ for some $v$. 

1961: If in addition we also choose the boundary condition $E_i^a=mA_i^a$, we get the non-abelian version of the self-dual

1962: equation:

1963: \be

1964: A_i^a={1\over2m}\,\e_{ijk}F_{jk}^a~.

1965: \ee

1966: This is the non-abelian generalization of \cite{DJ}.

1967: We have checked that usual 't Hooft instantons do not satisfy this condition but a modification of it. It would be interesting

1968: to see what boundary action gives rise to those solutions.

1969: 

1970: In section \ref{phys} we outlined a method to compute the two-point function at any of the points in the diagram in Figure \ref{diagram}. In particular, to compute the $S$-dual two-point functions we use electric rather than magnetic bulk sources, that is we couple the source to the gauge field rather than to the electric field. This method can be generalized to the non-abelian case. That is why we expect that some of our results may easily generalize to that case.

1971: 

1972: As we have noted, the particular compactification we have used has a non-abelian extension where the gauge symmetry is $SU(2)$,

1973: with ${\cal N}=2$ supersymmetry. As mentioned in the introduction, it is an interesting question whether this theory still

1974: has some sort of electric-magnetic duality and what is the holographic image of it. One may be able to shed light on this

1975: question by analyzing instanton solutions considered above. These have $T_{\m\n}=0$, and therefore gravity and the gauge

1976: fields basically decouple. It would be interesting to see whether the theory around such configurations exhibits definite

1977: electric-magnetic properties of the type in \cite{SW}.

1978: 

1979: 

1980: \section{Discussion}

1981: 

1982: The off-shell bulk electromagnetic action is invariant under electric-magnetic duality up to boundary terms. In this paper we 

1983: have analyzed the way these terms transform for arbitrary choice of boundary conditions. The fact that the action is 

1984: invariant off-shell implies that the boundary conditions transform as well. In the simplest example, Neumann and 

1985: Dirichlet boundary conditions for the gauge field are interchanged by $S$-duality. Classically, there is a one-parameter

1986: family of deformations connected by bulk duality. Quantum mechanically only a discrete subgroup survives. It would be interesting

1987: to check whether the latter statement also follows from quantization of Chern-Simons couplings. Operators of higher dimension

1988: are also mapped into each other by electric-magnetic transformations. Again, quantum mechanically the coefficients within a

1989: particular $SL(2,{\mathbb{Z}})$ orbit do not appear in arbitrary but only in specific  combinations. In this way, 

1990: electric-magnetic duality gives a way of probing the moduli space of deformations of effective CFT actions for operators of given dimension. This agrees with interesting experimental results in quantum Hall systems \cite{BD1,BD2}. In the massive case, $S$-duality can change dimension.

1991: 

1992: We presented a bulk computation of the finite renormalization of dimensions of operators under RG flow of the 

1993: deformed three-dimensional SCFT. The conserved current has IR dimension 2 but the expansion of its two-point function

1994: contains an $S$-dual dimension 1 gauge field in the UV. In other words, the 2-point function contains a

1995: contribution of the 2-point function of the dual current. This is in agreement

1996: with the field theory expectations \cite{tassos}. We also discussed the similarity of the effective action and the flow with those of 2+1 dimensional QED at large $N_f$. It may turn out that the high-energy limit of this theory is relevant to the SCFT in a phase where the non-abelian gauge symmetry is broken, or the theory has become classical due to the large $N$ limit. We notice, however, that whenever we have bulk instanton solutions we will get Chern-Simons terms on the boundary, which as stressed in \cite{schwarz} are not present in SYM theory.

1997: It would be interesting to study this further.

1998: 

1999: The form of the $S$-transformation worked out in \cite{Wittensl2z} depends on the choice of boundary 

2000: terms. As it turns out, it corresponds to pure Dirichlet and Neumann boundary conditions. In this paper we wrote down the

2001: $S$-generator for arbitrary boundary conditions.

2002: We found that generic choices of boundary conditions are actually invariant under electric-magnetic duality. A particularly 

2003: interesting case is the self-dual massive bulk solution. In this case, the boundary conditions do not entirely determine the

2004: solution, and there is a single remaining boundary degree of freedom. This degree of freedom in fact corresponds to the 

2005: self-dual topologically massive theory in three dimensions \cite{PTvN,DJ}. This should shed some new light on the meaning of 

2006: self-duality of topologically massive theories in three dimensions: it in fact corresponds to electric-magnetic duality in the bulk of 

2007: AdS$_4$! It would be interesting to extend this analysis to the non-abelian case, and we offered some thoughts to analyze

2008: this question. In particular, it seems that the method of doing AdS/CFT with both electric and magnetic sources should work also in that case. One could then test to what extent the effective non-abelian CFT's have dual properties.\\

2009: \\

2010: {\it Note added}: While this paper was being completed, \cite{herzog} appeared, where some related questions are studied.

2011: 

2012: 

2013: \section*{Acknowledgements}

2014: 

2015: \addcontentsline{toc}{section}{Acknowledgements}

2016: 

2017: We thank Stanley Deser, Marc Henneaux, Don Marolf, and Ioannis Papadimitriou, for useful comments and discussions, and especially Tassos Petkou for collaboration during the early stages of this work.

2018: 

2019: \appendix

2020: 

2021: \section{Another proof of $S$-duality}\label{2nds}

2022: 

2023: We now give the second proof, which relies on integrating out some fields (see also \cite{wittenabelian}). The equations of motion of \eq{1orderwcoupling} are:

2024: \bea\label{eomtiE}

2025: \d_E&:&\pa_rA_i-\pa_iA_r+g^2\,{\sf E}_i+{g^2\th\over4\pi^2}\,F_i=0~,\nn

2026: \d_A&:&-\pa_rE+{1\over g^2}\left(1+{g^4\th^2\over(4\pi^2)^2}\right)*\dd F+{g^2\th\over4\pi^2}\,*\dd\sf E=0~,\nn

2027: \d_{A_r}&:&\pa^i{\sf E}_i=0~.

2028: \eea

2029: 

2030: We first integrate out $\sf E$ using its equation of motion and the fact that it is conserved, getting the usual second-order action for

2031: the boundary component of the gauge field $A$:

2032: \be\label{Aaction}

2033: S[A;g,\th]=\int\dd^4x\left(-{1\over 2g^2}\,(\pa_rA)^2+{1\over2g^2}\,F^2-{\th\over4\pi^2}\,(\pa_rA)F\right)~.

2034: \ee

2035: This is the usual second order action \eq{actionwtheta}.

2036: 

2037: Next we use the equation of the gauge field to integrate out $A$ instead. We make use of conservation of $\sf E$ to introduce a gauge field

2038: $\sf v$:

2039: \bea

2040: \sf E&=&*\dd \sf v\nn

2041: \sf v&=&v-{\th\over4\pi^2}A~.

2042: \eea

2043: 

2044: Then the second equation in \eq{eomtiE} is integrated to:

2045: \be

2046: F={g^2\over1+{g^4\th^2\over(4\pi^2)^2}}\,\pa_r{\sf v}-{g^4{\th\over4\pi^2}\over1+{g^4\th^2\over(4\pi^2)^2}}\,{\sf E}~.

2047: \ee

2048: Observe that this is nothing but the $S$-dual of the equation of motion for $\sf E$! We get:

2049: \be\label{Sd}

2050: S[{\sf E}]=-\int\dd^4x\left(-\half{g^2\over1+{g^4\th^2\over(4\pi^2)^2}}\,(\pa_r{\sf v})^2+ \half{g^2\over1+{g^4\th^2\over(4\pi^2)^2}}\,{\sf E}^2+

2051: {g^4{\th\over4\pi^2}\over1+{g^4\th^2\over(4\pi^2)^2}}\,(\pa_r{\sf v}){\sf E}\right)~.

2052: \ee

2053: This is the $S$-dual of \eq{Aaction}! 

2054: 

2055: In conclusion, the gauge fields $A$ and ${\sf v}$ are each other's $S$-duals: integrating out $\sf v$ we get the usual second form of the

2056: action; integrating out $A$ gives its $S$-dual version where the gauge field is $\sf v$. The first order form of the action, on the other hand,

2057: \eq{1orderwcoupling} or equivalently \eq{1storderform}, interpolate between both. In particular, \eq{1orderwcoupling} has manifest $S$-duality.

2058: 

2059: The on-shell action now is:

2060: \be

2061: S_{\sm{on-shell}}=-\half\int\dd^3x\,\sf E A~,

2062: \ee

2063: which is obviously the same as \eq{onshellwth}. Notice however that in terms of the redefined $\sf E$, the on-shell action takes the same

2064: form as the on-shell action without the $\th$-term.

2065: 

2066: 

2067: 

2068: \section{Regularity of the solutions}\label{regularityapp}

2069: 

2070: In this appendix we study the regularity condition on bulk solutions and give some other formulas that were used in the main text.

2071: 

2072: \subsection{Gauge invariance}

2073: 

2074: We first discuss gauge invariance.

2075: The longitudinal, pure gauge part $\vf=-{i\over p^2}\,p^iA_i$ is not fixed by the equations 

2076: of motion. 

