Re: [AMBER] Choosing the number of replicas in NEB

From: Ross Walker <>
Date: Fri, 3 Jun 2011 09:32:23 -0700

Hi Chris,

> I am trying to test the relative amount of time needed to run NEB by
> varying
> the amount of cpus I use. I am following the A5 Nudged Elastic Band
> method
> tutorial and I don't quite understand where it is decided to use 32
> replicas
> (which would require a minimum of 32 cpus.) I was hoping to run the
> simulation first on 16 cpus, but I don't understand how to make a group
> file
> which contains only 16 replicas. Any help would be greatly
> appreciated.

This is somewhat of a black art and I am not sure if there is a defined way
of choosing the number of replicas needed. In some ways the more you can use
the better but if you have too many you run the risk of seeing lots of
irrelevant motions on flat areas of your energy surface. Your best bet is
just to check the literature and see what others have used.

With regards to the number of CPUs it is not strickly this. You just need to
run that many MPI threads. So for example you could run 32 replicas on an 8
core machine. You would just run with mpirun -np 32 and each core would
process 4 of the replicas. Obviously you don't get the speed up of having 32
real cores but it is not too bad.

To run with 16 replicas just go through and change all 32's to 16. I.e. 16
structures in your inpcrd file, 16 lines in your group file etc etc.

All the best

|\oss Walker

| Assistant Research Professor |
| San Diego Supercomputer Center |
| Adjunct Assistant Professor |
| Dept. of Chemistry and Biochemistry |
| University of California San Diego |
| NVIDIA Fellow |
| | |
| Tel: +1 858 822 0854 | EMail:- |

Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.

AMBER mailing list
Received on Fri Jun 03 2011 - 10:00:02 PDT
Custom Search