betaFeedback

WorksPhysics and Astronomy

Benchmark of GW\boldsymbol{GW} Methods for Core-Level Binding Energies

Abstract

The GWGW approximation has recently gained increasing attention as a viable method for the computation of deep core-level binding energies as measured by X-ray photoelectron spectroscopy (XPS). We present a comprehensive benchmark study of different GWGW methodologies (starting-point optimized, partial and full eigenvalue-self-consistent, Hedin shift and renormalized singles) for molecular inner-shell excitations. We demonstrate that all methods yield a unique solution and apply them to the CORE65 benchmark set and ethyl trifluoroacetate. Three GWGW schemes clearly outperform the other methods for absolute core-level energies with a mean absolute error of 0.3 eV with respect to experiment. These are partial eigenvalue self-consistency, in which the eigenvalues are only updated in the Green's function, single-shot GWGW calculations based on an optimized hybrid functional starting point and a Hedin shift in the Green's function. While all methods reproduce the experimental relative binding energies well, the eigenvalue self-consistent schemes and the Hedin shift yield with mean errors <0.2<0.2 eV the best results.

INTRODUCTION

X-ray photoelectron spectroscopy (XPS) is a standard characterization tool for materials, [1] liquids[2, 3] and molecules. [4] In XPS, the core-level binding energies (CLBEs) are measured, [5] which are element-specific, but also sensitive to the chemical environment. However, establishing the link between between the measured spectrum and the atomic structure is challenging, in particular for complex materials with heavily convoluted XPS signals. [6, 7] Guidance from theory is thus often necessary to interpret XPS spectra.

Theoretical XPS methods can be distinguished into Delta and response approaches. In Delta-based methods, the CLBE is computed as the total energy difference between the neutral and core-excited system. These calculations can be performed at different levels of theory, for example, with high-level wave-function methods, such as Delta coupled cluster (Δ\DeltaCC)[8, 9] or with Kohn-Sham density functional theory[10, 11, 12] (KS-DFT). The most popular DFT-based approach is the Delta self-consistent field (Δ\DeltaSCF) method, [13] which has been thoroughly benchmarked. [14, 15, 16, 17, 18, 19] While high accuracy can be achieved with these approaches, the explicit optimization of a core-ionized wave function leads to conceptual problems, e.g., regarding periodicity, constraining spin-orbit coupled states or, in the case of DFT, deteriorating accuracy for larger structures, which was already discussed and demonstrated elsewhere.[20, 21, 22] [23, 24]

An explicit orbital optimization of core-ionized systems and the related conceptual issues are avoided in response theories, where electron propagators are applied to transform the ground into an excited state. Recently, wave-function-based methods, such as linear response and equation-of-motion CC methods25–28 and the algebraic diagrammatic construction method27 were reassessed for absolute CLBEs, yielding partly promising results. Another promising approach in the realm of response methods is the GWGW approximation[29, 30, 31] to many-body perturbation theory, which is derived from Hedin’s equation[32] by omitting the vertex correction. The GWGW approximation is considered the “gold standard” for the computa tion of band structures of materials, [30, 33] but it has also been successfully applied to valence excitations of molecules. [30, 34, 35, 36]

Due to its primary application to solids, GWGW was traditionally implemented in plane wave codes that typically use pseudo potentials for the deeper states. With the increasing availability of the GWGW method in localized basis set codes, [37, 38, 39, 40, 41, 42, 43, 44] core states moved into focus. Core-level binding energy calculations have emerged as a recent trend in GWGW. [21, 43, 44, 45, 46, 47, 48, 49, 50, 51] [23] By extension to the Bethe-Salpeter equation (BSE@GWGW) also KK-edge transition energies measured in X-ray absorption spectroscopy can be calculated. [52] These studies focused primarily on molecules. However, GWGW is one of the most promising methods for core-level predictions of materials because the scaling with respect to system size is generally smaller than for wave-function based response methods and the method is well-established for periodic structures. In addition, GWGW implementations for localized basis sets are advancing rapidly. Recently, periodic implementations [43, 53, 54] and low-scaling GWGW algorithms with O(N3)O(N^3) complexity emerged in localized basis sets formulations. [42, 55, 56, 57, 58]

The application of GWGW to core states is more difficult than for valence states. We showed that more accurate and computationally more expensive techniques for the frequency integration of the self-energy are required compared to valence excitations. [21] Furthermore, we found that the standard single-shot G0W0G_0W_0 approach performed on top of DFT calculations with generalized gradient approximations (GGAs) or standard hybrid functionals fails to yield a distinct solution, which is caused by a loss of spectral weight in the quasiparticle (QP) peak. [48] We demonstrated that eigenvalue self-consistency in GG or using a hybrid functional with 45% exact exchange as starting point for the G0W0G_0W_0 calculation restores the QP main excitation. Including also relativistic corrections, [49] an agreement of 0.3 eV and 0.2 eV with respect to experiment was reported for absolute and relative CLBEs, respectively. [48]

While G0W0G_0W_0 is the computationally least expensive GWGW flavor, it strongly depends on the density functional approximation (DFA). Tuning the exchange in the hybrid functional to, e.g., 45% is conceptually unappealing and introduces undesired small species dependencies, as discussed more in detail in this work. Self-consistency reduces or removes the dependence on the underlying DFT functional, but significantly increases the computational cost. The computationally least expensive self-consistent schemes are the so-called eigenvalue self-consistent approaches, where the eigenvalues are iterated in the Green’s function GG (evGW0GW_0) or alternatively in GG and the screened Coulomb interaction WW (evGWGW).[30] Higher-level self-consistency schemes, such as fully-self-consistent GWGW[59, 60, 61] (scGWGW) and quasiparticle self-consistent GWGW[62] (qsGWGW) remove, unlike evGW0GW_0 or evGWGW, the starting point dependence completely. However, these higher-level self-consistency schemes are much more expensive and not necessarily better because of the inherent underscreening due to the missing vertex correction.[63, 64]

Recently, we proposed the renormalized singles (RS) Green’s function approach, denoted as GRSW0G_{\mathrm{RS}}W_0, to reduce the starting point dependence in GWGW.[65] The RS concept was developed in the context of the random phase approximation (RPA) for accurate correlation energies,[66, 67] and termed renormalized single-excitation (RSE) correction. Following standard perturbation theory, single-excitations contribute to the second-order correlation energy. It was shown that their inclusion significantly improves binding energies.[66, 67] The RS Green’s function approach extends the RSE idea from correlation energies to GWGW QP energies. In the GRSW0G_{\mathrm{RS}}W_0 scheme, the RS Green’s function is used as a new starting point and the screened Coulomb interaction is calculated with the KS Green’s function. For valence excitations, we found that this renormalization process significantly reduces the starting point dependence and provides improved accuracy over G0W0G_0W_0.[65] The mean absolute errors obtained from the GRSW0G_{\mathrm{RS}}W_0 approach with different DFAs are smaller than 0.2 eV for predicting ionization potentials of molecules in the GW100GW100 set.[65] Unlike the self-consistent schemes, the RS Green’s function method hardly increases the computational cost compared to G0W0G_0W_0.

Recently, we employed the concept of RS in a multireference DFT approach for strongly correlated systems.[68] We also used the RS Green’s function in the T-matrix approximation (GRSTRSG_{\mathrm{RS}}T_{\mathrm{RS}}) in a similar manner as in GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}.[69] The T-matrix method scales formally as O(N6)O(N^6) with respect to system size NN, [69, 70] with reduced scaling possible using effective truncation of the active space. [71] In addition to the high computational cost, the performance of GRSTRSG_{\mathrm{RS}}T_{\mathrm{RS}} for core-levels is not particularly impressive. The error with respect to experiment is 1.5 eV for absolute and 0.3 eV for relative CLBEs. [69] In the present work, we focus thus on RS GWGW approaches for core-levels.

In this work, we benchmark GWGW approaches, which we consider computationally affordable and suitable for large-scale applications. This includes G0W0G_{0}W_{0} with tuned starting points, the eigenvalue self-consistent schemes evGW0\mathrm{ev}GW_{0} and evGW\mathrm{ev}GW and two new methods that we introduce in this work. One is based on the so-called Hedin shift [72] and can be understood as approximation of the evGW0\mathrm{ev}GW_{0} method. We refer to this scheme as GΔHW0G_{\Delta\mathrm{H}}W_{0}. The other is a different flavor of the RS Green’s function approach, where the the screened Coulomb interaction is also computed with the RS Green’s function (GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}).

The remainder of this article is organized as follows: We introduce the different GWGW approaches in § 2 and give the computational details for our calculations in § 3. The solution behavior of the different methods is discussed in § 4.1 by comparing self-energy matrix elements and spectral functions. In § 4.2 results are presented for the CORE65 benchmark set and in § 4.3 for the ethyl trifluoroacetate molecule and we finally draw conclusions in § 5.

THEORY

Single-shot G0W0G_{0}W_{0} approach

The most popular GWGW approach is the single-shot G0W0G_{0}W_{0} scheme, where the GWGW QP energies are obtained as corrections to the KS eigenvalues {ϵn0}\{\epsilon_n^0\}:

ϵnQP=ϵn0+Re⁡⟨ψn0|Σ(ϵnQP)−vxc|ψn0⟩.(1)\epsilon_n^{\mathrm{QP}} = \epsilon_n^0 + \operatorname{Re}\left\langle\psi_n^0 \middle| \Sigma\left(\epsilon_n^{\mathrm{QP}}\right) - v^{\mathrm{xc}} \middle| \psi_n^0 \right\rangle. \tag*{(1)}

{ψn0}\{\psi_n^0\} are the KS molecular orbitals (MOs) and vxcv^{\mathrm{xc}} is the KS exchange-correlation potential.

We use ii, jj for occupied orbitals and aa, bb for virtual orbitals, and mm, nn for general orbitals.

We omitted the spin index in all equations for simplicity and use the notation Σn=⟨ψn0∣Σ∣ψn0⟩\Sigma_n=\langle\psi_n^0|\Sigma|\psi_n^0\rangle and vnxc=⟨ψn0∣vxc∣ψn0⟩v_n^{\mathrm{xc}}=\langle\psi_n^0|v^{\mathrm{xc}}|\psi_n^0\rangle in the following. We can directly obtain the CLBE of state nn from the QP energies because they are related by CLBEn=−ϵnQP\mathrm{CLBE}_n=-\epsilon_n^{\mathrm{QP}}. The self-energy Σ\Sigma is given by

Σ(r,r′,ω)=i2π∫dω′G0(r,r′,ω+ω′)W0(r,r′,ω′)eiω′η(2)\Sigma(\mathbf{r},\mathbf{r}',\omega)=\frac{i}{2\pi}\int d\omega'G_0(\mathbf{r},\mathbf{r}',\omega+\omega')W_0(\mathbf{r},\mathbf{r}',\omega')e^{i\omega'\eta} \tag*{(2)}

where the non-interacting KS Green’s function is denoted G0G_0 and the screened Coulomb interaction W0W_0. η\eta is a positive infinitesimal. The KS Green’s function reads

G0(r,r′,ω)=∑mψm0(r)ψm0(r′)ω−ϵm0−iηsgn⁡(ϵF−ϵm0)(3)G_0(\mathbf{r},\mathbf{r}',\omega)=\sum_m\frac{\psi_m^0(\mathbf{r})\psi_m^0(\mathbf{r}')}{\omega-\epsilon_m^0-i\eta\operatorname{sgn}(\epsilon_{\mathrm{F}}-\epsilon_m^0)} \tag*{(3)}

where ϵF\epsilon_{\mathrm{F}} is the Fermi energy. The screened Coulomb interaction is calculated at the level of the RPA as

W0(r,r′,ω)=∫dr′′ϵ−1(r,r′′,ω)v(r′′,r′)(4)W_0(\mathbf{r},\mathbf{r}',\omega)=\int d\mathbf{r}''\epsilon^{-1}(\mathbf{r},\mathbf{r}'',\omega)v(\mathbf{r}'',\mathbf{r}') \tag*{(4)}

where ϵ(r,r′,ω)\epsilon(\mathbf{r},\mathbf{r}',\omega) is the dielectric function and v(r,r′)=1/∣r−r′∣v(\mathbf{r},\mathbf{r}')=1/|\mathbf{r}-\mathbf{r}'| the bare Coulomb interaction.

The calculation of the self-energy matrix elements Σn\Sigma_n is split into a correlation part Σnc\Sigma_n^c and an exchange part Σnx\Sigma_n^x, i.e., Σn=Σnc+Σnx\Sigma_n=\Sigma_n^c+\Sigma_n^x. The HF-like exchange part Σnx\Sigma_n^x is given by

Σnx=−∑iocc⟨ψn0ψi0∣v∣ψi0ψn0⟩.(5)\Sigma_n^x=-\sum_i^{\mathrm{occ}}\langle\psi_n^0\psi_i^0|v|\psi_i^0\psi_n^0\rangle. \tag*{(5)}

The correlation part Σc\Sigma^c is computed from W0c=W0−vW_0^c=W_0-v and is the part that we plot in Section 4.1 to investigate the GWGW solution behavior. The correlation part in its fully analytic form is given by

Σnc(ω)=∑m∑s∣⟨ψn0ψm0∣ρs0⟩∣2ω−ϵm0+(Ωs0−iη)sgn⁡(ϵF−ϵm0)(6)\Sigma_n^c(\omega)=\sum_m\sum_s\frac{\left|\langle\psi_n^0\psi_m^0|\rho_s^0\rangle\right|^2}{\omega-\epsilon_m^0+\left(\Omega_s^0-i\eta\right)\operatorname{sgn}(\epsilon_{\mathrm{F}}-\epsilon_m^0)} \tag*{(6)}

where Ωs0\Omega_s^0 are charge neutral excitations at the RPA level and ρs0\rho_s^0 are the corresponding transition densities. The fully analytic form of Σnc\Sigma_n^c directly shows the pole structure of the self-energy and is illustrative to understand the solution behavior of GWGW. However, the evaluation of Ωs0\Omega_s^0 scales with O(N6)O(N^6). In practice, the correlation self-energy is usually evaluated with a reduced scaling by using techniques such as analytical continuation[43, 50] or the contour deformation.[21, 43, 73]

The QP energies can be obtained by solving Equation (1), which is typically the computationally least expensive approach. In this work, we additionally employ two alternative approaches to obtain further insight into the physics and suitability of the different GWGW approaches. The first is the graphical solution of Equation (1), where we plot the self-energy matrix elements Σnc\Sigma_n^c and determine the QP solution by finding the intersections with the straight line ω−εn0+vnxc−Σnx\omega-\varepsilon_n^0+v_n^{xc}-\Sigma_n^x. The presence of several intersections would indicate that more than one solution exists. The second, computationally even more expensive alternative, is the computation of the spectral function,[48] which is given by

A(ω)=1π∑m∣Im⁡Σm(ω)∣[ω−εm−(Re⁡Σm(ω)−vmxc)]2+[Im⁡Σm(ω)]2(7)A(\omega)=\frac{1}{\pi}\sum_m\frac{\lvert\operatorname{Im}\Sigma_m(\omega)\rvert}{\left[\omega-\varepsilon_m-\left(\operatorname{Re}\Sigma_m(\omega)-v_m^{xc}\right)\right]^2+\left[\operatorname{Im}\Sigma_m(\omega)\right]^2} \tag*{(7)}

In Equation (7), we include also the imaginary part of the complex self-energy, which gives us direct access to the spectral weights and satellite spectrum. At the G0W0G_0W_0 level, we only use the diagonal elements of Σ\Sigma in Equation (7).

Eigenvalue-selfconsistent GW schemes

Including self-consistency in Hedin’s GWGW equations is a widely used strategy to go beyond G0W0G_0W_0. scGWscGW[30, 61] is conceptually the purest approach, but also the most expensive one. To reduce the computational demands, different lower-level self-consistent schemes were developed. The simplest approach is an eigenvalue self-consistent scheme, which comes in two different flavors. The first one is to iterate the eigenvalues only in GG and keep WW fixed at the W0W_0 level. This scheme is referred to as the evGW0GW_0 approach. In evGW0GW_0, we start by updating the KS eigenvalues in the Green’s function with the G0W0G_0W_0 QP energies, re-evaluate the QP equation (see Equation (1)) and iterate until self-consistency in GG is reached. The Green’s function in the eigenvalue self-consistent scheme reads

Gev(r,r′,ω)=∑mψm0(r)ψm0(r′)ω−εmQP−iηsgn⁡(εF−εmQP)(8)G_{\mathrm{ev}}(\mathbf{r},\mathbf{r}',\omega)=\sum_m\frac{\psi_m^0(\mathbf{r})\psi_m^0(\mathbf{r}')}{\omega-\varepsilon_m^{\mathrm{QP}}-i\eta\operatorname{sgn}(\varepsilon_{\mathrm{F}}-\varepsilon_m^{\mathrm{QP}})} \tag*{(8)}

with εmQP=εm0+Δεm\varepsilon_m^{\mathrm{QP}}=\varepsilon_m^0+\Delta\varepsilon_m, where Δεm\Delta\varepsilon_m is the GWGW correction, see Equation (1). The second flavor is evGWGW, where the KS eigenvalues are not only updated in GG, but also in the screened Coulomb interaction WW. The eigenvalue self-consistent calculations are computationally significantly more expensive than a G0W0G_0W_0 calculation, in particular in combination with the accurate self-energy integration techniques that are required for core-levels. [21] The computational demands are large because GG, the screened Coulomb interaction WW (in the case of evGWGW) and the self-energy have to be built repeatedly. In addition, Equation (1) must be solved not only for the states of interest, but for all states.

GW with Hedin shift

The cost of an evGW0GW_0 scheme can be drastically reduced by using a global shift ΔH\Delta_{\mathrm{H}} instead of an individual shift Δεm\Delta\varepsilon_m for each state mm. This scheme was first introduced by Hedin[32] and is referred to as GΔHW0G_{\Delta_{\mathrm{H}}W_0} in the following. The GΔHW0G_{\Delta_{\mathrm{H}}W_0} approach was discussed several times in the literature [29, 72, 74, 75] and the effect of evGW0GW_0 and the GΔHW0G_{\Delta_{\mathrm{H}}W_0} on the self-energy structure has been discussed for valence states in Ref. 30. In the Hedin-shift scheme, the Green’s function transforms into

GΔH(r,r′,ω)=∑mψm0(r)ψm0(r′)ω−(εm0+ΔH)−iηsgn⁡(εF−εm0)(9)G_{\Delta_{\mathrm{H}}}(\mathbf{r},\mathbf{r}',\omega)=\sum_m\frac{\psi_m^0(\mathbf{r})\psi_m^0(\mathbf{r}')}{\omega-(\varepsilon_m^0+\Delta_{\mathrm{H}})-i\eta\operatorname{sgn}(\varepsilon_{\mathrm{F}}-\varepsilon_m^0)} \tag*{(9)}

where G0(ω−ΔH)=GΔH(ω)G_0(\omega-\Delta H)=G_{\Delta H}(\omega). The QP equation with the Hedin-shift scheme then becomes

ϵnQP=ϵn0+Re⁡⟨ψn0∣Σ(ϵnQP−ΔH)−vxc∣ψn0⟩.(10)\epsilon_n^{\mathrm{QP}}=\epsilon_n^0+\operatorname{Re}\langle\psi_n^0|\Sigma(\epsilon_n^{\mathrm{QP}}-\Delta H)-v^{\mathrm{xc}}|\psi_n^0\rangle. \tag*{(10)}

Traditionally, ΔH\Delta H is determined with respect to the Fermi level of G0G_0 for metals or the valence band maximum for gapped solid-state systems. For the molecular case, ΔH\Delta H is evaluated with respect to the highest occupied molecular orbital (HOMO) by introducing the self-consistency condition ϵHOMOQP=ϵHOMO0+ΔH\epsilon_{\mathrm{HOMO}}^{\mathrm{QP}}=\epsilon_{\mathrm{HOMO}}^0+\Delta H, which is inserted in Equation (10) and yields

ΔHHOMO=Re⁡ΣHOMO(ϵHOMO0)−vHOMOxc.(11)\Delta H_{\mathrm{HOMO}}=\operatorname{Re}\Sigma_{\mathrm{HOMO}}(\epsilon_{\mathrm{HOMO}}^0)-v_{\mathrm{HOMO}}^{\mathrm{xc}}. \tag*{(11)}

As demonstrated in Ref. 30, evGW0\mathrm{ev}GW_0 and the GΔHW0G_{\Delta H}W_0 lead to a shift of the pole structure of the self-energy matrix elements Σn\Sigma_n to more negative frequencies. For ΣHOMO(ω)\Sigma_{\mathrm{HOMO}}(\omega), the shift is similar for evGW0\mathrm{ev}GW_0 and GΔHW0G_{\Delta H}W_0 yielding practically the same ϵHOMOQP\epsilon_{\mathrm{HOMO}}^{\mathrm{QP}}.

For core states, we found that the shift ΔH\Delta H computed as in Equation (11) is much smaller than the one from evGW0\mathrm{ev}GW_0. We propose here a new approach, where we calculate the shift ΔH\Delta H for the core state of interest as

ΔHcore=Re⁡Σcore(ϵcore0)−vcorexc.(12)\Delta H_{\mathrm{core}}=\operatorname{Re}\Sigma_{\mathrm{core}}(\epsilon_{\mathrm{core}}^0)-v_{\mathrm{core}}^{\mathrm{xc}}. \tag*{(12)}

For example, to obtain ϵC1sQP\epsilon_{\mathrm{C1s}}^{\mathrm{QP}} for the CO molecule, we solve Equation (10) with ΔHC1s\Delta H_{\mathrm{C1s}}, whereas we solve it with ΔHO1s\Delta H_{\mathrm{O1s}} for ϵO1sQP\epsilon_{\mathrm{O1s}}^{\mathrm{QP}}. In the case of HCOOH, we determine ΔHcore\Delta H_{\mathrm{core}} for each O separately.

In a GΔHW0G_{\Delta H}W_0 calculation, we calculate ΔH\Delta H once, insert it in Equation (10) and iterate the latter as in a regular G0W0G_0W_0 calculation. The shift ΔH\Delta H is kept constant during the iteration of the QP equation. Compared to a G0W0G_0W_0 calculation, the computation of ΔH\Delta H is the only computational overhead that we introduced. The computational cost of a GΔHW0G_{\Delta H}W_0 calculation is thus practically the same as for a G0W0G_0W_0 calculation. The GΔHW0G_{\Delta\mathrm{H}}W_0 method can be viewed as a one-diagonal element correction in the context of the evGW0evGW_0 method.

Renormalized singles excitations in RPA

The RS Green’s function approach is based on the same idea as the RSE corrections in RPA. Since we consider the introduction of the RS concept more illustrative for total than QP energies, we will briefly summarize the key equations derived by Ren et al.[66, 67, 76] in the context of Rayleigh-Schrödinger perturbation theory (RSPT), before proceeding with the RS Green’s function approaches in Section 2.5.

In RSPT, the interacting NN-electron Hamiltonian H^\hat{H} is partitioned into a non-interacting mean-field Hamiltonian H^0\hat{H}_0 and an interacting perturbation H^′\hat{H}'.

H^0=∑j=1Nh^0(j)=∑j=1N[−12∇j2+vext(rj)+vjHxc](13)\hat{H}_0 = \sum_{j=1}^{N}\hat{h}^{0}(j) = \sum_{j=1}^{N}\left[-\frac{1}{2}\nabla_j^2+v_{\mathrm{ext}}(\mathbf{r}_j)+v_j^{\mathrm{Hxc}}\right] \tag*{(13)}
H^′=∑j<kN1∣rj−rk∣−∑j=1NvjHxc(14)\hat{H}' = \sum_{j<k}^{N}\frac{1}{\lvert\mathbf{r}_j-\mathbf{r}_k\rvert}-\sum_{j=1}^{N}v_j^{\mathrm{Hxc}} \tag*{(14)}

where h^0\hat{h}_0 is the single-particle Hamiltonian of the mean field reference. The Hamiltonian h^0\hat{h}_0 includes the kinetic term, the external potential vextv_{\mathrm{ext}} and the mean-field potential vHxcv^{\mathrm{Hxc}}. The latter can be the single-particle potential from HF or KS-DFT and contains the Hartree and exchange-correlation terms. Following standard perturbation theory, single-excitations (SE) contribute to the second-order correlation energy and are given by [66, 67]

EcSE=∑iocc∑avirt∣⟨Ψ0∣H^′∣Ψia⟩∣2ϵi0−ϵa0(15)E_c^{\mathrm{SE}} = \sum_i^{\mathrm{occ}} \sum_a^{\mathrm{virt}} \frac{\left|\langle\Psi_0|\hat{H}'|\Psi_i^a\rangle\right|^2}{\epsilon_i^0-\epsilon_a^0} \tag*{(15)}
=∑iocc∑avirt∣⟨ψi0∣vHF−vHxc∣ψa0⟩∣2ϵi0−ϵa0(16)= \sum_i^{\mathrm{occ}} \sum_a^{\mathrm{virt}} \frac{\left|\langle\psi_i^0|v^{\mathrm{HF}}-v^{\mathrm{Hxc}}|\psi_a^0\rangle\right|^2}{\epsilon_i^0-\epsilon_a^0} \tag*{(16)}

where ∣Ψ0⟩|\Psi_0\rangle is the Slater determinant for the ground state configuration and ∣Ψia⟩|\Psi_i^a\rangle for the singly excited configuration. The orbitals ψi/a0\psi_{i/a}^0 and corresponding orbital energies ϵi/a0\epsilon_{i/a}^0 are the ones of the h^0(j)\hat{h}^0(j) operator. The derivation of Equation (16) from Equation (15) is given in detail in the supporting information (SI) of Ref. 66. As evident from Equation (16), the singles correction vanishes if vHxcv^{\mathrm{Hxc}} is the HF mean-field potential, which is a consequence of Brillouin’s theorem. [77]

The energy EcSEE_c^{\mathrm{SE}} is only the second-order correction to the correlation energy, as shown in Figure 1. The infinite summation of the higher-order diagrams yields the renormalized singles excitation (RSE) correction. The derivation of the RSE correlation energy is given in detail in Ref. 67. To summarize briefly the procedure, the Fock matrix is evaluated with the KS orbitals. Subsequently, the occupied and unoccupied blocks of this matrix are diagonalized separately (subspace diagonalization), which yields a new set of (RS) eigenvalues and orbitals. Replacing ψi/a\psi_{i/a} and ϵi/a\epsilon_{i/a} in Equation (16) with the RS eigenvalues and orbitals yields the RSE correlation energy.

Goldstone diagrams for renormalized singles contributions

Figure 1. Goldstone diagrammtic for the renormalized singles contributions. Dashed lines, which end with a cross denote the matrix elements ⟨ψp∣vHF−vHxc∣ψq⟩\langle\psi_p|v^{\mathrm{HF}}-v^{\mathrm{Hxc}}|\psi_q\rangle

RS Green’s function GWGW approaches

In analogy to the RPA RSE correction, the RS Green’s function GRSG_{\mathrm{RS}} is designed as an effective non-interacting reference system that includes all the single contributions. The RS Green’s function is defined as

GRS−1=G0−1−P(ΣHF[G0]−vHxc)P−Q(ΣHF[G0]−vHxc)Q,(17)G_{\mathrm{RS}}^{-1}=G_0^{-1}-P\left(\Sigma_{\mathrm{HF}}[G_0]-v^{\mathrm{Hxc}}\right)P-Q\left(\Sigma_{\mathrm{HF}}[G_0]-v^{\mathrm{Hxc}}\right)Q, \tag*{(17)}

where P=∑iocc∣ψi0⟩⟨ψi0∣P = \sum_i^{\mathrm{occ}} |\psi_i^0\rangle\langle\psi_i^0| is the projection into the occupied orbital space and Q=I−PQ = I-P is the projection into the virtual orbital space. ΣHF[G0]\Sigma_{\mathrm{HF}}[G_0] refers to a HF-like self-energy constructed with G0G_0, which is usually the KS Green’s function. ΣHF\Sigma_{\mathrm{HF}} is the sum of the Hartree self-energy ΣH\Sigma^{\mathrm{H}} and the exchange self-energy Σx\Sigma^{\mathrm{x}}, i.e., ΣHF=ΣH+Σx\Sigma_{\mathrm{HF}} = \Sigma^{\mathrm{H}} + \Sigma^{\mathrm{x}}, where ΣmnH=∑iocc⟨ψm0ψi0∣v∣ψn0ψi0⟩\Sigma_{mn}^{\mathrm{H}} = \sum_i^{\mathrm{occ}} \langle\psi_m^0\psi_i^0|v|\psi_n^0\psi_i^0\rangle and Σmnx=−∑i⟨ψm0ψi0∣v∣ψi0ψn0⟩\Sigma_{mn}^{\mathrm{x}} = -\sum_i \langle\psi_m^0\psi_i^0|v|\psi_i^0\psi_n^0\rangle. Note that both are built with the mean-field orbitals ψn0\psi_n^0 provided by, e.g., KS-DFT. vHxcv^{\mathrm{Hxc}} is the single-particle Hartree-exchange-correlation potential defined in Equation (13). If the potential vHxcv^{\mathrm{Hxc}} is the one from HF and if G0G_0 is the HF Green’s function, then the second and third term on the right-hand-side of Equation (17) vanish and GRSG_{\mathrm{RS}} corresponds to the HF Green’s function, which is again a consequence of Brillouin’s theorem. GRSG_{\mathrm{RS}} includes the singles contributions, which are one source of the starting point dependence. As we showed previously,[65] the dependence on the DFA is therefore reduced in the GRSW0G_{\mathrm{RS}}W_0 scheme.

The RS Green’s function is given as the solution of the two projected equations in the occupied orbital subspace[65]

P(GRS−1)P=P(G0−1)P+P(ΣHF[G0]−vHxc)P,(18)P(G_{\mathrm{RS}}^{-1})P = P(G_0^{-1})P + P(\Sigma_{\mathrm{HF}}[G_0] - v^{\mathrm{Hxc}})P, \tag*{(18)}

and the virtual orbital subspace

Q(GRS−1)Q=Q(G0−1)Q+Q(ΣHF[G0]−vHxc)Q.(19)Q(G_{\mathrm{RS}}^{-1})Q = Q(G_0^{-1})Q + Q(\Sigma_{\mathrm{HF}}[G_0] - v^{\mathrm{Hxc}})Q. \tag*{(19)}

In practice, GRSG_{\mathrm{RS}} is obtained by a similar subspace diagonalization procedure as used for the RSE total energy corrections. The KS density matrix is used to construct the HF Hamiltonian H^HF\hat{H}_{\mathrm{HF}}, which defines the RS Hamiltonian H^RS=H^HF[G0]\hat{H}_{\mathrm{RS}} = \hat{H}_{\mathrm{HF}}[G_0]. The equations for the occupied

P(H^HF[G0])P∣ψiRS⟩=ϵiRSP∣ψiRS⟩,(20)P(\hat{H}_{\mathrm{HF}}[G_0])P|\psi_i^{\mathrm{RS}}\rangle= \epsilon_i^{\mathrm{RS}}P|\psi_i^{\mathrm{RS}}\rangle, \tag*{(20)}

and virtual subspace

Q(H^HF[G0])Q∣ψaRS⟩=ϵaRSQ∣ψaRS⟩(21)Q(\hat{H}_{\mathrm{HF}}[G_0])Q|\psi_a^{\mathrm{RS}}\rangle=\epsilon_a^{\mathrm{RS}}Q|\psi_a^{\mathrm{RS}}\rangle \tag*{(21)}

are diagonalized separately.[65] The subspace diagonalization yields the RS eigenvalues ϵnRS\epsilon_n^{\mathrm{RS}} and corresponding eigenvectors ψnRS\psi_n^{\mathrm{RS}} and is performed only once. The RS Green’s function is computed with the RS eigenvalues and orbitals,

GRS(r,r′,ω)=∑mψmRS(r)ψmRS(r′)ω−ϵmRS−iηsgn⁡(ϵF−ϵmRS)(22)G_{\mathrm{RS}}(\mathbf{r},\mathbf{r}',\omega)=\sum_m\frac{\psi_m^{\mathrm{RS}}(\mathbf{r})\psi_m^{\mathrm{RS}}(\mathbf{r}')}{\omega-\epsilon_m^{\mathrm{RS}}-i\eta\operatorname{sgn}(\epsilon_{\mathrm{F}}-\epsilon_m^{\mathrm{RS}})} \tag*{(22)}

and is diagonal in the occupied subspace and the virtual subspace.

In the GRSW0G_{\mathrm{RS}}W_0 approach,[65] the RS Green’s function is used as a new starting point and the screened interaction is calculated with the KS Green’s function. The correlation part of the GRSW0G_{\mathrm{RS}}W_0 self-energy is[65]

Σnc,GRSW0(ω)=∑m∑s∣⟨ψn0ψm0∣ρs0⟩∣2ω−ϵmRS+(Ωs0−iη)sgn⁡(ϵmRS−ϵF)(23)\Sigma_n^{c,G_{\mathrm{RS}}W_0}(\omega)=\sum_m\sum_s\frac{\left|\langle\psi_n^0\psi_m^0|\rho_s^0\rangle\right|^2}{\omega-\epsilon_m^{\mathrm{RS}}+(\Omega_s^0-i\eta)\operatorname{sgn}(\epsilon_m^{\mathrm{RS}}-\epsilon_{\mathrm{F}})} \tag*{(23)}

where ρs0\rho_s^0 and Ωs0\Omega_s^0 are the transition densities and the RPA excitation energies calculated with the KS Green’s function. Then the QP equation for GRSW0G_{\mathrm{RS}}W_0 is

ϵnQP=ϵn0+Re⁡⟨ψn0∣ΣGRSW0(ϵnQP)−vxc∣ψn0⟩.(24)\epsilon_n^{\mathrm{QP}}=\epsilon_n^0+\operatorname{Re}\langle\psi_n^0|\Sigma^{G_{\mathrm{RS}}W_0}(\epsilon_n^{\mathrm{QP}})-v^{\mathrm{xc}}|\psi_n^0\rangle. \tag*{(24)}

Note that Equations (23) and (24) are slightly different from the original QP equation for GRSW0G_{\mathrm{RS}}W_0,[65] where we used the RS eigenvectors ψnRS\psi_n^{\mathrm{RS}}. Here we use for simplicity the KS instead of the RS orbitals because we found that the difference induced upon using RS orbitals is marginal.[65] Since we use the KS orbitals, the exchange part of the GRSW0G_{\mathrm{RS}}W_0 self-energy is the same as in G0W0G_0W_0, see Equation (5).

In this work we introduce a new approach that uses the RS Green’s function as a new starting point and also calculates the screened Coulomb interaction with the RS Green’s function, denoted as GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}. This means that WRSW_{\mathrm{RS}} is obtained by inserting the RS Green’s function into the RPA equation. Similar to GRSW0G_{\mathrm{RS}}W_0, the exchange part of the GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} self-energy is also the same as G0W0G_0W_0, but the correlation part of the GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} self-energy becomes

Σnc,GRSWRS(ω)=∑m∑s∣⟨ψn0ψm0|ρsRS⟩∣2ω−ϵmRS−(ΩsRS−iη)sgn⁡(ϵmRS−ϵF)(25)\Sigma_{n}^{c,G_{\mathrm{RS}}W_{\mathrm{RS}}}(\omega)=\sum_m\sum_s\frac{\left|\left\langle\psi_n^0\psi_m^0\middle|\rho_s^{\mathrm{RS}}\right\rangle\right|^2}{\omega-\epsilon_m^{\mathrm{RS}}-\left(\Omega_s^{\mathrm{RS}}-\mathrm{i}\eta\right)\operatorname{sgn}\left(\epsilon_m^{\mathrm{RS}}-\epsilon_{\mathrm{F}}\right)} \tag*{(25)}

where ρsRS\rho_s^{\mathrm{RS}} and ΩsRS\Omega_s^{\mathrm{RS}} are the transition densities and the RPA excitation energies calculated with the RS Green’s function. Then the QP equation for the GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} approach follows as

ϵnQP=ϵn0+Re⁡⟨ψn0|ΣGRSWRS(ϵnQP)−vxc|ψn0⟩.(26)\epsilon_n^{\mathrm{QP}}=\epsilon_n^0+\operatorname{Re}\left\langle\psi_n^0\middle|\Sigma_{G_{\mathrm{RS}}W_{\mathrm{RS}}}\left(\epsilon_n^{\mathrm{QP}}\right)-v_{\mathrm{xc}}\middle|\psi_n^0\right\rangle. \tag*{(26)}

The construction of the RS Green’s function scales only as O(N3)O(N^3) with respect to system size NN. The overall scaling of GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} depends on the frequency integration technique, which is O(N6)O(N^6) for a fully analytic evaluation of the self-energy using Equation (6), but O(N4)O(N^4) (valence states) or O(N5)O(N^5) (core states) with the contour deformation approach. [21, 43, 73] Given that the same frequency integration is used for G0W0G_0W_0 and GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}, the computational cost of a GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} calculation is only marginally larger than for a G0W0G_0W_0 calculation.

COMPUTATIONAL DETAILS

Core-level calculations were performed at the G0W0G_0W_0, evGW0GW_0, evGWGW, GΔHW0G_{\Delta\mathrm{H}}W_0 and GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} level for the CORE65 benchmark set, [48] which contains 65 1s excitations energies of 32 small molecules (30×\timesC1s, 21×\timesO1s, 11×\timesN1s and 3×\timesF1s). Geometries and experimental reference values were taken from Ref. 48. Additionally, we studied the ethyl trifluoroacetate molecule. The structure of the latter was obtained upon request from the authors of Ref. 18 and is available in the SI.

The G0W0G_0W_0, evGW0GW_0, evGWGW and GΔHW0G_{\Delta\mathrm{H}}W_0 calculations were carried out with the FHI-aims program package, [78, 79] which is based on numeric, atom-centered orbitals (NAOs). The G0W0G_0W_0 and evGW0GW_0 data were extracted from our previous work, [48] while the evGWGW and GΔHW0G_{\Delta\mathrm{H}}W_0 [page continues] data were generated for this benchmark study. In the FHI-aims calculations, the contour deformation technique [21, 30, 73] is used to evaluate the self-energy, using a modified Gauss-Legendre grid[38] with 200 grid points for the imaginary frequency integral part. The GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} calculations were performed with the QM4D program.[80] In QM4D, the GWGW self-energy integral is calculated fully analytically, see Equation (6). In FHI-aims and also in QM4D, the QP equation is always solved iteratively.

For evGW0evGW_{0}, evGWevGW and GΔHW0G_{\Delta_\mathrm{H}}W_{0}, we used the Perdew-Burke-Ernzerhof (PBE)[81] functional for the underlying DFT calculation, while the G0W0G_{0}W_{0} calculations employ the PBEh(α\alpha) hybrid functional[82] with 45% exact exchange (α=0.45\alpha= 0.45). The GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} calculations were performed with PBE and three different hybrid functionals, namely PBE0,[83, 84] PBEh(α=0.45\alpha= 0.45) and B3LYP.[85, 86] Note that PBE0 corresponds to PBEh(α=0.25\alpha= 0.25).

All GWGW results are extrapolated to the complete basis set limit using the Dunning basis set family cc-pVnnZ.[87, 88] Following Ref. 48, the extrapolation is performed by a linear regression with respect to the inverse of the total number of basis functions. A four-point extrapolation with n=3−6n = 3 - 6 is performed for G0W0G_{0}W_{0}, evGW0evGW_{0}, evGWevGW and GΔHW0G_{\Delta_\mathrm{H}}W_{0}. For GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}, we use only two points (n=3,4n = 3, 4) due to computational limitations. We verified that this two-point extrapolation deviates only by 0.1 eV from the four-point scheme on average. The cc-pVnnZ family are contracted Gaussian-type orbitals (GTOs), which can be considered as a special case of an NAO and are treated numerically in FHI-aims. Note that the cc-pVnnZ basis sets are treated as spherical GTOs in FHI-aims, whereas in QM4D they are processed as pure Cartesian GTOs. Both codes use the resolution-of-the-identity (RI) approach with the Coulomb metric (RI-V).[89] In FHI-aims, the RI auxiliary basis sets are generated on-the-fly as described in Ref. 38. For the QM4D calculations, the corresponding RI basis sets for cc-pVTZ and cc-pVQZ from Ref. 90 were used.

Relativistic effects were included for all calculations as post-processing step following the approach in Refs. 48 and 49, i.e., we performed a non-relativistic GWGW calculation on top of a non-relativistic KS-DFT calculation and added the corrective term derived in Ref. 49 to the GWGW QP energies. The magnitude of the corrections increase with the atomic number and ranges from 0.12 eV for C1s to 0.71 eV for F1s. The relativistic corrections were derived for a free neutral atom at the PBE level and were obtained by evaluating the difference between the 1s eigenvalues from the radial KS and the 4-component Dirac-KS equation.

RESULTS AND DISCUSSIONS

Solution behavior

In our previous work[48] we showed that standard G0W0G_0W_0 calculations starting from a GGA functional, which are routinely applied to valence states, lead to an erroneous multi-solution behavior for deep core states. It is thus important to confirm that the respective GWGW flavors yield indeed a unique solution. Only looking at the QP energies obtained by iterating Equation (1) is typically not enough to verify the latter. Detailed insight into the solution behavior of GWGW-based methods can be obtained by: i) plotting the real part of the correlation self-energy Σc\Sigma^{c} and ii) plotting the spectral function A(ω)A(\omega) as defined in Equation (7). In Figure 2, we investigate A(ω)A(\omega) and the diagonal matrix elements Σnc(ω)\Sigma^{c}_{n}(\omega) for the 1s oxygen orbital of a single water molecule. Results are shown for G0W0G_0W_0 and GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} with different starting points (PBE, PBE0, B3LYP, PBEh(α=0.45\alpha= 0.45)) as well as partial self-consistent schemes, namely evGW0\mathrm{ev}GW_0, evGW\mathrm{ev}GW and GΔHW0G_{\Delta_{\mathrm{H}}}W_0, using PBE for the underlying DFT calculation.

O1s excitation spectra and correlation self-energy for a single water molecule

Figure 2. O1s excitation for a single water molecule from G0W0G_{0}W_{0}, evGW0GW_{0}, GΔHW0G_{\Delta H}W_{0}, evGWGW and GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} computed at the cc-pVQZ level. (a) Spectral function A(ω)A(\omega) (Equation (7)) from G0W0G_{0}W_{0} using starting points with no exact exchange or a low amount. (b,c) Real part of the correlation self-energy Σc(ω)\Sigma^{c}(\omega) in evGW0GW_{0}@PBE, GΔHW0G_{\Delta H}W_{0} and evGWGW@PBE. Diagonal matrix elements Re⁡Σnc(ω)=⟨ψn∣Re⁡Σc(ω)∣ψn⟩\operatorname{Re}\Sigma^{c}_{n}(\omega)=\langle\psi_{n}|\operatorname{Re}\Sigma^{c}(\omega)|\psi_{n}\rangle for the oxygen 1s orbital. The intersection with the red line is the graphical solution of Equation (1). The vertical gray-dashed line indicates the QP solution of this graphical solution and Δ\Delta indicates the shift with respect to G0W0G_{0}W_{0}@PBE (gray) (d) Spectral functions in evGW0GW_{0}@PBE, GΔHW0G_{\Delta H}W_{0}@PBE, evGWGW@PBE and G0W0G_{0}W_{0}@PBEh(α=0.45\alpha=0.45). (e) Self-energy matrix elements Re⁡Σnc(ω)\operatorname{Re}\Sigma^{c}_{n}(\omega) in GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}@PBE (green) compared to G0W0G_{0}W_{0}@PBE (gray). (f) Spectral functions A(ω)A(\omega) in GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} with different starting points.

We start our discussion with the G0W0G_0W_0 spectral functions and self-energy elements displayed in Figure 2(a,Figure 2b,Figure 2d), where we reproduced for convenience the G0W0G_0W_0@PBE, G0W0G_0W_0@PBE0 and G0W0G_0W_0@PBEh(α=0.45\alpha= 0.45) results, which were also presented in Ref. 48. Figure 2(b) shows the self-energy from a G0W0G_0W_0@PBE (gray line), which exhibits many poles. The poles are broadened by the η\eta-term in Equation (3) and thus appear as spikes in the self-energy. For G0W0G_0W_0@PBE we find that the poles are located in the frequency region where the QP solution is expected (around 540 eV). As already outlined in § 2.1, we can obtain the graphical solution of Equation (1) by finding the intersections with the straight line ω−εn0+vnxc−Σnx\omega- \varepsilon_n^0 + v_n^{\mathrm{xc}} - \Sigma_n^{\mathrm{x}}.

For G0W0G_{0}W_{0}, we observe many intersections, which are all valid solutions of the QP equation. The corresponding spectral function in Figure 2(a) shows many peaks with equal spectral weight, but no clear main peak that could be assigned to the QP excitation in contrast to the experiment, where a sharp peak at 539.7 eV[4] is observed. A main peak starts to emerge for hybrid functional starting points, such as PBE0 (α=0.25\alpha=0.25) and B3LYP (α=0.20\alpha=0.20). However, G0W0G_{0}W_{0}@PBE0 and G0W0G_{0}W_{0}@B3LYP still yield an unphysical second peak, which carries a large fraction of the spectral weight.

As already discussed in Ref. 48, the reason for this unphysical behavior is the overlap of the satellite spectrum with the QP peak. Satellites are due to multielectron excitations that accompany the photoemission process, e.g., shake-up satellites, which are produced when the core photoelectron scatters a valence shell electron to a higher unoccupied energy level. [91, 92] These peaks appear as series of smaller peaks at higher energies than the QP energy. For molecules, the spectral weight of these peaks is orders of magnitudes smaller than for the main excitation. [92] Satellites occur in frequency regions where the real part of the self-energy has poles. As demonstrated in, e.g., Ref. 30, the imaginary part of the self-energy exhibits complementary peaks at these frequencies (Kramers-Kronig relation). According to Equation (7), large imaginary parts lead to peaks with small spectral weight, i.e., peaks with satellite character.

The occurrence of pole features around the QP excitations for deep core states can best be understood by analyzing the denominator of the fully analytic expression of the self-energy given in Equation (6). Σnc(ω)\Sigma_n^c(\omega) has poles on the real axis for η→0\eta\to0 at ϵi0−Ωs\epsilon_i^0-\Omega_s (occupied states ii) and ϵa0+Ωs\epsilon_a^0+\Omega_s (virtual states aa). For occupied states, the eigenvalues ϵi0\epsilon_i^0 are too large (too positive) and the charge neutral excitations Ωs\Omega_s are underestimated at the PBE level. As a result, the poles ϵi0−Ωs\epsilon_i^0-\Omega_s are at too positive frequencies and the satellite thus too close to the QP peak. For virtual states, the same reasoning holds for the poles ϵa0+Ωs\epsilon_a^0+\Omega_s, but with reversed sign, i.e., the poles are at too small frequencies. The separation between satellites and QP peak is also too small for valence excitations. However, the problem gets progressively worse further away from the Fermi level since the absolute differences between PBE eigenvalues ϵi0\epsilon_i^0 and the QP excitation increases. We demonstrated this for semi-core states, [56] for which a distinct QP peak is still obtained. However, for deep core states the separation becomes finally so small that the satellites merge with the QP peak.

The proper separation between QP excitation and satellites can be restored by using an evGW0GW_0 scheme. The evGW0GW_0@PBE self-energy is shown in Figure 2(b) (reproduced from Ref. 48). Iterating the eigenvalues in GG shifts the on-set of the pole structure too more negative frequencies. The overall pole structure is very similar to G0W0@PBEG_{0}W_{0}@\mathrm{PBE}, but shifted by a constant value of Δ=−28.7\Delta= -28.7 eV. The GΔHW0@PBEG_{\Delta H}W_{0}@\mathrm{PBE} self-energy displayed in Figure 2(b) is almost identical to evGW0@PBE\mathrm{ev}GW_{0}@\mathrm{PBE}. The shift of the pole structure compared to G0W0@PBEG_{0}W_{0}@\mathrm{PBE} is with Δ=−28.8\Delta= -28.8 eV only slightly larger than for evGW0\mathrm{ev}GW_{0}. The rigid Δ\Delta-shift of the poles features can be understood as follows: In evGW0\mathrm{ev}GW_{0} and GΔHW0G_{\Delta H}W_{0}, the KS eigenvalue ϵm0\epsilon^{0}_{m} are replaced with ϵm0+Δϵm\epsilon^{0}_{m}+\Delta\epsilon_{m} and ϵm0+ΔH1s\epsilon^{0}_{m}+\Delta\mathrm{H}_{1s}, respectively, where Δϵm\Delta\epsilon_{m} is the self-consistent GWGW correction for state mm and ΔH1s\Delta\mathrm{H}_{1s} its non-self-consistent approximation for the O1s state (see Equation (12)). The poles are consequently located at ϵ1s+Δϵ1s−Ωs\epsilon_{1s}+\Delta\epsilon_{1s}-\Omega_{s} and ϵ1s+ΔH1s−Ωs\epsilon_{1s}+\Delta\mathrm{H}_{1s}-\Omega_{s}. Since both corrections, Δϵ1s\Delta\epsilon_{1s} and ΔH1s\Delta\mathrm{H}_{1s} are negative for PBE starting points, the poles, i.e., satellites move to more negative frequencies and are properly separated from the main excitation. The spectral function now exhibits a distinct QP peak as evidenced by Figure 2(d).

As discussed in detail previously, [48] the effect of eigenvalue self-consistency can be mimicked in a G0W0G_{0}W_{0} calculation by using a hybrid functional with a high amount of exact exchange α\alpha. We showed that increasing α\alpha progressively shifts the pole features to more negative frequencies. For α=0.45\alpha= 0.45, the evGW0@PBE\mathrm{ev}GW_{0}@\mathrm{PBE} self-energy is approximately reproduced and the spectral function shows a distinct peak as displayed in Figure 2(d). We note here again that values of α<0.3\alpha< 0.3 do not yield a clear main peak[48] and thus no unique solution, which is also demonstrated in Figure 2(a).

The evGW\mathrm{ev}GW approach and the GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} schemes lead to a significantly stronger shift of the pole features than evGW0@PBE\mathrm{ev}GW_{0}@\mathrm{PBE} or GΔHW0@PBEG_{\Delta H}W_{0}@\mathrm{PBE}, as shown in Figure 2(c) and (e). The spectral functions displayed in Figure 2(d) and (f) confirm that evGW\mathrm{ev}GW and the GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} yield a distinct peak in the spectrum. The RS eigenvalues of the occupied orbitals are more negative and the ones of the virtual orbitals are more positive than the KS eigenvalues. In addition, RPA evaluated with RS fundamental gaps provides larger excitation energies Ωs\Omega_{s}. In GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} the poles at ϵa+Ωs\epsilon_{a}+\Omega_{s} are shifted in the positive direction and the poles at ϵi−Ωs\epsilon_{i}-\Omega_{s} are shifted in the negative direction. Therefore, satellites from these poles are separated from the main peak. For GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}, a unique solution is obtained for all starting points. As we show in

Figure S1 (see SI), the GRSW0G_{\mathrm{RS}}W_0 approach suffers from a multi-solution behavior in the deep core region and cannot be applied for core-level calculations.

CORE65 benchmark

In the following, we discuss the CORE65 results for the GWGW schemes for which a physical solution behavior was confirmed in § 4.1, namely evGW0\mathrm{ev}GW_0@PBE, evGW\mathrm{ev}GW@PBE, G0W0G_0W_0@PBEh (α=0.45\alpha= 0.45), GΔHW0G_{\Delta H}W_0@PBE and GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} with four different starting points (PBE, PBE0, B3LYP, PBEh (α=0.45\alpha= 0.45)). The distribution of the errors with respect to experiment are shown in Figure 3 and Figure 4 for the absolute CLBEs and the relative CLBEs, respectively. The corresponding mean absolute errors (MAE) and the mean errors (ME) are given in Table 1. The error of excitation ii is defined as Errori=CLBEitheory−CLBEiexperiment\mathrm{Error}_i=\mathrm{CLBE}_i^{\mathrm{theory}}-\mathrm{CLBE}_i^{\mathrm{experiment}}.

Distribution of errors with respect to the experiment for absolute 1s binding energies of the CORE65 benchmark set

Figure 3. Distribution of errors with respect to the experiment for absolute 1s binding energies of the CORE65 benchmark set, where the error is defined as Errori=CLBEitheory−CLBEiexp\mathrm{Error}_i=\mathrm{CLBE}_i^{\mathrm{theory}}-\mathrm{CLBE}_i^{\mathrm{exp}}. The histogram is stacked. Eight GWGW approaches are compared: (a) evGW0\mathrm{ev}GW_0@PBE, (b) G0W0G_0W_0@PBEh (α=0.45\alpha= 0.45), (c) evGW\mathrm{ev}GW@PBE, (d) GΔHW0G_{\Delta H}W_0@PBE, (e) GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}@PBE, (f) GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}@PBE0, (g) GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}@B3LYP, (h) GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}@PBEh (α=0.45\alpha= 0.45).

Figure 4. Distribution of errors with respect to the experiment for relative 1s1s binding energies of the CORE65 benchmark set. The histogram is stacked. CH4_4, NH3_3, H2_2O and CH3_3F have been used as reference molecules for C1s, N1s, O1s and F1s respectively. Eight GW approaches are compared: (a) evGW0@PBE\mathrm{ev}GW_{0}@\mathrm{PBE}, (b) G0W0@PBEh(α=0.45)G_{0}W_{0}@\mathrm{PBEh}(\alpha= 0.45), (c) evGW@PBE\mathrm{ev}GW@\mathrm{PBE}, (d) GΔHW0@PBEG_{\Delta H}W_{0}@\mathrm{PBE}, (e) GRSWRS@PBEG_{\mathrm{RS}}W_{\mathrm{RS}}@\mathrm{PBE}, (f) GRSWRS@PBE0G_{\mathrm{RS}}W_{\mathrm{RS}}@\mathrm{PBE0}, (g) GRSWRS@B3LYPG_{\mathrm{RS}}W_{\mathrm{RS}}@\mathrm{B3LYP}, (h) GRSWRS@PBEh(α=0.45)G_{\mathrm{RS}}W_{\mathrm{RS}}@\mathrm{PBEh}(\alpha= 0.45).

core-levelevGW0@evGW_0@ PBEG0W0@G_0W_0@ PBEhevGW@evGW@ PBEGΔHW0@G_{\Delta H}W_0@ PBEGRSWRS@G_{\mathrm{RS}}W_{\mathrm{RS}}@ PBE
MAEMEMAEMEMAEMEMAEMEMAEMEMAEMEMAEMEMAEME
Absolute CLBEs
all0.30-0.290.33-0.081.531.530.25-0.204.884.885.325.324.944.945.725.72
C1s0.27-0.270.240.191.371.370.20-0.133.973.974.504.504.244.245.035.03
N1s0.30-0.300.16-0.011.581.580.14-0.135.135.135.595.595.125.126.006.00
O1s0.32-0.280.48-0.401.701.700.35-0.315.915.916.246.245.715.717.487.48
F1s0.44-0.440.83-0.831.651.650.54-0.546.566.566.756.756.326.326.886.88
Relative CLBEs
all0.180.020.260.230.18-0.030.19-0.020.400.390.430.430.370.360.480.46
C1s0.18-0.050.290.250.19-0.010.200.070.360.360.330.330.300.300.410.41
N1s0.140.140.230.210.14-0.130.16-0.150.290.290.390.390.300.300.400.40
O1s0.220.080.250.240.18-0.030.20-0.060.560.560.660.660.560.560.680.65
F1s0.05-0.050.110.100.110.020.16-0.160.16-0.100.160.000.200.020.220.00

Table 1. Mean absolute error (MAE) and mean error (ME) in eV with respect to experiment for absolute and relative CLBEs of the CORE65 benchmark set. The error for excitation ii is defined as Error⁡i=CLBE⁡itheory−CLBE⁡iexp\operatorname{Error}_i = \operatorname{CLBE}^{\mathrm{theory}}_i - \operatorname{CLBE}^{\mathrm{exp}}_i. The relative CLBEs are the shifts with respect to a reference molecule, ΔCLBE⁡=CLBE⁡−CLBE⁡ref_mol\Delta\operatorname{CLBE} = \operatorname{CLBE} - \operatorname{CLBE}_{\mathrm{ref\_mol}}. CH4_4, NH3_3, H2_2O and CH3_3F have been used as reference molecules for C1s, N1s, O1s and F1s respectively.

