[AMBER] questions on pmemd.cuda for thermodynamic integration

From: Song-Ho Chong <songho.chong.gmail.com>
Date: Sat, 6 Oct 2018 00:06:49 +0900

Dear Amber developers,

I have questions on the GPU version of pmemd (pmemd.cuda)
for doing thermodynamic integration (TI) simulations.

I'm doing TI simulations for the system of my interest following
the Amber tutorial (decharge/vdw_bonded/recharge steps and
lambda = 0.0, 0.1, ..., 0.9, 1.0, and minimization/equilibration
are done for each lambda before entering production runs), and
encountered some errors during simulations.

1. vdw_bonded specific question on minimization

For decharge and recharge, I can do minimization using the
steepest descent and then conjugate gradient.
However, for vdw_bonded, I receive "input errors occurred"
with the same input, and I'm forced to use ntmin = 2 (i.e.,
only the steepest descent).

Is this the way minimization should be done for vdw_bonded?
(So far, I could not find any related description in the manual.)

2. vdw_bonded specific question on shake

ntc = 2 and ntf = 2 work for decharge and recharge steps, but
fail for the vdw_bonded step.
On the other hand, ntc = 2 and ntf = 1 work for all the steps.
Again, ntf=1 should always be used for vdw_bonded?

3. difficulty in handling lambda = 1.0

In most of the cases (not all the cases), simulations fail for
lambda = 1.0, even though there is no problem up to 0.9.

What usually happens is density significantly drops close to
zero during the equilibration at constant pressure.
Making the time step interval shorter did not help.

I just wonder why this happens at lambda = 1.0, though
it does not happen at lambda = 0.0.
(States at lambda = 0.0 and 1.0 are more or less symmetric.)

Is there any suggestion to avoid this problem?

Song-Ho Chong
AMBER mailing list
Received on Fri Oct 05 2018 - 08:30:02 PDT
Custom Search