[gmx-developers] SHAKE / LINCS crash

[David van der Spoel]

On 2015-01-28 18:13, "Pablo García Risueño" wrote:
>
>> On Tue, Jan 27, 2015 at 11:26 AM, "Pablo García Risueño" <
>> Risueno at physik.hu-berlin.de> wrote:
>>
>>> Dear Gromacs developers
>>>
>>> My name is Pablo G. Risueño, I am a postdoc researcher at the Humboldt
>>> University of Berlin. I am currently working with some colleagues in the
>>> development of new constrained MD schemes.
>>>
>>> A few years ago I had a conversation with prof. B. Hess, where he told
>>> me
>>> that both SHAKE and LINCS can make a MD simulation to crash (specially
>>> for
>>> large systems with many time steps). We would like to analyse the
>>> problem,
>>> so we would like to have some example(s) of simulation where such crash
>>> happens. We would need the Gromacs input file, and the
>>> coordinates/velocities files at the step before the crashing. May
>>> anybody
>>> help us?
>>>
>>
>> Hi,
>>
>> This is one of the easier problems in molecular dynamics. :-) You can
>> probably take anybody's MD tutorial, multiply the time step by 10 and
>> observe the simulation blow up from numerical instability. The constraint
>> algorithm is often the first thing to fail, but it is not itself the
>> problem. It seems to me that 99+% of the time the user has not prepared
>> the
>> system or model physics well enough, and most of the rest of the cases are
>> code bugs.
>>
>> It is not clear to me that even if a constraint algorithm could be
>> designed
>> to satisfy such cases (of poor user input), that having it would be a good
>> thing. It is irritating to have a simulation blow up because you were too
>> rough with it, but it could well be worse to have your simulation succeed
>> with only subtle evidence of the original problem.
>>
>> Cheers,
>>
>> Mark
>
> Dear Mark
>
> Thank you very much for your reply. However, if I understand it correctly,
> the crash of SHAKE I am speaking about is not a consequence of poor
> definition of the input, but of nearly-singularity in the constraint
> system of equations to solve. I guess that the (statistical) fails in
> LINCS may be due to the fact that the correction that LINCS does may not
> suffice to satisfy the constraints for some given configurations.
>

Both shake and lincs only correct bond lengths along the direction of 
the original bond. However, if due to an unphysical situation, e.g. 
large forces or too long time steps, the bond rotates too much, let us 
say 90 degrees, the problem becomes unsolvable. Mathematically that 
could correspond to a singularity in the equations, but the equations 
become unstable due to other problems as Mark indicated.

> Best regards.
>
>
>> --
>> 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.
>
>
> Beste Grüße.
>
>
> --
>
> Dr. Pablo García Risueño
>
> Institut für Physik und IRIS Adlershof, Humboldt Universität zu Berlin,
> Zum Großen Windkanal 6, 12489 Berlin, Germany
>
> Tel. +49 030 209366369
>


-- 
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