Re: Running bias exchange metadynamics simulations with NAMD

From: Jérôme Hénin (
Date: Mon Jan 17 2011 - 10:35:20 CST

On 17 January 2011 15:40, Ajasja Ljubetič <> wrote:
> Thank you, an interesting reading.
> Crude summary for anyone else reading this:
> On simple systems both method give similar free energy surfaces (FES).
> Metadynamics is designed as a nonequilibrium method, while ABF
> uses equilibrium sampling (from the canonical ensemble).
> So in simple fast-relaxing systems metadynamics explores the FES more
> rapidly, while ABF is more "gentle", but my converge more slowly.
> There is also a difference in memory time of the two methods. ABF has
> "infinite" memory time, but  metadynamics does not (which may or may not be
> an advantage).
> Lastly metadynamics has currently more colvars avalible in NAMD than ABF.

Thanks for the summary!

> I have a question (that my be of no importance what so ever):
> In figure 4B why are there so many accumulated samples around 9 Angstroms? Is
> it because in this region (<12 A) the end to end distance is not a good
> colvar anymore? (and there are also some spikes in 5B)

That's right. The conformational space is more complex there, so the
1D (and even 2D) descriptions are ambiguous: one value of these
coordinates corresponds to several, completely different
conformations. The adaptive bias then converges very slowly and the
system gets "stuck" at these locations.


> On Fri, Jan 14, 2011 at 17:37, Jérôme Hénin <>
> wrote:
>> On 14 January 2011 16:31, Ajasja Ljubetič <>
>> wrote:
>> > Dear Giacomo,
>> > Thank you for a great explanation!
>> > Could I ask another question:
>> > What is the difference between metadynamics and ABF? Or more
>> > specifically,
>> > what are the advantages/disadvantages of each method?
>> Indeed, there are similarities between the two methods. One of the
>> goals of this paper:
>> was to answer your question.
>> Cheers,
>> Jerome

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:19:43 CST