THE FREYD-MITCHELL EMBEDDING THEOREM

JULIAN KÜLSHAMMER

1. Definitions
Definition 1.1. A category A is a class of objects Ob(A) together with a set of
morphisms HomA (M, N ) for every pair of objects M, N ∈ Ob(A) and a map ◦ :
HomA (M, N ) × HomA (L, M ) → HomA (L, N ) for each triple of objects L, M, N ∈ Ob(A)
such that:
• ∃1M ∈ HomA (M, M ) such that 1M ◦ g = g and f ◦ 1M = f whenever this makes
sense, and
• (f ◦ g) ◦ h = f ◦ (g ◦ h) whenever this makes sense.
Example 1.2. Let R be a unitary ring. Then Mod R, the category of R-modules with
R-module homomorphisms is a category.
Definition 1.3. A category A is called small if Ob(A) is a set.
Definition 1.4. Let A, B be categories. A covariant (resp. contravariant) functor
F : A → B is given by an object F (M ) for every M ∈ A and for every morphism f ∈
HomA (M, N ) a morphism F (f ) ∈ HomB (F (M ), F (N )) (resp. F (f ) ∈ HomB (F (N ), F (M ))),
such that:
• F (1M ) = 1F (M )
• F (g ◦ h) = F (g) ◦ F (h) (resp. F (g ◦ h) = F (h) ◦ F (g)).
Example 1.5. Let R be a unitary ring, M an R-module. Then HomR (M, −) : Mod R →
Mod Z is a covariant functor.
Definition 1.6. (i) A category A is called preadditive if HomA (M, N ) is an abelian
group for all M, N ∈ Ob(A) and ◦ is bilinear.
(ii) An object 0 ∈ A is called zero object if for every object M ∈ A there is only one
morphism 0 → M and only one morphism M → 0. If there is a zero object, the
composition of M → 0 and 0 → N is also denoted by 0 for every pair M, N ∈ Ob(A).
(iii) For objects M, N ∈ A the product M × N is defined as the object M × N , such
that:
p2
M bEo p1 MO × /N
<
EE z
EE  h2 zzz
EE∃!h  zz
h1 E zz
X

Date: June 5, 2012.

1
2 JULIAN KÜLSHAMMER

(iv) A preadditive category is called additive if it has a zero object and M × N exists
for every pair M, N ∈ Ob(A).
(v) Let f : M → N . Then ker f , the kernel of f is defined by the following diagram:
f
ker fbE
i / MO / N
E
E
∃!h E
X
where f i = 0 and f g = 0.
(vi) Let f : m → N . Then Coker f , the cokernel of f is defined by the following diagram:
f
M / N
π / Coker f
v
v
g v
 {v v ∃!h
X
(vii) Let A be additive. A morphism i : M → N is called monomorphism if ig = 0
implies g = 0 for all g : L → M .
(viii) Let A be additive. A morphism p : M → N is called epimorphism if gp = 0 implies
g = 0 for all g : N → X.
(ix) A category A is called abelian if A is additive, every morphism has a kernel and a
cokernel and every monomorphism is a kernel and every epimorphism is a cokernel.
(x) Let A be abelian. Define Im(f ) := ker Coker f and Coim f := Coker ker f .
f g
(xi) A sequence L → M → N is called exact if ker g = Im f , i.e. Im f satisfies the
universal property of ker g:
h0/ ker π
X II
II h
II
II
II  i
f $ g
L /M / N
π

Coker f
where gi = 0 and gh = 0. A sequence · · · → Mi → Mi+1 → Mi+2 → . . . is exact if
every piece with three terms is exact.
(xii) A functor F : A → B is called left exact if for every exact sequence 0 → L → M →
N the sequence 0 → F (L) → F (M ) → F (N ) is exact.
(xiii) A functor F : A → B is called right exact if for every exact sequence L → M →
N → 0 the sequence F (L) → F (M ) → F (N ) is exact.
(xiv) A functor F : A → B is called exact if it is both left and right exact.
(xv) A functor F : A → B is called full if F : HomA (M, N ) → HomB (F (M ), F (N )) is
surjective.
 THE FREYD-MITCHELL EMBEDDING THEOREM 3

(xvi) A functor F : A → B is called faithful if F : HomA (M, N ) → HomB (F (M ), F (N ))

is injective.

2. Main result and how to apply it

Theorem 2.1 (Freyd-Mitchell embedding theorem). Let A be a small abelian category.
Then there is a unitary ring R and an exact fully-faithful functor F : A → Mod R.
Roughly speaking this theorem implies that every statement in an arbitrary abelian cat-
egory which can be stated in terms of diagrams can be proven via diagram chasing.

The next example is a bit cheating since we will use this lemma in the cause of the
proof of the embedding theorem, but it explains how to use the Freyd-Mitchell embedding
theorem in practice.
Lemma 2.2 (Snake lemma). Let

ψ
X / Y / Z / 0
f g h
 ϕ  
0 / L / M / N

be a commutative diagram with exact rows. Then there exists a morphism δ such that the
following sequence is exact:

ker f / ker g / ker h

δ / Coker f / Coker g / Coker h
Proof. First prove the lemma in the module category of a ring R via diagram chasing, i.e.
just taking elements and pushing them around with the maps, e.g. define δ(v) := ϕgψ −1 (v).
For proving this theorem in A, construct a small abelian category A our of the diagram
inductively, i.e. take the full subcategory of A with Ob(A1 ) := {X, Y, Z, L, M, N }, then in
each step insert all kernels and cokernels. The limit process yields a small abelian category.
Now apply F to A. But since F is fully faithful the sequence which we know exists in
Mod R comes from a sequence in A. Since F is exact, we have that F commutes with
kernels and cokernels and hence the sequence in A is also exact. 

The proof strategy of the embedding theorem is the following:

/ o R / /
A B I L Mod R

We will now explain what B, L and R are:

4 JULIAN KÜLSHAMMER

Definition 2.3. An abelian category B is called complete if arbitrary products (resp.

coproducts) exist, i.e. the universal properties of the following diagrams hold:

pi Q
Mi odH i∈I Mi
HH O
HH
HH ∃!h 
hi HHH 
X

resp.
ιi `
/
Mi H i∈I Mi
HH 
HH  ∃!h
HH
hi HHH 
$ 
X
Definition 2.4. (i) An object P ∈ B is called projective generator if HomB (P , −) is
exact and faithful.
(ii) An object I ∈ B is called injective cogenerator if HomB (−, I) is exact and faithful.
What is B? B = Fun(A, Mod Z) is the category of functors A → Mod Z (as objects)
and the following morphisms:
Definition 2.5. A natural transformation ϕ between two functors F, G : A → A0 is
given by a morphism ϕM : F (M ) → G(M ) for every M ∈ Ob(A) such that the following
diagram commutes for every f : M → N :

ϕM
F (M ) / G(M )
F (f ) G(f )
 ϕN

F (N ) / G(N )
What is L? This is just the subcategory of left exact functors. It has some sort of
”inverse” R and the nice property that it is complete and has a projective generator P . So
we will take R := End(⊕P )op . The following sections will now construct each of the given
functors.

3. The Yoneda lemma

Lemma 3.1 (Yoneda). Let F : A → Sets be a functor and let M ∈ A. Then there is an
isomorphism (in Sets) HomFun(A,Sets) (HomA (M, −), F ) ∼
= F (M ) given via ϕ 7→ ϕM (1M ),
which is natural in M and F , i.e. HomFun(A,Sets) (HomA (•, −), F ) ∼
= F (•) as functors
∼
C → Sets and HomFun(A,Sets) (HomA (M, −), •) = • as functors Fun(C, Sets) → Sets.
 THE FREYD-MITCHELL EMBEDDING THEOREM 5

Proof. We start with the injectivity of the given map. For a natural transformation consider
the following commutative diagram:
Hom(M,f )
Hom( M, M ) / Hom(M, N )
ϕM ϕN
 F (f ) 
F (M ) / F (N )
Hence we have F (f )(ϕM (1M )) = ϕN (Hom(M, f )(1M )) = ϕN (f ). Hence ϕN (and hence ϕ)
is determined by ϕM (1M ).
For the surjectivity we can use this law as the definition. Let x ∈ F (M ). We want
to define a natural transformation such that ϕ 7→ x. Define ϕN (f ) := F (f )(x). Then
ϕM (1M ) = F (1M )(x) = 1F (M ) (x) = x. We now have to prove that this law indeed defines
a natural transformation, i.e. that the following diagram commutes:
ϕN
Hom(M, N ) / F (N )
Hom(M,f ) F (f )
 ϕN 0 
Hom(M, N 0 ) / F (N 0 )
Indeed, we have F (f )(ϕN (g)) = F (f )F (g)(x) = F (f g)(x) = ϕN 0 (f g) = ϕN 0 (Hom(M, f )(g)).
Now we come to the naturalities. First convince yourself that for every f : M → N we
have that Hom(f, −) : Hom(N, −) → Hom(M, −) given by Hom(f, −)X = Hom(f, X) :
Hom(N, X) → Hom(M, X), g 7→ gf is a natural transformation. Thus the naturality is
given has follows:
ψ7→ψM (1M )
Hom(Hom(M, −), F ) / F (M )
ψ7→ψ◦Hom(f,−) F (f )
 ψ7→ψN (1N ) 
Hom(Hom(N, −), F ) / F (N )
Hence if we go first down and then to the right we get ψ 7→ ψN (Hom(f, N )(1N )) and if
we go first to the right and then down we get ψ 7→ F (f )(ψM (1M )). These two expressions
coincide since ψ is a natural transformation.
For the second naturality let ϕ : F → G and consider the following diagram:
ρ7→ρM (1M )
Hom(Hom(M, −), F ) / F (M )
ρ7→ϕ◦ρ ϕM
 ρ7→ρM (1M ) 
Hom(Hom(M, −), G) / G(M )
It is easy to see that ρ is mapped to (ϕ ◦ ρ)M (1M ) for both ways. 
Corollary 3.2. H : A → Fun(A, Mod Z), M 7→ Hom(M, −) is contravariant fully-faithful
left exact (we will see in the next chapter what the abelian structure on Fun(A, Mod Z) is).
6 JULIAN KÜLSHAMMER

Proof. It is fully-faithful since by the Yoneda lemma we have

HomFun(A,Mod Z) (Hom(M, −), Hom(N, −)) ∼= Hom(N, M ).
And left exact since Hom(−, X) is left exact for every X, hence if L → M → N → 0 is an
exact sequence, then 0 → Hom(N, −) → Hom(M, −) → Hom(L, −) is exact. 
But we have to solve the problem that this functor is not exact.

4. Nice properties of Fun(A, Mod Z)

Proposition 4.1. Let A be small abelian, then Fun(A, Mod Z) is abelian and complete.
Furthermore A has a projective generator.
Proof. This is true for Mod Z and everything can be defined componentwise, so it is true
for Fun(A, Mod Z). ` `
The
Q projective generator is given by M ∈A Hom(M, −). This is true because Hom( M ∈A Hom(M, −), •)
M ∈A •(M ) by the Yoneda lemma, so exactness and faithfulness are proven easily: If
F → G Q is an epimorphism,
Q then F (M ) → G(M ) is also an epimorphism for every M ,
hence F (M ) → G(M ) is an epimorphism, so via the Yoneda isomorphism we get that
the functor is exact. Similarly faithfulness follows since if we assume Qϕ, ψ : FQ→ G satisfy
Hom( Hom(M, −), ϕ) = Hom( Hom(M, −), ψ) this implies that ϕM ∼
` `
= ψM via the
Yoneda isomorphism. Hence ϕ = ψ. 
Definition 4.2. (i) A category B is well-powered if the family of subobjects of a given
object is a set. S `
(ii) Let Mi → M be subobjects, then I Mi is defined as the image of Mi → M .
(iii) Let I be a partially ordered set. A family of objects Mi → M is called linearly
ordered if for all i ≤ j we have a morphism πij : Mi → Mj , such that for every
i ≤ k ≤ j we have πkj πik = πij .
(iv) A complete well-powered category is called Grothendieck if for every linearly or-
dered
S familySof subobjects Mi of an object M and each other subobject N we have
N ∩ Mi = N ∩ Mi , where ∩ denotes the pullback of the two morphisms (one with
a minus sign).
Definition 4.3. Let A be an abelian category.
(i) An essential extension is a monomorphism M → N such that for every non-zero
monomorphism L → N the pullback of M → N and L → N is non-zero.
(ii) An injective envelope of M is an injective essential extension.
Proposition 4.4. The category Fun(A, Mod Z) is Grothendieck and has injective en-
velopes.
Proof. The well-powered-ness follows from the existence of a generator P , since for a sub-
object M 0 → M we have a monomorphism of sets Hom(P, M 0 ) → Hom(P, M ) and as P is
a generator, two different subobjects will yield different set monomorphisms.
The Grothendieck property follows pointwise from the fact that Mod Z is Grothendieck.
 THE FREYD-MITCHELL EMBEDDING THEOREM 7

The proof that this category has injective envelopes is more difficult and relies on the axiom
of global choice. 

5. Digression: Colimits and adjoint functors

Definition 5.1. Let I, A be categories. The colimit of a functor F : I → A is defined
as an object C = colimi∈I F (i) with maps ιi : F (i) → C such that for every α : j → i in
I we have ιj = ιi F (α) and the universal property that for all A ∈ A such that we have
morphisms fi : F (i) → A with fj = fi F (α) there exists a unique γ : C → A such that
fi = γιi for all i:
ιi
C o F (i)
{
 ∃!γ {{{
 {{{ fi
}{
A
Remark 5.2. (i) The coproduct is a colimit by taking for a set I the discrete category
(no non-identity morphisms) with objects i ∈ I. Then the diagram of the colimit is
the diagram of the coproduct.
(ii) The cokernel is a colimit by taking I = {0, 1, 2} with non-identity morphisms as given
in the following diagram:
0O

1 /2

Now a functor F can be given by mapping this diagram to the following diagram in
a category A with a zero object 0:
0O

f
M / N
Then ι0 = 0 and ι1 = 0 and ι2 = π : N → Coker f satisfy the given property and the
universal property is the same as for the cokernel as F (0) → F (1) = 0.
Definition 5.3. Two functors R : B → L and S : L → B are called adjoint if there is a
natural bijection in X and Y , Hom(R(X), Y ) ∼
= Hom(X, S(Y )).
Proposition 5.4. Left adjoint functors R preserve colimits, i.e. R(colim F ) ∼
= colim RF .
Proof. Let C = colim F . Then the diagram

C oaB F (i)
O
B BB
ιi
BFB (α)
ιj BB

F (j)
8 JULIAN KÜLSHAMMER

commutes and if we apply R to it we get the commuting diagram

RC cGo RF (i)
O
GGRιi
GG
GG (α)
RF
Rιj GG
RF (j)
Hence the first property is satisfied for the morphisms ι0i = Rιi . Next consider a diagram
RC o RF (i)
fi www
Rιi
ww
www
w{
A
Applying S we arrive at the diagram
SRC o SRF (i)
uu
SRιi
uuu
uu
zuu Sfi
S(A)
But SRM ∼ = M naturally by the adjunction homomorphism Hom(M, SRM ) ∼ = Hom(RM, RM )
(Check that it is an isomorphism!). By the universal property of the colimit, there now
exists a morphism γ : SRC → SA and applying R the morphism R(γ) makes the diagram
we started with commutative. 

6. Localizing subcategories
Definition 6.1. Let B be abelian. A non-empty subcategory C is called Serre subcate-
gory if for all exact sequences L → M → N we have M ∈ C iff L, N ∈ C.
Definition 6.2. Let C be a Serre subcategory of B and let f : M → N be in B. Then
f is called C-monomorphism if ker f ∈ C, C-epimorphism if Coker f ∈ C and C-
isomorphism if it is both a C-monomorphism and a C-epimorphism.
Definition 6.3. Let B be abelian. C ⊆ A be a Serre subcategory and L ∈ B. Then
B is called C-closed if for every C-isomorphism u : M → N we have that Hom(u, L) :
Hom(N, L) → Hom(M, L) is a bijection. The full subcategory of all C-closed objects is
denoted by L.
Definition 6.4. Let M ∈ B. A morphism f : M → L is called C-envelope if f is a
C-isomorphism and L is C-closed.
Definition 6.5. A Serre subcategory is called localizing if very object in B has a C-
envelope.
Theorem 6.6. Let C be a local subcategory of B, L ⊆ B the full subcategory of C-closed
objects. Then we have:
(i) The inclusion I : L → B has a left adjoint R : B → L. In particular R is fully-faithful.
 THE FREYD-MITCHELL EMBEDDING THEOREM 9

(ii) L is abelian (but in general not an abelian subcategory of B).

(iii) R is exact.
(iv) If B is complete, then L is complete.
Proof. (i) Define R on objects by choosing a C-envelope RM for every M ∈ B. Let uM :
M → RM . We have to prove that HomB (RM, L) ∼ = HomL (M, IL) ∼= HomB (M, L).
But this follows from the definition of C-envelope. Since HomB (um , L) provides an
isomorphism. In particular HomL (RM, RN ) ∼ = HomL (M, RN ), so there is a unique
map Rf making the following diagram commutative:

f
M / N
uM uN
 
RM _∃RF
_ _/ RN

(ii)-(iv) That L is closed under kernels can be seen as follows: Let 0 → X → L → L0 be exact
with L, L0 ∈ L. And let u : M → N be a C-isomorphism. Then there is the following
commutative diagram:

0 / Hom(M, X) / Hom(M, L) / Hom(M, L0 )

O O O

0 / Hom(N, X) / Hom(N, L) / Hom(N, L0 )

The 5-Lemma now implies that Hom(N, X) → Hom(M, X) is also an isomorphism

like the other vertical maps.
Next we want to prove that with B, also L is closed under products:
Let Li ∈ LQ and let u : M → N be a C-isomorphism. Then the following diagram
shows that Li is also in L:

/
Q Q
Hom(N, Li ) Hom(M, Li )
O


/
Q Q
Hom(N, Li ) Hom(M, Li )

As R is a left adjoint, we have that R preserves colimits. In particular cokernels and

coproducts. Hence it only remains to prove that R preserves kernels, then we are
done.
First note that via a dual argument (to the argument that R preserves cokernels) we
have that I preserves kernels.
f g
The key step is now to prove that IR preserves kernels. For that let 0 → M 0 → M →
M 00 → 0 be exact. Then we have Rg ◦ Rf = R0 = 0 as R is an additive functor. We
10 JULIAN KÜLSHAMMER

have the following commutative diagram with exact rows:

f g
0 / M0 / M / M 00 / 0
uM 0

0 uM uM 00
RM
 I II Rf
 II
∃ II
 II
$  Rg 
0 / ker Rg / RM / RN / 0
Denote the map M 0 → ker Rg by h. Then the Snake Lemma implies that we have
the following exact sequence:
0 → ker h → ker uM → ker uM 00 → Coker h → Coker uM → Coker uM 00
As C is a Serre subcategory we have that ker h and Coker h are like the other terms of
this exact sequence objects of C. Hence h is a C-isomorphism. By applying the Snake
lemma (Check!) we can somehow C-invert h to get that the map RM 0 → ker Rg is
also a C-isomorphism. Hence there is an exact sequence 0 → C → RM 0 → ker Rg →
D → 0 with C, D ∈ C. We now prove that C = 0: The fact that C ∈ C implies that
C → 0 is a C-isomorphism. Hence Hom(C, RM 0 ) ∼ = Hom(0, RM 0 ) = 0. And as the
sequence is exact we have C = 0. Thus we have a short exact sequence 0 → RM 0 →
ker Rg → D → 0. But this sequence splits as Hom(ker Rg, RM 0 ) → Hom(RM 0 , RM 0 )
is an isomorphism as RM 0 → ker Rg is a C-isomorphism. In particular, a splitting of
the sequence exists. This implies that D is a subobject of ker Rg, which is a subobject
of RM ∈ L and hence D = 0. Hence RM 0 → ker Rg is an isomorphism. Hence IR
preserves kernels.
But I is just the inclusion and preserves kernels as it is a right adjoint.

