Feature #1966

Hydrogen mass repartiniong

Added by Flaviyan Jerome Irudayanathan over 4 years ago. Updated over 4 years ago.

preprocessing (pdb2gmx,grompp)
Target version:


Dear Developers,

Just a suggestion here for the integration of Hydrogen mass re-partitioning(HMR) as a feature in PDB2GMX with compatible force fields; initially reported in

K. Anton Feenstra, Berk Hess, and Herman J. C. Berendsen, “Improving efficiency of large timescale molecular dynamics simulations of hydrogen-rich systems,”
Journal of Computational Chemistry, vol. 20, 1999, pp. 786–798.

Which was then picked up recently whit some modifications in the following for AMBER.

Long-Time-Step Molecular Dynamics through Hydrogen Mass Repartitioning
Chad W. Hopkins, Scott Le Grand, Ross C. Walker, and Adrian E. Roitberg
Journal of Chemical Theory and Computation 2015 11 (4), 1864-1874
DOI: 10.1021/ct5010406

I am aware that there is already a implementation of this in pdb2gmx, but the new modifications uses a 3 amu mass scaling so there would be some changes in the code level as how this is handled.


#1 Updated by Erik Lindahl over 4 years ago

Gromacs already does this in a far better way than mere hydrogen mass repartitioning.

While there might not be obvious effects on protein structure, as already shown in the original Feenstra article there will be differences in kinetic properties (although one could argue they are not too extreme).
However, in Gromacs I would strongly recommend you to simply use virtual interaction sites for hydrogens instead: This allows you to go all the way to 5fs timesteps, with much smaller artefacts than hydrogen mass repartitioning.

#2 Updated by Flaviyan Jerome Irudayanathan over 4 years ago

I looked into the -viste option, Thanks for the suggestions.
I guess this ticket can be closed then.

#3 Updated by Roland Schulz over 4 years ago

The vsite option is probably what is best in most cases. But for some hardware the constraints required for vsites are expansive making that option not very efficient. In those cases HMR might actually be more efficient and the better option if the larger artifact is OK for the use case.

#4 Updated by Berk Hess over 4 years ago

  • Priority changed from Normal to Low

A difference of sqrt(3/4) is dynamic properties is very small, so I would suggest to just use the default factor 4 that GROMACS has always used.

#5 Updated by Szilárd Páll over 4 years ago

HMR does have performance benefits wrt vsites (especially related to scaling, right?), so better documenting its support could be useful (just searching for "hydrogen mass repartitioning" in the manual did not return any results). If nothing else people may want to repro AMBER results with GROMACS and they'd likely appreciate being able to do that easily. Are there any simple changes (other than docs) that would facilitate its use HMR in GROMACS?

Also available in: Atom PDF