Colloquium Schedule for 2000-1 |

Numerical Variational Analysis


Dr. Asen L. Dontchev
Department of Mathematics
University of Michigan at Ann Arbor




For deriving error bounds or estimating convergence rates,the classical numerical analysis extensively uses the Taylorexpansion. I will talk about variational problems with constraints to which, as a rule, Taylor's theorem cannot be applied. First, a"neoclassical" interpolation problem will be discussed: find a convex function which interpolates given points and has a minimal L2 norm of the second derivative; by duality this problem reduces to a system of equations involving nonsmooth functions. A Newton-type methodfor this system was proposed in 1985 but its quadratic convergence was provedonly recently.

Next, a Runge-Kutta approximation to a boundary-value problem coupledwith a variational inequality is considered. Tight error bounds will bepresented depending on the regularity of the solution. The third topic is aboutconditioning of systems of inequalities and variational inequalities. I will show that Renegar's distance to infeasibility is equal to the reciprocal of the modulusof metric regularity.