[gmx-developers] Output volume for replica exchange?

Mark Abraham Mark.Abraham at anu.edu.au
Tue Sep 18 04:08:22 CEST 2012

On 18/09/2012 11:00 AM, Shirts, Michael (mrs5pt) wrote:
> Hi, all-
> While Mark Abraham is helping me clean up the replica exchange, I was
> reminded about one issue that I mentioned a while back.
> If one performs replica exchange switches very frequently (say, 20 to 50
> steps), which is becoming the current best practice, then the detailed
> information about the replica switches at every exchange is (and was in 4.5)
> printed out to the log file, which  makes these log files get extremely
> large. This has caused some storage problems for some people.

Yes it's currently unscaleable.

> How could this behavior be modified?  Should there be a 'verbose' replica
> exchange option that suppresses this output?  Should this information be
> part of a separate file that is only printed occasionally? Any other ideas?

Some of that information is needed for demultiplexing, but much of it 
doesn't have an application that I can see. Creating energy file entries for
* delta (the energy difference from which the probability is constructed),
* a boolean for whether exchange occurred
is one possibility, but it doesn't scale to the kind of 
multiple-partners scenarios implemented in 4.6 so perhaps a new output 
file is in order.

Also, the demux capability of trjconv should be extended to eneconv. I 
have that in a private branch if anyone else feels keen about it.


More information about the gromacs.org_gmx-developers mailing list