Re: Clarification regarding some finer details in ABF simulations

From: Jerome Henin (
Date: Tue Aug 18 2009 - 16:01:56 CDT


Yes, that i what should happen one enough samples are collected. More
precisely, the algorithm does not specifically aim for a particular
amount of sampling or histogram. It just applies a time-dependent bias
computed in a certain way (independent of the current sampling
histogram). Then, the rules of stat mech state that in the long run,
sampling should be uniform.


On Tue, Aug 18, 2009 at 11:01 AM, Hugh Martin<> wrote:
> Thanks indeed, a small follow up question: Once the reaction coordinate is
> sufficiently sampled (according to the parameter 'abf fullSamples'), and the
> simulation is still running (due to a high 'numsteps' parameter), will the
> system then attempt to evenly add further samples? From my simulations this
> appears to be the case, but I wanted to be sure. This then appears to be a
> trivial way of sampling the reaction coordinate further, rather than
> drastically increasing the 'abf fullSamples' parameter, which the paper
> "exploring the free-energy landscape of a short peptide using an average
> force" indicates can be problematic, if I've understood correctly.
> Many thanks,
> Hugh
> 2009/8/15 Jerome Henin <>
>> Hugh,
>> Sorry for the delayed reply.
>> On Mon, Aug 10, 2009 at 1:33 PM, Hugh Martin<> wrote:
>> > Hello,
>> >
>> > I was hoping someone could clarify a few minor points regarding ABF
>> > simulations, after reading related papers and the documentation in
>> > detail
>> > there are a couple of minor points still unclear to me:
>> >
>> > 1. The user guide states that parameter forceConst is applied at xiMin
>> > and
>> > xiMax (the borders of the reaction coordinate) and not in between them.
>> > My
>> > simulations show that this force is acting within the xiMin and xiMax
>> > boundaries (along the whole reaction coordinate), this is not supposed
>> > to
>> > occur, correct?
>> No, it should not occur. I would need more details to understand
>> what's happening. Incidentally, you seem to be using the NAMD 2.6
>> version of ABF, which is deprecated now. If you run into more
>> problems, you may consider switching to the new code. You will also
>> get better performance out of it.
>> > 2. When the ABF force is applied, the moving atom makes its way to the
>> > next
>> > reaction coordinate bin via diffusion, it moves mainly due to diffusion
>> > because at this point the net force acting on the atom is roughly zero
>> > when
>> > considering both the free energy landscape and the applied ABF force.
>> > Once
>> > the atom is in the next bin, the applied ABF force is 'switched' off,
>> > does
>> > this imply that the moving atom could then diffuse out of the bin in
>> > either
>> > direction before sufficient sampling has taken place?
>> Yes, the ABF coordinate may always diffuse either way.
>> > Or will the ABF force
>> > be reapplied if the atom leaves the bin before sufficient sampling has
>> > occured?
>> Sorry, you lost me there... If it moves to a bin where the sampling is
>> sufficient, then the bias will be applied.
>> > 3. As for the 'sampling' itself (abf fullSamples), I am unclear how the
>> > scripts determine the frequency at which a 'sample' is taken? I was
>> > expecting this to be a controllable parameter, but it appears that this
>> > is
>> > not the case. I presume that a snapshot measurement of the force acting
>> > along the reaction coordinate is taken, and I'm currently assuming that
>> > these snapshots are taken at regular time intervals, is this the case?
>> Yes, the force samples are measured at every step. This is a sensible
>> thing to do because the force fluctuations have a large, fast
>> oscillating component that averages out quickly if frequent samples
>> are taken.
>> Best,
>> Jerome

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:53:12 CST