Introduction to Measure

You already know that the interval $[a, b]$ has length $b - a$. But what is the “length” of the set of all rationals in $[0, 1]$? Or of a set built by an adversarial construction that avoids any obvious description? Measure theory is the branch of mathematics that answers these questions rigorously — and in doing so, it becomes the foundation of modern probability, integration, and analysis.

Why naïve length is not enough

For a finite union of disjoint intervals, length is obvious: add up the lengths of the pieces. The trouble starts when you try to assign lengths to more exotic subsets of $\mathbb{R}$.

The rational numbers have measure zero

The rationals $\mathbb{Q} \cap [0, 1]$ are dense in $[0, 1]$: between any two reals you can find a rational. Yet there are only countably many of them — as you saw in Countable Set. This tension is resolved by measure theory:

$$\lambda(\mathbb{Q} \cap [0,1]) = 0. \tag{1}$$

The intuition: arrange the rationals in a sequence $q_1, q_2, q_3, \ldots$ Cover $q_k$ with an open interval of length $\varepsilon / 2^k$. The union of all these intervals covers every rational in $[0,1]$, yet has total length at most

$$\sum_{k=1}^{\infty} \frac{\varepsilon}{2^k} = \varepsilon.$$

Since $\varepsilon$ can be made arbitrarily small, the rationals occupy zero length — even though they are everywhere.

You cannot measure every subset

The real surprise is that there exist subsets of $\mathbb{R}$ to which you cannot consistently assign a length. The Vitali set is the canonical example. Partition $[0, 1]$ by the equivalence relation $x \sim y \iff x - y \in \mathbb{Q}$. By the Axiom of Choice, pick exactly one representative from each class to form a set $V$. If you try to assign $V$ a length $\lambda(V) \geq 0$, you run into a contradiction: the translates $V + q$ (for $q \in \mathbb{Q} \cap [-1, 1]$) are pairwise disjoint, their union covers $[0, 1]$, and — if length were countably additive — the total length would have to equal $1$ while each summand is the same number, which is impossible whether that number is $0$ or positive.

The lesson: not every set can be measured. Measure theory’s first task is to identify which sets can be measured.

What measure theory builds

The fix has two parts.

First, instead of trying to assign a size to every subset, you restrict attention to a carefully chosen family of subsets called a σ-algebra — a collection that is closed under complement and countable union, making it stable under all the set operations you actually need. The Vitali set lies outside this family.

Second, you define a function $\mu$ on the σ-algebra that assigns a non-negative number (possibly $+\infty$) to each measurable set. The key property is countable additivity: for pairwise disjoint measurable sets $E_1, E_2, \ldots$,

$$\mu\!\left(\bigsqcup_{k=1}^{\infty} E_k\right) = \sum_{k=1}^{\infty} \mu(E_k). \tag{2}$$

Together, a σ-algebra $\mathcal{F}$ on a set $X$ and a measure $\mu \colon \mathcal{F} \to [0, +\infty]$ satisfying $(2)$ form a measure space $(X, \mathcal{F}, \mu)$.

The road ahead

In the next few checkpoints you will build the Lebesgue measure on $\mathbb{R}$ step by step:

1. σ-Algebra — the precise definition of the domain of a measure.
2. Outer Measure — a preliminary “size” defined on all subsets by covering with intervals; it is monotone and sub-additive but not fully additive.
3. Carathéodory’s Criterion — the rule that picks out exactly the sets on which the outer measure becomes a true measure.
4. Lebesgue Measure — the outer measure from step 2 restricted to the measurable sets from step 3, giving the definitive notion of length on $\mathbb{R}$.

Summary

• The length of an interval is the starting point, but extending length to arbitrary subsets of $\mathbb{R}$ requires care.
• The rationals have measure zero despite being dense: a countable set can always be covered by intervals of arbitrarily small total length, as in equation $(1)$.
• The Vitali set shows that not every subset of $\mathbb{R}$ can be assigned a consistent length: the domain of a measure must be restricted.
• The solution is a measure space $(X, \mathcal{F}, \mu)$: a set $X$, a σ-algebra $\mathcal{F}$ of measurable subsets, and a countably additive function $\mu$ on $\mathcal{F}$ — see equation $(2)$.
• The Lebesgue measure, built in the next checkpoints, is the canonical example on $\mathbb{R}$.