[gmx-developers] Re:Re:Re: energy calculation fails in "-rerun" mode

[Igor Leontyev]

Well, for most of users "-rerun" might be indeed infrequently used option,
but I have published paper based on exclusively this feature. If you have an
obtained trajectory, than it's possible to recalculate the energies with
higher level force-field. Such an approach is legitimate if energy
difference fluctuation between force-fields is lower than kT. There is an
analog in quantum perturbation theory: you have an energy perturbation in
the first order but the wave-function is perturbed only in the second order
of the small parameter. Thus "-rerun" option provides powerful tool for
combined force-field methods.    And, of course, the most valuable
quantities which you can extract from simulation are average values . With
additional effort one can be obtained by "g_energy" utility, but it's just
preferable to have it right in the log-file. Thanks for your help.


>> Leontyev Igor wrote:
>>>>> The problem appears when I am trying to recalculate energies and dhdl
>>>>> values
>>>>> using "-rerun" option.
>>>>
>>>> There were bugfixes for -rerun between 4.0.4 and 4.0.5. Please check
>>>> release notes for this sort of thing :-) I don't know if you might
>>>> still
>>>> have a real issue, but please update to 4.0.5 and try again.
>>>
>>> The upgrade to 4.0.5 version fixed issues 2) and 3), but still the
>>> average values are not printed at the end of the log-file (see the
>>> issue 1).
>>
>> I think that's intended as a feature, rather than a bug, e.g.
>> http://www.gromacs.org/pipermail/gmx-commit/2009-May/001656.html. I
>> don't know of a good reason either way. You can get the data you want
>> from g_energy on the .edr file, of course.
>>
>> Mark
>>> The averages were incorrect.
> mdrun and g_energy use a fancy way to determine exact averages and
> fluctuations over all the md steps.
> Since with -rerun you usually do not have all md steps, we would need to
> implement extra functionality
> for -rerun. But g_energy will give you the averages, so that is not
> worth the effort and the risk of more
> bugs because of infrequently used code.

> Berk