[gmx-users] Same starting structure but different trajectories
mark.abraham at anu.edu.au
Wed Feb 17 11:46:09 CET 2010
----- Original Message -----
From: Emanuel Peter <Emanuel.Peter at chemie.uni-regensburg.de>
Date: Wednesday, February 17, 2010 20:46
Subject: Re: [gmx-users] Same starting structure but different trajectories
To: gmx-users at gromacs.org
> Dear Gromacs users,
> I have another question.
> Does that mean that I cannot compare my simulations if I have
> performed just
> 2 different jobs?
You can compare them. Whether that makes sense is another question. If they were identical, then there's no reason to compare them. If they've diverged slightly then you have to defend how much extra information you've gained compared with a simulation from a different starting point.
> How many runs should I then start to compare my chaotic
> trajectories?Thanks in advance.
That depends why you'd want to compare them.
> I still have some trouble to understand.
> Could you give me a more detailed answer why I get two different
> trajectories on the same machine? It should have the same
> registers and
> numerical chaos should be exactly the same in that case.
The algebrae of real numbers and floating-point numbers are not the same. As Erik said, you can add up differently-sorted large lists of floating-point numbers and get different results. MD adds up lots of such lists. The default mdrun algorithm in GROMACS is not intended to be reproducible. For example, dynamic load-balancing diverges runs almost immediately. Speed is a more important objective than reproducibility in most cases. You can use mdrun -reprod to try to reproduce the same calculation on the same machine, but there can be subtle differences that are outside the control of GROMACS...
More information about the gromacs.org_gmx-users