Correlated optimized effective-potential treatment of the derivative discontinuity and of the highest occupied Kohn-Sham eigenvalue: A Janak-type theorem for the optimized effective-potential model Mark E. Casida* De ´partement de Chimie, Universite ´ de Montre ´al, Case Postale 6128, Succursale Centre-Ville, Montre ´al, Que ´bec, Canada H3C 3J7 Received 27 July 1998 A Janak theorem is derived for the correlated optimized effective-potential model of the Kohn-Sham exchange-correlation potential v xc . It is used to evaluate the derivative discontinuity DDand to show that the highest occupied Kohn-Sham eigenvalue, H -I , the negative of the ionization potential, when relaxation and correlation effects are included. This reconciles an apparent inconsistency between the ensemble theory and fractional occupation number approaches to noninteger particle number in density-functional theory. For finite systems, H =-I implies that v xc =0 independent of particle number, and that the DD vanishes asymp- totically as 1/r . The difference in behavior of the DD in the bulk and asymptotic regions means that the DD affects the shape of v xc , even at fixed, integer particle number. S0163-18299904907-3 I. INTRODUCTION Hohenberg-Kohn-Sham density-functional theory 1,2 DFTis an important workhorse for electronic structure cal- culations in both physics and chemistry. Since the quality of the results is determined by the approximation used for the exchange-correlation functional, the development of im- proved practical, approximate functionals is of central im- portance in DFT. For example, the functionals widely used for molecular calculations yield exchange-correlation poten- tials v xc , whose asymptotic behavior is qualitatively incor- rect. This is critical for the calculation of high-lying bound excitations from time-dependent DFT Refs. 3 and 4and is important for other properties that are sensitive to the outer part of the charge density, such as static and dynamic polar- izabilities. Furthermore, the shortcomings of approximate functionals have meant that molecular first ionization poten- tials and band gaps in solids are, in practice, not computed from the Kohn-Sham eigenvalues but instead from total- energy-difference-based procedures and post-DFT Green- function approximations, respectively. Work on improving approximate functionals is often guided by properties dem- onstrated to hold for the exact functional. The concept that the exactDFT exchange-correlation energy E xc has a discontinuity in its derivative at integer particle electronnumber, v xc + r E xc r N+0 + E xc r N-0 + v xc - r, 1.1 was originally introduced to explain the discrepancy between calculated and measured band gaps in solid-state physics 5,6 and the dissociation of diatomic molecules into neutral atoms. 7–9 In contrast to the well-established place of the de- rivative discontinuity DDin the theory for infinite systems, the problem of the DD in finite systems has often been re- garded as abstruse, due to questions of consistency between different methods of introducing noninteger particle number into DFT, and as irrelevant for practical applications. How- ever, several recent works suggest that the DD should be taken into account in designing functionals with the correct asymptotic behavior. 10,11,3,12,13 Indeed, the derivatives with respect to particle number are intimately connected to the relation between the highest occupied Kohn-Sham eigen- value H and the ionization potential I and, hence, to the asymptotic value v xc lim r v xc ( r) of the exchange- correlation potential. The proof of these latter relations has been the subject of recent controversy. 14,9,15 This paper re- conciles the apparent inconsistencies in the DD obtained from different methods of handling noninteger particle num- ber and gives an independent proof confirming that H = -I . For finite systems, this means v xc =0 and that the DD affects the shape of v xc . Thus, a proper consideration of the DD is important for the design of improved practical func- tionals. There are two main approaches for introducing noninteger particle number into DFT. In the fractional occupation num- ber approach, the usual Kohn-Sham equation, which was de- rived only for integer occupation number, is used, but the orbital occupation numbers f i entering into the total-energy expression E =- 1 2 i f i i | 2 | i + v rrd r + 1 2  r 1 r 2 | r 1 -r 2 | d r 1 d r 2 +E xc 1.2 through the kinetic energy and the charge density ( r) = j f j | j ( r) | 2 are allowed to be fractional. In this formal- ism, Janak’s theorem states that E f i v = i , 1.3 wherever the derivative exists. 16 The fractional occupation number approach can be viewed as simply providing a well-defined smooth extension of the charge density and of the energy expression, which PHYSICAL REVIEW B 15 FEBRUARY 1999-I VOLUME 59, NUMBER 7 PRB 59 0163-1829/99/597/46945/$15.00 4694 ©1999 The American Physical Society