Computational Mathematics at Sandia National Laboratories hosted by UNM Mathematics and Statistics

Event Type: 
Seminar
Speaker: 
Denis Ridzal
Event Date: 
Monday, February 29, 2016 - 3:30pm
Location: 
SMLC 356
Audience: 
General PublicFaculty/StaffStudentsAlumni/Friends

Event Description: 

Title:

Optimization with simulation constraints: Formulations, numerical algorithms, and applications at Sandia National Labs

Abstract:

Optimal design, optimal control and inverse problems are at the core of practice of engineering. In many cases, they can be stated elegantly and succinctly as optimization problems with partial differential equation (PDE), differential algebraic equation (DAE) or other simulation constraints. At the same time, their numerical solution poses a number of interesting mathematical, algorithmic and software challenges.
In this talk, we review recent research in simulation-constrained optimization at Sandia Labs. We focus on algorithmic challenges that are unique to our large-scale modeling and simulation environments, and present formulations, methods and software that enable the solution of optimization problems with billions of design parameters and thousands of uncertain model inputs. Central to the methods’ effectiveness is the rigorous use of inexact computations, e.g., enabled through adaptive numerical models and iterative linear system solves. An efficient software realization is based on matrix-free programming interfaces and available in Sandia’s Rapid Optimization Library (ROL). We conclude the presentation with numerical results in the areas of optimal control of radiative heat transfer, inverse problems in structural dynamics, and calibration of electrical circuit models.
 
Tea, Coffee, Cookies and conversation at 3:00pm in the lounge

 

Event Contact

Contact Name: Daniel Appelo

Contact Phone: (505) 277-3310

Contact Email: appelo@math.unm.edu