1: %%      File: cpamshell.tex     Version: 2004/05/13 ( LaTeX2e[1994/06/01] )

2: %%      ---------------------------------------------------------------------

3: %%      Input file for Articles to be published in Communications on Pure and

4: %%      Applied Mathematics issued by The Courant Institute of Mathematical

5: %%      Sciences, New York University Published by John Wiley and Sons

6: 

7: %%      ---------------------------------------------------------------------

8: %%      ---------------------- PUBLISHER'S AREA -----------------------------

9: %%      ---------------------------------------------------------------------

10: 

11: %%      For input by typesetter.

12: %%      Authors, please do not alter the following section.

13: 

14: \documentclass{c-art2}     %% Input Standard CPAM class.

15: %\received{Month 200X}       %\revised{date of revision}

16:  %\yearofpublication{200X}

17: \volume{000}

18: \startingpage{1}

19: 

20: 

21: %\usepackage{pdfsync}

22: 

23: 

24: %User-ID: 0707.0442

25: %Password: 8ysgx 

26: 

27: 

28: \usepackage{amsthm}

29: 

30: %%      ----------------------- END OF PUBLISHER'S AREA ---------------------

31: 

32: % Insert names as you want them to appear on top of pages

33: \authorheadline{M.~Adler,  J.~Del\'epine, and  P.~van Moerbeke}

34: % Insert shortened title as you want them to appear on top of pages

35: \titleheadline{Airy process with outliers}

36: 

37: %%      ---------------------------------------------------------------------

38: %%      ------------------------- AUTHOR'S PACKAGES -------------------------

39: %%      ---------------------------------------------------------------------

40: 

41: %% Insert TeX and LaTeX packages that you would like to use here;

42: %% please note that amsmath is loaded by default and need not be specified.

43: 

44: % \usepackage{amssymb}

45: % \usepackage{graphicx}

46: % \usepackage{}

47: 

48: %%      ---------------------------------------------------------------------

49: %%      --------------------- CHOOSING THE DEFAULT FONTS --------------------

50: %%      ---------------------------------------------------------------------

51: 

52: % CPAM uses the Times Roman font; most installations can use the

53: % mathptmx package below, which will put almost all text in Times Roman

54: % except for some mathematical symbols. If that package doesn't work on

55: % your system, use the times package, which will put only running text

56: % and not mathematics in Times Roman (math will be in the default

57: % Computer Modern font). To use Times Roman for all math expressions,

58: % you need the Mathtime package; if you have that package installed,

59: % comment out or delete \usepackage{mathptmx} and uncomment the next

60: % instruction for using the Mathtime fonts. This is what CPAM uses. If

61: % none of these work, comment out all three instructions to use the

62: % default Computer Modern fonts.

63: 

64: % Uncomment one of the following if you have Times Roman fonts available

65: %\usepackage[mtbold,subscriptcorrection,T1]{mathtime}

66: \usepackage{mathptmx}

67: %\usepackage{times}

68: 

69: 

70: \usepackage{graphicx}

71: 

72: %%      ---------------------------------------------------------------------

73: %%      -------- COMMANDS FOR NUMBERED THEOREMLIKE ENVIRONMENTS--------------

74: %%      ---------------------------------------------------------------------

75: 

76: % This is the setup for AMS-LaTeX theorem environments; you may change

77: % the word in the first set of braces; this is the word you use to

78: % call up the environment in text. Do not change any other text in the

79: % list below.

80: 

81: \newtheorem{theorem}{Theorem}[section]

82: \newtheorem{lemma}[theorem]{Lemma}

83: \newtheorem{corollary}[theorem]{Corollary}

84: \newtheorem{proposition}[theorem]{Proposition}

85: % Do not remove the following line

86: \theoremstyle{definition}

87: \newtheorem{definition}[theorem]{Definition}

88: % Do not remove the following line

89: \theoremstyle{remark}

90: \newtheorem{remark}[theorem]{Remark}

91: 

92: %%      ---------------------------------------------------------------------

93: %%      -------------------- AUTHORS' MACROS ---------------------

94: %%      ---------------------------------------------------------------------

95: 

96: \newcommand{\rg}{\rightarrow}

97: \newcommand{\lrg}{\longrightarrow}

98: \newcommand{\Rg}{\Rightarrow}

99: \newcommand{\DF}{\Longleftrightarrow}

100: \newcommand{\df}{\longleftrightarrow}

101: \newcommand{\pp}{\ldots}

102: \newcommand{\TT}{\tilde\tau}

103: %\newcommand{\BR}{\mathcal {B}}

104: \newcommand{\AR}{\mathcal {A}}

105: \newcommand{\CR}{\mathcal {C}}

106: \newcommand{\DR}{\mathcal {D}}

107: \newcommand{\FR}{\mathcal {F}}

108: \newcommand{\GR}{\mathcal{G}}

109: \newcommand{\HR}{\mathcal {H}}

110: \newcommand{\JR}{\mathcal{J}}

111: \newcommand{\LR}{\mathcal {L}}

112: %\newcommand{\NR}{\mathcal{N}}

113: \newcommand{\OR}{\mathcal{O}}

114: \newcommand{\PR}{\mathcal{P}}

115: \newcommand{\SR}{\mathcal{S}}

116: \newcommand{\VR}{\mathcal{V}}

117: \newcommand{\WR}{\mathcal{W}}

118: \newcommand{\BB}{\mathcal {B}}

119: \newcommand{\BT}{\mathcal{T}}

120: \newcommand{\XR}{\mathcal{X}}

121: \newcommand{\BC}{\mathbb{C}}

122: \newcommand{\BD}{\mathbb{D}}

123: \newcommand{\BF}{\mathbb{F}}

124: \newcommand{\BH}{\mathbb{H}}

125: \newcommand{\BL}{\mathbb{L}}

126: \newcommand{\BP}{\mathbb{P}}

127: \newcommand{\BQ}{\mathbb{Q}}

128: %\newcommand{\BR}{\mathbb{R}}

129: \newcommand{\BS}{\mathbb{S}}

130: \newcommand{\BV}{\mathbb{V}}

131: \newcommand{\BX}{\mathbb{X}}

132: \newcommand{\BY}{\mathbb{Y}}

133: \newcommand{\BZ}{\mathbb{Z}}

134: \newcommand{\Sg}{\Sigma}

135: \newcommand{\iy}{\infty}

136: \newcommand{\pl}{\partial}

137: \newcommand{\al}{\alpha}

138: %\newcommand{\t}{\tau}

139: 

140: \newcommand{\tp}{\tilde\partial}

141: %\newcommand{\tx}{\tilde x}

142: \newcommand{\gs}{{\bf s}}

143: \newcommand{\no}{\nonumber}

144: \newcommand{\oiint}{\frac{1}{(2\pi

145: i)^2}\oint\limits_{\iy}\!\!\oint\limits_{\iy}}

146: \newcommand{\tx}{\tilde x}

147: \newcommand{\ty}{\tilde y}

148: \newcommand{\ba}{{\backslash}}

149: 

150: \newcommand{\acht}{{\rm ht}}

151: 

152: \newcommand{\shade}{/\!\!/\!\!/\!\!/\!\!%/\!\!

153: }

154: \newcommand{\bshade}{\backslash\!\!

155: \backslash\!\!\backslash\!\!\backslash%\!\!\backslash~~~

156: }

157: 

158: \def\inn#1#2{\left\langle#1\,\left\vert\,#2\right.\right\rangle}

159: \def\INN{\langle\cdot\,\vert\,\cdot\rangle}

160: 

161: \newcommand{\vv}{||}

162: 

163: 

164: \def\be#1\ee{\begin{equation}#1\end{equation}}

165: \def\bea#1\eea{\begin{eqnarray}#1\end{eqnarray}}

166: \def\bean#1\eean{\begin{eqnarray*}#1\end{eqnarray*}}

167: 

168: \newcommand{\Pf}{\operatorname{\rm Pfaff}}

169: \newcommand{\Tr}{\operatorname{\rm Tr}}

170: \newcommand{\Mat}{\operatorname{\rm Mat}}

171: \newcommand{\OneOrTwo}{{\left\{\substack{1\\2}\right\}}}

172: 

173: \newcommand{\om}{\omega}

174: \newcommand{\vp}{\varphi}

175: \newcommand{\la}{\langle}

176: \newcommand{\ra}{\rangle}

177: \newcommand{\ga}{\gamma}

178: \newcommand{\Ga}{\Gamma}

179: \newcommand{\dt}{\delta}

180: \newcommand{\Dt}{\Delta}

181:  \newcommand{\vr}{\varepsilon}

182: \newcommand{\sg}{\sigma}

183: \newcommand{\BR}{\mathbb{R}}

184: \newcommand{\lb}{\lambda}

185: \newcommand{\Lb}{\Lambda}

186: \newcommand{\tr}{\mbox{tr}}

187: \newcommand{\dis}{\displaystyle}

188: 

189: \newcommand{\Bn}{\mathfrak{n}}

190: \newcommand{\Bk}{\mathfrak{k}}

191: \newcommand{\BG}{\mathbb{G}}

192: \newcommand{\BK}{\mathbb{K}}

193: \newcommand{\BJ}{\mathbb{J}}

194: \newcommand{\BI}{\mathbb{I}}

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

196: \newcommand{\diag}{\operatorname{diag}}

197: \newcommand{\Res}{\operatorname{Res}}

198: 

199: % \new

200: 

201: % Please enter all macros that you have used in the body of the paper

202: % here. Use \newcommand for new macros and \renewcommand for

203: % redefining existing macros; avoid using \def if possible

204: 

205: %%      ---------------------------------------------------------------------

206: %%      -------------------------- BEGIN ARTICLE ----------------------------

207: %%      ---------------------------------------------------------------------

208: 

209: \newcommand\ig[1]{{}}

210: 

211: \begin{document}                        %% Standard LaTeX command

212: 

213: %%      ---------------------------------------------------------------------

214: %%      -------------------------------- TITLE -----------------------------

215: %%      ---------------------------------------------------------------------

216: 

217: \title{Dyson's Nonintersecting Brownian Motions \\ with a Few Outliers}

218: 

219: %%      ---------------------------------------------------------------------

220: %%      ------------------------------- AUTHORS -----------------------------

221: %%      ---------------------------------------------------------------------

222: \author{Mark Adler}{Brandeis University}

223: % EXAMPLE: \author{Paris Hilton}{Universit‰ Paris-Sorbonne (Paris IV)}

224: % Uncomment the following lines as needed

225: \author{Jonathan Del\'epine}{Universit\'e Catholique de Louvain}

226:  \author{Pierre van Moerbeke}{Universit\'e Catholique de Louvain \& Brandeis University}

227: %\author{*** FOURTH AUTHOR'S NAME ***}{*** FOURTH AUTHOR'S AFFILIATION WHEN ARTICLE WAS WRITTEN ***}

228: %\author{*** FIFTH AUTHOR'S NAME ***}{*** FIFTH AUTHOR'S AFFILIATION WHEN ARTICLE WAS WRITTEN ***}

229: % Add additional names and affiliations as necessary using above format

230: %%      ---------------------------------------------------------------------

231: %%      --------------------------- DEDICATION  (OPTIONAL)-------------------

232: %%      ---------------------------------------------------------------------

233: 

234: %       Uncomment the following line to insert a dedication.

235: 

236: %\dedication{ *** DEDICATION *** }        %% Enter dedication between braces.

237: 

238: 

239: %%      ---------------------------------------------------------------------

240: %%      --------------------------- ABSTRACT (OPTIONAL)----------------------

241: %%      ---------------------------------------------------------------------

242: 

243: %% ***** UNCOMMENT THE FOLLOWING TO INSERT AN ABSTRACT

244: 

245:  \begin{abstract}

246:  Consider $n$ non-intersecting \mbox{Brownian} particles

247: on $\BR$ (Dyson Brownian motions), all starting from the

248:  origin at time $t=0$, and forced to return to $x=0$

249:  at time $t=1$. For large $n$, the average mean density of particles

250: has its support, for each $0<t<1$, on the interval $\pm

251: \sqrt{2nt(1-t)}$. The Airy process $\mathcal{A}(\tau)$

252: is

253:  defined as the motion of these non-intersecting

254:  Brownian motions for large $n$, but viewed from the

255:  curve $\mathcal{C}:\quad y=\sqrt{2nt(1-t)}$ with

256:  an appropriate space-time rescaling.

257: %

258:  Assume now a finite number $r$ of these particles are

259: forced to a different target point, say

260: $a=\rho_0\sqrt{n/2}>0$. Does it affect the Brownian

261: fluctuations along the curve $\mathcal{C}$ for large

262: $n$? In this paper, we show that no new process appears

263: as long as one considers points $(y,t)\in \mathcal{C},

264: \mbox{ such that } 0<t<({1+\rho_0^2})^{-1}

265:  $, which is the $t$-coordinate of

266: the point of tangency of the tangent to the curve

267: passing through $(\rho_0\sqrt{n/2},1)$. At this point of tangency 

268: the fluctuations obey a new statistics, which we call

269: the {\bf Airy process with $r$ outliers} $\mathcal

270: {A}^{(r)}(\tau)$ (in short: {\bf $r$-Airy process}).

271: %

272:    The log of the probability

273:    that at time $\tau$ none of the particles in the 

274:  cloud exceeds $x$ is given by the Fredholm

275:  determinant of a new kernel (extending the Airy kernel)

276:  and it satisfies a non-linear PDE in $x$ and $\tau$,

277:  from which the asymptotic behavior of the process can be

278:  deduced for $\tau\rg -\iy$. This kernel is closely

279:  related to one found by Baik, Ben Arous and P\'ech\'e in

280:  the context of multivariate statistics.

281: 

282: 

283:  \end{abstract}

284: 

285: % With AMS-LaTeX, \maketitle follows the abstract

286: \maketitle

287: 

288: %%      ---------------------------------------------------------------------

289: %%      ------------------- TABLE OF CONTENTS (OPTIONAL) --------------------

290: %%      ---------------------------------------------------------------------

291: 

292: %% ***** IF YOUR PAPER IS OVER 40 PAGES AND YOU WISH TO HAVE A TABLE

293: %% ***** OF CONTENTS, PLEASE UNCOMMENT THE FOLLOWING LINE

294: 

295:   \tableofcontents

296: 

297: \ig{

298: \newcommand{\MAT}[1]{\left(\begin{array}{*#1c}}

299: \newcommand{\mat}{\end{array}\right)}

300: \newcommand{\qed}{\leavevmode\unskip\nobreak\penalty200\hskip2pt\null

301: \nobreak\hfill\rule{1.1ex}{1.1ex}%\parfillskip=0pt

302:  \medbreak }

303: 

304: \newcommand{\rg}{\rightarrow}

305: \newcommand{\lrg}{\longrightarrow}

306: \newcommand{\Rg}{\Rightarrow}

307: \newcommand{\DF}{\Longleftrightarrow}

308: \newcommand{\df}{\longleftrightarrow}

309: \newcommand{\pp}{\ldots}

310: \newcommand{\TT}{\tilde\tau}

311: \newcommand{\AR}{\mathcal {A}}

312: \newcommand{\CR}{\mathcal{C}}

313: \newcommand{\DR}{\mathcal{D}}

314: \newcommand{\FR}{\mathcal{F}}

315: \newcommand{\GR}{\mathcal{G}}

316: \newcommand{\HR}{\mathcal {H}}

317: \newcommand{\JR}{\mathcal{J}}

318: \newcommand{\LR}{\mathcal{L}}

319: \newcommand{\MR}{\mathcal{M}}

320: \newcommand{\NR}{\mathcal{N}}

321: \newcommand{\OR}{\mathcal{O}}

322: \newcommand{\PR}{\mathcal{P}}

323: \newcommand{\SR}{\mathcal{S}}

324: \newcommand{\VR}{\mathcal{V}}

325: \newcommand{\WR}{\mathcal{W}}

326: \newcommand{\BB}{\mathcal{B}}

327: \newcommand{\BT}{\mathcal{T}}

328: \newcommand{\XR}{\mathcal{X}}

329: \newcommand{\BC}{\mathbb{C}}

330: \newcommand{\BD}{\mathbb{D}}

331: \newcommand{\BF}{\mathbb{F}}

332: \newcommand{\BH}{\mathbb{H}}

333: \newcommand{\BL}{\mathbb{L}}

334: \newcommand{\BP}{\mathbb{P}}

335: \newcommand{\BQ}{\mathbb{Q}}

336: \newcommand{\BS}{\mathbb{S}}

337: \newcommand{\BV}{\mathbb{V}}

338: \newcommand{\BX}{\mathbb{X}}

339: \newcommand{\BY}{\mathbb{Y}}

340: \newcommand{\BZ}{\mathbb{Z}}

341: \newcommand{\Sg}{\Sigma}

342: \newcommand{\iy}{\infty}

343: \newcommand{\pl}{\partial}

344: \newcommand{\al}{\alpha}

345: %\newcommand{\t}{\tau}

346: 

347: \newcommand{\tp}{\tilde\partial}

348: %\newcommand{\tx}{\tilde x}

349: \newcommand{\gs}{{\bf s}}

350: \newcommand{\no}{\nonumber}

351: \newcommand{\oiint}{\frac{1}{(2\pi

352: i)^2}\oint\limits_{\iy}\!\!\oint\limits_{\iy}}

353: \newcommand{\tx}{\tilde x}

354: \newcommand{\ty}{\tilde y}

355: \newcommand{\ba}{{\backslash}}

356: 

357: \newcommand{\acht}{{\rm ht}}

358: 

359: \newcommand{\shade}{/\!\!/\!\!/\!\!/\!\!%/\!\!

360: }

361: \newcommand{\bshade}{\backslash\!\!

362: \backslash\!\!\backslash\!\!\backslash%\!\!\backslash~~~

363: }

364: 

365: 

366: \def\inn#1#2{\left\langle#1\,\left\vert\,#2\right.\right\rangle}

367: \def\INN{\langle\cdot\,\vert\,\cdot\rangle}

368: 

369: \newcommand{\vv}{||}

370: 

371: % \newcommand{\proof}{\underline{\sl Proof}: }

372: % \newcommand{\remark}{\underline{\sl Remark}: }

373: % \newcommand{\example}{\underline{\sl Example}: }

374: 

375: \newenvironment{proof}{\medskip\noindent{\it Proof:\/} }{\qed}

376: \newenvironment

377:         {remark}{\medskip\noindent\underline{\it Remark:\/} }{\medbreak}

378: \newenvironment

379:         {example}{\medskip\noindent\underline{\it Example:\/} }{\medbreak}

380: 

381: \newcommand{\om}{\omega}

382: \newcommand{\vp}{\varphi}

383: \newcommand{\la}{\langle}

384: \newcommand{\ra}{\rangle}

385: \newcommand{\ga}{\gamma}

386: \newcommand{\Ga}{\Gamma}

387: \newcommand{\dt}{\delta}

388: \newcommand{\Dt}{\Delta}

389:  \newcommand{\vr}{\varepsilon}

390: \newcommand{\sg}{\sigma}

391: \newcommand{\BR}{\mathbb{R}}

392: \newcommand{\lb}{\lambda}

393: \newcommand{\Lb}{\Lambda}

394: \newcommand{\tr}{\mbox{tr}}

395: \newcommand{\dis}{\displaystyle}

396: 

397: \newcommand{\Bn}{\mathfrak{n}}

398: \newcommand{\Bk}{\mathfrak{k}}

399: \newcommand{\BG}{\mathbb{G}}

400: \newcommand{\BK}{\mathbb{K}}

401: \newcommand{\BJ}{\mathbb{J}}

402: \newcommand{\BI}{\mathbb{I}}

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

404: \newcommand{\diag}{\operatorname{diag}}

405: \newcommand{\Res}{\operatorname{Res}}

406: 

407: % some macros are not liked by amsmath package:

408: 

409: % \newcommand{\be}{\begin{equation}}

410: % \newcommand{\ee}{\end{equation}}

411: % \newcommand{\bea}{\begin{eqnarray}}

412: % \newcommand{\eea}{\end{eqnarray}}

413: 

414: % so replace:

415: 

416: \def\be#1\ee{\begin{equation}#1\end{equation}}

417: \def\bea#1\eea{\begin{eqnarray}#1\end{eqnarray}}

418: \def\bean#1\eean{\begin{eqnarray*}#1\end{eqnarray*}}

419: 

420: \newcommand{\Pf}{\operatorname{\rm Pfaff}}

421: \newcommand{\Tr}{\operatorname{\rm Tr}}

422: \newcommand{\Mat}{\operatorname{\rm Mat}}

423: \newcommand{\OneOrTwo}{{\left\{\substack{1\\2}\right\}}}

424: 

425: % \newtheorem{definition}{Definition}[section]

426: % \newtheorem{theorem}[definition]{Theorem}

427: % \newtheorem{lemma}[definition]{Lemma}

428: % \newtheorem{corollary}[definition]{Corollary}

429: % \newtheorem{proposition}[definition]{Proposition}

430: 

431: %\makeatletter

432: %\def\ps@X{\let\@mkboth\@gobbletwo

433: %        \def\@oddhead{\tt A-K-vM:%

434: %        Pfaff \& Jack\hfil \today\hfil\S\thesection, p.\thepage}

435: %       \def\@oddfoot{\rm\hfil\thepage\hfil}

436: %       \let\@evenhead\@oddhead

437: %       \let\@evenfoot\@oddfoot}

438: %\makeatother

439: 

440: %\textwidth16.1cm

441:  %\textheight228mm

442:  %\oddsidemargin-0mm    %-0mm

443:  %\evensidemargin-0mm%-5mm

444:  %\voffset-12mm

445: 

446: %beginning of macros

447: 

448: %Make ! a letter so we can use it in the names of control sequences

449: \catcode `!=11

450: 

451: \newdimen\squaresize

452: \newdimen\thickness

453: \newdimen\Thickness

454: \newdimen\ll! \newdimen \uu! \newdimen\dd! \newdimen \rr! \newdimen

455: \temp!

456: 

457: %parameters are left up, down, right

458: \def\sq!#1#2#3#4#5{%

459: \ll!=#1 \uu!=#2 \dd!=#3 \rr!=#4

460: \setbox0=\hbox{%

461: %left edge

462:  \temp!=\squaresize\advance\temp! by .5\uu!

463:  \rlap{\kern -.5\ll!

464:  \vbox{\hrule height \temp! width#1 depth .5\dd!}}%

465: %

466: %up edge

467:  \temp!=\squaresize\advance\temp! by -.5\uu!

468:  \rlap{\raise\temp!

469:  \vbox{\hrule height #2 width \squaresize}}%

470: %

471: %down edge

472:  \rlap{\raise -.5\dd!

473:  \vbox{\hrule height #3 width \squaresize}}%

474: %

475: %right edge

476:  \temp!=\squaresize\advance\temp! by .5\uu!

477:  \rlap{\kern \squaresize \kern-.5\rr!

478:  \vbox{\hrule height \temp! width#4 depth .5\dd!}}%

479: %

480: %contents

481:  \rlap{\kern .5\squaresize\raise .5\squaresize

482:  \vbox to 0pt{\vss\hbox to 0pt{\hss $#5$\hss}\vss}}%

483: }%end of \hbox

484:  \ht0=0pt \dp0=0pt \box0

485: }%end of \sq!

486: 

487: 

488: \def\vsq!#1#2#3#4#5\endvsq!{\vbox to \squaresize{\hrule

489: width\squaresize height 0pt%

490: \vss\sq!{#1}{#2}{#3}{#4}{#5}}}

491: 

492: \newdimen \LL! \newdimen \UU! \newdimen \DD! \newdimen \RR!

493: 

494: \def\vvsq!{\futurelet\next\vvvsq!}

495: \def\vvvsq!{\relax

496:   \ifx     \next l\LL!=\Thickness \let\continue=\skipnexttoken!

497:   \else\ifx\next u\UU!=\Thickness \let\continue=\skipnexttoken!

498:   \else\ifx\next d\DD!=\Thickness \let\continue=\skipnexttoken!

499:   \else\ifx\next r\RR!=\Thickness \let\continue=\skipnexttoken!

500:   \else\def\continue{\vsq!\LL!\UU!\DD!\RR!}%

501:   \fi\fi\fi\fi

502:   \continue}

503: 

504: \def\skipnexttoken!#1{\vvsq!}

505: 

506: \def\place#1#2#3{\vbox to 0pt{\vss

507: \rlap{\kern#1\squaresize

508:   \raise#2\squaresize\hbox{$#3$}}

509: \vss}}

510: 

511: 

512: \def\Young#1{\LL!=\thickness \UU!=\thickness \DD! = \thickness \RR! =

513: \thickness \vbox{\smallskip\offinterlineskip

514: \halign{&\vvsq! ##

515: \endvsq!\cr #1}}}

516: 

517: \def\blank{\omit\hskip\squaresize}

518: \catcode `!=12

519: 

520: 

521: \squaresize = 35pt \thickness = 1pt \Thickness = 3pt

522: %end of macro's

523: 

524: }

525: 

526: 

527: Dyson \cite{Dyson} made the important observation that

528: putting dynamics into random matrix models leads to finitely many

529: non-intersecting Brownian motions (on ${\mathbb R}$) for the eigenvalues.

530: Applying scaling limits to the random matrix models,

531: combined with Dyson's dynamics, then leads to

532: interesting infinitely many diffusions for the

533: eigenvalues. This paper studies a model, which stems

534: from multivariate statistics and which interpolates

535: between the {\bf Airy} and {\bf Pearcey} processes.

536: 

537: Consider $n$ non-intersecting \mbox{Brownian} particles on the

538: real line $\BR$,

539: %

540: $$

541: -\iy <x_1(t)<\ldots<x_n(t)<\iy,$$

542:  with (local) Brownian transition probability given by

543: \be p(t ; x, y)  :=  \frac{1}{\sqrt{\pi t}} e^{-

544: \frac{(y-x)^{2}}{t}}, \label{Normal}\ee  all starting

545: from the origin $x=0$ at time $t=0$, and forced to

546: return to $x=0$ at time $t=1$.

547: For very large $n$, the average mean density of particles

548: has its support, for each $0<t<1$, on the interval

549: $(-\sqrt{2nt(1-t)},\sqrt{2nt(1-t)})$, as sketched in Figure \ref{fig1}.

550: 

551: The {\bf Airy process} ${A}(\tau)$ is defined as the

552: motion of these non-intersecting Brownian motions for

553: large $n$, but viewed from an observer on the (right

554: hand) edge-curve $$\mathcal{C}:\quad

555: \{y=\sqrt{2nt(1-t)}>0\}$$ of the set of particles, with

556: space stretched by the customary GUE-edge rescaling $

557: n^{1/6}$ and time rescaled by the factor $ n^{1/3}$ in

558: tune with the Brownian motion space-time rescaling; this

559: is to say that in this new scale, slowed down

560: microscopically, the left-most particles appear

561: infinitely far and the time horizon $t=1$ lies in the

562: very remote future. Thus, the Airy process describes the

563: fluctuations of the Brownian particles near the

564: edge-curve $\mathcal{C}$, looked at through a magnifying glass, as shown in Figure  \ref{fig1}.

565: %Thus the maximum $\sqrt{n/2}$ of this edge lies at $t=1/2$.

566: 

567:  The Airy process was introduced by Pr\"ahofer and Spohn \cite{Spohn} and further

568:   investigated in

569:   \cite{Johansson2,Johansson3,TW-Dyson,AvM-Airy-Sine}.

570:   Notice that

571: in this work the Airy process is not viewed as the

572: motion of the largest particle (point process), but as

573: the motion of the cloud of particles, which will be

574: described as a determinantal process. Giving

575: a pathwise description of this motion remains an open

576: problem.

577: 

578: \begin{figure}

579: \hspace*{-2.9cm}\includegraphics[width=180mm,height=120mm]{Armenia_Airy.pdf}

580: 

581: \vspace{-3.5cm}

582: 

583: \caption{Airy process}\label{fig1}

584: \end{figure}

585: 

586:  \vspace*{-4cm}

587: 

588: $$\hspace*{9cm}\mathcal{C}:=\{x=\sqrt{2nt(1-t)}\}$$

589: 

590: \vspace*{3cm}

591: 

592: %Consider a collection of $n$ non-intersecting

593: %\mbox{Brownian} particles on ${\mathbb R}$,

594: %  with transition probability given by

595: %$$

596: %p(t ; x, y)  :=  \frac{1}{\sqrt{\pi t}} e^{-

597: %\frac{(y-x)^{2}}{t}},  $$ all starting from the origin

598: %at time \,$t=0$, \,and evolving up to time \,$t=1$.

599: Assume now that, among those $n$ paths, $0\leq r\leq n$

600: are forced to reach a given final target $a \geq 0$,

601: \,while the $(n-r)$ remaining particles return to the

602: position $x=0$. Consider the probability that, at any

603: given time $0 < t < 1$ , all of the particles avoid a

604: window $E \subset \BR$, namely (the superscript in the

605: probability $\BP_{Br}^{0a}$ here and later refers to the

606: target points)

607: %

608: \be  \BP_{Br}^{0a}\left(\mbox{all  $x_j(t)\in

609: E^c$}\right)  := \! \BP\!\!

610: \left(\!\!\!\begin{tabular}{c|c}

611: & all $x_j(0) =0$\\

612: all $x_j(t)\in E^c$ &

613:  $r$  right paths end up at  $a$  at $t=1$\\

614:   &

615: $(n-r)$ paths end up at $0$ at $t=1$

616: \end{tabular}\!\!\!\right).

617:   \label{Brownian target}\ee

618: %

619: % \\

620: %

621: %

622: %

623: %

624: %\medbreak

625: %

626: %

627: Does the fact that a finite number $r$ of particles are

628: forced to a different target point, in particular the

629: target point $a=\rho_0\sqrt{n/2}>0$ for some arbitrary

630: parameter $\rho_0$, affect the Brownian fluctuations

631: along the curve $\mathcal{C}$ for very large $n$? It is

632: understood here that near the points of the curve (under

633: consideration), one uses the same scaling as the Airy

634: process. In this paper, we show that no new process

635: appears as long as one considers points \be (y,t)\in

636: \mathcal{C}, \mbox{ such that }

637: 0<t<\frac{1}{1+\rho_0^2}.

638:  \label{curve}\ee

639: Observe that $t=t_0=(1+\rho_0^2)^{-1}$ corresponds to the point of tangency $(y_0,t_0)$ of the tangent to the curve  passing through the

640:    point $(\rho_0\sqrt{n/2},1)$ of the $(t=1)$-axis; i.e.

641:  \be

642:  (y_0,t_0)= \left(\frac{\rho_0 \sqrt{2n}}{1+\rho_0^2},

643:  \frac{1}{ 1+\rho_0^2 }\right)\in  \mathcal{C}

644: .\label{tangency}

645:  \ee

646:    %

647:    At this point of tangency $(y_0,t_0)$ the fluctuations obey a new

648: statistics, which we call the {\bf Airy process with

649: $r$ outliers} $\mathcal {A}^{(r)}(\tau)$

650:   (in short: {\bf $r$-Airy process}); see Figure 0.2

651:     In particular, for

652: the target point $a=\sqrt{n/2}$, the $r$-Airy process

653: occurs at the the maximum of the edge curve

654: $\mathcal{C}$; i.e., $\rho_0=1$ and thus $t_0=1/2$.

655: %

656:  Notice, the $r$-Airy process is an extension of the Airy

657: process; the $r$-Airy process coincides with the Airy

658: process, when the

659:   target point $a$ coincides with $0$ or,

660:   what is the same, when $r=0$. The Airy process

661:   is stationary, whereas the $r$-Airy process ceases to be stationary.

662: %     \begin{figure}

663: %\includegraphics[width=90mm,height=85mm]{r_airy.pdf}

664: %\caption{The $r$-Airy process}\label{fig2}

665: %\end{figure}

666: %

667: \begin{figure}

668: \hspace*{-1cm}\includegraphics[width=170mm,height=105mm]{r-AiryProcess.pdf}

669: 

670: 

671: \vspace*{-2.9cm}

672: 

673: \caption{The $r$-Airy process}\label{fig2}

674: \end{figure}

675: 

676:  %\vspace*{-2cm}

677:  

678:  

679: Given a target point $a=\rho_0\sqrt{n/2}$,

680:   the point of tangency of the tangent to the curve

681:   $\mathcal{C}$, passing through $a$, can be written, in

682: accordance with (\ref{tangency}),

683:    \be  (y_0,t_0)=\left(\frac{\rho_0 \sqrt{2n}}{1+\rho_0^2},

684:  \frac{1}{ 1+\rho_0^2 }\right)=(y_0,t_0)= \left(\frac{\sqrt{2n}    }

685:    {2\cosh\tau_0 }  ,\quad  \frac{e^{ \tau_0 }}

686:  {2\cosh \tau_0    }\right)\in \mathcal{C}

687: %

688:    %

689:   ,\label{y0,t0}\ee

690: upon introducing a new parameter $\tau_0$, whose

691: significance as a new time parameter will be clear from

692: section \ref{sect1} (formula (\ref{time-change})), and

693: which is defined by

694: %

695:    \be \label{rho0}  e^{ -\tau_0}:=\rho_0=\sqrt{\frac{1-t_0}{t_0}}.\ee

696: Given a subset $E\subset {\mathbb R}$, the $r$-Airy

697: process in the new time $\tau$ will be obtained by taking the following limit:%

698:   \bea

699:   \lefteqn{ {~~~~~~~~~\BP(\mathcal {A}^{(r)}(\tau)\cap E%(x,\iy)

700:   =\emptyset)}}%\no

701:    \label{r-AiryP}\\&&

702:   \! := \!

703:   %

704:    \lim_{n\rightarrow \iy}

705: \BP^{(0,\rho_0\sqrt{n/2})}_{Br}\left(\mbox{all}~

706: x_i\left( \frac{1}

707:  {1+e^{-2(\tau_0+\frac{\tau}{n^{1/3}})}}\right) \in

708:   \frac{\sqrt{2n}+\frac{E^c}{\sqrt{2}n^{1/6}}   }

709:    {2\cosh(\tau_0+\frac{\tau}{n^{1/3}})}

710:   % ~~\mbox{with}~e^{ \al_0}=\sqrt{\frac{s_0}{1-s_0}}

711:    \right)

712:  .\no\eea

713: %

714:  Notice that for $\tau=0$ and upon ignoring the set $E^c$, space and time on the right hand side of the formula above

715:   equal $(y_0,t_0)$ as in (\ref{y0,t0}). In section \ref{sect2}, the limit above will be shown to exist, independently of the parameter $\rho_0$, at the same time establishing some universality.

716: 

717: 

718:  % \noindent

719:   In order to state Theorem \ref{Theo1} below, define the functions\footnote{$C$ is a contour running from $\iy

720:  e^{5i\pi/6}$ to $\iy

721:  e^{i\pi/6}$, such that $-i\tau$ lies above the contour.}

722:  \be

723:  A^{\pm}_r(u;\tau):=

724:   \int_C   e^{\frac{1}{3} ia^3+iau }

725:  \left(\mp ia-\tau\right)^{\pm r}

726:  \frac{da}{2\pi}

727:   \label{k-Airy0}\ee

728: %

729: and the standard

730: Airy function $A(u):=A^-_0(u;\tau)=A^+_0(u;\tau)$,  satisfying the differential equation

731:  $

732: A'' (x)=xA(x).$

733: Given these functions, the Airy and $r$-Airy kernels %appearing in

734: %(\ref{0.3}) 

735:  are defined by (\ref{r-kernel}) below and will be used in Theorem \ref{Theo1}:

736:    \begin{equation}\label{r-kernel}\begin{split}

737:     K^{(0)}(u,v)&= \int_0^{\iy}

738:  dw A(w+u)A(w+v)=\frac{A(u)A'(v)-A'(u)A(v)}{u-v}

739:  \\

740:  K^{(r)}_\tau(u,v)&=\int_0^{\iy}

741:  dw A_r^-(w+u; \tau)A_r^+(w+v; \tau). 

742:  \end{split}\end{equation}

743:  We now state:

744: 

745: 

746: 

747: \begin{theorem} \label{Theo1}

748: Consider non-intersecting Brownian motions as above with

749: $r$ particles forced to a target point

750: $\rho_0\sqrt{n/2}>0$ at time $t=1$. For large $n$, the

751: average mean density of particles as a function of $t$

752: has its support on a region bounded to the right by the

753: curve $ \mathcal{C}$ (defined just below (\ref{Normal})). The

754: tangent line to $ \mathcal{C}$, passing through

755: $(\rho_0\sqrt{n/2},1)$ has its point of tangency at

756: $(y_0,t_0)$, given by (\ref{tangency}). Letting $n\rg

757: \iy$ and given an arbitrary point $(y,t)\in \CR$ with

758: $t\leq t_0$, a phase transition occurs at $t=t_0$, which

759: is conveniently expressed in terms of the new parameters

760: defined by\footnote{Remember from (\ref{rho0}), one has

761: $\rho_0=e^{-\tau_0}$.} $t=(1+e^{-2\sg})^{-1}$ and

762: $t_0=(1+e^{-2\tau_0})^{-1}$; one has the following limit:

763: %

764: % and experiences a phase transition at

765: %$\sigma=\tau_0$. Namely, given an arbitrary point

766: %$(y,t)\in \CR$ with $t\leq t_0$,

767: % $$  (y,t)=%\left(\frac{\rho_0 \sqrt{2n}}{1+\rho_0^2},

768: %% \frac{1}{ 1+\rho_0^2 }\right)=

769: % \left(\frac{\sqrt{2n}    }

770: %   {2\cosh(\sg)}  ,\quad  \frac{1}

771: % {1+e^{-2 \sg }}\right)\in \mathcal{C},\quad \mbox{with $t\leq t_0$},

772: %

773:    %

774: % $$% ,\label{y0,t0}\ee

775: %

776: %

777: %one has

778:  \bea \lefteqn{\label{0.3}

779:    \qquad\lim_{n\rightarrow \iy}

780: \BP^{(0,\rho_0\sqrt{n/2})}_{Br}\left(\mbox{all}~

781: x_i\left( \frac{1}

782:  {1+e^{-2(\sigma+\frac{\tau}{n^{1/3}})}}\right) \in

783:   \frac{\sqrt{2n}+\frac{E^c}{\sqrt{2}n^{1/6}}   }

784:   {2\cosh(\sigma+\frac{\tau}{n^{1/3}})}

785:   % ~~\mbox{with}~e^{ \al_0}=\sqrt{\frac{s_0}{1-s_0}}

786:    \right) } \\

787:   %&=&\BP(\mathcal {A}^{(r)}(t)\cap (x,\iy)=\emptyset)\\

788:   %

789:  %

790:  \no\\

791:  %

792: &=&\left\{\begin{array}{lll}

793:  \BP(\mathcal {A}(\tau)\cap E=\emptyset)=

794:  \det\left(I-K^{(0)}\right)_E

795:   &~~\mbox{for}\quad  0\leq t<t_0

796:   &\\

797:   & \quad\quad(\mbox{ i.e., }0 \leq\sigma<\tau_0) ,\\

798:  \\

799:  \BP(\mathcal {A}^{(r)}(\tau)\cap E=\emptyset)=

800:   \det\left(I-K_\tau^{(r)}\right)_E

801:   &\mbox{for}\quad t=t_0 \quad&\\

802:   & \qquad(\mbox{ i.e., }

803:   \sigma=\tau_0)

804:  . \end{array}\right. %\label{xx}

805:    \no%

806:    \eea

807: %

808: These Fredholm determinants are genuine probability distributions for each value of $\tau$.  Notice that for $0 \leq\sigma<\tau_0 $, the

809: limit above is independent of the time $\tau$ and the number of outliers $r$, unlike the case $\sigma=\tau_0$.

810:  

811:     

812: 

813: 

814: \end{theorem}

815: 

816: 

817: \begin{figure}

818: \includegraphics[width=95mm,height=90mm]{asym_pearcy.pdf}

819: 

820: \vspace*{-1cm}

821: 

822: \caption{Pearcey process}\label{fig3}

823: \end{figure}

824: 

825: 

826: %\newpage

827: 

828: 

829:   \bigbreak

830: 

831: As is clear from Figure 0.2 and for $t>t_0$, the $r$

832: outliers separate from the bulk as a group; thus after

833: an appropriate shift to take into account their

834: displacement from the curve $\mathcal{C}$ and using a

835: different scaling, they behave probabilistically like

836: the $r$ eigenvalues of an $r \times r$ matrix in the GUE

837: ensemble.

838: 

839: 

840:   The $r$-Airy process can also be viewed as an {\bf interpolation}

841:   between the Airy and Pearcey processes. The {\bf Pearcey process}

842:   \cite{TW-Pearcey, Okounkov, AvM-Pearcey,AvM-Pearcey-joint} is

843:   defined as the limit of non-intersecting Brownian motions (for large

844:   $n$), all leaving from $0$ at time $t=0$, with $(1-p)n$ paths forced

845:   to end up at $0$ and $pn$ paths forced to end up at $a\sqrt{n/2}$ at

846:   time $t=1$; see Figure 0.3. The Pearcey process then

847: describes this Brownian cloud of particles (for $n\rg

848: \iy$) near the time $t$ of bifurcation, where the

849: support of the average mean density goes from one

850: interval into two intervals, with stretched space and

851: time. Then the boundary of the support of the average

852: mean density of particles in $(y,t)$-space has a cusp.

853: In section \ref{inter} it will be shown sketchily how

854: the location of the cusp, when the proportion $p$ of

855: particles tends to $0$ as $r/n$, tends to the precise

856: place $(y_0,t_0)$ where the $r$-Airy process occurs.

857: 

858: 

859: \begin{remark} To simplify the notation, one often writes

860:  \be

861: \BP(\sup\mathcal {A}^{(r)}(\tau)\leq x)

862: : =\BP( \mathcal{A}^{(r)}(\tau)

863:  \cap (x,\iy)=\emptyset)\label{notation}

864:   .\ee

865: \end{remark}

866: 

867:  % \includegraphics[width=90mm,height=81mm]{PvM2.jpg}

868:    % ~\hspace*{3cm}\includegraphics[width=66mm,height=93mm]{Pearcey.jpg}

869: 

870: % \vspace{-4cm}

871: % \

872: % \hspace*{7cm}~

873: %Pearcey Process replace picture!!!

874: 

875: %\begin{figure}

876: %\includegraphics[width=95mm,height=90mm]{asym_pearcy.pdf}

877: %\caption{Pearcey process}\label{fig3}

878: %\end{figure}

879: 

880:  %\includegraphics[width=180mm,height=162mm]{r_airy.pdf}

881: 

882: 

883: \begin{remark} The joint probabilities for the $r$-Airy

884: processes for a finite number of times can be defined in a similar way and lead to a

885: matrix Fredholm determinant, to be discussed in a later

886: paper.

887: \end{remark}

888: 

889: 

890: 

891: This phenomenon is closely related to statistical work

892: by Baik-Ben Arous-P\'ech\'e \cite{BBP},

893: Baik \cite{Baik} and P\'ech\'e \cite{Peche}.%\cite{BBP}.

894:  ~Indeed, consider a (complex) Gaussian population

895: $\overrightarrow{y}\in \BC^N$, with covariance matrix

896: $\Sigma$. Given $M$ samples

897: $\overrightarrow{y_1},\ldots,\overrightarrow{y_M}$, the

898: (centered) sample covariance matrix $S:=\frac{1}{M}

899: X\bar X^{\top}$, where

900: $$

901:  X=\left(

902:  \overrightarrow{y_1}-\frac{1}{M} \sum_1^M

903:  \overrightarrow{y_i},\ldots,

904:  \overrightarrow{y_M}-\frac{1}{M} \sum_1^M

905:  \overrightarrow{y_i}\right)

906:  ,$$

907:  is a positive definite matrix and is an estimator of the true covariance matrix

908: $\Sigma$. One may test the statistical hypothesis that

909: $\Sigma=I$ or that $\Sigma$ has all eigenvalues $=1$,

910: except for a few outliers. When all the eigenvalues of

911: $\Sigma$ are $=1$, then the limit distribution of the

912: largest eigenvalue of the sample covariance matrix $S$,

913: for $N$ and $M$ tending to $\iy$ in the same way, is

914: given by the Tracy-Widom distribution. Then

915: Baik-BenArous-P\'ech\'e \cite{BBP} noticed that, this is

916: still so, if the eigenvalues of $\Sigma$ contain some

917: outliers, which are not too large. There is a critical

918: point at which and beyond which the largest eigenvalue

919: of the sample covariance matrix $S$ will be different

920: from the Tracy-Widom distribution. At this point of

921: phase transition Baik-BenArous-P\'ech\'e \cite{BBP} have

922: found an Airy-type distribution in $x$, which is given

923: by the Fredholm determinant $$ \det

924: {(I-K_\tau^{(r)})}_{_{(x,\iy)}}\Bigr|_{\tau=0} ,$$ where

925: $r$ denotes the number of eigenvalues of $\Sigma$ that

926: are equal to $1+\gamma^{-1}$, while all the others are

927: $=1$; $\gamma$ is such that $M/N=\gamma^2$ for $M$ and

928: $N$ very large. This distribution was further

929: generalized in \cite{BBP} to the case where $\tau \neq

930: 0$ in the kernel $K_\tau^{(r)}$. Baik proved in

931: \cite{Baik} that the Fredholm determinant of

932: $K_\tau^{(r)}$ is a genuine probability distribution. In

933: the statistical problem above, the covariance matrix $S$

934: is positive definite and therefore its eigenvalues

935: satisfy Laguerre-type distributions; this idea was

936: extended to GUE-type distributions by S. P\'ech\'e

937: \cite{Peche}. In the present paper one finds that the

938: shift $\tau$ appearing in the kernel $K_\tau^{(r)}$ is

939: precisely the rescaled time of the non-intersecting

940: Brownian motion model! The arguments will appear in

941: sections \ref{sect1} and \ref{sect2}.

942: 

943: 

944: % the  work and work of P\'ech\'e

945: %show that in the Brownian motion model above the shift

946: %$\tau$ in the $K_\tau^{(r)}$ is precisely the rescaled

947: %time of the non-intersecting Brownian motion model.

948: 

949: This paper also shows that the probability of the

950: $r$-Airy process or, what is the same, the Fredholm

951: determinant $\det\left(I-K_\tau^{(r)}\right)_{(x,\iy)}$

952: satisfies a non-linear PDE in $x$ and $\tau$, depending

953: on the number $r$ of outliers, as established in section

954: \ref{sect6}:

955: 

956: 

957: %Note the PDE \ref{PDE0} is invariant under the

958: %involution $(r,\tau,x)\rg (-r,-\tau,x)$

959: 

960: \begin{theorem}\label{Theo2}

961:  The logarithm of the probability

962:  $$Q(\tau,x):=\log\BP(\sup\mathcal {A}^{(r)}(\tau)\leq x)

963:  = \log\det {(I-K_\tau^{(r)})}_{_{(x,\iy)}}$$

964: satisfies the following non-linear PDE\footnote{The

965: Wronskian $\{f,g\}_x$ with regard to the variable $x$ is

966: defined as $f'g-fg'$.}, with both, the function

967: $Q(\tau,x)$ and the PDE, being invariant\footnote{The

968: invariance under the involution is obvious for the

969: equation; for the function $Q(\tau,x)$, see Lemma 7.5.}

970: under the involution $(\tau,x,r)\rg (-\tau,x,-r)$,

971: \be \label{PDE0}\ee

972:    \bean

973:  &&\left\{\frac{\pl^3 Q}{\pl \tau\pl x^2}

974:  ,

975:   \displaystyle{\left[

976:    \begin{array}{l} \left( r-{\frac {\partial ^{2}Q}{\partial \tau\partial x}} \right) ^{2}

977: \left( {\frac {\partial ^{3}Q}{\partial {x}^{3}}}

978:  \right) +

979: \left( r-{\frac {\partial ^{2}Q}{\partial \tau\partial x}} \right)

980: \frac{\pl}{\pl \tau}\left(2\tau\frac{\pl^2Q}{\pl x\pl \tau}+\frac{\pl^2Q}{\pl \tau^2}\right)

981: \\

982:   +  {\frac {\partial ^{3}Q}{\partial \tau\partial x^2}}    \left( 2r{\frac {\partial ^{2}Q}{

983: \partial {x}^{2}}} +2{\frac {\partial Q}

984: {\partial \tau}}  -xr\right)

985: +\frac{\pl}{\pl x}\left(\frac{1}{4}(\frac {\partial ^{2}Q}{\partial {\tau}^{2}})^2+\tau

986:  \frac {\partial ^{2}Q}{\partial {\tau}^{2}} \frac {\partial ^{2}Q}{\partial {\tau}\partial x} \right)

987:  % +2   {\frac {\partial ^{3}Q}{\partial x^2

988: %\partial t}}    {\frac {\partial Q}

989: %{\partial t}}

990:  %+\tau\frac{\pl}{\pl x}(\frac{\pl Q}{\pl \tau\pl x}\frac{\pl Q}{\pl \tau ^2})

991:  \end{array}\right]}

992:   \right\}_x

993:   \\\no\\

994:  %\be

995: \lefteqn{~~~~~~~ -\frac{1}{2}\left( {\frac {\pl

996: ^{3}Q}{\pl \tau\pl x^2}} \right) ^{2}\left({\frac

997: {\partial ^{3}Q}{\partial {\tau}^{ 3}}}

998:  -4\frac{\pl^2Q}{\pl \tau\pl x}  \frac {\pl ^{3}Q}{\pl

999: x^3}\right)=0,} %\label{one-time PDE}

1000: \no\eean

1001: %

1002: with ``initial condition", given by the log of the

1003: Tracy-Widom distribution,

1004:  \be

1005:  Q_0(x):=\lim_{\tau\rg -\iy} Q(\tau,x):=

1006:  \log\BP(\sup\mathcal {A}(\tau)\leq x)

1007:   =-\int_x^{\iy}(\al-x)g^2(\al)d\al, \label{ic}\ee

1008: where

1009:   $g(\al)$ is the Hastings-MacLeod solution of Painlev\'e

1010:   II,

1011:   \be

1012:   g''=\al g+2g^3,~~~~\mbox{with}~~~ g(\al)\cong

1013:   \frac{

1014:   e^{-\frac{2}{3}  \al^{\frac{3}{2}}}}{2\sqrt \pi \al^{1/4}}

1015: %~ \mbox{for}~ x \rightarrow \iy

1016: \mbox{~~~for~~~}\al\nearrow \infty .\label{Painleve}\ee

1017: 

1018: \end{theorem}

1019: 

1020: %ZZZZZZZZZZZ: general E

1021: 

1022: \begin{remark} Obviously, the PDE (\ref{PDE0})

1023: has the following structure

1024:   \be

1025:  \frac{\pl}{\pl

1026: x}\left(\frac{\ldots}{\frac{\pl^{3}Q}{\pl \tau

1027:   \pl x^2}}\right)

1028: =  \frac{\partial^{3} Q}{\partial \tau^3 } -4

1029: \frac{\partial^{2} Q}{\partial \tau

1030:  \partial x} \frac{\partial^{3} Q}{ \partial x^3}

1031: ~.%

1032: \label{structure}\ee %where the $\ldots$ in the numerator

1033: %of the ratio above is an expression given in equation

1034: %(\ref{one-time PDE}).

1035: \end{remark}

1036: 

1037: 

1038: 

1039: %It is useful to rewrite the PDE (\ref{PDE0})

1040: %  as the ratio of the

1041: %   Wronskian of the third derivative

1042: %  $\frac{\pl^3 Q}{\pl \tau\pl x^2}$ versus some expression,

1043: %  and the square of this third derivative and on the other side

1044: %  a rather simple expression, as follows

1045: %  \bea

1046: % &&\left\{\frac{\pl^3 Q}{\pl \tau\pl x^2}

1047: % ,

1048: %  \displaystyle{\left[\begin{array}{l}

1049: %   %\displaystyle

1050: %   r^{2}

1051: %  \frac{\pl^3 Q}{\pl x^3}

1052: % +r\left(\frac{\pl^2}{\pl \tau^2}(\frac{\pl F}{\pl \tau}

1053: %  +2t\frac{\pl Q}{\pl x}

1054: % )

1055: %   -\frac{\pl^3 (x Q)}{\pl \tau\pl x^2}

1056: %   +2\left\{

1057: %   \frac{\pl^2Q}{\pl \tau\pl x},\frac{\pl^2Q}{\pl x^2}

1058: %   \right\}_x

1059: %    %

1060: % \right)\\ \\

1061: %  +\frac{1}{2}

1062: % \frac{\pl^3Q}{\pl \tau^2\pl x}\frac{\pl^2 Q}{\pl \tau^2}

1063: %  -\frac{\pl^2Q}{\pl \tau\pl x}\frac{\pl^3Q}{\pl \tau^3}

1064: %  +\left(\frac{\pl^2Q}{\pl \tau\pl x}\right)^2\frac{\pl^3Q}

1065: %  {\pl x^3}+

1066: %  \left\{ \frac{\pl^2Q}{\pl \tau\pl x},

1067: %   \frac{\pl (tQ)}{\pl \tau^2}

1068: %  \right\}_x

1069: %  \end{array} \right]}

1070: %  \right\}_x

1071: % \no\\\no\\

1072: % %\be

1073: %\lefteqn{~~~~~~~-\frac{1}{2}\left( {\frac {\pl

1074: %^{3}Q}{\pl \tau\pl x^2}} \right) ^{2}\left({\frac

1075: %{\partial ^{3}Q}{\partial {\tau}^{ 3}}}

1076: % -4\frac{\pl^2Q}{\pl \tau\pl x}  \frac {\pl ^{3}Q}{\pl

1077: %x^3}\right)=0.} \label{one-time PDE} \eea%; see

1078: %%(\ref{one-time PDE}).

1079: 

1080: \begin{remark} The following simple recipe gives the PDE for $\BP(\mathcal {A}^{(r)}(\tau)\cap

1081: E=\emptyset)$ for a general set $E=\cup_1^\ell

1082: [x_{2i-1},x_{2i}]$, replacing the PDE (\ref{PDE0}) for

1083: $E=(x,\iy)$; indeed, perform the replacements  

1084:   $$\frac{\pl}{\pl x}\mapsto\sum_i \frac{\pl}{\pl

1085: x_i}~~\mbox{ and }~~ x\left(\frac{\pl}{\pl

1086: x}\right)^k\mapsto\left(\sum_i x_i\frac{\pl }{\pl

1087: x_i}\right)\left(\sum_i  \frac{\pl}{\pl

1088: x_i}\right)^{k-1},$$ with the understanding that

1089: differentiation must always be pulled to the right.

1090: 

1091: \bigbreak

1092: 

1093: 

1094: 

1095: Although the average mean density of the particles is insensitive to the presence of outliers,

1096: the presence of the $r$ particles forced to reach the

1097: target $a>0$ at $t=1$ is already felt, when $t\rg 0$ in

1098: the $t$-scale; that is when $\tau\rg -\iy$ in the

1099: $\tau$-scale. The net effect is that it pulls the edge

1100: of the cloud of particles in the average towards the right to first

1101: order like $|r/\tau|$; i.e., the more so when $r$ gets

1102: large; this edge then behaves like the Airy process

1103: shifted in space, up to and including order $1/\tau^4$.The

1104: PDE in Theorem \ref{Theo2} is a convenient instrument to

1105: extract the remote past asymptotics, as shown in

1106: Section~\ref{sect7} and stated in the theorem below.

1107: \end{remark}

1108: 

1109: \begin{theorem}\label{Theo3}

1110: 

1111: The PDE with the initial condition $Q_0(x)$ as in

1112: (\ref{ic}) admits the asymptotic solution, for $\tau\rg

1113: -\iy$, of the form \bean

1114:  Q(\tau,x)&=&\sum_0^{\iy}\frac{Q_i(x)}{\tau^i}%\\

1115:  %

1116:    =%&=&

1117:    Q_0\left((x+\frac{r}{\tau})(1+\frac{r}{3\tau^3})+

1118:     \frac{r^2}{4\tau^4}\right)

1119:    + \frac{r}{5\tau^5}{\mathcal{F}}_5

1120:    +O(\frac{1}{\tau^6}).\eean

1121: For the probability itself, one has\footnote{with %$c_5$

1122: %an unimportant constant (which can be determined) and

1123: %with

1124:  $ \mathcal{F}_5:=

1125:   x^2Q_0'+4xQ_0+Q_0^{\prime 2}

1126:    %

1127:    +10\int_x^{\iy}Q_0

1128:     -

1129:    6\int_x^{\iy}dy\int_y^{\iy}du Q_0^{\prime\prime 2}

1130:    %

1131:    .$ }

1132:    \bean

1133:    \lefteqn{\BP(\sup\mathcal {A}^{(r)}(\tau)\leq x)}

1134:    %\BP(\mathcal {A}^{(r)}(\tau)\leq x)

1135:   \\

1136:    &=&

1137:     \BP\left(\sup\mathcal {A}(\tau)\leq (x+\frac{r}{\tau})(1+\frac{r}{3\tau^3})+

1138:     \frac{r^2}{4\tau^4}\right)

1139:     %\\

1140:     %&&~~~~~

1141:     %~\times ~

1142:     \left(1+\frac{r}{5\tau^5}\mathcal{F}_5

1143:      +O(\frac{1}{\tau^6})\right)

1144:     % \\

1145:     % &=&

1146:     %  e^{Q_0}\left(x(1+\frac{r}{3t^3})+\frac{r}{t}(1+

1147:     % \frac{7r}{12 t^3})\right)+O(\frac{1}{t^5})

1148:      .\eean

1149:      For $\tau\rg -\iy$, the mean and variance of the right edge of the process

1150:       behave as

1151:       \bean

1152:       {\mathbb E}

1153:        (\sup \mathcal {A}^{(r)}(\tau))&=&

1154:  {\mathbb E}(\sup\mathcal {A}^{(0)}(\tau))\left(1-\frac{r}{3\tau^3}

1155:  \right)

1156:  -\frac{r}{\tau}-\frac{r^2}{4\tau^4} +O(\frac{1}{\tau^5})

1157:       \\

1158:        \mbox{\em var}(\sup \mathcal {A}^{(r)}(\tau))

1159:        &=&

1160:  \mbox{\em var}(\sup\mathcal {A}^{(0)}(\tau))

1161:  \left(1-\frac{2r}{3\tau^3}\right)

1162:   +O(\frac{1}{\tau^5}).\eean

1163: 

1164: \end{theorem}

1165: 

1166: The probability (\ref{Brownian target}) is related, via

1167: a change of variables, to a Gaussian matrix model with

1168: external potential, \bean\BP_{n} (\alpha , \tilde{E})

1169:  &=& \frac{1}{Z_n}\int_{\HR_n(\tilde E)}

1170:  dM e^{-\frac{1}{2}\Tr (M^2-2AM)}

1171:  ,\eean

1172: where $\HR_n(\tilde E)$ denotes the Hermitian matrices

1173: with eigenvalues in $\tilde E\subset {\mathbb R}$ and with a

1174: diagonal matrix \be

1175:  A:= \left(\begin{array}{cccccc}

1176:    \al\\

1177:    &\ddots& & & &{\bf O}\\

1178:    & &\al\\

1179:    & & &0\\

1180:    &{\bf O}& & &\ddots\\

1181:    & & & & &0

1182: \end{array}\right)\begin{array}{l}

1183: \updownarrow r\\

1184: \\

1185: \\

1186: \\

1187: \updownarrow n-r

1188: \end{array}.

1189:   \label{diagmatrix}\ee

1190:   The relationship between the Karlin-McGregor

1191:   non-intersecting Brownian motions and matrix models

1192:    has been developped

1193:   by Johansson \cite{Johansson1} and, for the Gaussian

1194:   matrix model with external potential, by

1195:   Aptekarev-Bleher-Kuijlaars \cite{AptBleKui}.

1196: %

1197:   The latter model has come up in many other situations;

1198:   among others, see

1199:   \cite{Pastur,Brezin2,Brezin3,Brezin4,Zinn1,Zinn2,Johansson2,

1200:   BleKui1, TW-Pearcey, AvM-Pearcey}.

1201:   Following the method of Adler-van Moerbeke \cite{AvM-Pearcey}, using the multi-component KP hierarchy (section \ref{sect:KP}) and the

1202:   Virasoro constraints (section \ref{sect:Vir}),

1203:   it is shown in section

1204:   \ref{sect:PDE} that this matrix model satisfies a

1205:   non-linear PDE in $\al$ and the boundary points of $\tilde E$.

1206:   The PDE of Theorem \ref{Theo2} then

1207:   follows from making an asymptotic analysis on that

1208:   PDE. The asymptotic behavior of the $r$-Airy process

1209:   for $\tau\rg -\iy$ follows from solving the

1210:   partial differential equation near $\tau=-\iy$ for the

1211:   Airy-kernel initial condition; this process introduces

1212:   some free constants, which then can be determined by the

1213:   asymptotic properties of the $r$-Airy kernel for $x \rg \iy$. % enables

1214:   %one to determine the constants.

1215:    It is an interesting question whether these results

1216:   can be deduced via the Riemann-Hilbert methods, in the

1217:   style of \cite{Deift}.

1218: 

1219:  %\newpage

1220: 

1221: 

1222: 

1223: 

1224: %\end{document}

1225: 

1226: \section{A constrained Brownian motion with a few

1227: outliers}\label{sect1}

1228: 

1229: The purpose of this section is to motivate the scaling

1230: limit (\ref{r-AiryP}) leading to the definition of the

1231: $r$-Airy process. The Airy process was originally

1232: defined as an edge scaling limit of Dyson's

1233: non-intersecting Brownian motions, in the same way that

1234: the Tracy-Widom distribution was obtained as an edge

1235: scaling limit of GUE. The non-intersecting Brownian

1236: motions (\ref{Brownian target}) with target point $a=0$

1237: can be transformed to the Dyson Brownian motion, as will

1238: be explained in this section. This transformation will

1239: be used in the case of a $a\neq 0$ target. In the course

1240: of doing this, one must consider a Gaussian matrix model

1241: with external potential. 

1242: %

1243: \subsection{Dyson Brownian motion} 

1244: %

1245: The Dyson

1246: process (\cite{Dyson}) describes the motion of the eigenvalues

1247: $\lb_i(t')$ of an $n\times n$ Hermitian random matrix $B(t')$ whose

1248: real and imaginary entries perform independent

1249: Ornstein-Uhlenbeck processes, given that the initial

1250: distribution is given by invariant measure for the

1251: process, namely

1252:  $$

1253:   Z^{-1}\int_{\mathcal {H}_n(E')} dB e^{-\Tr B^2}

1254:   .$$

1255: Then the process is stationary and its probability

1256: distribution at any time is given by

1257:  %

1258: \be\BP_{Dy}(\mbox{all}~\lb_i(t')\in E')=

1259:  Z^{-1}\int_{\mathcal {H}_n(E')} dB e^{-\Tr B^2}=

1260:  Z^{-1}\int_{\mathcal {H}_n(\sqrt{2}E')} dM e^{-\frac{1}{2}\Tr M^2}

1261:  .\label{Dyson1}\ee

1262: The probability for two times $0<t_1'<t_2'$ is then

1263: given by (set $c'=e^{-(t'_2-t'_1)}$)

1264: %=

1265: %\frac{\sqrt{\frac{t_1}{1-t_1}}}{\sqrt{\frac{t_2}{1-t_2}}}

1266: %$

1267: %

1268: %$$

1269: %P(\mbox{all}~\lb_i(t')\in E')=Z^{-1}\int_{{\cal

1270: %H}_n(E')}dM~e^{-\Tr M^2}dM

1271: %$$

1272: %\vspace{1cm}

1273: %

1274:  \bea

1275: \lefteqn{\BP_{Dy}(\mbox{all~}\lb_i(t'_1)\in E'_1,

1276:        \mbox{~all~}\lb_i(t'_2)\in E'_2)} %\\

1277: \no\\ \no\\%\no\\

1278: &=& P(\mbox{all~}(B(t'_1)

1279:  \mbox{-eigenvalues})\in E'_1,\mbox{~all~}(B(t'_2)

1280:   \mbox{-eigenvalues})\in E'_2) %\\

1281: \no\\ \no\\%\no\\

1282: &=&  \int\!\!\!\int_{\mbox{\scriptsize{all~}}B _1

1283: -\mbox{\scriptsize{eigenvalues~}}\in

1284: E'_1\atop{\mbox{\scriptsize{all~}}B _2

1285:  -\mbox{\scriptsize{eigenvalues~}}\in E'_2}}%\\

1286: %\\

1287: %& &\hspace*{4cm}

1288:  Z^{-1}\frac{dB_1dB_2}{(1-c'^2)^{n^2/2}}

1289: e^{-\frac{1}{1-c^{\prime 2}}\Tr(B_1^2+B_2^2-2c'B_1B_2)}\no\\

1290: \no\\%\no\\

1291: &=&  \int\!\!\!\int_{

1292:  M_1\in

1293: \mathcal {H}_n(\frac{\sqrt{2}E'_1}{\sqrt{1-c'^2}}) \atop{

1294: M_2 \in \mathcal {H}_n (\frac{\sqrt{2}E'_2}{ \sqrt{1-c'^2}}}

1295: )}

1296: %\\

1297: %\\

1298: %& &\hspace*{4cm}

1299: Z^{'-1}dM_1dM_2~e^{-\frac{1}{2}

1300:    \Tr(M_1^2+M_2^2-2c'M_1M_2)}

1301: \label{Dyson2}\eea 

1302: %

1303: %\newpage

1304: %

1305: %\emph{Proof of Theorem 0.2 \,:} \,\,\quad

1306: \subsection{Constrained Brownian motion with target

1307: $a>0$} As in the introduction, consider the $n$

1308: non-intersecting \mbox{Brownian} particles on ${\mathbb R}$, all

1309: starting from the origin at time $t=0$, where among

1310: those paths, $1 \leq r \leq n$ are forced to end up at

1311: the target $a > 0$, while the $(n-r)$ other paths return

1312: to the position $x=0$ at time $t=1$. Remember

1313: $\BP_{Br}^{0a}$ denotes the transition probability. Then

1314: the probability that all the particles belong to some

1315: window $E \subset {\mathbb R}$ at a given time $0 < t < 1$, can

1316: be expressed in terms of a Gaussian matrix model with an

1317: external potential, using the Karlin-McGregor formula

1318: \cite{Karlin} for non-intersecting Brownian motions (see

1319: \cite{Grabiner, Johansson1, AptBleKui, AvM-Pearcey}),

1320: itself involving the Brownian transition probability

1321: (\ref{Normal}),

1322: %\begin{equation}

1323: %p(t ; x, y)  :=  \frac{1}{\sqrt{\pi t}} e^{-

1324: %\frac{(y-x)^{2}}{t}}, \label{C036} \end{equation}

1325: namely:

1326: %

1327: %   $\BP(a;E)$, using Karlin-McGregor:

1328: %

1329: %

1330: %

1331: \bean    %\lefteqn

1332: { \BP_{Br}^{0 a} \left( \mbox{all

1333: $x_j(t)\in E$} \right)

1334:  }%\no\\ \no\\

1335:  &:=&

1336:   %\BP_{Br}^{0a}\left(\mbox{all \,$x_j(t)\in E$}\right)

1337: %\,:=\,

1338: \BP \left(\begin{tabular}{c|c}

1339: & all $x_j(0) =0$\\

1340: all $x_j(t)\in E$ &

1341:  $r$ right paths end up at $a$ at $t=1$\\

1342:   &

1343: $(n-r)$ paths end up at $0$ at $t=1$

1344: \end{tabular}\right)

1345: \\

1346:  &=&\lim_{\begin{array}{c}

1347:   \mbox{  all $~\gamma_i\rightarrow 0$}

1348:   \\

1349:  \mbox{$\delta_1, \ldots , \delta_{n-r}\rightarrow 0$}\\

1350:   \mbox{$\delta_{n-r+1}, \ldots , \delta_{n}

1351:   \rightarrow a$}

1352:   \end{array}} \int_{E^n}\frac{1}{Z_{n}(\ga,\delta)}

1353:   \prod_1^n dx_i\no\\%}

1354:    \no\\ &&~~~\hspace*{0cm}

1355:   \det\left(p(t;\gamma_i,x_j)\right)_{1\leq i,j\leq

1356: n}\det(p(1-t;x_{i'},\delta_{j'}))_{1\leq i',j'\leq n}

1357:   ,

1358:  \\%  \eean\bean

1359:   &=& \BP_{n}\left(a \sqrt{\frac{2t }{1-t}};

1360: E\sqrt{\frac{2}{t(1-t)}}\right)

1361:  % \no\\

1362:   \eean

1363: %

1364: %

1365: %

1366: %

1367: where\footnote{$\Dt_r(x)$ denotes the Vandermonde

1368: determinant, with regard to the $r$ variables

1369: $x=(x_1,\ldots,x_{r})$ and $\Dt_n(x,y)$ denotes the

1370: Vandermonde determinant, with regard to the $n$

1371: variables $x=(x_1,\ldots,x_{r})$ and

1372: $y=(y_1,\ldots,y_{n-r})$.}

1373:  \bean\BP_{n} (\alpha ,

1374: \tilde{E})

1375:  &=& \frac{1}{Z_n}\int_{\HR_n(\tilde E)}

1376:  dM e^{-\frac{1}{2}\Tr (M^2-2AM)}

1377:  \\%\eean\bean

1378:   &=& \frac{1}{Z'_{n}}

1379:   \int_{ \tilde E  %\sqrt{\frac{2}{t (1-t)}}

1380:    ^{n}

1381:    }

1382:    \Dt_{n}(x ,y )

1383:  \left(\Dt_r(x )  \prod^r_{i=1}

1384:   e^{

1385:  -\frac{x_i^{   2}}{2}  +  { \al x_i

1386:  }}

1387:  dx_i\right)

1388: %%

1389: %\\

1390: %&& ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

1391: \left(\Dt_{n-r}(y  )  \prod^{n-r}_{i=1}e^{ -\frac{y_i^{

1392:  2}}{2} }dy_i  \right) , \eean

1393: with $A$ as in (\ref{diagmatrix}).

1394: % \be

1395: % A:= \left(\begin{array}{cccccc}

1396: %   \al\\

1397: %   &\ddots& & & &{\bf O}\\

1398: %   & &\al\\

1399: %   & & &0\\

1400: %   &{\bf O}& & &\ddots\\

1401: %   & & & & &0

1402: %\end{array}\right)\begin{array}{l}

1403: %\updownarrow  r\\

1404: %\\

1405: %\\

1406: %\\

1407: %\updownarrow n-r

1408: %\end{array}.

1409: %  \label{diagmatrix}\ee

1410:   %

1411: % \be

1412: %\BP_n(\al;b):=\BP(~\mbox{all eigenvalues}~\in \tilde E)

1413: %=

1414: % \frac{1}{Z_n} \int_{\mathcal {H}_{n}(\tilde E)}

1415: %   e^{-\Tr(\frac{1}{2} M^2-AM) }dM,

1416: %  \label{0.4}

1417: %   \ee

1418: %

1419: %    is the

1420: %probability () associated to the random \mbox{matrix}

1421: %model of theorem $0.1$, \,with \,$k_{1}=k$\, and

1422: %\,$k_{2}=n-k$. \,

1423: In short, the conditioned Brownian motion is related to

1424: a Gaussian matrix model with an external potential $A$

1425: as follows:

1426: %

1427: %two functions are related by the simple change of

1428: %variables \,:

1429: %

1430: \begin{equation}

1431: \BP_{Br}^{0a}\left(\mbox{all $x_j(t)\in E$}\right)

1432:  = \BP_{n} \bigl(\al, \tilde E %a  \sqrt{\frac{2\,t}{1-t}}

1433: %\,\,\,,\,\, E  \sqrt{\frac{2}{t\,( 1-t)}} \,

1434: \bigr) ~\mbox{  with  }~

1435:  \tilde E=E  \sqrt{\frac{2}{t\,( 1-t)}} ,~~\al=a

1436: \sqrt{\frac{2\,t}{1-t}} .\label{CBM1}

1437: \end{equation}

1438: %with $$ \tilde E=E  \sqrt{\frac{2}{t\,( 1-t)}} ,~~\al=a

1439: %\sqrt{\frac{2\,t}{1-t}} .$$

1440: %

1441: The joint probability for the constrained Brownian

1442: motion at two times is related to a chain of two

1443: Gaussian matrix models (see \cite{AvM-Pearcey-joint})

1444: with an external potential $A$, which again by

1445: Karlin-McGregor reads as follows: \be

1446:   {\BP_{Br}^{0a}(\mbox{all~}x_i(t_1)\in

1447: E_1,\mbox{all~}x_i(t_2)\in  E_2)}

1448: =\BP_n\bigl(\alpha;c';\tilde

1449: E_1,\tilde E_2\bigr),%

1450: \label{CBM2}\ee

1451:  where ($A$ is the same diagonal matrix as in

1452:  (\ref{diagmatrix}))

1453:   \bean

1454: % \lefteqn{\BP_n\left(\alpha;c';\tilde

1455: %E_1,\tilde E_2\right) }\\

1456: %\\

1457: %&:=&

1458: % \frac{1}{Z_n}\int_{\prod^m_{\ell=1}{\cal

1459: %H}_n(\tilde E_{\ell})}\!\!\!\!\!\!\!e^{-\frac{1}{2}\Tr

1460: %(M_1^2+\ldots+M_m^2-2c'_1M_1M_2-\ldots-2c'_{m-1}M_{m-1}M_{m}-2AM_m)}

1461: %dM_1\ldots dM_m

1462: %\\

1463:  \BP_n\left(\alpha;c';\tilde

1464: E_1,\tilde E_2\right)&=&

1465:  \frac{1}{Z_n}\int_{ \mathcal{H}_n(\tilde E_{1})\times\mathcal {H}_n(\tilde E_{2})

1466: }\!\!\!\!\!\!\!e^{-\frac{1}{2}\Tr

1467: (M_1^2+M_2^2-2c'M_1M_2-2AM_2)} dM_1  dM_2

1468:  ,\eean

1469: % upon setting ($A$ is diagonal with $\al$)

1470: %$$

1471: %\tilde E_{\ell}%=\tilde b_i^{(\ell)}\sqrt{\frac{2}{\sg_{\ell}}}

1472: %=

1473: %   E_\ell\sqrt{\frac{2(t_{\ell+1}-t_{\ell-1})}

1474: %  {(t_{\ell+1}-t_{\ell})(t_{\ell}-t_{\ell-1})}}

1475: % ~%\frac{x_i^{(j)}}{\sqrt{\sg_j}}=\frac{y_i^{(j)}}{\sqrt{2}},\qquad

1476: %,~c'_j%=\frac{\sqrt{\sg_j\sg_{j+1}}}{t_{j+1}-t_j}

1477: % =\sqrt{ \frac{(t_{j+2}-t_{j+1})(t_{j}-t_{j-1})}

1478: %   {(t_{j+2}-t_{j})(t_{j+1}-t_{j-1})}}\label{changeof

1479: %   variables1}$$

1480: %    \mbox{and}

1481: %   \be\al

1482: %=a\frac{\sqrt{2\sg_m }}{1-t_m}

1483: %=a\sqrt{\frac{2(t_{m}-t_{m-1})}{(1-t_m)(1-t_{m-1})}}

1484: %  .%~,~~\mbox{for}~1\leq j

1485: %\leq m-1

1486: % \label{changeof

1487: %   variables2} \ee %$\mbox{for}~1\leq j \leq m-1$

1488: %

1489: %where

1490: %

1491: with%when $m=2$, one has

1492: $$

1493: \tilde  E_1=E_1 \sqrt{\frac{2t_2}{(t_2-t_1)t_1}} ,~~~~

1494:  \tilde  E_2=E_2 \sqrt{\frac{2(1-t_1)}{(1-t_2)(t_2-t_1)}}

1495:   $$

1496:   \be

1497:   c'=\sqrt{\frac{(1-t_2)t_1}{(1-t_1)t_2}}=

1498:    \frac{\sqrt{\frac{t_1}{1-t_1}}}{\sqrt{\frac{t_2}{1-t_2}}}

1499:  ,~~~~~~

1500:  \al=a\sqrt{\frac{2(t_2-t_1)}{(1-t_2)(1-t_1)}}

1501:   %= a\sqrt{2}\sqrt{e^{2t'_2}-e^{2t'_1}}

1502:   \label{map1}\ee

1503: %

1504: %

1505: \subsection{Comparing the Dyson and constrained Brownian

1506: motions with target $a=0$}

1507: %

1508: From the identities (\ref{Dyson1}) and (\ref{CBM1}), one

1509: deduces the following set of identities for one time

1510: $t'$ (in the time $t'$ of the Dyson process), setting

1511: the point $a=0$,

1512:  \be

1513: \BP_{Dy}\left(\mbox{all \,$\lb_j(t')\in

1514: E'$}\right)=\BP_n(\al, \tilde E)\Bigr|_{\al =0} =

1515: \BP_{Br}^{0a}\left(\mbox{all \,$x_j(t)\in E$}\right)

1516: \Bigr|_{a =0}\label{iso} ,\ee

1517:   where

1518: $$

1519: \sqrt{2}E'=\tilde E=E \sqrt{\frac{2}{t\,( 1-t)}} .$$

1520: %

1521: Similarly, for two times $t'_1$ and $t_2'$, one deduces

1522: from the identities (\ref{Dyson2}) and (\ref{CBM2}),

1523:  \bean %\lefteqn

1524:  {

1525: \BP_{Dy}(\mbox{all~}\lb_i(t'_1)\in E'_1,

1526:        \mbox{~all~}\lb_i(t'_2)\in E'_2)}%\\

1527:        &=&

1528:        \BP_n\left(\alpha;c';\tilde

1529: E_1,\tilde E_2\right)\Bigr|_{\al =0} \\

1530:  &=&

1531:  {\BP_{Br}^{0a}(\mbox{all~}x_i(t_1)\in

1532: E_1,\mbox{all~}x_i(t_2)\in  E_2)}\Bigr|_{a =0}

1533:  ,\eean

1534: where, as follows from (\ref{Dyson2}) and (\ref{map1}),

1535: one deduces

1536: %

1537: \bea

1538: \frac{\sqrt{2}E'_1}{\sqrt{(1-c'^2)}}&=&\tilde E_1=E_1

1539: \sqrt{\frac{2t_2}{(t_2-t_1)t_1}}

1540: %$$

1541: %\be

1542: \no\\

1543: \frac{\sqrt{2}E'_2}{\sqrt{(1-c'^2)}}&=&\tilde

1544: E_2=E_2 \sqrt{\frac{2(1-t_1)}{(1-t_2)(t_2-t_1)}}

1545: \label{A}\eea with $$

1546: c'=\frac{e^{t'_1}}{e^{t'_2}}%\sqrt{\frac{(1-t_2)t_1}{(1-t_1)t_2}}=

1547:    =\frac{\sqrt{\frac{t_1}{1-t_1}}}{\sqrt{\frac{t_2}{1-t_2}}}

1548:  .$$

1549: %

1550: %\newpage

1551: %

1552: Thus the processes are related by a clock change

1553:  \be

1554: e^{t'_i}= \sqrt{\frac{t_i}{1-t_i}}\Longleftrightarrow

1555: t_i=\frac{1}{1+e^{-2t'_i}}

1556:   \Longleftrightarrow

1557:   \frac{1}{\sqrt{t_i(1-t_i)}}=2\cosh

1558:   t'_i,\label{time-change}\ee

1559:  and thus $

1560: 1-c^{\prime 2} =\frac{t_2-t_1}{t_2(1-t_1)}.$

1561: %

1562: %

1563: Comparing extremities in (\ref{A}), one finds

1564: $$

1565: E'_1=E_1 \sqrt{\frac{2t_2}{(t_2-t_1)t_1}}

1566:  \sqrt{(1-c'^2)/2}=E_1\frac{1}{\sqrt{t_1(1-t_1)}}

1567: $$

1568: $$

1569: E'_2=E_2 \sqrt{\frac{2(1-t_1)}{(1-t_2)(t_2-t_1)}}

1570:  \sqrt{(1-c'^2)/2}=E_2\frac{1}{\sqrt{t_2(1-t_2)}}

1571: .$$

1572: %also:

1573: %$$

1574: %E'_i\frac{1}{2 \cosh t'_i}=E_i'\sqrt{t_i(1-t_i)}=E_i$$

1575: This fact, combined with (\ref{time-change}), yields

1576: % $$

1577: %\sqrt{2}E'_i=E_i\sqrt{\frac{2}{{t_i(1-t_i)}}}

1578: %$$

1579: %$$

1580: %\sqrt{2}E'_2=E_2\sqrt{\frac{2}{ {t_2(1-t_2)}}}

1581: %$$

1582: %and so

1583: \be E_i=E'_i \sqrt{t_i(1-t_i)}=\frac{E'_i}{2\cosh t'_i}

1584: .\label{space-change} \ee

1585: %

1586: % \bean

1587: %\lefteqn{ \BP_{Br}^{0 a}\left(\mbox{all $x_j(t)\in

1588: %E$}\right)

1589: %  }

1590: %   \\

1591: %&:=& \BP \left(\begin{tabular}{c|c}

1592: %   & all $x_j(0) =0$\\

1593: %  all $x_j(t)\in E$  & $k$ left paths end up at $-a$ at time $t=1$,\\

1594: %   &

1595: %$k$ right paths end up at $+a$ at time $t=1$

1596: %\end{tabular}\right)

1597: %%\\  \\

1598: %\eean

1599: So, summarizing (\ref{time-change}) and

1600: (\ref{space-change}), one has the relation between the

1601: parameters of the Dyson process and constrained Brownian

1602: motions,

1603: %

1604: \be e^{t'_i}= \sqrt{\frac{t_i}{1-t_i}}

1605:  ,~~~~E_i\sqrt{\frac{2}{t_i(1-t_i)}}=\sqrt{2}E'_i

1606:  .\label{space-time-change}

1607:  \ee

1608: 

1609:  %\vspace*{.2cm}%\bigbreak

1610: 

1611:  %\noindent

1612: 

1613: 

1614: %\newpage

1615: 

1616: \begin{theorem}\label{Theo:1.1}

1617:   (Tracy-Widom \cite{TW-Airy},

1618:   Adler-van Moerbeke \cite{AvM-Airy-Sine})

1619: Taking a limit on the Dyson process, in an appropriate

1620:  time and space scale,

1621:  one finds the Airy process, which is stationary:

1622: $$

1623: \lim_{n\rightarrow \iy}

1624:  \BP_{Dy}\left(\mbox{all}~ \lb_i\left(\frac{\tau}{n^{1/3}}\right) \in

1625:   (-\iy, \sqrt{2n}+\frac{x}{\sqrt{2}

1626:   n^{1/6}})\right)=\BP(\mathcal {A}(\tau)\leq x)=F(x),

1627: $$

1628:  $F(x)$ being the Tracy-Widom distribution.

1629: Similarly the limit of the joint probability for the

1630: Dyson process yields the joint probability for the Airy

1631: process:

1632:  $$ \lim_{n\rightarrow \iy}

1633:  \BP_{Dy}\left(\begin{array}{l}

1634:    \mbox{all}~ \lb_i\left(\frac{\tau_1}{n^{1/3}}\right) \in

1635:   (-\iy, \sqrt{2n}+\frac{x_1}{\sqrt{2}

1636:   n^{1/6}}) \\

1637:    \mbox{all}~ \lb_i\left(\frac{\tau_2}{n^{1/3}}\right) \in

1638:   (-\iy, \sqrt{2n}+\frac{x_2}{\sqrt{2}

1639:   n^{1/6}})\end{array}\right)

1640:   =

1641:    \BP\left(\begin{array}{l} \mathcal {A}(\tau_1)\leq x_1

1642:       ,\\

1643:      \mathcal {A}(\tau_2)\leq x_2 \end{array}  \right)

1644:     . $$

1645:      Its logarithm  (setting

1646: $s=(\tau_2-\tau_1)/2$)

1647: %

1648: $$

1649:  H(s;x,y):=

1650:   \log \BP\left(\mathcal {A}(\tau_1) \leq {x+y} , \mathcal {A}(\tau_2) \leq

1651:    {x-y} \right),

1652:  $$

1653: %$$ g(x,y)=g(u-v,u+v)=h(u,v)

1654: %.$$ Then $g(x,y) $

1655: satisfies the Airy PDE %

1656: \be

1657:   2s\frac{\pl^3 H}{\pl s\pl x \pl y}= \bigl( 2s^2\frac{\pl}{\pl y}-

1658: y\frac{\pl}{\pl x}\bigr)\bigl(\frac{\pl^2 H}{\pl y^2}-

1659:  \frac{\pl^2 H}{\pl x^2}\bigr)+ \left\{ \frac{\pl^2 H}{\pl x\pl y},

1660:   \frac{\pl^2 H}{\pl x^2} \right\}_x.

1661:  \label{Airy PDE1}\ee

1662: \end{theorem}

1663: 

1664:  %\remark  Note for the solution $H(t;x,y)$,

1665: %  $$

1666: %  \lim_{t\searrow 0}H\left(t;x,y

1667: %   \right)=\log  F_2

1668: %  \left(\min(\frac{y+x}{2},

1669: %  \frac{y-x}{2})\right)

1670: %  .$$

1671: 

1672: 

1673: 

1674: 

1675: 

1676: 

1677: \begin{corollary}\label{Cor:k-Airy} Taking an appropriate scaling limit on the constrained Brownian

1678: motion, we have

1679: %

1680:  \bean { \lim_{n\rightarrow \iy}

1681: \BP^{0a}_{Br}\left(\mbox{all}~

1682: x_i\left(\frac{1}{1+e^{-2\tau/n^{1/3}}}\right) \in

1683:   \frac{(-\iy, \sqrt{2n}+\frac{x}{\sqrt{2}

1684:   n^{1/6}})}{2\cosh( \tau/n^{1/3})}\right)\Bigr|_{a=0}

1685:   }=\BP(\mathcal {A}(\tau)\leq x)%=F(x)

1686:   \eean

1687:  and similarly for two times. \end{corollary}

1688: \begin{proof} This statement follows immediately from Theorem

1689: 1.1 and the correspondence (\ref{iso}), upon using the clock

1690: change (\ref{time-change}) and the space change

1691: (\ref{space-change}), with the appropriate time and

1692: space scalings of Theorem \ref{Theo:1.1}.\end{proof}

1693: 

1694: \bigbreak

1695: 

1696: \noindent Remembering the definition

1697:  $$\BP_{n} (\alpha ,

1698: \tilde{E})

1699:  = \frac{1}{Z_n}\int_{\HR_n(\tilde E)}

1700:  dM e^{-\frac{1}{2}\Tr (M^2-2AM)},$$

1701:   for the diagonal matrix $A$ as in (\ref{diagmatrix}),

1702:    we now state a Theorem of P\'ech\'e, which is closely

1703:   related to the multivariate statistical problem

1704:   mentioned in the introduction; see \cite{BBP}.

1705: 

1706: \begin{theorem}  \label{Peche}(P\'ech\'e \cite{Peche})

1707: \be \lim_{n\rightarrow \iy}

1708:  \BP_n\left(\rho\sqrt{n},

1709:  (-\iy,2\sqrt{n}+\frac{x}{n^{1/6}})\right)=

1710:  \left\{

1711:   \begin{array}{l}

1712:   F(x)\mbox{  for} ~~\rho <1\\ \\

1713:   F^{(r)}(x)\mbox{  for} ~~\rho =1

1714:   \end{array}

1715:   \right.

1716:   \label{PecheF}\ee

1717:  where

1718:  \be

1719:  F^{(r)}(x)=\det (I-K^{(r)}_\tau{\chi_{

1720:  }}_{(x,\iy)}(y))\Bigr|_{\tau=0}

1721:  \label{r-Airy1}\ee

1722:   is the Fredholm determinant of the kernel (see

1723:   (\ref{k-Airy0}))

1724:     \be

1725:  K^{(r)}_\tau(x,y)=\int_0^{\iy}

1726:  du A_r^-(x+u; \tau)A_r^+(y+u; \tau).

1727:  \label{r-Airy kernel}\ee

1728:  %and where\footnote{$C$ is a contour running from $\iy

1729: % e^{5i\pi/6}$ to $\iy

1730: % e^{i\pi/6}$, such that $-i\tau$ lies above the contour.}

1731: % $$

1732: %A_r^-(x; \tau) =

1733: % \int_C   e^{\frac{1}{3} iu^3+iux }

1734: % \left(\frac{1}{iu-\tau}\right)^r

1735: % \frac{du}{2\pi}

1736: %  $$

1737: %  $$

1738: %A_r^+(x; \tau) =

1739: % \int_C ~e^{\frac{1}{3} iv^3+ivx }

1740: % \left( {-iv-\tau}\right)^r

1741: %  \frac{dv}{2\pi}

1742: % .$$

1743:  For $r=0$, (\ref{r-Airy1}) yields the

1744:  Tracy-Widom distribution $F(x)$.

1745: 

1746: \end{theorem}

1747: 

1748: 

1749: 

1750:  The Airy process will now be deformed in a way

1751:   which is compatible with P\'ech\'e's Theorem, especially concerning the target point $a$,

1752:  %

1753:   $$

1754:    %

1755: \BP^{(0a)}_{Br}\left(\mbox{all}~

1756: x_i\left(\frac{1}{1+e^{-2\tau/n^{1/3}}}\right) \in

1757:   \frac{(-\iy, \sqrt{2n}+\frac{x}{\sqrt{2}

1758:   n^{1/6}})}{2\cosh( \tau/n^{1/3})}\right)

1759:   .$$

1760:    Since, from (\ref{CBM1}),

1761: $$

1762:  \BP_{Br}^{0 a} \left( \mbox{all $x_j(t)\in E$} \right)

1763:  %

1764:  %

1765: =\BP_{n}\left(a \sqrt{\frac{2t }{1-t}};

1766: E\sqrt{\frac{2}{t(1-t)}}\right)

1767: $$

1768:  holds, we have

1769: \bean

1770:    %

1771:  \lefteqn

1772:  {\BP^{(0a)}_{Br}\left(\mbox{all}~

1773: x_i\left(\frac{1}{1+e^{-2\tau/n^{1/3}}}\right) \in

1774:   \frac{ (-\iy,\sqrt{2n}+\frac{x}{\sqrt{2}

1775:   n^{1/6}})}{2\cosh( \tau/n^{1/3})}\right)

1776:  }  \hspace*{4cm} \\

1777:     && ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

1778:   =\BP_n\left(a\sqrt{2}e^{\tau/n^{1/3}};

1779:  (-\iy,2\sqrt{n} +\frac{x}{n^{1/6}})

1780:  \right).\eean

1781:   Comparing this formula at time $t=1/2$, or what is the same at $\tau=0$,

1782:    with formula (\ref{PecheF})

1783:  suggests the choice \be a=\rho\sqrt{\frac{n}{2}}.\label{choice a}\ee

1784: %

1785:   We thus define the {\bf Airy process with $r$ outliers}

1786:   (in short: $r$-Airy process) $\mathcal {A}^{(r)}(\tau)$

1787:   by means of the Airy scaling as in Corollary

1788:   \ref{Cor:k-Airy} and with the choice of $a$ above for $\rho=1$. Notice

1789:   from (\ref{CBM1}) and using (\ref{space-time-change}),

1790:   this can also be expressed in terms of the matrix

1791:   model with external potential:

1792:   %

1793:   \bean {\BP(  \mathcal {A}^{(r)}(\tau)\cap

1794:     E=\emptyset)

1795:   }

1796:   %\\

1797:   &:=& \lim_{n\rightarrow \iy}

1798: \BP^{(0,\sqrt{n/2})}_{Br}\left(\mbox{all}~

1799: x_i\left(\frac{1}{1+e^{-2\tau/n^{1/3}}}\right) \in

1800:   \frac{ \sqrt{2n}+\frac{E^c}{\sqrt{2}

1801:   n^{1/6}} }{2\cosh( \tau/n^{1/3})}\right)

1802:   \\ \\

1803:   &=&

1804:    \lim_{n\rightarrow \iy}

1805:   \BP_{n}\left({}\sqrt{n}e^{\tau /n^{1/3}};

1806:   2\sqrt{n} +\frac{E^c}{n^{1/6}}   \right).%

1807:   \eean

1808:   In a similar way, one defines the joint probability

1809:   of $\mathcal {A}^{(r)}(\tau)$

1810:   for any number of times. The next section deals with

1811:   this limit expressed in terms of a Fredholm

1812:   determinant.

1813: 

1814:  %\newpage

1815: 

1816:  \section{The existence of the limit to the $r$-Airy kernel}\label{sect2}

1817: 

1818: The first part of this section deals with a sketch of the proof of

1819:  Theorem \ref{Theo1}; details and rigor can be found in \cite{BBP} and

1820:  \cite{Peche}. In the second part,

1821:  the $r$-Airy kernel (\ref{r-Airy kernel})

1822:   will be expanded for large time $\tau$.

1823: 

1824: 

1825: 

1826: \begin{proof}[Proof of Theorem \ref{Theo1}]

1827: From the explicit Brownian motion transition

1828: probability, one shows (see Johansson \cite{Johansson1}

1829: and Tracy-Widom \cite{TW-Pearcey})

1830: 

1831: $$

1832: \BP_{Br}^{0a}\left(\mbox{all $x_j(t)\in E$}\right) =

1833:   {\det(I-H_n^{(r)})}_{L^2(E^c)}

1834:   $$

1835:   where

1836: %

1837: %

1838: %

1839:  \newline\noindent$

1840: H_n^{(r)}(x,y)dy =%\hfill

1841:   $

1842: %

1843: %

1844: \be -\frac{dy}{2 \pi^2 (1-t)} \int_{ \mathcal{D}} dz

1845: \int_{\Gamma_{L}} dw ~e^{ -\frac{ tz^2}{1-t}  +

1846: \frac{2xz}{1-t} + \frac{tw^2}{1-t} - \frac{2wy}{1-t} }

1847: \bigg(\frac{w}{z}\bigg)^{n-r}

1848: \bigg(\frac{w-a}{z-a}\bigg)^{r} \frac{1}{w-z}

1849: ;\label{C032} \ee $\mathcal{D}$ is a closed contour

1850: containing the points $0$ and $a$, which is to the left

1851: of the line $\Gamma_L:=L+i{\mathbb R}$ by picking $L$ large

1852: enough. So, $\Re

1853: (w-z)>0$.

1854: %

1855:  Consider now an arbitrary point $(y,t)$

1856:  on the curve $\mathcal{C}$, parametrized by (\ref{tangency1})

1857:   and the point $(\rho\sqrt{n/2},1)$,

1858:  which is the point of intersection of the tangent to

1859:  $\mathcal{C}$ at $(y,t)$ with the axis $(t=1)$; as pointed out before, it is convenient to parametrize $\rho$ by $\rho=e^{-\sigma}$ and thus

1860:   \be

1861:  (y,t)= \left(\frac{\rho \sqrt{2n}}{1+\rho^2},

1862:  \frac{1}{ 1+\rho^2 }\right) = \left(\frac{\sqrt{2n}    }

1863:    {2\cosh(\sigma)}  ,\quad  \frac{1}

1864:  {1+e^{-2 \sigma }}\right)\in \mathcal{C}

1865: .\label{tangency1}

1866:  \ee

1867:   Consider the Brownian motions with $r$ outliers forced to a point $\rho_0\sqrt{n/2}$ with $0<\rho_0\leq \rho$ at time $t=1$; here also parametrize $\rho_0$ by $\rho_0=e^{-\tau_0}$ and set $\alpha=\rho_0 / \rho=e^{\sigma-\tau_0}$.

1868: 

1869: 

1870: The main issue is to compute the following limit, for $0<\rho_0\leq \rho$

1871: %Remembering the relation $\rho_0=e^{-\tau_0}$, as in (\ref{rho0}), between $%%%%\rho_0$ and $\tau_0$, the following limit% In view of

1872: \bean %\lefteqn

1873: %{\hspace*{-1cm}\BP(\mathcal {A}^{(r)}(\tau)\cap E=\emptyset)}%\hspace*{5cm}

1874: %\\

1875: %  &:=&

1876: \lim_{n\rightarrow \iy}

1877: \BP^{(0,\rho_0\sqrt{n/2})}_{Br}\left(\mbox{all}~

1878: x_i\left(\frac{1}{1+e^{-2(\sigma+\frac{\tau}{n^{1/3}})}}\right) \in

1879:   \frac{  \sqrt{2n}+\frac{E^c}{\sqrt{2}

1880:   n^{1/6}} }{2\cosh(\sigma+ \frac{\tau}{n^{1/3}})}\right)

1881:   %

1882:   ,\eean

1883:  which dictates the space- and

1884:   time-scale to be used in the kernel

1885:  $H_n^{(r)}$ for large $n$. Let $t$ and $y$ be the time and space variables for the Brownian motion, which in terms of the new time and space scale $\mathcal{L}$ reads:

1886:  \be \mathcal{L}:\quad

1887:   t=\frac{1}{1+e^{-2(\sigma+\tau/n^{1/3})}},\quad

1888:   x= \frac{

1889: \sqrt{2n}+\frac{u}{\sqrt{2}

1890:   n^{1/6}} }{2\cosh(\sigma+ \frac{\tau}{n^{1/3}})},\quad y= \frac{  \sqrt{2n}+\frac{v}{\sqrt{2}

1891:   n^{1/6}}}{2\cosh(\sigma+ \frac{\tau}{n^{1/3}} )}

1892:  , \label{substit}\ee

1893:   with target point $a=\rho_0\sqrt{n/2}=e^{-\tau_0}\sqrt{n/2}$. Putting this rescaling in the integral (\ref{C032}) suggests changes of integration variables

1894:  $$ z:=\tilde z\sqrt{\frac{n}{2}}e^{-\sigma-\tau/n^{1/3}}\mbox{    and    }

1895:   w:=\tilde w\sqrt{\frac{n}{2}}e^{-\sigma-\tau/n^{1/3}}$$

1896:  in the integral (\ref{C032}); the exponential will contain a function $ F(z)$, with Taylor series at $z=1$: \be F(z)=\frac{z^2}{2}-2z+\log z

1897:   = F(1)+\frac{1}{3}(z-1)^3+ { O}(z-1)^4\label{F}

1898:  . \ee   Also set $\zeta

1899:  :=1+\frac{\gamma}{n^{1/3}}$ for some parameter $\gamma$ and

1900:  \be

1901:   Z_n=\left\{\begin{array}{lll}n^{-r/3}e^{nF(1)} &\mbox{ for }&\alpha=1\\

1902:                                      (1-\alpha)^r e^{nF(1)}

1903:                                      & \mbox{ for } &0<\alpha<1

1904:                                      \end{array}\right.

1905:      .\label{Zn}\ee

1906:  %

1907:  %

1908:  Then, using,

1909:  $$

1910: \frac{1}{n^{1/3}(\tilde w-\tilde z)}=\int_0^{\iy}

1911:  e^{-yn^{1/3}(\tilde w-\tilde z)}dy

1912:  ,~~~~~~\mbox{for}~~\Re

1913: (\tilde w-\tilde z)>0,$$

1914:  one checks\footnote{ by elementary computation$$

1915:  \frac{t}{1\!-\!t}=e^{2(\sigma+\frac{\tau}{n^{1/3}})}, \quad \frac{dy}{1-t} \frac{dz dw}{w-z}=

1916:   \frac{n^{\frac{1}{3}}d\tilde z d\tilde w dv}{\tilde w-\tilde z}

1917:  ,\quad  \frac{2xz}{1\!-\!t}=(2n+un^{\frac{1}{3}})\tilde z,\quad \frac{tz^2}{1\!-\!t} =\frac{n}{2}\tilde z^2.

1918:    $$}, using the rescaling (\ref{substit}):

1919: %

1920:    \bean

1921:  \lefteqn

1922: {  \left.

1923:  H_n^{(r)}\left(% \frac{

1924: %\sqrt{2n}+\frac{u}{\sqrt{2}

1925: %  n^{1/6}} }{2\cosh(\tau_0+ \tau/n^{1/3})}, \frac{  \sqrt{2n}+\frac{v}{\sqrt{2}

1926: %  n^{1/6}})}{2\cosh(\tau_0+ \tau/n^{1/3}) }

1927: x,y\right)dy

1928:   \right|_{\mathcal{L}}

1929:  }

1930: % \\

1931: % &&\hspace*{7cm}

1932:  % d\left(\frac{  \sqrt{2n}+\frac{v}{\sqrt{2}

1933:  % n^{1/6}}}{2\cosh(\tau_0+ \tau/n^{1/3}) }\right)

1934:    \\

1935:   &=&

1936:  -\left(\frac{n^{1/3}}{2\pi}\right)^2

1937:  dv\int_{\mathcal{D}} d\tilde z\int_{\Gamma_L}

1938:  d\tilde w

1939:  e^{(-\frac{n}{2}\tilde z^2+\tilde z(2n+n^{1/3}u))}

1940:  e^{-(-\frac{n}{2}\tilde w^2+\tilde w(2n+n^{1/3}v))}

1941:  \\

1942:  &&\left(\frac{\tilde w}{\tilde z}\right)^n

1943:  \left(  \frac{\tilde w-\alpha e^{\tau/n^{1/3}}}{\tilde

1944:  w}\right)^r

1945:  \left(\frac{\tilde z-\alpha e^{\tau/n^{1/3}}}{\tilde z} \right)^{-r}

1946:  \frac{1}{n^{1/3}(\tilde w-\tilde z)}

1947: \\%  \eean\bean\\

1948: %

1949: %$$

1950: &=&- e^{\zeta n^{1/3}(u-v)}dv\int_0^{\iy}dy \frac{Z_n

1951: n^{1/3}}{2\pi}\int_\mathcal{D} d\tilde z~e^{-nF(\tilde z)~}

1952:  \left(\frac{\tilde z}{\tilde

1953: z-\alpha e^{\tau/n^{1/3}}}\right)^r

1954:  e^{n^{1/3}(u+y)(\tilde z-  \zeta)}

1955: \\

1956: &&%\eean

1957: %$$

1958: \frac{ n^{1/3}}{2\pi

1959: Z_n}\int_{\Gamma_L} d\tilde w~e^{ nF(\tilde w)~}

1960:  \left(\frac{\tilde w}{\tilde

1961: w-\alpha e^{\tau/n^{1/3}}}\right)^{-r}

1962:  e^{-n^{1/3}(v+y)(\tilde w-  \zeta)}

1963: .\eean

1964: %

1965: Conjugating the kernel, which leaves invariant the

1966: Fredholm determinant, one finds

1967: $$

1968:  e^{  n^{1/3}(v-u)} \left.

1969: H_n^{(r)}\left(x,y\right)\right|_{\LR}dy =-dv~e^{\gamma(u-v)}\int_0^{\iy}dy~\mathcal{I}_{\tau}^{(n)}(u+y)\mathcal{J}_{\tau}^{(n)}(v+y)

1970: $$

1971: with

1972: %

1973: \bea

1974:  \label{Integrals}

1975:  \mathcal{I}_{\tau}^{(n)}(x)&=&\frac{n^{1/3}Z_n}{2\pi}\int_\mathcal{D} d\tilde z~e^{-nF(\bar z)}\left( \frac{\tilde

1976: z}{\tilde z-\alpha e^{\tau/n^{1/3}}}\right)^re^{

1977: n^{1/3}x(\tilde z-\zeta)} \\

1978: \no\\

1979: \mathcal{J}_{\tau}^{(n)}(y)&=&\frac{n^{1/3}}{2\pi~Z_n}\int_{\Gamma_L}d\tilde

1980: w~e^{nF(\tilde w)}\left( \frac{\tilde w}{\tilde

1981: w-\alpha e^{\tau/n^{1/3}}}\right)^{-r}e^{ -n^{1/3}y(\tilde

1982: w-\zeta)} .\no\eea

1983: %

1984: Using the Taylor series (\ref{F}) for $F(z)$ and the value (\ref{Zn}) for $Z_n$, %defined above %(\ref{F})

1985: %

1986: %$$F(z)=\frac{z^2}{2}-2z+\log z,

1987: %$$

1988: %$nF(z)$

1989: % has a Taylor expression about $z=1$ of the form

1990: %

1991: %$nF(z)=nF(1)+\frac{n}{3}(z-1)^3+n{\bf O}(z-1)^4; $

1992:  one is led naturally to pick a new variable $u$ such that

1993: %

1994: $ \tilde z-1=\frac{u}{n^{1/3}}. $ Then

1995: %

1996: \bean \lefteqn

1997: {n^{1/3}Z_nd\tilde z~e^{-nF(\tilde z)}\left(\frac{\tilde z}{\tilde z-\alpha e^{\tau/n^{1/3}}}\right)^re^{n^{1/3}

1998: x(\tilde z-\zeta)}}%\eean\bean\\

1999: \\

2000: &=&n^{1/3}Z_n%n^{-r/3}e^{nF(1)}

2001: \frac{du}{n^{1/3}}\left(

2002: e^{-nF(1)-\frac{u^3}{3}}+ {

2003: O}\left(\frac{1}{n^{1/3}}\right)\!\!\right)

2004:  %\\

2005:  %\\

2006:  %&

2007: %&\qquad\qquad\qquad\qquad

2008: \\

2009: &&

2010: \hspace*{4cm} \left(\frac{u+n^{1/3}}{u+n^{1/3}\!-\!\alpha \left(

2011: 1\!+\!\frac{\tau}{n^{1/3}}\!+\!\ldots\right)n^{1/3}}

2012: \right)^re^{x(u-\gamma)}\\

2013: \\

2014: &=&\left\{\begin{array} {lll}e^{-x\gamma}du~e^{-\frac{u^3}{3}+xu}\left(

2015: \frac{1}{u-\tau}\right)^r+\mbox{lower order terms}&\mbox{ for } &

2016:  \alpha=1\\

2017:   e^{-x\gamma}du~e^{-\frac{u^3}{3}+xu}+\mbox{lower order terms}&\mbox{ for } &

2018: 0< \alpha<1.

2019: \end{array}\right.

2020: \eean

2021: %

2022: %\begin{picture}(40,40)

2023: % \put(40,40){\oval(40,40)[t1]}

2024: % \end{picture}

2025: %

2026: Upon remembering the definition of the functions $A^\pm _r(x,\tau)$ and the Airy function $A(x)$ and also the fact that $\alpha=\rho_0/\rho$, the rigorous saddle point argument, given in

2027: \cite{BBP,Peche}, yields

2028: %

2029: %

2030: \bea \lefteqn{ \lim_{n\rg\iy}\mathcal{I}_{\tau}^{(n)}(x)=}\no\\

2031: &&

2032: \left\{\begin{array}{l} \frac{1}{2\pi}

2033:  \int_{\mathcal{D}'}\frac{1}{(z-\tau)^r}e^{-\frac{z^3}{3}}e^{

2034:  x(z-\gamma)}dz=

2035: \mathcal{I}_{\tau}(x):=e^{-x\gamma}A^-_r(x,\tau)\\

2036: \qquad\qquad\qquad\qquad\qquad\qquad \mbox{ for }\quad

2037: \rho_0=\rho

2038:   \\

2039:    \frac{1}{2\pi}

2040:  \int_{\mathcal{D}'} e^{-\frac{z^3}{3}}e^{

2041:  x(z-\gamma)}dz= \mathcal{I}_{ }(x):=e^{-x\gamma}A(x)\\

2042:  \qquad\qquad\qquad\qquad\qquad\qquad\mbox{ for }\quad 0<\rho_0<\rho

2043:    \end{array} \right.\label{I}\\ \no\\

2044: \lefteqn{\lim_{n\rg\iy} \mathcal{J}_{\tau}^{(n)}(y)=}\no

2045: \\& &

2046: \left\{ \begin{array}{l}\frac{1}{2\pi}

2047: \int_{\mathcal{D}''}

2048: (z-\tau)^r

2049: e^{\frac{z^3}{3}}e^{-y(z-\gamma)}dz=\mathcal{J}_{\tau}(y):=-e^{y\gamma}A^+_r(y,\tau),\\

2050: \qquad\qquad\qquad\qquad\qquad\qquad\mbox{for }\quad

2051: \rho_0=\rho

2052:  \\

2053:  \frac{1}{2\pi}

2054: \int_{\mathcal{D}''}

2055: e^{\frac{z^3}{3}}e^{-y(z-\gamma)}dz=\mathcal{J}(y):=-e^{y\gamma}A(y),\\

2056: \qquad\qquad\qquad\qquad\qquad\qquad\mbox{for }\quad

2057: 0<\rho_0<\rho

2058: \end{array}\right.

2059: \label{J}

2060:  \eea

2061:   where

2062:  $\mathcal{D}'$ is a contour running from $\iy

2063:  e^{4i\pi/3}$ to $\iy

2064:  e^{2i\pi/3}$, with an indentation to the right of $\tau$

2065:  , such that $\tau$ is to the left of the contour,

2066: and where

2067:   $\mathcal{D}''$ is a contour running from $\iy

2068:  e^{-i\pi/3}$ to $\iy

2069:  e^{i\pi/3}$, with an indentation to the right of $\tau$,

2070:  such that $\tau$ lies also to the left of the contour.

2071:  Upon rotating the two contours and deforming $\mathcal{D}''$

2072:  slightly, since the integrand is pole-free, one

2073:  gets the final identities in the equations above.

2074: 

2075:  Therefore

2076:  $ \BP(\sup \mathcal {A}^{(r)}(\tau)\cap

2077:  E=\emptyset)=\det(I-K_\tau^{(r)})_E ,$

2078: %

2079: with

2080: %

2081: \bean

2082:   \lefteqn{

2083:  {K_\tau^{(r)}(u,v)dv}

2084: =\lim_{n\rg\iy}%e^{\gamma(u-v)}

2085:  e^{n^{1/3}(v-u)}

2086: H_n^{(r)}\left(x,y\right)dy\Bigr|_{\LR}

2087:  }\\

2088:     &=&

2089:     \left\{\begin{array}{l}

2090:    dv

2091:  \int^{\iy}_0 dw~A^-_r(u+w,\tau)A^+_r(v+w,\tau)

2092:  \qquad\mbox{  for  } \rho_0=\rho \\ \\

2093:    %

2094:    dv

2095:  \int^{\iy}_0 dw~A (u+w,\tau)A (v+w,\tau)

2096:  \qquad\mbox{  for  } 0<\rho_0<\rho.% \\

2097: \end{array}\right.\eean

2098: %

2099:  Baik \cite{Baik} has shown that the Fredholm determinant of

2100: the $r$-Airy kernel is a probability distribution, i.e.,

2101: $$\lim_{x\rg \pm \iy}

2102: \det\left(I-K_\tau^{(r)}\right)_{(x,\iy)}=\left\{{1

2103: \atop 0 .}\right.$$ This establishes Theorem

2104: \ref{Theo1}.\end{proof}

2105: 

2106: 

2107: 

2108: 

2109: %\newpage

2110: 

2111: 

2112: 

2113: % and \\

2114: %$lim_{x\rg -\iy} \det\left(I-K_\tau^{(r)}\right)_x=0.$

2115: %\end{proof}

2116: 

2117: \begin{remark} In this section one lets $\tau\rg -\iy$, which

2118: implies that $-i\tau$ remains above the contour $C$ and

2119: is thus compatible with the contour mentioned above.

2120: Letting $\tau\rg +\iy$ would require a drastic change of

2121: the functions $A^{\pm}_r$.

2122: \end{remark}

2123: %

2124: 

2125: 

2126: The next statement concerns the asymptotic

2127: behavior of the $r$-Airy kernel for $\tau\rg -\iy$,

2128:    \be

2129:  K^{(r)}_\tau(u,v)=\int_0^{\iy}

2130:  dw A_r^-(u+w; \tau)A_r^+(v+w; \tau)

2131:  ,\label{r-kernel2}\ee

2132:  where (remember)

2133: %

2134:  \be

2135: A_r^\pm(u; \tau) =

2136:  \int_C ~e^{\frac{1}{3} ia^3+iau }

2137:  \left( {\mp ia-\tau}\right)^{\pm r}

2138:   \frac{da}{2\pi}

2139:  ,\label{k-Airy}\ee

2140: %

2141:  where $C$ is a contour running from $\iy

2142:  e^{5i\pi/6}$ to $\iy

2143:  e^{i\pi/6}$, such that $-i\tau$ lies above the contour.

2144: %

2145: % \be

2146: %  K^{(r)}_t(u,v)=\int_0^{\iy}

2147: % dw A_r^-(u+w; t)A_r^+(v+w; t)

2148: % \label{t-kernel}\ee

2149: %$

2150: %for $t\rg -\iy$.

2151: This limit is compatible with the contour $C$

2152:  appearing in the definition of the functions

2153:  $A_r^{\pm}$, since then $-i\tau$ remains

2154:  above the contour $C$, as required.

2155: 

2156: 

2157: 

2158:  \begin{lemma}\label{lemma:kernels} Given the

2159:  ``initial condition" $$\lim_{\tau\rg

2160:  -\iy}K^{(r)}_\tau(u,v)=K^{(0)}(u,v):=

2161:  \frac{A(u)A'(v)-A'(u)A(v)}{u-v}=

2162:   \mbox{``Airy kernel"}

2163:  ,$$

2164:    the kernel

2165:  $K^{(r)}(u,v)$ behaves asymptotically for

2166:  $\tau\rg -\iy$, as

2167: $$ K^{(r)}(u,v)%&=&\int_0^{\iy}

2168: % du A_r^-(u+w; t)A_r^+(v+w; t)

2169: %\\

2170: =

2171:  K_{0}+\frac{K_1^{(r)}}{\tau}+\frac{K_2^{(r)}}{\tau^2}+

2172:  \frac{K_3^{(r)}}{\tau^3}+\ldots

2173: ,$$

2174:  where\footnote{Whenever $\pl /

2175:  \pl w$ appears with a negative exponent in the formula below, it is set $=0$.} for $n\geq 1$,

2176:  {%\footnotesize

2177:  \bean

2178: {K_n^{(r)}(u,v)}

2179: &=&-\frac{r^n}{n!}\left(\frac{\pl}{\pl w}\right)^{n-1}A(u+w)A(v+w)\Big|_{w=0}\\

2180: \\

2181: & &-\frac{r^{n-1}}{2(n-2)!}\left(\frac{\pl}{\pl

2182: w}\right)^{n-2}\!\!\Big(A'(u+w)A(v+w)\!-\!A

2183: (u+w)A'(v+w)\Big)\Big|_{w=0}\\

2184: \\

2185: & &-\frac{r^{n-2}}{(n-3)!}\left(

2186: \begin{array}{l}

2187: \frac{3n-1}{24}\left(\frac{\pl}{\pl w}\right)^{n-1}A(u+w)A(v+w)\\

2188: -\frac{n-1}{2}\left(\frac{\pl}{\pl

2189: w}\right)^{n-3}A'(u+w)A'(v+w)

2190: \end{array}\right)\Bigg|_{w=0}%\\

2191:  \\

2192:  & &

2193: +~(\mbox{polynomial of degree $n-3$ in $r$}).

2194:  \eean}

2195: %

2196: Although the kernel $K_\tau^{(r)}(u,v)$ involves

2197: integration, the terms $K_i^{(r)}(u,v)$ in the expansion

2198: never involve integration, they are quadratic in the

2199: Airy function and its derivatives; also the

2200: $K_i^{(r)}(u,v)$ are polynomials in $r$ of degree $i$,

2201: divisible by $r$, with alternately symmetric and

2202: skew-symmetric coefficients in $u$ and $v$, the top

2203: coefficient being symmetric.

2204: 

2205: \end{lemma}

2206: 

2207: 

2208: 

2209: 

2210: \medbreak

2211: %\newpage

2212: 

2213: \begin{proof} In order to expand the kernel (\ref{r-kernel})

2214: with regard to $\tau$ for $\tau\rightarrow -\iy$, set

2215: the expressions (\ref{k-Airy}) into the kernel

2216: (\ref{r-kernel}), which then becomes a triple integral.

2217: Set $\al=ia$ and $\beta=ib$ and consider the following

2218: Taylor expansions about $\tau= -\iy$, {%\footnotesize

2219:  \bea

2220: %\lefteqn

2221: {\frac{1}{(\al-\tau)^r(-\beta-\tau)^{-r}}}%\no\\

2222:  &=&\left(1+\frac{\frac{\al+\beta}{\tau} }{ 1-\frac{\al}{\tau}}\right)^r \no\\

2223: \no\\

2224:  &=&

2225:   1+\sum_{n=1}^\iy \frac{1}{\tau^n}

2226:      \sum_{j=0}^{\left[\frac{n-1}{2}\right]}

2227:    \left(\begin{array}{l}

2228:    (r(\al+\beta))^{n-2j}Q_{2j}(\al,\beta)\\

2229:    +(r (\al+\beta))^{n-2j-1} (\al-\beta)\tilde

2230:    Q_{2j}(\al,\beta)\end{array}

2231:    \!\!\!\!\!\right)

2232:    %\right)

2233: %

2234:  \no\\

2235:  &=&1\!+\!\sum_{n=1}^\iy \frac{1}{\tau^n}

2236:   \left(\!\!

2237:   \begin{array}{l}

2238:   \frac {(r(\al+\beta))^n}{n!}

2239:   +

2240:     \frac {(r (\al+\beta))^{n-1}(\al-\beta)}{2(n-2)!}

2241:    \\

2242:    +

2243:     \frac {(r(\al+\beta))^{n-2}}{(n-3)!}

2244:     \left(\frac{(3n-1)(\al+\beta)^2}{24}

2245:      -\frac{(n-1)\al\beta}{2}\right)+\ldots

2246:   \end{array}\!\!\!\!\!\right)

2247:  % \no\\%

2248: %  \eean%\bea%\no\\ \no\\

2249: %&=&1+\frac{r}{\tau}(\al+\beta)%\no\\

2250: %\no\\

2251: %& &

2252: %+\frac{r}{2!\tau^2}(\al+\beta)(r(\al+\beta)+\al-\beta)

2253: % \no\\

2254: % &&

2255: %+\frac{1}{3!\tau^3}\left(\begin{array}{l}

2256:  %r^3(\al+\beta)^3+3r^2(\al+\beta)^2(\al-\beta)

2257:  %\\+2r(\al+\beta)((\al+\beta)^2-3\al \beta)\end{array}\right)

2258: %

2259: %\no\\

2260: %&& +\frac{1}{4!\tau^4}

2261: % \left(\begin{array}{l}

2262: %  r^4(\al+\beta)^4+6r^3(\al+\beta)^3(\al-\beta)\\

2263: %  +r^2(\al+\beta)^2(11(\al+\beta)^2-36\al \beta)

2264: %  +6r( \al ^4-\beta^4)\end{array}\right)

2265:  %\no\\&&

2266: %  \!+\!{\bf

2267: %O}\left(\frac{1}{\tau^5}\right) %\no\\

2268: \label{powerseries},\eea}

2269: %

2270: where $Q_{2j}$ and $\tilde Q_{2j}$ are symmetric

2271: homogeneous polynomials of degree $2j$ in the arguments,

2272: since the first expression is invariant under the

2273: involution $r\mapsto -r$ and $\al\mapsto -\beta$. The

2274: coefficients of $1/\tau^n$ are divisible by $r$, for the

2275: simple reason that for $r=0$, the expression above

2276: equals $1$.

2277: %

2278: %

2279: %Notice that $\al+\beta$ and $r$ must factor out of every

2280: %term in the expansion (\ref{powerseries}), since

2281: %$$\left.\frac{1}{(\al-t)^r(-\beta-t)^{-r}}\right|_{\al=-\beta}=

2282: % \left.\frac{1}{(\al-t)^r(-\beta-t)^{-r}}\right|_{r=0}=1$$

2283: %Also $\al-\beta$ appears alternately in the coefficients

2284: %of $1/t^k$, because

2285: % $$\left.\frac{1}{(\al-t)^r(-\beta-t)^{-r}}\right|_{\al=\beta}$$

2286: % has an expansion in $r^it^{-j}$, with only those

2287: % monomials

2288: % surviving with $i+j=$even.

2289: 

2290:    Also notice

2291: multiplication by $ia$ of the integrand in the kernel

2292: (\ref{r-kernel}),

2293: %

2294:  \bean K^{(r)}_\tau(u,v)\!\!\!&=&\!\!\!\int_0^{\iy}\!\!

2295:  dw \!\!\int_C   e^{\frac{1}{3} ia^3+ia(w+u) }

2296:  \left(\frac{1}{ia-\tau}\right)^r

2297:  \frac{da}{2\pi}

2298:   \int_C ~e^{\frac{1}{3} ib^3+ib(w+v) }

2299:  \left( {-ib-\tau}\right)^r

2300:   \frac{db}{2\pi}\\

2301:   \!\!\!&=&\!\!\!

2302:   \int_0^{\iy}

2303:  \!\!

2304:  dw \!\!\int_C  \int_C %\frac{da}{2\pi}%\int_C

2305:   \frac{dadb}{4\pi^2}

2306: e^{\frac{1}{3} ib^3+ib(w+v)}  e^{\frac{1}{3}

2307: ia^3+ia(w+u) }(1+\frac{r}{\tau}(ia+ib)+\ldots)

2308:   \eean

2309:  can be realized by taking $\frac{\pl K^{(0)}

2310: }{\pl u}$ and similarly multiplication of the integrand by $ib$ is realized by taking

2311: $\frac{\pl K^{(0)} }{\pl v}$; thus we have the following

2312: recipe

2313: \bean %&&

2314: ia &\leftrightarrow& \frac{\pl K^{(0)}

2315: }{\pl

2316: u}=\int_0^{\iy}dwA'(u\!+\!w)A(v\!+\!w)   \\

2317:   %,~~~~~\hspace*{.3cm}~ 

2318:   ib &\leftrightarrow& \frac{\pl K^{(0)} }{\pl

2319: v}=\int_0^{\iy}dwA(u\!+\!w)A'(v\!+\!w), \eean and so in

2320: particular, \bean

2321: %\lefteqn

2322: {(ia)^{k_1}(ib)^{k_2}(ia\!+\!ib)^n}%\\ \\

2323: %

2324:  &\Leftrightarrow&

2325:   \int_0^{\iy}\left(\frac{\pl}{\pl u}\!+\!\frac{\pl}{\pl v}\right)^n

2326:  \left(\frac{\pl }{\pl u }\right)^{k_1}

2327:  \left(\frac{\pl}{\pl v}\right)^{k_2}

2328:   A(u\!+\!w)A(v\!+\!w)dw\\

2329: %  &&=

2330: %   \int_0^{\iy}\left(\frac{\pl}{\pl

2331: %   w}\right)^nA^{(k_1)}(u+w)A^{(k_2)}(v+w)dw\\

2332:    &&=

2333:    -\left(\frac{\pl}{\pl

2334:    w}\right)^{n-1}A^{(k_1)}(u+w)A^{(k_2)}(v+w)\Bigr|_{w=0}

2335:   . \eean

2336: Notice that, since $ia+ib$ factors out of every term in

2337: the expansion (\ref{powerseries}), the kernels obtained

2338: never contain integration. In addition, since $ia-ib$

2339: factors out of every other term, every other term in

2340: $K^{(r)}_i(u,v)$ must be skew; in particular it vanishes

2341: for $u=v$.

2342:  One then reads off the $K^{(r)}_i(u,v)$'s from the expansion

2343: (\ref{powerseries}) and the recipe above, upon using

2344: occasionally the differential equation $A''(x)=xA(x)$

2345: for the Airy function, thus

2346: %

2347:  ending the proof of Lemma \ref{lemma:kernels}.

2348: \end{proof}

2349: 

2350: 

2351: \begin{remark} As an example, we give explicit expressions for

2352: the first few $K_i^{(r)}(u,v)$'s:

2353: %

2354: {% \footnotesize

2355: \bea

2356: && \\

2357: \label{Ki's} K_0^{(r)}(u,v)&=&K^{(0)}(u,v)=\frac{A(u)A'(v)-

2358: A'(u)A(v)}{u-v}

2359: %\eean} {\bean%

2360:   \no\\ %\no\\

2361: K_1^{(r)}(u,v)&=&-r~A(u)A(v)%\no\\

2362: \no\\

2363: K_2^{(r)}(u,v)&=&-\frac{r^2}{2}(A'(u)A(v)+A(u)A'(v))+\frac{r}{2}(u-v)K^{(0)}(u,v)

2364: \no\\ %\no\\

2365: K_3^{(r)}(u,v)&=&-\frac{r^3}{6}(A''(u)A(v)+2 A

2366: '(u)A'(v)+A(u)A''(v)) +\frac{r^2}{2}(v-u)A(u)A(v)

2367: \no\\%\eean\bea%\no\\ \no\\

2368: & &-\frac{r}{3}\left( A''(u)

2369: A(v)+A(u)A''(v)-A'(u)A'(v)\right).

2370: \no   % \no\\

2371: % \no\\

2372: % K_4^{(r)}(u,v)

2373: %&=&

2374: % -\frac{r^4}{24}(A'''(u)A(v)+3 A ''(u)A'(v)+3 A '(u)A''(v)

2375:  %+A(u)A'''(v))\no\\

2376: % &&+\frac{r^3}{4}(A(u)A'''(v)+A'(u)A''(v)-A'''(u)A(v)-A''(u)A'(v))

2377: %\no\\&&+\frac{r^2}{24}

2378: % \left( 3(A''(u)A'(v)+A'(u)A''(v))

2379: %  -11(A'''(u)A(v)+A(u)A'''(v))\right)\no\\

2380:   %

2381: %&&+\frac{r}{4}

2382: % (A(u)A'''(v)+A''(u)A'(v)-A'''(u)A(v)-A'(u)A''(v)).

2383: \eea}

2384: \end{remark}

2385: 

2386: 

2387: In order to find the PDE for the transition probability, one will need an estimate on how the actual transition probability for the finite problem converges for $n\rg \iy$. This will be used in (\ref{6.3}).

2388: 

2389: \begin{corollary}\label{cor:estimate2}

2390:  For $x\in {\mathbb R} $ sufficiently large, one has for some constant $C>0$,

2391: %

2392: \bea

2393: \lefteqn{\hspace*{-2cm} \left|\log \BP^{(0,\rho_0\sqrt{\frac{n}{2}})}_{Br}\left(\mbox{all}~

2394: x_i\left(\! \frac{1}{1\!+\!e^{-2(\tau_0\!+\!\frac{\tau}{n^{1/3}})}}\!\right)

2395: \! \leq \!

2396:   \frac{  \sqrt{2n}+\frac{x}{\sqrt{2}

2397:   n^{1/6}}}{2\cosh(\tau_0\!+\! \frac{\tau}{n^{1/3}})}\right)

2398:   %}\right.

2399: \right. }     \label{estimate1} \\

2400:  &&\no

2401:  %\hspace*{5cm} ~~~~~~~~~~~~~~~

2402:  ~~~\left.-\log \BP(\sup \mathcal {A}^{(r)}(\tau)\leq x)\right|  \leq Cn^{-1/3}.

2403:   %

2404:   \eea

2405: %   \be\hspace*{9cm}\leq Cn^{-1/3}

2406: %   \ee

2407: \end{corollary}

2408: 

2409: 

2410: \begin{proof}

2411: For any trace class operators $K_n$ and $K_\iy$, set

2412: \footnote{Define the three norms on a Hilbert space: the sup, the trace and the Hilbert-Schmidt norms, with $ \vv T \vv \leq  \vv T \vv_{_{HS}} \leq  \vv T \vv_{_{\tr}}$ :

2413: $$\vv T \vv=\sup_{v \in \mathcal {H}}\frac{|Tv|}{|v|}

2414: =\sup |\lb_i|,~

2415:  \vv T \vv_{_{\tr}}=\Tr (T^*T)^{1/2}=

2416:   \sum |\lb_i|,~ \vv T \vv_{_{HS}}=

2417:   (\Tr T^*T)^{1/2}=(\sum |\lb_i|^2)^{1/2}.$$}

2418:  %

2419:   $$

2420:  M:= \max ( \vv K_n  \vv, \vv K_\iy  \vv).

2421:   $$

2422:   Then assuming $M<1$,

2423:   one checks, using $K_n^i-K_\iy^i=\sum_{\ell=0}^{i-1}

2424:    K_\iy^\ell (K_n-K_\iy)K_n^{i-1-\ell}$ and $\vv AB\vv_{_{\tr}}\leq \vv A\vv~ \vv B\vv_{_{\tr}}$ and $\vv AB\vv_{_{\tr}}\leq\vv A\vv_{_{\tr}} \vv B\vv

2425:    $,

2426: \bea

2427:   \label{estimate} \hspace*{.4cm}\left| \log \det (I-K_n)-\log \det (I-K_\iy)\right|

2428:   &=&

2429:     \left| \tr(\log (I-K_n)-\log  (I-K_\iy))\right|

2430:    \\

2431:      &\leq&

2432:    \sum_1^{\iy}

2433:        \left|

2434:         \tr(\frac{K_n^i-K_\iy^i}{i})

2435:         \right|

2436:         \no\\

2437:         &\leq&

2438:         \sum_1^{\iy} \frac{1}{i}\vv

2439:       K_n^i-K_\iy^i    \vv_{_{\tr}}

2440:       \no\\

2441:        &\leq&

2442:       \vv

2443:       K_n-K_\iy   \vv_{_{\tr}}

2444:       \sum_0^{\iy}

2445:        M^i

2446:        \no\\

2447:         &\leq&

2448:          \frac{\vv

2449:      K_n-K_\iy   \vv_{_{\tr}}}{1-M}.

2450:     \no

2451: \eea

2452: %

2453: Setting (see notation (\ref{Integrals}))

2454: \bean

2455:  K_n(u,v)&:=&  \int_0^\iy dy \mathcal{I}_\tau^{(n)} (u+y)

2456:                       \mathcal{J}_\tau^{(n)} (v+y)\\

2457:   K_\iy(u,v)&:=&

2458:                        e^{(v-u)\gamma} \int_0^\iy dy A_r^-(u+y,\tau)  A_r^+(v+y,\tau)

2459:             ,  \eean

2460: one checks

2461: \bean

2462: \lefteqn{\left |

2463: K_n(u,v)-K_\iy(u,v) \right |

2464: }\\

2465: %

2466: &\leq &

2467:  \left |

2468:  \int_0^\iy dy \Bigl(\mathcal{I}_\tau^{(n)} (u+y)-e^{-(u+y)\gamma} A_r^-(u+y,\tau)\Bigr)

2469:    \mathcal{J}_\tau^{(n)} (v+y)\right |

2470:  \\

2471: &&+\left|\int_0^\iy dy e^{-(u+y)\gamma}A_r^-(u+y,\tau)

2472:  \Bigl(\mathcal{J}_\tau^{(n)} (v+y)-e^{(v+y)\gamma} A_r^+(v+y,\tau)\Bigr)\right|

2473: .\eean

2474:  %

2475:   An argument similar to the one of Baik-BenArous-P\'ech\'e \cite{BBP,Peche} shows that

2476: for given $x_0\in {\mathbb R}$, there are constants $C>0,~N>0$,

2477: \bean

2478:  \left | \mathcal{I}_\tau^{(n)} (x)-e^{-x\gamma} A_r^-(x,\tau) \right |

2479:   &\leq& \frac{Ce^{-cx}}{n^{1/3}} \\

2480:   \left | \mathcal{J}_\tau^{(n)} (x)-e^{x\gamma} A_r^+(x,\tau) \right |

2481:   &\leq& \frac{Ce^{-cx}}{n^{1/3}} ~,~~\mbox{for $n\geq N$ and $x\geq x_0$}.

2482:      \eean

2483: %for $N$ and $x$ sufficiently large.

2484: %

2485:  Viewing the functions in the integrals on the right hand side

2486:  as kernels representing H\"ankel-like

2487: integral operators on $(x_0,\iy)$, one has, using the

2488: inequality $\vv AB  \vv_{_{\tr}}\leq \vv A

2489: \vv_{_{HS}} \vv B \vv_{_{HS}}$ for

2490: Hilbert-Schmidt operators,

2491:   \bean

2492:    \vv K_n-K_\iy  \vv_{_{\tr}}

2493:     &\leq &

2494:     \vv   \mathcal{I}_\tau^{(n)} (u+y)-e^{-(u+y)\gamma} A_r^-(u+y,\tau)    \vv_{_{HS}}   \vv     \mathcal{J}_\tau^{(n)} (v+y)       \vv_{_{HS}}

2495:   \\  \\ && +

2496:      \vv   e^{-(u+y)\gamma}A_r^-(u+y,\tau)     \vv_{_{HS}}

2497:       \vv    \mathcal{J}_\tau^{(n)} (v+y)-e^{(v+y)\gamma} A_r^+(v+y,\tau)       \vv_{_{HS}}

2498:   \\

2499:   &\leq & 2C'n^{-1/3},

2500:       \eean

2501:      where $$C'= \max (\vv  e^{-(u+y)\gamma}A_r^-(u+y,\tau)   \vv_{_{HS}}, \sup_{n\geq N}\vv    \mathcal{J}_\tau^{(n)} (v+y)    \vv_{_{HS}})Ce^{-cx_0}.$$

2502: 

2503: 

2504: 

2505:      Then, from (\ref{estimate}), one has that

2506:       $$\left| \log \det (I-K_n)-\log \det (I-K_\iy)\right|\leq \frac{2C'}{1-M} n^{-1/3}

2507:       .$$

2508:    Remembering the representation of the probabilities in the statement (\ref{estimate1}) in terms of Fredholm determinants

2509:    %

2510:    establishes Corollary \ref{cor:estimate2}.\end{proof}

2511: 

2512: 

2513: %Then for $n$ sufficiently large, we have, using the integrals (\ref{Integrals})

2514: %$$\vv \mathcal{I}_{\tau}-{\cal

2515: %I%}_{\tau}^{(n)}\vv_{hs}\leq C  n^{-1/3}

2516: % ~~\mbox{and}~~

2517: % \vv \mathcal{J}_{\tau}-{\cal

2518: %J}_{\tau}^{(n)}\vv_{hs}\leq C  n^{-1/3}

2519: % $$

2520: % and thus

2521: % $$\vv  \mathcal{I}_{\tau}\mathcal{J}_{\tau}-

2522: %  \mathcal{I}^{(n)}_{\tau}\mathcal{J}^{(n)}_{\tau}\vv_{_{\tr}}

2523: %  \leq 2C \max(\vv{\cal

2524: %I}_{\tau} \vv,\vv{\cal

2525: %J}_{\tau} \vv)n^{-1/3}$$

2526: 

2527: %  Therefore, combining with (\ref{estimate}), one finds for some appropriate %constant $C'$

2528: %  \bean

2529: % \left| \log \det (I-\mathcal{I}_{\tau}\mathcal{J}_{\tau})-\log \det (I-\mathcal{I}^{(n)}_{\tau}\mathcal{%J}^{(n)}_{\tau})\right|

2530: % &\leq&

2531: %    2C' n^{-1/3}

2532:  % %

2533:  % \eean

2534: 

2535: 

2536: 

2537: %trace norm of T is the sum of the eigenvalues of $(T* T)^(1/2).$

2538: 

2539: 

2540: 

2541: 

2542: 

2543: 

2544: 

2545: %\newpage

2546: 

2547: 

2548: 

2549: 

2550: 

2551: %\vspace*{-1cm}

2552: 

2553: \section{An integrable deformation of Gaussian random

2554: ensemble with external source and 3-component

2555: KP}\label{sect:KP}

2556: 

2557: 

2558: The connection between the Gaussian random ensemble with

2559: external source and the multi-component KP hierarchy is

2560: explained in \cite{AvM-Pearcey} and \cite{AvM-Mops}. The

2561: main ideas are sketched in this section. For the multicomponent KP hierarchy, see [30].

