Numerical methods for large-scale minimization

Jean Charles Gilbert (1)

1 INRIA Rocquencourt

This paper recalls some basic tools for unconstrained minimization problems. It includes a comparison of performance of automatic differentiation codes and methods when they are used to compute a gradient. Some algorithms for minimizing a function with a large number of variables are reviewed and the impact of automatic differentiation on them is discussed.

Mots clés: Automatic differentiation; Conjugate gradient; Large-scale optimization; Limited memory quasi-Newton method; Partitioned quasi-Newton method; Unconstrained optimization

Domaines: Mathématiques [math]

Jean Charles Gilbert. Numerical methods for large-scale minimization. Numerical Methods in Atmospheric Models, ECMWF, Sep 1991, Reading, Berkshire, United Kingdom. pp.93-113. ⟨hal-04140523⟩