Many-body benchmarking of DFT local-registry energetics in bilayer InSe

arXiv:2607.01682 · cond-mat.mtrl-sci · Submitted 2026-07-02 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.

Kai: We're looking at the paper: "Many-body benchmarking of DFT local-registry energetics in bilayer InSe".

Mira: Here's the gist of the paper. Density functional theory (DFT) is widely used to model twisted bilayers, but the accuracy of the local stacking energetics underlying such models remains uncertain Here, we benchmark the localregistry landscape of bilayer InSe using diffusion quantum Monte Carlo (DMC).

Kai: Benchmarking DFT Predictions The study benchmarks the local-registry landscape of bilayer InSe using DMC to test DFT predictions.

Mira: DFT predicts that AB, AAr, and ABr stackings, which share the same interfacial Se registry, are nearly degenerate within 1.5 meV/f.u.

Kai: and exhibit nearly indistinguishable DFT charge-density responses.

Paper summary: Kai: So, to wrap up the "Many-body benchmarking of DFT local-registry energetics in bilayer InSe" paper by Ahn and colleagues. They used DMC to show that DFT underestimates the energy scale driving moiré relaxation in bilayer InSe >

Mira: The authors highlight that DFT predicts AB, AAr, and ABr stackings are nearly degenerate within one point five meV per formula unit, but DMC separates them significantly more by eight point five and forty one point four meV per formula unit <ref:2607.01682#pg1>. Lev What does this mean for the actual physics we’re trying to model in moiré systems? It means that the weak twist dependence DFT predicted shouldn't be interpreted as intrinsically weak registry dependence, but rather a result of a flattened local-registry landscape >

Kai: Exactly. The paper shows that DFT can artificially compress energy differences among locally similar registries <ref:2607.01682#pg3>. Mira So the big point is that reliable moiré modeling requires validating the local-registry hierarchy and its full energetic scale, not merely relying on relaxed geometry or what DFT identifies as the minimum >

Lev: From an engineering standpoint, if we're building systems that rely on moiré patterns, we need to incorporate this many-body physics because it dictates the actual energy landscape and how things will relax in a real device >

Conclusion: Kai: So, this paper, "Many-body benchmarking of DFT local-registry energetics in bilayer InSe," it’s basically using diffusion quantum Monte Carlo to test how well density functional theory predicts the stacking energies in these twisted bilayers >

Mira: Right. The authors are showing that DFT predicts AB, AAr, and ABr stackings are nearly the same energy within one point five meV per formula unit, but DMC shows they actually separate by eight point five and forty one point four meV per formula unit >

Lev: That separation is huge for a quantum error correction system because it means the true energy scale is much bigger than what standard DFT gives you >

Kai: So what does this mean for the whole moiré problem? It suggests that we can't just trust the geometry or which stacking DFT picks as its minimum anymore >

Mira: Exactly. The local registry, like AB versus AAr, isn't just about the atoms touching; it’s about how charge actually moves between the layers and stays in the structure >

Lev: From an engineering standpoint, if we design a moiré device based on DFT predictions alone, we’re missing a big part of the physics that governs how stable those patterns actually are >

Kai: It means that when DFT says there's this weak twist dependence, it’s probably not because the registry is inherently weak; it’s because the local-registry landscape itself is flattened >

Mira: That’s right. The authors argue that DFT can artificially compress those energy differences among similar-looking registries, which messes up our moiré models >

Lev: So what this paper really tells us is that for any real moiré simulation, you need to validate the full energetic scale of the local registry, not just the relaxed geometry or whatever minimum DFT finds >

Jeonghwan Ahn, Abdulgani Annaberdiyev, Jovan Nelson, Nathaniel P. Stern, Hyeondeok Shin

Anthony J Leggett Institute for Condensed Matter Theory, University of Illinois at Urbana-Champaign · Center for Nanophase Materials Sciences, Oak Ridge National Laboratory · Applied Physics Program, Northwestern University · Department of Physics and Astronomy, Northwestern University · Computational Science Division, Argonne National Laboratory

cond-mat.mtrl-sci

Submitted: 2026-07-02

Updated: 2026-07-02

DOI: 10.1103/1mnm-51c5

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 91/100

The gist: Density functional theory (DFT) is widely used to model twisted bilayers, but the accuracy of the local stacking energetics underlying such models remains uncertain Here…

Key concepts

Local Registry Landscape
This refers to how different atomic arrangements (like AB vs. AAr) at a specific interface affect the energy of a material. DFT often predicts this landscape is shallow and functional-dependent, meaning its predictions for stacking stability can be unreliable.
Diffusion Quantum Monte Carlo (DMC)
DMC is a high-accuracy computational method used to calculate the true ground state energies of materials by simulating quantum mechanical particles. It was used here to test and correct the less accurate DFT predictions regarding stacking energies.
Many-Body Hierarchy
This describes the actual, detailed energy ordering among different atomic stackings, which DMC reveals is much more complex than DFT suggests. The many-body hierarchy shows that certain arrangements (like AAr) are significantly more stable or less stable than DFT predicts.
Charge Redistribution
This analysis looks at how electrons move between the layers when the bilayer forms. The study found that AAr stabilizes by moving charge away from the interlayer gap, while ABr shows an opposite trend, providing a microscopic reason for its different energy stability.

