We are pleased to announce the release of Artelys Knitro 10.1. The substantial R&D effort made in recent months leads to many new great features!
Hence an innovative MISQP algorithm for solving nonlinear problems with integer variables is offered. This algorithm is particularly useful for the minimization of functions that can be evaluated only at certain points and through relatively expensive simulations in terms of computation time.
In addition, all the Artelys Knitro algorithms can now be directly called from R thanks to a new dedicated interface. Finally the version 10.1 brings numerous functional enhancements that make Artelys Knitro easier to use and more efficient. Among them, user options for scaling variables/constraints individually or for relaxing and reformulating binary/integer variables automatically are included.

New MISQP algorithm

New MISQP algorithm

• Designed for nonlinear mixed-integer models with possibly expensive function evaluations (typically black-box simulations).

• Complements the existing branch-and-bound methods.

• Handles functions that cannot be evaluated when integer variables take fractional values (“non-relaxable” integer variables).

• Requires fewer function evaluations to find a solution estimate.

• Can search for global solutions of nonconvex mixed integer models using a parallel multistart procedure.

 

New R interface

New R interface

• Enables R users to get started quickly with Knitro from R.

• Supports all Knitro features (including integer variables and complementarity constraints).

• Efficient and dedicated interface for nonlinear least-squares.

• Library of examples to get started.

 

Additional features:

• New custom scaling features: easy definition of scaling factors for variables, constraints and objective. Users may now easily modify problem scaling without reformulating their model.

• Automatic reformulation of binary variables as complementarity constraints.

• Ability to relax integer variables as continuous variables.

• Significant speed and robustness improvements when using BFGS or L-BFGS Hessian approximations with the IP/Direct interior-point method.

• MATLAB interface: improved memory management and ability to specify individual feasibility tolerances.

 

Interested in trying Artelys Knitro on your optimization problem? Download a free trial version here.