Introduction

Measure theory begins with a simple question:

What does it mean to assign a size to a set?

For an interval on the real line, the answer seems obvious. The interval \([2,5]\) has length \(3\), and the interval \((2,5)\) also has length \(3\). A rectangle in the plane has area, a solid has volume, and an event in probability has probability. At first, these look like different ideas. Measure theory shows that they are examples of one common structure.

A measure is a rule that assigns a nonnegative size to certain sets. Length, area, volume, mass, and probability are all measures when they are formulated carefully. The word “certain” is important: in modern measure theory, we usually do not assign size to every possible subset of a space. Instead, we first specify which sets are allowed to be measured. These sets are called measurable sets.

This may feel strange at the beginning. If a set exists, why should it not have a size? One of the main lessons of the subject is that infinite sets can behave in ways that defeat ordinary geometric intuition. If we demand too many natural properties at once—especially compatibility with translations and countable addition—then not every subset of the real line can consistently be assigned a length. The construction of nonmeasurable sets, such as Vitali sets, is a classical expression of this limitation; it depends on the axiom of choice and is a central reason why measure theory must be built carefully rather than guessed from pictures (Folland, 1999; Royden and Fitzpatrick, 2010).

The goal of this book is to build that careful theory from first principles.

Why calculus needs measure theory

In calculus, integration is often introduced through area. For a nonnegative function \(f\) on an interval \([a,b]\), the integral

\[ \int_a^b f(x)\,dx \]

is described as the area under the graph of \(f\). This idea is powerful, but it hides a question: what kinds of “areas” are we allowed to measure?

For continuous functions, the picture is usually well behaved. For example, the function

\[ f(x)=x^2 \]

on \([0,1]\) has a smooth graph, and the area under it is

\[ \int_0^1 x^2\,dx=\frac13. \]

The Riemann integral handles such functions very well. It also handles many discontinuous functions. But it does not handle all naturally occurring limits of simple functions.

Consider the function

\[ \mathbf{1}_{\mathbb{Q}\cap[0,1]}(x) = \begin{cases} 1, & x\in \mathbb{Q}\cap[0,1],\\ 0, & x\notin \mathbb{Q}. \end{cases} \]

This is the indicator function of the rational numbers in \([0,1]\). An indicator function takes the value \(1\) on a set and \(0\) outside it. In this case, the function asks a yes-or-no question: “Is \(x\) rational?”

This function is discontinuous at every point of \([0,1]\). Every interval, no matter how small, contains both rational and irrational numbers. Therefore the Riemann upper sums see value \(1\), while the lower sums see value \(0\). The function is not Riemann integrable.

But from a measure-theoretic point of view, the rationals are small inside \([0,1]\). They are countable, meaning they can be listed as a sequence

\[ q_1,q_2,q_3,\dots. \]

A central fact in Lebesgue measure is that every countable subset of \(\mathbb{R}\) has length zero. Thus \(\mathbb{Q}\cap[0,1]\) has Lebesgue measure zero, while the interval \([0,1]\) has measure \(1\). The Lebesgue integral therefore gives

\[ \int_{[0,1]} \mathbf{1}_{\mathbb{Q}\cap[0,1]}\,d\lambda = 0, \]

where \(\lambda\) denotes Lebesgue measure. The function equals \(1\) only on a set of length zero, so its accumulated contribution is zero.

This example is not merely a trick. It reveals a change in viewpoint. The Riemann integral asks how a function behaves on intervals. The Lebesgue integral asks how large the sets are on which the function takes its values. Henri Lebesgue’s original work on integration, length, and area introduced this modern viewpoint and reshaped real analysis in the early twentieth century (Lebesgue, 1902).

The central idea: measure before integration

In calculus, integration often appears before a formal theory of size. In measure theory, the order is reversed:

1. first define which sets are measurable;
2. then define a measure on those sets;
3. then define measurable functions;
4. then define integration using the measure.

This order may seem slower at first, but it gives a much stronger theory.

For example, suppose \(E\subseteq \mathbb{R}\) is measurable. Its indicator function is

\[ \mathbf{1}_E(x) = \begin{cases} 1, & x\in E,\\ 0, & x\notin E. \end{cases} \]

For such a function, the Lebesgue integral should satisfy

\[ \int \mathbf{1}_E\,d\lambda = \lambda(E). \]

In words: integrating the indicator of a set gives the measure of the set. This is one of the guiding ideas of the whole subject.

From there, we build more complicated functions out of simple ones. A simple function is a function that takes only finitely many values. For example,

\[ s(x)=2\mathbf{1}_A(x)+5\mathbf{1}_B(x) \]

takes value \(2\) on \(A\), value \(5\) on \(B\), and possibly value \(0\) elsewhere, depending on the sets \(A\) and \(B\). If \(A\) and \(B\) are measurable and disjoint, it is natural to define

\[ \int s\,d\mu = 2\mu(A)+5\mu(B). \]

This formula already contains the heart of Lebesgue integration: values are weighted by the measures of the sets on which they occur. General nonnegative measurable functions are then approached by increasing sequences of simple functions. Signed and complex-valued functions are handled later by decomposing them into nonnegative pieces.

This construction may look abstract, but it is designed to make limit operations behave well. The Monotone Convergence Theorem, Fatou’s Lemma, and the Dominated Convergence Theorem are among the major rewards. They explain when limits and integrals can be interchanged, a question that appears constantly in analysis, probability, differential equations, and mathematical physics (Folland, 1999; Royden and Fitzpatrick, 2010).

Countable additivity: the rule that changes everything

The defining property of a measure is not just additivity. It is countable additivity.

If \(A\) and \(B\) are disjoint measurable sets, meaning \(A\cap B=\varnothing\), we want

\[ \mu(A\cup B)=\mu(A)+\mu(B). \]

This is finite additivity. It says that the size of two non-overlapping pieces is the sum of their sizes.

Measure theory requires more. If

\[ A_1,A_2,A_3,\dots \]

are pairwise disjoint measurable sets, then

\[ \mu\left(\bigcup_{n=1}^{\infty} A_n\right) = \sum_{n=1}^{\infty}\mu(A_n). \]

This is countable additivity.

The difference between finite and countable additivity is profound. Finite additivity can describe ordinary cutting and recombining. Countable additivity allows us to pass to limits. Since analysis is full of limiting processes—sequences, infinite series, approximations, convergence of functions—countable additivity is the property that makes measure theory compatible with analysis.

For example, each singleton set \(\{x\}\subseteq\mathbb{R}\) has Lebesgue measure zero:

\[ \lambda(\{x\})=0. \]

Since the rational numbers are countable, we may write

\[ \mathbb{Q}=\{q_1,q_2,q_3,\dots\}. \]

Then countable additivity gives

\[ \lambda(\mathbb{Q}) = \lambda\left(\bigcup_{n=1}^{\infty}\{q_n\}\right) \leq \sum_{n=1}^{\infty}\lambda(\{q_n\}) = 0. \]

So \(\lambda(\mathbb{Q})=0\). This is impossible to understand from finite additivity alone. The smallness of countable sets is one of the first places where countable additivity shows its power.

Measurable spaces: deciding which questions are observable

A measure is not defined on a bare set alone. It is defined on a pair

\[ (X,\mathcal{A}), \]

where \(X\) is a set and \(\mathcal{A}\) is a collection of subsets of \(X\). The collection \(\mathcal{A}\) is called a sigma-algebra.

A sigma-algebra is a family of sets closed under the operations needed for countable logic:

• if \(A\in\mathcal{A}\), then its complement \(X\setminus A\) is also in \(\mathcal{A}\);
• if \(A_1,A_2,A_3,\dots\in\mathcal{A}\), then \(\bigcup_{n=1}^{\infty}A_n\in\mathcal{A}\);
• consequently, countable intersections are also in \(\mathcal{A}\).

The pair \((X,\mathcal{A})\) is called a measurable space. It is a set equipped with a chosen collection of measurable questions.

