Abstract Nonsense

Crushing one theorem at a time

Natural Transformations (Pt. I)

Point of Post: In this post we introduce the notion of natural transformations, natural equivalences, and prove some basic theorems concerning them. We also give a few examples of natural transformations between some specific functors, as well as cover the classic natural equivalences (e.g. the natural equivalence of the double dual functor and the identity functor on \mathbf{FinVect}_k).

\text{ }


\text{ }

In this post we discuss a notion in category theory which will finally make rigorous such wildly tossed around statements such as “____ is  naturally/canonically isomorphic to ___”. That phrase, or some variant thereof, is heard so often in elementary mathematics (e.g. basic group theory, even in books as basic as Pinter or Fraleigh, as well in non-algebraic books such as Simmons) but is rarely (if ever!) actually given meaning in those settings. So, what precisely does this mean? Often times one hears explanations of ‘canonically’ as meaning “without having to choose”. For example, an isomorphism in linear algebra is ‘canonical’ if we can formulate it without having to fix a basis in the source or target space. But, of course this has no meaning when one is dealing with isomorphisms in, say, \mathbf{Top}, as there are no choices ever really made. So, what does it mean for two things to be ‘naturally’ isomorphic? We begin by phrasing the problem in more categorical language. Often times one speaks of ‘natural isomorphisms’ between certain constructions. For example, in \mathbf{FinVect}_k (the category of finite dimensional vector spaces over some field k) it’s often stated that “every finite dimensional vector space V  is ‘naturally’ isomorphic to its double dual V^{\ast\ast}“. Really what is being said, in a sense, is that the construction mapping V to itself and the construction mapping V to V^{\ast\ast} are “the same”. Or, putting things in phrasing we know, that there is some notion of equivalence between the identity functor and double dual functor on \mathbf{FinVect}_k. What is this equivalence? While there are many explanations of the intuition behind the kind of equivalence we’re interested, there is a particularly beautiful description which, at the risk of alienating some readers, I’d like to explain. In topology given two maps f,g:X\to Y we call them homotopic if there exists a continuous path from f to g. Said differently, if there exists a continuous map F:X\times I\to Y (where I=[0,1]) such that F(x,0)=f and F(x,1)=g. One thinks of maps as being homotopic if one can nicely slide one path onto the other path without hitting any snags/tearing. One can think of this process as be indexed by time, and at each second there is a curve, and as time progresses this goes from looking like f to looking like g. Intuitively, natural transformations are the homotopies of category theory. Namely, a natural transformation from a functor F to a functor G is a ‘continuous’ (i.e. respects the category, doesn’t “tear” anything) deformation of the functor F into the functor G. Of course, doesn’t “tear” anything and respects the category should be further explained. By this we shall see, as I’d like to defer to the discussion below, an associated shifting of the “image” of F to the “image” of G, and by not “tearing” and “respecting the category” we mean that this diagram commutes–that we can figure out information in the image of G by working through the image of F and this transformation between them.

\text{ }

Natural Transformations

\text{ }

Let \mathcal{C} and \mathcal{D} be two given categories and assume that that we have functors \mathcal{C}\overset{\displaystyle \overset{F}{\longrightarrow}}{\underset{G}{\longrightarrow}}\mathcal{D}. A natural transformation between F and G is a set \eta=\{\eta_X\} of morphisms \eta_X:F(X)\to G(X) for each X\in\text{obj}(\mathcal{C}) such that the following square is commutative for all X,Y and all morphisms X\xrightarrow{f}Y

\text{ }

\begin{matrix} F(X) & \overset{\eta_X}{\longrightarrow} & G(X)\\ _{F(f)}\big\downarrow & & \big\downarrow _{G(f)}\\ F(Y) & \overset{\eta_Y}{\longrightarrow} & G(Y)\end{matrix}

\text{ }

We call this the naturality condition. We notate the fact that \eta is a natural transformation from F to G by \eta:F\implies G. A natural transformation \eta:F\implies G is called a natural equivalence if F(X)\xrightarrow{\eta_X}G(X) is an isomorphism for each X\in\text{obj}(\mathcal{C}). If \eta:F\implies G is a natural equivalence, we are apt to notate this as \eta:F\cong G.

\text{ }

We see then that, as stated in the motivation, the set \eta of transformations creates a natural (no pun intended) “shifting” of the image of F to the image of G as the following diagram (copied [laboriously I might add] and badly from [1]) indicates:

\text{ }

\begin{matrix}  F(x) & & & \overset{\eta_x}{\longrightarrow} & G(x) & & \\  & \searrow^{F(g)} & & & &\searrow^{G(g)} \\  _{F(f)}\big\downarrow& & F(y) & \overset{\eta_y}{\longrightarrow} & \big\downarrow_{G(f)} & & G(y)\\  & \nearrow_{F(h)} & & & & \nearrow_{G(h)}& \\  F(z)& & & \overset{\eta_z}{\longrightarrow} & G(z) & & \\  & & & & &  \end{matrix}

\text{ }

Moreover, the naturality conditions is precisely the statement that this diagram commutes! Thus, the idea that this is a homotopy becomes more clear, even visually. Namely, we see that \eta serves to shift the image of F to the image of G just as a homotopy shifts curves to other curves, and the fact that this shifting jives with the category (i.e. makes the diagram commute) can be interpreted as the necessity of continuity of the shifting in homotopy.

\text{ }

Let’s take a look at some examples of natural transformations.

\text{ }

Suppose we fix a group G. We claim then that H\mapsto G\times H and A\xrightarrow{f}B\mapsto A\times G\xrightarrow{f\times1} B\times G defines a functor P_G:\mathbf{Grp}\to\mathbf{Grp}. Indeed, clearly P_G(1_A)=1_{A\times G} if A\xrightarrow{j}B\xrightarrow{k}C then we see that P_G(k\circ j)(a,g)=(k(j(a)),g)P_G(k)(j(a),g)=P_G(k)(P_G(j)(a,g)) and so P_G(k\circ j)=P_G(k)\circ P_G(j). We claim now that any group homorphism G\xrightarrow{f}H induces a natural transformation \eta:P_G\implies P_H with \eta_A:A\times G\to A\times H given by (a,g)\mapsto (a,f(g)). Indeed, we must check that for any two groups A,B and any homomorphism A\xrightarrow{j}B the equation P_H(j)\circ \eta_A=\eta_B\circ P_G(j) holds. To do this we just plug and chug: P_H(j)(\eta_A(a,g)))=P_H(j)((a,f(g))=(j(a),f(g)) and \eta_B(P_G(j)(a,g))=\eta_B((j(a),g))=(j(a),f(g)). Thus, \eta really is a natural transformation. Moreover, it’s clear that if f is an isomorphism then \eta:P_G\cong P_H. This then allows us to make such statements as, if A\cong G, B\cong H then A\times G is naturally isomorphic to B\times H.

\text{ }

We now show that “every group is naturally isomorphic to its opposite group , or in our new language the identity functor 1_{\mathbf{Grp}} is naturally isomorphic to the opposite functor F which sends G\mapsto G^{\text{op}} and every homomorphism to the associated antihomomorphism defined by f^{\text{op}}(g)=f(g)^{-1}. Indeed, for each group G define \eta_G:G\to G^{\text{op}} by g\mapsto g^{-1}–this is readily verified to be a group homomorphism. Since \eta_G is, in fact, an isomorphism, to prove it’s a natural equivalence it suffices to show that the naturality condition holds. This amounts to showing that f^{\text{op}}\circ\eta_G=f, but this is trivially true.

\text{ }

As a last classic example, we show that the identity functor is naturally equivalent to the double dual functor on \mathbf{FinVect}_k for some field k. Indeed, for each finite dimensional k-space V define \eta_V:V\to V^{\ast\ast} by \eta_V(v)=\text{ev}_v where \text{ev}_v(f)=f(v) for f\in V^\ast. It is a standard result of linear algebra that since V is finite dimensional, that \eta_V really is an isomorphism for each V. Thus, if we can show that \eta_V satisfies the naturality condition, it will automatically be a natural equivalence. To do this we must verify that given any two spaces V,W and any morphism V\xrightarrow{T}W we have that T^{\ast\ast}\circ\eta_V=\eta_W\circ T. To do this we pick v\in V and evaluate it at both elements to get T^{\ast\ast}(\eta_V(v))=T^{\ast\ast}(\text{ev}_v)=\text{ev}_v\circ T^\ast and \eta_W\circ T=\text{ev}_{T(v)}. To check that these are the same, we pick f\in W^\ast and check that (\text{ev}_v\circ T^\ast)(f)=\text{ev}_v(f\circ T)=f(T(v)) and \text{ev}_{T(v)}(f)=f(T(v)). Thus, we see that the naturality condition, and thus the fact that the identity functor and double dual functor are naturally equivalent follows.

\text{ }

We now show that there is a way in which we can compose natural transformations. Indeed, suppose that we have categories \mathcal{C,D} and functors F,G,H:\mathcal{C}\to\mathcal{D}. Suppose then that we have natural transformations \eta:F\implies G and \varepsilon:G\implies H. Then, we can form the composition \varepsilon\circ\eta:F\implies H  by the rule (\varepsilon\circ\eta)_A=\varepsilon_A\circ\eta_A or all objects A\in\text{obj}(\mathcal{C}). Note then that a natural transformation \eta:F\implies G is a natural equivalence if and only if there exists a natural transformation \varepsilon:G\implies F such that \varepsilon\circ\eta=1_F and \eta\circ\varepsilon=1_G where, in this context 1_F is the map which is identity on each F(X) and likewise for 1_G.

\text{ }

\text{ }


[1] Mac, Lane Saunders. Categories for the Working Mathematician. New York: Springer-Verlag, 1994. Print.

[2] Adámek, Jirí, Horst Herrlich, and George E. Strecker. Abstract and Concrete Categories: the Joy of Cats. New York: John Wiley & Sons, 1990. Print.

[3] Berrick, A. J., and M. E. Keating. Categories and Modules with K-theory in View. Cambridge, UK: Cambridge UP, 2000. Print.

[4] Freyd, Peter J. Abelian Categories. New York: Harper & Row, 1964. Print.

[5] Rotman, Joseph J. Introduction to Homological Algebra. Springer-Verlag. Print.


December 28, 2011 - Posted by | Algebra, Category Theory | , , ,


  1. […] Natural Transformations (Pt. II) Point of Post: This is a continuation of this post. […]

    Pingback by Natural Transformations (Pt. II) « Abstract Nonsense | December 28, 2011 | Reply

  2. […] theorem of bifunctors, is whether an assignment of mappings for bifunctors such that and are natural transformations  and is necessarily a bifunctor . This turns out to be the case, as the following […]

    Pingback by Product of Categories (Pt. II) « Abstract Nonsense | December 31, 2011 | Reply

  3. […] Theorem (Identity): Let be a ring and a left -module. Then, if we consider as an -bimodule in the usual way so that is considered as a left -module. Then, . Similarly, if is a right -module then as right -modules. Moreover, these isomorphisms are natural. […]

    Pingback by Basic Properties of the Tensor Product (Pt. I) « Abstract Nonsense | January 5, 2012 | Reply

  4. […] which to make the set of all functors into a category, with the morphisms being nothing more than natural transformations. This may not sound at all important at first, but in fact functor categories play a pivotal role […]

    Pingback by Functor Categories (Pt. I) « Abstract Nonsense | January 7, 2012 | Reply

  5. […] systems in over . Basically this entails showing that the morphisms in are really just natural transformations between the associated functors in the sense just described. Indeed, by definition a morphism […]

    Pingback by Preordered Sets as Categories, and their Functor Categories « Abstract Nonsense | January 10, 2012 | Reply

  6. […] range over the appropriate set of modules makes the corresponding set of isomorphisms and into natural equivalences between the associated […]

    Pingback by Adjoint Isomorphism for Tensor and Hom « Abstract Nonsense | January 10, 2012 | Reply

  7. […] as stated before, we can use comma categories to restate the definition of natural transformation. Indeed,  let and be categories with (indexed different just to tell them apart). Then, if we […]

    Pingback by Comma Categories (Pt. II) « Abstract Nonsense | January 10, 2012 | Reply

  8. […] differently? Well, I claim that an object of is an equalizer for if and only if the functor is naturally isomorphic to the contravariant Hom functor . Indeed, suppose for a second that was such an object. What we […]

    Pingback by Equalizers and Coequalizers (Pt. I) « Abstract Nonsense | February 22, 2012 | Reply

  9. […] be a finite dimensional vector space. Then, the map as defined above is a natural […]

    Pingback by Some Natural Identifications « Abstract Nonsense | August 14, 2012 | Reply

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: