WorksPhysics and Astronomy
Koopmans spectral functionals for excitons in materials
Abstract
Accurate first-principles characterization of excitons typically relies on the Bethe-Salpeter equation (BSE) combined with quasiparticle energies from computationally demanding GW calculations. Here, we replace the GW step with quasiparticle energies from Koopmans-compliant spectral functionals, an approach we call BSE@qKI, and implement it in an automated, reproducible AiiDA workflow. For a set of benchmark semiconductors, BSE@qKI reproduces experimental band gaps and first exciton energies with mean absolute relative errors of 13% and 4%, respectively, comparable to BSE@G0W0 for the gaps and markedly better for the excitation energies. We also apply the method to the layered magnetic semiconductor CrI3, obtaining an absorption spectrum that reproduces the main experimental features. These results establish BSE@qKI as a GW-free alternative to BSE@G0W0 for systematic studies of excitonic properties in complex materials.
paper.pdf
Koopmans spectral functionals for excitons in materials
Abstract
Introduction
Excitons — bound states formed by electron-hole pairs — are the quasiparticles that govern the optical response of materials [1]. They play a central role in several applications including photocatalysis [2, 3, 4], photovoltaics [5, 6, 7], plasmonics [8, 9], and chemical sensing [10]. Describing excitonic effects is therefore essential both for understanding the optical properties of materials and for designing of devices with tailored functionalities. From a first-principles perspective, many-body perturbation theory (MBPT), and in particular the Bethe-Salpeter equation (BSE) [11, 12], provides a state-of-the-art framework for describing neutral excitations. Conventionally, the BSE is solved using quasiparticle (QP) energies from the approximation [12], often within the single-shot scheme [13] (the well-known BSE@ approach). This route yields highly accurate optical spectra and excitonic properties, but its computational cost can quickly become prohibitive with material size (the scaling is typically , where is the number of particles) or for systematic studies over many materials [14, 15, 16, 17, 18]. The computational burden of the calculation is compounded by the subsequent construction and diagonalization of the BSE Hamiltonian. As such, it is desirable to decrease the cost of these calculations, without compromising the accuracy of the resulting excitonic description.
Koopmans spectral functionals [19, 20, 21, 22] are designed to provide accurate spectral properties at a computational cost closer to that of standard density-functional theory (DFT) [22], which makes them a promising source of QP energies. By feeding their eigenvalues directly to the BSE Hamiltonian, one can bypass the calculations altogether. We call this approach BSE@qKI. It aims to retain the predictive power of the BSE while substantially reducing the computational cost of the standard BSE@ workflow. In this work, we assess the accuracy of the BSE@qKI approach. We first benchmark it against BSE@ and against experiment for five well-characterized bulk semiconductors, and we then apply it to the layered magnetic semiconductor [23], a system in which non-collinear magnetism and spin-orbit coupling are both important. The optical spectra and excitonic properties agree well with those of BSE@; for the energy of the first excitonic peak, BSE@qKI agrees better with experiment than BSE@ does for every benchmark material considered. The BSE@qKI approach also lends itself to automation: Koopmans functionals are implemented in the KOOPMANS package [22], which we integrate here with the AiiDA workflow engine [24, 25, 26]; together, they make BSE@qKI calculations reproducible and systematic with limited user intervention. This provides a practical framework for applying the method across different classes of materials, and for carrying out high-throughput or large-scale studies of excitonic properties.
The rest of the paper is organized as follows. Section II introduces the theoretical framework underlying Koopmans functionals and the BSE, with particular emphasis on how they are combined in the BSE@qKI approach. Section III illustrates the computational workflow and assesses its performance for a set of benchmark bulk semiconductors and for . Finally, Section IV summarizes the main findings and discusses the prospects of Koopmans-based approaches for efficient first-principles exciton characterization.
Methods
Koopmans spectral functionals
Koopmans spectral functionals [19, 20, 21, 22, 27] describe ground-state properties and charged excitations simultaneously and accurately through a functional approach (as opposed to a perturbative one). They give accurate spectral properties for model [28], finite [20, 29, 30, 31, 32], and extended systems [27, 33, 34, 35, 36, 37, 38, 39] at a computational cost and complexity well below those of other state-of-the-art approaches such as quantum chemistry or Green’s function methods. This is achieved by augmenting standard density-functional approximations with simple, physically motivated corrections that force the total energy to be piecewise-linear (PWL) when an electron is added to, or removed from, any orbital in the system. This is an heuristic generalization — to the entire electronic manifold — of the well-known piecewise linearity of the total energy as a function of the number of particles [40] (or, equivalently, as a function of the fractional occupation of the highest occupied state). Once screening and relaxation effects are correctly accounted for [21, 27], this restores the physical meaning of single-particle Koopmans-compliant (KC) energies as total-energy differences between the neutral system and the system with plus or minus one electron, and therefore amounts to a correct description of charged excitations. In practice, the generalized PWL condition is imposed orbital by orbital: the dependence of the total energy on the occupation of each orbital is removed and replaced by a linear term,
where is the DFT Hamiltonian evaluated at the density in which orbital has occupation . The slope of the linear term determines the flavor of the Koopmans functional [19, 20, 22]. In the Koopmans-integral (KI) functional, is the DFT total-energy difference between the systems with orbital fully occupied and empty, . In the frozen-orbital approximation, the resulting corrections can be expressed as explicit orbital-density-dependent functionals, defining the bare (unscreened) Koopmans corrections. Screening and relaxation effects are accounted for a posteriori via orbital-dependent screening coefficients . The KC total energy functional reads:
where is the total charge density and is the orbital density associated with orbital at occupation . In this work, we adopt the quadratic approximation to the KI functional, denoted quadratic Koopmans-integral (qKI) [27, 35]. This approximation is obtained by expanding the KI corrections to second order in [27, 35]:
where is the orbital density at full occupation, and is the Hartree-plus-exchange-correlation kernel — that is, the second derivative of the underlying DFT functional with respect to the density. The screening coefficients in Eq. (2) account for the response of the system to the ionization process [21, 32]. They are defined in terms of the dielectric response of the material as
where is the microscopic static dielectric function of the system evaluated at the underlying DFT level. As for other orbital-density-dependent approaches [41], KC spectral functionals are not invariant under unitary rotations, and are minimized by a set of variational orbitals that are typically localized in space. Imposing the generalized PWL condition on this set of localized orbitals is what allows the method to be extended seamlessly to periodic systems [21]. Moreover, in most cases the variational orbitals are well approximated by maximally localized Wannier functions (MLWFs) [21, 35, 42] that are faster to compute and which bypass the energy-minimization procedure entirely. This Koopmans-Wannier formulation [35, 37, 43] is deployed as a one-shot correction to DFT eigenvalues and eigenvectors in the KCW module of Quantum ESPRESSO [44, 45]. In practice, one computes the matrix elements of the qKI orbital-density-dependent potential on a set of MLWFs,
where is the qKI potential acting on the th variational orbital. The KC orbital-density-dependent potential acts on the variational state as a simplified self-energy — hence the notation in Eq. (5) — that is static and local in space [46], but that includes approximate vertex corrections [32, 37]. In the standard procedure, the contribution in Eq. (5) is added to the DFT Hamiltonian, represented in the same MLWF basis, and the resulting KC Hamiltonian is diagonalized to give the canonical orbitals. These have a natural interpretation as Dyson orbitals, and their energies as particles additional/removal energies [29, 46, 47].
Bethe-Salpeter equation
Neutral excitations arise from the creation of electron-hole pairs, which can be induced by external perturbbations such as photon absorption or inelastic scattering [12]. The electron and the hole interact via the Coulomb interaction and can propagate through the system as a correlated pair; this correlated pair is the exciton. Describing the optical properties of materials from an ab initio perspective therefore requires excitonic effects to be included. Within MBPT and the Green’s functions formalism, this is achieved through the Bethe-Salpeter equation (BSE) [11, 12, 48], which can be cast as an effective two-particle excitonic Hamiltonian . Within the resonant approximation and in the long-wavelength limit , both used throughout this work, its matrix elements read
