Derivative-free Nonlinear Programming

Joshua Griffin and Tamara Kolda

Project Goals
- Extend generating set search (GSS) methods for global optimization
- Practical methods that get good results in few function evaluations
Technical Approach
- Hybrid optimization
- Perform multiple optimizations in parallel with automatic load balancing

Framework for hybrid methods automatically load balances function
evaluationsand shares the information across processors

Many scientific applications use optimization for parameter tuning and design. Here
we see design of a microfluidic mixing device that used APPSPACK optimization.

Accomplishments
- Hybrid DIRECT + GSS is an extremely robust approach to global optimization
- Collaborating on incorporating other globalization schemes such as TGP
- Linear & nonlinear constraints in APPSPACK leading to publications in ACM TOMS, SIOPT, SISC (one accepted, one submitted)

(Return to Applied Math program list)