OT/Supercell vs. Kpoints: Abnormal pressure difference

[Nicholas Winner]

Hello all,

I performed two cell optimizations on bulk Si, the first with a primitive cell and kpoints and the second using OT and a 250 atom supercell. Both structures started from the Materials Project (mp-149) lowest energy Fm-3m polymorph.

I found that while OT only took 3 cell opt steps, the kpoint calculation took 10 steps, which I found a little strange.

Upon investigation I found two things: 

(a) the kpoint calculation, had a large initial pressure deviation of -92570 bar, compared to OT, which had an initial deviation of  -708 bar at DZVP level and -2934 at TZV2PX level. Again these are the same structure, just kpoints vs supercell. It is a little odd to me that DZVP is closer to converged pressure, but I don't think its too strange.

(b) the kpoint calculation relaxed from Fm-3m to Im-3m space group, which is not the ground state polymorph.

My question is whether or not this issue is due to CP2K's kpoint implementation, or if this is something more fundamental, and how (if at all) it can be resolved. Below is the abridged DFT section for the kpoint calculation, which I assume is where the problem lies, but I can provide more info if needed.

Best,

Nick

&DFT

                BASIS_SET_FILE_NAME BASIS_MOLOPT

                BASIS_SET_FILE_NAME BASIS_MOLOPT_UCL

                POTENTIAL_FILE_NAME GTH_POTENTIALS

                UKS True

                MULTIPLICITY 0

                CHARGE 0

                &QS

                        METHOD GPW

                        EPS_DEFAULT 1e-12

                        EXTRAPOLATION PS

                &END QS

                &SCF

                        MAX_SCF 100

                        EPS_SCF 1e-07

                        SCF_GUESS RESTART

                        MAX_ITER_LUMO 400

                        ADDED_MOS 100

                        &DIAGONALIZATION

                                &DAVIDSON

                                        PRECONDITIONER FULL_ALL

                                &END DAVIDSON

                        &END DIAGONALIZATION

                        &MIXING

                                METHOD BROYDEN_MIXING

                                ALPHA 0.2

                                NBUFFER 5

                        &END MIXING

                        &SMEAR

                                ELEC_TEMP 300

                                METHOD FERMI_DIRAC

                                FIXED_MAGNETIC_MOMENT -100.0

                        &END SMEAR

                &END SCF

                &KPOINTS

                        SCHEME MONKHORST-PACK 13 13 13

                        EPS_GEO 1e-06

                        FULL_GRID False

                        PARALLEL_GROUP_SIZE -1

                        SYMMETRY False

                        UNITS B_VECTOR

                        VERBOSE True

                        WAVEFUNCTIONS COMPLEX

                &END KPOINTS

&END DFT

[Lucas Lodeiro]

Hi Nicholas,

I am seeing some definitions which are not totally comparable. For example, the primitive cell of Si has 2 atoms, and if you use a 13x13x13 kpoint grid, the comparable supercell using just the gamma point is a 13x13x13 supercell containing 4394 atoms. As you mention, for gamma point + supercell calculation, you use a cell with 250 atoms (5x5x5 supercell) which is comparable with a primitive cell calculation with 5x5x5 kpoint grid.

Abot the difference using different basis sets, well is not weird, the stress tensor calculation is highly dependent on the scf convergence and the accuracy of the wavefunction. 

Another thing that caught my attention was the OT vs diagonalization comparison. Both probably give a pretty similar solution, but no equal... I would f¿prefer compare the primitive and supercell calculation using the same wavefunction solver, especially if you use smearing, which is not implemented for OT method, and could be a source of discrepancies (albeit in a semiconductor this is not so important, if you are using a GGA or LDA functional, the band gap is highly underestimated and the smearing affects the calculation).

Why are you using UKS instead of RKS?

Regards - Lucas Lodeiro

--
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/b7c861a2-eb23-4fac-8a42-9846e53f6f24n%40googlegroups.com.

[Nicholas Winner]

Hi Lucas, good points you made.

The biggest thing that surprised me was not that 250 gamma point was not equivalent to 13x13x13 kpoint calculation on a primitive cell. I expected some difference. The biggest surprise was that the OT calculation converged to something very similar to Materials Project, while the kpoint calculation converged to a complete different polymorph. The Materials Project structure was calculated with kpoint grid and high plane wave basis cutoff, which confirmed Fm-3m is the ground state polymorph in agreement with experiments. While Materials Project's VASP calculations should not be equivalent to CP2K, with high quality basis/cutoffs/and the same kpoint grid, the results should be similar, as they both use good pseudopotentials, but it wasn't.

That's why I thought there had to be something off about how I was using the kpoints in CP2K. 

Also I just forgot to turn UKS off. Probably wasted some core-hours, but won't make the results worse.

Any further thoughts?

-Nick

[Lucas Lodeiro]

Hi, 

without a full input and output is difficult to say more. But, I insist with the smearing + UKS... maybe the system is creating an abnormal magnetic state. Try with RKS + kpoints and without smearing. I do not believe that PW programs make a huge difference with CP2K. If you want to try, CP2K has the SIRIUS package, which computes electronic structure as Quantum ESPRESSO. I tried it later, and gave pretty similar results with QE and VASP.

Regards

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/e071d68c-279e-4028-83cb-6d9f73662270n%40googlegroups.com.