Hi,
Ross has summed up my worries, and yet at a practical level Thomas seems to be getting results that agree very closely with experiment, whilst following a protocol with a few unshaken Hs and dt=0.002 (to my understanding of e.g. Steinbrecher et al J Comp Chem 2011 3253-3263). If we are talking about one unshaken hydrogen, one can see that the error due to this might be extremely small compared to other sources of error in a TI simulation - e.g. not running the simulation for long enough. So we are back to the trade-off between the accuracy of individual time points against the ability to sample enough of phase space. I've not started these simulations yet, but doubling the computational cost for the benefit of one unshaken hydrogen is something I'd like to avoid, should there be support for such a position. Unfortunately the unshaken hydrogen is obviously at the heart of the change.
Anybody else have any thoughts?
Thanks,
Peter.
-----Original Message-----
From: Ross Walker [mailto:ross.rosswalker.co.uk]
Sent: 02 November 2011 14:37
To: 'AMBER Mailing List'
Subject: Re: [AMBER] Softcore TI.
Hi,
> the general assumption is that having one or two fast vibrations un-
> SHAKED
> (as opposed to tens of thousands in non-SHAKE water solvation) will
> have
> little impact on your simulation. Depending on the actual system at
> hand,
> you have to check that this very localized integrator inaccuracy does
> not
> mess with your structure and dynamics.
This cannot be true in general. If you have a single frequency in your
system that is faster than the nyquist limit for your integrator and
timestep you have to reduce your time step. Simply saying it won't matter
for that atom makes no sense whatsoever, you will have integration issues
for this atom and ultimately stability problems. You might not see it
because it is hidden by the rest of the noise of the system but it is there.
Thus I do not believe you can defend using a 2fs timestep for TI-MD if you
don't shake ALL hydrogens in your system. I'd be VERY skeptical reviewing a
manuscript that tried to do otherwise without establishing that ALL
frequencies in the system are appropriately slow for the use of a 2 fs
timestep.
> As far as energy conservation goes, do you see a difference between 1fs
> and 2fs timestep TI-MD runs? If you are using a thermostate, this
> should
> not be noticeable.
Because it is lost in the noise. It does not necessarily mean the results
are correct. You would need to very carefully follow the unshaken atoms to
check on their stability. What do their average and RMS bond lengths come
out like?
It is possible that the local environment in some way damps the vibrations
of the hydrogens in question in which case it might be appropriate to use a
2fs timestep with shaking these bonds but this seems unlikely to me and
would have to be proven. Simply asserting that it does not appear to cause a
problem seems hokey to me.
I would just use a 1.0fs timestep, or ideally a 0.8 to 0.5fs timestep to be
on the safe side and resign yourself to running the simulations for longer.
It will certainly cast less doubt on the validity of your simulations when
it comes time to publish.
All the best
Ross
/\
\/
|\oss Walker
---------------------------------------------------------
| Assistant Research Professor |
| San Diego Supercomputer Center |
| Adjunct Assistant Professor |
| Dept. of Chemistry and Biochemistry |
| University of California San Diego |
| NVIDIA Fellow |
|
http://www.rosswalker.co.uk |
http://www.wmd-lab.org/ |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
---------------------------------------------------------
Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Nov 02 2011 - 08:00:06 PDT
