Core Concepts
The authors develop a novel approach to equality in substructural logic that preserves its natural quantitative interpretation as a non-trivial distance, using the categorical language of Lawvere's doctrines and graded modalities.
Abstract
The paper addresses the challenge of introducing equality in substructural logics, such as fragments of Linear Logic, in a way that supports a quantitative interpretation. In standard First Order Logic, equality is modeled by equivalence relations, which cannot be directly translated to the substructural setting as it would lead to undesirable properties like the ability to freely delete or duplicate formulas.
The authors propose a solution by working in a minimal fragment of Linear Logic enriched with graded modalities. This allows them to rephrase the substitutive property of equality in a resource-sensitive way, preventing the equality predicate from being duplicated. They introduce the notion of Lipschitz doctrine, which provides a sound and complete categorical semantics for this quantitative equality.
The key aspects of the work are:
Defining R-graded doctrines, which model the (⊗, 1)-fragment of Linear Logic enriched with graded modalities indexed by a resource semiring R.
Introducing R-Lipschitz doctrines as R-graded doctrines with a quantitative equality predicate satisfying Lipschitz-style conditions.
Presenting a core deductive calculus for quantitative equality and proving its soundness and completeness with respect to the R-Lipschitz doctrine semantics.
Analyzing the 2-categorical properties of R-Lipschitz doctrines, showing they arise as coalgebras for a 2-comonad and relating quantitative equality to the standard one defined by left adjoints.
Extending the results to richer fragments of Linear Logic up to full Linear Logic with quantifiers.
The work provides a formal framework for reasoning about quantitative equality in substructural settings, with potential applications in areas like program metrics, differential privacy, and quantitative semantics.
Stats
1 ⊢ t ≖ t
φ[t/x] ⊗ t ≖ u ⊢ φ[u/x]