Note that by the proof of the theorem we have that RC = 0 for every C ∈ C since C → 0
is a C-envelope of C.
We now apply this general theory to the case we are interested, i.e. we let B = Fun(A, Mod Z).
Definition 6.7. A functor F ∈ Fun(A, Mod Z) is called weakly effaceable if for all
M ∈ A and each x ∈ F (M ) there is a monomorphism f : M → N such that F (f )(x) = 0.
Defnote the full subcategory of all weakly effaceable functors by C.
Proposition 6.8. The category of all weakly effaceable functors C is localizing.
Proof. left out 
Proposition 6.9. If C is the category of all weakly effaceable functors, then L, the category
of all C-closed objects is the category of left exact functors.
Proof. left out 
Proposition 6.10. The functor RH : A → Lex(A, Mod Z) is exact fully-faithful.
 THE FREYD-MITCHELL EMBEDDING THEOREM 11

Proof. RH is fully-faithful as it is the composition of two fully-faithful functors.

To prove that it is exact consider an exact sequence 0 → M 0 → M → M 00 → 0 in A.
Applying the Yoneda isomorphism H and extending by the image of the last map we get
an exact sequence in Fun(A, Mod Z):
0 → Hom(M 00 , −) → Hom(M, −) → Hom(M 0 , −) → Q → 0.
If we show that Q is weakly effaceable, then we are done, since by a previous remark, then
RQ = 0. Therefore pick x ∈ Q(X). Since Hom(M 0 , X) → Q(X) is surjective we can find
g ∈ Hom(M 0 , X) which is mapped to x via this map. Now form the pushout diagram:
0 / M0 / M / M 00 / 0
g h
 f 
0 / X / P0 / M 00 / 0
Applying Hom(−, X) and Hom(−, P 0 ) to the left part of the upper exact sequence we get
a commutative diagram
Hom(M, X) / Hom(M 0 , X) / Q(X) / 0

Hom(M, P 0 ) Hom(M 0 , P 0 ) / Q(P 0 ) / 0

Recall that (Hom(M 0 , X) → Q(X))(g) = x. Furthermore (Hom(M 0 , X) → Hom(M 0 , P 0 ))(g) =
f g. But f g ∈ Im(Hom(h, P ). As the rows are exact we get that (Hom(M 0 , P 0 ) →
Q(P 0 ))(f g) = 0. Thus by commutativity of the diagram we have (Q(X) → Q(P 0 ))(x) = 0.
Thus Q is weakly effaceable. Therefore RQ = 0 and hence RH is exact. 

7. Nice properties of Lex(A, Mod Z)

In the foregoing section we have already proven:
Proposition 7.1. The category Lex(A, Mod Z) is abelian and complete and RH : A →
Lex(A, Mod Z) is exact fully-faithful.
To construct an embedding of A into Mod R for some ring R in the next section we need
an additional property of Lex(A, Mod Z):
Proposition 7.2. Lex(A, Mod Z) has an injective cogenerator.
Proof. Again this proof will be left out here. The arguments are similar to those we have
already seen. 

8. The embedding into the module category

Theorem 8.1. Let L be a complete abelian category with a projective generator P . Let
A ⊆ L be a small subcategory. Then there is a fully-faithful exact embedding F : A →
Mod R for some ring R.
12 JULIAN KÜLSHAMMER

Proof. Consider the following commutative diagram for every A ∈ A:

o
`
Hom(P ,A) P tP  if t
tt
 ∃pA tttt
 tt f
yttt
A
We have that pA is an epimorphism: Assume that hpA = 0 for some h : A → B. Then
Hom(P , h)(f ) = hf = hpA if = 0 for all f . Hence Hom(P , h) = 0. But as P is a projective
generator, we have that Hom(P , −) is faithful,`hence h = 0.
Now take I = ∪A∈A Hom(P , A). Define P := I P . We claim that P is again a projective
generator and for every A ∈ A there exists an epimorphism P → A.
The last claim follows by the same argument as showing that pA was an epimorphism.
For showing that it is projective consider the following diagram:

P

 ιi
 
∃hi 

`
P
 I

 ∃h
f
 | 
X / Y / 0
g

The hi exist because P is projective and the h exists because of the universal property of
the coproduct. Hence the diagram commutes. In ` particular we have ghιi = ghi = f ιi and
as ιi is a monomorphism we have gh = f . Hence P is projective. `
For proving that it is a generator we have to prove that Hom( I P , −) is faithful, i.e.
` `
that the function Hom(M, N ) → Hom(Hom( P , M ), Hom( P , N )), f 7→ (g 7→ f ◦ g)
is injective, i.e. that for every f 6= 0 there exists g such that f ◦ g 6= 0. We have that
Hom(P , −) is faithful, i.e. for f ∈ Hom(M, N ) there exists g 0 : P → M such that f g 0 6= 0.
Fix a particular j ∈ I and define g by the universal property of the coproduct with respect
to the following diagram:
ιj ` ` ιi
P LLL / P i∈I\{j} P
o
rr P
LLL  rr
LLL
LLL  ∃g rrrrr
g0 LL&  rrrr 0
xr
M
Hence we have f gιj = f` g 0 6= 0 and as ιj is injective we have f g 6= 0.
We now let R := End( I P )op and define F : A → Mod R via M 7→ Hom(P, M ). We
check that it is well-defined:
• Hom(P, M ) is an R-module via r · f = f ◦ r, and
• Hom(P, f ) : Hom(P, M ) → Hom(P, N ) is an R-module homomorphism as r ·
Hom(P, f )(h) = (f h)r = f (hr) = Hom(P, f )(h ◦ r) = Hom(P, f )(r · h).
 THE FREYD-MITCHELL EMBEDDING THEOREM 13

It remains to prove that this functor is exact and fully-faithful. The exactness and the
faithfulness follow from the fact that P is a projective generator. So we are just left with
the fact that Hom(P, −) is full.
To prove this take an arbitrary R-module homomorphism α : Hom(P, M ) → Hom(P, N ).
We have proven that there exist epimorphisms P → M and P → N . Define K := ker(P →
M ). Then we have the following commutative diagram:
0 / Hom(P, K) / R / Hom(P, M ) / 0

∃α0 α
 
R / Hom(P, N ) / 0
The α0 exists because R is a projective R-module. But Rop ∼
= End(R) via r 7→ ρr , the right
multiplication with r. Hence α0 = ρx for some x ∈ R = End(P )op . Consider the following
diagram:
0 /K /P /M /0

x

P /N /0

Applying Hom(P, −) to the composition K → P → P → N we get zero (see the above

diagram). Hence as Hom(P, −) is faithful the composition K → P → P → N has to be
zero. Hence the universal property of the cokernel implies the existence of a map h, such
that the following diagram commutes:
0 / K / P / M / 0
x

  ∃h
P / N / 0
Now apply Hom(P, −) to get the diagram
0 / Hom(P, K) / R / Hom(P, M ) / 0
α0
 
R / Hom(P, N )
The right-most vertical map making the diagram commutative can be either α or Hom(P, h).
But R → Hom(P, M ) is an epimorphism, hence α = Hom(P, h). 

9. Literature
The following references might be useful to learn more about the Freyd-Mitchell embed-
ding theorem and categories in general. I used parts of them for this paper:
• Aly: Abelian categories and the Freyd-Mitchell embedding theorem, 2008
• Keller: Introduction to abelian and derived categories
• Mitchell: The full imbedding theorem, 1964
14 JULIAN KÜLSHAMMER

• Freyd: Abelian categories, 1964

• Swan: Algebraic K-theory, 1964
• Mac Lane: Categories for the working mathematician, 1969-1997
• Gabriel: Des categories abeliennes, 1962