1: \section{Riemann-Hilbert problem}\label{sec-rhp}
2:
3: In this section, we will prove Proposition \ref{prop3.2}.
4:
5: For \eqref{en3} and \eqref{e3.5}, note that
6: $\phi^{(k)}_n(t)\ge \phi^{(1)}_n(t)$ for all $k\ge 1$
7: as $\lambda_1\ge\lambda_2\ge\cdots$.
8: But for $k=1$ the estimates \eqref{en3}, \eqref{e3.5}
9: were proved in \cite{BDJ} (Lemma 7.1 (i), (ii)),
10: and so we have the same bounds for all $k\ge 1$.
11: On the other hand, since $\Prob^{\Plan}_N(\lambda\le n)\le 1$,
12: we always have $0\le \phi^{(k)}_n(t)\le 1$.
13:
14: %We will focus on \eqref{e3.6} in t
15: The rest of this section is devoted to proving \eqref{e3.6}
16: and \eqref{en4}.
17: We start from the formulae (see \cite{BOO, kurtj:disc, Ok99, Rains:corr})
18: that (recall $\phi^{(k)}_n(t)=\Prob^{Pois}_t(\lambda_k\le n)$)
19: \begin{eqnarray}
20: \phi^{(1)}_n(t)&=& \det(1-\S_n),\\
21: \label{ez1} \phi^{(k+1)}_{n+k}(t)&=& \phi^{(k)}_{n+k-1}(t) +
22: \biggl(-\frac{d}{dr}\biggr)^k\bigg|_{r=1} \det(1-r\S_n),
23: \qquad k\ge 1,
24: \end{eqnarray}
25: with
26: \begin{equation}\label{e5.3}
27: \varphi(z)=e^{t(z-z^{-1})}.
28: \end{equation}
29: This follows from, for example, Theorem 3.1 of \cite{Rains:corr}
30: with $p_+=p_-=(t,0,0)$ (see also \cite{Rains:mean}) which states that
31: for any finite subset $A$ of $\Z$,
32: \begin{equation}
33: \Prob^{Pois}_t( A\subset \{ \lambda_j-j\} ) = \det ( S(i,j))_{i,j\in A},
34: \qquad S(i,j) = \sum_{k=1}^\infty (\varphi^{-1})_{i+k}\varphi_{-j-k},
35: \end{equation}
36: with $\varphi$ given by \eqref{e5.3},
37: where $\Prob^{Pois}_t$
38: denotes the Plancherel measure for $S_N$ with $N$ being the
39: Poisson variable with mean $t^2$ (Poissonized Plancherel measure).
40: Recall from \eqref{ewr}
41: that $\phi^{(k)}_{n}(t)=\Prob^{Pois}_t(\lambda_k\le n)$.
42: In \cite{BDJ}, the authors obtained the estimates (stronger than)
43: \eqref{e3.6} and \eqref{en4} in the case $k=1$ (see the Remark at the end
44: of Section \ref{sec-tail}).
45: % $\det(1-\S_n)\le Ce^{-c|x|^{3/2}}$, $x\le 0$.
46: Hence we need to prove that for any fixed $k\in\N$,
47: \begin{eqnarray}
48: &\biggl| \frac{d^k}{dr^k}\big|_{r=1} \det(1-r\S_n)
49: \biggr| \le Ce^{-c|x|^{3/2}}, \qquad & 1<\frac{2t}{n}<1+\delta, \\
50: &\biggl| \frac{d^k}{dr^k}\big|_{r=1} \det(1-r\S_n)
51: \biggr| \le Ce^{-ct}, \qquad &1+\delta\le \frac{2t}{n}.
52: \end{eqnarray}
53: By Cauchy's theorem, we write for $0<\epsilon<1$,
54: \begin{equation}
55: \frac{d^k}{dr^k}\big|_{r=1} \det(1-r\S_n)
56: =\frac{k!}{2\pi i} \int_{|s-1|=\epsilon}
57: \frac{\det(1-s\S_n)}{(s-1)^{k+1}} ds.
58: \end{equation}
59: By Remark 3 in Section \ref{sec-iden}, for
60: $\varphi(z)$ as in \eqref{e5.3},
61: $\S_n$ is positive and $\|\S_n\|\le 1$.
62: Hence the eigenvalues $a_j$ of (the trace class operator)
63: $\S_n$ satisfies $0\le a_j\le 1$ (actually one can show
64: that $0\le a_j< 1$).
65: For $|s-1|=\epsilon$,
66: \begin{equation}
67: \bigl|\det(1-s\S_n)\bigr|= \prod_j \bigl|1-sa_j\bigr|
68: \le \prod_j ( 1-(1-\epsilon)a_j )
69: = \det(1-(1-\epsilon)\S_n).
70: \end{equation}
71: Therefore we have
72: \begin{equation}
73: \biggl| \frac{d^k}{dr^k}\big|_{r=1} \det(1-r\S_n) \biggr|
74: \le \frac{k!}{\epsilon^k} \det(1-(1-\epsilon)\S_n).
75: \end{equation}
76: Thus it is sufficient to prove that for fixed $0<r<1$,
77: \begin{eqnarray}
78: &\det(1-r\S_n)\le Ce^{-c|x|^{3/2}}, \qquad &1<\frac{2t}{n}<1+\delta,\\
79: &\det(1-r\S_n)\le Ce^{-ct}, \qquad &1+\delta\le \frac{2t}{n},
80: \end{eqnarray}
81: or by Theorem \ref{mainthm}, we need to prove that
82: for any fixed $0<s<1$, (note that $\varphi$ given
83: by \eqref{e5.3} has no winding)
84: \begin{eqnarray}
85: \label{eq5.9}
86: &(1+s)^{-n}\det(1-s\K_n)\le Ce^{-c|x|^{3/2}}, \qquad
87: &1<\frac{2t}{n}<1+\delta,\\
88: \label{en5.1}
89: &(1+s)^{-n}\det(1-s\K_n)\le Ce^{-ct}, \qquad
90: &1+\delta\le \frac{2t}{n},
91: \end{eqnarray}
92: where $x$ is defined in \eqref{e3.4}
93: \begin{equation}\label{eq5.14}
94: \frac{2t}{n}= 1-\frac{x}{2^{1/3}n^{2/3}}.
95: \end{equation}
96:
97:
98:
99: Since $\K_n$ is an integrable operator,
100: there is a naturally associated Riemann-Hilbert problem
101: (see \cite{IIKS, deiftint}).
102: Let $m(z;k)$ be the $2\times 2$ matrix function
103: which solves the following Riemann-Hilbert problem (RHP) :
104: with contour $\Sigma = \{ |z|=1\}$,
105: oriented counterclockwise,
106: \begin{equation}\label{eq5.11}
107: \begin{cases}
108: m(z;k) \qquad \text{is analytic in $z\in\C\setminus\Sigma$,}\\
109: m_+(z;k)=m_-(z;k) \begin{pmatrix}
110: 1-s^2&-sz^{-k}e^{-t(z-z^{-1})}\\
111: sz^{k}e^{t(z-z^{-1})}&1.
112: \end{pmatrix}
113: \qquad \text{for $z\in\Sigma$,}\\
114: m(z;k) \to I \qquad \text{as $z\to\infty$.}
115: \end{cases}
116: \end{equation}
117: Here and also in the following,
118: the notation $f_+(z)$ (resp., $f_-(z)$)
119: denotes the limiting value
120: of $\lim_{z'\to z} f(z')$ from the left (resp., right)
121: of the contour in the direction of the orientation.
122: In the above case, $m_+(z;k;s)$ (resp., $m_-(z;k;s)$) means
123: $\lim_{z'\to z} m(z';k;s)$
124: with $|z'|<1$ (resp., $|z'|>1$.)
125: In (52) of \cite{BDJ2}, it is shown that
126: \begin{equation}\label{eq5.10}
127: (1+s)^{-n}\det(1-s\K_n)
128: = \prod_{k=n}^\infty m_{11}(0;k+1),
129: \end{equation}
130: where $m_{11}(0;k)$ is the (11)-entry of $m(z;k)$ evaluated at $z=0$.
131: Therefore, in order to prove \eqref{eq5.9} and \eqref{en5.1},
132: we need asymptotic results for $m_{11}(0;k)$
133: as $k,t\to\infty$.
134: In the special case when $s=1$, this RHP is algebraically equivalent to
135: the RHP for the orthogonal polynomials on the unit circle
136: with respect to the measure $e^{t(z+z^{-1})}dz/(2\pi i z)$,
137: whose asymptotics as $k,t\to\infty$ was investigated in \cite{BDJ}.
138: The RHP \eqref{eq5.11} was introduced in \cite{BDJ2}.
139:
140: There is a critical difference in the asymptotic analysis
141: depending whether $s=1$ or $0<s<1$.
142: In the former case, the jump matrix in
143: \eqref{eq5.11} has a upper/lower factorization,
144: but not a lower/upper factorization,
145: while in the later case, the jump matrix has both factorizations.
146: This difference makes the later case much easier
147: to analyze asymptotically.
148: In the former case, we need a WKB type analysis which involves
149: the construction of a parametrix in terms of the equilibrium measure
150: of a certain variational problem,
151: and in the case where $\frac{2t}{n}>1$,
152: the main asymptotic contribution to the RHP
153: comes from the part of the circle near $z=-1$.
154: But in the later case, due to the existence of both factorization
155: of the jump matrix,
156: the RHP localizes in the limit just to two points on the circle.
157: We refer the reader to \cite{DZ1} for an example of the second type,
158: and to \cite{DZ2, DVZ, DKMVZ3} for examples of the first type.
159:
160: \begin{rem}
161: The different analysis for $s=1$ and $0<s<1$ gives us different estimates.
162: Indeed, when $s=1$, instead of \eqref{ee1} below, we have
163: (see (6.42) of \cite{BDJ})
164: \begin{equation}
165: \log m_{11}(0;k) \le k(-\frac{2t}{k}+\log \frac{2t}{k} +1),
166: \end{equation}
167: which imply \eqref{q1}, \eqref{q2}.
168: Thus, in order to obtain the better estimates \eqref{q1}, \eqref{q2},
169: \eqref{e3.3}
170: for general row $\xi_k$ (see Remark at the end of Section \ref{sec-tail}),
171: we need to analysis the RHP \eqref{eq5.11} as $s\to 1$ instead of fixed $s<1$.
172: \end{rem}
173:
174: Fix $0<s<1$.
175: We will prove the following estimate.
176: \begin{lem}\label{lem5.1}
177: There are positive constants $M_0$ and $c$ such that
178: when $\frac{2t}{k}\ge 1+\frac{M_0}{2^{1/3}k^{2/3}}$,
179: we have
180: \begin{equation}\label{ee1}
181: \log m_{11}(0;k) \le -c \sqrt{1-\biggl(\frac{k}{2t}\biggr)^2},
182: %+ C\frac{1}{k^{2/3}}.
183: \end{equation}
184: for large $t$.
185: \end{lem}
186: Assuming this result, we will prove \eqref{eq5.9} and \eqref{en5.1},
187: which completes the proof of Proposition \ref{prop3.2}.
188:
189: \begin{proof}[Proof of Proposition \ref{prop3.2}]
190: We need to prove \eqref{eq5.9} and \eqref{en5.1}.
191:
192: {\bf (a) Estimate \eqref{eq5.9} : }
193: It is enough to prove \eqref{eq5.9} for $-2^{1/3}n^{2/3}\delta< x\le -M$
194: where $M:=M_0(1+2\delta)$ with $M_0$ in Lemma \ref{lem5.1}.
195: As noted above, since $\S_n$ is positive and $\|\S_n\|\le 1$,
196: its eigenvalues $a_j$ satisfies $0\le a_j\le 1$.
197: Hence for $0<s<1$, by \eqref{eq5.10} and Theorem \ref{mainthm},
198: \begin{equation}
199: 0\le \prod_{k=n}^\infty m_{11}(0;k+1)
200: = \det(1-s^2\S_n) = \prod_j (1-s^2a_j) \le 1,
201: \end{equation}
202: for \emph{any} $n$.
203: Therefore in order to prove \eqref{eq5.9},
204: it is enough to show that
205: \begin{equation}\label{eq5.13}
206: \sum_{(*)} \log m_{11}(0;k+1)
207: \le -c|x|^{3/2} + C,
208: \end{equation}
209: for some constants $c,C>0$
210: where the summation is over the set (note $x+M\le 0$ in \eqref{eq5.13})
211: \begin{equation}\label{eq5.15}
212: (*) : \quad n\le k\le n-\frac{x+M_1}{2^{1/3}}n^{1/3},
213: \end{equation}
214: with $M_1:=M_0(1+\delta)$.
215: We will show that for $k$ in $(*)$,
216: \begin{equation}\label{eq5.16}
217: 1+\frac{M_0}{2^{1/3}k^{2/3}} \le \frac{2t}{k}.
218: \end{equation}
219: Since $n\le k$, \eqref{eq5.16} follows from
220: \begin{equation}\label{ee5.17}
221: \frac{2t}{k} \ge 1+ \frac{M_0}{2^{1/3}n^{2/3}}.
222: \end{equation}
223: In order to show \eqref{ee5.17}, since
224: \begin{equation}\label{ee5.18}
225: \frac{2t}{k} = \frac{2t}{n}\cdot \frac{n}{k}
226: \ge \bigl( 1-\frac{x}{2^{1/3}n^{2/3}}\bigr)
227: \bigl( 1-\frac{x+M_1}{2^{1/3}n^{2/3}}\bigr)^{-1},
228: \end{equation}
229: it is enough to checking that
230: \begin{equation}\label{ee5.19}
231: \bigl( 1-\frac{x}{2^{1/3}n^{2/3}}\bigr)
232: \ge
233: \bigl( 1-\frac{x+M_1}{2^{1/3}n^{2/3}}\bigr)
234: \biggl( 1+\frac{M_0}{2^{1/3}n^{2/3}} \biggr),
235: \end{equation}
236: which is equivalent to check that
237: \begin{equation}\label{ee5.20}
238: M_1 \ge \bigl( 1-\frac{x+M_1}{2^{1/3}n^{2/3}}\bigr) M_0.
239: \end{equation}
240: But since $-x\le 2^{1/3}\delta n^{2/3}$,
241: \begin{equation}
242: \bigl( 1-\frac{x+M_1}{2^{1/3}n^{2/3}}\bigr) M_0
243: \le \bigl( 1+\delta -\frac{M_1}{2^{1/3}n^{2/3}}\bigr)
244: \frac{M_1}{1+\delta}
245: \le M_1,
246: \end{equation}
247: and hence \eqref{eq5.16} is proved.
248:
249: Now using \eqref{ee1}, the sum in \eqref{eq5.13} satisfies
250: \begin{equation}
251: \begin{split}
252: \sum_{(*)} \log m_{11}(0;k+1)
253: &\le \sum_{(*)}
254: -c\sqrt{1-\biggl(\frac{k}{2t}\biggr)^2}\\
255: &\le \sum_{(*)} -c \sqrt{1-\frac{k}{2t}}\\
256: &\le -c\int_{n}^{n-\frac{x+M_1}{2^{1/3}}n^{1/3}}
257: \sqrt{1-\frac{s}{2t}} ds \\
258: &= c\biggl(\frac{4t}3\biggr)
259: \biggl[ \biggr( 1-\frac{n-\frac{x+M_1}{2^{1/3}}n^{1/3}}{2t}
260: \biggr)^{3/2}
261: - \biggr( 1-\frac{n}{2t} \biggr)^{3/2} \biggr]\\
262: &= \frac23c \biggl( \frac{2t}{n}\biggr)^{-1/2} \biggl[
263: \biggr( n^{2/3} \bigl( \frac{2t}{n}-1\bigr) + \frac{x+M_1}{2^{1/3}}
264: \biggr)^{3/2}
265: - \biggr( n^{2/3} \bigl( \frac{2t}{n}-1 \bigr) \biggr)^{3/2} \biggr]\\
266: &= \frac23c \biggl( \frac{2t}{n}\biggr)^{-1/2} \biggl[
267: \biggr( \frac{M_1}{2^{1/3}} \biggr)^{3/2}
268: - \biggr( \frac{-x}{2^{1/3}} \biggr)^{3/2} \biggr]
269: \end{split}
270: \end{equation}
271: where the second inequality is due to the monotonicity of the function
272: $f(y)=\sqrt{1-\frac{y}{2t}}$.
273: Since $1<\frac{2t}{n}<1+\delta$, we obtain \eqref{eq5.9}.
274:
275:
276: {\bf (b) Estimate \eqref{en5.1} : }
277: By a similar argument as in (a), it is enough to show that
278: \begin{equation}\label{ee2}
279: \sum_{(**)} \log m_{11}(0;k+1) \le -ct+C,
280: \end{equation}
281: for some constants $c,C>0$ where $(**)$ is the set
282: \begin{equation}
283: (**): \quad \frac{2t}{1+\delta}\le k\le \frac{2t}{1+\delta/2}.
284: \end{equation}
285: For $k$ in $(**)$, we have
286: \begin{equation}
287: \frac{k}{2t}
288: \le \frac{1}{1+\delta/2}< 1.
289: \end{equation}
290:
291: Now using \eqref{ee1}, the sum in \eqref{ee2} satisfies
292: \begin{equation}
293: \begin{split}
294: \sum_{(**)} \log m_{11}(0;k+1)
295: &\le \sum_{(**)}
296: -c\sqrt{1-\biggl(\frac{k}{2t}\biggr)^2}\\
297: &\le -c \sqrt{1-\biggl(\frac{1}{1+\delta/2}\biggr)^2}
298: \biggl( \frac{2t}{1+\delta/2}-\frac{2t}{1+\delta}+1 \biggr) \\
299: & \le -ct +C.
300: \end{split}
301: \end{equation}
302: \end{proof}
303:
304:
305: \subsection*{RHP Asymptotics and Proof of Lemma \ref{lem5.1}}
306: In the rest of this section, we prove Lemma \ref{lem5.1}
307: by asymptotic analysis of the RHP \eqref{eq5.11}.
308:
309: Set
310: \begin{equation}
311: \eta := \frac{k}{2t}.
312: % \gamma:= \frac{2t}{k}= 1-\frac{x}{2^{1/3}k^{2/3}}.
313: \end{equation}
314: Under the condition of Lemma \ref{lem5.1},
315: we have $\eta<1$.
316: We denote by $v(z)$ the jump matrix in the second condition
317: of the RHP \eqref{eq5.11}.
318: Note that the (21)-entry of $v$ is $se^{2tf(z;\eta)}$ where
319: \begin{equation}
320: f(z;\eta):=\frac{1}2(z-z^{-1})+\eta\log z,
321: \end{equation}
322: where $\log z\in\R$ for $z>0$.
323: The critical points of this function are
324: $\xi:=e^{i\theta_c}$ and $\xi^{-1}=\overline{\xi}$ where
325: \begin{equation}\label{er5.34}
326: \xi=-\eta+i\sqrt{1-\eta^2}.
327: \end{equation}
328: Note that $-\pi/2<\theta_c<\pi$.
329: For $z=\rho e^{i\theta}$, consider
330: \begin{equation}\label{er5.35}
331: F_\theta(\rho) := Re f(z) = \frac{1}2(\rho-\rho^{-1})
332: \cos\theta + \eta \log \rho.
333: \end{equation}
334: Its derivative at $\rho=1$ is
335: \begin{equation}\label{ee5.35}
336: \frac{d}{d\rho}F_\theta(1)= \cos\theta+\eta,
337: \end{equation}
338: which is positive for $|\theta|< \theta_c$, and is negative
339: for $\theta_c < |\theta|\le \pi$.
340: \begin{figure}[ht]
341: \centerline{\epsfig{file=fun1.eps, width=5cm, height=5cm}
342: \epsfig{file=fun2.eps, width=5cm, height=5cm}
343: \epsfig{file=fun3.eps, width=5cm, height=5cm}}
344: \caption{Graph of $F_\theta(\rho)$ for $\theta=\frac\pi6$ (case (i)),
345: $\frac{5\pi}6$ (case (ii)) and $\pi$ (case (iii))
346: when $\eta=15/16$}
347: \label{fig-fun}
348: \end{figure}
349: Indeed it is easy to check that :
350: \begin{enumerate}
351: \item When $|\theta| \le \frac{\pi}2$,
352: $F_\theta(\rho) < 0$ for $0<\rho <1$,
353: and $F_\theta(\rho) > 0$ for $\rho >1$.
354: \item When $\frac{\pi}2< |\theta| <\theta_c$,
355: $F_\theta(\rho)>0$ for $0<\rho<\rho_0$,
356: $F_\theta(\rho) <0$ for $\rho_0 <\rho <1$,
357: $F_\theta(\rho) >0$ for $1<\rho <\rho_o^{-1}$, and
358: $F_\theta(\rho)<0$ for $\rho>\rho_0^{-1}$.
359: Here $\rho_0$ is a number satisfying
360: $0<\rho_0<\rho_\theta$, where
361: \begin{equation}\label{ee5.36}
362: \rho_\theta:= \frac{\eta-\sqrt{\eta^2-\cos^2\theta}}{-\cos\theta} <1,
363: \end{equation}
364: and $\frac{d}{d\theta}F_\theta(\rho_\theta)
365: =\frac{d}{d\theta}F_\theta(\rho_\theta^{-1})=0$.
366: \item When $\theta_c < |\theta| \le \pi$,
367: $F_\theta(\rho) >0$ for $0< \rho<1$,
368: and $F_\theta(\rho)<0$ for $\rho>1$.
369: \item The curve $\{ \rho e^{i\theta} : F_\theta(\rho)=0 \}$
370: crosses the circle at $90$ degree.
371: \end{enumerate}
372: Typical graphs of $F_\theta(\rho)$ is given in Figure \ref{fig-fun}
373: for the value $\eta=15/16$ for $\theta$ in the three different
374: cases (i)-(iii).
375: Figure \ref{fig-bound} is a signature table for $Re(f(z))$ when $\eta=15/16$.
376: The solid curve is $Re (f(z))=0$, and the dotted rays
377: represent the lines $\cos\theta=-\eta$.
378: The $\pm$ signs denote
379: the signature of $Re (f(z))$ in each of the four components.
380: The curve $Re (f(z))=0$ and the lines $\cos\theta=-\eta$ meet
381: on the unit circle at the points $\xi$ and $\xi^{-1}$.
382: \begin{figure}[ht]
383: \centerline{\epsfig{file=bd6.eps, width=8cm, height=8cm}}
384: \caption{Curve $Re (f(z))=0$ when $\eta=15/16$}
385: \label{fig-bound}
386: \end{figure}
387: \begin{figure}[ht]
388: \centerline{\epsfig{file=crit.eps, width=8cm, height=8cm}}
389: \caption{Solid curve represents the set
390: $\{ \rho_\theta e^{i\theta} : \frac\pi2<\theta\le \theta_c \}$
391: when $\eta=15/16$}
392: \label{fig-crit}
393: \end{figure}
394:
395:
396: Let $\Sigma=\Sigma_1\cup\Sigma_2$,
397: where $\Sigma_1=\{ e^{i\theta} : |\theta| < \theta_c\}$,
398: and $\Sigma_2=\Sigma\setminus\Sigma_1$.
399: Define the function
400: \begin{equation}\label{ee5.37}
401: \delta(z):=
402: \biggl( \frac{z-\xi}{z-\xi^{-1}}\biggr)^{-\frac1{2\pi i} \log(1-s^2)},
403: \end{equation}
404: which is analytic in $\C\setminus\Sigma_2$ : we choose the branch so that
405: $\delta(z)\to 1$ as $z\to\infty$ along the positive real axis.
406: Then it solves the scalar Riemann-Hilbert problem
407: \begin{equation}
408: \begin{split}
409: &\delta_+(z)=\delta_-(z)(1-s^2), \qquad z\in\Sigma_2, \\
410: & \delta(z) \to 1, \qquad \text{as $z\to\infty$,}
411: \end{split}
412: \end{equation}
413: where $\delta_\pm$ has the same meaning as in the RHP \eqref{eq5.11}.
414: Note that
415: \begin{equation}
416: \delta(0)= (1-s^2)^{1-\frac{\theta_c}{\pi}}.
417: \end{equation}
418: Now set
419: \begin{equation}
420: m^{(2)}(z):= m(z) \delta^{-\sigma_3},
421: \qquad \sigma_3= \begin{pmatrix} 1&0\\0&-1 \end{pmatrix}.
422: \end{equation}
423: Then (i) $m^{(2)}(z)$ is analytic in $\C\setminus\Sigma$,
424: (ii) $m^{(2)}(z)\to I$ as $z\to\infty$, and (iii)
425: $m^{(2)}_+(z)=m^{(2)}_-v^{(2)}(z)$ for $z\in\Sigma$, where
426: \begin{equation}
427: v^{(2)}(z) = \begin{cases}
428: \begin{pmatrix} 1-s^2& -se^{-2tf(z)}\delta^2(z)\\
429: se^{2tf(z)}\delta^{-2}(z)&1 \end{pmatrix},
430: \qquad &z\in\Sigma_1, \\
431: \begin{pmatrix} 1& -\frac{s}{1-s^2} e^{-2tf(z)}\delta_+^2(z)\\
432: \frac{s}{1-s^2} e^{2tf(z)}\delta_-^{-2}&1-s^2 \end{pmatrix},
433: \qquad &z\in\Sigma_2.
434: \end{cases}
435: \end{equation}
436: Also we have
437: \begin{equation}
438: m_{11}(0)= m^{(2)}_{11}(0)(1-s^2)^{1-\frac{\theta_c}{\pi}}.
439: \end{equation}
440: Note that the jump matrix has the factorizations
441: \begin{equation}
442: v^{(2)}(z) = \begin{cases}
443: \begin{pmatrix} 1&-se^{-2tf(z)}\delta^2(z)\\0&1 \end{pmatrix}
444: \begin{pmatrix} 1&0\\se^{2tf(z)}\delta^{-2}(z)&1 \end{pmatrix}
445: , \qquad &z\in\Sigma_1, \\
446: \begin{pmatrix} 1&0\\
447: \frac{s}{1-s^2} e^{2tf(z)}\delta_-^{-2}&1-s^2 \end{pmatrix}
448: \begin{pmatrix} 1&-\frac{s}{1-s^2} e^{-2tf(z)}\delta_+^2(z)\\0&1 \end{pmatrix}
449: , \qquad &z\in\Sigma_2.
450: \end{cases}
451: \end{equation}
452:
453: From (i)-(iii) in \eqref{ee5.36}, we can take an oriented closed curve
454: $\Sigma^{(3)}_-$
455: surrounding $0$ and $-1$,
456: and passing through $\xi$ and $\xi^{-1}$
457: (the solid curve in Figure \ref{fig-contour}) on which
458: $Re f(z) <0$ except at $z=\xi, \xi^{-1}$.
459: \begin{figure}[ht]
460: \centerline{\epsfig{file=contour.eps, width=15cm}}
461: \caption{$\Sigma^{(3)}_\pm$ and $\Omega^{(3)}_j$, $j=1,\cdots,6$}
462: \label{fig-contour}
463: \end{figure}
464: Let $\Sigma^{(3)}_-=\overline{\Sigma^{(3)}_{-1}\cup\Sigma^{(3)}_{-2}}$
465: where $\Sigma^{(3)}_{-1}$ is the open subset of $\Sigma^{(3)}_-$
466: satisfying $|\arg(z)|<\theta_c$ and
467: $\Sigma^{(3)}_{-2}=\Sigma^{(3)}_-\setminus \overline{\Sigma^{(3)}_{-1}}$.
468: Similarly, we can take an oriented closed curve
469: $\Sigma^{(3)}_+$ surrounding $0$, but not $-1$
470: and passing through $\xi, \xi^{-1}$
471: (the dashed curve in Figure \ref{fig-contour}) on which
472: $Re f(z)>0$ except at $z=\xi, \xi^{-1}$.
473: Again let $\Sigma^{(3)}_+=\overline{\Sigma^{(3)}_{+1}
474: \cup\Sigma^{(3)}_{+2}}$
475: where $\Sigma^{(3)}_{+1}$ is the open subset of $\Sigma^{(3)}_+$
476: satisfying $|\arg(z)|<\theta_c$ and
477: $\Sigma^{(3)}_{+2}=\Sigma^{(3)}_+\setminus \overline{\Sigma^{(3)}_{+1}}$.
478: The shape of $\Sigma^{(3)}_\pm$ will be specified further below
479: (see the third case for the estimation of $|v_R(z)-I|$
480: between \eqref{en1} and \eqref{ec2}).
481: Let $\Sigma^{(3)}=\overline{\Sigma^{(3)}_-\cup\Sigma^{(3)}_+}$.
482: Let $\Omega^{(3)}_j$, $1\le j\le 6$ be open regions
483: as in Figure \ref{fig-contour}.
484: Define
485: \begin{equation}
486: m^{(3)}(z) := \begin{cases}
487: m^{(2)}(z) \begin{pmatrix} 1&0\\ se^{2tf}\delta^{-2}&1
488: \end{pmatrix}^{-1}, &\qquad z\in\Omega^{(3)}_1,\\
489: m^{(2)}(z) \begin{pmatrix} 1&-se^{-2tf}\delta^2\\0&1
490: \end{pmatrix}, &\qquad z\in\Omega^{(3)}_2,\\
491: m^{(2)}(z) \begin{pmatrix} 1& -\frac{s}{1-s^2}e^{-2tf}\delta^2\\0&1
492: \end{pmatrix}^{-1}, &\qquad z\in\Omega^{(3)}_3,\\
493: m^{(2)}(z) \begin{pmatrix} 1&0\\ \frac{s}{1-s^2}e^{2tf}\delta^{-2}&1
494: \end{pmatrix}, &\qquad z\in \Omega^{(3)}_4.\\
495: m^{(2)}(z), &\qquad z\in\Omega^{(3)}_5\cup\Omega^{(3)}_6.
496: \end{cases}
497: \end{equation}
498: Then (i) $m^{(3)}(z)$ is analytic in $\C\setminus\Sigma^{(3)}$,
499: (ii) $m^{(3)}(z)\to I$ as $z\to\infty$, and
500: $m^{(3)}_+(z)=m^{(3)}_-(z)v^{(3)}$ for $z\in\Sigma^{(3)}$,
501: where
502: \begin{equation}\label{e5.46}
503: v^{(3)}(z)= \begin{cases}
504: \begin{pmatrix} 1&0\\ se^{2tf}\delta^{-2}&1
505: \end{pmatrix}, &\qquad z\in\Sigma^{(3)}_{-1},\\
506: \begin{pmatrix} 1&-se^{-2tf}\delta^2\\0&1
507: \end{pmatrix}, &\qquad z\in\Sigma^{(3)}_{+1},\\
508: \begin{pmatrix} 1& -\frac{s}{1-s^2}e^{-2tf}\delta^2\\0&1
509: \end{pmatrix}, &\qquad z\in\Sigma^{(3)}_{+2},\\
510: \begin{pmatrix} 1&0\\ \frac{s}{1-s^2}e^{2tf}\delta^{-2}&1
511: \end{pmatrix}, &\qquad z\in\Sigma^{(3)}_{-2}.
512: \end{cases}
513: \end{equation}
514: Also we have
515: \begin{equation}\label{ee5.46}
516: m_{11}(0)= m^{(3)}_{11}(0)(1-s^2)^{1-\frac{\theta_c}{\pi}}.
517: \end{equation}
518: Observe that $v^{(3)}(z)\to I$ as $t\to\infty$
519: for $z\in\Sigma^{(3)}\setminus\{\xi, \xi^{-1}\}$.
520: Thus we expect that $m^{(3)}(z)\to I$ as $t\to\infty$.
521: If this were indeed true, we would have
522: \begin{equation}\label{ee5.47}
523: \log m_{11}(0) \sim \frac{\sin^{-1}\sqrt{1-\eta^2}}{\pi}\log(1-s^2)
524: \le \frac{\log(1-s^2)}{\pi}\sqrt{1-\biggl(\frac{k}{2t}\biggr)^2},
525: \qquad t\to\infty.
526: \end{equation}
527: But the difficulty, however, is that $v^{(3)}$ does not converge to $I$
528: uniformly on $\Sigma^{(3)}$.
529: As in \cite{DZ1}, we overcome this difficulty by constructing a parametrix
530: for the solution of the RHP $(\Sigma^{(3)},v^{(3)})$
531: around the points $\xi, \xi^{-1}$.
532:
533:
534:
535:
536: Let $\tau$ be a complex number satisfying $0<|\tau|<1$.
537: Following \cite{DZ1}, set
538: \begin{equation}
539: \nu:= -\frac1{2\pi}\log(1-|\tau|^2), \qquad a:=i\nu.
540: \end{equation}
541: Define
542: \begin{equation}
543: \beta_{12}:= \frac{\sqrt{2\pi}e^{\frac{\pi}{4}i}
544: e^{-\frac{\pi}2\nu}}{\tau\Gamma(-a)},
545: \qquad \beta_{21}:=\overline{\beta_{12}}
546: =\frac{\sqrt{2\pi}e^{-\frac{\pi}{4}i}
547: e^{-\frac{\pi}2\nu}}{\overline{\tau}\Gamma(a)}.
548: \end{equation}
549: Note that
550: \begin{equation}
551: \beta_{12}\beta_{21}=\nu,
552: \end{equation}
553: as $|\Gamma(iv)|^2=\frac{\pi}{\nu\sinh(\pi\nu)}$ for real $\nu\neq 0$.
554: Let $D_a$ be the parabolic-cylinder function (see, e.g. \cite{AS, WW})
555: which solves
556: \begin{equation}
557: \frac{d^2}{d\zeta^2} D_a(\zeta)
558: +\biggl(\frac12-\frac{\zeta^2}{4} +a\biggr)D_a(\zeta)=0.
559: \end{equation}
560: We note that $D_a(\zeta)$ is an entire function.
561:
562: Let the matrix
563: \begin{equation}
564: \Psi(w)=\begin{pmatrix}
565: \Psi_{11}(w)&\Psi_{12}(w)\\ \Psi_{21}(w)&\Psi_{22}(w)
566: \end{pmatrix},
567: \qquad w\in\C\setminus\R,
568: \end{equation}
569: be defined as follows (see \cite{DZ1} Section4) : for $Im(w)>0$,
570: \begin{eqnarray}
571: \Psi_{11}(w) &:=& e^{-\frac34\pi\nu}D_a(e^{-\frac34\pi i}w),\\
572: \Psi_{12}(w) &:=&
573: (\beta_{21})^{-1}e^{\frac14\pi \nu}
574: \biggl(\frac{d}{dw} D_{-a}(e^{-\frac14\pi i}w)
575: -\frac{iw}{2}D_{-a}(e^{-\frac14\pi i}w)\biggr), \\
576: \Psi_{21}(w) &:=&
577: (\beta_{12})^{-1}e^{-\frac34\pi \nu}
578: \biggl(\frac{d}{dw} D_{a}(e^{-\frac34\pi i}w)
579: +\frac{iw}{2}D_{a}(e^{-\frac34\pi i}w)\biggr), \\
580: \Psi_{22}(w) &:=& e^{\frac14\pi\nu}D_{-a}(e^{-\frac14\pi i}w),
581: \end{eqnarray}
582: and for $Im(w)<0$,
583: \begin{eqnarray}
584: \Psi_{11}(w) &:=& e^{\frac14\pi\nu}D_a(e^{\frac14\pi i}w),\\
585: \Psi_{12}(w) &:=&
586: (\beta_{21})^{-1}e^{-\frac34\pi \nu}
587: \biggl(\frac{d}{dw} D_{-a}(e^{\frac34\pi i}w)
588: -\frac{iw}{2}D_{-a}(e^{\frac34\pi i}w)\biggr), \\
589: \Psi_{21}(w) &:=&
590: (\beta_{12})^{-1}e^{\frac14\pi \nu}
591: \biggl(\frac{d}{dw} D_{a}(e^{\frac14\pi i}w)
592: +\frac{iw}{2}D_{a}(e^{\frac14\pi i}w)\biggr), \\
593: \Psi_{22}(w) &:=& e^{-\frac34\pi\nu}D_{-a}(e^{\frac34\pi i}w),
594: \end{eqnarray}
595: The function $\Psi$ satisfies
596: \begin{itemize}
597: \item $\Psi(w)$ is analytic in $w\in\C\setminus\R$.
598: \item For $w\in\R$,
599: \begin{equation}
600: \Psi_+(w)= \Psi_-(w) \begin{pmatrix}
601: 1-|\tau|^2&-\overline{\tau}\\ \tau&1
602: \end{pmatrix},
603: \end{equation}
604: where $\Psi_+(w)$ (resp., $\Psi_-(w)$)
605: is the limit of $\Psi(s)$ as $s\to w$ with $Im(s)>0$ (resp., $Im(s)<0$).
606: \item As $w\to\infty$,
607: \begin{equation}\label{ea5.63}
608: \Psi(w)e^{\frac14iw^2\sigma_3}w^{-i\nu\sigma_3} = I + O(w^{-1})
609: \end{equation}
610: where $w^{-i\nu}$ denotes the branch which
611: is analytic in $\C\setminus(-\infty,0]$
612: and has modulus $1$ for $w\in (0,\infty)$.
613: \end{itemize}
614: These properties can be found in \cite{DZ1} Section 4.
615:
616: Let $\Gamma$ be the union of four rays, labeled by
617: $\Gamma_j$, $j=1,\cdots,4$,
618: with the orientation as indicated in
619: Figure \ref{fig-Gamma}.
620: All the rays and $\R$ meet at the angle $\pi/3$.
621: \begin{figure}[ht]
622: \centerline{\epsfig{file=figGamma.eps, width=7cm}}
623: \caption{$\Gamma_j$ and $\Omega_j$}
624: \label{fig-Gamma}
625: \end{figure}
626: Denote the components of $\C\setminus(\Gamma\cup\R)$ by $\Omega_j$,
627: $j=1,\cdots,6$ as in Figure \ref{fig-Gamma}.
628: Define $H(w)$, analytic in $\C\setminus\Gamma$, by
629: \begin{equation}
630: H(w) := \bigl(\begin{smallmatrix} a(\xi)^{-1}&0\\0&a(\xi)
631: \end{smallmatrix}\bigr)
632: \bigl(\begin{smallmatrix} 0&1\\1&0 \end{smallmatrix}\bigr)
633: \Psi(w) e^{\frac14 iw^2\sigma_3} w^{-i\nu\sigma_3}
634: \bigl(\begin{smallmatrix} 0&1\\1&0 \end{smallmatrix}\bigr)
635: \phi(w)
636: \bigl(\begin{smallmatrix} a(\xi)&0\\0&a(\xi)^{-1}
637: \end{smallmatrix}\bigr),
638: \end{equation}
639: where $a(\xi)$ is
640: \begin{equation}
641: a(\xi)=e^{tf(\xi)} \biggl(
642: \frac{-i\xi}{(\xi-\xi^{-1})\sqrt{2t}(1-\eta^2)^{1/4}} \biggr)^{i\nu},
643: \end{equation}
644: and $\phi(w)$ is defined by
645: \begin{equation}
646: \phi(w):=
647: \begin{cases}
648: \begin{pmatrix} 1&-\tau e^{\frac12 iw^2} w^{-2i\nu}\\ 0&1
649: \end{pmatrix} , \qquad
650: &w\in \Omega_1, \\
651: \begin{pmatrix} 1&0\\
652: -\overline{\tau}e^{-\frac12 iw^2} w^{2i\nu}&1
653: \end{pmatrix} , \qquad
654: &w\in \Omega_6, \\
655: \begin{pmatrix} 1&0\\ \frac{\overline{\tau}}{1-|\tau|^2}
656: e^{-\frac12 iw^2} w^{2i\nu}&1 \end{pmatrix} ,
657: \qquad &w\in \Omega_3, \\
658: \begin{pmatrix} 1&\frac{\tau}{1-|\tau|^2}
659: e^{\frac12 iw^2} w^{-2i\nu} \\ 0&1
660: \end{pmatrix} , \qquad
661: &w\in \Omega_4, \\
662: I
663: &w\in \Omega_2, \Omega_5.
664: \end{cases}
665: \end{equation}
666: Then by recalling that $w^{-i\nu\sigma_3}$ is analytic in
667: $\C\setminus(-\infty,0]$, one can directly check that
668: $H_+(w)=H_-(w)v_H(w)$ for $w\in\Gamma$, where $v_H(w)$ is given by
669: \begin{equation}\label{e5.67}
670: v_H(w):=
671: \begin{cases}
672: \begin{pmatrix} 1&-\tau a(\xi)^{-2} w^{-2i\nu}e^{\frac12 iw^2}
673: \\ 0&1 \end{pmatrix} , \qquad
674: &w\in \Gamma_1, \\
675: \begin{pmatrix} 1&0\\ \overline{\tau} a(\xi)^{2} w^{2i\nu}e^{-\frac12 iw^2}
676: &1 \end{pmatrix} , \qquad
677: &w\in \Gamma_4, \\
678: \begin{pmatrix} 1&0\\
679: \frac{\overline{\tau}}{1-|\tau|^2}a(\xi)^{2} w^{2i\nu}e^{-\frac12 iw^2}
680: &1 \end{pmatrix} ,
681: \qquad &w\in \Gamma_2, \\
682: \begin{pmatrix} 1&\frac{-\tau}{1-|\tau|^2} a(\xi)^{-2}
683: w^{-2i\nu}e^{\frac12 iw^2} \\ 0&1
684: \end{pmatrix} , \qquad
685: &w\in \Gamma_3.
686: \end{cases}
687: \end{equation}
688: Also, from \eqref{ea5.63}, we have
689: \begin{equation}\label{ee5.67}
690: H(w) = I + O(w^{-1}), \qquad \text{as $w\to\infty$.}
691: \end{equation}
692: As $|a(\xi)|=e^{-\nu\theta_c}$, $-\pi/2<\theta_c<\pi$, we see that the error
693: term $O(w^{-1})$ in \eqref{ee5.67} is uniform for $\frac{2t}{k}>1$.
694: Similarly, $|H(w)|$ is uniformly bounded in the $w$ plane for
695: $\frac{2t}{k}>1$.
696:
697: \begin{figure}[ht]
698: \centerline{\epsfig{file=map.eps, width=10cm}}
699: \caption{map $z\mapsto w(z)$}
700: \label{fig-map}
701: \end{figure}
702: Define the map
703: \begin{equation}
704: z \mapsto w(z):=\sqrt{2t}(1-\eta^2)^{1/4}i\xi^{-1}(z-\xi).
705: \end{equation}
706: It maps $\xi$ to $0$, and the tangent line $L_\xi$ to the
707: unit circle $\Sigma$ at $\xi$, to the real line as in Figure \ref{fig-map}.
708: Let $\calO_\xi$, $\calO_{\overline{\xi}}$
709: be the disjoint sets $\{z : |z-\xi|<\varrho \}$,
710: $\{z : |z-\overline{\xi} |<\varrho \}$, respectively,
711: where $\varrho$
712: is defined by
713: \begin{equation}\label{ee5.69}
714: \varrho= \begin{cases}
715: \frac{\epsilon}{2}|\xi-\overline{\xi}| = \epsilon\sqrt{1-\eta^2}, \qquad
716: & 1+\frac{M_0}{2^{1/3}k^{2/3}}\le \frac{2t}{k}< 1+\delta,\\
717: \epsilon, \qquad & 1+\delta \le \frac{2t}{k}.
718: \end{cases}
719: \end{equation}
720: The (small) parameter $0<\epsilon<1$ will be specified below (see
721: \eqref{ee5.97} below).
722: We note that one may choose the curves in $\Sigma^{(3)}$ above so that
723: in $\calO_\xi$, $\calO_{\overline{\xi}}$, they
724: are straight lines
725: which map under $z\mapsto w(z)$ to (finite subsets of )
726: the rays $\Gamma_j$, $j=1,\cdots, 4$,
727: $\Sigma^{(3)}_{-1}\cap\calO_\xi \to \Gamma_4$,
728: $\Sigma^{(3)}_{-2}\cap\calO_\xi \to \Gamma_2$,
729: $\Sigma^{(3)}_{+1}\cap\calO_\xi \to \Gamma_1$,
730: $\Sigma^{(3)}_{+2}\cap\calO_\xi \to \Gamma_3$,
731: and similarly for the neighborhood of $\calO_{\overline{\xi}}$.
732: For $\tau=s$, we define
733: \begin{equation}
734: m_p(z) := \begin{cases}
735: H(w(z)), \qquad &z\in \calO_\xi\setminus \Sigma^{(3)}, \\
736: \overline{H(w(\overline{z}))},
737: &z\in \calO_{\overline{\xi}}\setminus \Sigma^{(3)}, \\
738: I, &z\in \C\setminus
739: \overline{(\calO_\xi\cup \calO_{\overline{\xi}})}. \\
740: \end{cases}
741: \end{equation}
742: \begin{figure}[ht]
743: \centerline{\epsfig{file=sigmar.eps, width=8cm}}
744: \caption{$\Sigma_R:= \Sigma^{(3)}\cup\partial \calO_\xi\cup\partial
745: \calO_{\overline{\xi}}$}
746: \label{fig-sigmar}
747: \end{figure}
748: Let $\Sigma_R:= \Sigma^{(3)}\cup\partial \calO_\xi\cup\partial
749: \calO_{\overline{\xi}}$ as in Figure \ref{fig-sigmar}
750: where $\partial \calO_\xi$ and $\calO_{\overline{\xi}}$ are
751: oriented counterclockwise.
752: Clearly $m_p$ solves a RHP on $\Sigma_R$ : $m_p(z)$ is
753: analytic in $\C\setminus \Sigma^{(3)}$, $m_p(z)\to I$ as $z\to\infty$,
754: and $m_{p+}(z)=m_{p-}(z)v_p(z)$ for $z\in\Sigma_R$
755: for a suitable jump matrix $v_p$.
756: Set $R(z):= m^{(3)}(z)m_p(z)^{-1}$.
757: Then $R_+(z)=R_-(z)v_R(z)$ for $z\in\Sigma_R$ where
758: $v_R=m_{p-}v^{(3)}v_p^{-1}m_{p-}^{-1}$.
759: Now we estimate $|v_R(z)-I|$.
