Optimizing MD parametrization on the basis of QM

Post date: Aug 2, 2012 3:32:49 PM

How to check if MD geometry (caused by the current parametrization) is consistent with the QM one.

Note that QM is computed in vacuo with gaussian03 and may be very slow.

MD part

  • Parametrize the fragment
  • Run a short MD run, preferably of solvated system
  • Compute (or check visually) the average structure

QM part (based on Gaussian)

module add ambertools module add gaussian antechamber -i BEN.pdb -fi pdb -o BEN.gin -fo gcrt -nc 1 -pf y # optionally add:    -gm "%mem=1000MB" -gn "%nproc=8"

edit BEN.gin and change the first line in… (not necessary?) #HF/6-31G* Test opt. Then:

g03 BEN.gin antechamber -i BEN.log -fi gout -o BEN-QM.pdb -fo pdb -nc 1 -pf y

This gives you BEN-QM.pdb, that you should compare with the structure computed in the first step.

IMPORTANT 1. In presence of symmetry, geometry can converge to a wrong minimum. In doubt, start from non-symmetric structures (e.g. add noise)!

IMPORTANT 2. If you get “Write error in NtrExt1: Bad address” or similar: gaussian only works on mio and a few other machines due to the “va address randomization”. Ask sysadmins to disable.

Runs on 4 cores by default: 30’ for BEN on mio.

If you want to improve the parametrization of the MD structure:

  • Identify the torsion angle to be changed (e.g. ca-ca-c2-n3 for BEN)
  • Check that it is used only once in the system (parameters are defined by atom type)
  • Identify the direction of the change required (e.g.: multiplicity? minimum? harmonic constant?)
  • Put the changed parameter in a xxx​.frcmod file, e.g.
      • DIHE ca-ca-c2-n3 1 1.500 000.000 4.000
  • Load the frcmod together with the others, and build the system as usual
  • Repeat the MD step until sufficient accuracy.