Problems encountered when using eABF method to calculate PMF in colvars module of LAMMPS

LAMMPS LAMMPS Beginners
,
1

Dear LAMMPS users
I am using LAMMPS’ Colvars module to calculate the free energy between two nanoparticles in a fluid as a function of distance, but I have obtained unacceptable results. I ran 0.5ns, 0.75ns, and 1ns under fix colvars and found significant changes in the shape and magnitude of the PMF curve. I think the change in the shape of the curve is the reason why my simulation did not converge due to insufficient running time. But I don’t think there should be such a big change in the PMF value, and I don’t know if it’s related to me not writing the colvars file correctly. I also want to know if there are any other methods to determine convergence besides periodically outputting PMF observations. The PMF curve and colvars file generated by running the output are shown below:

0.5ns
0.5ns3216×2462 52.4 KB

0.75ns
0.75ns3216×2462 45.1 KB

1ns
1ns3216×2462 41.8 KB

Colvarstrajfrequency 1000
Colvarsrestartfrequency 500000
indexFile non1.ndx
indexFile non2.ndx

colvar {
name d
width 0.20

extendedlagrangian on
extendedFluctuation 0.1
extendedTemp 1600
extendedTimeConstant 200

subtractAppliedForce on

distance {
    group1 {
        indexGroup non1
    }
    group2 {
        indexGroup non2
    }
    forceNoPBC yes
}
lowerBoundary 0.2
upperBoundary 30

}
harmonicWalls {
name mywlls
colvars d
lowerWalls 0.2
upperWalls 30
forceConstant 20
}

abf {
name abf1
colvars d
applyBias yes
fullSamples 200
outputEnergy on
writeCZARwindowFile on
CZARestimator on
}
Thank you for your help!

2

Hi @youkuenheng if I understand correctly the code is producing a physical result (the PMF) but you are skeptical of the latter.

I think you are right on this point, but my own take would be to consider that you ran for a very short time. You should find out what’s a reasonable correlation time for this system, and set fullSamples accordingly.

A 1 ns PMF computation would not be considered converged by most.

Giacomo

3

Thank you for your reply. I should try running for more than ten nanoseconds and check the reliability of the data