About restarting metadynamics run (in Gromacs)

[Suman Chakrabarty]

Hi,

I have two questions regarding restarting a metadynamics run:

1. Can I change the hill height, width, stride, biasfactor (for well-
tempered metadynamics) etc. (or any other input parameter for that
matter) while restarting the metadynamics run? Will it read all of the
input parameters afresh?

Moreover, will the sum_hills utility read the mixed HILLS file
correctly if there is a change in the parameters in between?

2. Is it important to restart the (interrupted) simulation exactly at
the same step where the last hill was added? I am a little confused
because there can be many methods to restart a simulation in Gromacs,
including checkpoint file system etc. What is the ideal way to restart
a metadynamics run within Gromacs (4.0.5)?


Thanking you,
Suman Chakrabarty.

[Giovanni Bussi]

On Thu, Jul 23, 2009 at 9:46 PM, Suman
Chakrabarty<chakraba...@gmail.com> wrote:
>
> Hi,
>
> I have two questions regarding restarting a metadynamics run:
>
> 1. Can I change the hill height, width, stride, biasfactor (for well-
> tempered metadynamics) etc. (or any other input parameter for that
> matter) while restarting the metadynamics run? Will it read all of the
> input parameters afresh?

Yes you can for all the options of the HILLS or of the WELLTEMPERED
directive. You can even change the width of the hills.

(Obviously, you should not change the definition of the collective variables.)

However, if you change the biasfactor with welltempered metadynamics,
there is a slightly subtle point: the bias is automatically adapted by
plumed at the new value of biasfactor, but the target probability
distribution is changed suddenly. Thus, it could be safer (at least,
this is what I'm doing in a practical case) to first "re-equilibrate"
the system for a while by restarting metadynamics with hills height =
0, then restart adding hills.

This is similar to equilibrating a system at T1, and then restarting
at T2: you would discard the statistic coming from the first part of
the T2 run. Since in metadynamics the history is used to optimize the
potential, it would be better to discard the history immediately
following the change in the biasfactor.

>
> Moreover, will the sum_hills utility read the mixed HILLS file
> correctly if there is a change in the parameters in between?

Yes. This is the reason why plumed is writing all the hills parameter
at every line of the HILLS file: to use them later!

>
> 2. Is it important to restart the (interrupted) simulation exactly at
> the same step where the last hill was added? I am a little confused
> because there can be many methods to restart a simulation in Gromacs,
> including checkpoint file system etc. What is the ideal way to restart
> a metadynamics run within Gromacs (4.0.5)?

If you restart "rarely", it is irrelevant. Of course, if the atomic
configuration is not as up-to-date as the HILLS, you will slightly
overestimate the bias in the region where the system is at the moment
of restart (since you will add twice the hills). However, this error
is expected to be small, and is going to be smeared out as the
simulation proceed.

Of course, if you restart very often the error will not be negligible.

Giovanni

>
>
> Thanking you,
> Suman Chakrabarty.
>
> >
>