1: \documentclass[12pt]{article}

2: 

3: \usepackage{psfig,amssymb}

4: 

5: \setlength{\oddsidemargin}{.0001in}

6: \setlength{\evensidemargin}{.0001in}

7: \setlength{\textwidth}{6.5in}

8: \setlength{\textheight}{8.5in}

9: \setlength{\topmargin}{.0001in}

10: \setlength{\parskip}{.06in}

11: \setlength{\parindent}{.3in}

12: %\newcommand{\sss}{\setcounter{equation}{0}}

13: 

14: 

15: \makeatletter

16: \def\eqalign#1{\null\vcenter{\def\\{\cr}\openup\jot\m@th

17:   \ialign{\strut$\displaystyle{##}$\hfil&$\displaystyle{{}##}$\hfil

18:       \crcr#1\crcr}}\,}

19: \makeatother

20: \newcommand{\lra}{\leftrightarrow}

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

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

23: \newcommand{\beq}{\begin{eqnarray}}

24: \newcommand{\eeq}{\end{eqnarray}}

25: \newcommand{\bt}{\begin{theorem}}

26: \newcommand{\et}{\end{theorem}}

27: \newcommand{\bl}{\begin{lemma}}

28: \newcommand{\el}{\end{lemma}}

29: \newcommand{\bc}{\begin{corollary}}

30: \newcommand{\ec}{\end{corollary}}

31: %\newcommand{\bp}{\begin{prop}}

32: %\newcommand{\ep}{\end{prop}}

33: \newcommand{\ba}{\begin{array}}

34: \newcommand{\ea}{\end{array}}

35: \newcommand{\bs}{\backslash}

36: 

37:  

38: \newcommand{\la}{\label}

39: \newcommand{\ci}{\cite}

40: \newcommand{\proof}{\noindent{\bf Proof: \ }}

41: \newcommand \qed {\hskip 6pt\vrule height6pt width5pt depth1pt \bigskip}

42: \newtheorem{theorem}{Theorem}

43: \newtheorem{lemma}[theorem]{LEMMA}

44: \newtheorem{corollary}[theorem]{COROLLARY}

45: \newtheorem{prop}[theorem]{PROPOSITION}

46: \newtheorem{remark}[theorem]{REMARK}

47: %\renewcommand{\theequation}{\arabic{section}.\arabic{equation}}

48: 

49: \newcommand{\ti}{\tilde}

50: \newcommand{\de}{\delta}

51: \newcommand{\De}{\Delta}

52: \newcommand{\al}{\alpha}

53: \newcommand{\ga}{\gamma}

54: \newcommand{\Ga}{\Gamma}

55: \newcommand{\ch}{\chi}

56: \newcommand{\si}{\sigma}

57: \newcommand{\Si}{\Sigma}

58: \newcommand{\om}{\omega}

59: \newcommand{\Om}{\Omega}

60: \newcommand{\ka}{\kappa}

61: \newcommand{\lb}{\lambda}

62: \newcommand{\ze}{\zeta}

63: \renewcommand{\th}{\theta}

64: \newcommand{\Th}{\Theta}

65: \newcommand{\vp}{\varphi}

66: \newcommand{\ph}{\phi}

67: \newcommand{\ps}{\psi}

68: \newcommand{\Ph}{\Phi}

69: \newcommand{\La}{\Lambda}

70: \newcommand{\ep}{\varepsilon }

71: 

72: \newcommand{\e}{\rm e}

73: \newcommand{\R}{{\cal R}}

74: \newcommand{\I}{{\cal I}}

75: \newcommand{\cH}{\cal H}

76: \newcommand{\F}{{\cal F}}

77: \newcommand{\K}{{\cal K}}

78: \newcommand{\T}{{\cal T}}

79: \newcommand{\cB}{{\cal B}}

80: \newcommand{\A}{{\cal A}}

81: \newcommand{\M}{{\cal M}}

82: \newcommand{\X}{{\cal X}}

83: \newcommand{\cV}{{\cal V}}

84: \newcommand{\cW}{{\cal W}}

85: \newcommand{\cL}{{\cal L}}

86: 

87: 

88: \newcommand{\bi}{\bibitem}

89: 

90: \newfont{\msbm}{msbm10 scaled\magstep1}%blackboardbold

91: \newfont{\msbms}{msbm7 scaled\magstep1} %blackboardbold   subscript

92: %\newcommand{\msbm}{\bf}

93: %\newcommand{\msbms}{\bf}

94: 

95: \newcommand{\bbr}{\mbox{$\mbox{\msbm R}$}}

96: \newcommand{\bbn}{\mbox{$\mbox{\msbm N}$}}

97: \newcommand{\bbi}{\mbox{$\mbox{\msbm I}$}}

98: \newcommand{\bbc}{\mbox{$\mbox{\msbm C}$}}

99: \newcommand{\bbk}{\mbox{$\mbox{\msbm K}$}}

100: 

101: \newcommand{\bbe}{\mbox{$\mbox{\msbm E}$}}

102: \newcommand{\bbz}{\mbox{$\mbox{\msbm Z}$}}

103: \newcommand{\bbp}{\mbox{$\mbox{\msbm P}$}}

104: \newcommand{\bbt}{\mbox{$\mbox{\msbm T}$}}

105: 

106: \newcommand{\bbrs}{\mbox{$\mbox{\msbms R}$}} %for subscripts

107: \newcommand{\bbns}{\mbox{$\mbox{\msbms N}$}}

108: \newcommand{\bbis}{\mbox{$\mbox{\msbms I}$}}

109: \newcommand{\bbcs}{\mbox{$\mbox{\msbms C}$}}

110: \newcommand{\bbks}{\mbox{$\mbox{\msbms K}$}}

111: \newcommand{\bbes}{\mbox{$\mbox{\msbms E}$}}

112: \newcommand{\bbzs}{\mbox{$\mbox{\msbms Z}$}}

113: \newcommand{\bbps}{\mbox{$\mbox{\msbms P}$}}

114: 

115: \newcommand{\B}{\mbox{$\mbox{\msbm B}$}}

116: \newcommand{\Bs}{\mbox{$\mbox{\msbms B}$}} %for subscripts

117: \newcommand{\E}{\mbox{$\mbox{\msbm E}$}}

118: 

119: \newcommand{\eps}{\epsilon}

120: \newcommand{\Gd}{$G_\delta$}

121: \newcommand{\Fs}{$F_\sigma$}

122: 

123: 

124: 

125: \begin{document}

126: \def\wt{\widetilde}

127: %\hfill {\small November 19, 2003}

128: \bigskip\bigskip\bigskip

129: \begin{center}

130: {\Large\bf 

131: Gap probability in the spectrum of random matrices

132: and asymptotics of polynomials orthogonal on an arc of the unit circle. 

133: }\\

134: \bigskip\bigskip\bigskip

135: I. V. Krasovsky\\

136: \bigskip Technische Universit\"at Berlin\   

137: Institut f\"ur Mathematik MA 7-2\\

138: Strasse des 17. Juni 136, D-10623, Berlin, Germany\\

139: E-mail: ivk@math.tu-berlin.de\\

140: \bigskip\bigskip

141: \end{center}

142: \bigskip\bigskip\bigskip

143: \noindent{\bf Abstract.} We obtain uniform asymptotics for polynomials

144: orthogonal on a fixed and varying arc of the unit circle with a 

145: positive analytic weight function. We also

146: complete the proof of the large $s$ asymptotic expansion for the

147: Fredholm determinant with the kernel $\sin z/(\pi z)$ on the interval 

148: $[0,s]$, verifying a conjecture of Dyson for the

149: constant term in the expansion.

150: In the Gaussian Unitary Ensemble of random matrices, this

151: determinant describes the probability for an interval of length $s$

152: in the bulk scaling limit to be free from the eigenvalues.

153: \bigskip\bigskip\bigskip

154: 

155: 

156: 

157: 

158: 

159: 

160: 

161: 

162: 

163: \section{Introduction}

164: One problem in the random matrix theory is estimation of the probability

165: for a given interval to be free from the eigenvalues. In the Gaussian

166: Unitary Ensemble this probability for any interval of length $2s$

167: in the bulk scaling limit is equal to the following

168: Fredholm determinant:

169: \begin{equation}

170: \De(s)=\det[I-K],

171: \label{De}

172: \end{equation}

173: where $K$ is the integral operator on $L^2(0,2s)$ given by

174: \[

175: (Kg)(x)=\int_0^{2s}{\sin (x-y)\over \pi(x-y)}g(y)dy.

176: \]

177: 

178: The probability of a gap in the spectrum of random matrices from

179: orthogonal and symplectic ensembles 

180: is also expressed in terms of $\De(s)$ (see \cite{Mehta,TW,Dbook}). 

181: 

182: It was shown by Jimbo,  Miwa, M\^ori, and Sato \cite{JMMS} 

183: that $(d/ds)\ln \De(s)$ satisfies a 

184: modified  Painlev\'e V equation (for simpler proofs of this see 

185: \cite{TW,DIZ}).  

186: 

187: An interesting question is 

188: calculation of the asymptotics of $\De(s)$ for large $s$ 

189: (the small $s$ series are easy to obtain). 

190: The first two terms in the expansion

191: of $\ln\De(s)$ were found by des Cloizeaux and Mehta \cite{CM} who used

192: a connection with the spheroidal functions. The full asymptotic

193: expansion was obtained by Dyson \cite{Dyson} with the help of the inverse

194: scattering techniques for Schr\"odinger operators.  

195: These calculations were partly conjectural. A rigorous derivation of the

196: main term was given by Widom \cite{WidomAs} using continuous analogues of

197: orthogonal polynomials. Finally, Deift, Its, and Zhou \cite{DIZ} 

198: (see also that work for a

199: more extensive introduction) calculated, as a particular case of a

200: more general result, the full asymptotics of the derivative

201: $(d/ds)\ln\De(s)$ using techniques of matrix Riemann-Hilbert problems.

202: This settled the question up to the constant term

203: in the expansion of $\ln\De(s)$.

204: The first 3 terms in the Dyson expansion are as follows:

205: \be

206: \ln \De(s)=-{s^2\over 2}-{1\over 4}\ln s+c_0+O\left({1\over s}\right),

207: \qquad s\to\infty,\label{exp}

208: \ee

209: where the constant term $c_0=(1/12)\ln 2 +3\zeta'(-1)$, and

210: $\zeta'(x)$ is the derivative of Riemann's zeta function.

211: Thus, justification of $c_0$ here remained the only problem and it is

212: solved in the present paper. Actually, we obtain the first 

213: 3 terms:\footnote{As this paper was being prepared for publication,

214: an announcement by T. Ehrhardt claiming the same result as Theorem 1

215: (by a different method) was posted on the internet.

