LAMMPS WWW Site - LAMMPS Documentation - LAMMPS Mailing List Archives
Re: [lammps-users] Questions about the output of Colvars and Multiple-replica metadynamics
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [lammps-users] Questions about the output of Colvars and Multiple-replica metadynamics

From: Giacomo Fiorin <giacomo.fiorin@...24...>
Date: Wed, 14 Jun 2017 12:54:03 +0200

Hi Xinyi, I'm copying the mailing list on my reply to share the information.  Please do the same.

On Mon, Jun 12, 2017 at 9:32 PM, Xinyi Shen <xinyis@...1683...> wrote:
Dear Prof. Giacomo Fiorin,

I am Xinyi Shen, a first year PhD student in Princeton University. I am learning to use your collective variables module(colvars) for LAMMPS in my research, your work is really cool and helps me a lot! However, as a beginner, I meet some problems. Could you please help me?

I read the manual for Colvars, and find out the output file output.pmf is only used for one-dimensional calculations. But in my research, I need to use two different CVs, distance and angle, at the same time. What should I do if I want to get the result of PMF with two variables?

That depends on what you are using: metadynamics will write the 2D PMF automatically; ABF will need post-processing for anything above 1 dimension; umbrella sampling will also require post-processing.

(My one-dimensional calculation works well, but when I do the same thing to my multiple variables problem, the PMF in the output file is a constant for every point, which is impossible.)

How did you obtain the 2D PMF?  If by metadynamics, try plotting it as a heat map or 3D surface.  You should see a small region that is non-constant, but everything else is probably just not sampled yet.

Another question is, could you please tell me when shall we use Multiple-replica metadynamics? It is a little hard for me to understand the goal to run multiple replica of the same system at the same time.

That is only necessary when you have reached the strong scaling limit for a single replica but you still have compute nodes available, so you add more replicas for better statistics.  Coupling the replicas (as opposed to averaging independent runs) is supposed to give you a quicker estimate of the optimal biasing potential to enhance the sampling.

Could you please help me? I have been troubled with the two problems for a long time. 

Thank you very much!

Best wishes, 

Giacomo Fiorin
Associate Professor of Research, Temple University, Philadelphia, PA
Contractor, National Institutes of Health, Bethesda, MD