Operator Algebras and Dynamics Seminars
Fall 2019
All talks are from 3:45-4:45 p.m. in the Seminar room, unless otherwise specified.
-
Nov04
-
Norming for Discrete Crossed ProductsVrej ZarikianTime: 03:45 PM
-
Oct28
-
Unique Extension Problems for C*-InclusionsVrej ZarikianTime: 03:45 PM
-
Oct21
-
Mixing on random rank-one transformationsDarren CreutzUSNATime: 03:45 PM
-
Sep30
-
Staircase transformations are mixingDarren CreutzUSNATime: 03:45 PM
View Abstract
Ergodic theory includes, among other things, the study of iterated maps of [0,1] to itself in terms of probability. I will introduce various mixing notions that such maps can possess and discuss a method of constructing examples of maps with various mixing properties, culminating in a presentation of my work, joint with C. Silva, that general staircase transformations are mixing. Two weeks later I will continue the presentation on more recent work in these directions.
-
Sep16
-
On the Koopman operator of algorithmsFelix DietrichJohns Hopkins University & Princeton UniversityTime: 03:45 PM
View Abstract
Most numerical algorithms are acting iteratively on a state variable. In this case, it is possible to treat the algorithm as a dynamical system, where the variable parametrizing integral curves is the iteration number instead of the time. This conceptual shift allows us to employ tools usually used for the study of dynamical systems in the study of numerical algorithms. One powerful tool is the Koopman operator associated to the system, which encodes the evolution of observables on the state. The operator acts linearly on the function space of observables, and thus is amenable to spectral decomposition. Its spectrum allows insight into the qualitative behavior of the underlying system, and even quantitative predictions are possible if enough eigenfunctions of the operator are available. In this talk, we explore the qualitative properties of several important algorithms in numerical optimization through the spectrum of their associated Koopman operator. We provide analytical formulations of the spectrum and eigenfunctions for the Newton-Raphson algorithm on the complex plane, and discuss a numerical approximation of the continuous spectrum for chaotic algorithms. Literature: https://arxiv.org/abs/1907.10807
