AMERICAN MATHEMATICAL SOCIETY COLLOQUIUM PUBLICATIONS VOLUME XXV LATTICE THEORY BY GARRETT BIRKHOFF PUBLISHED BY THE AMERICAN MATHEMATICAL SOCIETY ProvipEnce, RuopE Isr4NpFirst Edition, 1940 Second (Revised) Edition, 1948 Third Edition, Second Printing, 1973 COPYRIGHT, 1940, 1948, 1967, BY THE AMERICAN MATHEMATICAL SOCIETY All Rights Reserved No portion of this book may be reproduced without the written permission of the publisher. Library of Congress Catalog Card Number 66-23707 Printed in the United States of AmericaPREFACE TO THE THIRD EDITION The purpose of this edition is threefold: to make the deeper ideas of lattice theory accessible to mathematicians generally, to portray its structure, and to indicate some of its most interesting applications. As in previous editions, an attempt is made to include current developments, including various unpublished ideas of my own; however, unlike previous editions, this edition contains only a very incomplete bibliography. Iam summarizing elsewhere} my ideas about the role played by lattice theory in mathematics generally. I shal] therefore discuss below mainly its logical structure, which I have attempted to reflect in my table of contents. The beauty of lattice theory derives in part from the extreme simplicity of its basic concepts: (partial) ordering, least upper and greatest lower bounds. In this respect, it closely resembles group theory. These ideas are developed in Chapters 1-V below, where it is shown that their apparent simplicity conceals many subtle variations including for example, the properties of modularity, semimodularity, pseudo-complements and orthocomplements. At this level, lattice-theoretic concepts pervade the whole of modern algebra, though many textbooks on algebra fail to make this apparent. ‘Thus lattices and groups provide two of the most basic tools of “ universal algebra”, and in particular the structure of algebraic systems is usually most clearly revealed through the analysis of appropriate lattices. Chapters VI and VII try to develop these remarks, and to include enough technical applications to the theory of groups and loops with operators to make them convincing. A different aspect of lattice theory concerns the foundations of set theory (including general topology) and real analysis. Here the use of various (partial) orderings to justify transfinite inductions and other limiting processes involves some of the most sophisticated constructions of all mathematics, some of which are even questionable! Chapters VIIT—XII deseribe these processes from a lattice- theoretic standpoint. Finally, many of the deepest and most interesting applications of lattice theory concern (partially) ordered mathematical structures having also a binary addition or multiplication: lattice-ordered groups, monoids, vector spaces, rings, and fields (like the real field). Chapters XIII-XVII describe the properties of such systems, +G. Birkhoff, What can lattices do for you?, an article in Trends in Lattice Theory, James C. Abbot, ed., Van Nostrand, Princeton, N.J., 1967. iiiv PREFACE and also those of positive linear operators on partially ordered vector spaces. ‘The theory of such systems, indeed, constitutes the most rapidly developing part of lattice theory at the present time. The labor of writing this book has been enormous, even though I have made no attempt at completeness. I wish to express my deep appreciation to those many colleagues and students who have criticized parts of my manuscript in various stages of preparation. In particular, I owe a very real debt to the following: Kirby Baker, Orrin Frink, George Gratzer, C. Grandjot, Alfred Hales, Paul Halmos, Samuel H. Holland, M. F. Janowitz, Roger Lyndon, Donald MacLaren, Richard 8. Pierce, George Raney, Arlan Ramsay, Gian-Carlo Rota, Walter Taylor, and Alan G. Waterman. My thanks are also due to the National Science Foundation for partial support of research in this area and of the preparation of a preliminary edition of notes, and to the Argonne National Laboratory and the Rand Corporation for support of research into aspects of lattice theory of interest to members of their staffs. Finally, I wish to thank Laura Scblesinger and Lorraine Doherty for their skillful typing of the entire manuscript.TABLE OF CONTENTS CHAPTER I TYPES OF LATTICES CHAPTER IT POSTULATES FOR LATTICES . CHAPTER IIT STRUCTURE AND REPRESENTATION THEORY CHAPTER IV GEOMETRIC LATTICES . ‘ . CHAPTER V COMPLETE LATTICES CHAPTER Vi UNIVERSAL ALGEBRA CHAPTER VII APPLICATIONS TO ALGEBRA. CHAPTER VIII TRANSFINITE INDUCTION... : . . CHAPTER IX APPLICATIONS TO GENERAL TOPOLOGY . . CHAPTER X METRIC AND TOPOLOGICAL LATTICES. CHAPTER XI BOREL ALGEBRAS AND VON NEUMANN LATTICES CHAPTER XII APPLICATIONS TO LOGIC AND PROBABILITY v iu 132 169 180 2u1 230 254 277vi CONTENTS CHAPTER XII LATTICE-ORDERED GROUPS . CHAPTER XIV LATTICE-ORDERED MONOIDS CHAPTER XV VECTOR LATTICES . CHAPTER XVI POSITIVE LINEAR OPERATORS CHAPTER XVII LATTICE-ORDERED RINGS BIBLIOGRAPHY INDEX 287 319 347 380, 397 411CuarTzr 1 TYPES OF LATTICES 1. Posets; Chains Pure lattice theory is concerned with the properties of a single undefined binary relation <, to be read “is contained in”, “is a part of”, or “‘is less than or equal to”. This relation is assumed to have certain properties, the most. basic of which lead to the following concept of a “partially ordered set”, alias “partly ordered set” or “ poset”. Dermrmon. A poset is a set in which a binary relation x $ y is defined, which satisfies for all x, y, z the following conditions: Pl. Forallz,zSz. (Reflexive) P2. Ife x. The above notation and terminology are standard. There are countless familiar examples of partly ordered sets—i.e., of mathe- matical relations satisfying P1-P3. Three of the simplest are the following. Example 1. Let D(I) consist of all subsets of any class I, including I itself and the void class @; and let x < y mean x is a subset of y. Example 2. Let Z* be the set of positive integers; and let x < y mean that x divides y. ; Example 3. Let F consist of all real single-valued functions f(z) defined on -1S 2 ljand let f 2 g mean that f(x) 2 g(x) for every x with -1 SxS 1. We now state without proof two familiar laws governing inclusion relations, which follow from P1-P3. Lemma 1. In any poset, x < x for no x, while x < y and y < z imply x < z. Conversely, if a binary relation < satisfies the two preceding conditions, define x Sy to mean that x < y or x = y; then the relation < satisfies P1-P3. In other words, strict inclusion is characterized by the anti-reflexive and transi- tive laws. It is easily shown that a poset P can contain at most one element a which satisfies a S x for all ze P. For if a and b are two such elements, then a @ from a poset P to a poset Q is called order-preserving or isotone if it satisfies a) 2S y implies (2) < Oy).§2 ISOMORPHISM; DUALITY 3 An isotone function which has an isotone two-sided inverse is called an isomorphism. In other words, an isomorphism between two posets P and Q is a bijection which satisfies (1) and also a’) Oz) S Oy) implies x < y. An isomorphism from a poset P to itself is called an automorphism. ‘Two posets P and @ are called isomorphic (in symbole, P ~ Q), if and only if there exists an isomorphism between them. The converse of a relation p is, by definition, the relation p such that xpy (read, ‘‘ is in the relation 3 to y”) if and only if ypr. Thus the converse of the relation ‘‘includes” is the relation “‘is included in”; the converse of “greater than” is “less than”. It is obvious from inspection of conditions P1-P3 that TuzoreM 2 (Duatary Princreue). The converse of any partial ordering is iteelf a partial ordering. Dermirion. The dual of a poset X is that poset X defined by the converse par- tial ordering relation on the same elements. Since X ~ X, this terminology is legitimate: the relation of duality is symmetric. Dermrrion. A function 6: P — Q is antitone if and only if (2) xy implies 6(z) = Oy), (2') x) $ Hy) implies x2 y. A bijection (one-one correspondence) which satisfies (2)-(2') is called a dual We shall refer to systems isomorphic with X as “dual” to X. Obviously posets are dual in pairs, whenever they are not self-dual. Similarly, definitions and theorems about posets are dual in pairs whenever they are not self-dual; and if any theorem is true for all posets, then so is its dual. ‘As we shall see later, this Duality Principle applies to algebra, to projective geometry, and to logic. Many important posets are self-dual (i.e., anti-isomorphic with themselves), ‘Thus Example 1 of §2 is self-dual; the correspondence which carries each subset into its complement is one-to-one and inverts inclusion. Similarly the set of all linear subspaces of n-dimensional Euclidean space which contains the origin is self-dual: the correspondence carrying each subspace into its orthogonal comple- ment is one-to-one and inverts inclusion. In these cases the self.duality is of period two: the image (2’)' of the image 2’ ofany zisagaine. Such self-dualities (dual automorphisms) are called involutions. Exercises for §§1-2: 1. Prove Lemma 1. 2. Prove Lemme 2, 3. Show that there are exactly three different ways of partly ordering a set of two elements. 4. (a) Show that there are just two nonisomorphic posets of two elements, both of which are self-dual.4 TYPES OF LATTICES I (b) Show that there are five nonisomorphie posets of three elements, throe of which are self-dual. +5. (a) Let Gin) donoto the number of nonisomorphic posets of 2 elements, Show that G4) = 16, G(5) = 63, (6) = 318. (I. Rose - R. T. Sasaki) (0) @*(n) denote the number of different partial orderings of n elements. Show G*(2)=3, G*(3) = 19, G*(4) = 219, G*(5) = 4231, @*(6) = 130,023, G*(7) = 6,129,859. (c) How many of the precoding give self-dual posets ? (d) Is @*(n) odd for all nt? Justify. 3. Diagrams; Graded Posets The notion of “immediate superior poset, as follows. Derinition. By “a covers 8” in a poset P, it is meant that a > 6, but that a>x>bfornoze P. By the order n(P) of a poset P is meant the (casdinal) number of its elements. When this number is finite, P is called a “finite” poset. Using the covering relation, one can obtain a graphical representation of any finite poset P as follows. Draw a small circle to represent each element of P, placing a higher than 6 whenever a > 6. Draw a straight segment from a to b whenever a covers b. The resulting figure is called a diagram of P: examples are shown in Figures la—le below. Since a > 6 if and only if one can move from a to b downward along some broken line, it is clear that any finite poset is defined up to isomorphism by its diagram. It is also clear that the diagram of the dual P of a poset P is obtained from that of P by turning the latter upside down. in a hierarchy can be defined in any (a) {by te) (d) (e) Froure 1. Examples of diagrams Dermrrox. By a least element of any subset X of P, we mean an element aeX such that a < x for all xe X. By a greatest element of X, we mean an element b eX such that b > x for all ze X. ‘The preceding concepts are not to be confused with the concepts of minimal and maximal elements. A minimal element of a subset X of a partly ordered set P is an element a such that a < z for no ze X; maximal elements are defined dually. Clearly, a least element must be minimal and a greatest element maximal, but the converse is not true. Tarorem 3. Any finite nonempty subset X of a poset has minimal and maximal members.§3 DIAGRAMS; GRADED POSETS 5 Proor. Let X consist of 2, -++,2,. Define m, = x, and m, asx, if, < my and m,., otherwise. Then m, will be minimal. Similarly, X has a maximal element. TurorEm 4. In chains, the notions of minimal and least (maximal and greatest) clement of @ subset are effectively equivalent. Hence any finite chain has a least (first) and greatest (last) element. Proor. If x < a for no xe X, then by P4 2 2 a for every re X. Tunorem 5. Every finite chain of n elements is isomorphic with the ordinal number n (the chain of integers 1, «++, 7). That is, there is a bijection ¢ from the chain X onto {1, ---, n} such that x, < 2, if and only if ¢(z1) S $(x2); finite chains are finite ordinal numbers. Proor. Let ¢ map the least x X into 1, and the least of the remaining x ¢ X into 2, ete, ‘The length of a finite chain n is defined to be » — 1 (see its diagram). More ' generally, the length UP) of a poset P is defined as the least upper bound of the lengths of the chains in P. When [(P) is finite, P is said to be of finite length. Any poset of finite length is defined up to isomorphism by its covering relation: a>) if and only if a finite sequence x, 2, --+,2, exists, such that a = 2, b = 2,, and %_, covers x fori = 1,+++,n. The isomorphism or nonisomorphism of two finite posets can often be tested most simply by drawing their diagrams. Any isomorphism must be one-to-one onto