EUROPT 2024
Abstract Submission

190. Random Newton-type Iterations with Application to Electronic Structure Determination

Invited abstract in session WE-5: Randomized optimization algorithms part 1/2, stream Randomized optimization algorithms.

Wednesday, 14:10 - 15:50
Room: M:N

Authors (first author is the speaker)

1. Titus Pinta
Mathematics, ENSTA

Abstract

We combine two separate theories in the context of molecular electronic structure determination. On the one hand, we present a general framework of Newton-type methods for root-finding problems, with constraints, that characterizes classes of functions for which superlinear or better convergence is guaranteed. On the other hand we present a theory of random function iterations, where the functions are randomly selected mappings generated from familiar optimization problems. Putting these two ideas together yields a Markov operator whose kernel is given by the Newton-type mapping of a random variable to the critical point set of a randomly generated optimization problem. The "fixed points" that we seek are invariant measures of the corresponding Markov operator. In the context of molecular electronic structure determination, the optimization problem is a maximum liklihood estimation problem. We present the foundations of both theoretical threads and preliminary results of their combination toward reconstructing molecular electronic structures from X-ray free electron laser measurements.

Keywords

Status: accepted


Back to the list of papers