Release 0.4.1

Release Notes for Minotaur v0.4.1

Features:

• Simple presolve for all handlers.
• Products of binary variables and products of a binary variable with continuous variable are linearized.
• Fix and propagate heuristic.
• Linear constraints are classified based on specific structures.
• Cut selection and filtering of simplex based cuts.
• Quadratic functions are stored as a dictionary of keys instead of computational graphs for all solvers.
• Root relaxer can provide a feasible solution to the solution pool.
• Heuristics to run on original problem instead of transformed problem.
• Option to assume a problem is nonconvex and hence disable convexity detection.
• mglob will call mbnb for convex mixed-binary qp.
• Heuristic to find solution based on lock numbers.

Fixes:

• Fixes in LP Engine when objective is absent.
• Fixed a bug in calculating problem size when a new constraint is added.
• Fixed a numerical issue while OBBT.
• Fixed variable addition in AMPL interface.
• Bug fixes in nonlinear presolve.
• Checking feasibility of solution provided by NLP engine when no candidates to branch in mglob.
• Fix in calculating upper bound of the tree after root relaxation is created.
• Correctly do backward bound propagation for product of variables.
• Added default bounds in transformer.

Release 0.4

Release Notes for Minotaur v0.4

• Several enhancements to mglob, the solver for global optimization of quadratic optimization including convexity detection, heuristics, RLT cuts and bound tightening
• Bug fixes and improvements in interfacing with AMPL and Pyomo
• Upgraded third-party software
• Solvers available out-of-the box include
  • mqg for Quesada-Grossmann (LO/NLO) on convex MINLO
  • mbnb for Nonlinear Branch-and-Bound on convex MINLO
  • mglob for branch-and-reduce on nonconvex MIQCQO
  • mmultistart heuristic for nonconvex MINLO
• Updated log format
• Updated website
• Minotaur is now a part of Coin-OR

Regards
Mustafa Vora, Krushna Salunke, Mohan Katta and Ashutosh Mahajan

Release 0.3

---

Release Notes for Minotaur v0.3

New features

• a new shared-memory parallel version of Quesada-Grosmann algorithm (mqgpar) with
  extra linearization schemes
• a new solver, mglob for global optimization of MIQCQPs
• new solvers moa (outer-approximation) and mlstoa (lazy cuts based
  single-tree outer-approximation) for convex MINLP using Cplex MILP solver

Libraries, binaries and interfaces

• merged multiple source and interface related libraries into a single
  (libminotaur) library
• nl files can be written
• mps files (linear only) can be read
• C-interface with Cplex's LP and MILP solvers (CplexEngine)

Improvements and bug fixes

• several bugs fixed in the nonlinear branch-and-bound (mbnb) and
  Quesada-Grossmann (mqg) solvers
• removed all boost shared pointers; using normal pointers now
• website moved to https://minotaur-solver.github.io/

Source code

• moved the subdirectory algorithms in src/ to solvers
• consolidated interfaces to third-party software like ASL,
  LP, MILP and NLP subsolvers in src/interfaces

Third-party software

• upgraded third-party software and their dependencies

Regards
Meenarli Sharma, Prashant Palkar, Mustafa Vora and Ashutosh Mahajan

MPC

This release marks the software version corresponding to the paper "Minotaur: A Mixed-Integer Nonlinear Optimization Toolkit".