For the sake of completenss I begin with

Def. If $S:D\rightarrow C$ is a functor and $c$ is an object of $C,$ a universal arrow form $c$ to $S$ is a pair $<r,u>$ consisting of an object $r$ of $D$ and arrow $u:c\to Sr$ of $C,$ such for every $d$ of $D$ and arrow $f:c\to Sd$ of $C,$ there is a unique arrow $f':r\to d$ of $D$ with $Sf'\circ u=f.$

I am interested in logical characterisation of universal arrow. That is, if we read the definition carefully we shall notice $\forall\exists !$ part and come up with

Thm. If $S:D\rightarrow C$ is a functor and $c$ is an object of $C,$ then $<r,u:c\to Sr>$ is universal arrow from $c$ to $S$ if and only if function sending each $f':r\rightarrow d$ into $Sf'\circ u$ is a bijection of hom-sets (this bijection will be denoted $\phi_d$) $$D(r,d)\cong C(c,Sd).$$ This bijection is natural in $d.$

This is precesely the first part of Mac Lane's Proposition 1. (from III.2). The naturallity in $d$ is clear. As a result we end up with natural isomorphism $\phi_d:D(r,d)\rightarrow C(c,Sd)$ of parallel functors $$D(r,\cdot),C(r,S\cdot):D\to Ens.$$

Second part of Mac Lane's Proposition 1. is following

Thm. Given functor $S:D\rightarrow C,$ object $c$ of $C,$ object $r$ of $D$ and natural isomorphism $\phi$ $$D(r,d)\cong C(c,Sd),$$ there is unique arrow $u:c\rightarrow Sr$ such that $<r,u>$ is universal arrow form $c$ to $S$ and $$\forall(f':r\to d)\hspace{10pt}\phi_d(f')=Sf'\circ u.$$

Proof is important, because in Yoneda Lemma Mac Lane is referring to it. So we define $u$ to be $\phi_r(1_r).$ Then we just take arbitrary $f':r\to d$ and by commutativity of diagram enter image description here

we see that $\phi_d(f')=Sf'\circ u.$

Question 1. Why is $u$ unique? I see no reason.

Then Mac Lane says

The argument for Proposition 1 rested on the observation that each natural transformation $\phi:D(r,\cdot)\to K$ is completely determined by the image under $\phi$ of the identity $1_r.$ This fact may be stated as follows:

And here goes the Yoneda Lemma

Lemma (Yoneda) If $K:D\to Set$ is a functor and $r$ is an object of $D,$ then there is a bijection $$y:Nat(D(r,\cdot),K)\cong Kr$$ which sends each natural transformation $\alpha$ to $\alpha_r(1_r)$ The proof is indicated by the following commutative diagramenter image description here

Here I see the same problem as previously. I take arbitrary natural transformation $\alpha$ then arbitrary $f:r\to d$ and by commutativity of above diagram I end up with $\alpha_d(f)=K(f)(\alpha_r(1_r)).$

Question 2. Where this bijectivity of $y$ comes from?


Question 1: Taking $d=r$ and $f'=1_r$, the condition $\phi_d(f')=Sf'\circ u$ says that $\phi_r(1_r)=S(1_r)\circ u=u$, since $S(1_r)=1_{Sr}$. So the only possible value of $u$ that could work is $u=\phi_r(1_r)$.

Question 2: You can recover $\alpha$ from $y(\alpha)$ via the formula $\alpha_d(f)=K(f)(y(\alpha))$, so $y$ is injective. It is also surjective, because for any $x\in K(r)$, you can just define $\alpha_d$ by the formula $\alpha_d(f)=K(f)(x)$, and you can easily check that these are the components of a natural transformation $\alpha$ such that $y(\alpha)=x$.

  • $\begingroup$ I felt like I miss something obvious and general. Namy if we have a function $f:A\to B$ and we are said that a=something(f(a)) (i.e $1_A=g\circ f$), then it implies injectivity of $f.$ However surjectivity of $y$ isn't so evident from MacLane's book. $\endgroup$ – Fallen Apart Nov 7 '15 at 20:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.