Third M.I.T. Conference on Computational Fluid and Solid Mechanics June 14–17, 2005  

Condor, a parallel, direct, constrained optimizer for highcomputing-load, black-box objective functions

F. Vanden Berghen*
Université Libre de Bruxelles, IRIDIA laboratory, Brussels, 1050 Belgium

  Full Text
This paper presents CONDOR, an algorithmic extension of Powell's UOBYQA algorithm ('Unconstrained Optimization BY Quadratical Approximation'). CONDOR stands for 'COnstrained, Non-linear, Direct, parallel Optimization using trust Region method for high-computing load noisy objective functions'. CONDOR opens new possibilities in the field of industrial shape optimization based on CFD (Computation Fluid Dynamic) codes or PDE (partial differential equations) solvers where the evaluations of the objective functions are very CPU intensive and very noisy. We also report comparative numerical results between UOBYQA, DFO and CONDOR. The experimental results are very encouraging and validate the approach. Finally, we present a new, free, easily comprehensible and fully stand-alone implementation in C++ of CONDOR.

Keywords:  Non-linear optimization; Lagrange interpolation; Trust region method; Parallel optimization; Constrained optimization; Noisy optimization; High-computing-load optimization

* Tel.: +32 (479) 992768; E-mail: