OpenMM + Plumed Restraint "Spiral out of control"

[XingchengLin]

Hi,

I encountered a strange issue when using OpenMM + Plumed for an umbrella simulation between a protein and a DNA. I used the plumed to add a distance bias. But no matter which (groups of) atoms I chose, the system would run for a short while, and then both protein and DNA would start to spiral away from each other, very similar to a simulation that has not been equilibrated well.

Strangely, this only happens in a specific computer I used to run the simulation. In all other computers with the same code, my simulation works fine. If I turn off the plumed input in OpenMM or use a spring constant of 0.0, the simulation in that computer will also work fine.

I am very puzzled by this issue, and I am not sure how to debug this. I installed the latest OpenMM and the OpenMM-plumed plugin using Conda for all computers I tested. Has anyone encountered a similar issue before? Is there anything I can check on the computer to look into this error?

Thanks!

[XingchengLin]

I want to update everyone on this issue as I have found a solution. The problem was because, in this computer, the OpenMM-plumed applied the restraint in the initial time steps but stopped operating later. Although the bias energy was updated along with the simulation, this bias energy does not seem to be added to the system, causing the system to "remember" the initial bias and not be updated.

This is indeed a version issue. After I restored the Openmm-Plumed v1.0 associated with Openmm 8.0.0, the restraint functions correctly. Why was Openmm-Plumed v2.0 not working on this computer, however, is still unclear.

[peastman]

I think this is probably a duplicate of #78. Can you test the newest release and see if it works?

[XingchengLin]

Thanks, Peter! Unfortunately, the machine that used to have this error was a loaned one that was recently deactivated. I will keep monitoring the performance of OpenMM+Plumed in other machines and keep the community posted.