216: 

217: A third solution to the problem by a Riemann-Hilbert approach (related to

218: the present one) is in preparation by P. Deift, A. Its, and X. Zhou.

219: }

220: \bt

221: The large $s$ asymptotics of $\ln \De(s)$ are given by (\ref{exp}).

222: \et

223: 

224: The proof is based on a formula by Deift \ci{D} which connects the

225: determinants of two Toeplitz matrices, a formula by Widom \ci{Warc} for

226: asymptotics of Toeplitz determinants on a circular arc, and on

227: asymptotics for orthogonal polynomials on a circular arc which are

228: computed here.

229: 

230: Let $f_\al(\th)$ be a weight function on an arc 

231: $\al\le\th\le 2\pi-\al$, $z=e^{i\th}$, 

232: $0<\al<\pi$ of the unit circle $|z|=1$, 

233: and $\phi_n(z,\al)=\chi_n z^n+\cdots$, $n=0,1,\dots$ 

234: the corresponding system of orthonormal polynomials:

235: \be

236: {1\over 2\pi}\int_\al^{2\pi-\al}

237: \phi_k(e^{i\th},\al)\overline{\phi_m(e^{i\th},\al)}f_\al(\th)d\th=

238: \de_{km},\qquad k,m=0,1,\dots.\la{phi1}

239: \ee

240: 

241: Such polynomials in the case of the circle ($\al$=0) were first 

242: studied by Szeg\H o (see \ci{Sz}) who, in

243: particular, found several important asymptotics for them as $n\to\infty$.  

244: Afterwards, asymptotic analysis of such polynomials was carried

245: out by many authors. Specifically for the case of an arc (whose study

246: was initiated by Akhieser \ci{Akh}), see

247: \ci{GN,GAkh,BL,RL} and references therein. 

248: However, the full asymptotic expansion

249: at all points $z\in\bbc$ for a wide class of weights

250: became a feasible task only after recent development

251: of Riemann-Hilbert problem methods. 

252: It was observed by Fokas, Its, and Kitaev \ci{fik} that 

253: orthogonal polynomials satisfy certain matrix Riemann-Hilbert problems.

254: An efficient method for their asymptotic solution (steepest descent

255: techniques) was developed by Deift and Zhou \ci{dz,Dbook}

256: and applied for analysis of polynomials orthogonal on the real axis

257: in \ci{DK1,DK2} (see also \ci{BI} for a different Riemann-Hilbert

258: approach) and on the unit circle in \ci{bdj}. 

259: The case of polynomials orthogonal on $[-1,1]$

260: (especially relevant for the present work) 

261: was considered by Kuijlaars, McLauphlin, Van Assche, and Vanlessen

262: \ci{kuimcl,kuivan} who found full asymptotics at all points in the

263: case of a positive analytic weight on $[-1,1]$ with 

264: power-type singularities at the end-points. 

265: 

266: In the present paper, we shall give a procedure to obtain full

267: asymptotics for all $z$ for polynomials $\phi_n(z,\al)$ and their leading

268: coefficients as $n\to\infty$ in the case of a positive analytic

269: weight $f_\al(\th)$. 

270: The argument will be similar to that of  \ci{kuimcl}.

271: We consider $2s/n\le\al<\pi$, $n>s$, $s\to\infty$, which includes 

272: both the cases of a fixed arc ($\al$ is

273: independent of $n$) and a varying arc.

274: The asymptotics for $\phi_n(z,\al)$ we obtain are in the inverse

275: powers of $n\sin(\al/2)$. 

276: The remainder after $k$ terms is uniform in $\al$.

277: The general solution is given by equations

278: (\ref{i2}--\ref{aschi}). The first 2 asymptotic terms for any $z$ can

279: be easily written using (\ref{R1},\ref{A1B1}). An example is given in

280: (\ref{p1},\ref{c1}). 

281: 

282: Our solution can be generalized to the following cases:

283: (1) the weight $f_\al(\th)$ has power-type singularities at the

284: end-points of the arc (this can be done following \ci{kuimcl});

285: (2) the weight $f_\al(\th)$ is not analytic but only smooth enough and

286: positive (one can approximate it then by its Fourier series).

287: 

288: If the weight is symmetric $f_\al(\th)=f_\al(2\pi-\th)$, there exist

289: relatively simple formulas of Szeg\H o type \ci{Sz,Zh,fredh}

290: connecting polynomials on an arc with those on an interval. In this

291: case and for a fixed arc one could try to obtain our results

292: from those of \ci{kuimcl, kuivan}. 

293: The present argument, however, is more direct.

294: 

295: For the proof of Theorem 1, we shall only need asymptotics of

296: polynomials with the weight $f_\al(\th)=1$ and only at the

297: point $e^{i\al}$. What we need is summarized in the following theorem

298: proved in Section 2 (after the argument in the general case is given):

299: 

300: \bt

301: Let $0<\al<\pi$, $\ga=\cos(\al/2)$,

302: \be\eqalign{

303: r^1_+={e^{-i\al/2}\over 3\cdot2^4 i}(1+e^{-i\al}-2e^{i\al}),\qquad

304: r^1_-={e^{-i\al/2} \over 2^4 i}(1+3e^{i\al}),\\

305: r^2_+={1\over 3\cdot2^9}(16-9e^{i\al}+43e^{-i\al}-2e^{-2i\al}),\qquad

306: r^2_-={1\over 2^9}(-6+7e^{i\al}-17e^{-i\al}),\\

307: {r^1_-}'={e^{i\al/2} \over 4}\cos^2(\al/2),\qquad

308: \tau={1+2\cos\al\over 6i},\qquad

309: \rho=n\sin(\al/2),\qquad \ep>0.}

310: \ee

311: Let $f_\al(\th)=1$.

312: Then the polynomial $\phi_n(z,\al)$ admits an asymptotic expansion 

313: for large $\rho$ in the inverse powers of $\rho$. We have for $z=e^{i\al}$

314: \be

315: \phi_n(e^{i\al},\al)=\chi_n

316: \ga^n e^{i\al(n/2-1/4)}\sqrt{\pi i \rho}

317: \left[1+{r^1_- \over \rho}+

318: {r^2_- \over \rho^2}+ {r^3_- \over \rho^3}+

319: O\left({1\over\rho^4}\right)\right],\la{phin}

320: \ee

321: where $\chi_n$ is the leading coefficient of 

322: $\phi_n(z,\al)=\chi_n z^n+\cdots$ for which we have

323: \be

324: \chi_{n-1}^2=\ga^{-2n+1}\left[1+{1\over 4n}+{5\over

325:     2^5n^2}+O\left({1\over n^3}\right)\right],\la{chin}

326: \ee

327: and $r^3_-$ is a bounded function of $\al$.

328: The derivative of the polynomial

329: $\phi_n'(z,\al)=(d/dz)\phi_n(z,\al)$

330: at $e^{i\al}$ can be written as

331: \be\eqalign{

332: \phi_n'(e^{i\al},\al)={n\over 2}\phi_n(e^{i\al},\al)e^{-i\al}+

333: \chi_n\ga^n e^{i\al(n/2-5/4)}{\sqrt{\pi i \rho}\over 2\sin\al}

334: \left[i\rho^2+e^{i\al/2}\rho+\tau+\right. \\ \left.

335: {1\over\rho}(r^1_-(i \rho^2+\tau)+r^1_+e^{i\al/2}\rho+{r^1_-}')+

336: {1\over\rho^2}(r^2_- i\rho^2+r^2_+e^{i\al/2}\rho)+

337: {i r^3_-\over\rho} +O\left({1\over\rho^2}\right)\right].}\la{phinp}

338: \ee

339: There exists $s_0>0$ such that

340: all the remainder terms are valid and uniform 

341: in $\al$, $s$, and $n$

342: for $\al\in[2s/n,\pi-\ep]$, $s>s_0$, and $n>s$.

343: \et

344:  

345: \noindent

346: {\bf Remark}

347: The uniformity of the remainders here is crucial for 

348: the proof of Theorem 1.

349: 

350: 

351: After constructing asymptotics for polynomials in Section 2, we give a

352: proof of Theorem 1 in Section 3. Note that the present method could

353: also be used to obtain the full asymptotic expansion of $\De(s)$.

354: 

355: 

356: 

357: 

358: 

359: 

360: 

361: 

362: 

363: 

364: 

365: 

366: 

367: 

368: \section{Asymptotics of polynomials on an arc}

369: In the present section we construct asymptotics for polynomials

370: $\phi_n(z)=\chi_n z^n+\cdots$ orthonormal with 

371: a weight $f(z)=f_\al(\th)$ on an arc

372: $\al\le\th\le2\pi-\al$, $z=e^{i\th}$, for $2s/n\le\al\le\pi-\ep$, $\ep>0$, 

373: $n>s$, $s\to\infty$. This includes both the cases of a fixed arc

374: $0<\al<\pi$ and the varying arc $\al=2s/n$. The function $f(z)$ is

375: assumed positive and analytic on the arc for a fixed arc case, and on

376: the whole circle in the general case. In the general case, we obtain

377: $\al$-uniform asymptotics in the inverse powers of $n\sin(\al/2)$. 

378: 

379: Consider the following $2\times 2$ matrix 

380: \begin{equation} \label{RHM}

381:     Y(z) =

382:     \pmatrix{

383: \chi_n^{-1}\phi_n(z) & 

384: \chi_n^{-1}\int_{\Si}{\phi_n(\xi)\over \xi-z}

385: {f(\xi)d\xi \over 2\pi i \xi^n} \cr

386: \chi_{n-1}\phi^*_{n-1}(z) & 

387: \chi_{n-1}\int_{\Si}{\phi^*_{n-1}(\xi)\over \xi-z}

388: {f(\xi)d\xi \over 2\pi i \xi^n}

389:     },

390: \end{equation}

391: where $\Si$ is the arc $\al \le\th\le 2\pi-\al$ of the unit circle

392: traversed in the direction from $2\pi-\al$ to $\al$, and

393: $\phi^*_n(z)=z^n\overline{\phi_n(1/{\bar z})}$.

394: As is easy to verify (see \ci{bdj,kuimcl}),

395: $Y(z)$ is the unique solution of the following Riemann-Hilbert problem:

396: \begin{enumerate}

397:     \item[(a)]

398:         $Y(z)$ is  analytic for $z\in\bbc \setminus\Si$.

