VE10

Bound constrained nonlinear least squares with an emphasis on large-scale problems

VE10 is a line search method in which the search direction is obtained by a truncated conjugate gradient technique. The bounds are handled by bending the search direction on the boundary of the feasible domain. VE10 also contains provision for estimating gradients by finite difference, if they are unavailable, or to check the analytic gradients otherwise. It features both Newton and quasi-Newton algorithms.

As well as using the least-squares nature of the problem, VE10 exploits the partially separable structure of many large-scale problems to obtain good efficiency. In particular, it uses a Brown-Dennis partitioned updating technique as the quasi-Newton option for partially separable problems.

VE10 is a standard ANSI Fortran subroutine in double precision. Machine dependencies are restricted to arithmetic constants, which can easily be modified by the user.

Need more info?

Contact:

Ph. Toint 
Department of Mathematics 
FUNDP 
61 rue de Bruxelles 
B5000 Namur BELGIUM 
pht@math.fundp.ac.be

References:

Ph. L. Toint, On large-scale nonlinear least squares calculations, SIAM J. Sci. Statist. Comput. 8 (1987), pp. 416--435.

Ph. L. Toint, VE10AD: A routine for large-scale nonlinear least squares, Harwell Subroutine Library, 1987.

[ Optimization Software Guide | OTC Home Page | NEOS Server | NEOS Guide ]