Length is clear for an interval, but a subset of the line can have an intricate boundary or no intervals inside it at all. Covering a set by intervals suggests an upper bound for its length. Taking the best such bound defines a size for every subset of . Countable additivity, however, will require us to restrict the sets we call measurable.
From interval covers to outer measure
Definition 1.1.1
(Lebesgue outer measure)
For , define
where is the interval's length and the sum may be infinite.
For every , the empty set has a countable cover by nonempty intervals of total length at most ; hence . Every set has a cover, for example by the intervals , although its total length may be infinite. The infimum need not be attained by a particular cover.
Proposition 1.1.2
(Monotonicity and countable subadditivity)
If , then , and for any sequence of subsets of ,
Proof
(Pay for the -th set)
Every cover of covers , proving monotonicity. If the sum on the right is infinite, subadditivity is immediate. Otherwise, for choose a countable interval cover of each with total length less than . All these intervals together form a countable cover of the union, with total length less than . Let .
Proposition 1.1.3
(Outer measure of an interval)
For , . Every countable set has outer measure zero, and so of a bounded interval does not depend on whether its endpoints are included.
Proof
(Compactness reduces to finite covers)
The upper estimate follows by covering with and sending to zero. Conversely, every open interval cover has a finite subcover by compactness. The sum of lengths of finitely many intervals covering is at least : order and merge overlapping intervals, then compare the length of the merged component containing with the sum. A singleton has outer measure zero by covering it with arbitrarily short intervals, and countable subadditivity gives outer measure zero for every countable set, for instance . Changing the endpoints of a bounded interval changes the set by at most two points, so by monotonicity and subadditivity the outer measure is unchanged.
Worked example 1.1.4
(A countable set by a direct cover)
Let be countable and let . Cover by the open interval of length centered at . These intervals cover and have total length
so for every , that is, . A finite set is the edge case: use the first intervals, or pad the enumeration by repeating ; repeated points only add more intervals and leave the total at most .
Common mistake
(Dense does not mean positive outer measure)
A dense set need not have positive outer measure. Indeed is dense in and has outer measure zero, whereas the interval has outer measure one. Density concerns closure, while outer measure concerns efficient covers.
Carathéodory's splitting test
Definition 1.1.5
(Lebesgue measurable set)
A set is Lebesgue measurable if for every ,
Subadditivity always gives the direction . The definition demands the reverse inequality for every test set , including nonmeasurable ones. Infinite values are allowed; no expression of the form is used.
Common mistake
(Testing only )
Checking the splitting identity for the single test set proves nothing. Since , at least one of the two terms is infinite, so the identity holds for every subset . The test sets must include bounded ones, and in fact all of them.
Theorem 1.1.6
(The measurable sets form a complete measure space)
The Lebesgue measurable sets form a -algebra, and the restriction of to them is a complete countably additive measure.
Proof
(Split repeatedly and keep the residual)
The empty set is measurable, and the splitting equality is symmetric under replacing by , so complements are measurable. If and are measurable, split an arbitrary first by and then split by . Monotonicity and subadditivity imply
so is measurable. Thus finite unions and differences are measurable.
Let be pairwise disjoint measurable sets, and let . Repeated splitting by the first sets gives
Letting and applying countable subadditivity to gives the reverse inequality in the splitting test for . Thus disjoint countable unions are measurable. For an arbitrary sequence, replace each set by the part outside the previous ones; finite differences make the new sets measurable and disjoint. Their union is unchanged, proving closure under countable unions.
Taking in the displayed finite splitting equality yields for every . The opposite inequality follows from countable subadditivity, so , including when the sum is infinite. Finally, if has outer measure zero and , monotonicity gives . For every , subadditivity and monotonicity then give ; hence is measurable and null. This proves completeness.
Proof X-ray
How the splitting test becomes a measure
Only two properties of are used: monotonicity and countable subadditivity. Nothing about intervals enters, so the same argument works for any outer measure. Subadditivity is available before measurability and always supplies the easy inequality; measurability is used only to split a test set exactly, one set at a time. The key move is the finite identity with the residual kept as a separate term. Monotonicity then replaces the residual by the smaller , the only lossy step, and its loss runs in the harmless direction. This is why no limit of residuals has to be computed: letting and applying subadditivity once closes the inequality. Choosing the test set turns the same identity into countable additivity. Completeness needs no splitting at all, because a set of outer measure zero contributes nothing to either side.
Completeness says that every subset of a measurable null set is measurable. It is stronger than simply saying that countable sets are measurable.
Continuity of measure for nested sets
Countable additivity also lets us pass from a nested sequence of sets to its union or intersection. This is a property of a measure on measurable sets, and it will be used when proving convergence theorems. The two directions have different hypotheses: the theorem below assumes a finite starting measure for a decreasing sequence.
Theorem 1.1.7
(Continuity of measure along nested sequences)
Let be any measure space, and let all sets below be measurable.
- If , then , with infinity allowed.
- If and , then .
Proof
(Disjoint increments and complements in )
Put and for . The are pairwise disjoint, their first terms have union , and their full union is . Countable additivity gives
This argument permits an infinite sum. For decreasing sets, increases to . By the first part,
Both and the analogous identity for the intersection involve finite quantities because . Subtracting from gives the second result. No subtraction of is involved.
The intervals used below are measurable, as shown in the next section.
Worked example 1.1.8
(Shrinking intervals)
On with Lebesgue measure, is decreasing, , and : a point leaves once , and . The theorem therefore predicts , and indeed .
Counterexample mode
Decreasing sets of infinite measure
Claim: for every decreasing sequence of measurable sets, the measure of the intersection is the limit of the measures. Take with Lebesgue measure. The sets decrease, and , since any real is excluded from once . So , yet for every and the limit is . The repair is the hypothesis in the theorem: some set of the sequence, which may be taken as , has finite measure. Writing does not repair the proof.
Quick check
(Let and . Find each limiting set and the limit of its measures.)
Solution · Answer
The union of the is and . The intersection of the is and . A limiting intersection need not be empty for its measure to tend to zero.
Why intervals are measurable
The splitting test quantifies over arbitrary , so an interval's measurability still needs proof.
Theorem 1.1.9
(Intervals and Borel sets are measurable)
Every interval, every open subset of , and every set in the Borel -algebra, the -algebra generated by the open sets, is Lebesgue measurable.
Proof
(Split each covering interval at the cut point)
Fix and write , . If there is nothing to prove; otherwise take a countable open-interval cover of of finite total length. Split each interval with into and , choosing small enough and with total added length below a prescribed . Intervals lying entirely to one side need no splitting. The resulting left and right collections cover and respectively, and their total length is at most the original cover's total plus . Taking the infimum over covers and then proves
The reverse inequality is subadditivity, so is measurable. Complements, finite intersections, and countable unions now show that every interval is measurable. Every open subset of is a countable union of open intervals with rational endpoints, so all open sets are measurable. Since the measurable sets form a -algebra, they contain the Borel -algebra.
Concept lensStructural
Measurable sets are approximable from outside by open sets
A set is Lebesgue measurable if and only if for every there is an open with . If is measurable with , an interval cover of total length below has a union with ; for general apply this to with . Conversely, choose for and put : then is null, so is measurable by completeness.
The Lebesgue -algebra is complete; the Borel -algebra alone is not complete for Lebesgue measure. Establishing a concrete non-Borel subset of a Borel null set requires an additional construction or cardinality argument, not merely the observation that null subsets are measurable. It is also false that every subset of must be Lebesgue measurable; the construction of a nonmeasurable set uses a choice principle and is beyond this section.
Practice with solutions
Quick check
(Give two dense subsets of with different outer measures. If has , must it be dense in ?)
Solution · Answer
The rational points of are dense and null, while itself is dense and has outer measure 1. Any of outer measure 1 must be dense: if a nonempty relatively open interval were disjoint from , then , and the at most two intervals left over have total length , so .
Quick check
(If is measurable and is a subset of a null set, prove that is measurable and has the same measure as .)
Solution · Answer
Completeness makes measurable with ; closure under unions makes measurable. Monotonicity and subadditivity give , including the case .
Affine images of Lebesgue measurable sets
Fix , , and let . We use the convention for .
Images of covering intervals. If , then . If , then reverses order, so the endpoints swap: . Its length is . In both cases an open interval is sent to an open interval, and its length is multiplied by . An unbounded open interval and its image both have infinite length, so the same statement holds with extended values.
Theorem 1.1.10
(Affine scaling of outer measure)
For , and every , .
Proof
(Push covers forward and use the inverse map)
Let be a countable cover of by open intervals. Then is a countable open-interval cover of , so . Taking the infimum over all covers gives . This bound holds trivially when . The inverse map is affine with nonzero coefficient . The same inequality therefore gives . Multiplying by yields the reverse inequality. At no point do we subtract quantities that might be infinite.
The theorem is an equality of outer measures, and it holds for every set, including nonmeasurable ones. It does not claim that every set is measurable.
Corollary 1.1.11
(Affine images of measurable sets are measurable)
If is Lebesgue measurable, and , then is Lebesgue measurable and .
Proof
(Transport the splitting test)
Let be an arbitrary test set, and put . Because is a bijection, and . Scaling both pieces and splitting by gives
So satisfies the Carathéodory criterion, and the measure formula is the theorem restricted to measurable sets.
The case . If , then . Otherwise . Both sets are measurable and have outer measure , whatever is. The scaling formula is not used in this case.
Worked example 1.1.12
(A reflection that stretches by two)
Let , , . The set is countable, hence null. So is measurable, and , giving . The map reflects and stretches: and , so . Meanwhile is still a countable null subset. Thus has measure , as the corollary predicts.
Quick check
(For , find .)
Solution · Answer
The image of a countable set is countable, so its outer measure is .
Quick check
(Find when .)
Solution · Answer
We have , whose measure is .