First-principles molecular dynamics (FPMD) is emerging as a very
powerful atomistic simulation approach, which combines a classical
description of nuclei with a quantum mechanical description of
electrons. Recent advances in electronic structure methods, notably
Density Functional Theory (DFT), have made FPMD a truly predictive
approach, which provides information on the structural, dynamical and
electronic properties of a physical system. FPMD has been
succesfully applied to several areas of research in materials
science, chemistry and biochemistry. The computational cost of FPMD
simulations is high, due to the detailed description of electronic
structure that is required. Straightforward implementations of FPMD
incur a computational cost of O(N3) for N atoms. More recent
approaches have been proposed to reduce this cost to O(N). We will
present recent progress in the development of O(N) algorithms based
on real-space, finite-difference methods, as well as the challenges
that arise when implementing conventional O(N3) FPMD on large,
massively parallel computers.