399:     \item[(b)]

400: For $\th \in (\al,2\pi-\al)$, 

401: $Y$ has continuous boundary values

402: $Y_{+}(x)$ as $z$ approaches $x=e^{i\th}$ from

403: the outside of the circle, and $Y_{-}(x)$, from the inside. 

404: They are related by the jump condition

405: \begin{equation}\label{RHPYb}

406:             Y_+(x) = Y_-(x)

407:             \pmatrix{

408:                 1 & x^{-n}f(x) \cr

409:                 0 & 1},

410:             \qquad\mbox{$x=e^{i\th},\qquad \th\in (\al,2\pi-\al)$.}

411:         \end{equation}

412:     \item[(c)]

413:         $Y(z)$ has the following asymptotic behavior at infinity:

414:         \begin{equation} \label{RHPYc}

415:             Y(z) = \left(I+ O \left( \frac{1}{z} \right)\right)

416:             \pmatrix{

417:                 z^{n} & 0 \cr

418:                 0 & z^{-n}}, \qquad \mbox{as $z\to\infty$.}

419:         \end{equation}

420:     \item[(d)]

421:        Near the end-points of the arc $e^{\pm i\al}$ 

422:         \begin{equation}\label{RHPYd}

423:             Y(z) =  O\pmatrix{ 

424:                     1 & \log|z-e^{\pm i\al}| \cr

425:                     1 & \log|z-e^{\pm i\al}|},

426:         \end{equation}

427: as $z \to e^{\pm i\al}$, $z \in \bbc \setminus \Si$.

428: \end{enumerate}

429: 

430: The function

431: \be

432: \psi(z)={1\over2\ga}\left(z+1+\sqrt{(z-e^{i\al})(z-e^{-i\al})}\right),

433: \qquad \ga=\cos(\al/2),\la{psi}

434: \ee

435: which conformally maps the outside of the arc $\Si$ into the outside

436: of the unit circle, will have an 

437: important role in what follows (cf. (6.18) of \ci{DIZ}). 

438: Here we take the branch of the square

439: root which is positive for positive arguments.

440: Note that the boundary values of $\psi(z)$, $\psi_+(x)$ as $z$

441: approaches $x\in\Si$ from the outside of the circle, and  $\psi_-(x)$, from

442: the inside, are related as:

443: \be

444: \psi_+(x)\psi_-(x)= x.\la{bvf}

445: \ee

446: Let $\mu(z)$ be defined by the equation (\ref{psi}) but with the minus

447: sign in front of the square root. Then $\mu(z)$ is the mapping of the

448: outside of the arc into the {\it inside} of the unit circle. 

449: Hence $|\mu(z)|<1$, whereas $|\psi(z)|>1$ for 

450: $z\in\bbc\setminus\Si$.

451: Therefore we have that first, for $|z|<1$  $|z/\psi(z)^2|<1$, and

452: second, for $|z|>1$

453: \[

454: \left|{z\over\psi(z)^2}\right|=

455: \left|{z\mu(z)^2\over\psi(z)^2\mu(z)^2}\right|=

456: \left|{z\mu(z)^2\over z^2}\right|=\left|{\mu(z)^2\over z}\right|<1.

457: \] 

458: Thus,

459: \be\la{ineq}

460: \left|{z\over\psi(z)^2}\right|<1

461: \qquad\mathrm{for}\qquad |z|\neq 1.

462: \ee

463: This inequality will be useful later on.

464: 

465: We now replace the original Riemann-Hilbert problem with an equivalent one

466: which is normalized to unity at infinity and has oscillating elements

467: of the jump matrix.

468: Namely, set

469: \be

470: T(z)=\ga^{-n\si_3}Y(z)\psi(z)^{-n\si_3},\qquad \si_3=\pmatrix{1&0\cr 0&-1}.

471: \ee

472: Then, as is easy to verify, $T(z)$ satisfies the same problem

473: as $Y(z)$ (\ref{RHM}) but with the changed conditions (b) and (c):

474: \begin{enumerate}

475: \item[(b)]

476: \begin{equation}

477:             T_+(x) = T_-(x)

478:             \pmatrix{

479:                 x^n\psi_+(x)^{-2n} & f(x) \cr

480:                 0 & x^n\psi_-(x)^{-2n}},

481:             \qquad\mbox{$x\in\Si$,}

482:         \end{equation}

483:     \item[(c)]

484:         \begin{equation} 

485:             T(z) = I+ O \left( \frac{1}{z} \right), 

486:      \qquad \mbox{as $z\to\infty$.}

487: \end{equation}

488: \end{enumerate}

489: 

490: Following the idea of the steepest descent method of Deift and Zhou, 

491: we now replace the Riemann-Hilbert problem for $T(z)$ with 

492: an equivalent one on a system of 3 contours where some of the 

493: jump matrix elements are exponentially small.

494: Divide the complex plane into 3 regions as shown on Figure 1

495: (the contours $\Si_{1,3}$ lie sufficiently close to $\Si_2\equiv\Si$

496: for $f(z)$ to remain nonzero and analytic in regions 1 and 2)

497: \begin{figure}

498: \centerline{\psfig{file=contour.ps,width=4.0in,angle=-90}}

499: \vspace{0.1cm}

500: \caption{

501: The contour for Riemann-Hilbert problems.}

502: \label{fig1}

503: \end{figure}

504: and define the matrix-valued function $S(z)$ by the formulas:

505: 

506: \noindent

507: 1) in region 1

508: \be

509: S(z)=T(z)\pmatrix{1&0\cr -{z^n\over f(z)\psi(z)^{2n}}& 1},

510: \ee 

511: 2) in region 2

512: \be

513: S(z)=T(z)\pmatrix{1&0\cr {z^n\over f(z)\psi(z)^{2n}}& 1},

514: \ee

515: 3) in region 3

516: \be

517: S(z)=T(z).

518: \ee

519: 

520: 

521: The condition (c) in the problem for $S(z)$ is the same as

522: for $T(z)$, the conditions (a), (b) and (d) are different. Namely,

523: \begin{enumerate}

524: \item[(a,b)]

525: $S(z)$ is analytic in $\bbc\setminus(\Si_1\cup\Si_2\cup\Si_3)$ with the

526: following jump conditions on the contours:

527: \begin{equation}

528:             S_+(x) = S_-(x)

529:             \pmatrix{1&0\cr {x^n\over f(x)\psi(x)^{2n}}& 1},

530: \qquad\mbox{$x \in \Si_1\cup\Si_3$,}

531: \end{equation}

532: \begin{equation}

533: S_+(x) = S_-(x)

534:             \pmatrix{0&f(x)\cr

535:               -f(x)^{-1}&0},

536: \qquad\mbox{$x \in \Si_2\equiv\Si$,}

537: \end{equation}

538: \item[(c)]

539: as $z\to\infty$

540: \be

541:  S(z) = I+ O \left( \frac{1}{z} \right), 

542: \ee

543: \item[(d)]

544: near the end-points of the arc

545: \begin{equation}

546:             S(z) =  O\pmatrix{ 

547:                    \log|z-e^{\pm i\al}| & \log|z-e^{\pm i\al}| \cr

548:                    \log|z-e^{\pm i\al}| & \log|z-e^{\pm i\al}|},\la{scS}

549:         \end{equation}

550: as $z \to e^{\pm i\al}$, $z \in \bbc \setminus (\Si_1\cup\Si_2\cup\Si_3)$.

551: \end{enumerate}

552: Recalling (\ref{ineq}),

553: we see that, for $n$ large, for $x$ outside some neighborhoods of the

554: endpoints of the arc, the jump matrix on $\Si_1\cup\Si_3$ is uniformly

555: exponentially close to the identity. We therefore approximate the

556: function $S(z)$ with parametrices inside the mentioned neighborhoods

557: and in the outside region where we neglect the jumps on

558: $\Si_1\cup\Si_3$. The Riemann-Hilbert problems for the parametrices

559: can be solved and the solution closely resemble that in case of

560: polynomials on an interval \ci{kuimcl}.

561: 

562: The parametrix for the outside region is defined as the solution of the

563: following Riemann-Hilbert problem:

564: \begin{enumerate}

565:     \item[(a)]

566:         $N(z)$ is  analytic for $z\in\bbc \setminus\Si\equiv\Si_2$,

567:     \item[(b)]

568: with the jump condition on $\Si$

569: \be

570: N_+(x) = N_-(x)

571:             \pmatrix{0&f(x)\cr

572:               -f(x)^{-1}&0},

573: \qquad\mbox{$x \in \Si$},

574: \ee

575: \item[(c)]

576: and the following behavior at infinity

577: \be

578: N(z) = I+ O \left( \frac{1}{z} \right), 

579:      \qquad \mbox{as $z\to\infty$.}

580: \ee

581: \end{enumerate}

582: The solution $N(z)$ is found in the same way as in \ci{kuimcl}.

583: Consider the Szeg\H o function:

584: \be

585: {\cal D}(z)=\exp\left[{\sqrt{(z-e^{i\al})(z-e^{-i\al})}\over 2\pi i}\int_\Si

586: {\ln f(\xi)\over \sqrt{(\xi-e^{i\al})(\xi-e^{-i\al})}}

587: {d\xi \over \xi-z}\right].\la{Sf}

588: \ee

589: This function is analytic outside the arc, and its boundary values

590: satisfy

591: \be

592: {\cal D}_+(x){\cal D}_-(x)=f(x),\qquad x\in\Si.

593: \ee

594: Denote

595: \be

596: {\cal D}_\infty=\lim_{z\to\infty} {\cal D}(z)=

597: \exp\left[-{1\over 2\pi i}\int_\Si{\ln f(\xi) d\xi \over 

598: \sqrt{(\xi-e^{i\al})(\xi-e^{-i\al})}}\right].\la{Sfinf}

599: \ee

600: 

601: Then the solution of the above Riemann-Hilbert problem is as follows:

602: \be

603: N(z)={1\over2}({\cal D}_\infty)^{\si_3}

604: \pmatrix{a+a^{-1}&-i(a-a^{-1})\cr i(a-a^{-1})& a+a^{-1}}

605: {\cal D}(z)^{-\si_3},

606: \qquad

607: a(z)=\left({z-e^{i\al}\over z-e^{-i\al}}\right)^{1/4},\la{N}

608: \ee

609: where the value of the root satisfies the condition $a(z)\to 1$ as

610: $z\to\infty$. Note that $\det N(z)=1$, which allows, in particular, to

611: write a simple expression for the inverse $N(z)^{-1}$.

612: 

613: Now consider a $\de$-neighborhood $U_\de$ of the point $e^{i\al}$,

614: small enough so that $f(z)$ is analytic and nonzero there.

615: The jump

616: matrices on $\Si_1$, $\Si_3$ are not close to the identity in this

617: region, so we need to construct a separate local parametrix.

618: We look for a matrix-valued function $P(z)$ which is analytic in

619: $U_\de$, satisfies the same jump relations on 

620: $(\Si_1\cup\Si_2\cup\Si_3)\cap U_\de$ as $S(z)$, has the same

621: behavior at $z=e^{i\al}$ as $S(z)$, and matches $N(z)$ at the

622: boundary:

623: \be

624: P(z)N(z)^{-1}=I+O\left({1\over n\sin(\al/2)}\right),\qquad z\in 

625: \partial U_\de\setminus (\Si_1\cup\Si_2\cup\Si_3),\la{match}

626: \ee

627: where $\rho=n\sin(\al/2)\to\infty$.

628: We seek $P(z)$ in the form

629: \be

630: P(z)=E(z)\hat P(z)\left({\psi(z)\over\sqrt{z}}\right)^{-n\si_3}

631: f(z)^{-\si_3/2},\la{P1}

632: \ee

633: where $E(z)$ is invertible and analytic in a neighborhood of $U_\de$.

634: The function $E(z)$ does not affect jump relations and will 

635: be chosen later so that $P(z)$ satisfies (\ref{match}).

636: Using the boundary-value property (\ref{bvf}), we obtain 

637: (cf. \ci{kuimcl}):

638: \be\eqalign{

639: \hat P(x)_+=\hat P(x)_-

640: \pmatrix{1&0\cr 1&1}, \qquad \mbox{$x\in (\Si_1\cup\Si_3)\cap U_\de,$}\\

641: \hat P(x)_+=\hat P(x)_-

642: \pmatrix{0&1\cr -1&0}, \qquad \mbox{$x\in \Si_2\cap U_\de.$}}\la{jtP}

643: \ee

644: Consider the function $\om(z)$ defined by the equation:

645: \be

646: e^{\sqrt{\om(z)}}={\psi(z)\over\sqrt{z}}\la{g}

647: \ee

648: Using (\ref{bvf}), we have for the boundary values of $\om$ on the arc:

649: \be

650: \sqrt{\om(x)}_+=\ln{\psi_+(x)\over\sqrt{x}}=-\sqrt{\om(x)}_-,

651: \ee

652: therefore $\om(z)$ is analytic in $U_\de$. 

653: For $z$ near $e^{i\al}$, we obtain uniformly for all $\al$

654: \be\eqalign{

655: {\psi(z)\over\sqrt{z}}=

656: 1+\left(\sqrt{i\sin(\al/2)\over\cos(\al/2)}e^{-i\al/2}

657: \sqrt{1+{u\over 2i\sin\al}}\sqrt{u}+

658: (e^{i\al/2}\cos(\al/2)^{-1}-1)e^{-i\al}{u\over 2}\right)(1+O(u)),

659: \\

660: u=z-e^{i\al}.}\la{psi_as}

661: \ee

662: If also $|u|<2\sin\al$, we get for the function $\om(z)$

663: the following (nonuniform) expansion at $z=e^{i\al}$:

664: \be

665: \om(z)={i\sin(\al/2)\over \cos

666:   (\al/2)}e^{-i\al}u\left(

667: 1-{1-2e^{-i\al}-2e^{-2i\al}\over 6i\sin\al}u+O(u^2)\right).\la{g_as1}

668: \ee

669: Denote

670: \be

671: \hat P(z)=Q(\zeta),\qquad \zeta=n^2 \om(z).

672: \ee

673: Now we reached a crucial moment.

674: The circle $\partial U_\de$ is transformed in

675: the $\zeta$ variable into a curve $\partial\hat U_\de$ 

676: whose minimal distance from zero is

677: $n^2 \min_{0\le t\le 2\pi}|\om(e^{i\al}+\de e^{it})|$.

678: In order to construct a solution, we need this distance to be large.

679: This is so for large $n$ if $\de$ and $\al$ 

680: are independent of $n$ (see (\ref{g_as1})).

681: In the general case of $2s/n\le\al\le\pi-\ep$, there exists some 

682: small $\al_0$ (depending on $f(z)$, $\ep$) such that we can assume 

683: \be

684: \de=\cases{\sin(\al_0/2),& for $\al_0\le\al\le\pi-\ep$\cr

685: \sin(\al/2),& for $2s/n\le\al\le\al_0$}.\la{del}

686: \ee

687: Putting $u=\de e^{it}$ in (\ref{psi_as}) and choosing $\al_0$

688: sufficiently small, we obtain after simple analysis:

689: \be\la{g_as2}

690: n^2\min_{0\le t\le 2\pi} |\om(e^{i\al}+\de e^{it})|\ge C(n\sin(\al/2))^2

691: \ee

692: for some constant $C$ which is larger than zero. We see that for large

693: $s$ this distance remains uniformly large (not less than of order

694: $s^2$) for any $\al\in[2s/n,\pi-\ep]$, where $n>s$.

695: 

696: 

697: In $\hat U_\de$ the image $\hat\Si_{1,2,3}$ of the cuts can be 

698: considered as 3 direct

699: lines emanating from zero. (The image of $\Si$ is a line, and the exact

700: form of $\Si_1$ and $\Si_3$ can be chosen at will.) 

701: The analytic matrix-valued function $Q(\zeta)$ which satisfies the jump

702: conditions (\ref{jtP}) on $\hat\Si_{1,2,3}$ and singularity conditions

703: (\ref{scS}) at $\zeta=0$ was constructed in \ci{kuimcl}.

704: Namely, we have in $\hat U_\de$ (the regions 1, 2, and 3 correspond to the

705: $\zeta$-variable images of the regions in Figure 1) the following

706: expressions in terms of modified Bessel and Hankel functions (see,

707: e.g., \ci{Abr}):

708: 

709: 1) region 1

710: \begin{equation}\label{Q1}

711:     Q(\zeta) ={1\over 2}

712:     \pmatrix{

713:         H_{0}^{(1)}(e^{-i\pi/2}\zeta^{1/2}) &

714:         H_{0}^{(2)}(e^{-i\pi/2}\zeta^{1/2}) \cr

715:         \pi \zeta^{1/2} \left(H_{0}^{(1)}\right)'(e^{-i\pi/2}\zeta^{1/2}) &

716:         \pi \zeta^{1/2} \left(H_{0}^{(2)}\right)'(e^{-i\pi/2}\zeta^{1/2})

717:     },

718: \end{equation}

719: 

720: 2) region 2 

721: \begin{equation}\label{Q2}

722:     Q(\zeta) ={1\over 2}

723:     \pmatrix{

724:         H_{0}^{(2)}(e^{i\pi/2}\zeta^{1/2}) &

725:         -H_{0}^{(1)}(e^{i\pi/2}\zeta^{1/2}) \cr

726:         -\pi \zeta^{1/2} \left(H_{0}^{(2)}\right)'(e^{i\pi/2}\zeta^{1/2}) &

727:         \pi \zeta^{1/2} \left(H_{0}^{(1)}\right)'(e^{i\pi/2}\zeta^{1/2})

728:     },

729: \end{equation}

730: 

731: 3) region 3

732: \begin{equation}\label{Q3}

733:     Q(\zeta) =

734:     \pmatrix{

735:      I_{0} (\zeta^{1/2}) & \frac{i}{\pi} K_{0}(

736:         \zeta^{1/2})\cr

737:      \pi i \zeta^{1/2} I_{0}'(\zeta^{1/2}) & 

738:      -\zeta^{1/2} K_{0}'(\zeta^{1/2})},

739: \end{equation}

740: where $-\pi<\arg(\zeta)<\pi$.

741: 

742: We now have to choose $E(z)$ so that the matching condition

743: (\ref{match}) is satisfied. For that we can use the first term in 

744: the asymptotic expansion of Bessel and Hankel functions

745: for large $\zeta$. 

746: The expansion for $Q(\ze)$ is the same in all the 3 regions. We can

747: write down an arbitrary number of terms. Below we shall make use only

748: of the first three. We have:

749: \be\eqalign{

750: Q(\ze)={1\over\sqrt{2}}(\pi\sqrt\ze)^{-\si_3/2}\pmatrix{1&i\cr i&1}

751: \left[I+{1\over 8\sqrt\ze}\pmatrix{-1&-2i\cr -2i&1}-\right.\\ \left.

752: {3\over 2^7\ze}\pmatrix{1&-4i\cr 4i&1}+

753: O(\ze^{-3/2})\right]e^{\sqrt\ze\si_3}}\la{Qas}

754: \ee

755: uniformly on the boundary of $\hat U_\de$.

756: We now define $E(z)$ as follows:

757: \be

758:  E(z)={1\over\sqrt{2}}N(z)f(z)^{\si_3/2}

759: \pmatrix{1&-i\cr-i&1}(\pi n\sqrt{\om(z)})^{\si_3/2}.\la{E}

760: \ee

761: We verify exactly as in \ci{kuimcl} that it is an analytic function 

762: in $U_\de$.

763: Using (\ref{P1}), (\ref{g}), and (\ref{g_as2})

764: (and estimating $N(z)$ as below)

765: we see that the matching 

766: condition (\ref{match}) is now satisfied.

767: Thus we have

768: \be

769: P(z)=E(z)Q(n^2 \om(z))\left({\psi(z)\over\sqrt{z}}\right)^{-n\si_3}

770: f(z)^{-\si_3/2}.\la{P}

771: \ee

772: Solution in the neighborhood $\ti U_\de$ of $e^{i(2\pi-\al)}$ 

773: is similar (but

774: note the reversed direction of the contours). We have there:

775: \be\eqalign{

776: \ti P(z)=\ti E(z)\si_3 Q(n^2 \om(z))\si_3

777: \left({\psi(z)\over\sqrt{z}}\right)^{-n\si_3}

778: f(z)^{-\si_3/2},\\

779: \ti E(z)={1\over\sqrt{2}}N(z)f(z)^{\si_3/2}

780: \pmatrix{1&i\cr i&1}(\pi n\sqrt{\om(z)})^{\si_3/2}.}\la{P2}

781: \ee

782: We are now ready for the last transformation of the Riemann-Hilbert

783: problem. Let

784: \be\eqalign{

785: R(z)=S(z)N^{-1}(z),\qquad

786: z\in\bbc\setminus(\overline{U_\de\cup\ti U_\de}\cup\Si_{1,2,3}),\\

787: R(z)=S(z)P^{-1}(z),\qquad

788: z\in U_\de\setminus\Si_{1,2,3},\\

789: R(z)=S(z)\ti P^{-1}(z),\qquad

790: z\in \ti U_\de\setminus\Si_{1,2,3}.}\la{Rb}

791: \ee

792: It is easy to see that this function has jumps only on 

793: $\partial U_\de$, $\partial\ti U_\de$, and parts of $\Si_1$, and

794: $\Si_3$ lying outside of the neighborhoods $U_\de$, $\ti U_\de$ (we

795: denote these parts $\Si_{1,3}^\mathrm{out}$).

