KNITRO is an especially versatile nonlinear solver, using state-of-the-art algorithmic options to accommodate varied objective and constraint nonlinearities in continuous and integer variables. It is designed for large-scale problems with up to hundreds of thousands of variables.
Developer: Ziena Optimization LLC
Current version: 9.1
Problem types supported: Linear, quadratic, and general smooth nonlinear optimization in continuous and integer variables.
Special forms detected: Complementarity and equilibrium constraints using the AMPL “complements” operator.
Algorithms available: For continuous problems, a choice of direct interior-point, conjugate gradient interior-point, and sequential linear-quadratic active-set algorithms. For mixed-integer problems, a choice of general and special convex branch-and-bound procedures using any of the three continuous algorithms to solve subproblems.
Special features: Extensive use of shared-memory multi-core computing: concurrent optimization to determine the best choice among multiple algorithms; a parallel multistart procedure for finding high-quality locally optimal solutions to nonconvex problems; parallel linear algebra and finite-difference gradient computations. Options to keep iterates feasible with respect to bounds and inequalities.
Ziena Optimization LLC website