Nexus: Update QE Input for New Hubbard+U Syntax

[hgriffin-ux]

Problem: Starting with Quantum Espresso (QE) version 7.1 (mid-2022 release), Hubbard+U calculations have a new required syntax. The current version (qmcpack 4.1.0) of the QE input conversion script for Nexus (pwscf_input.py) produces the older syntax. When versions of QE >= 7.1 run the Nexus generated input, the calculation crashes with a message that says: “DFT+Hubbard input syntax has changed since v7.1”.

Solution suggested: Could the Nexus QE input script get updated to produce the newer Hubbard+U input file syntax if detects QE versions >=7.1?

Alternatives: in the meanwhile for my current project, I am planning to run QE separately from Nexus and then get the Nexus script to read in the QE output data.

[kayahans]

Hello,
Nexus should be able to produce QE input files using the newer Hubbard +U syntax. There is an example in
nexus/examples/quantum_espresso/02_diamond_self_consistent_U that uses the new syntax. Please check that example and its README file for more information.

Nexus uses the hubbard tag for the newer QE syntax, instead of the legacy hubbard_u, which corresponds to older QE versions. Keep in mind that the two classes of tags — the legacy version (hubbard_u, hubbard_j0, hubbard_j, U_projection_type) and the newer version (hubbard, hubbard_proj) — should not be used at the same time. For example, if you use both hubbard_u and hubbard_proj in the generate_pwscf function, Nexus will raise an error. The hubbard tag populates the Hubbard card in the pw.x input, while hubbard_u writes lines such as hubbard_u(1) = X.X under the &SYSTEM namelist.

For reference, here are some example input formats accepted by hubbard:

hubbard = {'V': {('C-2p', 'C-2p'): 1e-8}}
# Other use examples
# hubbard = {'V': {('C-2p', 'C-2p'): [{'indices': (1, 2), 'value': 1e-8},
#                                    {'indices': (2, 1), 'value': 1e-8}]}}
# hubbard = {'U': {'C-2p': 1.0}, 'V': {('C-2p', 'C-2p'): 1e-8}}

Another important note: New hubbard format is more flexible in the sense that hubbard operators can act on different orbital manifolds (s,p,d). This means that the manifold names in the pseudopotentials should follow the QE nomenclature exactly. For example, for the C-diamond example to work, I have modified the C.BFD.upf file converting the “0s”->“2S” and “0p”->“2P” in the PP_HEADER and PP_PSWFC sections. There is an open PR on this if you want to follow its process, where these changes will be made on all the PP in pseudopotential library: QMCPACK/pseudopotentiallibrary#105 . For now though, you would need to make these changes yourself, by hand unfortunately.

Currently, if you want to use a V parameter, these examples above will give an incomplete assignment because Nexus only loops over atom pairs inside the primitive cell. For a proper +V assignment, pairs that include atoms in periodic image cells must also be considered.

For example, suppose you want to add a Ni–d / O–p V interaction in a conventional NiO cell, between every Ni atom and its first-nearest O neighbors. For a given Ni, some of the neighboring O atoms will be in image cells. In this context, QE treats each Ni–O pair as unique (i.e., it does not apply symmetry), so each pair must be written explicitly for the V interaction. This requires knowing the index numbers for each Ni–O pair, which Nexus cannot determine automatically.

To obtain these indices, you need to perform a quick hp.x calculation (which Nexus can run) and then use the results of that run as a dependency for the subsequent QE calculation, where the index numbers are transferred. More details can be found in the hp.x paper to better understand the procedure (https://arxiv.org/pdf/2203.15684). I realize this may not be directly relevant to your use case, but I wanted to provide additional context on what Nexus can currently do with respect to +V, as it may be helpful for others reading this issue in the future.

[prckent]

Short version:

Instead of

    U_projection_type = 'ortho-atomic',
    hubbard_u        = obj(Fe=5.5),

use

    hubbard          = {'U':{'Fe-3d': 5.5}},
    hubbard_proj     = 'ortho-atomic',

(copied from the just updated https://github.com/QMCPACK/qmcpack/blob/develop/nexus/examples/qmcpack/rsqmc_quantum_espresso/04_iron_dft_dmc_gcta/iron_ldaU_dmc_gcta.py )

As Kayahan noted above, the pseudopotentials need to be compatible with the QE 7.1+ labeling scheme.

[hgriffin-ux]

Thank you both for your explanations, directions on where to locate the examples, distinguishing the QE legacy vs. 7.1>= case keywords, and pseudopotential library update directions. For my current project, I just the Hubbard+U version. It is now generating properly from Nexus and QE is performing as expected.