1: \documentclass[12pt]{article}
2:
3: \usepackage{amsmath,amsfonts,amssymb,amsthm,pstricks,pst-plot,pst-node,eucal}
4: \usepackage[dvips]{graphics}
5:
6: %\def\Sb{_\multilimits@}
7: %\def\endSb{\crcr\egroup\egroup\egroup}
8:
9:
10:
11: \newcommand{\Z}{\mathbb{Z}}
12: \newcommand{\R}{\mathbb{R}}
13: \newcommand{\oh}{\tfrac12}
14: \newcommand{\fS}{\mathfrak{S}}
15: \newcommand{\tfS}{\widetilde{\mathfrak{S}}}
16: \newcommand{\fM}{\mathfrak{M}}
17: \newcommand{\bK}{\mathsf{K}}
18: \newcommand{\bE}{\mathsf{E}}
19: \newcommand{\bB}{\mathsf{B}}
20: \newcommand{\cL}{\mathcal{L}}
21: \newcommand{\bS}{\mathsf{S}}
22: \newcommand{\bfS}{\boldsymbol{\mathfrak{S}}}
23: \DeclareMathOperator{\Prob}{Prob}
24: \DeclareMathOperator{\Ad}{Ad}
25: \DeclareMathOperator{\Var}{Var}
26: \DeclareMathOperator{\Ai}{Ai}
27: \DeclareMathOperator{\Res}{Res}
28: \newcommand{\cdd}{\,\cdot\,}
29: \newcommand{\cl}{\{\lambda(t)\}}
30: \newcommand{\vac}{v_\emptyset}
31: \newcommand{\Dt}{\Delta t}
32: \newcommand{\Dh}{\Delta h}
33: \newcommand{\Pl}{\text{Planch}}
34: \newcommand{\qd}[1]{(#1;q)_\infty}
35:
36: \newcommand{\timeprod}{\operatornamewithlimits{\overleftarrow{\prod}}}
37:
38: \newpsobject{showgrid}{psgrid}{subgriddiv=1,griddots=5,gridlabels=0pt}
39:
40:
41: \newtheorem{lemma}{Lemma}
42: \newtheorem{theorem}{Theorem}
43: \newtheorem{corollary}{Corollary}
44: \DeclareMathOperator{\dil}{dilog}
45: \DeclareMathOperator{\dilq}{dilog_q}
46: \DeclareMathOperator{\grad}{grad}
47:
48: \newcommand{\LV}{\Lambda^{\frac\infty2}V}
49: \newcommand{\ul}{\underline}
50: \newcommand{\al}{\alpha}
51:
52:
53:
54:
55: \theoremstyle{definition}
56: \newtheorem{remark}{Remark}
57: \newtheorem{definition}{Definition}
58:
59: \begin{document}
60:
61: \title{Correlation function of Schur process with
62: application to local geometry of a random
63: 3-dimensional Young diagram}
64: \author{Andrei Okounkov and Nikolai Reshetikhin\thanks{
65: Department of Mathematics, University of California at
66: Berkeley, Evans Hall \#3840,
67: Berkeley, CA 94720-3840. E-mail: okounkov@math.berkeley.edu,
68: reshetik@math.berkeley.edu }
69: }
70: \date{}
71: \maketitle
72:
73: \begin{abstract}
74: Schur process is a time-dependent analog of the Schur measure on
75: partitions studied in \cite{O}. Our first result is that the correlation
76: functions of the Schur process are
77: determinants with a kernel that has a nice contour integral
78: representation in terms of the parameters of the process.
79: This general result is then applied to a particular specialization
80: of the Schur process, namely to random 3-dimensional Young diagrams.
81: The local geometry of a large random 3-dimensional diagram is described
82: in terms of a determinantal point process on a 2-dimensional lattice with
83: the incomplete beta function kernel (which generalizes the discrete sine
84: kernel). A brief discussion of the universality of this answer
85: concludes the paper.
86: \end{abstract}
87:
88: \section{Introduction}
89:
90: \subsection{Schur measure and Schur process}
91:
92: \subsubsection{}
93:
94: This paper is a continuation of \cite{O}. The Schur measure, introduced
95: in \cite{O}, is a measure on partitions $\lambda$ which weights a
96: partitions $\lambda$ proportionally to $s_\lambda(X) \, s_\lambda(Y)$,
97: where $s_\lambda$ is the Schur function and $X$ and $Y$ are two
98: sets of variables.
99:
100:
101:
102: Here we consider
103: a time-dependent version of the Schur measure which we call
104: the Schur process, see Definition \ref{defS}.
105: This is a measure on sequences $\cl$ such
106: that
107: $$
108: \Prob(\cl)\,\propto\, \prod \bS^{(t)}(\lambda(t),
109: \lambda(t+1))\,,
110: $$
111: where the time-dependent weight $\bS^{(t)}(\lambda,\mu)$ is
112: a certain generalization of a skew Schur function. It is given
113: by a suitably regularized infinite minor of a certain Toeplitz
114: matrix. This determinants can be also interpreted as a Karlin-McGregor
115: non-intersection probability \cite{KM}. The distribution of each
116: individual $\lambda(t)$ is then a Schur measure with suitable
117: parameters.
118:
119: In this paper, we consider the Schur process only in
120: discrete time, although a continuous time formulation is
121: also possible, see Section \ref{cont}.
122:
123: The idea to make the Schur measure
124: time dependent was inspired, in part, by the paper
125: \cite{J3} which deals with some particular instances of
126: the general concept of the Schur processes introduced here.
127: For another development of the ideas of \cite{J3}, see
128: the paper \cite{PS} which appeared after the results of
129: the present paper were obtained.
130:
131:
132: \subsubsection{}
133: Our main interest in this paper are the correlation functions
134: of the Schur process which, by definition, are
135: the probabilities that the random set
136: $$
137: \fS( \{\lambda(t)\}) = \{(t,\lambda(t)_i-i+\oh)\}\,, \quad t,i\in Z\,,
138: $$
139: contains a given set $U\subset \Z \times (\Z+\oh)$.
140:
141: Using the same infinite wedge machinery as in \cite{O}
142: we prove that the correlation functions of the Schur
143: process have a determinantal form
144: $$
145: \Prob\left(U\subset \fS(\cl)\right) =
146: \det\big(K(u_i,u_j)\big)_{u_i,u_j\in U}\,,
147: $$
148: and give an explicit
149: contour integral representation for the correlation kernel $K$,
150: see Theorem \ref{t1}. This theorem is a generalization
151: of Theorem 2 in \cite{O}.
152:
153: \subsection{Asymptotics}
154:
155: \subsubsection{}
156:
157: The contour integral representation for correlation
158: kernel is particularly convenient for
159: asymptotic investigations. Only very elementary
160: means, namely residue calculus and basic saddle-point
161: analysis are needed to derive the asymptotics.
162:
163: \subsubsection{}
164:
165: We illustrate this in
166: Section \ref{secA} by computing explicitly the
167: correlation functions asymptotics for 3-dimensional
168: Young diagram in the bulk of their limit shape.
169: Three-dimensional diagrams, also known as plane
170: partition, is an old subject which has received
171: recently a lot of attention.
172: More specifically, the
173: measure on 3D diagrams $\pi$ such that
174: %
175: \begin{equation}
176: \Prob(\pi)\, \propto \, q^{|\pi|}\,, \quad 0<q<1\,,
177: \label{qpi}
178: \end{equation}
179: %
180: where $|\pi|$ is the volume of $\pi$, is
181: a particular specialization of the Schur process for which
182: the time parameter has the meaning of an extra spatial
183: dimension.
184: As $q\to 1$, a typical partition, suitably scaled,
185: approaches a limit
186: shape which was described in \cite{CK} and can be seen
187: in Figure \ref{f7} below. The existence of this limit
188: shape and some of its properties were
189: first established by A.~Vershik \cite{V},
190: who used direct combinatorial methods.
191:
192:
193: It is well known
194: that 3D diagrams are in bijection with certain rhombi
195: tiling of the plane, see for example Section
196: \ref{rtil} below. Local statistics of random domino tilings
197: have been the subject of intense recent studies, see
198: for example \cite{BP,CEP,CKP,K1} and the survey \cite{K2}.
199:
200:
201: The authors of \cite{CKP}
202: computed local correlations for domino tilings
203: with periodic boundary conditions
204: and conjectured that the same formula holds in the
205: thermodynamic limit for domino tilings of more
206: general regions, see Conjecture 13.5 in \cite{CKP}.
207:
208: {}From the point of view of statistical mechanics,
209: the argument
210: behind this conjecture can be the belief that
211: in the thermodynamic limit and away from the
212: boundary the local correlations depend
213: only on macroscopic parameters,
214: such as the density of tiles of a given kind.
215: In particular, in the case of 3D diagrams the densities
216: of rhombi of each of the 3 possible kinds
217: are uniquely fixed by the
218: tilt of the limit shape at the point in question.
219:
220: \subsubsection{}
221:
222:
223: In this paper, we approach the local geometry of
224: random 3D diagrams using the general exact formulas
225: for correlation functions of the Schur process. For 3D diagrams, these
226: general formulas specialize to contour integrals
227: involving the quantum dilogarithm function \eqref{qdil}, see Corollary
228: \ref{cor3D}.
229:
230:
231: We compute the $q\to 1$ limit of the
232: correlation kernel explicitly. The result turns out to
233: be the discrete incomplete beta kernel,
234: see Theorem \ref{t2}. One can show, see Section \ref{iint}, that this kernel is
235: a specialization of the kernel of \cite{CKP} when one of the
236: parameters vanishes in
237: agreement with Conjecture 13.5 of \cite{CKP}.
238:
239:
240: The incomplete beta kernel is also a bivariate generalization of the
241: discrete sine kernel which appeared in \cite{BOO} in the
242: situation of the Plancherel specialization of the
243: Schur measure, see also \cite{J3}.
244:
245:
246:
247: \subsection{Universality}
248:
249: Although we focus on
250: one specific asymptotic problem, our methods are both
251: very basic and completely general.
252: They should apply, therefore, with little or no
253: modification in a much wider variety of situations
254: yielding the same or analogous results.
255: In other words, both the methods and the results
256: should be universal in a large class of asymptotic
257: problems.
258:
259: As explained in Section \ref{eqtime},
260: this universality should be especially robust for
261: the equal time correlations, in which case the
262: discrete sine kernel should appear.
263: This is parallel to the situation with random
264: matrices \cite{J2}.
265:
266: \subsection{Acknowledgments}
267:
268: We are grateful to R.~Kenyon and A.~Vershik for fruitful
269: discussions. A.O.\ was partially supported by NSF grant DMS-0096246
270: and a Sloan foundation fellowship. N.~R.\
271: was partially supported by NSF grant DMS-0070931.
272: Both authors were partially supported by CRDF grant RM1-2244.
273:
274:
275: \section{Schur process}
276:
277: \subsection{Configurations}
278:
279: \subsubsection{}
280:
281: Recall that a partition is a sequence
282: $$
283: \lambda=(\lambda_1 \ge \lambda_2 \ge \lambda_3 \ge \dots \ge 0)
284: $$
285: of integers such that $\lambda_i=0$ for $i\gg 0$. The zero, or
286: empty, partition is denoted by $\emptyset$. The book \cite{M}
287: is a most comprehensive reference on partitions and symmetric functions.
288:
289: Schur process is a measure on sequences
290: $$
291: \{ \lambda(t)\}\,, \quad t \in \Z\,,
292: $$
293: where each $\lambda(t)$ is a partition and $\lambda(t)=\emptyset$
294: for $|t|\gg 0$. We call the variable $t$ the time, even though
295: it may have a different interpretation in applications.
