The Riemann integral partitions the domain into intervals and asks whether the oscillation of a function is small on most of them. This is poorly suited to a function that changes on a dense set of negligible size. Lebesgue integration instead permits measurable sets as building blocks and approximates a nonnegative function from below by functions with finitely many values. Throughout, is a measure space; may take the value .
Measurability is a condition on inverse images
Definition 1.2.1
(-algebra and measure)
A -algebra on contains and is closed under complements and under countable unions. Consequently it also contains the empty set and is closed under countable intersections. A measure is a function with that is countably additive on pairwise disjoint measurable sets.
Definition 1.2.2
(Measurable function and almost everywhere)
An extended-real-valued function on is measurable if for each real . A property holds almost everywhere if its failure set is contained in a measurable null set.
Strict and weak thresholds are interchangeable: for instance . Thus is measurable. The wording "contained in a measurable null set" also works on a measure space that is not complete, where a subset of a null set need not be measurable.
Proposition 1.2.3
(Countable operations preserve measurability)
If are measurable extended-real functions, then , , , and are measurable. In particular a pointwise limit is measurable whenever it exists everywhere.
Proof
(Threshold sets of suprema and limits)
The set is . The infimum follows by negation, and the two limit functions are respectively and . Each is obtained by countable operations just shown to preserve measurability.
Almost-everywhere limits and measurable representatives
In applications, convergence often fails on a null set. There are two separate questions: can we find a measurable function representing the limit, and is an arbitrarily specified value on the exceptional set also measurable? The first answer is always yes; the second requires care if the measure space is not complete.
Proposition 1.2.4
(Measurable representatives of almost-everywhere limits)
Let be measurable and let an extended-real function satisfy almost everywhere. There is a measurable equal to almost everywhere. If the measure is complete, itself is measurable.
Proof
(Redefine the limit on a measurable null cover)
Choose a measurable null set outside which the asserted convergence holds. By the preceding proposition is measurable everywhere. Define on and on . Its threshold sets are unions of measurable pieces, so is measurable; outside , . If the measure is complete, every subset of is measurable. For any real , the symmetric difference between and is a subset of , hence is measurable. Thus every threshold set of is measurable too.
The measurable null cover is essential in the construction: the set on which an arbitrary differs from the measurable representative need not itself belong to the original -algebra. Choosing the values of on the whole measurable set avoids that difficulty. In a complete space there is no such difficulty, and any function equal almost everywhere to a measurable function is measurable.
Worked example 1.2.5
(An exceptional point of almost-everywhere convergence)
On with Lebesgue measure, let , and specify , for . For we have as soon as , so off , a null set. At convergence fails: for all . The limit representative obtained by replacing the value at by is measurable and equals almost everywhere. Completeness of Lebesgue measure also ensures that the specified is measurable. Almost-everywhere convergence does not mean convergence at every point, and the value at an exceptional point cannot be recovered from it.
Counterexample mode
Completeness cannot be dropped
Claim: an almost-everywhere limit of measurable functions is always measurable. Let , , and . Define when and for the other two measurable sets. This is a measure, with measurable and null, but a nonempty proper subset of is not in . Put everywhere and . Then outside , so convergence holds almost everywhere, yet is not measurable because . The repaired statement is the proposition above: only a measurable representative, here simply , is guaranteed without completeness.
Quick check
(Suppose is measurable and outside a measurable null set in a complete measure space. Why is measurable?)
Compare the threshold sets of and .
Solution · Answer
For every real , the symmetric difference of and lies in the null set. Completeness makes that difference measurable; hence is measurable.
Quick check
(In the Lebesgue-measure example with , can one conclude from almost-everywhere convergence?)
Solution · Answer
No. Here and for every ; the point is exceptional.
For real-valued measurable , the sum is measurable: . The density of the rationals supplies a between and when the sum exceeds . Products and absolute values are measurable as well, using continuous operations. For extended values, expressions such as must first be excluded; they are not defined by measurability.
Why continuous operations preserve measurability
The sum argument used a countable union of threshold sets. To justify products and more complicated formulas, we need the same idea in several coordinates. Here all functions are finite and real-valued.
Definition 1.2.6
(Borel measurable map)
A map is measurable into the Borel sets if for every Borel set .
Theorem 1.2.7
(Coordinate theorem)
The map is Borel measurable exactly when each coordinate is measurable. Consequently, if is continuous, then is measurable.
Proof
(Pull back a countable basis)
For a real measurable , inverse images of open intervals are measurable, because
Every open subset of is a countable union of intervals with rational endpoints. Thus inverse images of all open sets are measurable; since inverse images preserve complements and countable unions, the same holds for all Borel sets.
If is measurable, each is measurable because the coordinate projection is continuous. Conversely, if all are measurable, the inverse image of an open rectangle is , a measurable set. Every open subset of is a countable union of rectangles with rational endpoints: these rectangles form a countable basis. Therefore inverse images of open, and hence Borel, sets under are measurable. Finally, for an open , continuity makes open, so is measurable.
Proof X-ray
Where countability and continuity enter
The measurability of each is used only once: the inverse image of a rational rectangle is a finite intersection of measurable sets. The key move is the countable basis. An open is the union of the rational rectangles it contains, and there are countably many of them, so is a countable union of measurable sets. A -algebra need not be closed under uncountable unions, so writing as the union of all small balls inside it would prove nothing. The passage from open to Borel sets closes the argument: the sets with form a -algebra containing the open sets. In the composition step, continuity of is used only to make open. For a merely Lebesgue-measurable , the set could fail to be Borel, and a continuous inner map need not pull back every Lebesgue-measurable set to a measurable set. The theorem concerns several functions on the same domain; it neither assumes nor proves a product-measure theorem.
Consequences. Addition, multiplication, absolute value, maximum and minimum are continuous functions of their coordinates; their outputs are measurable. For division, is measurable. Define on and on . To see that is measurable without dividing by zero, set
The denominator is positive and the displayed formula is a continuous function of , so is measurable. Pointwise, where and where ; the measurable-limit theorem proves the claim. The chosen value zero is a convention on the zero set, not a value for division by zero.
Worked example 1.2.8
(A quotient that is unbounded but zero almost everywhere)
On with Lebesgue measure let and . Both are measurable. Put when and . The argument above proves measurability even though is unbounded and discontinuous at every point of : at positive rationals and at irrationals. In fact off the countable set of positive rationals, so almost everywhere and . A function can have unbounded pointwise values yet represent zero in .
Quick check
(Let be finite measurable functions. Why are and the function equal to on and to on measurable? Must this quotient be integrable?)
Solution · Answer
The map is continuous, so the coordinate theorem applies. The zero-extended quotient is measurable by the preceding construction, and the requested function is , which is measurable. Integrability does not follow: on with Lebesgue measure take , . The quotient is , whose integral is infinite.
Worked example 1.2.9
(The indicator of the rationals)
On with Lebesgue measure, is measurable because the rationals are countable: each threshold set is , a countable set, or . It is discontinuous at every point, hence not Riemann integrable, but its Lebesgue integral is . The difference lies in the null set of rational points, not in a special convention for a dense set.
Approximation by simple functions
Definition 1.2.10
(Nonnegative simple function and its integral)
A nonnegative simple function is a measurable function with finitely many finite values, written canonically as with distinct and pairwise disjoint measurable level sets . Its integral is
with the convention .
Refining two partitions into their intersections shows that the integral is additive on nonnegative simple functions and monotone: if , then . If and are written on finite measurable partitions of , the intersections form a common partition. On each such piece, has value . Finite additivity of measure therefore gives
All terms are nonnegative, so this remains valid if some measures are infinite. The same refinement shows that two different finite representations of the same simple function have the same integral. If , their values satisfy on every nonempty intersection; summing the corresponding inequalities proves monotonicity without subtracting infinite integrals.
Theorem 1.2.11
(Increasing approximation by simple functions)
For each measurable there are nonnegative simple such that pointwise.
Proof
(Dyadic truncation)
Put when , and when . The dyadic level sets are measurable, and only finitely many levels occur. Equivalently . Dyadic floor approximations increase as increases, and for finite their error is less than once ; at an infinite value .
Uniform approximation of bounded real and complex functions
The preceding theorem uses an order and gives an increasing approximation to a nonnegative function. In later arguments we also need a uniform approximation to a bounded signed or complex function; complex numbers have no compatible order to use here. A complex simple function is a measurable function with finitely many complex values.
Theorem 1.2.12
(Uniform approximation of bounded functions)
Let be a measurable space and let be bounded and measurable, meaning that its real and imaginary parts are real measurable functions. There are complex simple functions with . For real-valued , the same construction gives real simple functions with error at most . No measure or finite-measure assumption is needed.
Proof
(A grid of half-open cells)
Write , choose such that , and for each integer define
For integers , the set on which is
It is measurable. Since both and lie in , only finitely many pairs occur, so is simple. Each coordinate differs from its lower grid value by less than , including for negative coordinates; hence at every point. For a real , omit the imaginary coordinate. Also for all , so the approximants share a bound.
Quick check
(On with its Borel -algebra, let and use the grid above with . Find and ; why is measurable?)
Solution · Answer
At the coordinate floors are and , giving . At , they are and , giving . The grid-cell preimages are intersections of measurable inverse images, as in the proof. Endpoint values are assigned consistently by the half-open cells.
For later use, uniform convergence alone does not imply convergence on an infinite-measure space. For a fixed , on with Lebesgue measure let . Each lies in and , but . On a finite-measure space the uniform bound does imply ; this is why the later arguments specify the size of the space.
Definition 1.2.13
(Integral of a nonnegative measurable function)
For measurable ,
This value may be infinite.
The integral is monotone immediately from the definition. The supremum makes the construction independent of any one choice of approximating simple functions; the monotone convergence theorem provides the important further fact that an increasing simple approximation has integrals converging to this supremum.
Concept lensAnalytic
Slicing the range instead of the domain
For measurable , the integral is the area under the distribution function:
a Lebesgue integral in of a nonincreasing function. For simple both sides equal , because contributes to exactly for . For general take simple ; then , and monotone convergence on both sides gives the formula, with no finiteness assumption on .
Why the size of the domain matters. On a set of finite measure, if measurable and , then monotonicity and the integral of a constant give
provided is bounded, so that all these integrals are finite. The equality uses nonnegative additivity: approximate and from below by simple functions and apply monotone convergence to their sums.
Common mistake
(A uniformly small error can have infinite integral)
Uniform closeness alone does not give convergence of integrals on an infinite-measure space. On with Lebesgue measure, the simple functions converge uniformly to zero, but for every : a small error is spread over an infinite domain. The increasing approximation theorem avoids this difficulty through monotone convergence, not through uniform convergence.
Signed functions and integrability
For real measurable , let and . Then and .
Definition 1.2.14
(Integrable function)
A real measurable is integrable if ; then .
This excludes the ambiguous subtraction , since both are finite.
Common mistake
(Finite at every point is not integrable)
A finite value almost everywhere does not by itself imply integrability: on is finite at every point, but its integral is infinite.
For integrable functions, linearity and follow from additivity for nonnegative functions, proved via monotone approximation. Modifying an integrable function on a null set leaves its integral unchanged, provided the modified function is measurable. Indeed, a nonnegative simple function supported on a null set has integral zero; taking a supremum proves that every nonnegative measurable function supported there has integral zero.
What a zero integral tells us
Integration ignores changes on null sets, but how much information does the integral retain? For a nonnegative function it retains enough to detect whether the function is zero almost everywhere. The key is to look at sets on which the function has a definite positive size, rather than at the entire set where it is merely positive.
Proposition 1.2.15
(Level-set estimate)
If is measurable and , then
Consequently if and only if almost everywhere. No assumption that is finite is needed.
Proof
(Compare with one simple function)
Set . The simple function lies below , so monotonicity gives the estimate, including when either side is infinite. If , each has measure zero. Since
this measurable set is null. Conversely, if off a measurable null set , any nonnegative simple is supported in and has integral zero. Taking the supremum in the definition gives .
Common mistake
(A zero integral does not make a signed function vanish)
The nonnegativity matters: on with Lebesgue measure, has integral zero by cancellation but is nonzero except at . In contrast, , so the proposition applied to detects that is not zero almost everywhere.
Worked example 1.2.16
(Zero distance in the integral norm)
Suppose and are integrable and . Apply the proposition to : it follows that almost everywhere. Conversely, if they agree almost everywhere, their absolute difference is zero off a null set and so has zero integral. This is why measures separation only up to null sets, an observation used when defining as a space of equivalence classes.
Worked example 1.2.17
(An unbounded integrable step function)
Let on , and zero elsewhere in . Its disjoint level sets are measurable, and the interval has length . The finite partial sums increase to , so monotone convergence gives
The unboundedness of near zero does not prevent integrability. The series identity follows by differentiating the geometric series inside its interval of convergence, or by summing its finite partial sums.
Practice with solutions
Exercise 1.2.18
(Prove that the set on which a sequence of real measurable functions converges is measurable.)
CoreProof
Solution(3)
- 1
Replace convergence by the Cauchy condition
The sequence converges exactly when it is Cauchy.
Because: is complete.
- 2
Write the Cauchy set with countable operations
Key step
- 3
Conclude
∎Each is measurable by the coordinate theorem, so each set is measurable, and countable unions and intersections keep the whole set in .
Exercise 1.2.19
(On with Lebesgue measure, let . Find its pointwise limit and its integrals. What does this show?)
Foundation
Solution(3)
- 1
Pointwise limit
Each fixed lies outside once , so .
- 2
Integrals
for every .
- 3
Interpret
: pointwise convergence alone cannot justify interchanging limit and integral.
Exercise 1.2.20
(For a measurable of finite measure and , when is integrable?)
Foundation
Solution(2)
- 1
Compute the integral of
, so .
- 2
Read off the condition
Since by assumption, is integrable exactly when .
Exercise 1.2.21
(If are integrable and , show that for every . Does the converse hold?)
Core
Solution(2)
- 1
Apply the level-set estimate to
Key step
- 2
Test the converse
It fails. On with Lebesgue measure take : for each the set has measure at most , but every .