Force-matching pair potentials

Automatic force field optimization. Given a force field and a set of reference data (e.g. energies and forces from QM calculations, experimental measurements), we tune the force field parameters such that it accurately reproduces the reference data.
POST REPLY
User avatar
Lee-Ping Wang
Posts: 102
Joined: Sun Jun 19, 2011 5:14 pm

Force-matching pair potentials

Post by Lee-Ping Wang » Tue Mar 18, 2014 1:23 am

Continued from https://simtk.org/forums/viewtopic.php? ... 7&start=10 .
ponadto wrote:OK, ForceBalance works like a charm, but at the moment I only had time to check the 001_water_tutorial.

I have a different project and I was wondering if I could ForceBalance for a quite different purpose than the one we were discussing so far.
I'll ask a few questions without thoroughly reading the doc:
1. Can I use ForceBalance to parameterize pair interactions between (any) two groups of atoms?
Yes, this should be possible. ForceBalance works by passing force field files to other MD codes (e.g. OpenMM or Gromacs) and reading their outputs, so as long as you can create a force field file to give pair interactions, you should be in business.
ponadto wrote:2. My system is quite large (~1000 atoms), and the two groups of atoms I'm considering both consist of 5 amino-acid residues. Also, my full atom trajectory is long (say, 10^6 snapshots). Would that be palpable for ForceBalance?

Have I overinterpreted the features of ForceBalance, or am I still in the scope of potential usage?
Your system is indeed larger than the ones I've tackled. I don't think it's outside the scope of ForceBalance though I haven't done a calculation of this size before. I think the main limitations will come from array size, as ForceBalance stores a number of large arrays in memory. Some arrays are size (3*N_atoms*N_parameters*N_snapshots), others are (9*N_atoms^2), and others are (3*N_atoms*N*parameters*N_parameters). Some of these large arrays are allocated in order to speed up the calculation in other ways, but they aren't 100% necessary. Thus, I would suggest first trying with a small number of snapshots and then increasing it.

Also, ForceBalance is intended to be usable for large systems, so if you do run into problems with system size I would be happy to improve the code to address these issues.

Let me know if this is something you're interested in, I think it would be a medium-sized project in itself.

POST REPLY