Overview

A natural transformation is a map between functors. Its components compare the outputs for individual objects, but those comparisons must also respect every map between the inputs. Naturality converts an informal assertion such as “this construction is compatible with change of input” into a testable equation.

Prerequisites

Understand covariant functors and composition of arrows. The two functors must have a common source and a common target. A family of maps between their values has the right shape to be a transformation, but shape alone does not guarantee naturality.

Definition and conventions

For F,G:C→DF,G:\mathcal C\to\mathcal D, a natural transformation η:F⇒G\eta:F\Rightarrow G consists of arrows ηX:F(X)→G(X)\eta_X:F(X)\to G(X) for every object XX such that, for every f:X→Yf:X\to Y,

G(f)∘ηX=ηY∘F(f).G(f)\circ\eta_X=\eta_Y\circ F(f).

The subscript XX names a component. The double arrow records a transformation between functors rather than a morphism between objects of C\mathcal C. If every component is invertible, η\eta is a natural isomorphism. Componentwise invertibility without the naturality equation is insufficient.

Intuition

There are two routes from F(X)F(X) to G(Y)G(Y). One first compares the constructions at XX and then transports along ff; the other first transports inside FF and then compares at YY. Naturality says the routes agree, so the comparison does not depend on which route a calculation takes.

Examples and non-examples

On sets, let F(A)=A×AF(A)=A\times A and let GG be the identity functor. First-coordinate projections pA(a,b)=ap_A(a,b)=a form a natural transformation. Indeed, for any f:A→Bf:A\to B, both f∘pAf\circ p_A and pB∘(f×f)p_B\circ(f\times f) send (a,b)(a,b) to f(a)f(a).

For a non-example, use the category whose objects are sets of size two and whose morphisms are all functions between them. Choose one distinguished element aAa_A in each object and define ηA\eta_A to be the constant map with that value. This is a family of endomorphisms of the identity functor. At a two-element set AA, the swap map ss gives sηAs\eta_A constant at s(aA)s(a_A) but ηAs\eta_As constant at aAa_A. The maps differ, so the family is not natural.

Important properties

Natural transformations compose vertically by composing components. If η:F⇒G\eta:F\Rightarrow G and θ:G⇒H\theta:G\Rightarrow H, then (θη)X=θXηX(\theta\eta)_X=\theta_X\eta_X. Substituting the two naturality equations proves naturality of the composite.

If η\eta is natural and its components invertible, rearranging G(f)ηX=ηYF(f)G(f)\eta_X=\eta_YF(f) gives F(f)ηX−1=ηY−1G(f)F(f)\eta_X^{-1}=\eta_Y^{-1}G(f). Thus the inverse components are automatically natural. This is the proof that a componentwise isomorphism is an isomorphism in the functor category.

Connections and further reading

Natural transformations are morphisms in functor categories and supply the unit and counit of adjunctions. Yoneda shows that some transformations are determined by just one element at one object. Read Riehl, §1.4, Definition 1.4.1 and Li, the natural-transformation definition in “函子与自然变换”.