PHYSICAL REVIEW B 69, 085104 共2004兲 Weighted-density-approximation description of rare-earth trihydrides Zhigang Wu and R. E. Cohen Geophysical Laboratory, Carnegie Institution of Washington, 5251 Broad Branch Road NW, Washington, DC 20015, USA D. J. Singh Center for Computational Materials Science, Naval Research Laboratory, Washington, DC 20375, USA R. Gupta Commissariat a l’Energie Atomique, Centre d’Etudes de Saclay, Service de Recherches de Métallurgie Physique Commissariat à l’Energie Atomique, 91191 Gif Sur Yvette Cedex, France M. Gupta Institut des Sciences des Materiaux, Batiment 415, Universite Paris-Sud, 91405 Orsay, France 共Received 24 October 2003; published 19 February 2004兲 Density-functional calculations within the weighted density approximation 共WDA兲 are presented for YH3 and LaH3 . We investigate some commonly used pair-distribution functions G. These calculations show that within a consistent density-functional framework a substantial insulating gap can be obtained while at the same time retaining structural properties in accord with experimental data. Our WDA band structures agree with those of GW approximation very well. It shows that there is no strong correlation in H 1s states, and the absence of self-interaction in H 1s is crucial to obtain correct band structures of these rare-earth trihydrites. But the calculated band gaps are still 1.0–2.0 eV smaller than experimental findings. DOI: 10.1103/PhysRevB.69.085104 PACS number共s兲: 71.15.Mb, 71.30.⫹h, 78.20.Bh I. INTRODUCTION The rare-earth trihydrides, M H3 , where M ⫽Y, La–Lu, are the subject of considerable practical and theoretical interests. The rare earths absorb hydrogen readily and in the trihydride form have among the largest hydrogen to metal ratios of the elemental hydrides.1 The fully hydrogenated materials are insulating, while those with less hydrogen are metallic, and in thin film form they can be used as switchable mirrors.2 It is important to demonstrate a correct description of these trihydrides in first-principles approaches in order to validate calculations for more complex metal hydrogen systems. However, standard first-principles calculations based on density-functional theory 共DFT兲 within the local-density 共LDA兲 and generalized gradient 共GGA兲 approximations incorrectly predict metallic behavior for the rare-earth trihydrides. Here we explore application of the weighted density approximation 共WDA兲. We find a WDA that works for these materials 共and others兲, indicating that the gap problem here is not due to unusual correlations or quasiparticle corrections, but is simply a problem with the LDA and GGA exchange correlation 共xc兲 functionals. The rare-earth dihydrides, M H2 , are metallic. The trihydrides on the other hand are insulating with optical gaps of the order of 2 eV, with a metal insulator transition at a hydrogen content correlated with rare-earth ionic radius.2 This insulating gap of the trihydrides may be understood in terms of simple picture, in which H forms a negative ion, H⫺ , fully ionizing the metal atoms, thus forming a gap between H bands and higher lying metal sd conduction bands. This simple picture is supported by infrared and photoemission measurements.3,4 It is consistent with the fact that all the rare-earth trihydrides are insulators despite differences in the 0163-1829/2004/69共8兲/085104共6兲/$22.50 details of their crystal structures. These range from the f cc derived structure of LaH3 to the not fully solved hexagonal structure of YH3 . 