Re: Trouble using rotateReference, centerRefrence and refPositionsGroup in the colvar module

From: JťrŰme Hťnin (
Date: Fri Feb 03 2012 - 05:20:31 CST

Here is a complement to Giacomo's remarks.

On 2 February 2012 16:06, Ajasja Ljubetińć <> wrote:
>> Regarding why the PDB file doesn't work, you need to match the number of
>> coordinates of the atoms that you use for fitting (refPositionsGroup), not
>> the number of atoms of group1.  The problem in providing a file with the
>> whole system is that the PDB reader will by default read the coordinates of
>> group1, not those of refPositionsGroup.
> Aha, I thought that it would take the indices of the refPositionsGroup and
> only load those from the PDB.

And you were right, it should, but it doesn't due to a bug that
Giacomo just discovered (and that was my code too, mea culpa). That
will work as documented in future revisions of the code.

> From the manual under  rotateReference/centerRefrence:
> "Note:the derivatives of the colvars with respect to the
> rotation/translation are usually neglected (except
> by rmsd and eigenvector)."
> What exactly does this mean? Does this come into play when
> the Cartesian biasing force is calculated (eq 4) from the colvar derivative?
> (BTW there is still a small typo in that equation - the gradient should  be
> along Cartesian coordinates and not along colvars space).

When a colvar depends on rotated coordinates, its gradient in the lab
frame is the "counter-rotated" gradient in the rotated frame, plus
another term that depends on derivatives of the optimal rotation with
respect to Cartesian coordinates of the fitted atoms. That is handled
explicitly in the RMSD and eigenvector coordinates, which are require
fitted coordinates by nature, but it is awfully complicated to handle
in a general way. So in other variables, the second term is neglected,
and empirically it seems to work but it could become non-negligible in
some non-intuitive cases.


This archive was generated by hypermail 2.1.6 : Tue Dec 31 2013 - 23:21:37 CST