A measurable set need not be open or closed, and a measurable function may be discontinuous everywhere. Nevertheless, on a finite interval these objects can be approximated in measure by much more regular ones. The word “approximately” matters: the exceptional set can be made arbitrarily small, but it need not be possible to make it empty or null.
We work with Lebesgue measure on a fixed finite interval , . Complements below are relative to unless stated otherwise. We use the interval-cover construction of Lebesgue outer measure, measurability of open sets, simple-function approximation, Egorov's theorem and completeness of the real numbers.
Open and compact approximations
Theorem 3.4.1
(Regularity of Lebesgue measure on an interval)
If is Lebesgue measurable and , there are a compact set and an open set with
The outer approximation comes straight from the definition of outer measure; the inner one is obtained by applying the outer one to the complement.
Proof
(Outer covers, then complements)
By the outer-measure definition, cover by countably many open intervals whose total length is less than . Their union is open and measurable, contains , and has measure at most that total. Since and , we obtain .
Apply the same argument to the measurable set , obtaining an open containing it with . Then is closed in the compact interval , hence compact, and . Moreover , so its measure is below .
The finite measure in the subtraction above is important; the proof does not manipulate . A compact subset of approximates from inside in measure, while an open superset approximates from outside. Neither assertion says itself is open or compact.
Continuous cutoffs
Lemma 3.4.2
(Continuous cutoff between a compact set and an open set)
If with compact and open, there is a continuous function with on and outside .
The proof is a formula built from two distance functions.
Proof
(An explicit distance quotient)
For nonempty and , put
Both distance functions are continuous. The denominator is positive: it could vanish only at a point belonging to both the closed set and the closed set . On the second distance is zero and the first is positive; outside the first is zero. If is empty, take ; if is nonempty and , take . In the applications below we replace by its intersection with , so is nonempty.
A concrete case shows what the quotient looks like.
Worked example 3.4.3
(The cutoff for inside )
Here . For we have and , so the denominator is and . For , and . For , and , so . Outside the numerator vanishes and . Thus is the trapezoid that rises linearly on , equals on and falls linearly on . It agrees with except on , a set of measure ; shrinking to reduces that exceptional set to measure .
For a measurable , choose with small. The cutoff then agrees with outside . Thus geometry turns set approximation into a continuous function that is exactly correct away from a controlled exceptional set.
Lusin's theorem
Continuity of a restriction is a statement about the subspace alone, so it is worth fixing precisely before the theorem.
Definition 3.4.4
(Continuity of a restriction)
Let and . Then is continuous on (in the relative topology) if for every and there is such that for all with . For , the restriction is continuous when this holds with ; points are never tested.
With this notion, a measurable function becomes continuous after a set of arbitrarily small measure is removed.
Theorem 3.4.5
(Lusin's theorem on an interval)
If is measurable and finite almost everywhere, then for every there is a compact such that and the real-valued restriction is continuous in the relative topology.
The proof turns simple functions into continuous ones with the cutoff lemma, then upgrades pointwise convergence to uniform convergence.
Proof
(Cutoffs, Egorov and a final compact set)
Include the null set where is not finite among the exceptions. Choose finite-valued simple functions , indexed by , converging pointwise to at every remaining point; for example truncate to and round downward to a grid of mesh . Write with disjoint measurable level sets.
For each , with , put and apply regularity with error to get and . Intersect with so it is bounded without losing or increasing the error. If a level set is empty, take and its cutoff zero. Let be the corresponding cutoff and set . Each is continuous on .
Outside , every cutoff equals the corresponding indicator, so . The union of these sets and the initial null set has measure below . On , consequently pointwise. This set has finite measure; Egorov's theorem supplies a measurable of measure below such that convergence is uniform on .
By inner regularity choose a compact inside whose complement there has measure below . Then . The restrictions are continuous and converge uniformly to ; the uniform-limit theorem proves continuous.
Proof X-ray
Where each hypothesis and each quarter of the error budget is spent
The hypothesis that is finite almost everywhere is used once, at the start: it lets the null set of infinite values be discarded so that the truncated, rounded functions are finite-valued and converge to a real limit. The finite measure of is used twice: in the regularity theorem, where is a legitimate subtraction, and in Egorov's theorem, which fails on sets of infinite measure.
The error budget is a double geometric series. Because , infinitely many cutoffs cost only one quarter; Egorov costs a second quarter and the final compact approximation a third, so the total stays below . The key move is that off for every simultaneously, which is why one exceptional set makes a sequence of continuous functions converge pointwise. The last compact approximation is not cosmetic: is only measurable, and the conclusion asks for a compact set.
Common mistake
(Continuity of the restriction is not continuity of the function)
The theorem concerns continuity on : neighborhoods and convergent sequences are taken within . It does not say that , as a function on all of , is continuous at the points of . The rationals example below has constant while is discontinuous at every point of . Nor can the complement of in general be chosen null.
Concept lensAnalytic
Agreeing with a continuous function off a small set
Lusin's theorem has an equivalent global form: for every there is a continuous with . Given , extend linearly across each bounded open interval of and constantly on the end pieces; the result is continuous on , agrees with on and satisfies . Conversely, given such a , inner regularity yields a compact with , and is continuous.
A density consequence
Corollary 3.4.6
(Continuous functions are dense in )
Let . For every and there is a continuous with .
To see the role of regularity, first use the standard density of finite simple functions in and take . Choose compact with as small as needed, and let be continuous cutoffs. Since ,
The right-hand side can be made arbitrarily small; the triangle inequality then combines this approximation with that of by . This is approximation, not uniform approximation of every measurable function.
Common mistake
(Density in does not extend to )
The estimate above uses , which has no analogue when . Indeed with has distance at least in from every continuous : near a continuous takes values close to on both sides, and cannot be within less than of both and .
Examples and limits of the conclusion
Worked example 3.4.7
(Everywhere discontinuous, yet continuous on a large compact set)
Let on a nondegenerate interval . Every interval contains rationals and irrationals, so is discontinuous at every point of . Enumerate its rational points as and cover each by an open interval of length less than . The complement of their union in is compact, has , and contains no rationals. Thus is continuous. Every point of is nevertheless a discontinuity of on , because rationals accumulate there. Relative continuity on does not imply continuity at the same point in the original domain.
The indicator of an arbitrary measurable set needs no Egorov step at all.
Worked example 3.4.8
(Lusin's theorem for an indicator )
Let be measurable and . By regularity choose compact and with and . Then is compact and . The sets are disjoint and compact, so they lie at a positive distance apart; for , every with lies in the same piece as . Hence is locally constant, and so continuous on . In the edge case we may take , and .
Counterexample mode
The exceptional set need not be removable completely
For the same , no compact subset of full measure can make continuous. A compact full-measure subset is closed; if it missed a point, its nonempty relatively open complement would contain a nondegenerate interval and have positive measure. Thus , on which is nowhere continuous. Lusin's arbitrary positive error cannot be replaced in general by zero. The obstruction is closedness: on the full-measure but non-closed set the restriction is the constant .
Practice with solutions
Quick check
(If is measurable and , find a compact and an open satisfying the regularity theorem for a prescribed .)
The inner set may be empty.
Solution · Answer
Take , so . Since , the outer-measure definition gives open intervals covering with total length below ; their union satisfies .
Quick check
(Why does the cutoff equal outside when ?)
Split the complement of into two pieces.
Solution · Answer
The complement of is . On both and equal , because . Outside both equal , because .
Quick check
(For , why is measurable, and why is its restriction continuous on some compact set of measure greater than ?)
Use countability of the rationals.
Solution · Answer
The rationals are countable, hence a measurable null set, so is measurable. Cover the rationals of by open intervals of total length below and let be the complement of their union in . Then is compact, , and is continuous.
Quick check
(Does Lusin's theorem assert that the continuous approximations converge uniformly on all of ?)
Recall what a uniform limit of continuous functions must be.
Solution · Answer
No. The proof obtains uniform convergence only after removing exceptional sets, then restricts to a compact subset. An everywhere discontinuous function such as cannot be a uniform limit of continuous functions on the whole interval, because such a limit is continuous.