2077: %Fourier transforming, we find the non-local expression

2078: %\be\label{Aipos}

2079: %A_i^{\tiny{\mbox{T}}}(r,x)=\cos(r\sqrt{-\Box})A_i^{\tiny{\mbox{T}}}(x)+{1\over \sqrt{-\Box}}\sin(r\sqrt{-\Box})\,f_i(x)

2080: %\ee

2081: %In position space, we have 

2082: \bea\label{decomp}

2083: A_i(r,x)&=&A_i^{\tiny{\mbox{T}}}+\pa_i\vf\nn

2084: A_r(r,x)&=&\pa_r\vf\nn

2085: F_{ri}(r,x)&=&\pa_rA_i^{\tiny{\mbox{T}}}~.

2086: \eea

2087: From this, the radial gauge $A_r=0$ is simply $\pa_r\vf=0$, leaving the residual 

2088: $r$-independent gauge freedom $\vf(x)$. $\vf(x)$ realizes the gauge 

2089: symmetry of the dual CFT, where the background field has been gauged and a dual current is constructed from the

2090: dual gauge field $A$. By construction, $A_i^{\tiny{\mbox{T}}}$ and $f_i$ 

2091: are invariant under such transformations.

2092:  

2093: \subsection{Regularity in two different bulk gauges}

2094: As the topological term does not effect equations of motion, we will take the Maxwell action without $\theta$-term

2095: \be\label{emaction}

2096: S_{\sm{bulk}}(A)={1\over2}\int d^4x\sqrt{g}\,F_{\m\n}F^{\m\n}

2097: \ee

2098: 

2099: We will study two different gauges. The first is $\nabla_{\m}A^{\m}=0$, where the bulk equations are 

2100: \be\label{emeq}

2101: \nabla_{\m}F^{\m\n}=0,\quad \nabla_{\m}A^{\m}=0

2102: \ee

2103:  %After some algebra they become

2104: %\bea

2105: %&& \left(\partial_r-\frac{2}{r}\right)A_0(x) +\partial_i A_i(x) =0\nonumber\\

2106: %&& \bar\partial^2 A_0(x)-\partial_r\partial_i A_i(x)=0\nonumber\\

2107: %&& \partial_r^2 A_i(x) -\partial_i\partial_r A_0(x)+\bar\partial^2 A_i(x)=\partial_i \partial_j A_j(x)

2108: %\eea

2109: Making use of the following identities

2110: \bea

2111: && \partial_r^2-{2\over r}\partial_r+{2\over r^2}=(\partial_r-{1\over r})^2\nonumber\\

2112:  &&(\partial_r-{1\over r})^n(r\phi)=r\partial_r^n\phi

2113:  \eea

2114: and denoting ${\cal A}={A_0(r,\bar x)/r}$, we rewrite the equations 

2115: \bea\label{emeq2}

2116:  (\partial_r^2+\bar\partial^2){\cal A}&=&0\nonumber\\

2117:  (\partial_r^2+\bar\partial^2)A_i&=&2\partial_i{\cal A}

2118:  \eea

2119: which are solved by

2120: \bea\label{sol}

2121:  {\cal A}(r,\bar p)&=& {\cal A}_0(\bar p)\cosh(|\bar p|r)+{1\over|\bar p|}{\cal A}_1(\bar p)\sinh(|\bar p|r)\nn

2122:   A_i(r,\bar p)&=&A_i^{(0)}\cosh(|\bar p|r)+{1\over|\bar p|}{A}_i^{(1)}\sinh(|\bar p|r)\nn

2123:                &+& ir{\bar p_i\over|\bar p|}\left({\cal A}_0\sinh(|\bar p|r)+{1\over|\bar p|}{\cal A}_1\cosh(|\bar p|r)\right)\nn

2124: \eea

2125: The gauge fixing $\nabla_{\m}A^{\m}=0$ implies

2126: \be\label{aux}

2127:  {\cal A}_0(\bar p)=i\bar p_iA_i^{(0)}(\bar p),\quad  {\cal A}_1(\bar p)=i\bar p_iA_i^{(1)}(\bar p)

2128: \ee

2129: %from which follows 

2130:  %\bea\label{Ai1}

2131:  %A_i(r,\bar p)&=&A_i^{(0)}\cosh(|\bar p|r)+{1\over|\bar p|}\left({A}_i^{(1)}\sinh(|\bar p|r)-r{\bar p_i\bar p_j\over|\bar p|}A_j^{(1)}\cosh(|\bar p|r)\right)\nn

2132:  %&-&r{\bar p_i\bar p_j\over|\bar p|}A_j^{(0)}\sinh(|\bar p|r)

2133:  %\eea

2134: The two boundary functions $A_i^{(0)},A_i^{(1)}$ completely determine the bulk solutions. 

2135: 

2136: Near the boundary $r\!\rightarrow\!0$ we find

2137:  \bea\label{r0}

2138:  &&A_i(r,\bar p)=A_i^{(0)}(\bar p)+r\,\Pi_{ij}A_j^{(1)}(\bar p)+O(r^2)\nonumber\\

2139:  &&{E}_i=F_{0i}=\Pi_{ij}A_j^{(1)}(\bar p)+r\,\bar p_i\bar p_jA_j^{(0)}+O(r^2)

2140:  \eea

2141:  where

2142: \be\label{proj}

2143: \Pi_{ij}=\delta_{ij}-{p_ip_j\over p^2}

2144: \ee

2145: By construction, this matrix has a null eigenvector $p_i$. It's clear that $f_i=E_i^{(0)}$ is the transverse part of $A_i^{(1)}$.

2146:  In position space

2147:  \be\label{Air0}

2148:  A_i(r,\bar x)=A_i^{(0)}(\bar x)+r{E}_i^{(0)}(\bar x)+...

2149:  \ee

2150:  where

2151:  \be

2152: f_i(\bar x)=E_i^{(0)}(\bar x)=\int d^3\bar y {A_j^{(0)}(\bar y)\over |\bar x-\bar y|^4}\left(\delta_{ij}-{(\bar x_i-\bar y_i)(\bar x_j-\bar y_j)\over |\bar x-\bar y|^2}\right)

2153: \ee

2154:  

2155:  To find regular solutions, we expand the general solution (\ref{sol})

2156:  \bea\label{regular}

2157:  %{\cal A}&=&{1\over2}\left({\cal A}_0+{1\over|\bar p|}{\cal A}_1\right)e^{|\bar p|r}+{1\over2}\left({\cal A}_0-{1\over|\bar p|}{\cal A}_1\right)e^{-|\bar p|r}\nonumber\\

2158:  {A}_i&=&{1\over2}\left({A}_i^{(0)}+{1\over|\bar p|}{A}_i^{(1)}\right)e^{|\bar p|r}+{1\over2}\left({A}_i^{(0)}-{1\over|\bar p|}{A}_i^{(1)}\right)e^{-|\bar p|r}\nonumber\\

2159:  &&+i{\bar p_i\over|\bar p|}\left[{1\over2}\left({\cal A}_0+{1\over|\bar p|}{\cal A}_1\right)re^{|\bar p|r}-{1\over2}\left({\cal A}_0-{1\over|\bar p|}{\cal A}_1\right)re^{-|\bar p|r}\right]

2160:  \eea

2161: To remove the divergent terms, it's enough to require 

2162:  \be\label{sol1}

2163:  {A}_i^{(0)}(\bar p)+{1\over|\bar p|}{A}_i^{(1)}(\bar p)=0

2164:  \ee

2165: 

2166: 

2167: 

2168: Next we study the radial gauge $A_r=0$, with bulk equations

2169: \be

2170: \nabla_{\m}F^{\m\n}=0,\quad A_{r}=0

2171: \ee

2172: These reduce to 

2173: \be\label{rgauge}

2174: (\partial_r^2+\bar\partial^2)A_i(r,\bar x)=\partial_i\partial_jA_j(r,\bar x),\quad \partial_r\partial_iA_i=0

2175: \ee

2176: which is simply a rewriting of (\ref{emeq2}) if one replaces ${\cal A}$ by $\partial_jA_j$. Notice this replacement is consistent with (\ref{aux}). The second equation in (\ref{rgauge}) imposes the Gauss law $\partial^iE_i=0$. 

2177: 

2178: In summary, we found regularity requires in either gauge 

2179: \be\label{Af}

2180: \Pi_{ij}A_j(p)+{1\over{|p|}}f_i(p)=0

2181: \ee

2182: in terms of the boundary fields $A_i$ and $f_i$. In momentum space, we denote the transverse part of the boundary field

2183: \be\label{transverse}

2184: A_i^{\tiny{\mbox{T}}}(p)=\Pi_{ij}A_j(p)\,.

2185: \ee

2186: The regularity condition is then

2187: \be\label{tAf}

2188: A_i^{\tiny{\mbox{T}}}(p)+{1\over{|p|}}f_i(p)=0

2189: \ee

2190: 

2191: \subsection{Regularity without gauge fixing}

2192: We take advantage of the conformal invariance of the Maxwell equations. In Poincare coordinates they take the form

2193: \be

2194: \nabla_{\m}F^{\m\n}=0~,

2195: \ee

2196: where the metric is the flat Euclidean metric. These reduce to

2197: \bea\label{rgauge}

2198: \left(\partial_r^2+\Box\right)A_i(r,\bar x)&=&\pa_i\pa^jA_j+\pa_i\pa_rA_r\nn

2199: \Box A_r&=&\pa_r\pa^jA_j~,

2200: \eea

2201: where $\Box=\pa^j\pa_j$. The latter is the conservation equation for the electric field $E_i=F_{ri}$, $\pa^iE_i=0$.

2202: 

2203: To separate the different degrees of freedom, we decompose the gauge field in the following way

2204: \bea\label{varsep}

2205: A_i&=&\pa_i\vf+A_i^{\tiny{\mbox{T}}}\nn

2206: \Box\vf&=&\pa^iA_i

2207: \eea

2208: Clearly, $A^{\tiny{\mbox{T}}}$ is the transverse part of the gauge field $\pa^iA_i^{\tiny{\mbox{T}}}=0$,

2209: which carries the physical polarizations. Indeed, $A_i^{\tiny{\mbox{T}}}=A_i-\Box^{-1}\pa_i\pa^jA_j$

2210: is invariant under gauge transformations. We find

2211: \be

2212: E_i=\pa_rA_i^{\tiny{\mbox{T}}}~.

2213: \ee

2214: In momentum space, we have

2215: \be

2216: A_i^{\tiny{\mbox{T}}}(r,p)=\Pi_{ij}A_j(r,p)\,.

2217: \ee

2218: 

2219: 

2220: The longitudinal, pure gauge part of the gauge field is described by $\vf$. In momentum space, $\vf=-{i\over p^2}\,p^jA_j$.

2221: 

2222: Fill the above decomposition into the equations of motion. We get

2223: \be\label{Ar}

2224: A_r=\pa_r \vf~,

2225: \ee

2226: and so $A_r$ is not determined, since $\vf$ isn't. The remaining equation of motion reads

2227: \be

2228: \left(\pa_r^2+\Box\right)A_i^{\tiny{\mbox{T}}}=0~.

2229: \ee

2230: This is solved by

2231: \be

2232: A_i^{\tiny{\mbox{T}}}(r,\bar p)=A_{(0)i}^{\tiny{\mbox{T}}}\cosh(|\bar p|r)+{1\over|\bar p|}\,A_{i(1)}^{\tiny{\mbox{T}}}\sinh(|\bar p|r)~.

2233: \ee

2234: We recognize this as the homogeneous piece of the solution in (\ref{sol}), as the other piece there is pure gauge. 

2235: 

2236: 

2237: We will now impose regularity of the solution. Rewrite the solution as

2238: \be\label{solu}

2239: A_i^{\tiny{\mbox{T}}}={1\over2}\left(A_{i(0)}^{\tiny{\mbox{T}}}+{1\over|\bar p|}\,A_{i(1)}^{\tiny{\mbox{T}}}\right)e^{|\bar p|r}+{1\over2}\left(A_{i(0)}^{\tiny{\mbox{T}}} -{1\over|\bar p|}\,A_{i(1)}^{\tiny{\mbox{T}}}\right)e^{-|\bar p|r}~.

2240: \ee

2241: As before, regularity at $r\rightarrow\infty$ requires

2242: \bea\label{sol11}

2243: A_{i(0)}^{\tiny{\mbox{T}}}(\bar p)+{1\over|\bar p|}\,A_{i(1)}^{\tiny{\mbox{T}}}(\bar p)&=&0~.

2244: \eea

2245: We can rewrite this in terms of the full gauge field as:

2246: \bea\label{regularity1}

2247: \Pi_{ij}\left(A_{(0)j}+{1\over|p|}\,A_{(1)j}\right)&=&0\nn

2248: A_{(0)i}+{1\over|p|}\,A_{(1)i}&=&ip_i\left(\vf_{(0)}+{1\over|p|}\,\vf_{(1)}\right)~,

2249: \eea

2250: where we expanded $\vf$ in the usual way. Hence, regularity relates $A_{(0)}$ and $A_{(1)}$ up to an arbitrary gauge transformation.

2251: 

2252: The self-duality boundary condition \eq{sdbc} can be rewritten as

2253: \be

2254: A_{(1)i}^{\tiny{\mbox{T}}}=i\ve\,\e_{ijk}p_jA_{(0)k}

2255: \ee

2256: which is again rewritten as

2257: \be

2258: \Pi_{ij}\left(A_{(1)j}-i\ve\,\e_{jkl}p_kA_{(0)l}\right)=0~.

2259: \ee

2260: Combining regularity \eq{regularity1} and self-duality, we get

2261: \be

2262: \Pi_{ij}\left(A_{(0)j}+{i\ve\over|p|}\,\e_{jkl}p_kA_{(0)l}\right)=0~.

2263: \ee

2264: Again, the combination between brackets is set to zero up to a gauge transformation.

2265: 

2266: 

2267: Imposing the gauge $A_r=0$ sets $\pa_r\pa^jA_j=0$, which at lowest order simply sets the longitudinal part of $A_{(1)}$ to zero, $\vf_{(1)}=0$. The longitudinal part of $A_{(0)}$ is left undetermined. This corresponds to the residual gauge symmetry of \eq{rgauge} corresponding to $r$-independent gauge transformations. These act purely on the boundary value of the gauge field. In that case, the gauge invariant form of the regularity condition is

2268: \be

2269: A_{(0)i}(p)+{1\over|p|}\,A_{(1)i}=ip_i\vf_{(0)}~.

2270: \ee

2271: Combined with self-duality thus gives

2272: \be

2273: A_{(0)i}+{i\ve\over|p|}\,\e_{ijk}p_jA_{(0)k}=ip_i\vf_{(0)}~.

2274: \ee

2275: 

2276: \subsection{Some used identities}\label{ids}

2277: 

2278: 

2279: 

2280: \bea\label{id4}

2281: \Pi^2&=&\Pi\nn

2282: *\dd F&=&*\dd*\dd A=-t\Box\Pi A\nn

2283: \e_{ikm}\e_{jlm}\pa_k\pa_l&=&t\,\Box\Pi_{ij}\nn

2284: \Pi*\dd&=&*\dd\Pi=*\dd\nn

2285: (*\dd*\dd)_{ij}&=&-M^2\Pi_{ij}=-t\Box\Pi A\nn

2286: *\dd(M\Pi)^{-1}*\dd(M\Pi)^{-1}*\dd&=&-*\dd\nn

2287: \left(\Pi-{\th\over M}\,*\dd\right)^{-1}&=&{\Pi+{\th\over M}\,*\dd\over1+\th^2}~.

2288: \eea

2289: 

2290: Using regularity, we get

2291: \be

2292: \Pi_{ij}v_j={it\over M^2}\,\e_{ijk}p_jf_k

2293: \ee

2294: so

2295: \be

2296: v_if_i=t\,A_iF_i~.

2297: \ee

2298: 

2299: 

2300: 

2301: 

2302: 

2303: 

2304: 

2305: \section{Solution of the boundary self-dual equation}\label{solutionSD}

2306: 

2307: 

2308: 

2309: 

2310: 

2311: \subsection{General regular solution}

2312: 

2313: 

2314: Let us consider the general action \eq{generalaction}. We rewrite its equations of motion \eq{generaleom}, combined with the regularity condition \eq{gaugeinvreg}:

2315: \be\label{reg}

2316: f=-M\,\Pi A~,

2317: \ee

2318: and obtain equations for regular solutions

2319: \bea\label{regeom}

2320: (\a-\m M)\,F+(1-\g)M\,\Pi\cdot A+\d\,*\dd F&=&\ti J\nn

2321: -\b M\,A+(\g-\e M)\,F+\m\,*\dd F&=&J~.

2322: \eea

2323: 

2324: We now compute the on-shell action using regularity:

2325: \bea\label{osefa}

2326: S_{\sm{on-shell}}&=&\int\dd^3x\left(\left(\half\a-\m M\right)AF-\half\b Mv\Pi A+(\half\e M^2-M\g+\half M)\,A\Pi A+\half\,\d F^2\right.\nn

2327: &-&\left.A(\ti J-M\Pi{\cal J})\right)~.

2328: \eea

2329: We can partial integrate to find

2330: \bea

2331: -{1\over2}\b\int d^3x\, v_iM\Pi_{ij}A_j={t\over2}\b\int d^3x\,A_iF_i~.

2332: \eea

2333: and also $\int F^2=t\int M^2A\Pi A$. The effective action is finally

2334: \bea\label{osefa1}

2335: S_{\sm{on-shell}}&=&\int\dd^3x\left(\left(\half\a-\m M+\half\b\right)AF+\left(\half\e M^2-M\g+\half M+\half\,t\d\,M^2\right)\,A\Pi A\right.\nn

2336: %&+&\left.\half\,\d F^2

2337: &-&\left.A(\ti J-M\Pi{\cal J})\right)~.

2338: \eea

2339: This expression, although not yet entirely explicit because it still depends on $A$, and was used in the main text.

2340: 

2341: We now solve the equations of motion in general, for regular Euclidean bulk solutions. Using regularity in the equations of motion \eq{generaleom}, after some algebra we get:

2342: \bea

2343: -M(\b+\m tM)\Pi A+(\g-\e M)\,F&=&*\dd J\nn

2344: -M(\g-1+\d tM)\Pi A+(\a-\m M)\,F&=&*\dd\ti J~.

