AMBER Archive (2007)

Subject: Re: AMBER: large RMS fluctuations in turning off VDW interactions by TI

From: steinbrt_at_scripps.edu
Date: Wed Jul 04 2007 - 07:51:08 CDT


Hi,

sorry, I am only transiently accesing the net right now and it looks like
I deleted half of this discussion before I realised that its on my topic.

I think David and Dave have pointed out the important points already,
menaing that the fluctuations you see in turning of a vdw-particle
(disappearing an atom) are absolutely to be expected. Without klambda>1
you have no hope of ever getting a free energy curve that is suitable for
numerical integration.

The recent paper to which David pointed you indeed discusses precisely
that and compares several different scaling schemes, as does the van
Gunsteren paper mentioned earlier. From my experience on the solvation
free energies, using separation shifted 'softcore' potentials looks like
the best option you can have in terms of curve shape that is good for
integration and dvdl-histograms that make it likely to get a sound average
without doing huge amount of sampling. However, at least for the simple
small molecule solvation free energies I looked at, the klambda=6 option
is not much worse and should easily give you good results if you dont have
access to softcore potentials. I am not sure if this is still the case if
a more complex conformational space like a ligand-protein complex is
studied (as David also pointed out already). We will do more on a test
system there in the near future, but for now I would expect that you have
a good chance at least trying to study your transformation with just the
klambda option.

If I remember right (and correct me if I dont) you mentioned dvdl-rms of
ca 30kcal/mol. While this is more than I saw in a disappearing toluene, it
is by no means prohibitively large. If you get dvdl-correlation times of
say 100fs on a 1 ns run, then this would correspond to a standard error in
the range of < 1kcal/mol and that only for the worst part of your free
energy curve. So chances are good that you will get a decent free energy
with a reasonable standard error.

Kind Regards,

Thomas

>> minor. Thomas is travelling right now, but maybe when he reads this, he
>> can
>> post the RMS fluctuations from this studies.

Sorry, I dont have my data here, so I'll have to refer you to the preprint
of our publication available on David Mobley's homepage.

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu