1: \section{Proof}\label{sec-walk}
2:
3:
4: Let $d_\lambda(N)$ be the number of semistandard Young tableaux
5: of shape $\lambda$ with fillings taken from $\{1.2.\cdots,N\}$,
6: and let $\ell(\lambda)$ be the number of rows of $\lambda$
7: (parts of $\lambda$, or the length of the first column).
8: From the bijection of path configurations and tableaux, the number of path
9: $\pi\in\pathcon(N,k)$ satisfying $L_1(N,k)(\pi)\le l$ is equal to
10: \begin{equation}\label{eq;walk3}
11: \sum_{\substack{\lambda\vdash k \\ \ell(\lambda)\le l}} d_\lambda(N).
12: \end{equation}
13: In our analysis (see also \cite{BR1, BR2}),
14: it turns out that in addition to
15: the number of rows, the number of odd columns plays an important role
16: in describing a tableau.
17: For a partition $\lambda$, we define
18: $\lambda'$ to be the transpose of $\lambda$,
19: $f(\lambda)$ to be the number of odd row in $\lambda$,
20: and $|\lambda|$ to be the size of $\lambda$.
21: Let $b(N,j,m,l)$ be the number of semistandard Young tableaux
22: of size $2j+m$ with $m$ odd columns with at most $l$ columns
23: with fillings taken from $\{1,2,\cdots,N\}$ :
24: \begin{equation}\label{eq;walk4}
25: b(N,j,m,l)= \sum_{\substack{\lambda\vdash 2j+m \\ f(\lambda')=m \\
26: \ell(\lambda)\le l}} d_\lambda(N).
27: \end{equation}
28: We use the notation $b(N,j,m,\infty)$ for the sum above without
29: restriction on $\ell(\lambda)$.
30: Now we define a generating function
31: \begin{equation}\label{eq-walk2}
32: \begin{split}
33: \phi(N,l,t,\beta) &:=
34: (1-t^2)^{N(N-1)/2}(1-\beta t)^N
35: %\sum_{j,m=0}^\infty t^{2j+m}\beta^m
36: %\sum_{\substack{\lambda\vdash 2j+m \\ f(\lambda')=m \\ \ell(\lambda)\le l}}
37: \sum_{\ell(\lambda)\le l} t^{|\lambda|}\beta^{f(\lambda')}
38: d_\lambda(N)\\
39: &=(1-t^2)^{N(N-1)/2}(1-\beta t)^N
40: \sum_{j,m\ge 0} t^{2j}(\beta t)^m b(N,j,m,l),
41: \end{split}
42: \end{equation}
43: where the sum in the first expression is taken
44: over all the partitions $\lambda$ satisfying
45: $\ell(\lambda)\le l$.
46:
47: The starting point of our analysis is the following result of \cite{BR1}.
48: \begin{lem}\label{lem0}
49: Let $\phi(N,l,t,\beta)$ be defined as in \eqref{eq-walk2}.
50: We have
51: \begin{equation}\label{eq-walk3}
52: \phi(N,2l+1,t,\beta)
53: = (1-t^2)^{N(N-1)/2} \det(H_l),
54: %= (1-t^2)^{N(N-1)/2}
55: %\Exp_{U\in Sp(2l)} \det(1-tU)^{-N}.
56: \end{equation}
57: where $H_l=(h_{jk})_{0\le j,k<l}$ is the $l\times l$ Hankel determinant
58: with
59: \begin{equation}\label{eq.walk7}
60: h_{jk}=h_{jk}(N,t)=\frac{2^{j+k+1}}{\pi} \int_{-1}^1
61: x^{j+k} (1+t^2-2tx)^{-N}(1-x^2)^{1/2}dx.
62: \end{equation}
63: \end{lem}
64:
65: \begin{rem}
66: Note that the right hand side of \eqref{eq-walk3} does not depend on $\beta$.
67: \end{rem}
68: \begin{proof}
69: This proof is in \cite{BR1} in a slightly different form.
70: Let $s_\lambda(x)$, $x=(x_1,x_2,\cdots)$, be the Schur function,
71: and define $H(u;y)$ with $y=(y_1,y_2,\cdots)$ by
72: \begin{equation}\label{eq;lem01}
73: H(u;y)=\prod_j (1-uy_j)^{-1}.
74: \end{equation}
75: In (5.65) of \cite{BR1}, it is proved that
76: \begin{equation}\label{eq;lem02}
77: \sum_{\ell(\lambda)\le 2l+1} \beta^{f(\lambda')} s_\lambda(x)
78: = H(\beta;x) \Exp_{U\in Sp(2l)} \det(H(U;x)),
79: \end{equation}
80: which is an identity as a formal power series in $x$.
81: But the combinatorial definition of the Schur function is
82: (see, e.g. Chapter 7.10 of \cite{Stl})
83: \begin{equation}
84: s_\lambda(x) =\sum_T x_1^{\alpha_1(T)} x_2^{\alpha_2(T)}\cdots,
85: \end{equation}
86: where the sum is over all semistandard Young tableaux $T$ of shape $\lambda$,
87: and $\alpha_j(T)$ is the number of parts of $T$ equal to $j$
88: (type of $T$).
89: Since $\sum_j \alpha_j(T)=|\lambda|$,
90: if we take the special case $x=(t,t,\cdots,t,0,0,\cdots)$
91: where the first $N$ elements are $t$ and the rest are $0$,
92: then $s_\lambda(x)$ becomes
93: \begin{equation}
94: t^{|\lambda|}d_\lambda(N).
