Re: well-tempered metadynamics deforms the water box and the system becomes unstable..

From: Giacomo Fiorin (
Date: Mon Oct 19 2020 - 13:34:20 CDT

Hi Zeynab, in the well-tempered method statistical convergence is
artificially enforced by scaling the newly added Gaussians based not on
simulation time elapsed, but on how many Gaussians have been added at the
same location in the collective variables' space.

The bias will effectively decrease over time, but the exact rate will
depend on how quickly the landscape gets explored: more sampled regions
will have more biasing energy, and their Gaussians will be scaled down more
quickly. So even if you add a new Gaussian on the same spot as the
previous, its weight would be decreased by virtue of the preceding Gaussian
being counted.

If alternatively, you simply want to decrease the weight over time
according to a fixed rate or any schedule of your choice, you may also do
that. For that, I would suggest just changing the weight in the
configuration file. Since you probably don't want to produce many
different Colvars configuration files for each segment, keep just the
variable's definition in an external file (named variable.conf below),
while adding the metadynamics bias directly in the NAMD script and define
the new weight via a Tcl script variable:

# NAMD script
colvars on

# Tcl commands at the end
cv configfile variable.conf
cv config "
metadynamics {
    name mtd
    colvars var
    hillWeight [expr 0.01*exp(-1.0*${rate}*${ijob})]
where "rate" measures the decay rate that you want to have, and "ijob" is
the numeric index of the job.


On Mon, Oct 19, 2020 at 1:39 PM zeynab hosseini <>

> Hi Giacomo,
> Thank you for the quick response and apologies for the delayed response. I
> was trying different hill weights as you suggested that Iā€Œā€Œ noticed the
> hill weights aren't updated as soon as the simulation restarts. More
> specifically, I perform well-tempered metadynamics though consecutive
> simulations each taking 1 nano second. Iā€Œ expect that the hill weight
> reported on the first line of the .hills.traj file to be the same as that
> of the last line of .hill.traj file from the previous nanosecond of the
> simulation, but I see each nanosecond starts from initial hill weight, i.e.
> 0.1 kcal/mol. The value of the collective variable is updated appropriately
> but the hill weight does not. Iā€Œ tried almost all options on the NAMD user
> guide on the metadynamics section to get the hill weight updating but it
> didn't work. That might be due to my ignorance as I'm new in metadynamics--00000000000000ca9705b20a5f94--

This archive was generated by hypermail 2.1.6 : Fri Dec 31 2021 - 23:17:10 CST