Measure Theory Blueprint

A set is simply a well-defined collection of objects, called its elements or members. The notation x ∈ A means that x is an element of the set A, while x ∉ A means it is not. Sets can contain numbers, functions, other sets, or any mathematical objects you care to work with. In measure theory, the sets you work with are almost always subsets of some fixed background space, called the sample space or universal set, typically denoted Ω or X. Every construction in this book takes place relative to such a background space.

The most fundamental set operations are the ones you likely already know well, but it is worth stating them precisely because measure-theoretic arguments depend on their exact properties. Given two sets A and B inside a universal set X:

• Union: A ∪ B = {x : x ∈ A or x ∈ B}, the set of elements in at least one of the two sets.
• Intersection: A ∩ B = {x : x ∈ A and x ∈ B}, the set of elements in both sets simultaneously.
• Complement: Ac = {x ∈ X : x ∉ A}, the set of elements in X that are not in A.
• Set difference: A \ B = A ∩ Bc, the elements of A that are not in B.
• Symmetric difference: A △ B = (A \ B) ∪ (B \ A), the elements in exactly one of the two sets.

Two sets are called disjoint if their intersection is the empty set ∅. A collection of sets is called pairwise disjoint if every pair of distinct sets in the collection is disjoint. This concept is central to measure theory because measures are defined to be additive over disjoint collections. When you assign a measure to a union of disjoint sets, you simply add up the individual measures. This additivity property is what makes the whole theory work.

Measure theory frequently requires working with countably infinite collections of sets, not just finite ones. For a sequence of sets A1, A2, A3, ..., the countable union is written ⋃n=1∞ An and the countable intersection is written ⋂n=1∞ An. These infinite operations are where classical finite intuition can break down, and where careful definitions become essential. The distinction between countable and uncountable infinity plays a surprisingly important role throughout the theory.

Two algebraic identities appear so frequently in measure-theoretic proofs that they deserve special attention. De Morgan's Laws state that the complement of a union equals the intersection of complements, and the complement of an intersection equals the union of complements. In symbols:

• (⋃n An)c = ⋂n Anc
• (⋂n An)c = ⋃n Anc

These laws let you convert between unions and intersections by taking complements, which is an extremely useful technique when you need to show that a family of sets is closed under certain operations. You will use De Morgan's Laws in almost every sigma-algebra argument you encounter.

Another concept that appears constantly is the notion of a partition. A partition of a set X is a collection of nonempty, pairwise disjoint subsets whose union is all of X. Partitions formalize the idea of dividing a space into non-overlapping pieces that together cover everything. For example, the partition of the real line into rational and irrational numbers is a simple two-element partition. In probability theory, a partition of the sample space corresponds to a set of mutually exclusive and exhaustive events, which is exactly the structure needed to apply the law of total probability.

It is also important to understand the concept of a power set. The power set of X, written 𝒫(X) or 2X, is the collection of all subsets of X, including the empty set and X itself. If X has n elements, its power set has 2n elements. For infinite sets, the power set is always strictly larger than the original set in the sense of cardinality. This is relevant because sigma-algebras are always subsets of the power set, and the power set itself is always a valid sigma-algebra. However, as you will see shortly, the power set is often too large to be useful.

Example: Set operations on a concrete space. Let X = {1, 2, 3, 4, 5}, A = {1, 2, 3}, and B = {2, 3, 4}. Then A ∪ B = {1, 2, 3, 4}, A ∩ B = {2, 3}, Ac = {4, 5}, A \ B = {1}, and A △ B = {1, 4}. The power set 𝒫(X) has 25 = 32 elements. A partition of X could be {{1, 2}, {3}, {4, 5}}, three disjoint sets whose union is all of X.