2345: \eea

2346: This has now become purely a system of equations for $A$. Since there is but one independent equation for $A$, both equations

2347: have to be consistent. Without loss of generality, let us take this to be the first equation above, which we rewrite as

2348: \be\label{genmassive}

2349: -a \Pi A+bF=J~,

2350: \ee

2351: After some algebra, we can rewrite this as a massive equation for $A$, as we did before:

2352: \be\label{generalmassive}

2353: \left(a^2-t\,b^2M^2\right)\Pi A=j~,

2354: \ee

2355: where we defined a new conserved current

2356: \be

2357: j=-b*\dd J-a\,J~.

2358: \ee

2359: The on-shell effective action can now be written entirely in terms of $j$.

2360: 

2361: 

2362: \subsection{General topologically massive solution}

2363: 

2364: In this appendix we explicitly solve the equations of motion of the massive topological boundary theory.

2365: 

2366: We showed in Appendix \ref{regularityapp} that the self-duality solution combined with regularity give the following equation in momentum space:

2367: \be

2368: A_i^{\mbox{\tiny{T}}}(p)=-{i\ve\over|p|}\,\e_{ijk}p_jA_k^{\mbox{\tiny{T}}}(p)~.

2369: \ee

2370: The solution is expanded in a basis of polarization vectors $p_i$, $p_i^*$, $k_i$, where

2371: $k\cdot p=k\cdot p^*=0$. Choosing coordinates $p_i=(p_0,p_1,p_2)$, we have $p_i^*=(-p_0,p_1,p_2)$, and we can choose for example $k_i=(0,-p_1,p_2)$. Expanding

2372: \be\label{Asolution}

2373: A_i^{\mbox{\tiny{T}}}(p)={p_i^*\over|p|}\,f(p)+{k_i\over|p|}\,g(p)~,

2374: \ee

2375: we find the constraints

2376: \bea\label{B3}

2377: p\cdot p^*&=&0\nn

2378: g(p)&=&-{4i\ve p_1p_2\over\sqrt{2}p_0^2}\,{\mbox{sgn}}(p_0)\,f(p)~.

2379: \eea

2380: Imposing the massive boundary condition \eq{dtbc} simply adds to this the constraint $p^2=m^2$ to the above solution.

2381: 

2382: It will be useful to have the general solution of the self-dual theory

2383: \be\label{AeA}

2384: A_i={1\over m}\,\e_{ijk}\pa_jA_k~.

2385: \ee

2386: in position space. First of all, \eq{AeA} automatically implies the gauge fixing condition 

2387: \be\label{gf}

2388: \pa^iA_i=0~.

2389: \ee

2390: We can rewrite \eq{AeA} as follows:

2391: \be

2392: \left(\Box+m^2\right)A_i=0~,

2393: \ee

2394: where $\Box$ is with respect to the flat Euclidean 3-dimensional metric. The general solution of this is best written in terms of Bessel functions of the first kind:

2395: \be

2396: A_i(R,\Omega_2)={1\over\sqrt{R}}\sum_{lm}\left(a_{i(1)lm}\,J_{l+\half}(mr)+a_{i(2)lm}\,J_{-l-\half}(mr)\right) Y_{lm}(\Omega_2)

2397: \ee

2398: in polar coordinates $(R,\Omega_2)$, for some constants $a_{i(1)}$ and $a_{i(2)}$. These constants are however not all arbitrary. Indeed, it is easy to show that the theory has a single massive degree of freedom, described by a {\it single} pair of polarizations $a_{i(1)}$, $a_{i(2)}$. One of the three $A_i$'s is solved for from \eq{AeA}. The remaining one is solved from \eq{gf}.

2399: 

2400: Solving the equation in momentum space as in \eq{Asolution}, we find

2401: \bea

2402: p\cdot p^*&=&0\nn

2403: g(p)&=&{4\sqrt{2}i\over m}\,{\mbox{sgn}}(p_0)\,p_1p_2\,f(p)~.

2404: \eea

2405: 

2406: \subsection{Boundary terms for the self-dual problem}\label{sdactions}

2407: 

2408: As remarked in the main text, because of the invariant of the self-dual solutions, self-dual boundary conditions do not lead to unique boundary terms. We give here two other actions that lead to the same boundary problem:

2409: \bea\label{moresdac}

2410: S_{\sm{bdy}}[A,f]&=&\int\dd^3x\left(Af+{\ve\over2m}\,F^2-{1\over m}\,fF\right)\nn

2411: S_{\sm{bdy}}[A,f]&=&\int\dd^3x\left(Af+\half\,\ve\,AF-{1\over m}\,fF\right)~.

2412: \eea

2413: It would be interesting to check that these are connected by an $SO(1,1)$ transformation.

2414: 

2415: We finally also give the values of the constants in \eq{generaleom} and \eq{generalmassiveeq}:

2416: \bea\label{a123}

2417: a_1&=&\b\d-(\g-1)\m\nn

2418: a_2&=&\g\d-\a\m\nn

2419: a_3&=&\d\e-\m^2\nn

2420: b_1&=&\b\m-(\g-1)\e\nn

2421: b_2&=&\g\m-\a\e~,

2422: \eea

2423: and

2424: \bea\label{ab}

2425: a&=&{a_3^2\over a_1b_2-a_2b_1}\nn

2426: b&=&{a_3(a_1-b_2)\over a_1b_2-a_2b_1}~,

2427: \eea

2428: both in Euclidean signature.

2429: 

2430: \subsection{Near-extremal solutions}\label{nearextremal}

2431: 

2432: A particularly interesting boundary condition is the generalization of \eq{sdbc} to non-extremal solutions where the electric and magnetic field are still aligned:

2433: \be\label{flF}

2434: f=\l\,F~.

2435: \ee

2436: This boundary condition preserves conformal invariance. Combining it with the massive boundary condition \eq{dtbc}, we get the following action:

2437: \be

2438: S_{\sm{bulk}}^{\sm{on-shell}}=\int\dd^3x\left(\g\,Af+\half\,\ve\,AF+{\e\over2}\,f^2\right)~,

2439: \ee

2440: The parameter $\l$ and the mass are given by:

2441: \bea

2442: \l&=&{1\over1-\g}\nn

2443: m&=&-{\g\over\e}~.

2444: \eea

2445: Near extremality, $\l\simeq\ve$, we get that $\g$ is small. $m$ is kept finite by making $\e$ small as well.

2446: 

2447: 

2448: \section{Holographic renormalization of abelian gauge fields}\label{holren}

2449: 

2450: The total Euclidean action is:

2451: 

2452: \be\label{Euclaction}

2453: S=\int\dd^4x\sqrt{g}\,\left(-{R-2\L\over16\pi G_N}+{1\over4}F_{\m\n}F^{\m\n}\right)-{1\over8\pi G_N}\int\dd^3x\sqrt{\g}\,K+S_{\sm{bdy}}[A]

2454: \ee

2455: where $\L=-3/\ell^2$, $K_{ij}=\half\pa_r\g_{ij}$, and $\g={\ell^2\over r^2}\,g$ where $g$ is the metric \eq{FG}. The gravity equations of motion are

2456: \bea

2457: G_{\m\n}+\L\,g_{\m\n}&=&8\pi G_N\,T_{\m\n}\nn

2458: T_{\m\n}&=&F_{\m\l}F_\n{}^\l-{1\over4}\,g_{\m\n}\,F_{\l\s}F^{\l\s}~.

2459: \eea

2460: For conformally invariant matter, the stress-energy tensor is traceless. Therefore, it is most convenient to write Einstein's equations as follows:

2461: \be\label{Einsteineq}

2462: R_{\m\n}+{d\over\ell^2}\,g_{\m\n}=8\pi G_N\,T_{\m\n}~.

2463: \ee

2464: 

2465: The Ricci tensor is:

2466: \bea\label{Ricci}

2467: R_{rr}&=&-\half\,\Tr(g^{-1}g'')+{1\over4}\,\Tr(g^{-1}g')^2+{1\over2r}\,\Tr(g^{-1}g')-{d\over r^2}\nn

2468: R_{ri}&=&\half[\nabla^jg_{ij}'-\nabla_i\Tr(g^{-1}g')]\nn

2469: R_{ij}&=&R_{ij}[g]-\half\,g_{ij}''+\half\,(g'g^{-1}g')_{ij}-{1\over4}\,g_{ij}'\,\Tr(g^{-1}g')+{1\over2r}\left(g_{ij}\Tr(g^{-1}g')+(d-1)g'_{ij}\right)-{d\over r^2}\,g_{ij}~,\nn

2470: \eea

2471: where $d$ is the dimension of the boundary. From the fact that the curvature scalar is always constant we get

2472: \be

2473: \Tr(g^{-1}g'')-{d\over r}\,\Tr(g^{-1}g')-{3\over4}\,\Tr(g^{-1}g')^2+{1\over4}(\Tr(g^{-1}g'))^2-R[g]=0~.

2474: \ee

2475: 

2476: As usual in holographic renormalization, we solve Einstein's equations perturbatively in the distance to the boundary:

2477: \be\label{FGexp}

2478: g(r,x)=g_{(0)}(x)+r^2g_{(2)}(x)+r^3g_{(3)}+r^4g_{(4)}+\cdots

2479: \ee

2480: We find \cite{dHSS}

2481: \be

2482: g_{(2)}=-{1\over d-2}\left(\mbox{Ric}[g_{(0)}]-{1\over2(d-1)}\,g_{(0)}\,R[g_{(0)}]\right)=0~,

2483: \ee

2484: where we used the fact that in our applications we have the boundary condition $g_{(0)ij}=\d_{ij}$. At the next order, from the first and third of \eq{Ricci}-\eq{Einsteineq} we find the equation 

2485: \bea

2486: \Tr g_{(3)}&=&0\nn

2487: (d-3)\,g_{(3)}&=&0~,

2488: \eea

2489: which leaves the traceless part of $g_{(3)}$ undetermined, as it should, since at this order we find the holographic stress-energy tensor \cite{dHSS}:

2490: \be

2491: g_{(3)}={16\pi G_N\over3\ell^2}\,\bra T_{ij}\ket~.

2492: \ee

2493: The second equation then gives

2494: \be

2495: \nabla^jg_{(3)ij}={16\pi G_N\over3\ell^2}\,\t_{ri}|_{r=0}~,

2496: \ee

2497: therefore

2498: \be\label{Wardid}

2499: \nabla^j\bra T_{ij}\ket=f^jF_{ij}~.

2500: \ee

2501: Since $f^i=-\bra O^i_{\Delta=2}\ket$, where $O^i_{\Delta=2}$ is an operator of dimension 2, this is in exact 

2502: agreement with the Ward identity derived in field theory (see formula (4.19) of \cite{BFS}). Indeed, one considers 

2503: the effective action

2504: \be

2505: \d S_{\sm{eff}}=\int\dd^3x\sqrt{g_{(0)}}\left(\half\d g^{ij}_{(0)}\bra T_{ij}\ket+\bra O^i_{\Delta=2}\ket\d 

2506: A_i\right)~.

2507: \ee

2508: Invariance of the action under diffeomorphism then implies \eq{Wardid}. As remarked earlier, in this paper, instead of regarding $O^i_{\Delta=2}$ as an operator, we will regard it as a conserved current.

2509: 

2510: At the next order, we find

2511: \bea

2512: g_{(4)}&=&-{4\pi G_N\over\ell^2}\left(\t_{ij}-\half\,\d_{ij}\,\Tr\,\t\right)|_{r=0}\nn

2513: &=&-{4\pi G_N\over\ell^2}\left(f_if_j-F_iF_j-{1\over4}\,g_{ij}(f^2-F^2)\right)~.

2514: \eea

2515: 

2516: Regularity of the matter solution amounted to a relation between $f$ and $A$. When we include gravity, we have to 

2517: demand regularity of the coupled gravity-Maxwell system. This will in particular involve a special choice of the 

2518: boundary stress-energy tensor, which is left undetermined by \eq{Wardid}. We will not pursue this further here, but 

2519: simply assume that such a choice indeed exists and is enough to make the solutions well-behaved at $r=\infty$. 

2520: 

2521: We now consider the effect of back-reaction on the asymptotic expansion of the on-shell action, which determines the divergent part of the action. The matter part of the action has  been analyzed in the main text. Including 

2522: back-reaction does not change this, as the on-shell action was shown to be a total derivative. For the gravitational 

2523: part of the action we can now use the formula in \cite{dHSS}

2524: \be\label{regaction}

2525: S=\int_{\r\geq\e}\dd\r\,{3\over\r^{5/2}}\,\sqrt{\det g(\r,x)}+{1\over\r^{3/2}}\left( -6\sqrt{\det g(\r,x)}+4\r\pa_\r\sqrt{\det g(\r,x)}\right)\,

2526: \ee

2527: where $\r=r^2$. This is the regulated part of the action. Now it is easy to check that the contribution from $\r=\e$ is

2528: \be

2529: S=-4\e^{-3/2}~.

2530: \ee

2531: This is the leading term near the conformal vacuum. As usual, this is cancelled by adding a covariant counterterm 

2532: $2(1-d)/\e^{d/2}$. 

2533: 

2534: From \eq{regaction} we also see that $g_{(4)}$ and any higher order pieces only contribute terms that do not contribute to 

2535: two-point functions. On the other hand, there is a potential contribution from $g_{(3)}$, which contains the 

2536: stress-energy tensor. This gives the usual coupling to the metric \eq{Wardid}

2537: \be

2538: \half\int\dd^3x\sqrt{g_{(0)}}\,g^{ij}_{(0)}T_{ij}~,

2539: \ee

2540: but since with our choice of $g_{(0)}$ the boundary stress-energy tensor is traceless, this vanishes. Therefore, 

2541: only the contribution from the matter part of the action is present in the two-point functions. In fact, 

2542: there will be additional contributions since we will be considering a problem with generalized boundary conditions.

2543: 

2544: \section{Analysis of supersymmetric boundary conditions}\label{susyapp}

2545: 

2546: 

2547: 

2548: The bosonic fields we discuss here are the bosonic fields of the four dimensional gauged $\CN=2$ supergravity \cite{Freedman:1976aw}. The gauge symmetry is $SO(2)=U(1)$ and can be embedded in the $SO(5)^V$ truncation of the gauged maximal $SO(8)$ supergravity \cite{Romans:1983qi} where all scalar fields are projected out.  The full field contents are the graviton, two real gravitini $\psi_{\m}^i,\,i=1,2$ and the abelian gauge field, which form the $\CN=2$ gravity multiplet. The complete action with the two real gravitini combined into $\psi_\m=\psi^1_\m+i\psi^2_\m$ reads (setting $\kappa=1$) \cite{Caldarelli:2003pb}

2549: \bear

2550: S&=&\int\dd^4x\, \sqrt{g}\left(-{R\over4}+{\L\over2}+{1\over4}F_{\m\n}F^{\m\n}+\half \bar\psi_\m\g^{\m\n\rho}D_\n\psi_{\rho}-{1\over2\ell}\bar\psi_\m\g^{\m\n}\psi_\n \right.\nonumber\\

2551: &&\left.+{i\over4}(F^{\m\n}-\half {\rm{Im}}(\bar\psi^\m\psi^\n))\bar\psi_{\rho}\g_{[\m}\g^{\rho\sigma}\g_{\n]}\psi_{\sigma}\right)

2552: \eear

2553: where

2554: \bear

2555:  D_\m&=&\nabla_\m-{i\over\ell}A_\m\nn

2556: \nabla_\m&=&\pa_\m+{1\over4}\lambda_\m{}^{ab}\g_{ab}

2557: \eear

2558: and the spin connection is given by 

2559: \bear

2560: \lambda_{\m ab}&=&\O_{\m ab}-\O_{\m ba}-\O_{ab\m}\nn

2561: \O_{\m\n}{}^a&=&\pa_{[\m}e^a_{\n]}-\half{\rm{Re}}(\bar\psi_\m\g^a\psi_\n)

2562: \eear

2563: Define the super-covariantized field strength 

2564: \be

2565: \hat F_{\m\n}=F_{\m\n}-{\rm Im}(\bar\psi_\m\psi_\n)

2566: \ee

2567: and the super-covariant derivative

2568: \be

2569: {\cal D}_\m=D_\m+{1\over2\ell}\g_\m+{i\over4}\hat F_{ab}\g^{ab}\g_\m

2570: \ee

2571: we can write the supersymmetry transformations as 

2572: \bea\label{susy}

2573: \d e^a_\m&=&{\rm Re}(\bar\epsilon\g^a\psi_\m)\nn

2574: \d A_\m&=&{\rm Im}(\bar\epsilon\psi_{\m})\nn

2575: \d\psi_\m&=&{\cal D}_\m\epsilon

2576: \eea

2577: where $\epsilon=\epsilon^1+i\epsilon^2$. The $U(1)$ gauge transformation is 

2578: \be

2579: \d A_\m=\pa_\m\a,\quad\d\psi_\m=-i\a\psi_\m

2580: \ee

2581: 

2582: To determine the supersymmetric boundary conditions, we take the approach recently studied in \cite{Hollands:2006zu}. First, we look for boundary conditions on the bosonic and fermionic fields which render the symplectic flux through the boundary vanishing. As we showed in Appendix \ref{holren}, we can expand close to the boundary $r=0$

2583: \be

2584: g(r,x)=\d+r^3g_{(3)}+r^4g_{(4)}+\cdots

2585: \ee

2586: which gives the boundary condition on metric, esp. $\d g=O(r^3)$. The gauge field contributes to the symplectic current

2587: \bea

2588: *\o_r[A]=-\d_1A^\m \d_2(F_{r\m}+{i\over2}\bar\psi_{\rho}\g_{[r}\g^{\rho\sigma}\g_{\m]}\psi_{\sigma})+\d_2A^\m \d_1(F_{r\m}+{i\over2}\bar\psi_{\rho}\g_{[r}\g^{\rho\sigma}\g_{\m]}\psi_{\sigma})

2589: \eea

2590: Notice that the pure gauge field contribution vanishes provided the matrices in \eq{sym} are symmetric. The fermion bilinear terms contribute to the symplectic flux together with those involving only the gravitini, which we now discuss.

