Re: Running multiple-replicas metadynamics

From: Vermaas, Joshua (Joshua.Vermaas_at_nrel.gov)
Date: Fri Nov 03 2017 - 19:05:18 CDT

Hi Prapasiri,

The big thing you are missing is that NAMD has a mechanism to launch
many loosely coupled jobs from a single namd executable. See
http://www.ks.uiuc.edu/Research/namd/2.12/ug/node69.html

Based on the example in the user guide, you would create 5 directories
(most folks are uncreative and call them 0, 1, 2, 3, 4) where the things
like the dcd and the colvar.state file for each independent replica
would live. Then your single mpi-based namd executable would run each of
these replicas simultaneously, and as each replica advances in time, it
will deposit its contribution to the overall pmf in a shared file, which
is the bias applied to all replicas. So now you have 5 data-collectors
rather than 1, so you in principle sample 5x as fast.

-Josh

On 11/03/2017 05:40 PM, Prapasiri Pongprayoon wrote:
> Hi All,
>
> I’m new to NAMD and need your valuable help. I ‘m now doing metadynamics of drug translocation through membrane protein. The simulations go well, but I suspect that it will require large CPU time for my drug to explore all configuration space. So, I decide to move to the multiple-replicas metadynamics. Based on the manual, it seems that I need to:
>
> 1. turn on “multipleReplicas”
> 2. Add replicaID, replicasRegistry, replicasUpdateFrequency, and dumpPartialFreeEnergyFile
>
> After going through the manual, I still don’t understand the process clearly. So, it would be very appreciated, if you could explain how to set up the run.
>
> From my understanding: (Pls correct me if it’s wrong)
> 1. To run multiple-replicas meta dynamics, I need a number of pdbs with different drug’s positions. Each system is called “replica” where each is defined as replicaID. 5 systems in their individual folder.
> 2. I still need to have a single file (put in replicasRegistry) containing the paths of “colvar.state" and “hill.traj” files that will be generated when all five start running.
> 3. If I have 5 systems = 5 replicas (everything is the same except position of drug in each system), I need to run them separately with its own .conf and colvar files. The only difference among them are “replicaID”. Is this correct?
> 4. When all are run, they will talk to each other via “colvar.state" and “hill.traj” files defined in “replicasRegistry”. The file just has lines showing the location of state and hill files.
> 5. 5 runs will generate their own outputs and .pmf, but the pmf obtained from each replica is generated by combining data among 5 replicas. So, 5 pmfs from 5 replicas are generated, but there are the same. Is this correct? For the partial.pmf, does this file reflect the influence of individual run on the overall pmf file?
> 6. To restart the runs, I just add “colvarsInput input.colvars.state" in .conf of all five.
>
> These are what I understand from the manual and NAMD list.
>
> If these are correct, I still have some questions
> 1. If I have 5 replicas, I have to run 5 replicas independently. How do they communicate if they don’t start running at the same time?
> 2. Based on the recipe above, does it mean that each replica explores its own configuration space and then the data obtained from each replica will be combined and used to get the overall pmf?
>
> Is there any tutorial for multiple-replicas metadynamics that I can go through?
>
> Thanks for your help and patience in advance.
>
> Regards,
> Prapasiri
>
>

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2018 - 23:20:41 CST