An increasing function may jump, and even a continuous increasing function may rise on a set of Lebesgue measure zero. Nevertheless, its ordinary derivative exists and is finite almost everywhere. Stieltjes measure explains the result: the absolutely continuous part supplies the derivative, while the singular part has zero local density almost everywhere.
Let be nondecreasing and finite, with . We use the earlier Stieltjes-measure construction, finite Lebesgue decomposition, the Lebesgue differentiation theorem for functions, and the theorem that a finite measure singular to Lebesgue measure has zero ball density almost everywhere. Those results are inputs here; this argument is not their proof.
A right-continuous representative
Definition 2.3.1
(Right-continuous representative)
For put , and put .
Monotonicity and boundedness make the limits finite. The function is nondecreasing and right-continuous on : if , then whenever . The lower bound tends to , while the upper bound can be taken over all , so .
Extend constantly as to the left of and as to the right of . The Stieltjes construction then gives a finite nonnegative Borel measure supported on , with no atom at , such that
Decompose on , where is integrable and is singular to Lebesgue measure . Extend both by zero outside when applying the differentiation theorems.
The derivative of the right-continuous representative
Lemma 2.3.2
(The density is the derivative of )
At almost every , is differentiable and .
Proof
(Split each increment into density and singular mass)
At almost every , is a Lebesgue point of and
At such an , for small enough, (1) gives
The first term tends to : its deviation is at most twice the average of over . The second is nonnegative and at most , which tends to zero by (2). For , use ; the same two bounds apply with . Thus .
Proof X-ray
Separate mass before taking a quotient
The proof uses its two differentiation inputs at the same point , so it needs the intersection of two full-measure sets, which still has full measure. Identity (3) is the key move: the Stieltjes relation (1) splits every one-sided increment of into a density part and a singular part before any quotient is taken. The density part is compared with a symmetric average, at the cost of a factor , because Lebesgue points are defined by centered intervals. The singular part is handled only by monotonicity of measure: lies in the ball of radius , and nonnegativity of makes enlarging harmless. No claim that vanishes near is made; only its density does.
Return to the original monotone function
The original differs from only at jumps, and a jump cannot occur where is differentiable.
Lemma 2.3.3
( inherits the derivative of )
If and is differentiable at , then and .
Proof
(Trap between one-sided limits of )
For , monotonicity gives
Indeed for every , while . If is differentiable at , it is continuous there, so (4) gives .
Differentiability of at says that for each , all sufficiently close to satisfy
For a nearby , letting shows that the bound also holds with in place of , and it holds for . Since lies between these two values by (4), . Therefore .
Common mistake
(Replacing by everywhere)
The right-continuous representative agrees with only where is right-continuous. At a jump point may be any value between and . The lemma uses only where it is differentiable, and there (4) forces ; at the countably many jumps no claim about is made, and none is needed for an almost-everywhere statement.
The two lemmas combine into the main theorem.
Theorem 2.3.4
(Lebesgue's theorem on monotone functions)
Every finite nondecreasing function on has a finite nonnegative derivative almost everywhere in . Its derivative is integrable and
Here the derivative may be assigned any measurable value on its null exceptional set.
Proof
(Combine the lemmas and compare masses)
By the two lemmas, almost everywhere in , and is finite and integrable. Moreover
The last display also identifies the gap in (5): it equals . The first term is the possible jump at the left endpoint, which the convention excludes from ; the second is the singular mass.
Why the inequality can be strict
Worked example 2.3.5
(Smooth recovery)
If on , then , , , and everywhere. Both sides of (5) equal . Absolute continuity, treated in the fundamental-theorem unit, is a condition that guarantees such recovery.
Worked example 2.3.6
(An interior jump)
On let for , , and for . Then for and for , so , consistent with (4): . The measure is , which is singular, so and . Hence at every , , and the gap equals .
Worked example 2.3.7
(A jump at the left endpoint)
On let and for . Then and , so , and . The derivative is on , and the whole gap comes from the term . This is the edge case that the convention "no atom at " moves out of .
Jumps are not the only way to lose variation; a continuous function can lose all of it.
Counterexample mode
Continuity does not give equality in the monotone theorem
Claim: if is continuous and nondecreasing on , then . For the Cantor function on , the associated Stieltjes measure is the singular Cantor probability measure. Its absolutely continuous density is , so almost everywhere and , yet . Here , and the entire gap is . The repaired statement requires absolute continuity of , which forces and no endpoint jump.
Thus an almost-everywhere derivative need not reconstruct an arbitrary monotone function by integration.
Practice
Quick check
(What does the theorem say about the derivative of a finite nondecreasing function outside a null set?)
List the three properties.
Solution · Answer
Outside a null set the derivative exists, is finite and is nonnegative; moreover it is integrable with integral at most .
Quick check
(Why does the Cantor function show that need not hold?)
Compare the derivative with the endpoint increase.
Solution · Answer
Its derivative is zero almost everywhere, so , but its endpoint increase is . All of its variation is carried by the singular part .
Quick check
(If on , calculate both sides of (5).)
Differentiate and integrate.
Solution · Answer
, so , and . Both sides are ; equality holds because has no singular part.
Quick check
(Why is the ball in the singular estimate allowed to be larger than the one-sided interval?)
Use the sign of .
Solution · Answer
The singular measure is nonnegative, so enlarging to only increases its mass. Since and the ball density tends to zero, the quotient still tends to zero.