3
$\begingroup$

Does System Configuration on which we run MD simulation (to obtain a trajectory for the system under investigation) affects the analysis results.

If it does, then could anybody explain me to what extent (significant figures) does it affects.

I tried to look for answers but couldn't find it.
I came across the results, that doing simulation with different simulation software does affects the results to some significant extent but regarding system configuration I couldn't find.

Since I got to know (from some sources), that if one needs to compare different force field parameters for some molecule (let say water), one has to run simulation in a single system in order to not affect the results by any means (effects of system configuration such as cpu or gpu).

Hope someone could clarify my doubts.

$\endgroup$

1 Answer 1

6
$\begingroup$

If you are running a particular algorithm, then different codes should in principle give the same results. However, often algorithms require random numbers that are generated with seeds within the code. This can lead to problems with reproducibility - especially if the seeds cannot be set manually.

I think the best approach here is use statistical tools to estimate the error of a given calculation. If anyone ever does a single MD simulation (a single computation) the result is meaningless, as the data obtained could have been an outlier. The experiment should be repeated multiple times and then the appropriate statstical analysis done. I'm a big fan of bootstrapping. One would hope that results from different codes will then match-up (up to certain errors). If not, then it would be a major problem in field.

I don't directly work in MD, but I know they use ensembles to estimate error. For example see: https://pubs.acs.org/doi/10.1021/acs.jctc.1c00526

$\endgroup$
1
  • $\begingroup$ Thank you Sir for your valuable comment. I went through one of the MD books (Computer Simulation of Liquids by Allen and Tidsley) where as you mentioned about bootstrapping, they mentioned about block averaging method. In this method one can do multiple simulation with different initial state and later analysis was done with averaging the data obtained on all the trajectories obtained. $\endgroup$
    – anupama
    Jan 6, 2023 at 6:55

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .