HEAVY_H masses in the CHARMM36m GROMACS force field

User discussions
1

Dear GROMACS developers,

I recently noticed what appears to be a potential inconsistency in the Hydrogen Mass Repartitioning (HMR) implementation of the GROMACS version of the CHARMM36 force field (downloaded from MacKerell Lab ) and would appreciate your clarification.

In older CHARMM36 GROMACS force field releases (e.g., charmm36-jul2020.ff and earlier), the HEAVY_H section of tip3p.itp assigns the following masses:

  • OT = 9.9514 amu
  • HT = 4.0323 amu

giving a total water mass of approximately 18.015 amu, as expected.

However, starting from charmm36-jul2021.ff (and continuing in subsequent releases), the HEAVY_H section instead assigns:

  • OT = 7.9356 amu
  • HT = 4.0323 amu

This results in a total water mass of only about 16.0 amu, rather than the physical value of 18.015 amu.

My understanding of HMR is that hydrogen masses are increased while the bonded heavy atom masses are decreased by the same amount, such that the total molecular mass is preserved. For TIP3P water with hydrogen masses of ~4 amu, I would therefore expect the oxygen mass to be approximately 10 amu rather than 8 amu.

I initially contacted Prof. Alex MacKerell about this, and he suggested that I raise the question here.

Could someone clarify:

  1. Was the change of the TIP3P oxygen mass from ~10 amu to ~8 amu intentional?
  2. If so, what is the rationale for not preserving the total molecular mass of water?
  3. If not, should this be considered a bug in the distributed CHARMM36 GROMACS force field?

Thank you very much for your time and clarification.

2

The OT mass is a bug. Correct it in the topology; we have already fixed the code and will fix the force field files soon.

3

Thank you for the quick clarification, Justin!

4

Note that force fields no longer need to specify different masses as we now have the mass-repartition-factor mdp option.

5

Good to know. Would this option alter the mass of both the solvent and the ligand as well? This point isn’t clearly stated in the manual.

6

The option only looks at masses and bonds, not at any other properties.

7

Thanks for the clarification, Berk.

It seems this option will also modify mass distributions inside water molecules, which could be undesirable for some cases. For example, for simulations where water viscosity is relevant, we would likely want to avoid using the mass-repartition-factor option.

A related question: If masses are explicitly modified in the force field parameters (e.g., global mass rescaling applied to an entire molecule) and the mass-repartition-factor option is enabled simultaneously, how will GROMACS handle this combination? What will the final mass values be? Also, is there a straightforward way to see these final masses?

8

I don’t understand. You scale masses to achieve a larger time step. That means you dynamics are always affected. And you need to scale the masses of all hydrogens to retain accuracy, even though hydrogens in water models are a bit less critical due to the rigid geometry.

9

Thanks, Berk. I may not have made my point clear enough. I agree that HMR inevitably affects dynamics.

For many biomolecular simulations focusing on protein conformational sampling, full-system HMR (including water) is a reasonable and widely used approach. However, for applications where solvent dynamics are important (e.g., diffusion or viscosity), it may be desirable to avoid modifying solvent masses.

For example, solute-only HMR has been explored in the literature. Nagae et al. showed that applying HMR only to solute molecules (without changing water masses) better preserved translational and rotational diffusion properties compared with schemes involving water mass modification in their tested systems.

I hope this clarifies my point and provides the context for my questions above.

Reference: