NVT vs NPT for Non-Equilibrium Switching in pmx/GROMACS

pmx
, , , ,
1

I’m using pmx + GROMACS to run non-equilibrium alchemical free energy calculations for large protein–protein interactions (e.g., ~150 k atoms, like Fc–FcRn). My workflow is essentially:

  • Equilibrate both end states (start with NVT → NPT → production NPT),

  • Extract snapshots,

  • Run fast-growth non-equilibrium transitions from those snapshots,

  • Analyze with Crooks/BAR.

In tutorials and some practice examples I’ve seen, the non-equilibrium switching is run in NPT (pressure coupling on), even for very short transitions (e.g., 50–100 ps in the standard pmx tutorials). However, many methodological discussions suggest that barostat fluctuations add noise to work distributions during driven perturbations.

I want to gather community experience specifically on the ensemble choice for the switching step:

Main Questions

  1. Do experienced pmx users typically run NE switching in NVT or NPT?

    • Is there a de facto best practice?

    • If NPT is used, is it because the tutorial mdp simply has it “on by default,” or is there a theoretical/empirical advantage?

  2. For short switching times (≤1 ns or ~2 ns):

    • Does NPT add significant variance to work distributions?

    • Does NVT give noticeably tighter overlap / lower BAR uncertainty?

  3. For longer switching times (~5–8 ns):

    • Do people find NPT and NVT give similar convergence?

    • Is pressure coupling less of a concern when the switching is slow relative to barostat timescales?

  4. Could you share practical rules of thumb or MDP settings (e.g., barostat settings, thermostat choices) that have worked well for your pmx NE runs in GROMACS?

2
  1. This depends on what ensemble you want to simulate. Most of the experimental measurements are performed at constant pressure, thus you may want to calculate Gibbs free energy, i.e. NPT ensemble
  2. I have not observed significant difference in convergence due to switching on/off the thermostat. I would expect NVT to lead to higher work dissipation, e.g. when coupling a molecule (or a part of it) to the system cavity needs to be created, water pushes against the box walls which are not rescaled.
  3. Time constant for barostat is always at least an order of magnitude shorter than a switching timescale.
  4. I use Parrinello-Rahman barostat with tau-p=5ps. c-rescale barostat should also be a good choice
1 Like