Lecture Notes in Ordered Sets by DR Tero Harju

Lecture Notes in
Ordered Sets
Tero Harju
Department of Mathematics
University of Turku, Finland
[email protected]
2006
Good order is the foundation of all things.
Edmund Burke (1729 1797)
Contents
1. Posets 3
1.1. First notions 3
1.2. Chains 14
1.3. Extensions and dimension 20
1.4. M obius function 26
2. Lattices 32
2.1. Denition 32
2.2. Isomorphism and sublattices 36
2.3. Congruences and ideals 40
2.4. Fixed points 46
3. Special Lattices 50
3.1. Distributive and modular lattices 50
3.2. Algebraic lattices 59
3.3. Complemented lattices 68
3.4. M obius function for lattices 71
Literature 74
Index 75
1
CONTENTS 2
Notation
N = 0, 1, . . . , Z, Q, R and C denote the sets of nonnegative integers, integers,
rational numbers, real numbers and complex numbers.
For positive integers k and n, k[n means that k divides n.
For a nite set X, [X[ denotes its size, that is, the number of elements in X.
2
X
denotes the power set of X, that is, the family of all subsets of X including
the empty set , and X, itself.
The identity relation on a set X is dened by
X
= (x, x) [ x X
and the universal relation on X is dened by
X
= (x, y) [ x, y X .
We often omit the subscript X from these notations.
For a family A 2
X
of subsets,
_
A =
_
AA
A and
A =
AA
A.
A partition of a set X is a family = X
i
[ i I of nonempty subsets of
X such X
i
X
j
= whenever i ,= j and X =
iI
X
i
. The elements X
i
are
blocks of the partition.
For a mapping : X Y , and a subset A X, let (A) = (x) [ x A.
Some early history
Partially ordered sets and lattices can be said to have their origin in the work
of G. Boole when he tried to axiomatize propositional logic. Lattices were in-
troduced at the end of the 19th century by C.S. Peirce and E. Schroder (Die
Algebra der Logik) in the context of Boolean algebras. In 1890s R. Dedekind
considered lattices (under the name dualgruppen) in the context of ideals gener-
ated by algebraic numbers, and subgroups abelian groups. Modularity of lattices
is Dedekinds product and this notion comes from subgroup considerations. Af-
ter Dedekinds pioneering studies nothing much happened in lattice theory until
1930s when G. Birkho and O. Ore revived the theory. Modern lattice theory
owes much to Birkho who systematized their study in the 1930s. He also showed
the connection of lattice theory to universal algebra.
1 Posets
1.1. First notions
Ordered sets
Let X and Y be sets, and R X Y a relation between the elements of X and
Y , i.e., R consists of some pairs (x, y) with x X and y Y . We often write
xRy instead of (x, y) R. We let
R
1
= (y, x) [ (x, y) R
be the inverse relation of R. For two relations R X Y and S Y Z,
their composition is the relation R S X Z dened by
R S = (x, z) [ y X : (x, y) R and (y, z) S .
Let R
0
=
X
, R
1
= R, R
2
= R R, and inductively R
n+1
= R
n
R for n 1.
In the order theory we usually consider relations R X X on the set X.
Such a relation R is called
reexive if (x, x) R for all x X;
symmetric if (x, y) R implies (y, x) R;
antisymmetric if (x, y) R and (y, x) R imply x = y;
transitive if (x, y) R and (y, z) R imply (x, z) R.
The above denitions have the following equivalent formulations:
Lemma 1.1. Let R X X be a relation. Then
(1) R is reexive if and only if
X
R.
(2) R is symmetric if and only if R = R
1
.
(3) R is transitive if and only if R
2
R.
For a relation R, let R
+
be its transitive closure:
R
+
=
_
n=1
R
n
.
Hence (x, y) R
+
if and only if there exists a nite sequence z
0
, z
1
, . . . , z
n
for
some n such that x = z
0
, z
i
Rz
i+1
for i = 0, 1, . . . , n 1, and z
n
= y.
A relation R X X is called
a quasi-order (also known as a preorder) if it is reexive and transitive;
a partial order if it is reexive, transitive and antisymmetric;
a linear order (or a chain or a total order) if it is a partial order and
xRy or yRx for all x, y X ;
an equivalence relation if it is reexive, transitive and symmetric.
3
1.1. FIRST NOTIONS 4
If R is an equivalence relation on the set X, then the equivalence classes
xR = y [ xRy for x X
form a partition of X, that is,
X =
_
xX
xR and xR yR = for x ,= y .
Theorem 1.2. Let R be a quasi-order on a set X, and let be dened by
x y xRy and yRx.
Then is an equivalence relation that satises: if x
1
y
1
and x
2
y
2
, then
x
1
Rx
2
if and only if y
1
Ry
2
.
Proof. Exercise.
Example 1.3. Let : X Y be a function. Then the preimages
1
(y) = x [
(x) = y form a partition of X, and the kernel dened by
ker() = (x, y) [ (x) = (y)
is an equivalence relation on X.
Posets
If R is a partial order on the set X, then P = (X, R) is a partially ordered
set, or a poset for short. Later we write mostly x P and A P instead of
x X and A X, respectively. Also, we denote by
P
the order of the poset P.
Let us write x <
P
y if x
P
y and x ,= y. The relation <
P
is the strict partial
order associated to
P
.
We let
[x, y]
P
= z [ x
P
z and z
P
y
be the interval (or segment) of the elements x, y X. If x
P
y then the
interval is empty.
A poset P is said to be locally nite if every interval [x, y]
P
is nite.
Example 1.4. The poset (N, ) of natural numbers is innite, and we have
0 1 2 . . . . Hence this poset is locally nite.
If
P
is a partial order, then we write x
P
y, or simply x y, if x and
y are comparable elements, that is, if x
P
y or y
P
x. If x and y are not
comparable, i.e., x , y, then they are incomparable. The relation is reexive
and symmetric, but it need not be transitive. To see this, consider the 3-element
poset where x
P
y and x
P
z, but y and z are incomparable.
A poset P is said to be connected if for each pair of elements x, y P there
is a sequence x = z
0
z
1
z
2
y of comparable elements.
Let P be a poset, and let A P be a nonempty subset.
An element x A is minimal in A, if, for all y A with y
P
x, we
have y = x. If A has a unique minimal element then it is called the minimum
element of A. If the full poset P has a minimum element, it is called the
bottom element of P, and it is often denoted by 0
P
.
An element x A is maximal in A, if for all y A with x
P
y, we
have y = x. If A has a unique maximal element then it is called the maximum
element of A. If the full poset P has a maximum element, it is called the
top element of P, and it is often denoted by 1
P
.
We say that an element y P covers x P, denoted by x
P
y, if [x, y] =
x, y, that is, x ,= y, x
P
y and there exists no element z ,= x, y such that
x
P
z
P
y.
Example 1.5. (1) Note that the covering relation may be empty. Indeed, con-
sider the poset (Q, ) of rational numbers with their usual linear ordering. Then
no element covers any other element. Notice that the poset of rational numbers
is not locally nite.
(2) A locally nite poset P is completely determined by its cover relations.
In P if x <
P
y then there exists an element z such that x z and an element z
such that z
y. Of course, it can be that z = z
or that z = y and z
= x.
Hasse diagrams
When P is nite, one can represent it conveniently using the covering relation
(which determines P). In the Hasse diagram of P if x
P
y then we draw a
line connecting x and y so that y is above x in the picture.
Each locally nite poset P can be represented by a 0, 1-matrix M, where
M
xy
= 1 x = y or x
P
y .
Here the powers M, M
2
, . . . tell how many chains there are from each x to each y:
M
k
xy
is the number of ascending chains (with repetitions):
x = x
0

P
x
1

P

P
x
k
= y
where x
i
= x
i+1
or x
i

P
x
i+1
.
In particular, for a nite P, there is a power M
k
, with k [P[, such that
M
k
xy
> 0 x
P
y .
Example 1.6. Consider the 5-element poset on P =
1, 2, 3, 4, 5, where the order is dened by the following
matrix M:
M =
_
_
_
_
_
_
1 1 1 0 0
0 1 0 1 1
0 0 1 0 1
0 0 0 1 0
0 0 0 0 1
_
_
_
_
_
_
The covering relation is represented by the Hasse dia-
gram on the right.
1
2 3
4 5
Example 1.7 (Subsets). Let X be a set, and consider the family 2
X
of its subsets
with respect to inclusion. Then (2
X
, ) is a poset that is called the subset poset
(or the Boolean poset) on X.
For the eight subsets of the nite set X = 1, 2, 3, the Hasse diagram of 2
X
is given in Figure 1.1.
1 2 3
1, 2 1, 3 2, 3
1, 2, 3
Fig. 1.1. The subset poset of 1, 2, 3.
Subposets
Let P be a poset on a set X, and A X a subset of the underlying set X. We
dene the suborder
A
as the restriction of
P
to pairs of A:
x
A
y x
P
y and x, y A.
Then (A,
A
) is a subposet of P.
A subposet C is called a chain in P, if C is a linearly ordered set. A chain C
is maximal if for all x, y C with x <
P
y, [x, y]
P
= x, y, i.e., the interval
does not contain other elements of P.
A subposet A of P consisting of only incomparable elements is called an
antichain. An antichain A of P is maximal if for all x P, x is comparable
to an element in A (including the possibility that x A).
Example 1.8. Let P
i
, i I, be a set of posets on a common set X. Then also
their intersection P =
iI
P
i
is a poset. Indeed, xPy holds if and only if xP
i
y
holds for all i I, and the poset conditions for P follow from the conditions for
each individual poset P
i
.
On the other hand, a union Q =
iI
P
i
need not be a poset, since it can
be that xP
i
y and yP
j
x for some i ,= j and x ,= y, and in Q we would have to
have both xQy and yQx. Also, Q need not be transitive and thus not even a
quasi-order.
For a subset A P, let
A = y P [ x A: x
P
y ,
A = y P [ x A: y
P
x
be the up-set (or the order lter) and the down-set (or the order ideal) of
P generated by A. If A = x is a singleton set, then we write x and x instead
of x and x. These are the principal up-set and the principal down-set
of P generated by the element x.
Bounds
Let A P be a subset of the poset P.
An element x P is an upper bound of A if x A
u
, i.e., if y
P
x for all
y A. Let
A
u
= x P [ y
P
x for all y A
be the set of all upper bounds of A. We say that x A
u
is a least upper
bound (or a supremum) of A, if x
P
y for all upper bounds y A
u
. It
will be denoted by
_
A, if it exists.
An element x P is a lower bound of A if x A
l
, i.e., if x
P
y for all
y A. Let
A
l
= x P [ x
P
y for all y A
be the set of all lower bounds of A. Also, x A
l
is a greatest lower bound
(or an inmum) of A, if y
P
x for all lower bounds y A
l
. It will be
denoted by
_
A, if it exists.
Afterwards, for instance, we let A
ul
denote
_
A
u
_
l
. (It is the set of all lower
bounds of the upper bounds of A.) Note that usually A
ul
,= A
lu
.
The sets A
u
and A
l
can be empty. In general,
A
u
=
xA
x and A
l
=
xA
x.
If A = x is a singleton, then clearly x = x
u
and x = x
l
.
Example 1.9. Recall the poset of Example 1.6. There 2 = 2, 4, 5 = 2, 4,
and 5 = 1, 2, 3, 5. Note that 2, 4
u
= 4 , = 2, 4.
Lemma 1.10. Let A P be a nonempty subset of a poset P. If A does have a
least upper bound, then it is unique. Similarly, if A has a greatest lower bound,
then it is unique.
Proof. Let x be a least upper bound of A, and let y A
u
, then x
P
y. Hence
if also y is a least upper bound then x = y. The proof for the greatest lower
bound is similar.
Examples
Example 1.11 (Subgroups). For a group G, its subgroups form a poset under
the subgroup relation. Notice that the normal subgroups of G need not form a
poset under the normal subgroup-relation , since a normal subgroup HN of
a normal subgroup NG need not be a normal subgroup of G, and so transitivity
condition fails for this relation.
Example 1.12 (Projective geometry). Let V be a vector space over a eld F.
The projective geometry PG(V ) is the poset on the subspaces of V with
respect to the inclusion relation.
Example 1.13 (Divisibility poset). The positive nat-
ural numbers N 0 form an innite poset under the
divisibility relation: n
N
m if and only if n divides m.
Among the nite subposets of this are the divisibility
posets N
n
= (1, 2, . . . , n , [ ). On the right there is
the Hasse diagram of N
10
.
1
2 3
4
5
6
7
8
9 10
Example 1.14 (Divisor poset). On the other hand, the divisor poset T
n
of a
postive integer n consists of the divisors of n, and the relation is the divisibility
relation. Thus T
n
is a subposet of N
n
.
Example 1.15 (Partitions). The set (X) of all partitions of a set X forms a
poset under the partition order:
1

2
, if each block Y of
2
is a union of
blocks of
1
. Equivalently, each block of
1
is contained in some block of
2
.
The poset of partitions goes hand-in-hand with the poset of equivalence
relations. The set Eq(A) of equivalence relations on a set A forms a poset under
inclusion.
Example 1.16 (Words). Let be a set of symbols, called an alphabet. Each
sequence of symbols a
1
a
2
. . . a
n
, for n 1, is a word over . The empty word,
which has no symbols, is denoted by . The set of all words over is denoted by
. A word u is a factor of v, denoted u v if v = v

1
uv
2
, where v
1
and v
2
can
be empty. Then (
, ) is a poset, the factor poset of
.
Example 1.17. For a poset P, consider the set P
of all words over P. The

lexicographic order
P
on P
is dened as follows:
u
P
v v = uw for some w, or
u = wau
and v = wbv
where a <
P
b .
Moreover, we put
P
u for all u P
.
With respect to the lexicographic order P
is a poset. Also, if P is linearly

ordered, so is P
. Note that although P can be nite, P
is always innite, and

it has no maximal elements. The empty word is the bottom element.
Example 1.18 (Subsequences). Also,
is a poset under the subsequence

order:
x = x
1
x
2
. . . x
n
y = y
1
x
1
y
2
x
2
. . . y
n
x
n
y
n+1
_
= x y .
Words can be used in many contexts to describe (or code) sequences. For
instance, let = , + be a binary alphabet. If P is a poset then each
sequence s = (x
0
, . . . , x
n
), where x
i
x
i+1
, can be associated with a word
(s) = (
0
, . . . ,
n1
) where
i
=
_
+ if x
i

P
x
i+1
,
if x
i

P
x
i+1
.
Isotone mappings
Let P and Q be two posets.
A function : P Q is an isotone mapping (also called monotone or
order preserving) if for all x, y P,
x
P
y = (x)
Q
(y) .
An injective isotone mapping is called an order embedding.
Moreover, is an order isomorphism if it is bijective and also its inverse
1
: Q P is isotone. Two posets P and Q are isomorphic, denoted by
P

= Q, if there exists an order isomorphism between them.
Example 1.19. For a poset P, let P
P
be the set of all mappings : P P.
Then P
P
forms a poset under the operation

P
P (x)
P
(x) for all x P .
It is straightforward to check the poset conditions for P
P
. Also, the isotone
mappings P P form a poset with respect to the same ordering.
Lemma 1.20. Two posets P and Q are isomorphic if and only if there exists a
surjective mapping : P Q such that, for all x, y P,
(1.1) x
P
y (x)
Q
(y) .
Proof. Notice rst that if satises (1.1), then it is necessarily a bijection
from P onto Q. Indeed, if (x) = (y) then (x)
Q
(y) gives x
P
y, and
symmetrically we obtain y
P
x , and thus x = y. Hence is injective. It is
surjective by hypothesis.
Now by (1.1), is isotone, and so is
1
, because (1.1) is equivalent to the
condition
1
(u)
P

1
(v) u
Q
v .
Hence is an order isomorphism.
Conversely, assume then that : P Q is an order isomorphism. Then both
and
1
are isotone, and hence (1.1) holds.
Example 1.21. On has to be careful in the above, since an isotone bijection
: P Q need not be an order isomorphism. Indeed, consider the two posets
of Figure 1.2. There : P Q is an isotone bijection, but it is not an order
isomorphism. In fact, P and Q are not isomorphic.
For isomorphic posets the Hasse diagrams look the same except for the names
of the elements.
a
1
b
1
c
1
a
2
b
2
c
2
Fig. 1.2. Two posets with an isotone mapping.
Example 1.22 (Abian). Consider the posets P and Q with Hasse diagrams of
Figure 1.3. The mapping : P Q dened by (x
i
) = y
i
is isotone and bijective.
Also, the mapping : Q P dened by
(y
1
) = x
2
, (y
3
) = x
0
,
(y
i
) = x
i+4
, i 2 even ,
(y
i
) = x
i4
, i 5 odd
is isotone and bijective. However, these posets are not isomorphic.
We shall now prove that all posets can be represented as set systems.
. . .
x
3
x
1
x
2
x
6
x
10
x
0
x
4
x
8
. . .
P
. . .
y
3
y
1
y
2
y
6
y
10
y
0
y
4
y
8
. . .
Q
Fig. 1.3. Nonisomorphic posets with bijective isotone mapping
P Q and Q P.
Theorem 1.23. Each poset P is isomorphic to the poset of its principal down-
sets under set inclusion. To be more precise, let : P 2
P
be dened by
(x) = x.
Then is an order isomorphism from P onto the set of all principal down-sets
of P.
Proof. First, is a bijection to the principal down-sets:
(x) = (y) x = y x
P
y and y
P
x x = y .
To show that is an order isomorphism, observe that if x
P
y, then also
(x) (y). Also, since x x, (x) (y) implies x
P
y. Therefore x
P
y
if and only if (x) (y), and the claim follows.
Direct product
Let P and Q be two posets. Dene their direct product P Q as the cartesian
product (x, y) [ x P, y Q together with the partial order
PQ
:
(x
1
, y
1
)
PQ
(x
2
, y
2
) x
1

P
x
2
and y
1

Q
y
2
.
Theorem 1.24. The direct product of two posets is a poset.
Proof. Exercise.
Example 1.25. The Hasse diagram of the direct product of two chains of length
two is the square as depicted in Figure refg:dirprod.
As usual, the direct products
(1.2)
k
i=1
P
i
= P
1
P
2
P
k
a
1
a
2
b
1
b
2
(a
1
, b
1
)
(a
1
, b
2
) (a
2
, b
1
)
(a
2
, b
2
)
Fig. 1.4. Direct product of two posets.
are dened inductively so that
k
i=1
P
i
=
_
k1
i=1
P
i
_
P
k
. This poset is isomorphic
to P
1

_
k
i=2
P
i
_
, and this allows the notation (1.2).
Example 1.26 (Subset poset). Let X be a nite set of n elements, and consider
the subset poset 2
X
. We show that 2
X
is isomorphic to the n-fold direct product
n
C
2
= C
2
C
2
C
2
of the 2-element chain C
2
= 0, 1 where 0 < 1. Indeed, let X = x
1
, x
2
, . . . , x
n
,
and let : 2
X

n
C
2
be such that the ith component of (A) is 1 just in case
x
i
A. For instance, if n = 5, then (x
2
, x
3
, x
5
) = (0, 1, 1, 0, 1). In
n
C
2
,
(a
1
, a
2
, . . . , a
n
) (b
1
, b
2
, . . . , b
n
) a
i
b
i
for all i ,
and this corresponds to the subset relation in 2
X
.
Dual posets
For a poset P, its dual poset P
d
reverses the order:
x
P
d y y
P
x.
It is immediate that P
d
is a poset, and that (P
d
)
d
= P.
Example 1.27. The Hasse diagram of a nite dual
poset P
d
is obtained from the diagram for P by turning
it over.
a
b
c a
b
c
Many notions and results concerning a poset P have dual statements for the
poset P
d
. For instance, we immediately have the following dual representation
theorem:
Theorem 1.28. Each poset P is isomorphic to the poset of its principal up-sets
under inclusion.
Finite posets
In Figure 1.5 all connected posets are represented with four elements.
Fig. 1.5. Nonisomorphic connected posets of 4 elements. There
are also 6 disconnected posets of four elements.
Table 1 gives the number of all nonisomorphic posets of n elements for n =
1, 2, . . . , 14. The values are from Sequence A000112 in Sloanes integer sequence
web-page: http://www.research.att.com/ njas/sequences/Seis.html .
n posets n posets
1 1 8 16 999
2 2 9 183 231
3 5 10 2 567 284
4 16 11 46 749 427
5 63 12 1 104 891 746
6 318 13 33 823 827 452
7 2 045 14 1 338 193 159 771
Table 1. The number of nonisomorphic posets.
No formula is known for the number p(n) of nite posets, but it is known
(Kleitman and Rothschild 1970) that we have the following asymptotic bound
for the number p(n):
lim
n
log p(n)
n
2
4
log 2
= 1 .
The following is a nite version of the so called Whaleys theorem. We say
that a subset A P of a poset is a super-antichain if for all dierent x, y A,
x y = and x y = , i.e., distinct elements of A do not have a common
upper bound or a common lower bound.
Theorem 1.29 (Freese, Hyndman, Nation). Let P be a nite poset of n elements.
Then P has a subset A such that [A[ n
1/3
|, where A = x or A = x for
some x P or A is a super-antichain of P.
1.2. CHAINS 14
Proof. Let m be the maximum size of a principal down-set or a principal up-set
of P, and assume that m n
1/3
|.
For each element x P, let
N
x
=
_
u
P
x
u
_
v
P
x
v .
Denote a = [ x x[ and b = [ x x[. Then
[N
x
[ =
_
u>
P
x
(u x)
_
v<
P
x
(v x)
+ 1
b(m(a + 1)) +a(m(b + 1)) + 1 = (m1)(a +b) 2ab + 1
= (m1)
2
+ 1 (m1 a)(m1 b) ab (m1)
2
+ 1 .
We now dene a super-antichain A: Let x
1
P be arbitrary, and suppose
x
1
, . . . , x
i
form a super-antichain. Then let x
i+1
/ N
x
1
N
x
i
, if the union is
a proper subset of P. Finally, we obtain an index k such that P =
k
i=1
N
x
i
, from
which it follows that n k((m1)
2
+ 1), and hence m n
1/3
| or k n
1/3
|,
since (m1)
2
+ 1 m
2
and so n km
2
.
Order-preserving functions behave nicely on nite posets.
Theorem 1.30 (Abian). Let P be a nite poset, and : P P an order em-
bedding. Then is an isomorphism.
Proof. Exercise.
1.2. Chains
Zorns lemma
Let X be a family of nonempty sets. A choice function on X is a mapping
: X

X such that for each A X, (A) A. For innite families X, the
existence of a choice function is not self-evident.
Axiom of Choice. For any family of nonempty sets there exists at least one
choice function.
This principle was rst stated and used (in a dierent form) by Zermelo in
1904 in his proof that any set can be well-ordered.
This axiom has many important equivalent formulations; more than 200 of
these are recorded in the book of Rubin and Rubin. For instance, the following
results are equivalent to the Axiom of Choice:
Every vector space has a basis. Proved by Blass in 1984 to be equivalent to
the Axiom of Choice.
Every eld has an algebraic closure (Steinitz 1910).
Every commutative ring with identity has a maximal ideal (Hodges 1979).
1.2. CHAINS 15
Tychonovs Theorem stating that the product of compact spaces is compact
(Kelley 1950).
Krein-Milman Theorem which states that the unit ball B of the dual of a
real normed linear space has a point which is not an interior point of any line
segment in B (Bell and Fremlin 1972).
The following theorem is considered to be an axiom for us. Zorns Lemma
plays an important role in the foundations of mathematics. It is equivalent to
the Axiom of Choice.
Theorem 1.31 (Zorns Lemma). Let P be a poset, where every chain has an
upper bound. Then P contains a maximal element.
Proof. Omitted.
We have the following result using Zorns lemma.
Theorem 1.32 (Kuratowski). Let P be a poset, and C be a chain in P. Then
there exists a maximal chain M of P such that C M.
Proof. Let C be the set of all chains containing C. It is nonempty since C C.
We consider C as a poset with respect to the set inclusion, and we show that C
has a maximal element, which then proves the claim.
Consider an ascending chain B of sets in C, and regard B = B as a subposet
of P. We show that B is a chain in P. To this end, let x, y B, and let
B
x
, B
y
B be such that x B
x
and y B
y
. Then either B
x
B
y
or B
y
B
x
,
since B is a chain. Without loss of generality, we can assume that B
x
B
y
, and
so x, y B
y
, which means that x and y are comparable in P, since B
y
is a chain
in P. Therefore B C.
Clearly, B is an upper bound of B, and hence C has a maximal element by
Zorns lemma.
Well-orders
Later in this section we study partial well-orders. We start with the total notion
of this order.
A poset P is said to be well-ordered if every nonempty subset X P has
a minimum element, i.e., an element x X such that x <
P
y for all y X x.
Lemma 1.33. Every well-ordered poset P is a chain.
Proof. If x, y P are any two dierent elements, then the subset x, y has a
minimum element, and therefore either x <
P
y or y <
P
x, and thus P is linearly
ordered.
The converse of this lemma does not hold, since, for instance, P = (Z, ) is
a chain but it has no minimum element. Note, however, that every nite chain
is well-ordered.
1.2. CHAINS 16
Theorem 1.34. Let P be well-ordered. Then each element x P is either
maximum or there exists a unique y P such that x
P
y.
Proof. If x is not a maximum element of P, the set (x) x has a unique
minimum element y, which is as claimed.
The following theorem states that one can dene an ordering of each set such
that the result is a well-order. For instance, for the integers we can take the
unorthodox order 0
+1
+2
. . . .
Theorem 1.35 (Well-Ordering). Every set X can be given a well-ordering.
Proof. Omitted.
Chain conditions
Let P be a poset on X. Then P satises
the nite chain condition or FCC if every chain in P is nite.
the ascending chain condition or ACC if every strictly increasing chain
x
1
<
P
x
2
<
P
. . . is nite.
the descending chain condition or DCC if every strictly descending chain
x
1
>
P
x
2
>
P
. . . is nite. A poset that satises DCC is well-founded.
the nite antichain condition or FAC if every antichain of P is nite.
Note that a poset P satisfying the nite chain condition can still have unbounded
height, since there can be arbitrarily long chains in P. Obviously the FCC implies
both ascending and descending chain conditions.
Lemma 1.36. A poset P satises the ascending chain condition if and only if
every nonempty subset X P has a maximal element.
Proof. Let A P be nonempty such that A has no maximal elements. Fix
an element x
0
A. Since x
0
is not maximal, there exists an x
1
A such that
x
0
<
P
x
1
, and so forth, x
0
<
P
x
1
<
P
x
2
<
P
. . . , which gives a innite ascending
chain. (Note that the so forth part needs Axiom of Choice: one can always
pick x
i
.)
Assume then that every nonempty subset has a maximal element, and let
x
1

P
x
2

P
. . . be an ascending chain in P. Write X = x
i
[ i = 1, 2, . . . ,
and let x
k
X be a maximal element in X. Hence x
i
= x
k
for all i k, and
therefore X is nite. It follows that P satises the ACC.
A dual argument shows that a poset P satises the descending chain condition
if and only if every nonempty subset of P has a minimal element. In other words,
Lemma 1.37. A poset P is well-founded if and only if every nonempty subset of
P has a minimal element.
1.2. CHAINS 17
Theorem 1.38 (Konig). Let P be a poset satisfying both the nite chain and
antichain conditions (FCC and FAC). Then P is nite.
Proof. The set M of all maximal elements of P consists of incomparable ele-
ments, and thus M is nite by the nite antichain condition. Since every max-
imal chain contains a maximal element, there are at most [M[ maximal chains,
and, by the nite chain condition, there is a constant bounding their lengths.
Moreover, every element of P belongs to some maximal chain, and this proves
the claim.
Example 1.39. Consider the cartesian product ZZ of integers with the partial
order (x
1
, y
1
) (x
2
, y
2
) if and only if x
1
x
2
and y
1
y
2
. This poset is clearly
locally nite. However, it does not satisfy the chain conditions ACC, DCC, nor
FAC. Indeed, the sequence < (2, 0) < (1, 0) < (0, 0) < (1, 0) < . . . gives
an counter-example to both ACC and DCC, and (n, n) [ n = 0, 1, . . . is an
innite set of incomparable elements in the poset.
Graded posets
Let P be a poset. A function r : P N is a rank function of P if it satises
the following conditions
(1) if x is minimal, then r(x) = 0.
(2) if x
P
y then r(y) = r(x) + 1.
Not all posets possess a rank function; for instance, Z with
its natural order does not, because it has an innite de-
scending chain. Also, there are nite posets that do not
have a rank function: Consider the poset N
5
give on the
right. On the other hand, each nite chain C possesses a
unique rank function, and as is rather immediate, the rank
function of C is unique.
N
5
We say that a poset P is graded of rank n if every maximal chain in P has
the same length n, i.e., each maximal chain has n + 1 elements.
Theorem 1.40. Each graded poset of rank n has a unique rank function.
Proof. Let C be a maximal chain in P, say x
0
<
P
x
1
<
P
<
P
x
n
. Then
x
0
is a minimal element of P and x
n
is a maximal element of P. Let r
C
(x
i
) =
i be the rank function of the chain C. Now if C
is another maximal chain,

x
0
<
P
x
1
<
P
<
P
x
n
, then also r
C
(x
i
) = i. If C and C
have a common
element, x
i
= x
j
, then necessarily i = j. Indeed, if j < i, then the chain
x
0
<
P
x
1
<
P
<
P
x
i
= x
j
<
P
. . . x
n
is longer than C, contradicting the
fact that P is graded. Therefore the rank functions of the maximal chains are
compatible with each other. Since P is the union of its maximal chain, the claim
follows.
1.2. CHAINS 18
Higmans Theorem
We consider a special case of Higmans theorem for posets. The general theorem
is stated for quasi-orders in algebras. A well-founded poset P satisfying the nite
antichain condition is called partially well-ordered. Thus a poset P is partially
well-ordered if and only if it satises the descending chain condition and the nite
antichain condition.
Theorem 1.41. The following are equivalent for a poset P.
(1) P is partially well-ordered.
(2) If x
1
, x
2
, . . . is an innite sequence of elements in P, then x
i

P
x
j
for some
i < j.
(3) Every innite sequence of elements of P has an innite ascending subsequence.
(4) P is well-founded and any subset has only nitely many minimal elements.
Proof. Exercises
We show a particular case of the above result.
Lemma 1.42. Let P be a partially well-ordered set, and A an innite sequence
x
1
, x
2
, . . . of elements from P. Then there exists an innite ascending subse-
quence x
i
1

P
x
i
2

P
. . . such that i
1
< i
2
< . . . .
Proof. Let A = x
1
, x
2
, . . . P, and let
M = m [ x
m

P
x
i
for all i > m .
Then M is nite, since P is partially well-ordered, say m k for all m M. For
an index i
1
> k such that i
1
/ M dene recursively i
j+1
such that i
j+1
> i
j
and
x
i
j+1

P
x
i
j
. Then this sequence is an innite ascending sequence as required by
the claim.
Next we extend P to P
. The subsequence order on P
is dened by
u
P
v v = y
1
y
2
. . . y
n
and u = x
i
1
x
i
2
. . . x
i
m
where 1 i
1
< i
2
< < i
m
n, and
x
i
k

P
y
i
k
for all k = 1, 2, . . . , m.
This generalizes the subsequence relation dened in Example 1.18.
Lemma 1.43. The set of sequences P
is a well-founded poset under the relation

subsequence order
P
.
Proof. That
P
is a partial order is an exercise. For well-foundedness, assume
that there exists an innite strictly decreasing chain v
1
>
P
v
2
>
P
. . . in P
.
Now since 0 < [v
i+1
[ [v
i
[ for all i, we can assume that, for all i m, they are
all of the same length n, and so
v
i
= x
i1
x
i2
. . . x
in
.
1.2. CHAINS 19
for each i m, where all x
ij
are in P. Since v
i+1
<
P
v
i
, we have x
i+1,k

P
x
ik
for all k = 1, 2, . . . , n, and there exists an index k
i
n with x
i+1,k
i
<
P
x
ik
i
.
Let k
i
be the rst one with this property. The sequence v
i
is innite, and hence
there exists an index k for which k = k
i
for innitely many i. Thus the sequence
x
ik
, i = 1, 2, . . . , contains a strictly decreasing subsequence. This contradicts the
well-foundedness of
P
.
For a poset P, a bad sequence is an innite sequence x
1
, x
2
, . . . with
x
i

P
x
j
for all i < j. A bad sequence is minimal if there is no bad sequence
x
1
, x
2
, . . . , x
i1
, y
i
, y
i+1
, . . . where y
i
<
P
x
i
for some i. Notice that an innite
subsequence of a bad sequence is bad.
Lemma 1.44. Let P be a well-founded poset that is not partially well-ordered.
Then P has a minimal bad sequence.
Proof. The proof is by induction on the positions in sequences. First of all there
exists at least one bad sequence, since P is not partially well-ordered. Since each
subset of P has a minimal element, there exists an element x
1
P which is
minimal among the rst elements of bad sequences. Inductively, we can choose
x
i
P such that it is a minimal element extending the sequence x
1
, x
2
, . . . , x
i1
to a bad sequence. The resulting sequence is a minimal bad sequence.
Lemma 1.45. Let x
1
, x
2
, . . . be a minimal bad sequence of a poset P. Then
P
is a partial well-order on the subposet
A = x [ x <
P
x
i
for some i .
Proof. Assume contrary to the claim that A is not partially well-ordered and let
y
1
, y
2
, . . . be a bad sequence in A. We have that y
i
<
P
x
j
i
for some x
j
i
from the
bad sequence. Let m be such that the index j
m
is the smallest. The sequence
y
m
, y
m+1
, . . . is bad as a subsequence of a bad sequence.
We show that
(1.3) x
1
, . . . , x
j
m
1
, y
m
, y
m+1
, . . .
is bad. This proves the lemma, since the sequence x
i
was assumed to be minimal,
but now y
m
<
P
x
j
m
.
If (1.3) is not bad then x
r
<
P
y
k
for some r < j
m
and k m. Also, y
k
<
P
x
j
k
,
and j
k
j
m
by the choice of j
m
. Therefore x
r
<
P
y
k
<
P
x
j
k
and r < j
m
. This
contradicts the assumption that the sequence x
i
is bad.
The present proof of Higmans theorem is due to NashWilliams.
Theorem 1.46 (Higman). Let P be a well-founded poset. Then the subsequence
order is a partial well-order on P
.
Proof. The proof is by contradiction. Let v
1
, v
2
, . . . be a minimal bad sequence
given by Lemma 1.44. Write
v
i
= x
i
u
i
,
1.3. EXTENSIONS AND DIMENSION 20
where x
i
P. Now [u
i
[ < [v
i
[ and so u
i
<
P
v
i
. By Lemma 1.45, the set
u
i
[ i = 1, 2, . . . is partially well-ordered. By Lemma 1.41, we can assume that
u
1

P
u
2

P
. . . . Since P is partially well-ordered, there are indices i < j such
that x
i

P
x
j
. Hence v
i

P
v
j
; a contradiction.
Example 1.47 (RobertsonSeymour). One of the most impressive results on
discrete mathematics during the last decades is the RobertsonSeymour Theorem
that solved the famous Wagner Conjecture. For two graphs, dene the minor
relation by G _ H if G can be obtained from H by contracting (i.e., identifying
the ends) or deleting zero or more edges. Then _ is a well quasi-order on the
family of graphs. The present proof of this result requires more than 500 pages.
1.3. Extensions and dimension

Linear extensions
Let P and Q be two posets on a common domain X. Then Q is an extension
of P if x
P
y implies x
Q
y for all x, y X. That is, if the relation
P
is
contained in
Q
. An extension of P that is a linear order is a linear extension
of P.
Lemma 1.48. Let P be a poset, and let x , y be incomparable elements in P.
Then there exists a poset Q extending P such that x
Q
y.
Proof. Let Q = P (x y), i.e., Q is obtained from P by adding all pairs
(u, v) for which u
P
x and y
P
v. Since x
Q
y, we need to show that Q is a
poset.
First of all, x y = , since if z were a common element, y
P
z and
z
P
x would yield y
P
x; a contradiction.
That Q is reexive is clear, since P Q.
For antisymmetry, suppose that u
Q
v and v
Q
u for some elements u ,= v.
By the denition of Q, either u
P
v or u
P
x and y
P
v. Similarly either
v
P
u or v
P
x and y
P
u. There are the following cases to be considered:
Case u
P
v. In this case v
P
u, since otherwise u = v. Hence v
P
x and
y
P
u, However, now y
P
u
P
v
P
x gives a contradiction: y
P
x.
Case v
P
u. This is symmetric to the previous case.
Case u
P
x and y
P
v. Now also v
P
x and y
P
u, which contradicts the
fact that x and y are disjoint.
For transitivity, suppose u
Q
v and v
Q
w. Now either u
P
v or u
P
x
and y
P
v, and either v
P
w or v
P
x and y
P
w. Again there are cases to
be considered.
Case u
P
v. If also v
P
w, then the claim follows from transitivity of P.
Suppose thus that v
P
x and y
P
w. Now also u
P
x, and hence u x and
w y, and so u
Q
w as required.
Case u
P
x and y
P
v. Now if v
P
w, then also y
P
w, and so u x and
w y, and hence u
Q
w. The case v
P
x and y
P
w is not possible, since
x and y are disjoint.
This concludes the proof.
Theorem 1.49 (Szpilrajn). Every nite partially ordered set has a linear exten-
sion.
Proof. The claim follows by applying Lemma 1.48 inductively to incomparable
elements.
Theorem 1.50. Every poset P on a nite set X is the intersection of a set of
linear orders on X.
Proof. Consider the intersection
P =
PR
R
of all linear extensions R of P. By Theorem 1.49, the intersection is nonempty.
It is also clear that

P is a partial order and P

P. Assume that x and y are
incomparable in P. By Lemma 1.48, there is a linear order P
x
for which P P
x
and x
P
x
y, and a linear order P
y
for which P P
y
and y
P
y
x. Therefore the
elements x and y are incomparable in

P. It follows that

P = P.
The above proof is constructive for nite posets, and it does not generalize
to innite partial orders. In the innite case we need Zorns Lemma.
Theorem 1.51 (DushnikMiller). Every poset P on a set X is the intersection
of a set of linear orders on X.
Proof. Consider the poset of all partial orders R that are extensions of P. If
R
1
R
2
. . . is an ascending chain of such posets then the union
i1
R
i
is
also a poset and an extension of P. Hence every chain of such posets has an
upper bound, and by Zorns Lemma, there exists a maximal poset R extending
P. Now R is a linear order by Lemma 1.48.
The dimension of a poset P, denoted by dim(P), is the least number of
linear orders for which P is the intersection. We can have that dim(P) = . If
dim(P) = 1, then P, itself, is a linear order.
Example 1.52 (Crowns). The crown P = Cr
2n
on 2n elements
X = a
1
, . . . , a
n
b
1
, . . . , b
n
.
is the poset, where for all i ,= j, we have a
i

P
b
j
and the other pairs are incomparable. We show that
dim(P) = n.
Indeed, dene a linear order
i
by setting
a
1

i

i
a
i1

i
a
i+1

i

i
a
n

i
b
i

i
a
i

i
b
1

i

i
b
i1

i
b
i+1

i
b
n
.
a
1
a
2
a
3
b
1
b
2
b
3
Let L
i
= (X,
i
) be the poset for
i
. Each of these is a linear extension of the
crown order P, and P is the intersection
n
i=1
L
i
. Hence dim(P) n.
On the other hand, assume that P =
k
i=1
L
i
for some linear extensions L
i
.
For each i, there exists an L
j
such that b
i

j
a
i
in L
j
. (Otherwise, a
i

P
b
i
.)
Suppose there exists an L
j
for which also b
r

j
a
r
for another index r, say i < r.
Now, however, b
i

j
a
i

P
b
r

j
a
r
implies b
i

P
a
r
; a contradiction.
Example 1.53. An alternating cycle of a poset P is a subset S = (x
i
, y
i
) [
i = 1, 2, . . . , k of pairs such that
x
1
, y
1

P
x
2
, y
2

P

P
x
k
, y
k

P
x
1
,
where k 1. As an exercise we state: Let S be a subset of pairs of incomparable
elements. If S contains no alternating cycles, then the transitive closure (P S)
+
is a poset. In fact this is true also in converse.
Theorem 1.54 (Hiraguchi). Let C be a chain in a poset P. Then there are
linear extensions D and U of P such that, for all x , y with x C and y P,
y
D
x and x
U
y .
Proof. Let P
D
= (P S
D
)
+
be the transitive closure of P S
D
, where S
D
=
(y, x) [ x C and x , y. It is clear that S
D
contains no alternating cycles,
and therefore P
D
is a poset by Example 1.53. Let D be a linear extension of P
D
.
Similarly, let P
U
= (P S
U
)
for S
U
= (x, y) [ x C and x , y. Also P
U
is a
poset, and let U be a linear extension of P
U
.
Theorem 1.55. Let P be a poset. If for every nite subposet R of P we have
dim(R) n, then also dim(P) n.
Proof. Omitted.
Dilworths Theorem
In this section we consider nite posets P.
The height of P, denoted by h(P), is the number of elements in a longest
chain in P.
The width of P, denoted by w(P), is the largest number of elements in an
antichain of P.
Clearly, if C is a chain in P and A is an antichain, then [CA[ 1. Therefore
in every set of antichains, whose union is the full poset P, there are at least h(P)
antichains.
Theorem 1.56. Let P be a nite poset. Then P is partitioned into h(P) an-
tichains.
Proof. We prove the claim by induction on h(P). For h(P) = 1 the claim is
obvious, since then P consists of incomparable elements. Assume then that the
claim holds for posets of height less than that of P.
Let M be the set of all maximal elements of P. Then M forms an antichain
in P, and every maximal chain contains an element from M (as an end point).
Therefore the poset P M has height h(P) 1, and by the induction hypothesis,
P M is partitioned into h(P) 1 antichains. Therefore together with M, the
poset P is partitioned into h(P) antichains.
The following result is a dual statement of Theorem 1.56. The present proof
is due to Tverberg.
Theorem 1.57 (Dilworth). Let P be a nite poset. Then there is a partition of
P into exactly w(P) chains.
Proof. Again, if C is a chain in P and A is an antichain, then [C A[ 1.
Therefore in every set of chains partitioning P there are at least w(P) chains.
Thus we need to prove that there exists a set of w(P) chains such that every
element of P is in exactly one of these chains.
We proceed by induction on [P[. The case [P[ = 1 is trivial. Suppose then
that the claim holds for posets of size at most n, and let P be a poset of size n+1.
If w(P) = 1, then the claim obviously holds. Assume thus that w(P) > 1.
Let C be a maximal chain in P. It is clear that
w(P) 1 w(P C) w(P) .
If the subposet P C has width w(P) 1, then, by the induction hypothesis,
P C can be partitioned into w(P) 1 chains, and together with C we have a
partition of P into w(P) chains.
Assume now that w(P C) = w(P), and let A = a
1
, a
2
, . . . , a
w(P)
be an
antichain in P C. Notice that A A = P, since [A[ = w(P) and thus all
elements of P are comparable with an element from A. Since C is a maximal
chain, its maximum element is not in A, for, otherwise the chain C could be
extended by one element of A. By the induction hypothesis, Acan be partitioned
into w(P) chains, C
1
, C
2
, . . . , C
w(P)
, where a
i
C
i
for each i.
For each x A A, there exists an index i such that x
P
a
i
, and hence
a
j

P
x for all j. This means that a
j
is the maximal element of the chain C
j
.
Similarly, the minimal element of C is not in A, which can then be parti-
tioned into w(P) chains C
+
1
, C
+
2
, . . . , C
+
w(P)
, where, moreover, a
j
is the minimal
element of C
+
j
. Combining the chains C
j
and C
+
j
, we obtain a partition of P
into w(P) chains as required.
Theorem 1.58 (Dilworth). For each poset P, dim(P) w(P).
Proof. By Dilworths Theorem 1.57, P can be partitioned into w(P) chains
C
1
, C
2
, . . . , C
w(P)
. Let L
i
(= D) be a linear extension of P provided by Hi-
raguchis Theorem 1.54 for the chain C
i
. Then P =
i
L
i
(and so dim(P)
w(P)). Indeed, if x , y in P, then y
L
i
x, where i is such that x C
i
, and
x
L
j
y, where y C
j
. Hence also x , y in the intersection.
Applications
Example 1.59 (Monotone sequences). Consider the set N = 1, 2, . . . , n
2
+ 1
of integers, and let = (k
1
, k
2
, . . . , k
n
2
+1
) be any permutation of them. We show
that this sequence contains a monotonic subsequence of length n + 1.
Consider the cartesian product P = N N together with the partial order
(x, i)
P
(y, j) if and only if x y and i j. A subsequence
of is
ascending if and only if
is a chain in the poset P. By Theorem 1.57, P

can be partitioned into w(P) chains, and thus can be partitioned into w(P)
ascending subsequences. Thus has an ascending subsequence of length at least
[N[/w(P).
Observe that if k
i
1
, k
i
2
, . . . , k
i
t
is an antichain in P where i
1
< i
2
< < i
t
,
then k
i
1
> k
i
2
> > k
i
t
. Thus if w(P) > n + 1, then has a descending
subsequence of length n + 1. Otherwise, if w(P) n, [N[/w(P) (n
2
+ 1)/n
implies that has an ascending subsequence of length at least n + 1.
Example 1.60 (Sperner). Consider the poset 2
N
of the subsets of N = 1, . . . , n
under inclusion. We show that if A
1
, A
2
, . . . , A
m
2
N
are such that A
i
A
j
for
all i ,= j, then
m
_
n
n/2|
_
.
Such a family of sets is an antichain of 2
N
. The proof is due to Lubell.
Each maximal chain C of 2
N
consists of sets B
0
, B
1
, . . . , B
n
, where B
0
=
and B
i+1
= B
i
x
i
for some x
i
N. There are n! maximal chains, because at
stage i + 1 we can choose x
i
from a set of n i remaining elements. Also, there
are k!(nk)! maximal chains that contain a xed subset A with [A[ = k. Indeed,
rst you can choose the elements of A in k! dierent orders and then the rest of
the elements in (n k)! dierent ways.
Let t be the number of pairs (i, C), where C is a maximal chain containing A
i
,
and let r
k
denote the number of the sets A
i
of size k. Then
t =
n
k=0
r
k
k!(n k)! .
Counting with respect to the maximal chains, we obtain that t n!, since each
maximal chain contains at most one element from an antichain. Therefore
n
k=0
r
k
_
n
k
_ 1 ,
where the binomial coecient obtains its maximum at k = n/2|, and so
1
1
_
n
n/2
_
n
k=0
r
k
=
m
_
n
n/2
_ .
which proves the claim.
Example 1.61 (Halls Marriage Theorem). Let S = S
1
, S
2
, . . . , S
n
be a family
of subsets of a nite set X such that X =
n
i=1
S
i
. (The sets S
i
can intersect with
each other.) A function : 1, 2, . . . , n X is called a disjoin representative
function for S if it is injective and, for all 1 i n, there exists an integer
j such that (j) S
i
. Here (i) represents S
i
and only S
i
. For an index set
I 1, 2, . . . , n, denote
S(I) =
_
iI
S
i
.
We show that the family S has a distinct representative function if and only
if the following Halls condition holds
[S(I)[ [I[
for all index sets I 1, 2, . . . , n.
If [I[ < [S(I)[ then clearly no distinct representative function can exist. To
prove the suciency of Halls condition, consider the relation on the set P =
1, 2, . . . , n X dened by
x i x S
i
.
Then P is is a partial order (with height h(P) = 2). We show that w(P) n.
To this end, let A be an antichain of P, and let I = A 1, 2, . . . , n. Now A
does not intersect with S(I) X, and hence [A[ [I[ + (n [S(I)[) n, since
[I[ [S(I)[ by Halls condition. Therefore w(A) n, and by Dilworths theorem,
P can be partitioned into n chains. Since the antichain 1, 2, . . . , n is maximal,
each chain contain an element of this set. This proves the suciency of Halls
condition.
1.4. M
OBIUS FUNCTION 26
1.4. Mobius function
The incidence algebra of a poset
For a locally nite poset P denote by
I(P) = f : P P R [ f(x, y) = 0 if x
P
y
the set of all real-valued functions for which f(x, y) = 0 if x / y. The sum and
the scalar product in I(P) are dened in the usual way,
(f +g)(x, y) = f(x, y) +g(x, y) ,
(cf)(x, y) = c f(x, y)
for c R.
Lemma 1.62. If f, g I(P) and c R, also f +g I(P) and cf I(P).
Proof. Exercise.
In the summations we usually leave out the index P, and write [x, y] instead
of [x, y]
P
.
The convolution (or matrix product) of two functions f, g I(P) is
dened by
(f g)(x, y) =
_
_
_
z[x,y]
f(x, z)g(z, y) if x
P
y ,
0 if x
P
y .
This product is well dened, since P is locally nite and hence the sum is over a
nite interval z [x, y]
P
.
The incidence algebra of the locally nite poset P is the set I(P) together
with the operations +, and the scalar product.
Theorem 1.63. Let P be a locally nite poset. The operation of convolution is
associative on I(P), that is, (f g) h = f (g h).
Proof. The claim follows from
((f g) h)(x, y) =
z[x,y]
(f g)(x, z)h(z, y)
=
z[x,y]
_

t[x,z]
f(x, t)g(t, z)
_
h(z, y)
=
t[x,y]
f(x, t)
_

z[t,y]
g(t, z)h(z, y)
_
=
t[x,y]
f(x, t)(g h)(t, y)
= (f (g h))(x, y) .
1.4. M
OBIUS FUNCTION 27
Let P be a locally nite poset, and let be the identity element of the
incidence algebra I(P):
(x, y) =
_
1 if x = y ,
0 if x ,= y .
The function is also known as the delta function and the Kronecker func-
tion of the poset P. It satises the condition f = f = f for all f. A
function f I(P) has an inverse f
1
I(P) if
f f
1
= = f
1
f .
The inversion formula
The zeta function of the poset P is the characteristic function of the poset P,
i.e., it is dened by
(x, y) =
_
1 if x
P
y ,
0 otherwise .
As shown in the next lemma, the zeta function has an inverse, which is called the
M obius function of P, denoted by (=
1
): (x, y) = 0 if x y, and
(1.4) (x, y) =
_
_
_
1 if x = y ,

x
P
z<
P
y
(x, z) if x <
P
y .
In the above the summation is over the half open interval. Note that, by (1.4),
we always have that if x <
P
y, then
(1.5)
z[x,y]
(x, z) = 0 .
Lemma 1.64. Let P be a locally nite poset. The M obius function of P is the
inverse of the zeta function .
Proof. For the cases x <
P
y, we have, by (1.5),
( )(x, y) =
z[x,y]
(x, z)(z, y) =
z[x,y]
(x, z) 1 = 0 .
Also (x, x)(x, x) = 1, and hence = . Similarly, = , and hence
=
1
.
1.4. M
OBIUS FUNCTION 28
Theorem 1.65 (Mobius inversion formula). Let P be a locally nite poset having
a bottom element 0, and let f, g : P R be functions. Then
(1.6) g(x) =
z[0,x]
f(z)
if and only if
(1.7) f(x) =
z[0,x]
g(z)(z, x) .
Proof. Let g be as in (1.6). Then
f(x) =
t[0,x]
f(t)(t, x) =
t[0,x]
f(t)( )(t, x)
=
t[0,x]
_
f(t)
z[t,x]
(t, z)(z, x)
_
=
t[0,x]
z[t,x]
f(t)(t, z)(z, x)
=
z[0,x]
_

t[0,z]
f(t)(t, z)
_
(z, x)
=
z[0,x]
_

t[0,z]
f(t)
_
(z, x) =
z[0,x]
g(z)(z, x) ,
which proves the claim.
When we apply Theorem 1.65 for the dual poset P
d
, we have
Theorem 1.66 (Dual Mobius inversion formula). Let P be a locally nite poset
with a top element 1, and let f, g : P R be functions. Then
g(x) =
z[x,1]
f(z)
if and only if
f(x) =
z[x,1]
(x, z)g(z) .
1.4. M
OBIUS FUNCTION 29
Poset of subsets
For the chain (N, ) of integers, we have the following characterization of its
Mobius function. It follows directly from (1.4).
Theorem 1.67 (Chains). For the chain (N, ), we have
(k, n) =
_
_
1 if k = n,
1 if k + 1 = n,
0 otherwise .
Proof. Assume that k < n. Then (k, n) =
n1
i=k
(i, n). Hence, by (1.5),
if k < n 1, then (k, n) = 0. If k = n 1, then the claim follows from
(n 1, n) +(n, n) = 0.
In this case, the Mobius inversion formula states a rather unsurprising result:
for n > 0,
g(n) =
n
i=0
f(i) f(n) = g(n) g(n 1) .
Theorem 1.68. The M obius function
PQ
of the direct product P Q is the
product of the M obius functions
P
and
Q
of P and Q, that is,
PQ
((x
1
, y
1
), (x
2
, y
2
)) =
P
(x
1
, x
2
)
Q
(y
1
, y
2
) .
Proof. Exercise
For the poset of subsets, we obtain
Theorem 1.69. Consider the poset P = (2
X
, ) for a nite set X. The M obius
function for P is
(Z, Y ) =
_
(1)
|Y ||Z|
if Z Y ,
0 otherwise .
Proof. Recall from Example 1.26, that the poset 2
X
of subsets of X with [X[ = n
is isomorphic to the n-fold direct product
n
C
2
= C
2
C
2
C
2
of the
2-element chains C
2
= 0, 1 where 0 < 1.
The poset C
2
is a chain, and its Mobius function is
C
2
(x, y) = (1)
yx
for x < y ,
(There is only one such pair: (0, 1).)
Let then A and B be subsets of X, and let u = (a
1
, a
2
, . . . , a
n
) = (A) and
v = (b
1
, b
2
, . . . , b
n
) = (B) be their corresponding n-tuples in
n
C
2
. Then, by
1.4. M
OBIUS FUNCTION 30
Theorem 1.68, we have the claim: for A B,
(A, B) =
n
C
2
(u, v) =
n
i=1
C
2
(a
i
, b
i
) = (1)
b
i
a
i
= (1)
|B||A|
.
From Theorem 1.65 we have

Theorem 1.70. Let g, f : 2
X
R be mappings from a nite set X such that
(1.8) f(Y ) =
ZY
g(Z) .
Then for all Y X,
(1.9) g(Y ) =
ZY
(1)
|Y ||Z|
f(Z) .
Example 1.71 (The divisor poset). Consider the divisor poset (N
+
, [) of positive
integers (with the bottom element 1). In this case, we browse through intervals
[k, n] w.r.t. divisibility. An element z is in this interval if and only if z[k and k[n.
Assume rst that n = p
i
for a prime number p. Then the poset D
p
i is a chain
of the i +1 elements 1, p, . . . , p
i
. Hence the corresponding Mobius function
p
i is
p
i (p
k
, p
j
) =
_
_
1 if k = j ,
1 if k + 1 = j ,
0 otherwise .
Now let
n = p
i
1
1
p
i
2
2
. . . p
i
m
m
be the factorization of n 2 into prime numbers. The poset D
n
is isomorphic to
D
p
i
1
1
D
p
i
2
2
D
p
i
m
m
in a natural way. By Theorem 1.68, the Mobius function of the divisor poset is
given by
(k, n) =
_
_
1 if k = n
(1)
t
if n = k p
1
p
2
. . . p
t
for distinct primes p
i
0 otherwise.
The ordinary number theoretic Mobius function is obtained from this, since
(n/k) = (1, n/k) = (k, n), if k[n.
1.4. M
OBIUS FUNCTION 31
Structure of the Mobius function
Theorem 1.72. Let P be a locally nite poset. Then
k
(x, y) is the number of
chains x = x
0

P
x
1

P
x
2

P

P
x
k
= y, where equalities x
i
= x
i+1
are
allowed.
Proof. Exercise.
We have
( )(x, y) =
_
1 if x <
P
y ,
0 if x = y ,
Theorem 1.73. Let P be a locally nite poset. Then ( )
k
(x, y) is the number
of chains x = x
0
<
P
x
1
<
P
x
2
<
P
<
P
x
k
= y in P.
Proof. Exercise.
By the following Halls theorem, the value (x, y) is a local property of the
poset: it depends only on the interval [x, y]
P
.
Theorem 1.74 (Hall). Let P be a locally nite poset, and denote by c
i
(x, y) the
number of chains x = x
0
<
P
x
1
<
P
<
P
x
i
= y of length i. Then
(x, y) =
i0
(1)
i
c
i
(x, y) .
Proof. We use the fact that, in general, (1 +z)
1
=
i=0
(1)
i
z
i
. Now
(x, y) =
1
(x, y) = ( + ( ))
1
(x, y)
= ( ( ) + ( )
2
. . . )(x, y)
= (x, y) ( )(x, y) + ( )
2
(x, y) . . .
= c
0
(x, y) c
1
(x, y) +c
2
(x, y) . . . ,
as required.
Theorem 1.74 can be restated as follows.
Corollary 1.75. Let x, y P for a locally nite poset P. Then
(x, y) =
C
(1)
|C|
,
where the summation is over all chains C from x to y, and where [C[ denotes the
length of the chain C.
2 Lattices
2.1. Denition
Recall that if the least upper bound of A exists in a poset, it is denoted by
_
A, and if the greatest lower bound of A exists, it is denoted by
_
A. If A =
x
1
, x
2
, . . . , x
n
is a nite subset, then we can also write
_
A = x
1
x
2
. . . x
n
,
and
_
A = x
1
x
2
. . . x
n
. We also adopt the notation
_
xA
x =
_
A and
_
xA
x =
_
A, if these exist.
A poset L is a lattice if x y and x y exist for all elements x, y L. In
a lattice and can be regarded as binary operations, called join and meet,
: L L L and : L L L.
Example 2.1. All chains (linearly ordered posets) such as (N, ), (Z, ), (Q, )
and (R, ), are lattices. In these x y = maxx, y and x y = minx, y.
Example 2.2. The divisibility poset (N 0, [ ) is a lattice, where x y =
lcm(x, y), i.e., the least common multiple and x y = gcd(x, y), i.e., the greatest
common divisor.
Example 2.3. The poset of subsets 2
X
is a lattice, where x y = x y and
x y = x y.
Example 2.4. The continuous functions f : [0, 1] R form a lattice, when the
order relation is dened pointwise:
f g f(x) g(x) for all 0 x 1 .
In this case (f g)(x) = maxf(x), g(x) and (f g)(x) = minf(x), g(x).
Example 2.5. The poset Sub(G) of subgroups of a group forms a lattice. Here
G
1
G
2
is the subgroup generated by G
1
G
2
, and G
1
G
2
= G
1
G
2
.
Example 2.6. The factor poset of the set of words
is not a lattice. Indeed, for instance, the incom-

parable words ab and ba are both lower bounds of
A = aba, bab, and A does not have a greatest lower
bound. ab ba
aba bab
In the next lemma it is essential that the subset A is assumed to be nite.
Indeed, in the lattice (N, ) the subset A = 1, 3, 5, . . . does not have a least
upper bound.
Lemma 2.7. Let L be a lattice, and A L a nite nonempty subset. Then the
least upper bound
_
A and the greatest lower bound
_
A both exist.
Proof. We proceed by induction on the size n = [A[. The claim holds for n 2
by the denition. Assume it holds for all subsets of n elements, and let A =
32
2.1. DEFINITION 33
x
1
, x
2
, . . . , x
n+1
. By the induction hypothesis, the least upper bound z =
_
(A x
n+1
) exists, and so does x = z x
n+1
. It is now easy to show that x is
the least upper bound of A.
The case for the greatest lower bound is dual to the above case.
Example 2.8. The Hasse diagram on the right repre-
sents a poset on eight elements. Since it is nite, the
bottom element 0 = O
L
and the top element 1 = 1
L
exist in it. On can check that it is a lattice by observing
that every two elements x, y have a meet x y and a
join x y. Also, for instance,
x (y z) = x 0 = 0 , (x y) z = a z = 0 ,
and so x (y z) = (x y) z.
0
a
x
y
b
1
z
c
In the next lemma some important special cases are mentioned.
Lemma 2.9. Let L be a lattice. If
_
exists in L, then
_
=
_
L is the bottom
element
_
L of L. If
_
exists in L, then
_
=
_
L is the top element
_
L of L.
Proof. Suppose z =
_
exists. Then, by denition, z
L
x for all x L,
and thus z = L. Therefore z is the bottom element of L, which is, again by
denition,
_
L. The case for
_
is similar.
Example 2.10 (Equivalence relations). Let Eq(X) denote the set of all equiva-
lence relations on the set X. It is clear that Eq(X) forms a poset under inclusion.
The set Eq(X) is a lattice where
R
1
R
2
= R
1
R
2
,
R
1
R
2
= (R
1
R
2
)
+
the transitive closure of the union. To see this, one needs to observe that if
R
1
, R
2
Eq(X) then also R
1
R
2
and (R
1
R
2
)
+
are equivalence relations, and
the claim easily follows from this.
In particular,
x(R
1
R
2
)y there is a sequence x
1
, x
2
, . . . , x
n
such that
x
i
(R
1
R
2
)x
i+1
for all i = 1, 2, . . . , n 1
with x = x
1
, y = x
n
there is a sequence x
1
, x
2
, . . . , x
n
such that
x
2i1
R
1
x
2i
and x
2i
R
2
x
2i+1
for all i
with x = x
1
, y = x
n
.
The latter equivalence follows because both R
1
and R
2
are equivalence relations,
and so one can shorten an instance x
i
R
k
x
i+1
and x
i+1
R
k
x
i+2
to x
i
R
k
x
i+2
(for
both k = 1, 2).
2.1. DEFINITION 34
Dual lattices
Lemma 2.11. Let L be a lattice. Then its dual poset L
d
is also a lattice.
Proof. The denition of a lattice is symmetric with respect to the dual order,
and this gives the claim.
In the dual lattice L
d
of a lattice L we denote the join and meet by
d
and
d
, respectively. Then
x
d
y = x y and x
d
y = x y
(where and are the join and meet of L).
Duality is an important tool for lattices. Indeed, for every proposition there
corresponds a dual proposition, which is obtained by changing the order and the
operations and . For instance, the dual proposition of xy
L
y is xy
L
y.
Complete lattices
A lattice L is complete if both
_
A and
_
A exist for all subsets A L. Thus if
L is a complete lattice, it does have a top and a bottom element:
1
L
=
L and 0
L
=
L.
In literature, these are denoted also by and .
Theorem 2.12. Let P be a poset where every subset has a least upper bound.
Then P is a complete lattice.
Proof. Notice rst that, by assumption,
_
exists, and it is the bottom element
0
P
of P as in Lemma 2.9. For a nonempty subset A P, dene
A
= x P [ x
P
y for all y A .
Since 0
P
exists, it is in A
, and so A
,= . By assumption,
_
A
exists and it is
a lower bound of A, since if y A, then, by denition,
_
A

P
y. Also,
_
A
is the greatest lower bound of A by its denition. Therefore the greatest lower
bound
_
A exists for each subset A, and consequently P is a complete lattice.
Therefore in the denition of a complete lattice we might have dropped o
the requirement that
_
A should always exist.
The following result is a dual version of Theorem 2.12.
Theorem 2.13. Let P be a poset where every subset has a greatest lower bound.
Then P is a complete lattice.
2.1. DEFINITION 35
Lattices as algebras
The join and meet are both binary operations in a lattice, and thus we can
consider the lattice (L,
L
) also as an algebra (L, , ) with respect to these
operations.
Theorem 2.14. The following identities are valid in every lattice L:
x x = x , x x = x, (L1)
x y = y x , x y = y x, (L2)
x (y z) = (x y) z , x (y z) = (x y) z , (L3)
x (x y) = x , x (x y) = x. (L4)
Proof. Most of the claims are obvious by the denitions of the operations. We
prove only the rst part of (L4).
First, we have x
L
x and x
L
x y, and hence x
L
x (x y). Also,
if z
L
x (x y), then obviously z
L
x, and therefore x is the greatest lower
bound of x and x y. This shows (L4).
In Theorem 2.14 the properties on the same line are dual to each other. By
(L1), the elements of L are idempotent; by (L2), L is commutative; by (L3),
L is associative (a semigroup) with respect to both operations. The conditions
(L4) are the absorption laws that weave the two operations together.
The associative laws (L3) allows us to remove the brackets from meets (and
joins) so that we can write x
1
x
2
. . . x
n
instead of (. . . (x
1
x
2
) . . . ) x
n
.
Example 2.15. The set of laws (L1) (L4) is not minimal. Indeed, (L1) follows
from the absorption law: First, x = x (x (x x)), when we put y = x x in
(L4). Here x (x x) = x by (L4), and thus x = x x.
Example 2.16 (One law). One can even reduce the laws to one law, which is a
bit more complicated than (L1) (L4). There are several such laws. The rst one
was given by R. McKenzie in 1970. The length of the law was 300 000 involving
34 variables. The following is due to McCune, Padmanabhan, and Vero:
(((y x) x) (((z (x x)) (u x)) v)) (w ((s x) (x t))) = x.
Therefore an algebra (X, , ) satises this law if and only if it is a lattice.
Lemma 2.17. In a lattice L we have, for all x, y L,
x
L
y x y = x x y = y .
Proof. These are obvious by the denitions of meet and join.
2.2. ISOMORPHISM AND SUBLATTICES 36
For two binary operations , : L L L state the following conditions
from Theorem 2.14:
x x = x , x x = x, (A1)
x y = y x , x y = y x, (A2)
x (y z) = (x y) z , x (y z) = (x y) z , (A3)
x (x y) = x , x (x y) = x. (A4)
Theorem 2.18. Let L be a set with two binary operations and that satisfy
the conditions (A1) (A4). Then L is a lattice with respect to the order relation
dened by
x y x = x y .
Proof. It is an exercise to show that is a partial order. We show that x y =
x y under the relation .
Now,
x (x y)
(A3)
= (x x) y
(A1)
= x y ,
and hence x y x. Similarly, using (A2), we obtain that x y y, and so
x y is a lower bound of x, y. Assume then that for an element z L with
z x and z y. Then z = z x and z = z y, and we have
z = z y = (z x) y = z (x y)
and so z x y. This shows that x y is the greatest lower bound of x and y.
For the least upper bound, we need (A4). It guarantees that x x y for
all x and y. Then dually to the above we can show that x y is the least upper
bound of x and y.
Notice that, by the preceding proof, the conditions (A1), (A2) and (A3) for
the operation imply that the dened order is a partial order for which the
meet x y always exists. Such posets are called (meet) semilattices. The
dual conditions (A1), (A2) and (A3) for imply that is also a poset, a (join)
semilattice. Thus a semilattice is a commutative semigroup (A, ), where all
elements are idempotent.
2.2. Isomorphism and sublattices
Isomorphism
Two lattices are isomorphic if they are order isomorphic as posets.
Lemma 2.19. Let : L K be a mapping between the lattices L and K. Then
the following conditions are equivalent for :
(a) is an isotone mapping.
(b) (x y)
K
(x) (y) for all x, y L.
(c) (x y)
K
(x) (y) for all x, y L.
Proof. We always have that x
L
y implies x
L
xy. Hence if is isotone, then
(x)
K
(x y) and similarly, (y)
K
(x y). These prove the implication
(a) (b).
On the other hand suppose that (b) holds, and let x
L
y. Now, y = x y,
and by (b), (y) = (x y)
K
(x) (y), which gives that (x)
K
(y).
Hence we have also (b) (a).
The equivalence of the cases (a) and (c) is dual to the above case.
A mapping : L K between two lattices is
join preserving if (x y) = (x) (y) ;
meet preserving if (x y) = (x) (y) ;
a (lattice) homomorphism if it is both join and meet preserving;
a (lattice) isomorphism if it is a bijective (lattice) homomorphism;
an automorphism if L = K and is an isomorphism.
Theorem 2.20. Let L and K be lattices. Then L

= K (as posets) if and only if
there exists a lattice isomorphism : L K.
Proof. Suppose rst that L and K are order isomorphic, and let : L K be
such that, as in Theorem 1.20,
(2.1) x
L
y (x)
K
(y) .
Let x, y L. Then x
L
x y and y
L
x y give (x) (y)
K
(x y).
On the other hand, since maps L onto K, there exists a z L such that
(z) = (x) (y). We have (x)
K
(z) and by (2.1), also x
L
z. Similarly,
y
L
z, and so x y
L
z. Finally, by (2.1), (x y)
K
(z) = (x) (y) ,
and thus (x y) = (x) (y) as required. A dual argument shows that
(x y) = (x) (y), and hence is a lattice homomorphism. Since it is
bijective, it is an isomorphism.
On the other hand, if is a lattice isomorphism, then
x
L
y x y = y (x y) = (y)
(x) (y) = (y) (x)
K
(y) ,
and hence is an order isomorphism by Theorem 1.20.
If : L K is an isomorphism, then also its inverse
1
: K L is an
isomorphism. The relation

= is an equivalence relation among lattices. As usual
in algebra, isomorphic lattices are often identied with each other.
Example 2.21. In the following gures there are the Hasse diagrams of the nite
lattices, up to isomorphism, of at most ve elements.
Sublattices
Let L be a lattice. Then a nonempty subset K L is a sublattice of L if K is
closed under the join and meet operations of L:
x y K and x y K whenever x, y K .
By the denition of a sublattice, for all x, y K, x
K
y if and only if x
L
y.
We denote by Sub(L) the set of all sublattices of the lattice L. We also say that
a lattice K embeds into L, if K is isomorphic to a sublattice of L. In this case
we often say just that K is a sublattice of L.
Example 2.22. Each chain C
n
of height n is a sublattice of C
m
for m n. Note
that C
n
can be embedded in many dierent ways into C
m
for large n < m.
Example 2.23. If L is not a chain then it has a square as a sublattice: there
are incomparable elements x, y L such that x y, x, y, x y form a square.
Example 2.24. Let L be a lattice and x
L
y, then the (closed) interval [x, y]
L
=
z [ x
L
z
L
y is a sublattice of L. The open intervals need not be sublattices
as can be seen from the four element square lattice.
Example 2.25. Consider the lattice (T
12
, [ ) of divisors
of 12. Then T
12
has, among others, the following sub-
lattices: all singleton sets d with d[12; all sets 1, d
with d[12; the sets 1, 3, 4, 12 and 2, 4, 6, 12.
1
2
4
3
6
12
Example 2.26. A sublattice of a lattice is always a lattice, but a lattice that
is a subset of a lattice is not necessarily a sublattice! To see this, consider the
lattice L = (2
{1,2,3}
, ) and the chain K consisting of the sets , 1, 2, 1, 2, 3
under inclusion. The lattice K is a subset of L, but
1 2 = 1, 2 in L and 1 2 = 1, 2, 3 in K .
In this example, the lattice operations and are not the same in K and L.
However, the lattice K can be embedded into L by the homomorphism dened
by () = , (1) = 1, (2) = 2, (1, 2, 3) = 1, 2.
As seen the equivalence relations Eq(X) and the subgroups Sub(G) of a group
G form a lattice. In fact, these lattices have the general form:
Theorem 2.27 (Whitman). Every lattice L can be embedded into the lattice of
equivalence relations Eq(X) on some set X.
Proof. Omitted.
Theorem 2.28 (Birkho). Every lattice L can be embedded into the lattice of
subgroups Sub(G) of some group.
Proof (Idea). Consider L as a lattice of equivalence relations with the embed-
ding : L Eq(X) with (x) =
x
. The group G in the claim can be chosen
to consist of the permutations on L that move, (x) ,= x, only nitely many
elements of L. Then
(x) = G [ (y, (y))
x
for all y L
is a required embedding.
Theorem 2.29. For each lattice L, the set Sub
0
(L) = Sub(L) forms a
complete lattice under inclusion.
Proof. Let S = K
i
[ i I be a family of sublattices of L such that K =
iI
K
i
. Then, if K ,= , it is a sublattice of L. (This can be quickly shown
relying only on the denition of a sublattice.) In this case,
_
S = K, and the
claim follows from Theorem 2.13.
In particular, if A L is any subset of the lattice L, then there exists the
smallest sublattice of L containing A:
Sg(A) =
K [ K Sub(L) and A K .
Sg(A) is called the sublattice of L generated by A.
Example 2.30. Consider the divisor lattice T
12
. Here Sg(4, 6) = 2, 4, 6, 12.
It is the smallest subset that is closed under both operations and .
2.3. CONGRUENCES AND IDEALS 40
2.3. Congruences and ideals
Congruences
Let L be a lattice, and L L an equivalence relation of L. Then is a
congruence if for all x, y, z L,
xy = (x z)(y z) and (x z)(y z) .
The set of all congruences of L is denoted by Con(L). The equivalence class
x = y [ xy is called the congruence class of the element x.
For a subset A L is convex if
x, y A = [x, y]
L
A.
Notice that a convex subset need not be a sublattice.
Theorem 2.31. Let Con(L) for a lattice L. Then the congruence class x
is a convex sublattice for all x L. Moreover, if xy then [x y, x y]
L
x.
Proof. First of all, x is a sublattice, since if a, b x then
(a b)(x b) and (x b) = (b x)(x x) = x
and so (ab)x, which gives that ab x. Similarly, we obtain that ab x.
For convexity, suppose that a, b x and a
L
z
L
b. Then ab and thus
also (a z)(b z). Here a z = a and b z = z, and so az, which means that
z a = x. Hence [a, b]
L
x. This shows that x is convex.
For the second claim, since x is a sublattice, if y x then also x y x
and x y x, and so [x y, x y]
L
x as required.
Example 2.32. Consider the 4-element square lattice L.
We can represent L as a table, where the element in the
crossing of (x, y) is x y/x y.
/ 0 a b 1
0 0/0 0/a 0/b 0/1
a 0/a a/a 0/1 a/1
b 0/b 0/1 b/b b/1
1 0/1 a/1 b/1 1/1 0
a
b
1
Let be a congruence of L. Then the following cases hold:
0a b1 , 0b a1 , (2.2)
ab = 0a and so =
L
, (2.3)
01 = =
L
. (2.4)
Using these properties we deduce that there are four congruences in Con(L): the
identity congruence
L
, the universal congruence
L
, and the congruences
1
and
2
that have the following congruence classes:
1
: 0, a, b, 1 and
2
: 0, b, a, 1 .
Theorem 2.33. Let

i
[ i I be a set of congruences of a lattice. Then the
intersection
iI

i
is also a congruence of L. In particular, if R is an equiv-
alence relation on L, then there there exists the smallest congruence in Con(L)
containing R.
Proof. Notice that
L

i
for all i, and hence also the intersection contains
L
,
and so the intersection is nonempty. The claim reduces then to the individual
congruences.
For an equivalence relation R, let R) denote the smallest congruence con-
taining R, that is,
R) =
[ Con(L), R .
Theorem 2.34. The set Con(L) of congruences forms a complete lattice under
inclusion. The operations of this lattice are
1

2
=
1

2
and
2

2
=
1

2
) .
Proof. The claim follows from Theorem 2.13.
Theorem 2.35. The congruence lattice Con(L) is a sublattice of the lattice Eq(L)
of all equivalence relations on L.
Proof. Exercise.
Theorem 2.36. Let
1
,
2
Con(L) for a lattice L. Then
1

2
= , where
(x, y) there is a nite sequence
x y = x
0

L
x
1

L

L
x
n
= x y
where (x
i
, x
i+1
)
1

2
for all i .
Proof. Exercise.
Example 2.37. Consider the given 6-element lattice L together with its table
for operations.
/ 0 1 2 3 4 5
0 0/0 1/0 2/0 3/0 4/0 5/0
1 1/0 1/1 3/0 3/1 5/0 5/1
2 2/0 3/0 2/2 3/2 4/2 5/2
3 3/0 3/1 3/2 3/3 5/2 5/3
4 4/0 5/0 4/2 5/2 4/4 5/4
5 5/0 5/1 5/2 5/3 5/4 5/5
0
1
2
3
4
5
The congruences of L are the following in addition to the identity
L
and the
universal congruence
L
:
0, 1, 2, 3, 4, 5, 0, 2, 1, 3, 4 5
0, 1, 2, 3, 4, 5, 0, 2, 4, 1, 3, 5,
2, 4, 3, 5, , 0, 1, 0, 1, 2, 3, 4, 5 .
If is a congruence, then we denote its congruence classes by i . . . j, where
i, . . . , j are the elements in the class. Then the congruence lattice Con(L) is given
in Figure 2.1.
01, 23, 45 02, 13, 4, 5 24, 35, 0, 1

0123, 45 01, 2345 024, 135
L
Fig. 2.1. The congruence lattice of the example.
The congruence classes x, x L, partition the lattice L. The quotient

lattice (modulo ) is dened as the set
Q = L/ = x [ x L
of all congruence classes together with the operations
x y = (x y) and x y = (x y) .
These operations are well dened, and hence L/ is a lattice.
Ideals
A nonempty subset I L of a lattice L is an ideal if,
for all x, y I,
x y I and x I .
The set of all ideals of a lattice L is denoted by Id(L).
I
Notice that each ideal is a down-set and a sublattice. Indeed, I = I and for
all x, y I, x y I and x y I.
The lattice L is its own ideal, L Id(L). If L has other ideals, they are called
proper. Note that if the lattice L has the bottom element 0
L
then 0
L
I for
every ideal of L.
We notice that always
x = x y [ y L ,
and therefore
Lemma 2.38. Let I ,= be a subset of a lattice L. Then I is an ideal of L if
and only if I is closed under nite joins and, moreover,
x I and y L = x y I .
Lemma 2.39. Let I
i
be a set of ideals of a lattice L for all i A. Then also
iA
I
i
is an ideal of L if it is nonempty. In particular, every subset X L has
the smallest ideal containing X:
(X] =
I [ I Id(L) and X I .
Proof. Exercise.
The ideal (X] is called the ideal generated by X. The ideal generated by
a singleton set x is a principal ideal, and it is denoted by (x] (i.e., without
set brackets). Hence (x] = x, since if y
L
x and z
L
x then also y z
L
x.
Example 2.40. In the subset lattice 2
X
, the ideal generated by a subset Y X
is simply (Y ] = 2
Y
.
Lemma 2.41. Let X be a nonempty subset of a lattice L. Then for all x L,
x (X] if and only if there exists a nite subset Y X such that x
L
_
Y .
Proof. If Y X is a nite subset then
_
Y exists and
_
Y (X], since ideals
are closed under nite joins. Hence if x
L
_
Y then also x (X]. Now, the set
I = x L [ x
L
F for some nite F X

is an ideal, since if x, y I, say x
L
_
F
x
and y
L
_
F
y
for some nite subsets
F
x
, F
y
X, then x y
L
__
F
x
F
y
_
, and hence x y I. It is clear that if
x I and y
L
x, then also y I. Now X I, and therefore I = (X].
Theorem 2.42 (Ideal Lattice). The set Id(L) of ideals of a lattice L is a lattice
under set inclusion (called the ideal lattice of L). The operations in Id(L) are
the following:
I
1
I
2
= I
1
I
2
= x y [ x I
1
, y I
2
,
I
1
I
2
= (I
1
I
2
] = z L [ z x y for some x I
1
, y I
2
.
Proof. First of all, always I
1
I
2
,= , since x y I
1
I
2
whenever x I
1
and
y I
2
. Moreover, if x I
1
I
2
then x = x x where x I
1
and x I
2
. Hence
x y [ x I
1
, y I
2
= I
1
I
2
. By Lemma 2.39, I
1
I
2
Id(L), and, surely
then I
1
I
2
= I
1
I
2
.
For the join, let
J = z L [ z x y for some x I
1
, y I
2
.
The J is obviously an ideal, and I
1
I
2
J. Therefore also (I
1
I
2
] J. Now,
if xy J for some x I
1
and y I
2
then x, y I
1
I
2
and so xy (I
1
I
2
].
This shows that J = (I
1
I
2
] and hence J = I
1
I
2
.
Theorem 2.43. If L is a nite lattice then every ideal of L is principal.
Proof. If X L, then (X] = (
_
X] by the above results.
Example 2.44. If X is an innite set then its nite
subsets form an ideal of the subset lattice 2
X
that is
not a principal ideal.
Example 2.45. Consider the lattice L on the right.
There (2, 4] = 1, 2, 4 = (4], and (4, 5] =
1, 2, 4, 5, 10] = (10]. Of course, (20] = L the full lattice,
since 20 is the top element in L.
1
2
4 5
10
20
Recall that a lattice L embeds into a lattice K if there exists an injective
homomorphism : L K.
Theorem 2.46 (Embedding result). Every lattice L can be embedded into its
ideal lattice. Moreover, if L is a nite lattice then L

= Id(L).
Proof. Let, for all x L,
(x) = (x] .
The mapping : L Id(L) is clearly injective, and it is a homomorphism, since,
by Theorem 2.42, (x] (y] = (x y] and (x] (y] = (x y].
The second claim follows from Theorem 2.43.
Prime ideals
A proper ideal I, I ,= L, is said to be a prime ideal if
(2.5) x y I = x I or y I .
In the following (2.5) is changed to equivalence.
Lemma 2.47. A proper subset I of a lattice L is a prime ideal if I is closed
under nite joins and
(2.5) x y I x I or y I .
Proof. Exercise.
Let C
2
be the chain on two elements 0 and 1.
0
1
Theorem 2.48 (Prime Embedding). Let I Id(L) be an ideal of a lattice L.
Then I is a prime ideal if and only if there exists a surjective homomorphism
: L C
2
such that
1
(0) = I.
Proof. Suppose rst that I is a prime ideal, and dene
(x) =
_
0 x I ,
1 x / I ,
for all x L. Then is surjective, since every prime ideal is proper and nonempty.
That is a homomorphism, follows from
(x y) = 0 x y I x, y I
(x) = 0 = (y)
(x) (y) = 0 ,
and
(x) (y) = 0 x y I
x I or y I
(x) = 0 or (y) = 0
(x) (y) = 0 .
2.4. FIXED POINTS 46
Conversely, suppose that : L C
2
is a surjective homomorphism. Then
1
(0) is a nonempty proper subset of L. If (x) = 0 = (y) then (x y) =
(x) (y) = 0, and so x y
1
(0). Furthermore,
x y
1
(0) (x y) = 0
(x) (y) = 0
(x) = 0 or (y) = 0 .
Hence
1
(0) is a prime ideal.
Filters
Dually, we say that a nonempty subset F L of a lattice L is a lter if, for all
x, y F,
x y F and x F .
We dene proper lters and prime lters accordingly.
A lter is always an up-set of L, F = F. The lter generated by a subset
X L is denoted by [X). It is the smallest lter containing X.
The results for ideals can be easily modied for lters by taking dual state-
ments of the claims. Indeed, a lter F of L is an ideal of the dual lattice L
d
.
Theorem 2.49. Let A be a subset of a lattice L. Then A is convex if and only
if A = I F for some ideal I and lter F (with nonempty intersection).
Proof. Assume that A is convex, and let I = (A] and F = [A). Then A I F.
On the other hand, if x I F then there are elements y and z in A such that
y
L
x
L
z, and so x A. Hence A = I F.
Suppose then that A = I F for an ideal I and lter F. Note rst that
A is a sublattice. Indeed, if x, y A then x y I and x y F, and so
x y I F = A. Similarly, x y A. Now if x, y A and x
L
z
L
y then
z I F, and we conclude that A is convex.
2.4. Fixed points
Continuous functions
Let : L L be a function on the lattice L. An element x L is a xed point
of if (x) = x. The set of all xed points of is denoted by Fix().
Let : L L be a function on L. Then is continuous if preserves
existing least upper bounds: if A L is such that
_
A exists, then also
_
(A)
exists and
_
(A) =
_ _
A
_
.
Especially if L is a complete lattice, then
_
A exists for all subsets A, and
then a continuous function satises
_
(A) =
_ _
A
_
for all A.
Lemma 2.50. Every continuous function on a lattice L is isotone.
Proof. Let be continuous on L, and let x
L
y. Denote A = x, y. Then
(x)
L
(A) =
_
A
_
= (y) .
Here, of course, the joins do exist, since A is nite.
KnasterTarski
Theorem 2.51 (KnasterTarski). Let L be a complete lattice, and : L L be
an isotone mapping. Then has a xed point. In fact, has a largest xed point
z
max
and a least xed point z
min
given by:
z
max
=
x L [ x
L
(x) ,
z
min
=
x L [ (x)
L
x .
Proof. We show that z
max
is the largest xed point of . The case for z
min
can
be proved dually. Let
A = x L [ x
L
(x) .
So that z
max
=
_
A. Observe at this point that A ,= , since at least 0
L
A.
We show that z
max
is a xed point of .
First of all, for every x A, we have x
L
z
max
. As is an isotone mapping,
it follows that (x)
L
(z
max
), an so, for every x A, x
L
(x)
L
(z
max
).
Thus (z
max
) is an upper bound for A, and so z
max

L
(z
max
). From this it fol-
lows that (z
max
)
L
((z
max
)), and hence (z
max
) A. Therefore (z
max
)
L
z
max
, as z
max
is an upper bound for A. We conclude that z
max
= (z
max
).
To show that z
max
is the largest xed point of , let x Fix(). Then
x
L
(x), and thus x A and therefore x
L
_
A = z
max
.
Example 2.52. In the above proof we need the existence of the bottom ele-
ment 0
L
. Indeed, consider the chain . . . , 1, 0, 1, 2, . . . , m. Here
_
A exists for
all subsets A, but the isotone mapping dened by (k) = k 1 does not have
any xed points.
Example 2.53 (SchroderBerstein). Let X and Y be any sets such that there
are injective mappings : X Y and : Y X. Then there exists a bijection
: X Y .
We prove this using the xed-point theorem. Let : 2
X
2
X
be the mapping
dened by
(A) = X
_
Y (A)
_
.
Then is an isotone mapping: if A B then (A) (B). Since 2
X
is a
complete lattice with respect to union and intersection, has xed point A:
A = X
_
Y (A)
_
.
Now, X = AB for B = X A, and Y = (A) C, where C = Y (A), and so
(C) = B, since A was a xed point. Now, maps C bijectively onto B. Dene
then : X Y as follows:
(x) =
_
(x) if x A,
1
(x) if x B.
Then is a required bijection.
Theorem 2.54 (DavisTarski). A lattice L is complete if and only if every iso-
tone mapping : L L has a xed point.
In the other direction this is Theorem 2.51. We omit the proof of the converse.
AbianBrown
A mapping : P P is said to be increasing if x
P
(x) for all x P. A
poset P is chain-complete if
_
C exists for every chain C in P. The following
theorem is also known as the BourbakiWitt theorem.
Theorem 2.55 (AbianBrown). If P is a chain-complete poset then every in-
creasing function has a xed point.
Proof. We say that a subset A P is ()closed if (A) A and
_
C A for
all chains C A inside A. Let z P be any element. Clearly every intersection
of closed sets is closed, and so
Z =
A [ z A and A is closed
is also closed. The full poset P is closed by denition, and therefore Z ,= . Also,
if A is closed and z A, then the set A z satises these conditions, since is
increasing. Hence Z z, i.e., z
P
y for all y Z.
An element x Z is normal, if for all y Z
(2.6) y <
P
x = (y)
P
x.
Claim 1. Let x be normal. Then for all y Z, either y
P
x or y
P
(x).
Proof of Claim 1. We show that the set
A = y Z [ y
P
x or y
P
(x)
is closed and it contains the element z, and thus that A = Z, by the denition
of Z. This will prove the present claim.
It is clear that z A, since z
P
y for all y Z.
Let then y A. We show that also (y) A. If y <
P
x then (y)
P
x
by (2.6), and hence (y) A. If y = x then (y)
P
(x), and so (y) A.
If y >
P
x then y
P
(x), since y A, and (y)
P
y
P
(x), since is
increasing. Hence in all cases, (y) A, and so A is closed.
Let then C be a chain in A. We show that
_
C A. If y
P
x for all y C
then
_
C
P
x, and so
_
C A. Suppose thus that there exists an element y C
such that x <
P
y. Then y
P
(x), since y A. Hence
_
C
P
y
P
(x) and
therefore
_
C A as required.
Claim 2. All elements in Z are normal.
Proof of Claim 2. Denote by N = x [ x Z is normal. We show that N is
closed and it contains z.
First of all, z N, since z
P
y for all y Z.
Let then x N. We show that also (x) N. Now y
P
(x) implies
y
P
x by Claim 1. Therefore (y) = (x) or (y)
P
x
P
(x) since x N.
This implies that (x) N.
Let C be a chain in N. Now y <
P
_
C implies that y <
P
x for some x C,
and so (y) <
P
x by normality of x. Hence (y)
P
_
C. Since the set N is
closed, N = Z.
We proceed the proof of the theorem. For all x, y Z, we have that y
P
x or
y
P
(x)
P
x. Therefore Z is a chain. As Z is closed,
_
Z Z, which implies
that also (
_
Z) Z and consequently (
_
Z)
P
_
Z, that is, (
_
Z) =
_
Z
since is increasing.
3 Special Lattices
3.1. Distributive and modular lattices
Distributive law
Lemma 3.1. Let L be any lattice. Then
x (y z)
L
(x y) (x z)
for all x, y, z L.
Proof. Exercise.
A lattice L is said to be distributive if there is an equality in the above
formula, i.e., if L satises the distributive law: for all x, y, z L,
(D1) x (y z) = (x y) (x z) .
Example 3.2. Not all lattices are distributive. The following two basic lattices
N
5
(pentagon) and M
5
(diamond) are not distributive. In N
5
we have
d (b c) = d ,= c = (d b) (d c)
and in M
5
,
b (c d) = b ,= a = (b c) (b d) .
a
b
c
d
e
N
5
a
b
c
d
e
M
5
Example 3.3. The subset lattice 2
X
of a set X is distributive. The law (D1) is
well known to hold for the subsets with respect to the operations and . Also,
all chains are trivially distributive.
Lemma 3.4. A lattice L is distributive if and only if it satises
(D2) x (y z) = (x y) (x z) .
for all x, y, z L.
50
3.1. DISTRIBUTIVE AND MODULAR LATTICES 51
Proof. Note that (D2) is the dual of (D1). Assume (D1). We have
(x y) (x z)
(D1)
=
_
(x y) x
_
_
(x y) z
_
(L4)
= x
_
(x y) z
_
(L2)
= x
_
z (x y)
_
(D1)
= x
_
(z x) (z y)
_
(L3)
=
_
x (z x)
_
(z y)
(L4)
= x (z y)
(L2)
= x (y z) .
The converse is proved similarly.
Theorem 3.5 (FunayamaNakayama). The congruence lattice Con(L) of a lat-
tice is distributive.
Proof. Exercise.
Modular lattices
Lemma 3.6. Let L be any lattice. Then
x (y z)
L
y (x z)
for all x, y, z L with x
L
y.
Proof. Exercise.
If in (D2) we assume that x
L
y then x y = y and there is an equality in
Lemma 3.6: x (y z)
(D2)
= (x y) (x z) = y (x z).
We say that the lattice L is modular, if it satises
(M) if x
L
y then x (y z) = y (x z)
for all x, y, z L.
Example 3.7. The lattice L = N
5
of Example 3.2 is not modular, since c
L
d
but
c (d b) = c ,= d = d (c b) .
All distributive lattices are modular, and so is the nondistributive lattice M
5
.
Lemma 3.8. A lattice L is modular if and only if for all x, y, z L:
(M) (x y) (x z) = x (y (x z)) .
Proof. Exercise.
Theorem 3.9. A lattice L is modular if and only if it has no sublattices (iso-
morphic to) N
5
.
Proof. If N
5
is a sublattice of L then clearly the condition (M) is not satised
in L.
Assume then that L does not satisfy (M), and let x, y, z L be such that
x
L
y and x (y z) <
L
y (x z) .
(Recall that, by Lemma 3.6, for all lattices we do have
L
in the above.)
Denote
a = x (y z) and b = y (x z) .
Then a <
L
b, and
z a = z (x (y z))
(L2,3)
= (z (y z)) x
(L4)
= z x,
z b = z (y (x z))
(L2,3)
= (z (y z)) y
(L4)
= z y .
Moreover z y
L
x (z y)) = a <
L
b, and so
z y
L
z a
L
z b = z y ,
that is,
z y = z a = z b .
Similarly we obtain that
z x = z a = z b ,
and therefore we have an N
5
in L.
z y
z
a
b
z x
Theorem 3.10. A lattice L is distributive if and only if it has no sublattices N
5
and M
5
.
Proof. It is clear that if either N
5
or M
5
is a sublattice then L is not distributive.
Assume then that L is not distributive, and suppose that L does not contain
N
5
. Thus, by Theorem 3.9, L is modular. By Lemma 3.1 and the denition of a
distributive lattice, there are elements x, y, z L such that
(x y) (x z) <
L
x (y z) .
Dene then the elements
u = (x y) (x z) (y z) ,
v = (x y) (x z) (y z) ,
a = (x v) u,
b = (y y) u,
c = (z y) u.
These elements generate an M
5
in L as can be seen from
the following claims the proofs of which are exercises.
Claim. u
L
a
L
v.
Similar claims hold for b and c.
Claim. u <
L
v.
Claim. a b = u and a b = v.
Similar claims hold for the other cases of a, b, c.
u
a
b
c
v
Example 3.11. It can be shown that a lattice L is distributive if and only if it

satises the law:
(D3) (x y) (y z) (z x) = (x y) (y z) (z x) .
It is rather obvious that the special lattices M
5
and N
5
do not satisfy (D3), and
therefore it remains to be shown that every distributive lattice does satisfy (D3).
We omit this proof.
Modularity of normal subgroups
Let G be a group. Then a subgroup H is a normal subgroup, denoted by
H G, if for all g G and h H, also g
1
hg H. Denote by N(G) the set of
the normal subgroups of G.
Theorem 3.12. N(G) is a modular lattice under inclusion.
Proof. Let H
1
, H
2
N(G). Then H
1
H
2
G and hence H
1
H
2
= H
1
H
2
.
Also, H
1
H
2
G and H
1
H
2
H
1
H
2
, since the identity element is in both H
1
and H
2
. This gives
H
1
H
2
= H
1
H
2
.
(Notice that if G
1
and G
2
are just subgroups of G then G
1
G
2
might not be a
subgroup. The case is favourable when these are normal subgroups: if h
1
, g
1
H
1
and h
2
, g
2
H
2
then h
1
h
2
g
1
g
2
= h
1
g
1
(g
1
1
h
2
g
1
)g
2
H
1
H
2
.)
Recall now Dedekinds law which states that if H
1
, H
2
, H
3
are subgroups of
G and H
1
H
2
then H
1
(H
2
H
3
) = H
2
H
1
H
3
. (Here, in general, H
1
(H
2
H
3
)
and H
1
H
3
need not be subgroups.)
So for normal subgroups H
1
, H
2
, H
3
with H
1
H
2
, we have
H
1
(H
2
H
3
) = H
2
(H
1
H
3
) ,
and so N(G) is modular.
JordanDedekind condition
Theorem 3.13. Let L be a modular lattice and x, y
L. Then
[x y, x]
L

= [y, x y]
L
,
where the isomorphism is given by (z) = z y.
x y
x
y
x y
z
z y
Proof. Consider the mapping in the reverse direction:
: [y, x y]
L
[x y, x]
L
with (z) = z x.
Then we have (z) = z for all z [x y, x]
L
, since
(z) = (z y) = (z y) x
(M)
= (x y) z = z .
Similarly, (z) = z for all z [y, x y]
L
, and hence =
1
, which shows that
is a bijection. Clearly, both and
1
are isotone, and so the claim follows.
We have then the following corollary:
Theorem 3.14. If L is a modular lattice and x, y L then, for the cover relation,
x y x y x y .
Proof. If x y x then [x y, x]
L
is a chain of two elements, and thus, by
Theorem 3.13, so must be [y, x y]
L
, that is, y x y. The converse goes
similarly.
For a lattice L, a sequence
x = x
0
x
1
. . . x
n
= y
is a cover chain from x to y of length n.
Theorem 3.15 (JordanDedekind). Let L be a modular lattice. Then any two
cover chains from x to y have the same length.
Proof. We prove the claim by induction on the length of cover chains. If x y
then this is the only cover chain from x to y. Assume then that the claim holds
for chains of length at most n 1, and let
x = x
0
x
1
. . . x
n
= y ,
x = y
0
y
1
. . . y
m
= y
be two cover chains from x to y. If x
1
= y
1
, then the claim follows by the
induction hypothesis. Suppose then that x
1
,= y
1
. Now x = x
1
y
1
and so
x
1
y
1
x
1
, which gives by Theorem 3.14 that y
1
x
1
y
1
. Symmetrically, we
have that x
1
x
1
y
1
. Again if x
2
= y
2
, then the claim follows by the induction
hypothesis, and thus assume that x
2
,= y
2
. By symmetry, we can suppose that
x
2
,= x
1
y
1
. (Otherwise, y
2
,= x
1
y
1
.) Then x
2
/ [x
1
y
1
, y]
L
, since x
1
x
2
and x
1
x
1
y
1
both hold.
The interval [x
1
y
1
, y]
L
is nite, since otherwise the in-
terval [x
1
, y]
L
would contain a forbidden sublattice M
5
.
(Take an element z from the interval [x
1
y
1
, y]
L
that is
not in the chain of x
i
s, and the elements x
1
, x
2
, x
1
y
1
and y.) By the induction hypothesis, every cover chain
from x
1
to y has length n 1 including the ones that
go through x
1
y
1
, and every cover chain from y
1
to
y has length m1 including the ones through x
1
y
1
.
Therefore n = m as was required.
x
x
1
y
1
x
1
y
1
y
Ideals of distributive lattices
Lemma 3.16. Let L be a distributive lattice. Then
x y = x z and x y = x z = y = z .
Proof. Indeed, suppose that x y = x z and x y = x z. Then
y = y (x y) = y (x z)
(D1)
= (y x) (y z)
= (x z) (y z) = z (x y)
= z (x z) = z .
Theorem 3.17. A lattice L is distributive if and only if for all ideals I, J Id(L),
I J = x y [ x I, y J .
Proof. Assume rst that L is distributive, and let x I J. By Theorem 2.42,
there are elements a I and b J such that x
L
a b. Therefore x =
x (a b)
(D1)
= (x a) (x b), where x a I and x b J by the denition
of an ideal. Hence x has the required form.
Suppose now that L is not distributive. Then L contains the sublattice N
5
or M
5
.
Case N
5
. Let I = (b] and J = (c]. Then a
L
b c,
and hence a I J. Suppose that there are elements
x I and y J such that a = x y. In this case,
y
L
a. Since y J = (c], also y
L
c, and so y
L
a c <
L
b. Therefore y I = (b]. Now x, y I imply
that a = x y I, that is, a
L
b; a contradiction.
a c
c
b
a
b z
Case M
5
. Let I = (b] and J = (c]. Then a
L
b c
and hence a I J. As in the previous case, we obtain
a contradiction from the assumption that a = x y for
some x I and y J.
u
a
b
c
v
Theorem 3.18. Let L be a distributive lattice, I an
ideal and F a lter of L such that I F = . Then
there exists a prime ideal P of L such that I P and
P F = .
I
P
F
Proof. Let
S = J Id(L) [ I J, J F = .
Then S is nonempty, because I S. Let C be any ascending chain in S, and let
M =
C. Then M is an ideal of L, I M and M F = , that is, M S.

Therefore every ascending chain in S has an upper bound in S. By Zorns lemma,
there exists a maximal element P in S.
Of course, P is an ideal. Suppose that P is not a prime ideal. Then there
are elements x, y / P such that x y P, and hence P P (x] properly.
By maximality of P, we have that (P (x]) F ,= . Let z F be in P (x].
Then by Theorem 3.17, there exists an a P such that z = a x. Similarly
(P (y]) F ,= , and there exists a b P such that b y F. Since F is a
lter, (a x) (b y) F. However,
(a x) (b y)
(D1)
= (a b) (a y) (x b) (x y) P ,
gives that P F ,= ; a contradiction.
Corollary 3.19. Let I be an ideal in a distributive lattice L, and let x / I. Then
there is a prime ideal P of L such that I P and x / P.
Proof. Consider the principal lter F = [x). Then I [x) = , and the claim
follows from Theorem 3.18.
Corollary 3.20. Let L be a distributive lattice. Then every ideal I of L is the
intersection of some prime ideals.
Proof. We have
I =
P [ P prime ideal, I P .
Corollary 3.21. Let L be a distributive lattice L, and let x, y L be dierent

elements. Then there exists a prime ideal P that contains exactly one of them.
Proof. Apply the previous corollary to the principal ideals (x] and (y].
BirkhoStone Theorem
Let R 2
X
be a set of subsets of X. Then R is a set ring, if A B R and
A B R for all A, B R. Clearly, every set ring is a distributive lattice.
Theorem 3.22 (BirkhoStone). Each distributive lattice L is isomorphic to a
set ring.
Proof. For each element x L, dene
(x) = P [ P a prime ideal, x / P .
Let R = (x) [ x L.
Claim. The mapping p: L R is a homomorphism and R is a set ring.
First, (x y) = (x) (y):
P (x y) x y / P x / P and y / P P (x) (y) .
Secondly, (x y) = (x) (y):
P (x y) x y / P x / P or y / P P (x) (y) .
Claim. The mapping : L R is an isomorphism.
By its denition, is surjective, and by the rst claim it is a homomorphism.
For injectivity, let x ,= y. By Corollary 3.21, there exists a prime ideal P that
separates x and y, and thus (x) ,= (y).
Geometric lattices
Recall that for a modular lattice L, the intervals [x y, x]
L
and [y, x y]
L
are
isomorphic, and so are the intervals [x y, y]
L
and [x, x y]
L
. We say that a
lattice L is semimodular, if
(S) x y x = y x y ,
for all elements x, y L.
Example 3.23. The lattice on the right is the smallest
semimodular lattice that is not modular. Notice that
the elements 0, x, y, z and 1 form an N
5
which is forbid-
den in a modular lattice. By Theorem 3.14, all modular
lattices are also semimodular.
0
a
b
x
z
y
1
Let L be a lattice that has the bottom element 0
L
. Then an element a L
is an atom, if 0
L
a. Let Atom(L) be the set of the atoms of a lattice L. We
say that L is
an atomic lattice if for all x L with x ,= 0
L
, there exists an atom a such
that a
L
x.
a point lattice if the atoms generate L, that is, for all x L with x ,= 0
L
,
x =
a L [ a
L
x, a an atom .
a geometric lattice if it is semimodular point lattice that satises the nite
chain condition (FCC).
Example 3.24. Every nite lattice is atomic. The atoms of the lattice L in
Example 3.23 are a and x. However, L is not a point lattice, since, for instance,
the element z is not a join of atoms.
X
of a set X is always a point lattice. Its
atoms are the singleton sets x, and every set A X is the join
_
xA
x. In
2
X
we have A B if and only if [A B[ = 1, and A B = A B A implies
that B AB = AB. Thus 2
X
is semimodular. If X is innite, 2
X
does not
satisfy the FCC. Every nite lattice does satisfy FCC, and so for nite sets X,
2
X
is geometric.
Lemma 3.26. Every geometric lattice is complete.
Proof. If A = x
1
, x
2
, . . . L is innite, then
x
1
, x
1
x
2
, . . . , x
1
x
1
. . . x
n
, . . .
forms an ascending chain, which by FCC can contain only nitely many terms.
3.2. ALGEBRAIC LATTICES 59

Lemma 3.27. Let L be a geometric lattice. If a Atom(L) and x L are such
that a
L
x then x x a.
Proof. Since a
L
x, we have x a <
L
a and, since a is an atom, it follows that
x a = 0
L
a. From the condition (S) we obtain that x x a which proves
the present claim.
Theorem 3.28 (Exchange law). Let L be a geometric lattice, and a, b Atom(L)
and x L. Then
a
L
x and a
L
x b = b
L
x a .
Proof. Let a
L
x b and a
L
x. Now a
L
x implies that x x a by
Lemma 3.27. If b
L
x then b
L
x a.
Let us assume thus that b
L
x. In this case, by
Lemma 3.27, x x b. On the other hand, a
L
x b
and therefore also x a
L
x b, and there must be
equality here. Consequently,
x x a = x b
which shows that b
L
x a.
x
x a x b
3.2. Algebraic lattices
Compact elements
An element x of a complete lattice L is called compact if whenever x
L
_
A
for a set A L then there exists a nite subset F A such that x
L
_
F.
Theorem 3.29. Let L be a complete lattice. Then all elements of L are compact
if and only if L satises the ascending chain condition.
Proof. Suppose that the ACC holds for L, and thus every nonempty subset of L
has a maximal element; see Theorem 1.36. Let x
L
_
A for a subset A L. The
set
_
F [ F A, [F[ < has a maximal element, say
_
M. Since y
L
_
M
for all y A, also
_
M =
_
A and hence x
L
_
M as was required.
Assume then that all elements of L are compact, and let A L. Now
_
A
L
_
A, and hence there exists a nite subset F A such that
_
A
L
_
F.
Each nite set has a maximal element, and such an element is also maximal in A.
This shows that every subset has a maximal element.
Lemma 3.30. If x and y are compact elements in L, then so is their join x y.
Proof. Exercise.
Example 3.31. The meet of two compact elements need not
be compact. Consider the poset on the right where between
a and b there are innite chain of elements a z
1
z
2
. . .
with z
i

L
b for all i. Then both x and y are compact, but
b = x y is not. Indeed, let Z = z
i
[ i = 1, 2, . . . . Then
b
L
_
Z, but if F Z is a nite subset, then
_
F
L
z
m
for the maximum element in F.
A complete L is algebraic (or compactly generated) if all
its elements are joins of compact elements, that is, if x L
then there exists a set A of compact elements of L such that
x =
_
A. In particular, in an algebraic lattice
a
z
1
z
2
.
.
.
b
x
y
x =
c [ c
L
x, c compact
for all x L.
X
is algebraic. The compact elements of
2
X
are the nite subsets of X, and each subset A X is the union of the nite
subsets of A.
Example 3.33. The subgroup lattice of a group G is a compact lattice. The
compact elements in this lattice are the nitely generated subgroups.
Theorem 3.34. Let L be an algebraic lattice, and let x <
L
y. Then there exists
a compact element c such that c
L
y but c
L
x.
Proof. Let A
c
be a set of compact elements such that y =
_
A
c
. If all c A
c
were such that c
L
x then
_
A
c

L
x would imply that x = y; a contradiction.
Thus at least one c A
c
satises c
L
x.
We say that a subset D of a lattice L is (upward) directed if
x, y D = z D: x y
L
z .
It follows that if D is a directed subset and F D is a nite subset, then there
exists an element z D such that
_
F
L
z.
Theorem 3.35 (Generalized distributive law). Let L be an algebraic lattice and
D its directed set. Then, for all x L,
x
D =
yD
(x y) .
Proof. Clearly,
yD
(x y)
L
x and
yD
(x y)
L
D,
and hence
yD
(x y)
L
x
D.
In converse, let c be a compact element such that c
L
x
_
D. Then c
L
x
and c
L
_
D. Let F D be a nite subset such that c
L
_
F. Since F is nite
and D is directed, there exists an element a D such that
_
F
L
a. Therefore
(3.1) c
L
x a
L
yD
(x y) .
Hence all compact c
L
x
_
D satisfy c
L
_
yD
(x y). By Theorem 3.34,
these elements must be equal.
Suppose, moreover, that L is distributive in the above, and let
_
F =
_
n
i=1
x
i
for some x
i
. Then x
_
F
(D1)
=
_
n
i=1
(x x
i
), and (3.1) can be rewritten in the
form
c
L
x
F
L
yD
(x y) .
where one did not need the fact that D is directed. Therefore we have shown
Theorem 3.36. Let L be a distributive and algebraic lattice. Then
x
A =
yA
(x y)
for all A L and x L.
Closure operations
Let X be a set. A mapping C: 2
X
2
X
is a closure operation if for all
A, B X,
A C(A) , (C1)
C
2
(A) = C(A) , (C2)
A B = C(A) C(B) . (C3)
If C: 2
X
2
X
is a closure operation then C(A) is the closure of the subset
A X. We say that a subset A X is closed w.r.t. the closure operation C if
A = C(A). The condition (C1) is extension condition; (C2) is idempotency
condition; (C3) is isotonicity condition.
Example 3.37. The operations C that maps a subset of an algebra to its gen-
erated subalgebra is a closure operation. For a group/lattice, C maps a subset A
to the subgroup/sublattice Sg(A).
Example 3.38. The topological closure operations satisfy the above conditions.
Moreover, the topological closure concerns union which is not required in the
general case.
Example 3.39. The operation that adjoins the smallest equivalence relation to
a relation is a closure operation.
Lemma 3.40. A mapping C: 2
X
2
X
is a closure operation if and only if it
satises the conditions (C1) and
(C2+3) A C(B) = C(A) C(B) .
Proof. Exercise.
Lemma 3.41. Let C: 2
X
2
X
be a closure operation, and A
i
[ i I a family
of sets. Then
_
iI
C(A
i
) C
_
_
iI
A
i
_
.
Proof. Since U = C
_
iI
A
i
_
is closed and A
i
U for all i, also C(A
i
) U for
all i, and thus the claim follows.
The closed subsets form a poset Clo(C) under inclusion. Such a poset is
called a closure system.
Lemma 3.42. Let C be a closure operation and let A
i
[ i I Clo(C) be a
family of closed sets. Then also the intersection
iI
A
i
is closed.
Proof. Write A =
iI
A
i
(=
A). Now, A A
i
for all i, and hence by (C3)
C(A) C(A
i
) = A
i
for all i I. Therefore, C(A) A C(A), and so C(A) = A, which proves the
claim.
By Theorem 2.13, we have
Theorem 3.43. Let C be a closure operation. Then Clo(C) is a complete lattice
with
iI
C(A
i
) =
iI
C(A
i
) and
iI
C(A
i
) = C
_
_
iI
A
i
_
.
Also the converse holds:
Theorem 3.44. Let L be a complete lattice. Then L is isomorphic to a closure
system Clo(C) for some closure operation C.
Proof. Dene
C(A) = (
A] .
Then, as is easy to show, C: 2
L
2
L
is a closure operation, and a subset is
closed if and only if its is an ideal. (Notice that in a complete lattice every ideal
is principal. Indeed, I = (
_
I].) By (the proof of) Theorem 2.46 the mapping
: L Clo(C) dened by (x) = (x] is a required isomorphism.
A closure operation C: 2
X
2
X
is algebraic if for all subsets A X,
(C4) C(A) =
_
C(F) [ F A, [F[ < .
Lemma 3.45. A closure operation C: 2
X
2
X
is algebraic if and only if it is
locally nite: if x C(A) then there exists a nite subset F A such that
x C(F).
Proof. If C is algebraic and x C(A) then the claim follows from the condition
(C4), since C(A) is a union of closed sets generated by nite sets. On the other
hand, if C is locally nite then the claim follows from the denition.
The next lemma is a reformulation of the denition.
Lemma 3.46. Let C: 2
X
2
X
be an algebraic closure operation. Then
C(A) =
C(F) [ F A, [F[ <

for all A X.
Proof. Indeed,
C(F) [ F A, [F[ < = C

_
_
FA
|F|<
C(F)
_
=
_
FA
|F|<
C(F) .
Example 3.47. Let G be a group and Sg(A) be the subgroup generated by A.

Then Sg: 2
G
2
G
is an algebraic closure operation. Indeed, g Sg(A) if and
only if it is a nite product g = a
e
1
1
a
e
2
2
. . . a
e
n
n
of elements, where a
i
A and
e
i
= 1. Therefore g Sg(a
1
, a
2
, . . . , g
n
).
Similarly, if R is a ring, then the operation I : 2
R
2
R
that maps a subset
A to the ideal I(A) generated by A is an algebraic closure operation.
Theorem 3.48. Let C: 2
X
2
X
be an algebraic closure operation. Then the
closure system Clo(C) is an algebraic lattice. The compact elements of Clo(C)
are the closed sets C(F) for nite subsets F X.
Proof. We show rst that a closed set A is compact if and only if there exists a
nite set F such that A = C(F).
Let F X be a nite subset, and suppose that C(F)
_
iI
A
i
, where
each A
i
is a closed set, C(A
i
) = A
i
for all i I. Denote U =
iI
A
i
. Then
C(U) =
_
C, and since C is algebraic,
C(F) C(U) =
_
C(B) [ [B[ < , B U
=
_
C(B) [ [B[ < , B
_
iJ
A
i
, [J[ < .
Since F is nite and F C(F), there are nite sets B
1
, B
2
, . . . , B
k
U such
that F
k
i=1
C(B
i
). Each B
i
is nite, and so
B
i

_
jI
i
A
j
for some index set [I
i
[ <
for all i = 1, 2, . . . , k. Therefore
F
k
_
i=1
C(B
i
)
k
i=1
C
_
_
jI
i
A
j
_
=
C(A
j
) [ j I
i
, i = 1, 2, . . . , k .
We have now that C(F)
_
jJ
C(A
j
) for the nite index set
J =
k
_
i=1
I
i
.
Thus C(F) is compact.
Suppose then that the closed subset A = C(A) is compact. Now
C(A) =
_
C(B) [ [B[ < , B A
=
C(B) [ [B[ < , B A
i=1
C(B
i
) ,
for some nite B
1
, B
2
, . . . , B
k
A, by compactness of A. Moreover,
C(A)
k
i=1
C(B
i
) = C
_
k
_
i=1
B
i
_
C(A) ,
where there must be equalities. The claim follows since
k
i=1
B
i
is nite.
That the closure system Clo(C) is algebraic follows directly from the fact that
C is algebraic, that is, from the condition (C4).
We say that a subset A generates a closed set Y if C(A) = Y . Moreover, Y
is said to be nitely generated if there exists a nite set F such that C(F) = Y .
We leave the following corollaries as exercises.
Corollary 3.49. Let C: 2
X
2
X
be an algebraic closure operation. Then the
nitely generated subsets of X form the compact elements of the lattice Clo(C).
Corollary 3.50. Every algebraic lattice is isomorphic to some closure system
Clo(C) where C is an algebraic closure operation.
Recall that in a directed family of sets D for every two sets A, B A there
exists a set C D such that A B C.
Theorem 3.51. Let C: 2
X
2
X
be a closure operation. Then C is algebraic if
and only if for every directed family A = A
i
[ i I of subsets of X,
C
_
_
iI
A
i
_
=
_
iI
C(A
i
) .
Proof. Assume that C is algebraic and let A
i
[ i I be a directed family. Let
U =
iI
A
i
. By (C4),
C(U) =
_
C(F) [ F U, [F[ < .
Every nite subset F U is included in some element A
F
of the directed fam-
ily A
i
[ i I. Now, C(F) C(A
F
), and hence
C(U) =
_
C(F) [ F U, [F[ <
_
iI
C(A
i
) .
On the other hand, by Lemma 3.41, the inclusion is always valid in the other
direction.
Conversely, suppose that C satises the condition for directed families. For
each subset A, let
M(A) = C(F) [ F A, [F[ < .
Then
M(A) C(A). The family M(A) is directed, and hence
M(A) is closed.
On the other hand, A
M(A) and therefore

C(A) C(
_
M(A)) =
_
M(A) ,
which was the claim.
Example 3.52. [Exchange law] Theorem 3.28 can be reformulated using closure
operations as follows. Let L be a geometric lattice L, and dene
C(A) = a Atom(L) [ a
L
A
for all A Atom(L). Then C is a closure operation on the atoms of L that
satises the following exchange property:
a C(A b) with a / C(A) = b C(A a) .
DedekindMacNeille completions
Completions of posets were motivated by Dedekinds method to construct the real
numbers from rational numbers. This approach was generalized by MacNeille for
arbitrary posets.
Let P be a poset, and L a complete lattice. An order embedding : P L
is called a completion of P.
By Theorem 1.23, every poset P is isomorphic to the poset of its principal
down-sets under inclusion. Hence the mapping : P 2
P
dened by (x) = x
is an order embedding of P into the complete lattice 2
P
(under inclusion). The
disadvantage of this completion is in the large size of the power set 2
P
compared
to the size of P.
Recall that, for a subset A P, let
A
l
= x P [ x
P
A and A
u
= x P [ x
P
A
be the sets of lower and upper bounds in the poset P.
Lemma 3.53. The following hold for subsets A, B P:
A A
ul
and A A
lu
, (3.2)
if A B then B
u
A
u
and B
l
A
l
, (3.3)
A
l
= A
lul
and A
u
= A
ulu
. (3.4)
Also, A
l
is a down-set and A
u
is an up-set.
Proof. Exercise.
The DedekindMacNeille completion of a poset P is dened to be
DM(P) = A [ A P, A = A
ul
together with inclusion. Here x A

ul
if and only if it is a lower bound of all
upper bounds of A.
Example 3.54. Let P = (N, ). If A is nite, then A
u
= n [ n max(A), and
DM(A) = 0, 1, . . . , max(A). On the other hand, if A is innite, then A
u
= ,
and then DM(A) = N. Also, DM() = 0.
Theorem 3.55. Let P be a poset. Then DM(P) is a complete lattice.
Proof. First of all, DM(P) has the minimum element
ul
. By Theorem 2.12, it
suces to show that each nonempty subset A of DM(P) has a least upper bound.
Let
A = A
i
[ i I and A =
_
iI
A
i
.
By Lemma 3.53, we have that A
ul
DM(P). Also, A
ul
is an upper bound of A
in DM(P). Consider then a subset B DM(P) that is also an upper bound
of A. Then A B, and therefore, by Lemma 3.53, B
u
A
u
, and, furthermore,
A
ul
B
ul
. Hence A
ul
is the least upper bound of A, and this proves that DM(P)
is complete.
Lemma 3.56. Let x P for the ordered set P. Then x DM(P).
Proof. One needs to show that x = (x)
ul
. This is an exercise.
Lemma 3.57. Let P be an ordered set and A P such that
_
A exists in P.
Then A
ul
= (
_
A)
l
.
Proof. Exercise.
Theorem 3.58. Let L be a complete lattice. Then L and DM(L) are isomorphic.
Proof. By Theorem 3.55, DM(L) is a complete lattice. Dene the mapping
: L DM(L) by
(3.5) (x) = x.
(Notice that x = x
l
for all elements x L.) Let A L. Since L is complete,
_
A exists. Using Lemma 3.57, we obtain
A
ul
=
_
A
_
l
=
_
A
_
=
_
A
_
.
It is clear that A
ul
is an upper bound in DM(L) for (x) [ x A. Let
B DM(L) be another upper bound. Since x (x) B for all x A, we
have A B, and thus A
ul
B
ul
= B. This shows that A
ul
=
_
(A), and
so
_ _
A
_
=
_
(A) as required. Dual argumentation shows that preserves
greatest lower bounds.
Also, A
ul
=
_ _
A
_
=
_ _
A
_
, and hence is surjective. Surely it is injective,
and thus it is an isomorphism.
3.3. COMPLEMENTED LATTICES 68
3.3. Complemented lattices
Complements
Let L be a lattice with the bottom and top elements 0
L
and 1
L
. An element
x L is a complement of x L if
(C) x x = 0
L
and x x = 1
L
.
The lattice L is complemented if each of its elements has a complement. Clearly,
if x has a complement x then x is a complement of x.
Example 3.59. The lattice L on the right is comple-
mented. We always have that 0
L
and 1
L
are comple-
ments of each other. Notice that an element may have
several complements. For instance, in the given lattice,
x has the complements a, b, c and d.
0
a
x
c
b d
1
X
is complemented in a natural way. On
the other hand, a divisor lattice T
n
need not be complemented. For instance, the
integer 2 does not have a complement in L = T
12
: We do have that 2 3 = 1 (=
0
L
), but 2 3 = 6 (, = 1
L
).
A lattice L (which possibly does not have a bottom nor a top element) is rel-
atively complemented, if every interval [x, y]
L
with x
L
y is a complemented
sublattice.
Example 3.61. It is clear that every relatively com-
plemented lattice is complemented if it has a top and a
bottom element. The converse is not true as shown in
the lattice L on the right. This lattice is complemented
but the sublattice [b, d]
L
is not complemented.
0
a
b
c
d
1
Theorem 3.62. Let L be a distributive lattice with 0
L
and 1
L
. Then every
element can have at most one complement in L.
Proof. Assume both y and z are complements of x L. Then
y = y 1
L
= y (z x)
(D1)
= (y z) (y x)
= (y z) 0
L
= y z ,
and hence y
L
z. Symmetrically, we have z
L
y, and therefore y = z.
Example 3.63. A point lattice (where all elements are
joins of atoms) need not be complemented. The element
x in the point lattice L on the right has no complement.
0
a
b
x
c
d
1
Theorem 3.64. Each geometric lattice is relatively complemented, and thus also
complemented.
Proof. Recall that every geometric lattice L is complete, and thus it has the
bottom and the top elements.
Let x
L
y for x, y L, and let z [x, y]
L
and choose an element z
1
[x, y]
such that z z
1
= x. (Such an elements exists, e.g., z
1
= x.) If z z
1
= y, we
are done: z
1
is a complement of z in [x, y]
L
.
Suppose thus that z z
1
<
L
y. Since each geometric lattice is a point lattice,
there exists an atom a L such that a
L
y and a
L
z z
1
. (Since y is the
join of the atoms b
L
y and z z
1
is the join of the atoms b
L
z z
1
.) Let
z
2
= z
1
a. Clearly, z
2
[x, y]
L
and so z z
2

L
y. Moreover,
z z
1
<
L
z z
2
.
We show now that z z
2
= x.
Suppose contrary to this claim that x <
L
z z
2
. Then there exists an atom
b of L such that b
L
z z
2
but b
L
x. We have
x = z z
1
<
L
(z z
1
) b
L
z z
2
= z (z
1
a) .
Since b
L
z and b
L
x = z z
1
, it follows that b
L
z
1
. Also, b
L
z
2
= z
1
a,
and Lemma 3.28 gives that a
L
z
1
b, and so z
1
a
L
z
1
b. Also z
1
b
L
z
1
a,
since b
L
z
1
a, and hence z
1
b = z
1
a = z
2
, which then implies that
z z
1
b = z z
2
.
However, b
L
z and b
L
z
1
a, and hence z z
2
= z z
1
b = z z
1
, which is
a contradiction. Thus, indeed, z z
2
= x.
Now we have found an element z
2
for which
z z
2
= x and z z
1
<
L
z z
2

L
y .
Again, if z z
2
= y, we have found a complement z
2
of z. Otherwise we repeat
the above process for z
2
, and obtain in this fashion a sequence z
1
, z
2
, . . . such
that
z z
i
= x and z z
i1
<
L
z z
i

L
y .
Since every chain in L is nite, the above construction will end in an element z
i
for which z z
i
= y, and thus z
i
is a complement of z in the interval [x, y]
L
.
Boolean lattices
A complemented distributive lattice B with 0
B
,= 1
B
is called Boolean lattice.
Example 3.65. The 2-element chain C
2
, also often de-
noted by B
2
, is a Boolean lattice, and so are the direct
products B
2
n of B
2
that have 2
n
elements.
0
x
x
1
Example 3.66. A nonempty family K of subsets of X is a set algebra
A, B K = A B, A B, X A K.
Every set eld is a Boolean lattice.
Theorem 3.67. Let B be a Boolean lattice. Then every element x B has a
unique complement x. Moreover,
(1) 0
B
= 1
B
and 1
B
= 0
B
;
(2) x = x for all x B;
(3) x y = x y and x y = x y;
(4) x
B
y y
B
x.
Proof. The uniqueness of complements follows from Theorem 3.62. The rest of
the cases are exercises.
In addition to the lattice laws (L1) (L4), the Boolean lattices B satisfy the
distributive laws (D1) and (D2), and the law for complements.
Note that from these laws it can be derived that 0
B
is the bottom element of
B and 1
B
is the top element of B. Indeed, for all x B,
x 0
B
(C)
= x (x x)
(L3)
= (x x) x
(L1)
= x x
(C)
= 0
B
,
and similarly x 1
B
= 1
B
.
Example 3.68 (Huntington). The following three axioms determine Boolean
lattices, that is, they are equivalent to the previous set of axioms:
x (y z) = (x y) z (H1)
x y = y x (H2)
x y x y = x (H3)
The rst two of these are (L3) and (L2). The last one is the Huntington
equation. Note that these axioms do not use .
3.4. M
OBIUS FUNCTION FOR LATTICES 71

Example 3.69 (Robbins Conjecture). It was conjectured by Robbins that the
following three axioms determine Boolean lattices:
x (y z) = (x y) z (R1)
x y = y x (R2)
x y x y = x (R3)
The rst two are again (L3) and (L2). The last one is the Robbins equa-
tion. This conjecture was proved by McCune in 1997 by an extensive usage of
computers. Note that these axioms do not use .
Example 3.70 (Single axiom). The following single axiom determines the Boolean
lattices:
(3.6) x y z u u z x = z .
This is due to McCune, Vero, Fitelson, Harris, Feist and Wos (2000).
In the nite case we have the following characterizations of Boolean lattices.
Theorem 3.71. Let L be a nite distributive lattice. Then the following condi-
tions are equivalent:
(1) L is a boolean algebra,
(2) L is complemented,
(3) L is relatively complemented,
(4) L is atomic,
(5) 1
L
is the join of the atoms of L,
(6) L is geometric.
Proof. Exercise.
3.4. Mobius function for lattices
Theorems of Hall and Weisner
Recall that the Mobius function of a locally nite poset P is the inverse of the
zeta function:
(x, y) =
_
1 if x
L
y ,
0 otherwise .
Then
(3.7) (x, y) =
_
_
0 if x y ,
1 if x = y ,

x
P
z<
P
y
(x, z) if x <
L
y .
In this section we study the Mobius function for lattices.
3.4. M

The following tool is needed in the proof of Theorem 3.73. It states that each
nonempty nite set X has equally many subsets of even size and of odd size.
Lemma 3.72. Let X be a nonempty nite set. Then
AX
(1)
|A|
= 0 .
Proof. Let n = [X[ 1. For each i = 0, 1, . . . , n there are
_
n
i
_
subsets A X
such that [A[ = i. Hence
AX
(1)
|A|
=
n
i=0
(1)
i
_
n
i
_
=
n
i=0
(1)
i
1
ni
_
n
i
_
= (1 1)
n
= 0
by the binomial theorem.
Second proof. The second proof goes by induction on [X[. The claim is true
for singleton sets, since is has even size and X has odd size. Let then x X,
where [X[ = n + 1. By the induction hypothesis X x there are equally many
even and odd subsets of X that do not contain x. To obtain all subsets, add x
to the previous subsets; even subsets turn to odd, and odd to even, and so there
are equally many of subsets of even and odd elements that contain x.
Theorem 3.73 (Hall). Let L be a nite lattice. Then
(0, x) =
AY (x)
(1)
|A|
,
where
Y (x) = A Atom(L) [ x =
A .
Proof. Let x L, and denote (x) =
AY (x)
(1)
|A|
. Let also A(x) = a
Atom(L) [ a
L
x, and (x) =
y
L
x
(y). Then
(x) =
y
L
x
(y) =
AA(x)
(1)
|A|
=
_
1 if A(x) = ,
0 if A(x) ,= ,
=
_
1 if x = 0 ,
0 if x ,= 0 .
By the Mobius inversion formula, we have the claim:
(x) =
y
L
x
(y)(y, x) = (0, x) .
3.4. M

Corollary 3.74. Let L be a nite lattice. If x L is not a join of atoms, then
(0, x) = 0.
Applying the previous corollary to the sublattice [x, y]
L
we obtain
Corollary 3.75. Let L be a nite lattice, and let A
x
= z [ x z. Then
(x, y) = 0 if y is not a join of elements from A
z
.
Theorem 3.76 (Weisner). Let L be a nite lattice, and let x L and a L with
a ,= 0. Then
za=x
(0, z) = 0 .
Proof. Let (x) =
za=x
(0, z), and let (x) =
y
L
x
(y). Then
(x) =
y
L
x
(y) =
za
L
x
(0, z) .
If a
L
x then obviously (x) = 0. Suppose that a
L
x. Since a ,= 0, (x) =
z
L
x
(0, z) = 0.
Closure operations
We generalize the notion of a closure operation to all lattices as follows. A
function c: L L is a closure operation if, for all x, y L,
x
L
c(x) ,
x
L
y = c(x)
L
c(y) ,
c(x) = c(c(x)) .
The set of all closed elements, i.e., xed points of c, is denoted by
Clo(c) = x [ c(x) = x .
Example 3.77. Let a be a xed element of the lattice L. Then the function
c: L L dened by c(x) = c a is a closure operation.
Lemma 3.78. Let L be a lattice and c: L L a closure operation on L. Then
Clo(c) is a complete lattice under inclusion.
Proof. It is easy to show that Clo(c) is closed under arbitrary intersections, and
thus the claim follows from Theorem 2.13.
Theorem 3.79 (Rota). Let L be a locally nite lattice and c: L L a closure
operation on L, and let K = Clo(c) be the lattice of closed elements for c. Then
LITERATURE 74
for all x, y L,
zL
c(z)=c(y)
(x, z) =
_
Q
(c(x), c(y)) if x = c(x) ,
0 if x <
L
c(x) .
Proof. For brevity we write x instead of c(x) for all x. Also and refer to the
lattice L, and
K
,
K
and
K
refer to the lattice K. Then
zL
z=y
(x, z) =
zL
(x, z)
K
(z, y)
=
zL
(x, z)
_
uK
K
(z, u)
K
(u, y)
_
=
zL
uK
(x, z)
K
(z, u)
K
(u, y)
()
=
zL
uK
(x, z)(z, u)
K
(u, y)
=
uK
_
zL
(x, z)(z, u)
_
K
(u, y) =
uK
(x, u)
K
(u, y) ,
where in (*) we used the fact that z
L
u if and only if z
K
u.
Literature
[1] Aigner, M, Combinatorial Theory, Springer-Verlag, 1997.
[2] Birkho, G., Lattice Theory, Amer. Math. Sco., 1948
[3] Crawley, P. and R.P. Dilworth, Algebraic Theory of Lattice, Prentice-Hall,
1973.
[4] Davey, B.A. and H.A. Priestley, Introduction to Lattices and Order, Cam-
bridge Univ. Press, 2002 (2nd edition).
[5] Gratzer, G., Lattice Theory, Freeman, 1971.
[6] Gratzer, G., General Lattice Theory, Academic Press, 1978.
[7] Gratzer, G., Universal Algebra, Springer-Verlag, 1979.
[8] Rubin, H. and J. Rubin, Equivalents of the Axiom of Choice, II. Amster-
dam: North-Holland, 1985.
[9] Schroder, Ordered Sets: An Introduction, Birkhauser, 1997.
[10] Stanley, R., Enumerative Combinatorics, Cambride Uni. Press, 1997.
Index
ACC, 16
algebraic
closure, 63
lattice, 60
alphabet, 8
alternating cycle, 22
antichain, 6
antisymmetric, 3
atom, 58
atomic lattice, 58
automorphism, 37
bad sequence, 19
block, 2
Boolean lattice, 70
bottom element, 5
chain, 3, 6
chain complete, 48
chain condition, 16
choice function, 14
closure, 61
operation, 61, 73
system, 62
comparable, 4
complement
element, 68
complemented lattice, 68
complete lattice, 34
completion, 66
composition, 3
congruence, 40
connected, 4
continuous function, 46
convex subset, 40
convolution, 26
cover, 5
cover chain, 54
crown, 22
DCC, 16
Dedekind-MacNeille completion, 66
delta nction, 27
diamond M
5
, 50
dimension, 21
direct product, 11
directed subset, 60
distributive, 50
divisibility poset, 8
divisor poset, 8
down-set, 7
dual poset, 12
embedding, 44
equivalence, 3
exchange property, 66
extension, 20
extension condition, 61
FAC, 16
factor, 8
factor poset, 8
FCC, 16
lter, 46
xed point, 46
generated sublattice, 39
geometric lattice, 58
graded poset, 17
greatest lower bound, 7
Halls condition, 25
Hasse diagram, 5
height, 23
homomorphism, 37
Huntington equation, 70
ideal, 43
lattice, 44
idempotency condition, 61
75
INDEX 76
incidence algebra, 26
incomparable, 4
increasing mapping, 48
inmum, 7
interval, 4
inverse relation, 3
isomorphism, 37
isotone, 9
isotonicity condition, 61
join, 32
join preserving, 37
JordanDedekind theorem, 54
kernel, 4
Kronecker function, 27
lattice, 32
least upper bound, 7
lexicographic order, 9
linear extension, 20
linear order, 3
locally nite, 4
closure, 63
lower bound, 7
Mobius function, 27
maximal
antichain, 6
chain, 6
element, 5
maximum element, 5
meet, 32
meet preserving, 37
minimal element, 5
minimum lement, 5
minor relation, 20
modular lattice, 51
normal subgroup, 53
order
embedding, 9
isomorphism, 9
preserving, 9
partial order, 3
partially well-ordered, 18
partition, 2
partition order, 8
pentagon N
5
, 50
point lattice, 58
poset, 4
prime
lter, 46
ideal, 45
principal
down-set, 7
ideal, 43
up-set, 7
proper
lter, 46
ideal, 43
quaotient lattice, 42
quasi-order, 3
rank function, 17
reexive, 3
relatively complemented, 68
Robbins equation, 71
semimodular lattice, 58
set algebra, 70
set ring, 57
strict partial order, 4
sublattice, 38
subposet, 6
subsequence order, 9, 18
subset poset, 6
super-antichain, 13
supremum, 7
symmetric, 3
top element, 5
transitive, 3
transitive closure, 3
up-set, 7
upper bound, 7
well-order, 15
width, 23
zeta function, 27

Lecture Notes in Ordered Sets by DR Tero Harju

Uploaded by

Copyright:

Available Formats

Lecture Notes in Ordered Sets by DR Tero Harju

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lecture Notes in Ordered Sets by DR Tero Harju

Uploaded by

Copyright:

Available Formats

Lecture Notes in

y. Of course, it can be that z = z

. A word u is a factor of v, denoted u v if v = v

, ) is a poset, the factor poset of

of all words over P. The

is a poset. Also, if P is linearly

. Note that although P can be nite, P

is always innite, and

is a poset under the subsequence

is another maximal chain,

. The subsequence order on P

is a well-founded poset under the relation

1.3. Extensions and dimension

is a chain in the poset P. By Theorem 1.57, P

From Theorem 1.65 we have

is not a lattice. Indeed, for instance, the incom-

Theorem 2.33. Let

01, 23, 45 02, 13, 4, 5 24, 35, 0, 1

The congruence classes x, x L, partition the lattice L. The quotient

F for some nite F X

Example 3.11. It can be shown that a lattice L is distributive if and only if it

C. Then M is an ideal of L, I M and M F = , that is, M S.

Corollary 3.21. Let L be a distributive lattice L, and let x, y L be dierent

3.2. ALGEBRAIC LATTICES 59

C(F) [ F A, [F[ <

C(F) [ F A, [F[ < = C

Example 3.47. Let G be a group and Sg(A) be the subgroup generated by A.

C(B) [ [B[ < , B A

M(A) C(A). The family M(A) is directed, and hence

M(A) and therefore

together with inclusion. Here x A

OBIUS FUNCTION FOR LATTICES 71

OBIUS FUNCTION FOR LATTICES 72

OBIUS FUNCTION FOR LATTICES 73

You might also like