Most probably what Sikandar proposed should solve the issue. You
should always make sure, that each bin of the spline grid is at least
sampled a few times within each block. If you choose a too big
interval, that leads to problems since the outer parts are not sampled
at all. Also if a block used for averaging contains not sufficient
statistics (for small block sizes) or if there is a gap in the middle
(e.g. due to some very rare cis/trans flip), you will run into
problems with the current implementation. To the background:

For doing the least squares fit of the forces, A matrix is constructed
and inverted to obtain the solution. If now a few grid points in the
spline are not sampled, the matrix contains zero lines and the inverse
does not exist. The "this should be a number" is a very stupid check
whether the inversion has - at least not completely - failed. It
prints an arbitrary entry of the solution, if it's NAN the inversion
obviously failed. This simple check seemed to work most times but for
sure is not sufficient and I guess platform and system dependent.

A more elegant way would be to ensure that the inverse of the matrix
is defined by stripping out the empty lines. Maybe also using the
pseudo-inverse instead of the inverse might help?

Another idea, slightly connected to the above mentioned problem, was
to go for an iterative solver since the scaling of the direct sovler
we currently use is very bad for bigger systems. However, for the
systems we looked at so far, the direct solver was sufficient and it
went to the low priority todo list...

Best,
Victor

2011/9/14 Marc-André Dubois <[email protected]>:
> Again your my savior. Thanks,
> Marc-André
> Le 2011-09-14 à 11:19, Sikandar Mashayak a écrit :
>
> Hey
> Marc-André
>
> Even I am not sure what "This should be a number" output field means?. One
> of the reasons for NAN values in .force file, I think, is due to too small
> value for minimum r in your fmatch.xml file. Many times I was able to
> resolve NAN issue by slightly increasing minimum value of r. Try it and see
> whether that resolves the issue.
>
> --
> sikandar
>
> On Wed, Sep 14, 2011 at 9:49 AM, Marc-André Dubois <[email protected]>
> wrote:
>>
>> Hi guys,
>>               I try to do some force matching… Everything seems to work
>> fine except that the .force file shows a long list of nan values… The "This
>> should be a number: " field in the output of my terminal show a number,
>> which suppose that my files are ok… Any ideas?
>>
>> Marc-André
>>
>> --
>> You received this message because you are subscribed to the Google Groups
>> "votca" group.
>> To post to this group, send email to [email protected].
>> To unsubscribe from this group, send email to
>> [email protected].
>> For more options, visit this group at
>> http://groups.google.com/group/votca?hl=en.
>>
>
>
> --
> You received this message because you are subscribed to the Google Groups
> "votca" group.
> To post to this group, send email to [email protected].
> To unsubscribe from this group, send email to
> [email protected].
> For more options, visit this group at
> http://groups.google.com/group/votca?hl=en.
>
> --
> You received this message because you are subscribed to the Google Groups
> "votca" group.
> To post to this group, send email to [email protected].
> To unsubscribe from this group, send email to
> [email protected].
> For more options, visit this group at
> http://groups.google.com/group/votca?hl=en.
>

-- 
You received this message because you are subscribed to the Google Groups 
"votca" group.
To post to this group, send email to [email protected].
To unsubscribe from this group, send email to 
[email protected].
For more options, visit this group at 
http://groups.google.com/group/votca?hl=en.

Reply via email to