Re: [AMBER] Fwd: NVE GPU question

From: Jason Swails <>
Date: Thu, 22 Jan 2015 17:50:31 -0500

On Thu, Jan 22, 2015 at 5:22 PM, Joseph Baker <> wrote:

> Hi Jason,
> Thanks for the reply. I will try playing with the SHAKE tolerance to see
> what happens with dt=0.002. I guess I should also point out that we we were
> also seeing worse-than-typical energy conservation for the cases where we
> were using dt=0.001 and no SHAKE at all, so that wouldn't explain those
> cases I guess?

​0.001 might be too large of a time step for an ionic liquid. The charges
are quite a bit larger in the IL than regular water, right? So the forces
are probably correspondingly larger?​

> I tried adjusting rsum_tol just now and was given the error that rsum_tol
> is only used if ew_type = 1, so I tried setting ew_type to 1 (default is
> 0), and then get the error that ew_type must be 0 for pmemd. So apparently
> this is not changeable when using pmemd? I'll try adjusting just dsum_tol
> down a few orders of magnitude and see how that influences the drift.

​Ah, ew_type=1 means pure Ewald (no PME). So rsum_tol doesn't do what I
thought it did... I guess dsum_tol is the setting to play with, although
it's not clear to me that that will improve overall accuracy.

You could also try to set the PME order to 5 or so (I don't think that will
work with pmemd.cuda, but it should work with pmemd, and hopefully give
better accuracy in the reciprocal forces).

If the discrepancy is occurring due to integration errors, I would expect
that would come as a result of forces that are larger on average than the
simulation giving better energy conservation. You can check this
explicitly by setting the "ntwf" flag to write forces to a trajectory.

All the best,

P.S. -- Have you run these simulations on CPUs with pmemd to verify that
you get the same (bad) energy conservation there? It would be a good
control to have.

Jason M. Swails
Rutgers University
Postdoctoral Researcher
AMBER mailing list
Received on Thu Jan 22 2015 - 15:00:03 PST
Custom Search