Closed cryos closed 8 years ago
cryos
commented
16 years ago Logged In: YES user_id=21420 Originator: NO
From Henry Rzepa: I think most people look for an energy that has bottomed out. So displaying that prominently would allay their concerns
Original comment by: @ghutchis
cryos
commented
16 years ago Logged In: YES user_id=1682851 Originator: NO
The AutoOpt tool was not really designed to minimize until convergence, it minimizes until the user clicks the stop button. This allows you to drag atoms or draw additional atoms while minimizing. However, in future version more feedback could be integrated.
Examples:
There is a convergence criteria 1e-6 (dE), but when this is reached AutoOpt just loops, waiting for the user to disrupt the energy surface again.
The steps option was added recently, before it would take 2 steps and conjugate gradients was not really better in this case. Now it would make sense to add cg back to the algorithm option.
Original comment by: @timvdm
cryos
commented
16 years ago Logged In: YES user_id=1682851 Originator: NO
I've added the energy to the output on the 3D canvas together with dE. Since this isn't really a bug I'll close this one.
Original comment by: @timvdm
On a PPC G5 Mac, I have one observation. Minimising (eg propane) using the steepest descent seems to never stop, ie Autoopt runs for ever. Might there perchance be an open implementation of eg the BFGS variable metric method? I ask because steepest descent is notorious for meandering for ever? Or is it roundoff error perchance that is preventing it from stopping? I have not yet found what the quitting criteria might be, or how close the system is to achieving them.
Certainly, in Ghemical, this step rarely takes more than a second or two.
Reported by: *anonymous