RE: Linear force ramp increments in ABF

From: Matthew Ralph Adendorff (mraden_at_mit.edu)
Date: Thu Aug 29 2013 - 14:56:30 CDT

Hi Jerome

I will definitely have a look at the code. In contrast, would the Metadynamics method follow a more uniform (dare I say more trivial) addition of force to the system in the form of controlling the hillWeight and hillWidth variables?

Best,

Matthew
________________________________________
From: Jérôme Hénin [jerome.henin_at_ibpc.fr]
Sent: 29 August 2013 03:15 PM
To: Matthew Ralph Adendorff
Cc: namd-l_at_ks.uiuc.edu
Subject: Re: namd-l: Linear force ramp increments in ABF

Hi Matthew,

If you are into coding, the calculations are done in the file colvarbias_abf.C - look for the variables full_samples and min_samples. What happens to the biasing force over time is non-trivial, as the coordinate may move between bins at the same time as the value of the multiplying factor changes in each bin.

Cheers,
Jerome

----- Original Message -----
> Hi Jerome
>
> Many thanks for the information, it will be highly useful. Might I
> ask how the value of F is calculated? From the formulation in the
> NAMD manual it appears to be determined from the running estimate of
> the free energy gradient at the current point in the collective
> variable subspace. Therefore, would the value of F be defined at
> FS/2 from the gradient estimate and then slowly ramped up to F over
> the n/2 remaining time steps? Or is the target F continually
> refined? What I am essentially looking for is a way to put an upper
> bound on the amount of force added per time-step increment to assess
> the impact this force ramp will have on the slowly relaxing degrees
> of freedom in the structure, thereby getting a good value for
> fullSamples in subsequent runs.
>
> Thanks again for the assistance,
>
> Matthew
> ________________________________________
> From: Jérôme Hénin [jerome.henin_at_ibpc.fr]
> Sent: 29 August 2013 01:06 PM
> To: Matthew Ralph Adendorff
> Cc: namd-l_at_ks.uiuc.edu
> Subject: Re: namd-l: Linear force ramp increments in ABF
>
> Hi Matthew,
>
> To be precise, the ramp is linear in the number of samples in the
> bin, and starts at half of fullSamples.
>
> If FS is the value of fullSamples, the current number of samples is
> n, and the full biasing force is F, the force that is actually
> applied is:
>
> n < FS/2 : 0
> FS/2 < n < FS : (n - FS/2) / (FS/2) * F
> n > FS : F
>
> Best,
> Jerome
>
>
> ----- Original Message -----
> > Good day,
> >
> > During an ABF run, I understand that once the minimum number of
> > samples has been achieved for a given bin the biasing forces are
> > added following a linear ramp. During this ramp is the force
> > applied
> > in discrete increments of a fixed value per time-step or does the
> > applied quantity vary? If so, what is this increment or is there a
> > procedure to determine how much force is added per step in the
> > linear ramp?
> >
> > Many thanks for the support.
> >
> > Best,
> >
> > Matthew
> >
> >
>

This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:21:35 CST