296:
297: \subsubsection{}
298:
299: An example of such an object is a plane partition
300: which, by definition, is a 2-dimensional array of nonnegative numbers
301: $$
302: \pi=(\pi_{ij})\,, \quad i,j=1,2,\dots\,,
303: $$
304: that are nonincreasing as function of both $i$ and $j$ and such that
305: $$
306: |\pi|=\sum \pi_{ij}
307: $$
308: is finite. The plot of the function
309: $$
310: (x,y)\mapsto \pi_{\lceil x \rceil,\lceil y \rceil}\,, \quad x,y>0\,,
311: $$
312: is a 3-dimensional Young diagram with volume $|\pi|$.
313: For example, Figure \ref{f1} shows the
314: 3D diagram corresponding to the plane partition
315: %
316: \begin{equation}
317: \label{pi}
318: \pi =
319: \left (\begin {array}{cccc} 5&3&2&1\\\noalign{\medskip}4&3&1&1
320: \\\noalign{\medskip}3&2&1\\\noalign{\medskip}2&1\end {array}
321: \right )\,,
322: \end{equation}
323: %
324: where the entries that are not shown are zero.
325:
326: \begin{figure}[htbp]
327: \begin{center}
328: \scalebox{0.3}{\includegraphics{3dpart.ps}}
329: \caption{A 3-dimensional diagram $\pi$}
330: \label{f1}
331: \end{center}
332: \end{figure}
333:
334: We associate to $\pi$ the sequence $\cl$ of its diagonal slices,
335: that is, the sequence of partitions
336: %
337: \begin{equation}
338: \label{pila}
339: \lambda(t)=(\pi_{i,t+i})\,, \quad i\ge\max(0,-t) \,.
340: \end{equation}
341: It is easy to see that a configuration $\cl$ corresponds
342: to a plane partition if and only if it satisfies the
343: conditions
344: %
345: \begin{equation}
346: \label{laint}
347: \dots \prec \lambda(-2) \prec \lambda(-1) \prec \lambda(0)
348: \succ \lambda(1) \succ \lambda(2) \succ \dots\,,
349: \end{equation}
350: %
351: where $\lambda \succ \mu$ means that $\lambda$ and $\mu$ interlace,
352: that is,
353: $$
354: \lambda_1 \ge \mu_1 \ge \lambda_2 \ge \mu_2 \ge \lambda_3 \ge \dots \,.
355: $$
356: In particular, the configuration $\cl$ corresponding to
357: the diagram \eqref{pi} is
358: $$
359: (2) \prec (3,1) \prec (4,2) \prec (5,3,1) \succ (3,1)\succ (2,1)\succ (1)\,.
360: $$
361:
362: \subsubsection{}
363:
364: The mapping
365: $$
366: \lambda \mapsto \fS(\lambda)=\{\lambda_i - i +\oh\} \subset \Z+\oh
367: $$
368: is a bijection of the of the set of partitions and the
369: set $\bfS$ of subsets $\fS\subset \Z+\oh$ such that
370: $$
371: | \fS \setminus (\Z+\oh)_{<0} | = | (\Z+\oh)_{<0} \setminus \fS | <
372: \infty \,.
373: $$
374: The mapping
375: %
376: \begin{equation}
377: \label{fS}
378: \{\lambda(t)\}\mapsto
379: \fS( \{\lambda(t)\}) = \{(t,\lambda(t)_i-i+\oh)\} \subset \Z \times (\Z+\oh)\,,
380: \end{equation}
381: %
382: identifies the configurations of the Schur process
383: with certain subsets of $\Z \times (\Z+\oh)$. In other words, the mapping
384: \eqref{fS} makes the Schur process a random point field on
385: $\Z \times (\Z+\oh)$.
386:
387: For example, the subset $\fS( \{\lambda(t)\})$
388: corresponding to the 3D diagram from Figure \ref{f1} is shown in
389: Figure \ref{f2}. One can also visualize $\fS( \{\lambda(t)\})$ as a
390: collection of nonintersecting paths as in Figure \ref{f2}.
391:
392: \begin{figure}[htbp]\psset{unit=0.5cm}
393: \begin{center}
394: \begin{pspicture}(-5,-5)(5,5)\showgrid
395: \scriptsize
396: \psaxes[axesstyle=frame,Ox=-5,Oy=-5,Dx=2,Dy=2,ticks=none](-5,-5)(5,5)
397: \psdots[dotscale=1.5](-5,-.5)(-5,-1.5)(-5,-2.5)(-5,-3.5)(-5,-4.5)(-4,-.5)(-4,-1.5)(-4,-2.5)(-4,-3.5)(-4,-4.5)(-3,1.5)(-3,-1.5)(-3,-2.5)(-3,-3.5)(-3,-4.5)(-2,2.5)(-2,-.5)(-2,-2.5)(-2,-3.5)(-2,-4.5)(-1,3.5)(-1, .5)(-1,-2.5)(-1,-3.5)(-1,-4.5)(0,4.5)(0,1.5)(0,-1.5)(0,-3.5)(0,-4.5)(1,2.5)(1,-.5)(1,-2.5)(1,-3.5)(1,-4.5)(2,1.5)(2,-.5)(2,-2.5)(2,-3.5)(2,-4.5)(3, .5)(3,-1.5)(3,-2.5)(3,-3.5)(3,-4.5)(4,-.5)(4,-1.5)(4,-2.5)(4,-3.5)(4,-4.5)(5,-.5)(5,-1.5)(5,-2.5)(5,-3.5)(5,-4.5)
398:
399: \psline(-5,-.5)(-4,-.5)(-4,-.5)(-3,1.5)(-3,1.5)(-2,2.5)(-2,2.5)(-1,3.5)(-1,3.5)(0,4.5)(0,4.5)(1,2.5)(1,2.5)(2,1.5)(2,1.5)(3, .5)(3, .5)(4,-.5)(4,-.5)(5,-.5)
400: \psline(-5,-1.5)(-4,-1.5)(-4,-1.5)(-3,-1.5)(-3,-1.5)(-2,-.5)(-2,-.5)(-1, .5)(-1, .5)(0,1.5)(0,1.5)(1,-.5)(1,-.5)(2,-.5)(2,-.5)(3,-1.5)(3,-1.5)(4,-1.5)(4,-1.5)(5,-1.5)
401: \psline(-5,-2.5)(-4,-2.5)(-4,-2.5)(-3,-2.5)(-3,-2.5)(-2,-2.5)(-2,-2.5)(-1,-2.5)(-1,-2.5)(0,-1.5)(0,-1.5)(1,-2.5)(1,-2.5)(2,-2.5)(2,-2.5)(3,-2.5)(3,-2.5)(4,-2.5)(4,-2.5)(5,-2.5)
402: \psline(-5,-3.5)(-4,-3.5)(-4,-3.5)(-3,-3.5)(-3,-3.5)(-2,-3.5)(-2,-3.5)(-1,-3.5)(-1,-3.5)(0,-3.5)(0,-3.5)(1,-3.5)(1,-3.5)(2,-3.5)(2,-3.5)(3,-3.5)(3,-3.5)(4,-3.5)(4,-3.5)(5,-3.5)
403: \psline(-5,-4.5)(-4,-4.5)(-4,-4.5)(-3,-4.5)(-3,-4.5)(-2,-4.5)(-2,-4.5)(-1,-4.5)(-1,-4.5)(0,-4.5)(0,-4.5)(1,-4.5)(1,-4.5)(2,-4.5)(2,-4.5)(3,-4.5)(3,-4.5)(4,-4.5)(4,-4.5)(5,-4.5)
404: \end{pspicture}
405: \caption{Point field and nonintersecting paths corresponding to the diagram in Figure \ref{f1}}
406: \label{f2}
407: \end{center}
408: \end{figure}
409:
410:
411: \subsection{Probabilities}
412:
413:
414:
415:
416: \subsubsection{}
417:
418: Schur process, the formal definition of which is given
419: in Definition \ref{defS} below, is a measure on sequences $\cl$ such that
420: $$
421: \Prob(\{\lambda(t)\}) \,\propto\,
422: \prod_{t\in \Z} \bS^{(t)}(\lambda(t),\lambda(t+1))\,,
423: $$
424: where $\bS^{(t)}(\mu,\lambda)$ is a certain time-dependent transition
425: weight between the partition $\mu$ and $\lambda$ which will be
426: defined presently.
427:
428: The coefficient $\bS^{(t)}(\mu,\lambda)$
429: is a suitably regularized infinite minor
430: of a certain Toeplitz matrix. It can be viewed as a generalization of the
431: Jacobi-Trudy determinant for a skew Schur function or as a form
432: of Karlin-McGregor non-intersection probability \cite{KM}.
433:
434: \subsubsection{}
435:
436: Let a function
437: $$
438: \phi(z) = \sum_{k\in\Z} \phi_k \, z^k\,,
439: $$
440: be nonvanishing on the unit circle
441: $|z|=1$ with winding number $0$ and geometric mean $1$.
442: These two conditions mean that $\log \phi$ is a well defined
443: function with mean $0$ on the unit circle. For simplicity,
444: we additionally assume that $\phi$
445: is analytic in some neighborhood of the
446: unit circle. Many of the results below
447: hold under weaker assumptions on $\phi$
448: as can be seen, for example, by an approximation
449: argument. We will not pursue here the greatest analytic
450: generality.
451:
452:
453: \subsubsection{}
454:
455: Given two subsets
456: $$
457: X=\{x_i\},\, Y=\{y_i\} \in \bfS\,,
458: $$
459: we wish to assign a meaning to the following infinite
460: determinant:
461: %
462: \begin{equation}
463: \label{dT}
464: \det(\phi_{y_i-x_j}) \,.
465: \end{equation}
466: %
467: We will see that, even though there is no canonical way
468: to evaluate this determinant, different regularizations
469: differ by a constant which depends only on
470: $\phi$ and not on $X$ and $Y$.
471:
472:
473:
474: Since our goal is to
475: define probabilities only up to a constant factor,
476: it is clear that different regularizations lead to
477: the same random process.
478:
479: \subsubsection{}
480:
481: The function $\phi$ admits a Wiener-Hopf factorization
482: of the form
483: $$
484: \phi(z)=\phi^+(z)\, \phi^-(z)
485: $$
486: where the functions
487: $$
488: \phi^\pm(z)= 1 + \sum_{k\in\pm\mathbb{N}}
489: \phi^\pm_{k} \, z^{k}
490: $$
491: are analytic and nonvanishing in some neighborhood of the interior (resp.,
492: exterior) of the unit disk.
493:
494: There is a special case when the meaning of \eqref{dT} in
495: unambiguous, namely, if $\phi^-(z)=1$ (or $\phi^+(z)=1$)
496: then the matrix is almost unitriangular and the determinant
497: in \eqref{dT} is essentially a finite determinant. This
498: determinant is then a Jacobi-Trudy determinant for a skew Schur
499: function
500: $$
501: \det\left(\phi^+_{\lambda_i-\mu_j+j-i}\right) = s_{\lambda/\mu} (\phi^+)\,,
502: $$
503: where $s_{\lambda/\mu} (\phi^+)$ is the skew Schur function
504: $s_{\lambda/\mu}$ specialized so that
505: %
506: \begin{equation}
507: \label{phih}
508: h_k = \phi^+_k\,,
509: \end{equation}
510: %
511: where $h_k$ are the complete homogeneous symmetric functions.
512: Note that
513: %
514: \begin{equation}
515: \label{svan}
516: s_{\lambda/\mu}(\phi^+)=0\,, \quad \mu\not\subset\lambda \,.
517: \end{equation}
518: %
519:
520: \subsubsection{}
521:
522: \begin{definition}\label{defW}
523: We define the transition weight by the following formula
524: %
525: \begin{equation}
526: \label{T}
527: \bS_\phi(\mu,\lambda)=\sum_\nu s_{\mu/\nu}(\phi^-)\, s_{\lambda/\nu}(\phi^+) \,,
528: \end{equation}
529: %
530: where $s_{\lambda/\nu}(\phi^+)$ is defined by \eqref{phih} and $s_{\mu/\nu} (\phi^-)$ is the skew Schur function $s_{\mu/\nu}$
531: specialized so that
532: %
533: \begin{equation*}
534: h_k = \phi^-_{-k}\,.
535: \end{equation*}
536: %
537: \end{definition}
538:
539: Note that if the determinant $\det(\phi_{\lambda_i-\mu_j+j-i})$ were
540: unambiguously defined and satisfied the Cauchy-Binet formula then
541: it would equal \eqref{T} because
542: $$
543: \big(\,\phi_{i-j}\,\big) = \big(\,\phi^+_{i-j}\,\big) \, \big(\,\phi^-_{i-j}\,\big)\,.
544: $$
545: Also note that because of \eqref{svan}
546: the sum in \eqref{T} is finite, namely, it ranges over all
547: $\nu$ such that $\nu\subset\mu,\lambda$.
548:
549: \subsubsection{}
550:
551: In what follows, we will assume that the reader is familiar
552: with the basics of the infinite wedge formalism, see for
553: example Chapter 14 of the book \cite{K} by V.~Kac.
554: An introductory account of this formalism, together
555: with some probablistic applications, can be found
556: in the lectures \cite{O2}. We will
557: use the notation conventions of the appendix to \cite{O}
558: summarized for the reader's convenience in the appendix to
559: this paper.
560:
561: Consider the following vertex operators
562: $$
563: \Gamma_\pm(\phi)=\exp\left(
564: \sum_{k=1}^\infty (\log \phi)_{\mp k} \, \, \alpha_{\pm k} \right)\,,
565: $$
566: where $(\log \phi)_k$ denotes the coefficient of $z^{k}$
567: in the Laurent expansion of $\log \phi (z)$. In particular, if
568: the algebra of symmetric functions is specialized as in
569: \eqref{phih} then
570: $$
571: (\log \phi)_k = \frac{p_k}{k}\,,\quad k=1,2,3,\dots\,,
572: $$
573: where $p_k$ is the $k$th power-sum symmetric function.
574:
575: It is well known, see for example Exercise 14.26 in \cite{K},
576: that the matrix coefficient of the operators $\Gamma_\pm$
577: are the skew Schur functions, namely
578: $$
579: (\Gamma_-(\phi) \, v_\mu,
580: v_\lambda) = s_{\lambda/\mu}(\phi^+)\,.
581: $$
582: It follows that
583: %
584: \begin{equation}
585: \label{Tver}
586: \bS_\phi(\mu,\lambda) = (\Gamma_-(\phi) \, \Gamma_+(\phi) \, v_\mu,
587: v_\lambda) \,.
588: \end{equation}
589: %
590: It is clear from the commutation relation
591: %
592: \begin{equation}
593: \label{commG}
594: \Gamma_+(\phi) \Gamma_-(\phi) =
595: e^{\sum k \, (\log \phi)_k (\log \phi)_{-k}} \,
596: \Gamma_-(\phi) \, \Gamma_+(\phi)
597: \end{equation}
598: %
599: that different ordering prescriptions in \eqref{Tver},
600: which produce different regularizations of \eqref{dT}, all
601: differ by a constant independent of $\mu$ and $\lambda$.
602:
603:
604:
605: \subsubsection{}
606:
607: Now suppose that for any half-integer $m\in \Z+ \oh$ we choose,
608: independently, a function
609: $$
610: \phi[m](z)=\sum_{k\in \Z} \phi_k[m] \, z^k
611: $$
612: as above so that the series
613: $$
614: \sum_{m\in\Z} \log\phi[m](z)
615: $$
616: converges absolutely and uniformly in some neighborhood of the unit disk.
617: This assumption is convenient but can be weakened.
618: The functions $\phi[m]$
619: will be the parameters of the Schur process.
620:
621: \begin{definition}\label{defS}
622: The
623: probabilities of the Schur process are given by
624: $$
625: \Prob(\{\lambda(t)\}) = \frac1Z \, \prod_{m\in \Z+1/2}
626: \bS_{\phi[m]}\left(\lambda(m-\oh),\lambda(m+\oh)\right)\,,
627: $$
628: where the transition weight $\bS_\phi$ is defined in
629: Definition \ref{defW} and $Z$ is the normalizing factor (partition function)
630: $$
631: Z=\sum_{\cl} \prod_{m\in \Z+1/2}
632: \bS_{\phi[m]}\left(\lambda(m-\oh),\lambda(m+\oh)\right) \,.
633: $$
634: \end{definition}
635:
636:
637: \subsubsection{}
638:
639: It follows from \eqref{Tver} that $Z$ is given by the
640: following matrix coefficient
641: %
642: \begin{equation}
643: \label{Zver}
644: Z=\left(\timeprod_{m \in \Z+1/2}
645: \Gamma_-(\phi[m]) \, \Gamma_+(\phi[m])
646: \, v_\emptyset, v_\emptyset\right) \,,
647: \end{equation}
648: %
649: where $\timeprod$ denotes the time-ordered product, that is,
650: the product in which operators are ordered from right to
651: left in the increasing time order.
652:
653:
654: Using \eqref{commG} and the following consequence of
655: \eqref{ann}
656: %
657: \begin{equation}
658: \label{fix}
659: \Gamma_+(\phi) \, \vac = \vac \,.
660: \end{equation}
661: %
662: we compute the matrix coefficient \eqref{Zver}
663: as follows
664: $$
665: Z=\exp\left(\sum_{m_1<m_2} \sum_k k \, (\log\phi[m_1])_k \,
666: (\log\phi[m_2])_{-k} \right)\,.
667: $$
668: Our growth assumptions on the functions $\phi[m]$
669: ensure the convergence of $Z$.
670:
671: \subsubsection{}
672: It is clear from the vertex-operator description that
673: %
674: \begin{equation}
675: \Prob(\lambda(t)=\mu) \, \propto \, s_\mu\left(\prod_{m<t} \phi^+[m]
676: \right) \, s_\mu\left(\prod_{m>t} \phi^-[m]
677: \right) \,,\label{Schr}
678: \end{equation}
679: %
680: where, for example, the first factor is the image of the
681: Schur function under the specialization that sets $h_k$ to
682: the coefficient of $z^k$ in the product of $\phi^+[m](z)$
683: over $m<t$.
684:
685: This means that the distribution of each individual $\lambda(t)$
686: is a Schur measure \cite{O} with parameters \eqref{Schr}.
687:
688: \subsubsection{}\label{restrict}
689:
690: More generally, the restriction of the Schur process to
691: any subset of times is again a Schur process with suitably
692: modified parameters. Specifically, let a subset
693: $$
694: \{t_k\} \subset \Z\,,\quad k\in \Z\,,
695: $$
696: be given and consider the restriction of the Schur process to this
697: set, that is, consider the process
698: $$
699: \{\widetilde\lambda(k)\}=\{\lambda(t_k)\}\,.
700: $$
701: It follows from the vertex operator description that
702: this is again a Schur process with parameters
703: $$
704: \widetilde\phi[l]=\prod_{t_{l-\frac12} < m <t_{l+\frac12}} \phi[m]\,,
705: \quad l,m\in\Z+\oh\,.
706: $$
707: The case of a finite set $\{t_k\}$ is completely analogous
708: and can be dealt with formally by allowing infinite values
709: of $t_k$'s.
710:
711: \subsubsection{}\label{cont}
712:
713: The restriction property from Section \ref{restrict}
714: forms a natural basis for
715: considering the Schur process in continuous time.
716:
717: For this we need a function $\cL(z,s)$ of a
718: continuous variable $s\in\R$ which will
719: play the role of the density of $\log\phi$. The restriction
720: $\{\lambda(t_k)\}$
721: of the process $\lambda(t)$, $t\in\R$, to any discrete
722: set of times $\{t_k\}$ is the Schur process with
723: parameters
724: $$
725: \phi[m](z) = \exp\left(\int_{t_{m-\frac12}}^{t_{m+\frac12}} \cL(z,s) \, ds\right)\,,
726: \quad m\in \Z+\oh\,.
727: $$
728:
729: \subsubsection{}
730:
731: Let $q\in (0,1)$ and consider the probability measure
732: $\fM_q$ on the
733: set of all 3D diagrams such that
734: $$
735: \Prob(\pi) \,\propto\, q^{|\pi|} \,.
736: $$
737: We claim that there exists a particular choice of the
738: parameters of the Schur process which yield this measure
739: under the correspondence \eqref{pila}. Concretely, set
740: %
741: \begin{equation}
742: \label{specp}
743: \phi_{3D}[m](z)=
744: \begin{cases}
745: (1-q^{|m|} z)^{-1}\,, & m<0\,,\\
746: (1-q^{|m|} z^{-1})^{-1}\,, & m>0 \,,
747: \end{cases}
748: \qquad m\in\Z+\oh\,.
749: \end{equation}
750: %
751: It is well known that if the algebra of the symmetric
752: functions is specialized so that
753: $$
754: h_k = c^k \,, \quad k=1,2,\dots \,,
755: $$
756: for some constant $c$, then
757: $$
758: s_{\lambda/\mu}=
759: \begin{cases}
760: c^{|\lambda|-|\mu|}\,, & \mu \prec \lambda\,,\\
761: 0\,, & \mu \not\prec \lambda\,.
762: \end{cases}
763: $$
764: Therefore, we have
765: %
766: \begin{multline*}
767: \bS_{\phi_{3D}[m]}\left(\lambda(m-\oh),\lambda(m+\oh)\right)=\\
768: \begin{cases}
769: q^{m(|\lambda(m-\frac12)|-|\lambda(m+\frac12)|)}\,, &
770: m<0,\, \lambda(m-\frac12) \prec \lambda(m+\frac12)\quad\textup{or}\\
771: &
772: m>0,\, \lambda(m-\frac12) \succ \lambda(m+\frac12)\,,\\
773: 0\,, & \textup{otherwise} \,.
774: \end{cases}
775: \end{multline*}
776: %
777: It follows that for the specialization
778: \eqref{specp} the Schur process is
779: supported on configurations of the shape \eqref{laint} and
780: the weight of a configuration \eqref{laint} is proportional to
781: $$
782: q^{\sum |\lambda(t)|}=q^{|\pi|} \,.
783: $$
784: In particular, the partition function $Z$ becomes
785: %
786: \begin{multline*}
787: Z_{3D}=\exp\left(\sum_{m_1,m_2=1/2}^\infty \sum_k \frac{q^{k(m_1+m_2)}}k \,\right)=\\
788: \prod_{m_1,m_2} (1-q^{m_1+m_2})^{-1} = \prod_{n=1}^\infty (1-q^n)^{-n} \,,
789: \end{multline*}
790: %
791: which is the well-known generating function, due to McMahon,
792: for 3D diagrams.
793:
794:
795:
796: \subsection{Correlation functions}
797:
798: \subsubsection{}
799:
800: \begin{definition}
801: Given a subset $U\subset \Z\times (\Z+\oh)$, define
802: the corresponding correlation function by
803: $$
804: \rho(U)=\Prob\left(U\subset \fS(\cl)\right)\,.
805: $$
806: These correlation functions depend on the parameters $\phi[m]$ of the Schur process.
807: \end{definition}
808:
809:
810: In this section we show that
811: %
812: \begin{equation}
813: \label{rhodet}
814: \rho(U)=\det\big(K(u_i,u_j)\big)_{u_i,u_j\in U}\,,
815: \end{equation}
816: %
817: for a certain kernel $K$ which will be computed explicitly.
818:
819: \subsubsection{}
820:
821: Suppose that
822: $$
823: U=\{u_1,\dots,u_n\}\,, \quad u_i=(t_i,x_i)\in \Z\times (\Z+\oh)\,,
824: $$
825: and the points $u_i$ are ordered so that
826: $$
827: t_1 \le t_2 \le t_3 \dots \le t_n \,.
828: $$
829: For convenience, we set
830: $t_0 =- \infty$. From \eqref{Tver}
831: and \eqref{yn} it is clear that
832: %
833: \begin{equation}
834: \label{rver}
835: \rho(U)=\frac1Z\, (R_U \, v_\emptyset,v_\emptyset)\,,
836: \end{equation}
837: %
838: where $R_U$ is the following operator
839: $$
840: R_U = \timeprod_{m>t_n} \Gamma_-(\phi[m]) \, \Gamma_+(\phi[m])
841: \timeprod_{i=1..n} \left(\psi_{x_i} \psi_{x_i}^* \,
842: \timeprod_{t_{i-1} < m < t_i} \Gamma_-(\phi[m]) \, \Gamma_+(\phi[m])
843: \right)
844: $$
845:
846: \subsubsection{}
847:
848: Define the operator
849: $$
850: \Psi_{x}(t)=\Ad\left(\prod_{m>t} \Gamma_+(\phi[m]) \,
851: \prod_{m<t} \Gamma_-(\phi[m])^{-1} \right) \cdot \psi_x \,,
852: $$
853: where $\Ad$ denotes the action by conjugation, and define
854: the operator $\Psi^*_x(t)$ similarly. Note the ordering of
855: the vertex operators inside the $\Ad$ symbol is immaterial
856: because the vertex operators commute up to a central element.
857:
858: It follows from \eqref{rver}, \eqref{fix}, and \eqref{Zver} that
859: $$
860: \rho(U)=\left(\timeprod \Psi_{x_i}(t_i)\, \Psi^*_{x_i}(t_i) \, \vac,\vac
861: \right) \,.
862: $$
863: Now we apply Wick formula in the following form
864:
865: \begin{lemma}[Wick formula]
866: Let $A_i = \sum_k a_{i,k} \, \psi_k$ and
867: $A^*_i = \sum_k a^*_{i,k}\, \psi^*_k$. Then
868: %
869: \begin{equation}
870: \label{Wick}
871: \left(\prod A_i \, A^*_i \, \vac, \vac\right)=
872: \det\big(K_A(i,j)\big)\,,
873: \end{equation}
874: %
875: where
876: $$
877: K_A(i,j)=
878: \begin{cases}
879: (A_i \, A^*_j \, \vac,\vac)\,, & i \ge j\,,\\
880: -(A^*_j \, A_i \, \vac, \vac)\,, & i < j \,.
881: \end{cases}
882: $$
883: \end{lemma}
884:
885: \begin{proof}
886: Both sides of \eqref{Wick} are linear in $a_{i,k}$ and
887: $a^*_{i,k}$, therefore it suffices to verify
888: \eqref{Wick} for some linear basis in the space of possible
889: $A_i$'s and $A^*_j$'s. A convenient linear basis is formed
890: by the series \eqref{psiz} as the parameter $z$ varies. Using the canonical
891: anticommutation relation satisfied by $\psi(z)$ and $\psi^*(z)$ one then
892: verifies \eqref{Wick} directly.
893: \end{proof}
894:
895: We obtain the formula \eqref{rhodet} with
896: $$
897: K((t_1,x_1),(t_2,x_2))=
898: \begin{cases}
899: \left(\Psi_{x_1}(t_1)\, \Psi^*_{x_2}(t_2) \, \vac,\vac\right)\,,
900: & t_1 \ge t_2 \,,\\
901: -\left(\Psi^*_{x_2}(t_2)\, \Psi_{x_1}(t_1) \, \vac,\vac\right) \,,
902: & t_1 < t_2 \,.\\
903: \end{cases}
904: $$
905: Note that for $t_1\ne t_2$ the operators
906: $\Psi_{x_1}(t_1)$ and $\Psi^*_{x_2}(t_2)$ do not, in general,
907: anticommute so the time ordering is important. However,
908: for $t_1=t_2$ and $x_1\ne x_2$, these operators do
909: anticommute, so at equal time the ordering is immaterial.
910:
911:
912:
913:
914: \subsubsection{}
915:
916: A convenient generating function for the kernel $K$ can be
917: obtained as follows. Set
918: %
919: \begin{equation}
920: \psi(z)=\sum_{k\in\Z+1/2} z^k \, \psi_k \,,
921: \quad
922: \psi^*(z)=\sum_{k\in\Z+1/2} z^{-k} \, \psi^*_k \,.\label{psiz}
923: \end{equation}
924: %
925: Set also
926: $$
927: \Psi(t,z)=\Ad\left(\prod_{m>t} \Gamma_+(\phi[m]) \,
928: \prod_{m<t} \Gamma_-(\phi[m])^{-1} \right) \cdot \psi(z) \,,
929: $$
930: and define $\Psi^*(t,z)$ similarly.
931:
932: We have from \eqref{adal}
933: %
934: \begin{alignat*}{2}
935: \Ad(\Gamma_\pm(\phi))\,\cdot\, &\psi(z) &&= \phi^\mp(z^{-1})\, \psi(z)\,,\\
936: \Ad(\Gamma_\pm(\phi))\,\cdot\, &\psi^*(z) &&= \phi^\mp(z^{-1})^{-1}\, \psi(z)\,.
937: \end{alignat*}
938: %
939: and therefore
940: %
941: \begin{equation*}
942: \Psi(t,z)=\Phi(t,z) \,
943: \psi(z)\,, \quad
944: \Psi^*(t,z)=\Phi(t,z)^{-1} \,
945: \psi(z)\,,
946: \end{equation*}
947: %
948: where
949: %
950: \begin{equation}
951: \label{Phi}
952: \Phi(t,z)=\frac{\prod_{m>t} \phi^-[m](z^{-1})}{\prod_{m<t} \phi^+[m](z^{-1})} \,.
953: \end{equation}
954: %
955: Finally, it is obvious from the definitions that
956: %
957: \begin{align*}
958: (\psi(z) \, \psi^*(w) \, \vac,\vac)&=
959: \frac{\sqrt{zw}}{z-w}\,, \quad |z|>|w| \,, \\
960: -(\psi^*(w) \, \psi(z) \, \vac,\vac)&=
961: \frac{\sqrt{zw}}{z-w}\,, \quad |z|<|w| \,.
962: \end{align*}
963: %
964: Putting it all together, we obtain the following
965:
966: \begin{theorem}\label{t1}
967: We have
968: $$
969: \rho(U)=\det\big(K(u_i,u_j)\big)_{u_i,u_j\in U}\,,
970: $$
971: where the kernel $K$ is determined by the following
972: generating function
973: \begin{align}
974: \bK_{t_1,t_2}(z,w)&=\sum_{x_1,x_2\in\Z+\frac12} z^{x_1} \, w^{-x_2} \,
975: K((t_1,x_1),(t_2,x_2))\,, \label{bK1}\\
976: &=\frac{\sqrt{zw}}{z-w} \, \frac{\Phi(t_1,z)}
977: {\Phi(t_2,w)}\,. \label{bK2}
978: \end{align}
979: %
980: Here the function $\Phi(t,z)$ is defined by \eqref{Phi}, and
981: \eqref{bK1} is the expansion of \eqref{bK2}
982: in the region $|z|>|w|$ if $t_1 \ge t_2$ and
983: $|z|<|w|$ if $t_1 < t_2$.
984: \end{theorem}
985:
986:
987: \subsubsection{}
988:
989: In the special case of the measure $\fM_q$ on the
990: 3D diagrams the function
991: \eqref{Phi} specializes to the following function
992: %
993: \begin{equation}
994: \label{Phip}
995: \Phi_{3D}(t,z)=\frac
996: {\prod_{m > \max(0,-t)} (1-q^{m}/z)}
997: {\prod_{m > \max(0,t)} (1-q^{m} z)}\,, \quad m\in\Z+\oh\,.
998: \end{equation}
999: %
1000: Consider the following function
1001: %
1002: \begin{equation}
1003: \label{qdil}
1004: \qd{z}=\prod_{n=0}^\infty (1-q^n z) \,.
1005: \end{equation}
1006: %
1007: For various reasons, in particular because of the relation
1008: \eqref{qda}, this function is sometimes called the quantum
1009: dilogarithm function \cite{FK}.
1010:
1011: It is clear that the function \eqref{Phip} has the
1012: following expression in terms of the quantum dilogarithm
1013: $$
1014: \Phi_{3D}(t,z)=
1015: \begin{cases}
1016: \dfrac{\qd{q^{1/2}/z}}
1017: {\qd{q^{1/2+t} z}}\,, & t \ge 0\,,\vspace{5pt} \\
1018: \dfrac{\qd{q^{1/2-t}/z}}{\qd{q^{1/2} z}}\,, & t \le 0\,.
1019: \end{cases}
1020: $$
1021:
1022: \subsubsection{}\label{rtil}
1023:
1024: In order to make a better connection with the geometry of
1025: 3D diagrams, let us introduce a different encoding of
1026: diagrams by subsets in the plane. Given a plane
1027: partition
1028: $$
1029: \pi=(\pi_{ij})\,, \quad i,j=1,2,\dots\,,
1030: $$
1031: we set
1032: $$
1033: \tfS(\pi)=\{(j-i,\pi_{ij}-(i+j-1)/2)\}\,,
1034: \quad i,j=1,2,\dots \,.
1035: $$
1036: There is a well-known correspondence between 3D diagrams and
1037: tilings of the plane by rhombi. Namely, the tiles are the
1038: images of the faces of the 3D diagram under the projection
1039: %
1040: \begin{equation}
1041: (x,y,z)\mapsto (t,h)=(y-x,z-(x+y)/2)\,.\label{xyz}
1042: \end{equation}
1043: %
1044: The tiling corresponding to the diagram in Figure \ref{f1}
1045: is shown in Figure \ref{f3}.
