[gmx-developers] reproducability of md sims

[Bogdan Costescu]

On Sun, Nov 30, 2014 at 9:05 AM, David van der Spoel
<spoel at xray.bmc.uu.se> wrote:
> Hi,
>
> I'm trying to test reproducibility of simulations of liquids. The simulation
> systems are flexible (no constraints) and run with the -reprod flag,
> starting from the same tpr, on a single core. However after a few hundred fs
> the energy starts to diverge.
>
> Are there factors impacting the reproducibility that are not controlled by
> the -reprod flag?

About 3 years ago, I have run extensive tests of binary
reproducibility with GROMACS 4.5.3, also with no constraints, for up
to 1ns. However I was interested in parallel runs, not single core
ones, for performance reasons. With '-reprod -dlb no' I was able to
achieve what I wanted, after I was told that with DLB enabled this is
not possible:

https://mailman-1.sys.kth.se/pipermail/gromacs.org_gmx-developers/2011-July/005323.html

Apart from these flags, reproducibility also requires keeping the same
number of ranks, architecture of nodes, libs, etc. to avoid the
rounding and order-of-operations problems mentioned earlier. I have
not redone these tests with 4.6.x or newer.

Cheers,
Bogdan