5– 8 Computations within the LDA predict a metallic band structure with substantial overlap between H and metal derived bands.8 –11 This severe gap problem could be due to several possible causes: 共1兲 use of an incorrect crystal structure; 共2兲 an extreme case of LDA error in predicting gaps; 共3兲 unusual correlation effects; or 共4兲 self-interaction errors in LDA. Early attempts at resolving this discrepancy focused on crystal structure. Indeed it was shown that with optimized crystal structures, DFT calculations could yield insulating band structures for YH3 . 12,13 Unfortunately, these relaxed structures are not consistent with diffraction measurements, and such an explanation would not explain the similar insulating properties of the other M H3 compounds. DFT is a theory of ground-state properties, and the eigenvalues of the Kohn-Sham equations cannot be equated to quasiparticle excitations.14 Nevertheless, any nonpathological accurate approximation to the exact DFT should predict an insulating Kohn-Sham band structure, because static responses and other ground-state properties depend on the presence or absence of a gap. For example, LO-TO splitting would be present in the phonon spectrum of ionic insulator like YH3 but not in a metal. Recent studies on rare-earth trihydrides have considered electron correlation effects,15–19 particularly electron correlations on the H sites. However, since there are no partially occupied d or f orbitals in LaH3 and YH3 , nor is there evidence that these materials are near a quantum critical point, these hydrides represent a novel class of strong correlated materials. Calculations done within the GW approximation showing band gaps that are reconcilable with experiment are 69 085104-1 ©2004 The American Physical Society PHYSICAL REVIEW B 69, 085104 共2004兲 WU, COHEN, SINGH, GUPTA, AND GUPTA obtained, both with the cubic BiF3 structure (LaH3 ) and the hexagonal HoD3 structure 共a simplified YH3 structure兲.17–19 However in the GW calculations, the dielectric function is inherently wrong since it comes from the LDA, where the ground state is metallic and not insulating. Also, the GW calculations are very cumbersome, and it would be advantageous to a simple and fast method that gives reasonable results. The GW method approaches the problem as being most simply viewed as due to the treatment of excited states, but if the problem is due to self-interaction, it is fundamentally a ground-state problem. The LDA makes large errors in description of the one-electron hydrogen atom because of incomplete cancellation of the Hartree and xc self-interactions. One way to remove the spurious self-interactions in LDA and GGA calculations is the so-called