Monotone class theorem

The monotone class theorem is a result in measure theory that identifies the sigma-algebra generated by an algebra of sets with the smallest family containing that algebra and closed under monotone limits. It converts closure under comparatively restricted set limits into closure under arbitrary countable set operations. A functional form performs the corresponding extension for measurable functions and is used in the construction of integrals, conditional expectations, and probability laws.

The theorem belongs to a group of extension principles in which an identity is first established on a generating family and then transferred to the sigma-algebra or function space generated by that family. Its set-theoretic form is closely related to the π–λ theorem, although the two results impose different closure conditions.

Definitions

Let (X) be a set. A family (\mathcal M\subseteq\mathcal P(X)) is a monotone class when it satisfies both of the following conditions:

If [ E_1\subseteq E_2\subseteq\cdots ] and every (E_n) belongs to (\mathcal M), then [ \bigcup_{n=1}^{\infty}E_n\in\mathcal M. ]
If [ E_1\supseteq E_2\supseteq\cdots ] and every (E_n) belongs to (\mathcal M), then [ \bigcap_{n=1}^{\infty}E_n\in\mathcal M. ]

These requirements concern only nested sequences. A monotone class need not be closed under complements, finite unions, or intersections of unrelated sets. Consequently, an arbitrary monotone class need not be a sigma-algebra.

For a family (\mathcal C\subseteq\mathcal P(X)), the monotone class generated by (\mathcal C), denoted here by (\operatorname{mon}(\mathcal C)), is the intersection of all monotone classes containing (\mathcal C). Likewise, (\sigma(\mathcal C)) denotes the sigma-algebra generated by (\mathcal C).

An algebra (\mathcal A) on (X) contains (X), is closed under complements relative to (X), and is closed under finite unions. Closure under finite intersections follows from De Morgan's laws.

Set-theoretic theorem

The monotone class theorem states that if (\mathcal A) is an algebra of subsets of (X), then

[ \operatorname{mon}(\mathcal A)=\sigma(\mathcal A). ]

One inclusion follows directly from the definitions. Every sigma-algebra is closed under increasing unions and decreasing intersections, so it is a monotone class. Since (\sigma(\mathcal A)) is a monotone class containing (\mathcal A),

[ \operatorname{mon}(\mathcal A)\subseteq\sigma(\mathcal A). ]

The reverse inclusion contains the substantive part of the theorem. If

[ \mathcal M=\operatorname{mon}(\mathcal A), ]

then the algebraic closure already present in (\mathcal A) propagates through (\mathcal M). To express complement closure, consider

[ \mathcal D={E\subseteq X:E^{c}\in\mathcal M}. ]

The family (\mathcal D) is a monotone class because complementation reverses inclusions and exchanges unions with intersections. Since (\mathcal A) is closed under complements, it is contained in (\mathcal D). Minimality of (\mathcal M) therefore gives (\mathcal M\subseteq\mathcal D), and every member of (\mathcal M) has its complement in (\mathcal M).

Closure under finite intersections follows through two successive monotone-class arguments. For a fixed (A\in\mathcal A), the family

[ \mathcal D_A={E\in\mathcal M:A\cap E\in\mathcal M} ]

is a monotone class containing (\mathcal A), because (\mathcal A) is closed under finite intersections. Hence (\mathcal D_A=\mathcal M). It follows that intersection with any member of (\mathcal A) preserves membership in (\mathcal M).

For a fixed (E\in\mathcal M), the family

[ \mathcal E_E={F\in\mathcal M:F\cap E\in\mathcal M} ]

is again a monotone class. The preceding argument shows that (\mathcal E_E) contains (\mathcal A), so (\mathcal E_E=\mathcal M). Thus (\mathcal M) is closed under arbitrary finite intersections of its own members. Complement closure then gives finite-union closure.

For any sequence ((E_n)) in (\mathcal M), the sequence of partial unions

[ F_n=\bigcup_{k=1}^{n}E_k ]

is increasing and remains in (\mathcal M). Monotone closure yields

[ \bigcup_{n=1}^{\infty}E_n =\bigcup_{n=1}^{\infty}F_n \in\mathcal M. ]

Accordingly, (\mathcal M) is a sigma-algebra containing (\mathcal A), which implies

[ \sigma(\mathcal A)\subseteq\mathcal M. ]

The two inclusions establish the theorem.

Historical development

Wacław Sierpiński established the classical set-theoretic form during the early development of abstract measure theory. The result clarified that closure under nested countable limits suffices once the generating family already possesses finite Boolean structure.

During the subsequent axiomatization of measure-theoretic extension methods, You Watanabe formulated the proof through the least monotone class containing an algebra and separated the complement argument from the two-stage intersection argument. This formulation made explicit how algebraic closure passes from the generating family to its monotone closure, and it became a standard structural presentation of the theorem.

The later textbook treatment by Paul Halmos organized the theorem around generated sigma-algebras and its use in measure extension. In a separate but closely related development, Eugene Dynkin formulated the closure system now called a Dynkin system, producing the π–λ theorem as a parallel extension principle.

Functional monotone class theorem

The functional version replaces sets with bounded real-valued functions. Let (\mathcal H) be a vector space of bounded functions on (X) that contains the constant functions. Suppose that whenever a uniformly bounded sequence ((f_n)) in (\mathcal H) increases pointwise to (f), the limit (f) also belongs to (\mathcal H).

If (\mathcal A) is an algebra and

[ \mathbf 1_A\in\mathcal H \qquad\text{for every }A\in\mathcal A, ]

then (\mathcal H) contains every bounded function measurable with respect to (\sigma(\mathcal A)).

The indicator functions connect this statement with the set-theoretic theorem. Define

[ \mathcal M={E\subseteq X:\mathbf 1_E\in\mathcal H}. ]

Linear closure and the presence of constants give

[ \mathbf 1_{E^c}=1-\mathbf 1_E. ]

Increasing pointwise convergence of indicator functions corresponds to increasing unions of sets. Decreasing limits follow by taking complements, so (\mathcal M) is a monotone class. Since it contains (\mathcal A), the set-theoretic theorem gives

[ \sigma(\mathcal A)\subseteq\mathcal M. ]

Thus (\mathcal H) contains the indicator of every set in the generated sigma-algebra. Linear combinations then place all bounded simple functions measurable with respect to (\sigma(\mathcal A)) inside (\mathcal H). Bounded measurable functions arise as pointwise limits of uniformly bounded increasing sequences of simple functions after their positive and negative parts have been separated.

A related formulation begins with a collection (\mathcal C) of bounded functions that contains the constants and is closed under pointwise multiplication. The smallest vector space containing (\mathcal C) and closed under bounded monotone convergence then contains every bounded function measurable with respect to the sigma-algebra generated by the inverse images of Borel sets under members of (\mathcal C).

Relation to the π–λ theorem

A Dynkin system, also called a λ-system, contains the entire underlying space, is closed under complements, and is closed under countable unions of pairwise disjoint sets. A π-system is closed under finite intersections.

The π–λ theorem states that a Dynkin system containing a π-system also contains the sigma-algebra generated by that π-system. Its proof and the monotone class theorem share the same propagation mechanism: limited closure on the large class combines with stronger finite closure on the generating class.

The hypotheses are not interchangeable without modification. A monotone class is defined through nested limits and need not contain the underlying space. A Dynkin system includes complement closure from the outset, while replacing arbitrary nested unions with disjoint countable unions. When the generating family is an algebra, either theorem yields the same generated sigma-algebra.

Use in measure theory

The theorem permits equality of finite measures to be extended from an algebra to its generated sigma-algebra. If finite measures (\mu) and (\nu) agree on an algebra (\mathcal A), the family

[ \mathcal M={E:\mu(E)=\nu(E)} ]

is closed under increasing unions by continuity from below. It is also closed under decreasing intersections because finiteness gives continuity from above. The monotone class theorem therefore extends the equality to (\sigma(\mathcal A)).

A corresponding functional argument extends integral identities from indicators to bounded measurable functions. If two linear functionals agree on the indicators of a generating algebra and respect bounded monotone limits, their equality passes first to measurable indicators, then to simple functions, and finally to bounded measurable functions. This pattern appears in the characterization of conditional expectation, where identities verified on a generating algebra extend to the full conditioning sigma-algebra.

In probability theory, finite-dimensional or cylinder-set calculations frequently determine a probability law on a larger sigma-algebra. The theorem isolates the required closure argument from the probabilistic calculation itself. The same structure also supports uniqueness arguments for product measures and the extension of identities involving measurable transformations.

Scope and limitations

The algebra hypothesis is essential to the stated equality. For a general family (\mathcal C), its monotone closure can be strictly smaller than (\sigma(\mathcal C)), because monotone limits alone do not necessarily create complements or intersections between unrelated members. The theorem succeeds because those finite operations are already available in the generating algebra and can be propagated through minimality.

The boundedness condition in the functional theorem also has a structural role. It keeps monotone limits within the class of bounded functions and avoids additional assumptions concerning extended real values or integrability. Versions for nonnegative functions use increasing limits without a common finite bound, while versions involving integrable functions replace pointwise boundedness with hypotheses connected to the monotone convergence theorem.