Re: [AMBER-Developers] [AMBER] problem with randomizing velocities

From: Ross Walker <ross.rosswalker.co.uk>
Date: Thu, 15 Oct 2015 10:00:41 -0700

Hi Andrew,

I only had a chance to quickly glance at this but I suspect the adjustment of the number of degrees of freedom for shake and the randomization of velocities is occurring in the wrong order. The degrees of freedom adjustment is in subroutine degcnt in degcnt.F90.

This is called from subroutine runmd at the very beginning (line 258). However it looks like the random velocities (if irest=0 and tempi>0) happens inside pme_alltasks_setup:

174 if (ntx .eq. 1 .or. ntx .eq. 2) then
275 call all_atom_setvel(natom, atm_vel, atm_mass_inv, tempi)
276 #ifdef CUDA
277 call gpu_upload_vel(atm_vel)
278 #endif
279 end if

pme_alltasks_setup is called from pmemd.F90 line 270

269 if (using_pme_potential) then
270 call pme_alltasks_setup(num_ints, num_reals)
271 else if (using_gb_potential) then
272 call gb_alltasks_setup(num_ints, num_reals)
273 end if

while runmd is called on line 445

445 #ifdef MPI
446 call runmd(natom, atm_crd, atm_mass, atm_frc, atm_vel, atm_last_vel, &
447 gbl_my_atm_lst, remd_method, numexchg)
448 #else
449 call runmd(natom, atm_crd, atm_mass, atm_frc, atm_vel, atm_last_vel)
450 #endif /* MPI */

So that's the problem.

Volunteers (hint hint to my students!!!) to fix AND test this?

All the best
Ross

> On Oct 15, 2015, at 8:41 AM, Kalenkiewicz, Andrew (NIH/NICHD) [F] <andrew.kalenkiewicz.nih.gov> wrote:
>
> Dear Dr. Case,
>
> Thanks so much for your response. I wanted to send this earlier but figured I'd do a little more troubleshooting first. Your suspicion looks to be correct, as I tried a simulation without shake and did not get the large temperature spike as before. I did get some +/- 30K fluctuations near the beginning (even with a time step of 0.1 fs) but I presume this is due to the fact that the system was equilibrated with shake turned on?
>
> To minimize temperature fluctuations I suppose one can carry out heating and volume equilibration followed by short production runs from different random seeds all with shake turned off. The runs can then be restarted with shake turned on. This option still yields a slight temperature discrepancy (quick drop of ~10K followed by steady convergence back to 300K). Is this just an unavoidable consequence of switching shake on?
>
> I see that the randomization procedure for pmemd is defined in dynamics.F90 under vrand_set_velocities. Are you suggesting something should be changed in this procedure in order to account for shake?
>
>
> Very best wishes,
> Andrew
>
>
> From: David A Case [david.case.rutgers.edu]
> Sent: Monday, October 05, 2015 2:17 AM
> To: AMBER Mailing List
> Cc: amber-developers.ambermd.org
> Subject: Re: [AMBER] problem with randomizing velocities
>
> On Fri, Oct 02, 2015, Kalenkiewicz, Andrew (NIH/NICHD) [F] wrote:
>>
>> I am trying to start an accelerated MD simulation from an equilibrated
>> system using a random seed (ig=-1) for the velocities, but for some
>> reason there is a large temperature spike at the beginning of my
>> simulation when I use the following input file:
>>
>>
>> The temperature starts at ~430K, drops to around 270K, and then
>> equilibrates to 300K. But according to the manual:
>
> I suspect that the velocity randomization procedure doesn't properly recognize
> the fact that bonds to hydrogen are being shaken. Try a short test simulation
> without shake to see if you get good initial temperatures.
>
> cc-ing to the Amber developers' list, since this is arguably something that
> ought to be fixed. I think(?) there is code in libsff that addresses this,
> but that doesn't help much for pmemd users. (Also worth checking mdgx....)
>
> Volunteers are welcome...this is how you get established in a community.
> My start in Amber was to volunteer to add second derivative code. I
> polished up that code so faithfully, that now I am the ruler of the
> Queen's Navy....
>
> ....dac
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber


_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Thu Oct 15 2015 - 10:30:02 PDT
Custom Search