1: \section{RHP analysis}\label{sec-rhp}
2:
3:
4:
5: Set
6: \begin{equation}
7: \Gamma = \overline{\Gamma_1\cup\Gamma_2},
8: \end{equation}
9: oriented counter-clockwise.
10: It is a simple closed curve which has $0$ and $-t$ inside and
11: $-t^{-1}$ outside.
12: Since the jump matrix
13: $V_Y=\bigl( \begin{smallmatrix} 1 & z^{-k}\varphi(z) \\ 0 & 1
14: \end{smallmatrix} \bigr)$
15: for $Y$ in \eqref{e-Y} is analytic in
16: $\C\setminus \{0\}$, we can deform the contour $\Sigma$ for $Y$ to $\Gamma$,
17: as follows.
18: Set
19: \begin{equation}
20: \widetilde{Y}(z)=
21: \begin{cases}
22: Y(z) \qquad \text{for $z$ inside both $\Gamma$ and $\Sigma$,
23: and for $z$ outside both $\Gamma$ and $\Sigma$,} \\
24: Y(z)V_Y(z) \qquad \text{for $z$ inside $\Gamma$ and outside $\Sigma$,} \\
25: Y(z)V_Y^{-1}(z) \qquad \text{for $z$ outside $\Gamma$ and inside $\Sigma$.}
26: \end{cases}
27: \end{equation}
28: Then $\widetilde{Y}$ is analytic in $\C\setminus\Gamma$ and continuous
29: up to the boundaries, satisfies $\widetilde{Y}_+(z)=\widetilde{Y}_-V_Y(z)$
30: for $z\in \Gamma$, and $\widetilde{Y}(z)z^{-k\sigma_3}= I+O(z^{-1})$
31: as $z\to\infty$.
32:
33: Now (see the Introduction) we define
34: \begin{equation}
35: M(z) = e^{-\frac12\ell k \sigma_3} \widetilde{Y}(z)
36: e^{-(g(z)-\frac12\ell)k\sigma_3}.
37: \end{equation}
38: Then from Proposition \ref{prop-g} (1), (2),
39: $M$ satisfies
40: \begin{equation}\label{e-M}
41: \begin{cases}
42: M(z) \text{ is analytic in $z\in\C\setminus \Gamma$}, \\
43: M_+(z)= M_-(z) \begin{pmatrix} e^{-k(g_+(z)-g_-(z))}
44: & e^{k(g_+(z)+g_-(z)-W(z)-\ell)} \\ 0&e^{k(g_+(z)-g_-(z))}
45: \end{pmatrix},
46: \qquad z\in\Gamma, \\
47: M(z) = I + O(1/z), \qquad \text{as $z\to\infty$}.
48: \end{cases}
49: \end{equation}
50: From the Proposition \ref{prop-g} (3), (4), the jump matrix $V$ for $M$ is now
51: \begin{equation}
52: V(z) = \begin{pmatrix} e^{-k\Psi_{1+}(z)} & 1 \\ 0 &
53: e^{k\Psi_{1+}(z)}
54: \end{pmatrix},
55: \qquad z\in\Gamma_1,
56: \end{equation}
57: %using $\Phi_+(z)=-\Phi_-(z)$ for $z\in\Gamma_1$,
58: and from the Proposition \ref{prop-g} (1), (5), we have
59: \begin{equation}
60: V(z) = \begin{pmatrix} 1 & e^{k\Psi_2(z)} \\ 0 & 1 \end{pmatrix},
61: \qquad z\in\Gamma_2.
62: \end{equation}
63:
64: For the jump matrix on $z\in\Gamma_1$, note that $\Psi_+(z)=-\Psi(z)_-$ and
65: \begin{equation}
66: V(z)=
67: \begin{pmatrix} e^{-k\Psi_{1+}(z)} & 1 \\ 0 & e^{-k\Psi_{1-}(z)}
68: \end{pmatrix}
69: = \begin{pmatrix} 1&0 \\ e^{-k\Psi_{1-}(z)} & 1 \end{pmatrix}
70: \begin{pmatrix} 0 & 1 \\ -1 & 0 \end{pmatrix}
71: \begin{pmatrix} 1&0 \\ e^{-k\Psi_{1+}(z)} & 1 \end{pmatrix}.
72: \end{equation}
73: Clearly $\Psi_{1+}$ has an analytic continuation to the $(+)$-side
74: of the contour $\Gamma_1$.
75: Now for $z\in\Gamma_1$,
76: it is easy to see that $Re(\Psi_{1_+}(z))=0$, and hence
77: from the Proposition \ref{prop-h} (c) and \eqref{e-hPhi1},
78: the derivative $\frac{d}{dz} \Psi_{1+}(z)= \Phi_+(z)$
79: along the contour $\Gamma_1$ satisfies
80: $Im( \frac{d}{dz}\Psi_{1+}(z))<0$, $z\in\Gamma_1$.
81: Thus the Cauchy-Riemann condition for the analyticity
82: implies that $Re(\Psi_{1}(z))>0$ for $z$ on the $(+)$-side of $\Gamma_1$
83: and close to the contour.
84: Therefore we can take a contour $\Gamma_1^{(1)}$ with endpoints
85: $\xi$, $\overline{\xi}$ for which $Re(\Psi(z))>0$
86: for $z\in int(\Gamma^{(1)})$.
87: Similarly, $\Psi_{1-}$ has an analytic continuation to the $(-)$-side
88: of $\Gamma_1$ and its real part is positive for
89: $z$ on the $(+)$-side of $\Gamma_1$ close to the contour $\Gamma_1$
90: and we take a contour $\Gamma_1^{(2)}$ for which $Re(\Psi(z))>0$
91: on its interior.
92: We take the orientation of $\Gamma_1^{(j)}$, $j=1,2$ to be from
93: $\overline{\xi}$ to $\xi$.
94: See Figure \ref{fig-deformed} for the general shape of the contours
95: $\Gamma_1^{(1)}$, $\Gamma_1^{(2)}$.
96: \begin{figure}[ht]
97: \centerline{\epsfig{file=deformed.eps, width=7cm}}
98: \caption{The contours $\Gamma_1^{(1)}$ and $\Gamma_1^{(2)}$.}
99: \label{fig-deformed}
100: \end{figure}
101:
102:
103: Define $\widetilde{M}(z)$ to be $M(z)$ for $z$ in the region bounded by
104: $\Gamma_2$ and $\Gamma_1^{(1)}$ and also in the unbounded region.
105: For the region bounded by $\Gamma_1$ and $\Gamma_1^{(1)}$, define
106: \begin{equation}
107: \widetilde{M}= M
108: \begin{pmatrix} 1&0 \\ -e^{-k\Psi_{1}(z)} & 1 \end{pmatrix},
109: \end{equation}
110: and for the region bounded by $\Gamma_1$ and $\Gamma_1^{(2)}$, define
111: \begin{equation}
112: \widetilde{M}= M
113: \begin{pmatrix} 1&0 \\ e^{-k\Psi_{1}(z)} & 1 \end{pmatrix}.
114: \end{equation}
115: Set $\Gamma'=\overline{\Gamma_1\cup\Gamma_1^{(1)}
116: \cup\Gamma_1^{(2)}\cup\Gamma_2}$.
117: Then $\widetilde{M}$ satisfies the new RHP
118: \begin{equation}\label{e-wideM}
119: \begin{cases}
120: \widetilde{M}(z) \text{ is analytic in $z\in\C\setminus \Gamma'$}, \\
121: \widetilde{M}_+(z)= \widetilde{M}_-(z) \widetilde{V}(z)
122: \qquad z\in\Gamma', \\
123: \widetilde{M}(z) = I + O(1/z), \qquad \text{as $z\to\infty$},
124: \end{cases}
125: \end{equation}
126: where the jump matrix $\widetilde{V}$ is
127: \begin{equation}
128: \widetilde{V}(z)=
129: \begin{cases}
130: \begin{pmatrix} 0&1\\-1&0 \end{pmatrix},
131: \qquad &z\in\Gamma_1, \\
132: \begin{pmatrix} 1&0 \\ e^{-k\Psi_{1}(z)} & 1 \end{pmatrix},
133: \qquad &z\in\Gamma_1^{(1)}\cup\Gamma_1^{(2)}, \\
134: \begin{pmatrix} 1& e^{k\Psi_{2}(z)} \\ 0 & 1 \end{pmatrix},
135: \qquad &z\in\Gamma_2.
136: \end{cases}
137: \end{equation}
138:
139: Now we take the limit $k\to\infty$ with $\gamma\ge 1$ in a compact set
140: and $a_0<a\le a^*$, for some $a^*$.
141: From the signature of $Re(\Psi_1(z))$ on $\Gamma_1^{(j)}$, $j=1,2$,
142: we see that the jump matrix
143: $\widetilde{V} \to I$.
144: For $z\in\Gamma_2$, we have $\Psi_2(z)<0$
145: from the Proposition \ref{prop-g} and \ref{prop-h}.
146: Indeed $\Psi_2(z)$ is decreasing as $z$ follows from $\xi$ to $-z_0$
147: along $\Gamma_2$, and then increasing as $z$ follows from $-z_0$ to
148: $\overline{\xi}$ along $\Gamma_2$.
149: On the other hand,
150: $\Psi_2(\xi)=0$ and
151: $\Psi_2(\overline{\xi})=\int_{\xi}^{\overline{\xi}}\Phi(s)ds =0$,
152: as $\Phi(s)ds=\overline{\Phi(s)ds}=-\Phi(\overline{s})d\overline{s}$,
153: and the negativity of $\Psi_2(z)$ on $\Gamma_2$ follows.
154: Hence as $k\to\infty$, $\widetilde{V} \to I$ on $\Gamma_2$.
155: Therefore we have $\widetilde{V} \to V^\infty$ where
156: \begin{equation}
157: V^\infty=
158: \begin{cases}
159: \begin{pmatrix} 0&1\\-1&0 \end{pmatrix}, \qquad &z\in \Gamma_1,\\
160: I \qquad &z\in\Gamma_1^{(1)}\cup\Gamma_1^{(2)}\cup\Gamma_2.
161: \end{cases}
162: \end{equation}
163: Let $M^\infty$ be the solution to the RHP with the jump matrix $V^\infty$
164: and normalized at infinity.
165: The solution is given by
166: \begin{equation}
167: M^\infty(z) = \begin{pmatrix}
168: \frac{\beta+\beta^{-1}}2 & \frac{\beta-\beta^{-1}}{2i} \\
169: - \frac{\beta-\beta^{-1}}{2i} & \frac{\beta+\beta^{-1}}2
170: \end{pmatrix},
171: \end{equation}
172: where
173: \begin{equation}
174: \beta(z) = \biggl( \frac{z-\xi}{z-\overline{\xi}} \biggr)^{1/4}
175: \end{equation}
176: which is defined to be analytic $\C\setminus\Gamma_1$
177: and satisfies $\beta(z)\sim 1$ as $z\to\infty$.
178: We expect that $\widetilde{M} \sim M^\infty$ as $k\to\infty$, and hence
179: by tracking the algebraic transformations $Y\to M\to \widetilde{M}$,
180: we expect that
181: \begin{equation}\label{e-YMinfty}
182: Y_{21}(0)e^{-k(g(0)-\ell)} \sim M^\infty_{21}(0),
183: \qquad k\to\infty,
184: \end{equation}
185: for $\gamma\ge 1$ in a compact set and $a_0<a\le a^*$.
186:
187: Indeed we have:
188: \begin{prop}
189: Let $1\le \gamma\le \gamma_1$ for any fixed $\gamma_1\ge 1$.
190: There are $L_1, \delta_1>0$ such that
191: for
192: \begin{equation}
193: a_0 + \frac{L_1}{k^{2/3}} \le a\le (1+\delta_1)a_0,
194: \end{equation}
195: we have
196: \begin{equation}\label{e-YMinfty1}
197: Y_{21}(0)e^{-k(g(0)-\ell)} = M^\infty_{21}(0)
198: \bigl( 1+ O(\frac1{k|a-a_0|})\bigr),
199: \end{equation}
200: for sufficiently large $k$.
201: \end{prop}
202:
203: The convergence $\widetilde{V} \to V^\infty$ is not uniform on
204: $\Gamma'$ and this considerably complicates the analysis. As in
205: \cite{DKMVZ3} in order to obtain the above error bound, we need to
206: introduce local parametrices for the solution of the RHP near each of
207: the endpoints $\xi$ and $\overline{\xi}$. As in \cite{BDJ}, a
208: suitable local parametrix near each endpoint can be obtained in terms
209: of Airy functions. Also since $a$ is not fixed, but is allowed to
210: approach $a_0$, we need to vary the magnitude of the parametrix
211: according to the size of $a-a_0$.
212: %This is exactly the situation
213: %illustrated in Figure~\ref{fig:contours_aseq}; as the endpoints
214: %coalesce and come within a distance proportional to $1/k$ the local
215: %analysis needs to be revised and a single local parametrix must be
216: %constructed in terms of parabolic cylinder functions to simultaneously
217: %account for nonuniformities near both endpoints.
218: A similar situation
219: arises in Lemma 6.2 (ii) of \cite{BDJ}. The proof of the above
220: Proposition is parallel to that of Lemma 6.2 (ii), \cite{BDJ}
221: ($\gamma$, $q$ in \cite{BDJ} play the same role as $a$, $k$ in this
222: paper, respectively), and we do not repeat the argument here.
223:
224: \bigskip
225:
226: \begin{lem}\label{lem-real}
227: We have $\Delta:=g(0+i0)-\ell\in \R$.
228: In particular, $e^{g(0)-\ell}=e^{g(0+i0)-\ell}>0$.
229: \end{lem}
230:
231: \begin{proof}
232: For $x\in\R\setminus\{p_i\}$, by Remark \ref{rem-real},
233: $g(x+i0)=\overline{g(x-i0)}$, and hence by Proposition \ref{prop-g} (1),
234: $e^{g(x+i0)}=e^{g(x-i0)}=e^{\overline{g(x+i0)}}$,
235: and so $e^{g(x)}=e^{g(x+i0)}=e^{g(x-i0)}$ is real.
236: In particular, it follows that $Im(g(x+i0))\in \Z \pi$,
237: and hence by continuity, $Im(g(x+i0))$ is constant for $x<p_i$.
238: From \eqref{e-l} and the proof of Proposition \ref{prop-g} (3),
239: $\ell=g_+(\xi)+g_-(\xi)-W(\xi)= g_+(p_i+i0)+g_-(p_i)-W(p_i)$.
240: Hence $e^{g(0)-\ell}=e^{g(0+i0)-g_+(p_i+i0)}e^{-g_-(p_i)+W(p_i)}$.
241: But by the above, $Im(g(0+i0))=Im(g_+(p_i+i0))$.
242: Hence $g(0+i0)-g_+(p_i+0i)\in\R$.
243: Clearly $g_-(p_i)$ and $W(p_i)$ are also real, and this proves the lemma.
244: \end{proof}
245:
246: If we set $\xi=|\xi|e^{i\theta_c}$, $0<\theta_c<\pi$, then we can check
247: $\beta(0)= e^{i\theta_c/2}$ and
248: $M^\infty_{21}(0)= -\frac1{2i}(\beta(0)-\beta(0)^{-1})
249: = -\sin \frac{\theta_c}2$.
250: Hence the above proposition yields that
251: \begin{equation}\label{e-Yest}
252: -Y_{21}(0) = e^{k\Delta}
253: \sin\frac{\theta_c}2 \bigl(1+O(\frac1{k|a-a_0|})\bigr).
254: \end{equation}
255: Note that $-Y_{21}(0)$ is indeed real and positive from Lemma \ref{lem-GandY}.
256: This is consistent with Lemma \ref{lem-real}.
257:
258: Now we compute $e^{g(0)-\ell}$.
259: Let
260: \begin{equation}
261: \alpha = \frac{\xi+\overline{\xi}}2 = |\xi|\cos\theta_c.
262: \end{equation}
263: From the formula \eqref{e-hprop} for $h$, one can check directly that
264: an anti-derivative of $2(h(z)-\frac1z)$ is
265: \begin{equation}\label{e-hantider}
266: \begin{split}
267: &-\gamma a \log(z+t^{-1})-a\log(z+t) + (a-1)\log(z) \\
268: &+ \biggl( -\frac{\gamma a}{x} -\frac{a}{y} + \frac{a+1}{r} \biggr) R(z)
269: - \biggl( -\frac{\gamma a}{x}(\alpha+t^{-1}) - \frac{a}{y}(\alpha+t)
270: +\frac{a+1}{r}\alpha \biggr) \log(z-\alpha+R(z)) \\
271: &- \gamma a \log\biggl(\frac{z+R(z)+t^{-1}-x}{z+R(z)+t^{-1}+x} \biggr)
272: - a \log\biggl(\frac{z+R(z)+t-y}{z+R(z)+t+y} \biggr)
273: + (a+1) \log\biggl(\frac{z+R(z)-r}{z+R(z)+r} \biggr)
274: \end{split}
275: \end{equation}
276: where the logarithms are taken to be the analytic
277: in $\C\setminus(-\infty, 0]$ and
278: $\log z= \log|z|$ for $z>0$,
279: and $r=-R(0)$, $x=-R(-t^{-1})$, $y=-R(-t)$ as in \eqref{e-rxyxi}.
280: It is straightforward, but tedious, to check that \eqref{e-hantider} is
281: analytic in $\C\setminus(\overline{\Gamma_1}\cup(-\infty, p_i])$
282: as in the definition of $g$ in \eqref{e-g}.
283: Using the endpoint conditions \eqref{e-endprop}, or\eqref{e-contemp},
284: \eqref{e-hantider} is equal to
285: \begin{equation}
286: \begin{split}
287: &-\gamma a \log(z+t^{-1})-a\log(z+t) + (a-1)\log(z)
288: + (1+\gamma a)\log(z-\alpha+R(z)) \\
289: &- \gamma a \log\biggl(\frac{z+R(z)+t^{-1}-x}{z+R(z)+t^{-1}+x} \biggr)
290: - a \log\biggl(\frac{z+R(z)+t-y}{z+R(z)+t+y} \biggr)
291: + (a+1) \log\biggl(\frac{z+R(z)-r}{z+R(z)+r} \biggr).
292: \end{split}
293: \end{equation}
294: Evaluating the asymptotics as $z\to\infty$, we see that
295: \begin{equation}\label{e-gexpl}
296: \begin{split}
297: 2g(z)
298: = & -\gamma a \log(z+t^{-1})-a\log(z+t) + (a+1)\log(z) \\
299: &+ (1+\gamma a)\log\bigl((z-\alpha+R(z))/2\bigr)
300: - \gamma a \log\biggl(\frac{z+R(z)+t^{-1}-x}{z+R(z)+t^{-1}+x} \biggr) \\
301: & - a \log\biggl(\frac{z+R(z)+t-y}{z+R(z)+t+y} \biggr)
302: + (a+1) \log\biggl(\frac{z+R(z)-r}{z+R(z)+r} \biggr),
303: \end{split}
304: \end{equation}
305: and
306: \begin{equation}\label{e-g0expl}
307: \begin{split}
308: 2Re(g(0+i0))
309: = & (\gamma-1) a \log t
310: + (1+\gamma a)\log((\alpha+r)/2) \\
311: &- \gamma a \log\biggl|\frac{-r+t^{-1}-x}{-r+t^{-1}+x} \biggr|
312: - a \log\biggl|\frac{-r+t-y}{-r+t+y} \biggr|
313: + (a+1) \log\biggl(\frac{2r}{1+\alpha/r} \biggr).
314: \end{split}
315: \end{equation}
316: Also from \eqref{e-gexpl} and \eqref{e-l}, we have
317: \begin{equation}\label{e-lexpl}
318: \begin{split}
319: \ell = &\gamma a \log t + (1+\gamma a)\log(\xi-\alpha)/2 \\
320: & - \gamma a \log\biggl(\frac{\xi+t^{-1}-x}{\xi+t^{-1}+x} \biggr)
321: - a \log\biggl(\frac{\xi+t-y}{\xi+t+y} \biggr)
322: + (a+1) \log\biggl(\frac{\xi-r}{\xi+r} \biggr).
323: \end{split}
324: \end{equation}
325: %To emphasize the dependence on $a$ for fixed $\gamma\ge 1$,
326: %we write $\Delta=\Delta(a)$.
327: %We will regard $\Delta(a)$ as a function of $a$ with fixed $\gamma\ge 1$.
328: %Then \eqref{e-Yest} becomes
329: %\begin{equation}
330: % -Y_{21}(0) \le e^{k\Delta(a)}\bigl( 1+ \frac{C}{k|a-a_0|} \bigr).
331: %\end{equation}
332: %Now we use \eqref{e-g0expl}, \eqref{e-lexpl} for $\Delta(a)$ and
333: %put $\xi= re^{i\theta_c}$ (thus $\alpha=r\cos\theta_c$).
334: %From the relation \eqref{e-con3} between $x, y, r$, we have the formula
335: By \eqref{e-xiextra},
336: %\eqref{e-xiextra} for $\cos\theta_c$:
337: \begin{equation}
338: \cos\theta_c= \frac{x^2-r^2-t^{-2}}{2rt^{-1}}
339: = \frac{y^2-r^2-t^2}{2rt}.
340: \end{equation}
341: Thus using \eqref{e-g0expl} and \eqref{e-lexpl},
342: we can express $\Delta$ in terms of $r, x, y$.
343: After some algebra, we find
344: \begin{equation}\label{e-Delta}
345: \begin{split}
346: \Delta =
347: &-\gamma a\log t +(2+a+\gamma a)\log 2 + (1+a)\log r
348: -\frac12 \log(r+t^{-1}-x) \\
349: &-\frac12(1+2\gamma a)\log(r+t^{-1}+x)
350: -\frac12 \log(r+t-y) -\frac12(1+2a)\log(r+t+y).
351: \end{split}
352: \end{equation}
353:
354: \medskip
355:
356: To emphasize the dependence on $a$,
357: we write $\Delta=\Delta(a)$, etc.
358:
359: \begin{lem}\label{lem-Delta}
360: Fix $1\le \gamma_2<\infty$.
361: Then there exists $\delta_2>0$ such that for $a_0\le a\le (1+\delta_2)a_0$
362: and $1\le \gamma\le \gamma_2$,
363: we have
364: \begin{equation}\label{e-Delest}
365: \Delta(a) = -c_2(a-a_0)^2 \bigl(1 +O(|a-a_0|\bigr)
366: \end{equation}
367: where
368: \begin{equation}
369: c_2=\frac{t^2(t+t\gamma+2\sqrt\gamma)^3\sqrt\gamma}
370: {4(1+t\sqrt\gamma)^2(t+\sqrt\gamma)^2},
371: \end{equation}
372: and the order term $O(|a-a_0|)$ is uniform for
373: $a$ and $\gamma$ as above.
374: \end{lem}
375:
376:
377: For the proof, we need the following lemma,
378: which considers the case when $a=a_0$.
379: This case is specifically excluded from Lemma \ref{lem-H1}.
380:
381: \begin{lem}\label{lem-H2}
382: For fixed $0<t<1$, $\gamma\ge 1$,
383: and $a=a_0=\frac{1-t^2}{t((\gamma+1)t+2\sqrt\gamma)}$,
384: there is a unique solution $r$ to \eqref{e-Heq} satisfying
385: $r_1< r <r_2$, given by
386: \begin{equation}\label{e-rata0}
387: r=r(a_0)=r_0:=\frac{1+t\sqrt\gamma}{t+\sqrt\gamma}.
388: \end{equation}
389: The function $a\mapsto r(a)$ is smooth for all $a\ge a_0$ and
390: \begin{equation}\label{e-rpata0}
391: r'(a_0)= -\frac{3(\gamma-1)(t\gamma+t+2\sqrt\gamma)^2t^2}
392: {4(1+t\sqrt\gamma)(t+\sqrt\gamma)^3}.
393: \end{equation}
394: \end{lem}
395:
396: \begin{proof}
397: When $a=a_0$, we have $r_1= tr_0^2$ and $r_2=\frac1{t}r_0^2$.
398: Since $t<r_0<\frac1t$, $r_0$ satisfies $r_1<r_0<r_2$.
399: It is then a direct calculation to check that $r_0$ satisfies the
400: equation \eqref{e-Heq}.
401: Now we want to show the uniqueness of the solution.
402: Let $H$ be as in \eqref{e-H} in the proof of Lemma \ref{lem-H1}.
403: Now when $a=a_0$, the value $r_*$ of \eqref{e-minr0} is $r_0$.
404: Thus we have
405: \begin{equation}
406: H(r)+\frac{r}{2}H'(r) \ge H(r_0)+\frac{r_0}{2}H'(r_0)=0,
407: \end{equation}
408: for $r_1<r<r_2$, and the inequality is strict for $r\neq r_0$.
409: Hence if there is a zero $r_c\neq r_0$, it should satisfy
410: $H'(r_c)>0$.
411: On the other hand, by direct calculation, we have
412: \begin{equation}
413: H'(r_0)=H''(r_0)=0,
414: \qquad H^{(3)}(r_0) =
415: \frac{24t(t+\sqrt\gamma)^5}{\sqrt\gamma(1-t^2)^2(1+t\sqrt\gamma)^3} >0,
416: \end{equation}
417: and hence $H$ is also increasing at the zero $r=r_0$.
418: As in Lemma \ref{lem-H1}, there is no other zero $r_c\neq r_0$
419: in $(r_1, r_2)$.
420:
421: Now consider $H=H(a, r)$.
422: %Now we consider the function $a\mapsto r(a)$, $a\ge a_0$.
423: By direct calculations, we find
424: \begin{equation}
425: H(a_0, r_0)=H_r(a_0, r_0)=H_a(a_0, r_0)= H_{rr}(a_0,r_0)=0
426: \end{equation}
427: and
428: \begin{eqnarray}
429: H_{ra}(a_0,r_0)&=&\frac{4(t+\sqrt\gamma)(t\gamma+t+2\sqrt\gamma)^2t^2}
430: {(1+t\sqrt\gamma)^3(1-t^2)}\neq 0, \\
431: H_{aa}(a_0,r_0)&=& \frac{6(\gamma-1)(t\gamma+t+2\sqrt\gamma)^4t^4}
432: {(1+t\sqrt\gamma)^4(t+\sqrt\gamma)^2(1-t^2)}.
433: \end{eqnarray}
434: Hence near $(a_0, r_0)$, the power series of $H(a,r)$ has the form
435: \begin{equation}
436: H(a,r)= H_{ra}(a_0,r_0)(r-r_0)(a-a_0) + \frac12H_{aa}(a_0,r_0)(a-a_a)^2
437: + \frac16H_{rrr}(a_0,r_0)(r-r_0)^3+\cdots.
438: \end{equation}
439: Motivated by this expansion, we set
440: \begin{equation}
441: \eta:= \frac{r-r_0}{a-a_0},
442: \end{equation}
443: and substitute $r=r_0+\eta(a-a_0)$ in $H$, and define
444: \begin{equation}
445: F(a, \eta) := \frac{H(a, r_0+\eta(a-a_0))}{(a-a_0)^2}.
446: \end{equation}
447: Setting
448: \begin{equation}
449: \eta_0:= -\frac{H_{aa}(a_0,r_0)}{2H_{ra}(a_0, r_0)}
450: = -\frac{3(\gamma-1)(t\gamma+t+2\sqrt\gamma)^2t^2}
451: {4(1+t\sqrt\gamma)(t+\sqrt\gamma)^3},
452: \end{equation}
453: a direct calculation shows that
454: $F(a, \eta)$ is a smooth function near $(a_0, \eta_0)$ and
455: \begin{equation}
456: F(a_0, \eta_0)=0, \qquad F_{\eta}(a_0, \eta_0)=H_{ra}(a_0, r_0)\neq 0.
457: \end{equation}
458: Therefore by the implicit function theorem, there is the smooth function
459: $\eta=\eta(a)$, $a_0\le a<a_0(1+\delta)$ for some $\delta>0$ such that
460: $F(a,\eta(a))=0$.
461: Then $H(a, r_0+\eta(a)(a-a_0))=0$, and by the uniqueness of the solution,
462: $r=r_0+\eta(a)(a-a_0)$ is smooth in $a$ for $a_0\le a<a_0(1+\delta)$.
463: But for $a>a_0$, $H_a(a, r(a))>0$, and the smoothness of $r=r(a)$
464: is elementary.
465: Hence $r(a)$ is a smooth function for $a\ge a_0$.
466: \end{proof}
467:
468:
469: Let
470: \begin{equation}\label{e-xyata0}
471: x_0:= x(a_0)=\frac{\sqrt\gamma(1-t^2)}{t(t+\sqrt\gamma)},
472: \qquad y_0:=y(a_0)=\frac{1-t^2}{t+\sqrt\gamma},
473: \end{equation}
474: which are obtained by setting $a=a_0$, $r=r(a_0)=r_0$ in
475: \eqref{e-con1}, \eqref{e-con2}.
476: Then
477: \begin{equation}
478: \frac{x_0^2-r_0^2-t^{-2}}{2r_0t^{-1}}=-1,
479: \end{equation}
480: and hence from \eqref{e-cosxi}, the point $\xi$ is on the negative
481: real line. Thus when $a=a_0$, the two endpoints $\xi$ and
482: $\overline{\xi}$ collapse to the point $-r_0$ on the real line.
483: This is an extreme case of the deformation illustrated in
484: Figure~\ref{fig:contours_aseq}.
485:
486: \begin{proof}[Proof of Lemma \ref{lem-Delta}]
487: When $a=a_0$, we have $r=r_0$, $x=x_0$, $y=y_0$, and we can direct check
488: from \eqref{e-Delta} that $\Delta(a_0)=0$.
489: We have
490: \begin{equation}
491: \begin{split}
492: \Delta'(a)=
493: &-\gamma \log t +(1+\gamma )\log 2 + \log r
494: - \gamma\log(r+t^{-1}+x) -\log(r+t+y) \\
495: &+ (1+a)\frac{r'}{r} -\frac12 \frac{r'-x'}{r+t^{-1}-x}
496: -\frac12(1+2\gamma a)\frac{r'+x'}{r+t^{-1}+x}
497: -\frac12 \frac{r'-y'}{r+t-y} -\frac12(1+2a)\frac{r'+y'}{r+t+y}.
498: \end{split}
499: \end{equation}
500: At $a=a_0$, from from \eqref{e-rata0}, \eqref{e-xyata0}, we have
501: \begin{equation}\label{e-Deltapptemp1}
502: -\gamma \log t +(1+\gamma )\log 2 + \log r_0- \gamma\log(r_0+t^{-1}+x_0)
503: -\log(r_0+t+y_0) =0
504: \end{equation}
505: and hence again from \eqref{e-rata0}, \eqref{e-xyata0}, after some algebra,
506: \begin{equation}
507: \Delta'(a_0)= -\frac{(\gamma-1)(1-t^2)}
508: {(t+t\gamma+2\sqrt\gamma)(1+t\sqrt\gamma)}
509: \bigl[(1+t\sqrt\gamma)r'(a_0)+t\sqrt\gamma x'(a_0)-y'(a_0)\bigr].
510: \end{equation}
511: Now from the relation \eqref{e-con3} between $r,x,y$, we have
512: \begin{equation}\label{e-rxyp}
513: rr'= \frac1{1-t^2}(yy'-t^2xx').
514: \end{equation}
515: This implies by \eqref{e-rata0}, \eqref{e-xyata0},
516: \begin{equation}\label{e-rxypata0}
517: (1+t\sqrt\gamma)r'(a_0)=y'(a_0)-t\sqrt\gamma x'(a_0),
518: \end{equation}
519: and hence
520: \begin{equation}
521: \Delta'(a_0)=0.
522: \end{equation}
523:
524: Now we compute $\Delta''(a_0)$.
525: We have
526: \begin{equation}\label{e-Deltapptemp3}
527: \begin{split}
528: \Delta''(a)=
529: &2\frac{r'}{r} +(1+a) \biggl( \frac{r''}{r}-\bigl(\frac{r'}{r}\bigr)^2\biggr)
530: -\frac12 \biggl( \frac{r''-x''}{r+t^{-1}-x}
531: -\bigl(\frac{r'-x'}{r+t^{-1}-x}\bigr)^2 \biggr) \\
532: &-2\gamma \frac{r'+x'}{r+t^{-1}+x}
533: -\frac12(1+2\gamma a)\biggl( \frac{r''+x''}{r+t^{-1}+x}
534: - \bigl( \frac{r'+x'}{r+t^{-1}+x} \bigr)^2 \biggr) \\
535: & -\frac12 \biggl( \frac{r''-y''}{r+t-y}
536: -\bigl( \frac{r'-y'}{r+t-y} \bigr)^2 \biggr)
537: -2 \frac{r'+y'}{r+t+y}
538: -\frac12(1+2a)\biggl( \frac{r''+y''}{r+t+y}
539: - \bigl( \frac{r'+y'}{r+t+y} \bigr)^2 \biggr).
540: \end{split}
541: \end{equation}
542: First consider the terms with double derivatives. At $a=a_0$,
543: \begin{equation}\label{e-Deltapptemp2}
544: \begin{split}
545: &\biggl( (1+a) \frac{r''}{r} -\frac12 \frac{r''-x''}{r+t^{-1}-x}
546: - \frac12(1+2\gamma a) \frac{r''+x''}{r+t^{-1}+x}
547: -\frac12 \frac{r''-y''}{r+t-y}
548: -\frac12(1+2a) \frac{r''+y''}{r+t+y} \biggr)(a_0) \\
549: &= -\frac{(\gamma-1)(1-t^2)}
550: {4(t+t\gamma+2\sqrt\gamma)(1+t\sqrt\gamma)}
551: \bigl[(1+t\sqrt\gamma)r''(a_0)+t\sqrt\gamma x''(a_0)-y''(a_0)\bigr].
552: \end{split}
553: \end{equation}
554: From \eqref{e-rxyp},
555: \begin{equation}
556: rr'' - \frac{1}{1-t^2} (yy''-t^2xx'')
557: = -(r')^2 +\frac1{1-t^2} \bigl( (y')^2-t^2(x')^2\bigr),
558: \end{equation}
559: and hence the right-hand side of \eqref{e-Deltapptemp2} is equal to
560: \begin{equation}\label{e-Deltapptemp4}
561: -\frac{(\gamma-1)(1-t^2)(t+\sqrt\gamma)}
562: {4(t+t\gamma+2\sqrt\gamma)(1+t\sqrt\gamma)}
563: \bigl[ -(r'(a_0))^2
564: +\frac1{1-t^2} \bigl( (y'(a_0))^2-t^2(x'(a_0))^2\bigr) \bigr].
565: \end{equation}
566: Thus $\Delta''(a_0)$ is given by \eqref{e-Deltapptemp3} at $a=a_0$
567: where the terms with
568: double derivatives are replaced by \eqref{e-Deltapptemp4},
569: which involves only the first derivatives of $r, x, y$ at $a_0$.
570: From \eqref{e-rpata0} and \eqref{e-con1}, \eqref{e-con2},
571: we have
572: \begin{eqnarray}
573: r'(a_0) &=& -\frac{3(\gamma-1)(t\gamma+t+2\sqrt\gamma)^2t^2}
574: {4(1+t\sqrt\gamma)(t+\sqrt\gamma)^3} \\
575: x'(a_0) &=& \frac{(t+4\sqrt\gamma +3t\gamma)(t+t\gamma+2\sqrt\gamma)^2t}
576: {4(1+t\sqrt\gamma)(t+\sqrt\gamma)^3} \\
577: y'(a_0) &=& \frac{(4t\sqrt\gamma +\gamma +3)(t+t\gamma+2\sqrt\gamma)^2t^2}
578: {4(1+t\sqrt\gamma)(t+\sqrt\gamma)^3},
579: \end{eqnarray}
580: and we obtain, after some calculation,
581: \begin{equation}
582: \Delta''(a_0) = - \frac{t^2(t+t\gamma+2\sqrt\gamma)^3\sqrt\gamma}
583: {2(1+t\sqrt\gamma)^2(t+\sqrt\gamma)^2}.
584: \end{equation}
585: By Taylor's formula, for $a\ge a_0$,
586: \begin{equation}
587: \Delta(a)= \frac12 \Delta''(a_0)(a-a_0)^2 +
588: \frac16 \Delta'''(\widetilde{a}) (a-a_0)^3,
589: \end{equation}
590: for some $a_0\le \widetilde{a}\le a$.
591: For $\delta>0$ and $1\le \gamma_2<\infty$, set
592: \begin{equation}
593: C:= \sup \{ |\Delta'''(\widetilde{a},\gamma)| :
594: 1\le \gamma\le \gamma_2,
595: a_0(\gamma) \le \widetilde{a}\le a_0(\gamma)(1+\delta) \},
596: \end{equation}
597: where we have made the dependence on $\gamma$ of $\Delta'''$
598: explicit.
599: It follows from the smooth dependence of $\Delta(a,\gamma)$
600: on $\gamma$, as well as on $a$, that given $\gamma_2$,
601: we can choose $\delta=\delta_2$ such that $C<\infty$.
602: Therefore
603: \begin{equation}
604: \Delta(a)= \frac12 \Delta''(a_0)(a-a_0)^2
605: \bigl(1+O(|a-a_0|\bigr),
606: \end{equation}
607: where $O(|a-a_0|)$ is uniform for $1\le \gamma\le \gamma_2$
608: and $a_0\le a\le a_0(1+\delta_2)$.
609: Indeed $O(|a-a_0|) \le C|a-a_0|$.
610: \end{proof}
611:
612: %Now from \eqref{e-Yest} and Lemma \ref{lem-Delta}, we have
613: %for $1\le\gamma\le \gamma_3$ with fixed $\gamma_3\ge 1$,
614: %and for $a_0+\frac{L_1}{k^{2/3}}\le a\le (1+\delta_3)a_0$
615: %with some $\delta_3>0$, as $k\to\infty$,
616: %\begin{equation}
617: % -Y_{21}(0) \le e^{-c_0 k(a-a_0)^2}
618: %\end{equation}
619: %for some $c_0>0$, where $c_0$ is independent of $\gamma_3$.
620: %This proves Proposition \ref{prop1}.
621: %From Lemma \ref{lem-GandY}, after a similar procedure
622: %as in \cite{BDJ}, \cite{BDR} Proposition 4.1, this estimate implies
623: %that
624: %for $1\le \gamma\le\gamma_0$ with any fixed $\gamma_0\ge 1$,
625: %there are $C, c>0$ such that
626: %\begin{equation}
627: % \Prob( G([\gamma N], N)\le n) \le Ce^{-c|x|^3}
628: %\end{equation}
629: %when $N$ is large enough.
630:
631: In order to prove Proposition \ref{prop1}, we use \eqref{e-Yest}.
632: As
633: \begin{equation}
634: \cos\theta_c= \frac{x(a)^2-r(a)^2-t^{-2}}{2r(a)t^{-1}},
635: \end{equation}
636: and $\cos\theta_c\to -1$ as $a\to a_0$,
637: we see that $\sin\frac{\theta_c}2=\sqrt{\frac{1-\cos\theta_c}2}$
638: is a smooth function of $a$ in $[a_0,\infty)$.
639: Thus $\sin\frac{\theta_c}2= 1+O(|a-a_0|)$ for $a$ near $a_0$,
640: $a\ge a_0$.
641: Inserting this information into \eqref{e-Yest} and using \eqref{e-Delest},
642: we obtain \eqref{e-Yexactest}.
643: