School of Mathematics and Physics – UNIVERSITY OF PORTSMOUTH

BASIC INTRODUCTION TO CATEGORIES

DR SERGI SIMON
Part II of module TOPICS IN ALGEBRA AND GEOMETRY
(M31448, year 2021/2022)

FURTHER READING:
Mac Lane, S., Categories for the Working Mathematician
Lang, S., Algebra (Graduate Texts in Mathematics, Springer 211)
Atiyah, M. et al, Introduction To Commutative Algebra
Adámek, J. et al, Abstract and Concrete Categories: The Joy of Cats
Simon, S., Algebra and Topology Lecture Notes (Moodle)

Portsmouth, November 29, 2021

I Introduction
● In Zermelo-Fraenkel Set Theory and its extensions (especially those due to Gödel) we have the
following notions which we will define very roughly:
– a class is any entity having members (i.e. in which the notion of belonging makes sense) with
some common property;
– a proper class is a class that is not a member of another entity.
● Obviously, in this context a set is a particular case of a class.
● A proper class is not a set (otherwise it would be a member of the larger class of all sets).
● More specific examples of classes that are not sets:
– C = { all sets }; otherwise the set A of all sets that are not members of themselves would
satisfy A ∈ A and A ∉ A at the same time, leading to the well-known Russell’s paradox.
– {all groups}: you can define a group structure on P (S ) for any set S, and it takes you back
to Russell’s paradox.
– {all rings}: same reason because you can define a ring operation on any power set of a set.

1
 Definition 1. A category C consists of: C
(i) a class Obj(C ) whose elements are called objects of the category C;
(ii) a class Hom(C ) of (homo)morphisms, arrows or maps between the objects.
Fixing any A, B ∈ Obj(C ), HomC (A, B ) is the class of all morphisms from A to B, written

A Ð→ B f :A→B
f
or for every f ∈ HomC (A, B );
A is called the domain or source and B is called the codomain or target of the morphism; if
(A, B ) ≠ (A′, B ′), then HomC (A, B ) ∩ HomC (A′, B ′) = ∅;
(iii) and a law of composition for any A, B, C ∈ Obj(C )

HomC (A, B ) × HomC (B, C ) /

HomC (A, C )
(f, g ) ✤ /
g○f
such that
– (Associativity) (h ○ g ) ○ f = h ○ (g ○ f ) for any morphisms f, g, h, and
– there exists 1A ∈ HomC (A, A), called the identity (over A) such that 1A ○ f = f and
g ○ 1A = g for every f ∈ HomC (B, A) and g ∈ HomC (A, B ).
2
Definition 2. Let C be a category and A, B ∈ Obj (C ). We say A and B are isomorphic if there exist
f ∈ HomC (A, B ) and g ∈ HomC (B, A) such that g ○ f = 1A and f ○ g = 1B . We write A ≅ B
and call f and g isomorphisms.

Definition 3. Let C be a category. A morphism f of C is called monic or a mono(morphism) (resp.

epic or an epi(morphism)) if f ○ g = f ○ h implies g = h (resp. g ○ f = h ○ f implies g = h)

Definition 4. C is a small category if both Obj (C ) and Hom (C ) are sets and not proper classes, and
large otherwise. It is locally small if for every A, B ∈ Obj (C ), HomC (A, B ) is a set (called a homset).

Definition 5. A concrete category C is ( for the time being) one whose objects are sets, potentially with
additional structures (i.e. have underlying sets) and whose morphisms are maps between these sets.

Definition 6. Let C and C ′ be two categories. We say C ′ is a subcategory of C if

– Obj (C ′) ⊂ Obj (C ).
– For every A′, B ′ ∈ Obj (C ′), HomC ′ (A′, B ′) ⊂ HomC (A′, B ′).
– morphism composition in C ′ coincides with that in C.
– for every A′ ∈ Obj (C ′), 1A′ in HomC (A′, A′) is equal to 1A′ in HomC ′ (A′, A′).

Definition 7. We say a subcategory C ′ of C is full if for every A′, B ′ ∈ Obj (C ′) we have

HomC ′ (A′, B ′) = HomC (A′, B ′).
First examples
● Let us start with a minimalist approach. The empty category, 0 has no objects and morphisms:
Obj (0) = {∅} , Hom (0) = {∅} .
● Define now the category with one object and one morphism, 1:
Obj (1) = {A} , Hom1 (A, A) = {1A} .
This is obviously small because both Obj (1) and Hom (1) are sets (both of them having size 1).
● We can also define a category 2 with two objects and only one non-identity morphism:
⎧
⎪
⎪ Hom2 (A1, A2) = {f } ,
Obj (2) = {A1, A2} , ⎨ Hom2 (A2, A1) = {∅} , (1)
⎪
⎪
⎩ Hom2 (Ai, Ai) = {1A } , i = 1, 2.
i

– Again, it is small because Obj (2) is a set (of size 2) and Hom (2) is a set (of size 3).
– Bear in mind that these objects and morphisms are primitive notions in that A1, A2 are not
expected to be sets, and f, 1A1 , 1A2 are not expected to be functions. We can visualize this
with a diagram (i.e. a directed graph where vertices are objects and edges are morphisms):

A2 Of course, for this to define a category we still need to define a

A1 law of composition on (1) but it is simple here:
f
1A2 1A1 ○ 1A1 := 1A1 , 1A2 ○ 1A2 := 1A2 , f ○ 1A1 := f, 1A2 ○ f := f
1A1 and we now have a category.
● For general N ∈ N, we would like to define a category N with N elements such that
⎧
⎪
⎪ {fi,j } if i < j,
Obj (N) = {A1, . . . , AN } , HomN Ai, Aj = ⎨ 1A if i = j, (2)
⎪
⎪
⎩ ∅ if i ≥ j.
i

and we can represent this with a diagram as well, e.g. for N = 4:

1A 2
A2 f2,3 A3 1A 3
We call a diagram of objects and arrows commutative if any
f1,3
possible path (i.e. sequence of compositions) between two
f1,2 f3,4 vertices yields the same composition.
f2,4 So far, we do not have a composition so we do not know
A1 A4
whether this particular diagram is commutative, but...
f1,4
1A1 1A4
... given that there is only one morphism between any two objects, defining the composition law
fj,k ○ fi,j := fi,k , for every i ≤ j ≤ k, fi,j ○ fk,l := ∅ in all other instances
renders N with (2) a category and diagrams such as the above commutative, at the same time.
Exercise: check associativity of ○ yourselves.
● Again, we are not assuming Ai to be sets or fi,j to be functions between them, a priori.
● The next few slides will show examples of categories that will be infinite (thereby making diagrams
a bit less practical) but whose objects and morphisms you will find it easier to relate to.
Examples of large, concrete categories
● Set, the category of sets, whose objects and morphisms are
Obj (Set) = {all sets} , HomSet (A, B ) = {all functions A → B} .
– We need a law of composition for this to be a category; we got one, namely the usual function
composition: given S, T, U ∈ Obj (Set), f ∈ HomSet (S, T ) and g ∈ HomSet (T, U ),
g ○f

f g %
S /
T U /

s✤ /
f (s ) ✤ /
g (f (s))

Again, associativity is easy to check: s ↦ t ↦ u ↦ v will always have v as the final image
f g h

regardless of how you assemble f, g, h in sequences of binary compositions.

– It is large, because Obj (Set) is not a set – otherwise this would lead to Russell’s paradox.
– But it is also locally small because HomSet (S1, S2) is a set, for every S1, S2 ∈ Obj (Set).
– It is concrete because its objects are sets (in this case with no additional structure expected).
– given a set S ∈ Obj (Set), the identity element 1S is nothing but the identity function,
idS
1S = idS ∈ HomSet (S, S ) defined by: S✤ /
S
x /
x
– in Set, monic is equivalent to injective and epic means surjective. Let us prove the first state-
ment and leave the second as an Exercise:
⇒) assume f : S → T is a monomorphism and let x, y ∈ S such that f (x) = f (y ).
Consider any singleton (i.e. set of cardinality 1) {a} ∈ Obj (Set) and the diagram
⎧
⎪
⎪
⎪
⎪ {a}
λx /
⎪
f ○λx
{a} ❚❚❚❚λ
S
⎪
⎪
⎪
⎪
❚❚❚x❚ a ✤ /
x
where λx, λy are defined by ⎨
❚❚❚❚ f #

⎪
❥❥5 S ;T
)

⎪
/

⎪
⎪
❥

{a}
❥

{a}
❥❥ ❥

⎪
❥
❥
λy /
⎪
❥❥❥
λ
⎪
y S
⎪
⎪
⎩ a
f ○λy ✤ /
y
We know that f ○ λx ≡ f ○ λy :
a ✝✩❋❋❋ f ○λx
❋❋
λ
❋❋ x
❋❋
❋❋
❋❋
❋# f
x✤ /
f (x)

=
①; y✤ f
/
f> (y )
①①①
①
①①
①①
✽①①①
①①
λ y
a ✚
f ○λy
thus the assumption that f is a mono implies λx ≡ λy , hence both functions map a to the
same element, which implies x = y □ 7
 Let us prove the converse:
(⇐ assume f : S → T is injective. We want to prove it monic in Set. Let A ∈ Obj (Set) and
g, h ∈ HomSet (A, S ). We have the diagram
f ○g
A ❚❚❚❚❚❚g❚❚
❚❚❚❚ f #
❥❥❥❥4
*
S /
< T
❥❥❥❥
❥❥❥❥
A h
f ○h
assume f ○ g ≡ f ○ h. That means f (g (x)) = f (h (x)) for every x ∈ A. But we know
f is injective, thus this implies g (x) = h (x) for every x ∈ A. That is the same as saying
g ≡ h. Thus, f injective implies f mono. □
– As said in the previous page, epic is also equivalent to surjective in morphisms of Set.
– A morphism that is both epic and monic is called a bimorphism.
– Thus, it follows from the above that bijective is equivalent to bimorphism in Set.
– How does this being a bimorphism relate to being an isomorphism?
∗ In Set, they are equivalent concepts. We call Set a balanced category because of this.
More on this in the next slide.
∗ In general, though, isomorphism implies bimorphism, but the converse need not be true.
Later on we will see a category and a morphism exemplifying this.
8
● So in a nutshell: when are two sets S1, S2 considered equivalent in the category Set?
(
Whenever S1 ≅ S2 in Set i.e. there exist isomorphisms S1 h S2 in Set, or functions
⎫
S2 ⎪⎪
⎪ ⎧
⎪
f
⎪
⎪ ⎪
⎪
S1 g ○f
⎪ ⎪
/

f (x) ⎪⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
f g
x i.e.
⎪ ⎪
✤ '
/
g ○ f = 1 S S S1
⎪ ⎪
/ /
S1
⎪ ⎪
1 2

⎪
⎪ ⎪
⎪
f (x) g (f (x)) = x
✤ ✤
x / /

and ⎬ such that ⎨ and

⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
f ○g

⎪
⎪
⎪ ⎪
⎪
⎪
S1 ⎪⎪ ⎪
⎪
f i.e.
g f '

⎪ ⎪
f ○ g = 1 S S S2
⎪ ⎪
/ /
S2 /
S2 2 1
⎪
g (y ) ⎪ ⎩
⎭
y ✤ / y ✤ /
g ( y ) ✤ /
f (g (y )) = y

in other words: whenever there exists at least one bijective function between S1 and S2
(because an invertible function is the same as a bijective function, Exercise).
● Set Theory studies properties conserved by isomorphisms of sets, i.e. bijective functions, e.g.
– cardinality because if there exists S1 → S2 bijective, then #S1 = #S2;
– countability, i.e. whether or not there exists a bijection N → given set;
if there exists f : S1 → S2 bijective, then S1 is countable if and only if S2 is countable.
– subsets, i.e. if there exists S1 → S2 in Set, then S1 and S2 have similar subset lattices.
≅

– very long etcetera

9
● Grp, the large, concrete category of groups, whose objects and morphisms are

Obj (Grp) = {all groups} , HomGrp (G1, G2) = {all group morphisms G1 → G2} .

Definition 8. A group (G, ⋆) is a set G endowed with a binary operation ⋆ such that:
– ⋆ is internal, i.e. x ⋆ y ∈ G for every x, y ∈ G,
– ⋆ is associative, i.e. x ⋆ (y ⋆ z ) = (x ⋆ y ) ⋆ z for every x, y, z ∈ G,
– ⋆ has a neutral (identity) element eG ∈ G: eG ⋆ x = x ⋆ eG = x for every x ∈ G,
– every element x ∈ G has an inverse, i.e. x−1 ∈ G such that x−1 ⋆ x = x ⋆ x−1 = eG.

Definition 9. A morphism or homomorphism of groups (G1, △) → (G2, □) is a function

f

between the underlying sets G1 → G2 such that f (x △ y ) = f (x) □f (y ) for every x, y ∈ G1.
f

– The composition rendering this a category is the same as in Set, i.e. composition of the
underlying functions of group morphisms: (g ○ f ) (x) = g (f (x)).
– It is large, because if Obj (Grp) were a set, we would arrive to Russell’s paradox.
– It is also locally small, for the same reason as Set.
– it is concrete because each of its objects has an underlying set.
– it is a subcategory of Set, but it is not a full subcategory as we will see in a moment:
∗ Obj (Grp) ⊂ Obj (Set) because every group is also set;
∗ HomGrp (G1, G2) ⊂ HomSet (G1, G2); every group morphism is, after all, a function
● When are two groups (G1, ⋆), (G2, ⊙) considered equivalent in the category Grp?
Whenever (G1, ⋆) ≅ (G2, ⊙) in Grp i.e. whenever there exist
f ∈ HomGrp (G1, G2), g ∈ HomGrp (G2, G1) that are mutual inverses:
⎫
⎪ ⎧
⎪
⎪
⎪
⎪
⎪
⎪
f g ○f
⎪ ⎪
G1 G2
⎪ ⎪
/

a✤ f (a ) ⎪
⎪
⎪ ⎪
⎪
⎪ i.e.
f g '

f ( x1 ⋆ y 1 ) = f ( x1 ) ⊙ f ( y 1 ) ⎪ ⎪
/
g ○ f = 1 G G G1
⎪ ⎪
/ /
G1
⎪ ⎪
1 2

⎪
⎪ ⎪
⎪
a✤ f (a) ✤ / g (f (a)) = a
/

and ⎬ such that ⎨ and

⎪
⎪
⎪ ⎪
⎪
⎪
f
⎪
⎪
⎪ ⎪
⎪
⎪
f ○g

⎪ ⎪
G2 G1
⎪ ⎪
/

⎪
⎪
⎪ ⎪
⎪
⎪
g f
i.e.
'

⎪ ⎪
b ✤ /
g (b ) f ○ g = 1 G G G2
g ( x2 ⊙ y 2 ) = g ( x2 ) ⋆ g ( y 2 ) ⎪ ⎪
/ /
G2 2 1
⎪
⎭ ⎩ g (b ) f (g (b)) = b
✤ ✤
b / /

meaning whenever there exists a morphism between them which is bijective as a function
(the inverse of a bijective morphism is also a morphism, Exercise).
● Group Theory studies properties conserved by isomorphisms of groups, i.e. bijective group
morphisms, e.g.
– abelianity because if there exists G1 → G2, then G1 is abelian if and only if G2 is.
≅

– being cyclic, solvable, finite, long etcetera, for the same reason: if G1 ≅ G2 then one has
the given property if, and only if, the other has it too.
– all the properties Grp inherits from the larger category Set, e.g. cardinality, countability, etc.
Why is Grp not a full subcategory of Set?.
We know every morphism of groups is a morphism of sets (i.e. a function). Is the converse true?

{a, b, c, d} ⎫
⎪ {●, △, ,}
f
⎪
⎪
/

⎪
⎪
⎪
⎪
⎪ this is a function between sets,
a ✚❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩❩ ❥❥❥❥❥❥❥5 ●
⎬
❩❩❥❩
❥❥❥❥ ❩❩❩❩❩,

hence f ∈ HomSet ({a, b, c, d} , {●, △, ,})

b ❨❨❥❨❥❨❥❨❥❨❥❨❥❨❨❨❨❨ ❦❦❦❦❦❦5 △
⎪
❥

⎪
✙❨❨❨❨ ❥❥

⎪
⎪
❥ ❨❨❨ ❦

⎪
✯❥❥❥❥❥ ❦❦❦❦❨❨❨❨❨❨❨,

⎪
c ❦❦❦❦❦❦❦❦ ❦ ❦
,
⎪
❦

⎪
⎭
✰ ❦ ❦❦
d ❦❦
if we define group structures on both sets with the following Cayley tables,
⋆ a b c d
+ ● △ ,
Exercise: prove ({a, b, c, d} , ⋆) and
a a b c d
● ● △ ,
({●, △, ,} , +) are indeed groups
b b c d a
△ △ , ●
c c d a b
, , ● △
d d a b c
then if the group morphism fails at least in one pair of elements, f will not be a group morphism:

} Ô⇒ f (b ⋆ c) ≠ f (b) + f (c) ,
f (b ⋆ c) = f (d ) = △
f (b ) + f (c ) = , + ● = ,
hence f ∉ HomGrp (({a, b, c, d} , ⋆) , ({●, △, ,} , +)), in other words:

Every group morphism is a set morphism, but the converse is generally not true.
Another example. Sets S := {a, b, c, d} and T := {x, y, z, t} are isomorphic in Set:
all we need to find is at least one isomorphism of Set between them, i.e. a bijective function:

{a, b, c, d} / {x, y, z, t}
f

a ✚❩❩❩❩❩❩❩❩❩❩❩❩❩❩ ♦7 x
❩❩❩❩❩♦❩♦♦♦♦
❩
b ✚❩❩❩❩❩❩❩❩❩❩❩❩♦❩♦❩♦♦♦♦♦♦ ❩❩❩, y
❩❩❩❩❩❩
♦♦♦ ❩❩❩❩,
c ✚❩❩❩♦❩♦❩♦❩♦❩♦❩♦❩♦❩❩❩❩❩❩❩ z
✴♦♦♦ ❩❩❩❩❩❩❩
❩❩,
d t
which we know is bijective because it has an inverse (equally bijective of course):
{x, y, z, t} / {a, b, c, d}
g

x ✎❖❖❖❖ ❞❞❞❞❞❞ a
❞2
❖❖❞❖❞❞❞❞❞❞❞❞❞❞❞
❞
y ✩❞❞❞❞ ❖❖❖❖❖❖❞❖❞❖❞❞❞❞❞❞❞❞❞❞❞2 b
❞ ❖❖
❞❞❞❞❞❞ ❖❖❖
z ✩❞❞❞❞ ❞❞❞❞❞ ❞
❞ ❞❞❞❖❞❖❞❖❞❖❞ c
❖ ❞2
❞ ❞❞ ❞❞❞ ❖❖ '
t ✩❞❞❞❞ d
That they are mutual inverses can be checked very easily: we check g ○ f = 1{x,y,z,t},
{a, b, c, d} / {x, y, z, t} / {a, b, c, d}
f g

a ✚❩❩❩❩❩❩❩❩❩❩❩❩❩❩ ♦7 x ✎❖❖❖❖ ❞❞❞❞❞❞ a

❞2
❩❩❩❩❩♦❩♦♦♦♦
❩
❖❖❞❖❞❞❞❞❞❞❞❞❞❞❞
❞ and you can check f ○ g = 1{a,b,c,d} yourselves
b ✚❩❩❩❩❩❩❩❩❩❩❩❩♦❩♦❩♦♦♦♦♦♦ ❩❩❩, y ✩❞❞❞❞ ❖❖❖❖❖❖❞❖❞❖❞❞❞❞❞❞❞❞❞❞❞2 b
♦ ❩❩❩❩❩❩❩❩ ❞❞❞ ❖❖❖
♦♦♦ ❩❩, ✩❞❞❞❞❞❞❞❞
c ✚ ♦♦❩♦❩♦❩❩❩❩❩❩❩❩❩❩
❩❩❩❩ ♦♦ z ❞❞❞❞❞❞❞❞❞❞❖❞❖❞❖❞❖❞ c
❖❖❖ ❞2
✴♦♦♦ ❩❩❩❩❩❩❩ ❞ ❞❞❞ ❖❖
❩❩, ✩❞❞❞❞❞❞ '
d t d

Thus {x, y, z, t} ≅ {a, b, c, d} in Set.

However: what happens if S and T are endowed with group operations?
Then they may, or may not, be isomorphic in Grp.
– For instance, if we define ⋆ and □ with the Cayley tables
⋆ a b c d □ x y z t
a a b c d x t z x y
b b c d a y z t y x
c c d a b z x y z t
d d a b c t y x t z
● (S, ⋆) and (T, □) are both groups (Exercise);
● (S, ⋆) ≅ (T, □); all we need to prove this is finding at least one bijective morphism
between the groups. In this case, for instance,
{a, b, c, d} / {x, y, z, t}
f

a ✔❚❚❚❚❚❚❚ ♦♦7 x
❚❚❚❚
♦♦♦♦♦
b✤ ❚❚❚❚ ♦♦
♦♦♦♦❚♦❚❚❚❚❚
/y
♦♦♦ ❚❚❚❚
*
c ✚❩❩❩♦❩♦❩♦❩♦❩♦❩♦❩♦❩❩❩❩❩❩❩ z
✴♦♦♦ ❩❩❩❩❩❩❩
❩❩,
d t
is bijective and checking the morphism property on each pair of elements is easy, e.g.
f (b ⋆ c) = f (d) = x, f (b ) □ f (c) = y □ t = x ⇒ f (b ⋆ c ) = f (b ) □ f (c )
(and you can check the rest of cases yourselves, Exercise). 14
– But assume we change one of the group operations, say the one on T = {x, y, z, t}:
⋆ a b c d ⋅ x y z t
a a b c d x y x t z
b b c d a y x y z t
c c d a b z t z y x
d d a b c t z t x y
● T is also a group with the new operation ⋅ (Exercise);
/ (T, ⋅); you can find bijective functions between the underlying sets, e.g.
● but (S, ⋆) ≅
{a, b, c, d} / {x, y, z, t} {a, b, c, d} / {x, y, z, t} ⎫
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
a ✚❩❩❩❩❩❩❩❩❩❩❩❩❩❩ ♦7 x a ✔❚❚❚❚❚❚❚ 7x
functions defined in
⎬
❩❩❩❩❩♦❩♦♦♦♦ ❚❚❚❚ ♦♦ ♦♦♦

⎪
❩ ❚❚❚❚ ♦♦♦♦
b ✚❩❩❩❩❩❩❩❩❩❩❩❩♦❩♦❩♦♦♦♦♦♦ ❩❩❩, y
⎪
b✤ /y
the previous two slides
⎪
♦♦❚♦❚❚❚❚❚
⎪
♦

⎪
♦ ❩❩❩❩❩❩❩❩ ♦ ❚
♦♦♦
⎪
❚❚❚*
♦♦♦
⎪
❩❩,
c ✚ ♦❩♦❩♦❩♦❩❩❩❩❩❩❩❩❩❩ z c ✚❩❩❩♦❩♦❩♦❩♦❩♦❩♦❩♦❩❩❩❩❩❩❩ z
⎪
❩❩❩ ♦♦

⎭
✴♦♦♦ ❩❩❩❩❩❩❩
❩❩, ✴♦♦♦ ❩❩❩❩❩❩❩
❩❩,
d t d t
but no such bijection is a morphism between (S, ⋆) and (T, ⋅). This is easy to prove:
the identity elements are e(S,⋆) = c and e(T,⋅) = y; the smallest number of times we have
to operate b with itself to obtain e(S,⋆) (called the order of b) is 4, whereas no element in
(T, ⋅) has order 4 . Any bijective morphism would preserve order (Exercise) hence
– you can find bijective maps S → T but they will not be morphisms (S, ⋆) → (T, ⋅),
– and morphisms (S, ⋆) → (T, ⋅) exist (e.g. map all of S to y) but none are bijective.

Thus the heuristic conclusion is: being isomorphic in Grp is harder than in Set.
● Grp has an important subcategory of its own: Ab, the large, concrete category of abelian groups,

Obj (Ab) = {all abelian groups} , HomAb (G1, G2) = HomGrp (G1, G2) .

Definition 10. A group (G, ⋆) is abelian or commutative if x ⋆ y = y ⋆ x for every x, y ∈ G.

– it is large, locally small and concrete for the same reasons as Set or Grp.
– it is not full as a subcategory of Set (the example a few slides ago still works here).
– it is full as a subcategory of Grp.
Examples
∗ with the usual addition, (Z, +) , (Q, +) , (R, +) , (C, +) ∈ Obj (Ab).
∗ with the usual multiplication, (Q ∖ {0} , ⋅) , (R ∖ {0} , ⋅) , (C ∖ {0} , ⋅) ∈ Obj (Ab).
∗ defining the quaternion group Q8 = {±1, ±i, ±j, ±k} by
i2 = j2 = k2 = ijk = −1, 1 neutral element and (−1)2 = 1,
then Q8 ∈ Obj (Grp) but Q8 ∉ Obj (Ab) because e.g. ij ≠ ji (Exercise).
∗ defining the following maps

R ∖ {0}
f 1 / f2 / f f4
R R (0, ∞) Z 3 / Q8 Z /
Z
/ x / x / 2
x✤ e x✤ e k✤ i
/ k k✤ k
then f1, f2, f3, f4 ∈ HomSet, f1, f2, f3 ∈ HomGrp, f1, f2 ∈ HomAb and
16
f2 is an isomorphism in Ab.
● Ring, the large, concrete category of rings, is given by
Obj (Ring) = {rings} , HomRing (R1, R2) = {ring morphisms R1 → R2} .

Definition 11. A ring (R, +, ⋅) is a set R endowed with two binary operations +, ⋅ such that:
– +, ⋅ are internal, i.e. x + y and x ⋅ y ∈ R for every x, y ∈ R,
– (R, ⋅) is an abelian group (see the preceding slide),
– ⋅ is associative, i.e. x ⋅ (y ⋅ z ) = (x ⋅ y ) ⋅ z for every x, y, z ∈ R,
– +, ⋅ satisfy the distributive properties:
x ⋅ (y + z ) = (x ⋅ y ) + (x ⋅ z ) and (x + y ) ⋅ z = (x ⋅ z ) + (y ⋅ z ) , for every x, y, z ∈ R,

Additional properties:
– If ⋅ has a neutral element, 1R such that 1R ⋅ x = x ⋅ 1R = x for every x ∈ R, we call
R unitary or unital.
– If x ⋅ y = y ⋅ x for every x, y ∈ R, we call R commutative.

Definition 12. A morphism or homomorphism of rings (R1, +, ⋅) → (R2, ⊕, ⊙) is a function

f

between the underlying sets R1 → R2 such that for every x, y ∈ R1,

f

– f (x + y ) = f (x) ⊕ f (y ) i.e. f is a group morphism (R1, +) → (R2, ⊕).

– f (x ⋅ y ) = f (x) ⊙ f (y ).
– it is large and concrete (and not full as a subcategory of Set) for the same reasons as Grp.
● Relevant subcategories of Ring:
– URing, the category of unitary rings, is not full in Ring:
Obj (URing) = {rings (R, +, ⋅) with multiplicative identity, i.e. neutral element 1R for ⋅} ,
HomURing ((R1 , □, △) , (R2 , ⊕, ⊙)) = {f ∈ HomRing ((R1 , □, △) , (R2 , ⊕, ⊙)) : f (1R1 ) = 1R2 } ;

Examples
∗ with the usual + and ⋅, (Z, +, ⋅) , (Q, +, ⋅) , (R, +, ⋅) , (C, +, ⋅) ∈ Obj (URing);
∗ the set of n × n matrices with coefficients in a ring R, with the usual matrix sum and
product, Matn (R) ∈ Obj (URing) and 1Matn(R) = Idn, the n × n identity matrix.
∗ the subset of even numbers 2Z = {2k : k ∈ Z} ⊂ Z is a ring with same operations as Z, i.e.
(2Z, +, ⋅) ∈ Obj (Ring), but (2Z, +, ⋅) ∉ Obj (URing) .
∗ with the usual operations, (N, +, ⋅) ∉ Obj (Ring) (let alone Obj (URing)).
– CRing, the category of commutative rings, is full in Ring:
Obj (CRing) = {commutative rings} ,
HomCRing ((R1, +, ⋅) , (R2, ,, /)) = HomRing ((R1, +, ⋅) , (R2, ,, /)) ;
Examples
∗ with the usual + and ⋅, (Z, +, ⋅) , (Q, +, ⋅) , (R, +, ⋅) , (C, +, ⋅) ∈ Obj (CRing);
∗ Matn (R) ∉ Obj (CRing) because matrix product is not commutative if n > 1.
An important full subcategory of Ring, URing and CRing:
– Int, the category of integral domains, is given by
Obj (Int) = {integral domains} , HomInt (R1, R2) = HomRing (R1, R2) ;
Definition 13. An integral domain (or simply a domain) is a commutative, unitary ring
(R, +, ⋅) that has no divisors of zero (other than 0R itself ), i.e. for any x, y ∈ R, x ⋅ y implies
x = 0 or y = 0.
Examples (see References and Algebra Notes for examples and definitions):
∗ with the usual + and ⋅, (Z, +, ⋅) , (Q, +, ⋅) , (R, +, ⋅) , (C, +, ⋅) ∈ Obj (Int);
∗ the ring of polynomials R [X ] with coefficients in R (whenever R is a domain itself ) is
also a domain;
∗ Euclidean domains, i.e. domains allowing a division process (with a quotient and a
remainder) similar to that of Z and polynomials over Q, R or C;
∗ principal ideal domains (PID), i.e. domains all of whose ideals are principal, expressible
as multiples of a single generator, e.g. (Z, +, ⋅), R [X ], Q [X ] are PID but Z [X ] is not;
∗ unique factorization domains (UFD), i.e. domains all of whose elements can be decom-
posed in unique fashion in terms of products of irreducible elements; e.g. all PID are UFD.
19
– Let n ∈ Z. The set of possible remainders upon division by n, {0, 1, . . . , n − 1}, yields the
set of congruence classes of integers modulo n, Zn := {0, 1, . . . , n − 1} (also written
Z/ (n) or Z/nZ) by way of the equivalence relation (expressed in different ways)
k1 − k2 is a k1, k2 have the same remainder
k1 = k2 ⇔ k1 ≡ k2 mod n ⇔ ⇔
multiple of n upon division by n
for instance, 3 = 7 = 11 = −1 in Z4.
∗ we can define the following operations on Zn,
k1 + k2 := r, r, q ∈ Z being such that k1 + k2 = nq + r and 0 ≤ r < n,
k1 ⋅ k2 := R, R, Q ∈ Z being such that k1k2 = nQ + R and 0 ≤ R < n.
and it can be proved that (Zn, +, ⋅) ∈ Obj (CRing) , Obj (URing), for instance (Z4, +, ⋅)
has the following operation tables:
+ 0 1 2 3 ⋅ 0 1 2 3
0 0 1 2 3 0 0 0 0 0
1 1 2 3 0 1 0 1 2 3
2 2 3 0 1 2 0 2 0 2
3 3 0 1 2 3 0 3 2 1
also (Zn, +, ⋅) ∈ Obj (Int) if and only if n = 0 or n is a prime number.
For instance (Z4, +, ⋅) described above is not a domain. 20
An important full subcategory of Ring, URing, CRing and Int:
– Field, the category of fields, is given by
Obj (Field) = {fields} , HomField (F1, F2) = HomRing (F1, F2) ;

Definition 14. A field is a commutative, unitary ring (R, +, ⋅) in which every nonzero element
is invertible (with respect to ⋅): if x ∈ R ∖ {0R }, then there exists y ∈ R such that x ⋅ y = 1R .
Examples
∗ with the usual + and ⋅, (Q, +, ⋅) , (R, +, ⋅) , (C, +, ⋅) ∈ Obj (Field);
∗ (Z, +, ⋅) ∉ Obj (Field) because the only invertible elements in Z are ±1.
∗ (Zn, +, ⋅) ∈ Obj (Field) if and only if n is a prime number.
∗ (H, +, ⋅), Hamilton’s quaternions, endowing 4-dimensional space with a ring structure:

dk bi + dk
i2 = j2 = k2 = ijk = −1

a + dk (H, +, ⋅) ∈ Obj (Ring) , Obj (URing)

a + bi + cj + dk (H, +, ⋅) ∉ Obj (CRing) , Obj (Field)
bi
thus (H, +, ⋅) is a more general type of
cj ring (called a division ring or skew field)
a containing (R, +, ⋅),(C, +, ⋅)
a + cj 21
– In all of the above examples of ring categories, the law of composition is obviously the same
as in Set, i.e. ○, and properties defining morphisms f : (R1, +, ⋅) → (R2, ⊕, ⊙) are always
the same (exception for unitary ring morphisms, where one more property is required):
for every x, y ∈ R1, f (x + y ) = f (x) ⊕ f (y ) and f (x ⋅ y ) = f (x) ⊙ f (y )
and if R1, R2 ∈ Obj (URing), then we also expect f 1R1 = 1R2 .

– See further examples of ring morphisms in §2.5 of the Algebra Notes.

– Given a bijective morphism f : R1 → R2 of rings, its inverse f −1 is also a morphism (see

Solved Exercise 197 in the Algebra Notes or prove it yourselves as an Exercise).
However, monic and epic does not imply isomorphism, thus Ring is not balanced–see the
Exercises at the end of this section for an example.

– The category of groups, Grp, is balanced, however. In other words every epic and monic
morphism of groups is also an isomorphism.

22
● Given a ring R, the large, concrete category of R-modules is R − Mod given by
Obj (R − Mod) = {R-modules} , HomR−Mod (M1, M2) = {R-linear maps M1 → M2} .

Definition 15. Given a ring (R, +, ⋅), an R-module or module over R is a set M endowed with
two operations (one internal, one external)
M ×M /M R×M M
/

(m 1 , m 2 ) ✤ /m
1 m2 (r, m) ✤ /r m

– (M, ) is an abelian group,

– is associative with ring product ⋅: r1 (r2 m) = (r1 ⋅ r2) m, ∀r1, r2 ∈ R, m ∈ M ;
– +, , satisfy the corresponding distributive properties:
r (m n ) = (r m ) (r n) and (r1 + r2) m = (r1 m) (r 2 m ) , ∀r1 , r2 ∈ R, ∀m, n ∈ M ;
– 1R m = m for every m ∈ M .
If R is a field, then an R-module is called a vector space over R or an R-vector space
Definition 16. A morphism or homomorphism of R-modules or an R-linear map
(M1, , ) → (M2, , ) is a function M1 → M2 such that
f f

– f (m1 n1) = f (m1) f (n1), ∀m1, n1 ∈ M1, i.e. f ∈ HomAb ((M1, ) , (M2, ));
– f (r m) = r f (m), for every r ∈ R, m ∈ M1.

In particular, the category of Z-modules “is” the category of abelian groups. More on this later.
● Ring Theory studies properties conserved by ring isomorphisms and modules over them, i.e.
ring or module morphisms that are bijective as functions e.g.
– the ones inherited from Set, e.g. cardinality, etc.
– the ones inherited from Grp and Ab for the first operation in a ring, e.g. cyclic ideals.
– commutativity: if R1 ≅ R2 then R1 is commutative if, and only if, R2 is.
– their ideal lattices, or being division rings, factorial, principal, long etcetera.
● Commutative Algebra studies properties conserved by isomorphisms of commutative rings
and modules over commutative rings, e.g.
– all of the above,
– being finitely generated as modules, being Artinian, Noetherian modules, long etcetera.
– being Artinian rings, Noetherian rings, fields, domains, Euclidean, PID, UFD, long etcetera.
● Linear Algebra studies the category of vector spaces K −Vect (i.e. modules over a field K),
thus properties conserved by isomorphisms of K-vector spaces, e.g.
– the ones inherited from Set or, for the vector sum , category Ab,
– dimension, i.e. the size of a basis: if V ≅ W in K −Vect, then dimK V = dimK W .
– number and structure of vector subspaces and direct summands,
– long etcetera
24
A few more examples of large, concrete categories
● Let K be a field. Intuitively speaking, an affine space is any set of points A such that if we fix an
a ∈ A, the segments (properly defined) between a and the rest of elements of A form a K-vector
Ð
→
space A having a as its additive neutral element (i.e. the zero vector, 0, or abelian group neutral
element for ):
⎫
⎪
a5 ⎪
⎪
⎪
⎪
a3
a4 ⎪
⎪
⎪ And choosing another
a2 a6
⎪
⎪
⎪
a7
⎪ origin instead of a0
⎬
⎪
⎪
⎪
⎪
yields a vector space of
⎪
⎪ Ð
→
⎪
⎪
⎪ the same dimension as A
⎪
ak
⎪
a1
⎪
⎭
a0
The large, concrete category of affine spaces is K −Aff given by

Obj (K −Aff) = {K-affine spaces} , HomK−Aff (A1, A2) = {affine maps A1 → A2} .
Ð
→
where a function f : A1 → A2 is called affine if it is consistent with a similar linear map f ∈
Ð→ Ð→
HomK−Vect(A1, A2) in a certain way.
25
– For instance, Rn = {(x1, . . . , xn) : xi ∈ R} can be seen
∗ as a vector space (which we can still denote as Rn), in which case 0 = (0, . . . , 0) is
the additive neutral element, and each of its vector or linear subspaces is the kernel (or
nullspace) of a linear map, i.e. the solution to a set of linear equations:

W linear subspace of Rn ⇔ W = {x ∈ Rn : Ax = 0m} , for some A ∈ Matm×n (R) ;

∗ and as an affine space (in which case we write An R instead of R n even though the

underlying set is the same), if we choose the origin to be any other point (e.g.
p0 = (1, 1, . . . , 1)) and the arrows joining p0 to the other points form a vector space with
the same structure (i.e. isomorphic) to the vector space Rn. In this case, affine subspaces
are the set of solutions to a set of affine equations – in this context, an affine map (as de-
scribed in the previous slide) is defined as a linear map followed by a possible translation:

R ⇔ S = {x ∈ R : Ax = b} ,
S affine subspace of An for some A ∈ Matm×n (R) .
n

26
– Examples of this are easy to find. An affine subspace of An
R of dimension d can be described
by a system of n − d affine equations:
AFFINE PLANE (d = 2) x
{ x − y + 2z = 1 }
1 −1 2 y = 1
´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹¶ z
´¸¶
A ´¹¹ ¹ ¸¹ ¹ ¹¶ b
x

AFFINE LINE (d = 1)

{ }
x + 2y = 1
x + 3 y − 2z = 3
x
1 2 0 y
1
1 3 −2 = 3
z
´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹¸¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¶ ´¹¹ ¹ ¸¹ ¹ ¹¶ ´¹¸¹ ¶
A x b

AFFINE POINT (d = 0)
1 2 0 x 1 AFFINE AND VECTOR LINE (d = 1)
1 3 −2 y = 3
{ }
1 −1 2 z 1 x + 6y + 11z = 0
´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹¶ ´¹¹ ¹ ¹ ¹ ¹¸¹ ¹ ¹ ¹ ¹ ¹¶ ´¹¹ ¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹ ¹ ¶ 3y + 5 z = 0
A x b x
1 6 11 y
0
0 3 5 = 0
z
´¹¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¸ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹ ¹¶ ´¹¹ ¹ ¸¹ ¹ ¹¶ ´¹¸¹ ¶
A x 0
– When are two affine subspaces S1, S2 considered “equivalent” in K − Aff? Whenever
S1 ≅ S2 in this category, i.e. mutually inverse affine functions S1 i S2 exist.
)

– In the context C = R − Aff (e.g. that of the figures in the previous slide) this entails the
existence of an affine isomorphism, meaning a function S1 → S2 which is:
f

∗ affine, i.e. expressible as a linear part A ⋅ ∗ followed by a translation ∗ + b

f
S1 /
S2 A constant matrix and b constant vector
x ✤ / Ax + b
∗ and bijective, i.e. has an inverse g = f −1 : S2 → S1 such that g ○ f = idS1 , f ○ g = idS2 :

which { ●● only needs A to be invertible for g to be well-defined

g
S2 /
S1 is automatically affine if A is invertible
y✤ / A−1 y − A−1 b

– Affine Geometry studies properties left invariant by affine isomorphisms. Which are
basically the same as those of Linear Algebra , plus some others, e.g. parallelism, simplicial
medians and barycenters, cross ratio, etcetera.
– Affine maps x ↦ Ax + b are usually mistaken with linear maps x ↦ Ax. Linear implies
affine (it is affine with trivial translation b = 0) but affine does not necessarily imply linear.
28
– For instance, spaces S1 and S2, seen as affine subspaces of A2R, are isomorphic in R−Aff: a
translation can transform one set into the other, and a translation is an affine bijection:

x = 2xx+ 1 f (x) = x2+x 13/2

+1 = Id2 x + 13/2
0

S1 = {y − 2x = 1} S2 = {y − 2x = −12}

Needless to say you can find infinitely many such translations, e.g.

x = 2xx+ 1
F (x) = x2+x 6 = Id2 x + 6
−1

S1 = {y − 2x = 1} S2 = {y − 2x = −12}

but the existence of at least one, and the fact it is bijective, is enough to establish S1 ≅ S2 in
the category of affine spaces. Meaning: they are “essentially” the same object in this category.
– R−Aff is not full as a subcategory of, say, Set, because not every morphism of sets (i.e. func-
tion) A1 → A2 need be affine, for instance,

+ 2 x2
f (x) = 27 + 4 x2
x = 2xx+ 1

S1 = {y − 2x = 1} S2 = {y − 2x = −12}

is a function between the two sets S1 and S2, i.e. f ∈ HomSet (S1, S2), but it is not an affine
function; it is actually the restriction F ∣S1 where

A2R F /
A2R
x ✤ / 2 x2 + 7
y y + 4 x2 − 2 x + 1

which is ostensibly non-affine because it cannot be expressed as a linear map followed by a

translation, ⋅ ↦ A ⋅ +b.
30
– these two will never be isomorphic in the category of affine spaces R−Aff:

they have different

dimensions and dimension
AFFINE PLANE ≅/ is an invariant of the
isomorphism class of
an affine space, hence
AFFINE LINE
we do not even need to
x
a b c
x
b1 make an effort to find an
a b c y = b1
d e f
y = b2 isomorphism
z z

– these two, on the other hand, are isomorphic, thus categorically “the same”:
x + 5 y − 4z = 4 x + 3y − 4z = −6
x
f (x)

f −1 (X) X

x f ∈ Hom (R−Aff) 1 2 3 x 1
x= y 0 1 −1 y + 3
z 0 0 1 z 2

1 2 3 −1 x 1 f −1 ∈ Hom (R−Aff) X
0 1 −1 y − 3 X= Y
0 0 1 z 2 Z
● An algebraic variety is a subset of An n
R or AC described by a system of algebraic, i.e. polynomial
equations in n variables, i.e. the set of zeros of a finite number of polynomial functions.
Examples:
– any affine (including linear) subspace of Rn or Cn, such as the ones in the previous slides, is
an algebraic set, because an affine map, e.g. x + 5y − 4z − 4, is a polynomial of degree one.
– not every algebraic manifold is an affine subspace, however:

{x2 + y 2 + z 2 − 1 = 0} { } {(x, y ) ∈ R2 : y − x2 = 0} {(x, y, z ) ∈ R3 : x2 − y 2 − z = 0}

x2 + y 2 + z 2 − 1 = 0 ,
(x − 1)2 + y 2 − 1 = 0

– When will we consider two algebraic varieties V1, V2 equivalent in a consistent manner?
Whenever a biregular function V1 → V2 exists, i.e. such that
f

∗ f (x1, . . . , xn) is regular, i.e. polynomial with respect to all variables;

∗ f is bijective;
∗ f −1 is regular as well.
 – With this in mind, and given the fact that function composition is always associative it is clear
we have new subcategory of Set: the category of (real) algebraic varieties R−Var given by
Obj (R−Var) = {algebraic varieties of AnR , n ≥ 1} , HomR−Var (V1 , V2 ) = {regular functions V1 → V2 } .
and obviously a similar category C−Var arises in the complex case.
– For example: any line and any parabola are always isomorphic, thus equivalent in R−Var,
because bijections with polynomial coordinate functions exist between them:

f /
g / h /
≅ ≅ ≅

f g h
(x, ax + b) ✤ /
(x, 0) ✤ /
(x, x2) ✤ /
Exercise
f −1 g −1 h−1 (x, y ) satisfying the
(t, at + b) o ✤
(t, 0) o ✤
(t, t2) o Exercise
✤
equation of a parabola

– In Algebraic Geometry properties conserved by biregular functions are relevant, e.g. di-
mension (also known as Krull dimension in Commutative Algebra).
– If we add one property, namely that a variety “resembles” Rp at each point locally, then an
algebraic variety is called an algebraic manifold and is also studied by Algebraic Geometry.
– R−Aff is not full as a subcategory of R−Var, which is not full itself in Set. Exercise.
 So far we have
Affine Spaces
studied by
⊂ Algebraic Varieties and Manifolds
studied by
Affine Geometry ⊅ Algebraic Geometry

● A larger category: differentiable (or smooth, or C k manifolds), i.e. subsets of An R that can
be defined, at least locally (by way of a tool called an atlas), by C k functions, i.e. functions
f (x1, . . . , xn) at least k times differentiable and having continuous kth partial derivatives:
Obj Diffk = {all C k manifolds M ⊂ Rn} ,
HomDiffk M1, M2 = { k times continuously differentiable functions M1 → M2}
Every algebraic manifold is a differentiable manifold because polynomials are C k for every k (in
fact they are C ∞ and C ω ), but the converse is generally not true:

LINE ELLIPSOID ax2 + by 2 + cz 2 = d2

Affine subspace z = sin (x + y )
Algebraic manifold Algebraic manifold
Differentiable manifold Differentiable manifold Differentiable manifold
– Two C k manifolds M1, M2 are equivalent in Diffk if they are isomorphic in this category,
by which we mean there exists a diffeomorphism M1 → M2, i.e. a function such that:
∗ f ∈ C k (M1, M2), i.e. k times differentiable with continuous kth partial derivatives;
∗ f is bijective;
∗ f −1 ∈ C k (M2, M1).
– A relevant part of Differential Geometry is the study of properties conserved by
diffeomorphisms, e.g.
∗ existence and uniqueness of a tangent space Tp (M ) at any point p ∈ M ;
∗ dimension (defined in terms of affine dimension Tp (M ) at any point), e.g.

Tq S
Tp C

C q
p ≇ S

because their dimension, i.e. the dimension of their respective tangent spaces at any point,
is not the same: one is a manifold of dimension 1 (a differentiable curve), the other is a
manifold of dimension 2 (a differentiable surface);
∗ geodesics and their distribution within the manifold, long etcetera.
 A pattern is clear: every topic studies a particular category
Affine Spaces
studied by
⊂ Algebraic Varieties and Manifolds
studied by
⊂ Differentiable Manifolds
studied by
Affine Geometry ⊅ Algebraic Geometry ⊅ Differential Geometry

● Define the category of topological spaces (see Topology Notes, Chapter 2) is Top by
Obj (Top) = {topological spaces} , HomTop (X1, X2) = {continuous functions X1 → X2} .

Definition 17. Let X be a set and τ ⊂ P (X ) a collection of subsets of X. We say τ is a topology

on X and (X, τ ) is a topological space if the following properties are met:
T1: ∅, X ∈ τ ;
T2: for every collection {Ui}i∈I of elements in τ , ⋃i∈I Ui is an element of τ ;
T3: for every finite collection {U1, . . . , Un} of elements in τ , ⋂n
i=1 Ui is an element of τ .
Elements of τ are called open sets of (X, τ ) (or of X if there is no ambiguity about the topology).
Definition 18. A continuous function of topological spaces f : (X, τX ) → (Y, τY ) is one such
that the pre-image of an open set in Y is open in X: f −1 (U ) ∈ τY for every U ∈ τY .

36
 Example: relevant to our previous examples is the Euclidean topology on subsets S of An R,
whose open sets are those containing an open “circular” disk centered on any element of S:
OPEN

OPEN NOT OPEN

– affine spaces, algebraic manifolds and differentiable manifolds are Euclidean topological
spaces, but the converse is not true in general, for instance this topological curve is not a
differentiable curve:
The curve has a point with no unique
tangent line; the functions defining the
curve are continuous but they are not
differentiable at that point

– obviously, Top is not full in Set; think of examples of functions R → R with discontinuities.
– Topology has an interest in properties conserved by the isomorphisms of Top, which are
called homeomorphisms; think of the topological analogues of diffeomorphisms in Diffk ,
biregular functions in R−Var or affine isomorphisms in R−Aff; homeomorphisms are f :
(X, τX ) → (Y, τY ) such that
∗ f is continuous,
∗ f is bijective,
∗ f −1 is continuous.
there are many such properties, e.g.
∗ the possibility of deciding how “close” points are (i.e. the existence of a metric or distance),
∗ connectedness, i.e. being “in one piece”,
∗ compactness, intuitively (in the Euclidean case) being closed as well as the ability to see
the entire space if we are far enough,
∗ long etcetera including separability, countability axioms, ...
– We know that in a concrete category, f isomorphism implies f a bijective function. We
now have an example in which the converse is not true: not every continuous bijective map
between topological spaces is an isomorphism in Top. In other words:
Top is not a balanced category.
Exercise: find one such continuous and bijective (but not having a continuous inverse)
function, or look for an example in the Topology notes, e.g. Exercise 33 (p. 50).
● It is clear that we have a nested sequence of subcategories of Set:

Affine Algebraic Differentiable Topological

Spaces ⊂ Manifolds
studied by
⊂ Manifolds ⊂ Spaces
studied by
studied by
Affine Geometry
⊅ Algebraic Geometry
⊅ studied by
Differential Geometry
⊅ Topology

● The smaller the category, the more rules, objects and properties it must preserve, thus the
more constraints it has – and thus the easier it becomes, which is why Affine Geometry allows
much less freedom, hence is much easier to learn than Topology or Set Theory. It also explains
why Group Theory is much easier than Set Theory,
Groups
studied by
⊂ Sets
studied by
Group Theory ⊅ Set Theory

or why commutativity simplifies the study of rings a great deal:

Commutative Rings and Modules over them
studied by
⊂ Rings and Modules over them
studied by
Commutative Algebra ⊅ Ring Theory

39
Examples of small categories
● Let G be a group. Define category C by

Obj (C ) = {G} , HomC (G, G) = G.

