This paper is published as part of a PCCP Themed Issue on: Time-Dependent Density-Functional Theory

Guest Editors: Miguel A. L. Marques and Angel Rubio

Editorial
Time-dependent density-functional theory Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b908105b

Towards a gauge invariant method for molecular chiroptical properties in TDDFT Daniele Varsano, Leonardo A. Espinosa-Leal, Xavier Andrade, Miguel A. L. Marques, Rosa di Felice and Angel Rubio, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b903200b Second-order nonlinear optical properties of transition metal clusters [MoS4Cu4X2Py2] (M = Mo, W; X = Br, I) Qiaohong Li, Kechen Wu, Yongqin Wei, Rongjian Sa, Yiping Cui, Canggui Lu, Jing Zhu and Jiangang He, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b903582f Absorption and fluorescence properties of oligothiophene biomarkers from long-range-corrected time-dependent density functional theory Bryan M. Wong, Manuel Piacenza and Fabio Della Sala, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b901743g Time-dependent current-density functional theory for generalized open quantum systems Joel Yuen-Zhou, Csar Rodrguez-Rosario and Aln AspuruGuzik, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b903064f Optical and magnetic properties of boron fullerenes Silvana Botti, Alberto Castro, Nektarios N. Lathiotakis, Xavier Andrade and Miguel A. L. Marques, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b902278c Inhomogeneous STLS theory and TDCDFT John F. Dobson, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b904385n Bound states in time-dependent quantum transport: oscillations and memory effects in current and density E. Khosravi, G. Stefanucci, S. Kurth and E.K.U. Gross, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b906528h Time-dependent density functional theory for resonant properties: resonance enhanced Raman scattering from the complex electric-dipole polarizability Abdelsalam Mohammed, Hans gren and Patrick Norman, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b903250a

Perspective
Time-dependent density functional theory of high excitations: to infinity, and beyond Meta van Faassen and Kieron Burke, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b901402k

Papers
Time-dependent density functional theory versus BetheSalpeter equation: an all-electron study Stephan Sagmeister and Claudia Ambrosch-Draxl, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b903676h TD-DFT calculations of electronic spectra of hydrogenated protonated polycyclic aromatic hydrocarbon (PAH) molecules: implications for the origin of the diffuse interstellar bands? Mark Hammonds, Amit Pathak and Peter J. Sarre, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b903237a TDDFT diagnostic testing and functional assessment for triazene chromophores Michael J. G. Peach, C. Ruth Le Sueur, Kenneth Ruud, Maxime Guillaume and David J. Tozer, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b822941d An ab initio and TD-DFT study of solvent effect contributions to the electronic spectrum of Nile Red Patrick Owen Tuck, Robert Christopher Mawhinney and Manit Rappon, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b902528f

On the proton transfer mechanism in ammonia-bridged 7-hydroxyquinoline: a TDDFT molecular dynamics study Matteo Guglielmi, Ivano Tavernelli and Ursula Rothlisberger, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b903136g Chemical and protein shifts in the spectrum of the photoactive yellow protein: a time-dependent density functional theory/molecular mechanics study Eneritz Muguruza Gonzlez, Leonardo Guidoni and Carla Molteni, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b902615k Excitation energies from ground-state densityfunctionals by means of generator coordinates E. Orestes, A. B. F. da Silva and K. Capelle, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b902529d A time-dependent density-functional approach to nonadiabatic electron-nucleus dynamics: formulation and photochemical application Hirotoshi Hirai and Osamu Sugino, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b901144g Wavepacket basis for time-dependent processes and its application to relaxation in resonant electronic transport Peter Bokes, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b902501d Can phthalocyanines and their substituted -para(methoxy)phenyl derivatives act as photosensitizers in photodynamic therapy? A TD-DFT study Angelo Domenico Quartarolo, Ida Lanzo, Emilia Sicilia and Nino Russo, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b819064j Substituent effects on the light-induced CC and CBr bond activation in (bisphosphine)( 2-tolane)Pt0 complexes. A TD-DFT study Daniel Escudero, Mariana Assmann, Anne Pospiech, Wolfgang Weigand and Leticia Gonzlez, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b903603b Photodegradation mechanism of the common nonsteroid anti-inflammatory drug diclofenac and its carbazole photoproduct Klefah A. K. Musa and Leif A. Eriksson, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b900144a

