\[ \newcommand{\C}{\mathbb{C}} \newcommand{\haar}{\mathsf{m}} \newcommand{\P}{\mathcal{P}} \newcommand{\R}{\mathbb{R}} \newcommand{\N}{\mathbb{N}} \newcommand{\Z}{\mathbb{Z}} \newcommand{\g}{>} \newcommand{\l}{<} \newcommand{\intd}{\,\mathsf{d}} \newcommand{\Re}{\mathsf{Re}} \newcommand{\area}{\mathop{\mathsf{Area}}} \newcommand{\met}{\mathop{\mathsf{d}}} \newcommand{\emptyset}{\varnothing} \newcommand{\B}{\mathscr{B}} \DeclareMathOperator{\borel}{\mathsf{Bor}} \]

Integration

Fix a measurable space $(X,\B)$. A measure on $(X,\B)$ allows us to assign sizes to subsets of $X$. In this section we are going to understand how measures give rise to integrals. The key is the we can associate to each measurable set a function on $X$ whose integral can reasonably be thought of as the measure of $X$. We will then integrate more complicated functions by approximations.

Simple functions

Fix a set $X$ and $A \subset X$. The indicator function of $A$ is the function \[ 1_A(x) = \begin{cases} 1 & x \in A \\ 0 & x \notin A \end{cases} \] from $X$ to $\R$.

Lemma

Fix a measurable space $(X,\B)$. For every $B \in \mathscr{B}$ the function $1_B : X \to \R$ is $(\B,\P(\R))$ measurable.

Proof:

Fix $E \subset \R$. The set $(1_B)^{-1}(E)$ can only be $\emptyset$, $B$, $X \setminus B$ or $X$ according to which of $0,1$ belong to $E$. All four sets belong to $\mathscr{B}$ because $B \in \mathscr{B}$. Therefore $1_B$ is $(\B,\P(\R))$ measurable.

Definition

Fix a measurable space $(X,\B)$. A function $f : X \to \R$ is simple if \[ f = a_1 1_{B(1)} + \cdots + a_k 1_{B(k)} \] for sets $B(1),\dots,B(k)$ in $\B$ and numbers $a_1,\dots,a_k$.

The expression \[ a_1 1_{B(1)} + \cdots + a_k 1_{B(k)} \] is not unique. Indeed, when $B(1)$ and $B(2)$ are disjoint we can write \[ 1_{B(1)} + 1_{B(2)} = 1_{B(1) \cup B(2)} \] and when $B(1) \cap B(2)$ is non-empty we can write \[ a_1 1_{B(1)} + a_2 1_{B(2)} = a_1 1_{B(1) \setminus B(2)} + (a_1 + a_2) 1_{B(1) \cap B(2)} + a_2 1_{B(2) \setminus B(1)} \] however, it is always possible to write a simple function in the form \[ a_1 1_{B(1)} + \cdots + a_k 1_{B(k)} \] where the sets $B(1),\dots,B(k)$ are disjoint.

Lemma

A function $f : X \to \R$ is simple if and only if its range is finite.

Proof:

If $f$ can be written as \[ a_1 1_{B(1)} + a_2 1_{B(2)} + \cdots + a_k 1_{B(k)} \] then its range is finite, as $f$ can only take values obtained from adding together at most $k$ of the terms $a_1,\dots,a_k$. On the other hand, if the only values $f$ takes are $c_1,\dots,c_k$ then \[ f = c_1 1_{f^{-1}(\{c_1\})} + \cdots + c_k 1_{f^{-1}(\{c_k\})} \] expresses $f$ as a simple function.

Integrating simple functions

Fix a measure space $(X,\B,\mu)$. We define the integral of a simple, measurable function $f : X \to \R$ with respect to $\mu$ to be \[ \int f \intd \mu = \sum_{t \in \R} t \cdot \mu(f^{-1}(\{t\})) = \sum_{i=1}^n a_i \cdot \mu(f^{-1}(\{a_i\})) \] where $a_1,\dots,a_n$ are the values that $f$ takes.

We will first check that the integral of simple functions is linear.

Theorem

Let $(X,\B,\mu)$ be a measure space. Let $f,g : X \to \R$ be simple and measurable. Then \[ \begin{gathered} \int f + g \intd \mu = \int f \intd \mu + \int g \intd \mu \\ \alpha \int f \intd \mu = \int \alpha f \intd \mu \end{gathered} \] for all $\alpha \in \R$.

Proof:

The second property is immediate. For the first, write \[ f = \sum_{i=1}^m a_i 1_{A(i)} \qquad g = \sum_{j=1}^r b_j 1_{B(j)} \] with $A(1),\dots,A(m)$ pairwise disjoint and $B(1),\dots,B(r)$ also pairwise disjoint. We can write \[ A(i) = \bigcup_{j=1}^r A(i) \cap B(j) \qquad B(j) = \bigcup_{i=1}^m B(j) \cap A(i) \] with both unions consisting of pairwise disjoint collections. Thus \[ \begin{gathered} f(x) = \sum_{i=1}^m \sum_{j=1}^r a_i 1_{A(i) \cap B(j)} \\ g(x) = \sum_{j=1}^r \sum_{i=1}^m b_i 1_{B(j) \cap A(i)} \\ (f+g)(x) = \sum_{i=1}^m \sum_{j=1}^r (a_i + b_j) 1_{A(i) \cap B(j)} \end{gathered} \] and adding the integrals of the first two simple functions gives the integral of the third simple function.

Integrating non-negative functions

Now that we know how to integrate simple functions, we move on to non-negative measurable functions.

Definition

Fix a measure space $(X,\B,\mu)$ and a measurable function $f : X \to [0,\infty]$. Define the integral of $f$ with respect to $\mu$ to be \[ \int f \intd \mu = \sup \left\{ \int \phi \intd \mu : \phi \text{ a simple function with } 0 \le \phi \le f \right\} \] the idea being to approximate the area under the graph of $f$ by the integral of a simple function.

Proposition

Fix $f,g : X \to [0,\infty]$ measurable.

  1. If $f \le g$ then $\displaystyle\int f \intd \mu \le \int g \intd \mu$
  2. If $\alpha \ge 0$ then $\displaystyle \alpha \int f \intd \mu = \int \alpha f \intd \mu$
Proof:

The first of these is immediate because every simple function $0 \le \phi \le f$ satisfies $0 \le \phi \le g$. For the second, if $\alpha = 0$ both sides are zero, and if $\alpha > 0$ then the result follows because it is true for simple functions.

We next want to prove that the integral is linear. We will not do this directly, instead proving first the fundamental convergence result on integration with respect to a measure.

Monotone convergence theorem

Theorem

For any measure space $(X,\B,\mu)$ and any non-decreasing sequence $f_n : X \to [0,\infty]$ of measurable functions the equality \[ \lim_{n \to \infty} \int f_n \intd \mu = \int \lim_{n \to \infty} f_n \intd \mu \] holds.

Proof:

Define $g$ to be the pointwise limit of the sequence $n \mapsto f_n$ of functions. It is well-defined and measurable as a function from $X$ to $[0,\infty]$. We have \[ \int f_n \intd \mu \le \int g \intd \mu \] for all $n \in \N$ by monotonicity. Therefore \[ \lim_{n \to \infty} \int f_n \intd \mu \le \int \lim_{r \to \infty} f_r \intd \mu \] holds. It remains to prove the reverse inequality.

For the reverse inequality fix $0 \l \alpha \l 1$ and fix $0 \le \phi \le g$ a simple function. Put \[ E(n) = \{ x \in X : f_n(x) \ge \alpha \phi(x) \} \] and note that $\phi \le f$ together with $0 \l \alpha \l 1$ implies the sequence $n \mapsto E_n$ is increasing and has union $X$. Writing $\{ a_1,\dots,a_r \}$ for the range if $\phi$ we can then calculate that \[ \begin{aligned} \int 1_{E(n)} \phi \intd \mu &= \sum_{i=1}^r a_r \mu( \phi^{-1}(a_r) \cap E(n)) \\ &\to \sum_{i=1}^r a_r \mu(\phi^{-1}(a_r)) = \int \phi \intd \mu \end{aligned} \] where the convergence follows from continuity of measures. But then \[ \lim_{n \to \infty} \int f_n \intd \mu \ge \lim_{n \to \infty} \int 1_{E(n)} f_n \intd \mu \ge \lim_{n \to \infty} \int 1_{E(n)} \alpha \phi \intd \mu = \alpha \int \phi \intd \mu \] which finishes the proof because $0 \l \alpha \l 1$ was arbitrary and $0 \le \phi \le f$ was arbitrary.

As a consequence of the monotone convergence theorem we get linearity and a useful convergence result.

Theorem

Whenever $f,g : X \to [0,\infty]$ are measurable \[ \int f+g \intd \mu = \int f \intd \mu + \int g \intd \mu \] holds.

Proof:

Fix sequences $n \mapsto \phi_n$ and $n \mapsto \psi_n$ of simple functions that take non-negative values and satisfy $\phi_n \to f$ and $\psi_n \to g$. Then $n \mapsto \phi_n + \psi_n$ is a sequence of non-negative simple functions that converges to $f + g$. We have already checked that \[ \int \phi_n + \psi_n \intd \mu = \int \phi_n \intd \mu + \int \psi_n \intd \mu \] and the monotone convergence theorem allows us to take the limits on both sides by evaluating the limits inside the integrals, giving the desired conclusion.