For example, let \(X\) be the set of outcomes of rolling a fair die:

\[ X=\{1,2,3,4,5,6\}. \]

If we take \(\mathcal{A}\) to be the collection of all subsets of \(X\), then every event is measurable. The event “the result is even” is

\[ \{2,4,6\}, \]

and the event “the result is at least 5” is

\[ \{5,6\}. \]

A probability measure \(P\) assigns each event a number between \(0\) and \(1\), with

\[ P(X)=1. \]

For a fair die,

\[ P(\{2,4,6\})=\frac{3}{6}=\frac12. \]

Probability theory is therefore measure theory with total measure \(1\). This viewpoint was axiomatized in modern form by Andrey Kolmogorov, who formulated probability spaces using measure-theoretic language (Kolmogorov, 1956).

In more complicated spaces, such as \(\mathbb{R}\), we usually do not start with all subsets. We often begin with open intervals and generate the Borel sigma-algebra, the smallest sigma-algebra containing the open sets. Later, Lebesgue measure enlarges this structure by completing it, meaning that all subsets of null sets become measurable. This distinction between Borel measurable sets and Lebesgue measurable sets will become important in Chapters 6 and 7.

“Almost everywhere” and the discipline of ignoring null errors

One of the most useful phrases in measure theory is almost everywhere.

A statement holds almost everywhere if it fails only on a set of measure zero. For example, with respect to Lebesgue measure on \(\mathbb{R}\), the functions

\[ f(x)=0 \]

and

\[ g(x)=\mathbf{1}_{\mathbb{Q}}(x) \]

are equal almost everywhere, because they differ only on \(\mathbb{Q}\), which has measure zero.

At first, this may seem like a dangerous habit: how can two functions be considered essentially the same if they differ at infinitely many points? The answer is that measure theory cares not only about how many points are involved, but how large the set of points is in measure. The rationals are infinite and dense, but they have Lebesgue measure zero. Their topological size and their measure-theoretic size are different.

This idea becomes essential in \(L^p\) spaces. In those spaces, functions that agree almost everywhere are identified. For instance, \(f=0\) and \(g=\mathbf{1}_{\mathbb{Q}}\) represent the same element of \(L^1([0,1])\), because

\[ \int_0^1 |f-g|\,d\lambda = \int_0^1 \mathbf{1}_{\mathbb{Q}\cap[0,1]}\,d\lambda = 0. \]

This is not a technical inconvenience; it is one of the reasons the theory becomes clean. Many convergence theorems and functional-analytic structures work properly only after null-set differences are treated as negligible.

The path of the book

The book begins with the set-theoretic language needed for measure theory. We review sets, functions, images, preimages, countability, and sequences of sets. This foundation matters because measure theory is built from operations on sets, especially countable unions and intersections.

We then ask why length needs a theory. The familiar idea of length is not enough by itself; we must specify which properties length should satisfy and how those properties interact with infinite operations. This leads to algebras, sigma-algebras, measurable spaces, and measures.

After that, we construct Lebesgue measure. The construction proceeds through outer measure, a preliminary way to estimate the size of any set from outside by covering it with intervals. Carathéodory’s criterion then identifies which sets are measurable. This method is one of the standard routes to Lebesgue measure and remains central in abstract measure theory (Halmos, 1974; Folland, 1999).

Once measurable sets are available, we study measurable functions. These are functions whose inverse images of measurable sets are measurable. For a real-valued function \(f\), measurability means that questions such as

\[ \{x : f(x)>a\} \]

are measurable for appropriate real numbers \(a\). In other words, the function is compatible with the measurable structure of the domain.

The next major stage is integration. We build the Lebesgue integral first for nonnegative functions, then for signed and complex-valued functions. This leads to the main convergence theorems, including the Monotone Convergence Theorem and the Dominated Convergence Theorem. These theorems are among the most important tools in modern analysis.

The middle chapters compare Riemann and Lebesgue integration, organize different modes of convergence, and introduce \(L^p\) spaces. The notation \(L^p\) refers to spaces of measurable functions whose \(p\)-th powers are integrable, with functions identified when they agree almost everywhere. These spaces form a bridge from measure theory to functional analysis.