Computation of accurate excitation energies for large organic molecules with double-hybrid density functionals Lars Goerigk, Jonas Moellmann and Stefan Grimme, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b902315a Time-dependent current density functional theory via time-dependent deformation functional theory: a constrained search formulation in the time domain I. V. Tokatly, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b903666k Photoabsorption spectra from adiabatically exact timedependent density-functional theory in real time Mark Thiele and Stephan Kmmel, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b902567g Double excitation effect in non-adiabatic time-dependent density functional theory with an analytic construction of the exchangecorrelation kernel in the common energy denominator approximation Oleg V. Gritsenko and Evert Jan Baerends, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b903123e Physical signatures of discontinuities of the timedependent exchangecorrelation potential Daniel Vieira, K. Capelle and C. A. Ullrich, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b902613d Autoionizing resonances in time-dependent density functional theory August J. Krueger and Neepa T. Maitra, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b902787d The polarizability in solution of tetra-phenyl-porphyrin derivatives in their excited electronic states: a PCM/TDDFT study Roberto Improta, Camilla Ferrante, Renato Bozio and Vincenzo Barone, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b902521a A new generalized KohnSham method for fundamental band-gaps in solids Helen R. Eisenberg and Roi Baer, Phys. Chem. Chem. Phys., 2009 DOI: 10.1039/b902589h

PAPER

www.rsc.org/pccp | Physical Chemistry Chemical Physics

A new generalized KohnSham method for fundamental band-gaps in solids

Helen R. Eisenberg and Roi Baer*
Received 9th February 2009, Accepted 20th April 2009 First published as an Advance Article on the web 5th May 2009 DOI: 10.1039/b902589h We developed a method for calculating the ground-state properties and fundamental band-gaps of solids, using a generalized KohnSham approach combining a local density approximation (LDA) functional with a long-range explicit exchange orbital functional. We found that when the range parameter is selected according to the formula g = A/(eN ~) where eN is the optical dielectric e constant of the solid and ~ = 0.84 and A = 0.216 a01, predictions of the fundamental band-gap e close to the experimental values are obtained for a variety of solids of dierent types. For most solids the range parameter g is small (i.e. explicit exchange is needed only at long distances) so the predicted values for lattice constants and bulk moduli are similar to those based on conventional LDA calculations. Preliminary calculations on silicon give a general band structure in good agreement with experiment.

I.

Introduction
1

Density functional theory (DFT), applied via the KohnSham (KS) approach,2 is routinely used for the successful determination and interpretation of the structural and cohesive properties of a broad variety of solid state systems. However, solid-state band-gaps are typically underestimated by a factor of 2. In the KS approach the system of electrons is mapped onto a system of non-interacting fermions (the KS system) governed by a local potential (in practical calculations this local potential is only approximately known). In this non-interacting system a one-particle Schrodinger equation is set up and its eigenstates and eigenvalues (called KS orbital energies) are determined. The lowest orbital energies are then used as estimates for ionization potentials. However, such a procedure is only known to be rigorously correct for the Fermi level.3,4 The fundamental band-gap of the solid, Eg, is often approximated by the KohnSham band-gap (the dierence between the bottom of the conduction band and the top of the valence band). However, this too is unjustied. For a system of N electrons the fundamental band-gap is a ground state quantity in the sense that it can be expressed in terms of the ground-state energies, Egs, of the N 1, N and N + 1 electron systems as follows: Eg IP EA lim Egs N 1 2Egs N Egs N 1
N!1

1:1

where IP is the ionization potential and EA the electron anity. For a nite electron system the band-gap can be computed by considering the KS systems of N and N + 1 particles and can be expressed as follows: Eg = eN,N + eN + 1,N+1 = (eN,N+1 eN,N) + DXC (1.2)

Institute of Chemistry and the Fritz Haber Center for Molecular Dynamics, The Hebrew University of Jerusalem, Jerusalem 91904, Israel. E-mail: [email protected]

Where eN,M is the Mth KS orbital energy of the N-particle KS system and DXC = eN + 1,N + 1 eN,N+1 is called the derivative discontinuity (because it does not go to zero in the N - N limit) and the term in parentheses is the KS band-gap. Evidently, even if the exact local potential of the KS systems was known, the common procedure of taking the KS band-gap would not give the correct solid-state gap as one would still need to add to the orbital gap the derivative discontinuity DXC.57 The popular local and semi-local approximations, such as the local density approximation (LDA) and the generalized gradients approximations (GGAs), are believed to give local potentials and orbital energies which are not bad approximations to the exact KS quantities.8 However, because of their inherent semi-local density structure, they predict a zero derivative discontinuity (DXC = 0).5 Therefore, the fundamental bandgaps inferred from them, are similar to the exact KohnSham band gaps but are poor approximations to the experimental gaps (by a factor of B2). The fact that derivative discontinuities are missing from semi-local functionals has been associated with the existence of spurious electron self-repulsion.5,9,10 The complication of adding the derivative discontinuity to the KS band can be circumvented in the generalized KohnSham (GKS) framework11 which uses explicit orbital functionals (i.e. the orbitals are explicitly contained in the functional as opposed to the functional being only explicitly dependent on the density). With explicit orbital functionals, the orbital energy band-gap already incorporates some or all of the derivative discontinuity and so may be used to directly approximate the experimental band-gap. For example, if the orbital functional includes a HartreeFock like exchange operator the derivative discontinuity DXC can be decomposed into the sum of corresponding exchange (DX) and correlation (DC) parts. It was demonstrated, using perturbative arguments, that the orbital energy gap in the GKS equation incorporates much of DX while the correlation derivative discontinuity DC was estimated as a much smaller correction.11
This journal is
 c

