Abstract
For more than three decades, nearly freeelectron elemental metals have been a topic of debate because the computed bandwidths are significantly wider in the local density approximation to densityfunctional theory (DFT) than indicated by angleresolved photoemission (ARPES) experiments. Here, we systematically investigate this using first principles calculations for alkali and alkalineearth metals using DFT and various beyondDFT methods such as metaGGA, G_{0}W_{0}, hybrid functionals (YSPBE0, B3LYP), and LDA + eDMFT. We find that the static nonlocal exchange, as partly included in the hybrid functionals, significantly increase the bandwidths even compared to LDA, while the G_{0}W_{0} bands are only slightly narrower than in LDA. The agreement with the ARPES is best when the local approximation to the selfenergy is used in the LDA + eDMFT method. We infer that even moderately correlated systems with partially occupied s orbitals, which were assumed to approximate the uniform electron gas, are very well described in terms of shortrange dynamical correlations that are only local to an atom.
Similar content being viewed by others
Introduction
Materials are often loosely categorized into weakly and strongly correlated systems depending on the strength of electronelectron correlation. Densityfunctional theory (DFT) in the local density approximation (LDA), which maps each point in space inside a crystal to a uniform electrongas problem through the electron density, is found to be fairly successful in the qualitative description of weakly correlated materials. However it is a groundstate theory, an accurate description of the excitedstate properties even in moderately correlated materials lies outside the realm of conventional DFT. On the other hand, over the last few decades, there has been a longstanding effort to develop either perturbative, stochastic, or hybridfunctional approaches to understanding the strongly correlated and excitedstate properties^{1,2,3,4,5,6}. As a result, metaGGAs, hybrid functionals, the GW approximation^{7,8}, and DFT+ dynamical meanfield theory (DMFT) methods have become quite popular for understanding correlated solids. These methods are commonly referred to as “beyondDFT” methods.
BeyondDFT methods are often applied to openshell delectron systems, where the corrections to DFT are large and often qualitative, but very little is known about the applicability of these methods to solids that are only moderately correlated. In particular, the DFT + eDMFT method, which has gained wide popularity for describing the electronic structure of localized d and felectron systems, has so far mainly been applied to study strongly correlated systems^{3,9,10,11,12,13}, while its extension to the weak coupling limit is rarely discussed in the literature^{14,15}. Similarly other “beyondDFT” methods such as metaGGAs, hybrid functionals, and GW have mostly been applied to insulating materials^{7,8} with strong or moderate correlations, but it is unclear if these methods are equally successful when applied to weakly or moderately correlated metallic systems^{16,17}.
In a recent work^{18}, we have systematically tested “beyondDFT” methods on insulating transitionmetal monoxides, and showed that the hybrid functionals and eDMFT methods perform well, while other methods are inadequate in at least a couple of monoxides. Here, we use the same set of methods to investigate the alkali and alkalineearth metals, whose electronic structure has been studied since the inception of quantum mechanics^{19}. They are often assumed to be the closest natural analogs to the uniform electron gas (UEG). DFT either with LDA or the generalized gradient approximation (GGA) is found to be fairly successful in the qualitative description of these metals, making them exemplary systems for weakly correlated electrons. However, the excitedstate spectra and consequently the occupied bandwidths often quantitatively disagree with angleresolved photoemission spectroscopy (ARPES) experiments. For example, ARPES studies by Lyo and Plummer^{20} indicate that the bandwidth of the first occupied band of sodium is substantially narrower than predicted by LDA. This has triggered a large amount of experimental^{21,22,23,24,25,26} and theoretical^{27,28,29,30,31,32,33,34,35,36,37,38,39,40} work over the last three decades.
An earlier study using the G_{0}W_{0} approximation^{41}, which truncates the selfenergy to first order in the Green’s function G_{0} and the screened Coulomb interaction W_{0}, pointed to the importance of manybody effects and suggests that the lowestorder perturbative term may not be sufficient to describe the excitation spectrum in these materials^{28}. This motivated further GW studies to include higherorder perturbative corrections to the band limit^{33,39,40,42,43,44}. Mohan and Sernelius^{27} and more recently Kutepov^{36} found that the inclusion of vertex corrections (GWΓ) modifies the GW selfenergy, but the loworder vertex corrections to the selfenergy and the dielectric function nearly cancel each other, resulting in a bandwidth renormalization not very different from the G_{0}W_{0} prediction.
Recent work using a sophisticated unbiased reptation Monte Carlo method further indicates that the vertex corrections and selfconsistency aspects of GW cancel to a large degree near the Fermi surface^{45}. This result also hints that the elemental metallic systems might be moderately correlated. An opposite trend of increasing the bandwidth compared to LDA is found in recent studies using selfconsistent GW^{33,39} and variational and fixednode diffusion quantum Monte Carlo^{46} techniques. Contrary to the indications from the GW approximation, Zhu and Overhauser^{47} predicted that the spin fluctuations within a paramagnon pole model could account for the bandwidth reduction in Na, but a more recent study^{38} found that this effect is negligible. Interestingly, very recent ARPES results^{48} have ruled out the proposed^{38} strong coupling between the conduction electrons and spin fluctuations in Na and demands new theoretical insights. The entire situation remains unresolved, and the reason for the discrepancies between ARPES measurements and the theoretically predicted bandwidths in these simple metals remains one of the fundamental questions in condensed matter physics.
On a different tack, the narrowing of the ARPES spectra has alternatively been ascribed to finalstate effects, which would require treating the outgoing electron as embedded in an interacting uniform electron gas inside the solid, rather than as a freeelectron leaving the solid^{39}. A similar approach was taken in refs. ^{49,28}, but with the inclusion of surface effects. Such an interpretation was challenged in ref. ^{29} (see also ref. ^{40}), as it would invalidate the accepted interpretation of the ARPES experiments as measuring the singleparticle spectral function weighted by matrixelement effects^{50}. This would have farreaching implications for the interpretation of all ARPES data to date.
To answer such questions, our computational study is carried out in the framework of a broader ongoing program involving the systematic performance and curation of electronic structure calculations using a range of methodologies and applied to a range of materials^{18,51}. The premise of this approach is that for a proper evaluation of the strengths and weaknesses of various first principles methods applied to a given class of materials, a methodologically heterogeneous literature is not enough. Rather, a set of calculations for different functionals performed on a consistent footing is essential. This is especially true in the context of highthroughput computation, where a desire for accuracy has to be carefully weighed against issues of consistency and modest computational load. Moreover, the availability of these results in a materials database, such as the “beyondDFT” component (https://jarvis.nist.gov/jarvisbdft) of the JARVIS database (https://jarvis.nist.gov/jarvisdft) used here, makes the comparisons between different functionals broadly available to the materials science community, providing a guide for future calculations on related systems.
Working in this context, we systematically apply several DFT and beyondDFT methods to the elemental metallic systems from the first and second columns of the periodic table (Li to Cs and Be to Sr). We resolve a controversy over the disagreement between theory and experiment for the occupied bandwidths of such systems, showing how they depend on the effects of local and nonlocal exchange and correlations. We find that the band narrowing is surprisingly well described with nonperturbative dynamical correlations modeled as local to an atom rather than to a point in 3D space, emphasizing the importance of umklapp contributions to the electron selfenergy at higher order in perturbation theory, beyond GW, seem to have a significant effect even in these systems. In particular, in this letter we show that the elemental metals with partially occupied s orbitals, which are usually assumed to be nearlyfreeelectron metals, are in fact moderately correlated, thus forcing a reconsideration of longheld notions about these simple metals.
Results
All computations are performed for the roomtemperature experimental crystal structures obtained from the ICSD database. Most of the elemental metals studied here crystallize in the bcc structure at room temperature, except for Be and Mg which crystallize in hcp, and Ca and Sr in fcc. In the following we compare the electronic band structures using the abovementioned methods with ARPES data, which are available for Na, K, and Mg. We describe each of these compounds in detail here, and direct the reader to the Supplementary Information for the others.
Sodium
We first describe our results for Na, which has been most widely discussed in the literature as a prototypical UEG elemental metal. In Fig. 1, ARPES data are obtained from earlier experiment by Lyo and Plummer^{20} and shown as pink dots. Very recently an ARPES experiment for Na was repeated with more resolution by Potorochin et al.^{48} (shown as gray dots), indicating that the bandwidth is still underestimated as compared to LDA, and implying that a more advanced theoretical understanding is still needed. The computed band structures in LDA, mBJ, B3LYP, G_{0}W_{0}, and eDMFT are shown in Fig. 1. We directly compare with the ARPES data (shown in pink and gray dots), which are adopted from refs. ^{20,48}. Since the LDA and mBJ bands are found to be almost identical, we only show the LDA bands in the figure. While the bands are very similar to each other close to the Fermi surface near the Npoint for all the methods, the differences between them become more evident near the Γ point.
Among these methods, the discrepancy between the B3LYP bands and the experiment is large. The agreement with ARPES for simple LDA is substantially better than for B3LYP. The screened hyrbid (YSPBE0) performs slightly better than the unscreened one, but still substantially worse than LDA. We also compute the G_{0}W_{0} bands, which is challenging for metallic systems due to numerical difficulties associated with the treatment of the Fermi surface singularity, often leading to less accurate results on the Matsubara axis, and consequently, extreme difficulty in the analytic continuation to real frequencies. This has been discussed in detail in ref. ^{52}, where the algorithm for the frequency convolution on the Matsubara axis was improved. This allows a stable analytic continuation of the imaginaryaxis data via a Pade approximation, which is carried out for GW bandstructure calculations throughout the Brillouin zone (BZ).
It is evident in Fig. 1 that the G_{0}W_{0} band (yellow) is very close to the LDA bands, narrowing only slightly relative to the LDA even at the Γpoint. The agreement between theory and earlier experiment is best for the eDMFT method; the spectral function (green) reproduces the ARPES data well throughout the BZ, except near the Γ point where the agreement is slightly worse. This is again reconfirmed in the recent ARPES experiment^{48}, which was performed with more resolution than the earlier experiment by the Plummer group. The agreement now is better with LDA + eDMFT.
Potassium
We perform similar comparisons for bcc K in Fig. 2, where we plot the band structures as computed in LDA, mBJ, B3LYP, YSPBE0, G_{0}W_{0}, and eDMFT. The ARPES data (shown in pink dots) are taken from ref. ^{23}. For K, the ARPES data are not available for the entire BZ, extending only about halfway from Γ to N. We again see that the bands obtained using various methods cross the Fermi energy at almost the same k point, but the dispersion towards Γ is quite different for the various methods. The B3LYP band disperses the most, reaching −2.76 eV at Γ, while LDA and mBJ disperse only to −2.15 eV. As was the case for Na, the LDA and mBJ bands are almost indistinguishable, so we display only the LDA in Fig. 2. The G_{0}W_{0} bands narrow only slightly, reducing the bandwidth to 2.00 eV from the LDA value of 2.15 eV. The eDMFT narrows the bandwidth to 1.42 eV at Γ, as compared with the experimental value of 1.6 eV. (Slightly away from Γ, the ARPES data are asymmetric which could be due to inaccuracy in the experimental data^{23}. As a result, the eDMFT band crosses the ARPES band only on the right half of the figure. A similar artifact appears for Na in Fig. 1).
Magnesium
Unlike most of the elemental metals, Mg crystallizes in the hcp instead of the bcc structure at room temperature and has multiple occupied bands. We display the bands in Fig. 3. The ARPES data (pink dots, taken from ref. ^{22}) indicate two bands within this energy range, one with small and one with large dispersion from the Fermi energy to Γ. In particular, one band disperses from Γ at −0.82 eV up to the Fermi energy (E = 0), and another disperses from −1.85 eV down to −6.15 eV along Γ–A–Γ. Here we also find a similar trend in the computed bands using various methods as noticed for Na and K. The B3LYP bands disperse the most and lie furthest from the experimental bands. Once again, the LDA and mBJ bands are both very similar. The first G_{0}W_{0} band disperses to −1.29 eV at Γ, while in eDMFT it is −0.82 eV and in the experiment, it is −0.9 eV. The second occupied band crosses the Γ point at −6.66, −6.18, and −6.15 eV in G_{0}W_{0}, eDMFT, and experiment, respectively. We again find the best overall agreement with ARPES for the eDMFT spectral function. We find similar trends in the band structures of other elemental metals in this family as well, as discussed in the Supplementary Figures. The sharp spectral function in eDMFT, which is common in Figs. 1–3 reflects that electronelectron scattering is weak in these metals.