95: \end{equation}
96: Hence for this special choice of $x$ \eqref{eq;lem02} is now
97: \begin{equation}\label{eq;lem05}
98: \sum_{\ell(\lambda)\le 2l+1} \beta^{f(\lambda')} t^{|\lambda|}
99: d_\lambda(N)
100: = (1-\beta t)^{-N} \Exp_{U\in Sp(2l)} \det((1-tU)^{-N}).
101: \end{equation}
102:
103: Now using Weyl's integration formula for $Sp(2l)$ (see, e.g. \cite{Simon}),
104: the expectation in \eqref{eq;lem05} becomes
105: \begin{equation}\label{eq;lem06}
106: \Exp_{U\in Sp(2l)} \det((1-tU)^{-N})
107: = \frac{2^{l^2}}{l!(2\pi)^l}
108: \int_{[0,2\pi]^l} \prod_{1\le j<k\le l} (\cos\theta_j-\cos\theta_k)^2
109: \prod_{j=1}^{l} \sin^2\theta_j
110: (1+t^2-2t\cos\theta_j)^{-N}d\theta_j.
111: %\det (\iota_{j-k}-\iota_{j+k+2})_{0\le j,k< l},
112: \end{equation}
113: Standard Vandermonde determinant manipulations yield
114: \begin{equation}
115: \frac{2^{l^2}}{(2\pi)^l} \det\biggl(
116: \int_0^{2\pi} \cos^{j+k}\theta \sin^2\theta(1+t^2-2t\cos\theta)^{-N}d\theta
117: \biggr)_{0\le j,k<l},
118: \end{equation}
119: which again after change of variables $x=\cos\theta$,
120: is equal to
121: %where
122: %\begin{equation}\label{eq;lem07}
123: % \iota_j=\frac1{2\pi}\int_0^{2\pi} e^{ij\theta}
124: %(1-te^{i\theta})^{-N} (1-te^{-i\theta})^{-N} d\theta.
125: %\end{equation}
126: %Note that $\iota_j$ is real, so we can replace $e^{ij\theta}$
127: %in the integrand by $(e^{ij\theta}+e^{-ij\theta})/2$.
128: %Hence
129: %the entry in the determinant of \eqref{eq;lem06} is
130: %\begin{equation}
131: % \iota_{j-k}-\iota_{j+k+2}
132: %= \frac2{\pi} \int_0^{\pi} \sin((j+1)\theta)\sin((k+1)\theta)
133: %(1+t^2-2t\cos\theta)^{-N} \theta.
134: %\end{equation}
135: %Now by change of variable $\cos\theta\mapsto x$, it is equal to
136: \begin{equation}
137: \frac{2^{l^2}}{\pi^l}\det\biggl(
138: \int_{-1}^1 U_j(x)U_k(x)(1+t^2-2tx)^{-N}(1-x^2)^{1/2}dx
139: \biggr)_{0\le j,k<l}
140: \end{equation}
141: where $U_j(x)=\frac{\sin((j+1)\theta)}{\sin\theta}$, $x=\cos\theta$,
142: is the Chebyshev polynomial of the second kind.
143: Note that $U_j(x)=2^jx^j+\cdots$.
144: Hence elementary row and column operations yield Lemma \ref{lem0}.
145: \end{proof}
146:
147: Using this expression, we first obtain the asymptotic result
148: for the generating function.
149: The limit is insensitive to $\beta$ since so is $\phi$.
150: \begin{prop}\label{prop1}
151: Let $0<t<1$ and $\beta>0$ be fixed satisfying $0<\beta t<1$.
152: For each $l$ and $N$, define $x\in\R$ by
153: \begin{equation}\label{eq.walk16}
154: x=\frac{l-\eta(t)N}{\rho(t)N^{1/3}},
155: \end{equation}
156: where $\eta(t)$ and $\rho(t)$ are defined in \eqref{eq;walk1}
157: Then there exits a positive constant $M_0$ such that for $M>M_0$,
158: there are constants $C, c>0$, independent of $M$, and $C(M)$
159: which may depend on $M$ so that
160: \begin{equation}\label{eq;walk7}
161: |\phi(N,l,t,\beta)- F_1(x)| \le
162: \frac{C(M)}{l^{1/3}}+ Ce^{-cM^{3/2}},
163: \qquad -M<x<M.
164: \end{equation}
165: Also we have
166: \begin{eqnarray}
167: \label{walk16}
168: 0\le1-\phi(N,l,t,\beta)\le Ce^{-cx^{3/2}}, \qquad &x\ge M_0,\\
169: \label{walk17}
170: 0\le \phi(N,l,t,\beta)\le Ce^{-c|x|^3}, \qquad &x\le -M_0.
171: \end{eqnarray}
172: \end{prop}
173:
174: \begin{proof}
175: It is enough to consider the limit for $\phi(N,2l+1,t,\beta)$
176: since from the definition \eqref{eq-walk2}, $\phi$ is monotone
177: increasing in $l$.
178: First we related the determinant in \eqref{eq-walk3} with certain quantities
179: of orthogonal polynomials on the circle.
180:
181: Let $p_j(x)=x^j+\cdots$ be the $j^{\text{th}}$ monic orthogonal
182: polynomial with respect to the weight $w(x)dx=(1+t^2-2tx)^{-N}(1-x^2)^{1/2}dx$
183: on the interval $(-1,1)$, and let $C_j$ be the norm of $p_j$ :
184: \begin{equation}
185: \int_{-1}^1 p_j(x)p_k(x) w(x)dx= C_j\delta_{jk}.
186: \end{equation}
187: It is a well known result of orthogonal polynomial theory
188: (see, e.g. \cite{Szego})
189: that $C_j=\det(\tilde{H}_{j+1})/\det(\tilde{H}_j)$,
190: where $\tilde{H}_l=(\tilde{h}_{jk})_{0\le j,k<l}$
191: with
192: \begin{equation}
193: \tilde{h}_{jk}=\int_{-1}^1 x^{j+k}(1+t^2-2tx)^{-N}
194: (1-x^2)^{1/2}dx.
195: %= \frac{\pi}{2^{j+k+1}}h_{jk}.
196: \end{equation}
197: which is equal to $\frac{\pi}{2^{j+k+1}}h_{jk}$ (recall \eqref{eq.walk7}).
198: Hence $\det(\tilde{H}_j)=\frac{\pi^j}{2^{j^2}}\det(H_j)$.
199: Since the Szeg\"o strong limit theorem for Hankel determinants
200: (see, e.g. \cite{Jo1}) implies that
201: $\lim_{l\to\infty}\det(H_l)= (1-t^2)^{-N(N-1)/2}$ for fixed $N$,
202: we have
203: \begin{equation}\label{eq.walk23}
204: \phi(N,2l+1,t,\beta)=\lim_{k\to\infty}
205: \prod_{j=l}^{k}\frac{\det(H_j)}{\det(H_{j+1})}
206: =\prod_{j=l}^{\infty} \frac{\pi}{2^{2j+1}}C_j^{-1}.
207: \end{equation}
208:
209: Now we use the relation between orthogonal polynomials on the unit circle
210: and those on the interval $(-1,1)$.
211: Let $\pi_j(z)=z^j+\cdots$ be the $j^{\text{th}}$ monic orthogonal polynomials
212: on the unit circle $\{|z|=1\}$ with respect to the weight
213: \begin{equation}
214: \varphi(z)\frac{dz}{2\pi i}=(1-tz)^{-N}(1-tz^{-1})^{-N}\frac{dz}{2\pi iz},
215: \end{equation}
216: and let
217: $N_j$ be the norm of $\pi_j(z)$ :
218: \begin{equation}
219: \int_{|z|=1} \pi_j(z)
220: \overline{\pi_k(z)}\varphi(z) \frac{dz}{2\pi iz}
221: = N_j \delta_{jk}.
222: \end{equation}
223: There is a simple relation between orthogonal polynomials $p_j$ on
224: the unit circle and orthogonal polynomials $\pi_j$ on the interval
225: (see the forth equation of (11.5.2) in \cite{Szego}) :
226: \begin{equation}
227: C_j^{-1/2}p_j(x) = \sqrt{\frac{2}{\pi}} \bigl(1+\pi_{2j+2}(0)\bigr)^{-1/2}
228: N_{2j+1}^{-1/2}
229: \frac{z^{-j}\pi_{2j+1}(z)-z^{j}\pi_{2j+1}(z^{-1})}{z-z^{-1}},
230: \quad x=\frac12(z+z^{-1}).
231: \end{equation}
232: Especially comparing the coefficient of the leading term $x^j$,
233: we have the relation
234: \begin{equation}\label{eq.walk26}
235: C_j=\frac{\pi}{2^{2j+1}} \bigl(1+\pi_{2j+2}(0)\bigr)N_{2j+1}.
236: \end{equation}
237: But we also have
238: $(1-\pi_{n+1}(0)^2)N_n=N_{n+1}$ (see (11.3.6) in \cite{Szego}).
239: Hence \eqref{eq.walk26} is equal to
240: \begin{equation}
241: C_j=\frac{\pi}{2^{2j+1}} \bigl(1-\pi_{2j+2}(0)\bigr)^{-1}N_{2j+2}.
242: \end{equation}
243: Therefore \eqref{eq.walk23} becomes
244: \begin{equation}
245: \phi(N,2l+1,t,\beta) =
246: \prod_{j=l}^\infty \bigl(1-\pi_{2j+2}(0)\bigr)N_{2j+2}^{-1}.
247: \end{equation}
248: This argument appeared in Corollary 2.7 of \cite{BR1}.
249: Now using Proposition \ref{prop-rhp}, computations similar to
250: Lemma 7.1 of \cite{BDJ} (also Corollary 7.2 and
251: Corollary 7.6 of \cite{BR2})
252: yield the result.
253: \end{proof}
254:
255: Interpreting the notation $b(N,j,m,\infty)$ as the sum without
256: constraints on $\ell(\lambda)$ in \eqref{eq;walk4},
257: the number of path configuration in $\pathcon(N,k)$ is equal to
258: \begin{equation}\label{eq;walk8}
259: |\pathcon(N,k)| = \sum_{2j+m=k} b(N,j,m,\infty),
260: \end{equation}
261: and the probability of interest that $L_1(N,k)\le l$ for
262: $\pi\in\pathcon(N,k)$ is equal to
263: \begin{equation}\label{eq;walk9}
264: \frac1{|\pathcon(N,k)|}
265: \sum_{2j+m=k} b(N,j,m,l)
266: = \frac1{|\pathcon(N,k)|}
267: \sum_{2j+m=k} p(N,j,m,l)b(N,j,m,\infty),
268: \end{equation}
269: where
270: \begin{equation}\label{eq.walk30}
271: p(N,j,m,l)=\frac{b(N,j,m,l)}{b(N,j,m,\infty)}.
272: \end{equation}
273:
274: For fixed $N$, as $l\to\infty$,
275: the Szeg\"o strong limit theorem for Hankel determinants
276: (see, e.g. \cite{Jo1}) implies that
277: \eqref{eq-walk3} becomes $1$.
278: Thus we have the identity
279: \begin{equation}\label{eq;walk10}
280: \sum_{j,m\ge 0} t^{2j}(\beta t)^{m} b(N,j,m,\infty)
281: = (1-t^2)^{-N(N-1)/2}(1-\beta t)^{-N}.
282: \end{equation}
283: By taking Taylor expansion of the right hand side in $t$ and $\beta$,
284: we obtain
285: \begin{equation}\label{eq-walk8}
286: %\sum_{\substack{\lambda\vdash 2k+m \\ f(\lambda')=m}} d_\lambda(N)
287: b(N,j,m,\infty)
288: = \binom{\frac{N(N-1)}2+j-1}{j}\binom{N+m-1}{m}.
289: \end{equation}
290: There is a more direct way to see this.
291: See the remark after Lemma \ref{lem3} below.
292: Now from \eqref{eq;walk8}, the total number of paths
293: in $\pathcon(N,k)$ is equal to
294: \begin{equation}\label{eq;walk12}
295: |\pathcon(N,k)|=
296: \sum_{2j+m=k} \binom{\frac{N(N-1)}2+j-1}{j}\binom{N+m-1}{m}.
297: \end{equation}
298: Now it is straightforward to obtain the following result on the number
299: of all paths.
300:
301: \begin{lem}\label{lem1}
302: Let $0<t<1$
303: and let
304: \begin{equation}\label{eq;walk13}
305: k= \bigl[ \frac{t^2}{1-t^2}N^2+o(N^{4/3})\bigr].
306: \end{equation}
307: As $N\to\infty$, we have
308: \begin{equation}\label{eq.walk33}
309: |\pathcon(N,k)| =
310: \frac{\exp\bigl\{-\frac{N^2t^2}{1-t^2}\log t
311: -\mu N\log t- \frac14\bigl(\frac{\mu(1-t^2)}{t}-1\bigr)^2\bigr\}}
312: {\sqrt\pi tN(1-t)^{N}(1-t^2)^{N(N-1)/2-1}}
313: \bigl(1+o(1)\bigr)
314: \end{equation}
315: where the term $o(1)$ vanishes as $N\to\infty$,
316: and $\mu$ is defined by
317: \begin{equation}\label{eq.walk34}
318: \mu:= \frac{k}{N}-\frac{t^2}{1-t^2}N
319: \end{equation}
320: which is of order $o(N^{1/3})$ from \eqref{eq;walk13}.
321: Moreover, the main contribution to the sum comes from
322: $|m - \frac{t}{1-t}N| \le N^{1/2+\epsilon/2}$ for some $0<\epsilon<\frac13$ ;
323: precisely, there is a constant $c>0$ such that
324: for any $0<\epsilon<\frac13$, we have
325: \begin{equation}\label{eq.walk37}
326: |\pathcon(N,k)|=
327: \biggl[ \sum_{\substack{|m-\frac{t}{1-t}N|\le N^{1/2+\epsilon/2} \\
328: k-m \text{ is even}}} b(N,\frac{k-m}{2},m,\infty) \biggr]
329: \bigl(1+O(e^{-cN^{\epsilon}})\bigr).
330: \end{equation}
331: \end{lem}
332:
333: \begin{proof}
334: From \eqref{eq;walk12}, we have
335: \begin{equation}
336: |\pathcon(N,k)|=\sum_{m=0}^{[\frac{k}2]} a(m),
337: \qquad a(m)= \binom{\frac{N(N-1)}{2}+\frac{k-m}2-1}{\frac{k-m}2}
338: \binom{N+m-1}{m}.
339: \end{equation}
340: The ratio of $a(m)$ is
341: \begin{equation}\label{eq.walk38}
342: \frac{a(m+2)}{a(m)}=\frac{(N+m+1)(N+m)(k-m)}{(m+2)(m+1)(N(N-1)+k-m-2)}.
343: \end{equation}
344: One can directly check that under the condition \eqref{eq;walk13},
345: the above ratio is decreasing in $m$, and becomes closest to $1$ at
346: \begin{equation}
347: m_c= \bigl[\frac{t}{1-t}N + o(N^{1/3})\bigr].
348: \end{equation}
349: Hence $a(m)$ is unimodal: it is increasing for $m<m_c$
350: and is decreasing for $m>m_c$.
351: Now consider the neighborhood $\mathcal N$
352: of $m_c$ of size $N^{1/2+\epsilon/2}$
353: for some fixed $0<\epsilon<\frac13$.
354: For $m$ in $\mathcal N$, set
355: \begin{equation}\label{eq.walk39}
356: m= \frac{t}{1-t}N+x, \qquad |x|\le N^{1/2+\epsilon/2}.
357: \end{equation}
358: For any $M,x>0$, Stirling's formula yields
359: \begin{equation}\label{2.41}
360: (M+x)!=\sqrt{2\pi M}M^{M+x}e^{-M+\frac{x^2}{2M}}\
361: \biggl(1+O\bigl(\frac{x}{M}\bigr)+O\bigl(\frac{x^3}{M^2}\bigl)\biggr).
362: \end{equation}
363: Using \eqref{eq;walk13}, \eqref{eq.walk39} and \eqref{2.41},
364: we have for $m$ in $\mathcal N$,
365: \begin{equation}
366: \binom{N+m-1}{m}=\frac{(\frac{1}{1-t}N+x-1)!}{(N-1)!(\frac{t}{1-t}N+x)!}
367: =\frac{\exp\bigl\{-(\frac{tN}{1-t}+x)\log t-\frac{(1-t)^2x^2}{2tN}\bigr\}}
368: {\sqrt{2\pi tN}(1-t)^{N-1}}
369: \biggl(1+O(N^{-\frac12+\frac32\epsilon})\biggr)
370: \end{equation}
371: and
372: \begin{equation}
373: \begin{split}
374: &\binom{\frac{N(N-1)}{2}+\frac{k-m}2-1}{\frac{k-m}2}
375: =\frac{\bigl(\frac{N(N-1)}{2(1-t^2)}+(\mu-\frac{t}{1-t^2})\frac{N}2
376: -\frac{x}2-1\bigr)!}
377: {\bigl(\frac{N(N-1)}{2}-1\bigr)!
378: \bigl(\frac{t^2N(N-1)}{2(1-t^2)}+(\mu-\frac{t}{1-t^2})\frac{N}2
379: -\frac{x}2\bigr)!}\\
380: &\qquad\qquad
381: =\frac{\exp\bigl\{ -(\frac{t^2}{1-t^2}N^2+(\mu-\frac{t}{1-t})N-x)\log t
382: -\frac{1}{4}(\frac{(1-t^2)\mu}{t}-1)^2 \bigr\}}
383: {\sqrt{\pi}tN(1-t^2)^{N(N-1)/2-1}} \bigl(1+o(1)\bigr).
384: \end{split}
385: \end{equation}
386: Thus we have
387: \begin{equation}\label{eq.walk40}
388: a(m) =
389: \frac{\exp\bigl\{-\frac{N^2t^2}{1-t^2}\log t
390: -N\mu\log t- \frac14\bigl(\frac{\mu(1-t^2)}{t}-1\bigr)^2
391: -\frac{(1-t)^2x^2}{2tN}\bigr\}}
392: {\sqrt{2}\pi t^{3/2}N^{3/2}(1-t)^{N}(1-t^2)^{N(N-1)/2-1}}
393: \bigl(1+o(1)\bigr).
394: \end{equation}
395: Let
396: \begin{equation}
397: |\pathcon(N,k)| =\sum_{*} a(m)+\sum_{**} a(m),
398: \end{equation}
399: where $*$ denotes the set $\mathcal N$ of $m$ satisfying \eqref{eq.walk39}
400: and $**$ denotes the rest of the range of $m$.
401: From \eqref{eq.walk40}, the first sum over $*$ is equal
402: to the right hand side of \eqref{eq.walk33}.
403: Also from the unimodality, $a(m)$ in $**$ is less than or equal to
404: the largest of $a(m_+)$ and $a(m_-)$ where
405: $m_\pm=[\frac{t}{1-t}N \pm N^{1/2+\epsilon}]$.
406: The number of summand in $**$ is of order $N^2$.
407: Hence using \eqref{eq.walk40} again for $m_\pm$,
408: if we take $c=\frac{(1-t)^2}{4t}$, for large $N$,
409: we obtain
410: \begin{equation}
411: \sum_{**} a(m)= \bigl(\sum_{*} a(m)\bigr) e^{-cN^{\epsilon}},
412: \end{equation}
413: which establishes the proof.
414: \end{proof}
415:
416:
417: Now we rewrite \eqref{eq-walk2} as
418: \begin{equation}
419: \phi(N,l,t,\beta)
420: = (1-t^2)^{N(N-1)/2}(1-\beta t)^N
421: \sum_{j,m\ge 0} t^{2j}(\beta t)^m b(N,j,m,\infty) p(N,j,m,l).
422: \end{equation}
423: The asymptotics of $\phi(N,l,t,\beta)$ and $p(N,j,m,l)$
424: are related as follows.
425: \begin{lem}\label{lem-dep}
426: For any $d>0$, there are constants $C_0, c_0>0$
427: such that
428: for all $l\ge 0$,
429: \begin{equation}\label{eq.walk46}
430: p(N,\mu_+(N),\nu_+(N),l)-\frac{C_0}{N^d}\le
431: \phi(N,l,t,\beta)\le p(N,\mu_-(N),\nu_-(N),l)+\frac{C_0}{N^d}
432: \end{equation}
433: where
434: \begin{eqnarray}
435: \label{eq.walk47}
436: \mu_\pm(N)&=&\biggl[\frac{t^2}{2(1-t^2)}N^2 \pm c_0N\sqrt{\log N}\biggr],\\
437: \label{eq.walk48}
438: \nu_\pm(N)&=&\biggl[\frac{\beta t}{1-\beta t} N \pm c_0\sqrt{N\log N}\biggr].
439: \end{eqnarray}
440: \end{lem}
441: The proof follows by using the following
442: Lemma twice for $j$ and $m$ indices together with
443: the Lemma \ref{lem3}. (Recall the \eqref{eq-walk8}).
444:
445: \begin{lem}\label{lem2}
446: For a sequence $\{q_j\}_{j\ge 0}$, we define the following
447: generating function
448: \begin{equation}\label{eq46}
449: G(N)=(1-a)^N\sum_{j=0}^\infty a^j\binom{N+j-1}{j} q_j,
450: \qquad 0<a<1,\ \ N=1,2,\cdots.
451: \end{equation}
452: For each $d>0$, there are constants $C_1,c_1\ge 0$ such that
453: for any sequence $\{q_j\}_{j\ge 0}$ satisfying
454: (i) $q_j\ge q_{j+1}$ and (ii) $0\le q_j\le 1$,
455: \begin{equation}
456: q_{N^{**}}-\frac{C_1}{N^d}\le G(N)\le q_{N^*}+\frac{C_1}{N^d},
457: \qquad N\ge 1,
458: \end{equation}
459: where
460: \begin{eqnarray}
461: N^*&=&\frac{a}{1-a}N-c_1\sqrt{N\log{N}},\\
462: N^{**}&=&\frac{a}{1-a}N+c_1\sqrt{N\log{N}}.
463: \end{eqnarray}
464: \end{lem}
465:
466: \begin{proof}
467: This proof is parallel to that of the de-Poissonization
468: lemma in \cite{Jo2}.
469: We have
470: \begin{equation}
471: G(N)=\sum_{j=0}^\infty f_jq_j, \qquad f_j=(1-a)^Na^j\binom{N+j-1}{j}.
472: \end{equation}
473: Stirling's formula yields for $n,m\ge 1$,
474: \begin{equation}
475: \binom{m+n-1}{m}\le C\exp\biggl\{ m\bigl[(1+\frac{n}{m}\log(1+\frac{n}{m})
476: -\frac{n}{m}\log\frac{m}{n}\bigr] \biggr\},
477: \end{equation}
478: with some constant $C$.
479: Thus we have
480: \begin{equation}
481: f_j\le C\exp\{Nh(j/N)\},
482: \qquad h(x)=(1+x)\log(1+x)-x\log x +x\log a+\log(1-a).
483: \end{equation}
484: One can directly check the following estimates of $h$ :
485: \begin{eqnarray}
486: \label{eq53}
487: h(x) &\le& -\frac{(1-a)^2}{2a}\bigl(x-\frac{a}{1-a}\bigr)^2,
488: \qquad \qquad 0\le x\le \frac{2a}{1-a},\\
489: \label{eq54}
490: h(x) &\le& -\bigl[ \log 2-\frac{1+a}{2a}\log(1+a)\bigr] x,
491: \qquad x\ge \frac{2a}{1-a}.
492: \end{eqnarray}
493: We take a constant $c_1>0$ satisfying $\frac{(1-a)^2}{2a}c_1^2-1\ge d$.
494: From \eqref{eq53} and the condition (ii), we have
495: \begin{equation}
496: \sum_{j\le N^*} f_jq_j \le \frac{a}{1-a}CN
497: e^{-\frac{(1-a)^2}{2a}c^2\log N}
498: \le \frac{C}{N^d},
499: \end{equation}
500: with a new constant $C$.
501: Similarly,
502: \begin{equation}
503: \sum_{N^{**}\le j\le \frac{2a}{1-a}N} f_jq_j \le \frac{C}{N^d}.
504: \end{equation}
505: Also using \eqref{eq54}, we have
506: \begin{equation}
507: \sum_{j\ge \frac{2a}{1-a}N} f_jq_j\le C'e^{-c'N}
508: \end{equation}
509: for some constants $c',C'$.
510: Thus we have
511: \begin{equation}\label{eq58}
512: \biggl| G(N)-\sum_{N^*\le j\le N^{**}}f_jq_j \biggr| \le \frac{C}{N^d},
513: \end{equation}
514: with a possibly different constant $C$.
515: Now from the monotonicity condition (i), we have
516: \begin{equation}
517: \sum_{N^*\le j\le N^{**}} f_jq_j
518: \le \biggl( \sum_{N^*\le j\le N^{**}} f_j \biggr) q_{N^*}
519: \le q_{N^*},
520: \end{equation}
521: and
522: \begin{equation}
523: \sum_{N^*\le j\le N^{**}} f_jq_j
524: \ge \biggl( \sum_{N^*\le j\le N^{**}} f_j\biggr) q_{N^{**}}
525: \ge \bigl(1-\frac{C}{N^d}\bigr) q_{N^{**}}
526: \ge q_{N^{**}} -\frac{C}{N^d},
527: \end{equation}
528: using the equality \eqref{eq58} for the second equality.
529: Thus we obtained the desired result.
530: \end{proof}
531:
532: To use the above Lemma to $\phi$, we need monotonicity in $l$.
533: It is more convenient now to view semistandard Young tableaux (SSYT)
534: as generalized permutations.
535: A two-rowed array
536: \begin{equation}
537: \pi=\begin{pmatrix}
538: i_1&\cdots&i_k\\ j_1&\cdots&j_k
539: \end{pmatrix}
540: \end{equation}
541: is called a generalized permutation if either $i_r<i_{r+1}$
542: or $i_r=i_{r+1}$, $j_r\le j_{r+1}$.
543: Suppose the elements in the upper row of $\pi$ come from
544: $\{1,2,\cdots,M\}$ and the elements in the bottom row
545: come from $\{1,2,\cdots,N\}$.
546: One can represent a generalized permutation as a $M\times N$ matrix
547: $(a_{ik})$ where $a_{ik}$ is the number of times when
548: $\binom{i}{k}$ occurs in $\pi$.
549: For example, the generalized permutation
550: \begin{equation}\label{eq;walk68}
551: \begin{pmatrix}
552: 1&1&1&2&2&2&2&3&3\\ 1&3&3&2&2&2&4&3&4
553: \end{pmatrix}
554: \end{equation}
555: corresponds to
556: \begin{equation}
557: \begin{pmatrix}
558: 1&0&2&0\\ 0&3&0&1\\ 0&0&1&1
559: \end{pmatrix}.
560: \end{equation}
561: In the proof of the Lemma below, we regard a generalized permutation
562: as a $M\times N$ square board with stacks of $a_{ik}$
563: balls in each position $(i,k)$.
564: We denote by $L(\pi)$ the length of the longest strictly decreasing
565: subsequence of $\pi$.
566: In the example \eqref{eq;walk68}, $L(\pi)=2$.
567:
568:
569: Let $M_{j,m}$ be the set of $N\times N$ matrices $\pi=(a_{ik})$
570: which is symmetric $a_{ik}=a_{ki}$, and satisfies
571: $\sum_{i=1}^N a_{ii}= m$ and $\sum_{1\le i<k\le N} a_{ik}= j$.
572: This is a certain subset of the set of generalized permutations.
573: The celebrated Robinson-Schensted-Knuth correspondence
574: \cite{Knuth:correspondence}
575: establishes a bijection between $M_{j,m}$
576: and the set of SSYT of size $2j+m$
577: with $m$ odd columns with fillings taken from
578: $\{1,2,\cdots, N\}$.
579: Moreover, under this bijection,
580: $L(\pi)$ for $\pi\in M_{j,m}$ (viewed as a generalized permutation)
581: is equal to the number of rows of the corresponding SSYT.
582:
583: With this preliminary, we can prove the following.
584:
585: \begin{lem}[monotonicity]\label{lem3}
586: For any $j,m\ge 0$, we have
587: \begin{equation}
588: p(N,j+1,m,l)\le p(N,j,m,l), \qquad p(N,j,m+1,l)\le p(N,j,m,l).
589: \end{equation}
590: \end{lem}
591:
592: \begin{proof}
593: We first consider the second inequality.
594: From \eqref{eq.walk30}, we need to show that
595: \begin{equation}
596: (m+1)b(N,j,m+1,l)\le (N+m)b(N,j,m,l).
597: \end{equation}
598: By the definition \eqref{eq;walk4} and
599: the Robinson-Schensted-Knuth correspondence,
600: $b(N,j,m,l)$ is equal to the number of $\pi\in M_{j,m}$
601: satisfying $L(\pi)\le l$.
602:
603: Consider all possible distinct (strict) upper triangular parts
604: of elements in $M_{j,m}$.
605: It is equal to putting $j$ identical balls into $N(N-1)/2$ boxes ;
606: $K=\binom{N(N-1)/2+j-1}{j}$ distinct ways.
607: Hence we have a disjoint union $M_{j,m}=\cup_{i=1}^{K} S_{i,m}$
608: where each $S_{i,m}$ consists of $\pi\in M_{j,m}$ with same upper
609: triangular part, and elements in $S_{i,m}$ and $S_{i',m}$
610: have different upper triangular parts when $i\neq i'$.
611: Similarly, $M_{j,m+1}=\cup_{i=1}^{K} S'_{i,m+1}$
612: where $\sigma\in S'_{i,m+1}$ has the upper triangular part same as
613: that of $\pi\in S_{i,m}$.
614:
615: Now for each $\pi=(a_{rs})\in S_{i,m}$, we generate
616: $N+m$ elements in $S'_{i,m+1}$ as follows.
617: For $1\le r\le N$, assign $a_{rr}+1$ identical $\pi'=(a'_{kl})$
618: such that $a'_{rr}=a_{rr}+1$ and $a'_{kl}=a_{kl}$
619: for $(k,l)\neq (r,r)$.
620: (One can think this as adding a new ball in an array of $a_{rr}$
621: balls ; there are $a_{rr}+1$ ways.)
622: Since $\sum_{1\le r\le N} a_{rr}+1= m+N$,
623: there result $(m+N)|S_{i,m}|$ (many identical) elements of $S_{i,m+1}$.
624: Note that under this assignment,
625: \begin{equation}\label{eq;walk70}
626: L(\pi')\ge L(\pi).
627: \end{equation}
628: Now fix $\sigma=(b_{kl})\in S_{i,m+1}$.
629: Since there are $m+1$ diagonal entries, there are exactly
630: $m+1$ (many identical) elements of $S_{i,m}$ from which
631: $\sigma$ is generated under the above assignment.
632: (Considering each entry as a ball, each $m+1$ balls on the diagonal
633: can be a newly added one.)
634: Thus we have the identity $(m+N)|S_{i,m}|=(m+1)|S_{i,m+1}|$.
635: Furthermore, from the remark regarding \eqref{eq;walk70},
636: we have $(m+1)|R(i,m+1,l)|\le (m+N)|R_{i,m,l}|$,
637: where $R_{i,m,l}$ is the subset of $\pi\in S_{i,m}$ satisfying
638: $L(\pi)\le l$.
639: Therefore the second inequality in the Lemma is obtained.
640:
641: The first inequality follows from a similar argument.
642: \end{proof}
643:
644: \begin{rem}
645: As mentioned before, using the generalized permutation interpretation
646: of SSYT, we can see \eqref{eq-walk8} directly.
647: From non-negative integer matrix representation of generalized permutations,
648: $b(N,j,m,\infty)$ is the number of $N\times N$ matrices $(a_{rs})$
649: with non-negative integer entries such that
650: $\sum_{r=1}^N a_{rr}=m$ and
651: $\sum_{1\le r<s\le N}a_{rs}=j$.
652: It is equivalent to placing $m$ identical balls of color 1 into $N$ boxes
653: and $j$ identical balls of color 2 into $N(N-1)/2$ boxes.
654: Therefore we obtain \eqref{eq-walk8}.
655: \end{rem}
656:
657: Now we give the proof of the theorem.
658: \begin{proof}[proof of Theorem \ref{mainthm}]
659: In \eqref{eq;walk9}, we have
660: \begin{equation}
661: %\begin{split}
662: \Prob(L_1(N,k)\le l) = \frac{1}{|P(N,k)|}
663: \sum_{2j+m=k} p(N,j,m,l)b(N,j,m,\infty).
664: %&= \frac{1}{|P(N,k)|}\sum_{m=0}^{[\frac{k}2]}
665: %p(N,j,m,l) a(m)
666: %, \qquad
667: %a(m)= \binom{\frac{N(N-1)}{2}+\frac{k-m}2-1}{\frac{k-m}2}
668: %\binom{N+m-1}{m}.
669: %\end{split}
670: \end{equation}
671: %where $a(m)$ is defined in \eqref{eq.walk38}.
672: where $P(N,k)$, $p(N,j,m,l)$ and $b(N,j,m,\infty)$
673: are given in \eqref{eq;walk12}, \eqref{eq-walk8} and \eqref{eq.walk30},
674: and $b(N,j,m,l)$ is given in \eqref{eq;walk4}.
675: We split the above sum into two pieces.
676: One part is the sum over (1) $|m-\frac{t}{1-t}N|\le N^{1/2+\epsilon/2}$
677: and (2) the rest, where $0<\epsilon<\frac13$ is fixed.
678: Then since $0\le p(N,j,m,l)\le 1$, we have from \eqref{eq.walk37}
679: \begin{equation}
680: \frac{1}{|P(N,k)|} \sum_{(2)} p(N,j,m,l) b(N,j,m,\infty)
681: = O(e^{-cN^{\epsilon}})
682: \end{equation}
683: for some $c>0$.
684:
685: We use Lemma \ref{lem-dep} to estimate $p(N,j,m,l)$ for $(j,m)$ in (1).
686: Set
687: \begin{eqnarray}
688: \tilde{t}^2=\frac{k-m-2c_0N\sqrt{\log N}}{N^2+ k-m-2c_0N\sqrt{\log N}},
689: \qquad
690: \beta=\frac{m-c_0\sqrt{N\log N}}{N+ m-c_0\sqrt{N\log N}}
691: \tilde{t}^{-1} ,
692: \end{eqnarray}
693: where $k$ satisfies the condition in \eqref{eq;walk2},
694: and we take $\tilde{t}>0$.
695: For $(j,m)$ in (1), they satisfy
696: \begin{eqnarray}\label{eq.walk74}
697: \tilde{t}=t+o(N^{-2/3}), %O(\frac{\sqrt{\log N}}{N}),
698: \qquad \beta=1+O(N^{-1/2+\epsilon/2}).
699: \end{eqnarray}
700: The first inequality of \eqref{eq.walk46} yields
701: $p(N,j,m,l)\le \phi(N,l,\tilde{t},\beta)+C_0N^{-d}$.
702: Set $\tilde{x}$ by \eqref{eq.walk16} where $t$ is replaced by $\tilde{t}$
703: and $l$ is given by $l=[\eta(t)N+x\rho(t)N^{1/3}]$.
704: Let $M?M_0$ satisfies $-M<2x<M$ where $M_0$ is given in
705: Proposition \ref{prop1}.
706: From \eqref{eq.walk74}, we have
707: \begin{equation}\label{2.76}
708: \tilde{x}=x+o(1),
709: \end{equation}
710: Proposition \ref{prop1} implies that
711: \begin{equation}
712: |\phi(N,l,\tilde{t},\beta) - F_1(\tilde{x})|\le
713: \frac{C(M)}{l^{1/3}}+Ce^{-cM^{3/2}},
714: \qquad l=[\eta(t)N+x\rho(t)N^{1/3}]
715: \end{equation}
716: for large $N$.
717: %Since $\tilde{x}=x+O(\frac{\sqrt{\log N}}{N})$ from \eqref{eq.walk74},
718: %and
719: Since $F_1'(x)=-\frac12(u(x)+v(x))F_1(x)$ is bounded for $x\in\R$,
720: we have from \eqref{2.76} that
721: $F_1(\tilde{x})=F_1(x)+ o(1)$. %O(\frac{\sqrt{\log N}}{N})$.
722: Thus we have for large $N$,
723: \begin{equation}
724: p(N,j,m,l)\le F_1(x) + o(1),
725: \end{equation}
726: where $o(1)$ term is independent of $(j,m)$ in (1)
727: and vanishes as $N\to\infty$.
728: Thus using Lemma \ref{lem1}, we have for large $N$,
729: \begin{equation}
730: \begin{split}
731: \Prob(L_1(N,k)\le l) &\le \frac{1}{|P(N,k)|}
732: \sum_{(1)} (F_1(x)+o(1)) b(N,j,m,\infty) + O(e^{-cN^{2\epsilon}})\\
733: &= F_1(x) + o(1),
734: \qquad l=[\eta(t)N+x\rho(t)N^{1/3}]
735: \end{split}
736: \end{equation}
737:
738: Similarly, we obtain the lower bound using the second inequality
739: of \eqref{eq.walk46}.
740: Thus we proved \eqref{eq;walk2}.
741:
742: The convergence of moments is also similar
743: using \eqref{walk16} and \eqref{walk17}
744: (cf. Section 8 of \cite{BR2}).
745: \end{proof}
746:
747:
748:
749:
750: