yoneda lemma

References

Summary

Presheafs

Presheafs are basically generalizations of right modules. So a presheaf F :β€„π’žop → Set can be thought of as a right representation of π’ž, or as a right π’ž-module structure on Set, where the action of an "element" (ie a morphism) f is fβ€…β‹…β€…x = F(f)(x) for all xβ€„βˆˆβ€„F(f)s.

In the case that we take functors BG → Vec, we get ordinary representation theory of groups, as left k[G] -modules. If we do it with presheaves, we get right modules.

This further implies that topos theory is a generalization of representation theory.

Definitions

Yoneda Embedding

This is essentially the regular representation

Density Theorem

All presheafs are colimits of representable presheaves This is basically saying that all modules are sums and quotients of regular representations

Irreducibles

You can decompose presheafs, like you would ordinary representations. This is sheafification

Giraud's Theorem

Subobject Classifier

This is basically Schur's lemma

If you have two irreducible representations V, W, then $$ \operatorname{Mor}(V, W) \cong \begin{cases} k, \text{if} V \cong W, \\ 0, \text{else} \end{cases}. $$

Mor Functors

Mor functors are very nice to work with; in some ways they're some of the simplest functors, so it's nice to be able to understand them.

β€’

Intuition: There should be something underneath the Yoneda Lemma that (to me) is more important: that natural transformations out of mor functors are completely determined by where they send the identity.

There should be some generalization of this

Definitions

Representable Functors

My Definition: A functor F :  ⁀ C →  ⁀ D is representable, if there is some object A and map $f_A: \newmoon \to F(A)$ such that if Ξ·, θ : F ⇒ G and Ξ·AfA = θAfA then η = θ

So in other words, the functor is rigid in some sense; natural transformations are determined by where they send a single element of the functor

This looks like enough to prove the yoneda lemma. I think that you can generalize $\newmoon$ to an object X, as long as you get some outer horn filling condition: for any two maps X → F(B) and X → G(B), you can make the diagram commute with a map F(B) → G(B). This holds if X is the terminal object, and maybe this characterizes the terminal object; I'll have to see

This also seems related to universal morphisms; so maybe that's what I'm looking for

Results

representable iff satisfies the Yoneda Lemma

\item

Surjective: Let $\alpha: \newmoon \to G(A)$. Then we have

$\newmoon \xrightarrow{x_F} F(A) \xrightarrow{F(f)} G(A)$

Use the fact that it's the terminal object to get Ξ·AxF = α

F(A) & F(B)
G(A) & G(B)

We can use the terminal object to demonstrate $\eta_B f = G(f) \alpha: \newmoon \to G(A)$

\item Say F satisfies the Yoneda Lemma. Then we have for any functor, G, [F,G] ≅ G(A)

Let F, G :  ⁀ C →  ⁀ D

They are both representable with the same object iff they're naturally isomorphic

Representable: We have F, G with the special object AF ≅ AG and special morphisms $x_F: \newmoon \to F(A_F)$ and $x_G: \newmoon \to G(A_G)$.

[F,G] ≅ G(AF) and [G,F] ≅ F(AG). Since we have an iso α : AF → AG, we have that F(Ξ±) : F(AF) → F(AG) is an iso. Now the map $\newmoon \xrightarrow{x_F} F(A_F) \xrightarrow{F(\alpha)} F(A_G)$ must correspond to some natural transformation Ξ·F : F ⇒ G (since F satisfies the Yoneda Lemma). Similarly, $\newmoon \xrightarrow{x_G} G(A_G) \xrightarrow{G(\alpha)^{-1}} G(A_F)$ gives us a natural transformation Ξ·G : G ⇒ F

Let $F: \cat{C} \to \textsc{Set}$ be representable in my definition. Then it is representable

(iff?)

