Artelys Knitro 13.1: solving MINLP faster than ever!

5 July 2022

— We are pleased to announce that Artelys Knitro 13.1 is now available! Check out its improvements, especially on mixed-integer nonlinear problems (MINLP).

Mixed-integer nonlinear problems offer a very natural way to model many engineering problems where nonlinear functions can capture the complex behavior of physical processes and integer variables can be used to model discrete decisions.

Artelys Knitro 13.0 introduced a new parallel branch and bound to solve our customers’ most complex MINLP. Artelys Knitro 13.1 pushes it further by adding several key features:

  • New heuristics to quickly find good feasible solutions.
  • Possibility for the user to pass an initial integer solution. This solution can even be repaired automatically if infeasible.
  • Introduced an automatic restart functionality for higher performances. This allows a MINLP solve to reset the tree search back to the root node after a partial search and restart with an optimized set of solver parameters and cutoff values.
  • Significant performance improvements across all discrete and continuous algorithms.

Relying on these new features dedicated to MINLPs, Artelys Knitro 13.1 is 3 times faster on average than version 13.0 on the MINLP benchmarks such as MINLPLib2.

Artelys Knitro 13.1 also introduces a new noise estimation routine for nonlinear objective functions to adapt the derivatives approximation. This is particularly useful when solving black-box optimization problems and relying on an external simulator (network model, finite element methods, chemical reaction, etc.) in which numerical inaccuracies can lead to noisy black-box optimization models.

Additional features of Artelys Knitro 13.1:

  • 40% performance improvement on Apple Silicon thanks to the integration of new specialized mathematical subroutines.
  • Updated Knitro-MATLAB interface to support the new MATLAB “problem-based” interface.
  • New heuristic for solving models with complementarity constraints (MPEC) when using the barrier/interior-point algorithm.
  • General performance improvements on nonlinear continuous models in particular when using BFGS or L-BFGS Hessian approximation.

If you want to know more about Artelys Knitro, you can contact us or visit the dedicated web page.

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