[gmx-developers] Conserved energy drifts horribly with md-vv

[David van der Spoel]

On 09/06/15 19:11, Shirts, Michael R. (mrs5pt) wrote:
>
>> we compared the md and the md-vv integrators in single and double
>> precision with the v-rescale thermostat. The md integrator does not
>> conserve energy in sp but is find in dp with sufficiently short time
>> steps (no constraints used anywhere). For md-vv the drift is horrible
>> even in dp.
>
> Then something someone did at some point in the integrator screwed up
> md-vv, because I did pretty careful analysis of energy conservation when I
> merged it in, down to checking the variance as a function of step size and
> making sure it's quadratic, both with and without constraints (as did md).
>
> The only difference should be is that vv should converge to the small-step
> limit faster (less bias with increasing time step), but with more noise
> than md (more noise with increasing timestep).
>
> Unfortunately, it's incredibly easy to screw up integrators without
> meaning to because of the structure of the inner loop now (cleaning that
> up is one of the projects over the next few months, after grants get in at
> the end of June).
>
> I'll try to look at this over the next week or so, but will be traveling
> the end of this week.

Or could it be an AVX2_256 double precision inner loop issue?
I guess the kernel tests would find that?
>
>
> Best,
> ~~~~~~~~~~~~
> Michael Shirts
> Associate Professor
> Department of Chemical Engineering
> University of Virginia
> michael.shirts at virginia.edu
> (434) 243-1821
>
>>
>> we compared the md and the md-vv integrators in single and double
>> precision with the v-rescale thermostat. The md integrator does not
>> conserve energy in sp but is find in dp with sufficiently short time
>> steps (no constraints used anywhere). For md-vv the drift is horrible
>> even in dp. Plotted is the drift in the conserved-energy term from the
>> edr file. 1000 Molecules, 1 ns.
>>
>>
>>
>> Dt(fs)  md/sp   md/dp   md-vv/sp  md-vv/dp
>> 2       486     501      5350      5287
>> 1       40      47       4915      4981
>> 0.5     -30     18       4769      4938
>> 0.25    -189    3        4306      4963
>> 0.1     -996    5        2496      4956
>> 0.05    -2783   4       -1464      4953
>>
>> Clues?
>> --
>> David van der Spoel, Ph.D., Professor of Biology
>> Dept. of Cell & Molec. Biol., Uppsala University.
>> Box 596, 75124 Uppsala, Sweden. Phone:	+46184714205.
>> spoel at xray.bmc.uu.se    http://folding.bmc.uu.se
>> --
>> Gromacs Developers mailing list
>>
>> * Please search the archive at
>> http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_List before
>> posting!
>>
>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>
>> * For (un)subscribe requests visit
>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-developers
>> or send a mail to gmx-developers-request at gromacs.org.
>


-- 
David van der Spoel, Ph.D., Professor of Biology
Dept. of Cell & Molec. Biol., Uppsala University.
Box 596, 75124 Uppsala, Sweden. Phone:	+46184714205.
spoel at xray.bmc.uu.se    http://folding.bmc.uu.se