A numerical comparison between the LANCELOT and MINOS packages
	       for large-scale constrained optimization

 I. Bongartz  A.R. Conn N. I. M. Gould M. A. Saunders Ph. L. Toint

                          Report 97-13

We present the results of a numerical comparison of two nonlinear
optimization packages capable of handling large problems, MINOS 5.5
and LANCELOT (Release A).  The comparison was performed using over
900 constrained and unconstrained problems from the CUTE collection.

With the default options, LANCELOT makes use of first and second
derivatives, while MINOS requires gradients but cannot use higher
derivatives.

We conclude that LANCELOT is usually more efficient in terms of the
number of function and derivative evaluations.  If the latter are
inexpensive, MINOS may require less CPU time unless there are many
degrees of freedom.  LANCELOT proves to be less reliable than MINOS
on linear programming problems, but somewhat more reliable on
problems involving nonlinear constraints.