A parameter can change an integral in two ways: it changes the integrand at points already included, and it can move the limits of integration. This chapter separates these effects and proves a formula that keeps both.
Fixed endpoints: one estimate for every integration point
Theorem 5.4.1
(Differentiation under the integral sign)
Let be an open interval, , and suppose and its partial derivative are jointly continuous on . Define
Then is and
Proof
(A uniform bound on the difference quotient)
Fix and choose a closed parameter interval around it contained in . Joint continuity makes uniformly continuous on its product with . For sufficiently small nonzero , the one-variable mean value theorem, applied to for each fixed , gives
The intermediate point from the mean value theorem may depend on , but the right side controls all at once and tends to zero. Integrate the inequality: the error in the difference quotient of is at most times that supremum. Thus it tends to zero, proving the formula. Uniform continuity similarly shows that the integral of depends continuously on , so is in fact .
Proof X-ray
Why the moving intermediate point is harmless
The one delicate point is that the mean value theorem produces an intermediate parameter that depends on the integration variable, and one cannot integrate a quantity that depends on an unknown, possibly non-measurable choice . The proof never integrates : it bounds the error pointwise by a supremum over all admissible and , and that supremum is a single number.
Joint continuity is what makes the supremum small: on the compact set (parameter interval) it gives uniform continuity, so the supremum tends to zero as . Compactness of supplies the finite factor when the pointwise bound is integrated. Without joint continuity the supremum need not tend to zero, which is exactly how the example with a moving jump below breaks the argument.
A proper compact interval matters in this proof. An infinite endpoint or an unbounded integrand needs an additional theorem or a separate uniform tail estimate; neither is covered merely by writing this formula.
Moving endpoints and orientation
Theorem 5.4.2
(Leibniz rule with moving endpoints)
Let be open intervals, and continuous on , and let take values in . Using oriented integrals, put
Then
The upper boundary contributes with a plus sign, the lower boundary with a minus sign, and the remaining integral accounts for changes in the integrand at fixed . The formula remains valid when the limits coincide or reverse: is part of the definition.
Proof
(Reduce to a function of two variables)
Choose any fixed and define . Locally near any , the relevant values lie in a compact subinterval of . The fixed-endpoint result gives , and the fundamental theorem gives . These partial derivatives are continuous: for , split a change into a parameter change, controlled by uniform continuity on a compact rectangle, and an endpoint change, controlled by a local bound times the change in endpoint. Thus is . Since
the multivariable chain rule gives the two endpoint terms and . No assumption that is needed.
Worked examples
Worked example 5.4.3
(Fixed endpoints, including the parameter value zero)
Let . Both and are continuous on , so . At this is . For , integration by parts gives
For an independent check, for , and the quotient rule gives the same expression. The closed form is undefined at , but the integral formula supplies , and the closed form tends to as (expand ).
Worked example 5.4.4
(All three contributions)
For real , let
The polynomial integrand and its partial are continuous everywhere, and both endpoint functions are . The formula gives
For an independent check, direct integration with the given order of endpoints yields
whose derivative is the same. At , the interval has zero length and the integral term vanishes, yet : the upper endpoint moves twice as fast as the lower one. Equal endpoints at one instant do not mean that the derivative of the changing integral is zero.
Worked example 5.4.5
(A parameter in the integrand and in the upper limit)
Define, for every real ,
Find for all , including . Both and are continuous on , and the endpoints are . The moving-endpoint formula, with oriented integrals, gives
At the integral has equal endpoints, giving . For , direct integration gives , so the quotient rule yields
This agrees with the integral formula. The quotient expression is undefined at zero, so its absence there is not evidence that the derivative fails: the original integral formula gives the value directly. For negative , the integral is oriented from down to ; the same derivative formula applies without inserting an absolute value.
Common mistake
(Forgetting the speed of the endpoint)
When only the upper limit depends on , the contribution is the integrand at the endpoint times the endpoint's derivative. For the correct answer is , not : the factor is the chain rule. Likewise the lower limit enters with a minus sign, so .
When an apparently harmless derivative loses a moving jump
Counterexample mode
A jump that travels with the parameter
Claim: if exists and equals zero except at one point of each -slice, one may ignore that exceptional point and integrate the derivative. For , let if and if . Each integral is proper and equals , so . At a fixed the parameter derivative is zero, but it fails to exist at . Replacing it with zero at that exceptional point and integrating would give zero, whereas the derivative of the integral is . The joint-continuity and existence hypotheses fail precisely on the moving boundary . Repair: write with the continuous integrand , and use the moving-endpoint formula: . The jump of the original integrand reappears as the endpoint term.
Practice
Quick check
(Compute with the moving-endpoint formula, and check it by integrating first.)
Only the lower limit moves.
Solution · Answer
The integrand does not depend on , so only the lower endpoint contributes: . Directly, the integral equals , whose derivative is .
Quick check
(Let for . The integrand is not differentiable in at , so the fixed-endpoint theorem does not apply. Is differentiable?)
Split the integral at .
Solution · Answer
Yes. For , , so . This equals , so the formula still holds. The hypotheses of the theorem are sufficient, not necessary: they fail only at the single point of each slice, which carries no integral mass here, unlike the moving jump above.
Exercise 5.4.6
(On set . Find its pointwise limit, decide whether convergence is uniform, and justify the limit of its integrals.)
Core
Solution(3)
- 1
Find the pointwise limit
For every term is zero; for fixed the denominator tends to infinity, so the pointwise limit is zero.
- 2
Bound all points at once
Key step
This single bound shows that , so the convergence to zero is uniform.
Because: The same bound must serve every .
- 3
Pass the limit through the integral
. Here the uniform estimate directly justifies the integral limit , rather than an unsupported interchange of pointwise limits and integration.
Because: Each is continuous, hence Riemann integrable.