Altanis

0.3Outer Measure on the Reals

Updated 2 Sep 2026Chapter (PDF)

This is not a section in Folland and content is drawn from a variety of places, such as Chapters 1-2 of MIRA.

[0.3.1]Definition(Length of Open Interval)#

We denote the length of an open interval by

(I)={baI=(a,b),0I=,I=(,a),(a,), or (,).\ell(I) = \begin{cases} b - a & I = (a, b), \\ 0 & I = \emptyset, \\ \infty & I = (-\infty, a), (a, \infty), \text{ or } (-\infty, \infty). \end{cases}

If we fully cover a set AA with open intervals in the most conservative way possible, the length of all intervals in the cover would seem to be a good notion of “size” for AA. And so it is.

[0.3.2]Definition(Outer Measure)#

The outer measure of a set ARA \subseteq \bR is defined by

μ(A)=inf{k=1(Ik):I1,I2, are open intervals such that Ak=1Ik}.\mu^*(A) = \inf\Big\{ \sum_{k = 1}^\infty \ell(I_k): I_1, I_2, \dots \text{ are open intervals such that } A \subseteq \bigcup_{k = 1}^\infty I_k \Big\}.

The measure takes \infty as a value—that is, μ:R[0,]\mu^*: \bR \to [0, \infty]. Thus, for example, μ(R)=\mu^*(\bR) = \infty. The outer measure has many desirable properties.

[0.3.3]Theorem(Countable Sets Are Measure Zero)#

Suppose ARA \subseteq \bR is countable. Then μ(A)=0\mu^*(A) = 0.

Proof.

Suppose A={a1,a2,}A = \{a_1, a_2, \dots\} is a countable subset of R\bR. Fix some ε>0\epsilon > 0. For each kZ+k \in \bZ_+, define Ik=(akε/2k,ak+ε/2k)I_k = (a_k - \epsilon/2^k, a_k + \epsilon/2^k). Then note

μ(A)k=1(Ik)=2εk=112k=2ε.\mu^*(A) \le \sum_{k = 1}^\infty \ell(I_k) = 2\epsilon \sum_{k = 1}^\infty \frac{1}{2^k} = 2\epsilon.

Since ε\epsilon is arbitrary, we have that μ(A)=0\mu^*(A) = 0.

[0.3.4]Theorem(Outer Measure Preserves Set Inclusion Order)#

Suppose A,BA, B are subsets of R\bR such that ABA \subseteq B. Then μ(A)μ(B)\mu^*(A) \le \mu^*(B).

[0.3.5]Theorem(Outer Measure is Translation Invariant)#

Let AA be any subset of R\bR and tt any arbitrary number in R\bR. Then μ(A)=μ(t+A)\mu^*(A) = \mu^*(t + A).

Let (1,4)(1, 4) and (3,5)(3, 5) be open intervals, whose union is (1,5)(1, 5). Then note that

μ((1,4)(3,5))<μ((1,4))+μ((3,5)),\mu^*((1, 4) \cup (3, 5)) < \mu^*((1, 4)) + \mu^*((3, 5)),

because the LHS is 44 and the RHS is 55. The intuition behind this is that the RHS counts (3,4)(3, 4) twice, whereas LHS counts it only once. We will generalize this to show the outer measure on R\bR is countably subadditive.

[0.3.6]Theorem(Outer Measure is Countably Subadditive)#

Suppose A1,A2,A_1, A_2, \dots is a sequence of subsets of R\bR. Then

μ(k=1Ak)k=1μ(Ak).\mu^*\left( \bigcup_{k = 1}^\infty A_k \right) \le \sum_{k = 1}^\infty \mu^*(A_k).
Proof.

If μ(Ak)=\mu^*(A_k) = \infty for any kZ+k \in \bZ_+, the result would be immediate, so suppose μ(Ak)<\mu^*(A_k) < \infty for every kZ+k \in \bZ_+. Fix ε>0\epsilon > 0. For each kZ+k \in \bZ_+, define a sequence of open intervals A1,k,A2,k,A_{1, k}, A_{2, k}, \dots that cover AkA_k in such a way that

j=1(Aj,k)ε2k+μ(Ak).\sum_{j = 1}^\infty \ell(A_{j, k}) \le \frac{\epsilon}{2^k} + \mu^*(A_k).

Simply consider the collection {Aj,k}j,kZ+\{A_{j, k}\}_{j, k \in \bZ_+}. Note the sum over these indices yields

k=1j=1(Aj,k)ε+k=1μ(Ak).\sum_{k = 1}^\infty \sum_{j = 1}^\infty \ell(A_{j, k}) \le \epsilon + \sum_{k = 1}^\infty \mu^*(A_k).

Note that this collection also covers k=1Ak\bigcup_{k = 1}^\infty A_k by construction, and so

μ(k=1Ak)k=1j=1(Aj,k)ε+k=1μ(Ak).\mu^*\left( \bigcup_{k = 1}^\infty A_k \right) \le \sum_{k = 1}^\infty \sum_{j = 1}^\infty \ell(A_{j, k}) \le \epsilon + \sum_{k = 1}^\infty \mu^*(A_k).

Since ε>0\epsilon > 0 is arbitrary, we have achieved that

μ(k=1Ak)k=1μ(Ak).\mu^*\left( \bigcup_{k = 1}^\infty A_k \right) \le \sum_{k = 1}^\infty \mu^*(A_k).

Countable subadditivity implies finite subadditivity, since we can let Ak=A_k = \emptyset for every k>nk > n, then we have

μ(k=1nAk)k=1nμ(Ak).\mu^*\left( \sum_{k = 1}^n A_k \right) \le \sum_{k = 1}^n \mu^*(A_k).
[0.3.7]Theorem(Outer Measure of Closed Intervals)#

μ([a,b])=ba\mu^*([a, b]) = b - a.

Proof.

Note that, for any ε>0\epsilon > 0, the sequence of open sets (aε,b+ε),,,(a - \epsilon, b + \epsilon), \emptyset, \emptyset, \dots forms an open cover for [a,b][a, b], and so

μ([a,b])(ba)+2εba.\mu^*([a, b]) \le (b - a) + 2\epsilon \le b - a.

We simply show that μ([a,b])ba\mu^*([a, b]) \ge b - a. Let {In}nZ+\{I_n\}_{n \in \bZ_+} be any countable open cover for [a,b][a, b]. By compactness of [a,b][a, b], there exists a finite subcover {Ik}k=1n\{I_k\}_{k = 1}^n. We prove by induction that

k=1nμ(Ik)ba,\sum_{k = 1}^n \mu^*(I_k) \ge b - a,

and so k=1μ(Ik)k=1nμ(Ik)ba\sum_{k = 1}^\infty \mu^*(I_k) \ge \sum_{k = 1}^n \mu^*(I_k) \ge b - a, meaning μ([a,b])=ba\mu^*([a, b]) = b - a.

If [a,b]I1[a, b] \subseteq I_1, then μ(I1)ba\mu^*(I_1) \ge b - a, of course. Now suppose we know that, if [a,b]=I1I2In[a, b] = I_1 \cup I_2 \cup \cdots \cup I_n, then

k=1nμ(Ik)ba.\sum_{k = 1}^n \mu^*(I_k) \ge b - a.

Suppose [a,b]I1In+1[a, b] \subseteq I_1 \cup \cdots \cup I_{n + 1}. Write In+1=(c,d)I_{n + 1} = (c, d), and without loss of generality suppose bIn+1b \in I_{n + 1}. Note that if cac \le a, then μ(In+1)b+a\mu^*(I_{n + 1}) \ge b + a, so we are done. Thus suppose a<c<b<da < c < b < d, and so [a,c]I1In[a, c] \subseteq I_1 \cup \cdots \cup I_n. By inductive hypothesis, note that

k=1n+1μ(Ik)(ca)+μ(In+1)(ca)+(dc)=daba.\sum_{k = 1}^{n + 1} \mu^*(I_k) \ge (c - a) + \mu^*(I_{n + 1}) \ge (c - a) + (d - c) = d - a \ge b - a.

Thus μ([a,b])=ba\mu^*([a, b]) = b - a.

[0.3.8]Corollary(Intervals in R\bR are Uncountable)#

Every interval in R\bR that contains at least two distinct elements is uncountable.

Proof.

Suppose II is an interval containing a,bRa, b \in \bR with a<ba < b. Then

μ(I)μ([a,b])=ba0.\mu^*(I) \ge \mu^*([a, b]) = b - a \ne 0.

Countable sets are measure zero, and so II is not countable.

The outer measure does have some unpleasant properties, however. We would expect that, for any disjoint sets A,BA, B, the outer measure of ABA \cup B is equal to the sum of the outer measure of AA and BB respectively. Unfortunately, this is not true in general.

[0.3.9]Theorem(Nonadditivity of Outer Measure)#

There exist disjoint subsets A,BRA, B \subseteq \bR such that μ(AB)μ(A)+μ(B)\mu^*(A \cup B) \ne \mu^*(A) + \mu^*(B).

Proof.

Consider the set [1,1][-1, 1]. For any a,b[1,1]a, b \in [-1, 1], define the equivalence relation

ab    abQ.a \sim b \iff a - b \in \bQ.

Then the equivalence relation \sim partitions [1,1][-1, 1] in such a way that

[1,1]=a[1,1][a].[-1, 1] = \bigcup_{a \in [-1, 1]} [a]_\sim.

For each coset, choose exactly one point, then let VV be the set of all these points. That is, V[a]V \cap [a]_\sim has a single, unique element for each a[1,1]a \in [-1, 1].

Let r1,r2,r_1, r_2, \dots be a sequence of points such that [2,2]Q={r1,r2,}[-2, 2] \cap \bQ = \{r_1, r_2, \dots\}. Then note

[1,1]k=1(rk+V).[-1, 1] \subseteq \bigcup_{k = 1}^\infty \, (r_k + V).

Indeed, for any a[1,1]a \in [-1, 1], note there is some vV[a]v \in V \cap [a]_\sim. Then avQa - v \in \bQ, meaning there is some kZ+k \in \bZ_+ such that a=rk+va = r_k + v, meaning ark+Va \in r_k + V as desired. By order properties and countable subadditivity of the outer measure, note

2=μ([1,1])μ(k=1rk+V)k=1μ(rk+V)=k=1μ(V).2 = \mu^*([-1, 1]) \le \mu^*\left( \bigcup_{k = 1}^\infty r_k + V \right) \le \sum_{k = 1}^\infty \mu^*(r_k + V) = \sum_{k = 1}^\infty \mu^*(V).

Thus μ(V)>0\mu^*(V) > 0.

Let nZ+n \in \bZ_+. Note that

k=1n(rk+V)[3,3],\bigcup_{k = 1}^n \, (r_k + V) \subseteq [-3, 3],

since V[1,1]V \subseteq [-1, 1] and rk[2,2]r_k \in [-2, 2] for each kk. Again, by order properties of the outer measure, note

μ(k=1n(rk+V))k=1nμ(V)6.\mu^*\left( \bigcup_{k = 1}^n \, (r_k + V) \right) \le \sum_{k = 1}^n \mu^*(V) \le 6.

But also note that

k=1nμ(V)=nμ(V).\sum_{k = 1}^n \mu^*(V) = n \mu^*(V).

Simply choose nZ+n \in \bZ_+ such that nμ(V)>6n \mu^*(V) > 6, and so

μ(k=1n(rk+V))k=1nμ(rk+V).\mu^*\left( \bigcup_{k = 1}^n \, (r_k + V) \right) \le \sum_{k = 1}^n \mu^*(r_k + V).

Note each rk+Vr_k + V is disjoint. Since μ(A1An)μ(A1)++μ(An)\mu^*(A_1 \cup \cdots \cup A_n) \ne \mu^*(A_1) + \dots + \mu^*(A_n) in general for disjoint A1,,AnA_1, \dots, A_n, by induction we have μ(AB)μ(A)+μ(B)\mu^*(A \cup B) \ne \mu^*(A) + \mu^*(B) in general.

[0.3.10]Remark(Intuition behind Disproof of Additivity of Outer Measure)#

We start by taking [1,1][-1, 1] and forming equivalence classes that group numbers that differ by a rational together. From each equivalence class, we choose one point, and we take the set of these points to be VV. Crucially, note that no elements of VV differ by a rational, otherwise two elements of VV would be in the same equivalence class, contrary to construction. We then make copies of the set, shifting VV by all the rationals in [2,2][-2, 2], written as a sequence r1,r2,r_1, r_2, \dots. The translates r1+V,r2+V,r_1 + V, r_2 + V, \dots are all disjoint from eachother, since if they overlapped, there would be two numbers in VV that differed by a rational. We will refer to rk+Vr_k + V as a “copy” of VV, since it is still isomorphic to VV.

If we take infinitely many of these copies and union them together, we note it forms a cover for [1,1][-1, 1]. Thus the outer measure of this union is greater than 22, establishing the outer measure of VV is nonzero.

Now, we look at finitely many copies of VV, say nn of them. The union of nn many copies of VV lie in [3,3][-3, 3], and so the outer measure of this union is bounded above by 66. But the sum of nn many outer measures of VV is given by nn multiplied by the outer measure of VV, which is nonzero. This sum grows without bound, so there exists some nn such that nμ(V)>6n\mu^*(V) > 6, completing the disproof.