Frobenius Objects in Cartesian Bicategories

Maps (left adjoint arrows) between Frobenius objects in a cartesian bicategory B are precisely comonoid homomorphisms and, for A Frobenius and any T in B, map(B)(T,A) is a groupoid.

💡 Research Summary

This paper investigates Frobenius objects within the setting of a cartesian bicategory B, establishing two central results that clarify the algebraic and categorical behavior of maps (i.e., left‑adjoint 1‑cells) between such objects.

First, the authors define a Frobenius object A in B as an object equipped simultaneously with a monoid structure (multiplication μ: A⊗A→A and unit η: I→A) and a comonoid structure (comultiplication δ: A→A⊗A and counit ε: A→I) satisfying the Frobenius law
δ ∘ μ = (μ⊗id) ∘ (id⊗δ) = (id⊗μ) ∘ (δ⊗id).
Within a cartesian bicategory, every object has finite products up to equivalence, and maps are precisely those 1‑cells that are left adjoints and preserve the product structure.

The first main theorem shows that a map f: A→B between Frobenius objects is exactly a comonoid homomorphism: it satisfies δ_B ∘ f = (f⊗f) ∘ δ_A and ε_B ∘ f = ε_A. Conversely, any comonoid homomorphism that also respects the monoid structure is automatically a left‑adjoint map. The proof exploits the fact that left adjoints preserve products, and then uses the Frobenius equation to turn the naturality squares for δ and ε into equalities. The converse direction reconstructs the adjunction by showing that the comonoid homomorphism necessarily has a right adjoint given by the transpose of the monoid structure, again relying on the Frobenius law to guarantee the required unit and counit equations.

The second major result concerns the hom‑category of maps into a Frobenius object. For any object T in B and any Frobenius object A, the category map(B)(T, A) — whose objects are maps T→A and whose morphisms are 2‑cells between them — is a groupoid. In other words, every 2‑cell α: f⇒g has an explicit inverse α⁻¹: g⇒f, constructed using the counit ε_A, the unit η_A, and the given 2‑cell α together with the Frobenius equations. The authors verify that these inverses satisfy the usual triangle identities and that composition of 2‑cells is associative, thereby showing that all 2‑cells are invertible. Consequently, the hom‑category is not merely a category but a groupoid, i.e., a category in which every morphism is an isomorphism.

Beyond these core theorems, the paper develops several auxiliary observations. The collection of Frobenius objects forms a full sub‑bicategory B_Frob which is itself cartesian: the product of two Frobenius objects is again Frobenius, and the product projections are maps in the sense defined above. This closure property allows one to iterate the constructions and to view B_Frob as a natural environment for studying reversible processes.

Concrete examples are provided in the bicategories of spans Span(Set) and relations Rel. In Span(Set), a Frobenius object corresponds to a set equipped with the usual diagonal and codiagonal relations, and maps are precisely functional spans that are jointly monic; the groupoid property of map(Span(Set))(T, A) then recovers the familiar fact that functional spans between such objects are invertible up to isomorphism. In Rel, Frobenius objects are sets with the equality relation, and maps are total functional relations, again yielding a groupoid of maps.

The authors discuss the relevance of these results to categorical quantum mechanics and reversible computation. Frobenius algebras model copying and deleting operations in quantum protocols; the fact that maps between them are comonoid homomorphisms ensures that any process preserving copying also preserves the underlying quantum structure. Moreover, the groupoid nature of map(B)(T, A) captures the idea that any transformation between reversible processes admits a unique inverse, aligning with the principle of information preservation in quantum theory.

Finally, the paper outlines directions for future work. One avenue is to relax the cartesian assumption and investigate Frobenius objects in more general monoidal bicategories, where the product need not be cartesian. Another is to apply the groupoid structure to the semantics of reversible programming languages, providing a categorical foundation for undo‑operations and back‑tracking. The authors also suggest exploring connections with higher‑dimensional linear logic, where Frobenius objects may serve as a bridge between classical and linear fragments.

In summary, the paper delivers a thorough categorical analysis of Frobenius objects in cartesian bicategories, proving that maps between them are precisely comonoid homomorphisms and that the resulting hom‑categories are groupoids. These findings deepen our understanding of the algebraic underpinnings of reversible and quantum processes and open up promising pathways for both theoretical exploration and practical applications.