different energy after constant hard-coding in custom force

[lslumass]

Hi,

I'm running simulation starting form CHARMM-style force field, and trying to add an custom nonbonded force which is

f1 = f"""(ke/er*charge1*charge2)/r*exp(-r*kf)"""
or
f2 = f"""(ke/20*charge1*charge2)/r*exp(-r*kf)"""

, a coulomb interaction (er is 20.0 as the relative dielectric constant). For both f1 and f2, I got the same initial potential energy (1434.24 kj/mol). But after minimization I got -273.9 kj/mol for f1 and -243.3 kj/mol for f2. I'm confused why they are different. any suggestions?

BTW, starting from CHARMM force field, if NBFIX was set, I found that all the epsilon in default nonbonded force was set as 0. Instead, all the interactions were defined in a new CustomNonbondedForce. Is that correct?

Thank you.

[peastman]

The energy minimizer is not deterministic. It finds a local minimum, but it doesn't guarantee which one it will find. If you call it several time from the same starting conformation, it may find different minima.

I found that all the epsilon in default nonbonded force was set as 0. Instead, all the interactions were defined in a new CustomNonbondedForce. Is that correct?

Correct. NonbondedForce doesn't support NBFIX, so it uses a CustomNonbondedForce to calculate the LJ interactions.

[lslumass]

Agree, it should be. That's why I was confused. My test system was so simple that I got identical energies after minimization for several times. I hve repeated the minimization five times, and every time I got -273.9 kj/mol and -243.3 kj/mol for f1 and f2, respectively.

[peastman]

Try taking the two different ending conformations and evaluate their energies with the two versions of the force. Do they agree?