The central question is when a pointwise limit can be taken through an integral. A moving spike shows why a hypothesis is necessary: converges pointwise to zero on while each integral is one. Monotonicity and integrable domination provide two different remedies. We work on a measure space .
Monotone convergence
Theorem 1.3.1
(Monotone convergence theorem)
If measurable functions satisfy pointwise, then
where either side may be infinite.
Proof
(Approximating a simple function from below)
Monotonicity of the integral gives . For the reverse inequality take any nonnegative simple and . The measurable sets increase and cover : where , the pointwise limit is at least , and where the inequality already holds. For each level set of , continuity from below gives ; this follows from countable additivity by decomposing the increasing union into disjoint increments. Consequently . Since , . Let , then take the supremum over . This also covers infinite integrals.
The theorem immediately gives countable additivity of integration for nonnegative measurable functions: apply it to partial sums. It also gives additivity of the nonnegative integral by approximating both summands from below by increasing simple functions. In particular, no integrability assumption is needed for a sum of nonnegative terms, although its value may be infinite.
Worked example 1.3.2
(A monotone series with an infinite integral)
On , increases to . Thus
This calculation does not rely on uniform convergence near ; the infinite answer is permitted by the theorem.
A decreasing sequence can also pass through the integral, but only after a finiteness condition makes subtraction legitimate.
Corollary 1.3.3
(Monotone convergence for decreasing sequences)
Suppose almost everywhere and . Then .
Proof
(Subtracting from an integrable first term)
After changing values on a null set, take finite everywhere and the monotonicity pointwise. Then , and monotone convergence gives . Because is finite, each of these integrals equals minus the integral of the subtracted function, and subtracting from gives .
The finiteness of is the only point where the decreasing case differs from the increasing one.
Counterexample mode
A decreasing sequence with infinite integrals
Claim: if , then . On with Lebesgue measure take . For each , as soon as , so pointwise and . Yet for every . The repair is the hypothesis : for one has and , as the corollary predicts.
Fatou and dominated convergence
Lemma 1.3.4
(Fatou's lemma)
For nonnegative measurable ,
Proof
(Monotone convergence applied to tail infima)
Put . Then and for . Monotone convergence yields .
Common mistake
(Reading Fatou's lemma as an equality)
Fatou gives a one-sided bound; it does not promise equality or convergence of integrals. The inequality can be strict: the moving spike has pointwise limit zero, so the left side is , while every integral is one.
Equality needs a hypothesis that controls the mass that may escape. Integrable domination is the standard one.
Theorem 1.3.5
(Dominated convergence theorem)
Suppose and are measurable, almost everywhere, and almost everywhere for all , where is integrable. Then is integrable and . In particular .
Proof
(Fatou applied to the complementary quantity)
A countable union of the exceptional null sets is null. After changing functions there if necessary, all inequalities and convergence hold everywhere; this does not alter their integrals. The limit is measurable and , so it is integrable. Apply Fatou to . The integrable upper bound makes all terms finite and the pointwise limit of the displayed expression is . Fatou therefore gives
The nonnegative error integrals have limsup zero. The final assertion follows from the triangle inequality for integrals.
The proof uses each hypothesis exactly once, and the order matters.
Proof X-ray
Where integrability of the dominator enters
Fatou only bounds integrals of nonnegative functions from below, so the proof feeds it the nonnegative quantity , whose pointwise limit is known. The bound is used once, to make this quantity nonnegative. Integrability of is used at the subtraction step: the identity requires to be finite, and only then can be canceled from both sides. With on the displayed inequality reads and proves nothing, which is exactly what the translated indicators in the next paragraph exploit. The argument closes because a limsup of nonnegative numbers that is at most zero forces the limit to be zero.
Common mistake
(A bound must be one integrable function)
The word integrable matters. The constant bound dominates on but is not integrable there; these functions converge pointwise to zero while their integrals remain one. Pointwise boundedness by a different function for each also does not suffice: the same must work for every .
Worked example 1.3.6
(A limit with an exceptional endpoint)
For , the integrands converge to for , and are bounded by the integrable function . Dominated convergence gives . At every integrand equals , so the limit function differs there, but a single point has measure zero and is immaterial.
Integrating a series with signed terms
Monotone convergence allows a nonnegative series to pass through an integral, even if the answer is infinite. A series with changing signs needs a further argument: cancellation in its partial sums does not by itself control the integral of its limit. A useful sufficient condition is that the total of the absolute integrals is finite.
Theorem 1.3.7
(Termwise integration of an absolutely integrable series)
Let be any measure space, and let be real- or complex-valued integrable measurable functions. If
then converges absolutely almost everywhere. Defining its sum to be zero where it fails to converge absolutely gives a measurable integrable function , and
Proof
(Monotone convergence on the absolute series)
Monotone convergence applied to the nonnegative partial sums gives
Consequently is finite almost everywhere. Outside the measurable null set where , the scalar series converges absolutely. Its sum, set to zero on that null set, is measurable and satisfies almost everywhere, so it is integrable. For the partial sum , the pointwise tail estimate holds almost everywhere. Integrating and using monotone convergence again gives the displayed error bound. Finally, linearity for finite sums and yield the identity. No finite-measure assumption on is needed.
Worked example 1.3.8
(An alternating logarithmic series)
On with Lebesgue measure put for . Then
The geometric series sums to , so the theorem justifies
Pointwise convergence of the partial sums is not a substitute for the absolute-integral condition.
Counterexample mode
Telescoping spikes with the wrong termwise integral
Claim: if each is integrable and converges pointwise, then . On let , put , and for . The partial sums telescope to pointwise, so the integral of the sum is . Yet and for , so the sum of the integrals is . The theorem's hypothesis fails: equals on and on , so for , and .
Quick check
(On let for . Why may this series be integrated term by term, and must the sum of the absolute integrals be finite whenever a particular termwise integration happens to be valid?)
Compute , then look for a series where dominated convergence works without the absolute condition.
Solution · Answer
We have , and , so the theorem applies. The condition is sufficient, not necessary. For example, has , but its partial sums are , bounded by on and converging to . Dominated convergence still justifies exchanging the limit of partial sums and the integral.
Bounded convergence on a finite-measure space
Dominated convergence is especially easy to recognize when the same constant bounds every function. The size of the domain then decides whether that constant is integrable.
Theorem 1.3.9
(Bounded convergence theorem)
Let . Suppose are measurable real- or complex-valued functions, almost everywhere, and there is a finite constant such that almost everywhere for every . Then has an integrable measurable representative. Use that representative for and its displayed integrals, including on a noncomplete space. Then
Proof
(A constant becomes an integrable dominator)
The countably many exceptional sets for the bounds and the exceptional set for convergence fit inside one measurable null set. Outside it the pointwise limit has modulus at most . Define to be zero on that null set and to be the pointwise limit elsewhere; this is a measurable representative. The constant function is integrable because , so dominated convergence applies to and gives convergence of their integrals. It also applies to , dominated almost everywhere by , proving the stronger conclusion.
The proof is short because all the work was done by dominated convergence; what remains is to see which hypothesis supplies the dominator.
Proof X-ray
Why a constant bound is enough here
The theorem does not make boundedness a substitute for integrability. It turns a constant bound into an integrable dominator using the separate assumption , since . The common constant is used twice: as the dominator for and as for , after the limit has been shown to obey the same bound. On an infinite-measure space the step from to an integrable function fails, and the translated indicators on show the conclusion fails with it.
The same must work for the entire sequence; being bounded separately for each is weaker. Nor is the hypothesis merely that every takes finite values. On , tends to zero pointwise, and each function is bounded, but all its integrals equal .
Worked example 1.3.10
(The limit of )
For , put on . The logarithmic limit gives , with the case causing no difficulty for the limit. Also . Bounded convergence yields
Here is an independent calculation that checks the result: integrating first gives , which has the same limit. The theorem justifies passing the limit through the integral without requiring this antiderivative.
Worked example 1.3.11
(Moving measurable sets)
Let , and suppose almost everywhere. Since , the conclusion gives . In particular . The converse implication from measures alone is false: on with Lebesgue measure, alternating and have the same measure as for every , but their indicators do not converge to almost everywhere.
Quick check
(With Lebesgue measure on , let for , and . Show that pointwise on , and find .)
Treat the endpoints separately from interior points.
Solution · Answer
At either endpoint both indicators are zero. For , sufficiently large satisfies , so both indicators are eventually one. Since and the omitted two intervals have total length , ; endpoints do not affect the measure.
Differentiating an integral with respect to a parameter
Theorem 1.3.12
(Differentiation under the integral sign)
Let be an open interval and . Suppose is such that is measurable and integrable for every , and there is a single null set such that for every and every , exists and , for some . Then is differentiable at and
Proof
(Difference quotients under a dominator)
For any sequence with and , the difference quotients converge almost everywhere to . The mean value theorem bounds their absolute values by , even though its intermediate point can depend on and . Set the derivative to zero on . It is then a measurable function: it agrees off with the pointwise limit of the measurable difference quotients. Dominated convergence applies. The sequential criterion for limits proves the derivative formula.
Worked example 1.3.13
(Moments of the exponential by differentiation)
For , take on with Lebesgue measure. Near a fixed take ; the derivative is dominated by , which is integrable. Differentiating gives , so . The local dominating function depends on , which is sufficient: differentiability is a local property.
A cancellation integral: choosing the parameter carefully
The theorem becomes especially useful when an integral is hard to evaluate directly but its parameter derivative is simpler.
Worked example 1.3.14
(The integral of )
For , consider
The two terms obtained by splitting the numerator are separately divergent at zero, so subtracting their integrals is not legitimate. The numerator must stay together. For , the mean value theorem applied to bounds the absolute integrand by . For , it is at most , which is integrable. Thus the combined integrand is absolutely integrable.
Fix and let for . Pointwise differentiation gives . At an arbitrary , all these derivatives for are bounded by the single integrable function . The parameter theorem therefore gives
Since , the ordinary fundamental theorem of calculus yields
This also covers : the integral and logarithm are then negative. The cancellation near zero establishes that exists; the separate local bound on the derivative justifies moving differentiation inside the integral. Neither step follows from merely observing a pointwise derivative.
Quick check
(Evaluate .)
Identify and in the cancellation integral.
Solution · Answer
The argument applies with , , giving . Reversing the two exponentials changes the answer to .
This parameter theorem is distinct from the Lebesgue differentiation theorem. For , that deeper theorem states that for almost every , averages over intervals centered at converge to . Consequently an indefinite integral has derivative almost everywhere. Its proof requires additional covering arguments and is not established by dominated convergence alone.
Practice with solutions
Exercise 1.3.15
(Let be measurable and . Prove almost everywhere.)
FoundationProof
Solution(3)
- 1
Integrate the series term by term
.
Because: Monotone convergence applies to the increasing nonnegative partial sums.
- 2
A finite integral forces finiteness almost everywhere
Key step
Let , a measurable set. For every , , so . If , letting would make the integral infinite.
- 3
Conclude
∎Hence , that is, almost everywhere.
Exercise 1.3.16
(Evaluate .)
Core
Solution(3)
- 1
Find the pointwise limit
For , , so the integrand tends to ; at it is , a null set.
- 2
Dominate every integrand by one integrable function
Key step
, which is integrable on .
Because: for .
- 3
Apply dominated convergence
The limit equals .
Quick check
(Can the monotone convergence theorem be applied to on with Lebesgue measure?)
Check every hypothesis of the theorem, not only monotonicity.
Solution · Answer
No: the sequence is increasing and tends pointwise to zero, but its terms are negative. Indeed each extended integral is , so the asserted equality would fail. Nonnegativity is essential in the stated theorem.