Core Structure Theory
Congruences and the Substitution Property
Congruences: the equivalence relations compatible with the operations. The substitution property, why it is the right condition, and the failure modes when it is absent.
Learning objectives
- State the compatibility condition defining a congruence
- Verify the condition in examples and construct a non-example
- Explain why congruences and not arbitrary equivalences give quotients
The definition
An equivalence relation θ on the universe of an algebra A is a congruence if it has the substitution property: for every n-ary basic operation f and all elements with ⟨ai, bi⟩ ∈ θ for each i, one has ⟨f(a1,…,an), f(b1,…,bn)⟩ ∈ θ.
In words: replacing arguments by θ-equivalent arguments leaves the value θ-equivalent. Equivalence classes can be substituted for one another without the operations noticing.
Why the condition is exactly right
The purpose of a congruence is to allow operations to be defined on the quotient set. Given a class a/θ, one wants to define
f(a1/θ,…,an/θ) = f(a1,…,an)/θ
This is well defined precisely when the choice of representatives does not matter — which is precisely the substitution property. The condition is not merely sufficient; it is necessary.
Congruences are not an arbitrary refinement of equivalence relations. They are exactly those equivalences for which the quotient construction works, so the definition is determined by what it is for.
A non-example
An equivalence that is not a congruence
Take the integers under addition and partition them into {0} and the non-zero integers. This is an equivalence relation with two classes.
It fails the substitution property: 1 and 2 are equivalent, as are −1 and −1. But 1 + (−1) = 0 while 2 + (−1) = 1, and 0 is not equivalent to 1. Addition does not respect the partition, so no quotient algebra exists.
The general lesson: most equivalence relations on an algebra are not congruences. Con(A) is typically a very small subset of Eq(A).
Term operations respect congruences
If θ is a congruence on A and p is any term, then ⟨ai, bi⟩ ∈ θ for all i implies ⟨pA(a), pA(b)⟩ ∈ θ.
The proof is induction on the structure of the term: the base case is the reflexivity of θ on variables and constants, and the inductive step is the substitution property applied to a basic operation. So checking basic operations suffices, and derived operations come free.
The familiar cases
| Algebra | Congruences correspond to | Determined by one class? |
|---|---|---|
| Group | Normal subgroups | Yes — the class of e |
| Ring | Two-sided ideals | Yes — the class of 0 |
| R-module | Submodules | Yes — the class of 0 |
| Lattice | Congruences proper | No |
| Semigroup | Congruences proper | No |
| Boolean algebra | Filters (equivalently ideals) | Yes — the class of 1 |
Groups, rings and modules are congruence-permutable, and permutability is what forces a congruence to be recoverable from a single class. Lattices and semigroups are not permutable, so their congruences carry genuinely more information than any one class. Mal'cev's theorem identifies exactly when the reduction is available.
Frequently asked questions
Why are normal subgroups the congruences of a group?
Given a congruence θ, the class of the identity is a normal subgroup, and θ is recovered as the relation a θ b iff ab⁻¹ lies in that subgroup. Conversely every normal subgroup yields a congruence this way, and the correspondence is a lattice isomorphism.
Can an algebra have only the two trivial congruences?
Yes — such algebras are called simple, and they are the building blocks in several structure theorems. Simple groups are the group-theoretic instance.
Source. S. Burris and H. P. Sankappanavar, A Course in Universal Algebra, The Millennium Edition — a corrected re-typesetting of Springer-Verlag Graduate Texts in Mathematics 78 (1981). Section II.5, book pages 38-39.
This page is an original exposition prepared for the KEVOS® knowledge library. It restates and reorganises mathematical results; it is not a reproduction of the source text.