4674 | Phys. Chem. Chem. Phys., 2009, 11, 46744680

the Owner Societies 2009

There are several previous works which use a generalized KS framework for band-gaps. One approach splits the exchange energy into an explicit short-range exchange operator and a local functional for the long-range exchange.1113 This resulted in signicant improvements over the LDA fundamental bandgap for some of the materials studied. A similar approach was applied using the HSE functional with much improved results, probably due to the use of more advanced semi-local exchangecorrelation functionals.14 Both these approaches are not expected to include the entire exchange discontinuity DX because the orbital functional they use does not include a longrange self-repulsion correction. Therefore, one cannot expect the fundamental band-gap to be fully contained in their GKS orbital gap. This problem can be xed by using an explicit exchange operator for the long-range (instead of the short range) part of the exchange. This might thus be a more appropriate way to circumvent the derivative discontinuity contribution. Such an attempt was reported recently but huge gaps were found for most materials.15 In this work we present a generalized KohnSham method which deploys an explicit orbital exchange that incorporates the long-range self-repulsion correction exactly. The procedure introduces into the DFT correlation energy expression a short-range but non-local exchange-like orbital functional which eliminates the detrimental eects of the full HartreeFock-like exchange.1619 In this sense our approach is similar to that of ref. 15. The crucial dierence is that here the range-parameter g is not considered universal. It depends on the density of the system and must be tuned separately for each system.1922 We show how such a tuning can be done for solid state systems: the range-parameter g was found empirically to correlate very well with the optical dielectric constant eN of the solid. Based on this relation we developed a method that describes the usual ground-state properties of solids (lattice parameter and bulk modulus) with LDA quality while simultaneously describing the fundamental band-gaps extremely well. Our theory for the range-separated hybrid is discussed in section II. The relation between the range parameter and the dielectric constant is the topic of section III. The performance of the method is demonstrated in section IV followed by summary and discussion in section V.

P 1 ^ U 1 Ne rnm are, respectively, the kinetic energy, the internam 2 action with the external potential and the electron Coulomb repulsion operators (in atomic units). The last term EC|n| is the negative correlation energy functional which, because the two systems have the same density, can be written as: EC[n] = hcgs|T + U |cgsi hcS|T + U |cSi (2.2)

As we explain below, there exists a g for which the correlation energy can also be expressed as:19 EC[n] = hcgs|Yg|cgsi hcS|Yg|cSi where Yg is a shielded Coulomb interaction energy: 1X ^ Yg yg jrn rm j 2 nam With the pair potential: yg r erfcgr r40 r
g!1

(2.3)

2:4

2:5

This function has the properties that lim yg r 0 and y0 r 1, which can be used to show that for each groundr state density n(r) there exists a g for which eqn (2.3) holds exactly.19 Even when g is known expression (2.3) is not practical for calculations since we have no access to cgs. We thus follow the spirit of the local density approximation and approximate it as: R g EHyb[n] = fXC(n(r))d3r C R + 1 |r(r1,r2)|2yg(|r1 r2|)d3r1d3r2 (2.6) 2 in which r(r1,r2) is the density matrix of the non-interacting system and fg (n(r),|rn(r)|, ) = [eHEG(n(r)) + eg (n(r))]n(r) XC C X eHEG(n) C (2.7)

II.

The range-separated hybrid functional

In the KohnSham approach DFT the ground-state energy of a system of Ne electrons in an external potential u(r), with particle-density n(r) and a many-electron wave function cgs is expressed using quantities calculated for a system of non-interacting fermions with an identical ground-state density. The wave function of this non-interacting system is a Slater determinant and is denoted by cS.2 Both wave functions, cgs and cS are unique (up to global phase) functionals of the density. The energy of the interacting system is written as: E = hcgs|H|cgsi = hcS|H|cSi + EC[n] (2.1)

is the correlation energy per electron of the homogeneous gas (HG) of electrons at density n and eg is the X exchange energy of the HG of particles interacting with the potential yg (r).23 Eqn (2.6) depicts the correlation energy of DFT as a functional of the density. However, in GKS it can be considered a functional of the orbitals. This is our intention here. The details of the approach given here are described in ref. 19. We should note that similar approaches to the present method were conceived earlier1618,24 but view g as a universal parameter that is obtained by tting to a database of molecular properties. Recently we used a similar method to the one described here for molecules which are problematic for conventional DFT.20 In that paper it was shown that ab-initio tuning of the range parameter is necessary in order to describe the symmetric radical cation R + R+ in the ground state.21 Furthermore, range parameter tuning is also required for charge-transfer excitations.25

III.
A

Application for solids

Technical computational details

where H = T + V + U is the Hamiltonian of the interacting P P ^ ^ system and T Ne 1 1 r2 , V Ne u rn and n n1 2 n

This journal is
 c

The previous section has briey reviewed the range separated functional; we now describe its use for predicting the properties of solids. The calculations described below were
Phys. Chem. Chem. Phys., 2009, 11, 46744680 | 4675

the Owner Societies 2009

carried out using the Quantum-ESPRESSO package26 which we modied to include the functional described in eqns (2.6)(2.7). The local correlation functional used was the PZ81 LDA functional.27 The local exchange functional is the LDA exchange of a homogeneous gas of particles interacting with the yg(r) pair potential.16,18 All calculations used a plane-wave basis and were converged for kinetic energy cuto for wave-functions (varied with material) and k-point grid density (4 4 4 grid was used). Calculations were carried out at the minimum energy lattice constant for each value of g. Norm-conserving pseudopotentials based on the PZ81 exchange correlation functional were used and all the calculations were fully self-consistent. We ignored spinorbit (SO) splitting eects when calculating the fundamental band-gap. These eects are negligible in all the systems we studied except AlSb and to a lesser extent AlAs. Following ref. 11 the correction to the DFT calculated fundamental gap due to spinorbit splitting is approximately DSO/3 at g (where DSO is the spinorbit splitting energy). For AlSb this gives a SO correction of 0.23 eV (13% of the band gap) and for AlAs a SO correction of 0.09 eV (4% of the band gap). In future works we intend to investigate the eects of large SO splitting and determine whether or not corrections for this need to be included in our method. B Band-gap dependence on the range parameter

As also found for molecules20,21,25 the range parameter g in the functional must undergo a tuning stage. In order to determine a method for this we study the dependence of the fundamental gap on g by considering the following non-dimensional quantity which we call the normalized band gap: gg
g 0 Eg Eg 0 LDA 1 HF ; Eg  Eg ; Eg % Eg 1 E0 Eg g

3:1

where Eg is the gap computed with the range-separated hybrid g functional with parameter g. By denition, gN = 1 and this was found to be close to the results of HartreeFock (HF) theory, while g0 = 0 is the LDA limit for this equation. We plot the values of gg for several materials, diamond (C), silicon, SiC, and MgO in Fig. 1a. It can be seen that the normalized gap energy is sensitive to g for 0.01 o g o 1. The experimental band gap values typically coincide with calculated ones in the range 0.01 o g o 0.1. A dierent behavior is seen for the lattice constant a: it is insensitive to g when 0.01 o g o 0.1 and takes the LDA value. This is clearly seen in Fig. 1b, where the normalized lattice constant, dened by: ag ag a0 ; a0  aLDA ; a1 % aHF a1 a0 3:2

Fig. 1 (a) The normalized band gap (eqn (3.1)) as a function of g for several materials. Direct (indirect) band gaps are denoted dir (indir). The horizontal lines are the experimental results (exp). (b) The normalized lattice constant (eqn (3.2)) of SiC.

fundamental band-gap is equal to the corresponding experimental band-gap. C The relation between c* and the dielectric constant

is shown for SiC (other materials show similar behavior) as a function of g. Change in the lattice constant starts at g 4 0.1. We conclude that the material band-gaps are very sensitive to g in the experimentally relevant range while the lattice constants take the LDA values there. Furthermore each material requires a dierent value of g in order to obtain the experimental band-gap. For any given material we designate by g* the value of g for which the calculated
4676 | Phys. Chem. Chem. Phys., 2009, 11, 46744680

The basic problem is to nd a method for determining g*. Thus, we have to nd an additional relation between g and another internal property of the solid, which can be calculated from the electronic structure. This way one can determine g* computationally without any material-dependent external input parameters. We found that a simple relation exists between g* and the optical dielectric constant eN. We plot in Fig. 2 the value of g* as a function of the experimental eN for several solids. It can be seen that the relation between g* and eN can be described approximately by a simple functional relation with two parameters A and ~: e g* e1 % A e e1 ~
 c

3:3

This journal is

the Owner Societies 2009

Fig. 3 Fig. 2 The value of g giving experimental fundamental band-gaps, g*, plotted against the experimental value of the optical dielectric constants eN for dierent materials. The line depicts the relation given in eqn (3.3) with A = 0.216 a01 and ~ = 0.84. e

The dependence of the calculated dielectric constant on g.

IV.