One more concept deserves careful attention: the notion of limsup and liminf of a sequence of sets. For a sequence A1, A2, ..., the limit superior is limsup An = ⋂n=1∞ ⋃k=n∞ Ak, which is the set of points that belong to infinitely many of the An. The limit inferior is liminf An = ⋃n=1∞ ⋂k=n∞ Ak, the set of points that belong to all but finitely many of the An. These constructions appear in the Borel-Cantelli lemma and in the study of convergence of random variables, making them indispensable tools for any serious work in probability theory.^[1]

A practical way to build fluency with these operations is to practice translating between verbal descriptions and set notation. For instance, the event that at least one of two experiments succeeds corresponds to a union, while both experiments succeed corresponds to an intersection. The event that the first experiment succeeds but the second fails is a set difference. Developing this translation skill makes it much easier to follow measure-theoretic proofs, which often move fluidly between verbal reasoning and symbolic manipulation. The more automatic this translation becomes, the less cognitive effort you spend on notation and the more you can focus on the mathematical ideas themselves.

Before moving on, it is worth pausing to appreciate why cardinality matters in this context. A set is countable if its elements can be listed in a sequence indexed by the natural numbers, even if that sequence is infinite. The integers, the rationals, and any finite set are all countable. The real numbers, however, are uncountable: no sequence can exhaust them. This distinction matters because sigma-algebras are defined using countable operations. Allowing only countable unions and intersections, rather than arbitrary ones, is precisely what makes sigma-algebras manageable and measure theory consistent. The restriction to countable operations is not a limitation but a deliberate design choice that keeps the theory coherent.

With set theory in hand, you are ready to meet the central object of this chapter: the sigma-algebra. A sigma-algebra (also written σ-algebra) on a set X is a collection ℱ of subsets of X that satisfies three axioms:

1. Contains the whole space: X ∈ ℱ.
2. Closed under complements: If A ∈ ℱ, then Ac ∈ ℱ.
3. Closed under countable unions: If A1, A2, A3, ... ∈ ℱ, then ⋃n=1∞ An ∈ ℱ.

From these three axioms, you can derive that ∅ ∈ ℱ (since ∅ = Xc), and that ℱ is also closed under countable intersections (by De Morgan's Laws). The pair (X, ℱ) is called a measurable space. The sets in ℱ are called measurable sets. These are the sets to which a measure can be assigned.

Why these three axioms? The answer comes from thinking about what you need in order to do probability or integration consistently. If an event A is measurable, its complement not A should also be measurable. If a countable sequence of events is measurable, their union at least one of them occurs should also be measurable. These are the minimal closure properties needed to make probability calculations well-defined. The sigma-algebra axioms are not arbitrary: they encode exactly the operations you need to perform in practice.^[2]

Let's look at some concrete examples to build intuition. Consider X = {a, b, c}. The smallest possible sigma-algebra on X is {∅, X}, called the trivial sigma-algebra. The largest is the power set 𝒫(X) = {∅, {a}, {b}, {c}, {a,b}, {a,c}, {b,c}, X}. Between these extremes, you can check that {∅, {a}, {b,c}, X} is also a sigma-algebra: it contains X, it is closed under complements (the complement of {a} is {b,c}, which is present), and it is closed under countable unions. However, {∅, {a}, {b}, X} is not a sigma-algebra because {a} ∪ {b} = {a,b} is not in the collection.

Checking the sigma-algebra axioms: a step-by-step example. Let X = ℝ and consider the collection ℱ = {∅, ℝ, (-∞, 0), [0, ∞)}. Does this satisfy the axioms? (1) ℝ ∈ ℱ: yes. (2) Complements: the complement of ∅ is ℝ (present), the complement of (-∞, 0) is [0, ∞) (present), and vice versa. (3) Countable unions: the only nontrivial union is (-∞, 0) ∪ [0, ∞) = ℝ (present). So yes, this is a valid sigma-algebra on ℝ, though a very coarse one that only distinguishes negative from non-negative numbers.

A key construction is the sigma-algebra generated by a collection. Given any collection 𝒞 of subsets of X, the sigma-algebra generated by 𝒞, written σ(𝒞), is the smallest sigma-algebra containing all sets in 𝒞. It exists because the intersection of any family of sigma-algebras is again a sigma-algebra, so you can take the intersection of all sigma-algebras that contain 𝒞. This generated sigma-algebra is the minimal structure that makes all sets in 𝒞 measurable. The concept of generation is enormously useful because it lets you specify a sigma-algebra by describing a simple generating collection rather than listing all its members explicitly.

The most important example of a generated sigma-algebra is the Borel sigma-algebra on the real line, denoted ℬ(ℝ). It is defined as the sigma-algebra generated by all open intervals (a, b) with a < b. Equivalently, it is generated by all open sets in ℝ, or by all closed sets, or by all half-open intervals of the form (-∞, x]. The sets in ℬ(ℝ) are called Borel sets. They include all open sets, all closed sets, all countable unions of closed sets (called F_σ sets), all countable intersections of open sets (called G_δ sets), and much more. For practical purposes, every set you are likely to encounter in analysis or probability is a Borel set.^[3]

It is natural to ask: why not just use the power set as the sigma-algebra and be done with it? The answer is that the power set is too large. In the early twentieth century, mathematicians discovered that if you assume the Axiom of Choice, you can construct subsets of the real line that cannot be assigned a consistent length (measure). These are called non-measurable sets, and their existence shows that no measure can be defined on all subsets of ℝ while satisfying the natural properties you would want. The Vitali set, constructed by Giuseppe Vitali in 1905, is the classic example.^[4] By restricting attention to a sigma-algebra smaller than the power set, you avoid these pathological sets and keep the theory consistent.

There is an important hierarchy of sigma-algebras worth knowing. Given two sigma-algebras ℱ1 and ℱ2 on the same space X, you say ℱ1 is coarser than ℱ2 (or ℱ2 is finer than ℱ1) if ℱ1 ⊆ ℱ2. A coarser sigma-algebra contains fewer measurable sets and therefore carries less information about the space. In probability theory, this hierarchy corresponds directly to the concept of information: a coarser sigma-algebra represents less information about the outcome of an experiment. This connection between sigma-algebras and information is one of the deepest ideas in modern probability theory, underpinning the entire theory of conditional expectation and martingales.

Another important construction is the product sigma-algebra. Given two measurable spaces (X, ℱ) and (Y, 𝒢), the product sigma-algebra on X × Y is generated by all measurable rectangles of the form A × B where A ∈ ℱ and B ∈ 𝒢. This construction is essential for defining measures on product spaces, which in turn is the foundation for Fubini's theorem and the theory of joint distributions in probability. When you work with a random vector (X, Y), the natural sigma-algebra on the joint sample space is exactly this product sigma-algebra.

To solidify your understanding, here is a practical checklist for verifying that a collection ℱ is a sigma-algebra:

1. Check that X ∈ ℱ (or equivalently, that ∅ ∈ ℱ and X ∈ ℱ).
2. For each set A you claim is in ℱ, verify that Ac is also in ℱ.
3. For any countable sequence of sets in ℱ, verify that their union is also in ℱ. It suffices to check this for pairwise disjoint sequences if you have already verified closure under finite unions.
4. Use De Morgan's Laws to confirm closure under countable intersections as a consistency check.

The concept of a sigma-algebra might seem abstract at first, but it becomes concrete when you connect it to familiar examples. In classical probability with a finite sample space, the sigma-algebra is typically the full power set, and every subset is an event. In continuous probability, the Borel sigma-algebra on ℝ is the natural choice, and it contains all the intervals, open sets, and closed sets you would ever want to assign probabilities to. In statistics, when you observe data, the sigma-algebra generated by your observations captures exactly the information those observations provide. The sigma-algebra is not just a technical device: it is a precise mathematical model of the concept of information itself.

One final point deserves emphasis. The sigma-algebra is always defined relative to a specific underlying space X. The same collection of sets might be a sigma-algebra on one space but not on another. For example, {∅, {1, 2}, {3, 4}, {1, 2, 3, 4}} is a sigma-algebra on X = {1, 2, 3, 4}, but it is not a sigma-algebra on X = {1, 2, 3, 4, 5} because it does not contain X itself. Always keep the underlying space in mind when working with sigma-algebras. This context-dependence is not a complication but a feature: it reminds you that measurability is always relative to a chosen structure, not an intrinsic property of sets in isolation.