nlopt icon indicating copy to clipboard operation
nlopt copied to clipboard

Published 20 hours ago • verified publisher icon stevengj

MMA generates NaN

Open HGuillemet opened this issue 6 years ago • 3 comments

Hello,

In some iterations of an optimization problem, MMA feeds my objective function with some x values being NaN. Could this be a "normal behavior" or shall we track a bug ?

Version is 2.6.1, no constraints but lower and upper bounds, stopping criteria is ftol_re (1e-12).

HGuillemet avatar May 13 '19 22:05 HGuillemet

Precision: my objective function doesn't generate NaN, neither as a function value nor as a gradient value.

HGuillemet avatar May 14 '19 10:05 HGuillemet

It turns out that the rho internal parameter becomes larger and larger, until it generates some infinite numbers. I'm tempted to add a stop criterium on the rho value, but I'm not sure if this makes sense.

HGuillemet avatar May 18 '19 16:05 HGuillemet

I'm seeing the same behavior. @stevengj any thoughts on this?

I'll try and provide a MWE that quickly reproduces the issue.

smartalecH avatar Apr 17 '20 13:04 smartalecH