Year: 2025
Type: preprint
Abstract: Containers conveniently represent a wide class of inductive data types. Their derivatives compute representations of types of one-hole contexts, useful for implementing tree-traversal algorithms. Derivatives of discrete containers (whose positions have decidable equality) satisfy a universal property with respect to linear morphisms of containers. This induces a well-behaved notion of linear substitution for one-hole contexts. Working in Univalent Foundations, we extend the derivative operation to untruncated containers (whose shapes and positions are arbitrary types). We prove that this derivative, defined in terms of a set of isolated positions, satisfies an appropriate universal property in the wild category of untruncated containers and linear morphisms, as well as basic laws with respect to constants, sums and products. A chain rule exists, but is in general non-invertible. In fact, a globally invertible chain rule is inconsistent in the presence of non-set types, and equivalent to a classical principle when restricted to set-truncated containers. We derive a rule for derivatives of least fixpoints from the chain rule, and characterize its invertibility. All of our results are formalized in Cubical Agda.
Source: arXiv (Cornell University)
Authors: Joram, Philipp, Veltri, Niccolò
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7116961979
Cites:
Cited by:
Related to: