A proof of Parisi’s conjecture on the random assignment problem, Probab. Theory Relat. Fie.pdf

a r X i v : m a t h / 0 3 0 3 2 1 4 v 1 [ m a t h .C O ] 1 8 M a r 2 0 0 3 A PROOF OF PARISI’S CONJECTURE ON THE RANDOM ASSIGNMENT PROBLEM SVANTE LINUSSON AND JOHAN WA?STLUND Abstract. An assignment problem is the optimization problem of find- ing, in an m by n matrix of nonnegative real numbers, k entries, no two in the same row or column, such that their sum is minimal. Such an optimization problem is called a random assignment problem if the ma- trix entries are random variables. We give a formula for the expected value of the optimal k-assignment in a matrix where some of the en- tries are zero, and all other entries are independent exponentially dis- tributed random variables with mean 1. Thereby we prove the formula 1+1/4+1/9+· · ·+1/k2 conjectured by G. Parisi for the case k = m = n, and the generalized conjecture of D. Coppersmith and G. B. Sorkin for arbitrary k, m and n. 1. Introduction The problem of minimizing the sum of k elements in a matrix of non- negative real numbers under the condition that no two of them may be in the same row or column is called an assignment problem. A set of matrix positions no two in the same row or column is called an independent set. An independent set of k matrix positions will also be called a k-assignment. A random assignment problem, or RAP for short, is given by a number k, and an m by n matrix (min(m,n) ≥ k) of random variables. If P is a random assignment problem, we denote by E(P ) the expected value of the minimal sum of an independent set of k matrix elements. In this article we prove the following. Theorem 1.1 (Parisi’s Conjecture [P98]). Let P be the RAP where k = m = n and the matrix entries are independent exponential random variables with intensity 1. Then E(P ) = 1 + 1 4 + 1 9 + · · ·+ 1 k2 . We also prove the following two generalizations. Theorem 1.2 (Conjectured by D. Coppersmith and G. B. Sorkin [CS98]). Let P be an RAP where the matrix entries are independent exponential ran- dom variables with