Large-Scale Scientific and Engineering Design Optimization

Description

Nonlinear optimization problems arise in a wide range of science and engineering applications at Sandia. These traditionally include such areas as optimal design and control, computational chemistry, and materials characterization. In addition, we are starting to encounter a new generation of problems arising from model-based safety assessment applications developed under the Accelerated Strategic Computing Program (ASCI). Optimization problems arising from these applications often require running a complex, PDE-based computer simulation in order to compute a function value. As a result, they present many challenges to optimization algorithms, including computationally expensive function evaluations, low-accuracy function values, and lack of analytic gradients. The same may be true of the constraints. In this situation, the constraints are likely to be nonlinear, and while there are good methods for bound and linear constraints, handling nonlinear constraints is a difficult problem. Furthermore, it may not be possible to quantify all of the constraints. For example, a function evaluation could fail for a particular set of parameters. This is a type of constraint that cannot be described mathematically and that must be handled on the fly. Other issues that are becoming increasingly important include non-convex functions, non-smooth functions, uncertainty in the input parameters, and multiple levels of parallelism.

Our goals are to address the challenges described here by developing nonlinear optimization algorithms that are both theoretically sound and computationally efficient and to incorporate them in object-oriented software libraries.

Software

• APPSPACK, a C++ implementation of asynchronous parallel pattern search
• OPT++, a C++ library of nonlinear optimization methods for engineering optimization

Project Members

• Paul Boggs
• Genetha Gray
• Patty Hough
• Tammy Kolda
• Kevin Long
• Monica Martinez-Canales
• Pam Williams

Students (Summer 2002)

• Sarah Brown, University of Maryland
• Sherae Daniel, University of Maryland
• Jill Reese, North Carolina State University

Related Links

• Global Optimization, Sandia's MICS-supported research on global optimization methods
• Discrete Optimization, Sandia's MICS-supported research on discrete optimization methods
• The DAKOTA Toolkit, Sandia's optimization framework
• Sandia MICS Program, an overview of Sandia's MICS-supported research projects
• Optimization Technology Center, a joint optimization research center between Argonne and Northwestern
• Computer Science Research Institute, an ASCI-supported institute that fosters collaborations between Sandia and universities

Contact

For more information, please contact Patty Hough (pdhough@ca.sandia.gov).