[gmx-users] What is the most reliable way to run repeats for reproducibility?
jalemkul at vt.edu
Thu Jan 11 14:07:31 CET 2018
On 1/10/18 9:59 AM, ZHANG Cheng wrote:
> Hi Mark,
> Thank you very much.
> ) For the link you provide, I think I could not manipulate most of the computer resources, as I submit my jobs to our cluster, and the jobs are distributed to different available cores randomly.
> ) For "random seed" of velocity, I found here and I enabled this option:
> gen_vel = yes
> So does it mean that it is better to use the same em.tpr and run different NVT,NPT,etc. for different repeats, so as to initialise it with different velocities?
This is common practice. Minimize the system once (there's usually no
variation here, the coordinates always move downhill on the potential
energy gradient) and then initiate however many simulations you want
with different starting velocities (requiring different gen_seed
values). This generates independent simulations.
> ) How the "natural chaotic divergence during equilibration" is reflected at which step?
> The link says: "The Central Limit Theorem tells us that in the case of infinitely long simulation all observables converge to their equilibrium values". But I think this "equilibrium" is not practical for protein in MD. For example, if I am running a protein at 370K, ultimately it will unfold, like boiling an egg in water, it takes 10 min. But in MD, the time scale is way more shorter, i.e. usually a few hundred ns scale. We could "never" see the proteins converges within that short period.
> So my understanding about "equilibrium" is the equilibration for temperature/pressure/density, but not the protein itself. Is that correct?
Yes, quantities like temperature and pressure converge relatively
quickly, but the dynamics of the system tend to take much longer, orders
Justin A. Lemkul, Ph.D.
Virginia Tech Department of Biochemistry
303 Engel Hall
340 West Campus Dr.
Blacksburg, VA 24061
jalemkul at vt.edu | (540) 231-3129
More information about the gromacs.org_gmx-users