self-interaction correction.20 In this approach an orbital dependent selfinteraction is calculated at each step and used to create an orbital dependent single-particle Hamiltonian. This method suffers from basis set dependence, e.g., the computed correction is zero for a Bloch state representation. We explore whether one can remain within conventional ground-state DFT and obtain the proper electronic structure of rare-earth trihydrides. of G,27 and another G-J type for k⫽4 (G-J4), which was tested by Rushton et al. for silicon.28 One subtle issue in WDA implementations is shell partitioning.23 Although the WDA has the exact form of xc energy, it does not describe the exchange energy between core and valence states very well due to the approximate model G function. Outside core regions, the core density vanishes, so does the core-valence exchange energy. However in the WDA scheme, because the range of integration of G is similar to the size of atoms, and there is no distinction between core and valence electrons, core and valence electrons would dynamically screen valence electrons equally, and it could cause the core-valence exchange energy to be exaggerated. If the exact G function for an arbitrary electron system is known, it will naturally describe intershell interactions correctly. On the other hand, the LDA can give correct intershell contributions, since the LDA depends only on local density. Based on this observation, a shell partitioning approach was proposed23,30 to guarantee the correct behavior of core-valence xc interaction, in which the valence-valence interactions are treated with the WDA, while core-core and core-valence with the LDA. If n v(r) and n c(r) denote valence and core electron density, respectively, the xc energy II. WEIGHTED DENSITY APPROXIMATION LDA WDA LDA E xc关 n 兴 ⫽E xc 关 n 兴 ⫹E xc 关 n v ,n 兴 ⫺E xc 关 n v ,n 兴 , The weighted density approximation 共WDA兲 共Refs. 21– 23兲 has the exact nonlocal form of DFT expression of xc energy, E xc关 n 兴 ⫽ 冕冕 n 共 r兲 n 共 r⬘ 兲 兩 r⫺r⬘ 兩 关 ḡ xc共 r,r⬘ ,n 兲 ⫺1 兴 drdr⬘ . where WDA E xc 关 n v ,n 兴 ⫽ 共2兲 where the weighted density n̄(r) can be determined from the xc hole condition: 冕 n 共 r⬘ 兲 G 关 兩 r⫺r⬘ 兩 ,n̄ 共 r兲兴 dr⬘ ⫽⫺1. 冕冕 n v共 r兲 n v共 r⬘ 兲 兩 r⫺r⬘ 兩 G 关 兩 r⫺r⬘ 兩 ,n̄ 共 r兲兴 drdr⬘ , 共5兲 共1兲 Since the exact pair-distribution function ḡ xc(r,r⬘ ,n) is elusive, a model WDA function G is made, ḡ xc共 r,r⬘ ,n 兲 ⫺1⫽G 关 兩 r⫺r⬘ 兩 ,n̄ 共 r兲兴 , LDA E xc 关 n v ,n 兴 ⫽ 冕冕 n v共 r兲 n v共 r兲 兩 r⫺r⬘ 兩 G 关 兩 r⫺r⬘ 兩 ,n 共 r兲兴 drdr⬘ . 共6兲 The weighted density n̄(r) can be chosen from the following sum rule 冕兵 共3兲 This assures no self-interaction in a one-electron system. The WDA also gives correct results for the uniform electron gas because it reduces to the LDA in this case. However because WDA is incorrect, G WDA(r,r⬘ )⫽G WDA(r⬘ ,r), the form of V xc and it causes wrong asymptotic behavior at large distance away from nucleus.23 Fortunately this error is not substantial when calculations are done at experimental structure. In the WDA, an approximation is made for the function G, a natural choice of G would be of a uniform system, but Mazin and Singh25 emphasized, based on the xc kernel, that G of the uniform electron gas is not optimal. Several model forms of G have been suggested, such as the GunnarssonJones 共G-J兲 共Ref. 24兲 ansatz, G G-J(r,n)⫽c 兵 1 ⫺k ⫺exp(⫺关r/兴 )其, with k⫽5; and the Gritsenko et al. 共GRBA兲 共Ref. 26兲 ansatz, G GRBA(r,n)⫽c•exp(⫺关r/兴k), with k⫽1.5. In this paper, we also used a homogeneous type 共4兲 n v共 r⬘ 兲 G 关 兩 r⫺r⬘ 兩 ,n̄ 共 r兲兴 ⫹n c共 r兲 G 关 兩 r⫺r⬘ 兩 ,n 共 r兲兴 其 dr⬘ ⫽⫺1. 共7兲 The xc potential for core and valence states are different in that the xc energy depends on core and valence electron density explicitly. The derivation is attached in the appendix. An accurate potential makes the calculations fully consistent, so that atomic forces can be determined by the HellmannFeynmann theory. We found that the calculated forces within the WDA agree with numerical results of finite energy difference very well. III. RESULTS AND DISCUSSION A. Calculation details The calculations were done using a plane-wave basis with pseudopotential method. Pseudopotentials were generated by Troullier and Martins29 scheme. We treated semicore states 085104-2 PHYSICAL REVIEW B 69, 085104 共2004兲 WEIGHTED-DENSITY-APPROXIMATION DESCRIPTION . . . TABLE I. With the calculated atomic positions and fixed c/a ratio, the optimized equilibrium volumes 共in Å3 ) of hexagonal YH3 (Y2 H6 ). Here a, b, c, and d denote uniform, GRBA, G-J, and G-J4 forms of G, respectively. Numbers in parentheses are the percentage deviations from experiment. LDA WDA 共a兲 WDA 共b兲 WDA 共c兲 WDA 共d兲 Expt. 72.8 (⫺6.2) 75.0 (⫺3.4) 77.4 (⫺0.3) 75.6 (⫺2.6) 79.4 (⫹2.3) 77.6 of Y and La with LDA, higher states and H 1s states with WDA, and lower states are pseudized. The ionic configurations and cutoff radii are 4s 2 4p 6 4d 2 and r s,p,d c ⫽1.55,1.55,1.67 Bohr for Y, 5s 2 5p 6 5d 0.5 and r s,p,d c ⫽1.77,1.77,1.77 Bohr for La, and 1s 0.92p 0.1 and r s,p c ⫽1.15,1.15 Bohr for H. Well converged plane-wave basis sets were used; the energy cutoffs are 156 Ry for hexagonal YH3 , 121 Ry for cubic YH3 , and LaH3 . We used a 48⫻48 ⫻84 real space grid and 12 k-point mesh for hexagonal YH3 , and a 48 real space grid and 12⫻12⫻12 k-point mesh for cubic YH3 and LaH3 to obtain the band structure.31 B. Hexagonal YH3 As mentioned before, although the exact structure of hexagonal YH3 is still under investigation, its insulating property is believed to be an electronic property, i.e., not sensitive to the crystal structure.32,33 We studied a simplified hexagonal LaF3 structure with a Y2 H6 unit cell,11 in which Y atoms form a hexagonal close-packed 共hcp兲 structure. The measured lattice constants are a⫽3.672 Å and c/a⫽1.81. We optimized the H positions using LDA and WDA 共with four types of G), and all the results are consistent with the previous LDA result.11 We also optimized the equilibrium volume with fixed c/a ratio. Table I shows that the WDA volumes deviate by only about 2–3 % from experimental data. The calculated bands of the LDA and WDA 共with the G-J4 type of G) are shown in Fig. 1. The LDA predicts a direct band gap of 1.0 eV at ⌫ and an overlap of 0.93 eV between ⌫ and K, which leads to a semimetal. Our present LDA results are a little different from previous results 共0.6 eV for the ⌫ gap and 1.3 eV for the overlap11兲 because we used the Hedin-Lundqvist xc function, whereas the Wigner ansatz was used previously. The WDA 共with the G-J4 type of FIG. 1. LDA and WDA 共with the G-J4 type of G) electron bands of the hexagonal YH3 . TABLE II. Calculated LDA and WDA 共a, b, c, and d as seen in Table I兲 fundamental and ⌫ gaps 共in eV兲, compared with GW results.17 A negative band gap means band overlap. LDA WDA 共a兲 WDA 共b兲 WDA 共c兲 WDA 共d兲 GW fd. gap ⫺0.93 ⌫ gap 1.0 ⫺0.74 1.2 ⫺0.35 1.5 ⫺0.72 1.2 0.41 2.2 0.6 2.9 G) predicts a direct ⌫ gap of 2.2 eV, and opens up a fundamental gap of 0.41 eV between ⌫ and K. Thus, the WDA predicts a substantial insulating gap. Our WDA band gaps are close to the GW calculations17 of 0.6 eV for the fundamental gap and 2.9 eV for the direct ⌫ gap. In the Brillouin zone of the more complicated tripled HoD3 structure of YH3 , this indirect ⌫-K gap will be folded to form a direct gap at ⌫ point, and they are roughly of the same order of magnitude.11,17 Thus, a small gap between ⌫ and K in the simple hexagonal structure makes YH3 a semiconductor even in the complicated HoD3 structure. We also calculated the WDA bands with the other three G functions, and the band gaps are shown in Table II. We found that the WDA with all these G’s gives better band gaps than the LDA. Except for the G-J4 form, the GRBA 共Ref. 26兲 form is superior to the other two, because it predicts a larger direct ⌫ gap and smaller overlap between ⌫ and K. These results illustrate that the WDA band structure is sensitive to the choice of G. We compare these pair-distribution functions by drawing them together in Fig. 2, where n ⫽3/(4 r s3 ), and in the WDA calculations of YH3 , 0.8⬍r s ⬍2.3. We found the G-J4 form has the shortest range, and the range of GRBA ansatz is shorter than the other two. Our calculations indicate that a shorter range G gives larger band gaps. Interestingly, such functions, particularly GRBA, yield static responses closer to Monte Carlo results for the uniform electron gas.25 C. Cubic YH3 and LaH3 We performed LDA and WDA 共with G-J4 only兲 calculations on cubic BiF3 structure of YH3 and LaH3 , in which the Y or La atoms constitute a face-centered cubic 共fcc兲 lattice and H atoms are located at the tetrahedral and octahedral sites. Recently the insulating fcc YH3 was stabilized by existence of MgH2 , 33 but there are large discrepancies in the FIG. 2. Four types of the pair-distribution function G(r,n) with r s ⫽0.8 and r s ⫽2.3. 085104-3 PHYSICAL REVIEW B 69, 085104 共2004兲 WU, COHEN, SINGH, GUPTA, AND GUPTA Although our DFT-WDA band gaps of YH3 and LaH3 agree with GW quasiparticle calculations very well, the measured optical gap is ⬃2.8 eV for YH3 and ⬃1.8 eV for LaH3 , which are still 1–2 eV larger. Gelderen et al.17 argued that the fundamental band gap of hexagonal YH3 is not the measured optical gap due to the forbidden transition between the highest valence and the lowest conduction bands. They concluded that the measured value corresponds to the direct gap at ⌫ of the hexagonal LaF3 structure. However the real hexagonal YH3 is more complex and it may have low symmetry so that the dipole matrix element could may well not be zero. Our DFT-WDA method may underestimate band gaps due to the DFT gap discontinuity at the Fermi level.36 IV. SUMMARY FIG. 3. LDA and WDA 共with the G-J4 type of G) electron bands of the cubic YH3 and LaH3 . measured volumes of fcc YH3 . Ahuja et al.13 expected 32.2 Å3 , while van der Molen et al.33 extrapolated 36.9 Å3 and Gogh et al.32 obtained 38.0 Å3 . Our LDA and WDA equilibrium volumes are 32.4 Å3 and 35.2 Å3 , respectively. We calculated band structure at experimental volume of 36.9 Å3 . For LaH3 the measured volume 44.2 Å3 is close to both LDA and WDA results, and we used this volume to calculate band structure. In Fig. 3 are the calculated band structures of cubic YH3 and LaH3 . The LDA predicts that both of them are metals with direct overlaps of 1.0 eV for YH3 and 0.46 eV for LaH3 at ⌫ point, which agree with previous calculations.19,34 Contrary to the LDA, the WDA predicts that both trihydrides are semiconductors with direct ⌫ gaps of 0.19 eV and 0.80 eV for YH3 and LaH3 , respectively. However the fundamental gaps are about 0.1 eV smaller, because the energy of lowest conduction band at L is slightly lower than that at ⌫. Our WDA band gaps are in good agreement with previous GW calculations, e.g., Alford et al.34 predicted a ⌫ gap of 0.2–0.3 eV for YH3 and 0.8 –0.9 eV for LaH3 , and Chang et al.35 obtained 0.8 eV for LaH3 . We found that the hexagonal and cubic YH3 have similar small gaps, suggesting that the insulating properties are of electronic origin rather than structural. Another interesting feature of the WDA band structures is that compared with the LDA, the overall valence bandwidth hardly changes. It supports a view that rare earth hydrites are not strong correlated systems. This is consistent with previous GW calculations,17,19,34 but in contrast to the many-body model Hamiltonian calculations which show a large valence bandwidth decrease.15 Thus, our results support the previous conclusion17,34 that rare-earth trihydrides are just band insulators, and that strong correlations are not required. In conclusion, we have studied the band structures of the hexagonal YH3 and cubic YH3 and LaH3 using the firstprinciples WDA method. The calculations showed that YH3 and LaH3 are both semiconductors, and this insulating property does not rely on the detailed crystal structure. The calculated band gaps agree with GW results very well, and they are also consistent with experiment. Our WDA calculations predicted good ground-state structures as well. The success of WDA may be attributed to the absence of self-interaction in H 1s states. ACKNOWLEDGMENTS We are grateful for helpful discussions with I. Geibels, P. Vajda, and K. Yvon. This work was supported by the Office of Naval Research under ONR Grant Nos. N000149710052, N00014-02-1-0506, and N0001403WX20028. Calculations were done on the Center for Piezoelectrics by Design 共CPD兲 computer facilities and on the Cray SV1 at the Geophysical Laboratory supported by NSF EAR-9975753, and the W. M. Keck Foundation. We also thank the Institut du Developpement et des Ressources en Informatique Scientifique 共IDRIS兲 for a grant of computer time. D.J.S. is grateful for the hospitality of the University of Paris Orsay during part of this work. APPENDIX: XC POTENTIAL OF THE WDA The xc potential V xc(r) in the DFT is the functional derivative of the xc energy, V xc共 r兲 ⫽ ␦ E xc关 n 兴 . ␦ n 共 r兲 共A1兲 In the shell-partitioning scheme as described in Sec. II, the xc potential of core states xc V core 共 r兲 ⫽ ␦ E xc关 n 兴 LDA ⫽V xc 共 r兲 ⫹V 1 共 r兲 ⫺V 2 共 r兲 , ␦ n c共 r兲 共A2兲 where the first term is the ordinary LDA potential, and 085104-4 V 1 共 r兲 ⫽ 冕 n v共 r兲 n v共 r⬘ 兲 ␦ G 关 兩 r⫺r⬘ 兩 ,n̄ 共 r兲兴 dr⬘ , 共A3兲 ␦ n c共 r兲 兩 r⫺r⬘ 兩 PHYSICAL REVIEW B 69, 085104 共2004兲 WEIGHTED-DENSITY-APPROXIMATION DESCRIPTION . . . V 2 共 r兲 ⫽ 冕 n 2v 共 r兲 ␦ G 关 兩 r⫺r⬘ 兩 ,n 共 r兲兴 dr⬘ . ␦ n 共 r兲 兩 r⫺r⬘ 兩 V 5 共 r兲 ⫽ 共A4兲 In order to evaluate V 1 (r), we need the expression for ␦ n̄(r)/ ␦ n c(r). This can be done by considering the variation of sum rule Eq. 共7兲 with respect to n c(r), ␦ n̄ 共 r兲 ⫽ ␦ n c共 r兲 n 2 共 r兲 冕 n v共 r兲 n v共 r⬘ 兲 ␦ G 关 兩 r⫺r⬘ 兩 ,n̄ 共 r兲兴 ␦ n̄ 共 r兲 冕冕 V 6 共 r兲 ⫽ 冕 n v共 r⬘ 兲 ␦ G 关 兩 r⫺r⬘ 兩 ,n̄ 共 r兲兴 冕 h 2 共 r兲 ⫽ ␦ n̄ 共 r兲 兩 r⫺r⬘ 兩 n v共 r⬘ 兲 ␦ G 关 兩 r⫺r⬘ 兩 ,n̄ 共 r兲兴 ␦ n̄ 共 r兲 冕 dr⬘ V 7 共 r兲 ⫽2 dr⬘ , dr⬘ , 共A6兲 冕 n 2v 共 r兲 h 1 共 r兲 . n 2 共 r兲 h 2 共 r兲 共A8兲 ␦ n̄ 共 r兲 ⫽⫺ ␦ n v共 r兲 ␦ E xc关 n 兴 xc LDA ⫽V xc V vln 共 r兲 ⫽ 共 r兲 ⫹V 3 共 r兲 ⫹V 4 共 r兲 ⫹V 5 共 r兲 ␦ n v共 r兲 共A9兲 V 3 共 r兲 ⫽ V 4 共 r兲 ⫽ 1 冕 冕 兩 r⫺r⬘ 兩 n v共 r⬘ 兲 兩 r⫺r⬘ 兩 V 5 共 r兲 ⫽ G 关 兩 r⫺r⬘ 兩 ,n̄ 共 r兲兴 dr⬘ , 共A10兲 G 关 兩 r⫺r⬘ 兩 ,n̄ 共 r⬘ 兲兴 dr⬘ , 共A11兲 G. G. Libowitz and A. J. Mealand, in Handbook on the Physics and Chemistry of Rare Earths, edited by K. A. Gschneidner, Jr. and L. Eyring 共North-Holland, Amsterdam, 1979兲, Vol. 3, p. 299. 2 J.N. Huiberts, R. Griessen, J.H. Rector, R.J. Wijngaarden, J.P. Dekker, D.G. de Groot, and N.J. Koeman, Nature 共London兲 380, 231 共1996兲; J.N. Huiberts, R. Griessen, R.J. Wijngaarden, M. Kremers, and C. Van Haesendonck, Phys. Rev. Lett. 79, 3724 共1997兲. 3 M. Rode, A. Borgschulte, A. Jacob, C. Stellmach, U. Barkow, and J. Schoenes, Phys. Rev. Lett. 87, 235502 共2001兲. 4 J. Osterwalder, Z. Phys. B: Condens. Matter 61, 113 共1985兲. 5 N.F. Miron, V.I. Shcherbak, V.N. Bykov, and V.A. Levdik, Kristallografiya 17, 404 共1972兲 关Sov. Phys. Crystallogr. 17, 342 共1972兲兴. 共A14兲 冕 G 关 兩 r⫺r⬘ 兩 ,n̄ 共 r⬘ 兲兴 n v共 r⬙ 兲 n 共 r兲 2 冕 ␦ G 关 兩 r⬘ ⫺r⬙ 兩 ,n̄ 共 r⬘ 兲兴 ␦ n̄ 共 r⬘ 兲 , dr⬙ 共A15兲 n c共 r兲 n v共 r⬘ 兲 ␦ G 关 兩 r⫺r⬘ 兩 ,n̄ 共 r兲兴 ␦ n̄ 共 r兲 . dr⬘ 共A16兲 And the final expressions for V 5 and V 6 are where n v共 r⬘ 兲 G 关 兩 r⫺r⬘ 兩 ,n 共 r兲兴 dr⬘ ⫹V 2 共 r兲 . 兩 r⫺r⬘ 兩 ␦ n̄ 共 r⬘ 兲 ⫽⫺ ␦ n v共 r兲 共A7兲 Similarly the xc potential of valence states is ⫹V 6 共 r兲 ⫺V 7 共 r兲 , n v共 r兲 共A13兲 Again we can obtain ␦ n̄(r⬘ )/ ␦ n v(r) and ␦ n̄(r)/ ␦ n v(r) from variation of Eq. 共7兲, we can simplify V 1 共 r兲 ⫽ n v共 r兲 n v共 r⬘ 兲 ␦ G 关 兩 r⫺r⬘ 兩 ,n̄ 共 r兲兴 dr⬘ , ␦ n v共 r兲 兩 r⫺r⬘ 兩 . 共A5兲 If we denote h 1 (r) and h 2 (r) as h 1 共 r兲 ⫽ n v共 r⬘ 兲 n v共 r⬙ 兲 ␦ G 关 兩 r⬘ ⫺r⬙ 兩 ,n̄ 共 r⬘ 兲兴 dr⬘ dr⬙ , ␦ n v共 r兲 兩 r⬘ ⫺r⬙ 兩 共A12兲 冕 n v共 r⬘ 兲 h 1 共 r⬘ 兲 ␦ G 关 兩 r⫺r⬘ 兩 ,n̄ 共 r⬘ 兲兴 h 2 共 r⬘ 兲 V 6 共 r兲 ⫽⫺ ␦ n̄ 共 r⬘ 兲 n v共 r兲 n c共 r兲 h 1 共 r兲 . n 2 共 r兲 h 2 共 r兲 dr⬘ , 共A17兲 共A18兲 A. Pebler and W.E. Wallace, J. Phys. Chem. 66, 148 共1962兲. T.J. Udovic, Q. Huang, R.W. Erwin, B. Hjorvarsson, and R.C.C. Ward, Phys. Rev. B 61, 12 701 共2000兲. 8 O.J. Zogal, W. Wolf, P. Herzig, A.H. Vuorimaki, E.E. Ylinen, and P. Vajda, Phys. Rev. B 64, 214110 共2001兲. 9 An early calculation did yield a gap, which is not reproduced in more recent LDA and GGA calculations, A.C. Switendick, Int. J. Quantum Chem. 5, 459 共1971兲. 10 J.P. Dekker, J. van Ek, A. Lodder, and J.N. Huiberts, J. Phys.: Condens. Matter 5, 4805 共1993兲. 11 Y. Wang and M.Y. Chou, Phys. Rev. Lett. 71, 1226 共1993兲; Phys. Rev. B 51, 7500 共1995兲. 12 P.J. Kelly, J.P. Dekker, and R. Stumpf, Phys. Rev. Lett. 78, 1315 共1997兲. 13 R. Ahuja, B. Johansson, J.M. Wills, and O. Eriksson, Appl. Phys. Lett. 71, 3498 共1997兲. 6 7 085104-5 PHYSICAL REVIEW B 69, 085104 共2004兲 WU, COHEN, SINGH, GUPTA, AND GUPTA 14 G. Onida, L. Reining, and A. Rubio, Rev. Mod. Phys. 74, 601 共2002兲. 15 K.K. Ng, F.C. Zhang, V.I. Anisimov, and T.M. Rice, Phys. Rev. Lett. 78, 1311 共1997兲; Phys. Rev. B 59, 5398 共1999兲. 16 R. Eder, H.F. Pen, and G.A. Sawatzky, Phys. Rev. B 56, 10 115 共1997兲. 17 P. van Gelderen, P.A. Bobbert, P.J. Kelly, and G. Brocks, Phys. Rev. Lett. 85, 2989 共2000兲. 18 P. van Gelderen, P.J. Kelly, and G. Brocks, Phys. Rev. B 63, 100301 共2001兲. 19 T. Miyake, F. Aryasetiawan, H. Kino, and K. Terakura, Phys. Rev. B 61, 16 491 共2000兲. 20 J.P. Perdew and A. Zunger, Phys. Rev. B 23, 5048 共1981兲. 21 O. Gunnarsson, M. Jonson, and B.I. Lundqvist, Phys. Lett. 59A, 177 共1976兲; Solid State Commun. 24, 765 共1977兲. 22 J.A. Alonso and L.A. Girifalco, Phys. Rev. B 17, 3735 共1978兲. 23 O. Gunnarsson, M. Jonson, and B.I. Lundqvist, Phys. Rev. B 20, 3136 共1979兲. 24 O. Gunnarsson and R.O. Jones, Phys. Scr. 21, 394 共1980兲. 25 I.I. Mazin and D.J. Singh, Phys. Rev. B 57, 6879 共1998兲. 26 O.V. Gritsenko, A. Rubio, L.C. Balbás, and J.A. Alonso, Chem. Phys. Lett. 205, 348 共1993兲. P. Gori-Giorgi and J.P. Perdew, Phys. Rev. B 66, 165118 共2002兲. 28 P.P. Rushton, D.J. Tozer, and S.J. Clark, Phys. Rev. B 65, 235203 共2002兲. 29 N. Troullier and J.L. Martins, Phys. Rev. B 43, 1993 共1991兲. 30 D.J. Singh, Phys. Rev. B 48, 14 099 共1993兲. 31 Zhigang Wu, R.E. Cohen, and D.J. Singh, in Proceedings of Fundamental Physics of Ferroelectrics, 2003, edited by P. K. Davies and D. J. Singh, p. 276. 32 A.T.M. van Gogh, D.G. Nagengast, E.S. Kooij, N.J. Koeman, J.H. Rector, R. Griessen, C.F.J. Flipse, and R.J.J.G.A.M. Smeets, Phys. Rev. B 63, 195105 共2001兲. 33 S.J. van der Molen, D.G. Nagengast, A.T.M. van Gogh, J. Kalkman, E.S. Kooij, J.H. Rector, and R. Griessen, Phys. Rev. B 63, 235116 共2001兲. 34 J.A. Alford, M.Y. Chou, E.K. Chang, and S.G. Louie, Phys. Rev. B 67, 125110 共2003兲. 35 E.K. Chang, X. Blase, and S.G. Louie, Phys. Rev. B 64, 155108 共2001兲. 36 R.W. Godby, M. Schlüter, and L.J. Sham, Phys. Rev. Lett. 56, 2415 共1986兲. 27 085104-6