Relation Algebras and the Converse
Introduction
The binary relations on a set carry three operations: the Boolean operations of union, intersection and complement, the composition $R ; S$ of two relations, and the converse $R^{-1}$ that reverses each ordered pair. Composition is associative, has the diagonal $\Delta = 1'$ as identity, and distributes over unions; the converse is an involution on the elements, reversing the order of the factors of a composite and commuting with the Boolean operations. A relation algebra is the abstract structure with those operations and the axioms they satisfy, of which the relations on a set form the standard example. This article states the axioms, proves that the relations on a set satisfy them, isolates the converse as the involution and examines its fixed part, the symmetric relations, and it treats the residuals of composition and the Peircean law that links the converse to composition through the complement.
The article presupposes Sets, Functions and Relations — relations, their composition and their converse — The Converse Relation as an Operator, in this category, where the passage $R \mapsto R^{-1}$ is studied as an operator, and Order Theory and Lattices and Boolean Algebras with an Involution, in this group, for the Boolean operations and the complement. The article uses the notation of The Converse Relation as an Operator for the converse and writes it $R^{-1}$; the literature mark $\breve{R}$ is not used.
Three boundaries are observed. No topology is used. The algebra of matrices over a field and
the transpose, which is the other standard instance of a converse, are Linear Spaces and Linear
Algebras, later in this Part, and are named only; the article works with relations on a set. The
adjoint in the involutive sense, and the reading of the converse as an adjoint, belong to the
* Operator Theory group of this category and are named only. Nothing linear and no form is used.
The Algebra of Binary Relations
Composition, Converse and the Diagonal
Let $X$ be a set and let $\mathcal{P}(X \times X)$ be the set of binary relations on $X$. For $R, S \subseteq X \times X$ put
$$ R ; S = \{(x,z) : \text{there is } y \in X \text{ with } (x,y) \in R \text{ and } (y,z) \in S\}, \qquad R^{-1} = \{(y,x) : (x,y) \in R\}, $$
and let $\Delta = \{(x,x) : x \in X\}$ be the diagonal, written $1'$.
Proposition. Composition is associative, $\Delta$ is a two-sided identity for it, and composition distributes over unions on both sides:
$$ (R ; S) ; T = R ; (S ; T), \qquad R ; \Delta = R = \Delta ; R, \qquad R ; (S \cup T) = (R ; S) \cup (R ; T), \qquad (S \cup T) ; R = (S ; R) \cup (T ; R) . $$
Proof. For the associativity, $(x,w) \in (R;S);T$ if and only if there are $y,z$ with $(x,y) \in R$, $(y,z) \in S$, $(z,w) \in T$, which is symmetric in the two groupings. The identity and the distributivity are immediate from the definition, the latter because a witness $y$ for the union is a witness for one of the two relations.
Proposition. The converse reverses composition and commutes with the Boolean operations:
$$ (R^{-1})^{-1} = R, \qquad (R ; S)^{-1} = S^{-1} ; R^{-1}, \qquad (R \cup S)^{-1} = R^{-1} \cup S^{-1}, \qquad (\overline{R})^{-1} = \overline{R^{-1}}, $$
where $\overline{R} = (X \times X) \setminus R$. Hence $R \mapsto R^{-1}$ is an order isomorphism of $\mathcal{P}(X \times X)$ with its opposite, of order two, and it is the involution of the relation algebra.
Proof. The first identity is the definition read twice. For the second, $(z,x) \in (R;S)^{-1}$ means $(x,z) \in R;S$, so there is $y$ with $(x,y) \in R$, $(y,z) \in S$; then $(z,y) \in S^{-1}$ and $(y,x) \in R^{-1}$, so $(z,x) \in S^{-1};R^{-1}$. The congruence for the union is the same reading, and the one for the complement follows because the converse is a bijection of $X \times X$.
The Relation Algebra Axioms
Definition. A relation algebra is a set $A$ with two binary operations $\vee$ and $;$ and a unary operation ${}^{-1}$ and two constants $0$ and $1'$, such that
- $(A, \vee)$ is a join-semilattice with least element $0$, and the associated order has meets and a greatest element $1$, making $A$ a Boolean algebra;
- $;$ is associative, has identity $1'$, and distributes over $\vee$ on both sides;
- ${}^{-1}$ is an involution with $(x \vee y)^{-1} = x^{-1} \vee y^{-1}$ and $(x ; y)^{-1} = y^{-1} ; x^{-1}$;
- the Peircean law holds: $(x ; y) \wedge \overline{z} = 0$ implies $(x^{-1} ; z) \wedge \overline{y} = 0$ and $(z ; y^{-1}) \wedge \overline{x} = 0$.
Theorem. The relations on a set $X$ form a relation algebra under union, composition, the converse and the constants $\emptyset$ and $\Delta$; the greatest element is $X \times X$, and the complement is the set complement.
Proof. The Boolean part is Sets, Functions and Relations and the power-set lattice is treated in Order Theory and Lattices; the first three axioms are the two propositions above, with the converse of a union and of a composite already computed. The Peircean law is proved in the next section.
The axioms are those of Tarski. The relation algebra $\mathcal{P}(X \times X)$ is the standard proper relation algebra; an abstract relation algebra need not be of this form, and the representation theorem saying which are is a deeper result and is not used here.
The Converse as the Involution and the Symmetric Part
The map $R \mapsto R^{-1}$ is the involution on the elements of the relation algebra; its fixed elements are the symmetric relations. The symmetry of a relation is a property of the relation, not of the algebra, and the fixed elements do not form a subalgebra of the relation algebra: they are closed under union, intersection and complement, but not under composition.
Proposition. The fixed elements of the involution are closed under the Boolean operations and contain $\emptyset$, $X \times X$ and $\Delta$, but they are not closed under composition whenever $X$ has at least three elements.
Proof. If $R = R^{-1}$ and $S = S^{-1}$ then $(R \cup S)^{-1} = R^{-1} \cup S^{-1} = R \cup S$ and likewise for the intersection and the complement, so the fixed set is a Boolean subalgebra. For the composition, let $X = \{0,1,2\}$, let $R$ be the relation $\{(0,1),(1,0)\}$ and $S$ the relation $\{(1,2),(2,1)\}$, both symmetric; then $R ; S$ contains $(0,2)$ but not $(2,0)$, so it is not symmetric.
Proposition. Every relation $R$ decomposes as $R = R_{\mathrm{s}} \cup R_{\mathrm{a}}$ with $R_{\mathrm{s}} = R \cap R^{-1}$ symmetric and $R_{\mathrm{a}} = R \setminus R^{-1}$ satisfying $R_{ \mathrm{a}} \cap R_{\mathrm{a}}^{-1} = \emptyset$.
Proof. The two parts are disjoint and their union is $R$; the first is fixed by the involution because the involution is order-reversing and of order two, and the second is disjoint from its converse by construction.
The Residuals and the Peircean Law
The Residuals of Composition
Composition of relations is a residuated map in each variable, in the sense of Operators on a Poset, and the residuals are computed by a universal condition.
Definition. For relations $R, T \subseteq X \times X$ put
$$ R \backslash T = \{(y,z) : (x,y) \in R \Rightarrow (x,z) \in T \text{ for all } x \in X\}, \qquad T / R = \{(x,y) : (y,z) \in R \Rightarrow (x,z) \in T \text{ for all } z \in X\} . $$
Theorem. For all relations $R, S, T$,
$$ R ; S \subseteq T \quad\Longleftrightarrow\quad S \subseteq R \backslash T \quad\Longleftrightarrow\quad R \subseteq T / S . $$
Proof. $R;S \subseteq T$ says that every $(x,z)$ with a witness $y$, $(x,y) \in R$, $(y,z) \in S$, lies in $T$. This holds if and only if for every $(y,z) \in S$ and every $x$ with $(x,y) \in R$ one has $(x,z) \in T$, which is $S \subseteq R \backslash T$. The second equivalence is the same statement with the roles of $R$ and $S$ exchanged.
Corollary. The maps $S \mapsto R;S$ and $R \mapsto R;S$ are residuated; the residual $R \backslash T$ is the greatest $S$ with $R;S \subseteq T$, and $T/R$ the greatest $R$ with $R;S \subseteq T$. The two residuals are related to the converse by $(R \backslash T)^{-1} = T^{-1} / R^{-1}$.
Proof. The residuals are the greatest solutions by the theorem and the definition of the order; the last identity is the definition read through the converse.
The Peircean Law
Theorem (Peircean law). For relations $R, S, T$ on $X$ the following are equivalent:
$$ (R ; S) \cap T = \emptyset, \qquad (R^{-1} ; T) \cap S = \emptyset, \qquad (T ; S^{-1}) \cap R = \emptyset . $$
Proof. (1) $\Rightarrow$ (2): suppose $(R;S) \cap T = \emptyset$ and suppose some $(a,b)$ lies in both $R^{-1};T$ and $S$. From $(a,b) \in R^{-1};T$ there is $y$ with $(a,y) \in R^{-1}$ and $(y,b) \in T$, that is, $(y,a) \in R$ and $(y,b) \in T$; with $(a,b) \in S$ this gives $(y,b) \in R;S$ and $(y,b) \in T$, contradicting $(R;S) \cap T = \emptyset$. (1) $\Rightarrow$ (3): suppose $(a,b) \in T;S^{-1}$ and $(a,b) \in R$; from $(a,b) \in T;S^{-1}$ there is $y$ with $(a,y) \in T$ and $(y,b) \in S^{-1}$, that is, $(b,y) \in S$, so $(a,y) \in R;S$ and $(a,y) \in T$, again a contradiction. The implications (2) $\Rightarrow$ (1) and (3) $\Rightarrow$ (1) are the same arguments read backwards, so the three statements are equivalent.
Corollary. The residual identity $R;S \subseteq T \Leftrightarrow S \subseteq R \backslash T$ and the Peircean law are equivalent in a relation algebra, in the presence of the other axioms; the Peircean law is the form in which the converse enters the residual calculus.
Proof. The law is the instance $T = \overline{U}$ of the residual identity, and conversely the residual identity follows from the law by complementing and using the Boolean axioms; the two are two readings of the same triangle condition.
Summary
The relations on a set form a relation algebra: a Boolean algebra with an associative composition that distributes over the joins and has the diagonal as identity, and with the converse $R \mapsto R^{-1}$ as an involution on the elements that reverses composites and commutes with the Boolean operations. The fixed elements of the involution are the symmetric relations, a Boolean subalgebra that is not closed under composition; every relation is the disjoint union of its symmetric part $R \cap R^{-1}$ and its asymmetric part $R \setminus R^{-1}$.
Composition is residuated in each variable: $R;S \subseteq T$ if and only if $S \subseteq R \backslash T$ if and only if $R \subseteq T/S$, with the residuals computed by a universal condition and the residuals exchanged by the converse. The Peircean law is the form of that residual calculus that uses the complement, $(R;S) \cap T = \emptyset$ implying $(R^{-1};T) \cap S = \emptyset$; it is one of the axioms of the abstract relation algebra, and the relations on a set satisfy it.
Summary of Notation
| Symbol | Meaning |
|---|---|
| $R ; S$ | Composition of relations: $(x,z)$ with a witness $y$ |
| $R^{-1}$ | Converse of $R$; the involution on the elements |
| $\Delta = 1'$ | The diagonal, the identity of composition |
| $\overline{R}$ | Complement of $R$ in $X \times X$ |
| $R \backslash T$, $T / R$ | Right and left residuals of composition |
| symmetric | $R^{-1} = R$; the fixed elements of the involution |
| $R_{\mathrm{s}}, R_{\mathrm{a}}$ | Symmetric and asymmetric parts $R \cap R^{-1}$ and $R \setminus R^{-1}$ |
Further Reading
- Alfred Tarski, "On the calculus of relations", Journal of Symbolic Logic 6 (1941), 73–89, for the axioms of a relation algebra and the calculus of relations.
- Roger D. Maddux, Relation Algebras, Studies in Logic 150 (Elsevier, 2006), for the Peircean law, the residuals and the representation theory.
- Bjarni Jónsson, "Varieties of relation algebras", Algebra Universalis 15 (1982), 273–298, for the equational theory and the relation-algebra identities.
- Chris Brink, Wolfram Kahl and Gunther Schmidt, Relational Methods in Computer Science (Springer, 1997), for composition, the residuals and the residuated structure of relations.
- Gunther Schmidt and Thomas Ströhlein, Relations and Graphs (Springer, 1993), for the calculus of relations, the Peircean law and the residual operations.