Composition is defined by

HomC (G, G) × HomC (G, G) /

HomC (G, G)
(a, b) ✤ /
b ○ a := b ⋅ a
– it is small because both {G} and G are sets (of sizes 1 and ∣G∣ respectively);
we can also write elements of G in arrow form here, e.g. a ∈ G as G → G;
a
–
– ○ is associative because associativity is one of the conditions on the internal operation ⋅ on G;
– 1G is nothing but the identity element in G: 1G ○ x = 1G ⋅ x = x for every x ∈ G;
– also every morphism f ∈ HomC (G, G) is an isomorphism, because f is an element of
G and every element in a group is invertible with respect to ⋅: there exists y ∈ G (thus y ∈
HomC (G, G)) such that x ⋅ y = y ⋅ x = 1G.

40
● If we do not require each element to be invertible, we have something more general than a group:
Definition 19. A monoid (M, ⋆) is a set M endowed with a binary operation ⋆ such that:
– ⋆ is internal, i.e. x ⋆ y ∈ M for every x, y ∈ M ,
– ⋆ is associative, i.e. x ⋆ (y ⋆ z ) = (x ⋆ y ) ⋆ z for every x, y, z ∈ M ,
– ⋆ has a neutral (identity) element eM ∈ M : eM ⋆ x = x ⋆ eM = x for every x ∈ M ,
Let M be a monoid. Define category C in the same manner as we did in the previous slide:

Obj (C ) = {M } , HomC (M, M ) = M,

again with composition defined by the operation on M :
HomC (M, M ) × HomC (M, M ) / HomC (M, M ) , again, 1M is nothing but the identity element in M .
(a, b) ✤ / b ○ a := b ⋅ a
– ○ is associative and C is a small category for the same reasons as the previous example;
– the converse is also true: if C is a category with a single object A, HomC (A, A) is a monoid
(trivially easy exercise).
– we can still write elements of M in arrow form here, e.g. a ∈ M as M → M ;
a

– however, not every morphism f ∈ HomC (M, M ) is an isomorphism. For instance, in the
monoid (Z, ⋅), element 2 is not invertible with respect to ⋅ (its multiplicative element would
be 12 ∉ Z) but it is a morphism Z → Z of C.
2
41
● Let (S, ≤) be a poset, i.e. a partially ordered set. See the folder Background Material for
a small introduction to what that means. We define the small, non-concrete category C by

{(a, b)} , a ≤ b,
Obj (C ) := S, HomC (a, b) = {
∅ a>b
– Associative law is observed: a ≤ b and b ≤ c imply a ≤ c due to the transitive property:

HomC (a, b) × HomC (b, c) /

HomC (a, c)
((a, b) , (b, c)) ✤ /
(a, c)
And the identity on a is a ≤ a.
– Taking the trivial order ≤t into account (i.e. if x ≤t y iff x = y), the conclusion is that

Every set S can be understood as a category.

– C is small, obviously, because Obj (Set) and HomC (a, b) ∀a, b are sets.
– You will find that for finite, total ordered sets (S, ≤), this category is “essentially the same”
as the categories N defined in the first lecture.

42
Examples of categories defined from other categories
● Let C be a category. We define the arrow category D := Arr (C ) by
Obj (D) = Hom (C ) , i.e. every object is a morphism f ∈ HomC (A, B ) for some A, B ∈ Obj (C ),

and for any f ∈ HomC (A, B ) and g ∈ HomC (A′, B ′),

HomD (f, g ) = {(α, β ) ∈ HomC A, A′ × HomC B, B ′ : β ○ f = g ○ α} .
i.e. rendering the following diagram commutative (see §1.5.3 in Algebra Notes):
f
A /
B
sometimes we denote commutative
α β diagrams with ⟲ or – or omit these

≡
≡

symbols it commutativity is clear

 
g
A′ B′ /

– we still need to define a law of composition on D. This will be simply given by placing com-
mutative squares next to one another: given the following two morphisms in D = Arr (C ),

{A → B} Ð→ {A → B } , {A → B } Ð→ {A → B }
f ′ f′ ′(α,β ) ′ ′ (α′,β ′) ′′ f ′′ ′′
and

we define their composition as

{A → B} , {A → B } .
f f ′′ ′′
(α′, β ′) ○ (α, β ) := (α′ ○ α, β ′ ○ β ) ∈ HomD ′′
– Of course, we need to check the final claim, i.e. that (α′, β ′) ○ (α, β ) is indeed a morphism
in D. We have two commutative diagrams in C,
A α /
A′ A′ α′ /
A′′ A A′ A′′
f f′ f′ f ′′ where f , f′ , f ′′ ∈ Obj (D)
β′
  

β   
B /
B′
B′ /
B ′′ B B ′′ B′
and studying α′ ○ α and β ′ ○ β equates to pasting these two diagrams together:
A α /
A′ α′ /
A′′
f f′ f ′′

β 
β′ 
B /
B ′′ /
B′
associativity of ○ and commutativity of the two original diagrams implies
f ′′ ○ α′ ○ α = f ′′ ○ α′ ○ α = β ′ ○ f ′ ○ α = β ′ ○ f ′ ○ α = β ′ ○ β ○ f = β ′ ○ β ○ f
thus the following is commutative:
A α′○α A′′
/
A A′′
f ⟲ f ′′ ⇒ α′, β ′ ○ (α, β ) = (α′ ○ α, β ′ ○ β ) ∈ HomD f , f ′′
 

B

β ′○β /

B ′′
B B ′′
Exercise: prove associativity of the composition in D.
● See §3.4.1, Lemma 3.4.4 (the Five Lemma), Example 3.4.5, Definition 3.4.8 in the typeset
Algebra Notes. Given two R-modules M and N , define the large, concrete category C by
Obj (C ) = {all extensions of M by N , i.e. short exact sequences 0 → M → E → N → 0}
and given any two such extensions, define exact sequence i.e. short exact sequences of the form
⎧
⎪
⎪
⎪ E : 0 → M → E → N → 0,
f g
⎨
⎪
⎪
⎪ → →
f˜
→
g̃
N →0
⎩ F : 0 M F
morphisms E → F are given by ϕ ∈ HomR−Mod (E, F ) yielding a commutative diagram:
f g
0 /
M✤ /
E✤ /
N✤ /
0
✤ ✤ ✤

idM
✤
✤
✤
✤ ϕ
✤
✤ idN (3)
✤ ✤ ✤

f˜
✤ ✤ ✤
g̃
0 /
M /
F /
N /
0
in other words (if we take the two identity morphisms for granted in an abridged notation):
⎧
⎪ all morphisms E → F closing the diagram: ⎫
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪
=E✤ ❇
⎪
⎪
HomC (E, F) = ⎨ ⎬
④④④ ✤ ❇❇❇
④ ❇❇

⎪ ⎪
④④ ❇❇

⎪ ⎪
④④ ✤

⎪
⎪
0 N /0 ⎪
⎪
✤
M ✤ϕ
⎪ ⎪
/

⎪ ⎪
❈❈ ⑤>

⎪
⎪ ⎪
⎪
❈❈❈ ✤ ⑤⑤

⎪ ⎪
❈❈ ✤ ⑤⑤

⎩ ⎭
❈ ⑤⑤
!  ⑤
F 45
– All morphisms are isomorphisms in this category; indeed, the Five Lemma automatically
implies every morphism ϕ such as the one appearing in (3) will always be an isomorphism in
R−Mod, thus inducing an isomorphism in C.
– there may be objects in C such that there is no morphism between them. For instance,
Exercise, defining morphisms of Z-modules
(2) = {2k : k ∈ Z} i π τ pr
/ Z Z / Z2 (2) / (2) × Z2 (2) × Z2 / Z2
2k ✤ / 2k k✤ /k
k✤ / k, 0 (2k, m) ✤ /m

there is no morphism closing this diagram of Z-modules,

ss
9 Z✤ ❏❏❏❏
i
ss
ssss ✤
✤
❏❏ π
❏❏
❏❏
sss ❏❏
s ✤ %
0 /
(2) ■■ ✤
✤ ∄ : Z2 /
0
■■ ✉✉
■■ ✤ ✉✉✉
pr
■■ ✉
τ ■■
$
✤
 ✉✉
✉✉

(2) × Z2
where Cartesian product (2) × Z2 is endowed with Z-module structure with operations

(2k1, n1) + (2k2, n2) = (2k1 + 2k2, n1 + n2) , m ⋅ (2k, n) = (2mk, mn) ;
we will see a general version of × later on.
46
● Let C be a category. We define the opposite or dual category C op by inverting all the morphisms:

Obj (C op) := Obj (C ) , HomC op (A, B ) := HomC (B, A) .

Composition works in reverse order:

HomC op (A, B ) × HomC op (B, C ) /

HomC op (A, C )
(f, g ) ✤ /
f ○g
A category C equal (or equivalent, to be defined in the future) to its own dual C op is self-dual.
● For any statement E (C ) regarding a certain category C, the dual statement E (C op) = E (C )op
results from interchanging the source and target of every morphism as well as the order of com-
position.
● Thus if E is valid given certain axioms {A1, . . . , An}, then the dual E op is valid as long as
{A1 , . . . , An } are valid in C op.
op op

● Any statement such that E = E op is called self-dual.

47
Examples
– (S, ≤)op = (S, ≥).
– Isomorphism is a self-dual categorical statement.
– Validity for every category implies validity of the dual for every category.
– Here is an example of how to find the dual E op from any categorical definition E:
M : a morphism f : M → N such that f g = f h implies g = h.
M op: a morphism f : N → M such that gf = hf implies g = h.
in other words: the Definitions of monic and epic in 3 are duals of one another.
– Another example:
Lemma 20. For a morphism f : A → B in Set, the following are equivalent:
1. f is monic;
2. f is left-invertible, i.e. there exists g : B → A such that g ○ f = idA.
The dual of Lemma 20 is the Lemma 21 below (Exercise). Observe how we operate with
compositions and the interchange of notions M , M op in the previous item:
Lemma 21. For a morphism f : A → B in Set, the following are equivalent:
1. f is epic;
2. f is right-invertible, i.e. there exists g : B → A such that f ○ g = idB .
Thus all you need to do (Exercise) is prove one of these Lemmae, knowing that by doing so
you will also prove its dual.
Exercises
● Prove everything marked as Exercise in this Section.
● Prove the following. If you can prove a result by merely proving its dual, make sure you do.
1. In an arbitrary category C, every isomorphism is monic and epic.
2. The inclusion of Z in Q is a monic and epic morphism in the category of rings but it is not
an isomorphism. Thus Ring is not balanced.
3. The composition of monic (resp. epic) morphisms is a monic (resp. epic) morphism.
4. If g ○ f is monic (resp. epic) then f is monic (resp. g is epic).
5. If g ○ f is an isomorphism, then f is monic and g is epic.

49
II Universal objects
Initial and final objects

Definition 22. Let C be a category and A ∈ Obj (C ).

(i) A is an initial (or universal, or coterminal) object if, for every B ∈ Obj (C ), HomC (A, B )
has only one element.
(ii) dually, A is a final (or terminal) object if, for every B ∈ Obj (C ), HomC (B, A) has only one
element.
(iii) A is a zero or null object if it is both initial and final.
(iv) C is a pointed category if it contains a zero object.

We leave the proof of the following as an Exercise.

Lemma 23. A is an initial (resp. final) object of C iff, it is a final (resp. initial) object of C op.
Proposition 24. Every initial (resp. final) object is unique up to a unique isomorphism.
Corollary 25. If a zero object exists, then every initial object and every final object is isomorphic to
the zero object.
50
Definition 26. Let C be a category with a zero object, which we will denote o. For every pair of
objects A, B ∈ Obj (C ), we define the zero morphism 0AB : A → B by the composition of unique
morphisms (remember Def. 22 (iii)) A → o → B.

Lemma 27. Morphism 0AB : A → B does not depend on the zero object o chosen to define it. The
composition of any morphism with 0XY is a zero morphism as well.

Definition 28. Let C be a category with a zero object. Let f ∈ HomC (A, B ). A kernel of f ,
written ker f , is a morphism i : K → A such that f ○ i is the zero morphism 0KB and for
every morphism j : X → A such that such that f ○ j = 0XB , there exists a unique morphism
h : X → K such that i ○ h = j:

KO✤ ❆❆❆
✤ ❆❆
❆❆
f ○i=0
✤
✤
❆❆ i
❆❆
✤ ❆❆
❆❆
✤ ❆
✤ f $
h✤ >A B
/
:
✤ ⑥⑥
⑥⑥
✤
⑥⑥⑥
✤ ⑥⑥
⑥⑥
✤
⑥⑥
⑥ j
✤
⑥⑥ f ○j =0
X 51
● The dual of a kernel is called a cokernel, cokerf of f : A → B in a category C with a zero
object. You can fill out the blanks yourselves (Exercise):

⑧⑧?
1 ●✤
⑧⑧ ✤
⑧⑧⑧ ✤
⑧⑧ ✤
⑧⑧⑧ ✤
⑧⑧ ✤
⑧⑧
● /
● ❄❄
❄❄
✤
✤
❄❄ ✤
❄❄ ✤
❄❄ ✤
❄❄
❄❄ ✤
❄❄ ✤

-
●
The following are left as Exercises as well:

Lemma 29.
(i) If it exists in C, every kernel (resp. cokernel) is unique up to isomorphism, i.e. any two kernels
(resp. cokernels) of f : A → B are isomorphic in a certain category.
(ii) Every kernel is a monic morphism. Thus every cokernel is an epic morphism.
(iii) ker (coker (ker f )) = ker f and coker (ker (cokerf )) = cokerf .
52
Example: in categories Grp, Ab, Ring, R−Mod, K−Vect kernels and cokernels exist and coincide
with the definitions familiar to each of those categories:
– in Grp and Ab, given a morphism f : G1 → G2, the inclusion morphism

K  i / G1 for the subgroup K = {x ∈ G1 : f (x) = eG2 } ⊂ G1



x ✤ /
x
is a kernel (according to Def. 28 above) of f . In the usual context of Group Theory, we use
the same terminology kernel for both K (which we also label ker f ) and i : K → G1.
– Same applies to f : A → B in Ring, R−Mod, K−Vect and in general, categories whose ob-
jects have an underlying additive abelian group (more on this in the Exercises section later):
ker f refers to either K = {x ∈ A : f (x) = 0B }, or injective morphism K ↪ A.
i

– For instance, a linear map between K-vector spaces f : V1 → V2 can be expressed (fixing
bases on each space) with a matrix A; the kernel of f coincides with the equally-called kernel
(or nullspace) of matrix A: ker f = ker A = {v ∈ V1 : Av = 0} or its inclusion morphism
in V1:

ker f = ker A = {v ∈ V1 : Av = 0} Ð→ V1
i

– In Ab, R−Mod and K −Vect, a cokernel of morphism f : M → N is the quotient object

N /f (M ) or the quotient map N → N /f (M ) defined by n ↦ [n] (see slide 58).
First examples of initial and final objects
1. If they exist, kernels (resp. cokernels) defined in the previous pages are final (resp. initial) objects
of certain categories. Exercise
2. In Set, ∅ is an initial object and every set with a single element is a final but not an initial object:

{a} this is the only possible function between

f
any given set S and a fixed singleton {a}
S /

x✤ /
a

{a} ❴*4
S but between a fixed {a} and any S
a✤ /
... there are #S functions
3. All groups with a single element G = {a} are zero objects of the category Grp (as well as Ab):

{a} this is the only possible morphism between

f
any given group S and group {a}
G /

x✤ /
a

{a} /
G and a group morphism must map the identity element
a✤ /
eG a of {a} to the identity element of G
4. In a poset (S, ≤) (see slide 42), the smallest element, if it exists, is an initial object, and the largest
element, if it exists, is a final object. Exercise
54
5. Let (G, ⋅) be a group and H be subgroup of H (i.e. x ⋅y ∈ H for every x, y ∈ H). The following
is one of the most important constructs in Group Theory and has counterparts in Ring, R−Mod
and other categories as well.
– Define the category C whose objects are morphisms to other groups, vanishing on elements
of H, and whose arrows are those rendering the resulting diagrams commutative:

Obj (C ) = {f ∈ HomGrp (G, G1) : H ⊂ ker f, i.e. f (H ) = {0G1 }}

⎧
⎪ ⎫
⎪
⎪
⎪
⎪
G ⎪
⎪
⎪
⎪ ⎪
2
⎪
⎪ ⎪
⎪
f = O✤

⎪ ⎪
2 ⑤⑤⑤⑤
✤

HomC G → G1, G → G2 = ⎨ϕ ∈ HomGrp (G1, G2) : ϕ ○ f1 = f2, i.e. G ❇❇❇ ✤ ϕ ⎬

⑤
f1 f2 ⑤⑤ ✤

⎪
⎪ ⎪
⎪
✤

⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
❇❇ ✤

⎪ ⎪
❇❇

⎪ ⎪
❇! ✤
f
⎩ ⎭
1
G 1
These indeed map objects of C to objects of C because if H ⊂ ker f1, then f1 (H ) = {0G1 }
thus f2 (H ) = ϕ (f1 (H )) = {0G2 }. Compositions work as well:
G k
f3rrrrrr9 3 a❈❈❈❈❈
r ❈❈
rrr
rrr f2 ψ ❈/
G ▲▲▲▲ G2 ψ○ϕ because (ψ ○ ϕ) ○ f1 = ψ ○ f2 = f3
▲▲▲
▲▲▲
ϕ ④④
④④=
▲▲▲ ④④
f1 % ④④
G1
and associativity in Hom (C ) follows from that of ○ in Grp. Thus C is a category.
– If H is a normal subgroup of G, i.e, its left cosets are equal to its right cosets,

a ⋅ H := {a ⋅ h : h ∈ H} = {h ⋅ a : h ∈ H} =: Ha, for every a ∈ G,

C defined above has an initial object: the one given by the quotient group (G/H, ⋆), where

G/H := {[a] = a ⋅ H : a ∈ G} and (a ⋅ H ) ⋆ (b ⋅ H ) := (a ⋅ b) ⋅ H

the actual initial object of C is the morphism (usually called canonical map)
πG / H
G /
G /H which does belong to Obj (C ) because H ⊆ H = ker πG/H
g✤ /
g⋅H
and the fact that it is indeed an initial object of C is called the Universal Property of the
Quotient Group (Theorem 1.7.1 in the Algebra Notes): for every morphism to an-
other group G → G1, there exists a unique morphism ϕ closing the diagram:
f

f
G ❊❊❊ ;✇
/
G1
❊❊ ✇
❊❊
π G /H "
❊❊
✇
✇
ϕ
G /H
 – For instance if G = Z with usual addition (abelian, thus all subgroups are normal) and
H = 3Z = (3) = {3k : k ∈ Z}, then the quotient is the group whose three elements are
three infinite subsets of Z:

G/H = Z/3Z = {{3k : k ∈ Z} , {3k + 1 : k ∈ Z} , {3k + 2 : k ∈ Z}} = {[0] , [1] , [2]}

and the addition table is exactly the same as for (Z3, +); they are indeed the same group.
– Any other morphism f : Z → G1 vanishing on multiples of 3 would have to factor through
πZ/3Z in a unique manner according to the Universal Property. For instance, morphism

f
(Z, +) (S3, ○)
/

1 2 3 n where S3 = {{1, 2, 3} → {1, 2, 3} bijective}

n ✤ /
2 3 1
factors uniquely through the quotient map πZ/3Z : Z → Z3 as follows:
n
f 1 2 3
n ✞✤●●● /
for instance 2 ≠ 5 but [2] = [5] and
●● 2 3 1
●●
●●
●● ♣♣♣♣♣
8 they will both be mapped to the
●●
same element by ϕ because
♣♣
●● ♣♣♣
π ●
Z3 ●●●●● ♣♣♣♣
ϕ
♣♣♣
●● ♣♣
1 2 3 2 1 2 3 5 1 2 3
✵♣♣♣♣ = =
[n] = {3k + n : k ∈ Z}
●●
#
2 3 1 2 3 1 3 1 2
– The same construction and the same Universal Properties will apply to quotients of R-
modules by their submodules, with the knowledge that left cosets will always be equal to
right cosets (because (M, +) will always be an abelian group),

m + N = {m + n : n ∈ N } = {n + m : n ∈ N } = N + m,
and thus the quotients will always exist as modules and produce initial objects of C defined
three slides ago, but this time with R−Mod:
f
M ●●● /
M1
M /N = {[m] = m + N : m ∈ M } , ●● :✈
●● ✈
●● ✈
π M /N #
● ✈
✈
∃!f˜ if N ⊂ ker f
M /N
– And given that a ring R is always a module over itself and its submodules are the ideals of R,
the same applies in Ring:
f
R ❈❈❈ /
R1
R/I = {[r] = r + I : r ∈ R} , ❈❈ ②<
❈❈ ②
π R /I !
❈❈
②
②
∃!f˜ if I ⊂ ker f
R /I
58
6. Let M, N be two R-modules. Let us define an important algebraic construct categorically.
– The Cartesian product M ×N can be endowed with an R-module structure with operations

(m 1 , n 1 ) + (m 2 , n 2 ) = (m 1 + m 2 , n 1 + n 2 ) , r ⋅ (m, n) = (r ⋅ m, r ⋅ n) ;
we will see more on this when we talk about products in categories.
– A bilinear map M × N → P is any function between R-modules M × N → P such that
1. f (m1 + m2, n) = f (m1, n) + f (m2, n) for every m1, m2 ∈ M, n ∈ N ;
2. f (m, n1 + n2) = f (m, n1) + f (m, n2) for every m ∈ M, n1, n2 ∈ N ; and
3. f (r ⋅ m, n) = f (m, r ⋅ n) for every m ∈ M, n ∈ N and r ∈ R.
– Define C by

Obj (C ) = {bilinear maps M × N → P : P ∈ Obj (R−Mod)}

⎧
⎪ ⎫
⎪
⎪
⎪
⎪
ϕ ∈ HomR−Mod (M, N ) : ϕ ○ f = g ⎪
⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪ f s9 P✤
⎪
⎪
HomC M × N → P, M × N → P ′ = ⎨ ⎬
ss ✤
f g ss s

⎪ ⎪
ss

⎪ ⎪
ss ✤

⎪
⎪ ⎪
⎪
✤
M × N ✤ϕ
⎪
⎪ ⎪
⎪
❑❑❑

⎪
⎪ ⎪
⎪
❑❑❑❑ ✤

⎪ ⎪
❑❑❑
g ❑% ✤
⎩ P′ ⎭
Then C has an initial object π⊗ : M × N → M ⊗R N . This R-module M ⊗R N is called
the tensor product of M and R.
– What does this mean? It means any bilinear map M × N → P factorizes in a unique way
as a composition of the distinguished bilinear map π⊗ : M × N → M ⊗R N and a linear
map (i.e. an R-module morphism) M ⊗R N → P . Thus the initial nature of M ⊗R N is
evident (pay attention to the arrows, and see π⊗ : M × N → M ⊗R N as a single entity):

e ar❤❤❤❤4 M ⊗✤ R N
in
bil ❤❤❤❤
π ❤⊗❤❤❤❤❤ ✤
❤❤❤❤ ✤
M × N ❱❱❱❱❱❱ ✤
✤ ∃ ! f˜ linear
❱❱❱❱ ✤
❱
f biline❱❱❱❱❱❱❱❱❱❱❱ ✤
ar ❱❱❱+
P
– The actual definition of M ⊗R N and π⊗ can be seen e.g. in §3.6 in the Algebra Notes
but suffice it to say that π⊗ maps every pair (m, n) to a single element m ⊗ n (hence the
reduction from bilinear to linear) in such a way that
(m1 + m2)⊗n = m1⊗n+m2⊗n, m⊗(n1 + n2 ) = m⊗n1 +m⊗n2 , r⋅(m ⊗ n) = (rm)⊗n = m⊗(rn)

– That C does have an initial object π⊗ : M × N → M ⊗R N (defined in §3.6.1 and §3.6.2

of the Algebra Notes) is called the Universal Property of the Tensor Product and
proved in Prop. 3.6.12 of said notes.
– An example that is easier to relate to at the beginning would be Rn ∈ Obj (R−Vect): given
two vectors, expressible in the canonical basis {e1, . . . , en},
x1 y1
x= ⋮ = x1 e 1 + ⋅ ⋅ ⋅ + xn e n , y= ⋮ = y 1 e 1 + ⋅ ⋅ ⋅ + yn e n ∈ R n ,
xn yn

you can express π⊗ (x, y) in the following form

x1 y 1 x1 y 2 ⋯ x1 y n
x2 y 1 x2 y 2 ⋯ x2 y n
x ⊗ y := ⋮ ⋮ ⋱ ⋮ , (4)
xn y 1 xn y 2 ⋯ xn yn

in which case a basis of Rn ⊗R Rn will consist of n2 matrices expressed in the following form
1 0 ⋯ 0 0 1 ⋯ 0 0 0 ⋯ 1
0 0 ⋯ 0 0 0 ⋯ 0 0 0 ⋯ 0
e1 ⊗ e1 = ⋮ ⋮ ⋱ ⋮ , e1 ⊗ e2 = ⋮ ⋮ ⋱ ⋮ , ⋯ e1 ⊗ en = ⋮ ⋮ ⋱ ⋮ ,
0 0 ⋯ 0 0 0 ⋯ 0 0 0 ⋯ 0
0 0 ⋯ 0 0 0 ⋯ 0 0 0 ⋯ 0
1 0 ⋯ 0 0 1 ⋯ 0 0 0 ⋯ 1
e2 ⊗ e1 = ⋮ ⋮ ⋱ ⋮ , e2 ⊗ e2 = ⋮ ⋮ ⋱ ⋮ , ⋯ e2 ⊗ en = ⋮ ⋮ ⋱ ⋮ ,
0 0 ⋯ 0 0 0 ⋯ 0 0 0 ⋯ 0
⋮ ⋮ ⋱ ⋮
0 0 ⋯ 0 0 0 ⋯ 0 0 0 ⋯ 0
0 0 ⋯ 0 0 0 ⋯ 0 0 0 ⋯ 0
en ⊗ e1 = ⋮ ⋮ ⋱ ⋮ , en ⊗ e2 = ⋮ ⋮ ⋱ ⋮ , ⋯ en ⊗ en = ⋮ ⋮ ⋱ ⋮ ,
1 0 ⋯ 0 0 1 ⋯ 0 0 0 ⋯ 1
and any element of the form (4) can be expressed in a unique manner as a linear combination
of the above: x ⊗ y = x1y1 (e1 ⊗ e1) + x1y2 (e1 ⊗ e2) + . . . .
– Does that mean that every element in Rn ⊗R Rn can be expressed as x ⊗ y ?
Not at all; the set S = {x ⊗ y : x, y ∈ Rn} generates Rn ⊗R Rn, meaning every element
in the vector space Rn ⊗R Rn can be expressed as a combination of elements of S, albeit not
in a unique manner–you need a basis for that additional uniqueness property, and you have
one in the preceding slide.
– For instance: if n = 3, the following element of R3 ⊗R R3 can be expressed in terms of the
elements of S in different ways
4 3 11 1 0 1 4 1 4 0 8
T= 12 6 30 = 0 ⊗ 1 + 3 ⊗ 2 = 1 ⊗ 3 + 1 ⊗ 3
4 6 14 4 1 1 10 3 11 −1 19

whereas in the basis {ei ⊗ ej } it can only be expressed in one way:

T = 4e1 ⊗ e1 +3e1 ⊗ e2 +11e1 ⊗ e3 +12e2 ⊗ e1 +6e2 ⊗ e2 +30e2 ⊗ e3 +4e3 ⊗ e1 +6e3 ⊗ e2 +14e3 ⊗ e3

and it cannot be expressed as a single generator x ⊗ y :

4 3 11 x1 y 1 x1 y 2 x1 y 3 try solving this system for the xi, yj and
12 6 30 ≠ x2 y 1 x2 y 2 x2 y 3
4 6 14 x3 y 1 x3 y 2 x3 y 3 you will realize there is no solution
– Needless to say, you can express elements ∑i xi ⊗ y i ∈ Rn ⊗R Rm, n ≠ m, in matrix form as
well; only difference will be the matrices xiyj will not be square. Exercise: try some
i,j
examples yourselves.
– An equivalent method of representing these elements of Rn ⊗Rm is by way of single column
vectors instead of matrices (pay attention to the order in this particular representation):
x1 y 1
x1 y 2
⋮
x1 y m
x2 y 1
x2 y 2
x1 y1
⋮
x⊗y = ⋮ ⊗ ⋮ = x2 y m
x3 ym
⋮
⋮
xn y 1
xn y 2
⋮
xn y m

and this will also be reflected in the expression of the basis elements (which will also be vec-
tors of length nm) and the rest of the arguments written before will apply here as well. For
instance, in this expression the above example for n = m = 3 reads
4
12
4
1 0 1 4 1 4 0 8 3
0 ⊗ 1 + 3 ⊗ 2 = 1 ⊗ 3 + 1 ⊗ 3 = 6
4 1 1 10 3 11 −1 19 6
11
30
14

i.e. the matrix of the previous slide stacked in one column (this is not the only way to do it).
– The reason why both methods (matrices or long vectors) are equivalent to one another and
equally useful to represent elements of Rn ⊗R Rm will not be dealt with in too much detail
here; suffice it to say that they are a consequence of certain isomorphisms that hold true for
spaces of finite dimension (some of those can be found in §3.6 of the Notes).
– What happens if we have a bilinear function defined on Rn × Rm?
To make it easier, let us do it with an example: the following is a bilinear map (Exercise)

R2 × R3 B /
R4
x1 y 1 + x1 y2 + x2 y 3
y1
x1 2 x1 y 3 + x2 y 1 + 3 x 2 y 2 + x 2 y 3
x2 , y2 ✤ /
B ( x, y ) = 4 x1 y2 + x1 y 3 + x2 y 1 + x2 y 2 + x2 y 3
y3
x 2 y 1 + x2 y 3

the Universal Property of the Tensor Product (in other words, the fact that π⊗ : R2 × R3 →
R4 is an initial object of the category C defined earlier) implies the existence of a unique linear
map B̃ : R2 ⊗R R3 → R4 such that

B̃ (x ⊗ y) = B (x, y) for every (x, y) ∈ R2 × R3.

Exercise: find that map B̃. If you express x ⊗ y as elements of R6 it will be easier for you.
– Needless to say, things were easy here because Rn are vector spaces of finite dimension.
For general R-modules, visualizing tensor products will not be so simple.
Free objects
● Vector spaces over a field K always have a basis, i.e. a set of linearly independent generators.
● In the more general case of a module over any ring R, the existence of a basis is not guaranteed.
● Whenever an R-module does have a basis, though, we call it a free R-module.
● Among the many equivalent definitions of a free R-module, one stands to simple categorical
scrutiny. Let X be a set. Define DX by its objects and morphisms,
Obj (DX ) = {X → M : f ∈ HomSet (X, M )}
f

⎧
⎪ ⎫
⎪
⎪
⎪
⎪
M ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
= ✤
f ④④④ ✤
⎪ ⎪
④④

X → M, X → N = ⎨ ϕ ∈ HomR−Mod (M, N ) : ϕ ○ f = g, i.e X ❈❈❈ ✤ ϕ ⎬

f g ④④
④ ✤
HomDX
⎪
⎪ ⎪
⎪
✤

⎪
⎪
⎪ ⎪
⎪
⎪
⎪ N ⎪
❈❈ ✤

⎪ ⎪
❈❈ ✤
g ❈! 
⎩ ⎭
An initial object (Def. 22) of this category, X → R(X ), provides precisely a free R-module
with a basis having the same cardinality as X.
● The converse is true: every free module equals R(X ) for some set X (Th. 3.3.3 in Algebra Notes).
● Let us see this with an example whose outcome we already know: the R-vector space M = R3.
To prove that R3 is a free module of dimension 3, let us choose an arbitrary set whose cardinality
equals the dimension, i.e. X = {a, b, c}. Define τ to be the following set function:

X τ /
R3
1
a ✤ / 0
0
We deliberately choose these vectors because we know
0 they form a basis (commonly known as the canonical (5)
b ✤ / 1 basis)
0
0
c ✤ / 0
1

Consider any other vector space, e.g. N = R2, and any other set function, e.g.
f
X /
R2
1
a✤ /
2
2
b✤ /
3
−1
c✤ /
7

66
● Then there exists a unique linear (i.e. ∈ HomR−Mod) function f˜ : R3 → R2 closing the diagram
= ✤
⑤⑤ ✤
R3
τ ⑤⑤
⑤⑤
✤
⑤⑤
✤
X ❇❇❇ ✤ f˜
❇❇ ✤
❇❇ ✤
❇!
f
R2
We know we can find this function because Linear Algebra provides us with the tools to do so:
f˜
R3 /
R2
x x
y ✤ / 1 2 −1
y
z
2 3 7 z
What happens if we choose some other vector space N and some other function in X → N ?
We will still have a unique linear function f˜ : R3 → N closing the corresponding diagram.
Exercise: prove this as formally as you can.
● Thus X → R3 is an initial object of the category D{a,b,c}. This is how Category Theory
τ

characterizes the notion “R3 has a basis”. It also relates well to practical problems, because
τ ({a, b, c}) is indeed a basis.
● In the definition of τ , you could have chosen another set of 3 linearly independent vectors instead
of the canonical basis chosen in (5); everything stated above would still be true (albeit with a
different definition of f˜, think why) and τ ({a, b, c}) would still be a basis of R3.
● The logical question is: can we extend this construction to a general concrete category C?
Let X be a set and define, once again, DX by its objects and morphisms,

Obj (DX ) = {X → A : f ∈ HomSet (X, A) , A ∈ C}

f

⎧
⎪ ⎫
⎪
⎪
⎪
⎪
A ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
f ⑤> ✤

⎪ ⎪
⑤⑤ ✤
⑤⑤

X → A, X → B = ⎨ ϕ ∈ HomC (A, B ) : ϕ ○ f = g X ❇❇ ✤✤ ϕ ⎬
f g ⑤⑤⑤ ✤
HomDX
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
❇❇

⎪ ⎪
✤

⎪ ⎪
❇❇
g ❇ ✤

⎩ ⎭
❇
B

Definition 30. A free object generated by X in C is defined to be an initial object X → F in DX ,

i

i.e. for every function X → A there is a unique ψ ∈ HomC (F, A) such that ψ ○ i = f ,
f

F
> ✤
i⑤⑤⑤⑤⑤⑤ ✤
⑤⑤ ✤
✤
X ❇❇❇ ✤ ψ
❇❇ ✤
❇❇ ✤
f ❇ 
A
and if it exists, F ∈ Obj (C ) is unique save for isomorphism in C (or in DX ).
68
Examples
– It is clear from Th. 3.3.3 in Algebra Notes that R−Mod, hence also Ab, have free objects.
See Exercise 406 (p. 396) in those Notes for a proof of uniqueness up to isomorphism in
R−Mod, and think of how you would extend it to an arbitrary category.
– Same applies to the category of groups, Grp. Let X = {xi}i∈I be any set; define X =
{xi}i∈I to be a set of the same cardinality as X. Define Y := X ∪ X. A word in X is any
finite concatenation of elements of Y , i.e. elements of the form

a1⋯an, where for every k, ak = xi or ak = xi for some i ∈ I.

For instance, if X = {a, b, c, d}, then Y = {a, b, c, d, a, b, c, d} and examples of words
would be aabac, daccb, etc. We adopt the following reduction conventions on words:
∗ if xi and xi appear consecutively, they can be erased simultaneously, e.g. daccb → dab
∗ a word only consisting of an element xi and its counterpart xi (in any order) becomes the
empty word e; in other words, xi and xi are inverses of each other in the operation that
is concatenation. For instance, aa or aa can be reduced to the empty word.
The set of reduced words thus constructed is a free group FX where the operation ⋅ is
concatenation and the identity element is the empty word e. A basis of the group is X.
Exercise: write a formal proof of this.
69
Products
● Let C be a category and A1, A2 ∈ Obj (C ). Let A ∈ Obj (C ) and pi : A → Ai such that for any
B → Ai in C there exists a unique ϕ ∈ HomC (B, A) such the following diagram commutes:
fi

⑥⑥
B✤ ❆❆❆
⑥⑥ ✤ ❆❆
f1 ⑥⑥⑥⑥⑥ ✤ ❆❆
f
❆❆ 2
⑥⑥
✤ ϕ ❆❆
❆❆
⑥⑥⑥ ✤ ❆❆
⑥ ✤ ❆❆
~⑥
⑥ p p2
1
A1 o
A /
A2
We call A a product of A1 and A2 and write A = A1 × A2.
● A product of A1 and A2, if it exists, is unique up to isomorphism: given two of them, A → Ai
fi

f˜i
and à → Ai, the unique morphisms ϕ and ψ resulting from the above definition yield
A✤ ❅❅❅
⑦⑦
⑦⑦ ✤ ❅❅
⑦ ❅❅
f 1⑦⑦⑦⑦
⑦ ✤ ❅❅ 2f
⑦ ψ ✤✤ ❅❅
❅❅ and the same uniqueness applied to the outer diagram
⑦⑦⑦ ✤ ❅❅
f˜ f˜
⑦ ❅❅
implies ϕ ○ ψ = 1A . Same thing if you now place A in the
⑦
⑦  
1 2 /
A1 o
Ã✤ A2
_❅❅
❅❅
❅❅
✤ ⑦⑦⑦
?
middle row: you will obtain ψ ○ ϕ = 1Ã
❅❅ ✤ ⑦⑦
⑦⑦
❅❅
❅ ϕ ✤✤ ⑦⑦
⑦
f 1 ❅❅❅❅❅ ✤ ⑦⑦⑦⑦ f 2
⑦⑦
A
● More in general, let C be a category and {Ai}i∈I ⊂ Obj (C ) a set of objects. We define C ′ by

Obj C ′ = {{B → Ai}

fi
: B ∈ Obj (C ) , f ∈ HomC (B, Ai) , i ∈ I} ,
i∈I
HomC ′ {B → Ai} , {C → Ai} = {ψ ∈ HomC (B, C ) : fi = gi ○ ψ, ∀i ∈ I}
fi gi
i∈I i∈I
i.e. morphisms B → C in C rendering all diagrams below commutative:
B✤ ❇❇❇
✤ f
❇❇ i
❇❇
✤ ❇
for every i ∈ I.
✤
ψ ✤✤ Ai
⑤>
✤
⑤⑤⑤
✤
⑤⑤
⑤ g i
C

Definition 31. We say {C → Ai} ∈ Obj (C ) is a product of {Ai}i∈I if it is a final object of C ′.

πi
i∈I
We write C := ∏i∈I Ai. In other words, for any {B → Ai}
fi
∈ Obj (C ′) there exists a unique
i∈I
f ∈ HomC (B, ∏i∈I Ai) such that fi = πi ○ f for every i ∈ I:
fi
B ■■ 9
/
Ai
■ f tt
tt
■ tt
■
■$ tt
tt
π i
71
∏i∈I Ai
● Needless to say, in virtue of Proposition 24 (uniqueness of initial and final objects if they exist),
Lemma 32. If it exists, ∏i∈I Ai is unique up to isomorphisms in C.
Examples
– Set has a product ∏i∈I Ai→Ai: the Cartesian product with its projections on each set:
projj
∏ Ai := {(ai)i∈I : ai ∈ Ai} , ∏i∈I Ai /
Aj
i∈I (a i )i ✤ /
aj
– Grp, Ab, Ring, R−Mod have a product {proji : ∏i Ai → Ai}i∈I called the direct product,
which happens to coincide with the product they inherit as subcategories Set. This is proved
in the Algebra Notes in Theorem 2.6.21 for rings, Proposition 3.2.2 for modules, and is
left as an Exercise for groups. In all cases, its underlying set is the Cartesian product.
– Top has products as well: they are given by the product topology defined in §4.1.2 in the
Topology Notes.
– Let C = S where (S, ≤) is an ordered set. There exists a product of x1, x2 ∈ Obj (C ). Indeed,
we need x such that x ≤ x1, x ≤ x2 and if z ≤ x1, x2 then z ≤ x. A clear choice for x is
x = inf (x1, x2):
z◆
f1♣♣♣♣♣♣♣ ✤✤ ◆◆◆◆◆◆f◆ 2
✤ϕ
♣♣♣ ◆◆◆
x♣♣♣ π1  π ◆
2 ◆/ &
x1 o
x x2
– Cyclic groups are groups with a single generator, i.e. groups that can be expressed in the form

G = ⟨a⟩ = {a1 ⋯ an : n ∈ N, ai ∈ {a, a−1} for every i} .

They are easily proven to be objects of a full subcategory C of Ab.

There is no product in this category. We will prove this by verifying that there is no
product Z2 × Z2 in C. Such a product P → Z2 ought to be cyclic as well, hence P ≅ Zn
p 1 ,2

for some n, and there should be a unique ϕ rendering the following commutative:

idZ Z✤ 2 ▼▼▼ id
rrr
r ▼▼▼ Z
2
rrr ✤
ϕ 2 ▼▼▼
rrrrr ✤ ▼▼▼
▼&
x r
Z2 o
p1 Zn p2
/
Z2

implying pi 1 ≠ 0 for i = 1, 2, and a unique ψ rendering the following commutative:

idZ r
Z✤ 2 ▼▼▼
2 rrrr ▼▼▼
0
✤ ψ
rr ✤ ▼▼▼
rrr ▼▼▼
x rr
r ▼&
Z2 o
p1 Zn p2
/
Z2

Let m = ψ 1 . Then 1 = mp1(1) and 0 = mp2(1), thus m ≠ 0 and m = 0, absurd.

Coproduct
● Let us study the dual notion of a product. Let C be any category and {Ai}i∈I ⊆ Obj (C ).
We define the category C ′ such that

Obj C ′ := {{Ai → B} : B ∈ Obj (C ) , fi ∈ HomC (Ai, B ) , i ∈ I} ,

fi

and a morphism in C ′ between any two such objects, {Ai → B} Ð→ {Ai → C} , is

fi ψ gi
i∈I i∈I
defined as a morphism ψ ∈ HomC (B, C ) such that gi = ψ ○ fi for every i ∈ I, i.e.

> B
fi ⑤⑤⑤⑤⑤
⑤
⑤⑤
Ai ❇ ψ
❇❇
❇❇
gi ❇❇

C
Definition 33. We say {Ai → C} ∈ Obj (C ′) is a coproduct of {Ai}i∈I if it is an initial object
hi
i∈I
of C ′. We write C = ∐i∈I Ai . In other words, for any {Ai → B}
fi
∈ Obj (C ′) there exists a
i∈I
unique f ∈ HomC (∐i∈I Ai, B ) such that fi = f ○ hi for every i ∈ I:
fi
Ai ❏❏❏ ✉:
/
B
❏❏ ✉
❏❏ ✉
❏❏ ✉
✉
hi ❏% ✉ f
∐i∈I Ai
● Again, in virtue of Proposition 24,
Lemma 34. ∐i∈I Ai is unique up to isomorphisms in C.
Examples
– In the category C = Set the coproduct is the disjoint union with the obvious injections:
given X1, X2 ∈ Obj (Set), X1 ∐ X2 := X1 ⊎ X2–think of how you would inject X1, X2
into X1 ⊎ X2 with adequate morphisms.
– In Top the coproduct is the disjoint union with the corresponding natural topology.
– Ab has a coproduct: direct sum. Given a family of abelian groups, {Gi}i∈I , we define

⊕ Gi := {(xi)i∈I : xi ∈ Gi ∀i ∈ I, and only finitely many xi ≠ 0} .

i∈I
The injections
τj
Gj /
⊕i∈I Gi
where xj = gj and xi = eGi for i ≠ j
gj ✤ /
(xi)i∈I
render (⊕i∈I Gi, {τi}) the coproduct in Ab. Obviously if {Gi}i is finite then direct sum
and direct product coincide: G1 ⊕ ⋯ ⊕ Gn = G1 × ⋯ × Gn.
– In general, R−Mod has a coproduct: direct sum of R-modules defined as above, along with
the same injections τi : Mi → ⊕i Mi.
75
– Let us prove ⊕i∈I Gi cannot be the coproduct in Grp. Assume (Z ⊕ Z, {h1, h2}) is a
coproduct; these h1, h2 cannot be just anything, they are the following (Exercise: why?)
h1 h2
Z /
Z⊕Z Z /
Z⊕Z
m✤ /
(m, 0) n✤ /
(0, n)
Let f1, f2 : Z → S3 be given by n ↦ (1, 2, 3)n and n ↦ (1, 3)n respectively.
The existence of a morphism ψ : Z ⊕ Z → S3 such that f1 = ψ ○ h1, f2 = ψ ○ h2,

S
idZ♦♦♦♦♦♦♦ O✤✤ 3 id
7 g❖❖❖
❖❖❖Z
❖❖❖

♦♦♦
♦♦ ψ✤ ❖❖❖
❖❖❖
♦♦♦
Z h1
/
Z⊕Z h o
Z
2

would imply
(1, 2, 3) = f1 (1) = ψ (h1 (1)) = ψ (1, 0) , (1, 3) = f2 (1) = ψ (h2 (1)) = ψ (0, 1) ,
but (1, 1) = (1, 0) + (0, 1) = (0, 1) + (1, 0), hence
ψ (1, 1) = (1, 2, 3) (1, 3) = (2, 3) , ψ (1, 1) = (1, 3) (1, 2, 3) = (1, 2) ,

contradiction.
– Grp does have a coproduct, though; it is called the free product but it is constructed in a
different way and is usually not abelian even if the groups it is built from are.
Exercises
● Prove everything marked as Exercise in this Section.
● A pre-additive category C is one satisfying the following:
– for every A, B ∈ Obj (C ), HomC (A, B ) is an abelian group, written additively;
– composition of morphisms is Z-bilinear, i.e.

g + g ′ ○ f + f ′ = g ○ f + g ○ f ′ + g ′ ○ f + g ′ ○ f ′.
Prove that if C is a pre-additive category and A ∈ Obj (C ), the following are equivalent:
(i) A is an initial object;
(ii) A is a final object;
(iii) A is a zero object;
(iv) 1A = 0, i.e. the neutral element of (HomC (A, A) , +);
(v) (HomC (A, A) , +) is a trivial group, i.e. consisting of only one element.
● Let C be a pre-additive category. Given A1, A2 ∈ Obj (C ), a biproduct of A1, A2 is a family of
morphisms pj : B → Aj , ij : Aj → B such that pj ○ ij = 1Aj and i1 ○ p1 + i2 ○ p2 = 1B .
Prove that for a pair of objects {A1, A2} in a pre-additive category, the following are equivalent:
(i) {A1, A2} has a product;
(ii) {A1, A2} has a biproduct; 77
(iii) {A1, A2} has a coproduct.
● An additive category is a pre-additive category with a zero object such that every pair of objects
has a biproduct. An additive category such that
(i) every morphism has a kernel and a cokernel,
(ii) every monic morphism is a kernel,
(iii) and every epic morphism is a cokernel,
is called abelian.
Let C be an abelian category. Prove the First Isomorphism Theorem: every morphism f in C
factorizes in the form f = k ○ c, where k is monic and c is epic.
● Let R be a ring.
(i) Prove that for a morphism f : M → N in R−Mod, the following are equivalent:
(a) f is monic;
(b) ker f = {0M }, where ker f = {m ∈ M : f (m) = 0N } is the kernel of f .
(ii) State the dual of the above statement (which will automatically be true).

78
II Functors
First notions
Definition 35. Let C1 and C2 be two categories. A covariant functor (resp. contravariant functor)
is an assignation F : C1 → C2 defined as follows:
– for every A ∈ Obj (C1), F (A) ∈ Obj (C2);
– for every A, B ∈ Obj (C1), we have a correspondence (a class function)

HomC1 (A, B ) /
HomC2 (F (A) , F (B ))
f✤ /
F (f )

HomC1 (A, B ) /
HomC2 (F (B ) , F (A))
resp.
f✤ /
F (f )
such that for any morphisms f, g, and whenever composition makes sense,
∗ F (g ○ f ) = F (g ) ○ F (f ) (resp. F (g ○ f ) = F (f ) ○ F (g )), i.e.

F (f ) F (g ) F (f ) F (g )
F (A ) /
F (B ) /
2 F (C ) resp. F (A ) o l F (B ) o F (C )
F (gf ) F (gf )

∗ F (1A) = 1F (A) for every A ∈ Obj (C1).

● We can write a functor as F : C1 → C2. C1 is the domain of F and C2 is the codomain of F .
● These functions are, obviously, only set functions if Obj and Hom are sets for both categories.
Every contravariant functor C1 → C2 can be seen as a covariant functor C1 → C2 . Exercise.
op
●
● Functors preserve isomorphisms, i.e. if f is an isomorphism then so is F (f ). Exercise.
● Functors also preserve commutative diagrams. Exercise.
● Given two functors F : C → D and G : D → E, there exists a new functor G ○ F : C → E called
the composition of F and G, and such that:
– for every object A ∈ Obj (C ), (G ○ F ) (A) = G (F (A));
– for every morphism f ∈ HomC (A, B ), (G ○ F ) (f ) = G (F (f )).
The proof that this is indeed a functor this is very simple and can be completed by you
(Exercise).

80
Examples
● A constant functor C → C ′ maps all objects in C to a fixed object in C ′, and all morphisms in C
to the identity morphism of said fixed object: fixing C ∈ Obj (C ′),

C /
C′
A✤ /
C
f 1C
 
B C ✤ /

● Let C ′ ⊂ C be a subcategory. The inclusion functor, called the identity functor if C = C ′,

idC
C /
C′ C /
C
A✤ /
A A✤ /
A
f f f f
   
B ✤
B / B B ✤ /

is obviously covariant: F (f ) = f and arrow directions are maintained.

● The covariant functor F : Grp → Set that maps every group to its underlying set and every group
morphism to its underlying function,

Grp F /
Set
(G1, ⋅) ✤ /
G1
f f
 
(G2, △) ✤ /
G2
is called the forgetful functor because it “forgets” or “chooses to ignore” the structure of the
groups and morphisms and sees them as naked sets and functions.
● Similarly, there are forgetful functors on other categories we have seen, e.g. Ab → Set,
Top → Set, R−Mod → Set, R−Var → Set, etc.
● Forgetful functors also exist if we ignore one of the operations, e.g. Ring → Ab, R−Mod → Ab.
● Same applies to forgetful functor Top● → Top (we will define Top● later) or to “forgetting
about the metric” Met → Top where Met is the category of metric spaces whose morphisms
are continuous functions according to the classical δ − ϵ definition you already know.
82
● For a given ring R and a ∈ R, the following is a covariant functor:

R−Mod a⋅ /
R−Mod

M aM where f ′ (am) = af (m).


f 
f′
N aN
● Reduction modulo an ideal I ⊂ R is also a covariant functor:
R−Mod a⋅ / R/I−Mod

M M /IM where f˜ (x) = f (x).


f 
f˜
N N /IN
● Define R⋆ to be the set of units, i.e. invertible elements of a given ring R. This is a group with
respect to the second operation in R, thus we have a functor

Ring U /
Grp

R1 U (R1) = R1⋆ where U (f ) is the restriction f ∣R⋆ of f to R1⋆.

1
f U (f )
 
R2 U (R2 )
● Let C be any locally small category (remember Definition 4). Fix A ∈ Obj (C ).
Correspondence C → Set defined by B → HomC (A, B ) and
F F

C F /
Set
B1 /
HomC (A, B1) HomC (A, B1)
f⋆ /
HomC (A, B2)
where
f F (f ):=f⋆ g✤ /
f ○g
 
B2 /
HomC (A, B2)
is a covariant functor (Exercise) that is usually written F = HomC (A, ⋅).
We will see this is part of a more general type of functors: those called representable.
● A special case of the above is when C = R − Mod, the category of R-modules. Fixing M ∈
Obj (C ), assignation F = HomR (M, ⋅) : C → C given by M1 → HomR (M, M1) and
F (f ) = f⋆ : g ↦ f ○ g is thus covariant. In particular, if C = Ab then HomZ (G1, G2)
is an abelian group and we have have a functor Ab → Ab as well.

84
● Let us find the dual to the previous general example. Let C be locally small and A ∈ Obj (C ). The
assignation C → Sets defined by F (B ) = HomC (B, A) and
F

C F /
Set
B1 /
HomC (O B1, A) HomC (B2, A)
f⋆ /
HomC (B1, A)
where
f F (f ):=f ⋆ g✤ /
g○f

B2 /
HomC (B2, A)
is written F = HomC (⋅, A) and is a contravariant functor (check the arrows). Again, this is a
special case of what will be defined to be a representable functor.
● An example of the above. Let A = R with the Euclidean topology. We could see F =
HomTop (⋅, R) as a functor Top → Set, but given the existence of ring operations in R, we
can restrict the codomain (Exercise: why?):

Top F /
CRing
X /
HomTop(X, R) g ○O f
O

f F (f )

85
HomTop (Y, R)
❴
Y /
g
● Given C = R−Mod and M ∈ Obj (C ), F = M ⊗R ⋅ : C → C given by N ↦ M ⊗R N and
f
N❴ 1 ✤ /
N❴ 2
F F
 F (f )=idM ⊗f 
M ⊗R N1 ✤ /
M ⊗R N2
m ⊗ n1 ✤ /
m ⊗ f (n 1 )
● The dual version: in the same hypotheses as above, assignation F = ⋅ ⊗R M given by N →
N ⊗R M and f → f ⊗ idM is a contravariant functor.
● Let f : R → S be a ring morphism. Let C = R − Mod. Assignation F = S ⊗R ⋅ : C → C given
by
f
M1 /
M2
M → S ⊗R M, F F (6)
 F (f )=idS ⊗f 
S ⊗R M1 /
S ⊗R M2
is a covariant functor. Exercise: you can build the contravariant dual.
86
● A pointed topological space is simply a pair (X, x0) consisting of a topological space (X, τ )
and a distinguished point x0 ∈ X. The category of pointed spaces, Top●, is defined by
Obj (Top●) := {((X, τ ) , x0 ) : (X, τ ) ∈ Obj (Top) , x0 ∈ X} ,
HomTop● (((X, τX ) , x0 ) , ((Y, τY ) , y0 )) := {f ∈ HomTop ((X, τX ) , (Y, τY )) : f (x0 ) = y0 } .

– A loop or closed path based at x0 ∈ X is a continuous function γ : [0, 1] → (X, x0) such
that γ (0) = γ (1) = x0 (with the obvious understanding that τ[0,1] = τEuclidean):

γ X
x0

0 1

We usually call the image γ ([0, 1])of the loop (i.e. the curve) a loop and label it γ as well.
– Two loops γ1, γ2 based at x0 ∈ X are called homotopical to one another, γ1 ∼ γ2, if they
can be deformed into one another continuously without ceasing to be subsets of X.

87
– For instance all loops on the 2-sphere S2 are homotopical to the constant loop x0:

x0

– The torus T2, on the other hand, has loops that are not homotopical to one another:

γ x0

88
– Also loop γ ∗ γ which is basically γ run twice, is not homotopic to the original γ, but it is to
the loop β shown below:

x0
β

– This operation ∗ can also be extended to two different loops, e.g. taking γ and then α would
be homotopic to this:

89
– Important facts (not proven here):
○ operation ∗ is well-defined as long as all loops start and end in the same point;
○ homotopy is a relation of equivalence;
○ ∗ works well with equivalence classes, i.e. if γ1 ∼ γ2 and β1 ∼ β2, then β1γ1 ∼ β2γ2;
○ thus, if we denote the class of equivalence of all paths homotopic (deformable) to γ as [γ ]
and all paths homotopic to β as [β ], we can define an product of paths [γ ] ∗ [β ];
○ the set of equivalence classes by ∼ is a group with ∗ and is called the fundamental group
π1 (X, x0). There are higher-order homotopy groups πn (X ) but we will not study them.
– Examples
∗ All paths in S2 can be contracted to a single point. Thus π1 S2, x0 ≅ {1}, the group of a
single element.
∗ The torus T2 had two non-homotopic paths, γ and α, and everything else was homotopic
to a product of γ k and αj , k, j ≥ 0. An isomorphism of groups can be easily built:
ϕ
π1 T 2 , x0 , ∗ /
(Z × Z, +)
[ γ ] k ∗ [ α] j ✤ /
(k, j ) ,
i.e. mapping [γ ] ↦ (1, 0), [α] ↦ (0, 1), and path product transforms into additions of
integers, e.g. instance [γα] ↦ (1, 1), γ 2α ↦ (2, 1), γ −1α3 ↦ (−1, 3), etc.
– It appears as though we are building an assignation Top● → Grp which:
∗ maps every pointed topological space (X, x0) to its homotopy group π1 (X, x0);
∗ works well with morphisms in each category if every continuous function
f : (X, τX ) → (Y, τY ) such that f (x0) = y0, is corresponded with this function:
f∗
π1 (X, x0) /
π1 (Y, y0)
[γ ] ✤ /
[f ○ γ ]
which happens to be a morphism of groups and is well-defined because f ○ γ : [0, 1] → Y
is a closed loop based at y0 ∈ Y (I will leave the details to you if you are interested).
∗ maps compositions in Hom (Top●) to group morphism compositions: (g ○ f )∗ = g∗ ○f∗

In other words: we have a covariant functor, named the fundamental group functor:
π1
Top● /
Grp
(X, x0) /
π1 (X, x0) [γ❴ ]
f π1 (f )=f∗
  
(Y, y0) /
π1 (Y, y0) [f ○ γ ]
91
– This functor is an important example of a category probing into another–in this case it shows
light on what Algebra can “see” from Topology and translate into its own inner workings.
– A consequence of this: proofs of otherwise complicated results may become simpler. For
instance: the 2-sphere is not homeomorphic to the torus, i.e. S2 ≇ T2 in Top. Indeed,
∗ If they were, then the pointed topological spaces would be isomorphic in Top● as well:
S2, x0 ≅ T2, f (x0) for some homeomorphism f : S2 → T2.
∗ But if ≅ in Top●, given that functors preserve isomorphisms, their
fundamental groups would also have to be isomorphic in Gr.
∗ However, the fundamental groups are not isomorphic:

π1 = {1} ≇ Z × Z = π1

∗ Thus ≇ in Top● and that means ≇ in Top.

– A similar argument proves that the 1-sphere (i.e., the circle) S1 = is not homeomorphic
to or , because π1 = Z.

92
Faithful, full and dense functors
Definition 36. Let F : C1 → C2 be a covariant functor. For every A, B ∈ Obj (C1), we have a class
function
HomC1 (A, B ) /
HomC2 (F (A) , F (B ))
(7)
f✤ /
F (f )
We say
1. F is a faithful functor if (7) is injective;
2. F is a full functor if (7) is surjective;
3. F is fully faithful if the two above items hold;
4. F is dense or essentially surjective if for every B ∈ Obj (C2) there exists A ∈ Obj (C1) such
that F (A) ≅ B.

● If F is faithful and full, F reflects isomorphisms – in other words: F (f ) isomorphism implies

f isomorphism. Exercise
● Now we can properly define something that was left unfinished in Def. 5:
Definition 37. A concrete category is a pair (C, F ) where C is a category and F : C → Set is a
faithful functor.
93
Examples
● The forgetful functor F : C → Set whenever C = Grp, Ring, R − Mod is:
– faithful (thus Def. 37 above), because F (f ) = f for every morphism f in C, thus different
morphisms in C are mapped to different functions in Set.
– not full, because not every set function between two groups (or rings, or R-modules, etc) can
be a morphism in each of those categories. We saw an example of this in page 15.
● Forgetful functor Ring → Ab (i.e. discarding the second operation on rings) is faithful, but
neither full nor essentially surjective. Exercise.
● Let C ′ ⊂ C be a full subcategory. The inclusion functor,
C /
C′
A✤ /
A
f f
 
B ✤ /
B
is fully faithful.
● For instance, inclusion functor Ab ↪ Gr is fully faithful. It is not essentially surjective, however,
because not every group is abelian.
94
● Let Z → Q be a ring morphism. Let C = Z−Mod. Assignation F = Q ⊗Z ⋅ : C → C given by
f

M → Q ⊗Z M in (6) in page 86 is a covariant functor. It is faithful, dense and not full.

– Let M1 = M2 = Z in (6). We have a commutative diagram for every f ∈ HomC (Z, Z):
f
Z❴ ✤ /
Z❴
F F
 F (f )=idQ ⊗f 
Q ⊗Z Z ✤ /
Q ⊗Z Z
q ⊗ n1 ✤ /
q ⊗ f (n 1 )
and there is a number of ways of proving Q ⊗Z Z ≅ Q. The question stands on whether
F : HomAb (Z, Z) → HomAb (Q, Q) is surjective, and the answer is no. For instance,
homothety g = 21 ⋅ ⋆ : Q → Q cannot arise from any morphism f : Z → Z. Indeed, if
g = idQ ⊗ f then 12 = idQ ⊗ f (1 ⊗ 1) = 1 ⊗ f (1) , absurd. Hence Q ⊗Z ⋅ is not full.
– Any f ∈ HomQ (Q, Q) is determined over Z by f (1): f (n) = nf (1). Hence Q ⊗Z ⋅ is
faithful.
– Given any Q-vector space V , it is also an Z-module and V ≅ Q ⊗Z V . hence Q ⊗Z ⋅ is dense.
Note Q ⊗Z Q = Q.
95
● Correspondence ⋅ ⊗Z Z/ (2) : C → C given by M → M ⊗Z Z/ (2) is a non-faithful, full and
essentially surjective contravariant functor. The situation is, taking M1 = M2 = Z,
f
Z /
Z
F F
 
Z ⊗Z Z / (2)f ⊗ido
Z ⊗Z Z / (2)
O
Z/(2) O
≅ ≅
Z/ (2) Z/ (2)
and although ⋅2, ⋅4 : Z → Z are different morphisms, they both lead to the same morphism
⋅2 = ⋅4 = 0 : Z/ (2) → Z/ (2).
● Let Q → Q [X ] be a ring morphism. Let C = the category of Q-vector spaces. Correspondence
⋅ ⊗Q Q [X ] : C → C given by M → M ⊗Q Q [X ] is a contravariant functor which is non-dense,
faithful and full. Indeed, Q [X ] / (X ) = Q as a Q [X ]-module. The fact Q is a field implies
any Q-module F can be expressed F ≅ ⊕i∈I Q = Q(I ), hence F ⊗Q Q [X ] ≅ ⊕i∈I Q [X ] =
Q [X ](I ) and X ⋅ F ⊗Q Q [X ] ≠ 0.

96
 Definition 38.
– Let F : C → D a functor and X ∈ Obj (D). A universal arrow from object X to functor
F is a pair (UX , υ ) consisting of UX ∈ Obj (C ) and υ ∈ HomD (X, F (UX )), having the
following, called a universal property: for every A ∈ Obj (C ), if ϕ ∈ HomD (X, F (A)), there
exists a unique Φ : UX → A in C such that F (Φ) ○ υ = ϕ, i.e.

UX X ❋❋❋ υ /
F (U X )
✤ ❋❋
✤ ❋❋
❋❋
✤ ❋❋
✤ Φ ❋❋ F (Φ)
✤ ϕ ❋❋
❋❋
❋❋
✤ " 

A F (A )
in C in D
– In the special case in which D = Set, X = {x} and υ : {x} → F (UX ) is defined by
υ (x) = ι, we can drop the subindex from U and call (U, ι) a universal element of F .
The above definition can be dualized into that of a universal arrow from functor F to object X
(Exercise: write it in detail) and the property will still be called universal.
● The constructions in Section II can be characterized as universal arrows or elements as defined
above, for instance tensor products, quotient groups, quotient R-modules, free modules, etc.
Exercises
● Prove every statement marked Exercise in the Section.
● Prove that every faithful functor reflects commutative diagrams.
● Express the defining property of a free object in a category Cas a universal property according to
the last definition in this Section.
● Repeat the above with the tensor product of two R-modules.
● Prove that the functor F : URing → Grp defined by R ↦ R⋆ (the set of all invertible elements
of R) is not essentially surjective.
● Prove that the inclusion functor URing ↪ Ring is faithful but is neither full nor essentially
surjective.

98
IV. Natural transformations
● We have already seen that we can compose functors F : C1 → C2 and G : C2 → C3,

A ↦ (G ○ F ) (A) := G (F (A)) , f ↦ (G ○ F ) (f ) := G (F (f )) ,
and a “rule of signs” applies concerning their covariant-contravariant character (Exercise).
● Recall the identity functor defined at the beginning of the Examples subsection in III:
idC
C /
C
A✤ /
A
f f
 
B ✤ /
B
Definition 39. We say a functor F : C1 → C2 is an isomorphism of categories if a functor
G : C2 → C1 exists such that

G ○ F = idC1 and F ○ G = idC2 .

We then say the categories are isomorphic: C1 ≅ C2
For instance Z − Mod and Ab are isomorphic (Exercise).
99
● The above definition, however, can be restrictive and in reality we need a more general definition
characterizing two categories as being “basically the same”. We need to define a few tools first.
Definition 40. Let F, G : C1 → C2 be two covariant (resp. contravariant) functors. A
correspondence ϕ : F → G, sometimes written
F
ϕ : F Ô⇒ G or C1 ϕ C2
G
is a natural transformation or morphism of functors between F and G if for every A ∈ Obj (C1)
there exists ϕA ∈ HomC2 (F (A) , G (A)) (called the component of ϕ at A) such that for every
f ∈ HomC1 (A1, A2) the following diagram (called the naturality square or diagram) commutes
ϕA
1
F (A
O 1
) /
G (A
O 1
)
F (f ) G (f )
 ϕA 
2
F (A 2 ) /
G (A 2 )
double-sided arrows obviously adaptable to the situation, e.g. for covariant F and contravariant G,
/
O

 /
Examples
● For every functor F : C1 → C2, there exists an identity natural transformation idF : F ⇒ F
defined by mapping every object A ∈ Obj (C1) to (idF )A := 1F (A) (the identity morphism
F (A) → F (A) in C2). The naturality diagram is nearly trivial in this case:
(idF )A =1F (A )
1 1 /
F (A 1 ) F (A 1 )
F (f ) F (f )
 
F (A 2 ) /
F (A 2 )
(idF )A =1F (A )
2 2

101
● Let Mon be the category of monoids (Def. 19), i.e. same definition as groups but without
guaranteed invertibility of every element
Obj (Mon) = {all monoids (M, ⋆)} ,
HomMon ((M1 , ⋆) , (M2 , ⋅)) = {f : M1 → M2 : f (x ⋆ y ) = f (x) ⋅ f (y ) ∀x, y ∈ M1 , f (eM1 ) = eM2 }

Let n ∈ N, C the category of commutative and unital rings, and define the following two functors:
Matn U
C /
Mon C /
Mon
R1 ✤ /
Matn (R1) R1 ✤ /
U (R 1 )
f Matn (f ) f U (f )
   
R2 ✤ /
Matn (R2) R2 ✤ /
U (R 2 )
where
– Matn (R) is the monoid of n × n matrices with coefficients in ring R,
– Matn (f ) is the morphism that maps every matrix ai,j to matrix f ai,j ,
i,j i,j
– U ((R, +, ⋅)) := (R, ⋅), because every ring is a monoid with its second operation,
– U (f ) = f for every ring morphism.
Our intention is to find a natural transformation Matn ⇒ U .
– Define the following correspondence between the two functors
Matn

C det Mon (8)

U
where for every ring (R, +, ⋅), detR : Matn (R) → U (R) is the function mapping every
matrix A = ai,j to its determinant det A (defined similarly as it would if R = R).
i,j
– The fact that detR (AB ) = detR (A) detR (B ) and detR (Idn) = 1 (easy to check from
the definition of a determinant) implies detR : Matn (R) → U (R) is a monoid morphism.
– The naturality diagram whose commutativity we wish to prove is
detR
1
Matn (R1) /
U (R1 )
Matn (f ) U (f ) (9)
 
Matn (R1) det /
U (R2 )
R2

for every A ∈ Matn (R1), U (f ) detR1 (A) = f detR1 (A) ; basic Linear Algebra
implies f detR1 (A) = detR1 f (A) for every ring morphism f : R1 → R2. Thus
detR2 ○Matn (f ) = U (f ) ○detR1 , hence (9) commutes and (8) is a natural transformation.
● For every R-module M , define M ⋆ = HomR−Mod (M, R) to be the dual of M and being a set
of morphisms between two modules, it naturally has an R-module structure itself with operations

(f1 + f2) (m) := f1 (m) + f2 (m) , (r ⋅ f1) (m) := r ⋅ f1 (m) .

Let us prove that there is a natural transformation between the bi-dual functor F : M ↦ M ⋆⋆,
and the identity functor idR−Mod. To be more precise, for every M ∈ Obj (R−Mod), this entails
F (M ) = HomR (HomR (M, R)). For every f : M → N , passage to the dual is contravariant:
f⋆
M⋆ o N⋆
α○f o ✤
α
Hence F (f ) = f ⋆⋆, F (M ) = M ⋆⋆ is covariant. For us to have a natural transformation to 1
we need τ such that for every M ∈ Obj (R−Mod) there exists τM : HomR (M, F (M )) such
that for every f ∈ HomR−Mod (M, N ) the following diagram commutes:
τM
M /
F (M )
f F (f )
 
N τ N F (N )
/

104
For every M ∈ Obj (R−Mod), define
τM τ M (m )
M /
F (M ) = M ⋆⋆ M⋆ R
where
/

m✤ /
τ M (m ) f✤ /
f (m )
τM (m) is a morphism of R-modules:
τ M ( m ) ( f 1 + f 2 ) = ( f 1 + f 2 ) ( m ) = f 1 ( m ) + f 2 ( m ) = τ M ( m ) ( f 1 ) + τ M (m ) ( f 2 ) ,
τM (m) (af ) = (af ) (m) = af (m) = aτM (m) (f ) .
Hence τM is a morphism of R-modules: τM (m1 + m2) = τM (m1) + τM (m2) and
τM (am) = aτM (m) . Indeed, let us show this for the first identity,
τM (m1 + m2 ) (f ) = f (m1 ) + f (m2 ) = τM (m1 ) (f ) + τM (m2 ) (f ) = [τM (m1 ) + τM (m2 )] (f ) .

For every f ∈ HomR−Mod (M, N ) we have a commutative diagram

τM
M /
M ⋆⋆ m
❴
✤ /
τM ❴(m)
f 
f ⋆⋆
 
N τN
/
N ⋆⋆ f (m ) ✤ /
τN (f (m))

Indeed, every f : M → N induces M ← N given by α ○ f α. Hence M → N ⋆⋆,

⋆ f⋆ ⋆ ⋆⋆ f ⋆⋆

ω ↦ f ⋆⋆ (ω ) : N ⋆ → R by g ↦ ω (g ○ f ). For every g ∈ N ⋆,
f ⋆⋆ (τM (m)) (g ) = τM (m) (g ○ f ) = g (f (m)) and τN (f (m)) (g ) = g (f (m)) ,

hence both are equal, thus τN (f (m)) = f ⋆⋆ (τM (m)).

● The following example is slightly harder than the ones above. Given ideal I ⊂ R and F = ⊗R
R/I, G = /I : R−Mod → R−Mod define M ↦ ϕM by
ϕ
M ⊗R R/I ≅M / M /IM
m⊗1✤ /
m
This is a natural transformation between F and G. We have a commutative diagram

M ⊗R R/I ≅ /
M /IM
f ⊗1 f
 
N ⊗R R/I ≅ /
N /IN
This is precisely the naturality diagram for the correspondence.

106
Definition 41. We say natural transformation ϕ : F → G is an natural isomorphism or a natural
equivalence or an isomorphism of functors if ϕA is an isomorphism for every A ∈ Obj (C1). In
such case we call F and G (naturally) isomorphic and write F ≅ G, and we can define the inverse
natural equivalence
ϕ−A1
1
F (A
O 1 ) o
G (A
O 1
)
F (f ) G (f )
 
F (A 2 ) o G (A 2 )
ϕ−A1
2
Lemma 42.
(i) The composition of natural transformations is a natural transformation.
(ii) The composition of natural isomorphisms is a natural isomorphism.
(iii) ϕ−1 ○ ϕ = idF and ϕ ○ ϕ−1 = idG.
We can now write that general characterization of two categories as “basically the same”:
Definition 43. Let C1, C2 be two categories and F : C1 → C2 a functor. We say F is an equivalence
of C1 to C2 if there exists a functor G : C2 → C1 such that G ○ F ≅ idC1 and F ○ G ≅ idC2 . We say
C1 and C2 are equivalent and write C1 ≈ C2.
More examples
● The identity transformation idF : F ⇒ F is always a natural isomorphism.
● Let (G, ⋆) ∈ Obj (Grp) be any group. Define the opposite group (Gop, ⋆op) as having the same
underlying set as G and operation x ⋆op y := y ⋆ x. This is a group as well (Exercise) and we
can always define the following functor:
op
Grp /
Grp
op
G1 ✤ /
G1
where f op := f.
f f op
op
 

G2 ✤ /
G2
Let us prove functors idGrp and op are naturally isomorphic. All we need for every G is an
isomorphism ηG : G → Gop such that the naturality diagram
ηG op
1
G1 /
G1

f f op
op
 

G2 ηG
/
G 2
2
108
is commutative.
Hence we need to find an isomorphism of groups ηG : G → Gop for every G, such that
f op ○ ηG1 = ηG2 ○ f for every group morphism f : G1 → G2.
– For every group G, define ηG : G → Gop by ηG (g ) := g −1.
= g implies ηG is a morphism G → Gop and is
−1
– The fact (gh)−1 = h−1g −1 and g −1
equal to its own inverse (ηG ○ ηG = idG), thus is an isomorphism as well.
– For every g ∈ G1, group morphism properties imply that
−1
η G 2 ○ f (g ) = f (g ) = f g −1 = f ηG1 (g ) = f ○ ηG1 (g ) = f op ○ ηG1 (g )
and this means the naturality square is always commutative, thus idGrp ≅ op

109
● Let V be an n-dimensional K-vector space. Let C be the following category with a single element

Obj (C ) = {V } , HomC (V, V ) = HomK (V, V ) = {all linear maps V → V } ,

and C ′ = K−Vect the category of all n-dimensional K-vector spaces. Then inclusion

C F /
C′
V ✤ /
F (V ) = V
f F (f )=f
 
V ✤ /
F (V ) = V
is an equivalence of categories. Indeed, define G : C ′ → C by G (W ) := V for every W ∈
Obj (C ′). There always exists an isomorphism of vector spaces of dimension n (again, basic Linear
Algebra) ϕW : W → V and G (f ) are defined to render the diagram below commutative,

W ϕ≅ / V = G (W )
W
f G (f ) i.e. G (f ) := ϕW ′ ○ f ○ ϕ−W1
 
W ′ϕ ≅ / V = G (W ′ )
W′

and F G ≅ idC ′ and GF ≅ idC are now an easy Exercise. 110

● Recall the forgetful functor F : Grp → Set defined by G ↦ G on objects (every group is sent
to its underlying set) and f ↦ f on morphisms (every group morphism is sent to the underlying
set function). Then F is naturally equivalent to the functor HomA (Z, ⋅).
● This can be generalized to a very important type of functors that were already mentioned in the
previous Section:

Definition 44. Let C be a category. We call covariant (resp. contravariant) functor

F : C → Sets representable if it is naturally isomorphic to HomC (A, ⋅) (resp. HomC (⋅, A)) for
some A ∈ Obj (C ). A representation of F is a pair (A, ϕ) such that ϕ : HomC (A, ⋅) → F (resp.
ϕ : HomC (⋅, A) → F ) is a natural isomorphism.
Example
– Let G be a group and H ⊂ G a subgroup. Define the functor F of the category of groups
defined by F (G′) := {f ∈ Hom (G, G′) : f ∣H = 0}. Then F is a representable functor.
Exercise.

111
 Definition 45. Let C be a category. We define a skeleton of C as any full (as in Def. 7) subcategory
constructed by picking one representative from every isomorphism class of objects of C; these represen-
tatives will be the objects of the skeleton.
● The following is a very important and useful characterization:
Theorem 46. Let F : C1 → C2 a covariant functor. The following are equivalent:
1. F is an equivalence of categories.
2. F is a full, faithful and essentially surjective functor.
In particular, every category is equivalent to its skeleton and two categories are equivalent if and only
if their skeletons are isomorphic.
● For instance, this makes the proof that the inclusion functor for finite-dimensional vector spaces
in slide 110 is an equivalence of categories, much easier. Exercise.
● We will prove 1. ⇒ 2. in Theorem 46 and leave the other implication as an Exercise. Let
F : C1 → C2 and G : C2 → C1 such that G○F ≅ idC1 and F ○G ≅ idC2 . For every B ∈ Obj (C2),
ϕ τ
τB
F (G (B )) ≅ B and G (B ) ∈ Obj (C1), hence F is dense. Similarly so is G.

112
– Given A1, A2 ∈ Obj (C1) we have a commutative diagram
ϕA
1
G (F (A1)) ≅
/
A1
G(F (f )) G(F (g )) f g
 ϕA 
G (F (A2)) ≅ 2 /
A2
with the understanding GF (f ) = ϕ−A1 ○ f ○ ϕA1 .
2
– Assume F (f ) = F (g ). Then G (F (f )) = G (F (g )), hence f = g. Thus F is faithful.
Similarly G is faithful.
– Let g : F (A1) → F (A2). We need h : A1 → A2 such that F (h) = g. For h to exist the
diagram
ϕA
G (F (A1)) ≅ 1 /
A✤ 1 o ≅ G (F (A ))
ϕA 1
G (g )  ✤ 1 G(F (h))
ϕA ✤ h 
G (F (A2)) ≅ 2 /
A2 o ≅ G (F (A ))
ϕA 2
2

must be commutative, and to that effect we define h := ϕA2 ○ G (g ) ○ ϕ−A1 which is the
1
same bijection HomC1 (A1, A2) ≅ HomC1 (GF (A1) , GF (A2)) we used earlier. Then
G (g ) = G (F (h)), and the fact G is faithful implies g = F (h). Hence F is full.
Exercises
● Prove everything marked as an Exercise in this section.
● Let C be the category of sets and define, for every A ∈ Obj (C ) and any set map f : A → B,

P (A) := {B : B ⊂ A} , P (f ) = f −1
(i) Prove P defines a contravariant functor P : C → C.
(ii) Let X = {0, 1} ∈ Obj (C ). Prove P is naturally equivalent to functor HomC (⋅, X ).
● Let R be a ring, I an ideal and M an R-module. Let F (M ) := {x ∈ M : Ix = 0}.
(i) Prove that F (M ) is a submodule of M isomorphic to HomR (R/I, M ).
(ii) Prove that F induces a covariant functor in the category of R-modules, and that it is natu-
rally equivalent to HomR (R/I, ⋅).
● Let R be a ring. Then the following covariant functors of the category of associative, commu-
tative R-algebras with neutral elements to the category Ab of abelian groups are representable
(Exercise):
(i) Gm (B ) := B ⋆, the multiplicative group of invertible elements of B.
(ii) µn (B ) := {x ∈ B : xm = 1}, multiplicative group of n-th roots of unity in B.
(iii) Ga : (B ) = B ⋆, the additive group of B.
114
● Prove that the natural transformation between F ⇒ idR−Mod between the bi-dual functor F :
M ↦ M ⋆⋆ and the identity functor shown in the earlier Examples is a natural isomorphism.
● Let A, B, C, D be categories and H : C → A, F, G : A → B, T : B → D covariant functors. Let
τ : F → G be a natural transformation of functors. Prove that correspondences C ↦ τH (C ),
C ∈ Obj (C ) and A ↦ T (τA), A ∈ Obj (A), define natural transformations, respectively, σ :
F H → GH, ρ : T F → T G.
● If R is a unital ring, we have R-module isomorphisms between M and HomR−Mod (R, M )
given by m ↦ ϕm : r ↦ r ⋅ m and ϕ ↦ ϕ (1). Prove they define a natural isomorphism
idR−Mod ≅ HomR−Mod (R, ).
● Prove that categories R − Mod and Matn (R) − Mod are equivalent.

115
 TOPICS IN ALGEBRA
AND GEOMETRY
Second Coursework
M31448 –Academic Session 2021–2022

Instructions
a) Answer all four exercises.
b) This is worth 50% of your total mark for this unit.
c) You must undertake this assignment individually.
d) All claims and statements must be justified.
e) Bonus marks will only be counted if the rest of the given exercise has ≥ 20 marks.
f) Submission method: by Moodle through the available dropbox.
g) Submission deadline: January 28, 2022, at 4PM

1. Let Grp be the category of all groups. Consider the correspondence

G 7→ F (G) = {a ∈ G : ag = ga, for every g ∈ G} , for every G ∈ Obj (Grp) . (1)

(i) (10 marks) Show that the above definition on objects cannot be extended to morphisms in such a
way as to define a covariant functor in F : Grp → Grp.

(ii) (5 marks) Define C with the same objects, but having epimorphisms as the only morphisms. Prove
C is a subcategory of Grp.

(iii) (10 marks) Prove there does exist a functor C → Grp with the same definition on objects as (1).

2. Let R be the category of unitary rings and G the category of groups.

(i) (15 marks) Prove that the functor F1 : R → G defined on objects by F1 (R) := R⋆ and restriction of
morphisms is neither faithful nor full.

(ii) (10 marks) Now define a functor F2 : R → G which is faithful, using GLn (R), the general linear
group of degree n ≥ 2, as a tool.

(iii) (Bonus 10 marks) Is the functor you found in (ii) an equivalence of categories? Prove your claim in
the most economical way possible.
2 ACADEMIC SESSION: 2021–2022

3. Let n ∈ N. For every group G, define On (G) := {g ∈ G : g n = eG }.

(i) (10 marks) Extend the above definition on objects to a functor On : Grp → Set.

(ii) (15 marks) Prove On is naturally isomorphic to HomGrp (Zn , ·)

4. Let R be an integral domain and R − Mod the category of R-modules. Let T be the torsion functor on
this category which takes every M to T (M ).

(i) (15 marks) Build a natural transformation τ of T to the identity functor on R − Mod.

(ii) (10 marks) Study whether τ can be a natural isomorphism for some ring R.

(iii) (Bonus 10 marks) Let F be the functor which maps every module M to the quotient M/T (M ).
Build a natural transformation σ of the identity functor on R − Mod to F and study for which rings
this is a natural isomorphism.