Error in BS calculations with HSE06 functional: required libint was not linked with CP2K

[Megaa M]

Dear Community,

I'm trying to calculate the band structure of a cubic bulk material using the HSE06 hybrid functional with CP2K 2025.2.

With relaxed structure, I first performed a PBE energy calculation with k-point sampling and obtained a wavefunction file named ENERGY-PBE.kp. I then attempted to use this file as the initial wavefunction guess for the HSE06 calculation.

However, when the HSE06 calculation reads the ENERGY-PBE.kp file and tries to run, CP2K terminates with the following error:

"This CP2K executable has not been linked against the required library libint."

The complete output is attached as cp2k_out.txt.

Interestingly, when I run the same HSE06 calculation without reading the PBE wavefunction file, the calculation starts successfully. However, convergence is much slower and the calculation requires substantially more computational time.

I contacted the staff responsible for the HPC system, but they were unable to determine the cause of the error. They provided details of the installed CP2K and Libint configurations, which are attached as "cp2k_2025.2_libint.txt". Based on the available information, it seems that the hfx related libint appears to absent (Tho the HSE06 calculation without reading PBE.kp file can still run).

Can you please advise whether this problem may be related to the CP2K compilation, the compatibility of the PBE wavefunction file with the HSE06 calculation, or a particular input setting? Is there anything wrong during the installation? Did they miss some files that were supposed to be downloaded? If yes, is there any way for them to get the files?

Much appreciate for any guidance on this.

Kind regards,

Weiyi 

[Frederick Stein]

Dear Weiyi,
I am not quite sure. Officially, CP2K supported Libint version 2.6.0 at that time. It is also not clear why the calculation should fail with restart file but not without because the libint wrapper should be called in both cases. Have you or your admin team the regression test suite after compilation?
Best,
Frederick

[Megaa M]

Hi Frederick,

It seems that the admin team has done the regression test during the installation of the package. They excuted:

make ARCH=Linux-x86-64-gnu-nci VERSION=psmp test TESTOPTS+=\"--mpiranks 4 --ompthreads 1 --maxtasks 4 --timeout 900\" make ARCH=Linux-x86-64-gnu-nci VERSION=psmp test TESTOPTS+=\"--mpiranks 2 --ompthreads 1 --maxtasks 4 --timeout 900\"

Just wondering that do you have any advise for that?

Many thanks !

Cheers,

Weiyi

[Frederick Stein]

Hi Weiyi,

Something seems off. Have you run both calculations with exactly the same executable? Can you show both output files? This kind of behavior (runs smoothly without restart file and fails with missing libraries with the given error message with restart file) cannot happen because CP2K needs to call Libint in both cases and both code required for this behavior paths exclude each other for the same executable. It must have been two different executables (conditional compilation).

I have just tried with a current development version to restart a HF calculation (tests/QS/regtest-kp-hfx-ri-2/diamond_gapw_tc.inp) from the corresponding PBE calculation and was successful. Can you try that?

Best,

Frederick

[Megaa M]

Hi Frederick

Thank you for the advice.

I tried to run the regtest as you suggested, it runs successfully. I checked my input scripts and I guess the error showed up was because I used different basis sets in the PBE and HSE06 input scipts.

Then I changed the basis set used in the PBE calculation, make it aligns with the HSE06 script. However, at this time it gave me errors like "CPASSERT failed". I've uploaded the log files in the attachment. Please have a look.

Just wondering that do you have any idea about this error?

I reckon this happened was probably because I added the auxiliary basis set ADMM in the HSE06 basis set. So I deleted the ADMM basis set and then it returned some errors like "AUX_FIT basis set is not defined." I guess to do this calculation, the ADMM basis set is compulsory. 

Please let me know how you feel.

Cheers,

Weiyi

[Frederick Stein]

Dear Megaa,

This bug is known and was fixed in version 2026.1. So I recommend you to switch to version 2026.1 or 2026.2. If this is not possible, try a different number of MPI ranks or a different value for EPS_DEFAULT (see https://groups.google.com/g/cp2k/c/y_tap__2q4M/m/3LBtjlWdEQAJ?utm_medium=email&utm_source=footer).

Best,

Frederick

[Megaa M]

Hi Frederick,

Thank you for the advice. Will try on 2026 version then.

Cheers,

Weiyi

[Megaa M]

Hi Frederick,

I've switched to cp2k/2026.1 and run the input scripts. Unfortunately, it still returned error like "CPASSERT failed". Then I tried to decrease EPS_DEFAULT from 1.0E-11 to 1.0E-13 as suggested. It can run but terminate at some point. 

For your reference, I've added the log files in the attachment. Please have a look.

Just wondering that do you have any advice on this?

Cheers,

Weiyi

[Stein, Dr. Frederick]

Hi Weiyi,
Do you also have a CP2K output file? Just throwing a `CPASSERT failed` is
not a useful error message (we used it everywhere in the code) and according
to the stack trace, all cores fail within MKL.
Best,
Frederick

Am Tue, 11 Aug 2026 21:17:48 -0700 (PDT) schrieb Megaa M
<chrisz...@gmail.com>:

> --
> 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 visit
>https://groups.google.com/d/msgid/cp2k/df8b39a6-df13-4bbe-b836-48edf68df54en%40googlegroups.com.

Dr. Frederick Stein
Scientific Computing Core
CASUS - Center for Advanced Systems Understanding, Görlitz
Helmholtz-Zentrum Dresden-Rossendorf e.V. (HZDR)

[Megaa M]

Hi Frederick,

Thanks for pointing out. I've added the output and log files in the attachment.

When I used the EPS_DEFAULT = 1.0E-11, the error message showed up with  `CPASSERT failed`. For more details, please refer to the document named 'EPS_DEFAULT_E-11_cp2k_out.txt'. Then I decreased the parameter to EPS_DEFAULT = 1.0E-13, there was no error messages in the out file ('EPS_13_cp2k_out.txt'), but all cores failed within MKL as you mentioned in the previous group chat. Could you please share some suggestions about that? 

Many thanks !

Cheers,

Weiyi

[Frederick Stein]

Hi Weiyi,

Sorry, you need version 2026.2 instead of version 2026.1. The fix was merged soon after the release of 2026.1. Otherwise, I do not know what is going on with MKL. It could be an invalid value which cannot be investigated without any further information or a debug build. Because CP2K was not compiled with support for COSMA, you also cannot try a different library.

Best,

Frederick