2
$\begingroup$

Suppose, I am running a Monte Carlo simulation of protein folding using harmonic potential, function.

The inner loop of Monte Carlo has 100 iterations, and the outer loop has 1000 iterations.

What are some symptoms that could tell me that my simulation is not running properly?

$\endgroup$
4
  • 7
    $\begingroup$ I'm afraid this question is too vague. Are you using MD or MC, or perhaps some AI (like AlphaFold)? Are you using an all-atom force field, a coarse grained force field or a machine learning potential? While theoretically someone may be able to exhaustively enumerate all possible symptoms even without you providing these information, the number of people capable (and willing) to answer your question will dramatically increase if you can make your question more specific. $\endgroup$
    – wzkchem5
    Jun 3, 2023 at 11:21
  • 2
    $\begingroup$ Please add more details: specifically what system are you studying, what FF are using and what timestep is used. It's difficult to give a specific answer otherwise. $\endgroup$ Jun 3, 2023 at 11:34
  • $\begingroup$ Well, very crude approach - check protein geometries during the optimization.If they get distorted to some unphysical shape(s), you are on the wrong track. $\endgroup$ Jun 3, 2023 at 19:00
  • 1
    $\begingroup$ Check that energy, pressure/density and temperature converge to a plateau that oscillates slightly. Do replicate runs. 10 or more is best. Do way more steps (not enough steps will show up in energy/pressure/temp profile) $\endgroup$
    – B. Kelly
    Jun 9, 2023 at 19:22

1 Answer 1

5
$\begingroup$

There are generic self-consistency checks (SCCs) that check if a simulation set has converged. The most basic of these is to use block analysis to examine the convergence of standard error in observables (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2865156/) and a very recent example of detailed application of SCCs can be found here: https://pubs.acs.org/doi/10.1021/acs.jctc.2c01140.

Another simple check is to perform two Monte Carlo runs at slightly different temperatures and compare the resulting probability distributions of observables. These should be Boltzmann distributions and hence be related by inverse exponential temperature.

$\endgroup$
1
  • 1
    $\begingroup$ Replicate runs >>>>>>> Block averages $\endgroup$
    – B. Kelly
    Jun 9, 2023 at 19:20

You must log in to answer this question.

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