Closed Lisa-Phan closed 2 weeks ago
So regarding convergence, the optimizer by default has a max-iterations setting of 250. Can be changed like this: Optimizer(..., maxiter=500). Btw, for QM/MM optimizations of a protein like this, you may need a lot iterations initially because all coordinates are in bad positions coming from either X-ray structure of a classical MD job. This particularly applies when you have a large active region, e.g. 1000 atoms.
Your broken-symmetry job is almost correctly setup, there is just one detail that applies to the case of broken-symmetry singlets. When you tell ASH to request a broken-symmetry job, it creates an ORCA inputfile containing that input for an open-shell broken-symmetry Flipspin procedure and ORCA will then converge to the open-shell singlet, creating a GBW file containing the orbitals for the broken-symmetry job. This happens in Optimization number 1. In the next optimization step, ASH has turned off the broken-symmetry procedure because there is no good reason to go through the Flipspin procedure again (unnecessarily timeconsuming). Instead ASH creates an ORCA inputfile for a singlet job in the next step and tries to read the GBW file from the previous step; but what goes wrong here is that ORCA by default, for singlets (multiplicity 1), tries to converge a restrickted (RKS), closed-shell electronic structure, instead of unrestricted, UKS. This means that while it reads the broken-symmetry orbitals it interprets them as closed-shell and converges to a useless closed-shell solution. For som reason things go wrong from there.
The simple solution to this is to add "UKS" keyword to your input_line, i.e. input_line = "! BP86 def2-SVP tightscf UKS".
I have updated the documentation, warning about this. I will try to add an option to ORCATheory so that this is automatically performed if the user forgets the UKS keyword.
I think this should solve the problem.
Thank you Ragnar for the detailed explanation!
I believe this is the second time I've made this mistake... My apologies for not checking the set up more closely.
Good evening,
I have a few questions about ASH's default convergence choice when running QM/MM. I recall ORCATheory has an option for convergence threshold parameters, but the overall default behavior for QM/MM is that the run stops after 250 iterations?
Additionally, I'm encountering a strange behavior for broken symmetry QM/MM optimization and was wondering if it was due to user error. The system is the same as one mentioned the previous thread. Compared to the singlet or highspin treatment, which converges stably, broken symmetry system seems much less stable. The xyz shows expansion of the QM region, along with energy initially lowering then rising in a parabolic pattern.
Comparing the orca_last.out file for the two jobs, only the highspin run has both mulliken atomic and spin population printed whereas the broken symmetry run only has atomic population. Does the slight reduction in mulliken atomic population suggested incorrect set up?
My apologies for the rough graphs.
Attached beneath is the script I'd used for the broken symmetry run, which used a partially optimized .pdb as the starting point. This is because using the original unoptimized structure to start causes things to not converge.
and this is the script used for the full high-spin treatment
For the broken symmetry run
For the all-highspin run
I'd really appreciate any pointers on how to troubleshoot the run.
Thank you, Lisa Phan