Re: Issues with 'Gyration' in Colvars module

From: Jérôme Hénin (jhenin_at_ifr88.cnrs-mrs.fr)
Date: Fri May 25 2012 - 11:08:58 CDT

Hi Shaon,

As Giacomo noted, there is a discrepancy between what's implemented
and what is documented - this probably escaped us because our typical
test cases have atoms with identical masses. That may well be the
cause of your problem, because the Jacobian used for the ABF
implementation is definitely that of the unweighted function. For this
kind of variable, the Jacobian correction always yields a biasing
force that pushes toward small values, and increases as 1/Rg.

Did you include hydrogen atoms in your definition of Rg? That would
make it more sensitive to this issue.

In any case, the code calculating this should be changed. We have a
fix in the works: the mass weighting will become optional to
accommodate all cases. In the meantime, I am attaching a source file
that disables it, if that's any help.

Best,
Jerome

On 25 May 2012 03:53, Shaon Chakrabarti <shaonc_at_gmail.com> wrote:
> Hi Aron, Giacomo
> Thanks so much for all the suggestions...Problem 1) is almost
> resolved, but Problem 2) is still creating problems.
> Here's what I find:
>
> PROBLEM 1)
> I defined Rg using the mass weighting definition as follows:
>
> Rg = sqrt[ 1/M * sum over all N atoms
> {m*(x-COMx)^2+m*(y-COMy)^2+m*(z-COMz)^2} ]
> and similarly for the COG case.
>
> Now, Rg (COM) is 5.5785031 while Rg (COG) is  5.5799608
>
> So indeed, Rg (COG) defined with the mass-weighted formula matches
> the colvars.traj result, as Giacomo had said. However, the formula
> most closely matching the VMD rgyr result seems to be the one
> with COG (without mass-weighting)! I will check once again to see
> if I'm making a mistake however.

I think

>
> PROBLEM 2)
> This is still causing problems. First let me mention that my system
> is a 25-mer alkane in water, so though it does have a global minimum
> at around Rg = 4 angstroms, it is expected to collapse from an extended
> state on a timescale of 1 ns. Without any biases, that's precisely what I
> observe in my simulation. I've applied HARMONIC biases (with spring
> constant as low as 1 kcal/mol.A^2) to the Rg colvar with perfectly fine
> results.
> I have also used ABF with the 'distance' component (order parameter
> end-to-end distance) and the results are fine, even with wall constant 10--20cf307f362ee3b1db04c0de9673--

This archive was generated by hypermail 2.1.6 : Tue Dec 31 2013 - 23:22:00 CST