760:
761: \begin{itemize} %% main estimates %%
762: \item
763: For $z\in\partial \calO_\xi$, we have
764: from \eqref{ee5.69},
765: when $1+\frac{M_0}{2^{1/3}k^{2/3}}\le \frac{2t}{k}\le 1+\delta$,
766: \begin{equation}
767: \begin{split}
768: |w(z)| &= \epsilon(1-\eta^2)^{3/4}\sqrt{2t} \\
769: &\ge \epsilon \biggl(\frac{2t}{k}\biggr)^{-1/4}
770: \biggl( k^{2/3}\bigl(\frac{2t}{k}-1\bigr)\biggr)^{3/4} \\
771: &\ge \frac{\epsilon M_0^{3/4}}{2^{1/4}(1+\delta)^{1/4}}.
772: \end{split}
773: \end{equation}
774: When $1+\delta\le \frac{2t}{k}$,
775: \begin{equation}
776: |w(z)| = \sqrt{2t}(1-\eta^2)\epsilon
777: \ge \biggl( \frac{\delta}{1+\delta}\biggr)^{1/4}\epsilon \sqrt{2t}.
778: \end{equation}
779: Thus if we have taken $M_0$ large, and $t$ is large, we have
780: for $z\in \partial \calO_\xi$,
781: from \eqref{ee5.67},
782: \begin{equation}
783: m_{p+}(z) = I+ O\biggl(\frac{1}{\min(M_0^{3/4},\sqrt{t})}\biggr).
784: \end{equation}
785: But as $v^{(3)}=I$ on $\partial\calO_\xi$,
786: $v_R(z)=v_p(z)^{-1}= m_{p+}(z)^{-1}$, and hence
787: \begin{equation}\label{ee5.72}
788: \|v_R(z)-I\|_{L^\infty(\partial\calO_\xi)}
789: \le \frac{C}{\min(M_0^{3/4},\sqrt{t})}
790: \qquad z\in \partial \calO_\xi.
791: \end{equation}
792: We are using here the standard fact that if $\det v_p=1$, then $\det m_p=1$.
793: Similarly, we have the same estimate \eqref{ee5.72} on $v_R(z)$ for
794: $z\in \partial \calO_{\overline{\xi}}$.
795: \item For $z\in \Sigma^{(3)}\cap \calO_\xi$,
796: since $m_p$ and $m_p^{-1}$ are uniformly bounded,
797: $|v_R(z)-I| \le C|v^{(3)}(z)v_p(z)^{-1} -I|$.
798: For $z\in \Sigma^{(3)}_{-1} \cap \calO_\xi$,
799: by \eqref{e5.46}, \eqref{e5.67},
800: \begin{equation}\label{ee5.74}
801: \begin{split}
802: |v^{(3)}(z)v_p(z)^{-1} -I|
803: &\le | v^{(3)}_{21}(z)- (v_H)_{21}(w(z))|\\
804: &= s| e^{2tf(z)}\delta^{-2}(z)-a(\xi)^2w(z)^{2i\nu}e^{-\frac12iw(z)^2}|
805: =:s |\Delta|.
806: \end{split}
807: \end{equation}
808: Setting $u:=i\xi^{-1}(z-\xi)$, we have
809: \begin{equation}
810: \begin{split}
811: \Delta &= e^{2t(\frac12(\xi(1-iu)-\frac1{\xi(1-iu)})+\eta\log \xi(1-iu))}
812: \biggl( \frac{-i\xi u}{\xi(1-iu)-\xi^{-1}} \biggr)^{2i\nu} \\
813: &\qquad - e^{2t(\frac12(\xi-\xi^{-1})+\eta\log \xi)-ti(1-\eta^2)^{1/2}u^2}
814: \biggl( \frac{-i\xi u}{\xi-\xi^{-1}} \biggr)^{2i\nu} \\
815: &=\bigl( e^{2th(u)}j(u)-1 \bigr)
816: e^{2t(\frac12(\xi-\xi^{-1})+\eta\log \xi)-ti(1-\eta^2)^{1/2}u^2}
817: \biggl( \frac{-i\xi u}{\xi-\xi^{-1}} \biggr)^{2i\nu},
818: \end{split}
819: \end{equation}
820: where
821: \begin{equation}
822: \begin{split}
823: h(u) &= \frac12 ( -i\xi u -\frac{i u}{\xi(1-iu)} )
824: + \eta\log(1-iu) + \frac12 i(1-\eta^2)^{1/2}u^2 \\
825: &= (-\frac16 i\eta+\frac12\sqrt{1-\eta^2}) u^3 +O(u^4),
826: \end{split}
827: \end{equation}
828: and
829: \begin{equation}
830: j(u)= \biggl( \frac{\xi-\xi^{-1}}{\xi-\xi^{-1}-i\xi u} \biggr)^{2i\nu}
831: = 1+ O(\frac{u}{\xi-\xi^{-1}}) = 1+O(\frac{u}{\sqrt{1-\eta^2}}).
832: \end{equation}
833: Also, as $|\frac{u}{\sqrt{1-\eta^2}}| \le c\epsilon$ for $z\in \calO_\xi$,
834: we have
835: \begin{eqnarray}
836: | h(u)| &\le & c |u|^3, \\
837: %\frac12\sqrt{1-\eta^2} |u|^3+ O(|u|^4)
838: %= c\sqrt{1-\eta^2}|u|^3, \\
839: j(u) &=& 1+O(\frac{u}{\sqrt{1-\eta^2}}) = 1+O(\epsilon).
840: \end{eqnarray}
841: On the other hand, for $z\in\Sigma^{(3)}_{-1}\cap\calO_\xi$,
842: \begin{equation}
843: Re (-it\sqrt{1-\eta^2} u^2) \le -ct\sqrt{1-\eta^2} |u|^2,
844: \qquad c=\cos\frac\pi6>0.
845: \end{equation}
846: Therefore, we obtain
847: \begin{equation}
848: \begin{split}
849: |\Delta|
850: &\le C|(e^{2th(u)}-1)+e^{2th(u)}(j(u)-1)| e^{-ct\sqrt{1-\eta^2} |u|^2}\\
851: &\le C\bigl(|2th(u)| + |j(u)-1|\bigr)
852: e^{-ct\sqrt{1-\eta^2} |u|^2 + |Re (2th(u))|} \\
853: &\le C(t|u|^3 + \epsilon) e^{-ct\sqrt{1-\eta^2} |u|^2}\\
854: &\le \frac{C}{\sqrt{2t}(1-\eta^2)^{3/4}} + C\epsilon,
855: \end{split}
856: \end{equation}
857: where for the last inequality, we have used the fact that
858: $|x^3e^{-x^2}|$ is uniformly bounded for $x\in\R$.
859: Now for $1+\frac{M_0}{2^{1/3}k^{2/3}} \le \frac{2t}{k} <1+\delta$,
860: we have
861: \begin{equation}
862: \sqrt{2t}(1-\eta^2)^{3/4}
863: > \sqrt{2t}(1-\eta)^{3/4}
864: = \biggl( \frac{k}{2t} \biggr)^{1/4}
865: \biggr( k^{2/3}(\frac{2t}{k}-1)\biggr)^{3/4} \ge
866: \frac{1}{(1+\delta)^{1/4}}\biggl( \frac{M_0}{2^{1/3}} \biggr)^{3/4},
867: \end{equation}
868: and for $1+\delta \le \frac{2t}{k}$, we have
869: \begin{equation}
870: \sqrt{2t}(1-\eta^2)^{3/4} \ge \sqrt{2t} \biggl( \frac{\delta}{1+\delta}
871: \biggr)^{3/4}.
872: \end{equation}
873: Thus we obtain (recall \eqref{ee5.74})
874: \begin{equation}\label{er5.86}
875: \|v_R-I\|_{L^{\infty}(\Sigma^{(3)}_{-1}\cap\calO_\xi)}
876: \le Cs\|\Delta\|_{L^{\infty}(\Sigma^{(3)}_{-1}\cap\calO_\xi)}
877: \le \frac{C}{\min (M_0^{3/4}, \sqrt{t})} + C\epsilon,
878: \end{equation}
879: which is small if we take $M_0, t$ large and $\epsilon$ small.
880: For other parts of $\Sigma^{(3)}\cap\calO_\xi$, by a similar argument,
881: we obtain the same estimate.
882: By the symmetry $m^{(3)}(z)=\overline{m^{(3)}(\overline{z})}$
883: and $m_p(z)=\overline{m_p(\overline{z})}$, we obtain the same estimate
884: for $\Sigma^{(3)}\cap\calO_{\overline{\xi}}$.
885: \item
886: Let $\calO:=\calO_\xi\cup\calO_{\overline{\xi}}$.
887: For $z\in \Sigma^{(3)}_{-1}\cap(\C\setminus\overline\calO)$,
888: $v_R(z)=v^{(3)}(z)$.
889: Thus we need an estimate for $v^{(3)}_{21}(z)=se^{2tf(z)}\delta^{-2}(z)$.
890: Since $|\delta(z)|= e^{-\nu \theta}$
891: where $\theta=arg (\frac{z-\xi}{z-\overline{\xi}})$,
892: $|\delta(z)|$ and $|\delta^{-1}(z)|$ are uniformly bounded.
893: When $1+\delta\le \frac{2t}{k}$,
894: $dist(\Sigma^{(3)}_{-1}\cap\overline{\calO}^{c}, \{ \xi, \xi^{-1}\}$
895: is uniformly bounded below.
896: From this fact, one can check that we can take $\Sigma^{(3)}_{-1}$
897: so that $Re (f(z))\le -c_0(\epsilon)$
898: for $z\in\Sigma^{(3)}_{-1}\cap\overline{\calO}^{c}$
899: for some constant $c_0(\epsilon)>0$ depending on $\epsilon$.
900: Hence we have
901: \begin{equation}\label{en1}
902: \| v_R -I \|_{L^\infty(\Sigma^{(3)}_{-1}\cap\overline{\calO}^{c})}
903: \le Ce^{-c_0(\epsilon) t},
904: \qquad 1+\delta\le \frac{2t}{k}.
905: \end{equation}
906: On the other hand, when $1+\frac{M_0}{2^{1/3}k^{2/3}}\le
907: \frac{2t}{k}<1+\delta$, we take
908: $\Sigma^{(3)}_{-1}=\{ \rho(\theta) e^{i\theta} : |\theta|< \theta_c\}$
909: such that
910: \begin{enumerate}
911: \item[(i)] For $z\in\Sigma^{(3)}_{-1}$ with
912: $\frac{2\pi}3< | arg(z)| < \theta_c$, $\Sigma^{(3)}_{-1}$ is
913: a pair of straight
914: lines which meet the unit circle at $\xi$ and $\overline{\xi}$, respectively,
915: with angle $\pi/3$.
916: \item[(ii)] For $\rho e^{i\theta}\in \Sigma^{(3)}_{-1}$
917: with $| arg(z)|\le \frac{2\pi}3$,
918: $Re (f(\rho e^{i\theta})) \le Re (f(\rho' e^{i\theta}))$
919: for $\rho\le \rho'\le 1$.
920: Also $\rho(\theta)$ is an increasing function for $0<\theta<\frac{2\pi}3$
921: and is a decreasing function for $-\frac{2\pi}3<\theta<0$.
922: \end{enumerate}
923: (Here the precise value $\frac{2\pi}3$ is of no importance : any angle
924: between $\pi/2$ and $\pi$ will do.)
925: Condition (ii) can be achieved by choosing $\Sigma^{(3)}_{-1}$
926: always to be above the curve $\{ \rho_\theta e^{i\theta} : \frac{\pi}2 <
927: \theta \le \theta_c \}$ (recall \eqref{ee5.36} and Figures
928: \ref{fig-fun}, \ref{fig-crit}).
929: Condition (i) can be achieved as the curve
930: $\{ \rho_\theta e^{i\theta} : \frac{\pi}2 < \theta \le \theta_c \}$
931: crosses the unit circle at $90$ degree (see Figure \ref{fig-crit})
932: For $z$ in (i) satisfying $arg(z)>0$, we have
933: $z=\xi(1-i r e^{-\frac{\pi}3 i})$ for some real $r>0$.
934: We note that $r\le \frac{\sqrt{3}}2<\frac{2}{\sqrt{3}}$.
935: For such $z$, we have (recall \eqref{er5.34})
936: \begin{equation}\label{ec2}
937: Re (f(z))= A(r)\sqrt{1-\eta^2} + B(r) \eta,
938: \end{equation}
939: where
940: \begin{equation}
941: A(r)= \frac{r^2(r-\sqrt3)}{4(1-\sqrt3 r +r^2)}, \qquad
942: B(r)= -\frac{r(r-\sqrt3)(2-\sqrt3 r)}{4(1-\sqrt3 r +r^2)}
943: + \frac12 \log(1-\sqrt3 r +r^2).
944: \end{equation}
945: One can easily check that $A(r)<0$ for $0<r<\sqrt{3}$
946: and $B(r)<0$ for $0<r<\frac{2}{\sqrt{3}}$.
947: Thus for $z$ in (i) satisfying $arg(z)>0$, we have
948: for some $c>0$,
949: \begin{equation}\label{ez2}
950: Re (f(z)) \le A(r)\sqrt{1-\eta^2}\le -cr^2\sqrt{1-\eta^2}.
951: \end{equation}
952: For $z$ in (ii), note first that
953: for fixed $0<\rho<1$, $Re (f(\rho e^{i\theta})$ is an increasing function
954: in $0\le \theta<\pi$.
955: Let $z_b$ be the point on $\Sigma^{(3)}_{-1}$ satisfying $arg(z)=\frac{2\pi}3$.
956: Thus together with the condition (ii),
957: we obtain for $z$ in (ii) satisfying $arg(z)>0$,
958: \begin{equation}
959: Re (f(z)) \le Re (f(z_b)) \le -c|z_b-\xi|^2\sqrt{1-\eta^2}
960: \le -c|z-\xi|^2\sqrt{1-\eta^2}.
961: \end{equation}
962: Here the second inequality follows from \eqref{ez2}.
963: Thus we have
964: for $z \in \Sigma^{(3)}_{-1}\cap\overline{\calO_\xi}^c$
965: with $arg(z)>0$,
966: \begin{equation}\label{er92}
967: |v_R(z)-I| \le
968: \begin{cases}
969: Ce^{-c_0(\epsilon) t}, \qquad &1+\delta\le \frac{2t}{k},\\
970: Ce^{-ct|z-\xi|^2\sqrt{1-\eta^2}}, \qquad &1+\frac{M_0}{2^{1/3}k^{2/3}}
971: \le \frac{2t}{k}< 1+\delta.
972: \end{cases}
973: \end{equation}
974: By symmetry, we have similar estimates for
975: $z\in \Sigma^{(3)}_{-1}$ with $arg(z)<0$.
976: Since $|z-\xi|> \varrho$ for $z\in\overline{\calO}^c$,
977: the above estimates imply in particular that
978: \begin{equation}\label{er5.89}
979: \| v_R -I\|_{L^{\infty}(\Sigma^{(3)}_{-1}\cap(\C\setminus\overline\calO))}
980: \le
981: \begin{cases}
982: Ce^{-c\epsilon^2 M_0^{3/2}}, \qquad
983: & 1+\frac{M_0}{2^{1/3}k^{2/3}}\le \frac{2t}{k}<1+\delta, \\
984: Ce^{-c_0(\epsilon) t}, \qquad & 1+\delta\le \frac{2t}{k},
985: \end{cases}
986: \end{equation}
987: For $\Sigma^{(3)}_{+1}\cap(\C\setminus\overline\calO)$,
988: by the symmetry $Re (f(\rho e^{i\theta})) = Re (f(\rho^{-1} e^{i\theta})$,
989: we have the same estimate.
990: Also by a similar argument, we obtain a similar estimate
991: for $(\Sigma^{(3)}_{-2}\cup\Sigma^{(3)}_{+2})\cap(\C\setminus\overline\calO)$.
992: \end{itemize}
993:
994: %%%%%% new %%%%%%%%%%
995: %Let $z_b=\rho_b e^{i\theta_b}$ be
996: %the intersection point of $\Sigma^{(3)}_{-1}$
997: %and $\partial\calO_\xi$.
998: %Now from \eqref{er5.35}, one sees that
999: %for fixed $0<\rho<1$, $Re (f(\rho e^{i\theta})$ is an increasing function
1000: %in $0\le \theta<\pi$.
1001: %Also we can take $\Sigma^{(3)}_{-1}$ so that
1002: %for fixed $0\le \theta \le \theta_b$,
1003: %$Re (f(\rho e^{i\theta})) \le Re (f(\rho_b e^{i\theta}))$
1004: %for $\rho e^{i\theta}\in \Sigma^{(3)}_{-1}$.
1005: %(This can be done by choosing $\Sigma^{(3)}_{-1}$ always to be
1006: %above the curve
1007: %$\{ \rho_{\theta}e^{i\theta} : \frac{\pi}2<\theta \le \theta_c \}$.)
1008: %Thus, by using the symmetry $f(\rho e^{i\theta})= f(\rho e^{-i\theta})$,
1009: %we have
1010: %\begin{equation}
1011: % Re (f(z)) \le Re (f(z_b)),
1012: %\qquad z\in \Sigma^{(3)}_{-1}\cap (\C\setminus \overline\calO).
1013: %\end{equation}
1014: %Now since we take $\Sigma^{(3)}\cap\partial\calO_\xi$ as straight lines
1015: %which intersect at $\xi$ at angle $\pi/3$ (this precise value is
1016: %of no importance :
1017: %any fixed angle between $0$ and $\pi/2$ would work),
1018: %the point $z_b$ is given by $z_b= \xi(1-i\varrho e^{-i\pi/3})$
1019: %(recall \eqref{ee5.69}).
1020: %Then one can easily check that
1021: %if we take $\epsilon$ small,
1022: %\begin{equation}
1023: % Re (f(z_b))\le -c\sqrt{1-\eta^2}\varrho^2.
1024: %\end{equation}
1025: %Thus we obtain
1026: %\begin{equation}\label{er5.89}
1027: % \| v_R -I\|_{L^{\infty}(\Sigma^{(3)}_{-1}\cap(\C\setminus\overline\calO))}
1028: %\le Ce^{-ct\sqrt{1-\eta^2}\varrho^2} \le
1029: %\begin{cases}
1030: %Ce^{-c\epsilon^2 M_0^{3/2}}, \qquad
1031: %& 1+\frac{M_0}{2^{1/3}k^{2/3}}\le \frac{2t}{k}<1+\delta, \\
1032: %Ce^{-c\epsilon^2 t}, \qquad & 1+\delta\le \frac{2t}{k},
1033: %\end{cases}
1034: %\end{equation}
1035: %which is small if we take $t, M_0$ large after we fix $\epsilon$.
1036: %For $\Sigma^{(3)}_{+1}\cap(\C\setminus\overline\calO)$,
1037: %by the symmetry $Re (f(\rho e^{i\theta})) = Re (f(\rho^{-1} e^{i\theta})$,
1038: %we have the same estimate.
1039: %Also by a similar argument, we obtain the same estimate
1040: %for $(\Sigma^{(3)}_{-2}\cup\Sigma^{(3)}_{+2})\cap(\C\setminus\overline\calO)$.
1041: %\end{itemize}
1042:
1043: As usual, define an operator on $L^2(\Sigma_R)$,
1044: \begin{equation}
1045: C_{v_R}(f) = C_-(f(v_R-I))
1046: \end{equation}
1047: where $C_-$ is the Cauchy operator
1048: \begin{equation}
1049: (C_-f)(z)= \lim_{z'\to z} \frac1{2\pi i}\int_{\Sigma_R} \frac{f(s)}{s-z'} ds,
1050: \qquad \text{$z\in\Sigma_R$,\ \ $z'$ on the $-$ side of $\Sigma_R$.}
1051: \end{equation}
1052: As the Cauchy operator is scale invariant,
1053: $C_-$ is bounded from $L^2(\Sigma_R)\to L^2(\Sigma_R)$
1054: uniformly for $\frac{2t}{k}\ge 1+\frac{M_0}{2^{1/3}k^{2/3}}$,
1055: %and $\|v_R-I\|_{L^\infty(\Sigma_R)}\le Ct^{-1/2}+Ce^{-ct^{1/3}}<1$
1056: %when $t$ is large,
1057: and we have $\|C_{v_R}\|<\frac12$
1058: for $t$, $M_0$ sufficiently large by \eqref{ee5.72}, \eqref{er5.86}
1059: and \eqref{er5.89}.
1060: Hence $1-C_{v_R}$ is invertible.
1061: By standard facts in Riemann-Hilbert theory (see \cite{CG, BC}),
1062: the solution $R(z)$ to the RHP $(\Sigma_R,v_R)$ is given by
1063: \begin{equation}
1064: R(z)= I+ \frac1{2\pi i}
1065: \int_{\Sigma_R} \frac{(I+(1-C_{v_R})^{-1}C_{v_R}I)(v_R-I)(s)}{s-z} ds.
1066: \end{equation}
1067: As $m_p(0)=I$, we have $m^{(3)}_{11}(0)=R_{11}(0)$.
1068: By using $dist (0, \Sigma_R)>0$,
1069: $\|(1-C_{v_R})^{-1}\|\le c$,
1070: and $\|C_-\|\le c$, we have
1071: \begin{equation}\label{ee5.92}
1072: \begin{split}
1073: |m^{(3)}_{11}(0)-1|
1074: &\le c\|v_R-I\|_{L^1(\Sigma_R)}
1075: + c\|(1-C_{v_R})^{-1}C_{v_R}I\|_{L^2(\Sigma_R)}\|v_R-I\|_{L^2(\Sigma_R)}\\
1076: &\le c\|v_R-I\|_{L^1}
1077: + c\|(1-C_{v_R})^{-1}\|_{L^2\to L^2}\|C_-(v_R-I)\|_{L^2}
1078: \|v_R-I\|_{L^2}\\
1079: &\le c\|v_R-I\|_{L^1}
1080: + c \|v_R-I\|^2_{L^2}\\
1081: &\le c\|v_R-I\|_{L^1}
1082: + c \|v_R-I\|_{L^\infty} \|v_R-I\|_{L^1}\\
1083: &\le c\|v_R-I\|_{L^1(\Sigma_R)}
1084: \end{split}
1085: \end{equation}
1086: as $\|v_R-I\|_{L^\infty}$ is bounded.
1087: We estimate $\|v_R-I\|_{L^1}$ in each part of $\Sigma_R$.
1088: First, for $\partial\calO$ and $\Sigma^{(3)}\cap\calO$,
1089: since the length of the contour is of order $\varrho$, we obtain
1090: by \eqref{ee5.72}, \eqref{er5.86}
1091: %(we take $\sqrt{t}>M_0^{3/4}$)
1092: \begin{equation}\label{ee5.93}
1093: \|v_R-I\|_{L^1(\Sigma_R\cap\overline{\calO})}
1094: \le C\varrho\biggl(\frac1{\min (M_0^{3/4}, \sqrt{t})}+\epsilon\biggr).
1095: \end{equation}
1096: When $1+\frac{M_0}{2^{1/3}k^{2/3}}\le \frac{2t}{k}<1+\delta$,
1097: by \eqref{ee5.69}, $\varrho= \epsilon \sqrt{1-\eta^2}$.
1098: When $1+\delta \le \frac{2t}{k}$,
1099: $\sqrt{1-\eta^2}=\sqrt{1-(k/(2t))^2} \ge C$,
1100: and hence we have $\varrho=\epsilon\le c\sqrt{1-\eta^2}$.
1101: Thus in both cases, we obtain
1102: \begin{equation}\label{ee5.94}
1103: \|v_R-I\|_{L^1(\Sigma_R\cap\overline{\calO})}
1104: \le C\sqrt{1-\eta^2}\biggl(\frac1{\min (M_0^{3/4}, \sqrt{t})}+\epsilon\biggr).
1105: \end{equation}
1106: Now we compute $\|v_R-I\|_{L^1(\Sigma_R\cap\overline{\calO}^c)}$.
1107: We first focus on $\Sigma^{(3)}_{-1}\cap\overline{\calO}^c\cap\{ Im(z)>0\}$
1108: When $1+\delta\le \frac{2t}{k}$,
1109: by \eqref{er92},
1110: \begin{equation}
1111: \|v_R-I\|_{L^1(\Sigma^{(3)}_{-1}\cap\overline{\calO}^c\cap\{ Im(z)>0\})}
1112: \le Ce^{-c_0(\epsilon)t}
1113: \le \frac{C}{\sqrt{t}} \sqrt{1-\eta^2},
1114: \end{equation}
1115: for large $t$
1116: as $\sqrt{1-\eta^2}\ge C$ in this case.
1117: When $1+\frac{M_0}{2^{1/3}k^{2/3}}\le \frac{2t}{k}<1+\delta$,
1118: from \eqref{er92},
1119: \begin{equation}
1120: \begin{split}
1121: \|v_R-I\|_{L^1(\Sigma^{(3)}_{-1}\cap\overline{\calO}^c\cap\{ Im(z)>0\})}
1122: &\le \int_{\Sigma^{(3)}_{-1}\cap\overline{\calO}^c\cap\{ Im(z)>0\}}
1123: Ce^{-ct\sqrt{1-\eta^2}|z-\xi|^2}|dz|\\
1124: &\le C\int_{\varrho}^{\infty} e^{-ct\sqrt{1-\eta^2}r^2} dr\\
1125: %&\le C\int_{(t\sqrt{1-\eta^2})^{1/2}\varrho}^\infty
1126: %e^{-cs^2} \frac{ds}{(t\sqrt{1-\eta^2})^{1/2}}\\
1127: &\le \frac{C}{(t\sqrt{1-\eta^2})^{1/2}}
1128: e^{-ct\sqrt{1-\eta^2}\varrho^2}.
1129: \end{split}
1130: \end{equation}
1131: But since, for $1+\frac{M_0}{2^{1/3}k^{2/3}}\le \frac{2t}{k}<1+\delta$,
1132: \begin{equation}
1133: \sqrt{1-\eta^2}\ge \sqrt{1-\frac{k}{2t}}
1134: =\frac1{(2t)^{1/3}}\biggl(\frac{k}{2t}\biggr)^{1/6}
1135: \sqrt{k^{2/3}\biggl(\frac{2t}{k}-1\biggr)}
1136: \ge \frac{CM_0^{1/3}}{t^{1/3}},
1137: \end{equation}
1138: we obtain
1139: \begin{equation}\label{er103}
1140: \|v_R-I\|_{L^1(\Sigma^{(3)}_{-1}\cap\overline{\calO}^c\cap\{ Im(z)>0\})}
1141: \le \frac{C}{M_0^{3/4}}
1142: e^{-c\epsilon^2M_0^{3/2}}\sqrt{1-\eta^2}.
1143: \end{equation}
1144: By a similar computation, we obtain the same estimate
1145: for the other parts of $\Sigma_R\cap\overline{\calO}^c$.
1146: Thus if we take $\epsilon$ small, and then take $M_0$, $t$ large,
1147: we obtain by \eqref{ee5.92}, \eqref{ee5.93} and \eqref{er103},
1148: \begin{equation}\label{ee5.97}
1149: \|m^{(3)}_{11}-1\|_{L^1(\Sigma_R)} \le \alpha\sqrt{1-\eta^2},
1150: \end{equation}
1151: with a constant $\alpha>0$ which can be taken to be arbitrarily small.
1152: Therefore, from \eqref{ee5.46}, \eqref{ee5.92},
1153: using \eqref{ee5.97},
1154: we obtain (note \eqref{ee5.47})
1155: for large $t$,
1156: \begin{equation}
1157: \log m_{11}(0)= \log m^{(3)}_{11}(0)+(1-\frac{\theta_c}{\pi})\log(1-s^2)
1158: \le \alpha \sqrt{1-\eta^2} -c \sqrt{1-\eta^2}
1159: \le -C\sqrt{1-\eta^2},
1160: \end{equation}
1161: for some $C>0$, which is \eqref{ee1}.
1162: