How to Check Convergence in Biased Simulations with OPES in PLUMED?

[WJiangH]

Dear plumed developers,

I’m currently running biased simulations using PLUMED (v2.9) to study transitions between two states (A and B) with a designed collective variable. I’m using OPES_META for enhanced sampling (in LAMMPS 2024). I have two questions regarding convergence and parameter settings:

1. Convergence Check:

When using a biased method like OPES_META, is it recommended to assess convergence:
• Based on a single, sufficiently long simulation that includes multiple back-and-forth transitions between A and B?
• Or can convergence be assessed by combining several independent, shorter simulations that may each only capture one full transition?

In other words, is the quality of sampling tied to the continuity of one long trajectory, or can it be reliably built from multiple independent shorter ones?

2. Energy Units and Barrier Parameter:

If I set the units in plumed.dat as: UNITS LENGTH=A TIME=ps ENERGY=96.485, which means the energy is in eV, should the BARRIER parameter in OPES also be provided in eV, or should it still be given in kJ/mol? I’m asking because the BARRIER value is involved in controlling the bias factor in OPES, and I want to ensure I’m providing the correct unit.

Thank you so much for your help!

[invemichele]

Hi,

Based on a single, sufficiently long simulation that includes multiple back-and-forth transitions between A and B?

This would be the preferred method

Or can convergence be assessed by combining several independent, shorter simulations that may each only capture one full transition?

We have shown in https://pubs.acs.org/doi/10.1021/acs.jctc.2c00152 that this can also work in a toy system, provided that you do have at least one transition in each of the replicas. This can happen because in OPES_METAD the bias quickly becomes quasi-static. Notice that for complex systems this approach is not guaranteed to work and multiple transitions in a single replica remains the gold standard.
There are several other approaches aimed at obtaining a free energy difference without back-and-forth sampling, e.g. here they use multiple forward-only and backward-only transitions.

If I set the units in plumed.dat as: UNITS LENGTH=A TIME=ps ENERGY=96.485, which means the energy is in eV, should the BARRIER parameter in OPES also be provided in eV, or should it still be given in kJ/mol?

The BARRIER parameter is in plumed energy units. so if you change units from kJ/mol to eV and want to obtain the same simulation you must properly convert the BARRIER value.