Crossed pseudopotential − - functional calculations made simple: An extended Kohn-Sham framework

Kuiyu Ye, Jiale Shen, Haitao Liu, Yuanchang Li · 2025

Read Crossed pseudopotential − - functional calculations made simple: An extended Kohn-Sham framework by Kuiyu Ye, Jiale Shen, Haitao Liu, Yuanchang Li (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-07269.txt.

Introduction

Kuiyu Ye and co-authors argue that the helper potentials used in standard materials calculations are a hidden source of error, in work from 2025. First-principles codes lean hard on these stand-ins for core electrons, yet few studies check their cost. The team shows most stand-ins shift atomic energy levels enough to break the theory's own ground rule.

They use adjusted potentials tuned at the atomic level, paired with the usual energy formulas as a pragmatic fix. Tests on fifty-four semiconductors removed wrong metal calls for eleven compounds. Average error in band gap forecasts fell from eighty percent to twenty. Accuracy passed standard hybrid methods and costly many-body runs.

Catalog record

CREATORKuiyu Ye, Jiale Shen, Haitao Liu, Yuanchang Li
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE26 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-07269

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Crossed pseudopotential − - functional calculations made simple: An extended Kohn-Sham framework Kuiyu Ye, Jiale Shen, Haitao Liu, Yuanchang Li Abstract Modern density-functional-theory (DFT) calculations rely heavily on pseudopotentials, yet their impact on accuracy is barely addressed. In this work, we derive from the Kohn-Sham equation that the use of pseudopotentials invariably introduces a “dropping” error, which leads to a deviation from the Hohenberg-Kohn theorem. Crossed pseudopotential-functional calculations provide a pragmatic way to balance accuracy and efficiency, enabling the right results for the right reasons. This paradigm goes beyond the (generalized) Kohn-Sham framework, which we name the extended Kohn-Sham framework. We support our assertion with a bandgap study on 54 monovalent-Cu semiconductors. The crossed calculations, compared to consistent ones, not only remove all 11 erroneous metal predictions, but also drastically reduce the mean relative error from 80% to 20%. The accuracy even exceeds that of the hybrid functionals and GW due to the role of pseudopotentials in modelling the external potentials of Cu-valence electrons that cannot be compensated by exchange-correlation. Density functional theory (DFT) has become a cornerstone of modern electronic structure studies Jones ; Huang ; Lejaeghere . By reformulating the intricate interactions among electrons in terms of densities, DFT strikes a better balance between computational efficiency and accuracy than solving the wavefunctions directly. DFT is mostly realized with Kohn-Sham (KS) equations, which use electron densities constructed from independent single-particle wavefunctions to describe interacting many-body systems KS . In principle, the accuracy of KS-DFT depends entirely on the exchange-correlation functionals. However, the actual accuracy also depends on the numerical method Lejaeghere , where the basis set is a key factor. Various DFT codes make extensive use of plane-wave bases. In order to accurately handle the rapid oscillations of wavefunctions near the nucleus, plane waves are used in conjunction with pseudopotentials Milman ; Garrity . Pseudopotentials predate DFT and highlight the difference between core and valence electrons Hellmann ; Cohen1983 . The former are tightly bound to the nucleus and chemically inert, while the latter are high-energy, chemically active and determine macroscopic properties. Characterizing the interaction of the nucleus and core electrons on the valence electrons as a whole in terms of pseudopotentials greatly simplifies the calculations by eliminating the rapid oscillations of wavefunctions near the nucleus. Although pseudopotentials can be empirical, those prevalent today are virtually from first-principles Cohen ; Zunger . Early study showed that when the pseudopotential is not generated from the functional employed in the calculation, it increases the error Fuchs1998 . For this reason, it has long been assumed in the DFT community that results are more accurate when the pseudopotential is generated by the same functional employed, which is known as pseudopotential-functional consistency Yang2023 . However, two facts challenge this. On the one hand, the effect of pseudopotentials on DFT accuracy is considered less compared to exchange-correlation functionals. So, compared to the fast-growing amount of functionals (There are more than 500 in the LIBXC software library Lehtola ), the number of pseudopotentials is an order of magnitude smaller. In condensed matter physics, using LDA/GGA pseudopotentials is still common even with advanced meta-GGA functionals, which leads to more practical inconsistency Borlidojctc ; Borlidonpj ; Mazdziarz . On the other hand, more and more studies have revealed that pseudopotentials have a far from negligible effect on the DFT accuracy, such as bandgaps, barrier heights, and exciton binding energies, etc TanJCP ; Tanpccp ; Vogel96 ; Vogel97 ; Shen2023 ; Shen2024 ; WuYJ ; Bu ; Rossomme ; Filippetti . In some cases, pseudopotentials play a decisive role and crossed pseudopotential-functional calculations are instead more accurate TanJCP ; Tanpccp ; Vogel96 ; Vogel97 ; Shen2023 ; Shen2024 ; WuYJ ; Bu ; Rossomme ; Filippetti . In view of these, how the pseudopotential affects the accuracy and the underlying physics behind the better accuracy of the crossed paradigm, have become urgent theoretical issues for the DFT community. In this work, we derive exact conditions for the pseudopotentials directly from the KS equations, which show that pseudopotentials constructed from different functional approximations introduce “dropping” errors of unequal magnitude. Although pseudopotentials and functionals are communicable with each other, they affect the accuracy in their own unique ways. Thus, the accuracy of pseudopotential-DFT is essentially a joint result of the two variables, and crossed pseudopotential-functional calculations that abandon the consistency constraint can better balance accuracy and efficiency. The crossed paradigm falls out of the (generalized) KS framework, which we refer to as the extended KS framework. We support the assertion with a bandgap study of 54 monovalent-Cu semiconductors that are notorious examples of KS-DFT failures. With the same PBE functional, consistent PBE pseudopotentials incorrectly predict about a quarter as metals, and more than three-quarters having bandgap underestimation > 1 eV. Replacing the PBE with hybrid pseudopotentials not only opens the bandgap for all but also about three-quarters of the error is within 0.4 eV. The mean relative error is significantly reduced from 80% to 20%. This accuracy statistically outperforms even the HSE hybrid functional and GW, but with a significant advantage in computational efficiency. Finally, we show that this improvement belongs to the right results for the right reasons, originating from the hybrid pseudopotential correctly describing the external potential of the Cu-valence electrons. The KS equation KS reads \displaystyle\{-\frac{1}{2}\nabla^{2}+v_{nu}({\bf r})+v_{H}([n];{\bf r})+v_{xc}([n];{\bf r})\}\phi_{i}({\bf r})=\varepsilon_{i}\phi_{i}({\bf r}),\ \ [n]=\sum_{i}^{N}|\phi_{i}({\bf r})^{2}|. (1) Here v_{nu}({\bf r}) , v_{H}([n];{\bf r}) and v_{xc}([n];{\bf r}) are the nuclear, Hartree and exchange-correlation potentials, respectively. Inserting a potential v_{ps}({\bf r}) one can rewrite Eq. (1) as \displaystyle\{-\frac{1}{2}\nabla^{2}+v_{ps}({\bf r})+v_{H}([n];{\bf r})+v_{xc}([n];{\bf r})\}\phi_{i}({\bf r})+\{v_{nu}({\bf r})-v_{ps}({\bf r})\}\phi_{i}({\bf r})=\varepsilon_{i}\phi_{i}({\bf r}). (2) If \{v_{nu}({\bf r})-v_{ps}({\bf r})\}\phi_{i}({\bf r})=0\ \ \ (r>r_{0}), (3) a set of wavefunctions \psi_{i}({\bf r}) can be found such that \displaystyle\{-\frac{1}{2}\nabla^{2}+v_{ps}({\bf r})+v_{H}([n];{\bf r})+v_{xc}([n];{\bf r})\}\psi_{i}({\bf r})=\varepsilon_{i}\psi_{i}({\bf r}),\ \ \psi_{i}({\bf r})=\phi_{i}({\bf r})\ \ (r>r_{0}). (4) This is exactly the equation that the pseudopotential-DFT self-consistently solves Cohen1983 ; Hellmann . Therefore, any v_{ps}({\bf r}) satisfying Eq. (3) is a pseudopotential. Physically, Eq. (3) states that at r>r_{0} , the attraction of the nucleus to the outer valence electrons is completely cancelled out by v_{ps}({\bf r}) , thus making them nearly-free-electrons and suitable for the plane-wave basis. The v_{ps}({\bf r}) derived from Eq. (3) is exact and does not introduce additional errors into the solution of the KS equations. However, solving Eq. (3) requires \phi_{i}({\bf r}) , which comes from solving Eq. (1). The exact exchange-correlation is unknown, so is the corresponding \phi_{i}({\bf r}) . Consequently, the actual v^{ac}_{ps}({\bf r}) derived from an approximate functional leads to \{v_{nu}({\bf r})-v^{ac}_{ps}({\bf r})\}\phi_{i}({\bf r})\neq 0\ \ \ (r>r_{0}). (5) This means an incomplete cancellation of v^{ac}_{ps}({\bf r}) to the nuclear attraction. As a result, a non-zero term is dropped in the process of transferring from the exact KS Eq. (2) to the actually solved pseudopotential Eq. (4), resulting in a “dropping” error. Although essentially derived from the functional approximation, the “dropping” error is physically different from the error induced by the functional approximation in Eq. (4). The former is embedded in the pseudopotential construction and depends only on the accuracy of the functional used to compute atoms. More precisely, it depends on the accuracy of the functional used to characterize the fact that the attraction of the nucleus to the outer

Read the complete text