This proposal is concerned with simulation-based approaches to sound synthesis, in the interest of generating very high quality synthetic sound of a natural acoustic character—partly to emulate real instruments, but also to explore classes of sounds which cannot be produced using conventional synthesis methods, or acoustic instruments. A further goal is to introduce such physical modelling synthesis methods definitively into the world of electronic music, virtual environments, and to the greater public.Target systems to be studied and simulated include: highly nonlinear acoustic systems (brass instruments, and percussion based on plate and shell vibration; electromechanical instruments; full 3D acoustic spaces; embedding of instruments within 3D spaces in order to achieve fully virtual and spatialized audio; and finally modular connections of systems in order to allow the eventual user, a composer, an instrument design environment. Such complex systems, including strong nonlinear effects, have never before seen a rigorous exploration from a numerical synthesis perspective. This proposed project is of an interdisciplinary nature, and rooted in music, numerical analysis, time-domain simulation, and high-performance computing.Work will be carried out at various levels: a) theoretical work and time domain algorithm design, with special attention paid to the appropriate choice of model, efficiency and real time operation, and various issues critical in audio, including: adequate perceptual rendering of system responses at audio sample rates; aliasing; robust algorithm design, ensuring numerical stability under highly nonlinear conditions; and modular constructions. b) large-scale parallel implementations on multicore processors and general purpose graphics processing units (GPGPUs), and c) experimental testing through collaborative work with established composers of electronic music, leading to performances original multichanel and fully synthetic music.