Starting the discussion with the absolute CLBEs, we find that evGW0@PBE\mathrm{ev}GW_{0}@\mathrm{PBE}, G0W0@PBEh(α=0.45)G_{0}W_{0}@\mathrm{PBEh}(\alpha= 0.45) and GΔHW0@PBEG_{\Delta H}W_{0}@\mathrm{PBE} yield the best results with error distributions close to zero and MAEs of ≈0.3\approx0.3 eV. The smallest overall MAE of 0.25 eV is obtained with GΔHW0@PBEG_{\Delta H}W_{0}@\mathrm{PBE}. Figure 3(a) and (d) show that the errors from evGW0@PBE\mathrm{ev}GW_{0}@\mathrm{PBE} and GΔHW0@PBEG_{\Delta H}W_{0}@\mathrm{PBE} are tightly distributed, but mostly negative, i.e., the computed CLBEs are slightly underestimated. Generally, we find that the GΔHW0@PBEG_{\Delta H}W_{0}@\mathrm{PBE} scheme reproduces the evGW0@PBE\mathrm{ev}GW_{0}@\mathrm{PBE} results almost perfectly. The slight underestimation of the absolute CLBEs by evGW0@PBE\mathrm{ev}GW_{0}@\mathrm{PBE} and GΔHW0@PBEG_{\Delta H}W_{0}@\mathrm{PBE} might be due to insufficiencies in the cc-pVnnZ basis sets, which are not captured by the extrapolation procedure. A very recent study [51] with G0W0@PBEhG_{0}W_{0}@\mathrm{PBEh} showed that increasing the number of core functions by, e.g., uncontracting the cc-pVnnZ basis sets, yields larger absolute CLBEs. The reported increase is in the range of 0.25 to maximal 0.5 eV, indicating that the CLBEs from evGW0\mathrm{ev}GW_{0} and GΔHW0@PBEG_{\Delta H}W_{0}@\mathrm{PBE} might be even closer to experiment with core-rich basis sets.

The error distribution of G0W0@PBEh(α=0.45)G_{0}W_{0}@\mathrm{PBEh}(\alpha= 0.45), which is displayed in Figure 3(b), is centered around zero, yielding also the smallest overall ME, see Table 1. Compared to evGW0@PBE\mathrm{ev}GW_{0}@\mathrm{PBE} and GΔHW0@PBEG_{\Delta H}W_{0}@\mathrm{PBE}, the spread of the G0W0@PBEhG_{0}W_{0}@\mathrm{PBEh} errors is larger and a clustering by species can be observed. The C1s binding energies (BEs) are overestimated, while the N1s, O1s and F1s are increasingly underestimated. This is due to the species dependence of the α\alpha parameter, which we discussed in Ref. 48. As we showed in Ref. 48, including relativistic effects reduces the species dependence of α\alpha, but does not remove it completely. The optimal α\alpha value, αopt\alpha_{\mathrm{opt}}, increases from 0.44 (C1s) to 0.49 (F1s), after including relativistic corrections. For α<αopt\alpha< \alpha_{\mathrm{opt}}, the CLBEs are too small and for α>αopt\alpha> \alpha_{\mathrm{opt}} too large. As a result, the C1s BEs are overestimated for α=0.45\alpha= 0.45, and O1s and F1s BEs are underestimated.

evGW@PBE\mathrm{ev}GW@\mathrm{PBE} systematically overestimates the absolute CLBEs by 1-2 eV since iterating also the eigenvalues in WW effectively leads to an underscreening. At the PBE level, the fundamental gap is underestimated and inserting the PBE eigenvalues in WW consequently yields an overscreened potential. However, the overscreening in WW compensates the underscreening introduced by the missing vertex corrections. Comparing evGW0@PBE\mathrm{ev}GW_{0}@\mathrm{PBE} and evGW@PBE\mathrm{ev}GW@\mathrm{PBE}, our observation for deep core-levels agrees with previous work, which found that evGW0\mathrm{ev}GW_{0} improves upon G0W0G_{0}W_{0}, [93, 94] while evGW\mathrm{ev}GW yields too large band gaps [93] and overly stretched spectra. [95] Furthermore, the performance of evGW@PBE\mathrm{ev}GW@\mathrm{PBE} for deep core-levels seems to be comparable to higher-level self-consistency schemes such as qsGW\mathrm{qs}GW. [62] An exploratory study by Setten et al. [46] reported that qsGW\mathrm{qs}GW overestimates the absolute 1s BEs of small molecules by 2 eV. This indicates that qsGW\mathrm{qs}GW suffers from similar underscreening effects and that the orbitals inserted in the GWGW scheme have a minor effect on core-level QP energies.

The error distributions for the absolute CLBEs from GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} are shown in Figure 3(e-Figure 3h) for four different starting points. GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} overestimates the absolute CLBEs by 3-8 eV with an MAE between 5-6 eV. The reason for the large overestimation is that the RS fundamental gap is too large, which then leads, similarly as in G0W0@HFG_0W_0@\mathrm{HF}[95], to an underscreening in WW.

One way to reduce the underscreening is to include corrections for the electron correlation in the RS Hamiltonian, which is dominated by exchange interactions. An alternative strategy is to compensate the underscreening by including vertex corrections, e.g., to use the T-matrix formalism[29, 69, 70], where the two-point screened interaction WW is replaced with a four-point effective interaction TT.

However, methods such as the T-matrix are computationally much more expensive than GWGW due to their higher complexity. We recently applied the GRSTRSG_{\mathrm{RS}}T_{\mathrm{RS}} scheme to the CORE65 benchmark set.[69] Comparing GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} and GRSTRSG_{\mathrm{RS}}T_{\mathrm{RS}}, the overestimation is indeed reduced by GRSTRSG_{\mathrm{RS}}T_{\mathrm{RS}}, which yields an ME of ≈1.5\approx1.5 eV. However, the errors for the absolute CLBEs are still an order of magnitude larger than for computationally cheaper schemes such as evGW0evGW_0 or GΔHW0G_{\Delta H}W_0, which rely on a very fortunate error cancellation effect that leads to a balanced screening.

Furthermore, we find that the overestimation with GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} increases with the atomic number, i.e., from the C1s to the F1s excitations. This species dependence is inherited from the KS-DFT calculation. As shown in Table S8 in the SI, the deviations of the CLBEs obtained from KS-DFT eigenvalues (CLBEn=−εn0\mathrm{CLBE}_{n}=-\varepsilon_{n}^{0}) generally increase from C1s to F1s for all DFAs. We expect that, e.g., adding correlation contributions to the RS Hamiltonian would also reduces this undesired species dependence.

The motivation of the RS approach is the reduction of the starting point dependence. As evident from Figure 3(e-Figure 3h), the large overestimation is observed for all DFAs. Based on the MAEs and MEs in Table 1, it can be shown that the starting point dependence is on average <1<1 eV for α=0.0−0.45\alpha=0.0-0.45. As shown in Figure S2 (SI) for a single water molecule, evGW0evGW_{0} seems to reduce the starting point dependence less. For the same α\alpha range, the CLBE changes by ≈2\approx2 eV. The direct comparison with G0W0G_{0}W_{0} is difficult because a unique solution is only obtained for α>0.3\alpha>0.3. However, we can study the change of the CLBEs for α=0.3−1.0\alpha=0.3-1.0, see Figure S2 (SI), which shows that the starting point dependence is 10 eV with G0W0G_{0}W_{0} compared to 2.8 eV with evGW0evGW_{0}.

Moving now to the relative CLBEs, we observe that evGW0evGW_{0}@PBE, G0W0G_{0}W_{0}@PBEh, evGWevGW@PBE and GΔHW0G_{\Delta\mathrm{H}}W_{0}@PBE yield MAEs of ≈0.2−0.3\approx0.2-0.3 eV, and GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} MAEs of 0.4−0.50.4-0.5 eV, see Table 1. The errors of the relative CLBEs are centered and tightly distributed around zero for evGW0evGW_{0}@PBE, evGWevGW@PBE and GΔHW0G_{\Delta\mathrm{H}}W_{0}@PBE. The perturbative schemes G0W0G_{0}W_{0}@PBEh and GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} slightly overestimate the relative CLBEs. The latter is evident from the positive MEs and the error distributions in Figure 3(b,Figure 3e-Figure 3h), which are not centered at zero, but exhibit a small offset towards positive values. The RS results show a larger spread compared to G0W0G_{0}W_{0}@PBEh and the self-consistent schemes. Furthermore, outliers with errors >1>1 eV are observed for G0W0G_{0}W_{0}@PBEh and in particular for GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}. The largest outliers are primarily O1s excitations, which originate from the underlying DFT calculation, as evident from Table S8 (see SI), which shows the MAEs of the relative CLBEs from the KS-DFT eigenvalues. For all four functionals (PBE, PBE0, B3LYP, PBEh), we obtained the largest MAE at the KS-DFT level for the O1s excitations. These errors are inherited in the one-shot G0W0G_0W_0 and GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} approaches because of their perturbative nature.

The chemical shifts between CLBEs of the same atomic type can be smaller than 0.5 eV for second row elements[4] and even as small as 0.1 eV for C1s.[96] Therefore, the errors for absolute CLBEs from evGWGW and GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} are too large to align or resolve experimental XPS spectra, for which reference data are not available. The most promising methods are evGW0GW_0@PBE, G0W0G_0W_0@PBEh and GΔHW0G_{\Delta H}W_0@PBE. With MAEs between 0.2 − 0.3 eV for absolute and relative CLBEs, the accuracy is well within the chemical resolution required to interpret most XPS data. The disadvantage of the G0W0G_0W_0@PBEh(α\alpha) scheme is the need for tuning the α\alpha parameter.

In addition, the species dependence of αopt\alpha_{\mathrm{opt}} cannot be completely removed. Conversely, the accuracy of evGW0GW_0@PBE and GΔHW0G_{\Delta H}W_0@PBE is species independent. In addition, the already very good agreement of evGW0GW_0@PBE and GΔHW0G_{\Delta H}W_0@PBE with experiment might further improve with core-rich basis sets, as already mentioned before.

Comparing evGW0GW_0, G0W0G_0W_0@PBEh and GΔHW0G_{\Delta H}W_0, evGW0GW_0 is the computationally most expensive approach. In evGW0GW_0, the eigenvalues are iterated in GG (outer loop) and in each step of the outer loop we iterate Equation (1) (the QP equation) not only for the core state of interest, but for all states. Using exact methods such as contour deformation, the self-energy needs to be re-evaluated at each iteration step of the QP equation, which usually converges within 10 steps. This implies that even for small molecules we evaluate the self-energy in the evGW0GW_0 procedure several hundred times. The G0W0G_0W_0@PBEh and GΔHW0G_{\Delta H}W_0@PBE schemes are computationally significantly less expensive. For G0W0G_0W_0, the QP equation is iterated only once for the core state of interest. Given that the QP equation converges within 10 steps, the total number of self-energy evaluations amounts to 10. The computational cost of GΔHW0G_{\Delta H}W_0@PBE is only marginally larger than for G0W0G_0W_0. The Hedin shift Δ\DeltaH (see Equation (12)) is computed from Σ1s(ϵ1s0)\Sigma_{1s}(\epsilon^0_{1s}) once before the iteration of Equation (10). Given that the latter converges also in 10 steps, the self-energy needs to be calculated 11 instead of 10 times.

ETFA molecule

We further examine the performance of the eight GW approaches, which we applied to the CORE65 benchmark set in § 4.2, for C1s excitations of the ethyl trifluoroacetate (ETFA) molecule, which is also referred to as the ‘ESCA molecule’ in the literature.[97] The ETFA molecule was synthesized to demonstrate the potential of XPS for chemical analysis in the late 1960s. It contains four carbon atoms in various chemical environments, see inlet of Figure 5(a). The ETFA molecule presents a challenge because of the extreme variations of the chemical shifts, which range up to 8.0 eV and decrease from the CF3_{3} to the CH3_{3} end. The four C1s signals are separated by several electronvolts due to the widely different electronegativities of the substituents on the carbon atoms. ETFA is thus an important reference system and was very recently used to benchmark the performance of different functionals in Δ\DeltaSCF calculations[18, 99] and GW approaches.[44]

Figure 5. Comparison of the C1s XPS spectrum of ETFA (a) for absolute CLBEs and (b) relative CLBEs obtained from ev⁡GW0@PBE⁡\operatorname{ev}GW_{0}@\operatorname{PBE}, G0W0@PBEh⁡(α=0.45)G_{0}W_{0}@\operatorname{PBEh}(\alpha= 0.45), ev⁡GW@PBE⁡\operatorname{ev}GW@\operatorname{PBE}, GΔHW0@PBE⁡G_{\Delta\mathrm{H}}W_{0}@\operatorname{PBE}, GRSWRS@PBE⁡G_{\mathrm{RS}}W_{\mathrm{RS}}@\operatorname{PBE}, GRSWRS@PBE0⁡G_{\mathrm{RS}}W_{\mathrm{RS}}@\operatorname{PBE0}, GRSWRS@B3LYP⁡G_{\mathrm{RS}}W_{\mathrm{RS}}@\operatorname{B3LYP} and GRSWRS@PBEh⁡(α=0.45)G_{\mathrm{RS}}W_{\mathrm{RS}}@\operatorname{PBEh}(\alpha= 0.45). Dashed lines indicate the experimental reference.

