forum.alglib.net

Hi,
I would like to minimize some function which depends on a very large number of variables (1 to 8 millions is the typical number) using the LBFGS method. I am wondering how to estimate the total memory occupation for the alglib library. With the original fortran code, which I used many years ago, I remember that the total memory occupation given N, the number of variables, was typically 8-10 times N. Is this the case also for minlbfgsoptimize and related routines?
Sincerely
Max

For L-BFGS, memory occupation is very roughly ~N*M*2 double precision numbers. Here N is number of variables, M is number of L-BFGS updates stored by optimizer (both parameters are specified during optimizer creation).