Hi Dmitry,
Yes you are mostly correct in that it is unlikely to have a measurable effect. However if you use the same random see for each restart with thermostats (and barostats) that rely on the random number stream you can introduce statistical biasing that could ultimately influence your results. The following paper is a nice read that highlights this issue:
http://pubs.acs.org/doi/abs/10.1021/ct8002173
If you were using the langevin thermostat you may want to consider re-running your simulations that were started with the same random seed each time just to be on the safe side.
All the best
Ross
> On Aug 16, 2016, at 1:27 PM, Dmitry Suplatov <genesup.gmail.com> wrote:
>
> Bill, thank you again for you time and help
>
> I want to clarify one last point.
>
> Setting ig of a restart run to a positive value used in the initial run is not incorrect, just not an entirely correct way to restart. Is it?
>
> I mean that since I use the rst file from the last step of the crashed run I will get a new trajectory (not a repetition of the first part of the run) even with the same sequence of "random" values. However, this new trajectory will be biased from a fundamental statistical point of view. Still, it most likely will be a fine MD trajectory. Would it?
>
> I simply wanna know if I should delete previous trajectories which were restarted by preserving ig from the previous run.
>
> Dmitry
>
>> On 16 Aug 2016, at 22:40, Bill Ross <ross.cgl.ucsf.edu> wrote:
>>
>> Putting it another way, I would likely start fresh from my last restrt with ig=-1, or rerun if I wanted tidyness.
>>
>> Bill
>>
>>> On 8/16/16 12:12 PM, Bill Ross wrote:
>>> The 'ig' value determines where the random number generator starts, so
>>> if the run restarts in the middle with ig from the beginning, the random
>>> numbers for the first part of the run will repeat.
>>>
>>> Is perfect reproducibility important?
>>>
>>> Bill
>>>
>>>
>>>> On 8/16/16 9:43 AM, DmitryASuplatov wrote:
>>>> Dear Amber users,
>>>>
>>>> I want to continue a crushed run. I use Langevin thermostat.
>>>> Do I understand correctly that in order to continue the trajectory
>>>> exactly as it would have ran if no crush occurred I need:
>>>>
>>>> - offset "nstlim" for the remaining number of steps
>>>> - set "ig" to the exact value used in the previous run (if I used -1 in
>>>> the previous run then I have to open the .out file and look for the
>>>> particular ig value there).
>>>>
>>>> Please comment.
>>>>
>>>> Thank you for your time.
>>>> Best,
>>>> Dmitry
>>>>
>>>> _______________________________________________
>>>> AMBER mailing list
>>>> AMBER.ambermd.org
>>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>
>>> _______________________________________________
>>> AMBER mailing list
>>> AMBER.ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Aug 16 2016 - 16:00:02 PDT
