1: \documentclass[12pt,letterpaper]{article}

2: 

3: \usepackage{epsfig}

4: \usepackage{amsmath}

5: \usepackage[psamsfonts]{amssymb}

6: \usepackage{amsthm}

7: \usepackage{fullpage}

8: \usepackage{indentfirst}

9: 

10: \numberwithin{equation}{section}

11: 

12: % MACROS FROM ANDY 

13: \newcommand{\nid}{\noindent}

14: \newcommand{\for}{\quad \text{for }}

15: \newcommand{\deltabar}{\overline{\delta}}

16: 

17: \newcommand{\R}{\ensuremath{\mathbb R}}

18: \newcommand{\C}{\ensuremath{\mathbb C}}

19: \newcommand{\PP}{\ensuremath{\mathbb P}}

20: \newcommand{\Z}{\ensuremath{\mathbb Z}}

21: \newcommand{\Q}{\ensuremath{\mathbb Q}}

22: \newcommand{\half}{\ensuremath{\frac{1}{2}}}

23: \newcommand{\qtr}{\ensuremath{\frac{1}{4}}}

24: \newcommand{\N}{{\mathcal N}}

25: \newcommand{\dirac}{\!\!\not\!\partial}

26: \newcommand{\Dirac}{\!\!\not\!\!D}

27: \newcommand{\M}{{\mathcal M}}

28: \newcommand{\Mstab}[2]{{\overline{\M_{0,#1,#2}}}}

29: \newcommand{\Mint}{{\M_{{\rm int}}}}

30: \newcommand{\Mintgamma}{{\M^\Gamma_{{\rm int}}}}

31: \newcommand{\Mintlambda}{{\M^\Lambda_{{\rm int}}}}

32: \newcommand{\Mintone}{{\M^1_{{\rm int}}}}

33: \newcommand{\Mintgammaprime}{{\M^{\Gamma'}_{{\rm int}}}}

34: \newcommand{\MintK}{{\M^{K}_{{\rm int}}}}

35: \newcommand{\Mlines}{{\M_{{\rm lines}}}}

36: \newcommand{\Mlinesgamma}{{\M^\Gamma_{{\rm lines}}}}

37: \newcommand{\mulines}{{\mu_{{\rm lines}}}}

38: \newcommand{\muint}{{\mu_{{\rm int}}}}

39: 

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

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

42: \newcommand{\IP}[1]{\langle#1\rangle}

43: \newcommand{\dwrt}[1]{\frac{\partial}{\partial#1}}

44: \newcommand{\eps}{\epsilon}

45: \newcommand{\ci}{\oint}

46: 

47: \newcommand{\la}{\label}

48: \newcommand{\ti}[1]{\textit{#1}}

49: 

50: \newcommand{\mfn}[1]{%

51:    \footnote{#1}

52:    \marginpar{\hfill {\sf\thefootnote}}%

53: }

54: 

55: \newcommand{\CPP}{{\C\PP^{3|4}}}

56: 

57: 

58: \newcommand{\prodprime}[2]{{\prod_{#1}^{#2} \!\!\!\!\begin{array}{c}\prime\\ \,\end{array}}}

59: 

60: 

61: 

62: % MACROS FROM LUBOS

63: \def\IZ{{\mathbb Z}}           % \mathfrak - gothic, \mathcal - script

64: \def\IR{{\mathbb R}}

65: \def\IC{{\mathbb C}}

66: \def\A#1{{\cal A}_{[#1]}}        % Amplitude - for example \A{++++---}

67: \def\Beta{B}

68: \def\calD{{\cal D}}

69: \def\calH{{\cal H}}

70: \def\calN{{\cal N}}

71: \def\calO{{\cal O}}

72: \def\Box{\square}

73: \def\IA{{\mathbb{A}}}

74: \def\IB{{\mathbb{B}}}

75: \def\IP{{\mathbb{P}}}

76: 

77: 

78: \def \tilde{\widetilde}

79: \def \tb#1{\left(\begin{array}#1\end{array}\right)}

80: \def \cpt{\mathbb{CP}^{(3|4)}}

81: \def \abs#1{\left|#1\right|}

82: \def \lr{\leftrightarrow}

83: \def \eqref#1{(\ref{#1})}

84: \def \bra#1{\left\langle #1\right\vert}

85: \def \ket#1{\left\vert #1\right\rangle}

86: \def \dslash{{\not\!\partial}}

87: \def \exp{\mbox{exp}}

88: \def \tp{\mbox{tp\,}}

89: \def \ignoruj#1{}

90: \def \eqn#1#2{\begin{equation}#2\label{#1}\end{equation}}

91: \def \ha{{1\over 2}}

92: \def \Tr{\mbox{Tr\,}}

93: \def \tr{\mbox{Tr\,}}

94: \def \ham{\mathcal H}

95: \def \Lag{\mathcal L}

96: \def \gs{g_{\mathrm{string}}}

97: \def \inte{{\mathrm{interaction}}}

98: \def \ls{l_{string}}

99: \def \lpl{l_{\mathrm{Planck}}}

100: \def \action{\mathcal A}

101: \def \Ga{\Gamma}

102: \def \a{\alpha}

103: \def \half{{1\over 2}}

104: \def \s{\sigma}

105: \def \l{\lambda}

106: \def \g{\gamma}

107: \def \d{\delta}

108: \def \de{{\rm d}}

109: \def \e{\epsilon}

110: \def \ad{{\dot a}}

111: \def \bd{{\dot b}}

112: \def \cd{{\dot c}}

113: \def \dd{{\dot d}}

114: \def \CH{\mathfrak {CH}}

115: \def \b{\beta}

116: \def \vt{\vartheta}

117: \def \ttheta{{\tilde \theta}}

118: \def \tSigma{{\tilde \Sigma}}

119: %

120: \def\ignorethis#1{}

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

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

123: \def\ar#1#2{\begin{array}{#1}#2\end{array}}

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

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

126: \def\p{\partial }

127: \def\Rads{R_{\mathrm{AdS}}}

128: \def\AdS{\mathrm{AdS}}

129: 

130: 

131: 

132: %---------------------------------------------------------------------------

133: 

134: 

135: \begin{document}

136: 

137: \bibliographystyle{utphys}

138: 

139: \setcounter{page}{1}

140: \pagestyle{plain}

141: 

142: 

143: \begin{titlepage}

144: 

145: \begin{center}

146: \hfill HUTP-04/A016\\

147: \hfill HEP-UK-0021\\

148: \hfill hep-th/0404085

149: 

150: \vskip 1.5 cm

151: {\huge \bf Equivalence of twistor prescriptions\\

152: {\small~\newline} for super Yang-Mills}

153: \vskip 1.3 cm

154: {\large Sergei Gukov, Lubo\v{s} Motl, and Andrew Neitzke}\\

155: \vskip 0.5 cm

156: {Jefferson Physical Laboratory,

157: Harvard University,\\

158: Cambridge, MA 02138, USA}

159: \vskip 0.3cm

160: {

161: {\tt gukov@feynman.harvard.edu}\\

162: {\tt motl@feynman.harvard.edu}\\

163: {\tt neitzke@fas.harvard.edu}}

164: \end{center}

165: 

166: 

167: \vskip 0.5 cm

168: \begin{abstract}

169: There is evidence that one can compute tree level super Yang-Mills amplitudes

170: using either connected or completely disconnected curves in twistor space.

171: We argue that the two computations 

172: are equivalent, if the integration contours are chosen in a specific way, 

173: by showing that they can both be reduced to the same integral over a moduli space 

174: of singular curves.  

175: We also formulate a class of new ``intermediate'' 

176: prescriptions to calculate the same amplitudes.

177: \end{abstract}

178: 

179: \end{titlepage}

180: 

181: \renewcommand{\baselinestretch}{1.4}

182: \small\normalsize

183: 

184: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

185: 

186: \tableofcontents

187: 

188: \line(1,0){450}

189: 

190: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

191: 

192: \section{Introduction}

193: 

194: Recently in \cite{Witten:2003nn} Witten proposed a new approach to

195: perturbative gauge theories in four dimensions which, among other things,

196: implies remarkable regularities in the perturbative scattering amplitudes

197: of $\N=4$ super Yang-Mills and leads to new ways of computing them.

198: The scattering amplitudes in question depend on the momentum

199: and polarization vectors of the external gluons,

200: and are devilishly difficult to compute using the standard Feynman diagram

201: techniques.

202: For example, even computing a tree level amplitude with

203: 4 external gluons of positive helicity and 3 gluons of negative helicity

204: (such an amplitude will be denoted $\A{++++---}$)

205: requires summing over hundreds of different diagrams!

206: 

207: According to the conjecture of \cite{Witten:2003nn},

208: perturbative $\N=4$ super Yang-Mills theory can be described

209: as a string theory in twistor space $\CPP$.

210: In this reformulation, the Yang-Mills scattering amplitudes

211: are given by certain integrals over moduli spaces

212: of holomorphic curves in $\CPP$, which can be interpreted as D1-brane instantons.

213: More precisely, for a tree level process involving $q$ negative

214: helicity gluons, the amplitude is given by an integral over

215: moduli of curves of total degree $d$, where

216: %

217: \eqn{dviaq}{ d=q-1. }

218: %

219: For example, the simplest non-vanishing amplitude with $q=2$

220: gluons of negative helicity\footnote{We follow the conventions

221: of \cite{Witten:2003nn} where a $n$-gluon scattering amplitude

222: is called MHV if $n\!-\!2$ external gluons have positive helicity,

223: and $\overline{{\rm MHV}}$ (or ``googly'') if $n\!-\!2$ gluons have

224: negative helicity.} --- the so-called maximally helicity

225: violating (MHV) amplitude \cite{Parke:1986gb,Mangano:1988xk} --- can be computed by integrating over

226: the moduli space of degree $1$ curves in $\CPP$ \cite{Witten:2003nn}.

227: 

228: However, when one considers the next simplest case, $q=3$, there is a puzzle.

229: In the prescription of \cite{Witten:2003nn} this amplitude seems to 

230: involve a sum over two distinct contributions:

231: one from an integral over connected degree $2$ curves,

232: and another from an integral over disconnected pairs

233: of degree $1$ curves; see Figure \ref{linesab}.

234: Surprisingly, in the case of $\A{++---}$, it was found

235: that the contribution from connected degree $2$ curves alone

236: gives the full Yang-Mills amplitude,

237: at least up to a multiplicative constant \cite{Roiban:2004vt}.

238: This computation was extended to all googly \cite{Roiban:2004ka}

239: and some non-MHV \cite{Roiban:2004yf} amplitudes,

240: again with the surprising result that connected degree $d$ curves already

241: account for the full Yang-Mills amplitude, without adding any disconnected curves.

242: 

243: \begin{figure}

244: \begin{center}

245: \epsfig{file=linesab.eps,width=130mm}

246: \end{center}

247: \caption{An instanton contribution: {\bf (a)} from a connected

248: curve of degree 2;\quad {\bf (b)} from a pair of degree 1 curves.

249: The dotted line represents a propagator in holomorphic

250: Chern-Simons theory.}

251: \label{linesab}

252: \end{figure}

253: 

254: On the other hand, there is some evidence that these tree level

255: amplitudes can also be computed by considering only the contribution

256: of curves which are ``maximally disconnected,'' namely,

257: they consist of $d$ distinct degree $1$ lines.

258: Since degree 1 curves are associated with MHV amplitudes,

259: this result suggests an alternative method of computing

260: generic tree amplitudes from graphs with MHV vertices \cite{Cachazo:2004kj}.

261: The number $v$ of vertices is determined by the number

262: of gluons with negative helicity; it is actually equal to the degree

263: \eqref{dviaq},

264: %

265: \eqn{vviaq}{ v=q-1. }

266: %

267: This approach leads to a spectacular simplification of the computations.

268: For example, the 7-gluon amplitude $\A{++++---}$ mentioned earlier can be

269: computed using only 8 diagrams with MHV vertices.

270: However, it also leads to a puzzle.

271: 

272: As we just discussed, the evidence so far in the literature

273: suggests that rather than one prescription

274: for Yang-Mills amplitudes there are at least two:

275: one involving connected curves only, another involving maximally disconnected ones.

276: We will refer to these as the ``connected prescription''

277: and the ``disconnected prescription'' respectively.  

278: These different prescriptions have so far not been related directly.

279: In a sense, they seem to have complementary virtues:  the connected prescription

280: expresses the whole amplitude as a single integral, and 

281: from this form it is easier to prove some properties of the amplitude,

282: such as the parity symmetry; on the other hand, the disconnected prescription

283: leads to concrete and immediately useful formulas for the tree level amplitudes.

284: 

285: The purpose of this note is to argue that the connected and disconnected prescriptions

286: are equivalent, at least for an appropriate choice of the integration contours,

287: and to give an \ti{a priori} explanation for this agreement.

288: The explanation is that, in both prescriptions, the integral

289: over the moduli space is localized to poles on a particular submoduli space. 

290: This submoduli space parameterizes configurations of intersecting degree $1$ curves.

291: 

292: Let us illustrate this explanation in

293: the simplest case of degree $2$ curves.

294: We have two different moduli spaces,  $\Mstab{n}{2}$ and $\Mlines$,

295: parameterizing respectively connected degree $2$ curves in $\CPP$

296: and disconnected pairs of lines in $\CPP$,

297: and integrands $\omega_{{\rm conn}}$

298: and $\omega_{{\rm disc}}$ on the two spaces (we will review the construction of these integrands

299: in Section \ref{sec-review}).  Our job is to explain the equality

300: \begin{equation}

301: \int_\Mstab{n}{2} \omega_{{\rm conn}} = \int_\Mlines \omega_{{\rm disc}}.

302: \end{equation}

303: The explanation begins by noting that both

304: $\Mlines$ and $\Mstab{n}{2}$ contain a codimension-one

305: ``degeneration locus'' $\Mint$ parameterizing the moduli of

306: pairs of intersecting lines in $\CPP$.

307: In the case of $\Mlines$ we get such a degenerate configuration

308: just by taking two lines in $\CPP$ which happen to intersect.

309: For $\Mstab{n}{2}$ we get such a degeneration by

310: considering a hyperbola $xy = C$ in the limit $C \to 0$,

311: appropriately embedded in $\CPP$.

312: The crucial point is that both

313: $\omega_{{\rm conn}}$ and $\omega_{{\rm disc}}$

314: turn out to have a simple pole along $\Mint$,

315: and furthermore the residue is the same in both cases.\footnote{We learned of

316: the possibility of such an explanation from Edward Witten.} Therefore, 

317: provided that the integration contours on

318: $\Mlines$ and $\Mstab{n}{2}$ are chosen compatibly

319: (so that they both encircle $\Mint$ and reduce

320: to the same contour along it), the desired agreement follows.

321: 

322: The argument for general degree $d$ proceeds along similar lines.  In the moduli space

323: $\Mstab{n}{d}$ we find a pole where a degree $d$ curve degenerates into two intersecting

324: curves of degrees $d_1$ and $d_2$; the integral over $\Mstab{n}{d}$ localizes to this sublocus;

325: then inside this sublocus there is a pole where one of the two curves degenerates further, and

326: so on until we reduce finally to the moduli space

327: $\Mint$ of connected trees built from degree $1$ curves.

328: On the other hand, the integral over $\Mlines$ also reduces to the same $\Mint$, 

329: because the propagators connecting the different lines have poles when the lines intersect.

330: Furthermore it turns out that the integrands on $\Mint$ coming from the two prescriptions are

331: proportional.  This establishes the agreement between these two prescriptions, again provided

332: that the contours are chosen appropriately, and up to an overall constant which we do not fix.

333: 

334: This iterative argument pays a surprising dividend:  

335: for any $K = 0, \dots, d-1$, we can define an ``intermediate prescription,''

336: in which we integrate over configurations of $K+1$ curves with total degree $d$.

337: We will show that all of these intermediate prescriptions agree with

338: the connected and disconnected prescriptions.

339: They can also be understood diagrammatically:

340: one sums over tree diagrams with $K+1$ vertices,

341: where each vertex is decorated with a degree.

342: In these notations,

343: vertices of degree 1 are the MHV vertices of \cite{Cachazo:2004kj},

344: whereas vertices with $d>1$ could be called ``non-MHV vertices''.

345: These intermediate prescriptions deserve further study.

346: 

347: For other recent work on the twistor string approach to Yang-Mills, see

348: \cite{Roiban:2004vt,Roiban:2004ka,Roiban:2004yf,Berkovits:2004hg,Berkovits:2004tx,Witten:2004cp} 

349: for the connected prescription, 

350: \cite{Cachazo:2004kj,Zhu:2004kr,tree-scalar-mhv} for the disconnected prescription,

351: and \cite{Aganagic:2004yh,Nekrasov:2004js,Neitzke:2004pf} for related

352: topics.

353: 

354: 

355: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

356: 

357: \subsection{Notation and moduli spaces}

358: 

359: We always consider scattering amplitudes of $n$ external gluons

360: associated with the particular trace factor $\tr (T_1 T_2 \ldots T_n)$.

361: 

362: We use a coordinate representation

363: for the super twistor space $\C^{4|4}$.

364: We unify the bosonic and fermionic indices into

365: a superspace index $\IA$ taking values in

366: \eqn{superindex}{\IA\in

367: \{1,\,2,\,3,\,4\, | 1',2',3',4'\}.}

368: The components of all objects with bosonic values of the 

369: superspace index are commuting, while components with fermionic

370: (primed) values of the superspace index are anticommuting.

371: The coordinates on the super twistor space will be denoted by $Z^\IA$,

372: which are related to the coordinates in the literature by

373: \eqn{twistorcomponents}{(Z^1,Z^2,Z^3,Z^4|Z^{1'},Z^{2'},Z^{3'},Z^{4'})

374: =(\lambda^1,\lambda^2,\mu^1,\mu^2 |

375: \psi^{1}, \psi^{2}, \psi^{3}, \psi^{4})

376: \in {\IC^{4|4}}.} 

377: 

378: We will also be considering various moduli spaces of

379: curves in $\CPP$  with marked points.  We use the standard notation 

380: \begin{equation}

381: \M_{0,n,d}(\CPP)

382: \end{equation}

383: for the moduli space of ``genus $0$, $n$-pointed curves of degree $d$ in $\CPP$.''

384: This moduli space has dimension $(4d+n)|(4d+4)$.

385: As in \cite{Witten:2003nn} we realize it as the space of

386: automorphism classes of maps \,$\C\PP^1 \to \CPP$, of degree $d$, with $n$ marked points on $\C\PP^1$.

387: Since the target space is always $\CPP$ in this paper, 

388: sometimes we abuse notation and write simply $\M_{0,n,d}$.

389: 

390: \vspace{3mm}

391: 

392: \begin{figure}

393: \begin{center}

394: \epsfig{file=deg-lines.eps,width=120mm}

395: \end{center}

396: \caption{A curve of degree 2 can degenerate into a pair

397: of intersecting lines.}

398: \label{deg-lines}

399: \end{figure}

400: 

401: We will be interested in integrating 

402: over $\M_{0,n,d}(\CPP)$, so we need to understand the

403: properties of this moduli space.

404: First, $\M_{0,n,d}(\CPP)$ is non-compact, due

405: to certain degenerations that a degree $d$ curve with $n$ marked

406: points can have which are not simply described by a map

407: $\C\PP^1 \to\CPP$. 

408: One type of degeneration that will be important

409: below is when a curve develops a node, {\it i.e.}\ splits into two components.  

410: There is a standard way of incorporating these degenerate curves into 

411: our moduli space of maps; one then obtains a larger compact space 

412: $\overline{\M_{0,n,d}}(\CPP)$, called the ``moduli space of

413: stable maps.''  This moduli space is a smooth algebraic variety, except for

414: certain orbifold points which will not play an important role in this paper.\footnote{Strictly

415: speaking this theorem has been proven when the

416: target space is $\C\PP^3$ \cite{MR98m:14025}, not for the supermanifold $\CPP$, 

417: but we do not expect any important differences.}

418: 

419: In particular, the ``boundary'' of this moduli space,

420: \begin{equation}

421: \overline{\M_{0,n,d}} (\CPP) \setminus \M_{0,n,d}(\CPP),

422: \end{equation}

423: contains a codimension $1$ divisor

424: which parameterizes curves which have split into two components.

425: Similarly, for any $K$ there is a subspace $\MintK$

426: of codimension $K$ that parameterizes reducible curves

427: with $K$ nodes, {\it i.e.}\ curves which have split up into $K\!+\!1$ intersecting 

428: components which intersect in a tree.

429: This $\MintK$ can be further decomposed into irreducible pieces,

430: \begin{equation}

431: \Mint = 

432: \bigcup_\Gamma \Mintgamma,

433: \end{equation}

434: where the different $\Gamma$ label 

435: different shapes of the tree, together with different decompositions

436: of $d$ into individual degrees $\{d_i\}$, ~

437: $i=1,2,\dots K\!+\!1$, ~

438: $d_i \geq d_{i+1}$,\quad and

439: different ways in which the $n$

440: marked points can be distributed over the 

441: $K\!+\!1$

442: components.

443: Some of these $\Mintgamma$ will play an important role in our discussion below.

444: 

445: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

446: 

447: \section{Review of connected and disconnected prescriptions} \label{sec-review}

448: 

449: Suppose we want to use the twistor prescription of \cite{Witten:2003nn}

450: to evaluate a Yang-Mills amplitude with $q=d+1$ negative helicity gluons.

451: All contributions to this amplitude are expected to involve holomorphic

452: curves of total degree $d$, but \ti{a priori} these can be either

453: connected or disconnected. In this section we review the contributions

454: which would be expected from the two most extreme cases: connected degree $d$

455: curves and completely disconnected families of $d$ degree $1$ curves.

456: 

457: In both cases we will consider the Yang-Mills amplitude with arbitrary 

458: external scattering states.

459: Via the Penrose transform these scattering states are described by

460: twistor space wavefunctions,\footnote{Actually,

461: the wavefunctions are not defined on all of $\CPP$, but this distinction 

462: will not be important for us.} which are

463: $\overline{\partial}$-closed $(0,1)$ forms $\phi_i$ ($i=1, \dots, n$) on $\CPP$.  We always treat these $\phi_i$ as generic.

464: In our computation, we will be focusing on 

465: poles which arise in integrals over moduli spaces of curves; 

466: we emphasize that the poles in question never come from the $\phi_i$.

467: 

468: The prescriptions as we write them below are not gauge invariant.

469: To make the amplitudes gauge invariant we would

470: probably have to include additional diagrams in both

471: prescriptions, involving cubic Chern-Simons interaction vertices.

472: Nevertheless, both prescriptions make sense provided 

473: we choose a specific gauge for

474: the gauge field, such as an axial gauge.

475: In this gauge one expects

476: that the cubic vertices do not contribute \cite{Witten:2003nn}.\footnote{We

477: thank Peter Svr\v{c}ek for reminding us of this point.}

478: 

479: \subsection{Connected prescription} \label{review-connected}

480: 

481: We first review the connected prescription for computation

482: of $n$-point Yang-Mills amplitudes.

483: The amplitude is obtained as an integral over degree $d$ maps 

484: \begin{equation}

485: P: \C\PP^1 \to \CPP.

486: \end{equation}

487: Such a map $P$ can be written explicitly, in terms of the inhomogeneous

488: coordinate $\sigma$ on $\C\PP^1$, as

489: \eqn{polyno}{

490: P^\IA(\sigma)=Z^\IA = \sum_{k=0}^d \beta^\IA_k \sigma^k

491: }

492: The supermoduli of the degree $d$ map $P$

493: are $\beta^\IA_k$; these span a space $\C^{4d+4|4d+4}$, which comes

494: equipped with the natural measure 

495: \begin{equation}

496: \mu_d = \prod_{k,\IA} \de \beta^\IA_k.

497: \label{natural}

498: \end{equation}

499: We also have a holomorphic $n$-form on $(\C\PP^1)^n$ given by the free-fermion correlator, 

500: \begin{equation} \label{free-fermions}

501: \omega(\sigma_1, \dots, \sigma_n) = 

502: \prod_{i=1}^n \frac{\de\sigma_i}{\sigma_i - \sigma_{i+1}},\qquad

503: \sigma_{n+1}\equiv \sigma_1.

504: \end{equation}

505: 

506: Note that both $\mu$ and $\omega$ are invariant under the group

507: $GL(2,\C)$ that acts linearly on the homogeneous coordinates on $\C\PP^1$. 

508: Its action on $\sigma$ is given by the usual expression

509: \eqn{modular-sigma}{ \sigma \mapsto \sigma' = \frac{a\sigma + b}{c\sigma+ d},

510: \qquad

511: ad-bc\neq 0}

512: while its action on $\beta^\IA_k$ is dictated by the invariance of $Z^\IA$

513: in \eqref{polyno}: the coefficients $\beta^\IA_k$ may be reorganized 

514: (up to some combinatorial factors suppressed for simplicity)

515: into a rank $d$ tensor under $GL(2,\IC)$,

516: \eqn{ranktensor}{ \{\beta^\IA_k\} = \{\beta^\IA_{I_1 I_2 \dots I_d} \},

517: \qquad

518: I_l = 1,2,}

519: where the number of indices $I_l=2$ equals $k$,

520: so that the action of $GL(2,\C)$ on $\beta^\IA_k$ becomes

521: \eqn{action-on-beta}{ \beta^\IA_{I_1 I_2 \dots I_d}

522: \mapsto {\beta'}^\IA_{I_1 I_2 \dots I_d} =

523: M_{I_1}^{\,I'_1}

524: M_{I_2}^{\,I'_2}\dots

525: M_{I_d}^{\,I'_d}

526: \beta^\IA_{I'_1 I'_2 \dots I'_d},\quad

527: M_{I}^{\,I'}=\left(\!\!

528: \begin{array}{rr}d&-b\\ -c&a\end{array}

529: \right)

530: .}

531: Along with $\mu$ and $\omega$ we also have to include the external wavefunctions,\footnote{We write

532: $\phi(P(\sigma_i))$ for the pullback of $\phi$ to moduli space via

533: the evaluation map sending $P$ to $P(\sigma_i)$.}

534: \begin{equation}

535: \Phi = \prod_{i=1}^n \phi_i(P(\sigma_i)).

536: \end{equation}

537: Putting everything together,

538: the Yang-Mills amplitude is formally\footnote{Here and below,

539: by $\mathrm{vol}(GL(2,\C))$ we really mean the volume form

540: on that group; this is just the standard quotient, when written

541: in terms of an integral over the quotient space.}

542: \begin{equation} \label{formal-yma}

543: \int_{\Mstab{n}{d}} 

544: \frac{\mu_d \wedge \omega(\sigma_1, \dots, \sigma_n)}{\mathrm{vol}(GL(2,\C))}

545: \wedge \Phi.

546: \end{equation}

547: 

548: The expression \eqref{formal-yma} is formal for several reasons.

549: The first and most serious reason is that we have to choose a contour

550: for the integral over the coordinates $\beta^\IA_k$ in 

551: $\Mstab{n}{d}$, and the proper choice of contour

552: is not yet well understood. (We do not have

553: to choose a contour for the integrals over $\sigma$,

554: because the integrand includes both $\de \sigma$ from $\omega$

555: and $\de \bar{\sigma}$ from the external wavefunctions.)

556: We will have more to say about the contour below; to match the

557: disconnected prescription we will 

558: essentially use a contour around infinity (suitably defined) so that {\it all}

559: residues are counted.

560: 

561: Second, we have to divide out by the action of $GL(2,\C)$.

562: A convenient gauge-fixing will be chosen below, but of course the amplitude

563: is independent of the choice of gauge.

564: We should perhaps mention that we consider $GL(2,\C)$ over $\C$, {\it i.e.}\  

565: we divide by the ``holomorphic'' volume form.  This means that

566: \begin{itemize}

567: \item this symmetry will always be fixed by a set of holomorphic

568: conditions; 

569: \item we will sum over all inequivalent solutions;

570: \item only the holomorphic Jacobian will be included in the integrals.

571: \end{itemize}

572: These rules are compatible with the computations of

573: \cite{Roiban:2004vt,Roiban:2004ka,Roiban:2004yf}.

574: 

575: 

576: 

577: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

578: 

579: 

580: \vspace{3mm}

581: 

582: \begin{figure}[t]

583: \begin{center}

584: \epsfig{file=twistor1.eps,width=140mm}

585: \end{center}

586: \caption{A contribution to Yang-Mills amplitudes with $5$ positive

587: and $5$ negative helicity gluons, represented

588: {\bf (a)} as four disconnected lines in twistor space,

589: {\bf (b)} as a graph $\Gamma$ with four MHV vertices.

590: }

591: \label{fig-mhv-tree}

592: \end{figure}

593: 

594: 

595: 

596: \subsection{Disconnected prescription} \label{review-disconnected}

597: 

598: Now we describe the disconnected prescription for the same amplitudes, 

599: formulated in twistor space along the lines of

600: the derivation given in \cite{Cachazo:2004kj}.

601: In this prescription a tree level amplitude involving $d+1$ negative helicity

602: gluons, with a particular cyclic ordering, is obtained as a sum over

603: various tree diagrams with $d$ vertices.

604: In Figure \ref{fig-mhv-tree} we show a representative example of a 

605: diagram $\Gamma$ which contributes to

606: amplitudes with $5$ positive and $5$ negative helicity gluons.

607: The $10$ external gluons are arranged cyclically around the index loop,

608: and since there are $5$ negative helicity gluons there are $5-1=4$ vertices.

609: The vertices have arbitrary

610: valence.\footnote{Ultimately, it turns out that any diagram containing

611: a vertex of valence $\le 2$ does not contribute to the amplitude \cite{Cachazo:2004kj}.}

612: We have not specified which gluons have which helicities;

613: the twistor space computation yields superspace expressions

614: which generate the answers for all possible choices

615: when suitably expanded in the fermionic coordinates.

616: 

617: \vspace{3mm}

618: 

619: \begin{figure}

620: \begin{center}

621: \epsfig{file=tree-circles.eps,width=70mm}

622: \end{center}

623: \caption{A different version of Figure \ref{fig-mhv-tree},

624: representing the same single-trace amplitude

625: with the index line made manifest. The circles represent

626: degree 1 curves in twistor space.}

627: \label{fig-mhv-tree-index-line}

628: \end{figure}

629: 

630: Each vertex of $\Gamma$ corresponds to a $\C\PP^1$ in $\CPP$, equipped with

631: marked points corresponding to internal or external lines attached to the vertex.

632: To compute the contribution of $\Gamma$ to the amplitude we have

633: to integrate over the moduli of these curves,

634: given by $d$ degree $1$ maps

635: \begin{equation}

636: Q_i: \C\PP^1 \to \CPP.

637: \end{equation}

638: Each such map can be written

639: \begin{equation}

640: Q_i^\IA(\sigma) = \sum_{k=0}^1 \beta^\IA_{k,i} \sigma^k

641: \end{equation}

642: so there are a total of $8d|8d$ supermoduli $\beta^\IA_{k,i}$ for these $d$ maps,

643: reduced to $4d|8d$ by the $GL(2,\C)^d$ symmetry.  We also have to integrate over the

644: moduli for the marked points; if in the diagram $\Gamma$ there

645: are $n_i$ marked points on the $i$-th $\C\PP^1$, 

646: then the full moduli space is

647: \begin{equation}

648: \Mlinesgamma = \prod_{i=1}^d \M_{0,n_i,1}(\CPP). 

649: \end{equation}

650: 

651: As in the connected case there is a natural measure

652: for the moduli of the curves,

653: \begin{equation}

654: \mulines = \prod_{k,\IA,i} \de \beta^\IA_{k,i}.

655: \end{equation}

656: There are several factors in the integrand which depend on the marked points.

657: First, there is a free-fermion correlator for each curve; the points on the $i$-th

658: $\C\PP^1$ come with a cyclic ordering as indicated in Figure \ref{fig-mhv-tree}, and if

659: we label them $\sigma_1, \dots, \sigma_{n_i}$, they contribute

660: \begin{equation}

661: \omega_i = \omega(\sigma_1, \dots, \sigma_{n_i})

662: \end{equation}

663: with $\omega$ defined in \eqref{free-fermions}.

664: These free-fermion correlators contain $\de \sigma$ for each

665: marked point.

666: 

667: Next we have to include the external wavefunctions:

668: each external wavefunction $\phi_j$ is connected

669: to a marked point $\sigma$ on the $i$-th $\C\PP^1$,

670: for some $i$, and the integrand includes the factor 

671: \begin{equation} \label{external}

672: \phi_j(Q_i(\sigma))

673: \end{equation}

674: just as in the connected prescription.

675: But unlike the connected prescription, here we also have some marked points

676: which are connected to internal propagators.

677: Let us write $D(\cdot, \cdot)$ for the twistor space propagator,

678: which is a $(0,2)$-form on $\CPP \times \CPP$.

679: Each internal propagator is connected to two

680: marked points $\sigma$, $\sigma'$ on the $i$-th

681: and $i'$-th $\C\PP^1$'s respectively,

682: for some $i$, $i'$, and contributes to the integrand a factor

683: \begin{equation} \label{internal}

684: D(Q_i(\sigma), Q_{i'}(\sigma')).

685: \end{equation}

686: Let us write $\Phi \wedge D$ for the product of all the wavefunctions

687: and propagators from \eqref{external}, \eqref{internal}.

688: Since every marked point is attached either

689: to a propagator or to an external wavefunction, this $\Phi \wedge D$

690: includes one factor $\de \bar{\sigma}$ for each marked point.

691: 

692: Then the amplitude in the disconnected prescription is

693: given by the sum over tree diagrams,

694: \begin{equation}

695: \sum_{\Gamma} \int_{\Mlinesgamma} 

696: \frac{\mulines

697: \wedge \left(\prod_{i=1}^d \omega_i \right) 

698: \wedge \Phi\wedge D}{\mathrm{vol}(GL(2,\IC))^d}

699: \label{discpr}.

700: \end{equation}

701: 

702: As with the connected prescription, to make this integral

703: concrete we have to do two more things.

704: First, we must gauge-fix the symmetry $GL(2,\C)^d$ which

705: acts separately on each $\C\PP^1$.  Second, we must choose

706: a contour for the integrals over the moduli $\beta^\IA_{k,i}$.

707: 

708: In \cite{Cachazo:2004kj} it was argued that if one makes a particular choice

709: of contour, and chooses external wavefunctions corresponding to gluons of fixed helicity

710: and momentum, then the integral over $\Mlinesgamma$ in \eqref{discpr} can be evaluated by a simple rule.

711: Namely, one first assigns $(+)$ and 

712: $(-)$ helicities to the endpoints of each propagator,

713: consistent with the rule that each vertex should have exactly two $(-)$ helicities on it;

714: for given $\Gamma$, 

715: there is at most one way to do this.  (If there is no way to do it, then the diagram $\Gamma$

716: just contributes zero.)  Then each vertex gives a copy of the MHV 

717: amplitude --- continued off-shell in a specific way to accommodate the internal lines ---

718: while each propagator carrying momentum $q$ gives $1/q^2$.

719: 

720: For future use in section \ref{csw-generalized} we also mention a natural

721: generalization of the disconnected prescription:  instead of using $d$ degree $1$ curves

722: we could use $K+1$ curves for some $K$, with total degree $d$, 

723: connected into a tree by $K$ propagators.  The integrand is then defined in a way

724: precisely analogous to \eqref{discpr}, except that the sum over $\Gamma$ includes 

725: all choices for the degrees of the curves in addition to distributions of the marked points.

726: 

727: 

728: 

729: 

730: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

731: 

732: \section{Matching the prescriptions in degree $2$ case}

733: 

734: 

735: \subsection{The argument in degree 2 case}

736: 

737: How can the disconnected and connected prescriptions give the same result?

738: Let us consider next-to-maximally helicity violating amplitudes, $q=3$,

739: which come from degree $2$ curves.  We postpone the discussion

740: of curves of higher degree to section 

741: \ref{higherdegree}.

742: 

743: The contribution of disconnected

744: instantons comes from pairs of degree $1$ curves connected by a

745: single propagator, with $n$ marked points distributed over the pair of

746: curves.  This moduli space has dimension $(8+n)|16$ (which includes

747: $4|8$ for each degree $1$ curve

748: plus $n$ for the marked points.)

749: Different distributions of the marked points correspond to

750: different MHV diagrams $\Gamma$.\footnote{There 

751: are $n(n+1)/2$ such diagrams, although once

752: we fix the external wavefunctions not every diagram gives a nonzero contribution

753: to the sum \eqref{discpr}; 

754: if the helicities are $---+++ \cdots ++$, then there are $2(n-3)$

755: diagrams which contribute.}

756: 

757: \begin{figure}[t]

758: \begin{center}

759: \epsfig{file=branches2.eps,width=145mm}

760: \end{center}

761: \caption{A degenerate configuration of two intersecting lines

762: in $\CPP$ can be deformed into a smooth connected curve

763: of degree 2 or into two disconnected lines.

764: The transition between the two branches of

765: moduli space is reminiscent of

766: a conifold transition.

767: }

768: \label{branches}

769: \end{figure}

770: 

771: It was shown in \cite{Cachazo:2004kj} that for each $\Gamma$ the

772: integrand in \eqref{discpr} has a simple pole on the submoduli space $\Mintgamma$,

773: parameterizing degenerate configurations of intersecting lines of degree 1.  

774: This submoduli space has dimension $(7+n)|12$, because

775: the condition that there exists an intersection in the bosonic space 

776: removes one bosonic modulus, and the condition that all four fermionic 

777: coordinates of the two lines coincide at this point removes four fermionic 

778: moduli.\footnote{This fermionic delta-function guarantees the opposite helicity of

779: the two endpoints of the propagators when one expands in fermions to evaluate a 

780: particular amplitude.}

781: 

782: After contour-integrating to localize to $\Mintgamma$, the sum \eqref{discpr} can be written as

783: %

784: \eqn{dtwoi}{

785: \sum_{\Gamma} \int_{\Mintgamma} 

786: \frac{1}{\mathrm{vol}(GL(2,\C))^{2}}

787: \left( \muint \wedge

788: \left( \prod_{i=1}^{n_1} \frac{\de\sigma_i}{\sigma_i - \sigma_{i+1}} \right)

789: \wedge

790: \left( \prod_{j=1}^{n_2} \frac{\de\sigma_j'}{\sigma_j' - \sigma_{j+1}'} \right) \right)

791: \wedge \Phi.

792: }

793: %

794: Here $i$ and $j$ run over the marked points on each $\C\PP^1$,

795: including the point of intersection;

796: so for a diagram with $m$ external wavefunctions

797: attached to the first line, $n_1 = m + 1$

798: and $n_2 = n - m + 1$.  

799: Also, $\sigma_{n_1+1} \equiv \sigma_1$ and $\sigma'_{n_2+1} \equiv \sigma'_1$.

800: The measure $\muint$ is completely determined by the symmetries

801: of $\CPP$.

802: 

803: On the other hand, from the connected

804: prescription \eqref{formal-yma} we find

805: %

806: \eqn{dtwoc}{

807: \int_{\Mstab{n}{2}}

808: \frac{1}{\mathrm{vol}(GL(2,\C))}

809: \left( \mu_2 \wedge

810: \left( \prod_{i=1}^{n} \frac{\de\sigma_i}{\sigma_i - \sigma_{i+1}} \right) \right)

811: \wedge 

812: \Phi.

813: }

814: %

815: We will reorganize the integral \eqref{dtwoc} over the $(8+n)|12$-dimensional

816: space $\Mstab{n}{2}$ of conics in the following way:

817: Locally, to any conic we will associate a pair of intersecting lines which 

818: are its ``asymptotes.''  The moduli space of pairs of intersecting lines 

819: with $n$ marked points is  

820: the $\Mint$ which occurred in the disconnected prescription.  

821: This $\Mint$ has 

822: dimension $(7+n)|12$, so in $\Mstab{n}{2}$ there is one more coordinate, which we call $C$;

823: $C=0$ corresponds to the singular conics, which coincide with their

824: asymptotes. 

825: This $C$ can be thought of as a ``deformation parameter'' which resolves the singularity.

826: We will find that the integrand has a pole at $C=0$, {\it i.e.}\ along

827: $\Mint$.

828: 

829: More precisely, $\Mint$ includes only those degenerations in which the marked

830: points are distributed 

831: in a way corresponding to some MHV tree graph $\Gamma$.  This just means the points are broken

832: into two groups which are cyclically ordered --- so e.g.\ if $n=6$, there is a component of $\Mint$ with points $1,2,3$ 

833: on one line and $4,5,6$ on the other, but we do not include the degeneration which has $1,2,4$ on one line and $3,5,6$ on the other.  Indeed, we will see that the latter degeneration 

834: does \ti{not} give a pole.

835: We will find poles only along $n(n+1)/2$ distinct components $\Mintgamma$, 

836: which are in one-to-one correspondence with the diagrams $\Gamma$ contributing to \eqref{dtwoi}.

837: 

838: Moreover, we will show that the residue along $\Mintgamma$ is precisely

839: such that the integral \eqref{dtwoc} agrees with \eqref{dtwoi} after localizing.

840: This will complete the argument for the equivalence in the degree $2$ case.

841: 

842: 

843: \subsection{Computing the residue in degree 2 case\label{lower-residue}}

844: 

845: In this section we show that the integral \eqref{dtwoc} over the moduli

846: space $\Mstab{n}{2}$ of genus zero, degree 2 curves in $\CPP$ with $n$ marked points 

847: has a pole at the subspace $\Mint$ describing

848: pairs of intersecting lines, and that

849: it has the desired residue as discussed in the last section.

850: 

851: Let us start by fixing part of the $GL(2,\C)$ symmetry reviewed in section

852: \ref{review-connected}.

853: We use three generators of $GL(2,\C)$ to impose

854: the constraints

855: \eqn{gaugechoice}{

856: P^4(\sigma) = \sigma

857: \mbox{\qquad {\it i.e.}\qquad}

858: (\beta^4_0,\beta^4_1,\beta^4_2) = (0,1,0).

859: }

860: In other words, we are imposing the conditions that the two intersections 

861: of the hyperplane $Z^4\!=\!0$\, with the curve have

862: coordinates\footnote{The point $\sigma=\infty$ can be written

863: as $(1:0)$ in homogeneous coordinates, and therefore is 

864: completely nonsingular.}

865: $\sigma\!=\!0$ and $\sigma\!=\!\infty$, and normalizing

866: the coefficients $\beta^4_{0,1,2}$.

867: There is one more generator of $GL(2,\C)$

868: to be fixed, the matrix

869: \begin{equation}

870: M = \begin{pmatrix}

871: \lambda & 0 \\

872: 0 & \lambda^{-1}

873: \end{pmatrix},

874: \end{equation}

875: which acts as

876: \eqn{unfixed}{\beta^\IA_k \to \lambda^{2-2k} \beta^\IA_k,

877: \qquad

878: \sigma\to \lambda^2 \sigma.

879: }

880: This transformation preserves the gauge choice \eqref{gaugechoice}.

881: 

882: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

883: 

884: \subsubsection{Factors from the measure on the moduli space}

885: 

886: Using the freedom to divide all twistor coordinates $Z^\IA$ by $\sigma$,

887: we can write \eqref{polyno} as

888: \eqn{polynoo}{

889: P^\IA(\sigma)=Z^\IA = \sum_{k=0}^{2} \beta^\IA_{k} \sigma^{k-1} =

890: \beta^\IA_0\sigma^{-1} + \beta^\IA_1 + \beta^\IA_2 \sigma,

891: }

892: which using \eqref{gaugechoice} implies $P^4(\sigma)=1$.  As $\sigma \to \infty$ or $\sigma \to 0$,

893: we can neglect the first or the last term in \eqref{polynoo}, respectively.

894: So \eqref{polynoo} describes a hyperbola that approaches

895: two asymptotic lines in the superspace $\IC^{3|4}$:

896: \eqn{asymplines}{Z^\IA=

897: \beta^\IA_0\sigma^{-1} + \beta^\IA_1,

898: \qquad

899: Z^\IA = \beta^\IA_1 + \beta^\IA_2 \sigma.

900: }

901: These two lines intersect at the point $Z^\IA=\beta^\IA_1$, while

902: $\beta^\IA_0$ and $\beta^\IA_2$ with $\IA\neq 4$

903: are the tangent vectors along 

904: these lines.  It is important that for every conic

905: $\Sigma := P_* (\C\PP^1) \subset \CPP$ we can find a singular

906: conic $\Sigma'$ (a pair of intersecting lines) in $\Mint$

907: defining the asymptotes of $\Sigma$.

908: This rule is not canonical; 

909: it depended on our choice to single out 

910: the points at infinity, {\it i.e.}\ the hyperplane $Z^4\!=\!0$.

911: 

912: We want to express $\Mstab{n}{2}$ locally

913: as a product of $\Mint$ and $\C$, with the extra $\C$ parameterized

914: by the deformation parameter $C$.  What are the appropriate coordinates?

915: The $3|4$ parameters 

916: \eqn{firstmoduli}{\beta^\IA_1,\qquad \IA\neq 4,} 

917: describing the position of the intersection of the asymptotes,  

918: give coordinates on $\Mint$.  The remaining $4|8$ 

919: coordinates on $\Mint$ are the directions of the two asymptotes; 

920: each asymptote gives us $2|4$ moduli.  

921: We want to describe these directions by ``unit vectors'' in a suitable sense.

922: As we approach a generic point of $\Mint$, 

923: $\beta^3_0$ and $\beta^3_2$ are nonzero, and we 

924: may use them to normalize the direction vectors. 

925: In other words, the remaining $2|4$ plus $2|4$ coordinates on $\M_{{\rm int}}$ may be 

926: chosen as

927: \eqn{remainingm}{

928: \frac{\beta^\IA_0}{\beta^3_0}

929: \mbox{\quad and \quad}

930: \frac{\beta^\IA_2}{\beta^3_2},\qquad

931: \IA\in\{1,\,2\,|1',2',3',4'\}.

932: }

933: (Choosing different coordinates on $\Mint$ instead of \eqref{firstmoduli}

934: and \eqref{remainingm} would not change the result below; the only change 

935: would be a $C$-independent Jacobian.)

936: 

937: Looking at our original coordinates on $\Mstab{n}{2}$, we still have

938: two more bosonic components of $\beta$ which are independent of our

939: coordinates on $\Mint$, namely $\beta^3_0$ and $\beta^3_2$ themselves.

940: We also have one unfixed generator of $GL(2,\C)$ given in

941: \eqref{unfixed}.  This generator simply multiplies the ratio 

942: $\beta^3_0 / \beta^3_2$ by $\lambda^4$, so we can use it to fix that

943: ratio to a constant, such as

944: \eqn{gaugechoicee}{\frac{\beta^3_0 }{\beta^3_2} = 1.} 

945: Now having fixed the full $GL(2,\C)$ symmetry we can write the measure $\mu_2$

946: from \eqref{natural} as

947: \begin{equation} \label{measure-deltas}

948: (J/4)\ \prod_{k, \IA} \de \beta_k^\IA\ \delta(\beta_0^3 / \beta_2^3 -1)\ \delta(\beta_0^4)\ \delta(\beta_1^4 - 1)\ \delta(\beta_2^4).

949: \end{equation}

950: Here $J$ is the determinant of the Jacobian matrix of variations of the constraints

951: with respect to the $GL(2,\C)$ generators.

952: If we parameterize the generators of $GL(2,\C)$ by

953: \begin{equation}

954: M = \begin{pmatrix}1+a & b \\ c & 1+d\end{pmatrix}

955: \end{equation}

956: then this matrix is

957: \eqn{fixingmatrix}{

958: \delta

959: \left(

960: \begin{array}{c}

961: \beta^4_0\\

962: \beta^4_1\\

963: \beta^4_2\\

964: \beta^3_0 / \beta^3_2

965: \end{array}

966: \right)

967: =

968: \left(

969: \begin{array}{rrrr}

970: \!\!1&0&0&0\\

971: 0&1&0&0\\

972: 0&0&1&1\\

973: *&\phantom{-}*&\phantom{-}2&-2

974: \end{array}

975: \right)

976: \left(

977: \begin{array}{c}

978: b\\

979: c\\

980: a\\

981: d

982: \end{array}

983: \right)

984: }

985: and hence we get simply

986: \begin{equation}

987: J = -4.

988: \end{equation}

989: The factor $J/4$ in \eqref{measure-deltas} represents

990: $1/\mathrm{vol}(GL(2,\C))$; we had to divide by $4$

991: because the $\Z_4 \subset GL(2,\C)$ generated by

992: \begin{equation} \label{unfixed-discrete-matrix}

993: M = 

994: \left(\begin{array}{rr}

995:   i    & 0 \\

996:   0    & -i

997: \end{array}\right)

998: \end{equation}

999: is left unfixed by our gauge condition.

1000: 

1001: The three delta functions in \eqref{measure-deltas} involving

1002: $\beta_k^4$ just eliminate the integrals over those variables,

1003: imposing \eqref{gaugechoice}.  Let us

1004: also use $\delta(\beta_0^3 / \beta_2^3 -1)$ to eliminate $\beta_0^3$,

1005: imposing \eqref{gaugechoicee}.

1006: Integrating over $\beta_0^3$ gives

1007: a factor $\beta_2^3$, so the measure becomes

1008: \eqn{natura}{

1009: - \beta_2^3\,\de \beta_2^3

1010: \prod_{\IA\neq 4} \de \beta_1^\IA

1011: \prod_{k\in\{0,2\}}  \prod_{\IA\neq 3,4}

1012: \de

1013: \beta_k^\IA.

1014: }

1015: We rewrite this as a measure for

1016: the single transverse coordinate $\beta_2^3$, times

1017: a measure on $\Mint$, for which a full set of $7|12$ coordinates were given 

1018: in \eqref{firstmoduli}, \eqref{remainingm}:

1019: \eqn{naturab}{

1020: \left(- (\beta_2^3)^{1-4} \de \beta_2^3\right)

1021: \times \left( \prod_{\IA\neq 4} \de \beta_1^\IA

1022: \prod_{k\in\{0,2\}}  \prod_{\IA\neq 3,4}

1023: \de

1024: \left(

1025: \frac{\beta_k^\IA}{\beta_k^3}

1026: \right) \right).

1027: }

1028: The extra power $(-4)$ in $(\beta_2^3)^{-4}$ was calculated as

1029: $2_{k=0,2}\times (2_B-4_F)$; the terms $2_B$ and $-4_F$ arise from the

1030: redefined bosonic {\it and} 

1031: fermionic measures involving $\beta_k^\IA$, respectively.

1032: 

1033: The coordinate $\beta_2^3$ is related to 

1034: the deformation parameter $C$ --- we will see

1035: that the natural definition of $C$ is $(\beta_2^3)^{2}$.

1036: The measure

1037: $(\beta_2^3)^{-3} \de \beta_2^3$ occurring in \eqref{naturab} 

1038: will be corrected to

1039: $\de \beta_2^3 / \beta_2^3$ --- the desired pole --- once we include an extra factor

1040: $(\beta_2^3)^{2}$ which comes from the free-fermion correlator $\omega$.

1041: We now turn to the analysis of this factor.

1042: 

1043: 

1044: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1045: 

1046: 

1047: \subsubsection{Factors from the fermion correlator}

1048: 

1049: The integrand \eqref{dtwoc} contains the factor

1050: \begin{equation} \label{free-fermions-copy}

1051: \omega(\sigma_1, \dots, \sigma_n) =

1052: \prod_{i=1}^n \frac{\de\sigma_i}{\sigma_i - \sigma_{i+1}},\qquad

1053: \sigma_{n+1}\equiv \sigma_1.

1054: \end{equation}

1055: We would like to investigate how this form behaves on

1056: conics that are degenerating into a pair of lines ({\it i.e.}\ near $\Mint$.)

1057: The result will be that along $\Mint$, $\omega$ factorizes into a product

1058: of two copies of $\omega$ defined on the two lines separately (with an extra

1059: marked $\sigma$ on each line at the point of intersection), while transverse

1060: to $\Mint$, $\omega$ vanishes like $(\beta_2^3)^{2}$.

1061:  

1062: As the curve degenerates to

1063: a pair of lines, some of the $n$ insertions approach one line and some approach the

1064: other.

1065: We consider the case where

1066: \eqn{firstgroup}{\sigma_1, \dots,\sigma_m}

1067: approach one asymptote while the remaining

1068: $(n-m)$ insertions 

1069: \eqn{secondgroup}{\sigma_{m+1}, \dots, \sigma_n}

1070: approach the other.  This is not the most general choice, since the $\sigma_i$ come

1071: with a fixed cyclic ordering which is built into \eqref{free-fermions-copy}; our

1072: choice is characterized by the fact that as we run through the cyclic ordering we jump from the first

1073: line to the second and back only once.

1074: We will comment on other possibilities at the end.

1075: 

1076: With the $GL(2,\C)$ gauge-fixing we chose above, 

1077: as we approach some point of $\Mint$, the coordinates $\sigma_i$ do not remain finite;

1078: one of the lines is $\sigma \to 0$ while the other line is $\sigma \to \infty$.

1079: So we need to rescale the $\sigma_i$ to get new coordinates $\hat \sigma_i$ on $\Mint$

1080: which label the positions of the marked points;

1081: we define $\hat \sigma_i$ so that 

1082: $Z^\IA$ defined in \eqref{asymplines} remains constant

1083: as $\hat\sigma_i$ is kept fixed and $\beta_0^3,\beta_2^3 \to 0$. 

1084: The correct redefinition is

1085: \eqn{newsigma}{\sigma_i

1086: =

1087: \left\{

1088: \begin{array}{lcl}

1089: (\beta_2^3)^{-1} \hat\sigma_i &\mbox{~for~}& i\in\{1,2,\dots m\}\\

1090: \beta_0^3 (\hat\sigma'_i)^{-1} &\mbox{~for~}& i\in\{m+1, 

1091: m+2,\dots n\}\\

1092: \end{array}

1093: \right\}.

1094: }

1095: (We use two different symbols $\hat\sigma_i$ and $\hat\sigma'_i$

1096: to distinguish the coordinates on the two different lines.)

1097: Rewriting $\omega$ from \eqref{free-fermions-copy} in terms of

1098: $\hat\sigma_i$ and $\hat\sigma'_i$, we obtain

1099: \eqn{newsigmaomega}{

1100: \omega(\hat\sigma_1, \dots \hat \sigma'_n)

1101: =

1102: \beta_0^3\beta_2^3

1103: \left(\prod_{i=1}^{m-1}

1104: \frac{\de \hat\sigma_i}{\hat\sigma_i-\hat\sigma_{i+1}}\right)

1105: \frac{\de \hat\sigma_m}{\hat\sigma_1\hat\sigma_m}

1106: \left(\prod_{i=m+1}^{n-1}

1107: \frac{\de \hat\sigma'_i}{\hat\sigma'_i-\hat\sigma'_{i+1}}

1108: \right)

1109: \frac{\de\hat\sigma'_n}{\hat\sigma'_{m+1}\hat\sigma'_n}+ \dots

1110: }

1111: where the intersection was defined to be at $\hat\sigma=\hat\sigma'=0$.

1112: The dots in \eqref{newsigmaomega}

1113: indicate terms suppressed by powers of $\beta_0^3\beta_2^3$.

1114: 

1115: Most of the powers of $\beta^3_0$ and $\beta^3_2$ have canceled,

1116: but there is an extra factor of $\beta_0^3\beta_2^3$,

1117: which equals $(\beta_2^3)^2$ because of our gauge

1118: choice \eqref{gaugechoicee}.  Also, we obtained the expected free fermion contractions, 

1119: including the $2+2$ 

1120: contractions 

1121: involving the intersection of the

1122: two lines at

1123: $\hat\sigma=\hat\sigma'=0$.

1124: 

1125: Note that $\beta_0^3$ and $\beta_2^3$ always appeared in the combination

1126: \eqn{ccombi}{C=\beta_0^3\beta_2^3}

1127: that is invariant under 

1128: \eqref{unfixed}.

1129: This is the same $C$ that we used 

1130: in Figure \ref{branches};

1131: in fact, one can rewrite our curve in the form

1132: \eqn{xyc}{xy=C}

1133: where $x$,$y$ are coordinates on a plane in $\CPP$.

1134: The limit $C\to 0$ describes the singular conics.

1135: Note that it is $C$ rather than $\beta_2^3$ that is a good coordinate ---

1136: this is because a simultaneous sign flip on $\beta_0^3$ and $\beta_2^3$ 

1137: is the gauge transformation 

1138: \eqref{unfixed} with $\lambda = i$,

1139: which preserves our gauge choices \eqref{gaugechoicee}.

1140: 

1141: Finally, it is easy to check that if we choose a different distribution of

1142: the marked points, the result comes out suppressed by additional powers of $C$.

1143: We are only interested in the leading terms, which are linear in $C$ and will give 

1144: the coefficient of \,$\de C / C$.

1145: 

1146: 

1147: \subsection{Finishing the proof in degree 2 case} \label{finishing}

1148: 

1149: Now we can collect the results from the previous two 

1150: subsections.  

1151: The powers of $\beta_2^3$ from 

1152: \eqref{naturab} and \eqref{newsigmaomega} combine to give

1153: $\smallint \de \beta_2^3 / \beta_2^3$, which is proportional to $\smallint

1154: \de C / C$.  So as advertised, the integral \eqref{dtwoc} localizes

1155: after contour integration to an integral over $\Mint$.  The symmetries of 

1156: $\CPP$ determine the measure for the moduli of the two lines in $\Mint$, which therefore agrees

1157: with the measure $\muint$ in \eqref{dtwoi} up to an overall constant; as for

1158: the integral over the marked points, comparing \eqref{newsigmaomega} and

1159: \eqref{dtwoi} we see that these measures are also identical.  This completes

1160: the argument for equivalence in the $d=2$ case.

1161: 

1162: Incidentally, one can also compare the measures on $\Mint$ directly, without

1163: recourse to a symmetry argument.  

1164: We have already computed

1165: the measure which arises from the connected prescription, in \eqref{naturab},

1166: so the job is to compute the measure $\muint$ which arises from the disconnected

1167: prescription.  This computation is given (in greater generality) in section \ref{finishing-higher}.

1168: 

1169: 

1170: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1171: 

1172: \section{Higher degree \label{higherdegree}}

1173: 

1174: Now let us consider the connected prescription for general degree $d$.

1175: We will see that the fully disconnected description and the fully connected

1176: prescription are not only equivalent, they are

1177: just two extreme cases of a more general class of

1178: rules to calculate the amplitude.

1179: We will find $d$ \ti{a priori} different expressions for

1180: the scattering amplitude with $d\!+\!1$ negative-helicity

1181: gluons,

1182: \eqn{ways}{{\cal A}_{[K]},\qquad K=0,1,2,\dots , d-1,}

1183: where $K\!+\!1$ denotes the total number of curves involved in the 

1184: prescription.\footnote{Later we will see that $K$ also represents the codimension in moduli space on which

1185: the prescription is localized, or equivalently the number of internal propagators

1186: which appear in the prescription.}

1187: 

1188: The organization of this section is as follows:

1189: 

1190: \begin{itemize}

1191: 

1192: 

1193: \item subsection \ref{plethora} outlines the argument that

1194: the completely connected and completely disconnected prescriptions

1195: agree;

1196: 

1197: \item subsection \ref{csw-generalized} discusses the intermediate prescriptions

1198: with arbitrary $K$ and their diagrammatic interpretation;

1199: 

1200: \item subsection \ref{higher-residue} generalizes the residue

1201: calculation of subsection \ref{lower-residue} to the case of a degree $d$

1202: curve splitting into two curves of degrees $d_1$ and $d_2$;

1203: 

1204: \item subsection \ref{finishing-higher} shows that the residues occurring

1205: for any degeneration are actually independent of the chosen prescription, 

1206: completing the argument.

1207: 

1208: \end{itemize}

1209: 

1210: 

1211: \subsection{The proof in higher degree case\label{plethora}}

1212: 

1213: Rather than showing directly that the connected prescription arising from a single

1214: connected degree $d$ curve is equivalent to the disconnected

1215: prescription involving $d$ lines, we will first show that it is

1216: equivalent to a computation 

1217: involving two disconnected components of degrees $d_1$, $d_2$, such that 

1218: \begin{equation} \label{dd}

1219: d_1 + d_2 = d.

1220: \end{equation}

1221: The proof is a 

1222: generalization of the computation we did in section \ref{lower-residue}:  

1223: namely, in subsection \ref{higher-residue} we will find a pole 

1224: on each boundary divisor $\Mintgamma$, corresponding to a degeneration into

1225: intersecting curves,

1226: \begin{equation}

1227: \Sigma_d \longrightarrow \Sigma_{d_1} \cup \Sigma_{d_2},

1228: \quad \quad \quad d_1 + d_2 = d,

1229: \label{dddcurves}

1230: \end{equation}

1231: with a particular distribution of the marked points.  

1232: 

1233: \begin{figure}[t]

1234: \begin{center}

1235: \epsfig{file=d3curves.eps,width=120mm}

1236: \end{center}

1237: \caption{A degeneration of a degree 3 curve into three intersecting

1238: lines can be viewed as a two-step process.

1239: The moduli space of degree 3 maps with 5 marked points, $\Mstab{5}{3}$,

1240: contains divisors, $\M^{\Lambda_1}_{{\rm int}}$ and $\M^{\Lambda_2}_{{\rm int}}$,

1241: associated with degenerations into a degree 2 curve and a line,

1242: shown at the intermediate stages. The moduli space $\Mintgamma$

1243: of three intersecting lines (shown in the lower right corner) can be

1244: identified with the intersection $\M^{\Lambda_1}_{{\rm int}} \cap \M^{\Lambda_2}_{{\rm int}}$.}

1245: \label{d3curves}

1246: \end{figure}

1247: 

1248: Next we want to show iteratively that

1249: this integral over curves with $2$ irreducible components is equivalent to one over curves with $3$ components, and so on until eventually we reach $d$ components (all of which must

1250: have degree $1$.)  The idea which makes this iteration possible is the following:  

1251: consider some locus

1252: $\Mintgamma$, corresponding to a particular degeneration of $\Sigma$ into

1253: $K\!+\!1$ components, with a particular distribution of the marked points.

1254: This locus

1255: can be obtained as an intersection of $K$ boundary divisors, $\M^{\Lambda_j}_{{\rm int}}$,

1256: each of which is associated with a degeneration of $\Sigma_d$

1257: into two irreducible components,\footnote{We use $\Gamma$ to denote a general 

1258: degeneration into $K\!+\!1$ components, and $\Lambda$ to denote a degeneration into just two components.}

1259: \begin{equation}

1260: \Mintgamma = \M^{\Lambda_1}_{{\rm int}} \cap \cdots \cap \M^{\Lambda_{K}}_{{\rm int}}.

1261: \label{mintascap}

1262: \end{equation}

1263: An example is shown in Figure \ref{d3curves}.  

1264: In this sense, the problem 

1265: of studying a general degeneration boils down to

1266: understanding the basic process \eqref{dddcurves}.

1267: 

1268: So let's start with the integral over 

1269: $K$-component curves and try to prove it agrees

1270: with an integral over $(K+1)$-component curves.  

1271: In the $K$-component case we

1272: have to integrate over various loci $\Mintgamma$ as in

1273: \eqref{mintascap}.

1274: Since the various divisors $\Mintlambda$ meet transversally \cite{MR98m:14025},

1275: in integrating over each such $\Mintgamma$ we will encounter poles

1276: wherever $\Mintgamma$ intersects another divisor $\Mintlambda$.\footnote{One

1277: way to understand this is to note that if we start with

1278: the full $\Mstab{n}{d}$ and look near such an intersection of $K$ divisors,

1279: the integrand looks like

1280: \begin{equation}

1281: \frac{\de C_1}{C_1} \wedge \cdots \wedge \frac{\de C_{K}}{C_{K}}

1282: \wedge ({{\rm regular}}).

1283: \end{equation}

1284: We have already contour-integrated 

1285: over $C_1, \dots, C_{K-1}$ and thus restricted to $C_1 = \cdots = C_{K-1} = 0$, \ti{i.e.}\ 

1286: to $\Mintgamma$; after doing this we get simply $\de C_{K}/C_{K}$, with a pole at $C_K = 0$, 

1287: \ti{i.e.}\ at $\Mintgamma \cap \Mintlambda$.}

1288: We choose our contour so that

1289: it picks up the residues at these poles.

1290: In this way we reduce the integral over $\Mintgamma$ to the sum

1291: of integrals over all intersections $\Mintgamma \cap \Mintlambda$.

1292: Then we have to sum over all $\Gamma$ describing $K$-component degenerations.

1293: What is the result of all this summation? From the perspective

1294: of the $(K\!+\!1)$-component degenerations --- which we label by $\Gamma'$ ---

1295: the answer is clear: given some 

1296: \begin{equation}

1297: \Mintgammaprime = \M^{\Lambda_1}_{{\rm int}} \cap \ldots \cap \M^{\Lambda_{K}}_{{\rm int}},

1298: \label{mintascap2}

1299: \end{equation}

1300: there are $K$ ways to make it by intersecting

1301: some $\Mintgamma$ with some $\M^{\Lambda_i}_{{\rm int}}$.

1302: Therefore we get a sum over all $(K\!+\!1)$-component

1303: degenerations, with an \ti{overall} multiplicative factor $K$.

1304: 

1305: Finally, after repeating this process $d-1$ times, we arrive at an 

1306: integral over the moduli space of connected trees consisting of $d$ lines,

1307: with all possible shapes for the tree and all allowed distributions of

1308: marked points.  But the arguments of \cite{Cachazo:2004kj} show that the disconnected prescription also reduces to such an integral, by a similar process of localization

1309: to poles.  Furthermore, in section \ref{higher-residue}

1310: we will see that the residues in these two computations agree;

1311: this will complete the proof.

1312: 

1313: 

1314: 

1315: \subsection{Intermediate prescriptions} \label{csw-generalized}

1316: 

1317: In subsection \ref{plethora} we encountered $d-1$ different moduli spaces $\MintK$ of singular curves, characterized by the number $K+1$ of components, 

1318: which interpolated between the nonsingular degree $d$ curve ($K=0$) and the tree of degree $1$

1319: curves ($K=d-1$).  Furthermore we obtained integrals over each $\MintK$ by starting with the connected

1320: prescription ($K=0$) and successively localizing to poles.  As a result of this localization

1321: all these integrals are equal; now we want to argue that the intermediate cases $K=1, \dots, 

1322: d-2$ can be naturally interpreted as coming from ``intermediate prescriptions,'' involving

1323: integrals over the moduli of $K+1$ disconnected curves with $K$ propagators connecting them.  

1324: We defined these prescriptions at the end of section \ref{review-disconnected}.

1325: 

1326: The argument is a generalization of the ``heuristic'' derivation of

1327: the computational rules for the disconnected prescription, given 

1328: in \cite{Cachazo:2004kj}.

1329: Namely, starting from the intermediate prescription,

1330: note that the propagator $D(\cdot, \cdot)$ by definition satisfies

1331: \begin{equation} \label{prop-def}

1332: \bar{\partial} D = \Delta.

1333: \end{equation}

1334: Here $\Delta$ is a $(0,3)$-form on $(\CPP)^2$

1335: which is concentrated on the diagonal $\CPP$:

1336: in inhomogeneous coordinates with $Z^4 = Z'^4 = 1$ it may be written

1337: \begin{equation}

1338: \Delta = \deltabar(Z^1 - Z'^1)\,\deltabar(Z^2 - Z'^2)\,

1339: \deltabar(Z^3 - Z'^3)\,\delta^{4}(\psi - \psi'),

1340: \qquad

1341: \deltabar(f) := \delta^2(f) \de\bar{f}.

1342: \end{equation}

1343: The equation \eqref{prop-def} means that $D(\cdot, \cdot)$

1344: is meromorphic with a pole along the diagonal.  

1345: The integral over $\MintK$ in the disconnected prescription contains $K$

1346: propagators \eqref{internal}; these factors therefore

1347: have poles when $Q_i(\sigma) = Q_{i'}(\sigma')$.  

1348: 

1349: As in \cite{Cachazo:2004kj}, we assume that $K$ of the integrals over

1350: moduli of the disconnected curves are evaluated on contours

1351: which encircle these poles, in a suitable sense.

1352: Using \eqref{prop-def}, performing these contour

1353: integrals is equivalent to filling in the contour and 

1354: replacing $D$ by $\Delta$.

1355: This localizes the integral to the sublocus of moduli

1356: space where all propagators have shrunk to zero length,

1357: which is exactly $\MintK$.

1358: 

1359: So finally we have $d$ different prescriptions, involving summing 

1360: over configurations with $1$ curve (connected case), $2$, $3$, \dots, $d$ curves (maximally disconnected case); and we have argued that each of these prescriptions is equivalent, up to an overall rescaling.  In this sense any of them can be used to calculate the Yang-Mills

1361: amplitudes.

1362: 

1363: Of course, another possibility is that the correct amplitudes are obtained by summing

1364: different contributions from various sorts of diagrams with various numbers of curves.  We have argued that all such contributions are proportional to one 

1365: another, so such a modified rule would only change the overall prefactor.

1366: Although we will not try to make the final verdict in this paper, we believe that a more

1367: detailed analysis of the prescriptions (including the coefficients) should be able

1368: to resolve this uncertainty.

1369: 

1370: 

1371: 

1372: \subsubsection{Diagrammatic interpretation and an example}

1373: 

1374: Now let us discuss the diagrammatic interpretation of the

1375: intermediate prescriptions.

1376: We have seen that the $K$-th intermediate prescription is naturally localized

1377: on $\MintK$, which is a union of various $\Mintgamma$.  Here $\Gamma$

1378: describes the decomposition of the curve $\Sigma_d$ into $K+1$ components

1379: and the distribution of marked points along these components.  

1380: Equivalently, we could say that

1381: $\Gamma$ describes a slight generalization of an MHV tree diagram:

1382: namely, it is a tree diagram with $K+1$ vertices, where each vertex now carries

1383: an internal index $d_i$, subject to the rule that $\sum d_i = d$.  The MHV

1384: diagrams are the case where all $d_i = 1$.

1385: 

1386: \begin{figure}

1387: \begin{center}

1388: \epsfig{file=twistor3.eps,width=100mm}

1389: \end{center}

1390: \caption{An MHV tree diagram contributing to $\A{+-+---}$.}

1391: \label{twistor3}

1392: \end{figure}

1393: 

1394: It would be very useful if we could give a compact formula for the contribution

1395: of a general vertex with arbitrary $d_i$, analogous to the off-shell

1396: continuation of the MHV amplitude given in \cite{Cachazo:2004kj}.

1397: At the moment we do not possess such a formula, so we can only define the diagram

1398: $\Gamma$ to be the integral over $\Mintgamma$ which we considered above.  In this

1399: language, our localization argument relating different prescriptions 

1400: becomes the statement that the contribution from a diagram $\Gamma$ agrees with

1401: the sum over all $\Gamma'$ obtained by ``splitting a vertex'' in $\Gamma$.  In

1402: other words, $\Gamma'$ should be obtained by replacing a 

1403: vertex with index $d$ by a pair of vertices with indices

1404: $d_1$, $d_2$, such that $d_1 + d_2 = d$, with a propagator connecting them.  This is

1405: the diagrammatic analog of a degree $d$ curve which degenerates into two curves

1406: with degrees $d_1$, $d_2$.

1407: 

1408: We can also repeat the combinatorics from

1409: subsection \ref{plethora} in this language.  Start with a diagram with $K\!+\!1$

1410: vertices.  This diagram contains $K$ propagators.

1411: Therefore there are $K$ ways to shrink a single propagator

1412: and obtain a ``parent'' diagram with $K$ vertices.

1413: Because a diagram with $K\!+\!1$ vertices has $K$ parents, 

1414: the sum over the daughters 

1415: with $K\!+\!1$ vertices equals $K$ times the sum over

1416: the parents with $K$ vertices. 

1417: 

1418: \begin{figure}[t]

1419: \begin{center}

1420: \epsfig{file=twistor2.eps,width=110mm}

1421: \end{center}

1422: \caption{Two types of tree diagram with one MHV and one

1423: non-MHV (degree 2) vertex that contribute to the $\A{+-+---}$ amplitude.

1424: In total, there are six diagrams of each kind.

1425: The number attached to each vertex represents the degree

1426: of the corresponding curve in twistor space.}

1427: \label{twistor2}

1428: \end{figure}

1429: 

1430: To illustrate how all this works when external wavefunctions of fixed

1431: helicity are included, let us consider a 6-gluon amplitude $\A{+-+---}$.

1432: If we were to use the connected prescription, we would have

1433: to integrate over the moduli space $\Mstab{6}{3}$ of degree 3 curves.

1434: On the other hand, in the disconnected prescription one has

1435: to consider three degree 1 curves, which can be interpreted

1436: as MHV vertices in Yang-Mills theory \cite{Cachazo:2004kj}.

1437: Therefore, in this case one has to sum over all tree graphs with

1438: three MHV vertices connected by Yang-Mills propagators --- see Figure \ref{twistor3}.

1439: In total, there are 19 such graphs contributing to $\A{+-+---}$.

1440: 

1441: Now let us consider the intermediate prescription with $K=1$. 

1442: This prescription leads

1443: to a sum over tree graphs with two vertices, one MHV

1444: and one non-MHV (the non-MHV vertex involves three insertions of negative helicity).

1445: Examples of such graphs with non-MHV vertices are shown in Figure \ref{twistor2}.

1446: There are 12 such diagrams which contribute to $\A{+-+---}$.

1447: Since each non-MHV vertex itself can be represented

1448: as a sum over tree diagrams with two MHV vertices, we should be

1449: able to reproduce the disconnected prescription if we split

1450: all non-MHV vertices into MHV ones.

1451: More precisely, in this decomposition we should encounter each

1452: MHV diagram twice (since in the disconnected prescription $K=2$).

1453: Indeed, it is straightforward to check that the 12 non-MHV diagrams

1454: lead to 38 MHV graphs, in agreement with the general rule.

1455: 

1456: 

1457: 

1458: \subsection{Computing the residue in higher degree case\label{higher-residue}}

1459: 

1460: Returning from our digression to discuss the intermediate prescriptions, 

1461: in this section we show that the integral \eqref{formal-yma} over the

1462: moduli space $\Mstab{n}{d}$ which arises in the connected prescription

1463: has a pole along the codimension 1 divisor $\Mintone$ describing

1464: curves that are degenerated into 2 components.  We further verify

1465: that the residue is the same as that which arises after localization of the $K=1$ 

1466: prescription on $\Mintone$, thus establishing the equivalence between connected

1467: and $K=1$ prescriptions.

1468: 

1469: We want to study a degeneration in which 

1470: the curve $\Sigma_d$ degenerates into a pair of intersecting curves,

1471: $\Sigma_{d_1}$ and $\Sigma_{d_2}$, of degree $d_1$ and $d_2$,

1472: as in \eqref{dddcurves}.

1473: Using the projective symmetry to divide by $\sigma^{d_1}$,

1474: we can write the degree $d$ map \eqref{polyno} as

1475: \eqn{polynood}{

1476: Z^\IA(\sigma) = \sum_{k= -d_1}^{d_2} \beta^\IA_{d_1+k} \sigma^k.

1477: }

1478: We fix the $GL(2,\C)$ symmetry similarly to 

1479: the degree 2 case,

1480: namely by conditions based on \eqref{gaugechoice} and \eqref{gaugechoicee}:

1481: \eqn{fixing-d}{

1482: (\beta^4_{d_1-1},\beta^4_{d_1},\beta^4_{d_1+1}) = (0,1,0),\qquad

1483: \frac{\beta^3_{d_1-1}}{\beta^3_{d_1+1}} = 1,

1484: }

1485: and define the deformation parameter $C:= \beta_{d_1-1}^3 \beta_{d_1+1}^3$.

1486: As in degree $2$, the singular limit will be $C \to 0$, or equivalently

1487: $\beta_{d_1+1}^3 \to 0$, and the question is

1488: how the other coefficients should scale in this limit.

1489: 

1490: \begin{figure}

1491: \begin{center}

1492: \epsfig{file=table-beta.eps,width=140mm}

1493: \end{center}

1494: \caption{The organization of the coefficients $\beta^\IA_k$

1495: for a degree $d$ curve degenerating into curves of degrees $d_1$ and $d_2$.

1496: The symmetry $GL(2,\C)$ is fixed by setting three bosonic coefficients to the values

1497: $(0,1,0)$ and two others to $\sqrt{C}$; 

1498: this $C$ is the deformation parameter, which approaches zero in the degeneration

1499: limit.

1500: }

1501: \label{table-beta}

1502: \end{figure}

1503: 

1504: The correct scaling is as follows:  we take $\beta_{d_1-1}^3 = \beta_{d_1+1}^3 \to 0$ while holding

1505: finite the quantities

1506: \eqn{coef-done-dtwo}{

1507: \alpha_k^\IA := \frac{\beta^\IA_{d_1-k}}{(\beta^3_{d_1-1})^k}, \quad 0\leq k \leq d_1; \qquad

1508: \alpha'^\IA_k := \frac{\beta^\IA_{d_1+k}}{(\beta^3_{d_1+1})^k}, \quad 0\leq k \leq d_2.}

1509: In that limit we obtain two curves,

1510: \begin{eqnarray} \label{curves-new-coords}

1511: \Sigma_{d_1}~: && Z^\IA(\hat\sigma) = \sum_{~k= 0~}^{d_1} 

1512: \alpha_k^\IA \hat\sigma^{k}, \nonumber \\

1513: \Sigma_{d_2}~: && Z^\IA(\hat\sigma') = \sum_{~k= 0~}^{d_2} 

1514: \alpha'^\IA_k \hat\sigma^{\prime k}.\label{two-curves-b}

1515: \end{eqnarray}

1516: Namely, we obtain the points $Z^\IA(\hat\sigma)$ on $\Sigma_{d_1}$ by holding fixed

1517: $\hat\sigma = \sigma / \beta^3_{d_1-1}$ in the limit, 

1518: and we obtain the points $Z^\IA(\hat\sigma')$ on $\Sigma_{d_2}$ by holding fixed

1519: $\hat\sigma' = \sigma \beta^3_{d_1+1}$ in the same limit.

1520: See Figure \ref{table-beta}.

1521: 

1522: Therefore the parameters $\alpha_k^\IA, \alpha'^\IA_k$ give coordinates on $\Mintone$,

1523: specifying the moduli of the degenerated curve.

1524: (Note

1525: that $\alpha_0^\IA = \alpha'^\IA_0$; these shared coordinates 

1526: specify the intersection

1527: point of $\Sigma_{d_1}$ and $\Sigma_{d_2}$.)

1528: 

1529: Now we want to study how our integral \eqref{formal-yma} behaves near $\Mintone$.

1530: As in section \ref{lower-residue}, we have to compute the Jacobian $J$ from the

1531: gauge-fixing of $GL(2,\C)$.  The matrix of variations generalizing \eqref{fixingmatrix}

1532: is

1533: \eqn{fixingmatrix-2}{

1534: \delta

1535: \left(

1536: \begin{array}{c}

1537: \beta^4_{d_1-1}\\

1538: \beta^4_{d_1}\\

1539: \beta^4_{d_1+1}\\

1540: \beta^3_{d_1-1} / \beta^3_{d_1+1}

1541: \end{array}

1542: \right)

1543: =

1544: \left(

1545: \begin{array}{ccrr}

1546: d_2&\!\!\!\!(d_1+2)\beta^4_{d_1+2}&0&0\\

1547: \!\!\!\!(d_2+2)\beta^4_{d_1-2}\!\!&d_1 &0&0\\

1548: 0&0&d_1\!&d_2\! \\

1549: *&*&2&-2

1550: \end{array}

1551: \right)

1552: \left(

1553: \begin{array}{c}

1554: b\\

1555: c\\

1556: a\\

1557: d

1558: \end{array}

1559: \right).

1560: }

1561: In the singular limit, the $\beta^4_{d_1 \pm 2}$ terms in \eqref{fixingmatrix-2} 

1562: vanish, and we get

1563: \begin{equation} \label{jac-1}

1564: J \to -2 d_1 d_2 d.

1565: \end{equation}

1566: The gauge-fixed integral includes the factor $J / 2d$; the $2d$ comes from

1567: an unfixed subgroup of $GL(2,\C)$, analogous to \eqref{unfixed-discrete-matrix}, 

1568: which is $\Z_2 \times \Z_d$ if both

1569: $d_1$ and $d_2$ are even and $\Z_{2d}$ otherwise.

1570: Next we have to rewrite the integrand in terms of the new coordinates \eqref{coef-done-dtwo}.

1571: One might be worried that switching to these coordinates

1572: will generate extra powers of $C$ beyond what we had in the degree 2 case,

1573: spoiling the conclusion that there is a pole along $\Mintone$.

1574: But this does not occur; if we increase $d_1$ by 

1575: $1$, for example, the integrand just acquires an extra integral over $4|4$ variables:

1576: \eqn{measure-new}{ \mu \to \mu \wedge \prod_\IA \de \beta^\IA_{0} =

1577: \mu \wedge

1578: \prod_\IA \de \alpha^\IA_{d_1}

1579: }

1580: The powers of $\beta^3_{d_1+1}$ simply cancel between the 4 bosons and 4 fermions!

1581: Unlike the coefficients $\beta^\IA_{d_1}$ and

1582: $\beta^\IA_{d_1\pm 1}$, among which $5$ special bosonic components

1583: have been used to gauge-fix the $GL(2,\C)$ symmetry or to describe the parameter $C$,

1584: the additional moduli $\beta^\IA_{d_1\pm k}$ with $k>1$ come in full ``supermultiplets''

1585: containing 4 bosons and 4 fermions.  Therefore no new powers of $C$ are generated in

1586: rescaling $\beta$'s to $\alpha$'s,

1587: so the measure for the moduli of the degenerating curve still 

1588: behaves as $\de C / C^2$ near $C=0$.  Similarly, the free

1589: fermion correlator $\omega$ factorizes,

1590: \begin{equation}

1591: \omega(\sigma) \to C\,\omega_1(\hat\sigma) \wedge \omega_2(\hat\sigma'), 

1592: \end{equation}

1593: just as in degree $2$.

1594: 

1595: So we have a pole along $\Mintone$, as in the degree $2$ case, and after integrating

1596: around this pole the fully gauge-fixed measure

1597: for the moduli of the degenerate curve is

1598: \begin{equation} \label{measure-degenerate}

1599: -d_1 d_2 \prod_{\IA} \left( {\prodprime{k=0}{d_1}}\! \de \alpha_k^\IA\,{\prodprime{k=1}{d_2}}\! \de \alpha'^\IA_k \right).

1600: \end{equation}

1601: Here the symbol $\Pi'$ indicates that we omit the $5$ factors

1602: \begin{equation} \label{fixed-components}

1603: \de \alpha^4_1,\, \de \alpha'^4_1,\, \de \alpha^4_0,\, \de \alpha^3_1,\, \de \alpha'^3_1;

1604: \end{equation}

1605: there are no such $\alpha$'s among the coordinates on $\Mintone$, because their corresponding $\beta$'s

1606: were used up in the gauge-fixing and in the transverse coordinate $C$, as shown in Figure \ref{table-beta}.  