Terminology

Summary

Density functional theory (DFT) is widely used to model twisted bilayers, but the accuracy of the local stacking energetics underlying such models remains uncertain Here, we benchmark the localregistry landscape of bilayer InSe using diffusion quantum Monte Carlo (DMC). p. 0

Where the paper says it

Leggett Institute for Condensed Matter Theory, Department of Physics, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA 2Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, TN 37830, USA 3Applied Physics Program, Northwestern University, Evanston, Illinois 60208, USA 4Department of Physics and Astronomy, Northwestern University, Evanston, Illinois 60208, USA 5Computational Science Division, Argonne National Laboratory, Argonne, Illinois 60439, USA (Dated: July 3, 2026) Abstract Density functional theory (DFT) is widely used to model twisted bilayers, but the accuracy of the local stacking energetics underlying such models remains uncertain. Here, we benchmark the localregistry landscape of bilayer InSe using diffusion quantum Monte Carlo (DMC).

Benchmarking DFT Predictions

The study benchmarks the local-registry landscape of bilayer InSe using DMC to test DFT predictions. DFT predicts that AB, AAr, and ABr stackings, which share the same interfacial Se registry, are nearly degenerate within 1. p. 0

Where the paper says it

Here, we benchmark the localregistry landscape of bilayer InSe using diffusion quantum Monte Carlo (DMC). DFT predicts that AB, AAr, and ABr stackings, which share the same interfacial Se registry, are nearly degenerate within 1.

5 meV/f.u. and exhibit nearly indistinguishable DFT charge-density responses. p. 0
Where the paper says it

DFT predicts that AB, AAr, and ABr stackings, which share the same interfacial Se registry, are nearly degenerate within 1. and exhibit nearly indistinguishable DFT charge-density responses.

p. 0
Where the paper says it

DFT predicts that AB, AAr, and ABr stackings, which share the same interfacial Se registry, are nearly degenerate within 1. and exhibit nearly indistinguishable DFT charge-density responses.

DMC instead separates these stackings by 8(5) and 41(4) meV/f. p. 0
Where the paper says it

and exhibit nearly indistinguishable DFT charge-density responses. DMC instead separates these stackings by 8(5) and 41(4) meV/f.

p. 0
Where the paper says it

and exhibit nearly indistinguishable DFT charge-density responses. DMC instead separates these stackings by 8(5) and 41(4) meV/f.

u., while the energy difference between the most stable and least stable registries reaches 60(7) meV/f. p. 0
Where the paper says it

DMC instead separates these stackings by 8(5) and 41(4) meV/f. , while the energy difference between the most stable and least stable registries reaches 60(7) meV/f.

p. 0
Where the paper says it

DMC instead separates these stackings by 8(5) and 41(4) meV/f. , while the energy difference between the most stable and least stable registries reaches 60(7) meV/f.

u. These large energy separations show that the stacking energetics are not determined by the interfacial atomic motif alone but depend on the full registry and its associated many-body electronic response. p. 0
Where the paper says it

, while the energy difference between the most stable and least stable registries reaches 60(7) meV/f. These large energy separations show that the stacking energetics are not determined by the interfacial atomic motif alone but depend on the full registry and its associated many-body electronic response.

p. 0
Where the paper says it

, while the energy difference between the most stable and least stable registries reaches 60(7) meV/f. These large energy separations show that the stacking energetics are not determined by the interfacial atomic motif alone but depend on the full registry and its associated many-body electronic response.

DFT Landscape Limitations

DFT predicts a shallow and functional-dependent registry landscape in which AB, AAr, and ABr remain nearly degenerate within 1.5 meV/f.u., although they differ in their full atomic registries, and the selected minimum depends on the exchange-correlation functional. p. 4

Where the paper says it

, making the three stackings effectively nearly degenerate at the DFT level. The identity of the lowest-energy stacking also depends on the exchange-correlation functional: vdW-optB88, vdW-optB86b, rev-vdW-DF2, and vdW-DF2 predict ABr to be the lowest-energy stacking, whereas r2SCAN+rVV10, SCAN+rVV10, vdW-DF, PBE+MBD, and PBE+D3 favor AB.

This behavior persists when the high-symmetry comparison is extended to representative lower and upper DFT estimates of the full sliding landscape. p. 4
Where the paper says it

2 as approximate lower and upper DFT estimates of the local-registry landscape. This extension from selected stackings to the full lateral-displacement space is required for twisted bilayers because a moir´e pattern samples both favorable and unfavorable local registries.

The DFT local-registry landscape is therefore not a settled reference for twisted bilayer InSe. p. 6
Where the paper says it

The nearly flat vdW-DF landscape and the more strongly corrugated rev-vdW-DF2 landscape would lead to different descriptions of moir’e-scale interlayer response. The DFT local-registry landscape is therefore not a settled reference for twisted bilayer InSe.

Many-Body Hierarchy Revelation

The DMC benchmark gives a qualitatively different hierarchy and energy scale. DMC lifts the apparent degeneracy among AAr, AB, and ABr, separating them by 8(5) and 41(4) meV/f. p. 8

Where the paper says it

The DMC interlayer binding energies strongly lift the DFT near-degeneracy within the common-Se-registry group. First, the β-like AAr registry becomes the lowest-energy bilayer configuration, with Eb = −70(3) meV/f.

u. relative to AAr, and expands the energy difference between the most stable and least stable registries to 60(7) meV/f. p. 0
Where the paper says it

DMC instead separates these stackings by 8(5) and 41(4) meV/f. , while the energy difference between the most stable and least stable registries reaches 60(7) meV/f.

u. The charge-redistribution analysis shows that this hierarchy is associated with registry-specific rearrangement between the interlayer gap and the intralayer bonding network, which is largely smoothed out in DFT. p. 2
Where the paper says it

Density analysis supports this energetic hierarchy by revealing registry-dependent charge redistribution that is largely smoothed out in DFT. Since twisted bilayer InSe samples the same local stacking landscape, the DMC benchmark implies that the weak twist dependence suggested by DFT reflects a flattened local-registry landscape rather than the full many-body interlayer coupling.

Implications for Twisted Bilayers

The enlarged DMC contrast between AB- and AA-like environments shows that DFT underestimates the registry-energy scale entering the moir´e problem. The weak twist dependence found in DFT should therefore not be interpreted as intrinsically weak registry dependence, but as a consequence of a flattened local-registry landscape. p. 2

Where the paper says it

Density analysis supports this energetic hierarchy by revealing registry-dependent charge redistribution that is largely smoothed out in DFT. Since twisted bilayer InSe samples the same local stacking landscape, the DMC benchmark implies that the weak twist dependence suggested by DFT reflects a flattened local-registry landscape rather than the full many-body interlayer coupling.

More generally, the AAr, AB, and ABr comparison shows that DFT can artificially compress the energy differences among locally similar registries. p. 4
Where the paper says it

The comparison in Fig. 1(b) tests whether DFT assigns a clear energetic hierarchy within this structurally motivated subset of locally favorable stackings.

Microscopic Origin of Differences

The density analysis provides a microscopic explanation for the difference between the DMC hierarchy and the DFT landscape. The enhanced stability of AAr is associated with a redistribution that removes charge from the interlayer overlap region and reinforces the intralayer bonding environment. p. 2

Where the paper says it

Density analysis supports this energetic hierarchy by revealing registry-dependent charge redistribution that is largely smoothed out in DFT. Since twisted bilayer InSe samples the same local stacking landscape, the DMC benchmark implies that the weak twist dependence suggested by DFT reflects a flattened local-registry landscape rather than the full many-body interlayer coupling.

ABr shows the opposite redistribution, with charge transferred into the interlayer gap and depleted from the intralayer In–Se and In–In regions. p. 2
Where the paper says it

Density analysis supports this energetic hierarchy by revealing registry-dependent charge redistribution that is largely smoothed out in DFT. Since twisted bilayer InSe samples the same local stacking landscape, the DMC benchmark implies that the weak twist dependence suggested by DFT reflects a flattened local-registry landscape rather than the full many-body interlayer coupling.

The many-body correction to the local-registry landscape is therefore not a uniform shift of the DFT energies. p. 6
Where the paper says it

The DFT local-registry landscape is therefore not a settled reference for twisted bilayer InSe. The DMC calculations below test which parts of this functional-dependent picture survive when the interlayer binding is treated at the many-body level.

Conclusion

The DMC benchmark shows that reliable moir´e modeling requires validation of the local-registry hierarchy and its full energetic scale, not merely the relaxed geometry or the identity of the DFT minimum. The main implication is that DFT-based predictions for twisted bilayer InSe can miss the many-body local-registry physics even when the relaxed geometry appears reasonable. p. 0

Where the paper says it

These large energy separations show that the stacking energetics are not determined by the interfacial atomic motif alone but depend on the full registry and its associated many-body electronic response. More broadly, these results show that DFT-based moir´e models can substantially underestimate local stacking-energy corrugation, with direct consequences for predicted structural relaxation, domain formation, and electronic reconstruction in twisted layered materials.

The DMC benchmark shows that DFT underestimates the registry-energy scale driving moir´e relaxation. p. 2
Where the paper says it

Since twisted bilayer InSe samples the same local stacking landscape, the DMC benchmark implies that the weak twist dependence suggested by DFT reflects a flattened local-registry landscape rather than the full many-body interlayer coupling. More broadly, these results show that underestimating the local-registry energy scale can propagate into moir´e models and affect predicted structural relaxation, domain formation, and electronic reconstruction in twisted layered materials.

More generally, reliable moir´e modeling requires validating the local-registry hierarchy and its full energetic scale, not merely the relaxed geometry or the identity of the DFT minimum. p. 2
Where the paper says it

Since twisted bilayer InSe samples the same local stacking landscape, the DMC benchmark implies that the weak twist dependence suggested by DFT reflects a flattened local-registry landscape rather than the full many-body interlayer coupling. More broadly, these results show that underestimating the local-registry energy scale can propagate into moir´e models and affect predicted structural relaxation, domain formation, and electronic reconstruction in twisted layered materials.

Methods

We used DMC within the fixed-node approximation as implemented in QMCPACK. Single-determinant Slater–Jastrow trial wave functions were used, including one-body electron–ion, two-body electron–electron, and three-body electron–electron–ion Jastrow terms. p. 7

Where the paper says it

We used the Hessian obtained from DFT-PBE calculations to define the search directions, because the DFT-PBE Hessian provides a cost-effective approximation for structural optimization. For the monolayer calculation, the DMC potential-energy surface was sampled using a 288-electron supercell at the Γ point, and the reported structure was obtained after three line-search iterations.

The singleparticle orbitals used in the Slater determinants were generated from PBE [19] calculations performed using QUANTUM ESPRESSO. p. 7
Where the paper says it

In this approach, the search direction is generated from an approximate Hessian, while the energy minimization is carried out on the DMC potential-energy surface using DMC total energies. We used the Hessian obtained from DFT-PBE calculations to define the search directions, because the DFT-PBE Hessian provides a cost-effective approximation for structural optimization.

Data Availability

The data that support the findings of this study are available this article, its supplementary material, and the materials data facility[??] at [link to be provided upon acceptance of this manuscript].

Acknowledgements

J. Ahn acknowledges support from U. p. 0

Where the paper says it

Many-body benchmarking of DFT local-registry energetics in bilayer InSe Jeonghwan Ahn,1 Abdulgani Annaberdiyev,2 Jovan Nelson,3 Nathaniel P. Stern,3, 4 and Hyeondeok Shin5, ∗ 1The Anthony J.

p. 0
Where the paper says it

Many-body benchmarking of DFT local-registry energetics in bilayer InSe Jeonghwan Ahn,1 Abdulgani Annaberdiyev,2 Jovan Nelson,3 Nathaniel P. Stern,3, 4 and Hyeondeok Shin5, ∗ 1The Anthony J.

S. Department of Energy, Office of Science, Office of Basic Energy Sciences, Computational Materials Sciences Award No. p. 0
Where the paper says it

Stern,3, 4 and Hyeondeok Shin5, ∗ 1The Anthony J. Leggett Institute for Condensed Matter Theory, Department of Physics, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA 2Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, TN 37830, USA 3Applied Physics Program, Northwestern University, Evanston, Illinois 60208, USA 4Department of Physics and Astronomy, Northwestern University, Evanston, Illinois 60208, USA 5Computational Science Division, Argonne National Laboratory, Argonne, Illinois 60439, USA (Dated: July 3, 2026) Abstract Density functional theory (DFT) is widely used to model twisted bilayers, but the accuracy of the local stacking energetics underlying such models remains uncertain.

p. 0
Where the paper says it

Stern,3, 4 and Hyeondeok Shin5, ∗ 1The Anthony J. Leggett Institute for Condensed Matter Theory, Department of Physics, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA 2Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, TN 37830, USA 3Applied Physics Program, Northwestern University, Evanston, Illinois 60208, USA 4Department of Physics and Astronomy, Northwestern University, Evanston, Illinois 60208, USA 5Computational Science Division, Argonne National Laboratory, Argonne, Illinois 60439, USA (Dated: July 3, 2026) Abstract Density functional theory (DFT) is widely used to model twisted bilayers, but the accuracy of the local stacking energetics underlying such models remains uncertain.

DE-SC0020177 for calculations, analysis, and writing of the paper. p. 2
Where the paper says it

and substantially enlarging the energy range between the most stable and least stable registries. Density analysis supports this energetic hierarchy by revealing registry-dependent charge redistribution that is largely smoothed out in DFT.

p. 2
Where the paper says it

and substantially enlarging the energy range between the most stable and least stable registries. Density analysis supports this energetic hierarchy by revealing registry-dependent charge redistribution that is largely smoothed out in DFT.

A.A. (analysis, pseudopotentials) and H. p. 2
Where the paper says it

and substantially enlarging the energy range between the most stable and least stable registries. Density analysis supports this energetic hierarchy by revealing registry-dependent charge redistribution that is largely smoothed out in DFT.

p. 2
Where the paper says it

and substantially enlarging the energy range between the most stable and least stable registries. Density analysis supports this energetic hierarchy by revealing registry-dependent charge redistribution that is largely smoothed out in DFT.

S. (mentorship, calculations, analysis, writing) was supported by the U. p. 2
Where the paper says it

and substantially enlarging the energy range between the most stable and least stable registries. Density analysis supports this energetic hierarchy by revealing registry-dependent charge redistribution that is largely smoothed out in DFT.

p. 2
Where the paper says it

and substantially enlarging the energy range between the most stable and least stable registries. Density analysis supports this energetic hierarchy by revealing registry-dependent charge redistribution that is largely smoothed out in DFT.

S. Department of Energy, Office of Science, Basic Energy Sciences, Materials Sciences and Engineering Division. p. 0
Where the paper says it

Stern,3, 4 and Hyeondeok Shin5, ∗ 1The Anthony J. Leggett Institute for Condensed Matter Theory, Department of Physics, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA 2Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, TN 37830, USA 3Applied Physics Program, Northwestern University, Evanston, Illinois 60208, USA 4Department of Physics and Astronomy, Northwestern University, Evanston, Illinois 60208, USA 5Computational Science Division, Argonne National Laboratory, Argonne, Illinois 60439, USA (Dated: July 3, 2026) Abstract Density functional theory (DFT) is widely used to model twisted bilayers, but the accuracy of the local stacking energetics underlying such models remains uncertain.

J.N. (analysis, writing) and N. p. 2
Where the paper says it

and substantially enlarging the energy range between the most stable and least stable registries. Density analysis supports this energetic hierarchy by revealing registry-dependent charge redistribution that is largely smoothed out in DFT.

p. 2
Where the paper says it

and substantially enlarging the energy range between the most stable and least stable registries. Density analysis supports this energetic hierarchy by revealing registry-dependent charge redistribution that is largely smoothed out in DFT.

P. S. (mentorship, analysis, writing) acknowledge support from the National Science Foundation under Award No. p. 0
Where the paper says it

Stern,3, 4 and Hyeondeok Shin5, ∗ 1The Anthony J. Leggett Institute for Condensed Matter Theory, Department of Physics, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA 2Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, TN 37830, USA 3Applied Physics Program, Northwestern University, Evanston, Illinois 60208, USA 4Department of Physics and Astronomy, Northwestern University, Evanston, Illinois 60208, USA 5Computational Science Division, Argonne National Laboratory, Argonne, Illinois 60439, USA (Dated: July 3, 2026) Abstract Density functional theory (DFT) is widely used to model twisted bilayers, but the accuracy of the local stacking energetics underlying such models remains uncertain.

p. 0
Where the paper says it

Stern,3, 4 and Hyeondeok Shin5, ∗ 1The Anthony J. Leggett Institute for Condensed Matter Theory, Department of Physics, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA 2Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, TN 37830, USA 3Applied Physics Program, Northwestern University, Evanston, Illinois 60208, USA 4Department of Physics and Astronomy, Northwestern University, Evanston, Illinois 60208, USA 5Computational Science Division, Argonne National Laboratory, Argonne, Illinois 60439, USA (Dated: July 3, 2026) Abstract Density functional theory (DFT) is widely used to model twisted bilayers, but the accuracy of the local stacking energetics underlying such models remains uncertain.

DMR-1905986. An award of computer time was provided by the Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program. p. 0
Where the paper says it

Stern,3, 4 and Hyeondeok Shin5, ∗ 1The Anthony J. Leggett Institute for Condensed Matter Theory, Department of Physics, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA 2Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, TN 37830, USA 3Applied Physics Program, Northwestern University, Evanston, Illinois 60208, USA 4Department of Physics and Astronomy, Northwestern University, Evanston, Illinois 60208, USA 5Computational Science Division, Argonne National Laboratory, Argonne, Illinois 60439, USA (Dated: July 3, 2026) Abstract Density functional theory (DFT) is widely used to model twisted bilayers, but the accuracy of the local stacking energetics underlying such models remains uncertain.

p. 0
Where the paper says it

Stern,3, 4 and Hyeondeok Shin5, ∗ 1The Anthony J. Leggett Institute for Condensed Matter Theory, Department of Physics, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA 2Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, TN 37830, USA 3Applied Physics Program, Northwestern University, Evanston, Illinois 60208, USA 4Department of Physics and Astronomy, Northwestern University, Evanston, Illinois 60208, USA 5Computational Science Division, Argonne National Laboratory, Argonne, Illinois 60439, USA (Dated: July 3, 2026) Abstract Density functional theory (DFT) is widely used to model twisted bilayers, but the accuracy of the local stacking energetics underlying such models remains uncertain.

References

[1] Yoo, H. et al. Atomic and electronic reconstruction at the van der Waals interface in twisted bilayer graphene

[2] Carr, S., Fang, S. & Kaxiras, E. Electronic-structure methods for twisted moir´e layers

[3] Zhou, J. et al. InSe monolayer: synthesis, structure and ultra-high second-harmonic generation

[4] Hilse, M. et al. Mixed polytype/polymorph formation in InSe films grown by molecular beam epitaxy on GaAs(111)B

[5] Song, S. et al. Indium selenides for next-generation low-power computing devices

[6] Mudd, G. W. et al. Tuning the bandgap of exfoliated InSe nanosheets by quantum confinement

[7] Sun, Y. et al. InSe: a two-dimensional material with strong interlayer coupling

[8] Yao, X. & Zhang, X. Electronic structures of twisted bilayer InSe/InSe and heterobilayer graphene/InSe

[9] Pike, N. A. et al. Understanding the origin and implication of the indirect-to-direct bandgap transition in multilayer InSe

[10] Magorrian, S. J., Siddiqui, A. & Hine, N. D. Strong atomic reconstruction in twisted bilayers of highly flexible InSe: Machine-learned interatomic potential and continuum model approaches

[11] Foulkes, W. M. C., Mitas, L., Needs, R. J. & Rajagopal, G. Quantum monte carlo simulations of solids

[12] Han, L. et al. Strong interlayer interaction in two-dimensional layered PtTe2

[13] Ahn, J., Kang, S.-H., Yoon, M. & Krogel, J. T. Exploring interlayer coupling in the twisted bilayer PtTe2

[14] Li, J. et al. Layer-dependent band gaps of platinum dichalcogenides

[15] Ahn, J. et al. Stacking polymorphism of PtSe2: its implication to layer-dependent metalinsulator transitions

[16] Politano, A. et al. Indium selenide: an insight into electronic band structure and surface excitations

[17] Shin, H. et al. Optimized structure and electronic band gap of monolayer GeSe from quantum Monte Carlo methods

[18] Kim, J. et al. QMCPACK: An open source ab initio Quantum Monte Carlo package for the electronic structure of atoms, molecules, and solids

[19] Perdew, J. P., Burke, K. & Ernzerhof, M. Generalized gradient approximation made simple

[20] Giannozzi, P. et al. QUANTUM ESPRESSO: a modular and open-source software project for quantum simulations of materials

[21] Burkatzki, M., Filippi, C. & Dolg, M. Energy-consistent pseudopotentials for quantum Monte Carlo calculations

[22] Casula, M., Moroni, S., Sorella, S. & Filippi, C. Size-consistent variational approaches to nonlocal pseudopotentials: Standard and lattice regularized diffusion Monte Carlo methods revisited

[23] Lin, C., Zong, F. H. & Ceperley, D. M. Twist-averaged boundary conditions in continuum quantum Monte Carlo algorithms

[24] Kresse, G. & Hafner, J. Ab initio molecular dynamics for liquid metals

[25] Kresse, G. & Furthm¨uller, J. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set

[26] Kresse, G. & Joubert, D. From ultrasoft pseudopotentials to the projector augmented-wave method

[27] Dion, M., Rydberg, H., Schr¨oder, E., Langreth, D. C. & Lundqvist, B. I. Van der Waals density functional for general geometries

[28] Lee, K., Murray, E. D., Kong, L., Lundqvist, B. I. & Langreth, D. C. Higher-accuracy van der Waals density functional

[29] Hamada, I. van der Waals density functional made accurate

[30] Klimeˇs, J., Bowler, D. R. & Michaelides, A. Van der Waals density functionals applied to solids

[31] Klimeˇs, J., Bowler, D. R. & Michaelides, A Chemical accuracy for the van der Waals density functional

[32] Peng, H., Yang, Z.-H., Perdew, J. P. & Sun, J Versatile van der Waals density functional based on a meta-generalized gradient approximation

[33] Ning, J. et al. Workhorse minimally empirical dispersion-corrected density functional with tests for weakly bound systems: r2SCAN+rVV10

[34] Grimme, S., Antony, J., Ehrlich, S. & Krieg, H. A consistent and accurate ab initio parametrization of density functional dispersion correction (DFT-D) for the 94 elements H-Pu

--- Page 1 ---

Many-body benchmarking of DFT local-registry energetics in bilayer InSe is a study that uses diffusion quantum Monte Carlo (DMC) to test the accuracy of DFT predictions regarding the stacking energetics in bilayer InSe, which matters because it reveals that DFT underestimates the registry-energy scale driving moir´e relaxation.

Pruning redundant reasoning tokens

DFT predicts that AB, AAr, and ABr stackings, which share the same interfacial Se registry, are nearly degenerate within 1.5 meV/f.u. and exhibit nearly indistinguishable DFT charge-density responses. DMC instead separates these stackings by 8(5) and 41(4) meV/f.u., while the energy difference between the most stable and least stable registries reaches 60(7) meV/f.u. These large energy separations show that the stacking energetics are not determined by the interfacial atomic motif alone but depend on the full registry and its associated many-body electronic response.

How the training signal is built

The study establishes a many-body reference for testing the shallow and functional-dependent DFT local-registry landscape. This is done by first defining an isolated-layer reference used for the DMC binding energies. p. 6

Where the paper says it

The DMC calculations below test which parts of this functional-dependent picture survive when the interlayer binding is treated at the many-body level. Many-body monolayer reference Before benchmarking the bilayer local-registry energetics, we first define the isolated-layer reference used for the DMC binding energies.

The DMC structural optimization was performed using the surrogate Hessian line-search method [17]. p. 7
Where the paper says it

at the DMC level using the in-plane lattice constant a, the vertical In–In separation z1, and the vertical Se–Se separation z2 as structural variables. The DMC structural optimization was performed using the surrogate Hessian line-search method [17].

Where the gains come from

The many-body benchmark reshapes the entire local-registry hierarchy relative to DFT. The DMC equilibrium distances are broadly consistent with the structural expectations from the interfacial Se registry. p. 7

Where the paper says it

This procedure provides a direct many-body reference for testing the shallow and functional-dependent DFT local-registry landscape established above. The DMC equilibrium distances are broadly consistent with the structural expectations from the interfacial Se registry.

The AAr and AB remain strongly bound, while ABr is much less stable, showing that the many-body hierarchy cannot be explained by the relaxed interlayer distance or by the coarse AB-type classification of the facing Se sublattices alone. p. 9
Where the paper says it

AAr and AB remain strongly bound, whereas ABr is much less stable. Thus, the many-body hierarchy cannot be explained by the relaxed interlayer distance or by the coarse AB-type classification of the facing Se sub10

Registry-dependent density redistribution at the InSe interface

The study analyzes charge redistribution induced by bilayer formation to identify the microscopic origin of the DMC registry hierarchy. AAr responds to interlayer contact by removing charge from the overlap region between the two layers and redistributing it into the intralayer bonding network. p. 1

Where the paper says it

Even when the constituent monolayers are individually well understood, bringing two layers together introduces a new degree of freedom through the lateral registry between them. Local stacking registry can reshape the equilibrium spacing, charge redistribution, and low-energy electronic structure [1, 2].

ABr shows the opposite response, with charge accumulating in the central interlayer region, while nearby In–Se and In–In intralayer environments show depletion. p. 0
Where the paper says it

, while the energy difference between the most stable and least stable registries reaches 60(7) meV/f. These large energy separations show that the stacking energetics are not determined by the interfacial atomic motif alone but depend on the full registry and its associated many-body electronic response.

This provides a natural explanation for why ABr loses the apparent low-energy stability suggested by DFT despite belonging to the same short-distance group as AAr and AB. p. 9
Where the paper says it

The ABr result is especially important because it shows that the short-distance Seinterface motifs are not energetically equivalent at the many-body level. AAr, AB, and ABr have comparable optimized interlayer distances within the DMC uncertainties, and the DFT relaxations also place them in the same short-distance structural group.

Implications for twisted bilayer InSe

The DMC point at θ = 0◦ gives the many-body AA-to-AB energy difference on the same relative-energy scale. Since the twisted-bilayer energy is governed by the local-registry window defined by the AB and AA limits, the larger DMC separation shows that DFT underestimates the registry-energy scale entering the twist problem. p. 2

Where the paper says it

Since twisted bilayer InSe samples the same local stacking landscape, the DMC benchmark implies that the weak twist dependence suggested by DFT reflects a flattened local-registry landscape rather than the full many-body interlayer coupling. More broadly, these results show that underestimating the local-registry energy scale can propagate into moir´e models and affect predicted structural relaxation, domain formation, and electronic reconstruction in twisted layered materials.

The limitation of the DFT-based moir´e description is therefore already apparent from the local-registry DMC benchmark. p. 2
Where the paper says it

These considerations motivate a many-body benchmark for the high-symmetry bilayer registries, not only to assess DFT at the bilayer level, but also to evaluate whether DFT provides an appropriate local-registry landscape for twisted bilayer InSe. Here, diffusion quantum Monte Carlo (DMC) [11] reveals the stacking dependence hidden by the DFT description of bilayer InSe.

The DMC benchmark gives a qualitatively different hierarchy and energy scale. The AAr, AB, and ABr comparison further demonstrates that DFT can artificially compress the energy differences among locally similar registries. p. 4

Where the paper says it

The comparison in Fig. 1(b) tests whether DFT assigns a clear energetic hierarchy within this structurally motivated subset of locally favorable stackings.

The weak twist dependence predicted by DFT should therefore be interpreted as a consequence of this flattened registry landscape. p. 2
Where the paper says it

Since twisted bilayer InSe samples the same local stacking landscape, the DMC benchmark implies that the weak twist dependence suggested by DFT reflects a flattened local-registry landscape rather than the full many-body interlayer coupling. More broadly, these results show that underestimating the local-registry energy scale can propagate into moir´e models and affect predicted structural relaxation, domain formation, and electronic reconstruction in twisted layered materials.

More broadly, the DMC results show that reliable moir´e modeling requires validation of the local-registry hierarchy and its full energetic scale, not merely the relaxed geometry or the identity of the DFT minimum. p. 2
Where the paper says it

Since twisted bilayer InSe samples the same local stacking landscape, the DMC benchmark implies that the weak twist dependence suggested by DFT reflects a flattened local-registry landscape rather than the full many-body interlayer coupling. More broadly, these results show that underestimating the local-registry energy scale can propagate into moir´e models and affect predicted structural relaxation, domain formation, and electronic reconstruction in twisted layered materials.

J. Ahn acknowledges support from U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, Computational Materials Sciences Award No. DE-SC0020177 for calculations, analysis, and writing of the paper. A.A. (analysis, pseudopotentials) and H.S. (mentorship, calculations, analysis, writing) was supported by the U.S. Department of Energy, Office of Science, Basic Energy Sciences, Materials Sciences and Engineering Division as part of the Computational Materials Sciences Program and Center for Predictive Simulation of Functional Materials. p. 0

Where the paper says it

Stern,3, 4 and Hyeondeok Shin5, ∗ 1The Anthony J. Leggett Institute for Condensed Matter Theory, Department of Physics, University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA 2Center for Nanophase Materials Sciences, Oak Ridge National Laboratory, Oak Ridge, TN 37830, USA 3Applied Physics Program, Northwestern University, Evanston, Illinois 60208, USA 4Department of Physics and Astronomy, Northwestern University, Evanston, Illinois 60208, USA 5Computational Science Division, Argonne National Laboratory, Argonne, Illinois 60439, USA (Dated: July 3, 2026) Abstract Density functional theory (DFT) is widely used to model twisted bilayers, but the accuracy of the local stacking energetics underlying such models remains uncertain.

J.N. (analysis, writing) and N. P. S. (mentorship, analysis, writing) acknowledge support from the National Science Foundation under Award No. DMR-1905986. An award of computer time was provided by the Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Many-body energetic accuracy for twisted bilayer simulations can be improved by replacing DFT local registry models with diffusion quantum Monte Carlo (DMC) benchmarks to accurately capture stacking-dependent energy scales, as DMC separates these stackings by 8(5) and 41(4) meV/f.u.

  2. The improved AI system can predict the true energy hierarchy of InSe bilayer stackings, which is currently underestimated by DFT because it predicts AB, AAr, and ABr stackings, which share the same interfacial Se registry, are nearly degenerate within 1.5 meV/f.u.

  3. The system will be able to distinguish between structurally similar but energetically distinct local registries in twisted bilayers by analyzing the charge-redistribution response; specifically, it can identify that AAr responds to interlayer contact by removing charge from the overlap region between the two layers and redistributing it into the intralayer bonding network, whereas ABr shows an opposite response.

  4. For moire pattern modeling, the AI system will incorporate a registry-dependent energy scale derived from DMC, ensuring that the twist angle energy should be more sensitive to the many-body registry contrast than suggested by the DFT curves, leading to more accurate predictions for structural relaxation and domain formation.

  5. The improved system can correct artificial near-degeneracies among locally similar registries identified by DFT, specifically resolving how stacking affects charge transfer between the interlayer gap and the intralayer bonding network to achieve a reliable moir´e modeling that validates the local-registry hierarchy itself.

Abstract

Density functional theory (DFT) is widely used to model twisted bilayers, but the accuracy of the local stacking energetics underlying such models remains uncertain. Here, we benchmark the local-registry landscape of bilayer InSe using diffusion quantum Monte Carlo (DMC). DFT predicts that AB, AAr, and ABr stackings, which share the same interfacial Se registry, are nearly degenerate within 1.5 meV/f.u. and exhibit nearly indistinguishable DFT charge-density responses. DMC instead separates these stackings by 8(5) and 41(4) meV/f.u., while the energy difference between the most stable and least stable registries reaches 60(7) meV/f.u.. These large energy separations show that the stacking energetics are not determined by the interfacial atomic motif alone but depend on the full registry and its associated many-body electronic response. More broadly, these results show that DFT-based moiré models can substantially underestimate local stacking-energy corrugation, with direct consequences for predicted structural relaxation, domain formation, and electronic reconstruction in twisted layered materials.

Related papers