Bandwidth of elemental metals
To quantify the observed trends in alkali and alkalineearth metals, we compare the bandwidths of the occupied bands, defined as the depth of the energy at Γ below E_{F}, as computed using LDA, mBJ, B3LYP, YSPBE0, G_{0}W_{0}, and eDMFT, in Table 1. We describe the bandwidths within a 0–9 eV window in Fig. 4a and the relative error from the experimental value in Fig. 4b.
We also compare them with available experiments from ARPES as well as other photoemission spectroscopies. As the Fermi surface is almost exactly spherical in these compounds, and the band structure is close to a renormalized freeelectron solution in the proper periodic potential, the most relevant number here is the bandwidth. LDA overestimates the bandwidths across the family as expected, except for Be. The relative error in LDA as compared to experiment goes as high as 56% (Fig. 4b). The hybrid functional, which corrects some of the selfinteraction error by incorporating a fraction of exact exchange, correcting the nonlocal part of the selfenergy, has been found to significantly improve the descriptions of many delectron systems^{18,53,54,55,56}, in particular Mott insulators. Here we find that the static nonlocal corrections included either in B3LYP or YSPBE0 have the opposite effect, worsening the agreement with ARPES as they significantly overestimate the bandwidth for all the systems studied here, with relative error, which can reach as high as 80% (Fig. 4b). This clearly shows that the hybrid functionals do not improve on LDA for metallic systems^{57}. mBJ performs better than B3LYP and gives very similar bandwidths as LDA, except for heavy elements such as Sr and Cs, where mBJ performs better than LDA. Next, we notice that the bandwidths as computed in the G_{0}W_{0} method are in better agreement than LDA or mBJ, but are still substantially wider than those measured by ARPES. In G_{0}W_{0}, the relative error varies from 1 to 43% across the compounds studied here. Various implementations of the GW approximation, with and without vertex corrections, give the bandwidth for Na in the range of 2.5–3.2 eV^{37,58}. Here we obtain 3.15 eV, quite similar to recently reported selfconsistent quasiparticle GW values of 3.17 eV^{58}. This result indicates that singleshot GW or G_{0}W_{0} perform similar to that of selfconsistent quasiparticle GW and explains that for the elemental metals selfconsistency may not be necessary. The improvement of GW as compared to static hybrid methods, which incorporate some HartreeFock selfenergy, shows that the dynamical nature of the correlations, with proper description of screening, is certainly an important factor in the band narrowing in Na and other alkali metals.
Finally, the bandwidth narrowing is even stronger in eDMFT due to the dynamic correlations incorporated in the strong frequency dependence of the selfenergy, which further improves the agreement with ARPES experiments. This indicates that the umklapp processes at higher order in perturbation theory, beyond GW, seem to have a significant effect even in these relatively weakly interacting solids. The relative error in eDMFT varies between 0.5 to 11% (Fig. 4b) across the compounds studied here. While within DFT + U, the static analog of LDA + eDMFT, the bandwidth is insensitive to the value of U (not shown here), in the LDA + eDMFT results here, the bandwidth does depend on the local Coulomb interaction U (see the Supplementary Information).
Discussion
Here, we have systematically studied the electronic structure of elemental metals from the first and second columns of the periodic table using LDA and various beyondDFT methods, such as the mBJ metaGGA and B3LYP and YSPBE0 hybrid functionals, GW, and LDA + eDMFT (eDMFT), and compared with ARPES experiments where available. We have found that B3LYP significantly overestimates the bandwidths, even compared to LDA and mBJ, which are close to each other. G_{0}W_{0} reduces the bandwidths further, but still not sufficiently, while eDMFT narrows the bands the most, producing spectral functions that match rather well with the ARPES experiments. These trends were found to follow almost uniformly over the elemental metallic family of compounds studied here.
It is well known that the dispersion relation for the HartreeFock (HF) equations applied to the ground state of the uniform electron gas or the jellium introduces a logarithmic divergence of the derivative of the band energy at the Fermi level^{59,60}. This gives an infinite velocity and zero density of states at the Fermi energy for the UEG. Manybody correlation effects screen the Coulomb potential and eliminate the divergence.
Since hybrid functionals contain a portion of HartreeFock exchange, it is expected to perform poorly in the context of metals, which is evident from our computations for two different hybrid functionals, namely B3LYP and YSPBE0 (similar to HSE06). In particular, since B3LYP contains the bare (unscreened) exactexchange potential, the comparison with experiment is worse than with YSPBE0 or HSE06 where the interaction is screened. The fact that YSPBE0 improves only slightly over B3LYP, suggests that screening in metals is far stronger than assumed in the standard hybrid functionals, and from comparing to other methods, we can conclude that the dynamic umklapp processes rather than static nonlocal (to an atom) effects dominate.
On the other hand, metaGGAs such as the mBJ method, which is again popular for insulating or semiconducting materials, can be seen as a kind of “hybrid” potential whose amount of “exact exchange” is controlled by a parameter that controls the separation between short and longrange exchange in the screened hybrid functional^{61}. As a result, nonlocality in this method is restricted, and it performs much better than hybrid functionals for the simple metals. The performance of DFT was known to improve if we go up in the famous “Jacob’s ladder” for the exchangecorrelation energy^{62}. Here we find an opposite trend for simple metals, where metaGGA performs much better than hybrid functionals (screened or unscreened), and the difference between LDA and mBJ is not significant. It is worth mentioning that the metaGGAs, in particular the mBJ method, which recovers the LDA bandwidth for metallic system, is computationally as expensive as LDA and considerably cheaper than either hybrid functionals or GW method. Whether this method is suitable for metals was not clear, as it overestimates magnetic moments in ferromagnetic metals^{63}. Here we demonstrate that the mBJ method is applicable to nonmagnetic simple metallic systems, especially for the heavy alkali or alkalinemetals where it performs better than LDA.
In LDA, mBJ, or hybrid functionals the correlation effect is static. It is well known that the dynamic nature of correlation incorporated in GW, which also contains nonlocal selfenergy, can overcome the barrier of failure of conventionalDFT methods. Here we find that the bandwidths computed in GW can improve over LDA, but not to the extent needed to reach agreement with the experimental bandwidths.
It is intriguing to understand why the bands in LDA and G_{0}W_{0} have almost identical slopes near the Fermi energy for almost all elemental metals. As obtained from Landau Fermi liquid theory, the expressions for the difference in band mass in LDA and G_{0}W_{0}, are^{19,64}
where V_{xc}, Σ, and Z_{k} are the exchangecorrelation potential in DFT, GW selfenergy, and quasiparticle renormalization amplitude in GW, respectively. m is the bare band mass, and k_{F} is the Fermi wavevector. Since most of the metals studied here have very similar bandwidths in LDA and GW, we can infer from the above equations that the changes coming from Z and ∂Σ/∂k, largely cancel each other to give similar slopes in LDA and GW near the Fermi energy. Indeed, V_{xc} is local in LDA, hence we can conclude that ∂Σ/∂k ≈ (1/Z_{k} − 1)k_{F}/m is positive and it reduces the mass, while Z_{k} < 1 always increases the mass.
In LDA+eDMFT the mass is renormalized only by Z, as the selfenergy has no momentum dependence. Hence, the cancellation between the momentum and frequency derivative does not occur, and the band narrowing is stronger. Given the much better agreement of eDMFT with the ARPES, this would suggest that the momentum dependence of the selfenergy in this moderately correlated regime might be overestimated by singleshot G_{0}W_{0}, and that a better nonperturbative treatment of correlation effects should increase the frequency dependence and reduce the momentum dependence of the selfenergy. Perhaps the selfconsistent GWΓ calculation with vertex corrections can solve this problem. Based on our study, however, we cannot exclude the possibility that the conventional interpretation of the ARPES experiments is invalid, and that the experimental narrowing of the bands is an artifact of finalstate interactions^{39} or surface effects^{28,49}, opening up new possibilities and challenges for ARPES experiments in future^{48}. A recent ARPES measurement, which tries to eliminate some of these effects in the experiment, agrees with LDA + eDMFT more strongly than the earlier experiment and reconfirms which predictive method is powerful for simple metals. Our work also shows that the umklapp contributions to the electron selfenergy at higher order in perturbation theory (beyond GW) are important and have a significant effect in alkali and alkalineearth metals.
Our study suggests that the elemental metals with partially occupied s orbitals are better described in terms of shortrange dynamical correlations (local to an atom rather than to a point in 3D space) with predominantly large momentum umklapp contributions than in the weakly interacting perturbative picture, or in nonlocal hybrid schemes that incorporate purely local correlations and nonlocal HartreeFock exchange. Such a shortrange dynamical description was already very successful for strongly correlated systems with partially filled d or f bands. Here, we conclude that it is also remarkably successful in describing moderately correlated simple metals as well.
Methods
In this work we have used the full potential linear augmented plane wave (LAPW) method, as implemented in the WIEN2k^{65} software and its extensions. The following DFT functionals in WIEN2k software are used: LDA/GGA, BeckeJohnson (mBJ) potential^{61} for metaGGA, and B3LYP^{53,66} for hybrid functionals. The GW extensions of WIEN2k was implemented in PyGW software^{52} following the Hedin’s GW formalism, and embedded dynamical meanfield method (eDMFT) functional is implemented in LDA+eDMFT software^{67}. For G_{0}W_{0}, mBJ, and B3LYP hybrid functionals (B3LYP and YSPBE0) we construct the initial wavefunction and eigenvalues with the LDA functional as a starting point, and then fully selfconsistent calculations were achieved in all but G_{0}W_{0}, which requires a single shot beyond LDA. For YSPBE0, which is very similar to HSE06 functional, we use a screening parameter of 0.165. In B3LYP there is no screening parameter. The standard DFT computations with the hybrid functionals were first performed selfconsistently, and then the band structures were obtained non selfconsistently. The interpolation methods, as implemented in WIEN2k, assume smooth energy bands in hybrid functionals to avoid any divergence. Thus, the anomalous band dispersion for B3LYP is not seen in the band structures.
