search

scychon / openmm_drudeNose

Dual Nose-Hoover thermostat with extended Lagrangian integration for drude polarizable MD in OpenMM
6 stars 0 forks source link

The stability of drudeNose #6

Open qiuzy029 opened 4 years ago

qiuzy029 commented 4 years ago

Hi, scychon! I come back again. After installing this plugin, I have used it in my test system. My test system is very simple which is a short peptide in bulk water and on water/air interface. Each system has five independent runs. But the simulations fails sometimes.

For the cases that peptide in bulk water, evergything goes well. But in the case of peptide on the water/air interface, the simulations often fail with error "Energy is NaN" after the simulations had run for ~100 ns or even longer or at least tens of ns. And it can be found sometimes the energy deviated from the normal fluctuation and became apparently lower. I have checked the drude particles' distance to their parent atoms. At the moment of simulation going wrong, some drude particles reached the upper distance limite (0.25 Ang.). Besides this, I don't kown what I can do to debug. Did u have met this problem?

Thanks for your time!!!

scychon commented 4 years ago

Hi quizy029,

I'm sorry for my delayed response. Your description seems to be related to a rather unstable propagation of the Drude particles. This typically happens when the timestep is too large, and usually polarizable simulations require shorter timesteps than nonpolarizable simulations. What timestep are you using?

And it also can be the case that your particular force field is not suitable for simulations on air/water surface, since many force fields are developed and benchmarked under bulk solution condition. In many cases, putting a protein on the air/water interface would destabilize the surface protein residues, and if the charge of the Drude particles are too large then the Drude simulation can be unstable. You might be able to avoid such problem using shorter timesteps, but it's not always guaranteed to work. I have not experienced such case with DrudeNoseHoover thermostat but I had similar experience with SCF type of integration in Gromacs simulation package. Hope you could figure out a way to avoid this problem.