796: Namely,

797: \be\eqalign{

798: R_+(x)=R_-(x)N_-(x)\pmatrix{1&0\cr {x^n\over f(x)\psi(x)^{2n}}&1}

799: N_-(x)^{-1},\qquad x\in\Si_{1,3}^{\mathrm out},\\

800: R_+(x)=R_-(x)P(x)N(x)^{-1},\qquad x\in \partial U_\de,\\

801: R_+(x)=R_-(x)\ti P(x)N(x)^{-1},\qquad x\in \partial\ti U_\de.}\la{Rj}

802: \ee

803: As is easy to verify, the

804: matrix elements of $N(x)$ and $N(z)^{-1}$ remain bounded 

805: for $|x-e^{\pm  i\al}|\ge\de\ge\sin(s/n)$ and all $\al$.

806: For example,

807: \be\la{a4}

808: |a^4|=(1+8\ga\sin t+16\ga^2)^{-1/2}\le(1-4\ga)^{-1}

809: \qquad\mathrm{for}\qquad z=e^{i\al}+\sin(s/n)e^{it}\in\partial U_{\sin(s/n)}.

810: \ee

811: (For evaluation of ${\cal D}(z)$ near $e^{\pm i\al}$ see a similar

812: calculation in Lemma 6.4 of \ci{kuimcl}.)

813: Therefore the jump matrix on $\Si_{1,3}^\mathrm{out}$

814: can be uniformly (both in $z$ and $\al$)

815: estimated by (\ref{g},\ref{g_as2}) as

816: $I+O(\exp{(-C_1 n\sin(\al/2))})$, where $C_1$

817: is a positive constant. Note that in the case of a fixed arc, it is

818: sufficient to use (\ref{ineq}). 

819:  

820: 

821: The jump matrices on $\partial U_\de\cup\partial\ti U_\de$

822: admit an asymptotic expansion in powers of $1/\sqrt{\ze}$ (which turns

823: into expansion in powers of

824: $1/n$ for a fixed arc or $1/s$ for a varying arc case). 

825: First, (\ref{Qas}), (\ref{E}), and (\ref{P}) yield

826: \be\eqalign{

827: P(z)N(z)^{-1}=

828: I+N(z)f(z)^{\si_3/2}\left[

829: {1\over 8\sqrt\ze}\pmatrix{-1&-2i\cr -2i& 1}-

830: {3\over 2^7\ze}\pmatrix{1&-4i\cr 4i& 1}+\right.\\ \left.

831: O(\ze^{-3/2})\right]

832: f(z)^{-\si_3/2}N(z)^{-1}=

833: I+\De_1+\De_2+O(\ze^{-3/2}),\qquad z\in \partial U_\de,}\la{lastjump}

834: \ee

835: where $\De_1$ and $\De_2$ denote the terms with $\sqrt\ze$ and $\ze$,

836: respectively (the remainder term will be justified below).

837: We write them down explicitly for the case of $f(z)=1$

838: needed later on:

839: \be

840: \De_1={1\over 2^4\sqrt\ze}\pmatrix{

841: -(3a^2-a^{-2})& -i(3a^2+a^{-2})\cr

842: -i(3a^2+a^{-2})& 3a^2-a^{-2}},

843: \qquad

844: \De_2=-{3\over 2^7\ze}\pmatrix{1& -4i\cr 4i& 1},\la{de12}

845: \ee

846: where $f=1$. The functions $\De_1(z)$

847: and $\De_2(z)$ for $z\in\partial\ti U_\de$ are given by the same

848: expressions with $a$ exchanged with $1/a$ and $i$ replaced by $-i$.

849:  

850: Using the expansion for Bessel functions, we can write a general term

851: $\De_j$ in (\ref{lastjump}) which is of order $1/\ze^{j/2}$. 

852: Indeed,

853: apart from the prefactor with $\ze^{j/2}$, the matrix elements

854: of $\De_j$ are obviously $O(1)$ as $\ze\to\infty$ for a fixed arc.

855: It is also true in the general case, because of 

856: the remark after equation (\ref{Rj}).

857: Now it is clear that (\ref{lastjump}) is an asymptotic expansion in $\ze$.

858: 

859: 

860: Since for $z\in\partial U_\de$ by (\ref{g_as1},\ref{g_as2})

861: \be

862: \sqrt\ze=n\sqrt{\om(z)}=O(n\sin(\al/2))=

863: \cases{O(n),& $\al$ fixed\cr

864: O(s),& $\al=2s/n$,} 

865: \ee

866: the component $\De_j$ on this boundary is of order $1/n^j$ (fixed arc)

867: and $1/s^j$ (varying arc), and 

868: the remainder term in (\ref{lastjump}) is

869: uniform for all $\al$ between a fixed positive value and $2s/n$,

870: all $n>s$, and all $s$ larger than some $s_0$. (As we shall see this

871: uniformity persists for the remainder in the asymptotics of our polynomials.)

872: We show as in \ci{kuimcl}

873: that $\De_j(z)$ is an analytic function in $U_\de$ with a 

874: pole at $e^{i\al}$ of order less than or equal to $[(j+1)/2]$.

875: The same reasoning also holds for the neighborhood $\ti U_\de$.

876: We shall denote the components of the jump matrix there by the same

877: symbols $\De_j(z)$.

878: 

879: If $R(z)$ is known, we can trace the sequence $Y\mapsto T\mapsto

880: S\mapsto R$ backwards, and obtain an expression for the polynomials. 

881:  

882: We look for $R(z)$ asymptotically in the 

883: form $R(z)\sim R_0(z)+R_1(z)+R_2(z)+\cdots$,

884: where $R_j(z)$ is of the same order as $\De_j$ (in our case, of order

885: $(n\sin(\al/2))^{-j}$).

886: For more discussion and

887: justification of this expansion see \ci{DK1,kuimcl}. More precisely,

888: it can be shown as in Theorems 7.8--7.10 of \ci{DK1} that

889: for any $k\ge1$

890: \be

891: R(z)=I+\sum_{j=1}^{k-1}R_j(z)+O((n\sin(\al/2))^{-k})\la{R}

892: \ee

893: uniformly for all $z$ if the arc is fixed, and for $z$ outside a

894: neighborhood of $z=1$ if the arc is varying ($\al=2s/n$).

895: The proof of this expansion in the general case $2s/n\le\al<\pi$

896: for $z$ close to

897: $1$ requires a special argument since the contours then are close to $z$

898: and can not be trivially deformed. Such a proof will be given later on

899: for the case of $f=1$, $z\in U_\de$, we need below.

900: The argument in the general case is similar.

901: Moreover, it follows directly from the proofs that the remainder 

902: term in (\ref{R}) has the same uniformity property

903: as that in (\ref{lastjump}).

904: 

905: 

906: 

907: Substituting this asymptotic expansion

908: into (\ref{Rb}) and collecting the terms of the same order, we obtain:

909: \be\eqalign{

910: R_{0+}(x)+R_{1+}(x)+\cdots\sim

911: (R_{0-}(x)+R_{1-}(x)+\cdots)(I+\De_1(x)+\cdots),

912: \qquad x\in \partial U_\de\cup\partial\ti U_\de.\\

913: R_{0+}(x)=R_{0-}(x)\quad\Rightarrow\quad R_0(z)=I,\\

914: R_{1+}(x)-R_{1-}(x)=\De_1(x),\\

915: R_{2+}(x)-R_{2-}(x)=R_{1-}(x)\De_1(x)+\De_2(x),\\

916: R_{k+}(x)-R_{k-}(x)=\sum_{j=1}^k R_{k-j,-}(x)\De_j(x),\qquad k=1,2,\dots}

917: \ee

918: The main term in the asymptotics of polynomials 

919: is given therefore by

920: the parametrices at the appropriate points $z$.

921: The expressions for $R_k(z)$ follow from the Sokhotsky-Plemelj

922: formulas:

923: \be

924: R_1(z)={1\over 2\pi i}\int_{\partial U_\de\cup\partial\ti U_\de}

925: {\De_1(x)dx\over x-z},\qquad

926: R_2(z)={1\over 2\pi i}\int_{\partial U_\de\cup\partial\ti U_\de}

927: {R_{1-}(x)\De_1(x)+\De_2(x)\over x-z}dx,\quad\dots\la{plem}

928: \ee

929: Note that the contours are traversed in the negative direction (see

930: Figure 1).

931: 

932: Following \ci{kuimcl}, we can also obtain the expressions for $R_j(z)$ 

933: in a different way.

934: As mentioned,  $\De_1(z)$ is analytic in $U_\de\cup\ti

935: U_\de$ with simple poles at the end-points of the arc. Thus,

936: \be

937: \De_1(z)=\frac{A^{(1)}}{z-e^{i\al}}+O(1),\quad\mbox{as } z\to e^{i\al}, 

938: \qquad

939:     \De_1(z)=\frac{B^{(1)}}{z-e^{-i\al}}+O(1),\quad\mbox{as } z\to

940:     e^{-i\al},

941: \ee 

942: where the constant matrices $A^{(1)}$ and $B^{(1)}$ are obtained by expanding

943: various functions in (\ref{lastjump}) at $z=e^{i\al}$ and 

944: $z=e^{-i\al}$. It is easy to verify directly that 

945: the Riemann-Hilbert problem for $R_1(z)$ has the solution: 

946: \be

947: R_1(z)=

948: \cases{\frac{A^{(1)}}{z-e^{i\al}}+\frac{B^{(1)}}{z-e^{-i\al}},&

949: for $z\in\bbc\setminus(\overline{U_\de\cup\ti U_\de})$\cr

950: \frac{A^{(1)}}{z-e^{i\al}}+\frac{B^{(1)}}{z-e^{-i\al}}-\De_1(z),&

951: for $z\in U_\de\cup\ti U_\de$.}\la{R1}

952: \ee

953: 

954: Expanding the functions in (\ref{lastjump}), we obtain:

955: \be

956: A^{(1)}={\cos(\al/2)\over 8n}\pmatrix{1&-i{\cal D}_\infty^2\cr

957: -i{\cal D}_\infty^{-2}& -1}e^{i\al/2},\qquad

958: B^{(1)}=\overline {A^{(1)}},\la{A1B1}

959: \ee

960: where $\overline {M}$ means complex conjugation applied

961: to every matrix element of $M$.

962: The general term $R_k(z)$ is obtained similarly provided we have 

963: the expressions for $R_j(z)$, $j=1,2,\dots,k-1$.  

964: Since $R_j(z)$ are analytic in $U_\de\cup\ti U_\de$

965: and $\De_j(z)$ have poles at $e^{\pm i\al}$ of order 

966: at most $[(j+1)/2]$, we see that

967: \be

968: \sum_{j=1}^k R_{k-j,-}(z)\De_j(z)=

969: \frac{A^{(k)}_p}{(z-e^{i\al})^p}+

970: \frac{A^{(k)}_{p-1}}{(z-e^{i\al})^{p-1}}+\cdots+

971: \frac{A^{(k)}_1}{z-e^{i\al}}+O(1),\quad \mbox{as } z\to e^{i\al},\la{i1}

972: \ee

973: where $p=[(k+1)/2]$, and similar expressions hold with matrices $A$

974: replaced with some matrices $B$ in a neighborhood of $e^{-i\al}$.

975: Then

976: \be

977: R_k(z)=

978: \cases{

979: \sum_{j=1}^p\left(\frac{A^{(k)}_j}{(z-e^{i\al})^j}+

980: \frac{B^{(k)}_j}{(z-e^{-i\al})^j}\right),&

981: for $z\in\bbc\setminus(\overline{U_\de\cup\ti U_\de})$\cr

982: \sum_{j=1}^p\left(\frac{A^{(k)}_j}{(z-e^{i\al})^j}+

983: \frac{B^{(k)}_j}{(z-e^{-i\al})^j}\right)-

984: \sum_{j=1}^k R_{k-j,-}(z)\De_j(z),&

985: for $z\in U_\de\cup\ti U_\de$}.\la{i2}

986: \ee

987: 

988: Recalling the definitions of $R$, $S$, and $T$, we finally get

989: \be

990: Y(z)=\ga^{n\si_3}R(z)M(z)\psi(z)^{n\si_3},\qquad

991: R(z)\sim I+R_1(z)+R_2(z)+\cdots,

992: \ee

993: where for $z$ restricted to region 3 $M(z)=N(z)$, $P(z)$, or $\ti P(z)$ if 

994: $z\in\bbc\setminus(\overline{U_\de\cup\ti U_\de})$, $U_\de$, or $\ti U_\de$,

995: respectively (the expressions in regions 1 and 2 can also be

996: readily written).

997: Therefore, by (\ref{RHM}),

998: \be

999: \chi_n^{-1}\phi_n(z)=Y_{11}(z)=\ga^n\psi^n(z) 

1000: [R_{11}(z)M_{11}(z)+R_{12}(z)M_{21}(z)].\la{aspol}

1001: \ee

1002: Furthermore,

1003: \be

1004: \chi_{n-1}^2=Y_{21}(0)=\ga^{-2n}

1005: [R_{21}(0)N_{11}(0)+R_{22}(0)N_{21}(0)].\la{aschi}

1006: \ee

1007: Since we know the expressions for $M(z)$ and can obtain $R(z)$ with an

1008: arbitrary precision, the last 2 equations give an implicit solution

1009: for asymptotics of polynomials $\phi_n(z)$ and their leading

1010: coefficients. These are the asymptotic series in the

1011: inverse powers of $n\sin(\al/2)$. The error

1012: after $k$ terms is $O(n\sin(\al/2))^{-k-1}$

1013: and remains uniform for all $\al$, $s$, and $n$, provided

1014: $\al\in[2s/n,\pi-\ep]$, $n>s$, $s>s_0$. 

1015: 

1016: As an example, we give below the first two terms in the asymptotics of

1017: $\phi_n(z)$ valid for $z$ outside a fixed arbitrary small

1018: $\epsilon$-neighborhood of the arc $\Si$:

1019: \be

1020: \chi_n^{-1}\phi_n(z)\sim \ga^n\psi^n(z){{\cal D}_\infty\over {\cal

1021:   D}(z)}\left[{a(z)+a(z)^{-1} \over 2}+{\ga\over 8n}

1022: \left({a(z)e^{i\al/2}\over z-e^{i\al}}+

1023: {a(z)^{-1}e^{-i\al/2}\over z-e^{-i\al}}\right)\right],\la{p1}

1024: \ee

1025: \be

1026: \chi_{n-1}^2\sim {\ga^{-2n+1}\over {\cal D}(0){\cal D}_\infty}

1027: \left[1+{1\over 4n}\right],\la{c1}

1028: \ee

1029: where all the quantities are defined as above (see (\ref{psi},

1030: \ref{Sf}, \ref{Sfinf}, \ref{N})).

1031: 

1032: 

1033: 

1034: 

1035: 

1036: We now give an explicit solution for the first 3 terms in the 

1037: asymptotics of $\phi_n(z)$

1038: and its first derivative at an end-point of the arc for the weight $f(z)=1$. 

1039: 

1040: Recall that we still have to prove the remainder term in the expansion (\ref{R})

1041: for $z\in U_\de$, $2s/n\le\al<\al_0$.

1042: Since $f(z)=1$ we see that ${\cal D}(z)=1$.

1043: Each matrix element of $\De_k(z)$ can be written as 

1044: $(\al_k a^2+\beta_k a^{-2}+\ga_k)\ze^{-k/2}$, where $\al_k$, $\beta_k$,

1045: and $\ga_k$ are independent of $\al$. In a neighborhood of $e^{i\al}$,

1046: we have the series:

1047: \be\eqalign{

1048: \ze^{-k/2}=(n^2 u\sin\al)^{-k/2}\sum_{j=0}^\infty b_j(k,\al)

1049: \left({u\over\sin\al}\right)^j,\qquad

1050: a(z)^2=\sqrt{u\over\sin\al}

1051: \sum_{j=0}^\infty c_j \left({u\over\sin\al}\right)^j,\\

1052: |u|<\sin(\al/2),\qquad u=z-e^{i\al},

1053: \qquad 0<|c_j|\le1,\quad j\ge0,}

1054: \ee

1055: where $b_j(k,\al)$ are bounded functions of $\al$.

1056: Since $\De_k(z)$ is single-valued in $U_\de$, its matrix

1057: elements do not contain terms with $\sqrt{u}$. Hence, 

1058: $\al_k,\beta_k=0$ if $k$ is even, and $\ga_k=0$ if $k$ is odd.

1059: We have in the same neighborhood:

1060: \be

1061: \De_k(z)={1\over (n\sin\al)^k}\sum_{j=-[(k+1)/2]}^\infty C_j(k,\al)

1062: {u^j\over \sin^{j}\al}.

1063: \ee

1064: Now using (\ref{i1}) and the second formula in (\ref{i2}), 

1065: it is easy to show by induction that

1066: \be

1067: R_k(z)={1\over (n\sin\al)^k}\sum_{j=0}^\infty \hat C_j(k,\al){u^j\over

1068:   \sin^{j}\al}.\la{Rser}

1069: \ee

1070: The matrix elements of $C_j$ and $\hat C_j$ are bounded functions of $\al$.

1071: Considering the remainder term in the asymptotic 

1072: expansion of Bessel functions, we see that

1073: $R(z)-I-\sum_{j=1}^{k-1}R_j(z)$

1074: is given by the same series (\ref{Rser}) with different 

1075: matrices $\hat C_j(k,\al)$ (but also bounded in $\al$). 

1076: By analyticity, these series and their derivative w.r.t. $u$ 

1077: converge in $U_\de$. This is also true for $\al>\al_0$.

1078: Thus,

1079: \be\eqalign{

1080: R(z)=I+\sum_{j=1}^{k-1}R_j(z)+O((n\sin(\al/2))^{-k}),\\

1081: {d\over dz}R(z)=\sum_{j=1}^{k-1}{d\over dz}R_j(z)+

1082: O(n^{-k}\sin(\al/2)^{-k-1}), \qquad z\in U_\de,}\la{RRp}

1083: \ee

1084: where the remainder terms are uniform 

1085: for $z\in U_\de$, $2s/n<\al\le\pi-\ep$, $n>s$, $s>s_0$.

1086: 

1087: 

1088: Using 

1089: (\ref{aspol},\ref{P},\ref{E},\ref{Q3}), 

1090: we obtain for $z$ in the intersection of $ U_\de$ and

1091: region 3 an asymptotic equivalence: 

1092: \be\eqalign{

1093: \chi_n^{-1}\phi_n(z)\sim\ga^n z^{n/2}\sqrt{{\pi n\over

1094:     2}\sqrt{\om(z)}}

1095: \left[a^{-1}I_0+aI_0'+

1096: \sum_{j=1}^\infty\left\{

1097: (R_{j,\,11}(z)-iR_{j,\,12}(z))a^{-1}I_0+\right.\right. \\ \left.\left.

1098: (R_{j,\,11}(z)+iR_{j,\,12}(z))a I_0'\right\}\right],}\la{polas}

1099: \ee

1100: where the Bessel functions $I_0$ and $I_0'=I_1$ are taken at 

1101: $n\sqrt{\om(z)}$. We now estimate the ``$R$'' terms and some of their 

1102: derivatives at the point $e^{i\al}$. 

1103: From (\ref{plem}) and (\ref{de12}) we get:

1104: \[\eqalign{

1105: R_{1,\,11}(e^{i\al})-iR_{1,\,12}(e^{i\al})=

1106: {1\over 2\pi i}\int_{\partial U_\de\cup\partial\ti U_\de}

1107: {\De_{1,\,11}(x)-i\De_{1,\,12}(x)\over x-e^{i\al}}dx=\\

1108: {1\over 2\pi i}\oint {3a^2\over 8\sqrt{\ze}}{du\over u}+

1109: {1\over 2i\sin\al}\mathrm{Res}_{x=e^{-i\al}}{a^2\over 8\sqrt{\ze}},}

1110: \]

1111: where $u=z-e^{i\al}=\epsilon e^{it}$ describes a circle of a small

1112: radius $\epsilon$ in the positive direction. 

1113: The expansion of $\ze$ near $e^{i\al}$ is given by

1114: (\ref{g_as1}), and that near $e^{-i\al}$ is obtained from it by

1115: changing the sign of $\alpha$. Expanding also $a^2(z)$ near $e^{\pm

1116:   i\al}$ and calculating residues, we obtain 

1117: \be

1118: R_{1,\,11}(e^{i\al})-iR_{1,\,12}(e^{i\al})=

1119: {3 e^{i\al/2}+e^{-i\al/2} \over 16i n\sin(\al/2)}.

1120: \ee

1121: Similarly, we calculate

1122: \be

1123: R_{1,\,11}(e^{i\al})+iR_{1,\,12}(e^{i\al})=

1124: e^{-i\al/2}{1+e^{-i\al}-2e^{i\al}\over 3\cdot 16i n\sin(\al/2)}.

1125: \ee

1126: 

1127: 

1128: Now differentiating (\ref{plem}) w.r.t. $z$, we get

1129: \[

1130: R'_1(z)={1\over 2\pi i}\int_{\partial U_\de\cup\partial\ti U_\de}

1131: {\De_1(x)\over (x-z)^2}dx.

1132: \]

1133: From here we obtain as above

1134: \be

1135: R'_{1,\,11}(e^{i\al})-iR'_{1,\,12}(e^{i\al})=

1136: {e^{-i\al/2}\cos(\al/2)\over 16 n \sin^2(\al/2)}.

1137: \ee

1138: 

1139: To estimate $R_2(z)$ using (\ref{plem}), we need to know 

1140: $L(z)=R_1(z)\De_1(z)+\De_2(z)$ in the neighborhoods of $e^{\pm

1141:   i\al}$. Here $\De_1(z)$, $\De_2(z)$ are given by (\ref{de12}) and 

1142: for $R_1(z)$ we use the second formula in (\ref{R1}).

1143: Then we obtain in the same way as above:

1144: \be\eqalign{

1145: R_{2,\,11}(e^{i\al})-iR_{2,\,12}(e^{i\al})=

1146: {-6+7e^{i\al}-17e^{-i\al}\over 2^9 n^2\sin^2(\al/2)};\\

1147: R_{2,\,11}(e^{i\al})+iR_{2,\,12}(e^{i\al})=

1148: {16-9e^{i\al}+43e^{-i\al}-2e^{-2i\al}\over 3\cdot 2^9

1149:   n^2\sin^2(\al/2)}.}

1150: \ee

1151: 

1152: Substituting these expressions into 

1153: (\ref{polas}), expanding $a(z)$, Bessel functions, and $\om(z)$ at

1154: $e^{i\al}$ (see (\ref{g_as1})), and recalling (\ref{RRp}),

1155: we obtain (\ref{phin}).

1156: A simple calculation of (\ref{aschi}) yields (\ref{chin}). 

1157: Taking the derivative of (\ref{polas}) at $z=e^{i\al}$ and using

1158: the expressions for $R(e^{i\al})$ and $R'(e^{i\al})$ we complete the

1159: proof of Theorem 2. $\Box$

1160:  

1161: 

1162: 

1163: 

1164: 

1165: 

1166: 

1167: 

1168: 

1169: 

1170: 

1171: 

1172: 

1173: 

1174: 

1175: 

1176: 

1177: \section{Proof of Theorem 1}

1178: 

1179: Consider the following weight function on the unit circle:

1180: \[

1181: f_\al(\th)=\cases{1, & $\al\le\th\le2\pi-\al$,\cr 0, & otherwise.},

1182: \qquad 0<\al<\pi

1183: \]

1184: and the corresponding orthonormal polynomials 

1185: $\phi_k(z,\al)=\chi_k z^k+b_{k-1}z^{k-1}+\dots+b_0$

1186: satisfying (\ref{phi1}).

1187: Since $f_\al(\th)=f_\al(2\pi-\th)$, the coefficients of the

1188: polynomials $\phi_k(z,\al)$ are real. 

1189: 

1190: Associated with $f_\al(\th)$ is 

1191: an $(n+1)\times (n+1)$ Toeplitz matrix $T_n(\al)$

1192: whose matrix elements are as follows:

1193: \[

1194: (T_n(\al))_{jk}=

1195: {1\over 2\pi}\int_0^{2\pi}e^{-i(j-k)\theta}f_{\al}(\theta)d\theta=

1196: \cases{1-\al/\pi, & $j=k$,\cr -{\sin(\al(j-k))\over\pi(j-k)},& $j\neq

1197:   k$},

1198: \qquad j,k=0,1,\dots,n.

1199: \]

1200: Putting $\al=2s/n$ and taking the limit $n\to\infty$, we easily obtain

1201: \be

1202: \De(s)=\lim_{n\to\infty}\det T_{n-1}\left({2s\over n}\right).\la{tf}

1203: \ee

1204: If $\al$ is fixed, the large $n$ asymptotics of $\det T_n(\al)$ were

1205: obtained by Widom \ci{Warc} (see also \ci{asto} for an alternative

1206: derivation), namely,

1207: \be

1208: \det T_{n-1}(\al)=

1209:  \cos^{n^2}(\al/2)\left(n\sin{\al\over

1210:   2}\right)^{-1/4}2^{1/12}e^{3\zeta'(-1)}(1+o(1)),\qquad 0<\al<\pi.

1211: \la{arc}

1212: \ee

1213: An idea of Dyson \ci{Dyson} was to put $\al=2s/n$ in these

1214: asymptotics. Then, in view of (\ref{tf}), one formally obtains

1215: the first 3 terms of (\ref{exp}). 

1216: However, as noted in \ci{Dyson}, since the remainder 

1217: term in (\ref{arc}) is unknown, we cannot say if this expansion 

1218: is uniform in $\al$ and therefore cannot justify such a limit.

1219:  

1220: Recently, Deift \ci{D} found a formula which connects the determinants

1221: of two different Toeplitz matrices. A proof of it we need below was

1222: noticed by Simon. We shall

1223: use this formula together with equation (\ref{arc}) for  

1224: $\det T_{n-1}(\al)$ ($\al$ fixed) to obtain an expression 

1225: for $\det T_{n-1}(2s/n)$.

1226: The variant of Deift's formula we shall use is the following:

1227: \be

1228: {d\over d\al}\ln\det T_{n-1}(\al)=

1229: -\sum_{k=0}^{n-1}{1\over 2\pi}\int_\al^{2\pi-\al}

1230: {d\over d\al}|\phi_k(e^{i\th},\al)|^2 d\th,\qquad n=1,2,\dots\la{2det}

1231: \ee

1232: Indeed, it is well-known (e.g., \ci{Sz}) that the Toeplitz determinant has the

1233: following representation in terms of the leading coefficients of

1234: $\phi_n(z,\al)$:

1235: \be

1236: \det T_{n-1}(\al)=\prod_{k=0}^{n-1}\chi_k^{-2},\qquad n=1,2,\dots

1237: \ee

1238: Therefore 

1239: \[

1240: {d\over d\al}\ln\det T_{n-1}(\al)=-2\sum_{k=0}^{n-1}{\chi_k'(\al)\over

1241: \chi_k(\al)}.

1242: \]

1243: On the other hand, the orthogonality of our polynomials implies

1244: \[\eqalign{

1245: -\sum_{k=0}^{n-1}{1\over 2\pi}\int_\al^{2\pi-\al}

1246: {d\over d\al}|\phi_k(e^{i\th},\al)|^2 d\th=\\

1247: -\sum_{k=0}^{n-1}{1\over 2\pi}\int_\al^{2\pi-\al}

1248: \left(\phi_k(e^{i\th},\al)(\chi_k'(\al)e^{-ik\th}+b_{k-1}'(\al)e^{-i(k-1)\th}

1249: +\dots)+\mathrm{\, c.c. \,}\right) d\th=

1250: -2\sum_{k=0}^{n-1}{\chi_k'(\al)\over

1251: \chi_k(\al)},}

1252: \]

1253: and (\ref{2det}) is obtained.

1254: A corollary of it is the following 

1255: 

1256: \bl

1257: Let $\phi_n(z,\al)$ and $T_{n-1}(\al)$ be defined as at the beginning

1258: of the section, and

1259: $\phi_n'(z,\al)=(d/dz)\phi_n(z,\al)$. Then for any $n=1,2,\dots$ 

1260: \be

1261: {d\over d\al}\ln\det T_{n-1}(\al)=

1262: {n\over\pi}|\phi_n(e^{i\al},\al)|^2-{1\over\pi}\left\{

1263: \phi_n(e^{-i\al},\al)e^{i\al}\phi_n'(e^{i\al},\al)+ \mathrm{

1264:   c.c.}\right\}.\la{2det2}

1265: \ee

1266: \el

1267: 

1268: \noindent 

1269: {\it Proof } From the identity

1270: \[

1271: {d\over d\al}\left(

1272: \sum_{k=0}^{n-1}{1\over 2\pi}\int_\al^{2\pi-\al}

1273: |\phi_k(e^{i\th},\al)|^2  d\th\right)={d\over d\al} n=0

1274: \]

1275: we obtain using (\ref{2det}),

1276: \[

1277: {d\over d\al}\ln\det T_{n-1}(\al)=- {1\over\pi}

1278: \sum_{k=0}^{n-1}|\phi_k(e^{i\al},\al)|^2.

1279: \]

1280: As is known (e.g., \ci{Sz}), any system of orthonormal

1281: polynomials on the circle

1282: satisfies an analogue of the Christoffel formula which we write for $x$

1283: and $y$ {\it on} the unit circle in the form:

1284: \[

1285: \sum_{k=0}^{n-1}\phi_k(x)\overline{\phi_k(y)}=

1286: {\phi_n^*(x)\overline{\phi_n^*(y)}-\phi_n(x)\overline{\phi_n(y)}

1287: \over 1- x{\bar y}}=

1288: {\phi_n(x)\overline{(\phi_n(x)-\phi_n(y))}-

1289: \phi_n^*(x)\overline{(\phi_n^*(x)-\phi_n^*(y))}

1290: \over x\overline{(x-y)}},

1291: \]

1292: where $\phi_n^*(x)=x^n \overline{\phi_n(x)}$.

1293: Letting $y\to x$ along the unit circle and noting that

1294: $\overline{(d/dx)\phi_n^*(x)}=n x^{-n+1}\phi_n(x)-

1295:   x^{-n+2}\phi_n'(x)$, we obtain

1296: \be

1297: \sum_{k=0}^{n-1}|\phi_k(x)|^2=

1298: x\overline{\phi_n(x)}\phi_n'(x)+{1\over

1299:   x}\phi_n(x)\overline{\phi_n'(x)}-n|\phi_n(x)|^2,\qquad |x|=1,

1300: \ee

1301: which, after substituting $x=e^{i\al}$ and recalling that 

1302: the coefficients of our polynomials are real,

1303: completes the proof. $\Box$ 

1304: 

1305: Note \cite{DIZ,TW2} that the logarithmic derivative (\ref{2det2})

1306: satisfies a $\tau$-function version of Painlev\'e VI equation.

