We present a method of computing Casimir forces for arbitrary geometries, with any desired accuracy, that can directly exploit the efficiency of standard numerical-electromagnetism techniques. Using the simplest possible finite-difference implementation of this approach, we obtain both agreement with past results for cylinder-plate geometries, and also present results for new geometries. In particular, we examine a piston-like problem involving two dielectric and metallic squares sliding between two metallic walls, in two and three dimensions, respectively, and demonstrate non-additive and non-monotonic changes in the force due to these lateral walls.
PACS numbers:Casimir forces arise between macroscopic objects due to changes in the zero-point energy associated with quantum fluctuations of the electromagnetic field [1]. This spectacular effect has been subject to many experimental validations, as reviewed in Ref. 2. All of the experiments reported so far have been based on simple geometries (parallel plates, crossed cylinders, or spheres and plates). For more complex geometries, calculations become extremely cumbersome and often require drastic approximations, a limitation that has hampered experimental and theoretical work beyond the standard geometries.In this letter, we present a method to compute Casimir forces in arbitrary geometries and materials, with no uncontrolled approximations, that can exploit the efficient solution of well-studied problems in classical computational electromagnetism. Using this method, which we first test for geometries with known solutions, we predict a non-monotonic change in the force arising from lateral side walls in a less-familiar piston-like geometry (Fig. 2). Such a lateral-wall force cannot be predicted by "additive" methods based on proximity-force or other purely two-body-interaction approximations, due to symmetry, and it is difficult to find a simple correction to give a non-monotonic force. We are able to compute forces for both perfect metals and arbitrary dispersive dielectrics, and we also obtain a visual map of the stress tensor that directly depicts the interaction forces between objects.The Casimir force was originally predicted for parallel metal plates, and the theory was subsequently extended to straighforward formulas for any planar-multilayer dielectric distribution ε(x, ω) via the generalized Lifshitz formula [3]. In order to handle more arbitrary geometries, two avenues have been pursued. First, one can employ approximations derived from limits such as that of parallel plates; these methods include the proximity-force approximation (PFA) and its refinements [4], renormalized Casimir-Polder [5] or semi-classical interactions [6], multiple-scattering expansions [7], classical ray optics [8], and various perturbative techniques [9,10]. Such methods, however, involve uncontrolled approximations when applied to arbitrary geometries outside their range of applicability, and have even been observed to give qualitatively incorrect results [11,12]. Therefore, re...