where the electron–hole basis states are , with and the valence and conduction single-particle states with energies and . The interaction kernel consists of the bare exchange term and the screened direct interaction ; it is this kernel that carries the electron–hole interaction responsible for excitonic effects. The screened Coulomb interaction is calculated within the random-phase approximation (RPA).
The BSE can be recast as an effective two-particle eigenvalue problem,
with excitonic states expanded as
The exciton binding energy (EBE) of a given state is usually computed as the difference between the fundamental band gap and the exciton eigenvalue . That definition is not appropriate here, however, because the lowest excitonic state may involve transitions away from the point in the Brillouin zone (BZ) at which the minimum gap occurs. We therefore evaluate the EBE as the difference between the weighted average of the independent-particle transition energies contributing to , , and the excitonic eigenvalue itself:
With this definition [49], the EBE measures the red shift produced by the electron-hole interaction alone, and can be compared between different quasiparticle schemes on an equal footing. Within this framework, the inverse macroscopic dielectric function can be expressed in terms of the excitonic eigenvalues and oscillator strengths . Considering only the resonant contributions, we obtain:
where is the volume of the unit cell and is the number of points in the BZ sampling. The oscillator strengths
measure the optical activity of the excitonic states and determine their contribution to the absorption spectrum. From an experimental perspective, excitonic effects can be probed using different spectroscopic techniques depending on the transferred momentum. In the optical limit () they are seen in photoabsorption measurements, while finite-momentum excitations are probed by electron energy-loss spectroscopy (EELS) or non-resonant inelastic X-ray scattering (NRIXS). These experiments give access, respectively, to the absorption spectrum and the loss function , which are related to the macroscopic dielectric function by [12]
As stated above, we restrict the analysis to the optical limit; for treatments beyond the regime we redirect the reader to Refs. 48, 50.
Finally, converged results require careful control of several numerical parameters, notably the plane-wave expansion of the microscopic dielectric matrix that enters the screened interaction . We return to these parameters in Section III A; further details are provided in Refs. 16, 17.
Combining the two: BSE@qKI
The quasiparticle eigenvalues and entering (6) can be obtained in several ways: from a simple scissor-and-stretching operator applied to the Kohn-Sham (KS) eigenvalues, or from one-shot , i.e. the first iteration of Hedin’s self-consistent loop [13]. The latter choice defines the standard BSE@ approach. Here we propose BSE@qKI, the direct analogue in which the eigenvalues are replaced by qKI ones.
Making this replacement requires some care. As discussed in Section II A, diagonalizing the qKI Hamiltonian yields eigenvectors that in general differ from the DFT ones, whereas BSE@ reconstructs the DFT eigenvalues perturbatively and builds the BSE basis out of products of occupied and empty KS-DFT eigenvectors. As an original contribution of this work, we extend qKI to a perturbative formulation that preserves the DFT orbitals while incorporating the qKI corrections at first order. Starting from the matrix elements of the qKI potential on the variational orbitals, Eq. (5), we define the Hermitian operator
whose matrix elements can be computed in any basis. The perturbative qKI quasiparticle energies then follow as the first-order correction to the KS-DFT eigenvalues,
where the are the KS-DFT eigenvectors of the underlying DFT energy functional. All the qKI results reported below are obtained in this perturbative form. BSE@qKI thus mirrors BSE@ step by step, while retaining the computational advantages of a Koopmans-compliant description of quasiparticle energies. Since the two schemes share the same KS wavefunctions and the same BSE kernel, we expect them to capture the same excitonic physics and, in particular, to give comparable exciton energies; any difference in the excitation energies should then be traceable to the quasiparticle energies alone. § 2 B tests this expectation. A pioneering implementation of BSE on top of Koopmans-compliant energies was proposed by Elliott et al. [31]. There, quasiparticle energies from the KI functional are used while the explicit calculation of the screening matrix is bypassed through an iterative minimization scheme on an MLWF basis, as implemented in the Gwl code [31] of Quantum ESPRESSO. That formulation was, however, limited to finite systems. Here we extend the combination of Koopmans-compliant quasiparticle energies and BSE to periodic solids and provide an automated workflow for the entire calculation, enabling its systematic application to crystalline materials. All the BSE and calculations reported in this work were performed with the Yambo code [51, 52].
Results
Automated workflow for BSE@qKI
The BSE@qKI workflow, sketched in Fig. 1, consists of the following steps:
convergence of the BSE eigenvalues with respect to relevant YAMBO parameters;
calculation of the Koopmans screening coefficients ;
calculation of the qKI quasiparticle eigenvalues on the converged k-point mesh;
conversion of the Koopmans data into YAMBO-readable format via the k2y interface [53];
the final BSE calculation using the qKI quasiparticle energies;
optionally, Wannier interpolation of the qKI band structure.
The first step establishes the numerical parameters required for a converged BSE calculation. In particular, the excitonic energies are converged with respect to the number of bands , the plane-wave cutoff and the k-point sampling used for the screening matrix and the BSE kernel (Eq. (6)) [16]. The k-point convergence is particularly important, because it controls both the sampling of the BZ and the resolution with which electron-hole transitions are represented in reciprocal space: the BSE Hamiltonian is formulated in the space of electron-hole transitions, which is defined by the selected valence and conduction bands and by the k-point mesh. The maximum energy difference between the KS states included in this transition space is chosen according to the spectral region of interest. Once the BSE parameters have been converged, the resulting k-point mesh is used for the subsequent Koopmans calculation. The screening coefficients can be determined on a coarser k-point grid, since their convergence is independent of the numerical parameters governing the BSE calculation. The qKI Hamiltonian is then solved on the converged BSE mesh, yielding the quasiparticle eigenvalues required for the final BSE calculation. These eigenvalues are then converted into the YAMBO format using the dedicated k2y interface. Starting from the original KS wavefunctions stored in the ns.db1 YAMBO database and from the qKI output, k2y generates a new ndb.QP database containing the qKI quasiparticle energies. These can then be incorporated directly into the diagonal part of the BSE Hamiltonian within the standard YAMBO workflow, without modifying the underlying KS wavefunctions or the BSE kernel. Finally, a full BSE calculation is performed using the qKI quasiparticle energies. The BSE Hamiltonian is constructed from the qKI-corrected electron-hole transition energies and the corresponding interaction kernel; diagonalizing it provides the excitonic eigenstates and hence the optical observables, including absorption and energy-loss spectra. As an optional post-processing step, the qKI band structure can be interpolated using MLWFs. Wannier interpolation enables the evaluation of the quasiparticle band structure along arbitrary paths in reciprocal space, including conventional high-symmetry directions. This step does not affect the BSE or the resulting optical properties, as it is performed independently on the converged qKI band structure and serves only as a means of obtaining a smooth representation of the quasiparticle dispersion.
This workflow closely parallels a conventional BSE@ calculation, but the computational cost is distributed differently. In BSE@, the cost is dominated by the step, whose convergence depends on several coupled parameters, including the number of empty states and the plane-wave cutoff entering the calculation of the polarizability and the self-energy [16]. In practice, the parameters required to converge the quasiparticle energies are typically reused in the subsequent BSE calculation, which reduces the additional convergence effort associated with the BSE step; the overall cost nonetheless remains dominated by . Even when plasmon-pole approximations are employed [54, 55], the calculation involves frequency-dependent quantities, whereas the BSE kernel considered here is static, so the BSE calculation is comparatively cheap. Koopmans-compliant functionals, by contrast, provide quasiparticle corrections within a variational functional framework and never require the explicit evaluation of the frequency-dependent self-energy. This makes quasiparticle energies much cheaper to obtain, while retaining — as demonstrated in the following sections — a level of accuracy suitable for the description of optoelectronic properties.