1307: 

1308: Now we want to integrate (\ref{2det2}) over $\al$ between $\al_1=2s/n$

1309: and any fixed $\al_2<\pi$ for large $n$ (for $\al_2$ asymptotics

1310: (\ref{arc}) are valid).

1311: We therefore need to know large $n$ asymptotics of $\phi_n(z,\al)$ 

1312: and its derivative at

1313: the endpoint $z=e^{i\al}$ of the orthogonality arc. 

1314: These are given by Theorem 2.

1315: Substituting them into (\ref{2det2}), we see that the

1316: term ${n\over\pi}|\phi_n(e^{i\al},\al)|^2$ cancels at once, and 

1317: purely imaginary terms in the rest of the expression also disappear

1318: (in particular, the terms with $r^3_-$ drop out).

1319: As a result we have:

1320: \be

1321: {d\over d\al}\ln\det T_{n-1}(\al)=

1322: -n^2{\sin(\al/2)\over 2\cos(\al/2)}-{\cos(\al/2)\over 8\sin(\al/2)}+

1323: O\left({1\over n\sin^2(\al/2)}\right),

1324: \ee 

1325: where the remainder term is uniform for $2s/n\le\al\le\pi-\ep$, $s>s_0$,

1326: $n>s$, $\ep>0$. 

1327: Integrating this expression over $\al$ from $2s/n$ to $\al_2$ and

1328: using Widom's asymptotics (\ref{arc}) for $\det T_{n-1}(\al_2)$, we

1329: get

1330: \be

1331: \ln\det T_{n-1}(2s/n)=

1332: n^2\ln\cos{s\over n}-{1\over 4}\ln n\sin{s\over n}+c_0+

1333: O\left({1\over n\sin(s/n)}\right)+o(1)

1334: \ee

1335: as $n\to\infty$ with the first remainder term turning into $O(1/s)$ valid

1336: for all $s>s_0$.

1337: By (\ref{tf}), this equation yields (\ref{exp}).

1338: %Note that the remainder term can be explicitly estimated using that

1339: %in the asymptotics of Bessel functions.

1340: $\Box$

1341: 

1342: \noindent

1343: {\bf Remark} The Riemann-Hilbert problem methods allow us to calculate

1344: asymptotics of orthogonal polynomials to arbitrary precision.

1345: Because of the integral identity (Theorem 2b and equation (4) of

1346: \ci{H}, see also \ci{Wsing}) that expresses Toeplitz determinants in

1347: terms of orthogonal polynomials,

1348: an arbitrary number of terms in asymptotics of

1349: Toeplitz determinants could also be computed.

1350: Thus one should be able to estimate the remainder term in (\ref{arc})

1351: and therefore give another proof of Theorem 1.

1352: 

1353: 

1354: 

1355: 

1356: \section{Acknowledgements} 

1357: I am grateful to P. Deift for helpful discussions. I also thank

1358: A. Its and H. Widom for useful comments and to 

1359: S. Jitomirskaya for inviting me to the AMS conference at Snowbird,

1360: where I heard \ci{D}. This work was partly supported by Sfb 288.

1361: 

1362: 

1363: 

1364: 

1365: 

1366: \begin{thebibliography}{99}

1367: 

1368: \bibitem{Mehta} M. L. Mehta: Random matrices. San Diego: Academic

1369:   1990

1370: 

1371: \bibitem{JMMS} M. Jimbo, T. Miwa, Y. M\^ori, and M. Sato: Density matrix of

1372:   an impenetrable Bose gas and the fifth Painlev\'e transcendent,

1373: Physica {\bf 1D}, 80--158 (1980) 

1374: 

1375: \bibitem{TW} C. A. Tracy, H. Widom: Introduction to random matrices

1376: in "Geometric and Quantum Aspects of Integrable Systems'' (Ed. G. F.

1377: Helminck), Lecture Notes in Physics {\bf424} (1993) 103-130, Springer

1378: 

1379: \bibitem{DIZ} P. Deift, A. Its, and X. Zhou: A Riemann-Hilbert

1380:   approach to asymptotic problems arising in the theory of random

1381:   matrix models, and also in the theory of integrable statistical

1382:   mechanics. Ann.Math {\bf 146}, 149--235 (1997)

1383: 

1384: \bibitem{CM} J. des Cloizeaux and M. L. Mehta: 

1385: Asymptotic behavior of spacing distributions for the eigenvalues of

1386: random matrices, J.Math.Phys. {\bf14} (1973) 1648--1650

1387: 

1388: \bibitem{Dyson} F. Dyson: Fredholm determinants and inverse

1389:   scattering problems. Commun.Math.Phys. {\bf 47}, 171--183 (1976)

1390: 

1391: \bibitem{WidomAs} H. Widom: The asymptotics of a continuous analogue

1392:   of orthogonal polynomials. J.Approx.Th. {\bf 77}, 51--64 (1994)

1393: 

1394: \bibitem{Warc} H. Widom: The strong Szeg\H o limit theorem for

1395:   circular arcs.  Indiana Univ.Math.J. {\bf 21}, 277--283 (1971)

1396: 

1397: \bibitem{D} P. Deift: Talk at the conference ``Spectral theory and

1398:   inverse spectral theory for Jacobi operators'', Snowbird, June 2003.

1399: 

1400: \bibitem{Sz} G. Szeg\H o: Orthogonal polynomials. AMS Colloquium

1401:   Publ. {\bf 23}. New York: AMS 1959

1402: 

1403: \bibitem{Akh} N. I. Akhieser: On polynomials orthogonal on a circular

1404:   arc. Sov.Math.-Doklady {\bf 1}, 31--34 (1960) 

1405: 

1406: \bibitem{GN} L. Golinskii, P. Nevai, W. Van Assche: Perturbation of

1407:   orthogonal polynomials on an arc of the unit circle. J.Approx.Theory

1408:   {\bf 83}, 392--422 (1995)

1409: 

1410: \bibitem{GAkh} L. Golinskii: Akhiezer's orthogonal polynomials and

1411:   Bernstein-Szeg\H o method for a circular arc. J.Approx.Th. {\bf 95},

1412:   229--263 (1998)

1413: 

1414: \bibitem{BL} Bello Hern\'andez, M.; L\'opez Lagomasino:

1415: Ratio and relative asymptotics of polynomials orthogonal on an arc of the

1416: unit circle.

1417: J. Approx. Theory {\bf 92}, 216--244 (1998)

1418: 

1419: \bibitem{RL} 

1420: Barrios Rolan\'ia,  L\'opez Lagomasino:

1421: Ratio asymptotics for polynomials orthogonal on arcs of the unit circle.

1422: Constr. Approx. {\bf 15}, 1--31 (1999)

1423: 

1424: 

1425: \bi{fik} A. S. Fokas, A. R. Its, A. V. Kitaev: The isomonodromy

1426: approach to matrix models in 2D quantum gravity. 

1427: Commun.Math.Phys. {\bf 147}, 395--430 (1992)

1428: 

1429: \bi{dz} P. Deift and X. Zhou: A steepest descent method for

1430: oscillatory Riemann-Hilbert problem. Ann. Math. {\bf 137}, 295--368

1431: (1993)

1432: 

1433: \bi{Dbook} P. Deift: Orthogonal polynomials and random matrices: a

1434: Riemann-Hilbert approach. Courant Lecture Notes in Math. 1998 

1435: 

1436: \bi{DK1} P. Deift, T. Kriecherbauer, K. T-R McLaughlin,   

1437: S. Venakides, X. Zhou: Strong asymptotics for orthogonal polynomials

1438: with respect to exponential weights. Commun. Pure Appl.Math. {\bf 52},

1439: 1491--1552 (1999)

1440: 

1441: \bi{DK2} P. Deift, T. Kriecherbauer, K. T-R McLaughlin,   

1442: S. Venakides, X. Zhou: Uniform asymptotics for polynomials

1443: orthogonal with respect to varying exponential weights

1444: and applications to universality questions in random matrix theory.

1445: Commun. Pure Appl.Math. {\bf 52},

1446: 1335--1425 (1999)

1447: 

1448: \bi{BI} P. Bleher, A. Its: Semiclassical asymptotics of orthogonal 

1449: polynomials, Riemann-Hilbert problem, and universality in the Matrix

1450: Model: Ann.Math. {\bf 150}, 185--266 (1999)

1451:  

1452: 

1453: \bi{bdj} J. Baik, P. Deift, K. Johansson: 

1454: On the distribution of the length of the longest increasing

1455: subsequence of random permutations. 

1456: J.Amer.Math.Soc. {\bf 12}, 1119--1178 (1999)

1457: 

1458: \bi{kuimcl} A. B. J. Kuijlaars, K. T-R McLaughlin, W. Van Assche,

1459: M. Vanlessen: The Riemann-Hilbert approach to strong asymptotics for

1460: orthogonal polynomials on $[-1,1]$. math.CA/0111252.

1461: To appear in Adv. Math.

1462: 

1463: \bi{kuivan} A. B. J. Kuijlaars, M. Vanlessen: Universality for

1464: eigenvalue correlations from the modified Jacobi unitary ensemble. 

1465: Int.Math.Res.Not., no. 30, 1575--1600 (2002)  

1466: 

1467: \bi{Zh} A. Zhedanov: On some classes of polynomials

1468:   orthogonal on arcs of the unit circle connected with symmetric

1469:   orthogonal polynomials on an interval. J.Approx.Th. {\bf 94},

1470:   73--106 (1998)

1471: 

1472: \bi{fredh} I. V. Krasovsky: Some computable Wiener-Hopf determinants and

1473: polynomials orthogonal on an arc of the unit circle.

1474: math-FA/0310172

1475: 

1476: \bi{Abr} Handbook of Mathematical Functions (ed. M. Abramowitz,

1477: I. Stegun), NY:Dover, 1965

1478: 

1479: \bibitem{asto} I. V. Krasovsky: Asymptotics for Toeplitz determinants 

1480: on a circular arc. math-FA/0401256

1481: 

1482: \bi{TW2} C. A. Tracy and H. Widom: Fredholm determinants, differential

1483: equations, and matrix models. Commun.Math.Phys. {\bf 163}, 33--72 (1994)

1484: 

1485: \bi{H} I. I. Hirschman, Jr.: On a formula of Kac and

1486: Achiezer. J.Math.Mech, N.16, 167--196 (1967)

1487: 

1488: \bi{Wsing} H. Widom:  Toeplitz determinants with singular generating

1489: functions. Amer. J. Math. {\bf 95}, 333--383 (1973)

1490: 

1491: 

1492: 

1493: \end{thebibliography}

1494: 

1495: 

1496: \end{document}

1497: 

1498: