Not able to read forces into TAMkin from cp2k.

[John Low]

I am using branch 2.3 of cp2k.  TAMkin can not find the forces in the cp2k output.  The tamkin.io.cp2k.load_molecule_cp2k function is looking for a "Forces|" flag which is not in the cp2k output.  It appears that this function was written for a different version of cp2k.  Which version of cp2k will generate output compatible with TAMkin?

[Toon Verstraelen]

Hi John,

That routine was written about three years ago, so it may need to be
updated. For the single-point input file one must activate the
computation of the forces. (ENERGY_FORCE job in cp2k) I see that this is
not explicitly mentioned in the doc strings, so that may also be the
reason for this error.

Best Regards,

Toon

On 12/24/2012 03:25 PM, John Low wrote:
> I am using branch 2.3 of cp2k. TAMkin can not find the forces in the
> cp2k output. The tamkin.io.cp2k.load_molecule_cp2k

> <http://molmod.github.com/tamkin/reference/io.html?highlight=cp2k#tamkin.io.cp2k.load_molecule_cp2k>function

> is looking for a "Forces|" flag which is not in the cp2k output. It
> appears that this function was written for a different version of cp2k.
> Which version of cp2k will generate output compatible with TAMkin?
>

> --
> You received this message because you are subscribed to the Google
> Groups "TAMkin" group.
> To post to this group, send email to tam...@googlegroups.com.
> To unsubscribe from this group, send email to
> tamkin+un...@googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msg/tamkin/-/MGMQFmmYArQJ.
> For more options, visit https://groups.google.com/groups/opt_out.
>
>

[John J. Low]

Toon,

Cp2k has been modified so the the "FORCE|" string is not include the force output, even if you run an "ENERGY_FORCE" job. I have reviewed the cp2k source and I can find the old routines which wrote the "FORCE|" string. Unfortunately, cp2k has been modified so that these routines are no longer used to write out the force. It is not a simple fix to use the old routines because they involve a complicated usage of modules.

Unfortunately, I got distracted by errors in the forces generated by CP2K, a new supercomputer, and vacation. I have not been able to modify TAMKIN or cp2k to get useful thermodynamics.

I am planning to modify cp2k to be compatible with TAMKIN, it seems like the simplest path for me. Although it would not help anyone else, unless the cp2k authors include the modification to their source.

Ultimately, the best fix would be to modify TAMKIN. Which I might do when I get to it. Could you fix TAMKIN, if I sent you some sample input and output from cp2k?

By the way, the errors in forces from cp2k are pretty significant if you don't use a huge cutoff for the plane wave basis in the Gaussian Plane Wave (GPW) approximation. The Gaussian Augmented Plane Wave (GAPW) approximation appears to be better but one needs to fiddle with the augmenting gaussians to get acceptable noise in the forces.

Thanks for the reply.

John J. Low
Computational Scientist-STA
Mathematics and Computer Science
Building 240, 1.G.10
9700 South Cass Avenue
Argonne National Laboratory
Argonne, IL 60439.
630-252-0045

[Toon Verstraelen]

On 01/09/2013 04:35 PM, John J. Low wrote:
> Toon,
>
> Cp2k has been modified so the the "FORCE|" string is not include the force
> output, even if you run an "ENERGY_FORCE" job. I have reviewed the cp2k
> source and I can find the old routines which wrote the "FORCE|" string.
> Unfortunately, cp2k has been modified so that these routines are no longer
> used to write out the force. It is not a simple fix to use the old
> routines because they involve a complicated usage of modules.
>
> Unfortunately, I got distracted by errors in the forces generated by CP2K,
> a new supercomputer, and vacation. I have not been able to modify TAMKIN
> or cp2k to get useful thermodynamics.
>
> I am planning to modify cp2k to be compatible with TAMKIN, it seems like
> the simplest path for me. Although it would not help anyone else, unless
> the cp2k authors include the modification to their source.
>
> Ultimately, the best fix would be to modify TAMKIN. Which I might do when
> I get to it. Could you fix TAMKIN, if I sent you some sample input and
> output from cp2k?

Hi John

I can definitely do that. If possible, please send files for a rather small
system (both single point and frequency). That would allow me to include them
in the unit tests.

> By the way, the errors in forces from cp2k are pretty significant if you
> don't use a huge cutoff for the plane wave basis in the Gaussian Plane Wave
> (GPW) approximation. The Gaussian Augmented Plane Wave (GAPW)
> approximation appears to be better but one needs to fiddle with the
> augmenting gaussians to get acceptable noise in the forces.

Aha, that is interesting. It will probably also affect the computation of the
Hessian as it is obtained with finite differences from several force computations.

B.t.w., the reason we read in the forces is to make sure that certain NMA
methods are applicable. Some of these NMA methods may give weird results if
the Hessian is not computed at a stationary point.

Best Regards,

Toon

[John J. Low]

Toon,

This is the test I have been working with (disilic acid plus a water molecule) is it small enough? See attached files.

Yes! I am aware that not being at a stationary point will cause problems.

It looks like the geometry in final.xyz is not a local minimum judging by the frequencies in the cp2k output.

I believe that the geometry is converged to a stationary point as well as can be expect for this cutoff, pseudopotential etc.

I hope that this case is converged well enough to test the tamkin's input routines, even if the predicted thermodynamics might be bogus.

John

----- Original Message -----
From: "Toon Verstraelen" <Toon.Ver...@UGent.be>
To: tam...@googlegroups.com
Sent: Wednesday, January 9, 2013 9:58:26 AM
Subject: Re: [tamkin] Not able to read forces into TAMkin from cp2k.

[Toon Verstraelen]

Hi John,

Thanks for the files. I managed to let the CP2K routine work with both
the old and the new files. Could you try the update code on github? I
have also changed the load_cp2k routine. It no longer requires the XYZ
file, which should make it easier to use.

Best Regards,

Toon