Dear Dmitry
What you suggest is possible and people have called that serial exchange
I think, or even serial tempering, depending on he literature. What you
are missing here (I think) is that you would have ONE replica traversing
temperature space. Of course, given infinite time it works, but at each
exchange point you get to move to the next higher or lower temperature.
What replica exchange (as usually understood) has are N replicas doing
this at the same time. So, you can exchange between T1 and T2 and AT THE
SAME TIME exchange between T10 and T11 for instance. That added
dimension is crucial for things to work well.
Adrian
On 8/23/11 3:59 PM, Dmitry Mukha wrote:
> )) ok, I definitely have more CPUs than just one, but also I am also sure
> that their number is not enough. the system is quite large, so I want to
> have a lot of replicas covering high temperature values and the performance
> good enough to see the effect of a traveling across temperature. Some kind
> of exhaustive phase space search.
>
> Maybe, it would be rather difficult [to rewrite the code], but memory data
> can be saved, thus decreasing the time for restart (without dumping to
> disk). I guess there is a lot of tricks here possible.
>
> 2011/8/23 Carlos Simmerling<carlos.simmerling.gmail.com>
>
>> it usually takes programs some time before they become efficient. the
>> spatial decompositino and load balancing are not ideal at the start.
>> additionally, much work needs to be done to start running MD, such as
>> reading and writing to disk, setting up the arrays, and so on.
>> stopping the program each ps or less would be quite slow. are you in a
>> situation where you have no way to access more than 1 CPU?
>>
>>
>
--
Dr. Adrian E. Roitberg
Professor
Quantum Theory Project, Department of Chemistry
University of Florida
roitberg.ufl.edu
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Aug 23 2011 - 13:30:03 PDT
