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

From: Giacomo Fiorin (
Date: Wed Oct 21 2020 - 14:08:35 CDT

Hi Zeynab, thanks for checking.

I presume that you generated this using a NAMD prior to 2.14. There should
not be two Gaussian hills added at the same step, i.e. step 500000. The
Colvars version included in NAMD 2.14 corrects for this error.

Given your setup the error results in one extra Gaussian hill added every
2000 hills, which on average is similar to using 0.1 * 2001/2000 = 0.10005
kcal/mol as initial weight. I don't think that this will affect
convergence significantly for as long as you are adding the bias slowly
enough to get a PMF (either with standard metadynamics or well-tempered).

The second hill is slightly smaller than the previous one due to the
well-tempered scaling, so that is fully expected.


On Wed, Oct 21, 2020 at 2:07 PM zeynab hosseini <>

> Hi Giacomo,
> Thank you so much for your guidance.
> On Wed, Oct 21, 2020 at 11:26 AM Giacomo Fiorin <>
> wrote:
>> Hi Zeynab, the bias isn't correctly restarted because there is no command
>> in there to load the state file.
>> Quite simply, that tutorial shows how to do a quick metadynamics run for
>> demonstration purposes, so it did not provide you with instructions on how
>> to restart it. You can find these here:
>> and pretty much it boils down to using either "colvarsInput" or "cv load
>> ...." in the NAMD script to give restart information to Colvars.
> I did as you said; including the "colvarsInput" command in the namd .conf
> file to load the state file from previous simulation:
> colvars on
> colvarsConfig
> colvarsInput cnt-ser_meta_29.colvars.state
> I checked for the .hills.traj outputs of two dependent jobs. Here is the
> result where I compared the last line of the .hill.traj file for the 29ns
> and the first line of the same for the 30ns:
> -tail cnt-ser_meta_*29*.colvars.meta-distance.hills.traj
> 499000 2.31048448549163e+01 4.00000000000000e-03
> 3.25361527452210e-02
> 499250 2.35892630640388e+01 4.00000000000000e-03
> 8.04196673987633e-02
> 499500 2.26889661523101e+01 4.00000000000000e-03
> 9.33115557260331e-02
> 499750 2.31630224461575e+01 4.00000000000000e-03
> 3.87756645748246e-02
> 500000 2.30916666324611e+01 4.00000000000000e-03
> 4.50554516054642e-02
> -head cnt-ser_meta_*30*.colvars.meta-distance.hills.traj
> 500000 2.30916666324611e+01 4.00000000000000e-03
> 4.27859620060334e-02
> 500250 2.30432359543864e+01 4.00000000000000e-03
> 5.27299888358347e-02
> 500500 2.30443231967936e+01 4.00000000000000e-03
> 5.27151736910929e-02
> 500750 2.34415977600652e+01 4.00000000000000e-03
> 3.45533862110254e-02
> As it is seen, the hill weight on the 500000th time step (30ns) is not
> started from the first step hill weight (0.1 kcal/mol) but it changed a
> little from 29ns to 30ns. For 29ns, it is 4.50554516054642e-02, and for the
> 30ns, it is 4.27859620060334e-02. I wonder why they aren't exactly the same
> and if this affects the convergence.
>> This information is a bit harder to find in the NAMD manual, because the
>> entire Colvars doc is crammed into one section of the NAMD manual. Also,
>> in the NAMD website the links to the same sections are not consistent
>> between NAMD versions, which throws off the search engines when searching
>> the mailing list archives. For this reason, using the Colvars doc page
>> linked above or the PDF version
>> <> would be more
>> appropriate.
> Thank you.
> All the best,
> Zeynab

This archive was generated by hypermail 2.1.6 : Thu Dec 31 2020 - 23:17:14 CST