Commutative monads as a theory of distributions

The theory of commutative monads on cartesian closed categories provides a framework where aspects of the theory of distributions and other extensive quantities can be formulated and some results proved. We make explicit a link between our theory and the theory of Schwartz distributions of compact support. We also discuss probability distributions.

💡 Research Summary

The paper proposes a categorical framework for “distributions” based on commutative monads on a cartesian closed category E. The author begins by recalling the classical notion of a distribution as a continuous linear functional on a space of test functions (Schwartz distributions of compact support) and points out that the usual construction relies on a double‑dualization process in functional analysis. To avoid this analytic dependence, the paper introduces a monad T = (T, η, μ) on E, interpreting the object T(X) as the space of “distributions” on X, with the unit η_X :X → T(X) playing the role of the Dirac delta at x.

A key technical ingredient is the notion of strength (or “tensorial strength”) for the endofunctor T. In a cartesian closed setting, for any objects X, Y ∈ E there is a natural map

 t″_{X,Y} : X × T(Y) → T(X × Y)

which is required to satisfy coherence conditions expressing the strong naturality of the monad multiplication μ and unit η. The paper shows that t″ is 2‑linear: it is linear in the second argument when the second argument carries a T‑algebra structure. Moreover, t″ is initial among all such 2‑linear maps, i.e. any 2‑linear map f : X × T(Y) → B (with B a T‑algebra) uniquely factors through t″ via a T‑linear extension. This universal property follows from the free‑algebra nature of T(X) and the strong naturality of η and μ.

Using the same ideas, the author defines a partial linear extension: given any ordinary map f : X × Y → B, there exists a unique 2‑linear map \tilde f : X × T(Y) → B extending f through η_Y. Dually, there is a unique 1‑linear extension through η_X. These extensions are expressed explicitly using the strength t″ and the algebra structure on B.

From these partial extensions the paper constructs the Fubini maps

 ⊗ : T(X) × T(Y) → T(X × Y)

and its “dual” \tilde⊗. The map ⊗ is obtained by first extending the map t′_{X,Y} : T(X) × Y → T(X × Y) (the 1‑linear counterpart of t″) to a 2‑linear map, and similarly for \tilde⊗. In general the two extensions need not coincide; the monad T is called commutative precisely when they do (i.e. ψ = \tildeψ). When this commutativity holds, ⊗ equips T with a monoidal structure, making T a strong monoidal functor. The equality of the two Fubini maps is interpreted as a categorical version of Fubini’s theorem for iterated integrals.

The paper then introduces an integration pairing

 ⟨ –, – ⟩ : T(X) × (X ⊸ B) → B

where X ⊸ B denotes the internal hom (the space of “test functions” from X to a T‑algebra B). This pairing is defined as the 1‑linear extension of the evaluation map ev : X × (X ⊸ B) → B along η_X. Consequently, for a “distribution” P ∈ T(X) and a “test function” φ ∈ X ⊸ B, the value ⟨P, φ⟩ behaves exactly like the integral ∫_X φ dP. The pairing satisfies an extraordinary naturality in X: for any f : Y → X, ⟨T(f)(P), φ⟩ = ⟨P, φ ∘ f⟩, mirroring the change‑of‑variables formula for integrals.

Finally, the author discusses probability distributions. When T is instantiated as the probability‑measure monad (or any commutative monad modeling stochastic processes), the unit η gives Dirac measures, the Fubini map ⊗ realizes the product of independent random variables, and the pairing ⟨ –, – ⟩ yields expectations. Thus the categorical framework simultaneously captures Schwartz distributions, Radon measures, and probability measures, all as instances of the same abstract structure.

In summary, the paper demonstrates that commutative monads equipped with appropriate strength provide a unified, purely categorical foundation for various notions of distribution. It replaces the analytic double‑dual construction with universal properties of monads, recovers linearity, bilinearity, Fubini’s theorem, and integration in a coherent way, and shows how probability theory fits naturally into this picture. This approach opens the door to applying categorical logic and algebraic techniques to problems traditionally treated with functional analysis.