VOLUME 76, NUMBER 11 PHYSICAL REVIEW LETTERS 11 MARCH 1996 Constraint Satisfaction in Local and Gradient Susceptibility Approximations: Application to a van der Waals Density Functional John F. Dobson and Bradley P. Dinte Faculty of Science and Technology, Griffith University, Nathan, Queensland 4111, Australia (Received 31 October 1995) We show how charge conservation and reciprocity can be built into local density or gradient approximations for density-density response functions (susceptibilities). We apply these ideas to derive from first principles a variant of the Rapcewicz-Ashcroft formula for the van der Waals interaction. We also discuss how improved formulas may be obtained. PACS numbers: 31.15.Ew, 34.20.–b, 71.45.Gm Overall, electron density functional theory (DFT) has been gratifyingly successful in approximating the energies of inhomogeneous interacting electronic systems [1,2]. Part of this success is due to the satisfaction of the exchange-correlation (xc) hole normalization condition, often achieved with considerable effort [3]. The usual lo- cal density approximation (LDA) [4] and its various gra- dient extensions [2] do not, however, give an adequate description of dispersion or van der Waals (vdW) forces [5]. The approach to be introduced here both simplifies the problem of achieving hole normalization, and facili- tates the derivation of van der Waals functionals. The difficulty of describing vdW forces in the LDA or gradient approaches is not surprising since these forces depend on correlations between distant density fluctua- tions, which may be different from those in the uniform or near-uniform electron gas upon which the above approxi- mations are based. Rather general methods have been proposed [6,7] for treating these long-ranged correlations by explicit solution of nonlocal screening equations, while still making the local density approximation for a suit- able intermediate quantity. These methods are expected to work well for a wide variety of situations including both overlapping and nonoverlapping electron distribu- tions. Such approaches will usually require substantial computation, however. We therefore first consider a less ambitious problem, the efficient calculation of the vdW interaction between a pair of nonoverlapping neutral systems, using as input only the ground-state electron densities n 1 r and n 2 r of the two systems. Rapcewicz and Ashcroft [8] have al- ready given an expression for this limit of the vdW inter- action, using arguments based on Feynman diagrams and three-point functions, plus a conjecture about an appropri- ate average density computed between two distant points. We will here derive a similar but not identical expres- sion from a straightforward local density approximation for a suitable quantity. The result will depend crucially on satisfaction of some constraints, and the methods used to achieve this should be useful elsewhere. Consider a pair of nonoverlapping many-electron systems so that electrons in system 1 can be considered distinguishable from those in system 2. Under these circumstances it is well known that second-order per- turbation theory in the Coulomb interaction between the two systems yields a good approximation to the van der Waals or dispersion interaction. Zaremba and Kohn [9] reexpressed this second-order energy, without further approximation, in the form E 2 2 ¯ h 2p Z dr 1 dr 2 dr 0 1 dr 0 2 e 2 jr 1 2 r 2 j e 2 jr 0 1 2 r 0 2 j 3 Z ` 0 x 1 r 1 , r 0 1 , iux 2 r 2 , r 0 2 , iudu . (1) Here x 1 r , r 0 , vand x 2 r , r 0 , vare the exact density- density response functions (in the Kubo sense) of each separate system in the absence of the other. x 1 is defined by the linear density response dn 1 r exput of the electrons in system 1 to an externally applied electron potential energy perturbation dV ext r exput , dn 1 r Z x 1 r , r 0 , iudV ext r 0 dr 0 , (2) and similarly for x 2 . It is important to note that x 1 includes the electron-electron interaction amongst the electrons of system 1 to all orders, and similarly for x 2 . [Note also that, unlike Ref. [9], we have referred the space arguments of x 1 and x 2 in (1) to a common origin.] To simplify (1), we will approximate x 1 and x 2 via a form of local density approximation. That is, we will appeal to a simple model of the density-density response of a homogeneous electron gas, and modify it suitably to approximate the response of the inhomogeneous gas in each system. The simplest such homogeneous response model is a nondispersive form of hydrodynamics in which we eliminate the fluid velocity v between the linearized continuity equation, Newton’s second law (with no pres- sure term), and Poisson’s equation. In Fourier repre- sentation, this gives a density perturbation dnq, v x q, vdV ext q, vwhere dV ext is the bare external po- tential energy and x hom q ! 0, v q 2 n 0 mv 2 2v 2 P n 0  , (3) 1780 0031-90079676(11) 1780(4)$10.00 © 1996 The American Physical Society