On Tue, Mar 18, 2008, David Cerutti wrote:
> In fact, I tried setting nscm=0 and also the dsum_tol an order of
> magnitude lower, but by this morning the energy loss was pretty much the
> same. I set the SHAKE tolerance to 1.0e-6, so a full order of magnitude
> lower rather than half its default value, and that seems to have finally
> worked--the total energy is now stable. I actuallly had two systems I was
> trying to do this for--the second had 136 glycerols and 1024 waters.
>
> That second system was also lossing energy, though only about half as
> severely; half as many glycerols, half as many SHAKE-constrained
> hydrogens, half the rate of energy loss. (Water was constrained by SETTLE,
> though I think I recall that pure water systems are also sensitive to the
> SHAKE tolerance for some reason.) I don't know if anyone might find that
> interesting...
Yes, many are interested in this, and there is a moderately lively dicsussion
among developers about what the default values should be for the next release.
I see no downside to tightening the SHAKE tolerance; it's a little harder to
know about the plusses and minuses of turning off nscm by default -- it seems
a good thing to do for your situation (although I'm not clear if you tried
duplicate simulations where the only difference was in nscm).
Changing dsum_tol without changing anything else is a hard call. Tom Darden
keeps promising to look into this, but hasn't been able to free up any time to
do the needed experiments. To be (ultra?)-careful, you might force the
program to use a slightly tighter reciprocal grid (nfft1-3), since reducing
dsum_tol puts more weight on the reciprocal part of the calculation. But it's
not obvious to me that this has to be done, and I've run longish simulations
with your parameters -- but we both could be missing some very subtle error.
Probably in amber 10, the defaults won't change, but we may add a discussion
in the User's Manual about what to do if you want better energy conservation.
...thanks for your reports....dave
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Fri Apr 18 2008 - 21:12:26 PDT
