Convergence problem

[Soree Kim]

Hi,

I want to know how the VOTCA program calculates the convergence.

And what value is the reasonable cut-off?

Thank you.

Soree Kim.

[Victor Ruehle]

Hey,

by default, the iterations stop after the maximum number of iterations
has been reached. If you want a real convergence check, e.g. if the
distributions match, you will have to add that to your options.xml. In
particular, check the options

cg.inverse.convergence_check.type
and also add convergence to postadd.

I just realized the documentation on that is not very good and I had
to lookup in the code again. I will add a section to the manual.

About the cutoff: there are different oppinions on how to best choose
the cutoff. It depends on the system, we usually choose where the RDF
has a maximum (-> minimum in the potential) and is already close to
one. For water that was around 0.9nm, for methanol around 1.54. You
can check the tutorials to get a feeling for the values we had chosen.
As far as I know there is no general way.

Victor

2012/7/27 Soree Kim <ksr...@gmail.com>:

> --
> You received this message because you are subscribed to the Google Groups
> "votca" group.
> To view this discussion on the web visit
> https://groups.google.com/d/msg/votca/-/0_RVF-7Dh8IJ.
> To post to this group, send email to vo...@googlegroups.com.
> To unsubscribe from this group, send email to
> votca+un...@googlegroups.com.
> For more options, visit this group at
> http://groups.google.com/group/votca?hl=en.

[Soree Kim]

Thank you for your answer.

However, I think I didn't ask clearly.

When I said 'cut-off', it means the value of the convergence.

<convergence_check>default</convergence_check>

<convergence_check_options>

    <limit>0.001</limit>                          <======== This value!

    <name_glob>*.conv</name_glob>

</convergence_check_options> 

I want to know how small the limit should be to obtain reasonable rdf. 
And the first question is also about the calculation of that convergence value.

Thank you.

Soree Kim.

2012년 7월 27일 금요일 오후 7시 33분 13초 UTC+9, Victor Rühle 님의 말:

> votca+unsubscribe@googlegroups.com.

[Sikandar Mashayak]

Hi
Soree Kim

To answer your question I am taking example of spce water coarse-graining by ibi, as in tutorial.

To check for convergence in rdf you can specify following options

<cg>
<non-bonded>
---
---
<inverse>
 ---
---
 <post_add>convergence</post_add>
 <post_add_options>
  <convergence>
  <what>dist</what>
  <weight>1.0</weight>
  </convergence>
</post_add_options>
---
---
</inverse>
</non-bonded>

<inverse>
---

<convergence_check>default</convergence_check>

<convergence_check_options>

    <limit>0.001</limit>

</convergence_check_options>
---
</inverse>
</cg>

with the above options at each step of ibi, new rdf (dist) would be computed and
for convergence check the 1-st norm of ( CG-CG.dist.new - CG-CG.dist.tgt) will be computed
and compared against the limit to decide whether to stop iteration or not.

Now, about the limit value, it depends on what level of accuracy you want from cg potentials.
To get an idea you can check what is this value at diiferent steps of ibi of spce in tutorial and decide on
what would be good enough value for your case.

thanks
sikandar

To view this discussion on the web visit https://groups.google.com/d/msg/votca/-/cKQjLBnghQkJ.

To post to this group, send email to vo...@googlegroups.com.

To unsubscribe from this group, send email to votca+un...@googlegroups.com.