This page uses the Darboux and stability topics of the chapter on continuous operations to test where composition preserves integrability and where it does not.
Revision route and checklist
First separate small oscillation on most partition cells from large oscillation confined to a small total length. Next distinguish a continuous outer operation from one that merely has an integral. Read the order of composition before attempting a counterexample.
- For bounded , integrability is equivalent to arbitrarily small weighted oscillation sums .
- A continuous function on a compact value interval is bounded and uniformly continuous.
- If bad cells satisfy , their total length is at most .
- Finite exceptional points can be enclosed in short intervals. Put interior exceptional points strictly inside their intervals, so they do not leak into neighboring uncovered cells through partition endpoints.
- Dense rational and irrational inputs force full oscillation for the rational indicator on every nondegenerate cell.
Preserving information or losing it
Exercise 18.3.1
(Let be Riemann integrable on , with , and suppose its values lie in a compact interval . Let be continuous on . Prove directly from the Darboux criterion that is integrable. Choose the uniform-continuity scale explicitly, split good and bad cells, and bound their contributions.)
SynthesisProof
Solution(4)
- 1
Fix the constants in dependency order
Put , and choose with on ; the composite is bounded by . Given , uniform continuity supplies such that inputs in less than apart have output difference less than . Choose a partition with .
- 2
Bound the good cells
Key step
Their composite oscillation is at most , so their total contribution is at most .
Because: On cells with , every pair of input values satisfies the uniform-continuity condition.
- 3
Bound the bad cells
On the remaining cells , so their total length is at most , while each composite oscillation is at most .
- 4
Add the two contributions
∎The Darboux criterion proves integrability, with every tolerance chosen in dependency order. Continuity of the inner function was never required.
Exercise 18.3.2
(For a bounded real function on , prove that is Riemann integrable if and only if is.)
CoreProof
Solution(3)
- 1
Forward direction
Choose with . If is integrable, composing with the continuous cube function on proves integrability of .
- 2
Reverse direction
Key step
If is integrable, its values lie in . Composing with the real cube root gives back .
Because: Real cube root is continuous on the compact interval .
- 3
Conclude
∎Both implications hold. No bounded derivative of the cube root at zero is needed, because uniform continuity on a compact interval suffices.
Worked example 18.3.3
(Squaring forgets the sign)
Define
Then is continuous with , so it is integrable. But every nondegenerate partition cell contains a rational point where and an irrational point where , so every cell oscillation is at least . Consequently every partition satisfies
where are the cell lengths, and the Darboux criterion fails: is not integrable. Squaring identifies opposite signs and so has no inverse on the interval of possible signed values. Applying a positive square root would recover , not . The cube argument retains that information and can be reversed.
Worked example 18.3.4
(Cubing keeps the sign)
For the same , the cube satisfies at every rational point and at every irrational point, since is increasing and or respectively. So every cell has and for every partition: is not integrable, in agreement with the equivalence of integrability of and . Squaring turned the nonintegrable into the integrable ; cubing cannot do that because it is invertible.
Two integrable functions, a nonintegrable composite
Let list the rational points of without repetition. Define
and define the outer function on by
Theorem 18.3.5
(A function with countably many small values is integrable)
The function is bounded and Riemann integrable on with .
Proof
(Finite head and small tail)
The enumeration gives . Every nondegenerate cell contains an irrational input, so every lower sum is zero. To make an upper sum arbitrarily small, fix and choose with .
The finite head is . Enclose its points in a finite union of small closed intervals of total length less than . Choose their interior endpoints irrational, place every interior point of strictly inside a covering interval, and merge any overlapping intervals. For a point of at 0 or 1, use a one-sided interval with an irrational interior endpoint. Because is finite, the lengths can be made arbitrarily small while satisfying these conditions.
Use all covering endpoints in a partition. Covered cells contribute at most their total length to the upper sum, since . Every uncovered cell, including its endpoints, avoids : an interior covering endpoint is irrational, and any domain endpoint in belongs only to the covered portion. Hence on uncovered cells the values are at most . Their lengths sum to at most one. Thus
Together with , this proves integrability and integral zero.
Proof X-ray
Finite head, uniform tail, and the partition endpoints
The proof is not merely that the nonzero set is countable; a countable modification can destroy integrability, as the rational indicator shows. It uses a specific splitting: finitely many values that may be large (the head, where is as large as ) are trapped in intervals of tiny total length, and every other value is uniformly tiny because decays. The decay is the hypothesis that makes the tail uniformly small; a function equal to at every rational would have no such splitting.
The non-obvious details are the endpoints. If a covering interval ended exactly at a head point, that point would also be an endpoint of a neighboring uncovered cell, where only the tail bound is available and the value can be as large as . Choosing covering endpoints irrational forces there, and putting head points strictly inside their intervals keeps them out of uncovered cells. For the interval is one-sided and its other end is irrational. The argument closes with because the covered length is below and the tail bound is below .
Worked example 18.3.6
(A concrete tolerance with the endpoints in the head)
Take . Then requires , since . Suppose the enumeration starts , , so the head contains both endpoints of the domain. Cover by and by with irrational and . The covered length is . Partition at . The upper sum is at most , because the middle cell has all its values at most , giving . Hence with this explicit partition.
Quick check
(Prove that is Riemann integrable on and calculate its integral.)
Compare upper sums with the lower sum of the partition with nodes .
Solution · Answer
Every partition has upper sum one, since the supremum on every cell is . The partition with nodes , where , has lower sum : the infimum on the first cell is zero and on the second it is one. Letting shows both integrals equal one.
Counterexample mode
The composite of two integrable functions need not be integrable
Claim: if and are Riemann integrable then so is . At every rational input, is strictly positive, so . At every irrational input , so . The composite is the rational indicator: on every cell its supremum is and its infimum is , so every Darboux gap equals and the composite is not integrable. For example, on the rational point for some has and so gives , while the irrational point gives . Although the inner and outer functions each have a proper Riemann integral, the outer function is discontinuous at zero, a value attained by the inner function and approached by its positive values. It therefore does not satisfy the continuity hypothesis of the composition theorem on the compact value interval containing the range. Integrability of the outer function cannot replace that hypothesis.
Common mistake
(Composition order matters)
With the same and , the composite is not integrable, but is: takes only the values and , so equals at and on , a function with at most one point of discontinuity. Both functions being integrable does not determine integrability of a composite. Check which function is outer, and whether it is continuous on the compact interval containing the values supplied by the inner one.
Common mistake
(Countable modification preserves integrability only with control)
Changing a function on a countable set does not in general preserve Riemann integrability: the rational indicator is the zero function modified on the countable set , yet it is not integrable. The integrability of above rests on the decay of its values, as the finite-head argument shows.