The later chapters develop product measures, Fubini’s and Tonelli’s theorems, signed and complex measures, and the Radon–Nikodym theorem. The Radon–Nikodym theorem explains when one measure can be represented as an integral density with respect to another measure. In probability, this is closely related to likelihood ratios and changes of measure; in analysis, it is a fundamental structural theorem about measures (Folland, 1999).

The final chapters connect measure theory to probability, weak convergence, regular Borel measures, and broader areas of modern analysis. The aim is not only to prove theorems, but to help you recognize the patterns that make measure-theoretic arguments work.

What to expect from the subject

Measure theory is precise. It asks you to distinguish between statements that are true everywhere, true almost everywhere, true for every measurable set, or true only after passing to a subsequence. These distinctions can feel subtle at first, but they are part of the power of the subject.

Measure theory is also generous. Once the definitions are in place, many results become clearer than their calculus versions. For example, instead of asking whether a complicated improper integral converges by a special trick, we can ask whether the positive and negative parts are integrable. Instead of guessing whether limits and integrals may be exchanged, we can look for monotonicity, domination, or convergence in \(L^1\).

A useful guiding principle is this:

Measure theory turns questions about size, limit, and accumulation into a single language.

Here are three examples of that principle.

First, length is a measure. The interval \([0,1]\) has measure \(1\), while the rationals inside it have measure \(0\).

Second, probability is a measure. The probability of an event is its measure inside a probability space.

Third, integration accumulates values according to measure. A function contributes significantly only on sets of significant measure.

These examples are different in interpretation but identical in structure. That unity is why measure theory appears across analysis, probability, geometry, and applied mathematics.

A first mental model

Before entering the formal development, it is helpful to hold the following mental model.

A measurable space \((X,\mathcal{A})\) tells us which subsets of \(X\) are observable.

A measure \(\mu\) assigns sizes to those observable sets.

A measurable function \(f:X\to\mathbb{R}\) is a function whose level-set questions are observable.

The integral

\[ \int_X f\,d\mu \]

is the accumulated value of \(f\), weighted by the measure \(\mu\).

For a simple example, let

\[ X=\{a,b,c\} \]

and define a measure by

\[ \mu(\{a\})=2,\qquad \mu(\{b\})=1,\qquad \mu(\{c\})=4. \]

Let \(f:X\to\mathbb{R}\) be given by

\[ f(a)=10,\qquad f(b)=3,\qquad f(c)=5. \]

Then the integral should be

\[ \int_X f\,d\mu = 10\cdot 2+3\cdot 1+5\cdot 4 = 43. \]

This finite example already contains the general idea. The integral is not merely adding function values; it is adding function values weighted by the sizes of the points or regions where they occur. Lebesgue integration extends this idea to infinite spaces, continuous spaces, and functions that are limits of simpler functions.

The promise of the theory

By the end of this book, you should be able to read and write arguments involving measurable sets, null sets, measurable functions, convergence almost everywhere, convergence in measure, \(L^p\) norms, product measures, iterated integrals, and Radon–Nikodym derivatives. More importantly, you should understand why these ideas belong together.

The beginning will be careful. We will define sets, countability, sigma-algebras, and measures before proving major theorems. This is not delay; it is construction. Measure theory rewards patient foundations. Once the language is built, it becomes a remarkably efficient way to think.

We now begin with the basic tools of sets, logic, and countability.

References

Folland, Gerald B. Real Analysis: Modern Techniques and Their Applications. 2nd ed., Wiley, 1999.

Halmos, Paul R. Measure Theory. Springer, 1974.

Kolmogorov, A. N. Foundations of the Theory of Probability. 2nd English ed., Chelsea Publishing Company, 1956.

Lebesgue, Henri. “Intégrale, longueur, aire.” Annali di Matematica Pura ed Applicata, Series 3, vol. 7, 1902, pp. 231–359.

Royden, H. L., and P. M. Fitzpatrick. Real Analysis. 4th ed., Pearson, 2010.