CHARMM Development Project Forums User Discussion & Questions Performance and Benchmarking Performance of full-explict v. GSBP runs with REMD

Dear CHARMM users:

I am considering purchasing CHARMM c39b1 for a specific use and would be grateful for some help gauging simulation performance with the GSBP method before the purchase.

Specifically, I would like to use the GSBP method ( http://scitation.aip.org/content/aip/journal/jcp/114/7/10.1063/1.1336570 ) to study a spherical region with ~ 30 A radius (volume ~ 110K A^3; ~ 11K atoms) in Replica exchange (REMD) simulations. This will reduce the simulation box by a factor of five compared to explicitly simulating my whole system of interest.

I suspect that the GBSP method will reduce the simulation speed for two reasons:
(a) computing the reaction field for the implicit region -- noted as ~ 30% in the Im, Berneche, and Roux paper cited above.
(b) Loss of speed in paralle from applying domain decomposition to a spherical region -- this might be quite large and is the part that I am least able to estimate

Furthermore, the main reason that I want to use the GSBP method relates to my plan to combine GSBP with REMD. Therefore, my main goal in using GSBP is not to increase simulation speed per se (although of course I welcome that), but is mainly to reduce the overall potential energy of the system so that I can use fewer replicas in REMD.

Thank you for your time,
Chris.

To the best of my knowledge, the domain decomposition methods in CHARMM are limited to fully solvated systems with an orthogonal lattice, usually CUBIC or TETRAGONAL.

Keeping in mind that this work was done at least 15 years ago, you may wish to contact the authors directly; Wonpil Im is now at U. Kansas, and Benoit Roux is at U. Chicago.

Also, I don't think the replica exchange methods in CHARMM are compatible with domain decomposition yet, but I believe that work is underway to fix that.

I suspect GBSP and replica exchange should work, but may not be particularly fast.