Large energy difference between md step and single point calculation

[Massimo Bocus]

Dear CP2K users/developers,

I am extracting some snapshots from an umbrella sampling simulation (with PLUMED as dependency of CP2K 5.1) and I would like to recalculate their energy at a higher level of theory. So to start I performed a single point calculation at the same level of theory of the US simulation (PBE-D3) and, surprisingly for me, I got very different results in the energy. In particular, the energy of the MD step is:

ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.):            -3520.987951048081868

While for the single point calculation:

ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.):            -3520.903314334958850

Which is an enormous difference. Attached the two input scripts, the main differences being the RUN_TIPE, the cell parameters (for the single point they are just taken from the md output at the selected step) and the lack of the MOTION section.

Does anybody have an idea on where this large energy difference could come from?

Thank you in advance, 

Massimo Bocus

[HB H]

Hi  Massimo,

I think in MD simulations, the energy of one step is calculated not only based on the geometry, but also wavefunction of the last step. While the single point calculation only relies on the geometry.

That being said, you should save the wfn files of each MD steps and use them as the restart wfn for single point calculations.(i.e. add "WFN_RESTART_FILE_NAME XXXXX-RESTART.wfn" in "DFT" section)

Let know if I was wrong or not.

Cheers,

Haibei

Massimo Bocus <massib...@gmail.com> 于2020年9月22日周二 下午4:55写道：

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/aa1a788c-5a07-4d01-9a15-c038c0026fcfn%40googlegroups.com.

[Massimo Bocus]

Dear Haibei, 

Thank you for your kind reply. I completely agree with you, however I think that if reading the wavefunction makes the final output energy change by 200 kJ/mol with respect to a calculation "from scratch" then there must be something conceptually wrong in how the code treats the two cases. Moreover, it is completely impossible for me to save the wavefunction for every step as the size of the file is about 30MB and I cannot know in advance which steps I am going to take for the single point calculations.

If you have any other idea on how the problem could be solved, I would be happy to know.

Best regards,

Massimo

You received this message because you are subscribed to a topic in the Google Groups "cp2k" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/cp2k/WvWHl09a5c8/unsubscribe.
To unsubscribe from this group and all its topics, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/CAF%2BoEkroCTdiqY7v%2Biop1KQk7vZtKMWiDWDt%3D7X8hVtytJPN7w%40mail.gmail.com.

[Marcella Iannuzzi]

Dear Massimo

In the MD the reference cell is used, which is quite different from the cell of the single point calculation.

Kind regards

Marcella

[Massimo Bocus]

Dear Marcella,

I used that reference cell parameters as they derive from a static optimization, then they relax and equilibrate during the npt run. Does cp2k actually maintain the plane waves basis set size during the md run so that once I do a single point calculation with the same pw energy cutoff the number of pw is different? And could this in your opinion justify the observed energy difference? 

Thank you very much for the suggestion, 

Massimo

--

You received this message because you are subscribed to a topic in the Google Groups "cp2k" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/cp2k/WvWHl09a5c8/unsubscribe.
To unsubscribe from this group and all its topics, send an email to cp2k+uns...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/27523c6d-ae84-4fc5-9da7-0e66b9c713e4n%40googlegroups.com.

[HB H]

Hi Massimo,

I believe the easiest way to figure out what is going on here, is to run a few MD steps with their wavefunction saved. And try single point calculations with and without wavefunction from scratch.

This should be very quick and looking forward to your reply~.~

Cheers,

Haibei

Massimo Bocus <massib...@gmail.com> 于2020年9月22日周二 下午5:34写道：

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/CAKuX9WnF0urzek8d8dd7CEw3KN1r8%2BBBMGOwL8yH2bZOtsESXQ%40mail.gmail.com.

[qi zhang]

Dear Massimo：

Recently I’m trying to use CP2K and the PLUMED plugin to run umbrella sampling MD and build the PMF curve for my system. I also restraint the distance of two atoms in my system. From your md.inp I see you use plumed at MOTION / FREE_ENERGY / METADYN section,  but when I look over the manual of cp2k, I find METADYN is the section that sets parameters to set up a calculation of metadynamics. I get confused.  Does this mean that the part is designed for metadynamics? Or when using CP2K to run dynamics, just look at the methods in Plumed.dat and don't worry about the limitations of the section? 

Could you please give me some guidance? Thank you in advance.

qzhang

[Massimo Bocus]

Dear Qi,

Normally that section simply imposes cp2k to call plumed in order to calculate the biased forces for the simulation, so it does not really matter what is actually the content of the plumed file and it can be used for any type of biased simulation.

Regards,

Massimo

--

You received this message because you are subscribed to a topic in the Google Groups "cp2k" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/cp2k/WvWHl09a5c8/unsubscribe.
To unsubscribe from this group and all its topics, send an email to cp2k+uns...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/638b43f1-7c24-4738-831f-a39814c91f0cn%40googlegroups.com.

[qi zhang]

Dear Massimo： 

 

   So that's it! Thank you very much for your answer！

qzhang