Artelys Knitro 12.2: solve your nonlinear problems faster than ever!

• Up to 60% performance improvement on customer benchmark for very large-scale general nonlinear instances (more than 100 000 variables) solved with Knitro interior point algorithms.
• Average performance improvement of 50% on medium sized general nonlinear instances (between 1000 and 10 000 variables). This enhancement derives from the refinement of Knitro Sequential Quadratic Programming (SQP) algorithm which proves particularly useful for customers solving problems with expensive function evaluations (e.g. complex nonlinear expressions, external simulator, black-box model, etc.).
• Up to 75% performance improvement on very large-scale unconstrained models (more than 250 000 variables) such as machine learning applications. This significant enhancement derives from the refinement of Knitro quasi-Newton methods (namely the dense and limited-memory Quasi-Newton BFGS).

• Default parallelism when using the multi-start, multi-algorithm, or tuner features.
• Average performance improvement of 14% on general MINLP instances.
• Support for ARM processors! Artelys Knitro can now be used on embedded systems. Contact us for more information.
• Significant speedup can be observed on some Optimal Power Flow (OPF) instances using a new presolve option.
• An updated C++ interface offering efficiency improvements, particularly when solving models with quadratic structures (e.g. QPs and QCQPs).
• Increased performance on problems with complementarity constraints (i.e. MPECs) when using quasi-Newton methods.
• Faster resolution of non-smooth unconstrained models (such as machine learning applications) relying on the new weak Wolfe linesearch.