Re: abf: multiple distance

From: Jerome Henin (jhenin_at_cmm.chem.upenn.edu)
Date: Fri Nov 30 2007 - 09:01:39 CST

Hi Luca,

On Nov 30, 2007 4:22 AM, luca <bellucci14_at_unisi.it> wrote:
> On Friday 30 November 2007 00:37:54 you wrote:
> Hi Jerome,
> Thank for replay.
> > Dear Luca,
> > If your "more advanced reaction coordinate" is a single variable that
> > depends on several distances, you just have to work out a few
> > equations (mostly the gradient of that variable in Cartesian
> > coordinates). I can point you to a few papers if you need it.
> I am interested on this argument. I have been reading some article this week , but
> I think that this is a difficult work to implement ...in fast and easy way.

That is true. Implementing new coordinates takes some time and effort,
except in the simplest cases. That is why we encourage people who
write them to contribute them so that others in the community do not
duplicate the work.

> > If what you really want is to define a multidimensional free energy
> > surface, function of several distances, then the NAMD ABF code cannot
> > currently do that.
> Yes, I'm sorry I had this in mind .
> > We are working on making this possible in future
> > versions, but there is no release schedule yet.
>
> Ok. I will see http://www.edam.uhp-nancy.fr/ABF/index.html
> At this moment I can run multiple md with different reaction coordinates and
> make FES manually. Do You have a reference/suggestion for this approach?

I can see two cases:
1) The different coordinates are not significantly correlated. In this
case, the free energy will be additive:
A(x1, x2, x3) = A1(x1) + A2(x2) + A3(x3)
and all the information is contained in the 1-D profiles.

2) The coordinates are correlated. That can still be treated if you
can define discrete states for most of the coordinates, e.g. if the
1-D profiles have a small number of free energy basins. Then you could
compute 1-D PMFs with all other coordinates restrained to a specific
state, and repeat this procedure for all states you consider
interesting.
Of course, this only works if you have a good intuitive understanding
of the system already, and it gets impractical if there are many
coordinates/states, but it may help get "semi-quantitative" results.

Jerome

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:45:37 CST