A function can have derivative zero almost everywhere and still increase: the Cantor function does so. The missing hypothesis in the familiar formula is not merely continuity. Absolute continuity controls the total change over a collection of short intervals, preventing variation from hiding on a null set. We prove the converse fundamental theorem using that control, while identifying both deeper inputs: almost-everywhere differentiability of monotone functions and the Lebesgue differentiation theorem used for indefinite integrals.
Throughout, is real-valued on a finite interval , , and all integrals and null sets are Lebesgue. A complex-valued version follows by applying the result to real and imaginary parts.
Absolute continuity and bounded variation
Definition 2.2.1
(Absolute continuity and total variation)
is absolutely continuous (AC) if for every there is such that every finite collection of pairwise disjoint intervals with satisfies
The total variation of is
Taking a single interval shows that an absolutely continuous function is uniformly continuous. The first consequence is a bound on variation.
Lemma 2.2.2
(Absolute continuity gives finite variation)
Every absolutely continuous on has .
Proof
(Refine by a fixed fine subdivision)
Choose for , and partition into finitely many, say , subintervals each of length less than . Given an arbitrary partition, refine it to include these fixed subdivision points; refinement cannot reduce the sum of absolute increments. Within each fixed subinterval, the partition intervals have total length below , so their increment sum is below . Adding over the fixed subintervals bounds every partition sum by .
The converse is false: a step function has finite variation but is discontinuous. Even a continuous function of finite variation need not be absolutely continuous; the Cantor function is a continuous nondecreasing counterexample.
Differentiability and integrability of the derivative
The existence of rests on a separate differentiation theorem for monotone functions, proved in the unit on monotone differentiability via measures. We do not derive it from the conclusion we are proving.
Theorem 2.2.3
(Differentiability of monotone functions)
A real nondecreasing function on a finite interval has a finite derivative almost everywhere.
If has finite variation, put and . Variation is additive across adjacent intervals, so is nondecreasing and for . Hence is also nondecreasing, and . The monotone theorem gives a finite almost everywhere. Since is continuous, its difference quotients with are measurable, and their pointwise limit on the measurable set where it exists is measurable. Assigning zero on the remaining null set gives a measurable representative of , which we fix from now on.
Lemma 2.2.4
(The derivative is integrable)
If is absolutely continuous, its almost-everywhere derivative belongs to , with
Proof
(Average the difference quotients and apply Fatou)
Extend and constantly outside ; the inequality for persists. For put . Changing variables,
because on and on . Along , almost everywhere, so Fatou's lemma gives .
Why an absolutely continuous function with zero derivative is constant
The following lemma is the step that fails for the Cantor function.
Lemma 2.2.5
(Zero derivative forces a constant)
If is absolutely continuous on and almost everywhere, then is constant.
Proof
(Gauge partition with two kinds of tags)
Fix in and . Choose from absolute continuity so that disjoint intervals of total length below have total absolute increment below . Let contain all points where the derivative is absent or nonzero. By outer regularity cover by an open set of measure less than .
For each , differentiability with derivative zero supplies a radius such that whenever and . For each , choose so that its neighborhood within lies in .
There exists a finite tagged partition with tags such that each interval lies inside the radius- neighborhood of its tag. This elementary gauge partition lemma follows by bisection: if an interval has no such partition, at least one half also has none. Repeating produces nested closed intervals whose lengths tend to zero; their common point has a positive radius, so a sufficiently small interval in the sequence admits a one-tag partition, a contradiction.
Split the partition intervals into those with tags in and those with tags outside it. The first group lies in , so its total length is at most , and absolute continuity bounds the sum of its increments by . For a tag outside , the triangle inequality through bounds the interval's increment by . Therefore
Since was arbitrary, . As were arbitrary, is constant.
Proof X-ray
Where the exceptional set is controlled
The proof meets two kinds of points. Off the hypothesis is used pointwise: it bounds the increment over a short interval by times its length, and these bounds add up to . On nothing pointwise is known, so absolute continuity takes over. Its hypothesis concerns total length, which is why the open cover with is needed: saying that each bad interval is short would be useless, since there may be many of them, and may be dense. The gauge partition is the move that closes the argument: a single finite partition lets every interval be assigned to exactly one of the two controls, without discarding an uncountable set of points. For the Cantor function lies inside the Cantor set, and the step that fails is the second control, because that function is not absolutely continuous.
The converse fundamental theorem
With the three lemmas in hand, the indefinite integral of can be compared with .
Theorem 2.2.6
(Converse fundamental theorem of calculus)
If is absolutely continuous on , then exists almost everywhere, belongs to , and for every ,
The value assigned to on its exceptional null set does not affect the integral.
Proof
(Subtract the indefinite integral)
The preceding lemmas give existence and integrability. Define . The integral-to-derivative theorem of the unit on Lebesgue points and indefinite integrals says that the indefinite integral of an function is absolutely continuous and has that integrand as its derivative almost everywhere; it uses the Lebesgue differentiation theorem. Thus is absolutely continuous and almost everywhere. Then is absolutely continuous, has derivative zero almost everywhere and satisfies . The zero-derivative lemma gives everywhere, which is the formula.
Combining the two directions gives an exact characterization: a function on is absolutely continuous if and only if it is a constant plus the indefinite integral of an function. The integrand is unique almost everywhere: if two integrands have the same indefinite integral, the integral-to-derivative theorem identifies each with the derivative of that common function almost everywhere.
Products and integration by parts
There are two separate questions: does the product have enough regularity for the fundamental theorem, and is its derivative the expected expression? Answering the second alone would not answer the first.
Theorem 2.2.7
(Integration by parts for absolutely continuous functions)
Let be real-valued and absolutely continuous on . Then is absolutely continuous and, almost everywhere, . Both terms on the right belong to , and for every ,
Proof
(Increment estimate, then the fundamental theorem)
Absolute continuity implies continuity, so compactness makes and bounded. Write and . For a finite collection of pairwise disjoint intervals , add and subtract to obtain
Given , use absolute continuity of and to make their respective sums less than and whenever the total interval length is small enough. Taking the smaller of the two thresholds makes the displayed sum less than . Thus is absolutely continuous.
By the converse fundamental theorem, and exist and are integrable. Away from the union of their two exceptional null sets, the pointwise product rule gives . Boundedness of and integrability of make both products integrable. Apply the fundamental theorem to between and , and rearrange.
Worked example 2.2.8
(An unbounded derivative is allowed)
Take and on . Although is unbounded near zero, , so is absolutely continuous; is Lipschitz. The endpoint value of is irrelevant. Integration by parts gives
Check: the left integrand is , whose integral is . The example needs integrability of the derivative, not its boundedness or continuity.
Absolute continuity is used in the proof only to make recoverable from its derivative, and it cannot be weakened to continuity with an almost-everywhere derivative.
Counterexample mode
Integration by parts with the Cantor function
Claim: if are continuous on and differentiable almost everywhere with integrable derivatives, the integration-by-parts formula holds. Take the Cantor function and . Then almost everywhere and everywhere, so both integrals in the formula are , while . The formula fails. The repaired statement assumes that and are absolutely continuous, which the Cantor function is not.
Quick check
(Put and on . Verify the hypotheses and use integration by parts to find .)
Split the integral of at the corner.
Solution · Answer
Both functions are Lipschitz, hence absolutely continuous. Almost everywhere , while on and on ; the value at the corner is immaterial. Therefore
Examples and practice
Worked example 2.2.9
(A corner is allowed)
On , is Lipschitz, hence absolutely continuous. Its derivative is on and on , and is undefined at zero. The formula gives : for the integral is , and for it is , irrespective of the value assigned to the integrand at zero. A point of nondifferentiability is compatible with the theorem.
Worked example 2.2.10
(Continuity alone is insufficient)
Define and for . This is continuous at zero. Let for positive integers . Then , so and successive absolute increments equal . Since diverges by comparison with the harmonic series, has infinite variation and cannot be absolutely continuous. Continuity controls small individual changes but not their accumulated variation.
Quick check
(Let on . Verify its absolute continuity and recover it by integrating its derivative.)
Look for a Lipschitz bound.
Solution · Answer
on , so is Lipschitz and therefore absolutely continuous: take . Its derivative is , and .
Quick check
(Suppose are absolutely continuous on and almost everywhere. What is the relationship between them?)
Apply the zero-derivative lemma to a suitable function.
Solution · Answer
is absolutely continuous with derivative zero almost everywhere, hence constant: for some constant .
Quick check
(Explain why the Cantor function does not contradict the converse fundamental theorem.)
Ask which hypothesis it lacks.
Solution · Answer
The Cantor function is continuous, nonconstant and has derivative zero almost everywhere, so the zero-derivative lemma shows that it cannot be absolutely continuous. Directly: at stage the retained intervals have length each, so their interiors are disjoint open intervals of total length , while the function rises by on each, a total increment of .
Lipschitz composition preserves absolute continuity
Composition with a Lipschitz outer function is the simplest way to produce new absolutely continuous functions.
Proposition 2.2.11
(Lipschitz composition)
Let be absolutely continuous, and let satisfy for all . Then is absolutely continuous.
Proof
(Transfer the increment bound)
If , then is constant, so is constant and trivially AC. Now let and . Choose such that for every finite family of disjoint subintervals with . For any such family,
The hypothesis cannot be weakened to continuity of . Let be the Cantor function and set for , for , for . Since and , is continuous on . With the AC function on , , which is not AC.
A derivative bound for
Take , so and is AC. Let be the set of points where both and exist; by the converse fundamental theorem applied to and , is null. For and small , the reverse triangle inequality gives
Letting gives on , hence almost everywhere.
Common mistake
(Applying the chain rule through a corner)
Do not write with : does not exist. For , is not differentiable at . The ordinary chain rule applies only where is differentiable at and is differentiable at ; only the inequality above is claimed.
Worked example 2.2.12
(Total variation from the absolute derivative)
On let . Then for and for ; at the one-sided derivatives are and in some order, so fails to exist there. Thus almost everywhere, and
Check: is monotone on with changes , summing to .
Quick check
(If are absolutely continuous on , prove that and are absolutely continuous.)
Express both through .
Solution · Answer
is AC, so is AC by the Lipschitz composition proposition with . Since sums and scalar multiples of AC functions are AC, so are and .
Smooth outer functions: the chain rule
When the outer function is continuously differentiable, the expected formula does hold almost everywhere. Again there are two questions: is the composite absolutely continuous, and is its derivative the expected expression?
Theorem 2.2.13
(Chain rule for a smooth outer function)
Let be AC and let be continuously differentiable. Then is AC. At every where exists, is differentiable at and ; in particular this holds almost everywhere. Moreover , and for every ,
Proof
(Lipschitz on the range, then a difference-quotient factorization)
Step 1: a bounded derivative on the compact range. need not be Lipschitz on all of ; is not. Since is continuous on , its range is a compact interval (a point if is constant). Because is continuous, is finite, and the mean value theorem gives for . Let be the nearest-point projection onto ; it satisfies . Then is -Lipschitz on and , so the Lipschitz proposition shows that is AC.
Step 2: the pointwise chain rule. Fix at which exists, and write . Define
Differentiability of at says exactly that is continuous at , and for every , including . Put and divide by :
As , , hence , while the last quotient tends to .
Step 3: the exceptional set is harmless. Let consist of , and the points where fails to exist; it is null. With the measurable representative of fixed earlier, is continuous and , so is integrable. By Step 1 and the converse fundamental theorem, is the integral of over . By Step 2 this integrand equals outside , and changing an integrand on a null set does not change its integral.
Proof X-ray
Which hypothesis each step uses
Continuity of is used only in Step 1, to make Lipschitz on the compact range; the projection then lets the Lipschitz proposition be quoted. Without such a bound Step 1 is lost: the continuous Cantor-type outer function of the previous section gives a composite that is not even AC. Step 2 uses only differentiability of at . Its key move is the factor , which avoids dividing by , a quantity that may vanish for arbitrarily small . For Step 2 breaks at every with , and that set need not be null (take zero on a subinterval). With , every exceptional point comes from itself, so Step 3 only has to discard a null set before the converse fundamental theorem closes the proof.
Substitution without monotonicity
Definition 2.2.14
(Oriented integral)
For a continuous function and real numbers , write for the integral over when , and for the negative of the integral over when .
With this convention, for all real , whatever their order.
Theorem 2.2.15
(Substitution)
Let be AC, and let be a real continuous function on an interval containing . Then and
No monotonicity of is assumed.
Proof
(Apply the chain rule to a primitive)
Let and let be the projection onto used above. Replacing by changes no value of on , which contains every value and every point between and ; so we may assume that is continuous on . Define the oriented primitive
Additivity gives for all , and because is continuous, the classical fundamental theorem gives everywhere. Thus is continuously differentiable. The smooth chain rule with and shows that is integrable and
The primitive is defined on the whole range, independently of how moves through it. The left side weights by the signed velocity , so a stretch of values crossed upward and later downward contributes two canceling amounts; only the net passage from to survives. The proof uses continuity of to make differentiable at every point and well defined. For a merely measurable neither is available, since may map a set of positive measure into a null set; no claim is made for such .
Worked example 2.2.16
(A non-monotone substitution)
Let on and . The function is Lipschitz, hence AC, and it is not monotone: it decreases from to and then increases to . Almost everywhere for and for . Directly,
The substitution theorem gives the same value from endpoint data alone:
The values in are traversed twice, first downward and then upward, and these two contributions cancel.
Common mistake
(Integrating over the range instead of between endpoint values)
In the last example the range of is , and is not the answer. The right side of the substitution formula is the oriented integral from to . An integral over the range, weighted by , belongs to a different statement for strictly monotone maps.
Quick check
(Let on . Show that is AC although is unbounded, and compute both by substitution and directly.)
Write as a constant plus an indefinite integral.
Solution · Guided solution
Let for , for , and . Then and , so . For , , and for , ; hence for every . By the characterization after the converse fundamental theorem, is AC, and away from . With , substitution gives
Directly, for the integrand is , and for it is . Hence the integral is .
Quick check
(Let be AC on with for all . Show that , that , and that is AC.)
Find an interval on which is continuous.
Solution · Answer
The range is compact, and because the minimum is attained at a point where is positive. Since is continuous on , substitution gives and . The same argument on gives , a constant plus an indefinite integral of an function; hence is AC. Positivity on the whole closed interval is what keeps the range inside an interval where is continuous.