LDA + eDMFT
In this work, we use the embedded implementation of the allelectron DFT+DMFT method, where the selfenergy is approximated by a quantity local to the atom in the unit cell. The nonperturbative form of such selfenergy is obtained by solving the quantum impurity problem in the presence of a selfconsistent meanfield environment. This work uses the fully selfconsistent DFTDMFT implementation developed at Rutgers by one of the coauthors. The quantum impurity method is solved by the continuoustime quantum Monte Carlo method^{68}. The selfenergy for the sorbital is embedded by the Dyson equation into the Hilbertspace of all electrons in the solid using projection/embedding technique^{67}. This projection step connects the atomic degrees of freedom with the continuum degrees of freedom of solid, and leads to causal DMFT equations, which can capture spectral weight of the all electrons in the solid. Unlike DFT functional, the DFT+DMFT functional is based on the BaymKadanoff functional, which delivers not only the ground state, but also the excitedstate properties of solids. The eDMFT implementation allows one to write down the total energy in terms of the stationary form of the energy functional. Unlike many other DFT+DMFT implementations, we do not downfold to Wannier orbitals, but rather use projectors to the very localized orbitals contained within the muffintin spheres. The purpose of the projection, which is done on correlated orbitals within Fermienergy ±10 eV, is to extract the local Green’s function from the full Green’s function. For the case of simple metals, we have considered s orbitals to be correlated. We have also tested this with including the porbitals. In this work LDA+ embeddedDMFT(LDA+eDMFT) method^{67} means that we used the combination of DMFT and the LDA functional in the LAPW basis set as implemented in WIEN2k^{65}. Using the exact double counting between LDA and eDMFT^{69}, we obtain the selfenergy on Matsubara frequency, which is then analytically continued with the maximum entropy method from the imaginary to the real axis, continuing the local cumulant function, to obtain the partial density of states and the spectral functions. In eDMFT, where all such higherorder Feynman diagrams are explicitly calculated by the impurity solver, the amount of the screening by the degrees of freedom not included in the method, is substantially reduced, and consequently the values of U are larger and more universal across similar set of material, and are quite successfully predicted by the selfconsistent constrained method, similar to the constrained LDA, but using the LDA + eDMFT functional to evaluate total energies. A fine kpoint mesh of at least 15 × 15 × 15 kpoints in MonkhorstPack kpoint grid and a total of 20 million Monte Carlo steps for each iteration are used for the elemental metals at T = 300 K. To avoid fine tuning parameters, the Coulomb interaction U and Hund’s coupling J_{H} are fixed at 5.0 and 0.3 eV respectively. To estimate these values of the Coulomb repulsion, we used the constrainedDMFT method, in which one computes the total energy in a supercell for N − 1, N, and N + 1 constrained electrons, where N is the number of constrained electrons in the localized orbital. Then we take the difference of energies E(N + 1) − E(N) and E(N) − E(N − 1) to compute the value of U, just like in constrained LDA. The computed values of U for the presented metallic systems within the selfconsistent constrainedeDMFT method are close to U = 5 eV, a value which we adopt for all the compounds studied here. Thus, U is not treated as a tuning parameter. The computed values of U do not vary much among these metallic compounds. We nevertheless studied the dependence of the spectral function on U as shown in Fig. S1. for Na. Surprisingly, a similar value of U was obtained for strongly correlated metals like FeSe, and using U = 5 eV, the spectral functions matched remarkably well with experiments in various Fechalcogenides and pnictide compounds^{70}.
GW
We perform singleshot GW (G_{0}W_{0}) using PyGW software package^{52}, an allelectron GW implementation, where GW selfenergy is computed within the allelectron LAPW basis of WIEN2K. In this newly developed GW code, special attention is paid to the metallic systems and proper treatment of deep laying core states. Prior to GW calculation, we perform LDA calculation to obtain the input singleparticle KohnSham wavefunctions (ψ_{k,i}) as well as the eigenvalues (\({\varepsilon }_{{{{\bf{k}}}}}^{0}\)) in the LAPW basis. In the LAPW method, the space is partitioned into nonoverlapping atomcentered muffin tin (MT) spheres and the interstitial region (IR). We included core electrons in our GW computations. The radius of MT for the elemental metals were considered to be 2.5 Bohr. A 12 Ry cutoff energy for the plane wave was used and RMT*KMAX was fixed at 7 atomic unit. The number of core states and number of the empty bands used for construction of polarizability matrix are described in Supplementary Table II in the Supplementary Information for all the compounds studied here. In the G_{0}W_{0} method, we obtain the screened interaction W_{0} from the Polarization P, which is computed by the convolution of the two singleparticle KohnSham Green’s functions \({G}_{0}\,=\,1/(\omega \,+\,\mu \,\,{\varepsilon }_{{{{\bf{k}}}}}^{0})\) in the form P = −iG_{0}*G_{0}. Here μ is the chemical potential and ω is the frequency. Polarization P and the screened interaction W are expressed in the product basis^{71}, which is further rotated to diagonalize the Coulomb interaction matrix. This product basis allows one to express P, W and the Coulomb interaction in the matrix form. The product basis^{71} is an orthogonal overcomplete basis that can faithfully represent products of two KohnSham orbitals, \({\xi }_{\alpha }^{{{{\bf{q}}}}}({{{\bf{r}}}})\). Here r stands for the real space vector, and q is momentum in the first Brillouin zone. In this implementation within the LAPW basis, the product basis functions \(\left{\xi }_{\alpha }\right\rangle\) are not only orthonormal in the muffintin part, they are also made orthonormal in the interstitial part, unlike in most other implementations^{71,72,73}. More technical details are available in ref. ^{52}.
Once such product basis \({\xi }_{\alpha }^{{{{\bf{q}}}}}({{{\bf{r}}}})\) is constructed, we compute the matrix elements between two KohnSham orbitals and this basis functions: \({M}_{\alpha ,ij}({{{\bf{k}}}},{{{\bf{q}}}})\,\equiv\, \langle {\xi }_{\alpha }^{{{{\bf{q}}}}} {\psi }_{{{{\bf{k}}}},i}{\psi }_{{{{\bf{k}}}}{{{\bf{q}}}},j}^{* }\rangle\) as well as M\({}_{\beta ,i^{\prime} j^{\prime} }^{* }({{{\bf{k}}}}^{\prime} ,{{{\bf{q}}}})\,\equiv\,\)\(\langle {\psi }_{{{{\bf{k}}}}^{\prime} ,i^{\prime} }{\psi }_{{{{\bf{k}}}}^{\prime} {{{\bf{q}}}},j^{\prime} }^{* } {\xi }_{\beta }^{{{{\bf{q}}}}}\rangle\), where ψ_{k,i}, \({\psi }_{{{{\bf{k}}}}^{\prime} {{{\bf{q}}}},j^{\prime} }\) are incoming, and \({\psi }_{{{{\bf{k}}}}{{{\bf{q}}}},j}^{* }\), \({\psi }_{{{{\bf{k}}}}^{\prime} ,i^{\prime} }^{* }\) are outgoing electrons. Similarly, we compute the matrix elements of the Coulomb repulsion in this basis \({v}_{\alpha \beta }({{{\bf{q}}}})\,=\,\langle {\xi }_{\alpha }^{{{{\bf{q}}}}} {V}_{C}({{{\bf{q}}}}) {\xi }_{\beta }^{{{{\bf{q}}}}}\rangle\), and then compute the square root of the Coulomb repulsion in its eigenbasis \({\sqrt{v({{{\bf{q}}}})}}_{\alpha ,\beta }\,=\,{T}_{\alpha ,l}\sqrt{{v}_{l}}\,{T}_{l,\beta }^{{\dagger} }\), where v_{l} are eigenvalues and T_{α,l} are eigenvectors of the Coulomb repulsion.
The dielectric function in matrix form is computed as \(\varepsilon \,=\,1\,\,\sqrt{{V}_{C}}P\sqrt{{V}_{C}}\). The advantage of the product basis is elaborately described in refs. ^{52,72}. Once the dielectric matrix ε is calculated, we invert it in this eigenbasis of the Coulomb repulsion to compute the screened interaction W. The dynamical correlation selfenergy within GW approximation is then obtained by convolution of W and the singleparticle KohnSham Green’s function G_{0}. In PyGW code, the dynamic part of the screened interaction is computed on the Matsubara axis, which performs much better than classical plasmonpole method^{74}. For the analytic continuation to the real frequency axis we used standard Pade method^{75,76}, which is accurate at low to moderate frequencies, provided we have very accurate imaginaryaxis data. This was achieved in our calculations up to 5 eV, which is sufficient to plot reliable band structures of these metallic materials. The Pade approximation is forced to go exactly through all Matsubara frequencies calculated on logarithmic mesh (between 32 and 64), hence the number of poles in such analytic function is large (between 30 and 62) since for metals, a few pole approximation in Padetype fitting is usually not sufficient. We then carefully converge our results using a very fine kpoint mesh as the convergence for metallic systems can be difficult in GW method. Here, even though we use tetrahedron analytic integration over momentum points, we find that a large number of momentum points is necessary for convergence. For example, although the 4 × 4 × 4 grid gives approximate spectra not too different from LDA, the convergence is reached only at 16 × 16 × 16 momentum mesh. We have used the interpolation method from refs. ^{77,78} for computing spectra along the highsymmetry lines. The input data at momentum points in the calculation are of high quality, but the smooth interpolation at other momentum points is not very straightforward in the delocalized basis sets, especially for interstitial charge and such interpolation is not extremely precise at points where cusps occur in the band structure, and sometimes leads to small computational artifacts near Fermi surface singularity. For this reason, we notice the GW bands have some dispersion anomalies close to the Fermi energy in some of the computed spectra. The Fourier transformation of the selfenergy from momentum space to the real space, and its evaluation at the generic momentum point along the momentum path, will improve on the current simple interpolating scheme. More sophisticated and more accurate Wannier interpolation might also ameliorate this issue and is beyond the scope of this work. For more technical details about methodology, implementation of GW in LAPW basis as well as the Kpoint convergence, we refer the reader to ref. ^{52}.
Note added in proof
While this paper was under review, we became aware of a recent ARPES experiment^{48}, that reconfirms our prediction with LDA + eDMFT for Na.
Data availability
The data that support the findings of this study are available from the corresponding authors upon request. In the future, the data will be available in the form of an opensource database that is currently under construction.
Code availability
We have used the WIEN2K^{65} DFT + eDMFT, and PyGW codes to generate the data. WIEN2K is a commercially available software package. LDA + eDMFT and PyGW are freely available and can be obtained from http://hauleweb.rutgers.edu/tutorials and https://github.com/ruccmt/PyGW respectively.
References
Onida, G., Reining, L. & Rubio, A. Electronic excitations: densityfunctional versus manybody Green’sfunction approaches. Rev. Mod. Phys. 74, 601 (2002).
Georges, A., Kotliar, G., Krauth, W. & Rozenberg, M. J. Dynamical meanfield theory of strongly correlated fermion systems and the limit of infinite dimensions. Rev. Mod. Phys. 68, 13–125 (1996).
Kotliar, G. et al. Electronic structure calculations with dynamical meanfield theory. Rev. Mod. Phys. 78, 865–951 (2006).
Choudhary, K. et al. Computational screening of highperformance optoelectronic materials using OptB88vdW and TBmBJ formalisms. Sci. Data 5, 180082 (2018).
Stevanović, V., Lany, S., Zhang, X. & Zunger, A. Correcting density functional theory for accurate predictions of compound enthalpies of formation: fitted elementalphase reference energies. Phys. Rev. B 85, 115104 (2012).
Li, W., Walther, C. F. J., Kuc, A. & Heine, T. Density functional theory and beyond for bandgap screening: performance for transitionmetal oxides and dichalcogenides. J. Chem. Theory Comput. 9, 2950–2958 (2013).
Hybertsen, M. S. & Louie, S. G. Electron correlation in semiconductors and insulators: band gaps and quasiparticle energies. Phys. Rev. B 34, 5390–5413 (1986).
Larson, P., Dvorak, M. & Wu, Z. Role of the plasmonpole model in the GW approximation. Phys. Rev. B 88, 125205 (2013).
Biermann, S., Poteryaev, A., Lichtenstein, A. I. & Georges, A. Dynamical singlets and correlationassisted peierls transition in VO_{2}. Phys. Rev. Lett. 94, 026404 (2005).
Kuneš, J. et al. Collapse of magnetic moment drives the Mott transition in MnO. Nat. Mat. 7, 198–202 (2008).
Park, H., Millis, A. J. & Marianetti, C. A. Siteselective mott transition in rareearthelement nickelates. Phys. Rev. Lett. 109, 156402 (2012).
Mandal, S., Zhang, P., IsmailBeigi, S. & Haule, K. How correlated is the FeSe/SrTiO_{3} system? Phys. Rev. Lett. 119, 067004 (2017).
Ferber, J., Foyevtsova, K., Valentí, R. & Jeschke, H. O. LDA + DMFT study of the effects of correlation in lifeas. Phys. Rev. B 85, 094505 (2012).
Bisti, F. et al. Weaklycorrelated nature of ferromagnetism in nonsymmorphic CrO_{2} revealed by bulksensitive softXray arpes. Phys. Rev. X 7, 041067 (2017).
Taranto, C. et al. Comparing quasiparticle GW + DMFT and LDA + DMFT for the test bed material SrVO_{3}. Phys. Rev. B 88, 165119 (2013).
Marini, A., Onida, G. & Del Sole, R. Quasiparticle electronic structure of copper in the GW approximation. Phys. Rev. Lett. 88, 016403 (2001).
Bernardi, M., Mustafa, J., Neaton, J. B. & Louie, S. G. Theory and computation of hot carriers generated by surface plasmon polaritons in noble metals. Nat. Commun. 6, 7044 (2015).
Mandal, S., Haule, K., Rabe, K. M. & Vanderbilt, D. Systematic beyonddft study of binary transition metal oxides. npj Comput. Mater. 5, 115 (2019).
Mahan, G. D. Manyparticle physics (Plenum Press,1981).
Lyo, I.W. & Plummer, E. W. Quasiparticle band structure of na and simple metals. Phys. Rev. Lett. 60, 1558–1561 (1988).
Jensen, E. et al. Angleresolved photoemission study of the electronic structure of beryllium: Bulk band dispersions and manyelectron effects. Phys. Rev. B 30, 5500–5507 (1984).
Bartynski, R. A., Gaylord, R. H., Gustafsson, T. & Plummer, E. W. Angleresolved photoemission study of the surface and bulk electronic structure of Mg(0001) and Mg(112^{−}0). Phys. Rev. B 33, 3644–3656 (1986).
Itchkawitz, B. S., Lyo, I.W. & Plummer, E. W. Experimental band structure of potassium as measured by angleresolved photoemission. Phys. Rev. B 41, 8075–8084 (1990).
Sashin, V. A., Bolorizadeh, M. A., Kheifets, A. S. & Ford, M. J. Electronic band structure of metallic calcium measured by electron momentum spectroscopy. J. Condens. Matter Phys. 12, 9407–9423 (2000).
Shek, M.L., Hrbek, J., Sham, T. K. & Xu, G.Q. Corelevel photoemission from alkali metals on Ru(001). Phys. Rev. B 41, 3447–3454 (1990).
Elatresh, S. F. et al. Fermi surface studies of the lowtemperature structure of sodium. Phys. Rev. B 101, 220103 (2020).
Mahan, G. D. & Sernelius, B. E. Electronelectron interactions and the bandwidth of metals. Phys. Rev. Lett. 62, 2718–2720 (1989).
Shung, K. W. K., Sernelius, B. E. & Mahan, G. D. Selfenergy corrections in photoemission of Na. Phys. Rev. B 36, 4499–4502 (1987).
Ku, W., Eguiluz, A. G. & Plummer, E. W. Comment on “why is the bandwidth of sodium observed to be narrower in photoemission experiments?”. Phys. Rev. Lett. 85, 2410–2410 (2000).
Vogt, M., Zimmermann, R. & Needs, R. J. Spectral moments in the homogeneous electron gas. Phys. Rev. B 69, 045113 (2004).
Campillo, I., Pitarke, J. M., Rubio, A., Zarate, E. & Echenique, P. M. Inelastic lifetimes of hot electrons in real metals. Phys. Rev. Lett. 83, 2230–2233 (1999).
Maezono, R., Towler, M. D., Lee, Y. & Needs, R. J. Quantum Monte Carlo study of sodium. Phys. Rev. B 68, 165103 (2003).
Higuchi, M., Yoshinaga, S. & Yasuhara, H. Effects of electronelectron interactions on the bandwidth of simple metals. J. Phys. Soc. Jpn. 68, 3473–3476 (1999).
Craco, L. & Leoni, S. LDA + DMFT approach to electronic structure of sodium metal. Phys. Rev. B 100, 115156 (2019).
Nilsson, F., Boehnke, L., Werner, P. & Aryasetiawan, F. Multitier selfconsistent GW + EDMFT. Phys. Rev. Mater. 1, 043803 (2017).
Kutepov, A. L. Electronic structure of Na, K, Si, and LiF from selfconsistent solution of hedin’s equations including vertex corrections. Phys. Rev. B 94, 155101 (2016).
Northrup, J. E., Hybertsen, M. S. & Louie, S. G. Quasiparticle excitation spectrum for nearlyfreeelectron metals. Phys. Rev. B 39, 8198–8208 (1989).
Lischner, J., Bazhirov, T., MacDonald, A. H., Cohen, M. L. & Louie, S. G. Effect of spin fluctuations on quasiparticle excitations: firstprinciples theory and application to sodium and lithium. Phys. Rev. B 89, 081108 (2014).
Yasuhara, H., Yoshinaga, S. & Higuchi, M. Why is the bandwidth of sodium observed to be narrower in photoemission experiments? Phys. Rev. Lett. 83, 3250–3253 (1999).
Yasuhara, H., Yoshinaga, S. & Higuchi, M. Yasuhara, Yoshinaga, and Higuchi reply. Phys. Rev. Lett. 85, 2411–2411 (2000).
Hedin, L. New method for calculating the OneParticle green’s function with application to the ElectronGas problem. Phys. Rev. 139, A796 (1965).
Holm, B. & von Barth, U. Fully selfconsistent GW selfenergy of the electron gas. Phys. Rev. B 57, 2108–2117 (1998).
Cazzaniga, M. gw and beyond approaches to quasiparticle properties in metals. Phys. Rev. B 86, 035120 (2012).
Okumura, H., Sato, K. & Kotani, T. Spinwave dispersion of 3d ferromagnets based on quasiparticle selfconsistent GW calculations. Phys. Rev. B 100, 054419 (2019).
Holzmann, M. et al. Momentum distribution of the homogeneous electron gas. Phys. Rev. Lett. 107, 110402 (2011).
Maezono, R., Towler, M. D., Lee, Y. & Needs, R. J. Quantum Monte Carlo study of sodium. Phys. Rev. B 68, 165103 (2003).
Zhu, X. & Overhauser, A. W. Plasmonpole and paramagnonpole model of an electron liquid. Phys. Rev. B 33, 925–936 (1986).
Potorochin, D. V. et al. Lifetime of quasiparticles in the nearlyfree electron metal sodium. Preprint at https://arxiv.org/pdf/2112.00422.pdf (2021).
Shung, K. W. K. & Mahan, G. D. Calculated photoemission spectra of Na. Phys. Rev. Lett. 57, 1076–1079 (1986).
Damascelli, A., Hussain, Z. & Shen, Z.X. Angleresolved photoemission studies of the cuprate superconductors. Rev. Mod. Phys. 75, 473–541 (2003).
Choudhary, K. et al. The joint automated repository for various integrated simulations (JARVIS) for datadriven materials design. npj Comput. Mater. 6, 173 (2020).
Haule, K. & Mandal, S. All electron gw with linearized augmented plane waves for metals and semiconductors. Preprint at https://arxiv.org/pdf/2008.07727.pdf (2020).
Tran, F., Blaha, P., Schwarz, K. & Novák, P. Hybrid exchangecorrelation energy functionals for strongly correlated electrons: applications to transitionmetal monoxides. Phys. Rev. B 74, 155108 (2006).
Heyd, J., Scuseria, G. E. & Ernzerhof, M. Hybrid functionals based on a screened coulomb potential. J. Chem. Phys. 118, 8207–8215 (2003).
Perdew, J. P., Ernzerhof, M. & Burke, K. Rationale for mixing exact exchange with density functional approximations. J. Chem. Phys. 105, 9982–9985 (1996).
Lee, C., Yang, W. & Parr, R. G. Development of the ColleSalvetti correlationenergy formula into a functional of the electron density. Phys. Rev. B 37, 785–789 (1988).
Paier, J., Marsman, M. & Kresse, G. Why does the B3LYP hybrid functional fail for metals? J. Chem. Phys. 127, 024103 (2007).
Kutepov, A., Oudovenko, V. & Kotliar, G. Linearized selfconsistent quasiparticle GW method: application to semiconductors and simple metals. Comput. Phys. Commun. 219, 407–414 (2017).
Ashcroft, N. W. & Mermin, N. D. Solid State Physics (Saunders College Publishing, Harcourt, Inc., 1976).
Blair, A. I., Kroukis, A. & Gidopoulos, N. I. A correction for the HartreeFock density of states for jellium without screening. J. Chem. Phys. 142, 084116 (2015).
Tran, F. & Blaha, P. Accurate band gaps of semiconductors and insulators with a semilocal exchangecorrelation potential. Phys. Rev. Lett. 102, 226401 (2009).
Perdew, J. P. & Schmidt, K. Jacob’s ladder of density functional approximations for the exchangecorrelation energy. AIP Conf. Proc. 577, 1–20 (2001).
Koller, D., Tran, F. & Blaha, P. Merits and limits of the modified BeckeJohnson exchange potential. Phys. Rev. B 83, 195134 (2011).
Haule, K. & Chen, K. Singleparticle excitations in the uniform electron gas by diagrammatic Monte Carlo. Sci. Rep. 12, 2294 (2022).
Blaha, P., Schwarz, K., Madsen, G., Kvasnicka, D. & Luitz, J. An augmented plane wave plus local orbitals program for calculating crystal properties (Vienna University of Technology, 2001).
Becke, A. D. Densityâfunctional thermochemistry. III. the role of exact exchange. J. Chem. Phys. 98, 5648–5652 (1993).
Haule, K., Yee, C.H. & Kim, K. Dynamical meanfield theory within the fullpotential methods: electronic structure of CeIrIn_{5}, CeCoIn_{5}, and CeRhIn_{5}. Phys. Rev. B 81, 195107 (2010).
Haule, K. Quantum monte carlo impurity solver for cluster dynamical meanfield theory and electronic structure calculations with adjustable cluster base. Phys. Rev. B 75, 155113 (2007).
Haule, K. Exact double counting in combining the dynamical mean field theory and the density functional theory. Phys. Rev. Lett. 115, 196403 (2015).
Yin, Z. P., Haule, K. & Kotliar, G. Kinetic frustration and the nature of the magnetic and paramagnetic states in iron pnictides and ironÂ chalcogenides. Nat. Mater. 10, 932 (2011).
Aryasetiawan, F. & Gunnarsson, O. Productbasis method for calculating dielectric matrices. Phys. Rev. B 49, 16214–16222 (1994).
Friedrich, C., Blügel, S. & Schindlmayr, A. Efficient implementation of the GW approximation within the allelectron flapw method. Phys. Rev. B 81, 125102 (2010).
Jiang, H. et al. FHIgap: a GW code based on the allelectron augmented plane wave method. Comput. Phys. Commun. 184, 348–366 (2013).
Engel, G. E. & Farid, B. Generalized plasmonpole model and plasmon band structures of crystals. Phys. Rev. B 47, 15931–15934 (1993).
Beach, K. S. D., Gooding, R. J. & Marsiglio, F. Reliable padé analytical continuation method based on a highaccuracy symbolic computation algorithm. Phys. Rev. B 61, 5147–5157 (2000).
Baker, G. A. J. Essentials of Padé Approximants (Academic Press, 1975).
Pickett, W. E., Krakauer, H. & Allen, P. B. Smooth fourier interpolation of periodic functions. Phys. Rev. B 38, 2721–2726 (1988).
Koelling, D. & Wood, J. On the interpolation of eigenvalues and a resultant integration scheme. J. Comput. Phys. 67, 253–262 (1986).
Acknowledgements
This research was funded by NSF DMREF DMR1629059 and NSF DMREF DMR1629346. The computations were performed at the Frontera supercomputer at the Texas Advanced Computing Center (TACC) at The University of Texas at Austin, which is supported by National Science Foundation grant number OAC1818253 and at the Extreme Science and Engineering Discovery Environment (XSEDE), which is supported by National Science Foundation grant number ACI1548562.
Author information
Authors and Affiliations
Contributions
S.M. carried out the calculations. All authors discussed the results and cowrote the paper.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Mandal, S., Haule, K., Rabe, K.M. et al. Electronic correlation in nearly free electron metals with beyondDFT methods. npj Comput Mater 8, 181 (2022). https://doi.org/10.1038/s41524022008678
Received:
Accepted:
Published:
DOI: https://doi.org/10.1038/s41524022008678
This article is cited by

The role of electron correlations in the electronic structure of putative Chern magnet TbMn6Sn6
npj Quantum Materials (2023)