This page applies the Darboux and stability foundations of the chapter on continuous operations to two outer functions: the square root, which is not Lipschitz, and a clipping function, which is. In both cases the weights given by partition-cell lengths must be kept.
Facts used
- On a partition with cell lengths , .
- A bounded function is Riemann integrable exactly when such gaps can be made arbitrarily small.
- Linear combinations, absolute values and products of bounded Riemann-integrable functions are Riemann integrable.
- Cauchy–Schwarz for finite sums retains all the weights: .
A non-Lipschitz outer function still works
Throughout, let , let be bounded and Riemann integrable, and write .
Lemma 18.2.1
(Square-root inequality)
For , . Consequently, on every partition cell . No bounded derivative at zero is needed.
Proof
(Rationalizing the difference)
Suppose . If both are zero the result is immediate. Otherwise,
because . Interchanging the inputs proves the absolute-value version. The same argument includes , with equality there.
On a cell let and . Its square-root values lie between and , so its oscillation is at most . This uses bounds rather than assuming either extremum is attained.
Theorem 18.2.2
(Square-root gap bound)
For every partition of , . Consequently is Riemann integrable, and if then .
Proof
(Cauchy–Schwarz with the cell lengths as weights)
For any partition,
Choose a partition with . Its square-root gap is then strictly less than . The square root is bounded because is bounded and nonnegative. The Darboux criterion proves integrability.
Proof X-ray
Why the weights must be split as they are
The pointwise inequality gives , but the gap is not controlled by directly, because enlarges small numbers. The key move is to split each term as , so that the first factor is a square root of exactly the quantity that makes up , and the second carries only the total length . Cauchy–Schwarz then trades a sum of square roots for the square root of a sum. Applying Cauchy–Schwarz to instead would produce the unweighted sum , which is not controlled by .
The argument closes by an explicit choice of tolerance: demanding in advance makes the right side , so the target gap for is quadratically smaller than the target for . This quadratic price is the cost of not being Lipschitz at .
Common mistake
(Square root is not Lipschitz on an interval containing zero)
The difference quotient of between and is . For it equals , so no Lipschitz constant works on . A proof that composes with by claiming is therefore invalid; the weaker estimate is the correct replacement.
Counterexample mode
The gap bound needs the interval length
Claim: , with no factor . Take , on and on , and the one-cell partition . Then , so and , while . The claim fails because . The correct bound holds, with equality here.
Proposition 18.2.3
(Converse for square roots)
If is bounded and is Riemann integrable, then is Riemann integrable.
Proof
(Squaring back)
Put . Product stability gives integrability of . The nonnegative assumption is essential to the meaning of the real square root.
Clipping without enlarging an error
Definition 18.2.4
(Clipping function)
Define the clipping function
which replaces every value below by and every value above by .
Proposition 18.2.5
(Clipping is 1-Lipschitz and stable in the integral)
for all real . For bounded Riemann-integrable on , and are integrable and
Proof
(Composing two clamps)
For a real constant , both and are 1-Lipschitz: for instance , by checking the cases (equality), (difference ) and (then ). Hence , a composition of two such maps, satisfies .
The Lipschitz composition theorem proves integrability of and , so their difference and its absolute value are integrable. Using the integral triangle inequality and the pointwise clipping bound,
Worked examples
Worked example 18.2.6
(The square-root gap bound on the identity function)
Let on and take equal cells. Then , so . For the oscillations add up, after multiplying by and telescoping, to . The bound gives , and indeed . The estimate holds but is not sharp for this function.
Worked example 18.2.7
(An equality case at zero)
Let , and let on and on . On a cell of length containing in its interior, and : equality holds in the oscillation bound because the cell contains a point where . Here and , and the gap bound reads , true since . Also becomes arbitrarily large as , which shows again that no Lipschitz constant exists.
Worked example 18.2.8
(The exact integral of a clipped line)
Take . The two clipping thresholds occur at and . Thus
It is continuous at both joins. Direct integration gives
The middle contribution is the interval length times the average of its linear endpoint values, . Check: . Notice that : clipping does not preserve the integral.
Common mistake
(Clipping bounds the error, not the value)
The stability estimate controls by . It does not say that the clipped function equals the original function, nor that clipping preserves the integral; the example above has while .
Exercises
Quick check
(Is the real square root Lipschitz on ? Give a reason, and say what replaces the Lipschitz bound for oscillations.)
Compute the difference quotient against .
Solution · Answer
No: is unbounded as , so no constant works. The replacement is the Hölder-type estimate , which gives .
Exercise 18.2.9
(Let be Riemann integrable on with , where . Give an explicit interval guaranteed to contain .)
Core
Solution(3)
- 1
Use the exact value for the model function
From the worked example, , and is integrable.
- 2
Apply the stability estimate
Key step
Because: Clipping is 1-Lipschitz, so pointwise.
- 3
State the enclosure
This controls the integral error; it does not claim that the clipped function equals the original one.
Exercise 18.2.10
(Let be bounded on and suppose is Riemann integrable. Prove that is Riemann integrable, and explain why the argument does not need any Lipschitz property.)
CoreProof
Solution(3)
- 1
Name the integrable function
Put . It is bounded and, by assumption, Riemann integrable.
- 2
Use product stability
Key step
because makes .
Because: Products of bounded Riemann-integrable functions are integrable.
- 3
Conclude
∎Hence is Riemann integrable. Only the algebraic identity and product stability are used, so no Lipschitz bound on is involved; the Lipschitz question arises only for the forward direction.