The force output of RETRAJ module under a r2scan functional is zero

[qin li]

Hi,

I'm trying to train a deep potential model of the MgCl2 solution based on r2scan functional, where the positions and forces are needed. I employed the r2scan functional to rerun the trajectory resulting from the BLYPD3 functional, but the force output of the RETRAJ module comes out to be all zero. However, the RETRAJ module works fine for the BLYPD3 functional. Does anyone have an idea of how to resolve this?

The cp2k version is cp2k2022.1, inputs of the r2scan and force output are as follows and the attachment:

&XC
  &XC_FUNCTIONAL
    &MGGA_X_R2SCAN
    &END MGGA_X_R2SCAN
    &MGGA_C_R2SCAN
    &END MGGA_C_R2SCAN
  &END XC_FUNCTIONAL
  &XC_GRID
    XC_DERIV SPLINE3
    XC_SMOOTH_RHO NONE
  &END XC_GRID
&END XC

&REFTRAJ
  EVAL_ENERGY_FORCES
  TRAJ_FILE_NAME ${REFTRAJ}
&END REFTRAJ

&PRINT
  &FORCES
    &EACH
      MD 1
    &END EACH
  &END FORCES

&END PRINT

[qin li]

The force output of the RETRAJ module works problematiclly in version cp2k2022.1.

qin li <flytomoo...@gmail.com> 于2023年9月7日周四 10:20写道：

[Tianhua Wang]

In fact, according to my tests, this bug exists in almost all CP2K versions later than 6.1, i.e., 7.1, 8.1, 8.2, 9.1, 2022.1, and 2023.2 (I did not test 2023.1), for both SCAN and r2SCAN functionals. Instead, CP2K 6.1 can output the forces normally. It seems to be a systemic problem.

[Krack Matthias]

Hi

 

Add the keyword EVAL_FORCES to the &REFTRAJ section or use the latest CP2K development version after PR2974 has been merged successfully. Alternatively, you can apply the same patch to your local CP2K version.

 

HTH

 

Matthias

 

--
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/760142b3-db53-4c9a-8ca2-882cccc136ban%40googlegroups.com.

[qin li]

Thanks, the keyword EVAL_FORCES  does work.

Krack Matthias <matthia...@psi.ch> 于2023年9月7日周四 21:50写道：

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/ZRAP278MB0827DF9B11ACCAF76AD4DE17F4EEA%40ZRAP278MB0827.CHEP278.PROD.OUTLOOK.COM.

[Ari Paavo Seitsonen (apsi)]

Hello Matthias et co,

  Thank you, it indeed seems to work - this was exactly the same problem that I asked for one week ago.

  Just one question: Why a new keyword, "EVAL_FORCES", has been introduced? Will the "EVAL_ENERGY_FORCES" be removed in the future? And was there a specific reason why the latter did not work since a long time? My test seems to indicate that both of them give the same output, with the present version.

    Greetings from Paris,

       apsi

[Krack Matthias]

Hi Ari

 

The RUN_TYPE ENERGY_FORCE was introduced very early in CP2K. It never worked for me as one would expect intuitively. I guess (without having checked the archives), that the EVAL_FORCES keyword was introduced much later to fix the force output while keeping the EVAL_ENERGY_FORCES keyword, because removing a keyword breaks backward compatibility or might cause unexpected side effects in other parts of the code.

 

Greetings

 

Matthias

 

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/27aff72e-a773-4aa2-b7ae-5b81753aa96en%40googlegroups.com.

[Ole Schütt]

Hi Ari,

I've refactored the input structure and added a new keyword:

  https://manual.cp2k.org/trunk/CP2K_INPUT/MOTION/MD/REFTRAJ.html#EVAL

The old keywords still work (with the known quirks), but are marked as deprecated.

-Ole