I've run into this problem often when dealing with log probabilities in maximum likelihood estimation where the optimizer can try some numbers that are way off

I've run into this problem often when dealing with log probabilities in maximum likelihood estimation where the optimizer can try some numbers that are way off from the optimal value and you still want to return an accurate number. There is an package in R for handling these operations automatically [1]. [1] http://cran.r-project.org/web/packages/Brobdingnag/index.html http://cran.r-project.org/web/packages/Brobdingnag/index.htm...