[gmx-users] how to repeat simulation correctly?

Erik Marklund erikm at xray.bmc.uu.se
Thu Nov 22 11:49:35 CET 2012


MD is chaotic regardless of how differences, however small, are created in the first place. This was just one example.

Stochastic terms in e.g. the v-rescale thermostat will rely on the same sequence of random numbers in separate simulations if the random number generator is seeded in the same way. But sure, choosing different seeds would be one way of having diverging simulations.

I think your point is correct in principle, e.g. when pressure is concerned. That equilibration time must however also be put in relation to how long it takes for two simulations to become uncorrelated.

Best,

Erik
22 nov 2012 kl. 11.30 skrev Felipe Pineda, PhD:

> Not to forget about the additional stochastic term in the V-rescale thermostat, when it's used. Since the equations are evidently deterministic, is the chaotic nature of MD just a numerical effect?
> 
> The practical point: if the velocities are reset upon a restart from an equilibrated frame in order to generate multiple, independent trajectories for statistical purposes, the equilibration will be probably lost and a new equilibration phase will be needed. Is this correct?
> 
> Best,
> 
> Felipe
> 
> On 11/22/2012 11:12 AM, Erik Marklund wrote:
>> It will depend on the integration algorithms, parallelization, etc. The equations are deterministic, but numerical differences may arise e.g. from different ordering of floating point numbers being added together in different simulations. The chaotic nature of MD would then have the simulations diverge over time, but the question is how long it takes for such differences to really manifest.
>> 
>> Best,
>> 
>> Erik
>> 
>> 22 nov 2012 kl. 10.13 skrev Felipe Pineda, PhD:
>> 
>>> Would "non-deterministic" be correct to characterize the nature of MD as well? There is also deterministic chaos ... And what about the outcome of starting several trajectories from the same equilibrated frame as continuation runs, i.e., using its velocities? Could they be considered independent and used to extract that valuable statistics mentioned in a previous posting?
>>> 
>>> Felipe
>>> 
>>> On 11/22/2012 10:04 AM, Erik Marklund wrote:
>>>> Stochastic and chaotic are not identical. Chaotic means that differences in the initial state will grow exponentially over time.
>>>> 
>>>> Erik
>>>> 
>>>> 22 nov 2012 kl. 09.52 skrev Felipe Pineda, PhD:
>>>> 
>>>>> Won't this same stochastic nature of MD provide for different, independent trajectories even if restarted from a previous, equilibrated frame even without resetting velocities, i.e., as a continuation run using the velocities recorded in the gro file of the selected snapshot?
>>>>> 
>>>>> Felipe
>>>>> 
>>>>> On 11/22/2012 12:55 AM, Mark Abraham wrote:
>>>>>> Generating velocities from a new random seed is normally regarded as good
>>>>>> enough. By the time you equilibrate, the chaotic nature of MD starts to
>>>>>> work for you.
>>>>>> 
>>>>>> Mark
>>>>>> On Nov 21, 2012 1:04 PM, "Felipe Pineda, PhD" <luis.pinedadecastro at lnu.se>
>>>>>> wrote:
>>>>>> 
>>>>>>> So how would you repeat the (let be it converged) simulation from
>>>>>>> different starting conditions in order to add that valuable statistics you
>>>>>>> mention?
>>>>>>> 
>>>>>>> I think this was Albert's question
>>>>>>> 
>>>>>>> Felipe
>>>>>>> 
>>>>>>> On 11/21/2012 12:41 PM, Mark Abraham wrote:
>>>>>>> 
>>>>>>>> If a simulation ensemble doesn't converge reliably over a given time
>>>>>>>> scale,
>>>>>>>> then it's not converged over that time scale. Repeating it from different
>>>>>>>> starting conditions still adds valuable statistics, but can't be a
>>>>>>>> replicate. Independent replicated observations of the same phenomenon
>>>>>>>> allow
>>>>>>>> you to assess how likely it is that your set of observations reflect the
>>>>>>>> underlying phenomenon. The problem in sampling-dependent MD is usually in
>>>>>>>> making an observation (equating a converged simulation with an
>>>>>>>> observation).
>>>>>>>> 
>>>>>>>> Mark
>>>>>>>> 
>>>>>>>> On Wed, Nov 21, 2012 at 8:12 AM, Albert <mailmd2011 at gmail.com> wrote:
>>>>>>>> 
>>>>>>>>  hello:
>>>>>>>>>    I am quite confused on how to repeat our MD in Gromacs. If we started
>>>>>>>>> from the same equilibrated .gro file with "gen_vel        = no" in
>>>>>>>>> md.mdp,
>>>>>>>>> we may get "exactly" the same results which cannot be treated as
>>>>>>>>> reasonable
>>>>>>>>> repeated running. However, if we use "gen_vel=yes" for each round of
>>>>>>>>> running, sometimes our simulation may not converged at our simulated time
>>>>>>>>> scale and we may get two results with large differences.
>>>>>>>>> 
>>>>>>>>>    So I am just wondering how to perform repeated MD in Gromacs in a
>>>>>>>>> correct way so that our results can be acceptably repeated?
>>>>>>>>> 
>>>>>>>>> thank you very much.
>>>>>>>>> Albert
>>>>>>>>> --
> 
> -- 
> gmx-users mailing list    gmx-users at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-users
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
> * Please don't post (un)subscribe requests to the list. Use the www interface or send it to gmx-users-request at gromacs.org.
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists

-----------------------------------------------
Erik Marklund, PhD
Dept. of Cell and Molecular Biology, Uppsala University.
Husargatan 3, Box 596,    75124 Uppsala, Sweden
phone:    +46 18 471 6688        fax: +46 18 511 755
erikm at xray.bmc.uu.se
http://www2.icm.uu.se/molbio/elflab/index.html




More information about the gromacs.org_gmx-users mailing list