[gmx-users] Re: BAR calculations with tabulated non-bonded potentials

[Justin A. Lemkul]

On 5/19/12 5:42 PM, jrustad wrote:
>
> Justin A. Lemkul wrote
>>
>> I don't know how complex the system is you're dealing with, but at this
>> point,
>> isn't it just far more straightforward to construct a rudimentary force
>> field
>> with a few atom types rather than rely on (slow) tabulated functions and
>> potential funny business to get the free energy code to work?
>>
>
> Yes, well, I don't want to invent and validate a whole new forcefield, if
> that's what you mean.
> I would sooner write a BAR code from scratch than do that.
> Do you think the approach above won't work for some reason?

As far as nonbonded interactions go, I was simply suggesting constructing a new 
force field for better performance.  Doing table lookups for all the pairs you 
listed before will be a serious performance hit.

> Its unusual maybe to set a constraint to zero bond length in the pull code
> (which I think is the only way in GMX to set constraints between atoms in
> different molecules).
> But other than that, it looks straightforward.

You don't need the pull code if the two atoms are in the same [moleculetype] - a 
"molecule" in Gromacs is any set of atoms; they don't necessarily have to make 
chemical sense.  The only problem I see is that if you have two atoms in the 
same location, while simultaneously turning one "on" and the other "off," you 
will likely develop strong repulsive interactions between them such that the 
constraints may fail and your atoms will be displaced from the desired location. 
  I can't say for sure whether it will be a problem, but if you've got atom A 
existing at 20% and atom B trying to exist at 80% in the exact same location, 
that sounds like a recipe for trouble.

-Justin

-- 
========================================

Justin A. Lemkul, Ph.D.
Research Scientist
Department of Biochemistry
Virginia Tech
Blacksburg, VA
jalemkul[at]vt.edu | (540) 231-9080
http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin

========================================