[gmx-developers] g_sdf, who developed it?

[Ondrej Marsalek]

Hi,

sorry, I missed this thread when it was alive. I would just like to clarify.

On Thu, Jun 3, 2010 at 08:18,  <chris.neale at utoronto.ca> wrote:
> - g_sdf does a fitting based on (I think) 3 atoms defined by the user.
> - g_spatial requires the user to do all of the fitting via trjconv.
>
> In my opinion, it is easier and more robust to require the user to first
> pre-process the trajectory with trjconv. Nevertheless, I see more users on
> the list inquiring about g_sdf; It is impossible to discern if this is
> because g_sdf is harder to use or because it is more often used.
>
> The reasons to use g_spatial, are that one might want to fit based on some
> arbitrary routine, and this should always be possible via trjconv. In my
> opinion, the only thing that favours g_sdf is that one doesn't need to have
> the disk space to store a fit trajectory prior to running g_sdf -- and this
> could be a big plus for g_sdf in some cases.

The crucial difference here is that g_sdf can work with multiple
triples of reference atoms, like g_rdf can do with multiple reference
atoms. g_spatial, while useful for some tasks, does distribution in
absolute space and therefore requires the preprocessing mentioned
above. If I am interested in the SDF of one atomtype around water
molecules, I can get this from one run of g_sdf. With g_spatial, this
would mean having to do one trajectory fitting (and one run of
g_spatial) for each reference water molecule that I want contributing
to my sampling. Plus some postprocessing to combine the distributions.

Best,
Ondrej