Georgia Institute of Technology College of Sciences

Individual chemical reactions between molecules are inherently stochastic, although for a large collection of molecules, the overall system behavior may appear to be deterministic. When deterministic chemical reaction models are sufficient to describe the behavior of interest, they are a compact way to describe chemical reactions. However, in other cases, these mass-action kinetics models are not applicable, such as when the number of molecules of a particular type is small, or when no closed-form expressions exist to describe the dynamic evolution of overall system properties. The former case is common in biological systems, such as intracellular reactions. The latter case may occur in either small or large systems, due to a lack of smoothness in the reaction rates. In both cases, kinetic Monte Carlo simulations are a useful tool to predict the evolution of overall system properties of interest. In this talk, an approach will be presented for generating approximate low-order dynamic models from kinetic Monte Carlo simulations. The low-order model describes the dynamic evolution of several expected properties of the system, and thus is not a stochastic model. The method is demonstrated using a kinetic Monte Carlo simulation of atomic cluster formation on a crystalline surface. The extremely high dimension of the molecular state is reduced using linear and nonlinear principal component analysis, and the state space is discretized using clustering, via a self-organizing map. The transitions between the discrete states are then computed using short simulations of the kinetic Monte Carlo simulations. These transitions may depend on external control inputs―in this application, we use dynamic programming to compute the optimal trajectory of gallium flux to achieve a desired surface structure.

One of the basic problems arising in many pure and applied areas of mathematics is to solve a system of polynomial equations. Numerical Algebraic Geometry starts with addressing this fundamental problem and develops machinery to describe higher-dimensional solution sets (varieties) with approximate data. I will introduce numerical polynomial homotopy continuation, a technique that is radically different from the classical symbolic approaches as it is powered by (inexact) numerical methods.

The purpose of this talk is to highlight some versions of the Krein-Rutman theorem which have been widely and deeply applied in many fields (e.g., Mathematical Analysis, Geometric Analysis, Physical Sciences, Transport theory and Information Sciences). These versions are motivated by optimization theory, perturbation theory, bifurcation theory, etc. and give rise to some simple but useful comparison methods, in ordered Banach spaces, such as the Dodds-Fremlin theorem and the De Pagter theorem.

We discuss joint work with J.-M. Martell, in which werevisit the ``extrapolation method" for Carleson measures, originallyintroduced by John Lewis to proveA_\infty estimates for certain caloric measures, and we present a purely real variable version of the method. Our main result is a general criterion fordeducing that a weight satisfies a ReverseHolder estimate, given appropriate control by a Carleson measure.To illustrate the useof this technique,we reprove a well known theorem of R. Fefferman, Kenig and Pipherconcerning the solvability of the Dirichlet problem with data in some L^p space.