[gmx-developers] tpic with large cavities

Berk Hess hess at cbr.su.se
Wed Nov 17 09:37:02 CET 2010

On 11/17/2010 09:13 AM, David van der Spoel wrote:
> On 2010-11-17 08.12, hess at sbc.su.se wrote:
>>> Hi all,
>>> Currently I'm looking into the cavity test particle insertion (TPIC)
>>> code
>>> since I want to make a modified version of the algorithm. Due to
>>> obtaining
>>> strange results I started to do some testing of the original tpic
>>> algorithm (using the current git version). My test system is a plain
>>> water
>>> box into which I insert a helium atom. I calculate the energy in a
>>> spherical cavity at the center of the box and when the radius of
>>> insertion
>>> (rtpi in the mdp file) is increased I expect the energies to converge
>>> towards the energy calculated using the standard TPI algorithm for the
>>> whole box.
>>> Small radii gives quite accurate energies, but large radii give much
>>> lower
>>> energies than expected. A clue might be that the threshold for my
>>> system
>>> lies around the size of the cutoff. When I tried to debug the code I
>>> noticed that (as far as I can see) do_forces are using the center of
>>> the
>>> cavity instead of the actual insertion point. But setting
>>> "fr->cg_cm[top->cgs.nr-1]" to "x_tp" instead of "x_init" did not help.
>>> This should not be a sampling problem since insufficient sampling would
>>> overestimate the energy rather than underestimate it. (And the energies
>>> seems to converge nicely too.)
>> I guess your problem problem is that r_tpi is large compared to rlist.
>> As the manual says, neighbor searching is done only once per step.
>> The neighborlist is made with respect to the center of the cavity,
>> but the energies are determined for the actual location of the molecule.
> So this means the algorithm can only be used in practice for very
> small cavities?
No, you can simply increase the cut-off.

I never thought someone would want to use tpic for cavities of the size
of or larger
than the cut-off. For those cases it should not be much work to modify
tpi or tpic
to do what you want.

>>> Another problem in the TPIC code is that the box volume is used to
>>> calculate the energies instead of the volume of the insertion sphere
>>> (which is constant and cancels out in the equation). This might give
>>> large
>>> errors if the box volume changes a lot due to pressure rescaling.
>> The volume issue is always tricky.
>> This corrects for the probability of encountering the certain part
>> of space where you are inserting. We should not correct for the volume
>> of the whole box. But I don't see immediately if and if so,
>> what type of volume correction is required for tpic.
>> Berk
>>> Ideas and comments are most welcome!
>>> Best regards,
>>> Daniel
>>> -- 
>>> gmx-developers mailing list
>>> gmx-developers at gromacs.org
>>> http://lists.gromacs.org/mailman/listinfo/gmx-developers
>>> Please don't post (un)subscribe requests to the list. Use the
>>> www interface or send it to gmx-developers-request at gromacs.org.

More information about the gromacs.org_gmx-developers mailing list