🔬 materials science

Efficient Interpolation of the Bethe-Salpeter Hamiltonian at Arbitrary Momenta

This paper introduces an efficient reciprocal-space interpolation method that reconstructs the Bethe-Salpeter Hamiltonian at arbitrary momenta from coarse grids, achieving significant computational speedups while accurately capturing excitonic properties and enabling new applications in exciton-phonon interactions.

Original authors: Stefano Mocatti, Xiangzhou Zhu, Giulio Volpato, Pierluigi Cudazzo, Matteo Calandra, Giovanni Marini

Published 2026-11-03

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Stefano Mocatti, Xiangzhou Zhu, Giulio Volpato, Pierluigi Cudazzo, Matteo Calandra, Giovanni Marini

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the solid world of materials science, light does not simply bounce off a surface or pass straight through; it often gets caught in a delicate dance between electrons. When a photon strikes a solid, it can knock an electron loose, leaving behind a vacancy known as a "hole." Because opposite charges attract, the freed electron and the hole often remain bound together, forming a temporary, lightweight particle called an exciton. These excitons are crucial for understanding how materials absorb and emit light, governing everything from the efficiency of solar cells to the color of LEDs. To predict how these particles behave, scientists use a complex mathematical framework called the Bethe–Salpeter equation. This tool allows researchers to calculate the energy and movement of excitons, but it comes with a heavy price: to get an accurate answer, the calculation must sample the material's internal structure at an incredibly fine level of detail. If the sampling is too coarse, the results are blurry; if it is fine enough to be sharp, the calculation becomes so massive that even the fastest supercomputers struggle to finish it.

A team of researchers at the University of Trento has developed a new method to bypass this bottleneck, allowing them to reconstruct the behavior of these excitons at any point in a material without needing to run a full, expensive calculation for every single point. Instead of brute-forcing the problem by calculating every detail from scratch, they created a smooth, continuous map of the material's electronic landscape. They achieved this by taking a coarse, low-resolution set of data points—like a few scattered dots on a map—and using a mathematical technique involving "Wannier functions" to fill in the gaps with high precision. This approach effectively decouples the need for a dense, expensive calculation from the desire for a detailed result. The researchers tested their method on three distinct materials: a single layer of hexagonal boron nitride, a single layer of molybdenum disulfide, and a chunk of bulk gallium arsenide. In every case, the reconstructed results matched the expensive, high-resolution calculations almost perfectly, reproducing the exact energy levels and light-absorption patterns of the excitons.

The power of this new technique lies in its flexibility. Because the method creates a smooth map, scientists can now ask questions about the excitons at momenta that were previously impossible to reach. For instance, in two-dimensional materials like the single layers of boron nitride and molybdenum disulfide, the excitons exhibit a strange, non-smooth behavior when they move very slowly, a phenomenon caused by long-range forces between the electron and the hole. Standard calculations, which rely on a uniform grid of points, often miss this subtle behavior because the grid points are too far apart to catch the rapid change. The new method allowed the researchers to probe these materials at momenta seven times smaller than what a standard grid could access, revealing the precise way the exciton energy splits into different paths. This level of detail was achieved without running a single new, heavy calculation; the researchers simply interpolated the existing data to find the answer.

The efficiency gains are substantial. The researchers estimated that for the materials they studied, their method could speed up the process by a factor of up to twenty-two. This means a calculation that might have taken days or weeks on a supercomputer can now be done in a fraction of the time. Furthermore, the method allows for "local refinement," where scientists can focus their computational power only on the specific areas of the material's internal structure that matter most. When applied to gallium arsenide, a common semiconductor, this focused approach allowed them to calculate the binding energy of the lowest exciton state with high precision using only a few thousand data points, whereas a standard uniform approach would have required over one hundred thousand points to reach the same level of accuracy. The result matched experimental values and previous high-level benchmarks, confirming that the method is not just faster, but also reliable.

By providing a way to access exciton properties at arbitrary points with high efficiency, this work opens the door to more complex studies that were previously out of reach. The researchers note that having a smooth, consistent description of excitons at any momentum is essential for understanding how these particles interact with vibrations in the material's crystal lattice, known as phonons. This interaction is key to understanding how excitons move, lose energy, and eventually decay. With this new framework, scientists can now combine the behavior of excitons with the behavior of electrons and the vibrations of the lattice in a single, coherent picture. This capability paves the way for simulating how materials behave under non-equilibrium conditions, such as when they are hit by a laser pulse, and could lead to better designs for future optoelectronic devices. The study demonstrates that by changing how we look at the data, we can solve problems that were once thought to be too computationally expensive to tackle.

Technical Summary: Efficient Interpolation of the Bethe–Salpeter Hamiltonian at Arbitrary Momenta

Problem Statement
The Bethe–Salpeter equation (BSE) provides a standard first-principles description of neutral excitations in solids, including optical spectra and exciton binding energies. However, its predictive accuracy is hindered by substantial computational costs, particularly the need for dense momentum sampling to converge electron–hole transitions and screened Coulomb interactions. Standard implementations rely on regular electronic momentum meshes, making calculations at arbitrary transferred momenta ( $Q$ ) restrictive. An arbitrary $Q$ often requires a new, denser underlying calculation, which is computationally prohibitive. This limitation is acute in two-dimensional systems near the $\Gamma$ point, where long-range electron–hole exchange induces nonanalytic exciton dispersions requiring momenta far below the spacing of typical ab initio meshes. Furthermore, coupling excitons to lattice degrees of freedom (e.g., for exciton–phonon interactions or nonequilibrium dynamics) requires a common, smooth electronic representation for both excitonic states and interaction vertices, which is difficult to achieve with disparate sampling grids.

Methodology
The authors develop a reciprocal-space framework to reconstruct the static BSE Hamiltonian at arbitrary crystal momenta from coarse first-principles grids. The method decouples the electron–hole transition momenta ( $k$ ) and the transferred momentum ( $Q$ ) from the underlying ab initio sampling. The approach, implemented in the code EPIq, proceeds in four main steps:

Wannier Interpolation of Bloch States: Bloch states are interpolated within an optimally smooth Wannier subspace (OSS). The periodic parts of the Bloch functions are expanded in plane waves and transformed into the OSS using maximally localised Wannier functions (MLWFs). This ensures a smooth representation in reciprocal space, allowing direct interpolation of wavefunction coefficients at arbitrary momenta without intermediate real-space transformation. The interpolated states are then rotated back to the band-eigenstate representation to construct transition densities.
Reconstruction of Screened Interaction: Instead of interpolating the inverse dielectric matrix ( $\epsilon^{-1}$ ) directly—which retains rapid variations from the bare Coulomb interaction—the method interpolates a smoother auxiliary function, $f_{GG'}(q)$ . This function removes the explicit Coulomb dependence. The screened interaction $W_{GG'}(q)$ is then reconstructed analytically at the required exchanged momentum $q = k - k'$ by re-introducing the Coulomb factors. The bare Coulomb interaction is evaluated directly at the required momentum $Q$ and never interpolated.
Hamiltonian Construction: Using the interpolated transition densities and the reconstructed screened interaction, the direct and exchange electron–hole kernels are calculated. The BSE Hamiltonian is assembled for arbitrary transition momenta $k, k'$ and transferred momentum $Q$ .
Solution: The Hamiltonian is solved using iterative solvers (Lanczos–Haydock for optical spectra or Davidson for individual exciton states), avoiding full diagonalization.

Key Contributions

Decoupling of Sampling: The method allows the transition-momentum grid ( $k$ ) and the transferred momentum ( $Q$ ) to be chosen independently of the coarse source grids used for the initial ab initio calculations (Bloch states and dielectric response).
Common Smooth Representation: By utilizing the OSS for both Bloch states and the dielectric response, the framework ensures that exciton wavefunctions and interaction vertices exist in a common smooth electronic representation. This facilitates the consistent combination of excitonic effects with other interaction vertices, such as electron–phonon couplings.
Arbitrary Momentum Access: The framework enables direct access to the long-wavelength regime of exciton dispersion without requiring a uniform refinement of the entire Brillouin zone.

Results and Benchmarks
The method was benchmarked against explicit ab initio calculations (using Yambo) for monolayer hBN, monolayer MoS $_2$ , and bulk GaAs.

Interpolation Accuracy: The interpolated Bloch-state plane-wave components and reconstructed dielectric screening showed excellent agreement with explicit calculations on dense target grids (e.g., $36\times36\times1$ or $48\times48\times1$ ), even when derived from significantly coarser source meshes (e.g., $12\times12\times1$ ). The relative error for Bloch states was below 1%, while dielectric screening errors were largest near $\Gamma$ but remained manageable (around 5% for coarse grids) and decreased with source mesh refinement.
Optical Spectra: Reconstructed BSE optical spectra closely reproduced reference spectra, including excitonic line shapes, peak positions, and oscillator strengths. Convergence analysis using the Wasserstein distance indicated that refining the screening source mesh was the dominant factor for accuracy, particularly in low-dimensional systems.
Computational Efficiency: The approach yielded significant speedups. For monolayer MoS $_2$ , a speedup of up to 22 $\times$ was estimated for dense grids ( $96\times96\times1$ ), and 10 $\times$ for bulk GaAs ( $28\times28\times28$ ). These gains arise because the expensive first-principles calculations (NSCF and dielectric response) are performed only once on the coarse source mesh, while the interpolation and BSE construction scale more favorably.
Long-Wavelength Exciton Dispersion: The method resolved the nonanalytic behavior of exciton dispersions in hBN and MoS $_2$ at momenta approximately seven times smaller than the spacing of practical uniform grids. In hBN, this allowed the resolution of the linear splitting between transverse and longitudinal branches induced by long-range exchange. In MoS $_2$ , it resolved the splitting of the bright A-exciton doublet.
Locally Refined Sampling: For bulk GaAs, locally refined transition-momentum sampling (concentrating points near $\Gamma$ ) converged the binding energy of the threefold-degenerate lowest exciton to 4.0 meV using only a few thousand BSE points. This matched experimental values (3.8–4.2 meV) and previous benchmarks, whereas uniform meshes required over $10^5$ points and remained unconverged.

Significance
The paper claims that this framework makes flexible momentum sampling practical for BSE calculations, removing the need to repeat expensive ab initio calculations for every target momentum. By providing exciton energies and amplitudes at arbitrary crystal momenta within a common smooth electronic representation, the method opens the way to incorporating ab initio excitonic effects into nonlocal electron–phonon vertices and extending calculations to exciton–phonon interactions and nonequilibrium exciton dynamics. The authors emphasize that the method is applicable to a broad range of materials and electronic-structure codes, as it relies only on the availability of Bloch states and the microscopic inverse dielectric matrix on a coarse grid.

Technical Summary: Efficient Interpolation of the Bethe–Salpeter Hamiltonian at Arbitrary Momenta

More like this