Re: AMBER: Solvation net forces

From: Nicolas Lux Fawzi <fawzin.berkeley.edu>
Date: Tue, 25 Sep 2007 14:24:05 -0700

Sorry to break into the conversation -- here is a suggestion:
It seems like you could simply strive to keep the error in your force
less than or equal to the error in the force as calculated by the PME
approximation (with the set of parameters that the community [default
parameters?], or you in particular, are comfortable with) to the Ewald
sum for each atom. If your error is bigger than the PME vs Ewald
difference for the commonly accepted set of parameters, it might be too
much error to tolerate -- parameters leading to that order of magnitude
error have presumably in use for a reason, unless you want to undertake
the additional task of convincing everyone that more error is acceptable.
-Nick

Tyler Luchko wrote:
>
> On 18-Sep-07, at 6:14 PM, David A. Case wrote:
>
>> On Mon, Sep 17, 2007, Tyler Luchko wrote:
>>>
>>> I am implementing an implicit solvation model in Sander called 3D-
>>> RISM. The solvation forces for each solute atom are integrated off
>>> of a grid. By looking at several systems we have confirmed that we
>>> are getting the expected forces except for what appears to be a
>>> random drift force. I.e. there is a non-zero net solvation force on
>>> the solute. These forces scale linearly with the number of solute
>>> atoms and are roughly 0.1 kcal/mol/Angstrom in each direction. I
>>> would like to simply apply an equal and opposite force to the center-
>>> of-mass to remove the error.
>>
>> This looks like a pretty large force error to me. Can you get rid of it
>> by decreasing the grid size? (Even if this is expensive, it seems
>> like a good
>> test in order to make sure that there are not other bugs lying around.)
>>
>
> Yes, decreasing the grid spacing by 50% does make a difference and the
> drift force comes down to 4e-3 kcal/mol/angstrom per atom. However,
> this adds considerable cost in terms of memory and time. Decreasing
> the grid spacing again by 50% gives a drift error of 2e-3
> kcal/mol/angstrom per atom.
>
>> I don't have a particular opinion about subtracting it out, expect
>> that it
>> *sounds* dangerous....
>>
>
> I agree, it does sound dangerous, which is why I am interested in
> other opinions.
>
> In Essman et al., J. Chem. Phys., 103, 8577 they state that the net
> force on the atoms calculated from PME is not zero and on the the
> order of the RMS error in the force. In particular they suggest
> removing the average net force from each atom to avoid a slow Brownian
> drift about the centre of mass. If this is an acceptable approach I
> would like to know what is an acceptable drift error to subtract off.
>
> Thank you,
>
> Tyler
>
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber.scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Wed Sep 26 2007 - 06:07:26 PDT
Custom Search