NUMERICAL EXPERIMENTS WITH AN INFINITE ALGORITHM FOR SOLVING A MATRIX GAME
Abstract
This paper is concerned with an algorithm of von Neumann's for solving a matrix game. Numerical experiments are described which deal with some basic smoothing techniques designed to improve the algorithm by increasing its rate of convergence. Certain additional calculations are described which deal with the deletion of obvious dominance in the pay-off matrix as a possible aid in further increasing the rate of convergence. Some numerical comparisons are made between this algorithm and three others from the literature. In general, the performance of von Neumann's algorithm is quite inferior to other methods for solving matrix games: Dantzig's Simplex Method converges in fewer iterations and Brown's Fictitious Play requires a much simpler arithmetic process and fewer iterations for a similar convergence. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 27, 1962
- Accession Number
- AD0292691
Entities
People
- Martin Hershkowitz
Organizations
- George Washington University