Artelys Knitro 12.4 further improves the interface versatility

• Improvement on NLP:
  • New dynamic scaling of the limited-memory Quasi-Newton BFGS, more robust and efficient for all algorithms. This leads to an average performance improvement of 20% on general nonlinear instances when using this Hessian approximation.

 

• New Knitro API functions:
  • Allow deleting or changing linear or constant structures of a problem and then re-solving.
  • Allow to retrieve information about the feasibility of the variables and constraints after both the presolve and solving process, giving a detailed analysis of the complex infeasibilities in nonlinear problems.
  • Allow users to provide their own customized linear algebra solver, which is useful when the problem has a very specific structure and can be solved with highly specialized routines. This feature was recently applied with success to an industrial problem and enabled a 2x overall speedup.

• Improvement on MINLP:
  • Improved heuristics strategies
  • Early integer-feasible solution found on 15% more instances

 

• Newly supported platforms:
  • Support for the new Apple M1 chip