In this setting the special map fA is really just an element. Let A be the special object of A and let xβ€„βˆˆβ€„F(A) be the special element. The Yoneda lemma will hold for F, so $[F, \mor{A}] = \mor{A}{A}$

And $[\mor{A}{-}, F] = F(A)$

Now consider the natural transformation corresponding to xβ€„βˆˆβ€„F(A). The natural transformation is Ξ·B(f) = F(f)(x)

I think we can show that if two functors satisfy the yoneda lemma, they're naturally isomorphic

Takeaway: This gives a different perspective to the Yoneda Lemma (and generalizes to categories that have a terminal object). It shows that representable functors are rigid, in the sense that some special element of them determines how every natural transformation out of it acts

$$\mor{Y_A}{F} \cong F(A)$$

YYAF = F(A)

$Y_A^B \equiv \mor{A}{B}$ $Y_A^{F(-)} \equiv \mor{A}{F(-)}$

Let $F: \cat{C} \to \textsc{Set}$ and $A \in \obj_{\cat{C}}$. Then

Ο†: [, F] F(A)

is a bijection, with Ο•(Ξ·) = ηA(idA).

Also: If you consider functors $G,H: \cat{C} \times \textsc{Set}^{\cat{C}}$ with

G(A, F) &= [, F]
H(A,F) &= F(A)

Then Ο• is a natural isomorphism in A and F.

Well-defined: Let $\eta \in [\mor{A}{-}, F]$ be a natural transformation. Then $\eta_A: \mor{A}{A} \to F(A) \in \obj_{\textsc{Set}}$. So Ο•(Ξ·) = ηA(idA)β€„βˆˆβ€„F(A).

Injective: Say Ο•(Ξ·) = ηA(idA) = ϕ(ΞΈ) = θA(idA). Now let $B \in \obj_{\cat{C}}$. We need to show that Ξ·B = θB. Now we have

Ξ·B, ΞΈB: β†’ F(B)

Let f : A → B.

Natural transformation: Because Ξ·, θ are natural transformations, we get the commutative square

So if we take f : A → B then $\mor{A}{-}(f): \mor{A}{A} \to \mor{A}{B}$. And thus from the above commutative square, we get $\eta_B \mor{A}{-}(f) = F(f)\eta_A$ and $\theta_B \mor{A}{-}(f) = F(f) \theta_A$. Now apply these maps to idA. We have $\eta_B \mor{A}{-}(f)(id_A) = \eta_B(f \circ id_A) = \eta_B(f) = F(f)\eta_A(id_A) = F(f) \theta_A(id_A) = \theta_B \mor{A}{-}(f)(id_A) = \theta_B(f \circ id_A) = \theta_B(f)$. So

Ξ·B(f) = ΞΈB(f)

for all f : A → B. So Ξ·B = θB. And since this hold for all $B \in \obj_{\cat{C}}$ we have η = θ. So we are injective.

Surjective: Now let xβ€„βˆˆβ€„F(A). Define a natural transformation by Ξ·A(idA) = x and for any $f \in \mor{A}{B}$ let Ξ·B(f) = F(f)(x).

This is a natural transformation: Consider the square

Let $g \in \mor{A}{A}$. We have $\eta_B \mor{A}{-}(f)(g) = \eta_B(f \circ g) \equiv F(f \circ g)(x) = (F(f) \circ F(g)) (x) = F(f)(F(g)(x))$. And F(f)Ξ·A(g) ≑ F(f)(F(g)(x)). So the above square commutes. Thus we are surjective.

So we have a bijection.

Natural isomorphism: We have two types of squares to check:

\item Natural in $F$: We need

Now let f : X → Y and η : F ⇒ F. Then the upper arrow on the diagram is

\item Natural in $A$:

Let $F: \cat{C}\to \textsc{Set}$ be contravariant. Then

[,F] β‰… F(A)

This is basically just by dualizing everything; a contravariant functor $F: \cat{C} \to \textsc{Set}$ corresponds to a covariant $F^{op}: \cat{C}^{op} \to \textsc{Set}$ (this is why I usually just write $F: \cat{C}^{op} \to \textsc{Set}$ for contravariant functors). And $\mor{A}{-}^{op} = \mor{-}{A}$. So dualizing everything, you get the result.

Corollary: [Mor(βˆ’,A),Mor(βˆ’,B)] ≅ Mor(A,B)

The functor

Y-:  ⁀ C β†’ Set ⁀ C

is the covariant Yoneda embedding, that sends A → YA. And

Y-:  ⁀ C β†’ Set ⁀ Cop

What's the use? It turns out that cats of the form $\textsc{Set}^{\cat{C}}$ are really nice: they have all limits and colimits. So it can be really nice to work in these cats. But even better, it turns out that the Yoneda embedding is actually full and faithful (meaning it's essentially a bijection on the morphisms β€” in other words,  ⁀ C embeds into $\textsc{Set}^{\cat{C}}$; so it's a subcategory)

In this way: We can view $\textsc{Set}^{\cat{C}}$ and $\textsc{Set}^{\cat{C}^{op}}$ as ways of extending  ⁀ C so that it's complete and cocomplete.

Consider  ⁀ C⁀D. If  ⁀ D has (co)limits, then so does  ⁀ C⁀D

Let F :  ⁀ J →  ⁀ C⁀D. The limit is a functor. We want to show that

lim F ≅ lim F(βˆ’)

YΞ”(βˆ’)F

Ylim F ≅ Y

We're looking at

Ylim F ≅ Y*Yβˆ’F

Let $\mor{A}{-}: \cat{C} \to \textsc{Set}$. Consider the Yoneda bijection

[, ]

Define: Let $Y: \cat{C}^{op} \to \textsc{Set}^{\cat{C}}$ act as

Y(A) &=
Y(f: B β†’ A) &= Ο†-1(f)

Results

Yoneda Embedding is Fully Faithful

Consider the Yoneda embedding hβˆ’β€„:β€„π’žβ€„β†’β€„Setπ’žop. Applying the Yoneda lemma, we get Nat (hA,hB) ≅ Mor (A,B). Since the isomorphism in the category of sets is a bijection, this tells us that the Yoneda embedding is fully faithful.

Ninja Yoneda Lemma

Let hA = Mor(-, A) and hA = Mor(A, -)

When F is covariant ∫AhAβ€…βŠ—β€…F(A) ≅ F and ∫AHom(hA,F(A)) ≅ F

When F is contravariant ∫AhAβ€…βŠ—β€…F(A) ≅ F and ∫AHom(hA,F) ≅ F

Uses

Proving Isomorphisms

Often instead of proving A β‰… B directly, you can just prove that Mor(X,A) ≅ Mor(X,B) for all X, naturally. In other words, you show hA β‰… hB.

Let hA β‰… hB, witnessed by the natural transformation Ξ·. Consider Ξ·A: Mor(A, A) β†’ Mor(A, B) and Ξ·B-1: Mor(B, B) β†’ Mor(B, A). Now we take the diagram

Mor(A, A) β†’ Mor(B, A) Ξ·A ↓ ↓ Ξ·B Mor(A, B) β†’ Mor(B, B)

Mor(f, A) = - Λ† f, for any f: B β†’ A. Now taking Ξ·B-1(1B) as our f, we have Ξ·A(1A)β€…βˆ˜β€…Ξ·Bβˆ’1(1B) = (Mor(Ξ·Bβˆ’1(1B),B)∘ηA)[1A] = (Ξ·B∘Mor(Ξ·Bβˆ’1(1B),A))[1A] = ηBβ€…βˆ˜β€…1Aβ€…βˆ˜β€…Ξ·Bβˆ’1 = 1B. Similarly, Ξ·Bβˆ’1(1B)β€…βˆ˜β€…Ξ·A(1A) = (Mor(Ξ·A(1A),A)∘ηBβˆ’1)[1B] = ηAβˆ’1β€…βˆ˜β€…(Mor(Ξ·A(1A),B))[1B] = ηAβˆ’1β€…βˆ˜β€…1Bβ€…βˆ˜β€…Ξ·A(1A) = 1A. Thus Ξ·A(1A) and Ξ·B-1(1B) are inverse pairs, showing that A ≅ B.

The yoneda embedding hβˆ’ is fully faithful. We have hA ≅ hB. Then by the fact that fully faithful functors reflect isomorphisms, we get that A ≅ B.

Grothendieck Topology

Use the Yoneda embedding to send A β†’ hA. Then inside of the presheaf category, all limits and colimits exist. You perform the (co)limit there, and then check whether your result is representable. If it is, you drop back down into the original category.