Release · 1.1
SHOT 1.1 released
SHOT can now solve its own integer-fixed NLP problems, partition nonseparable quadratics, and handle semicontinuous variables and special ordered sets.
SHOT 1.1 is available on GitHub.
Major new features
- SHOT no longer needs an external NLP solver in its primal strategy, since it can call itself to solve the integer-fixed NLP problems. Activate this with
Primal.FixedInteger.Solver=2. For nonconvex problems, an external NLP solver is still recommended. - Convex nonseparable quadratic functions can be partitioned into separate constraints with a reformulation based on an eigenvalue decomposition. Activate this with
Model.Reformulation.Quadratics.UseEigenValueDecomposition=true. - An initial polyhedral approximation of the nonlinear feasible set can be computed before feasibility-based bound tightening. Activate this with
Model.BoundTightening.InitialPOA.Use=true. - Support for semicontinuous and semi-integer variables.
- Support for special ordered sets.
Minor improvements and bug fixes
- Supporting hyperplanes can be generated for the entire nonlinear feasible set instead of the feasible sets of the individual constraint functions. Activate this with
Dual.ESH.Rootsearch.UseMaxFunction=true. - Improved support for passing nonconvex quadratic functions directly to the MIP solver, if it supports them.
- Bug fixes in the AMPL interface.