NEGF in CP2K

[Michael LaCount]

I noticed in the 2026.1 release of CP2K a new feature 'NEGF: New method to extract the matrix Hamiltonians for electrodes'. I have been running NEGF calculations using an earlier version of CP2K 2024.2, is there a notable performance or accuracy difference with the new method (or from other changes to NEGF)?

I also just found the SMEAGOL NEGF interface option, I am unfamiliar with SMEAGOL and would like to ask if there is any advantage to using that interface vs the internal CP2K NEGF routines?

[Dmitry Ryndyk]

Dear Michael,

One example. If you run regtest-negf-2, you may find the result in regtest-negf-2_2024.png, which is incorrect. New code gives regtest-negf-2_2024.png.

Best wishes,

Dmitry

[Michael LaCount]

Fair enough I would definitely assume the 2026.1 version makes more sense than the 2024 version. For reference, I was interested in what the SMEAGOL interface would give, I translated (to the best of my ability) the regtest-negf-2 job into a set of jobs based on the examples given from "https://chemwiki.ch.ic.ac.uk/wiki/Potential_control_and_current_induced_forces_using_CP2K%2BSMEAGOL". I attached the results here. I may not have used SMEAGOL properly, as there are a number of available keywords for the CP2K+SMEAGOL interface that I didn't use, but the results are fairly close to CP2K internal routines. I would intuitively think the 2026 version of CP2K looks most reasonable between the 3: SMEAGOL, CP2K 2024 and CP2K 2026.

I used the new version of CP2K 2026.1 on a previous job. The force/energy calculations are within numerical noise, but the fermi energy of the contacts and the shift in the HARTREE POTENTIAL were substantial.

Version        Ef Contact          Shift in Hartree

2024.2         0.170                   -0.011

2026.1         0.159                    0.075

Is this behavior consistent with the changes made to the CP2K NEGF routine (with regards to which properties changed more so than how they changed)?