1046: %
1047: \begin{figure}[htbp]\psset{unit=0.5cm}
1048: \begin{center}
1049: \begin{pspicture}(-7,-4)(7,5)
1050: \scriptsize
1051: \showgrid
1052: \psset{dimen=middle}
1053: \psaxes[axesstyle=frame,Ox=-7,Oy=-4,Dx=2,Dy=2,ticks=none](-7,-4)(7,5)
1054: \psdiamond[fillstyle=solid,fillcolor=lightgray](0,4.5)(1,.5)
1055: \psdiamond[fillstyle=solid,fillcolor=lightgray](1,2.0)(1,.5)
1056: \psdiamond[fillstyle=solid,fillcolor=lightgray](2, .5)(1,.5)
1057: \psdiamond[fillstyle=solid,fillcolor=lightgray](3,-1.0)(1,.5)
1058: \psdiamond[fillstyle=solid,fillcolor=lightgray](4,-2.5)(1,.5)
1059: \psdiamond[fillstyle=solid,fillcolor=lightgray](5,-3.0)(1,.5)
1060: \psdiamond[fillstyle=solid,fillcolor=lightgray](6,-3.5)(1,.5)
1061: \psdiamond[fillstyle=solid,fillcolor=lightgray](-1,3.0)(1,.5)
1062: \psdiamond[fillstyle=solid,fillcolor=lightgray](0,1.5)(1,.5)
1063: \psdiamond[fillstyle=solid,fillcolor=lightgray](1,-1.0)(1,.5)
1064: \psdiamond[fillstyle=solid,fillcolor=lightgray](2,-1.5)(1,.5)
1065: \psdiamond[fillstyle=solid,fillcolor=lightgray](3,-3.0)(1,.5)
1066: \psdiamond[fillstyle=solid,fillcolor=lightgray](4,-3.5)(1,.5)
1067: \psdiamond[fillstyle=solid,fillcolor=lightgray](-2,1.5)(1,.5)
1068: \psdiamond[fillstyle=solid,fillcolor=lightgray](-1,0.0)(1,.5)
1069: \psdiamond[fillstyle=solid,fillcolor=lightgray](0,-1.5)(1,.5)
1070: \psdiamond[fillstyle=solid,fillcolor=lightgray](1,-3.0)(1,.5)
1071: \psdiamond[fillstyle=solid,fillcolor=lightgray](2,-3.5)(1,.5)
1072: \psdiamond[fillstyle=solid,fillcolor=lightgray](-3,0.0)(1,.5)
1073: \psdiamond[fillstyle=solid,fillcolor=lightgray](-2,-1.5)(1,.5)
1074: \psdiamond[fillstyle=solid,fillcolor=lightgray](-1,-3.0)(1,.5)
1075: \psdiamond[fillstyle=solid,fillcolor=lightgray](0,-3.5)(1,.5)
1076: \psdiamond[fillstyle=solid,fillcolor=lightgray](-4,-2.5)(1,.5)
1077: \psdiamond[fillstyle=solid,fillcolor=lightgray](-3,-3.0)(1,.5)
1078: \psdiamond[fillstyle=solid,fillcolor=lightgray](-2,-3.5)(1,.5)
1079: \psdiamond[fillstyle=solid,fillcolor=lightgray](-5,-3.0)(1,.5)
1080: \psdiamond[fillstyle=solid,fillcolor=lightgray](-4,-3.5)(1,.5)
1081: \psdiamond[fillstyle=solid,fillcolor=lightgray](-6,-3.5)(1,.5)
1082: \end{pspicture}
1083: \caption{Horizontal tiles of the tiling corresponding to diagram in Figure \ref{f1}}
1084: \label{f3}
1085: \end{center}
1086: \end{figure}
1087:
1088: It is clear that under this
1089: correspondence the horizontal faces of a 3D diagram are
1090: mapped to the horizontal tiles and that the positions
1091: of the horizontal tiles uniquely determine the tiling and
1092: the diagram $\pi$. The set
1093: $$
1094: \tfS(\pi) \subset \Z \times \oh \, \Z
1095: $$
1096: is precisely the set of the centers of the horizontal tiles.
1097: It is also clear that if $\cl$ corresponds to the diagram
1098: $\pi$, then
1099: $$
1100: (t,h) \in \tfS(\pi) \Leftrightarrow (t,x+|t|/2)\in \fS(\cl) \,.
1101: $$
1102: Theorem \ref{t1} specializes, therefore, to the following
1103: statement
1104:
1105: \begin{corollary}\label{cor3D}
1106: For any set $\{(t_i,h_i)\}$, we have
1107: $$
1108: \Prob\left(\{(t_i,h_i)\} \subset \tfS(\pi)\right) =
1109: \det\big[K_{3D} ((t_i,h_i),(t_j,h_j)) \big]\,,
1110: $$
1111: where the kernel $K_{3D}$ is given by the following formula
1112: %
1113: \begin{multline}
1114: \label{K3D}
1115: K_{3D} ((t_1,h_1),(t_2,h_2)) = \\
1116: \frac1{(2\pi i)^2}
1117: \int_{|z|=1\pm\epsilon} \int_{|w|=1\mp\epsilon}
1118: \frac1{z-w} \frac{\Phi_{3D}(t_1,z)}{\Phi_{3D}(t_2,w)}
1119: \frac{dz \, dw}{z^{h_1+\frac{|t_1|+1}2} w^{-h_2-\frac{|t_2|-1}2}}\,.
1120: \end{multline}
1121: %
1122: Here the function $\Phi_{3D}(t,z)$ is defined by
1123: \eqref{Phip}, $0<\epsilon\ll 1$, and one picks the plus sign if $t_1 \ge t_2$
1124: and the negative sign otherwise.
1125: \end{corollary}
1126:
1127:
1128:
1129: \section{Asymptotics}\label{secA}
1130:
1131: \subsection{The local shape of a large 3D diagram}
1132:
1133: \subsubsection{}
1134:
1135: Our goal in this section is to illustrate how suitable
1136: is the formula \eqref{bK2} for asymptotic analysis. In
1137: order to be specific, we work out one concrete
1138: example, namely the local shape of a 3D diagram
1139: distributed according to the measure $\fM_q$ as $q\to 1$.
1140: Our computations, however, will be of a very abstract and general
1141: nature and applicable to a much wider variety of
1142: specialization.
1143:
1144:
1145: The reader will notice that the passage to the
1146: asymptotics in \eqref{bK2} is so straightforward that even
1147: the saddle-point analysis is needed in only a very weak
1148: form, namely as the statement that
1149: %
1150: \begin{equation}\label{saddle}
1151: \int_\gamma e^{M S(x)} \, dx \to 0 \,, \quad M\to +\infty\,,
1152: \end{equation}
1153: %
1154: provided the function $S(x)$ is smooth and $\Re S(x)< 0$
1155: for all but finitely many points $x\in\gamma$.
1156:
1157:
1158: \subsubsection{}
1159: Let $q=e^{-r}$ and $r\to +0$. We begin with the following
1160:
1161: \begin{lemma}
1162: We have the following convergence in probability
1163: $$
1164: r^3 |\pi| \to
1165: 2\zeta(3) \,, \quad r\to +0 \,,
1166: $$
1167: where $|\pi|$ is
1168: the volume of a 3D diagram
1169: $\pi$ sampled from the measure $\fM_q$.
1170: \end{lemma}
1171: \begin{proof}
1172: First consider the expectation of $|\pi|$
1173: \begin{equation*}
1174: \bE|\pi| = \frac{q\frac{d}{dq} Z_{3D}}{Z_{3D}} =
1175: \sum_{n\ge 1} \frac{n^2 q^n}{1-q^n} =
1176: \sum_{n,k\ge 1} n^2 q^{nk} =
1177: \sum_k \frac{q^k(1+q^k)}{(1-q^k)^3}
1178: \sim
1179: \frac{2\zeta(3)}{r^3} \,.
1180: \end{equation*}
1181: %
1182: Similarly, the variance of $|\pi|$ behaves like
1183: $$
1184: \Var |\pi| = q\frac{d}{dq} \, \bE|\pi| =
1185: o(r^{-6})\,,
1186: $$
1187: whence $\Var (r^3|\pi|) \to 0$, which concludes the proof.
1188: \end{proof}
1189:
1190: \subsubsection{}
1191:
1192:
1193: It follows that as $r\to +0$ the typical 3D diagram $\pi$,
1194: scaled by $r$ in all directions, approaches the suitably scaled
1195: limit shape for typical 3D diagrams of a large volume described in \cite{CK}.
1196: Below we will
1197: also see this limit shape appear from our calculations.
1198:
1199: We are interested in the $r\to+0$ limiting local structure of $\pi$
1200: in the neighborhood of various points in the limit shape. In
1201: other words, we are interested in the limit of the
1202: kernel \eqref{K3D} as
1203: $$
1204: r t_i \to \tau \,, \quad r h_i \to \chi \,,
1205: $$
1206: where the variables
1207: $\tau$ and $\chi$ describe the global position on the limit
1208: shape,
1209: in such a way that the relative distances
1210: $$
1211: \Dt=t_1-t_2\,, \quad \Dh=h_1-h_2 \,,
1212: $$
1213: remain fixed. This limit is easy to obtain by a combination
1214: of residue calculus with saddle-point argument. Since the
1215: measure $\fM_q$ is obviously symmetric with respect to
1216: the reflection $t\mapsto -t$, we can without loss of generality
1217: assume that $\tau \ge 0$ in our computations.
1218:
1219: \subsubsection{}
1220:
1221: We have the following $r\to+0$ asymptotics:
1222: %
1223: \begin{equation}
1224: \label{qda}
1225: \ln \qd{z} \sim r^{-1} \, \int_0^z \frac{\ln(1-w)}{w} \, dw =
1226: - r^{-1} \, \dil(1-z) \,,
1227: \end{equation}
1228: %
1229: where
1230: $$
1231: \dil(1-z)=\sum_{n} \frac{z^n}{n^2}\,, \quad |z|\le 1\,,
1232: $$
1233: analytically continued
1234: with a cut along $(1,+\infty)$.
1235:
1236: Introduce the following
1237: function
1238: $$
1239: S(z;\tau,\chi)=-(\tau/2+\chi)\ln z - \dil(1-1/z) +
1240: \dil(1-e^{-\tau} z)
1241: $$
1242: and recall that we made the assumption that $\tau\ge 0$.
1243: The function $S(z;\tau,\chi)$ is analytic in the complex
1244: plane with cuts along $(0,1)$ and $(e^\tau,+\infty)$.
1245:
1246:
1247: As $r\to +0$, the exponentially large term in the integrand
1248: in \eqref{K3D} is
1249: %
1250: \begin{equation}
1251: \label{expb}
1252: \exp\left(\frac1r(S(z;\tau,\chi)-S(w;\tau,\chi))\right)\,.
1253: \end{equation}
1254: %
1255: The saddle-point method suggests, therefore, to look at the
1256: critical points of the function $S(z;\tau,\chi)$. Since
1257: $$
1258: z \frac{d}{dz} S(z;\tau,\chi) = -\tau/2 - \chi -
1259: \log(1-1/z)(1-e^{-\tau} z) \,,
1260: $$
1261: the critical points of $S$ are the roots of the quadratic
1262: polynomial
1263: $$
1264: (1-1/z)(1-e^{-\tau} z)=e^{-\tau/2-\chi}\,.
1265: $$
1266: The two roots of this polynomial are complex conjugate
1267: if
1268: %
1269: \begin{equation}
1270: \label{ineqchi}
1271: |e^{\tau/2}+e^{-\tau/2}-e^{-\chi}| < 2\,,
1272: \end{equation}
1273: %
1274: which can be expressed equivalently as
1275: %
1276: \begin{equation}
1277: - 2 \ln \left(2 \cosh \frac\tau4\right) <
1278: \chi < - 2 \ln \left(2 \sinh \frac\tau4\right) \,.
1279: \label{ineqchi2}
1280: \end{equation}
1281: %
1282: In the case when the roots are complex conjugate they lie on the circle
1283: $$
1284: \gamma=\{|z|=e^{\tau/2}\}
1285: $$
1286: As we shall see below, the inequality \eqref{ineqchi}
1287: describes precisely the possible values
1288: of $(\tau,\chi)$ that correspond to the bulk of
1289: the limit shape.
1290:
1291:
1292: \subsubsection{}
1293: The following elementary properties of the function
1294: $S(z;\tau,\chi)$
1295: %
1296: \begin{gather*}
1297: S(\bar z; \tau,\chi) = \overline{S(z;\tau,\chi)}\,,\\
1298: S(z;\tau,\chi) + S(e^{\tau}/z;\tau,\chi) = - (\tau/2+\chi) \tau \,,
1299: \end{gather*}
1300: %
1301: imply that on the circle $\gamma$ the real part
1302: of $S$ is constant, namely,
1303: $$
1304: \Re S(z;\tau,\chi) = - (\tau/2+\chi) \tau/2 \,, \quad z\in\gamma\,.
1305: $$
1306: On $\gamma$ we also have
1307: $$
1308: z \frac{d}{dz} S(z;\tau,\chi) = 3\tau/2 - \chi - \ln|e^\tau-z|^2 \,,
1309: \quad z\in\gamma\,,
1310: $$
1311:
1312:
1313: {}From this it is clear that when the critical points of $S$
1314: are complex conjugate, they are the points of intersection of
1315: two following circles
1316: %
1317: \begin{equation}
1318: \label{intc}
1319: \{z_c,\bar z_c\}=\{|z|=e^{\tau/2}\} \cap
1320: \{|z-e^\tau|=e^{3\tau/4-\chi/2}\} \,.
1321: \end{equation}
1322: %
1323: This is illustrated in Figure \ref{f4} which also shows the vector
1324: field
1325: $$
1326: \nabla\left(\Re S(z;\tau,\chi)\right) = \frac{z^2}{e^\tau} \, \frac{d}{dz} S(z) \,,
1327: \quad z\in\gamma\,.
1328: $$
1329:
1330: \begin{figure}[htbp]
1331: \begin{center}
1332: \begin{pspicture}(-3.000000,-4.000000)(5.000000,4.000000)%\showgrid
1333: \uput[l](-2.117000,0){$\gamma$}
1334: \uput{10pt}[u](.995800,1.868000){$z_c$}
1335: \uput{10pt}[d](.995800,-1.868000){$\bar z_c$}
1336: \uput[l](2.117000,0){$e^{\tau/2}$}
1337: \uput[r](4.482000,0){$e^{\tau}$}
1338: \pcline{<-*}(2.504500,3.426000)(4.482000,0)
1339: \Aput{$e^{\frac34\tau-\frac12\chi}$}
1340: \SpecialCoor\pscircle(0,0){2.117000}
1341: \psarc(4.482000,0){3.955000}{120}{240}
1342: \psline{->}(2.117000;0.000000)(3.146000;0.000000)
1343: \psline{->}(2.117000;18.000000)(2.992000;18.000000)
1344: \psline{->}(2.117000;36.000000)(2.646000;36.000000)
1345: \psline{->}(2.117000;54.000000)(2.270000;54.000000)
1346: \psline{->}(2.117000;72.000000)(1.938000;72.000000)
1347: \psline{->}(2.117000;90.000000)(1.665000;90.000000)
1348: \psline{->}(2.117000;108.000000)(1.452000;108.000000)
1349: \psline{->}(2.117000;126.000000)(1.291000;126.000000)
1350: \psline{->}(2.117000;144.000000)(1.180000;144.000000)
1351: \psline{->}(2.117000;162.000000)(1.115000;162.000000)
1352: \psline{->}(2.117000;180.000000)(1.093000;180.000000)
1353: \psline{->}(2.117000;198.000000)(1.115000;198.000000)
1354: \psline{->}(2.117000;216.000000)(1.180000;216.000000)
1355: \psline{->}(2.117000;234.000000)(1.291000;234.000000)
1356: \psline{->}(2.117000;252.000000)(1.452000;252.000000)
1357: \psline{->}(2.117000;270.000000)(1.665000;270.000000)
1358: \psline{->}(2.117000;288.000000)(1.938000;288.000000)
1359: \psline{->}(2.117000;306.000000)(2.270000;306.000000)
1360: \psline{->}(2.117000;324.000000)(2.646000;324.000000)
1361: \psline{->}(2.117000;342.000000)(2.992000;342.000000)
1362: \end{pspicture}
1363: \caption{Gradient of $\Re S(z)$ on the circle $\gamma=\{|z|=e^{\tau/2}\}$}
1364: \label{f4}
1365: \end{center} \end{figure}
1366:
1367:
1368: \subsubsection{}
1369:
1370: Now we are prepared to do the asymptotics in \eqref{K3D}.
1371: We can deform the contours of the integration in \eqref{K3D}
1372: as follows
1373: $$
1374: K_{3D} ((t_i,h_i),(t_j,h_j)) =
1375: \frac1{(2\pi i)^2}
1376: \int_{(1\pm\epsilon)\gamma} dz \int_{(1\mp\epsilon)\gamma} dw \quad
1377: \dots
1378: $$
1379: where dots stand for the same integrand as in \eqref{K3D},
1380: $0<\epsilon\ll 1$, and we pick the plus sign if $t_1 \ge t_2$
1381: and the negative sign otherwise.
1382:
1383: Now we define the contours $\gamma_>$, $\gamma_<$, $\gamma_+$,
1384: $\gamma_-$. This definition will be illustrated by Figure \ref{f5}.
1385: The contour $\gamma_>$ is the circle $|z|=e^{\tau/2}$ slightly
1386: deformed in the direction of the gradient of $\Re S$, see Figure
1387: \ref{f4}. Similarly, the contour $\gamma_{<}$ is the
1388: same circle $|z|=e^{\tau/2}$ slightly pushed in the opposite
1389: direction. The contours $\gamma_\pm$ are the arcs of the
1390: circle $|z|=e^{\tau/2}$ between $\bar z_c$ and $z_c$, oriented
1391: toward $z_c$.
1392:
1393: \begin{figure}[htbp]
1394: \begin{center}
1395: \begin{pspicture}(-4,-4)(4,4)%\showgrid
1396: \psarc[arrowsize=5pt]{<-*}(0,0){3}{60}{300}
1397: \psarc[arrowsize=5pt]{*->}(0,0){3}{-60}{60}
1398: %\pscircle[linestyle=dashed](1.5,0){2.5981}
1399: %\pscircle[linestyle=dotted](-1.5,0){3.9686}
1400: \psccurve[linestyle=dashed](-2,0)(0,2.5)(1.5,2.5981)(4,0)(1.5,-2.5981)(0,-2.5)
1401: \psccurve[linestyle=dotted]
1402: (-4,0)(0,3.5)(1.5,2.5981)(2,0)(1.5,-2.5981)(0,-3.5)
1403: \uput{3pt}[l](-4,0){$\gamma_<$}
1404: \uput{3pt}[r](-2,0){$\gamma_>$}
1405: \uput{3pt}[l](-3,0){$\gamma_-$}
1406: \uput{3pt}[r](3,0){$\gamma_+$}
1407: \uput{5pt}[ur](1.5,2.5981){$z_c$}
1408: \uput{5pt}[dr](1.5,-2.5981){$\bar z_c$}
1409: \psline(3,0)(0,0)(1.5,2.5981)
1410: \psarc(0,0){1}{0}{60}
1411: \SpecialCoor
1412: \uput[ur](1;30){$\theta_*$}
1413: \end{pspicture}
1414:
1415: \caption{Contours $\gamma_>$(dashed), $\gamma_<$(dotted), and
1416: $\gamma_\pm$} \label{f5}
1417: \end{center}
1418: \end{figure}
1419:
1420: Deforming the contours, and picking the residue at $z=w$, we
1421: obtain
1422: $$
1423: K_{3D} ((t_i,h_i),(t_j,h_j)) = \int^{(1)} + \int^{(2)} \,,
1424: $$
1425: where
1426: $$
1427: \int^{(1)} =
1428: \frac1{(2\pi i)^2}
1429: \int_{\gamma_<} dz \int_{\gamma_>} dw \quad
1430: \dots
1431: $$
1432: with the same integrand in as in \eqref{K3D} and
1433: $$
1434: \int^{(2)} = \frac1{2\pi i} \int_{\gamma_\pm}
1435: \frac
1436: {\qd{q^{1/2+t_2} w}}
1437: {\qd{q^{1/2+t_1} w}}
1438: \frac{dw}{w^{\Dh+\Dt/2+1}} \,,
1439: $$
1440: where we choose $\gamma_+$ if $t_1 \ge t_2$ and
1441: $\gamma_-$ otherwise. As we shall see momentarily,
1442: $$
1443: \int^{(1)} \to 0 \,, \quad r\to +0\,,
1444: $$
1445: while $\int^{(2)}$ has a simple limit.
1446:
1447: \subsubsection{}
1448:
1449: It is obvious that
1450: $$
1451: \int^{(2)} \to \frac1{2\pi i} \int_{\gamma_\pm}
1452: (1-e^{-\tau} w)^{\Dt}
1453: \frac{dw}{w^{\Dh+\Dt/2+1}}\,, \quad r\to +0 \,.
1454: $$
1455: The change of variables $w=e^\tau w'$ makes this integral
1456: a standard incomplete beta function integral:
1457: $$
1458: \int^{(2)} \to \frac{e^{-\tau(\Dh+\Dt/2)}}
1459: {2\pi i} \int_{e^{-\tau} \gamma_\pm}
1460: (1-w)^{\Dt} \frac{dw}{w^{\Dh+\Dt/2+1}}\,, \quad r\to +0 \,.
1461: $$
1462: Now notice that the prefactor $e^{-\tau(\Dh+\Dt/2)}$ will cancel
1463: out of any determinant with this kernel, so it can be ignored.
1464: We, therefore, make the following
1465:
1466:
1467:
1468: \begin{definition}
1469: Introduce the following incomplete beta function kernel
1470: $$
1471: \bB_\pm(k,l;z)= \frac{1}{2\pi i}\int_{\bar z}^{z} (1-w)^{k}
1472: w^{-l-1} \, dw\,,
1473: $$
1474: where the path of the integration crosses $(0,1)$ for the plus
1475: sign and $(-\infty,0)$ for the minus sign.
1476: \end{definition}
1477:
1478: \subsubsection{}
1479:
1480: It is also obvious from our construction that the function
1481: $\Re S(z)$ reaches its maximal value on the contour $\gamma_<$
1482: precisely at the points $z_c$ and $\bar z_c$. The same
1483: points are the minima of the function $S(z)$ on the contour
1484: $\gamma_>$. The behavior of the integral $\int^{(1)}$
1485: is thus determined by the term \eqref{expb} and, by the
1486: basic principle \eqref{saddle}, the limit of $\int^{(1)}$
1487: vanishes.
1488:
1489:
1490: \subsubsection{}
1491:
1492: We can summarize our discussion as follows. Set
1493: $z_*=e^{-\tau} z_c$, in other words,
1494: %
1495: \begin{equation}
1496: \label{zst}
1497: \{z_*, \bar z_*\} = \{|z|=e^{-\tau/2}\} \cap
1498: \{|z-1|=e^{-\tau/4-\chi/2}\}\,, \quad \Im z_*>0\,.
1499: \end{equation}
1500: %
1501: It is convenient to extend the meaning of
1502: $z_*$ to denote the point on the circle $|z|=e^{-\tau/2}$
1503: which is the closest point to the circle $|z-1|=e^{-\tau/4-\chi/2}$
1504: in case when the two circles do not intersect. In other
1505: words, we complement the definition \eqref{zst} by setting
1506: %
1507: \begin{equation}
1508: \label{degz}
1509: z_*=
1510: \begin{cases}
1511: e^{-\tau/2} \,, & \quad e^{-\chi/2} < e^{\tau/4}-e^{-\tau/4} \,, \\
1512: -e^{-\tau/2}\,, & \quad e^{-\chi/2} > e^{\tau/4}+e^{-\tau/4} \,. \\
1513: \end{cases}
1514: \end{equation}
1515:
1516: We have established the following
1517:
1518:
1519: \begin{theorem}\label{t2}
1520: Let $U=\{(t_i,h_i)\}$ and suppose that as $r\to+0$
1521: $$
1522: r t_i\to \tau\ge 0\,, \quad r h_i \to \chi\,,\quad i=1,2,\dots\,,
1523: $$
1524: in such a way that the differences
1525: $$
1526: \Dt_{ij}=t_i-t_j\,, \quad \Dh_{ij}=h_i-h_j\,,
1527: $$
1528: remain fixed. Then, as $r\to+0$,
1529: $$
1530: \Prob\{U\subset\tfS(\pi)\}\to
1531: \det\left[
1532: \bB_\pm\left(\Dt_{ij},\Dh_{ij}+\frac{\Dt_{ij}}2;z_*\right)
1533: \right]\,,
1534: $$
1535: where the point $z_*=z_*(\tau,\chi)$ is defined in \eqref{zst}and \eqref{degz}
1536: and the
1537: choice of the plus sign corresponds to $\Dt_{ij} =t_i - t_j\ge 0$.
1538: \end{theorem}
1539:
1540: \begin{remark}
1541: These formulas can be transformed, see Section \ref{iint}, into a
1542: double integral of form
1543: considered in \cite{CKP}, Proposition 8.5 and Conjecture 13.5.
1544: \end{remark}
1545:
1546:
1547:
1548: \begin{remark}
1549: Observe that the limit correlation are trivial in the cases
1550: covered by \eqref{degz}. In other words, they are nontrivial
1551: unless the inequality \eqref{ineqchi} is satisfied.
1552: This means that the inequality \eqref{ineqchi} describes
1553: the values of $(\tau,\chi)$ that correspond to the bulk of
1554: the limit shape.
1555: \end{remark}
1556:
1557:
1558:
1559: \subsubsection{}
1560:
1561: In particular, denote by $\rho_*(\tau,\chi)$ the
1562: limit of $1$-point correlation function
1563: $$
1564: K_{3D}((t,h),(t,h)) \to \rho_*(\tau,\chi)\,, \quad
1565: rt\to\tau, \, r h \to \chi\,.
1566: $$
1567: This is the limiting density of the horizontal tiles at
1568: the point $(\tau,\chi)$. We have the following
1569:
1570: \begin{corollary}\label{c2}
1571: The limiting density of horizontal tiles is
1572: $$
1573: \rho_*(\tau,\chi) = \frac{\theta_*}{\pi} \,,
1574: $$
1575: where $\theta_*=\arg z_*$ (see Figure \ref{f5}), that is,
1576: %
1577: \begin{equation}
1578: \theta_*=
1579: \arccos \left( \cosh\frac\tau2 - \frac{e^{-\chi}}2\right) \,. \label{th*}
1580: \end{equation}
1581: %
1582: \end{corollary}
1583:
1584: The level sets of the density as functions of
1585: $\tau$ and $\chi$ are plotted in Figure \ref{f6}. More
1586: precisely, Figure \ref{f6} shows the curves
1587: $$
1588: \theta_*(\tau,\chi)= \frac{k\pi}8\,, \quad k=0,\dots,8\,.
1589: $$
1590: The knowledge of density $\rho_*$ is equivalent to the
1591: knowledge of the limit shape, see Sections \ref{limsh1} and
1592: \ref{limsh2}. We point out, however, that additional
1593: analysis is needed to prove the convergence to the limit
1594: shape in a suitable metric as in \cite{CK}.
1595:
1596:
1597: \begin{figure}[htbp]
1598: \begin{center}
1599: \scalebox{0.6}{\includegraphics{3ddens.ps}}
1600: \caption{Level sets of the density of horizontal tiles}
1601: \label{f6}
1602: \end{center}
1603: \end{figure}
1604:
1605:
1606:
1607: \subsubsection{}
1608:
1609: Corollary \ref{c2} can be generalized as follows:
1610:
1611: \begin{corollary}
1612: The equal time correlations are given by the
1613: discrete sine kernel, that is, if
1614: $t_1=t_2=\dots$ then
1615: $$
1616: \Prob\{U\subset\tfS(\pi)\}\to
1617: \det\left[
1618: \frac{\sin(\theta_*(h_i-h_j))}{\pi(h_i-h_j)}
1619: \right]\,,
1620: $$
1621: as $r\to+0$.
1622: \end{corollary}
1623:
1624: \subsubsection{}\label{iint}
1625:
1626: The incomplete beta kernel $\bB_\pm$ can be transformed
1627: into a double integral of the form considered in \cite{CKP}
1628: as follows.
1629:
1630: Using the following standard integral
1631: $$
1632: \frac1{2\pi i}\int_{|z|=\alpha} \frac{z^{-k-1} \, dz}{1-\beta z}
1633: =
1634: \begin{cases}
1635: \beta ^k \,, & k\ge 0,\, |\alpha \beta|<1 \,,\\
1636: -\beta^k \,, & k<0, \, |\alpha\beta|>1 \,, \\
1637: 0\,, & \text{otherwise}\,,
1638: \end{cases}
1639: $$
1640: we can replace the condition
1641: $$
1642: |w-1|\lessgtr e^{-\tau/4-\chi/2}\,,
1643: $$
1644: in the definition of $\bB_\pm(k,l;z_*)$ by an extra
1645: integral. We obtain
1646: %
1647: \begin{equation}
1648: \bB_\pm(k,l;z_*)=\frac1{(2\pi i)^2}
1649: \iint_
1650: {
1651: \begin{subarray}{l}
1652: |w|=e^{-\tau/2}\\ |z|=e^{\tau/4+\chi/2}
1653: \end{subarray}
1654: }
1655: \frac{z^{-k-1} w^{-l-1}}{1-z+zw}\, dz\, dw\,,
1656: \label{eii}
1657: \end{equation}
1658: %
1659: where the plus sign corresponds to $k\ge 0$.
1660:
1661: \subsubsection{}\label{limsh1}
1662:
1663: Let $z(\tau,\chi)$ denote the $z$-coordinate of
1664: the point on the limit shape corresponding to the point
1665: $(\tau,\chi)$. This function can be obtained
1666: by integrating the density $\rho_*(\tau,\chi)$ as
1667: follows.
1668:
1669: Consider a tiling such as the one in the Figure
1670: \ref{f3} and the corresponding 3D diagram, which
1671: for the tiling in Figure \ref{f3} is shown in
1672: Figure \ref{f1}. It is clear that the $z$-coordinate
1673: of the face corresponding to a given horizontal tile equals
1674: the number of holes (that is, positions not occupied by a horizontal
1675: tile) below it. It follows that
1676: %
1677: \begin{equation}
1678: z(\tau,\chi)=\int_{-\infty}^\chi (1-\rho_*(\tau,s)) \, ds \,. \label{z}
1679: \end{equation}
1680: %
1681: Here, of course, the lower limit of integration can
1682: be any number between $-\infty$ and the lower boundary of the
1683: limit shape given by the equation \eqref{ineqchi2}.
1684:
1685: Integrating \eqref{th*} we obtain the following formula
1686: %
1687: \begin{equation}
1688: z(\tau,\chi)=\frac1\pi\int_0^{\pi-\theta_*}
1689: \frac{s \, \sin s \,ds}{\cos s + \cosh \frac\tau2}\,.\label{z1}
1690: \end{equation}
1691: %
1692: This integral can be evaluated in terms of the
1693: dilogarithm function. From \eqref{xyz} we can now compute
1694: the other two coordinates as follows
1695: %
1696: \begin{equation}
1697: x(\tau,\chi)=z(\tau,\chi)-\chi-\frac\tau2\,,
1698: \quad
1699: y(\tau,\chi)=z(\tau,\chi)-\chi+\frac\tau2\,,\label{xy}
1700: \end{equation}
1701: %
1702: which gives a parametrization of the limit shape.
1703: A plot of the limit shape is shown in Figure \ref{f7}
1704:
1705: \begin{figure}[htbp]
1706: \begin{center}
1707: \scalebox{0.6}{\includegraphics{3dlimshape.eps}}
1708: \caption{The limit shape}\label{f7}
1709: \end{center}
1710: \end{figure}
1711:
1712: \subsubsection{}\label{limsh2}
1713:
1714: To make the connection to the parametrization of the
1715: limit shape given in \cite{CK},
1716: let us now substitute for $\rho_*$ in the integral
1717: \eqref{z} the formula \eqref{eii}.
1718: After a simple coordinate change we obtain
1719: $$
1720: \rho_*(\tau,\chi) = \frac1{4\pi^2} \iint_0^{2\pi}
1721: \frac{du \,dv}{1+e^{\chi/2+\tau/4+iu}+e^{\chi/2-\tau/4+iv}} \,.
1722: $$
1723: Integrating this in $\chi$ we obtain
1724: $$
1725: z(\tau,\chi)=\frac1{2\pi^2} \iint_0^{2\pi}
1726: \ln \left|1+e^{\chi/2+\tau/4+iu}+e^{\chi/2-\tau/4+iv}\right| \,
1727: du \,dv \,.
1728: $$
1729: This together with \eqref{xy} is equivalent to following parametrization
1730: of the limit shape found in \cite{CK}
1731: $$
1732: (x,y,z)=\left(f(A,B,C)-2\ln A,f(A,B,C)-2\ln B, f(A,B,C)-2 \ln C
1733: \right)\,,
1734: $$
1735: where $A,B,C>0$ and
1736: $$
1737: f(A,B,C)=\frac1{2\pi^2} \iint_0^{2\pi}
1738: \ln \left|A+ B e^{iu}+C e^{iv}\right| \,
1739: du \,dv \,.
1740: $$
1741: This parametrization is manifestly symmetric in $x$, $y$, and $z$.
1742: However, it involves integrals that are more complicated to
1743: evaluate than the parametrization \eqref{z1}.
1744:
1745: Note that the shape considered in \cite{CK} differs from
1746: our by a factor of $2$ due to different scaling conventions.
1747:
1748:
1749: \subsection{Universality}
1750:
1751: \subsubsection{}
1752: The reader has surely noticed that all that we really needed
1753: for the asymptotics is to be able to deform the
1754: contours of integration as in Figure \ref{f5} so that
1755: the points of the intersection of $\gamma_>$ and $\gamma_<$
1756: the minima of the function $\Re S(z)$ on one curve and
1757: the maxima --- on the other. The intersection points
1758: are then forced to be the critical points of the function
1759: $S(z)$.
1760:
1761: It is clear that this
1762: principle is very general and applicable in a potentially
1763: very large variety of situations, such as, for example,
1764: in the case of Plancherel measure, see below. In particular,
1765: the existence of contours of a certain kind is a property
1766: which is preserved under small perturbations.
1767:
1768: \subsubsection{}
1769: One such perturbation would be to consider anisotropic
1770: partitions. The anisotropy in the $t$-direction is
1771: especially easy to introduce: one just should replace
1772: $q^{|m|}$ in \eqref{specp} by $q^{V(m)}$ for some
1773: function $V$.
1774:
1775:
1776: \subsubsection{}\label{eqtime}
1777: Observe that the asymptotics of the equal time correlations
1778: are especially easy to obtain in our approach. This is
1779: because
1780: $$
1781: \Res_{z=w}\frac{dz}{z-w} \, \frac{\Phi(t_1,z)}
1782: {\Phi(t_2,w)}=1\,, \quad t_1=t_2\,,
1783: $$
1784: and hence the analog of the integral $\int^{(2)}$ becomes simply
1785: the integral
1786: $$
1787: \int^{(2)} = \frac1{2\pi i} \int_{\gamma_\pm}
1788: \frac{dw}{w^{\Delta x+1}} \,,
1789: $$
1790: which leads to the discrete sine kernel in the variable
1791: $\Delta x$.
1792:
1793: \subsubsection{}
1794: Let us illustrate these general remarks by briefly discussing
1795: the asymptotics for
1796: the poissonized Plancherel measure \cite{BOO}. It corresponds
1797: to the following specialization of the Schur process
1798: $$
1799: \phi_{\Pl}[m]=
1800: \begin{cases}
1801: e^{\sqrt{\alpha} \, z}\,, & m=-\frac12\,,\\
1802: e^{\sqrt{\alpha} \, z^{-1}}\,, & m=\frac12\,,\\
1803: 1\,, & \text{otherwise}\,,
1804: \end{cases}
1805: $$
1806: where $\alpha>0$ is the poissonization parameter.
1807: In particular, only the partition $\lambda(0)$ is
1808: nontrivial. The correlation kernel at $t=0$ specializes
1809: to
1810: $$
1811: K_{\Pl}(x,y)=\frac1{(2\pi i)^2} \iint
1812: \frac{z^{-x-1/2} w^{y-1/2}}{z-w} \, e^{\sqrt{\alpha} (z-z^{-1}-w+w^{-1}) } \,
1813: dz\,dw \,,
1814: $$
1815: which can be expressed in terms of Bessel functions of the
1816: argument $2\sqrt{\alpha}$, see \cite{BOO,J1}.
1817:
1818:
1819:
1820:
1821: \subsubsection{}
1822:
1823: The $\alpha\to\infty$ asymptotics
1824: of the kernel $K_{\Pl}$ is easy to obtain from the
1825: classical asymptotics of the Bessel functions, see \cite{BOO}.
1826: It is, however, instructive to see
1827: how this can be done even more quickly in our
1828: framework. Assume that
1829: $$
1830: \frac{x}{\sqrt{\alpha}}, \frac{y}{\sqrt{\alpha}} \to \xi
1831: $$
1832: in such a way that $\Delta=x-y$ remains fixed. The critical
1833: points of the action
1834: $$
1835: S_{\Pl} = z-z^{-1}-\frac{\xi}{2}\, \ln z
1836: $$
1837: are complex precisely when $|\xi|<2$,
1838: in which case they are the points $e^{\pm i\theta}$, where
1839: $$
1840: \theta = \arccos(\xi/2) \,.
1841: $$
1842: So, the same argument as we employed above immediately yields
1843: the following formula from \cite{BOO}.
1844: $$
1845: K_{\Pl}(x,y) \to \frac{\sin \theta \Delta}{\pi \Delta}\,.
1846: $$
1847:
1848: \subsubsection{}
1849:
1850: Of course, a finer analysis (which was carried out in \cite{BOO})
1851: is needed to justify depoissonization in the asymptotics.
1852: In our situation, a similar problem is to pass from the
1853: $q\to 1$ asymptotics of the measure $\fM_q$ to the asymptotics
1854: of the uniform measures on partitions of a given volume $N$
1855: as $N\to\infty$. In other words, further work is needed to
1856: verify the equivalence of ensembles in the asymptotics.
1857:
1858: \subsubsection{}
1859:
1860: Similarly, finer analysis is needed to work with the asymptotics
1861: at the edges of the limit shapes where one expects to
1862: see the Airy kernel appear. In the edge scaling, the following
1863: equivalent version of the formula \eqref{bK2}
1864: %
1865: \begin{equation}\label{sumK}
1866: K((t_1,x_1),(t_2,x_2))=
1867: \sum_{m=1/2}^{\infty}
1868: [z^{x_1+m} w^{-x_2-m}] \, \frac{\Phi(t_1,z)}
1869: {\Phi(t_2,w)}\,,
1870: \end{equation}
1871: %
1872: may be helpful because, by analogy to the situation with the
1873: Plancherel measure \cite{BOO,J1}, one could expect to see this
1874: sum to become the integral representing the Airy kernel
1875: $$
1876: \frac{\Ai(x) \Ai'(y) - \Ai'(x) \Ai(y)}{x-y}=
1877: \int_0^\infty \Ai(x+s) \, \Ai(y+s) \, ds \,.
1878: $$
1879: Further discussion of the Airy-type asymptotics of the
1880: integrals of the form \eqref{bK2}, \eqref{K3D} can be
1881: found in \cite{O2}.
1882:
1883: \subsubsection{}
1884:
1885: Recently, the techniques of this paper were used in
1886: \cite{FS} to prove that, indeed, the boundary of a random 3D
1887: Young diagram converges to the Airy process.
1888:
1889: \appendix
1890:
1891: \section{Summary of the infinite wedge formulas}
1892:
1893: Let the space $V$ be spanned by $\ul{k}$, $k\in\Z+\oh$.
1894: The space $\LV$ is, by definition, spanned by vectors
1895: $$
1896: v_S=\ul{s_1} \wedge \ul{s_2} \wedge \ul{s_3} \wedge \dots\,,
1897: $$
1898: where $S=\{s_1>s_2>\dots\}\subset \Z+\oh$ is such a subset that
1899: both sets
1900: $$
1901: S_+ = S \setminus \left(\Z_{\le 0} - \oh\right) \,, \quad
1902: S_- = \left(\Z_{\le 0} - \oh\right) \setminus S
1903: $$
1904: are finite. We equip $\LV$ with the inner product
1905: in which the basis $\{v_S\}$ is orthonormal. In particular,
1906: we have the vectors
1907: $$
1908: v_\lambda=\ul{\lambda_1-\tfrac12} \wedge \ul{\lambda_2-\tfrac32}
1909: \wedge \ul{\lambda_4-\tfrac52} \wedge \dots \,,
1910: $$
1911: where $\lambda$ is a partition.
1912: The vector
1913: $$
1914: \vac = \ul{-\tfrac12} \wedge \ul{-\tfrac32} \wedge \ul{-\tfrac52} \wedge \dots
1915: $$
1916: is called the vacuum vector.
1917:
1918: The operator $\psi_k$ is the exterior multiplication by $\ul{k}$
1919: $$
1920: \psi_k \left(f\right) = \ul{k} \wedge f \,.
1921: $$
1922: The operator $\psi^*_k$ is the adjoint operator.
1923: These operators satisfy the canonical anti-commutation relations
1924: $$
1925: \psi_k \psi^*_k + \psi^*_k \psi_k = 1\,,
1926: $$
1927: all other anticommutators being equal to $0$. We have
1928: %
1929: \begin{equation}\label{yn}
1930: \psi_k \psi^*_k \,\, v_S =
1931: \begin{cases}
1932: v_S\,, & k \in S \,, \\
1933: 0 \,, & k \notin S \,.
1934: \end{cases}
1935: \end{equation}
1936: %
1937:
1938: The operators $\al_n$ defined by
1939: $$
1940: \al_n = \sum_{k\in\Z+\frac12} \psi_{k-n}\, \psi^*_k \,, \quad n=\pm 1,\pm2, \dots \,,
1941: $$
1942: satisfy the Heisenberg commutation relations
1943: $$
1944: \left[\al_n, \al_m\right] = n \, \delta_{n,-m} \,,
1945: $$
1946: see the formula. Clearly, $\al^*_n=\al_{-n}$. It is clear from definitions that %
1947: \begin{equation}\label{adal}
1948: [\al_n,\psi(z)]=z^n\, \psi(z)\,, \quad [\al_n,\psi^*(w)]=- w^{n}\, \psi^*(w)
1949: \end{equation}
1950: %
1951: and also that
1952: %
1953: \begin{equation}
1954: \label{ann}
1955: \al_n \, \vac = 0 \,, \quad n\le 0 \,.
1956: \end{equation}
1957:
1958:
1959: \begin{thebibliography}{99}
1960:
1961:
1962:
1963: \bibitem{BOO}
1964: A.~Borodin, A.~Okounkov, and G.~Olshanski,
1965: \emph{On asymptotics of the Plancherel measures for
1966: symmetric groups}, J.\ Amer.\ Math.\ Soc.\ \textbf{13} (2000), no.\ 3,
1967: 481--515.
1968:
1969: \bibitem{BP}
1970: R.~Burton and R.~Pemantle,
1971: \emph{Local characteristics, entropy and limit theorems for
1972: spanning trees and domino tilings via transfer-impendances},
1973: Ann.\ Prob.\ \textbf{21} (1993), 1329--1371.
1974:
1975: \bibitem{CK}
1976: R.~Cerf and R.~Kenyon,
1977: \emph{The low-temperature expansion of the Wulff
1978: crystal in the 3D Ising model}, preprint (2001).
1979:
1980:
1981: \bibitem{CEP}
1982: H.~Cohn, N.~Elkies, and J.~Propp,
1983: \emph{Local statistics for random domino tilings of
1984: the Aztec diamond},
1985: Duke Math.\ J.\ \textbf{85} (1996), no.\ 1, 117--166.
1986:
1987: \bibitem{CKP}
1988: H.~Cohn, R.~Kenyon, and J.~Propp,
1989: \emph{A variational principle for domino tilings},
1990: math.CO/0008220.
1991:
1992: \bibitem{FK}
1993: L.~Faddeev and R.~Kashaev,
1994: \emph{Quantum dilogarithm},
1995: hep-th/9310070.
1996:
1997: \bibitem{FS}
1998: P.~Ferrari and H.~Spohn,
1999: \emph{Step fluctuations for a faceted crystal},
2000: cond-mat/0212456.
2001:
2002: \bibitem{J1}
2003: K.~Johansson,
2004: \emph{Discrete
2005: orthogonal polynomials and the Plancherel measure},
2006: math.CO/9906120.
2007:
2008: \bibitem{J2}
2009: K.~Johansson,
2010: \emph{Universality of the local spacing distribution in
2011: certain ensembles of Hermitian Wigner matrices},
2012: math.ph/0006020.
2013:
2014: \bibitem{J3}
2015: K.~Johansson,
2016: \emph{Non-intersecting paths, random tilings, and
2017: random matrices},
2018: math.PR/0011250.
2019:
2020: \bibitem{K}
2021: V.~Kac,
2022: \emph{Infinite dimensional Lie algebras},
2023: Cambridge University Press.
2024:
2025: \bibitem{KM}
2026: S.~Karlin and G.~McGregor,
2027: \emph{Coincidence probabilities}, Pacific J.\ Math
2028: \textbf{9} (1959), 1141--1164.
2029:
2030: \bibitem{K1}
2031: R.~Kenyon,
2032: \emph{Local statistics of lattice dimers},
2033: Ann.\ Inst.\ H.\ Poincar\'e, Prob.\ et Stat.\
2034: \textbf{33} (1997), 591--618.
2035:
2036: \bibitem{K2}
2037: R.~Kenyon,
2038: \emph{The planar dimer model with a boundary: a survey},
2039: to appear in Proceedings CRM.
2040:
2041: \bibitem{M}
2042: I.~G.~Macdonald,
2043: \emph{Symmetric functions and Hall
2044: polynomials},
2045: Clarendon Press, 1995.
2046:
2047: \bibitem{O}
2048: A.~Okounkov,
2049: \emph{Infinite wedge and random partitions},
2050: Selecta Math., New Ser., \textbf{7} (2001), 57--81,
2051: math.RT/9907127.
2052:
2053: \bibitem{O2}
2054: A.~Okounkov,
2055: \emph{Symmetric functions and random partitions},
2056: Symmetric functions 2001: Surveys of Developments and
2057: Perspectives, edited by S.~Fomin, Kluwer Academic
2058: Publishers, 2002.
2059:
2060: \bibitem{PS}
2061: M.~Praehofer and H.~Spohn,
2062: \emph{Scale Invariance of the PNG Droplet and the Airy Process},
2063: math.PR/0105240.
2064:
2065: \bibitem{V}
2066: A.~Vershik,
2067: talk at the 1997 conference on Formal Power Series and Algebraic
2068: Combinatorics, Vienna.
2069:
2070:
2071: \end{thebibliography}
2072:
2073:
2074: \end{document}
2075:
2076: Schur process is a time-dependent analog of the
2077: Schur measure on partitions studied in math.RT/9907127.
2078: Our first result is that the correlation functions of
2079: the Schur process are determinants with a kernel that
2080: has a nice contour integral representation in terms of
2081: the parameters of the process. This general result is
2082: then applied to a particular specialization of the
2083: Schur process, namely to random 3-dimensional Young
2084: diagrams. The local geometry of a large random 3-dimensional diagram
2085: is described in terms of a determinantal point process
2086: on a 2-dimensional lattice with the incomplete beta function kernel
2087: (which generalizes the discrete sine kernel).
2088: A brief discussion of the universality of this answer
2089: concludes the paper.
2090:
2091: