UWashington Distinguished Seminar in Optimization and Data: Ting-Kei Pong

  • Date: 06/05/2023
  • Time: 15:30
Ting-Kei Pong, Hong Kong Polytechnic University

University of Washington


Error bounds for conic feasibility problems: case studies on the exponential cone


A linear conic programming problem aims at minimizing a linear objective over the intersection of an affine set and a closed convex cone. The conic feasibility problems we study in this talk naturally arise from the optimality conditions of linear conic programming problems, which boil down to finding a point in the intersection of an affine set and a closed convex cone. While the distance to the intersection gives a precise measure of proximity of a point to being feasible, this distance can be much harder to compute than the distances to the affine set and the cone respectively. Thus, establishing bounds on the distance to the intersection based on the latter two distances (a.k.a. the error bound problem) is instrumental in the design of termination criteria for conic solvers and the study of convergence rate of algorithms. In this talk, we present a general framework for deriving error bounds for conic feasibility problems. Our framework is based on the classical concept of facial reduction, which is a fundamental tool for handling degeneracy in conic programs, and a new object called one-step facial residual function. We develop tools to compute these facial residual functions, which are applicable even when the projections onto the cones under study are not easy to analyze. We illustrate how our framework can be applied to obtain error bounds for the exponential cone. Exponential cone is a recent addition to the MOSEK commercial conic solver and allows the modeling of optimization problems involving power, exponential, logarithmic and entropy functions in the objective or constraint set. If time permits, we will also use our new error bound results to derive interesting new results in the study of Kurdyka-Lojasiewicz property. This is joint work with Scott B. Lindstrom and Bruno F. Lourenco.

Other Information: 

Time: 3:30 pm Pacific

Location: Gates Commons (CSE 691) Allen Center


Seminar webpage: https://sites.math.washington.edu/~thomas/DSOS.html