I have very little experience programming, so I hope that someone

who actually knows anything about C will take the program

and do a better job.

The main problem in implementing the algorithm is dealing with

"large states." A state M is large if it contains many strategy profiles.

Large states will slow down the algorithm considerably. In my program

I kept down the size of the states by discarding, in each iteration, all but the

minimal elements of M. This makes the algorithm search strategy profiles

unnecessarily, but the gain in keeping M small is large enough to justify

this waste.

My program is for two-player games. It is not difficult to do it for n-player

games, but I made a bad choice when I started writing the program

(I had one pointer for each player, instead of using an array), and

then I was locked in.

One could try to implement the algorithm using MATLAB, Gauss,

or something like that. Its probably slower, I haven't tried.

by the simulations that I report in the paper.

The C-programs I used for doing the simulations are trivial.c and simulate.c.

Program trivial.c simulates games and finds all equilibria using the trivial

algorithm; simulate.c simulates games and finds all equilibria using the

algorithm I propose in the paper. Note that the seed is set in the programs

so that both algorithms work on the same set of games---this is crucial

for making the comparison between the algorithms, and to check that

my algorithm is actually working!

Please send any feedback to fede@hss.caltech.edu