Exact expectations and distributions for the random assignment problem

by

Sven Erick Alm and Gregory B. Sorkin

Uppsala University and IBM T.J. Watson Research Center

U.U.D.M. Report 1999:27 ISSN 1101-3591
In Combinatorics, Probablity and Computing 11 (2202), 217-248.

Postscript

Abstract

A generalization of the random assignment problem asks the expected cost of the minimum-cost matching of cardinality $k$ in a complete bipartite graph $\Kmn$, with independent random edge weights. With weights drawn from the exponential(1) distribution, the answer has been conjectured to be $\sum_{ i,j \ge 0 , \ i+j < k } \frac{1}{(m-i)(n-j)}$.
Here, we prove the conjecture for $k \leq 4$, $k=m=5$, and $k=m=n=6$, using a structured, automated proof technique that results in proofs with relatively few cases. The method yields not only the minimum assignment cost's expectation but the Laplace transform of its distribution as well. From the Laplace transform we compute the variance in these cases, and conjecture that with $k=m=n \ra \infty$, the variance is $2/n + O(\log n / n^2)$. We also include some asymptotic properties of the expectation and variance when $k$ is fixed.

2002-07-05, Sven Erick Alm, sea@math.uu.se