[gmx-developers] Fluctuations printed by mdrun

[Anton Feenstra]

On 07/12/10 23:09, Justin A. Lemkul wrote:
>
> Christian Mötzing wrote:
>>
>> The reason why I am asking is that our results did not turn out as I
>> expected. I simulated the waterbox with stepsize (dt) going from 0.00025
>> to 0.004. I was told that the fluctuation should go up with growing
>> stepsize. But it does not. It stays perfectly constant.
>>
>
> Fluctuations are dependent upon the input .mdp settings. I suppose that
> if you use accurate methods for nonbonded interactions and (especially)
> contraints for bonds, then I don't see how the fluctuations should
> inherently go up with the step size.

They will, as the stepsize determines the accuracy of the numeric 
integration (of Newtons equations of motion). But 4 fs step (0.004ps) 
for a waterbox is already very accurate, and you need a few tricks to 
actually see increased accuracy at smaller timesteps. I've explored this 
a while ago in J. Comput. Chem. 20 (8), pp 786-798, 1999 (link below).

I used shifted potentials for Coulomb and Lennard-Jones interactions to 
eliminate 'cut-off' noise. I'm not sure if (accurate) Ewald long-range 
would do the job well enough. Also, you need double precision; IIRC, at 
single precision the average energy drift over 10 short simulations 
(which is a reasonable proxy to fluctuations) already levels out 
somewhere between 5 and 6 fs timestep. As Fig 2 in my ref shows, with 
shifted potentials and double precision you see the drift going down up 
to .5fs timestep.


http://onlinelibrary.wiley.com/doi/10.1002/(SICI)1096-987X(199906)20:8%3C786::AID-JCC5%3E3.0.CO;2-B/abstract


-- 
Groetjes,

Anton
  _____________ _______________________________________________________
|             |                                                       |
|  _   _  ___,| K. Anton Feenstra                                     |
| / \ / \'| | | IBIVU/Bioinformatics - Free University  Amsterdam     |
|(   |   )| | | De Boelelaan 1083A - 1081 HV Amsterdam - Netherlands  |
| \_/ \_/ | | | Tel +31 20 59 87783 - Fax +31 20 59 87653 - Room P136 |
|             | Feenstra at few.vu.nl - www.few.vu.nl/~feenstra/         |
|             | "Confirmed" (Star Trek)                               |
|_____________|_______________________________________________________|