Enabling reduced-order dynamics for graphics, haptics and sound
D. James (CS, Cornell)

Reduced-order dynamics methods hold promise for approximating the behavior of complex N-dimensional discrete systems, such as deformable objects, in parsimonious, low r-dimensional subspaces (r << N). Ideally, this model reduction can be exploited to provide a principled speed-accuracy trade-off, e.g., for interactive PDE applications.  Unfortunately, numerous complications arise in practice: the lack of fast N-independent subspace integration algorithms; kinematic and material nonlinearities; high subspace rank; and the "chicken and egg" problem of determining the subspace before you determine the solution.  This talk will describe our on-going work on leveraging reduced-order dynamics for novel applications in computer animation, haptic force-feedback rendering, and physically based sound rendering.