2562: %

2563: %\bigbreak

2564: %

2565: \subsection{Two sets of weights and the $p+q$-KP

2566: hierarchy} Define two sets of weights

2567: %

2568: $$

2569:  \psi_1(x),\ldots,\psi_q(x)~~~\mbox{and}~~~

2570:  \varphi_1(y),\ldots,\varphi_p(y),~~\mbox{with}~x,y\in

2571:  {\mathbb R},

2572: $$

2573: %

2574: and deformed weights depending on {\it time} parameters

2575: $s_\al=(s_{\al 1},s_{\al 2},\ldots)$ ($1\leq \al\leq q$)

2576: and $t_\beta=(t_{\beta 1},t_{\beta 2},\ldots)$ ($1\leq

2577: \beta \leq p$), denoted by

2578: %

2579: \be

2580:   \psi_\al^{-s}(x):=\psi_\al(x)e^{-\sum_{k=1}^\iy s_{\al k}x^k}

2581:    ~~~~~\mbox{and}~~~~~

2582:   \varphi_\beta^{t}(y):=\varphi_\beta (y)e^{\sum_{k=1}^\iy t_{\beta

2583:    k}y^k}.

2584: \label{weights}\ee

2585: %

2586: That is, each weight goes with its own set of times. For

2587: each set of positive integers\footnote{$\vert m

2588: \vert=\sum_{\al=1}^{q}m_\al$ and $\vert

2589: n\vert=\sum_{\beta=1}^{p}n_\beta$.}

2590: %

2591: $

2592:   m=(m_1,\ldots,m_q),~n=(n_1,\ldots,n_p)\mbox{~ with~}\vert m\vert=\vert n\vert,%\sum_1^qm_i=\sum_1^pn_i,

2593: $

2594: %

2595: consider the determinant of a moment matrix $T_{mn}$ of

2596: size $\vert m\vert=\vert n\vert$, composed of $pq$

2597: blocks of sizes $m_in_j$; the moments are taken with

2598: regard to a (not necessarily symmetric) inner product

2599: $\INN$

2600: %

2601: \bea%\label{intro1}

2602:   \label{Tmn_def}  \lefteqn{\hspace*{.5cm}\tau_{mn}(s_1,\ldots,s_q;t_1,\ldots,t_p)}

2603:   \hskip1cm \\ %\nonumber\\ \nonumber

2604:    &&  \hspace*{-1.5cm}

2605:   := \det\left(\!\!\!\!\begin{array}{ccc}

2606:   \Bigl(\!

2607:   \inn{x^i\psi^{-s}_1(x)} {y^j\varphi^{t}_1(y) }

2608:   \Bigr)_{0\leq i<m_1\atop{0\leq j<n_1}}&\ldots&\left(

2609:   \inn{x^i\psi^{-s}_1(x)} {y^j\varphi^{t}_p(y) }

2610:   \!\right)_{0\leq i<m_1\atop{0\leq j<n_p}}\\

2611:   %& & \\

2612:   \vdots& &\vdots\\

2613:   & &\\

2614:   \left(\!\inn{x^i\psi^{-s}_q(x)}

2615:   {y^j\varphi^{t}_1(y) } \right)_{0\leq i<m_q\atop{0\leq j<n_1}}&\ldots&\left(\inn{x^i\psi^{-s}_q(x)}

2616:   {y^j\varphi^{t}_p(y)}\! \right)_

2617:   {0\leq i<m_q\atop{0\leq j<n_p}}

2618:   \end{array}\!\!\!\!\right).

2619:  \no

2620: %

2621: \eea

2622: 

2623: %\vspace{-.5cm}

2624: 

2625: We now state a non-trivial Theorem involving a relationship between the determinants of the block moment matrices above, by increasing or decreasing the sizes of the blocks by one. Modifying the size $n_\beta$ in $n=(n_1,\ldots, n_p)$ by $1$ is indicated by $n\mapsto n\pm e_\beta$, where $e_\beta=(0,\ldots,0, 1,0,\ldots,0)$, with $1$ at place $\beta$. The proof and many simple examples can be found in \cite{AvM-Mops}:

2626: 

2627: \begin{theorem}\label{Theo:p+q-KP}{\em (Adler, van Moerbeke and Vanhaecke \cite{AvM-Mops})}

2628: %

2629: Then the block matrices $\tau_{mn}$ satisfy the $(p+q)$-KP hierarchy; to be precise, the functions $\tau_{mn}$ satisfy the bilinear relations\footnote{The integrals are contour integrals along a small circle about $\iy$, with formal Laurent series as the integrand. Also, for $z \in \BC$, we define $[z^{-1}]:= (\frac{z^{-1}}{1},\frac{z^{-2}}{2},\frac{z^{-3}}{3},\ldots)$. For a given polynomial $p(t_1,t_2,\dots)$, the Hirota symbol between functions $f=f(t_1,t_2,\ldots)$ and

2630: $g=g(t_1,t_2,\ldots)$ is defined by

2631: %

2632: $

2633:   p(\frac{\pl}{\pl t_1},\frac{\pl}{\pl t_2},\dots)f\circ g:=

2634:   p(\frac{\pl}{\pl y_1},\frac{\pl}{\pl y_2},\dots)

2635:   f(t+y)g(t-y)\Bigl|_{y=0}.

2636: $

2637: %

2638: We also need the elementary Schur polynomials

2639: ${\gs}_{\ell}$, defined by

2640: $e^{\sum^{\iy}_{1}t_kz^k}:=\sum_{k\geq 0} {

2641: \gs}_k(t)z^k$ for $\ell\geq 0$ and ${\gs}_{\ell}(t)=0$

2642: for $\ell<0$; moreover, set

2643: %

2644: $

2645:   \gs_{\ell}(\tilde \pl_t):={\gs}_{\ell}(\frac{\pl}{\pl t_1},

2646:   \frac{1}{2}\frac{\pl}{\pl t_2},\frac{1}{3}\frac{\pl}{\pl t_3},

2647:   %

2648:   \ldots).

2649: $

2650:  }

2651: %

2652:     \be \label{tau-bilinear identity}\ee

2653:    \vspace*{-.7cm}

2654:       \bean

2655: %

2656:   \sum_{\beta=1}^p\oint_\infty

2657:   % \scriptstyle

2658:   (-1)^{\sigma_\beta(n)} \tau_{m,n-e_{\beta}}

2659:     (t_{\beta}-[z^{-1}])\tau_{m^*,n^*+e_{\beta}}(t^*_{\beta}+[z^{-1}])

2660:           e^{\scriptscriptstyle\sum_1^{\iy} (t_{\beta k}-t_{\beta k}^*)z^k}z^{n_\beta-n_\beta^*-2}\,dz=\nonumber\\

2661: %

2662:   \sum_{\al=1}^q\oint_\infty

2663:  % \scriptstyle

2664:  (-1)^{\sigma_\al(m)}\,\tau_{m+e_\al,n}(s_{\al}-

2665:   [z^{-1}])\tau_{m^*-e_\al

2666:     ,n^*}(s^*_{\al}+[z^{-1}])

2667:     e^{\sum_1^{\iy} (s_{\al k}-s_{\al k}^*)z^k}\,

2668:     z^{m_{\al}^*-m_\al-2}\,dz,

2669: %

2670:   \eean

2671: %

2672:   for all $m,n,m^*,n^*$ such that

2673:   $\vert m^*\vert=\vert n^*\vert+1$

2674: and $\vert m\vert=\vert

2675:   n\vert-1$ and all $s,t,s^*,t^*

2676: \in \BC^{\iy}$ and where

2677: %

2678: %

2679: %

2680:    $

2681:      \sigma_\al(m)={\sum_{\al'=1}^\al (m_{\al'}

2682: -m_{\al'}^*)}\quad\hbox{and}\quad

2683:      \sigma_\beta(n)={\sum_{\beta'=1}^\beta(n_{\beta'}-n_{\beta'}^*)}.

2684:    $%

2685:    \end{theorem}

2686: 

2687: Computing the residues in the contour integrals above,

2688: the functions $\tau_{mn}$, with $\vert m\vert=\vert

2689: n\vert$, satisfy the following PDE's %conveniently

2690: %expressed

2691: in terms of the Hirota symbol, defined in the footnote 11:

2692: %above:

2693: %

2694: {%\footnotesize

2695:  \bea

2696:   \tau_{mn}^2\frac{\pl^2}{\pl t_{\beta,\ell+1}\pl t_{\beta',1}}

2697:   \log \tau_{mn}

2698:   &=&

2699:   \gs_{\ell+2\delta_{\beta\beta'}}\bigl(\tilde\pl_{t_{\beta}}

2700:   \bigr) \tau_{m,n+e_{\beta}-e_{\beta'}}

2701:   \circ\tau_{m,n+e_{\beta'}-e_{\beta}}  \nonumber\\

2702:   %\eean\bea

2703:   \tau_{mn}^2 \frac{\pl^2}{\pl s_{\al,\ell+1}\pl

2704:   s_{\al',1}}\log

2705:   \tau_{mn}

2706:   &=&\gs_{\ell+2\delta_{\al\al'}}(\tilde\partial_{s_{\al}})

2707:      \tau_{m+e_{\al'}-e_{\al},n}

2708:      \circ\tau_{m+e_{\al}-e_{\al'},n} %\nonumber\\

2709: \no\\ \label{introa}%

2710:   \tau_{mn}^2 \frac{\pl^2}{\pl s_{\al,1}\pl t_{\beta,\ell+1}}

2711:   \log \tau_{mn}

2712:   &=&-\gs_{\ell}(\tilde\partial_{t_{\beta}})\tau_{m+e_{\al},

2713:   n+e_\beta}\circ\tau_{m-e_{\al},n-e_{\beta}}\nonumber\\

2714: %

2715:   \tau_{mn}^2\frac{\pl^2}{\pl t_{\beta,1}\pl s_{\al,\ell+1}}

2716:   \log \tau_{mn}

2717:   &=&-\gs_{\ell}(\tilde\partial_{s_{\al}})

2718:   \tau_{m-e_{\al},n-e_\beta}\circ\tau_{m+e_{\al},n+e_{\beta}}.

2719: %

2720: \eea}

2721: %

2722: %\newpage

2723: %

2724: \subsection{Gaussian ensemble with external source}

2725: Consider an ensemble of $n\times n$ Hermitian matrices

2726: with an external source, given by a diagonal matrix

2727:  $

2728:  A=\diag(a_1,\ldots,a_n)

2729:   $

2730:  and a general potential $V(z)$, with density

2731: %

2732:   $$

2733:   \BP_n(M\in [M,M+dM])=

2734:   \frac{1}{Z_{n}}

2735:   e^{-\Tr(V(M)-AM) }dM.

2736:   $$

2737: For a subset $E\subset {\mathbb R}$, %the disjoint union of

2738: %intervals

2739:   %$

2740:  % E =\bigcup^r_{i=1}[b_{2i-1},b_{2i}]

2741: %  $,

2742: %

2743: the following probability can be transformed by the

2744: Harish-Chandra-Itzykson-Zuber formula, with

2745: $D:=\diag(z_1,\ldots,z_n$), $$\Dt_n(z):=\prod_{1\leq

2746: i<j\leq n}(z_i-z_j),$$ and all distinct $a_i$,

2747:  %

2748:  \bea

2749:   % \lefteqn{

2750:  \label{1.1}  \BP_n(\mbox{spectrum} ~M\subset E)

2751:  &=&\frac{1}{Z_n} \int_{\mathcal{H}_{n}(E)} e^{-\Tr(V(M)-AM) }dM

2752:   % }

2753:   \\

2754:    && \no\\

2755:   &=&\frac{1}{Z_n} \int_{E^n} \Dt^2_n(z)\prod_1^n

2756: e^{-V(z_i)}dz_i\int_{U(n) }e^{\Tr AUD

2757: U^{-1}}dU    \no\\

2758:   %\eean \bea

2759: &=&\frac{1}{Z'_n} \int_{E^n} \Dt^2_n(z)\prod_1^n

2760: e^{-V(z_i)}dz_i\frac{\det[e^{a_iz_j}]_{1\leq i,j\leq

2761: n}}{\Dt_n(z)\Dt_n(a)}  \no\\

2762:   &=&\frac{1}{Z''_n}

2763: \int_{E^n}\Dt_n(z)\det[e^{-V(z_j)+a_iz_j}]_{1\leq

2764: i,j\leq n}\prod_1^n dz_i,%\no\\

2765: \no\eea

2766: %

2767: %Consider this integral, when

2768: % $$

2769: % A=\diag ().

2770: % $$

2771: with $a_i\neq a_j$ and the Vandermonde

2772: $\Dt_n(z)=\prod_{1\leq i<j\leq n}(z_i-z_j)$. The formula

2773: remains valid in the limit, when some $a_i$'s coincide,

2774: upon making differences of rows and dividing by the

2775: appropriate $(a_i-a_j)$'s. In the following Proposition,

2776: we consider a general situation, of which (\ref{1.1})

2777: with $A=\diag (a,\ldots,a,0,\ldots,0)$ is a special

2778: case, by setting $\varphi^+=e^{az}$ and $\varphi^-=1$.

2779:  Consider the Vandermonde determinant %of the variables $

2780: % x_1,\ldots,x_{k_1}

2781:  %,y_1,\\ \ldots, y_{k_2}$, namely

2782:  $\Dt_n(x,y)

2783:  :=\Dt_n(x_1,\ldots,x_{k_1}

2784:  ,y_1,\ldots, y_{k_2}). $

2785: Then we have the following (see \cite{AvM-Pearcey}):

2786: 

2787: \begin{proposition}

2788: Given an arbitrary potential $V(z)$ and arbitrary

2789: functions

2790:  $\varphi^+(z)$ and $\varphi^-(z)$, define

2791:  ($n=k_1+k_2$)

2792: \bean

2793: (\rho_1,\ldots,\rho_n)&:=&e^{-V(z)}\left(\varphi^+(z),z\varphi^+(z)

2794: ,\ldots,   z^{k_1-1}\varphi^+(z),  \right.

2795:  \\

2796:  &&\hspace*{2cm} \left.\right.%\\

2797: %&&~~~~~~~~~~~~~~~~~~~~~~~

2798: \left.~\varphi^-(z),

2799: ~z\varphi^-(z),\ldots,z^{k_2-1}\varphi^-(z)\right)

2800:  .

2801:  \eean

2802:  We have

2803: %

2804:  \bea

2805:  \label{1.3}\lefteqn{\hspace*{.6cm}\frac{1}{n!}\int_{E^n}\Dt_n(z) \det(\rho_i(z_j))_{1\leq i,j\leq

2806: n}

2807:    \prod_1^n%e^{-z_i^2/2}

2808:     dz_i}

2809:      \\

2810: &=&\frac{ 1}{k_1!k_2!}\int_{E^n}\Dt_n(x,y)

2811: \Dt_{k_1}(x)\Dt_{k_2}(y)\prod_1^{k_1}\varphi^+(x_i)

2812: e^{-V(x_i)}dx_i

2813: \prod_1^{k_2}\varphi^-(y_i)e^{-V(y_i)}dy_i

2814:  %

2815: \no\\%\eea\bea %\no\\

2816: &=& \det\left(\begin{array}{c}

2817: \left(\displaystyle{\int_{E}}z^{i+j}\vp^+(z)e^{-V(z)}\right)_{{\begin{array}{l}

2818:      0\leq i\leq k_1-1\\

2819:      0\leq j\leq k_1+k_2-1

2820:      \end{array}}} \\

2821:   \\

2822: \left(\displaystyle{\int_{E}}z^{i+j}\vp^-(z)e^{-V(z)}\right)_{{\begin{array}{l}

2823:      0\leq i\leq k_2-1\\

2824:      0\leq j\leq k_1+k_2-1

2825:      \end{array}}}

2826: \end{array}\right)

2827:  .\no

2828:  \eea

2829: 

2830: 

2831: \end{proposition}

2832: 

2833: 

2834:  %\newpage

2835: 

2836: \subsection {Adding extra-variables $t,~s~u,~ \mbox{and}~ \beta$}

2837: 

2838: We add the extra-variables $t=(t_1,t_2,\ldots),~~

2839:                        s=(s_1,s_2,\ldots),~~

2840:                        u=(u_1,u_2,\ldots)~~\mbox{and}~~

2841:                        \beta$ in the

2842: exponentials, as follows (%$E

2843: %=\bigcup^r_{i=1}[c_{2i-1},c_{2i}],~~

2844: $n=k_1+k_2$),

2845:  %\be

2846: \bea

2847:    V(z):=\frac{z^2}{2}-\sum_1^{\iy}t_i z^i  \hspace*{2cm}

2848:   \no\\

2849:    \varphi^+(z)= e^{ az+\beta z^2

2850:   -\sum_1^{\iy}s_iz^i},

2851:    ~~~~~~~\varphi^-(z)= e^{%-az-\beta z^2

2852:   -\sum_1^{\iy}u_iz^i}.\label{extra-var}

2853:  \eea

2854:  The determinant of the moment matrix (\ref{Tmn_def}) with regard to

2855:   the inner-product

2856:   $\la f,g\ra=\int_E f(z)g(z)e^{-z^2/2}dz$, with $p=1$,

2857:   $q=2$, $n_1=k_1+k_2$, $m_1=k_1$, $m_2=k_2$, and

2858:    $\varphi_1(x)=1,~~\psi_1(y)=e^{ay+\beta y^2},

2859:    ~~\psi_2(y)=1 $

2860:    is the same as the determinant

2861:    (\ref{1.3}), with the expressions $V(z)$ and

2862:    $\varphi^{\pm}(z)$ as in (\ref{extra-var}), and setting

2863:     $s_{1i}:=s_i$, $s_{2i}=u_i$,

2864:    $t_{1i}=t_i$. Therefore by virtue of Theorem

2865:    \ref{Theo:p+q-KP}, the expression below satisfies

2866:    the 3-KP hierarchy, since $p+q=3$, namely,

2867: %

2868: %

2869: %

2870: %

2871: %   In both cases, we have $q=1$ and $p=2$, and $m:=n_1+n_2$, leading to the introduction of three sets of times

2872: %$s_i:=-s_{1i}$, $u_i:=-t_{1i}$ and $v_i:=-t_{2i}$.Thus,

2873: %from the general theory, the numerator of both

2874: %probabilities,

2875: %

2876:    \bea

2877:     %\lefteqn

2878:     {\tau_{k_1k_2}(t,s,u;\al,\beta;E)}%\no\\

2879:     &:=&

2880:    \det

2881:   %

2882:     \left( \!\!\begin{array}{c}

2883: \!

2884: \left(\displaystyle{\int_{E}}z^{i+j}e^{-\frac{z^2}{2}+\al

2885: z+\beta

2886:  z^2}e^{\sum_1^{\iy}(t_k-s_k)z^k}dz\!\!

2887:   \right)_{\!\!\scriptsize{\begin{array}{l}

2888:       0\leq i\leq k_1-1\\

2889:       0\leq j\leq k_1\!+\!k_2\!-\!1

2890:       \end{array}}}

2891:   \\

2892:  \!\!\!\left(\displaystyle{\int_{E}}z^{i+j}e^{-\frac{z^2}{2}

2893:  %-az-\beta  z^2

2894:  }

2895:  e^{\sum_1^{\iy}(t_k-u_k)z^k}dz

2896:  \right)_{\scriptsize{\begin{array}{l}

2897:        0\leq i\leq k_2-1\\

2898:       0\leq j\leq k_1 + k_2 - 1

2899:       \end{array}}}

2900:  \end{array}\!\!\!\!\!\!\right)

2901:  %\no\\

2902: %

2903:  \no\\

2904:  \label{1.5}  &=&\frac{1}{k_1!k_2!}

2905:   \int_{E^{n }}\Dt_{n }(x,y)

2906:   \prod_{j=1}^{k_1}e^{\sum_1^{\iy}t_i x_j^i  }

2907:    \prod_{j=1}^{k_2}e^{\sum_1^{\iy}t_i y^i_j  }

2908:   \\%\eean\bea% \no\\

2909: & &~~~  \left(\Dt_{k_1}(x)\prod^{k_1}_{j=1}

2910:   e^{-\frac{x_j^2}{2}+\al x_j+\beta

2911: x_j^2 }

2912: % \rho_1(x_j)

2913: e^{-\sum_1^{\iy}s_ix^i_j}dx_j\right)

2914:  \no\\ &&~~~

2915: \left(\Dt_{k_2}(y)    \prod^{k_2}_{j=1}

2916: %\rho_2(y_j)

2917: e^{-\frac{y_j^2}{2}%-ay_j-\beta y_j^2

2918: } e^{-\sum_1^{\iy}u_iy^i_j}dy_j\right)

2919:   \no.~~~~~\eea

2920: %

2921: %where

2922: 

2923: 

2924: 

2925: \begin{corollary}\label{bilinear-identity}

2926: %

2927: % for all integer $k_1,k_2,\ell_1,\ell_2>0$ and for all $t,t',s,s',u,u'\in \BR^{\iy}$.

2928: 

2929: 

2930: %  Upon specializing, these identities

2931: The functions $\tau_{k_1k_2}(t,s,u)$ satisfy the

2932: identities

2933: %

2934: %

2935: %

2936: %

2937: % \footnotesize

2938:  {\begin{equation}

2939:   \frac{\partial}{\partial

2940: t_{1}} \log \frac{\tau_{k_{1}+1, k_{2}}}{\tau_{k_{1}-1,

2941: k_{2}}} \ = \frac{  \frac{\partial^{ 2}}{\partial t_{2}

2942: \partial s_{1}}  \log  \tau_{k_{1},  k_{2}}

2943:  }{  \frac{\partial^{ 2}}{\partial t_{1}

2944: \partial s_{1}}   \log  \tau_{k_{1}, \,k_{2}}  },~~~

2945: %\label{C007}

2946: %\end{equation}

2947: %

2948: %\begin{equation}

2949: -  \frac{\partial}{\partial s_{1}} \log

2950: \frac{\tau_{k_{1}+1, \,k_{2}}}{\tau_{k_{1}-1, \,k_{2}}}

2951:    =  \frac{  \frac{\partial^{ 2}}{\partial t_{1}

2952: \partial s_{2}}  \log  \tau_{k_{1},  k_{2}}  }{

2953: \frac{\partial^{ 2}}{\partial t_{1}

2954: \partial s_{1}}   \log  \tau_{k_{1}, k_{2}} }

2955:  \label{C008}

2956:  \end{equation}

2957: %

2958:  \begin{equation}

2959: \frac{\partial}{\partial t_{1}} \log \frac{\tau_{k_{1},

2960: k_{2}+1}}{\tau_{k_{1},  k_{2}-1}}

2961:   = \frac{  \frac{\partial^{ 2}}{\partial t_{2}

2962: \partial u_{1}}  \log  \tau_{k_{1},  k_{2}}

2963:  }{  \frac{\partial^{ 2}}{\partial t_{1}

2964: \partial u_{1}}  \log  \tau_{k_{1},  k_{2}}  }

2965:  %\label{C009}

2966: %\end{equation}

2967: %

2968: %\begin{equation}

2969: ,~~~~- \frac{\partial}{\partial u_{1}} \log

2970: \frac{\tau_{k_{1}, k_{2}+1}}{\tau_{k_{1},  k_{2}-1}}

2971:   = \frac{  \frac{\partial^{2}}{\partial t_{1}

2972: \partial u_{2}}  \log  \tau_{k_{1},  k_{2}}  }{

2973: \frac{\partial^{ 2}}{\partial t_{1}

2974: \partial u_{1}}   \log  \tau_{k_{1},  k_{2}}  } .\label{C010}

2975: \end{equation}

2976: }

2977: 

2978: \end{corollary}

2979: 

2980: 

2981: \begin{proof}

2982:  The

2983: bilinear identities (\ref{tau-bilinear identity}) imply

2984: the PDE's (\ref{introa}) for $$

2985: \tau_{k_1k_2}(t,s,u):=\tau_{k_1,k_2,k_1+k_2}(t,s,u),$$

2986: (in the notation of (\ref{Tmn_def}), setting

2987: $m=(k_1,k_2)$ and $n=k_1+k_2$) expressed in terms of

2988: Hirota's symbol, for $j=0,1,2,\ldots$,

2989: %

2990: {%\footnotesize 

2991: \bea

2992: \gs_j(\tilde\pl_t)\tau_{k_1+1,k_2}\circ\tau_{k_1-1,k_2}&=&

2993:  -\tau^2_{k_1k_2}

2994: \frac{\pl^2}{\pl  s_1\pl t_{j+1}}\log\tau_{k_1k_2}

2995:   \label{bilinear identity1}

2996: \\

2997: \gs_j(\tilde\pl_s)\tau_{k_1-1,k_2}\circ\tau_{k_1+1,k_2}&=&

2998:  -\tau^2_{k_1k_2}

2999: \frac{\pl^2}{\pl t_1\pl s_{j+1}}\log\tau_{k_1k_2}

3000:   \label{bilinear identity2}

3001:  \no\\

3002:  \gs_j(\tilde\pl_t)\tau_{k_1,k_2+1}\circ\tau_{k_1,k_2-1}&=&

3003:  -\tau^2_{k_1k_2}

3004: \frac{\pl^2}{\pl  u_1\pl t_{j+1}}\log\tau_{k_1k_2}

3005:  \no\\

3006: \gs_j(\tilde\pl_u)\tau_{k_1,k_2-1}\circ\tau_{k_1,k_2+1}&=&

3007:  -\tau^2_{k_1k_2}

3008: \frac{\pl^2}{\pl t_1\pl u_{j+1}}\log\tau_{k_1k_2}. \no\eea}

3009: %

3010: In particular for $j=0$, one finds the following

3011: expressions

3012: %

3013:  \bea \frac{\pl^2\log \tau_{k_1 ,k_2 }}{\pl

3014: t_1\pl s_1} &=&

3015:    -\frac{\tau_{k_1 +1,k_2 }\tau_{k_1 -1,k_2 }}

3016:          {\tau_{k_1 ,k_2 }^2}

3017:        \no \\

3018:          \frac{\pl^2\log \tau_{k_1 ,k_2 }}{\pl

3019: t_1\pl u_1} &=&

3020:    -\frac{\tau_{k_1 ,k_2+1 }\tau_{k_1 ,k_2-1 }}

3021:          {\tau_{k_1 ,k_2 }^2}

3022: \label{14}

3023:  \eea

3024:  and another set of expressions for $j=1$. Then taking

3025:  appropriate ratios of these expressions yields the

3026:  formulae of Corollary \ref{bilinear-identity}.\end{proof}

3027: 

3028: 

3029: %\newpage

3030: 

3031: 

3032: 

3033:  %\newpage

3034: 

3035: \section{Virasoro constraints}\label{sect:Vir}

3036: 

3037:  Define the differential operators $\mathcal{B}_m$ involving the boundary points of the set $E$,

3038: $$

3039: \mathcal{B}_m := \sum_{i=1}^{2r} b_i^{m+1} \frac{\pl}{\pl

3040: b_i} , \quad\mbox{for} \quad

3041: E=\bigcup_{i=1}^{r}~[b_{2i-1},b_{2i}] \subset {\mathbb R}  ,$$

3042: %

3043:  and the differential operators $\BV_{-1}$ and $\BV_{0}$ involving differentiation with respect to the auxiliary variables $t_i, ~s_i, ~u_i$ and $\beta$:

3044: %

3045:  %

3046:  \bean

3047: \BV_{-1}&:=&  - \frac{\partial}{\partial

3048: t_{1}} - 2 \beta \frac{\partial}{\partial s_{1}} + \sum_{i\geq2} \left( it_{i} \frac{\partial}{\partial

3049: t_{i-1}} + is_{i} \frac{\partial}{\partial s_{i-1}} +

3050: iu_{i} \frac{\partial}{\partial u_{i-1}} \right)

3051: \\

3052:  &&+k_1(t_1-s_1)+k_2(t_1-u_1)%- \bigg(k_{1}  s_{1} + k_{2}  u_{1} - \big(k_{1} +

3053: %k_{2}\big) t_{1} \bigg)  +

3054: +\alpha  k_{1}

3055: \\

3056:  %

3057: \BV_{0}&:=&   -\frac{\partial  }{\partial

3058: t_{2}}    - 2\beta\frac{\partial }{\partial

3059: s_{2}} - \alpha\frac{\partial  }{\partial s_{1}}

3060:   + \sum_{i\geq1} \left( it_{i} \frac{\partial}{\partial

3061: t_{i}} + is_{i} \frac{\partial}{\partial s_{i}} + iu_{i}

3062: \frac{\partial}{\partial u_{i}} \right)

3063: \\

3064: &&+ k_{1}^{2} + k_{2}^{2} + k_{1} k_{2}

3065:  \label{C005}.

3066: \eean

3067: 

3068: 

3069:  \begin{theorem}\label{Theo:Virasoro}

3070: The integral \,$\tau_{k_1k_2}(t,s,u ; \alpha, \beta ;

3071: E)$\, defined in (\ref{1.5}) satisfies :

3072:  \be

3073:   \mathcal{B}_m \,\tau_{k_1k_2} \,=\, \BV_m%^{k_1k_2}_m \,

3074:   \tau_{k_1k_2} ~~~~~\mbox{for}~~ m=-1 \mbox{ and } 0,

3075:    \label{Virasoro constraints}\ee

3076:   with $\BV_m$ as in (\ref{C005}).  \end{theorem}

3077: 

3078: 

3079: \begin{proof}  Let  $I_{k_1k_2} (x,y) $ be the integrand of the integral $\tau_{k_1k_2}(t,s,u ; \alpha, \beta ; E)$\, defined in (\ref{1.5}):

3080: %

3081: $$

3082: \tau:=\tau_{k_1k_2}(t,s,u ; \alpha, \beta ; E)=\int_{E^n} I_{k_1k_2} (x,y) \prod_1^{k_1}dx_i

3083:  \prod_1^{k_2}dy_j.

3084:  $$

3085: To obtain (\ref{Virasoro constraints}), one uses the fundamental Theorem of calculus and the commutation relation $$\sum_1^n (\frac{\pl}{\pl x_i} x_i- x_i \frac{\pl}{\pl x_i} )=n;$$ one uses the fact that for any Vandermonde, $$\sum_1^k \frac{\pl}{\pl x_i}\Dt(x_1,\ldots,x_k)=0$$  and $$\sum_1^k  x_i\frac{\pl}{\pl x_i}\Dt(x_1,\ldots,x_k)=\frac{n(n-1)}{2}\Dt(x_1,\ldots,x_k).$$  Also one uses the fact that the sum of the $x_i$ and $y_i$ derivatives of the integrand $I_{k_1,k_2}$ translates into the $t_i,~s_i$ and $u_i$-derivatives of $I_{k_1,k_2}$, i.e., the auxiliary parameters $t,s,u ; \alpha, \beta$ were precisely added for this very purpose! So we find:

3086: \bean

3087:  \mathcal{B}_{-1}\tau%\pl_{_{\!E}} \tau

3088:  &=&  \int_{E^n}  \left(\sum_1^{k_1}\frac{\pl}{\pl x_j}+

3089:      \sum_1^{k_2}\frac{\pl}{\pl y_j}\right) I_{k_1,k_2}    \prod_1^{k_1}dx_i     \prod_1^{k_2}dy_j

3090:       \\

3091:     &=&   \int_{E^n}  \BV_{-1} (I_{k_1,k_2})    \prod_1^{k_1}dx_i     \prod_1^{k_2}dy_j

3092:     \\

3093:     &=&   \BV_{-1} \left( \int_{E^n}  I_{k_1,k_2} \prod_1^{k_1}dx_i     \prod_1^{k_2}dy_j\right)

3094: %

3095: =   \BV_{-1}   \tau

3096:  \eean

3097:    and

3098:  \bean

3099:  \mathcal{B}_{0}\tau%\vr_{_{\!E}} \tau

3100:  &=&  \int_{E^n}  \left(\sum_1^{k_1}\frac{\pl}{\pl x_j}x_j+

3101:      \sum_1^{k_1}\frac{\pl}{\pl y_j}y_j\right) I_{k_1,k_2}    \prod_1^{k_1}dx_i     \prod_1^{k_2}dy_j

3102:       \\

3103:     &=&   \int_{E^n} \BV_{0} (I_{k_1,k_2})    \prod_1^{k_1}dx_i     \prod_1^{k_2}dy_j

3104:     \\

3105:     &=&    \BV_{0} \left( \int_{E^n}  I_{k_1,k_2} \prod_1^{k_1}dx_i     \prod_1^{k_2}dy_j\right)

3106: %

3107: =  \BV_{0}  \tau,

3108:  \eean

3109:   establishing Theorem \ref{Theo:Virasoro}. Aother way of computing this has appeared in [2].\end{proof}

3110: 

3111:   \bigbreak

3112: 

3113: \noindent We also have the following identities, valid

3114: when acting on $\tau_{k_{1}k_{2}} (t,s,u;\alpha, \beta;

3115: E)$:

3116: %

3117: %

3118: \begin{equation}

3119:  \frac{\pl}{\pl t_n} = -\frac{\pl}{\pl s_n} - \frac{\pl}{\pl u_n}

3120: , \label{C046} \end{equation}

3121: %

3122: \be

3123: \frac{\partial}{\partial s_{1}} =

3124: -\frac{\partial}{\partial \alpha} ,~~~~~%~~~~~~;~~~~~~~~~~

3125: \frac{\partial}{\partial t_{1}} =

3126: \frac{\partial}{\partial \alpha} -

3127: \frac{\partial}{\partial u_{1}},~~~~~

3128: %\end{displaymath}

3129: %\begin{equation}

3130: \frac{\partial}{\partial s_{2}} =

3131: -\frac{\partial}{\partial \beta}, ~~~~~%~~~~~;~~~~~~~~~~

3132: \frac{\partial}{\partial t_{2}} =

3133: \frac{\partial}{\partial \beta} -

3134: \frac{\partial}{\partial u_{2}}  \label{C004} .\ee

3135: 

3136: \begin{corollary}

3137: On the locus $\mathcal{L} = \{ t = s = u = \beta = 0

3138: \}$, the function \\ $f := \log \tau_{k_{1}k_{2}}

3139: (t,s,u;\alpha, \beta; E)$ satisfies the Virasoro

3140: constraints :

3141: %

3142:  \bea

3143: \frac{\partial f}{\partial s_{1}} &=& -\frac{\partial

3144: f}{\partial \alpha} ,\qquad\qquad \frac{\partial

3145: f}{\partial s_{2}} = -\frac{\partial f}{\partial \beta}

3146: %

3147: \label{C003}

3148: \\

3149:  \frac{\partial f}{\partial t_{1}} &=&  -

3150: \mathcal{B}_{-1} f + \alpha k_{1},\qquad

3151: %\end{displaymath}

3152: %\begin{displaymath}

3153: \frac{\partial f}{\partial u_{1}} = \left(

3154: \mathcal{B}_{-1} + \frac{\partial}{\partial \alpha}

3155: \right) f - \alpha k_{1}

3156:   %

3157:  %

3158: \no\\

3159: \frac{\partial f}{\partial t_{2}} &=& \left(

3160: -\mathcal{B}_{0} + \alpha \frac{\partial}{\partial

3161: \alpha} \right) f + \bigg( k_{1}^{2} + k_{2}^{2} + k_{1}

3162: k_{2} \bigg) ~~~~~~

3163: \no\\

3164: \frac{\partial f}{\partial u_{2}} &=& \left(

3165: \mathcal{B}_{0} - \alpha \frac{\partial}{\partial

3166: \alpha} + \frac{\partial}{\partial \beta} \right) f -

3167: \bigg( k_{1}^{2} + k_{2}^{2} + k_{1} k_{2} \bigg)

3168: \no%

3169: \eea

3170: \vspace*{-.5cm}

3171: \bea

3172:   \label{C002}

3173:   \frac{\partial^{2} f}{\partial t_{1} \partial

3174: u_{1}}& =& -\mathcal{B}_{-1} \left( \mathcal{B}_{-1} +

3175: \frac{\partial}{\partial \alpha} \right) f - k_{2}

3176:  \\

3177: \frac{\partial^{2} f}{\partial t_{1} \partial u_{2}}& =&

3178: -\mathcal{B}_{-1} \left( \mathcal{B}_{0} -

3179: \alpha\frac{\partial}{\partial \alpha} +

3180: \frac{\partial}{\partial \beta} \right) f + 2 \left(

3181: \mathcal{B}_{-1} + \frac{\partial}{\partial \alpha}

3182: \right) f - 2\alpha k_{1}\no\\

3183: %

3184: %\end{displaymath}

3185: %\begin{equation}

3186: \frac{\partial^{2} f}{\partial t_{2} \partial u_{1}}& =&

3187: \left( -\mathcal{B}_{0} + \alpha\frac{\partial}{\partial

3188: \alpha} + 1 \right) \left( \mathcal{B}_{-1} +

3189: \frac{\partial}{\partial \alpha} \right) f  - 2\alpha

3190: k_{1} ~~~~~~~~~~~~~~~~~\no \eea

3191: %

3192: \vspace*{-.5cm}

3193: %

3194: \bea \label{C001}

3195: \frac{\partial^{2} f}{\partial t_{1} \partial s_{1}}

3196: &=& \mathcal{B}_{-1} \left( \frac{\pl}{\pl \alpha} \right)

3197: f - k_{1}

3198: \\%

3199:  \frac{\partial^{2} f}{\partial t_{1} \partial

3200: s_{2}}& =&\left( \mathcal{B}_{-1}  

3201: \frac{\partial}{\partial \beta}   - 2  

3202: \frac{\partial }{\partial \alpha}

3203:  \right)f  \no\\

3204:   \frac{\partial^{2} f}{\partial t_{2} \partial

3205: s_{1}} &=& \left( \mathcal{B}_{0} - \alpha

3206: \frac{\partial}{\partial \alpha} -1 \right) \left(

3207: \frac{\partial}{\partial \alpha} \right) f ~.~~~

3208:  \no\eea

3209: \end{corollary}

3210: 

3211: \begin{proof} Upon dividing by $\tau$, equations (\ref{C003})

3212: are a direct consequence of (\ref{C005}) and

3213: (\ref{C004}), when evaluated on the locus $\mathcal{L}$. To derive

3214: equations (\ref{C002}) and (\ref{C001}), we use the fact

3215: that the boundary operators

3216: $$\mathcal{B}_{m}=\sum_{j=1}^{2r} b_{j}^{m+1}

3217: \frac{\partial}{\partial b_{j}}$$ commute with pure

3218: \mbox{time-differential} operators. For example, the

3219: calculation of the first equation in (\ref{C002}) goes

3220: as follows. We know from (\ref{C005}) and (\ref{C004})

3221: that :

3222: %

3223: $$

3224: -\mathcal{B}_{-1} f = \left( \frac{\partial f}{\partial

3225: t_{1}} - \alpha k_{1} + L_{1}f + \ell_{1} \right) ,~~~

3226: \left( \mathcal{B}_{-1} + \frac{\partial}{\partial

3227: \alpha} \right) f = \left( \frac{\partial f}{\partial

3228: u_{1}} + \alpha k_{1} + L_{2}f + \ell_{2} \right) ,

3229: $$

3230: %

3231: where the $L_{i}$ are linear differential operators

3232: vanishing on $\mathcal{L}$ (commuting with

3233: $\mathcal{B}_{m}$) and the $\ell_{i}$ are functions

3234: vanishing on $\mathcal{L} ~~ (i=1,2)$. Therefore:

3235: %

3236: %

3237: \bean

3238:  \lefteqn{-\mathcal{B}_{-1} \left( \mathcal{B}_{-1} +

3239: \frac{\partial}{\partial \alpha} \right) f

3240: \Bigr|_{\mathcal{L}}}\\&&= -\mathcal{B}_{-1} \left(

3241: \frac{\partial f}{\partial u_{1}} + \alpha k_{1} + L_{2}

3242: f + \ell_{2} \right) \Bigr|_{\mathcal{L}}\\

3243:  &&= -\mathcal{B}_{-1}

3244: \left( \frac{\partial f}{\partial u_{1}} + L_{2} f

3245: \right) \Bigr|_{\mathcal{L}} = - \left(

3246: \frac{\partial}{\partial u_{1}} + L_{2} \right)

3247: \mathcal{B}_{-1} f \Bigr|_{\mathcal{L}} = - \left(

3248: \frac{\partial}{\partial u_{1}} \right) \mathcal{B}_{-1}

3249: f \Bigr|_{\mathcal{L}}

3250: \\

3251: &&= \frac{\partial}{\partial u_{1}} \left(

3252: \frac{\partial f}{\partial t_{1}} - \alpha k_{1} +

3253: L_{1}f + \ell_{1} \right) \Bigr|_{\mathcal{L}} = \left(

3254: \frac{\partial^{2} f}{\partial t_{1} \partial u_{1}} +

3255: \frac{\partial}{\partial u_{1}} L_{1}f + \frac{\partial

3256: \ell_{1}}{\partial u_{1}} \right) \Bigr|_{\mathcal{L}}.

3257: \eean

3258: 

3259: %\begin{verbatim}

3260: %\end{verbatim}

3261: 

3262: Since  $$L_{1} = -2 \beta

3263: \frac{\partial}{\partial \alpha} - \sum_{i\geq2} \left(

3264: i t_{i} \frac{\partial}{\partial t_{i-1}} + i s_{i}

3265: \frac{\partial}{\partial s_{i-1}} + i u_{i}

3266: \frac{\partial}{\partial u_{i-1}} \right)$$ and

3267: $\ell_{1} = k_{1} s_{1} + k_{2} u_{1} -

3268: \left( k_{1} + k_{2} \right) t_{1} $, one checks

3269: that, along the locus $\LR$, one has $$\frac{\partial}{\partial u_{1}}L_1f=0\qquad \mbox{    ~~~and~~~   }\qquad\frac{\partial

3270: \ell_{1}}{\partial u_{1}} =k_2,$$ yielding the expression for $\frac{\partial^{2} f}{\partial t_{1} \partial

3271: u_{1}}$ in (\ref{C002}).

3272: %

3273: %the second term \mbox{vanishes} on $\mathcal{L}$,

3274: %while the third one yields $k_{2}$ as expected. A

3275: A similar procedure \mbox{applies} to establish all the

3276: identities above. \end{proof}

3277: 

3278: 

3279: 

3280: 

3281: %\newpage

3282: 

3283: 

3284: 

3285: \section{A PDE for the Gaussian ensemble with \mbox{external source}}

3286: \label{sect:PDE}

3287: %\begin{verbatim}

3288: %\end{verbatim}

3289: 

3290: 

3291: 

3292: 

3293: Consider the Gaussian Hermitian random matrix ensemble

3294: $\mathcal {H}_n$ with external source $A$, given by the

3295: diagonal matrix (\ref{diagmatrix}) (set $n=k_1+k_2$)

3296:  %

3297: % \be

3298: % A:= \left(\begin{array}{cccccc}

3299: %   \al\\

3300: %   &\ddots& & & &{\bf O}\\

3301: %   & &\al\\

3302: %   & & &0\\

3303: %   &{\bf O}& & &\ddots\\

3304: %   & & & & &0

3305: %\end{array}\right)\begin{array}{l}

3306: %\updownarrow k_1\\

3307: %\\

3308: %\\

3309: %\\

3310: %\updownarrow k_2

3311: %\end{array},

3312: %  \label{diagmatrix5}\ee

3313:   and density

3314:   \be

3315:   \frac{1}{Z_n} %\int_{\mathcal {H}_{k}(E)}

3316:   e^{-\Tr(\frac{1}{2} M^2-AM) }dM.

3317:   \ee

3318: %

3319:     Given a disjoint union of intervals

3320:  $

3321:  E:= \bigcup^r_{i=1}[b_{2i-1},b_{2i}]\subset {\mathbb R}

3322:    ,$

3323: %  and a

3324: %diagonal matrix  (set $k=k_1+k_2$)

3325: %

3326: %

3327: %

3328: define the algebra of differential operators, generated

3329: by

3330:  \be\mathcal{B}_k=\sum_{i=1}^{2r}b_i^{k+1}\frac{\pl}{\pl b_i}.\ee

3331: %

3332: %

3333: Consider the following probability:

3334:   \be

3335: \BP_n(\al;E):=\BP(~\mbox{all eigenvalues}~\in E) =

3336:  \frac{1}{Z_n} \int_{\mathcal {H}_{n}(E)}

3337:    e^{-\Tr(\frac{1}{2} M^2-AM) }dM,

3338:   \label{0.4}

3339:    \ee

3340: where $\mathcal {H}_{n}(E)$ is the set of all Hermitian

3341: matrices with all eigenvalues in $E$. The purpose of

3342: this section is to prove the following theorem:

3343: %

3344: 

3345: 

3346: \begin{theorem} \label{Theo:PDE}

3347: The log of the probability $\BP_n(a;E)$

3348: %   \be

3349: %\BP_n(a;E):=\BP(~\mbox{all eigenvalues}~\in E) =

3350: % \frac{1}{Z_n} \int_{\mathcal {H}_{k}(E)}

3351: %   e^{-\Tr(\frac{1}{2} M^2-AM) }dM

3352: %  \label{0.4}  \ee

3353:    satisfies

3354:   a  \underline {fourth-order} PDE in $a$ and in the endpoints $b_1,..., b_{2r}$

3355:   of the set $E$, with \underline {quartic

3356:   non-linearity}:

3357: %

3358: %

3359: \bea && \bigg( F^{+}    \mathcal{B}_{-1} G^{-}   + F^{-}

3360:   \mathcal{B}_{-1} G^{+}   \bigg) \bigg( F^{+}

3361: \mathcal{B}_{-1} F^{-}   - F^{-}    \mathcal{B}_{-1}

3362: F^{+}   \bigg)

3363: \no\\

3364: &&- \bigg( F^{+} G^{-} + F^{-} G^{+} \bigg) \bigg( F^{+}

3365:    \mathcal{B}_{-1}^{ 2} F^{-}   - F^{-}

3366: \mathcal{B}_{-1}^{ 2} F^{+}   \bigg)  =  0 \label{C017a}

3367: , \eea

3368:  or what is the same

3369:  \begin{equation}

3370: \det \left( \begin{array}{cccc}

3371:  G^+& \mathcal{B}_{-1}F^+  & - F^+ &0\\

3372:  -G^-& \mathcal{B}_{-1}F^-  & - F^- &0\\

3373:  \mathcal{B}_{-1} G^+ &\mathcal{B}_{-1}^2F^+&0&-F^+\\

3374:  -\mathcal{B}_{-1} G^- &\mathcal{B}_{-1}^2F^-&0&-F^-\\

3375:  \end{array}\right) =0 , \label{C021}

3376: \end{equation}

3377:   where

3378: $$

3379:   F^{+}  = \mathcal{B}_{-1}

3380:  \frac{\partial}{\partial \alpha}   \log

3381: \BP_{n} - k_{1} ,\qquad%\\

3382:  F^{-} =  -\mathcal{B}_{-1} \left( \mathcal{B}_{-1}  +

3383: \frac{\partial}{\partial \alpha} \right) \log  \BP_{n} -

3384: k_{2}

3385: $$

3386: 

3387: \vspace*{-.6cm}

3388: 

3389: \bean 

3390:   H_{1}^{+} & =& 4  \frac{\partial}{\partial

3391: \alpha}   \log \BP_{n} + 4 \alpha k_{1} +

3392: \frac{4k_{1}k_{2}}{\alpha}

3393:  ,\\

3394:  H_{1}^{-}

3395:  &=&  -  2 \left( \mathcal{B}_{0} - \alpha

3396: \frac{\partial}{\partial \alpha} + 1 \right)

3397: \frac{\partial}{\partial \alpha}   \log \BP_{n}

3398:  -   \frac{4k_{1}k_{2}}{\alpha}

3399: \\

3400: H_{2}^{+} &=& 2 \left( \!\mathcal{B}_{0}\! -\! \alpha

3401: \frac{\partial}{\partial \alpha}\! -\! 1\! - 2 \alpha

3402: \mathcal{B}_{-1} \!\right) \frac{\partial}{\partial

3403: \alpha}   \log \BP_{n}

3404:  \\

3405:  H_{2}^{-} &=&  -  2 \left( \mathcal{B}_{0}\! -\! \alpha

3406: \frac{\partial}{\partial \alpha} \!-\! 1 \right)   \bigg(

3407: \mathcal{B}_{-1} \!+\! \frac{\partial}{\partial \alpha}

3408: \bigg) \log

3409:  \BP_{n}

3410:  \\ \

3411:  2G^{\pm}&=&\left\{H_1^{\pm},F^{\pm}\right\}_{\mathcal{B}_{-1}}\mp

3412:  \left\{H_2^{\pm},F^{\pm}\right\}_{\pl/ \pl \al} .

3413:  \eean 

3414: %

3415: \end{theorem}

3416: 

3417: 

3418: \begin{remark}   It is not surprising that the PDE

3419: (\ref{C017a}) has exactly \emph{the same form} as the

3420: PDE \mbox{derived} in \cite{AvM-Pearcey} and

3421: \cite{AvM-Pearcey-joint}, associated to the Gaussian

3422: Unitary Ensemble with an \mbox{external} source, in the

3423: case where the source matrix admits \emph{two}

3424: eigenvalues of \mbox{\emph{opposite}} signs. The only

3425: difference is that the expressions for the functions

3426: $F^{\pm}, H_{1}^{\pm}, H_{2}^{\pm}$ and $G^{\pm}$

3427: \mbox{obtained} here, differ from those in

3428: \cite{AvM-Pearcey}. The reason is that corresponding

3429: $\tau$-functions satisfy the same integrable equation

3430: (the 3-KP hierarchy, as in Section 3.3), whereas the Virasoro constraints \mbox{leading}

3431: to (\ref{C017a}) are different. In particular, there is

3432: no more involution relating the variables $s_{k}$ and

3433: $u_{k}$.

3434: \end{remark}

3435: 

3436: 

3437: 

3438: \begin{proof}[Proof of Theorem~\ref{Theo:PDE}]

3439: Remember:

3440: \begin{displaymath}  \BP_n (\alpha , E) := \BP_n (\textrm{spec}  M \subseteq E) =

3441: \frac{ \tau_{k_1k_2} (t, s, u ; \alpha, \beta ; E) } {

3442: \tau_{k_1k_2} (t, s, u ; \alpha, \beta ; {\mathbb R}) } \Bigr|_{

3443: \mathcal{L} \equiv \{ t = s = u = \beta = 0 \} }.

3444: \end{displaymath}

3445: %

3446: The denominator, that is the integral (\ref{1.5}) over

3447: the whole range, can be expressed in terms of moments,

3448: which contain standard {Gaussian} {integrals} (the

3449: reader is referred to the \mbox{Appendix $1$} in

3450: \cite{AvM-Pearcey}), leading to an exact evaluation,

3451: with $c_{k_{1}k_{2}}$ a constant, depending on $k_{1} ,

3452: k_{2}$ only:

3453: %

3454: \begin{equation}

3455: \tau_{k_{1},k_{2}} (t, s, u ; \alpha, \beta ;{\mathbb R})

3456: \Bigr|_{\mathcal{L}} = c_{k_{1}k_{2}} \alpha^{k_{1}

3457: k_{2}}e^{ k_{1} \left( \alpha^{2} / 2 \right)}.

3458: ~~~~~~~~~~~ \label{C018}

3459: \end{equation}

3460: %\begin{equation}

3461: %\log \tau_{k_{1},k_{2}} (t, s, u ; \alpha, \beta ; \BR)

3462: %\Bigr|_{\mathcal{L}} = \frac{k_{1}}{2} \alpha^{2} +

3463: %k_{1} k_{2} \log \left(\alpha\right) + C_{k_{1}k_{2}} ,

3464: % \label{C019}

3465: %\end{equation}

3466: %

3467: %with $c_{k_{1}k_{2}}$ and $C_{k_{1}k_{2}}$ are

3468: %coefficients depending on $k_{1} , k_{2}$ only. \\

3469: Consequently (with $C_{k_{1}k_{2}}$ a constant,

3470: depending on $k_{1} , k_{2}$ only):

3471: \begin{equation}

3472: \log \BP_{n} (\alpha, E) = \log \tau_{k_{1},k_{2}} (t,

3473: s, u ; \alpha, \beta ; E) \Bigr|_{\mathcal{L}} -

3474: \frac{k_{1}}{2} \alpha^{2} - k_{1} k_{2} \log

3475: \left(\alpha\right) - C_{k_{1}k_{2}} \label{C020}.

3476: \end{equation}

3477: 

3478: Then, we turn our attention to the numerator

3479: $\tau_{k_1k_2} (t, s, u ; \alpha, \beta ; E)$, more

3480: briefly noted as $\tau_{k_{1}k_{2}}$. On the locus

3481: $\mathcal{L}$, by (\ref{C008}), and (\ref{C003}), one

3482: finds:

3483: %

3484: \bea \label{C011} \frac{ \frac{\partial^{2}}{\partial t_{2}

3485: \partial s_{1}} \log \tau_{k_{1}, k_{2}}

3486: }{ \frac{\partial^{2}}{\partial t_{1}

3487: \partial s_{1}}  \log \tau_{k_{1}, k_{2}} } &=& \frac{\partial}{\partial t_{1}}

3488: \log \left( \frac{\tau_{k_{1}+1, k_{2}}}{\tau_{k_{1}-1,

3489: k_{2}}} \right) = -\mathcal{B}_{-1} \log \left(

3490: \frac{\tau_{k_{1}+1, k_{2}}}{\tau_{k_{1}-1, k_{2}}}

3491: \right) + 2\alpha ~

3492:  \\

3493: \frac{ \frac{\partial^{2}}{\partial t_{1}

3494: \partial s_{2}} \log \tau_{k_{1}, k_{2}}

3495: }{ \frac{\partial^{2}}{\partial t_{1}

3496: \partial s_{1}}  \log \tau_{k_{1}, k_{2}} } &=& -\frac{\partial}{\partial s_{1}} \log \left(

3497: \frac{\tau_{k_{1}+1, k_{2}}}{\tau_{k_{1}-1, k_{2}}}

3498: \right) = \frac{\partial}{\partial \alpha} \log \left(

3499: \frac{\tau_{k_{1}+1, k_{2}}}{\tau_{k_{1}-1, k_{2}}}

3500: \right) . \no \eea

3501: %

3502: %Left equations are a rewriting of the KP-identities

3503: %(\ref{C007}) and (\ref{C008}), whereas right

3504: %\mbox{equations} reflect the first Virasoro constraints

3505: %of (\ref{C003}), valid on

3506: %$\mathcal{L}$, and adapted to the indices $k_{1} \pm 1 , k_{2}$ . \\

3507: %

3508: In (\ref{C011}), acting with $  {\partial}/{\partial

3509: \alpha} $ on the first expression, then acting with

3510: $\mathcal{B}_{-1}$ on the second expression, and adding

3511: the two, yields:

3512: 

3513: \begin{displaymath}

3514: \frac{\partial}{\partial \alpha} \left( \frac{

3515: \frac{\partial^{2}}{\partial t_{2}

3516: \partial s_{1}} \log \tau_{k_{1}, k_{2}} }{

3517: \frac{\partial^{2}}{\partial t_{1}

3518: \partial s_{1}}  \log \tau_{k_{1}, k_{2}} } - 2\alpha \right)

3519: + \mathcal{B}_{-1} \left( \frac{

3520: \frac{\partial^{2}}{\partial t_{1}

3521: \partial s_{2}} \log \tau_{k_{1}, k_{2}} }{

3522: \frac{\partial^{2}}{\partial t_{1}

3523: \partial s_{1}}  \log \tau_{k_{1}, k_{2}} } \right)

3524: = 0 . \end{displaymath}

3525: %

3526: This identity can conveniently be expressed as

3527: Wronskians:

3528: %

3529: \begin{equation}

3530: \left\{ \mathcal{B}_{-1} \frac{\partial}{\partial \beta}

3531: \log \tau_{k_{1} k_{2}} \Bigr|_{\mathcal{L}} , F^{+}

3532: \right\} _{\mathcal{B}_{-1}} = \bigg\{ H_{1}^{+} ,

3533: \frac{F^{+}}{2} \bigg\}_{\mathcal{B}_{-1}} - \bigg\{

3534: H_{2}^{+} , \frac{F^{+}}{2}

3535: \bigg\}_{\frac{\partial}{\partial \alpha}} =: G^{+},

3536: \label{C012}

3537: \end{equation}

3538: %

3539: where

3540: %

3541: \bea \label{C013}

3542:  F^{+}  &:=&  \frac{\partial^{2}}{\partial t_{1}

3543: \partial s_{1}} \log \tau_{k_{1} , k_{2}}

3544:    \\%~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\end{displaymath}

3545: %\begin{displaymath}

3546: H_{1}^{+} &:=& -2 \frac{\partial^{2}}{\partial t_{1}

3547: \partial s_{2}} \log \tau_{k_{1} , k_{2}} + 2\mathcal{B}_{-1}\frac{\partial}{\partial \beta} \log

3548: \tau_{k_{1} k_{2}} \no\\%~~~~~~~~~~  \end{displaymath}

3549: %\begin{equation}

3550: H_{2}^{+} &:=& 2 \left( \frac{\partial^{2}}{\partial

3551: t_{2}

3552: \partial s_{1}} \log \tau_{k_{1} , k_{2}} -2\alpha

3553: \frac{\partial^{2}}{\partial t_{1} \partial s_{1}} \log

3554: \tau_{k_{1} , k_{2}} \right) .~~ \no\eea

3555: %

3556: %\begin{verbatim}

3557: %\end{verbatim}

3558: %

3559: In this way, by virtue of (\ref{C001}) and (\ref{C020})

3560: one obtains, along the locus $\mathcal{L}$, explicit

3561: expressions for $F^{+}$, $H_{1}^{+}$ and $H_{2}^{+}$

3562: that are free of partials in $\beta$; namely:

3563: {  \bea

3564:  \label{Hplus} F^{+}& =& \mathcal{B}_{-1}

3565: \left(\frac{\partial}{\partial \alpha} \right) \log

3566: \tau_{k_{1},k_{2}} - k_{1} = \mathcal{B}_{-1}

3567: \left(\frac{\partial}{\partial \alpha} \right) \log

3568: \BP_{n} - k_{1} ~~

3569:   \\

3570: H_{1}^{+} &=& 4 \left(\frac{\partial}{\partial \alpha}

3571: \right) \log \tau_{k_{1},k_{2}} = 4

3572: \left(\frac{\partial}{\partial \alpha} \right) \log

3573: \BP_{n} + 4 \alpha k_{1} + 4 \frac{k_{1}k_{2}}{\alpha}

3574:    \no\\

3575: H_{2}^{+}& =& 2 \left( \mathcal{B}_{0} - \alpha

3576: \frac{\partial}{\partial \alpha} - 1 - 2 \alpha

3577: \mathcal{B}_{-1} \right)  \bigg(

3578: \frac{\partial}{\partial \alpha} \bigg) \log

3579: \tau_{k_{1},k_{2}} + 4 \alpha k_{1} ~~

3580:    \no\\

3581:  &=&

3582:  2 \left( \mathcal{B}_{0}\!  -  \!\alpha

3583: \frac{\partial}{\partial \alpha}\!  -  \!1 \! -  \!2

3584: \alpha \mathcal{B}_{-1} \right)  \bigg(

3585: \frac{\partial}{\partial \alpha} \bigg) \log \BP_{n} .

3586:  \no \eea}%\\

3587: %

3588: Subsequently, one repeats exactly the same operations

3589: for the Virasoro and KP-identities

3590: involving the $t_{k}$ and $u_{k}$ variables. From

3591: (\ref{C010}) and (\ref{C003}), one finds on

3592: $\mathcal{L}$:

3593: %

3594: \bea \label{C014}

3595:   %

3596: &&\hspace*{.7cm}\frac{ \frac{\partial^{2}}{\partial t_{2}

3597: \partial u_{1}} \log \tau_{k_{1}, k_{2}}

3598: }{ \frac{\partial^{2}}{\partial t_{1}

3599: \partial u_{1}}  \log \tau_{k_{1}, k_{2}} } = \frac{\partial}{\partial t_{1}}

3600: \log \left( \frac{\tau_{k_{1}, k_{2}+1}}{\tau_{k_{1},

3601: k_{2}-1}} \right) = -\mathcal{B}_{-1} \log \left(

3602: \frac{\tau_{k_{1}, k_{2}+1}}{\tau_{k_{1}, k_{2}-1}}

3603: \right)   \\

3604: &&\hspace*{.7cm}\frac{ \frac{\partial^{2}}{\partial t_{1}

3605: \partial u_{2}} \log \tau_{k_{1}, k_{2}}

3606:  }{  \frac{\partial^{ 2}}{\partial t_{1}

3607: \partial u_{1}}   \log  \tau_{k_{1},  k_{2}} } = - \frac{\partial}{\partial u_{1}} \log \left(

3608: \frac{\tau_{k_{1},  k_{2}+1}}{\tau_{k_{1},  k_{2}-1}}

3609: \right) = - \left( \mathcal{B}_{-1} +

3610: \frac{\partial}{\partial \alpha} \right) \log \left(

3611: \frac{\tau_{k_{1},  k_{2}+1}}{\tau_{k_{1},  k_{2}-1}}

3612: \right) . \no\eea

3613: 

3614: So, in (\ref{C014}), acting on the first equation with

3615: $\left( \mathcal{B}_{-1} + \frac{\partial}{\partial

3616: \alpha} \right)$ and on the second equation with

3617: $\mathcal{B}_{-1}$ , then subtracting the two yields:

3618: %

3619: \begin{displaymath}

3620: \mathcal{B}_{-1} \left( \frac{  \left( \frac{\partial^{

3621: 2}}{\partial t_{2} \partial u_{1}} - \frac{\partial^{

3622: 2}}{\partial t_{1} \partial u_{2}} \right) \log

3623:  \tau_{k_{1},  k_{2}}

3624:   }{  \frac{\partial^{2}}{\partial t_{1}

3625: \partial u_{1}}   \log  \tau_{k_{1},  k_{2}}  } \right)

3626:  +  \frac{\partial}{\partial \alpha} \left( \frac{

3627: \frac{\partial^{ 2}}{\partial t_{2}

3628: \partial u_{1}}  \log  \tau_{k_{1}, k_{2}}  }{

3629: \frac{\partial^{ 2}}{\partial t_{1}

3630: \partial u_{1}}  \log  \tau_{k_{1},  k_{2}}  } \right)

3631: = 0  , \end{displaymath}

3632: %

3633: or equivalently (remember the brackets are Wronskians) :

3634: %

3635: \begin{equation}

3636: -\left\{ \mathcal{B}_{-1} \frac{\partial}{\partial

3637: \beta} \log \tau_{k_{1} k_{2}} \Bigr|_{\mathcal{L}}

3638: ,F^{-} \right\} _{\mathcal{B}_{-1}} =  \bigg\{ H_{1}^{-}

3639: , \frac{F^{-}}{2} \bigg\}_{\mathcal{B}_{-1}} + \bigg\{

3640: H_{2}^{-} , \frac{F^{-}}{2}

3641: \bigg\}_{\frac{\partial}{\partial \alpha}} \,\,=:\,\,

3642: G^{-} \label{C015} \end{equation}

3643: %

3644: in terms of the functions

3645: %

3646: {%\footnotesize

3647: \bea

3648:    \label{C016}  F^{-} &:=&

3649: \frac{\partial^{\,2}}{\partial t_{1}

3650: \partial u_{1}} \,\log \tau_{k_{1} , k_{2}}

3651:  \\

3652: H_{1}^{-} &:=&  2  \left( \,\frac{\partial^{

3653: 2}}{\partial t_{2}

3654: \partial u_{1}} -

3655: \frac{\partial^{ 2}}{\partial t_{1} \partial u_{2}}

3656: \right) \log \tau_{k_{1},k_{2}} - 2 \mathcal{B}_{-1}

3657: \frac{\partial}{\partial \beta} \log

3658: \tau_{k_{1} k_{2}}  \no\\

3659: %

3660:  H_{2}^{-} &:=& 2  \left(

3661: \frac{\partial^{ 2}}{\partial t_{2}

3662: \partial u_{1}} \right)  \log \tau_{k_{1} , k_{2}}  .

3663:  \no\eea}

3664: %

3665: Using the Virasoro constraints \,(\ref{C002}), \,as well

3666: as (\ref{C020}), we obtain \mbox{explicit} formulae for

3667: \,$F^{-}$, \,$H_{1}^{-}$ and $H_{2}^{-}$ (which do not

3668: contain partials in $\beta$):

3669: % \be \label{Hminus}\ee

3670:  % \vspace*{-.6cm}

3671:     %

3672:  \bea

3673:   %

3674:   F^{-} &=&

3675: -\mathcal{B}_{-1} \left( \mathcal{B}_{-1} +

3676: \frac{\partial}{\partial \alpha} \right) \log

3677: \tau_{k_{1},k_{2}} - k_{2} 

3678: \no\\

3679: &=& -\mathcal{B}_{-1} \left(

3680: \mathcal{B}_{-1} + \frac{\partial}{\partial \alpha}

3681: \right) \log  \BP_{n} - k_{2}  \no\\%\eea\bea

3682: %\no\\

3683: H_{1}^{-} &=& -  2 \left( \mathcal{B}_{0} - \alpha

3684: \frac{\partial}{\partial \alpha} + 1 \right) \bigg(

3685: \frac{\partial}{\partial \alpha} \bigg) \log

3686: \tau_{k_{1},k_{2}}  \no\\

3687:   &=&

3688:    -  2 \left(

3689: \mathcal{B}_{0} - \alpha \frac{\partial}{\partial

3690: \alpha} + 1 \right) \bigg( \frac{\partial}{\partial

3691: \alpha} \bigg) \log  \BP_{n} - 4

3692:  \frac{k_{1}k_{2}}{\alpha}\no\\

3693:  %\eean

3694: %\begin{displaymath}

3695: H_{2}^{-}  &=&  -  2 \left( \mathcal{B}_{0} - \alpha

3696: \frac{\partial}{\partial \alpha} - 1 \right)   \bigg(

3697: \mathcal{B}_{-1} + \frac{\partial}{\partial \alpha}

3698: \bigg) \log \tau_{k_{1},k_{2}} - 4  \alpha  k_{1}

3699: %~~~~~~~~~~~~~~~~~~~~~~ \end{displaymath}

3700:  \no\\ %\bea

3701:   &=&

3702:   -  2 \left( \mathcal{B}_{0} - \alpha

3703: \frac{\partial}{\partial \alpha} - 1 \right) \, \bigg(

3704: \mathcal{B}_{-1} + \frac{\partial}{\partial \alpha}

3705: \bigg) \log

3706:  \BP_{n} .\label{Hminus} \eea %\\

3707:  %

3708: Equations (\ref{C012}) and  (\ref{C015})\, form a linear

3709: system in  $$\BB_{-1}  \frac{\partial  \log

3710:  \tau_{k_{1} k_{2}}}{\partial  \beta}

3711: \Bigr|_{\mathcal{L}} \qquad\mbox{and}\qquad\BB_{-1}^{2}

3712:  \frac{\partial  \log  \tau_{k_{1} k_{2}}}{\partial

3713:  \beta} \Bigr|_{\mathcal{L}},$$ which can be solved to

3714: extract the quantities:

3715: %

3716: \bea \mathcal{B}_{-1} \,\frac{\partial \log

3717: \tau_{k_{1},k_{2}}}{\partial \beta} \Bigr|_{\mathcal{L}}

3718: &=& \frac{ G^{-} F^{+} + G^{+} F^{-} }{ -F^{-} \left(

3719: \mathcal{B}_{-1} F^{+} \right)  +\, F^{+} \left(

3720: \mathcal{B}_{-1} F^{-} \right)  }   \label{C034} \\

3721: \mathcal{B}_{-1}^{ 2} \,\frac{\partial \log

3722: \tau_{k_{1},k_{2}}}{\partial \beta} \Bigr|_{\mathcal{L}}

3723:  &=&  \frac{ G^{-} \left( \mathcal{B}_{-1} F^{+}

3724: \right) + G^{+} \left( \mathcal{B}_{-1} F^{-} \right) }{

3725: -F^{-} \left( \mathcal{B}_{-1} F^{+} \right) + F^{+}

3726: \left( \mathcal{B}_{-1} F^{-} \right)  }  . \label{C035}

3727: \eea

3728: %

3729: Finally, subtracting the second relation from

3730:  $\mathcal{B}_{-1}$ of the first equation,

3731:  establishes the expected PDE (\ref{C017a}) and Theorem \ref{Theo:PDE}.

3732: %

3733: %

3734: %\bea  &&\bigg( F^{+}   \mathcal{B}_{-1} G^{-}

3735: % + F^{-}   \mathcal{B}_{-1}

3736: %G^{+}  \bigg) \bigg( F^{+}   \mathcal{B}_{-1} F^{-}   -

3737: %F^{-}   \mathcal{B}_{-1} F^{+}   \bigg)

3738: %\no\\

3739: %&&- \bigg( F^{+} G^{-} + F^{-} G^{+} \bigg) \bigg( F^{+}

3740: %  \mathcal{B}_{-1}^{ 2} F^{-}  - F^{-}

3741: %\mathcal{B}_{-1}^{ 2} F^{+}   \bigg)= 0. \label{C017}

3742: %\eea

3743: %

3744: To prove the second equation (\ref{C021}), set 

3745: $$

3746:  X:=\left. \mathcal{B}_{- 1}\frac{\pl  }{\pl \beta }

3747:     \log \tau_{k_1k_2}\right|_{\LR};

3748:  $$

3749: then the matrix in (\ref{C021}) annihilates the column

3750: $(

3751:       1,

3752:       X,

3753:       \mathcal{B}_{-1} X,

3754:       \mathcal{B}_{-1}^2X

3755:       )^{\top}$, and thus the determinant vanishes, concluding the proof of Theorem

3756: \ref{Theo:PDE}.\end{proof}

3757: 

3758: 

3759: 

3760: \section{A PDE for the transition probability of the $r$-Airy process}

3761: %Brownian motion model

3762: \label{sect6}

3763: 

3764: %\begin{verbatim}

3765: %\end{verbatim}

3766: 

3767: \begin{proof}[Proof of Theorem \ref{Theo2}]

3768: %The purpose of this section is to prove Theorem \ref{Theo2}.

3769:  Remember from section \ref{sect1}

3770:  the identity

3771:  \begin{multline*}

3772: \lim_{n\rightarrow \iy}

3773: \BP^{(0,\rho\sqrt{n/2})}_{Br}\left(\mbox{all}~

3774: x_i\left(\frac{1}{1+e^{-2\tau/n^{1/3}}}\right) \in

3775:   \frac{(-\iy, \sqrt{2n}+\frac{x}{\sqrt{2}

3776:   n^{1/6}})}{2\cosh( \tau/n^{1/3})}\right) = \\*

3777:      \lim_{n\rightarrow \iy}

3778:   \BP_{n}\left({\rho}\sqrt{n}e^{\tau /n^{1/3}};

3779:   2\sqrt{n} +\frac{(-\iy,x)}{n^{1/6}}   \right),%

3780:   \end{multline*}

3781:   which for $0\leq \rho\leq 1$ leads by

3782:  Theorem~\ref{Theo1} to a phase transition at $\rho=1$, for which

3783:   the expression above reduces to

3784:   $\BP(\sup \mathcal {A}^{(r)}(\tau)\leq x)$,

3785:   according to ({\ref{r-AiryP}). The

3786:   above scaling suggests the choice $z=n^{-1/6}$ as small

3787:   parameter and considering the map

3788:   %

3789:   $ (\tau,x) \longmapsto(\al,b) , $

3790: given by \bean

3791:  \al={\rho}\sqrt{n}e^{\tau /n^{1/3}}=%\pi\sqrt{n}

3792:   %{e^{\tau/n^{1/3}}}

3793:     \frac{\rho}{z^3}e^{\tau z^2},\qquad

3794:  b=

3795:    2\sqrt{n} +\frac{x}{n^{1/6}}

3796:    =\frac{2}{z^3}+xz  %{2\sqrt{n}+\frac{ x }{n^{1/6}}}  ,

3797:   %

3798:  \eean

3799:  with inverse map

3800:   $(\al,b)\longmapsto (\tau,x),  $

3801:  given by

3802:  $$

3803:  \tau=\frac{1}{z^2}\log\left(\frac{\al

3804:  z^3}{\rho}\right),~~~

3805:  x =\frac{b}{z}-\frac{2}{z^4}

3806:  .$$

3807: %

3808:  Setting

3809: \bea  \label{C024}

3810:  \tilde Q(\tau,x) &:=&

3811:  %=

3812:     \log\BP_{n}\left({\rho}\sqrt{n}e^{\tau /n^{1/3}};

3813:   2\sqrt{n} +\frac{(-\iy,x)}{n^{1/6}}   \right)

3814:     \\

3815:    &=&

3816:   \log\BP_{n}\left(\frac{\rho}{z^3}e^{\tau z^2};

3817:    \frac{2}{z^3}+(-\iy,x)z  \right)

3818:    %

3819:    ,\no\eea

3820: yields, via the inverse map, \be

3821:  \tilde Q\left(\frac{1}{z^2}\log\bigl(\frac{\al

3822:  z^3}{\rho}\bigr),\frac{b}{z}-\frac{2}{z^4}

3823:  \right)=\log\BP_n(\al,(-\iy,b)).\label{C025}

3824:   \ee

3825: From Corollary \ref{cor:estimate2} (section \ref{sect2}) it follows that for $z\rg 0$,

3826:  \be

3827:  \tilde

3828:  Q(\tau,x)=Q(\tau,x)+O(z^2),

3829:  \label{6.3}\ee

3830: with $Q(\tau,x)$ independent of $z$.

3831: 

3832: 

3833: 

3834: 

3835: 

3836: Since we have shown that $\log \BP_{n} (\alpha,

3837: (-\infty, b))$ satisfies the PDE (\ref{C017a}) of

3838: Theorem~\ref{Theo:PDE}, with $\mathcal{B}_{-1} \equiv

3839: \frac{\partial}{\partial b}$ and $\mathcal{B}_{0} \equiv

3840: b \frac{\partial}{\partial b}$ , it follows that, to get

3841: a PDE for the limiting case, we just need to estimate:

3842: %

3843: { \footnotesize \begin{equation}

3844: %\hspace{-2.5cm}0&=&\!\!

3845: \left.\left\{\begin{array}{l} \Bigl(F^ + \BB_{-1}G^-+F^-

3846: \BB_{-1}G^+

3847: \Bigr)     \Bigl(F^+ \BB_{-1}F^- -F^- \BB_{-1}F^+ \Bigr) \\

3848:  - \Bigl(F^+ G^- +F^- G^+

3849: \Bigr)   \Bigl(F^+ \BB_{-1}^2F^- -F^- \BB_{-1}^2F^+

3850: \Bigr)

3851: \end{array}\right\}

3852:  \right|_{\footnotesize\begin{array}{l}

3853:      \alpha  \mapsto  \left( \rho / z^{ 3} \right)

3854:       e^{ \tau  z^{2} } \\

3855:      b \,\mapsto  x z  +  \frac{2}{z^{ 3}} \\

3856:      n  \mapsto  \frac{1}{z^{ 6}}\end{array} }

3857: \label{C027}.\end{equation} }

3858: %

3859: To do this, the various expressions in the bracket must

3860: be computed in terms of the function $\tilde Q (\tau, x)

3861: %:= \log  F(\tau, x)

3862: $.

3863: %

3864: By (\ref{C024}) and (\ref{C025}), one immediately gets:

3865: %

3866: $$

3867:  \frac{\partial  \log \BP_{n}}{\partial  \alpha}

3868: = \frac{1}{\alpha z^2} \bigg(\frac{\partial  \tilde

3869: Q}{\partial \tau}\bigg) = \frac{z}{\rho} e^{- \tau z^2}

3870: \bigg( \frac{\partial  \tilde Q}{\partial \tau} \bigg)

3871: %\label{C029}  \\

3872: %\end{equation} \bea

3873:  , \qquad \frac{\partial  \log

3874:  \BP_{n}}{\partial  b} = \frac{1}{z}

3875:  \bigg(\frac{\partial \tilde Q}{\partial

3876:   x}\bigg) .

3877: %~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

3878: %\label{C030}

3879: $$

3880: %

3881: Hereafter, to shorten notation, we will write the partials as %shall make a

3882: %systematic use of the following notation: an index

3883: %$\tau$ or $x$ stands for a partial derivative with

3884: %respect to the associated variable (for example

3885: $$

3886: \tilde Q_{\tau x} \equiv \frac{\partial^{2}}{\partial

3887: \tau

3888: \partial x} \tilde Q ,~\tilde Q_{\tau \tau x}  \equiv

3889: \frac{\partial^{3}}{\partial \tau^{2} \partial x} \tilde

3890: Q,~~ \mbox{ etc.}

3891: $$ 

3892: Without taking a limit yet, but expanding

3893: asymptotically the expressions in powers of $z$, we find

3894: %the expressions in Appendix 2,

3895: %

3896: %

3897: %from which one deduces

3898: %

3899: {\footnotesize

3900:  \bean

3901:   \bigg(F^+\BB_{-1}F^- -F^-\BB_{-1}F^+\bigg)

3902:    &=&  \frac{ \tilde Q_{\tau x x} }{\rho z^7}

3903: - \frac{\tau \tilde  Q_{\tau x x} }{\rho z^5}

3904:  %\\&&

3905:   +\frac{\tilde

3906: Q_{xxx}(r\rho-\tilde Q_{\tau x})+\tilde Q_{\tau

3907: xx}(\tilde

3908: Q_{xx}+\frac{\tau^2}{2})}{\rho z^3}  + O(\frac{1}{z}) \\

3909:   \bigg(F^+\BB_{-1}^2F^- -F^-\BB_{-1}^2F^+\bigg)

3910:      &=&\frac{ \tilde Q_{\tau x x x} }{\rho z^8}

3911: - \frac{\tau  \tilde Q_{\tau x x x} }{\rho z^6}

3912:  %\\&&

3913:  +

3914:    \frac{\tilde Q_{xxxx}(r\rho-\tilde Q_{\tau x})+\tilde

3915: Q_{\tau xxx}(\tilde Q_{xx}+\frac{\tau^2}{2})}{\rho z^4}

3916:  + O(\frac{1}{z^2}). \eean}

3917: %

3918: In order to compute the expansions of $G^{\pm}$ (which

3919: are respectively defined in (\ref{C012}) and

3920: (\ref{C015}))  and of $\BB_{-1} G^{\pm}$, we need the

3921: asymptotics of $H_i^{\pm}$, as defined in (\ref{Hplus})

3922: and (\ref{Hminus}). In the end, one finds

3923: %

3924: %

3925: %implying

3926: {\footnotesize \bean

3927:  %\lefteqn

3928:  %{F^+\BB_{-1}G^-+F^-\BB_{-1}G^+}\\

3929:  F^+ G^- + F^- G^+

3930:  &=&  \frac{2 \tilde Q_{\tau x

3931: x} }{\rho^2} \bigg[  2 \tilde Q_{\tau x}  +

3932: r\left(\rho-1\right)^2 \bigg] \frac{1}{z^{10}}

3933:  \\ &&-  \frac{2}{\rho^3 z^{8}}

3934:  \left(   \begin{array}{l}

3935:    \tilde Q_{\tau x}  \bigg( \left(\rho-1\right)

3936:    \tilde  Q_{\tau \tau x}

3937:     + 4 \rho \tau

3938:  \tilde  Q_{\tau x x}  \bigg)

3939:       %

3940:     \\ \\

3941:   -  2 r \rho  (\rho-1 ) ( \tilde Q_{\tau \tau x}

3942:      %\\ \\

3943:      +

3944:        \tau    \tilde Q_{\tau x x})

3945:      \end{array}

3946:      \!\! \right)

3947:     \! +\! \frac{\mathcal{T}}{z^{6}}\! +O(\frac{1}{z^4})

3948:      \\

3949:      F ^+ ~ \BB_{-1}G^-+F^-\BB_{-1}G^+&=&  \frac{2  \tilde Q_{\tau x x x}

3950: }{\rho^2} \bigg[ 2  \tilde Q_{\tau x}  +

3951: r\left(1-\rho\right)^2 \bigg] \frac{1}{z^{11}}

3952:  \\ &&- \frac{2}{\rho^3 z^{9}}

3953:  \left( \!\!  \begin{array}{l}

3954:    \tilde Q_{\tau x}  \bigg( \left(\rho\!-\!1\right) \tilde  Q_{\tau \tau x x}

3955:    + 4 \rho \tau

3956:   \tilde Q_{\tau x x x}  \bigg)

3957:       %

3958:     \\ \\

3959:     -  2 r \rho  (\rho-1 )  (\tilde  Q_{\tau \tau x x}

3960: %

3961:      + \tau  \tilde Q_{\tau x x x})

3962: %

3963:      \end{array}

3964:      \! \!\!\!\right)\!+\! \frac{\mathcal{T}'}{z^7}

3965:   %\\

3966:    %&&

3967:    \!\! + \!\! O(\frac{1}{z^{5}})

3968:   ,\eean}

3969: where $\mathcal{T}$ and $\mathcal{T}^\prime$ are given by the

3970: following expressions and where $\mathcal{T}_1$ and $\mathcal{T}_1'$ denote further expressions in the derivatives of

3971: $\tilde Q$,

3972: %(notice one only needs them for

3973: %$\rho=1$),

3974:  {\footnotesize \bean  \mathcal{T}%\Bigr|_{\rho=1}

3975:  &=&

3976:  2r^2\tilde Q_{xxx}

3977:  %

3978:  +2r(\tilde Q_{\tau xx}(\tilde Q_{xx}+\tau ^2-x)+2\tau \tilde

3979:  Q_{\tau \tau x}+2\tilde Q_{\tau x}+\tilde Q_{\tau \tau \tau })

3980:  -2\tilde Q_{\tau x}^2(\tilde Q_{xxx}+2)\\

3981: &&

3982:  %-2\tilde Q_{\tau x}^2(\tilde Q_{xxx}+2)

3983:   -2\tilde Q_{\tau x}(\tau \tilde

3984:   Q_{\tau \tau x}+\tilde Q_{\tau \tau \tau }-\tilde Q_{\tau xx}

3985:   (\tilde Q_{xx}+4\tau ^2+x))

3986:  %\\&&

3987:  +2\tilde Q_{\tau xx}\tilde Q_{\tau }+ \tilde Q_{\tau \tau }

3988:   \tilde Q_{\tau \tau x}

3989:    +(\rho-1)\mathcal{T}_1,

3990:  \eean}

3991: %

3992:  {%\footnotesize

3993:  \bean \mathcal{T}'%\Bigr|_{\rho=1}

3994:   \!&=&\!2r^2 \tilde Q_{xxxx}

3995:  +2r(\tilde Q_{\tau xxx}(\tilde Q_{xx}+\tau ^2-x)+2\tau \tilde

3996:  Q_{\tau \tau xx}+\tilde Q_{\tau xx}+\tilde Q_{\tau \tau \tau x})

3997:  \\

3998:  &&-2\tilde Q_{\tau x}^2\tilde Q_{xxxx}-2\tilde Q_{\tau x}

3999:  (\tau \tilde Q_{\tau \tau xx}+\tilde

4000:  Q_{\tau \tau \tau x}+2\tilde Q_{\tau xx}

4001:  -\tilde Q_{\tau xxx}

4002:   (\tilde Q_{xx}+4\tau ^2+x))

4003:    \\

4004:   &&

4005: +  \tilde Q_{\tau \tau x}^2+ \tilde Q_{\tau \tau }\tilde

4006: Q_{\tau \tau xx}

4007:   -\tilde Q_{\tau xx}\tilde Q_{\tau \tau \tau }

4008:   +2\tilde Q_{\tau xxx}\tilde Q_{\tau }+(\rho-1)\mathcal{T}_1'.

4009: \eean}}

4010: %

4011: %$$

4012: %T3=\frac{\tilde Q_{xxxx}(r\rho-\tilde Q_{\tau x})+\tilde

4013: %Q_{\tau xxx}(\tilde Q_{xx}+\frac{\tau^2}{2})}{\rho z^4}

4014: % $$

4015: %

4016:  %$$

4017:  % +2\tilde Q_{tt}\tilde Q_{ttx}

4018:  %  +4\tilde Q_{txx}\tilde Q_{t}

4019: %$$

4020: %\newpage

4021:  %

4022: Consequently, using the two leading orders for the

4023: expressions above, one obtains for small $z$:

4024: \be \label{expansion6}\ee

4025: \vspace*{-1cm}

4026: {\footnotesize\bean

4027:  0 &=&  \!\!\!\!\!\!\!\left.\left\{\!\!\begin{array}{l}

4028: \Bigl(F^+\BB_{-1}G^-+F^-\BB_{-1}G^+ \Bigr)

4029: \Bigl(F^+\BB_{-1}F^- -F^-\BB_{-1}F^+ \Bigr)  \\

4030:  \,-\,

4031:  \Bigl(F^+ G^- +F^- G^+

4032: \Bigr) \Bigl(F^+\BB_{-1}^2F^- -F^-\BB_{-1}^2F^+ \Bigr)

4033: \end{array}\!\!\!\right\}

4034:  \right|_{\begin{array}{l}

4035:      \alpha \,\mapsto\, \left( \rho / z^{ 3} \right)  e^{ \tau \,z^{2} } \\

4036:      b  \mapsto  x z + \frac{2}{z^{ 3}} \\

4037:      n  \mapsto  \frac{1}{z^{ 6}}\end{array} }

4038:  \\  %&&\no\\

4039:  &=& \frac{4(\rho-1)}{\rho^4 z^{16}}

4040:     %\frac{4}{\rho^4} \big(\rho-1\big)

4041:      \bigg( 2 r\rho  -  \frac{\partial^{2}\tilde Q}{\partial \tau \partial x}

4042:     \bigg)  \left\{\frac{\partial^{3}  \tilde Q}{\partial \tau^2 \,\partial x} \,,\,

4043:    \frac{\partial^{3}\tilde  Q }

4044:    {\partial \tau  \partial x^2} \right\}

4045:    _{x}  +

4046:     \frac{\mathcal{E}(\tilde Q)

4047:     +(\rho-1)\mathcal{F}(\tilde Q)}{z^{14}}

4048:     + O\left(\frac{1}{z^{12}}\right) ,

4049:    \no

4050:     \eean }

4051: %

4052: %

4053: where $\mathcal{E}(\tilde Q)$ is given by

4054: % \be\label{PDE6} \ee

4055:  %\vspace*{-1cm} 

4056:   \bea     \mathcal{E}(\tilde Q) &=&  2 \bigg( r - \frac{\partial^{2} \tilde

4057: Q}{\partial \tau

4058: \partial x} \bigg)^{2} \bigg\{\frac{\partial^{3}

4059: \tilde Q}{\partial \tau

4060:  \partial x^2} ,  \frac{\partial^{3}   \tilde Q}{\partial x^3}

4061:  \bigg\}_{x} \no\\%\eean\bea%\no\\

4062:  & &+ 2 \bigg( r - \frac{\partial^{2}  \tilde Q}

4063:  {\partial \tau  \partial x}

4064: \bigg)  \bigg\{ \frac{\partial^{3} \tilde  Q}{\partial

4065: \tau

4066:  \partial x^2}  ,  \frac{\partial}{\partial \tau}

4067: \Bigg( \frac{\partial}{\partial \tau}  \bigg(

4068: \frac{\partial \tilde  Q}{\partial \tau} + \tau

4069: \frac{\partial

4070:  \tilde Q}{\partial x} \bigg)  -  x \frac{\partial^2

4071:  \tilde Q}{\partial x^2} \Bigg)

4072:  \bigg\}_{x} \no\\

4073:  &&+ \bigg\{ \frac{\partial^{3} \tilde  Q}

4074:  {\partial \tau  \partial x^2}

4075:  ,

4076: \frac{\partial^{3}  \tilde Q}{\partial \tau^2 \partial

4077: x} \bigg( 2r \tau + \frac{\partial^2   \tilde

4078: Q}{\partial \tau^2} \bigg) \bigg\}_{x} +

4079: \bigg(\frac{\partial^3 \tilde Q}{\partial \tau

4080: \partial x^2}\bigg)^{2}  \frac{\partial}{\partial \tau}

4081: \bigg( \frac{\partial^2   \tilde Q}{\partial \tau^2} + 2

4082:  \frac{\partial   \tilde Q}{\partial x} \bigg)  ,

4083:  \label{PDE6}   \eea

4084: %

4085: %

4086: %precisely the equation (\ref{PDE6}) below for $\tilde Q$

4087: where $\mathcal{F}(\tilde Q)$ is a similar expression,

4088: that will not be needed, and where the bracket is a

4089: Wronskian associated to the ``space''

4090: operator ${\partial}/{\partial x}$. %\\

4091: 

4092: \vspace{.4cm}

4093: 

4094: Hence, \underline{\em for $0\leq \rho < 1$}, taking the

4095: limit $z\rightarrow 0$ above and using (\ref{6.3}),

4096: yields the equation

4097:  $$\bigg( 2 r\rho  -  \frac{\partial^{2}

4098: Q}{\partial \tau \partial x}

4099:     \bigg)  \left\{\frac{\partial^{3}    Q}{\partial \tau^2

4100:     \partial x} ~,~

4101:    \frac{\partial^{3}  Q }

4102:    {\partial \tau  \partial x^2} \right\}

4103:    _{x}  =0

4104:    $$

4105: which is trivially satisfied; indeed, from other

4106: considerations we know that $e^Q$ is the Tracy-Widom

4107: distribution, which of course is $\tau$-independent.

4108: 

4109: %() of Theorem 0.2. \\

4110: 

4111: However, \underline{\em in the critical case $\rho=1$}, the

4112: leading term has order $1/ z^{14}$, with coefficient

4113: $\mathcal{E}(\tilde Q(\tau,x;z))$, as in (\ref{PDE6}). Then

4114: taking a limit in (\ref{expansion6}) when $z\rg 0$ and

4115: using (\ref{6.3}), forces upon us the equation

4116: $$\lim_{z\rg 0}\mathcal{E}(\tilde Q(\tau,x;z))

4117: =\mathcal{E}(Q(\tau,x))=0,$$

4118:  which an easy computation shows can be written

4119:  as

4120:  \begin{equation} \label{one-time PDE}

4121: \begin{split}

4122: &\left\{\frac{\pl^3 Q}{\pl \tau\pl x^2},

4123:   \left[\begin{array}{l}

4124:    r^{2}

4125:   \frac{\pl^3 Q}{\pl x^3}

4126:  +r\left(\frac{\pl^2}{\pl \tau^2}(\frac{\pl Q}{\pl \tau}

4127:   +2\tau \frac{\pl Q}{\pl x}

4128:  )

4129:    -\frac{\pl^3 (x Q)}{\pl \tau\pl x^2}

4130:    +2\left\{

4131:    \frac{\pl^2Q}{\pl \tau\pl x},\frac{\pl^2Q}{\pl x^2}

4132:    \right\}_x

4133:  \right) \\

4134:   +\frac{1}{2}

4135:  \frac{\pl^3Q}{\pl \tau^2\pl x}\frac{\pl^2 Q}{\pl \tau^2}

4136:   -\frac{\pl^2Q}{\pl \tau\pl x}\frac{\pl^3Q}{\pl \tau^3}

4137:   +\left(\frac{\pl^2Q}{\pl \tau\pl x}\right)^2\frac{\pl^3Q}

4138:   {\pl x^3}+

4139:   \left\{ \frac{\pl^2Q}{\pl \tau\pl x},

4140:    \frac{\pl (\tau Q)}{\pl \tau^2}

4141:   \right\}_x

4142:   \end{array} \right]

4143:   \right\}_x

4144:   \\

4145:  %\be

4146: & -\frac{1}{2}\left( {\frac {\pl

4147: ^{3}Q}{\pl \tau\pl x^2}} \right) ^{2}\left({\frac

4148: {\partial ^{3}Q}{\partial {\tau}^{ 3}}}

4149:  -4\frac{\pl^2Q}{\pl \tau\pl x}  \frac {\pl ^{3}Q}{\pl

4150: x^3}\right)=0,

4151: \end{split}

4152: \end{equation}   %; see %(\ref{one-time PDE}).

4153:  and further rewritten as

4154:    equation (\ref{PDE0}), ending the proof of Theorem~0.4.

4155: \end{proof}

4156: %

4157: %Using (\ref{6.3}), the PDE obtained when

4158: %\mbox{$z\rightarrow 0$} then expresses the fact that the

4159: %coefficient of this term identically vanishes, replacing

4160: %$\tilde Q (\tau, x ; z)$

4161: %by $Q (\tau, x)$.%is also replaced by its limit when \,$z \rightarrow 0$.  \\

4162: 

4163: \begin{remark} This ``phase transition'' at $\rho=1$ is

4164: completely analogous to the results found in \cite{BBP}

4165: and \cite{Peche} for small rank perturbations of random

4166: Hermitian matrices.

4167: \end{remark}

4168: 

4169: 

4170: 

4171: 

4172: 

4173: \section{Remote past asymptotics}\label{sect7}

4174: 

4175: The aim of this section is to study the behavior of the

4176: $r$-Airy process $\mathcal {A}^{(r)}(t)$ for $t\rg -\iy$, as

4177: stated in Theorem \ref{Theo3}. In this section $\tau$

4178: will be systematically replaced by $t$. This theorem

4179: will be rephrased as Theorem \ref{theo:one-time

4180: expansion}, which includes some additional details.

4181: 

4182: 

4183: %\section{}

4184: %\newpage

4185: 

4186: 

4187: 

4188: %\begin{proposition}

4189: 

4190: 

4191: %\end{proposition}

4192: 

4193: 

4194: 

4195: 

4196: \begin{theorem}\label{theo:one-time expansion}

4197: 

4198: %$$

4199: %{\frac {r^2}{90}}  \left(

4200: % \begin{array}{l} 23\,{x}^{2}

4201: %  {   Q_0''} \left( x \right)

4202: %+173 x   Q_0'   +36     {Q_0'}

4203: %    {Q_0''} \\

4204: %     -108 {

4205: %Q_0}  +108 \int _{x}^{\infty } \left(  {Q_0''} \right)

4206: %^{2}{du}+{c_6} {Q_0'}\end{array}

4207: %  \right)

4208: %

4209: %$$

4210: %

4211: The log of the probability of the $r$-Airy process $

4212:  Q(t,x)= \\  \log\BP  (\sup\mathcal {A}^{(r)}(t)\leq x)

4213: $ admits an asymptotic expansion, as $t\rightarrow

4214: -\iy$, having the following form

4215: $$Q(t,x)=Q_0(x)+\frac{1}{t}Q_1(x)+\frac{1}{t^2}Q_2(x)+...

4216: ,$$ for the initial condition

4217: %

4218: %

4219: %Given the initial condition for $t\rightarrow -\iy$,

4220: $$

4221: \lim _{t\rightarrow -\iy}Q(t,x):=Q_0(x):=

4222:  \log\BP(\sup\mathcal {A}(t)\leq x)=%\log \det

4223: %(I-K^{(0)})=

4224: -\int_x^{\iy}(\al-x)g^2(\al)d\al ,$$

4225:  and where%solution to the PDE is given by

4226: %

4227: {\be %

4228:  \label{Qi}\ee}

4229: \vspace*{-1cm}{\footnotesize\bean Q_1&=&rQ'_0

4230: %\\

4231: ,~\quad Q_2=\frac{r^2}{2!}Q''_0

4232:  %\\

4233:  ,~\quad Q_3=\frac{r^3}{3!}Q'''_0+\frac{r}{3 } xQ'_0,

4234:  ~\quad Q_4=%\frac{r^4}{4!}F_0^{iv}

4235:   \frac{r^4}{4!}

4236:    Q_0^{iv}+\frac{r^2}{3}xQ_0''+\frac{7r^2}{12} Q_0',~~~

4237:  \\ %

4238: %\eean}\vspace*{-.51cm}{\footnotesize\bean%\\

4239: %

4240:  %\\

4241:    Q_5&=&

4242:    \frac{r^5}{5!}

4243:    Q_0^{v}+\frac{r^3}{3~2!}xQ_0'''+\frac{7r^3}{12} Q_0''

4244:    +\frac{r}{5} \mathcal{F}_5

4245: %   \left(\begin{array}{l}x^2Q_0'+4xQ_0+Q_0^{\prime 2}

4246: %   %

4247: %   +10\int_x^{\iy}Q_0

4248: %   \\-

4249: %   6\int_x^{\iy}dy\int_y^{\iy}Q_0^{\prime\prime 2}+c_5Q'_0

4250: %   \end{array}\right)%\eean

4251: %\\ \\

4252: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

4253: %\eean\bean

4254: \\

4255: Q_6&=&\frac{r^6}{6!}

4256:    Q_0^{vi}+\frac{r^4}{3~3!}xQ_0^{iv}+

4257:    \frac{7r^4}{12~2!} Q_0'''+{\frac {r^2}{5}}  \left(

4258:  %

4259:  \mathcal{F}_5'+\frac{5}{18}(x^2Q_0''+13(x+c_6)Q_0' ) \right)

4260: %

4261:   \\

4262:  \vdots\\

4263: Q_n&=&

4264:    \frac{r^n}{n!}

4265:    Q_0^{(n)}+\frac{r^{n-2}x}{3~(n-3)!} {Q_0^{(n-2)}}

4266:     +\frac{7r^{n-2}}{12(n-4)!}  {Q_0^{(n-3)}}

4267:    +\sum_{i=2}^{\left[\frac{n-1}{2}\right]}{r^{n-2i}}

4268:    Q_{n,n-2i}(x)%G_{ni}(x)%\left(\begin{array}{l}x^2F_0'+4xF_0+F_0^{\prime 2}

4269:    %

4270:    %+10\int_x^{\iy}F_0

4271:    \eean}

4272:    for some constant $c_6$ and with

4273:     $$

4274: \mathcal{F}_5:=  x^2Q_0'+4xQ_0+Q_0^{\prime 2}

4275:    %

4276:    +10\int_x^{\iy}Q_0

4277:    -

4278:    6\int_x^{\iy}dy\int_y^{\iy}du Q_0^{\prime\prime 2}

4279:     %

4280:    .$$

4281: %

4282: Also

4283:    \bean

4284:    \BP(\sup\mathcal {A}^{(r)}(t)\leq x)&=&

4285:     \BP\left(\sup\mathcal {A}(t)\leq (x+\frac{r}{t})(1+\frac{r}{3t^3})+

4286:     \frac{r^2}{4t^4}\right)

4287:     \\

4288:     &&~~~~~~~~~~~~~~~~~~~~\times ~\left(1+\frac{r\mathcal{F}_5}{5t^5}

4289:     %(\mathcal{F}_5+c_5Q'_0)

4290:      +O(\frac{1}{t^6})\right)

4291:     % \\

4292:     % &=&

4293:     %  e^{Q_0}\left(x(1+\frac{r}{3t^3})+\frac{r}{t}(1+

4294:     % \frac{7r}{12 t^3})\right)+O(\frac{1}{t^5})

4295:      .\eean

4296:      The mean and variance of the right edge of the process

4297:       behave as

4298:       \bean

4299:       {\mathbb E}

4300:        (\sup \mathcal {A}^{(r)}(t))&=&

4301:  {\mathbb E}(\sup\mathcal {A}^{(0)}(t))\left(1-\frac{r}{3t^3}\right)

4302:  -\frac{r}{t}-\frac{r^2}{4t^4} +O(\frac{1}{t^5})

4303:       \\

4304:        \mbox{\em var}(\sup \mathcal {A}^{(r)}(t))

4305:        &=&

4306:  \mbox{\em var}(\sup\mathcal {A}^{(0)}(t))\left(1-\frac{2r}{3t^3}\right)

4307:   +O(\frac{1}{t^5}).\eean

4308: 

4309:      \end{theorem}

4310: 

4311: 

4312: 

4313: \bigbreak

4314: %\newpage

4315: 

4316: \noindent Remember the $r$-Airy kernel

4317:    $$

4318:  K^{(r)}_t(u,v)=\int_0^{\iy}

4319:  dw A_r^-(u+w; t)A_r^+(v+w; t)

4320:  %\label{r-kernel2}

4321:  ,$$ as in (\ref{r-kernel}),

4322:  where $A^{\pm}_r(u;\tau)$ is given by (\ref{k-Airy0}),

4323: %

4324: % $$

4325: %A_r^-(u; t) =

4326: % \int_C   e^{\frac{1}{3} ia^3+iau }

4327: % \left(\frac{1}{ia-t}\right)^r

4328: % \frac{da}{2\pi}

4329: %  $$

4330: %  \be

4331: %A_r^+(u; t) =

4332: % \int_C ~e^{\frac{1}{3} ib^3+ibu }

4333: % \left( {-ib-t}\right)^r

4334: %  \frac{db}{2\pi}

4335: % ,\label{k-Airy2}\ee

4336: %

4337:  where $C$ is a contour running from $\iy

4338:  e^{5i\pi/6}$ to $\iy

4339:  e^{i\pi/6}$, such that $-it$ lies above the contour.

4340:  In this section one lets $t\rg -\iy$, which, of course,

4341: implies that $-it$ will remain above the contour $C$ and thus

4342: this limit is compatible with the contour just

4343: mentioned. Letting $t\rg +\iy$ would require a drastic

4344: change of the functions $A^{\pm}_r$.

4345:  In this section the subscript $t$ will occasionally be omitted from

4346:  the $r$-Airy kernel $K^{(r)}_t(u,v)$.

4347: Note that $K^{(0)}_t(u,v)=K^{(0)}(u,v)$ is the Airy

4348: kernel, which is independent of $t$,

4349: %$$K^{(0)}(x,y)=

4350: % \int^{\iy}_0du A(x+u)A(y+u)

4351: % .$$

4352:  $$K^{(0)}(u,v):=\frac{A(u)A'(v)-A'(u)A(v)}{u-v}

4353:   =%\\&=&

4354:  \int^{\iy}_{0}A(w+u)A(w+v)dw

4355:  ,$$

4356:   where $$A(u)=A_0^{\pm}(u)=\int_C   e^{\frac{1}{3} ia^3+iau }

4357:  %

4358:  \frac{da}{2\pi}$$ is the Airy function, satisfying the

4359:  ordinary

4360: differential equation $ A'' (u)=uA(u)$ and behaving

4361: asymptotically as

4362:  \be

4363: A(x)=\frac{e^{-\frac{2}{3}x^{3/2}}}{2\sqrt{\pi}x^{1/4}}

4364: (1+\sum_1^{\iy}\frac{\al_i}{(x^{3/2})^i}+\ldots),~~~\mbox{as

4365: $x\rg \iy$.}

4366: \label{Airy-asymptotics}

4367: \ee

4368:  The ODE and $\lim_{x\rg \iy}A(x)=0$ imply the following formulae,

4369: upon differentiation by $x$,

4370:  \bea 

4371:   \int_x^{\iy}A^2(u)du&=&(A^{'2}-A A'')(x) \no\\

4372:   \int_x^{\iy}A^{'2}(u)du&=&

4373:   -\frac{1}{3}\left( (A^2)'+x(A^{'2}-AA'')\right)(x)

4374: . \label{RemF}\eea

4375: 

4376: 

4377: 

4378: 

4379: \noindent Also remember the Tracy-Widom distribution

4380: \cite{TW-Airy}

4381:    \bean

4382:       \det

4383:   \left(I-K^{(0)}\right )_x=\exp\left(-\int^{\iy}_{x}(\al-x)g^2(\al)

4384: d\al \right)%=:F_2(u)

4385: , \eean%}

4386:  %where

4387: % $$K_0(u,v):=\frac{A(u)A'(v)-A'(u)A(v)}{u-v}

4388: %  =%\\&=&

4389: % \int^{\iy}_{0}A(u+y)A(v+y)dy

4390: % $$

4391: %  where $A(x)$ is the Airy functions, which satisfies the

4392: %differential equation

4393: %$$

4394: %A'' (x)=xA(x).$$

4395:  where

4396:   $g(\al)$ is the Hastings-MacLeod solution (\ref{Painleve})

4397:    of Painlev\'e II.

4398: %  behaving at infinity like the Airy function,

4399: % \be

4400: %\left\{\begin{array}{l} g''=\al g+2g^3\\ g(\al)\cong

4401: %  \frac{

4402: %  e^{-\frac{2}{3}  \al^{\frac{3}{2}}}}{2\sqrt \pi \al^{1/4}}

4403: %%~ \mbox{for}~ x \rightarrow \iy

4404: %\mbox{\,\,for\,\,}\al\nearrow \infty.

4405: %\end{array}\right. %(\mbox{{\bf Painlev\'e II}})

4406: %\label{Painleve}\ee

4407: %{\footnotesize(Tracy-Widom)\\(Adler-Shiota-van

4408: %Moerbeke)}

4409: %

4410: %

4411: The following shorthand notation will be used for

4412: integers $\ell\geq 1$,

4413: $$

4414: O(A^\ell)=O\left(  \left(\frac{

4415:   e^{-\frac{2}{3}  x^{\frac{3}{2}}}}{2\sqrt \pi x^{1/4}}

4416:   \right)^\ell x^k\right),~~\mbox{for $x\rg \iy$},

4417:    $$

4418: whatever be the power $k\in {\mathbb R}$.

4419: 

4420: 

4421: 

4422: \begin{lemma}\label{lemma:Qprime}

4423: Given

4424: $$

4425: Q_0(x)=\log \det \left(I-K^{(0)}\right

4426: )_x=-\int_x^{\iy}(\al-x)g^2(\al)d\al

4427: ,$$

4428: one checks \bean

4429: %

4430:  Q_0'(x)&=&\int_x^{\iy}

4431: g^2(u)du=A^{'2}(x)-A''(x)A(x)%\int_x^{\iy} A^2(u)du

4432:  +O(A^3)%

4433: %\\

4434: % Q_0''(x)&=&-g^2(x)= -A^2(x)+O(A^3)

4435: % \\

4436: %  Q_0'''(x)&=&-(g^{2 })'(x)= -(A^{2})'(x)+O(A^3)

4437: % \\

4438: %  Q_0^{iv}(x)&=&-(g^{2 })''(x)= -(A^{2})''(x)+O(A^3)

4439: %  \\

4440: %  \vdots

4441:   \\

4442:    Q_0^{(n)}(x)&=&-(g^{2 })^{(n-2)}(x)

4443:    = -(A^{2})^{(n-2)}(x)+O(A^3),~~~~\mbox{for $n\geq 2$}.

4444: %  \\

4445: %  \vdots%

4446:   \eean

4447:  \end{lemma}

4448: \begin{proof} The estimates follow from (\ref{RemF}) and an

4449: improved version of the estimate (\ref{Painleve}) by

4450: Hastings-McLeod \cite{HM}, namely:

4451: %

4452: $$

4453: g(x)= \mbox{A}(x)+ { 

4454: O}\left(\frac{e^{-\frac{4}{3}x^{3/2}}}{x^{1/4}}\right)

4455: \mbox{~for~}x\nearrow\iy,

4456: $$

4457: which is to be interpreted as a genuine asymptotic formula; i.e., it can be both integrated and differentiated.% combined with (\ref{RemF}).

4458:  \end{proof}

4459: 

4460:   \bigbreak

4461: %

4462: \noindent For future use, one needs the following

4463: estimates for the Airy function:

4464: 

4465: \begin{lemma}\label{lemma:Airy}For $x\rg \iy$, one has

4466: the estimates

4467: %

4468: \bean

4469:   \int_x^{\iy} A(u)du&=&

4470:  \frac{e^{-\frac{2}{3}x^{3/2}}}{2\sqrt{\pi} x^{3/4}}

4471:   (1+\sum_1^{\iy}\frac{c'_i}{(x^{3/2})^i} )=O(A)

4472: \\

4473: \int_x^{\iy}\!\! A^2(u)du&=&(A^{'2}-A''A)(x)=

4474:  \frac{e^{-\frac{4}{3}x^{3/2}}}{8\pi x}

4475:   (1+\sum_1^{\iy}\frac{c''_i}{(x^{3/2})^i} )=O(A^2)

4476:  .\eean

4477: %$$

4478: %\int_x^{\iy} dy\int_y^{\iy} A^2(u)du=

4479: % \frac{e^{-\frac{4}{3}x^{3/2}}}{16\pi x^{3/2}}

4480: %  (1+\sum_1^{\iy}\frac{c'''_i}{(x^{3/2})^i} )

4481: %$$

4482: 

4483: %$$

4484: % -A''(x)A(x)+A^{'2}(x)=\frac{e^{-\frac{4}{3}x^{3/2}}}{8\pi x}

4485: %  (1+\sum_1^{\iy}\frac{c^{i v}_i}{(x^{3/2})^i} )

4486: %$$

4487: %$$

4488: %\int_x^{\iy}(A''(u)A(u)-A^{'2}(u) )du=

4489: %-\frac{e^{-\frac{4}{3}x^{3/2}}}{16\pi x^{3/2}}

4490: %  (1+\sum_1^{\iy}\frac{c^{'''}_i}{(x^{3/2})^i} )

4491: %$$

4492: 

4493: 

4494: \end{lemma}

4495: 

4496: \begin{proof} Upon using integration by parts and upon substituting the asymptotic formula (\ref{Airy-asymptotics}) for the Airy function, one computes for

4497: instance,

4498:  \bean

4499:   \int_x^{\iy} A^2(u)du&=&

4500:  \int_x^{\iy}\frac{-1}{8\pi u}

4501:  (1+\sum_1^{\iy}\frac{\tilde c'_i}{(u^{3/2})^i})

4502: d(e^{-\frac{4}{3}u^{3/2}})\\

4503: &=&

4504:  \frac{e^{-\frac{4}{3}x^{3/2}}}{8\pi x}

4505:   (1+\sum_1^{\iy}\frac{\tilde c'_i}{(x^{3/2})^i} )-

4506:   \!\frac{1}{8\pi}\!

4507:    \int_x^{\iy}\frac{e^{-\frac{4}{3}u^{3/2}}}{ u^2}

4508:   (1+\sum_1^{\iy}\frac{\tilde c''_i}{(u^{3/2})^i} ).

4509: \eean Further terms in the expansion can be obtained by

4510: differentiation by parts % Concerning the fourth

4511: %expression, it is more convenient to compute \bean

4512: % A''A-A^{'2}=A^2(\log A)''&=&

4513: %  -\frac{e^{-\frac{4}{3}x^{3/2}}}{8\pi x}

4514: %  (1+\sum_1^{\iy}\frac{c'_i}{(x^{3/2})^i} )

4515: %  ,\eean

4516: %

4517: and similarly for the first expression, thus ending the proof

4518: of Lemma~\ref{lemma:Airy}.\end{proof}

4519: 

4520: 

4521: 

4522: 

4523:  Given a kernel $F(y,u)$ acting on

4524: $L^2(E)$ with $E\subset {\mathbb R}$ and a bounded continuous

4525: real function $f$ on $E$, define the norms

4526: %

4527: %\newcommand{\vv}{||}

4528: %

4529: \bean \vv F\vv_1=\sup_{y\in E}\int_E

4530: |F(y,u)|du~~~~\mbox{and}~~ ~~\vv f\vv_{\iy}=\sup_{y\in

4531: E}|f(y)| .\eean

4532:  If $f$ is a function of several variables, the sup is

4533:  taken over all variables.

4534:  Then

4535:   \bean

4536: \vv FG\vv_1&\leq&\sup_{y\in E}\int_E dz\int_E du |F(y,u)G(u,z)|\\

4537: \\

4538:   &\leq&\sup_{y\in E}\int_E du |F(y,u)|\sup_{u\in E}\int_E dz|G(u,z)|=\vv

4539: F\vv_1\vv G\vv_1 \eean

4540: %

4541: Hence\footnote{Obviously the estimate below requires $\vv

4542: F\vv_1<1$. In the application, this is achieved by restricting the domain of the operator to the interval $(x,\iy)$ for sufficiently large $x$.}

4543: %

4544: \bean \vv \sum_1^{\iy}F^n\vv_1&\leq&\sum_1^{\iy}\vv

4545: F\vv_1^n=\frac{\vv F\vv_1}{1-\vv F\vv_1} \eean

4546:   and

4547:    \bean

4548: \vv Ff\vv_{\iy}=\sup_{y\in E}\int_E |F(y,u)  f(u)|du

4549: &\leq&\sup_{u\in E} |f(u)|\sup_{y\in E}\int_E |F(y,u)|du

4550:  %\\

4551:  %\\

4552: %&=&

4553: =\vv F\vv_1\vv f\vv_{\iy}. \eean

4554: %

4555: %

4556: Given the kernel $K^{(0)}$, define the resolvent kernel

4557: $R$ by \be I+R:=(I-K^{(0)})^{-1}\label{resolvent}.\ee

4558: Then readily

4559: %

4560: \be

4561:  R-K^{(0)}= K^{(0)2}(I+K^{(0)}+K^{(0)2}+\ldots)

4562:   = (I-K^{(0)})^{-1}K^{(0)2}.

4563:  \label{R-K}\ee

4564: 

4565:  \bigbreak

4566: 

4567: \begin{lemma}\label{functA}One has the following

4568: estimates \bean

4569:  \vv K^{(0)}\vv_{1}=O(A^2),~~%\\

4570:  \vv K^{(0)}\vv_{\iy}=O(A^2),\qquad\mbox{and}\qquad%\\

4571:  \vv R-K^{(0)}\vv_{\iy}=O(A^4).

4572: \eean

4573: \end{lemma}

4574: \begin{proof} Using the fact that $A(u)\geq 0$ is monotonically

4575: decreasing for $u\geq 0$, and setting $E=(x,\iy)$,

4576: %

4577: \bean \vv K^{(0)}\vv_1&=&\sup_{u\in

4578: (x,\iy)}\int_x^{\iy}dv\left|\int_0^{\iy} dw

4579: ~A(u+w)A(v+w)\right|\\

4580: %\eean\bean%\\

4581: &\leq&\sup_{u\in (x,\iy)}\int_x^{\iy}dv~A(v)\int_0^{\iy} dw

4582: ~A(u+w)  \\

4583:  \\

4584: & \leq & \left(\int_x^{\iy}dv~A(v)\right)^2={ O}(A^2)

4585: \eean by Lemma \ref{lemma:Airy}, while

4586: %

4587: \bean

4588: \vv K^{(0)}\vv_{\iy}=\sup_{u,v\in (x,\iy)}|K(u,v)|%\\

4589: %\\

4590: &=&\sup_{u,v\in (x,\iy)}\left|\int_0^{\iy}  A(u+w) A(v+w)dw\right| \\

4591: \\

4592: &\leq& \int_0^{\iy} A(x+w)^2dw\\

4593: \\

4594: &=&\int_x^{\iy}A(w)^2dw={ O}(A^2), \eean also by

4595: Lemma \ref{lemma:Airy}.

4596: %

4597: %Remember

4598: %

4599: %$$

4600: %R-K^{(0)}=(I-K^{(0)})^{-1}K^{(0)2}.

4601: %$$

4602: Considering the function $R(\cdot,v)$ for fixed $v$, one

4603: has, using (\ref{R-K}),

4604: %

4605: \bean \vv R(\cdot,v)-K^{(0)}(\cdot,v)\vv_{\iy}&=&\vv

4606: (I-K^{(0)})^{-1}(K^{(0)2}(\cdot,v))\vv_{\iy} \\

4607: \\

4608: &\leq&\vv (I-K^{(0)})^{-1}\vv_1~\vv K^{(0)2}(\cdot,v)\vv_{\iy} \\

4609: \\

4610: &\leq&\frac{1}{1-\vv K^{(0)}\vv_1}\vv K^{(0)2}(\cdot,v)\vv_{\iy}\\

4611: \\

4612: &\leq&\frac{1}{1-\vv K^{(0)}\vv_1}\vv K^{(0)}\vv_1~\vv

4613: K^{(0)}(\cdot,v)\vv_{\iy}. \eean

4614: %

4615: Hence, with $\vv K\vv_{\iy}:=\displaystyle{\sup_{u,v \in

4616: E} |K(u,v)|}$,

4617: %

4618: \vspace{-.4cm}

4619: 

4620: $$

4621: \vv R-K^{(0)}\vv_{\iy}:=\sup_{u,v\in

4622: E}|R(u,v)-K^{(0)}(u,v)|\leq\frac{\vv K^{(0)}\vv_1}{

4623: 1-\vv K^{(0)}\vv_1}\vv K^{(0)}\vv_{\iy}={ O}(A^4) ,$$

4624: ending the proof of Lemma \ref{functA}.

4625: \end{proof}

4626: 

4627: 

4628: 

4629: 

4630: The next point is to get an asymptotic expansion for the

4631: Fredholm determinant $\det (I-K_t^{(r)})_{x,\iy}$ in

4632: $t$, using the asymptotic expansion  of the kernel

4633: $K_t^{(r)}(u,v)$ in $t$ (Lemma \ref{lemma:kernels}); this

4634: leads to the next Lemma:% have

4635: %

4636: %\bean K^{(r)}(u,v) &=&

4637: % K_0^{(r)}+\frac{K_1^{(r)}}{t}+\frac{K_2^{(r)}}{t^2}+\frac{K_3^{(r)}}{t^3}+\ldots

4638: %\eean Then we have

4639: 

4640: \begin{lemma}\label{lemma:F}

4641: The following probability for the $r$-Airy process has

4642: an asymptotic expansion in $1/t$ of the following form

4643:  \bean

4644: Q(t,x)&=& \log\BP(\sup\mathcal {A}^{(r)}(t)\leq x)=

4645: \log\det\left(I-K_t^{(r)}\right)_x%\\

4646: %&=&

4647: = Q_0(x)+\sum^{\iy}_{n=1}\frac{Q_n(x)}{t^n} ,\eean

4648:  where

4649:  \be

4650:  Q_n(x)=

4651:  \sum_{i=0}^{\left[\frac{n-1}{2}\right]}

4652:   r^{n-2i}Q_{n,n-2i}(x)%\no\\

4653:   =

4654:    -\Tr K_n^{(r)}+O(A^4)

4655:   ,

4656:   \label{Qn}

4657:   \ee

4658: %and

4659:  where $Q_0=\log\BP(\sup\mathcal {A}(t)\leq x)$ and where the $Q_i\rightarrow 0$ and have all their

4660: derivatives $\rightarrow 0$ for $x \rightarrow \iy$.

4661: Moreover, for $x \rightarrow \iy$, one has

4662: {  \bean

4663: Q_1(x)&=& r(A^{'2}-AA''  +O(A^4))\\

4664: Q_2(x)&=& -\frac{r^2}{2}(A^2+O(A^4))\\

4665: Q_3(x)&=& -\frac{r^3}{3!}((A^2)'  +O(A^4))

4666:  %\\

4667:  %&&

4668: -\frac{r}{3}

4669: \bigl((A^2)'+3\int_x^{\iy}A'^2(u)du+O(A^4)\bigr)+O(A^4)\\

4670:  \vdots

4671:  \\

4672: Q_n(x)&=& -\frac{r^n}{n!}((A^2)^{(n-2)} +O(A^4))

4673:   \\

4674:   &&

4675: -\frac{r^{n-2}} {(n-3)!} \left( \frac{3n-1}{24}(A^{2})''

4676: -\frac{n-1}{2}(A')^2 +O(A^4)\right)

4677:  ^{(n-4)}

4678:  \\

4679: %

4680: % Q_n(x)

4681:  && %- \Tr

4682:  %K_n^{(r)}+O(A^4)

4683:  +\sum_{i=2}^{\left[\frac{n-1}{2}\right]}

4684:   r^{n-2i}T_{n,n-2i}(A)+O(A^4)

4685: %%

4686: %

4687: %  \sum_{i=4 \atop i~even}^{n-1}

4688: %r^{n-i}(T_i(A) +O(A^4))

4689:   ,\eean}

4690:  with $T_{n,n-2i}(A)=$ quadratic polynomial of $A$

4691: and $ ~A'$, with coefficients depending on $x$ $+$ $

4692: \int_x^\iy $(quadratic polynomial of $A$ and $ ~A'$,

4693: with coefficients depending on $x$).

4694: \end{lemma}

4695: 

4696: \begin{proof} We shall always operate in $L^2(x,\iy)$, so that

4697: occasionally the $x$ will be suppressed. Then, using the

4698: asymptotics for the kernel $K_t^{(r)} $ as in Lemma

4699: \ref{lemma:kernels}, one has the following:

4700: %

4701:  $$ I-K_t^{(r)} =

4702: I-K^{(0)}-\frac{K^{(r)}_1}{t}-\frac{K^{(r)}_2}{t^2}-\ldots

4703: =(I-K^{(0)})(I-\sum_{i\geq 1} \frac{L_i}{t^i})

4704:  %~\mbox{with} ~~%(I-K_0)^{-1}K_1+(I-K_0)^{-1}K_2+\ldots

4705:   $$

4706:   with (the resolvent operator $R$ of the Airy kernel is

4707: defined in (\ref{resolvent}))

4708:   \be

4709:    L_i=(I-K^{(0)})^{-1}K_i^{(r)}=(I+R)K_i^{(r)}.

4710:    \label{L}\ee

4711: %

4712: %as $I+R=(I-K^{(0)})^{-1}$.%

4713:  Using $\log(1-z)=-z-\frac{z^2}{2}-\frac{z^3}{3}-\ldots$,

4714: one finds

4715: %

4716: $$

4717: Q(t,x)=\log\det\left(I-K_t^{(r)}\right)_x=\Tr\log(I-K_t^{(r)})=

4718: \Tr\log(I-K^{(0)})+\sum^{\iy}_{i=1}\frac{Q_i}{t^i} ,$$

4719: where

4720:  \bea  \label{Q's}

4721:   Q_1&=&-\Tr L_1,\quad Q_2=-\Tr

4722: (L_2+\frac{L_1^2}{2}), \quad Q_3=-\Tr \Bigl(L_3+{L_1L_2}

4723: +\frac{L_1^3}{3}\Bigr),

4724:  \\~~Q_4&=&-\Tr\Bigl(L_4+\frac{1}{2}L_2^2+L_1^2L_2+L_1L_3

4725:  +\frac{1}{4}L_1^4

4726: \Bigr),\ldots.\no\eea

4727:   More generally, the $Q_n$'s are weight-homogeneous polynomials of degree $n$

4728:    in the $L_i$, with weight$(L_i)=i$, having the form

4729:    below, which can further be expressed in terms of the

4730:    $K_i^{(r)}$ and $R$, using

4731:    expression (\ref{L}) for the $L_i$,

4732: %

4733:   \bea  \label{Qn7}

4734:   Q_n&=&-\Tr

4735: L_n+\Tr P_n(L_1,L_2,\ldots,L_{n-1})

4736:  \\

4737: &=&

4738:  -\Tr K^{(r)}_n-\Tr(RK^{(r)}_n)+\Tr P_n(L_1,L_2,\ldots,L_{n-1})\no\\

4739:  &=&

4740:   -\Tr K^{(r)}_n+\Tr S_n(K^{(r)}_1,\ldots,K^{(r)}_{n-1},R)

4741:   ,\no\eea

4742: %

4743: %  Notice the $Q_n$ are polynomials of degree $n$ in

4744: %$r$, missing every other term and containing no constant

4745: %term, as follows from the homogeneous structure of the

4746: %$Q_i$'s in terms of the $L_i$'s and the $L_i$'s; this

4747: %establishes (\ref{Qn}) .

4748: %

4749:     with $P_k$ and $S_k$ polynomials of non-commutative variables

4750: with no linear or independent terms, but with quadratic

4751: terms and higher.

4752: %

4753: %

4754: From Lemma \ref{lemma:kernels}, the kernels $K_i^{(r)}$

4755: and hence the $L_i$ are polynomials in $r$ of degree $i$

4756: having no constant terms; hence the $Q_n$'s, by their

4757: weight-homogeneity, are polynomials of degree $n$ in

4758: $r$, having no constant term and so (here one must

4759: indicate the $r$-dependence)

4760:  \be

4761:  Q_r(t,x):=Q(t,x)=\sum_{n=0}^{\iy}

4762:  \left(\frac{r}{t}\right)^n\sum_{i=0}^{n-1}

4763:  \frac{Q_{n,n-i}(x)}{r^i}.\label{star}\ee

4764:  The claim is that only the terms $Q_{n,n-2i}(x)$

4765:  appear. Observe from

4766:  (\ref{r-kernel}) and (\ref{k-Airy0}),

4767:  that $

4768:   K^{(-r)}_{-t}(u,v)=K^{(r)}_{t}(v,u)

4769:    ,$

4770:     and thus

4771:     \bean

4772:      Q_r(t,x)=\det (I-K_t^{(r)}(u,v))_x

4773:            =\det (I-K_t^{(r)}(v,u))_x

4774:            &=&\det (I-K_{-t}^{(-r)}(v,u))_x\no\\

4775:            &=&Q_{-r}(-t,x)

4776:            .\eean

4777: So from (\ref{star}), one has

4778:   $$

4779:   Q_r(t,x)=\frac{1}{2}(Q_r(t,x)+Q_{-r}(-t,x)),$$

4780:   implying that only the even terms appear in the

4781:   sum $\sum_i$ in (\ref{star}), thus proving the

4782:   statement (\ref{Qn}).

4783: 

4784: 

4785: 

4786: 

4787: 

4788: 

4789: 

4790: 

4791: 

4792: 

4793: 

4794:  Next we now proceed to estimate the

4795: two traces above:

4796: 

4797: \noindent$\bullet$ At first

4798:  $$

4799:    \Tr S_k(K^{(r)}_1,\ldots,K^{(r)}_{k-1},R)=O(A^4)

4800:  ,$$

4801:  which we now illustrate on a typical example,

4802:  like $\Tr( RK_1^{(r)})$

4803: . Note $A(u)\geq 0$ for $u\geq 0$, and so by Lemma

4804: \ref{functA},  \bean

4805:  | \Tr RK^{(r)}_1|&=&|\Tr ((R-K^{(0)})K^{(r)}_1)|+|\Tr (K^{(0)}K^{(r)}_1)

4806: |  \\

4807:   &\leq& (|\!|R-K^{(0)}|\!|_{\iy}+|\!|K^{(0)}|\!|_{\iy})

4808:    \int\!\!\!\!\int_x^{\iy}|K_1^{(r)}(u,v)|du dv\\

4809:    &\leq& (O(A^4)+O(A^2))

4810:     \int\!\!\!\!\int_x^{\iy}A(u)A(v)du dv\\

4811:     %&=&

4812:     &\leq&O(A^2)\Bigl(\int_x^{\iy} A(u)du\Bigr)^2 \\

4813:     &\leq& O(A^4);

4814:   \eean

4815:   the last estimate follows from Lemma \ref{lemma:Airy}. More generally, the trace

4816:   of a monomial of degree $\ell$ has order

4817:   $O(A^{2\ell})$.

4818: 

4819: 

4820: \noindent $\bullet$ Then we evaluate $\Tr K^{(r)}_n$; in

4821: order to do so, it suffices to evaluate the kernels

4822: $K_n^{(r)}$ of Lemma \ref{lemma:kernels} along the

4823: diagonal, and to notice that a

4824: skew-symmetric operator vanishes on the diagonal. Since the domain of the operator is unbounded, one needs to consider $K_n^{(r)} \chi_{_{(x,m)}}$; the trace is then obtained by integrating  on the diagonal and by taking the limit $m\rg \iy$, upon using the decay of the kernel at $\iy$. %

4825: Therefore, on the diagonal, $K_n^{(r)}(u,v)$ is a

4826: polynomial of degree $n$ in $r$, skipping every other

4827: term,

4828: %

4829: \bean {K_n^{(r)}(u,u)}

4830: &=&-\frac{r^n}{n!}\Big(A^2(u)\Big)^{(n-1)}\\

4831: \\

4832: &

4833: &-\frac{r^{n-2}}{(n-3)!}\left(\frac{3n-1}{24}\Big(A^2(u)\Big)^{(n-1)}-\frac{n-1}{2}

4834: \Big(A^{' 2}(u)\Big)^{(n-3)}\right)+ ...

4835:  \eean

4836: %

4837:   That the

4838: $Q_i\rightarrow 0$ and that $Q_0=\log\BP(\sup \mathcal {A}(t)\leq

4839: x)$ has all their derivatives $\rightarrow 0$ for $x

4840: \rightarrow \iy$ follows from the statement on $T_{n,n-2i}(A)$ (which are the coefficients of $r^{n-2i}$ appearing in $Q_n(x)$ as in the statement of Lemma \ref{lemma:F} )

4841: and the asymptotics of the Airy function, from which

4842: Lemma \ref{lemma:F}  follows. \end{proof}

4843: 

4844: 

4845: %\begin{remark} Using now and then the differential equation for

4846: %the Airy function, here are the $K_n^{(r)}(u,u)$ for a

4847: %few

4848: %values $n=1,2,3,4$; namely%Then it suffices to

4849: %evaluate the kernels $K_i^{(r)}$ of Lemma

4850: %\ref{lemma:kernels} along the diagonal, using for

4851: %$K_4^{(r)}(u,u)$

4852: %

4853: %

4854: %\bean

4855: % K_1^{(r)}(u,u)&=&-rA^2(u),~~~~~~

4856: %  K_2^{(r)}(u,u)=-\frac{r^2}{2} (A^2(u))'\\ \\

4857: %   K_3^{(r)}(u,u)&=&-\frac{r^3}{6} (A^2(u))''

4858: % -\frac{r}{3}((A^2(u))''-3(A'(u))^2) \\ \\

4859: %   K_4^{(r)}(u,u)&=&-\frac{r^4}{24} (A^2(u))'''

4860: % -r^2(\frac{11}{24}(A^{2}(u))'''-\frac{3}{2}(A'^2(u))')

4861: % ,\eean

4862: 

4863: % \vspace{.3cm}

4864: %

4865: 

4866: %The Airy function behaves asymptotically as $$

4867: %A(x)=\frac{e^{-\frac{2}{3}x^{3/2}}}{2\sqrt{\pi}x^{1/4}}

4868: %(1+\sum_1^{\iy}\frac{\al_i}{(x^{3/2})^i}+\ldots)$$

4869: %

4870: %\newpage

4871: %

4872: 

4873: %\end{remark}

4874: 

4875: 

4876: 

4877: 

4878: 

4879:  \begin{proof}[Proof of Theorem~\ref{theo:one-time expansion}]

4880:  From Section~\ref{sect6}, we know that $Q(t,x)=\log   \BP(\sup\mathcal {A}^{(r)}(t)\leq x)$ satisfies the non-linear

4881:   PDE (\ref{PDE0}); it is more convenient here to use version

4882:   (\ref{one-time PDE}) of the equation.

4883: %

4884:   Also remember $Q(t,x)\rg 0$, when $x\rg \iy$. Then,

4885: assigning weight$=1$ to both variables $t$ and $r$, one

4886: readily checks that the PDE (\ref{one-time PDE}) can be

4887: graded as follows:

4888:   \bean

4889:  \hspace{-.7cm} 0\!\!\!

4890:    &=&   \left\{\frac{\pl^3Q}{\pl t\pl x^2}

4891:   ,r^{2}

4892:    \frac{\pl^3Q}{\pl

4893:    x^3}\right\}~~~~~~~~~~~~~~~~~

4894:    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~(\mbox{weight$=1$})

4895:   \\

4896:   &&+\left\{\frac{\pl^3Q}{\pl t\pl x^2},r\left(2\frac{\pl^2}{\pl t^2}(

4897:      t\frac{\pl Q}{\pl x}

4898:   )

4899:     -\frac{\pl^3 (x Q)}{\pl t\pl x^2}

4900:     +2\left\{

4901:     \frac{\pl^2Q}{\pl t\pl x},\frac{\pl^2Q}{\pl x^2} \right\}

4902:      %

4903:   \right)\right\}%+\dots

4904:   ~(\mbox{weight$=-1$})\\

4905:   &&+  \mbox{other terms of weight$<-1$}.

4906:   \eean

4907: Since, by Lemma \ref{lemma:F}, the solution has the

4908: following general form

4909:   $$

4910:   Q(t,x)=\sum_0^\iy \frac{Q_n(x)}{t^n}=

4911:    \sum_0^\iy \left(\frac{r}{t}\right)^n\left( Q_{nn}(x)+

4912:    \frac{1}{r^2}

4913: Q_{n,n-2}(x)+ \frac{1}{r^4} Q_{n,n-4}(x)+\ldots\right)

4914: ,$$

4915:   it follows that one can compute inductively all

4916:   the $Q_{nn}(x)$ and then inductively all the $Q_{n,n-2}(x)$ and so

4917:     the $Q_n$ will be as announced in Theorem

4918:   \ref{theo:one-time expansion}.

4919: 

4920: Setting this solution in the PDE above, yields a series

4921: of descending weights, namely $ 0=W_1+W_{-1}+W_{-3}+

4922: ..., $ which holds for $t\rg -\iy$, all $x\in {\mathbb R}$ and

4923: all integers $r>0$; this implies $W_1=W_{-1}=W_{-3}=...

4924: =0$; one then checks the explicit expressions

4925: %

4926: \bean

4927: W_1&=&-\sum^{\iy}_{n=1}\frac{r^{n+2}}{t^{n+1}}\sum_{j=0}^{n-1}\left\{(n-j)Q''_{n-j,n-

4928: j},Q'''_{jj}\right\}\\

4929: \\

4930: W_{-1}&=&\!-\sum^{\iy}_{n=1}\frac{r^{n}}{t^{n+1}}\sum_{j=0}^{n-1}\left(

4931: \!\!\!\begin{array}{l}

4932: \left\{(n-j)Q''_{n-j,n-j},\left(\begin{array}{l}

4933: Q'''_{j,j-2}+2(j-1)^2Q'_{j-1,j-1}\\

4934: +(j-1)x  Q''_{j-1,j-1}\\

4935: -2\displaystyle{\sum_{\ell

4936: +k=j-1}}\ell\{Q'_{\ell,\ell},Q''_{kk}\}

4937: \end{array}\right)\right\}\\

4938: \\

4939: +\left\{(n-j)Q''_{n-j,n-j-2},Q'''_{jj}\right\}

4940: \end{array}\!\!\!\right)

4941: .\eean

4942: %

4943: Since this holds for all $t\searrow -\iy$ and $r>0$

4944: integer, one must have for all $x$

4945:  \be

4946: \sum^{n-1}_{j=0}(n-j)\left\{Q''_{n-j,n-j},Q'''_{jj}\right\}=0,\qquad

4947: n=1,2,... \label{leading} \ee with $Q_{00}=Q_0(x)$.

4948: 

4949: %\bigbreak

4950: 

4951: \noindent $\bullet$ For $n=1$, this is $

4952: \{Q''_{11},Q'''_0\}=0, $ leading to

4953: $

4954: Q_{11}=c_0Q'_0+\al x+\beta.

4955: $

4956: %

4957: Considering the asymptotics for $x\nearrow\iy$ and using

4958: Lemmas \ref{lemma:Qprime} and \ref{lemma:F}, $\al$ and

4959: $\beta$ must $=0$, leading to the equation

4960: %

4961: $$

4962: 0=Q_{11}-c_0Q'_0=(A^{'2}-AA'')(1-c_0)+{ O}(A^3)

4963: =(1-c_0){ O}(A^2)+{ O}(A^3), $$ implying $c_0=1$

4964: and so $Q_{11}=Q'_0$.

4965: 

4966: %\bigbreak

4967: 

4968: \noindent $\bullet$ For $n=2$, the equation, with the

4969: previous data introduced, reads

4970: %

4971: \bean

4972: 0&=&\left\{2Q''_{22},Q'''_0\right\}+

4973: \left\{Q''_{11},Q'''_{11}\right\}%\\

4974: =\left\{Q'''_0,-2Q''_{22}+Q^{iv}_0\right\} ,\eean which

4975: upon solving leads to $

4976: Q_{22}=\frac{1}{2}Q''_0+c_1Q'_0+\al'x+\beta'. $ For the

4977: same reason as before $\al'=\beta' =0$. Then again using

4978: Lemma \ref{lemma:Qprime} and Lemma \ref{lemma:F}, one

4979: finds for $x\nearrow \iy$,

4980: %

4981: \bean

4982: 0=Q_{22}-\frac{1}{2}Q''_0-c_1Q'_0 

4983: &=&-\frac{A^2}{2}+\frac{A^2}{2}-c_1(A^{' 2}-A'' A)+{  O}(A^3) 

4984: \\ &=&c_1{  O}(A^2)+{  O}(A^3), \eean implying $c_1=0$

4985: and thus $Q_{22}=\frac{1}{2}Q''_0$.

4986: 

4987: \noindent $\bullet$ By induction, assume $

4988: Q_{ii}=\frac{1}{i!}Q_0^{(i)},\mbox{ for~} 0\leq i\leq

4989: n-1. $ Then substituting this identity into equation

4990: (\ref{leading}) and setting $

4991: Q_{nn}=\frac{1}{n!}Q_0^{(n)}+R_n $ leads to pairwise

4992: cancellations in equation (\ref{leading}) with only one

4993: remaining contribution $ \{Q_0''', R''_n\}=0, $

4994: %

4995: with solution $ R_n=c_nQ'_0+\al'' x+\beta '' ,$ where

4996: $\al '' =\beta '' =0$, and thus, by the asymptotics of

4997: Lemmas \ref{lemma:Qprime} and \ref{lemma:F},

4998: %

4999: \bean

5000: 0=Q_{nn}-\frac{1}{n!}Q^{(n)}_0-c_nQ'_0%\\

5001: &=&-\frac{1}{n!}(A^2)^{(n-2)}+\frac{1}{n!}(A^2)^{(n-2)}-c_n(A^{'

5002: 2}-AA'')+{ O}(A^3)\\

5003: &=&c_n{ O}(A^2)+{ O}(A^3) \eean leading to

5004: $c_n=0$, completing the proof that

5005: $Q_{nn}=\frac{1}{n!}Q^{(n)}_0$ for all $n=1,2,... $.

5006: This proves the form of the leading term (coefficient of

5007: $r^n$) in formulae (\ref{Qi}) for the $Q_n$'s. Since

5008: from Lemma \ref{lemma:kernels}, from the form

5009: (\ref{Qn7}) of the $Q_n$ and the fact that the

5010: coefficients $K_n^{(r)}(u,v)$ in the expansion of

5011: $K^{(r)}$ are divisible by $r$, the $Q_i$ themselves are

5012: divisible by $r$. Since they skip every other degree in

5013: $r$, this shows the formulae for $Q_1$ and $Q_2$; in

5014: particular $Q_{20}=0$.

5015: 

5016: 

5017: 

5018: Setting this information $Q_{nn}=\frac{1}{n!}Q^{(n)}_0$

5019: into the equation $W_{-1}=0$ and noticing that the

5020: following term vanishes automatically, $ \sum_{\ell

5021: +k=j-1}\ell\{Q'_{\ell\ell},Q''_{kk}\}=0, $ one finds for

5022: $n\geq 3$,

5023: \be \label{next} \ee

5024: \vspace*{-1cm}

5025: \bean

5026: 0&=&\sum^{n-1}_{j=2}\left\{\frac{Q_0^{(n-j+2)}}{(n-j-1)!},\left(Q'_{j,j-2}+\frac{1}{(

5027: j-2)!}

5028: \left(xQ_0^{(j-1)}+2(j-2)Q_0^{(j-2)}\right)\right)''\right\}\nonumber\\

5029: \nonumber\\

5030: & &-\sum^{n-1}_{j=0}\left\{\frac{Q_0^{(j+3)}}{j!},

5031: (n-j)Q''_{n-j,n-j-2}\right\}

5032: \no\\%\eean\bea%\no\\

5033: %

5034:  &=&\sum^{n-3}_{j=0}\left\{\frac{Q_0^{(j+3)}}{j!},

5035:   \left(\begin{array}{l}

5036:    Q'_{n-j-1,n-j-3}-(n-j)Q_{n-j,n-j-2}\\  \\

5037:    +\frac{1}{(n-j-3)!}\left(xQ_0^{(n-j-2)}+2(n-j-3)Q_0^{(n-j-3)}\right)

5038:    \end{array}

5039:    \right)''\right\}

5040:   %\no\\

5041:    \eean

5042: 

5043: 

5044: \noindent $\bullet$ For $n=3$, by using the fact that

5045: $Q_{20}=0$, the equation reads  $

5046: \left\{Q'''_0,(xQ'_0-3Q_{31})''\right\}$ $=0 $ yielding $

5047: Q_{31}=\left(\frac{x}{3}+c'_3\right)Q'_0+\al'''x+\beta'''

5048: $ with $\al'''=\beta'''=0$. Thus, using Lemmas

5049: \ref{lemma:F} and \ref{lemma:Qprime}, and the identity

5050: (\ref{RemF}),

5051: %

5052: \bean

5053: 0&=&Q_{31}-\left(\frac{x}{3}+c'_3\right)Q'_0

5054: \\

5055: &=&-\frac{1}{3}\left((A^2)'+3\int^{\iy}_xA^{'2}(u)du\right)-

5056: \left(\frac{x}{3}+c'_3\right)(A^{' 2}-AA'')+{ O}(A^3)

5057: %\\

5058:  \\

5059: %\eean\bean

5060:  &=& -\left( \int_x^{\iy}A^{'2}(u)du

5061:   +\frac{1}{3}\left(

5062:   (A^2)'+x(A^{'2}-AA'')\right)\right)-c'_3(A^{'2}-AA'')+{ O}(A^3)

5063:   %

5064: \\

5065: &=&-c'_3(A^{'2}-AA'')+{ O}(A^3)

5066:  =c'_3{\bf O}(A^2)+{ O}(A^3)

5067:   ,~~~~\mbox{using

5068:   (\ref{RemF})}%\\

5069: %\\

5070: %&=&c'_3{\bf O}(A^2)+{\bf O}(A^3),

5071:   \eean

5072:   %

5073:    yielding $c'_3=0$, and thus $Q_{31}=\frac{x}{3}Q'_0$.

5074: 

5075: \medbreak

5076: 

5077: \noindent $\bullet$ For $n=4$, using the formula for

5078: $Q_{31}$, the equation reads

5079: $$

5080: 0=\left\{Q'''_0,\left(-4Q_{42}+\frac{1}{3}\left(

5081: xQ_0'\right)'+xQ''_0+2Q'_0\right)'' \right\},

5082: $$

5083: with solution

5084: $

5085: Q_{42}=\frac{1}{3}xQ''_0+\left(

5086: \frac{7}{12}+c'_4\right)Q'_0+\al^{{\it iv}}x+\beta^{iv}

5087: $

5088: and thus $\al^{iv}=\beta^{iv}=0$, and by the same Lemmas

5089: \ref{lemma:F} and \ref{lemma:Qprime},

5090: %

5091: \bean

5092: 0&=&Q_{42}-\frac{1}{3}xQ''_0-\left( \frac{7}{12}+c'_4\right)Q'_0\\

5093: \\

5094: &=&\frac{3}{2}A^{'2}-\frac{11}{24}(A^2)''

5095: +\frac{1}{3}AA'' -\left(

5096: \frac{7}{12}+c'_4\right)(A^{'2}-AA'')+

5097: { O}(A^3)\\

5098: \\

5099: &=&\frac{3}{2}A^{'2}-\frac{11}{12}(A^{'2}+AA'')+\frac{1}{3}AA''

5100: -\left(

5101: \frac{7}{12}+c'_4\right)(A^{'2}-AA'')+ { O}(A^3)\\

5102: \\

5103: &=&-c'_4(A^{'2}-AA'')+{ O}(A^3)%\\

5104: %\\

5105: %&=&

5106: =c'_4{ O}(A^2)+{ O}(A^3), \eean implying

5107: $c'_4=0$.

5108: 

5109: \noindent $\bullet$ Using induction, assume

5110: $$

5111: Q_{i,i-2}=\frac{x}{3(i-3)!}Q_0^{(i-2)}+\frac{7}{12(i-4)!}Q_0^{(i-3)}

5112: $$

5113: holds for $i=3,..., n-1$. Then setting \be

5114: Q_{n,n-2}=\frac{x}{3(n-3)!}Q_0^{(n-2)}+\frac{7}{12(n-4)!}Q_0^{(n-3)}+R_n

5115: \label{ansatz}\ee into equation (\ref{next}) gives the

5116: simple equation for $R_n$, namely $ \{Q_0''',R''_n\}=0,

5117: $ and so $R_n=c'_nQ'_0. $ Then, rewriting (\ref{ansatz})

5118: and using the asymptotics for $Q_{n,n-2}$ (Lemma

5119: \ref{lemma:F}), and for the derivatives $Q_0^{(i)}$

5120: (Lemma \ref{lemma:Qprime}), and using the ODE for the

5121: Airy function $xA=A''$, we get

5122: \bean

5123: 0&=&Q_{n,n-2}-\frac{1}{12(n-3)!}(4xQ''_0+(3n-5)Q'_0)^{(n-4)}-c'_nQ'_0\\

5124: \\

5125: &=&\frac{1}{12(n-3)!}\left(\begin{array}{l}

5126: 6(n-1)A^{'2}-(3n-1)(A^{'2}+AA'')\\

5127: +4AA''-(3n-5)(A^{'2}-AA'')

5128: \end{array}

5129: \right)

5130: %

5131: ^{(n-4)}%\\

5132:  \\

5133:  & &

5134: -c'_n(A^{' 2}-AA'')+{ O}(A^3)%\\

5135: \\

5136: &=&-c'_n(A^{' 2}-AA'')+{ O}(A^3)%\\

5137: %\\

5138: =c'_n { O}(A^2)+{ O}(A^3) ,\eean implying

5139: $c'_n=0$. Thus

5140: %

5141: % it follows one can compute inductively all

5142: % the $Q_{nn}(x)$ and then inductively all the $Q_{n,n-2}(x)$ and so

5143:    the $Q_n$'s are as announced in Theorem

5144:  \ref{theo:one-time expansion}, namely

5145: %

5146:  $$

5147:  Q_n=

5148:    \frac{r^n}{n!}

5149:    Q_0^{(n)}+\frac{r^{n-2}x}{3}\frac{Q_0^{(n-2)}}{(n-3)!}

5150:     +\frac{7r^{n-2}}{12} \frac{Q_0^{(n-3)}}{(n-4)!}

5151:    +{r^{n-4}}G(x)+\left(\begin{array}{l}

5152:    \mbox{lower degree\!\!\!}\\

5153:    \mbox{terms in $r$}

5154:    \end{array}\right).%\left(\begin{array}{l}x^2F_0'+4xF_0+F_0^{\prime 2}

5155:    %

5156:   $$

5157: %

5158: %

5159: In the same fashion we compute $Q_{51}$ and $Q_{62}$;

5160: for example, setting

5161: $$ Q_5=

5162:  \frac{r^5}{5!}

5163:    Q_0^{(v)}+\frac{r^{3}x}{3}\frac{Q_0^{'''}}{2!}

5164:     +\frac{7r^{3}}{12} \frac{Q_0^{''}}{1!}

5165:    +rQ_{51}%\mbox{~~with~}Q_{51}=\frac{1 }{5} R(x)

5166:    $$

5167:    into the equation (\ref{one-time PDE}), one finds the following

5168:    differential equation for $Q_{51}$, namely

5169:    $$ \left\{Q_0''', 5Q_{51}''-8xQ_0''+4Q_0^{\prime\prime 2}  \right\}

5170: +2Q_0^{\prime\prime\prime 2}(Q_0^{\prime\prime } +x) =0,

5171:  $$

5172: % \bean

5173: %(xF'_0)''&=&2F''_0+xF'''_0\\

5174: %xF'''_0&=&(xF'_0-2F_0)''

5175: % \eean

5176: % $$

5177: %x^2 F^{iv}_0=(x^2F_0''-4xF_0'+6F_0)''

5178: %$$

5179: %

5180: %

5181: %

5182: %

5183:  which upon solving leads to

5184:  $$Q_{51}=

5185: %

5186: \frac{1}{5}\left(\begin{array}{l}(x^2+c_5)Q_0'+4xQ_0

5187: +Q_0^{\prime 2}

5188:    %

5189:    %

5190:    \\+10\int_x^{\iy}Q_0-

5191:    6\int_x^{\iy}dy\int_y^{\iy}du Q_0^{\prime\prime 2}

5192:     \end{array}\right)

5193:   =:\frac{1}{5} (\mathcal{F}_5+c_5Q'_0) ,$$

5194: with a constant $c_5$, which has been shown by Aminul

5195: Huq (private communication, 2008) to be $0$. Similarly one finds a differential

5196: equation for $Q_{62}$ and upon solving one finds, for some integration constant $c_6$,

5197: $$

5198: Q_{62}=

5199:  {\frac {1}{5}}  \left(

5200:  %

5201:  \mathcal{F}_5'+\frac{5}{18}(x^2Q_0''+13(x+c_6)Q_0' ) \right)

5202:  .$$

5203: 

5204:  \medbreak

5205: 

5206:  %\bigbreak

5207: 

5208:  Assembling all the pieces, one notices that

5209: two Taylor series in $Q_0$ and $Q_0'$ make their

5210: appearance in the $1/t$-expansion of $Q(t,x)$, leading

5211: to shifts in the argument of $Q_0(x)$ up to order $5$:

5212: %

5213: \be \label{Q}\ee

5214: \vspace*{-1.1cm}

5215: \bean

5216:  Q(t,x)&=&\sum_0^{\iy}\frac{Q_i(x)}{t^i}\\

5217:  &=&

5218:   \sum_0^{\iy} \left(\frac{r}{t}\right)^n\frac{Q_0^{(n)}(x)}{n!}

5219:  +\left(\frac{xr}{3t^3}+\frac{7r^2}{12t^4}\right)

5220:   \sum_0^{\iy}

5221:   \left(\frac{r}{t}\right)^n\frac{Q_0^{(1+n)}(x)}{n!}

5222:   \no\\&& + \frac{r}{5t^5}\mathcal{F}_5+O(\frac{1}{t^6})

5223:     \no\\

5224:    &=&

5225:    Q_0\Bigl(x+\frac{r}{t}\Bigr)+

5226:     \Bigl(\frac{xr}{3t^3}+\frac{7r^2}{12t^4}\Bigr)

5227:     Q_0'\Bigl(x+\frac{r}{t}\Bigr)+

5228:     \frac{r}{5t^5}\mathcal{F}_5+O(\frac{1}{t^6})

5229:     \no\\

5230:    % \eean\bea

5231:    %\\

5232:    &=&

5233:    Q_0\Bigl(x+\frac{r}{t}+\frac{xr}{3t^3}+\frac{7r^2}{12t^4}\Bigr)

5234:    + \frac{r}{5t^5}\mathcal{F}_5+c_5Q_0')+O(\frac{1}{t^6})

5235:   .\no\eean

5236: %

5237: %

5238: %

5239:  Exponentiating (\ref{Q}), remembering

5240: that $e^{Q_0(y)}=

5241:  \BP\left(\sup\mathcal {A}(t)\leq y\right)$,

5242: taking a derivative $\frac{d}{dy}\BP\left(\sup\mathcal{A}(t)\leq y\right)$ and setting

5243: $P_0(x):=\BP\left(\sup\mathcal {A}(t)\leq x\right)$ yields

5244: %

5245: %

5246: \bea

5247:  %\lefteqn

5248:  {\frac{d}{dx}

5249:  \BP(\sup\mathcal {A}^{(r)}(t)\leq x)}%\no\\

5250:  &=&\Bigl(1+\frac{r}{3t^3}\Bigr)

5251:  \frac{d}{dy}\BP\left(\sup\mathcal {A}(t)\leq

5252:  y\right)\Bigr|_{y=

5253:  (x+\frac{r}{t})(1+\frac{r}{3t^3})+

5254:     \frac{r^2}{4t^4}}+O(\frac{1}{t^5})

5255:   \no\\

5256:   &=&

5257:    P_0'  +\frac{r}{t}P_0'' +

5258:     \frac{r^2}{2t^2}P_0'''

5259:     +\frac{r}{6t^3}(r^2P_0^{(iv)}+(2xP_0')')

5260:    \no \\

5261:     &&

5262:     +\frac{r^2}{24 t^4}(r^2 P_0^{(v)}+14P_0''+8(xP_0'')')

5263:     +O(\frac{1}{t^5}).

5264:    \label{7.7} \eea

5265: For the moments with regard to the density $\frac{d}{dx}

5266:  \BP(\sup\mathcal {A}^{(r)}(t)\leq x)$,

5267: $$

5268: \mu_{\ell}^{(r)}(t)=\int_{-\iy}^\iy   x^\ell

5269: \frac{d}{dx}

5270:  \BP(\sup\mathcal {A}^{(r)}(t)\leq x) dx ,

5271:  $$

5272:   one reads off from (\ref{7.7}) the following expansion

5273:   for $\mu_{\ell}^{(r)}(t)$ in terms of the $t$-independent moments

5274:   $$

5275: \mu_{\ell}  =\int_{-\iy}^\iy   x^\ell \frac{d}{dx}

5276:  \BP(\sup\mathcal {A} (t)\leq x) dx =

5277:  \int_{-\iy}^\iy   x^\ell P_0^{\prime}(x) dx

5278:  ,

5279:  $$

5280:   namely,

5281: \bean \mu_{\ell}^{(r)}(t)&=&

5282:  \mu_{\ell}  -\ell\frac{r}{t}\mu_{\ell-1} +

5283:     \frac{r^2}{2t^2}\ell(\ell-1)\mu_{\ell-2}

5284:     +\frac{r}{6t^3}(-r^2\ell(\ell-1)(\ell-2)\mu_{\ell-3}-2\ell\mu_\ell)

5285:  \\&&   +\frac{r^2}{24 t^4}(r^2 \ell(\ell-1)(\ell-2)(\ell-3)

5286:  \mu_{\ell-4}+\ell(8\ell-14)\mu_{\ell-1})

5287:  +O(\frac{1}{t^5}).\eean

5288: In particular, the mean and second moment behave as

5289: $$

5290: \mu_1^{(r)}(t)={ \mu_1}-{\frac {r}{t}}-  {\frac {r{

5291: \mu_1}}{3{t}^{3}}}-  {\frac

5292: {{r}^{2}}{4{t}^{4}}}+O\Bigl(\frac{1}{t^5}\Bigr)

5293: $$

5294: and

5295: $$

5296: \mu_2^{(r)}(t)=

5297:  {  \mu_2}-  {\frac {2r{  \mu_1}}{t}}+{\frac

5298:  {{r}^{2}}{{t}^{2}}}-

5299:  {

5300: \frac {2r{  \mu_2}}{3{t}^{3}}}+  {\frac {{r}^{2}{

5301: \mu_1}}{6{t}^{4}}}+O\Bigl(\frac{1}{t^5}\Bigr).

5302: $$

5303: Hence the variance of the right edge of the process

5304:       behaves as

5305: %$$

5306: %var()=

5307: % ({  \mu_2}-{{  \mu_1}}^{2})\left(1

5308: %  -

5309: %  \frac{2r}{3t^{3}}\right)+O \left( \frac{1}{t^{5}}

5310: %  \right)

5311: %$$%

5312: %

5313: %The variance

5314:      \bean \mbox{var}(\sup \mathcal{A}^{(r)}(t))=%&=&

5315:      (\mu_2^{(r)}-\mu_1^{(r)2})(t)  &=&

5316:     ( \mu_2^{ }-\mu_1^{

5317:     2})\left(1-\frac{2r}{3t^3}\right)+O\Bigl(\frac{1}{t^5}\Bigr)\\&=&

5318:  \mbox{var}(\sup\mathcal {A}^{(0)}(t))\left(1-\frac{2r}{3t^3}\right)

5319:   +O\Bigl(\frac{1}{t^5}\Bigr),\eean

5320: %

5321: %

5322: ending the proof of Theorem~\ref{theo:one-time expansion}.

5323: \end{proof}

5324: 

5325: 

5326: 

5327:   \section{The $r$-Airy process, an interpolation between the Airy and Pearcey processes}\label{inter}

5328: 

5329: %  {\bf MARK: This will be section 8. But for the time being, I put %it here.}

5330: 

5331: 

5332: 

5333: 

5334:    Consider $n$ non-intersecting Brownian motions on ${\mathbb R}$, with

5335:  $0<p<1$ and $b<a$:

5336: %$$(x_1(t),\ldots,x_n(t))$$ in $\BR$ all leaving from the

5337: %origin,

5338: %

5339: %(with

5340: %$b_1<\ldots<b_p$ and

5341: %$\sum_{i=1}^p n_i=n$)

5342: %

5343: \bean   %\hspace*{-1.3cm}

5344: \BP^{(a,b)}_{n} \left(\begin{tabular}{c|c}

5345: & all $x_j(0) =0$\\

5346: all $x_j(t)\in E$ for $1\leq j\leq n$ &

5347:  $pn$ paths end up at $a$ at $t=1$\\

5348:   %& \\

5349:   &

5350: $(1-p)n$ paths end up at $b$ at $t=1$

5351: \end{tabular}\right)

5352: \eean

5353: %

5354: %

5355: %

5356: It is intuitive that, when $n\rg \iy$, the mean density of Brownian particles

5357: has its support on one interval for $t\sim  0$ and on

5358: two intervals for $t\sim  1$, so that a bifurcation

5359: appears for some intermediate time $t_0$, where one

5360: interval splits into two intervals. At this point the

5361: boundary of the support of the mean density in $(x,t)$-space has a cusp; see Figure 0.3.

5362: The Pearcey process describes this cloud of particles near the point of bifurcation, with time and space stretched in such a way that the outer particles appear infinitely far and such that the time horizon $t=1$ is at infinity. In \cite{AvM-asymmPearcey} it is shown that the same Pearcey process appears in the neighborhood of this cusp, independently of the target points $a$ and $b$, and the number $np$ of paths forced to $a$, showing ``universality" of the Pearcey process. It

5363: is convenient to introduce the parametrization of $p$,

5364:  \be

5365:   p=\frac{1}{1+q^3} \mbox{~with~}  0< q< \iy . %\mbox{~and let~}

5366:   %r:=\sqrt{q^2-q+1}.

5367:   \label{inter1}\ee

5368: Setting for simplicity $b=0$, one has the following:

5369: 

5370: 

5371:   \begin{proposition} \cite{AvM-asymmPearcey}

5372:   For $n\rg \iy$, the cloud of Brownian particles lies

5373: within a region, having a cusp at location

5374: $(x_0\sqrt{n},t_0)$, with

5375: \be

5376:   x_0 =\frac{(2q-1)a}{q+1}t_0%\frac{2a-b-3\frac{a-b}{q+1}  }

5377:  %{ 1+2 ( {q}^{2}-q+1 )  \left(\frac{ a-b}{q+1}\right) ^2

5378: %} = (2a-b-3\frac{a-b}{q+1})t_0

5379: ,\qquad\qquad

5380: t_0=\left({1+2a^2\frac{q^2-q+1}{(q+1)^2}}\right)^{-1}%\left(\frac{a-b}{q+1}\right)^2}.

5381: . \label{inter2}\ee

5382:   Moreover, the following Brownian motion probability tends to the

5383: probability for the Pearcey process $\mathcal{P}(t)$:

5384:  \bea

5385: \lefteqn{\lim_{n\rg\iy}\BP_{BR}^{(0,a\sqrt{n})}

5386:  \left(\mbox{all~}%\frac{1}{c_0}

5387: x_j \left(t_0+ (c_0\mu)^2\frac{2\tau}

5388: {n^{1/2}}\right)\in  {x_0}  n^{1/2}+c_0 A

5389: \tau+c_0\mu\frac{E^c}{n^{1/4}}\right)

5390:  }\no

5391:  \\& &\hspace*{9cm}

5392:   =\BP^\mathcal{P}\left(\mathcal{P}(\tau ) \cap E =

5393: \emptyset\right), \label{Pearcey limit}\eea

5394:  with constants expressed in terms of (\ref{inter1}) and (\ref{inter2}),

5395:  %

5396:  $$

5397: %

5398:    \mu:=\left(\frac{q^2-q+1

5399:  }{q}\right)^{1/4}>0,

5400: %

5401: \qquad 

5402: c_0:=\sqrt{\frac{ t_0(1-t_0)}{2}}>0,%.\eean

5403:  $$ $$

5404:  A := {q^{1/2}(1-\frac{x_0}{a})-q^{-1/2}\frac{x_0}{a}} .$$

5405: \end{proposition}

5406: 

5407: 

5408: 

5409: The $r$-Airy process is an interpolation between the Pearcey process and the Airy process, which can easily be described by looking at Figure 0.3: 

5410: 

5411: \begin{theorem}\label{Interpolation Theorem} When $p\rg 0$ and for $n$ very large, such that $pn$ equals a fixed integer $r>0$, the tip $(x_0\sqrt{n},t_0)$ of the cusp (as given by (8.2)) moves towards the right hand boundary of the picture, and, in particular, to the tangency point of the line through $(a\sqrt{n},1)$ tangent to the curve $y =\sqrt{2nt(1-t)}$:

5412: $$

5413:  (x_0\sqrt{n},t_0)\rg\left(\frac{2a}{1+2a^2}\sqrt{n},\frac{1}{1+2a^2}\right)\in \mbox{curve}\{y=\sqrt{2nt(1-t)}\}

5414: . $$

5415: Also the Pearcey process near the cusp tends to the r-Airy process in the neighborhood of the point of tangency above.

5416: \end{theorem}

5417: 

5418: 

5419: 

5420: %  The $r$-Airy process is an interpolation between the Pearcey process and the Airy process in the following sense:

5421: 

5422:  % \begin{theorem}\label{Interpolation Theorem}

5423:   % For $n$ very large and when the proportion $p$ of particles (forced to $a\sqrt{n}$) tends to $0$, then the probability on the left hand side of (\ref{Pearcey limit}) (leading in the limit to the Pearcey process),

5424: % with the target point located at $a\sqrt{n}$ and the cusp located at $(x_0\sqrt{n},t_0)$ for $(x_0,t_0)$ as in (\ref{inter2}),

5425: %    $$ (x_0,t_0)= \left( \frac{(2q-1)a\sqrt{n}}{q+1+2a^2\frac{q^2-q+1}{(q+1)}}, \frac{1}%{1+2a^2\frac{q^2-q+1}{(q+1)^2}}\right)

5426: %   $$

5427: %  tends to the $r$-Airy process occurring in the neighborhood of the point

5428:   %

5429: %  $$(x_0\sqrt{n},t_0)= \left( \frac{2a}{1+2a^2}\sqrt{n}, \frac{1}{1+2a^2}   \right)\in \mbox{curve}\left \{y\sqrt{n}=\sqrt{2ns(1-s)}\right\},$$

5430: %  with the point of phase transition $\rho_0\sqrt{n/2}$ (from Airy to $r$-Airy), mentioned in Theorem \ref{Theo1}, coinciding with the Pearcey target point $a\sqrt{n}$ on the axis $s=1$. Thus this corresponds to a phase transition from Pearcey to the $r$-Airy process.

5431: 

5432: 

5433: 

5434: 

5435: %  \end{theorem}

5436: 

5437: 

5438: 

5439: 

5440: 

5441: \begin{proof}

5442: Indeed, letting $p\rg 0$, or what is the same from (\ref{inter1}), letting $q\rg \iy$, one sees from formula (\ref{inter2}) that the cusp is located at the point

5443:  $$x_0\sqrt{n}= 2at_0\sqrt{n} \mbox{~~and~~} t_0=\frac{1}{1+2a^2}.$$

5444:   This implies that the point $(y,t)=(x_0\sqrt{n},t_0)$

5445:   belongs to the curve $y=\sqrt{2nt(1-t)}$ and that

5446:    $$a\sqrt{n}=\sqrt{\frac{1-t_0}{2t_0}}\sqrt{n}=\rho_0\sqrt{\frac{n}{2}},$$

5447:    establishing the first part of Theorem \ref{Interpolation Theorem}. That the Pearcey process tends to the $r$-Airy process will be done elsewhere.  \end{proof}

5448: 

5449: 

5450: 

5451: 

5452: 

5453: %\section{Appendix}

5454: 

5455: \section{Appendix}

5456: 

5457: 

5458: 

5459: %\newpage

5460: %\cite{}

5461: 

5462: 

5463: 

5464: 

5465: 

5466: 

5467: %%      ---------------------------------------------------------------------

5468: %%      ---------------------------- BODY OF PAPER --------------------------

5469: %%      ---------------------------------------------------------------------

5470: 

5471: %%      Please input or insert the body of your paper here.

5472: 

5473: 

5474: 

5475: 

5476: %%      ---------------------------------------------------------------------

5477: %%      ------------------------- APPENDIX (OPTIONAL) -----------------------

5478: %%      ---------------------------------------------------------------------

5479: 

5480: %%      If you have one appendix, uncomment the line \appendix and add

5481: %%      a \section{ *** APPENDIX TITLE ***}. If you have more than

5482: %%      one, uncomment the line \appendices and add a \section{ ***

5483: %%      APPENDIX TITLE ***} command for each appendix title.

5484: 

5485: %\appendix

5486: 

5487: %Here we derive the identities using functional analysis:

5488: 

5489: 

5490: %\newpage

5491: %$$

5492: %(I-K)=(I-K_0)(I+(I-K_0)^{-1}K_1+(I-K_0)^{-1}K_2+\ldots$$

5493: 

5494: %Using $\log(1-z)=-z-\frac{z^2}{2}-\frac{z^3}{3}$, one

5495: %finds

5496: 

5497: The purpose of this appendix is to show that the first

5498: few $Q_i$ of Theorem \ref{theo:one-time expansion} can

5499: be obtained, {\em with tears}, by functional analytical

5500: methods, in the style of Widom \cite{Widom}. The proof

5501: requires many intricate identities involving the kernels

5502: $K_i^{(r)}$, some of which can be found in Tracy-Widom

5503: \cite{TW-Airy}. This section should convince the reader

5504: of the usefulness of the PDE's in computing the

5505: asymptotics for $t\rg -\iy$.

5506: 

5507: Remember the $L_i =(I+R_x)K_i^{(r)}$ from (\ref{L}),

5508: where we now indicate the explicit dependency of the

5509: resolvent $R_x=K^{(0)}(I-K^{(0)})^{-1}$ on $x$, since

5510: all operators act on $L^2(x,\iy)$. Then $Q(t,x)$ has an

5511: expansion, with $Q_i$'s given in (\ref{Q's}),

5512: $$

5513: Q=\Tr\log(I-K_t^{(r)})=\sum^{\iy}_{i=0}\frac{Q_i}{t^i}=

5514:  \Tr\log(I-K^{(0)})+\sum^{\iy}_{i=1}\frac{Q_i}{t^i}

5515: .$$

5516: %

5517: Throughout this section, we shall be using the

5518: inner-product $$\la f,g\ra:= \int_{\mathbb R}

5519: \raisebox{1mm}{$\chi$}{}_{(x,\iy)}(u)f(u)g(u)du .$$

5520: 

5521: %\newpage

5522: 

5523: \begin{proposition}\label{Theo:8.1}

5524: 

5525: \bean Q_1=rQ'_0 ,~~ Q_2=\frac{r^2}{2!}Q''_0

5526:  ,~~

5527:  Q_3=\frac{r^3}{3!}Q'''_0+\frac{r}{3 } xQ'_0.

5528: \eean

5529: 

5530: 

5531: 

5532: 

5533: \end{proposition}

5534: 

5535: 

5536: 

5537: %\newpage

5538: 

5539: \begin{lemma}\label{lemma:8.1}

5540: $$

5541: \Tr L_1^n=(-r)^n\la (I+R_x)A ,A \ra^n.

5542: $$

5543: \end{lemma}

5544: 

5545: \begin{proof} Indeed,

5546: %

5547: \bean

5548: \lefteqn{\Tr((I+R_x)K_1)^n}\\

5549: &=&(-r)^n\int_{(x,\iy)^n}du_1\ldots

5550: du_n\left(\left((I+R_x) A(u_1)\right)A(u_2)\right)

5551: \left(\left((I+R_x)A(u_2)\right)A(u_3)\right)%\\

5552: \\

5553: & &\qquad\qquad\qquad \qquad\qquad\qquad\ldots

5554: \left(\left((I+R_x)A(u_n)\right)

5555: A(u_1)\right) %\\

5556: \\

5557: &=&(-r)^n\la (I+R_x)A(u),A(u)\ra^n .\eean\end{proof}

5558: 

5559: The

5560: identities in the following Lemma can be found in or

5561: deduced from Tracy-Widom \cite{TW-Airy}.

5562: 

5563: \begin{lemma} \label{TW}

5564: \bea

5565: R_x(x,x)&=&\la(I+R_x)A,A\ra \label{TW1}\\

5566: %\no\\

5567: \left(\frac{\pl}{\pl x}\!+\!\frac{\pl}{\pl

5568: u}\!+\!\frac{\pl}{\pl v}\right)R_x(u,v)&=&-((I+R_x)A(u))

5569: ((I+R_x)A(v))\label{TW2}%\\

5570: \\

5571: %\eea\bea%\no\\

5572: \left(\frac{\pl}{\pl x}+\frac{\pl}{\pl

5573: u}\right)(I+R_x)A(u)&=&(I+R_x)A'(u)-(I+R_x)A(u)%\\

5574: %\\

5575: %& &\qquad\qquad\qquad

5576:  \la (I+R_x)A,A\ra .%\no\\

5577:  \label{TW3}\eea

5578: %

5579: %(TW), (2.18)

5580: %

5581: \bea 2\la (I+R_x)A',A\ra -\la

5582: (I+R_x)A,A\ra^2&=&-((I+R_x)A(x))^2 \label{TW4}%\ee

5583: %

5584: %\be

5585: \\ 2\la (I+R_x)A,A''\ra -\la

5586: (I+R_x)A',A'\ra&=&x\la(I+R_x)A,A\ra \label{TW4'}%\ee

5587: \\%

5588: %\be

5589:   \frac{d}{dx}(I+R_x)A'(u)&=&-R_x(u,x)(I+R_x)A'(x)

5590: \label{TW5}\eea

5591: %

5592: \bea \label{TW6}&&\frac{\pl}{\pl u}(I+R_x)A'(u)=u(I+R_x)A(u)-2\la

5593: (I+R_x)A',A\ra

5594: (I+R_x)A(u) \\

5595: %\nonumber\\

5596: & &\qquad\qquad +\la (I+R_x)A,A\ra

5597: (I+R_x)A'(u)+R(u,x)(I+R_x)A'(x).\no%\\

5598: %\nonumber\\

5599: \eea

5600: 

5601: 

5602: \end{lemma}

5603: %\newpage

5604: %

5605: \begin{lemma} \label{Q'}We have

5606: %

5607: \bean

5608: Q'_0(x)&=&\la (I+R_x)A,A\ra\qquad \mbox{   and   } \qquad

5609:  %

5610: Q''_0(x)=-((I+R_x)A(x))^2\\

5611: \\

5612: \frac{1}{2}Q'''_0(x)&=&-((I+R_x)A(x))((I+R_x)A'(x))

5613: +((I+R_x)A(x))^2\la

5614: (I+R_x)A,A\ra .\\

5615: \eean

5616: \end{lemma}

5617: %

5618: \begin{proof} One computes

5619: %

5620: \bean Q'_0(x)&=&%\frac{\pl}{\pl x}\log\det(I-K_0

5621: %\raisebox{1mm}{$\chi$}{}_{(x,\iy)} ) \\

5622: %\\

5623: %&=&

5624: \frac{\pl}{\pl x}\Tr\log(I-K^{(0)}\raisebox{1mm}{$\chi$}{}_{(x,\iy)} )\\

5625: \\

5626: &=&-\frac{\pl}{\pl x}\Tr\left(K^{(0)}(u,v)+

5627: \frac{1}{2}\int^{\iy}_xK^{(0)}(u,w)K^{(0)}(w,v)dw+\ldots\right)\\

5628: \\

5629: &=&-\frac{\pl}{\pl x}\left(\begin{array}{l}

5630: \displaystyle{\int_x^{\iy}K^{(0)}(u,u)du+

5631: \frac{1}{2}\int^{\iy}_x\int^{\iy}_x}K^{(0)}(u,w)K^{(0)}(

5632: w,u)dudw\\

5633: \\

5634: \!\!+\displaystyle{\frac{1}{3}\int^{\iy}_x\!\!\!\int^{\iy}_x\!\!\!\int^{\iy}_x}K_0(u,w_1)K_0(w_1,w_2)

5635: K^{(0)}(w_2,u)dudw_1dw_2+\ldots

5636: \end{array}

5637: \right)%\\

5638: %\\

5639: \\%\eean\bean

5640: &=&K^{(0)}(x,x)+\int_x^{\iy}K^{(0)}(x,v)K^{(0)}(v,x)dv%\\

5641: \\

5642: & &\qquad

5643: +\int_x^{\iy}\!\!\!\int_x^{\iy}K^{(0)}(x,w_1)K^{(0)}

5644: (w_1,w_2)K^{(0)}(w_2,x)dw_1dw_2+\ldots\\

5645: \\

5646: &=&R_x(x,x)

5647: =\la (I+R_x)A,A\ra,\mbox{~using (\ref{TW1})} \eean by

5648: the Neumann series for $R_x=(I-K^{(0)})^{-1}-I$.

5649: %

5650:  Moreover, using the previous result,%again using Lemma \ref{TW},

5651: %

5652: %Note that (TW)

5653: %$$

5654: %R_x(x,x)=\la A,(I-K_0)^{-1}A\ra

5655: %$$

5656: %

5657: %\newpage

5658: %

5659: \bean Q''_0(x)=\frac{d}{dx}R_x(x,x)&=&

5660: \left(\frac{\pl}{\pl x}+\frac{\pl}{\pl u}+\frac{\pl}{\pl

5661: v} \right)R_x(u,v) \Bigl|_{u=v=x}%\eean

5662: %\bean%\\

5663:  \\

5664: &=&-((I+R_x)A(u))(I+R_x)A(v))\Bigl|_{u=v=x}

5665: ~~ \mbox{using (\ref{TW2})}\\

5666: %\\

5667: &=&-((I+R_x)A(x))^2, \eean

5668:  %

5669:  and again using the result just obtained and using identity

5670:  (\ref{TW3}),

5671:  %

5672:  %

5673: \bean

5674: Q'''_0(x)&=&-2((I+R_x)A(x))\frac{d}{dx}(I+R_x)A(x)\\

5675: \\

5676: &=&-2((I+R_x)A(x))\left(\frac{\pl}{\pl x}+\frac{\pl}{\pl

5677: u}\right)

5678: (I+R_x)A(u)\Bigl|_{u=x}\\

5679: \\

5680: &=&-2((I+R_x)A(x))(I+R_x)A'(x)%\\

5681: %\\

5682: %& &\qquad\qquad\qquad

5683: +2((I+R_x)A(x))^2\la (I+R_x)A,A\ra

5684: ,\eean

5685:  proving Lemma \ref{Q'}.\end{proof}

5686: 

5687: 

5688: 

5689: \begin{lemma}\label{Lemma:8.5}

5690: 

5691: \bean Q_1(x)&=&r\la (I+R_x)A,A\ra\\

5692: %&=&rR(x,x),%=rF'_0(x)

5693:  Q_2(x)&=&

5694:   -\frac{r^2}{2}( (I+R_x)A(x))^2

5695: \\

5696:  Q_3(x)&=& \frac{r^3}{3}(((I+R_x)A(x))^2\la

5697: (I+R_x)A,A\ra

5698: -(I+R_x)A(x)(I+R_x)A'(x))\\&&+\frac{rx}{3}\la

5699: (I+R_x)A,A\ra. \eean

5700: 

5701: \end{lemma}

5702: 

5703: 

5704: \begin{proof} Indeed, by (\ref{Q's}) and (\ref{Ki's}),

5705: %

5706: \bea \label{Q1}

5707: Q_1(x)=-\Tr L_1%\no\\%&=&-\Tr(I-K_0)^{-1}K_1\no\\

5708: %\\

5709: %&=&-\Tr(I+R)K_1\\

5710: %\\

5711: &=&-\Tr K^{(r)}_1-\Tr R_xK^{(r)}_1  \\

5712: %\\

5713: %&=&r\int_x^{\iy}A^2(u)du+r\int_x^{\iy}\int_x^{\iy}dudv ~R(u,v)A(u)A(v)\\

5714: %\\

5715: &=&r\int_x^{\iy}du~A(u)\left(A(u)+

5716: \int_x^{\iy}R_x(u,v)A(v)dv\right)\no\\

5717: \no\\

5718: &=&r\int_x^{\iy}A(u)((I+R)A)(u)du%\\

5719: %\\

5720: %&=&r\int_x^{\iy}A(u)((I-K_0)^{-1}A)(u)du\\

5721: %\\

5722: %&=&r\int_x^{\iy}A(u)Q_x(u)du\\

5723: %\\

5724: %&=&r\la A\raisebox{1mm}{$\chi$}{}_{[x,\iy)} ,(I-K_0)^{-1}A\ra \\

5725: %\no\\

5726: %&=&

5727: =r\la (I+R_x)A,A\ra.\no%\\

5728: %&=&rR(x,x),%=rF'_0(x)

5729: \eea

5730: % using Lemma \ref{TW} again in the last

5731: %identity.

5732: %

5733: %\newpage

5734: %Notice the trace of the product of a symmetric with a

5735: %skew-symmetric operator vanishes

5736: %

5737: %$$

5738: %F_1(x)=-\Tr(I-K_0)^{-1}K_1=-r\la (I-K_0)^{-1}A,A\ra

5739: %$$

5740: %

5741: %\newpage

5742: %\bean

5743: %K_2^{(r)}(u,v)&=&-\frac{r^2}{2}(A'(u)A(v)+A(u)A'(v))

5744: %+\frac{r}{2}(u-v)K_0(u,v) ,\eean

5745: %

5746:  %

5747:  %

5748: %

5749: Computing $Q_2$ by (\ref{Q's}) involves $\Tr L_2$ and

5750: $\Tr L_1^2$. Since $K_2^{(r)}(u,v)$ has a symmetric and

5751: skew-symmetric part, and since $ I+R_x $ is symmetric,

5752: and remembering the form of $K_2^{(r)}(u,v)$ in

5753: (\ref{Ki's}), we have (by symmetry) and the fact that a

5754: symmetric times a skew-symmetric operator is traceless,

5755: %

5756: %

5757:  \bean

5758:  \Tr L_2=\Tr(I+R_x)K^{(r)}_2&=&

5759:  -\frac{r^2}{2}\Tr(I+R_x)(A'(u)A(v)+A(u)A'(v))%\\

5760:  \\

5761:  &=&-r^2\la (I+R_x)A',A\ra .\eean

5762: %

5763: %\newpage

5764: %

5765: Hence, combining this result with the computation of

5766: $\Tr L_1^2$ in Lemma \ref{lemma:8.1}, one finds, using

5767: (\ref{TW4}),

5768: %

5769: \bea \label{Q2}

5770: Q_2=-\Tr(L_2+\frac{L_1^2}{2})%=-\Tr\left((I-K_0)^{-1}K_2+\frac{1}{2}((I-K_0)^{-1}K_1)^2\right)

5771: %\\

5772: %

5773: %&=&-\Tr\left((I+R_x)K^{(r)}_2+\frac{1}{2}((I+R_x)K^{(r)}_1)^2\right)\\

5774: %\\

5775: %&=&

5776: % -\frac{r^2}{2}\Tr(I+R_x)(A'(u)A(v)+A(u)A'(v))

5777: % -\frac{r^2}{2}

5778: %\la (I+R_x)A,A\ra ^2

5779: %\\

5780: &=&\frac{r^2}{2}(2\la (I+R_x)A',A\ra -\la (I+R_x)A,A\ra ^2)

5781: %\no\\

5782:  \\

5783: &=&-\frac{r^2}{2}( (I+R_x)A(x))^2%=+

5784: %\frac{r^2}{2}F''_0(x),\mbox{~using Lemma}

5785: .\no\eea

5786: %

5787: %\newpage

5788: %

5789: The computation of $Q_3$, by (\ref{Q's}), involves $\Tr

5790: L_3$,~$\Tr L_1L_2$ and $\Tr L_1^3$. Using again the fact

5791: that a symmetric times a skew-symmetric operator is

5792: traceless, one reads off from the form of $K_3^{(r)}$

5793: (see (\ref{Ki's})) the following:

5794: %

5795: %Using the fact that the coefficient of $k^2$ is a

5796: %skew-symmetric operator and using the symmetry of

5797: %$(I-K_0)^{-1}$

5798: %

5799: %{\footnotesize

5800: \bean \Tr L_3 ={\Tr(I+R_x)K^{(r)}_3}

5801: &=&-\frac{r^3}{3}(\la(I+R_x)A,A''\ra +\la (I+R_x)A',A'\ra )%\\

5802: \\

5803: &&\qquad\qquad -\frac{r}{3}(2\la (I+R_x)A,A''\ra -\la

5804: (I+R_x)A',A'\ra )

5805:  \eean%}

5806: %

5807: %

5808: %\end{document}

5809: %

5810: and using

5811: \vspace*{-.7cm}

5812:  { \bean

5813:  {(I+R_x)K^{(r)}_2}

5814: &=&-\frac{r^2}{2}\left(((I+R_x)A'(u))A(v)

5815: +((I+R_x)A(u))A'(v)\right )\\

5816: \\

5817: &&\qquad\qquad -\frac{r}{2}\left(((I+R_x)A'(u))A(v)-

5818: ((I+R_x)A(u))A'(v)\right ), \eean}

5819: % \vspace*{-.7cm}

5820: %

5821:  one computes

5822: %

5823: %

5824: %\vspace*{-.17cm}

5825: %

5826: {%\footnotesize 

5827:  \bean

5828: %\lefteqn{\Tr L_1L_2}\\

5829: \Tr L_1L_2 &=& \Tr(I+R_x)K_1^{(r)}(I+R_x)K_2^{(r)}\\

5830: &=&\frac{r^3}{2}\int\!\!\!\int_{(x,\iy)^2}((I+R_x)A)(w_1)A(w_2)

5831: ((I+R_x)A')

5832: (w_2)A(w_1)dw_1dw_2\\

5833: \\

5834: & &+\frac{r^3}{2}\int\!\!\!

5835: \int((I+R_x)A)(w_1)A(w_2)((I+R_x)A)(w_2)A'(w_1)dw_1dw_2

5836: \\

5837: \\

5838: & &+\frac{r^2}{2}\int\!\!\int((I+R_x)A)(w_1)A(w_2)

5839: ((I+R_x)A')(w_2)A(w_1)dw_1dw_2\\

5840: \\

5841: & &-\frac{r^2}{2}\int\!\!\!\int((I+R_x)A)(w_1)A(w_2)

5842:  ((I+R_x)A)(w_2)A'(w_1)dw_1dw_2%\\

5843:  \\

5844: %\eean\bean\\

5845: &=&\frac{r^3}{2}\left(\begin{array}{c}\la

5846: (I+R_x)A,A\ra\la

5847: (I+R_x)A',A\ra \\

5848: \\

5849: +\la (I+R_x)A,A\ra\la A',(I+R_x)A\ra\end{array}\right)\\

5850: \\

5851: & &+\frac{r^2}{2}\left(\begin{array}{c}\la

5852: (I+R_x)A,A\ra\la

5853: (I+R_x)A',A\ra \\

5854: \\

5855: -\la (I+R_x)A,A\ra\la A',(I+R_x)A\ra\end{array}\right)\\

5856: \\

5857: &=&r^3\la (I+R_x)A,A\ra \la (I+R_x)A',A\ra .\eean}

5858: %

5859: %\newpage

5860: %

5861: Putting the pieces together and using (\ref{Q's}) and

5862: using Lemmas \ref{lemma:8.1} and \ref{Q'}, one obtains

5863: %

5864: {%\footnotesize 

5865:  \bea \label{Q3}

5866:  \qquad Q_3&=&-\Tr \Bigl(L_3+ {L_1L_2}

5867: +\frac{1}{3}L_1^3\Bigr) \\

5868:  &=&\frac{r^3}{3}\left(\begin{array}{c}\la

5869: (I+R_x)A,A''\ra +\la

5870: (I+R_x)A',A'\ra  \no\\

5871: -3\la (I+R_x)A,A\ra \la (I+R_x)A',A\ra\no\\

5872: +\la (I+R_x)A,A\ra ^3\end{array}\right)

5873:  \\

5874:  &&+\frac{r}{3}(2\la (I+R_x)A,A''\ra -

5875:  \la (I+R_x)A',A'\ra )\no\\

5876: \no\\

5877: &=&\frac{r^3}{3}\left(\begin{array}{l}\la

5878: (I+R_x)A,A''\ra +\la

5879: (I+R_x)A',A' \ra  \no\\

5880: %&&

5881: -\la (I+R_x)A,A\ra\la (I+R_x)A',A\ra %\\

5882: \no\\

5883: +\la (I+R_x)A,A\ra \Bigl(\la  (I+R_x) A,A\ra^2-2\la

5884: (I+R_x)A',A\ra \Bigr)

5885: \end{array}\right)\no\\

5886: \no\

5887: & &+\frac{r}{3}(2\la (I+R_x)A,A''\ra -\la (I+R_x)A',A'\ra )\no\\

5888: \no\\

5889: &{=}&\frac{r^3}{3}(-(I+R_x)A(x)(I+R_x)A'(x)+

5890: ((I+R)A(x))^2\la

5891: (I+R_x)A,A\ra )%\no\\

5892:  \no\\& &

5893: +\frac{rx}{3}\la (I+R_x)A,A\ra

5894:  ,%,~~\mbox{}.

5895: %\\

5896: %\\

5897: %&=&\frac{r^3}{6}Q_0'''(x)+\frac{rx}{3}Q'_0(x)

5898: % \no \\

5899: \no\eea}

5900: %

5901: %\newpage

5902: %

5903: using in the last equality (\ref{TW4}), (\ref{TW4'})

5904: combined with

5905:  (\ref{Identity3}) below. Then, using the differential equation $uA(u)=A''(u)$, one

5906: checks:

5907: %

5908: \be  \label{Identity3}\ee

5909: \vspace{-.8cm}

5910: \bean

5911: \lefteqn{\la (I+R_x)A',A'\ra} \\

5912: &=&-A(x)(I+R_x)A'(x)-\la\frac{\pl}{\pl u}(I+R_x)A',A\ra

5913: \mbox{~~(by integration by parts)}\no\\

5914: %\no\\

5915: %\eea \bea

5916: &=&-A(x)(I+R_x)A'(x)-\la (I+R_x)A(u),uA(u)\ra

5917:  %\no\\&&

5918:  +\la (I+R_x)A',A\ra \la (I+R_x)A,A\ra\no\\

5919: %

5920: & & -\la R(u,x),A(u)\ra

5921:  (I+R)A'(x)

5922: ,~~~\mbox{using (\ref{TW6}),} \no

5923: \\

5924:  &=&-(I+R_x)A(x)(I+R_x)A'(x)-\la (I+R_x)A,A''\ra +\la (I+R_x)A',A\ra \la (I+R_x)A,A\ra.

5925: \no\eean

5926: %

5927: %using

5928: %

5929: % \bea

5930: %&&\frac{\pl}{\pl u}(I+R_x)A'(u)=u(I+R_x)A(u)-2\la

5931: %(I+R_x)A',A\ra

5932: %(I+R_x)A(u)\nonumber\\

5933: %\nonumber\\

5934: %& &\qquad\qquad +\la (I+R_x)A,A\ra (I+R_x)A'(u)

5935: %+R(u,x)(I+R_x)A'(x) \label{8.7}

5936: %&\bullet&uA(u)=A''(u)

5937: %\eea

5938: \end{proof}

5939: 

5940: \begin{proof}[Proof of Proposition~\ref{Theo:8.1}]

5941:  The formulae follow immediately from comparing the

5942:   formulae of Lemmas~\ref{Q'} and~\ref{Lemma:8.5}.

5943: \end{proof}

5944: 

5945: 

5946: 

5947: 

5948: 

5949: 

5950: 

5951: 

5952: 

5953: 

5954: 

5955: 

5956: 

5957: 

5958: %\appendices

5959: %\section{}

5960: 

5961: %%      Type body of appendix/-ices here.

5962: 

5963: 

5964: %%      ---------------------------------------------------------------------

5965: %%      ---------------------------ACKNOWLEDGMENTS (OPTIONAL) ---------------

5966: %%      ---------------------------------------------------------------------

5967: 

5968: %% ***** UNCOMMENT THE FOLLOWING LINE TO ADD ACKNOWLEDGMENTS.

5969: 

5970:   \ack

5971:   %

5972:   Pierre van Moerbeke thanks G\'erard Ben Arous for a

5973: useful conversation (May 2006) and thanks Patrik Ferrari for a very interesting discussion (January 2008) concerning Theorem \ref{Theo1}. Mark Adler and Pierre van Moerbeke gratefully acknowledge the support of a National Science Foundation grant \#~DMS-07-04271. This work was partially done while PvM was a member of the Miller Institute for Research in Science, Berkeley, California. The support of National Science Foundation grant \#~DMS-07-0427, a European Science Foundation grant (MISGAM), a Marie Curie Grant (ENIGMA), a FNRS grant and a "Interuniversity Attraction Pole" (Belgium) grants are gratefully acknowledged.

5974: 

5975: 

5976: %%      Type acknowledgments here.

5977: 

5978: 

5979: %%      ---------------------------------------------------------------------

5980: %%      --------------------------- BIBLIOGRAPHY ----------------------------

5981: %%      ---------------------------------------------------------------------

5982: 

5983: \frenchspacing

5984: \bibliographystyle{plain}

5985: \begin{thebibliography}{99}

5986: 

5987: 

5988: %\bibitem{AvM0}%AvM0

5989: %. Adler, T. Shiota and P. van Moerbeke: {\em Random

5990: %matrices, vertex operators and the Virasoro algebra},

5991: %Phys. Lett. {\bf A 208}, 67-78, (1995).

5992: 

5993: %\bibitem{AvM-coupled}AvM-coupled

5994: %M. Adler and P. van Moerbeke: {\sl  The spectrum of

5995: %coupled random matrices}, Annals of Math., {\bf 149},

5996: %921--976  (1999).

5997: 

5998:  \bibitem{AvM-asymmPearcey}%AvM-Pearcey-joint

5999:  M. Adler, N. Orantin and P. van Moerbeke: {\em Universality for the

6000:  Pearcey process and a PDE for the transition probabilities}, Preprint 2008. %Comm. Pure Appl Math. (2007)

6001:   %(arXiv:math/0612393).

6002: 

6003: \bibitem{AvM-PDE-GUE}%AvM-PDE-GUE

6004:   ~M. Adler and P. van Moerbeke:

6005:   {\sl  Hermitian, symmetric and symplectic random ensembles: PDE's

6006:   for the distribution of the spectrum}, Annals of Math., {\bf

6007:   153}, 149--189 (2001).

6008: 

6009: % \bibitem{AvM-Airy} AvM-Airy M. Adler and P. van Moerbeke:

6010: % {\sl  A PDE for the joint distribution of the Airy

6011: % process}, arXiv:math.PR/0302329 (2003).

6012: 

6013: 

6014: \bibitem{AvM-Airy-Sine}%AvM-Airy-Sine

6015: M. Adler and P. van Moerbeke: {\sl  PDE's for the joint

6016: distributions of the Dyson, Airy and Sine processes},

6017: The Annals of Probability, {\bf 33}, 1326-1361 (2005).

6018: (arXiv:math.PR/0302329 and math.PR/0403504)

6019: 

6020: 

6021: \bibitem{AvM-Pearcey}%AvM-Pearcey~

6022: M. Adler and P. van Moerbeke: {\sl PDE's for the

6023: Gaussian ensemble with external source and

6024:   the Pearcey distribution}, Comm. Pure and Appl. Math,

6025:   {\bf 60} 1-32 (2007) (arXiv:math.PR/0509047)

6026: 

6027: 

6028:   \bibitem{AvM-Pearcey-joint}%AvM-Pearcey-joint

6029:  M. Adler and P. van Moerbeke: {\em Joint probability for the

6030:  Pearcey process}, %Comm. Pure Appl Math. (2007)

6031:   (arXiv:math/0612393).

6032: 

6033: 

6034:  

6035: 

6036: 

6037: %\bibitem{AvM2} M. Adler and P. van Moerbeke:

6038: %{\sl  A PDE for the joint distribution of the Airy

6039: %process}, arXiv:math.PR/0302329 (2003).

6040: 

6041: \bibitem{AvM-Mops}%AvM-Mops

6042:  M. Adler, P. van Moerbeke and P. Vanhaecke:

6043: {\sl  Moment matrices and multicomponent KP, with

6044: applications to random matrix theory}, (2006)

6045: (arXiv:math-ph/0612064)

6046: 

6047: %\bibitem{Ambjorn} J. Ambjorn and Yu. Makeenko:

6048: %{\sl Properties of loop equations for the Hermitian

6049: %matrix model and for two-dimensional quantum gravity},

6050: %Modern Physics letters {\bf A 5} 1753--1763 (1990)

6051: 

6052: 

6053: 

6054: \bibitem{AptBleKui} %AptBleKui

6055: A. Aptekarev, P. Bleher and A. Kuijlaars: {\em Large $n$

6056: limit of Gaussian random matrices with external source}.

6057: II. {Comm. Math. Phys.}  {\bf 259} 367--389 (2005)

6058: (arXiv: math-ph/0408041)

6059: 

6060: 

6061: \bibitem{BBP}%BBP

6062: Jinho Baik, G\'erard Ben Arous, Sandrine P\'ech\'e: {\em

6063: Phase transition of the largest eigenvalue for non-null

6064: complex sample covariance matrices}, Ann. Probab.  {\bf

6065: 33}, no. 5, 1643--1697 (2005) (arXiv:math/0403022)

6066: 

6067: %Baik, Jinho; Ben Arous, Gérard; Péché, Sandrine Phase

6068: %transition of the largest eigenvalue for nonnull complex

6069: %sample covariance matrices.  Ann. Probab.  {\bf 33}, no.

6070: %5, 1643--1697 (2005)

6071: 

6072: 

6073: \bibitem{Baik}%Baik

6074:  Jinho Baik: {\em Painlev\'e formulas of the limiting distributions for

6075: non-null complex sample covariance matrices}, Duke Math.

6076: J.  {\bf 133}, no. 2, 205--235 (2006)

6077: (arXiv:math/0504606)

6078: 

6079: %Baik, Jinho Painlevé formulas of the limiting

6080: %distributions for nonnull complex sample covariance

6081: %matrices.  Duke Math. J.  {\bf 133}, no. 2, 205--235

6082: %(2006)

6083: 

6084: 

6085: %\bibitem{Van Assche}%Van Assche

6086: %A. Aptekarev, A. Branquinho

6087: % and W. Van Assche: {\sl Multiple orthogonal polynomials for

6088: % classical weights}, Trans. Amer. Math. Soc., {\bf 355}, 3887--3914 (2003)

6089: 

6090: 

6091: \bibitem{BleKui1}%BleKui1

6092: P. Bleher and A. Kuijlaars: {\sl  Random matrices with

6093: external source and multiple orthogonal polynomials},

6094: Internat. Math. Research Notices {\bf 3}, 109--129

6095: (2004) (arXiv:math-ph/0307055).

6096: 

6097: 

6098: %\bibitem{BleKui2} %BleKui2

6099: %P. Bleher and A. Kuijlaars: {\sl  Large $n$ limit of

6100: %Gaussian random matrices with external source, Part I},

6101: %Comm. Math. Phys., {\bf 252}, 43--76  (2004).

6102: 

6103: 

6104: 

6105: %\bibitem{BleKui3}%BleKui3

6106: %P. Bleher and A. Kuijlaars: {\sl  Large $n$ limit of

6107: %Gaussian random matrices with external source, Part II},

6108: %Comm. Math. Phys., {\bf 252}, 43--76  (2004).

6109: 

6110: 

6111: 

6112: %\bibitem{Brezin1}Brezin1 E. Br\'ezin and S. Hikami:

6113: %{\sl  Spectral form factor in a random matrix theory},

6114: %Phys. Rev., {\bf E 55}, 4067--4083 (1997).

6115: 

6116: 

6117: \bibitem{Brezin2} %Brezin2

6118: E. Br\'ezin and S. Hikami: {\sl  Correlations of nearby

6119: levels induced by a random potential}, Nuclear Physics

6120: {\bf B 479}, 697--706 (1996).

6121: 

6122: 

6123: 

6124: \bibitem{Brezin3} %Brezin3

6125: E. Br\'ezin and S. Hikami: {\sl  Extension of level

6126: spacing universality}, Phys. Rev., {\bf E 56}, 264--269

6127: (1997).

6128: 

6129: 

6130: 

6131: \bibitem{Brezin4} %Brezin4

6132: E. Br\'ezin and S. Hikami: {\sl  Universal singularity

6133: at the closure of a gap in a random matrix theory},

6134: Phys. Rev., {\bf E 57}, 4140--4149 (1998).

6135: 

6136: 

6137: \bibitem{Brezin5} %Brezin5

6138: E. Br\'ezin and S. Hikami: {\sl  Level spacing of random

6139: matrices in an external source}, Phys. Rev., {\bf E 58},

6140: 7176--7185 (1998).

6141: 

6142:  \bibitem{Deift}

6143:   P. Deift, T. Kriecherbauer, K. T-R McLaughlin,

6144:  S. Venakides and X. Zhou: {\sl  Strong asymptotics of

6145:  orthogonal polynomials with respect to exponential

6146:  weights}, Comm. of Pure and Appl. Math. {\bf 52},

6147:  1491--1552 (1999)

6148: 

6149: 

6150:  \bibitem{Dyson} %Dyson

6151:  F.J. Dyson: {\sl  A Brownian-Motion Model for the

6152:  Eigenvalues of a Random Matrix}, Journal of Math. Phys.

6153:  {\bf 3}, 1191--1198 (1962)

6154: 

6155: %\bibitem{Bateman} A. Erdelyi: {Higher transcendental

6156: %functions, Vol. II}, McGraw-Hill (1953)

6157: 

6158: 

6159: 

6160: %\bibitem{Forrester} P.J. Forrester, T. Nagao and G. Honner:

6161: %{\sl   Correlations for the orthogonal-unitary and

6162: %symplectic-unitary transitions at the hard and soft

6163: %edges},

6164: % Nucl. Phys. B, {\bf 553}, 601--643 (1999).

6165: % %ArXiv. Math. 9811142 (1998).

6166: 

6167: 

6168:  \bibitem{Grabiner} %Grabiner

6169:  D.J. Grabiner: {\sl  Brownian-Motion in a Weyl chamber,

6170:  non-colliding particles and random matrices}, Ann.

6171:  Institut H. Poincar\'e {\bf 35}, 177--204 (1999)

6172: 

6173: \bibitem{HM} %HM

6174: S. P. Hastings and J. B. McLeod: {\sl  A boundary value

6175: problem associated with the second Painlev\'e

6176: transcendent and Korteweg-de Vries equation}, Arch.

6177: Rational Mech. Anal. {\bf 73}, 31--51 (1980)

6178: 

6179: %\bibitem{JMMS}  M. Jimbo, T.

6180: %Miwa, Y. Mori and M. Sato: {\em Density matrix of an

6181: %impenetrable Bose gas and the fifth Painlev\'e

6182: %transcendent}, Physica 1D, 80--158 (1980).

6183: 

6184: \bibitem{Johansson1} %Johansson1

6185: K. Johansson: {\sl  Universality of the Local Spacing

6186: distribution in certain ensembles of Hermitian Wigner

6187: Matrices}, Comm. Math. Phys. {\bf 215}, 683--705 (2001)

6188: 

6189: 

6190:    \bibitem{Johansson2}%Johansson2

6191:    K. Johansson:

6192:   {\sl   Discrete Polynuclear Growth and Determinantal

6193:   Processes},  Comm. Math. Phys. {\bf 242}, 277--329

6194:   (2003).

6195: 

6196:  \bibitem{Johansson3}%Johansson3

6197:   K. Johansson: {\sl   The Arctic circle boundary and

6198:   the Airy process}, Ann. Probab.  {\bf 33}, no. 1, 1--30 (2005)

6199:  (arXiv: Math. PR/0306216)

6200: 

6201:  %Johansson, Kurt The arctic circle boundary and the

6202:  %Airy process.  Ann. Probab.  {\bf 33}, no. 1, 1--30 (2005)

6203: 

6204: 

6205: 

6206: 

6207: %\bibitem{Gross}D. J. Gross and M.J. Newman: {\sl Phys.

6208: %Letters}, {\bf B 266} 291--299 (1991)

6209: 

6210: 

6211: %\bibitem{Kazakov}%Kazakov

6212: %V.A. Kazakov: Nucl. Phys. B (Proc. Suppl.){\bf 4}, 93

6213: %(1988).

6214: 

6215: 

6216: % \bibitem{Johansson} K. Johansson:

6217: %{\sl   Discrete Polynuclear Growth and Determinantal

6218: %Processes},  Comm. Math. Phys. {\bf 242}, 277--329

6219: %(2003).

6220: 

6221: %\bibitem{Johansson1} K. Johansson:

6222: %{\sl   The Arctic circle boundary and the Airy process},

6223: %ArXiv. Math. PR/0306216 (2003).

6224: 

6225: 

6226: \bibitem{Karlin} %Karlin

6227: S. Karlin and J. McGregor: {\sl  Coincidence

6228: probabilities}, Pacific J. Math. {\bf 9}, 1141--1164

6229: (1959).

6230: 

6231: \bibitem{Okounkov} %Okounkov

6232: A. Okounkov and N. Reshetikhin: {\sl Random skew plane

6233: partitions and the Pearcey process}, Comm. Math. Phys.

6234: {\bf 269}, no. 3, 571--609 (2007)

6235: (arXiv:math.CO/0503508)

6236: 

6237: %Okounkov, Andrei; Reshetikhin, Nicolai Random skew plane

6238: %partitions and the Pearcey process.  Comm. Math. Phys.

6239: %{\bf 269}, no. 3, 571--609 (2007)

6240: 

6241: 

6242: 

6243: \bibitem{Pastur} %Pastur

6244: L.A. Pastur: {\sl  The spectrum of random matrices

6245: (Russian)}, Teoret. Mat. Fiz. {\bf 10}, 102--112 (1972).

6246: 

6247: 

6248: 

6249: %\bibitem{Pearcey} %Pearcey

6250: %T. Pearcey: {\sl  The structure of an electromagnetic

6251: %field in the neighborhood of a cusp of a caustic},

6252: %Philos. Mag. {\bf 37}, 311--317 (1946).

6253: 

6254:  \bibitem{Spohn}%Spohn

6255:   M. Pr\"ahofer and H. Spohn:

6256:  {\sl   Scale Invariance of the PNG Droplet and the Airy

6257:  Process}, J. Stat. Phys. {\bf 108}, 1071--1106

6258:  (2002). (arXiv:Math. PR/0105240)

6259: 

6260: 

6261: 

6262: 

6263: 

6264: %\bibitem{Spohn}{Spohn} M. Pr\"ahofer and H. Spohn:

6265: %{\sl   Scale Invariance of the PNG Droplet and the Airy

6266: %Process}, J. Stat. Phys. {\bf 108}, 1071--1106

6267: %(2002).%ArXiv. Math. PR/0105240 (2001).

6268: 

6269: 

6270: 

6271: 

6272: %\bibitem{Szego} G. Szeg\"o: {  Orthogonal

6273: %polynomials}, Amer. Math. Soc., Providence 1959.

6274: 

6275:  \bibitem{TW-Airy}%TW-Airy

6276:  C.A. Tracy and H.~Widom : {\em Level-spacing

6277:  distributions and the {A}iry kernel}, Comm. Math. Phys.,

6278:   {\bf 159}, 151--174, (1994).

6279: 

6280: 

6281: %\bibitem{TW4}

6282: %C.A. Tracy and H.~Widom : {\em Fredholm determinants,

6283: %Differential equations and Matrix models}, Comm. Math.

6284: %Phys. {\bf 163},  33--72, (1994).

6285: %%ArXiv. Math. PR/0309082 (2003).

6286: 

6287: 

6288: %\bibitem{TW2}

6289: %C.A. Tracy and H.~Widom : {\em A system of differential

6290: %equations for the Airy process}, Elect. Comm. in Prob.

6291: %{\bf 8}, 93--98 (2003). %ArXiv. Math. PR/0302033 (2002).

6292: 

6293:  \bibitem{TW-Dyson}%TW-Dyson

6294:  C.A. Tracy and H.~Widom : {\em Differential equations

6295:  for Dyson processes}, Comm. Math. Phys.  {\bf 252}, no.

6296: 1-3, 7--41 (2004) (ArXiv:Math. PR/0309082)

6297: 

6298: %  \bibitem{TW-Pearcey}%TW

6299: % C. Tracy and H. Widom: {\em The Pearcey process}, Comm.

6300: % Math. Phys.  {\bf 263}  (2006),  no. 2, 381--400.

6301: % arXiv:math. PR /0412005.

6302: 

6303: 

6304: 

6305: %\bibitem{Widom}

6306: %H.~Widom : {\em On asymptotics for the Airy process},

6307: %ArXiv. Math. PR/0308157 (2003).

6308: 

6309: 

6310: 

6311: 

6312: 

6313: 

6314: %\bibitem{AvM2} M. Adler and P. van Moerbeke:

6315: %{\sl  "Hermitian, symmetric and symplectic random ensembles: PDE's

6316: %for the distribution of the spectrum}, Annals of Mathematics, {\bf

6317: %153}, 149--189 , 2001 .

6318: 

6319: %\bibitem{AvM2} Adler,~M.,~ Shiota,~T. and van Moerbeke, ~P. : {\em

6320: %Random matrices, vertex operators and the Virasoro algebra}, Phys.

6321: %Lett. A208, 67--78  (1995) and {\em Random matrices, Virasoro

6322: %algebras and ``non-commutative" KP}, Duke Math. J., {\bf 94},

6323: %379-431 (1998).

6324: 

6325: 

6326: 

6327: %\bibitem{Dyson} F.J. Dyson:

6328: %{\sl  A Brownian-Motion Model for the Eigenvalues of a

6329: %Random Matrix}, Journal of Math. Phys. {\bf 3},

6330: %1191--1198 (1962)

6331: 

6332: %\bibitem{Bateman} A. Erdelyi: {Higher transcendental

6333: %functions, Vol. II}, McGraw-Hill (1953)

6334: 

6335: 

6336: 

6337: %\bibitem{Forrester} P.J. Forrester, T. Nagao and G. Honner:

6338: %{\sl   Correlations for the orthogonal-unitary and

6339: %symplectic-unitary transitions at the hard and soft

6340: %edges},

6341: % Nucl. Phys. B, {\bf 553}, 601--643 (1999).

6342:  %ArXiv. Math. 9811142 (1998).

6343: 

6344: 

6345: %\bibitem{JMMS}  M. Jimbo, T.

6346: %Miwa, Y. Mori and M. Sato: {\em Density matrix of an

6347: %impenetrable Bose gas and the fifth Painlev\'e

6348: %transcendent}, Physica 1D, 80--158 (1980).

6349: 

6350: 

6351: %\bibitem{Krasikov} I. Krasikov: {\em New bounds on the

6352: %Hermite polynomials,} arXiv:math.CA/0401310.

6353: 

6354: %\bibitem{Macedo} A. M. S. Mac\^edo: {\sl   Universal parametric

6355: %correlations at the soft edge of the spectrum of random

6356: %matrix ensembles}, Europhysics letters, {\bf 26}, 641,

6357: %(1994). (ArXiv: cond-mat/9404038)

6358: 

6359: 

6360: %\bibitem{Szego} G. Szeg\"o: {  Orthogonal

6361: %polynomials}, Amer. Math. Soc., Providence 1959.

6362: 

6363: %\bibitem{TW}

6364: %C.A. Tracy and H.~Widom : {\em Level-spacing

6365: %distributions and the {A}iry kernel}, Comm. Math. Phys.,

6366: %{\bf 159}, 151--174, (1994).

6367: 

6368: %\bibitem{TW4}

6369: %C.A. Tracy and H.~Widom : {\em Fredholm determinants,

6370: %Differential equations and Matrix models}, Comm. Math.

6371: %Phys. {\bf 163},  33--72, (1994).

6372: %ArXiv. Math. PR/0309082 (2003).

6373: 

6374: 

6375: %\bibitem{TW2}

6376: %C.A. Tracy and H.~Widom : {\em A system of differential

6377: %equations for the Airy process}, Elect. Comm. in Prob.

6378: %{\bf 8}, 93--98 (2003). %ArXiv. Math. PR/0302033 (2002).

6379: 

6380: % \bibitem{TW1}

6381: % C.A. Tracy and H.~Widom : {\em Differential equations

6382: % for Dyson processes}, ArXiv. Math. PR/0309082 (2003).

6383: 

6384: \bibitem{TW-Pearcey}%TW-Pearcey

6385: C. A. Tracy and H. Widom: {\sl The Pearcey Process},

6386: Comm. Math. Phys.  {\bf 263}, no. 2, 381--400 (2006).

6387: (arXiv:math. PR /0412005)

6388: 

6389: %Tracy, Craig A.; Widom, Harold The Pearcey process.

6390: %Comm. Math. Phys.  {\bf 263}, no. 2, 381--400 (2006).

6391: 

6392: 

6393: 

6394: %\bibitem{Widom}

6395: %H.~Widom : {\em On asymptotics for the Airy process},

6396: %ArXiv. Math. PR/0308157 (2003).

6397: 

6398: \bibitem{Peche}%Peche

6399: Sandrine P\'ech\'e: {\em

6400:     The largest eigenvalue of small rank perturbations

6401:     of Hermitian random matrices}

6402:     Probab. Theory Related Fields  {\bf 134},  no. 1,

6403:     127--173 (2006).

6404: (arXiv:math/0411487)

6405: 

6406:     %Péché, S. The largest eigenvalue of small rank

6407:     %perturbations of Hermitian random matrices.

6408:     %Probab. Theory Related Fields  134  (2006),  no. 1, 127--173.

6409: 

6410: \bibitem{UT}%UT

6411:  K. Ueno and K. Takasaki: {\em Toda Lattice

6412: Hierarchy}, Adv. Studies in Pure Math. {\bf 4}, 1--95

6413: (1984).

6414: 

6415: \bibitem{Widom}

6416: H.~Widom : {\em On asymptotics for the Airy process},

6417:  J.

6418:  Statist. Phys.  {\bf 115}, 1129--1134 (2004).

6419: (ArXiv. Math. PR/0308157)

6420: 

6421: 

6422: 

6423:   \bibitem{Zinn1} %Zinn1

6424:   P. Zinn-Justin:

6425:  {\sl   Random Hermitian matrices in an external field},

6426:  Nuclear Physics {\bf B 497}, 725--732

6427:  (1997).

6428: 

6429: 

6430: \bibitem{Zinn2} %Zinn2

6431: P. Zinn-Justin: {\sl   Universality of correlation

6432: functions in Hermitian random matrices in an external

6433: field}, Comm. Math. Phys. {\bf 194}, 631--650 (1998).

6434: 

6435: 

6436: 

6437: 

6438: \end{thebibliography}

6439: 

6440: \vspace*{1cm}

6441: 

6442: adler@brandeis.edu

6443: 

6444: jonathan.delepine@uclouvain.be

6445: 

6446: pierre.vanmoerbeke@uclouvain.be

6447: 

6448: vanmoerbeke@brandeis.edu

6449: 

6450: 

6451: 

6452: %%      For each reference, provide the following information:

6453: 

6454: % \bibitem{ *** LABEL *** }             %% Give a reference label.

6455: % * Name(s) of Author(s) *              %% Enter author(s) names.

6456: % EXAMPLE:  Gray, M., Black, F., and White, A.

6457: 

6458: %%      Use the following template for a journal article:

6459: % * Title of article *.                 %% Example: Existence and uniqueness.

6460: % \textit{* Abbreviated journal name *} %% Example: \textit{Comm. Pure Appl. Math.}

6461: % \textbf{* Volume number *}            %% Example: \textbf{72}

6462: % (* Year of publication *),            %% Example: (1993),

6463: % * Issue number [optional],            %% Example: no. 6,

6464: % * Page range *.                       %% Example: 675--690.

6465: 

6466: %%      Use the following template for a book:

6467: % \textit{* Title of book *}.           %% Example: \textit{Ancient Topology}.

6468: % * Publisher *,                        %% Example: Wiley-Interscience,

6469: % * City of publisher *,                %% Example: New York,

6470: % * Year of publication *.              %% Example: 1993.

6471: 

6472: 

6473: %%      ---------------------------------------------------------------------

6474: %%      ------------------------ CONTACT INFORMATION ------------------------

6475: %%      ---------------------------------------------------------------------

6476: 

6477: %      Place contact information for each author between

6478: %      the \begin{comment} and \end{comment} commands. Include

6479: %      preferred mailing address and e-mail addresses. Please note

6480: %      that these will not print. We will format them for printing

6481: %      during the editing stage.

6482: 

6483: \begin{comment}

6484: 

6485: \author{Mark Adler\thanks{2000

6486: {\em Mathematics Subject Classification}. Primary:

6487: 60G60, 60G65, 35Q53; secondary: 60G10, 35Q58. {\em Key

6488: words and Phrases}: Dyson's Brownian motion, Airy

6489: process, extended kernels, random Hermitian ensembles,

6490: coupled random matrices. \newline

6491:  Department of Mathematics, Brandeis University,

6492: Waltham, Mass 02454, USA. E-mail: adler@brandeis.edu.

6493: The support of a National Science Foundation grant \#

6494: DMS-04-06287 is gratefully acknowledged.} ~~~~ Jonathan

6495: Del\'epine\thanks{ Department of Mathematics,

6496: Universit\'e de Louvain, 1348 Louvain-la-Neuve,

6497: Belgium.}~~~~~~~~~~~ Pierre van Moerbeke\thanks{

6498: Department of Mathematics, Universit\'e de Louvain, 1348

6499: Louvain-la-Neuve, Belgium and Brandeis University,

6500: Waltham, Mass 02454, USA. This work was done while PvM

6501: was a member of the Miller Institute for Research in

6502: Science, Berkeley, California. E-mail:

6503: pierre.vanmoerbeke@uclouvain.be and

6504: vanmoerbeke@brandeis.edu. The support of a National

6505: Science Foundation grant \# DMS-04-06287, a European

6506: Science Foundation grant (MISGAM), a Marie Curie Grant

6507: (ENIGMA), FNRS and ``Inter-University Attraction Pole

6508: (Belgium)" grants is gratefully acknowledged.}}

6509: 

6510: 

6511: 

6512: 

6513: %Sample:

6514: 

6515: Native New Yorker

6516: Courant Institute

6517: 251 Mercer St., Room 1414

6518: New York, NY 10012

6519: natnewyork@nyu.edu

6520: 

6521: \end{comment}

6522: \end{document} %%<=== do not delete. This will end paper properly.

6523: 

6524: %%------------------------ END OF PAPER ------------------

6525: 

6526: 

6527: 

6528: 

6529: 

6530: @@@@@@@@@@@

6531: 

6532:  The {\bf Airy process with $r$ outliers}

6533:   (in short: {\bf $r$-Airy process}) $\mathcal {A}^{(r)}(\tau)$

6534:   is defined as the limit of the non-intersecting

6535:   Brownian motions (\ref{Brownian target})

6536:   for $n\rg \iy$ and fixed $r\geq 1$, with the same

6537:   rescaling of time and

6538:   space, but where the target point $a$ coincides with

6539:   $\sqrt{n/2}$, the maximum of the edge for the Airy process.

6540:    This is an extension of the Airy process

6541:    $\mathcal {A} (\tau)$, when the

6542:   target point $a$ coincides with $0$ or,

6543:   what is the same, where $r=0$. The Airy process

6544:   is stationary, whereas the $r$-Airy process ceases to be stationary.

6545:   Given a subset $E\subset {\mathbb R}$, and rescaling

6546:   the target point as $a=\sqrt{n/2}$,

6547:   the $r$-Airy process is then

6548:   defined by

6549: %

6550:   \bea

6551:   \lefteqn{ {~~~~~~~~~\BP(\mathcal {A}^{(r)}(\tau)\cap E%(x,\iy)

6552:   =\emptyset)}}%\no

6553:    \label{r-AiryP}\\&&

6554:   \! := \! \lim_{n\rightarrow \iy}

6555: \left.\BP^{(0,\rho\sqrt{n/2})}_{Br}\left(\mbox{all}~

6556: x_i\left(\frac{e^{  {\tau}/{n^{1/3}}}}

6557:  {2\cosh(\tau/n^{1/3})}\right) \in

6558:   \frac{  \sqrt{2n}+\frac{E^c}{\sqrt{2}

6559:   n^{1/6}} }{2\cosh( \tau/n^{1/3})}\right)\right|_{\rho=1}

6560:   .~~~~ %

6561:   \no%\no\\

6562:   \eea

6563: %

6564:   %ZZZZZZZZZZZZZZZZ: E

6565: %

6566:   Notice that the time-origin $\tau=0$ of the $r$-Airy process

6567:   corresponds to $t=1/2$ in the original time-scale. The

6568:   $\cosh$ appearing in the denominator takes into account the

6569:   constraint that the paths must leave from $0$ and that

6570:   most return to $0$.

6571: 

6572:   Consider now the target point $a=\rho\sqrt{n/2} $, for an arbitrary

6573:   parameter $0\leq \rho\leq1$. %We now multiply

6574:   %the target point $a$ with a parameter $0\leq \rho\leq 1$;

6575:   The limit will now experience a phase transition

6576:   at $\rho=1$. In other terms, if the target point is at $\sqrt{n/2}$, then the $r$-Airy process occurs in the neighborhood of $(y,t)=(\sqrt{n/2},1/2)$. More generally, for an arbitrary target point located at $a=\rho_0\sqrt{n/2}$ for any $\rho_0>0$, the $r$-Airy process occurs in the neighborhood of the point of tangency of the tangent to the curve $y=\sqrt{2nt(1-t)}$ passing through the point $(\rho_0\sqrt{n/2},1)$ in the $(y,t)$-plane. This more general fact will be stated in Theorem \ref{Theo1}.

6577: 

6578: 

6579: %{\bf Mark: the figures will have to be redrawn}

6580: %\vspace*{-.3cm}

6581: 

6582: 

6583: %       ~~~~~~\includegraphics[width=60mm,height=55mm]{Brownianmotions2.jpg}

6584: 

6585: 

6586: 

6587: 

6588: %\vspace{2cm}%%%%%

6589: 

6590: % \vspace*{.1cm} ~~~~~~~~~~~~~~~~~~~~Airy

6591: %process~~~~~~~~~~~~~~~~~~~~~~~~~~~~~$r$-Airy process

6592: %(for $\rho=1$)

6593: 

6594: %\vspace*{.3cm}

6595: 

6596: 

6597: 

6598: \noindent Define\footnote{$C$ is a contour running from $\iy

6599:  e^{5i\pi/6}$ to $\iy

6600:  e^{i\pi/6}$, such that $-i\tau$ lies above the contour.}

6601:  \be

6602:  A^{\pm}_r(u;\tau):=

6603:   \int_C   e^{\frac{1}{3} ia^3+iau }

6604:  \left(\mp ia-\tau\right)^{\pm r}

6605:  \frac{da}{2\pi}

6606:   \label{k-Airy0}\ee

6607: %

6608: and $A(u):=A^-_0(u;\tau)=A^+_0(u;\tau)$, the standard

6609: Airy function, satisfying the differential equation

6610:  $

6611: A'' (x)=xA(x).$

6612: 

6613: 

6614: \begin{theorem}\label{Theo1}

6615:  The limit of the probability for the non-intersecting Brownian motions, viewed from an arbitrary point $(y_0,t_0)$ of the curve

6616: $y=\sqrt{2nt(1-t)}$ for $0<t<1$ with time and space rescaled as below,  exists for $n\rg \iy$ and $0\leq \rho\leq 1$. It equals a

6617: Fredholm determinant, which defines a probability

6618: distribution\footnote{$\det(I-K)_E$ defines a Fredholm

6619: determinant restricted to the subset $E$.} for each time $\tau$ and experiences a phase transition at %$\rho=1$. Setting

6620:  \be

6621:  \rho=\rho_0:=\sqrt{\frac{1-t_0}{t_0}}=: e^{ -\tau_0}

6622:   .

6623:     \label{rho0}\ee

6624:     Namely,

6625: %

6626: %

6627:  \bea \lefteqn{

6628:    \lim_{n\rightarrow \iy}

6629: \BP^{(0,\rho\sqrt{n/2})}_{Br}\left(\mbox{all}~

6630: x_i\left( \frac{1}

6631:  {1+e^{-2(\tau_0+\frac{\tau}{n^{1/3}})}}\right) \in

6632:   \frac{\sqrt{2n}+\frac{E^c}{\sqrt{2}n^{1/6}}   } {2\cosh(\tau_0+\frac{\tau}{n^{1/3}})}

6633:   % ~~\mbox{with}~e^{ \al_0}=\sqrt{\frac{s_0}{1-s_0}}

6634:    \right) }\no\\

6635:   %&=&\BP(\mathcal {A}^{(r)}(t)\cap (x,\iy)=\emptyset)\\

6636:   %

6637:  %

6638:  \no\\

6639:  %

6640: &&\left\{\begin{array}{l}

6641:  \BP(\mathcal {A}^{(r)}(\tau)\cap E=\emptyset)=

6642:   \det\left(I-K_\tau^{(r)}\right)_E ~~\mbox{for}~ \rho=\rho_0\\

6643:  \\ \BP(\mathcal {A}(\tau)\cap E=\emptyset)=

6644:  \det\left(I-K^{(0)}\right)_E

6645:   ~~\mbox{for}~0 \leq\rho<\rho_0 ,%

6646:   \end{array}\right. %\label{xx}

6647:    \label{0.3}

6648:   \eea

6649: %

6650:  where the $r$-Airy and Airy kernels are defined by

6651:    \begin{equation}\label{r-kernel}\begin{split}

6652:  K^{(r)}_\tau(u,v)&=\int_0^{\iy}

6653:  dw A_r^-(w+u; \tau)A_r^+(w+v; \tau)\\

6654:  K^{(0)}(u,v)&= \int_0^{\iy}

6655:  dw A(w+u)A(w+v)=\frac{A(u)A'(v)-A'(u)A(v)}{u-v}.

6656:  \end{split}\end{equation}

6657:  Notice that for $0\leq \rho <\rho_0$, the

6658: limit above is independent of the time $\tau$ and the number of outliers $r$, unlike the case $\rho=\rho_0$.

6659: Also notice that the point of phase transition $(\rho_0\sqrt{\frac{n}{2}},1)=\left(\sqrt{\frac{1-t_0}{ {2}t_0}}\sqrt{n},1\right)=\left(\frac{y_0}{2t_0},1\right)$ on the $t=1$ axis in $(y,t)$-space is precisely the point where

6660:   the line tangent to the curve $y=\sqrt{2nt(1-t)}$ at the point $(y_0,t_0)$ intersects the line $t=1$.

6661: \end{theorem}

6662: 

6663: 

6664: 

6665: For a point $(y_0,t_0)$ on the curve $y=\sqrt{2nt(1-t)}$, the space scaling in the probability (\ref{0.3}) %

6666: can be written as:

6667: $$

6668:  \frac{\sqrt{2n}+\frac{E^c}{\sqrt{2}n^{1/6}}   } {2\cosh(\tau_0+\frac{\tau}{n^{1/3}})}

6669:   =\frac{y_0+\frac{E^c\sqrt{t_0(1-t_0)}}{\sqrt{2} n^{1/6}}}

6670:  {t_0 e^{ {\tau}/{n^{1/3}}}+(1-t_0)e^{- {\tau}/{n^{1/3}}} }

6671: .$$

6672: The maximum of the curve $y=\sqrt{2nt(1-t)}$ takes place at $t_0=1/2$, for which $\rho_0=1$ and $\tau_0=0$. For $\rho_0<1$, we have $\tau_0>0$ and $t_0>1/2$; for $\rho_0>1$, we have $\tau_0<0$ and $t_0<1/2$.

6673: 

6674: 

6675: @@@@@@@@@@@@@@@@

6676: