Technopedia

The technologist’s encyclopedia

Measurable functions

A measurable function is a map between two measurable spaces that “respects” the measurable structure: preimages of measurable sets are measurable. It is the bridge that allows integration to be defined. It answers the question which functions can we integrate? In probability, a random variable is precisely a measurable function from a probability space to R\mathbb{R}, which motivates the concept for readers coming from that direction.

Authoritative sources

Definition

Let (X,F)(X, \mathcal{F}) and (Y,G)(Y, \mathcal{G}) be measurable spaces. A function f:X→Yf: X \to Y is measurable if f−1(B)∈Ffor all B∈Gf^{-1}(B) \in \mathcal{F} \quad \text{for all } B \in \mathcal{G}

In the most common case, Y=RY = \mathbb{R} with its Borel sigma-algebra B(R)\mathcal{B}(\mathbb{R}).

Details

Equivalent characterizations for f:X→Rf: X \to \mathbb{R}:

ff is measurable if and only if any (hence all) of the following hold:

  1. {x:f(x)>a}∈F\{x : f(x) > a\} \in \mathcal{F} for all a∈Ra \in \mathbb{R}
  2. {x:f(x)≥a}∈F\{x : f(x) \geq a\} \in \mathcal{F} for all a∈Ra \in \mathbb{R}
  3. {x:a≤f(x)<b}∈F\{x : a \leq f(x) < b\} \in \mathcal{F} for all a,b∈Ra, b \in \mathbb{R}

Key facts:

Note on the nature of measurability. Measurability is a relationship between two measurable structures (the sigma-algebra on the domain and the sigma-algebra on the codomain), not a property of the function in isolation. The same function f:X→Rf: X \to \mathbb{R} may be measurable with respect to one sigma-algebra on XX and not another.

Non-measurable functions. If F={∅,X}\mathcal{F} = \{\varnothing, X\} (the trivial sigma-algebra), then the only measurable functions f:X→Rf: X \to \mathbb{R} are the constant functions: for any non-constant ff, there exists a Borel set BB such that f−1(B)f^{-1}(B) is a proper non-empty subset of XX, which is not in F\mathcal{F}. More dramatically, on (R,L)(\mathbb{R}, \mathcal{L}) (Lebesgue-measurable sets), the indicator of a Vitali set VV is not measurable because V∉LV \notin \mathcal{L}.

Simple functions

A simple function is a measurable function taking only finitely many values: f=∑i=1nai 1Aif = \sum_{i=1}^{n} a_i \, \mathbf{1}_{A_i}

where ai∈Ra_i \in \mathbb{R} are distinct and Ai∈FA_i \in \mathcal{F} are pairwise disjoint.

Details

Simple functions are the building blocks of the Lebesgue integral. Every nonnegative measurable function ff is the pointwise limit of an increasing sequence of simple functions fn↑ff_n \uparrow f.

Example. On R\mathbb{R}, the floor function f(x)=⌊x⌋f(x) = \lfloor x \rfloor is not simple (it takes infinitely many values). However, restricted to [0,3)[0, 3) it is simple: f=0⋅1[0,1)+1⋅1[1,2)+2⋅1[2,3)f = 0 \cdot \mathbf{1}_{[0,1)} + 1 \cdot \mathbf{1}_{[1,2)} + 2 \cdot \mathbf{1}_{[2,3)}

The integral of a simple function is defined algebraically (see Lebesgue integral for the full construction): ∫f dμ=∑i=1nai μ(Ai)\int f \, d\mu = \sum_{i=1}^{n} a_i \, \mu(A_i)

This is independent of the particular representation of ff as a sum of indicator functions.

Composition of measurable functions

Theorem. If f:X→Yf: X \to Y is measurable and g:Y→Zg: Y \to Z is measurable, then g∘f:X→Zg \circ f: X \to Z is measurable.

Corollary. If f:X→Rf: X \to \mathbb{R} is measurable and g:R→Rg: \mathbb{R} \to \mathbb{R} is Borel-measurable (in particular, if gg is continuous), then g∘fg \circ f is measurable.

Details

Proof. For any C∈HC \in \mathcal{H} (the sigma-algebra on ZZ): (g∘f)−1(C)=f−1(g−1(C))(g \circ f)^{-1}(C) = f^{-1}(g^{-1}(C)) Since gg is measurable, g−1(C)∈Gg^{-1}(C) \in \mathcal{G}. Since ff is measurable, f−1(g−1(C))∈Ff^{-1}(g^{-1}(C)) \in \mathcal{F}.

Why the corollary matters. Every continuous function g:R→Rg: \mathbb{R} \to \mathbb{R} is Borel-measurable (preimages of open sets are open, hence Borel). So if ff is measurable, then sin⁡(f)\sin(f), efe^f, ∣f∣|f|, f2f^2, etc. are all measurable. This is the practical workhorse: you rarely need to check measurability from scratch; you compose known measurable functions.

Pointwise limits and measurability

Theorem. If f1,f2,…f_1, f_2, \ldots are measurable functions X→RX \to \mathbb{R}, then:

Details

Proof sketch. For sup⁡nfn\sup_n f_n: {x:sup⁡nfn(x)>a}=⋃n{x:fn(x)>a}\{x : \sup_n f_n(x) > a\} = \bigcup_n \{x : f_n(x) > a\}, which is a countable union of measurable sets.

This theorem is crucial because it means the class of measurable functions is “closed under the operations we actually use”: taking limits, composing, adding, multiplying.

Connection to the Riemann integral. Every Riemann-integrable function is Lebesgue-measurable (it is the a.e. limit of step functions, which are simple functions). The converse is false: the Dirichlet function 1Q\mathbf{1}_{\mathbb{Q}} is measurable (hence Lebesgue-integrable, with integral 0) but not Riemann-integrable.

Interactive exploration

Example problem

Problem. Let (R,B(R))(\mathbb{R}, \mathcal{B}(\mathbb{R})) be the Borel measurable space. Show that f(x)=1Q(x)f(x) = \mathbf{1}_{\mathbb{Q}}(x) (the Dirichlet function) is Borel-measurable, and find a sequence of simple functions converging to it pointwise.

Solution

Step 1: Show measurability.

ff takes only the values 0 and 1. For any Borel set B⊆RB \subseteq \mathbb{R}:

So ff is Borel-measurable.

Step 2: Construct simple functions.

Let fn(x)=1{q1,…,qn}(x)f_n(x) = \mathbf{1}_{\{q_1, \ldots, q_n\}}(x) where {q1,q2,…}\{q_1, q_2, \ldots\} is an enumeration of Q\mathbb{Q}. Each fnf_n is a simple function (it takes values 0 and 1 on the measurable sets {q1,…,qn}\{q_1, \ldots, q_n\} and its complement). And fn(x)→1Q(x)=f(x)f_n(x) \to \mathbf{1}_{\mathbb{Q}}(x) = f(x) for every xx.

- If $`0 \in B` and $`1 \in B`: $`f^{-1}(B) = \mathbb{R} \in \mathcal{B}(\mathbb{R})` ✓ - If $`0 \in B`, $`1 \notin B`: $`f^{-1}(B) = \mathbb{R} \setminus \mathbb{Q} \in \mathcal{B}(\mathbb{R})` (complement of a countable set) ✓ - If $`0 \notin B`, $`1 \in B`: $`f^{-1}(B) = \mathbb{Q} \in \mathcal{B}(\mathbb{R})` (countable set) ✓ - If $`0 \notin B`, $`1 \notin B`: $`f^{-1}(B) = \varnothing \in \mathcal{B}(\mathbb{R})` ✓

So ff is Borel-measurable.

Step 2: Construct simple functions.

Let fn(x)=1{q1,…,qn}(x)f_n(x) = \mathbf{1}_{\{q_1, \ldots, q_n\}}(x) where {q1,q2,…}\{q_1, q_2, \ldots\} is an enumeration of Q\mathbb{Q}. Each fnf_n is a simple function (it takes values 0 and 1 on the measurable sets {q1,…,qn}\{q_1, \ldots, q_n\} and its complement). And fn(x)→1Q(x)=f(x)f_n(x) \to \mathbf{1}_{\mathbb{Q}}(x) = f(x) for every xx.