VASP PBE 6.4 pseudopotentials

[janosh]

Re-posted here from @Andrew-S-Rosen's commit comment for better visibility and to solicit further discussion.

@janosh, @esoteric-ephemera: Out of curiosity, are there any planned changes for the PBE .64 set, or is everything going to match the .54 set in terms of the "flavor" of POTCAR chosen? Curious to know what your unofficial recommendations are.

Summary of current state

@Andrew-S-Rosen short answer: no planned changes for which symbol we use for a given element between the 6.4 and 5.4 potentials for now.

Long answer: I'll post here (with permission) a summary of a private email thread with @MichaelWolloch, @sudarshanv01 (VASP developers) and @esoteric-ephemera on the topic.

• PBE 64 potentials are strongly recommended, even just for the lanthanide PSPs which Georg Kresse recreated from scratch after the authors of https://arxiv.org/abs/2305.17274 noticed issues in earlier PSP versions (though the recreation process explicitly did not involve fitting to the AE data in that study)
• @MichaelWolloch performed a recheck of 960 systems with vasp.6.4.1, expecting differences in results due to 6 new potentials. Key findings:

  • 15 out of 60 changed EOSs show less than excellent agreement (eps<=0.06) betwen 54 and 64, with 13 improved and 2 slightly worse compared to AE reference.

  • Significant improvements in all Cu systems and specific He and Ba compounds.

  • Slight deterioration in Rn results compared to AE.

  • @MichaelWolloch ran these checks with the potential mapping list of the paper, so mostly GW PSPs. the improved C_h, N_h, F_h, as well as Cs_sv and the non_h Lanthanides were not part of his testing. See YAML below for exact POTCARs used.

    POTCAR mapping

    RECOMMENDED_ACWF: {
      'H': 'H_GW',
      'He': 'He_GW',
      'Li': 'Li_sv_GW',
      'Be': 'Be_sv_GW',
      'B': 'B_GW',
      'C': 'C_GW',
      'N': 'N_GW',
      'O': 'O_h_GW',
      'F': 'F_GW',
      'Ne': 'Ne_GW',
      'Na': 'Na_sv_GW',
      'Mg': 'Mg_sv_GW',
      'Al': 'Al_GW',
      'Si': 'Si_GW',
      'P': 'P_GW',
      'S': 'S_GW',
      'Cl': 'Cl_GW',
      'Ar': 'Ar_GW',
      'K': 'K_sv_GW',
      'Ca': 'Ca_sv_GW',
      'Sc': 'Sc_sv_GW',
      'Ti': 'Ti_sv_GW',
      'V': 'V_sv_GW',
      'Cr': 'Cr_sv_GW',
      'Mn': 'Mn_sv_GW',
      'Fe': 'Fe_sv_GW',
      'Co': 'Co_sv_GW',
      'Ni': 'Ni_sv_GW',
      'Cu': 'Cu_sv_GW',
      'Zn': 'Zn_sv_GW',
      'Ga': 'Ga_d_GW',
      'Ge': 'Ge_d_GW',
      'As': 'As_GW',
      'Se': 'Se_GW',
      'Br': 'Br_GW',
      'Kr': 'Kr_GW',
      'Rb': 'Rb_sv_GW',
      'Sr': 'Sr_sv_GW',
      'Y': 'Y_sv_GW',
      'Zr': 'Zr_sv_GW',
      'Nb': 'Nb_sv_GW',
      'Mo': 'Mo_sv_GW',
      'Tc': 'Tc_sv_GW',
      'Ru': 'Ru_sv_GW',
      'Rh': 'Rh_sv_GW',
      'Pd': 'Pd_sv_GW',
      'Ag': 'Ag_sv_GW',
      'Cd': 'Cd_sv_GW',
      'In': 'In_d_GW',
      'Sn': 'Sn_d_GW',
      'Sb': 'Sb_d_GW',
      'Te': 'Te_GW',
      'I': 'I_GW',
      'Xe': 'Xe_GW',
      'Cs': 'Cs_sv_GW',
      'Ba': 'Ba_sv_GW',
      'La': 'La_GW',
      'Ce': 'Ce_GW',
      'Pr': 'Pr_h',
      'Nd': 'Nd_h',
      'Pm': 'Pm_h',
      'Sm': 'Sm_h',
      'Eu': 'Eu_h',
      'Gd': 'Gd_h',
      'Tb': 'Tb_h',
      'Dy': 'Dy_h',
      'Ho': 'Ho_h',
      'Er': 'Er_h',
      'Tm': 'Tm_h',
      'Yb': 'Yb_h',
      'Lu': 'Lu',
      'Hf': 'Hf_sv_GW',
      'Ta': 'Ta_sv_GW',
      'W': 'W_sv_GW',
      'Re': 'Re_sv_GW',
      'Os': 'Os_sv_GW',
      'Ir': 'Ir_sv_GW',
      'Pt': 'Pt_sv_GW',
      'Au': 'Au_sv_GW',
      'Hg': 'Hg_sv_GW',
      'Tl': 'Tl_d_GW',
      'Pb': 'Pb_d_GW',
      'Bi': 'Bi_d_GW',
      'Po': 'Po_d_GW',
      'At': 'At_d_GW',
      'Rn': 'Rn_d_GW',
      'Fr': 'Fr_sv',
      'Ra': 'Ra_sv',
      'Ac': 'Ac',
      'Th': 'Th',
      'Pa': 'Pa',
      'U': 'U',
      'Np': 'Np',
      'Pu': 'Pu',
      'Am': 'Am',
      'Cm': 'Cm'
    }

Re new hard Lanthanide PSPs

• @MichaelWolloch noted issues with non-_h potentials in older PSP releases which are now resolved
• the new hard (_h) potentials for lanthanides should be more (chemically) transferable, because they handle short bond length better. they might also transfer better between functionals which could be relevant for MP given same PSPs used for PBE and r2SCAN but that would require careful testing to make sure.
  • highest required ENMAX by any of the hard rare earth PSPs seems to be 773 eV for F_h which isn't that far from the 680 eV ENCUT MP already uses for r2SCAN
• New docs on VASP PSPs are underway (covering when to use which ones), aiming to improve clarity/readability

[Andrew-S-Rosen]

Excellent! Thank you for this wonderful writeup. I greatly appreciate it.

highest ENMAX needed seems to be 773 eV

As a reminder, don't forget that we need to do 1.3*ENMAX for volume relaxes, so this one might hurt a bit.

[JaGeo]

How about the GW potentials in general? In the test against AE, this was mostly used and the VASP wiki also recommends them for accurate computations.

[mkhorton]

Which element is this? I understand the desire to have a constant ENCUT for all calculations/materials, but what is the theoretical basis for this; are we sure this is strictly required?

[esoteric-ephemera]

Agree with @Andrew-S-Rosen here: we'd need a 1000 eV ENCUT for consistency. Per @mkhorton's comment, it's not strictly necessary to have a constant ENCUT across all elements. Some elements will be better-converged at a lower ENCUT than others, and a different chemical environment might require a different ENCUT for similarly-well-converged energies

The vacuum level likely changes with ENCUT

[mkhorton]

How is the vacuum level shifted by ENCUT?

[esoteric-ephemera]

Since the vacuum level is defined as long-range limit of the KS potential, it is also the zero-wavevector limit of the KS potential in reciprocal space. Calculating that quantity numerically requires the charge density (see here, for example), and would have some dependence on the plane-wave cutoff

More or less trying to say: we should just ensure our calculations are well-converged "enough"

[mkhorton]

This is fantastic info, thanks for sharing.I would be in favour of adopting these for sure.To avoid ambiguity, do you have a list of specific pseudopotentials used for these new tests?

[janosh]

@mkhorton @JaGeo @Andrew-S-Rosen Just fyi, I received some more comments and clarifications from @MichaelWolloch. See updated initial post above.

[JaGeo]

@janosh Thank you for the update! Yes, sounds like a very good idea to adopt this.

Any suggestions for the ENCUT based on this initial benchmark?

[mkhorton]

Thanks Aaron. Separate from benchmarking, could it make sense then to have an input set with a sensible default for most elements, and have it be increased only for some? I’d fear a 1000 eV cut off might be quite prohibitive for large systems in the general case.

[esoteric-ephemera]

It might make sense to, but I think benchmarking is unavoidable if we want to see how the hard PSPs compare. 1000 eV would not be feasible for many systems, especially since there are fluorinated organometallic complexes that are huge

[mkhorton]

Thanks Aaron. Yes, just to clarify, was asking more for the final input set rather than the benchmarking.