Poor convergence using Fe HMOs for FMO in Gamess

Click For Summary

Discussion Overview

The discussion centers on the challenges of achieving convergence in calculations involving a heme group and ligands using Fragment Molecular Orbital (FMO) methods in the GAMESS quantum chemistry software. Participants explore various strategies to improve convergence, particularly in the context of oscillating values during calculations.

Discussion Character

  • Technical explanation
  • Debate/contested
  • Mathematical reasoning

Main Points Raised

  • One participant describes the use of HMOs for the FMOBND section in GAMESS, noting that while monomer calculations converge, the values oscillate between two points, preventing progression to the dimer stage.
  • Another participant suggests that the oscillation might indicate that the ground state and the first excited state are close in energy, complicating the convergence process.
  • There is a proposal to improve convergence by using a better initial guess for orbital occupation numbers, particularly if the ground state is known.
  • DIIS damping is mentioned as a potential method to aid convergence, with a suggestion to turn it off completely if necessary, despite the longer computation time.
  • Participants express uncertainty about their expertise in solving SCF convergence problems, describing it as somewhat of an art.
  • One participant acknowledges their limited experience with GAMESS compared to other quantum chemistry programs, which may affect their ability to troubleshoot effectively.

Areas of Agreement / Disagreement

Participants express a shared understanding of the difficulties associated with SCF convergence problems, but there is no consensus on the best approach to resolve the oscillation issue. Multiple strategies are proposed, reflecting differing levels of familiarity with GAMESS and computational techniques.

Contextual Notes

Participants note limitations in their experience with GAMESS and SCF convergence, which may affect the applicability of their suggestions. The discussion also highlights the complexity of distinguishing between ground and excited states in the context of convergence.

Who May Find This Useful

This discussion may be useful for researchers and practitioners working with GAMESS or similar quantum chemistry software, particularly those facing challenges with SCF convergence in complex molecular systems.

WillCawdell
Messages
2
Reaction score
2
Hello!

I'm trying to model an interaction between ligands and a heme group using FMO in Gamess.

I've tried to make HMOs for the FMOBND section of the input file (using an Fe-F complex), the HMOs are shown below:
STO-3G 19 5
0 1 0.992624, 0.019366, 0.000000, 0.000000,-0.000001,-0.014838, 0.000000,
0.000000, 0.000004, 0.005535, 0.005534, 0.005534, 0.000000, 0.000000,
0.000000,-0.000595, 0.000000, 0.000000,-0.000029
1 0 0.420492,-1.033033, 0.000000, 0.000000,-0.000111,-0.314157, 0.000000,
0.000000, 0.000062, 0.110327, 0.111004, 0.109981,-0.000339, 0.000000,
0.000000,-0.010802, 0.000000, 0.000000,-0.000514
0 1 -0.000040, 0.000094, 0.000000, 0.000000,-0.990859, 0.000170, 0.000000,
0.000000,-0.026700,-0.000085,-0.000086,-0.000094, 0.000001, 0.000000,
0.000000, 0.000139, 0.000000, 0.000000, 0.002284
0 1 0.000000, 0.000000, 0.928325, 0.346200, 0.000000, 0.000000, 0.025225,
0.009407, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000, 0.000008,
0.000003, 0.000000,-0.001955,-0.000729, 0.000000
0 1 0.000000, 0.000000,-0.346473, 0.929057, 0.000000, 0.000000,-0.008635,
0.023155, 0.000000, 0.000000, 0.000000, 0.000000, 0.000000,-0.000003,
0.000009, 0.000000, 0.000689,-0.001848, 0.000000

These HMOs work to a certain extent, allowing monomer calculations on the heme group, cysteine donor and my ligand donor to converge, however the value they converge to is oscillating between 2 points thus never converging according to Gamess. So the calculation cannot proceed to the dimer stage.

Does anyone have any ideas how to make my convergence better? U currently use DIIS as a start an initial convergence mode then switch to SOSCF to fine tune the convergence value

Thannks a lot in advance!
 
Physics news on Phys.org
WillCawdell said:
These HMOs work to a certain extent, allowing monomer calculations on the heme group, cysteine donor and my ligand donor to converge, however the value they converge to is oscillating between 2 points thus never converging according to Gamess. So the calculation cannot proceed to the dimer stage.

Does anyone have any ideas how to make my convergence better? U currently use DIIS as a start an initial convergence mode then switch to SOSCF to fine tune the convergence value
Caveat 1: I've never been great at solving SCF convergence problems, and I'm not sure anyone actually is. It's a bit of an art.
Caveat 2: I never really used GAMESS as extensively as some other QC programs, so I'm not as familiar with its syntax and its capabilities.

If your convergence value is oscillating between two points, it might be the case that the ground state and the first excited state are fairly close in energy, and the program can't decide which one is which. If you know which one is your ground state, and if its orbital occupation numbers are significantly different from the first excited state, you might be able to resolve the problem with a better initial guess. Absent that, you could try DIIS damping or turn off DIIS completely. This will take longer but might help convergence. Good luck.
 
Reply
  • Informative
Likes   Reactions: berkeman
TeethWhitener said:
Caveat 1: I've never been great at solving SCF convergence problems, and I'm not sure anyone actually is. It's a bit of an art.
Caveat 2: I never really used GAMESS as extensively as some other QC programs, so I'm not as familiar with its syntax and its capabilities.

If your convergence value is oscillating between two points, it might be the case that the ground state and the first excited state are fairly close in energy, and the program can't decide which one is which. If you know which one is your ground state, and if its orbital occupation numbers are significantly different from the first excited state, you might be able to resolve the problem with a better initial guess. Absent that, you could try DIIS damping or turn off DIIS completely. This will take longer but might help convergence. Good luck.
Thanks a lot for your help! I agree convergence is a black art! I appreciate what you're saying about the ground and excited states I should have thought of that sooner! I think I'll try the DIIS changes you suggested! Thanks!
 
Reply
  • Like
Likes   Reactions: jim mcnamara and TeethWhitener
I'll keep my fingers crossed for you. If that doesn't work, it's not a bad idea to poke around google or some of the QC program forums for advice on convergence problems.