[gmx-users] Run time for energy minimization, nvt, and npt equilibration

Justin Lemkul jalemkul at vt.edu
Tue Aug 2 13:33:27 CEST 2016



On 8/1/16 7:13 PM, Thanh Le wrote:
>> Hi guys, I have a question regarding the run time for energy minimization,
>> nvt and npt equilibration steps. Currently, I am doing a RNA-protein
>> simulation (using ff amber14 and spce water model) in a 10 A octahedron
> box
>> containing almost 400k solute atoms.
>> For the energy minimization step, I did it for 2 ns and received a
>
>> There is no time in EM, only number of steps.
> I used 1,000,000 steps for EM (I have always assumed 1,000,000 step to be
> 2ns)

If dt = 0.002 and using MD, then yes, a million steps will get you 2 ns.  Don't 
make assumptions; understand the differences in the processes you're carrying 
out.  There is no time aspect of EM.

>> potential E of nearly -2.25e-7.
>
>> Check your exponent :)  Surely your system does not have an energy of
> (effectively) zero.
> You are correct. Hahahaha. I didnt even the error. My potential E is
> actually -2.25e-7. Is that a good number?
>

Assuming you didn't intend to write the same number with the likely typo, I'll 
wager that -2.25e7 is reasonable for a system of 400k atoms.  What's more 
important is the value of Fmax, which should have converged to a nice, low value 
within the specified tolerance.

>> I am also doing 2ns for nvt and npt equilibration steps.
>> My question is: is 2ns too long for these steps?
>
>> You need to run long enough for the observable(s) of interest to
> converge.  Have
> they?
> The observable of interest? Do you mean temperature and density by it?
> I am still running it and it has been runing over 4 days (nvt run) on my
> mbp.

I would not subject a laptop to such an intense calculation.  Your system is 
quite large and frankly I don't know if a laptop can handle the heat that the 
job is going to throw off.  The run will certainly proceed at a fairly glacial 
pace since the system is so large.

-Justin

>> Also another question regarding running simulation on a cluster: do I need
>> to install gromacs mpi to run it on parallel? Is it necessary since I read
>> in the literature gromacs would use all the available cores to run if I
>> dont specify the number of cores.
>
>> Depends on the configuration of the cluster, how many cores you intend to
> use,
>> etc.  Everything is here:
>> http://www.gromacs.org/Documentation/Acceleration_and_parallelization
>> Consult with your sysadmin for best performance on whatever hardware you
> have.
> I will talk to my sysadmin about it.
> Thanks,
> Thanh Le
>

-- 
==================================================

Justin A. Lemkul, Ph.D.
Ruth L. Kirschstein NRSA Postdoctoral Fellow

Department of Pharmaceutical Sciences
School of Pharmacy
Health Sciences Facility II, Room 629
University of Maryland, Baltimore
20 Penn St.
Baltimore, MD 21201

jalemkul at outerbanks.umaryland.edu | (410) 706-7441
http://mackerell.umaryland.edu/~jalemkul

==================================================


More information about the gromacs.org_gmx-users mailing list