An Affine Semigroup from Orbifold Boundary Conditions: cut, phylogenetic and hierarchical models in the unit-weight sector, and weighted configurations beyond them

What the math gives to ML

The paper identifies orbifold boundary-condition classes with affine semigroups of finite-group label configurations satisfying a local conservation law: each branch label lies in a prescribed subgroup and the labels sum to zero. The transferable asset is not the orbifold application itself, but a principled mixed-domain constraint that couples heterogeneous categorical variables while retaining an explicitly enumerable configuration set. This can be turned into a differentiable constrained fusion or MoE-routing layer whose support contains only valid tuples, reducing the joint state space and enforcing an exact symmetry or conservation invariant. The main risk is inductive-bias mismatch, so the construction should be tested on tasks with genuine compositional or modular structure rather than imposed universally.