1607: 

1608: 

1609: 

1610: \subsection{Finishing the proof in higher degree case} \label{finishing-higher}

1611: 

1612: Finally we

1613: have to check that the measure \eqref{measure-degenerate} agrees with the one coming from localization

1614: of the $K=1$ prescription.  From section \ref{csw-generalized} we know that the latter measure

1615: is obtained as follows:  start with two curves of degree $d_1$, $d_2$,

1616: \begin{align} 

1617: Q^\IA(\sigma) &= \sum_{k=0}^{d_1} \alpha^\IA_{k} \sigma^k, \nonumber \\

1618: Q'^\IA(\sigma') &= \sum_{k=0}^{d_2} \alpha'^\IA_{k} \sigma'^k. \label{dcurves}

1619: \end{align} 

1620: (The notation $\alpha^\IA_{k}$, $\alpha'^\IA_k$ we use here

1621: agrees with the notation we

1622: used above for the moduli of the curves obtained by a degeneration; compare \eqref{dcurves} with \eqref{curves-new-coords}, \eqref{coef-done-dtwo}.  The only difference is that here we do not necessarily have $\alpha^\IA_0 = \alpha'^\IA_0$.)

1623: Then we have the standard measure \eqref{natural} on the two curves separately, which before gauge-fixing is

1624: \begin{equation} \label{standard}

1625: \mu_{d_1} \wedge \mu_{d_2} = \prod_{\IA} \left( \prod_{k=0}^{d_1} \de \alpha^\IA_k \prod_{k=0}^{d_2} \de \alpha'^\IA_k \right).

1626: \end{equation}

1627: As explained in section \ref{csw-generalized}, the requirement that the two curves actually intersect is enforced by a delta function which is coupled to one marked point on each curve,

1628: \begin{equation} \label{deltafunc}

1629: \Delta(Q(\sigma), Q'(\sigma')).

1630: \end{equation}

1631: To compare the measures (including this delta function) we have to gauge-fix

1632: the $GL(2,\C)^2$ symmetry acting on the coefficients $\alpha^\IA_k, \alpha'^\IA_k$.

1633: There are many ways to do this;

1634: we choose a way that is as similar as possible to the gauge-fixing we used for the degenerating

1635: degree $d$ curve, so that the unfixed moduli will match directly.

1636: Namely, we take

1637: \begin{align} \label{gauge-conditions1}

1638: \alpha^i_0 &= \alpha'^i_0 \quad \textrm{for}\ i \in \{2,3\}, \\

1639: \alpha^4_0 &= \alpha'^4_0 = 1, \label{gauge-conditions2} \\

1640: \alpha^4_1 &= \alpha'^4_1 = 0, \\

1641: \alpha^3_1 &= \alpha'^3_1 = 1.

1642: \end{align}

1643: The matrix of variations from this gauge-fixing is similar to \eqref{fixingmatrix-2},

1644: but since it is an $8 \times 8$ matrix we just write the answer here:

1645: \begin{equation} \label{jac-2}

1646: J = (d_1 d_2)^2 (\alpha^2_1 - \alpha'^2_1).

1647: \end{equation}

1648: The gauge-fixing factor is $J / {d_1 d_2}$, because of the subgroup $\Z_{d_1} \times \Z_{d_2} \subset

1649: GL(2,\C) \times GL(2,\C)$,

1650: roots of unity acting on each curve separately; since this subgroup

1651: acts trivially it is unfixed by our gauge condition. 

1652: Next we must include the integral over the delta function \eqref{deltafunc}, which we write as

1653: \begin{equation}

1654: \int \de \sigma\ \de \sigma'\ \delta^{(3|4)} \left( \frac{Q^\IA(\sigma)}{Q^4(\sigma)} - \frac{Q'^\IA(\sigma')}{Q'^4(\sigma')} \right).

1655: \end{equation}

1656: With our gauge choice, it is easy to study the behavior of this delta function in the vicinity

1657: of $\sigma = \sigma' = 0$.\footnote{Although our gauge choice was rigged so that studying 

1658: $\sigma = \sigma' = 0$ would recover the desired moduli space of intersecting curves, 

1659: it is not clear \ti{a priori} 

1660: from our arguments why one should consider only this region; this is related to the issue of the

1661: exact contour choice in the intermediate prescription, which we will not settle here.  We are also integrating over the delta function as if it were real instead of holomorphic;

1662: similar manipulations were used in \cite{Witten:2004cp}.}  

1663: One uses the $Z^2$ and $Z^3$ components of the delta function to set $\sigma = \sigma' = 0$, 

1664: obtaining

1665: \begin{equation} \label{deltas}

1666: \frac{1}{(\alpha^2_1 - \alpha'^2_1)}\ \delta(\alpha^1_0 - \alpha'^1_0) \prod_{\IA=1'}^{4'} \delta(\alpha^\IA_0 - \alpha'^\IA_0).

1667: \end{equation}

1668: Note that the $1|4$ delta functions in \eqref{deltas}, combined with the gauge conditions \eqref{gauge-conditions1}, \eqref{gauge-conditions2}, 

1669: are enough to set all $\alpha'^\IA_0 = \alpha^\IA_0$.  This was

1670: the main motivation for this gauge-fixing; the point $\alpha^\IA_0$ represents the intersection 

1671: of the two curves, and the remaining moduli are precisely the ones we had for the degenerating degree

1672: $d$ curve in \eqref{measure-degenerate}.  Therefore we easily see that the measures agree, including

1673: the prefactor $d_1 d_2$.

1674: (Although we have not been careful about overall constant factors,

1675: the absence of a relative factor here is important --- it corresponds to the absence of prefactors

1676: weighing different diagrams in the intermediate prescriptions.)

1677: 

1678: This completes the argument for the equivalence between the connected and $K=1$ prescriptions.

1679: It also sets up the iteration we described in section \ref{plethora} to prove

1680: the equivalence of all prescriptions, by successive localization to poles in higher

1681: and higher codimension, corresponding to more and more degenerate curves.

1682: 

1683: One detail remains:

1684: we have to check that the residues we obtain are always independent of which

1685: prescription we started with.  

1686: In other words, we have to prove that

1687: the measure for the integral over $K+1$-component trees obtained by some degeneration process

1688: always agrees with the measure coming from the disconnected prescription.  

1689: As we know from section \ref{csw-generalized}, the latter measure

1690: can be written as a product of measures for the individual curves, with delta-functions

1691: that guarantee the curves intersect.  

1692: We just proved the agreement for $K=1$.  For general $K$ we can work 

1693: inductively; given a $K+1$-component tree

1694: on which some curve is further degenerating, just focus on the measure for that curve, and 

1695: note that the delta-functions from the other curves are well behaved on moduli space

1696: near the degeneration we are studying.  In this sense the degenerating curve can

1697: be isolated from the rest of the tree.  The computation done above in the $K=1$ case then 

1698: shows that the measure after this degeneration agrees with that from the disconnected 

1699: prescription.  This then completes the argument for the equivalence of all prescriptions.

1700: 

1701: 

1702: 

1703: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

1704: \section{Conclusions and open questions}

1705: 

1706: We have argued for the equivalence of the connected and disconnected

1707: twistorial formulae for the tree level scattering amplitudes of $\N=4$ super Yang-Mills,

1708: provided that the contours are appropriately chosen.

1709: Using this equivalence we can now exploit the complementary virtues of the two

1710: prescriptions simultaneously.  As we remarked in the introduction, the connected prescription

1711: minimizes the number of diagrams one has to sum, namely, there is only one; the amplitude 

1712: is expressed as a single integral, which was the starting point for several theoretical developments

1713: \cite{Berkovits:2004hg,Berkovits:2004tx,Witten:2004cp}.  The disconnected

1714: prescription involves more diagrams, but still a manageable number for some interesting

1715: amplitudes, and the contribution from each diagram can be immediately written down.  

1716: 

1717: To conclude, we summarize some of the many remaining open problems in this area:

1718: 

1719: \begin{itemize}

1720: 

1721: \item {\bf Contours I.}

1722: Is there a rigorous justification of the choice of contours in all these calculations?

1723: In our argument for the equivalence between connected and disconnected

1724: prescriptions we identified specific poles in the integral

1725: over moduli, and we roughly wanted a contour which encircles all of these poles.

1726: We believe it should be possible to show by a deformation argument that 

1727: our choice of contour is equivalent to the one used in \cite{Roiban:2004yf},

1728: thus completing the proof of equivalence,

1729: but this seems to be nontrivial; the 

1730: computations in \cite{Roiban:2004yf} depend on a particular method of evaluating the integral

1731: in the connected prescription by saturating delta-functions, and it is difficult to see which contour 

1732: it corresponds to.

1733: 

1734: \item {\bf Contours II.} Once the residues are isolated in both prescriptions,

1735: we must still integrate over the degeneration locus 

1736: $\Mint$, which requires yet another choice of

1737: contour; for example, the integration over $t$ from $0$ to $\infty$ in section

1738: 6 of \cite{Cachazo:2004kj} should have some \ti{a priori} justification. 

1739: This paper has not addressed this question.

1740: Our argument for the equivalence requires that the contours on $\Mint$ are chosen

1741: to be equivalent in all prescriptions.

1742: 

1743: \item {\bf Explicit external wavefunctions.}

1744: Our derivation was rather formal. It did not depend on the particular form of the wavefunctions.

1745: Of course, it would be interesting to verify the picture by calculating the amplitudes 

1746: involving particles with well-defined momenta {\it i.e.}\ $(\lambda,\tilde\lambda,\psi)$

1747: using our generalized prescriptions.  Unlike the MHV vertices in \cite{Cachazo:2004kj},

1748: one might expect that the $d>1$ vertices will be ratios of polynomials involving both $\lambda$ and $\tilde\lambda$.  (Of course, it is also possible that one will not 

1749: obtain any compact formula for the $d>1$ vertices in this way.)

1750: 

1751: \item {\bf Derivation from the B-model.}

1752: Both connected and disconnected contributions seem to arise in the topological

1753: B-model of \cite{Witten:2003nn} as long as we use not only the degree $d$ 

1754: D1-instantons but also the propagators (and vertices) of the holomorphic Chern-Simons theory.

1755: Does our equivalence suggest that the D1-instantons are not independent

1756: of the Chern-Simons degrees of freedom?

1757: 

1758: \item {\bf Real versions.}

1759: The framework first proposed by Berkovits \cite{Berkovits:2004hg} 

1760: and the topological A-model of \cite{Neitzke:2004pf} seem to prefer

1761: the real version of the twistor space, $\mathbb{RP}^{3|4}$, and correspondingly

1762: real values of the moduli. Is there a real variation of our procedures?

1763: One can imagine that the disconnected rules for the amplitudes might be derived

1764: from the cubic twistorial string field theory of \cite{Berkovits:2004tx} if $K$

1765: stringy propagators are expanded in component fields, so that the different parts

1766: of the worldsheet become effectively disconnected.

1767: 

1768: \item {\bf Choice of prescriptions.}

1769: According to our analysis, there is significant freedom to choose a 

1770: twistor prescription for tree diagrams; we gave $d$ different rules,

1771: all of which agree up to overall prefactors.  Is this all one can say, or

1772: would a more sensitive study give more information

1773: about which is the ``correct'' prescription?  Does this proliferation of

1774: prescriptions persist beyond tree level?

1775: 

1776: \item {\bf Loops and higher genus.} 

1777: We only studied tree diagrams, corresponding to genus zero curves.  

1778: What are the exact rules and equivalences between various formulae for 

1779: loop and nonplanar amplitudes?

1780: Our analysis suggests that an investigation of possible degenerations of genus $g$ curves 

1781: should be relevant for the understanding of loop diagrams in the twistor string.

1782: 

1783: \end{itemize}

1784: 

1785: 

1786: 

1787: \section*{Acknowledgements}

1788: We are grateful to Michal Fabinger, Peter Svr\v{c}ek, Cumrun Vafa, Anastasia Volovich and 

1789: Edward Witten for very useful discussions.

1790: This work was conducted during the period S.G.\ served

1791: as a Clay Mathematics Institute Long-Term Prize Fellow.

1792: S.G. is also supported in part by RFBR grant 01-02-17488.

1793: The work of L.M.\ was supported in part by Harvard DOE grant

1794: DE-FG01-91ER40654 and the Harvard Society of Fellows.  The

1795: work of A.N.\ was supported by NSF grants PHY-0255841 

1796: and DMS-0244464.

1797: 

1798: \renewcommand{\baselinestretch}{1}

1799: \small\normalsize

1800: 

1801: \bibliography{physics}

1802: 

1803: \end{document}

1804: