Brownian Gibbs property for Airy line ensembles

Abstract

Consider N Brownian bridges B_i:[-N,N] -> R, B_i(-N) = B_i(N) = 0, 1 <= i <= N, conditioned not to intersect. The edge-scaling limit of this system is obtained by taking a limit as N -> infinity of these curves scaled around (0,2^{1/2} N) horizontally by a factor of N^{2/3} and vertically by N^{1/3}. If a parabola is added to each limit curve, an x-translation invariant process sometimes called the multi-line Airy process is obtained. We prove the existence of a version of this process (which we call the Airy line ensemble) in which curves are a.s. everywhere continuous and non-intersecting. This process naturally arises in the study of growth processes and random matrix ensembles, as do related processes with "wanderers" and "outliers". We formulate our results to treat these relatives as well. Note that the law of the finite collection of Brownian bridges above has the property -- called the Brownian Gibbs property -- of being invariant under the following action. Select an index 1 <= k <= N and erase B_k on a fixed time interval (a,b) subset of (-N,N); then replace this erased curve with a new curve on (a,b) according to the law of a Brownian bridge between the two existing endpoints (a,B_k(a)) and (b,B_k(b)), conditioned to intersect neither the curve above nor the one below. We show that this property is preserved under the edge-scaling limit and thus establish that the Airy line ensemble has the Brownian Gibbs property. An immediate consequence is a proof of M. Prahofer and H. Spohn's prediction that the lines of the Airy line ensemble are locally absolutely continuous with respect to Brownian motion. We also prove the conjecture of K. Johansson that the top line of the Airy line ensemble minus a parabola attains its maximum at a unique point, thus establishing the asymptotic law of the transversal fluctuation of last passage percolation with geometric weights.