In equilibrium, ETFA exists in two dominating conformations (anti-anti and anti-gauche), i.e., each peak in the spectrum is a superposition of the signals from both conformers.[97]

However, including the different conformations is primarily important when resolving the vibrational profiles of the peaks, which is not the scope of our benchmarking effort. The experimental conformational shifts are <0.1< 0.1 eV.[97] To ensure direct comparability with the computational data from Ref. 18, we include thus only the anti-anti conformer.

The first high-quality experimental spectrum of the free ETFA molecule in gas-phase was reported in 1973,[98] while new results were published by Travnikova et al. in 2012,[97] see Table 2. Both results were referenced in the most recent studies.[18, 44] The newer results have a higher resolution and are vibrationally resolved. More relevant for this work is that the chemical shifts of the more de-screened carbon atoms are significantly larger smaller than in the older spectrum. The difference between the experimental spectra was attributed to missing correction techniques in early multi-channel plate detectors. We follow here the reasoning of Ref. 44, pointing out that coupled-cluster results[8] are significantly closer to the new experimental data (in particular the chemical shifts). We use thus the data by Travnikova et al. as experimental reference.

C1C2C3C4MEMAE
experiment (Gelius et al.[98])299.45296.01293.07291.20
experiment (Travnikova et al.[97])298.93295.80293.19291.47
ev⁡GW0@PBE⁡\operatorname{ev}GW_{0}@\operatorname{PBE}−0.70-0.70−0.54-0.54−0.19-0.19−0.12-0.12−0.36-0.360.36
G0W0@PBEh⁡(α=0.45)G_{0}W_{0}@\operatorname{PBEh}(\alpha= 0.45)0.560.540.300.160.390.39
GΔHW0@PBE⁡G_{\Delta\mathrm{H}}W_{0}@\operatorname{PBE}−0.53-0.53−0.44-0.44−0.10-0.10−0.02-0.02−0.27-0.270.27
ev⁡GW@PBE⁡\operatorname{ev}GW@\operatorname{PBE}0.931.011.351.401.171.17
GRSWRS@PBE⁡G_{\mathrm{RS}}W_{\mathrm{RS}}@\operatorname{PBE}3.563.704.134.143.883.88
GRSWRS@PBE0⁡G_{\mathrm{RS}}W_{\mathrm{RS}}@\operatorname{PBE0}4.414.534.654.514.524.52
GRSWRS@B3LYP⁡G_{\mathrm{RS}}W_{\mathrm{RS}}@\operatorname{B3LYP}4.204.294.334.234.264.26
GRSWRS@PBEh⁡(α=0.45)G_{\mathrm{RS}}W_{\mathrm{RS}}@\operatorname{PBEh}(\alpha= 0.45)5.025.135.024.825.005.00
ev⁡GW0@PBE⁡\operatorname{ev}GW_{0}@\operatorname{PBE}[44]−0.41-0.41−0.18-0.18−0.04-0.04−0.09-0.09−0.18-0.180.18
ΔSCAN⁡\Delta\operatorname{SCAN}[18]−0.15-0.15−0.08-0.080.050.170.000.11
ΔCCSD(T)⁡\Delta\operatorname{CCSD(T)}[8]−0.35-0.35−0.24-0.24−0.31-0.31−0.23-0.23−0.28-0.280.28

Table 2. Absolute C1s CLBEs for ethyl trifluoroacetate (ETFA), see Figure 5(a) for the labels. The deviation Δexp\Delta_{\mathrm{exp}} with respect to experiment and the corresponding mean (ME) and mean absolute errors (MAE) are computed with the experimental data by Travnikova et al.[97], where Error⁡i=CLBE⁡itheory−CLBE⁡iexp\operatorname{Error}_{i}=\operatorname{CLBE}_{i}^{\mathrm{theory}}-\operatorname{CLBE}_{i}^{\mathrm{exp}}.

The comparison between the experimental spectrum and calculated spectra are shown in Figure 5. The differences to the experimental peak positions are reported in Table 2 and 3. For the absolute CLBEs, the ETFA predictions are inline with the CORE65 benchmark results. evGW0@PBE\mathrm{ev}GW_{0}@\mathrm{PBE} and GΔHW0@PBEG_{\Delta\mathrm{H}}W_{0}@\mathrm{PBE} slightly underestimate the CLBEs, while evGW@PBE\mathrm{ev}GW@\mathrm{PBE} and the RS schemes severely overestimate them, see Figure 5. G0W0@PBEh(α=0.45)G_{0}W_{0}@\mathrm{PBEh}(\alpha= 0.45) also overestimates the C1s energies slightly. As discussed in § 4.2, this is due to the fact that αopt\alpha_{\mathrm{opt}} is slightly smaller than 0.45 for C1s excitation. C1s excitations are consequently slightly overestimated in G0W0@PBEhG_{0}W_{0}@\mathrm{PBEh}. GΔHW0@PBEG_{\Delta\mathrm{H}}W_{0}@\mathrm{PBE} provides the best accuracy followed by evGW0@PBE\mathrm{ev}GW_{0}@\mathrm{PBE} with an MAE of 0.27 eV and 0.36 eV, respectively, which is consistent with the conclusion for the CORE65 set benchmark.

Turning to the relative CLBEs, the three self-consistent schemes and the G0W0@PBEhG_{0}W_{0}@\mathrm{PBEh} approach yield MAEs around 0.3 eV, which is only slightly worse than the CORE65 MAEs for C1s. evGW0\mathrm{ev}GW_{0}, evGW\mathrm{ev}GW and GΔHW0G_{\Delta\mathrm{H}}W_{0} underestimate all shifts, while G0W0@PBEhG_{0}W_{0}@\mathrm{PBEh} overestimates the shifts of the more "descreened" C atoms (C1 and C2), see Figure 5(b). GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} also provides good shifts for ETFA. The RS results are again not completely independent on the starting point. With PBE, the C1s shifts are similarly underestimated as for the self-consistent schemes, while they are slightly too large with PBEh. The RS schemes with the conventional hybrid functionals PBE0 and B3LYP yield, with MAEs below 0.1 eV, the best overall result of the 8 investigated schemes. Furthermore, except for GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} with hybrid starting points, we find that our predictions worse progressively with the electronegativity of the substituents at the C atoms. The relative CLBEs of C1 (CF3_3) and C2 (carbonyl) seem to be the ones that are more difficult to predict, while the predictions of the C3 (CH2_2) shifts are mostly within 0.1 eV of the experimental references. However, this trend can be also found in the deviations between the experimental references, see Table 3, i.e., deviations between the experiment values increase with the "descreening" of the C atoms.

Finally, we compare our results to previously published computational XPS data.[8, 18, 44, 97, 99, 100, 101] We focus here on the results from i) the same method (evGW0\mathrm{ev}GW_0), ii) one of the most successful functionals for Δ\DeltaSCF (Δ\DeltaSCAN) and iii) from a higher-level method (coupled cluster), see Tables 2 and 3. All three literature results include relativistic effects. The previous evGW0\mathrm{ev}GW_0@PBE study[44] uses the same correction scheme[49] as employed in this work. The Δ\DeltaSCAN results were obtained with the atomic zeroth-order regular approximation (ZORA).[78] For the delta coupled cluster doubles results with triples correction (Δ\DeltaCCSD(T)),[8] scalar relativistic effects were taken into account by an exact tow-component theory. The Δ\DeltaSCAN and coupled cluster results were computed for the anti-anti conformer, whereas the evGW0\mathrm{ev}GW_0 literature data are actually an average of the CLBEs of both conformers.

The absolute evGW0\mathrm{ev}GW_0@PBE CLBEs from Ref. 44 are less underestimated with respect to experiment in comparison to our results. The largest differences are observed for the descreened C1 and C2 atoms. The previously published evGW0\mathrm{ev}GW_0 results[44] are not extrapolated, but obtained with a basis sets with more core functions. As already discussed for the CORE65 results, substantially increasing the amount of core functions, seems to cure the (small) systematic underestimation of evGW0\mathrm{ev}GW_0 for absolute CLBEs. The predictions upon including more core functions should change on average by 0.25 eV for C atoms, [51] which is close to the (maximal) 0.3 eV difference we observe here. However, similar to our results, the deviation from experiments are larger for C1/C2 than for C3/C4. Since adding the core functions seems to be more relevant for the descreened environments, the chemical shifts improve, too, see Table 3. Some of the difference must be also attributed to the inclusion of the anti-gauche conformer in Ref. 44, which has slightly higher C1s BEs. [97]

C1C2C3C4MEMAE
experiment (Gelius et al.)[98]8.254.811.870.00
experiment (Travnikova et al.)[97]7.464.331.720.00
evGW0GW_{0}@PBE−0.58−0.42−0.070.00−0.360.36
G0W0G_{0}W_{0}@PBEh(α=0.45\alpha= 0.45)0.400.380.140.000.310.31
GΔHW0G_{\Delta\mathrm{H}}W_{0}@PBE−0.50−0.42−0.070.00−0.330.33
evGWGW@PBE−0.47−0.39−0.050.00−0.300.30
GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}@PBE−0.58−0.44−0.010.00−0.340.34
GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}@PBE0−0.100.020.140.000.020.09
GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}@B3LYP−0.030.050.100.000.040.06
GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}@PBEh(α=0.45\alpha= 0.45)0.210.310.200.000.240.24
evGW0GW_{0}@PBE[44]−0.32−0.090.050.00−0.120.15
Δ\DeltaSCAN[18]−0.32−0.25−0.120.00−0.230.23
Δ\DeltaCCSD(T)[8]−0.12−0.01−0.080.00−0.070.07

Table 3. Relative C1s CLBEs for ethyl trifluoroacetate (ETFA), see Figure 5 for the labels. The deviation Δexp\Delta_{\mathrm{exp}} with respect to experiment and the corresponding mean (ME) and mean absolute errors (MAE) are computed with the experimental data by Travnikova et al.[97], where Δexp=ΔCLBEitheory−ΔCLBEiexp\Delta_{\mathrm{exp}} = \Delta\mathrm{CLBE}_{i}^{\mathrm{theory}} - \Delta\mathrm{CLBE}_{i}^{\mathrm{exp}}. The labels of the C atoms are given in the inset of Figure 5(a).

We note here that Ref. 44 contains also ETFA results with G0W0@PBEh(α=0.45)G_{0}W_{0}@\mathrm{PBEh}(\alpha= 0.45) employing core-rich basis sets. However, the α\alpha values was tuned with respect to experiment using an extrapolation scheme with the cc-pVnZ basis sets. [48] An insufficiency in the basis set description (i.e. here a systematic underestimation) would be partly absorbed in the α\alpha value, which is the reason why our G0W0@PBEhG_{0}W_{0}@\mathrm{PBEh} results will agree better with experiment.

It has been recently shown that the SCAN functional yields excellent absolute and relative CLBEs for molecules[17] and solids. [19] We find that this is also true for the ETFA molecule. Δ\DeltaSCAN yields the best MAE for absolute CLBEs, which is, however, very close to the evGW0@PBE\mathrm{ev}GW_{0}@\mathrm{PBE} results[44] with core-function-rich basis set. For the relative CLBEs, the Δ\DeltaSCAN is outperformed by partially self-consistent and RS GWGW approaches as well as coupled cluster.

The absolute Δ\DeltaCCSD(T) core excitations[8] are underestimated by 0.23-0.35 eV. However, these results were obtained at the cc-pVTZ level and are probably not fully converged. 27 It is thus difficult to judge the performance of the method for absolute CLBEs. The chemical shifts on the contrary are often less affected by the basis set choice and Δ\DeltaCCSD(T) yields together with GRSWRS(@PBE0G_{\mathrm{RS}}W_{\mathrm{RS}}(@\mathrm{PBE0} or @B3LYP)@\mathrm{B3LYP}) MAEs <0.1< 0.1 eV.

Conclusion

We have presented a benchmark study of different GWGW approaches for the prediction of absolute and relative CLBEs. In addition to the evGW0@PBE\mathrm{ev}GW_{0}@\mathrm{PBE} and G0W0@PBEh(α=0.45)G_{0}W_{0}@\mathrm{PBEh}(\alpha= 0.45) methods, which were already investigated in Ref. 48, we have included evGW@PBE\mathrm{ev}GW@\mathrm{PBE} and two new methods, namely GΔHW0G_{\Delta H}W_0 and GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}, in our study. GΔHW0G_{\Delta H}W_0 is an adaption of the "Hedin shift"[72, 74] to core levels and can be considered as computationally less expensive approximation to evGW0\mathrm{ev}GW_0. In the GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} approach, the RS Green’s function is used as a new starting point and, in contrast to our previous work[65], also used to compute the screened Coulomb interaction. The purpose of introducing the RS scheme is to reduce the dependence on the starting point and the method has thus been tested with four different DFAs (PBE, PBE0, B3LYP and PBEh(α=0.45\alpha= 0.45).

By investigating the self-energy matrix elements and spectral functions, we have confirmed that evGW0\mathrm{ev}GW_0@PBE, G0W0G_0W_0@PBEh, GΔHW0G_{\Delta H}W_0@PBE, GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}@PBE, GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}@PBE0, GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}@B3LYP and GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}}@PBEh yield a unique solution. G0W0G_0W_0 schemes starting from a GGA or hybrid DFT calculation with a low amount of exact exchange do not yield a distinct QP solution. A meaningful physical solution can thus not be obtained with standard approaches such as G0W0G_0W_0@PBE, G0W0G_0W_0@PBE0 and G0W0G_0W_0@B3LYP for CLBEs.

We have studied the CORE65 benchmark set and the C1s excitations of the ETFA molecule with all 8 approaches, for which a physically reasonable solution behavior was confirmed. For the CORE65 set, evGW0\mathrm{ev}GW_0@PBE, G0W0G_0W_0@PBEh and GΔHW0G_{\Delta H}W_0@PBE yield with MAEs of 0.30, 0.33 and 0.25 eV, respectively the best results. evGW\mathrm{ev}GW and GRSWRSG_{\mathrm{RS}}W_{\mathrm{RS}} overestimate the absolute CLBEs by several electronvolts and are thus not suitable for the prediction of the absolute binding energies. Nevertheless, the RS approaches significantly reduces the starting point dependence as intended. The relative CLBEs are reasonably reproduced with all methods, but in particular with the eigenvalue self-consistent schemes and GΔHW0G_{\Delta H}W_0@PBE (MAEs <0.2< 0.2 eV). The methods exhibit a similar performance for the ETFA molecule, except that the RS approaches with standard hybrid functionals yield here the best chemical shifts.

The G0W0G_0W_0@PBEh(α\alpha) approach was introduced in our previous work[48] as computationally affordable alternative to evGW0\mathrm{ev}GW_0 that can mimic to some extent the effect of eigenvalue self-consistency in GG. However, the α\alpha-tuning is methodologically unsatisfying and the optimal α\alpha is dependent on the atomic species. We therefore recommend to use the GΔHW0G_{\Delta\mathrm{HW}_{0}}@PBE approach instead, which is in terms of computational cost comparable to G0W0G_{0}W_{0}.

Finally, we found that evGW0\mathrm{ev}GW_{0}@PBE and GΔHW0G_{\Delta\mathrm{HW}_{0}}@PBE systematically underestimate the experiment. Our comparison to the ETFA literature results and very recent work [51] suggest that this slight, but systematic underestimation can be cured by very large, core-rich basis sets, which might improve the agreement with experiment even further. Future work will consider this and focus on the development of compact and computationally efficient NAO basis sets for core-level GWGW calculations.

SUPPORTING INFORMATION

See the Supporting Information for CORE65 benchmark results, solution behavior of GRSW0G_{\mathrm{RS}}W_{0}, errors of KS-DFT for predicting absolute CLBEs and relative CLBEs, starting point dependence on the tuning parameter in PBEh, geometry of ETFA.

Acknowledgement

J. L. acknowledges the support from the National Institute of General Medical Sciences of the National Institutes of Health under award number R01-GM061870. W.Y. acknowledges the support from the National Science Foundation (grant no. CHE-1900338). D.G. acknowledges the Emmy Noether Programme of the German Research Foundation under project number 453275048 and P. R. the support from the European Union’s Horizon 2020 research and innovation program under Grant Agreement No. 951786 (The NOMAD CoE). Computing time from CSC – IT Center for Science (Finland) is gratefully acknowledged. We also thank Levi Keller for providing preliminary data for the ETFA molecule.

Data Availability Statement

The data are available in the SI and the input and output files of the FHI-aims calculations are available from the NOMAD data base. [102]

References

Graphical TOC Entry

Figure 6 (PDF p. 45). MAE (eV). absolute CLBE; relative CLBE. G_RS W_RS@. G_H W_0 @PBE; evGW_0 @PBE; G_0 W_0 @PBEh; evGW @PBE; PBE; B3LYP; PBE0; PBEh. 0; 2; 4; 6.

References

  1. [1]Bagus, P. S.; Ilton, E. S.; Nelin, C. J. The Interpretation of XPS Spectra: Insights into Materials Properties. Surf. Sci. Rep. 2013, 68, 273–304.DOI
  2. [2]Höfft, O.; Bahr, S.; Himmerlich, M.; Krischok, S.; Schaefer, J. A.; Kempter, V. Electronic Structure of the Surface of the Ionic Liquid [EMIM][Tf2N] Studied by Metastable Impact Electron Spectroscopy (MIES), UPS, and XPS. Langmuir 2006, 22, 7120–7123.pubmed.ncbi.nlm.nih.gov/16893200
  3. [3]J. Villar-Garcia, I.; F. Smith, E.; W. Taylor, A.; Qiu, F.; J. Lovelock, K. R.; G. Jones, R.; Licence, P. Charging of Ionic Liquid Surfaces under X-ray Irradiation: The Measurement of Absolute Binding Energies by XPS. Phys. Chem. Chem. Phys. 2011, 13, 2797–2808.
  4. [4]Sieg­bahn, K. ESCA Applied to Free Molecules; North-Holland Publishing: Amsterdam; London, 1969.
  5. [5]Barr, T. L. Modern ESCA: The Principles and Practice of X-Ray Photoelectron Spectroscopy; CRC Press: Boca Raton, 2020.DOI
  6. [6]Aarva, A.; Deringer, V. L.; Sainio, S.; Laurila, T.; Caro, M. A. Understanding X-ray Spectroscopy of Carbonaceous Materials by Combining Experiments, Density Functional Theory, and Machine Learning. Part I: Fingerprint Spectra. Chem. Mater. 2019, 31, 9243–9255.DOI
  7. [7]Aarva, A.; Deringer, V. L.; Sainio, S.; Laurila, T.; Caro, M. A. Understanding X-ray Spectroscopy of Carbonaceous Materials by Combining Experiments, Density Functional Theory, and Machine Learning. Part II: Quantitative Fitting of Spectra. Chem. Mater. 2019, 31, 9256–9267.
  8. [8]Zheng, X.; Cheng, L. Performance of Delta-Coupled-Cluster Methods for Calculations of Core-Ionization Energies of First-Row Elements. J. Chem. Theory Comput. 2019, 15, 4945–4955.DOI
  9. [9]Sen, S.; Shee, A.; Mukherjee, D. Inclusion of Orbital Relaxation and Correlation through the Unitary Group Adapted Open Shell Coupled Cluster Theory Using Non-Relativistic and Scalar Relativistic Hamiltonians to Study the Core Ionization Potential of Molecules Containing Light to Medium-Heavy Elements. J. Chem. Phys. 2018, 148, 054107.DOI
  10. [10]Hohenberg, P.; Kohn, W. Inhomogeneous Electron Gas. Phys. Rev. 1964, 136, B864–B871.
  11. [11]Kohn, W.; Sham, L. J. Self-Consistent Equations Including Exchange and Correlation Effects. Phys. Rev. 1965, 140, A1133–A1138.DOI
  12. [12]Parr, R. G.; Yang, W. Density-Functional Theory of Atoms and Molecules; Oxford University Press, 1989.
  13. [13]Bagus, P. S. Self-Consistent-Field Wave Functions for Hole States of Some Ne-Like and Ar-Like Ions. Phys. Rev. 1965, 139, A619–A634.DOI
  14. [14]Pueyo Bellafont, N.; Bagus, P. S.; Illas, F. Prediction of Core Level Binding Energies in Density Functional Theory: Rigorous Definition of Initial and Final State Contributions and Implications on the Physical Meaning of Kohn-Sham Energies. J. Chem. Phys. 2015, 142, 214102.DOI
  15. [15]Pueyo Bellafont, N.; Álvarez Saiz, G.; Viñes, F.; Illas, F. Performance of Minnesota Functionals on Predicting Core-Level Binding Energies of Molecules Containing Main-Group Elements. Theor Chem Acc 2016, 135, 35.DOI
  16. [16]Viñes, F.; Sousa, C.; Illas, F. On the Prediction of Core Level Binding Energies in Molecules, Surfaces and Solids. Phys. Chem. Chem. Phys. 2018, 20, 8403–8410.DOI
  17. [17]Kahk, J. M.; Lischner, J. Accurate Absolute Core-Electron Binding Energies of Molecules, Solids, and Surfaces from First-Principles Calculations. Phys. Rev. Mater. 2019, 3, 100801.DOI
  18. [18]Klein, B. P.; Hall, S. J.; Maurer, R. J. The Nuts and Bolts of Core-Hole Constrained Ab Initio Simulation for K-shell x-Ray Photoemission and Absorption Spectra. J. Phys.: Condens. Matter 2021, 33, 154005.DOI
  19. [19]Kahk, J. M.; Michelitsch, G. S.; Maurer, R. J.; Reuter, K.; Lischner, J. Core Electron Binding Energies in Solids from Periodic All-Electron Δ-Self-Consistent-Field Calculations. J. Phys. Chem. Lett. 2021, 12, 9353–9359.DOI
  20. [20]Pinheiro, M.; Caldas, M. J.; Rinke, P.; Blum, V.; Scheffler, M. Length Dependence of Ionization Potentials of Transacetylenes: Internally Consistent DFT/GW Approach. Phys. Rev. B 2015, 92, 195134.DOI
  21. [21]Golze, D.; Wilhelm, J.; van Setten, M. J.; Rinke, P. Core-Level Binding Energies from GW: An Efficient Full-Frequency Approach within a Localized Basis. J. Chem. Theory Comput. 2018, 14, 4856–4869.DOI
  22. [22]Michelitsch, G. S.; Reuter, K. Efficient Simulation of Near-Edge x-Ray Absorption Fine Structure (NEXAFS) in Density-Functional Theory: Comparison of Core-Level Constraining Approaches. J. Chem. Phys. 2019, 150, 074104.DOI
  23. [29]Martin, R. M.; Reining, L.; Ceperley, D. M. Interacting Electrons; Cambridge University Press, 2016.
  24. [30]Golze, D.; Dvorak, M.; Rinke, P. The GW Compendium: A Practical Guide to Theoretical Photoemission Spectroscopy. Front. Chem. 2019, 7.DOI
  25. [31]Reining, L. The GW Approximation: Content, Successes and Limitations. WIREs Comput. Mol. Sci. 2018, 8, e1344.DOI
  26. [32]Hedin, L. New Method for Calculating the One-Particle Green’s Function with Application to the Electron-Gas Problem. Phys. Rev. 1965, 139, A796–A823.DOI
  27. [33]Rasmussen, A.; Deilmann, T.; Thygesen, K. S. Towards Fully Automated GW Band Structure Calculations: What We Can Learn from 60.000 Self-Energy Evaluations. npj Comput Mater 2021, 7, 1–9.DOI
  28. [34]van Setten, M. J.; Caruso, F.; Sharifzadeh, S.; Ren, X.; Scheffler, M.; Liu, F.; Lischner, J.; Lin, L.; Deslippe, J. R.; Louie, S. G.; Yang, C.; Weigend, F.; Neaton, J. B.; Evers, F.; Rinke, P. GW100: Benchmarking G0W0 for Molecular Systems. J. Chem. Theory Comput. 2015, 11, 5665–5687.DOI
  29. [35]Stuke, A.; Kunkel, C.; Golze, D.; Todorović, M.; Margraf, J. T.; Reuter, K.; Rinke, P.; Oberhofer, H. Atomic Structures and Orbital Energies of 61,489 Crystal-Forming Organic Molecules. Sci Data 2020, 7, 58.arxiv.org/abs/2001.08954
  30. [36]Bruneval, F.; Dattani, N.; van Setten, M. J. The GW Miracle in Many-Body Perturbation Theory for the Ionization Potential of Molecules. Front. Chem. 2021, 9.DOI
  31. [37]Blase, X.; Attaccalite, C.; Olevano, V. First-Principles GW Calculations for Fullerenes, Porphyrins, Phthalocyanine, and Other Molecules of Interest for Organic Photovoltaic Applications. Phys. Rev. B 2011, 83, 115103.DOI
  32. [38]Ren, X.; Rinke, P.; Blum, V.; Wieferink, J.; Tkatchenko, A.; Sanfilippo, A.; Reuter, K.; Scheffler, M. Resolution-of-Identity Approach to Hartree–Fock, Hybrid Density Functionals, RPA, MP2 andGWwith Numeric Atom-Centered Orbital Basis Functions. New J. Phys. 2012, 14, 053020.DOI
  33. [39]van Setten, M. J.; Weigend, F.; Evers, F. The GW-Method for Quantum Chemistry Applications: Theory and Implementation. J. Chem. Theory Comput. 2013, 9, 232–246.
  34. [40]Bruneval, F.; Rangel, T.; Hamed, S. M.; Shao, M.; Yang, C.; Neaton, J. B. Molgw 1: Many-body Perturbation Theory Software for Atoms, Molecules, and Clusters. Comput. Phys. Commun 2016, 208, 149–161.
  35. [41]Wilhelm, J.; Del Ben, M.; Hutter, J. GW in the Gaussian and Plane Waves Scheme with Application to Linear Acenes. J. Chem. Theory Comput. 2016, 12, 3623–3635.DOI
  36. [42]Förster, A.; Franchini, M.; van Lenthe, E.; Visscher, L. A Quadratic Pair Atomic Resolution of the Identity Based SOS-AO-MP2 Algorithm Using Slater Type Orbitals. J. Chem. Theory Comput. 2020, 16, 875–891.DOI
  37. [43]Zhu, T.; Chan, G. K.-L. All-Electron Gaussian-Based G0W0 for Valence and Core Excitation Energies of Periodic Systems. J. Chem. Theory Comput. 2021, 17, 727–741.DOI
  38. [44]Mejia-Rodriguez, D.; Kunitsa, A.; Aprà, E.; Govind, N. Scalable Molecular GW Calculations: Valence and Core Spectra. J. Chem. Theory Comput. 2021, 17, 7504–7517.
  39. [45]Aoki, T.; Ohno, K. Accurate Quasiparticle Calculation of X-Ray Photoelectron Spectra of Solids. J. Phys.: Condens. Matter 2018, 30, 21LT01.DOI
  40. [46]van Setten, M. J.; Costa, R.; Viñes, F.; Illas, F. Assessing GW Approaches for Predicting Core Level Binding Energies. J. Chem. Theory Comput. 2018, 14, 877–883.DOI
  41. [47]Voora, V. K.; Galhenage, R.; Hemminger, J. C.; Furche, F. Effective One-Particle Energies from Generalized Kohn–Sham Random Phase Approximation: A Direct Approach for Computing and Analyzing Core Ionization Energies. J. Chem. Phys. 2019, 151, 134106.DOI
  42. [48]Golze, D.; Keller, L.; Rinke, P. Accurate Absolute and Relative Core-Level Binding Energies from GW. J. Phys. Chem. Lett. 2020, 11, 1840–1847.DOI
  43. [49]Keller, L.; Blum, V.; Rinke, P.; Golze, D. Relativistic Correction Scheme for Core-Level Binding Energies from GW. J. Chem. Phys. 2020, 153, 114110.DOI
  44. [50]Duchemin, I.; Blase, X. Robust Analytic-Continuation Approach to Many-Body GW Calculations. J. Chem. Theory Comput. 2020, 16, 1742–1756.DOI
  45. [51]Mejia-Rodriguez, D.; Kunitса, A.; Aprà, E.; Govind, N. On the Basis Set Selection for Molecular Core-Level GW Calculations. arXiv:2203.10169 2022,
  46. [52]Yao, Y.; Golze, D.; Rinke, P.; Blum, V.; Kanai, Y. All-Electron BSE@GW Method for K-Edge Core Electron Excitation Energies. J. Chem. Theory Comput. 2022,DOI
  47. [53]Wilhelm, J.; Hutter, J. Periodic GW Calculations in the Gaussian and Plane-Waves Scheme. Phys. Rev. B 2017, 95, 235123.
  48. [54]Ren, X.; Merz, F.; Jiang, H.; Yao, Y.; Rampp, M.; Lederer, H.; Blum, V.; Scheffler, M. All-Electron Periodic G₀W₀ Implementation with Numerical Atomic Orbital Basis Functions: Algorithm and Benchmarks. Phys. Rev. Mater. 2021, 5, 013807.
  49. [55]Wilhelm, J.; Golze, D.; Talirz, L.; Hutter, J.; Pignedoli, C. A. Toward GW Calculations on Thousands of Atoms. J. Phys. Chem. Lett. 2018, 9, 306–312.DOI
  50. [56]Wilhelm, J.; Seewald, P.; Golze, D. Low-Scaling GW with Benchmark Accuracy and Application to Phosphorene Nanosheets. J. Chem. Theory Comput. 2021, 17, 1662–1677.DOI
  51. [57]Duchemin, I.; Blase, X. Cubic-Scaling All-Electron GW Calculations with a Separable Density-Fitting Space–Time Approach. J. Chem. Theory Comput. 2021, 17, 2383–2393.DOI
  52. [58]Förster, A.; Visscher, L. Low-Order Scaling Quasiparticle Self-Consistent GW for Molecules. Front. Chem. 2021, 9.DOI
  53. [59]Schöne, W.-D.; Eguiluz, A. G. Self-Consistent Calculations of Quasiparticle States in Metals and Semiconductors. Phys. Rev. Lett. 1998, 81, 1662–1665.DOI
  54. [60]Caruso, F.; Rinke, P.; Ren, X.; Scheffler, M.; Rubio, A. Unified Description of Ground and Excited States of Finite Systems: The Self-Consistent GW Approach. Phys. Rev. B 2012, 86, 081102.
  55. [61]Caruso, F.; Rinke, P.; Ren, X.; Rubio, A.; Scheffler, M. Self-Consistent GW: All-electron Implementation with Localized Basis Functions. Phys. Rev. B 2013, 88, 075105.DOI
  56. [62]van Schilfgaarde, M.; Kotani, T.; Faleev, S. Quasiparticle Self-Consistent GW Theory. Phys. Rev. Lett. 2006, 96, 226402.
  57. [63]Caruso, F.; Dauth, M.; van Setten, M. J.; Rinke, P. Benchmark of GW Approaches for the GW100 Test Set. J. Chem. Theory Comput. 2016, 12, 5076–5087.pubmed.ncbi.nlm.nih.gov/27631585
  58. [64]Grumet, M.; Liu, P.; Kaltak, M.; Klimeš, J.; Kresse, G. Beyond the Quasiparticle Approximation: Fully Self-Consistent GW Calculations. Phys. Rev. B 2018, 98, 155143.DOI
  59. [65]Jin, Y.; Su, N. Q.; Yang, W. Renormalized Singles Green’s Function for Quasi-Particle Calculations beyond the G0W0 Approximation. J. Phys. Chem. Lett. 2019, 10, 447–452.DOI
  60. [66]Ren, X.; Tkatchenko, A.; Rinke, P.; Scheffler, M. Beyond the Random-Phase Approximation for the Electron Correlation Energy: The Importance of Single Excitations. Phys. Rev. Lett. 2011, 106, 153003.pubmed.ncbi.nlm.nih.gov/21568551
  61. [67]Ren, X.; Rinke, P.; Scuseria, G. E.; Scheffler, M. Renormalized Second-Order Perturbation Theory for the Electron Correlation Energy: Concept, Implementation, and Benchmarks. Phys. Rev. B 2013, 88, 035120.arxiv.org/abs/1212.3674
  62. [68]Li, J.; Chen, Z.; Yang, W. Multireference Density Functional Theory for Describing Ground and Excited States with Renormalized Singles. J. Phys. Chem. Lett. 2022, 13, 894–903.DOI
  63. [69]Li, J.; Chen, Z.; Yang, W. Renormalized Singles Green’s Function in the T-Matrix Approximation for Accurate Quasiparticle Energy Calculation. J. Phys. Chem. Lett. 2021, 12, 6203–6210.DOI
  64. [70]Zhang, D.; Su, N. Q.; Yang, W. Accurate Quasiparticle Spectra from the T-Matrix Self-Energy and the Particle–Particle Random Phase Approximation. J. Phys. Chem. Lett. 2017, 8, 3223–3227.DOI
  65. [71]Zhang, D.; Yang, W. Accurate and Efficient Calculation of Excitation Energies with the Active-Space Particle–Particle Random Phase Approximation. J. Chem. Phys. 2016, 145, 144105.pubmed.ncbi.nlm.nih.gov/27782522
  66. [72]Pollehn, T. J.; Schindlmayr, A.; Godby, R. W. Assessment of the GW approximation using Hubbard chains. J. Phys.: Condens. Matter 1998, 10, 1273–1283.
  67. [73]Govoni, M.; Galli, G. Large Scale GW Calculations. J. Chem. Theory Comput. 2015, 11, 2680–2696.
  68. [74]Lee, J. D.; Gunnarsson, O.; Hedin, L. Transition from the Adiabatic to the Sudden Limit in Core-Level Photoemission: A Model Study of a Localized System. Phys. Rev. B 1999, 60, 8034–8049.DOI
  69. [75]Rinke, P.; Qteish, A.; Neugebauer, J.; Freysoldt, C.; Scheffler, M. Combining GWcalculations with Exact-Exchange Density-Functional Theory: An Analysis of Valence-Band Photoemission for Compound Semiconductors. New J. Phys. 2005, 7, 126–126.
  70. [76]Ren, X.; Rinke, P.; Joas, C.; Scheffler, M. Random-Phase Approximation and Its Applications in Computational Chemistry and Materials Science. J Mater Sci 2012, 47, 7447–7471.DOI
  71. [77]Szabo, A.; Ostlund, N. S. Modern Quantum Chemistry: Introduction to Advanced Electronic Structure Theory; Courier Corporation, 2012.
  72. [78]Blum, V.; Gehrke, R.; Hanke, F.; Havu, P.; Havu, V.; Ren, X.; Reuter, K.; Scheffler, M. Ab Initio Molecular Simulations with Numeric Atom-Centered Orbitals. Comput. Phys. Commun 2009, 180, 2175–2196.DOI
  73. [79]Havu, V.; Blum, V.; Havu, P.; Scheffler, M. Efficient O(N) Integration for All-Electron Electronic Structure Calculation Using Numeric Basis Functions. J. Comput. Phys. 2009, 228, 8367–8379.
  74. [80]See http://www.qm4d.info for an in-house program for QM/MM simulations.
  75. [81]Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized Gradient Approximation Made Simple. Phys. Rev. Lett. 1996, 77, 3865–3868.pubmed.ncbi.nlm.nih.gov/10062328
  76. [82]Atalla, V.; Yoon, M.; Caruso, F.; Rinke, P.; Scheffler, M. Hybrid Density Functional Theory Meets Quasiparticle Calculations: A Consistent Electronic Structure Approach. Phys. Rev. B 2013, 88, 165122.DOI
  77. [83]Adamo, C.; Barone, V. Toward Reliable Density Functional Methods without Adjustable Parameters: The PBE0 Model. J. Chem. Phys. 1999, 110, 6158–6170.DOI
  78. [84]Ernzerhof, M.; Scuseria, G. E. Assessment of the Perdew–Burke–Ernzerhof Exchange-Correlation Functional. J. Chem. Phys. 1999, 110, 5029–5036.DOI
  79. [85]Becke, A. D. Density-functional Thermochemistry. III. The Role of Exact Exchange. J. Chem. Phys. 1993, 98, 5648–5652.DOI
  80. [86]Lee, C.; Yang, W.; Parr, R. G. Development of the Colle-Salvetti Correlation-Energy Formula into a Functional of the Electron Density. Phys. Rev. B 1988, 37, 785–789.pubmed.ncbi.nlm.nih.gov/9944570
  81. [87]Dunning, T. H. Gaussian Basis Sets for Use in Correlated Molecular Calculations. I. The Atoms Boron through Neon and Hydrogen. J. Chem. Phys. 1989, 90, 1007–1023.DOI
  82. [88]Wilson, A. K.; van Mourik, T.; Dunning, T. H. Gaussian Basis Sets for Use in Correlated Molecular Calculations. VI. Sextuple Zeta Correlation Consistent Basis Sets for Boron through Neon. J. Mol. Struct. THEOCHEM 1996, 388, 339–349.DOI
  83. [89]Vahtras, O.; Almlöf, J.; Feyereisen, M. W. Integral Approximations for LCAO-SCF Calculations. Chem. Phys. Lett 1993, 213, 514–518.DOI
  84. [91]Sankari, R.; Ehara, M.; Nakatsuji, H.; De Fanis, A.; Aksela, H.; Sorensen, S. L.; Piancastelli, M. N.; Kukk, E.; Ueda, K. High Resolution O 1s Photoelectron Shake-up Satellite Spectrum of H2O. Chem. Phys. Lett 2006, 422, 51–57.DOI
  85. [92]Schirmer, J.; Angonoa, G.; Svensson, S.; Nordfors, D.; Gelius, U. High-Energy Photoelectron C 1s and O 1s Shake-up Spectra of CO. J. Phys. B: Atom. Mol. Phys. 1987, 20, 6031–6040.DOI
  86. [93]Shishkin, M.; Kresse, G. Self-Consistent GW Calculations for Semiconductors and Insulators. Phys. Rev. B 2007, 75, 235102.DOI
  87. [94]Byun, Y.-M.; Öğüt, S. Practical GW Scheme for Electronic Structure of 3d-Transition-Metal Monoxide Anions: ScO-, TiO-, CuO-, and ZnO-. J. Chem. Phys. 2019, 151, 134305.DOI
  88. [95]Marom, N.; Caruso, F.; Ren, X.; Hofmann, O. T.; Körzdörfer, T.; Chelikowsky, J. R.; Rubio, A.; Scheffler, M.; Rinke, P. Benchmark of GW Methods for Azabenzene­s. Phys. Rev. B 2012, 86, 245127.DOI
  89. [96]Pireaux, J. J.; Svensson, S.; Basilier, E.; Malmqvist, P.-Å.; Gelius, U.; Caudano, R.; Siegbahn, K. Core-Electron Relaxation Energies and Valence-Band Formation of Linear Alkanes Studied in the Gas Phase by Means of Electron Spectroscopy. Phys. Rev. A 1976, 14, 2133–2145.DOI
  90. [97]Travnikova, O.; Børve, K. J.; Patanen, M.; Söderström, J.; Miron, C.; Sæthre, L. J.; Mårtensson, N.; Svensson, S. The ESCA Molecule—Historical Remarks and New Results. J. Electron Spectrosc. Relat. Phenom. 2012, 185, 191–197.DOI
  91. [98]Gelius, U.; Basilier, E.; Svensson, S.; Bergmark, T.; Siegbahn, K. A High Resolution ESCA Instrument with X-ray Monochromator for Gases and Solids. J. Electron Spectrosc. Relat. Phenom. 1973, 2, 405–434.DOI
  92. [99]Delesma, F. A.; Van den Bossche, M.; Grönbeck, H.; Calaminici, P.; Köster, A. M.; Pettersson, L. G. M. A Chemical View on X-ray Photoelectron Spectroscopy: The ESCA Molecule and Surface-to-Bulk XPS Shifts. ChemPhysChem 2018, 19, 169–174.DOI
  93. [100]Van den Bossche, M.; Martin, N. M.; Gustafson, J.; Hakanoglu, C.; Weaver, J. F.; Lundgren, E.; Grönbeck, H. Effects of Non-Local Exchange on Core Level Shifts for Gas-Phase and Adsorbed Molecules. J. Chem. Phys. 2014, 141, 034706.DOI
  94. [101]Travnikova, O. et al. Energy-Dependent Relative Cross Sections in Carbon 1s Photoionization: Separation of Direct Shake and Inelastic Scattering Effects in Single Molecules. J. Phys. Chem. A 2019, 123, 7619–7636.DOI
  95. [102]Golze, D. Dataset in NOMAD repository. 2022; DOI will be added after revision.

Paper details

vlnc:W_fekaq8rp
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents