53 views

Uploaded by VitorBRz

save

- 10.1.1.91
- Application of Fuzzy Algebra in Automata theory
- MOebius Function Dense Subgraphs
- Answers to Questions Asked by Students in Math 104
- Schneider--Medium-Voltage-Equipment-Design-Guide.pdf
- Curso de linguistica general 2007.pdf
- 9_-5-_The_effect_of_mirror_therapy_on_balance_ability_of_subacute_stroke_patients_-Bella_Syafitri-.pdf
- Doc1.docx
- R$ 350,00 POR TCC OU MONOGRAFIA WHATSAPP (21) 3942-6556 tccmonografia44@gmail.com (18).pdf
- Burj Khalifa
- Comunicación de una Organización
- sleep.docx
- esrfh.docx
- Norma Tecnica de Metrados
- 2017년 12월 온라인쇼핑 동향 (1)
- Kodak
- B.E Regulation 2017 Data Structure Lab Manual
- Coseme Beret
- Padrões e processos ecológicos e evolutivos em escala regional - 2009.pdf
- Easy Play Rock Internacional
- Ambiental II
- investment law
- 14_15_forms
- bon-jovi-always.pdf
- ANALISIS CHAKRA AURA ANDA.doc
- 352165033-triangulos-del-cuello-pdf.pdf
- 4891-3166-1-PB
- 3197Ramon.pdf
- 7933-25093-1-PB
- teorias pedagogicas .pdf
- Roteiro Aula Pratica5lkkkkkkkkmmmmmmmmmmbg534
- Engineering IP Topic Proposal
- Alcaldia Libro Cbba 7
- Penilaian
- Programming in Stardard ML
- Pierre Boulez's Structures
- The Concept of Time Single Pages
- H55M-LE

You are on page 1of 21

Johan G. Belinfante

Revised 1995 May 21

Introduction. These notes were originally prepared in July 1972 as a handout for a class

in modern algebra taught at the Carnegie-Mellon University in Pittsburgh, Pennsylvania.

A proof of Zorn’s lemma from the axiom of choice is given along the lines presented by

Paul R. Halmos in his book Naive Set Theory and by Kenneth S. Miller in the appendix

to his book Elements of Modern Abstract Algebra. The pretty theorem on the connection

between comparability and covering is original. The notes were revised in October 1984

and were used again in classes taught at the Georgia Institute of Technology in Atlanta,

Georgia. The original notes were handwritten because of the many ﬁgures and specialized

symbols that were required. In March 1993 the notes were typeset using PCT

E

X and

P

I

CT

E

X.

Partial Orderings. A relation ≤ on a class P is a partial ordering if it is reﬂexive,

symmetric and transitive:

Reﬂexive: (∀x ∈ P) x ≤ x,

Antisymmetric: (∀x, y ∈ P) x ≤ y & y ≤ x =⇒ x = y,

Transitive: (∀x, y, z ∈ P) x ≤ y & y ≤ z =⇒ x ≤ z.

A poset (P, ≤) consists of a set P and a partial ordering ≤ on P. We deﬁne x < y to mean

x ≤ y & x = y, and we deﬁne a ≥ b to mean b ≤ a. Note that ≥ is also a partial ordering

relation, but < is not. Although the transitive law does hold for <, neither the reﬂexive

law nor the antisymmetric law is true.

Bibliography. A slightly diﬀerent approach for the axiomatics is used in the following

reference.

N. Bourbaki, Theory of Sets, Chapter III.

Examples of Posets. Many examples of posets are already familiar.

Example. A trivial example of a poset is (S, =), where S is any set. (Any set can be

partially ordered by equality.)

Example. Any subset of the real line is partially ordered by the usual ≤ relation.

Example. A given set may be partially ordered in more than one way. For example, any

set P of non-negative integers is partially ordered not only by the usual ≤ relation, but

also by the divisibility relation [. If a, b ∈ P, we write a [ b if and only if there exists an

integer m such that b = ma.

Example. Any collection ( of sets can be partially ordered by inclusion. That is, if ( is

a set whose elements are sets, then ((, ⊂) is a poset. A simple case is the power set of a

set: ( = PowS = ¦A [ A ⊂ S¦.

Example. The dual of a poset (P, ≤) is the poset (P, ≥).

1

Example. If L is any collection of statements, then (L, =⇒) satisﬁes the reﬂexive and

transitive properties, but may fail to satisfy the antisymmetric law. Two distinct state-

ments A, B ∈ L may be logically equivalent: A =⇒ B and B =⇒ A.

Chains and Antichains. Elements x, y ∈ P of a poset (P, ≤) are comparable, written

x

................

................

................

................

y, if x ≤ y or y ≤ x.

A subset C of a poset (P, ≤) is a chain if any pair of elements in C are comparable:

(∀x, y ∈ C) x

................

................

................

................

y.

Chains are also called totally ordered subsets.*

A subset A of a poset (P, ≤) is an antichain if no two distinct elements of A are

comparable:

(∀x, y ∈ C) x

................

................

................

................

y =⇒ x = y.

Example. In any poset (P, ≤), the empty set ∅ and every singleton ¦x¦ is both a chain

and an antichain.

Example. Any subset of the real line (RRR, ≤), with the usual order, is a chain. In partic-

ular, ω = ¦0, 1, 2, . . .¦ and ZZ Z = ¦. . . , −2, −1, 0, 1, 2, . . .¦ are chains.

Proposition. Any subset of a chain is a chain.

Proof: This is clear. q.e.d.

Intervals and Half-Intervals. In any poset (P, ≤) we can deﬁne certain subsets which

we call intervals and half-intervals. If a, b ∈ P, then the following sets are called closed

half-intervals:

[a, ) = ¦x ∈ P [ a ≤ x¦,

( , b] = ¦x ∈ P [ x ≤ b¦.

The closed half-interval ( , b] is also called a weak initial segment. The sets

(a, ) = ¦x ∈ P [ a < x¦,

( , b) = ¦x ∈ P [ x < b¦.

are called open half-intervals.

Lemma. If (P, ≤) is any poset, then

(∀a, b ∈ P) a ≤ b =⇒ ( , a] ⊂ ( , b],

(∀a, b ∈ P) a = b =⇒ ( , a] = ( , b].

Proof: This is an easy exercise which uses all three of the poset axioms. q.e.d.

* The terms linearly ordered and completely ordered are used by some authors.

2

The set of all elements in a poset (P, ≤) which are comparable with a given element

x ∈ P is the union of two half-intervals:

¦y ∈ P [ x

................

................

................

................

y¦ = ( , x] ∪ [x, ).

There are four kinds of intervals, obtained by intersecting various types of half-

intervals. Sets of the form

[a, b] = [a, ) ∩ ( , b],

obtained by intersecting two closed half-intervals are called closed intervals. Similarly, sets

of the form

(a, b) = (a, ) ∩ ( , b),

obtained by intersecting two open half-intervals are called open intervals. Finally, we deﬁne

two types of half-open intervals

[a, b) = [a, ) ∩ ( , b),

(a, b] = (a, ) ∩ ( , b],

by intersecting a closed half-interval and an open half-interval.

Lemma. For any poset (P, ≤),

(∀x ∈ P) [x, x] = ¦x¦.

Proof: Use the antisymmetric law. q.e.d.

Greatest and Least Elements. An element g ∈ S belonging to a subset S ⊂ P of a

poset (P, ≤) is a greatest element of S if

(∀s ∈ S) s ≤ g.

Similarly, l ∈ S is a least element of S if

(∀s ∈ S) l ≤ s.

A poset need not have any greatest or least element.*

Lemma. A subset S ⊂ P of a poset (P, ≤) can have at most one greatest element, and at

most one least element.

Proof: Use the antisymmetric law. q.e.d.

Note that g is the greatest element of S if g ∈ S and S ⊂ ( , g]. Similarly l is the least

element of S if l ∈ S and S ⊂ [l, ).

* Some authors prefer to use the terms top element or last element for what we call a greatest element.

Similarly, the terms bottom element or ﬁrst element are used for a least element.

3

Well-ordered Sets. A poset (W, ≤) is well-ordered if every nonempty subset has a least

element. Well-ordered posets are totally ordered:

Proposition. Every well-ordered poset is a chain.

Proof: If x, y ∈ W are elements of a well-ordered poset (W, ≤), then the set ¦x, y¦ has

a least element. Hence x ≤ y or y ≤ x. That is, any two elements of W are comparable.

q.e.d.

Example. Any ﬁnite chain is well-ordered. The natural numbers with the usual ordering

is a well-ordered set (ω, ≤).

The well-ordering of the natural numbers is equivalent to the principle of complete

induction. One can derive the principle of induction from the well-ordering property, and

conversely.

Not every inﬁnite chain is well-ordered.

Example. The set RR R of real numbers with the usual ordering is a chain, but not a well-

ordered set. The same is true for the set QQ Q of rational numbers.

For example, in both QQ Q and RR R, the half-interval (0, ) has no least element. There is no

least positive rational or real number: if p is positive, then so is p/2.

Monotone Mappings and Poset Isomorphisms. A mapping f: P

1

→ P

2

from a poset

(P

1

, ≤) to a poset (P

2

, ≤) is said to be monotone if

(∀x, y ∈ P

1

) x ≤ y =⇒ f(x) ≤ f(y).

Monotone mappings are also known as poset homomorphisms or simply poset morphisms.

It is of interest to investigate which concepts in the theory of partially ordered sets are

preserved by monotone mappings. Monotone mappings preserve greatest elements, and

chains, for example.

Proposition. If f: P

1

→ P

2

is a monotone mapping mapping from a poset (P

1

, ≤) to a

poset (P

2

, ≤), and if x ∈ S is the greatest element of some subset S ⊂ P

1

, then f(x) is the

greatest element of the subset

f[S] = ¦f(s) [ s ∈ S¦.

Proof: This is an easy exercise. q.e.d.

Proposition. The image of a chain under any monotone mapping is a chain.

Proof: This is clear. q.e.d.

A mapping f: P

1

→ P

2

from a poset (P

1

, ≤) to a poset (P

2

, ≤) is said to be a poset

isomorphism if f is one-to-one and onto, and both f and f

−1

are monotone. Posets (P

1

, ≤)

and (P

2

, ≤) are isomorphic if there exists an isomorphism f: P

1

→ P

2

.

4

Counterexample. A one-to-one and onto monotone mapping need not have a monotone

inverse.

Construction: Let (P

1

, ≤) and (P

2

, ≤) be the posets pictured below, with the mapping

f as indicated.

•

• •

a

b

c

(P

1

, ≤)

...........................................

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

................................................... . . . . . . . . . . .

. . . . . . . . . . .

f

•

•

•

f(a)

f(b)

f(c)

...................................................

...................................................

(P

2

, ≤)

The inverse f

−1

is not monotone since f(c) ≤ f(a) does not imply c ≤ a. q.e.f.*

Proposition. For any poset (P, ≤), the mapping x → ( , x] is a one-to-one monotone

mapping from the poset (P, ≤) to the poset (PowP, ⊂).

Proof: See the results of the section on Intervals and Half-Intervals. q.e.d.

Corollary. Any poset (P, ≤) is isomorphic to a poset whose members are sets, partially

ordered by inclusion.

Hasse Diagrams and the Covering Relation in a Poset. An element x ∈ P is

covered by an element y ∈ P in a poset (P, ≤), written x < y, if x is strictly less than y,

and there is no element strictly between x and y. That is, x <y if x < y and (x, y) = ∅.

Example. In the poset of natural numbers (ω, ≤), we have (∀n ∈ ω) n < n + 1. In

the poset of rational numbers (QQQ, ≤) and in the poset of real numbers (RRR, ≤), no element

covers any other.

A poset (P, ≤) is ﬁnite if the set P has a ﬁnite number of elements. Any ﬁnite poset

(P, ≤) can be represented by a Hasse diagram in which the elements of P are represented

as vertices, and the lines represent the covering relation. Note that x ≤ y if there is a path

leading upward from x to y.

•

•

• •

x

y

...................................................

..................................................

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Example of a Hasse Diagram.

* The abbreviation q.e.f. stands for quod erat faciendum. This marks the end of a construction, just

as q.e.d., which stands for quod erat demonstrandum, marks the end of a proof.

5

Examples of Finite Posets. The empty set and any set with a single element can be

partially ordered in only one way. A posets with two elements must be isomorphic to one

of the two posets whose Hasse diagrams are depicted below. One is a chain, and the other

is an antichain.

•

•

• • chain antichain

Show below are Hasse diagrams of the ﬁve types of posets with three elements. Note that

three of these posets are isomorphic to their own duals; one of these is a chain, and another

is an antichain. Of the remaining two, each is isomorphic to the dual of the other. Note

also that three of these have connected Hasse diagrams, and the remaining two are made

up of disconnected pieces.

•

•

•

•

•

•

•

•

•

•

•

•

• • •

................................................................................ ................................................................................

................................................................................ ................................................................................

There are sixteen posets with 4 elements, of which 10 have connected Hasse diagrams. The

number of nonisomorphic posets with n elements grows very rapidly with increasing n.*

n

number of number of connected

distinct posets Hasse diagrams

0 1 1

1 1 1

2 2 1

3 5 3

4 16 10

5 63 44

6 318 238

7 2045 1650

8 16999 14512

Number of Nonisomorphic Posets

* These numbers can be found in a paper by Shawpawn Kumar Das, 1977, Journal of the Association

for Computing Machinery, Vol. 24, pp. 676–692.

The number of connected Hasse diagrams is related to the total number of posets by a general

formula. See for example:

Albert Nijenhuis and Herbert S. Wilf, 1975, Combinatorial Algorithms, Academic Press. Cf. p. 79.

6

Comparability and Covering. We shall now discuss a relation between the concepts of

comparability and covering. For this purpose it is convenient to introduce the following

notation:

•

•

•

•

•

•

•

•

x

x

x

x

y

y

y

y

means

means

means

means

<

<

≤

x

................

................

................

................

y,

x <y,

x < y,

x ≤ y.

Dashed lines indicate assertions to be proved.

Lemma (Comparability and Covering.) Let x, y, x

, y

**be elements of a poset (P, ≤).
**

If x <x

and y <y

, and if x

................

................

................

................

y, x

................

................

................

................

y

and x

................

................

................

................

y, then either x = y or x

................

................

................

................

y

, or both.

•

•

•

•

x

x

y

y

..........................................................................................................................................................................................................

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

........... ........... ........... ...........

Proof: If x

≤ y, then x

≤ y < y

and hence x

................

................

................

................

y

**, and so we are done in this case.
**

Similarly, by symmetry, we are done if y

≤ x. So we may as well assume that both x < y

and y < x

**. At this point we have the picture:
**

x

x

y

y

•

•

•

•

..........................................................................................................................................................................................................

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

........... ........... ........... ...........

Because the picture is symmetric, we may also assume without loss of generality that

x ≤ y. We are done if x = y, and if x < y, then x < y < x

, that is, y ∈ (x, x

), violating

the assumption that x <x

. q.e.d.

Minimal and Maximal Elements. An element m ∈ P of a poset (P, ≤) is said to be a

maximal element if there is no greater element in P, that is, if

(∀x ∈ P) m ≤ x =⇒ m = x.

Similarly, an element m ∈ P of a poset (P, ≤) is a minimal element if there is no lesser

element in P.

7

Proposition. An element m ∈ P in a poset (P, ≤) is maximal if and only if [x, ) = ¦x¦.

Similarly, m ∈ P is minimal if and only if ( , x] = ¦x¦.

Proof: This is clear from the deﬁnitions. q.e.d.

Proposition. If a poset has a greatest element, then that element is the one and only

maximal element in that poset.

Proof: Easy exercise. q.e.d.

Example. Minimal and maximal elements need not exist in general. If they do exist,

they need not be unique. A maximal element need not be a greatest element.

Construction: If ω = ¦0, 1, 2, 3, . . .¦ denotes the system of natural numbers, then the

poset (ω, ≤) has a minimal element, but no maximal element. If S is any set, then every

element of the poset (S, =) is both minimal and maximal. If S holds more than one

element, there is no greatest or least element. q.e.f.

Maximal elements, unlike greatest elements, do not behave well under monotone map-

pings, but we do have a weak result:

Proposition. If f: P

1

→ P

2

is a one-to-one and onto monotone mapping from a poset

(P

1

, ≤) to a poset (P

2

, ≤), and if P

2

has a maximal element, then so does P

1

.

Proof: If y = f(x) is a maximal element of P

2

, then x is a maximal element of P

1

.

q.e.d.

The above result, in conjunction with Zorn’s lemma to be discussed later, could possibly

be used to prove existence of maximal elements in certain posets.

Covering Operations. A covering operation f on a poset (P, ≤) is a mapping f: P → P

for which

(∀x ∈ P) x <f(x).

A covering operation need not be monotone.

An example of a covering operation is the successor operation n → n + 1 on the

poset (ω, ≤) of natural numbers. A nonempty poset with a covering operation must have

inﬁnitely many elements; if a is one of them, we have

a <f(a) <f(f(a)) < .

In general, a poset with a covering operation f can not have a maximal element. For if m

were a maximal element, then f(m) would be strictly greater than m, which is impossible.

Proposition. If a well-ordered set (W, ≤) has no greatest element, then there is a unique

covering operation on W.

Proof: Let a ∈ W be any element of a well-ordered set (W, ≤). If (a, ) = ∅, then a would

be the greatest element of W. If W has no greatest element, then the open half-interval

(a, ) is not empty, and therefore has a least element, say b. Then a < b, and clearly b is

the only element which covers a. The unique covering operation assigns a → b. q.e.d.

8

Proposition. If x ∈ P is any element of a poset (P, ≤), then the closed half-interval [x, )

is invariant under any covering operation on P.

Proof: If y ∈ [x, ), then x ≤ y <f(y), and thus f(y) ∈ [x, ). q.e.d.

Upper and Lower Bounds. An element b ∈ P is called an upper bound for a subset

S ⊂ P of a poset (P, ≤) if

(∀s ∈ S) s ≤ b.

Similarly, an element a ∈ P is called an lower bound for S ⊂ P if

(∀s ∈ S) a ≤ s.

These deﬁnitions can be restated as follows:

Lemma. An element b of a poset P is an upper bound for a subset S of P if and only

if S ⊂ ( , b]. Similarly, an element a in P is a lower bound for a subset S if and only if

S ⊂ [a, ).

Proof: This follows directly from the deﬁnitions of upper and lower bounds and the sets

[a, ) and ( , b]. q.e.d.

We sometimes draw pictures of upper and lower bounds as follows:

•

•

a

b

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

........................................................................................................................................................................................................................................................... ...........................................................................................................................................................................................................................................................

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. .. ... .................................................................................................................................................................. ... .. . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.................................................................................. ... .. . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. .. ... .................................................................................

S

S

Pictures of Upper and Lower Bounds.

A subset S ⊂ P of a poset (P, ≤) is said to be bounded above if there exists an upper

bound for S in P. Similarly, a subset S ⊂ P is said to be bounded below if there is a lower

bound for S in P. A subset S ⊂ P is bounded if it is bounded both above and below.

Example. The half-intervals ( , b) and ( , b] are bounded above. The half intervals (a, )

and [a, ) are bounded below. The intervals (a, b), (a, b], [a, b) and [a, b] are all bounded

subsets.

The empty set is, of course, a bounded subset of every poset. Also, any singleton is a

bounded subset.

Counterexample. In general, a ﬁnite subset of a poset need not be bounded.

Construction: Any subset with two or more elements in the poset (S, =) fails to be

bounded. q.e.f.

9

Directed Posets. A poset is said to be directed if every ﬁnite subset is bounded above.

(A better term, perhaps, would be directed upward. The term directed downward could

then be applied to the dual concept of a poset in which every ﬁnite subset is bounded

below.)

Proposition. A poset is directed if and only if every set of two elements is bounded

above. That is, a poset (P, ≤) is directed if and only if for all x, y ∈ P, there exists an

element z ∈ P such that x ≤ z and y ≤ z.

Proof: Easy exercise. q.e.d.

What happens if we require not only that every ﬁnite subset is bounded above but that

every subset is bounded above? We remark that in this case we have the following simple

result.

Proposition. A poset (P, ≤) has a greatest element if and only if every subset of P is

bounded above.

Proof: If P itself has an upper bound, then that upper bound must be the greatest

element of P. Conversely, if P has a greatest element, then that greatest element is an

upper bound for every subset of P. q.e.d.

Corollary. A ﬁnite poset is directed if and only if it has a greatest element.

Bounded subsets are well-behaved under monotone mappings.

Proposition. Let f: P

1

→ P

2

be a monotone mapping from a poset P

1

to a poset P

2

. If

a subset S ⊂ P

1

is bounded above in P

1

, then its image f[S] ⊂ P

2

under f is bounded

above in P

2

.

Proof: If b is an upper bound for the subset S, then f(b) is an upper bound for the

image f[S]. Indeed, since s ≤ b for all s in S, then f(s) ≤ f(b). q.e.d.

Proposition. The image of a directed subset under a monotone mapping is a directed

subset.

Proof: Let f: P

1

→ P

2

be monotone, and let S be a subset of P

1

. Any two elements of

f[S] can be represented as f(x) and f(y) where x and y belong to S. Since S is directed,

there exists an element z in S which is an upper bound for both x and y. Since f is

monotone, then f(z) is an upper bound for f(x) and f(y). q.e.d.

Proposition. If a directed poset has a maximal element, then that element is the greatest

element.

Proof: If m is a maximal element, and if x is any element, then since the poset is directed,

there is an element z which is an upper bound for both m and x. Since m is maximal,

then we must have m = z. Hence m is an upper bound for every element x in P. q.e.d.

10

Least Upper and Greatest Lower Bounds. Let (P, ≤) be a poset. An element b ∈ P

is the least upper bound of a subset S ⊂ P if b is the least element of the set of all upper

bounds for S. If a set S ⊂ P has a least upper bound, we denote it by lub S or

S.

Similarly, an element is the greatest lower bound of a subset S if it is the greatest element

of the set of all lower bounds for S. If a set S ⊂ P has a greatest lower bound, we denote

it by glb S or

**S. In some books, the least upper bound of S is called the supremum,
**

denoted by sup S and the greatest lower bound of S is called the infemum, denoted by

inf S.

Counterexample. A minimal upper bound need not be a least upper bound.

Construction: In the poset depicted below, elements c and d are both minimal upper

bounds for the set S = ¦a, b¦, but there is no least upper bound. q.e.f.

• •

• •

a

b

c d

.....................................................................................................

.....................................................................................................................................................................................

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.....................................................................................................

Proposition. If x ∈ P is any element of a poset (P, ≤), then

( , x] = x.

Proof: This is clear. q.e.d.

For open half-intervals, the results are more complicated. For example, in the poset

(ω, ≤) of the natural numbers, if n > 0, then

**( , n) = n −1, whereas in the poset (RRR, ≤),
**

where RR R is the set of all real numbers, we have

**( , x) = x for all x ∈ RR R. More generally:
**

Proposition. If (C, ≤) is a chain in which no element covers another, then (∀x ∈

C)

( , x) = x.

Proof: Clearly x is an upper bound for the half-interval ( , x). If y ∈ C is another upper

bound for the half-interval ( , x), then ( , x) ⊂ ( , y]. Since (C, ≤) is a chain, either y < x

or x ≤ y. If y < x, then ( , x) = ( , y]. If z ∈ (x, y), then z ∈ ( , x) and z / ∈ ( , y], which is

absurd. Hence (x, y) = ∅, and so x <y, contrary to the hypothesis that no element covers

another in (C, ≤). Hence x ≤ y. That is, x is the least upper bound of the half-interval

( , x). q.e.d.

Proposition. If a subset S ⊂ P of a poset (P, ≤) has a greatest element g, then g =

S.

Proof: Clearly S ⊂ ( , g], so g is an upper bound for S. If b is also an upper bound for

S, then since g ∈ S, we must have g ≤ b. Hence g is the least element of the set of all

upper bounds for S. q.e.d.

Counterexample. Monotone mappings need not preserve least upper bounds.

Construction: Let RR R be the set of all real numbers, ordered in the usual way. Let

f: RRR →RR R be the staircase function deﬁned by

(∀x ∈ RR R) f(x) = x| = greatest integer n such that n ≤ x.

11

1

2

3

1 2 3

•

•

•

•

x

f(x)

If S = [0, 1), then

S = 1 and f[S] = ¦0¦, so that

f[S] = 0, whereas f (

S) = f(1) = 1.

Thus

f[S] = f (

S). q.e.f.

Semi-Lattices. A poset (P, ≤) is a join-semilattice if for any pair of elements x, y ∈ P,

there is a least upper bound in P for the set ¦x, y¦. We write x ∨ y =

¦x, y¦. The

symbol ∨ is called join or cup. Similarly, A poset (P, ≤) is a meet-semilattice if for any

pair of elements x, y ∈ P, there is a greatest lower bound in P for the set ¦x, y¦. We write

x ∧ y =

**¦x, y¦. The symbol ∧ is called meet or cap.
**

Proposition. Any join-semilattice is a directed poset.

Proof: This is clear. q.e.d.

Example. If S is any set, and PowS is its power set, then (PowS, ⊂) is both a join-

semilattice and a meet-semilattice, with ∨ and ∧ being the set-theoretic union and inter-

section, respectively.

Lemma. If (P, ≤) is a join-semilattice, then

(∀x, y ∈ P) x ≤ x ∨ y & y ≤ x ∨ y,

(∀x, y ∈ P) x ≤ z & y ≤ z =⇒ x ∨ y ≤ z.

Proof: The element x ∨ y is the least upper bound of ¦x, y¦. q.e.d.

Theorem. If (P, ≤) is a join-semilattice, then

Idempotent Law: (∀x ∈ P) x ∨ x = x,

Commutative Law: (∀x, y ∈ P) x ∨ y = y ∨ x,

Associative Law: (∀x, y, z ∈ P) (x ∨ y) ∨ z = x ∨ (y ∨ z).

Proof: Since x is obviously the greatest element of ¦x, x¦, then x ∨ x = x. Since

¦x, y¦ = ¦y, x¦, then x ∨y = y ∨x. The associative law follows from the lemma. q.e.d.

12

Lattices. A poset (L, ≤) is a lattice if it is both a join-semilattice and a meet-semilattice.

That is, any pair of elements x, y ∈ L has both a least upper bound x ∨ y ∈ L and a

greatest lower bound x ∧ y ∈ L.

Corollary. Any lattice is a directed poset.

Proposition. Any chain (C, ≤) is a lattice.

Proof: If x, y ∈ C then either x ≤ y or y ≤ x. If x ≤ y, then x ∧y = x and x ∨y = y. If

y ≤ x, then x ∧ y = y and x ∨ y = x. q.e.d.

We may summarize this as:

well-ordered set =⇒ chain =⇒ lattice =⇒ join-semilattice =⇒ directed poset.

Theorem (Absorptive Laws). If (L, ≤) is a lattice, then

(∀x, y ∈ L) x ∨ (x ∧ y) = x,

(∀x, y ∈ L) x ∧ (x ∨ y) = x.

Proof: Clearly x ≤ x ∨ (x ∧ y). Since x ≤ x and x ∧ y ≤ x, then x is an upper

bound for the set ¦x, x ∧ y¦. Since x ∨ (x ∧ y) is the least upper bound of this same set,

then x ∨ (x ∧ y) ≤ x. By the antisymmetric property of partial ordering, it follows that

x ∨ (x ∧ y) = x. The second statement is proved in a similar fashion. q.e.d.

A comment on axiomatics is in order. We have deﬁned a lattice as a certain type of

poset. Alternatively, one can view a lattice as a set L equipped with two binary operations

∨ and ∧ subject to six axioms: the two commutative laws, the two associative laws, and

the two absorptive laws. One can then reintroduce the partial ordering ≤ on L by deﬁning

x ≤ y to mean x = x ∧ y. Note that, on account of the absorptive laws, the statements

x = x ∧ y and x ∨ y = y are equivalent. Incidentally, the idempotent laws need not be

explicitly postulated since they actually follow from the absorptive laws. Indeed:

x = x ∨ [x ∧ (x ∨ x)] = x ∨ x,

x = x ∧ [x ∨ (x ∧ x)] = x ∧ x.

Bibliography. There are lots of books on lattices, and some of them treat posets as well,

but most deal with special topics.

G. Birkhoﬀ, Lattice Theory, First edition: 1940; Second edition: 1948; Third

edition: 1961.

G. Szasz, 1963, Academic Press, Introduction to Lattice Theory.

D. E. Rutherford, 1965, Hafner, New York, Introduction to Lattice Theory.

G. Gr¨ atzer, 1971, Lattice Theory.

These are just a few of the classics; there are lots more!

13

A monotone mapping f: L

1

→ L

2

from one lattice to another is called a lattice homo-

morphism if

(∀x, y ∈ L) f(x ∨ y) = f(x) ∨ f(y),

(∀x, y ∈ L) f(x ∧ y) = f(x) ∧ f(y).

We say that f preserves the lattice operations ∨ and ∧. In general, a monotone mapping

need not preserve the lattice operations.

Complete Lattices. In a lattice, any nonempty ﬁnite set ¦x

1

, . . . , x

n

¦ has a least upper

bound and a greatest lower bound, because

¦x

1

, x

2

, . . . , x

n

¦ = ( ((x

1

∨ x

2

) ∨ x

3

) ∨ ∨ x

n

),

¦x

1

, x

2

, . . . , x

n

¦ = ( ((x

1

∧ x

2

) ∧ x

3

) ∧ ∧ x

n

).

The empty set and inﬁnite subsets, however, need not have least upper and greatest lower

bounds. For example, the poset (ZZZ, ≤) of all the integers, with the usual order, is a chain,

and therefore a lattice, but neither ∅ nor ZZ Z has a least upper bound. In the case of ∅,

every integer is an upper bound, but there is no least integer; in the case of ZZZ, there is no

upper bound at all.

We say that a poset (P, ≤) is a complete lattice if every subset S ⊂ P has a least upper

bound

S ∈ P, and a greatest lower bound

**S ∈ P. Actually, the existence of greatest
**

lower bounds could be omitted from this deﬁnition in view of the following theorem.

Theorem. If every subset S of a poset (P, ≤) has a least upper bound

S, then every

subset S has a greatest lower bound

S.

Proof: The set L of all lower bounds for S has a least upper bound

L ∈ P. Any

element s ∈ S is an upper bound for L, and hence (∀s ∈ S)

L ∈ P ≤ s. It follows that

L ∈ L, that is,

L is the greatest element of L; that is,

**L is the greatest lower bound
**

for S. q.e.d.

The converse is also true; if every subset has a greatest lower bound, then every subset has

a least upper bound. The proof is similar; the least upper bound of a set S is the greatest

lower bound

**U of the set U of all upper bounds of S.
**

Proposition. A complete lattice (L, ≤) has a least element and a greatest element.

Proof: Since any element x ∈ L is an upper bound for ∅, the least upper bound

∅ is

the least element of L. Similarly,

**∅ is the greatest element of L. q.e.d.
**

Corollary. Any ﬁnite nonempty lattice is complete.

Proof: Any ﬁnite nonempty subset of a lattice has a least upper bound and a greatest

lower bound. q.e.d.

Proposition. For any set S, the poset (PowS, ⊂) is a complete lattice.

Proof: If A ⊂ S, we have

A =

A. If A is not empty, then

A =

A. If A = ∅,

then

A = S. q.e.d.

14

Example. Any closed interval [a, b] on the real line (RRR, ≤) is a complete lattice if a < b.

A closed interval [a, b] on the rational line (QQQ, ≤) is not a complete lattice if a < b.

For example, because

√

2 is irrational, the set

¦x ∈ QQ Q [ x > 0 & x

2

< 2¦ ⊂ [0, 2]

has no least upper bound in the closed interval [0, 2] ⊂ QQ Q.

A somewhat more interesting example is provided by topology. Recall that a topology T

on a set S is a collection of subsets of S, called open subsets such that

S ∈ T ,

(∀o ⊂ T )

o ∈ T ,

(∀G

1

, G

2

∈ T ) G

1

∩ G

2

∈ T .

This deﬁnition of a topology is essentially equivalent to that given on page 37 in John L.

Kelley’s book, General Topology.

Note that since ∅ ⊂ T , then ∅ =

**∅ ∈ T . The interior of any subset A ⊂ S is
**

deﬁned to be the union of the collection of all open subsets contained in A.

Proposition. If T is a topology on a set S, then (T , ⊂) is a complete lattice.

Proof: If o ⊂ T , then

**o ∈ T is the least upper bound for o. Note that if o is
**

not empty, then the greatest lower bound

o is not the intersection

**o, but rather the
**

interior of the intersection. When o = ∅, we have

o =

T = S. q.e.d.

The Complete Lattice Fixed Point Theorem. An element x ∈ L is a ﬁxed point for

an operation f: L → L if f(x) = x. A subset S ⊂ L is invariant under an operation f if

f[S] ⊂ S; that is, S is invariant under f if

(∀x ∈ S) f(x) ∈ S.

In particular, x is a ﬁxed point of f if and only if the singleton ¦x¦ is invariant under f.

Theorem. Any monotone operation f: L → L on a complete lattice (L, ≤) has a ﬁxed

point.

Proof: The subset S = ¦x ∈ L [ x ≤ f(x)¦ is invariant under f because

x ≤ f(x) =⇒ f(x) ≤ f(f(x)).

Since L is a complete lattice, then

S ∈ L and (∀x ∈ S)x ≤

S, so that x ≤ f(x) ≤

f(

S). Hence f(

S) is an upper bound for S, and

S ≤ f(

S). Therefore

S ∈ S.

Since S is invariant, then f(

S) ∈ S. Hence f(

S) ≤

S. It follows that

S is a ﬁxed

point of f. q.e.d.

Davis proves that the converse is also true.

A. C. Davis, 1955, Paciﬁc J. Math. 5, 311–319, A Characterization of Complete

Lattices.

Another ﬁxed point theorem (due to Bourbaki) will be proved later; the Bourbaki ﬁxed

point theorem uses Zorn’s lemma.

15

Filters and Ideals in Lattices. A subset S ⊂ L of a lattice (L, ≤) is a sublattice if

(∀x, y ∈ S) x ∧ y ∈ S,

(∀x, y ∈ S) x ∨ y ∈ S.

Proposition. The intersection of any nonempty collection of sublattices of a lattice is a

sublattice.

Proof: This is elementary. q.e.d.

A subset I ⊂ L of a lattice (L, ≤) is an ideal if

(∀x ∈ I)(∀y ∈ L) x ∧ y ∈ I,

(∀x, y ∈ I) x ∨ y ∈ I.

An ideal is clearly a sublattice, but a sublattice need not be an ideal.

Example. The empty set ∅ and L itself are ideals of any lattice (L, ≤).

The ideals ∅ and L are called improper ideals of L.

Example. If x ∈ L is any element of a lattice (L, ≤), then the closed half-interval ( , x]

is an ideal of L.

The ideals ( , x] are called principal ideals of L.

Ideals can be characterized in another way which does not make use of the lattice

operations. A subset H ⊂ P of a poset (P, ≤) is hereditary if

(∀h ∈ H) ( , h] ⊂ H.

That is, H ⊂ P is hereditary if and only if

(∀x ∈ P)(∀h ∈ H) x ≤ h =⇒ x ∈ H.

Recall also that a subset D ⊂ P of a poset (P, ≤) is directed if

(∀x, y ∈ D)(∃z ∈ D) x ≤ z & y ≤ z.

Proposition. A subset I ⊂ L of a lattice (L, ≤) is an ideal of L if and only if I is directed

and hereditary.

Proof: We leave the proof as an easy exercise. q.e.d.

The concept of a hereditary directed subset makes sense in any poset, thereby generalizing

the concept of an ideal in a lattice. This concept is interesting mainly in inﬁnite posets in

view of the following theorem.

16

Proposition. Every nonempty hereditary directed subset of a ﬁnite poset (P, ≤) is a

closed half-interval ( , x].

Proof: Any ﬁnite directed subset has a greatest element. If a hereditary subset H has a

greatest element x, then H = ( , x]. q.e.d.

Proposition. The set of all ideals of a lattice (L, ≤), partially ordered by inclusion, is a

complete lattice.

Proof: If ( is a nonempty collection of ideals, then

**( is an ideal and this ideal is the
**

greatest lower bound for (. Also

∅ = L. q.e.d.

Note that if ( is a collection of ideals, then the least upper bound

**( need not be the
**

union

(, but instead

**( is the intersection of the set of all ideals which contains the
**

union.

The mapping x → ( , x] is a one-to-one monotone mapping which allows us to embed any

lattice (L, ≤) in its lattice of ideals. Thus, any lattice can be embedded in a complete

lattice. (Is this mapping a lattice homomorphism?)

By a complete chain we mean a chain which is a complete lattice. For example, any

closed interval on the real line [a, b] ⊂ RRR is a complete chain.

Proposition. The lattice of ideals of a chain (C, ≤) is a complete chain.

Proof: If I and J are ideals of C, and if I is not contained in J, then there exists an

element x ∈ I which does not belong to J. If y ∈ J, then ( , y] ⊂ J and since x

................

................

................

................

y and

x / ∈ J, we must have y < x. Since y is an arbitrary element of J, then J ⊂ ( , x) ⊂ I.

q.e.d.

Corollary. Every chain can be monotonically embedded in a complete chain.

Proof: Use the mapping x → ( , x]. q.e.d.

A comment is in order here. Starting with the rational line (QQQ, ≤), this completion proce-

dure does not yield the real line (RRR, ≤). Indeed, the real line fails to be a complete chain

because it lacks a least and greatest element.

A subset D of a lattice (L, ≤) is a dual ideal if

(∀x ∈ D)(∀y ∈ L) x ∨ y ∈ D,

(∀x, y ∈ D) x ∧ y ∈ D.

An dual ideal is clearly a sublattice. A proper dual ideal in a lattice L is called a ﬁlter.

The ﬁlters of a lattice are partially ordered by inclusion. A maximal ﬁlter is called an

ultraﬁlter.

17

Various Chain Conditions. We shall say that a subset S ⊂ P of a poset (P, ≤) is

inductive if every nonempty chain C ⊂ S has an upper bound in S. (This is precisely the

hypothesis of Zorn’s lemma.) We say that a subset S ⊂ P of a poset (P, ≤) is strictly

inductive if every nonempty chain C ⊂ S has a least upper bound in P, and

C ∈ S.

Finally, we say that a subset S ⊂ P of a poset (P, ≤) is Noetherian or satisﬁes the ascending

chain condition (often abbreviated as a.c.c. if every nonempty chain C ⊂ S has a greatest

element. The following implications are clear:

ﬁnite =⇒ Noetherian =⇒ strictly inductive =⇒ inductive.

Note that the empty set satisﬁes all three conditions.

The above terminology for chain conditions on a poset is not completely standard. The

term inductive is used with the present meaning by Bourbaki in Theory of Sets, Chapter

III, Section 2.4. However, the term inductive set is also used in ordinal number theory

with a diﬀerent meaning which need not concern us here. The term strictly inductive is my

own invention. The term Noetherian is used in the present sense by Bourbaki in Theory

of Sets, Chapter III, Section 6.5. The same term is used with a similar meaning in ring

theory; an abelian ring is Noetherian if its lattice of ideals satisﬁes the ascending chain

condition.

Any poset with a greatest element, of course, is inductive.

Proposition. A chain is inductive if and only if it has a greatest element.

Proof: If (C, ≤) is an inductive chain, then C itself must have an upper bound b ∈ C.

Since x ≤ b for all x ∈ C, the element b is the greatest element of C. q.e.d.

The connection between ordinary mathematical induction and inductive sets is slightly

topsy turvy. The set of positive integers, of course, is not even an inductive set because it

is a chain with no greatest element. We need to turn the integers upside down:

Example. The set of negative integers, with the usual ordering, is an example of an

inﬁnite Noetherian poset.

We say that a poset (P, ≤) satisﬁes the descending chain condition, abbreviated d.c.c., if

every nonempty chain C ⊂ P has a least element.

Example. A well-ordered set is a chain which satisﬁes the d.c.c.

Example. Any complete lattice is strictly inductive; for example, any closed interval

[a, b] ⊂ RR R with a < b is strictly inductive. A closed interval of the rational numbers, with

the usual ordering, is an example of an inductive poset which is not strictly inductive.

Counterexample. A subset of an inductive set need not be inductive.

Construction: The closed interval [0, 1] ⊂ RR R is inductive, but the subset

¦0, 1/2, 3/4, 7/8, . . .¦ ⊂ [0, 1]

is a chain with no greatest element, and is therefore not inductive. q.e.f.

The importance of the concept of a strictly inductive poset is due to the following:

18

Theorem. The set Ch P of all chains in a poset (P, ≤), partially ordered by inclusion, is

a strictly inductive poset.

Proof: Suppose ( is a nonempty chain of chains; that is, ( is a nonempty chain in the

poset (Ch P, ⊂). If x, y ∈

**(, then there exist chains C
**

1

, C

2

∈ ( with x ∈ C

1

and y ∈ C

2

.

Since ( is a chain, either C

1

⊂ C

2

or C

2

⊂ C

1

. In either case, both x and y belong to the

bigger one of these chains. Hence x

................

................

................

................

y, and it follows that

( is a chain in P. Clearly

**( is the least upper bound for the chain ( ⊂ Ch P. q.e.d.
**

Proposition. If a poset (P, ≤) is strictly inductive, and if x ∈ P, then the subsets ( , x]

and [x, ) are strictly inductive.

Proof: We leave this as an exercise. q.e.d.

Proposition. The intersection of any nonempty collection of strictly inductive subsets is

strictly inductive.

Proof: We leave this as an exercise. q.e.d.

Spinal Elements. The following terminology is not standard, but will prove useful in

connection with Zorn’s lemma, to be discussed later. We shall say that an element x ∈ P

of a poset (P, ≤) is a spinal element if it is comparable with every element in P. The spine

of a poset (P, ≤) is the set of all spinal elements; we denote it by

Spine P = ¦x ∈ P [ (∀y ∈ P)x

................

................

................

................

y¦.

The spine of any poset is a chain, possibly empty. Note that x ∈ Spine P if and only if

P = ( , x] ∪ [x, ).

Example. If a poset (P, ≤) has a least or greatest element, then these are spinal elements.

Example. Any chain (C, ≤) is its own spine.

We shall prove a theorem about the spine of a strictly inductive poset. For this we

need the following lemma.

Lemma. If (P, ≤) is a strictly inductive poset, and if x ∈ P, then ( , x] ∪ [x, ) is a strictly

inductive subset.

Proof: Any nonempty chain C ⊂ ( , x] ∪ [x, ) is the union of two chains C = C

+

∪ C

−

where C

+

⊂ [x, ) and C

−

⊂ ( , x]. If C

+

is not empty, then

C =

C

+

∈ [x, ). If

C

+

= ∅, then

C =

C

−

∈ ( , x]. q.e.d.

Theorem. The spine of a strictly inductive poset is either empty or has a greatest element.

Proof: If (P, ≤) is a strictly inductive poset, and if its spine S = Spine P is not empty,

then

S ∈ P. If x ∈ P, then S ⊂ ( , x] ∪ [x, ), and hence

S ∈ ( , x] ∪ [x, ). Hence

(∀x ∈ P) x

................

................

................

................

S.

Hence

S ∈ S. q.e.d.

19

A Theorem due to Zermelo. In this section we shall prove a theorem due to Zermelo

which is a close relative of Zorn’s lemma, but which does not require the axiom of choice.

We begin with a technical lemma which will be needed later.

Lemma. If f is a covering operation on a poset (P, ≤), and if x ∈ P is a spinal element of

P, then the set of elements comparable to f(x) is invariant under f.

x

f(x)

y

f(y)

•

•

•

•

..........................................................................................................................................................................................................

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

........... ........... ........... ...........

Proof: We have to show that

(∀y ∈ P) y

................

................

................

................

f(x) =⇒ f(y)

................

................

................

................

f(x).

Since x is a spinal element, this follows from the diagram. q.e.d.

A strictly inductive subset of a poset which is invariant under a covering operation f

is called an f-tower. We will prove theorem due to Zermelo which asserts that f-towers

do not exist, (except in the trivial case of the empty set.) “Towers topple.” For this we

need the following lemma.

Lemma. Let f be a covering operation on a poset (P, ≤). The intersection of any non-

empty set of f-towers is itself an f-tower.

Proof: Invariant subsets and strictly inductive sets are preserved under intersections.

q.e.d.

Theorem (Zermelo). A nonempty strictly inductive poset (P, ≤) has no covering op-

erations.

Proof: Suppose f were a covering operation on (P, ≤), and b ∈ P. Then [b, ) is an

f-tower, and b ∈ [b, ). The intersection M of all f-towers to which b belongs is an f-tower,

and b ∈ M. This minimal f-tower is contained in any f-tower to which b belongs, so

M ⊂ [b, ). Clearly b is the least element of M, so the spine S of M is not empty. We

proved earlier that the spine of a non-empty strictly inductive poset has a greatest element.

Let g ∈ S be the greatest element of the spine S. By a technical lemma proved earlier, the

set T ⊂ M of all elements in M which are comparable to f(g) is invariant under f. Note

that b ∈ T since b ≤ g <f(g). Since M and ( , f(g)] ∪ [f(g), ) are both strictly inductive,

so is their intersection T. Thus T is an f-tower to which b belongs. Since M is the smallest

f-tower to which b belongs, then M ⊂ T. Since T ⊂ M and M ⊂ T, then T = M. Hence

f(g) is a spinal element of M. This is absurd because g is the greatest spinal element of

M and f(g) is strictly greater than g. q.e.d.

20

Zorn’s Lemma. Up to now, we have not made use of the axiom of choice; Zermelo’s

theorem on the nonexistence of covering operations on a strictly inductive poset does not

require the axiom of choice. From it we can deduce a more useful result, called Zorn’s

lemma, whose proof does make essential use of the axiom of choice.

Theorem (Zorn’s Lemma). Every nonempty inductive poset has a maximal element.

Proof: Suppose (P, ≤) is a nonempty inductive poset which has no maximal element. If

C ⊂ P is a chain in P, then by the hypothesis that P is inductive, there exists an upper

bound for C, say x. (Note that if C is the empty set, then any element of P is an upper

bound for C.) Since x is not a maximal element of P, there exists a strictly larger element,

say y. Since C ⊂ ( , x] and x < y, then y / ∈ C. Therefore C ∪ ¦y¦ is a chain. Clearly the

chain C is covered by the chain C ∪ ¦y¦ in the poset (Ch P, ⊂) of all chains in P. By the

axiom of choice, there is a function f: ChP → Ch P which assigns to each chain C ∈ Ch P

some such chain f(C) = C ∪ ¦y¦ which covers C. Then f is a covering operation on the

poset of chains in P. Since the poset (Ch P, ⊂) is strictly inductive, by Zermelo’s theorem,

there can be no covering operation on (Ch P, ⊂). This is a contradiction. q.e.d.

The converse of Zorn’s lemma is not valid.

Counterexample. A poset (P, ≤) having a maximal element need not be inductive.

Construction: In the poset shown below, the element m is maximal, but the chain

¦0, 1, 2, . . .¦ has no upper bound. q.e.d.

•

•

•

•

•

0

1

2

3

m

........................................................................ ...................................................................................................................................................................................................................... .............

.............

.............

Corollary. Any chain in a poset is contained in a maximal chain.

Proof: The poset (Ch P, ⊂) of all chains in a poset (P, ≤) is strictly inductive. It is

not empty since ∅ ∈ Ch P. If C ∈ Ch P, then the half-interval [C, ) ⊂ Ch P is a strictly

inductive subset of Ch P. Hence there is a maximal element in [C, ). q.e.d.

Corollary. Any antichain in a poset is contained in a maximal antichain.

Proof: The union of any chain of antichains is an antichain. Hence the poset of all

antichains in a given poset is strictly inductive. q.e.d.

Corollary (Bourbaki Fixed Point Theorem.) Any operation f: P → P on a

nonempty inductive poset P satisfying (∀x ∈ P) x ≤ f(x) has a ﬁxed point.

Proof: If x is a maximal element of P, then x ≤ f(x) implies that f(x) = x. q.e.d.

21

Example. If L is any collection of statements, then (L, = satisﬁes the reﬂexive and ⇒) transitive properties, but may fail to satisfy the antisymmetric law. Two distinct statements A, B ∈ L may be logically equivalent: A = B and B = A. ⇒ ⇒ Chains and Antichains. Elements x, y ∈ P of a poset (P, ≤) are comparable, written . .. .... .... x ........... y, if x ≤ y or y ≤ x. ... ... A subset C of a poset (P, ≤) is a chain if any pair of elements in C are comparable:

... . ... .... ... (∀x, y ∈ C) x ........... y. . ..

Chains are also called totally ordered subsets.* A subset A of a poset (P, ≤) is an antichain if no two distinct elements of A are comparable: .... .... ... ... ⇒ (∀x, y ∈ C) x .............. y = x = y. Example. In any poset (P, ≤), the empty set ∅ and every singleton {x} is both a chain and an antichain. R Example. Any subset of the real line (R, ≤), with the usual order, is a chain. In particular, ω = {0, 1, 2, . . .} and Z = {. . . , −2, −1, 0, 1, 2, . . .} are chains. Proposition. Any subset of a chain is a chain. Proof: This is clear. q.e.d.

Intervals and Half-Intervals. In any poset (P, ≤) we can deﬁne certain subsets which we call intervals and half-intervals. If a, b ∈ P , then the following sets are called closed half-intervals: [a, ) = {x ∈ P | a ≤ x}, ( , b] = {x ∈ P | x ≤ b}. The closed half-interval ( , b] is also called a weak initial segment. The sets (a, ) = {x ∈ P | a < x}, ( , b) = {x ∈ P | x < b}. are called open half-intervals. Lemma. If (P, ≤) is any poset, then (∀a, b ∈ P ) a ≤ b = ( , a] ⊂ ( , b], ⇒ (∀a, b ∈ P ) a = b = ( , a] = ( , b]. ⇒ Proof: This is an easy exercise which uses all three of the poset axioms. *

The terms linearly ordered and completely ordered are used by some authors.

q.e.d.

2

. . Proof: Use the antisymmetric law.. obtained by intersecting various types of halfintervals.. Finally.. ) ∩ ( . by intersecting a closed half-interval and an open half-interval. b] = (a... For any poset (P... b) = (a.. ≤). . ) ∩ ( . * Some authors prefer to use the terms top element or last element for what we call a greatest element. sets of the form (a.. b).. Similarly. the terms bottom element or ﬁrst element are used for a least element. and at most one least element.d... ). q. y} = ( . ≤) which are comparable with a given element x ∈ P is the union of two half-intervals: . ) ∩ ( . q. x] ∪ [x. (a..e. x] = {x}. b) = [a. obtained by intersecting two closed half-intervals are called closed intervals. (∀x ∈ P ) [x.. A poset need not have any greatest or least element.e. Note that g is the greatest element of S if g ∈ S and S ⊂ ( . A subset S ⊂ P of a poset (P. we deﬁne two types of half-open intervals [a. Lemma. ≤) is a greatest element of S if (∀s ∈ S) s ≤ g. ). An element g ∈ S belonging to a subset S ⊂ P of a poset (P.. {y ∈ P | x . Proof: Use the antisymmetric law.* Lemma... Greatest and Least Elements. ) ∩ ( . Sets of the form [a. Similarly. Similarly.. l ∈ S is a least element of S if (∀s ∈ S) l ≤ s.. . b]. ≤) can have at most one greatest element. 3 .. b). g].d. . b] = [a. obtained by intersecting two open half-intervals are called open intervals. There are four kinds of intervals.. b]. Similarly l is the least element of S if l ∈ S and S ⊂ [l.The set of all elements in a poset (P.

Well-ordered Sets. Proof: This is an easy exercise. If f : P1 → P2 is a monotone mapping mapping from a poset (P1 . ≤) are isomorphic if there exists an isomorphism f : P1 → P2 . ) has no least element. ≤) to a poset (P2 .e. y ∈ P1 ) x ≤ y = f (x) ≤ f (y). for example. and if x ∈ S is the greatest element of some subset S ⊂ P1 . q. ≤). Proposition. The well-ordering of the natural numbers is equivalent to the principle of complete induction. There is no least positive rational or real number: if p is positive. The image of a chain under any monotone mapping is a chain. Proof: This is clear. One can derive the principle of induction from the well-ordering property.d. q. ≤) is said to be a poset isomorphism if f is one-to-one and onto. ≤) is said to be monotone if (∀x. ≤) to a poset (P2 . then the set {x. ≤) is well-ordered if every nonempty subset has a least element.d. ⇒ Monotone mappings are also known as poset homomorphisms or simply poset morphisms. y} has a least element. Not every inﬁnite chain is well-ordered. but not a wellordered set. ≤) to a poset (P2 . Example. The set R of real numbers with the usual ordering is a chain. Well-ordered posets are totally ordered: Proposition. 4 .e. A mapping f : P1 → P2 from a poset (P1 . A poset (W. and chains. Example.d.e. A mapping f : P1 → P2 from a poset (P1 . Proposition. ≤). the half-interval (0. Proof: If x. Monotone Mappings and Poset Isomorphisms. The same is true for the set Q of rational numbers. Any ﬁnite chain is well-ordered. The natural numbers with the usual ordering is a well-ordered set (ω. and both f and f −1 are monotone. then so is p/2. y ∈ W are elements of a well-ordered poset (W. For example. Hence x ≤ y or y ≤ x. then f (x) is the greatest element of the subset f [S] = {f (s) | s ∈ S}. Every well-ordered poset is a chain. q. Monotone mappings preserve greatest elements. and conversely. in both Q and R. Posets (P1 . It is of interest to investigate which concepts in the theory of partially ordered sets are preserved by monotone mappings. ≤). That is. any two elements of W are comparable. ≤) and (P2 .

. .. . Any poset (P. . .. .* Proposition. ...e.. . . .d.. For any poset (P. . . .e. . . . . b .. just as q. ≤) be the posets pictured below.... q. .. . . • f (b) .. . .. Corollary.... In the poset of natural numbers (ω. Construction: Let (P1 . .. . ⊂). .. .f. This marks the end of a construction.. . .. (P1 . . .. In the poset of rational numbers (Q . Hasse Diagrams and the Covering Relation in a Poset.. .Counterexample. ≤) and in the poset of real numbers (R. ≤) is ﬁnite if the set P has a ﬁnite number of elements.e. ≤) to the poset (Pow P. •. ... q.. .. A poset (P.d.. ≤) and (P2 .. . . .. Note that x ≤ y if there is a path leading upward from x to y... y) = ∅.e.. . . and the lines represent the covering relation.. . . the mapping x → ( . Proof: See the results of the section on Intervals and Half-Intervals. marks the end of a proof. .f. and there is no element strictly between x and y. . • • .. .. . . .. . .. with the mapping f as indicated.. . . An element x ∈ P is covered by an element y ∈ P in a poset (P.. ≤) is isomorphic to a poset whose members are sets. if x is strictly less than y. . That is. . Any ﬁnite poset (P.... a . partially ordered by inclusion. . . * The abbreviation q. .. .. .... Example. . .. we have (∀n ∈ ω) n < n + 1. . ≤). . . .. ≤)... which stands for quod erat demonstrandum. . no element Q R covers any other. •y . ≤) f . ≤). . ≤). . x < y if x < y and (x.. . . stands for quod erat faciendum. . . . x • • • Example of a Hasse Diagram... . . A one-to-one and onto monotone mapping need not have a monotone inverse. . x] is a one-to-one monotone mapping from the poset (P. .... ≤) can be represented by a Hasse diagram in which the elements of P are represented as vertices.. . . . ≤) The inverse f −1 is not monotone since f (c) ≤ f (a) does not imply c ≤ a.. .. . 5 . . . (P2 . c • f (a) • f (c) .. written x < y..

. .. The empty set and any set with a single element can be partially ordered in only one way. • • • • • There are sixteen posets with 4 elements.. .. . The number of nonisomorphic posets with n elements grows very rapidly with increasing n. . . . . 1975.... . . 1977. .. . . ... . .. .. Of the remaining two.. and the other is an antichain. . . Wilf..Examples of Finite Posets. A posets with two elements must be isomorphic to one of the two posets whose Hasse diagrams are depicted below. The number of connected Hasse diagrams is related to the total number of posets by a general formula. One is a chain... . Note also that three of these have connected Hasse diagrams.. . . . . .. . . . .. . each is isomorphic to the dual of the other... Combinatorial Algorithms.. ... .. 24.. . 79.. and another is an antichain. . . . . Journal of the Association for Computing Machinery. Academic Press.. • • • . Note that three of these posets are isomorphic to their own duals..* n 0 1 2 3 4 5 6 7 8 number of distinct posets 1 1 2 5 16 63 318 2045 16999 number of connected Hasse diagrams 1 1 1 3 10 44 238 1650 14512 Number of Nonisomorphic Posets * These numbers can be found in a paper by Shawpawn Kumar Das. .. . • • • • • • . p. . . . pp... .. .. 676–692. .. . 6 . . . . of which 10 have connected Hasse diagrams... .. . •. .. . and the remaining two are made up of disconnected pieces. . one of these is a chain. .. See for example: Albert Nijenhuis and Herbert S. . . Cf. chain • • • • antichain Show below are Hasse diagrams of the ﬁve types of posets with three elements. . Vol.... .

.... We are done if x = y. .. ... .. .. . . . . .. . . x .. . ... y.. .. y. .... . x ).. . .. ..Comparability and Covering. and so we are done in this case.. ... y be elements of a poset (P. .. . .. .. . . x < y.. .. . .... .. . if (∀x ∈ P ) m ≤ x = m = x.. ... . .. we may also assume without loss of generality that x ≤ y.. . then either x = y or x . .. . .. .... ..... .... ....e.... ... . x •... .. .. .. .. that is. or both... . that is. ......... .... . ... Minimal and Maximal Elements.... ...... At this point we have the picture: .. 7 ...... y ... ..) Let x.................................. x . . .... ... .... ..... then x < y < x ..... . .. . .. . .. ... ...... . ........ ..... ... x < y.. ... .. An element m ∈ P of a poset (P. .. . x .. •y x• •y ... x ≤ y. ≤) is said to be a maximal element if there is no greater element in P ... ... ........ . . ...... . . ....... . we are done if y ≤ x.. ... . . We shall now discuss a relation between the concepts of comparability and covering......... ... ... . then x ≤ y < y and hence x ... ... and if x < y... . Lemma (Comparability and Covering. x •. . .. y and x .. ... . .... ... ....... . ....... y.... .. ....... ... . . Proof: If x ≤ y. y..... .. . Dashed lines indicate assertions to be proved... ... ..... Similarly...... an element m ∈ P of a poset (P. .. . . ≤) is a minimal element if there is no lesser element in P . .. For this purpose it is convenient to introduce the following notation: x• x• x• x• < < ≤ •y •y •y •y means means means means . .. .. •y •y Because the picture is symmetric.. ..... . So we may as well assume that both x < y and y < x .. ...... ..... . ....... . .. . . . .. by symmetry.. . .... . ..... .. ≤)........ ⇒ Similarly.... . .... and if x ... y .. ... .. ........ .. q................ .. . ....... ........ ....... .. x• . ....... .. . . .. . .......... If x < x and y < y ..d.. .... .. ... ..... . ... ... ... .... .... violating the assumption that x < x . y ∈ (x...... . .. . . ... ... .. . .. .

we have a < f (a) < f (f (a)) < · · · . then f (m) would be strictly greater than m. then x is a maximal element of P1 . If (a. For if m were a maximal element. ) = {x}. ≤) is maximal if and only if [x. If S is any set. Similarly. Then a < b. and if P2 has a maximal element. ≤) has a minimal element. then the poset (ω.e. ≤) is a mapping f : P → P for which (∀x ∈ P ) x < f (x). and therefore has a least element. then every element of the poset (S. An element m ∈ P in a poset (P. then the open half-interval (a. If S holds more than one element. unlike greatest elements. q. in conjunction with Zorn’s lemma to be discussed later. q. m ∈ P is minimal if and only if ( . If W has no greatest element. then so does P1 . If a well-ordered set (W. then that element is the one and only maximal element in that poset.e.e. but we do have a weak result: Proposition. Minimal and maximal elements need not exist in general.e. Proof: Easy exercise. which is impossible. ≤) to a poset (P2 . Covering Operations.f. 1.Proposition. ≤). 8 . In general. The above result.d. there is no greatest or least element. q. A covering operation f on a poset (P. q. =) is both minimal and maximal. Example. ≤). Proof: If y = f (x) is a maximal element of P2 . do not behave well under monotone mappings. a poset with a covering operation f can not have a maximal element.d. Construction: If ω = {0. then a would be the greatest element of W . . 3.d. A nonempty poset with a covering operation must have inﬁnitely many elements. A covering operation need not be monotone. could possibly be used to prove existence of maximal elements in certain posets. Proof: Let a ∈ W be any element of a well-ordered set (W. Proof: This is clear from the deﬁnitions. if a is one of them. q. Proposition. If a poset has a greatest element. ≤) has no greatest element. A maximal element need not be a greatest element. x] = {x}. then there is a unique covering operation on W . The unique covering operation assigns a → b. An example of a covering operation is the successor operation n → n + 1 on the poset (ω.d. . but no maximal element. Proposition. ) is not empty. and clearly b is the only element which covers a. they need not be unique. . 2. ≤) of natural numbers. If they do exist. say b. Maximal elements. ) = ∅.} denotes the system of natural numbers. If f : P1 → P2 is a one-to-one and onto monotone mapping from a poset (P1 .e.

... .. ... .. ... .... ... . ... ...... . Similarly. . . . .. . . ...... ...d. .. . .. . .. .. .. A subset S ⊂ P of a poset (P.... ... . . ..... . b • S S • a Pictures of Upper and Lower Bounds. . ..... . .. . . . .. . ... .. .. . b) and ( . .... .. ).. . .. .. .... .. . . In general. . .. . .. ... .. .... [a... . . .... . . then x ≤ y < f (y). . ... . . .. ) are bounded below. ... .. . .. . . .... ... .. . .. .. . . . .. q. .. . . Also.. . . ....f. .. .. .. . . . .. .... .. .. .. . . . ... The intervals (a... . . . ... .. . . .. . . . .. . .. .. . . .. .. . ≤) is said to be bounded above if there exists an upper bound for S in P .. . .. These deﬁnitions can be restated as follows: Lemma.. An element b of a poset P is an upper bound for a subset S of P if and only if S ⊂ ( . . ..... ) and ( .. .. b] are all bounded subsets. . A subset S ⊂ P is bounded if it is bounded both above and below..... .. .. . .. The half intervals (a... . ). . .... . . .... .. .. .... . ..... . .... .. ... .... An element b ∈ P is called an upper bound for a subset S ⊂ P of a poset (P. a ﬁnite subset of a poset need not be bounded. ... . . =) fails to be bounded. . . .. ... .. .. .. b)... . ) and [a.. .... . b]. . ≤) if (∀s ∈ S) s ≤ b. .. . ). . . .. ... . . . .. .. .. b]....... .. . .. The half-intervals ( . . .. . ... . . ...Proposition.. Counterexample... .... ..... .. .. .. .. . .. Proof: This follows directly from the deﬁnitions of upper and lower bounds and the sets [a. .. . . ... . If x ∈ P is any element of a poset (P.. . Similarly.. . .. . . .. . .. .. . .. 9 ... .. .. ... ) is invariant under any covering operation on P . . ... .. . .. b) and [a. . .. . q.. ... The empty set is.. a bounded subset of every poset. . .. . . ...... .... any singleton is a bounded subset. . an element a in P is a lower bound for a subset S if and only if S ⊂ [a. ...... ..e.. ..e.. . .. ... . .. ..... . ... . . .. . . . Upper and Lower Bounds. ..... .. ... .. . Example.. . . . a subset S ⊂ P is said to be bounded below if there is a lower bound for S in P ...... ... . . . . . . . . (a. .d... . .. of course. . ... . . .. ... . .. ..... q. . .. .. . an element a ∈ P is called an lower bound for S ⊂ P if (∀s ∈ S) a ≤ s. . .. ...... . .. . .. . ... ... then the closed half-interval [x. Similarly... . . . . .. . .. . ... and thus f (y) ∈ [x. ... Proof: If y ∈ [x. ≤).. We sometimes draw pictures of upper and lower bounds as follows: . . . .. . . . Construction: Any subset with two or more elements in the poset (S..... . . . .. . . ...... .. . . b]. . .e... . . . b] are bounded above..

then f (b) is an upper bound for the image f [S]. would be directed upward. ≤) is directed if and only if for all x.d. then f (z) is an upper bound for f (x) and f (y). A poset is directed if and only if every set of two elements is bounded above. Proposition. What happens if we require not only that every ﬁnite subset is bounded above but that every subset is bounded above? We remark that in this case we have the following simple result. Since f is monotone. Proposition. If a directed poset has a maximal element. and if x is any element. Conversely. (A better term. y ∈ P .d. q. then that element is the greatest element. Let f : P1 → P2 be a monotone mapping from a poset P1 to a poset P2 . Proposition. there is an element z which is an upper bound for both m and x. then that greatest element is an upper bound for every subset of P . Proof: If b is an upper bound for the subset S.e.) Proposition. if P has a greatest element. Proof: Let f : P1 → P2 be monotone. 10 . then we must have m = z. there exists an element z in S which is an upper bound for both x and y. Indeed. then that upper bound must be the greatest element of P . Proof: Easy exercise. A ﬁnite poset is directed if and only if it has a greatest element. The image of a directed subset under a monotone mapping is a directed subset. Since S is directed. Hence m is an upper bound for every element x in P . ≤) has a greatest element if and only if every subset of P is bounded above. That is. perhaps. q. Proof: If m is a maximal element. The term directed downward could then be applied to the dual concept of a poset in which every ﬁnite subset is bounded below.d. A poset (P. A poset is said to be directed if every ﬁnite subset is bounded above. since s ≤ b for all s in S. then since the poset is directed. and let S be a subset of P1 . q. there exists an element z ∈ P such that x ≤ z and y ≤ z. Corollary.e. then f (s) ≤ f (b). Any two elements of f [S] can be represented as f (x) and f (y) where x and y belong to S. Proof: If P itself has an upper bound.e. q. Since m is maximal. Bounded subsets are well-behaved under monotone mappings. q. Proposition. then its image f [S] ⊂ P2 under f is bounded above in P2 . If a subset S ⊂ P1 is bounded above in P1 .e.e. a poset (P.Directed Posets.d.d.

.. q. . . then ( . If y < x. . . . .. x) = ( ...... Construction: Let R be the set of all real numbers. x).. Similarly.. If a set S ⊂ P has a least upper bound.e... x) = x for all x ∈ R. so g is an upper bound for S. If a subset S ⊂ P of a poset (P.. More generally: Proof: Clearly x is an upper bound for the half-interval ( ... ≤) is a chain.. . . .. denoted by sup S and the greatest lower bound of S is called the infemum. .. . .. . ... elements c and d are both minimal upper bounds for the set S = {a. . .. if n > 0.. •. then since g ∈ S. ... then z ∈ ( . . in the poset (ω. x) ⊂ ( . .. . . . If a set S ⊂ P has a greatest lower bound. ... . . x) = x. That is. . then ( ... ... . .. .e.. .Least Upper and Greatest Lower Bounds... .. . .. An element b ∈ P is the least upper bound of a subset S ⊂ P if b is the least element of the set of all upper bounds for S.. Hence (x.. If y ∈ C is another upper bound for the half-interval ( ... but there is no least upper bound. . ordered in the usual way. . ... ≤). . ....e. A minimal upper bound need not be a least upper bound.. ≤). • • d a • b Proposition.. .. . In some books. For example. . Proposition. .. g]. . . If x ∈ P is any element of a poset (P.. . x] = x. If z ∈ (x. then ( . .. .. .. . . either y < x or x ≤ y. ... 11 Proposition. . . . the least upper bound of S is called the supremum. . y]..d.. Hence x ≤ y. . . . ......... y]. .. . .... . . . .. . Since (C. we have ( . y) = ∅. the results are more complicated. ... we denote it by glb S or S. Let f : R → R be the staircase function deﬁned by (∀x ∈ R) f (x) = x = greatest integer n such that n ≤ x. ... .. .. If b is also an upper bound for S. ... q. ≤) is a chain in which no element covers another.. then (∀x ∈ C) ( . .d. . Construction: In the poset depicted below. then ( .. then g = S.. .... Let (P. Proof: Clearly S ⊂ ( . ...... b}. . q.. For open half-intervals. .. an element is the greatest lower bound of a subset S if it is the greatest element of the set of all lower bounds for S. denoted by inf S. Hence g is the least element of the set of all upper bounds for S. . n) = n − 1.. ≤).. ... . contrary to the hypothesis that no element covers another in (C. which is / absurd. . c . . . Monotone mappings need not preserve least upper bounds. ≤) of the natural numbers.f. . .d.. .. . ≤) be a poset. q. . x) and z ∈ ( . . whereas in the poset (R. x). Proof: This is clear. . . .. x).... ≤) has a greatest element g. . . R where R is the set of all real numbers. Counterexample. . x is the least upper bound of the half-interval ( .e.. .... . Counterexample. we denote it by lub S or S.. If (C.. and so x < y. . . ..... .. . . . ..... . y). . . y]. .. we must have g ≤ b..

≤) is a meet-semilattice if for any pair of elements x. ≤) is a join-semilattice. The associative law follows from the lemma. there is a least upper bound in P for the set {x. Similarly. q. q. y} = {y. Since {x. q. y}. then S = 1 and f [S] = {0}. Semi-Lattices.e. A poset (P.e. ≤) is a join-semilattice if for any pair of elements x. y ∈ P . then x ∨ y = y ∨ x. (∀x. We write x ∧ y = {x. The symbol ∨ is called join or cup. Example. ≤) is a join-semilattice. Proposition. If (P. then Idempotent Law: (∀x ∈ P ) x ∨ x = x. If S = [0. y}. with ∨ and ∧ being the set-theoretic union and intersection. y}. z ∈ P ) (x ∨ y) ∨ z = x ∨ (y ∨ z). Any join-semilattice is a directed poset. 1). then x ∨ x = x. then (∀x. y. x}. A poset (P. whereas f ( S) = f (1) = 1. respectively. y}. Proof: This is clear. If (P. Associative Law: Proof: Since x is obviously the greatest element of {x. y ∈ P ) x ≤ z & y ≤ z = x ∨ y ≤ z. ⊂) is both a joinsemilattice and a meet-semilattice.e. The symbol ∧ is called meet or cap. there is a greatest lower bound in P for the set {x. We write x ∨ y = {x. Lemma. y}. y ∈ P . Commutative Law: (∀x. 12 . then (Pow S.d. and Pow S is its power set. Theorem.d.e. x}.d. q. (∀x.f. so that Thus f [S] = f ( S). ⇒ Proof: The element x ∨ y is the least upper bound of {x. y ∈ P ) x ∨ y = y ∨ x. If S is any set. y ∈ P ) x ≤ x ∨ y & y ≤ x ∨ y.3 2 1 • f (x) • • • x 1 2 3 f [S] = 0.

one can view a lattice as a set L equipped with two binary operations ∨ and ∧ subject to six axioms: the two commutative laws. y ∈ L) x ∨ (x ∧ y) = x. Hafner. Lattice Theory. Proposition. A poset (L. G. ≤) is a lattice if it is both a join-semilattice and a meet-semilattice. That is. and some of them treat posets as well. a These are just a few of the classics. then x ∧ y = x and x ∨ y = y. Theorem (Absorptive Laws). Note that. but most deal with special topics. A comment on axiomatics is in order. x ∧ y}. then x ∧ y = y and x ∨ y = x. If (L. Szasz. y ∈ C then either x ≤ y or y ≤ x. q. There are lots of books on lattices. Any lattice is a directed poset. x = x ∧ [x ∨ (x ∧ x)] = x ∧ x. the idempotent laws need not be explicitly postulated since they actually follow from the absorptive laws. Since x ∨ (x ∧ y) is the least upper bound of this same set. Any chain (C. y ∈ L has both a least upper bound x ∨ y ∈ L and a greatest lower bound x ∧ y ∈ L. Corollary. G. By the antisymmetric property of partial ordering. y ∈ L) x ∧ (x ∨ y) = x. Incidentally.d. One can then reintroduce the partial ordering ≤ on L by deﬁning x ≤ y to mean x = x ∧ y. there are lots more! 13 . Second edition: 1948. Academic Press.e. the two associative laws. (∀x.d.Lattices. Birkhoﬀ. ≤) is a lattice. The second statement is proved in a similar fashion. Alternatively. then x ∨ (x ∧ y) ≤ x. We have deﬁned a lattice as a certain type of poset. Proof: If x. Gr¨tzer. First edition: 1940. any pair of elements x. ≤) is a lattice. 1965. the statements x = x ∧ y and x ∨ y = y are equivalent. E. Since x ≤ x and x ∧ y ≤ x. then (∀x. on account of the absorptive laws. Rutherford. We may summarize this as: ⇒ ⇒ ⇒ ⇒ well-ordered set = chain = lattice = join-semilattice = directed poset.e. q. Bibliography. Third edition: 1961. D. Proof: Clearly x ≤ x ∨ (x ∧ y). and the two absorptive laws. Introduction to Lattice Theory. G. New York. Introduction to Lattice Theory. Lattice Theory. If y ≤ x. 1963. 1971. Indeed: x = x ∨ [x ∧ (x ∨ x)] = x ∨ x. If x ≤ y. then x is an upper bound for the set {x. it follows that x ∨ (x ∧ y) = x.

there is no upper bound at all. L is the greatest lower bound for S. y ∈ L) f (x ∧ y) = f (x) ∧ f (y). the poset (Z.A monotone mapping f : L1 → L2 from one lattice to another is called a lattice homomorphism if (∀x. Proposition. y ∈ L) f (x ∨ y) = f (x) ∨ f (y). with the usual order. then A= A. If A is not empty. . Theorem. The converse is also true. we have then A = S. Any element s ∈ S is an upper bound for L. q. ⊂) is a complete lattice. .d.d. The empty set and inﬁnite subsets.d. a monotone mapping need not preserve the lattice operations. but neither ∅ nor Z has a least upper bound. ≤) is a complete lattice if every subset S ⊂ P has a least upper bound S ∈ P . We say that a poset (P. If every subset S of a poset (P. For example. It follows that L ∈ L. Proposition. Proof: Any ﬁnite nonempty subset of a lattice has a least upper bound and a greatest lower bound. any nonempty ﬁnite set {x1 . . Complete Lattices. in the case of Z . If A = ∅. q. however. . .e. Any ﬁnite nonempty lattice is complete. ∅ is 14 .e.e. For any set S. x2 . . In general. . that is. Z and therefore a lattice. xn } = (· · · ((x1 ∨ x2 ) ∨ x3 ) ∨ · · · ∨ xn ). the poset (Pow S. every integer is an upper bound. Proof: If A ⊂ S. Corollary. Proof: Since any element x ∈ L is an upper bound for ∅. q. . the existence of greatest lower bounds could be omitted from this deﬁnition in view of the following theorem. if every subset has a greatest lower bound.d. In a lattice. is a chain. A= A. . but there is no least integer. . . the least upper bound the least element of L. ∅ is the greatest element of L. q. (∀x. S. .e. ≤) has a least element and a greatest element. xn } has a least upper bound and a greatest lower bound. Similarly. that is. We say that f preserves the lattice operations ∨ and ∧. need not have least upper and greatest lower bounds. Actually. then every Proof: The set L of all lower bounds for S has a least upper bound L ∈ P . the least upper bound of a set S is the greatest lower bound U of the set U of all upper bounds of S. xn } = (· · · ((x1 ∧ x2 ) ∧ x3 ) ∧ · · · ∧ xn ). x2 . then every subset has a least upper bound. The proof is similar. {x1 . ≤) of all the integers. ≤) has a least upper bound subset S has a greatest lower bound S. L is the greatest element of L. and a greatest lower bound S ∈ P . and hence (∀s ∈ S) L ∈ P ≤ s. A complete lattice (L. In the case of ∅. because {x1 .

Therefore S ∈ S. we have S = T = S. S is invariant under f if (∀x ∈ S) f (x) ∈ S. q. Any closed interval [a. The interior of any subset A ⊂ S is deﬁned to be the union of the collection of all open subsets contained in A. b] on the rational line (Q . the Bourbaki ﬁxed point theorem uses Zorn’s lemma. Kelley’s book. has no least upper bound in the closed interval [0.d. A. 2] ⊂ Q.e. ⊂) is a complete lattice. 5. ≤) has a ﬁxed point. 15 . Since S is invariant. C. If T is a topology on a set S. x is a ﬁxed point of f if and only if the singleton {x} is invariant under f . It follows that S is a ﬁxed point of f .d. ≤) is a complete lattice if a < b. Davis proves that the converse is also true. Another ﬁxed point theorem (due to Bourbaki) will be proved later. ≤) is not a complete lattice if a < b. Davis. In particular. but rather the interior of the intersection. Hence f ( S) ≤ S. Proof: The subset S = {x ∈ L | x ≤ f (x)} is invariant under f because x ≤ f (x) = ⇒ f (x) ≤ f (f (x)). A subset S ⊂ L is invariant under an operation f if f [S] ⊂ S. This deﬁnition of a topology is essentially equivalent to that given on page 37 in John L. Hence f ( S) is an upper bound for S. Paciﬁc J. 311–319. 1955. Since L is a complete lattice. because 2 is irrational. 2] A somewhat more interesting example is provided by topology. Proof: If S ⊂ T . The Complete Lattice Fixed Point Theorem. An element x ∈ L is a ﬁxed point for an operation f : L → L if f (x) = x. (∀S ⊂ T ) S ∈ T . called open subsets such that S ∈T. Recall that a topology T on a set S is a collection of subsets of S. Proposition. Note that if S is not empty. then (T . (∀G1 . General Topology. Q √ For example. so that x ≤ f (x) ≤ f ( S). q.R Example. A Characterization of Complete Lattices. When S = ∅. then ∅ = ∅ ∈ T . A closed interval [a. and S ≤ f ( S). then S ∈ T is the least upper bound for S.e. Note that since ∅ ⊂ T . that is. G2 ∈ T ) G1 ∩ G2 ∈ T . Theorem. Math. then f ( S) ∈ S. the set {x ∈ Q | x > 0 & x2 < 2} ⊂ [0. b] on the real line (R. then the greatest lower bound S is not the intersection S. Any monotone operation f : L → L on a complete lattice (L. then S ∈ L and (∀x ∈ S)x ≤ S.

A subset H ⊂ P of a poset (P. A subset S ⊂ L of a lattice (L. This concept is interesting mainly in inﬁnite posets in view of the following theorem. y ∈ S) x ∨ y ∈ S. The intersection of any nonempty collection of sublattices of a lattice is a sublattice. The ideals ∅ and L are called improper ideals of L. y ∈ I) x ∨ y ∈ I. x] are called principal ideals of L. ≤) is directed if (∀x. (∀x. An ideal is clearly a sublattice. q. The empty set ∅ and L itself are ideals of any lattice (L. A subset I ⊂ L of a lattice (L. Proposition. ≤). Proposition. ≤) is hereditary if (∀h ∈ H) ( . 16 . ≤). Proof: We leave the proof as an easy exercise. h] ⊂ H. x] is an ideal of L.e. ≤) is an ideal if (∀x ∈ I)(∀y ∈ L) x ∧ y ∈ I. The concept of a hereditary directed subset makes sense in any poset. That is. thereby generalizing the concept of an ideal in a lattice. ≤) is a sublattice if (∀x. The ideals ( . then the closed half-interval ( . Example.Filters and Ideals in Lattices. ≤) is an ideal of L if and only if I is directed and hereditary.d. (∀x. Ideals can be characterized in another way which does not make use of the lattice operations. H ⊂ P is hereditary if and only if (∀x ∈ P )(∀h ∈ H) x ≤ h = x ∈ H. Example. but a sublattice need not be an ideal. A subset I ⊂ L of a lattice (L.d. q. If x ∈ L is any element of a lattice (L. y ∈ S) x ∧ y ∈ S. y ∈ D)(∃z ∈ D) x ≤ z & y ≤ z. Proof: This is elementary. ⇒ Recall also that a subset D ⊂ P of a poset (P.e.

The mapping x → ( . Every nonempty hereditary directed subset of a ﬁnite poset (P. ≤). x] is a one-to-one monotone mapping which allows us to embed any lattice (L.. Indeed. Every chain can be monotonically embedded in a complete chain. ≤) is a dual ideal if (∀x ∈ D)(∀y ∈ L) x ∨ y ∈ D. C is an ideal and this ideal is the Note that if C is a collection of ideals. A subset D of a lattice (L.. then there exists an . partially ordered by inclusion. x ∈ J . x]. b] ⊂ R is a complete chain. ≤). A proper dual ideal in a lattice L is called a ﬁlter. then H = ( . ≤) is a closed half-interval ( .e. y and .. then ( . The set of all ideals of a lattice (L. A comment is in order here. Corollary. is a complete lattice. we must have y < x. this completion proceQ dure does not yield the real line (R... For example.Proposition. Proof: If I and J are ideals of C.. then the least upper bound C need not be the union C. If a hereditary subset H has a greatest element x. q.. (Is this mapping a lattice homomorphism?) By a complete chain we mean a chain which is a complete lattice. but instead C is the intersection of the set of all ideals which contains the union.. . (∀x. Proof: Use the mapping x → ( ...e. Proof: If C is a nonempty collection of ideals.. If y ∈ J . element x ∈ I which does not belong to J . Since y is an arbitrary element of J .. . . Proposition.d... ≤). 17 ... then J ⊂ ( . ≤) in its lattice of ideals. y] ⊂ J and since x . A maximal ﬁlter is called an ultraﬁlter. q. the real line fails to be a complete chain R because it lacks a least and greatest element.e.. then greatest lower bound for C... x]. any lattice can be embedded in a complete lattice. Also ∅ = L. The ﬁlters of a lattice are partially ordered by inclusion.d. x]. Proof: Any ﬁnite directed subset has a greatest element. The lattice of ideals of a chain (C. y ∈ D) x ∧ y ∈ D. / q. x) ⊂ I. ≤) is a complete chain. Proposition. any closed interval on the real line [a.e. Thus. q. and if I is not contained in J .d.. Starting with the rational line (Q ..d. .. . . An dual ideal is clearly a sublattice..

with the usual ordering. then C itself must have an upper bound b ∈ C.c.e. Construction: The closed interval [0.c.d. The importance of the concept of a strictly inductive poset is due to the following: 18 . . with the usual ordering. Counterexample. is an example of an inﬁnite Noetherian poset. Example. A closed interval of the rational numbers. Chapter III. but the subset {0. We say that a poset (P. The same term is used with a similar meaning in ring theory. The term Noetherian is used in the present sense by Bourbaki in Theory of Sets. ≤) satisﬁes the descending chain condition. Proof: If (C.5.c. any closed interval [a. and C ∈ S. for example. A chain is inductive if and only if it has a greatest element. The term strictly inductive is my own invention. abbreviated d. . b] ⊂ R with a < b is strictly inductive..c. The set of negative integers. 3/4. The following implications are clear: ﬁnite = Noetherian = strictly inductive = inductive. Proposition. A subset of an inductive set need not be inductive. q. and is therefore not inductive.c. Finally. ≤) is Noetherian or satisﬁes the ascending chain condition (often abbreviated as a. Any poset with a greatest element. is an example of an inductive poset which is not strictly inductive. We shall say that a subset S ⊂ P of a poset (P. we say that a subset S ⊂ P of a poset (P. if every nonempty chain C ⊂ P has a least element. 1] is a chain with no greatest element. ≤) is strictly inductive if every nonempty chain C ⊂ S has a least upper bound in P . The above terminology for chain conditions on a poset is not completely standard. ≤) is an inductive chain. The connection between ordinary mathematical induction and inductive sets is slightly topsy turvy. q. The term inductive is used with the present meaning by Bourbaki in Theory of Sets.c. A well-ordered set is a chain which satisﬁes the d. Section 6. if every nonempty chain C ⊂ S has a greatest element. of course. of course. is not even an inductive set because it is a chain with no greatest element. Since x ≤ b for all x ∈ C. The set of positive integers. We need to turn the integers upside down: Example.e. Any complete lattice is strictly inductive. the term inductive set is also used in ordinal number theory with a diﬀerent meaning which need not concern us here.f. ⇒ ⇒ ⇒ Note that the empty set satisﬁes all three conditions. However.} ⊂ [0. (This is precisely the hypothesis of Zorn’s lemma.4.) We say that a subset S ⊂ P of a poset (P. 1] ⊂ R is inductive. an abelian ring is Noetherian if its lattice of ideals satisﬁes the ascending chain condition. Example.Various Chain Conditions. ≤) is inductive if every nonempty chain C ⊂ S has an upper bound in S. is inductive. Section 2. Chapter III. the element b is the greatest element of C. 1/2. . 7/8.

and if its spine S = Spine P is not empty. and it follows that C is a chain in P . bigger one of these chains. . possibly empty.. either C1 ⊂ C2 or C2 ⊂ C1 .. . Theorem. If a poset (P..d... ≤) is a spinal element if it is comparable with every element in P . Example. If x ∈ P . We shall say that an element x ∈ P of a poset (P.. . Example. Proof: We leave this as an exercise... We shall prove a theorem about the spine of a strictly inductive poset.. ≤) is a strictly inductive poset. that is.e. y}. ) are strictly inductive.. For this we need the following lemma.. x] ∪ [x.. . Lemma. ). Clearly . q. ) is the union of two chains C = C+ ∪ C− where C+ ⊂ [x. Proposition. x] and [x.. If a poset (P. but will prove useful in connection with Zorn’s lemma. Proof: If (P. Hence S ∈ S... In either case. and hence S ∈ ( .e.e. to be discussed later.. Hence . and if x ∈ P . we denote it by . .. ≤) is strictly inductive. If x. partially ordered by inclusion. Since C is a chain.e... The set Ch P of all chains in a poset (P... x] ∪ [x... . .d.. If C+ is not empty. Proof: Suppose C is a nonempty chain of chains. ≤) is the set of all spinal elements. .. is a strictly inductive poset... The spine of any poset is a chain. then there exist chains C1 ... y ∈ C. The following terminology is not standard. x] ∪ [x. If (P.. Spine P = {x ∈ P | (∀y ∈ P )x . ) is a strictly inductive subset.. .. then S ∈ P . and if x ∈ P ...d.. C2 ∈ C with x ∈ C1 and y ∈ C2 . x]. ..... then the subsets ( .. then C = C+ ∈ [x. x]. q.. then S ⊂ ( .e.. . The spine of a poset (P.. Proof: We leave this as an exercise. The intersection of any nonempty collection of strictly inductive subsets is strictly inductive.. Spinal Elements. ≤) has a least or greatest element.. . ≤) is a strictly inductive poset. Hence x . . x] ∪ [x.. Proof: Any nonempty chain C ⊂ ( ... q.. The spine of a strictly inductive poset is either empty or has a greatest element. Any chain (C..Theorem. then these are spinal elements. Proposition.. then ( . (∀x ∈ P ) x . then C = C− ∈ ( . x] ∪ [x. ). . ⊂). C is a nonempty chain in the poset (Ch P.. If C+ = ∅.. q. ).... 19 . ). Note that x ∈ Spine P if and only if P = ( . ... C is the least upper bound for the chain C ⊂ Ch P ... ≤). ) and C− ⊂ ( .d. both x and y belong to the .d. q. .... S. ≤) is its own spine. y. .

. Lemma....) “Towers topple..... . We proved earlier that the spine of a non-empty strictly inductive poset has a greatest element.... Then [b.... Since M is the smallest f -tower to which b belongs... .... Clearly b is the least element of M . .. . We will prove theorem due to Zermelo which asserts that f -towers do not exist. ..d.. In this section we shall prove a theorem due to Zermelo which is a close relative of Zorn’s lemma.. .. ...... .. .. .... . then the set of elements comparable to f (x) is invariant under f . ... ≤) has no covering operations.. Since M and ( . ... . .... but which does not require the axiom of choice.. Lemma. This minimal f -tower is contained in any f -tower to which b belongs... • f (y) x• Proof: We have to show that •y .. . ... .. . A nonempty strictly inductive poset (P. .. .d. ..... .. .... .... q... then M ⊂ T . so is their intersection T . . f (g)] ∪ [f (g). Proof: Invariant subsets and strictly inductive sets are preserved under intersections...” For this we need the following lemma.. ..e.. ... This is absurd because g is the greatest spinal element of M and f (g) is strictly greater than g. ≤)... ... f (x) •. Since T ⊂ M and M ⊂ T . ) are both strictly inductive.e... ....... .. .......... .. . We begin with a technical lemma which will be needed later.. ) is an f -tower. . ... . this follows from the diagram.. . so the spine S of M is not empty. . . . . . ... . ... . f (x) = f (y) . ⇒ (∀y ∈ P ) y . ... f (x). . Note that b ∈ T since b ≤ g < f (g).. .. q... . .. Hence f (g) is a spinal element of M ...... . By a technical lemma proved earlier. . ≤).. the set T ⊂ M of all elements in M which are comparable to f (g) is invariant under f .......A Theorem due to Zermelo.. .. ... Thus T is an f -tower to which b belongs. ).. A strictly inductive subset of a poset which is invariant under a covering operation f is called an f -tower. ≤). . ).. 20 ... .. .. .... If f is a covering operation on a poset (P... . .. (except in the trivial case of the empty set.. . ... ...... Proof: Suppose f were a covering operation on (P. . and b ∈ P . Let f be a covering operation on a poset (P.. and if x ∈ P is a spinal element of P .. .. Theorem (Zermelo)... then T = M .....d. .. .e.. .. q. . .... ... .. The intersection M of all f -towers to which b belongs is an f -tower. .. . Since x is a spinal element.......... . .. Let g ∈ S be the greatest element of the spine S... .... .. . and b ∈ [b. .... The intersection of any nonempty set of f -towers is itself an f -tower. so M ⊂ [b. .... .. .. and b ∈ M ..

.. ≤) having a maximal element need not be inductive. Hence there is a maximal element in [C. . Theorem (Zorn’s Lemma)... . say x.. . If C ⊂ P is a chain in P . Any antichain in a poset is contained in a maximal antichain. Up to now. (Note that if C is the empty set. .) Since x is not a maximal element of P .e.. . Hence the poset of all antichains in a given poset is strictly inductive... ) ⊂ Ch P is a strictly inductive subset of Ch P . . .. . ⊂) is strictly inductive.. .. . A poset (P. 2. Proof: If x is a maximal element of P . then the half-interval [C. Every nonempty inductive poset has a maximal element. the element m is maximal.d. q.. . ). there exists a strictly larger element. . . there exists an upper bound for C.. q.. . ... .. . Corollary (Bourbaki Fixed Point Theorem. .. Therefore C ∪ {y} is a chain. then by the hypothesis that P is inductive. ⊂) of all chains in P .d. . . . then x ≤ f (x) implies that f (x) = x. Proof: Suppose (P.) Any operation f : P → P on a nonempty inductive poset P satisfying (∀x ∈ P ) x ≤ f (x) has a ﬁxed point. then any element of P is an upper bound for C.. . . q. . whose proof does make essential use of the axiom of choice... . This is a contradiction.. •3 •2 m• •1 •0 Corollary.e..Zorn’s Lemma. .e.. q. ... .. . . Corollary. . .d. . . . From it we can deduce a more useful result. . . .d. but the chain {0.. say y. ..d. If C ∈ Ch P . ⊂) of all chains in a poset (P. Any chain in a poset is contained in a maximal chain. The converse of Zorn’s lemma is not valid. . .. . x] and x < y. by Zermelo’s theorem. . . ... Proof: The union of any chain of antichains is an antichain. 21 q. Since C ⊂ ( . . ≤) is a nonempty inductive poset which has no maximal element. we have not made use of the axiom of choice.... Then f is a covering operation on the poset of chains in P .. ... Clearly the / chain C is covered by the chain C ∪ {y} in the poset (Ch P. Proof: The poset (Ch P. Since the poset (Ch P. .. . 1. Zermelo’s theorem on the nonexistence of covering operations on a strictly inductive poset does not require the axiom of choice. .} has no upper bound.e.. . there can be no covering operation on (Ch P. there is a function f : Ch P → Ch P which assigns to each chain C ∈ Ch P some such chain f (C) = C ∪ {y} which covers C. ... then y ∈ C. called Zorn’s lemma.. By the axiom of choice. .. . It is not empty since ∅ ∈ Ch P . . ⊂)... ≤) is strictly inductive.. Construction: In the poset shown below. . Counterexample.. .e. .