Notes for Introduction to Lattice theory

Yilong Yang
May 18, 2013

Abstract
This is a note for my talk Introduction to Lattice Theory. I have a talk in Math DUG about this
topic. In that talk I managed to introduce the section 2,3 and 4.

Contents
1 Introduction to Category Theory 2

2 Introduction to Lattice 3

3 Modular Lattice and Distributive Lattice 7

4 Some Relation to Group Theory 10

5 Some relation to topology 11

6 Birkhoff s Representation Theorem 12

7 Stone Duality for Bounded Distributive Lattice 15

1
1 Introduction to Category Theory
Definition 1.1. A category C consists of a class (not necessarily a set) of objects Obj(C), a class of
morphism Hom(C) and two functions domain : Hom(C) Obj(C), codomain : Hom(C) Obj(C), and a
partially defined function composition : Hom(C) Hom(C) Hom(C) (usually represented by a circle ),
such that the following holds:
1. f, g Hom(C), composition is defined for the pair (f, g) iff domain(f ) = codomain(g). Then
domain(f g) = domain(g) and codomain(f g) = codomain(f ).
2. composition is associative.
3. A Obj(C), idA Hom(C) such that domain(idA ) = codomain(idA ) = A, idA f = f for all f
with codomain A, and f idA = f for all f with domain A.
Definition 1.2. For any A, B C, we shall use Hom(A, B) to denote the class of morphisms of Hom(C)
with domain A and codomain B.
Definition 1.3. Two objects A, B C are said to be isomorphic iff f Hom(A, B), g Hom(B, A)
such that f g = idB , g f = idA . Morphisms f, g are called isomorphisms
Definition 1.4. A (covariant) functor F from a category C to a category D is a function of class
F : Obj(C) Obj(D) and a function of class F : Hom(C) Hom(D) such that the following hold:
1. domain(F) = F(domain), codomain(F) = F(codomain). i.e. for any A, B C, F(Hom(A, B))
Hom(F(A), F(B)). (F respects domain and codomain.)
2. For each A C, F(idA ) = idF (A) . (F respects the identity.)
3. For any f, g Hom(C) with domain(f ) = codomain(g), F(f g) = F(f ) F(g). (F respects
composition.)
Definition 1.5. A contravariant functor F from a category C to a category D is a function of class
F : Obj(C) Obj(D) and a function of class F : Hom(C) Hom(D) such that the following hold:
1. domain(F) = F(codomain), codomain(F) = F(domain). i.e. for any A, B C, F(Hom(A, B))
Hom(F(B), F(A)). (F interchanges domain and codomain.)
2. For each A C, F(idA ) = idF (A) . (F respects the identity.)
3. For any f, g Hom(C) with domain(f ) = codomain(g), F(f g) = F(g) F(f ). (F interchanges
composition.)
Definition 1.6. Let F, G : C D be covariant functors. A natural transformation is a function
: Obj(C) Hom(D) such that the following hold:
1. For any A Obj(C), (A) Hom(F(A), G(A)).
2. For any f Hom(A, B) Hom(C), (B) F(f ) = G(f ) (A).
Definition 1.7. Let F, G : C D be contravariant functors. A natural transformation is a function
: Obj(C) Hom(D) such that the following hold:
1. For any A Obj(C), (A) Hom(F(A), G(A)).
2. For any f Hom(A, B) Hom(C), (A) F(f ) = G(f ) (B).
Definition 1.8. Two (covariant or contravariant) functors F, G : C D are said to be naturally isomor-
phic if there exist natural transformations : F G, 0 : G F such that 0 (A) = idG(A) , 0 (A) =
idF (A) for any A Obj(C). and 0 are called natural isomorphisms.

2
 To show natural isomorphism, the following criteria is often used.
Proposition 1.9. A natural transformation : F G for functors F, G : C D is a natural isomorphism
iff X is an isomorphism for each X Obj(C).
0
Proof. Since X : F(X) G(X) is an isomorphism for each X Obj(C), we can define X = (X )1 :
0 0
G(X) F(X). Then clearly (A) = idG(A) , (A) = idF (A) for any A Obj(C).
Proposition 1.10. Compositions of covariant functors or of contravariant functors are covariant functors.
Proof. Suppose F : C D, G : D E are covariant functors. Then G F(domain) = G(domain(F)) =
domain(G F), G F(codomain) = G(codomain(F)) = codomain(G F), G F(idA ) = G(idF (A) ) =
idGF (A) , G F(f g) = G(F(f ) F(g)) = (G F(f )) (G F(g)). So G F : C E is a covariant
functor.
Suppose F : C D, G : D E are contravariant functors. Then G F(domain) = G(codomain(F)) =
domain(G F), G F(codomain) = G(domain(F)) = codomain(G F), G F(idA ) = G(idF (A) ) = idGF (A) , G
F(f g) = G(F(g) F(f )) = (G F(f )) (G F(g)). So G F : C E is a covariant functor.
Definition 1.11. The identity functor for a category C is a functor idC : C C such that idC : Obj(C)
Obj(C) and idC : Hom(C) Hom(C) are both identity functions.
Definition 1.12. Two categories C, D are equivalent if there exist covariant functors F : C D, G : D C
such that G F is natually isomorphic to the identity functor idC , and F G is naturally isomorphic to idD .
Two categories are dual if there exist contravariant functors F : C D, G : D C such that G F is
natually isomorphic to the identity functor idC , and F G is naturally isomorphic to idD .

From now on, we shall abuse the notation and use A C to mean that A Obj(C).

2 Introduction to Lattice
Let us first define a lattice in an algebraic way.

Definition 2.1. An algebraic lattice (L, , ) is a set L with two binary operations meet and join, and
, such that both operations are commutative and associative, and the absorption law holds. i.e. a, b, c L,
1. a b = b a, a b = b a (Commutivity)
2. (a b) c = a (b c), (a b) c = a (b c) (Associativity)

3. a = a (a b) = a (a b) (Absorption law)
Proposition 2.2. In an algebraic lattice, both operations are idempotent. i.e. a L, a a = a a = a.
Proof. For any a L, pick any b L and let c = a b. Then a a = a (a (a b)) = a (a c) = a. The
same for a a = a.

Proposition 2.3. In an algebraic lattice L, for any a, b L, we have a b = a iff a b = b.

Proof. If a b = a, then a b = (a b) b = b by absorption law. Conversely if a b = b, then
a b = a (a b) = a by absorption law.
Now, to have the category of algebraic lattice, we need to define algebraic lattice homomorphism.

Definition 2.4. A function of algebraic lattices f : L L0 is a (algebraic lattice) homomorphism iff

f (ab) = f (a)f (b), f (ab) = f (a)f (b). A homomorphism is an isomorphism if it has a homomorphism
g : L0 L such that f g = idL0 and g f = idL .

3
Definition 2.5. The category of algebraic lattices La has algebraic lattices as its objects, and homo-
morphisms as the morphisms, with domain, codomain, composition defined in the obvious way.
Finally, whenever we have an algebraic structure, we can always seek to define the associated algebraic
substructures.

Definition 2.6. A sublattice is a nonempty subset L0 of an algebraic lattice L, such that L0 is closed under
joint and meet.
Now we shall proceed to give another definition of lattice, and show that the two definition gives us the
same thing, in the sense that the category of these new objects would be equivalent to the category of lattice
in a canonical way. First of all we shall construct the category of posets.

Definition 2.7. A partial order is a relation on a set S that is reflexive, antisymmetric and transitive.
i.e. a, b, c S,
1. a a
2. a b, b a implies a = b

3. a b, b c implies a c
Definition 2.8. A poset (partially ordered set) (P, ) is a set P together with a partial order on it.
For any S P , we have the following definition:
1. an upper bound of S is any element a P such that a b, b S.

2. an lower bound of S is any element a P such that b a, b S.

3. the maximum of S, max S, is any element a S such that a b, b S.
4. the minimum of S, min S, is any element a S such that b a, b S.

5. the suprimum of S, sup S, is any element a P such that a = min{a0 P : a0 is an upper bound of S}.
6. the infimum of S, inf S, is any element a P such that a = max{a0 P : a0 is a lower bound of S}.
Note that by antisymmetry, max, min, sup and inf are all unique if exist.
Remark 2.9. As a convention, every element of P is an upper bound and a lower bound of the empty set.

Definition 2.10. An order-preserving function is a function of posets f : P Q such that f (a) f (b)
whenever a b. An order-reversing function is a function of posets f : P Q such that f (a) f (b)
whenever b a. An order-preserving function f : P P 0 is an isomorphism of posets if there is an
order-preserving function g : P 0 P such that f g = idP 0 and g f = idP .
Definition 2.11. The category of posets, P os, consists of posets as objects and order-preserving functions
as morphisms.
Now we proceed to find a subcategory of P os.
Definition 2.12. An order lattice is a poset (L, ) such that for any a, b L, sup{a, b} and inf{a, b}
exist.

Proposition 2.13. Let L be a lattice. Then for any finite subset S L, sup S and inf S exist.
Proof. Done by induction.
Now we want morphisms for the category of order lattice.

4
Definition 2.14. A (order lattice) homomorphism is a function f : L L0 preserving supremum and
infimum of finite sets. i.e. if S L is a finite set, then f (sup S) = sup f (S). A homomorphism of order
lattices f : L L0 is an isomorphism of there is a homomorphism g : L0 L such that f g = idL0 and
g f = idL .
Proposition 2.15. A homomorphism of order lattice is order-preserving.

Proof. Suppose a b. Then sup{a, b} = a. So sup{f (a), f (b)} = f (sup{a, b}) = f (a). So f (a) f (b).
The converse of the above proposition is in general false. However, it is true under some special cases.
One of the cases is shown below.
Proposition 2.16. If a map of order lattices f : L L0 is an isomorphism of posets, then it is a homo-
morphism of order lattices.
Proof. Let S be a finite subset of L. We know sup S a for all a S. As f is order-preserving, we have
f (sup S) f (a) for all a S. So f (sup S) sup f (S).
Then similarly we can show that g(sup f (S)) g f (sup S). Now since g f = idL , we have g(sup f (S))
sup S. So f (sup S) f g(sup f (S)) = sup f (S). This gives us the inequality in the other direction. So
f (sup S) = sup f (S). Similarly we have f (inf S) = inf f (S).
Definition 2.17. The category of order lattices Lo is the subcategory of P os with order lattices as
objects and order lattice homomorphisms as morphisms.
Theorem 2.18. La and Lo are cannonically equivalent.

Proof. We shall define functors F : La Lo and G : Lo La .

Step 1: For each L La , we let F(L) be the same set, but we define a b iff a b = a. Note that by
Lemma 2.3, we also have a b iff a b = b. Let us show that this is an order lattice.
First, since is idempotent, we have a a = a, so a a, is reflexive. If a b and b a, then
a = ab = ba = b. So is antisymmetric. If a b and b c. Then ac = (ab)c = a(bc) = ab = a.
So a c, and is transitive. So F(L) is indeed a poset.
Let a, b be any elements of L. Then (a b) b = a (b b) = a b = (a a) b = a (a b), so
a b a, b. So a b is an upper bound of {a, b}. On the other hand, if c a, b, then c a = c and c b = c.
Then c (a b) = (c a) b = c b = c. So c a b. So a b = sup{a, b}.
On the other hand, by absorption law we have a (a b) = a, b (a b) = b, so a b is a lower bound
of {a, b}. Now if a c and b c, then c a = c and c b = c. Thus c (a b) = (c a) b = c b = c. So
a b c. So a b = inf{a, b}.
So we indeed have F(L) Lo .
Step 2: For any f Hom(L, L0 ) Hom(La ), we define F(f ) Hom(F(L), F(L0 )) Hom(Lo ) to be
the same function as f as function of sets. Let S = {a1 , ..., an } be any finite subset of L. Then by induction
we have sup S = a1 ... an . So f (sup S) = f (a1 ) ... f (an ) = sup f (S). So F(f ) is an order lattice
homomorphism.
F(idL ) = idF (L) and F(f g) = F(f ) F(g) are clear from definition. So F : La Lo is a covariant
functor.
Step 3: For each L Lo , we let G(L) be the same set, but we define a b = sup{a, b}, a b = inf{a, b}.
Then the commutativity is clear, and (a b) c = sup{a, b, c} = a (b c) and (a b) c = inf{a, b, c} =
a (b c), so we have associativity. Finally, we know a b = sup{a, b} a. So a (a b) = inf{a, a b} = a.
Similarly we have a (a b) = sup{a, a b} = a since a inf{a, b} = a b. So G(L) La .
Step 4: For any f Hom(L, L0 ) Hom(Lo ), we define G(f ) Hom(G(L), G(L0 )) Hom(La ) to be the
same function as f as function of sets. Then f (ab) = f (sup{a, b}) = sup{f (a), f (b)} = f (a)f (b), f (ab) =
f (inf{a, b}) = inf{f (a), f (b)} = f (a) f (b).
G(idL ) = idG(L) and G(f g) = G(f ) G(g) are clear from definition. So G : Lo La is a covariant
functor.

5
 Step 4: G F : Hom(La ) Hom(La ) and F G : Hom(Lo ) Hom(Lo ) are identities by definition.
So we only need to show that G F : Obj(La ) Obj(La ) and F G : Obj(Lo ) Obj(Lo ) are identities as
well.
For any L La , then F(L) has a partial order , and G F(L) has operations 0 and 0 . Then
a b = sup{a, b}. On the other hand, we know (a b) a = a and (a b) b = b, so a b sup{a, b} = a 0 b.
0

For any c a, b, we have c a = c and c b = c. So c (a b) = (c a) b = c b = c. So c a b. So

a 0 b = sup{a, b} a b. So = 0 . Similarly = 0 . So G F(L) = L
For any L Lo , then G(L) has operations , . Then F G(L) has a partial order 0 . Then suppose
a 0 b. Then a b = a. So sup{a, b} = a, and therefore a b. On the other hand suppose a b. Then
a b = sup{a, b} = a, and thus a 0 b. So =0 . So F G(L) = L.
So Lo and La are canonnically equivalent.

The above theorem shows that algebraic lattice and order lattice are the same thing. From now on, we
shall only talk about lattices and lattice homomorphism, and the category of lattices L as a subcategory of
P os, and we shall talk about operations and partial orders interchangably.
We will conclude this section by giving several important examples of lattices. We need to clarify some
terminology of sublattices first.

Proposition 2.19. Arbitrary nonempty intersections of sublattices are sublattices

Proof. Let L0 = iI Li be an intersection of sublattices with arbitrary index set I. Then if a, b L0 , we
T
must have a, b Li for each i. Then a b, a b Li for each i. So a b, a b L0 . So L0 is a sublattice.
Definition 2.20. A sublattice generated by a subset S L of a lattice L is the intersection of all
sublattices containing S.

Definition 2.21. An interval [a, b] in a lattice L with a, b L is the subset {c L : a c b}.

Proposition 2.22. An interval [a, b] is nonempty iff a b.
Proof. If a b, then because a a and b b, we must have a, b [a, b].
Conversely, if [a, b] is nonempty, then there exists c such that a c b. So a b.

Proposition 2.23. A nonempty interval is a sublattice.

Proof. If c, d [a, b], then c, d b and c, d a. So c d = sup{c, d} b and c d = inf{c, d} a. So
b c d c d a. So c d, c d [a, b].
Here are some example of lattices

Example 2.24.

1. The subsets of a set form a lattice, with inclusion being the partial order. The join and meet are union
and intersection.
2. The finite subsets of a set form a lattice, with inclusion being the partial order. The join and meet are
union and intersection.
3. The partitions of a set form a lattice, where a b iff a is a refinement of b.

4. The subgroups of a group form a lattice, with inclusion being the partial order. The join of two
subgroups is the subgroup generated by the two subgroups, and the meet of two subgroups are their
intersection.
5. The normal subgroups of a group form a lattice, with inclusion being the partial order. The join
of two normal subgroups is the product of the two normal subgroups, and the meet of two normal
subgroups are their intersection.

6
 6. The subrings of a ring form a lattice, with inclusion being the partial order. The join of two subgroups
is the subgroup generated by the two subgroups, and the meet of two subgroups are their intersection.
7. The ideals of a ring form a lattice, with inclusion being the partial order. The join of two ideals is
their sum, and the meet of two ideals are their intersection.

8. The open subsets of a topological space form a lattice, with inclusion being the partial order. The join
and meet are union and intersection.
9. The closed subsets of a topological space form a lattice, with inclusion being the partial order. The join
and meet are union and intersection.
10. Integers form a lattice, with its usual ordering. In fact, any set with a total order is a lattice.

11. Positive integers form a lattice, where a b iff a is a multiple of b. The join and meet are the least
common multiple and greatest common divisor.
12. Pairs of integers form a lattice, where (a, b) (c, d) iff a c and b d. We have (a, b) (c, d) =
(max{a, c}, max{b, d}), (a, b) (c, d) = (min{a, c}, min{b, d}).

13. Any finite poset is a lattice iff it has a maximum and a minimum.
14. The sublattices of a lattice together with the empty set form a lattice, with inclusion being the partial
order. The join of two sublattices is the sublattice generated by the two sublattices, and meet of two
sublattices is their intersection.

15. The intervals of a lattice form a lattice, with inclusion being the partial order. We have [a, b] [c, d] =
[a c, b d], [a, b] [c, d] = [a c, b d].
16. For any lattice L with partial order , we can define a partial order 0 on L such that a b iff b 0 a.
Then (L, 0 ) will be lattice, the dual lattice of L.

3 Modular Lattice and Distributive Lattice

Note that for lattice of subsets of a set, we have (A B) (A C) = A (B C) and (A B) (A C) =
A (B C), i.e. we have the distributive law. On the other hand, consider the lattice of subgroups of the
Klein four grou V4 = ha, b|a2 = b2 = (ab)2 = ei. Let c = ab. Then hai (hbi hci) = hai V4 = hai, but
(hai hbi) (hai hci) = hei hei = hei the trivial subgroup. So the distributive law fails.
To see better what the distributive law does, we first prove a general identity for lattices.
Proposition 3.1. For any a, b, c in a lattice, (a b) (a c) = a ((a b) (a c)) and (a b) (a c) =
a ((a b) (a c)).
Proof. We know a a b, a c. Therefore a sup{a b, a c} = (a b) (a c). So a ((a b) (a c)) =
(a b) (a c). The other identity can be shown in exactly the same way.
For the first identity, the statement of the law of distribution is that we can replace the a b and a c
on the right hand side of the above identity to b, c. As we have seen, this is not true for general lattices.
However, for some lattice we may have half of the distributive law, i.e. we may replace one of a b, a c
with b, c, but not both at the same time. This is called the modular law.

Definition 3.2.

1. The distributive law is (a b) (a c) = a (b c) and (a b) (a c) = a (b c). A lattice where

the distributive law holds is a distributive lattice.

7
 2. The modular law is (a b) (a c) = a ((a b) c) = a (b (a c)) and (a b) (a c) =
a ((a b) c) = a (b (a c)). A lattice where the modular law holds is a modular lattice.
We first explore the modular law. Note that the current modular has four identities, making it somewhat
cubersome. We shall show that all four identities are equivalent, and show some another identity equivalent
to the modular law.
Proposition 3.3. In a lattice, the following identies are equivalent:
1. (a b) (a c) = a ((a b) c).
2. (a b) (a c) = a (b (a c)).
3. (a b) (a c) = a ((a b) c).
4. (a b) (a c) = a (b (a c)).
5. a (b c) = (a b) c whenever a c.
6. a (b c) = (a b) c whenever a c.
If any one of the identities above holds for a lattice, then the lattice is modular.
Proof. Identity 6 is just identity 5 with left hand side and right hand side swapped and positions of a, c
swapped. So they are trivially equivalent. We can also see that identity 2 is just identity 1 with b, c
swapped, and identity 4 is just identity 3 with b, c swapped. So identity 2 is equivalent to 1, and identity 4
is equivalent to 3.
Assume identity 5 and 6. Then because a ab, we have a((ab)c) = a(c(ab)) = (ac)(ab).
So identity 1 holds, and identity 2 holds. Conversely, assume identity 1 and 2. Then suppose a c, we have
a (b c) = (c a) (c b) = c ((c a) b) = c (a b) = (a b) c. So identity 5 holds. So identities
1,2,5,6 are equivalent. Similarly identities 3,4,5,6 are equivalent. So we are done.
Now we will explore yet another equivalent condition to the modular law, the diamond isomorphism.
Note that for each pair of elements a, b in a lattice L, we always have a a b and a b b. So we have
sublattices [a, a b] and [a b, b]. We can define fa : [a b, b] [a, a b] such that fa (c) = c a, and we can
define gb : [a, a b] [a b, b] such that gb (c) = c b. It is clear that fa , gb are order-preserving functions,
but they are not necessarily homomorphisms, and they are in general not inverse to each other.
Definition 3.4. In a lattice L, we say that the diamond isomorphism holds for a, b L if fa , gb defined
above are isomorphisms of sublattices [a, a b], [a b, b].
Proposition 3.5. A lattice L is modular iff the diamond isomorphism is true for all a, b L.
Proof. Suppose the diamond isomorphism is true for all pairs of elements in L. For any a, b, c L with
a c, then we have b a b c b. So a b [b, b c]. So gc (a b) = (a b) c. Now since fb is the
inverse of gc , we have a b = fb ((a b) c). On the other hand, we have b c a (b c) c (b c) = c.
So a (b c) [b c, c]. And we have fb (a (b c)) = b a (b c) = b a = fb ((a b) c). Since fb is an
isomorphism, it must be injective. So a (b c) = (a b) c. So the lattice is modular.
Conversely, suppose the lattice is modular. For any a, b L, for any c [a, a b], we have fa gb (c) =
a(bc). Since c [a, ab], we have a c. So by modular law aw have fa gb (c) = a(bc) = (ab)c = c
since c a b. For any c [a b, b], we have gb fa (c) = b (a c) = (c a) b. Since c b, by modular
law we have gb fa (c) = (c a) b = c (a b) = c since c a b. So fa , gb are inverse of each other, and
thus they are isomorphisms.
Recall that we have second isomorphism theorem of groups, stating that N H/N ' H/(N H) for normal
subgroups N, H of a group G. This is indeed the diamond isomorphism. We also have similar theorems for
rings and modules. So we have the following corollaries.

8
Corollary 3.6. The lattice of normal subgroups, the lattice of ideals and the lattice of submodules are always
modular.
Again we now start with a bunch of equivalent identities for any lattice.
Proposition 3.7. In a lattice, the following identies are equivalent:

1. (a b) (a c) = a (b c).
2. (a b) (a c) = a (b c).
3. (a b) (b c) (c a) = (a b) (b c) (c a).

If any one of the identities above holds for a lattice, then the lattice is distributive.
Proof. Suppose identity 1 is true. Then for any a c, we have a (b c) = (a c) (b c) = (a b) c, so the
lattice is modular. Then for any a, b, c, we have (ab)(bc)(ca) = (b(ac))(ca) = ((b(ac))c)
((b(ac))a) = (c(b(ac)))(a(b(ac))). Now we know a, c ac. So by the modular law we have
(ab)(bc)(ca) = (c(b(ac)))(a(b(ac))) = (cb)(ac)(ab)(ac) = (ab)(bc)(ca).
So identity 3 is true. Similarly identity 2 implies identity 3 as well.
Suppose identity 3 is true. Suppose a c. Then a (b c) = (a c) (b c) = (a c) (b c)
(a b) = (a c) (b c) (a b) = (a c) (a b) = (a b) c. So the lattice is modular. Now
for any a, b, c, we consider the diamond isomorphism for the pair a, b c. Let f = fbc and g = ga .
We clearly have a b c (a b) (a c) a a = a. So (a b) (a c) [a b c, a]. Then
f ((a b) (a c)) = (a b) (b c) (c a) = (a b) (b c) (c a). Because g is the inverse of f , we
have (a b) (a c) = g((a b) (b c) (c a)) = (a b) (b c) (c a) a = (b c) a. So identity 1
holds. So identity 3 implies 1. And similarly identity 3 also implies 2.
Definition 3.8. A lattice satisfies the cancellation law if whenever a b = c b and a b = c b, we have
a = c.
Proposition 3.9. A distributive lattice satisfies the cancellation law.

Proof. Suppose the lattice is distributive, and suppose a b = c b and a b = c b. Then c = c (c b) =

c (a b) = (c a) (c b) = (c a) (c b) (c b) = (c a) (c b) (a b), and the same is true for
a as well. So a = c.
In fact, the converse to above is also true, as we shall show later.
For now, all the criteria we have for modular lattice and distributive lattice are highly algebraic. We
shall try to work out some more geometric criteria as well. Let us first show some examples of non-modular
lattice and non-distributive lattice.
Definition 3.10. The lattice N5 is defined to be a set {a, b, b0 , c, d} where a is larger than everything else,
and d is smaller than everything else. b b0 , and the pair (b, c), (b0 , c) have no ordering relation. The lattice
M3 is defined to be the lattice isomorphic to the subgroup lattice of the Klein four group V4 , i.e. it is a
set {a, b, c, d, e} where a is larger than everything else, e is smaller than everything else, and b, c, d have no
ordering relation between them.
Proposition 3.11. N5 is not modular (and thus not distributive), and V4 is modular but not distributive.
Proof. N5 is specifically constructed such that the diamond isomorphism fails for the pair b, c. So it is not
modular. M3 is modular because it is the subgroup lattice of an abelian group (which is the same as the
normal subgroup lattice of that group). But M3 is specifically constructed such that the cancellation law for
the triple b, c, d fails. So it is not distributive.
Theorem 3.12 (Dedekind). A lattice is modular iff it contains no sublattice isomorphic to N5 .

9
Proof. The necessity is trivial, as the sublattice isomorphic to N5 will immediately give us a counterexample
to the modular law. Let us show sufficiency. Suppose that the lattice L is not modular. Then there exist
x, y, z L with x z such that x (y z) 6= (x y) z. Let a = x y, b = (x y) z, b0 = x (y z),
c = y, d = y z. Then I claim that {a, b, b0 , c, d} form a sublattice isomorphic to N5 .
We have to first show that a, b, b0 , c, d are distinct, and they have the right ordering. Note that if x y,
then x y z. So x (y z) = y z = (x y) z, contradicting our assumption. If we have y z, then
x y z. Then (x y) z = x y = x (y z), contradicting our assumption.
Now b, b0 are distinct by assumption. Since x y x and x y y y z, we have x y x (y z).
We also have z x and z y z. So z x (y z). So we have b = (x y) z x (y z) = b0 .
We clearly have d = y z x (y z) = b0 . If d = b0 , then x y z y, contradiction. So d 6= b0 . We
also have a = x y (x y) z = b. If a = b, then x y z, and thus y z, contradiction. So a 6= b. So
we have a b b0 d, all distinct.
Now clearly a = x y y y z = d. So we have a c d. If a = c, then y x, contradiction. So
a 6= c. Suppose c = d. Then y z, contradiction. So c 6= d.
Now we only need to show that b, c have no order and b0 , c have no order. Suppose b0 c, then x y,
contradiction. Suppose b c, then y z, contradiction. So there are no possible order relations. So
{a, b, b0 , c, d} is indeed a sublattice isomorphic to N5 .

Theorem 3.13 (Birkhoff). A lattice is distributive iff it contains no sublattices isomorphic to N5 or M3 .

Proof. The necessity is trivial. For sufficiency, suppose the lattice is not distributive. If the lattice is not even
modular, the by last theorem there is a sublattice isomorphic to N5 . Suppose the lattice is modular. Since it
is not distributive, we can find elements x, y, z such that (x y) (y z) (z x) 6= (x y) (y z) (z x).
Now let a = (x y) (y z) (z x), e = (x y) (y z) (z x), b = (x a) e, c = (y a) e,
d = (z a) e. I claim that {a, b, c, d, e} form a sublattice isomorphic to M3 .
Now x y x y, x z, z y, so x y e. The same is for y z and z x. Thus we have a e. I
claim that b c = c d = d b = a, b c = c d = d b = e. If this is the case, then a b, c, d e. And
suppose b = a, then e = b c = c and similarly e = d, but then c d = e 6= a, contradiction. So b 6= a.
Similarly c 6= a, d 6= a, b 6= e, c 6= e, d 6= e. Then it must follows that b, c, d must be mutually distinct and
have no order relation. Then indeed {a, b, c, d, e} is a sublattice isomorphic to M3 .
We see that b = (x a) e = (x (x y) (y z) (z x)) e = (x (y z)) e = e (x (y z)).
Because e y z, by modular law we have b = (e x) (y z) = (x (y z)) (y z), and again by
modular law we have b = (x (y z)) (y z) as well.
Now bc = (x(yz))(yz)(y(xz))(xz) = (x(yz))(y(xz)) = ((yz)x)(y(xz)) =
(yz)(x(y(xz))) = (yz)(xy)(xz) = e. Similarly cd = db = e, and bc = cd = db = a.

Corollary 3.14. A lattice is distributive iff the cancellation law holds.

Proof. If the lattice is not distributive, then we can find a sublattice {a, b, c, d, e} isomorphic to M3 . Then
b c = b d and b c = b d, but c 6= d. So the cancellation law fails.
Corollary 3.15. The set of positive integers with a b iff a is a multiple of b is a distributive lattice.

Proof. We shall show that the cancellation law holds. Suppose a c = b c and a c = b c. Then let
m = a c = b c, and let n = a c = b c. Then ac = gcd(a, c)lcm(a, c) = mn and similarly bc = mn. So
a = b.

4 Some Relation to Group Theory

Many properties of groups are in fact determined by their subgroup lattices. Here we shall prove the Ores
Theorem as an example.
Definition 4.1. A group G is locally cyclic if all its finitely generated subgroups are cyclic.

10
Example 4.2. The group of rational numbers under addition is locally cyclic but not cyclic.
Theorem 4.3 (Ore). A group G is locally cyclic iff its subgroup lattice is distributive.
Proof. Suppose the subgroup lattice is distributive. It is enough to show that for any a, b G, the subgroup
ha, bi is cyclic.
Now we know haihbi must be cyclic. Let c be a generator. Then hci is in the center of ha, bi, and therefore
a normal subgroup of ha, bi. So hc, abi = hcihabi. We also have hc, abi = hci habi = (hai hbi) habi =
(hai habi) (hbi habi) = ha, abi hb, abi = ha, bi ha, bi = ha, bi. So ha, bi = hcihabi.
Now by second isomorphism theorem, we have ha, bi/hci = (hcihabi)/hci = habi/(hci habi), which is
cyclic. So we conclude that ha, bi/(hai hbi) is cyclic for any a, b G.
Now since ha, bi is abelian and generated by two generators, by the structural theorem of finitely generated
abelian groups we have ha, bi = H1 H2 for some cyclic group H1 , H2 . Let k1 , k2 ha, bi be the generators
for H1 , H2 respectively. Then ha, bi = hk1 , k2 i. Now we know hk1 , k2 i/(hk1 i hk2 i) is cyclic, but hk1 i hk2 i
is the trivial subgroup by our definition of k1 , k2 . So hk1 , k2 i is cyclic. So ha, bi is cyclic.
Suppose now that G is locally cyclic. Then in particular G must be abelian, and its subgroup lattice will
coincide with its normal subgroup lattice, which must be modular.
For any subgroup A, B, C of G. We know C(A B) AC, BC, and therefore we have C(A B)
AC BC. We only need to prove the other direction of the inclusion. Now for any element h AC BC, then
g = ac = bc0 for some a A, b B, c, c0 C. Because G is locally cyclic, ha, b, c, c0 i must be cyclic, and we can
1
find g G generating ha, b, c, c0 i. Then hgi (A hgi)(C hgi) haihc, c0 i = ha, c, c0 i = hgi since b = acc0 .
So the inclusion above shall all be equalities. So hgi = (Ahgi)(C hgi), and similarly hgi = (Bhgi)(C hgi).
Let x, y, z be the indices of A hgi, B hgi, C hgi in hgi, i.e. A hgi = hg x i, B hgi = hg y i, C hgi = hg z i.
Then the equality hgi = (A hgi)(C hgi) = (B hgi)(C hgi) implies that gcd(x, z) = gcd(y, z) = 1. Then
gcd(xy, z) = 1. So hgi = hg xy ihg z i = (A B hgi)(C hgi) C(A B). So ac = bc0 hgi C(A B). So
C(A B) AC BC.

5 Some relation to topology

The final two sections will dig a little into the field of pointless topology (or pointfree topology). The ideal
of pointless topology is that topology is more about the structure of open sets than the actual points of the
topological spaces. So to study topology, we can restrict our attention to the study of special kinds of lattices
(those corresponding to lattice of open subsets of topological spaces). To justify this, we need to show that
for many topological spaces, am isomorphism of the lattice of open subset will induce a homeomorphism of
the topological spaces. In this section we show it for sober spaces.In the next two section, we will go further
to show that the category of coherent spaces is dual to the category of bounded distributive lattices.
Definition 5.1. A topological space X is irreducible if Y Z = X for closed subsets Y, Z implies that
X = Y or X = Z. A subset of X is irreducible if it is irreducible with the subspace topology.
Definition 5.2. A point p of a topological space X is a generic point if its closure is the whole space. It
is a generic point of a subspace iff its closure is the whole subspace.
Definition 5.3. A topological space X is sober if every irreducible closed subset has a unique generic point.
Definition 5.4. A homomorphism is upper complete if it preserves arbitrary supremum.
Proposition 5.5. An isomorphism of lattices are upper complete.
Proof. This is trivial.
Recall that 2 is the lattice with two elements {0, 1} with 0 < 1.
Proposition 5.6. For a sober space X, let T (X) be the lattice of open subset of X. There is a bijection
between points in X and upper complete bounded homomorphism from T (X) to 2.

11
Proof. We already know that points in X has a one-to-one correspondence with continuous maps f : {p}
X, where {p} is a singleton with the obvious topology. Then T ({pt}) = 2. For each x X, we can define
continuous map fx : {pt} X, then it induces an upper complete bounded homomorphism (fx )1 : T (X)
T ({p}) = 2. Note that (fx )1 (U ) = 0 = {} iff x / U . Conversely, for each upper complete bounded
homomorphism h : T (X) 2. Let U = sup h1 (0). Note that h(U ) = h(sup h1 (0)) = sup h(h1 (0)) = 0.
So h1 (0) = {V T (X) : V U }. I claim that U c is irreducible, and therefore has a unique generic point.
Suppose U1 , U2 are two open sets such that (U1 )c (U2 )c U c . Then U1 U2 U . So h(U1 ) h(U2 ) =
h(U1 U2 ) = 0. So h(U1 ) = 0 or h(U2 ) = 0. So U1 U or U2 U . So (U1 )c U c or (U2 )c U c . So U c
/ V . So x (fx )1 and
is irreducible. Let xh be the unique generic point of U c . Then hx (V ) = 0 iff x
h xh give the desired bijection.
Theorem 5.7. Two sober spaces with isomorphic lattices of open subsets are homeomorphic.
Proof. Let X, Y be two sober spaces with isomorphic lattices of open subsets. Let h : T (Y ) T (X) be
the isomorphism. For each x X, then we have a upper complete bounded homomorphism hx = (fx )1 :
T (X) 2, where fx : {pt} X : pt 7 x. Then hx h is an upper complete bounded homomorphism.
So we can find a unique y Y such that hy = (fy )1 = hx h, where fy : {pt} Y : pt 7 y. We define
f : X Y such that f (x) = y. Then for each open set U Y , then f 1 (U ) = {x X : f (x) U } = {x
X : hf (x) (U ) = (ff (x) )1 (U ) = 1} = {x X : hx h(U ) = 1} = {x X : x h(U )} = h(U ) is open. So f is
continuous. Conversely we can build f 1 from h1 , and it will also be continuous for the same reason, and
it is the inverse of f by construction. So X, Y are homeomorphic.

6 Birkhoff s Representation Theorem

We shall perform a deeper study of distributive lattices. We know that if a collection of sets form a lattice,
with join and meed defined as union and intersection, then this lattice is necessarity distributive. These two
section will show that all bounded lattices arise this way. This section will prove it for the finite case, and
the next section will prove it for the general case.
In this section we shall show the Birkhoffs representation theorem, or sometimes called the fundamental
theorem of finite distributive lattices. Note that a finite lattice is always bounded. More precisely, we
shall show that the category of finite distributive lattice F Dist is the dual of the category of finite posets
F P os. And the contravariant functor I : F P os F Dist will map posets into distributive lattices whose
elements are sets. Then by definition of duality of categories, it is clear that every finite distributive lattice
is isomorphic to some lattice in the image of I.
First we shall clarify our definition of F Dist and F P os.
Definition 6.1. An upper bound of a lattice L, usually denoted as 1, is an element that is larger than
or equal to all other elements of L. An lower bound of a lattice L, usually denoted as 0, is an element
that is smaller than or equal to all other elements of L. A lattice is bounded if both 0 and 1 exist. A
homomorphism between bounded lattices, f : L L0 , is a bounded homomorphism if f (0) = 0 and
f (1) = 1.
Proposition 6.2. All finite lattices are bounded.
Proof. Since L is finite, by definition of lattices, sup L and inf L exist. Then again by definition we have
0 = inf L, 1 = sup L.
Definition 6.3. The category of finite distributive lattice F Dist is a category with finite distributive
lattices as objects, and bounded homomorphisms as morphisms. The category of finite posets F P os is a
category with finite posets as objects, and order-preserving functions as morphisms.
Now we can proceed to construct our functors. We first construct J : F Dist F P os.
Definition 6.4. An element of a lattice a L is join irreducible if a 6= 0, and whenever a = b c for
some b, c, then a = b or a = c.

12
Definition 6.5. For each lattice L, we define J (L) to be the set of all join irreducible elements, with partial
order inherited from L, i.e. a b in J (L) iff a b in L.
Proposition 6.6. J : F Dist F P os is a contravariant functor.
Proof. For any L F Dist, since L is finite, we have J (L) a finite poset. For any bounded homomorphism
f : L L0 . Then we define J (f ) : J (L0 ) J (L) such that J (f )(b) = inf f 1 (b). Note that f 1 (b) is a
subset of L and is therefore finite, so its infimum exists. I claim that this is a well-defined order-preserving
function.
For any b J (L0 ), we only need to show that inf f 1 (b) is non-zero and join irreducible. We know
f (J (f )(b)) = f (inf f 1 (b)) = inf f (f 1 (b)) = inf{b} = b 6= 0, and f (0) = 0. Therefore J (f )(b) 6= 0.
Suppose aa0 = J (f )(b). Then f (a)f (a0 ) = f (aa0 ) = f (J (f )(b)) = b. Since b is join irreducible, we have
f (a) = b or f (a0 ) = b. Then a or a0 is in f 1 (b). Then a inf f 1 (b) = J (f )(b) or a0 inf f 1 (b) = J (f )(b).
But since a, a0 aa0 = J (f )(b), we have either a = J (f )(b) or a0 = J (f )(b). So J (f )(b) is join irreducible.
So J (f ) is indeed well-defined. Suppose a b in J (L0 ). Then J (f )(a) = inf f 1 (a) inf f 1 (b) = J (f )(b).
So J (f ) is order-preserving.
For any L F Dist and b J (L), we have J (idL )(b) = inf idL 1 (b) = inf{b} = b, and thus J (idL ) =
idJ (L) . For any bounded homomorphism f : L L0 , g : L0 L00 , and for any c J (L00 ), then J (g
f )(c) = inf(g f )1 (c) = inf f 1 g 1 (c) inf f 1 (inf g 1 (c)) = J (f ) J (g)(c). On the other hand, if
b f 1 (inf g 1 (c)), then g f (b) = g(inf g 1 (c)) = g J (g)(c) = c. So f 1 (inf g 1 (c)) f 1 g 1 (c). So
J (f ) J (g)(c) = inf f 1 (inf g 1 (c)) inf f 1 g 1 (c) = J (g f )(c). So both direction of the inequality is
proven, and we conclude that J (g f ) = J (f ) J (g). So J is indeed a contravariant functor.
Now we constructe the other functor I : F P os F Dist.

Definition 6.7. We shall use 2 to denote the poset with two element {0, 1} and with order relation 0 < 1.
Definition 6.8. For a finite poset P , an initial segment is a subset S of P such that there is an order-
preserving function f : P 2 such that S = f 1 (0).
Definition 6.9. For each poset P , we define I(P ) to be the set of all initial segments.

Proposition 6.10. A subset S P is an initial segment iff for each a b, b S implies a S.

Proof. Suppose S is an initial segment. Let f : P 2 be the corresponding order-preserving function for
S. Suppose a b and b S. Then f (a) f (b) = 0. So f (a) = 0 and thus a S.
Suppose for each a b, b S implies a S. Define f : P 2 such that f (a) = 0 if a S and f (a) = 1
if a
/ S. Suppose f is not order-preserving. Then we can find a b such that f (a) is not smaller than
or equal to f (b). Since in 2 every pair of elements has order relation, we must have f (a) larger than f (b).
Then f (a) = 1 and f (b) = 0. But then b S and a b implies a S, so f (a) = 0, contradiction. So f is
order-preserving.
Proposition 6.11. I(P ) is a distributive lattice, with intersection and union as its meet and join.
Proof. It is enough to show that the union and intersection of two initial segments are initial segments. Let
S, S 0 be two initial segments. Suppose a b and b S S 0 . If b S, then we have a S S S 0 , and if
b S 0 , we have a S 0 S S 0 . So S S 0 is an initial segment. Similarly, if a b and b S S 0 . Then
b S and b S 0 . Then a S and a S 0 . So a S S 0 . So S S 0 is an initial segment.
Proposition 6.12. I : F P os F Dist is a contravariant functor.

Proof. For each finite posets P , the it has only finitely many subsets. So it must have finitely many initial
segments. So I(P ) is a finite distributive lattice. Now let g : P P 0 be an order-preserving function. For
each S I(P 0 ), define I(g)(S) = g 1 (S). I claim that I(g) is a well-defined bounded homomorphism from
I(P 0 ) to I(P ).

13
 Let S I(P 0 ) be an initial segment, and let f : P 0 2 be the corresponding order-preserving function.
Then g 1 (S) = g 1 f 1 (0) = (f g)1 (0), where we know f g : P 2 is an order-preserving function.
So as a result g 1 (S) is the initial segment corresponding to the function f g. We also have g 1 (S S 0 ) =
g 1 (S) g 1 (S 0 ) and g 1 (S S 0 ) = g 1 (S) g 1 (S 0 ). So I(g) is a homomorphism. Finally, the upper
bound and lower bound of I(P 0 ) are P 0 and , and g 1 (P 0 ) = P, g 1 () = . So I(g) is a bounded
homomoprhism.
For any P F P os, then I(idP )(S) = idP 1 (S) = S for all S I(P ). So I(idP ) = idI(P ) . For any
order-preserving functions f : P P 0 , g : P 0 P 00 . Then I(g f ) = (g f )1 = f 1 g 1 = I(f ) I(g).
So I is a contravariant functor.
Now we shall show that these functors induces duality between these two categories. We first establish
a lemma about finite distributive lattices.

Lemma 6.13. In a finite distributive lattice L, for any a L, a = sup{b J (L) : b a}. For any initial
segment S of J (L), then S = {b J (L) : b sup S}.
Proof. If a = 0, then sup{b J (L) : b a} = sup = 0 = a. If a is join irreducible, then the statement is
trivial as well. Suppose now a 6= 0 and not join irreducible, but the statement is true for all a0 < a. Then
we can find x, y L such that x, y 6= a and x y = a. Then x, y < a. Then a = x y = sup{b J (L) : b
x} sup{b J (L) : b y} sup{b J (L) : b x y} = sup{b J (L) : b a} a. So all inequalities
above are equalities. Since L is a finite lattice, we are done by induction.
Now for any c sup S and c J (L), We have c = c sup S = c sup{b J (L) : b S} = sup{b c : b
S}, where the last equality is the distribution law. Then since c is join irreducible, we have c = b c for some
b S. Then c b. Since S is an initial segment, we must have c S. So S = {b J (L) : b sup S}.

Definition 6.14. For each finite distributive lattice L, we define (L) : I J (L) L such that for any
S I J (L), (L)(S) = sup S.
Proposition 6.15. : I J idF Dist is a natural isomorphism of functors.
Proof. First let us show that (L) is an isomorphism for each L F Dist. Clearly (L) is order preserving.
Let 0 (L) : L I J (L) such that for any a L, ; (L)(a) = {b J (L) : b a}. Then clearly 0 (L), (L) are
both order-preserving, and (L) 0 (L) = idL , 0 (L) (L) = idIJ (L) by the Lemma above. So 0 (L), (L)
are isomorphism of posets and thus isomorphism of lattices.
Now let us show that is a natural transformation. For any bounded homomorphism f : L L0 , and
for any initial segment S I J (L), we have (L0 ) (I J (f ))(S) = (L0 )((J (f ))1 (S)) = (L0 )({b
J (L0 ) : J (f )(b) S}) = sup{b J (L0 ) : J (f )(b) S} = sup{b J (L0 ) : inf f 1 (b) S}. On the other
hand (idF Dist (f )) (L)(S) = f (sup S) = sup f (S). I claim that f (S) = {b J (L0 ) : inf f 1 (b) S}. Then
would indeed be a natural transformation.
If inf f 1 (b) S, then b = f (inf f 1 (b)) f (S). So {b J (L0 ) : inf f 1 (b) S} f (S). Conversely, if
b f (S), then f 1 (b)S 6= . Say c f 1 (b)S. Now we know inf f 1 (b) = J (f )(b) J (L) is irreducible.
Then since inf f 1 (b) c and S is an initial segment, we have inf f 1 (b) S. So we are done.
Now we prove a lemma for finite posets

Lemma 6.16. For a finite poset P , if S is a join irreducible element of I(P ), then sup S exists and
S = {b P : b sup S}. Conversely, for any a P , then Sa = {b S : b a} is a join irreducible element
for I(P )
S S
Proof. We know S = bS {b} = bS {c P : c b} by definition of an initial segments. If S is join
irreducible, we must conclude that S = {c P : c b} for some b S. Then clearly b = sup S.
Conversely, suppose S S 0 = Sa for initial segments S, S 0 . Then a S S 0 , and thus either a S or
a S 0 . Then be definition of initial segments, Sa S or Sa S 0 . Since we also have S, S 0 S S 0 = Sa ,
we must conclude that S = Sa or S 0 = Sa . So Sa is join irreducible.

14
Definition 6.17. For each finite poset P , we define (P ) : J I(P ) P such that (P )(S) = sup S.
Proposition 6.18. : J I idF P os is a natural isomorphism of functors.
Proof. We start by showing that (P ) is an isomorphism of posets for each P F P os. Clearly is
order-preserving. Let 0 (P ) : P J I(P ) such that 0 (P )(a) = {b P : b a}, so 0 (P ) is clearly
order-preserving as well. Then by the lemma above we have (P ) 0 (P ) = idP and 0 (P ) (P ) = idJ I(P ) .
So (P ) is an isomorphism of posets for each P F P os.
Now let us show that is a natural transformation. Consider any order-oreserving function f : P P 0 ,
and any S J I(P ). Let a = sup S, and then S = {b P : b a}. Then J I(f )(S) = inf I(f )1 (S) =
inf{S 0 I(P 0 ) : I(f )(S 0 ) = S} = inf{S 0 I(P 0 ) : f 1 (S 0 ) = S}. Let Sf (a) = {b P 0 : b f (a)}.
Then clearly f 1 (Sf (a) ) = S. On the other hand, for any S 0 I(P 0 ) such that f 1 (S 0 ) = S, then we have
f (a) S 0 . So Sf (a) S 0 . So J I(f )(S) = Sf (a) . So (P 0 ) (J I(f ))(S) = (P 0 )(Sf (a) ) = sup Sf (a) =
f (a) = f (sup S) = idF P os (f ) (P )(S). So is a natural transformation.
The above propositions proves our major theorem of the section.
Theorem 6.19. The categories F Dist and F P os are dual to each other through functors I, J .

7 Stone Duality for Bounded Distributive Lattice

Now the above approach fails for infinite bounded distributive lattice. Say, for the following example.
Example 7.1. Let N be the lattice of natural numbers, with a partial order such that a b iff b is a multiple
of a. This is an infinit distributive lattice with upper bound 1 and lower bound 0. However, the set J (N)
is empty. For each natural number a 6= 0, we can find p, q distinct prime numbers coprime to a. Then
a = gcd(pa, qa) = pa qa. So no natural number is join irreducible. So I J (N) = I() = {} is the trivial
lattice with one element, which is clearly not isomorphic to N.
For the infinite case, we shall show that the category of bounded distributive lattice is dual to the category
of coherent spaces, a special kind of topological spaces. This shall imply that every bounded distributive
lattice is isomorphic to the lattice of compact open subsets of some coherent space. This is a generalization
of the last theorem, since finite coherent spaces is exactly the same as finite posets with topology induced by
the partial order (treat initial segments as closed subsets). We start by building the definition of coherent
spaces.
Definition 7.2. Let X be a topological space and let K(X) be the set of all compact open subsets of X.
Then X is coherent if X is sober and compact, and K(X) is closed under finite intersection, and K(X) is
a basis for the topology.A coherent map between coherent spaces is a continuous function such that the
preimage of every compact open subsets are compact open. The category of coherent space Coh is a
category with coherent spaces as objects and coherent maps as morphisms.
Definition 7.3. The category of bounded distributive lattices Dist is a category with bounded dis-
tributive lattices as objects and bounded homomoprhism as morphisms.
Proposition 7.4. K : Coh Dist is a contravariant functor.
Proof. For each coherent space, K(X) is clearly partially ordered by inclusion, and it is closed under finite
intersection. It is also closed under finite union since union of compact sets are compact and union of open
sets are open. Finally K(X) has upper bound X and lower bound . So K(X) is a bounded distributive
lattice.
Let f : X Y be any coherent map. Then we define K(f ) : K(Y ) K(X) such that K(f )(S) = f 1 (S)
for each S K(Y ). This is well-defined by the definition of coherent map. We also know that f 1 (Y ) = X
and f 1 () = , and that f 1 (S S 0 ) = f 1 (S) f 1 (S 0 ), f 1 (S S 0 ) = f 1 (S) f 1 (S 0 ). So K(f ) is a
bounded homomorphism. We also trivially have K(idX ) = idK(X) , and K(g f ) = (g f )1 = f 1 g 1 =
K(f ) K(g). So K is a contravariant functor.

15
 Now we go for the other direction, from bounded distributive lattices to coherent spaces. We shall first
define ideals and prime ideals of a bounded distributive lattice. Recall that 2 is the poset {0, 1} with 0 < 1.
This is clearly a bounded distributive lattice.
Definition 7.5. A subset of a bounded distributive lattice L is an ideal if it is the preimage of the lower
bound for any bounded homomorphism from L to a bounded distributive lattice L0 . A prime ideal is the
preimage of the lower bound of any bounded homomorphism from L to 2.
Proposition 7.6. A subset of a bounded distributive lattice is an ideal iff it is nonempty, proper, and its a
sublattice and an initial segment.
Proof. Suppose a subset I is an ideal for the bounded homomorphism f : L L0 . Then since f (0) = 0, we
have 0 f 1 (0) = I, and thus I 6= . Since f (1) = 1, we have 1 / I, and thus I is proper. If a, b I, then
f (a b) = f (a) f (b) = 0 and f (a b) = f (a) f (b) = 0, so a b, a b I and I is a sublattice. Finally if
a b and b I, then f (a) f (b) = 0, and thus f (a) = 0 and a I. So I is an initial segment.
Conversely, suppose I is a nonempty sublattice which is also an initial segment. Define L0 = L/ where
we claim that b b0 iff a b = a0 b0 for some a, a0 I. This relation is reflexive since b 0 = b 0 for
any b L and 0 I. This relation is symmetric by its definition. Finally if b b0 and b0 b00 , then we
can find a, a0 , c, c0 I such that a b = a0 b0 , c b00 = c0 b0 . Then a0 c b00 = a0 c0 b0 = c0 a b,
and a c0 , a0 c I since I is a sublattice. So is transitive. Therefore is an equivalent relation and L0
is a well-defined set. We shall use [b] to denote the equivalence class in L0 for b L, and let f : L L0 be
the map b 7 [b]. I claim that L0 is a bounded distributive lattice, and that f is a bounded homomorphism.
Suppose a a0 , b b0 . Then we can find c, c0 , d, d0 I such that a c = a0 c0 , b d = b0 d0 . Then
a b c d = a0 b c0 d = a0 b0 c0 d0 . So a b a0 b0 . So we can define [a] [b] = [a b], and this
will be a well defined binary operation on L0 . Similarly we have (a b) (c d) = (a c d) (b c d) =
(a0 c0 d) (b0 c d0 ) = (a0 b0 ) e for some e c d c0 d0 I. Since I is a lower segment, we
have e I. So a0 b0 a b. So we can define [a] [b] = [a b], and this will be a well defiend binary
operation on L0 . It follows easily from the definition that , on L0 is commutative and associative, and
the absorption and distribution law holds. So L0 is a distributive lattice. Finally, [0], [1] are clearly lower
and upper bounds of L0 . So L0 is a bounded distributive lattice. And since f (b) = [b], it is clear that f is
a bounded homomorphism. Now for each a I, we have a a = 0 a, and thus [a] = [0]. Conversely, if
[a] = [0], we have a c = 0 c0 = c0 for some c, c0 I. Then a c0 , and thus a I. So we conclude that
I = f 1 (0).
Proposition 7.7. An ideal is a prime ideal iff a b I implies that a I or b I.
Proof. Suppose I is a prime ideal for the bounded homomorphism f : L 2. Then if a b I, then
f (a) f (b) = f (a b) = 0. Clearly we cannot have f (a) = f (b) = 1. So either f (a) = 0 or f (b) = 0. So
a I or b I.
Suppose that I is an ideal such that a b I implies that a I or b I. Define function f : L 2
such that f (a) = 0 if a I and f (a) = 1 if a / I. Clearly f 1 (0) = I, so we only need to show that
f is a bounded homomorphism. If a, b I, we have f (a b) = f (a b) = 0 since I is a sublattice. If
a I, b / I, then a b / I, a b I since I is an initial segment. So f (a b) = 1 = 0 1 = f (a) f (b),
and f (a b) = 0 = 0 1 = f (a) f (b). If a, b / I, then clearly a b
/ I, and a b
/ I because otherwise
we would have a I or b I. So f (a b) = 1 = f (a) f (b) and f (a b) = 1 = f (a) f (b). So f is a
homomorphism. Finally we have f (0) = 0, f (1) = 1 since I is proper and nonempty.
Definition 7.8. The spectrum of a bounded distributive lattice L is the set of all prime ideals, denoted
by Spec(I). It is a topological space with its topology generated by the collection of subsets {Ua = {I
Spec(L) : a
/ I}}aL .
Proposition 7.9. Ua Ub = Uab and Ua Ub = Uab .

16
Proof. If a / I and b / I for a prime ideal I, then clearly a b / I since I is prime. If a I or b I, then
since a b a, b and I is an initial segment, we have a b I. So Ua Ub = Uab .
If a
/ I or b
/ I, then since a b a, b, we must have a b
/ I. If a I and b I, then clearly a b I.
So Ua Ub = Uab .
The following is an equivalent variation of the famous Boolean Prime Ideal Theorem, the proof of which
requires axiom of choice. It is unprovable without axiom of choice. In fact, if treated as an axiom, it is
strictly weaker than axiom of choice.
Theorem 7.10 (Prime Ideal Theorem for bounded distributive lattices). Let I be an ideal of a bounded
distributive lattice L, and suppose a
/ I. Then there is a prime ideal P such that I P and a
/ P.

Proof. Let C(I) = {I 0 L : I 0 is an ideal containing I, and a / I 0 }, partially ordered

S by inclusion. Let
{Ii }iN be a chain of ideals in C(I), i.e. I0 I1 I2 .... Then consider I 0 = iN Ii . It is clearly
nonempty and containing I. Since a / Ii for each Ii , we have a / I 0 , and in particular I 0 is proper. If
a b I , then b Ii for some i, and thus a Ii I . So I is an initial segment. Finally, if a, b I 0 , we
0 0 0

have a Ii , b Ij for some i, j. Find k > i, j, then we have Ii , Ij Ik . So a, b Ik and thus a b, a b Ik .

So a b, a b I 0 . So I 0 C(I). So every chain in C(I) has an upper bound. By Zorns lemma (Axiom
of choice), C(I) must have a maximum element. Say J is a maximum element. Suppose x y J and
x, y / J. Let Jx be the intersection of all ideals containing J and x, and Jy be the intersection of all ideals
containing J and y. Then Jx , Jy are ideals strictly contain J. By maximality of J in C(I), we would have
a Jx , Jy . Then I claim that a x x0 , a y y 0 for some x0 , y 0 J by Lemma ??. If this is true, then
a (x x0 ) (y y 0 ) = (x y) (x0 y) (x y 0 ) (x0 y 0 ) (x y) x0 y 0 J. So a J, contradiction.
So x J or y J.
To see our claim, consider Jx0 = {b L : b x x0 for some x0 J}. It is clearly nonempty and proper,
and it is an initial segment by definition. If b, b0 Jx0 , then b x x0 and b0 x x00 for some x0 , x00 J.
Then b b0 b b0 x x0 x00 where x0 x00 J. So b b0 , b b0 Jx0 . So Jx0 is an ideal containing J
and x. Then by definition of Jx we have Jx Jx0 . On the other hand, clearly x x0 J 0 for any x0 J all
ideals J 0 containing both J and x. So x x0 Jx for all x0 J, and thus Jx0 Jx . So we conclude that
Jx = Jx0 = {b L : b x x0 for some x0 J}. And since a Jx , we can find x0 J such that a x x0 .
We can show a y y 0 for some y 0 J in exactly the same way.
Proposition 7.11. Spec(L) is a coherent space for each bounded distributive lattice L.
Proof. I claim that K(Spec(L)) = {Ua }aL . If this is true, then we also know that it is closed under finite
intersection since Ua Ub = Uab . Since no prime ideal contains the upper bound 1, we also know that
Spec(L) is compact since Spec(L) = U1 .
First of all, Ua is clearly open. For any open cover of Ua for some a L, since {Ub }bL is a base for
topology, weTcan assume that this open cover is {Ub }bS for a subset S of L. Note that (Ub )c = {I Spec(L) :
b I}. So bS (Ub )c = {I Spec(L) : S I}. Let IS be the intersection of all ideals containing S. It is
easy to check that IS = {b L : b s1 s2 ... sn for some positive integer n and some s1 , s2 , ..., sn S}.
If a IS , then a s1 s2 ... sn for some
Sn n and some s1 , ...,Ssn S. Then (Ua )c = {I Spec(L) :
c n
a I} {I Spec(L) : s1 , ..., sn I} = i=1 (Usi ) . So Ua i=1 Usi , covered by a finite subcover.
On the other hand, if a / IS . By the prime ideal theorem above, we can find Ia a prime ideal containing
IS but not a. Then Ia Ua but Ia / {Ub }bS , contradiction. So Ua is compact. So, we have shown that
K(Spec(L)) {Ua }aL
Conversely,
S let U be a compact open subset. Then since {Ua }aL is a base for topology, Snwe have
U = aS Ua for a subset S L. Then by conpactness we can find a1 , ..., an S such that U = i=1 Uai =
Ua1 a2 ...an . So K(Spec(L)) {Ua }aL .
It S
remains to show T that Spec(L) is sober. Let C be a closed irreducible subset of Spec(L). Then
C = ( aS Ua )c = aS (Ua )c = {I Spec(L) : S I} = {I Spec(L) : IS I} for a subset S L, and
IS is the intersection of all ideals containing S. Suppose ab IS for some a, b L. Then any ideal containing
IS will contain ab. So C (Uab )c = (Ua )c (Ub )c . Since C is irreducible, we have C (Ua )c or C (Ub )c .

17
 c
T
Then a IS or b IS . So IS is a prime ideal. Then (IS ) = (Ua )c IS (Ua ) = {I Spec(L) : IS I} = C.
So IS is a generic point for C.
Proposition 7.12. Spec : Dist Coh is a contravariant functor.
Proof. Let f : L L0 be any bounded homomorphism. Then define Spec(f )(I) = f 1 (I) for any prime
ideal I L0 . Let us first show that Spec(f )(I) is a prime ideal of L. Since f (0) = 0, f (1) = 1, we clearly
have 0 f 1 (I), 1 / f 1 (I). So f 1 (I) is proper and nonempty. If a b f 1 (I), then f (b) I. Then
f (a) f (b), and thus f (a) I, and thus a f 1 (I). So f 1 (I) is an initial segment. If a, b f 1 (I), then
f (a), f (b) I. Then f (a b) = f (a) f (b) I and f (a b) = f (a) f (b) I. Then a b, a b f 1 (I). So
f 1 (I) is a sublattice and thus an ideal. Finally, if ab f 1 (I), then f (a)f (b) = f (ab) I. So f (a) I
or f (b) I, and thus a f 1 (I) or b f 1 (I). So f 1 (I) is a prime ideal. So Spec(f ) : Spec(L0 ) Spec(L)
is a well-defined function.
For any a L, consider Spec(f )1 (Ua ) = {I Spec(L0 ) : a / f 1 (I)} = {I Spec(L0 ) : f (a) / I} =
Uf (a) . So Spec(f ) is continuous and coherent.
We also have Spec(idL )(I) = (idL )1 (I) = I for any L Dist and any I Spec(L). So Spec(idL ) =
idSpec(L) . Finally, Spec(g f ) = (g f )1 = f 1 g 1 = Spec(f ) Spec(g). So Spec is a contravariant
functor.
Finally we can proceed to build the duality between categories.
Definition 7.13. For each L Dist, we define (L) : L K Spec(L) such that (L)(a) = Ua .
Proposition 7.14. : idDist K Spec is a natural isomorphism.
Proof. For any L Dist, (L) is clearly surjective by definition, and it is a homomorphism since Ua Ub =
Uab , Ua Ub = Uab . We also have U1 = Spec(L) and U0 = . So it remains to show injectivity. If Ua = Ub ,
then Uab = Ua Ub = Ua = Ua Ub = Uab . Let I = {x L : x a b}, which is clearly an ideal. If
ab / I, then by our prime ideal theorem we can find prime ideal I 0 containing I and not a b. Then
I Uab but I 0
0
/ Uab , contradiction. So a b I. Then a b a b, implying that a b = a b, and
thus a = b. So (L) is injective and therefore an isomorphism.
Let f : L L0 be any bounded homomorphism. Then (K Spec(f )) (L)(a) = K Spec(f )(Ua ) =
Spec(f )1 (Ua ) = Uf (a) = (L0 ) idDist (f )(a). So is a natural transformation and thus a natural isomor-
phism.
Proposition
S 7.15. For any prime ideal I of K(X), then we know elements of I are compact open sets of X.
Let UI = U I U . Then (UI )c has a unique generic point xI , and I = {U K(X) : U UI }. Conversely,
for any point x X, let Ux = (x)c . Then Ix : {U K(X) : U Ux } is a prime ideal of K(X).
Proof. We know (UI )c is closed, so we only need to show that it is irreducible. Suppose U1 , U2 are two
open subset of X such that (U1 )c (U2 )c (UI )c . Then U1 U2 UI . Suppose U1 6 UI . Then we can
find a U1 UI , and we can find V1 a compact
S open neighborhood of a. Let V2 be any compact open set
contained in U2 . Then V1 V2 US I = U I U . Then since V1 V2 must also be compact open, we can
n
find a finite subcover, so V1 V2 i=1 Ui I for some U1 , ..., Un I. Then since I is a prime ideal and
V1 / I, we have V2 I. So V2 UI . Note that this isStrue for all compact open V2 U2 . Since compact
open sets form a base for the topology, we have U2 = V2 U,V2 K(X) V2 UI . Then (U2 )c (UI )c . So UI
is irreducible. Thus it has a unique generic point xI . Note that the above argument for V2 I shows that
for all U K(X) contained in K(X), U I. So I = {U K(X) : U UI }.
Conversely, Ix is clearly a proper initial segement of K(X). Its nonempty since Ix . If U1 , U2 Ix ,
then U1 , U2 Ux , and thus U1 U2 , U1 U2 Ux . So U1 U2 , U1 U2 Ix . Finally, if U1 U2 Ix . Then
x / U1 U2 , and thus x / U2 . Then x U1 = or x U2 = . Then U1 Ux or U2 Ux . So
/ U1 or x
U1 Ix or U2 Ix . So Ix is a prime ideal of K(X).
Definition 7.16. For each X Coh, we define (X) : X Spec K(X) such that (X)(x) = Ix for each
x X.

18
Proposition 7.17. : idCoh Spec K is a natural isomorphism.
Proof. By the last proposition, (X) has a well defined inverse 0 (X) : I 7 xI . We only need to show
that both are coherent.
S For any U K(X), then 0 (X)1 (U ) = {I Spec K(X) : xI U } = {I
Spec K(X) : U 6 V I V } = {I Spec K(X) : U / I} K(Spec K(X)) by definition of the topology
on the spectrum. So 0 (X) is coherent. Conversely, for any V K(Spec K(X)), we can find U K(X)
such that V = {I Spec K(X) : U / I}. Then (X)1 (V ) = {x X : Ix V } = {x X : U / Ix } =
{x X : x U } = U . So (X) is coherent.
Let f : X X 0 be any coherent map. Then (SpecK(f ))(X)(x) = (SpecK(f ))(Ix ) = K(f )1 (Ix ) =
{U K(X 0 ) : K(f )(U ) Ix } = {U K(X 0 ) : f 1 (U ) Ix } = {U K(X 0 ) : x / f 1 (U )} = {U K(X 0 ) :
f (x)
/ U } = If (x) = (X)(f (x)) = (X) (idCoh (f ))(x). So is a natural transformation and thus a natural
isomorphism.

So we have proven the following theorem:

Theorem 7.18. The categories Dist, Coh are dual to each other through functors Spec and K.

19