Tutorials › Real Analysis › Generated Sigma-Algebras

Measure Theory · Tutorial 819 of 1000

Generated Sigma-Algebras

Use the minimality property of generated sigma-algebras and identify the exact sets generated by any finite collection of subsets.

Advanced 10 min read

What You'll Learn

  • State the defining minimality property of a generated sigma-algebra
  • Use a universal property to compare generated sigma-algebras
  • Determine when adding a set leaves a generated sigma-algebra unchanged
  • Construct the atoms associated with finitely many generating sets
  • Describe every member of a finitely generated sigma-algebra as a union of atoms

What Does “Generated” Mean?

The Examples of Sigma-Algebras tutorial showed several ways to build sigma-algebras. A different question is often more useful: given a collection of subsets of a space, what is the smallest sigma-algebra that contains them? The collection may not itself be a sigma-algebra, so it must be enlarged until it has the required closure properties. The result is called the sigma-algebra generated by that collection.

Let \(X\) be a set and let \(\mathcal{A}\subseteq\mathcal{P}(X)\) be a collection of subsets of \(X\). We use the existence result from the Sigma-Algebras tutorial: there is a smallest sigma-algebra on \(X\) containing \(\mathcal{A}\). Its defining minimality is the key tool in working with generators.

Definition: The sigma-algebra generated by \(\mathcal{A}\) on \(X\), denoted \(\sigma_X(\mathcal{A})\), is the smallest sigma-algebra on \(X\) that contains every member of \(\mathcal{A}\). When the underlying space is clear, we write \(\sigma(\mathcal{A})\).

Equivalently, the generated sigma-algebra is the intersection of all sigma-algebras on \(X\) that contain \(\mathcal{A}\). This family is nonempty because \(\mathcal{P}(X)\) is a sigma-algebra containing \(\mathcal{A}\). The existence theorem ensures that the intersection gives the smallest such sigma-algebra. This description is useful conceptually, but it does not usually list the generated sets explicitly.

The Minimality Property

To show that a sigma-algebra contains \(\sigma_X(\mathcal{A})\), it is enough to show that it contains the generators. Conversely, every member of \(\sigma_X(\mathcal{A})\) must belong to any sigma-algebra that contains those generators. This simple equivalence is the universal property of generation.

Theorem (Universal Property of a Generated Sigma-Algebra): Let \(\mathcal{A}\subseteq\mathcal{P}(X)\), and let \(\mathcal{F}\) be a sigma-algebra on \(X\). Then $$ \mathcal{A}\subseteq\mathcal{F} \quad\Longleftrightarrow\quad \sigma_X(\mathcal{A})\subseteq\mathcal{F}. $$

Proof. Suppose first that \(\mathcal{A}\subseteq\mathcal{F}\). Then \(\mathcal{F}\) is a sigma-algebra containing every generator in \(\mathcal{A}\). By the minimality in the definition of \(\sigma_X(\mathcal{A})\), we have \(\sigma_X(\mathcal{A})\subseteq\mathcal{F}\).

For the reverse implication, every member of \(\mathcal{A}\) belongs to \(\sigma_X(\mathcal{A})\) by definition. Thus \(\sigma_X(\mathcal{A})\subseteq\mathcal{F}\) implies \(\mathcal{A}\subseteq\mathcal{F}\). Both directions hold. \(\square\)

One immediate consequence is a convenient test for equality. If two collections of generators each lie in the sigma-algebra generated by the other, they generate the same sigma-algebra. Also, if \(\mathcal{A}\subseteq\mathcal{B}\), then \(\sigma_X(\mathcal{A})\subseteq\sigma_X(\mathcal{B})\): the latter is a sigma-algebra containing \(\mathcal{A}\), so the universal property applies.

Worked Example: The Sigma-Algebra Generated by One Set

Let \(A\subseteq X\). If \(A\) is neither empty nor all of \(X\), the sigma-algebra generated by the single set \(A\) is

$$ \sigma_X(\{A\})=\{\varnothing,A,X\setminus A,X\}. $$

To verify this, denote the displayed family by \(\mathcal{F}\). It contains \(X\). Taking complements exchanges \(\varnothing\) with \(X\) and exchanges \(A\) with \(X\setminus A\), so \(\mathcal{F}\) is closed under complements. Any countable union of members of \(\mathcal{F}\) is the union of some selection from these four sets. If the selection contains \(X\), its union is \(X\). If it contains both \(A\) and \(X\setminus A\), its union is \(X\). Otherwise, the union is one of \(\varnothing,A,X\setminus A\). Thus \(\mathcal{F}\) is a sigma-algebra containing \(A\).

Now let \(\mathcal{G}\) be any sigma-algebra containing \(A\). It contains \(X\), \(\varnothing\), and \(X\setminus A\), by the sigma-algebra properties. Hence \(\mathcal{F}\subseteq\mathcal{G}\). The family \(\mathcal{F}\) is therefore the smallest sigma-algebra containing \(A\). If \(A=\varnothing\) or \(A=X\), the generated sigma-algebra is instead the trivial sigma-algebra \(\{\varnothing,X\}\), which is already the family in the formula after repeated sets are removed.

Atoms from Finitely Many Generators

For finitely many generating sets, the generated sigma-algebra has a precise description. Each point of \(X\) has a membership pattern: for every generator, the point either belongs to it or does not. Points with the same pattern cannot be separated by any set in the generated sigma-algebra. The nonempty groups of points with identical patterns are called the atoms associated with the generators.

Definition: Given \(A_1,\ldots,A_n\subseteq X\), an atom associated with these sets is a nonempty set of the form $$ \bigcap_{i=1}^{n} C_i, \qquad \text{where each }C_i\text{ is either }A_i\text{ or }X\setminus A_i. $$ The choices of \(C_i\) specify one membership pattern. Empty intersections of this form are discarded.

There are at most \(2^n\) such patterns. Different nonempty atoms are disjoint: if two patterns differ at index \(j\), one atom is contained in \(A_j\) and the other in \(X\setminus A_j\). Every point has exactly one membership pattern, so the atoms partition \(X\).

Theorem (Finite Generators and Their Atoms): Let \(A_1,\ldots,A_n\subseteq X\), and let \(\mathcal{Q}\) be the collection of their nonempty atoms. Then \(\sigma_X(\{A_1,\ldots,A_n\})\) consists exactly of all unions of subcollections of \(\mathcal{Q}\), including the empty union.

Proof. By the Sigma-Algebra of Unions of Partition Blocks theorem, the family \(\mathcal{F}_{\mathcal{Q}}\) of all unions of atoms is a sigma-algebra on \(X\). Each generator \(A_i\) is a union of atoms: it consists precisely of the atoms whose membership pattern selects \(A_i\), rather than \(X\setminus A_i\), at index \(i\). Therefore \(\mathcal{F}_{\mathcal{Q}}\) contains every generator. By the universal property, \(\sigma_X(\{A_1,\ldots,A_n\})\subseteq\mathcal{F}_{\mathcal{Q}}\).

For the other inclusion, each atom is a finite intersection of generators or their complements. The generated sigma-algebra contains the generators, is closed under complements, and is closed under finite intersections, so it contains every atom. There are at most \(2^n\) atoms, so each union of atoms is a finite union. The generated sigma-algebra is closed under finite unions, and consequently contains every member of \(\mathcal{F}_{\mathcal{Q}}\). Thus \(\mathcal{F}_{\mathcal{Q}}\subseteq\sigma_X(\{A_1,\ldots,A_n\})\), proving equality. \(\square\)

Worked Example: Two Generators on a Six-Point Space

Let \(X=\{a,b,c,d,e,f\}\), \(A=\{a,b,c\}\), and \(B=\{c,d\}\). The four membership patterns give these atoms:

$$ A\cap B=\{c\},\qquad A\cap(X\setminus B)=\{a,b\}, $$
$$ (X\setminus A)\cap B=\{d\},\qquad (X\setminus A)\cap(X\setminus B)=\{e,f\}. $$

These sets are disjoint and their union is \(X\). The theorem says that every member of \(\sigma_X(\{A,B\})\) is a union of some of these four atoms. For example, \(A\) is the union \(\{c\}\cup\{a,b\}=\{a,b,c\}\), and \(B\) is the union \(\{c\}\cup\{d\}=\{c,d\}\). The complement of \(A\) is \(\{d\}\cup\{e,f\}=\{d,e,f\}\), also a union of atoms.

There are \(2^4=16\) unions because each of the four atoms can be included or excluded independently. The generated sigma-algebra is not the power set of \(X\): for instance, \(\{a\}\) is not a union of these atoms, since the atom \(\{a,b\}\) must be taken as a whole. The generators distinguish the four atoms from one another, but they do not distinguish \(a\) from \(b\), or \(e\) from \(f\).

Adding a Generator and Replacing Generators

The atom description also explains when an additional set changes a generated sigma-algebra. If the added set is already a member of the existing sigma-algebra, it adds no new information. If it cuts an atom into smaller pieces, the resulting sigma-algebra can become larger.

Worked Example: An Added Set That Changes Nothing

Continue with \(X,A,B\) from the preceding example. Let \(C=\{a,b,d\}\). This is a union of atoms: \(C=\{a,b\}\cup\{d\}\). Therefore \(C\in\sigma_X(\{A,B\})\). Since the existing sigma-algebra contains \(A\), \(B\), and \(C\), the universal property gives

$$ \sigma_X(\{A,B,C\})\subseteq\sigma_X(\{A,B\}). $$

On the other hand, \(\{A,B\}\subseteq\{A,B,C\}\), so monotonicity of generation gives the reverse inclusion. Hence the two sigma-algebras are equal. By contrast, if we add \(D=\{a\}\), then \(D\) splits the former atom \(\{a,b\}\) into \(\{a\}\) and \(\{b\}\). The finite-atoms theorem shows that the new generated sigma-algebra contains \(\{a\}\), which the old one did not. Thus this added generator makes the sigma-algebra strictly larger.

A useful form of the same reasoning is that a collection of generators can be replaced by any collection that generates the same sigma-algebra. In particular, once a set is known to belong to \(\sigma_X(\mathcal{A})\), adding it to \(\mathcal{A}\) does not enlarge the generated sigma-algebra. More generally, if \(\mathcal{A}\subseteq\sigma_X(\mathcal{B})\), then \(\sigma_X(\mathcal{A})\subseteq\sigma_X(\mathcal{B})\). These comparisons let us choose whichever description of a sigma-algebra is easiest to use.

Why Generation Is Useful

The word “smallest” is more than a label: it turns a construction problem into a pair of inclusion arguments. To prove that a candidate family is the generated sigma-algebra, one can show both that it is a sigma-algebra containing the generators and that every one of its members must lie in any sigma-algebra containing those generators. In finite settings, the atom theorem makes the candidate family explicit. In general settings, the universal property is often the more practical tool.

A common pitfall is to assume that listing the generators lists all the sets in the generated sigma-algebra. Even one generator usually forces its complement and the empty set and the whole space to be included. With several generators, intersections of generators and their complements produce the membership-pattern atoms, and unions of those atoms must also be included. Another pitfall is to assume that every added set enlarges the sigma-algebra: if it is already generated by the existing collection, it changes nothing.

For finitely many generators, no further distinctions are possible beyond the atoms: every set in the generated sigma-algebra is a union of whole atoms. This gives an efficient way to count its members as well. If there are \(m\) nonempty atoms, then there are exactly \(2^m\) unions, because each atom is either selected or not selected. The next tutorial will apply the idea of generation to a central family of subsets of the real line.

Check Your Understanding

Use the minimality property and the atom description to answer these questions.

  1. If \(\mathcal{F}\) is a sigma-algebra containing every member of \(\mathcal{A}\), what inclusion must hold between \(\sigma_X(\mathcal{A})\) and \(\mathcal{F}\)?
  2. Why do two distinct atoms associated with finitely many generators have empty intersection?
  3. For two generators \(A\) and \(B\), how many membership patterns are possible before empty atoms are discarded?
  4. When does adjoining a set \(C\) to a generator collection leave the generated sigma-algebra unchanged?
  5. If there are \(m\) nonempty atoms, why does their union structure give exactly \(2^m\) members?