2591: 

2592:  Compared to the minimal $\CN=1$ theory, the $\CN=2$ action contains in addition coupling to the gravi-photon field and extra fermion interactions. However, it's straightforward to show that these couplings do not contribute to the gravitini symplectic current. Hence we find the symplectic current from the spin-2 and 3/2 fields \cite{Hollands:2006zu}

2593:  \bea

2594: *\o^r[g,\psi] &=& r^3P^{ijklm r} 

2595: [\delta_1 g_{ij} \delta_2 \lambda_{k\, lm} - 

2596: (1 \leftrightarrow 2)]

2597: \nonumber\\

2598: &+& r^{-2}\epsilon^{rijk} [\delta_1 \ti{\bar \psi_i}

2599: \g_5 \ti\g_j \d_2 \ti\psi_k + \half \d_1 g_j{}^m 

2600: \ti{\bar \psi_i} \g_5 \ti\g_m \d_2 \ti\psi_k

2601: -(1 \leftrightarrow 2)] \nn

2602: &-&r{i\over2}g^{rr}[\d_1A^m \d_2(\ti{\bar \psi_{\rho}}\ti\g_{[r}\ti\g^{\rho\sigma}\ti\g_{m]}\ti\psi_{\sigma})-(1 \leftrightarrow 2)]

2603:  \eea

2604:  where

2605:  \bea

2606:  P^{\alpha\beta\gamma\tau\sigma\mu} = -\frac{1}{2}

2607: g^{\alpha\beta}g^{\gamma\tau}g^{\mu\sigma}

2608: + \frac{1}{2}  

2609: g^{\alpha\mu}g^{\gamma\tau}g^{\beta\sigma}

2610: + \frac{1}{2}  

2611: g^{\alpha\gamma}g^{\beta\tau}g^{\mu\sigma}~.

2612:  \eea

2613: The symplectic from is written in terms of rescaled metric $g_{ij}$ and fermions $\ti\psi_k=r^{1/2}\psi_k$. Now it can be seen that the graviton contribution to the symplectic flux vanishes 

2614: \be

2615: \int_{r=0}\o[g]=0

2616: \ee

2617: where the dual symplectic current

2618: \be

2619: \o[g]=\epsilon_{ijk\,r}*\o^r\,\dd x^i\wedge\dd x^j\wedge\dd x^k=O(r^2)~.

2620: \ee 

2621: In the last equation we used $\d g=r^3\d g_{(3)}$. We also see that the gravitini contribution 

2622: \be\label{symptini}

2623: \int_{r=0}\o[\psi]=\int_{r=0}\left({1\over r^2}{\delta_1 \ti{\bar \psi_{i}}} \ti\g_j \g_5 \d_2 \ti\psi_{k} -{i\over2r}\ti\epsilon_{ijk}\d_1A^m \d_2(\ti{\bar \psi_{\rho}}\ti\g_{[r}\ti\g^{\rho\sigma}\ti\g_{m]}\ti\psi_{\sigma})\right)\, \dd x^i \wedge \dd x^j \wedge \dd x^k

2624: \ee

2625: is finite but non-vanishing if $\ti\psi_\m=O(r)$.  In order that the symplectic flux vanishes, we need $\ti\psi_i=O(r^2)$ and $\ti\psi_r=O(r)$. In terms of $\psi_\m$ we impose \be\label{bcvitini}\psi_r=O(r^{1/2}),\quad \psi_i=O(r^{3/2})~.\ee

2626: 

2627: Having found the boundary conditions consistent with quantization, it's interesting to see whether supersymmetry is preserved asymptotically. This is equivalent to that supersymmetry transformations keep the symplectic current invariant on the boundary.  Plugging \eq{bcvitini} into \eq{susy} leads to

2628: \be

2629:  \d_{\epsilon} A_r=O(r^{1/2}),\quad \d_{\epsilon} A_i=O(r^{3/2})~.

2630: \ee

2631: From this we see that $\d_{\epsilon} A_i$ contributes a vanishing term to the gauge field symplectic current. Evaluating the transformation of $g_{\m\n}$ is slightly more complicated, but one can show \cite{Hollands:2006zu} that up to an infinitesimal diffeomorphism the susy variation takes the form $\d_{\epsilon} g=O(r^2)$. The infinitesimal diffeomorphism $\pounds_{V_{\epsilon}}$ changes $g_{\m\n}$ at order $O(r)$ and simply signals a departure from the Gaussian normal gauge. In practice, one can take the supersymmetry generator to be $s_{\epsilon}-\!\pounds_V$. It now remains to show that the $\psi_\m$ boundary conditions are also preserved by supersymmetry. 

2632: 

2633: For convenience of notation we define the operator ${\cal D}_0={\cal D}|_{A=0}$. Using the result of Appendix \ref{holren} the stress tensor $T_{\m\n}=O(r^2)$, one can show following \cite{Hollands:2006zu} that for $\eta$ a ${\rm AdS_4}$ killing spinor 

2634: \be

2635: \d_{\eta}^0\psi_\m={\cal D}_{0\,\m}\eta=O(r^{3/2})~.

2636: \ee

2637: What remains is to find the behavior of 

2638: \be

2639: \d_{\eta}^A\psi_\m=({\cal D}_\m-{\cal D}_{0\,\m})\eta=-\left({i\over\ell}A_\m-{i\over4}\g^{\a\b}\g_\m F_{\a\b}\right)\eta~.

2640: \ee

2641: For generic values of $A$ and $F$, the first term is order $O(r^0)$ while the second is $O(r)$. Moreover since $A_\m$ is a $U(1)$ gauge field, both $A_\m$ and $F_{\a\b}$ are just numbers and the two terms are linearly independent due to the $\g$-matrix.  Recall $\ti\psi_\m=r^{1/2}\psi_\m$ and \eq{symptini} we find $A_\m=O(r)$ in order that the symplectic current is finite at the boundary. This suggests the only supersymmetric boundary condition for the gauge field is the Dirichlet boundary condition with 

2642: \be

2643: A_\m=0~.

2644: \ee

2645: Neglecting the radial component this is then $A_i=0$.

2646: 

2647: 

2648: 

2649: \begin{thebibliography}{1}

2650: 

2651: \bibitem{SW}  N.~Seiberg and E.~Witten,

2652:   ``Electric - magnetic duality, monopole condensation, and confinement in N=2

2653:   supersymmetric Yang-Mills theory,''

2654:   Nucl.\ Phys.\ B {\bf 426} (1994) 19

2655:   [Erratum-ibid.\ B {\bf 430} (1994) 485]

2656:   [arXiv:hep-th/9407087].

2657:   %%CITATION = HEP-TH 9407087;%%

2658: 

2659: \bibitem{VW}  C.~Vafa and E.~Witten,

2660:   ``A Strong coupling test of S duality,''

2661:   Nucl.\ Phys.\ B {\bf 431} (1994) 3

2662:   [arXiv:hep-th/9408074].

2663:   %%CITATION = HEP-TH 9408074;%%

2664: 

2665: \bibitem{green}  M.~B.~Green,

2666:   ``Interconnections between type II superstrings, M theory and N = 4

2667:   Yang-Mills,''

2668:   arXiv:hep-th/9903124.

2669:   %%CITATION = HEP-TH 9903124;%%

2670: 

2671: \bibitem{KW}

2672: A.~Kapustin and E.~Witten,

2673:   ``Electric-magnetic duality and the geometric Langlands program,''

2674:   arXiv:hep-th/0604151.

2675:   %%CITATION = HEP-TH 0604151;%%

2676: 

2677: \bibitem{CJ}

2678:   E.~Cremmer and B.~Julia,

2679:   ``The N=8 Supergravity Theory. 1. The Lagrangian,''

2680:   Phys.\ Lett.\ B {\bf 80} (1978) 48.

2681:   %%CITATION = PHLTA,B80,48;%%

2682:  ``The SO(8) Supergravity,''

2683:   Nucl.\ Phys.\ B {\bf 159} (1979) 141.

2684:   %%CITATION = NUPHA,B159,141;%%

2685: 

2686: \bibitem{HT}

2687:   M.~Henneaux and C.~Teitelboim,

2688:   ``Duality in linearized gravity,''

2689:   Phys.\ Rev.\ D {\bf 71} (2005) 024018

2690:   [arXiv:gr-qc/0408101].

2691:   %%CITATION = GR-QC 0408101;%%

2692: 

2693: \bibitem{hull}

2694:   C.~M.~Hull and P.~K.~Townsend,

2695:   ``Unity of superstring dualities,''

2696:   Nucl.\ Phys.\ B {\bf 438} (1995) 109

2697:   [arXiv:hep-th/9410167].

2698:   %%CITATION = HEP-TH 9410167;%%

2699: 

2700: \bibitem{Wittensl2z}

2701:   E.~Witten,

2702:    ``SL(2,Z) action on three-dimensional conformal field theories with Abelian

2703:   symmetry,''

2704:   arXiv:hep-th/0307041.

2705:   %%CITATION = HEP-TH 0307041;%%

2706: 

2707: \bibitem{BD1} C.~P.~Burgess and B.~P.~Dolan,

2708:   ``Particle-vortex duality and the modular group: Applications to the  quantum

2709:   Hall effect and other 2-D systems,''

2710:   arXiv:hep-th/0010246.

2711:   %%CITATION = HEP-TH 0010246;%%

2712: 

2713: \bibitem{BD2}

2714:   C.~P.~Burgess and B.~P.~Dolan,

2715:   ``The Quantum Hall Effect in Graphene: Emergent Modular Symmetry and the

2716:   Semi-circle Law,''

2717:   arXiv:cond-mat/0612269.

2718:   %%CITATION = COND-MAT 0612269;%%

2719: 

2720: 

2721: \bibitem{KS}

2722:   A.~Kapustin and M.~J.~Strassler,

2723:   ``On mirror symmetry in three dimensional Abelian gauge theories,''

2724:   JHEP {\bf 9904} (1999) 021

2725:   [arXiv:hep-th/9902033].

2726:   %%CITATION = HEP-TH 9902033;%%

2727: 

2728: \bibitem{BKW}

2729:   V.~Borokhov, A.~Kapustin and X.~k.~Wu,

2730:    ``Topological disorder operators in three-dimensional conformal field

2731:   theory,''

2732:   JHEP {\bf 0211} (2002) 049

2733:   [arXiv:hep-th/0206054].

2734:   %%CITATION = HEP-TH 0206054;%%

2735: 

2736: \bibitem{tassos}

2737:   R.~G.~Leigh and A.~C.~Petkou,

2738:   ``SL(2,Z) action on three-dimensional CFTs and holography,''

2739:   JHEP {\bf 0312}, 020 (2003)

2740:   [arXiv:hep-th/0309177].

2741:   %%CITATION = HEP-TH 0309177;%%

2742: 

2743: %\cite{deHaro:2006wy}

2744: \bibitem{sebastassos}

2745:   S.~de Haro and A.~C.~Petkou,

2746:   ``Instantons and conformal holography,''

2747:   arXiv:hep-th/0606276, to appear in {\bf JHEP}

2748:   %%CITATION = HEP-TH 0606276;%% 

2749: 

2750: \bibitem{HH}

2751:   T.~Hertog and G.~T.~Horowitz,

2752:   ``Holographic description of AdS cosmologies,''

2753:   JHEP {\bf 0504} (2005) 005

2754:   [arXiv:hep-th/0503071].

2755:   %%CITATION = HEP-TH 0503071;%%

2756: 

2757: \bibitem{dHPP}

2758:   S.~de Haro, I.~Papadimitriou and A.~C.~Petkou,

2759:   ``Conformally coupled scalars, instantons and vacuum instability in AdS(4),''

2760:   JHEP {\bf 12}, 076 (2006)

2761:  arXiv:hep-th/0611315.

2762:   %%CITATION = HEP-TH 0611315;%%

2763: 

2764: \bibitem{PTvN}

2765:   P.~K.~Townsend, K.~Pilch and P.~van Nieuwenhuizen,

2766:   ``Selfduality In Odd Dimensions,''

2767:   Phys.\ Lett.\  {\bf 136B} (1984) 38

2768:   [Addendum-ibid.\  {\bf 137B} (1984) 443].

2769:   %%CITATION = PHLTA,136B,38;%%

2770: 

2771: 

2772: \bibitem{DJ}

2773:   S.~Deser and R.~Jackiw,

2774:   ``'Selfduality' Of Topologically Massive Gauge Theories,''

2775:   Phys.\ Lett.\ B {\bf 139} (1984) 371.

2776:   %%CITATION = PHLTA,B139,371;%%

2777: 

2778: \bibitem{dHSS}

2779:   S.~de Haro, S.~N.~Solodukhin and K.~Skenderis,

2780:    ``Holographic reconstruction of spacetime and renormalization in the  AdS/CFT

2781:   correspondence,''

2782:   Commun.\ Math.\ Phys.\  {\bf 217} (2001) 595

2783:   [arXiv:hep-th/0002230].

2784:   %%CITATION = HEP-TH 0002230;%%

2785: 

2786: 

2787: \bibitem{HUY}

2788:   H.~U.~Yee,

2789:   ``A note on AdS/CFT dual of SL(2,Z) action on 3D conformal field theories

2790:   with U(1) symmetry,''

2791:   Phys.\ Lett.\ B {\bf 598}, 139 (2004)

2792:   [arXiv:hep-th/0402115].

2793:   %%CITATION = HEP-TH 0402115;%%

2794: 

2795: \bibitem{KOO}

2796: T.~Kitao, K.~Ohta and N.~Ohta,

2797:  ``Three-dimensional gauge dynamics from brane configurations with

2798:  (p,q)-fivebrane,''

2799:  Nucl.\ Phys.\ B {\bf 539} (1999) 79

2800:  [arXiv:hep-th/9808111].

2801:   %%CITATION = NUPHA,B539,79;%%

2802: 

2803: 

2804: \bibitem{isham}

2805:   S.~J.~Avis, C.~J.~Isham and D.~Storey,

2806:   ``Quantum Field Theory In Anti-De Sitter Space-Time,''

2807:   Phys.\ Rev.\ D {\bf 18} (1978) 3565.

2808:   %%CITATION = PHRVA,D18,3565;%%

2809: 

2810: \bibitem{bf}

2811:   P.~Breitenlohner and D.~Z.~Freedman,

2812:   ``Positive Energy In Anti-De Sitter Backgrounds And Gauged Extended

2813:   Supergravity,''

2814:   Phys.\ Lett.\ B {\bf 115} (1982) 197.

2815:   %%CITATION = PHLTA,B115,197;%%

2816: 

2817: %\cite{Breitenlohner:1982jf}

2818: \bibitem{Breitenlohner:1982jf}

2819:   P.~Breitenlohner and D.~Z.~Freedman,

2820:   ``Stability In Gauged Extended Supergravity,''

2821:   Annals Phys.\  {\bf 144}, 249 (1982).

2822:   %%CITATION = APNYA,144,249;%%

2823:   

2824: 

2825: \bibitem{klebanovwitten}

2826:   I.~R.~Klebanov and E.~Witten,

2827:   ``AdS/CFT correspondence and symmetry breaking,''

2828:   Nucl.\ Phys.\ B {\bf 556} (1999) 89

2829:   [arXiv:hep-th/9905104].

2830:   %%CITATION = HEP-TH 9905104;%%

2831: 

2832: \bibitem{Dobrev}

2833:   V.~K.~Dobrev,

2834:   ``Intertwining operator realization of the AdS/CFT correspondence,''

2835:   Nucl.\ Phys.\  B {\bf 553} (1999) 559

2836:   [arXiv:hep-th/9812194].

2837:   %%CITATION = NUPHA,B553,559;%%

2838: 

2839: 

2840: \bibitem{HM}

2841:   T.~Hertog and K.~Maeda,

2842:   ``Black holes with scalar hair and asymptotics in N = 8 supergravity,''

2843:   JHEP {\bf 0407} (2004) 051

2844:   [arXiv:hep-th/0404261].

2845:   %%CITATION = HEP-TH 0404261;%%

2846: 

2847: \bibitem{designer}

2848:   T.~Hertog and G.~T.~Horowitz,

2849:   ``Designer gravity and field theory effective potentials,''

2850:   Phys.\ Rev.\ Lett.\  {\bf 94} (2005) 221301

2851:   [arXiv:hep-th/0412169].

2852:   %%CITATION = HEP-TH 0412169;%%

2853: 

2854: \bibitem{wittendt}

2855:   E.~Witten,

2856:   ``Multi-trace operators, boundary conditions, and AdS/CFT correspondence,''

2857:   arXiv:hep-th/0112258.

2858:   %%CITATION = HEP-TH 0112258;%%  

2859: 

2860: 

2861: 

2862: \bibitem{MSS}

2863:   M.~Berkooz, A.~Sever and A.~Shomer,

2864:   ``Double-trace deformations, boundary conditions and spacetime

2865:   singularities,''

2866:   JHEP {\bf 0205} (2002) 034

2867:   [arXiv:hep-th/0112264].

2868:   %%CITATION = HEP-TH 0112264;%%

2869: 

2870: 

2871: \bibitem{don}

2872:   D.~Marolf,

2873:   ``States and boundary terms: Subtleties of Lorentzian AdS/CFT,''

2874:   JHEP {\bf 0505} (2005) 042

2875:   [arXiv:hep-th/0412032].

2876:   %%CITATION = HEP-TH 0412032;%%

2877: 

2878: 

2879: \bibitem{BY}

2880:   J.~D.~Brown and J.~W.~.~York,

2881:    ``Quasilocal energy and conserved charges derived from the gravitational

2882:   action,''

2883:   Phys.\ Rev.\ D {\bf 47} (1993) 1407.

2884:   %%CITATION = PHRVA,D47,1407;%%

2885: 

2886: \bibitem{BFS}

2887:  M.~Bianchi, D.~Z.~Freedman and K.~Skenderis,

2888:   ``Holographic renormalization,''

2889:   Nucl.\ Phys.\ B {\bf 631} (2002) 159

2890:   [arXiv:hep-th/0112119].

2891:   %%CITATION = HEP-TH 0112119;%%

2892: 

2893: \bibitem{vijayper}

2894:   V.~Balasubramanian, P.~Kraus and A.~E.~Lawrence,

2895:   ``Bulk vs. boundary dynamics in anti-de Sitter spacetime,''

2896:   Phys.\ Rev.\ D {\bf 59} (1999) 046003

2897:   [arXiv:hep-th/9805171].

2898:   %%CITATION = HEP-TH 9805171;%%

2899: 

2900: %\cite{Marolf:2006nd}

2901: \bibitem{marolf}

2902:   D.~Marolf and S.~F.~Ross,

2903:   ``Boundary conditions and new dualities: Vector fields in AdS/CFT,''

2904:   arXiv:hep-th/0606113.

2905:   %%CITATION = HEP-TH 0606113;%%

2906: 

2907: 

2908: \bibitem{herzogson}

2909:   C.~P.~Herzog and D.~T.~Son,

2910:   ``Schwinger-Keldysh propagators from AdS/CFT correspondence,''

2911:   JHEP {\bf 0303} (2003) 046

2912:   [arXiv:hep-th/0212072].

2913:   %%CITATION = HEP-TH 0212072;%%

2914: 

2915: \bibitem{maldacena}

2916:   J.~M.~Maldacena,

2917:   ``The large N limit of superconformal field theories and supergravity,''

2918:   Adv.\ Theor.\ Math.\ Phys.\  {\bf 2} (1998) 231

2919:   [Int.\ J.\ Theor.\ Phys.\  {\bf 38} (1999) 1113]

2920:   [arXiv:hep-th/9711200].

2921:   %%CITATION = HEP-TH 9711200;%%

2922: 

2923: \bibitem{unruh}

2924:   W.~G.~Unruh,

2925:   ``Absorption Cross-Section Of Small Black Holes,''

2926:   Phys.\ Rev.\ D {\bf 14} (1976) 3251.

2927:   %%CITATION = PHRVA,D14,3251;%%

2928: 

2929: \bibitem{GK}

2930:   S.~S.~Gubser and I.~R.~Klebanov,

2931:   ``Emission of charged particles from four- and five-dimensional  black

2932:   holes,''

2933:   Nucl.\ Phys.\ B {\bf 482} (1996) 173

2934:   [arXiv:hep-th/9608108].

2935:   %%CITATION = HEP-TH 9608108;%%

2936: 

2937: \bibitem{GKT}

2938:   S.~S.~Gubser, I.~R.~Klebanov and A.~A.~Tseytlin,

2939:   ``String theory and classical absorption by three-branes,''

2940:   Nucl.\ Phys.\ B {\bf 499} (1997) 217

2941:   [arXiv:hep-th/9703040].

2942:   %%CITATION = HEP-TH 9703040;%%

2943: 

2944: \bibitem{DT}

2945:   S.~Deser and C.~Teitelboim,

2946:   ``Duality Transformations Of Abelian And Nonabelian Gauge Fields,''

2947:   Phys.\ Rev.\ D {\bf 13} (1976) 1592.

2948:   %%CITATION = PHRVA,D13,1592;%%

2949: 

2950: \bibitem{dght}

2951:   S.~Deser, A.~Gomberoff, M.~Henneaux and C.~Teitelboim,

2952:    ``Duality, self-duality, sources and charge quantization in abelian  N-form

2953:   theories,''

2954:   Phys.\ Lett.\ B {\bf 400} (1997) 80

2955:   [arXiv:hep-th/9702184].

2956:   %%CITATION = HEP-TH 9702184;%%

2957: 

2958: \bibitem{ss}

2959:   J.~H.~Schwarz and A.~Sen,

2960:   ``Duality symmetric actions,''

2961:   Nucl.\ Phys.\ B {\bf 411} (1994) 35

2962:   [arXiv:hep-th/9304154].

2963:   %%CITATION = HEP-TH 9304154;%%

2964: 

2965: \bibitem{Deser:2004xt}

2966:   S.~Deser and D.~Seminara,

2967:   ``Duality invariance of all free bosonic and fermionic gauge fields,''

2968:   Phys.\ Lett.\ B {\bf 607} (2005) 317

2969:   [arXiv:hep-th/0411169].

2970:   %%CITATION = HEP-TH 0411169;%%

2971: 

2972: \bibitem{deser}

2973:   S.~Deser,

2974:   ``Black hole electromagnetic duality,''

2975:   arXiv:hep-th/9701157.

2976:   %%CITATION = HEP-TH 9701157;%%

2977: 

2978: 

2979: \bibitem{Deser82}

2980: S.~Deser, ``Off-shell electromagnetic duality invariance'',

2981: J. Phys. A: Math. Gen. 15 No 3 (1 March 1982) 1053-105

2982: 

2983: \bibitem{wittenabelian}

2984:   E.~Witten,

2985:   ``On S duality in Abelian gauge theory,''

2986:   Selecta Math.\  {\bf 1} (1995) 383

2987:   [arXiv:hep-th/9505186].

2988:   %%CITATION = HEP-TH 9505186;%%

2989: 

2990: \bibitem{yolanda}

2991:   Y.~Lozano,

2992:   ``S duality in gauge theories as a canonical transformation,''

2993:   Phys.\ Lett.\ B {\bf 364} (1995) 19

2994:   [arXiv:hep-th/9508021].

2995:   %%CITATION = HEP-TH 9508021;%%

2996: 

2997: \bibitem{YK}

2998:   I.~Papadimitriou and K.~Skenderis,

2999:   ``AdS / CFT correspondence and geometry,''

3000:   arXiv:hep-th/0404176.

3001:   %%CITATION = HEP-TH/0404176;%%

3002: 

3003: \bibitem{aharony}

3004:   O.~Aharony, A.~B.~Clark and A.~Karch,

3005:   ``The CFT/AdS correspondence, massive gravitons and a connectivity index

3006:   conjecture,''

3007:   Phys.\ Rev.\ D {\bf 74} (2006) 086006

3008:   [arXiv:hep-th/0608089].

3009:   %%CITATION = HEP-TH 0608089;%%

3010: 

3011:  %\cite{Freedman:1976aw}

3012: \bibitem{Freedman:1976aw}

3013:   D.~Z.~Freedman and A.~Das,

3014:   %``Gauge Internal Symmetry In Extended Supergravity,''

3015:   Nucl.\ Phys.\ B {\bf 120}, 221 (1977).

3016:   %%CITATION = NUPHA,B120,221;%%

3017: 

3018: 

3019:  %\cite{Romans:1983qi}

3020: \bibitem{Romans:1983qi}

3021:   L.~J.~Romans,

3022:   %``The SO(5) Sectors Of Gauged N=8 Supergravity,''

3023:   Phys.\ Lett.\ B {\bf 131}, 83 (1983).

3024:   %%CITATION = PHLTA,B131,83;%%

3025: 

3026:     

3027:   %\cite{Hollands:2006zu}

3028: \bibitem{Hollands:2006zu}

3029:   S.~Hollands and D.~Marolf,

3030:   ``Asymptotic generators of fermionic charges and boundary conditions

3031:   preserving supersymmetry,''

3032:   arXiv:gr-qc/0611044.

3033:   %%CITATION = GR-QC 0611044;%%

3034: 

3035: \bibitem{hagen}

3036:   C.~R.~Hagen,

3037:   ``Gauge Theories In Three-Dimensional Space-Time,''

3038:   Phys.\ Rev.\ D {\bf 36} (1987) 3773.

3039:   %%CITATION = PHRVA,D36,3773;%%

3040: 

3041: \bibitem{Skenderis}

3042:   K.~Skenderis,

3043:   ``Lecture notes on holographic renormalization,''

3044:   Class.\ Quant.\ Grav.\  {\bf 19} (2002) 5849

3045:   [arXiv:hep-th/0209067].

3046:   %%CITATION = CQGRD,19,5849;%%

3047: 

3048: \bibitem{DS}

3049:     S.~Deser and D.~Seminara,

3050:   ``Free spin 2 duality invariance cannot be extended to GR,''

3051:   Phys.\ Rev.\  D {\bf 71} (2005) 081502

3052:   [arXiv:hep-th/0503030].

3053:   %%CITATION = PHRVA,D71,081502;%%

3054: 

3055: \bibitem{schwarz}

3056:   J.~H.~Schwarz,

3057:   ``Superconformal Chern-Simons theories,''

3058:   JHEP {\bf 0411} (2004) 078

3059:   [arXiv:hep-th/0411077].

3060:   %%CITATION = HEP-TH 0411077;%%

3061: 

3062: \bibitem{herzog}

3063:   C.~P.~Herzog, P.~Kovtun, S.~Sachdev and D.~T.~Son,

3064:   ``Quantum critical transport, duality, and M-theory,''

3065:   arXiv:hep-th/0701036.

3066:   %%CITATION = HEP-TH 0701036;%%

3067: 

3068: 

3069:   %\cite{Caldarelli:2003pb}

3070: \bibitem{Caldarelli:2003pb}

3071:   M.~M.~Caldarelli and D.~Klemm,

3072:   %``All supersymmetric solutions of N = 2, D = 4 gauged supergravity,''

3073:   JHEP {\bf 0309}, 019 (2003)

3074:   [arXiv:hep-th/0307022].

3075:   %%CITATION = HEP-TH 0307022;%%

3076: 

3077: 

3078: 

3079: 

3080: 

3081: 

3082: \end{thebibliography}

3083: 

3084: 

3085: \end{document}   

3086: 

3087: 

3088: 

3089: 

3090: 

3091: 

3092: