Slide 9 of the 2020 deck: LBB derives no cuts from node subproblems and shares nothing between nodes — pure B&B. Add cut generation at nodes (OA cuts for convex, MC++ affine cuts for nonconvex, reusing LOA/GLOA machinery) with a tree-wide cut pool valid across siblings (cuts from a node's relaxation are valid for its subtree; disjunct-local placement makes them globally reusable).
Depends on: B2 (sane node handling), Pyomo#4014 (disjunctive bounds for cut strength), B6 (strengthening machinery). Novel algorithm — publishable alongside the implementation.
Part of roadmap epic #2 (item B3).
Slide 9 of the 2020 deck: LBB derives no cuts from node subproblems and shares nothing between nodes — pure B&B. Add cut generation at nodes (OA cuts for convex, MC++ affine cuts for nonconvex, reusing LOA/GLOA machinery) with a tree-wide cut pool valid across siblings (cuts from a node's relaxation are valid for its subtree; disjunct-local placement makes them globally reusable).
Depends on: B2 (sane node handling), Pyomo#4014 (disjunctive bounds for cut strength), B6 (strengthening machinery). Novel algorithm — publishable alongside the implementation.
Part of roadmap epic #2 (item B3).