Probability, Geometry and Integrable Systems MSRI Publications Volume 55, 2008
Nonintersecting Brownian motions, integrable systems and orthogonal polynomials
PIERRE VAN MOERBEKE
To Henry, teacher and friend, with admiration and gratitude
ABSTRACT. Consider n nonintersecting Brownian motions on ޒ, which leave from p definite points and are forced to end up at q points at time t 1. When n , the equilibrium measure for these Brownian particles hasD its support on!p 1intervals, for t 0, and on q intervals, for t 1. Hence it is clear that, when t evolves, intervals must merge, must disappear and be created, leading to various phase transitions between times t 0 and 1. Near these moments of phase transitions,D there appears an infinite-dimen- sional diffusion, a Markov cloud, in the limit n , which one expects to depend only on the nature of the singularity associated% 1 with this phase change. The transition probabilities for these Markov clouds satisfy nonlinear PDE’s, which are obtained from taking limits of the Brownian motion model with finite particles; the finite model is closely related to Hermitian matrix inte- grals, which themselves satisfy nonlinear PDE’s. The latter are obtained from investigating the connection between the Karlin-McGregor formula, moment matrices, the theory of orthogonal polynomials and the associated integrable systems. Various special cases are provided to illustrate these general ideas. This is based on work by Adler and van Moerbeke.
Mathematics Subject Classification: Primary: 60G60, 60G65, 35Q53; secondary: 60G10, 35Q58. Keywords: Dyson’s Brownian motion, Pearcey process, random matrices coupled in a chain, random matrices with external potential, infinite-dimensional diffusions. This work was done while PvM was a Miller visiting Professor at the University of California, Berkeley. The support of National Science Foundation grant #DMS-04-06287, a European Science Foundation grant (MISGAM), a Marie Curie Grant (ENIGMA), FNRS, Francqui Foundation and IUAP grants is gratefully acknowledged.
373 374 PIERREVANMOERBEKE
1. Introduction
This lecture in honor of Henry McKean forms a step in the direction of un- derstanding the behavior of nonintersecting Brownian motions on ޒ (Dyson’s Brownian motions), when the number of particles tends to . It explains a 1 novel interface between diffusion theory, integrable systems and the theory of orthogonal polynomials. These subjects have been at the center of Henry McK- ean’s oeuvre. I am delighted to dedicate this paper to Henry, teacher and friend, with admiration for his pioneering work in these fields. Consider n Brownian particles leaving from points a1 < < ap and forced to end up at b1 < < bq at time t 1. It is clear that, when n , the D ! 1 equilibrium measure for t 0 has its support on p intervals and for t 1 on q intervals. It is also clear that, when t evolves, intervals must merge, must disappear and be created, leading to various phase transitions, depending on the respective fraction of particles leaving from the points ai and arriving at the points bj . Therefore the region R in the space-time strip .x; t/ formed by the support . ޒ/ of the equilibrium measure as a function of time 0 t 1 will typically present singularities of different types. Near the moments, where a phase transition takes place, one would expect to find in the limit n an infinite-dimensional diffusion, a Markov cloud, % 1 having some universality properties. Universality here means that the infinite- dimensional diffusion is to depend on the type of singularity only. These Markov clouds are infinite-dimensional diffusions, which ‘in principle’ could be de- scribed by an infinite-dimensional Laplacian with a drift term. We conjecture that each of the Markov clouds obtained in this fashion is related to some in- tegrable system, which enables one to derive a nonlinear (finite-dimensional) PDE, satisfied by the joint probabilities. The purpose of this lecture is to show the intimate relationship between these subjects: nonintersecting Brownian mo- tions and integrable systems, via the theory of orthogonal polynomials. Special cases have also shown an intimate connection between the integrable system and the Riemann-Hilbert problem associated with the singularity. These ideas will then be applied to a simple model, where we show that the transition probabilities for the infinite-dimensional Brownian motions near a cusp satisfy a nonlinear PDE. The interrelations between all such equations, “initial” and “final” (t ) conditions, are interesting and challenging open problems. ! ˙1 Universality in this context is a largely open field. For references, see later. BROWNIAN MOTIONS, INTEGRABLE SYSTEMS AND ORTHOGONAL POLYNOMIALS 375
2. Biorthogonal polynomials and the 2-component KP hierarchy
Consider the inner product for the weight .x; y/ on ޒ2,
f g f.x/g.y/.x; y/dxdy: h j i WD ޒ2 ZZ and an inner product for this weight, augmented with an extra-exponential factor, depending on “time” parameters t .t1; t2;:::/ and s .s1; s2;:::/, WD WD 1 i i 1 .ti y si x / f g t;s f.x/g.y/.x; y/e dxdy: h j i WD ޒ2 ZZ P .1/ .2/ Construct monic biorthogonal polynomials pm .y/ and pn .x/ (also depending on the parameters t and s) with regard to this deformed weight,
1 i 1 i .2/ 1 si x .1/ 1 ti y pn .x/e pm .y/e P P D ˇ E 1 i i ˇ .2/ .1/ 1 .ti y si x / pn .x/pm .y/.x; y/e dx dy ˇ D ޒ2 ZZ P ınmhn; D and let n be the determinant of the moment matrix
1 i 1 i k 1 si x ` 1 ti y n.t; s/ det x e y e : WD P P 0 k;` n 1 D ˇ E ˇ The following theorem and its corollary,ˇ due to Adler and van Moerbeke [1997; 1999b] and inspired by Sato’s theory, establishes a link between the functions n and the biorthogonal polynomials:
THEOREM 2.1. Given these data, the determinant n.t; s/ and the biorthogonal polynomials are related by the following relations, where we have set Œ˛ WD .˛; 1 ˛2; 1 ˛3;:::/ for ˛ ރ: 2 3 2 1 n n.t Œz ; s/ .1/ z pn .z/; n.t; s/ D 1 n n.t; s Œz / .2/ z C pn .z/; n.t; s/ D 1 .2/ .t Œz ; s/ p .x/ 1 i i n 1 n 1 n 1 .ti y si x / z C C e .x; y/ dx dy; n.t; s/ D ޒ2 z y ZZ P 1 .1/ .t; s Œz / p .y/ 1 i i n 1 n 1 n 1 .ti y si x / z C e .x; y/ dx dy; (2-1) n.t; s/ D ޒ2 z x ZZ P 376 PIERREVANMOERBEKE with the n.t; s/ satisfying bilinear equations, for all integers n; m 0 and all t; t ; s; s ރ : 0 0 2 1 1 0 i 1 1 1 .ti t /z n m 2 n 1.t Œz ; s/m 1.t 0 Œz ; s0/e i z dz z C C P I D1 1 0 i 1 1 1 .si s /z m n n.t; s Œz /m.t 0; s0 Œz /e i z dz: D z C P I D1 Two-component KP hierarchy. Define the Hirota symbol between functions f f.t1; t2;:::/ and g g.t1; t2;:::/ by D D @ @ @ @ p ; ;::: f g p ; ;::: f.t y/g.t y/ : @t1 @t2 ı WD @y1 @y2 C y 0 ˇ D ˇ 1 t zi ˇ i The elementary Schur polynomials S are defined by e 1 i Si.t/z ` WD i 0 for ` 0 and S .t/ 0 for ` < 0; moreover, set for laterP use ` D P @ 1 @ 1 @ S`.@t / S` ; ; ; ::: : Q WD @t1 2 @t2 3 @t3
Finally, recall that the Wronskian f; g x of f and g is given by f g @f @g g.x/ f.x/: @x @x
COROLLARY. From Theorem 2.1, one deduces the equations
2 S @ 1 @ 2 @ j ; ;::: n 1 n 1 n log n; @t1 2 @t2 C ı D @s1@tj 1 C (2-2) 2 S @ 1 @ 2 @ j ; ;::: n 1 n 1 n log n; @s1 2 @s2 ı C D @t1@sj 1 C and finally a single partial differential equation for n in terms of Wronskians,
@2 log @2 log @2 log @2 log n ; n n ; n 0: (2-3) @t1@s2 @t1@s1 C @s1@t2 @t1@s1 D ( )t1 ( )s1
SKETCH OF PROOF OF THEOREM 2.1 AND ITS COROLLARY. The following double integral can be expanded in two different ways with regard to the param- eters a .a1; a2;:::/: WD BROWNIAN MOTIONS, INTEGRABLE SYSTEMS AND ORTHOGONAL POLYNOMIALS 377
n.t;s/n 1.t 0;s0/ C .2/ 1 i 0 i .1/ 1 .ti y si x / dx dy pn 1.t 0; s0 x/pn .t; s y/e .x;y/ ޒ2 I I t t a ZZ C P ˇ t‘0 t a ˇ 0‘ C ˇ s s ˇ D 1 @ 1 @ 1 @ 2 2aj 1Sj ; ; ;::: n 2 n O.a / D C @t1 2 @t2 3 @t3 C ı C j 0 XD 2 1 2 @ 2 2aj 1n 1 log n 1 O.a / ; (2-4) D C C @s1@tj 1 C C j 0 C XD using the fact that the space H span zi; i ޚ can be equipped with two WD f 2 g (formal) inner products:
(i) f; g f.z/g.z/ dz; h iD ޒ Z i H (ii) a residue pairing about z between f i 0 aiz C and h j 1 H D 1 D 2 D j ޚ bj z : 2 2 P P dz f; h f.z/h.z/ aibi: h i1 D z 2i D I D1 i 0 X The two inner products are related by g.u/ f; g f.z/g.z/ dz f; du : h iD ޒ D ޒ z u Z Z 1 Then the two expansions (2-4) are obtained, using the -function representa- tion (2-1) of the biorthogonal polynomials, transforming the double integral (2-4) into a contour integral about and finally computing the residues. Upon 1 equating the two series in (2-4) for arbitrary aj , one finds the first identity (2-2). Application of a similar shift s s a, s s a, t t yields the second ‘ 0 ‘ C 0 D identity (2-2). Then combining the identities (2-2) for j 0 and 1 leads to the D PDE (2-3). ˜
3. Orthogonal polynomials with regard to several weights and the n-component KP hierarchy
Now considering two sets of weights,
1;:::; q and '1;:::;'p; and deform each weight with its own set of times:
1 i 1 i s 1 ski x t 1 tki y k .x/ k .x/e and 'k .y/ 'k .y/e ; WD P WD P 378 PIERREVANMOERBEKE the time parameters being
s .s ; s ;:::/ for 1 k q and t .t ; t ;:::/ for 1 k p: k D k1 k2 k D k1 k2
Take a moment matrix consisting of p q blocks of sizes mi nj , formed of moments with regard to all the different combinations of i and 'j ’s; of course for the full matrix to be a square matrix, the integers m1; m2;::: 0 and q p n1; n2;::: 0 must satisfy mi ni. Define the determinant mn of 1 D 1 these moment matrices (the inner product is the same as in Section 2): P P
m1;:::;mq n1;:::;np .s1;:::; sq t1;:::; tp/ I I WD k s1 ` t1 k s1 ` tp x 1 .x/ y '1 .y/ 0k