Figure 1. Schematic representation of the BSE@qKI workflow. The workflow comprises the convergence of the BSE parameters, the determination of the Koopmans screening coefficients , the calculation and conversion of qKI quasiparticle energies, and the final BSE calculation. Wannier interpolation of the qKI band structure is optionally performed as a post-processing step.
The complete workflow involves several electronic-structure codes: Quantum ESPRESSO is used for the underlying DFT and Koopmans-compliant calculations, Wannier90 [56] for the construction of maximally localized Wannier functions, and Yambo for the many-body perturbation theory calculations. The k2y interface, developed within this work, provides the connection between the Koopmans and Yambo workflows by converting the qKI quasiparticle energies into a Yambo-readable ndb.QP quasiparticle database. Running these steps by hand is impractical: they involve data transfer between different codes, multiple convergence studies, and the execution of a substantial number of individual calculations (typically on the order of 30 per material). Automation is therefore essential for ensuring reproducibility and for reducing the burden associated with workflow management.
ated with workflow management. To this end, we employ the AiiDA workflow engine [24, 25, 26] and its ecosystem of plugins, including aiida-quantumespresso [57], aiida-wannier90-workflows [58], aiida-yambo [16, 52], and aiida-yambo-wannier90 [16]. The yambopy package [59] is used for post-processing and analysis of the Yambo results. Koopmans-compliant calculations are managed through the koopmans Python package [22], which provides automated workflows for both the reciprocal-space implementation based on density-functional perturbation theory (DFPT) [35, 60] and the finite-difference implementation based on the SCF approach [21]. In this work, we further extend this framework through the development of the aiida-koopmans plugin [61], which brings Koopmans calculations into the AiiDA ecosystem. This provides capabilities such as remote execution, parallelization of independent tasks (, the individual screening calculations), and automated error handling.
The resulting workflow is reproducible and, in principle, suitable for high-throughput studies. The main user-defined inputs are the atomic structure, the Wannier projections, and the available computational resources; most of the remaining parameters can be generated automatically from predefined protocols [62]. Two obstacles still stand in the way of complete automation. The first is the need for user-defined Wannier projections. In future work, this requirement will be removed by integrating the present workflow with recently developed automated Wannierization strategies based on projectability disentanglement and manifold remixing [63, 64], thereby enabling the automated construction of Wannier functions spanning the target valence and conduction manifolds. The second is that, although well-established protocols exist for many DFT parameters, the convergence parameters for many-body perturbation theory calculations remain strongly system-dependent, and so must still be converged case by case.
Validation on well-known semiconductors
In this section, we assess the performance of BSE@qKI for a representative set of well-known bulk semiconductors: silicon (Si), zinc oxide (ZnO), rutile titanium dioxide (TiO), diamond and hexagonal boron nitride (hBN). The central goal is to determine whether the use of qKI quasiparticle energies within the BSE framework provides excitonic properties comparable to those obtained from the established BSE@GW approach. For each material, we first calculate the quasiparticle band structure using both qKI and GW, and subsequently solve the BSE on top of the corresponding quasiparticle energies. All calculations employ PseudoDojo v0.4 pseudopotentials [65], within the local-density approximation (LDA) for the exchange-correlation functional. The default Quantum ESPRESSO parameters provided by the aiida-quantumespresso workflows are adopted through their predefined protocols [62]. The parameters entering the qKI calculations are selected to ensure convergence of both the Wannier representation and the Koopmans screening coefficients, whereas the GW calculations use parameters previously converged by the authors [16]. The BSE calculations employ the same numerical parameters as the corresponding GW calculations, with the exception of the Brillouin-zone sampling, whose convergence is treated separately as described in § 2 A. Within the automated YamboConvergence workflow, the BSE calculations are converged by imposing a relative threshold of on the lowest excitonic eigenvalue. This threshold provides a practical compromise between computational cost and accuracy and can be tightened when higher numerical accuracy is required. In addition to the convergence of the lowest excitonic state, the optical spectrum generally requires convergence with respect to the number of valence and conduction bands included in the BSE Hamiltonian. This condition depends on the spectral range of interest and is therefore treated separately. In the present work, we focus on the low-energy region of the optical spectra and retain a limited number of low-energy transitions. Furthermore, we restrict the analysis to optically active direct excitons, also for materials whose fundamental band gap is indirect. The detailed computational parameters used for each material are reported in the Supplementary Information (SI). The purpose of this benchmark is to assess the ability of the proposed BSE@qKI approach to reproduce the results of the established BSE@GW methodology across a chemically and structurally diverse set of systems. The convergence parameters adopted here provide the level of accuracy required for this comparison. For material-specific applications targeting finer, meV-scale energy differences, these parameters can be systematically tightened and a larger number of excitonic states considered.
The main results are summarized in tbl:I. For each material, we report the minimum quasiparticle band gap and the energy of the first excitonic peak. For comparison, the quasiparticle self-consistent with vertex corrections in the screened Coulomb interaction (QSG) band gaps are taken from the literature [66], while the remaining theoretical results are obtained in the present work. Since no explicit BSE@QSG calculations are available, the corresponding values are estimated by applying the QSG correction to the obtained from GW, thereby accounting for the change in the quasiparticle gap while keeping the exciton binding energy unchanged. We underline here that this is an auxiliary estimate, not a BSE calculation performed with QSG energies, orbitals and screening. Its use assumes that the relevant direct transitions change approximately uniformly. Experimental band gaps are corrected for zero-point renormalization (ZPR) by applying the corresponding ZPR reported in the literature [67]. No additional ZPR correction is applied to the exciton binding energies, as previous studies have shown that their ZPR is typically substantially smaller than the electronic contribution to the band-gap renormalization [68, 69, 70]. We first consider the minimum quasiparticle band gaps reported in tbl:I. Across the five materials, qKI and GW yield similar mean absolute relative error (MARE) with respect to experiment, respectively 13% and 10%, while the comparison between experiment and QSG gives a MARE of 6%. This similar average performance of qKI and GW, however, results from rather different material-dependent trends. In particular, qKI performs very well for ZnO, where its gap is close to both experiment and QSG, while GW substantially underestimates it. The opposite behavior is found for diamond, where GW is close to both experiment and QSG, whereas qKI significantly overestimates the gap. Overall, the gap comparison does not reveal a systematic advantage of one quasiparticle approach over the other. Importantly, the MARE reported here should be regarded only as a rough comparative metric and not as a statistically significant measure of the relative performance of the different approaches. Indeed, the comparison is sensitive to individual materials: when ZnO is excluded, GW shows substantially better agreement with experiment than qKI, with MAREs of 4% and 16%, respectively. The observed differences reflect the distinct approximations underlying the three quasiparticle approaches. The performance of Koopmans-compliant functionals for quasiparticle energies has been investigated in a range of benchmark studies from the authors [21, 34, 35]. The theoretical framework underlying the corresponding screening and quasiparticle corrections is discussed in Section II A, with further details provided in Refs. 27, 32, 37.
We next consider the energy of the first excitonic peak. For this analysis, it is important to distinguish the fundamental quasiparticle gap from the direct gap entering the BSE calculation. The former, reported in Table I (PDF p. 14), is the minimum gap over the Brillouin zone and can be indirect, whereas the BSE solved here describes vertical electron-hole transitions () and therefore depends on direct quasiparticle transitions. The corresponding minimum direct gaps, , are reported in the SI. A clear improvement of BSE@qKI over BSE@ emerges for the first excitonic peak. The MARE of with respect to experiment is 4% for BSE@qKI, compared with 16% for BSE@. The corresponding QSG estimate gives a MARE of 4%. Remarkably, BSE@qKI yields a first peak closer to experiment than BSE@ for all five materials considered. The origin of this systematic difference becomes clearer when considering the direct quasiparticle gaps. For Si, qKI and give and 3.25 eV, respectively, while the corresponding first excitonic peaks are 3.55 and 3.35 eV, so that the two differences, 0.19 and 0.20 eV, agree closely. For ZnO, the direct gap is 3.60 eV within qKI and 2.32 eV within , leading to first peaks at 3.43 and 2.16 eV, respectively, so that the two differences, 1.28 and 1.27 eV, again agree closely. Similarly, for hBN and diamond, the differences between the two approaches in closely follow those observed for the direct quasiparticle gaps: the direct gaps differ by 1.15 eV and 1.04 eV, respectively, while the corresponding differences in are 1.27 and 1.15 eV. A similar behavior is observed for TiO, where the direct gaps and the first peaks both differ by 0.82 eV. These results indicate that the difference between the BSE@qKI and BSE@ spectra is predominantly determined — as anticipated in Section II C — by the different quasiparticle transition energies entering the diagonal part of the BSE Hamiltonian. The electron-hole interaction kernel, by contrast, is constructed from the same underlying DFT quantities in the two approaches and therefore provides a similar excitonic renormalization (i.e. a similar EBE, as reported in the SI). The better agreement of BSE@qKI with the optical excitation energy is connected to its direct quasiparticle transition energies rather than to a uniformly more accurate fundamental gap.
| Material | k-mesh | ||
| Silicon | 400 | 16 | 12 x 12 x 12 |
| ZnO | 800 | 28 | 12 x 12 x 6 |
| hBN | 800 | 20 | 12 x 12 x 4 |
| Diamond | 300 | 20 | 8 x 8 x 8 |
| TiO | 600 | 12 | 8 x 8 x 12 |
| CrI | 800 | 28 | 4 x 4 x 4 |
Table I (PDF p. 14). Convergence parameters for the Yambo calculations.
As an illustrative example, Fig. 2 shows the absorption spectrum of bulk silicon obtained with BSE@qKI and BSE@ (shifted by the corresponding ), together with the experimental and RPA@qKI spectra. Both theoretical approaches reproduce the overall low-energy spectral structure, while BSE@qKI gives a first peak closer to the experimental position. The electron-hole transition amplitudes associated with this state are shown in the right panel, providing insight into the reciprocal-space character of the corresponding excitation. The equivalent plots for the remaining materials are collected in the SI. The EBEs, reported in the SI, are also in good agreement between qKI and , with a MARE of 10%, consistent with the use of the same underlying DFT electronic structure in the construction of the BSE electron-hole interaction. We do not compare the calculated EBEs with experiment at this stage, as this would require a more stringent convergence of the electron-hole transition space, particularly for excitons involving bands with high curvature near the band gap, such as in ZnO [73], and for systems with an indirect fundamental gap, such as Si and hBN.
Electronic and optical properties of bulk CrI
Chromium triiodide (CrI) is a prototypical two-dimensional magnetic material, and was the first material reported to retain intrinsic ferromagnetic order down to the monolayer limit [80, 81]. Its layered structure, held together by weak van der Waals interactions, makes it readily exfoliable and particularly attractive for spintronic and magnetoelectronic applications [81, 82, 83, 84, 85, 86, 87, 88, 89]. Among the chromium trihalides Cr (), CrI has the highest Curie temperature, K, and a strong magnetic anisotropy [23]. Bulk CrI also undergoes a temperature-driven structural transition, from a high-temperature monoclinic phase to a low-temperature rhombohedral phase with ABC stacking [23]. The electronic and optical properties of CrI have been investigated extensively using first-principles approaches [37, 90, 91, 92, 93, 94, 95], with reported quasiparticle band gaps showing a sizeable dependence on the theoretical framework and structural configuration. In particular, QSG calculations for the hexagonal hP structure have reported bulk band gaps in the range 2.2–2.5 eV [92, 94], while a Koopmans-compliant calculation for the rhombohedral structure reported a gap of 1.92 eV [37]. Part of this difference may therefore originate from the different structural configurations considered.

Figure 2. Electronic and optical properties of bulk silicon. Left: absorption spectrum obtained with BSE@qKI, BSE@, RPA@qKI and from experimental data. The BSE@ spectrum has been shifted by meV to align the gap with the QSG one, while the experimental one is shifted by meV to be compared with the theoretical results. The BSE spectra are obtained with a broadening of 100 meV, comparable with the full width at half maximum of the experimental peaks. The spectra have been normalized with respect to the highest detected peak. Right: electron-hole transition amplitudes contributing to the excitonic peak .
In the following we focus on the phase: it is the low-temperature structure, and hence the one relevant to the low-temperature optical measurements against which we compare. All the simulations are performed using non-collinear (NC) magnetism with spin-orbit coupling (SOC), which has been shown to reduce the band gap by meV [92]. Following the BSE@qKI approach, we first compute the band structure of CrI. At the qKI level of theory, we obtain an indirect band gap of eV, as shown in the right panel of Fig. 3 and in agreement with previous Koopmans predictions [37]. We then perform convergence tests for the optical properties, imposing a threshold of on the first excitonic eigenvalue. The resulting converged set of parameters is presented in the SI. The resulting absorption spectrum is shown in the left panel of Fig. 3, together with experimental differential reflection measurements, as presented in Ref. 96. To facilitate the comparison, the theoretical spectra are rigidly red-shifted by meV so that its onset aligns with the experimental one. This shift should be regarded as a crude alignment rather than
a quantitative correction to the theoretical spectrum; it may effectively account for effects not included in the present calculations, such as ZPR of the gap and other experimental or theoretical energy shifts. The experimental and theoretical spectra are normalized to their respective maximum intensities. Once is applied, the BSE@qKI results reproduce well the peak positions in the low-energy region of the spectrum. Structures at , 1.8, and 1.9–2.0 eV are correctly captured, with discrepancies smaller than 20 meV, an order of magnitude below the convergence threshold (160 meV). In contrast, a larger deviation of 200 meV is observed for the highest-energy peak, with the computed value at 2.43 eV compared to the experimental one at 2.65 eV. This larger discrepancy may partly reflect the more demanding convergence requirements associated with higher-energy excitonic states; a systematic description of the full high-energy spectrum, including the convergence of all relevant electron-hole transitions, is beyond the scope of the present work. Comparison with differential reflection measurements also becomes non-trivial above the continuum onset, due to the increasingly complex dependence of the reflectance on the full dielectric function, as well as possible interference and surface effects (e.g. plasmons).
The first calculated excitonic peak occurs at eV, which appears at 1.48 eV in the shifted spectrum of Fig. 3. The corresponding exciton has a large binding energy, of approximately 0.5 eV. Here, as for the indirect-gap systems discussed above, the binding energy must be referred to the relevant direct quasiparticle transitions entering the BSE rather than to the fundamental gap. Its magnitude is consistent with the quasi-two-dimensional character and the weak dielectric screening associated with the layered structure of CrI. The reciprocal-space composition of this excitonic state, shown in the right panel of Fig. 3, provides further insight into its character.

Figure 3. Absorption spectrum and E exciton composition for CrI. Left: comparison between the BSE@qKI (green) and RPA (blue) absorption spectra, both rigidly shifted down by meV, and the experimental (black) differential reflection measurements of CrI [96]. The spectra have been normalized with respect to the highest included peak. Right: band structure of CrI and electron-hole composition for the first peak of the Im(), E=1.68 eV.
Conclusions
In this work, we have developed and demonstrated an automated workflow for calculating excitonic properties within the BSE using Koopmans-compliant quasiparticle energies, thereby avoiding the computationally demanding step typically employed in standard BSE@ calculations. The complete BSE@qKI workflow has been implemented within the AiiDA ecosystem, enabling automated execution, provenance tracking, reproducibility, and sharing of the calculations, while individual workflow components can also be used independently of AiiDA. As part of this work, we have developed the k2y interface, which provides an efficient and direct conversion between Koopmans and Yambo data formats and allows qKI quasiparticle energies to be incorporated into the standard Yambo BSE workflow. We first validated the approach on a set of well-established bulk semiconductors. BSE@qKI provides excitonic peak energies in good agreement with experiment, and exciton binding energies in good agreement with BSE@. The differences between the two BSE schemes are primarily associated with the quasiparticle transition energies entering the diagonal part of the BSE Hamiltonian, whereas the electron-hole interaction gives a similar excitonic renormalization in both. Finally, we applied the workflow to bulk ferromagnetic CrI, including non-collinear magnetism and spin-orbit coupling. The BSE@qKI absorption spectrum reproduces the main features of the experimental low-energy spectrum, demonstrating that the approach can be extended beyond non-magnetic benchmark semiconductors to more complex magnetic materials. The first excitonic state is characterized by a large binding energy, highlighting the relevance of electron-hole interactions in this class of layered materials. Overall, these results establish BSE@qKI as a -free alternative to BSE@ for the calculation of excitonic and optical properties. Further work is needed to assess the performance of the approach across a broader range of materials and more demanding convergence regimes; the present workflow provides the automated and reproducible framework with which to carry out these systematic studies.
Data availability
The data presented in this work are available on the Materials Cloud Archive [97] at the URL: https://doi.org/10.24435/materialscloud:jp-07 [97].
Code availability
The aiida-koopmans plugin is openly available on GitHub at https://github.com/mikibonacci/aiida-koopmans, and the k2y code is openly available on GitHub at https://github.com/mikibonacci/k2y.
Acknowledgments
We thank Giovanni Pizzi, Xing Wang, Jusong Yu and Nataliya Paulish for helpful discussions and valuable comments on the manuscript. We acknowledge financial support from the NCCR MARVEL, a National Centre of Competence in Research, funded by the Swiss National Science Foundation (grant number 205602), and from the SwissTwins project, funded by the Swiss State Secretariat for Education, Research and Innovation (SERI). We acknowledge access to Piz Daint and Alps at the Swiss National Supercomputing Centre (CSCS) under PSI’s share with project ID psi15 and psi18 and under MARVEL share with project ID mr32.
Supplementary Material: Koopmans spectral functionals for excitons in materials
Miki Bonacci; Edward Linscott; Julian Geiger; Nicola Marzari; Nicola Colonna
paper_SI.pdf
Convergence parameters
Here we present the convergence parameters used for the Bethe-Salpeter equation (BSE) calculations. These are obtained via the aiida-yambo1 built-in workflows. In particular, the cutoffs for the summation over empty states and plane waves are taken from Ref. 1 (except for CrI, fully converged here), while the k-meshes are computed within this work. The calculations are converged by imposing a relative threshold of on the lowest excitonic eigenvalue . For this step, the quasiparticle corrections are simulated by applying a scissor-and-stretching operator to the Kohn-Sham (KS) eigenvalues, so as to reproduce the value of the band gap. This approximation is sufficient for the convergence tests.
Relation between the band gaps and the excitonic peaks
In Table II, we compare the minimum (fundamental) band gap , the minimum direct band gap , the energy of the lowest excitonic peak , and the EBE associated with this excitonic state. Here, the EBE is defined as the energy difference between the independent-particle electron–hole transition associated with the excitonic state and the corresponding interacting BSE eigenvalue, as described in the main text. It therefore does not necessarily coincide with the difference between the minimum direct gap and , since the lowest excitonic state may involve transitions away from the point in the Brillouin zone (BZ) at which the minimum direct gap occurs. Nevertheless, a clear correspondence emerges between the differences in the direct quasiparticle gaps and those in the excitonic peak energies. Since the BSE describes vertical electron-hole transitions, the quasiparticle transition energies entering its diagonal part correspond to direct transitions, while the electron-hole interaction gives rise to the excitonic renormalization. The EBE can therefore be viewed, to a good approximation, as the red shift of the relevant direct quasiparticle transition associated with the excitonic state. The close agreement between the EBE obtained with qKI and consequently indicates that the differences in primarily reflect differences in the corresponding direct quasiparticle transition energies. The hBN case provides a particularly clear illustration of this behavior. The minimum direct gaps obtained with qKI and differ by 1.15 eV, while the corresponding first excitonic peaks differ by 1.27 eV. In contrast, the associated EBEs differ by only 0.11 eV. The substantial discrepancy in is therefore due to the different direct quasiparticle transition energies entering the BSE, rather than to a significantly different electron-hole interaction. A similar trend is observed for diamond. These results further support the picture that, within the level of convergence adopted here, the difference between the BSE@qKI and BSE@ excitonic energies is predominantly inherited from the different quasiparticle transition energies, while the electron-hole interaction produces a similar renormalization in the two approaches. The origin of the remaining discrepancy between the qKI and direct gaps, particularly for diamond, requires further investigation and is left for future work.
| Material | (eV) qKI | (eV) | (eV) qKI | (eV) | (eV) qKI | (eV) | EBE (eV) qKI | EBE (eV) |
| Silicon | 1.32 | 1.17 | 3.44 | 3.25 | 3.55 | 3.35 | 0.08 | 0.09 |
| ZnO | 3.6 | 2.32 | 3.6 | 2.32 | 3.43 | 2.16 | 0.19 | 0.20 |
| hBN | 7.10 | 6.15 | 7.61 | 6.46 | 6.78 | 5.51 | 1.0 | 1.11 |
| Diamond | 7.0 | 5.58 | 8.32 | 7.28 | 8.13 | 6.98 | 0.35 | 0.30 |
| TiO | 4.07 | 3.21 | 4.07 | 3.25 | 4.73 | 3.91 | 0.10 | 0.11 |
Table II. Comparison of quadratic Koopmans-integral (qKI) and results for the minimum (fundamental) band gap , the minimum direct band gap , the energy of the lowest excitonic peak , and the corresponding exciton binding energy (EBE). The EBE is defined with respect to the independent-particle electron–hole transition associated with the excitonic state and therefore does not necessarily coincide with . The slightly indirect nature of the band gap of TiO has been previously reported.2
It is worth noting that the convergence requirements for the EBE can be more stringent than those for the excitonic peak energy, particularly for systems with small binding energies. A more systematic and tighter convergence of the electron-hole transition space would therefore be required for a quantitative assessment of the EBE, but this is beyond the scope of the present work. Our aim here is instead to establish that BSE@qKI can reproduce the excitonic properties obtained with the standard BSE@ approach within the accuracy and level of convergence adopted in this study.
Absorption spectra and lowest exciton composition for all the materials studied in this work
In the following we show the absorption spectra (the imaginary part of the macroscopic dielectric function ) obtained from the BSE, starting from the band structures computed via qKI and . The analogous random-phase approximation (RPA) spectrum (i.e., without the electron-hole interaction kernel) is also provided for the qKI case, together with experimental spectra as taken from the literature (see Table 1 of the main text).
Table 1. Minimum band gaps (), first exciton energies (), and zero-point renormalization (ZPR) for the validated semiconductors. Experimental and values are corrected for the ZPR, for comparison with theoretical results that do not include it. The QSG values are estimated by shifting the spectrum to align . The mean absolute relative error (MARE) is computed with experiment as reference. QSG and ZPR data are from Refs. 66 and 67, respectively, unless otherwise cited. Experimental values are extracted from plots in the cited references.

Figure 1 (PDF p. 15). Absorption spectrum (left panel) and composition of the first peak, (right panel), for Silicon, plotted on top of the qKI interpolated band structure. The experimental spectrum is shifted by -ZPR to be compared with theoretical results.

Figure 2 (PDF p. 16). Absorption spectrum (left panel) and composition of the first peak, (right panel), for ZnO, plotted on top of the qKI interpolated band structure. The experimental spectrum is shifted by -ZPR to be compared with theoretical results.

Figure 3 (PDF p. 16). Absorption spectrum (left panel) and composition of the first peak, (right panel), for hBN, plotted on top of the qKI interpolated band structure. The experimental spectrum is shifted by -ZPR to be compared with theoretical results.

Figure 4. Absorption spectrum (left panel) and composition of the first peak, (right panel), for Diamond, plotted on top of the qKI interpolated band structure. The experimental spectrum is shifted by -ZPR to be compared with theoretical results.

Figure 5. Absorption spectrum (left panel) and composition of the first peak, (right panel), for TiO, plotted on top of the qKI interpolated band structure. The experimental spectrum is shifted by -ZPR to be compared with theoretical results.

Figure 6. Absorption spectrum and composition of the first peak, , for CrI, plotted on top of the qKI interpolated band structure. Here, at variance with the other cases, the theoretical spectra are shifted by to align with the experimental measurement.
Wannier-interpolated band structure for all the materials studied in this work
Here we present the results for the interpolated band structures. The interpolation is done via the aiida-yambo-wannier901 plugin. We note, in particular, the inaccuracy of the interpolation for the band structure of ZnO, which is mostly due to the well-known inaccuracy of standard in this particular case.3 The BSE@ results are not affected by this, however, as they are mainly concentrated on the band structure close to the Fermi energy around the point.
References

Figure 7. Interpolated band structure of Silicon using different methods: DFT (QE), qKI Koopmans functionals (KI) and . The plot also shows the Quantum ESPRESSO bands computed on the same path.

Figure 8. Interpolated band structure of ZnO using different methods: DFT (QE), qKI Koopmans functionals (KI) and . The plot also shows the Quantum ESPRESSO bands computed on the same path.

Figure 9. Interpolated band structure of hBN using different methods: DFT (QE), qKI Koopmans functionals (KI) and . The plot also shows the Quantum ESPRESSO bands computed on the same path.

Figure 10. Interpolated band structure of Diamond using different methods: DFT (QE), qKI Koopmans functionals (KI) and . The plot also shows the Quantum ESPRESSO bands computed on the same path.

Figure 11. Interpolated band structure of TiO₂ using different methods: DFT (QE), qKI Koopmans functionals (KI) and . The plot also shows the Quantum ESPRESSO bands computed on the same path.

Figure 12. Interpolated band structure of CrI₃ using the qKI method. The plot also shows the Quantum ESPRESSO bands computed on the same path.
References
- [1]N. Marzari, A. Ferretti, and C. Wolverton, Electronic-structure methods for materials design, Nature Materials 20, 736–749 (2021).DOI
- [2]A. Balapure, J. Ray Dutta, and R. Ganesan, Recent advances in semiconductor heterojunctions: a detailed review of the fundamentals of photocatalysis, charge transfer mechanism and materials, RSC Applied Interfaces 1, 43 (2024).
- [3]W. Wang, S. Gao, J. He, J. Cui, and H. Ji, Exciton engineering in organic photocatalysis: mechanisms and applications, Environmental Functional Materials 5, 43 (2026).DOI
- [4]M. Kmichou, R. Masrour, L. Xu, E. Hlil, and A. Rezzouk, Evaluation of the phonon, optoelectronic, photovoltaic, thermoelectric, photocatalytic and thermodynamic properties of the inorganic perovskite cspbi2br using ab initio calculations, Solar Energy 279, 112812 (2024).
- [5]V. K. Solet and S. K. Pandey, Many-body ab initio study of quasiparticles, optical excitations, and excitonic properties in LiZnAs and ScAgC for photovoltaic applications, Phys. Rev. Appl. 23, 064040 (2025).
- [6]S. Sharifzadeh, Many-body perturbation theory for understanding optical excitations in organic molecules and solids, Journal of Physics: Condensed Matter 30, 153002 (2018).
- [7]M. Kumar, A. Singh, D. Gill, and S. Bhattacharya, Optoelectronic properties of chalcogenide perovskites by many-body perturbation theory, The Journal of Physical Chemistry Letters 12, 5301–5307 (2021).
- [8]H. Yu, Y. Peng, Y. Yang, and Z.-Y. Li, Plasmon-enhanced light–matter interactions and applications, npj Computational Materials 5, 45 (2019).
- [9]X. Yi, Plasmon–exciton coupling in two-dimensional materials: fundamentals and applications in nanosensors, Sensor Review 46, 604 (2026).DOI
- [10]T. M. R. Wolf, T. Xie, C. Jin, and A. H. MacDonald, Exciton-based sensing of remote electron correlations in 2d heterostructures, Phys. Rev. Lett. (2026).
- [11]G. Strinati, Application of the green’s functions method to the study of the optical properties of semiconductors, La Rivista del Nuovo Cimento (1978-1999) 11, 1 (1988).
- [12]G. Onida, L. Reining, and A. Rubio, Electronic excitations: density-functional versus many-body Green’s-function approaches, Rev. Mod. Phys. 74, 601 (2002).
- [13]L. Hedin, New Method for Calculating the One-Particle Green’s Function with Application to the Electron-Gas Problem, Phys. Rev. 139, 796 (1965).
- [14]M. J. van Setten, M. Giantomassi, F. Gonze, G.-M. Rignanese, and G. Hautier, Automation methodologies and large-scale validation for gw: Towards high-throughput gw calculations, Phys. Rev. B 96, 155207 (2017).
- [15]A. Rasmussen, T. Deilmann, and K. S. Thygesen, Towards fully automated gw band structure calculations: What we can learn from 60.000 self-energy evaluations, npj Computational Materials 7, 22 (2021).
- [16]M. Bonacci, J. Qiao, N. Spallanzani, A. Marrazzo, G. Pizzi, et al., Towards high-throughput many-body perturbation theory: efficient algorithms and automated workflows, npj Computational Materials 9, 74 (2023).
- [17]M. Großmann, M. Gruner, and E. Runge, A robust, simple, and efficient convergence workflow for GW calculations, npj Computational Materials 10, 135 (2024).DOI
- [18]M. Grunert, M. Großmann, and E. Runge, Machine learning climbs the jacob’s ladder of optoelectronic properties, Nature Communications 16, 8142 (2025).DOI
- [19]I. Dabo, A. Ferretti, N. Poilvert, Y. Li, N. Marzari, and M. Cococcioni, Koopmans’ condition for density-functional theory, Phys. Rev. B 82, 115121 (2010), One of the founding Koopmans functionals papers.
- [20]G. Borghi, A. Ferretti, N. L. Nguyen, I. Dabo, and N. Marzari, Koopmans-compliant functionals and their performance against reference molecular data, Phys. Rev. B 90, 075135 (2014), One of the founding Koopmans functionals papers.
- [21]N. L. Nguyen, N. Colonna, A. Ferretti, and N. Marzari, Koopmans-compliant spectral functionals for extended systems, Phys. Rev. X 8, 021051 (2018), Describes Koopmans functionals in periodic systems.
- [22]E. B. Linscott, N. Colonna, R. De Gennaro, N. L. Nguyen, G. Borghi, et al., Koopmans: An Open-Source Package for Accurately and Efficiently Predicting Spectral Properties with Koopmans Functionals, J. Chem. Theory Comput. (2023), Introduces the koopmans code.
- [23]M. A. McGuire, H. Dixit, V. R. Cooper, and B. C. Sales, Coupling of crystal structure and magnetism in the layered, ferromagnetic insulator cri3, Chemistry of Materials 27, 612–620 (2015).
- [24]G. Pizzi, A. Cepellotti, R. Sabatini, N. Marzari, and B. Kozinsky, AiiDA: automated interactive infrastructure and database for computational science, Computational Materials Science 111, 218 (2016).
- [25]P. Huber, S. Zoupanos, M. Uhrin, I. Talirz, L. Kahle, et al., AiiDA 1.0, a scalable computational infrastructure for automated reproducible workflows and data provenance, Scientific Data 7, 300 (2020).
- [26]M. Uhrin, S. P. Huber, Y. Yu, N. Marzari, and G. Pizzi, Workflows in AiiDA: Engineering a high-throughput, event-based engine for robust and modular computational workflows, Computational Materials Science 187, 110086 (2021).
- [27]N. Colonna, N. L. Nguyen, A. Ferretti, and N. Marzari, Screening in orbital-density-dependent functionals, Journal of Chemical Theory and Computation 14, 2549–2557 (2018).DOI
- [28]Y. Schubert, N. Marzari, and E. Linscott, Testing Koopmans spectral functionals on the analytically solvable Hooke’s atom, The Journal of Chemical Physics 158, 144113 (2023).
- [29]N. L. Nguyen, G. Borghi, A. Ferretti, I. Dabo, and N. Marzari, First-Principles Photoemission Spectroscopy and Orbital Tomography in Molecules from Koopmans-Compliant Functionals, Physical Review Letters 114, 166405 (2015).
- [30]N. L. Nguyen, G. Borghi, A. Ferretti, and N. Marzari, First-Principles Photoemission Spectroscopy of DNA and RNA Nucleobases from Koopmans-Compliant Functionals, Journal of Chemical Theory and Computation 12, 3948 (2016).
- [31]J. D. Elliott, N. Colonna, M. Marsili, N. Marzari, and P. Umari, Koopmans Meets Bethe–Salpeter: Excitonic Optical Spectra without GW, Journal of Chemical Theory and Computation 15, 3710 (2019).DOI
- [32]N. Colonna, N. L. Nguyen, A. Ferretti, and N. Marzari, Koopmans-compliant functionals and potentials and their application to the gw100 test set, Journal of Chemical Theory and Computation 15, 1905–1914 (2019).
- [33]J. M. de Almeida, N. L. Nguyen, N. Colonna, W. Chen, C. Rodrigues Miranda, et al., Electronic Structure of Water from Koopmans-Compliant Functionals, Journal of Chemical Theory and Computation 17, 3923 (2021).DOI
- [34]R. De Gennaro, N. Colonna, E. Linscott, and N. Marzari, Bloch’s theorem in orbital-density-dependent functionals: Band structures from koopmans spectral functionals, Phys. Rev. B 106, 035106 (2022).DOI
- [35]N. Colonna, R. De Gennaro, E. Linscott, and N. Marzari, Koopmans spectral functionals in periodic boundary conditions, Journal of Chemical Theory and Computation 18, 5435–5448 (2022).arxiv.org/abs/2202.08155
- [36]J. E. Ingall, E. Linscott, N. Colonna, A. J. Page, and V. J. Keast, Accurate and Efficient Computation of the Fundamental Bandgap of the Vacancy-Ordered Double Perovskite Cs2TiBr6, The Journal of Physical Chemistry C 128, 9217 (2024), Publisher: American Chemical Society.
- [37]A. Marrazzo and N. Colonna, Spin-dependent interactions in orbital-density-dependent functionals: Non-collinear koopmans spectral functionals, Phys. Rev. Res. 6, 033085 (2024).DOI
- [38]Y. Schubert, S. Luber, N. Marzari, and E. Linscott, Predicting electronic screening for fast Koopmans spectral functional calculations, npj Computational Materials 10, 299 (2024), Publisher: Nature Publishing Group.DOI
- [39]M. Stojkovic, E. Linscott, and N. Marzari, Predicting the suitability of photocatalysts for water splitting using Koopmans spectral functionals: The case of tio2 polymorphs, Physical Review Materials 10, 025403 (2026).DOI
- [40]J. P. Perdew, R. G. Parr, M. Levy, and J. L. Balduz, Density-Functional Theory for Fractional Particle Number: Derivative Discontinuities of the Energy, Physical Review Letters 49, 1691 (1982).DOI
- [41]J. P. Perdew and A. Zunger, Self-interaction correction to density-functional approximations for many-electron systems, Physical Review B 23, 5048 (1981).DOI
- [42]N. Marzari, A. A. Mostofi, J. R. Yates, I. Souza, and D. Vanderbilt, Maximally localized Wannier functions: Theory and applications, Rev. Mod. Phys. 84, 1419 (2012).DOI
- [43]J. Ma and L.-W. Wang, Using wannier functions to improve solid band gap predictions in density functional theory, Scientific Reports 6, 24924 (2016).DOI
- [44]P. Giannozzi, S. Baroni, N. Bonini, M. Calandra, R. Car, et al., QUANTUM ESPRESSO: a modular and open-source software project for quantum simulations of materials, Journal of Physics Condensed Matter 21, 395502 (2009).DOI
- [45]P. Giannozzi, O. Andreussi, T. Brumme, O. Bunau, M. Buongiorno Nardelli, et al., Advanced capabilities for materials modelling with Quantum ESPRESSO, Journal of Physics Condensed Matter 29, 465901 (2017).DOI
- [46]A. Ferretti, I. Dabo, M. Cococcioni, and N. Marzari, Bridging density-functional and many-body perturbation theory: Orbital-density dependence in electronic-structure functionals, Physical Review B 89, 195134 (2014).
- [47]M. Stengel and N. A. Spaldin, Self-interaction correction with Wannier functions, Physical Review B 77, 155106 (2008).DOI
- [48]M. Gatti and F. Sottile, Exciton dispersion from first principles, Phys. Rev. B 88, 155113 (2013).DOI
- [49]S. Gao, Y. Liang, C. D. Spataru, and L. Yang, Dynamical excitonic effects in doped two-dimensional semiconductors, Nano Letters 16, 5568–5573 (2016).DOI
- [51]A. Marini, C. Hogan, M. Grüning, and D. Varsano, yambo: An ab initio tool for excited state calculations, Computer Physics Communications 180, 1392 (2009).DOI
- [52]D. Sangalli, A. Ferretti, H. Miranda, C. Attaccalite, I. Marri, et al., Many-body perturbation theory calculations using the yambo code, J. Phys.: Condens. Matter 31, 325902 (2019).DOI
- [53]M. Bonacci, k2y code, https://github.com/mikibonacci/k2y, 2025.
- [54]R. W. Godby and R. J. Needs, Metal-insulator transition in kohn-sham theory and quasiparticle theory, Phys. Rev. Lett. 62, 1169 (1989).pubmed.ncbi.nlm.nih.gov/10039594
- [55]M. S. Hybertsen and S. G. Louie, Electron correlation in semiconductors and insulators: Band gaps and quasiparticle energies, Phys. Rev. B 34, 5390 (1986).pubmed.ncbi.nlm.nih.gov/9940372
- [56]G. Pizzi, V. Vitale, R. Arita, S. Blügel, F. Freimuth, et al., Wannier90 as a community code: new features and applications, Journal of Physics Condensed Matter 32, 165902 (2020).DOI
- [57]AiidaTeam, Official aiida plugin for quantum espresso., https://github.com/aiidateam/aiida-quantumespresso, 2025.
- [58]aiida-wannier90-workflows, https://github.com/aiidateam/aiida-wannier90-workflows, 2025.
- [59]F. Paleari, A. Molina-Sánchez, M. Nalabothula, R. Reho, M. Bonacci, et al., Yambopy, 2025.
- [60]S. Baroni, S. de Gironcoli, A. Dal Corso, and P. Giannozzi, Phonons and related crystal properties from density-functional perturbation theory, Rev. Mod. Phys. 73, 515 (2001).DOI
- [61]M. Bonacci, aiida-koopmans, https://github.com/mikibonacci/aiida-koopmans, 2025.
- [62]G. de Miranda Nascimento, F. J. dos Santos, M. Bercx, D. Grassano, G. Pizzi, and N. Marzari, Accurate and efficient protocols for high-throughput first-principles materials simulations, arXiv 2504.03962 (2025).DOI
- [63]J. Qiao, G. Pizzi, and N. Marzari, Projectability disentanglement for accurate and automated electronic-structure Hamiltonians, npj Computational Materials 9, 208 (2023).DOI
- [64]J. Qiao, G. Pizzi, and N. Marzari, Automated mixing of maximally localized Wannier functions into target manifolds, npj Computational Materials 9, 206 (2023).DOI
- [65]M. J. van Setten, M. Giantomassi, E. Bousquet, M. J. Verstraete, D. R. Hamann, et al., The PseudoDojo: Training and grading a 85 element optimized norm-conserving pseudopotential table, Computer Physics Communications 226, 39 (2018), Citation for the pseudo dojo standard v0.4.1 pseudopotential library.DOI
- [66]W. Chen and A. Pasquarello, Accurate band gaps of extended systems via efficient vertex corrections in gw, Phys. Rev. B 92, 041115(R) (2015).DOI
- [67]A. Miglio, V. Brousseau-Couture, E. Godbout, G. Antonius, Y.-H. Chan, et al., Predominance of non-adiabatic effects in zero-point renormalization of the electronic band gap, npj Computational Materials 6, 167 (2020).DOI
- [68]M. R. Filip, J. B. Haber, and J. B. Neaton, Phonon screening of excitons in semiconductors: Halide perovskites and beyond, Phys. Rev. Lett. 127, 067401 (2021).DOI
- [69]G. Ohad, S. E. Gant, D. Wing, J. B. Haber, M. Camarasa-Gómez, et al., Optical absorption spectra of metal oxides from time-dependent density functional theory and many-body perturbation theory based on optimally-tuned hybrid functionals, Phys. Rev. Mater. 7, 123803 (2023).DOI
- [70]M. Schebek, P. Pavone, C. Draxl, and F. Caruso, Phonon-mediated renormalization of exciton energies and absorption spectra in polar semiconductors, Journal of Physics: Materials 8, 025006 (2025).arxiv.org/abs/2409.15099
- [71]D. E. Aspnes and A. A. Studna, Dielectric functions and optical parameters of Si, Ge, GaP, GaAs, GaSb, InP, InAs, and InSb from 1.5 to 6.0 eV, Physical Review B 27, 985 (1983).DOI
- [72]M. Shishkin, M. Marsman, and G. Kresse, Accurate quasiparticle spectra from self-consistent gw calculations with vertex corrections, Phys. Rev. Lett. 99, 246403 (2007).pubmed.ncbi.nlm.nih.gov/18233465
- [73]D. A. Leon, C. Elgvin, P. D. Nguyen, O. Prytz, F. S. Hage, and K. Berland, Unraveling many-body effects in zno: Combined study using momentum-resolved electron energy-loss spectroscopy and first-principles calculations, Phys. Rev. B 109, 15153 (2024).arxiv.org/abs/2403.08385
- [74]B. Cunningham, M. Grüning, P. Azarhoosh, D. Pashov, and M. Van Schilfgaarde, Effect of ladder diagrams on optical absorption spectra in a quasiparticle self-consistent GW framework, Physical Review Materials 2, 034603 (2018).DOI
- [75]G. Cassabois, P. Valvin, and B. Gil, Hexagonal boron nitride is an indirect bandgap semiconductor, Nature Photonics 10, 262–266 (2016).DOI
- [76]C. Tarrio and S. E. Schnatterly, Interband transitions, plasmons, and dispersion in hexagonal boron nitride, Phys. Rev. B 40, 7852 (1989).pubmed.ncbi.nlm.nih.gov/9991213
- [77]T. Shen, X.-W. Zhang, H. Shang, M.-Y. Zhang, X. Wang, et al., Influence of high-energy local orbitals and electron-phonon interactions on the band gaps and optical absorption spectra of hexagonal boron nitride, Phys. Rev. B 102, 045117 (2020).DOI
- [78]A. Tal, P. Liu, G. Kresse, and A. Pasquarello, Accurate optical spectra through time-dependent density functional theory based on screening-dependent hybrid functionals, Phys. Rev. Res. 2, 032019(R) (2020).DOI
- [79]A. Thatribud, Electronic and optical properties of tio2 by first-principle calculation (dft-gw and bse), Materials Research Express 6, 095021 (2019).
- [80]H. Huang, G. Clark, E. Navarro-Moratalla, D. R. Klein, R. Cheng, et al., Layer-dependent ferromagnetism in a van der waals crystal down to the monolayer limit, Nature 546, 270–273 (2017).DOI
- [81]D. R. Klein, D. MacNeill, J. L. Lado, D. Soriano, E. Navarro-Moratalla, et al., Probing magnetism in 2d van der waals crystalline insulators via electron tunneling, Science 360, 1218–1222 (2018).DOI
- [82]S. Jiang, L. Li, Z. Wang, K. F. Mak, and J. Shan, Controlling magnetism in 2D CrI3 by electrostatic doping, Nature Nanotechnology 13, 549 (2018).DOI
- [83]S. Jiang, J. Shan, and K. F. Mak, Electric-field switching of two-dimensional van der Waals magnets, Nature Materials 17, 406 (2018).DOI
- [84]Z. Wang, I. Gutiérrez-Lezama, N. Ubrig, M. Kroner, M. Gibertini, et al., Very large tunneling magnetoresistance in layered magnetic semiconductor CrI3, Nature Communications 9, 2516 (2018).DOI
- [85]H. H. Kim, B. Yang, T. Patel, F. Sfigakis, C. Li, et al., One Million Percent Tunnel Magnetoresistance in a Magnetic van der Waals Heterostructure, Nano Letters 18, 4885 (2018).DOI
- [86]Z. Wang, T. Zhang, M. Ding, B. Dong, Y. Li, et al., Electric-field control of magnetism in a few-layered van der Waals ferromagnetic semiconductor, Nature Nanotechnology 13, 554 (2018).DOI
- [87]T. Song, X. X. Cai, M. W.-Y. Tu, X. Zhang, B. Huang, et al., Giant tunneling magnetoresistance in spin-filter van der Waals heterostructures, Science 360, 1214 (2018).DOI
- [88]S. Jiang, L. Li, Z. Wang, J. Shan, and K. F. Mak, Spin tunnel field-effect transistors based on two-dimensional van der Waals heterostructures, Nature Electronics 2, 159 (2019).DOI
- [89]T. Song, M. W.-Y. Tu, C. Carnahan, X. Cai, T. Taniguchi, et al., Voltage Control of a van der Waals Spin-Filter Magnetic Tunnel Junction, Nano Letters 19, 915 (2019).DOI
- [90]D. Soriano, C. Cardoso, and J. Fernández-Rossier, Interplay between interlayer exchange and stacking in cri3 bilayers, Solid State Communications (2018).DOI
- [91]M. Gibertini, Magnetism and stability of all primitive stacking patterns in bilayer chromium trihalides, Journal of Physics D: Applied Physics 54, 064002 (2020).DOI
- [92]S. Acharya, D. Pashov, A. N. Rudenko, M. V. Schilfgaarde, and M. I. Katsnelson, Real- and momentum-space description of the excitons in bulk and monolayer chromium tri-halides, npj 2D Materials and Applications 6, 33 (2022).DOI
- [93]T. Olsen, Unified treatment of magnons and excitons in monolayer cri3 from many-body perturbation theory, Phys. Rev. Lett. 127, 166402 (2021).DOI
- [94]S. Acharya, D. Pashov, B. Cunningham, A. N. Rudenko, M. Rösner, et al., Electronic structure of chromium tri-halides beyond density functional theory, Physical Review B 104, 155109 (2021).
- [95]A. Esquembre-Kučukalić, K. B. Le, A. García-Cristóbal, M. Bernardi, D. Sangalli, and A. Molina-Sánchez, Magnons in chromium trihalides calculated with the ab initio bethe-salpeter equation, Phys. Rev. B 112, 184412 (2025).
- [96]K. L. Seyler, D. Zhong, D. R. Klein, S. Gao, X. Zhang, et al., Ligand-field helical luminescence in a 2d ferromagnetic insulator, Nature Physics 14, 277–281 (2018).DOI
- [97]M. Bonacci, E. Linscott, J. Geiger, N. Marzari, and N. Colonna, Koopmans spectral functionals for excitons in materials, 2026.arxiv.org/abs/2611.22524