Real-space multigrid methods for large-scale electronic structure problems

We describe the development and applications of a new electronic structure method that uses a real-space grid as a basis. Multigrid techniques provide preconditioning and convergence acceleration at all length scales and therefore lead to particularly efficient algorithms. The salient points of our implementation include: (i) new compact discretization schemes in real space for systems with cubic, orthorhombic, and hexagonal symmetry and (ii) new multilevel algorithms for the iterative solution of Kohn-Sham and Poisson equations. The accuracy of the discretizations was tested by direct comparison with plane-wave calculations, when possible, and the results were in excellent agreement in all cases. These techniques are very suitable for use on massively parallel computers and in O(N) methods. Tests on the Cray-T3D have shown nearly linear scaling of the execution time up to the maximum number of processors (512). The above methodology was tested on a large number of systems, such as the C-60 molecule, diamond, Si and GaN supercells, and quantum molecular dynamics simulations for Si. Large-scale applications include a simulation of surface melting of Si and investigations of electronic and structural properties of surfaces, interfaces, and biomolecules. (C) 1997 John Wiley & Sons, Inc.