between lowest elements, between next lowest elements, and so on. Corre- sponding elements must be covered by equal numbers of elements, and the covering elements must also correspond. These principles make it easy to enumerate the different (i.e., nonisomorphic) posets having x = 4 elements (say); there are exactly 16 of them. In a poset P of finite length with O, the height or dimension h{[z] of an element xe P is, by definition, the Lu.b. of the lengths of the chains 0 = x < a, <--- < 2, = between O and z. If P has a universal upper bound J, then clearly HI] =P]. Clearly also, h[x] = 1 if and only if x covers 0; such elements are called “atoms” or “points” of P. ‘The height function is especially important in graded poses. These are defined as posets P with a function g: P > Z from P to the chain of all integers (in their natural order), such that: Gl. > y implies g[x] > gly] (strict isotonicity). G2. If. covers y, then g[x] = gly] + 1. Any graded poset satisfies the following JORDAN-DEDEKIND CHAIN CONDITION. All maximal chains between the same endpoints have the same finite length. Lemma 1. Let P be any poset with O in which all chains are finite. Then P satisfies the Jordan—Dedekind chain condition if and only tf it is graded by h{z].6 TYPES OF LATTICES I Proor. If P is graded by Alz], then the Jordan-Dedekind chain condition follows obviously: the length of any maximal chain joining the endpoints @ and 6 > ais h{b] — Ala]. Conversely, if the Jordan-Dedekind chain condition holds, then A(z] is the length of every maximal chain from 0 to z, from which G1 and G2 follow immediately. Exercises for §3: 1. (a) Show that the diagram of a poset is an oriented graph,t if we draw an arrow from x to y if and only if x covers y. grea _,(P) Show that finite oriented graph is associated with a poset if and only if apy, dy0ay “+ 4,144 is incompatible with aoa, {#c) Show that any oriented graph defines a quasi-ordered set, if ¢ 2 b is defined to mean a = b, ab, or nace owt «++, a,b for suitable 4;,++ +, a. 2. Show that the “covering relations” of any poset form a new poset if ('‘z covers y”) > {‘*u covers »”) means that y > w. 3. Show that any isotone transformation P — P, of one poset P onto another P, carries connected components of the graph of P into connected components of the graph of Pi. 4. Which of the diagrams of Figure 1 represent self-dual posets 1 Define two new self-dual posete by means of diagrams, §. Show that a chain can be defined as a set of elements in which a transitive relation z > y is defined, such that for any elements u, v, one and only one of the relations u > v, u = », v > uholds. 6. Show that chains are those posets, all of whose subsets are lattices. 7. Show that no finite poset of more than two elements is defined to within isomorphism by its graph. 8. Lot P be. poset of finite length. Show that any two clements of P have an upper bound if and only if P has a universal upper bound I. 9. Show that, in 9 poset P of finite length, a chain from a to 6 is maximal if end only if it is ‘connected in the graph of P. 4. Lattices An upper bound of a subset X of a poset P is an element a € P containing every xe X. The least upper bound is an upper bound contained in every other upper bound; it is denoted l.u.b. X or sup X. By P2, sup X is unique if it exists. The notions of lower bound of X and greatest lower bound (g.l.b. X or inf X) of X are defined dually. Again by P2, inf X is unique if it exists. Dzrrnrrion. A latticet is a poset P any two of whose elements have a g.l.b. or “meet” denoted by x A y, and a.u.b. or “join” denoted byx v y. A lattice L is complete when each of its subsets X has a 1.u.b. and a g.l.b. in L. Setting X = L, we see that any nonvoid complete lattice contains a least element Oand a greatest element J. Evidently, the dual of any lattice is a lattice, and the dual of any complete lattice is a complete lattice, with meets and joins inter- changed. Any finite lattice or lattice of finite length is complete. More sophisti- { For the concepts of graph and oriented graph (“digraph”), ef. O. Ore, Theory of graphs, Amer. Math. Soc., 1962. } The concept of a lattice (‘Dualgruppe”) was firet atudied in depth by Dedekind [1, pp. 113-14]. That of a complete lattice wes introduced by the author in [1, p. 442).