Artelys Knitro 12.3 solves nonlinear models with millions of constraints!

18 December 2020

— Artelys Knitro 12.3 release significantly improves the robustness of its algorithms for general nonlinear models.

The latest release of Artelys Knitro provides performance improvements on all types of optimization models (from linear and quadratic to general nonlinear models) allowing our customers to solve even larger problems with millions of constraints.

These problems are notably observed in several fields such as portfolio optimization in finance, trajectory optimization for robots or autonomous vehicles, economics or optimal power flow in energy.

This enhancement derives from two major developments:

  • An average performance improvement of 20% for large scale nonlinear models while using the default Interior/Direct algorithm. This improvement builds on the updates provided with Artelys Knitro 12.2 which refined internal linear systems for large scale models. These features have further been enhanced in this release, with a focus on numerical stability.
  • An average memory saving of 40% on all types of convex models when using the Interior/Direct algorithm.
  • In addition, users can now update the linear structure of their model (in the constraints and in the objective function) and re-solve without having to define a new Knitro model. This new feature is available for the linear structure of any general nonlinear model, enabling users to exploit advanced iterative approaches to solve particularly complex optimization problems.

    Additional features of Artelys Knitro 12.3:

  • Updated R, Java and C# interfaces. High speedups can be expected on all models with linear and quadratic structures as they are now processed through a dedicated API.
  • New C and Python single call functions to solve LP, QP and QCQP problems similar to functions already available for our Matlab users.
  • Memory savings up to 50% when using the least squares interface.
  • New diving heuristics to complement the feasibility pump heuristic for Mixed Integer Nonlinear Problems (MINLP).
  • General improvement on nonlinear models when using the BFGS Hessian approximation.
  • subscribe to our newsletters

    The collected data will be exclusively processed by the company Artelys for the purpose of keeping you informed about the services and products marketed by our company.


    © ARTELYS • All rights reserved • Legal mentions

    Pin It on Pinterest

    Share This