btrump1
I'm attempting to use Simulated Annealing to solve a structure with several rigid bodies. On newer versions of Topas it will often spit out the message: Breaking - Randomize on Errors, maybe 20-30 times in between each refinement.
I understand this is meant to correct the temperature [I typically use Auto_T(10)]; however, comparing this to older versions it seems to take longer to find a minimum and appears to get stuck on the same bad guess for longer. In Topas 5 I was able to work around it by refining an dummy atom (atom with occ=0), however this doesn't work in Topas 6. I've noticed thus far that I can get a statistically better fit in half the iterations in v5 versus v6, and I'm wondering if this somehow has something to do with it.
Any help in this matter would be significantly appreciated!
alancoelho
>Breaking - Randomize on Errors, maybe 20-30 times in between each refinement.
20 to 30 times seems excessive. Probably due to how the INP file is setup; If you send me the INP and data file I will have a look.
cheers
alan
alancoelho
Hi Ben
Thanks for sending the INP and data file.
This is a bug introduced in Version 5 and persisted into V6; thank you for picking this up.
It occurs when the following is seen:
prm !o1 0
prm o2 0
site O occ O = o1 o2; beq @ 1
Here, occ is zero and not refined due to !o1=0. The error occurs when beq is refined on a site that has a zero occ. This is rare as beq is typically not refined on a site that has zero occ. Of course it is still important to fix. The bug was introduced due to optimization.
I have fixed the problem and will get a version 6 out to all version 6 users. It may take a few days.
Once again thanks for picking this up.
cheers
alan