Results

This relation will naturally give g* - 0 for metallic systems (where eN - N), therefore LDA is recovered which is appropriate for metals. In order to avoid singularities and negative values of g*, ~ must be smaller than 1 (all materials have e eN Z 1). We tted to the data the constants A = 0.216 a01 and ~ = 0.84. We adopt these values as an empirical relation e and use them in this study. Algorithm. The way to use eqn (3.3) in a calculation is: (1) Perform an LDA calculation to determine eN. (2) Determine g* based on eN from eqn (3.3) and perform the DFT calculation with the range-separated hybrid functional with g*. (3) Determine the band-gap by measuring the energy dierence between the minimum energy of the conduction band orbital and the maximum energy of the valence band orbital. For small g the calculated dielectric constant eN does not e1 g signicantly depend on g. We plotted the ratio rg e1 0 as a function of g for Si and MgO (Fig. 3). It is seen that for Si the computed eN does not change signicantly from its LDA value for g o 0.05 and even until g = 0.1 the change is small (less than 6%). For MgO, the range at which eN takes the LDA value is even larger. We conclude that for materials with small g* (such as semiconductors) the proximity of the hybrid functional to LDA justies using the LDA value of eN. For larger gap materials this may need to be checked on a case by case basis and if needed the algorithm above can be modied to include an iteration step requiring self-consistency between eN and g*. An important point to note is that semiconducting systems (which are close to metals) have very small g* values, and so the predictions of the hybrid functional concerning ground-state properties become almost identical with those of LDA. Even the large band-gap systems, such as NaCl have a value of g* smaller by more than a factor of 2 than the typical values of g used for gas-phase molecules (which have g B0.520).
This journal is
 c

Our generalized KS DFT method consists of using as a correlation functional eqns (2.6)(2.7) with the value of g* determined by the relation (3.3). The dielectric constant in the latter equation is to be taken from the LDA calculation. However, we used the experimental dielectric constant instead in this paper (which is expected to be close to that of LDA) in order to demonstrate the concepts. We further discuss this issue later, and give a few results with calculated dielectric constants. We begin with ground-state properties: the lattice constant a and the isotropic bulk modulus B. We computed E(a), the SCF energy for several lattice constants near the optimal amin; we then t to the resulting data a low order polynomial Ep(E) = b0 + b1E + b2E2 + , using the polynomial for which we determined the exact value of the lattice constant amin (where E 0 p(amin) = 0)) and for the Bulk modulus we computed the 2nd derivative E00 p(amin) from which the bulk modulus is determined by: B p9V min where V is the unit cell volume. For several materials, where anharmonic eects are signicant, we noticed that the second derivative results were somewhat sensitive to the order and location of data points we used; thus we determined a condence interval for the value of B. The results of this procedure are depicted in Table 1 where the calculation results for the lattice constants and the bulk moduli are compared with various experimental measurements. The lattice constants we computed were almost identical with those of LDA. This is a result of the very small value of g* we used, which leaves us very close to the LDA limit. As for the bulk modulus, this quantity is more sensitive to g*, however in all cases we obtained values which are in agreement (or very closely so) with experimental results. The largest discrepancy was for NaCl where our result, although too large by 20%, was somewhat better than LDA. Next we consider the fundamental band-gaps. The values of g*(eN) shown in Table 1 were used to compute the fundamental band-gaps of dierent materials. We found good band-gaps, as shown in Fig. 4. This is not very surprising in view of the nice t eqn (3.3) gives for the data of g*. The only point lying signicantly below the curve of Fig. 2 is that
Phys. Chem. Chem. Phys., 2009, 11, 46744680 | 4677
a2 E 00 a

the Owner Societies 2009

Table 1 Dielectric constants, lattice constants and bulk moduli for the various solids considered in this study with comparison to experiment. LDA-g refers to the results obtained using the present method Experimental Material C Si SiC AlAs AlP AlN AlN AlSb GaP MgO NaCl Symmetry Diamond Diamond Zinc-blende Zinc-blende Zinc-blende Zinc-blende Wurtzite Zinc-blende Zinc-blende Cubic (rock salt) Cubic (rock salt) Solid type Insulator Semicond. Semicond. Semicond. Semicond. Semicond. Semicond. Semicond. Semicond. Ionic solid Ionic solid eNa 5.7 11.7 6.52 8.16 7.5 4.84 4.66 10.24 9.0 2.95 2.25 g*(eN) (a01) 0.045 0.020 0.038 0.031 0.033 0.054 0.057 0.023 0.027 0.104 0.157 Lattice constant (a0) LDA 6.67 10.23 8.21 10.61 10.23 8.07 a c u 11.55 10.10 7.94 10.83 LDA-g 6.67 10.23 8.22 10.61 10.23 8.08 5.76c 9.10c 0.388c 11.55 10.10 7.95 10.86 Experiment 6.74 10.26 8.24 10.68 10.31 8.28 5.88 9.41 0.382 11.58 10.28 7.96 10.58
b

Bulk Modulus (GPa) LDA 580585 9697 195245 7275 89 187196 200220 56 93 170 3038 LDA-g 510640 92 215225 7476 89 218 200220 56 93 163165 2831 Experimentd 440560 80100 220260 74 86e 191218 160210 5557 8591f 160-165g 24h

5.76 9.10 0.388

a Comments: C, Si, SiC, AlAs,28 AlP,29 AlN,30,31 AlSb,32 GaP,33 MgO,34 NaCl35 b C, Si, SiC,28 AlAs, AlP, AlN,AlSb, GaP,36 MgO,37 NaCl38 c For the wurtzite structure we took the LDA lattice parameters for u and c/a and minimized for a. d ref. 33 unless otherwise stated. e ref. 39 f ref. 40 g refs. 4143 h ref. 44,45

corresponding to AlN in the wurzite conguration (AlN(wur)). Indeed the band-gap predicted for this system deviates more signicantly from the experimental band-gap than in the other systems. In all cases the band gaps, which are much too small according to LDA, were greatly improved according to the new functional. Especially encouraging is the fact that small semiconductor gaps are described as well as large ionic gaps using the simple 2-parameter formula in eqn (3.3).

A detailed comparison of the relative band gap errors is shown in Table 2. The LDA band gaps are B40% too small, as is well known. GDFT and exact exchange typically overestimates the gaps by nearly 10%. HSE and GW show a good balanced description as they exhibit small mean relative errors but the mean absolute relative error is 56%. The new functional is also very balanced, showing vanishing mean relative errors and the mean absolute relative error is small as well, 3%.

Fig. 4 Comparison of calculated vs. experimental fundamental band-gaps using dierent methods. The straight line shows the experimental results. LDA refers to the results obtained using Quantum-ESPRESSO26 in the local density approximation; LDA-g refers to the results obtained using modied Quantum-ESPRESSO including our new functional; GW refers to results obtained using the GW approximation: AlN,46 Si,47 C,48 SiC,49 AlAs,50,51 GaP,52 AlP, AlSb,53; GWA refers to results obtained from the GW approximation,54 GDFT refers to results obtained using generalized-density-functional theory: Si, C, SiC, AlAs, AlP, AlN-zinc-blende, AlSb, GaP;55 NaCl,56 Exact-exchange refers to results obtained using the exact-exchange KohnSham formalism;57 GKS refers results obtained using the generalized KohnSham scheme;11 HSE refers to results obtained using the HeydScuseriaErnzerhof screened hybrid potential.14

4678 | Phys. Chem. Chem. Phys., 2009, 11, 46744680

This journal is

 c

the Owner Societies 2009

Table 2 Statistics on relative errors in the calculated fundamental band-gaps for the various methods considered here. MeanAbs refers to the mean absolute relative error. N is the number of results available for each method Method g GW4653 HSE14 Exact-exch57 GDFT55,56 GWA54 GKS11 LDA N 10 9 9 5 8 5 1 10 Mean 0% 1% 2% 7% 8% 13% 18% 39% MeanAbs 3% 5% 6% 8% 11% 13% 18% 39%

Table 3 Band-gaps (eV) in silicon. LDA-g* refers to calculations performed using our new functional with the range parameter g which gives the experimental fundamental gap; LDA-g refers to calculations with g determined by the relation (3.3). The gap energies are shown as arrows in Fig. 5 and are labeled by the gap symbols used in this table LDA Eg(G25 0 v-D1c) EX(G25 0 v-X1c) EL(G25 0 v-L1c) EG1(G25 0 v-G15c) EG2(G25 0 v-G2 0 c)
a

LDA-g* 1.17 1.31 2.20 3.24 4.07

LDA-g 1.10 1.23 2.13 3.17 3.99

Exper. 1.17a 1.2b 2.2a 3.4a 4.19a

0.48 0.62 1.51 2.55 3.37

b

From ref. 58.

From ref. 28.

LDA dielectric constants are somewhat higher than experiment. For example, in Si the experimental dielectric constant is eexp = 11.7 while our LDA calculation yielded eLDA = 13.1 N N This is a relative error of B12%. This does not mean that our estimate of the band gap has a relative error of 12%. It is about a factor of 2 smaller as can be explained by the following considerations: using the LDA dielectric constant in eqn (3.3) will lead to smaller than optimal g*, by B12%. This will cause the calculated normalized band gap gg (eqn (3.1)) to decrease by B12% (see Fig. 1). But the eect on the actual gap is seen to be (from eqn (3.1)):
g dEg g Eg

method. In Table 3 we give several specic gap energies predicted by our new method with two values of the range parameter, namely optimal g* and the value of g determined by eqn (3.3) using the optical dielectric constant. The values of the LDA calculation and experimental values are shown for comparison. The band-gaps predicted by our method are within 10% of the experimental values. The LDA band gaps are all shifted by almost the same amount as the fundamental gap.

V. Summary and discussion

We developed a theory for applying the range separated hybrid developed in ref. 19 for solid-state systems. We showed that the range parameter g must be selected according to a simple empirical equation depending on the optical dielectric constant of the material (eqn (3.3)). This leads to a new method which predicts ground-state properties (lattice constants, bulk moduli) with accuracy comparable to that of LDA and at the same time gives band-gaps which are close to experimental values. We have demonstrated the results on a series of systems spanning semiconductors, insulators and wide band gap materials such as ionic solids. We showed that this method is self-contained within the modied DFT code and does not rely on any external material-dependent input parameters. We have performed preliminary tests on the ability of our new functional to predict the general band-structure of solids as well as the fundamental band-gap. Initial results on silicon show an agreement with experiment of B10%. We are currently in the process of calculating the band-structure of other materials. We intend to test our new method on additional classes of solids, including those with large spin orbit splitting. Finally, an additional future research direction is the applicability to reduced dimensionality innite systems (1D polymers and 2D sheets) and large but nite systems, e.g. proteins.

LDA Eg g Eg

dgg gg

4:1

In Si (1 ELDA/Eg ) E 0.6 (i.e. LDA band gaps have a relative g g error of 60% with respect to the true band-gap) and so we expect the relative error in the gap, according to the new functional based on the LDA dielectric constant, to be about B7%. In MgO, where the LDA dielectric constant is closer to experiment the same type of calculation leads to a much smaller error, to 2.5%. In a future publication we will investigate the relation between the LDA dielectric constant and the experimental gaps more closely. The eect of the long-range correction on the band structure as a whole will be studied in detail in a future publication. Here we present the preliminary results for silicon. In Fig. 5 we show the detailed band structure predicted by our new

Acknowledgements
We gratefully acknowledge support of the USIsrael Binational Science Foundation.

References
Fig. 5 The band-structure of Si using the new functional. 1 P. Hohenberg and W. Kohn, Phys. Rev., 1964, 136, B864. 2 W. Kohn and L. J. Sham, Phys. Rev., 1965, 140, A1133.

This journal is

 c

the Owner Societies 2009

Phys. Chem. Chem. Phys., 2009, 11, 46744680 | 4679

3 M. Levy and J. P. Perdew, in Density Functional Methods in Physics, ed. R. Dreizler and J. Perovidencia, Plenum, New York, 1985, p. 11. 4 C.-O. Almbladh and U. von-Barth, Phys. Rev. B, 1985, 31, 3231. 5 J. P. Perdew, R. G. Parr, M. Levy and J. L. Balduz, Phys. Rev. Lett., 1982, 49, 1691. 6 L. J. Sham and M. Schluter, Phys. Rev. Lett., 1983, 51, 1888. 7 J. P. Perdew and M. Levy, Phys. Rev. Lett., 1983, 51, 1884. 8 K. Burke, J. P. Perdew and M. Ernzerhof, J. Chem. Phys., 1998, 109, 3760. 9 S. Kummel and L. Kronik, Rev. Mod. Phys., 2008, 80, 3. 10 A. J. Cohen, P. Mori-Sanchez and W. T. Yang, Science, 2008, 321, 792. 11 A. Seidl, A. Gorling, P. Vogl, J. A. Majewski and M. Levy, Phys. Rev. B, 1996, 53, 3764. 12 D. M. Bylander and L. Kleinman, Phys. Rev. B, 1990, 41, 7868. 13 C. B. Geller, W. Wolf, S. Picozzi, A. Continenza, R. Asahi, W. Mannstadt, A. J. Freeman and E. Wimmer, Appl. Phys. Lett., 2001, 79, 368. 14 J. Heyd, J. E. Peralta, G. E. Scuseria and R. L. Martin, J. Chem. Phys, 2005, 123. 15 I. C. Gerber, J. G. Angyan, M. Marsman and G. Kresse, J. Chem. Phys., 2007, 127, 054101. 16 A. Savin, in Recent Advances in Density Functional Methods, Part I, ed. D. P. Chong, World Scientic, Singapore, 1995, p. 129. 17 T. Leininger, H. Stoll, H.-J. Werner and A. Savin, Chem. Phys. Lett., 1997, 275, 151. 18 H. Iikura, T. Tsuneda, T. Yanai and K. Hirao, J. Chem. Phys., 2001, 115, 3540. 19 R. Baer and D. Neuhauser, Phys. Rev. Lett., 2005, 94, 043002. 20 E. Livshits and R. Baer, Phys. Chem. Chem. Phys., 2007, 9, 2932. 21 E. Livshits and R. Baer, J. Phys. Chem. A, 2008, 112, 12789. 22 T. Gershon, R. Baer and L. Kronik, work in progress 2008. 23 J. E. Robinson, F. Bassani, R. S. Knox and J. R. Schreier, Phys. Rev. Lett., 1962, 9, 215. 24 T. Yanai, D. P. Tew and N. C. Handy, Chem. Phys. Lett., 2004, 393, 51. 25 T. Stein, L. Kronik and R. Baer, J. Am. Chem. Soc., 2009, 131, 2818. 26 S. Baroni, A. Dal Corso, d. G. S. P. Giannozzi, C. Cavazzoni, G. Ballabio, S. Scandolo, G. Chiarotti, P. Focher, A. Pasquarello, K. Laasonen, A. Trave, R. Car, N. Marzari and A. Kokalj, http://www.pwscf.org/. 27 J. P. Perdew and A. Zunger, Phys. Rev. B, 1981, 23, 5048. 28 V. I. Siklitsky, New Semiconductor Materials Characteristics and Properties, http://www.ioe.rssi.ru/SVA/NSM2005. 29 A. DalCorso, F. Mauri and A. Rubio, Phys. Rev. B, 1996, 53, 15638. 30 Y. Goldberg, ed. M. E. Levinshtein, S. L. Rumyantsev and M. S. Shur, John Wiley & Sons Inc, New York, 2001, p. 31.

31 I. Akasaki and M. Hashimot, Solid State Commun., 1967, 5, 851. 32 B. W. Henvis and M. Haas, J. Phys. Chem. Solids, 1962, 23, 1099. 33 Landolt-Bornstein New Series III/41 A1, ed. O. Madelung, U. Rossler and M. Schulz, Springer-Verlag, Berlin; New York, 2001. 34 P. A. Thiry, M. Liehr, J. J. Pireaux and R. Caudano, Phys. Rev B, 1984, 29, 4824. 35 C. Kittel, Introduction to solid state physics, Wiley, New York, 1966. 36 I. Vurgaftman, J. R. Meyer and L. R. Ram-Mohan, J. Appl. Phys., 2001, 89, 5815. 37 Ja. Vanvecht, Phys. Rev., 1969, 182, 891. 38 American Institute of Physics and D. E. Gray, American Institute of Physics handbook. Section editors: Bruce H. Billings [and others] Coordinating editor: Dwight E. Gray (McGraw-Hill, New York, 1972). 39 T. Soma, J. Phys. C: Solid State Phys., 1978, 11, 2669. 40 A. Polian and M. Grimsditch, Phys. Rev., 1999, B60, 1468. 41 E. S. Zouboulis and M. Grimsditch, J. Geophys. Res., 1991, 96, 4167. 42 S. Speziale, C. S. Zha, T. S. Duy, R. J. Hemley and H. K. Mao, J. Geophys. Res., 2001, 106, 515. 43 A. Dewaele, G. Fiquet, D. Andrault and D. Hausermann, J. Geophys. Res., 2000, 105, 2869. 44 B. P. Singh, S. K. Srivastava and K. Dinesh, Phys. B: Condens. Matter, 2004, 349, 401. 45 H. Spetzler, C. G. Sammis and R. J. Oconnell, J. Phys. Chem. Solids, 1972, 33, 1727. 46 A. Rubio, J. L. Corkhill, M. L. Cohen, E. L. Shirley and S. G. Louie, Phys. Rev. B, 1993, 48, 11810. 47 M. S. Hybertsen and S. G. Louie, Phys. Rev. B, 1986, 34, 5390. 48 M. S. Hybertsen and S. G. Louie, Phys. Rev. Lett., 1985, 55, 1418. 49 M. Rohlng, P. Kruger and J. Pollmann, Phys. Rev. B, 1993, 48, 17791. 50 R. W. Godby, M. Schlueter and L. J. Sham, Phys. Rev. Lett., 1986, 56, 2415. 51 R. W. Godby, M. Schlueter and L. J. Sham, Phys. Rev. B, 1988, 37, 10159. 52 F. Gygi and A. Baldereschi, Phys. Rev. Lett., 1989, 62, 2160. 53 X. Zhu and S. G. Louie, Phys. Rev. B, 1991, 43, 14142. 54 M. Usuda, N. Hamada, K. Shiraishi and A. Oshiyama, Phys. Status Solidi C, 2003, 0, 2733. 55 I. N. Remediakis and E. Kaxiras, Phys. Rev. B, 1999, 59, 5536. 56 L. Fritsche and Y. M. Gu, Phys. Rev. B, 1993, 48, 4250. 57 M. Stadele, M. Moukara, J. A. Majewski, P. Vogl and A. Gorling, Phys. Rev. B, 1999, 59, 10031. 58 Landolt-Bornstein New Series III/17a, ed. O. Madelung, Springer-Verlag, Berlin; New York, 1982.

4680 | Phys. Chem. Chem. Phys., 2009, 11, 46744680

This journal is

 c

the Owner Societies 2009