AMBER Archive (2008)

Subject: Re: AMBER: Discontinuous Constant pH Restarts

From: Neil Bruce (Neil.Bruce_at_postgrad.manchester.ac.uk)
Date: Wed May 07 2008 - 15:21:28 CDT

David, thanks for your response.

I've since tried a few tests and found that this problem only occurs
when executing Sander in parallel, the serial code restarts as expected.

I've run tests on three computers and found the same behaviour (one
Linux/x86_64/ifort 9/mpich2, one Linux/ia64/ifort 9/bullmpi and
another that I don't have the details of at the moment but can get if
it helps). On the two computers I've given details for, the cnstph
test runs without error. 'Standard' GB simulations also restart
properly.

Could this be a compilation error?

Thanks, Neil

Quoting "David A. Case" <case_at_scripps.edu>:

> On Thu, May 01, 2008, Neil Bruce wrote:
>>
>> I am trying to perform constant pH simulations, but am having problems
>> when restarting simulations. Below is the energy data of the last step
>> of one trajectory and the first step of a restarted trajectory. As you
>> can see there is a large jump in the potential energy of the system.
>>
>>
>> NSTEP = 40000 TIME(PS) = 180.000 TEMP(K) = 299.71 PRESS
>> EELEC = -3128.8680 EGB = -2220.8243 RESTRAINT =
>> ------------------------------------------------------------------------------
>> NSTEP = 1 TIME(PS) = 180.002 TEMP(K) = 293.59 PRESS
>> EELEC = -3472.0209 EGB = -1146.9263 RESTRAINT =
>
> It's weird that the EGB energy is like a factor of 2 smaller...is there any
> possibility of a different intdiel or extdiel factor in the second run?
>
> You could explore this more by running two short simulations (say 50 steps
> each) where you set ntpr=1, and restart the second from the end of the first.
> Then compare that to a single 100 step run from the initial starting point.
> Using a smaller time step might also help identify restart problems.
>
> I'm cc-ing this to John Mongan, but I don't recognize the symptoms myself.
>
>>
>> Examining the cpout files shows that the protonation states are being
>> carried over into the restarted simulation as the initial states in
>> the restarted simulation match the end of the preceeding simulation. I
>> have tried using both formatted and unformatted restart files and
>> increasing cut and rgbmax, with the same result. Below is the mdin
>> file and the command used to run sander for the restarted simulation
>> (the mdin file for the initial simulation differs only in irest, ntx
>> and tempi):
>>
>> &cntrl
>> imin=0, irest=1, ntx=5,
>> icnstph=1, solvph=7.1, ntcnstph=5,
>> nstlim=40000, ntpr=1, ntwx=200,
>> cut=30.0, scee=1.2, igb=2, saltcon=0.1,
>> ntb=0, dt=0.002, nrespa=1,
>> ntt=1, temp0=300., tautp=2.0,
>> ntc=2, ntf=2, tol=0.000001,
>> /
>>
>> sander.MPI -O -i constph.in -p pep.prmtop -c constph.rst1 -cpin
>> constph.cprst1 -o constph.out2 -x constph.mdcrd2 -r constph.rst2
>> -cpout constph.cpout2 -cprestrt constph.cprst2
>>
>
> ...regards...dac
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu